INTRODUCTION

TOP ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION REFERENCES

THE INCIDENCE OF END-STAGE renal disease has risen at an annual rate of 7-9% for the last decade (3). This increase is due, at least in part, to incomplete understanding of renal disease pathophysiology and limited therapeutic options to prevent disease progression. Kidney disease-oriented research has primarily focused on candidate molecules, emphasizing fibrogenic pathways (7) and hemodynamic alterations in the setting of reduced nephron number (8) as key mechanisms regulating chronic renal disease initiation and progression. However, chronic renal failure results from complex interaction of genetic and environmental risk factors, and interruption of a single effector pathway is unlikely to result in significant therapeutic benefit (31).

In contrast to focusing on candidate effector pathways, gene expression profiling is an alternative but powerful tool to better understand renal disease pathogenesis by generating a detailed analysis of mRNA expression profiles. Novel molecular techniques used to generate transcript libraries simultaneously can determine net consequences of gene-gene and gene-environment interactions on expression of thousands of genes. Rather than applying a priori assumptions (i.e., hypothesis testing), kidney transcript profiles from normal animals and animals with progressive kidney disease could be "mined" by analytic methods developed to discover unexpected relationships between genes or pathways (i.e., "bottom-up approach") (6). Although hypothesis-driven experimentation remains critical for knowledge discovery, thoughtful analysis of expression profiles generated with other model systems has yielded unanticipated results and defined new paradigms. For example, cluster analysis of transcript profiles showed that serum specifically activated wound-healing processes in fibroblasts rather than serving as a general signal for cell growth (18). Global expression monitoring also demonstrated that receptor tyrosine kinases, with unique ligand specificities and distinct biological effects, induce broadly overlapping, rather than independent, sets of genes (13;27), suggesting that cellular responses depend less on the specific ligand than cellular differentiation state, signal strength, or combinatorial interactions between activated signaling pathways. Finally, clustering of transcripts with similar expression patterns can assign a precise biological pathway to a gene of "generic" function (e.g., kinase) and provide clues to function of an unknown gene by relating it to characterized genes (1, 10).

Gene expression profiles can be generated and compared by multiple techniques. Subtractive hybridization (33), subtraction libraries (14), and differential display (2, 5) are semiquantitative methods, which may not detect small, pathophysiologically important variations in gene expression. In contrast, DNA microarrays quantitatively assay expression levels of thousands of genes, using discrete DNA sequences robotically imprinted on glass microscope slides (9), but require technology and instrumentation not available to most investigators. Serial analysis of gene expression (SAGE) also allows simultaneous, quantitative analysis of a large number of transcripts (36) but does not require expensive instrumentation. Using SAGE, an individual investigator with access to an automated sequencer can quantify mRNAs, expressed at a level of 100 copies/cell, within months (36) and can identify transcripts expressed as low as 1 transcript/cell in larger SAGE libraries (37). To date, SAGE expression libraries have been generated predominantly from yeast, tumor specimens, and cultured cells (37, 42). Because quantitative analyses of transcript levels appear to be critical for a better understanding of normal cell function and cellular response to injury, we have utilized SAGE to generate a cDNA tag library, which is composed of unique 9- to 13-bp cDNA sequence signatures, from mouse kidney mRNA. The results provide feasibility of applying SAGE for comparison of gene expression between normal and diseased kidney to discover new mechanisms of renal disease pathogenesis and to identify novel targets for renal disease therapy.

	MATERIALS AND METHODS

TOP ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION REFERENCES

RNA preparation. As a prelude to comparing kidney expression libraries generated from wild-type animals and animals with progressive chronic renal failure, kidneys were harvested from 26-wk-old male ROP-Es/+ mice (Jackson Laboratories, Bar Harbor, ME), which are glomerulosclerosis prone but have normal kidney structure and function in the absence of a triggering event such as hyperglycemia or nephron reduction (12, 40). One and three-quarters kidney were used for polyadenylated RNA (A+) extraction by using RNeasy and Oligotx extraction kits (Qiagen, Valencia, CA) according to the manufacturer's recommendation.

