Global analysis of metastasis-associated gene expression in primary cultures from clinical specimens of clear-cell renal-cell carcinoma

Abstract

Metastatic clear-cell renal-cell carcinoma (ccRCC) has a poor prognosis and unpredictable course, and there are no molecular markers that reliably predict ccRCC metastasis. In this study, ccRCC specimens from 84 patients were directly cultured in vitro. Primary cultures from 38 of 94 specimens contained more than 90% tumor cells at the fourth passage. After identification by immunostaining, the primary cultures of metastatic and nonmetastatic ccRCC specimens from the age- and gender-matched patients were subjected to cDNA microarray assays. A total of 842 differentially expressed genes with a FDR (false discovery rate) of 4.79% were identified. Pathway analysis and co-occurrence with “cancer”, “metastasis” and “invasion” in the literature annotations functionally enriched the 842 genes and provided an indication of the reliability of our microarray assays. Novel genes associated with metastasis were selected based on protein-protein interactions between 205 differentially expressed genes that co-occurred with “metastasis” and those that did not co-occur with “metastasis” on Medline, and the results of co-expression analysis between the co-occurred genes and unpublished genes. FSTL1, AV722783, SLC15A1, DDX17, ORC2L and PKMYT1 were found to be potential ccRCC metastasis-associated novel genes, according to expression patterns in cultures and tumor tissues. Interestingly, the upregulated genes (CAV1, PKMYT1 and ORC2L) were also upregulated and the downregulated genes (FSTL1, GSTM3, CYR61, SLC15A1 and AV722783) were also downregulated in the primary ccRCC specimens compared with expression in adjacent renal tissues in 37 patients. This study has identified new candidate biomarkers and targets for the early diagnosis and treatment of ccRCC metastasis. © 2008 Wiley-Liss, Inc.

Renal-cell carcinoma (RCC) accounts for approximately 90−95% of neoplasms in the kidney. RCC is more common in men than in women (2:1), and it most often occurs in patients aged 50−70 years. The incidence of RCC has been increasing. One-third of patients present with metastatic disease and have a median survival of 7−11 months and a 5-year survival of 0−10%.1, 2 RCC is a pathologically heterogeneous disease, and can be subdivided into clear, papillary, granular, spindle, and mixed cell variants based on cytoplasmic features. Clear-cell RCC (ccRCC) is the most common type of RCC (approximately 80%) and accounts for most cases of metastatic disease.3 There have been no significant improvements in the mortality rate of ccRCC, most likely because currently available therapies for metastatic disease are relatively ineffective. A full understanding of the molecular genetics and signaling pathways involved in the metastatic process of ccRCC is important for early detection of metastasis and the development of innovative treatment options.

The prognostic and/or therapeutic potential of several molecular markers of metastatic RCC have been investigated. Carbonic anhydrase IX (CA9), caveolin-1 (CAV1), B7-H1/PD-1, B7-H4, vimentin, CD10, alpha B crystalline (ABC), survivin, vascular cell-adhesion molecule-1, hypoxia-inducible protein 2, adipose differentiation-related protein, cyclin B1, N-cadherin, kallikreins, minichromosome maintenance 2, gemnin and Ki67 have been suggested to be associated with metastasis and poor prognosis of ccRCC.4–19 Immunohistochemistry analysis has most frequently been used to evaluate the diagnostic and prognostic significance of the potential markers in limited numbers of patients with RCC. Unfortunately, there are no widely accepted molecular biomarkers for ccRCC aggressiveness. One of the most likely candidates, CA9, which is a von Hippel-Lindau (VHL)-mediated enzyme expressed in most (>85%) RCC samples, is not an independent prognostic marker. A recent study showed that low CA9 expression levels were not associated with RCC death after adjusting for nuclear grade or coagulative tumor necrosis; furthermore, CA9 expression was reported in extra-renal organs.20

Global gene-expression profiling by cDNA microarray assays can provide insights into the underlying molecular mechanisms of RCC, and lead to the identification of biomarkers for more accurate diagnosis and prognosis, and which could be drug targets for effective therapies. Differences in global gene expression between fresh-frozen RCC tissues and normal renal tissues have been described.21–26 Gene expression changes that resulted in aggressive behavior or metastatic potential have been identified in the primary RCC. A so-called “metastasis signature” derived from primary RCC with different prognosis could be used to classify tumors with and without metastases at the time of surgery.21, 22 However, published prognostic gene signatures in RCC have few genes in common.

Clinical RCC specimens usually contain some necrotic tissues and nontumor cells such as tumor-infiltrating leukocytes, endothelial cells and fibroblasts, which complicate the global gene-expression analysis. Many microarray assays are necessary to minimize the effects of the necrotic tissues and non-tumor cells in the tumor tissues on gene-expression profiling. To overcome this problem, fresh tumor specimens may be adapted to grow in vitro, as primary cell cultures, to provide homogeneous cellular material for gene-expression profiling. Short-term cultures allow enrichment of RCC cell types from tumor specimens. RCC primary cultures retain the proteomic profile of the corresponding tumor tissues.27, 28 Established RCC cell lines are not representative of clinical RCC specimens because the gene-expression profiles of RCC cell lines are different from those of RCC specimens.29

In this study, we isolated metastatic and nonmetastatic ccRCC cells from fresh specimens by transient primary culture. After removing the necrotic tissue and nontumor cells, we used the metastatic and nonmetastatic ccRCC primary cultures from specimens of the age- and gender-matched patients for global gene-expression analysis. A series of data-mining assays and validation methods were used to identify novel molecules related to ccRCC metastasis. To our knowledge, this is the first study to use the primary cultures of metastatic and nonmetastatic ccRCC specimens for global gene-expression analysis.

