Novel Integrative Methods for Gene Discovery Associated With Head and Neck Squamous Cell Carcinoma Development FREE

Head and neck squamous cell carcinoma (HNSCC) is the fifth most common cancer, and more than 500 000 cases are diagnosed annually worldwide. In the United States, it accounts for 3% of all cancers and 40 000 new cases each year.¹ Although considerable progress has been made in the areas of early detection, diagnosis, and treatment, the 5-year survival rate for patients has shown only modest improvement in the past 40 years.² Comprehensive analysis of clinical and treatment factors has shown tumor site–specific improvements in 5-year survival for cancers of the nasopharynx, oropharynx, and hypopharynx and late-stage laryngeal cancer.³

Exposure to tobacco and alcohol are the clinical factors most strongly associated with the development of tumors. In addition, many of the cancer genetic and epigenetic changes associated with sporadic HNSCC have been described, including p53 (GenBank 7157), VEGF (GenBank 7422), and cyclin D1 (GenBank 595).⁴ Yet, the molecular mechanisms of HNSCC carcinogenesis are still undergoing intensive investigation. Many heritable genetic alterations have been associated with cancer formation. Neoplastic cells undergo a variety of genetic alterations, from point mutations to chromosomal aberrations, that affect the function or expression of both oncogenes and tumor-suppressor genes. Chromosomal loss and gain resulting in gene loss of heterozygosity and gene amplification, respectively, have been shown for years to result in gene silencing (Knudson’s⁵ hypothesis) or gene overexpression.⁶ Amplification often results in increased messenger RNA (mRNA) transcription owing to an overabundance of promoter and template.⁷ Deletion can result in silencing via uniallelic or biallelic loss. Advances in techniques over the past few years, including array-based comparative genomic hybridization (aCGH) and array-based single-nucleotide polymorphism (aSNP), have allowed high-throughput, highly detailed studies of chromosomal loss or gain. For this study, we considered previously published areas of amplification (3q, 5p, 8q, 9q, and 20q)^8- 12 and deletion (3p, 8p, 13q, and 18q)^13- 14 and performed a comprehensive search for the best source of high-definition chromosomal loss and gains in HNSCC (aCGH/aSNP).

For the past few years, efforts have been made that begin to define the consensus cancer coding sequence.¹⁵ The first effort in publishing of the human consensus cancer coding sequence showed somatic mutations in many known cancer-causing genes (eg, p53 and APC), as well as discovery of somatic mutations in many new targets.¹⁶ This study group determined the sequence of well-annotated human protein-coding genes in 2 common tumor types and conducted an analysis of 13 023 genes in 11 breast and 11 colorectal cancers. A total of 189 genes (mean, 11 per tumor) were mutated at a notable frequency. Most of these genes were not previously known to be genetically altered in tumors. A wide range of cellular functions were implicated, including transcription, adhesion, and invasion.¹⁷ These types of studies find somatic DNA sequence mutations in cancers that provide a novel way to suggest function based on pressures of clonal selection in cancer. Similar efforts are under way by other organizations. New high-throughput screens of cancer genes are being developed at a rapid pace, creating the need for efficient approaches for integration of large data sets that use diverse technologies to describe genetic alterations in human cancers.¹⁶

Revelations of the human genome project and other recent advancements in technology have ushered in a new era of research. Genome-scale data sets of various forms are readily available to cancer researchers. Examples of such genome-scale data sets include aCGH (DNA loss or gain), RNA expression microarray (tissues or cell lines), small molecule cell line screens, and various proteomic approaches. Prior experience suggests that tumors may be susceptible to targeted therapies once their essential molecular alterations have been found. Integrative approaches to these genome-scale data sets allow multiple pieces of salient information to be combined in a manner that may yield novel and powerful new insights into biologic mechanisms of cancer.¹⁸ One prominent example of the usefulness of these integrative genetic approaches led to finding the oncogene MITF in melanoma. Researchers applied 2 genome-scale data sets (aSNP and expression microarray) to discover the oncogene MITF in melanoma. This gene may represent a new class of “lineage addiction oncogenes”—a fundamental tumor survival mechanism with important therapeutic implications.¹⁹ Other examples of integrative genomic approaches have also improved the genetic understanding of other cancers.^20- 23

