University of Groningen Novel biomarker panels in diabetic ... · There are however, few prospective studies of multiple biomarkers specifying type 2 diabetes as the cause of renal

University of Groningen

Novel biomarker panels in diabetic kidney diseasePena, Michelle Jo

IMPORTANT NOTE: You are advised to consult the publisher's version (publisher's PDF) if you wish to cite fromit. Please check the document version below.

Document VersionPublisher's PDF, also known as Version of record

Publication date:2015

Link to publication in University of Groningen/UMCG research database

Citation for published version (APA):Pena, M. J. (2015). Novel biomarker panels in diabetic kidney disease: Predicting disease progression andresponse to therapy, and monitoring drug effect. [Groningen]: University of Groningen.

CopyrightOther than for strictly personal use, it is not permitted to download or to forward/distribute the text or part of it without the consent of theauthor(s) and/or copyright holder(s), unless the work is under an open content license (like Creative Commons).

Take-down policyIf you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediatelyand investigate your claim.

Downloaded from the University of Groningen/UMCG research database (Pure): http://www.rug.nl/research/portal. For technical reasons thenumber of authors shown on this cover page is limited to 10 maximum.

Download date: 25-08-2020

https://www.rug.nl/research/portal/en/publications/novel-biomarker-panels-in-diabetic-kidney-disease(ad588591-3f1b-40a6-ad2e-85d882fd87e6).html

https://www.rug.nl/research/portal/en/persons/michelle-pena(6e7fd8a0-6aff-46dd-bb4e-11da70b884b5).html



Novel biomarker panels in diabetic kidney disease

Predicting disease progression and response to therapy, and monitoring drug effect

Michelle J. Pena

Novel biomarker panels in diabetic kidney disease

Predicting disease progression and response to therapy, and monitoring drug effect

PhD thesis

to obtain the degree of PhD at theUniversity of Groningenon the authority of the

Rector Magnificus Prof. E. Sterkenand in accordance with

the decision by the College of Deans.

This thesis will be defended in public on

Monday 21 December 2015 at 09.00 hours

by

Michelle Jo Pena

born on 23 October 1979 in California, United States of America

The research described in this thesis was supported by the European Community’s Seventh Framework Programme under grant agreement no. HEALTH–F2–2009–241544 (SysKID consortium).

Financial support for the printing of this thesis was kindly provided by the University of Groningen, University Medical Center Groningen, and Graduate School for Drug Exploration (GUIDE).

© Michelle Pena, Groningen 2015

Cover design: Marije Esselink, Hello Handsome (www.hellohandsome.nl) Molecular model of diabetic kidney disease adapted from Heinzel et al. Front Cell Dev Biol 2014.Layout: Tara Kinneging, Persoonlijk ProefschriftPrinted by: Ipskamp Drukkers, Enschede

ISBN: 978-90-367-8316-3 (printed version) 978-90-367-8315-6 (digital version)

Copyright by Michelle Pena, Groningen, the Netherlands. All rightsreserved. No part of this publication may be reproduced, stored on a retrieval system, or transmitted in any form or by any means, without permission of the author.

ParanymphsGiedrė GefenaitėSara Roscioni

Supervisors Prof. D. de Zeeuw Prof. H.J. Lambers Heerspink Assessment Committee Prof. B.H.R. Wolffenbuttel Prof. M. Kretzler Prof. G.J. Mayer

TABLE OF CONTENTS

Chapter 1 Introduction and aims 9

Part 1. Novel biomarker panels for predicting disease progression

Chapter 2 A panel of novel biomarkers representing different disease pathways improves prediction of renal function decline in type 2 diabetes 23

Chapter 3 Plasma proteomic classifiers improve risk prediction for renal disease in patients with hypertension or type 2 diabetes 53

Chapter 4 Urine and plasma metabolites predict the development of diabetic nephropathy in patients with type 2 diabetes mellitus 83

Part 2. Novel biomarker panels for predicting response to therapy and monitoring drug effect

Chapter 5 Serum metabolites predict response to angiotensin II receptor blocker therapy in diabetes mellitus 111

Chapter 6 The beneficial impact of atrasentan on a urinary metabolite panel previously associated with renal function decline 131

Chapter 7 Summary and future perspectives 153

Nederlandse samenvatting en toekomstperspectief 165

Acknowledgements 177

Curriculum Vitae 181

CHAPTER 1Introduction and aims

Modified from

Prognostic clinical and molecular biomarkers of renal disease in type 2 diabetes

Nephrol Dial Transplant. 2015; 30 Suppl 4: iv86-iv95

1110

Introduction and aimsChapter 1

1

INTRODUCTION

There is an urgency to better identify patients with type 2 diabetes mellitus at early stages of chronic kidney disease (CKD) [1]. Approximately 387 million adults around the world are currently living with diabetes, and due to a relentless increase in the incidence of type 2 diabetes, this estimate is projected to rise to 592 million by 2035 (Figure 1) [2]. Of those patients with type 2 diabetes, 20-40% will ultimately develop diabetic kidney disease (DKD). In addition, type 2 diabetes results in a high cardiovascular morbidity and mortality

and a decrease in the patients’ health-related quality of life.

Figure 1. Number of people by region with diabetes in 2013 and projected number of cases in 2035. Adapted from the IDF Diabetes Atlas 2013 [2].

DKD, traditionally referred to as diabetic nephropathy, is based in part on the finding of elevated urinary albumin excretion (UAE), progressive decline in glomerular filtration rate (GFR), an increase in systemic blood pressure, and a high risk of kidney failure [3]. DKD is also associated metabolic disturbances. DKD is now the leading cause of end-stage renal disease (ESRD), and accounts for approximately 50% of dialysis and renal transplantation in developed countries [4]. There could be a sharp rise in the prevalence of ESRD over

the next few decades [5], driven by population ageing and the increasing prevalence of diabetes (Figure 2). The costs for renal and cardiovascular related complications are extraordinarily high: costs for renal replacement therapies alone account for 3 to 5% of the total European Union (EU) health care budget and even more in other countries. The United States Renal Data System has reported that for patients aged 65 years and older with both CKD and diabetes, the total Medicare costs have increased more than 11 times in the past decade [6]. Additionally, in a group of patients with type 2 diabetes with early stage CKD in the United States, the 5-year healthcare costs were twice as high among those who progressed to a higher stage of CKD compared to who did not progress, and for patients with stage 3-4 CKD, the costs were more than threefold higher [7]. Thus, there is a strong economic and social imperative to improve the outcomes of type 2 diabetes. Early identification of patients with type 2 diabetes at risk of renal disease can lead to early intervention aimed at reducing the incidence of DKD and ultimately ESRD. There are many stakeholders that can benefit from early identification, number one being the patients themselves, their families, and society.

Figure 2. Estimated number of patients undergoing renal replacement therapy from 2010 to 2030 worldwide (A) and by region (B). 95% CIs shown as error bars. Adapted from Liyanage et al. Lancet 2015 [5].

A biomarker is “a characteristic that is objectively measured and evaluated as an indicator of normal biological processes, pathogenic processes, or pharmacologic responses to a therapeutic intervention” [8]. Estimated glomerular filtration rate (eGFR) and detection of albumin in urine (albuminuria) are the classical guideline-endorsed biomarkers for the classification of CKD [9]. These biomarkers are strong predictors of renal disease progression as well as cardiovascular disease and mortality. Reduction in eGFR and detection of microalbuminuria are considered the first clinical signs of renal disease. Reduced eGFR is the consequence of compromised kidney function and substantial loss

1312


1

and destruction of the glomeruli, and the presence of microalbuminuria already indicates a permeable glomerular basement membrane. Both point to possibly irreversible damage to the kidney. However, renal damage at early disease stages rarely shows clinical characteristics. Therefore, on the everyday clinical level, early stage diagnosis and tailored treatment of DKD are still inadequate. In order to improve patient outcomes and reduce associated health-care costs, timely detection and prevention of progression of renal disease are needed.

Novel biomarker panels can improve identification of renal disease at its early stages. The search for novel biomarker panels to improve the early identification of patients at high-risk for renal disease has been the priority of many researchers for many years. Novel biomarker panels can also have different roles for diagnosis, prognosis, and monitoring by improving risk stratification, help increase our understanding of renal disease pathophysiology, or provide insight into novel therapeutic targets.

Novel biomarker panels as predictors of renal diseaseThe past decade has produced a large number of papers published on novel biomarkers for renal disease. Many single proteins have been proposed as biomarkers of renal disease in type 2 diabetes and are measured by immunological assays [10-15]. Typically, these biomarkers capture one specific mechanism of disease such as inflammation, fibrosis, or tubular damage. These studies highlight the relevance of single disease mechanisms and provide important insight into the disease etiology. However, type 2 diabetes is a heterogeneous disease involving multiple pathophysiological mechanisms [16]. In theory, the measurement of several biomarkers simultaneously (a multi-marker approach) should improve risk stratification of patients at high risk for adverse events since it is unlikely that a single biomarker may possess useful diagnostic and prognostic power to fully capture the risk of renal disease in type 2 diabetes. Single biomarkers constantly face problems with individual, biological, and analytical variability.

To date, no one, single protein biomarker has been shown to significantly outperform albuminuria or eGFR as predictors of disease progression in longitudinal interventional studies. Alternatively, a panel of clearly defined biomarkers may provide a more robust and reproducible tool as a panel may tolerate changes in single biomarkers without jeopardizing their diagnostic precision and may offer a more realistic picture of disease and its underlying mechanisms. Multiple biomarker approaches are becoming more and more common in literature, though still not as prominent as single biomarker studies. There are however, few prospective studies of multiple biomarkers specifying type 2 diabetes as the cause of renal disease. Some studies consider many biomarkers, but test each biomarker one by one, instead of a combined biomarker panel approach [10,11,15,17]. There are only a few studies in literature that focus on biomarker panels where two or more novel biomarkers are tested in combination to predict renal disease progression [18-21]. Measuring multiple biomarkers at once is becoming more and more

realistic for clinical practice as advancing laboratory techniques with multiplex assays or mass-spectrometry technologies allow the simultaneous measurement of large number of biomarkers with minimal sample volume.

Multiple biomarker panels – Omics platformsThe measurement of multiple biological molecules has advanced significantly over the past years with the introduction of high-throughput omics screening platforms. An omics-based test is defined as an assay composed of, or derived from, multiple molecular measurements and interpreted by a fully specified computational model to produce a clinically meaningful result. Such assays can measure a full spectrum of peptides or metabolites in a short amount of time [22]. The measurement of peptides and metabolites, known as proteomics and metabolomics, have emerged as strong tools in biomarker discovery [22,23].

Figure 3. The conceptual relationship of the genome, transcriptome, proteome, and metabolome.

Adapted from Gerszten & Wang Nature 2008 [23].

Proteomics permit the rapid assessment of components of the proteome, which is the

complete inventory of proteins (or peptides) present within a biological sample. Biological

samples, such as urine, plasma, or serum, can be systematically analyzed with the goal

of identifying, quantifying and discerning the function of all observable proteins [24]. In

particular, urinary proteomics has gained much attention as a tool for the identification

of diagnostic and prognostic biomarkers of renal diseases [25], and may represent an

important step forward in the non-invasive diagnosis of renal diseases. Blood-derived

proteomics studies are not as common as urine proteomics, a few reasons being that

there is large heterogeneity and spread in abundance of proteins in blood and high

exposure to proteolytic activity [26], which complicates the analysis of the blood proteome.

Metabolomics, i.e. the measurement of low-weight intermediate metabolites (<500Da) and

end-products of cellular functions in biological fluids has emerged as another potential

1514


1

tool to discover novel biomarkers for renal disease. The metabolome can be viewed as

the down-stream integration of biological information of the genome, transcriptome,

proteome, and overall enzymatic reactions of an individual [23], and therefore enables

the detection of short and long-term physiological or pathological changes occurring in chronic diseases. Omics-based approaches hold promise for new diagnostic tests, better understanding of pathogenesis, and evolution of a disease.

Novel biomarker panels for predicting response to therapyDespite guideline recommended therapy for reduction of hypertension and albuminuria, not all patients with diabetes respond well to first line therapy intervening in the renin-angiotensin-aldosterone system (RAAS) [27]. Furthermore, there is large intra- and inter-individual variability in response to RAAS inhibiting therapy [28]. Many patients still have significant residual proteinuria [29]. In addition, a proportion of patients experience off-target effects [30-32], which may contribute to progressive renal function loss. The reasons behind these individual differences in response to therapy are unknown, and may be related to differences in systemic vs. renal tissue-specific renin-angiotensin system activity [33], dietary sodium consumption [34], or difference in genetic make-up [35, 36], among other factors. One strategy to improve the current state-of-the-art treatment is to tailor drug therapy by using a complementary approach to attribute drug response variability to individual variability in underlying molecular mechanisms involved in the progression of disease. On one hand, the interplay of different processes such as inflammation, fibrosis, angiogenesis, or oxidative stress, appears to drive disease progression, but the individual contribution of each process varies. On the other hand, drugs address specific targets and thereby interfere in certain disease associated processes. At this level novel biomarker panels may help gain insight into which specific pathophysiological processes are involved in an individual followed by a rational assessment whether a specific drug’s mode of action indeed targets the relevant process. In this context, novel biomarker panels can be used to identify a group of patients more likely to beneficially respond to therapy. This may reduce this inter-individual variation in response to medication. However, studies evaluating whether novel biomarker panels can be used as predictors of response to therapy have only been marginally explored.

Novel biomarker panels for monitoring drug effectA third option for using novel biomarker panels are to use changes in biomarkers to monitor the effect of therapy. This is important because it allows one to make a better estimate of the drug effect after the individual is exposed to the drug for a short period of time. In addition, results of such studies may also provide insight into the mechanisms through which drugs exert renoprotective effects and yield novel biomarkers to monitor response to therapy in patients with type 2 diabetes and DKD. Studies evaluating whether changes in novel biomarker panels can be used as predictors of renal disease are limited in existing literature.

AIMS OF THIS THESIS

This thesis examines several different approaches of utilizing novel biomarker panels in diabetic kidney disease that can be used to predict disease progression, predict response to therapy, or monitor effects of therapeutic intervention.

Part 1. Novel biomarker panels for predicting disease progression

Part 1 begins by investigating the predictive ability of novel biomarker panels for the progression of renal disease in patients with type 2 diabetes. Chapter 2 evaluates the ability of a panel of novel, assay-based biomarkers representing different disease pathways to improve prediction of renal function decline in type 2 diabetes, and to assess their combined predictive performance of accelerated renal function decline. In Chapter 3, proteomic analysis is used to identify plasma peptides associated with transitioning in stage of albuminuria in hypertension or type 2 diabetes, and examines whether two classifiers, one for hypertension and another for type 2 diabetes, are able to predict the transition of stage of albuminuria. In Chapter 4, metabolomics is performed to investigate the predictive ability of urine and plasma metabolites for the progression of renal dysfunction in patients with type 2 diabetes, and tests whether the metabolites are specific to type 2 diabetes by assessing the metabolites in individuals with hypertension without type 2 diabetes.

Part 2. Novel biomarker panels for predicting response to therapy and monitoring drug effect

Part 2 examines novel biomarker panels for predicting response to therapy in diabetes mellitus and monitoring the effect of therapeutic intervention. Chapter 5 first discovers and then validates a serum metabolite classifier that predicts response in albuminuria to angiotensin receptor blocker (ARB) therapy in patients with diabetes mellitus. Chapter 5 further integrates the identified metabolites in a molecular process model capturing disease pathophysiology at the interface of drug mechanism of action to decipher the underlying molecular processes driving albuminuria response to ARB. Chapter 6 assesses the correlation between a previously discovered metabolomics signature of diabetic kidney disease and eGFR in patients with type 2 diabetes and nephropathy, and evaluates the effect of atrasentan on these urinary metabolites.

This thesis ends by discussing future perspectives for using novel biomarker panels to improve on the status quo of choosing drugs for treatment of DKD in patients with type 2 diabetes and as a strategy to guide personalized medicine.

1716


1

REFERENCES

1. Gregg EW, Li Y, Wang J, Burrows NR, Ali MK, Rolka D, Williams DE, Geiss L. Changes in

diabetes-related complications in the United States, 1990-2010. N Engl J Med. 2014 Apr

17;370(16):1514-23.

2. International Diabetes Federation. IDF Diabetes Atlas, 6th edn. Brussels, Belgium:

International Diabetes Federation, 2013. http://www.idf.org/diabetesatlas.

3. National Kidney Foundation. KDOQI Clinical Practice Guideline for Diabetes and CKD:

2012 update. Am J Kidney Dis. 2012 Nov;60(5):850-86.

4. Tuttle KR. Linking metabolism and immunology: diabetic nephropathy is an inflammatory

disease. J Am Soc Nephrol. 2005 Jun;16(6):1537-8.

5. Liyanage T, Ninomiya T, Jha V, Neal B, Patrice HM, Okpechi I, Zhao MH, Lv J, Garg

AX, Knight J, Rodgers A, Gallagher M, Kotwal S, Cass A, Perkovic V. Worldwide access

to treatment for end-stage kidney disease: a systematic review. Lancet. 2015 May

16;385(9981):1975-82.

6. U.S. Renal Data System, USRDS 2013 Annual Data Report: Atlas of Chronic Kidney

Disease and End-Stage Renal Disease in the United States, National Institutes of Health,

National Institute of Diabetes and Digestive and Kidney Diseases, Bethesda, MD, 2013.

7. Vupputuri S, Kimes TM, Calloway MO, Christian JB, Bruhn D, Martin AA, Nichols GA.

The economic burden of progressive chronic kidney disease among patients with type 2

diabetes. J Diabetes Complications. 2014 Jan-Feb;28(1):10-6.

8. Biomarkers Definitions Working Group. Biomarkers and surrogate endpoints: preferred

definitions and conceptual framework. Clin Pharmacol Ther. 2001 Mar;69(3):89-95.

9. Kidney Disease: Improving Global Outcomes (KDIGO) CKD Work Group. KDIGO clinical

practice guideline for the evaluation and management of chronic kidney disease. Kidney

Int Suppl. 2013; 3(1):1-150.

10. Niewczas MA, Gohda T, Skupien J, Smiles AM, Walker WH, Rosetti F, Cullere X, Eckfeldt

JH, Doria A, Mayadas TN, Warram JH, Krolewski AS. Circulating TNF receptors 1 and 2

predict ESRD in type 2 diabetes. J Am Soc Nephrol. 2012 Mar;23(3):507-15.

11. Tam FW, Riser BL, Meeran K, Rambow J, Pusey CD, Frankel AH. Urinary monocyte

chemoattractant protein-1 (MCP-1) and connective tissue growth factor (CCN2) as

prognostic markers for progression of diabetic nephropathy. Cytokine. 2009 Jul;47(1):37-42.

12. Persson F, Rathcke CN, Gall MA, Parving HH, Vestergaard H, Rossing P. High YKL-40

levels predict mortality in patients with type 2 diabetes. Diabetes Res Clin Pract. 2012

Apr;96(1):84-9.

13. Hellemons ME, Mazagova M, Gansevoort RT, Henning RH, de Zeeuw D, Bakker SJ,

Lambers-Heerspink HJ, Deelman LE.. Growth-differentiation factor 15 predicts worsening

of albuminuria in patients with type 2 diabetes. Diabetes Care. 2012 Nov;35(11):2340-6.

14. Conway BR, Manoharan D, Manoharan D, Jenks S, Dear JW, McLachlan S, Strachan MW,

Price JF. Measuring urinary tubular biomarkers in type 2 diabetes does not add prognostic

value beyond established risk factors. Kidney Int. 2012 Oct;82(7):812-8.

15. Fufaa GD, Weil EJ, Nelson RG, Hanson RL, Bonventre JV, Sabbisetti V, Waikar SS, Mifflin

TE, Zhang X, Xie D, Hsu CY, Feldman HI, Coresh J, Vasan RS, Kimmel PL, Liu KD; Chronic

Kidney Disease Biomarkers Consortium Investigators. Association of urinary KIM-1,

L-FABP, NAG and NGAL with incident end-stage renal disease and mortality in American

Indians with type 2 diabetes mellitus. Diabetologia. 2015 Jan;58(1):188-98.

16. Fechete R, Heinzel A, Perco P, Mönks K, Söllner J, Stelzer G, Eder S, Lancet D, Oberbauer

R, Mayer G, Mayer B.Mapping of molecular pathways, biomarkers and drug targets for

diabetic nephropathy. Proteomics Clin Appl. 2011 Jun;5(5-6):354-66.

17. Agarwal R, Duffin KL, Laska DA, Voelker JR, Breyer MD, Mitchell PG. A prospective study

of multiple protein biomarkers to predict progression in diabetic chronic kidney disease.

Nephrol Dial Transplant. 2014 Dec;29(12):2293-302.

18. Persson F, Rossing P, Hovind P, Stehouwer CD, Schalkwijk CG, Tarnow L, Parving HH.

Endothelial dysfunction and inflammation predict development of diabetic nephropathy

in the Irbesartan in Patients with Type 2 Diabetes and Microalbuminuria (IRMA 2) study.

Scand J Clin Lab Invest. 2008;68(8):731-8.

19. Desai AS, Toto R, Jarolim P, Uno H, Eckardt KU, Kewalramani R, Levey AS, Lewis EF,

McMurray JJ, Parving HH, Solomon SD, Pfeffer MA. Association between cardiac

biomarkers and the development of ESRD in patients with type 2 diabetes mellitus,

anemia, and CKD. Am J Kidney Dis. 2011 Nov;58(5):717-28.

20. Wong MG, Perkovic V, Woodward M, Chalmers J, Li Q, Hillis GS, Yaghobian Azari

D, Jun M, Poulter N, Hamet P, Williams B, Neal B, Mancia G, Cooper M, Pollock CA.

Circulating bone morphogenetic protein-7 and transforming growth factor-beta1 are

better predictors of renal end points in patients with type 2 diabetes mellitus. Kidney Int.

2013 Feb;83(2):278-84.

21. Verhave JC, Bouchard J, Goupil R, Pichette V, Brachemi S, Madore F, Troyanov S. Clinical

value of inflammatory urinary biomarkers in overt diabetic nephropathy: a prospective

study. Diabetes Res Clin Pract. 2013 Sep;101(3):333-40.

22. Komorowsky CV, Brosius FC 3rd, Pennathur S, Kretzler M. Perspectives on systems biology

applications in diabetic kidney disease. J Cardiovasc Transl Res. 2012 Aug;5(4):491-508.

23. Gerszten RE, Wang TJ. The search for new cardiovascular biomarkers. Nature. 2008 Feb

21;451(7181):949-52.

24. Merchant ML, Perkins BA, Boratyn GM, Ficociello LH, Wilkey DW, Barati MT, Bertram CC,

Page GP, Rovin BH, Warram JH, Krolewski AS, Klein JB. Urinary peptidome may predict

renal function decline in type 1 diabetes and microalbuminuria. J Am Soc Nephrol. 2009

Sep;20(9):2065-74.

1918


1

25. Ben Ameur R, Molina L, Bolvin C, Kifagi C, Jarraya F, Ayadi H, Molina F, Granier C.

Proteomic approaches for discovering biomarkers of diabetic nephropathy. Nephrol Dial

Transplant. 2010 Sep;25(9):2866-75.

26. Kolch W, Neususs C, Pelzing M, Mischak H. Capillary electrophoresis-mass spectrometry

as a powerful tool in clinical diagnosis and biomarker discovery. Mass Spectrom Rev.

2005 Nov-Dec;24(6):959-77.

27. Bos H, Andersen S, Rossing P, De Zeeuw D, Parving HH, De Jong PE, Navis G. Role

of patient factors in therapy resistance to antiproteinuric intervention in nondiabetic and

diabetic nephropathy. Kidney Int Suppl. 2000 Apr;75:S32-7.

28. Schievink B, de Zeeuw D, Parving HH, Rossing P, Lambers Heerspink HJ. The renal

protective effect of angiotensin receptor blockers depends on intra-individual response

variation in multiple risk markers. Br J Clin Pharmacol. 2015 Apr 14. [Epub ahead of print].

29. De Zeeuw D, Remuzzi G, Parving HH, Keane WF, Zhang Z, Shahinfar S, Snapinn S, Cooper

ME, Mitch WE, Brenner BM. Proteinuria, a target for renoprotection in patients with type 2

diabetic nephropathy: Lessons from RENAAL. Kidney Int. 2004 Jun;65(6):2309-20.

30. Smink PA, Bakker SJ, Laverman GD, Berl T, Cooper ME, de Zeeuw D, Lambers Heerspink

HJ. An initial reduction in serum uric acid during angiotensin receptor blocker treatment is

associated with cardiovascular protection: a post-hoc analysis of the RENAAL and IDNT

trials. J Hypertens. 2012 May;30(5):1022-8.

31. Miao Y, Dobre D, Heerspink HJ, Brenner BM, Cooper ME, Parving HH, Shahinfar S,

Grobbee D, de Zeeuw D. Increased serum potassium affects renal outcomes: a post

hoc analysis of the Reduction of Endpoints in NIDDM with the Angiotensin II Antagonist

Losartan (RENAAL) trial. Diabetologia. 2011 Jan;54(1):44-50.

32. Mohanram A, Zhang Z, Shahinfar S, Lyle PA, Toto RD. The effect of losartan on hemoglobin

concentration and renal outcome in diabetic nephropathy of type 2 diabetes. Kidney Int.

2008 Mar;73(5):630-6.

33. Crowley SD, Gurley SB, Oliverio MI, Pazmino AK, Griffiths R, Flannery PJ, Spurney RF,

Kim HS, Smithies O, Le TH, Coffman TM. Distinct roles for the kidney and systemic

tissues in blood pressure regulation by the renin-angiotensin system. J Clin Invest. 2005

Apr;115(4):1092-9.

34. Vogt L, Waanders F, Boomsma F, de Zeeuw D, Navis G. Effects of dietary sodium and

hydrochlorothiazide on the antiproteinuric efficacy of losartan. J Am Soc Nephrol. 2008

May;19(5):999-1007.

35. Yasar U, Forslund-Bergengren C, Tybring G, Dorado P, Llerena A, Sjöqvist F, Eliasson

E, Dahl ML. Pharmacokinetics of losartan and its metabolite E-3174 in relation to the

CYP2C9 genotype. Clin Pharmacol Ther. 2002 Jan;71(1):89-98.

36. Parving HH, de Zeeuw D, Cooper ME, Remuzzi G, Liu N, Lunceford J, Shahinfar S, Wong

PH, Lyle PA, Rossing P, Brenner BM. ACE gene polymorphism and losartan treatment in

type 2 diabetic patients with nephropathy. J Am Soc Nephrol. 2008 Apr;19(4):771-9.

PART 1Novel biomarker panels for predicting disease progression

CHAPTER 2A panel of novel biomarkers representing different disease pathways improves prediction of renal function decline in type 2 diabetes

MJ PenaA HeinzelG Heinze

A AlkhalafSJL BakkerTQ Nguyen

R GoldschmedingHJG BiloP PercoB Mayer

D de Zeeuw HJ Lambers Heerspink

PLoS One. 2015 May 14;10(5):e0120995

2524

Biomarker panels and eGFR declineChapter 2

2

ABSTRACT

Objectives: We aimed to identify a novel panel of biomarkers predicting renal function decline in type 2 diabetes, using biomarkers representing different disease pathways speculated to contribute to the progression of diabetic nephropathy.

Methods: A systematic data integration approach was used to select biomarkers representing different disease pathways. Twenty-eight biomarkers were measured in 82 patients seen at an outpatient diabetes center in the Netherlands. Median follow-up was 4.0 years. We compared the cross-validated explained variation (R2) of two models to predict eGFR decline, one including only established risk markers, the other adding a novel panel of biomarkers. Least absolute shrinkage and selection operator (LASSO) was used for model estimation. The C-index was calculated to assess improvement in prediction of accelerated eGFR decline defined as ≤-3.0 mL/min/1.73m2/year.

Results: Patients’ average age was 63.5 years and baseline eGFR was 77.9 mL/min/1.73m2. The average rate of eGFR decline was -2.0 ± 4.7 mL/min/1.73m2/year. When modeled on top of established risk markers, the biomarker panel including matrix metallopeptidases, tyrosine kinase, podocin, CTGF, TNF-receptor-1, sclerostin, CCL2, YKL-40, and NT-proCNP improved the explained variability of eGFR decline (R2 increase from 37.7% to 54.6%; p=0.018) and improved prediction of accelerated eGFR decline (C-index increase from 0.835 to 0.896; p=0.008).

Conclusions: A novel panel of biomarkers representing different pathways of renal disease progression including inflammation, fibrosis, angiogenesis, and endothelial function improved prediction of eGFR decline on top of established risk markers in type 2 diabetes. These results need to be confirmed in a large prospective cohort.

INTRODUCTION

The growing prevalence of type 2 diabetes is a great global health problem. Type 2 diabetes is the leading cause of chronic kidney disease (CKD) in the United States and is associated with high cardiovascular risk [1, 2]. Optimizing treatment has been shown to improve life expectancy, reduce costs, and lower the risk of death in patients with type 2 diabetes [3, 4]. Despite important progress in improving therapy, many patients are still at risk for renal disease.

Early identification of patients with type 2 diabetes at risk for progressive renal function loss during the early stages of disease may lead to better patient outcomes. In clinical practice, estimated glomerular filtration rate (eGFR) and albuminuria are used to assess renal function when gold-standard measured GFR is not feasible or practical. The search for novel biomarkers that improve risk prediction models on top of established risk markers has been a priority of many researchers for many years. Various studies have assessed the performance of single biomarkers representing a single, disease-associated pathway to predict progression of renal function loss in type 2 diabetes [5, 6]. However, because type 2 diabetes is a multifactorial disease, several pathways involving pro-inflammatory, pro-fibrotic, and angiogenic processes, among others, are activated during the course of the disease [7]. Given the complexity of the multiple pathophysiological processes involved in progression of type 2 diabetes together with the intra-individual variability of biomarkers, it is questionable if a single biomarker may possess useful diagnostic and prognostic power. Alternatively, a combination of biomarkers that capture different pathways of renal damage may provide a more realistic picture of a patient’s actual pathophysiological status and hence may yield better assessment of disease prognosis performance.

Therefore, we aimed to identify a novel panel of biomarkers representing different disease pathways that are speculated to contribute to the progression of renal disease in type 2 diabetes, and to evaluate their combined predictive performance of accelerated renal function decline.

2726


2

METHODS

Patients and methodsThis observational cohort study was performed in Caucasian patients from Zwolle, the Netherlands, who participated in the PREvention of DIabetic ComplicaTIONS (PREDICTIONS) study [8]. Patients aged 35-75 with type 2 diabetes with a documented duration of ≥5 years were eligible for the PREDICTIONS study. Type 2 diabetes was defined according to World Health Organization criteria [9]. A total of 82 patients were recruited in 2007 - 2008 and followed for a median of 4.0 [1st, 3rd quartile 3.7 to 4.4] years. Follow-up information on urinary albumin:creatinine ratio (UACR), serum creatinine, cholesterol, and glycated hemoglobin (HbA1c) was obtained from electronic patient files from visits to the outpatient diabetes clinic during their annual visit to the diabetes specialist.

Ethics StatementThe PREDICTIONS study was approved by the ethical review boards of the medical ethics committees of the Isala Clinics in Zwolle and of the University Medical Center in Groningen, the Netherlands, and was conducted in accordance with the guidelines of the Declaration of Helsinki. All patients gave written, informed consent.

Selection of biomarkers, sample collection, preparation, and measurementTwenty-eight biomarkers were selected for testing using three distinct approaches, namely a literature review [10], identification of molecular processes and pathways [7], and ranking of consolidated Omics signatures [11]. A complete list of biomarkers is presented in Table 1, and the biomarker selection procedure is described in Supplemental Appendix 1.

Fasting serum and plasma samples were stored at -80°C. All samples were stored for 4 - 5 years and did not undergo any freeze-thaw cycles. Biomarkers were assayed on baseline samples by enzyme-linked immunosorbent assay (ELISA) or multiplex assay by Biomarker Design Forschungs GmbH (BDF), in Vienna, Austria, except for connective tissue growth factor (CTGF). CTGF was measured using specific antibodies (FibroGen Inc., San Francisco, USA) directed against distinct epitopes in the amino-terminal fragment of CTGF, as described previously [12]. All assays were used according to manufacturer’s instructions. A complete list of assays, and information on stability and determination of limits of detection are available in Supplemental Appendix 2. All biomarker analyses were performed blinded, and the results were then reported back to the study center for analysis.

Statistical analysisAnalyses were performed with SAS software (version 9.2; SAS Institute, Cary, NC) and R version 3.0.2 [13] using the packages mice and glmnet [14, 15]. Data are presented as mean (standard deviation) or median [1st, 3rd quartile] for skewed variables. Graphical techniques were used to detect outliers. The natural logarithm of UACR and the binary logarithm of all biomarkers were used to normalize their distributions. Log transformed variables were used in all regression analysis. Values below the detection limit were set to the detection limit. Variables with missing values were multiply imputed using chained

equations [16]. Five of the twenty-eight biomarkers had values with >10% missing or >25% below the detection limit were not used in analysis. Details on our implementation of multiple imputation can be found in Supplemental Appendix 3. All p-values were two-tailed, and values <0.05 were considered statistically significant.

The outcome of interest was eGFR decline, defined as the within-patient annual eGFR slope. EGFR decline was calculated using a minimum of 3 serum creatinine measurements during follow-up by fitting a straight line through the eGFR values using linear regression. The eGFR value at each time-point was estimated using the 4-variable Modification of Diet in Renal Disease (MDRD) Study Equation [17].

Statistical modeling consisted of several steps. First, established risk markers were selected as best predictors of eGFR decline using least absolute shrinkage and selection operator (LASSO) selection [18]. The LASSO is advantageous for small samples sizes because it places restrictions on the absolute sizes of the regression coefficients in the model while optimally selecting the subset of variables that best predicts the outcome. This restriction also controls for multicollinearity. LASSO involves the estimation of a tuning parameter controlling the amount of restriction, which was optimized by minimizing the leave-one-out cross-validated mean squared error (MSE) of prediction. The established risk markers listed in Table 2 were considered as potential predictors of eGFR decline. First, all established risk markers were included in a multivariable model using LASSO regression. The best predictors of eGFR decline were then identified from the multivariable model and are reported in the results section. Second, univariate linear models were fit for each of the novel biomarkers to assess a single biomarker association with eGFR decline. Third, multivariable models were then fit by linear regression with single novel biomarkers adjusting for the selected established risk markers. Fourth, a multivariable model including the selected established risk markers and all biomarkers was fit using the LASSO selection in order to find the best subset of predictors.

Bootstrap validation was performed to determine the validity of the model to assess the ability of the biomarker panel to predict renal function decline. The bootstrap (N=1000) was used to evaluate selection probabilities of each biomarker, and to construct 95% confidence intervals and two-sided p-values for the regression coefficients by the percentile method. A global p-value testing the global null hypothesis of no added value of the biomarkers was constructed by counting the number of bootstrap resamples in which the multivariable biomarker model led to a smaller cross-validated MSE than a model based on the established risk markers alone. In a simple bootstrap validation, LASSO models were fit to the 1000 bootstrap resamples, each time optimizing the cross-validated MSE as described above. These models were then applied to the original data without modification. The resulting MSE was calculated by averaging the squared average difference between the original outcome and the predicted outcome for each patient. This was done for models only considering the established risk markers, and for models considering clinical and biomarker predictors. From the MSEs, R2 measures were finally derived in order to determine whether the biomarkers significantly improved prediction.

2928


2

Tab

le 1

. Con

cent

ratio

ns o

f bio

mar

kers

* and

uni

varia

te a

nd m

ultiv

aria

ble

ass

ocia

tions

of s

ingl

e b

iom

arke

rs w

ith e

GFR

dec

line.

Co

ncen

trat

ions

Uni

vari

ate

asso

ciat

ion

Mul

tiva

riab

le a

sso

ciat

ion†

Pat

hway

Bio

mar

ker

Med

ian

[1st, 3

rd q

uart

ile]

β95

% C

Ip

-val

ueβ

95%

CI

p-v

alue

Infla

mm

atio

n

M

onoc

yte

chem

oatt

ract

ant

pro

tein

-1 (C

CL2

) (p

g/m

L)31

6.2

[258

.3, 3

86.4

]-1

.1-3

.6, 1

.50.

410.

1-2

.0, 2

.30.

89

T

umor

nec

rosi

s fa

ctor

rec

epto

r-1

(TN

FR1)

(ng/

mL)

3.8

[3.1

, 6.7

]-3

.2-5

.2, -

1.2

<0.

01-2

.1-4

.0, -

0.3

0.03

T

umor

nec

rosi

s fa

ctor

rec

epto

r-2

(TN

FR2)

(pg/

mL)

317.

6 [2

48.3

, 475

.0]

-2.4

-3.8

, -0.

9<

0.01

-2.2

-4.6

, 0.3

0.08

C

hitin

ase

3-lik

e 1

(YK

L-40

) (ng

/mL)

36.5

[21.

1, 8

7.1]

-1.1

-1.9

, -0.

20.

01-0

.5-1

.3, 0

.30.

20

C

hem

okin

e (C

-X-C

mot

if) 1

(CX

CL1

) (p

g/m

L)80

.0 [7

1.1,

94.

7]1.

6-1

.7, 4

.80.

33-0

.2-3

.0, 2

.60.

88

C

hem

okin

e (C

-X-C

mot

if) 1

0 (C

XC

L10)

(pg/

mL)

80.6

[58.

1, 1

21.7

]-0

.5-1

.7, 0

.60.

35-0

.5-1

.4, 0

.50.

35

Fib

rosi

s

C

onne

ctiv

e tis

sue

grow

th fa

ctor

(CTG

F) (n

mol

/L)

1.0

[0.8

, 1.5

]-3

.4-7

.8, 0

.30.

07-3

.8-7

.7, 0

.10.

05

M

atrix

met

allo

pep

tidas

e 1

(MM

P1)

(pg/

mL)

767.

4 [4

70.0

, 131

5.5]

-0.4

-1.4

, 0.7

0.49

0.2

-0.8

, 1.2

0.72

M

atrix

met

allo

pep

tidas

e 2

(MM

P2)

(ng/

mL)

38.3

[36.

1, 4

0.0]

9.6

0.4,

18.

70.

045.

6-1

.9, 1

3.1

0.14

M

atrix

met

allo

pep

tidas

e 7

(MM

P7)

(ng/

mL)

1.6

[0.6

, 3.0

]-1

.4-2

.1, -

0.7

<0.

01-0

.8-1

.5, -

0.04

0.04

M

atrix

met

allo

pep

tidas

e 8

(MM

P8)

(ng/

mL)

2.4

[1.5

, 4.8

]0.

1-0

.8, 0

.70.

880.

3-0

.3, 0

.90.

34

M

atrix

met

allo

pep

tidas

e 13

(MM

P13

) (p

g/m

L)12

0.0

[108

.5, 1

42.4

]-1

.2-3

.0, 0

.70.

23-0

.7-2

.4, 1

.00.

40

P

odoc

in (N

PH

S2)

(ng/

mL)

0.9

[0.3

, 1.2

]-2

.9-5

.0, -

0.7

0.01

-0.9

-3.0

, 1.3

0.44

L

eptin

(LE

P) (

ng/m

L)15

.9 [1

0.1,

33.

3]0.

2-0

.6, 0

.90.

66-0

.3-1

.1, 0

.60.

54

Ang

ioge

nesi

s

E

ndos

tatin

(Fra

g.C

OL1

8A1)

(pm

ol/L

)7.

6 [6

.3, 9

.7]

-2.6

-5.0

, -0.

30.

03-1

.3-4

.4, 1

.70.

39

T

yros

ine

kina

se (T

EK

) (p

g/m

L)66

6.9

[315

.7, 1

275.

8]-0

.6-1

.6,0

.30.

18-0

.9-1

.7, -

0.1

0.03

V

ascu

lar

end

othe

lial g

row

th fa

ctor

-A (V

EG

F-A

) (p

g/m

L)66

.7 [3

0.6,

155

.9]

-0.5

-1.1

, 0.2

0.13

-0.2

-0.8

, 0.4

0.49

H

epat

ocyt

e gr

owth

fact

or (H

GF)

(pg/

mL)

65.9

[35.

0, 1

20.3

]-0

.7-1

.5, 0

.20.

11-0

.2-0

.9, 0

.50.

61

End

othe

lial D

ysfu

nctio

n

A

min

o te

rmin

al p

ro C

-typ

e na

triu

retic

pep

tide

(NT-

pro

CN

P)(p

mol

/L)

2.9

[2.3

, 4.3

]-1

.3-2

.9, 0

.30.

12-0

.7-2

.4, 0

.90.

38

Min

eral

met

abol

ism

F

ibro

bla

st g

row

th fa

ctor

23

(FG

F23)

(pm

ol/L

)4.

0 [2

.6, 5

.5]

-0.8

-2.1

, 0.5

0.24

-0.2

-1.3

, 1.0

0.78

S

cler

ostin

(SO

ST)

(pm

ol/L

)42

.7 [3

3.4,

52.

