Genomic expression and single-nucleotide polymorphism profiling discriminates chromophobe renal cell carcinoma and oncocytoma

A total of 30 frozen primary kidney tumors (15 chRCC and 15 oncocytomas) were obtained from the French Kidney Tumors Consortium, University of Chicago, Northwestern University, and Spectrum Health Hospital (Grand Rapids, MI). Each sample was confirmed by pathologic analysis and anonymized prior to the study. A portion of the tumor sample was frozen in liquid nitrogen immediately after surgery and stored at -80°C. Total RNA was isolated from the frozen tissues using Trizol reagent (Invitrogen, Carlsbad, CA) and purified using the RNEasy kit (Qiagen). Gene expression profiling was performed as previously described using the HGU133 Plus 2.0 Affymetrix GeneChip platform, with 54,675 distinct transcripts assayed [10]. An external GEO data-set of gene expression profiles of oncocytomas and chRCC from Cornell University was obtained for validation (GSE12090) [11]. Data for this study has been uploaded publicly in the Gene Expression Omnibus, with the accession number GSE19982.

DNA from an independent set of 6 chRCC and 8 oncocytomas obtained from the Cooperative Human Tissue Network were isolated using a Jetquick DNA extraction kit (Genomed, Lohne, Germany) according to the manufacturer's protocol. The SNP assay was performed according to the manufacturer's instructions using the Affymetrix GeneChip Mapping 100 K array (Affymetrix, Santa Clara, CA). The raw SNP array data was processed by Affymetrix GeneChip Genotyping analysis (GTYPE v.3) and human genome reference of NCBI build 36 was used for analysis.

Statistical analyses were performed in the statistical environment R 2.6.0, utilizing packages from the Bioconductor project [12]. The robust multichip average (RMA) algorithm was used to perform pre-processing of the CEL files, including background adjustment, quartile normalization and summarization. For purposes of hierarchical analysis using complete linkage analysis, probe set filtering for coefficient of variation (≥0.05, with at least 2 samples showing log₂ value expression of 8) was performed. Significance analysis of microarrays (SAM) on unfiltered data based on two-class unpaired analysis, assumption of unequal group variances and 10,000 permutations was used to derive a list of probe sets differentially expressed between tumor subclasses, and ordered by relative fold-change [13]. A maximum false discovery rate threshold was defined as 0.05.

For derivation of a small gene classifier, we used prediction analysis of microarrays (PAM), an R implementation of nearest shrunken centroids methodology with 10-fold cross validation over 100 gene thresholds and an offset percentage of 30% on unfiltered data [14]. A maximum acceptable cross-validated misclassification error was defined as ≤ 10%. The smallest predictor corresponding to this cross-validated error was selected for external validation. We inferred cytogenetic profiles for the tumors through the use of a refinement of the comparative genomic microarray analysis (CGMA) algorithm [15], which predicts chromosomal alterations based on regional changes in expression. Briefly, relative expression profiles R were generated from the single channel tumor expression profiles (T) and the mean expression values of 12 single channel cortical kidney expression profiles (N) such that R = log₂(T) - log₂(N).

KEGG pathway and gene ontology (GO) analysis of enriched gene sets was performed using hypergeometric tests available in the GOstats package in Bioconductor after having identified unique genes with corresponding annotations. For KEGG pathway analysis, the p-value threshold was 0.01. For GO analysis, conditional testing was performed, and the threshold for p was 0.001. Molecular function, biologic process, and cellular component analyses were performed.

DNA copy number (CN) was calculated based on the allele intensity of each SNP probe on the array using dChip [16]http://biosun1.harvard.edu/complab/dchip/. Information about the cytobands and the physical position of all SNPs was obtained from Affymetrix and UCSC genome bioinformatics database (NCBI Build 36.1) http://genome.ucsc.edu. The working criteria for loss or gain are defined as the chromosomal region with at least four consecutive SNPs with CN < 1.6, or at least four consecutive SNPs with CN > 3.5, respectively. Copy number alteration (CNA) regions were identified when more than 30% of the samples showed copy number loss or gain.

