Cancers as Wounds that Do Not Heal: Differences and Similarities between Renal Regeneration/Repair and Renal Cell Carcinoma

Cancers have been described as wounds that do not heal, suggesting that the two share common features. By comparing microarray data from a model of renal regeneration and repair (RRR) with reported gene expression in renal cell carcinoma (RCC), we asked whether those two processes do, in fact, share molecular features and regulatory mechanisms. The majority (77%) of the genes expressed in RRR and RCC were concordantly regulated, whereas only 23% were discordant (i.e., changed in opposite directions). The orchestrated processes of regeneration, involving cell proliferation and immune response, were reflected in the concordant genes. The discordant gene signature revealed processes (e.g., morphogenesis and glycolysis) and pathways (e.g., hypoxia-inducible factor and insulin-like growth factor-I) that reflect the intrinsic pathologic nature of RCC. This is the first study that compares gene expression patterns in RCC and RRR. It does so, in particular, with relation to the hypothesis that RCC resembles the wound healing processes seen in RRR. However, careful attention to the genes that are regulated in the discordant direction provides new insights into the critical differences between renal carcinogenesis and wound healing. The observations reported here provide a conceptual framework for further efforts to understand the biology and to develop more effective diagnostic biomarkers and therapeutic strategies for renal tumors and renal ischemia. (Cancer Res 2006; 66(14): 7216-24)

Tissue regeneration and tumorigenesis are complex, adaptive processes controlled by cues from the host and from the tissue microenvironment. A variety of signals orchestrate the response to injury that results in regeneration and repair of a wound. Both tissue regeneration and carcinogenesis involve cell proliferation, survival, and migration that are controlled by growth factors and cytokines as well as inflammatory and angiogenic signals. Signals that promote cell proliferation, survival, and invasiveness derive from multiple cellular and extracellular sources in the microenvironment of wounds and cancer. Therefore, wounds and cancers share a number of phenotypic similarities in cellular behavior, signaling molecules, and gene expression. Haddow first recognized the similarities between wound healing and carcinogenesis, whereas Dvorak described cancer as wounds that do not heal (1, 2). Understanding the similarities between wounds and cancers may yield new insights into the malignant phenotype. Understanding the differences which relate to the “failure to heal” may provide insights into the loss of control in cancer, thereby providing the basis for novel diagnostic and therapeutic targets.

Microarray technology has allowed the characterization and comparison of global gene expression signatures of regenerating and malignant tissues. A microarray study comparing skin wounds and tumors provided molecular evidence that keratinocytes at wound margins have gene expression profiles similar to these of squamous cell carcinoma (3). Chang et al. studied changes in the global gene expression profiles of fibroblasts exposed to serum in vitro and compared those profiles with the publicly available gene expression data for numerous tumors (4, 5). That analysis suggested a similarity between the gene expression profile of fibroblasts, a cell type associated with the wound healing process, and that of the cancer. Furthermore, the serum response signature was predictive for survival of breast cancer patients. Our present study extends those observations to renal regeneration and renal carcinoma, and also for the first time examines comprehensively the differences between the two gene expression profiles as well as the similarities.

The kidney is a member of a restricted class of organs capable of regeneration and repair following damage events such as ischemic injury, a major cause of acute renal failure in both native (6) and transplanted organs (7). Clinically and biologically, ischemic acute renal failure is a complex but orderly continuum that, for simplification, can be separated into a series of four overlapping phases referred to as “initiation” (renal blood flow and cellular ATP decrease), “extension” (a prolonged hypoxia and continued production and release of inflammatory chemokines and cytokines after acute ischemia ceases), “maintenance,” (some cells undergo apoptosis whereas others proliferate, acquire the ability to migrate, and synthesize extracellular matrix proteins, which help reestablish and maintain the structural integrity of cells and tubules), and “recovery” (cellular function improves slowly, blood flow returns to normal or near normal, and epithelial cells establish intracellular and intercellular homeostasis; ref 8).

Renal cell carcinoma (RCC), which accounts for 3% of all adult male malignancies in the U.S. (9), is a clinicopathologically heterogeneous disease that includes several histologically distinct cellular subtypes (10). RCC is thought to originate in proximal renal tubules most of the time and in distal tubules occasionally (11). Five human genes are associated with predisposition to RCC: von Hippel-Lindau (VHL), met proto-oncogene (MET), fumarate hydratase (FH), Birt-Hogg-Dube (BHD/FLCN), and hyperparathyroidism 2 (HRPT2; ref. 12). RCC could develop following chronic renal regeneration and repair (RRR) in individuals with polycystic kidney disease or in renal allografts (13, 14).

