Abstract

   Aging is a major driver for chronic kidney disease (CKD) and the
   counterbalancing of aging processes holds promise to positively impact
   disease development and progression.

   In this study we generated a signature of renal age-associated genes
   (RAAGs) based on six different data sources including transcriptomics
   data as well as data extracted from scientific literature and dedicated
   databases. Protein abundance in renal tissue of the 634 identified
   RAAGs was studied next to the analysis of affected molecular pathways.
   RAAG expression profiles were furthermore analysed in a cohort of 63
   CKD patients with available follow-up data to determine association
   with CKD progression. 23 RAAGs were identified showing concordant
   regulation in renal aging and CKD progression. This set was used as
   input to computationally screen for compounds with the potential of
   reversing the RAAG/CKD signature on the transcriptional level. Among
   the top-ranked drugs we identified atorvastatin, captopril, valsartan,
   and rosiglitazone, which are widely used in clinical practice for the
   treatment of patients with renal and cardiovascular diseases. Their
   positive impact on the RAAG/CKD signature could be validated in an
   in-vitro model of renal aging.

   In summary, we have (i) consolidated a set of RAAGs, (ii) determined a
   subset of RAAGs with concordant regulation in CKD progression, and
   (iii) identified a set of compounds capable of reversing the proposed
   RAAG/CKD signature.

   Keywords: Renal aging, Chronic kidney disease progression,
   Computational compound screening, Drug repurposing, Anti-aging

Graphical abstract

   [37]Unlabelled Image
   [38]Open in a new tab

1. Introduction

   Global life expectancy increased over the last 35 years from 62 to
   72 years, while in the same time period causes of death and the profile
   of diseases significantly changed [[39]1]. More individuals reach
   higher age and develop chronic diseases. Among them, kidney diseases
   are still a rising cause of disability adjusted life years and amid the
   thirty leading causes of death. Age in this context is a heterogeneous
   term that needs to be differentiated into chronological age defined as
   counting time in days, months, and years from birth, and biological age
   as a functional decline of cellular function leading to alterations of
   organs and eventually the whole organism [[40]2]. Various triggers have
   been identified that can cause or accelerate the process of aging.
   Scientific effort has been invested to understand these mechanisms and
   to find ways to decelerate the aging process [[41][3], [42][4],
   [43][5]].

   The process of renal aging leads to mass loss of renal cortex as well
   as to loss of function including vascular abnormalities, reduced renal
   plasma flow and a decline in glomerular filtration rate (GFR) with a
   mean loss of GFR of 0.75 mL/min/1.73 m^2 per year. Histologically,
   renal aging manifests in interstitial fibrosis, glomerulosclerosis,
   tubular atrophy and thickening of the intima of arteries [[44]6].

   Several causative mechanisms involved in renal aging have been
   catalogued by Halloran and Melk in 2001 for the first time [[45]7].
   Cellular senescence triggers many of these pathological changes and
   senescence as well as aging promote and influence each other [[46]8].
   Two major pathways were defined, leading to growth arrest and cellular
   senescence, in part interacting with each other. Shortening of
   telomeres can be stopped by cell cycle arrest to protect the end of
   chromosomes of erosion. Besides telomeres the cell cycle inhibitors p16
   (CDKN2A) and p21 (CDKN1A) have been found to play major roles in
   cellular senescence. While CDKN1A induction is directly related to
   telomere shortening via the p53 pathway, CDKN2A has been shown to
   induce cellular senescence independently of telomere attrition [[47]9].
   Functional kidney decline results in an accumulation of urea and other
   substances interfering with many cellular pathways including cellular
   senescence [[48]10].

   High-throughput Omics technologies coupled with bioinformatics analyses
   have paved the way for systematic studies of molecular mechanisms
   associated with age-related diseases. Blankenburg and colleagues for
   example recently consolidated over thirty age-related Omics datasets to
   construct a catalogue of common aging mechanisms with a particular
   focus on mechanisms and molecular features linking different
   age-related processes [[49]11]. Data integration approaches also lead
   to the identification of novel age-related genes. Jiang and colleagues
   used RNA sequencing to screen for age-related genes being associated
   with IgA nephropathy to search for novel therapeutic targets [[50]12].
   Researchers have also started making use of available drug mechanism of
   action datasets as well as 3D drug models to identify compounds with
   potential beneficial effects on molecules and processes being
   deregulated in aging [[51]13,[52]14].

   In this work we generated a comprehensive set of renal age-associated
   genes (RAAGs) via literature mining and Omics data analysis. This
   constructed set of RAAGs was mechanistically analysed on the level of
   molecular pathways. RAAGs being significantly associated with CKD
   progression on the transcriptional level served as input for in-silico
   compound screening. Compounds with a positive impact on aging and
   disease mechanisms by reversing RAAG expression changes were identified
   and are discussed in this work.

2. Material and Methods

2.1. Constructing the Renal Age-Associated Gene (RAAG) Set

   We mined scientific literature for RAAGs using the following PubMed
   query: “(aging OR ageing OR replicative senescence) AND (renal OR
   kidney)” in February 2018. Human genes were extracted from the
   resulting set of publications using NCBI's gene2pubmed mapping file
   [[53]15]. Gene-publication associations of all genes with at least two
   publications were manually curated to only include true positive
   age-related genes for further analysis.

