By continuing to browse this site you agree to us using cookies as described in About Cookies
Notice: Wiley Online Library will be unavailable on Saturday 7th Oct from 03.00 EDT / 08:00 BST / 12:30 IST / 15.00 SGT to 08.00 EDT / 13.00 BST / 17:30 IST / 20.00 SGT and Sunday 8th Oct from 03.00 EDT / 08:00 BST / 12:30 IST / 15.00 SGT to 06.00 EDT / 11.00 BST / 15:30 IST / 18.00 SGT for essential maintenance. Apologies for the inconvenience.
Head and neck cancer (HNSCC) is one of the most distressing human cancers, causing pain and affecting the basic survival functions of breathing and swallowing. Mortality rates have not changed despite recent advances in radiotherapy and surgical treatment. We have compared the expression of over 13,000 unique genes in 7 cases of matched HNSCC and normal oral mucosa. Of the 1,260 genes that showed statistically significant differences in expression between normal and tumor tissue at the mRNA level, the three top ranking of the top 5% were selected for further analysis by immunohistochemistry on paraffin sections, along with the tumor suppressor genes p16 and p53, in a total of 62 patients including 55 for whom >4-year clinical data was available. Using univariate and multivariate survival analysis, we identified SPARC/osteonectin as a powerful independent prognostic marker for short disease-free interval (DFI) (p < 0.002) and poor overall survival (OS) (p = 0.018) of HNSCC patients. In combination with other ECM proteins found in our analysis, PAI-1 and uPA, the association with DFI and OS became even more significant (p < 0.001). Our study represents the first instance of SPARC as an independent prognostic marker in HNSCC.
Head and neck cancer (HNSCC) is the sixth most common neoplasm in the world today with approximately 900,000 cases diagnosed world-wide.1 In the United States there were 50,000 new cases and 15,600 deaths in 2000.2 When detected early, HNSCC has a 75% survival at 5 years but the majority present with metastatic disease decreasing survival at 5 years to 35%. The management of HNSCC involves a multi-disciplinary approach of surgery, radiotherapy and chemotherapy. Surgery involves resection in early stages of the disease but for later stages includes removal of regional neck nodes, reconstruction and adjuvant radiotherapy (XRT). Surgery has the best outcome but in large or later-staged tumors can be complicated with disfigurement, pain and minimal benefits to patients with short disease-free interval (DFI). XRT is less disfiguring but has complications such as pain, dry mouth, wound infection and osteonecrosis. There is a need to identify patients with low risk of recurrence to benefit from resection followed by reconstructive surgery, or to offer conservative treatment such as XRT alone to patients whose tumors can be predicted to recur. A pre-operative marker could help significantly in deciding the most appropriate treatment for a particular patient.
Studies published previously on HNSCC expression profiling used pooled HNSCC cell lines or unmatched normal tissue3, 4, 5 and none had validated their results with clinicoparameters or prognosis.6 In our study, candidate markers for prognosis were identified by analysing the differences in expression profiles of HNSCC cases matched with autologous normal oral mucosa and validated on a group of patients for whom prognosis was known. The aim of our study was to identify preoperative markers that could be used routinely to distinguish patients at risk of recurrence or aggressive disease. We found that SPARC was an independent prognostic marker for DFI and overall survival (OS) in HNSCC.
DFI, disease-free interval; ECM, extracellular matrix; ESTs, expressed sequence tags; HNSCC, head and neck cancer; OS, overall survival, PAI-1, urokinase plasminogen activator inhibitor type-1; SPARC, secreted protein that is acidic and rich in cysteine; uPA, urokinase plasminogen activator; XRT, radiotherapy.
Material and methods
Written consent was obtained for tumor banking for all patients undergoing investigation and treatment in the Head and Neck Clinic under a protocol approved by the Princess Alexandra Hospital Ethics Committee and the Queensland Institute of Medical Research Human Ethics Committee.