Material and methods

Tissue specimen

Tissue specimens were obtained from the Departments of Urology and Orthopedics at the 1st and 2nd Affiliated Hospitals of the Second Military Medical University from January 2004 to October 2007, according to a protocol that was approved by the Institutional Review Board. All hematoxylin & eosin (H&E) slides were reviewed by 1 investigator (Y.Y) and histologically unequivocal cases were included. Specimens from 169 Chinese patients with ccRCC were collected immediately after surgery. We obtained metastatic ccRCC specimens from 14 of the 169 patients, including 7 with bone metastases, 2 with liver metastases and 5 with lymph node metastases near the inferior vena and/or renal hilus. Paired metastatic and primary tumor specimens were harvested from 10 patients. Specimens of bone metastasis without paired primary tumors were harvested from 4 patients who had received nephrectomy several years previously. Primary tumor and paired adjacent kidney tissues were harvested from 155 patients. Specimens from 84 patients with ccRCC were directly subjected to primary cell culture. Data on the specimens and primary cultures are listed in Table I. All specimens were immediately frozen in liquid nitrogen and stored at −80°C.

Table I. Clear Cell Renal Cell Carcinoma Samples

	Frozen	Cultured	Growth in vitro
			<4 passages	≥4 passages
Mean age (range, yr)	53.9 (28−77)	51.4 (29−77)	50.3 (29−72)	47.4 (29−65)
Gender (male/female)	107/62	57/27	51/25	24/14
Stage (tumor-node-metastasis)
I(T1, N0−x, M0)	94	41	38	15
II(T2, N0−x, M0)	46	22	20	9
III(T3, N0−1−x, Mx)	20	12	11	7
IV(T1b−3b, N0−1−x, M1)	9	9	8	7
Primary tissue samples	165	80	72	29
Metastatic tissue samples	14	14	13	9
Samples for validation
Stage (tumor-node-metastasis)
I(T1, N0, M0)	21
II(T2, N0, M0)	9
III(T3, Nx, Mx)	7
IV(T1b, Nx, M1)	3
IV(T3b, N0, M1)	4

Primary cell cultures

Fresh ccRCC tissues were immediately immersed in cold RPMI 1640 medium (GIBCO, Invitrogen, Grand Island, NY) containing 1% penicillin/streptomycin (GIBCO), 0.5% glutamine (GIBCO) and kept at 4°C for 30 min. Following removal of the surrounding fat, blood vessels, and necrotic tissue, tumor tissue was mechanically minced with scissors in a sterile manner. Dissociated tumor cells were washed twice with phosphate-buffered saline (PBS), and then resuspended in RPMI 1640 (Life Technologies, Inc., Grand Island, NY) supplemented with 20% fetal calf serum (FCS; GIBCO), 2 mmol/L L-glutamine, 100 units/mL penicillin and 100 μg/mL streptomycin, and incubated at 37°C, 5% CO₂, 95% humidity. The cells remained in culture until sufficient confluence for a first tissue culture passage was reached. Aliquots of cells were cryopreserved in 90% FCS, 10% dimethyl sulfoxide (Sigma, Saint Louis, MO) and stored in liquid nitrogen after 1−2 passages. As a variation in phenotype can occur at higher passages,28–30 we selected the ccRCC cultures at the fourth passage for gene-expression analysis.

Cell characterization and immunostaining

Morphological characteristics, H&E staining, and karyotypic analysis were performed according to existing protocols.30–32 The doubling time of ccRCC primary cultures was subsequently determined as previously described.30, 32 The primary cells (5 × 10⁴ cells/ml) were plated onto 12-mm² glass coverslips and grown to approximately 60−70% confluence. Cells were fixed in ice-cold methanol for 5 min. CA9 expression was detected using a mouse anti-human CA9 monoclonal antibody (R&D System). Immunostaining for ABC was performed using anti-human monoclonal antibody (M-0008, Antibody Diagnostica Inc., Pleasanton, CA). Vimentin staining was performed using mouse monoclonal antibody V9 (DAKO, Denmark). The slides were then incubated sequentially with biotin-conjugated goat anti-mouse immunoglobulin G (IgG) secondary antibody (Boster Biological Technology, Wuhan, China), 0.02% hydrogen peroxide (Sigma) and 0.1% diaminobenzidine tetrahydrochloride (Sigma). Immunostaining for ABC and vimentin was counterstained with hemalaun.

