Molecular Profiling and the Identification of Genes Associated with Metastatic Oral Cavity/Pharynx Squamous Cell Carcinoma

ORIGINAL ARTICLE Molecular Profiling and the Identification of Genes Associated With Metastatic Oral Cavity/Pharynx Squamous Cell Carcinoma

Cecelia E. Schmalbach, MD, MS; Douglas B. Chepeha, MD, MSPH; Thomas J. Giordano, MD, PhD; Mark A. Rubin, MD; Theodoros N. Teknos, MD; Carol R. Bradford, MD; Gregory T. Wolf, MD; Rork Kuick, MS; David E. Misek, PhD; Douglas K. Trask, MD, PhD; Samir Hanash, MD, PhD

Objective: To investigate differences in gene expres- change Ն1.5; PϽ.01). Among this subset, 57 genes also sion profiles between oral cavity/oropharynx squamous exhibited significant differences between metastatic tu- cell carcinoma (OC/OP SCC) primary tumors that have mors and normal mucosa samples (fold change Ն1.5; metastasized to cervical lymph nodes and nonmeta- PϽ.05). This profile included genes related to the extra- static OC/OP SCC tumors. cellular matrix, adhesion, motility, inflammation, and protease inhibition. Collagen type 11 ␣-1 (COL11A1) dem- Design: Oligonucleotide microarray analysis of pri- onstrated the greatest differential expression between mary tumors was used to produce gene expression pro- metastatic and nonmetastatic OC/OP SCC tumors (fold files. Profile comparisons between metastatic and non- change=7.61; P=.002). Tissue inhibitor of metallopro- metastatic tumors were performed using principal teinase 1 (TIMP-1) also demonstrated increased expres- component analysis, t test, and fold change differences. sion in metastatic tumors (fold change=3.3; P=.003). A similar comparison between metastatic tumors and non- cancer oral mucosa samples was performed to ensure tu- Conclusions: Metastatic OC/OP SCC has a distinct gene mor origin. expression profile compared with nonmetastatic OC/OP SCC and normal oral mucosa. This metastatic profile in- Subjects: A prospective cohort of 20 patients with pre- cludes genes related to the extracellular matrix, adhe- viously untreated OC/OP SCC who underwent patho- sion, motility, and protease inhibition. Knowledge gained logic staging following surgical resection and lymphad- through tumor gene expression profiling may facilitate enectomy. early detection of aggressive tumors and targeted therapeutic interventions. Results: Of the approximately 9600 genes profiled, 101 demonstrated significant expression differences between the metastatic and nonmetastatic tumors (fold Arch Otolaryngol Head Neck Surg. 2004;130:295-302

ONVENTIONAL THERAPEU- gene profiles to predict patient survival tic modalities for oral cav- for early lung adenocarcinoma3 and to ity and oropharyngeal differentiate (1) high-grade from low- squamous cell carcinoma (OC/OP SCC) have CME course available at ledC to increased local tumor control, es- pecially for patients diagnosed as having www.archoto.com From the Departments of 1 Otolaryngology–Head and early disease. However, overall survival 4 Neck Surgery has not changed significantly because pa- grade gliomas and (2) clear cell ovarian (Drs Schmalbach, Chepeha, tients continue to die from disease at re- carcinomas from other ovarian tumors Teknos, Bradford, and Wolf), gional and distant sites. Improvement in portending a poor prognosis.5 Other Pathology (Drs Giordano and patient survival requires an increased un- investigators have successfully used pro- Rubin), and Pediatrics derstanding of tumor metastasis so that ag- filing studies to identify gene expression (Mr Kuick and Drs Misek and gressive tumors can be detected early in patterns unique to head and neck SCC Hanash), University of the disease process and targeted therapeu- (HNSCC).6-12 To gain insight into the Michigan Medical School, tic interventions can be developed. biology of aggressive tumors, we used oli- Ann Arbor; and Department of Otolaryngology–Head and Molecular profiling provides an gonucleotide microarrays to identify a set Neck Surgery, University of opportunity to simultaneously investigate of genes distinguishing OC/OP SCC pri- Iowa, Iowa City (Dr Trask). numerous potential markers and molecu- mary tumors in patients with cervical The authors have no relevant lar pathways associated with metastasis.2 lymph node metastasis from nonmeta- financial interest in this article. We previously used DNA microarray static OC/OP SCC tumors.

(REPRINTED) ARCH OTOLARYNGOL HEAD NECK SURG/ VOL 130, MAR 2004 WWW.ARCHOTO.COM 295

Histologic Tobacco Years Years Age, y/Sex Stage Site Differentiation Exposure, PY Tobacco Free* Alcohol Habit† Alcohol Free‡ Patients With Cervical Lymph Node Metastasis 49/M T2 N3 M0 Tonsil Poor 30 1 Abuse 8 65/M T2 N2b M0 Tongue Moderate 50 0 Abuse 0 51/M T2 N2c M0 FOM Poor 30 1 Abuse 0 64/M T4 N2b M0 FOM Poor 50 0 Daily 0 49/M T2 N2c M0 FOM Well 25 0 Social 0 61/M T2 N2c M0 FOM Moderate 45 0 Abuse 8 49/F T4 N2b M0 FOM Moderate 30 0 Abuse 12 59/M T3 N2b M0 Tonsil Moderate 20 0 Daily 0 64/F T4 N2b M0 RMT Moderate 0 NA Social 0 57/F T2 N1 M0 Tongue Moderate 30 0 Abuse 0 70/F T3 N2b M0 RMT Moderate 5 52 Social 1 73/M T3 N1 M0 Tongue Poor 30 0 Social 0 68/M T4 N1 M0 RMT Moderate 90 1 Abuse 22 Patients Without Lymph Node Metastasis 69/M T2 N0 M0 FOM Well 35 27 Daily 1 47/F T1 N0 M0 Tongue Moderate 0 NA Social 1 65/M T2 N0 M0 FOM Moderate 80 0 Daily 0 77/M T1 N0 M0 Tongue Moderate 50 5 Abuse 6 46/M T2 N0 M0 Tongue Moderate 0 NA None NA 59/M T2 N0 M0 Tongue Well 0 NA None NA 55/M T1 N0 M0 Tongue Well 0 NA None NA

Abbreviations: FOM, floor of mouth; NA, not applicable; PY, pack-years; RMT, retromolar trigone. *Number of years since quitting; 0 indicates that patient continues to smoke. †Social indicates nondaily alcohol consumption; Daily, more than 1 but less than 6 drinks per day; Abuse, more than 6 drinks per day. ‡Number of years since quitting; 0 indicates that patient continues to drink alcohol.