2]-0

.1-2

.4, 2

.30.

960.

3-1

.6, 2

.30.

75

Lip

id m

etab

olis

m

Z

inc-

bin

din

g al

pha

-2-g

lyco

pro

tein

1 (A

ZG

P1)

(ng/

mL)

13.4

[9.0

, 20.

2]-1

.1-2

.5, 0

.40.

14-0

.6-1

.8, 0

.70.

37

Glo

mer

ular

dam

age

G

row

th h

orm

one

1 (G

H1)

(pg/

mL)

330.

8 [5

3.4,

994

.2]

-0.2

-0.7

, 0.3

0.38

-0.3

-0.7

, 0.1

0.14

* Con

cent

ratio

ns o

f nep

hrin

(NP

HS

1), n

euro

pili

n-1

(NR

P1)

, int

erle

ukin

-1 a

lpha

(IL1

A),

inte

rleuk

in-1

bet

a (IL

1B),

and

ep

ider

mal

gro

wth

fact

or (E

GF)

wer

e m

issi

ng

in 1

0% o

f ob

serv

atio

ns o

r un

det

ecta

ble

in >

25%

of o

bse

rvat

ions

, and

the

se b

iom

arke

rs w

ere

ther

efor

e no

t us

ed in

ana

lysi

s.† A

dju

sted

for e

stab

lishe

d ri

sk m

arke

rs: b

asel

ine

UA

CR

, cur

rent

vs.

nev

er s

mok

er, s

ex, s

ysto

lic b

lood

pre

ssur

e, u

se o

f ora

l dia

bet

ic m

edic

atio

n, d

iast

olic

blo

od

pre

ssur

e, a

nd b

asel

ine

eGFR

.

3130


2

Table 2. Baseline characteristics in patients with type 2 diabetes (n=82).

Risk marker Baseline values

Age (years) 63.5 ± 9.4

Male Gender (%) 44 (53.7)

Current smoker (%) 8 (9.6)

Body mass index (kg/m2) 32.4 ± 6.3

Systolic blood pressure (mmHg) 135.2 ± 16.3

Diastolic blood pressure (mmHg) 72.7 ± 10.5

Duration of diabetes (years) 15.7 ± 7.3

Baseline laboratory measurements

UACR (mg/mmol) 1.2 [0.5, 57.7]

Serum creatinine (µmol/L) 88.4 ± 33.5

eGFR (mL/min/1.73m2) 77.9 ± 22.6

HDL Cholesterol (mmol/L) 1.3 ± 0.4

LDL Cholesterol(mmol/L) 2.0 ± 0.6

HbA1c (%) 7.7 ± 1.3

Medication use

RAAS* (%) 27 (42.9)

Insulin* (%) 58 (92.1)

Oral diabetic medication* (%) 35 (55.6)

Data are reported as mean ± standard deviation or number (percent) or median [1st, 3rd quartile].

*Data available for n = 63.

The added value of the biomarker panel was also evaluated using the discriminative index (C-index) by dichotomizing the observed outcome variables into accelerated or non-accelerated renal function decline (eGFR decline ≤-3 or >-3 mL/min/1.73m2/year, respectively) and comparing this with predicted probabilities of eGFR decline (see Supplemental Appendix 3). The C-index was also calculated using the simple bootstrap validation scheme, and the differences in the C-index between a model of only established risk markers and a model of established risk markers plus biomarkers were assessed. The threshold of -3 mL/min/1.73m2 was based on prior studies and its concurrence with the high quartile of eGFR decline [19, 20].

RESULTS

Baseline characteristics and association with eGFR declineBaseline characteristics are presented in Table 2. The average age of the cohort was 63.5 (SD 9.4) years and 53.7% were male. Type 2 diabetes was well established in the study population with average diabetes duration of 15.7 (SD 7.3) years. Renal function was relatively preserved in the cohort with an average eGFR of 77.9 (SD 22.6) mL/min/1.73m2 at baseline. Median UACR was 10.6 [1st, 3rd quartile: 4.42, 510.1] mg/g. The average rate of eGFR decline over the median of 4.0 [1st, 3rd quartile: 3.7, 4.4] years of follow-up was -2.1 (SD 4.5) mL/min/1.73m2/year.

The following best predictors of eGFR decline were selected from the LASSO selection: baseline UACR, current vs. never smoker, sex, systolic and diastolic blood pressure, use of oral diabetic medication, and baseline eGFR (Supplemental Figure 1).

Biomarker concentrations and associations with eGFR declineBaseline biomarker concentrations and univariate associations of the single biomarkers with eGFR decline are reported in Table 1. Higher concentrations of the individual biomarkers matrix metallopeptidases 2 (MMP2) (p = 0.04), matrix metallopeptidases 7 (MMP7) (p < 0.01), chitinase 3-like 1 (YKL-40) (p = 0.01), tumor necrosis factor receptor-1 (TNFR1) (p < 0.01), podocin (NPHS2) (p = 0.01), and endostatin (frag.COL18A1) (p = 0.03) were significantly associated with eGFR decline. When single biomarkers were modeled adjusting for established risk markers, MMP7, tyrosine kinase (TEK), and TNFR1 were independently associated with eGFR decline (Table 1). For every two-fold increase in the log concentration of MMP7, TEK, or TNFR1, a corresponding decrease of eGFR of 0.77 (p = 0.04), 0.90 (p = 0.02), and -2.1 (p = 0.03) mL/min/1.73m2/year, respectively, was observed.

When these three biomarkers were modeled on top of the established risk markers, they did not improve the explained variability (R2) of eGFR decline (35.7% compared to 37.7% of the reference model; p = 0.988). The three biomarkers also did not increase the C-index for prediction of accelerated renal function decline (0.860 compared to 0.835 of the reference model; p = 0.262).

Selection of optimal combination of established risk markers and biomarkersAlthough most individual biomarkers were not found to be independently associated with eGFR decline, we hypothesized that the combination of biomarkers representing different disease pathways may improve prediction of eGFR decline. In a multivariable LASSO selection, the optimal model for prediction of eGFR decline was achieved after inclusion of 19 variables (Figure 1). The model included a subset of 13 novel biomarkers representing fibrosis, angiogenesis, inflammation, mineral metabolism, and endothelial function that, when added to the established risk markers, more accurately predicted the rate of eGFR decline (Table 3). The explained variability of the model (R2) markedly increased from 37.7% to 54.6% (p = 0.018) and predicted a higher probability of accelerated renal

3332


2

function decline (Figure 2). There was also a significant improvement in the C-index of the optimal model for prediction of accelerated renal function decline (0.896 compared to 0.835 of the reference model; p = 0.008) (Figure 3).

To investigate the importance of each of the predictors in the optimal model, we omitted, one by one, variables from the full model. If a variable was omitted from the model, the other predictor variables could be selected instead. Only the omission of UACR or systolic blood pressure resulted in relevant inclusions of other novel biomarkers (Supplemental Table 1).

Table 3. Optimal model of established risk markers and biomarkers, results from LASSO selection and bootstrap resampling (N=1000).

Variable mean β 95% CI* pSelection

probability†

Baseline UACR -0.509 -0.834, -0.159 0.002 0.999

Systolic blood pressure 0.049 0.010, 0.085 0.012 0.994

MMP2 7.382 0.010, 0.085 0.018 0.993

TEK -0.793 -1.416, -0.139 0.018 0.993

Baseline eGFR -0.072 -0.130, -0.014 0.026 0.987

CTGF -5.911 -10.358, -0.913 0.026 0.987

MMP7 -0.540 -1.191, 0.0 0.078 0.966

Current vs. never smoker -1.593 -3.905, 0.0 0.144 0.943

MMP8 0.472 0.0, 1.036 0.134 0.935

NPHS2 -1.509 -3.667, 0.0 0.206 0.908

MMP1 0.392 -0.081, 1.051 0.298 0.897

TNFR1 -1.618 -4.037, 0.0 0.228 0.889

SOST 0.983 -0.014, 2.556 0.278 0.888

Oral diabetic medication -1.060 -2.673, 0.0 0.274 0.884

MMP13 -0.363 -1.835, 1.020 0.798 0.820

Sex 0.792 -0.905, 2.814 0.592 0.785

CCL2 0.461 -1.228, 2.672 0.854 0.781

YKL-40 -0.405 -1.358, 0.019 0.518 0.771

NT-proCNP 0.756 -0.002, 2.452 0.568 0.742

*95% confidence interval, estimated from the 2.5th and 97.5th percentiles of the bootstrap distribution. †The relative frequency of the marker being included in the model across 1000 bootstrap resamples.

Figure 1. LASSO selection of optimal model of established risk markers and biomarkers: cross validated mean squared error (Y-axis; black bullets; MSE) vs. amount of restriction (X-axis; log(Lambda)). Vertical bars refer to standard errors across the 82 cross-validations. The best cross-validated MSE was obtained after inclusion of 19 variables (step 31), which included baseline UACR, MMP7, current vs. never smoker, sex, TEK, MMP2, systolic blood pressure, baseline eGFR, TNFR1, NPHS2, CTGF, use of oral diabetic medication, YKL-40, MMP1, MMP13, MMP8, SOST, CCL2, and NT-proCNP.

Figure 2. Predicted probability of accelerated renal function decline (eGFR decline ≤-3 or >-3 mL/min/1.73m2/year) in patients with type 2 diabetes.

3534


2

Figure 3. C-index for prediction of accelerated renal function decline (eGFR decline <-3 or >-3 mL/min/1.73m2/year) for a) established risk markers (reference model: baseline UACR, current vs. never smoker, sex, systolic and diastolic blood pressure, use of oral diabetic medication, and baseline eGFR) (C-index = 0.835), b) 3-biomarker model (MMP7, TEK, and TNFR1 on top of reference model) (C-index = 0.835; p = 0.262 compared to reference model), and c) Optimal model (baseline UACR, MMP7, current vs. never smoker, sex, TEK, MMP2, systolic blood pressure, baseline eGFR, TNFR1, NPHS2, CTGF, use of oral diabetic medication, YKL-40, MMP1, MMP13, MMP8, SOST, CCL2, and NT-proCNP) (C-index = 0.896; p = 0.008 compared to reference model).

DISCUSSION

In this study, we established that a combination of different biomarkers representing different pathways that are speculated to be involved in the progression of renal disease improves prediction of eGFR decline. Although some biomarkers were not independently associated with eGFR decline, when combined into a multi-biomarker model, the combination of biomarkers improved renal risk stratification, suggesting that these biomarkers may possess synergistic effects in predicting renal function loss.

Diabetic kidney disease is characterized by the functional impairment and structural remodeling of the kidney and is linked to the changes in the kidney. Diabetic nephropathy is well characterized by glomerular hypertrophy and hyperfiltration, inflammation of glomeruli and tubuliointerstitial regions, and reduction of cell number by apoptosis and

accumulation of extracellular matrix (ECM). Each of the biomarkers selected in the optimal model has been associated with one of these pathophysiological processes involved in diabetic nephropathy.

First, chronic inflammation has long been identified in the pathogenesis of type 2 diabetes and progression of diabetic nephropathy, and inflammation is well represented by the biomarkers included in the optimal model. Tumor necrosis factor alpha is a key mediator of inflammation and plays a role in apoptosis. It mediates its signal via two distinct receptors, TNFR1 and TNFR2. Circulating forms of both TNF receptors were recently shown to predict end-stage renal disease (ESRD) in type 2 diabetes [6]. Monocyte chemoattractant protein-1 (CCL2), another marker of inflammation, is a potent C-C chemokine for monocyte/macrophages and T cells. Increased amounts of CCL2 have been detected in renal biopsies and urine from patients with diabetic nephropathy [21], and CCL2 has been shown to be a marker of late stage diabetic nephropathy [22]. Currently there are a couple of clinical trials ongoing that target CCL2 receptor as a means to delay progression of diabetic nephropathy (www.clinicaltrials.gov identifier NCT01712061, NCT01752985). Results of these studies will provide more insight whether CCL2 is a causal factor or consequence of renal function loss. Additionally, YKL-40, a proinflammatory marker, has been identified as an independent factor associated with albuminuria in early stage nephropathy in type 2 diabetes and might have a useful role as a noninvasive marker for the early diabetic nephropathy detection [5, 23]. High YKL-40 levels have been shown to predict mortality in patients with type 2 diabetes [24]. Future mechanistic studies exploring the interplay between different inflammatory markers will help determine which markers are causal factors or consequences in the progression of diabetic kidney disease.

Second, the optimal model included several biomarkers linked to pro-fibrotic processes. Fibrosis, resulting from expansion and change in composition of ECM in the kidney, is a well-known pathologic feature of diabetic complications. Altered expression of matrix metalloproteinases (MMPs) have been implicated in the progression of diabetic nephropathy by affecting the breakdown and turnover of ECM. In mice, the overexpression of MMP-9 has been shown to induce podocyte dedifferentiation, interrupt podocyte cell integrity, and promote podocyte monolayer permeability to albumin and extracellular matrix protein synthesis [25]. In humans, serum MMP7 has been shown to be increased in diabetic renal disease and diabetic diastolic dysfunction [26]. In support of this, our study showed that higher concentrations of MMP7 were independently associated with eGFR decline. CTGF is another well investigated pro-fibrotic biomarker that was included in the optimal model. CTGF, which is upregulated in diabetic nephropathy and contributes to extracellular matrix accumulation, has been associated with both early and late stage diabetic nephropathy [12, 22]. Down-regulation of CTGF and vascular endothelial growth factor-A (VEGF-A) in diabetic nephropathy is speculated to be a result of podocyte loss [27]. Our data, in conjunction with data from literature, support the importance of fibrotic pathways in the initiation and progression of diabetic kidney disease.

3736


2

Third, we included a marker representing angiogenesis. Angiogenesis is the formation of new blood vessels from pre-existing vasculature. Neovascularization has been implicated in the genesis of diverse diabetic complications such as retinopathy, impaired wound healing, neuropathy, and diabetic nephropathy. In both physiological and pathological angiogenesis, tyrosine kinase (TEK) plays a key role. TEK is principally expressed in endothelial cells and inhibits vascular permeability and tightens preexisting vessels [28]. Additionally, TEK plays a critical role in the angiogenesis of endothelial cells via binding to angiopoietin [29].

Finally the model included a marker representing endothelial function. Endothelial dysfunction is considered an initial step of the atherosclerotic process because diabetes substantially impairs vasodilating properties of the endothelium which leads to impaired vasodilation and ultimately endothelial dysfunction [30]. C-type natriuretic peptide (CNP), a member of the natriuretic peptide family, is produced in vascular endothelium. Our study implies that natriuretic (NT)-proCNP, the N-terminal fragment of the C-type natriuretic peptide precursor, contributes to prediction of eGFR decline. NT-proCNP has been shown to be associated with arterial stiffness, endothelial dysfunction, and early atherosclerosis [31], however the link of NT-proCNP to type 2 diabetes and nephropathy is still under investigation.

In our study, most biomarkers were not able to individually predict eGFR decline after adjustment for established risk markers, and the model of 3 biomarkers did not statistically improve prediction. Rather, the optimal model of 13 biomarkers yielded best and significant improvements in the C-index. Advancing laboratory techniques allowing simultaneous measurement of many biomarkers are becoming more and more realistic in clinical practice. Whether the biomarkers identified are either involved in the causal pathway contributing to CKD progression, or are markers of its risk, or are merely the end-product of existing pathological processes, remains an important and unresolved question that requires further exploration. A future study on etiology to examine the causal relationship between these biomarkers as risk factors of renal disease would be appropriate, and issues of confounding could then be addressed. Testing for confounding was beyond the scope of this prediction study; however, we were able to investigate the importance of each of the predictors in the optimal model. Baseline UACR was found to have the largest impact on eGFR decline, and only the omission of baseline UACR or systolic blood pressure allowed inclusions of other novel biomarkers into the model. The combination of multiple biomarkers in the final, optimal model appears to be more accurate in risk stratification for accelerated renal function decline in patients with type 2 diabetes.

There are some studies in literature that use a multi-biomarker approach for risk prediction in CKD. A recent study showed that the combination of a panel of biomarkers including inflammation, fibrosis, and cardiac stretch and injury improved prediction of death in a Canadian CKD cohort; however, this study was conducted in a cohort with different CKD etiology [32]. Additionally, in another study of multiple protein biomarkers, 17 urinary

and 7 plasma biomarkers were evaluated to predict progression. C-terminal FGF-23 and VEGF-A were found to be associated with the end point independent of urine albumin/creatinine. In that study many biomarkers were tested one by one, but did not use a combined biomarker approach to predict renal disease progression [33]. Furthermore, a panel of multiple urinary cytokines was found to predict rapid renal function decline in overt diabetic nephropathy [34]. However, that study included a heterogeneous population of patients with both type 1 and type 2 diabetes. Finally, in a post-hoc study from the IRMA-2 trial showed that multiple biomarkers of endothelial dysfunction and possibly inflammation were predictors of progression to diabetic nephropathy in patients with type 2 diabetes and microalbuminuria independent of traditional risk markers [35].

Advances in high throughput analytical methods has fueled novel biomarker discovery. Two such platforms, namely proteomics and metabolomics, have shown promise in multi-biomarker discovery for the diabetic CKD. A urinary peptide classifier, consisting of 273 defined urinary peptides, was recently discovered as a good classifier in patients with CKD [36] and validated in an independent cohort as a predictor of albuminuria progression in patients with type 2 diabetes [37]. Furthermore, a panel of 13 metabolites linked with mitochondrial metabolism was significantly reduced in CKD patients with diabetes compared to healthy controls [38], and the combination of plasma metabolites butenoylcarnitine and histidine, and urine metabolites glutamine, tyrosine, and hexoses were able to predict the progression from micro- to macroalbuminuria in patients with type 2 diabetes [39].

Interestingly in our study, HbA1c and duration of diabetes were not strong predictors of eGFR decline, whereas albuminuria was identified as the strongest predictor. The exclusion of HbA1c and duration of diabetes from the reference model may be due to small variations in these parameters within this population. Regarding albuminuria, there is evidence that demonstrates albuminuria as a strong risk predictor of renal function loss in patients with type 2 diabetes [40-43]. Moreover, experimental data show that increased albumin exposure to the tubuli causes tubulo-interstitial damage through activation of pro-inflammatory mediators, which leads to a progressive decline in glomerular and tubular function, ultimately culminating in end-stage renal disease [44, 45]. Our data on albuminuria as a strong predictor of eGFR decline are in line with this and highlight the importance of screening for high albuminuria to identify individuals at risk of progressive renal function loss. At the same time, it may be interesting to explore the predictive ability of urine biomarkers alongside albuminuria for renal disease progression as urine is considered quite a suitable substrate to measure biomarkers linked to kidney disease due to the practical advantages of collecting urine compared to blood samples. Since our study measured biomarkers in blood, we are unable to speculate if urine biomarkers, or the combination of both blood and urine markers, would yield similar predictive capabilities.

There are strengths and limitations to this study. A clear strength is the use of a multi-marker, multi-pathway approach for identifying and testing biomarkers in a population of patients with type 2 diabetes over approximately 4 years of follow-up. The clear limitation

3938


2

is the measurement of multiple biomarkers in a small sample size. However, as advancing laboratory techniques generate larger amounts of data, methods of data analysis to accommodate “big data” with smaller sample sizes are needed. The rigorous statistical method of the LASSO regression allowed for modeling many biomarkers in the small sample size, and multiple imputation was used to avoid truncating observations due to missing data. The true predictive capacity of the model could have been overestimated due to the prediction model being developed and tested in the same sample, and we do agree that external validation is necessary. In the absence of external validation, we performed internal bootstrap validation in an attempt to minimize this limitation [46]. GFR was estimated using a serum creatinine-based equation instead of by direct measurement, which may have contributed to misclassification bias. However, this could have only resulted in an underestimation of the strength of the reported associations. We chose to omit five biomarkers from our analysis due to many missing or below limit of detection values. While the exclusion of these biomarkers from our analysis may have resulted in an underrepresentation of pathways, the omission of biomarkers could have only underestimated the predictive ability of the biomarker panel. Additional limitations include the lack of information concerning insulin use, diet, and renin-angiotensin-aldosterone system medication type and dose, which clearly represent unmeasured confounders in our study.

In conclusion, novel biomarkers may provide deeper understanding into the pathophysiology of CKD or diabetic nephropathy but identification of progression-associated molecular pathways via biomarkers as proxy may also help to identify novel therapeutic targets. We identified a novel panel of biomarkers representing different pathways of renal damage, including inflammation, fibrosis, angiogenesis, and endothelial function. This combined panel improved prediction of accelerated renal function decline in patients with type 2 diabetes on top of established risk markers. The results of this study need to be validated in a large, prospective cohort to validate and assess its applicability in a broad type 2 diabetes population.

ACKNOWLEDGMENTS

We thank SS Roscioni, University Medical Center Groningen for her contributions to an earlier version of the article, R Datema and W Woloszczuk, Biomarker Design Forschung GmbH, G Mayer, Innsbruck Medical University, R Oberbauer, Medical University of Vienna, and G Stelzer and D Lancet, Weizmann Institute of Science for their help with the selection of biomarkers as a collaborative effort in the SysKid consortium, and J Benner and A Breitwieser, Biomarker Design Forschung GmbH for the measurement of the biomarkers.

FUNDING

The work leading to this paper received funding from the European Community’s Seventh Framework Programme under grant agreement no. HEALTH–F2–2009–241544 (SysKID consortium). The PREDICTIONS Study was supported by the Commission of the European Communities, 6th Framework Programme Priority 1, Life Sciences, Genomics and Biotechnology under grant agreement no. Health LSHM-CT-2005-018733. HJLH is supported by a VENI grant from the Netherlands Organisation for Scientific Research.

4140


2

REFERENCES

1. United States Renal Data System, 2014 Annual Data Report: Epidemiology of Kidney

Disease in the United States. National Institutes of Health, National Institute of Diabetes

and Digestive and Kidney Diseases, Bethesda, MD, 2014. Available: http://www.usrds.

org/adr.aspx.

2. Gregg EW, Li Y, Wang J, Burrows NR, Ali MK, Rolka D, Williams DE, Geiss L. Changes in

diabetes-related complications in the United States, 1990-2010. N Engl J Med. 2014 Apr

17;370(16):1514-23.

3. Tseng CL, Kern EF, Miller DR, Tiwari A, Maney M, Rajan M, Pogach L. Survival benefit of

nephrologic care in patients with diabetes mellitus and chronic kidney disease. Arch Intern

Med. 2008 Jan 14;168(1):55-62.

4. Palmer AJ, Annemans L, Roze S, Lamotte M, Lapuerta P, Chen R, Gabriel S, Carita P,

Rodby RA, de Zeeuw D, Parving HH. Cost-effectiveness of early irbesartan treatment

versus control (standard antihypertensive medications excluding ACE inhibitors, other

angiotensin-2 receptor antagonists, and dihydropyridine calcium channel blockers) or late

irbesartan treatment in patients with type 2 diabetes, hypertension, and renal disease.

Diabetes Care. 2004 Aug;27(8):1897-903.

5. Rondbjerg AK, Omerovic E, Vestergaard H. YKL-40 levels are independently associated

with albuminuria in type 2 diabetes. Cardiovasc Diabetol. 2011 Jun 22;10:54.

6. Niewczas MA, Gohda T, Skupien J, Smiles AM, Walker WH, Rosetti F, Cullere X, Eckfeldt

JH, Doria A, Mayadas TN, Warram JH, Krolewski AS. Circulating TNF receptors 1 and 2

predict ESRD in type 2 diabetes. J Am Soc Nephrol. 2012 Mar;23(3):507-15.

7. Fechete R, Heinzel A, Perco P, Mönks K, Söllner J, Stelzer G, Eder S, Lancet D, Oberbauer

R, Mayer G, Mayer B. Mapping of molecular pathways, biomarkers and drug targets for

diabetic nephropathy. Proteomics Clin Appl. 2011 Jun;5(5-6):354-66.

8. Alkhalaf A, Zürbig P, Bakker SJ, Bilo HJ, Cerna M, Fischer C, Fuchs S, Janssen B, Medek

K, Mischak H, Roob JM, Rossing K, Rossing P, Rychlík I, Sourij H, Tiran B, Winklhofer-Roob

BM, Navis GJ; PREDICTIONS Group. Multicentric validation of proteomic biomarkers in

urine specific for diabetic nephropathy. PLoS One. 2010 Oct 20;5(10):e13421.

9. World Health Organization. Dept. of Noncommunicable Disease Surveillance. Definition,

diagnosis and classification of diabetes mellitus and its complications: report of a WHO

consultation. Part 1, Diagnosis and classification of diabetes mellitus. Geneva, World

Health Organization. 1999. Available: http://apps.who.int/iris/handle/10665/66040.

10. Hellemons ME, Kerschbaum J, Bakker SJ, Neuwirt H, Mayer B, Mayer G, de Zeeuw D,

Lambers Heerspink HJ, Rudnicki M. Validity of biomarkers predicting onset or progression

of nephropathy in patients with type 2 diabetes: a systematic review. Diabet Med. 2012

May;29(5):567-77.

11. Stelzer G, Dalah I, Stein TI, Satanower Y, Rosen N, Nativ N, Oz-Levi D, Olender T, Belinky F,

Bahir I, Krug H, Perco P, Mayer B, Kolker E, Safran M, Lancet D. In-silico human genomics

with GeneCards. Hum Genomics. 2011 Oct;5(6):709-17.

12. Nguyen TQ, Tarnow L, Jorsal A, Oliver N, Roestenberg P, Ito Y, Parving HH, Rossing

P, van Nieuwenhoven FA, Goldschmeding R. Plasma connective tissue growth factor

is an independent predictor of end-stage renal disease and mortality in type 1 diabetic

nephropathy. Diabetes Care. 2008 Jun;31(6):1177-82.

13. R core team. R: A language and environment for statistical computing. R foundation for

statistical computing. Vienna, Austria. 2013. Available: http://www.R-project.org/.

14. van Buuren S, Groothuis-Oudshoorn K. mice: Multivariate imputation by chained

equations in R. J Stat Softw. 2011;45(3):1-67.

15. Friedman J, Hastie T, Tibshirani R. Regularization paths for generalized linear models via

coordinate descent. J Stat Softw. 2010;33(1):1-22.

16. White IR, Royston P, Wood AM. Multiple imputation using chained equations: Issues and

guidance for practice. Stat Med. 2011 Feb 20;30(4):377-99.

17. Levey AS, Bosch JP, Lewis JB, Greene T, Rogers N, Roth D. A more accurate method

to estimate glomerular filtration rate from serum creatinine: a new prediction equation.

Modification of Diet in Renal Disease Study Group. Ann Intern Med. 1999 Mar

16;130(6):461-70.

18. Tibshirani R. Regression shrinkage and selection via the lasso. J R Stat Soc Series B Stat

Methodol 1996;58(1):267-288.

19. Eriksen BO, Ingebretsen OC. The progression of chronic kidney disease: a 10-year

population-based study of the effects of gender and age. Kidney Int. 2006 Jan;69(2):375-82.

20. Shlipak MG, Katz R, Kestenbaum B, Fried LF, Newman AB, Siscovick DS, Stevens L,

Sarnak MJ. Rate of kidney function decline in older adults: a comparison using creatinine

and cystatin C. Am J Nephrol. 2009;30(3):171-8.

21. Morii T, Fujita H, Narita T, Shimotomai T, Fujishima H, Yoshioka N, Imai H, Kakei M, Ito S.

Association of monocyte chemoattractant protein-1 with renal tubular damage in diabetic

nephropathy. J Diabetes Complications. 2003 Jan-Feb;17(1):11-5.

4342


2

22. Tam FW, Riser BL, Meeran K, Rambow J, Pusey CD, Frankel AH. Urinary monocyte

chemoattractant protein-1 (MCP-1) and connective tissue growth factor (CCN2)

as prognostic markers for progression of diabetic nephropathy. Cytokine. 2009

Jul;47(1):37-42.

23. Lee JH, Kim SS, Kim IJ, Song SH, Kim YK, In Kim J, Jeon YK, Kim BH, Kwak IS. Clinical

implication of plasma and urine YKL-40, as a proinflammatory biomarker, on early

stage of nephropathy in type 2 diabetic patients. J Diabetes Complications. 2012 Jul-

Aug;26(4):308-12.

24. Persson F, Rathcke CN, Gall MA, Parving HH, Vestergaard H, Rossing P. High YKL-40

levels predict mortality in patients with type 2 diabetes. Diabetes Res Clin Pract. 2012

Apr;96(1):84-9.

25. Li SY, Huang PH, Yang AH, Tarng DC, Yang WC, Lin CC, Chen JW, Schmid-Schönbein

G, Lin SJ. Matrix metalloproteinase-9 deficiency attenuates diabetic nephropathy by

modulation of podocyte functions and dedifferentiation. Kidney Int. 2014 Aug;86(2):358-69.

26. Ban CR, Twigg SM, Franjic B, Brooks BA, Celermajer D, Yue DK, McLennan SV. Serum

MMP-7 is increased in diabetic renal disease and diabetic diastolic dysfunction. Diabetes

Res Clin Pract. 2010 Mar;87(3):335-41.

27. Baelde HJ, Eikmans M, Lappin DW, Doran PP, Hohenadel D, Brinkkoetter PT, van der

Woude FJ, Waldherr R, Rabelink TJ, de Heer E, Bruijn JA. Reduction of VEGF-A and CTGF

expression in diabetic nephropathy is associated with podocyte loss. Kidney Int. 2007

Apr;71(7):637-45.

28. Thurston G, Rudge JS, Ioffe E, Zhou H, Ross L, Croll SD, Glazer N, Holash J, McDonald

DM, Yancopoulos GD. Angiopoietin-1 protects the adult vasculature against plasma

leakage. Nat Med. 2000 Apr;6(4):460-3.

29. Ward NL, Dumont DJ. The angiopoietins and Tie2/Tek: adding to the complexity of

cardiovascular development. Semin Cell Dev Biol. 2002 Feb;13(1):19-27.

30. Avogaro A, Fadini GP, Gallo A, Pagnin E, de Kreutzenberg S. Endothelial dysfunction in

type 2 diabetes mellitus. Nutr Metab Cardiovasc Dis. 2006 Mar;16 Suppl 1:S39-45.

31. Vlachopoulos C, Ioakeimidis N, Terentes-Printzios D, Aznaouridis K, Baou K, Bratsas A,

Lazaros G, Stefanadis C. Amino-terminal pro-C-type natriuretic peptide is associated with

arterial stiffness, endothelial function and early atherosclerosis. Atherosclerosis. 2010

Aug;211(2):649-55.

32. Levin A, Rigatto C, Barrett B, Madore F, Muirhead N, Holmes D, Clase CM, Tang M,

Djurdjev O; CanPREDDICT Investigators. Biomarkers of inflammation, fibrosis, cardiac

stretch and injury predict death but not renal replacement therapy at 1 year in a canadian

chronic kidney disease cohort. Nephrol Dial Transplant. 2014 May;29(5):1037-47.

33. Agarwal R, Duffin KL, Laska DA, Voelker JR, Breyer MD, Mitchell PG. A prospective study

of multiple protein biomarkers to predict progression in diabetic chronic kidney disease.

Nephrol Dial Transplant. 2014 Dec;29(12):2293-302.

34. Verhave JC, Bouchard J, Goupil R, Pichette V, Brachemi S, Madore F, Troyanov S.Clinical

value of inflammatory urinary biomarkers in overt diabetic nephropathy: a prospective

study. Diabetes Res Clin Pract. 2013 Sep;101(3):333-40.



in the irbesartan in patients with type 2 diabetes and microalbuminuria (IRMA 2) study.

Scand J Clin Lab Invest. 2008;68(8):731-8.

36. Good DM, Zürbig P, Argilés A, Bauer HW, Behrens G, Coon JJ, Dakna M, Decramer

S, Delles C, Dominiczak AF, Ehrich JH, Eitner F, Fliser D, Frommberger M, Ganser A,

Girolami MA, Golovko I, Gwinner W, Haubitz M, Herget-Rosenthal S, Jankowski J, Jahn

H, Jerums G, Julian BA, Kellmann M, Kliem V, Kolch W, Krolewski AS, Luppi M, Massy Z,

Melter M, Neusüss C, Novak J, Peter K, Rossing K, Rupprecht H, Schanstra JP, Schiffer

E, Stolzenburg JU, Tarnow L, Theodorescu D, Thongboonkerd V, Vanholder R, Weissinger

EM, Mischak H, Schmitt-Kopplin P. Naturally occurring human urinary peptides for use in

diagnosis of chronic kidney disease. Mol Cell Proteomics. 2010 Nov;9(11):2424-37.

37. Roscioni SS, de Zeeuw D, Hellemons ME, Mischak H, Zürbig P, Bakker SJ, Gansevoort

RT, Reinhard H, Persson F, Lajer M, Rossing P, Lambers Heerspink HJ. A urinary peptide

biomarker set predicts worsening of albuminuria in type 2 diabetes mellitus. Diabetologia.

2013 Feb;56(2):259-67.

38. Sharma K, Karl B, Mathew AV, Gangoiti JA, Wassel CL, Saito R, Pu M, Sharma S, You YH,

Wang L, Diamond-Stanic M, Lindenmeyer MT, Forsblom C, Wu W, Ix JH, Ideker T, Kopp

JB, Nigam SK, Cohen CD, Groop PH, Barshop BA, Natarajan L, Nyhan WL, Naviaux RK.

Metabolomics reveals signature of mitochondrial dysfunction in diabetic kidney disease.

J Am Soc Nephrol. 2013 Nov;24(11):1901-12.J Am Soc Nephrol. 2013;24: 1901-1912.

39. Pena MJ, Lambers Heerspink HJ, Hellemons ME, Friedrich T, Dallmann G, Lajer M, Bakker

SJ, Gansevoort RT, Rossing P, de Zeeuw D, Roscioni SS. Urine and plasma metabolites

predict the development of diabetic nephropathy in individuals with type 2 diabetes

mellitus. Diabet Med. 2014 Sep;31(9):1138-47.

40. Ninomiya T, Perkovic V, de Galan BE, Zoungas S, Pillai A, Jardine M, Patel A, Cass A,

Neal B, Poulter N, Mogensen CE, Cooper M, Marre M, Williams B, Hamet P, Mancia G,

Woodward M, Macmahon S, Chalmers J; ADVANCE Collaborative Group. Albuminuria

and kidney function independently predict cardiovascular and renal outcomes in diabetes.

J Am Soc Nephrol. 2009 Aug;20(8):1813-21.

4544


2

41. Babazono T, Nyumura I, Toya K, Hayashi T, Ohta M, Suzuki K, Kiuchi Y, Iwamoto Y.

Higher levels of urinary albumin excretion within the normal range predict faster decline

in glomerular filtration rate in diabetic patients. Diabetes Care. 2009 Aug;32(8):1518-20.

42. Hellemons ME, Persson F, Bakker SJ, Rossing P, Parving HH, de Zeeuw D, Lambers

Heerspink HJ. Initial angiotensin receptor blockade-induced decrease in albuminuria is

associated with long-term renal outcome in type 2 diabetic patients with microalbuminuria:

a post hoc analysis of the IRMA-2 trial. Diabetes Care. 2011 Sep;34(9):2078-83.

43. Roscioni SS, Lambers Heerspink HJ, de Zeeuw D. Microalbuminuria: Target for

renoprotective therapy PRO. Kidney Int. 2014 Jul;86(1):40-9.

44. Remuzzi G, Bertani T. Is glomerulosclerosis a consequence of altered glomerular

permeability to macromolecules? Kidney Int. 1990 Sep;38(3):384-94.

45. Abbate M, Zoja C, Remuzzi G. How does proteinuria cause progressive renal damage? J

Am Soc Nephrol. 2006 Nov;17(11):2974-84.

46. Steyerberg EW. Validation of prediction models. In: Gail M, Krickeberg K, Samet J, Tsiatis

A, Wong W, editors. Clinical Prediction Models: A Practical Approach to Development,

Validation, and Updating. New York: Springer US; 2009. pp. 299-311.

SUPPLEMENTAL APPENDIX 1. SELECTION OF BIOMARKERS

Biomarker candidates were selected following three strategies in the realm of integrative evaluation, namely a manual consolidation effort and two algorithmic approaches. Common denominator of the procedures was evaluation of a candidate’s relevance in the context of molecular processes and pathways of putative relevance in diabetic nephropathy. Respective molecular pathways were extracted from the title and abstracts of publications retrieved with the following NCBI PubMed query: (“diabetic nephropathy”[ti] OR “diabetic nephropathies”[ti]) AND (pathway[ti] OR pathways[ti]).

Manual consolidation of biomarker candidates started with a literature review resting on PubMed MeSH annotation “biological markers” [mh] and “Diabetic Nephropathies” [mh]. Utilizing mechanistic context literature reviewing based on NCBI GeneRIF (ftp://ftp.ncbi.nih.gov/gene/GeneRIF/) and tissue specific expression information retrieved from a gene expression dataset on 32 human tissues (http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE7905) as well as from abundance information available in GeneCards (http://www.genecards.org) twelve biomarker candidates were selected for experimental evaluation utilizing ELISA-based assays: zinc-binding alpha-2-glycoprotein 1 (AZGP1); endostatin (fragment of COL18A1); connective tissue growth factor (CTGF); fibroblast growth factor 23 (FGF23); galectin-3 (LGALS3); monocyte chemoattractant protein-1 (CCL2); nephrin (NPHS1); podocin (NPHS2); neuropilin-1 (NRP1); amino terminal pro C-type natriuretic peptide (NT-proCNP); sclerostin (SOST), and tumor necrosis factor receptor 2 (TNFR2).

The second biomarker selection procedure utilized a molecular model representation of diabetic nephropathy [1]. In brief, the methodology consolidates Omics signatures associated with diabetic nephropathy on a molecular interaction network, and segments the induced subgraph into molecular units utilizing topological criteria. Based on this approach the following biomarker candidates were selected for multiplexed (Luminex-based) analysis: chitinase 3-like 1 (YKL-40); chemokine (C-X-C motif) 1 (CXCL1); chemokine (C-X-C motif) 10 (CXCL10); epidermal growth factor (EGF); growth hormone 1 (GH1); hepatocyte growth factor (HGF); interleukin 1 beta (IL1B); leptin (LEP); a set of matrix metallopeptidases (MMP13, MMP2, MMP7, MMP8); tyrosine kinase (TEK); tumor necrosis factor receptor superfamily, member 1A (TNFR1), and vascular endothelial growth factor A (VEGF-A). This selection procedure also included CCL2 already being derived as candidate from manual consolidation.

The third biomarker selection procedure rested on a ranking procedure of consolidated Omics signatures utilizing annotation from Genecards [2]. Next to the candidates CXCL1, CXCL10, EGF, IL1B and MMP2 also selected based on the molecular model analysis, CCL2 selected on the basis of the molecular model as well as on the basis of manual consolidation, and endostatin also selected based on manual evaluation, Interleukin 1 alpha (IL1A) and matrix metallopeptidases 1 (MMP1) were included as additional candidates for experimental evaluation.

4746


2

References. Supplemental Appendix 11. Heinzel A, Muhlberger I, Fechete R, Mayer B, Perco P. Functional molecular units for

guiding biomarker panel design. Methods Mol Biol. 2014;1159:109-33.

2. Stelzer G, Dalah I, Stein TI, Satanower Y, Rosen N, Nativ N, Oz-Levi D, Olender T, Belinky F,

Bahir I, Krug H, Perco P, Mayer B, Kolker E, Safran M, Lancet D. In-silico human genomics

with GeneCards. Hum Genomics. 2011 Oct;5(6):709-17.

SUPPLEMENTAL APPENDIX 2. BIOMARKER ASSAY INFORMATION

At the time of the study, most assays were available directly from the manufacturer, except for FGF23 and Endostatin (Biomedica/Biomarker Design Forschungs GmbH (BDF), Vienna, Austria) which were under development and validation according to the International Conference of Harmonization (ICH). Both assays met all ICH standards.

Regarding the limit of detection, for Biomedica/BDF assays, the Lower Limit of Detection (LLD) was defined as the calculated concentration of the lowest standard point plus three times the standard deviation. For Fibrogen (FibroGen Inc., San Francisco, USA), the detection limit of the CTGF assay was 4 pmol/l, and intra- and interassay variations were 6 and 20%, respectively. For R&D Systems (R&D Systems, Minneapolis, USA) assays, the Minimum Discriminatory Difference (MDD) was determined by adding two standard deviations to the mean optical density value of twenty zero standard replicates and calculating the corresponding concentration. For USCN (USCN Life Science Inc., Wuhan, Hubei, China) assays, the LLD was defined as the lowest protein concentration that could be differentiated from zero. It was determined by adding two standard deviations to the mean optical density value of twenty zero standard replicates and calculating the corresponding concentration.

Regarding stability of the biomarkers, we have no information about long term stability. USCN recommends to avoid repeated freeze/thaw cycles. R&D Systems and Biomedica/BDF have tested the freeze/thaw stability and state that the analytes are stable for at least 3 freeze/thaw cycles.

