We retrospectively selected 18 patients who were diagnosed with HNSCC and underwent primary tumor resection and neck dissection followed by CCRT at the Department of Otolaryngology-Head and Neck Surgery from April 2008 to June 2012. Tumor stage was determined according to the standard tumor-node-metastasis classification of AJCC guidelines (7th edition). Frozen tissues (tumor and adjacent normal tissue pairs) were obtained from the patients with previously untreated HNSCC. Indications for adjuvant CCRT are as follows: positive or close margins found on the resection, advanced T stage, lymphovascular invasion, perineural invasion, multiple nodal metastasis, or extracapsular spread. All 18 HNSCC patients received postoperative radiotherapy (RT) that consisted of conventionally fractionated doses of 1.8–2 Gy each in five weekly sessions. A volume including the primary tumor site and all the draining lymph nodes at risk received a dose of 50–55 Gy. Regions that had an inadequate resection margin, extracapsular nodal spread or the initially involved lymph node area were boosted up to 65 Gy. The patients received chemotherapy that consisted of 100 mg cisplatin/m² of body-surface area on days 1, 22 and 43 during RT or weekly 30 mg cisplatin/m² of body surface area. The Institutional Review Board of Seoul St. Mary's Hospital approved the retrospective review of the medical records and the use of archived tumor specimens.

For array-CGH analysis, 10-µm-thick frozen sections of tumor cell-rich areas (>60%) were microdissected. Genomic DNA was extracted from these sections using a DNeasy Blood and Tissue kit (Qiagen, Hilden, Germany). The Agilent SurePrint G3 Human CGH Microarray kit, 4×180K (Agilent technologies, Santa Clara, CA, USA) was used for the array-CGH analysis. This array contains 180,880 probes with a median probe spacing of 13 kb. The array hybridization procedure was performed as described elsewhere (12). In brief, 1 µg of genomic DNA (gDNA) from tumor tissue was labeled with Cy5-dCTP and 1 µg of gDNA from paired normal tissue with Cy3-dCTP. After purifying the labeled product by using Amicon Ultra-0.5 centrifugal Filter Device (Millipore, Billerica, MA, USA), the labeled DNA was applied on the Agilent 4×180K array with the blocking agent (Agilent technologies, Santa Clara, CA, USA) containing 50 µg of human Cot-1 DNA (HybMasker, ConnectaGen, Korea). Array slides were incubated for 24 h at 65°C in the hybridization oven. After the hybridization, the arrays were washed with wash buffer 1 and 2 according to the manufacturer's protocol and scanned using High-Resolution Microarray Scanner (Agilent technologies).

Microarray hybridization images were analyzed with Feature Extraction Software version 10.7.3.1 using CGH-107-Sep09 protocol for normalization as described previously (12). Data quality was assessed using CGH_QCTM_Sep09 QC metric set. Only the data that passed the quality check (in good or excellent range) was used for CNA calling. CNAs of samples were defined using the ADM-2 algorithm in Agilent Genomic Workbench Lite Edition 6.0 software (Agilent technologies). CNAs were defined using the following criteria; a minimum number of probes in region=3, a maximum number of probes in aberrant region=100,000, a signal intensity ratio of >0.3 in log₂ scale for gains, and of <−0.3 in log₂ scale for losses. High-level amplification was defined as a probe signal intensity ratio of 1 or higher in log₂ scale. Likewise, a homozygous deletion (HD) was defined as a ratio of -1 or lower in log₂ scale. After defining CNAs, recurrently altered regions (RAR) were defined using the CNVRuler program as chromosomal segments consisting of multiple overlapping CNAs found in ≥30% of the samples (13).

We performed qPCR validation of candidate RARs using the gDNA extracted from 18 pairs of head and neck cancer/normal tissues. As a diploid internal control, we used a genomic region on chromosome 13 (13q32.1) which was observed to have no genomic alteration in the array-CGH data. Sequence and location of primers used for the genomic qPCR experiment are as follows. RAR-G2 (7p12.1-p11.2 containing EGFR) forward, TTTGCCAAGGCACGAGTAA and reverse, CAAGGACCACCTCACAGTTATT; RAR-G8 (18p11.32 containing TYMS) forward, ATGTGGTGAACAGTGAGCTGTCCT and reverse, AATCA TGTACGTGAGCAGGGCGTA; RAR-L4 (9p21.3 containing CDKN2A/B) forward, TTTGGAGGACAAGCTGGTGCA ATG and reverse, TCCCAGATGAGGACAATGAGGCAA; Diploid control (16p12.2), forward, GACCTACCCACTTTCAACTCTG and reverse, CTCTCCAAGCGCAAGGTATT. Genomic qPCR were performed using ViiA 7 Real-Time PCR System and ViiA 7 RUO software (Applied Biosystems, Foster City, CA, USA) as described previously (14). In brief, qPCR mixture of 10 µl contains 10 ng of gDNA, Maxima^® SYBR-Green qPCR Master Mix (2X), ROX solution provided (Fermentas, Foster, CA, USA), and 5 pmol primers. Thermal cycling conditions consist of one cycle of 10 min at 95°C, followed by 40 cycles of 5 sec at 95°C, 10 sec at 61°C and 20 sec at 72°C. The dissociation value of the amplified PCR product was measured as the temperature inched up at a rate of 0.1°C/sec from 55°C to 95°C. All qPCR experiments were triplicated and relative quantification was performed by the 2^−ΔΔCT (15). Relative ratios were used to determine whether the status of a CNA is copy number gain (>2) or loss (<0.5).

