View Board at the Johns Hopkins Bloom- Following Procedure

Article

Metabolomic Alterations Associated with Cause of CKD

| Morgan E. Grams,*†‡ Adrienne Tin,†‡ Casey M. Rebholz,†‡ Tariq Shaﬁ,*†‡ Anna Ko¨ttgen,†§ Ronald D. Perrone, | | | Mark J. Sarnak, Lesley A. Inker, Andrew S. Levey, and Josef Coresh†‡

Abstract Background and objectives Causes of CKD differ in prognosis and treatment. Metabolomic indicators of CKD cause may provide clues regarding the different physiologic processes underlying CKD development and progression. *Division of fi Nephrology, Design, setting, participants & measurements Metabolites were quanti ed from serum samples of participants in Department of the Modification of Diet in Renal Disease (MDRD) Study, a randomized controlled trial of dietary protein Medicine, and ‡Welch restriction and BP control, using untargeted reverse phase ultraperformance liquid chromatography tandem mass Center for Prevention, spectrometry quantification. Known, nondrug metabolites (n=687) were log-transformed and analyzed to Epidemiology, and Clinical Research, Johns discover associations with CKD cause (polycystic kidney disease, glomerular disease, and other cause). Discovery Hopkins University, was performed in Study B, a substudy of MDRD with low GFR (n=166), and replication was performed in Study Baltimore, Maryland; A, a substudy of MDRD with higher GFR (n=423). †Department of Epidemiology, Johns Hopkins Bloomberg Results Overall in MDRD, average participant age was 51 years and 61% were men. In the discovery study School of Public Health, (Study B), 29% of participants had polycystic kidney disease, 28% had glomerular disease, and 43% had CKD of Baltimore, Maryland; another cause; in the replication study (Study A), the percentages were 28%, 24%, and 48%, respectively. In §Institute of Genetic the discovery analysis, adjusted for demographics, randomization group, body mass index, hypertensive Epidemiology, Faculty of medications, measured GFR, log-transformed proteinuria, and estimated protein intake, seven metabolites Medicine and Medical Center, University of (16-hydroxypalmitate, kynurenate, homovanillate sulfate, N2,N2-dimethylguanosine, hippurate, homocitrul- Freiburg, Freiburg, | line, and 1,5-anhydroglucitol) were associated with CKD cause after correction for multiple comparisons Germany; and Division (P,0.0008). Five of these metabolite associations (16-hydroxypalmitate, kynurenate, homovanillate sulfate, N2, of Nephrology, N2-dimethylguanosine, and hippurate) were replicated in Study A (P,0.007), with all replicated metabolites Department of Medicine, Tufts Medical exhibiting higher levels in polycystic kidney disease and lower levels in glomerular disease compared with CKD Center, Tufts University of other causes. School of Medicine, Boston, Massachusetts Conclusions Metabolomic profiling identified several metabolites strongly associated with cause of CKD. Clin J Am Soc Nephrol 12: 1787–1794, 2017. doi: https://doi.org/10.2215/CJN.02560317 Correspondence: Dr. Morgan E. Grams, 2024 East Monument, Room 2-638, Introduction function affects many aspects of metabolic health, Baltimore, MD 21205. Different causes of CKD have different pathologic fi Email: mgrams2@ abnormal metabolomic pro les may mediate the path- jhmi.edu signatures, require different therapies, and progress at ogenesis and prognosis of CKD (10–12). Metabolites that different rates (1,2). Because of these health implica- differ according to CKD cause above and beyond level of tions, the 2012 Kidney Disease: Improving Global GFR, proteinuria, diet, and medication use may provide Outcomes guideline for the evaluation and manage- novel, disease-specific treatment targets or the potential ment of CKD recommended incorporating cause of for early, noninvasive diagnostic techniques. CKD along with level of GFR and albuminuria in CKD Designed as a two-by-two factorial randomized clinical staging (2). Polycystic kidney disease (PKD) in partic- trial in two substudies, the Modification of Diet in Renal ular is characterized by a relentless decline in GFR, Disease (MDRD) Study is a unique population of with few effective therapies available (3). Recently, im- patients with expert-adjudicated cause of CKD, care- pairment in distinct metabolic processes such as fatty fully monitored protein intake, and measured GFR acid oxidation has been hypothesized as a factor in using urinary iothalamate clearance (13). Using global PKD pathogenesis and progression (4). metabolomic profiling of stored serum, we investi- Metabolite profiling, or an unbiased assessment of gated the associations of individual metabolites with small molecules in a biologic specimen, has recently CKD cause, classified as PKD, glomerular disease, been applied to studies of prognosis in CKD (5–9). A and CKD of other cause, excluding CKD attributed to metabolomic approach quantifies low–molecular diabetes mellitus. We used the smaller MDRD sub- weight biomarkers influenced by genetic variation, study (Study B) for discovery and the larger MDRD diet, medications, a person’s microbiome, liver func- substudy (Study A) for replication of metabolite tion,and,particularlyinCKD,GFR.Becausekidney associations with cause of CKD. www.cjasn.org Vol 12 November, 2017 Copyright © 2017 by the American Society of Nephrology 1787 1788 Clinical Journal of the American Society of Nephrology

