Homozygous Deletions and Chromosome Amplifications in Human Lung Carcinomas Revealed by Single Nucleotide Polymorphism Array Analysis

Mapping copy number alterations in the cancer genome has contributed to the subsequent identification of tumor suppressor genes and oncogenes. The delineation of cancer-specific homozygous deletions enabled the discovery of several different tumor suppressor genes, including RB1 (1), CDKN2A (2, 3), PTEN (4, 5), and SMAD4/DPC4 (6). Similarly, genes that are targets for high-level amplification in cancers are likewise often subject to oncogenic mutations. Examples where identification of cancer-specific amplifications preceded the identification of cancer somatic mutations include PDGFRA (7, 8), EGFR (9–14), ERBB2 (15, 16), and PIK3CA (17–19).

High-resolution genomic approaches now make it possible to screen for chromosomal copy number alterations in a systematic manner. Array comparative genomic hybridization can provide high-resolution detection of copy number changes (20, 21). Single nucleotide polymorphism (SNP) arrays can measure cancer-specific loss of heterozygosity (LOH) with high accuracy in a genome-wide fashion (22–26). Furthermore, SNP arrays that cover ∼10,000 SNP loci can be used to detect DNA copy number changes at the genome level, including high-level amplifications and homozygous deletions (27, 28).

Lung cancer is the leading cause of cancer death in the world and is estimated to result in ∼160,000 deaths annually in the United States alone (29). Genomic studies have begun to impact the diagnosis and treatment of human lung carcinoma. For example, mutations in the EGFR kinase domain in non–small cell lung carcinoma (NSCLC) specimens were found to correlate with patient responses to gefitinib and erlotinib (11–13).

To discover novel genomic changes in human lung carcinomas, in the hopes of identifying additional pathways active in these diseases, we have used SNP arrays covering ∼115,000 SNP loci to investigate copy number changes in a panel of DNA from 77 NSCLC and 24 small cell lung carcinoma (SCLC) lung cell lines and primary tumors. Given the high resolution of the SNP arrays, we have been able to identify several small homozygous deletions and amplifications that have not been detected by previous methods.

Primary tumor and cell line specimens. We obtained the following genomic DNA: lung adenocarcinoma (HOP-62, NCI-H23), large cell lung carcinoma (LC; HOP-92), and squamous cell lung carcinoma (NCI-H266) from the National Cancer Institute. We prepared genomic DNA from the following cell lines: adenocarcinoma (H1437, H1819, H1993, H2009, H2087, H2122, H2347, HCC193, HCC461, HCC515, HCC78, HCC827), adenosquamous lung carcinoma (HCC366), LC (H2126, HCC1359, HCC1171), unspecified NSCLC (H2882, H2887), squamous cell lung carcinoma (H157, HCC15, HCC95), bronchioloalveolar carcinoma (BAC; H358), and SCLC (H524, H526, H1184, H1607, H1963). The primary tumors were from anonymous patients and were surgically dissected and frozen at −80°C until use. All primary tumor specimens were examined histologically to ensure at least 70% neoplastic tissue, except SCLC samples that were considered to have high tumor contents. These tumors consisted of 19 SCLC and 51 NSCLC. These include SCLC (S0168T, S0169T, S0170T, S0171T, S0172T, S0173T, S0177T, S0185T, S0187T, S0188T, S0189T, S0190T, S0191T, S0192T, S0193T, S0194T, S0196T, S0198T, S0199T), lung adenocarcinoma (MGH1622T, MGH7T, MGH1028T, S0356T, S0372T, S0377T, S0380T, S0392T, S0395T, S0397T, S0405T, S0412T, S0464T, S0471T, S0479T, S0482T, S0488T, S0498T, S0500T, S0502T, S0514T, S0522T, S0524T, S0534T, S0535T, S0539T, AD157T, AD163T, AD309T, AD311T, AD327T, AD330T, AD334T, AD335T, AD336T, AD337T, AD347T), squamous cell lung carcinoma (S0446T, S0449T, S0458T, S0465T, S0480T, S0485T, S0496T, S0508T, S0515T, S0536T), adenocarcinoma/BAC (S0376T), and BAC (S0509T, AD338T, AD362T).

