Nephrol Dial Transplant (2015) 30: iv17–iv25 doi: 10.1093/ndt/gfv267

Full Review

Molecular presentation in diabetic nephropathy Downloaded from https://academic.oup.com/ndt/article/30/suppl_4/iv17/2324968 by guest on 25 September 2021 Andreas Heinzel1, Irmgard Mühlberger1, Gil Stelzer2, Doron Lancet2, Rainer Oberbauer3, Maria Martin4 and Paul Perco1

1emergentec biodevelopment GmbH, Vienna, Austria, 2Weizmann Institute of Science, Rehovot, Israel, 3Medical University of Vienna, Vienna, Austria and 4EMBL-EBI, Wellcome Trust Genome Campus, Hinxton, UK

Correspondence and offprint requests to: Paul Perco; E-mail: [email protected]

ABSTRACT INTRODUCTION

Diabetic nephropathy, as the most prevalent chronic disease of Kidney disease is a common problem in patients with diabetes. the kidney, has also become the primary cause of end-stage Depending on the degree of GFR impairment and/or albumin- renal disease with the incidence of kidney disease in type 2 dia- uria, roughly 30% of subjects are affected. As with most betics continuously rising. As with most chronic , the diabetes-associated comorbidities, the pathophysiology is pathophysiology is multifactorial with a number of deregulated multifactorial and the molecular pathways involved in the ini- molecular processes contributing to disease manifestation and tiation and progression constitute a wide and complex as well as progression. Current therapy mainly involves interfering in the redundant network of regulators. Considerable success has ––aldosterone system using angiotensin-con- been achieved in uncovering certain aspects of molecular regu- verting enzyme inhibitors or angiotensin-receptor blockers. lation of these processes. The most prominent examples are the Better understanding of molecular processes deregulated in identification of the causal contribution of the renin–angioten- the early stages and progression of disease hold the key for de- sin–aldosterone system (RAAS) as well as the matrix metallo- velopment of novel therapeutics addressing this complex dis- proteinase network (MMP) [1, 2]. However, this gain in ease. With the advent of high-throughput omics technologies, knowledge did not lead to an equal success in the discovery researchers set out to systematically study the disease on a mo- of actual drugs that could prevent or slow down the pace of fi lecular level. Results of the rst omics studies were mainly fo- this progressive disease. The only exception is the invention cused on reporting the highest deregulated molecules between of the blockade of the RAAS at several different sites by using diseased and healthy subjects with recent attempts to integrate angiotensin-converting enzyme (ACE) inhibitors or angioten- fi ndings of multiple studies on the level of molecular pathways sin-receptor blockers (ARBs) [3–5]. and processes. In this review, we will outline key omics studies There are several valid explanations for this uncomfortable on the genome, transcriptome, proteome and metabolome level situation but probably the most striking cause lies in the multi- in the context of DN. We will also provide concepts on how to factorial and highly redundant nature of the molecular pathways fi integrate ndings of these individual studies (i) on the level of of this disease. The revolutions in information technology as well functional processes using the -ontology vocabulary, (ii) as the omics technologies, however, offer nowadays new oppor- on the level of molecular pathways and (iii) on the level of tunities to uncover these complex molecular signal cascades thus – molecular models constructed based on laying the ground for potentially uncovering diagnostic and protein interaction data. prognostic as well as therapeutic leads to tackle this highly preva- Keywords: biological pathways, diabetic nephropathy, lent clinical problem. This review highlights the most recent re- functional annotation, molecular disease modeling, omics search in this area and provides rational strategies of how to profiling proceed in the fight against diabetic nephropathy (DN). If at

