Prognostic significance of cyclooxygenase-2 in laryngeal squamous cell carcinoma

Authors


Abstract

Epidermal growth factor receptor (EGFR) overexpression is an unfavorable prognostic marker in laryngeal squamous cell carcinoma (SCC). EGFR stimulates cyclooxygenase-2 (COX-2) expression in normal human keratinocytes and squamous carcinoma cells. Based on these observations a prognostic role of COX-2 expression in laryngeal SCC can be hypothesized. Consequently, COX-2 expression was studied in laryngeal SCC (median follow-up = 47 months; range: 2–87 months) by quantitative immunohistochemistry (n = 61) and EGFR by binding assay (n = 51). Well-differentiated regions of laryngeal SCC revealed strong COX-2 immunostaining, whereas histologically normal areas neighboring tumor as well as poorly-differentiated tumors were negative. Immunohistochemical results were confirmed by Western blot analyses. Cox's regression analysis showed that the combination of low levels of COX-2 integrated density and high levels of EGFR covariates provided strong prediction, at 5-year follow-up, of both poor overall survival (χ2 = 12.905; p = 0.0016) and relapse-free survival (χ2 = 9.209; p = 0.01). In vitro studies on CO-K3 cell line, obtained from an EGFR positive, COX-2 negative poorly-differentiated laryngeal SCC, revealed that EGF stimulation failed to induce COX-2 expression and PGE2 production suggesting a change in EGFR signaling pathway. These findings indicate that COX-2 is overexpressed in less aggressive, low grade laryngeal SCC, whereas its expression is lost when tumors progress to a more malignant phenotype. © 2001 Wiley-Liss, Inc.

Cyclooxygenase (COX) catalyzes the synthesis of prostaglandins (PG) from arachidonic acid. Two enzyme isoforms have been identified: COX-1 which is constitutively expressed as an housekeeping gene in most of the cells and COX-2 which is expressed as an early-response gene activated by many stimuli such as inflammatory cytokines as well as growth factors, and oncogenes.1 Several evidences indicate that COX-2 is involved in tumorigenesis being up-regulated in tumor cells.2–7 In particular, it has been recently reported that COX-2 is up-regulated in head and neck tumors at both mRNA and protein levels.8

The vast majority of laryngeal cancers are of the squamous cell type with a variable degree of differentiation ranging between well-differentiated and poorly-differentiated tumors. In laryngeal squamous cell carcinoma (SCC), epidermal growth factor receptor (EGFR) expression and amplification is a well recognized unfavorable prognostic marker.9, 10 Recently, it has been reported that activation of EGFR stimulates COX-2 expression in normal human keratinocytes and squamous carcinoma cells.11 On the basis of these observations a possible prognostic role of COX-2 expression in laryngeal SCC can be hypothesized.

In the present study the expression of COX-2 was studied by immunohistochemistry in laryngeal SCC in relationship with EGFR expression and clinical outcome.

MATERIAL AND METHODS

Patients

Our study included 61 untreated consecutive primary laryngeal SCC patients admitted to the Department of Otolaryngology of the Catholic University, Rome. Histological grading and TNM classification were performed on conventional paraffin sections according to the recommendations of the International Union Against Cancer.12 Accordingly, tumors were graded as well- (G1), moderately- (G2) or poorly- (G3-G4) differentiated. At our institution, all primary laryngeal cancer patients received standard therapeutic management including surgical treatment (curative surgery) of the primary tumor (T) related to the lesion extension; therapeutic neck dissection when there was lymph node involvement at clinical presentation (N+) according to the “wait and see” policy under strict follow-up conditions;13 post-surgical radiotherapy for locally advanced tumors (T4) and neck lymph node metastasis with extranodal spread according to the following treatment protocol: 180 cGy a day for 5 days in a week for a total of 70 Gy. All the patients in our study have been treated according to this standard procedure. Thirty-eight patients underwent total laryngectomy and 23 partial laryngectomy (supraglottic laryngectomy, hemilaryngectomy and cordectomy). At surgery, 8 patients with clinically positive neck nodes underwent a therapeutic neck dissection, that confirmed lymph-node positivity. The median follow-up period was 47 months (range: 2–87 months).