Single nucleotide polymorphism array. For each sample, SNPs were genotyped with two different arrays, CentXba and CentHind, in parallel (Affymetrix, Inc., Santa Clara, CA). Array experiments were done according to the manufacturer's recommendations. In brief, two aliquots of DNA (250 ng each) were first digested with XbaI or HindIII restriction enzyme (New England Biolabs, Boston, MA), respectively. The digested DNA was ligated to an adaptor before subsequent PCR amplification using AmpliTaq Gold (Applied Biosystems, Foster City, CA). Four 100 μL PCR reactions were then set up for each XbaI or HindIII adaptor-ligated DNA sample. The PCR products from four reactions were pooled, concentrated, and fragmented with DNase I to a size range of 250 to 2,000 bp. Fragmented PCR products were then labeled, denatured, and hybridized to the array. After hybridization, the arrays were washed on the Affymetrix fluidics stations, stained, and scanned using the Gene Chip Scanner 3000 and the genotyping software, Affymetrix Genotyping Tools Version 2.0.

Data analysis. Data were normalized to a baseline array with median signal intensity at the probe intensity level with the invariant set normalization method described by Li and Wong (30). After normalization, the signal values for each SNP in each array were obtained with a model-based (PM/MM) method (31). Signal intensities at each probe locus were compared with a set of normal reference samples representing 12 individuals. From raw signal data, the inferred copy numbers at each SNP locus was estimated by applying the hidden Markov model (HMM; ref. 27). We applied the HMM model based on the assumption of diploidy or triploidy; thus, possible normalized copy numbers are (0, 1, 2, 3, 4, …; diploid) or (0, 0.67, 1.33, 2, 2.67, 3.33, 4, …; triploid), leading to the possible copy number set (0, 0.67, 1, 1.33, 2, 2.67, 3, 3.33, 4, …). The analysis methods described above are implemented in the dChip software Version 1.3, which is freely available to academic users (http://biosun1.harvard.edu/~cli/dchip2004.exe). Mapping information of SNP locations and cytogenetic band are based on curation of Affymetrix and University of California Santa Cruz hg 16 (http://genome.ucsc.edu).

We applied the circular binary segmentation algorithm (32) to our raw log₂ ratio data. This algorithm recursively splits chromosomes into subsegments based on a maximum t statistic. The reference distribution for this statistic, estimated by permutation, is used to decide whether or not to split at each stage (see ref. 32 for details). We compared the (rounded) mean raw estimated copy for each segment to our HMM results.

Quantitative real-time PCR. Relative gene copy numbers and gene expression were determined by quantitative real-time PCR using a PRISM 7500 sequence Detection System (Applied Biosystems) and a QuantiTect SYBR Green PCR kit and a QuantiTect SYBR Green RT-PCR kit (Qiagen, Inc., Valencia, CA). The standard curve method was used to calculate target gene copy number in the tumor DNA sample normalized to a repetitive element Line-1 and normal reference DNA. The comparative threshold cycle method was used to calculate gene expression normalized to β-actin as a gene reference and normal human lung RNA as an RNA reference. Primers were designed using Primer 3 (http://frodo.wi.mit.edu/cgi-bin/primer3/primer3_www.cgi) and synthesized by Invitrogen (Carlsbad, CA). Primer sequences are available upon request.

