A Novel Nuclear Factor-κB Gene Signature Is Differentially Expressed in Head and Neck Squamous Cell Carcinomas in Association with TP53 Status

Purpose: To determine if gene signatures differentially expressed in head and neck squamous cell carcinomas (HNSCC) are related to alterations in transcription factors nuclear factor-κB (NF-κB) and TP53 previously associated with decreased cell death, response to therapy, and worse prognosis.

Experimental Design: Unique gene signatures expressed by HNSCC lines were identified by cDNA microarray, principal components, and cluster analyses and validated by quantitative reverse transcription-PCR (RT-PCR) and in situ hybridization. Bioinformatic analysis of the promoters and ontogeny of these clustered genes was done. Expression of proteins encoded by genes of a putative NF-κB signature, NF-κB p65, and TP53 were examined in HNSCC tissue specimens by immunostaining. Predicted promoter binding and modulation of expression of candidate NF-κB genes and cell survival were evaluated by p65 chromatin immunoprecipitation (ChIP) and small interfering RNA (siRNA) knockdown.

Results: Two groups of HNSCC exhibiting distinct gene signatures were identified: cluster A enriched for histone genes, with a higher prevalence of TP53 promoter binding motifs; and cluster B enriched for injury response genes with NF-κB regulatory motifs. Coexpression of cluster B proteins was observed with strong NF-κB phospho-p65 and weak TP53 staining, and NF-κB phospho-p65 was inversely associated with TP53 (P = 0.02). Promoter binding of the NF-κB signature genes was confirmed by p65 ChIP, and down-modulation of their expression and cell death were induced by p65 siRNA.

Conclusion: NF-κB promotes expression of a novel NF-κB–related gene signature and cell survival in HNSCC that weakly express TP53, a subset previously associated with inactivated wild-type TP53, greater resistance to chemoradiotherapy, and worse prognosis.

Development and malignant progression of cancer includes multiple steps and mechanisms, which are rarely due to a defect of a single gene or pathway. The global gene expression profile of certain cancer types or subsets has been analyzed in depth by advanced microarray technology and bioinformatics, revealing the identification of gene programs or signatures contributing to the heterogeneous characteristics and malignant phenotypes among different cancer samples, even of the same pathologic classification (1–3). Such clustering of genes in distinctive signatures may result from common epigenetic or genetic changes or alterations affecting activation of common signal pathways and transcription factors that regulate programmed gene responses.

In head and neck squamous cell carcinomas (HNSCC), gene expression profiling has been used in an attempt to identify biomarker signatures for diagnosis (4), differential sensitivity to chemotherapy (5), risk of recurrence (6), survival (7), malignant phenotype (8), and metastasis (9). Although considerable variability in the composition of gene signatures was observed in these studies, the data provided evidence for subsets and heterogeneity within HNSCC, which are possibly related to differences in molecular pathogenesis and malignant potential. However, the mechanisms underlying the regulation of common and different gene programs contributing to the malignant phenotype in HNSCC have not been well defined.

In murine and human squamous cell carcinoma (SCC), we and others have independently shown altered activation of transcription factors TP53, nuclear factor-κB (NF-κB), activator protein 1 (AP-1), and signal transducers and activators of transcription 3 (STAT3) and their functional roles in the expression of individual genes and phenotypic characteristics. Dysfunction of TP53 in HNSCC frequently results from mutation and is often associated with aberrant overexpression and immunostaining for TP53 protein, which occurs in ∼50% of HNSCC (10, 11). Inactivation of wild-type (wt) TP53 with weak expression and immunostaining is observed in most others, resulting from the inactivation of the p14ARF/p16INK4a locus regulating TP53 activation, or rapid protein degradation, from human papillomavirus (HPV) E6 gene expression (10, 12, 13). The mutation or loss of expression of TP53 has been implicated in a decrease in programmed cell death (14, 15). Interestingly, HNSCC retaining wt TP53 genotype but weakly expressing TP53 protein are associated with greater resistance to chemotherapy and/or radiation than those expressing mutant TP53 in clinical trials and studies in vitro (16, 17). However, the molecular basis for this observation and possible relationship to expression of various other genes previously associated with differences in malignant potential is not fully understood.

