Homozygous Deletions and Recurrent Amplifications Implicate New Genes Involved in Prostate Cancer^1,2

Cell Lines and DNA Isolation

PrECs are nonimmortalized early-passage human normal prostate epithelial cells maintained in serum-free defined medium (i.e., PrEC complete media). The cell and medium were obtained from Lonza Walkerville, Inc. (Walkerville, MD). The human DU145, LNCaP, PC3, and VCaP PCa cell lines were purchased from American Type Culture Collection (Manassas, VA). LNCaP is hypotetraploid, whereas DU145, PC3, LAPC4, and VCaP are near-triploid cell lines (American Type Culture Collection) [6]. BPH1 is a nontumorigenic, SV-40-immortalized, human prostate epithelial cell line generously provided by Simon Hayward [16]. CWR22Rv1 is derived from a human prostatic carcinoma xenograft that was serially propagated in mice after castration-induced regression and relapse of the parental, androgen-dependent CWR22 xenograft and was generously provided by Thomas G. Pretlow [17]. E006AA is a hypertriploid cell line established from primary PCa cells from an African-American patient, which demonstrates androgen-sensitive growth in culture; it was generously provided by Shahriar Koochekpour [18]. 975E/hTERT is a human telomerase reverse transcriptase-immortalized cell line derived from normal prostate cells generated from a radical prostatectomy (RP) specimen of a patient with a family history of PCa [19]. All of the cell lines were cultured in the RPMI-1640 medium with 10% fetal bovine serum or as previously described [19]. PC82 is a hormone-dependent transplantable tumor, which was maintained by serial xenograft transplantation in nude mice [20]; it was generously supplied by Fritz Schroeder [21]. We isolated genomic DNA using a DNA isolation kit from Gentra (Qiagen, Valencia, CA).

Clinical Samples

The clinical samples used in this study include 67 primary tumors and 5 lymph node metastases. All primary tumors were from PCa patients undergoing RP for treatment of clinically localized disease at Johns Hopkins Hospital. For the analysis of somatic DNA copy number alterations in these tumors, we selected cases from which genomic DNA of sufficient quantity (>5 µg) and purity (>70% cancer cells for cancer specimens; no detectable cancer cells for normal samples) could be obtained by macrodissection of matched nonmalignant (hereafter referred to as normal) and cancer-containing areas of prostate tissue as determined by histologic evaluation of hematoxylin and eosin-stained frozen sections of snap-frozen RP specimens. These include 9 with Gleason 6 tumors, 33 with Gleason 7 tumors, 6 with Gleason 8 tumors, 17 with Gleason 9 tumors, 1 with Gleason 10 tumors, and 1 having no Gleason score available.

GeneChip Mapping 500K SNP Assay

The GeneChipMapping 500K set comprisedNsp (∼262,000 SNPs) and Sty (∼238,000 SNPs) arrays. The median physical distance between SNPs is ∼2.5 kb, and the average distance between SNPs is ∼5.8 kb. The 500K SNP arrays were purchased from Affymetrix, Inc., Santa Clara, CA. All of the reagents used for the assay were obtained from manufacturers recommended by Affymetrix. We labeled and hybridized the 500K SNP mapping arrays according to the manufacturer's instructions [15]. We washed and stained the arrays using an Affymetrix Fluidics Station 450 and scanned the arrays using a GeneChip Scanner 3000 7G (Affymetrix, Inc.). The genotype of each SNP was generated by the GeneChip Genotyping analysis software (GTYPE) of Affymetrix.

DNA Copy Number, LOH, and Classification of Deletions and Gains

A database containing probe intensity and genotype data sets generated from the 500K SNP array using more than 200 DNA samples from both blood and normal prostate cells was used as reference samples for DNA copy number analysis in the cell lines. Because the ploidy levels vary among different cell lines, we normalized the data of log₂ ratio using most of the chromosomes as the baseline within each of the tumor genomes based on the ploidy levels of the cell lines. DNA copy number and LOH were calculated based on allele intensity using two different software packages: Copy Number Analyzer for Affymetrix GeneChip (CNAG2.0 [22]) and dChip analyzer (dChip [23]). The physical positions of the detected deletions were determined based on the Human hg17 Assembly (NCBI Build 35). Deletions and gains are deduced from the average of log₂ ratio with a smoothing window of 10-SNP probes using CNAG2.0 and dChip. The breakpoint SNPs were identified based on the median log2 ratio with a smoothing window of five SNP probes.