Our study tests the hypothesis that there are patterns of gene expression common to RRR and RCC. We used a mouse model of ischemia/reperfusion (in which the left renal artery was ligated transiently) to characterize gene expression changes at several time points during the first 2 weeks of RRR. Differential gene expression associated with RRR was then compared qualitatively with differential gene expression reported in the literature for human RCC. The results revealed two distinct genomic signatures: (a) a large group of genes (which we will call “concordant”) that are differentially expressed in the same direction in RRR and RCC, and (b) a smaller divergent group (“discordant”) that are differentially expressed in opposite directions in RRR and RCC. We analyzed concordant and discordant differentially expressed genes for biological significance by comparing categories and functional pathways. The concordant gene expression signature qualitatively reflects the normal regenerative phenotype, and the discordant signature provides new insight into critical differences between the malignancies and processes of tissue repair. The results could potentially lead to the development of more effective diagnostic and therapeutic strategies for cancer and for wound healing.

Animals. Five-week-old C57BL/6 female mice (20 g) were obtained from the National Cancer Institute. All animals had free access to water and food. Animal care and experiments were done according to protocols approved by the Animal Care and Use Committee of the National Cancer Institute.

Ischemia-reperfusion model. Regeneration was induced by a modification of the renal warm ischemia method (15). Mice were anesthetized with ketamine, xylazine, and acepromazine and placed on a heating table at 37°C to maintain body temperature. A left unilateral flank incision was made to allow exposure of the left kidney and renal artery. A nontraumatic vascular clamp was placed across the renal artery for 50 minutes. The mice were kept anesthetized during that time, with temporary closure of the abdomen. After the ischemic interval, the kidney was inspected for restoration of blood flow, and 1 mL of prewarmed (37°C) normal saline was instilled into the abdominal cavity. The abdomen was closed with wound clips (Roboz Surgical Instrument Co., Inc, RS-9262), and the animals were allowed to recover in a 37°C incubator. After the desired period of reperfusion (6-12 hours or 1, 2, 5, 7, or 14 days), the animals were anesthetized, and both kidneys were rapidly excised by midline abdominal incision. For microarray studies, the kidneys were flash-frozen in liquid nitrogen and stored at −70°C. Normal and ischemic kidneys were removed, processed, and frozen in an identical manner. For histologic studies, the kidneys were bivalved in the coronal plane and fixed in formalin (10%).

Immunohistochemistry. Fixed and paraffin-embedded tissue specimens were deparaffinized, rehydrated, subjected to antigen unmasking (16), and treated to block nonspecific staining. For the latter procedure, sections were incubated for 20 minutes at 24°C with 1% H₂O₂ in methanol, followed by blocking for 30 minutes with 5% normal horse serum in PBS. Polyclonal antibody against Ki67 (NCL-Ki67p; Novocastra Labs, New Castle upon Tyne, United Kingdom) or mouse glucose transporter (Glut-1; Alpha Diagnostic Int., San Antonio, TX) was added (1:1,000 dilution) for 16 hours at 4°C, followed by incubation for 30 minutes at room temperature with biotinylated secondary goat anti-rabbit IgG and incubation for 30 minutes with avidin-biotin peroxidase conjugate (1:50 dilution; Vectastain Elite Universal Kit; Vector Laboratories, Burlingame, CA). Color was developed using Vector Laboratories 3,3-diaminobenzidine kit for 10 minutes, followed by counterstaining with Mayer's hematoxylin. Negative controls were done with nonimmune serum or PBS. Three investigators evaluated the immunohistochemistry independently.

Microarray procedures. Mouse cDNA microarrays (NIH/NCI GEM2) containing 9,596 cDNA spots from the Integrated Molecular Analysis of Genomes and their Expression consortium were used to quantitate mRNA expression in the kidney samples. A reference sample consisting of an equal mixture of six normal mouse tissues (brain, heart, kidney, liver, lung, and spleen) was used in the competitive hybridization experiments. For the reference sample, 50 μg of total RNA was reverse transcribed using an oligo(dT)-primer. For experimental samples, 3.0 μg of polyadenylated RNA from whole kidney was reverse-transcribed using an oligo(dT)-primer. The labeling and remaining hybridization procedures have been described previously (17). Gene expression data are presented in their entirety in supplemental online material at the authors' web site.