Biomarker Assay Manufacturer

Endostatin (Frag.COL18A1) ELISA Biomedica/BDF

Amino terminal pro C-type natriuretic peptide (NT-proCNP) ELISA Biomedica/BDF

Fibroblast growth factor 23 (FGF23) ELISA Biomedica/BDF

Sclerostin (SOST) ELISA Biomedica/BDF

Connective tissue growth factor (CTGF) ELISA Fibrogen

Podocin (NPHS2) ELISA USCN

Zinc-binding alpha-2-glycoprotein 1 (AZGP1) ELISA USCN

Nephrin (NPHS1) ELISA USCN

Neuropilin-1 (NRP1) ELISA USCN

Tumor necrosis factor receptor-2 (TNFR2) ELISA R&D Systems

Monocyte chemoattractant protein-1 (CCL2) Multiplex R&D Systems

Tumor necrosis factor receptor-1 (TNFR1) Multiplex R&D Systems

Chitinase 3-like 1 (YKL-40) (ng/mL) Multiplex R&D Systems

Chemokine (C-X-C motif) 1 (CXCL1) Multiplex R&D Systems

Chemokine (C-X-C motif) 10 (CXCL10) Multiplex R&D Systems

Matrix metallopeptidase 1 (MMP1) Multiplex R&D Systems





Leptin (LEP) Multiplex R&D Systems

Tyrosine kinase (TEK) Multiplex R&D Systems

Vascular endothelial growth factor-A (VEGF-A) Multiplex R&D Systems

Hepatocyte growth factor (HGF) Multiplex R&D Systems

Growth hormone 1 (GH1) Multiplex R&D Systems

Interleukin-1 alpha (IL1A) Multiplex R&D Systems

Interleukin-1 beta (IL1B) Multiplex R&D Systems

Epidermal growth factor (EGF) Multiplex R&D Systems

4948


2

SUPPLEMENTAL APPENDIX 3. EXTENDED STATISTICAL METHODS

a. Pooling of results across the 20 imputed data setsFor univariate or single-biomarker models, we applied Rubin’s rules to pool results across the 20 imputed data sets [1].

To avoid the LASSO to select different variables in each of the imputed data set, we estimated regression coefficients after pooling the imputed data sets into one ‘stacked’ data set. Since the LASSO does not involve the estimation of standard errors of regression coefficients, this is a valid approach to estimate the posterior modes of the regression coefficients, since in standard linear regression analyses posterior modes obtained by proper pooling and by a ‘stacked data’ analysis are approximately similar.

However, when the tuning parameter which controls the amount of restriction in the LASSO is optimized by cross-validation, the dependency of multiple data lines contributed by a single patient must be accounted for. Thus, our implementation of cross-validation involved leaving out all 20 imputed data lines of a patient simultaneously, and repeating this for each patient in turn. We validated this approach by comparing the such-optimized tuning parameter with the distribution of optimized tuning parameters obtained when each imputed data is analyzed separately. In addition, we compared the final number of selected variables in the pooled analysis with the distribution of selected variables across the 20 individual analyses (data not shown). We became aware of the existence of more sophisticated ways of applying the LASSO to multiply imputed data only after completing data analysis of this project [2].

b. Computation of predicted probabilities of accelerated eGFR function declineFrom a predicted eGFR slope (Y) and its standard error for individual prediction (S), we derived the probability of accelerated eGFR decline (P), defined as a eGFR slope ≤ -3 mL/min/1.73m2/year, as follows, making use of a normal approximation: P=F({-3-Y}/S), where F(z) denotes the cumulative distribution function of the standard normal distribution, evaluated at z. S was defined as the bootstrap estimate of the root mean squared error of prediction. The probabilities P are shown and compared between patients with accelerated and non-accelerated eGFR decline in Figure 2.

References. Supplemental Appendix 31. White IR, Royston P, Wood AM. Multiple imputation using chained equations: Issues and

guidance for practice. Stat Med. 2011 Feb 20;30(4):377-99.

2. Chen Q, Wang S. Variable selection for multiply-imputed data with application to dioxin

exposure study. Stat Med. 2013 Sep 20;32(21):3646-59.

Sup

ple

men

tal T

able

1. I

mp

act

of o

mitt

ing

varia

ble

s fr

om t

he m

odel

. Col

umn

varia

ble

s ar

e th

e ne

wly

incl

uded

var

iab

le t

o th

e m

odel

with

the

ir st

and

ard

ized

b

etas

if t

he v

aria

ble

in t

he r

ow w

as o

mitt

ed f

rom

the

ful

l mod

el. S

tand

ard

ized

bet

as r

epre

sent

per

cent

cha

nge

in e

GFR

per

sta

ndar

d d

evia

tion

incr

emen

t in

b

iom

arke

r le

vel.

Elim

inat

ing

the

varia

ble

s b

asel

ine

eGFR

, the

use

of o

ral d

iab

etic

med

icat

ion,

NT-

pro

CN

P o

r C

CL2

did

not

incl

ude

any

new

pre

dic

tors

to

the

mod

el.

Onl

y th

e om

issi

on o

f b

asel

ine

UA

CR

or

syst

olic

blo

od p

ress

ure

resu

lted

in r

elev

ant

incl

usio

ns o

f ot

her

nove

l bio

mar

kers

into

the

mod

el (d

efine

d a

s in

clus

ions

of p

red

icto

rs w

ith a

sta

ndar

diz

ed c

oeffi

cien

t of

>5%

).

DB

PV

EG

F-A

GH

1LE

PH

GF

CX

CL1

CX

CL1

0T

NFR

2A

ZG

P1

FGF2

3Fr

ag.C

OL1

8A1

Bas

elin

e U

AC

R0.

067.

31.

15.

95.

317

.65.

20.

018

.69.

70.

0

Bas

elin

e eG

FR0.

00.

00.

00.

00.

00.

00.

00.

00.

00.

00.

0

CTG

F0.

00.

00.

50.

00.

00.

00.

00.

00.

00.

00.

0

TEK

0.0

0.0

1.0

0.0

2.0

0.0

0.0

0.0

0.0

0.0

0.0

Sys

tolic

blo

od p

ress

ure

0.8

3.1

0.8

0.0

7.1

1.5

0.3

0.0

8.7

0.0

0.0

MM

P2

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

2.5

0.0

MM

P7

0.0

0.0

0.0

0.0

1.0

0.0

0.0

0.0

0.0

0.0

0.0

Cur

rent

vs.

nev

er s

mok

er0.

00.

00.

04.

20.

20.

02.

50.

00.

00.

00.

0

TNFR

10.

00.

00.

80.

01.

10.

00.

00.

00.

00.

00.

0

MM

P1

0.0

0.0

0.0

0.0

0.8

0.0

0.0

0.0

0.0

0.0

0.0

MM

P8

0.0

0.0

0.4

1.4

00.

00.

20.

00.

00.

00.

0

NP

HS

20.

00.

00.

00.

10.

50.

00.

00.

00.

00.

00.

0

Ora

l dia

bet

ic m

edic

atio

n0.

00.

00.

00.

00.

00.

00.

00.

00.

00.

00.

0

Sex

0.0

0.0

0.0

3.0

0.5

0.0

0.0

0.0

0.0

0.0

0.0

SO

ST

0.0

0.0

0.0

1.4

0.3

0.0

0.0

0.0

0.0

0.0

0.0

YLK

-40

0.0

0.0

0.0

0.0

1.0

0.0

0.0

0.0

0.0

0.0

0.0

NT-

pro

CN

P0.

00.

00.

00.

00.

00.

00.

00.

00.

00.

00.

0

CC

L20.

00.

00.

00.

00.

00.

00.

00.

00.

00.

00.

0

MM

P13

0.0

0.0

0.0

0.2

0.3

0.0

0.0

0.0

0.0

0.0

0.0

5150


2

Supplemental Figure 1. LASSO selection of established risk markers: cross validated mean squared error (Y-axis; black bullets; MSE) vs. amount of restriction (X-axis; log(Lambda)). Vertical bars refer to standard errors across the 82 cross-validations. Predictive accuracy reached a plateau after step 21 (21st bullet from the right). At this step, the following established risk markers were selected: baseline UACR, current vs. never smoker, sex, systolic and diastolic blood pressure, use of oral diabetic medication, and baseline eGFR.

CHAPTER 3Plasma proteomic classifiers improve risk prediction for renal disease in patients with hypertension or type 2 diabetes

MJ PenaJ Jankowski

G HeinzeM Kohl

A HeinzelSJL Bakker

RT GansevoortP Rossing

D de ZeeuwHJ Lambers Heerspink

V Jankowski

Journal of Hypertension 2015 Jul 31. [Epub ahead of print]

5554

Plasma proteomics predict transition in stage of albuminuriaChapter 3

3

ABSTRACT

Objectives: Micro- and macroalbuminuria are strong risk factors for progression of nephropathy in patients with hypertension or type 2 diabetes patients. Early detection of progression to micro- and macroalbuminuria may facilitate prevention and treatment of renal diseases. We aimed to develop plasma proteomics classifiers to predict the development of micro- or macroalbuminuria in hypertension and type 2 diabetes.

Methods: Patients with hypertension (n=125) and patients with type 2 diabetes (n=82) were selected for this case-control study from the PREVEND cohort and the Steno Diabetes Center. Cases transitioned from normo- to microalbuminuria or micro- to macroalbuminuria. Controls, matched for age, gender, and baseline albuminuria stage, did not transition. Follow-up was 3.0±0.9 years. Plasma proteomics profiles were measured by LC-electrospray-trap mass-spectrometry. Classifiers were developed and cross-validated for prediction of transition in albuminuria stage. Improvement in risk prediction was tested on top of a reference model of baseline albuminuria, eGFR, and renin-angiotensin-aldosterone system intervention.

Results: In hypertensive patients, the classifier improved risk prediction for transition in albuminuria stage on top of the reference model (C-index from 0.69 to 0.78; p<0.01). In type 2 diabetes, the classifier improved risk prediction for transition from micro- to macroalbuminuria (C-index from 0.73 to 0.80; p=0.04). In both diseases, the identified peptides were linked to pathways recognized to contribute to nephropathy, including fibrosis, inflammation, angiogenesis, and mineral metabolism.

Conclusions: Plasma proteomics predict the transition in albuminuria stage beyond established renal risk markers in hypertension or type 2 diabetes. External validation is needed to assess reproducibility.

INTRODUCTION

Hypertension and type 2 diabetes are the leading causes of progressive chronic kidney disease (CKD) [1]. Microalbuminuria (urinary albumin excretion (UAE) between 30 and 300 mg/day) is considered a first sign of kidney dysfunction, and is an established and important predictor of progressive renal function loss in both hypertension and diabetes [2, 3].

Despite established therapies including renin-angiotensin-aldosterone system (RAAS) interventions, a substantial number of patients remain at risk for progressive loss of renal function [4, 5]. An explanation of this high residual renal risk might be that intervention is initiated after significant damage has already been established. Indeed, in the setting of RAAS intervention (RAASi), most risk reduction is observed when it is started in early stages of disease [6]. Thus, early detection of patients at risk to develop elevated albuminuria and subsequent initiation of appropriate treatment may be an efficacious approach to further reduce the incidence of end-stage renal disease (ESRD).

The identification of yet unknown biomarkers improving prediction of transitioning from normo- to microalbuminuria is currently a research priority [7]. Comprehensive analysis of peptides in biological fluids, known as peptidomics [8], has emerged as a tool to discover novel biomarkers at early stages of disease [9, 10]. Multiple peptides combined into one risk score may provide a tool to help classify which patients are at risk for progressive renal disease.

Therefore, the aim of the current study was the identification of plasma peptides associated with transitioning in stage of albuminuria and to develop two classifiers, one for hypertension and another for type 2 diabetes, to predict the transition of stage of albuminuria.

5756


3

METHODS

This case-control study was performed utilizing plasma samples from patients participating in the community-based “Prevention of REnal and Vascular ENd-stage Disease” (PREVEND - http://www.prevend.org/) study and from patients seen at the Steno Diabetes Center. The ongoing PREVEND study was initiated in Groningen, the Netherlands, in 1997, to study the natural course of urinary albumin excretion with serial follow-up measurements [11]. At the Steno Diabetes Center in Gentofte, Denmark, a subgroup of patients was identified from a study initiated in 2007, which follows 200 patients with type 2 diabetes with elevated urinary albumin excretion (UAE) (>30 mg/day), normal plasma creatinine, and no known history of cardiovascular disease on a yearly basis [12]. Both the PREVEND and Steno Diabetes Center studies were approved by local ethics committees and were conducted in accordance with the guidelines of the Declaration of Helsinki. All patients gave written, informed consent.

Hypertensive patientsHypertensive patients were drawn from the PREVEND study. Hypertension was defined as the use of antihypertensive treatment (self-reported or by information retrieved from the regional pharmacy database) or a systolic/diastolic blood pressure >140/90 mmHg at baseline.

Type 2 diabetes patientsPatients with type 2 diabetes at baseline were drawn from both the PREVEND and the Steno Diabetes Center cohorts. For PREVEND patients, individuals with type 2 diabetes, defined as the use of oral glucose-lowering treatment (self-reported or by information retrieved from the regional pharmacy database), a fasting plasma glucose >7.0 mmol/l (126 mg/dl) or non-fasting plasma glucose >11.1 mmol/l (>200 mg/dl) were eligible for the present study [13]. For the Steno Diabetes Center patients, type 2 diabetes was defined according to similar criteria.

Selection of cases and controlsCases were defined as patients who, had any progression in albuminuria stage (transition from normo- to microalbuminuria or from micro- to macroalbuminuria) with at least a 30% increase in UAE from baseline between two consecutive study visits. Controls were selected from patients who had stable normo- or microalbuminuria, respectively, during the same follow-up period. Cases and controls were matched for age, gender, baseline albuminuria class, and for the patients from the Steno Diabetes Center, also on duration of diabetes. For the patients from the PREVEND study, a case and a matched control had to be in the same baseline albuminuria stage during the same study visit period. Normoalbuminuria was defined as UAE <30 mg/24hr, microalbuminuria as UAE between 30 to 300 mg/24hr, and macroalbuminuria as UAE ≥300 mg/24hr. The use of anti-hypertensive RAASi medication was allowed, but the type of drug and its dose had to remain stable during the study period. Patients who discontinued RAASi medication or initiated new treatment during the follow-up period were excluded from this study.

In hypertension, 50 normo- to microalbuminuria case/control pairs and 25 micro- to macroalbuminuria case/control pairs were selected for biomarker discovery. However, due to insufficient sample availability, proteomics analysis was not possible for all samples. Therefore, plasma samples with sufficient quantity and quality were performed in 125 patients with hypertension (35 complete pairs in normo- microalbuminuria and 17 complete pairs in micro- to macroalbuminuria, and 14 patients in normo- to microalbuminuria and 7 patients in micro- to macroalbuminuria without a partner). In type 2 diabetes 24 normo- to microalbuminuria case/controls pairs and 21 micro- to macroalbuminuria case/controls pairs were selected for biomarker discovery. Due to insufficient sample availability, data from plasma proteomics analysis was not possible for all samples. Plasma samples with sufficient quantity and quality were available for 82 patients with type 2 diabetes (17 complete pairs in normo- microalbuminuria and 18 complete pairs in micro- to macroalbuminuria, and 8 patients in normo- to microalbuminuria and 4 patients in micro- to macroalbuminuria without a partner). The blood samples were centrifuged at 4°C directly after collection. The plasma was afterwards stored at a controlled –80°C. In the year 2010, the samples were thawed and divided into smaller aliquots, refrozen at –80°C before thawing for proteomic measurements. For the PREVEND study, samples were stored for median 6.8 [range 2.7 – 13.0] years. Samples from the Steno Diabetes Center were stored for approximately 3 years.

MeasurementsClinical, plasma, and urine chemistry measurements in PREVEND and at the Steno Diabetes Center were performed as previously described [11, 12].

Briefly, in the PREVEND study, during 2 outpatient office visits, blood pressure was measured in the supine position, every minute, over 10 and 8 minutes, respectively, with an automatic device (Dinamap XLModel 9300; Johnson-Johnson Medical Inc., Arlington, TX, USA). Systolic and diastolic blood pressure was calculated as the mean of the last two measurements of the two visits. Albuminuria status was based on the average of two consecutive-day measurements in 24hr urine collections, which was measured by nephelometry with a threshold of 2.3 mg/L and intra- and interassay coefficients of variation of 4.3% and 4.4%, respectively (Siemens, Munich, Germany). Plasma glucose and serum and urinary creatinine were measured by dry chemistry (Eastman Kodak, Rochester, New York, USA). For serum creatinine, the intra-assay coefficient of variation was 0.9% and inter-assay coefficient of variation was 2.9%.

In patients from the Steno Diabetes Center, blood pressure was measured in sitting position as the mean of three oscillometric measurements with an electronic blood pressure device (UA 779; A&D Instruments Ltd, Abingdon, UK). Albuminuria was determined by the average of three consecutive-day 24hr urine collections (Turbidimetry, Hitachi 912 system; Roche Diagnostics, Mannheim, Germany). Hemoglobin A1c (HbA1c) was measured by high-performance liquid chromatography (normal range: 4.1–6.4%, Bio-Rad Laboratories, Munich, Germany) [14] and serum creatinine concentration by an enzymatic method (Hitachi 912; Roche Diagnostics, Mannheim, Germany).

5958


3

For patients from both PREVEND and the Steno Diabetes Center, estimated glomerular filtration rate (eGFR) was estimated using the Modification of Diet in Renal Disease (MDRD) study equation, using gender, age, race, and serum creatinine [15].

Proteomics analysisThe plasma samples were analyzed at University Hospital RWTH, Institute for Molecular Cardiovascular Research (IMCAR) (Aachen, Germany), using LC-electrospray-trap mass-spectrometry. LC-MS-grade water was purchased from LAB-SCAN (Gliwice, Poland). LC-MS-grade acetonitrile was obtained from Honeywell (Seelze, Germany). HPLC-grade formic acid was purchased from Merck (Darmstadt, Germany). All other chemicals were obtained from Sigma-Aldrich (Hamburg, Germany).

Sample PreparationThe plasma samples were thawed and centrifuged at 13,000×g for 2 minutes, in order to remove denatured proteins. An aliquot of 400µL was transferred into a new sample tube and 1µg of the internal standard ([Sar 1, Thr 8]- angiotensin II) was added to each sample [16]. Next, the samples were deproteinized by using perchloric acid. After addition of the acid, the samples were vortexed for 30 seconds and centrifuged at 8,000×g at 4°C for 2 minutes. The supernatant was transferred into a new sample tube, and the pH value was adjusted at 9 or higher using 15M potassium hydroxide. The samples were vortexed for 30 seconds and frozen at –80°C for 24hr. After thawing, the samples were centrifuged at 3000×g at 4°C for 10 minutes. The supernatant was transferred into a new sample tube and stored at –80°C until fractionation.

Sample Fractionation Using Reverse-Phase-ChromatographyThe samples were diluted in 5mL 0.2% formic acid and transferred into a sample loop for fractionation. The separation was performed on C18-“Chromolith”® Performance Reversed-Phase column (100mm×4.6mm Merck, Darmstadt, Germany). The mobile phase consisted of 0.2% formic acid in water (v/v) (eluent A) and 80% ethanol in water (v/v) (eluent B). The flow rate was 1mL/min. The total run time was 37.5 min. The following step gradient was used for the separation: 0-6 min 0%B, 6-14 min 20%B, 14-22 min 40%B, 22-27 min 60%B, 27-29.5 min 100%B, 29.5-37.5 min 60%B. The samples were detected by using UV absorbance at 280nm. The 20% step-fraction, which contained the hydrophilic peptides of interest, was collected and lyophilized to dryness and used for further analysis.

Sample Analysis Using Electrospray Ionization Source-Liquid Chromatography Mass

SpectrometryAll analyses were performed using an Agilent 1200 series HPLC (Böblingen, Germany), a liquid chromatographic system, interfaced to HTC mass spectrometer (Bruker-Daltonic, Bremen, Germany), equipped with an electrospray ionization source. The lyophilized sample (20% step-fraction) was re-suspended in 50µL 0.1% formic acid and 2µL were injected on the column for analysis. The samples were separated on a Zorbax SB C18 Aq column (150mm×0.5mm, 5µm, Agilent Technologies, Santa Clara, USA). The column temperature was maintained at 50°C. The mobile phase consisted of 0.1% formic acid in

water (v/v) (eluent A) and 0.1% formic acid in acetonitrile (v/v) (eluent B). The flow rate was 60µL/min. The following gradient was used for the separation: 0-8.0 min 0-30%B, 8.0-11.5 min 30-100%B, 11.5-13.5 min 100%B , 13.5–14.0 min 100-0%B, 14.0-20.0 min 0%B. The electrospray ionization source was set in positive ion mode. The ion spray voltage was set to 4,000V. The dry temperature was set at 300°C and the dry gas at 8 l/min. The nebulizer gas was set at 20psi. The maximal accumulation time was set to 200ms. The mass spectrometer was tuned on m/z 800 in wide mode. The mass spectrometer was operated in enhanced mode to achieve maximal resolution and mass accuracy. The scan range was 100-1500 m/z. All data were acquired and processed using Compass 1.3 Software (Bruker-Daltonik, Bremen, Germany).

Data Pre-ProcessingAll acquired LC-MS raw data was pre-processed by using “Data-analysis” software (Bruker-Daltonik, Bremen, Germany). Therefore, an algorithm was used to select high quality signals (molecular features). The parameters used for filtering the data were as follows: signal-to-noise ratio: 3; correlation coefficient threshold: 0.7; minimum compound length: 10 spectra; smoothing width: 2. Furthermore, a spectral background subtraction was performed. To further simplify the data set for statistical analysis, the data was transferred to buckets using following parameters: 0.3 seconds and 0.3 m/z by using an profile analysis software (Bruker-Daltonik, Bremen, Germany). The data were normalized using the internal standard.

Statistical analysisAnalyses were performed with SAS software (version 9.3; SAS Institute, Cary, NC) and R version 2.12.2 (R Core Team 2012, Vienna, Austria) using the package glmnet [17]. A logistic regression analysis of a case-control study to evaluate the prognostic importance of a cross-validated classifier score with 50 cases and 50 controls will have approximately 80% power to detect an odds ratio of 2 per standard deviation of the classifier at a two-sided significance level of 5%, if adjusted for another strong confounding variable (odds ratio of 2 per standard deviation) which exhibits a correlation coefficient of 0.3 with the classifier. Given the assumptions used for sample size calculation, we would expect a C-index of 0.66 for the confounder-only model, and a C-index of 0.75 for confounder plus classifier model [18]. Due to some samples missing, the achieved power was 88% in hypertension and 70% in type 2 diabetes, given the same assumptions as the power calculation.

Data are presented as mean and standard deviation for normal distributions, characteristics with skewed distribution are reported as median and 1st and 3rd quartiles, and categorical variables are reported as frequencies and percentages. Logarithmic transformation of UAE and peptide values was used to normalize their distributions and were used in all regression analyses. Differences between cases and controls were tested with independent samples t-tests for continuous variables and χ² tests for categorical variables. All p-values were two-tailed, and values <0.05 were considered statistically significant.

6160


3

A classifier for hypertension and a separate classifier for type 2 diabetes, consisting of multiple molecular features combined into one risk score, were generated with the following steps. Mass-signals were extracted from about 10,500 mass-spectra accumulated from each group. Identifiable molecular features with a minimum of 12.5% non-zero values were selected for further analysis. Least absolute shrinkage and selection operator (LASSO) logistic regression models were fitted separately to the hypertensive and diabetic patient cohorts to select molecular features and estimate regression coefficients for the selected features [19]. Plasma-proteomic classifier scores were computed for each patient by internal leave-one-pair-out cross-validation. The hypertensive classifier was developed and cross-validated only within hypertension, and likewise, the type 2 diabetes classifier was developed and cross-validated only within type 2 diabetes. More details about development of the classifier can be found in Supplemental Appendix 1.

Prediction analysis was then performed to determine whether the plasma proteomics classifier improved risk prediction for transition of stage of albuminuria. We used unconditional logistic regression since, despite matched case-control sampling, there were some patients without a partner because proteomics measurement was not possible for some samples. We first fitted a reference model including baseline UAE, baseline eGFR and RAASi. Controlling for baseline UAE and baseline eGFR served to equalize residual differences between cases and controls in these variables persisting even after matching. In addition, we adjusted for use of RAASi medication at baseline since the number of patients on these drugs in the type 2 diabetes group was imbalanced between cases and controls. The differences in the C-index (p-value based on the likelihood ratio test) and the integrated discrimination improvement (IDI) between the reference model and a model additionally including the cross-validated plasma proteomics classifier were calculated. Because there were no significant differences between cases and controls in other clinical risk factors including ethnicity, body mass index (BMI), blood pressure, duration of diabetes, follow-up time, total cholesterol, fasting glucose, and HbA1c, these variables were not considered in prediction models.

Peptide and pathway identification The molecular features selected for the classifiers were further identified by matrix-assisted laser desorption/ionization (MALDI) time-of-flight/time-of-flight (TOF/TOF) mass-spectrometry, revaluated using Fourier transform-Orbitrap mass-spectrometry, and identified using the profile analysis approach. Identified molecular features were then mapped to pathways. Pathways were retrieved from PANTHER Pathway and the Kyoto Encyclopedia of Genes and Genomes. For metabolites, relevant enzymes were retrieved from the Human Metabolome Database, and these proteins were used for pathway retrieval. Based on literature mining, the resulting pathway list was divided into “nephropathy relevant” and “probably not nephropathy relevant” pathways.

RESULTS

Hypertension

Baseline characteristics The baseline characteristics of patients with hypertension (n=125) stratified by baseline stage of albuminuria (normo- or microalbuminuria) are shown in Table 1. We observed a significant difference in the baseline UAE level between hypertensive cases and controls despite the matching for albuminuria stage (Table 1). Furthermore, eGFR was lower in hypertensive cases compared to controls, but only in patients with macroalbuminuria at baseline this difference was statistically significant (Table 1). There were no significant differences in other baseline characteristics. The follow-up period for cases and controls was 3.1 ± 0.8 and 2.8 ± 0.9 years, respectively.

Hypertensive patients who transitioned from normo- to microalbuminuria or from micro- to macroalbuminuria had a median percentage increase in albuminuria of 281.9% [129.0, 526.9] and 373.4% [188.7, 717.9], respectively. In hypertensive controls with baseline normo- or microalbuminuria, median changes in albuminuria were -0.2% [-22.4, 18.5] and 37.8% [-1.3, 63.2], respectively (Table 1).

Overall performance of the model in hypertensionIn hypertension, 37 molecular features were selected for the hypertension classifier (Supplemental Table 1). The values of the hypertension classifier were significantly higher in cases than in controls (0.97 ± 3.09 vs. -1.40 ± 3.30, p<0.01) (Figure 1a). The hypertension classifier was independently associated with transitioning in stage of albuminuria (Odds ratio 1.29, 95% CI=1.13-1.48, p<0.01). Additionally, the hypertension classifier improved risk prediction for transitioning in stage of albuminuria on top of the baseline model (C-index from 0.69 to 0.78, p<0.01; IDI 0.06, p=0.03) (Figure 1b). We did not identify effect modification by baseline albuminuria stage, suggesting that the predictive performance of the classifier was similar in both normo- to microalbuminuria and micro- to macroalbuminuria subgroups.

Type 2 diabetes

Baseline characteristicsThe baseline characteristics patients with type 2 diabetes (n=82) stratified by baseline stage of albuminuria (normo- or microalbuminuria) are shown in Table 2. No differences were observed in baseline characteristics between cases and controls except in baseline UAE level between cases and controls (Table 2). There were no differences in eGFR between cases and controls (Table 2). The follow-up period for cases and controls was 3.1 ± 1.0 and 2.9 ± 0.8 years, respectively.

Patients with type 2 diabetes who transitioned from normo- to microalbuminuria and from micro- to macroalbuminuria had a median percentage increase in albuminuria of 122.5% [101.1, 206.6] and 394.9% [200.3, 891.6], respectively. Type 2 diabetes controls with baseline normo- or microalbuminuria had a median percentage change in albuminuria of 1.2% [-17.8, 22.1] and -11.6% [-35.8, 69.7], respectively.

6362


3

Tab

le 1

. Cha

ract

eris

tics

of p

atie

nts

with

hyp

erte

nsio

n w

ith t

rans

ition

in s

tage

of a

lbum

inur

ia (n

=12

5)

No

rmo

- to

Mic

roal

bum

inur

ia (n

=84

)M

icro

- to

Mac

roal

bum

inur

ia (n

=41

)

Cas

esC

ont

rols

Cas

esC

ont

rols

Num

ber

4242

2120

Age

at

bas

elin

e (y

ears

)66

.0 ±

9.1

66.2

± 8

.864

.4 ±

12.

563

.6 ±

9.7

Mal

e G

end

er (%

)27

(64.

3)28

(66.

7)16

(76.

2)16

(80.

0)

Cau

casi

an (%

)41

(97.

6)41

(97.

6)18

(85.

7)20

(100

)

Cur

rent

Sm

oker

at

bas

elin

e (%

)10

(23.

8)6

(14.

3)6

(28.

6)4

(20.

0)

BM

I at

bas

elin

e (k

g/m

2 )28

.6 ±

5.2

27.7

± 3

.928

.3 ±

3.3

28.2

± 3

.9

SB

P a

t b

asel

ine

(mm

Hg)

137.

0 ±

19.

713

8.4

± 1

4.5

140.

2 ±

21.

114

0.2

± 2

0.9

DB

P a

t b

asel

ine

(mm

Hg)

75.5

± 7

.178

.8 ±

9.2

79.4

± 9

.179

.2 ±

10.

0

Follo

w-u

p t

ime

(yea

rs)

2.8

± 0

.82.

8 ±

0.8

3.6

± 0

.82.

9 ±

1.2

Lab

orat

ory

par

amet

ers

Bas

elin

e U

AE

(mg/

24hr

)16

.5 [1

0.7,

22.

4]8.

1 [6

.5, 1

1.1]

***

111.

0 [8

0.0,

160

.0]

51.3

[34.

7, 1

12.2

]**

Bas

elin

e eG

FR (m

L/m

in/1

.73m

2 )72

.7 ±

17.

077

.1 ±

19.

956

.6 ±

21.

471

.3 ±

13.

9*

Bas

elin

e To

t. C

hole

ster

ol (m

mol

/L)

5.4

± 1

.15.

4 ±

1.1

5.2

± 0

.85.

3 ±

1.1

Bas

elin

e Fa

stin

g gl

ucos

e (m

mol

/L)

5.4

± 0

.75.

1 ±

0.9

5.2

± 0

.95.

0 ±

0.7

Fol

low

-up

SB

P (m

mH

g)14

7.8

± 2

0.5

140

±15

.914

4.4

± 2

0.2

138.

5 ±

16.

5

Fol

low

-up

UA

E (m

g/24

hr)

48.9

[39.

5, 8

6.8]

8.4

[6.2

, 11.

0]**

*56

9.4

[393

.2, 7

16.8

]71

.2 [4

8.9,

128

.7]*

**

Fol

low

-up

UA

E (%

cha

nge)

281.

9 [1

29.0

, 526

.9]

-0.2

[-22

.4, 1

8.5]

***

373.

4 [1

88.7

, 717

.9]

37.9

[-1.

3, 6

3.2]

***

Fol

low

-up

eG

FR (m

L/m

in/1

.73m

2 )75

.1 ±

19.

682

.2 ±

36.

358

.3 ±

28.

075

.8 ±

16.

3*

Med

icat

ion

usag

e at

bas

elin

e (%

)

RA

AS

i22

(52.

4)22

(52.

4)13

(61.

9)11

(55.

0)

Ant

ihyp

erte

nsiv

e38

(90.

5)40

(95.

2)20

(95.

2)20

(100

)

Lip

id-l

ower

ing

16 (3

8.1)

9 (2

1.4)

8 (3

8.1)

5 (2

5.0)

Dat

a re

por

ted

as

mea

n ±

sta

ndar

d d

evia

tion

or n

umb

er (p

erce

nt) o

r med

ian

[1st, 3

rd q

uart

iles]

. Cas

es v

s. C

ontr

ols:

*p

<0.

05, *

*p<

0.01

, ***

p<

0.00

1; B

MI:

bod

y m

ass

ind

ex (w

eigh

t kg/

heig

ht m

2 ); e

GFR

: est

imat

ed g

lom

erul

ar fi

ltrat

ion

rate

(4-v

aria

ble

Mod

ifica

tion

of D

iet i

n R

enal

Dis

ease

form

ula)

; RA

AS

i: an

giot

ensi

n-co

nver

ting

enzy

me

inhi

bito

rs (

AC

Ei)

oran

giot

ensi

n-2

rece

pto

r b

lock

ers

(AR

B);

UA

E:

urin

ary

alb

umin

exc

retio

n. P

erce

nt c

hang

e in

UA

E is

cal

cula

ted

as:

[(F

ollo

w-u

p U

AE

– B

asel

ine

UA

E)/

Bas

elin

e U

AE

]*10

0%.

6564


3

Tab

le 2

. Cha

ract

eris

tics

of p

atie

nts

with

typ

e 2

dia

bet

es w

ith t

rans

ition

in s

tage

of a

lbum

inur

ia (n

=82

)

No

rmo

-Mic

ro (n

=42

)M

icro

-Mac

ro (n

=40

)

Cas

esC

ont

rols

Cas

esC

ont

rols

Num

ber

2022

2020

Age

at

bas

elin

e (y

ears

)63

.4 ±

10.

364

.4 ±

9.5

67.4

± 6

.162

.5 ±

8.8

Mal

e G

end

er14

(48.

3)15

(68.

2)17

(85.

0)17

(85.

0)

Cau

casi

an (%

)18

(90.

0)21

(95.

5)19

20

Cur

rent

Sm

oker

at

bas

elin

e (%

)5

(25.

0)4

(18.

2)1

(10.

0)0

(0.0

)

BM

I at

bas

elin

e (k

g/m

2 )29

.5 ±

7.2

27.9

± 3

.630

.8 ±

4.2

29.4

± 5

.4

SB

P a

t b

asel

ine

(mm

HG

)13

8.2

± 1

4.2

133.

8 ±

21.

413

5.0

± 1

4.9

129.

3 ±

13.

9

DB

P a

t b

asel

ine

(mm

HG

)76

.1 ±

8.9

74.5

± 8

.871

.6 ±

9.5

74.0

± 8

.7

Dur

atio

n of

dia

bet

es †

n/a

n/a

17.3

± 9

.017

.3 ±

8.6

Follo

w-u

p t

ime

(yea

rs)

3.2

± 0

.92.

9 ±

0.8

2.8

± 1

.02.

7 ±

1.0

Lab

orat

ory

mea

sure

men

ts

Bas

elin

e U

AE

(mg/

24hr

s)20

.0 [1

6.1,

22.

6]7.

4 [6

.0, 1

2.5]

***

145.

3 [1

05, 1

92.4

]81

.3 [4

4.5,

122

.0]*

Bas

elin

e eG

FR (m

L/m

in/1

.73m

2)78

.8 ±

16.

483

.7 ±

16.

871

.1 ±

16.

082

.5 ±

21.

6

Bas

elin

e To

t. C

hole

ster

ol (m

mol

/L)

5.1

± 1

.35.

0 ±

1.4

4.3

± 1

.34.

6 ±

1.3

Bas

elin

e Fa

stin

g gl

ucos

e (m

mol

/L)‡

7.9

± 1

.97.

2 ±

1.1

7.2

± 2

.37.

1 ±

1.3

Bas

elin

e H

bA

1c (%

)†n/

an/

a7.

9 ±

1.4

8.2

± 1

.0

Fol

low

-up

SB

P (m

mH

g)14

2.4

± 1

5.7

130.

0 ±

16.

614

8.9

± 1

9.5

130.

4 ±

17.

0**

Fol

low

-up

UA

E (m

g/24

hrs)

44.6

[38.

6, 5

5.5]

8.9

[6.0

, 10.

8]**

*62

2.4

[402

.5, 7

92.6

]78

.6 [4

0.0,

138

.3]*

**

Fol

low

-up

UA

E (%

cha

nge)

122.

5 [1

01.1

, 206

.6]

1.2

[-17

.8, 2

2.1]

***

394.

9 [2

00.3

, 891

.6]

-11.

6 [-

35.8

, 69.

7]**

*

Fol

low

-up

eG

FR (m

L/m

in/1

.73m

2 )78

.7 ±

15.

486

.5 ±

20.

558

.8 ±

16.

380

.7 ±

22.

2**

Med

icat

ion

usag

e at

bas

elin

e (%

)

Ant

ihyp

erte

nsiv

e13

(65.

0)9

(40.

9)17

(85.

0)15

(75.

0)

RA

AS

i7

(35.

0)3

(13.

6)14

(70.

0)12

(60.

0)

Ora

l glu

cose

low

erin

g14

(70.

0)14

(63.

6)13

(65.

0)15

(75.

0)

Dat

a ar

e m

ean

(sta

ndar

d d

evia

tion)

or n

umb

er (p

erce

nt) o

r med

ian

[1st, 3

rd q

uart

iles]

is re

por

ted

. † S

teno

Dia

bet

es C

ente

r pat

ient

s; ‡ P

RE

VE

ND

pat

ient

s; C

ases

vs

Con

trol

s: *

p<

0.05

, **

p<

0.01

, **

*p<

0.00

1; B

MI:

bod

y m

ass

ind

ex (w

eigh

t kg

/hei

ght

m2 )

; eG

FR:

estim

ated

glo

mer

ular

filtr

atio

n ra

te (M

odifi

catio

n of

Die

t in

R

enal

Dis

ease

for

mul

a); R

AA

Si:

angi

oten

sin-

conv

ertin

g en

zym

e in

hib

itors

(AC

Ei)

or a

ngio

tens

in-2

rec

epto

r b

lock

ers

(AR

B);

UA

E: u

rinar

y al

bum

in e

xcre

tion.

C

hang

es in

UA

E a

re c

alcu

late

d a

s: [(

Follo

w-u

p U

AE

– B

asel

ine

UA

E)/

Bas

elin

e U

AE

]*10

0%.

6766


3

Figure 1a. Box-plot of cross-validated plasma proteomics classifier and transition in stage of albuminuria in patients with hypertension (n=125).

Figure 1b. C-index for prediction of transition in stage of albuminuria in patients with hypertension (n=125) for a) baseline UAE, baseline eGFR, RAASi (reference model) (C-index = 0.69) and b) cross-validated plasma proteomics classifier on top of reference model (C-index = 0.78; p <0.01 compared to reference model).

Figure 2a. Box-plot of cross-validated plasma proteomics classifier and transition in stage of albuminuria in patients with type 2 diabetes with microalbuminuria at baseline (n=40).

Figure 2b. C-index for prediction of transition from micro- to macroalbuminuria stage in patients with type 2 diabetes (n=40) for a) baseline UAE, baseline eGFR, RAASi (reference model) (C-index = 0.74) and b) cross-validated plasma proteomics classifier on top of reference model (C-index = 0.80; p = 0.04 compared to reference model).

6968


3

Overall performance of the model in type 2 diabetesIn type 2 diabetes, 18 molecular features were selected for the type 2 diabetes classifier (Supplemental Table 2). The values of the type 2 diabetes classifier were not significantly different between the cases and controls (-0.17 ± 2.2 vs. -0.87 ± 2.48, p=0.18). Additionally, the type 2 diabetes classifier was not independently associated with transition in albuminuria (Odds ratio 1.15, 95% CI=0.94-1.42, p=0.18). However, when testing an interaction term to assess whether the predictive performance was modified by baseline stage of albuminuria, the interaction was found to be significant (p-value for interaction = 0.04). In the normo- to microalbuminuria subgroup, the type 2 diabetes classifier was not associated with transitioning in stage of albuminuria (Odds ratio 1.21, 95% CI=0.72-2.04, p=0.46). In contrast, in patients with microalbuminuria at baseline, the type 2 diabetes classifier was independently associated with transitioning in stage of albuminuria (Odds ratio 1.53, 95% CI=1.03-2.26, p=0.04). The type 2 diabetes classifier improved risk prediction for transitioning in stage of albuminuria in the microalbuminuria subgroup on top of the baseline model (C-index from 0.73 to 0.80, p=0.04; IDI 0.05, p=0.31) (Figure 2b).

Identification of molecular features and pathways in hypertension and type 2 diabetesThe molecular features identified in hypertension were linked to mechanisms that are recognized to contribute to nephropathy including metabolic pathways, intracellular signaling pathways (PI3K-Akt, MAPK, VEGF, Wnt, mTOR), angiogenesis, cytokine-cytokine receptor interaction, and ECM-receptor interaction. The molecular features identified in type 2 diabetes were associated to several pathways recognized to contribute to renal disease including metabolic pathways, signaling pathways (PI3K-Akt, VEGF, mTOR, MAPK, p38 MAPK), angiogenesis, pentose phosphate, and cytokine-cytokine receptor interaction. Figure 3 represents pathways recognized to contribute to nephropathy in hypertension and type 2 diabetes.

