research article open access a gene signature in histologically normal surgical ... · 2018. 3....

RESEARCH ARTICLE Open Access

A gene signature in histologically normal surgicalmargins is predictive of oral carcinoma recurrencePatricia P Reis1,2†, Levi Waldron3†, Bayardo Perez-Ordonez4, Melania Pintilie5,6, Natalie Naranjo Galloni1,7, Yali Xuan1,Nilva K Cervigne1, Giles C Warner8, Antti A Makitie9, Colleen Simpson10, David Goldstein10, Dale Brown10,Ralph Gilbert10, Patrick Gullane10, Jonathan Irish10, Igor Jurisica3,11* and Suzanne Kamel-Reid1,4,12,13*

Abstract

Background: Oral Squamous Cell Carcinoma (OSCC) is a major cause of cancer death worldwide, which is mainlydue to recurrence leading to treatment failure and patient death. Histological status of surgical margins is acurrently available assessment for recurrence risk in OSCC; however histological status does not predict recurrence,even in patients with histologically negative margins. Therefore, molecular analysis of histologically normalresection margins and the corresponding OSCC may aid in identifying a gene signature predictive of recurrence.

Methods: We used a meta-analysis of 199 samples (OSCCs and normal oral tissues) from five public microarraydatasets, in addition to our microarray analysis of 96 OSCCs and histologically normal margins from 24 patients, totrain a gene signature for recurrence. Validation was performed by quantitative real-time PCR using 136 samplesfrom an independent cohort of 30 patients.

Results: We identified 138 significantly over-expressed genes (> 2-fold, false discovery rate of 0.01) in OSCC. Bypenalized likelihood Cox regression, we identified a 4-gene signature with prognostic value for recurrence in ourtraining set. This signature comprised the invasion-related genes MMP1, COL4A1, P4HA2, and THBS2. Over-expression of this 4-gene signature in histologically normal margins was associated with recurrence in our trainingcohort (p = 0.0003, logrank test) and in our independent validation cohort (p = 0.04, HR = 6.8, logrank test).

Conclusion: Gene expression alterations occur in histologically normal margins in OSCC. Over-expression of the 4-gene signature in histologically normal surgical margins was validated and highly predictive of recurrence in anindependent patient cohort. Our findings may be applied to develop a molecular test, which would be clinicallyuseful to help predict which patients are at a higher risk of local recurrence.

Keywords: oral squamous cell carcinoma, surgical resection margins, global gene expression profiling, prognosticsignature, recurrence

BackgroundOral Squamous Cell Carcinoma (OSCC) accounts for24% of all head and neck cancers [1]. Currently availableprotocols for treatment of OSCCs include surgery,radiotherapy and chemotherapy. Complete surgicalresection is the most important prognostic factor [2],since failure to completely remove a primary tumor is

the main cause of patient death. Accuracy of the resec-tion is based on the histological status of the margins,as determined by microscopic evaluation of frozen sec-tions. Presence of epithelial dysplasia or tumor cells inthe surgical resection margins is associated with a signif-icant risk (66%) of local recurrence [3]. However, evenwith histologically normal surgical margins, 10-30% ofOSCC patients will still have local recurrence [4], whichmay lead to treatment failure and patient death.Since histological status of surgical resection margins

alone is not an independent predictor of local recur-rence [5], histologically normal margins may harborunderlying genetic changes, which increase the risk of

* Correspondence: [email protected]; [email protected]† Contributed equally1Div. of Applied Molecular Oncology, Princess Margaret Hospital, OntarioCancer Institute, University Health Network, Toronto, ON, Canada3Ontario Cancer Institute and the Campbell Family Institute for CancerResearch, Toronto, ON, CanadaFull list of author information is available at the end of the article

Reis et al. BMC Cancer 2011, 11:437http://www.biomedcentral.com/1471-2407/11/437

© 2011 Reis et al; licensee BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative CommonsAttribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction inany medium, provided the original work is properly cited.

mailto:[email protected]:[email protected]://creativecommons.org/licenses/by/2.0

recurrence [6,7]. To date, candidate-gene approach stu-dies have identified genetic alterations in surgical resec-tion margins in head and neck squamous cell carcinoma(HNSCC) from different disease sites, e.g., oral cavity,pharynx/hypopharynx, larynx [6-16]. Genetic alterationsidentified in HNSCC included over-expression of eIF4E[6,9], TP53 [7,11] and CDKN2A/P16 proteins [7]. Otheralterations reported included promoter hypermethyla-tion of CDKN2A/P16 [13] and TP53 mutations [12,16].In addition, promoter hypermethylation of CDKN2A,CCNAI and DCC was associated with decreased time tohead and neck cancer recurrence [10].Despite all these studies, genetic alterations identified

to date have not been used clinically in the assessmentof surgical margins, and no study has developed a genesignature that can accurately predict which patients withOSCC are at a higher risk of disease recurrence. Thismay be due to the lack of studies using high-throughputanalysis of multiple surgical margins and matchedOSCCs to identify deregulated genes with prognosticvalue for recurrence. One probable reason for this defi-cit is that the application of strict inclusion criteria(tumors of a single anatomical site, with all marginsbeing histologically negative, and with multiple marginsof adequate RNA quantity and quality for expressionanalyses), presents a challenge for obtaining adequatesample size to develop a prognostic signature that canpredict recurrence in oral carcinoma. We addressed thischallenge through a hypothesis-driven approach, basedon the hypothesis that commonly over-expressed genesin OSCC and a subset of histologically normal marginsare biomarkers of recurrence. We used meta-analysis ofpublished microarray studies to help identify such geneswith high confidence, along with global gene expressionanalysis of histologically normal margins and their cor-responding OSCCs in our own study.

MethodsPatientsThis work was performed with the approval of the Univer-sity Health Network Research Ethics Board. All patientssigned their informed consent before sample collection,and had surgery as their primary treatment. Tissue sampleswere prospectively collected at time of surgery from theToronto General Hospital, Toronto, Ontario, Canada. Theinclusion criteria of patients were based on primary dis-ease, tumor site (oral cavity), histology of the primarytumor (squamous cell carcinoma), and histological statusof surgical resection margins. All patients included in thisstudy had a final pathology report of histologically normalresection margins, meaning none of the patients had apositive margin. An experienced head and neck pathologist(BP-O) performed histological evaluation of all surgicalmargins (frozen section analysis, as per standard of care) to

ensure that they were histologically normal. A section fromeach tumor sample (OSCC) was stained with hematoxylin-eosin, to confirm the presence of at least 80% tumor cells.The primary tumor, histologically normal margins and adistant normal oral tissue were collected from each patient,in both training and validation cohorts. All tissue sampleswere snap-frozen in liquid nitrogen until RNA extractionand gene expression analysis. We collected a total of 232tissue samples from 54 patients (divided into training andvalidation sets), as described below.

