Identification of nephropathy candidate genes by comparing sclerosis-prone and sclerosis-resistant mouse strain kidney transcriptomes

Animal experiments were conducted in accordance with the National Institute of Health guidelines and approved by the Institutional Animal Care and Utilization committee at Case Western Reserve University and the Medical College of Wisconsin. ROP-Os/+ and C57-Os/+ mice were purchased from Jackson Laboratories (Bar Harbor, ME) and housed in the animal facilities at the above institutions. We phenotyped mice at 6,10,12, and 16 weeks by comparing urine protein to creatinine ratio,glomerular surface area, and sclerosis score in Periodic Acid Schiff (PAS) stained sections as previously described [30]. Glomerular number was determined by the maceration method as previously described [31]. The kidneys were collected for SAGE library construction at 6 weeks of age to avoid identification of developmental genes (because mouse nephrogenesis continues up to 4 weeks of age), and prior to the onset of significant renal disease. After sacrifice, the kidneys were promptly removed and the renal medulla was dissected away and discarded. A small portion of each kidney was fixed in formaldehyde, sectioned, and stained with PAS for histopathological analysis.

Poly(A)⁺ RNA was isolated from renal cortices of ROP-Os/+ and C57Bl/6-Os/+ mice using ion-exchange column kits from Qiagen (Valencia, CA). The quality of the isolated RNA was analyzed by agarose gel electrophoresis and RNA concentration was determined by UV spectrophotometry [32]. RNA from three mice per strain was pooled and cDNA was synthesized from a starting amount of 5 μg poly(A)⁺ RNA using cDNA synthesis kit, according to manufacturer recommendation (Invitrogen). A fraction of first and second strand reactions was labeled with ³²P] to evaluate synthesis efficiency. cDNA quality was assessed using agarose gel electrophoresis and by measuring radioisotope incorporation. SAGE steps were performed as described previously [29] using the restriction endonuclease Nla-III as the anchoring enzyme [29]. We used SAGE software to analyze the SAGE produced concatemer sequences, which identifies the cDNA tags flanked by Nla-III restriction enzyme recognition sequence. In addition, the software identifies and excludes duplicate ditags, which are likely to be PCR amplification artifacts. Confirmation of SAGE tag gene identity, as well as relative expression levels were performed using northern blots, RNase protection assay or real-time RT-PCR. The confirmatory methods and data are in a Additional file 1.

We applied several statistical models, which employ Bayesian methods or Poisson distributions, to identify tags that were significantly different between the ROP-Os/+ and C57-Os/+ kidney libraries [33‐36]. The method allows comparison of unequal libraries based on the assumption that tag counts are binomially distributed. Regardless of the method used to identify statistically significant differentially expressed tags, rank orders and p-values were comparable.

Accurate annotation of SAGE tag libraries requires a combination of methods. Accordingly, we used multiple approaches to annotate SAGE tags including the online engine “SAGE genie” (http://​cgap.​nci.​nih.​gov/​SAGE/​mSEM), and manual nucleotide blast (blastn) searches. We also constructed genome-wide mouse cDNA data sets from NCBI-Ref-Seq and Riken-Fantom collections and designed a software program, which identifies and catalogues the 3’-most “CATG” sequence string (Nla-III recognition sequence) with the 3’ flanking 10 nucleotides. These tag-to-gene database tables were used to annotate the SAGE tags. Ambiguous tags, which mapped to multiple genes, and single copy tags were excluded. Tags were assigned gene identifiers if the tag mapped to a sequence that contained the most 3’-CATG sequence. Sequence databases were searched in the following priority: 1) our own libraries containing tags extracted from Ref-Seq and Fantom cDNA databases; 2) the mouse mSAGE expression matrix using SAGE Genie; or 3) the mouse mRNA (non-redundant [nr] and expressed sequence tags [dbest]) databases using blastn.