In this study, we used an integrative genomic method to consider genes that are altered in HNSCC. We used primary data from 3 general sources: (1) genes mutated in the consensus cancer coding sequence,¹⁷ (2) DNA/chromosomal deletion or amplification in primary HNSCC, and (3) RNA gene expression differences in HNSCC and normal tissues. These were then integrated in an attempt to define novel genetic alterations (gene deletion or amplification) in HNSCC that result in important alterations in associated gene expression. Three genes were studied in detail: RUNX1T1 (GenBank 862), a zinc finger transcription factor protein and translocation-activated oncogene in acute myeloid leukemia²⁴; RFC4 (GenBank 5984), a gene involved in DNA replication, repair, modification, and chromatin modeling²⁵ noted to be overexpressed in other cancers,²⁶ particularly human papillomavirus–associated cancers²⁷; and DLEC1) (GenBank 9940), a putative tumor suppressor involved in carcinogenesis of the lung, esophagus, kidney, and nasopharyngeal cancers.^28- 29

All samples were analyzed by the pathology department at Johns Hopkins Hospital, Baltimore, Maryland. Tissue samples were obtained via Johns Hopkins institutional review board–approved protocols under protocol No. 92-07-21-01. Normal samples were microdissected and DNA prepared from upper aerodigestive mucosa from healthy patients primarily undergoing uvulopalatopharyngoplasty procedures. Tumor samples were confirmed to be HNSCC and subsequently microdissected to separate tumor from stromal elements to yield at least 80% tumor cells. Tissue DNA was extracted as described in the subsection titled “DNA Extraction” in this section.

Total cellular RNA was isolated using the RNeasy kit (Qiagen, Valencia, California) according to the manufacturer's instruction. We performed oligonucleotide microarray analysis using the GeneChip U133A and U133B plates on the Affymetrix (Santa Clara, California) expression microarray, which assays 33 000 genes. Samples were converted to labeled, fragmented, complementary RNA (cRNA) per the Affymetrix protocol for use on the expression microarray. Signal intensity and statistical significance were established for each transcript using dChip, version 2006 (free, downloadable software available at http://biosun1.harvard.edu/complab/dchip). Default settings for dChip were used, including the perfect match/mismatch difference model, invariant set normalization, and check single/probe/array outlier algorithm.

The public databases used in this study were the University of California, Santa Cruz (UCSC), Human Genome reference sequence and the annotation database from the May 2004 freeze (hg17). For target discovery we used 3 data sets. First, we used all 1149 genes in the consensus cancer coding sequence from supplemental tables provided by Sjöblom et al¹⁷ from a survey of 13 023 genes with DNA sequencing. High-definition detailed locations of DNA and chromosomal loss and gain were considered from the published areas in the report by Sparano et al,³⁰ which coincided with analysis from the literature of reliable areas of chromosomal loss and gain in HNSCC. This group used aCGH to develop a genome-wide molecular profile of oral squamous cell carcinoma from 21 prospectively collected fresh-frozen specimens, at a 0.9-Mb resolution to identify distinct regions of genomic alteration and their associated genes.³⁰ The consensus cancer coding sequence¹⁷ genes were located with data from UCSC genome browser and joined to areas of loss or gain based on chromosome and base pair location coinciding with areas of loss or gain. Any coding sequence overlap with the amplification or deletion was considered important. Gene RNA expression microarray analysis of 6 primary normal samples and 8 primary tumor specimens on the Affymetrix U133A and U133B platform (33 000 genes) was conducted. Microarrays were studied with dChip and invariant-set normalized. Median tumor expression, T_exp, and median normal expression, N_exp, were calculated. P values were determined using Stata statistical software (version 9.0; StataCorp LP, College Station, Texas). Validation of targets was initially based on putative cancer causation from a literature search and subsequently quantitative real-time polymerase chain reaction (qRT-PCR) for expression on primary tumor and normal samples, followed by quantitative polymerase chain reaction (qPCR) for precise validation of gene amplification or deletion, and/or sequencing of primary tumors (see the following 2 subsections).

Samples were centrifuged and digested in a solution of detergent (sodium dodecylsulfate) and proteinase K for removal of proteins bound to the DNA. Samples were first purified and desalted with phenol/chloroform extraction. The digested sample was subjected twice to ethanol precipitation and subsequently resuspended in 500 μL of low-salt Tris hydrochloride–ethylenediamine tetraacetic acid buffer (ethylenediamine tetraacetic acid, 2.5 mM/L, and Tris, 10 mM/L) and stored at –80°C.