Cell lines

The human laryngeal carcinoma cell line Hep2 was originally purchased from the ATCC (Rockville, MD). CO-K3 is a laryngeal tumor cell line obtained in our laboratory from a 53-year-old Caucasian male with a histologically confirmed high grade (G4) primary laryngeal squamous cell carcinoma. The epithelial origin of CO-K3 cells was confirmed by immunoperoxidase staining. They stained strongly positive for cytokeratins and appeared negative for the intermediate filament vimentin. CO-K3 cells were used in passages 20–30. Cell lines were cultured as previously reported.14

Epidermal growth factor receptor assay

Representative frozen tumor tissues from 51 laryngeal SCC patients, histologically verified, were stored at −80°C until processed. The EGFR assay on tumor specimens was carried out as cxpreviously reported.10 EGFR on cell lines was measured by a whole cell assay. Cells were plated into Multiwell TM (Falcon 3047, Oxnard, CA) at a density of 100,000 cells/ml in RPMI-1640 medium (Gibco BRL, Grand Island, NY) supplemented with 10% heat-inactivated FCS. After plating, the cells were left to equilibrate in serum-free RPMI at 37°C, for 1 hr, in a CO2 incubator to allow the disengagement of bound EGF. Then, the medium was changed with fresh medium without serum and containing increasing concentrations (from 0.07–1.8 nM) of [125I]-epidermal growth factor (EGF) (NEN Dupont, DE Nemours; specific activity: 174 μCi/μg) with or without a 300-fold-molar excess of unlabeled EGF, to evaluate non-specific and total binding. After 1 hr incubation at 37°C, cells were rapidly washed with ice-cold medium and then incubated in 1 M NaOH for 30 min at 50°C. Radioactivity was measured in a gamma counter. Cells plated in parallel wells were trypsinized and counted at the end of washing procedure. Results were expressed as the number of binding sites per cell.

COX activity

Confluent cells were washed twice with RPMI 1640 medium (Eurobio, Milano, Italy) without serum, containing 1 mg/ml BSA (Sigma Chemicals, St. Louis, MO) and incubated for the indicated time intervals in the same medium with or without EGF at either 50 nM or 100 nM. To evaluate COX activity, cells were washed with Hanks' buffer, containing 1 mg/ml BSA, and incubated for 40 min in the same buffer containing 10 μM arachidonic acid (Cayman Chemical, Ann Arbor, MI). In experiments performed with COX inhibitors, cells were incubated for 40 min with various concentrations of indomethacin (Sigma), valeril salicilate (Cayman Chemical), or NS-398 (Cayman Chemical) and then washed with Hanks' buffer/BSA and incubated for 40 min in the same buffer with 10 μM arachidonic acid, as already described. Supernatants were collected for measurements of PGs. Cells were then harvested by trypsinization and counted. The experiments were always performed in duplicate. PGE2 was determined in the supernatants of the cell cultures using previously validated radioimmunoassay15

Western blot analysis

Adherent cells, washed twice with PBS and laryngeal SCC specimens were lysed in ice-cold lysis buffer (20 mM TRIS-HCl pH 7.5, 5 mM EDTA, 5 mM EGTA, 10 μg/ml leupeptin, 20 μg/ml aprotinin and 1 mM phenylmethylsulfonyl fluoride) for 15 min. The lysates were centrifuged at 4°C for 10 min at 10,000g. The protein content was determined using a Bio-Rad protein assay (Bio-Rad, Munich, Germany) using BSA as standard. Lysates (50–100 μg of protein) were mixed with Laemmli reagent under reducing conditions. SDS-PAGE was performed according to standard techniques, using 10% bis-acrylamide for the separation gel and transfer of proteins.16, 17 Nitrocellulose membranes were saturated for 2 hr at room temperature in 5% fat-free dry milk-Tris buffer saline (50 mM Tris-HCl, pH 7.5, 250 mM NaCl, and 0.1% Tween 20) (TBTT). Membranes were further incubated at RT for 1 hr with specific monoclonal antibodies for COX-1 (1:1,000), COX-2 (1:2,000)16, 17 or β-actin 1:5,000 (Sigma). Blots were washed 3 times with TBTT buffer and then incubated with anti-mouse horseradish peroxidase-linked secondary antibody in TBTT and 5% fat-free dry milk. Chemiluminescent substrates were used to reveal positive bands visualized after exposure for 1–2 min to Hyperfilm ECL (Amersham Italia Srl, Milano, Italy). Protein bands were quantified by laser densitometer (Amersham Italia Srl)