Histological sections were obtained from 62 patients with HNSCC that had complete clinical data and follow up. Seven of these patients had cDNA expression profiling and IHC validation carried out. The minimum and maximum follow-up in these 7 patients were 8 months and 2.2 years. The remaining 55 patients, who were treated predominantly during the period 1997–1999, all had a minimum of 4-year follow-up and a maximum of 6.9 years. There were 49 males and 13 females. The minimum and maximum ages were 20 and 88 years respectively with the average age of 60.2 ± 15 years. There were 5 patients with cancer arising from the oral buccal mucosa, 23 patients with floor of mouth cancer, 15 patients with oropharyngeal cancer and 19 patients with tongue cancer.
cDNA expression profiling and data analysis
Total RNA was extracted from matched HNSCC and autologous normal oral mucosa (lateral margin of oral mucosa after tumor excision and clearance confirmed on frozen section) in 3 patients with floor of mouth cancer, 2 patients with tongue cancer and 2 patients with oropharynx (tonsillar bed) cancer for cDNA expression profiling. RNA extraction was carried out using the RNeasy Midi Kit (Qiagen, Victoria, Australia) with combined proteinase K digestion as described by the manufacturer. Total RNA (20 μg) of sample (labeled with Cy5) or Universal Human Reference Total RNA (Stratagene, La Jolla, CA; labeled with Cy3) was mixed with 40 U RNasin (Promega, Sydney, NSW, Australia), 4 μg oligo d(T15) and 6 μg random hexamer primers (Invitrogen, Carlsbad, CA) and labeled using the amino-allyl (indirect) method as described previously.7 Hybridisation was carried out in 20 μg of Cot-1 DNA (Invitrogen), 20 μg of poly dA (Sigma) and 80 μl of DIG Easy Hybridisation solution (Roche Diagnostics, Castle Hill, NSW, Australia) for 14–16 hr at 37°C in humidified hybridisation chambers (TeleChem International Inc., Sunnyvale, CA). Microarrays were washed twice in a pre-heated (37°C) solution of 1× SSC, 0.1% SDS for 5 min then in 1× SSC for 3 min and finally in 0.1× SSC for 1 min before drying by centrifugation at 100g for 5 min. Microarray slides were immediately scanned in a GMS-418 Confocal Scanner (Genetic MicroSystems/Affymetrix Inc., Santa Clara, CA) and images imported into ImaGene 5.0 (BioDiscovery Inc., Marina Del Rey, CA) for data extraction. Mean signal pixel intensities and mean background pixel intensities for Cy3 and Cy5 channels were imported into GeneSpring 6 (Silicon Genetics, Redwood City, CA) and normalised using the Lowess algorithm to correct for intensity dependent bias, and data filtering to remove experimental noise and poor data before further analysis. The microarray chips used in our study were from the Ontario Cancer Institute (University Health Network, Toronto, Canada) containing 19,200 elements spotted in duplicate across 2 microarray slides, representing 18,107 separate genes/ESTs (expressed sequence tags), equating to 13,131 individual genes (http://www.microarrays.ca). Gene expression was calculated as the ratio to the reference RNA.
Sections (5 μm) of formalin-fixed and paraffin-embedded biopsies were dewaxed and rehydrated. Antigen retrieval was carried out by autoclaving in 0.01 M trisodium citrate buffer (pH = 6.0) at 105°C for 15 min. Immunohistochemistry was carried out as described previously.8 Non-specific binding sites blocked with normal 10% goat serum in PBS at room temperature for 30 min. Sections were incubated with primary antibodies overnight at 4°C at the following concentrations: osteonectin (SPARC; mouse monoclonal, Hematologic Technologies Inc, Essex Junction, VT), 1:5,000 dilution; p53 (D07 antibody, mouse monoclonal Ab, Novocastra Laboratories Ltd, Newcastle upon Tyne, UK) 1:50 dilution; p16 (mouse monoclonal, BD Biosciences Pharmingen, San Diego, CA) 1:50 dilution; PAI-1 (goat polyclonal, American Diagnostics, Stamford, CT), 1:60 dilution; uPA (mouse monoclonal, American Diagnostics), 1:100 dilution. All slides were counterstained with hematoxylin.
Scoring of IHC and analysis
The staining of sections was examined blind by a histopathologist. The overall percentage of cells stained in both tumor and normal tissues was determined for each sample. Staining patterns within the nucleus, cytoplasmic or secreted in the ECM (extracellular matrix) were also noted and an overall percentage score given. To study the possible relationship of the markers with DFI and OS, the overall percentage staining of tumor cells was divided in quartiles and the survival analysis was carried out. The hazard ratios using quartiles had an increasing trend suggesting the survival was inversely related to the intensity of staining. Compared to the lowest quartile, the survival function for the second and third quartiles was not significant and the overall significance was due to the fourth quartile. Therefore, we decided to combine the first 3 quartiles as the cut-off for the negative and positive staining for the marker. In addition, the likelihood ratio tests for the goodness of fit to the model indicated that 2 groups were as adequate fit as 4 groups. A similar approach was applied to uPA and PAI-1 markers when deciding the cut-off for positive and negative status (Table I). Disease-specific DFI and OS survival rates were calculated with the Kaplan-Meier method and their differences were evaluated by the log rank test. A p-value < 0.05 was considered statistically significant. Data analysis was carried out using the SPSS software, version 11.5 released for Windows (SPSS Inc., Chicago, IL).