METHODS and purified according to the manufacturer’s instructions. RNA samples were further purified on RNeasy spin columns (Qia- gen, Valencia, Calif). RNA quality was assessed by 1% agarose TISSUE SAMPLES AND CELL LINES gel electrophoresis in the presence of ethidium bromide. Tissue was obtained from a prospective cohort of 20 patients with newly diagnosed, pre viously untreated OC/OP SCC who present- MICROARRAY ANALYSIS ed to the University of Michigan Hospital, Ann Arbor, between Feb- Preparation of RNA and hybridization to commercially available ruary and December 2001. The University of Michigan institutional HG_U95Av2 microarrays (Affymetrix, Santa Clara, Calif) were reviewboardapprovedthestudy,andwrittenconsentwasobtained. performed according to the manufacturer’s protocols as previously All patients underwent pathologic staging following surgical tu- described.4,13 Microarrays were scanned, and probe intensities were mor resection and lymphadenectomy (13 metastatic and 7 non- extracted from the image (GeneArray scanner and Microarray Suite metastatic tumors). Demographic information for this prospective 4.0; Affymetrix). Each probe set on the array represented a gene cohort (Table 1) demonstrates that the 20 patients sampled are and typically consisted of 16 individual 25-base oligonucleotide representative of the HNSCC population. featurescomplementarytospecificDNA,calledperfectmatch(PM) At the time of surgical resection, a sample of the primary probes, and 16 identical probes whose sequence had been altered tumor was snap frozen in liquid nitrogen. Samples were em- at the central base, called mismatch (MM) probes. Publicly avail- bedded in OCT freezing media (Miles Scientific, Naperville, Ill) able software was used to process the probe intensities to obtain and stored at −80°C for RNA extraction. A surgical patholo- normalized results (software and documentation available as gist evaluated a cryotome section (5 µm) from each block us- “Supplementary Material” at http://dot.ped.med.umich.edu:2000 ing hematoxylin-eosin (H&E) staining. RNA was isolated from /pub/HeadNeck/index.html).4,13 areas of tissue containing at least 70% tumor cellularity. The microarray from a T2 N2 c floor of mouth tumor was Oral cavity SCC cell lines SCC-15 and SCC-25 were ob- selected as the standard because it was generalizable to the tained from the American Type Tissue Collection, Manassas, sample population, had low background measures, and had Va. The UM-SCC-1 and UM-SCC-17A cell lines were ob- bright signals. Probe pairs for which the PM-MM difference was tained from Thomas Carey, PhD, at the University of Michi- less than –200 on the standard were removed from the analy- gan. Normal controls included oral cavity mucosa samples from sis. The remaining PM-MM differences were averaged for each 4 patients without cancer who underwent surgical treatment probe set on each microarray by discarding the 25% highest for obstructive sleep apnea and primary cell explants of oral and lowest differences and averaging the remaining differ- cavity keratinocytes. ences. The intensities for each microarray were normalized to the standard using a piece-wise linear function that made 99 RNA ISOLATION evenly spaced quantiles agree with the corresponding quantiles in the distribution of the standard. Fold changes were com- Total cellular RNA was extracted from the tumor samples and puted as the ratio of group means, after first replacing the means cell lines as previously described.4,13 Samples were homog- that were less than 100 with 100. Normalized intensities were enized using Trizol reagent (Life Technologies, Gaithersburg, Md) log-transformed by mapping x to log(max[x+100,0] +100) prior