We discovered that NF-κB is constitutively activated in HNSCC and have shown that the activation of NF-κB/RelA (p50/p65) is one of the important factors controlling the expression of multiple genes that regulate cellular proliferation (Cyclin D1), apoptosis (cIAP1, Bcl-xL), angiogenesis (IL-8), immune and proinflammatory responses (IL-6), and therapeutic resistance of HNSCC and murine SCC (18–25). Variable increase in nuclear localization of NF-κB is also observed in tumors of ∼85% of patients with HNSCC, with stronger immunostaining associated with worst prognosis (26). NF-κB has become widely implicated in the pathogenesis of other diseases and cancers (27–32). However, the potential relationship between NF-κB regulated gene expression and TP53 status in HNSCC has not been previously reported.

Here, we examined global expression profile data for human HNSCC and primary keratinocyte lines by cDNA microarray and identified two subgroups of HNSCC cell lines with distinct gene expression signatures. Interestingly, bioinformatic analysis revealed that the subgrouping of the HNSCC cells was associated with differences in expression of known and putative NF-κB target genes. Known NF-κB target genes involved in cell proliferation, survival, and angiogenesis, such as IL-6, IL-8, and c-IAP1, were most strongly expressed in HNSCC cell lines with wt TP53 and tumor specimens with minimal or weak TP53 protein expression. Nuclear staining of the activated phospho-Ser⁵³⁶ form of p65 was inversely related to TP53 staining. These findings provide evidence that NF-κB activation and TP53 status are important determinants of distinct patterns of gene expression and linked in HNSCC. NF-κB and related genes could be important in the pathogenicity, resistance to chemoradiotherapy, and poor prognosis previously associated with the subset of HNSCC with impaired TP53 expression or function.

Cell lines and cell culture. Ten UM-SCC cell lines were from the University of Michigan series (Dr. T.E. Carey, University of Michigan, Ann Arbor, MI; Supplementary Methods and Supplementary Table S1). The cell lines were maintained in EMEM supplemented with 10% fetal bovine serum and penicillin/streptomycin. The TP53 mutation and expression status of UM-SCC cells lines were determined by bidirectional genomic sequencing of the exons 4–9 of TP53 and confirmed with immunocytochemistry using antibody of DO-1 (Supplementary Fig. S1; ref. 33). No mutation was detected in exons examined in four cell lines, UM-SCC 1, 6, 9, and 11A, and missense mutation of TP53 was detected in five cell lines, UM-SCC 5, 22A, 22B, 38, and 46 (Supplementary Fig. S1; ref. 33). A TP53 mutation was detected in UM-SCC 11B cells; however, immunocytostaining for TP53 protein indicated a weaker heterogeneous pattern of TP53 expression in UM-SCC 11B. The results of TP53 sequence analysis in UM-SCC 1, 5, 6, 11B, and 46 cells are consistent with those independently determined and previously reported by Bradford et al. (16). Human primary normal keratinocytes (HKC) derived from adult skin were obtained from Cascade Biologics Inc. and were maintained in keratinocyte serum-free medium 154CF containing 0.08 mmol/L Ca²⁺ with human keratinocyte growth supplements (HKGS). All HKC were used within five passages.

cDNA microarray and data analysis. Detailed methods for cDNA microarray and data analyses are recently published (34) and described in Supplementary Methods. Briefly, 50 to 100 μg of total RNA isolated using TRIzol reagent (Invitrogen) was reverse transcribed to fluorescence-labeled target cDNA with Cy3 or Cy5-dUTP (Amersham Pharmacia Biotech Inc.). Labeled targets were hybridized to 24K human array chips (National Human Genome Research Institute). Fluorescence images were captured by GenePix 4000 microarray scanner (Axon Instruments) and analyzed using the ArraySuite 2.1 extensions in IPLab program (Scanalytics, Inc.; ref. 35). Data analyses includes PCA analysis (Partek Pro 5.1, Partek Inc.), hierarchical clustering (36), Java Treeview (37), BRB-Array Tools,⁴

Gene Ontology, DAVID (Database for Annotation, Visualization, and Integrated Discovery, National Institute of Allergy and Infectious Diseases, NIH, Bethesda, MD), LocusLink [National Center for Biotechnology Information, NIH]. A mixed-model–based F test was used to analyze differential gene expression between different clusters (38, 39). This statistical model was carried out by using SAS 9.1 program (SAS Institute Inc.). P < 0.05 was designated as significant difference.