Materials and Methods 687 named compounds within 79 pathways were analyzed Study Population (Supplemental Appendix 2). Per protocol, metabolite The MDRD Study was a clinical trial of dietary protein values are normalized by run-day using spiked quality restriction and BP target implemented in a two-by-two- control standards to allow chromatographic alignment, factorial design (13). The trial was composed of two then divided by the median value of the metabolite (7). substudies on the basis of enrollment GFR: study A Missing values were imputed with the minimum value. consisted of patients with GFR between 25 and 55 ml/ All metabolites were log-transformed, after which 84.9% min per 1.73 m2, and study B consisted of patients with had a skewness between 21and1. GFR between 13 and 24 ml/min per 1.73 m2.StudyA randomized patients to usual protein diet or a low-protein Statistical Analyses diet (1.3 or 0.58 g of protein per kilogram of body weight Patient characteristics at the 12-month clinical visit were per day, respectively), and study B randomized patients compared by CKD cause using chi-squared, Kruskal– to a low-protein diet or a very low–protein diet with ketoacid Wallis, or t tests, as appropriate. To determine metabolites and amino acid supplementation (0.58 and 0.28 g/kg per associated with cause of CKD, linear regression was used, day, respectively). Both studies randomized patients to regressing log-transformed metabolites (dependent vari- usual versus low target BP. A heterogeneous group of able) on nonordered categoric cause of CKD (independent patients was recruited into the trial, with noteworthy variable), with adjustment for the following potential exclusions being patients with diabetes mellitus treated confounders: race, age, sex, log-transformed GFR measured with insulin and kidney transplant recipients; all partici- by urinary clearance of 125I-iothalamate, log-transformed pants provided informed consent. For the purpose of this total proteinuria, body mass index, diet randomization study, we selected stored serum samples from the 12-month group, BP randomization group, estimated protein intake, visit, 1 year after randomization. There were 697 MDRD diastolic BP, and angiotensin-converting enzyme inhibitor participants with sufficient serum at this visit for metab- and b blocker use, with the last three variables chosen due to olomic profiling. Of these, 19 were missing concomitant significant differences in baseline values by cause of disease. measured GFR, 27 were missing body mass index or Each analysis was performed separately in the discovery estimated protein intake, and an additional 62 had a cause study (study B) and replication study (study A). of disease distinct from the selected causes described below The threshold for statistical significance accounted for (e.g., diabetic nephropathy). This study was approved by multiple testing and intrametabolite correlation using the the institutional review board at the Johns Hopkins Bloom- following procedure. The 687 metabolites were included berg School of Public Health (Baltimore, MD) and is adher- in a principal component analysis, where 63 principal ent to the Declaration of Helsinki. components explained 90% of the metabolite variance. Statistical significance was thus set in the discovery cohort as a P value ,0.0008 (0.05 of 63 principal components) (17). Classification of Kidney Disease In the replication cohort, the threshold for statistical Cause of disease was classified as PKD, glomerular significance was set using a Bonferroni P value of ,0.007 disease, or CKD of other cause (interstitial nephritis, ves- (0.05 of 7, where 7 was the number of metabolites tested). icoureteral reflux, hypertensive nephropathy, single kidney, For metabolites significantly associated with cause of disease, and unknown), as adjudicated by experts (Supplemental we evaluated correlations with measured GFR and gener- Table 1) (14). To minimize overlap between CKD of other ated residuals from linear regression of metabolites on the cause and glomerular disease, patients adjudicated to CKD above covariates without adjusting for cause of CKD, and of other cause who had preadjudication diagnoses of focal estimated the Spearman correlations of the residuals. This sclerosis, chronic renal failure–proteinuria, and nephrotic procedure was performed separately within the discov- syndrome without a biopsy were excluded (n=46). We ery and the replication studies. further excluded 16 participants with a diagnosis of diabetic To assess the discriminatory capacity of candidate metab- nephropathy. Of note, diagnoses were biopsy-confirmed in olites for cause of CKD beyond clinical characteristics, a 131 of the 148 with glomerular disease and 32 of the 274 with model was built using the following clinical characteris- CKD of other causes. tics identified apriori: age, sex, race, systolic and diastolic BP, log-transformed measured GFR, and log-transformed Metabolomic Profiling proteinuria. To minimize confounding by randomization Metabolite profiling was performed by Metabolon, Inc. arm, study, diet and BP target assignment, and estimated (Durham, NC) in 2015 using serum samples collected at the protein intake were also included. The area under the 12-month postrandomization visit. These specimens were curve(AUC)forthismodelwascalculatedfortwo-way drawn in the original trial before GFR measurement, refriger- classification (i.e., PKD versus no PKD, and glomerular ated, and sent to the central laboratory in batches once a week. disease versus no glomerular disease) and then compared There, specimens were aliquoted for creatinine measurement, with a model that included clinical characteristics and the with the remaining sample frozen and stored at 280°C. replicated metabolites in the full MDRD study. Sensitiv- Metabolomic profiling was performed using an untargeted ity and specificity were assessed using the Youden metabolomic quantification protocol (Supplemental Ap- method for determining cut-points (18). Continuous net pendix 1) (15,16). reclassification index and integrated discrimination in- In order to focus on potential treatment targets, as- dex were also calculated to evaluate the two models. yet-unidentified metabolites and drug metabolites were ex- Analyses were performed using Stata/MP 14.1 (College cluded (n=427 and n=79, respectively). After these exclusions, Station, TX) and R. Clin J Am Soc Nephrol 12: 1787–1794, November, 2017 Metabolites and Cause of CKD, Grams et al. 1789

Results discovery study (Figure 1, Table 2, columns 1–4). These Baseline Characteristics of the Discovery Study (Study B) included metabolites involved in amino acid metabolism and Replication Study (Study A) (kynurenate, homovanillate sulfate, homocitrulline), lipid There were 166 participants in the discovery study: 48 with metabolism (16-hydroxypalmitate), and nucleotide me- PKD, 46 with glomerular disease, and 72 with other causes tabolism (N2,N2-dimethylguanosine); a carbohydrate (hypertensive nephrosclerosis, interstitial nephritis, vesiculo- (1,5-anhydroglucitol); and a xenobiotic (hippurate) (Sup- ureteral reflex, single kidney, unknown cause) (Table 1). plemental Table 3). Most demonstrated higher levels in Average age was 51 years, 84% were white, and 61% were PKD than glomerular disease, with CKD of other causes men. Mean GFR was 15 ml/min per 1.73 m2,andmedian having intermediate levels, with the exception of two, proteinuria was 0.4 g/d. Estimated protein intake was 0.6 g/kg homocitrulline and 1,5-anhydroglucitol, which were per day. In the replication study, there were 423 participants: higher in glomerular disease. Of all 687 tested metabolites, 119 with PKD, 102 with glomerular disease, and 202 with other slightly more (52%) were higher in PKD compared with causes. Average age was 52 years, 86% were white, and 61% other disease, and slightly fewer (49%) were higher in PKD were men. Mean GFR was 35 ml/min per 1.73 m2 and median compared with glomerular disease (Supplemental Figure proteinuria was 0.1 g/d. In both studies, participants with PKD 1). Residuals from the adjusted regressions showed pos- or glomerular disease were slightly younger than those with itive correlation between kynurenate, homovanillate sul- CKD of other cause (49, 48, and 55 years, respectively; fate, N2,N2-dimethylguanosine, and hippurate, with little P,0.001); white participants were more likely to have PKD, correlation with 16-hydroxypalmitate (Figure 2). 1,5- and black and Hispanic participants were more likely to have anhydroglucitol, a monosaccharide that decreases in the set- glomerular disease or CKD of other cause (Supplemental ting of hyperglycemia, and homocitrulline, a byproduct of Table 2). Median proteinuria was 0.1 g/d in PKD, 1.3 g/d in arginine and proline metabolism, were negatively corre- glomerular disease, and 0.1 g/d in CKD of other cause. Study lated with the others. randomization arm did not differ by CKD cause; however, there were differences in hypertension treatment, with fewer Replication of Metabolite Associations with Cause of CKD in participants with other CKD cause using angiotensin- Study A converting enzyme inhibitors, and fewer participants with All seven identified metabolites were tested for their glomerular disease using b blockers. associationwithcauseofCKDinstudyA.Therewerefive metabolites that remained significant in the adjusted model Discovery of Individual Metabolites Associated with Cause (Table 2, columns 5–8). These included 16-hydroxypalmitate, of CKD in Study B kynurenate, homovanillate sulfate, N2,N2-dimethylguanosine, In adjusted analyses, there were seven metabolites and hippurate. Of note, the direction of association comparing significantly associated with cause of disease in the PKD to CKD of other cause and PKD to glomerular disease was

Table 1. Characteristics of Modiﬁcation of Diet in Renal Disease Study participants, by substudy

Characteristic Discovery (Study B) Replication (Study A)

N 166 423 Mean age (SD), yr 51 (12) 52 (12) Race, N (%) Black 9 (5) 33 (8) Hispanic/other 17 (10) 25 (6) White 140 (84) 365 (86) Men, N (%) 102 (61) 259 (61) Diet assignment, N (%) Very low protein 77 (46) 0 (0) Low protein 89 (54) 208 (49) Usual protein 0 (0) 215 (51) Moderate BP assignment, N (%) 78 (47) 205 (49) CKD cause, N (%) Polycystic kidney disease 48 (29) 119 (28) Glomerular disease 46 (28) 102 (24) Other cause of CKD 72 (43) 202 (48) Mean systolic BP (SD), mmHg 132 (16) 129 (17) Mean diastolic BP (SD), mmHg 79 (10) 79 (9) Mean BMI (SD), kg/m2 25.4 (3.8) 27.3 (4.2) Mean estimated protein intake (SD), g/kg per day 0.60 (0.19) 0.95 (0.28) Mean GFR (SD), ml/min per 1.73 m2 15 (5) 35 (11) Median urine protein (ﬁrst, third quartiles), g/d 0.4 (0.2, 1.2) 0.1 (0.0, 0.8) Kidney biopsy, N (%) 49 (31) 119 (30) History of diabetes, N (%) 6 (4) 12 (3)