Fig

ure

3. B

ipar

tite

grap

h of

iden

tified

pep

tides

and

est

ablis

hed

nep

hrop

athy

pat

hway

s in

hyp

erte

nsio

n or

typ

e 2

dia

bet

es.

Pat

hway

s ar

e re

pre

sent

ed b

y b

lack

rec

tang

les

and

pro

tein

s ar

e re

pre

sent

ed b

y w

hite

rec

tang

les

with

a b

lack

bor

der

. Pep

tides

and

pat

hway

s in

onl

y hy

per

tens

ion

or o

nly

typ

e 2

dia

bet

es

are

loca

ted

at

the

top

and

bot

tom

of t

he fi

gure

, res

pec

tivel

y. P

athw

ays

in t

he m

idd

le h

old

pep

tides

from

bot

h hy

per

tens

ion

(HT)

and

typ

e 2

dia

bet

es (D

M).

7170


3

DISCUSSION

In this study we showed that newly developed plasma proteomics classifiers are able to predict transition in stage of albuminuria in hypertensive patients and transition from micro- to macroalbuminuria in patients suffering from type 2 diabetes. This was independent of established renal risk markers baseline UAE and eGFR, as well as used of RAAS intervention. The plasma peptides identified in hypertension and type 2 diabetes have been linked to pathways associated with established mechanisms of renal disease, including fibrosis, inflammation, angiogenesis, and mineral metabolism. Our findings suggest the use of these plasma proteomics classifiers as tools to identify patients with hypertension or type 2 diabetes at risk of renal disease progression in whom preventive treatment may be started in order to halt the devastating consequences of CKD.

Proteomics has been identified as a hypothesis-generating tool for biomarker discovery, though most proteomics studies of renal disease have focused on urinary proteomics [20-22]. For this study, we used plasma instead of urine. An advantage of plasma is that there are a higher numbers of substances compared to urine, so therefore there is a higher likelihood of detecting yet unknown biomarkers of CKD. Additionally, plasma still contains active mediators, in contrast to urine which is composed of metabolic waste products. There are a few proteomics studies on diabetic nephropathy conducted on blood-derived samples of patients with type 2 diabetes [23, 24], but none to our knowledge of studies on blood-derived samples for hypertensive renal disease. However, the aforementioned studies were conducted in serum, and peptides in serum have been shown to be subject to degradation due to high proteolytic activity immediately upon clotting [25]. The number of plasma proteomics studies is few. In a study of patients with type 1 or type 2 diabetes, twenty-eight unique plasma proteins mostly representing inflammatory response and coagulation proteins were identified for the screening and detection of diabetic retinopathy [26]. In a study conducted at the Steno Diabetes Center, ten single plasma peptides and three multi-peptide candidate biomarkers were identified for diabetic nephropathy in patients with type 1 diabetes ranging from having normoalbuminuria, microalbuminuria, and diabetic nephropathy [27]. However, these studies were cross-sectional in nature and did not evaluate the predictive value of plasma proteomics for disease progression.

CKD is a complex disease involving many underlying pathophysiological mechanisms. Prolonged hyperglycemia in type 2 diabetes leads to chronic metabolic and hemodynamic changes that activate many intracellular signaling pathways, transcription factors, cytokines, chemokines, and growth factors that are implicated in diabetic nephropathy. The pathophysiology underlying hypertension is not fully known, but many mechanisms including signaling pathways, RAAS, and vascular inflammation and remodeling have been identified as fundamental mechanisms responsible for the development of hypertensive nephropathy. The hypertension and type 2 diabetes plasma proteomics classifiers included peptides that have been associated to several mechanisms contributing to renal disease. Peptides relating to fibrosis, a well-known pathologic feature of nephropathy

resulting from expansion and change in composition of extracellular matrix in the kidney, were present in both classifiers. Peptides were related to PI3K-Akt, VEGF, mTOR, MAPK, and p38 MAPK signaling pathways points to a strong link to inflammation as a driver of progressive renal disease. Vascular endothelial growth factor A, a marker of angiogenesis, was observed in both hypertension and type 2 diabetes. An imbalance of angiogenesis-related factors impeding the development of new blood vessels from pre-existing vasculature has been shown to be involved in the progression of CKD. Finally, activation of the Wnt pathway suggests a role of mineral metabolism disorders, and studies have shown that vascular calcification is associated with renal disease [28].

The plasma peptides identified in hypertension did not overlap those found in type 2 diabetes and vice versa. This suggests that there may be disease-specific expressions of peptides in both hypertension and type 2 diabetes, and this is quite likely since the underlying pathology is different in both diseases. However, the peptides identified in the diseases were associated with similar pathways that drive nephropathy as shown in Figure 3. The question remains whether the different peptides truly represent differences between hypertension and type 2 diabetes or if these differences are related to the small sample size and case-control study design. Whether the peptides identified are either involved in the causal pathway contributing to CKD progression, or are markers of its risk, or are merely the end-product of existing pathological processes, remains an important and unresolved question that requires further exploration and cannot be addressed by our study. Additionally, assessing for confounding between these peptides and other risk factors of renal disease was beyond the scope of this prediction study and requires additional etiologic studies.

We additionally detected molecular features that were identified as amino acid sequences and included these features in the classifiers. We hypothesize that these molecular features are free amino acids and are not degradation products from sample processing since the signal intensity of these amino acids were different between cases or controls.

There are strengths and limitations to this study. One strength includes using a rigorous cross-validation approach with several nested cross-validation loops to avoid the reporting of over-optimistic model performance due to validation of our classifiers with the same data that was used for model development. All decisions made during model development were subjected to cross-validation. Limitations include a significant difference in the absolute values of UAE between cases and controls and in eGFR in the micro- macroalbuminuria group in hypertension, indicating that cases and controls were different at baseline. The imbalance in baseline UAE was unintended, and represents our best efforts in matching cases and controls with the same albuminuria range from the original cohorts. To control for the imbalance, we adjust for baseline albuminuria and baseline eGFR in our statistical models. Due to the early stage of disease being investigated and the strict criteria used for the matching of cases and controls, we were able to select only a limited number of patients from the original large PREVEND and Steno Diabetes Center cohorts, which limited statistical power. However, this is to the best of our knowledge, the first study to

7372


3

investigate prediction of renal disease progression in hypertensive and type 2 diabetes patients using plasma proteomics. Additional limitations include the lack of information concerning level of exercise, insulin use, diet, and other medication type and dose. The plasma proteomics classifier identified now requires validation in larger cohorts, ideally in large, prospective cohort studies.

The results of this plasma proteomics study support the growing evidence of using peptidomic platforms as a strategy for risk stratification of renal disease. As advancing laboratory techniques become more and more realistic in clinical practice, the complementary use of peptidomic techniques as a tool to reduce the burden of renal disease complications represents a strategy to improve disease diagnosis, prognosis, and treatment. In conclusion, our findings provide some evidence that proteomics analysis of plasma samples could be used to improve renal risk assessment of patients with hypertension and patients with type 2 diabetes for the prediction of transition of stage of albuminuria beyond established renal risk markers. Validation in an external study is needed to assess the reproducibility of these results.

ACKNOWLEDGMENTS

We thank M.E. Hellemons and S.S. Roscioni, University Medical Center Groningen, for their support in the early stages of this study.

FUNDING

The PREVEND Study has been made possible by grants of the Dutch Kidney Foundation. The work leading to this paper has received funding from the European Union under grant agreement no. HEALTH–F2–2009–241544 “SysKID consortium” and HEALTH.2011.2.4.2-2 (278249) “EU-MASCARA”.

REFERENCES

1. Hart PD, Bakris GL. Hypertensive nephropathy: prevention and treatment

recommendations. Expert Opin Pharmacother. 2010 Nov;11(16):2675-86.






3. Viazzi F, Leoncini G, Conti N, Tomolillo C, Giachero G, Vercelli M, Deferrari G, Pontremoli R.

Microalbuminuria is a predictor of chronic renal insufficiency in patients without diabetes

and with hypertension: the MAGIC study. Clin J Am Soc Nephrol. 2010 Jun;5(6):1099-106.

4. Brenner BM, Cooper ME, de Zeeuw D, Keane WF, Mitch WE, Parving HH, Remuzzi G,

Snapinn SM, Zhang Z, Shahinfar S; RENAAL Study Investigators. Effects of losartan on

renal and cardiovascular outcomes in patients with type 2 diabetes and nephropathy. N

Engl J Med. 2001 Sep 20;345(12):861-9.

5. Lewis EJ, Hunsicker LG, Clarke WR, Berl T, Pohl MA, Lewis JB, Ritz E, Atkins RC, Rohde

R, Raz I; Collaborative Study Group. Renoprotective effect of the angiotensin-receptor

antagonist irbesartan in patients with nephropathy due to type 2 diabetes. N Engl J Med.

2001 Sep 20;345(12):851-60.







7. Hellemons ME, Kerschbaum J, Bakker SJ, Neuwirt H, Mayer B, Mayer G, de Zeeuw D,

Lambers Heerspink HJ, Rudnicki M. Validity of biomarkers predicting onset or progression

of nephropathy in patients with Type 2 diabetes: a systematic review. Diabet Med. 2012

May;29(5):567-77.

8. Gerszten RE, Wang TJ. The search for new cardiovascular biomarkers. Nature. 2008 Feb

21;451(7181):949-52.

9. Rossing K, Christensen PK, Hovind P, Tarnow L, Rossing P, Parving HH. Progression of

nephropathy in type 2 diabetic patients. Kidney Int. 2004 Oct;66(4):1596-605.

10. Merchant ML, Perkins BA, Boratyn GM, Ficociello LH, Wilkey DW, Barati MT, Bertram CC,

Page GP, Rovin BH, Warram JH, Krolewski AS, Klein JB. Urinary peptidome may predict

renal function decline in type 1 diabetes and microalbuminuria. J Am Soc Nephrol. 2009

Sep;20(9):2065-74.

7574


3

11. Pinto-Sietsma SJ, Janssen WM, Hillege HL, Navis G, De Zeeuw D, De Jong PE. Urinary

albumin excretion is associated with renal functional abnormalities in a nondiabetic

population. J Am Soc Nephrol. 2000 Oct;11(10):1882-8.

12. Reinhard H, Hansen PR, Persson F, Tarnow L, Wiinberg N, Kjær A, Petersen CL, Winther

K, Parving HH, Rossing P, Jacobsen PK. Elevated NT-proBNP and coronary calcium

score in relation to coronary artery disease in asymptomatic type 2 diabetic patients with

elevated urinary albumin excretion rate. Nephrol Dial Transplant. 2011 Oct;26(10):3242-9.

13. Abbasi A, Peelen LM, Corpeleijn E, van der Schouw YT, Stolk RP, Spijkerman AM, van der

A DL, Moons KG, Navis G, Bakker SJ, Beulens JW. Prediction models for risk of developing

type 2 diabetes: systematic literature search and independent external validation study.

BMJ. 2012 Sep 18;345:e5900.

14. Rohlfing CL, Little RR, Wiedmeyer HM, England JD, Madsen R, Harris MI, Flegal KM,

Eberhardt MS, Goldstein DE. Use of GHb (HbA1c) in screening for undiagnosed diabetes

in the U.S. population. Diabetes Care. 2000 Feb;23(2):187-91.




16;130(6):461-70.

16. Schulz A, Jankowski J, Zidek W, Jankowski V. Absolute quantification of endogenous

angiotensin II levels in human plasma using ESI-LC-MS/MS. Clin Proteomics. 2014 Oct

27;11(1):37. eCollection 2014.

17. Friedman J, Hastie T, Tibshirani R. Regularization Paths for Generalized Linear Models via

Coordinate Descent. J Stat Softw. 2010;33(1):1-22.

18. Elashoff JD. NQuery Advisor Version 7.0 User’s Guide 2007. Statistical Solutions, Cork,

Ireland.

19. Tibshirani R. Regression Shrinkage and Selection via the Lasso. J R Stat Soc Series B

Stat Methodol 1996;58(1):267-288.

20. Bhensdadia NM, Hunt KJ, Lopes-Virella MF, Michael Tucker J, Mataria MR, Alge JL, Neely

BA, Janech MG, Arthur JM; Veterans Affairs Diabetes Trial (VADT) study group. Urine

haptoglobin levels predict early renal functional decline in patients with type 2 diabetes.

Kidney Int. 2013 Jun;83(6):1136-43.

21. Zürbig P, Jerums G, Hovind P, Macisaac RJ, Mischak H, Nielsen SE, Panagiotopoulos

S, Persson F, Rossing P. Urinary proteomics for early diagnosis in diabetic nephropathy.

Diabetes. 2012 Dec;61(12):3304-13.









23. Cho EH, Kim MR, Kim HJ, Lee DY, Kim PK, Choi KM, Ryu OH, Kim CW. The discovery of

biomarkers for type 2 diabetic nephropathy by serum proteome analysis. Proteomics Clin

Appl. 2007 Apr;1(4):352-61.

24. Kim HJ, Cho EH, Yoo JH, Kim PK, Shin JS, Kim MR, Kim CW. Proteome analysis of serum

from type 2 diabetics with nephropathy. J Proteome Res. 2007 Feb;6(2):735-43.

25. Kolch W, Neususs C, Pelzing M, Mischak H. Capillary electrophoresis-mass spectrometry

as a powerful tool in clinical diagnosis and biomarker discovery. Mass Spectrom Rev.

2005 Nov-Dec;24(6):959-77.

26. Lu CH, Lin ST, Chou HC, Lee YR, Chan HL. Proteomic analysis of retinopathy-related

plasma biomarkers in diabetic patients. Arch Biochem Biophys. 2013 Jan 15;529(2):146-56.

27. Hansen HG, Overgaard J, Lajer M, Hubalek F, Højrup P, Pedersen L, Tarnow L, Rossing P,

Pociot F, McGuire JN. Finding diabetic nephropathy biomarkers in the plasma peptidome

by high-throughput magnetic bead processing and MALDI-TOF-MS analysis. Proteomics

Clin Appl. 2010 Sep;4(8-9):697-705.

28. Moe SM, Chen NX. Mechanisms of vascular calcification in chronic kidney disease. J Am

Soc Nephrol. 2008 Feb;19(2):213-6.

7776


3

SUPPLEMENTAL APPENDIX 1. EXTENDED STATISTICAL METHODS

For both hypertension and type 2 diabetes, a separate plasma proteomics classifier was developed using regularized logistic regression in three steps. Log-base-2 transformed peptide values were used in all regression analyses. If a peptide did not exhibit any expression in a sample, it was treated as a ‘zero’ rather than a missing value. First, peptides with more than p% missing values across all samples were excluded. Second, the ‘zero’ values of a peptide were set to the minimum measured value of that peptide on the logarithmic scale minus a constant value of d. A least absolute shrinkage and selection operator (LASSO) logistic model was fitted to select molecular features,1 with the tuning parameter λ selected such that the cross-validated deviance was minimized. Third, the model was refitted using ridge logistic regression including two variables for each selected peptide: one binary variable distinguishing zero from non-zero measurements, and one continuous variable equal to the logarithm of the measured peptide expression. For zero measurements, this second variable was set to the mean logarithm of peptide expressions to maintain interpretability of the binary variable’s coefficient. The tuning parameters p% and d were selected among a few possible choices by a leave-one-out cross-validation loop around the whole model building process. This optimization gave the final regression coefficients which are described in Supplemental Tables 1 & 2. Since an independent validation sample was not available, we calculated cross-validated linear predictor scores (referred to as the plasma proteomics classifier) by repeating the complete model development process, including re-tuning of p% and d and selection of peptides, each time leaving out in turn one pair or a pair and a sample without partner from the analysis, and using the estimated model information (selected molecular features and model coefficients) to estimate the cross-validated plasma proteomics classifier for the left-out patients. The so-obtained plasma proteomics classifiers then used for model validation in the original respective patient samples (hypertension and diabetes mellitus).

Reference. Supplemental Appendix 11. Tibshirani R. Regression shrinkage and selection via the lasso. Journal of the Royal

Statistical Society. Series B (Methodological). 1996;58(1):267-288.

7978


3

Sup

ple

men

tal T

able

1. M

asse

s of

the

37

sele

cted

mol

ecul

ar f

eatu

res

with

>25

% n

on-z

ero

valu

es in

hyp

erte

nsio

n w

ith t

heir

pro

por

tion

of n

on-z

ero

valu

es

per

sta

tus

grou

p (c

ase/

cont

rol),

the

min

imum

non

-zer

o lo

g2 in

tens

ities

, the

uni

varia

te p

- an

d q

-val

ues,

the

coe

ffici

ents

for

the

cont

inuo

us a

nd d

icho

tom

ous

inte

nsiti

es in

the

mul

tivar

iab

le p

red

ictio

n m

odel

, the

diff

eren

ce in

log

odd

s b

etw

een

an a

vera

ge n

on-z

ero

log2

inte

nsity

and

a z

ero

inte

nsity

, and

the

par

tial

R-s

qua

red

. The

var

iab

les

are

rank

ed b

y th

eir

par

tial R

2 (i.

e. b

y th

eir

rela

tive

imp

orta

nce

in t

he p

red

ictio

n m

odel

). To

tal R

2 =

0.8

6.

Des

crip

tio

n o

f in

tens

ity

valu

esU

niva

riat

e

ELR

T t

ests

1M

ulti

vari

able

mo

del

Mo

lecu

lar

feat

ure

Mo

lecu

lar

Nam

e

Pro

po

rtio

nno

n-ze

roam

ong

ca

ses

Pro

po

rtio

nno

n-ze

roam

ong

co

ntro

ls

p-v

alue

q-v

alue

**β

cont

*

βd

ich†

non-

zero

vs. z

ero

‡

Par

tial

R2§

1147

.1s:

104.

75m

/zC

yste

in0.

476

0.71

00.

024

0.88

1-0

.187

1.38

7-1

.771

0.00

4

167.

2s:9

20.6

6m/z

Ser

ine-

Thre

onin

pro

teas

e0.

381

0.22

60.

133

1.00

00.

248

-2.2

321.

942

0.00

3

1147

.5s:

413.

03m

/zG

luco

se-6

Pho

spha

te T

rans

loca

se0.

333

0.21

00.

297

1.00

00.

191

-1.5

361.

515

0.00

2

164.

7s:8

68.6

7m/z

Klo

tho

0.42

90.

274

0.07

30.

827

0.18

2-1

.784

1.25

3-0

.001

344.

0s:1

92.2

4m/z

Urin

e gl

ycop

eptid

e0.

889

0.72

60.

021

0.89

70.

078

-0.9

340.

465

-0.0

01

3778

.1s:

285.

26m

/zM

yotu

bul

arin

-rel

ated

Pro

tein

12

0.31

70.

500

0.00

20.

590

-0.1

120.

673

-1.0

76-0

.002

730.

8s:1

56.6

5m/z

Arg

inin

0.22

20.

339

0.17

01.

000

-0.1

171.

092

-0.7

67-0

.003

163.

1s:8

04.7

2m/z

Friz

zled

-1 P

rote

in0.

175

0.37

10.

029

0.91

5-0

.128

1.05

6-1

.044

-0.0

03

858.

4s:1

25.7

3m/z

Rec

epto

r-ty

pe

tyro

sine

-pro

tein

-p

hosp

hata

se0.

460

0.27

40.

057

0.97

30.

155

-0.7

421.

585

-0.0

04

225.

9s:3

46.1

3m/z

Hum

anin

0.25

40.

306

0.56

21.

000

-0.1

341.

086

-1.1

18-0

.004

262.

1s:2

41.2

6m/z

Rec

epto

r-ty

pe

tyro

sine

-pro

tein

p

hosp

hata

se U

0.95

20.

790

0.01

00.

978

0.18

6-1

.392

2.04

5-0

.004

418.

3s:2

21.2

3m/z

Vasc

ular

end

othe

lial g

row

th fa

ctor

A0.

381

0.53

20.

198

1.00

0-0

.045

0.20

3-0

.566

-0.0

04

872.

0s:2

61.2

2m/z

Alp

ha-e

nola

se0.

825

0.69

40.

222

1.00

00.

058

-0.4

020.

615

-0.0

04

219.

1s:3

48.1

2m/z

Cyt

ochr

om P

450

A11

0.34

90.

177

0.04

81.

000

0.12

6-0

.707

1.29

7-0

.004

699.

4s:1

41.7

0m/z

Urin

e gl

ycop

eptid

e0.

413

0.27

40.

096

0.93

60.

104

-0.6

921.

080

-0.0

04

707.

6s:1

69.6

2m/z

Ost

eoca

lcin

0.41

30.

565

0.06

40.

942

-0.1

281.

452

-0.8

32-0

.004

3782

.9s:

358.

24m

/zN

IK r

elat

ed p

rote

in k

inas

e0.

667

0.79

00.

120

1.00

0-0

.108

0.77

9-1

.083

-0.0

04

897.

9s:1

87.6

2m/z

Tryp

tpha

n0.

429

0.29

00.

012

0.86

20.

120

-1.0

760.

854

-0.0

04

413.

7s:1

17.7

5m/z

WD

-rep

eat-

cont

aini

ng p

rote

in0.

286

0.48

40.

069

0.84

1-0

.083

0.41

2-0

.842

-0.0

04

164.

8s:8

26.6

9m/z

Fib

rob

last

gro

wth

fact

or R

ecep

tor

40.

190

0.32

30.

062

0.95

9-0

.144

1.06

9-1

.265

-0.0

05

257.

1s:1

11.7

7m/z

Mol

ecul

ar fe

atur

e st

ill t

o b

e id

entifi

ed0.

460

0.21

00.

001

0.57

00.

017

0.32

40.

576

-0.0

05

429.

4s:1

35.7

0m/z

Asp

arta

t0.

143

0.37

10.

011

0.92

8-0

.088

0.59

6-0

.752

-0.0

05

299.

6s:2

51.1

9m/z

Ank

yrin

rep

eat

dom

ain-

cont

aini

ng

pro

tein

0.36

50.

177

0.03

40.

876

0.17

5-1

.584

1.16

0-0

.005

170.

1s:1

134.

60m

/zS

trom

al in

tera

ctio

nmol

ecul

20.

333

0.17

70.

016

0.85

30.

105

-0.5

191.

172

-0.0

05

166.

6s:1

014.

62m

/zS

olut

e ca

rrie

r fa

mily

22

mem

ber

14

0.15

90.

435

0.00

20.

607

-0.1

541.

099

-1.4

84-0

.005

779.

5s:4

09.0

4m/z

Cal

cium

/cal

mod

ulin

-dep

end

ent

pro

tein

ki

nase

II in

hib

itor

0.27

00.

468

0.02

70.

883

-0.0

850.

698

-0.7

23-0

.006

177.

8s:1

38.5

4m/z

His

tidin

e0.

349

0.50

00.

201

1.00

0-0

.114

1.11

6-0

.773

-0.0

06

692.

6s:1

43.7

2m/z

Urin

e gl

ycop

eptid

e0.

397

0.59

70.

052

1.00

0-0

.041

-0.0

35-0

.837

-0.0

06

333.

4s:1

64.4

9m/z

Ost

eoca

lcin

0.36

50.

194

0.03

90.

956

0.11

9-0

.971

0.92

7-0

.006

721.

5s:1

81.3

0m/z

Tyro

sine

0.30

20.

500

0.06

70.

851

-0.0

350.

126

-0.4

91-0

.006

801.

1s:1

25.7

5m/z

Mol

ecul

ar fe

atur

e st

ill t

o b

e id

entifi

ed0.

635

0.48

40.

071

0.81

20.

067

-0.3

670.

720

-0.0

06

2223

.0s:

266.

18m

/zS

acol

ipin

(ass

ocia

ted

with

SE

RC

A1)

0.66

70.

500

0.14

51.

000

0.06

6-0

.440

0.77

1-0

.006

1173

.9s:

123.

77m

/zM

olec

ular

feat

ure

still

to

be

iden

tified

0.61

90.

403

0.05

11.

000

0.07

4-0

.188

1.11

1-0

.006

739.

2s:2

61.1

7m/z

Alp

ha-e

nola

se0.

508

0.32

30.

065

0.91

20.

060

-0.3

720.

620

-0.0

07

171.

1s:9

78.6

2m/z

Rec

epto

r ty

p t

yros

ine

pro

tein

p

hosp

hata

se0.

222

0.41

90.

003

0.53

4-0

.086

0.64

5-0

.788

-0.0

07

165.

9s:9

28.6

4m/z

Fib

rone

ctin

0.42

90.

226

0.01

20.

883

0.12

8-0

.991

1.19

2-0

.007

259.

7s:4

17.1

7m/z

Myo

ferin

0.63

50.

403

0.03

20.

864

0.10

6-0

.550

1.28

6-0

.007

**q

-val

ues

adju

sted

for

fals

e d

isco

very

rat

e; * β

con

t: r

egre

ssio

n co

effic

ient

of c

ontin

uous

log2

inte

nsity

val

ues;

† β d

ich:

reg

ress

ion

coef

ficie

nt o

f dic

hoto

mou

s (n

on-z

ero

vs. z

ero)

inte

nsity

; ‡ non

-zer

o vs

. zer

o: re

lativ

e lo

g od

ds

of a

n av

erag

e no

n-ze

ro lo

g2 in

tens

ity v

s. z

ero

inte

nsity

; § Par

tial R

2 of a

feat

ure

= T

otal

R2 –

R2

omitt

ing

that

feat

ure.

Ref

eren

ce: 1 T

aylo

r S

. an

d P

olla

rd K

. H

ypot

hesi

s te

sts

for

poi

nt-m

ass

mix

ture

dat

a w

ith a

pp

licat

ion

to ‘

omic

s d

ata

with

man

y ze

ro v

alue

s. S

tat

Ap

pl G

enet

M

ol B

iol.

2009

;8:A

rtic

le 8

.

8180


3

Sup

ple

men

tal T

able

2. M

asse

s of

the

18 s

elec

ted

mol

ecul

ar fe

atur

es w

ith >

33%

non

-zer

o va

lues

in ty

pe

2 d

iab

etes

with

thei

r p

rop

ortio

n of

non

-zer

o va

lues

p

er s

tatu

s gr

oup

(cas

e/co

ntro

l), t

he m

inim

um n

on-z

ero

log2

inte

nsiti

es, t

he u

niva

riate

p-

and

q-v

alue

s, t

he c

oeffi

cien

ts fo

r th

e co

ntin

uous

and

dic

hoto

mou

s in

tens

ities

in t

he m

ultiv

aria

ble

pre

dic

tion

mod

el, t

he d

iffer

ence

in lo

g od

ds

bet

wee

n an

ave

rage

non

-zer

o lo

g2 in

tens

ity a

nd a

zer

o in

tens

ity, a

nd t

he p

artia

l R

-sq

uare

d. T

he v

aria

ble

s ar

e ra

nked

by

thei

r p

artia

l R2

(i.e.

by

thei

r re

lativ

e im

por

tanc

e in

the

pre

dic

tion

mod

el).

Tota

l R2 =

0.74

.

Des

crip

tio

n o

f in

tens

ity

valu

esU

niva

riat

eE

LRT

tes

ts1

Mul

tiva

riab

le m

od

el

Mo

lecu

lar

feat

ure

Mo

lecu

lar

Nam

eP

rop

ort

ion

non-

zero

amo

ng c

ases

Pro

po

rtio

nno

n-ze

roam

ong

co

ntro

lsp

-val

ueq

-val

ue**

βco

nt*

βd

ich†

non-

zero

vs. z

ero

‡

Par

tial

R2§

680.

9s:1

37.7

1m/z

Urin

e gl

ycop

eptid

e0.

667

0.77

50.

390

1.00

0-0

.236

2.35

8-1

.702

0.01

5

673.

3s:1

09.7

7m/z

Mol

ecul

ar fe

atur

e st

ill t

o b

e id

entifi

ed0.

548

0.37

50.

165

1.00

00.

569

-7.3

841.

719

0.01

5

366.

9s:3

67.1

3m/z

Ang

io-a

ssoc

iate

d m

igra

tory

ce

ll p

rote

in0.

857

1.00

00.

008

1.00

0-0

.549

6.92

8-3

.251

0.01

3

2083

.6s:

788.

84m

/zP

hosp

hogl

ucom

utas

e 2

0.61

90.

850

0.04

31.

000

-0.3

084.

706

-1.0

760.

011

243.

9s:3

61.1

8m/z

Pro

tein

Pho

spha

tase

1F

0.92

90.

675

0.00

91.

000

0.30

9-3

.722

1.73

50.

010

424.

4s:2

20.1

9m/z

Vasc

ular

end

othe

lial g

row

th

fact

or A

0.33

30.

450

0.15

81.

000

-0.3

254.

061

-1.4

340.

008

256.

1s:3

75.1

9m/z

Mol

ecul

ar fe

atur

e st

ill t

o b

e id

entifi

ed0.

619

0.40

00.

135

1.00

00.

209

-2.3

081.

378

0.00

7

2874

.1s:

971.

43m

/zA

milo

-sen

sitiv

e-am

ino-

oxid

ase

0.21

40.

550

0.00

71.

000

-0.4

255.

346

-2.1

670.

007

244.

2s:7

49.1

3m/z

Fork

head

Box

J1

0.28

60.

425

0.41

41.

000

-0.1

982.

507

-0.8

930.

007

476.

5s:1

49.6

7m/z

Mol

ecul

ar fe

atur

e st

ill t

o b

e id

entifi

ed0.

524

0.70

00.

153

1.00

0-0

.211

2.21

1-1

.239

0.00

6

1707

.4s:

115.

76m

/zA

spar

agin

e0.

548

0.45

00.

524

1.00

00.

139

-1.2

241.

026

0.00

5

351.

1s:1

17.7

8m/z

Mol

ecul

ar fe

atur

e st

ill t

o b

e id

entifi

ed0.

881

0.72

50.

139

1.00

00.

241

-2.4

621.

778

0.00

5

869.

5s:2

61.2

1m/z

Alp

ha-e

nola

se0.

857

0.92

50.

104

1.00

0-0

.303

2.97

4-2

.404

0.00

4

730.

0s:1

89.2

6m/z

Ser

ine/

Thre

onin

e0.

548

0.32

50.

087

1.00

00.

297

-4.6

830.

873

0.00

3

414.

7s:1

57.6

5m/z

Arg

inin

0.47

60.

225

0.03

11.

000

0.27

9-2

.680

1.64

20.

000

3132

.2s:

844.

11m

/zP

eptid

yl p

roly

l iso

mer

ase-

like

50.

262

0.45

00.

132

1.00

0-0

.129

1.71

6-0

.897

-0.0

01

165.

6s:8

58.7

1m/z

Treh

alas

e0.

595

0.32

50.

030

1.00

00.

193

-2.2

080.

979

-0.0

03

166.

8s:9

10.6

8m/z

Ser

ine/

Thre

onin

e p

rote

ase

NE

K1

0.59

50.

325

0.03

11.

000

0.17

6-1

.873

1.03

0-0

.004

**q

-val

ues

adju

sted

for

fals

e d

isco

very

rat

e; * β

con

t: r

egre

ssio

n co

effic

ient

of c

ontin

uous

log2

inte

nsity

val

ues;

† β d

ich:

reg

ress

ion

coef

ficie

nt o

f dic

hoto

mou

s (n

on-z

ero

vs. z

ero)

inte

nsity

; ‡ non

-zer

o vs

. zer

o: re

lativ

e lo

g od

ds

of a

n av

erag

e no

n-ze

ro lo

g2 in

tens

ity v

s. z

ero

inte

nsity

; § Par

tial R

2 of a

feat

ure

= T

otal

R2 –

R2

omitt

ing

that

feat

ure.

Ref

eren

ce: 1 T

aylo

r S

. an

d P

olla

rd K

. H

ypot

hesi

s te

sts

for

poi

nt-m

ass

mix

ture

dat

a w

ith a

pp

licat

ion

to ‘

omic

s d

ata

with

man

y ze

ro v

alue

s. S

tat

Ap

pl G

enet

M

ol B

iol.

2009

;8:A

rtic

le 8

.

CHAPTER 4Urine and plasma metabolites predict the development of diabetic nephropathy in patients with type 2 diabetes mellitus

MJ PenaHJ Lambers Heerspink

ME HellemonsT Friedrich

G DallmannM Lajer

SJL BakkerRT Gansevoort

P RossingD de ZeeuwSS Roscioni

Diabet Med. 2014 Sep;31(9):1138-47

8584

Metabolomics predict development of diabetic nephropathyChapter 4

4

ABSTRACT

Objectives: Early detection of individuals with type 2 diabetes mellitus (T2DM) or hypertension at risk for micro- or macroalbuminuria may facilitate prevention and treatment of renal disease. We aimed to discover plasma and urine metabolites that predict the development of micro- or macroalbuminuria.

Methods: Patients with T2DM (n=90) and hypertension (n=150) were selected from the community-cohort PREVEND and the Steno Diabetes Center for this case-control study. Cases transitioned in albuminuria stage (from normo- to microalbuminuria or micro- to macroalbuminuria). Controls, matched for age, gender, and baseline albuminuria stage, remained in normo- or microalbuminuria stage during follow-up. Median follow-up was 2.9 years. Metabolomics were performed on plasma and urine. The predictive performance of a metabolite for albuminuria transition was assessed by the integrated discrimination index (IDI).

Results: In patients with T2DM with normoalbuminuria, no metabolites discriminated cases from controls. In patients with T2DM with microalbuminuria, plasma histidine was lower (fold change [FC]=0.87, p=0.02) and butenoylcarnitine was higher (FC=1.17, p=0.007) in cases versus controls. In urine, hexose, glutamine, and tyrosine were lower in cases versus controls (FC: 0.20, p<0.001; 0.32, p<0.001; 0.51, p=0.006, respectively). Adding the metabolites to a model of baseline albuminuria and estimated glomerular filtration rate improved risk prediction for macroalbuminuria transition (plasma IDI=0.28, p<0.001; urine IDI=0.43, p<0.001). These metabolites did not differ between hypertensive cases and controls without T2DM.

Conclusions: T2DM-specific plasma and urine metabolites were discovered that predict the development of macroalbuminuria beyond established renal risk markers. These results should be confirmed in a large, prospective cohort.

INTRODUCTION

Type 2 diabetes mellitus (T2DM) and hypertension are leading causes of progressive chronic kidney disease (CKD) and may lead to End-Stage Renal Disease (ESRD). Despite current, established therapies, many people with T2DM remain at risk for progressive renal function loss. Latest therapies including blockade of the Renin-Angiotensin-Aldosterone-System (RAAS) show significant risk reduction in late stage of disease [1]. However, most reduction in absolute and relative risk is observed when RAAS-intervention is tested in earlier stages of disease [2]. Microalbuminuria is considered the first sign of kidney dysfunction, often progressing to macroalbuminuria and ESRD [3,4]. Thus, early detection of individuals with T2DM at risk for micro- or macroalbuminuria by means of novel biomarkers and subsequent appropriate treatment may represent an effective strategy to reduce the incidence of ESRD.

Metabolomics, i.e. the measurement of low-weight intermediates and end-products of cellular functions in biological fluids, has emerged as a valuable tool to discover novel biomarkers for renal disease and in studying the pathophysiology of CKD [5]. The metabolome integrates the biological information of the genome, transcriptome, proteome, and overall enzymatic reactions of an individual, therefore enabling the detection of short and long-term physiological or pathological changes occurring in diseases.

A recent review by Zhao (2013) gives concise highlights of metabolomics applications in CKD [6], but there are a limited number of metabolomics studies in humans where T2DM is the etiology of CKD. One cross-sectional study found panels of plasma metabolites associated with different CKD stages of renal impairment [7]. However, the study was conducted in patients without diabetes. In a study conducted in patients with T2DM, serum metabolites were identified that were able to distinguish between diabetic nephropathy with macroalbuminuria and diabetic patients without albuminuria [8]. Another study showed that there were significant differences in serum metabolite levels of leucine, dihydrosphingosine, and phytosphingosine in patients with diabetic nephropathy from T2DM patients without nephropathy or healthy controls [9]. Additionally, a panel of 13 urine metabolites linked with mitochondrial metabolism was shown to be significantly reduced in diabetic CKD patients compared to healthy controls [10]. However, the cross-sectional design of these studies did not enable assessment of the predictive value of the metabolites, and to our knowledge, there have been no published, longitudinal, clinical studies assessing the ability of metabolomics to predict progressive renal function loss in people with T2DM.

Therefore, the aim of the current study was to discover potential metabolites that could predict the progression of renal dysfunction in T2DM. To test whether the metabolites were T2DM-specific, the same metabolites were tested in individuals with hypertension without T2DM.

8786


4

METHODS

Study participants and methodsThis prospective, case-control study was performed utilizing samples from individuals participating in the community-based “Prevention of REnal and Vascular End-stage Disease” (PREVEND) study in the Netherlands or who were seen at the Steno Diabetes Center in Denmark. A detailed description of these studies has been previously described [11,12].

Median follow-up was 2.9 years [interquartile range: 2.5-4.0]. T2DM was defined as the use of oral glucose-lowering treatment (self-reported or by information retrieved from the regional pharmacy database), a fasting plasma glucose >7.0 mmol/L, or non-fasting plasma glucose >11.1 mmol/L. Hypertension was defined as the use of antihypertensive treatment (self-reported or by information retrieved from the regional pharmacy database) or a systolic/diastolic blood pressure >140/90 mmHg. Study participants with albuminuria measurements during follow-up, and plasma and urine samples of sufficient quality, were eligible for selection for the present study. Both the PREVEND and Steno studies were approved by their local ethics committees and were conducted in accordance with the guidelines of the Declaration of Helsinki. All participants gave written, informed consent.

Selection of cases and controlsCases were defined as individuals who transitioned from normo- to microalbuminuria or from micro- to macroalbuminuria between two consecutive study visits with at least a 30% increase in urinary albumin excretion (UAE) from baseline. For each case that showed a transition in albuminuria stage, a matched control on T2DM or hypertension, age, gender, and baseline albuminuria stage (i.e. normoalbuminuria or microalbuminuria stage) was selected that most optimally resembled the case on these combined parameters. Steno patients were additionally matched for duration of diabetes. Normoalbuminuria was defined as UAE <30 mg/24hr, microalbuminuria as UAE = 30-300 mg/24hr, and macroalbuminuria as UAE ≥300 mg/24hr. Albuminuria status was based on the average of 2 consecutive measurements in 24-hour urine collections. The use of anti-hypertensive agents intervening in the RAAS was allowed, but the type of drug and dose had to remain stable during the study period. Individuals who discontinued or initiated new RAAS treatment during the follow-up period were excluded from this study. The final T2DM sample consisted of 24 normo- to microalbuminuria case/controls pairs (all from the PREVEND study) and 21 micro- to macroalbuminuria case/controls pairs (8 pairs from the PREVEND study and 12 pairs from the Steno cohort). The final hypertension sample included 50 normo- to microalbuminuria case/control pairs and 25 micro- to macroalbuminuria case/control pairs (all from the PREVEND study).

MeasurementsClinical, plasma and urine chemistry measurements in PREVEND and Steno were performed as described previously [12,13]. Estimated glomerular filtration rate (eGFR) was calculated using the 4-variable Modification of Diet in Renal Disease (MDRD) study equation [14].

Metabolomic analysisUrine samples were derived from 24-hour urine collections and stored in aliquots at minus 20°C. Plasma samples were stored at minus 80°C. All samples were stored for 2-5 years and underwent one freeze-thaw cycle. Metabolomics were performed only on baseline samples. The samples were prepared as previously described [15] and randomized on the plate prior to analysis to avoid potential confounding interaction between concentration and order of injection and to ensure a homogenous between-plate design in regard to study groups. Measurements on plasma and urine samples were performed blinded by Biocrates Life Sciences AG (Innsbruck, Austria). A concise summary of metabolomic analysis can be found in Supplemental Materials: Metabolomic Methods.

Acylcarnitines, glycerophospholipids, sphingomyelins, amino acids, hexoses and biogenic amines were analyzed in plasma und urine, whereas bile acids, eicosanoids and intermediates of energy metabolism were measured only in plasma. All methods have been validated considering FDA guidelines [16]. A total of 277 plasma and 234 urine metabolites were identified.

Statistical analysisAnalyses were performed using SAS version 9.2 and R version 2.14.0 (2011-10-31). Observations with missing data on covariates were not included. Baseline characteristics with normal distribution were reported as mean and standard deviation (SD), characteristics with skewed distribution were reported as median and inter quartile range [IQR], and categorical variables were reported as number and percentage. Variables with skewed distribution were log-transformed for regression analyses. Graphical methods were used to visualize normalization of the distribution after transformation.

Differences between case/control pairs were tested with independent samples t-tests for continuous variables and χ² tests on paired proportions for categorical variables. The median analytical variance of quality control (QC) samples was 8.4% within-plate coefficient of variation (CV) and 11.6% between-plates CV for plasma measurements and 8.5% within-plate CV and 12.0% between-plates CV for urine measurements. A total of 12 plasma and 3 urine metabolites were below the detection limit in >70% of participants and therefore not included in the analysis. Remaining concentration values of 265 plasma and 231 urine metabolites were quantified, log2-transformed, and batch corrected [17].