Interphase fluorescence in situ hybridization. Fluorescence in situ hybridization (FISH) probes were made from BAC clones RP11-805B16 and RP11-153K21 (Children's Hospital, Oakland Research Institute, Oakland, CA), identified to overlap the ASPH locus. BAC DNA was purified and 100 ng of each clone was labeled with digoxigenin-dUTP using random primers. The DNA was then purified with a MicroSpin S-200 HR column, ethanol precipitated, and resuspended in 100 μL hybridization solution. The control FISH probe, CEP 8 SpectrumOrange (Vysis, Downers Grove, IL), detects the centromeric region of chromosome 8. H2171 cells were grown in culture using standard methods and harvested by centrifugation after 3 days of growth. Slides for FISH analysis were prepared according to the control probe manufacturer's directions (Vysis). Briefly, cell pellets were resuspended in a fixative, a 3:1 solution of methanol and acetic acid. The cell suspension was dropped onto slides dipped in fixative and dried for 10 minutes over a 67°C water bath. The slides were pretreated with 2× SSC at 37°C for 1 hour, then digested for 5 minutes with a 1:25 solution of All III and rinsed with PBS. They were then incubated for 1 minute in 10% buffered formalin at room temperature, rinsed with PBS, dehydrated in an ethanol series (70%, 85%, 95%, and 100%), and air dried. Ten microliters of probe solution (6 μL hybridization buffer, 1 μL Cot-1 DNA, 1 μL centromere control probe, and 1 μL each of RP11-805B16 and RP11-153K21 probes) were incubated on the slide under a sealed coverslip for 3.5 minutes at 85°C and then placed in a humidified chamber overnight at 37°C. The slides were then washed in 0.5× SSC at 75°C for 5 minutes. The ASPH probes were detected using a 1:500 dilution of FITC antidigoxigenin in 10% normal goat serum, with 4′,6-diamidino-2-phenylindole as a counterstain.

Genome-wide analysis of lung cancer. One hundred one human lung carcinoma DNA samples, including 51 NSCLC primary tumor samples, 26 NSCLC cell line samples, 19 SCLC primary tumor samples, and 5 SCLC cell line samples, were hybridized to SNP arrays containing 115,593 mapped SNP loci. Two independent algorithms, the HMM in dChipSNP (27) and binary segmentation analysis (32), were used to infer copy number and thereby to identify genomic amplifications and deletions.

The genomes of lung carcinomas are often complex with numerous chromosomal alterations. For example, the lung adenocarcinoma cell line H2122 shows homozygous deletions on chromosome arms 2q, 8q, and 9p, and a significant amplification on chromosome arm 8q (Fig. 1A). Analyses of all tumor samples, throughout the genome, identifies recurrent regions of copy number gain and loss in lung carcinomas (Fig. 1B and C; Supplementary Fig. S1 shows copy number estimates for each sample across the entire genome). In both NSCLC and SCLC, the most frequent copy number gains were found in chromosome arm 5p, with ≥3 copies found in 25% and 43% of samples, respectively (Fig. 1B and C). However, the region of 5p copy number gain is usually large, and we were not able to identify any area of focal amplification from our data set.

Maximum degrees of copy number loss at a given locus, with an inferred copy number <1.5 based on HMM analysis, were found most often in chromosome arms 8p (33%) and 9p (26%) in NSCLC (Fig. 1B) and in chromosome arms 3p (68%) and 4q (58%) in SCLC (Fig. 1C). These findings are broadly similar to results reported by CGH (33); however, overall, we are reporting a somewhat lower frequency of these chromosome alterations. Possible reasons include the following: (a) the presence of stromal admixture in primary tumor samples; (b) copy number signal is attenuated either at the level of primary hybridization intensity or due to the application of the HMM; or (c) we are analyzing maximal regions of loss rather than tallying the presence of loss at any site within a whole chromosome arm. For example, the SCLC cell lines NCI-H524 and NCI-H526 only show partial loss of chromosome 3p in SNP array analysis (Supplementary Fig. S1); similar results are seen for the same cell lines upon analysis with cDNA array CGH.¹⁵

Homozygous deletions and amplifications are of particular interest because they may indicate tumor suppressor gene and oncogene loci, respectively. Regions of homozygous deletion were defined as segments of at least four SNP loci covering >5 kb with an inferred copy number of 0 by HMM. Similarly, regions of high-level amplification were defined as segments at least four SNP loci covering >5 kb with an inferred copy number ≥7.

Copy number alterations that recurred in multiple samples were verified by real-time PCR (Table 1). In general, copy number estimation was consistent between the HMM and binary segmentation approaches. Note that whereas quantitative real-time PCR results have been highly reproducible between duplicate experiments, the copy number increases detected by the arrays seem to saturate at a much lower level. Thus, an inferred copy number of ∼7 may correspond to a quantitative PCR copy number of as high as 175 or as low as 9 (MYC amplification; Table 1).