Real-time reverse transcription-PCR and in situ hybridization. Real-time reverse transcription-PCR (RT-PCR) and in situ hybridization for validation of microarray gene expression were done using standard methods as described in Supplementary Methods.

Bioinformatic prediction of promoter transcription factor binding motifs. To detect transcription factor binding sites in the promoter region on selected gene clusters, promoter analysis was done by Genomatix software suite 3.4.1 using the default matrix index. The proximal promoter sequences of each gene have, on average, ∼600 bp in length, located from ∼−500 bp to +100 bp of the transcription start site. We used all NF-κB binding matrices, including p65, p50, and c-Rel. The binding site of TP53 included the 20- to 21-bp decamer sequence. The predicted binding sites of AP-1 and STAT included the binding motifs for all the transcription factor members in the families. The statistical significance was examined by χ² analysis.

Immunohistochemical analysis of HNSCC tissue arrays and specimens, chromatin immunoprecipitation assay. Detailed information is described in Supplementary Methods.

Transient transfection with p65 RNAi plasmid or oligo nucleotides. Transient transfection of p65 RNAi plasmid (Imgenex) or p65 RNAi (SMARTpool, Dharmacom) was carried out using LipofectAMINE 2000 reagents according to manufacturer's protocol (Invitrogen). The transfection efficiency of plasmids was evaluated by RSV-LacZ construct (American Type Culture Collection). The effectiveness of p65 RNAi was confirmed in UM-SCC cells with high constitutive p65 activity by transient transfection, where p65 gene and protein expression and binding activity were decreased by more than 70% at 72 h after transfection (19).

Luciferase reporter gene assay. UM-SCC 1 and 9 cells were cotransfected with 5× NF-κB luciferase reporter plasmid (Clontech) and internal control plasmid RSV-LacZ using LipofectAMINE 2000 transfection reagent (Invitrogen) for 4 h in triplicates. Cell lysates were collected 48 h after transfection, and the luciferase activity was measured using Dual Light Reporter Gene assay (Tropix) on Monolight 2010 luminometer (Analytical Luminescence Laboratory). β-Galactosidase activities were determined to verify the reproducibility of each transfection experiment, and luciferase activities were normalized with β-galactosidase activity in each transfection experiment.

Gene expression profiling with principal components analysis and unsupervised hierarchical cluster analysis. Ten UM-SCC cell lines were used for gene expression profiling by cDNA microarray (Supplementary Table S1). Many of the molecular alterations and biological characteristics of these cell lines have been confirmed to reflect those identified in HNSCC tumors from patients in laboratory and clinical studies, including the role of activation of epidermal growth factor receptor (EGFR), interleukin 1 (IL-1) and IL-6 signal transduction pathways; altered activation of transcription factors TP53, NF-κB, AP-1, and STAT3; expression of cytokines and other genes; and radiation and chemosensitivity (5, 16, 20, 21, 23, 40–44). Nonmalignant human keratinocytes (HKC) used for comparison were from four donors, and these cells were harvested in the exponential growth phase to minimize the effects of differentiation. Gene expression profiles were obtained using a 24K cDNA microarray (National Human Genome Research Institute, NIH), and the expression of 9,273 of 12,270 evaluable known genes were submitted for principal components analysis (PCA; Fig. 1A). PCA classified the UM-SCC cell lines and HKC into two groups by component 1 on the X-axis and the UM-SCC lines into two subgroups by component 2 on the Y-axis (SCC_sub_1 and SCC_sub_2).

To investigate the critical genes contributing to the subgroups of cell lines, 942 genes were initially identified using at least a 2-fold difference in average expression ratio between UM-SCC and HKC (Student's t test, P < 0.01; refs. 45–48). This was followed by an additional global permutation test that excludes genes differing significantly due to chance alone. Genes that showed significant differences (P < 0.001) after 2,000 permutations were classified as differentially expressed genes. Using unsupervised hierarchical cluster analysis, three major hierarchical branches were identified, which included the group of four HKCs and two UM-SCC groups (Fig. 1B), each comprising the same cell lines that were segregated into subgroups by PCA. Of the 942 genes, 385 and 557 genes showed either increased or decreased expression, respectively, in UM-SCC cells when compared with that in HKC (Fig. 1B). The selected list of genes that are significantly up- or down-regulated in HNSCC cell lines are presented in Supplementary Table S2A and B. Two distinct gene expression patterns associated with each of the two subgroups of UM-SCC cells were observed and designated as cluster A (up-regulated in subgroup-1, including UM-SCC 5, 22A, 22B, 38, and 46) and cluster B (up-regulated in subgroup-2, including UM-SCC1, 6, 9, 11A, and 11B). Ontology annotation of the clusters A and B genes is presented in Supplementary Table S3. We used a mixed-model–based F test to examine statistical difference of gene expression among HKC and the subgroups of UM-SCC cells overexpressing clusters A and B genes in Fig. 1B. In cells expressing cluster A genes, a difference in expression with statistical significance with P < 0.0001 was observed when compared with expression in UM-SCC with the cluster B pattern or with HKC. In cells expressing cluster B genes, a difference in expression with P < 0.001 was obtained when compared with UM-SCC cells expressing cluster A or with HKC.