Immunohistochemistry

Tumor tissues obtained at surgery were fixed in formalin and paraffin-embedded according to standard procedures. Four-μm-thick tissue sections of representative blocks from each case, were deparaffinized in xylene, rehydrated, treated with 0.3% H2O2 in methanol for 10 min to block endogenous peroxidase activity, and subjected to heat-induced epitope retrieval in microwave oven using the Dako ChemMate detection kit (DAKO, Glostrup, Denmark) according to the manifacturer's instructions. Slides, from all cases studied, were then simultaneously processed for immunohistochemistry on the TechMate Horizon automated staining system (DAKO) using the Vectastain ABC peroxidase kit (Vector Laboratories, Burlingame, CA). Endogenous biotin was saturated by a biotin blocking kit (Vector Laboratories). Sections were incubated with normal rabbit serum for 15 min, then with rabbit antiserum against COX-217 diluted 1:500 for 1 hr. Negative controls were performed using nonimmunized rabbit serum, omitting the primary antibodies. Cytospin preparations from a batch of Hep2 cells, treated for 48 hr with 50 nM EGF, were used as reference positive control.

Quantification of immunohistochemical staining

The intensity of immunohistochemical staining was evaluated using image analysis based on Photoshop (version 5.0; Adobe Systems, San Jose, CA) together with “The image processing toolkit” (version 3.0, 1998, CRC Press, Boca Raton, FL) according to the method previously reported18 with some modifications. Briefly, the technical setup included a Zeiss Axioskop microscope (Zeiss; Jena, Germany) equipped with a Nikon Coolpix 950 digital camera (Nikon Corporation; Tokyo, Japan). Three ×20 fields were chosen from each section so as to best reflect the overall immunostaining of the tumor contained on the entire slide. After acquisition with digital camera, the files were saved in tagged-image file format which allows LZW compression without discarding any data. The files were opened in Photoshop using a Macintosh 400 MHz G3 workstation (Apple; Cupertino, CA). The immunostained regions of interest were automatically selected and highlighted using the “Magic Wand” tool and an appropriate color tolerance level. The mean density value and the area (in pixels) of the immunostained regions were measured by the “Brightness filter” tool and a built-in calibration curve constructed from the “Brightness filter” readings and the known optical density values of calibrated wedges digitized with the same camera. The rest of the tumor tissue was subsequently selected using the “Inverse” tool and the relative area in pixels calculated with the “Brightness filter” tool and added to the immunostained area to obtain the total measured area. Then the integrated density of the immunostaining was calculated as the product of the mean density value of the immunoreactive regions by the percentage of the immunostained tumor tissue. The computerized image analysis of all tissue sections were done by three different pathologists without prior knowledge of the clinical and pathological parameters.

Statistical analysis

For some statistical evaluation, different cut-off values of COX-2 integrated density (mean, median, 25% and 75% percentiles) were tested in the survival analyses. Although all cut-off values tested behaved as significant discriminators, the best prognostic cut-off point value was the the mean value of COX-2 integrated densities. Consequently, we decided to chose this value (≥435) to categorize tumors as positive or negative. Fisher's exact or χ2 tests for proportions were used to analyze the distribution of COX-2 status according to various clinico-pathological parameters. Survival data were available for all 61 patients. The Cox-Mantel method was used to evaluate the prognostic role of COX-2 integrated density and EGFR as continous variables.19 All medians and life tables were computed using the product-limit estimate by Kaplan and Meier,20 and the curves were examined by means of the log-rank test.21 Univariate and multivariate analysis was performed by Cox's proportional hazards model.19 Relapse-free survival was calculated from the date of first surgery to that of clinical or pathological recurrence. Overall survival was calculated from the date of first surgery to that of death. All p-values were two-sided.