The total DNA was measured and adjusted to the same amount for each tissue sample. The DNA was used as the templates for qRT-PCR with primers designed to specifically measure the DNA copy number of each candidate gene. β-Actin was examined to ensure accurate relative quantitation of copy number in qPCR. (Detailed PCR conditions and primer sequences are available from the authors on request.)

The total RNA was measured and adjusted to the same amount for each tissue sample, and then cDNA synthesis was performed by priming with oligo(dT) and the SuperScript First-Strand Synthesis kit (Invitrogen). The final cDNA products were used as the templates for subsequent PCR with primers designed specifically for each candidate gene. Glyceraldehyde-3-phosphate dehydrogenase or β-actin was studied to ensure accurate relative quantitation in qRT-PCR. (Detailed PCR conditions and primer sequences are available from the authors on request.) The lesions of the 36 patients evaluated for qRT-PCR were 100% primary cancers; the patients were without prior treatment, 80.5% were male, and their median age was 58 years. There were 14 current smokers, 14 previous smokers, and 8 nonsmokers. All statistical computations were made with Stata SE software (version 9.0; StataCorp LP).

We used an integrative bioinformatic approach to identify genes in the consensus cancer coding sequence,¹⁷ located in areas of chromosomal loss or gain, which had differential expression in HNSCC (Figure 1). The process had 2 parts: target discovery and target validation. Three data sets were incorporated into target discovery: (1) genes found in the consensus cancer coding sequence survey of 13 023 possible cancer-causing genes and DNA mutations,¹⁷ (2) regional DNA loss or gain of the whole cancer genome (using a literature search of known HNSCC amplicons and deletions with aCGH data from Sparano et al³⁰), and (3) RNA expression differences of 33 000 genes measured from expression microarray (Affymetrix, U133A and U133B plates) of 6 primary normal samples and 8 primary tumor specimens. Validation of targets was initially based on literature search, qRT-PCR for expression on primary tumor and normal samples, qPCR for validation of exact gene amplification or deletion, and/or primary tumor sequencing. The Table shows the 20 genes from the consensus cancer coding sequence,¹⁷ which are in areas of HNSCC chromosomal loss or gain. Of these 20 genes, 8 were part of areas of chromosomal deletion, and 12 were part of chromosomal amplification. Although all genes were validated somatic mutations found in breast or colon cancer, only 5 genes were CAN genes found in the final validation set by Sjöblom et al.¹⁷ These genes and their cancer mutation prevalence score, which reflects the probability that the number of mutations observed in a gene reflects a mutation frequency that is higher than that expected to be observed by chance (frequency >1.0 was considered significant), were CHL1 (1.26) (GenBank 10752), CMYA1 (1.36), ITGA9 (1.055), RUNX1T1 (2.42), and TGFBR2 (GenBank 7048) (2.85). These genes can be theoretically altered by mutation (missense, nonsense, etc), gene amplification or deletion, or via epigenetic activation or inactivation. All 20 genes may have functionally altering mutations in sporadic tumors, but gene amplification and deletion events tend to occur at a higher frequency in nonfamilial cancers, resulting in gene activation or inactivation. Five of these genes—BRCA2 (P = .003), CHL1 (P = .05), DLEC1 (P = .002), RTP1 (GenBank 132112) (P = .05), and TGFBR2 (P = .03)—showed statistically significant differences in T_exp, compared with N_exp, that correlated with amplification (upregulated) or deletion (downregulated) status, based on Mann-Whitney U test. These genes are shown in Figure 2. One additional gene from the Table, EIF4A2 (GenBank 1974), is contained within a well-described 3q amplicon in HNSCC but showed statistically significant downregulation (median T_exp/N_exp = 0.82; P = .02), which may reflect an area of chromosomal disruption. Also shown in Figure 2 are other genes with trends toward differential expression but which were not statistically significant: LIFR (median T_exp/N_exp = 0.84; P = .61), KCNB2 (median T_exp/N_exp = 1.23; P = .19), ITGA9 (median T_exp/N_exp = 0.80; P = .15), and RFC4 (median T_exp/N_exp = 1.54; P = .20).