Production of ROP-Es/+ kidney tag library. SAGE kidney libraries were generated as described by Velculescu et al. (36) with modifications to maximize the final yield. A schematic of SAGE can be found on the Internet (http://www.sagenet.org) and in Fig. 1. Briefly, biotin-TEG-oligo-dT (Integrated DNA Technologies, Coralville, IA) was used to drive first-strand cDNA synthesis of mouse kidney A+ RNA using a MMLV-RT cDNA synthesis kit (GIBCO-BRL, Gaithersburg, MD). Second-strand cDNA was then generated with a final yield efficiency of 75-85%. Biotinylated cDNA was digested with the restriction enzyme Nla III (New England Biolaboratories, Beverly, MA) and 3'-cDNA end fragments isolated with streptavidin-coated magnetic beads (Dynal, Oslo, Norway). 3'-cDNAs were split into two pools, and each pool was ligated by using T4 DNA ligase (GIBCO-BRL) to custom-made 40-bp SAGE oligonucleotide DNA linkers (L1 and L2; Integrated DNA Technologies). To release linker-cDNA tag hybrids, the ligation reactions were then digested with the class II-shift restriction endonuclease Bsm FI (New England Biolaboratories) at 60°C for 2 h with continuous rotation in a hybridization oven. Magnetic beads were pelleted, and released linker-tags were blunted with T4-DNA polymerase (New England Biolaboratories) in the presence of deoxynucleotides. The L1- and L2-linker-tag complexes were then ligated together to form linker-ditag-linker constructs. To determine the optimum input of the linker-ditag-linker for large scale PCR, aliquots of serially diluted (1:10-1:1,280) ligation products were PCR amplified for 25 cycles (95°C for 30 min, 55°C for 1 min, then 70°C for 1 min) and analyzed by 8% PAGE. Subsequently, a large-scale (96 tubes) PCR reaction was conducted at optimum template dilution, and PCR products were pooled and digested with Nla III to release kidney cDNA ditag sequences. Released ditags were separated from linkers by 12% PAGE, eluted, and ligated together to produce concatemers. Concatemers were size fractionated by using 8% PAGE, and those between 600 and 1,200 bp were isolated and cloned into pZero (Invitrogen, San Diego, CA), which had been digested with Sph I. After electroporation into DH10B (GIBCO-BRL), clones were analyzed for recombinants by direct PCR using M13 forward and M13 reverse primers. PCR products >500 bp in size were isolated by 1.2% agarose gel electrophoresis and sequenced by using a BigDye terminator cycle sequencing kit (Perkin-Elmer, Foster City, CA). Sequencing reactions were analyzed by using an ABI-377 automated sequencer (Perkin-Elmer).

View larger version (29K): [in this window] [in a new window]   Fig. 1.   Schematic of serial analysis of gene expression (SAGE). See MATERIALS AND METHODS for details. Adapted from Velculescu et al. (36, 37) and http://www.sagenet.org/.

Data analysis. Concatemer sequences were analyzed by using SAGE software v1.0 (provided by Dr. Kenneth Kinzler's laboratory, Johns Hopkins University, Baltimore, MD), which automatically detects and counts tags from sequence files. Tag counts directly reflect transcript abundance (36, 42). SAGE software excludes replicate ditags from the tag sequence catalogue, because the probability of any two tags being coupled in the same ditag is small, even for abundant transcripts (36). The identities of the genes corresponding to the tags were determined by homology searches of public databases at the National Center for Biotechnology web site (http://www.ncbi.nlm.nih.gov), including GenBank, European Molecular Biology Laboratory (EMBL), DNA Database of Japan (DDBJ), Protein Data Bank (PDB), and the expressed sequence tag (EST) division of GenBank, using basic alignment research tool (BLAST)-N v2.1 (cutoff 70, no filters) (35). This version of the BLAST search algorithm is no longer available, but our results can be replicated by using Advanced BLAST v2.0 [advanced search options: ungapped alignment, expect set to 100-10,000, no filters and word size set to 6 (see BLAST help file http://www.ncbi.nlm.nih.gov/BLAST/blast_help.htm)]. The nonredundant Mus musculus GenBank database was initially searched with the tag sequences. If no appropriate matches were obtained, the entire nonredundant GenBank database was next searched, and if necessary, the tag sequence was submitted to dbEST, the nonredundant GenBank, EMBL, DDBJ EST database. Frequently, SAGE tag sequences matched more than one transcript. In these cases, genes matching the SAGE tags were identified using the published algorithm (36, 39). First, GenBank entries from mammalian organisms were identified (for searches which were not limited to M. musculus). Matches with nonmammalian sequences were excluded. Second, genomic, non-mRNA sequences were eliminated. Finally, search results were analyzed for multiple entries for the same gene, and the final match was checked to verify that the SAGE sequence flanked the 3'-most Nla III restriction endonuclease recognition site.