We noted that the segregation of UM-SCC lines into two equivalent subgroups approximates the frequencies of ∼50% each for TP53 inactivation and mutation reported for HNSCC (49, 50). TP53 gene mutation and protein expression in these cell lines were examined (Supplementary Fig. S1; ref. 33).⁴ Briefly, missense mutations in TP53 were detected in 6 out of 10 cell lines, and TP53 nuclear protein expression was detected by immunostaining in 5/6 of these UM-SCC cell lines with mutant TP53 status (Supplementary Fig. S1B). Thus, in most cases, the subgrouping of UM-SCC by gene expression profiling were associated with their TP53 status, with the exception of UM-SCC 11B line, which exhibited a more heterogeneous pattern of TP53 immunostaining, and clustered with wt TP53 UM-SCC, including UM-SCC 11A derived from the same patient. Analysis of the expression and function of wt and mutation TP53 in these UM-SCC lines are under way to determine the basis for the association of cluster A and B gene expression with TP53 status, and for exceptions, such as UM-SCC-11B.

Validation of gene expression by real-time RT-PCR and in situ hybridization. A set of differentially expressed genes was further validated by real-time RT-PCR, including eight up-regulated and four down-regulated genes in UM-SCC (Fig. 2A). Concordant data were observed between microarray and real-time RT-PCR, indicating the reliability of the microarray analyses. Notably, the expression of cluster B genes IL-6, IL-8, c-IAP1, and YAP1 were increased mainly in subgroup 2 of UM-SCC lines. Conversely, expression of cluster A gene HIST1H2BN was increased mainly in subgroup 1 of UM-SCC lines. Other up-regulated genes, such as the antiapoptotic genes BAG2, TRAF2, and CCNB2, as well as the down-regulated genes IL-1R2, CDKN1A, CCNG1, and CSF2, were detected in most of the UM-SCC lines by both methods (Fig. 2A).

We further validated four up-regulated genes in eight human HNSCC tissues and two matched normal mucosa tissues by in situ hybridization. We selected CCND1 and PCNA, which have previously been reported to be overexpressed in SCC (6, 18, 22, 51, 52), and BAG2 and CCNB2, which have been less studied (Fig. 2B). In each experiment, we included tumor tissues from UM-SCC 11A xenograft using antisense probe as a positive control (a), and sense probe as the negative control (b). On the top left, the BAG2 antisense probe strongly hybridized to UM-SCC 11A xenograft tumors (white arrows, a) and patient HNSCC tumor tissues (white arrows, c), but not on the matched normal mucosa tissues (e). Sense control probes did not hybridize to any of the tissues (b, d, and f). Similar hybridization patterns were observed when using PCNA, CCND1, and CCNB2 probes (Fig. 2B, remaining graphs). In d for CCND1 (top right) and CCNB2 (bottom right), we used antisense probes in a HNSCC tumor tissue with adjacent mucosa. Most of the hybridization signals were found in tumors but not normal mucosa (white arrows). Our data indicated that the expressions of these genes identified from microarray studies in cell lines are similarly expressed in human HNSCC tissues.