RESULTS

As shown in Figure 1, in foci of neoplastic transformation, many COX-2 expressing cells could be observed whereas cells of histologically normal areas neighbouring tumor were not immunostained for COX-2 (Fig. 1a,b). In well-differentiated laryngeal SCC most cells displayed cytoplasmic COX-2 immunostaining with a perinuclear reinforcement. Moreover, keratinized areas amongst tumor cells were not immunostained (Fig. 1c,d). In high grade (G4), poorly differentiated tumors, cancer cells did not express COX-2 whereas some stromal cells displayed COX-2 immunoreactivity (Fig. 1e,f). Western blot analysis, with anti-COX-2 monoclonal antibody, showed immunoreactive bands of the expected COX-2 molecular weight (approximately 70 kDa), in the tissue lysates obtained from the immunohistochemically COX-2 positive tumors (Fig. 1g).

Figure 1.

Cyclooxygenase-2 (COX-2) immunohistochemistry (a,c,e), H&E staining (b,d,f) and Western blot analysis of COX-2 expression (g) in laryngeal squamous cell carcinoma (SCC). (a) In a focus of neoplastic transformation (arrow), cytoplasmic COX-2 immunostaining is observed, whereas the neighbouring normal mucosa is not immunoreactive. (c) In well differentiated (G2) laryngeal SCC most tumor cells display cytoplasmic COX-2 immunostaining with perinuclear reinforcement. Note that keratinized areas amongst the tumor cells are not immunostained for COX-2. (e) In high grade (G4) non keratinizing laryngeal SCC, tumor cells are not immunostained for COX-2 whereas some stromal cells display COX-2 immunoreactivity. (g) Western blot analysis of COX-2 expression on tumor samples (lanes 1, 2). The tissue samples assayed were from laryngeal SCC shown in (a,b) (lane 2), (c,d) (lane 1). β-actin was shown as a control for protein loading. Scale bar = 100 μm (a,e,f); 50 μm (c,d); 200 μm (b).

According to the cut-off criteria, the number of COX-2 positive and negative tumors was 19 and 42, respectively. Table I shows the distribution of COX-2 according to clinico-pathological characteristics in 61 primary laryngeal SCC. No significant relationship between COX-2 positivity and age, tumor stage, T-classification or tumor site, was observed. The vast majority of patients (53/61, 86.9%) resulted in N0 stage (clinically negative neck node tumors). The observation that COX-2 expression was associated with nodal status could be biased because clinical staging is not so accurate as staging neck dissection. It was interesting, however, that the 8 node positive cases were all COX-2 negative. A significant distribution of COX-2 positivity with respect of tumor grade was observed. The percentage of COX-2 positive tumors was 12/23 (52.2%) in low grade (G1+G2) and 7/38 (18.4%) in high grade (G3+G4) tumors, respectively.

Table I. Cyclooxygenase-2 Immunostaining According to Clinico-Pathological Characteristics in 61 Squamous Laryngeal Cancer Patients
 Tumor immunostaining
COX-2−COX-2+p-Value
Age (years)
 ≤6017 (71.4%)9 (28.6%)
 >6025 (23.7%)10 (76.3%)0.78
Histopathological  grading
 G1–G211 (47.8%)12 (52.2%)
 G3–G431 (81.6%)7 (18.4%)0.004
Stage
 I–II10 (55.6%)8 (44.4%)
 III–IV32 (74.4%)11 (25.6%)0.96
Lymph-node  involvement
 No34 (64.2%)19 (35.8%)
 N+8 (100%)00.04
T-classification
 1–217 (65.4%)9 (34.6%)
 3–425 (71.4%)10 (28.6%)0.78
Tumor site
 Glottic2 (50.0%)2 (50.0%)
 Supraglottic16 (84.2%)3 (15.8%)
 Transglottic24 (63.2%)14 (36.8%)0.17

During the follow-up period, regional relapses (lymph-node metastasis: 18/61, 30%) and local recurrences (13/61, 21%) were observed At the end of the study, 24/61 (39%) patients had died of cancer.

Cox univariate regression analysis, using COX-2 integrated density as a continuous covariate, indicated that the levels of tumor positivity were inversely associated with the risk of death (χ2 = 5.92, p = 0.015) and relapse (χ2 = 6.33, p = 0.012).