Quantitative real-time reverse transcription-PCR. RNA was isolated using Trizol Reagent (Invitrogen, CA). Total RNA (1 μg) was reverse transcribed in a volume of 50 μL. Five microliters of the resulting solution was then used for PCR according to the manufacturer's instructions (Applied Biosystems, Inc., Foster City, CA). Gene expression for IGFBP1, IGFBP3, CTGF, AKT, FRAP, MYC, NF-κB, HK1, and SIRT7 were quantified relative to the expression level of ribosomal 18s. PHD1, PHD2, and PHD3 were quantified relative to the expression level of filamin B (β-actin binding protein 278; FLNB). All probes were purchased from Applied Biosystems. Normalized data are presented as fold difference in log₂ gene expression.

Statistical analysis of microarray data. The experimental RNA was labeled with Cy3 (green) and the reference pool with Cy5 (red). Two different batches of reference were used for the two experiments. Log ratios used base 2 logarithms. There were 9,984 spots on each array, but 388 had Clone id = 0 and were excluded. Spots were filtered out if the log intensity in either channel was below two standard deviations from the mean for that channel on that array. For cluster analysis, genes present (not filtered) in at least 60% of the samples were included. Each array was normalized using a nonlinear Lowess smoother to provide intensity level–dependent normalization.¹²

The data were analyzed using principal component analysis.

Analysis and curation of pathways. Publicly available literature from 1966 to mid-2003 was surveyed using PubMed. The survey was complemented by information from publicly available databases, including OMIM, Entrez Gene (LocusLink), KEGG, GeneCard, MYC Cancer Gene,¹³

p53,¹⁴ Panomics,¹⁵ and Gilmore's Rel/NF-κB transcription factors.¹⁶ The survey was conducted with the goal of cataloguing genes reported to be expressed differentially. HUGO gene names were used for comparisons across databases. Only genes that were printed on the GEM2 microarray were considered for further analysis. If conflicting reports on gene expression were present in the literature, the genes were labeled “conflict.”

MatchMiner¹⁷

and SOURCE¹⁸ were used to translate among different types of identifiers for comparative analyses. The statistical significance of concordance or discordance in relative enrichment of gene subgroups was determined using a χ² test (Table 1; Supplemental Table S7). A 2 × 2 contingency table is shown below:

The P value for this 2 × 2 table was calculated using the statistical package R.

Ontology. GoMiner¹⁹

and EASE²⁰ analyses were done on the lists of differntially expressed genes (18). Enrichment of categories was determined using GoMiner (19), and P value thresholds were determined from EASE scores (18).

To understand RRR and its relationship to the gene expression in RCC, we established a murine model of warm unilateral renal ischemia and subsequent RRR (Fig. 1B), then we did an extensive five-step analysis (Fig. 1A): (a) histopathologic verification of RRR (Fig. 2A-C); (b) gene expression profiling of the regenerating kidneys relative to normal kidneys at different time points after ischemia. The resulting RRR gene expression data set served as the foundation for further analysis and comparison with RCC; (c) biological interpretation of the RRR differential gene expression by literature mining, gene ontology (GO) analysis, and pathway analysis; (d) comparative analysis of the differential gene expression patterns of RRR and RCC relative to normal renal tissue, resulting in the identification of concordant and discordant genes; and (e) bioinformatic analysis that included pathway analysis, GO analysis, and literature mining of the concordant and discordant genes. The results are presented in Figs. 1-3 and Tables 1,Table 2-3. Additional data (Supplemental Tables S4-9; Supplemental Fig. S4) and accompanying text are available in the online supplemental materials.