RNA extraction and cDNA microarray

Primary cultures from the metastatic ccRCC lesions were paired with those from the primary ccRCC lesions from the age- (≤2 years) and gender-matched ccRCC patients without distant metastasis. The metastatic and nonmetastatic ccRCC cells were cultured under identical conditions; media were changed and cells were passaged simultaneously. Total RNA was isolated from primary cultures using Trizol RNA isolation reagent (Life Technologies) according to the manufacturer's specifications. The integrity of the total RNA was assessed using a 2100 Bioanalyzer (Agilent Technologies, Palo Alto, CA). Owing to the limited cell numbers in primary cultures, total RNA of the primary cultures from 2 age- and gender-matched patients was pooled for each microarray assay. Each assay was repeated once. Cy3-labeled 1st strand cDNA synthesized from total RNA of the metastatic ccRCC pool was mixed with Cy5-labeled 1st strand cDNA synthesized from total RNA of the non-metastatic ccRCC pool. The resulting sample was then hybridized to 16 K cDNA chips (GEO Platform ID: GPL6259; SBC-R-HC-100-22, National Engineering Center for Biochip, Shanghai, China) that included 14,784 unigenes, 241 negative controls and 527 positive controls.

Microarray data analysis

The microarray data were first analyzed using GeneSpring 7.2 Software (Agilent Technologies, Santa Clara, CA). Data obtained from spots on each chip were normalized using the lowess implementation procedure. Cy3/Cy5 ratios were transformed into the logarithm base 2, and then median normalized per chip.33 Changes in gene expression between the metastastic and non-metastatic ccRCC primary cultures were assessed using Significance Analysis of Microarray (SAM) software.34 A false-discovery rate (FDR) of <5% was considered the threshold for selection of differentially expressed genes.

Pathway analysis

Pathway information was downloaded from the Kyoto Encyclopedia of Genes and Genomes (KEGG) database (www.genome.jp/kegg/pathway.html).35 All 14,784 genes in the cDNA chip were compared with those of different pathways in the KEGG database. Pathway enrichments of the differentially expressed genes were analyzed using the χ² test from the Statistical Program for Social Sciences software package (SPSS 12.0 for Windows, SPSS, Chicago, IL). Pathway enrichment with a p value of <0.05 was considered statistically significant.

Microarray literature-based annotation

Annotations of the genes identified as being differentially expressed from the microarray data were retrieved using defined terms from the Microarray Literature-based Annotation (MILANO) database (http://milano.md.huji.ac.il).36 The literature databases GeneRif and Medline (up to November 15, 2007) were used for retrieval of quick and comprehensive results. The numbers of co-occurrences of each gene on the list with “cancer”, “metastasis” and “invasion” in the literature databases were counted. The differentially expressed genes were divided into 4 categories: co-occurrence with “metastasis”; co-occurrence with “cancer” but not “metastasis”; with no co-occurrence with “cancer” (all according to Medline); and unpublished.

Prediction of metastasis-associated genes based on protein-protein interaction

To predict novel genes related to ccRCC metastasis, genes in the first 3 of the 4 categories listed above were analyzed using an Integrated Protein-protein Interaction Network Database (IntNetDB; http://hanlab.genetics.ac.cn/IntNetDB.htm).37 The functions of genes that did not co-occur with “metastasis” in Medline were predicted based on interactions with genes that did co-occur with “metastasis”. Gene pairs with likelihood ratio (LR) ≥10 (confidence ≥ 90.84%) were selected for further study.

Gene co-expression similarity among gene-expression profiles38 was used to indicate the level of concordance between the genes that co-occurred with “metastasis” and the unpublished genes across the 12 experiments using the 16 K cDNA chips in our laboratory (GEO Accession: from GSM253487 to GSM253498). The SPSS 12.0 software package was used to measure the Pearson coefficient of the similarity. Gene pairs with a Pearson coefficient ≥0.9 and a p value < 0.01 were selected. Unpublished genes that co-expressed with some known metastasis-associated genes were selected for validation.

Reverse transcription PCR

The PCR primers were designed using Primer Premier 5.0 software (PREMIER Biosoft International, Canada) and synthesized by Shanghai Invitrogen (Shanghai, China). Primer pairs are listed in Supplementary Table I. Genes encoding β-actin and GAPDH were used as endogenous controls in semi-quantitative and quantitative reverse transcription PCR (RT-PCR) analyses, respectively.

Quantitative RT-PCR (qRT-PCR) was performed using an ABI PRISM 7000 Sequence Detection System and ABI Power SYBR Green PCR Master Mix (ABI, USA). The RNA used for qRT-PCR was taken from the same specimens used for microarray analyses. Total RNA (2 μg) from different extractions was used for the RT reaction, and cDNA (2 μL) was then used for each 25 μL PCR reaction; this corresponds to 100 ng total RNA. A total of 42 amplification cycles were performed. Each cycle consisted of 15 sec at 95°C and 1 min at 60°C. Following the amplification, the same threshold was used for comparisons of Ct (threshold cycle) values from different experiments. The mean Ct values from each sample were normalized to the corresponding GAPDH Ct values calculated as (Ct_{experimental gene} − Ct_GAPDH).