© The Author 2015. Published by Oxford University Press iv17 on behalf of ERA-EDTA. All rights reserved. least some of these in-silico concepts of rational marker and drug understanding of the role of hyperglycaemia in the early pro- design will prove successful in clinical trials then a major step for- gression of glomerulosclerosis, a significant drawback was the ward will have been made towards an improved care of patients lack of annotation for more than a half of the identified mole- with diabetes. Needless to say, such concepts have the potential cules at that time. to be applied to other multifactorial entities and are by no means A few years later, the first oligonucleotide arrays were avail- limited to DN. In this manuscript, we will specifically outline key able to study transcriptional changes with an increased cover- omics studies conducted in the area of DN as well as analyse con- age of the full transcriptome, and new players as, for example, cepts on a functional level, on the level of molecular pathways members of the Wnt signalling pathway, were identified as and on the level of biological networks being applied to get a dee- contributors to the pathophysiology of DN [17]. per understanding of the pathophysiology of DN. A recurring conclusion in published DN transcriptomics studies is the importance of oxidative stress underlying fibrosis OMICS studies in diabetic nephropathy for DN progression. Morrison et al. [18] found an up-regulation of several thiol antioxidative in rat mesangial cells treated A number of omics studies have been conducted in the past Downloaded from https://academic.oup.com/ndt/article/30/suppl_4/iv17/2324968 by guest on 25 September 2021 couple of years with a focus on DN, as also indicated by the with high glucose concentrations, and Cheng et al. [19] identified MEDLINE query ‘diabetic nephropathies[majr] AND (gene increased concentrations of members of the system expression profiling[mh] OR microarray analysis OR proteo- under comparable experimental conditions in mice. mics[mh] OR metabolomics[mh] OR genome-wide associ- While in-vitro and animal studies mainly concentrated on the ation study[mh]) NOT review’ resulting in 151 publications. glomeruli, transcriptomics studies of microdissected human kid- Representative studies with a focus on disease pathophysiology ney biopsies from DN patients showed relevant changes in for each omics category were selected for this review, also com- gene-expression profiles also in the tubulointerstitial compart- plemented by the authors’ personal paper collections as well as ment, including several nuclear factor-kappaB targets as well as selected references of key articles. Supplementary Data, Table 1 genes linked to inflammation [20, 21] and angiogenesis [17]. holds the discussed omics studies with information on biologic- Comparisons of compartment-specific gene-expression changes al processes presented in the respective Results and Discussion identified common deregulated mechanisms in glomeruli and sections. The reader is also referred to the succeeding articles tubuli associated with proliferation [22] and Janus kinase within this special issue dedicated to the individual omics tracks signalling [23], pointing towards the role of hypertrophy and for an extended discussion of performed studies. fibrosis in DN. Meanwhile, transcriptomics studies on DN are no longer studies in DN. A significant number of Genome- confined to the protein-coding regions of the genome. Argryo- Wide Association Studies (GWAS) in the context of DN has poulos et al. [24] used miRNA arrays and could identify urinary concentrated on patients of African ancestry [6, 7], who have miRNA profiles that differ across the different stages of DN. a disproportionately high risk for developing type II diabetes studies in DN. FULL REVIEW mellitus [8]. Single-nucleotide polymorphisms in the MHY9 The majority of proteomics stud- gene are known to be highly associated with ESRD in this popu- ies in DN focus on the identification of biomarker panels. In lation [9, 10] and could also be linked to DN [6, 7]. More recently, particular, studies in the urine of DN patients resulted in a it was shown that MHY9 variants also mediate DN susceptibility number of potential markers showing altered concentration in European Americans [11]. Other SNPs discussed in the con- when compared with healthy controls like, e.g. the ribosomal text of DN are located in the FRMD3 gene [6, 12], and results ubiquitin fusion protein UBA52 [25] or alpha 1 antitrypsin, from GWAS based on samples from the multi-ethnic Family the latter also showing an up-regulation in diseased kidney Investigation of Nephropathy and Diabetes (FIND) study in- biopsies in areas of renal fibrosis [26]. clude SNPs near the genes CNDP1 [13], CACNB2, ARL5 and Jin et al. [27] reported on a diagnostic panel of three potential NEBL [14]. markers, namely alpha-1-antitrypsin, alpha-1-acid glycoprotein In search of genetic determinants explaining the lower inci- 1 and prostate stem cell antigen. Zuerbig et al. [29] showed that denceofDNindiabeticwomen,aGWASintheFinnish the previously identified prognostic classifier for chronic kidney Diabetic Nephropathy (FinnDiane) cohort identified variants disease consisting of 273 urinary peptides (CKD273) [28]was in a region between the two transcription factors SP3 (which capable of predicting development of nephropathy in a diabetes is known to interact with the ) and cohort before the onset of microalbuminuria. CDCA7 [15].Thereaderisreferredtothemanuscripton So far, proteomics studies on kidney in the context of ‘genome-wide studies to identify risk factors for kidney disease DN are rare and mainly conducted in animal models. Identified with a focus on patients with diabetes’ by Regele et al. within molecules are, e.g. PPARalpha which was identified as a hub this special NDT issue for a comprehensive overview on protein in the set of differentially expressed in diabetic identified SNPs in DN. animals [30] as well as a number of nephropathy-specificpro- teins including the phospholipids scramblase 3 and tropomy- Transcriptomics studies in DN. More than a decade ago, the osin 3 in the whole kidney tissue of STZ rats [31]. results of the first cDNA microarray study on DN, in a set of 81 genes differentially expressed in the kidneys of healthy and Metabolomics studies in DN. Since metabolomics is one of (STZ)-induced diabetic mice, were published by Wada et al. the latest advances in the field of omics, not least because of the [16]. Although these results contributed to a molecular technical challenges regarding the chemical diversity and the

