Allelotype of squamous cell carcinoma of the head and neck: fractional allele loss correlates with survival.

Allelic imbalance or loss of heterozygosity (LOH) studies have been used extensively to identify regions on chromosomes that may contain putative tumour-suppressor genes. We have undertaken an extensive allelotype of 80 specimens of squamous cell carcinoma of the head and neck (SCCHN) using 145 polymorphic microsatellite markers on 39 chromosome arms. Allelic imbalances were found most frequently on chromosome arms 3p, 9p, 17p and 18q with over 45% LOH and imbalances on 1p, 1q, 2p, 5q, 6p, 6q, 8p, 8q, 9q, 11q, 13q, 17q and 19q were found in more than 20% of SCCHN. These LOH data were analysed against a range of clinicopathological parameters which included previously untreated and previously treated tumours; correlations were found between LOH on 9q and nodes at pathology (P = 0.02) and between histopathological grade and LOH on 12q (P = 0.02) and 13q (P = 0.01). In the group of previously untreated tumours, a correlation was found between site of tumour and LOH on 3p (P = 0.019), and 8p (P = 0.029), while TNM staging correlated with LOH on 3p (P = 0.019) and 17p (P = 0.016). Fractional allele loss (FAL) was calculated for 52 tumours with LOH data on nine or more chromosomal arms and found to have a median value of 0.22 (range 0.0-0.80). Correlations were found between FAL > median value and nodes at pathology (P = 0.01) and tumour grade (P = 0.06), demonstrating that advanced tumours with lymph node metastasis often had LOH at multiple sites. FAL > median value was found to correlate with a poor survival (P < 0.03) and, furthermore, FAL > median value correlated with poor survival in the previously untreated patients (P < 0.019). These results indicate that assessment of the accumulation of genetic damage, as provided by allelotype data, provides a useful molecular indicator of the tumour behaviour and clinical outcome.

Sinnukry Allelic imbalance or loss of heterozygosity (LOH) studies have been used extensively to identify regions on chromosomes that may contain putative tumour-suppressor genes. We have undertaken an extensive allelotype of 80 specimens of squamous cell carcinoma of the head and neck (SCCHN) using 145 polymorphic microsatellite markers on 39 chromosome arms. Allelic imbalances were found most frequently on chromosome arms 3p, 9p, 17p and 18q with over 45% LOH and imbalances on lp, lq, 2p, Sq, 6p, 6q, 8p.
8q. 9q. lIq, 13q, 17q and 19q were found in more than 20% of SCCHN. These LOH data were analysed against a range of cinicopathological parameters which included previously untreated and previously treated tumours; correlations were found between LOH on 9q and nodes at pathology (P = 0.02) and between histopathological grade and LOH on 12q (P = 0.02) and 13q (P = 0.01). In the group of previously untreated tumours, a correlation was found between site of tumour and LOH on 3p (P = 0.019), and 8p (P = 0.029), while TNM staging correlated with LOH on 3p (P = 0.019) and 17p (P = 0.016). Fractional allele loss (FAL) was calculated for 52 tumours with LOH data on nine or more chromosomal arms and found to have a median value of 0.22 (range 0.0-0.80). Correlations were found between FAL> median value and nodes at pathology (P = 0.01) and tumour grade (P = 0.06), demonstrating that advanced tumours with lymph node metastasis often had LOH at multiple sites. FAL >median value was found to correlate with a poor survival (P<0.03) and, furthermore, FAL>median value correlated with poor survival in the previously untreated patients (P<0.019). These results indicate that assessment of the accumulation of genetic damage. as provided by allelotype data. provides a useful molecular indicator of the tumour behaviour and clinical outcome.
To date only two global analyses of the whole genome have been undertaken with the view to determine the fractional allele loss (FAL) of specific tumours and thus provide information concerning the 'genetic burden' of the disease during its progression as measured by clinicopathological parameters and survival data. This type of analysis has been undertaken in colorectal (Vogelstein et al., 1989) and bladder cancers (Knowles et al., 1994), and provides an indication of interacting genetic mechanisms in the development of these diseases. In addition, the results of such detailed allelotypes may aid the interpretation of carcinogenesis and the development of molecular progression models for specific tumours.
We have undertaken a very comprehensive allelotype of SCCHN using 145 microsatellite markers in order to identify common regions of allelic imbalance and to analyse the interactions of these regions by calculating the fractional allele loss (FAL) in these tumours.