8988


4

Differences in metabolite concentrations between cases and controls were tested with moderated F- and subsequent pairwise moderated t-tests [18]. Resulting p-values were corrected for multiple testing by false discovery rate [19]. All analyses were adjusted for baseline UAE and eGFR due to significant differences between cases and controls among these measurements. Conditional logistic regression was performed to investigate the association between metabolite concentrations and transition in albuminuria stage. Metabolites with significance of p-value <0.05 were used for prediction models. Additionally, the correlations of the metabolites with changes in UAE and eGFR were tested by linear regression analysis. Changes in albuminuria between follow-up and baseline were calculated as the logarithm of their ratio.

In order to determine whether the metabolites at baseline improved prediction of risk for albuminuria progression, the differences in the area under the receiver operating characteristic curve (AUROC) and the integrated discrimination improvement (IDI) were assessed by adding the metabolites to a reference model on top of baseline UAE and baseline eGFR) [20]. Because there were no significant differences between cases and controls in other clinical risk factors including ethnicity, body mass index (BMI), blood pressure, duration of diabetes, follow-up time, total cholesterol, fasting glucose, and glycated haemoglobin (HbA1c), these variables were not used in prediction models.

RESULTS

Type 2 diabetes mellitus

Baseline characteristicsIn total, 45 T2DM case/control pairs were available for analysis. Baseline characteristics of T2DM cases and controls are shown in Table 1. Despite matching for baseline albuminuria stage, a significant difference in the absolute value of baseline UAE between T2DM cases and controls persisted in both subgroups (Table 1). Furthermore, eGFR was higher in T2DM controls than in the respective cases, but only in the micro- to macroalbuminuria subgroup was this difference statistically significant (Table 1).

T2DM cases with normo- or microalbuminuria had a median percentage increase in albuminuria of 142.1% and 340.9%, respectively. T2DM controls with baseline normo- or microalbuminuria had a median percentage change in albuminuria of 1.2% and -8.2%, respectively.

Differences in metabolite concentrations between cases and controls in T2DMNo metabolites were found to differ significantly between T2DM cases and controls in the normo- to microalbuminuria subgroup after correcting for multiple testing and adjusting for baseline UAE and eGFR. There were significant differences in metabolite concentrations only in the micro- to macroalbuminuria subgroup (Table 2). Therefore, we focused our further analysis on the micro- to macroalbuminuria subgroup. Figure 1 presents exemplary chromatograms for these metabolites. In plasma, butenoylcarnitine was significantly higher (fold change 1.17, p=0.007, Figure 2a) whereas histidine was significantly lower (fold change 0.87, p=0.02, Figure 2b) in cases compared to controls. In urine, concentrations of hexose (fold change 0.20, p<0.001, Figure 3a), glutamine (fold change 0.32, p<0.001, Figure 3b), and tyrosine (fold change 0.51, p=0.006, Figure 3c) were significantly lower in cases compared to controls.

Predictive performance of metabolites in T2DM for transition in albuminuria stage To assess the predictive performance of the selected biomarkers in the T2DM micro- to macroalbuminuria subgroup, we calculated the AUROC and IDI of a reference model consisting of baseline UAE and eGFR (AUROC=0.84). We then added the urine or plasma metabolites to the reference model. Addition of the plasma metabolites to the reference model resulted in an AUROC increase of 0.10 (p=0.02, Table 3) and an IDI of 0.28 (p<0.001, Table 3). Addition of the urine metabolites to the reference model resulted in an AUROC increase of 0.15 (p=0.009, Table 3) and an IDI of 0.43 (p<0.001, Table 3). In terms of predictive performance, the combination of plasma and urine metabolites was not superior to the model containing the urine metabolites alone (not shown).

9190


4

Tab

le 1

. Bas

elin

e ch

arac

teris

tics

in p

atie

nts

with

typ

e 2

dia

bet

es m

ellit

us w

ith t

rans

ition

in a

lbum

inur

ia s

tage

No

rmo

- to

Mic

roal

bum

inur

iaM

icro

- to

Mac

roal

bum

inur

ia

Cas

esC

ontr

ols

Cas

esC

ontr

ols

Num

ber

2424

2121

Age

(yea

rs)

64.8

± 1

062

.3 ±

9.5

65.6

± 6

.963

.2 ±

9.1

Mal

e G

end

er (%

)17

(70.

8)17

(70.

8)18

(86)

18 (8

6)

Cau

casi

an (%

)22

(91.

7)22

(91.

7)20

(95)

21 (1

00)

Sm

okin

g (%

)5

(20.

8)5

(20.

8)6

(28.

6)5

(23.

8)

BM

I (kg

/m2 )

29.9

± 6

.927

.9 ±

4.1

31.6

± 4

.729

.1 ±

5.5

SB

P (m

mH

g)13

6.4

± 1

4.3

134.

4 ±

21.

413

5.8

± 1

4.8

130.

8 ±

15.

3

DB

P (m

mH

g)75

.0 ±

9.2

76.2

± 8

.172

.8 ±

9.7

74.7

± 9

.0

Dur

atio

n of

dia

bet

es (y

ears

)N

/A†

N/A

†17

.2 ±

8.7

17.3

± 8

.6

Follo

w-u

p t

ime

(yea

rs)

2.7

[2.4

-4.1

]2.

8 [2

.4-4

.1]

3.6

[2.8

-4.3

]2.

9 [2

.5-3

.4]

Lab

orat

ory

mea

sure

men

ts

UA

E (m

g/24

hr)

19.6

[15.

4-22

.4]

7.4

[6.0

-12.

8]**

*16

3.0

[112

.0-1

96.0

]78

.0 [4

5.0-

114.

0]*

eG

FR (m

L/m

in/1

.73m

2 )77

.6 ±

17.

183

.9 ±

16.

770

.8 ±

15.

382

.3 ±

21.

1*

Tot

. Cho

lest

erol

(mm

ol/L

)5.

1 ±

1.2

5.1

± 1

.44.

3 ±

1.2

4.6

± 1

.3

Fas

ting

gluc

ose

(mm

ol/L

)7.

7 ±

1.8

7.2

± 1

.27.

4 ±

2.3

‡7.

2 ±

1.2

‡

Hb

A1c

(%)

N/A

†N

/A†

8.1

± 1

.48.

2 ±

0.9

8

Hb

A1c

(mm

ol/m

ol)

N/A

†N

/A†

65 ±

866

± 1

3

UA

E c

hang

es (%

)14

2.1

[101

.1-2

06.7

]1.

2 [-

20.4

-30.

1]**

*34

0.9

[161

.1-8

42.4

]-8

.2 [-

28.7

-67.

6]**

Med

icat

ion

usag

e

AC

Ei/A

RB

10 (4

1.7)

2 (8

.3)*

15 (7

1.4)

12 (5

7.1)

Ant

ihyp

erte

nsiv

e16

(66.

7)8

(33.

3)*

19 (9

0.5)

15 (7

1.4)

Ora

l glu

cose

-low

erin

g18

(75.

0)15

(62.

5)14

(66.

7)16

(76.

2)

Lip

id-l

ower

ing

13 (5

4.2)

4 (1

6.7)

*15

(71.

4)15

(71.

4)

Dat

a re

por

ted

as

mea

n ±

SD

or n

umb

er (%

) or m

edia

n [IQ

R]. C

ases

vs.

Con

trol

s: *

p<

0.05

, **p

<0.

01, *

**p

<0.

001.

† Dur

atio

n of

dia

bet

es a

nd

Hb

A1c

wer

e no

t ava

ilab

le in

the

PR

EV

EN

D s

tud

y. O

nly

PR

EV

EN

D p

artic

ipan

ts w

ere

incl

uded

in th

e no

rmo-

to m

icro

alb

umin

uria

sub

grou

p.

‡ Fas

ting

pla

sma

gluc

ose

was

onl

y av

aila

ble

in P

RE

VE

ND

par

ticip

ants

. B

MI:

bod

y m

ass

ind

ex (

wei

ght

kg/h

eigh

t m

2 );

eGFR

: es

timat

ed

glom

erul

ar fi

ltrat

ion

rate

(4-v

aria

ble

Mod

ifica

tion

of D

iet i

n R

enal

Dis

ease

form

ula)

; AC

Ei:

angi

oten

sin-

conv

ertin

g en

zym

e in

hib

itors

; AR

B:

angi

oten

sin-

2 re

cep

tor

blo

cker

s; U

AE

: ur

inar

y al

bum

in e

xcre

tion.

Cha

nges

in U

AE

are

cal

cula

ted

as:

[(Fo

llow

-up

UA

E –

Bas

elin

e U

AE

)/B

asel

ine

UA

E]*

100%

.

9392


4

Figure 1. Exemplary extracted ion chromatograms (EIC) of histidine, glutamic acid and tyrosine in plasma and urine for T2DM cases (red) and controls (controls) in the micro- to macroalbuminuria subgroup. (a) EIC of histidine in plasma, (b) EIC of glutamine in urine, and (c) EIC of tyrosine in urine samples. As butenoylcarnitine and hexose were measured applying FIA-MS/MS (flow-injection coupled to mass spectrometry) no meaningful chromatogram is available for these metabolites.

Figure 2. Box plots of significant plasma metabolites corrected for multiple testing after adjusting for baseline UAE and eGFR. Concentrations of the two significant metabolites butenoylcarnitine (a) and histidine (b) in T2DM cases and controls in the micro- to macroalbuminuria subgroup are depicted in box plots.

9594


4

Figure 3. Box plots of significant urine metabolites corrected for multiple testing after adjusting for baseline UAE and eGFR. Concentrations of the three significant metabolites hexose (a), glutamine (b), and tyrosine (c) in T2DM cases and controls in the micro- to macroalbuminuria subgroup are depicted in box plots.

Tab

le 2

. Con

cent

ratio

ns [µ

M] o

f sig

nific

ant

pla

sma

and

urin

e m

etab

olite

s in

pat

ient

s w

ith t

ype

2 d

iab

etes

mel

litus

: Mic

ro-

to m

acro

alb

umin

uria

sub

grou

p (n

=42

).

Bio

flui

dM

etab

olit

eC

hem

ical

Cla

ssM

etab

olit

e co

ncen

trat

ion

[µM

] C

ases

Met

abo

lite

conc

entr

atio

n [µ

M]

Co

ntro

lsFo

ld C

hang

ep

-val

ue*

Pla

sma

But

enoy

lcar

nitin

eA

cylc

arni

tines

0.02

0 [0

.018

, 0.0

24]

0.01

7 [0

.015

, 0.0

18]

1.17

0.00

7

Pla

sma

His

tidin

eA

min

o ac

ids

64.3

[58.

8, 7

2.8]

73.9

[67.

3, 8

1.4]

0.87

0.02

Urin

eH

exos

eS

ugar

s11

69.0

[378

.2, 4

003.

7]58

44.5

[134

2.5,

129

88.1

]0.

20<

.001

Urin

eG

luta

min

eA

min

o ac

ids

73.7

[45.

2, 1

14.6

]23

2.7

[156

.1, 3

34.8

]0.

32<

.001

Urin

eTy

rosi

neA

min

o ac

ids

26.2

[15.

7, 4

7.3]

51.2

[33.

8, 9

4.7]

0.51

0.00

6

Con

cent

ratio

ns a

re e

xpre

ssed

as

med

ian

[IQR

].

*P-v

alue

s ar

e co

rrec

ted

for

mul

tiple

tes

ting

and

ad

just

ed fo

r b

asel

ine

UA

E a

nd e

GFR

.

Tab

le 3

. Are

a un

der

the

Rec

eive

r Op

erat

ing

Cha

ract

eris

tic c

urve

(AU

RO

C) a

nd In

tegr

ated

Dis

crim

inat

ion

Ind

ex (I

DI)

for c

ond

ition

al lo

gist

ic re

gres

sion

mod

els

pre

dic

ting

tran

sitio

n in

alb

umin

uria

sta

ge in

pat

ient

s w

ith t

ype

2 d

iab

etes

mel

litus

: Mic

ro-

to m

acro

alb

umin

uria

sub

grou

p (n

=42

)

AU

RO

C95

% C

Ip

-val

ueID

I95

% C

Ip

-val

ue

Ref

eren

ce m

odel

*0.

840.

72, 0

.96

ref.

ref.

ref.

ref.

+ P

lasm

a m

etab

olite

s†0.

940.

89, 1

.00

0.02

0.28

0.15

, 0.4

1<

0.00

1

+ U

rine

met

abol

ites‡

0.99

0.96

, 1.0

00.

009

0.43

0.28

, 0.5

8<

0.00

1

*Bas

elin

e U

AE

and

eG

FR; † B

uten

oylc

arni

tine

and

His

tidin

e; ‡ H

exos

e, G

luta

min

e, a

nd T

yros

ine

9796


4

Table 4. Correlations between the significant plasma and urine metabolites with changes in UAE and in eGFR in patients with type 2 diabetes mellitus: Micro- to macroalbuminuria subgroup (n=42).

β Coefficient Standard Error R2 p-value

Plasma

Butenoylcarnitine vs. Delta UAE 1.4 0.55 0.14 0.02

Butenoylcarnitine vs. Delta eGFR -15.9 6.1 0.15 0.01

Histidine vs. DeltaUAE -1.5 0.60 0.13 0.02

Histidine vs. Delta eGFR 8.0 7.1 0.03 0.26

Urine

Glutamine vs. Delta UAE -0.22 0.10 0.12 0.03

Glutamine vs. Delta eGFR 2.8 1.2 0.12 0.02

Tyrosine vs. Delta UAE -0.15 0.12 0.04 0.22

Tyrosine vs. Delta eGFR -0.67 1.4 0.01 0.63

Hexose vs. Delta UAE -0.07 0.06 0.03 0.30

Hexose vs. Delta eGFR 0.56 0.74 0.01 0.46

Delta UAE is measured as the log of the ratio between follow-up and baseline UAE values. Delta eGFR is computed as the difference between the follow-up and baseline eGFR values.

Correlations of the metabolites with changes in UAE and eGFR in T2DMIn the T2DM micro- to macroalbuminuria subgroup, significant correlations were found between the metabolites and changes in UAE and eGFR during follow-up. In plasma, butenoylcarnitine correlated positively with changes in UAE, and negatively with changes in eGFR (p=0.02 and p=0.01, respectively, Table 4). Histidine was negatively correlated with changes in UAE (p=0.02, Table 4). In urine, glutamine was negatively correlated with changes in UAE and positively correlated with changes in eGFR (p=0.03 and p=0.02, respectively, Table 4).

Hypertension

Metabolite concentrations in hypertensive cases and controlsIn total, 75 hypertensive case/control pairs were available for analysis. The baseline characteristics of T2DM cases and controls are shown in Supplemental Table 1. Hypertensive cases who transitioned from normo- to microalbuminuria or from micro- to macroalbuminuira had a median percentage increase in albuminuria of 281.6% and 373.4%, respectively. In hypertensive controls with baseline normo- or microalbuminuria, median changes in albuminuria were -2.6% and 40.5%, respectively (Supplemental Table 1). The concentrations of the significant metabolites observed in the T2DM subgroup are listed in Supplemental Table 2 for the hypertensive population. No significant differences in these metabolites were observed between cases and controls with hypertension (Supplemental Table 2).

DISCUSSION

We found 3 urine and 2 plasma metabolites to predict progression from micro- to macroalbuminuria on top of established renal risk markers in T2DM. These metabolites did not predict albuminuria progression in hypertensive patients, suggesting a T2DM-specific metabolite profile.

How do the specific metabolites relate to pathophysiological processes of T2DM? T2DM is characterized by a dysregulation of fatty acid, glucose, and amino acid metabolism induced by elevated lipid deposition in β-cells, liver and skeletal muscle [21,22].

First, regarding butenoylcarnitine, lipid deposition seems to contribute to insulin resistance and is associated with elevated plasma acylcarnitine concentrations [22]. Indeed, we found elevated concentrations of the acylcarnitine butenoylcarnitine in plasma of patients with T2DM who transitioned from micro- to macroalbuminuria. Acylcarnitines, especially short-chain acylcarnitine such as butenoylcarnitine, accumulate in plasma due to the excessive yet incomplete mitochondrial oxidation of fatty acids [22], possibly due to a lower mitochondrial number, volume, or reduced mitochondrial oxidation capacity in T2DM tissues [23]. Alteration in the acylcarnitine pattern has recently been demonstrated in a prospective study of patients with type 1 diabetes where a specific urinary acylcarnitine metabolite profile was able to clearly discriminate between patients who progressed in albuminuria stage after 5 years of follow-up; however, no plasma acylcarnitines were measured [24]. In a cross-sectional study in the general population, an inverse correlation was found between serum acylcarnitines and eGFR [25]. In agreement with this, we reported a negative correlation of plasma butenoylcarnitine with changes in eGFR in patients with T2DM. Additionally, butenoylcarnitine positively correlated with continuous changes in UAE indicating that butenoylcarnitine concentrations increase with progressive renal function loss.

Second, we found lower concentrations of urinary glutamine in patients with T2DM who transitioned from micro- to macroalbuminuria. Glutamine is extracted by the kidney and metabolized to produce urinary ammonium [26]. Ammonia released by renal cells has been shown to normalize the diabetic-associated tubular acidotic milieu [26]. We may therefore speculate that observed decreases in urinary glutamine concentration may be due to the worsening of acidosis during progressive renal impairment. This hypothesis is supported by the finding that urinary glutamine was inversely correlated with changes in UAE during follow-up, pointing at glutamine as a candidate for future mechanistic studies.

Third, plasma histidine was significantly lower in patients with T2DM who transitioned from micro- to macroalbuminuria. Histidine has been shown to exert anti-inflammatory and anti-oxidant activity based on its ability to scavenge free radicals and to chelate divalent metal ions [27]. Plasma histidine concentrations were found to be significantly lower in CKD patients with signs of ongoing inflammation and associated with greater mortality in the same patients [28]. Dietary histidine supplementation seems to reduce oxidative and

9998


4

inflammatory disease pathway, a strategy which holds promise in the treatment of renal disease [29]. A recent study demonstrated that albuminuria significantly correlated with oxidant and antioxidant marker levels in early diabetic nephropathy [30]. In agreement with this study, we found a negative correlation between the antioxidant marker histidine and changes in UAE during follow-up, suggesting that decreases in histidine may be also related to progressive renal dysfunction.

Fourth, urinary tyrosine was significantly lower in T2DM micro- to macroalbuminuria cases versus controls. Tyrosine is a semi-essential amino acid that is only synthetized by the hydroxylation of phenylalanine by the enzyme phenylalanine hydroxylase (PAH). Chronic renal impairment is associated with decreased plasma concentrations of tyrosine whereas plasma concentrations of phenylalanine are often slightly increased, indicating impairment in the conversion of phenylalanine to tyrosine due to dysfunctional activity of PAH [31]. Moreover, PAH activity depends on the availability of its essential co-factor tetrahydrobiopterin (BH4), and underlying oxidative stress occurring in diabetes may limit BH4 availability and impair PAH activity [32]. Further studies are needed to investigate these possibilities in T2DM.

Fifth, urinary hexose was decreased in patients with T2DM who transitioned from micro- to macroalbuminuria. However, the importance of hexose needs to be confirmed in larger cohort studies where correction for insulin and glucose concentrations is possible.

An important question to be answered, especially in view of therapeutic purposes, is whether the metabolites are involved in the causal pathway leading to the disease (and its progression) or are merely the consequence of existing pathological processes. Several studies have shown that certain amino acids may be either a cause or a very early manifestation of insulin resistance that precedes the onset of diabetes by years [33]. Acylcarnitines may reflect the abnormal lipid metabolism in diabetes and have been shown to directly activate inflammatory pathways leading to inhibition of insulin action [34]. Inflammation contributes to lower plasma amino acid concentrations in CKD [35]. However, some amino acids such as histidine may directly affect inflammatory and oxidative pathways [28], indicating that alterations in those amino acids may represent a cause of chronic inflammation, oxidative stress, and progressive renal disease. As both inflammation and (lipid) oxidative stress caused by hyperglycemia seem to contribute to the pathogenesis of micro- and macroalbuminuria, we cannot exclude that the urine and plasma metabolites may also be a direct cause of albuminuria worsening over time. Indeed, some metabolites correlated significantly with changes in UAE during follow-up. The precise mechanisms of this process are currently not known but may involve the activation of some established pathways of renal damage such as nuclear factor kappa-light-chain-enhancer of activated B cells (Nf-kB), protein kinase C (PKC) and mitogen-activated protein kinases (MAPK) signaling pathways [36].

In our study, the urinary metabolome profile performed better than the plasma profile, and addition of the plasma metabolites to the urinary metabolites did not improve outcome prediction. This may point at urinary metabolomics as a better clinical approach to identify people with T2DM at risk of progressive renal disease (also due to practical advantages of collecting urine compared to blood samples). However, the results from the plasma metabolomics may be useful to identify underlying mechanisms and pathways of disease and should not be neglected.

The same study sample was recently used to validate the ability of a proteomic classifier to predict transition in albuminuria [13]. A direct connection between the proteome and metabolome in T2DM still needs to be determined. However, we did find similarities in the inflammatory profile (histidine in metabolomics vs. α-2-HS-glycoprotein and α-1-antitrypsin in proteomics), which supports the complementary use of proteomic and metabolomic techniques as a tool to reduce the burden of diabetes complications. One may interpret our results as if metabolomics is a stronger tool than proteomics due to the higher observed AUROC. It must be noted, however, that significant metabolites were observed only in the micro- to macroalbuminuria subgroup, whereas the proteomic classifier was significant in both normo- to microalbuminuria and micro- to macroalbuminuria subgroups. Importantly, in the micro- to macroalbuminuria subgroup, proteomics seemed to perform at best with an AUROC=0.99 (unpublished data).

There are limitations to this study. Due to the early stage of disease being investigated and the strict criteria used for matching cases and controls, we were able to select only a limited number of participants from the original large cohorts, limiting statistical power. Nevertheless, this is to our knowledge, the largest longitudinal discovery study to investigate prediction of renal disease progression in T2DM by means of metabolomic analysis. We are aware that these findings need to be validated in a larger prospective cohort. Nearly all clinical baseline variables were similar between cases and controls; however, there was a significant difference in the absolute values of UAE and eGFR between cases and controls. The imbalance in baseline UAE was unintended and represents our best efforts in matching cases and controls with the same albuminuria stage from the original cohorts. To account for this imbalance, we consistently adjusted our analysis for baseline UAE and eGFR. Another limitation was that the true predictive capacity of the models may be overestimated due to the relative high event rate which occurs in case-control studies and due to the prediction model being developed and tested in the same sample. Absolute values for the AUROC should therefore be interpreted with caution. The additive value of the metabolites to the AUROC of the reference model is however unaffected by the case-control design of the study. Additional limitations include the lack of information concerning the insulin-state, diet, and medication type and dose, which clearly represent unmeasured confounders in our study.

101100


4

In conclusion, we discovered T2DM-specific plasma and urine metabolites as potential biomarkers for prediction of progression from micro- to macroalbuminuria. These metabolites had significant value in addition to conventional renal risk markers, and may contribute to the early identification and treatment of individuals with T2DM at risk for progression of renal disease. Our results, if confirmed in larger prospective cohort studies, could eventually open the door for an omic-based risk assessment in primary prevention of renal disease in T2DM.

ACKNOWLEDGMENTS

We thank U. Lundin and K. Weinberger for their support in the early stage of this study.

FUNDING

The PREVEND Study has been made possible by grants of the Dutch Kidney Foundation. The work leading to this paper has received funding from the European Community’s Seventh Framework Programme under grant agreement no. HEALTH–F2–2009–241544 (SysKID consortium).

REFERENCES










3. Viazzi F, Leoncini G, Conti N, Tomolillo C, Giachero G, Vercelli M, Deferrari G, Pontremoli R.

Microalbuminuria is a predictor of chronic renal insufficiency in patients without diabetes

and with hypertension: the MAGIC study. Clin J Am Soc Nephrol. 2010 Jun;5(6):1099-106.






5. Zhao YY, Cheng XL, Wei F, Bai X, Tan XJ, Lin RC, Mei Q. Intrarenal metabolomic

investigation of chronic kidney disease and its TGF-beta1 mechanism in induced-adenine

rats using UPLC Q-TOF/HSMS/MS(E). J Proteome Res. 2013 Feb 1;12(2):692-703.

6. Zhao YY. Metabolomics in chronic kidney disease. Clin Chim Acta. 2013 Jun 25;422:59-69.

7. Shah VO, Townsend RR, Feldman HI, Pappan KL, Kensicki E, Vander Jagt DL. Plasma

Metabolomic Profiles in Different Stages of CKD. Clin J Am Soc Nephrol. 2013

Mar;8(3):363-70.

8. Hirayama A, Nakashima E, Sugimoto M, Akiyama S, Sato W, Maruyama S, Matsuo

S, Tomita M, Yuzawa Y, Soga T. Metabolic profiling reveals new serum biomarkers for

differentiating diabetic nephropathy. Anal Bioanal Chem. 2012 Dec;404(10):3101-9.

9. Zhang J, Yan L, Chen W, Lin L, Song X, Yan X, Hang W, Huang B. Metabonomics research

of diabetic nephropathy and type 2 diabetes mellitus based on UPLC-oaTOF-MS system.

Anal Chim Acta. 2009 Sep 14;650(1):16-22.





J Am Soc Nephrol. 2013 Nov;24(11):1901-12.

11. Pinto-Sietsma SJ, Janssen WM, Hillege HL, Navis G, De Zeeuw D, De Jong PE. Urinary

albumin excretion is associated with renal functional abnormalities in a nondiabetic

population. J Am Soc Nephrol. 2000 Oct;11(10):1882-8.

103102


4

12. Reinhard H, Hansen PR, Persson F, Tarnow L, Wiinberg N, Kjær A, Petersen CL, Winther

K, Parving HH, Rossing P, Jacobsen PK. Elevated NT-proBNP and coronary calcium

score in relation to coronary artery disease in asymptomatic type 2 diabetic patients with

elevated urinary albumin excretion rate. Nephrol Dial Transplant. 2011 Oct;26(10):3242-9.




2013 Feb;56(2):259-67.




16;130(6):461-70.

15. Solberg R, Enot D, Deigner HP, Koal T, Scholl-Bürgi S, Saugstad OD, Keller M. Metabolomic

Analysis of Plasma Reveals New Insights into Asphyxia and Resuscitation in Pigs. PLoS

One. 2010 Mar 9;5(3):e9606.

16. Ulrika Lundin, Robert Modre-Osprian and Klaus M. Weinberger (2011). Targeted

Metabolomics for Clinical Biomarker Discovery in Multifactorial Diseases, Advances in

the Study of Genetic Disorders, Dr. Kenji Ikehara (Ed.).

17. Johnson WE, Rabinovic A, Li C. Adjusting batch effect in microarray expression data

using Empirical Bayes methods. Biostatistics. 2007 Jan;8(1):118-27.

18. Smyth G. Linear models and emprical Bayes methods for assessing differential expression

in microarray experiments. Stat Appl Genet Mol Biol. 2004;3:Article3.

19. Benjamini Y, Hochberg Y. Controlling the False Discovery Rate: A Practical and Powerful

Approach to Multiple Testing. J R Stat Soc Series B Stat Methodol 1995;57(1):289-300.

20. Pencina MJ, D’Agostino Sr. RB, D’Agostino Jr. RB, Vasan RS. Evaluating the added

predictive ability of a new marker: From area under the ROC curve to reclassification and

beyond. Stat Med. 2008 Jan 30;27(2):157-72.

21. Unger RH. Lipotoxic diseases. Annu Rev Med. 2002;53:319-36.

22. Mihalik SJ, Goodpaster BH, Kelley DE, Chace DH, Vockley J, Toledo FG, DeLany JP.

Increased levels of plasma acylcarnitines in obesity and type 2 diabetes and identification

of a marker of glucolipotoxicity. Obesity (Silver Spring). 2010 Sep;18(9):1695-700.

23. Kelley DE, He J, Menshikova EV, Ritov VB. Dysfunction of mitochondria in human skeletal

muscle in type 2 diabetes. Diabetes. 2002 Oct;51(10):2944-50.

24. van der Kloet FM, Tempels FW, Ismail N, van der Heijden R, Kasper PT, Rojas-Cherto M,

van Doorn R, Spijksma G, Koek M, van der Greef J, Mäkinen VP, Forsblom C, Holthöfer H,

Groop PH, Reijmers TH, Hankemeier T. Discovery of early-stage biomarkers for diabetic

kidney disease using ms-based metabolomics (FinnDiane study). Metabolomics. 2012

Feb;8(1):109-119.

25. Goek ON, Döring A, Gieger C, Heier M, Koenig W, Prehn C, Römisch-Margl W, Wang-

Sattler R, Illig T, Suhre K, Sekula P, Zhai G, Adamski J, Köttgen A, Meisinger C. Serum

metabolite concentrations and decreased GFR in the general population. Am J Kidney

Dis. 2012 Aug;60(2):197-206.

26. Van Slyke DD, Phillips RA, Hamilton PB, Archibald RM, Flucher PH, Hiller A. Glutamine as

source material of urinary ammonia. J Biol Chem 1943;150:481-482.

27. Babizhayev MA, Seguin MC, Gueyne J, Evstigneeva RP, Ageyeva EA, Zheltukhina GA.

L-Carnosine (Beta-Alanyl-L-Histidine) and Carcinine (Beta-Alanylhistamine) Act as Natural

Antioxidants with Hydroxyl-Radical-Scavenging and Lipid-Peroxidase Activities. Biochem

J. 1994 Dec 1;304 (Pt 2):509-16.

28. Watanabe M, Suliman ME, Qureshi AR, Garcia-Lopez E, Bárány P, Heimbürger O,

Stenvinkel P, Lindholm B. Consequences of low plasma histidine in chronic kidney disease

patients: associations with inflammation, oxidative stress, and mortality. Am J Clin Nutr.

2008 Jun;87(6):1860-6.

29. Lee YT, Hsu CC, Lin MH, Liu KS, Yin MC. Histidine and carnosine delay diabetic,

deterioration in mice and protect human low density lipoprotein against oxidation and

glycation. Eur J Pharmacol. 2005 Apr 18;513(1-2):145-50.

30. Shao N, Kuang HY, Wang N, Gao XY, Hao M, Zou W, Yin HQ. Relationship between

Oxidant/Antioxidant Markers and Severity of Microalbuminuria in the Early Stage of

Nephropathy in Type 2 Diabetic Patients. J Diabetes Res. 2013;2013:232404.

31. Kopple JD. Phenylalanine and tyrosine metabolism in chronic kidney failure. J Nutr. 2007

Jun;137(6 Suppl 1):1586S-1590S.

32. Werner ER, Blau N, Thoeny B. Tetrahydrobiopterin: biochemistry and pathophysiology.

Biochem J. 2011 Sep 15;438(3):397-414.

33. Wang TJ, Larson MG, Vasan RS, Cheng S, Rhee EP, McCabe E, Lewis GD, Fox CS, Jacques

PF, Fernandez C, O’Donnell CJ, Carr SA, Mootha VK, Florez JC, Souza A, Melander O,

Clish CB, Gerszten RE. Metabolite profiles and the risk of developing diabetes. Nat Med.

2011 Apr;17(4):448-53.

34. Adams SH. Emerging perspectives on essential amino acid metabolism in obesity and the

insulin-resistant state. Adv Nutr. 2011 Nov;2(6):445-56.

35. Suliman ME, Qureshi AR, Stenvinkel P, Pecoits-Filho R, Bárány P, Heimbürger O,

Anderstam B, Rodríguez Ayala E, Divino Filho JC, Alvestrand A, Lindholm B. Inflammation

contributes to low plasma amino acid concentrations in patients with chronic kidney

disease. Am J Clin Nutr. 2005 Aug;82(2):342-9.

36. Satchell SC, Tooke JE. What is the mechanism of microalbuminuria in diabetes: a role for

the glomerular endothelium? Diabetologia. 2008 May;51(5):714-25.

105104


4

SUPPLEMENTAL MATERIALS: METABOLOMIC METHODS

Amino acids, acylcarnitines, sphingomyelines, phosphtidylcholines, hexose, biogenic amines: The quantification of amino acids, acylcarnitines, sphingomyelins, phosphatidylcholines, hexose (glucose), and biogenic amines was performed using a AbsoluteIDQTM p180 kit [1]. The assay was based on PITC (phenylisothiocyanate)-derivatization in the presence of internal standards followed by FIA-MS/MS (acylcarnitines, lipids, and hexose) and LC/MS (amino acids, biogenic amines) using an API4000 QTrap® mass spectrometer (Applied Biosystems/MDS Analytical Technologies, Darmstadt, Germany) with electrospray ionization. Multiple reaction monitoring (MRM) detection was used for quantification applying the spectra parsing algorithm integrated into the MetIQ software (Biocrates Life Sciences AG, Innsbruck, Austria).

Eicosanoids and oxidized fatty acids (prostaglandins): Eicosanoids and other oxidized polyunsaturated fatty acids were extracted from samples with aqueous acetonitrile that contained deuterated internal standards. The metabolites were determined by HPLC-tandem mass spectrometry (LC-MS/MS) with Multiple Reaction Monitoring (MRM) in negative mode using a API4000 QTrap® mass spectrometer with electrospray ionization (Applied Biosystems/MDS Analytical Technologies, Darmstadt, Germany). The LC-MS/MS method used for the analytical determination of eicosanoids has been published [2].

Energy metabolism intermediates: For the quantitative analysis of energy metabolism intermediates (glycolysis, citrate cycle, pentose phosphate pathway, urea cycle) hydrophilic interaction liquid chromatography (HILIC)-ESI-MS/MS method in negative MRM detection mode was used. The MRM detection was performed using a API4000 QTrap® tandem mass spectrometry instrument (Applied Biosystems/MDS Analytical Technologies, Darmstadt, Germany). The sample was protein precipitated and extracted simultaneously with aqueous methanol in a 96 well plate format. Internal standards (ratio external to internal standard) and external calibration were used for highly accurate quantitation.

Bile acids: A reversed phase LC-MS/MS analysis method in negative MRM detection mode was applied to determine the concentration of bile acids in plasma and urine samples. Samples were extracted via dried filter spot technique in 96 well plate format. For quantification, internal standards and external calibration were applied. In brief, internal standards and 20 µL sample volume placed onto filter spots were extracted and simultaneously protein precipitated with aqueous methanol. These sample extracts were measured by LC-ESI-MS/MS with an AB SCIEX 4000 QTrap™ tandem mass spectrometry instrument (Applied Biosystems/MDS Analytical Technologies, Darmstadt, Germany) in negative MRM detection mode.

LC-MS/MS data were quantified with Analyst 1.5.1 software (Applied Biosystems, Darmstadt, Germany) and finally exported for comprehensive statistical analysis.

References. Supplemental Materials1. Biocrates Life Sciences AG. Research Products. AbsoluteIDQ p180 Kit. Available at http://

www.biocrates.com/products/research-products/absoluteidq-p180-kit. Last accessed 18

November 2013.

2. Unterwurzacher I, Koal T, Bonn GK, Weinberger KM, Ramsay SL. Rapid sample

preparation and simultaneous quantitation of prostaglandins and lipoxygenase derived

fatty acid metabolites by liquid chromatography-mass spectrometry from small sample

volumes. Clin Chem Lab Med 2008;46(11):1589-1597.

107106


4

Sup

ple

men

tal T

able

1. B

asel

ine

char

acte

ristic

s in

pat

ient

s w

ith h

yper

tens

ion

with

out

T2D

M w

ith t

rans

ition

in a

lbum

inur

ia s

tage

No

rmo

-Mic

roM

icro

-Mac

ro

Cas

esC

ontr

ols

Cas

esC

ontr

ols

Num

ber

5050

2525

Age

(yea

rs)

66.5

(±8.

6)66

.5 (±

8.2)

65.9

(±12

.0)

63.8

(±9.

4)

Mal

e G

end

er33

(66)

33 (6

6)20

(80)

20 (8

0)

Cau

casi

an (%

)49

(98)

49 (9

8)22

(88)

24 (9

6)

Sm

okin

g (%

)11

(22)

6 (1

2)7

(28)

7 (2

8)

BM

I (kg

/m2 )

28.7

(±5.

0)27

.9 (±

4.8)

28.1

(±3.

1)27

.9 (±

3.8)

SB

P (m

mH

G)

137.

9 (±

18.9

)13

6 (±

15.

5)14

1.0

(±19

.9)

140.

4 (±

19.2

)

DB

P (m

mH

G)

75.6

(±7.

1)77

.1 (±

9.6)

79.1

(±9.

4)79

.6 (±

9.3)

Follo

w-u

p t

ime

(yea

rs)

2.7

[2.2

-3.0

]2.

7 [2

.1-2

.9]

3.6

[2.8

-4.2

]2.

9 [2

.2-4

.1]

Lab

orat

ory

mea

sure

men

ts

UA

E (m

g/24

hrs

)16

.5 [1

1.0-

22.5

]8.

2 [6

.5-1

0.6]

***

120.

2 [8

3.1-

154.

1]51

.3 [3

7.3-

110.

3]**

eGFR

(mL/

min

/1.7

3m2)

73.6

(±17

.4)

75.1

(±18

.6)

60.2

(±21

.3)

69.1

(±14

.4)

Tot.

Cho

lest

erol

(mm

ol/L

)5.

4 (±

1.1)

5.3

(±1.

0)5.

2 (±

0.8)

5.5

(1.2

)

Fast

ing

gluc

ose

(mm

ol/L

)5.

3 (±

0.7)

5.1

(±0.

8)5.

3 (±

1.0)

5.2

(±1.

0)

UA

E c

hang

es (%

)28

1.6

[136

.1-5

26.8

]-2

.6 [-

18.7

-18.

5]**

*37

3.4

[188

.7-7

18.0

]40

.5 [-

1.6-

69.6

]***

Med

icat

ion

usag

e

Ant

ihyp

erte

nsiv

e (%

)46

(92)

49 (9

8)23

(92)

25 (1

00)

AC

Ei/A

RB

(%)

25 (5

0)25

(50)

14 (5

6)14

(56)

Lip

id lo

wer

ing

(%)

18 (3

6)12

(24)

9 (3

6)8

(32)

Dat

a re

por

ted

as

mea

n ±

SD

or n

umb

er (%

) or m

edia

n [IQ

R]. C

ases

vs.

Con

trol

s: *

p<

0.05

, **p

<0.

01, *

**p

<0.

001;

BM

I: b

ody

mas

s in

dex

(wei

ght k

g/he

ight

m2 )

; eG

FR:

estim

ated

glo

mer

ular

filtr

atio

n ra

te (4

-var

iab

le M

odifi

catio

n of

Die

t in

Ren

al D

isea

se f

orm

ula)

; A

CE

i: an

giot

ensi

n-co

nver

ting

enzy

me

inhi

bito

rs;

AR

B:

angi

oten

sin-

2 re

cep

tor

blo

cker

s; U

AE

: urin

ary

alb

umin

exc

retio

n. C

hang

es in

UA

E a

re c

alcu

late

d a

s: [(

Follo

w-u

p U

AE

– B

asel

ine

UA

E)/

Bas

elin

e U

AE

]*10

0%.

Sup

ple

men

tal

Tab

le 2

. C

once

ntra

tions

[µM

] of

pla

sma

and

urin

e m

etab

olite

s in

pat

ient

s w

ith h

yper

tens

ion

with

out

T2D

M:

Mic

ro-

to m

acro

alb

umin

uria

sub

grou

p (n

=50

).

Bio

flui

dM

etab

olit

eC

hem

ical

Cla

ssM

etab

olit

e co

ncen

trat

ion

[µM

] C

ases

Met

abo

lite

conc

entr

atio

n [µ

M]

Co

ntro

lsFo

ld C

hang

ep

-val

ue*

Pla

sma

But

enoy

lcar

nitin

eA

cylc

arni

tines

0.01

6 [0

.015

, 0.0

19]

0.01

8 [0

.016

, 0.0

20]

0.89

0.38

Pla

sma

His

tidin

eA

min

o ac

ids

63.7

[58.

1, 7

5.7]

72.0

[68.

7, 7

8.7]

0.88

0.25

Urin

eH

exos

eS

ugar

s67

8.1

[332

.3, 1

223.

8]10

89.8

[544

.8, 1

827.

3]0.

620.

95

Urin

eG

luta

min

eA

min

o ac

ids

94.1

[21.

3, 1

60.0

]12

9.7

[44.

8, 2

97.0

]0.

730.

95

Urin

eTy

rosi

neA

min

o ac

ids

37.2

[11.

3, 5

5.5]

61.2

[27.

2, 9

1.0]

0.61

0.86

Con

cent

ratio

ns a

re e

xpre

ssed

as

med

ian

[IQR

].*P

-val

ues

are

corr

ecte

d fo

r m

ultip

le t

estin

g an

d a

dju

sted

for

bas

elin

e U

AE

and

eG

FR.