On average, the number of annotated genes is greater for regions of recurrent, high-level amplification (copy number ≥7) than for recurrent homozygously deleted regions (14.6 versus 7.7 genes/Mb, respectively). Given the parameters used in these experiments, the HMM algorithm was able to identify several amplified regions that were not found by binary segmentation but were verified by real-time PCR analysis. (All homozygous deletions and high-level amplifications identified in this study are shown in Supplementary Tables S1 and S2, respectively.)

The frequencies of high-level amplification (copy number ≥7) and homozygous deletion (copy number = 0) across the genome in NSCLC and SCLC are displayed in Fig. 1D. Note that gene names shown are merely representative of the locus. They do not imply that the indicated gene is the only or key target of chromosomal alteration or that it is involved in cancer pathogenesis.

Six distinct recurrent homozygous deletions were identified (Table 1; Fig. 1D, blue bars below the line). The most common homozygously deleted locus (7 of 26 NSCLC lines) includes the cyclin-dependent kinase inhibitor gene, CDKN2A, on chromosome 9p21, well-known to be deleted in NSCLC. Other deletions comprise the phosphatase and tensin homologue PTEN tumor suppressor gene on chromosome 10q23, in 1 of 5 SCLC lines and 2 of 19 primary tumors, and the FHIT gene on chromosome 3p, also primarily in SCLC. A homozygous deletion of chromosome 2q22.1 was found in 4 of 26 NSCLC cell lines (Table 1). Each of these four deletions fall within a single known gene, the low-density lipoprotein-related protein 1B gene (LRP1B), but in no case is the entire LRP1B gene deleted; it is unknown whether LRP1B or some undescribed transcript is the target of these deletions. Homozygous deletions in LRP1B have previously been identified in ∼17% of NSCLC cell lines (34). Interestingly, every cell line with interstitial LRP1B deletion also has undergone complete deletion of CDKN2A (Table 1). Finally, we identified two recurrent homozygous deletions on chromosome segments 3q25 and 9p23 (see below).

Genes within loci that most frequently undergo copy number gain (copy number ≥4) include the Myc family members MYC, MYCL1, and MYCN (35–37); regions encompassing the EGFR (38), FGFR1, and CDK4 (39) kinase genes and the CCNE1 cyclin gene; and the PIK3CA gene (18). All of these amplifications have been previously reported in lung carcinoma except the loci containing FGFR1 and CCNE1, which have been reported in other tumor types (40–44).

The 8q12-13 locus, where we recently identified amplification (27), is amplified in a second SCLC sample in this study. We have also identified two novel amplicons on chromosome 12p11 and 22q11.

Homozygous deletions within chromosome segments 3q25 and 9p23. There were two recurrent regions of homozygous deletion within our data set that had not been previously reported, on chromosome segments 3q25 and 9p23. Homozygous deletions within 3q25 were found in two samples, the H2882 NSCLC cell line and the S0177T primary SCLC tumor, by array analysis and quantitative PCR (Table 1; Supplementary Fig. S2A). This 120 kb region includes only two annotated genes, AADAC and SUCNR1. AADAC encodes arylacetamide deacetylase, which is predicted to catalyze biotransformation pathways for arylamine and heterocyclic amine carcinogens. SUCNR1 is a G-protein–coupled receptor for the citric acid cycle intermediate, succinate, and may be involved in succinate-induced hypertension (45).

