Genetic Differences Detected by ComparativeGenomic Hybridization in Head and Neck SquamousCell Carcinomas From Different Tumor Sites:Construction of Oncogenetic Trees for TumorProgression

Qiang Huang,1 Guo Pei Yu,2 Steven A. McCormick,3 Juan Mo,4 Bhakti Datta,3 Manoj Mahimkar,1 Philip Lazarus,5

Alejandro A. Schaffer,6 Richard Desper,6 and Stimson P. Schantz1,7*1Department of Otolaryngology, The New York Eye and Ear Infirmary, New York Medical College, New York, New York2Service of Epidemiology, The New York Eye and Ear Infirmary, New York Medical College, New York, New York3Department of Pathology, The New York Eye and Ear Infirmary, New York Medical College, New York, New York4Division of Basic Sciences, New York University College of Dentistry, New York, New York5Interdisciplinary Oncology Program, Departments of Biochemistry and Pharmacology and Therapeutics, H. Lee Moffitt CancerCenter, University of South Florida, Tampa, Florida6Computational Biology Branch, National Center for Biotechnology Information, National Institutes of Health, Bethesda, Maryland7Strang Cancer Prevention Center, New York, New York

For a better understanding of genetic alterations in head and neck squamous cell carcinoma (HNSCC), we applied comparativegenomic hybridization (CGH) in the analysis of 75 HNSCCs, comprised of 18 pharyngeal squamous cell carcinomas (PSCCs),23 laryngeal squamous cell carcinomas (LSCCs), and 34 oral squamous cell carcinomas (OSCCs). The three subgroups ofHNSCC showed significant differences in genetic alteration patterns. Overall, PSCC and LSCC had more copy numberaberrations (CNAs) per tumor than did OSCC. Apparent differing patterns of high-level amplification were also observed. Thesmallest recurrent chromosomal regions of high-level amplification (�15% of cases) were 7q22, 8q24.1, and 11q12–13 inPSCC and 3q26.1–29 in OSCC. According to single frequency and combined frequencies of CNAs, we concluded that themost important chromosomal events for progression of head and neck cancer were �3q, �5p, �8q, and �3p for allsubgroups of HNSCC; additionally, �7q, �17q, �9p, and �13q for PSCC; �7p, �9q, �11q12–13, �14q, and �17q forLSCC; and �1p and �11q12–13 for OSCC. To identify further important genetic alterations and the relationships among thealterations, we constructed oncogenetic tree models for tumor progression of HNSCC from CGH data using branching anddistance-based tree models. The tree models predicted that: (1) �3q21–29 was the most important early chromosomal event,and �3p, which occurred after �3q21–29, was also an important chromosomal event for all subsites of HNSCC; (2) �8q isthe second most important early chromosomal event; (3) there may be at least three subgroups of HNSCC: one characterizedby �3p, �9p, �7p, and �13q; another by �5p, �9qter, and �17p; and the other by �8q and �18p. These results suggestthat different chromosomal aberrations may play a role in the initiation and/or progression of different subgroups of HNSCC.© 2002 Wiley-Liss, Inc.

INTRODUCTION

Head and neck squamous cell carcinoma(HNSCC), which represents nearly 95% of headand neck cancers, most commonly affects middle-aged or older men who indulge in smoking tobaccoand drinking alcohol (Krajina et al., 1975; Lowry,1975). Although abundant cytogenetic and molec-ular genetic data on HNSCC have been accumu-lated, the genetic mechanisms involved in thepathogenesis and progression of the disease are stillnot clear.

Comparative genomic hybridization (CGH) is apowerful molecular cytogenetic approach, whichwas first described in 1992 (Kallioniemi et al.,

1992). Using only a small amount of DNA and asingle hybridization, CGH can provide detailedinformation on gains and losses of tumor DNAthroughout the entire genome. CGH has beenwidely used for analysis of tumors, includingHNSCC (Brzoska et al., 1995; Speicher et al., 1995;Bockmuhl et al., 1996, 1997, 1998; Hermsen et al.,

*Correspondence to: Dr. Stimson P. Schantz, Department ofOtolaryngology, The New York Eye and Ear Infirmary, 310 East14th Street, New York, NY 10003. E-mail: sschantz@nyee.edu

Received 1 August 2001; Accepted 20 December 2001Published online 8 March 2002

GENES, CHROMOSOMES & CANCER 34:224–233 (2002)DOI 10.1002/gcc.10062

© 2002 Wiley-Liss, Inc.

1997; Komiyama et al., 1997; Weber et al., 1998;Wolff et al., 1998; Okafuji et al., 1999).

CGH analyses of various tumors revealed that atype of cancer has a specific pattern of geneticaberrations. In HNSCC, DNA copy number in-creases were frequently observed on chromosomearms 3q, 5p, 8q, band 11q13, and chromosomes 17and 19, and DNA copy number decreases on chro-mosome arms 1p, 3p, 5q, 6q, 8p, 9p, 13q, 18q, and21q and chromosome 4 (Brzoska et al., 1995;Speicher et al., 1995; Bockmuhl et al., 1996, 1997,1998). However, previous analyses of CGH datafrom HNSCC have several weaknesses: (1) Datafrom different tumor sites within the upper aero-digestive tract were combined. Such an analysisdoes not account for the fact that head and neckcancers are a group of cancers with varied clinical,pathologic, and biological features. Important ge-netic aberrations may be masked by using themixed data. (2) Emphasis has been placed only onthe frequencies of individual chromosomal alter-ations, which may neglect the important role ofcombinations of two or more alterations during theprogression of cancer. To overcome these weak-nesses, we analyzed CGH data according to the siteof disease within the upper aerodigestive tract andcalculated the combined frequencies of two orthree regions of chromosomal alterations (meaningthat at least one of two or three genetic alterationswas observed).

