Novel markers for poor prognosis in head and neck cancer

Abstract

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

Consent

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.

Patient samples

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.

Immunohistochemistry

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

Protein	IHC staining patterns in tumor cells1			Overall percentage staining cut off used
	Nucleus	Cytoplasmic	Secreted (ECM)
1 Staining patterns: + positive; ++ moderate; +++ strong; − absence of staining. Cut off definition: tumors below this value considered negative.
P16	++	++	−	50
P53	++	++	−	50
SPARC	−	+++	+++	30
PAI-1	−	+++	+++	20
uPA	+	+++	+++	10

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.

Results

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

Common name	Genbank number	Map	Description	Average of tumor/normal mucosa ratio
1 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
SERPINE1	R21221	7q21.3–q22	Plasminogen activator inhibitor type 1 (PAI-1)	6.222
SPP1	R97904	4q21–q25	Secreted phosphoprotein 1 (osteopontin)	5.790
EGFR	H03729	7p12	Epidermal growth factor receptor	4.868
OSF-2	W35228	13q13.1	Osteoblast specific factor 2 (fasciclin I-like)	4.766
COL1A2	N30461	7q22.1	Collagen, type I, alpha 2	4.036
	H80736	5	ESTs	3.399
THH	BQ023400	1q21.3	Trichohyalin	3.398
SLC2A1	BQ000388	1p35–p31.3	Solute carrier family 2 (facilitated glucose transporter)	3.360
COL5A2	N43019	2q14–q32	Collagen, type V, alpha 2	3.095
	H62723		1NFLS Homo sapiens cDNA clone IMAGE:206622 5′	3.004
MMP12	N41372	11q22.3	Matrix metalloproteinase 12 (macrophage clastase)	2.821
PKM2	BQ052715	15q22	Pyruvate kinase, muscle	2.752
SF3A1	H04610	22q12.2	Splicing factor 3a, subunit 1, 120kDa	2.703
FLJ10432	R26193	19p12	Hypothetical protein FLJ10432	2.682
BAL	AA151346	3q13–q21	B aggressive lymphoma gene	2.674
CXCL10	AA150307	4q21	Chemokine (C-X-C motif) ligand 10	2.648
ILIRAP	R37898	3q28	Interleukin 1 receptor accessory protein	2.566
LOC113251	R92261	12q13.12	c-Mpl binding protein	2.543
IGFBP3	N31417	7p13–p12	Insulin-like growth factor binding protein 3	2.481
TUBA3	BM722402	12q12–12q14.3	Tubulin, alpha 3	2.476
KCNS1	R87913	20q12	Potassium voltage-gated channel, delayed-rectifier	2.410
	R21229	1	ESTs	2.393
TRIP6	H45501	7q22	Thyroid hormone receptor interactor 6	2.388
ITGA6	H16046	2q31.1	Integrin, alpha 6	2.336
Downregulated in tumor
	H72616		1NFLS Homo sapiens cDNA clone IMAGE:232672 5′, mRNA sequence	0.362
PPP1R12B	H06909	1q32.1	Protein phosphatase 1	0.361
CLU	BQ045101	8p21–p12	Clusterin	0.361
COX7B	H49485	Xq13.2	Cytochrome c oxidase subunit VIIb	0.360
FLJ31606	H20390	16q24.3	Hypothetical protein FLJ31606	0.359
SORBS1	W31730	10q23.3–q24.1	Sorbin and SH3 domain containing 1	0.353
LOC56990	B1553961	5q	Homo sapiens clone 25186 mRNA sequence	0.350
DMN	R70164	15q26.3	Desmuslin	0.348
SLC25A3	BQ064618	12q23	Solute carrier family 25 (mitochondrial carrier; phosphate carrier)	0.343
NNT	H15871	5p13.1–5cen	Nicotinamide nucleotide transhydrogenase	0.342
	R60657	1	Homo sapiens cDNA FLJ36815	0.340
	R14580		Soares infant brain 1NIB Homo sapiens	0.331
TXNIP	BM876583	1q21.1	Thioredoxin interacting protein	0.327
MGC41924	R56760	6q22.33	Hypothetical protein MGC41924	0.322
	AA126193	2	Homo sapiens, clone IMAGE:4183312, mRNA, partial cds	0.320
NDRG2	N80357	14q11.2	NDRG family member 2	0.319
CCL2	R75975	17q11.2–q21.1	Chemokine (C-C motif) ligand 2	0.319
FEMIA	BM718878	19p13.3	fem-1 homolog a (C.elegans)	0.315
KIAA1029	R90895	5q33.1	Synaptopodin	0.314
HDAC5	R14363	17q21	Histone deacetylase 5	0.307
MRAS	H06847	3q22.3	Muscle RAS oncogene homolog	0.271
CRYAB	H27592	11q22.3–q23.1	Crystallin, alpha B	0.268
TPM2	W76573	9p13.2–p13.1	Tropomyosin 2 (beta)	0.228
	R87955		Homo sapiens mRNA full length insert cDNA clone	0.222
APOD	AA131720	3q26.2–qter	Apolipoprotein D	0.215
PKIA	R14129	8q21.11	Protein kinase (cAMP-dependent, catalytic) inhibitor alpha	0.168

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).