Increased prevalence of TP53 and NF-κB binding motifs within the proximal promoter regions of clustered genes. We hypothesized that the differences in clustered gene expression observed could be related to transcription factors previously shown to be altered in HNSCC, such as TP53, NF-κB, AP-1, and STAT3. To test this hypothesis, the proximal promoter sequences of the genes in both clusters were extracted and analyzed using algorithms by Genomatix Software Suite 3.4.1, which evaluates the average optimal length of ∼600 bp, from −500 to +100 bp of transcription start sites (Genomatix; Supplementary Table S3). The presence of putative recognition sites for the four transcription factors using this algorithm are summarized adjacent to the clusters in Fig. 3A and B, and the frequency of the predicted promoter binding sites were calculated and presented in Fig. 3C. In cluster A, binding sites for TP53 were present at increased frequency (9/17, 53%), when compared with genes in cluster B (6/27, 22%). Conversely, NF-κB binding sites (including all NF-κB family members) are highly represented in cluster B (20/27, 74%) when compared with cluster A genes. Interestingly, when we further analyzed the preferred binding motifs for different NF-κB subunits, a predominance of p65 binding motifs were observed in the promoters of cluster B genes, in contrast to the prevalence of p50 and c-REL binding motifs present in the promoters of cluster A genes with putative NF-κB binding sites.⁵

Significant differences in the prevalence of TP53 and NF-κB p65 binding sites were detected between clusters A and B (Fig. 3C, χ², P < 0.05). Consistent with promoter analysis, the genes in the cluster B include several genes previously shown to be regulated by NF-κB, such as IL-6, IL-8, c-IAP1, and ICAM1 (19, 23, 53–55). Higher frequencies of the binding motifs for AP-1 and STAT were also represented in cluster B genes, but did not reach statistical significance (Fig. 3C). The detailed analyses of predicted transcription factor binding sites of the clusters A and B from the up-regulated genes in UM-SCC cells are presented in Supplementary Table S3.

Immunostaining for putative NF-κB regulated proteins, NF-κB, and TP53 in HNSCC tissues. We recently confirmed that the UM-SCC cells in the panel with wt TP53 genotype stain weakly with anti-TP53 antibody DO-1 when compared with those with mutation TP53 (33). To explore the potential relevance of the apparent relationship between NF-κB regulated gene expression and TP53 status in UM-SCC cell lines to HNSCC tumors, we examined if a relationship exists between NF-κB p65 and TP53 staining patterns in fixed tumor tissues from patients with HNSCC of upper aerodigestive tract mucosal sites (oral cavity, pharynx, and larynx). Upon analysis of tissue microarrays of HNSCC and normal oral mucosa, 10 of 20 samples (50%) showed strong TP53 staining, and increased NF-κB p65 staining was observed in 14/20 (70%) of tumor samples, compared with normal mucosa (Fig. 4A). A total of 6/20 (30%) of the samples showed strong NF-κB staining with no detectable TP53 staining (top left); 8/20 (40%) showed some staining for both NF-κB p65 and TP53 (bottom left); 3/20 (15%) cases showed strong TP53 staining with no NF-κB p65 staining (top right); and three tumors (3/20, 15%) showed negative staining for both antibodies (data not shown). No or minimal staining for either factor was observed in six normal mucosa controls (bottom right). These observations suggested that a portion of HNSCC tumors exhibits a predominant immunostaining by either TP53 or NF-κB p65.

We further explored if a relationship exists between immunostaining for the activated Ser⁵³⁶ phosphorylated form of p65 RelA, expression of putative NF-κB regulated cluster B proteins, and TP53 in tumor tissues from patients with HNSCC. To optimally detect undenatured antigens recognized by antibodies to Ser⁵³⁶ phosphorylated form of p65 and cluster B proteins, we stained frozen HNSCC tissues from de-identified specimens from a series of 24 patients (including oral cavity, pharynx, and larynx). Again, a subset of HNSCC stained predominantly for either phospho-Ser⁵³⁶ NF-κB p65 (Fig. 4B) or TP53 (Fig. 4C). Strong nuclear NF-κB phospho-p65 staining was associated with co-staining for cluster B proteins c-IAP1, YAP, and CA9 (Fig. 4B). In contrast, HNSCC tumors with strong TP53 nuclear staining showed relatively weaker staining of phosphorylated NF-κB p65 and c-IAP1, CA9 proteins from cluster B (Fig. 4C). YAP protein was detected in this set of tumor sections, but with a different staining pattern (more cytoplasmic than nuclear staining; Fig. 4C). To quantify and determine the potential relationship of TP53 and phospho–NF-κB p65, we scored the intensity and percent positivity of the immunostaining and plotted and examined the results by linear regression and statistical analysis. An inverse and statistically significant correlation between increasing nuclear NF-κB phospho-p65 and TP53 staining was observed (Fig. 4D; R = −0.48 and P = 0.02, Spearman rank test).