iv18 A. Heinzel et al. wide range of concentrations of metabolites, the number of development’ (GO:0072006), and 12 child terms, including studies using non-targeted approaches for metabolomics signa- ‘renal vesicle morphogenesis’ (GO:0072077) and ‘mesonephric ture detection in the context of DN is small with blood (either nephron morphogenesis’ (GO:0061228) (Figure 2). serum or plasma) being the sample matrix of choice in most The GO annotation data set provided by the GO Consor- studies. tium is one of the most widely used resources in biomedical Identified molecular processes include amino acid and and biotechnological data analysis, assisting researchers in in- phospholipid metabolism [32] or purine metabolism [33]. terpreting, validating and forming hypotheses for their data Han et al. [34] found an association in concentration levels of [40]. The Renal GOA initiative allowed (i) generation of de- esterified and non-esterified fatty acids in plasma and different tailed manual GO annotation for renal systems and (ii) to de- stages of DN. velop and improve the terms in the to ensure Results from urine metabolomics studies in DN patients that the whole of renal biology is well represented. The data could elucidate different disease aspects than serum or plasma set is available through GO and provides consistent annotations studies, e.g. the role of mitochondrial metabolism [35], specific for renal-specific data sets enabling researchers to rapidly evalu- Downloaded from https://academic.oup.com/ndt/article/30/suppl_4/iv17/2324968 by guest on 25 September 2021 aspects of altered tryptophan metabolism that were detected in ate and interpret existing renal data and generate hypotheses to a subset of the FinnDiane cohort [36] or the contrary tendency guide future research with confidence. This initiative has asso- of citric acid cycle intermediates in urine from db/db mice when ciated 163 156 GO terms to 2810 distinct UniProtKB proteins compared with serum profiles [37]. from the prioritized renal-related list. Of these, 1025 prioritized Hirayama et al. [38] identified a signature of five metabolites proteins have been annotated using GO terms. Additionally, in serum capable of discriminating diabetic patients with over 600 new renal-specific GO terms have been created. and without macroalbuminuria, including γ-butyrobetaine, We annotated the DN omics studies listed in the previous symmetric dimethylarginine, azelaic acid and two unknown chapters (Supplementary Data, Table 1) with GO biological metabolites. process terms reflecting the major outcomes of the particular Methods and tools for integrating findings from individual work. Specific terms were assigned whenever (i) the process it- studies will be discussed in the next chapters contributing to a self or (ii) key molecular features of a process were the main better understanding of molecular pathophysiology in DN, subject of the Results or Discussion section in the respective art- mainly focusing on functional annotation of molecular fea- icle. Results in the context of ‘extracellular matrix organization’ tures, identification of deregulated molecular pathways as well were discussed in 14 omics studies followed by the GO term REVIEW FULL as identifying deregulated molecular processes on the level of ‘immune response’ which was discussed in 10 studies. Among constructed biological networks (Figure 1). the top-ranked terms based on paper counts were also ‘response to oxidative stress’, ‘angiogenesis’ or ‘apoptosis’. A graphical re- Gene ontology and the Renal Gene Ontology Annotation presentation within the GO hierarchy of those GO terms, which Initiative were discussed in at least three of the omics studies, is depicted The Renal Gene Ontology Annotation (Renal GOA) Initia- in Figure 3A. tive [http://www.ebi.ac.uk/GOA/kidney] was established as an effort to summarize the accumulated experimentally based in- Pathway representation and superpathways formation in a comprehensive public resource of high-quality Next to studying omics result profiles on the level of GO functional annotation for proteins involved in renal develop- terms, protein to molecular pathway assignments as stored in ment, function and disease [39]. These annotations were molecular pathway can be used in order to identify based in the structured Gene Ontology (GO) [http:// deregulated mechanisms on a functional level. Prominent path- geneontology.org/] vocabulary for both (i) improving the de- way databases include KEGG [http://www.kegg.jp], Reactome scriptiveness of terms representing renal processes and (ii) in- [http://www.reactome.org], or Wiki pathways [http://www. creasing the number of associations of proteins involved in the wikipathways] with recent efforts trying to consolidate path- renal system to informative GO terms. The initiative was a ways from various sources, such as PathwayCommons [http:// 3-year project funded by Kidney Research UK and based in www.pathwaycommons.org/], NCBI’s Biosystems [http://www. the European Institute (EMBL-EBI), aiming ncbi.nlm.nih.gov/biosystems/], Human Pathway to provide a unique public resource of comprehensive high- (HPD) [http://discovery.informatics.iupui.edu/HPD/], Con- quality functional annotation for proteins involved in renal sensusPathDB [http://consensuspathdb.org/]orPathJam[http:// development, function and disease. www.pathjam.org/]. The GeneCards system currently holds infor- GO uses structured controlled vocabulary terms to describe mation on molecular pathways from 12 sources covering a total of three aspects of a gene product: the ‘molecular function(s)’,or 3215 biological pathways and 11 478 unique genes [41]. The four the activity(ies) that the sequence can directly perform; the ‘bio- largest pathway sources (KEGG, Reactome, Wiki pathways and logical process(es)’ or ‘the operation(s) it’ contributes to; and Qiagen) used in GeneCards collectively contain 10 770 genes the subcellular locations (‘cellular components’) in which it is with 1413 being shared between all four sources. located. GO terms are organized into a graph, where each Within GeneCards, the main pathway clustering algorithm term is linked to one or more general ‘parent’ terms and one is based on the Jaccard similarity coefficient (J), which scores or more specific ‘child’ terms if applicable. For example, the the degree of gene sharing for two pathways taking into account GO term ‘nephron morphogenesis’ (GO:0072028) has two par- their combined pathway size. In the clustering approach, two ent terms, ‘kidney morphogenesis’ (GO:0060993) and ‘nephron separate algorithms are combined. One is keeping edges

