Use of Complete Coverage Array Comparative Genomic Hybridization to Define Copy Number Alterations on Chromosome 3p in Oral Squamous Cell Carcinomas

Deletion on 3p is a frequent event that is likely to play a significant role in the pathogenesis of many cancer types, including oral squamous cell carcinomas [OSCCs (1, 2)]. The current literature suggests that there are multiple regions of chromosomal loss on this arm. However, these regions are poorly defined and their prognostic significance is unknown, with the exception of one region (3p14) where allelic loss has been associated with risk of progression of oral premalignant lesions to cancer (3, 4, 5, 6). The recent development of array comparative genomic hybridization (CGH) technology and its application to microdissected archival tissue support the feasibility of high-resolution profiling of copy number changes on an entire chromosome arm in a single experiment (7). However, the only report of array CGH analysis of an entire chromosome arm involves the use of a full coverage array of chromosome 22 [∼450 bacterial artificial chromosome (BAC) clones (8)]. The present study represents the first use of this approach to demarcate regions of copy number alterations in OSCCs.

Formalin-fixed paraffin-embedded tissue blocks were obtained from the British Columbia Oral Biopsy Service and diagnoses were confirmed by an oral pathologist. Tumor cells were microdissected from H&E-stained sections and DNA was isolated and quantified as described previously (9). The clinical features of the 20 cases are shown in Table 1.

A minimal, near-overlapping tiling set of 535 human BAC clones spanning chromosome arm 3p (3p12.3 to 3p26.3) was selected from the RPCI-11 library. The choice of these clones was based on their location on the human physical map⁴ and their map location was verified using the University of California at Santa Cruz Biotechnology Human Genome Browser.⁵ This clone list is available publicly online.⁶ The identity of each BAC DNA sample was confirmed by its HindIII fingerprint. To generate sufficient quantities of DNA for array construction, we amplified each BAC DNA sample by linker-mediated PCR. As described previously (10), MseI-digested BAC DNA was ligated to linkers (5′-AGTGGGATTCCGCATGCTAGT-3′ and 5′-TAACTAGCATGC-3′) and amplified by two rounds of PCR. The amplified DNA was dissolved in a 20% DMSO solution, boiled for 10 min, and re-arrayed for robotic printing. Each clone was spotted in triplicate with Stealth Micro Spotting Pins onto amine-coated slides (Telechem/ArrayIT SMP2.5, Sunnyvale, CA) using a VersArray ChipWriter Pro system (Bio-Rad, Mississauga, Ontario, Canada). The clones were arranged to ensure that adjacent BACs on the tiling set were not placed next to each other on the array. To allow normalization of the hybridization signal intensities between dyes, linker-mediated PCR-amplified normal male human genomic DNA samples (Novagen, Madison, WI) were spotted on the array 48 times. In addition, 96 randomly selected BACs distributed throughout the genome were included on this array as internal control spots to detect hybridization artifacts such as hybridization signal gradients. The DNA was then covalently bonded to the slides by baking and UV cross-linking. Slides were washed to remove unbound DNA.

Probe labeling was as described previously (10). Test and reference DNA (100 ng each) were labeled separately using Cyanine 3 and Cyanine 5 dCTPs, respectively. The DNA probes were combined, denatured, and annealed in a solution containing 100 μg of human Cot-1 in 25 μl of DIG Easy hybridization solution (Roche, Quebec, Canada), 50 μg of sheared herring sperm DNA (Sigma-Aldrich, Oakville, Ontario, Canada), and 250 μg of yeast tRNA (Calbiochem). The arrays were prehybridized with DIG Easy hybridization buffer (Roche) containing 1% BSA and 2 μg/μl sheared herring sperm DNA at 42°C for 1 h. The probe mixture was applied to the slide surface and hybridized for 36 h at 42°C. Arrays were washed five times with 0.1× SSC and 0.1% SDS (5 min each at room temperature), rinsed five times with 0.1× SSC, and then dried by centrifugation.

A charge-coupled device-based imaging system (Arrayworx eAuto; API, Issaquah, WA) was used to determine signal intensities of the Cyanine 5/Cyanine 3 channels. Images were analyzed with Softworx array analysis software. A scale factor based on the signal intensities of the aforementioned 48 human genomic DNA control spots on the array was used to normalize spot signal data for each channel. SDs for each triplicate spot set were calculated. The array CGH profile was presented as a graph plotting normalized Cyanine 5/Cyanine 3 log2 signal ratios versus the relative tiling path position of the BAC clones. A log2 signal ratio of 0 at a spot represents equivalent copy number between the sample and reference DNA. Quality control was applied to each batch of arrays synthesized. Normal DNA versus normal DNA hybridizations revealed spots with aberrant signal intensity, and these were removed from the analysis. Spots that exhibited signal ratios outside of 3 SDs from 0 in these experiments were discarded, resulting in the establishment of a ±0.2 log2 ratio threshold for defining regions of copy number increase and decrease. In addition, alterations had to have more than one consecutive BAC with copy number change to be scored as a region of alteration.