Immunoblotting. Tissue samples were isolated from neonatal rat kidney and heart (as a positive control). DRK23 cells overexpressing rat Kv2.1 were analyzed as a negative control. Samples were homogenized in 5-10 vol of 0.3 M sucrose, 10 mM NaPO₄ (pH 7.4) supplemented with protease inhibitor mix (Complete; Roche, Indianapolis, IN). After removal of debris and nuclei (3,000 g, 10 min), the supernatant was spun at 50,000 g for 1 h at 4°C to pellet a crude membrane fraction. Protein concentrations were determined by the bicinchoninic acid method (Pierce, Rockford, IL). Total protein was separated on 11% SDS polyacrylamide gels and transferred to polyvinylidene difluoride membranes. Membranes were blocked overnight with 5% nonfat dry milk in PBS plus 0.1% Tween and immunoblotted with a commercially available mouse monoclonal anti-Kv1.5 antibody (1:250 dilution; 1 h at room temperature; Transduction Laboratories, Lexington, KY) followed by horseradish peroxidase-conjugated secondary antibody (1:3,000; 1 h at room temperature; Amersham Pharmacia Biotech, Piscataway, NJ). Western blots were developed with the ECL-Plus detection system (Amersham Pharmacia, Arlington Heights, IL), as previously described (32).

	RESULTS

TOP ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION REFERENCES

Tag library generation. As described in MATERIALS AND METHODS, a pilot PCR reaction was performed to determine the optimum dilution for the input DNA by using serial dilutions (1:10-1:1,280) of recovered linker-ditag-linker constructs (Fig. 2). As expected, amplified linker-ditag-linker constructs were ~102 bp. A 1:40 dilution of the ligation reaction gave optimum results, with the intensity of the ethidium bromide-stained, amplified linker-ditag-linker being greater than the background products of ~80 (ligated linkers without kidney tag sequence) and 90 bp (ligated linkers with a single kidney tag sequence). A large-scale PCR reaction using the previously determined optimum conditions was subsequently performed, and the 102-bp linker-ditag-linker products were digested with Nla III to free ditags from the linkers. Ditags contain two 9- to 13-bp sequences, which have been derived from kidney transcripts and are joined in the sequence 5'-CATG(N)_22-26CATG-3'. The first half of the ditag represents sense sequence of an ROP-Es/+ kidney RNA, and the second half represents the antisense sequence of a different kidney transcript. Ditags migrated with 22- to 26-bp-molecular-size marker, whereas linkers migrated at 40 bp (Fig. 3). In our experience, purification of the 102-bp band by preparative PAGE, before Nla III treatment, improves digestion efficiency and ultimately the quality of the tag library. Eluted ditags were ligated together to generate concatemers (Fig. 4), which were ligated into pZero and transformed into bacteria as described in MATERIALS AND METHODS. Bacterial clones were screened for plasmids containing a kidney tag concatemer insert by direct PCR (Fig. 5) and concatemer sequence determined as described in MATERIALS AND METHODS.