Molecular DN presentation iv19 Downloaded from https://academic.oup.com/ndt/article/30/suppl_4/iv17/2324968 by guest on 25 September 2021 FULL REVIEW

FIGURE 1: Schematic analysis workflow for delineating molecular DN disease presentations. With the phenotype at the centre, data are gathered from the scientific literature as well as from different omics studies in order to generate a phenotype molecular feature set. Bioinformatics analyses on the level of GO term enrichment analysis, molecular pathways as well as via identification of process units leads to the identification of de- regulated molecular processes and pathways as well as the construction of molecular models for the given disease.

between each pathway and its best matching pathway (i.e. near- when the highest score of a pathway is 0.3), yet still keep highly est neighbour), thus allowing grouping together two pathways similar pathways in the same cluster when the best matching not necessarily having a high Jaccard score but are the best pathway is very similar (e.g. if two pathways have a Jaccard matching (this is specifically important to connect similar path- score of 0.99 and there is another pathway with a Jaccards ways from different databases where the gene sets used could be score of 0.95). This approach resulted in a set of 1073 unified very different and thus lower the Jaccard score). The second al- pathways, termed SuperPaths, that connect maximally inform- gorithm implemented in the approach is a hierarchical cluster- ative pathway-related annotations to every human gene [42]. By ing algorithm with a relatively high threshold (this is important judiciously supervising the resulting SuperPath size (i.e. the to connect highly similar sets of genes that mainly originate number of its contained pathways), a high measure of annota- from the same database where slightly different sets of genes tion specificity was preserved while minimizing redundancy. are annotated as a separate pathway). The reason for using a SuperPaths are available in the GeneCards pathway section combinationofthesealgorithmsisinorderto,ontheone enabling a clear depiction of the set of unified pathways for each hand, reduce the number of singletons which are determined gene [41]. A new GeneCards companion database, called Path- by the threshold used in the nearest neighbour algorithm (e.g. Cards [http://pathcards.genecards.org/], was constructed to

iv20 A. Heinzel et al. Downloaded from https://academic.oup.com/ndt/article/30/suppl_4/iv17/2324968 by guest on 25 September 2021 ULREVIEW FULL

FIGURE 2: Gene ontology (GO) excerpt. An excerpt of the gene ontology for the selected specific term nephron morphogenesis and all its more general parent terms and its relations up to the root term biological process on top of the hierarchy. present a web card for quick in-depth analysis of each human fold-changes in the context of DN are, e.g. ‘Cell adhesion_ECM SuperPath in addition to basic search capabilities in a pathway- remodeling’ or ‘Degradation of the extracellular matrix’ [43]. In centric orientation. This online database enables a pathway con- addition, various signalling SuperPaths were associated with nectivity view for each SuperPath as well as access to the gene lists deregulated molecular features such as ‘VEGF signalling path- of the SuperPath and of each of its constituent pathways. way’, ‘PEDF Induced Signalling’ and ‘Chemokine Signalling’, Metabolic pathways are composed of genes whose proteins thereby indicating the complex interplay of signals involved are mostly connected via compounds that serve as substrates or in processes leading to DN. products, whereas signalling pathways are characterized by pro- We previously extracted molecular pathways from scientific teins that directly interact and thereby transfer signals within articles focusing on DN [44]. A pathway map was constructed cells. This prominent difference can clearly be seen using the based on the set of 27 identified molecular KEGG and PAN- interaction network derived from STRING that is displayed THER pathways where edges between individual pathways in PathCards. Upon eludication of DN-associated SuperPaths, were based on shared molecular features (Figure 3B). compounds found in metabolic pathways might be used as remedy targets, whereas signalling pathways might elicit a Biological networks and molecular models of disease treatment approach for altered protein–protein interactions Next to evaluating proteins in the context of molecular path- (PPIs). The list of top-ranked SuperPaths based on a set of ways as listed in the previous section, researchers started to use human omics-derived molecular features showing the highest interaction data sets such as PPIs in order to delineate relations

Molecular DN presentation iv21 Downloaded from https://academic.oup.com/ndt/article/30/suppl_4/iv17/2324968 by guest on 25 September 2021 FULL REVIEW

