Microenvironmental Genomic Alterations and Clinicopathological Behavior in Head and Neck Squamous Cell Carcinoma FREE

Despite its slowly declining incidence rate (~ 4% since 1980) and a modest improvement in 5-year survival (54.4% to 59.4% over the last 20 years), squamous cell carcinoma (SCC) of the head and neck continues to be a clinical challenge.^1- 2 With a worldwide prevalence of more than 1.6 million, it is estimated that in 2006, about 30 990 new cases will be diagnosed in the United States alone.^2- 3 Even with the utilization of all modern therapeutic options that include surgery, radiation therapy, and chemotherapeutic intervention, 50% of all patients will ultimately die of this disease, with more than 7400 projected for 2006 in the United States alone.^2- 3 Especially for patients diagnosed with advanced or relapsed disease, head and neck SCC is almost uniformly fatal.²

To improve patient management and identify novel compartments to target therapy, it is essential to further advance understanding of this disease at the etiologic level. It is an accepted concept that head and neck SCC arises from a successive accumulation of genetic alterations in the squamous epithelium of the mucosa that will allow a cell to obtain a growth advantage, escape apoptotic signaling, clonally expand, and ultimately invade and metastasize.^1,4- 6 Several groups have looked at those genetic alterations and identified mutations in key regulatory genes including TP53 and p16^INK4a as well as genetic instability in regions such as 3p, 9p, 11q, and 17p.^1,4,6- 8

Aggravating the clinical situation is the high rate of recurrent and multifocal disease in head and neck SCC.¹ This clinical and pathological observation was first addressed by Slaughter et al⁹ and the concept of field cancerization was coined. Over the years, it has been related to genetic observations and interpreted in different ways. The hypotheses include the following: that tumor or their progenitor cells migrate (both intraepithelial or luminal) to the secondary tumor sites or that tumors occur as independent events within genetically altered and expanding fields of preneoplastic epithelial cells.^10- 13 However, cancer is now known to be not only a disease of the transformed epithelium but also fundamentally influenced by and dependent on its microenvironment including the stroma in which it develops.^14- 15 The tumor stroma consists of fibroblasts, microvessels, and lymphatic cells and facilitates a physical and biochemical network that communicates closely with the epithelial cells. Genetic alterations in the stromal cells can lead to aberrant excretion of proteins and misinterpretation of incoming signals resulting in disruption of the physiologic interplay between epithelium and stroma.^14,16- 17 Our group and others have shown that the stromal fibroblasts of different neoplasias are rich in genetic alterations and can potentially define the tumor phenotype or potentially induce or sustain the transformation of the preneoplastic epithelium in sporadic and BRCA1/2-related breast cancers, prostate, and pancreatic cancers, and other solid tumors.^15,18- 22 To our knowledge, no study has looked at the tumor stroma on a comprehensive genomic level in order to address its role in head and neck SCC carcinogenesis.^23- 24 Using a whole genome approach, therefore, we sought to determine the extent of genomic alterations in the stroma of head and neck SCC and whether it correlated with presenting clinicopathologic features. Our purpose was to elucidate the stromal contribution to carcinogenesis and phenotypic differentiation of the squamous cell epithelium, with the potential to identify novel diagnostic and therapeutic options for new compartments.

A total of 122 formalin-fixed, paraffin-embedded, primary head and neck squamous cell carcinomas (SCC) from 122 patients have been analyzed for this study (Table 1). These samples were consecutively selected for squamous cell histology not being known to have received previous chemoradiotherapy in proximity to the resection and not within a clinical trial. Of these 122 samples, 63 (51.6%) were pharyngeal carcinoma and 55 (45.8%) were oral SCC (mainly lingual carcinomas). In addition, 1 laryngeal cancer and 2 carcinomas of unknown primary were analyzed. Among the pharyngeal SCC, 24 (38.1%) were located in the oropharynx and the remaining 39 (69.9%) in the hypopharynx. The distribution according to pTNM classification was as follows: 20.9% were T1; 40%, T2; 17.27%, T3; and 21.8%, T44, which is similar to that obtained for all patients at our academic institutions. The clinical staging followed the guidelines of the sixth edition of the American Joint Committee of Cancer Cancer Staging Manual²⁵ (Table 1). The study, which used anonymized unlinked samples, was approved, under exempt status, by the participating institutional review boards for human subjects' protection. Examination of the cancer registry information revealed that the participants were smokers.