(REPRINTED) ARCH OTOLARYNGOL HEAD NECK SURG/ VOL 130, MAR 2004 WWW.ARCHOTO.COM 296

©2004 American Medical Association. All rights reserved. Downloaded From: https://jamanetwork.com/ on 09/26/2021 to performing statistical tests. Unigene Cluster identifiers were computed using homology of the probe set sequences to the A 50 sequences of Unigene Cluster members (software and docu- Met OC/OP SCC mentation available as “Supplementary Material” at Nonmet OC/OP SCC http://dot.ped.med.umich.edu:2000/pub/HeadNeck/index.html). 40 SCCA Cell Lines Normal Explant One-sided signed-rank tests of PM-MM differences for each Normal Mucosa probe set on each array were used to determine the number of 30 detectable transcripts. 20 IMMUNOHISTOCHEMICAL ANALYSIS OF TIMP-1 10 Expression of increased tissue inhibitor of metalloproteinase 1 (TIMP-1) in primary metastatic and nonmetastatic tumors was 0 investigated using immunohistochemical analysis on a tissue microarray comprised 102 independent tumor samples. All pa- Principal Component 2 –10 tients for this validation group presented to the University of Michigan Hospital between 1997 and 2000 with newly diag- –20 nosed, previously untreated HNSCC. University of Michigan institutional review board approval and written consent were ob- –30 tained. Patients underwent pathologic staging following surgical tumor resection and lymphadenectomy (70 had metastatic SCC –40 and 32 had nonmetastatic SCC). The mean age for the meta- –60 –40–20 0 20 40 static and nonmetastatic validation group was 57 years and 58 Principal Component 1 years, respectively. The mean disease-free interval for the metastatic group was 20 months, with a mean follow-up of 22 months. B Met OC/OP SCC The nonmetastatic group had a mean disease-free interval of 28 Met OC/OP SCC months, with a mean follow-up of 29 months. Met OC/OP SCC A head and neck pathologist used H&E staining to evalu- Met OC/OP SCC Met OC/OP SCC ate cryotome sections (5 µm) from each paraffin-embedded pri- Met OC/OP SCC mary tumor block. Representative areas of tumor and normal Met OC/OP SCC Met OC/OP SCC squamous cell mucosa were marked. A high-density tissue mi- Met OC/OP SCC croarray was constructed from the marked areas using 3 rep- Met OC/OP SCC licate tumor cores (0.6-mm diameter) and 1 normal mucosa Met OC/OP SCC Met OC/OP SCC core per patient. Nonmet OC/OP SCC Immunohistochemical analysis of the paraffin- Nonmet OC/OP SCC Nonmet OC/OP SCC embedded tissue microarray was performed using standard avi- Nonmet OC/OP SCC 13 din-biotin complex method. Protein expression of TIMP-1 was Nonmet OC/OP SCC evaluated with a rabbit polyclonal anti–TIMP-1 antibody (sc- Nonmet OC/OP SCC SCC-25 5538; Santa Cruz Biotechnology, Santa Cruz, Calif) (1:50). A UM-SCC-1 pathologist, blinded to the metastatic status of the tumors, scored SCC-15 UM-SCC-17A the nuclear protein expression as negative (1), weak (2), mod- Normal Explant erate (3), and strong (4). The mean TIMP-1 score for each tu- Met OC/OP SCC mor was compared with normal controls using a Wilcoxon rank Nonmet OC/OP SCC Normal Mucosa sum test. Because the gene microarray analysis revealed in- Normal Mucosa creased levels of TIMP-1 in metastatic OC/OP SCC, tissue mi- Normal Mucosa croarray tumors were categorized into an abundant nuclear stain- Normal Mucosa ing group (mean expression score Ն3.0) and a nonabundant Ͻ Figure 1. Results of principal component and cluster analysis for all tumor, nuclear staining group (mean expression score 3.0). The Fisher normal, squamous cell carcinoma cell line, and explant samples. A, A total of exact test was then used to study the relationship of TIMP-1 2928 probe sets (mean Ͼ100 and coefficient of variance [SD/mean] Ͼ0.6) and cervical lymph node metastasis, sex, primary tumor site, were standardized by subtracting the mean and dividing by the SD for each histologic differentiation, pathologic stage, tumor size, pres- probe set. The first 2 principal components were plotted (3-dimensional ence of extracapsular spread, and tumor recurrence. Linear re- version of principal component and analysis is available at gression was used to investigate the relationship of TIMP-1 and http://dot.ped.med.umich.edu:2000/pub/HeadNeck/index.html). B, Dendro- gram from “average” clustering14 using the same 2928 probe sets, trans- age. The Kaplan-Meier method was used to investigate the re- formed by taking the logarithms of ratios to the mean. Met OC/OP SCC indi- lationship between TIMP-1 and disease-free survival. cates metastatic oral cavity/oropharynx squamous cell carcinoma (tumors metastatic to cervical lymph nodes); Nonmet, nonmetastatic.

RESULTS samples averaged 7731 detectable probe sets and normal samples averaged 7798 detectable probe sets (PϽ.05). Using the Affymetrix HG_U95Av2 GeneChip (12625 A pictorial representation of the gene expression pro- probe sets representing approximately 9600 unique files was constructed using principal component analy- genes), we obtained gene expression profiles for 13 OC/OP sis. The greatest variance between sample gene profiles SCC primary tumors from patients with cervical lymph is represented by the first 2 principal components, which node metastasis, 7 nonmetastatic OC/OP SCC tumors, we graphed on the x- and y-axis of Figure 1A. Simi- 4 SCC cell lines, 4 normal oral cavity mucosa samples, larly, a dendrogram based on the hierarchical clustering and normal oral cavity keratinocyte explants. Using the is depicted in Figure 1B. The normal mucosa samples, 4 comparison outlined in the “Methods” section, tumor SCC cell lines, and OC/OP tumors clustered into 3 dis-

(REPRINTED) ARCH OTOLARYNGOL HEAD NECK SURG/ VOL 130, MAR 2004 WWW.ARCHOTO.COM 297

©2004 American Medical Association. All rights reserved. Downloaded From: https://jamanetwork.com/ on 09/26/2021 metastatic nodal status. Therefore, the variation between A 60 tumor profiles could not be explained solely by genes 50 Floor of Mouth that were differentially expressed with respect to metas- RMT 40 Tongue tasis. However, the nonmetastatic tumors clustered Tonsil closer together as a group, indicating greater homogene- 30 ity in their gene expression compared with the meta- 20 static tumors (Figure 2B). A 2-sample t test using log-transformed data from 10 metastatic and nonmetastatic tumors was performed for 0 all 12625 probe sets represented on the Affymetrix Principal Component 2 –10 HG_U95Av2 GeneChip to identify genes associated with metastatic OC/OP SCC; 160 probe sets met statistical sig- –20 nificance at PϽ.01. To reduce the proportion of false- –30 positives among the selected 160 probe set list, the means –30 –20–100 10 20 304050 of the metastatic and nonmetastatic tumors were re- Principal Component 1 quired to differ by at least 1.5-fold. This fold change cri- B 60 terion was met by 101 of the 160 probe sets. To exclude genes that may have differed in expres- 50 Met OC/OP SCC Nonmet OC/OP SCC sion level between metastatic and nonmetastatic tumors 40 owing to differences in the proportion of normal tissue 30 elements, gene profiles were also compared between the metastatic and normal mucosa samples. Within the 101 20 probe sets, we identified genes with both an expression 10 level difference between metastatic tumors and normal samples at PϽ.05 (2-sample t test) and a minimum dif- 0 ference in mean expression of 1.5-fold (the direction of Principal Component 2 –10 change being the same as the metastatic and nonmeta-