To distinguish which allele is deleted or gained in the clinical samples, we used allele-specific analysis as described by Nannya et al. [22]. The algorithm of allele-specific analysis takes advantage of genotype information and allele-specific intensities from paired samples to estimate DNA copy numbers for each heterozygous SNP. Allelespecific analysis can also minimize the effect of normal DNA contamination from nonmalignant cells in the tumors on the identification of DNA copy number changes, because the probe intensity of the other allele can be used as an internal reference for the detection of hemizygous deletion or gain. When the allele intensities of the match nontumor DNA are used as references, the homozygous deletion or gain can also be detected by allele-specific analysis.

We used the criteria described previously to define deletions and gains [24]. Hemizygous deletion refers to the loss of one of the alleles, whereas homozygous (biallelic) deletion refers to the loss of both alleles identified by allele-specific analysis in the clinical samples. In cell lines, homozygous deletions were identified by CNAG2.0 with a copy number of the probes equal to zero in default settings. The absolute log₂ ratios of one- and two-copy losses varied depending on the ploidy levels of the cell lines but can be clearly distinguished according to the criteria described previously [24]. For example, in near-triploid cell lines, we used log2 ratios <-0.2 with an average of ∼-0.25 in the 10-SNP genomic smoothed data to define one-copy loss, whereas we used log₂ ratios <-0.5 with an average of ∼-0.7 to define twocopy loss and log₂ ratios <-0.9 with an average of ∼-1.8 to define three-copy (complete) loss.We used the same method to define gains. For example, in near-triploid cell lines, we used log₂ ratios >0.2 with an average of ∼0.24 in the 10-SNP genomic smoothed data to define one extra copy gain, whereas we used log₂ ratios >0.4 with an average of ∼0.6 to define two-extra-copy gain and >0.8 with an average of ∼1 to define three-extra-copy gain or more. Where the gains were three or more extra copies, we define these gains as amplifications. We used both quantitative polymerase chain reaction (PCR) and FISH data to validate and refine the criteria for the copy number estimations.

Quantitative PCR

On the basis of the 500K SNP array analysis, we selected two normal DNA samples, PrEC from a primary culture of prostate epithelial cells and 083 from the blood of an unaffected individual, as reference samples for quantitative PCR (qPCR). The reaction was performed using the ABI Prism 7500 Sequence Detection System (Foster City, CA). Primers were designed using PrimerQuest software and were synthesized by Integrated DNA Technologies, Inc. (Coralville, IA). Amplicons were designed against the putatively homozygous-deleted and -amplified loci in the tumor cells and a control locus of known normalDNAcopy number. The sequences of the primers for BNIP3L, PPP2R2A, CDKN2A, CDKN2B, RAB20, ING1, OTUB1, MARK1, and a locus at 8q24.11 are available on request. Polymerase chain reaction kinetics at the control locus with normal DNA copy number was used to control for sample-to-sample differences in genomic DNA purity and concentration. Polymerase chain reaction conditions and determination of deletion and gain were described previously [15].

Fluorescent In Situ Hybridization

Fluorescent in situ hybridization was performed to analyze the DNA copy number of the androgen receptor gene (AR) in E006AA cells. CEP x (DX21) spectrum green probe (for centromere control) and the AR (Xq12) spectrum orange probe were purchased from Vysis, Inc. (Downers Grove, IL). Cell preparation and hybridization were carried out according to the manufacturer's protocol. Both interphase and metaphase analyses were performed to estimate DNA copy number of AR in relationship to centromere in E006AA and cells from a normal male as control. Fluorescent in situ hybridization analysis was performed using fluorescent microscopy with the appropriate filters to visualize the probes. Slides were blinded, then for each sample, a total of 100 interphase nuclei were scored and a ratio of red signals (AR) to green signals (Xcen) was calculated.