FIGURE 3: DN presentation on the level of GO, molecular pathways as well as a molecular model. (A) Highlights the relevant GO biological processes which were mentioned as key results in at least three of the omics studies discussed in this manuscript. Dashed lines indicate that intermediate parent GO terms were omitted for clarity. Colour-coding from orange to red reflects the number of omics studies reporting on the specific GO biological process. (B) Depicts the set of 27 molecular pathways extracted based on a semi-automatic literature mining approach for pathways in the context of DN. Molecular pathways highlighted in red were also identified as relevant in one of the omics studies listed in the current manuscript. (C) Graphical representation of the DN molecular model constructed based on omics data sets as well as molecular features from the scientific literature taking into account interaction data from the omicsNET framework. The four largest process units are highlighted, annotated with the top four enriched molecular pathways. Already discussed pathways are written in grey, whereas pathways not previously identified in the DN GO process tree or pathway maps are given in orange.

on a molecular level thus being able to expand analyses to pro- A PPI depicts a physical interaction of different strength and teins currently not assigned to a dedicated pathway but being evidence between two proteins. Such interactions do not neces- linked to other molecules via an interaction data. sarily imply that there is any functional relationship between

iv22 A. Heinzel et al. the partners, but merely indicate that two proteins bind to each central roles in a PPI network of DN-associated genes. He other (e.g. direct physical interaction) or at least are found in et al. [52] were interested in glomerular-specific molecular pro- close proximity to each other (e.g. co-localization). The estab- cesses and constructed a PPI network in the glomerulus which lishment of technologies such as co-immunoprecipitation, they termed GlomNet in order to find relevant signalling net- yeast two-hybrid screens or affinity electrophoresis facilitating works and molecules of relevance in glomerular development high-throughput screening of PPIs made interaction data for and disease. Tang et al. [22] recently constructed a gene co- model as well as human readily available. These expression network based on transcriptomics data in order to data can be obtained from publicly accessible databases such identify key regulators of transcription both in tubuli and glom- as IntAct [http://www.ebi.ac.uk/intact/]orBioGRID[http:// eruli of diabetic nephropathy samples when compared with thebiogrid.org/]. samples from control subjects. Next to PPIs, there are other types of networks such as (i) ‘tran- Within the SysKid project, we constructed a molecular pro- scriptional regulatory networks’ specifically focusing on the regu- cess model of DN using the omicsNET dependency network as lation of DNA target gene transcription through transcription underlying biological network for multi-omics data integration, Downloaded from https://academic.oup.com/ndt/article/30/suppl_4/iv17/2324968 by guest on 25 September 2021 factors, (ii) ‘genetic interaction networks’, which are composed hypothesis generation, and biomarker selection [53]. In brief, from observations on simultaneous alteration of two (or more) molecular features reported as being differentially regulated genes, e.g. by causing an unexpected change in the re- in (i) genomics, (ii) transcriptomics, (iii) proteomics and (iv) sulting phenotype given the effect of each individual alteration metabolomics experiments on DN were retrieved. The set of alone, (iii) ‘metabolic networks’, which aim at capturing the com- omics-derived molecular features was complemented by a set plete set of metabolic reactions in an organism by recording the of genes extracted via an automatic literature search for diabetic interplay between enzymes, transporters and metabolites and nephropathy-associated genes. The set of in total 2466 features their involvement in reactions transforming substances and meta- was mapped on the aforementioned omicsNET biological net- bolites into specific substances, (iv) ‘co-expression networks’, work constructing a disease network by including direct edges which focus on relations between genes showing similar expres- between features of the input set and discarding molecular fea- sion patterns over a set of samples, or (v) ‘co-annotation net- tures not connected to any other features of the input list. The works’, which focus on co-mentioning of two proteins in the resulting DN-specific networks held 1921 molecular features scientific literature. For an extensive review on different types of with 28 748 interactions and served as input to the MCODE biological networks, the reader is referred to Vidal et al. [45]. algorithm for identifying clusters of tightly interconnected mo- REVIEW FULL Next to network visualization and analysis tools such as lecular features following the model forming procedure as out- Cytoscape [http://cytoscape.org/], Osprey [http://biodata. lined in Heinzel et al. [54]. Thirty-four MCODE clusters were mshri.on.ca/osprey/], of VisANT [http://visant.bu.edu/], algo- identified holding in total 688 proteins (Figure 3C). The net- rithms and methodologies have been developed in order to work molecular model was used in order to delineate biomarker expand the set of experimentally derived PPIs based on predic- candidates for the most relevant process units. The set of se- tions taking into account information on pathway assignment, lected biomarkers was shown to improve prediction of acceler- structural protein domain information or functional protein ated eGFR decline with a median follow-up time of 4 years in annotation among others [46, 47]. The STRING database inte- patients with type 2 diabetes on top of established clinical risk grates experimentally determined and predicted protein inter- factors [55]. A validation study in an independent cohort of actions to provide a comprehensive compendium of direct as about 2000 patients with type 2 diabetes to assess performance well as indirect protein interactions for more than 1000 organ- of the marker panel is currently ongoing. isms [http://string-db.org/]. Comparison of molecular models via feature overlap of key Biological network analysis has also been applied in the con- network components depicts another way of analysing text of DN as outlined in the examples given below. Using visual biological networks like in the study by Heinzel et al.[56]on network inspection methods as well as basic quantitative mea- the analysis of common mechanisms between chronic kidney sures, Kretzler et al. [48] could show in a network build from disease and cardiovascular disease. expression data on seven different chronic renal disease entities Biological network interference analysis is also a way to that the current clinical disease classification does not reflect delineate on-target as well as off-target drug effects as well as molecular similarities on a molecular level. Fechete et al. [49] supporting the identification of predictive biomarkers on demonstrated that while only minor direct feature overlap be- drug response as exemplified by Heinzel et al.[44] in a recent tween molecular features implicated in the pathophysiology of study where the analysis of model overlap between a molecular DN and those reported to be of clinical relevance existed, a mo- model of DN and a drug mechanism of action molecular model lecular feature landscape generated by consolidating those fea- for the group of ACE inhibitors was primary focus of the study. tures on an expanded pathway network reflected the clinical A set of seven biomarkers is in the end proposed serving as implications of DN. Agrawal et al. [50] used sub-networks proxies for the interference of the ACE-inhibitor mechanism identified in a network specifically constructed to cover type of action molecular model and the molecular model for diabetic 2 diabetes and related complications by combining PPIs with nephropathy. co-expression information to elaborate on the development of diabetes and associated complications such as DN. Grekas et al. Outlook and perspectives [51] identified functional network modules as well as hub and The wealth of publicly available omics data sets has opened bottleneck genes that due to their nature are likely to play new avenues of studying pathophysiological mechanisms in