Samples used for microarrays (training set)96 samples (histologically normal margins, OSCC andadjacent normal oral tissues) from 24 patients were usedfor oligonucleotide microarray analysis. Patient clinicaldata for the training set are shown in Table 1.

Samples used for quantitative real-time reverse-transcription PCR (RQ-PCR) (validation set)136 samples (histologically normal margins, OSCC andadjacent normal oral tissues) from an independentcohort of 30 patients were used for RQ-PCR validationanalysis. Patient clinical data for the validation set sam-ples are shown in Table 2.

RNA Isolation, Microarrays and Validation ExperimentsThe detailed protocol for RNA isolation and the experi-mental details for microarrays using the Human Gen-ome HG-U133A 2.0 plus oligonucleotide arrays(Affymetrix, Santa Clara, CA, USA), as well as validationexperiments by RQ-PCR are described in Additional file1, Methods S1.

Bioinformatic AnalysesAll bioinformatic analyses of array data were performedin the R language and environment for statistical com-puting (version 2.10.0) implemented on CentOS 5.1 onan IBM HS21 Linux cluster. We pre-processed eachpublic data set using GCRMA normalization [17] withupdated Entrez Gene-based chip definition files [18],using the affy R package (version 1.24.2) [19]. Geneswith evidence of tumor-normal differential expressionacross all public datasets, with a False Discovery Rate(FDR) of 0.01 and fold-change ≥ 2, were identified usingthe rankprod R package [20]. Gene Ontology enrich-ment analysis was performed with the GOstats R pack-age (version 2.12.0) [21]. We also used GOEAST (GoEnrichment Analysis Software Toolkit) [22] for graphicalrepresentation of GO annotations. Analyses are repro-ducible using the code (available upon request); datahave been deposited in NCBI’s Gene Expression Omni-bus [23] and are accessible through GEO series acces-sion number GSE31056 http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE31056.


Page 2 of 11

http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE31056http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE31056

Meta-analysis of published microarray studiesOur goal was to identify a prognostic signature forrecurrence in OSCC, based on the hypothesis that geneexpression deregulation occurring in OSCC would be an

early indicator of recurrence if gene expression changesare present in a subset of histologically normal surgicalresection margins. We performed gene expression pro-filing of both resection margins and tumors with thepurpose of (1) identifying over-expressed genes intumors as potential markers of recurrence in histologi-cally normal margins, and (2) finding a subset of thosegenes predictive of recurrence. In order to generate avery high-confidence gene set, we augmented the analy-sis of our data with a meta-analysis of five publishedmicroarray studies [24-28] to reliably identify a set ofgenes significantly deregulated in OSCC compared tonormal oral tissues. These five public data sets wereselected based on the availability of raw microarray data,as well as for the inclusion of both oral cavity tumorsand either adjacent normal tissues or oral tissues fromhealthy individuals. Although data from Pyeon et al. [28]included HPV positive and HPV negative head and neckcarcinomas from different anatomic sites, we selectedoral carcinomas only, which are mainly negative forHPV infection, as shown in a recent study performed byour group [29]. This meta-analysis sample set was com-posed of a total of 199 samples (141 OSCCs, 38 adjacentnormal tissues and 20 healthy normal tissues) from 141oral cancer patients and 20 healthy individuals (withoutcancer) (Table 3). We pre-processed data from the dif-ferent array platforms with updated chip definition files,as described above, to correct outdated probe mappinginformation from older platforms. We used a Rank Pro-duct analysis for the public studies, which consideredonly the ranking of genes by differential expressionbetween pairs of samples within studies [20,30], avoidingbatch and platform-related effects which would occurfrom directly combining expression values from the dif-ferent studies. Genes were selected with evidence of up-regulation in tumors with a False Discovery Rate (FDR)of 0.01 and fold-change ≥ 2. We chose to focus onover-expressed genes only, since histologically normalmargins may contain only a fraction of geneticallyaltered cells, and the presence of genetically normalcells would likely make down-regulated genes unreliablemarkers. By using the intersection of genes identifiedboth by meta-analysis and the in-house array trainingset, we retained only genes that were reproducibly over-expressed compared to normal oral tissues from healthypatients and histologically normal margins. These strictselection criteria for gene signature candidates, based onprior hypothesis, helped to reduce the risk of over-fittingduring Cox regression analysis.

Data analysis of in-house microarray studyMicroarray results from our study were normalized asdescribed above, along with normal oral tissue samplesfrom healthy individuals also assayed by Affymetrix

Table 1 Clinicopathological data, recurrence andoutcome data from 24 OSCC patients (N = 96 samples,training set)

Variables N (%)

Age (years)

Median (range) 58.5 (37-83)

Gender

Male 15 (62.5)

Female 9 (37.5)

Tobacco use

Yes 16 (67)

No 8 (33)

Alcohol use

Yes 17 (71)

No 7 (29)

Tumor site

Tongue 16 (67)

Floor of mouth 4 (17)

Buccal mucosa 2 (8)

Alveolar 2 (8)

T category

T1-T2 12 (50)

T3-T4 12 (50)

Nodal status

Negative (N0) 12 (50)

Positive (N1, N2b, N2c) 12 (50)

Tumor stage

I, II 7 (29)

III, IV 17 (71)

Tumor grade

Moderately differentiated 18 (75)

Poorly differentiated 6 (25)

Recurrence

Yes 9 (37.5)

No 15 (62.5)

Time to recurrence

Median (range) 32 (1.8-34)

Follow-up

Median (range) 13 (1.7-58.8)

Outcome

Alive with no evidence of disease 13 (55)

Alive with disease 2 (8)

Dead of disease 7 (29)

Dead of other causes 2 (8)


Page 3 of 11

Human Genome U133 Plus 2.0 oligonucleotide arrays(downloaded from GEO accession number GSE6791).Probesets with low expression (75th percentile belowlog2(100)) or low variance (IQR on log2 scale < 0.5)were filtered [31] as well as the quality control probesets. The treat function from LIMMA: Linear Modelsfor Microarray Analysis (version 3.2.1) [32] was used toidentify genes ≥ 2-fold up-regulated in tumors comparedto margins from our study, with FDR = 0.01. We tookthe intersection of genes identified in both the meta-analysis and in-house microarray experiment, and thisintersection analysis produced a list of 138 genes, whichwere used as the feature set for developing the gene sig-nature for recurrence.