Detection of predicted NF-κB binding sites in the promoters of cluster B genes by ChIP assay. To obtain evidence whether NF-κB p65 binds to predicted sites in the cluster B gene promoters (Fig. 2 and Supplementary Table S3), chromatin immunoprecipitation (ChIP) assay was done using anti–NF-κB p65 antibody with an isotype immunoglobulin G (IgG) as a negative control (Fig. 5). In Fig. 5, constitutive NF-κB p65 binding to the IL-8 promoter was observed, and treatment by tumor necrosis factor-α (TNF-α), a known stimulus for NF-κB binding activity, further increased the binding. The induction was greater in UM-SCC 11A cells, and the weakest effect was observed in HKC, consistent with our previous studies of IL-8 gene activation and expression patterns in these cell lines (19, 20, 23, 41, 44). Similar or weaker constitutive and TNF-inducible binding activities of NF-κB p65 to the IL-6, YAP1, or CA9 promoters were observed, with the induction being relatively stronger in UM-SCC 11A cells (Fig. 5). In each case, the binding activity with p65 antibody was greater than the background in negative control lanes using the isotype IgG.

Effects of modulation of NF-κB p65 on cluster B gene expression and cell death. To examine the regulatory role of NF-κB p65 activation on selected cluster B genes and their functional significance in HNSCC with undetectable expression of wt TP53 protein, we examined the effect of knocking down NF-κB p65 by RNAi on modulation of NF-κB transactivation, its downstream gene expression, and cell survival. As shown in Fig. 6A, a significant suppression of NF-κB reporter activity was observed in two wt TP53 cell lines (UM-SCC 1 and 9) transiently transfected with NF-κB p65 RNAi. The expression of three representative genes from each cluster was compared by real-time RT-PCR. Inhibition of NF-κB p65 activity resulted in significant modulation of the expression of cluster B genes IL-6, IL-8, and YAP1, but had no effect on the expression of the genes of cluster A (HIST1H2BD, HIST1H2BN, and CALR), consistent with the regulation of the former genes by NF-κB (Fig. 6B). Although the promoters of HIST1H2BD and CALR contain putative NF-κB binding motifs (Fig. 2A and Supplementary Table S3), both of them are preferred motifs for p50 but not p65 subunit of NF-κB.⁵

We next tested if knockdown of NF-κB p65 affects growth or the phenotype of HNSCC. Cells were transfected either with a control or NF-κB p65 RNAi, and morphology was monitored over 72 h. As shown in Fig. 6C, compared with the control-transfected cells, which exhibited a relatively low proportion of rounded and dead cells (top), cells transfected by NF-κB p65 RNAi showed increased cell round-up, membrane blebbing, fragmentation, and detachment, all indicative of cell death (bottom). Knockdown of NF-κB p65 by RNAi and induction of cell death in UM-SCC lines were confirmed by Western blot, DNA fragmentation, and DNA flow cytometry in independent experiments (19). Together, these data indicated that knocking down NF-κB p65 inhibited the expression of the cluster B genes examined and induced cell death, consistent with our previous data showing that inhibition of NF-κB by overexpression of a dominant negative IκBα or p65 small interfering RNA (siRNA) is associated with decreased cell survival of human or murine SCC (18–20, 22).

Previous molecular profiling studies of HNSCC and other cancers have linked distinct gene expression patterns to differences in malignant potential or response to therapy. However, the regulatory and functional relationships underlying different gene expression patterns have not been well defined. In this study, cDNA microarray expression profiling distinguished normal keratinocytes from malignant HNSCC cells and stratified HNSCC further into two subgroups, each exhibiting distinctive gene signatures (Fig. 1). A possible regulatory role of the interactions between NF-κB and TP53 was suggested to underlie the inverse expression of the two discrete gene signatures by bioinformatic analysis of transcription factor binding sites in the promoters of the two gene clusters (Fig. 3). In support of this hypothesis, we confirmed that increased staining for NF-κB p65 and its target gene products was inversely correlated with immunostaining for TP53 in subsets of HNSCC tissue specimens, indicating the relevance of the pattern and relationship observed in HNSCC lines (Fig. 4). NF-κB p65 binding to the promoters of sample genes in cluster B was detected by ChIP assay and was inducible by TNF-α (Fig. 5). We further confirmed that the expression of NF-κB signature genes and cell survival were inhibited by p65 RNAi (Fig. 6; ref. 19). These findings illustrate that bioinformatic analysis of gene promoter regulatory elements and ontology of clustered genes identified by genomic profiling can generate biologically meaningful hypotheses and reveal underlying molecular mechanisms that regulate the differentially expressed genes defining the malignant phenotypes in subsets of HNSCC.