Additionally, we constructed oncogenetic treemodels to identify likely early genetic abnormali-ties and cause-and-effect relationships among thegenetic abnormalities in HNSCC. CGH may revealmany copy number aberrations (CNAs) in a tumorspecimen. However, only some of these aberrationsmay be important for cancer development, and theothers may occur randomly because of the ge-netic instability of cancer cells. Some relevantquestions are:

1. Which CNAs are important and tend to oc-cur early in tumor progression?

2. Which aberrations tend to occur together?3. What are the cause-and-effect relationships

between two or more genetic abnormalities?To address these questions, we developed mathe-matical oncogenetic tree models and applied themto the analyses of CGH data (Desper et al., 1999,2000; Jiang et al., 2000; Kainu et al., 2000). Here,we analyzed our CGH data by using branchingand distance-based tree models. We found that�3q21–29 and �3p are important chromosomalalterations common to all three sites of HNSCC

and that there are other genetic alterations thatoccur predominantly at each site.

MATERIALS AND METHODS

Tumor Samples

This study was reviewed and approved by theInternal Review Board at the New York Eye andEar Infirmary. Seventy-five primary HNSCC sam-ples composed of 18 cases of pharyngeal squamouscell carcinoma (PSCC), 23 cases of laryngeal squa-mous cell carcinoma (LSCC), and 34 cases of oralsquamous cell carcinoma (OSCC) were used in thisstudy. The proportions of various grades of tumordifferentiation were similar in each subgroup. Wedid not include this information in our study be-cause not all staging information was uniformlyacquired. Normal DNA was prepared from humanplacenta (46,XY). Tumor specimens were obtainedfrom fresh-frozen or formalin-fixed, paraffin-em-bedded tissue. Tissue blocks were processed toensure a minimum of 75% tumor cells in eachspecimen. Sections from fresh/frozen tumor tissuewere verified by a pathologist. The tumor areascomposed of a minimum of 90% tumor cells weremicrodissected (Huang et al., 2000). DNAs wereprepared by proteinase K digestion and phenol/chloroform extraction (Sambrook et al., 1998). Forformalin-fixed, paraffin-embedded tumor samples,the DNA was prepared as previously described(Huang et al., 2000).

DOP-PCR

Minute specimens that were not sufficient foranalysis with conventional CGH were amplifiedby degenerate oligonucleotide-primed polymerasechain reaction (DOP-PCR) before performingCGH. DOP-PCR was performed in two separatesteps according to our previously detailed protocol(Huang et al., 2000). Briefly, in step I, the totalvolume was 10 �l, which consisted of 100 pg DNA;0.2 mM of each dNTP; 1 �M UNI-primer (Tele-nius et al., 1992); 0.4 unit Thermo Sequenase DNApolymerase (Amersham Life Science, Cleveland,OH); and 1 �l Thermo Sequenase reaction buffer(260 mM Tris–HCl, pH 9.5; 65 mM MgCl2). PCRconditions consisted of 3 min at 95°C, followed by4 cycles of 1 min at 94°C, 1 min at 25°C, 3 mintransition at 25–74°C, 2 min extension at 74°C, anda final extension of 10 min. In step II, the totalvolume was 50 �l, which consisted of 10 �l ofproduct from the first step; 0.16 mM of eachdNTP; 1.2 �M UNI-primer; 0.1 unit AmpliTaqDNA polymerase (Perkin Elmer, Branchburg, NJ);

225DISTINCT GENETIC PATTERNS OF HNSCC SUBSETS

Figure 1.

and 4 �l 10� PCR buffer (100 mM Tris–HCl, pH8.3; 500 mM KCl; 15 mM MgCl2; and 0.01% gela-tin; Perkin Elmer). The amplification conditionswere 3 min at 95°C, followed by 35 cycles of 1 minat 94°C, 1 min at 56°C, 2 min extension at 72°C,and a final extension of 10 min.

DNA Labeling

The DOP-PCR products were precipitated withethanol before labeling. Normal genomic or ampli-fied DNA was labeled with Texas Red-5-dUTP,and tumor genomic or amplified DNA was labeledwith fluorescein-12-dUTP (NEN Life ScienceProducts, Boston, MA) using standard nick-transla-tion procedures (Kallioniemi et al., 1994).