   We furthermore downloaded genes associated with the Gene Ontology (GO)
   term “aging” from the GO webpage at [54]www.geneontology.org [[55]16]
   as well as from the GenAge database [[56]17]. We used data from the
   Human Protein Atlas on protein abundance and RNA expression in renal
   tissue to limit the gene sets extracted from GO and GenAge to those
   being expressed in kidney tissue [[57]18]. We specifically focused on
   proteins with medium or high abundance in either glomerular or tubular
   tissue. We in addition determined the median RNA expression values for
   all three RNA datasets provided in the Human Protein Atlas of these
   proteins with medium or high abundance in glomerulus or tubules. Genes
   showing an RNA expression value above the calculated median RNA
   expression were also considered as relevant and included in the set of
   RAAGs. Another source for RAAGs was the Digital Aging Database
   [[58]19]. RAAGs linked to kidney aging were downloaded using the
   respective search forms at the Digital Aging Database. The set of RAAGs
   extracted from scientific literature or age-related databases was
   complemented by including genes from transcriptomics studies. The gene
   sets published by Rodwell and colleagues as well as by Melk and
   colleagues were retrieved from the supplementary data files from the
   two published articles [[59]20,[60]21].

   The rentrez R package
   ([61]https://cran.r-project.org/web/packages/rentrez/index.html) was
   used to annotate transcripts in the supplementary table of Melk et al.
   via mapping the provided GenBank accession numbers to the respective
   NCBI UniGene cluster IDs and subsequently to the official Gene Symbols.
   The Ensembl Gene ID was used as common gene identifier and RAAGs were
   furthermore annotated with their official Gene Symbols.

2.2. Determining Protein Abundance in Renal Tissue

   Protein abundance levels in renal tissue were retrieved from the kidney
   expression summary section of the Human Protein Atlas for all RAAGs
   [[62]18]. Immunohistochemistry images have been manually annotated by
   two independent researchers and protein abundance levels are reported
   using one of the following four categories: “not detected”, “low”,
   “medium”, or “high”. The protein abundance levels are based on antibody
   staining intensity and fraction of stained cells. For kidney tissue,
   there is one protein abundance level for tubules and one for glomeruli
   provided in the Human Protein Atlas.

2.3. Investigating Associations with CKD Progression

   The expression of RAAGs was evaluated in a transcriptomics data set
   consisting of 63 CKD patients [[63]22]. Follow-up data were updated in
   Q2 2018 and patients were divided into a group with stable course of
   disease and a group with progressive course of disease. Progression was
   defined as either reaching end-stage renal disease or experiencing
   doubling of serum creatinine during the follow-up time with a minimum
   follow-up time of six months. Patients in the stable group did not
   develop ESRD nor doubling of serum creatinine. We further only kept
   patients with at least one year of follow-up in the stable group.

   The institutional Review Board of the Medical University of Innsbruck
   has accredited the use of surplus material from routine kidney biopsies
   and anonymized patient data for research purposes.

   Expression data are available in NCBI's Gene Expression Omnibus with
   the accession number [64]GSE60861
   ([65]https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=gse60861). The
   R limma package
   ([66]https://bioconductor.org/packages/release/bioc/html/limma.html)
   was used for preprocessing and statistical analysis. Expression data
   were background corrected, quantile normalized, and probes with the
   same sequence were summarized. Differentially expressed RAAGs between
   the group of stable and progressive patients were identified using the
   Statistical Analysis of Microarrays (SAM) method setting the false
   discovery rate (FDR) to <2.5% with a fold-change cutoff of > |1.25|.

2.4. Analysing the Functional Context of RAAGs

   Enriched molecular pathways were determined based on the set of RAAGs
   using the Database of Annotation, Visualization and Integrated
   Discovery (DAVID) v6.8 tool [[67]23]. The Kyoto Encyclopedia of Gene
   and Genomes (KEGG) pathway set was used as underlying pathway resource.
   Enriched pathways with p-values <.05 after Bonferroni adjustment for
   multiple testing were considered as relevant. Disease-specific pathways
   such as “Type 2 diabetes mellitus” or “Pathways in cancer” were
   excluded from further analyses.

2.5. Identifying Compounds Reversing CKD Associated RAAGs

   The library of integrated network-based cellular signatures (LINCS)
   L1000 data set was used to identify compounds reversing the set of
   RAAGs showing significant association with CKD outcome [[68]24]. The
   LINCS L1000 data set holds over a million gene expression profiles for
   around 20,000 compounds tested in different concentrations and cell
   lines to study drug mechanism of action on a molecular level. We
   restricted our input signature for the L1000 Characteristic Direction
   Signature Search Engine ([69]http://amp.pharm.mssm.edu/L1000CDS2/) to
   those RAAGs showing a concerted way of expression in renal aging and
   renal disease, i.e. being either up- or downregulated in both, renal
   aging and the group of progressive CKD patients. The 50 top-ranked
   compounds based on the drug score were further evaluated focusing in
   particular on the individual compound-gene combinations leading to high
   scores in the compound-RAAG signature interaction. The drug score is
   calculated based on the overlap of input RAAGs and the drug signature
   genes normalized to the effective input size defined as the number of
   intersecting genes between the input gene set and the set of L1000
   genes.