The histopathology of RRR. Ischemic injury was induced in the left kidneys of female mice by restricting blood flow for 50 minutes with a vascular clamp (see Materials and Methods for details). Following reperfusion, the kidneys were allowed to recover for 0.5, 1, 2, 5, 7, and 14 days before harvesting (Fig. 1B). Apoptotic cells were observed in the outer medulla within 12 hours of reperfusion, and the number of apoptotic cells increased for ∼24 hours (data not shown). Histologic markers of ischemia were monitored, with the results shown in Fig. 2A-C. More than half of the cortical tubules stained for hypoxia-inducible factor 1α (HIF1α)-regulated glucose transporter-1 (Glut-1/Slc2A1) after 1 day (Fig. 2C). Acute tubular necrosis with complete loss of epithelium within individual tubules was observed within the first 12 to 24 hours. Some tubules had cells with enlarged, reactive, hyperchromatic nuclei and prominent nucleoli (Fig. 2A and B). Tubular epithelial cells stained for proliferation marker MiB-1. When that staining peaked at about 48 hours, most of the tubules and at least 50% of the tubular epithelial cells were positive for MiB-1 (Fig. 2B). After 2 weeks, most tubules were histologically normal, and there were only rare examples of degenerative or regenerative change (Fig. 2B). Those observations are consistent with previous studies of renal injury, regeneration, and recovery (8).

Differential gene expression in RRR. Transcript expression was analyzed using a cDNA microarray with 9,596 spots (corresponding to 5,796 murine genes) and RNA samples of normal (day 0), ischemic (50 minutes), and reperfused mouse kidney harvested 1, 2, 5 and 14 days postinjury. Differentially expressed microarray spots (1,675; P < 0.05), representing 1,325 genes, clustered the kidney samples into three groups. The first included samples of normal and ischemic kidney (“baseline” and 50 minutes ischemic); the second included the samples from the 1st and 2nd days postinjury (“early”); the third included the samples from the 5th and 14th days postinjury (“late”). The average differential expression (RRR relative to normal and 50-minute ischemic kidneys) was calculated for each gene. By principal component analysis, individual data points were highly reproducible because repeat measurements (4-16 arrays per time point) clustered in the same pattern (Fig. 3A).

Relative to the normal and 50-minute ischemic kidneys, the 1,325 RRR genes fell into three groups in their temporal patterns of differential expression. The first included 323 genes differentially expressed continuously during RRR [Fig. 3B, “continuous” or (*)]. Included were 189 up-regulated and 134 down-regulated genes. The second group included 629 “early” genes (336 up-regulated and 293 down-regulated) differentially expressed only during the first 2 days after injury [Fig. 3B, “early” or (A)]. The third included 373 “late” genes (227 up-regulated, 96 down-regulated) differentially expressed 5 and 14 days after injury [Fig. 3B, “late” or (B)]. A complete list of the genes differentially expressed during RRR can be found in Supplemental Table S4. The data were validated by reverse-transcription quantitative PCR analysis of 10 genes (Supplemental Fig. S4) and by mining the literature for an additional 81 genes out of 91 that were reported by others (Supplemental Table S4).

Comparison of genes differentially expressed in RRR and RCC. An extensive literature survey of gene expression data for human RCC identified 2,815 genes reported to be differentially expressed in RCC relative to normal human kidney.²¹

For 984 of those genes, a mouse orthologue was present on the microarray used in this study, and 361 of those genes were differentially expressed in both RRR and RCC (Fig. 1C and D; Supplemental Table S4). Of those 361 genes, 278 (77%) were up-regulated in RRR and RCC or down-regulated in both RRR and RCC. Those genes are referred to here as “concordant” genes. The remaining 83 genes (23%) were differentially expressed in opposite directions in RRR and RCC. Those are referred to here as “discordant” (Fig. 1C and D; Table 1). The probability of observing those percentages by chance would be extremely low under the null hypothesis that the RRR and RCC phenotypes are unrelated (P value, 2.2 × 10⁻¹⁶ by Fisher's exact test; Fig. 1C and D; Table 1). Of the concordant genes, 209 were up-regulated. Included were VCAM1, ICAM1, and MYC. The remaining 69 concordant genes were down-regulated (P < 0.001; Fig. 1C and D; Supplemental Table S4). Thirty discordant genes were up-regulated in RRR and down-regulated in RCC (e.g., CTGF, THBS1, and SMC1L1; Fig. 1C and D; Table 3; Supplemental Table S4). Fifty-three discordant genes were down-regulated in RRR and up-regulated in RCC (e.g., IGFBP1, IGFBP3, PHD2/EGLN1, and HK1; Fig. 1C and D; Table 3; Supplemental Table S4).