The survival curves according to COX-2 positivity status showed a significant relationship between low COX-2 integrated density (<435) and short relapse-free survival (Fig. 2a). The 5-year relapse-free survival was 84% (95% confidence interval: 63–104%) for patients with positive tumors compared to 30% (95% confidence interval: 12–47%) for those with negative tumors (p = 0.0012). Similarly, the overall survival curves (Fig. 2b) indicated that patients with positive tumors have a longer overall survival rate than those with negative tumors. Thus, the 5-year survival rate was 100% for patients with COX-2 positive tumors compared to 34% (95% confidence interval: 11–58%) for patients with COX-2 negative tumors (p = 0.0027).

Figure 2.

Survival rate according to cyclooxygenase-2 (COX-2) status ( COX-2 + = integrated density ≥ 435; COX-2 − = integrated density < 435) in 61 primary laringeal cancer patients (a,b) and in the subgroup (n = 53) of lymph node negative ones (c,d).

N0 stage cases (53/61, 86.9%) may be considered representative of those laryngeal SCC patients in which it is of outmost importance to find new prognostic parameters to identify high- and low-risk cases to address the therapeutic management. Consequently, we decided to assess whether COX-2 status is of prognostic significance in the subgroup of node negative patients. A significant relationship between COX-2 negative tumors (COX-2 integrated density <435) and shorter relapse-free (Fig. 2c) and overall survival (Fig. 2d) rate was found. In particular, the 5-year overall survival was 51% (95% confidence interval: 31–71%) for patients with COX-2 negative tumors compared to 100% for patients with COX-2 positive tumors (p = 0.012). The 5-year relapse-free survival rate was 31% (95% confidence interval: 12–51%) for patients with COX-2 negative tumors compared to 84% (95% confidence interval: 63–104%) for patients with COX-2 positive tumors (p = 0.0053).

Although statistically not significant, the correlation between COX-2 integrated density and EGFR levels displayed a trend toward an inverse correlation (n = 51; r = 0.24; p = 0.06). When the patient population was grouped in accordance with clinical outcome (uncensored [died or relapsed] and censored [alive or relapse-free or died from other causes]), a highly significant inverse correlation was found only in the uncensored patient subgroup (Fig. 3).

Figure 3.

Upper panel: Relationship between epidermal growth factor receptor (EGFR) (fmol/mg of protein) and cyclooxygenase-2 (COX-2) integrated intensity in uncensored (died or relapsed) and censored (alive and died from other causes or relapse-free) patients. Lower panel: plots of the estimates of the relapse-free survival and the overall survival at 5-year follow-up in 51 primary laringeal cancer patients as a function of the values of COX-2 integrated density and EGFR covariates. X-axes: case number; y-axes: on the same scale COX-2 integrated density, EGFR and survival were indicated in the appropriate measure units, as indicated in the legend. Note that COX-2 integrated density values were divided for 10 to fit them in the same y-axis scale of EGFR and survival values.

Cox's regression analysis showed that low levels of COX-2 integrated density combined with high levels of EGFR provided strong prediction, at 5-year follow-up, of both poor overall survival (χ2 = 12.905; p = 0.0016) and relapse-free survival (χ2 = 9.209; p = 0.01) (Fig. 3).

Table II shows the univariate analysis of prognostic variables for overall and relapse-free survival. Lymph node involvement, age ≤60 years, high grade (G3-G4) as well as high EGFR (≥16 fmol/mg of protein) and low COX-2 integrated density (<435), were associated with an increased risk of relapse and death.

Table II. Univariate Analysis of Prognostic Variables in Squamous Laryngeal Cancer Patients
CovariateRelapse-free survivalOverall survival
RR1(CI 95%)2pRR(CI 95%)p
  • 1

    Unadjusted relative risk.

  • 2

    95% confidence intervals.