2.6. Validating the Impact of Identified Compounds in Human Renal Proximal
Tubular Cells

   Proximal tubular human kidney cells (HK2) were purchased from American
   Type Culture Collection (CRL-2190, Wesel, Germany) and cultured in
   Keratinocyte-Serum Free Medium (KSFM) containing 10% fetal bovine serum
   (FBS), 5 ng/ml recombinant epidermal growth factor (rEGF), 0.05 mg/ml
   bovine pituitary extract (BPE), 100 U/ml penicillin and 100 μg/ml
   streptomycin. Cell culture supplies were purchased from ThermoFisher
   Scientific, Vienna, Austria. All cells were grown at 37 °C in a
   humidified atmosphere with 5% CO[2]. After growth to confluence, cells
   were pre-treated with 0.5 μM H[2]O[2] for two hours to induce oxidative
   stress. Subsequently cells were stimulated with atorvastatin,
   captopril, rosiglitazone (10 μM each) or valsartan (1 μM). All
   chemicals used for stimulation experiments were purchased from
   Sigma-Aldrich, Vienna, Austria and solubilized in DMSO prior to use.
   Drug concentrations were chosen based on prior studies [[70][25],
   [71][26], [72][27], [73][28]]. After 24 h, RNA was isolated with RNeasy
   Mini Kit (Qiagen, Valencia, CA, USA) according to the manufacturer's
   protocol. RNA yield and quality were determined using a DS-11 FX+
   spectrophotometer (DeNovix, Wilmington, DE, USA).

   For qPCR, mRNA was reverse transcribed into cDNA with the High Capacity
   cDNA reverse Transcription kit. Samples were analysed with the
   following TaqMan® Gene Expression Assays: C3 (Hs00163811_m1), GAPDH
   (Hs99999905_m1), EGF (Hs01099999_m1), CD52 (Hs00174349_m1), CFB
   (Hs00156060_m1), LTF (Hs00914334_m1), MMP7 (Hs01042796_m1), TNFRSF11B
   (Hs00900358_m1). All materials for qPCR were purchased from
   ThermoFisher Scientific, Vienna, Austria.

   Reactions were prepared in duplicate for each sample (technical
   replicates) and analysed on the 7500 Fast Real-Time PCR System (Applied
   Biosystems) under the following conditions: 50 °C for 2 min, 95 °C for
   10 min followed by 40 cycles of 95 °C for 15 s and 60 °C for 1 min.
   Three biological replicates per condition were performed. The 2^-ddCt
   Method was used to calculate fold change values with respect to
   untreated control cells using GAPDH as housekeeper transcript [[74]29].
   Unpaired t-test assuming equal group variances was used to compare fold
   change values of H[2]O[2] treated samples with H[2]O[2] and
   drug-treated samples.

3. Results

3.1. The Renal Age-Associated Gene (RAAG) Set

   Six different data sources were used to generate the RAAG signature
   consisting of 634 unique genes ([75]Fig. 1 and [76]Table 1).

Fig. 1.

   [77]Fig. 1
   [78]Open in a new tab

   Generation of the renal age-associated gene (RAAG) set.

   Numbers of RAAGs extracted from the six different sources contributing
   to the final set of 634 unique RAAGs are provided in [79]Fig. 1A.
   Transcriptomics-based data sets are highlighted in red with
   literature-based data sets shown in green. The overlap of individual
   RAAGs between the six different data sources is displayed in [80]Fig.
   1B.

Table 1.

   Listing of renal age-associated datasets.
   Dataset acronym Dataset description # Genes Ref
   PubMed Human genes were extracted from the gene2pubmed mapping file for
   publications found with the query: (aging OR ageing OR replicative
   senescence) AND (renal OR kidney) 49 [[81]15]
   GO Genes associated with the GO term “aging” were retrieved and
   filtered for those being expressed in either tubules or glomeruli based
   on data from the Human Protein Atlas. 196 [[82]16]
   GenAge DB Human genes were extracted from the GenAge DB and filtered
   for those being expressed in either tubules or glomeruli based on data
   from the Human Protein Atlas. 229 [[83]17]
   Digital Ageing DB Human genes associated with renal aging were
   retrieved from the Digital Ageing DB. 32 [[84]19]
   TX-RODWELL Genes associated with renal aging in the published
   transcriptomics dataset by Rodwell et al. were extracted from the
   supplementary datasets of the publication. 95 [[85]20]
   TX-MELK Genes associated with renal aging in the published
   transcriptomics dataset by Melk et al. were extracted from the
   supplementary datasets of the publication. 155 [[86]21]
   [87]Open in a new tab

   Overview and brief description of datasets used for generating the RAAG
   signature in the present study.

   Information from three sources, namely PubMed, GO and GenAge DB, was
   mainly based on literature information whereas information from the
   other three sources, namely the Digital Aging DB, TX-Rodwell, and
   TX-Melk, was based on high-throughput transcriptomics experiments. The
   largest contributions to the set of RAAGs were from the GenAge DB (229)
   and from the GO dataset (196), even after filtering for genes being
   expressed in kidney tissue based on information obtained from the Human
   Protein Atlas, as these sources are not restricted to and therefore not
   specific for renal aging per se. The overlap between these two
   literature-based sources was significant (p-value <.0001, chi-square
   test) with 49 genes being reported in both data sets. 49 RAAGs were
   extracted from resulting publications using the PubMed query “(aging OR
   ageing OR replicative senescence) AND (renal OR kidney)” after manual
   curation of human gene-to-pubmed entries. The transcriptomics datasets
   TX-Melk, TX-Rodwell, as well as the Digital Aging DB contributed 155,
   95, and 32 RAAGs respectively.

   Matrix metallopeptidase 7 (MMP7) showed up in five of the six data sets
   with its expression going up with increasing age. Next in the list was
   epidermal growth factor (EGF), which was found in four data sets, with
   its expression going down with increasing age. A number of genes was
   found in three data sets, among them the cyclin dependent kinase
   inhibitors 1A and 2A (CDKN1A and CDKN2A), klotho (KL), the mechanistic
   target of rapamycin kinase (MTOR), fibronectin 1 (FN1), or
   apolipoprotein E (APOE). The full list of RAAGs is available in
   [88]Supplementary Table S1 along with information on the individual
   sources.