Table I. IHC Staining Patterns and Overall Percentage Staining Used to Distinguish Between Positive and Negative Staining for Survival Analysis
Staining patterns: + positive; ++ moderate; +++ strong; − absence of staining. Cut off definition: tumors below this value considered negative.
Univariate survival analysis using the Kaplan-Meier survivorship function was carried out to compare the DFI and OS between different categories of clinicopathological and molecular markers. Furthermore, multivariate Cox regression analysis was used to obtain adjusted survival estimates and their statistical significance. Clinicopathological parameters that were statistically significant in the univariate survival analysis were included in the multivariate model. Although the association between age and DFI or OS was not statistically significant, it was an important confounder and changed the estimates of molecular markers by 10% or more, hence it was included.
Expression profiling of HNSCC tumors
We sought to examine differential gene expression in the transition from normal mucosa and primary HNSCC. A panel of 7 tumors with matched normal tissue were collected: 3 patients with floor of mouth cancer, 2 patients with tongue cancer and 2 patients with oropharynx cancer. Total RNA was extracted from the samples and analysed by hybridisation of labeled cDNA to expression microarrays containing over 13,000 individual genes.
Expression profiling data was filtered initially by taking the average of the duplicate clones for the intensity in the sample (Cy5 or red) channel and the reference (Cy3 or green) channel. Dye swap experiments with these microarrays and the same labeling and hybridisation procedure showed 99.5% agreement. Further, a duplicate array from the same RNA sample showed 98% agreement (data not shown). The logarithm of this ratio was then calculated and used as the expression value. As data points with low intensity tend to be noise dominated, we used a quality controlled criteria that required clones to have intensities (red and green for both duplicates) between 80 and 65,000 fluorescence units. Of 19,200 elements in duplicate, 14,303 survived this filter across all 7 normal mucosa and matched primary tumor samples.
Analysis of expression profiling data was carried out using a supervised approach, based on a non-parametric method to determine differential gene expression between normal mucosa samples and matched primary tumor samples. Using the Wilcoxon-Mann-Whitney test, we expected 715 of the 14,303 filtered clones to be different between the 2 groups (p < 0.05) by chance. The supervised analysis yielded 1,260 clones from the filtered list as being differentially expressed at the p < 0.05 level. The number of genes obtained was significantly greater than that expected by chance (p < 0.001, χ2 test). The average expression value for the primary tumor samples was calculated and divided by the average expression value for the normal mucosal samples. A selection of the largest expression differences between normal mucosa and primary tumor is shown in Table II.
Table II. List of 50 Genes With the Most Statistically Significant Differences Between the Averages of Matched HNSCC and Autologous Normal Oral Mucosa Samples1
Average of tumor/normal mucosa ratio
Based on a non-parametric method to determine differential gene expression (Wilcoxon-Mann-Whitney test with p-value cutoff at 0.05).
Upregulated in tumor
Plasminogen activator inhibitor type 1 (PAI-1)
Secreted phosphoprotein 1 (osteopontin)
Epidermal growth factor receptor
Osteoblast specific factor 2 (fasciclin I-like)
Collagen, type I, alpha 2
Solute carrier family 2 (facilitated glucose transporter)
1NFLS Homo sapiens cDNA clone IMAGE:232672 5′, mRNA sequence
Protein phosphatase 1
Cytochrome c oxidase subunit VIIb
Hypothetical protein FLJ31606
Sorbin and SH3 domain containing 1
Homo sapiens clone 25186 mRNA sequence
Solute carrier family 25 (mitochondrial carrier; phosphate carrier)
Nicotinamide nucleotide transhydrogenase
Homo sapiens cDNA FLJ36815
Soares infant brain 1NIB Homo sapiens
Thioredoxin interacting protein
Hypothetical protein MGC41924
Homo sapiens, clone IMAGE:4183312, mRNA, partial cds
NDRG family member 2
Chemokine (C-C motif) ligand 2
fem-1 homolog a (C.elegans)
Histone deacetylase 5
Muscle RAS oncogene homolog
Crystallin, alpha B
Tropomyosin 2 (beta)
Homo sapiens mRNA full length insert cDNA clone
Protein kinase (cAMP-dependent, catalytic) inhibitor alpha
Verification of SPARC, PAI-1 and uPA expression at the protein level by IHC using matched normal mucosa and tumor sections
Sections from samples for microarray analysis were selected that included normal mucosa and tumor from the same patient on the same slide, for comparison of protein level and localisation by IHC staining (Fig. 1).