To explore associations of chromosomal copy number changes and clinicopathologic phenotypes, phenotype variables were treated as categorical variables, that is, age, gender, primary site, histologic grade, T stage, cervical lymph node metastasis, presence or absence of lymphovascular invasion, extracapsular spread and recurrence. Differences in the distributions of RARs for each categorical variable were tested using two-sided Fisher's exact test. Disease-specific survival was determined using the Kaplan-Meier method. The Cox proportional hazards model with likelihood ratio statistics was used to identify variables significantly and independently related to survival. P<0.05 was considered to indicate a statistically significant difference, and all calculations were performed using SPSS version 16.0 (SPSS, Chicago, IL, USA).

The median age was 59.2 years (range, 49–77 years) and all patients were males. The sites of the original primary tumor were oral cavity (11.1%), oropharynx (44.4%), larynx (16.7%) and hypopharynx (27.8%). A majority of them (17/18, 94.4%) were stage T2 and T3, and sixteen patients (89.0%) were cervical lymph node positives. Thirteen patients (72.2%) were lymphovascular invasion positives and nine (50.0%) had extracapsular spread of a metastatic neck node. A total of nine patients (50.0%) had tumor recurrence after CCRT. There was no clinicopathological factors significantly associated with the treatment failure defined as tumor recurrence. Details of clinicopathological characteristics of the study subjects are summarized in Table I.

The overall frequency plot of the CNAs detected in the 18 HNSCCs (9 with and 9 without recurrence) showed that CNAs were not randomly distributed but rather clustered in certain chromosomal segments across the genome (Fig. 1A). The mean number of CNAs per each HNSCC genome was 41.3±18.6. Patients with recurrence had more CNAs than those without recurrence (mean frequency 57.3±24.9 vs. 25.3±7.2), but the difference did not reach statistical significance. As a whole, 8 chromosomal arms were recurrently (appeared in ≥2 cases) gained on: 2p (11%), 3q (28%), 5p (11%), 7p (11%), 8q (11%), 12p (11%), 18p (11%) and 22q (11%). Also, 8 chromosomal arms were recurrently deleted: 1p (11%), 3p (22%), 4p (11%), 5q (22%), 10p (17%), 13q (17%), 14q (11%) and 21q (17%). Of the recurrent chromosomal arm changes, copy number gains on 7p, 8q and 18p were observed only in the treatment failure group but these differences were not statistically significant.

In addition to the entire chromosomal arm changes, many of the CNAs were regional changes. Some of them were observed recurrently, suggesting their potential contribution to tumorigenesis and progression. To identify the commonly implicated CNAs, we defined RARs. In the present study, chromosomal segments as the union of overlapping CNAs in ≥20% of the 18 HNSCCs were defined as RARs (RAR-G for gains and RAR-L for losses, respectively). A total of 15 RARs (8 RAR-Gs and 7 RAR-Ls) were identified in the 18 HNSCC patients (Table II). Fig. 2 illustrates the RAR-G on 7p11.2 and RAR-L on 9p21.3 as examples. Of the 15 RARs, a RAR-G on 7p11.2 (RAR-G2: 4.22 Mb-sized, 55.6% in recurrence group vs. 0% in non-recurrence group, P=0.029), and a RAR-G on 18p11.32 (RAR-G8: 4.35 Mb-sized, 44.4% in recurrence group vs. 0% in non-recurrence group, P=0.029) were significantly more common in the HNSCC with recurrence group than in the HNSCC without recurrence group. The well-known oncogene EGFR (7p11.2) and the potential oncogene TYMS (18p11.32) are located in the recurrence-group dominant RAR-Gs. A RAR-L on 9p21.3 (0.42 Mb-sized RAR-L4), where CDKN2A/B are located, more commonly occurred in the HNSCC with recurrence group (44.4%) than in non-recurrence group (22.2%) but it was not statistically significant. To validate the three recurrence group dominant RARs identified by array-CGH, we performed genomic qPCR validations for those loci (7p11.2, 9p21.3 and 18p11.32). All the CNAs identified by array-CGH were consistently detected by the qPCR and as a whole 93.9% (46/49) of the qPCR results were consistent with the copy number status by array-CGH (data not shown).

Of the CNAs identified in the 18 HNSCCs, 25 loci were found to be high-level CNAs (defined as ≥1 for amplification or ≤−1 for HD on the log₂ scale), with 15 amplifications and 10 HDs (Table III). Five high-level CNAs were detected in >20% of the samples; amplifications on 7p11.2 and 11q13.3, and HDs on 1p31.1, 4q13.2 and 9p21.3, respectively (Table III). Well-known oncogenes (EGFR and CCND1) and tumor suppressor gene (CDKN2A/B) are located in the recurrent (>20%) high-level CNA regions.

Univariate survival analysis was performed to assess the implications of the RARs on patient survival. Univariate analysis was also performed to identify clinicopathologic features (age, gender, primary site, histologic grade, T stage, cervical lymph node metastasis, lymphovascular invasion, extracapsular spread and recurrence) for inclusion as covariates for Cox regression. Five-year disease-specific survival rates (DSSR) in our cohort were 66%. Among the RARs and clinicopathologic factors, cervical lymph node metastasis (P=0.008), recurrence (P=0.003), RAR-G2 (P<0.001), RAR-G8 (P<0.001), and RAR-L4 (P=0.037) were significantly associated with poor DSSR in the univariate analysis (Fig. 3). Multivariate analysis with the three significant RARs and two clinical variables identified on univariate analysis revealed that RAR-G2 [hazard ratio (HR)=40.68, 95% confidence interval (CI): 3.56–464.69, P=0.003] and cervical metastasis (HR=16.38, 95% CI: 1.20–223.11, P=0.036) were significantly associated with poor DSSR (Table IV).