We have constructed a high-resolution BAC array for CGH analysis of chromosome 3p. This array consists of 535 overlapping BACs, providing near complete coverage of chromosome 3p from telomere (3p26.3) to centromere (3p12.3).

Twenty oral tumors were microdissected and the DNA was analyzed for segmental copy number change using this array. Nineteen of these 20 tumors had profiles where the internal control DNA spots gave consistent signal ratios. Of these 19 tumors, 2 had no apparent copy number alteration (Fig. 1,A), 3 showed a whole arm deletion (Fig. 1,B), and the remaining 14 showed multiple segmental copy number alterations (Fig. 1,C). As shown in Fig. 2, the pattern of alteration in these tumors was complex, showing discontinuous copy number change with either several regions of deletion or combinations of decreases and increases in copy number.

A large region of copy number decrease (∼27 Mbp) occurred at 3p14 (Fig. 2, Region 1). This alteration is present in 7 of the 15 tumors with segmental alterations. This region contains the fragile histidine triad gene, or FHIT, which has been shown to be altered in many tumor types as well as in oral premalignant lesions (11). Loss of FHIT expression is a predictor of poor outcome in oral cancer (11). The wide extent of alteration in Region 1 supports the possibility of other tumor suppressor genes in addition to FHIT.

Although there were numerous regions of copy number change in the 14 cases with segmental alterations, three regions (Fig. 2, Regions 3–5) were present at frequencies equal to or greater than that of the FHIT-containing Region 1. These regions were small (<2 Mbp each; Table 2) and together contain only two known genes, RBMS3 and GRM7, neither of which has previously been linked to oral cancer. GRM7, in Region 5, is a metabotropic glutamate receptor and RBMS3, in region 4, is a MYC gene single strand-binding protein. MYC gene single strand-binding proteins are thought to cooperate with MYC to regulate DNA replication, gene transcription, apoptosis, and cell cycle progression (12).

To date, only deletions have been associated with 3p in oral cancer. The final region identified at 3p21.31 (Region 2) is unique in that it contains an increase in copy number. This alteration is present in 6 of 14 cases, with 5 cases defining a ∼0.7 Mbp segmental copy number increase. For the sixth case (474T), the region covers only 0.2 Mbp. Based on the University of California at Santa Cruz Biotechnology Human Genome Browser (April 2003 version), there are eight known genes within the 0.7 Mbp region. Although none have been implicated in oral cancer, two are putative oncogenes. Testes-Specific Protease 50 (TSP50) has been shown to be overexpressed in breast cancer (13) and Teratocarcinoma-Derived Growth Factor 1 (TDGF-1) is overexpressed in a number of cancer cell lines (14). This region also contains a site recently reported to be sensitive to carcinogens present in tobacco (15).

Interestingly, genetic alterations near or within Region 2 have been described previously. Immediately centromeric to Region 2 is RASSF1A, which is frequently silenced by methylation in many cancers (16), although it is infrequently silenced by methylation in head and neck cancers (17).

We compared the known and novel regions of copy number alterations with previously published regions defined by fine mapping efforts using microsatellite markers and CGH analysis (1, 2, 5, 18, 19, 20). Historically, fine mapping by loss of heterozygosity in tumors is difficult due to the paucity of microsatellite markers as well as the constant revision of the human genome map. We have positioned the microsatellite markers used in the OSCC 3p fine mapping studies (1, 2, 5, 19) and highlighted those that showed allelic loss in >66% of cases in Fig. 2. Interestingly, these markers fall within Region 1 (which contains FHIT) at 3p14, Region 4 at 3p24, and Region 5 at 3p26.1 as defined by array CGH.

In summary, this is the first report of complete profiles of copy number changes on 3p in OSCCs. The data show the complex type of alterations that occur on this arm, many of which are small in size, requiring high-resolution assays to define critical regions of change relevant to oral cancer. Using this approach, we detected a large region containing the extensively studied FHIT gene, but we also detected three novel small regions of deletions (3p22, 3p24.1, and 3p26.1) at high frequency. We also report the first observation of a region showing copy number increase on 3p.

It is known that 3p alterations occur early in the development of oral cancer (3, 4, 5, 6). Future studies will be targeted toward mapping early premalignant lesions through segmental copy number change and will begin to delineate minimal regions of alteration for gene discovery.

We thank Bradley Coe, Spencer Watson, and Chad Malloff for array synthesis and bioinformatics assistance and Michelle Morris for providing clinical information on the OSCC samples used in this study.