Pathway and gene ontology analyses of “concordant” and “discordant” genes. We next tested the association between the concordantly and discordantly expressed genes and pathways presumed to be involved in RRR and/or RCC. Concordant genes were significantly enriched (P < 0.05) in the VHL, MYC, p53, and NF-κB pathways and the hypoxia-regulated category (Table 1).²²

Discordant genes were significantly enriched in the VHL, hypoxia, HIF (HRE), insulin-like growth factor (IGF-I), and p53 pathways. The NF-κB pathway was significantly enriched with concordant, but not discordant genes. The HIF and IGF-I pathways were significantly enriched with discordant, but not concordant, genes. The discordant genes in the IGF-I pathway included CTCG, CYR61, IGFBP1, IGFBP3, TASCTD2, VEGFA, and COX6C. The discordant genes in the HIF pathway included HK1, IGFBP1, IGFBP3, MMP2, PGK1, EGLN1, and VEGFA (Table 1; Supplemental Table S5b).

Gene ontology²³

categories significantly enriched in concordant genes are listed in Table 2 and are listed in detail in Supplemental Tables S5b and S6. Among the gene categories for concordant genes, which were mostly up-regulated, are such biological processes and functions as immune response, proliferation, cell growth, translation (ribosome biogenesis), metabolism, and extracellular matrix structural constituent (Table 2; Supplemental Tables S5b and S6). When the same GO analysis method (i.e., P < 0.05) was used for the discordant genes, a different set of GO categories was found, and among those were IGF binding, heparin binding, extracellular space, angiogenesis, regulation of cell growth, and morphogenesis of renal tissue (Table 2; Supplemental Tables S5b and S6). Only a small number of GO categories were enriched for both concordant and discordant genes (Table 2; Supplemental Tables S5b and S6).

Based on our earlier pathway analysis of the concordant and discordant genes (Table 1; Supplemental Table S4), we next analyzed the genes in the significant pathways (e.g., the hypoxia pathway) for enrichment of GO categories (Supplemental Table S9). Interestingly, the concordant genes in the hypoxia pathway were enriched for the category of enzyme inhibitor activity, whereas discordant genes in the hypoxia, HIF, and IGF-I pathways were enriched for gene functions related to cell growth.

Based on the common biological characteristics of cancer, and extensive analysis of the literature, we also categorized the discordant genes on a nonprobabilistic, gene-by-gene basis (Table 3; Supplemental Table S8).

Earlier studies suggested that the complexities of the signaling and regulatory pathways involved in cancer present a significant barrier to understanding, preventing, and/or treating cancer (20). However, cancers share many features in common with tissue regeneration, including immune response, cell proliferation, cell migration, tissue remodeling, and cell death. Our results support the proposed hypothesis that cancer is an aberrancy of the physiologic processes of wound healing, and reveal for the first time, two distinct qualitative gene expression signatures, a concordant signature and a discordant signature that distinguishes RCC from RRR. Of the 361 genes that we found to be differentially expressed in both RRR and RCC, the majority (77%) were concordant, and the remaining 23% were discordant. Given those numbers, we can confidently reject the null hypothesis that there is no relationship in differential expression between RRR and RCC (P, 2.2 × 10⁻¹⁶ by Fisher's exact test). The biological functions of the concordant genes seem to reflect cancer as wounds (e.g., cell proliferation and immune response). On the other hand, the discordant signature seems to reflect the critical differences between malignancies and the processes of tissue repair (e.g., regulation of cell growth, morphogenesis, and angiogenesis). Those gene expression patterns may yield new insight into pathways, functions, and cellular locations of proteins that play multifaceted roles in wound healing and/or carcinogenesis.

Clinical RRR is relatively common, but there is no ethical possibility of obtaining biopsy specimens at different times during the process. In recent years, mouse (and other) model systems have shed new light on the nature and treatment of human RRR. Physiologic, pharmacologic, global gene expression, and gene inactivation studies have been included (21, 22). Therefore, we chose a mouse model (unilateral renal ischemia) to assess changes in gene expression during RRR. The predominant consequences of renal injury in the model include proximal tubule necrosis, as well as apoptotic death of a minority of the cells. The reversal of those changes coincides with the reestablishment of the normal renal epithelial barrier as regeneration of cells relines the denuded tubules (23). Our results for the RRR model are in accord with the expected RRR processes and further suggest three distinct temporal patterns of differential gene expression: continuous (days 1, 2, 5, and 14), early (days 1 and 2), and late (days 5 and 14; Fig. 3A and B). Our GO analysis of the differential gene expression suggests that metabolic and catabolic processes, as well as response to injury, are involved throughout renal recovery, whereas the first 2 days following ischemia are enriched with regeneration processes, and the late RRR stage is characterized by immune response (Supplemental Tables S5a-b).