Semi-quantitative RT-PCR was used to determine differential expression of genes of interest in the primary cultures and frozen tissue specimens. Total RNA (1 μg) was used for the assay. mRNA RT was performed using the reverse transcriptase XL (AMV) and oligo(dT)₁₈ primer (TaKaRa Biotechnology Co., Dalian, China) according to the manufacturer's instructions. Single-strand cDNA in 1 μL of a 20-μL reaction mixture was amplified by PCR under the following conditions: 25 pmol of each primer, 25 nmol of each deoxyribonucleoside triphosphate, 2.5 U Taq DNA polymerase (Promega, Shanghai, China), 5 μL 10× PCR buffer in a final volume of 50 μL. An Autorisierter thermocycler (Eppendorf AG, Humburg, Germany) was programmed for initial denaturation of the samples for 1 min at 94°C, followed by 25−30 cycles of 94°C for 45 sec, 55°C for 45 sec, 72°C for 75 sec and a final elongation step at 72°C for 10 min. PCR products were electrophoresed on a 2% agarose gel, stained with ethidium bromide and visualized under UV light.

Results

Primary cultures and characterization

Primary cultures were established from 85 (90.4%) of the 94 specimens from the 84 ccRCC patients. The primary cultures took 14 ± 2.5 days to reach subconfluence for the first split. Almost all leukocytes were eliminated after the first medium change. Fibroblasts can be morphologically distinguished from ccRCC cells. As in previous reports, fibroblast overgrowth was not a significant problem in these experiments.27, 28, 32 The same is true for endothelial cells that require specific culture conditions to proliferate. The primary cultures from 38 of the 94 specimens (84 cases) contained more than 90% tumor cells at the fourth passage (Table I). Other ccRCC primary cultures failed to grow for more than 4 passages or contained more fibroblast-like cells, which were not suitable for subsequent analysis.

Figure 1 shows the morphologic, pathologic and immunostaining features of the primary cultures and histopathologic features of original ccRCC tissues. The cells exhibited epithelioid morphology and clear-cell subtype. Cancer cells were those with large and irregular nuclei, and higher nuclear-to-cytoplasmic ratios, when compared with the adjacent renal cells. These characteristics were shared by both metastatic and nonmetastatic cells. Immunostaining assays showed that all the primary cultures analyzed by microarray assays expressed CA9, vimentin and ABC, the potential biomarkers for ccRCC.5, 10, 11 The mean doubling time was 86 hr (range 48−120 hr) within the first 4 passages. Karyotypic analysis showed multiple chromosome abnormalities, with an average chromosome number of 75 (range 65−115) in the metastatic cells and 55 (range 48−58) in the non-metastatic cells.

Figure 1. Histologic and morphologic features and results of immunostaining for CA9, vimentin and ABC of ccRCC primary cultures and corresponding tumor tissues. (a−c) original tumor tissues, H&E stain, original magnification ×250. (d−f) primary cultures, original magnification ×250. (g−i) primary cultures, H&E stain, original magnification ×400. (j−l) primary cultures, immunostained with CA9 antibody, original magnification ×250. (m−o) primary cultures, immunostained with vimentin antibody, original magnification ×250. (p−r) primary cultures, immunostained with ABC antibody, original magnification ×250. (a−p) Metastatic ccRCC (to vertebra) could grow in vitro for more than 4 passages. (b−q) Nonmetastatic ccRCC failed to grow in vitro for more than 4 passages. (c−r) Nonmetastatic ccRCC could grow in vitro more than 4 passages.

Differentially expressed genes

Of 9 microarray assays performed, 6 assays with ≥85% detection efficiency (GEO Accession: GSM253487, GSM253491, GSM253493, GSM253495, GSM253496, GSM253498) were used for subsequent analysis. The valid assays represented metastastic ccRCC specimens from 6 patients and non-metastatic primary ccRCC specimens from 6 patients. Differentially expressed genes from the metastastic and nonmetastatic ccRCC were identified using SAM software. Using a FDR of 4.79% (expression change > 1.5-fold), 842 differentially expressed genes were identified, including 277 that were upregulated and 565 that were downregulated in the metastatic ccRCC samples. Using a FDR of 0.89% (expression change > 1.8-fold), 341 differentially expressed genes were identified, including 68 upregulated and 273 downregulated genes. The downregulated genes were significantly more than the upregulated genes (χ² = 19.723, p = 0.000). This refers to the size of the change in expression level. The genes that were upregulated by ≥3.0-fold and those that were downregulated ≥4.0-fold are listed in Supplementary Table II. CAV1, STAT1, GMNN, CD70, MAGEA4 and JAK3, which had previously been demonstrated to be associated with metastasis and invasion of ccRCC,6, 19, 39–42 were upregulated genes. The genes TGM2,43CP110,44E1F5A245 and JAK342 were upregulated by 1.799, 1.810, 1.851 and 1.907 fold, respectively; these genes have important roles in cancer metastasis so the 842 differentially expressed genes were used for subsequent study.