–20 static comparison). Overall, 61 probe sets representing 57 genes fulfilled all selection criteria. Four of the genes –30 –30 –20–100 10 20 304050 were represented in duplicate on the Affymetrix Principal Component 1 GeneChip, 50 represented classified genes, and 7 are currently unnamed genes. The expression values for Figure 2. Results of principal component analysis for the 20 primary tumors. each tumor and normal sample are provided in Ͼ A total of 2036 probe sets (mean 100 and coefficient of variance Figure 3, along with t test and fold-change results. [SD/mean] Ͼ0.6) were standardized by subtracting the mean and dividing by the SD for each probe set. The first 2 principal components were plotted. Among the list were genes related to cellular adhesion, Samples are labeled according to primary tumor site (A) and according to cell growth and differentiation, motility, and angiogen- metastatic nodal status (B) (3-dimensional version of principal component esis (Table 2). and analysis is available at http://dot.ped.med.umich.edu:2000/pub /HeadNeck/index.html). Met OC/OP SCC indicates metastatic oral To estimate the overall number of false-positives, a cavity/oropharynx squamous cell carcinoma (tumors metastatic to cervical computer program randomly relabeled the 20 OC/OP gene lymph nodes); Nonmet, nonmetastatic; RMT, retromolar trigone. profiles 1000 times, irrespective of nodal status. Each of the 1000 permutations was statistically analyzed using the identical selection criteria for obtaining the 61 probe tinct groups, with tumors having the greatest degree of sets in Figure 3. On average, only 18 of the 12625 probe heterogeneity in expression profiles. sets met all selection criteria by chance alone, and only To gain further insight into the gene expression 31 of the 1000 random computer trials yielded 61 or more profile specifically associated with tumor metastasis, the probe sets meeting the selection criteria. This permuta- effects of cell line profiles and normal mucosa profiles tion test overestimated the false-positive rate because were removed, and a principal component analysis using analysis was based only on tumors. When all samples only OC/OP SCC samples was repeated. This explor- (OC/OP tumors and normal mucosa) were randomly per- atory data analysis did not uncover a clear pattern in mutated 1000 times, an average of 8 probe sets met the gene expression profiles related to primary tumor site selection criteria by chance alone. The overall false- (Figure 2A) or histologic differentiation. Specifically, positive rate of this study falls between these 2 esti- OC and OP tumors did not separate into 2 distinct clus- mates of 8 and 18 probe sets. ters. Two of the OC tumors (tongue and floor of mouth) Immunohistochemical analysis was used to con- shared greater profile similarity to the OP tumors than firm protein expression and tumor cell origin of the gene their OC counterparts. For this reason, both OC and OP that exhibited the greatest degree of differential expres- tumors were included in the final comparison of meta- sion, for which a commercial antibody was available. In static and nonmetastatic tumors. Figure 2 was con- the metastatic vs nonmetastatic tumors, TIMP-1 had a structed using the 2036 probe sets that demonstrated mean expression level 3 times higher. TIMP-1 also dem- the greatest degree of variance between the 20 tumor onstrated 76 times greater expression in the SCC cell lines samples, irrespective of cervical lymph node status. vs normal keratinocyte explants (data not shown). Im- Tumors did not cluster into 2 distinct groups based on munohistochemical analysis for nuclear TIMP-1 pro-

(REPRINTED) ARCH OTOLARYNGOL HEAD NECK SURG/ VOL 130, MAR 2004 WWW.ARCHOTO.COM 298

©2004 American Medical Association. All rights reserved. Downloaded From: https://jamanetwork.com/ on 09/26/2021 Met vs Met vs Nonmet Normal Fold Fold Gene Symbol Gene Title Probe Set Change P Value Change P Value Met Nonmet OC/OP SCC SCC Normal Up-regulated in Met OC/OP SCC COL11A1 Collagen Type II, α1 37892_at 7.61 .0019 25.18 .0002 CL683 Weakly Similar to Glutathione Peroxidase 2 36009_at 3.36 .0047 3.36 .0161 TIMP-1 Tissue Inhibitor of Metalloproteinase 1 1693_s_at 3.30 .0028 8.78 .0001 C1QB Complement Component 1, Q Subcomponent 38796_at 3.27 .0041 4.10 .0061 KIAA1077 KIAA1077 Protein 35832_at 3.25 .0055 22.39 <.0001 NINJ1 Ninjurin 1 41475_at 3.10 .0038 3.10 .0054 FCGR3B Fc Fragment of IgG, Low Affinity IIIb/Receptor for (CD16) 37200_at 2.79 .0073 4.05 .0067 DKFZP586L1 DKFZP586L151 Protein 36007_at 2.76 .0100 8.37 .0003 CSPG2 Chondroitin-Sulfate Proteoglycan 2 (Versican) 38112_g_at 2.52 .0087 24.98 <.0001 RARRES2 Retinoic Acid Receptor Responder (Tazarotene Induced) 2 34407_at 2.36 .0023 3.47 .0011 OSF-2 Osteoblast-Specific Factor 2 (Fasciclin I–like) 1451_s_at 2.30 .0080 17.21 <.0001 None Homo sapiens cDNA FLJ31388 fis, Clone NT2NE1000023 34112_r_at 2.14 .0004 1.87 .0110 TCEA2 Transcription Elongation Factor A (SII), 2 745_at 2.13 .0065 2.13 .0071 CDH11 Cadherin 11, Type 2, OB-Cadherin (Osteoblast) 36976_at 2.05 .0089 16.70 <.0001 SFRP4 Secreted Frizzled-Related Protein 4 41405_at 2.05 .0038 2.05 .0252 ACAA2 Acetyl-Coenzyme A Acyltransferase 2 41530_at 1.97 .0010 1.69 .0134 ADA Adenosine Deaminase 907_at 1.92 .0079 3.46 .0008 RUNX1 Runt-Related Transcription Factor 1 (AML1) 943_at 1.90 .0046 3.71 .0001 None Homo sapiens, Clone IMAGE:4179842, mRNA 34530_at 1.89 .0038 1.74 .0296 PCOLCE Procollagen C-Endopeptidase Enhancer 31609_s_at 1.87 .0093 3.07 .0010 PRDM2 PR Domain Containing 2, With ZNF Domain 33922_at 1.67 .0008 3.65 <.0001 ARHC RAS Homolog Gene Family, Member C 1395_at 1.61 .0069 2.32 .0006 RUNX1 Runt-Related Trascription Factor (AML1) 393_s_at 1.59 .0003 2.52 <.0001 EDR2 Early Development Regulator 2 36960_at 1.58 .0091 17.2 .0148