We utilized the Ingenuity Pathway Analysis (IPA) system (Ingenuity Systems, Mountain View, CA, USA, http://​www.​ingenuity.​com) to visualize the SAGE data as biological networks. To perform a “connectivity” analysis, a table containing selected gene identifiers, defined as Focus Genes, and their corresponding relative mRNA expression ratios between ROP-Os/+ and C57-Os/+ kidneys was uploaded into the application. Criteria for selection as Focus Genes included: 1) a greater than 2-fold difference in expression levels between the ROP-Os/+ and C57-Os/+ kidney libraries; and/or 2) a p-value < 0.05 for differential expression after correction for multiple testing. Three hundred and nine genes met these criteria, were mapped to their corresponding gene objects in the Ingenuity Pathway Knowledge Base and subsequently overlaid onto a global molecular network developed from information contained in Ingenuity Pathways Knowledge Base. Networks were then algorithmically generated based on the connectivity of the eligible genes. Twenty-eight Focus Genes failed to generate a shared molecular network or pathway. Function of the connectivity networks was inferred from the functions of the genes in the network as annotated in the Knowledge Base.

An additional set of networks was generated using the functional information in the Ingenuity Pathways Knowledge Base (“functional” analysis). Genes from the SAGE kidney libraries, which met the cutoffs defined above and that were associated with biological functions in the Ingenuity Pathways Knowledge Base, were considered for this analysis. Right-tailed Fisher’s exact test was used to calculate a p-value reflecting the probability that each biological function assigned to that network is due to chance alone. Networks were ranked according to the combinatorial p-value of differentially expressed genes represented in the network.

Western blot of kidney protein extracts were performed as previously described [37] using TGFβ specific antibody. Equal loading was verified by stripping the membrane and probing for actin expression.

TGFβ activity was determined using mink lung epithelial cells expressing the firefly luciferase reporter gene under the control of the minimal, TGFβ-responsive, plasminogen activator inhibitor-1 (PAI-1) promoter [38] (gift from Dr. Daniel Rifkin, New York University). The culture conditions were as previously described [39]. The effect of Itch overexpression on TGFβ was determined from two nearly confluent six-well tissue culture plates, which were transiently transfected with either mouse Itch cDNA which was cloned in pCDNA3.1 or empty vector control, using lipofectamine (Invitrogen) according to manufacturer recommendations. Twenty-four hours after transfection, TGFβ (R&D system) was added to the culture media at 200 ng/ml. The cultures were maintained for 8 h and luciferase activity was assayed using a Promega kit according to manufacturer instructions. Relative luminescence units were determined compared to control conditions (empty pCDNA3.1 plasmid transfected cells).

We compared 40 SAGE-derived transcripts with the most divergent expression levels between ROP-Os/+ versus C57-Os/+ kidneys, with the human microarray expression data, comprised of samples obtained from micro-dissected human kidney biopsies from patients with FSGS [40]. The fold change and significance (assessed by q-value, corrected for multiple testing) was calculated by the Significance Analysis of Microarrays method (SAM) [41] using the MeV software from the Institute of Genome Research (http://​www.​jcvi.​org).

To identify genes which are differentially expressed in our model system, and are known to be expressed in the glomerulus, we searched for common genes between the glombase [42] gene list, SAGE glomerular tags [43] and annotated transcripts with the most significant differential expression between ROP-Os/+ versus C57-Os/+ kidneys. We then used Pubmed and Ingenuity pathway analysis to mine the literature for possible links between these genes and nephropathy.

We found that the urine protein to creatinine ratio was significantly higher in ROP-Os/+ mice compared to both C57-Os/+ and ROP-+/+ at all time points. The proteinuria in ROP-Os/+ mice reached a peak at 12 weeks. The glomerular surface area at 6 weeks was similar between the ROP-Os/+ and C57-Os/+ (5646 ± 286 and 5708 ± 460 μm², respectively) and significantly larger than glomerular surface area of ROP-+/+ (3984 ± 456 μm²). However, by 16 weeks of age, the surface area had increased significantly in the ROP-Os/+ compared to C57-Os/+glomeruli (8990 ± 628 vs 6458 ± 1259 μm²). Furthermore, the ROP-Os/+ mice showed progressive increase in sclerosis score over time (Additional file 1).