Functional enrichment analysis of the differentially expressed genes

Pathway enrichment of the 842 genes was analyzed. Significant enrichments were found in 9 metastasis-associated pathways including “cell communication”, “cytokine-cytokine receptor interaction”, “focal adhesion”, “ECM-receptor interaction”, “T-cell-receptor signaling pathway”, “metabolism of xenobiotics by cytochrome P450”, “MAPK signaling pathway”, “apoptosis” and “p53 signaling pathway”, as shown in Table II. Of the 842 differentially expressed genes, 451 were included in the GeneRIF MILANO database. Of the 451 genes, 105 co-occurred with “cancer”, 36 with “metastasis” and 26 with “invasion” (Supplementary Table III). These functional enrichment analyses indicate that the genes identified from our microarray data are reliable and consistent with current knowledge on cancer metastasis.

Table II. Differentialy Expression Genes Involved in the Pathways

Pathway	p value	Gene symbol
Cell communication	0.001	Up: COL1A2, COL4A4, LAMA2, THBS1
		Down: COL6A2, DSC1, DSC2, FN1, INA, KRT7, LMNB2, TNC
Cytokine−cytokine receptor interaction	0.002	Up: BMP7, CD70, CXCL12, IL17RB, IL1A
		Down: CCL2, CXCL2, CXCL3, CXCL6, HGF, IL11, IL6, IL8, LIF, LTB, TNF, TNFRSF10D, TNFRSF11B
Focal adhesion	0.003	Up: CAV1, COL1A2, COL4A4, LAMA2, PRKCA, RAP1A, VAV3
		Down: CCND2, COL6A2, DIAPH1, FLNC, FN1, HGF, ITGA8, ITGAV, PAK3, PIK3CD, PPP1CB, THBS1, TNC
ECM-receptor interaction	0.003	Up: COL1A2, COL4A4, LAMA2
		Down: COL6A2, FN1, HMMR, HSPG2, ITGA8, ITGAV, THBS1, TNC
T cell receptor signaling pathway	0.003	Up: PDK1, PRKCQ, VAV3, ZAP70
		Down: BCL10, CHP, NFKBIA, PAK3, PIK3CD, PPP3CA, TNF
Metabolism of xenobiotics by cytochrome P450	0.017	Up: CYP1A2, CYP2C18, GSTA1, UGT1A10
		Down: CYP2S1, GSTM3, GSTP1
Apoptosis	0.021	Up: IL1A, IRAK1
		Down: CASP3, CHP, NFKBIA, PIK3CD, PPP3CA, TNF, TNFRSF10D
MAPK signaling pathway	0.033	Up: CACNA1C, FGFR3, IL1A, NR4A1, PLA2G10, PRKCA, RAP1A
		Down: CASP3, CHP, DUSP8, EVI1, FLNC, GNG12, PPP3CA, RAPGEF2, RRAS2, TNF
p53 signaling pathway	0.041	Up: CCNB1
		Down: BBC3, CASP3, CCND2, CD82, SESN2, THBS1

Prediction of novel genes associated with metastasis

Of the 842 differentially expressed genes, 596 were recorded on Medline and of these 205 co-occurred with “metastasis”. IntNetDB analysis showed that, of the 596 genes, 450 genes showed 556 interactions (LR = 7, confidence = 55.61%). We grouped the 205 genes that co-occurred with “metastasis” as Genes A, and those genes that interacted with the Genes A group but did not co-occur with “metastasis” on Medline as Genes B. A total of 35 gene pairs with LR ≥ 10 (confidence ≥ 90.84%) were identified (Table III). Of the Genes B group, FSTL1, CDC16, PKMYT1, ORC2L, SLC15A1 and KIF4A were selected for validation, based on the importance of the interacting ‘Genes A’ gene in cancer metastasis and the number of genes from the Genes B group with which it interacts. FSTL1 and other selected genes were then used as seed genes to search for the closely interacted genes from IntNetDB database. It was found that FSTL1 closely interacted with several molecules important to cancer metastasis (Supplementary Table IV).