Down-regulated in Met OC/OP SCC SPRR2C Small Proline-Rich Protein 2C 36242_at 0.12 .0001 0.15 .0022 TGM3 Transglutaminase 3 32868_at 0.19 .0038 0.04 <.0001 KLK13 Kallikrein 13 36406_at 0.20 .0007 0.08 <.0001 SERPINB2 Serine (or Cysteine) Proteinase Inhibitor 2 (Ovalbumin) 37185_at 0.21 .0018 0.09 <.0001 MAL MAL, T-cell Differentiation Protein 38051_at 0.22 .0015 0.02 <.0001 LCN2 Lipocalin 2 (Oncogene 24p3) 32821_at 0.22 .0057 0.12 .0003 KIAA0624 KIAA0624 Protein 35579_at 0.25 .0041 0.23 .0001 None ESTs, Weakly Similar to A37413 Calbindin D, 28K 32163_f_at 0.28 .0026 0.18 <.0001 SERPINB4 Serine Proteinase Inhibitor 4 (Ovalbumin) 1549_s_at 0.30 .0039 0.29 .0214 ECM1 Extracellular Matrix Protein 1 37600_at 0.30 .0002 0.07 <.0001 PLA2G4A Phospholipase A2, Group IVA 35938_at 0.32 .0077 0.25 .0023 ANXA9 Annexin A9 39669_at 0.35 .0090 0.05 <.0001 None Homo sapiens mRNA; cDNA DKFZp586F1223 39625_at 0.35 .0084 0.13 <.0001 IGFBP6 Insulin-like Growth Factor–Binding Protein 6 1736_at 0.36 .0092 0.29 .0218 KLK13 Kallikrein 13 36407_at 0.37 .0024 0.14 <.0001 P11 Protease, Serine, 22 37093_at 0.38 .0075 0.02 <.0001 HBP17 Heparin-Binding Growth Factor–Binding Protein 38489_at 0.39 .0089 0.25 .0046 None Homo sapiens cDNA FLJ31482 fis, Clone NT2NE2001666 37190_at 0.40 .0026 0.30 .0021 CDS1 CDP-Diacylglycerol Synthase 1 40217_s_at 0.43 .0028 0.31 .0008 APG-1 Heat Shock Protein (hsp110 Family) 40354_at 0.45 .0092 0.26 .0002 CDS1 CDP-Diacylglycerol Synthase 1 40218_at 0.46 .0009 0.36 .0002 CSTA Cystatin A (Stefin A) 39581_at 0.47 .0075 0.27 .0010 CRABP2 Cellular Retinoic Acid–Binding Protein 2 1057_at 0.48 .0084 0.14 <.0001 CRYM Crystallin, Mu 38285_at 0.48 .0094 0.29 .0013 IL1RN Interleukin 1 Receptor Antagonist 31343_at 0.49 .0003 0.04 <.0001 CRABP2 Cellular Retinoic Acid–Binding Protein 2 41783_at 0.50 .0074 0.15 <.0001 ABLIM Actin-Binding LIM Protein 1 40155_at 0.50 .0052 0.25 <.0001 PPT2 Palmitoyl-Protein Thioesterase 2 38109_at 0.51 .0066 0.30 .0011 SNCG Synuclein γ (Breast Cancer–Specific Protein 1) 36555_at 0.51 .0081 0.26 <.0001 ARHD ras Homolog Gene Family, Member 31846_at 0.53 .0077 0.50 .0144 RNTRE Related to the N Terminus of TRE 34594_at 0.55 .0051 0.18 <.0001 PPT2 Palmitoyl Protein Thioesterase 2 38108_at 0.55 .0029 0.39 .0002 None Homo sapiens mRNA; cDNA DKFZp5886K1318 36551_at 0.55 .0086 0.28 <.0001 ATP10B ATPase, Class V, Type 10B 37415_at 0.58 .0045 0.29 <.0001 None Homo sapiens mRNA; cDNA DKFZp564A072 37366_at 0.59 .0071 0.40 .0001 PARD3 PAR-3 (Partitioning Defective 3, Caenorhabditis elegans) 39499_s_at 0.60 .0031 0.45 .0003 TOP3B Topoisomerase (DNA) III β 32360_s_at 0.60 .0089 0.48 .0009

Fold Change From the Median

1/4 1/2 1 2 4

Figure 3. Molecular profile for metastatic oral cavity/oropharynx squamous cell carcinoma (OC/OP SCC; tumors metastatic to cervical lymph nodes) demonstrated using Treeview software (Eisen Lab, Stanford University, Stanford, Calif).14 Sixty-one probe sets representing 57 genes were differentially expressed by metastatic nodal status. Each column represents an individual tissue sample; each row represents 1 of the 61 probe sets that differed simultaneously between metastatic and nonmetastatic tumors (fold change Ն1.5; PϽ.01) and between metastatic tumors and normal mucosa samples (fold change Ն1.5; PϽ.05). The color of each cell represents the fold change above (red) and below (green) the median of all 24 tissue samples. ATPase indicates adenosine triphosphatase; cDNA, complementary DNA; ESTs, expression sequence tags; mRNA, messenger RNA; Nonmet, nonmetastatic.

tein expression revealed a statistically significant differ- COMMENT ence in mean expression between tumor and normal controls (Wilcoxon rank sum test; PϽ.0001). While most of the tumors demonstrating abundant TIMP-1 expres- Various prognostic markers for OC/OP SCC have been sion were metastatic (37/48), the difference between reported. These include amplification of cyclin D1 TIMP-1 abundance and metastatic nodal status did not (PRAD1) on chromosome 11q13, overexpression of the achieve statistical significance (P=.06). Similarly, differ- MYC family of oncogenes, mutations of p53, loss of p16 ences in TIMP-1 abundance based on age, sex, primary protein function, and overexpression of epidermal growth tumor site, histologic differentiation, pathologic stage, factor receptor and transforming growth factor ␣.15,16 Cur- tumor size, extracapsular spread, and tumor recurrence rently no individual marker has emerged as a fully in- were not statistically significant. Expression of TIMP-1 formative prognostic marker, and treatment decisions for did not correlate with disease-free survival. OC/OP SCC patients continue to rely exclusively on the