Copy Number Alterations and LOH Revealed by the 500K SNP Array in PCa Cell Lines

DNA copy number alterations in each of the 11 genomes for each chromosome are presented in Figures W1 to WX. The deletions and gains are inferred from the 10-SNP genomic smoothed log₂ ratios of the hybridization intensity for each of the 500K probes analyzed (presented in a supplementary data set for each of the cell lines, http://www1.wfubmc.edu/Genomics/Publications+and+Data). Among the cell lines analyzed, PC3 and VCaP harbor the most DNA copy number alterations. Within these two cell lines, chromosome 5 harbors the most changes (Figure W5). In contrast, there were no large DNA copy number aberrations found in PrEC, although some small copy number variations were observed, for example, on chromosomes 14q and 15q (Figures W14 and W15, red arrows) in these presumably normal prostate epithelial cells. We also documented deletions and gains in prostate cell lines that had not been analyzed previously by array CGH (e.g., 975E, E006AA, and PC82; Figures W1–WX).

Because the SNP array also generates genotypes at each of the 500K SNPs, it was possible to analyze LOH across the whole genome for each of the cell lines using CNAG2.0 (Figures W1–WX). As would be predicted from similar previous studies, some regions of LOH cover the entire chromosomes in some of the cell lines. These include chromosomes 3, 7, 9, 13, and 14 in E006AA, chromosome 13 in DU145, chromosomes 8, 9, 10, 16, and 22 in PC3, and chromosome 18 in VCaP, reducing hundreds to thousands of genes to homozygosity in these cell lines. Because some LOH events were not accompanied by copy number changes, LOH in these instances apparently occurred along with gain of the remaining chromosomal copy, either completely or partially (Figure 1). In addition, we analyzed the minimum common regions (MCRs) of recurrent LOH and listed the six that occurred in at least four cell lines in Table 1. Among the many genes affected by these recurrent alterations are known and suggested TSGs such as TUSC1, PTEN, and KLF5 and genes involved in cellular senescence (MORF4), DNA repair (HMGB2 and EPC1), and oxidative stress (MGST1).

Homozygous Deletions Implicate New Genes Involved in Prostate Cancer

We used three different approaches in the identification of homozygous deletions, including the value of log₂, SNP call (no call), and qPCR. Although, in general, it is not feasible to use default settings of CNAG2.0 to generate calls for a specific copy number alteration (other than homozygous deletion with a copy number equal to zero) in the presence of multiple levels of deletion in polyploid cells, there are obvious differences in log₂ ratios between levels of deletion within a given genome when it is analyzed using the 500K SNP array. Taking the deletions on chromosome 5 in near-triploid cell lines PC3, for example (Figure 2a), we were able to clearly identify one-copy deletions (light blue arrows), two-copy deletions (black arrows), and complete or homozygous deletions (green arrows). In the regions of homozygous deletion, we found the no-call rate increased in comparison to the regions without homozygous deletion, consistent with loss of all copies of the specific genomic interval. To validate the homozygous deletion calls, we performed qPCR and confirmed the results as shown in Figure 2b.

Using the 500K SNParray, we identified 12 homozygous deletions in five cell lines, with 7 apparently being novel (Figure 2, a and c). Half of these deletions occurred in a single cell line, PC3. The genes affected by these novel homozygous deletions include CDH18, AK130123, PPP2R2A, BNIP3L, MAT1A, LOC143241, AK125908, C10orf58, BC005871, TSPAN14, SH2D4B, MAP4K5, SPG3A, SAV1, USP10, CR593410, ZDHHC7, KIAA0513, FLJ44299, MGC22001, and PPAP2C. The cyclin-dependent kinase inhibitors CDKN2A and CDKN2B were found to be deleted in the E006AA line, the only line in this study derived from PCa arising in an African-American. Notably, each homozygous deletion was unique in that it was observed only once among the lines examined, although a subset of these deletions was observed additional times in the clinical specimens examined (see below). All of the homozygous deletions reside in larger regions of LOH (Figures W2, W5, W6, W8, W9, W10, W14, W16, W17, and W19). We confirmed five other regions that have been reported to be completely lost: on chromosome 2 in LNCaP; chromosomes 5, 10, and 17 in PC3; and on chromosome 6 in DU145, including the genes MSH2, KIAA0416, PTEN, SFTPA2, CTNNA1, STAT3, and STAT5B. Furthermore, using the 500K SNP array, we defined boundaries of previously reported homozygous deletions more closely and identified more affected genes, which include KCNK12, LRRTM2, C10orf59, LIPL1, LIPF, LGP2, GCN5L2, HSPB9, RAB5C, KCNH4, HCRT, AK075277, LGP1, STAT5B, STAT5A, PTRF, and ATP6V0A1 (Figure 2c).