Immunohistochemical staining was performed on an independent set of chRCC (n = 11) and oncocytomas (n = 7). Aquaporin 6 and synaptogrin 3 were selected from the PAM (Table 1). Parafibromin (218578_at) (2-fold expression difference) and cytokeratin 7 (209016_s_at) were selected from the SAM analysis of the gene expression profiles for validation. Candidate marker choice was determined by factors including fold-change, specificity, biological and clinical interest. CK7 was selected as a marker to ascertain the additional benefit of routine pathologic practice in the samples. Briefly, following blocking and antigen retrieval, 4-micron sections on coated slides were incubated with the following antibodies: a mouse anti-cytokeratin 7 monoclonal antibody (DakoCytomation, Carpinteria, CA, 1:50, cytoplasmic staining), a polyclonal rabbit anti-human aquaporin 6 (AQP6, Alpha Diagnostic International, San Antonio, TX, 1:100, overnight at 4°C, membranous staining), polyclonal goat anti-synaptogyrin 3 N-18 and C-18 antibodies (SYNGR3, Santa Cruz Biotechnology, Santa Cruz, CA, cytoplasmic and membranous staining), a mouse monoclonal antibody specific for parafibromin (1:250, 1 hour at room temperature, nuclear staining) [17], a rabbit monoclonal anti-HER2 antibody (Neomarker RM 9103-S clone SP3, 1:200, membranous staining), a phospho-AKT (Ser473) antibody (Cell Signaling Technology, 1:30, cytoplasmic and nuclear staining). For the latter two antibodies, 22 chromophobe RCC and 8 oncocytoma specimens were available. For p-AKT, staining in the stromal and the tumor cell compartments was separately assessed. Subsequent reactions were performed with biotin-free HRP enzyme labeled polymer of EnVision Plus detection system (DakoCytomation). All slides were examined by a pathologist in a blinded fashion.

We visualized the 30 expression profiles of chRCC and oncocytoma by hierarchical clustering upon a filtered data-set of 8,995 transcripts. Clear partitioning of the two entities into separate classes was observed (Figure 1). PAM yielded excellent cross-validated discrimination over a series of thresholds [Additional file 1: Figure S1]. A gene predictor comprising 14 probe sets was identified (Table 1), which yielded an overall accuracy of 93% in the internal data-set (28/30) (Table 2). The same predictor successfully classified 17 of 18 samples in the external dataset from Cornell University, corresponding to an overall accuracy of 94% (Table 2). 5,210 probe sets were found to be differentially expressed between the two entities as identified using SAM at a delta of 1.4, with a false discovery rate of 0.03 corresponding to an estimated 222 probe sets [Additional file 2: Table S1]. 2,564 number of probe sets were relatively overexpressed in chRCC, and 2,646 transcripts relatively underexpressed in chRCC.

Bild vergrößern

We report copy number gains that were detected in chromosomes 4, 7, 11, 12, 14q, and 18q (Figure 2). Whole chromosomal losses of chromosome 1, 2, 6, 10, 13, 17, and 21 represent a unique copy number loss profile for chRCC. For renal oncocytoma, losses of chromosome 1p were noted. A CGMA (Figure 3) derived from the expression data yielded regional expression biases consistent with that reported by DNA copy number profiling. We report in particular that our high throughput methods demonstrate that there is a common gene alteration to both tumors (loss of chromosome 1p), which may represent an early event common in the histogenesis of both tumors. In particular, in oncocytoma, this loss appears restricted to the terminal end of 1p.

Bild vergrößern

Bild vergrößern

Pathway and GO analysis was performed on the SAM analysis, demonstrating an enrichment of genes involved in metabolic pathways in oncocytomas relative to chRCC (Table 3). These metabolic pathways include oxidative phosphorylation, amino acid metabolism, and fatty acid metabolism. Conversely, high expression of genes involved in cell adhesion, immune receptor signaling as well as proliferative pathways such as c-erbB2 (Her-2/neu) and mammalian target of rapamycin (mTOR) signaling are detected in chRCC. GO analyses performed supported these results [Additional file 3: Table S2], highlighting that mitochondrial genes were highly overrepresented among genes relatively overexpressed in oncocytomas, whereas tight junction genes were similarly overrepresented among genes overexpressed in chRCC.

The immunohistochemical profiling is summarized in Table 4 and Figure 4, the results of which were consistent with the mRNA quantitation by microarrays. The immunoreactivity of chRCC to cytokeratin 7 was higher than that of oncocytomas and normal kidney. For parafibromin, clear differential staining was noted, with predominantly nuclear expression in oncocytomas, and absent expression in chRCC. For synaptogyrin-3, both N-18 and C-18 antibodies yielded a similar signal, but the N-18 antibody yielded a crisper result though the maximal signal was distinctly weaker compared to AQP6, for which crisp membranous staining was noted in oncocytoma, but not in chRCC. For p-AKT, there was an apparent, but non-significant higher immunoreactivity in chRCC than oncocytoma, particularly in the stromal cells relative to the tumor cells. For extracellular HER2, all samples were unreactive (Images for p-AKT and HER2 not included).

Bild vergrößern