Penalized Cox regressionLASSO penalized Cox regression was used to train apredictive risk score from the maximum expression ofthe 138 genes in any margin of each patient (all these138 genes being found to be over-expressed inOSCC). Normalized expression values were first con-verted to Z-scores for each gene, to ensure equaleffect of the penalty on all coefficients, as is standardfor penalized regression. We applied LASSO penalizedCox regression as implemented in the penalized Rpackage (version 0.9-27) [33], to condition a linearrisk score with time to recurrence as the event ofinterest. In keeping with the hypothesis that over-expression of these genes is predictive of recurrence,we constrained the regression coefficients to positivevalues, and selected the L1 penalty parameter by opti-mizing 10-fold cross-validated likelihood. Since thecross-validated likelihood depends on the randomizedfolding of the data, we repeated the regression 50times and selected the model with the highest cross-validated likelihood.

ResultsPatient characteristics: OSCC recurrenceAs shown in Table 1 and Table 2, 9/24 patients (train-ing set) and 7/30 patients (independent validation set)had disease recurrence. Median time to local recurrence(by Kaplan Meier estimate) of patients in the trainingset was 32 months (range 2-34 months). Similarly,patients from the validation set recurred within 2-36months. All patients had local recurrence, and somepatients also had regional and/or distant failure (detaileddata in Tables 1 and 2). Median (by reverse Kaplan-Meier estimate) and range of follow-up times of patientswere 20 months (1.4-57 months) in the training set and23 months (1-81 months) in the validation set.

Table 2 Clinicopathological data, recurrence data andoutcome data from 30 OSCC patients (N = 136 samples,validation set)

Variables N (%)

Age (years)

Median (range) 67 (48-81)

Gender

Male 21 (70)

Female 9 (30)

Tobacco use

Yes 25 (83)

No 5 (17)

Alcohol use

Yes 21 (70)

No 9 (30)

Tumor site

Tongue 18 (50)

Floor of mouth 8 (27)

Tongue + Floor of mouth 3 (10)

Buccal mucosa 2 (7)

Alveolar 1 (3)

Retromolar 1 (3)

T category

T1-T2 14 (47)

T3-T4 16 (53)

Nodal status

Negative (N0) 24 (80)

Positive (N1, N2b, N2c) 6 (20)

Tumor stage

I, II 13 (43)

III, IV 17 (57)

Tumor grade

Moderately differentiated 23 (77)

Poorly differentiated 7 (23)

Recurrence

Yes 7 (23)

No 23 (77)

Time to recurrence


Follow-up


Outcome

Alive with no evidence of disease 23 (77)

Alive with disease 4 (13)

Dead of disease 2 (7)

Dead of other causes 1 (3)


Page 4 of 11

Differentially expressed genes in margins, OSCC andnormal oral tissuesMeta-analysis of the five public data sets (Table 3) iden-tified 667 up-regulated genes in OSCC; results are pro-vided in Additional file 2, Table S1. Requiring two-foldup-regulation in tumors in both the meta-analysis ofpublic datasets and our microarray experiment, withFDR < 0.01, identified 138 over-expressed genes (Addi-tional file 3, Table S2). The expression patterns of thesegenes in tumors, margins, and normal oral tissue sam-ples from healthy individuals are shown as a heatmap inFigure 1. As seen in the hierarchical clustering, the 138genes accurately discriminate between the tumors andmargins of the in-house study, as well as normal oraltissues from a publicly available study (as described inMethods). We performed LASSO penalized regressionof these 138 genes and selected a total of six genes. Ofthese six genes, we retained four genes with the largestcoefficients (MMP1, COL4A1, P4HA2 and THBS2), andeliminated two genes with very small coefficients (PXDNand PMEPA1), which together contributed less than 5%of the magnitude of the risk score. These four genesconstituted a 4-gene signature of recurrence. In theheatmap, Gene cluster “B” contains three of the fourgenes in the signature (COL4A1, P4HA2 and THBS2),and it shows frequent up-regulation in the surgical mar-gins compared to the normal oral tissues. Strikingly,MMP1, found in gene cluster “A”, shows less frequentover-expression in the margins, but has extreme differ-ential expression between margins and OSCCs (400-foldup-regulation as detected by microarrays, and 800-foldup-regulation, validated by RQ-PCR). The proteinsencoded by the 138 genes are also shown in a proteininteraction network that highlights the most highlyinter-connected proteins (Figure 2 and Additional file 1,Methods S1). In the heatmap, the main features of clus-ters A and B are the large number of interacting MMPproteins in cluster A, which contains MMP1, and col-lagens plus TGFB1 in cluster B, which also containsP4HA2, THBS2 and COL4A1 genes of the signature.

The large number of MMPs and collagen proteins areclosely connected; in particular, MMP9 interacts withboth THBS2 and COL4A1, and indirectly with MMP1.Results of Gene Ontology (GO) enrichment analysis ofall 138 genes are presented in Additional file 4, TableS3. Graphical representations of GO annotations areshown in Additional file 5, GO biological processes,Additional file 6, GO cellular component and Additionalfile 7, GO molecular function.

Over-expressed genes in surgical margins: prognosticvalue for recurrenceWe examined these 138 genes (see above) for univariateassociation with recurrence in the microarray trainingdata, which, since recurrence was not used in the selec-tion of these genes, constitutes the first validation of ourstudy hypothesis. We found significant enrichment forunivariate association with recurrence in these 138genes (see Additional file 8, Figure S1 for comparison ofthe distribution of nominal p-values with a null distribu-tion empirically determined by permutation of the out-come labels).

Four-gene prognostic signature for OSCC recurrenceWe performed quantitative PCR validation of the 4-genesignature (MMP1, COL4A1, P4HA2 and THBS2) in aseparate patient cohort (Table 2). This validation analy-sis confirmed that all four genes were up-regulated inmargins and OSCC samples from patients with diseaserecurrence compared to margins and OSCCs frompatients who did not recur (Figure 3A). The risk scorewas calculated using the maximum expression in anymargin of each patient, and was dichotomized at themedian score. These risk groups were predictive ofrecurrence in an independent test cohort assayed byRQ-PCR (136 samples; N = 30 patients) (HR = 6.8, p =0.04, logrank test) (Figure 3B). In multivariate Cox ana-lysis, correction for T (tumor size) and N (nodal status)resulted only in a slight increase in significance level ofthe four-gene risk score (p = 0.06, likelihood ratio test).