BMI, body mass index. 1790 Clinical Journal of the American Society of Nephrology

similar for all seven metabolites using this model. In contrast, untargeted global metabolomic profiling and cause of the differences between glomerular disease and CKD of other CKD. Leveraging the substudies of the MDRD trial, a rigor- cause were, in general, smaller and not always maintained ously performed clinical trialwithmeasuredGFRandclose (Supplemental Table 4). The correlations between metabolite monitoring of protein intake, we identified and replicated five residuals in the replication study were also largely un- metabolite associations (kynurenate, homovanillate sulfate, hip- changed (Figure 2). All replicated metabolites were nega- purate, N2,N2-dimethylguanosine, and 16-hydroxypalmitate) tively correlated with measured GFR: 16-hydroxypalmitate that demonstrated consistently higher levels in PKD com- (r=20.11; P,0.01), kynurenate (r=20.66; P,0.001), homo- pared with glomerular disease and CKD of other causes. vanillate sulfate (r=20.77; P,0.001), N2,N2-dimethylguano- These distinctions in metabolites persisted after adjustment sine (r=20.82; P,0.001), and hippurate (r=20.53; P,0.001). for demographics, GFR, proteinuria, diet, and medication use, and thus might represent pathophysiologic differences Additional Discrimination using Identified Metabolites in the development of PKD. Using clinical variables only, two-way classification One promising metabolite with low correlation to the fi models (i.e., PKD versus no PKD and glomerular disease other identi ed metabolites was 16-hydroxypalmitate. 16- v versus no glomerular disease) showed good discrimina- hydroxypalmitate is the -oxidation product of palmitic acid, tion in the combined discovery and replication cohorts. For one of the most common saturated fatty acids found in v b example, the AUC from a clinical model of PKD versus no humans (19). -Oxidation is an alternative to -oxidation, the PKD was 0.81 (95% confidence interval [95% CI], 0.77 to preferred route for fatty acid metabolism, and is thought to b 0.85) and the AUC for glomerular disease versus no assume a more important role when -oxidation is defective glomerular disease was 0.83 (95% CI, 0.80 to 0.87). Adding (20). Thus, the higher levels of 16-hydroxypalmitate in the five replicated metabolites increased the AUC for both PKD could represent an impairment in mitochondrial b classification models, with an AUC of 0.89 (95% CI, 0.87 to -oxidation. This observation is consistent with previous 0.93) and 0.85 (95% CI, 0.82 to 0.89), respectively (P,0.001 studies implicating defects in lipid handling as a driver of CKD – and P=0.04 compared with the clinical model). Similarly, progression as well as in mouse models of PKD (4,21 23). the probability of PKD, glomerular disease, and CKD of Defects in fatty acid metabolism may affect tubular epithe- other cause differed by cause of disease (Figure 3). lial cells, which rely on fatty acid oxidation as an energy fi Sensitivity, continuous net reclassification index, and in- source (24). Tubulointerstitial brosis has been associated tegrated discrimination index were also improved for each with impaired fatty acid oxidation, with intracellular fatty classification when the five replicated metabolites were acid accumulation in the tubular epithelial cells (23). In a added to clinical variables, and specificity was improved for recent mouse model of PKD, Pkd1 knockout cells exhibited the classification of PKD compared with no PKD (Supple- defective palmitate oxidation, and restriction of dietary mental Table 5). lipids resulted in a small improvement in the Pkd1 knockout mouse kidney-to-body weight ratio (4). Several of the PKD-associated metabolites have been pre- Discussion viously named as uremic toxins, and there may be implica- To our knowledge, this is the first study to evaluate the tions for treatment (25,26). Hippurate is a glycine conjugate of associations between small molecules identified through benzoic acid, related to diet and the gut microbiome, that is

Figure 1. | Six metabolites met the discovery signiﬁcance threshold [dotted line, P<0.001 based on the likelihood ratio test (LRT)] in association between metabolites and cause of CKD in study B of the Modiﬁcation of Diet in Renal Disease Study. Red font indicates the metabolite associations that were replicated in study A. Black and gray colors alternate over the x axis to separate metabolites by the associated superpathway. lnJA o eho 2 7719,Nvme,2017 November, 1787–1794, 12: Nephrol Soc Am J Clin

Table 2. Metabolites significantly associated with cause of CKD in the Modification of Diet in Renal Disease Study, sorted by discovery statistical significance

Discovery (Study B) Replication (Study A)

Percent (%) Percent (%) Percent (%) Percent (%) Difference Percent (%) Difference Percent (%) Difference Difference (Polycystic Difference (Polycystic Likelihood Difference Likelihood Metabolite (Pathway) (Polycystic (Polycystic Kidney (Glomerular Kidney Ratio Test P (Glomerular Ratio Test Kidney Kidney Disease Disease Disease versus Value Disease P Value Disease versus Disease versus versus versus Other) Glomerular versus Other) Other) Other) Glomerular Disease) Disease)

16-hydroxypalmitatea (fatty acid, 48 23 51 1.5E-05 60 6 51 1.9E-22 monohydroxy) Homocitrulline (urea cycle, 220 40 243 2.7E-05 211 15 223 3.7E-02 arginine and proline metabolism) 1,5-anhydroglucitol (glycolysis, 240 24 238 8.8E-05 215 21 214 8.5E-02 gluconeogenesis, and pyruvate metabolism) Kynurenatea (tryptophan 22 221 54 2.9E-04 21 25 27 8.6E-05 metabolism) Homovanillate sulfatea 63 26 75 3.4E-04 67 0 67 1.1E-07 (phenylalanine and tyrosine metabolism) a 2 2

N2,N2-dimethylguanosine 23 4 27 4.2E-04 13 2 15 3.0E-04 1791 al. et Grams CKD, of Cause and Metabolites (purine metabolism, guanine containing) Hippuratea (benzoate 58 215 86 7.7E-04 40 1 38 5.1E-03 metabolism)

Percent difference reflects difference in metabolite levels in each cause relative to the other, and was calculated as 1003(exp[b]21), where b is the coefficient from the regression of log-transformed metabolite as the dependent variable on cause of CKD as a nonordered independent variable, adjusted for the following covariates: race, age, sex, log(GFR), body mass index, study diet and BP assignment, estimated protein intake, angiotensin-converting enzyme inhibitor, b blocker, and log(proteinuria). Statistical significance was determined by P,0.0008 (=0.05 divided by 63, the number of principal components explaining 90% of the metabolite variance) in the discovery cohort (n=166; study B) and P,0.007 (=0.05 divided by 7, the number of metabolites that were significant in the discovery analysis) in the replication cohort (n=423; study A). aMetabolite associations that were replicated in study A. 1792 Clinical Journal of the American Society of Nephrology