Laser capture microdissection was performed using the Arcturus PixCell II microscope (Arcturus Engineering Inc, Mountain View, Calif) in order to isolate the 2 compartments of the neoplastic tissue (epithelium and stroma) separately (Figure 1).²⁶ We specifically captured stromal fibroblasts adjacent to malignant epithelium (ie, the tumor stroma) under direct microscopic observation. These stromal fibroblasts resided either in between aggregations of epithelial tumor cells or no more than 0.5 cm distant from a tumor nodule. Corresponding normal DNA for each case was procured from normal tissue (preferentially tumor-negative lymph node), obtained from a different tissue block containing only normal tissue.

Genomic DNA was extracted as previously described.^19,26 Polymerase chain reaction was performed using DNA from each compartment (normal control, tumor epithelium, and tumor stroma) of each sample and one of 72 multiplex primer panels, which comprises 366 fluorescent-labeled microsatellite markers. Genomic location is based on the MapPairs genome-wide Human Markers set (version 10) (Invitrogen, Carlsbad, Calif) developed at the Marshfield Institute. This whole genome panel has an average 16.2 markers per chromosome (ranging from 7 to 29 markers per chromosome) or approximately a 9-cM-intermarker distance. Genotyping was performed with the ABI 377xl or 3700 semiautomated sequencer (Applied Biosystems, Perkin-Elmer Corp, Norwalk, Conn). The results were analyzed by automated fluorescence detection using the GeneScan collection and analysis software (GeneScan, Applied Biosystems). Scoring of loss of heterozygosity or allelic imbalance (LOH/AI) was performed by manual inspection of the GeneScan output (Figure 1). A ratio of peak heights of alleles between germline and somatic DNA of at least 1.5 was used to define LOH/AI as previously described.^19,27- 29 The methodological veracity of LOH/AI using multiplex-polymerase chain reaction on archived tissue was extensively validated.^19,26

In total, 366 microsatellite markers were analyzed in both epithelium and stroma samples from the 122 patients. First, we sought to determine regional LOH hot spots defined as a significantly higher frequency of LOH at a marker or markers compared with other markers along the same chromosome. Toward those ends, for each marker, the statistical significance of overall (across all samples) LOH frequency compared with the chromosome average was analyzed using the exact test of binomial proportions (R base package binom.test; http://www.r-project.org). Second, the association of LOH or retained heterozygosity in epithelium and stroma samples with presenting clinicopathologic parameters—such as location, pT, pN, grade, clinical stage, age, and sex—were analyzed using a binomial model with nested structures.^30- 31 Of note, the age was dichotomized into 2 classes using age 40 years as the cutoff. For associations with clinical stage, pT or pN, the statistical significance was tested using the test of trend for multiple proportions. Multiple testing adjustment has been applied by using false-positive report probability (FPRP)³² with a prior probability of 0.05 and 0.01, denotated as FPRP_0.05 and FPRP_0.01, respectively. The FPRP indicates the probability that a statistically significant finding is a false-positive by considering 3 factors: the P value magnitude, the statistical power, and the prior probability of true associations. Only those with P values <.05 and estimated FPRP values less than 50% (or P<.50), indicating a small probability of being a false-positive, are reported as statistically significant findings. For example, a significant value with a prior probability of 0.01 and an FPRP value less than 50% is denoted FPRP_0.01<0.5. The hierarchical clustering and pattern visualization were performed using PfCluster.³³ The R package (http://www.r-project.org) was used for the data mining and statistical analysis. Data analysis, interpretation, statistical analyses, and drafting of the original manuscript occurred between November 2005 and June 2006.