Figure 1. Verification of SPARC, PAI-1 and uPA expression at the protein level by IHC using matched normal mucosa and tumor sections. (a) SPARC positivity in tumor (T) and absence in normal mucosa (N) at (5×). Scale bar = 400 μm. (b) SPARC positivity in tumor and ECM staining (10×). Scale bar = 200 μm. (c) PAI-1 positivity in tumor (T) and absence in normal mucosa (N) (5×). Scale bar = 400 μm. (d) PAI-1 positivity in ECM of tumor (10×). Scale bar = 200 μm. (e) uPA positivity in tumor (T) and absence in normal mucosa (N) (5×). Scale bar = 400 μm. (f) uPA positivity in tumor and ECM of tumor (10×). Scale bar = 200 μm.

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.

Figure 2. Representative examples of IHC staining of tumors from HNSCC patients treated in 1997–99. (a) Tumor tissue with SPARC positivity. There is also prominent staining of stromal fibroblasts both within the tumor and adjacent to it. Tumor cell staining was predominantly in the cytoplasm and adjacent ECM (20×). (b) Absence of SPARC expression in tumor cells, fibroblast and ECM (20×). (c) Tumor tissue with PAI-1 positivity. Staining was predominantly in the cytoplasm of tumor cells and in the ECM. There is also prominent staining of stromal fibroblasts both within the tumor and adjacent tissue (20×). (d) Absence of PAI-1 expression in tumor cells, fibroblasts and ECM (20×). (e) Tumor tissue with uPA positivity. There is also prominent staining of stromal fibroblasts both within the tumor and adjacent cells. Tumor cell staining was predominantly in the cytoplasm and adjacent ECM (20×). (f) Absence of uPA expression in tumor cells, fibroblasts and ECM (20×). Scale bar = 100 μm.

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

Parameter	n	DFI in months			OS in months
		Mean (SE)	Median1	p	Mean (SE)	Median1	p
1 —, median not reached.
Gender
Male	49	50.0 (5.0)	50.3		46.3 (3.6)	43.8
Female	13	45.9 (3.7)	—	0.511	53.9 (4.8)	—	0.690
Age group
<40 yr	5	61.7 (4.3)	—		61.7 (4.3)	—
40–60 yr	27	47.0 (4.9)	—		51.1 (4.5)	—
>60 yr	30	44.9 (6.0)	38.9	0.188	47.2 (5.9)	38.9	0.135
Nodal status
Negative	35	54.3 (4.0)	—		54.6 (3.9)	—
Positive	27	41.6 (5.4)	34.9	0.005	46.3 (5.4)	38.9	0.016
T stage
T1	6	59.4 (6.0)	—		59.4 (6.0)	—
T2	20	56.0 (4.8)	—		56.2 (4.8)	—
T3	25	30.7 (4.2)	27.7		35.6 (4.0)	34.9
T4	11	47.3 (8.5)	35.5	0.009	48.6 (8.1)	35.5	0.033
Site
Buccal	5	46.1 (3.1)	40.5		46.1 (3.1)	40.5
FOM	23	44.7 (5.2)	50.3		49.4 (4.7)	—
Oropharynx	15	53.4 (9.4)	—		58.6 (8.8)	—
Tongue	19	39.9 (5.2)	32.1	0.760	40.2 (5.1)	32.8	0.498
Differentiation
Well	9	51.7 (0.5)	—		61.3 (3.1)	—
Moderate	38	51.2 (5.0)	40.5		54.3 (4.8)	43.8
Poor	15	35.5 (7.6)	22.6	0.092	39.28 (7.4)	24.2	0.122
Neural invasion
Negative	37	63.3 (5.0)	—		63.6 (4.9)	—
Positive	25	29.8 (4.2)	29.7	<0.001	35.0 (4.0)	32.8	0.004
Radiotherapy
No	52	52.9 (2.7)	—		52.9 (2.7)	—
Yes	10	47.6 (4.6)	38.9	0.033	51.2 (4.4)	42.4	0.048
Cigarette
Non-smokers	12	54.8 (5.2)	—		55.0 (5.1)	—
Smokers	50	48.2 (4.8)	38.9	0.132	52.0 (4.7)	50.3	0.207