Selection of target genes for further study was based on availability of antibody for immunohistochemistry. We noted SPARC and PAI-1 (urokinase plasminogen activator type 1 inhibitor) in the top 5% of genes with statistically significant differences between mucosa and tumor. uPA (urokinase plasminogen activator) was upregulated but was excluded by strict filtering during data analysis. We have included uPA in our IHC, however, because of its close association with PAI-1 and because both are prognostic markers in several cancers.9, 10 In addition, Pasini et al.11 using Northern analysis of 91 matched HNSCC patients have shown increased mRNA expression of uPA and PAI-1 in tumors compared to normal mucosa.
The pattern for SPARC expression in tumor tissue showed prominent staining of stromal fibroblasts both within the tumor and adjacent to it. Tumor cells staining were predominantly noted in the cytoplasm and adjacent ECM. Normal mucosa was negative. The staining pattern for PAI-1 and uPA were also similar, occurring predominantly in the cytoplasm and adjacent ECM of tumor cells. The staining pattern of uPA demonstrated a gradual increased of staining from the transition of normal mucosa to the area of tumor cells (Fig. 1e).
Correlation of HNSCC markers with prognosis
To test the clinical significance of the results obtained by expression profiling, IHC was carried out on a further independent group of 55 samples (Fig. 2) from patients with a minimum of 4-year follow-up period. Sixty-two patients were validated on clinicopathological parameters and survival time analysis.
The univariate Kaplan-Meier analysis of clinical parameters demonstrated a statistically significant difference for DFI and OS between categories of nodal status, neural invasion tumor size and XRT (Table III; Fig. 3a,b). Patients with negative nodes, smaller tumor size or no neural invasion had longer survival. After adjusting for age, nodal status and neural invasion, survival between different categories of tumor size (p = 0.65) and XRT (p = 0.25) was not statistically significant. At the molecular level (Table IV), patients with negative SPARC (Fig. 3c,d), negative PAI-1 or negative uPA expression had better survival. The median DFI and OS were highest in the uPA negatives (Table IV), but with only 7 cases involved no statistical significance could be reached. No significant association was found for p53 and p16 expression (Table IV).
Table III. Univariate Analysis of Clinicopathological Parameters for DFI and OS of the 62 HNSCC Patients Studied
Cox regression multivariate analysis was used when calculating the adjusted hazard ratio for the three molecular markers, after adjusting for clinicopathological parameters that were statistically significant in the univariate analysis (Table V). The association of uPA expression with DFI was not statistically significant after adjusting for age, nodal status and neural invasion (p = 0.20), however, SPARC (p = 0.007) and PAI-1 (p = 0.04) were still statistically significant. As SPARC and PAI-1 were correlated (Fisher's exact test p = 0.03), including both proteins in the multivariate model weakened the association for both markers although SPARC was still associated marginally with DFI (p = 0.06). With the OS, both SPARC (p = 0.08) and PAI-1 (p = 0.08) approached statistical significance after adjusting for age, nodal status and neural invasion.
Table V. Hazard Ratios and Their Confidence Interval for SPARC, PAI- 1 and uPA for DFI After Adjusting for Age, Nodal Status and Neural Invasion for DFI and OS
Hazard ratio (95% CI)
Hazard ratio (95% CI)
We determined if positive staining of these ECM markers, SPARC, uPA and PAI-1 in combination with each other (Fig. 3e,f) or with other clinicopathological parameters such as nodal status (Fig. 3g,h) would improve the ability to distinguish survival (Table VI). Patients (n = 13) with all ECM proteins staining positive had a median DFI of 15.4 months with 10/13 (76.9%) recurrences. In patients (n = 5) with all ECMs negative, 1/5 (20.0%) had recurrence whereas with one ECM positive or negative (n = 44) 18/44 (40.9%) had recurrences with mean DFI of 72.1 and 48.8 months respectively (p < 0.001) (Table VI).