Molecular DN presentation iv23 renal diseases and especially DN as the most prevalent form of 6. Freedman BI, Langefeld CD, Lu L et al. Differential effects of MYH9 and chronic renal diseases. Whereas the first omics studies focused APOL1 risk variants on FRMD3 Association with Diabetic ESRD in Afri- on the identification of the most differentially regulated indi- can Americans. PLoS Genet 2011; 7: e1002150 7. McDonough CW, Palmer ND, Hicks PJ et al. A genome-wide association vidual molecular entities, researchers have in the meantime study for diabetic nephropathy genes in African Americans. Kidney Int adopted this concept of integrated analysis in order to (i) iden- 2011; 79: 563–572 tify pathological processes in disease development and progres- 8. Brancati FL, Kao WH, Folsom AR et al. Incident type 2 diabetes mellitus in sion, (ii) derive novel biomarker candidates and drug targets as African American and white adults: the Atherosclerosis Risk in Communi- – well as (iii) study drug on-target and off-target effects. The need ties Study. JAMA 2000; 283: 2253 2259 9. Kao WHL, Klag MJ, Meoni LA et al. MYH9 is associated with nondiabetic for a better prediction of drug effects on a molecular level was end-stage renal disease in African Americans. Nat Genet 2008; 40: 1185–1192 recently outlined by Heerspink et al. [57] in order to increase 10. Bostrom MA, Lu L, Chou J et al. Candidate genes for non-diabetic ESRD in success regarding drug efficacy in clinical trials but also to African Americans: a genome-wide association study using pooled DNA. stop development of compounds which trigger adverse side Hum Genet 2010; 128: 195–204 et al effects down the road by interfering with molecules in the mo- 11. Cooke JN, Bostrom MA, Hicks PJ . Polymorphisms in MYH9 are asso- ciated with diabetic nephropathy in European Americans. Nephrol Dial Downloaded from https://academic.oup.com/ndt/article/30/suppl_4/iv17/2324968 by guest on 25 September 2021 lecular network off the prime target. We are currently just at the Transplant 2012; 27: 1505–1511 beginning of making full use of the available data and with im- 12. Pezzolesi MG, Poznik GD, Mychaleckyj JC et al. Genome-wide association proved methodologies for analysis and additional omics pro- scan for diabetic nephropathy susceptibility genes in type 1 diabetes. Dia- files on clinically well-characterized samples, success rates of betes 2009; 58: 1403–1410 clinical trials in drug development will hopefully again increase 13.IgoRP,IyengarSK,NicholasSBet al. Genomewide linkage scan for diabetic – in the near future. renal failure and albuminuria: the FIND study. Am J Nephrol 2011; 33: 381 389 14. Thameem F, Igo RP, Freedman BI et al. A genome-wide search for linkage of estimated glomerular filtration rate (eGFR) in the Family Investigation of Nephropathy and Diabetes (FIND). PLoS One 2013; 8: e81888 et al SUPPLEMENTARY DATA 15. Sandholm N, McKnight AJ, Salem RM . 2q31.1 associ- ates with ESRD in women with type 1 diabetes. J Am Soc Nephrol 2013; 24: 1537–1543 Supplementary data are available online at http://ndt.oxford 16. Wada J, Zhang H, Tsuchiyama Y et al. Gene expression profile in journals.org. streptozotocin-induced diabetic mice kidneys undergoing glomerulo- sclerosis. Kidney Int 2001; 59: 1363–1373 17. Cohen CD, Lindenmeyer MT, Eichinger F et al. Improved elucidation of biological processes linked to diabetic nephropathy by single probe-based ACKNOWLEDGEMENTS microarray data analysis. PLoS One 2008; 3: e2937 18. Morrison J, Knoll