The constitutive activation of NF-κB, inactivation or mutation of TP53, and their interaction in affecting the expression of NF-κB and TP53 regulated genes could represent important alternative mechanisms for malignant progression and help explain differences in pathogenicity, resistance to chemoradiotherapy, and poorer prognosis previously associated with the subset of HNSCC with inactivated TP53. With regard to NF-κB, we previously showed that NF-κB is aberrantly activated with malignant progression, modulates the expression of clusters of important oncogenes in murine SCC, and promotes survival, tumorigenesis, and therapeutic resistance of murine and human HNSCCs (18–25, 40, 56–58). Increased activation of NF-κB revealed by clustered coexpression of NF-κB related genes was also recently shown by array profiling in association with malignant progression in SCC arising from skin of α-catenin knock-out mice (59). Zhang et al. (26) confirmed in a clinical study that increased nuclear staining of NF-κB in tissue specimens is associated with the progression of premalignant lesions and decreased prognosis. Chung et al. (60) and Roepman et al. (9) recently identified some of the NF-κB regulated genes among gene signatures expressed by subsets of HNSCC at higher risk for recurrence and metastasis, which were also found in our previous murine (18, 22) and present gene profiling studies.

With regard to TP53 in this study, we recently have found that UM-SCC cells with wt TP53 genotype show lower TP53 RNA and protein expression with defective functionality (33).⁶

Hunter et al. recently published a study indicating that inactivation of the p16(INK4a) locus necessary for wt TP53 expression was associated with isolation of immortalized HNSCC lines and worse prognosis, whereas TP53 mutation was associated with isolation of mortal HNSCC lines and better prognosis (8). In their study, microarray analysis identified distinct gene expression profiles in the cell lines derived from premalignant squamous dysplasias and HNSCCs in association with these differences in phenotype, TP53 genotype, and protein expression. Interestingly, a review of their gene list reveals increased expression of some NF-κB regulated genes identified in our previous murine SCC and our present study in cluster B, in the subset associated with the inactivation of TP53 and greater malignant potential. In another study by Bradford et al. (16, 17), wt TP53 genotype with low TP53 protein expression has been associated with cisplatin resistance in vitro and chemoradiation resistance in patients. This was shown to be inversely related to the expression of BCL-xL (61), another gene which we have shown to be coregulated by NF-κB as well as STAT3 (19).⁷ Together, these observations are consistent with our hypothesis that differential expression of NF-κB regulated clustered genes associated with differences in phenotype and clinical outcome is linked to TP53 status in subsets of HNSCC.

The mechanism(s) underlying this apparent inverse relationship between expression of wt or mutation TP53 protein and NF-κB p65 nuclear localization and target genes in HNSCC cell lines and specimens remains to be determined. This relationship resembles that observed with wt TP53 in previous studies by Perkins (62, 63), who found that deficient expression of wt TP53 protein (TP53 null) is permissive for the transactivation of NF-κB target genes and cell survival in embryonic fibroblasts and sarcoma cell lines. Conversely, forced expression of wt TP53 protein inhibited transactivation by NF-κB of target genes and cell survival. They found that TP53 protein can compete for CBP/p300, resulting in the increased association of NF-κB with histone deacetylase and repression, with no reduction in NF-κB DNA binding activities (64). Conversely, strong activation of NF-κB was shown to have a reciprocal competing and repressive effect on TP53 stability and transactivation (65). Alternatively, Gurova et al. (66) reported that decreased expression of wt TP53 can result from NF-κB activation through a distinct but unknown mechanism in renal cell carcinoma cell lines. Whether TP53 inactivation or NF-κB activation are primary events and how NF-κB activation and its regulated cluster B gene expression are linked to TP53 in HNSCC is under investigation.