Our study included predominantly (97.5%) SCC of the oral cavity and pharynx of patients with a history of smoking. Overall, 244 samples (122 epithelium and 122 stroma samples of 122 patients) were analyzed for genomic instability using 366 microsatellite markers. In total, 43 591 informative (nonhomozygous) data points were obtained. Of these, 28 320 markers (65%) showed LOH/AI and 15 271 markers (35%) retained heterozygosity. There was no difference in the number of informative markers between the stroma and epithelium (48.4% vs 48.9%). For the epithelium, the frequency of LOH/AI per sample was 69.0% (range, 33.3%-93.7%) compared with an LOH/AI frequency of 64.4% (range, 25.8%-90.3%) observed in the stroma (P = .10). To confirm that the high frequency of LOH/AI observed in the stroma is not a result of epithelial contamination, we took a multilevel approach to provide conclusive evidence against an erroneous or artifactual finding (Figure 1 and Figure 2). First, for several cases, we noted markers with opposing LOH/AI calls in each compartment of a given tumor (ie, LOH/AI observed in the epithelium but not stroma and vice versa). Second, in some cases with concordant LOH/AI calls, we found that different alleles are lost in a compartment-specific manner. Third, we identified somatic mutations in some of these cases that were confined to either the epithelium or stroma but not in both (data not shown). Because all analyses have been performed from the same pool of extracted DNA, such observations exclude to a very high probability the possibility of tissue admixture or intercompartmental contamination.

As a control, we examined our samples for compartment-specific LOH/AI in the markers residing in the previously reported regions of LOH/AI on 3p, 9p, and 17p with LOH frequencies greater than 50% in whole or epithelium-only head and neck SCC. In our study, we observed strong hot spots of LOH/AI in the microdissected tumor epithelium for 2 distinct regions on chromosome 3. The first chromosome 3 hot spot maps to subband p25.2 to 25.3 (Table 2 and Table 3). The second 3p hot spot maps to 3p14.2 (D3S1766) and is even more significantly associated with stroma (Table 2). The stroma also had this same hot spot mapping to subband p25.2, and perhaps a broader region defined by markers D3S2432 and D3S2409 (Table 4). Among all loci, chromosome 9 harbored the second highest frequency of LOH/AI (95%) for the epithelium at 9p21.3 to 9p23 (84%-95%, data not shown). In our study, besides a hot spot at 17p13.1 to 17p13.3 (TP53 locus), we noticed a hot spot of LOH/AI at 17p13.3 (D17S1308), telomeric of the TP53 locus (Table 4). Of the 27 loci with the most significant LOH/AI in the epithelial component, 11 have been reported by other groups to harbor regional losses by comparative genomic hybridization.^34- 36 Thus, of the previously reported regions of LOH/AI, all were identified in our compartment-specific study and served as a positive control.

Hot spots are defined as markers that show a significantly higher frequency of LOH/AI compared with all other loci on the same chromosome. In total, we identified 70 hot spots (at P<.05 and FPRP_0.05 <0.5), 17 occurring only in the epithelium, 43 only in the stroma, and 10 in both epithelium and stroma (Table 2, Table 3, and Table 4). The most significant hot spot (P<.001; FPRP_0.05 <0.5) of LOH/AI observed exclusively in the epithelium was defined by D16S422 mapping to16q23.3 (Table 3). Eight additional highly significant hot spots of genomic instability (P<.01) were identified at 1q31.1 (D1S518), 1q43 (D1S1594), 3q13.3 (D3S2460), 15q25.3 (D15S655), 16p13.3 (D16S2616), 20p12.2 (D20S851), 21q22.2 (D21S2055), and 3p25.2 (D3S4545, see above, Table 3). Among the 43 hot spots of LOH/AI that were restricted to the stroma, 30 loci were highly significant (P<.01, FPRP_0.05 <0.5, Table 4). Highest ranked among these were markers D17S1308 (17p13.3) and D14S1434 (14q32.13) followed by D10S1230 (10q26.1), D2S1400 (2p25.1), and D2S1790 (2p11.2, Table 4). Although our data show that hot spots of LOH/AI are more diverse in the tumor stroma than in the epithelium (43 vs 17, P = .005) of head and neck SCC, the frequency of highly significant loci among all hot spots within each compartment was similar (9 of 17; 30 of 43; P = .56).