CGH Analysis

CGH analyses were performed as described pre-viously (Kallioniemi et al., 1994). Briefly, equalamounts of the labeled test DNA and referenceDNA were mixed and hybridized in the presenceof 20 �g (for standard CGH) to 40 �g (for DOP-PCR-CGH) of human Cot-1 DNA (GIBCO/BRL,Grand Island, NY) for 2 days to normal humanmetaphase chromosomes (Vysis, Downers Grove,IL). It is important that relevant reference DNAmust be amplified for cohybridizing with amplifiedtest DNA. After hybridization, the slides werewashed and the chromosomes were counterstainedwith DAPI (Oncor, Gaithersburg, MD). Imageswere captured with a cooled charge-coupled devicecamera (Photometrics, Tucson, AZ) connected to aZeiss Axioskop fluorescence microscope (Zeiss,Thornwood, NY). These images were analyzedwith Quips GGH software (Vysis). Regions ofDNA gain and loss were determined by calculatinggreen:red average ratio profiles from at least 10metaphase cells. If the average ratio was �1.2 or�0.8, the region of alteration was considered as again or a loss, respectively (Brzoska et al., 1995). Inour control CGH experiments using normal–nor-mal DNAs, the green:red ratios of all chromosomesdid not exceed the range of 0.8–1.2. A high-levelamplification was defined if the ratio exceeded 1.5(Bockmuhl et al., 1996). We ignored the Y chromo-some because it cannot be reliably analyzed byCGH.

Construction of Oncogenetic Trees

Based on the CGH data, we constructed branch-ing trees and distance-based trees for HNSCC asdescribed in previous publications (Desper et al.,1999, 2000). Briefly, we used CNAs (i.e., gains andlosses) from CGH analysis as the input for thetree-modeling procedure. A small set of events thatoccurred sufficiently frequently to appear nonran-dom was selected by a well-known method (Bro-deur et al., 1982). Brodeur’s method takes intoaccount the size of each 1-digit chromosomal band;these sizes were derived from the study by Daniel(1985). To give the data a consistent format and toenable us to generate the null distribution requiredby Brodeur’s method, we divided the genome intosingle-digit bands. For the tree construction, theCGH profile was recorded as presence/absence of again and presence/absence of a loss in each single-digit band. The selected nonrandom events wereused for the construction of branching tree anddistance-based tree models by using freely avail-able software (see http://www.ncbi.nlm.nih.gov/CBBresearch/Schaffer/cgh.html). We did not con-struct tree models for each subgroup of HNSCCsbecause of the limitation of small sample numbers.

RESULTS

Genetic Alteration Pattern of HNSCC

CGH was performed on 75 primary HNSCCsamples. The most frequent DNA copy numbergains were observed on chromosome arms 3q (50/75, 67%), 8q (36/75, 48%), 5p (30/75, 40%), and 17q(26/75, 35%), band 11q13 (24/75, 32%), and 9q(23/75, 31%), 7q (21/75, 28%), 22q (17/75, 23%), 7p(16/75, 21%), 12p (16/75, 21%), 1p (14/75, 19%), aswell as chromosome 19 (14/75, 19%). DNA copynumber decreases occurred most frequently at 3p(35/75, 47%), 9p (21/75, 28%), 4q (18/75, 24%), 13q(18/75, 24%), and 11q (15/75, 20%).

Genetic Alteration Patterns of Subsets of HNSCC

The CGH data for HNSCC were analyzed ac-cording to the site of disease within the upperaerodigestive tract. Patterns of genetic alterationfor PSCC and LSCC were similar and significantlydifferent from those for OSCC (Fig. 1). The mostcommon gains involved 3q, 7q, 8q, 5p, 11q12–13,and 17q in PSCC; 3q, 17q, 8q, 5p, 7p, 7q, and 9q inLSCC; and 3q, 8q, 5p, 11q12–13, and 9q in OSCC.The most common losses involved 3p, 9p, 13q, 4q,and 11q14–25 in PSCC; 3p and 9p in LSCC; and3p in OSCC. Figure 2 compares genetic alterationsamong PSCCs, LSCCs, and OSCCs. Although

Figure 1. Summary of all genetic alterations detected by CGH inpharyngeal squamous cell carcinoma, laryngeal squamous cell carci-noma, and oral squamous cell carcinoma. Vertical lines on the left of thechromosome ideograms indicate DNA copy number losses; lines on theright, DNA copy number gains; wide bars, high-level amplification.

227DISTINCT GENETIC PATTERNS OF HNSCC SUBSETS

most chromosomal alterations occurred with similarfrequencies in all three subsets of HNSCC, somechromosomal changes occurred predominantly inspecific subsites. This included gain of 7q (50%) inPSCC and gain of 17q (52%) in LSCC. The fre-quencies of these gains were significantly morefrequent compared to that of OSCC: 12 and 21%,respectively (P � 0.05, two-tailed Fisher’s exacttest). Furthermore, losses occurred more fre-quently in PSCC. For example, 50% of PSCCsshowed losses on 9p and 13q. In contrast, losses of9p and 13q occurred in only 15 and 12% of OSCCs,respectively (P � 0.05). Additionally, abnormalitiesinvolving chromosome 6 were more frequent inPSCCs; gains and/or losses were identified in 56%of PSCCs, 9% of LSCCs, and 18% of OSCCs. Lossof chromosome 6 was observed in 22% of PSCCsbut only 6% of OSCCs. In contrast to PSCCs, noneof the LSCCs showed loss of this chromosome.