Table III. Prediction of Novel Metastasis-Related Genes Using IntNetDB Database

Genes A	Genes B	LR	PPI	Phenotype	GI	GO	DDI	Microarray	Gene context
LR, likelihood ratio; PPI, physical interactions; Phenotype, phenotype similarity; GI, genetic interactions; GO, shared GO annotation; DDI, domain−domain interactions.
RPL22	RPS29	15.325	−	+	−	+	−	+	−
RPL22	RPL5	14.820	−	−	−	+	−	+	−
RPL22	RPL24	14.820	−	−	−	+	−	+	−
RPL22	RPL12	14.820	−	−	−	+	−	+	−
SOD2	NCF2	14.492	−	−	−	+	−	+	−
INSL4	PSG1	13.450	−	−	−	+	−	+	−
PLN	CASQ2	13.349	−	−	−	+	−	+	−
RPL22	RPS15A	13.163	−	+	−	+	−	+	−
RPL22	RPL17	13.163	−	+	−	+	−	+	−
CCNB1	KIF4A	13.126	−	−	−	+	−	+	−
FN1	FSTL1	12.671	−	−	−	+	−	+	−
CYR61	FSTL1	12.671	−	−	−	+	−	+	−
TTK	CDC16	12.372	−	−	−	+	−	+	−
CCNB1	PKMYT1	12.036	−	−	−	+	−	+	−
PCNA	ORC2L	11.716	−	−	−	+	−	+	−
CD74	FCGR2B	11.672	−	−	−	+	−	+	−
PMP22	UGT8	11.663	−	−	−	+	−	+	−
CDC25C	CDC16	11.663	−	−	−	+	−	+	−
CTGF	FSTL1	11.596	−	−	−	+	+	+	−
SAA4	SERPING1	11.027	−	−	−	+	−	+	−
FBN1	FSTL1	11.027	−	−	−	+	−	+	−
CDH11	FSTL1	11.004	−	−	−	+	−	+	−
PGC	CTSE	10.973	−	−	−	+	−	+	−
STK6	CDC16	10.973	−	−	−	+	−	+	−
CD74	LAPTM5	10.951	−	−	−	+	−	+	−
PTPRK	PTPRZ1	10.788	−	−	−	+	+	+	−
IRAK1	GTF2F1	10.707	−	−	−	+	−	+	−
ASPH	ACO1	10.554	−	−	−	+	+	+	−
TFF2	SLC15A1	10.454	−	−	−	+	−	+	−
GMNN	ORC2L	10.383	−	−	−	+	−	+	−
IRAK1	TLR1	10.229	−	−	−	+	+	+	−
DLG1	MPP1	10.166	−	−	−	+	+	+	−
CAPG	GSN	10.123	−	−	−	+	−	+	−
PAPOLA	RPL17	10.021	+	−	−	+	−	+	−
CDC25C	PKMYT1	10.015	−	−	−	+	−	+	−

Co-expression of the 205 genes with the 246 unpublished genes was evaluated using the microarray data. Gene pairs with Pearson coefficient ≥0.9 (p < 0.01) were IL6-SIPA1L2, KRT7-Hs.474805, CXCL2-Hs.492525, PIK3CD-BE9629645, PIK3CD-ATP6V1H, SCA2-AF187554, SCA2-AF203815, SCA2-PRO1073, CHP-PRO1073, CTSB-AL713763, CTSB-AV722783 and SCL39A6-AV722783 (Supplementary Table V). As CTSB is important in cancer metastasis,46 we selected AV722783 and AL713763 (DDX17), which co-expressed with CTSB, for validation.

Validation of gene expression in the primary cultures and frozen tissue specimens

Expression patterns of FSTL1, CDC16, PKMYT1, ORC2L, SLC15A1, KIF4A, AV722783 and DDX17 in the metastatic and the non-metastatic ccRCC cultures were examined using qRT-PCR and semi-quantitative RT-PCR. Expression patterns of 3 downregulated genes FSTL1, AV722783 and SLC15A1 were consistent with those in the microarray data in all samples tested. Expression patterns of PKMYT1, ORC2L, and DDX17 were consistent with those in the microarray data in 2 of 3 independent samples. Expression patterns of KIF4A and CDC16 were consistent with those in the microarray data in 1 of 3 independent samples. The same results were obtained with semi-quantitative RT-PCR and qRT-PCR methods (data not shown). Average fold changes in expression levels of the selected genes in primary cultures of metastatic and non-metastatic ccRCC by qRT-PCR, compared with microarray data, are shown in Figure 2. FSTL1, AV722783, SLC15A1, PKMYT1, ORC2L, and DDX17 were selected for further study.

Figure 2. Expression patterns of the selected genes in primary cultures of metastatic and nonmetastatic ccRCC by quantitative RT-PCR, compared with microarray data. Average fold change is a mean value of 3 independent total RNA samples from metastatic ccRCC cultures.

Expression patterns of FSTL1, AV722783, SLC15A1, PKMYT1, ORC2L and DDX17, as well as CYR61, which is often downregulated in cancer metastasis,47 were examined in 7 paired metastatic and primary ccRCC specimens of 10 patients (specimens of 3 patients were not used owing to low-quality RNA) by semi-quantitative RT-PCR. Expression patterns of these genes in the tissue specimens were consistent with those in our microarray data (Supplementary Fig. 1 and Table IV).

Paired specimens from 37 patients were used to investigate expression patterns of the selected genes in primary ccRCC and adjacent renal tissues using semi-quantitative RT-PCR. Expression patterns of 3 upregulated genes (CAV1, PKMYT1 and ORC2L) and 6 downregulated genes (GSTM3, CYR61, SLC15A1, FSTL1, DDX17 and AV722783) are shown in Figure 3. Expression patterns of the 3 upregulated genes and 5 downregulated genes in primary ccRCC samples compared with adjacent kidney tissues were consistent with those reported for metastatic ccRCC samples compared with primary ccRCC samples (Table IV).

Figure 3. Validation of gene expression in frozen tissue specimens of paired primary ccRCC (C) and adjacent kidney tissues (N) from the first 12 of 37 patients using semi-quantitative PCR. β-Actin was used as an endogenous control.