3.2. Renal Specific Protein Abundance

   We evaluated RAAG expression in renal tissue using information from the
   Human Protein Atlas on protein abundance levels in glomerular and
   tubular compartments. Information was available for 582 of the 634
   RAAGs. 485 RAAGs showed at least low levels of protein abundance in
   tubules, while at least low protein abundance in the glomerular
   compartment was observed in 323 RAAGs. Average protein abundance was
   higher in tubules as compared with glomeruli with around 250 RAAGs
   being not at all expressed in glomeruli. [89]Fig. 2 displays protein
   expression of RAAGs in tubular and glomerular compartment based on data
   from the Human Protein Atlas.

Fig. 2.

   Fig. 2
   [90]Open in a new tab

   Renal age-associated gene (RAAG) protein abundance in renal tissue.

   Overview of protein abundance in renal tubules and glomeruli for the
   634 RAAGs based on data from the Human Protein Atlas. The number of
   RAAGs in each of the different categories is indicated by segment sizes
   of the outer ring. Color-coding of the outer ring indicates high
   protein abundance (red), medium protein abundance (orange), or low
   protein abundance (yellow) in tubular and glomerular tissue
   respectively. The fraction of RAAGs being not detectable (ND) on the
   protein level are given in light grey. No immunohistochemistry
   information was available for 52 RAAGs in the Human Protein Atlas
   (GLOM_NA and TUB_NA), displayed in dark grey. Connecting line segments
   are color coded based on tubular protein abundance levels.

3.3. Association with CKD Progression

   We evaluated RAAG expression in renal tissue in a set of 63 CKD
   patients with available clinical follow-up data. Mean follow-up time of
   the whole cohort was 7.01 years and key clinical parameters are listed
   in [91]Table 2. The percentage of male patients was higher in the
   progressive group (p-value = .0142, chi-square test) and patients were
   significantly older (p-value = .00057, t-test) as compared with the
   patients in the stable group. No significant difference between
   treatment was observed (p-value = .3338, chi-square test).

Table 2.

   Patient characteristics of the CKD cohort.
   Cohort (N = 63) Progressive (N = 24) Stable (N = 39) p-value
   Age [years]
    Min 17.06 20.00 17.06
    Max 74.00 74.00 73.85
    Mean (sd) 47.22 ± 17.49 56.19 ± 13.90 41.70 ± 17.32 0.00057
   Creatinine [mg/dL]
    Min 0.44 0.44 0.49
    Max 6.21 6.21 4.18
    Mean (sd) 1.55 ± 0.95 1.96 ± 1.13 1.31 ± 0.73 0.01624
   eGFR [ml/min/1.73 m^2]
    Min 9.46 9.46 16.24
    Max 153.68 153.68 150.58
    Mean (sd) 63.28 ± 35.20 49.54 ± 32.74 71.73 ± 34.36 0.01335
   UPCR [g/g]
    Min 0.22 0.72 0.22
    Max 12.77 11.23 12.77
    Mean (sd) 4.36 ± 3.37 4.22 ± 2.80 4.45 ± 3.71 0.78860
   Follow-up time [years]
    Min 0.52 0.52 1.22
    Max 17.08 11.34 17.08
    Mean (sd) 7.01 ± 4.30 4.7 ± 3.24 8.44 ± 4.28 0.00024
   Gender
    Female 27 (43%) 5 (21%) 22 (56%)
    Male 36 (57%) 19 (79%) 17 (44%) 0.01418
   Medication at baseline [n (%)]
    RAAS blockade 29 (46) 13 (54) 16 (41)
    Antidiabetics 5 (8) 3 (13) 2 (5)
    Statins 13 (21) 6 (25) 7 (18)
    Steroids 12 (19) 3 (13) 9 (23)
    NSAIDs 2 (3) 0 (0) 2 (5)
    NA 10 (16) 6 (25) 4 (10) 0.33380
   Histological diagnoses [n (%)]
    DN 5 (8) 5 (21) 0 (0)
    FSGS 7 (11) 3 (13) 4 (10)
    GN 1 (2) 1 (4) 0 (0)
    HTN 7 (11) 4 (17) 3 (8)
    IGAN 4 (6) 3 (13) 1 (3)
    LN 11 (17) 1 (4) 10 (26)
    MCD 10 (16) 0 (0) 10 (26)
    MN 10 (16) 4 (17) 6 (15)
    MPGN 2 (3) 2 (8) 0 (0)
    VASC 5 (8) 1 (4) 4 (10)
    Other 1 (2) 0 (0) 1 (3) 0.56460
   [92]Open in a new tab

   Clinical parameters are given for the CKD cohort [n = 63] as well as
   for the subsets of patients with a stable [n = 39] and progressive
   [n = 24] course of disease. eGFR = estimated glomerular filtration
   rate; UPCR = urine protein to creatinine ratio; RAAS = renin
   angiotensin aldosterone system; NSAIDs = non-steroidal
   anti-inflammatory drugs; DN = diabetic nephropathy; FSGS = focal
   segmental glomerulosclerosis; GN = glomerulonephritis;
   HTN = hypertensive nephropathy; IGAN = IgA glomerulonephritis;
   LN = lupus nephritis; MCD = minimal change disease; MN = membranous
   nephropathy; MPGN = membranoproliferative glomerulonephritis;
   VASC = vasculitis.