Age (years)
 >6011
 ≤603.04(1.4–6.6)0.0051.83(0.8–4.2)0.15
Histopathological grading
 G1–G211
 G3–G42.80(1.1–6.9)0.0254.55(1.3–15.3)0.014
Stage
 I–II11
 III–IV1.53(0.6–3.6)0.332.31(0.8–6.3)0.10
Lymph-node involvement
 No11
 Yes3.10(1.2–7.8)0.0165.60(2.1–14.9)0.0005
T-classification
 1–211
 3–41.32(0.6–2.8)0.481.6(0.7–3.7)0.29
Tumor site
 Supraglottic11
 Transglottic1.17(0.3–4.9)0.832.36(0.3–17.6)0.40
EGFR (fmol/mg of protein)
 <1611
 ≥162.75(1.3–5.7)0.0063.13(1.4–6.9)0.005
COX-2 integrated density
 ≥43511
 <4357.73(1.8–33.6)0.00612.8(1.63–101.4)0.015

In the multivariate analysis, low COX-2 (<435 integrated density) and high EGFR (≥16 fmol/mg of protein) expression retained an independent negative prognostic significance relative to the overall survival. When the relapse-free survival was considered, only COX-2 retained an independent negative prognostic significance (Table III). At variance with COX-2, EGFR, when included in the multivariate analysis, lost its significance as prognostic factor probably as a consequence of its association with COX-2 status. EGFR status retained its prognostic significance in the mulivariate analysis (p = 0.047 and p = 0.036 for relapse-free and overall survival, respectively) when the COX-2 covariate was excluded from the model.

Table III. Multivariate Analysis of Prognostic Variables in Squamous Laryngeal Cancer Patients
CovariateRelapse-free survivalOverall survival
RR1(CI 95%)2pRR(CI 95%)p
  • 1

    Unadjusted relative risk.

  • 2

    95% confidence intervals.

Age (years)
 >6011
 ≤601.86(0.7–5.0)0.211.29(0.4–3.7)0.63
Histopathological grading
 G1–G211
 G3–G41.25(0.4–4.0)0.712.9(0.7–12.0)0.15
Stage
 I–II11
 III–IV1.46(0.2–9.6)0.702.7(0.4–19.5)0.32
T-classification
 1–211
 3–41.44(0.3–7.0)0.650.9(0.2–4.5)0.35
Tumor site
 Supraglottic11
 Transglottic0.25(0.03–2.4)0.230.26(0.02–4.3)0.35
EGFR (fmol/mg of protein)
 <1611
 ≥161.92(0.7–5.1)0.192.30(1.0–8.8)0.05
COX-2 positivity (integrated density)
 ≥43511
 <4358.60(1.7–42.5)0.00811.11(1.26–98.2)0.03

It has been shown that ligand activation of EGFR induces COX-2 expression both in normal keratinocytes and squamous carcinoma cells.11 Despite high levels of EGFR expression, the vast majority of poorly differentiated SCC appeared COX-2 negative. To assess whether EGFR activation induces COX-2 expression in these tumors, we utilized Hep2 and CO-K3 laryngeal squamous carcinoma cell lines. CO-K3 cells were obtained from a patient (belonging to this series) with an EGFR positive, COX-2 negative, poorly-differentiated tumor. In vitro, both cell lines expressed high affinity (∼0.5 nM) EGFR (Fig. 4a,b) and produced low basal levels of PGE2 (Fig. 4c,d). When stimulated with EGF, however, only Hep2 cells exhibited a time-dependent increase in PGE2 production (Fig. 4c,d). Moreover, EGF-stimulated PGE2 production was inhibited by both indomethacin (10 μM) and NS-398 (1 μM), but not by valeril salicilate, up to 1 mM (not shown), indicating that PGE2 increase was COX-2 mediated. As revealed by Western blot analysis, EGF stimulation produced a time-dependent induction of COX-2 expression in Hep2 (Fig. 4e) but not in CO-K3 cells (not shown). The kinetic of COX-2 induction paralleled that of PGE2 production (Fig. 4evs. 4c). As assessed by Western blot analysis, both Hep2 and CO-K3 constitutively expressed COX-1 whose expression was not modified by the presence of EGF (data not shown).

Figure 4.