Table IV. Validation of Gene Expression in Clear-Cellrenal-Cell Carcinoma Specimens by Semi-Quantitative RT-PCR

Genes symbol	Valid cases	Upregulated cases	Downregulated cases	p value*
* χ^2 text.
Expression in metastatic ccRCC specimens compared with its primary tumor
PKMYT1	7	6	0	0.005
ORC2L	7	7	0	0.001
CYR61	7	0	6	0.005
SLC15A1	7	0	7	0.001
FSTL1	7	0	6	0.005
DDX17	7	0	7	0.001
AV722783	7	0	7	0.001
Expression in primary ccRCC specimenscompared with its adjacent kidney
CAV1	24	18	2	0.000
PKMYT1	35	29	2	0.000
ORC2L	31	18	10	0.040
GSTM3	24	4	17	0.000
CYR61	30	9	21	0.002
SLC15A1	32	7	21	0.000
FSTL1	34	6	24	0.000
DDX17	33	10	16	0.129
AV722783	29	8	19	0.003

Discussion

The goal of the present study was to profile the expression of the metastasis-associated genes of human ccRCC. To attain this goal, the cytological composition should be as homogeneous as possible to help identify the differential transcriptional patterns in the metastatic and the non-metastatic ccRCC samples. Human ccRCC specimens are heterogeneous. Primary cultures of the clinical specimens were therefore adapted to remove any remaining necrotic tissue and minimize the influence of non-tumor cells on gene-expression profiling. This study demonstrated that ccRCC cells were successfully adapted to in vitro growth for a few passages. The success rate for establishing primary cultures was 90.4%. Tumor-infiltrating leukocytes were easily removed at the first medium change. Fibroblasts and normal renal epithelial cells have good viability after tissue dissociation with hyaluronidase type V, DNase type IV and collagenase type IV.30 In our preliminary ccRCC primary cultures that underwent enzyme treatment, actively growing fibroblasts divided rapidly after initial plating (data not shown). When the specimens were mechanically minced, in the absence of enzymes, fibroblast overgrowth was not a significant problem. Primary cultures from 38 of the 94 specimens contained more than 90% tumor cells at the fourth passage and exhibited ccRCC morphologic and immunostaining characteristics(Fig. 1). Normal renal epithelial cells only survived an average of 3 passages.27, 28, 32 These ccRCC primary cultures at low passages retained strong similarity with its tissues, based on the proteomic profiling and immunostaining assays.27, 28 Some antigenic changes of RCC cells were found at late passages (50−150+) but not at early passages (3–20).30 The gene-expression profiles of RCC cell lines were significantly different from those of RCC tissues.29 The primary cultures at early passages, rather than late passages, should be a suitable model for ccRCC, and be complementary to analysis of tissue specimens of ccRCC.

SAM software was used to identify genes with statistically significant changes in expression on several microarrays.34 By this method, we identified 842 differentially expressed genes with FDR = 4.79%. The differentially expressed genes represented 6 paired metastatic and non-metastatic ccRCC from different patients. Several data-mining methods were sequentially used to study functional enrichments of the differentially expressed genes and predict novel genes related to cancer metastasis. Strict criteria (GeneRIF and KEGG analysis) were used to enrich the genes known to be associated with cancer metastasis, whereas a loose criterion (PubMed database) was used to screen novel genes not known to be associated with cancer metastasis.

Pathway analysis revealed significant enrichments in some important pathways often associated with invasion and cancer metastasis, such as “cell communication”, “cytokine-cytokine receptor interaction”, “focal adhesion”, “ECM-receptor interaction”, “apoptosis” and “p53-signaling pathway” (Table II). The GeneRIF MILANO database contains more than 90,000 short summaries of curated articles relevant to known genes. The curation procedure extracts information relevant to the gene from the full text of a published article.36 In this study, of the 451 differentially expressed genes recruited from the GeneRIF database, 105 co-occurred with “cancer” and 36 co-occurred with “metastasis”. The differentially expressed genes included CAV1, STAT1, GMNN, CD70, MAGEA4 and JAK3, which have previously been shown to be associated with progression, metastasis, or poor prognosis of ccRCC. These results indicate the reliability of the differentially expressed genes identified from the primary cultures in predicting the prognosis of ccRCC.

IntNetDB contains 180,010 predicted protein-protein interactions among 9,901 human proteins. In this database, 27 heterogeneous genomic, proteomic and functional annotation datasets are integrated to predict the interactions. The potential of a protein-protein interaction is scored as the LR of protein pairs (true positive interactions versus true negative interactions).37 In this study, we used this tool to predict novel genes associated with metastasis. The 205 genes were used as seed genes to search for novel genes that closely interacted with the seed genes but did not co-occur with “metastasis”. Novel genes were selected according to strict statistical power (LR ≥ 10, confidence ≥ 90.84%) and previous knowledge of the known genes. As the IntNetDB database did not contain unpublished genes, co-expression of the 205 genes with the 246 unpublished genes was evaluated using the 12 microarray data. Based on the above analysis, novel genes FSTL1, CDC16, PKMYT1, ORC2L, SLC15A1, KIF4A, AV722783 and DDX17 were selected.