Furthermore, we found that the three subgroupsof HNSCC had distinct patterns of high-level am-plification. Cancers with high-level amplification,from one region to more than five regions, com-

prised 61, 35, and 50% of the samples in PSCC,LSCC, and OSCC, respectively. Cancers withhigh-level amplification (HA-positive) usually ex-hibited more CNAs than did those without high-level amplification (HA-negative). More than 50%of HA-positive tumors showed CNA numbershigher than the median of total CNAs. The differ-ence was statistically significant in OSCC (Table1). The mean number of CNAs was 6 in the entiresubset of OSCC, 11 in the HA-positive, and 5 inthe HA-negative group. The smallest common re-gions of high-level amplification are summarized inTable 2. Recurrent high-level amplification wasdefined as an amplification occurring in more thanor equal to 15% of a tumor group. Recurrent re-gions of high-level amplification are 7q22, 8q24.1,and 11q12–13 in PSCC, and 3q26.1–29 in OSCC.

Correlations Between Copy-Number Aberrations(CNAs) and Tumor Site

The number of CNAs ranged from 1 to 22(mean � SD: 11.4 � 5.9) for PSCC, from 2 to 20(9.5 � 4.9) for LSCC, and from 1 to 20 (7.6 � 5.5)

Figure 2. Comparison of genetic alterations among pharyngeal squamous cell carcinoma, laryngealsquamous cell carcinoma, and oral squamous cell carcinoma.

228 HUANG ET AL.

for OSCC (Fig. 3). These differences in CNA num-bers when comparing tumor sites were significant[ANOVA, F(2, 66) � 3.45, P � 0.0375]. UsingTukey’s studentized range test for multiple com-parisons, we found that there was significant dif-ference between PSCC and OSCC (P � 0.05), butnot between LSCC and PSCC or LSCC andOSCC.

Combined Frequencies for Multiple ChromosomalAlterations in HNSCC

We calculated the combined frequencies for twoor three regions of chromosomal alterations, accord-ing to the following equation: combined frequen-cies � positive cases total number of cases �

100%. A positive case is one that shows geneticalterations at one or more given regions. The pat-terns of high combined frequencies are shown inTable 3. Ninety-one percent of LSCCs showed again of 3q and/or a loss of 3p, which may implyabnormalities on these chromosomal arms are themost important chromosomal events for the initia-tion and/or progression of LSCC. It is of note that�1pter and �3q are mutually exclusive CNAs inOSCC. One hundred percent (18/18) of �3q-pos-itive (i.e., with gain of 3q) OSCCs were �1pter-negative (i.e., without gain of 1pter). Likewise,100% (7/7) �1pter-positive OSCC were �3q-neg-ative.

Oncogenetic Trees

We constructed both a branching tree and adistance-based tree for HNSCC. The relative ad-vantages and limitations of the two tree modelswere previously described (Desper et al., 2000). Inthe branching tree model, there is one node calleda root, and the other nodes are CNAs. A CNA nearthe root is predicted to be an early event. CNAsthat have occurred together in the tumors are clus-tered in a subtree of the entire tree. If two CNAs areconnected to each other by an edge, then theCNAs have a hypothetical cause-and-effect rela-tionship. The edge lengths are irrelevant, and theorder of edges coming out of a node is arbitrary; theleft-to-right order of nodes is a prediction as tothe preferred order of CNAs.

The distance-based tree model is based in parton well-known methods for fitting a distance ma-trix to a tree. To fit the resulting matrix to a phy-logenetic tree, we constructed distance-based treesusing the Fitch (Fitch and Margoliash, 1967) andNeighbor (Saitou and Nei, 1987) programs inPHYLIP (Felsenstein, 1989). We found that theNeighbor tree gave a better fit to the input matrix,and thus it was used in this study. In the distance-based tree model, the CNAs are all leaves of the

TABLE 1. Correlations Between CNAs and High-Level Amplification in HNSCC

Geneticalteration

Positivea Negativeb

Pvalues

Totalcases

Cases with CNAs� medianc (%)

Totalcases

Cases with CNAs� medianc (%)

PSCC 11 6 (54.5) 7 2 (28.6) 0.366LSCC 8 4 (50.0) 15 6 (40.0) 0.685OSCC 17 12 (70.6) 17 4 (23.5) 0.016

aPositive, cancers with high-level amplification.bNegative, cancers without high-level amplification.cMedian: 11 (for PSCC); 9 (for LSCC); and 6 (for OSCC).

TABLE 2. Regions of High-Level Amplification in HNSCC

Chromo-some PSCC LSCC OSCC

2 None 2p23–25 (1) 2q31–32 (1)3 3q26.1–29 (1)a 3q26.2 (2) 3q26.1–29b (6)5 None 5p15 (2) None6 6p12–21.1 (1) None None

6q22 (1)7 7q22 (3) 7q11.2 (1) 7q11.2–22 (1)8 8q24.1 (3) None 8q11.2 (1)

8q24.1–24.3 (3)9 9q33–34 (1) 9q34 (1) 9p21 (2)10 10p (1) None None11 11p11.2 (1) 11q13 (2) 11p11.2–13 (1)

11q12–13 (3) 11q12–13 (1)11q21 (1)

12 12p (1) 12q14 (1) 12p13 (2)12q14 (1) 12q15 (1) 12q14 (1)

13 None None 13q21 (1)14 14q24–32 (1) None17 17q25 (2) 17q11.2–12 (2) 17q24–25 (1)

17q24–25 (1)18 None None 18p11.3 (3)20 None 20p11.2 (1) NoneX None None Xq26–28 (1)

aThe numbers in parentheses are the numbers of cases of high-levelamplification.bThe bold letters represent the recurrent regions of high-level amplifi-cation.