   31 RAAGs were identified as differentially expressed in the group of 24
   progressive CKD Patients as Compared with the Group of 39 CKD Patients
   Showing a Stable Disease course ([93]Table 3). Patient Core Clinical
   Parameters and Information on Assignment to the Progressive and Stable
   Group Can Be Found in [94]Supplementary Table S2

Table 3.

   RAAGs associated with CKD progression.
   Symbol Gene name Aging evidence Protein abundance in glomeruli Protein
   abundance in tubules Fold-change CKD p-Value FDR (%)
   C3 Complement C3 TX-RODWELL Not detected Not detected 2.69 2.41E-05 <
   1%
   LTF Lactotransferrin Digital Aging DB; TX-RODWELL Not detected Not
   detected 2.61 3.47E-05 < 1%
   IL7R Interleukin 7 receptor GenAge DB Not detected Medium 2.22 3.09E-04
   < 1%
   NNMT Nicotinamide N-Methyltransferase Digital Aging DB; TX-RODWELL Not
   detected High 2.19 2.39E-04 < 1%
   CD52 CD52 Molecule TX-MELK NA NA 2.10 3.48E-04 < 1%
   UBD Ubiquitin D TX-RODWELL NA NA 1.82 1.43E-03 < 1%
   APBB1IP Amyloid beta precursor Protein binding family B Member 1
   interacting protein Digital Aging DB Not detected Not detected 1.73
   1.79E-05 < 1%
   CFB Complement factor B TX-RODWELL Not detected Not detected 1.73
   5.96E-03 < 1%
   MMP7 Matrix metallopeptidase 7 PubMed; GO; Digital Aging DB;
   TX-RODWELL; TX-MELK Not detected Low 1.63 2.22E-03 < 1%
   VCAN Versican TX-MELK Low Low 1.57 2.91E-04 < 1%
   CLDN1 Claudin 1 GO Not detected Medium 1.57 3.77E-05 < 1%
   ITGB2 Integrin subunit beta 2 GO Not detected Not detected 1.57
   1.37E-02 < 3%
   MICALL2 MICAL-like 2 TX-RODWELL Not detected Low 1.56 3.37E-05 < 1%
   INPP5D Inositol polyphosphate-5-phosphatase D GO Not detected High 1.54
   1.06E-04 < 1%
   CLU Clusterin GenAge DB Not detected Not detected 1.49 8.61E-03 < 2%
   ITPR3 Inositol 1,4,5-trisphosphate receptor type 3 TX-RODWELL Not
   detected Low 1.46 6.41E-06 < 1%
   TMPRSS4 Transmembrane serine protease 4 TX-RODWELL Low Low 1.41
   5.11E-04 < 1%
   WFDC2 WAP four-disulfide core domain 2 TX-RODWELL Not detected Low 1.40
   1.23E-02 < 2%
   CDH6 Cadherin 6 TX-RODWELL Low High 1.40 5.76E-03 < 1%
   TSPAN1 Tetraspanin 1 TX-RODWELL Not detected High 1.38 1.18E-02 < 2%
   ANXA1 Annexin A1 TX-RODWELL Medium Not detected 1.36 4.01E-03 < 1%
   GLS Glutaminase TX-MELK Not detected High 1.32 1.78E-03 < 1%
   TNFRSF11B TNF receptor superfamily member 11b TX-RODWELL Not detected
   High 1.32 1.57E-03 < 1%
   CXCL6 C-X-C motif chemokine ligand 6 TX-RODWELL NA NA 1.28 1.01E-03 <
   1%
   TPBG Trophoblast glycoprotein TX-RODWELL Medium High 1.27 6.49E-05 < 1%
   NOS3 Nitric oxide synthase 3 PubMed Low Not detected −1.27 3.32E-04 <
   1%
   CGNL1 CINGULIN like 1 TX-MELK Medium High −1.27 1.30E-03 < 2%
   STARD8 StAR related lipid transfer domain containing 8 Digital Aging DB
   Low High −1.34 2.54E-04 < 1%
   CDKN2A CYCLIN dependent kinase inhibitor 2A PubMed; GO; GenAge DB
   Medium Medium −1.38 4.08E-04 < 1%
   LPL Lipoprotein lipase Digital Aging DB; TX-RODWELL NA NA −1.79
   1.34E-03 < 1%
   EGF Epidermal growth factor GenAge DB; Digital Aging DB; TX-RODWELL;
   TX-MELK NA NA −2.06 2.11E-04 < 1%
   [95]Open in a new tab

   Listing of RAAGs being significantly associated with CKD progression
   based on the comparison of 24 progressive and 39 stable CKD patients.
   Next to the Gene Symbol and the gene name, we provide information on
   the aging data sets as well as protein abundance levels in tubules and
   glomeruli. In addition fold-change values in progressive CKD along with
   p-values and FDR (%) are provided. RAAGs are sorted based on
   fold-change values in descending order.

   The gene set being significantly deregulated in progressive CKD was
   enriched in RAAGs from the TX-RODWELL (p-value <.00001) and the Digital
   Aging DB (p-value = .0004) datasets, both transcriptomics-based
   datasets. In contrast a significantly lower number of RAAGs than
   expected was present from the GenAge DB (p-value = .0456) based on a
   chi-square test.