To compare RRR and RCC, we analyzed differential gene expression in RRR and compared it with RCC differential gene expression. That analysis required integration of data from different organisms, tissue pathologies, methods, and authors (24). Despite the heterogeneity of cell populations, transcriptional profiling of bulk tumors and wounds has yielded significant insights, such as those in this study (25, 26).

The comparison of mouse RRR with human RCC was accomplished by using the corresponding normal tissue in each original study as a reference point, and thus, the comparison was indirect (i.e., not RRR versus RCC). To reduce the noise in the study, the differential expression was catalogued and compared only qualitatively (not quantitatively), as expressed up or down from normal tissue (Supplemental Table S4). The feasibility of that comparison was supported by the fact that both RCC and RRR are predominantly proximal tubular processes, and proximal tubules make up the bulk of the kidney (11, 27). Moreover, comparative analysis of the literature is supported by a comparison of the RRR literature with our experimental RRR data set. Of the 91 genes appearing in both lists, 89% were differentially expressed in full agreement (up or down), despite the difference in organism (human versus mouse) and methods (Supplemental Table S4). Further methodologic considerations are addressed in the supplemental material.

Concordant genes comprised the majority (77%) of the 361 genes we identified as differentially changed in both RRR and RCC. Those genes and their pathways reflect the common mechanisms of cell proliferation, growth, metabolism, and defense that are pertinent to both RRR and RCC. For example, our GO analysis of the differential concordant gene expression suggests, in agreement with the literature, a significant enrichment of categories as DNA replication [the highly conserved minichromosome maintenance genes MCM2, MCM3, MCM4, and MCM7 and the human mismatch repair gene mutS homologue 2 (MSH2), cell adhesion (e.g., ICAM1 and VCAM1), and through 21 up-regulated concordant immune response genes (Supplemental Table S4; ref. 28–30)].

Nearly a fourth (23%) of the genes differentially expressed in RRR and RCC were discordant, i.e., differentially expressed in opposite directions. Although differences in some of those genes may be due to extraneous factors (including different methodologies, species differences, or chance), the functions of the genes support the conclusion that many of them do differentiate the RRR and RCC processes from each other. Our GO analysis indicated that 95% of the GO categories for the discordant genes are distinctly different from those predicted for the concordant genes (Table 2; Supplemental Table S5b; Fig. 1A-D). Including categories such as IGF binding, organic cation transporter activity, heparin binding, angiogenesis, regulation of cell growth, organogenesis, and morphogenesis. Interestingly, alterations in morphogenesis have been cited as a clear characteristic of cancer (Supplemental Table S5b; ref. 31).

Another characteristic of RCC are the alterations in glycolysis. Fast-growing tumors consume large amounts of energy in the form of ATP. In hypoxic tumors, ATP is partially generated by anaerobic glycolysis, even though that pathway is far less efficient than aerobic glycolysis (32). The glycolytic genes differentially expressed in both RRR and RCC are interesting. For example, hexokinase 1 (HK1), which carries out the essential first step in the glycolytic pathway, is down-regulated early in RRR and is up-regulated in RCC (Table 3; Supplemental Table S4). In the kidney, HK1 is expressed in the proximal renal tubule and is regulated by HIF and possibly by p53 (33–37). Phosphoglycerate kinase 1 (PGK1) is down-regulated early in RRR and is up-regulated in RCC. PGK1, which is expressed in the collecting duct, is regulated by HIF and possibly by NF-κB and MYC (35, 37, 38). Solute carrier family 16-member 7 (SLC16A7/MCT2) is up-regulated in RCC and down-regulated in RRR. Those observations are consistent with increased glycolysis in cancer cells that rely to a greater extent on glycolytic pathways than do normal cells (39, 40).

Previous studies have implicated altered processes and pathways as associated with RCC pathogenesis. Included are processes involving morphogenesis and glycolysis and the HIF-VHL and the IGF-I pathways. The discordant genes include genes that may play a critical role in those processes and pathways (12, 37, 41, 42).

The HIF-VHL pathway. Seventeen HIF-responsive or HIF-associated genes are differentially expressed during RRR (P < 0.05), and seven of those are differentially expressed in the opposite direction during RCC (P < 0.05). Six of the seven discordant HIF-responsive genes have been reported to be hypoxia-induced and are up-regulated in RCC. Their down-regulation in RRR must signify other control mechanisms in normal regeneration that are not operative in RCC. Among the biological functions of those genes are glycolysis (HK1 and PGK1) and the IGF-I pathway (IGFBP1 and IGFBP3; Tables 1 and 2; Supplemental Table S4).