229DISTINCT GENETIC PATTERNS OF HNSCC SUBSETS

tree, and the internal vertices are hypothetical hid-den events. Likewise, early events are those nearthe root, and the sets of events should clustertogether in subtrees. However, the edge lengths ofthe distance-based tree are meaningful. The dis-tance from the root to a node is the sum of thehorizontal edge lengths. The vertical part of eachedge is included only to show the tree clearly.

We compared the two tree models to make pre-dictions of early important chromosomal events,

late events, and potential causality between events.The tree models are shown in Figure 4. Using thesingle-digit chromosome band representation ofthe data, Brodeur’s method selected 11 CNAs asoccurring significantly more frequently than wouldbe expected at random: �3q2, �5p1, �7p2, �8q2,�9q3, �17p1, �18p1, �3p1, �3p2, �9p2, �13q3.Because of the single-digit band representation,�3p1 and �3p2 are treated as distinct CNAs fortree modeling, even though they occur as a singleaberration in many tumors. The event �3q2 isadjacent to the roots of both trees and is the root ofa large branching subtree. This predicts that �3q2is the most important early chromosomal event inthe development of HNSCC. The event �3p2 thatis adjacent to �3q2 and placed centrally in thetrees is predicted to be another important chromo-somal event, which occurred after �3q2. Theevent �8q2 is a leaf off the root in the branchingtree, and close to the root in the distance-basedtree. This indicates that �8q2 is the second mostimportant early chromosomal event. Because cor-related events are clustered together in a distance-based subtree, the tree model predicts that theremay be at least three subgroups of HNSCC: onecharacterized by the events of �3p2, �3p1, �9p2,

Figure 3. Distribution of the total number of CNAs in pharyngeal squamous cell carcinoma, laryngealsquamous cell carcinoma, and oral squamous cell carcinoma. Each dot represents one case. Horizontaldotted lines indicate the averages of CNAs.

TABLE 3. Comparison of Combined Frequencies (%) ofChromosomal Regions of Genetic Alterations in PSCC,

LSCC, and OSCC

Genetic alteration PSCC LSCC OSCC

Two regions�3q/�3p 83 91 65

Three regions�3q/�3p/�1p 89 91 82�3q/�3p/�5p 94 91 79�3q/�3p/�8q 94 96 74�3q/�3p/�7p 89 96 65�3q/�3p/�9q 89 96 71�3q/�3p/�11q 89 96 74�3q/�3p/�14q 83 100 65�3q/�3p/�17q 89 96 71

230 HUANG ET AL.

�7p2, and �13q3; another by �5p1, �9q3, and�17p1; and still another by �8q2 and �18p1.

DISCUSSION

As a powerful genomic screening technique intumor genetics, CGH has been widely used foranalyses of various tumors, including HNSCC. Inour study, the genetic alteration pattern in HNSCCis consistent with published CGH data (Brzoska etal., 1995; Speicher et al., 1995; Bockmuhl et al.,1996). The most frequent changes were loss of 3pand gains of 3q, 8q, 5p, and 17q; specifically, over-representation of 3q26–29.

To our knowledge, more than 20 CGH analyseson head and neck cancers and cell lines have beenpublished to date, in which the majority combinedcancer specimens from all subsites within the up-per aerodigestive tract. Actually, HNSCC is a het-erogeneous group of cancers. Significant geneticalterations relevant to the progression of diseasewithin a specific subsite may be masked by com-bining of data from all sites. Therefore, we ana-lyzed our CGH data on HNSCC on the basis of site

of disease within the upper aerodigestive tract. Ourfindings indicate that subsets of HNSCC have dif-ferent genetic patterns that are distinguishable bysite of disease within the upper aerodigestive tract.Specifically, the numbers of CNAs for PSCC andLSCC are similar to, but significantly higher than,those for OSCC (P � 0.05 for PSCC vs. OSCC).However, the numbers of CNAs in the entireOSCC subset in our study were much lower thanthose in previous studies (Gebhart et al., 1998;Weber et al., 1998). We found that high-level am-plification was associated with higher CNAs pertumor, especially in OSCC (P � 0.05). The meannumbers of CNAs are 11 and 5 in HA-positive andHA-negative OSCC, respectively. Both total CNAsper tumor and high-level amplifications of 8q and20q13 were observed to be associated with recur-rence of breast cancers (Isola et al., 1995). It may beworthwhile to study the relationship between spe-cific regions of high-level amplification or the num-ber of CNAs and the prognosis of HNSCC.

Increasing evidence indicates that tumorsprogress through the continuous accumulation of

Figure 4. Branching tree (A) and distance-based tree (B) for all HNSCCs.