Table 3 Description of the five publicly available data sets used for meta-analysis

Public Data Set Reference andAccession ID

Sample Size Array Platform

Toruner et al. 2004GDS1584

16 oral carcinoma and 4 adjacent normal tissues from 16 patients Human Genome U133A Plus 2.0(Affymetrix)

Ye et al. 2008GSE9844

26 oral carcinoma samples and 12 adjacent normal tissues from 26 patients Human Genome U133A Plus 2.0(Affymetrix)

Kuriakose et al. 2004GDS2520

22 head and neck carcinomas and 22 adjacent normal tissues from 22patients

Human Genome U95A(Affymetrix)

Sticht et al. 2008GSE10121

35 oral carcinoma samples from 35 patients and 6 normal oral tissues fromhealthy individuals

Human Oligo Set 4.0 (Operon)

Pyeon et al., 2007GSE6791

42 head and neck carcinoma samples from 42 patients and 14 normal oraltissues from healthy individuals

Human Genome U133A Plus 2.0(Affymetrix)


Page 5 of 11

Clinical variables, alone or in combination, were notpredictive of recurrence in either training or validationcohorts. The coefficients of the 4-gene risk score, foruse with Z-score scaled expression values, are summar-ized in Table 4.In addition, we performed a survival Receiver Operat-

ing Characteristic (ROC) analysis to demonstrate thesensitivity and specificity of the model in the test set.This analysis showed the effect of altering the cutoffthreshold between high and low-risk patients. Ourresults showed that the use of a very high threshold todefine high-risk patients predicts a majority of recur-rences at a low false-positive rate (20%). The area underthe ROC curve (AUC) for recurrence within 36 monthswas 0.73, an improvement over the expected AUC of0.5 for unpredictive models (Additional file 9, FigureS2). While we maintained the standard median cutoff

due to the limited sample size, a larger study in thefuture may be able to further fine-tune the cutoffthreshold to optimize sensitivity and specificity in thecontext of the relative risks that treatment optionsinformed by this prognostic score entail.

Effect of reducing the number of available marginsSimulating the selection of only a single margin fromeach patient, the 4-gene signature maintained a predic-tive effect in both the training and validation sets (med-ian HR = 2.2 in the training set and 1.8 in the validationset, with 82% and 99% of bootstrapped hazard ratiosgreater than the no-effect value of HR = 1) (Additionalfile 10, Figure S3). Results from the bootstrap simula-tions showed smaller hazard ratios, compared withhazard ratios obtained when using the maximumexpression value from several margins. This difference

Figure 1 Heatmap of 138 genes up-regulated in OSCC. Expression values for each row (gene) are scaled to z-scores for visualization. Marginsand tumors annotated with darker colors above the heatmap are from patients who experienced recurrence.


Page 6 of 11

could not be explained by association between the num-ber of margins and recurrence in the training set (HR =1.17, p = 0.69 logrank test) or the validation set (HR =1.30, p = 0.24 logrank test). However, a plausible expla-nation for this difference is that increasing the numberof observed margins increases the probability of obser-ving genetic alterations spreading asymmetrically fromthe tumor, since genetic alterations are not seen in allsurgical margins.

DiscussionIt is known that histologically normal margins may har-bor genetic changes also found in the primary tumor, asshown by studies in HNSCC, including oral carcinomas[7]. In oral carcinoma, local recurrence may arise fromcancer cells left behind after surgery, undetectable by

histopathology (minimal residual cancer), or from fieldsof genetically altered cells with the potential to give riseto a new carcinoma [34]; such fields precede the tumorand can be detected in the surrounding mucosa (surgi-cal resection margins). Molecular changes that are com-monly detected in margins as well as the correspondingtumor could indicate that pre-malignant or malignantclones were able to migrate to the surrounding tissue,giving rise to a primary tumor recurrence [35].Herein, we show that global gene expression analysis

of histologically normal margins and OSCC is a valuableapproach for the identification of deregulated genes andpathways associated with OSCC recurrence. We used amulti-step procedure including our in-house whole-gen-ome expression profiling experiment and a meta-analy-sis of five published microarray datasets to develop a 4-

Figure 2 Protein-protein interaction network of 138 genes. I2D version 1.72 was used to identify protein interactions for the 138 genesshown in the heatmap. The resulting network was visualized using NAViGaTOR 2.1.14 http://ophid.utoronto.ca/navigator. Color of nodescorresponds to Gene Ontology biological function, as described in the legend. Red-highlighted squares represent the four genes in the signatureof OSCC recurrence.


Page 7 of 11

http://ophid.utoronto.ca/navigator

gene signature (MMP1, COL4A1, THBS2 and P4HA2)for prediction of recurrence in OSCC. This signature isbased on genes found to be consistently over-expressedin OSCC as compared to normal oral mucosa; thesegenes are also over-expressed in a subset of histologi-cally normal surgical resection margins, and their over-expression in such margins provides an indication of thepresence of genetic changes before histological altera-tions can be detected by histology. Notably, our initialanalyses reveal that this 4-gene signature predictedrecurrence in two of the patients (Pts. 17 and 20, Table2, validation set) who had not recurred until the latestupdate of the clinical data for recurrence status. Both of

these patients had local recurrence, 8 and 19 monthsafter surgery, respectively.Genes identified in the 4-gene signature (MMP1,

COL4A1, THBS2 and P4HA2) play major roles in cell-cell and/or cell-matrix interaction, and invasion. Thedirect and indirect partners of these genes are illustratedin Figure 2. The functions of two genes (P4HA2 andTHBS2) in our signature of OSCC recurrence and theirroles in cancer are not well understood. P4HA2 encodesa key enzyme involved in collagen synthesis, and itsover-expression has been previously reported in papil-lary thyroid cancer [36]. THBS2 is a matricellular pro-tein that encodes an adhesive glycoprotein and interactswith other proteins to modulate cell-matrix interactions[37]. Interestingly, THBS2 is associated with tumorgrowth in adult mouse tissues [37]. The two other genesin our OSCC recurrence signature (COL4A1 andMMP1) are better characterized in cancer. COL4A1encodes the major type IV alpha collagen chain and isone of the main components of basement membranes.Basement membranes have several important biologicalroles, and are essential for embryonic development,proper tissue architecture, and tissue remodeling [38].COL4A1 binds other collagens (COL4A2, 3, 4, 5 and 6),as well as LAMC2 (laminin, gamma 2), TGFB1

(B)(A)

Figure 3 Heatmap of validation data and Kaplan-Meier plot of disease recurrence. (A) Unsupervised hierarchical clustering of thequantitative real-time PCR (validation data) showing the maximum expression levels of MMP1, P4HA2, THBS2 and COL4A1 in margins frompatients with and without recurrence and with a follow-up time ≥ 12 months. Margins annotated with darker green above the heatmap arefrom patients who experienced recurrence. Margins from patients with locally recurrent tumors show increased expression levels of the four-gene signature compared to patients who did not recur. (B) Kaplan-Meier plot of quantitative real-time PCR data for patients in the validationset. Patients are assigned to high or low risk based on their four-gene signature risk score. As seen in the Kaplan-Meier plot, patients with over-expression of the 4-gene signature are at high risk for disease recurrence; all patients who experienced recurrence in the validation set are in thehigh risk group, suggesting that over-expression of this signature was highly predictive of recurrence in the validation set.