Chromosome 9p undergoes frequent LOH in lung and other cancers, typically associated with homozygous deletion or other inactivation of CDKN2A. Our data have identified an additional region of homozygous deletion on chromosome 9p23-24.1, telomeric to CDKN2A, which includes sequence upstream of and in the 5′-most portion of the protein tyrosine phosphatase, receptor type, D (PTPRD) gene (Fig. 2A and B; Table 1). One SCLC primary tumor, S0177T, and one NSCLC cell line, H358, contain homozygous deletions confirmed by real-time PCR upstream of PTPRD and in the 5′ untranslated region, exon 1, and intron 1 of this gene (Fig. 2C). Two additional cell lines, H2347 and HCC1171, as well as H358, also contain homozygous deletions further upstream of the PTPRD coding region and more centromeric on chromosome 9 (Fig. 2C; Table 1). Whereas not all of the deletions remove exons of PTPRD, all of the homozygous deletions eliminate exons from an uncharacterized spliced transcript whose sequence and exonic structure are conserved between human and mouse (BC028038; corresponds to the position ∼8.52 to 10.60 Mb on human chromosome 9). This transcript contains a unique 5′ end, several central exons shared with PTPRD, and a unique 3′ end. An alignment of PTPRD, BC028038, and the similar mouse transcript is shown in Supplementary Fig. S3. No role for PTPRD or the BC028038 transcript has been described as yet in lung tumorigenesis; we seek to determine the frequency of alterations in these genes in lung cancer by further SNP array analysis, by quantitative PCR, and by sequencing.

8q amplification in small cell lung carcinoma. A high-level amplicon of chromosome 8q12.1-q13.11, 1.7 to 2.6 Mb in size in the SCLC cell line H2171, was identified in our previous study using SNP arrays representing ∼10,000 SNP loci (27). Interphase FISH analysis on the H2171 line confirmed the amplification of the 8q12-13 locus, with an estimated copy number of at least 12 to 20 (Fig. 3A). In this study, SNP array analysis of one SCLC primary tumor sample, S0177T, revealed high-level amplification of the 8q12-13 region near the ASPH locus (Fig. 3B). Quantitative real-time PCR revealed a copy number of 89.9, a 45-fold increase over normal genomic DNA. SNP array analysis revealed lower level copy number gains (≥4) of the 8q12-13 region in additional SCLC primary tumors, which were verified by quantitative real-time PCR (Fig. 3C) and in 3 of 22 NSCLC cell lines (not shown).

Whereas we have identified other novel regions of chromosome amplification at similar frequencies, this 8q12-13 region is of interest because of the small number of genes involved. The primary SCLC tumor sample, S0177T, contains an amplicon of 670 to 750 kb in size that does not contain the entire ASPH gene, but does include its catalytic domain and one additional open reading frame, MGC34646, encoding a protein containing a Sec14p-like lipid-binding domain (Fig. 3D). Real-time reverse transcription-PCR (RT-PCR) analysis showed the relative expression of ASPH was 12-fold higher in H2171 cell line than in normal lung. Quantitative PCR analysis of MGC34646 expression revealed it to be >100-fold higher in H2171 than in control SCLC cell lines that do not have 8q12-13 amplification (not shown); MGC34646 expression was not detectable from normal lung tissue. These data suggest but do not prove that MGC34646 may be a target of chromosomal amplification resulting in significantly increased gene expression.

Recurrent amplification on 12p11 and 22q11 in non–small cell lung carcinoma. Amplification of 12p11 was found in two NSCLC samples (Table 1; Supplementary Fig. S2B), with an overlapping region from 32.7 to 33 Mb. The minimally amplified region contains four genes, LOC283343, a pseudogene similar to argininosuccinate synthetase, and CGI-04, a provisional protein coding gene, as well as DNM1L and PKP2. DNM1L encodes dynamin like protein 1, which is a member of the dynamin family of GTPases, involved in the fission of organelles, such as mitochondria and peroxisomes (46, 47). PKP2 encodes plakaphilin 2, which may be involved in β-catenin signaling (48).