Figure 2. | Consistent direction in Spearman correlations of significant metabolites in discovery (n=166; study B) and replication (n=423; study A) cohorts in the Modification of Diet in Renal Disease Study. Metabolite levels used in correlation analyses were the standardized residuals adjusted for race, age, sex, log(measured GFR), body mass index, study diet and BPassignment, estimated protein intake, angiotensin- converting enzyme inhibitor use, b blocker use, and log(urine protein per day). *Denote replicated metabolites. secreted by the proximal tubule (27). Mouse models demon- Lower hippurate clearance has been associated with in- strated that treatment with the CIC-2 chloride channel creased risk of mortality and a trend toward faster CKD activator lubiprostone attenuated kidney failure–related progression, independent of urea and creatinine clearance changes in the gut microbiota, lessened the accumulation of and albumin excretion rate (31). Torres and colleagues (35) hippurate, and reduced kidney fibrosis and local inflamma- demonstrated for PKD that blood flow decline in the tion (28). As another example, kynurenate levels are elevated kidney parallels increases in total kidney volume, a marker in kidney failure, particularly in relation to tryptophan (29). of disease severity, and precedes the decline in GFR. One This imbalance might be induced by upregulation of the possibility is that the early decline in blood flow in the indoleamine-2,3-dioxygenase (IDO-1) enzyme in the pres- kidney in PKD compared with GFR differentially affects ence of inflammation, with improvement through the use of the secretion of small molecules by the proximal tubule. niacin supplementation or IDO-1 inhibitors (30). N2,N2-dimethylguanosine, a degradation product of The metabolites identified as higher in PKD compared transfer RNA excreted by the kidney, also has reduced with other causes of CKD might signify differences in the urinary excretion in patients with kidney failure and may location of the pathologic-anatomic findings. Hippurate relate to proximal tubule health (36). and kynurenate have been implicated as markers of the Beyond markers of disease processes and potential health of the proximal tubule, a construct that only partially treatment, other potential uses for disease-associated me- correlates with GFR (31). Hippurate, kynurenate, and ho- tabolites include early or minimally invasive detection of movanillate are thought to be substrates of the organic anion disease. In the case of PKD, a metabolomic signature may transporter family, located on the basolateral membrane of be particularly relevant in children and younger adults, the proximal tubule, with ATP–binding cassette transporters before ultrasound can detect cysts in the kidney, and in mediating urinary secretion on the luminal side (32–34). whom the genetic variant is unknown. Although the

Figure 3. | Differences in predicted probabilities of polycystic kidney disease (PKD) versus no PKD, glomerular disease (GD) versus no GD, and other cause versus no other cause from a model using clinical variables and metabolites in the full Modiﬁcation of Diet in Renal Disease Study (n=589). Clin J Am Soc Nephrol 12: 1787–1794, November, 2017 Metabolites and Cause of CKD, Grams et al. 1793

metabolomic associations differentiating glomerular dis- case of 16-hydroxypalmitate, provide some evidence to ease from CKD of other cause were weaker than that between support the hypothesis of impaired fatty acid metabolism PKD and other causes, metabolites associated with GN might as an underlying driver of PKD. We view this work as inform pretest probabilities of finding disease on biopsy or hypothesis-generating, which should be followed by future allow for diagnosis in persons in whom kidney biopsy is not studies in study populations with careful phenotyping and feasible. Alternatively, classification of disease on the basis using targeted assays for absolute quanti fication of metab- of metabolites might allow for etiologic classification in olites. With additional translational work, metabolites may existing research studies with stored specimens, enabling provide insight into processes underlying differential disease more rigorous testing of the risk associated with cause of development and progression. CKD. Additional discovery and replication work in popu- fi lations with earlier stages of CKD and ner disease classi- Acknowledgments fication as well as the development of targeted assays with M.E.G. receives support from the National Institute of Diabetes more accurate quantification may refine these approaches and Digestive and Kidney Diseases (NIDDK) (K08DK092287 and and bring them closer to clinical practice. R01DK108803). A.T. receives support from R01 DK108803. J.C., Our study design benefits from an unbiased approach to L.A.I., A.S.L., M.E.G., and M.J.S. receive support from CKD Bio- metabolite detection, expert-adjudicated classification of markers Consortium (NIDDK U01 DK085689). J.C., L.A.I., and A.S.L. disease, careful assessment of GFR and protein intake, receive support from CKD-Epi panel eGFR (R01 DK097020). A.K. and a rigorous study design with discovery and replication was supported by German Research Foundation grants KO 3598/3-1 of findings in separate substudies with different ranges of and KO 3598/4-1. T.S. is supported by R03-DK-104012 and R01-HL- GFR. Five out of seven discovered metabolites were repli- 132372. C.M.R. is supported by a mentored research scientist de- cated, all of which showed higher levels in PKD than CKD velopment grant from the NIDDK (K01 DK107782). of other causes. On the other hand, differences between glomerular disease and CKD of other cause were weaker. Disclosures This may represent imperfect classification between glomer- Assay costs were discounted as part of a collaboration agree- ular disease and CKD of other causes or heterogeneity within ment between Metabolon and J.C., L.A.I., and A.S.L. to develop categories. The MDRD Study was rigorous in categorizing metabolomic estimates of GFR and for which they have a provisional participants according to their cause of CKD, but a kidney patent filed on August 15, 2014, entitled “Precise estimation of GFR biopsy was not uniformly required (37). Future work should from multiple biomarkers” (no. PCT/US2015/044567). The technol- investigate the feasibility of metabolomic classification in ogy is not licensed in whole or in part to any company. populations with more homogeneous disease classification, as well as populations with greater minority representation (,10% of MDRD Study participants were black, a signif- References 1. Wong CS, Pierce CB, Cole SR, Warady BA, Mak RH, Benador NM, icant underrepresentation of the population with CKD Kaskel F,Furth SL, Schwartz GJ; CKiD Investigators: Association of stage G3 and G4). proteinuria with race, cause of chronic kidney disease, and glo- Unbiased metabolomic profiling is excellent for evaluating a merular filtration rate in the chronic kidney disease in children broad range of metabolites, but precision is limited because study. Clin J Am Soc Nephrol 4: 812–819, 2009 quantification is relative, not absolute. Samples in MDRD 2. Kidney Disease Improving Global Outcomes: Clinical practice guideline for the evaluation and management of chronic kidney were refrigerated up to a week before shipment, which could disease. Kidney Int.Suppl 3: 1–150, 2013 affect the profiling of some metabolites. Despite this, the 3. Spithoven EM, Kramer A, Meijer E, Orskov B, Wanner C, Caskey F, correlations between metabolite creatinine and clinical creat- Collart F, Finne P, Fogarty DG, Groothoff JW, Hoitsma A, Nogier inine were high, as were metabolite correlations among blind MB, Postorino M, Ravani P, Zurriaga O, Jager KJ, Gansevoort RT; ERA-EDTA Registry; EuroCYST Consortium; WGIKD; EuroCYST duplicates. Subsequent work with targeted assays and sam- Consortium; WGIKD: Analysis of data from the ERA-EDTA ples drawn expressly for the purpose of assessing etiologic Registry indicates that conventional treatments for chronickidney differences in the metabolome may provide stronger and disease do not reduce the need for renal replacement therapy in previously unidentified associations. autosomal dominant polycystic kidney disease. Kidney Int 86: As with all observation studies, causality cannot be 1244–1252, 2014 4. Menezes LF, Lin CC, Zhou F, Germino GG: Fatty acid oxidation is determined; this analysis is merely a discovery study and impaired in an orthologous mouse model of autosomal dominant proof-of-concept. Additional work should replicate findings polycystic kidney disease. EBioMedicine 5: 183–192, 2016 in external cohorts and investigate potential causal roles in 5. Hocher B, Adamski J: Metabolomics for clinical use and research animal models. There is no absolute standard for adjusting in chronic kidney disease. Nat Rev Nephrol 13: 269–284, 2017 6. Zhao YY: Metabolomics in chronic kidney disease. Clin Chim for multiple comparisons in untargeted metabolomic anal- Acta 422: 59–69, 2013 yses, where metabolites are often highly correlated (38). 7. YuB, Zheng Y,Nettleton JA, Alexander D, Coresh J, Boerwinkle E: Because we structured our study as a discovery and replication Serum metabolomic profiling and incident CKD among African cohort, we allowed a slightly more permissive approach in the Americans. Clin J Am Soc Nephrol 9: 1410–1417, 2014 discovery cohort (17) so as to minimize type II error. In the 8. Zheng Y, Yu B, Alexander D, Manolio TA, Aguilar D, Coresh J, Heiss G, Boerwinkle E, Nettleton JA: Associations between me- replication study, we used the conservative Bonferroni cor- tabolomic compounds and incident heart failure among African rection in order to minimize type 1 error. Americans: The ARIC Study. Am J Epidemiol 178: 534–542, 2013 In summary, we evaluated percent differences in metab- 9. Zheng Y, Yu B, Alexander D, Mosley TH, Heiss G, Nettleton JA, olites by cause of kidney disease beyond differences in GFR, Boerwinkle E: Metabolomics and incident hypertension among blacks: The atherosclerosis risk in communities study. proteinuria, diet, and medications, identifying a group of Hypertension 62: 398–403, 2013 metabolites strongly associated with cause of disease. The 10. Rhee EP,Clish CB, Wenger J, Roy J, Elmariah S, Pierce KA, Bullock associated metabolites are biologically plausible and, in the K, Anderson AH, Gerszten RE, Feldman HI: Metabolomics of 1794 Clinical Journal of the American Society of Nephrology