Table 4 Coefficients of the linear risk score for z-scorenormalized log2-expression values.

Gene Coefficient FC (microarray) FC (RQ-PCR) p-value (qPCR)

MMP1 0.63 405 798 9E-16

COL4A1 0.25 3.7 4.3 7E-09

P4HA2 0.45 2.7 2.8 1E-06

THBS2 0.34 3 1.9 6E-03

Fold-change (FC) is the geometric-average expression in tumors relative tosurgical resection margins. P-values are for tumor/margin differentialexpression in the qPCR (independent validation set) (Wilcoxon Rank Sum test)


Page 8 of 11

(transforming growth factor, beta 1), among other pro-teins (Figure 2) http://www.ihop-net.org, playing a rele-vant role in extracellular matrix-receptor interaction andfocal adhesion [39]. The extracellular matrix undergoesconstant remodeling; during this process, proteins suchas MMP1 can degrade the extracellular matrix proteins(e.g., collagen IV), and contribute to invasion and metas-tasis [40]. In cancer, combined over-expression ofCOL4A1 and LAMC2 can distinguish OSCC from clini-cally normal oral cavity/oropharynx tissues [41]; this lat-ter study suggests that COL4A1 over-expression may bea useful biomarker for early detection of malignancy.MMP1 belongs to the family of matrix metallopro-

teases, which are key proteases involved in extracellularmatrix (ECM) degradation [42]. MMP1 encodes a col-lagenase, which is secreted by tumor cells as well as bystromal cells stimulated by the tumor; this secretedenzyme is responsible for breaking down interstitial col-lagens type I, II and III in normal physiological pro-cesses (e.g., tissue remodeling) as well as diseaseprocesses (e.g., cancer) [42]. It is believed that themechanism of up-regulation of most of the MMPs islikely due to transcriptional changes, which may occurfollowing alterations in oncogenes and/or tumor sup-pressor genes [42]. Indeed, MMP1 has been previouslyreported as consistently over-expressed in oral carci-noma compared to adjacent normal tissues [43-45] andsuggested as a biomarker of malignant transformation inprecursor lesions in oral cancer [44,46,47]. In HNSCC,over-expression of several genes with roles in invasionand metastasis, including MMPs, were previously asso-ciated with treatment failure of HNSCC [48]. In ourstudy, MMP1 was over-expressed in a subset of marginsexclusively from patients with recurrent OSCC, andshowed the highest fold-change of up-regulation inOSCC compared to margins. Our results are in agree-ment with the literature findings of MMP1 over-expres-sion associated with progression of pre-malignant orallesions; in this context, our data showing MMP1 over-expression in histologically normal surgical marginsfrom patients who developed local recurrence supportthe notion that MMP1 may be involved in initial stepsof cellular transformation and tumorigenesis, as well asinvasion of oral carcinoma cells. Indeed, matrix metallo-proteinases play an important role not only in invasionand metastasis but also in early stages of cancer devel-opment/progression [42].Our data suggest that histologically normal surgical

resection margins that over-express MMP1, COL4A1,THBS2 and P4HA2 are indicative of an increased risk ofrecurrence in OSCC. Patients at higher risk of recur-rence could potentially benefit from closer disease moni-toring and/or adjuvant post-operative radiationtreatment, even in the absence of other clinical and

histopathological indicators, such as advanced diseasestage and perineural invasion.

ConclusionConsidering that our 4-gene signature showed prognos-tic value for recurrence in two independent patientcohorts, over-expression of this signature may be usedfor molecular analysis of histologically negative marginsin oral carcinoma.

Additional material

Additional file 1: Methods S1. Description of methods used for RNAisolation, oligonucleotide array experiments, quantitative real-timereverse-transcription PCR validation and protein-protein interactionnetwork analysis.

Additional file 2: Table S1. Results of meta-analysis of the five publicdata sets identified 667 up-regulated genes in OSCC.

Additional file 3: Table S2. List of 138 over-expressed genes, with FDR< 0.01, identified in both the meta-analysis of public datasets and ourmicroarray experiment.

Additional file 4: Table S3. Results of Gene Ontology enrichmentanalysis of the 138 genes identified as over-expressed in both the publicdatasets and our microarray experiment.

Additional file 5: GO biological function. Graphical representation ofGO annotation (biological function).

Additional file 6: GO cellular component. Graphical representation ofGO annotation (cellular component).

Additional file 7: GO molecular function. Graphical representation ofGO annotation (molecular function).

Additional file 8: Figure S1. Distribution of nominal p-values forunivariate association of the 138 genes identified as over-expressed inOSCC with recurrence. P-values were determined by Cox regressionusing the maximum expression in any margin of each patient. Theempirical null distribution was determined from association of thesesame genes with 1,000 permutations of the outcome labels. Theobserved nominal p-values are significantly enriched for small values (p= 0.001, Kolmogorov-Smirnov test). It is worth noting that recurrencewas not used at any stage in the selection of these genes.

Additional file 9: Figure S2. Survival Receiver Operating Characteristiccurve for recurrence at 36 months in the test set. Using a high thresholdto define high-risk patients predicts a majority of recurrences (truepositives) at a low false-positive rate (20%). While we maintained thestandard median cutoff for this study due to the limited sample size, alarger study in the future may be able to further tune the cutoffthreshold to optimize sensitivity and specificity in the context of therelative risks that treatment options informed by this prognostic scoreentail. The area under the ROC curve (AUC) for recurrence within 36months is 0.73, which is an improvement over the expected AUC of 0.5for non-predictive risk scores.

Additional file 10: Figure S3. Bootstrap validation of four-genesignature risk score in training and validation sets. Density lines representthe distribution of hazard ratios observed in 1,000 re-samplings of asingle margin, randomly chosen, from each patient.