High-level amplification of chromosome segment 22q11 was found in two adenocarcinoma cell lines, HCC515 and H1819, one primary adenocarcinoma tumor, S0380T, and one large cell carcinoma cell line, HCC1359, and confirmed by quantitative PCR (Table 1), with a minimal region of overlap from 19.45 to 20.31 Mb (Supplementary Table S3). Examples from the HCC515 cell line and S0380T primary tumor are shown in Supplementary Fig. S2C. High-level amplification on 22q11 has not been previously described in lung cancer. Although we have not yet identified the target gene within the 22q11 amplicon, genes of interest that map to this region include CRKL and PIK4CA. CRKL is a member of the CRK adapter protein family, which include homologues of the v-CRK oncogene (49). PIK4CA is the catalytic subunit of phosphatidylinositol 4-kinase α, which is responsible for the downstream production of certain cell signaling molecules. Real-time quantitative RT-PCR analysis of cell line–derived mRNA showed the relative expression of CRKL in cells with 22q11 amplification to be higher (5.32-fold in HCC515 and 3.75-fold in HCC1359) than PIK4CA expression (2.61-fold in HCC515 and to be 0.4-fold higher in HCC1359), compared with both normal human lung and lung cancer lines without 22q11 amplification. However, a significant increase in CRKL protein expression was not found in cell lines containing the amplicon (data not shown). Thus, the target of the amplification remains unknown.

Copy number aberrations in the CDK4/CDK6 pathway in non–small cell lung carcinoma. The RB/p16/CDK4/CDK6 pathway is often disrupted in tumorigenesis. Copy number alterations of CCND1, CDK4, CDK6, p16, RB1, as well as CCNE1, were present in our data and were for the most part nonoverlapping, as expected with genes in a pathway. However, clear target genes have yet to be identified, as these regions with copy number alterations contain several genes in addition to these candidate ones. The CDK2NA locus on chromosome 9p21 is frequently subject to homozygous deletion in NSCLC (50) as in many other cancers. Seven NSCLC cell lines in this study were found to undergo loss of both copies of the CDKN2A locus (Table 1), confirming frequent deletion of this region; we suspect that there were also primary tumors with homozygous deletion but cannot confirm this finding in the face of stromal admixture.

Cyclin D1 (CCND1) amplification and overexpression has been previously described in NSCLC (51). The region containing the cyclin D1 gene CCND1 was amplified in five NSCLC cell lines (Supplementary Table S3). High-level amplification (3- to 4-fold) of the region surrounding the cyclin E (CCNE1) gene (19q12) was also present in two primary tumor samples, one SCLC and one adenocarcinoma (Table 1). High-level amplification on chromosome 12q13-12q14, encompassing the CDK4 locus, was found in two samples (Table 1; Supplementary Fig. S2B, left). High-level amplification of this region has recently been reported in lung cancer (39). The amplified region contains from 20 to 34 genes, among which CDK4 is an intriguing candidate oncogene (Supplementary Table S3). Two- to three-fold amplification of CDK6 was also found in four adenocarcinoma samples (Supplementary Table S3).

Tyrosine kinase copy number alterations. A survey of tyrosine kinase gene copy number identified four receptor tyrosine kinase (RTK) genes, FGFR1, ERBB2 (Her-2/neu), MET (HGFR), and EGFR, as being highly amplified (copy number ≥8) in at least one sample. These kinase genes were found in regions containing several genes and, therefore, the targets of the amplicons are still unknown. Whereas some of these amplifications are rare, we mention them because of the possibility of targeted therapy directed against aberrantly expressed protein tyrosine kinases.

A novel amplification of chromosome 8p11.2-p11.1 was identified, which includes the FGFR1 gene. High-level amplification (copy number ≥8) was found in two NSCLC samples, S0449T and MGH1622T (Fig. 4A) and was confirmed by real time PCR (Table 1); lower level copy number gains (copy number ≥4) were found in four additional NSCLC samples. Amplification of the FGFR1 locus has been found in other tumor types, including breast and urinary carcinoma (41, 42, 52). However, the role of FGFR1 in lung tumorigenesis remains unknown and we have no evidence to implicate FGFR1 as the target of this amplicon.

The identification of ERBB2, an RTK shown to be overexpressed in many human tumors, including breast, colorectal, ovarian, and NSCLC (53), has led to the development of the targeted breast cancer therapy, trastuzumab (Herceptin; ref. 54). Over 4-fold amplification (copy number ≥8) of the region surrounding ERBB2 was found in one adenocarcinoma cell line, H1819 (Fig. 4B). Additional primary tumors, including two adenocarcinoma samples and one SCLC sample, had moderate copy number gains (copy number ≥4) of the ERBB2 locus. Overexpression and activating mutations of another RTK, MET, have been found in a variety of tumor types (53). A high-level amplification (copy number = 10) of chromosome 7q31.2 resolved into three peaks in the H1993 sample. One of these peaks was a 390 kb amplicon that included MET (Fig. 4C); lower-level gain of this region was found in two additional NSCLC samples.