Direct plot of epidermal growth factor receptor (EGFR) binding assay on Hep2 (closed circles = specific binding; open circles = non-specific binding) and CO-K3 (open squares = specific binding; closed squares = non-specific binding) cell lines (a). Scatchard plot of data in (a) (closed circles = Hep2; open squares = CO-K3) (b). The apparent equilibrium dissociation constant (Kd) were 0.51 nM and 0.55 nM and the number of binding sites per cell were 27,250 and 21,200, for Hep2 and CO-K3, respectively. Time course of epidermal growth factor (EGF) stimulation of PGE2 production in Hep2 (c) and CO-K3 (d) cell lines. Closed circles = 100 nM EGF; open circles = 50 nM EGF; open squares = no EGF addition. Western blot analysis of time-dependent EGF induction of cyclooxygenase-2 (COX-2) expression in Hep2 cells (e) and of constitutive cyclooxygenase-1 (COX-1) expression in Hep2 (lane 1) and CO-K3 (lane 2) (f).

DISCUSSION

As revealed by quantitative immunohistochemistry and Western blot analysis, COX-2 protein was weakly expressed in normal squamous epithelium neighbouring tumor. In laryngeal SCC, COX-2 was strongly expressed in the well differentiated areas, whereas it was apparently absent in the poorly differentiated tumor areas. These findings were in agreement with previous observations in esophageal,22 colon,23, 24 pulmonary and mammary carcinoma.24 A relationship between COX-2 positivity and tumor cell differentiation was also evidenced by the highly significant association of COX-2 expression with low histopathological grade (G1-G2) and S100A2 positivity.25

Survival analysis revealed that patients with COX-2 negative tumors had a worse clinical outcome as compared to patients bearing a COX-2 positive neoplasia. This finding suggests that COX-2 negativity is associated with a more aggressive behaviour of laryngeal SCC. COX-2 expression, together with MeHPLAase26 and S100A225 status, may represent a prognostic indicator that allows to discriminate high- and low-risk patients in the lymph node negative subgroup.

The expression of EGFR is a well recognized unfavorable prognostic marker in laryngeal SCC.9, 10 Indeed, high levels of EGFR were inversely correlated with COX-2 expression only in patients with an adverse clinical outcome. Interestingly, it has been recently pointed out that interferon-γ induces a 9-fold increase in COX-2 mRNA levels through the production of transforming growth factor-α and activation of EGFR in normal keratinocytes.11 On the contrary, in some squamous cell lines, either interferon-γ or transforming growth factor-α failed to induce COX-2 expression, probably as a consequence of alterations in the cell response pathways.11 This finding could explain why poorly differentiated laryngeal SCC, containing high levels of EGFR, do not express COX-2. In addition, the EGFR positive CO-K3 cells, obtained from an EGFR positive, COX-2 negative, poorly-differentiated tumor, did not express COX-2 when stimulated with EGF. Interestingly, in this cell line, EGF stimulation failed to activate p38 mitogen-activated protein kinase (unpublished results) which has been reported to be involved in stabilizing COX-2 mRNA.11

EGFR positive, COX-2 negative, CO-K3 cells did not change PGE2 production when stimulated with EGF, suggesting that cancer cells in poorly differentiated laryngeal SCC have a decreased ability to produce COX-2 dependent PGE2. It has been reported that in head and neck cancers, the majority of PGE2 measured is probably derived from host inflammatory or stromal cells.27, 28 Therefore, we cannot exclude that in COX-2 negative laryngeal SCC, PGE2 is produced by stromal and inflammatory cells. In addition, the prognostic significance of PGE2 levels in squamous cell carcinoma has not been adequately addressed. In fact, enhanced PGE2 production has been reported to have an adverse effect, increasing invasiveness and metastatic potential of esophageal squamous cell carcinoma lines growing in nude mice.29) On the other hand, a lowered tumor level of PGE2 is associated with a decreased 2-year disease-free survival in head and neck cancer patients.30

Our results showed that expression of COX-2 is upregulated in well differentiated laryngeal SCC, suggesting that its expression may be involved in the pathogenesis or growth of well differentiated tumors. Alternatively, COX-2 over-expression in these tumors may be a consequence of squamous differentiation in an abnormal setting. In this view, it is relevant, from our results, that COX-2 expression is lost when laryngeal SCC progress to a more aggressive phenotype, becoming less differentiated.

It would be important to investigate whether the loss of COX-2 expression occurs in other cancer types during tumor progression.

Ancillary