We constructed SAGE libraries from renal cortical RNA extracted from ROP-Os/+ and C57-Os/+ kidneys at 6 weeks of age (Table 1), a time point which permitted completion of differentiation, but prior to observation of significant glomerulosclerosis. The combined ROP-Os/+ and C57-Os/+ kidney SAGE libraries contained 49,588 cDNA sequence tags, of which 20,594 were unique. We did not observe “loss of AT-rich tags” bias [44] and the libraries demonstrated 54.32% A+T and 45.68% C+G nucleotides. Table 2 shows the 50 most statistically significant, differentially expressed and annotated tags. We compared the annotated genes in our SAGE libraries to publically available wild type mouse Nla-III-anchored SAGE libraries. Of the genes not previously described in SAGE kidney libraries, we observed significant expression of arginine-glutamic acid dipeptide (RE) repeats (Rere or Atrophin-2), insulin-like growth factor binding protein (Igfbp7), aminolevulinic acid synthase (Alas1), glutathione S-transferase, theta 1 (Gstt1), and propionyl coenzyme A carboxylase (Pccb). Interestingly, all of these transcripts were induced to a detectable level in ROP-Os/+, suggesting that deep SAGE coverage can unmask changes in kidney gene expression patterns induced in a disease model. To identify differentially expressed glomerular genes, we searched for common genes between glombase [42], SAGE glomerular tags [43] and 220 annotated transcripts with the most significant differential expression between ROP-Os/+ versus C57-Os/+ kidneys. This comparison yielded 67 genes, of which 53 were upregulated, and only 14 were downregulated in the ROP-Os/+ mouse (Table 3). We confirmed a reduction in glomerular number in 6 week old ROP-Os/+ mice (data not shown), suggesting that upregulated gene expression is not due to altered glomerular mass.

Using the Ingenuity Pathway Analysis software to perform a connectivity analysis (see Methods), we identified multiple molecular networks of genes differentially expressed in ROP-Os/+ and C57-Os/+ kidneys. These interactions are developed from published literature describing either physical or functional interactions between the molecules. The 13 most significant gene networks, based on connectivity and their imputed molecular and cellular functions, are shown in Table 4; corresponding diagrams of the actual connectivity networks are shown in Figure 1 and the supplemental data (Additional file 1).

Because TGFβ has been implicated in the pathogenesis of glomerulosclerosis, we next chose to overlay the canonical TGFβ signaling network onto our connectivity networks. TGFβ1 signaling pathway target genes, Hnf4a, Itch, and Map3K7ip1, were upregulated in the ROP-Os/+ kidneys, in agreement with published microarray analysis of laser-captured/microdissected human glomeruli from FSGS biopsies [45]. Overlaying TGFβ1-regulated signaling pathways on to the connectivity networks of differentially expressed genes demonstrated that directional changes in gene expression patterns (up or down) in multiple networks (1, 2, 6, 8, and 13) could be a result of TGFβ1 signaling. Figure 2 shows molecular network 1 (Table 4) with the TGFβ signature identified.

We also performed a functional analysis using the Ingenuity Pathway Analysis system (see Methods). These data demonstrated functional clustering of differentially expressed genes within known canonical pathways, and highlighted mitochondrial dysfunction and oxidative stress responses as mechanisms for disease (Figure 2). Interestingly, a Nrf2-dependent stress response, which is the target for a new class of diabetic nephropathy drug [46], appears to also be activated in ROP-Os/+ kidneys.

Surprisingly, TGFβ Tag was not differentially expressed between the sclerotic and non-sclerotic kidneys. Because the pathway analyses identified a prominent role for TGFβ in ROP-Os/+−dependent FSGS pathogenesis, we there fore compared TGFβ proteinbetween ROP-Os/+ and C57-Os/+ kidneys using Western blot analysis, and found no significant difference between the two strains (Figure 3).