EGFR mutation compared with EGFR amplification. We sought to analyze amplification of the EGFR region in an unbiased fashion and to determine the degree or correlation between amplification, mutation (12), and expression (55) of EGFR within the same lung carcinoma samples. Our analysis revealed amplification of the EGFR region on chromosome 7p11.2 to copy number ≥8 in one primary squamous cell carcinoma (S0480T), one adenocarcinoma cell line (HCC827; Fig. 4D), and one primary adenocarcinoma (AD347T; Tables 1 and 2). Interestingly, HCC827 also contains an EGFR mutation (Table 2), whereas AD347T and S0480T do not.

Comparisons of EGFR amplification, mutation, and expression data indicate that high and moderate level amplifications are not always associated with EGFR gene mutation (Table 2). In addition, 17 samples run on SNP arrays, with known EGFR mutation status, also had expression information available (Table 2). Samples with EGFR mutation and/or copy number gain (copy number ≥4) of the EGFR gene were shown to have higher average EGFR expression than samples with wild type, unamplified EGFR on average, as measured by two probe sets on Affymetrix U95AV2 arrays [1537_at (EGFR): P = 0.035; 37327_at (EGFR): P = 0.004]. Our study shows that mutations in the EGFR gene are at least partially independent of EGFR gene amplification and that EGFR expression and amplification are correlated.

The present study represents the first application of genome-wide copy number analysis in lung cancer by SNP array. Our group and others recently showed that this technology provided us with a unique opportunity to assess DNA copy number changes and LOH simultaneously throughout the entire genome (27, 28). Our results illustrate the application of SNP arrays to the analysis of chromosomal alterations across lung cancer genomes. The ability to identify aberrations in the chromosomal structure of cancer cells, and the genes affected by them, has been improved through the present analysis with high resolution. The power of SNP arrays to detect small homozygous deletions with high reliability has been shown in this study with the identification of the novel chromosome 9p23 deletion. As used on unpaired samples in this study, SNP array analysis is comparable with oligonucleotide array CGH, although the SNP arrays also offer the ability to identify copy-neutral LOH when coupled with paired normal samples.

Targeting therapies based on the identification of specific chromosomal abnormalities in tumor cells have already offered great potential; inhibitors to kinases, such as BCR/ABL, KIT, ERBB2, and EGFR, have proven to be valuable as cancer therapies. Thus, kinases, such as CDK4, FGFR1, and MET, identified in this study may be future therapeutic targets in lung cancers upon further validation.

However, it is important to note that the presence of DNA copy number changes does not provide absolute proof of an involvement of these genes within the regions. Therefore, a detailed characterization should be done to elucidate the significance of the regions we have detected in this study, many of which may contain candidate tumor suppressor genes and oncogenes. Novel loci including the recurrent deletion within chromosome segments 3q25 and 9p23 and amplifications within 8q12-13, 12p11, and 22q11 will be of particular interest.

Grant support: NIH grant 2P30 CA06516-39 (C. Li), American Cancer Society grant RSG-03-240-01-MGO (M. Meyerson), Flight Attendant Medical Research Institute (M. Meyerson), Arthur and Rochelle Belfer Foundation (C. Li and M. Meyerson), and Lung Cancer Specialized Programs of Research Excellence grant P50CA70907 (J. Minna and L. Girard).

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.

Raw data will be available at the web site http://research2.dfci.harvard.edu/dfci/snp/ and the latest version of the dChip program is available at http://biosun1.harvard.edu/~cli/dchip2004.exe.

We thank Pamela Mole, Maura Berkeley, and Dr. Ed Fox at Dana-Farber Cancer Institute microarray facility for valuable assistance with the SNP array hybridization and Dr. Adi Gazdar for discussions and tumor cell lines.