K, Hessner MJ et al. Effect of high glucose on gene ex- pression in mesangial cells: upregulation of the thiol pathway is an adapta- The research leading to these results has received funding – ’ tional response. Physiol Genomics 2004; 17: 271 282 from the European Community s Seventh Framework Pro- 19. Cheng DW, Jiang Y, Shalev A et al. An analysis of high glucose and gramme under grant agreement no. HEALTH-F2-2009-241544 glucosamine-induced gene expression and oxidative stress in renal mesan- FULL REVIEW (SysKid). gial cells. Arch Physiol Biochem 2006; 112: 189–218 20. Schmid H, Boucherot A, Yasuda Y et al. Modular activation of nuclear factor-kappaB transcriptional programs in human diabetic nephropathy. Diabetes 2006; 55: 2993–3003 CONFLICT OF INTEREST STATEMENT 21. Woroniecka KI, Park ASD, Mohtat D et al. Transcriptome analysis of human diabetic kidney disease. Diabetes 2011; 60: 2354–2369 et al Results presented in this paper have not been published previ- 22. Tang W, Gao Y, Li Y . Gene networks implicated in diabetic kidney dis- ease. Eur Rev Med Pharmacol Sci 2012; 16: 1967–1973 ously in whole or part, except in abstract format. A.H., I.M. and 23. Berthier CC, Zhang H, Schin M et al. Enhanced expression of Janus kinase- P.P. are employees of emergentec biodevelopment GmbH. signal transducer and activator of transcription pathway members in human diabetic nephropathy. Diabetes 2009; 58: 469–477 24. Argyropoulos C, Wang K, McClarty S et al. Urinary microRNA profiling in the nephropathy of type 1 diabetes. PLoS One 2013; 8: e54662 REFERENCES 25. Dihazi H, Müller GA, Lindner S et al. Characterization of diabetic nephro- pathy by urinary proteomic analysis: identification of a processed ubiquitin 1. Ichikawa I, Brenner BM. Glomerular actions of angiotensin II. Am J Med form as a differentially excreted protein in diabetic nephropathy patients. 1984; 76: 43–49 Clin Chem 2007; 53: 1636–1645 2. Afkarian M, Hirsch IB, Tuttle KR et al. Urinary excretion of RAS, BMP, and 26. Sharma K, Lee S, Han S et al. Two-dimensional fluorescence difference gel WNT pathway components in diabetic kidney disease. Physiol Rep 2014; 2: electrophoresis analysis of the urine proteome in human diabetic nephro- e12010 pathy. Proteomics 2005; 5: 2648–2655 3. Brenner BM, Cooper ME, de Zeeuw D et al. Effects of losartan on renal and 27. Jin J, Ku YH, Kim Y et al. Differential proteome profiling using iTRAQ in cardiovascular outcomes in patients with type 2 diabetes and nephropathy. microalbuminuric and normoalbuminuric type 2 diabetic patients. Exp N Engl J Med 2001; 345: 861–869 Diabetes Res 2012; 2012: 168602 4. Parving HH, Lehnert H, Bröchner-Mortensen J et al. The effect of irbesar- 28. Argilés Á, Siwy J, Duranton F et al. CKD273, a new proteomics classifier tan on the development of diabetic nephropathy in patients with type 2 dia- assessing CKD and its prognosis. PLoS One 2013; 8: e62837 betes. N Engl J Med 2001; 345: 870–878 29. Zürbig P, Jerums G, Hovind P et al. Urinary proteomics for early diagnosis 5. Wu H-Y, Huang J-W, Lin H-J et al. Comparative effectiveness of in diabetic nephropathy. Diabetes 2012; 61: 3304–3313 renin-angiotensin system blockers and other antihypertensive drugs in pa- 30. Tilton RG, Haidacher SJ, Lejeune WS et al. Diabetes-induced changes in the tients with diabetes: systematic review and bayesian network meta-analysis. renal cortical proteome assessed with two-dimensional gel electrophoresis BMJ 2013; 347: f6008 and mass spectrometry. Proteomics 2007; 7: 1729–1742