View larger version (36K): [in this window] [in a new window]   Fig. 2.   Determination of optimum template dilution for large-scale PCR. An ethidium bromide-stained, nondenaturing 8% polyacrylamide gel demonstrates PCR products obtained from serial dilutions (top) of the linker-ditag-linker template. The 1:10 dilution is not shown. Amplified linker-ditag-linker migrate at 102 bp. Background bands of other sizes stain with equal or lower intensity. The 1:40 dilution of template had an optimal signal-to-noise ratio and was used for the large scale PCR (see MATERIALS AND METHODS).

View larger version (56K): [in this window] [in a new window]   Fig. 3.   Nla III digestion of the 102-bp linker-ditag-linker large scale PCR product yields 40-bp band (linkers) and a 26-bp ditag bands. The Nla III digestion products were analyzed by 16% PAGE, and the gel was stained with Sybr green-I. The 26-bp ditag band was excised from the gel and concatemerized as described in MATERIALS AND METHODS.

View larger version (45K): [in this window] [in a new window]   Fig. 4.   Sybr green-I stained, nondenaturing 5% PAGE of 26-bp ditags concatemerized as described in MATERIALS AND METHODS. Concatemers ranging in size from 600 to 1,200 bp were subcloned in the pZero cloning vector.

View larger version (22K): [in this window] [in a new window]   Fig. 5.   A representative 1.2% agarose gel electrophoresis of cloned kidney SAGE tag concatemers, which were generated by direct PCR of recombinant bacterial clones. The gel has been ethidium bromide stained. The PCR products vary in size from 700 to 1,400 bp, which include ~300 bp derived from the multiple cloning site.

Tag library analysis. Our library contained 3,868 sequence tags, which represented 1,575 unique mRNA transcripts. Ditags, containing identical tag sequences, were encountered 90 times. In each case, the tags were counted only once in tag abundance calculations. Such ditags are potentially produced by biased PCR, because the probability of any two tags being coupled in the same ditag is small even for abundant transcripts (36). The actual number of unique genes (1,575) identified by 3,868 kidney tags is consistent with previous predictions (37). Assuming that a cell contains 15,000 mRNA molecules (16), the 3,868-tag sequence library provides 1-fold coverage for mRNA molecules present at a minimum of 4 copies/cell. However, previous analyses of SAGE libraries have suggested that three- to fourfold coverage is necessary to identify all the transcripts at any level of abundance with high certainty. The 100 most abundant tags (6.3% of the unique transcripts represented in this expression library) represented 36% of isolated mRNAs. As expected, the frequency distribution pattern demonstrated that a minority of genes were responsible for the majority of transcripts, indicating that most genes are expressed at low abundance.

View larger version (25K): [in this window] [in a new window]   Fig. 6.   Kv1.5 expression was analyzed by immunoblotting as described in MATERIALS AND METHODS. Lysates from rat neonatal kidney (lanes 1 and 2), heart (lane 3, positive control) and DRK23 cells overexpressing Kv2.1 (lane 4, negative control) were resolved by SDS-PAGE. Left: molecular mass markers (in kDa); right arrow: position of Kv1.5.

	DISCUSSION

TOP ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION REFERENCES

Our data demonstrate the feasibility of generating expression libraries from kidney, and in aggregate, identify transcripts whose expression would be predicted by the physiology of the normal kidney. Genes regulating energy generation and solute transport dominated identified transcripts with known functions, whereas, genes normally expressed in inflammation or ongoing tissue remodeling or injury were not identified, as expected, in normal (wild-type) kidney. These data also demonstrate the potential power of expression profiling, which not only catalogues expressed genes but also allows an assessment of how the expression pattern of one gene relates to others. We suggest analysis of expression profiles by SAGE or other methodologies, generated from normal and diseased kidneys, should provide valuable new insights into renal disease pathogenesis and could identify critical regulators of renal disease initiation and progression.

