Periostin, a Cell Adhesion Molecule, Facilitates Invasion in the Tumor Microenvironment and Annotates a Novel Tumor-Invasive Signature in Esophageal Cancer

The development of esophageal squamous cell carcinoma (ESCC) involves a multistep, progressive process starting with increased esophageal epithelial cell proliferation leading to basal cell hyperplasia, dysplasia, carcinoma in situ, and, finally, advanced invasive carcinoma (1–3). Several genetic alterations, such as amplification of the epidermal growth factor (EGF) receptor (EGFR), dysregulation of cyclin D1, and somatic mutations in the DNA binding domain of the tumor suppressor p53, are involved in initiation and progression of ESCC (4). Although EGFR overexpression and p53 inactivation are key genetic alterations associated with ESCC, how these genetic alterations contribute to ESCC progression remains to be elucidated. Previously, we had addressed this by modeling EGFR overexpression and p53 missense mutation (R175H) in primary human esophageal epithelial cells (EPC2), which have been immortalized by hTERT overexpression (designated as EPC2-hTERT-EGFR-p53^R175H cells). EPC2-hTERT-EGFR-p53^R175H cells were grown in three-dimensional organotypic culture, resulting in the invasion of these cells into the underlying extracellular matrix (ECM) compared with EPC2-hTERT-EGFR or EPC2-hTERT-p53^R175H cells, which did not invade (5). Combined expression of these genes also resulted in anchorage-independent growth and in tumor formation in xenograft models, which was not observed in control cells overexpressing either EGFR or mutant p53 alone (5).

Recent experimental results have provided mounting evidence that altered expression of cell adhesion molecules can contribute directly to tumor progression by modulating cell signaling. Therefore, we sought to identify genes involved in cell adhesion that were differentially expressed in invading tumor cells and could shed new insights into processes affecting tumor progression.

Deriving a novel invasive tumor signature from a gene expression profile analysis of invading EPC2-hTERT-EGFR-p53^R175H cells in three-dimensional culture and human ESCC tumor microarrays, we have identified periostin to be the most significantly upregulated gene triggering tumor cell invasion. Periostin (POSTN) is a secreted 90-kDa protein that was identified originally as a bone adhesion molecule responsible for differentially recruiting and attaching osteoblasts to the periosteum (6). Periostin is a member of the fasciclin family and has a NH₂-terminal signal peptide sequence, a cysteine-rich domain, four internal homologous repeats, and a hydrophilic COOH-terminal domain (6). These four internal repeat domains share structural homology with Fasciclin 1, an insect neuronal adhesion protein (6, 7), and big-h3, a transforming growth factor-β1 (TGFβ1)–inducible gene (7, 8). The high degree of structural and sequence similarity with these cell adhesion molecules suggests that periostin has a role in cell adhesion and migration. Indeed, not only is its expression induced in human ESCC and mouse tumor xenografts, but more importantly, genetic modulation of periostin also affects tumor invasion in a direct fashion. The novel identification of periostin as an essential mediator of tumor invasion in the microenvironment has implications upon tumor detection and therapy.

Primary esophageal cells (EPC2) were established from surgical specimens of normal esophagus as previously described (9) and immortalized by overexpression of the catalytic subunit of telomerase (hTERT; EPC2-hTERT cells; ref. 10). All cells were maintained in keratinocyte-SFM medium (Invitrogen) supplemented with 40 mg/mL BPE, 1.0 ng/mL EGF, 100 U/mL penicillin, and 100 mg/mL streptomycin (full keratinocyte-SFM medium). Cells were grown at 37°C in 5% CO₂.

The PFB-neo or pBABE-zeo retroviral vectors were used to overexpress human EGFR in EPC2-hTERT cells. Additionally, p53^R175H was subcloned into the pBABE-puro or the pBABE-zeo retroviral vectors. Periostin cDNA was used (Open Biosystems) and subcloned into the pFBneo retroviral vector. Two periostin short hairpin RNAs (shRNA#1 5′-CCCATGGAGAGCCAATTAT-3′and shRNA#2 5′-CTCTGACATCATGACAACAAAT-3′) were used (Open Biosystems). Plasmids were transfected into Phoenix-Ampho packaging cells (gift of G. Nolan, Stanford University, Palo Alto, CA) using the Lipofectamine 2000 reagent (Invitrogen). Supernatants of the retroviruses encoding EGFR, mutant p53, periostin overexpression, and shRNA vector constructs were collected 48 and 72 hours after transfection. Subconfluent EPC-hTERT cells were infected with retrovirus supernatant in 4mg/mL polybrene (Sigma). Forty-eight hours after infection, cells were selected in 300 μg/mL G418, 0.5 μg/mL puromycin, 2 mg/mL zeocin, or 5 mg/mL blastocidin for 5 days. Overexpression of EGFR, mutant p53, and periostin, as well as periostin knockdown was confirmed by Western blot analysis.