Besides the 2 hot spots of LOH/AI at markers D3S1766 and D3S2403 mentioned above (″Validating previous loci AI associcated with HNSCC oncogenesis”), genomic alterations at 14q31.1 (D14S606) and 12q24.32 (D12S2078) was found most frequently in both epithelium (P = .003 and P=.<001) and stroma (P < .001 and P =.01; Table 2). Furthermore, an additional 8 loci were identified as noncompartment-specific hot spots of LOH/AI (ie, occurring equally in both epithelium and stroma) with a cutoff at P<.05 and FPRP_0.05 <0.5 (Table 2). We also identified a locus that retained heterozygosity (ie, did not show genomic instability) at a frequency higher than what we would expect by chance: D14S599, representing chromosome subband 14q13.1, showed LOH/AI only in 16 of 58 informative samples (27.6%, P = .001) in the epithelium and 16 of 57 (28.1%, P<.001) in the stromal compartment.

Our data mining process allowed us to identify loci of LOH/AI that extended over 2 or more adjacent hot-spot markers, indicating larger regions of genomic alterations on chromosome arms 3p, 12q, and 14q. For instance, 12q24.32 (D12S2078) harbored a hot spot of LOH/AI for the epithelium (81.2%, P = .001) and stroma (75.0%, P = .01). A second hot-spot region on chromosome 12 was located at 12q13.13 (D12S297) affecting only stroma (80.3%, P<.001) and extends further centromeric, to 12q21.33 (D12S1294, 74.3%; P = .01) and to 12q24.23 (D12S395, 77.9%; P = .002). In addition, LOH/AI at 11q12.1 (D11S4459) was identified in 84.6% of the stroma (P = .002) samples.

We then performed data mining on our whole-genome LOH/AI scan in order to identify compartment-specific loci that showed a correlation between LOH/AI frequency and clinicopathological parameters. Interestingly, stromal-specific LOH/AI-clinicopathological correlations were more frequently observed than for the epithelium. First, we sought to identify LOH/AI at loci that were positively associated with aggressiveness of disease as reflected by clinical stage, grade, and pT and pN status (Figure 2, Table 5). We found that LOH/AI at D6S305 (6q26) in the epithelium occurred significantly more frequently in clinical stage III and IV head and neck SCC (88.6%) than in stage I and II tumors (58.3%, P = .01, Table 5). In addition, we observed a linear increase of LOH/AI frequencies from stage I (50%) and stage II (63%) to stage III (80%) and to stage IV (95%) tumors (P = .01) for the locus 6q26, which contains the common fragile site FRA6E. No such association with clinical stage was identified for LOH/AI in the stroma. Interestingly, LOH/AI at D4S2417 (4q34.3) in the stroma showed a positive correlation with increasing pT stage (P<.001) (Figure 2, Table 5). Furthermore, markers mapping to D3S3630, and D19S599 (3p26.3, P = .01; 19q13.31, P = .02) showed an increasing frequency of LOH/AI in stroma correlating with the degree of lymph node involvement (Table 5). For the epithelium-specific LOH/AI, we identified that genomic alterations at 18p11.22 (D18S843) was positively correlated with regional lymph node metastasis (pN) with 33% LOH/AI in N0 tumors compared with 79.4% in lymph node positive disease (P<.001). Importantly, no positive correlation between LOH/AI in the epithelium and pT stage was observed.