We chose a subset of targets to validate expression differences with qRT-PCR. Targets DLEC1, RFC4, and RUNX1T1 were chosen because of expression differences and noted involvement in the pathogenesis of other cancer types. Figure 3 shows qRT-PCR expression of DLEC1, RFC4, and RUNX1T1, all normalized by GAPDH expression in 36 primary tumor tissues, and 7 normal upper aerodigestive mucosal tissues. DLEC1 is part of the 3p22 deletion found in HNSCC, and our hypothesis was that a subset of HNSCC primary tumors would show reduced expression. Thirteen of 36 tumors (36%) showed expression levels below all normal specimens. RFC4 is found on an amplification of 3q27 in HNSCC, and we found that 9 of 36 tumors (25%) showed increased expression compared with the highest expression level found in the normal tissues. RUNX1T1 is also found in an amplification, located on chromosome 8q. Eleven of the 36 tumor specimens (31%) showed expression levels higher than any normal samples.

We conducted sequencing of the TGFBR2. Of note, missense mutations are responsible for Lynch syndrome/HNSCC type 6, which confers an increased risk of multiple malignant lesions, primarily colon cancer, but also larynx cancer.^31- 33 We were unable to find any mutations in the coding sequence of this gene in 12 primary HNSCCs. Significant expression differences were noted (see Figure 2 for P values).

Last, we tried to assess the relationship between DNA coding amplification or deletion and associated mRNA expression (Figure 4). Three genes were chosen for further study based on in silico genetic profiling: RUNX1T1 (acute myelogenous leukemia translocation 1), RFC4 (replication factor C4), and DLEC1 (deleted in esophageal cancer isoform 1). Twelve primary HNSCC tissues had tumor DNA and RNA, matched leukocyte DNA extracted to study gene amplification and deletion, and resultant mRNA expression for each gene of interest. Six of the 12 tumors (50%) demonstrated amplification of the RUNX1T1 locus of greater than 3.5 copies, which was not seen in the leukocyte samples (P = .01). Four of those 6 with amplification (67%) demonstrated upregulated transcription of this gene. Among the tissues with gene amplification, there was statistically significant overexpression of this gene (P = .02) by Mann-Whitney U test. Amplification of the RFC4 locus of more than 6 copies was detected in 5 of 12 of the tumors (42%) (P = .03 by χ² test). One of those 5 (20%) demonstrated associated upregulated mRNA transcription. This was not statistically significant (P = .60). For the DLEC1 locus, 4 of the 12 tumors (33%) vs 0 of 12 matched leukocyte DNA samples demonstrated deletion with a copy number of less than 1.5 (P = .05 by χ² test). This finding is consistent with previously published³⁴ 3p22 loss rates of 40%. Three of those 4 (75%) with copy-number reduction demonstrated mRNA expression below median levels for that gene. Samples with deletion of DLEC1 had reduced expression of the gene that was not statistically significant (P = .06).

With the advent of novel high-throughput techniques to assay changes in tumor genomics, new and exciting methods of gene discovery in cancer are being developed. In 2005, Garraway et al¹⁹ performed an integrative comparison of expression microarray and aSNP in melanoma cell lines, which resulted in the discovery of the MITF oncogene that works in conjunction with the canonical BRAF (V600E) mutation to transform primary human melanocytes. In our approach, we considered the integration of the consensus cancer coding sequence¹⁷ and gene deletion and amplification in conjunction with expression array differences. We were able to elucidate 20 genes from the putative consensus cancer coding sequence¹⁷ that were found in areas of chromosomal loss or gain. We chose to study 4 genes in detail: TGFBR2, RUNX1T1, RFC4, and DLEC1. BRCA2 has been consistently found deleted in approximately 30% of HNSCCs, but without missense mutations.^34- 35 Three genes, CHL1, RTP1, and EIF4A2, showed changes in expression, but little is known regarding their function. Other genes that were not associated with expression differences and had no known previously reported oncogenic function were CLSTN2, CMYA1, KL, MGC21688, PIK3R4, MCF2L2, FLJ10560, KCNB2, and EIF4G1. These genes all have somatic mutations in human neoplasms, so they could be involved in the carcinogenesis of HNSCC in a manner that does not result in expression differences. More interestingly, several other genes with possible oncogenic function in other cancer systems but no expression differences in HNSCC were found: DGKG, LIFR, and ITGA9.