231DISTINCT GENETIC PATTERNS OF HNSCC SUBSETS

genetic and epigenetic alterations, and that mosthuman cancers contain multiple mutations. It isestimated that HNSCC arises after the accumula-tion of 6 to 10 independent genetic events (Renan,1993). CGH analyses have revealed that multiplegenetic aberrations occur during tumorigenesis.However, it is unclear which changes are prereq-uisites for the formation and progression of cancersand which alterations are simply accumulated as aresult of an increasingly unstable genome. Basedon both single frequency (Fig. 1) and combinedfrequencies of CNAs (Table 3) we suggest that,except for gains on 3q, 5p, and 8q, and loss on 3pfor HNSCC within all sites within the upper aero-digestive tract, some chromosomal alterations maybe more important for the progression of differentsubsites: �7q, �17q, �9p, and �13q for PSCC;�7p, �9q, �11q12–13, �14q, and �17q forLSCC; and �1p and �11q12–13 for OSCC.

Our results were partially in agreement with theprevious observation for LSCC (Kujawski et al.,1999). More recurrent abnormalities, however,were observed in our study. A frequent gain onchromosome 17 for PSCC was not observed in aprevious study (Welkoborsky et al., 2000). Recur-rent gain of 11q13 for OSCC was also observed inprevious studies (Gebhart et al., 1998; Wolff et al.,1998). Interestingly, 100% (7/7) of �1p-positiveOSCC were �3q-negative. Moreover, although thefrequencies of �3q and �1pter were 53 and 21%,respectively, the combined frequencies of the twoalterations increased to 74%, the highest combinedfrequencies of any two regions within OSCC. Ourstudy suggested that gain of 1pter plays an impor-tant role in the carcinogenesis of �3q-negativeOSCC. Gain of 1pter was also observed frequentlyin previous CGH analyses of OSCC (Gebhart et al.,1998; Wolff et al., 1998).

To explore more thoroughly the association ofvarious genetic mutational events and their rela-tionship to disease progression, we applied mathe-matical oncogenetic tree models to analyze ourCGH data. Our tree models predicted the follow-ing outcomes: (1) �3q2 was the most importantearly chromosomal event for development ofHNSCC; (2) �3p2, �3p1, �8q2, and �5p1 werealso important chromosomal events; (3) there maybe at least three subgroups of HNSCC (Fig. 4).Interestingly, high-level amplifications of 5p and18p occurred exclusively in LSCCs and OSCCs,respectively. Losses of 9p and 13q were observedfrequently in PSCCs but in less than 15% of theOSCCs.

Our findings that �3q and �3p are early chro-mosomal events in tumor development are consis-tent with those of previous studies (Bockmuhl etal., 1996; Heselmeyer et al., 1996). Loss of theentire 3p accompanied by gain of the whole 3q armwas observed in 10 cases, suggesting the formationof an isochromosome i(3)(q10) in some cancers.Overrepresentation of 3q has also been observed inother tumor types, such as esophageal squamouscell carcinoma (Tada et al., 2000), mantle cell lym-phoma (Bea et al., 1999), prostate cancer (Sattler etal., 1999), lung cancer (Bjorkqvist et al., 1998; Lu etal., 1999), cervical squamous cell carcinoma (Kirch-hoff et al., 1999), and ovarian cancer (Arnold et al.,1996). This suggests that chromosome arm 3q maycontain one or more oncogenes important for tu-morigenesis and/or progression of these tumors.This region is known to contain oncogenes such asBCL6/LAZ3, ECT2, EVI1, RAB7, and THPO. How-ever, none of the above-mentioned genes has beendescribed to be involved in tumorigenesis ofHNSCC. The AIS gene, a TP53 homolog, wasrecently observed to be amplified in squamous cellcarcinomas of the lung and head and in neck cancercell lines (Hibi et al., 2000).

The high-level amplification regions are differ-ent between PSCC, LSCC, and OSCC (Table 2).Amplification of chromosome band 11q13 has beenfound to be associated with the tumor site; it oc-curred predominantly in hypopharyngeal squa-mous cell carcinoma (Muller et al., 1997). In thisstudy, high-level amplification of 11q13 was fre-quently seen in PSCCs, which is in agreement withthe previous observation. Further studies examin-ing these specific high-level amplification regionsare necessary to identify the oncogenes that playcrucial roles in tumor initiation and/or tumor pro-gression of different subsets of HNSCC.

Diversity in head and neck cancers may reflectdifferences in etiologic exposures, individual hostgenetic susceptibility, as well as differences in em-bryologic and histologic characteristics of the upperaerodigestive mucosa. Such an analysis is currentlybeing undertaken.

In summary, this study demonstrates that signif-icant genetic differences exist among PSCC,LSCC, and OSCC, suggesting that HNSCC is aheterogeneous group of tumors with different ge-netic alterations and etiology. Studies should beinitiated that examine the influence of heritablehost characteristics and environmental exposureson particular mutational profiles identified withinsquamous cell cancers of the upper aerodigestive

232 HUANG ET AL.

tract. CGH represents a powerful tool to furthersuch molecular epidemiologic analyses.

REFERENCES

Arnold N, Hagele L, Walz L, Schempp W, Pfisterer J, Bauknecht T,Kiechle M. 1996. Overrepresentation of 3q and 8q material andloss of 18q material are recurrent findings in advanced humanovarian cancer. Genes Chromosomes Cancer 16:46–54.

Bea S, Ribas M, Hernandez JM, Bosch F, Pinyol M, Hernandez L,Garcıa JL, Flores T, Gonzalez M, Lopez-Guillermo A, Piris MA,Cardesa A, Montserrat E, Miro R, Campo E. 1999. Increasednumber of chromosomal imbalances and high-level DNA ampli-fications in mantle cell lymphoma are associated with blastoidvariants. Blood 93:4365–4374.