Regulation of the HIF1α transcription factor in RCC is complex. For example, FK506 binding protein 12-rapamycin associated protein 1 (FRAP1/MTOR) is down-regulated continuously during RRR but is up-regulated in RCC (Supplemental Table S4; ref. 43), suggesting that mTOR signaling increases the translation of HIF1α in RCC but not in RRR (44).

Interestingly, prolyl hydroxylases PHD2/EGLN1 and PHD3/EGLN3 are up-regulated during RCC (30, 38) and down-regulated during RRR, together with PHD1/EGLN2 (Supplemental Table S4; Supplemental Fig. S4). In RRR, the down-regulation of PHD1, PHD2, and PHD3 is likely to prolong the half-life of HIF1α protein in the early hours following ischemia (45). In RCC, the induction of PHD2 and PHD3 is a consequence of a dysfunctional negative feedback loop. The PHD2 and PHD3 genes are induced by HIF1α which is continuously up-regulated in RCC. Solid tumors are often hypoxic and mutated in the VHL gene. Therefore, the proline-hydroxylated HIF1α cannot be mediated for oxygen-dependent ubiquitination. Thus, in RCC, the up-regulation of PHD2 and PHD3 cannot affect the already dysfunctional VHL-HIF pathway (46–48). Further examples of discordant genes involved in the HIF-VHL pathway are given in the supplemental material.

Our GO analysis of the discordant genes in the HIF-VHL pathway indicated that they are significantly enriched with biological process of glycolysis, regulation of cell growth, and IGF binding. Those biological processes are in agreement with the RCC literature (Supplemental Table S9; ref. 37, 38).

The IGF-I pathway. Several IGF-I pathway genes were differentially expressed during RRR (e.g., CTGF/IGFBP8, IGFBP1, IGFBP3, and IGFBP4; Table 2; Supplemental Table S5b). In contrast to RRR, IGFBP1 and IGFBP3 are up-regulated during RCC. The bioavailability of the IGFs is influenced by the concentrations of specific IGFBPs. In a different physiologic context, IGFBPs could either increase or decrease IGF signaling. This complexity is poorly understood; it could well be that IGFBP3 up-regulation in RCC prolongs the half-lives of the IGFs. Alternatively, IGFBPs may compete with receptors for free IGFs and IGF-II and thus disrupt these pathways (49); or IGFBPs may serve some unknown functions in RCC (for IGFBP8/CTGF—see supplemental material). The discordant genes regulated by the IGF-I pathway were enriched with GO categories such as regulation of cell growth, angiogenesis, morphogenesis/organogenesis, heparin binding, and IGF binding (Supplemental Table S9).

Prospective and future directions. We have identified three temporally different patterns of differential gene expression in RRR: early, late, and continuous. RRR can be viewed as a complex, ordered process involving tissue regeneration and repair. Comparison of the RRR gene expression profile with that of RCC reported in the literature reveals two expression signatures that strongly support the proposed hypothesis that cancers bear similarity to wounds: a predominant concordant signature and a lesser discordant one. The biological functions of the concordant genes indeed support the view of “cancer as a wound” and include genes and pathways that are tuned to maintain the regenerative and repair processes. The discordant signature, however, points to processes, pathways (e.g., HIF and IGF-I), and genes that differentiate cancer from wounds. Those observations provide a conceptual framework for further efforts to understand the biology of RCC and RRR. They also provide information for the development of more effective diagnostic biomarkers and therapeutic strategies for renal tumors, as well as strategies for improving recovery from renal ischemia without promoting RCC.

Funding: This research was supported by the Intramural Research Program of the NIH, National Cancer Institute, Center for Cancer Research.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

Microarray analyses were done using BRB ArrayTools developed by Dr. R. Simon and A. Peng Lam. We gratefully acknowledge Drs. H.F. Dvorak for his comments, L.M. Staudt for advice in microarray technology, A.M. Michalowska and R. Simon for help in biostatistics, B.R. Zeeberg for advice in bioinformatics, H. Cao for web site development, A.R. Kane for graphics support, and L.K. Fleming, S.F. Goldberg, and M. Sander for editing this manuscript.