(REPRINTED) ARCH OTOLARYNGOL HEAD NECK SURG/ VOL 130, MAR 2004 WWW.ARCHOTO.COM 299

©2004 American Medical Association. All rights reserved. Downloaded From: https://jamanetwork.com/ on 09/26/2021 could be achieved. Further reasons for the lack of simi- Table 2. Classification of Gene Subset Differentially larity between the profiles include differences in gene se- Expressed Between Metastatic lection criteria and statistical analysis. One criticism of and Nonmetastatic OC/OP SCC* our study is the use of a 1.5-fold change selection crite- Extracellular matrix/adhesion molecules rion, which is smaller than that applied in other studies. Tissue inhibitor of metalloproteinase 1 (TIMP-1) ↑ However, we believe that a smaller fold change is appro- Nerve injury induced protein (NINJ1) ↑ priate when comparing histologically identical tissue types, Chondroitin sulfate proteoglycan 2 (CSPG2) ↑ such as OC/OP SCC, with one another. Osteoblast-specific factor 2 (OSF-2) ↑ One of the strengths of our study design was that Cadherin 11, type 2 (CDH11) ↑ our comparisons did not rely on cell line profiles. Partitioning defective 3 (PARD3) ↓ Cell cycle/apoptosis Instead, we compared metastatic and normal mucosa Secreted frizzled-related protein 4 (SFRP4) ↑ samples to ensure that expression differences were due Partitioning defective 3 (PARD3) ↓ to genes from OC/OP SCC tumors as opposed to sur- Intracellular signaling rounding normal epithelium. Although cell lines pro- Chondroitin sulfate proteoglycan 2 (CSPG2) ↑ vide a more definitive means to confirm tumor cell ori- Immune/inflammatory response gin, an in vitro setting does not mirror the natural milieu ↑ Complement component 1 (C1QB) of tumors because inflammatory and stromal compo- Fc fragment of IgG (FCGR3B) ↑ Serine proteinase inhibitor 4 (SERPINB4) ↓ nents are lacking. Consequently, genes related to cellu- Phospholipase A2 (PLA2G4A) ↓ lar adhesion, mobility, and inflammation and genes Interleukin 1 receptor antagonist (IL1RN) ↓ requiring activation from surrounding nontumor cells Cellular differentiation would not be observed in a model incorporating cell line Chondroitin sulfate proteoglycan 2 (CSPG2) ↑ criteria. Indeed, in our study collagen type XI ␣-1 ↓ Cellular retinoic acid−binding protein 2 (CRABP2) (COL11A1) demonstrated the greatest difference Transcription factor Transcription elongation factor A (TCEA-2) ↑ between metastatic and nonmetastatic tumors. This dif- Angiogenesis ference would not be predictable from our cell line pro- Extracellular matrix protein 1 (ECM1) ↓ files (data not shown). Previous studies have demonstrated that increased COL11A1 expression requires Abbreviation: OC/OP SCC, oral cavity/pharynx squamous cell carcinoma. paracrine activation from surrounding stroma.17,18 Our *Arrow indicates directional change in gene expression of metastatic use of primary tumors may also account for the limited tumors with respect to both nonmetastatic tumors and normal mucosa samples. overlap of only 2 genes from our metastatic profile (IgG Fc receptor gene and serine protease inhibitor-2) with the findings by Dong et al,10 who used cell lines to iden- clinical TNM system. Overall, cervical lymph node me- tify genes differentially expressed between metastatic tastasis remains the most important prognostic indica- and nonmetastatic tumors. tor for tumor recurrence and survival. We used oligo- Molecular profiling provides complex data, which nucleotide microarrays to determine if regionally may be difficult to interpret. We recognize that genes do metastatic OC/OP SCC exhibits a distinct gene expres- not function independently and that gene interactions sion profile. Increased understanding of the molecular play a critical role in tumor progression. However, ex- biology of metastasis may allow aggressive tumors, which amination of individual gene functions and related path- warrant adjuvant therapy despite their early stage, to be ways is still informative. For example, several genes from recognized at the time of diagnosis. A molecular profile our profile of metastatic OC/OP SCC are involved in path- for metastasis may also lead to the development of novel ways related to the extracellular matrix (ECM). This find- targeted therapies for these aggressive tumors. ing is plausible given the defined role of the ECM in cel- We compared our metastatic profile with previ- lular morphogenesis, proliferation, differentiation, ously reported profiles of genes unique to HNSCC.6-12 As apoptosis, and migration.19 Collagen is a major struc- expected, the genes differentiating metastatic from non- tural component of the ECM. In our study, the mean ex- metastatic tumor had limited overlap with the genes dif- pression of COL11A1 was more than 7.5 times higher in ferentiating tumor from normal mucosa. Although we metastatic vs nonmetastatic tumors. Increased COL11A1 identified a set of genes differentially expressed with re- expression is related to uncontrolled epithelial prolifera- spect to cervical lymph node metastasis, a previous study tion through the ␤-cadherin/Wnt/WISP-1 pathway.17 An did not detect such a difference.12 This discrepancy can association between COL11A1 and colorectal carci- be explained by variations in study design. The most com- noma progression has previously been reported.18 pelling reason is that we used a more recent version of In addition, serine protease inhibitor-2 (SERPINB2) the Affymetrix GeneChip, which consists of almost twice is related to the ECM. This gene regulates ECM remod- as many probe sets compared with the older Hu- eling by inhibiting urokinase-type plasminogen activa- GeneFL chip. Patient selection may also have contrib- tor, a protein involved in tumor invasion and metasta- uted to differences observed. We only profiled patients sis.20 SERPINB2 expression was decreased in our metastatic with newly diagnosed, previously untreated SCC to elimi- OC/OP SCC tumors relative to the nonmetastatic tu- nate potential gene profile alterations resulting from ra- mors. Dong et al10 demonstrated similar findings using diation and chemotherapy. In addition, we only pro- SCC cell lines, and decreased SERPINB2 expression was filed patients who underwent lymphadenectomy so that observed in studies involving metastatic lung,21 breast,22 pathologic staging for cervical lymph node metastasis and gastric carcinoma.23