That p65 RNAi suppressed NF-κB reporter activity and the expression of cluster B genes including IL-6, IL-8, and YAP1 and induced cell death suggest potential functional relevance of cluster B genes (Fig. 6; ref. 19). These results are consistent with previous reports that link several NF-κB regulated genes identified in cluster B to malignancy, metastasis, resistance to chemoradiotherapy, and/or decreased survival. We previously showed that, in a murine SCC model, murine homologues of cluster B genes IL-8 (Gro-1), cIAP1 (cIap-1), and YAP1 (Yap-1) were coexpressed concomitantly with increased NF-κB activation and tumorigenic and metastatic potential, and they are down-modulated by the inhibition of NF-κB (18, 22, 57). Overexpression of Gro-1 alone promotes tumorigenesis and metastasis (58). A similar role for IL-8 has been shown in xenograft models of other human cancers (67). c-IAP1 is known to be an inhibitor of cell apoptosis and as a target gene regulated by NF-κB and TNF pathways (68). YAP1 has been reported to promote oncogenic activity, related to the interaction with TP53 family member p73 (69, 70). Both c-IAP1 and YAP1 are amplified and contribute to the development of hepatocellular carcinoma in murine models (71). The NF-κB regulated genes IL-6, IL-8, GRO-1, and/or c-IAP1 have also been detected as potential disease and prognostic markers in studies of tumor, serum, or saliva of HNSCC patients (72–74). IL-8 was detected by microarray in a subset of tumor samples from a series of 41 patients with recurrent HNSCC (6). Therapeutic resistance to chemotherapy agents paclitaxel and cisplatinum has been independently associated with gene expression of IL-6 and IL-8 or c-IAP1 and YAP1 (5, 75). We have shown increased IL-6 and IL-8 in cell line supernatants and serum from a subset of HNSCC patients and found that these cytokines increase in those resistant to chemoradiation therapy (42, 76). In a phase I clinical study, we observed the inhibition of activated nuclear phospho-p65, c-IAP1 in tumor, and IL-6 and IL-8 in serum in association with increased apoptosis and tumor reduction in a subset of patients given bortezomib, an inhibitor of proteasome-dependent activation of NF-κB (25). Together, these results provide evidence suggesting that cluster B genes are likely associated with NF-κB activation, progression, cell survival, and therapeutic response of SCC.

Less is known about the regulation, significance, and effects of increase in genes identified in the cluster A signature. TP53 binding motifs were present at a higher prevalence in the proximal promoter regions of cluster A genes, where histone H2A and H2B family genes are highly represented (Fig. 2 and Supplementary Table S3). A few other H2A and H2B family genes have been detected in HNSCC cells in prior microarray and proteomics studies (7, 60, 73, 77–79). However, little is known about the biological significance of increased H2A and H2B gene expression or the potential association with TP53 (Fig. 3). Previous studies indicate that repression of histone gene expression occurs with the activation of wt TP53 by UV radiation (80), whereas their expression is increased by the expression of the SV40 T antigen, which inactivates wt TP53 protein (81). It remains to be determined whether overexpression of cluster A genes in this subset of UM-SCC could result from altered TP53 function or activation or interactions with other TP53 family members or cofactors.

Bioinformatics analysis of the promoter binding motifs has been used mostly in the prokaryotic or lower eukaryotic organisms to identify transcriptomes, and limited utilities have been reported following the analysis of global gene expression in human cancers. Using this approach, we identified differences in the prevalence of TP53 or NF-κB binding motifs in the genes of prominent gene clusters. We found evidence for an inverse relationship between expression of these distinct clusters, inactivation or mutation of TP53, and NF-κB activation. A more comprehensive global and computational analysis of the gene expression profiles, binding motifs of promoters, and potential pathway networks have been recently carried out in the same set of microarray data from UM-SCC cells and presented elsewhere.⁸

Our findings provided evidence that bioinformatic analysis of gene promoter regulatory elements can reveal the critical control mechanisms governing the gene signatures identified from genome-wide gene expression profiling data, which could generate biologically meaningful hypothesis and clinically useful biomarkers to accelerate experimental investigation.

We are grateful to Drs. Thomas E. Carey and Carol R. Bradford (University of Michigan, Ann Arbor, MI) for providing UM-SCC cell lines and related clinical data; Dr. Paul S. Albert (Biometric Research Branch, National Cancer Institute/NIH) for his assistance in biostatistical analysis of the immunostaining data; and the Cooperative Human Tissue Network for providing the HNSCC tumor specimens. We express our gratitude to Drs. David Gius (NCI/NIH) and Paul Meltzer (NHGRI/NIH) for their critical reading and helpful comments.