Acknowledgements and fundingThe authors acknowledge Dr. Jerry Machado and Dr. Mahadeo Sukhai forconstructive criticism of this manuscript. This work was supported by theOntario Institute for Cancer Research (OICR) (SKR, JI, PG, IJ, and BPO) and theGalloway Fund (RG, SKR), the Canada Research Chair Program (IJ) and theCIHR Catalyst Grant #202370 (IJ). Computational facilities were supported bythe Canada Research Chair Program, Canada Foundation for Innovation


Page 9 of 11

http://www.ihop-net.orghttp://www.biomedcentral.com/content/supplementary/1471-2407-11-437-S1.DOChttp://www.biomedcentral.com/content/supplementary/1471-2407-11-437-S2.XLShttp://www.biomedcentral.com/content/supplementary/1471-2407-11-437-S3.XLShttp://www.biomedcentral.com/content/supplementary/1471-2407-11-437-S4.XLShttp://www.biomedcentral.com/content/supplementary/1471-2407-11-437-S5.PDFhttp://www.biomedcentral.com/content/supplementary/1471-2407-11-437-S6.PDFhttp://www.biomedcentral.com/content/supplementary/1471-2407-11-437-S7.PDFhttp://www.biomedcentral.com/content/supplementary/1471-2407-11-437-S8.PNGhttp://www.biomedcentral.com/content/supplementary/1471-2407-11-437-S9.PNGhttp://www.biomedcentral.com/content/supplementary/1471-2407-11-437-S10.PNG

(Grant #12301 and #203383), Ontario Research Fund (GL2-01-030), and IBM.This research was funded in part by the Ontario Ministry of Health and LongTerm Care (MOHLTC). The views expressed do not necessarily reflect thoseof the MOHLTC.

Author details1Div. of Applied Molecular Oncology, Princess Margaret Hospital, OntarioCancer Institute, University Health Network, Toronto, ON, Canada. 2Dept. ofSurgery and Orthopedics, Faculty of Medicine, São Paulo State University -UNESP, Botucatu, SP, Brazil. 3Ontario Cancer Institute and the CampbellFamily Institute for Cancer Research, Toronto, ON, Canada. 4Dept. ofPathology, Toronto General Hospital, Ontario Cancer Institute, UniversityHealth Network, Toronto, ON, Canada. 5Dept. of Biostatistics, PrincessMargaret Hospital, Ontario Cancer Institute, University Health Network,Toronto, ON, Canada. 6Dalla Lana School of Public Health Sciences,University of Toronto, Toronto, ON, Canada. 7Dept. of Otolaryngology,Hospital Calderon Guardia, San Jose, Costa Rica. 8Dept. of Otolaryngology/Head and Neck Surgery, Worcester Royal Hospital, Worcester, UK. 9Dept. ofOtolaryngology/Head and Neck Surgery, Helsinki University Central Hospital,University of Helsinki, Helsinki, Finland. 10Dept. of Otolaryngology/SurgicalOncology, Princess Margaret Hospital, Ontario Cancer Institute, UniversityHealth Network, Toronto, ON, Canada. 11Dept. of Computer Science,University of Toronto, Toronto, ON, Canada. 12Dept. of Laboratory Medicineand Pathobiology, University of Toronto, ON, Canada. 13Dept. of MedicalBiophysics, University of Toronto, Toronto, ON, Canada.

Authors’ contributionsPPR, BPO, IJ and SKR designed the study. LW, IJ and MP performedbioinformatics and statistical data analyses. PPR, LW, IJ and SKR drafted themanuscript. YX and NKC helped with quantitative PCR experiments and datainterpretation. NNG, GCW and AAM helped with study design, identifiedpatients and collected samples. NNG and AAM helped draft the manuscript.NNG, CS, DG, DB, RG, PG and JI helped identify patients, collected samplesand provided detailed clinical data of patients. All authors read andapproved the final manuscript.

Competing interestsThe authors declare that they have no competing interests.

Received: 17 January 2011 Accepted: 11 October 2011Published: 11 October 2011

References1. Parkin DM, Pisani P, Ferlay J: Global cancer statistics. CA Cancer J Clin 1999,

49(1):33-64, 31.2. Sawair FA, Irwin CR, Gordon DJ, Leonard AG, Stephenson M, Napier SS:

Invasive front grading: reliability and usefulness in the management oforal squamous cell carcinoma. J Oral Pathol Med 2003, 32(1):1-9.

3. Jones AS, Bin Hanafi Z, Nadapalan V, Roland NJ, Kinsella A, Helliwell TR: Dopositive resection margins after ablative surgery for head and neckcancer adversely affect prognosis? A study of 352 patients withrecurrent carcinoma following radiotherapy treated by salvage surgery.Br J Cancer 1996, 74(1):128-132.

4. Leemans CR, Tiwari R, Nauta JJ, van der Waal I, Snow GB: Recurrence atthe primary site in head and neck cancer and the significance of necklymph node metastases as a prognostic factor. Cancer 1994,73(1):187-190.

5. Brandwein-Gensler M, Teixeira MS, Lewis CM, Lee B, Rolnitzky L, Hille JJ,Genden E, Urken ML, Wang BY: Oral squamous cell carcinoma: histologicrisk assessment, but not margin status, is strongly predictive of localdisease-free and overall survival. Am J Surg Pathol 2005, 29(2):167-178.

6. Nathan CA, Amirghahri N, Rice C, Abreo FW, Shi R, Stucker FJ: Molecularanalysis of surgical margins in head and neck squamous cell carcinomapatients. Laryngoscope 2002, 112(12):2129-2140.

7. Bilde A, von Buchwald C, Dabelsteen E, Therkildsen MH, Dabelsteen S:Molecular markers in the surgical margin of oral carcinomas. J OralPathol Med 2009, 38(1):72-78.

8. Nathan CA, Liu L, Li BD, Abreo FW, Nandy I, De Benedetti A: Detection ofthe proto-oncogene eIF4E in surgical margins may predict recurrence inhead and neck cancer. Oncogene 1997, 15(5):579-584.

9. Nathan CA, Franklin S, Abreo FW, Nassar R, De Benedetti A, Glass J: Analysisof surgical margins with the molecular marker eIF4E: a prognostic factorin patients with head and neck cancer. J Clin Oncol 1999, 17(9):2909-2914.

10. Tan HK, Saulnier P, Auperin A, Lacroix L, Casiraghi O, Janot F, Fouret P,Temam S: Quantitative methylation analyses of resection marginspredict local recurrences and disease-specific deaths in patients withhead and neck squamous cell carcinomas. Br J Cancer 2008,99(2):357-363.