Specimens
Eighty SCCHN tumour specimens were collected at the Royal Liverpool University Hospital, Department of Otorhinolaryngology and at the Walton Hospital Liverpool, Maxillofacial Unit. Tumour samples obtained from surgical specimens were frozen in liquid nitrogen and stored at -70C. The clinicopathological data on the 52 SCCHN used in fractional allele loss analysis is given in Table I. This group of patients had LOH information on nine or greater (9-39) chromosome arms.

DNA extraction
All the tumour specimens used for LOH analysis were microdissected to yield at least 60% tumour cells before DNA preparation. Genomic DNA was extracted from tumour specimens using the Nucleon II DNA extraction kit (Scotlab) following the manufacturer's instructions. Genomic DNA samples were stored at 4°C.  non-denaturing polyacrylamide gel at 250 V and visualised by silver staining. LOH or allelic imbalance was scored by direct visual comparison of the allelic ratios of the normal and tumour specimens. Examples of heterozygous, homozygous and LOH in tumour/normal SCCHN specimens are given in Figure 1. Complete loss or reduced intensity of one allele in the tumour was considered as LOH. In the cases where there was only a reduced intensity of one allele this was considered to be due to contamination of tumour tissue by normal stroma. It has been previously noted by Ah-See et al. (1994) that the PCR techniques used by a number of authors in similar studies cannot readily distinguish between allelic duplication or low-level amplification leading to loss of heterozygosity. This caveat has to be taken into consideration when interpreting these results and thus LOH may not necessarily be indicative of the presence of a tumour-suppressor gene.

PCR and LOH analtvsis
Statistical analysis Quantitative data were analysed by x2 or Fisher's exact test where appropriate. Survival curves were drawn using the Kaplan-Meier (1958) product limit estimate. Differences between survival times were analysed by the log-rank method (Peto et al., 1976

Results
A total of 80 SCCHN tumours were investigated for LOH, using 145 microsatellite markers and loss on individual chromosome arms was calculated using the total data set (Table IL). A total of 1092 chromosomal arms were analysed, of which 956 (88%) were informative. The most frequent losses were found on chromosome arms 3p, 9p, 17p and 18q.
Loss of heterozygosity at specific loci The highest incidence of LOH was found on chromosome 9p (24/39) with a 62% loss and this allelic imbalance was especially concentrated between the D9S161 (9p21) and D9S156 (9p23-9p22) informative markers in this region (JK Field et al., in preparation), which agrees with the observations of van der Riet et al. (1994).
The second highest percentage of allelic imbalance (52%) was found on chromosome arm 3p from 61 informative tumours. Using 18 markers we found the greatest loss in the 3p24-p25 and 3pl3 regions and a very small incidence of LOH at the 3p2l site, including the D3SJ217 marker (3p2l) which had a LOH of only 10% JK Field et al., unpublished).
Chromosome 17p revealed an LOH of 50% with the highest loss at the CHRNBI locus (17pl2-pll.l). Furthermore, as previously reported by Adamson et al. (1994), LOH at this locus was found in 77% of the hypopharyngeal carcinomas studied.
Significant losses, 29% (13 45). were also found on chromosome arm 5q. eight markers were used, including D5S346 (5q21 -q22) in the APC MCC region which showed 35% LOH (9 26). Six patients in this group of tumours showed loss only in this region. which was bounded in each case by informative heterozygous markers centromeric and telomenrc of D5S346. thereby indicating that this region plays an important role in some SCCHN.
LOH has also been observed on chromosome arm lp in a range of tumours. with the lp3l.2-p21.3 region indicating that this may be a further target region in SCCHN. The cumulative loss of two markers in this region, DIS.159 (lp22.1-p21) and DIS167 (lp22-p21). was 24% (14,46).
Chromosome 11 contains an amplicon region at 11q13 which includes the int-2. cvclin D and EMS-I genes. We have found 23% LOH (9 39) on the 1llq arm and LOH at the int-2 locus was 17% (3 18).
In this data set, whole chromosome loss was seen only on chromosome 17 and in four tumours: 78. 192, 225 and 335 (11 % of cases). All of these chromosome arms showing LOH at greater than 20% (3p. 17p. 9p, 18q, Sq, 8p, lp and 1 Icq) have been previously shown to contain either known or putative tumour-suppressor genes. However, there are other arms in this study with greater than 20% LOH (lq, 2p, 6p, 6q, 8q. 13q. 17q and 19q) and these may also be target regions involved in the development of SCCHN.
LOH data analvsed against a range of clinicopathological parameters LOH data for each chromosomal arm were analysed against a range of clinicopathological parameters, including site of the tumour, histology, TNM staging, nodes at pathology and survival (Table I). These calculations were undertaken on the whole data set of 80 tumours (previously untreated and previously treated) and on the two subgroups separately. In the whole data set (80 SCCHN), correlations were found between nodes at pathology and LOH on 9q (P = 0.020) and between histopathological grading and LOH on 12q (P = 0.022) and on 13q (P = 0.012). In the group of previously untreated tumours, a correlation was found between site and LOH on 3p (P = 0.032) and 8p (P= 0.029), while TNM staging correlated with LOH on 3p (P = 0.019) and 17p (P = 0.016). Only one association was found in the group of previously treated patients, between nodes at pathology and LOH on lIp (P = 0.045) (Table III) Figure 3. The fractional allele loss (FAL) in a tumour is defined as the number of chromosomal arms on which allelic imbalance was observed divided by the number of chromosomal arms for which markers were informative in the patient's normal cells (Vogelstein et al., 1989).
The FAL values for this group of 52 SCCHN showed a median value of 0.22 and a mean of 0.25 (range 0.0-0.80).
FAL values were assessed against the clinicopathological data (tumour site, tumour grade, TNM staging, nodes at pathology) by dividing the tumours into those with FAL> median value and those with FAL <median value. A positive correlation was found between FAL and nodes at pathology (P = 0.01) and between FAL and tumour grade (P = 0.06) (Table IV). This demonstrates that advanced tumours with lymph nodal metastasis often had LOH at multiple sites. No correlation was found between FAL and the patient's history of smoking or drinking ( Table V).
The FAL data was also investigated for a possible association with clinical outcome using the log-rank analysis. It was found that a FAL >median value correlated with poor survival (P < 0.032), and furthermore that a FAL > median value also correlated with a poor survival in the previously untreated patients (P<0.019) when analysed separately. In order to analyse a homogeneous group of patients for FAL with clinical outcome, we calculated the log rank on the subset of 40 advanced tumours (TNM III and IV) and this also demonstrated a correlation between FAL and prognosis (P < 0.05).