Table VI. Enhanced Prediction of DFI and OS by Combining Clinical Parameters and Novel Markers With SPARC
SPARC was elevated consistently in our matched cDNA microarray analysis in the transition from normal mucosa to tumor tissues in the seven patients studied. We further validated these results with direct IHC staining for SPARC on paraffin sections. SPARC, also known as osteonectin or BM-40 (basement membrane), is a calcium-binding and collagen-binding glycoprotein associated with stress-related ECM. SPARC is released by both malignant and normal cells derived from all primordial germ layers. It is found in osteoblasts,12 fibroblasts,13 endothelial cells during angiogenesis,14 tissue remodeling, cell migration and proliferation.12 TGF-β1 plays an important role in the initiation and regulation of SPARC production.15, 16
SPARC increases the production of collagenase, stromelysin, gelatinase, fibronectin and laminin in fibroblasts.17 This causes degradation of both interstitial and basement membrane matrices18 and increases endothelial permeability with the appearance of intercellular gaps, which provide a pathway for extravasation of tumor cells.19 The exact role of SPARC in tumor angiogenesis is still unclear but may include a role in VEGF (vascular endothelial growth factor) functions and in extravasation of tumors cells mediated by increased permeability of the endothelial barrier.17
SPARC overexpression has previously been associated with human tumors.20 Porte et al.21 identified an association of SPARC with colorectal carcinoma, where neoplastic progression was associated with overexpression of the stromelysin-3 and SPARC genes. Massi et al.22 demonstrated in 188 patients with thin (<0.75 mm) melanoma that SPARC positivity on IHC staining correlated with significantly poorer survival.22 Our study is the first to our knowledge to associate SPARC expression with decreased survival in HNSCC.
SPARC is also causally involved with the ECM urokinase plasminogen system PAI-1 and uPA in tumorigenesis.23, 24 On our microarray analysis, PAI-1 expression was the most significant difference in expression found between normal mucosa and primary HNSCC tumors. PAI-1 is an important regulator of uPA in the urokinase plasminogen activator system, causally involved in metastasis. PAI-1 expression on its own was significant as an independent prognostic marker in DFI (log rank = 4.35, df = 1, p = 0.037) and OS (log rank = 3.85, df = 1, p = 0.049). Despite having the longest median in DFI and OS, uPA staining on its own did not reach statistical significant P levels in DFI (log rank = 3.22, df = 1, p = 0.072) and OS (log rank = 2.75, df = 1, p = 0.097) as only 7 of 62 patients had negative staining and all had good prognosis (Table IV). We feel however, that in a larger study uPA may prove to be a sensitive marker in predicting recurrence.
In combination, SPARC, uPA and PAI-1 (SUP) staining showed improved sensitivity as markers for prognosis. The combination outperformed the single factors as well as traditional prognostic markers such as nodal involvement and tumor size. It may help to distinguish which patients without nodal involvement are at risk of recurrence (p < 0.0001). Patients who had no nodal involvement (N−) with all 3 ECM stained positive (SUP+) had a poorer DFI than patients with nodal involvement (N+) and at least one of the ECM (SUP−) stained negative, indicating that the combination of SPARC staining with uPA and PAI-1 (SUP) is an extremely sensitive prognostic marker (Fig. 3g,h).
P53 and p16 IHC staining was carried out on HNSCC paraffin sections as a standard molecular reference with 33% and 20% positivity respectively. Mutation in the p53 gene causes stability of the protein, and detection of the abnormal protein with IHC in HNSCC has been reported to occur early in HNSCC.25 Overall alteration of the p16 protein occurs in 70% of all HNSCC,26 with somatic mutation of the gene (CDKN2A) occurring in 10% and homozygous deletion occurring in 50% of cases studied.27 Despite the prominence of mutation in these tumor suppressor genes in HNSCC progression, their expression in paraffin sections did not predict DFI or OS in our study or in similar studies.28, 29
Of all the data available, the most sensitive predictor of DFI and OS was the combination of SPARC positivity and evidence of neural invasion by the tumor, these patients having the poorest outcome regardless of nodal status. Neural invasion is a post-operative finding and determining neural involvement may not always be feasible by the histopathologist because of tumor site and size. Our primary aim was to determine a pre-operative marker to distinguish patients that will benefit from surgery. SPARC in combination with PAI-1 and uPA, as opposed to neural invasion status of the tumor, is equally sensitive.
These markers would help clinicians distinguish which node-negative patients should have a prophylactic neck dissection and or adjuvant XRT has enormous benefits in terms of morbidity, subsequent surgery and cost. It may also help decide which patients with operable lesions that would benefit from resection and reconstruction surgery. In patients where poor prognosis is expected regardless of treatment, a more conservative approach could be adopted or XRT offered. In the future if validated in large sample numbers these ECM proteins may have potential to be serum markers for early recurrences, and to be exploited for therapeutic purposes.
Mr. C. Winterford, Department of Histopathology, University of Queensland Graduate Medical School assisted in the IHC work.