iv24 A. Heinzel et al. 31. Hwang H-J, Baek Y-M, Kim S-W et al. Differential expression of kidney 44. Heinzel A, Perco P, Mayer G et al. From molecular signatures to predictive proteins in streptozotocin-induced diabetic rats in response to hypogly- biomarkers: modeling disease pathophysiology and drug mechanism of ac- cemic fungal polysaccharides. J Microbiol Biotechnol 2007; 17: 2005–2017 tion. Front Cell Dev Biol 2014; 2: 37 32. Zhang J, Yan L, Chen W et al. Metabonomics research of diabetic nephro- 45. Vidal M, Cusick ME, Barabási A-L. Interactome networks and human dis- pathy and type 2 diabetes mellitus based on UPLC-oaTOF-MS system. Anal ease. Cell 2011; 144: 986–998 Chim Acta 2009; 650: 16–22 46. Bernthaler A, Mühlberger I, Fechete ="tag">R et al. A dependency graph approach 33. Xia J-F, Liang Q-L, Liang X-P et al. Ultraviolet and tandem mass spectrom- for the analysis of differential gene expression profiles. Mol Biosyst 2009; 5: etry for simultaneous quantification of 21 pivotal metabolites in plasma 1720–1731 from patients with diabetic nephropathy. J Chromatogr B Analyt Technol 47. Schmitt T, Ogris C, Sonnhammer ELL. FunCoup 3.0: database of genome- Biomed Life Sci 2009; 877: 1930–1936 wide functional coupling networks. Nucleic Acids Res 2014; 42: D380–D388 34. Han L-D, Xia J-F, Liang Q-L et al. Plasma esterified and non-esterified fatty 48. Bhavnani SK, Eichinger F, Martini S et al. Network analysis of genes regu- acids metabolic profiling using gas chromatography-mass spectrometry lated in renal diseases: implications for a molecular-based classification. and its application in the study of diabetic mellitus and diabetic nephropa- BMC Bioinformatics 2009; 10 (Suppl 9): S3 – thy. Anal Chim Acta 2011; 689: 85 91 49. Fechete R, Heinzel A, Perco P et al. Mapping of molecular pathways, bio- et al 35. Sharma K, Karl B, Mathew AV . Metabolomics reveals signature of markers and drug targets for diabetic nephropathy. Proteomics Clin Appl mitochondrial dysfunction in diabetic kidney disease. J Am Soc Nephrol 2011; 5: 354–366 Downloaded from https://academic.oup.com/ndt/article/30/suppl_4/iv17/2324968 by guest on 25 September 2021 – 2013; 24: 1901 1912 50. Sengupta U, Ukil S, Dimitrova N et al. Expression-based network biology et al 36. Van der Kloet FM, Tempels FWA, Ismail N . Discovery of early-stage identifies alteration in key regulatory pathways of type 2 diabetes and asso- biomarkers for diabetic kidney disease using ms-based metabolomics ciated risk/complications. PLoS One 2009; 4: e8100 – (FinnDiane study). Metabolomics 2012; 8: 109 119 51. Ntemka A, Iliadis F, Papanikolaou N et al. Network-centric analysis of gen- et al 37. Li M, Wang X, Aa J . GC/TOFMS analysis of metabolites in serum and etic predisposition in diabetic nephropathy. Hippokratia 2011; 15: 232–237 urine reveals metabolic perturbation of TCA cycle in db/db mice involved 52. He L, Sun Y, Takemoto M et al. The glomerular transcriptome and a predicted in diabetic nephropathy. Am J Physiol Renal Physiol 2013; 304: protein-protein interaction network. J Am Soc Nephrol 2008; 19: 260–268 F1317–F1324 53. Fechete R, Heinzel A, Söllner J et al. Using information content for expand- 38. Hirayama A, Nakashima E, Sugimoto M et al. Metabolic profiling reveals ing human protein coding gene interaction networks. J Comput Sci Syst new serum biomarkers for differentiating diabetic nephropathy. Anal Bioa- Biol 2013; 6: 073–082 nal Chem 2012; 404: 3101–3109 54. Heinzel A, Mühlberger I, Fechete R et al. Functional molecular units for 39. Alam-Faruque Y, Dimmer EC, Huntley RP et al. The renal gene ontology guiding biomarker panel design. Methods Mol Biol 2014; 1159: 109–133 annotation initiative. Organogenesis 2010; 6: 71–75 55. Pena M, Heinzel A, Heinze G et al. A novel panel of biomarkers to predict 40. RamachandraRao SP, Zhu Y, Ravasi T et al. Pirfenidone is renoprotective in eGFR decline in type 2 diabetes. PLoS One 2015; 10: e0120995 diabetic kidney disease. J Am Soc Nephrol 2009; 20: 1765–1775 56. Heinzel A, Fechete R, Mühlberger I et al. Molecular models of the cardior-

41. Safran M, Dalah I, Alexander J et al. GeneCards Version 3: the human gene REVIEW FULL enal syndrome. Electrophoresis 2013; 34: 1649–1656 integrator. Database (Oxford) 2010; 2010: baq020 57. Heerspink HJL, Grobbee DE, de Zeeuw D. A novel approach for establish- 42. Belinky F, Nativ N, Stelzer G et al. PathCards: multi-source consolidation of ing cardiovascular drug efficacy. Nat Rev Drug Discov 2014; 13: 942 human biological pathways. Database (Oxford) 2015; 2015: bav006 43. Stelzer G, Dalah I, Stein TI et al. In-silico human genomics with Gene- Cards. Hum Genomics 2011; 5: 709–717 Received for publication: 12.1.2015; Accepted in revised form: 6.6.2015

Molecular DN presentation iv25