Bjorkqvist A-M, Husgafvel-Pursiainen K, Anttila S, Karjalainen A,Tammilehto L, Mattson K, Vainio H, Knuutila S. 1998. DNA gainsin 3q occur frequently in squamous cell carcinoma of the lung, butnot in adenocarcinoma. Genes Chromosomes Cancer 22:79–82.

Bockmuhl U, Schwendel A, Dietel M, Petersen I. 1996. Distinctpatterns of chromosomal alterations in high- and low-grade headand neck squamous cell carcinomas. Cancer Res 56:5325–5329.

Bockmuhl U, Petersen S, Schmidt S, Wolf G, Jahnke V, Dietel M,Petersen I. 1997. Patterns of chromosomal alterations in metasta-sizing and nonmetastasizing primary head and neck carcinomas.Cancer Res 57:5213–5216.

Bockmuhl U, Wolf G, Schmidt S, Schwendel A, Jahnke V, Dietel M,Petersen I. 1998. Genomic alterations associated with malignancyin head and neck cancer. Head Neck 20:145–151.

Brodeur GM, Tsiatis AA, Williams DL, Luthardt FW, Green AA.1982. Statistical analysis of cytogenetic abnormalities in humancancer cells. Cancer Genet Cytogenet 7:137–152.

Brzoska PM, Levin NA, Fu KK, Kaplan MJ, Singer MI, Gray JW,Christman MF. 1995. Frequent novel DNA copy number increase insquamous cell head and neck tumors. Cancer Res 55:3055–3059.

Daniel A. 1985. The size of prometaphase chromosome segments.Clin Genet 28:216–224.

Desper R, Jiang F, Kallioniemi O-P, Moch H, Papadimitriou CH,Schaffer AA. 1999. Inferring tree models for oncogenesis fromcomparative genome hybridization data. J Comput Biol 6:37–51.

Desper R, Jiang F, Kallioniemi O-P, Moch H, Papadimitriou CH,Schaffer AA. 2000. Distance-based reconstruction of tree modelsfor oncogenesis. J Comput Biol 7:789–803.

Felsenstein J. 1989. PHYLIP: Phylogeny Inference Package (ver-sion 3.2). Cladistics 5:164–166. [See also http://evolution.genetics.washington.edu/phylip.html]

Fitch WM, Margoliash E. 1967. Construction of phylogenetic trees.Science 155:279–284.

Gebhart E, Liehr T, Wolff E, Ries J, Fiedler W, Steininger H,Koscielny S, Girod S. 1998. Pattern of genomic imbalances in oralsquamous cell carcinomas with and without an increased copynumber of 11q13. Int J Oncol 12:1151–1155.

Hermsen MA, Joenje H, Arwert F, Braakhuis BJ, Baak JP, Wester-veld A, Slater R. 1997. Assessment of chromosomal gains andlosses in oral squamous cell carcinoma by comparative genomichybridisation. Oral Oncol 33:414–418.

Heselmeyer K, Schrock E, du Manoir S, Blegen H, Shah, K, Stein-beck R, Auer G, Ried T. 1996. Gain of chromosome 3q definesthe transition from severe dysplasia to invasive carcinoma of theuterine cervix. Proc Natl Acad Sci USA 93:479–484.

Hibi K, Trink B, Patturajan M, Westra WH, Caballero OL, Hill DE,Ratovitski EA, Jen J, Sidransky D. 2000. AIS is an oncogeneamplified in squamous cell carcinoma. Proc Natl Acad Sci USA97:5462–5467.

Huang Q, Schantz SP, Rao PH, Mo J, McCormick SA, Chaganti RSK.2000. Improving degenerate oligonucleotide primed PCR-compara-tive genomic hybridization for analysis of DNA copy numberchanges in tumors. Genes Chromosomes Cancer 28:395–403.

Isola JJ, Kallioniemi O-P, Chu LW Fuqua SAJ, Hilsenbeck SG,Osborne CK, Waldman FM. 1995. Genetic aberrations detectedby comparative genomic hybridization predict outcome in node-negative breast cancer. Am J Pathol 147:905–911.

Jiang F, Desper R, Papadimitriou CH, Schaffer AA, Kallioniemi O-P,Richter J, Schraml P, Sauter G, Mihatsch MJ, Moch H. 2000. Con-struction of evolutionary tree models for renal cell carcinoma fromcomparative genomic hybridization data. Cancer Res 60:6503–6509.

Kainu T, Juo SH, Desper R, Schaffer AA, Gillanders E, RozenblumE, Freas-Lutz D, Weaver D, Stephan D, Bailey-Wilson J, Kal-lioniemi O-P, Tirkkonen M, Syrjakoski K, Kuukasjarvi T,

Koivisto P, Karhu R, Holli K, Arason A, Johannesdottir G, Bergth-orsson JT, Johannsdottir H, Egilsson V, Barkardottir RB, Johanns-son O, Haraldsson K, Sandberg T, Holmberg E, Gronberg H,Olsson H, Borg Å, Vehmanen P, Eerola H, Heikkila P, PyrhonenS, Nevanlinna H. 2000. Somatic deletions in hereditary breastcancers implicate 13q21 as a putative novel breast cancer suscep-tibility locus. Proc Natl Acad Sci USA 97:9603–9608.