(REPRINTED) ARCH OTOLARYNGOL HEAD NECK SURG/ VOL 130, MAR 2004 WWW.ARCHOTO.COM 300

©2004 American Medical Association. All rights reserved. Downloaded From: https://jamanetwork.com/ on 09/26/2021 The integrity of the ECM is maintained by various Submitted for publication February 25, 2003; final revi- metalloproteinases, several of which have an association sion received May 29, 2003; accepted July 1, 2003. with tumor invasion, angiogenesis, and metastasis.24 There- This study was supported by grant T32 DC00024 from fore, it may seem counterintuitive that we observed in- the National Institute of Health Research Training in the creased expression of TIMP-1, a metalloproteinase inhibi- Communicative Sciences, Bethesda, Md, and in part by grant tor, in the setting of metastatic disease. However, recent NIH CA 46592 from the Tissue Core of the University of studies have reported a paradoxical role for TIMP-1.Inad- Michigan Comprehensive Cancer Center, Ann Arbor. dition to metalloproteinase inhibition, TIMP-1 rescues epi- We would like to thank Augusto Paulino, MD, for his pa- thelial cells from apoptosis25 and mediates up-regulation thology efforts; Barbra Lamb, Michelle L. Lizyness, Dafydd 26 of Bcl-XL, which is known to have antiapoptotic prop- Thomas, Douglas Selby, Donna Chuey, Enola Cushenberry, erties. Furthermore, an association between overexpres- Donita Sanders, Michele Le Blanc, Nancy McAnsh, and Yaron 27,28 sion of Bcl-XL and HNSCC has been reported. Al- Fridman for their technical support; Jeremy M. G. Taylor for though a previous HNSCC study found an association statistical consultation; and Tamara Miller for her excellent between TIMP-1 and T stage but not metastatic nodal sta- support as the University of Michigan otolaryngology oncol- tus,29 our study is consistent with colorectal30 and breast ogy nurse. We appreciate the participation of referring phy- carcinoma31 findings associating TIMP-1 expression with sicians Ronald Bogdasarian, MD, Gerald Ebmeyer, MD, Ken- metastasis. neth Markiewicz, MD, Stephen Robinson, MD, Michael Sterett, Validation of gene microarray analysis is often MD, Eric Hensen, MD, and Aykan Tadros, MD. achieved through quantitative reverse transcription poly- Corresponding author and reprints: Cecelia E. Schmal- merase chain reaction, and we have previously validated bach, MD, Department of Otolaryngology–Head and Neck our oligonucleotide microarray methodology in this man- Surgery, 1500 W Medical Center Dr, Room A520 MSRB-1, ner.3-5 However, application of quantitative polymerase Ann Arbor, MI 48109-0656 (e-mail: [email protected]). chain reaction to the same tumor samples simply con- firms that the microarray technique is accurate in detect- REFERENCES ing RNA. It does not provide additional information related to protein expression, nor does it confirm cellular origin. For this reason, we chose to investigate TIMP-1 us- 1. Petruzzelli GJ, Benefield J, Yong S. Mechanism of lymph node metastases: cur- rent concepts. Otolaryngol Clin North Am. 1998;31:585-599. ing tissue microarray immunohistochemical analysis in or- 2. Whipple ME, Kuo WP. DNA microarrays in otolaryngology–head and neck sur- der to confirm protein expression and to determine cel- gery. Otolaryngol Head Neck Surg. 2002;127:196-204. lular origin using a larger, independent cohort. We 3. Beer DG, Kardia SLR, Huang CC, et al. Gene-expression profiles predict survival demonstrated that TIMP-1 was expressed at the protein of patients with lung adenocarcinoma. Nat Med. 2002;8:816-824. level and confirmed that expression of TIMP-1 was from 4. Rickman DS, Bobek MP, Misek DE, et al. Distinctive molecular profiles of high- grade and low-grade gliomas based on oligonucleotide microarray analysis. Can- HNSCC tumor cells as opposed to fibroblasts, stroma, or cer Res. 2001;61:6885-6891. normal mucosa (PϽ.0001). As predicted by genomic pro- 5. Schwartz DR, Kardia SLR, Shedden KA, et al. Gene expression in ovarian cancer filing, patients with cervical lymph node metastasis had reflects both morphology and biological behavior distinguishing clear cell from higher levels of TIMP-1 expression by immunohistochemi- other poor-prognosis ovarian carcinomas. Cancer Res. 2002;62:4722-4729. 6. Belbin TJ, Singh B, Barber I, et al. Molecular classification of head and neck squa- cal analysis. Although the difference was not statistically mous cell carcinoma using cDNA microarrays. Cancer Res. 2002;62:1184- significant, a P value of .06 remains of interest because our 1190. sample size provided sufficient power to detect only large 7. Leethanakul C, Patel V, Gillespie J, et al. Gene expression profiles in squamous differences, and immunohistochemical analysis cannot pro- cell carcinomas of the oral cavity: use of laser capture microdissection for the vide the degree of quantitative measure generated using construction and analysis of stage-specific cDNA libraries. Oral Oncol. 2000;36: 474-483. gene microarray technology. 8. Villaret DB, Wang T, Dillon D, et al. Identification of genes overexpressed in head and neck squamous cell carcinoma using a combination of complementary DNA CONCLUSIONS subtraction and microarray analysis. Laryngoscope. 2000;110:374-381. 9. Alevizos I, Mahadevappa M, Zhang X, et al. Oral cancer in vivo gene expression profiling assisted by laser capture microdissection and microarray analysis. On- Cervical lymph node metastasis is the most important cogene. 2001;20:6196-6204. prognostic marker for both tumor recurrence and over- 10. Dong G, Loukinova E, Chen Z, et al. Molecular profiling of transformed and meta- all survival. We identified a gene expression profile for static murine squamous carcinoma cells by differential display and cDNA mi- regionally metastatic OC/OP SCC using oligonucleo- croarray reveals altered expression of multiple genes related to growth, apop- ␬ tide microarray analysis. Genes related to the extracel- tosis, angiogenesis, and the NF- B signaling pathway. Cancer Res. 2001;61: 4797-4808. lular matrix, cellular differentiation, intracellular signal- 11. Al Moustafa AE, Alaoui-Jamali MAA, Batist G, et al. Identification of genes as- ing, adhesion, and immune response were identified sociated with head and neck carcinogenesis by cDNA microarray comparison within this profile of 57 genes. TIMP-1, COL11A1, and between matched primary normal epithelial and squamous carcinoma cells. On- SERPINB2 were among the genes with the greatest de- cogene. 2002;21:2634-2640. 12. Mendez E, Cheng C, Farwell DG, et al. Transcriptional expression profiles of oral gree of differential expression between metastatic and non- squamous cell carcinomas. Cancer. 2002;95:1482-1494. metastatic tumors. Oral cavity/oropharynx SCC tu- 13. Giordano TJ, Shedden KA, Schwartz DR, et al. Organ-specific molecular classi- mors, which have a propensity to metastasize, remain a fication of primary lung, colon, and ovarian adenocarcinomas using gene ex- therapeutic challenge. The ability to develop a gene expression profiles. Am J Pathol. 2001;159:1231-1238. pression profile for metastatic tumors is an important first 14. Eisen MB, Spellman PT, Brown PO, Botstein D. Cluster analysis and display of genome-wide expression patterns. Proc Natl Acad Sci USA. 1998;95:14863- step because improved patient survival requires early de- 14868. tection of aggressive tumors, as well as targeted thera- 15. Quon H, Liu FF, Cummings BJ. Potential molecular prognostic markers in head peutic interventions. and neck squamous cell carcinomas. Head Neck. 2001;23:147-159.