To test the involvement of TGFβ-related candidate genes in glomerular fibrogenesis, we evaluated the effect of Itch, which was upregulated in ROP-Os/+ kidneys, on TGFβ signaling. This gene was selected because Itch, an E3 ubiquitin ligase, amplifies TGFβ signaling by facilitating the interaction between the TGFβ receptor and SMAD2 [47]. For these experiments we used mink lung epithelial cells which express the firefly luciferase under the control of minimal TGFβ-responsive PAI-1 promoter [38]. As seen in Figure 4 cells transfected with Itch cDNA showed higher luciferase activity in response to TGFβ, compared to control cells transfected with empty vector, suggesting that enhanced Itch expression may be relevant to the ROP-Os/+ phenotype by stimulating TGFβ signaling.

To determine if the comparative transcriptome analysis illuminated changes in gene expression in human FSGS, we compared the 40 most significant, differentially expressed genes derived from the ROP-Os/+ vs. C57-Os/+ analysis, with microarray data from kidney biopsy samples derived from patients with FSGS [48]. Of the 40 queried genes, corresponding oligonucleotides from six (ACSM2A, TSPAN33, ERRFI1, NDUFA12, TMEM27, and RHPN2) were not spotted on the array chip. Of the remaining 34 genes, 17 were concordantly regulated in the human glomerulus and ROP-Os/+ mouse kidney, and altered mRNA expression in the human samples was statistically significant for 13 of the 17 genes. Two of the 13 genes were also concordantly regulated in the human tubulointerstitium and ROP-Os/+ mouse kidney (Table 5). An additional concordantly regulated gene, ID2, which encodes inhibitor of DNA binding 2, localized exclusively to the human tubulointerstitium.

There are >100 potentially different expressed genes with a p-value of <0.05 between the mouse strains studied in this report. It is likely that the sclerosis phenotype of the ROP-Os/+ mice (or the resistance of the C57-Os/+ to sclerosis) can be attributed to the combined effect of multiple genes. The renal pathology is due to the expression pattern of these genes on top of the reduced nephron number with its consequent hyperfiltration. The FSGS phenotype is not merely due to difference in the expression of background genes, as we contrasted the wild type ROP versus C57BL/6 kidney transcriptome using microarray, and did not detect differential expression of the SAGE identified genes (data not shown).

In addition to the TGFβ-related genes, which were evaluated in detail in Results, SAGE data showed that neuropilin-2 (Nrp2), which is expressed in the glomerulus and regulates vascular endothelial growth factor (Vegf) signaling [51, 52], is downregulated in the sclerosis prone ROP-Os/+ mouse. Nrp2 knockout mice show progressive glomerular damage after injection of the podocyte toxin, adriamycin, in contrast to their wild type littermates, who recover after the initial injury (J. Sedor, unpublished data).

Pea15 gene, which is overexpressed in the sclerotic mouse kidney, is a death effector domain-containing protein and promoter of autophagy. PEA15 inhibits caspase activation and increases ERK activity [53]. Transgenic mice overexpressing PEA15 have glomerular mesangial expansion and a histological pattern similar to diabetic nephropathy [54]. PEA15 induces the expression of the glucose transporter Glut1 in skeletal muscle cells [55]. Interestingly, Glut1 is also overexpressed in glomeruli from the FvB-Os/+ mice (Os mutation bred on FvB genetic background) [24], suggesting that PEA15 regulation of Glut1 may play a role in non-diabetic glomerular scarring.