   25 of the 31 differentially expressed RAAGs were upregulated and six
   RAAGs were downregulated in the group of progressive patients as
   compared with the group of stable CKD patients. 19 of the 25
   upregulated RAAGs showed a concordant pattern of expression with aging,
   i.e. being significantly upregulated in progressive CKD patients and
   being also reported as significantly upregulated with increasing age.
   Four of the six downregulated RAAGs on the other hand showed a
   concordant pattern in the opposite direction, i.e. being downregulated
   in progressive CKD patients and being reported to show decreased
   expression with increasing age. The RAAGs with concordant expression
   patterns in CKD and aging are highlighted in [96]Fig. 3 with colored
   labels.

Fig. 3.

   [97]Fig. 3
   [98]Open in a new tab

   Renal age-associated gene (RAAGs) associated with CKD progression.

   Barplot of RAAG fold changes when comparing expression data from
   progressive CKD patients with CKD patients showing a stable course of
   disease. Upregulated and downregulated genes in progressive CKD
   patients are indicated by red and green bars respectively. Genes
   showing concordant regulation of expression in CKD and renal aging are
   indicated by colored labels. Grey labels are used for genes showing
   discordant regulation of expression in CKD and renal aging.

   EGF was downregulated 2.06-fold in progressive CKD patients. C3 and
   lactotransferrin (LTF) on the other hand showed the largest
   upregulations in progressive CKD patients as compared with CKD patients
   showing a stable disease course with fold-change values of 2.69 and
   2.61, respectively.

3.4. Enriched Molecular Pathways and Functional Context

   33 enriched molecular KEGG pathways were identified using the signature
   of 634 RAAGs as input for the DAVID tool with a Bonferroni corrected
   p-value of <0.05. Among them, we identified well-known age-related
   signaling cascades such as cell cycle, mTOR signaling, P53 signaling,
   insulin signaling, or focal adhesion. A representation of enriched
   molecular pathways is displayed in [99]Fig. 4. Next to the enriched
   pathways, those genes showing concordant expression in CKD and renal
   aging that could be assigned to at least one of the enriched molecular
   pathways, are displayed.

Fig. 4.

   [100]Fig. 4
   [101]Open in a new tab

   Pathway map based on renal age-associated gene (RAAGs).

   Enriched molecular pathways based on the set of 634 RAAGs are
   displayed. Pathway color-coding indicates number of assigned RAAGs.
   Inter-pathway links indicate pathways sharing a common set of genes.
   RAAGs showing concordant regulation in CKD progression that could be
   assigned to at least one of the enriched pathways are displayed in
   addition with the color-coding indicating fold-changes in the CKD
   transcriptomics dataset.

   EGF is linked to the largest number of the 33 significantly enriched
   molecular pathways based on the signature of 634 RAAGs. Next to the
   ErbB signaling pathway itself, also PI3K-Akt signaling and Jak-STAT
   signaling play a crucial role, which has also been shown for focal
   adhesion, prominently represented by MMP7, FOXO, or RAS.

   A number of other RAAGs found to be significantly associated with CKD
   progression could be assigned to the broader context of cell adhesion
   and extracellular matrix remodeling including VCAN, MMP7, ITGB2, CLDN1,
   TSPAN1, TPBG, or CGNL1 based on information from GO terms and
   scientific literature. RAAGs involved in immune response and
   inflammation furthermore included complement C3 (C3), IL7R, CD52, UBD,
   CLU, ANXA1, and CXCL6.

3.5. Compounds Reversing the RAAG/CKD Profile

   The set of 23 RAAGs showing concordant expression patterns in CKD and
   renal aging was used as input for in-silico drug screening using the
   L1000 Characteristic Direction Signature Search Engine. The 50
   top-ranked compounds reversing the input RAAG signature were evaluated
   in detail with respect to the impact on individual RAAG expression
   levels. The heatmap in [102]Fig. 5 displays those 18 RAAGs as column
   headers being affected by at least one of the top-ranked compounds. The
   compounds are sorted based on the resulting drug score from the L1000
   Search from top to bottom.

Fig. 5.

   [103]Fig. 5
   [104]Open in a new tab

   Heatmap of compound–renal age-associated gene (RAAG) interactions.

   The impact of the 50 top-ranked compounds on RAAG expression is
   indicated in this heatmap. Compounds in part reversing the RAAG/CKD
   signature are ranked from top to bottom on the right hand side.
   Affected RAAGs are ranked from left to right in the top panel. Red
   squares in the heatmap indicate genes being upregulated by a specific
   drug, whereas green squares indicate genes being downregulated by a
   particular drug. Compounds in use in the context of diabetes and kidney
   disease are highlighted in green. Compounds targeting epigenetic
   mechanisms and in particular histone deacetylase inhibitors are
   highlighted in yellow. Compounds showing up multiple times in the list
   have been found to be of relevance in different cell lines and/or
   different concentrations. Details on experimental conditions in the
   drug profiling experiments are provided in Supplementary Table S3.

   Among the set of identified compounds were four drugs being already
   used in the clinical setting in the context of diabetes and kidney
   disease, namely atorvastatin, valsartan, rosiglitazone, and captopril.
   Next to these four compounds there was a set of compounds from the
   class of histone deacetylase (HDAC) inhibitors in the list such as
   vorinostat, trichostatin A, apicidin, or I-BET151. The majority of the
   remaining identified compounds were investigational drugs. A detailed
   list of the top-ranked compounds with information on the underlying
   cell-line, dose, and time is available in [105]Supplementary Table S3.