PART 2Novel biomarker panels for predicting response to therapy and monitoring drug effect

CHAPTER 5Serum metabolites predict response to angiotensin II receptor blocker therapy in diabetes mellitus

MJ PenaA Heinzel

P RossingH-H ParvingG Dallmann

K RossingS Andersen

B MayerHJ Lambers Heerspink

Submitted

113112

Serum metabolomics predict ARB responseChapter 5

5

ABSTRACT

Objectives: Individual patients show a large variability in albuminuria response to Angiotensin Receptor Blockers (ARB). Identifying novel biomarkers that predict ARB response may help tailor therapy. We aimed to discover and validate a serum metabolite classifier that predicts albuminuria response to ARBs in patients with diabetes mellitus and micro- or macroalbuminuria.

Methods: Liquid chromatography-tandem mass spectrometry metabolomics was performed on serum samples. Data from patients with type 2 diabetes with microalbuminuria (n=49) treated with irbesartan 300mg/day were used for discovery. LASSO and ridge regression were performed to develop the classifier. Improvement in albuminuria response prediction was assessed by calculating differences in R2 between a reference model of clinical parameters and a model with clinical parameters and the classifier. The classifier was externally validated in patients with type 1 diabetes with macroalbuminuria (n=50) treated with losartan 100mg/day. Molecular process analysis was performed to link metabolites to molecular mechanisms contributing to ARB response.

Results: In discovery, median change in urinary albumin excretion (UAE) was -42% [Q25-Q75: -69 - -8]. The classifier, consisting of 21 metabolites, was significantly associated with UAE response to irbesartan (p<0.001) and improved prediction of UAE response on top of the clinical reference model (R2 increase from 0.10 to 0.70; p<0.001). In external validation, median change in UAE was -43% [Q25-Q75: -63 - -23]. The classifier improved prediction of UAE response to losartan (R2 increase from 0.20 to 0.59; p<0.001). Specifically ADMA impacting eNOS activity appears to be a relevant factor in ARB response.

Conclusions: A serum metabolite classifier was discovered and externally validated to significantly improve prediction of albuminuria response to ARBs in diabetes mellitus.

INTRODUCTION

Intervention in the renin-angiotensin-aldosterone system (RAAS) has convincingly shown to delay progression of renal disease in patients with diabetes mellitus with elevated urinary albumin excretion (UAE) in several large trials. However, individual patients show a large variability in long-term renoprotective response, which is linked to a large variability in the short-term response in albuminuria and blood pressure [1]. Consequently, a considerable proportion of patients still have significant residual albuminuria, which may contribute to progressive renal function loss [2]. The reasons behind these individual differences in response to therapy are not completely understood, but are in part related to renal tissue-specific RAAS activity, dietary salt intake, or genetic background [3-7]. Identifying novel biomarkers that predict the albuminuria lowering response to RAAS intervention may improve the current “trial-and-error” approach to choosing the optimal therapy for treatment of patients with diabetes mellitus. This would mark a step further to implementation of personalized medicine.

Biomarker discovery has advanced significantly over the past years with the use of high-throughput omics screening platforms. Omics profiling has emerged as a research area to expand beyond biomarker discovery to also unravel molecular pathways involved in disease pathophysiology. Integrating these data with clinical data may help give further insights in the underlying molecular mechanisms of drug response variability. Prospective metabolomics studies predicting disease progression in diabetes mellitus are becoming more common [8-10], but to our knowledge, there are to date no metabolomics studies for the prediction of drug response in diabetes mellitus.

Therefore, the aims of this study were to first discover and validate a serum metabolite classifier that predicts response in albuminuria to angiotensin II receptor blocker (ARB) therapy in patients with diabetes mellitus and micro- or macroalbuminuria, and secondly, to integrate the identified metabolites in a molecular process model capturing disease pathophysiology at the interface of drug mechanism of action to decipher the underlying molecular processes driving albuminuria response to ARB.

115114


5

METHODS

Patients and study designSerum samples from patients enrolled in two distinct clinical studies conducted at the Steno Diabetes Center (Gentofte, Denmark), assessing the albuminuria lowering effect of ARBs were used for the present study.

For the discovery cohort, we use data from a crossover clinical study in type 2 diabetes assessing the effect of the ARB irbesartan. This cohort has been previously described [11]. In short, 52 patients with type 2 diabetes, hypertension, microalbuminuria, and treated with antihypertensive medication were recruited for a double-masked randomized crossover trial. At inclusion, previous antihypertensive treatment was discontinued and replaced with bendroflumethiazide, 5 mg once daily, for the entire study. Following two months wash-out (baseline), patients were treated randomly with irbesartan 300, 600, and 900mg once daily. All treatment periods were of 10 weeks’ duration and consisted of an initial two-week dose titration period with irbesartan 300 mg once daily followed by eight weeks treatment with irbesartan 300, 600, and 900 mg once daily in random order. For the present study, patient data and metabolomics measurements were available for 49 patients with type 2 diabetes with persistent microalbuminuria. For this discovery cohort, we defined the outcome of interest as percent change in UAE after twelve weeks of treatment of ibestartan 300 mg/day compared to baseline UAE.

For the external validation cohort, we used data from a clinical study in type 1 diabetes assessing the effect of the ARB losartan. This cohort has been previous described [12], and included patients with type 1 diabetes, hypertension, and diabetic nephropathy. After a four-week washout, the patients received 100 mg losartan once daily and were followed prospectively with a mean follow-up period of 36 months. For the present study, patient data and metabolomics measurements were available for 50 patients with type 1 diabetes and macroalbuminuria. For the validation cohort, we defined the outcome of interest as percent change in UAE from baseline to UAE after sixteen weeks of treatment of losartan 100 mg/day. In this cohort, GFR was measured by plasma clearance of 51Cr-EDTA every six months.

We expected the treatment effect of irbesartan and losartan on albuminuria to be fully present after twelve weeks or sixteen weeks of treatment, respectively. We refer to this as the response period.

Metabolomics measurementsSerum metabolomics were measured by BIOCRATES Life Sciences (Innsbruck, Austria). Flow injection analysis and liquid chromatography-tandem mass spectrometry based targeted metabolomics measurements were performed on serum samples [13]. The full set of 185 metabolites from the following chemical classes were quantified: acylcarnitines, amino acids, biogenic amines, energy/sugar metabolism (Hexoses), lysophosphatidylcholines, phosphatidylcholines, and sphingomyelins.

Statistical analysesAnalyses were performed using SAS version 9.3. Baseline characteristics with normal distribution were reported as mean and standard deviation (SD), characteristics with skewed distribution were reported as median and inter quartile range [IQR], and categorical variables were reported as number and percentage. The natural log of UAE and log-2 of metabolite concentrations were used in all regression analysis.

Statistical modeling consisted of several steps using a previously described methodology for development of a classifier [14]. First, a least absolute shrinkage and selection operator (LASSO) regression model was fitted in the discovery cohort to the full metabolite set to select a subset of metabolites that best predicted UAE response to ARB therapy [15]. The LASSO is advantageous for small samples sizes because it places restrictions on the absolute sizes of the regression coefficients with a tuning parameter λ and controls for multicollinearity, thereby selecting the optimal subset of variables that best predicts the outcome. The tuning parameter was optimized by five-fold cross-validation, and bootstrap (N=1000) was used to evaluate selection probabilities of each metabolite. Next, the metabolites selected by the LASSO were refitted in a new model using ridge regression to generate the classifier. Cross-validation was performed to select a new tuning parameter for the ridge regression model that minimized the mean square error (MSE). Finally, the classifier was validated in an external cohort by applying the betas for each metabolite and the tuning parameter as estimated from the discovery cohort.

In both the discovery cohort and the validation cohort, the added value of the classifier was evaluated by deriving the explained variation of the model (R2) from the MSEs in order to determine whether the biomarkers significantly improved prediction on top of a model of baseline clinical parameters (age, sex, HbA1c, SBP, GFR, UAE). The area under the receiver operating characteristics (ROC) curve and integrated discrimination improvement (IDI) index were calculated to assess the discriminatory ability of the serum metabolites for a dichotomous outcome of >30% decrease in UAE during the response period. This threshold was used based on prior work [2, 16, 17].

For the validation cohort, we also determined whether the serum metabolite classifier was able to predict change in GFR after the initial response period. Patient-specific GFR change was calculated by fitting a straight line through the GFR values after the initial response period, i.e. from week 16 to the end of follow-up using a linear regression model, as was done in the original study [12]. A dichotomous outcome for GFR change ≤ or > -3.0 mL/min/1.73m2/year was created to assess the discriminatory ability of the serum metabolites for accelerated renal function decline. The threshold of -3 mL/min/1.73m2

was chosen based on prior studies [9, 18, 19] and was approximately the median GFR change in this cohort (-3.4 [Q25, Q75: -5.7, -1.4]).

Molecular model of ARB drug mechanism of actionIdentification of protein coding genes showing association with ARB mechanism of action was performed by querying NCBI PubMed and gene2pubmed. For both drugs, a PubMed

117116


5

search using the queries >>”irbesartan”[TIAB] OR irbesartan[nm]<< or >>”losartan”[TIAB] OR losartan[nm]<< was performed for identifying publications discussing irbesartan and losartan, respectively. Genes linked to identified publications were extracted from gene2pubmed. For irbesartan, 1471 publications associated to a total of 44 genes and for losartan 8166 publications linked to 101 genes in total were identified. The total set of 125 protein coding genes was used for deriving a mechanism of action molecular model as described in Heinzel et al. [20]. In short, molecular features were mapped on a human protein interaction network, and the induced subgraph was split into molecular process segments according to network topology. The resulting ARB mechanism of action molecular model holds 48 protein coding genes embedded in seven molecular process segments.

Interference of this ARB mechanism of action molecular model was performed with a previously identified diabetic kidney disease (DKD) molecular model holding 688 protein coding genes in 34 molecular process segments [21]. Interference was defined by an overlap of interacting protein coding genes being present in both the ARB mechanism of action molecular model and the DKD molecular model.

Assignment of metabolitesMetabolites selected for the classifier were assessed for being part of the DKD molecular model. Metabolite-to-enzyme assignments were identified utilizing the Human Metabolome Database (HMDB) and the Kyoto Encyclopedia of Genes and Genomes (KEGG) database. Of the 185 metabolites addressed in targeted metabolomics, 114 metabolites could be assigned to at least one enzyme. The respective number for the shortlist of 21 metabolites included in the classifier on drug response is 14. For these 14 metabolites, nine could be assigned to the molecular model representation of DKD involving 11 assigned enzymes.

RESULTS

Baseline characteristics are presented in Table 1. In the discovery cohort, patients were approximately 59 (standard deviation 10) years of age, mostly male (80%), had a known duration of type 2 diabetes of 13 (8) years, and median 24-hour UAE was 84 [Q25, Q75: 65, 200] mg/24hr. Median change in UAE was -42% [Q25, Q75: -69, -8] after twelve weeks of treatment of ibestartan 300 mg/day (Table 1).

In the validation cohort, patients were approximately 47 (9) years of age, mostly male (60%), had a known duration of type 1 diabetes for 33 (9) years, and median 24-hour UAE was 1211 [Q25, Q75: 598, 2023] mg/24hr. Median change in UAE was -43% [Q25, Q75: -23, -62] over 16 weeks of treatment with losartan 100 mg/day (Table 1). During approximately three years of follow-up, GFR change after the response period was -3.8 (3.6) mL/min/1.73m2/year.

There were no significant associations between baseline characteristics and change in UAE in either the discovery or validation cohorts (Supplemental Table 1).

Table 1. Patient characteristics

Type 2 diabetes

Discovery cohort (n=49)

Type 1 diabetes

Validation cohort (n=50)

Baseline

Age (years) 59.0 (10.0) 44.6 (8.9)

Male Sex (number (%)) 39 (80) 30 (60)

SBP (mmHg) 140.0 (15.4) 150.6 (17.7)

DBP (mmHg) 81.6 (8.8) 84.7 (10.9)

HbA1c (%) 8.3 (1.4) 8.9 (1.2)

HbA1c (mmol/mol) 67.2 (15.3) 73.8 (13.1)

Cholesterol (mmol/l) 5.3 (1.0) 5.2 (1.0)

HDL (mmol/l) 1.2 (0.3) 1.6 (0.5)

GFR (ml/min/1.73m2) 102.3 (19.2) 86.5 (23.4)

24-hour UAE (mg/24hr) 84 [65, 200] 1211 [598, 2023]

Follow-up

Change in SBP (mmHg) -6.4 (16.2) -8.7 (14.3)

Change in DBP (mmHg) -6.0 (8.8) -5.6 (8.6)

Percent change in UAE (%) -42% [-69, -8] -43% [-62, -23]

>30% decrease in UAE from baseline (number (%)) 31 (63) 34 (68)

GFR change after response period (mL/min/1.73m2/year)

not available -3.8 (3.6)

Data are reported as mean ± standard deviation (SD) or number (percent) or median [25th, 75th quartile].

Serum metabolite classifierOut of the total set of 185 metabolites, 21 metabolites were selected with LASSO as best predictors of UAE response to ARB therapy in the discovery cohort. These 21 metabolites were used for the classifier. The 21 metabolites are presented in Supplemental Table 2.

In the discovery cohort, the serum metabolites classifier was significantly associated with change in UAE in response to irbesartan 300 mg/day (p-value <0.001) and significantly improved prediction on top of clinical parameters (R2 increase from 0.10 to 0.70; p-value <0.001) (Figure 1A and 1B, Table 2). For the dichotomous outcome of >30% decrease in UAE during the response period, the control model area under the ROC curve was 0.72, and the addition of the serum metabolite classifier significantly increased the area under the ROC curve to 0.95 (p-value = 0.001) (Table 2). The IDI of the classifier was

119118


5

0.33 (p-value <0.001) (Table 2). The classifier improved prediction in SBP response (R2 increase from 0.63 to 0.68; p-value = 0.019).

In the validation cohort, the serum metabolite classifier was significantly associated with change in UAE in response to lorsartan 100 mg/day (p-value <0.001) and significantly improved prediction of change in UAE on top of a panel of clinical parameters (R2 increase from 0.20 to 0.53; p-value <0.001) (Figure 1C and 1D, Table 2). For the dichotomous outcome of >30% decrease in UAE during the response period, the area under the ROC curve for the control model was 0.74, and the addition of the serum metabolite classifier increased the area under the ROC curve to 0.89. This increase was not significant (Table 2). In The IDI of the classifier was 0.19 (p-value= 0.06) (Table 2). The classifier did not improve prediction in SBP response in external validation (R2 control = 0.24, R2 classifier 0.25; p-value = 0.54).

Figure 1. Prediction of change in UAE from baseline. A) Discovery cohort, clinical parameters model; B) Discovery cohort, clinical parameters + serum metabolite classifier model; C) Validation cohort, clinical parameters model; D) Validation cohort, clinical parameters + serum metabolite classifier model. The lines of identity are shown in light grey, and the regression lines are shown in dashed grey. In the case of perfect prediction, the regression line would be equal to the line of identity.

Tab

le 2

. Ris

k p

red

ictio

n of

cha

nge

in U

AE

in r

esp

onse

to

AR

B t

hera

py

Dis

crim

inat

ion

of

>30

% d

ecre

ase

in U

AE

R2

p-v

alue

*R

OC

95%

CI

p-v

alue

*ID

I95

% C

Ip

-val

ue*

Dis

cove

ry c

ohor

t

Clin

ical

par

amet

ers†

0.10

Ref

.0.

720.

570.

87R

ef.

Ref

.

+ S

erum

met

abol

ites

clas

sifie

r0.

70<

.001

0.95

0.89

1.00

0.00

10.

500.

360.

63<

.001

+ S

ubse

t of

7 m

etab

olite

s‡0.

50<

.001

0.90

0.81

0.99

0.01

20.

330.

190.

47<

.001

Valid

atio

n C

ohor

t

Clin

ical

par

amet

ers†

0.20

Ref

.0.

740.

590.

89R

ef.

Ref

.

+ S

erum

met

abol

ites

clas

sifie

r0.

53<

.001

0.89

0.79

0.99

0.06

30.

300.

150.

46<

.001

+ S

ubse

t of

7 m

etab

olite

s‡0.

350.

002

0.78

0.64

0.92

0.46

00.

190.

090.

470.

055

* C

omp

arin

g cl

inic

al p

aram

eter

s +

met

abol

ites

to o

nly

clin

ical

par

amet

ers

† B

asel

ine

Age

, Sex

, SB

P, H

bA

1c, G

FR, U

AE

‡ M

etab

olite

s as

sign

ed t

o b

oth

dru

g in

terf

eren

ce:d

irect

dis

ease

phe

noty

pe

and

dis

ease

pro

gres

sion

pro

cess

es (

seve

n m

etab

olite

s: A

DM

A,

citr

ullin

e,

lyso

PC

a C

16:0

, lys

oPC

A C

16:1

, PC

aa

C36

:0, P

C a

a C

42:2

, try

pto

pha

n).

121120


5

In the validation cohort, for prediction of GFR change after the response period, the combination of the 21 serum metabolites significantly improved prediction on top of clinical parameters (R2 increase from 0.15 to 0.60; p-value <0.001). For the dichotomous outcome for GFR change ≤ or > -3.0 mL/min/1.73m2/year, a significant increase was observed in the area under the ROC curve with the addition of the serum metabolites on top of clinical parameters (ROC increase from 0.71 (95% C.I. 0.57 – 0.86) to 0.88 (95% C.I. 0.79 – 0.98); p-value 0.010).

Metabolite assignmentTo study the molecular mechanisms linked to albuminuria response and ARB effect, molecular process analysis was conducted by assigning the metabolites included in the classifier to enzymes and further to molecular processes identified in a DKD molecular model. The combined irbesartan/losartan drug mechanism of action molecular model holding 48 protein coding genes is presented in Figure 2A. The model interference of the ARB drug mechanism of action molecular model with the DKD molecular model is shown in Figure 2B. In total, 20 interacting protein coding genes being reported as associated with ARB effect were also identified in the DKD molecular model. Key overlap of ARB effect with DKD pathophysiology are shown in Figure 2C, at first including the drug target angiotensin II receptor, type 2 (AGTR2) together with the bradykinin system and the NFκB/PPARγ axis.

Nine out of the 21 metabolites included in the classifier could be assigned to 11 enzymes also involved in the DKD molecular model, including enzymatic turnover as well as metabolite transport. Metabolite-enzyme links are provided in Supplemental Table 3.

Interference of ARB drug mechanism of action molecular model and the DKD molecular model identifies the enzyme nitric oxide synthase 3 (NOS3). The metabolites included in the classifier were assigned to NOS3 are asymmetric dimethylarginines (ADMA) and citrulline, with ADMA being the most frequently selected metabolite in the LASSO (Supplemental Table 2). Furthermore, overlap of enzymes at the molecular process level between the molecular models was observed involving amino acids (glutamine, asparagines and tryptophan), lysophosphatidylcholines (lysoPC a C16:0 and lysoPC a C16:1), and phosphatidylcholines (PC aa C36:0 and PC aa C42:2). This subset of seven metabolites assigned to direct drug-to-DKD interference (ADMA, citrulline) or to molecular processes with interference on the molecular model level (lysoPC a C16:0, lysoPC a C16:1, PC aa C36:0, PC aa C42:2, tryptophan) significantly improved the explained variation in albuminuria response (R2) in the discovery study (R2=0.50; p<0.001 versus clinical model) and validation study (R2=0.39; p=0.001 versus clinical model).

Figure 2. A) Combined irbesartan/losartan drug mechanism of action molecular model holding 48 protein coding genes (nodes) organized in seven molecular process segments (boxes). Protein interactions are indicated as edges, further interactions of proteins across process segments are omitted. B) Interference of ARB mechanism of action model on the DKD molecular model. Matching network segments are shown as red nodes. Nodes colored in blue identify enzymes associated with metabolites included in the classifier, nodes in pink indicate metabolite transport. C) Gene symbols for selected nodes matching in drug and DKD molecular models, and associated metabolites in case of enzymes according to Supplemental Table 3.

123122


5

DISCUSSION

This study discovered and externally validated a serum metabolite classifier that significantly improves prediction of albuminuria response to ARBs in patients with diabetes mellitus with micro- or macroalbuminuria on top of traditional clinical risk factors. Metabolites included in the classifier could be assigned to general molecular mechanisms of oxidative stress, inflammation, and fibrosis pathways. Specifically, NOS3 activity appears to be a relevant factor in predicting the albuminuria lowering response to ARBs. Our findings suggest the use of serum metabolites as a tool to tailor albuminuria lowering ARB treatment and illustrate the use of metabolomics to unravel underlying molecular mechanisms of ARB response.

Metabolomics, the measurement of exogenous or endogenous small molecules in a sample, is an emerging research area to identify novel biomarkers. The metabolome integrates the biological information of the genome, transcriptome, proteome, and overall enzymatic reactions of an individual, therefore enabling the detection of short and long-term physiological or pathological changes occurring in diseases [22]. Metabolomics can be used to unravel molecular pathways of biological processes in order to better understand disease progression [9, 23, 24], but to the best of our knowledge, this is the first study integrating metabolomics to study response to drug therapy. These types of studies are necessary to characterize the molecular mechanisms of drug effects and drug response variability.

Pharmacological blockade of angiotensin II activity by ARBs is currently the most widely used therapeutic option (next to angiotensin converting enzyme inhibitors) for treatment of hypertension and albuminuria in diabetes mellitus. Yet, approximately 25% of patients with diabetes mellitus do not respond in terms of albuminuria lowering to ARBs [1]. The serum metabolite classifier was able to predict the short-term albuminuria response to the ARBs. This observation is important from a therapeutic point of view as a poor anti-albuminuric response predicts poor long-term renal prognosis [25]. Importance of early albuminuria reduction is not only evident for renoprotection but also for cardiovascular protection [2]. The metabolite panel thus provides an indication which patients will be protected for long-term renal and cardiovascular outcomes. In addition to predicting UAE response, the classifier predicted systolic blood pressure response in the discovery cohort but not in the validation cohort. The failure to predict the systolic blood pressure response may be explained by prior observations that not all patients respond concordantly in terms of albuminuria-lowering or SBP-lowering to ARBs [16, 17]. Identification and validation of novel biomarkers that can predict response to therapy may improve on the current “trial-and-error” approach in prescribing medication, may ultimately help reduce the large individual variability in response to therapy, and could be a step forward to implement personalized medicine.

We performed a molecular process analysis aiming to further unravel the molecular mechanisms linked to albuminuria response to ARBs in diabetes mellitus. The network approach assumes that individual drug response variability can be in part attributed to individual variability in underlying molecular mechanisms involved in the progression

of disease in the light of personalized molecular pathophysiology. Indeed, coupled pathophysiological processes such as oxidative stress, inflammation, and fibrosis appear to drive disease progression, but the individual contribution of each process varies per individual. Drugs on the other hand, address specific targets and thereby interfere in specific disease associated processes. At this level, metabolites, or biomarkers more generally, may help to gain insight to which specific pathophysiological processes are driving disease progression and are targeted by a specific drug’s mode of action [26].

A key finding from the molecular process analysis is the relevance of nitric oxide (NO) in ARB response, as illustrated by the direct interference of ARB molecular mechanism of action and DKD molecular model by the enzyme NOS3, and further reflected by the inclusion of ADMA and citrulline in the serum metabolite classifier. ADMA, an endogeneous NO synthase inhibitor, is considered relevant in endothelial dysfunction contributing to extracellular matrix remodeling and being involved in inflammation [27]. Increased ADMA levels have been shown to contribute to increased risk of progressive diabetic kidney disease and predict fatal and nonfatal cardiovascular events in patients with type 1 diabetic nephropathy [28]. The positive beta values for ADMA in the regression models suggest that higher concentrations of the metabolite are associated with less albuminuria reduction. We therefore speculate that higher concentrations of ADMA lead to increased blockade of NOS3, resulting in a decrease in NO availability and diminished albuminuria response. Furthermore, citrulline, catalyzed by NOS3, has been shown to be inversely correlated to inflammatory parameters such as C-reactive protein [29]. Supplementation with citrulline has been shown to increase arginine/ADMA ratio, in turn decreasing blood pressure and improving vascular function [30]. Our study points to higher concentrations of citrulline associated with greater reductions in albuminuria. ADMA and citrulline assignments in the drug molecular model and DKD molecular model further suggest that albuminuria response to ARBs is reflected on the background of progressive disease as well as influenced by eNOS activity.

Some metabolites in the classifier could not be directly linked to the ARB mechanism of action molecular model, but were linked to processes driving progression of kidney disease. This indicates that in order to assess drug response, the interplay between disease progression molecular characteristics and specific drug molecular effects should be considered. This notion was further emphasized when testing a different subgroup of metabolites. The subset of seven metabolites being present in the DKD molecular model and being linked with oxidative stress response, inflammation and fibrosis (TGFB and downstream ECM remodeling) next to NOS3 activity significantly improved prediction of albuminuria response. Furthermore, interferences at the process units at the direct drug target together with the bradykinin system and on NFκB/PPARγ indicate involvement of inflammatory processes and lipid metabolism contributing to ARB response. These observations suggest that for assessing drug response, both disease progression status and specific drug molecular effects need to be taken into account. Metabolites of the renin-angiotensin-system were not included in the classifier, suggesting that markers of RAAS activity do not predict the response to RAAS inhibition, which is in line with prior studies [31, 32].

125124


5

Within the validation cohort, we were able to assess GFR change after the initial response period up to the end of follow-up. We observed an improvement in GFR over follow-up in patients who had a >30% decrease in UAE from baseline, compared to a patients who did not see such a benefit in albuminuria response (Supplemental Table 4). This is in line with previous literature showing that greater reduction in albuminuria is associated with a lesser decline in eGFR during long-term follow-up [17]. Of interest, the metabolites were able to improve prediction of GFR changes, indicating that the serum metabolite classifier enables the identification of a group of patients who do not have a good albuminuria response to ARB therapy and have a fast deterioration of renal function over time. Further studies in large, diverse patient cohorts is necessary for validating these findings.

Limitations of this study include lack of comprehensiveness of drug mechanism of action characterization. The combined losartan/irbesartan mechanism of action molecular model included a total of 48 protein coding genes retrieved from literature mining. Adding detailed omics profiling specifically on drug effect on kidney cells and tissue via in vitro or in vivo models would provide an improved representation of ARB molecular effect. In addition, the relative small sample sizes of the included trials may have decreased the precision of our effect estimates. Unfortunately, a type 2 diabetes validation cohort was not available. We therefore validated the metabolomics classifier in a type 1 diabetes cohort and demonstrated that the classifier was able to predict response to ARB therapy in this population as well. While the pathophysiology of the diseases are clearly different, the response to treatment and predictors of response may be similar. We showed that the classifier was able to predict the albuminuria response to two different ARBs. Whether the classifier predicts the response to other interventions in the RAAS has to be analyzed in future studies.

In conclusion, we discovered and externally validated a classifier of 21 serum metabolites that significantly improve prediction of albuminuria response to ARBs in diabetes mellitus. Metabolites included in the classifier were assigned to stress/inflammation pathways and downstream consequences of fibrosis and extra cellular matrix remodeling. Specifically, NOS3 activity appears to be a specific factor relevant in ARB response. These results indicate that for assessing drug response, both disease progression status and specific drug molecular effects need to be taken into account. Moreover, the results of this metabolomics study support the growing evidence of using omics tools as a strategy to improve molecular characterization of drug effect and disease pathophysiology. The complementary use of omics platforms, integrated into molecular process models and from there determining biomarker panels, makes implementation of personalized medicine increasingly realistic in clinical practice.

FUNDINGThe work leading to this paper received funding from the European Community’s Seventh Framework Programme under grant agreement no. HEALTH–F2–2009–241544 (SysKID consortium) and EU-MASCARA (project no. 278249). Funding was also received from the Novo Nordisk Foundation Grant number NNF14SA0003. HJ Lambers Heerspink is supported by a VIDI grant from the Netherlands Organisation for Scientific Research.

REFERENCES

1. Schievink B, de Zeeuw D, Parving HH, Rossing P, Lambers Heerspink HJ. The renal

protective effect of angiotensin receptor blockers depends on intra-individual response

variation in multiple risk markers. Br J Clin Pharmacol 2015;[Epub ahead of print]



diabetic nephropathy: Lessons from RENAAL. Kidney Int 2004;65(6):2309-2320



diabetic nephropathy. Kidney Int Suppl 2000;75:S32-7

4. Crowley SD, Gurley SB, Oliverio MI, Pazmino AK, Griffiths R, Flannery PJ, Spurney RF,

Kim HS, Smithies O, Le TH, Coffman TM. Distinct roles for the kidney and systemic

tissues in blood pressure regulation by the renin-angiotensin system. J Clin Invest

2005;115(4):1092-1099

5. Vogt L, Waanders F, Boomsma F, de Zeeuw D, Navis G. Effects of dietary sodium

and hydrochlorothiazide on the antiproteinuric efficacy of losartan. J Am Soc Nephrol

2008;19(5):999-1007

6. Yasar U, Forslund-Bergengren C, Tybring G, Dorado P, Llerena A, Sjoqvist F, Eliasson

E, Dahl ML. Pharmacokinetics of losartan and its metabolite E-3174 in relation to the

CYP2C9 genotype. Clin Pharmacol Ther 2002;71(1):89-98

7. Parving HH, de Zeeuw D, Cooper ME, Remuzzi G, Liu N, Lunceford J, Shahinfar S, Wong

PH, Lyle PA, Rossing P, Brenner BM. ACE gene polymorphism and losartan treatment in

type 2 diabetic patients with nephropathy. J Am Soc Nephrol 2008;19(4):771-779

8. van der Kloet FM, Tempels FWA, Ismail N, van der Heijden R, Kasper PT, Rojas-Cherto M,

van Doorn R, Spijksma G, Koek M, van der Greef J, Makinen VP, Forsblom C, Holthofer

H, Groop PH, Reijmers TH, Hankemeier T. Discovery of early-stage biomarkers for

diabetic kidney disease using ms-based metabolomics (FinnDiane study). Metabolomics

2012;8(1):109-119




mellitus. Diabet Med 2014;31(9):1138-1147

10. Niewczas MA, Sirich TL, Mathew AV, Skupien J, Mohney RP, Warram JH, Smiles A, Huang

X, Walker W, Byun J, Karoly ED, Kensicki EM, Berry GT, Bonventre JV, Pennathur S, Meyer

TW, Krolewski AS. Uremic solutes and risk of end-stage renal disease in type 2 diabetes:

Metabolomic study. Kidney Int 2014;85(5):1214-1224

127126


5

11. Rossing K, Schjoedt KJ, Jensen BR, Boomsma F, Parving HH. Enhanced renoprotective

effects of ultrahigh doses of irbesartan in patients with type 2 diabetes and microalbuminuria.

Kidney Int 2005;68(3):1190-1198

12. Andersen S, Tarnow L, Cambien F, Rossing P, Juhl TR, Deinum J, Parving HH. Long-term

renoprotective effects of losartan in diabetic nephropathy: Interaction with ACE insertion/

deletion genotype? Diabetes Care 2003;26(5):1501-1506

13. Ramsay SL, Stoegg WM, Weinberger KM, al e. Apparatus and method for analyzing a

metabolite profile. [EP 1875401 A211] 2007; 1–67. Ref type: Patent

14. Pena MJ, Jankowski J, Heinze G, Kohl M, Heinzel A, Bakker SJ, Gansevoort RT, Rossing

P, de Zeeuw D, Heerspink HJ, Jankowski V. Plasma proteomics classifiers improve risk

prediction for renal disease in patients with hypertension or type 2 diabetes. J Hypertens

2015;[Epub ahead of print]

15. Tibshirani R. Regression shrinkage and selection via the lasso. Journal of the Royal

Statistical Society.Series B (Methodological) 1996;58(1):267-288

16. Eijkelkamp WB, Zhang Z, Remuzzi G, Parving HH, Cooper ME, Keane WF, Shahinfar S,

Gleim GW, Weir MR, Brenner BM, de Zeeuw D. Albuminuria is a target for renoprotective

therapy independent from blood pressure in patients with type 2 diabetic nephropathy:

Post hoc analysis from the reduction of endpoints in NIDDM with the angiotensin II

antagonist losartan (RENAAL) trial. J Am Soc Nephrol 2007;18(5):1540-1546

17. Hellemons ME, Persson F, Bakker SJ, Rossing P, Parving HH, De Zeeuw D, Lambers

Heerspink HJ. Initial angiotensin receptor blockade-induced decrease in albuminuria is

associated with long-term renal outcome in type 2 diabetic patients with microalbuminuria:

A post hoc analysis of the IRMA-2 trial. Diabetes Care 2011;34(9):2078-2083

18. Eriksen BO, Ingebretsen OC. The progression of chronic kidney disease: A 10-year

population-based study of the effects of gender and age. Kidney Int 2006;69(2):375-382

19. Shlipak MG, Katz R, Kestenbaum B, Fried LF, Newman AB, Siscovick DS, Stevens L,

Sarnak MJ. Rate of kidney function decline in older adults: A comparison using creatinine

and cystatin C. Am J Nephrol 2009;30(3):171-178

20. Heinzel A, Perco P, Mayer G, Oberbauer R, Lukas A, Mayer B. From molecular signatures

to predictive biomarkers: Modeling disease pathophysiology and drug mechanism of

action. Front Cell Dev Biol 2014;2:37

21. Heinzel A, Muhlberger I, Stelzer G, Lancet D, Oberbauer R, Martin M, Perco P. Molecular

disease presentation in diabetic nephropathy. Nephrol Dial Transplant 2015;30 Suppl

4:iv17-iv25

22. Gerszten RE, Wang TJ. The search for new cardiovascular biomarkers. Nature

2008;451(7181):949-952





J Am Soc Nephrol. 2013;24(11):1901-12

24. Nkuipou-Kenfack E, Duranton F, Gayrard N, Argilés À, Lundin U, Weinberger KM, Dakna

M, Delles C, Mullen W, Husi H, Klein J, Koeck T, Zürbig P, Mischak H. Assessment of

metabolomic and proteomic biomarkers in detection and prognosis of progression of

renal function in chronic kidney disease. PLoS One. 2014;9(5):e96955

25. Lambers Heerspink HJ, Gansevoort RT. Albuminuria is an appropriate therapeutic target

in patients with CKD: The pro view. Clin J Am Soc Nephrol 2015;10(6):1079-1088

26. Lambers Heerspink HJ, Oberbauer R, Perco P, Heinzel A, Heinze G, Mayer G, Mayer

B. Drugs meeting the molecular basis of diabetic kidney disease: Bridging from

molecular mechanism to personalized medicine. Nephrol Dial Transplant 2015;30 Suppl

4:iv105-iv112

27. Kielstein JT, Zoccali C. Asymmetric dimethylarginine: A novel marker of risk and a

potential target for therapy in chronic kidney disease. Curr Opin Nephrol Hypertens

2008;17(6):609-615

28. Lajer M, Tarnow L, Jorsal A, Teerlink T, Parving HH, Rossing P. Plasma concentration of

asymmetric dimethylarginine (ADMA) predicts cardiovascular morbidity and mortality in

type 1 diabetic patients with diabetic nephropathy. Diabetes Care 2008;31(4):747-752

29. Suliman ME, Qureshi AR, Stenvinkel P, Pecoits-Filho R, Barany P, Heimburger O,

Anderstam B, Rodriguez Ayala E, Divino Filho JC, Alvestrand A, Lindholm B. Inflammation

contributes to low plasma amino acid concentrations in patients with chronic kidney

disease. Am J Clin Nutr 2005;82(2):342-349

30. Schwedhelm E, Maas R, Freese R, Jung D, Lukacs Z, Jambrecina A, Spickler W, Schulze

F, Boger RH. Pharmacokinetic and pharmacodynamic properties of oral L-citrulline and

L-arginine: Impact on nitric oxide metabolism. Br J Clin Pharmacol 2008;65(1):51-59

31. Parthasarathy HK, Alhashmi K, McMahon AD, Struthers AD, McInnes GT, Ford I, Connell

JM, MacDonald TM. Does the ratio of serum aldosterone to plasma renin activity

predict the efficacy of diuretics in hypertension? results of RENALDO. J Hypertens

2010;28(1):170-177

32. Mahmud A, Mahgoub M, Hall M, Feely J. Does aldosterone-to-renin ratio predict the

antihypertensive effect of the aldosterone antagonist spironolactone? Am J Hypertens

2005;18(12 Pt 1):1631-1635

129128


5

Supplemental Table 1. Associations with baseline characteristics and change in UAE

Type 2 diabetes

Discovery cohort (n=49)

Type 1 diabetes

Validation cohort (n=50)

β S.E. p-value β S.E. p-value

Age at baseline 0.58 0.53 0.281 0.8 0.9 0.386

Male sex 7.7 13.2 0.560 -24.9 15.2 0.109

Baseline SBP -0.02 0.4 0.961 0.5 0.4 0.257

Baseline DBP -0.6 0.6 0.362 0.2 0.7 0.788

Baseline HbA1c 1.6 3.9 0.683 7.1 6.3 0.264

Baseline Cholesterol -0.8 5.4 0.877 4.4 7.6 0.567

Baseline HDL 19.4 15.8 0.225 14.0 14.3 0.332

Baseline GFR -0.2 0.3 0.554 0.02 0.3 0.961

Baseline 24-hour logUAE -7.9 5.7 0.177 -12.6 9.1 0.174

Supplemental Table 2. LASSO-selection of best predictors. Results from LASSO regression of 21 metabolites selected for the serum metabolites classifier, 5-fold cross-validation, and bootstrap resampling (N=1000) in the discovery cohort of patients with type 2 diabetes with microalbuminuria (n=49).

MetaboliteMean

Estimate

Standard

Deviation95% C.I.

Selection

Percentage*

Asymmetric dimethylarginine (ADMA) 4.4 6.5 0, 22 52.4

Asparagine (Asp) -3.0 7.4 -26, 0 26.2

Carnitine (C0) -2.4 6.8 -24, 0 20.0

Acylcarnitine (C12-DC) 11.3 24.6 0, 88 31.4

Linoleoylcarnitine (C18:2) -3.0 7.5 -27, 0 22.8

Acylcarnitine (C5:1-DC) 2.9 7.4 0, 25 23.4

Glutarylcarnitine (C5-DC / C6-OH) 3.3 7.5 0, 28 27.2

Acylcarnitine (C6:1) 3.7 9.9 0, 34 23.5

Acylcarnitine (C7-DC) 2.39 6.99 0, 25 20.6

Octanoylcarnitine (C8) -3.0 8.8 -31, 0 20.6

Citrulline (Cit) -2.0 4.9 -17, 0 27.5

Glutamine (Gln) 3.7 7.8 0, 28 31.8

Histidine (His) 7.5 14.0 0, 48 36.2

Lysophosphatidylcholines (lysoPC a C16:0) -10.4 18.4 -62, 0 36.9

Lysophosphatidylcholines (lysoPC a C16:1) -2.9 6.4 -22, 0 26.2

Phosphatidylcholines (PC aa C36:0) 4.5 7.0 0, 23 44.1

Phosphatidylcholines (PC aa C42:2) 9.7 13.8 0, 46 51.6

Symmetric dimethylarginine (SDMA) 0.3 1.0 0, 3 22.5

Spermine -3.9 7.0 -24, 0 36.2

Tryptophan (Trp) 6.2 12.2 0, 41 35.1

Valine (Val) 3.5 8. 7 0, 31 24.4

* The relative frequency of the marker being included in the model across 1000 bootstrap resamples.

Supplemental Table 3. Metabolite-enzyme links as identified in the DKD molecular model and description of assigned reactions and transport function.

Metabolite-Enzyme LinkDescription

Metabolite Enzyme

ADMA NOS2, NOS3

Produces nitric oxide (NOS) which is a messenger molecule with diverse functions throughout the body. In macrophages, NO mediates tumoricidal and bactericidal actions. Also has nitrosylase activity and mediates cysteine S-nitrosylation of cytoplasmic target proteins such COX2.

ADMA is an endogeous inhibitor of eNOS (NOS3) function.

Aspargine

Glutamine

ASNS Asparagine Synthase: Adenosine triphosphate + L-Aspartic acid + L-Glutamine + Water → Adenosine monophosphate + Pyrophosphate + L-Asparagine + L-Glutamic acid

Asparagine SLC1A1 Solute carries family member 1, L-Asp cotransporter

Glutamine TGM2 Catalyzes the cross-linking of proteins and the conjugation of polyamines to proteins.

Citrulline NOS1, NOS2, NOS3

L-Arginine + NADPH + Oxygen → Citrulline + Nitric oxide + NADP + Water NADPH + N-(o)-Hydroxyarginine + Oxygen + Hydrogen Ion → NADP + Nitric oxide + Citrulline + Water L-Arginine + Oxygen + NADPH + Hydrogen Ion → Nitric oxide + Citrulline + NADP + Water

lysoPC a C16:0

lysoPC a C16:1

PC aa C36:0

PC aa C42:2

PLA2G1B PA2 catalyzes the calcium-dependent hydrolysis of the 2-acyl groups in 3-sn-phosphoglycerides.