The mucosa of the upper aerodigestive tract is exposed to an array of carcinogens that have been attributed to cause genetic and epigenetic changes in the squamous cell lining and ultimately lead to head and neck SCC genesis. It is evident that these carcinogens not only affect these epithelial cells but also the mesenchymal fibroblasts, the latter representing the largest component of the stroma. These results show that the stromal cells in head and neck SCC are subjected to selection for locus-specific LOH/AI events. The high frequency of LOH/AI, especially in the tumor stroma, might appear distracting at first. However, it does reflect the biological background behind head and neck SCC since in our study only patients with a history of smoking have been analyzed. In addition, technical aspects have to be considered as well. First, it is important to note our operational definition of a hot spot, which is defined as a locus having a significantly high frequency of LOH/AI compared with all other loci along the same chromosome. Thus, it is possible that other studies using a small set of markers might therefore find an apparently high frequency of LOH/AI in 1 marker and labeled this locus significant; however, other loci along the same chromosome, which may not have been examined, might actually have LOH/AI to a similar or even elevated degree than the selected marker. In addition, studies using array comparative genomic hybridization, although having the advantage of differentiating between allelic gain and loss, usually detect losses or gains of larger genomic regions, spanning several bacterial artificial chromosome clones. In contrast, microsatellite marker LOH analysis is able to accurately identify submicroscopic deletions or even single base-pair alterations, if those affect the microsatellite marker priming sites. However, it is important to recognize that in this study, we could recapitulate the common observation of early events (ie, those with a high frequency of LOH/AI) attributed to head and neck SCC oncogenesis that are lost at 3p, 9p, and 17p in the tumor epithelium (Table 2). This acts as a control that our data mining approach can correctly identify compartment-specific hot spots of genomic instability in microdissected epithelium and, more importantly, the stroma of head and neck SCC lesions.

Interestingly, we observed more LOH/AI hot spots in the stroma than epithelium. Even when the same LOH/AI hot spot markers were found in both the epithelium and stroma, overall, the frequencies of LOH/AI were much higher in the corresponding stroma (Table 2). This may indicate that only a very limited set of key genetic alterations within the epithelium are required to initiate head and neck SCC genesis and other alterations are downstream events or even bystander events. This has been addressed previously by Gotte et al³⁷ who reported on the intratumoral heterogeneity of head and neck SCC. In contrast, the multiplicity of stroma-specific hot spots, likely occurring along all steps of carcinogenesis, suggest that these play the fundamental role in influencing the biological diversity, and hence, clinical behavior, of the disease (Figure 2). Whether the accumulation of stromal alterations occurs concordant with the neoplastic transformation of the epithelium or in fact precedes the malignant transformation of the squamous epithelium is unknown. In breast cancers from individuals with germline BRCA1/2 mutations, the inherited dysfunction in these repair genes seems to dictate that stromal genomic alterations occur before or at least simultaneously with epithelial transformation.¹⁹

Besides several genes involved in oncogenesis or cell-cell communication mapping to these hot spots, we also find micro-RNAs that might become deregulated through allelic imbalance. It is an emerging concept that the deregulation of micro-RNAs participate not only in development but also cancer. For instance hsa-miR-181 (19p13.12) was identified as a stroma-specific hot spot and has been implicated in cellular differentiation through regulation of homeobox genes.³⁸ Given that hot spot and LOH/AI frequencies highest in stroma, we may even postulate that if field cancerization precedes invasive head and neck SCC, then the mesenchymal cells undergo genetic alterations first.

Evidently, the positively selected stromal cells acquire additional hits, presenting as multiple hot spots of LOH/AI, that can lead to aberrant excretion of proteins and misinterpretation of incoming signals resulting in disruption of the physiologic interplay between epithelium and stroma and provides the necessary microenvironment to sustain and promote tumor progression.^14- 15,39 Seemingly paradoxically, however, 1 locus mapping to 14q13.1 retained heterozygosity at a significant frequency in both epithelium and stroma, suggesting that genes mapping to those loci might be necessary for maintenance of cell integrity or key regulatory genes might be frequently affected by somatic sequence variants that will cause dominant negative acting transcripts. Interestingly, among the genes within this region is PHD3 (prolyl hydroxylase domains 3; equivalent to EGLN3) involved in oxygen sensing and regulation of especially HIF-2α.⁴⁰