3.6. Impact of Identified Compounds on CKD Associated RAAGs in Human Renal
Proximal Tubular Cells

   To validate the findings from our in-silico prediction workflow, we
   tested the impact of four compounds (atorvastatin, valsartan,
   rosiglitazone, and captopril) on gene expression levels in human renal
   proximal tubular cells of selected RAAGs that were also associated with
   CKD progression. We selected RAAGs for which either literature
   information was available on the impact for at least one of the four
   drugs or that should be affected by at least three of the four drugs
   based on the in-silico predictions. The selected genes were EGF, TNF
   receptor superfamily member 11b (TNFRSF11B), MMP7, complement factor B
   (CFB), CD52, LTF, and C3. No valid relative concentrations could be
   determined for CD52 via qPCR and CD52 was therefore excluded from
   further analyses.

   Expression of C3 and TNFRSF11B was upregulated 1.37 and 1.38-fold by
   H[2]O[2] stimulus as compared with untreated HK2 controls.
   Rosiglitazone and captopril had the strongest impact on gene expression
   levels significantly counterbalancing expression levels of five and
   four of the selected genes, respectively ([106]Fig. 6). Valsartan and
   atorvastatin both significantly reduced expression of two RAAGs. In
   particular the two RAAGs linked to the complement and coagulation
   cascade C3 and CFB were affected by the selected compounds, with CFB
   being significantly downregulated by all four compounds and C3 being
   significantly downregulated by three of the four tested compounds.

Fig. 6.

   [107]Fig. 6
   [108]Open in a new tab

   Drug impact on renal age-associated gene (RAAG) expression in HK2
   cells.

   The impact of atorvastatin, valsartan, rosiglitazone, and captopril on
   RAAG gene expression in HK2 cells is displayed. Data represent
   expression fold changes as compared with untreated HK2 controls after
   normalization to expression levels of the housekeeping gene GAPDH.
   T-test was used to assess significance between H[2]O[2] stimulated
   cells and those treated with the respective drugs on top of H[2]O[2]
   stimulation. p-value coding: * < 0.05; ** < 0.01; *** < 0.001;
   **** < 0.0001.

4. Discussion

   In this study, we established and applied an in-silico analysis
   approach to identify compounds reversing a set of RAAGs also being
   significantly associated with CKD progression. We (i) consolidated a
   set of 634 RAAGs based on six different data sets, (ii) determined
   protein abundance in renal compartments, (iii) analysed enriched
   molecular pathways based on the RAAG signature, (iv) generated a set of
   RAAGs being also significantly associated with CKD progression in an
   independent transcriptomics dataset, (v) identified a set of compounds
   with the potential of reversing the expression signature of renal
   age-associated CKD prognosis genes, and (vi) validated the impact of
   selected drugs on RAAG expression in a cell culture model system of
   renal aging.

   A number of studies on renal aging is available in scientific
   literature, but to the best of our knowledge no data driven
   meta-analysis has been conducted so far. With three literature-based
   data sets and three gene expression data sets, we generated a unique
   set of 634 genes being associated with renal aging. Interestingly the
   overlap between genes from the six different data sets was sparse with
   the largest overlap found between the two literature-based data sets
   from GO and the GenAge DB. Over one third of RAAGs from the Digital
   Aging DB however were also part of one of the two transcriptomics
   datasets TX-RODWELL or TX-MELK. The remaining datasets seemed to
   provide complementary findings.

   Among the set of 33 identified enriched molecular pathways based on the
   signature of 634 RAAGs are key age-related mechanisms such as the mTOR
   pathway, the p53 signaling cascade, or cell cycle regulation but also
   mechanisms like focal adhesion or insulin resistance. mTOR signaling
   modulates aging and age-related diseases [[109]30] while insulin
   resistance amplifies chronic inflammation leading to accelerated aging
   [[110]31]. Unresolved inflammatory processes in renal transplants
   trigger delayed graft function involving aging pathways suggesting a
   dominant role of inflammation in the process of renal aging [[111]32].
   Cell cycle regulation has been discussed widely as major factor in the
   process of cellular aging also in the field of nephrology
   [[112]33,[113]34], whereas the p53 family is involved in DNA repair
   interacting with cellular aging [[114]35].

   The hunt for compounds with a positive impact on molecular processes of
   aging is an intriguing area of research. A couple of substances have
   already been identified such as compounds reducing the effect of
   reactive oxygen species or compounds targeting the mTOR complex with
   different hypotheses to expand life span including specific translation
   of mRNAs, improving stem cell maintenance or anti-inflammatory
   mechanisms [[115]30,[116]36,[117]37]. Using the set of 23 RAAGs showing
   concordant regulation in aging and CKD progression we identified
   compounds with the potential to reverse the RAAG/CKD expression
   signature. Drugs used to treat diabetes, hypertension, or
   hypercholesteremia (rosiglitazone, captopril, valsartan, and
   atorvastatin) showed up in this list, all directly affecting renal
   function [[118]38,[119]39].

   We investigated the impact of these four drugs on RAAG expression
   levels in a cell culture model system of aging, namely treatment of
   renal proximal tubular HK2 cells with H[2]O[2] [[120]40]. All four
   investigated drugs showed significant beneficial impact on RAAG gene
   expression with the antidiabetic drug rosiglitazone having the
   strongest impact targeting five of the six selected genes. There is
   also evidence for the impact of rosiglitazone on LTF, MMP7, and
   TNFRSF11B supporting our findings [[121]41], [[122]42]. The two
   complement and coagulation factors C3 and CFB were affected by all
   compounds in our in-vitro model system. Activation of the alternative
   pathway of the complement system has been discussed as a link between
   obesity and metabolic disorders. Rosiglitazone leads to a significant
   reduction of CFB, which is positively correlated to a number of obesity
   associated parameters [[123]43]. Statins also showed beneficial impact
   on the level of components of the complement and coagulation cascade in
   a cohort of patients with systemic sclerosis [[124]44].