SAGE provides a technique for high-throughput evaluation of gene expression and is based on two principles. First, a short sequence of 9-13 bp can be generated from DNA digestion with appropriate combinations of restriction endonucleases. This sequence tag contains sufficient information to uniquely identify a transcript, provided that it derived from a specific location within the mRNA sequence. Second, many transcript tags can be concatenated and sequenced, revealing the sequence of multiple tags simultaneously (36). A major strength of the technique is that it provides quantitative gene expression data (37), in contrast to differential display and subtraction hybridization. The quantitative aspect of SAGE has been validated in other laboratories (36-38). For example, quantitative hybridization experiments by Iyer and Struhl determined that the SUP44/RPS4 is expressed at ~75 copies in yeast. A yeast transcript profile, generated by SAGE, determined the abundance of this gene to be 63 copies/cell (37). Reproducibility and reliability of tag libraries generated from different amounts of input RNA from the same kidney sample have recently been reported (38). The abundance of the same tags correlated between the two libraries, suggesting this technique and its modifications for small samples will provide similar results in different laboratories.

The SAGE technique has limitations, however. First, a small number of transcripts would be predicted to lack a Nla III restriction site and would not be detected. Second, transcripts expressed at low abundance, or only in a fraction of the cell population, may not be reliably detected. However, SAGE has been used to analyze gene expression in yeast arrested at different phases of the cell cycle, yielding transcripts as low as 0.3 copies/cell (37). Therefore, the size of the SAGE library depends on the level of confidence desired for detecting low- abundance mRNA molecules. Monte Carlo simulations suggest the probability of identifying a single-copy transcript in a library containing 30,000 tags is between 72 and 97% (37). In addition, in an organ with as many different cell types as the kidney, a large tag library will need to be screened to identify transcripts unique for a subpopulation of cells. However, SAGE has been recently adapted for small samples and validated in microdissected kidney tubules containing ~50,000 cells (38). Third, differential mRNA expression may not be reflected at the protein level (15). Finally, identities of transcripts may remain unassigned (i.e., no match in either the nr or dbEST databases), but the percentage of these unknown transcripts will diminish as large-scale sequencing of the mouse and human genomes nears completion. Alternative high-throughput screening techniques, such as cDNA or oligonucleotide microarrays, may ultimately render SAGE obsolete. Recently, commercially available cDNA arrays on nylon membranes containing 588 unique murine genes were used to identify changes in gene expression in an in vitro model of branching morphogenesis (26). At present, though, microarray technology is not widely available, and only at considerable cost if robotically generated glass chips are used. In addition, the repertoire of mouse genes on chips and nylon membranes is limited, and SAGE analysis can result in discovery of novel genes. No matter which method is used to generate expression profiles, tools for exploring gene expression libraries are in their infancy. Data analysis of expression profiles ultimately will depend on integrating kidney mRNA libraries with external information resources and will require software development, such as the VectorArray application used to analyze the gene expression profiles during branching morphogenesis (26). Necessary bioinformatics tools include links between kidney genes identified in an expression profile and Genbank, links to user-friendly biological pathway databases, and access to databases that can identify functionally important nucleotide (e.g., regulatory elements) or protein (e.g., kinase domain) motifs.

Some of the most abundant genes expressed in ROP-Es/+ kidney were also identified in a previously published abundance profile, which was produced by sequencing randomly selected cDNA clones from microdissected renal tubules (34). Examples of common transcripts identified in this study and ours, included the androgen-regulated protein and cytochrome-c oxidase. Not surprisingly, SAGE reliably identified common tubular transcripts, validating the technique in kidney given that proximal tubules account for ~60% of renal mass. As a more comprehensive approach, SAGE should emerge as a preferable method for generating kidney expression libraries. The profile of abundant genes in our library (Table 1) also compares favorably to the abundant, nonmitochondrial genes recently identified by SAGE in C57BL/6J kidney (38). Of these 26 genes, we have identified 21 in similar abundance in our partial, ROP-Es/+ kidney SAGE library.