PC aa C36:0 ATPSA1 ATP10A PLSCR1

Transport of aminophospholipids

Tryptophan IDO1 Catalyzes the cleavage of the pyrrol ring of tryptophan and incorporates both atoms of a molecule of oxygen.

Supplemental Table 4. GFR change after response period in the validation cohort (n=50)

≤30% decrease in UAE >30% decrease in UAE

GFR change after response period (mL/min/1.73m2/year)

-5.2 (3.2) -3.2 (3.7)

Data are reported as mean ± standard deviation (SD).

CHAPTER 6The beneficial impact of atrasentan on a urinary metabolite panel previously associated with renal function decline

MJ PenaD de Zeeuw

D AndressJJ Brennan

R Correa-RotterB Coll

DE KohanH Makino

V PerkovicG Remuzzi

SW TobeR Toto

H-H ParvingT Corringham

S SharmaK Sharma

HJ Lambers Heerspink

Submitted

133132

Effect of atrasentan therapy on metabolitesChapter 6

6

ABSTRACT

Objectives: A specific panel of 13 urinary organic acid metabolites reflecting mitochondrial function was previously identified to be reduced in patients with diabetic kidney disease. We aimed to assess correlations between these 13 urinary metabolites and eGFR, and the effect of atrasentan therapy on these metabolites in the previously conducted RADAR study.

Methods: RADAR was a placebo controlled double blind trial in patients with type 2 diabetes and nephropathy (n=150) randomly allocated to 12 weeks treatment with placebo or atrasentan 0.75-1.25 mg/d, as adjunct to renin-angiotensin-system inhibition. Urinary metabolomics was performed with gas-chromatography mass spectrometry. A combined average index of the metabolites was developed, producing a single-valued index called the Metabolomics Signature of Diabetic Kidney Disease (MSDKD) panel.

Results: Concentrations of four out of 13 metabolites were below detectable levels. Of the remaining nine metabolites, six were reduced in patients with eGFR <60 compared to ≥60 ml/min/1.73m2, and eight showed significant correlation with baseline eGFR (all p≤0.015, with Bonferroni correction). In patients with eGFR <60 ml/min/1.73m2, the MSDKD panel changed by -0.31 (95%CI -0.60 to -0.02; p=0.035), -0.08 ( -12 to 0.29; p=0.43) and 0.01 (-0.21 to 0.19; p=0.913) in placebo, atrasentan 0.75 mg/d and 1.25 mg/d, respectively after 12 weeks. The MSDKD panel difference compared to placebo was 0.13 (-0.17 to 0.43; p=0.40) and 0.35 (0.05 to 0.65; p=0.024) for atrasentan 0.75 mg/d and 1.25 mg/d, respectively. Change in eGFR correlated with changes in all nine metabolites and the MSDKD panel (R2= 0.42; p<0.001) in the atrasentan groups.

Conclusions: Urinary metabolites reflecting mitochondrial function correlated with eGFR in patients with type 2 diabetes and nephropathy. Treatment with atrasentan 1.25 mg/d for 12 weeks stabilized the levels of the metabolites while they declined with placebo treatment.

INTRODUCTION

Metabolomics, the measurement of small molecules in a biological sample, is an emerging science that holds promise for identifying novel biomarkers of kidney disease. The metabolome integrates the biological information of the genome, transcriptome, proteome, and overall enzymatic reactions of an individual, therefore enabling the detection of short and long-term physiological or pathological changes occurring in diseases. The systematic analyses of metabolites in urine of patients with kidney disease may help characterize the biochemical pathways of kidney disease progression. A recent urinary metabolomics study identified a novel metabolomics signature of diabetic kidney disease [1]. This signature, consisting of 13 metabolites, was found to be significantly and consistently reduced in patients with diabetic kidney disease compared to healthy controls. These metabolites are regulated by mitochondrial function, suggesting a reduced mitochondrial content and function in diabetic kidney disease [1].

Significant efforts are being invested in improving risk assessment of renal function decline using novel biomarker panels to improve on the available clinical parameters [2]. Of equal importance is to determine whether that risk of renal function decline is modifiable with drug therapy. The endothelin system is a promising target for intervention in diabetic kidney disease [3]. Blockade of the endothelin A receptor with selective inhibitors such as atrasentan have shown clinical meaningful reductions in albuminuria [4, 5], which predicts long-term renoprotection. Urinary metabolomics may aid in discovering novel biomarkers for monitoring of a potential long-term renoprotective drug efficacy.

The first aim of this study was to assess the correlation between the previously discovered metabolomics signature of diabetic kidney disease and estimated glomerular filtration rate (eGFR) in patients with type 2 diabetes and nephropathy participating in the Reducing Residual Albuminuria in Subjects With Diabetes and Nephropathy With AtRasentan (RADAR) trial, and the second aim was to evaluate the effect of atrasentan on these urinary metabolites.

135134


6

METHODS

Patients and study designFor the present study, data and urine samples from 150 participants from the RADAR study (NCT01356849) were used. The detailed design, rationale, and study outcome for this study has been previously published [4]. In brief, the RADAR study assessed the effect of the endothelin A receptor antagonist atrasentan on albuminuria. Patients at least 18 years of age with type 2 diabetes, and nephropathy were eligible for participation in the clinical trial. Estimated glomerular filtration rate (eGFR) was calculated with the CKD-EPI formula and ranged between 30 and 75 ml/min/1.73m2 [6]. All participants had a urinary albumin to creatinine ratio (UACR) between 300 and 3500 mg/g, received the maximum tolerated labeled daily dose of a renin-angiotensin-aldosterone-system inhibitor, and 90% of patients were receiving a diuretic. Main exclusion criteria were a history of moderate or severe edema, pulmonary edema, pulmonary hypertension, or heart failure. Patients were randomly allocated to 12 weeks treatment with atrasentan 0.75 mg/d, 1.25 mg/d, or matched placebo. The study protocol was approved by an Independent Ethics Committee and local and central review boards. All participants signed written informed consent before the start of any study-specific procedure.

Measurements and outcomesStudy visits were scheduled at week 1, 2, and every 2 weeks thereafter. Physical examination including measurement of blood pressure, body weight, and assessment of edema were performed at every visit. Three consecutive first morning void urine collections were collected at each visit, with the exception of week 1, for determination of urinary albumin and creatinine concentrations. Urine samples were collected and stored at -80°C until further processing. Albuminuria was determined by the geometric mean of three first morning voids.

Metabolomics measurementsUrine samples were analyzed with a quantitative gas chromatography-mass spectrometry platform to measure the following organic acid metabolites: 3-OH-isobutyrate, citric acid, homovanillic acid, 2-ET-3-OH-propionate, glycolic acid, 3-OH-propionate, aconitic acid, uracil, 3-OH-isovalerate, 3-methyl adipic acid, tiglylglycine, 3-methyl-crotonyl glycine, and 2-methyl acetoacetate. Urine samples were pre-processed as previously described [1] and subsequently applied by injection onto a 30 m 3 0.32 mm column (Agilent DB-5) in a gas chromatogram (Agilent 5890) and eluted with a 4°C/min gradient of 70–300°C; analytes were detected by electron impact mass spectrometry (Agilent 5973 mass selective detector). Each compound was identified by spectrum and confirmed ratio of a qualifying ion and quantifying ion. The quantifying ion’s integrated current was used to estimate concentration based on standard curves for targeted metabolites or based on a ratio to 4-nitrophenol or the oximated derivative of 2-ketocaproic acid. The results were expressed as µmol organic acid per mmol urinary creatinine.

A combined average index of the metabolites was developed to produce a single-valued index, henceforth referred to as the Metabolomics Signature of Diabetic Kidney Disease (MSDKD) panel. To compute the MSDKD panel per sample, the concentration of each of the metabolites was first normalized by subtracting its mean over the whole data set (i.e., for all time points and treatment groups), then dividing that value by its standard deviation (also over the whole data set), and then taking the average of the normalized metabolite concentrations. This methodology has been used previously to combine multiple biomarkers into an index panel [7, 8].

Statistical analysesAnalyses were performed using SAS version 9.3. Baseline characteristics with normal distribution were reported as mean and standard deviation (SD), characteristics with skewed distribution were reported as median and inter quartile range [IQR], and categorical variables were reported as number and percentage. UACR and metabolite concentrations were log transformed for analyses to account for their skewed distribution. P-values were two-tailed and values <0.05 with Bonferroni correction for multiple testing were considered statistically significant.

The correlations between the metabolites and eGFR at baseline were computed with Pearson’s correlation. In addition, metabolite concentrations at baseline were compared by independent samples t-test between groups stratified by baseline eGFR <60 ml/min/1.73m2 (n=121) or ≥ 60 ml/min/1.73m2 (n=29). In order to assess the effect of atrasentan on the metabolites from baseline to week 6 and week 12, the difference between the log transformed metabolite concentrations was calculated in the atrasentan treatment group and the placebo group and assessed with the Welch t-test. Simple binomial probabilities of the differences being positive were tested under the null hypothesis of no treatment effect, assuming no correlation between the metabolites. A mixed-effects model repeated-measures analysis for change in the metabolites from baseline to week 6 and week 12 was performed to ascertain whether there were differences the MSDKD panel between the placebo group and the atrasentan treatment group over follow-up. The model included treatment, visit, treatment-by-visit interaction as factors. Visits were included as repeated measure units from the same patients. To allow generality for the covariance structure for the repeated measures, the variance-covariance matrix was assumed to be unstructured, i.e., purely data dependent. The correlations between changes in metabolites with changes in eGFR were tested by linear regression in a combined treatment group.

137136


6

RESULTS

Correlation between metabolites and eGFRAt baseline, nine of the 13 urinary metabolites from the metabolomics signature of diabetic kidney disease were detectable in urine. Levels of the other four metabolites were below detection in the majority of patients and therefore excluded from further analysis. Lower concentrations of the nine measurable metabolites were significantly associated with lower baseline eGFR for eight metabolites (all p≤0.015 after Bonferroni correction for multiple testing; Table 1). Concentrations of six of the metabolites were significantly higher in patients with a baseline eGFR ≥60 ml/min/1.73m2 compared to patients with a baseline eGFR <60 ml/min/1.73m2 (p<0.05 after Bonferroni correction for multiple testing; Table 1). A MSDKD index was calculated based on the average values of the 9 and found to be significantly associated with the eGFR at baseline (R2=0.43, p<0.0001) and markedly different in patients with eGFR <60 vs. >60 ml/min/1.73m2 (Table 1).

Table 1. Pearson’s correlations between metabolites and eGFR at baseline in the overall RADAR

population (n=150).

Correlation with baseline eGFR

Metabolite concentrations

(µmol organic acid/mmol urinary creatinine)

Individual MSDKD Metabolites

Pearson’s Coefficient

P-valueeGFR <60 ml/

min/1.73m2

(n=121)

eGFR ≥60 ml/min/1.73m2

(n=29)P-value

Aconitic acid 0.38 <0.001 13.9 [10.9, 18.5] 20.0 [15.7, 24.8] 0.008

Citric acid 0.33 <0.001 141.6 [77.7, 217.2] 216.1 [102.5, .46.5] 0.135

Glycolic acid 0.35 <0.001 14.4 [11.9, 16.9] 18.2 [16.3, 22.1] 0.004

Homovanillic acid 0.06 0.440 0.9 [0.7, 1.2] 1.0 [0.7, 1.2] 0.999

2-ET-3-OH-Propionate

0.27 <0.001 0.3 [0, 0.7] 0.7 [0.2, 1.0] 0.027

3-OH-Isobutyrate 0.26 0.001 17.7 [13.1, 22.6] 24.7 [18.5, 34.5] 0.015

3-OH-Isovalerate 0.40 <0.001 5.2 [4.3, 6.9] 7.8 [6.6, 10.1] <0.001

3-OH-Propionate 0.40 <0.001 0.8 [0.6, 1.1] 1.6 [1.2, 2.0] <0.001

Uracil 0.20 0.015 2.1 [1.2, 2.8] 3.1 [2.6, 4.1] 0.163

MSDKD panel 0.43 <0.001 -0.07 [-0.5, 0.3] 0.6 [0.04, 0.9] <0.001

Concentrations are reported as median [25th quartile, 75th quartile]. The 4 non-detectable metabolites from the metabolomics signature of mitochondrial dysfunction: 3-Methyl adipic acid, Tiglylglycine, 3-methyl-crotonyl glycine, 2-Methyl acetoacetate.

Effect of atrasentan on the metabolites in patients with eGFR <60 ml/min/1.73m2

Because this panel of metabolites was significantly reduced in patients with a baseline eGFR <60 ml/min/1.73m2, we assessed whether atrasentan treatment would alter the metabolite concentrations in this subgroup. The baseline demographics, clinical biochemical characteristics as well as the baseline metabolite levels were well balanced between the placebo and treatment groups (Supplemental Table 1). The concentrations of the metabolites decreased during the 12 weeks follow up period in the placebo group. In contrast, the same metabolites remained stable in the atrasentan groups over the 12 weeks follow-up period. Accordingly, relative to placebo, eight of the nine metabolites increased in the atrasentan 0.75 mg/d group and all nine individual metabolites increased in the atrasentan 1.25 mg/d group at week 12 (p=0.0019 for increase in all nine metabolites versus placebo assuming no correlation between the metabolites; Figure 1).

In patients with eGFR <60 ml/min/1.73m2 at baseline, the MSDKD panel changed by -0.31 (95%CI -0.60 to -0.02; p=0.035) at week 12 in the placebo group (Figure 2). In the atrasentan 0.75 mg/d group, the MSDKD panel changed by -0.08 (95% CI -0.12 to 0.29; p=0.43) and it changed by 0.01 (95%CI -0.21 to 0.19; p=0.913) in the atrasentan 1.25 mg/d group. This resulted in a dose dependent difference in the MSDKD panel compared to placebo of 0.13 (95% CI -0.17 to 0.43; p=0.40) and 0.35 (95% CI 0.05 to 0.65; p=0.024) for the atrasentan 0.75 mg/d and 1.25 mg/d groups, respectively at week 12 (Figure 2). The effect of atrasentan on the metabolites was not different when the overall population was analyzed (Supplemental Figure 1).

Correlations between change in eGFR and changes in metabolitesIn the placebo group, there were no significant associations observed between change in eGFR and change in the metabolites. In contrast, changes in eGFR significantly and positively correlated with changes in all nine metabolites (Table 2) in the combined atrasentan treatment group. A change in the MSDKD panel was also significantly associated with change in eGFR (R2= 0.42; p<0.001; Table 2 and Supplemental Figure 2). The individual metabolites and the MSDKD panel did not correlate with baseline UACR (Supplemental Table 2) or changes in UACR at week 12 (Supplemental Table 3).

139138


6

Figure 1. Effects of atrasentan A) 0.75mg/d and B) 1.25 mg/d relative to placebo on individual metabolites at week 6 and week 12 in patients with baseline eGFR <60 mL/min/m2 (n=121). Figure 2. Mean values of the MSDKD panel and 95% confidence intervals for placebo, atrasentan

0.75 mg/d, and atrasentan 1.25 mg/d in patients with baseline eGFR <60 mL/min/m2 (n=121). * P-value 0.024 for comparison placebo vs. atrasentan 1.25 mg/d.

141140


6

Table 2. Correlations between change in eGFR and change in metabolites between baseline and 12 weeks of follow-up in patients with baseline eGFR <60 mL/min/m2 (n=121).

Change in eGFR

β S.E. R2 P-value

Placebo (n=24)

ΔAconitic acid 4.9 5.8 0.03 0.407

ΔCitric acid -0.2 4.4 0.00 0.964

ΔGlycolic acid -1.8 6.4 0.00 0.778

ΔHomovanillic acid 7.3 7.7 0.04 0.358

Δ2-ET-3-OH-Propionate 11.9 7.6 0.11 0.135

Δ3-OH-Isobutyrate 1.0 4.4 0.00 0.815

Δ3-OH-Isovalerate 2.6 7.0 0.01 0.712

Δ2-3-OH-Propionate 3.7 6.5 0.02 0.577

ΔUracil 0.2 4.7 0.00 0.963

ΔMSDKD panel 0.7 1.3 0.02 0.573

Atrasentan (n=97)

ΔAconitic acid 21.3 5. 5 0.16 <.001

ΔCitric acid 14.1 3.2 0.20 <.001

ΔGlycolic acid 30.8 5.6 0.27 <.001


Δ2-ET-3-OH-Propionate 22.1 5.6 0.17 <.001

Δ3-OH-Isobutyrate 18.7 4.0 0.22 <.001

Δ3-OH-Isovalerate 37.1 5.8 0.34 <.001

Δ2-3-OH-Propionate 23.9 4.6 0.25 <.001

ΔUracil 7.2 3.5 0.05 0.043

ΔMSDKD panel 8.6 1.1 0.42 <.001

DISCUSSION

This study demonstrated that concentrations of urinary metabolites from a metabolomics signature of diabetic kidney disease were significantly correlated with eGFR levels in patients with type 2 diabetes and nephropathy. The levels of the metabolites remained stable after 12 weeks of treatment with atrasentan 1.25 mg/d whereas they decreased with placebo. Individual changes in the metabolites after 12 weeks of treatment positively correlated with changes in eGFR. Since lower metabolite concentrations reflect reduced mitochondrial content and renal function [1], it is plausible that treatment with atrasentan may stabilize mitochondrial function.

The correlation between lower urinary metabolite concentrations and lower eGFR were consistent with the findings of the prior discovery study of the metabolomics signature of diabetic kidney disease in patients with diabetic nephropathy [1]. That study reported that urinary concentrations of 12 of the 13 metabolites were significantly decreased in patients with diabetic nephropathy compared to patients with type 2 diabetes and eGFR>60 ml/min/1.73m2. Systems biology analysis revealed that the reduced metabolite concentrations were associated with generalized reduction in mitochondrial function, a hypothesis confirmed by kidney biopsy staining samples from healthy volunteers and patients with diabetic nephropathy [4]. In this present study, four metabolites could not be detected. Since levels of the urine metabolites were found to be lower with reduced eGFR [1], it may be possible that eGFR was too low to detect these metabolites. In the current study we also investigated changes in metabolites over time during placebo or treatment with the endothelin A receptor antagonist atrasentan. We focused our investigation on patients with baseline eGFR <60 ml/min/1.73m2 as metabolites were reduced in these patients. Interestingly, atrasentan increased the individual metabolite levels relative to placebo, suggesting that atrasentan may preserve aspects of renal mitochondrial function. In the combined MSDKD panel, a significant difference between placebo and 1.25mg/d atrasentan was observed, but not between placebo and 0.75 mg/d atrasentan, indicating a dose-dependent effect on these metabolites. On one hand, it may be possible that atrasentan directly stabilizes mitochondrial function. On the other hand, endothelin receptor antagonists have been shown to decrease mitotic activity and have anti-inflammatory effects, which may affect mitochondrial function [3, 9]. In association with data from patients and animal models, an improvement in renal mitochondrial function could be expected to improve underlying renal structure and function [1, 10].

What is known about the interaction between the endothelin system and mitochondrial function? Recent studies suggest that endothelin regulates mitochondrial biogenesis by inhibiting the transcriptional regulator peroxisome proliferator-activated receptor gamma coactivator 1-alpha (PGC-1α) [11, 12]. Inhibition of endothelin by atrasentan could stimulate PGC-1α and increase mitochondrial biogenesis. Additionally, results from animal studies suggest that endothelin receptor blockade attenuates cardiac injury and impairment of mitochondrial biogenesis, and decreased endothelin-1 induced by endothelin-converting

143142


6

enzyme-1 inhibition directly leads to cardioprotection and preserved mitochondrial biogenesis [11]. Increased mitochondria content could result in increased or stabilized mitochondrial-derived metabolites. An overall stabilization in mitochondrial function may provide a new hypothesis for a beneficial renoprotective effect of atrasentan in human diabetic kidney disease as several of the reduced mitochondrial-derived metabolites are closely correlated with eGFR [1]. The individual metabolites and the MSDKD panel did not correlate with changes in UACR. This implies that the treatment effect on the metabolites is unrelated to the effect on albuminuria. Although of importance, these results should be interpreted with caution since the 12 weeks follow-up is too short to adequately assess drug effects on overall renal function. A long-term trial is currently ongoing (SONAR trial NCT01858532) that will assess the potential long-term renoprotective effect of atrasentan and will allow further study of atrasentan’s potential renoprotective mechanism of action.

To the best of our knowledge, the present study appears to be a first in examining metabolite profiles in response to therapy in type 2 diabetes and nephropathy. While an increasing number of studies are exploring the metabolomics profiles in relation to renal function decline in type 2 diabetes [13, 14], studies evaluating the effects of drug treatment on metabolomics profiles are lacking. Such studies are needed to identify novel biomarkers for monitoring drug efficacy as well as providing insights into molecular mechanisms of drug effects and drug response variability.

Limitations of this study include a short follow-up duration and thus inability to assess whether the change in metabolites are sustained after the 12 weeks follow-up period. Another limitation is the lack of plasma samples that would allow for an assessment of whether the metabolites regulated in urine were similarly regulated in plasma. This would help to assess whether the effect of the drug therapy was potentially systemic or primarily renal, and whether these urinary changes reflect systemic changes, renal changes only, or simply changes in renal hemodynamics or creatinine handling. Moreover, urinary metabolites could be filtered in the glomerulus and the glomerular barrier function can be improved by endothelin receptor antagonists, thereby potentially altering the urinary metabolite concentration [4, 15]. This study, however, is unable to address these potential confounders. Further analysis of samples from this study and future studies will be of great interest to further understand the role of targeted metabolomics in identifying potentially important biomarkers and possible surrogates for diabetic kidney disease progression. Finally, the relatively small number of participants limited the study power and strength of the conclusions. Validation of this urinary metabolomics panel in larger studies with hard clinical endpoints is required to support our hypothesis. Additional studies may provide the data for using metabolites as a potential companion diagnostic or as a surrogate marker for atrasentan treatment in type 2 diabetes complications.

In conclusion, a specific panel of urinary metabolites linked to mitochondrial function was significantly correlated with eGFR levels in patients with type 2 diabetes and nephropathy. Treatment with atrasentan 1.25 mg/d for 12 weeks stabilized the levels of the metabolites while they declined with placebo treatment, implying that treatment with atrasentan may prevent their reduction as compared to placebo. Long-term hard outcome trials are required to confirm these findings and assess whether the short-term changes in these metabolites portend an improvement in long-term renal function.

FUNDING

AbbVie participated in the study design along with funding of the RADAR study. HJ Lambers Heerspink is supported by a VIDI grant from the Netherlands Organisation for Scientific Research.

145144


6

REFERENCES





J Am Soc Nephrol 2013 Nov;24(11):1901-1912

2. Pena MJ, de Zeeuw D, Mischak H, Jankowski J, Oberbauer R, Woloszczuk W, Benner J,

Dallmann G, Mayer B, Mayer G, Rossing P, Lambers Heerspink HJ. Prognostic clinical and

molecular biomarkers of renal disease in type 2 diabetes. Nephrol Dial Transplant. 2015

Aug;30(suppl 4):iv86-iv95

3. Barton M, Yanagisawa M. Endothelin: 20 years from discovery to therapy. Can J Physiol

Pharmacol 2008 Aug;86(8):485-498

4. de Zeeuw D, Coll B, Andress D, Brennan JJ, Tang H, Houser M, Correa-Rotter R, Kohan

D, Lambers Heerspink HJ, Makino H, Perkovic V, Pritchett Y, Remuzzi G, Tobe SW, Toto R,

Viberti G, Parving HH. The endothelin antagonist atrasentan lowers residual albuminuria in

patients with type 2 diabetic nephropathy. J Am Soc Nephrol 2014 May;25(5):1083-1093

5. Kohan DE, Pritchett Y, Molitch M, Wen S, Garimella T, Audhya P, Andress DL. Addition

of atrasentan to renin-angiotensin system blockade reduces albuminuria in diabetic

nephropathy. J Am Soc Nephrol 2011 Apr;22(4):763-772

6. Levey AS, Stevens LA, Schmid CH, Zhang YL, Castro AF,3rd, Feldman HI, Kusek JW,

Eggers P, Van Lente F, Greene T, Coresh J, CKD-EPI (Chronic Kidney Disease Epidemiology

Collaboration). A new equation to estimate glomerular filtration rate. Ann Intern Med 2009

May 5;150(9):604-612

7. Schram MT, Chaturvedi N, Schalkwijk CG, Fuller JH, Stehouwer CD, EURODIAB Prospective

Complications Study Group. Markers of inflammation are cross-sectionally associated

with microvascular complications and cardiovascular disease in type 1 diabetes--the

EURODIAB prospective complications study. Diabetologia 2005 Feb;48(2):370-378



in the irbesartan in patients with type 2 diabetes and microalbuminuria (IRMA 2) study.

Scand J Clin Lab Invest 2008;68(8):731-738

9. Kohan DE, Barton M. Endothelin and endothelin antagonists in chronic kidney disease.

Kidney Int 2014 Nov;86(5):896-904

10. Dugan LL, You YH, Ali SS, Diamond-Stanic M, Miyamoto S, DeCleves AE, Andreyev A,

Quach T, Ly S, Shekhtman G, Nguyen W, Chepetan A, Le TP, Wang L, Xu M, Paik KP, Fogo

A, Viollet B, Murphy A, Brosius F, Naviaux RK, Sharma K. AMPK dysregulation promotes

diabetes-related reduction of superoxide and mitochondrial function. J Clin Invest 2013

Nov;123(11):4888-4899

11. Miyagawa K, Emoto N, Widyantoro B, Nakayama K, Yagi K, Rikitake Y, Suzuki T, Hirata K.

Attenuation of doxorubicin-induced cardiomyopathy by endothelin-converting enzyme-1

ablation through prevention of mitochondrial biogenesis impairment. Hypertension 2010

Mar;55(3):738-746

12. Garnier A, Zoll J, Fortin D, N’Guessan B, Lefebvre F, Geny B, Mettauer B, Veksler V,

Ventura-Clapier R. Control by circulating factors of mitochondrial function and transcription

cascade in heart failure: A role for endothelin-1 and angiotensin II. Circ Heart Fail 2009

Jul;2(4):342-350

13. Niewczas MA, Sirich TL, Mathew AV, Skupien J, Mohney RP, Warram JH, Smiles A, Huang

X, Walker W, Byun J, Karoly ED, Kensicki EM, Berry GT, Bonventre JV, Pennathur S, Meyer

TW, Krolewski AS. Uremic solutes and risk of end-stage renal disease in type 2 diabetes:

Metabolomic study. Kidney Int 2014 May;85(5):1214-1224




mellitus. Diabet Med 2014 Sep;31(9):1138-1147

15. Saleh MA, Pollock JS, Pollock DM. Distinct actions of endothelin A-selective versus

combined endothelin A/B receptor antagonists in early diabetic kidney disease. J

Pharmacol Exp Ther. 2011 Jul;338(1):263-70

147146


6

Sup

ple

men

tal T

able

1. B

asel

ine

Cha

ract

eris

tics

in p

atie

nts

with

bas

elin

e eG

FR <

60 m

L/m

in/m

2 (n

=12

1).

Pla

ceb

o (n

=24

)A

tras

enta

n 0.

75m

g/d

(n=

46)

Atr

asen

tan

1.25

mg

/d (n

=51

)p

-val

ue*

Age

, yea

rs (S

D)

62.9

(9.2

)66

.0 (8

.1)

64.3

(8.9

)0.

351

Sex

, num

ber

(%)

0.11

3

M

ale

17 (7

0.8)

39 (8

5)34

(67)

-

F

emal

e7

(29.

2)7

(15)

17 (3

3)-

Rac

e, n

umb

er (%

)0.

172

C

auca

sian

19 (7

9.2)

25 (5

7)32

(63)

-

O

ther

5 (2

0.8)

20 (4

3)19

(37)

-

SB

P, m

m H

g (S

D)

137.

6 (1

5.0)

137.

2 (1

2.8)

135.

3 (1

4.7)

0.72

5

DB

P, m

m H

g (S

D)

72.3

(11.

4)73

.6 (8

.8)

72.9

(9.3

)0.

839

eGFR

, ml/m

in/1

.73m

2 (S

D)

42.5

(9.6

)40

.8 (8

.9)

7.5

(44.

0)0.

119

Alb

umin

uria

, mg/

g [1

st, 3

rd q

uart

ile]

759.

3 [4

88.2

, 135

2.1]

855.

5 [5

22.4

, 168

8.3]

816.

7 [4

65.2

, 131

2.2]

0.63

3

Ind

ivid

ual M

SD

KD

Met

abol

ites

(µm

ol o

rgan

ic a

cid

/mm

ol u

rinar

y cr

eatin

ine)

Aco

nitic

aci

d14

.5 [1

1.2,

19.

5]13

.3 [9

.5, 1

7.9]

15.2

[12.

2, 1

8.6]

0.41

9

Citr

ic a

cid

105.

1 [7

0.9,

241

.2]

134.

2 [7

6.3,

192

.3]

147.

2 [8

4.4,

219

.2]

0.95

6

Gly

colic

aci

d13

.5 [1

2.8]

14.1

[11.

4, 1

7.2]

14.8

[12.

5, 1

7.9]

0.38

8

Hom

ovan

illic

aci

d1.

9 [1

.6, 2

.3]

0.8

[0.7

, 1.2

]1.

0 [1

.3]

0.77

2

2-E

T-3-

OH

-Pro

pio

nate

1.1

[1.0

, 1.6

]0.

3 [0

, 0.6

]0.

5 [0

.03,

0.8

]0.

248

3-O

H-I

sob

utyr

ate

19.1

[14.

2, 2

3.6]

17.3

[13.

0, 2

2.4]

18.5

[13.

4, 2

2.7]

0.84

0

3-O

H-I

sova

lera

te6.

2 [5

.2, 8

.3]

5.0

[4.2

, 6.5

]6.

0 [4

.7, 7

.5]

0.14

0

3-O

H-P

rop

iona

te1.

8 [1

.6, 2

.4]

0.7

[0.6

, 1.1

]1.

0 [0

.7, 1

.3]

0.24

7

Ura

cil

3.2

[2.1

, 3.8

]2.

0 [1

.3, 3

.2]

2.1

[1.0

, 2.7

]0.

701

MS

DK

D p

anel

-0.2

[-0.

6, 0

.3]

-0.2

[-0.

6, 0

.2]

0.1

[-0.

4, 0

.4]

0.39

6

Dat

a ar

e re

por

ted

as

mea

n ±

sta

ndar

d d

evia

tion

(SD

) or

num

ber

(per

cent

) or

med

ian

[25th

qua

rtile

, 75th

qua

rtile

]. *T

o te

st d

iffer

ence

s b

etw

een

grou

ps,

A

NO

VA w

as u

sed

for

par

amet

ric d

ata,

Kru

skal

-Wal

lis t

est

for

non-

par

amet

ric d

ata,

and

chi

-sq

uare

goo

dne

ss o

f fit

test

for

pro

por

tions

.

Supplemental Table 2. Pearson’s correlations between metabolites and UACR at baseline in the overall RADAR population (n=150).

Individual Metabolites Pearson’s Coefficient P-value

Aconitic acid -0.17 0.039

Citric acid -0.05 0.583

Glycolic acid 0.05 0.565

Homovanillic acid -0.13 0.106

2-ET-3-OH-Propionate 0.002 0.976

3-OH-Isobutyrate 0.02 0.801

3-OH-Isovalerate 0.04 0.612

3-OH-Propionate 0.09 0.282

Uracil -0.03 0.707

MSDKD panel -0.0359 0.663

149148


6

Supplemental Table 3. Correlations between change in UACR and change in metabolites between baseline and 12 weeks of follow-up in patients with baseline eGFR <60 mL/min/m2 (n=121).

Change in UACR

β S.E. R2 P-value

Placebo (n=24)

ΔAconitic acid 0.07 0.40 0.002 0.855

ΔCitric acid 0.57 0.27 0.18 0.050

ΔGlycolic acid 0.16 0.44 0.01 0.726

ΔHomovanillic acid -0.19 0.54 0.01 0.725


Δ3-OH-Isobutyrate 0.23 0.30 0.03 0.447

Δ3-OH-Isovalerate 0.12 0.48 0.003 0.810

Δ2-3-OH-Propionate 0.35 0.44 0.03 0.432

ΔUracil 0.12 0.32 0.01 0.700

ΔMSDKD panel 0.05 0.09 0.01 0.594

Atrasentan (n=97)

ΔAconitic acid 0.59 0.43 0.02 0.174

ΔCitric acid 0.83 0.24 0.13 <.001

ΔGlycolic acid 0.68 0.47 0.03 0.156



Δ3-OH-Isobutyrate 0.30 0.32 0.01 0.357

Δ3-OH-Isovalerate 0.57 0.51 0.02 0.268

Δ2-3-OH-Propionate 0.59 0.39 0.01 0.459

ΔUracil 0.12 0.26 0.003 0.633

ΔMSDKD panel 0.20 0.11 0.04 0.060

Supplemental Figure 1. Mean values of the MSDKD panel and 95% confidence intervals in the overall RADAR population (n=150) randomly assigned to placebo, atrasentan 0.75 mg/d, or atrasentan 1.25 mg/d.

151150


6

Supplemental Figure 2. Correlation of change in the MSDKD panel with change in eGFR from baseline to week 12 in the atrasentan group in patients with baseline eGFR <60 mL/min/m2 (n=97). The dotted horizontal grey line indicates the mean eGFR change during 12 weeks of atrasentan treatment, and the vertical dotted line indicates the mean change in the MSDKD panel. The dark grey diagonal line is the regression line.

CHAPTER 7Summary and future perspectives

155154

Summary and future perspectivesChapter 7

7

SUMMARY

The overall premise of this thesis is that earlier detection of diabetic kidney disease (DKD) in patients with type 2 diabetes and subsequent intervention for increased risk of DKD may ultimately help alleviate the burden of end-stage renal disease (ESRD). One strategy to accomplish this may be the use of novel biomarker panels to predict progression of renal disease. Furthermore, this thesis examined the use novel biomarker panels for predicting response to therapy and monitoring the effect of therapeutic intervention.

Type 2 diabetes is a multifactorial disease involving different pathophysiologic molecular processes with a heterogeneous histopathological structure [1]. We hypothesized that a combination of biomarkers that capture different pathogenic processes of renal damage may provide a more realistic picture of a patient’s actual pathophysiological status. This thesis therefore focused on multiple biomarkers in a panel instead of single biomarkers. Recent advances in laboratory and throughput technologies have helped generate an expansive inventory of potential biomarker panels for renal disease in type 2 diabetes [2].

Novel biomarker panels can be used not only for risk prediction at early disease stages but also importantly for drug response prediction and monitoring therapeutic effect. Many patients with type 2 diabetes treated with guideline recommended therapy still face a high risk of renal disease progression and variability in response to therapy. Therefore, the use of novel biomarker panels in the context of drug response prediction or monitoring therapy is extremely important to reduce variability drug response, minimize side effects, or off-target effects, which is observed with many established and novel drugs. Using novel biomarkers to improve on current practices to improve risk stratification, help increase our understanding of renal disease pathophysiology, and provide insight into novel therapeutic targets can ultimately help reduce the burden renal disease in type 2 diabetes.

Novel biomarker panels as predictors of renal diseaseIn Part 1, this thesis investigated the predictive ability of novel biomarker panels for the progression of renal disease in patients with type 2 diabetes. Novel biomarkers may provide deeper understanding into the pathophysiology of DKD. Furthermore, identification of progression-associated molecular pathways via biomarkers as proxy may also help to identify novel therapeutic targets. Given the complexity of the multiple pathophysiological processes involved in progression of renal disease in type 2 diabetes together with the intra-individual variability of biomarkers, it is questionable if a single biomarker may possess useful diagnostic and prognostic power. Alternatively, a panel of clearly defined biomarkers may provide a more robust and reproducible tool as it tolerates changes in single biomarkers without jeopardizing their diagnostic ability. A combination of biomarkers that capture different pathways of renal damage may provide a more realistic picture of a patient’s actual pathophysiological status and hence may yield better assessment of disease prognosis performance.

Chapter 2 presented an observational cohort study of 82 patients with type 2 diabetes followed for 4 years. A panel of 13 novel biomarkers was associated with accelerated renal function decline beyond established risk markers. This study identified a novel panel of biomarkers representing different pathways of renal damage, including inflammation, fibrosis, angiogenesis, and endothelial function. This combined panel improved prediction of accelerated renal function decline in patients with type 2 diabetes on top of established risk markers. There was however, no external validation of this biomarker panel, and the small sample size limit the study results. Therefore, these results need to be validated in a large, prospective cohort to validate and assess this novel biomarker panel’s applicability in a broad type 2 diabetes population.

The measurement of peptides and metabolites, known as proteomics and metabolomics, have emerged as strong tools in biomarker discovery [3]. Chapter 3 assessed the predictive ability of plasma proteomics classifiers to improve risk prediction of transition in albuminuria stage in patients with hypertension or type 2 diabetes. The developed plasma proteomics classifiers were able to predict transition in stage of albuminuria in hypertensive patients and transition from micro- to macroalbuminuria in patients suffering from type 2 diabetes. This was independent of established the renal risk markers urinary albumin excretion (UAE) and estimated glomerular filtration rate (eGFR), as well as use of renin-angiotensin-aldosterone system (RAAS) intervention. The plasma peptides identified in hypertension and type 2 diabetes were linked to pathways associated with established mechanisms of renal disease, including fibrosis, inflammation, angiogenesis, and mineral metabolism. These results support the growing evidence of using peptidomic platforms as a strategy for risk stratification of renal disease.

Metabolomics is another potential tool to discover novel biomarkers for renal disease. Chapter 4 investigated the predictive ability of urine and plasma metabolites for the development of diabetic nephropathy in patients with type 2 diabetes. In this discovery study, plasma metabolites histidine and butenolycarnatine and urine metabolites hexoses, glutamine, and tyrosine were able to predict progression from micro- to macroalbuminuria on top of established renal risk markers in type 2 diabetes. These metabolites did not predict albuminuria progression in patients with hypertension, suggesting a type 2 diabetes specific metabolite profile. In this study, the urinary metabolome profile performed better than the plasma profile, and addition of the plasma metabolites to the urinary metabolites did not improve outcome prediction. This may point at urinary metabolomics as a better clinical approach to identify people with type 2 diabetes at risk of progressive renal disease (also due to practical advantages of collecting urine compared to blood samples). However, the results from the plasma metabolomics may be useful to identify underlying mechanisms and pathways of disease and should not be neglected. Again, external validation of these metabolites is still needed to give firm conclusions for these metabolites’ predictive ability in a broader population.

157156


7

Novel biomarker panels for predicting response to therapyFirst choice, guideline recommended therapy for treatment of hypertension and albuminuria for patients with type 2 diabetes are either angiotensin converting enzyme inhibitors (ACEi) or angiotensin II receptor blockers (ARB) [4]. However, many patients still do not respond optimally to ACEi or ARBs [5-9]. This may result in progressive renal function loss. One approach to reduce this variation in response to therapy is to use novel biomarkers to predict response to therapy. Chapter 5 discovered and externally validated a serum metabolite classifier that significantly improves prediction of albuminuria response to ARBs in patients with diabetes mellitus on top of traditional clinical risk markers. Metabolites included in the classifier were related to oxidative stress, inflammation, and fibrosis pathways. Specifically, increased nitric oxide synthase 3 (NOS3) activity appears to be a factor in predicting the albuminuria lowering response to ARBs. These findings suggest the use of serum metabolomics as a tool to help optimize treatment of antiproteinuric ARB intervention. Metabolites included in the classifier were assigned to stress/inflammation pathways and downstream consequences of fibrosis and extra cellular matrix remodeling. Furthermore, NOS3 appears to be a specific factor relevant in ARB response. These results indicate that for assessing drug response, both disease progression status and specific drug molecular effects need to be taken into account. The results of this metabolomics study support the growing evidence of using omics tools as a strategy to improve treatment of renal disease in diabetes mellitus. The complementary use of omics platforms can help bring personalized medicine one step closer to implementation in clinical practice.

Novel biomarker panels for monitoring the effect of therapyNovel biomarker panels can be used to monitor the response to treatment. Chapter 6 investigated the effect of atrasentan on the previously identified metabolomics signature of diabetic kidney disease [10]. This study first demonstrated that concentrations of urinary metabolites from this panel were significantly correlated with eGFR levels in patients with type 2 diabetes and nephropathy. We then assessed the beneficial impact of atrasentan on this urinary metabolite panel. Treatment with atrasentan 1.25 mg/d for 12 weeks stabilized the levels of the metabolites while they declined with placebo treatment, implying that treatment with atrasentan may prevent their reduction as compared to placebo. Individual changes in the metabolites after 12 week treatment positively correlated with changes in eGFR. Lower metabolite concentrations have been previously shown to reflect reduced mitochondrial content and renal function [10], therefore, it is plausible that treatment with atrasentan may stabilize several aspects of mitochondrial function. Long-term hard outcome trials are required to assess whether these short-term beneficial effects portend in improvement in overall renal function. Results of further studies may help provide insight into the mechanisms through which atrasentan exerts renoprotective effects and may yield novel biomarkers to monitor response to therapy in patients with type 2 diabetes and DKD.