D6cuson
In this detailed allelotype of SCCHN we have demonstrated a complex set of genetic alterations, a finding that has also been described in a range of human cancers (Vogelstein et al., 1989;Sato et al., 1990Sato et al., , 1991Fujimori et al., 1991;Monrta et al., 1991;Tsuchiya et al., 1992;Yamaguchi et al., 1992;Aoki et al., 1994;Fujino et al., 1994;Knowles et al., 1994). The highest LOH was found on the chromosome arms 3p, 9p, 17p and 18q which is in general agreement with previous studies on SCCHN (Ah-See et al., 1994;.  F_gwe 3 Individual allelotypes for 52 squamous cell carcinomas of the head and neck. These SCCHN tumours were investigated on nine or more informative chromosomal arms (range 9-39). The FAL (fractional allele loss) data has been given for each tumour specimen (range 0.0-0.80). *, LOH; 0, retention of heterozygosity; chromosome arms which were uninformative or not done are not shown. Each square represents the summation of the LOH results on a single chromosome arm using all of the informative markers, i.e. if there was allelic imbalance for any one of the markers tested for that specific chromosomal arm, then it is indicated as a filled square.
undertaken in this study in comparison with those in the two other allelotypes undertaken on SCCHN tumours. Both of the previous studies used about one-third of the number of markers used in this analysis: 58 markers  and 52 markers (Ah-See et al., 1994) respectively. Similar LOH values have been found at 3p, Sq, and 9q between Ah-See et al. (1994) and these data (± 15%), in the cases where the results may be compared. Also, similar LOH findings may be seen between Nawroz et al. (1994) and this data set for chromosome arms, Ip, Iq, 3p, 5q, 8p, 8q, 9p, 9q, lIp, 17p, 17q and 19q (± 15%). fjowever, a number of differences (> ± 20%) do exist between the previous reports and our results on 9p, llq, 13q and 18q. The percentage LOH for llq varies from 23%, 45% to 61% in the three studies (Ah-See et al., 1994;Nawroz et al., 1994; these data respectively), however the two previous studies only used two markers on this chromosome arm. The data on 18q from Nawroz et al. (1994) based on one marker give an LOH of 29%, whereas eight markers have been used in this study, demonstrating an LOH of 49%. This analysis demonstrates   Ah-See et al. (1994). target regions in SCCHN containing putative tumoursuppressor genes on 3p and 9p as well as a high LOH associated with the p53 gene. In addition, this analysis provides evidence for regions of minimal loss in SCCHN on lp, 8p, 17p and 18q (Adamson et al., 1994;Kiaris et al., 1994;Rowley et al., 1995;K Taylor et al., in preparation). The lp minimal area of loss has been located at lp3l.2-p2l.3, a region previously shown by karyotype analysis to contain cytogenetic abnormalities (Jin et al., 1990(Jin et al., , 1993Owens et al., 1992). A minimal area of loss has also been identified on 8p between 8pl2 and 8p21.2-p II in this series of tumours, a region considered to contain a candidate tumour-suppressor gene in colonic and hepatocellular carcinomas (Fujiwara et al., 1993;Cunningham et al., 1994). We have recently described a novel region on 17p distinct from TP53, at CHRNBI (17q12-pl 1.1), in SCCHN which has a particularly high loss in hypopharyngeal carcinomas (77%) (Adamson et al., 1994). Furthermore the detailed analysis of 18q has allowed us to identify a region at 18q21.1-q21.3 as a target region in SCCHN, which does not appear to be the DCC (deleted in colon cancer) gene as we found a very low LOH at the DCC locus. Thus it may be argued that there is a second tumour-suppressor gene in this region on 18q that is involved in SCCHN.