chronic kidney disease progression: A Case-Control Analysis in 28. Mishima E, Fukuda S, Shima H, Hirayama A, Akiyama Y, the Chronic Renal Insufficiency Cohort Study. Am J Nephrol 43: Takeuchi Y, Fukuda NN, Suzuki T, Suzuki C, Yuri A, Kikuchi K, 366–374, 2016 Tomioka Y, Ito S, Soga T, Abe T: Alteration of the intestinal 11. Zhang ZH, Chen H, Vaziri ND, Mao JR, Zhang L, Bai X, Zhao YY: environment by lubiprostone is associated with Amelioration Metabolomic Signatures of chronic kidney disease of diverse of adenine-induced CKD. J Am Soc Nephrol 26: 1787–1794, etiologies in the rats and humans. J Proteome Res 15: 3802–3812, 2015 2016 29. Saito K, Fujigaki S, Heyes MP, Shibata K, Takemura M, Fujii H, 12. Kimura T, Yasuda K, Yamamoto R, Soga T, Rakugi H, Hayashi T, Wada H, Noma A, Seishima M: Mechanism of increases in Isaka Y: Identification of biomarkers for development of end-stage L-kynurenine and quinolinic acid in renal insufficiency. Am kidney disease in chronic kidney disease by metabolomic pro- J Physiol Renal Physiol 279: F565–F572, 2000 filing. Sci Rep 6: 26138, 2016 30. Chen Y, Stankovic R, Cullen KM, Meininger V, Garner B, Coggan 13. Klahr S, Levey AS, Beck GJ, Caggiula AW, Hunsicker L, Kusek JW, S, Grant R, Brew BJ, Guillemin GJ: The kynurenine pathway Striker G; Modification of Diet in Renal Disease Study Group: The and inflammation in amyotrophic lateral sclerosis. Neurotox Res effects of dietary protein restriction and blood-pressure control on 18: 132–142, 2010 the progression of chronic renal disease. N Engl J Med 330: 877– 31. Suchy-Dicey AM, Laha T,Hoofnagle A, Newitt R, Sirich TL, Meyer 884, 1994 TW, Thummel KE, Yanez ND, Himmelfarb J, Weiss NS, 14. Levey AS, AdlerS, Caggiula AW,England BK, Greene T,Hunsicker Kestenbaum BR: Tubular secretion in CKD. J Am Soc Nephrol 27: LG, Kusek JW, Rogers NL, Teschan PE: Effects of dietary protein 2148–2155, 2016 restriction on the progression of advanced renal disease in the 32. Bahn A, Ljubojevic M, Lorenz H, Schultz C, Ghebremedhin E, Modification of Diet in Renal Disease Study. Am J Kidney Dis 27: Ugele B, Sabolic I, Burckhardt G, Hagos Y: Murine renal organic 652–663, 1996 anion transporters mOAT1 and mOAT3 facilitate the transport of 15. Evans AM, DeHaven CD, Barrett T, Mitchell M, Milgram E: In- neuroactive tryptophan metabolites. Am J Physiol Cell Physiol tegrated, nontargeted ultrahigh performance liquid chromatogra- 289: C1075–C1084, 2005 phy/electrospray ionization tandem mass spectrometry platform for 33. Deguchi T, Kusuhara H, Takadate A, Endou H, Otagiri M, the identification and relative quantification of the small-molecule Sugiyama Y: Characterization of uremic toxin transport by or- complement of biological systems. Anal Chem 81: 6656–6667, ganic anion transporters in the kidney. Kidney Int 65: 162–174, 2009 2004 16. OhtaT,MasutomiN, Tsutsui N, SakairiT,Mitchell M, Milburn MV, 34. Jansen J, Fedecostante M, Wilmer MJ, Peters JG, Kreuser UM, Ryals JA, Beebe KD, Guo L: Untargeted metabolomic profiling as van den Broek PH, Mensink RA, Boltje TJ, Stamatialis D, Wetzels an evaluative tool of fenofibrate-induced toxicology in Fischer JF, van den Heuvel LP, Hoenderop JG, Masereeuw R: 344 male rats. Toxicol Pathol 37: 521–535, 2009 Bioengineered kidney tubules efficiently excrete uremic toxins. 17. McMahon GM, Hwang SJ, Clish CB, Tin A, Yang Q, Larson MG, Sci Rep 6: 26715, 2016 Rhee EP,Li M, Levy D, O’Donnell CJ, Coresh J, Young JH, Gerszten 35. Torres VE, King BF, Chapman AB, Brummer ME, Bae KT, Glockner RE, Fox CS; CKDGen Consortium: Urinary metabolites along with JF, Arya K, Risk D, Felmlee JP, Grantham JJ, Guay-Woodford LM, common and rare genetic variations are associated with incident Bennett WM, Klahr S, Meyers CM, Zhang X, Thompson PA, Miller chronic kidney disease. Kidney Int 91: 1426–1435, 2017 JP; Consortium for Radiologic Imaging Studies of Polycystic 18. Youden WJ: Index for rating diagnostic tests. Cancer 3: 32–35, Kidney Disease (CRISP): Magnetic resonance measurements of 1950 renal blood flow and disease progression in autosomal dominant 19. Hodson L, Skeaff CM, Fielding BA: Fatty acid composition of polycystic kidney disease. Clin J Am Soc Nephrol 2: 112–120, adipose tissue and blood in humans and its use as a biomarker of 2007 dietary intake. Prog Lipid Res 47: 348–380, 2008 36. Niwa T, Takeda N, Yoshizumi H: RNA metabolism in uremic 20. Wanders RJ, Komen J, Kemp S: Fatty acid omega-oxidation as a patients: Accumulation of modified ribonucleosides in uremic rescue pathway for fatty acid oxidation disorders in humans. FEBS serum. Technical note. Kidney Int 53: 1801–1806, 1998 J 278: 182–194, 2011 37. Peterson JC, Adler S, Burkart JM, Greene T, Hebert LA, Hunsicker 21. Li SY, Susztak K: Fat burning problem in cystic kidneys: An LG, King AJ, Klahr S, Massry SG, Seifter JL: Blood pressure emerging common mechanism of chronic kidney disease. control, proteinuria, and the progression of renal disease. The EBioMedicine 5: 22–23, 2016 Modification of Diet in Renal Disease Study. Ann Intern Med 123: 22. Herman-Edelstein M, Scherzer P, Tobar A, Levi M, Gafter U: 754–762, 1995 Altered renal lipid metabolism and renal lipid accumulation in 38. Yekutieli D, Benjamini Y: Resampling-based false discovery rate human diabetic nephropathy. J Lipid Res 55: 561–572, 2014 controlling multiple test procedures for correlated test statistics. 23. KangHM,AhnSH,ChoiP,KoYA,HanSH,ChingaF,ParkAS,TaoJ, J Stat Plan Inference 82: 171–196, 1999 Sharma K, Pullman J, Bottinger EP, Goldberg IJ, Susztak K: Defective fatty acid oxidation in renal tubular epithelial cells has a key role in kidney fibrosis development. Nat Med 21: 37–46, 2015 Received: March 7, 2017 Accepted: July 10, 2017 24. Meyer C, Nadkarni V, Stumvoll M, Gerich J: Human kidney free fatty acid and glucose uptake: Evidence for a renal glucose-fatty M.E.G. and A.T. are cofirst authors. acid cycle. Am J Physiol 273: E650–E654, 1997 25. Vanholder R, Baurmeister U, Brunet P, Cohen G, Glorieux G, Published online ahead of print. Publication date available at www. Jankowski J; European Uremic Toxin Work Group: A bench to cjasn.org. bedsideviewofuremictoxins.JAmSocNephrol19:863–870,2008 26. Rhee EP, Souza A, Farrell L, Pollak MR, Lewis GD, Steele DJ, See related editorial, “Metabolomics and Kidney Precision Thadhani R, Clish CB, Greka A, Gerszten RE: Metabolite profiling ” – identifies markers of uremia. JAmSocNephrol21: 1041–1051, Medicine, on pages 1726 1727. 2010 27. Lees HJ, Swann JR, Wilson ID, Nicholson JK, Holmes E: Hippu- Thisarticlecontainssupplementalmaterialonlineathttp://cjasn. rate: The natural history of a mammalian-microbial cometabolite. asnjournals.org/lookup/suppl/doi:10.2215/CJN.02560317/-/ J Proteome Res 12: 1527–1546, 2013 DCSupplemental. Supplemental material is neither peer-reviewed nor thoroughly edited by CJASN. The authors alone are responsible for the accuracy and presentation of the material.