We found 5 specific loci of LOH/AI associated with clinicopathologic features at presentation (Figure 2). Among all the hot spot loci associated with presenting clinicopathologic features, these specific 5 were identified with sequentially increasing LOH/AI frequencies significantly associated with increasing pT, pN and/or clinical stage and with a low likelihood of representing false-positive associations. Three specific loci occurred in the stroma, associated with tumoral attributes of aggressive disease and invasion, namely, size (pT status; 1 locus at 4q34.3) and regional lymph node status (pN, 2 loci at 3p26.3 and 19q13.31). One gene in the 4q34.3 region is NEIL3, which encodes a class of glycolases that initiate the first step in base excision repair. One therefore could postulate that loss of NEIL3 could be one of the first events leading to a cascade of genomic alterations in the stroma.⁴¹

It also appears that the stroma plays an important role in metastases for which 2 of the 3 hot-spot loci, at 3p26.3 and to 19q13.31, in the stroma are correlated with increasing pN status (Figure 2). There are likely several genes mapping to these regions. One relevant gene mapping to 3p26.3 is FANCD2, which encodes 1 of the enzymes in the Fanconi anemia pathway pivotal to DNA repair and which interacts with BRCA1 and BRCA2.^42- 43 The Fanconi anemia pathway is again targeted by the loss of a gene encoding FAZF on 19q13.3, the other stromal locus whose loss is associated with pN status. This zinc-finger protein binds to another Fanconi anemia pathway member FANCC in a region that is deleted in Fanconi anemia patients with a severe disease phenotype.^44- 45 This 19q locus is proximal to another DNA repair enzyme gene, ERCC2. The gene ERCC2 or XPD encodes an excision repair enzyme that has been identified to have an increased risk of cancer when mutated due to abrogation of its transcriptional activation of FBP, a regulator of MYC.⁴⁵ Our observations herein, therefore, suggest that these genes in concert may play a role in head and neck SCC and, in particular, relevant to regional metastases. It is tantalizing that the most promising candidate genes in the regions of loss associated with clinicopathologic features belong to the various repair pathways. The loss of FANCD2, FAZF, and ERCC2 together could additively and more severely result in additive loss of repair capabilities that result in a cascade of downstream genomic alterations, leading to genomic instability resulting in invasion and metastasis. This postulate is supported by our observations herein in the multiplicity of genomic alterations in head and neck SCC stroma (Table 2, Table 3, and Table 4). To further support this hypothesis, a quantitative trait locus for prostate cancer aggressiveness has been identified in this region by 2 groups,^46- 47 suggestive that a gene(s) is harbored in this location that may also be important in head and neck SCC aggressiveness, as our association of this locus to pN suggests. Equally significant is the locus reflected by marker D18S843 (18p11.2) in the epithelium. Allelic loss for this region has previously been implicated in other solid tumors and even associated with relapse in breast cancer.^48- 49 From the genes mapping to this loci, it is unclear what the likely candidate will be; of note is APCDD1 with suggested oncogenic properties in colorectal cancer. Importantly, this gene is expressed during development to regulate epithelial-mesenchymal interaction.⁵⁰ Only a single specific locus (D6S305) was independently identified as a hot spot of LOH/AI associated with clinical stage. Deletions of 6q26 (D6S305) have been reported to have a role in carcinogenesis. This region harbors the common fragile site FRA6E that spans 8 genes (IGF2R, SLC22A1, SLC22A2, SLC22A3, PLG, LPA, MAP3K4, and PARK2), which have been implicated in the development of solid cancer.⁵¹

Our observations suggest that the apparently nonmalignant stroma of head and neck SCC is rich in genomic alterations. The strong association of a limited number of specific loci with sequentially higher frequencies of LOH/AI in the stroma with clinical aggressiveness indicates that mesenchyme is affected by carcinogens to the same extent as the squamous cell epithelium, and even more importantly, contributes in a fundamental way to the clinical phenotype of head and neck SCC. Our data suggest that this genetically altered mesenchymal field might provide the soil that facilitates the head and neck SCC invasion and metastases. We hope that our genomic observations, which point to genomic regions that may harbor many genes, will guide future in-depth functional and mechanistic studies. Nevertheless, our current observations can be used to identify new biomarkers for prediction of clinical outcome and potentially novel compartments for targeted therapy and prevention.