11. van der Toorn PP, Veltman JA, Bot FJ, de Jong JM, Manni JJ, Ramaekers FC,Hopman AH: Mapping of resection margins of oral cancer for p53overexpression and chromosome instability to detect residual (pre)malignant cells. J Pathol 2001, 193(1):66-72.

12. van Houten VM, Leemans CR, Kummer JA, Dijkstra J, Kuik DJ, van denBrekel MW, Snow GB, Brakenhoff RH: Molecular diagnosis of surgicalmargins and local recurrence in head and neck cancer patients: aprospective study. Clin Cancer Res 2004, 10(11):3614-3620.

13. Goldenberg D, Harden S, Masayesva BG, Ha P, Benoit N, Westra WH,Koch WM, Sidransky D, Califano JA: Intraoperative molecular marginanalysis in head and neck cancer. Arch Otolaryngol Head Neck Surg 2004,130(1):39-44.

14. Franklin S, Pho T, Abreo FW, Nassar R, De Benedetti A, Stucker FJ,Nathan CO: Detection of the proto-oncogene eIF4E in larynx andhypopharynx cancers. Arch Otolaryngol Head Neck Surg 1999,125(2):177-182.

15. Taioli E, Ragin C, Wang XH, Chen J, Langevin SM, Brown AR, Gollin SM,Garte S, Sobol RW: Recurrence in oral and pharyngeal cancer isassociated with quantitative MGMT promoter methylation. BMC Cancer2009, 9:354.

16. van Houten VM, Tabor MP, van den Brekel MW, Kummer JA, Denkers F,Dijkstra J, Leemans R, van der Waal I, Snow GB, Brakenhoff RH: Mutatedp53 as a molecular marker for the diagnosis of head and neck cancer. JPathol 2002, 198(4):476-486.

17. Wu Z, Irizarry RA, Gentleman R, Martinez-Murillo F, Spencer F: A Model-Based Background Adjustment for Oligonucleotide Expression Arrays.Journal of the American Statistical Association 2004, 99(468):909-917.

18. Dai M, Wang P, Boyd AD, Kostov G, Athey B, Jones EG, Bunney WE,Myers RM, Speed TP, Akil H, et al: Evolving gene/transcript definitionssignificantly alter the interpretation of GeneChip data. Nucleic Acids Res2005, 33(20):e175.

19. Gautier L, Cope L, Bolstad BM, Irizarry RA: affy–analysis of AffymetrixGeneChip data at the probe level. Bioinformatics 2004, 20(3):307-315.

20. Hong F, Breitling R, McEntee CW, Wittner BS, Nemhauser JL, Chory J:RankProd: a bioconductor package for detecting differentially expressedgenes in meta-analysis. Bioinformatics 2006, 22(22):2825-2827.

21. Falcon S, Gentleman R: Using GOstats to test gene lists for GO termassociation. Bioinformatics 2007, 23(2):257-258.

22. Zheng Q, Wang XJ: GOEAST: a web-based software toolkit for GeneOntology enrichment analysis. Nucleic Acids Res 2008, , 36 Web Server:W358-363.

23. Edgar R, Domrachev M, Lash AE: Gene Expression Omnibus: NCBI geneexpression and hybridization array data repository. Nucleic Acids Res 2002,30(1):207-210.

24. Toruner GA, Ulger C, Alkan M, Galante AT, Rinaggio J, Wilk R, Tian B,Soteropoulos P, Hameed MR, Schwalb MN, et al: Association betweengene expression profile and tumor invasion in oral squamous cellcarcinoma. Cancer Genet Cytogenet 2004, 154(1):27-35.

25. Ye H, Yu T, Temam S, Ziober BL, Wang J, Schwartz JL, Mao L, Wong DT,Zhou X: Transcriptomic dissection of tongue squamous cell carcinoma.BMC Genomics 2008, 9:69.

26. Kuriakose MA, Chen WT, He ZM, Sikora AG, Zhang P, Zhang ZY, Qiu WL,Hsu DF, McMunn-Coffran C, Brown SM, et al: Selection and validation ofdifferentially expressed genes in head and neck cancer. Cell Mol Life Sci2004, 61(11):1372-1383.

27. Sticht C, Freier K, Knopfle K, Flechtenmacher C, Pungs S, Hofele C, Hahn M,Joos S, Lichter P: Activation of MAP kinase signaling through ERK5 butnot ERK1 expression is associated with lymph node metastases in oralsquamous cell carcinoma (OSCC). Neoplasia 2008, 10(5):462-470.

28. Pyeon D, Newton MA, Lambert PF, den Boon JA, Sengupta S, Marsit CJ,Woodworth CD, Connor JP, Haugen TH, Smith EM, et al: Fundamentaldifferences in cell cycle deregulation in human papillomavirus-positive


Page 10 of 11

http://www.ncbi.nlm.nih.gov/pubmed/10200776?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/12558952?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/12558952?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/8679446?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/8679446?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/8679446?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/8679446?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/8275423?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/8275423?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/8275423?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/15644773?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/15644773?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/15644773?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/12461330?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/12461330?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/12461330?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/19192052?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/9247311?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/9247311?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/9247311?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/10561370?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/10561370?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/10561370?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/18594522?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/18594522?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/18594522?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/11169517?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/11169517?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/11169517?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/15173067?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/15173067?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/15173067?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/14732766?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/14732766?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/10037284?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/10037284?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/19807915?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/19807915?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/12434417?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/12434417?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/16284200?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/16284200?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/14960456?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/14960456?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/16982708?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/16982708?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/17098774?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/17098774?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/11752295?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/11752295?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/15381369?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/15381369?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/15381369?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/18254958?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/15170515?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/15170515?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/18472963?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/18472963?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/18472963?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/17510386?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/17510386?dopt=Abstract

and human papillomavirus-negative head/neck and cervical cancers.Cancer Res 2007, 67(10):4605-4619.

29. Machado J, Reis PP, Zhang T, Simpson C, Xu W, Perez-Ordonez B,Goldstein DP, Brown DH, Gilbert RW, Gullane PJ, et al: Low prevalence ofhuman papillomavirus in oral cavity carcinomas. Head Neck Oncol 2010,2:6.

30. Breitling R, Herzyk P: Rank-based methods as a non-parametricalternative of the T-statistic for the analysis of biological microarraydata. J Bioinform Comput Biol 2005, 3(5):1171-1189.

31. Bioinformatics and computational biology solutions using R andBioconductor. [http://www.worldcat.org/isbn/0387251464].

32. McCarthy DJ, Smyth GK: Testing significance relative to a fold-changethreshold is a TREAT. Bioinformatics 2009, 25(6):765-771.