Supplement to:

Metabolomic Alterations Associated with Cause of Chronic Kidney Disease

Morgan E. Grams, MD PhD (1,2,3);* Adrienne Tin, PhD (2,3);* Casey M. Rebholz, PhD, MPH, MS (2,3); Tariq Shafi, MBBS MHS (1,2, 3); Anna Köttgen, MD MPH (2,4); Ronald D. Perrone, MD (5); Mark J. Sarnak, MD MS (5); Lesley A. Inker, MD MS (5); Andrew S. Levey, MD (5); Josef Coresh, MD PhD (2,3)

*co‐first authors

1) Division of Nephrology, Department of Medicine, Johns Hopkins University, Baltimore MD 2) Department of Epidemiology, Johns Hopkins University Bloomberg School of Public Health, Baltimore MD 3) Welch Center for Prevention, Epidemiology, and Clinical Research, Johns Hopkins University, Baltimore MD 4) Division of Genetic Epidemiology, Faculty of Medicine and Medical Center – University of Freiburg, Freiburg, Germany 5) Division of Nephrology, Tufts Medical Center, Boston, MA

Corresponding author: Morgan E. Grams, MD PhD 2024 E. Monument, Rm 2‐638 Baltimore, MD 21205 Phone: 443‐287‐1827 Fax: 410‐955‐0485 Email: [email protected]

Supplemental material is neither peer-reviewed nor thoroughly edited by CJASN. The authors alone are responsible for the accuracy and presentation of the material.

Supplemental Table 1. Expert‐adjudicated cause of chronic kidney disease cross‐tabulated with clinical center‐reported cause of disease in the Modification of Diet in Renal Disease Study participants included in the study population (N=589)

Expert‐Adjudicated Cause of Disease CKD of Other Etiology Self‐reported cause of disease PKD Glomerular Interstitial VU Reflux/ Hypertensive One Disease* Unknown Nephritis Obstruction Nephrosclerosis kidney

Polycystic kidney disease 167 0 0 0 0 0 0 Heredity nephritis 0 0 0 0 0 0 0 Analgesic nephropathy 0 0 3 1 0 0 3 Pyelonephritis 0 0 11 0 0 0 10 Other interstitial nephritis 0 3 12 0 0 0 15 Obstructive uropathy‐acquired 0 0 0 1 0 0 0 Obtructive uropathy‐congenital 0 0 0 4 0 0 0 Vesico ureteral reflux 0 0 0 13 0 0 0 Urinary tract stones 0 0 0 6 0 0 0 Hypertensive nephrosclerosis 0 2 0 1 39 0 69 Diabetic nephropathy 0 0 0 0 0 0 0 Renal artery stenosis 0 0 0 0 1 0 0 Membranous nephropathy 0 15 0 0 0 0 0 Focal sclerosis 0 44 0 0 0 0 0 Membranoproliferative glomerulonephritis 0 16 0 0 0 0 0 Mesangial proliferative glomerulonephritis 0 6 0 0 0 0 0 Chronic renal failure with proteinuria 0 5 0 0 0 0 0 Nephrotic syndrome without biopsy 0 2 0 0 0 0 0 One kidney 0 0 0 0 0 17 0 IgA nephropathy 0 29 0 0 0 0 1 Other glomerulonephritis 0 20 1 0 0 0 13 Other 0 4 4 0 0 0 12 Missing 0 2 0 0 0 0 37 Total 167 148 31 26 40 17 160

Supplemental material is neither peer-reviewed nor thoroughly edited by CJASN. The authors alone are responsible for the accuracy and presentation of the material.