In the ROP-Os/+ strain a number of ubiquitin-proteasome pathway genes are differentially expressed; Psmb5, Psma3, Fbxl12, and Itch gene expression was enhanced, while Siah1a was downregulated. The differential expression of these genes suggests a role for aberrant protein degradation in glomerulosclerosis phenotype of the Os mice. Itch has been linked to the regulation of a multitude of signaling cascades, including TGFβ and EGF through ubiquitin and non-ubiquitin mediated mechanisms [47, 56]. We tested the effect of Itch overexpression in vitro as a proof of concept that it regulates TGFβ signaling, which has been widely implicated in the pathogenesis of glomerulosclerosis. The experimental data support the hypothesis that Itch overexpression increases TGFβ signaling, independent of TGFβ ligand level (Figures 3 and 4). Itch expression is regulated by the Src kinase Fyn [57‐60], which regulates podocyte function through phosphorylation of nephrin [61, 62]. Bigenic Fyn/Cd2ap heterozygotes demonstrated an FSGS phenotype [63]. Because of the links with Fyn, TGFβ and other TGFβ-related signaling molecules, such as Pcbp1 [64], we speculate that Itch may function as a molecular rheostat, by regulating downstream TGFβ signaling (and FSGS pathophysiology) independent of ligand concentration.

Laser capture microdissection-microarray analysis of FSGS glomeruli demonstrated multiple changes consistent with activity of TGFβ signaling [45]. Although we did not identify differences in TGFβ ligand SAGE tag expression in our libraries, and Western blot analysis showed similar TGFβ protein expression between ROP-Os/+ and C57-Os/+ mouse kidneys, we postulate that the TGFβ pathway in the ROP-Os/+ kidney is upregulated by downstream molecules, such as Itch and SnoN.

Using the Ingenuity Pathway Analysis engine, the clustering of genes involved in oxidative stress response like the Nrf2 response genes and GPX suggests a role for an electrophile or oxidative stress in the mechanisms promoting renal injury in the ROP-Os/+ model [65]. Recently, bardoxolone methyl, which activates the Keap1-Nrf2 anti-oxidant pathway, was shown to protect kidney function in patients with type 2 diabetes [46].

Genes overexpressed in the C57-Os/+ mouse may be protective. For example, glutathione-S-transferase theta (Gstt1) tag counts were 18.6 fold higher in the C57-Os/+ compared to ROP-Os/+ mice, and it was shown that the deletion of this gene increases the likelihood of ESRD in diabetic patients [66]. Also, upregulation of Hsp90ab1, Vcp, Prdx1, and Prdx2 genes in the sclerosis resistant C57-Os/+ strain could indicate a robust protective response to the increased oxidative stress induced by the Os mutation.

Connective tissue growth factor (Ctgf), integrin-β1 (Itgb1), and secreted phosphoprotein 1 (Spp1) were all upregulated in the sclerosis-prone ROP-Os/+ mouse kidney. These genes are also upregulated in a other renal diseases characterized by fibrosis [67‐72]. On the other hand, complement factor H (Ctf) and prosaposin (Psap) are relatively downregulated in ROP-Os/+ kidneys, similar to observations in membranoproliferative glomerulonephritis (MPGN) and tubular damage [73, 74].

By comparing our ROP-Os/+SAGE libraries to previously published kidney SAGE libraries, we identified 13 genes, which were concordantly regulated in ROP-Os/+ kidneys and human FSGS kidney biopsies, all of which localize to the glomerulus. None of these have been previously detected using either SAGE or microarray libraries from normal kidneys, (http://​www.​ncbi.​nlm.​nih.​gov/​geo/​) [43, 75], suggesting that novel glomerular gene sets are induced as part of the pathophysiology of FSGS. Alternatively, the regulation of gene expression in fibrotic glomeruli could be a manifestation (rather than a cause) of fibrogenesis, though we attempted to minimize this possibility by selecting mice for study at an age prior to the appearance of glomerulosclerosis.

SAGE is a powerful tool for gene and pathway discovery. However, like other global gene expression analysis methods, it has inherent limitations. The lack of a uniform annotation tool for SAGE tags makes data analysis burdensome. Although SAGE permits assembly of a comprehensive transcriptome, the sequencing costs can be significant. As a result, most labs have resorted to microarray technology to achieve adequate, but less comprehensive coverage.