33. Goeman JJ: Penalized estimation in the Cox proportional hazards model.Biometrical journal Biometrische Zeitschrift 2010, 52(1):70-84.

34. Tabor MP, Brakenhoff RH, van Houten VM, Kummer JA, Snel MH, Snijders PJ,Snow GB, Leemans CR, Braakhuis BJ: Persistence of genetically alteredfields in head and neck cancer patients: biological and clinicalimplications. Clin Cancer Res 2001, 7(6):1523-1532.

35. Ha PK, Califano JA: The molecular biology of mucosal field cancerizationof the head and neck. Crit Rev Oral Biol Med 2003, 14(5):363-369.

36. Jarzab B, Wiench M, Fujarewicz K, Simek K, Jarzab M, Oczko-Wojciechowska M, Wloch J, Czarniecka A, Chmielik E, Lange D, et al: Geneexpression profile of papillary thyroid cancer: sources of variability anddiagnostic implications. Cancer Res 2005, 65(4):1587-1597.

37. Bornstein P, Kyriakides TR, Yang Z, Armstrong LC, Birk DE: Thrombospondin2 modulates collagen fibrillogenesis and angiogenesis. J InvestigDermatol Symp Proc 2000, 5(1):61-66.

38. Sado Y, Kagawa M, Naito I, Ueki Y, Seki T, Momota R, Oohashi T,Ninomiya Y: Organization and expression of basement membranecollagen IV genes and their roles in human disorders. J Biochem 1998,123(5):767-776.

39. Hoffmann R, Valencia A: A gene network for navigating the literature. NatGenet 2004, 36(7):664.

40. Tanzer ML: Current concepts of extracellular matrix. J Orthop Sci 2006,11(3):326-331.

41. Chen C, Mendez E, Houck J, Fan W, Lohavanichbutr P, Doody D, Yueh B,Futran ND, Upton M, Farwell DG, et al: Gene expression profiling identifiesgenes predictive of oral squamous cell carcinoma. Cancer EpidemiolBiomarkers Prev 2008, 17(8):2152-2162.

42. Egeblad M, Werb Z: New functions for the matrix metalloproteinases incancer progression. Nat Rev Cancer 2002, 2(3):161-174.

43. Suhr ML, Dysvik B, Bruland O, Warnakulasuriya S, Amaratunga AN,Jonassen I, Vasstrand EN, Ibrahim SO: Gene expression profile of oralsquamous cell carcinomas from Sri Lankan betel quid users. Oncol Rep2007, 18(5):1061-1075.

44. Yen CY, Chen CH, Chang CH, Tseng HF, Liu SY, Chuang LY, Wen CH,Chang HW: Matrix metalloproteinases (MMP) 1 and MMP10 but notMMP12 are potential oral cancer markers. Biomarkers 2009, 14(4):244-249.

45. Saleh A, Zain RB, Hussaini H, Ng F, Tanavde V, Hamid S, Chow AT, Lim GS,Abraham MT, Teo SH, et al: Transcriptional profiling of oral squamous cellcarcinoma using formalin-fixed paraffin-embedded samples. Oral Oncol2010, 46(5):379-386.

46. Jordan RC, Macabeo-Ong M, Shiboski CH, Dekker N, Ginzinger DG,Wong DT, Schmidt BL: Overexpression of matrix metalloproteinase-1 and-9 mRNA is associated with progression of oral dysplasia to cancer. ClinCancer Res 2004, 10(19):6460-6465.

47. Estilo CL, P Oc, Talbot S, Socci ND, Carlson DL, Ghossein R, Williams T,Yonekawa Y, Ramanathan Y, Boyle JO, et al: Oral tongue cancer geneexpression profiling: Identification of novel potential prognosticators byoligonucleotide microarray analysis. BMC Cancer 2009, 9:11.

48. Ginos MA, Page GP, Michalowicz BS, Patel KJ, Volker SE, Pambuccian SE,Ondrey FG, Adams GL, Gaffney PM: Identification of a gene expressionsignature associated with recurrent disease in squamous cell carcinomaof the head and neck. Cancer Res 2004, 64(1):55-63.

Pre-publication historyThe pre-publication history for this paper can be accessed here:http://www.biomedcentral.com/1471-2407/11/437/prepub

doi:10.1186/1471-2407-11-437Cite this article as: Reis et al.: A gene signature in histologically normalsurgical margins is predictive of oral carcinoma recurrence. BMC Cancer2011 11:437.

Submit your next manuscript to BioMed Centraland take full advantage of:

• Convenient online submission

• Thorough peer review

• No space constraints or color figure charges

• Immediate publication on acceptance

• Inclusion in PubMed, CAS, Scopus and Google Scholar

• Research which is freely available for redistribution

Submit your manuscript at www.biomedcentral.com/submit


Page 11 of 11

http://www.ncbi.nlm.nih.gov/pubmed/17510386?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/20226055?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/20226055?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/16278953?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/16278953?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/16278953?dopt=Abstracthttp://www.worldcat.org/isbn/0387251464http://www.ncbi.nlm.nih.gov/pubmed/19176553?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/19176553?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/19937997?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/11410486?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/11410486?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/11410486?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/14530304?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/14530304?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/15735049?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/15735049?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/15735049?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/11147677?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/11147677?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/9562604?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/9562604?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/15226743?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/16721539?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/18669583?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/18669583?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/11990853?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/11990853?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/17914555?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/17914555?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/19489686?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/19489686?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/20371203?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/20371203?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/15475433?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/15475433?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/19138406?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/19138406?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/19138406?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/14729608?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/14729608?dopt=Abstracthttp://www.ncbi.nlm.nih.gov/pubmed/14729608?dopt=Abstracthttp://www.biomedcentral.com/1471-2407/11/437/prepub

AbstractBackgroundMethodsResultsConclusion

BackgroundMethodsPatientsSamples used for microarrays (training set)Samples used for quantitative real-time reverse-transcription PCR (RQ-PCR) (validation set)RNA Isolation, Microarrays and Validation ExperimentsBioinformatic AnalysesMeta-analysis of published microarray studiesData analysis of in-house microarray studyPenalized Cox regression

ResultsPatient characteristics: OSCC recurrenceDifferentially expressed genes in margins, OSCC and normal oral tissuesOver-expressed genes in surgical margins: prognostic value for recurrenceFour-gene prognostic signature for OSCC recurrenceEffect of reducing the number of available margins

DiscussionConclusionAcknowledgements and fundingAuthor detailsAuthors' contributionsCompeting interestsReferencesPre-publication history

research article open access a gene signature in histologically normal surgical ... · 2018. 3....

Documents