Screening for these homozygous deletions in the clinical samples from 72 patients with PCa using allele-specific analysis, we found five recurrent biallelic losses on chromosomes 8, 9, 10, and 16, with examples shown in Figure 2d (green arrows). The most common complete loss of both alleles occurred at chromosome 10q23.31 and included PTEN, with a frequency of 21%, whereas the frequency of homozygous deletions at 9p21.3 and 16q24.1 was 4.2% (Figure 2c). The regions of complete loss at 9p21.3 in the clinical samples overlap with that observed in cell line E006AA, with an MCR affecting only CDKN2A and CDKN2B from 21,968 to 22,125 kb (Figure 2c). The MCR of complete loss at 16q24.1 in the VCaP cell line and clinical samples ranges from 83,300 to 83,541 kb, affecting the genes USP10 and CRISPLD2. We also observed single occurrences of homozygous deletions on 8p in the 72 clinical samples. The complete loss at 8p21.2 in a primary tumor affected AK130123, EBF2, PPP2R2A, and BNIP3L. We did not observe any complete losses at the other regions listed in Figure 2c in the clinical samples analyzed.

Amplifications with More Than Two Extra Copies

Using the same approach as described above for the analysis of multiple levels of deletion, we were able to define multiple levels of gain, as shown in Figure 3. Taking the gains on chromosomes 1 (Figure 3a) and 10 (Figure 2a) in PC3, for example, we identified gained regions with one, two, and three or more extra copies, marked by brown, pink, and red arrows, respectively, in this near-triploid genome. In comparison to homozygous deletions (Figure 2, a and c), the size of the amplifications, as defined by the presence of three or more extra copies, and ranging from approximately 88 kb to 7.8 Mb, was generally far larger than the homozygous deletions observed and affected a number of genes in various cell lines (Figure 3b). To confirm these amplifications, we selected five different regions on chromosomes 8, 11, and 13 and performed qPCR analysis of DNA copy number using GAPDH as the control locus. For example, in VCaP as shown Figure 4, the results of qPCR analysis revealed gains of multiple copies of RAB20, ING1, OTUB1, and MARK2. In addition, we also validated the amplification of an amplicon at 8q24.21 where several germ line PCa risk alleles have been identified in populations of both European and African descent. Whereas most of the regions with amplification were observed only once in this group of analyzed cell lines, the amplification at 8q24.21 was found in 3 of these 11 cell lines (Figure 3b).

Analyzing the gains in the 72 clinical samples, we did not find the amplifications that we had observed in the cell lines shown in Figure 3b, except at chromosome 8q (Figure 2d). Among these clinical samples, approximately 38% harbor gains on 8q with most of them being hemizygous and affecting the entire q arm. In the allelespecific analysis, we identified three tumors harboring biallele gains, ranging from 98,928 kb to 8qter (Figure 2d, red arrows), which overlap with the amplifications at 8q24. However, the MCR ranging from 128,776 to 136,826 kb in the clinical samples is located 3′ from the MCR in the cell lines and affected a number of genes including MYC.

In addition, we detected an amplification event on chromosome X, ranging from 65,893 to 66,811 kb in VCaP. This alteration harbored only the AR gene and was present in at least five extra copies (Figure 3). Amplification of AR was also observed in the cell line E006AA, although the affected region was larger, ranging from 64,829 to 76,599 kb, covering a number of genes (Figure WX). The gain of AR in E006AA cells was confirmed by FISH where metaphase analysis indicated two copies of the X centromere (Figure 4i, green), consistent with the previously reported karyotype [18] and what seems to be two to three or more copies of the AR gene per X chromosome (Figure 4i, red). Interphase analysis demonstrated multiple red signals in relation to the number of green signals that could be more easily counted than in the metaphase. The number of green signals per nucleus ranged from 1 to 7 and the number of red signals per nucleus ranged from 3 to 18 (Figure 4h), indicating extensive genetic heterogeneity within these cells. On average, we observed 5.5 copies of AR with 2.2 copies of CEP X signals per nucleus, which is consistent with the result from the SNP array analysis (Figure WX). In comparison to E006AA, VCaP apparently harbors a higher level of AR amplification, according to the log₂ ratios and metaphase analysis of FISH (Figure WX and data not shown).