Two further chromosomal regions considered to contain tumour-suppressor genes in other neoplasms have not been shown to play an important role in SCCHN. Even though there is frequent LOH on 13q (Yoo et al., 1994) there does not appear to be inactivation of the retinoblastoma gene, and it has been argued by these authors that there may be another tumour-suppressor locus at 13ql4. Also, the APC/ MCC locus on 5q, which has been demonstrated to be involved in colorectal carcinomas (Kinzler et al., 1991) and has previously been shown to have a high LOH in SCCHN (Ah-See et al., 1994), may not in fact be the target locus, as mutations in the APC gene have rarely been found in oral cancers (Uzawa et al., 1994).
Analysis of LOH for each chromosomal arm was assessed against a range of clinicopathological parameters (Table III). In particular, a correlation was found between site and LOH on 3p and 8p, while TNM staging correlated with LOH on 3p and 17p in previously untreated tumours. Also, in the group of previously untreated and previously treated tumours a correlation was found between LOH on 9q and positive nodes at pathology, and histological grading correlated with LOH on 12q and 13q. In a detailed study undertaken by Lee et al. (1994), on chromosome 13 (using 13 markers), a correlation was found between LOH on 13q and lymph node metastasis. Moreover, these authors reported that they found similar LOH in a subset of the tumours in the adjacent non-malignant mucosa. However, no correlation between LOH at 13q and lymph node metastasis was observed in the study described here.
The phenomenon of microsatellite instability (MI) (Mao et al., 1994;Field et al., 1995) has been demonstrated in some of these SCCHN tumours, but no correlation was found between MI and LOH on any chromosome arm in this study. MI is therefore considered to be a separate pathogenic mechanism in the development of SCCHN.
The fractional allele loss (FAL) data were assessed for 52 tumours on which there was LOH information on nine or more chromosome arms. In this group we found a median FAL value of 0.22, mean of 0.25 (range 0.0-0.80). This demonstrates that alleles were lost on average from 25% of the chromosome arms in these tumours; a figure that is comparable with that obtained in non-small-cell carcinoma and colorectal carcinomas (0.2), bladder and breast carcinomas (0.11) and osteosarcarcinomas (0.32) (Sato et al., 1990(Sato et al., , 1991Morita et al., 1991;Tsuchiya et al., 1992;Knowles et al., 1994). The FAL values were compared with the clinicopathological data based on FAL <median value and FAL> median value. A correlation was found between FAL and positive nodes at pathology (P = 0.01), a clinical parameter considered to be the most useful prognostic indicator in head and neck cancer. No statistical correlation was found between FAL and site, TNM staging or histological differentiation. A history of smoking and drinking has been correlated with overexpression and mutations in the p53 gene (Field et al., 1991Field, 1992;Brennan et al., 1995), but no correlation has been found between these carcinogens and FAL in this analysis. In colorectal carcinomas, Vogelstein et al. (1989) found no correlation between FAL and Dukes' classification or tumour size, whereas in the allotype on bladder carcinomas, a correlation Hied an cancer -JK F-ed et a * 11 Q7 was found between FAL and tumour grade but not with the stage of the disease (Knowles et al., 1994). Thus, all three analyses demonstrate no correlation between FAL and tumour stage.
We have demonstrated that a FAL>median value correlated with a poor prognosis in all 52 patients analysed (P<0.032) and also in the subset of previously untreated patients (P<0.019) calculated by the log-rank method. Vogelstein et al. (1989) also showed a relationship between FAL and prognosis for colon cancers with a similar number and distribution of patients (P <0.01) using Fisher's exact test. Thus the argument originally proposed by Vogelstein et al. (1989) that recognition of accumulated genetic damage, as provided by the allelotype, provides a useful molecular indicator of the tumour behaviour is supported by the findings of this study.