SAGE has been used to generate gene expression profiles from normal human skeletal muscle (39), pancreas (36), and Saccharomyces cerevisiae yeast (37). Integration of the yeast gene expression data with the S. cerevisiae genomic map allowed generation of chromosomal expression maps, which identified regions that were transcriptionally active and discovered genes that had not been predicted by sequence information alone (37). More recently, gene expression has been generated from in vitro models of normal and disease states, as well as from actual tissues, to gain insights into pathogenesis. SAGE expression libraries generated from cancerous and normal colonic epithelia and p53- and mock-transfected cells have compared 300,000 and 7,200 transcripts, respectively (29, 37). Fewer than 1% of the transcripts were expressed at significantly different levels in these comparative analyses. Because genes exhibiting the greatest difference in expression between normal and diseased tissue are likely to be the most relevant to disease processes (42), SAGE expression libraries have the potential to identify candidate pathways important in disease pathogenesis, novel diagnostic markers, and potential therapeutic targets.

Expression libraries are a powerful tool to apply to mechanisms of renal disease, particularly because available treatments are limited, only marginally effective, and rarely curative. Progressive kidney disease probably results from complex interactions of traits (both environmental and genetic) (31). Further understanding of chronic renal failure disease pathogenesis and development of new chronic renal failure therapies will require molecular tools, such as expression profiling, designed to dissect complex diseases that result from interactions of multiple genes. Expression profiling has already been applied to in vitro and in vivo models of kidney disease. Suppressive subtractive hybridization identified connective tissue growth factor in glucose-stimulated mesangial cells, a finding confirmed in streptozotocin-induced diabetic nephropathy (22). Differential display has been used to identify candidate diabetic nephropathy genes in the GK rat (25), and PCR-based subtractive hybridization has been used to analyze kidney gene profiles after 5/6 nephrectomy (41) and injection of anti-Thy1.1 antibody (23). In each case, transcripts were identified, whose induction in diseased kidney was previously unrecognized. Interestingly, the most abundant transcript identified in our ROP-Es/+ kidney library, mitochondrial cytochrome c-oxidase I, was shown to undergo a 4.5-fold induction in expression in 5/6 nephrectomized mice compared with sham-operated control animals (41). Previous reports also showed increased cytochrome oxidase I activity after unilateral nephrectomy (19). The compensatory increase in cytochrome-c oxidase I expression may be linked to kidney cell deletion through enhanced oxygen radical generation and apoptosis (40). However, the power of expression profiling for identification of mechanisms of kidney disease will really occur only when sophisticated analyses, such as Monte Carlo simulations and hierarchal cluster analysis, are applied (6, 11). These techniques will allow identification of key candidate genes or pathways regulating kidney disease pathogenesis by 1) assigning a precise biological pathway to a gene of generic function (e.g., kinase), 2) relating an unknown gene to characterized genes, and 3) revealing unexpected relationships between previously known gene(s) or pathways.

In summary, SAGE is a powerful tool that enables quantitative identification of differentially expressed genes. Alternative methods of gene expression analysis (differential display, chip microarray, subtractive hybridization) either lack the quantitative analytical capacity or require prohibitively expensive equipment. Our data confirm that it is feasible for a small laboratory to generate and analyze comprehensive mRNA expression libraries by using tissue from in vivo animal models and SAGE technology. On the basis of these studies, we believe analysis of kidney expression libraries could identify renal disease susceptibility genes and new pathogenic mechanisms by comparing mRNA expression profiles from diseased and control animals. Such an approach certainly should improve on simple models that appear sufficient to explain a pathogenic process, but fail to improve outcome due to the true complexity of disease mechanisms. Finally, the SAGE technique may also have potential implications for the diagnosis of human renal disease, if the methods can be modified to accommodate smaller tissue volumes from biopsy specimens. Expression profiles from human renal biopsies could be used to stage disease and to identify patients at risk for progression. Microarray technology already has been used to stratify human breast cancer tissues (28). Further understanding of kidney disease pathogenesis and development of new kidney disease therapies will require application of genetic and genomic tools, such as SAGE and other expression profiling methodologies, which are designed to dissect complex pathogenic mechanisms that result from interaction of multiple genes.