(REPRINTED) ARCH OTOLARYNGOL HEAD NECK SURG/ VOL 130, MAR 2004 WWW.ARCHOTO.COM 301

©2004 American Medical Association. All rights reserved. Downloaded From: https://jamanetwork.com/ on 09/26/2021 16. Bradford CR, Carey TE. Molecular biology of head and neck tumors. In: Thawley 25. Alexander CM, Howard EW, Bissell MJ, Werb Z. Rescue of mammary epithelial SE, ed. Comprehensive Management of Head and Neck Tumors. Philadelphia, cell apoptosis and entactin degradation by a tissue inhibitor of metalloprotein- Pa: WB Saunders; 1999:321-342. ases-1 transgene. J Cell Biol. 1996;135:1669-1677. 17. Pennica D, Swanson TA, Welsh JW, et al. WISP genes are members of the con- 26. Guedez L, Stetler-Stevenson WG, Wolff L, et al. In vitro suppression of pro- nective tissue growth factor family that are up-regulated in wnt-1-tranformed cells grammed cell death of B cells by tissue inhibitor of metalloproteinases-1. J Clin and aberrantly expressed in human colon tumors. Proc Natl Acad Sci USA. 1998; Invest. 1998;102:2002-2010. 95:14717-14722. 27. Trask DK, Wolf GT, Bradford CR, et al. Expression of Bcl-2 family proteins in 18. Fischer H, Salahshor S, Stenling R, et al. COL11A1 in FAP polyps and in spo- advanced laryngeal squamous cell carcinoma: correlation with response to che- radic colorectal tumors. BMC Cancer. 2001;1:17. motherapy and organ preservation. Laryngoscope. 2002;112:638-644. 19. Stracke ML, Murata J, Aznavoorian S, Liotta LA. The role of the extracellular ma- 28. Pena JC, Thompson CB, Recant W, Vokes EE, Rudin CM. Bcl-KL and Bcl-2 ex- trix in tumor cell metastasis. In Vivo. 1994;8:49-58. pression in squamous cell carcinoma of the head and neck. Cancer. 1999;85: 20. Baker MS, Bleakley P, Woodrow GC, Doe WF. Inhibition of cancer cell urokinase 164-170. plasminogen activator by its specific inhibitor PAI-2 and subsequent effects on 29. O-charoenrat P, Rhys-Evans PH, Eccles SA. Expression of matrix metallopro- extracelluar matrix degradation. Cancer Res. 1990;50:4676-4684. teinases and their inhibitors correlates with invasion and metastasis in squa- 21. Nagayama M, Sato A, Hayakawa H, Urano T, Takada Y, Takada A. Plasminogen mous cell carcinoma of the head and neck. Arch Otolaryngol Head Neck Surg. activators and their inhibitors in nonsmall cell lung cancer. Cancer. 1994;73: 2001;127:813-820. 1398-1405. 30. Zeng ZS, Cohen AM, Zhang ZF, Stetler-Stvevenson W, Guillem JG. Elevated 22. Sumiyoshi K, Serizawa K, Urano T, Takada Y, Takada A, Baba S. Plasminogen tissue inhibitor of metalloproteinase 1 RNA in colorectal cancer stroma corre- activator system in human breast cancer. Int J Cancer. 1992;50:345-348. lates with lymph node and distant metastasis. Clin Cancer Res. 1995;1:899- 23. Nakamura M, Konno H, Tanaka T, et al. Possible role of plasminogen activator 906. inhibitor 2 in the prevention of the metastasis of gastric cancer tissues. Thromb 31. Ree AH, Florenes VA, Berg JP, Maelandsmo GM, Nesland JM, Fodstad O. High Res. 1992;65:709-719. levels of messenger RNAs for tissue inhibitors of metalloproteinases (TIMP-1 24. Stetler-Stevenson WG. The role of matrix metalloproteinases in tumor invasion, and TIMP-2) in primary breast carcinomas are associated with development of metastasis, and angiogenesis. Surg Oncol Clin N Am. 2001;10:383-392. distant metastases. Clin Cancer Res. 1997;3:1623-1628.

(REPRINTED) ARCH OTOLARYNGOL HEAD NECK SURG/ VOL 130, MAR 2004 WWW.ARCHOTO.COM 302