FUTURE PERSPECTIVES

Before a biomarker or panel of biomarkers can be used in clinical practice, it needs to be extensively validated in large studies to assess accuracy, reproducibility, sensitivity, and specificity. The translation of a biomarker or a combination of biomarkers from discovery to clinical practice is a process full of pitfalls and limitations. In 2009, Hlatky et al. proposed a framework for the development of biomarkers in collaboration with the American Heart Association [11], as listed in Box 1. Unfortunately, many novel biomarker panel studies for renal disease in type 2 diabetes stop at the end of the third phase [12], and do not proceed to external validation or assessment of clinical utility. More awareness and investments need to be made to preform studies in the clinical utility and clinical outcome phases in order to start implementing novel biomarker panels in clinical practice. Better study designs that can test the biomarkers’ practical value to translate the predictive power of a biomarker panel into decisions for clinical practice would help address non-acceptance of novel biomarkers by professional communities. Furthermore, collaborations between academia and industry may be a strategy to share expertise from different areas, to promote effective dissemination of results, and to support implementation of these findings in clinical practice.

Box 1. Framework for the development of novel biomarkers. Adapted from Hlatky et al. Circulation 2009 [11]

1. Proof of conceptDo novel marker levels differ between subjects with and without outcome?

2. Prospective validationDoes the novel marker predict development of future outcomes in a prospective cohort or nested case-cohort/case-cohort study?

3. Incremental valueDoes the novel marker add predictive information to established, standard risk markers?

4. Clinical utilityDoes the novel risk marker change predicted risk sufficiently to change recommended therapy?

5. Clinical outcomesDoes use of the novel risk marker improve clinical outcomes, especially when tested in a randomized clinical trial?

6. Cost-effectivenessDoes use of the marker improve clinical outcomes sufficiently to justify the additional costs of testing and treatment?

Much of this thesis consists of research performed with the SysKid program (Systems Biology towards Novel Chronic Kidney Disease Diagnosis and Treatment, www.syskid.eu). SysKid was a large-scale European research project involving both academia and industry partners. SysKid aimed at improving characterization of the molecular mechanisms underlying diabetic kidney disease at the interference of the molecular impact of individual drugs in order to tailor optimal therapy to individual patients. Results and continuing

159158


7

work stemming from SysKid have deepened our understanding of chronic kidney disease (CKD) in term of prevention, new diagnostic strategies, and treatment options for renal disease in diabetes and hypertension. Furthermore, there is an ongoing validation study in a large, type 2 diabetes population that will assess the validity of the novel biomarker panels described in Chapter 2-4 of this thesis. Though this validation study will not be able to provide further knowledge toward clinical utility or cost-effectiveness, there is the opportunity to evaluate the predictive ability of these novel biomarkers panels for hard renal outcomes. Furthermore, this validation study provides the opportunity to compare the biomarkers at early stage, mid stage, and late stage DKD.

Use of biomarker panels for risk stratification in clinical practice is one strategy to guide treatment decisions and target interventions for patients at highest risk for progressive renal disease. Within SysKid, validation of a novel urinary peptide panel known as the CKD273 score was performed [13]. This CE-MS based urinary proteomics classifier was developed in a cross-sectional study in CKD patient groups with varying underlying etiologies of disease [14]. The clinical utility of the CKD273 score is currently being tested in the Proteomic Prediction and Renin Angiotensin Aldosterone System Inhibition Prevention Of Early Diabetic nephRopathy In TYpe 2 Diabetic Patients With Normoalbuminuria (PRIORITY) trial (Clinical Trials Identifier NCT02040441) [15]. The PRIORITY trial will assess the clinical utility of the CKD273 score for the purpose of guiding treatment of patients with type 2 diabetes at high risk for renal disease progression. The PRIORITY trial uses the CKD273 score to identify patients at high risk of renal disease while still in early disease stages. The primary objective of the PRIORITY trial is to confirm whether urinary proteomics can predict development of microalbuminuria. The PRIORITY study will also assess whether high-risk patients identified with the CKD273 score will benefit from spironolactone therapy.

Current treatment of type 2 diabetes relies on optimal glycemic, lipid, and blood pressure control. Past randomized control studies has shown that targeting HbA1c, lipid management, and blood pressure control delays the progression of DKD [16-20]. However, even in these successful studies, the residual renal risk is still quite high, implying that the current treatment regime is insufficient to prevent progression to ESRD in a substantial proportion of patients. Given the large heterogeneity in pathophysiology of DKD and the substantial variability in response to renoprotective drugs, treatment strategies need to go in a new direction. Using novel biomarker panels to identify patients who would respond well to treatment before therapy has even been started, may help decrease variability in treatment response. An ideal scenario to implement use of novel biomarkers in clinical practice would be to identify individuals at risk for progressive disease while at the same time identify those who would respond well to treatment. However there is currently a paucity of studies that evaluate whether novel biomarkers predict drug response. Chapter 5 of this thesis is just the start in utilizing novel biomarkers to predict response to therapy, and future studies are needed to push this field forward.

Studies that assess whether treatment induced changes in novel biomarkers are associated with renal risk are lacking. This will be an area of interest for the future as it may help us to identify novel biomarkers that can be used to monitor drug efficacy, generate information about the molecular mechanisms through which drugs exert their effects, and provide insight into novel drug targets. It may be possible that a patient benefits from a drug without changing the urinary albumin levels and thus other biomarkers are needed to monitor drug efficacy or safety. Chapter 6 of this thesis provides some evidence towards proof of concept for use of a novel biomarker panel to monitor the effect of drug therapy, though long-term hard outcome trials are still required to confirm these findings and assess whether the short-term changes in these metabolites portend an improvement in long-term renal function.

In order to go beyond the status quo, novel approaches to treat patients with renal disease are necessary to improve disease outcomes. As advancing laboratory techniques become more and more realistic in clinical practice, the use of omics techniques as a tool to reduce the burden of renal disease complications represents a real strategy to improve disease diagnosis, prognosis, and treatment. Rich omics datasets have improved molecular phenotyping and characterization of the underlying molecular mechanisms of DKD on the level of individual patients. These datasets should also be used in the future to assess mechanisms of drug response and discover biomarkers that predict drug response. Collectively the omics methodologies, molecular process mapping, and from there developing novel biomarkers would contribute to the identification of the optimal therapeutic approach to provide the greatest benefit with minimal side effects or off-target effects (Figure 1). This may help improve on the current “trial-and-error” approach to choosing drugs for treatment of DKD in patients with type 2 diabetes and bring personalized medicine one step closer to clinical practice.

Figure 1. Creating a systems medicine approach to study drug response variability by adding post-genomic information to the genotype information, thereby connecting the complex genotype with the phenotype.

161160


7

REFERENCES

1. Fioretto P, Stehouwer CD, Mauer M, Chiesura-Corona M, Brocco E, Carraro A, Bortoloso

E, van Hinsbergh VW, Crepaldi G, Nosadini R. Heterogeneous nature of microalbuminuria

in NIDDM: studies of endothelial function and renal structure. Diabetologia. 1998

Feb;41(2):233-6.




Aug;30 Suppl 4:iv86-iv95.

3. Komorowsky CV, Brosius FC,3rd, Pennathur S, Kretzler M. Perspectives on systems biology




Int Suppl. 2013; 3:1-150.

















2008 Mar;73(5):630-6.






11. Hlatky MA, Greenland P, Arnett DK, Ballantyne CM, Criqui MH, Elkind MS, Go AS, Harrell

FE Jr, Hong Y, Howard BV, Howard VJ, Hsue PY, Kramer CM, McConnell JP, Normand

SL, O’Donnell CJ, Smith SC Jr, Wilson PW; American Heart Association Expert Panel on

Subclinical Atherosclerotic Diseases and Emerging Risk Factors and the Stroke Council.

Criteria for evaluation of novel markers of cardiovascular risk: a scientific statement from

the American Heart Association. Circulation. 2009 May 5;119(17):2408-16.

12. Schutte E, Gansevoort RT, Benner J, Lutgers HL, Lambers Heerspink HJ. Will the future

lie in multitude? A critical appraisal of biomarker panel studies on prediction of diabetic

kidney disease progression. Nephrol Dial Transplant. 2015 Aug;30 Suppl 4:iv96-iv104.




2013 Feb;56(2):259-67.









15. Siwy J, Schanstra JP, Argiles A, Bakker SJ, Beige J, Boucek P, Brand K, Delles C, Duranton

F, Fernandez-Fernandez B, Jankowski ML, Al Khatib M, Kunt T, Lajer M, Lichtinghagen

R, Lindhardt M, Maahs DM, Mischak H, Mullen W, Navis G, Noutsou M, Ortiz A, Persson

F, Petrie JR, Roob JM, Rossing P, Ruggenenti P, Rychlik I, Serra AL, Snell-Bergeon J,

Spasovski G, Stojceva-Taneva O, Trillini M, von der Leyen H, Winklhofer-Roob BM,

Zürbig P, Jankowski J. Multicentre prospective validation of a urinary peptidome-based

classifier for the diagnosis of type 2 diabetic nephropathy. Nephrol Dial Transplant. 2014

Aug;29(8):1563-70.

163162


7

16. UK Prospective Diabetes Study (UKPDS) Group. Intensive blood-glucose control with

sulphonylureas or insulin compared with conventional treatment and risk of complications

in patients with type 2 diabetes (UKPDS 33). Lancet. 1998 Sep 12;352(9131):837-53.

17. Parving HH, Lehnert H, Bröchner-Mortensen J, Gomis R, Andersen S, Arner P; Irbesartan

in Patients with Type 2 Diabetes and Microalbuminuria Study Group. The effect of

irbesartan on the development of diabetic nephropathy in patients with type 2 diabetes. N

Engl J Med. 2001 Sep 20;345(12):870-8.




Engl J Med. 2001 Sep 20;345(12):861-9.




2001 Sep 20;345(12):851-60.

20. Colhoun HM, Betteridge DJ, Durrington PN, Hitman GA, Neil HA, Livingstone SJ,

Thomason MJ, Mackness MI, Charlton-Menys V, Fuller JH; CARDS investigators.

Primary prevention of cardiovascular disease with atorvastatin in type 2 diabetes in the

Collaborative Atorvastatin Diabetes Study (CARDS): multicentre randomised placebo-

controlled trial. Lancet. 2004 Aug 21-27;364(9435):685-96.

Nederlandse samenvatting en toekomstperspectief

167166

Nederlandse samenvatting en toekomstperspectiefNederlandse samenvatting en toekomstperspectief

SAMENVATTING

De veronderstelling die ten grondslag ligt aan dit proefschrift is dat vroegere opsporing van en interventie bij patiënten met diabetes mellitus type 2 die een verhoogd risico lopen op diabetische nefropathie uiteindelijk de ziektelast van eindstadium nierfalen (ESRD) kunnen verlichten. Een manier om dit doel te bereiken is het gebruik van nieuwe biomarkerpanels om progressie van nefropathie te voorspellen. Daarnaast werd in dit proefschrift de voorspellende waarde van nieuwe biomarkerpanels voor respons op therapie onderzocht. Tevens werd het monitoren van het effect van therapeutische interventie onderzocht.

Diabetes mellitus type 2 is een multifactoriële ziekte waarin verschillende pathofysiologische moleculaire processen met heterogene histopathologische structuren betrokken zijn [1]. Wij veronderstelden dat een combinatie van biomarkers, die de verschillende pathologische processen van nierschade reflecteren, mogelijk een meer realistische weerspiegeling zijn van de actuele pathofysiologische status van een patiënt. Dit proefschrift richtte zich daarom op panels van meerdere biomarkers in plaats van individuele biomarkers. Recente ontwikkelingen in laboratorium- en throughput-technieken hebben bijgedragen aan een uitgebreide lijst potentiële biomarkerpanels voor nefropathie in diabetes mellitus type 2 [2].

Nieuwe biomarkers kunnen naast risicovoorspelling in vroege ziektestadia ook gebruikt worden om respons op medicatie te voorspellen en om therapeutische effecten te monitoren. Veel patiënten met diabetes mellitus type 2, die volgens de richtlijnen worden behandeld, houden een hoog risico op progressie van nefropathie en vertonen en variabele respons op therapie. Het gebruik van nieuwe biomarkerpanels om het effect op therapie te voorspellen en te monitoren is van groot belang om de variatie in respons te verminderen, de kans op bijwerkingen te minimaliseren en de kans op off-targeteffecten te verkleinen die frequent optreden bij het gebruik van klassieke en nieuwe medicatie. Het gebruik van nieuwe biomarkers om de huidige klinische praktijk en risico-stratificatie te verbeteren draagt bij aan een beter begrip van de pathofysiologische processen van nefropathie, geeft inzicht in nieuwe farmaceutische aangrijpingspunten en kan uiteindelijk de ziektelast van diabetische nefropathie doen afnemen.

Biomarkers als voorspellers van nefropathieIn deel 1 van dit proefschrift werd de voorspellende waarde van nieuwe biomarkerpanels voor progressie van nefropathie bij patiënten met diabetes mellitus type 2 onderzocht. Nieuwe biomarkers kunnen een beter inzicht geven in de pathofysiologie van diabetische nefropathie. Daarnaast kan identificatie van biomarkers als proxy voor moleculaire processen die geassocieerd zijn met progressie bijdragen aan de identificatie van nieuwe therapeutische aangrijpingspunten. Gezien de complexiteit van het grote aantal pathofysiologische processen die betrokken zijn bij de progressie van diabetes mellitus type 2 naast de intra-individuele variabiliteit van de biomarkers is het maar de vraag of een enkele biomarker voldoende diagnostische en voorspellende waarde kan hebben.

Daarentegen kan een panel van duidelijk gedefinieerde biomarkers een robuuster en meer reproduceerbaar alternatief vormen, omdat het panel variatie in individuele biomarkers kan tolereren zonder de diagnostische kracht in gevaar te brengen. Een combinatie van biomarkers, die verschillende processen van nierschade weerspiegelen, kan een meer realistisch beeld geven van de actuele pathofysiologische status van een patiënt en daarmee een betere prognostische inschatting geven.

Hoofdstuk 2 toonde een observationele cohortstudie van 82 patiënten met diabetes mellitus type 2 die gedurende vier jaar werden gevolgd. Een panel van dertien nieuwe biomarkers was geassocieerd met versnelde achteruitgang van de nierfunctie, bovenop traditionele risicomarkers. Deze studie identificeerde een nieuw panel biomarkers die verschillende processen van nierschade representeren, waaronder inflammatie, fibrosering, angiogenese en endotheelfunctie. Dit gecombineerde panel had een betere voorspellende waarde van versneld nierfunctieverlies in patiënten met diabetes mellitus type 2 bovenop traditionele risicomarkers. Er vond echter geen externe validatie van dit biomarkerpanel plaats en de kleine studiegroep beperkt de resultaten van deze studie. De resultaten moeten daarom in een groot, prospectief cohort gevalideerd worden om de toepasbaarheid van dit nieuwe panel biomarkers in een brede populatie patiënten met diabetes mellitus type 2 vast te stellen.

Het meten van peptiden en metabolieten, ook bekend als proteomics en metabolomics, zijn sterke methoden gebleken om biomarkers te ontdekken [3]. Hoofdstuk 3 onderzocht de voorspellende waarde van plasma proteomics classifier om de risico-inschatting te verbeteren op verandering in albuminurie-stadium bij patiënten met hypertensie of diabetes mellitus type 2. De ontwikkelde plasma proteomics classifiers waren in staat om verandering in albuminurie-stadium te voorspellen in patiënten met hypertensie, daarnaast waren zij in staat de overgang van micro- naar macroalbuminurie te voorspellen in patiënten met diabetes mellitus type 2. Deze resultaten zijn onafhankelijk van zowel traditionele risicomarkers voor albuminurie en eGFR, als het gebruik van RAAS-medicatie. De plasma-peptiden die in hypertensie en diabetes mellitus type 2 werden geïdentificeerd, zijn gerelateerd aan processen die betrokken zijn bij bekende nefropathofysiologische processen, waaronder fibrosering, inflammatie, angiogenese en mineraalhuishouding. Deze resultaten ondersteunen het groeiende bewijs om peptidoomplatforms te gebruiken voor risicostratificatie in nefropathie.

Metabolomics is een andere potentiële methode om nieuwe biomarkers voor nefropathie the ontdekken. Hoofdstuk 4 onderzocht de voorspellende waarde van urine- en plasmametabolieten voor het ontwikkelen van diabetische nefropathie in patiënten met diabetes mellitus type 2. In deze exploratieve studie bleken de plasmametabolieten histidine en butenoylcarnatine en de urinemetabolieten hoxoses, glutamine en tyrosine in staat om progressie van micro- naar macroalbuminurie te voorspellen bovenop traditionele risicomarkers in diabetes mellitus type 2. Deze metabolieten voorspelden geen albuminurieprogressie in patiënten met hypertensie, hetgeen een specifiek metabolietenprofiel suggereert voor diabetes mellitus type 2. In deze studie had het urine-

169168


metabolietenprofiel betere testeigenschappen dan het plasmaprofiel, ook verbeterde het toevoegen van de plasmametabolieten aan de urinemetabolieten de voorspellende eigenschappen niet. Dit zou erop kunnen wijzen dat urine metabolomics een betere klinische benadering is om patiënten met diabetes mellitus type 2 die risico lopen op progressieve nefropathie te identificeren (mede vanwege de praktische voordelen van het verzamelen van urine vergeleken met het verkrijgen van een bloedmonster). Desalniettemin zouden de uitkomsten van de plasmametabolieten behulpzaam kunnen zijn in het identificeren van onderliggende ziektemechanismen en –processen, zij zouden daarom niet veronachtzaamd moeten worden. Wederom is externe validatie van deze metabolieten nodig om stevige conclusies te kunnen trekken aangaande de voorspellende waarde van deze metabolieten in een bredere populatie.

Nieuwe biomarkers als voorspeller voor respons op therapieDe therapie van eerste keuze volgens huidige richtlijnen voor de behandeling van hypertensie en albuminurie bij patiënten met diabetes mellitus type 2 bestaat uit angiotensine-converterend-enzym-remmers (ACEi) of angiotensine II-receptor blokkerende middelen (ARB) [4]. Veel patiënten reageren echter suboptimaal op ACEi of ARB [5-9]. Dit kan leiden tot progressief nierfunctieverlies. Een methode om deze variatie in therapierespons te verminderen is het gebruik van nieuwe biomarkers om de respons op therapie te voorspellen. In hoofdstuk 5 werd een serum metabolomics classifier ontdekt die de albuminurierespons op ARB significant beter voorspelde in patiënten met diabetes mellitus type 2 dan traditionele risicofactoren. Deze classifier werd in een extern cohort gevalideerd. De metabolieten die in deze classifier werden geïncludeerd, waren gerelateerd aan oxidatieve stress, inflammatie en fibrosering. Met name NOS3 activiteit lijkt een voorspellende factor te zijn voor afname van albuminurie als respons op ARB. Deze bevindingen ondersteunen het gebruik van serum metabolomics als middel om de antiproteïnurische behandeling met ARB te optimaliseren. Metabolieten die in de classifier werden geïncludeerd worden toegeschreven aan processen die betrokken zijn bij stress/inflammatie en aan de downstream-effecten van fibrosering en remodellering van extracellulaire matrix. Daarnaast lijkt NOS3 een specifieke factor te zijn die relevant is voor de respons op ARB. Deze resultaten duiden erop, dat zowel ziekteprogressie als specifieke medicatie-afhankelijke moleculaire effecten meegewogen moeten worden om respons op medicatie in te schatten. De resultaten van deze metabolomics-studie ondersteunen het groeiende bewijs voor het gebruik van -omics als een strategie om de behandeling van nefropathie in diabetes mellitus te verbeteren. Het complementaire gebruik van –omics-platforms kan geïndividualiseerde behandeling een stap dichterbij de klinische praktijk brengen.

Nieuwe biomarkers om behandeleffecten te monitorenNieuwe biomarkerpanels kunnen gebruikt worden om het effect van behandeling te monitoren. Hoofdstuk 6 onderzocht het effect van atrasentan op een eerder vastgesteld metabolietenprofiel van diabetische nefropathie [10]. Deze studie toonde eerst aan dat de concentratie van urinemetabolieten uit dit panel significant geassocieerd waren met de eGFR van patiënten met diabetes mellitus type 2 en nefropathie. Daarna werd onderzocht of atrasentan een gunstig effect had op dit urine-metabolietenprofiel. De metabolieten bleven stabiel na 12 weken behandeling met atrasentan, terwijl deze afnamen bij gebruik van placebo. Verandering in individuele metabolieten na 12 weken behandeling waren positief gecorreleerd met verandering in eGFR. Aangezien lagere concentraties van metabolieten een verminderde mitochondriële inhoud en nierfunctie weerspiegelen [10], is het aannemelijk dat behandeling met atrasentan verschillende aspecten van de mitochondriële functie stabiliseert. Harde uitkomsten op lange termijn zijn nodig om vast te stellen of deze gunstige korte-termijn effecten een verbetering van de nierfunctie voorspellen. De resultaten van vervolgonderzoeken kunnen inzicht geven in de nefroprotectieve werkingsmechanismen van atrasentan en zouden nieuwe biomarkers op kunnen leveren om de respons op therapie in patiënten met diabetes mellitus type 2 en diabetische nefropathie te bewaken.

TOEKOMSTPERSPECTIEF

Voordat een biomarker of een panel van biomarkers in de klinische praktijk gebruikt kan worden, moeten deze uitgebreid gevalideerd worden in grote studies, om precisie, reproduceerbaarheid, sensitiviteit en specificiteit vast te stellen. De vertaling van een biomarker of combinatie van biomarkers van ontdekking tot klinische praktijk is een weg vol valkuilen en beperkingen. In 2009 hebben Hlatky et al. in samenwerking met de American Heart Association [11] randvoorwaarden voorgesteld voor de ontwikkeling van biomarkers, zoals is weergegeven in Box 1. Helaas stoppen veel studies naar nieuwe biomarkerpanels voor nefropathie bij diabetes mellitus type 2 aan het einde van de derde fase [12] en worden de panels niet in een extern cohort gevalideerd, noch wordt het klinisch nut vastgesteld. Er moet meer bewustwording komen voor en investering in studies in de fase van klinische uitkomsten, zodat nieuwe biomarkerpanels in de klinische praktijk geïmplementeerd kunnen worden. Een betere studieopzet, die de praktische waarden van biomarkers kan testen, en die de voorspellende waarde van de biomarkerpanel kan vertalen naar klinische beslissingen, kan bijdragen aan de achterblijvende acceptatie van nieuwe biomarkerpanels door zorgprofessionals. Verder zou samenwerking tussen universitaire en farmaceutische onderzoeksgroepen een manier kunnen zijn om expertise in verschillende deelgebieden met elkaar te kunnen delen, om zo effectieve verspreiding van onderzoeksresultaten te bevorderen en de implementatie van de bevindingen in de klinische praktijk te ondersteunen.

171170


Box 1. Randvoorwaarden voor de ontwikkeling van nieuwe biomarkers. Naar Hlatky et al. Circulation 2009 [11]

1. Proof of conceptverschillen de concentraties van nieuwe biomarkers tussen studiedeelnemers met en zonder de uitkomstmaat?

2. Prospectieve validatievoorspellen de nieuwe biomarkers de ontwikkeling van de uitkomstmaat in een prospectief cohort, of in een geneste case-cohort/case-cohort studie?

3. Toegevoegde waardevoegt de nieuwe biomarker voorspellende waarde toe aan gevestigde, standaard risicomarkers?

4. Klinisch nutverandert de nieuwe biomarker het voorspelde risico voldoende om het therapeutisch beleid aan te passen?

5. Klinische uitkomstenverbetert het gebruik van de nieuwe biomarker klinische uitkomsten, met name in een gerandomiseerde klinische trial?

6. Kosteneffectiviteit verbetert het gebruik van de nieuwe biomarker de klinische uitkomsten voldoende om de extra kosten van testen en behandelingen te rechtvaardigen?

Een groot deel van dit proefschrift bestaat uit onderzoek dat uitgevoerd is binnen het SysKid-programma (Systems Biology toward Novel Chronic Kidney Disease Diagnosis and Treatment, www.syskid.eu). SysKid was een grootschalig Europees onderzoeksproject waarin zowel universiteiten als de farmaceutische industrie deelnamen. SysKid beoogde de onderliggende moleculaire mechanismen van diabetische nefropathie op het kruispunt met de moleculaire impact van individuele medicijnen beter te karakteriseren om zodoende individuele behandeling van patiënten te optimaliseren. Resultaten van eerder en huidig onderzoek van SysKid hebben ons begrip van chronische nierziekte vergroot, met name in het kader van preventie, nieuwe diagnostische strategieën en behandelopties voor diabetische en hypertensieve nefropathie. Momenteel vindt er een validatiestudie plaats in een groot cohort patiënten met diabetes mellitus type 2, waarin de validiteit getest wordt van de nieuwe biomarkerpanels die in de hoofdstukken 2-4 van dit proefschrift beschreven zijn. Hoewel deze validatiestudie geen nieuwe kennis in de zin van klinische toepasbaarheid of kosteneffectiviteit zal voortbrengen, geeft deze studie wel de mogelijkheid om de voorspellende waarde van deze nieuwe biomarkerpanels voor harde nefrologische uitkomsten vast te stellen. Verder geeft deze validatiestudie de mogelijkheid om de biomarkers in vroege, intermediaire en late stadia van diabetische nefropathie te vergelijken.

Het gebruik van biomarkerpanels voor risicostratificatie in de klinische praktijk is een manier om behandelbeslissingen te sturen en interventies toe te spitsen op patiënten met het hoogste risico op progressieve nierziekte. Binnen SysKid werd een nieuw biomarkerpanel, bekend als CKD273, gevalideerd [13]. Deze urine proteomics classifier, gebaseerd op CE-MS, was ontwikkeld in een cross-sectionele studie in patiëntengroepen met chronische nierziekte met verschillende onderliggende etiologie [14]. Het klinische nut

van de CKD273 score wordt momenteel onderzocht in de Proteomic Prediction and Renin Angiotensin Aldosterone System Inhibition Prevention Of Early Diabetic nephRopathy In TYpe 2 Diabetic Patients With Normoalbuminuria (PRIORITY) trial (Clinical Trials Identifier NCT02040441) [15]. De PRIORITY-trial zal het klinische nut van de CKD273 score onderzoeken in het sturen van de behandeling van patiënten met diabetes mellitus type 2 met een hoog risico op progressieve nefropathie. De PRIORITY-trial gebruikt de CKD273-score om patiënten met een hoog risico op nefropathie te identificeren terwijl zij nog in de vroege fase van de ziekte verkeren. Het primaire doel van de PRIORITY-trial is om vast te stellen of urine proteomics de ontwikkeling van microalbuminurie kan voorspellen. De PRIORITY-trial zal ook onderzoeken of hoog-risico-patiënten baat hebben bij behandeling met spironolacton.

De huidige behandeling van diabetes mellitus type 2 is gestoeld op optimale regulatie van serum glucose, lipiden en bloeddruk. In het verleden hebben gerandomiseerde studies aangetoond dat adequate controle van HbA1c, lipiden en bloeddruk het ontstaan en de progressie van diabetische nefropathie vertragen [16-20]. De absolute reductie van het risico op nefropathie is echter laag, en dit impliceert dat het huidige behandelregime in een substantieel aantal patiënten onvoldoende is om progressie tot eindstadium-nierfalen te voorkomen. Gezien de grote heterogeniteit in pathofysiologie van diabetische nefropathie en de grote variabiliteit in de respons op nefroprotectieve medicatie is het noodzakelijk een nieuwe richting te zoeken voor de behandelstrategieën. Het gebruik van nieuwe biomarkerpanels om patiënten te identificeren die een gunstige respons op therapie zullen vertonen voordat de therapie wordt begonnen, kan de variabiliteit in behandelrespons verminderen. In een ideaal scenario voor de implementatie van nieuwe biomarkers in de klinische praktijk worden patiënten geïdentificeerd die zowel een hoog risico lopen op ziekteprogressie, als waarschijnlijk een gunstige respons op behandeling zullen vertonen. Er is momenteel echter gebrek aan studies die onderzoeken of nieuwe biomarkers de respons op therapie kunnen voorspellen. In Hoofdstuk 5 van dit proefschrift wordt slechts een begin gemaakt met het gebruik van nieuwe biomarkers om behandelrespons te voorspellen. Nader onderzoek is nodig om die veld verder te ontwikkelen.

Er zijn geen studies waarin onderzocht werd of medicatie-geïnduceerde verandering in nieuwe biomarkers geassocieerd is met risico op nierschade. Dit is een interessant onderzoeksgebied voor de toekomst, aangezien op deze manier nieuwe biomarkers geïdentificeerd kunnen worden die ons in staat stellen om effectiviteit van medicatie te monitoren, en er meer informatie beschikbaar komt over de moleculaire mechanismen waardoor de medicatie hun effect bewerkstelligen en nieuwe medicamenteuze aangrijpingspunten ontdekt kunnen worden. Wellicht heeft een patiënt baat bij een medicament, terwijl er geen verandering optreedt in albuminurie. Daarom zijn andere biomarkers nodig om effectiviteit en veiligheid van medicijnen te monitoren. Hoofdstuk 6 van dit proefschrift voorziet in proof of concept voor het gebruik van nieuwe biomarkers om effectiviteit van medicatie te monitoren, hoewel onderzoek naar harde lange-termijn-

173172


eindpunten nodig zijn om de bevindingen van ons onderzoek te ondersteunen en vast te stellen of de korte-termijnveranderingen in de metabolieten een verbetering in nierfunctie op lange termijn voorspellen.

Nieuwe behandelmethoden zijn nodig om ons een stap verder te brengen in het verbeteren van ziekte-uitkomsten. Nu geavanceerde laboratoriumtechnieken gerealiseerd worden in de klinische praktijk wordt het gebruik van -omics technieken als manier om de ziektelast van de complicaties van nierziekten te verminderen een reële strategie om diagnostiek, prognose en behandeling van nefropathie te verbeteren. In een ideaal scenario om het gebruik van nieuwe biomarkers in de klinische praktijk te implementeren worden patiënten geïdentificeerd die zowel een hoog risico lopen op ziekteprogressie, als een gunstige respons op behandeling laten zien. Grote –omics-datasets hebben de moleculaire fenotypering en karakterisering van de onderliggende moleculaire processen van diabetische nefropathie in individuele patiënten verbeterd. Deze datasets zouden ook in de toekomst gebruikt moeten worden om mechanismen van medicatierespons vast te stellen en om nieuwe biomarkers te ontdekken die de respons op medicatie voorspellen. De –omics-methodologie zou samen met het in kaart brengen van moleculaire processen en vandaaruit het ontwikkelen van nieuwe biomarkers bij kunnen dragen aan het identificeren van de optimale therapeutische benadering om de maximale therapeutische winst te verenigen met minimale bijwerkingen en off-targeteffecten (Figuur 1). Deze benadering zou de huidige manier van trial-and-error in de keuze voor medicatie voor patiënten met diabetes mellitus type 2 kunnen verbeteren en zou geïndividualiseerde therapie een stap dichter bij de klinische praktijk kunnen brengen.

Figuur 1. De ontwikkeling van een systems medicine approach om de variatie in respons op medicatie te onderzoeken, door het toevoegen van post-genomic informatie aan het genotype, waardoor het complexe genotype verbonden wordt aan het fenotype.

REFERENTIES

1. Fioretto P, Stehouwer CD, Mauer M, Chiesura-Corona M, Brocco E, Carraro A, Bortoloso

E, van Hinsbergh VW, Crepaldi G, Nosadini R. Heterogeneous nature of microalbuminuria

in NIDDM: studies of endothelial function and renal structure. Diabetologia. 1998

Feb;41(2):233-6.




Aug;30 Suppl 4:iv86-iv95.

3. Komorowsky CV, Brosius FC,3rd, Pennathur S, Kretzler M. Perspectives on systems biology




Int Suppl. 2013; 3:1-150.

















2008 Mar;73(5):630-6.






175174


11. Hlatky MA, Greenland P, Arnett DK, Ballantyne CM, Criqui MH, Elkind MS, Go AS, Harrell

FE Jr, Hong Y, Howard BV, Howard VJ, Hsue PY, Kramer CM, McConnell JP, Normand

SL, O’Donnell CJ, Smith SC Jr, Wilson PW; American Heart Association Expert Panel on

Subclinical Atherosclerotic Diseases and Emerging Risk Factors and the Stroke Council.

Criteria for evaluation of novel markers of cardiovascular risk: a scientific statement from

the American Heart Association. Circulation. 2009 May 5;119(17):2408-16.

12. Schutte E, Gansevoort RT, Benner J, Lutgers HL, Lambers Heerspink HJ. Will the future

lie in multitude? A critical appraisal of biomarker panel studies on prediction of diabetic

kidney disease progression. Nephrol Dial Transplant. 2015 Aug;30 Suppl 4:iv96-iv104.




2013 Feb;56(2):259-67.









15. Siwy J, Schanstra JP, Argiles A, Bakker SJ, Beige J, Boucek P, Brand K, Delles C, Duranton

F, Fernandez-Fernandez B, Jankowski ML, Al Khatib M, Kunt T, Lajer M, Lichtinghagen

R, Lindhardt M, Maahs DM, Mischak H, Mullen W, Navis G, Noutsou M, Ortiz A, Persson

F, Petrie JR, Roob JM, Rossing P, Ruggenenti P, Rychlik I, Serra AL, Snell-Bergeon J,

Spasovski G, Stojceva-Taneva O, Trillini M, von der Leyen H, Winklhofer-Roob BM,

Zürbig P, Jankowski J. Multicentre prospective validation of a urinary peptidome-based

classifier for the diagnosis of type 2 diabetic nephropathy. Nephrol Dial Transplant. 2014

Aug;29(8):1563-70.

16. UK Prospective Diabetes Study (UKPDS) Group. Intensive blood-glucose control with

sulphonylureas or insulin compared with conventional treatment and risk of complications

in patients with type 2 diabetes (UKPDS 33). Lancet. 1998 Sep 12;352(9131):837-53.

17. Parving HH, Lehnert H, Bröchner-Mortensen J, Gomis R, Andersen S, Arner P; Irbesartan

in Patients with Type 2 Diabetes and Microalbuminuria Study Group. The effect of

irbesartan on the development of diabetic nephropathy in patients with type 2 diabetes. N

Engl J Med. 2001 Sep 20;345(12):870-8.




Engl J Med. 2001 Sep 20;345(12):861-9.




2001 Sep 20;345(12):851-60.

20. Colhoun HM, Betteridge DJ, Durrington PN, Hitman GA, Neil HA, Livingstone SJ,

Thomason MJ, Mackness MI, Charlton-Menys V, Fuller JH; CARDS investigators.

Primary prevention of cardiovascular disease with atorvastatin in type 2 diabetes in the

Collaborative Atorvastatin Diabetes Study (CARDS): multicentre randomised placebo-

controlled trial. Lancet. 2004 Aug 21-27;364(9435):685-96.

Acknowledgements

179178

Acknowledgements

A PhD thesis is quite an endeavor, and I could not have done it without the guidance and support of many people.

I would like to sincerely thank my promotors, Dick de Zeeuw and Hiddo Lambers Heerspink, for your teaching and supervision during my PhD project. I am forever indebted to you for giving me a chance to pursue my PhD, even if during my interview, I had “no clue” as to what you needed for this project. I have learned a lot from you, and it has been a great pleasure to work with you.

Dick, your leadership has pushed me further than I imagined for myself. Thank you for helping me see the big picture.

Hiddo, as my daily supervisor, I thank you for your constant availability during this project. I am grateful for you always understanding the demands of my personal life.

I would like to express my gratitude and thanks to all the collaborators within the SysKid consortium. A special thank you to Bernd Mayer, Gert Mayer, Rainer Oberbauer, Peter Rossing, Maria Lajer, Georg Heinze, Paul Perco, Andreas Heinzel, Joachim Jankowski, Vera Jankowski, Harald Mishak, Wolfgang Woloszczuk, Jacqueline Benner, Guido Dallmann, Torben Friedrich, Stephan Bakker, Merel Hellemons. Your collaborations on the SysKid project laid the foundation to make my thesis possible. I have learned a lot from all of you, and I am forever grateful.

To the reading committee assessing this thesis, Prof. B.H.R. Wolffenbuttel, Prof. M. Kretzler, and Prof. G.J. Mayer, your efforts are highly appreciated.

To my many co-authors, thank you for your collaborations, expertise, input, and previous studies that contributed to the articles in this thesis. Stephan Bakker, Ron Gansevoort, Bernd Mayer, Andreas Heinzel, Georg Heinze, Peter Rossing, thank you for your many contributions to this thesis.

I would like to show appreciation to the numerous patients who participate in research studies. Your willingness to join studies makes our research possible.

Thank you to all my colleagues in the Department of Clinical Pharmacy and Pharmacology for a pleasant and enjoyable working environment. Alexandra, Ardy, and Marja, thank you for your constant help with all the little matters. Wessel, thank you for your forever patience with my many database and computer questions. Jan, thank you for your help during my time “away from the computer” and for the integral work you do in the laboratory.

To the supervisors of the gang. Thank you for your guidance and feedback during our weekly gangoverleg meetings.

Acknowledgements

To the postdocs, AIOs, and students members of the gang: Thank you for your help, feedback, discussion in and out of the office, and for your collegiality. A special thanks to my roommates for the enjoyable discussions, to Tobias for all your methodological and SAS help, and to Sigrid for the shopping trips, cooking, crafting, and gardening fun, and most of all, your friendship.

Sara, I appreciate your help during the beginning of my PhD project. Your energy and enthusiasm helped me adjust to working in academia. I am grateful for the friendship that grew out of working together. Thank you for being my paranymph.

Giedrė, from our first meeting at the course in Rotterdam, to the many dance parties with the girls, cooking marathons, and past (and future) travel adventures, I am thankful for your friendship. Thank you for being my paranymph.

Apryl, Anna, Becky, Lauren, Raquel, Kristin, Dru, Alisa, thank you for your enduring friendship throughout many countries and many years.

Girstė, Elena, Giulia, Lea, Kate, your friendship has made Groningen a brighter place. To my mommy friends, thank you for the play dates and socializing that help keep me grounded.

To my family, thank you for your constant support for all of my endeavors. Mom and Dad, you have always believed in me and have been the biggest supporters in my life, education, and career choices. I thank you for your guidance and love. Mike, Mindy, Brandon, Mikey, Annika, Joshua, Mason, thank you for always making our trips home so enjoyable. I wish we could spend more time together. Mam en Peter, bedankt voor alles wat jullie voor ons doen. Gerrie, dank je voor de gezelligheid.

Jeroen, the past seven years has been full… long distance dating, marriage, moving to the Netherlands, my career change, your residency and own PhD project, having kids… I wouldn’t have been able to accomplish any of this without you. Your support at home gave me the time to focus my energy and complete my thesis. Mostly, thank you for the two best things in my life. Anna and Eva, you give me so much joy.

181

Curriculum Vitae

CURRICULUM VITAE

Michelle Pena has had a diverse career working to improve health as a registered nurse, a community health educator in grassroots development, and a researcher with a governmental agency, prior to pursuing her PhD.

Michelle obtained an Associate of Science in Nursing and a Bachelor of Arts in Spanish from Pacific Union College (2000 and 2002, respectively), studied Public Health Nursing at California State University Dominguez Hills (2009), and obtained a dual Master of Public Health in epidemiology and Master of Art in Latin American Studies from San Diego State University (2012).

Michelle has eight years of work experience as a registered nurse in bedside patient care on medical/surgical and cardiovascular units. From 2003-2005, she worked with the U.S. Peace Corps, living in a rural village in Bolivia and working with community groups to improve nutrition, as a lack of electricity, poor road conditions, severe drought, and over-farming had impacted food security. This work led her to pursue her masters’ studies, and during that time, Michelle was a research assistant with the Office of Binational Border Health, California Department of Public Health, in San Diego (2008-2009). Her project focused on surveillance of occupational pesticide exposure of migrant farmworkers who regularly crossed the U.S.-Mexico border.

In 2011, Michelle moved to the Netherlands. Her desire to continue learning motivated her to begin a “second career” in clinical research. In 2012, Michelle started her PhD project on novel biomarker panels in diabetic kidney disease, at the Department of Clinical Pharmacy and Pharmacology, University Medical Center Groningen. In August 2015, Michelle started working as a postdoctoral researcher in the same department.

Documents

University of Groningen Novel biomarker panels in diabetic ... · There are however, few prospective studies of multiple biomarkers specifying type 2 diabetes as the cause of renal