Supplemental Table 2. Characteristics of Modification of Diet in Renal Disease Study participants, by sub‐study and cause of chronic kidney disease Discovery (Study B) Replication (Study A)

Polycystic Glomerular Other Cause P‐Value Polycystic Glomerular Other Cause P‐Value kidney disease disease kidney disease disease N 48 46 72 119 102 202 Age, mean (SD) 50 (10) 48 (13) 54 (12) 0.04 48 (10) 48 (13) 56 (11) <0.001 Race, % (n) 0.64 0.12 Black 4.2 (2) 6.5 (3) 5.6 (4) 5 (6) 7.8 (8) 9.4 (19) Hispanic/other 6.2 (3) 15.2 (7) 9.7 (7) 4.2 (5) 10.8 (11) 4.5 (9) White 89.6 (43) 78.3 (36) 84.7 (61) 90.8 (108) 81.4 (83) 86.1 (174) Male, % (n) 60.4 (29) 67.4 (31) 58.3 (42) 0.59 58.8 (70) 71.6 (73) 57.4 (116) 0.04 Diet assignment, % (n) 0.73 0.40 Very low protein 43.8 (21) 43.5 (20) 50.0 (36) Low protein 56.2 (27) 56.5 (26) 50.0 (36) 45.4 (54) 47.1 (48) 52.5 (106) Usual protein 0 0 0 54.6 (65) 52.9 (54) 47.5 (96) Moderate blood pressure 45.8 (22) 41.3 (19) 51.4 (37) 0.53 45.4 (54) 48 (49) 50.5 (102) 0.66 assignment, % (n) Systolic blood pressure 132 (17) 132 (13) 132 (18) 0.78 127 (16) 131 (18) 129 (18) 0.27 Diastolic blood pressure 82 (10) 81 (10) 77 (10) 0.13 81 (9) 80 (9) 78 (9) 0.05 BMI 25.6 (3.8) 24.7 (3.2) 25.8 (4.2) 0.12 26 (4) 28 (4) 28 (4) 0.27 Estimated protein intake, 0.6 (0.23) 0.6 (0.16) 0.59 (0.17) 0.68 0.96 (0.27) 0.99 (0.29) 0.92 (0.27) 0.08 g/kg/day, mean (SD) GFR 12.9 (3.9) 14.7 (4.7) 16.2 (4.8) 0.45 32 (10) 35 (12) 37 (11) 0.59 Urine protein, g/day 0.2 (0.1, 0.4) 1.1 (0.4, 1.9) 0.5 (0.1, 1.3) <0.001 0.1 (0, 0.2) 1.5 (0.4, 2.4) 0.1 (0, 0.4) <0.001 Kidney biopsy, % (n)* 2.1 (1) 82.2 (37) 17.5 (11) <0.001 3.4 (4) 93.1 (94) 11.8 (21) <0.001 History of diabetes, % (n) 2.1 (1) 0 (0) 6.9 (5) 0.18 0.8 (1) 4.9 (5) 3.0 (6) 0.18 HbA1c, median (IQR)* 5.7 (5.4‐5.9) 5.7 (5.3‐5.9) 5.6 (5.3‐5.9) 0.42 5.6 (5.3‐5.9) 5.6 (5.2‐6.0) 5.6 (5.3‐6.1) 0.60 Albumin, median (IQR) 4.1 (4‐4.3) 4.2 (3.9‐4.4) 4.1 (3.9‐4.3) 0.96 4.2 (4‐4.4) 4.0 (3.8‐4.2) 4.1 (3.9‐4.3) <0.001 *Sample size for kidney biopsy N=156 and 398 in Study B and Study A, respectively. Sample size for HbA1c N=161 and 416, respectively.

Supplemental material is neither peer-reviewed nor thoroughly edited by CJASN. The authors alone are responsible for the accuracy and presentation of the material.

Supplemental Table 3. Metabolite identifiers of replicated metabolite hits in PubChem, ChemSpider, Kegg, and HDMB.

Metabolite Super Sub Pathway Mass PubChem ChemSpider Kegg HDMB ID Pathway 16‐hydroxypalmitate Lipid Fatty Acid, Monohydroxy 271.2279 10466 10034 C18218 HMDB06294 Phenylalanine and Tyrosine homovanillate sulfate Amino Acid Metabolism 261.0074 29981063 21896746 HMDB11719 kynurenate Amino Acid Tryptophan Metabolism 188.0353 3845 3712 C01717 HMDB00715 Purine Metabolism, Guanine N2,N2‐dimethylguanosine Nucleotide containing 312.1303 92919 83878 HMDB04824 hippurate Xenobiotics Benzoate Metabolism 178.051 464 451 C01586 HMDB00714

Supplemental material is neither peer-reviewed nor thoroughly edited by CJASN. The authors alone are responsible for the accuracy and presentation of the material.

Supplemental Table 4. P‐values of metabolites significantly associated with cause of chronic kidney disease in the adjusted models, by sub‐study in the Modification of Diet in Renal Disease Study

Discovery (Study B) Replication (Study A) Metabolite P‐value P‐value P‐value Likelihood P‐value P‐value P‐value Likelihood (Pathway) (Polycystic (Glomerular (Polycystic Ratio Test p‐ (Polycystic (Glomerular (Polycystic Ratio Test Kidney Disease vs. Kidney value Kidney Disease vs. Kidney p‐value Disease vs. Other) Disease vs. Disease vs. Other) Disease vs Other) Glomerular Other) Glomerular Disease) Disease) 16‐hydroxypalmitate 3.5E‐05 7.7E‐01 7.4E‐05 1.54E‐05 1.24E‐20 2.9E‐01 1.1E‐10 1.9E‐22 (Fatty Acid, Monohydroxy) Homocitrulline (Urea cycle, Arginine and Proline 5.0E‐02 2.1E‐03 1.2E‐05 2.74E‐05 1.17E‐01 1.3E‐01 1.1E‐02 3.7E‐02 Metabolism) 1,5‐anhydroglucitol (Glycolysis, Gluconeogenesis, and Pyruvate 6.2E‐05 7.7E‐01 7.8E‐04 8.82E‐05 3.07E‐02 8.8E‐01 1.2E‐01 8.5E‐02 Metabolism) Kynurenate 3.5E‐02 1.7E‐02 8.9E‐05 2.94E‐04 1.70E‐04 3.6E‐01 2.1E‐04 8.6E‐05 (Tryptophan Metabolism) Homovanillate sulfate (Phenylalanine and Tyrosine 5.5E‐04 6.5E‐01 4.9E‐04 3.36E‐04 9.37E‐08 9.9E‐01 3.6E‐05 1.1E‐07 Metabolism) N2,N2‐dimethylguanosine (Purine Metabolism, Guanine 8.9E‐04 5.3E‐01 4.4E‐04 4.15E‐04 3.14E‐04 5.3E‐01 8.4E‐04 3.0E‐04 containing) Hippurate 3.1E‐03 3.0E‐01 4.2E‐04 7.74E‐04 1.81E‐03 9.2E‐01 2.0E‐02 5.1E‐03 (Benzoate Metabolism) Covariates: race, age, sex, log(GFR), BMI, study diet and BP assignment, estimate protein intake, ACE inhibitor, beta blocker, log(proteinuria) Statistical significance determined by p< 7.9e‐4 (=0.05/63 principal components) in the discovery cohort and p<0.007 (=0.05/7) in the replication cohort. Bold reflects metabolite associations that were replicated in Study A. Abbreviations: Other, other cause of CKD.

Supplemental material is neither peer-reviewed nor thoroughly edited by CJASN. The authors alone are responsible for the accuracy and presentation of the material.

Supplemental Table 5. Additional measures of discrimination comparing a model with clinical variables only to one that uses both clinical variables and the five replicated metabolites in the Modification of Diet in Renal Disease Study.