Of the genes we studied in detail, RUNX1T1 is a zinc finger transcription factor protein and a demonstrated oncoprotein in acute myeloid leukemia (AML). It is the functional half of the AML1-ETO (RUNX1T1) fusion gene, under the control of the AML1 promoter.²⁴RUNX1T1 is a putative zinc finger transcription factor and oncoprotein. It resides on an amplification in 8q21.3-8q22.2. RUNX1T1 is one of the most common genetic abnormalities in AML, identified in 15% of all cases. Leukemias are often much less genetically diverse than solid tumors and at times require only single hits for malignant cellular transformation. In vivo studies on variations of RUNX1T1 showed rapid development of leukemia in a normal mouse subjected to retrovirus with this construct.³⁶ Initially, we showed overexpression (above that of normal controls) in 11 of 36 tumor tissues (31%). This was associated with a 50% incidence of amplification (copy number >3.5) and statistically significant overexpression in these samples (P  = .02).

RFC4 forms a complex with PCNA and is involved in DNA replication, DNA repair, DNA modification, and chromatin modeling.²⁵ Recent evidence in cervical cancer demonstrates it is overexpressed.²⁶ Interestingly, RFC4 overexpression has been found to be associated with human papillomavirus and HNSCC tumors.²⁷ We found RFC4 upregulated in 9 of 36 tumors (25%), but only 1 of 5 tumors with significant gene amplification had marked overexpression (20%). This may reflect many confounding features in gene regulation including, but not limited to, interaction with transcription factors, promoter repression, epigenetic alterations, or transcript degradation.

DLEC1 is a putative tumor suppressor involved in carcinogenesis of the lung, esophagus, kidney, and nasopharyngeal cancers.^28- 29 Recently it was found to be the central target controlled by regional epigenetic regulation by an approach that searched for copy-number–independent gene dysregulation.³⁷ This finding was confirmed in ovarian cancer.³⁸ It is a 1755-amino acid polypeptide at 3p21.3. Functionally, transfection of this gene suppressed growth in 4 cancer cell lines by colony focus assay.³⁹ We found deletion in 4 of 12 primary HNSCC (33%), and these had concomitant reduction in expression that was not statistically significant (P = .06).

Although much work remains to elucidate possible contributors to that malignant phenotype of HNSCC, we did find 20 genes in the consensus cancer coding sequence¹⁷ that were associated with areas of chromosomal loss or gain, of which 5 had statistically significant differences in expression (see Table for P values). One pitfall of whole-genomic approaches is the problem of identifying the true signal from the substantial baseline noise generated by such large amounts of data. We used separate discovery and validation cohorts for expression microarray and qRT-PCR to bolster the importance of our findings. This approach represents the attempted discovery and validation of genes in HNSCC that are part of the consensus cancer coding sequence. Further work in the areas of functional experiments remains. Future work includes functional studies of TGFBR2, RUNX1T1, RFC4, and DLEC1 in HNSCC systems. Small molecule inhibitors of these targets may prove to be useful for initial HNSCC tumor studies.

Correspondence: Joseph A. Califano, MD, Department of Otolaryngology–Head and Neck Surgery, Johns Hopkins Medical Institutions, 601 N Caroline St, Sixth Floor, Baltimore, MD 21287-0910 (jcalifa@jhmi.edu).

Submitted for Publication: May 1, 2007; final revision received March 3, 2008; accepted April 7, 2008.

Author Contributions: Drs Smith, Mithani, Liu, Chang, Begum, Westra, and Califano and Ms Dhara had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. Study concept and design: Smith, Sidransky, and Califano. Acquisition of data: Smith, Liu, Chang, Dhara, Westra, and Califano. Analysis and interpretation of data: Smith, Mithani, Chang, Begum, Westra, Sidransky, and Califano. Drafting of the manuscript: Smith, Chang, Dhara, and Califano. Critical revision of the manuscript for important intellectual content: Smith, Mithani, Liu, Begum, Westra, Sidransky, and Califano. Statistical analysis: Sidransky. Obtained funding: Westra and Califano. Administrative, technical, and material support: Smith, Liu, Begum, Dhara, Westra, and Califano. Study supervision: Westra and Califano.

Financial Disclosure: None reported.

Funding/Support: Dr Califano is supported by a Clinical Innovator Award from the Flight Attendant Medical Research Institute, the National Institute of Dental and Craniofacial Research (1R01DE015939-01), and the National Cancer Institute SPORE (5P50CA096784-05). Dr Smith is funded by an institutional NIH T32 grant.

Previous Presentation: This study was presented at the 2007 American Head and Neck Society Meeting at the Combined Otolaryngology Spring Meetings; April 27, 2007; San Diego, California.