   Next to these in-vitro tested compounds, epigenetic regulators such as
   the HDAC inhibitors vorinostat, trichostatin A, or apicidin along with
   the bromodomain and extra-terminal motif (BET) inhibitor I-BET151
   scored high in our in-silico ranking. Histone deacetylases are enzymes
   regulating the modification of chromatin arrangement, transcriptional
   activity, and are involved in enabling epigenetic integrity.
   Renoprotective effects of HDAC inhibitors have been described in
   combination with inhibitors of the RAAS [[125]45]. This group of
   compounds was also found to have a positive impact in the context of
   aging in the human brain [[126]14]. Epigenetic regulation as
   therapeutic option in kidney disease however seems to be a two-edged
   sword where the therapeutic benefit has to be weighed against the
   nephrotoxic potential as reviewed recently [[127]46].

   We compared our results to a list of potential anti-aging compounds,
   which was consolidated by Dönertaş and colleagues based on the results
   of twelve different studies [[128]47]. The two HDAC inhibitors
   vorinostat and trichostatin A were part of the consolidated list being
   among the few compounds with at least three supporting studies. The ACE
   inhibitors captopril and enalapril were also listed next to
   pravastatin, a member of the drug class of statins like atorvastatin,
   which was identified in our study. The relevance to human aging and
   disease however needs to be established, as the evidence for the link
   to aging for the ACE inhibitors as well as pravastatin was based on
   studies in rotifers [[129]48].

   Those RAAGs being associated with CKD progression and being affected by
   one of the compounds in clinical use for diabetes and/or kidney disease
   might help in this task and might all be considered as potential
   predictive marker candidates for the respective drugs. CKD patients
   with elevated levels of certain RAAGs might for example have a larger
   benefit from therapeutic intervention than patients with normal RAAG
   expression levels. Assessment of these proteins as predictive markers
   was however beyond the scope of the present work but will be the focus
   of future follow-up studies.

   Findings of the present study are also of relevance to all future
   nephrological biomarker studies. As soon as one of the RAAGs is
   investigated as biomarker candidate in the context of kidney disease,
   adjustment for age is probably even more important in order to draw
   correct conclusions from the respective biomarker study. It might turn
   out, that the value of the biomarker is merely due to its association
   with chronological age, which still is the strongest single predictor
   of overall mortality [[130]49].

   This exploratory study has its limitations. Despite the fact that we
   applied quite stringent criteria on the underlying age-related data
   sets there is a chance that a certain amount of genes in the RAAG set
   is of minor relevance to renal aging and has therefore to be considered
   as false positives. In particular the initial gene sets extracted from
   GO and the GenAgeDB needed another curation step, as neither of the two
   sources specifically focused on renal aging alone. To reduce the number
   of false positives we therefore applied a filter using data from the
   Human Protein Atlas on gene and protein abundance values in renal
   tissue. After the pathway enrichment analysis, we however felt very
   positive about the generated set of RAAGs, as a number of aging and
   kidney disease related molecular pathways showed up on top of the list.

5. Conclusions

   In summary, we consolidated a set of RAAGs based on six independent
   data sources. RAAGs being significantly associated with CKD progression
   in an independent transcriptomics dataset were used to screen for
   compounds reversing the RAAG/CKD signature. Atorvastatin, valsartan,
   rosiglitazone, and captopril – all drugs being widely used in clinical
   routine to treat hypertension and diabetes, two conditions causing up
   to 50% of all cases of end-stage renal disease – were among the
   top-ranked compounds. In particular rosiglitazone and captopril had a
   significant beneficial impact on RAAG expression levels in human renal
   proximal tubular cells. In addition, compounds targeting epigenetic
   regulatory mechanisms were identified impacting renal age-related
   mechanisms.

   The following are the supplementary data related to this article.
   Supplementary Table S1

   Complete renal age-associated gene (RAAG) set Supplementary Table 1
   holds the complete list of 634 RAAGs along with information on the
   underlying data sources. The Gene Symbol as well as the Ensembl GeneID
   is provided as gene identifiers.
   [131]mmc1.xlsx^ (32.1KB, xlsx)
   Supplementary Table S2

   CKD patient cohort The set of 63 CKD patients used in this study in
   order to identify RAAGs being associated with CKD progression is
   provided in supplementary table 2. The provided patient IDs can be used
   in order to retrieve the respective gene expression datasets from GEO
   with the accession number [132]GSE60861.
   [133]mmc2.xlsx^ (9.4KB, xlsx)
   Supplementary Table S3

   Compounds reversing the renal age-associated gene (RAAG)/CKD signature
   The detailed listing of the 50 top-ranked compounds reversing the
   RAAG/CKD signature is available in supplementary table 3 with
   information on the calculated drug scores as well as information on
   used cell-lines, doses and time points of the underlying compound
   profiling data.
   [134]mmc3.xlsx^ (10.4KB, xlsx)

Author Contributions

   CK and PP designed and coordinated the study. CK and PP led the
   analysis with JL and MR supporting data consolidation and analysis. MR
   and JK consolidated clinical data of the CKD patient cohort used within
   this study. JL performed cell culture experiments. CK and PP drafted
   the first version of the manuscript. All authors contributed to results
   interpretation and manuscript writing. All authors read and approved
   the final manuscript.

Declarations of Competing Interests

   All authors have declared that no conflict of interest exists.

Acknowledgements