Using qRT-PCR and semi-quantitative RT-PCR, we confirmed that expression patterns of 6 novel genes FSTL1, AV722783, SLC15A1, KIF4A, ORC2L and DDX17 in the metastatic and the non-metastatic ccRCC cultures were consistent with the microarray data (Fig. 2). Expression patterns of FSTL1, AV722783, SLC15A1, ORC2L, PKMYT1, DDX17 and CYR61 in the paired metastatic and the non-metastatic ccRCC specimens were consistent with the microarray data (Supplementary Fig. 1 and Table IV). We conclude that FSTL1, AV722783, SLC15A1, DDX17, ORC2L and PKMYT1 are potential ccRCC metastasis-associated novel genes.

Interestingly, the 3 genes (CAV1, PKMYT1 and ORC2L) were not only upregulated in the metastatic ccRCC compared with the matched primary ccRCC, but also upregulated in the primary ccRCC compared with the adjacent kidney tissues; the 5 genes (FSTL1, GSTM3, CYR61, SLC15A1 and AV722783) were not only downregulated in the metastatic ccRCC compared with the matched primary ccRCC, but also downregulated in the primary ccRCC compared with adjacent kidney tissue in 37 paired specimens (Table IV). We hypothesized that these genes have roles in progression from carcinogenesis to metastasis and/or that the composition of potential metastatic cells increases gradually during progression from primary ccRCC to metastastic ccRCC.

CAV1, GSTM3, PKMYT1, ORC2L, CYR61, SLC15A1, FSTL1 and DDX17 were analyzed by IntNetDB. Genes that closely interacted (LR ≥ 10) with PKMYT1, ORC2L, CYR61, SLC15A1, FSTL1 and DDX17 were involved in cell adhesion, proliferation, angiogenesis, tumor suppression and metastasis. FSTL1, a candidate tumor suppressor,48 was proposed to be a hub gene in the molecular pathways related to cancer metastasis because it closely interacts with several molecules important to cancer metastasis, such as the extracellular matrix gene COL5A1, FBN1,49FN1, the tumor-suppressor gene CDH11,50 the angiogenic factor gene FGF2,51THBS2,52 and the metastasis-related genes MMP2 and CYR61 (Supplementary Table IV). Low expression levels of CYR61 were associated with more advanced malignant disease.47GSTM3 is known to be involved in the detoxification of active metabolites of many potential carcinogens.53 This study indicates that GSTM3 may be associated with ccRCC metastasis. Our results showed that CAV1 was upregulated in ccRCC metastasis, which is consistent with current knowledge.6, 54

This study indicates that silencing of the FSTL1, GSTM3, CYR61, SLC15A1, and AV722783 genes might have an important role in ccRCC metastasis. Epigenetic alterations, such as promoter methylation at CpG islands of these genes, might be involved in the process of ccRCC metastasis. Knockdown of gene expression by siRNA experiments or their overexpression by transfection experiments is necessary to elucidate the biological functions of these genes in cancer metastasis.

This study has raised some questions that remain to be answered. First, although there is some indirect evidence that primary cultures at low passages retain similarity with corresponding RCC tissues,27, 28, 30 there is no direct evidence to support this similarity. Comparison of gene-expression profiles from the primary cultures with those from frozen primary tumor tissues of patients with different prognoses might help to address this issue. Secondly, it is also necessary to clarify that some primary cultures of ccRCC specimens persistently grow in vitro, even though most fail to grow for more than 4 passages under the same culture conditions.

Our research has several limitations. First, only 6 qualified microarray datasets were used for gene-expression analysis. As the patients with apparent metastasis were not suitable for surgery, only 14 metastatic specimens were available for tissue cultures. We cannot distinguish differences in expression patterns of the selected genes between primary ccRCC patients with and without metastasis at the time of surgery owing to the small number of patients with metastasis. Second, the 16 K cDNA chip used in the study only contained 14,784 unigenes; some metastasis-associated genes may not be included as probes, resulting in loss of data. Although we used the selected genes to identify many other genes associated with cancer metastasis in the IntNetDB, further experiments are necessary to test the expression of these genes in ccRCC metastasis. Third, only 12 microarray datasets were used for the co-expression analysis. The co-expression assay results are indicative but not conclusive.

Together, these results demonstrate the utility of a reliable, effective gene-profiling method at low cost. The potential ccRCC-metastasis-associated novel genes FSTL1, AV722783, SLC15A1, DDX17, ORC2L and PKMYT1 are identified. These genes could be new candidate biomarkers to predict the prognosis of patients with ccRCC and be candidate therapeutic targets for the treatment of metastatic ccRCC.

Acknowledgements

The authors thank Ms. Xiu Zhu, Mr. Jinsong Chen, Ms. Junwen Chen, Ms. Haitang Chen, Ms. Jinfeng Zhao and Mr. Min Xu (Department of Epidemiology, Second Military Medical University), Mr. Qi Wu, Mr. Wenbin Du (Department of Urology, The 1st Affiliated Hospital, Second Military Medical University) and Dr. Huasheng Xiao (National Engineering Center for Biochip) for their technical assistance. They also thank Ms. Jie Xu (Department of Foreign Language, Second Military Medical University) for proofreading this manuscript.