Clinical variables only Polycystic kidney Glomerular disease Other cause of kidney disease vs. no vs. no glomerular disease vs. non‐other polycystic kidney disease cause of kidney disease disease

Youden cutpoint 0.34 0.23 0.45

Sensitivity 0.72 0.80 0.72

Specificity 0.81 0.73 0.67

Clinical variables + metabolites

Youden cutpoint 0.35 0.22 0.41

Sensitivity 0.89 0.84 0.82

Specificity 0.84 0.73 0.67

Clinical variables + metabolites vs. clinical variables only Continuous Net Reclassification Index* 0.98 (0.83, 1.13) 0.61 (0.43, 0.78) 0.63 (0.48, 0.79) Continuous Integrated Discrimination Index* 0.23 (0.19, 0.26) 0.04 (0.02, 0.06) 0.12 (0.10, 0.15)

*All indices with p<0.001. Net reclassification index reports the proportion of participants who correctly moved up added to the proportion of participants correctly moved down (maximum value: 2). Integrated discrimination index reports the sum of the difference in predicted probabilities, or the sum of the improvement in sensitivity and specificity.

Supplemental material is neither peer-reviewed nor thoroughly edited by CJASN. The authors alone are responsible for the accuracy and presentation of the material.

Supplemental Figure 1. Smile plots of association with cause of CKD vs. statistical significance for 687 tested metabolites in the discovery cohort (Study B) in the Modification of Diet in Renal Disease Study. Red denotes the five replicated metabolites, and the % denotes the percent of metabolites that were increased or decreased.

Supplemental material is neither peer-reviewed nor thoroughly edited by CJASN. The authors alone are responsible for the accuracy and presentation of the material.

Supplemental Appendix 1. Information on Metabolomic Profiling Procedures

Samples were sent to Metabolon, Inc. on dry ice. On receipt, samples were inventoried and immediately stored at ‐80oC until processing. Samples were processed by first adding several recovery standards, then using methanol to precipitate proteins. The extract was divided into five fractions: two for separate reverse phase ultraperformance liquid chromatography tandem mass‐spectrometry (RP/UPLC‐MS/MS) with a positive ion mode electrospray ionization (ESI), one for RP/UPLC‐MS/MS with negative ion mode ESI, one for hydrophilic interaction ultra‐performance liquid chromatography (HILIC‐UPLC‐MS/MS) with negative ion mode ESI, and one sample for back‐up. These platforms are designed to provide complementary information regarding metabolites. Controls were analyzed concomitantly with the experimental samples, including a pooled matrix sample serving as a technical replicate, extracted water samples for negative controls, and certain QC standards that do not interfere with endogenous compounds. The latter were added to each analyzed sample to provide monitoring for instruments and chromatographic alignment. Instrument variability was assessed as the median relative standard deviation for the QC standards (5%) and process variability was assessed as the median relative standard deviation for the endogenous metabolites in the pooled matrix samples (11%). Both values met Metabolon’s internal acceptance criteria. After analyzing via the four methods, each of which used a Waters ACQUITY ultra‐performance liquid chromatography (UPLC) and a Thermo Scientific Q‐Exactive high resolution/accurate mass spectrometer interfaced with a heated electrospray ionization (HESI‐II) source and Orbitrap mass analyzer operated at 35,000 mass resolution, data were extracted and processed using internal hardware and software. Peaks were identified through comparison to the Metabolon library, which catalogs purified, authenticated standards (currently >3300 compounds) and recurrent unknown entities. Experimental peaks were compared to the library with respect to retention time/index, mass to charge ratio, and chromatographic data; only compounds that matched on all three criteria are identified. The area under the curve was then used to quantify peaks, with correction for run‐day blocks, since samples required more than one day of analysis. For the current analysis, 1193 biochemicals were matched with compounds in the Metabolon library, including 766 known biochemicals, authenticated using purified standards, and 427 recurrent unknown biochemicals. Biochemicals were classified into 79 pathways (see table, next page). Correlation between the metabolite creatinine and in‐study, clinically‐measured creatinine was 0.96. Per protocol, 20 blind duplicate pairs were analyzed. The median metabolite correlation between duplicates was 0.91, and 72% of all metabolites had a correlation >0.80. Correlations for hippurate, 16‐hydroxypalmitate, kynurenate, homovanillate sulfate, and N2,N2‐dimethylguanosine were 0.97, 0.58, 0.98, 0.93, and 0.92, respectively.

Supplemental material is neither peer-reviewed nor thoroughly edited by CJASN. The authors alone are responsible for the accuracy and presentation of the material.

Supplemental Appendix 2. Number of metabolites within 79 identified pathways.

Number of Number of Known, Non‐Drug Pathways Metabolites Known, Non‐Drug Pathways Metabolites Advanced Glycation End‐product 1 Leucine, Isoleucine and Valine Metabolism 28 Alanine and Aspartate Metabolism 6 Long Chain Fatty Acid 14 Aminosugar Metabolism 4 Lysine Metabolism 12 Ascorbate and Aldarate Metabolism 3 Lysolipid 25 Bacterial/Fungal 1 Lysoplasmalogen 4 Benzoate Metabolism 20 Medium Chain Fatty Acid 5 Carnitine Metabolism 2 Methionine, Cysteine, SAM and Taurine 20 Metabolism Chemical 22 Mevalonate Metabolism 1 Creatine Metabolism 3 Monoacylglycerol 14 Diacylglycerol 4 Nicotinate and Nicotinamide Metabolism 7 Dipeptide 9 Oxidative Phosphorylation 1 Dipeptide Derivative 1 Pantothenate and CoA Metabolism 1 Disaccharides and Oligosaccharides 1 Pentose Metabolism 4 Eicosanoid 3 Phenylalanine and Tyrosine Metabolism 36 Endocannabinoid 5 Phospholipid Metabolism 38 Fatty Acid Metabolism (Acyl Choline) 1 Plasmalogen 11 Fatty Acid Metabolism (also BCAA 3 Polyamine Metabolism 5 Metabolism) Fatty Acid Metabolism(Acyl Carnitine) 18 Polypeptide 1 Fatty Acid Metabolism(Acyl Glycine) 2 Polyunsaturated Fatty Acid (n3 and n6) 12 Fatty Acid Synthesis 3 Primary Bile Acid Metabolism 9 Fatty Acid, Amino 2 Purine Metabolism, (Hypo)Xanthine/Inosine 7 containing Fatty Acid, Branched 2 Purine Metabolism, Adenine containing 6 Fatty Acid, Dicarboxylate 16 Purine Metabolism, Guanine containing 2 Fatty Acid, Dihydroxy 2 Pyrimidine Metabolism, Cytidine containing 2 Fatty Acid, Monohydroxy 13 Pyrimidine Metabolism, Orotate containing 3 Fibrinogen Cleavage Peptide 1 Pyrimidine Metabolism, Thymine containing 2 Food Component/Plant 35 Pyrimidine Metabolism, Uracil containing 7 Fructose, Mannose and Galactose 3 Secondary Bile Acid Metabolism 17 Metabolism Gamma‐glutamyl Amino Acid 15 Sphingolipid Metabolism 26 Glutamate Metabolism 6 Steroid 34 Glutathione Metabolism 3 Sterol 3 Glycerolipid Metabolism 2 TCA Cycle 9 Glycine, Serine and Threonine Metabolism 9 Tobacco Metabolite 4 Glycogen Metabolism 2 Tocopherol Metabolism 6 Glycolysis, Gluconeogenesis, and Pyruvate 5 Tryptophan Metabolism 22 Metabolism Guanidino and Acetamido Metabolism 3 Urea cycle; Arginine and Proline Metabolism 18 Hemoglobin and Porphyrin Metabolism 6 Vitamin A Metabolism 1 Histidine Metabolism 13 Vitamin B6 Metabolism 2 Inositol Metabolism 2 Xanthine Metabolism 15 Ketone Bodies 1