Figure 3. Kaplan-Meier survival curves. (a) DFI with nodal status (N+, nodal involvement; N−, no nodal involvement), Log-rank p = 0.005. (b) OS with nodal status. Log-rank p = 0.016. (c) DFI with SPARC positivity (SPARC+, SPARC staining positive; SPARC−, SPARC staining negative). Log rank = p < 0.002. (d) OS with SPARC positivity. Log-rank p = 0.018. (e) DFI with combination of SPARC, uPA and PAI-1 (SUP) (All negative, all 3 markers stained negative; All positive, all 3 markers stained positive; One positive or negative, at least one marker stained positive or negative). Log-rank p < 0.001. (f) OS with combination of SPARC, uPA and PAI-1 (SUP) positivity. Log-rank P < 0.001. (g) DFI of patients with nodal involvement (N+, nodal involvement; N−, no nodal involvement) and SUP status. (SUP+, all 3 markers positive; SUP−, at least one of the combination negative). Note that N− with SUP+ group had a poorer DFI than N+ with SUP− indicating that the combination of SUP can distinguish at-risk patients in the node negative group. Log-rank p < 0.001. (h) OS of patients with nodal involvement (N+ nodal involvement; N− no nodal involvement). Note that N− with SUP+ group has a poorer OS than N+ with SUP−, again indicating that the combination of SUP is a sensitive prognosis marker even in the node negative group. Log-rank p < 0.001.

Table IV. Univariate Analysis of Molecular Markers for Disease Free Interval (DFI) and Overall Survival (OS)

Parameter	n	DFI in months			OS in months
		Mean (SE)	Median1	p	Mean (SE)	Median1	p
1 —, median not reached.
p53
Negative	48	55.3 (4.7)	—		58.32 (4.5)	—
Positive	14	35.7 (6.8)	25.5	0.117	39.26 (6.5)	38.9	0.133
p16
Negative	51	52.8 (4.6)	50.3		55.6 (4.4)	—
Positive	11	39.0 (8.3)	35.5	0.604	43.2 (8.2)	35.5	0.737
SPARC
Negative	44	59.4 (4.5)	—		59.84 (4.4)	—
Positive	18	28.3 (6.0)	19.4	<0.002	34.94 (6.1)	24.2	0.018
PAI-1
Negative	32	61.1 (5.4)	—		63.1 (5.2)	—
Positive	30	33.4 (3.8)	35.5	0.037	36.8 (3.5)	38.9	0.049
uPA
Negative	7	75.3 (7.3)	—		75.3 (7.3)	—
Positive	55	42.0 (3.5)	40.5	0.072	45.1 (3.3)	43.8	0.097

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

Protein	n	DFI		OS
		Hazard ratio (95% CI)	p	Hazard ratio (95% CI)	p
SPARC
Negative	44	1.0	0.007	1.0	0.078
Positive	18	3.2 (1.3–7.7)		2.3 (0.9–5.9)
PAI-1
Negative	32	1.0	0.036	1.0	0.080
Positive	30	2.3 (1.0–5.3)		2.1 (0.9–4.9)
uPA
Negative	7	1.0	0.199	1.0	0.213
Positive	55	3.9 (0.4–31.2)		3.7 (0.4–30.0)

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

Protein	n	DFI in months			OS in months
		Mean (SE)	Median1	p	Mean (SE)	Median1	p
1 —, median not reached. 2 SUP, SPARC and uPA and PAI-1.
uPA and SPARC
Both negative	5	58.5 (6.8)	—		58.5 (6.8)	—
One positive or negative	40	58.2 (4.8)	—		58.7 (4.7)	—
Both positive	17	25.5 (5.8)	15.4	<0.001	32.0 (6.0)	22.6	0.014
PAI-1 and SPARC
Both negative	27	62.4 (5.7)	—		62.4 (5.6)	—
One positive or negative	22	46.5 (5.0)	—		49.3 (4.5)	—
Both positive	13	20.0 (4.5)	15.4	<0.001	25.0 (4.3)	19.4	<0.001
uPA and PAI-1
Both negative	8	76.3 (6.4)	—		76.3 (6.4)	—
One positive or negative	25	48.3 (4.9)	—		50.4 (4.6)	—
Both positive	29	32.5 (3.8)	32.1	0.027	36.0 (3.6)	35.5	0.041
SUP2
All three negative	5	72.1 (9.9)	—		72.1 (9.9)	—
One positive or negative	44	48.8 (3.57)	—		50.3 (3.4)	—
All three positive	13	20.0 (4.5)	15.4	<0.001	25.0 (4.3)	19.4	<0.001
p53 and SPARC
Both negative	36	61.1 (4.9)	—		61.4 (4.8)	—
One positive or negative	20	37.1 (6.2)	40.5		42.4 (5.9)	40.5
Both positive	6	20.5 (6.7)	15.4	0.005	25.20 (6.1)	22.6	0.022
Neural invasion and SPARC
Both negative	29	66.2 (5.1)	—		66.6 (5.0)	—
One positive or negative	23	41.2 (5.1)	37.4		41.4 (5.0)	40.5
Both positive	10	16.7 (5.3)	10.8	<0.001	26.2 (5.0)	22.6	<0.001

Discussion

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.

Acknowledgements

Mr. C. Winterford, Department of Histopathology, University of Queensland Graduate Medical School assisted in the IHC work.