Kallioniemi A, Kallioniemi O-P, Sudar D, Rutovitz D, Gray JW,Waldman F, Pinkel D. 1992. Comparative genomic hybridizationfor molecular cytogenetic analysis of solid tumors. Science 258:818–821.

Kallioniemi O-P, Kallioniemi A, Piper J, Isola J, Waldman FM, GrayJW, Pinkel D. 1994. Optimizing comparative genomic hybridiza-tion for analysis of DNA sequence copy number changes in solidtumors. Genes Chromosomes Cancer 10:231–243.

Kirchhoff M, Rose H, Petersen BL, Maahr J, Gerdes T, LundsteenC, Bryndorf T, Kryger-Baggesen N, Christensen L, EngelholmSA, Philip J. 1999. Comparative genomic hybridization reveals arecurrent pattern of chromosomal aberrations in severe dysplasia/carcinoma in situ of the cervix and in advanced-stage cervicalcarcinoma. Genes Chromosomes Cancer 24:144–150.

Komiyama T, Matsumura K, Tsuchida N. 1997. Comparison ofDNA copy numbers in original oral squamous cell carcinomas andcorresponding cell lines by comparative genomic hybridization.Jpn J Cancer Res 88:476–483.

Krajina Z, Kulcar Z, Konic-Carnelutti V. 1975. Epidemiology oflaryngeal cancer. Laryngoscope 85:1155–1161.

Kujawski M, Aalto Y, Jaskula-Sztul R, Szyfter W, Szmeja Z, SzyfterK, Knuutila S. 1999. DNA copy number losses are more frequentin primary larynx tumors with lymph node metastases than intumors without metastases. Cancer Genet Cytogenet 114:31–34.

Lowry WS. 1975. Alcoholism in cancer of the head and neck.Laryngoscope 85:1275–1280.

Lu Y-J, Dong X-Y, Shipley J, Zhang R-G, Cheng S-J. 1999. Chro-mosome 3 imbalances are the most frequent aberration found innon-small cell lung carcinoma. Lung Cancer 23:61–66.

Muller D, Millon R, Velten M, Bronner G, Jung G, Engelmann A,Flesch H, Eber M, Methlin G, Abecassis J. 1997. Amplification of11q13 DNA markers in head and neck squamous cell carcinomas:correlation with clinical outcome. Eur J Cancer 33:2203–2210.

Okafuji M, Ita M, Hayatsu Y, Shinozaki F, Oga A, Sasaki K. 1999.Identification of genetic aberrations in cell lines from oral squa-mous cell carcinomas by comparative genomic hybridization.J Oral Pathol Med 28:241–245.

Renan MJ. 1993. How many mutations are required for tumorigenesis?Implications from human cancer data. Mol Carcinog 7:139–146.

Saitou N, Nei M. 1987. The neighbor-joining method: a new methodfor reconstructing phylogenetic trees. Mol Biol Evol 4:406–425.

Sambrook J, Fritsch EF, Maniatis T. 1998. Molecular cloning: alaboratory manual. Cold Spring Harbor, NY: Cold Spring HarborLaboratory Press.

Sattler H-P, Rohde V, Bonkhoff H, Zwergel T, Wullich B. 1999.Comparative genomic hybridization reveals DNA copy number gainsto frequently occur in human prostate cancer. Prostate 39:79–86.

Speicher MR, Howe C, Crotty P, du Manoir S, Costa J, Ward DC.1995. Comparative genomic hybridization detects novel deletionsand amplifications in head and neck squamous cell carcinomas.Cancer Res 55:1010–1013.

Tada K, Oka M, Tangoku A, Hayashi H, Oga A, Sasaki K. 2000.Gains of 8q23-qter and 20q and loss of 11q22-qter in esophagealsquamous cell carcinoma associated with lymph node metastasis.Cancer 88:268–273.

Telenius H, Carter NP, Bebb CE, Nordenskjold M, Ponder BAJ,Tunnacliffe A. 1992. Degenerate oligonucleotide-primed PCR:general amplification of target DNA by a single degenerateprimer. Genomics 13:718–725.

Weber RG, Scheer M, Born IA, Joos S, Cobbers JMJL, Hofele C,Reifenberger G, Zoller JE, Lichter P. 1998. Recurrent chromo-somal imbalances detected in biopsy material from oral premalig-nant and malignant lesions by combined tissue microdissection,universal DNA amplification, and comparative genomic hybrid-ization. Am J Pathol 153:295–303.

Welkoborsky HJ, Bernauer HS, Riazimand HS, Jacob R, Mann WJ,Hinni ML. 2000. Patterns of chromosomal aberrations in metastasiz-ing and nonmetastasizing squamous cell carcinomas of the orophar-ynx and hypopharynx. Ann Otol Rhinol Laryngol 109:401–410.

Wolff E, Girod S, Liehr T, Vorderwulbecke U, Ries J, Steininger H,Gebhart E. 1998. Oral squamous cell carcinomas are characterizedby a rather uniform pattern of genomic imbalances detected bycomparative genomic hybridisation. Oral Oncol 34:186–190.

233DISTINCT GENETIC PATTERNS OF HNSCC SUBSETS