Keratinocytes were grown in organotypic culture to recreate their microenvironment by supplying ECM components including collagen, and laminin and fetal esophageal fibroblasts; its detailed procedures were previously described (5). Cells grown in organotypic culture were processed for histology by fixing in 10% formaldehyde and were paraffin embedded for laser capture microdissection (LCM) studies by embedding in OCT medium followed by freezing in liquid nitrogen and then stored at −80°C.

Frozen organotypic cultures were sectioned to 8 μm onto membrane-mounted metal frame slides (MMI) using a Microm HM 505E cryostat (Richard Allen Scientific). Sections were immediately fixed, stained, and dehydrated before laser microdissection. Microdissection was performed with a Nikon Eclipse TE 2000-5 microscope with a UV laser (MMI). Microdissected cells were collected, and 50 μL of extraction buffer (Arcturus) were added and incubated for 30 minutes at 42°C, followed by centrifugation at 800 × g for 2 minutes. RNA was extracted using the Arcturus PicoPure RNA isolation kit following the manufacturer's instructions. The integrity and concentration of the RNA was assessed using the Agilent Bioanalyzer 2100 (Agilent Technologies).

The RNA obtained from the LCM studies was amplified using the Affymetrix GeneChip Expression 3′-Amplification Two-Cycle cDNA Synthesis kit (Affymetrix) followed by the Affymetrix GeneChip Hybridization Wash and Stain kit (Affymetrix). The resulting cRNA (200 ng) was used as template for random-primed cDNA synthesis and a second round of in vitro transcription, which incorporated biotinylated CTP and UTP. The cRNA products were fragmented to 200 bp or less, heated at 99°C for 5 minutes, and hybridized for 16 hours at 45°C to Affymetrix U133Plus 2.0 oligonucleotide microarrays (Affymetrix). Microarrays were subsequently washed at low (6× saline-sodium phosphate-EDTA) and high (100 mmol/L MES, 0.1 mol/L NaCl) stringency, and stained with streptavidin-phycoerythrin. The fluorescence signal was amplified by the addition of biotinylated anti-streptavidin and an additional aliquot of streptavidin-phycoerythrin stain. A confocal scanner was used to acquire the fluorescent signal after excitation (570 nm).

LCM was repeated to isolate invading and noninvading EPC-hTERT-EGFR-p53^R175H cells grown in organotypic culture. Amplification and cDNA synthesis was performed using WT-Ovation RNA Amplification System (NuGen Technologies) according to the manufacturer's instructions. Real-time PCR was performed and analyzed using the ABI PRISM 7000 sequence detection system software (PE Applied Biosystems) and using the Power SYBR Green PCR Master Mix (PE Applied Biosystems) for β-actin according to the manufacturer's instructions. Taqman assays for periostin were run using the Taqman Universal PCR Master Mix (PE Applied Biosystems) according to the manufacturer's instructions.

Esophageal tumor tissue specimens and adjacent normal tissue were surgically procured from patients at the Okayama University Hospital (Drs. Shirakawa and Naomoto, Japan). All tumor specimens were pathologically diagnosed as ESCCs (grade III) and obtained from patients who signed informed consents in accordance with Institutional Review Board standards and guidelines.

Specimens were immediately snap frozen for RNA and protein analyses. The human ESCC tissue microarray was subjected to immunohistochemistry analysis using a polyclonal anti-periostin antibody and scored for periostin expression as follows; negative (0), marginal considered negative (0.5), mild positive stain (1), moderate positive stain (2), and intense positive stain (3). Scores of >0.5 are considered positive. Each case on the tissue microarray comprises of two cores and the mean scores of two cores were taken.

The following antibodies were used for immunoblotting: EGFR (NeoMarkers, Ab-12), p53 (Oncogene Research Products, Ab-6), phospho-EGFR (Cell Signaling, Tyr 1068), rabbit polyclonal periostin (Abcam, ab 14041), and p21 (Oncogene Research Products), WAF1 (ab-1); β-Actin (Sigma) was used as a loading control. For immunohistochemistry, periostin (Abcam, ab 14041) was used.

To confirm protein expression, Western blot analysis was performed as previously described (5). To confirm secreted periostin expression, conditioned media was collected, protein concentration was determined by Bio-Rad protein assay (Bio-Rad), and Western blot analysis was performed as described above.

Migration and invasion assays were performed as previously described (5). All experiments were performed at least three times in triplicate.

Affymetrix array cel files were processed in R using an algorithm that uses genes in the least variant set to normalize the expression data (11). In the analysis of invasive and noninvasive cells, we then identified 6,106 probe sets for subsequent statistical analysis that had a minimum expression level (expression of >100 in all samples) and that had significant variability among the 11 cell lines (SD was >150). The background correction, summarization, and normalization of Affymetrix cel files were done in the same manner independently for each of the two tumor/normal studies. Gene Expression Omnibus database accession number is GSE21293.

All statistical analyses used natural log-transformed expression data. Using the arrays of cells from the organotypic culture model, a multivariate regression model was fit for each gene that included two independent binary variables. The first variable indicated cells with both EGFR overexpression and P53 mutation only and the second indicated cells with EGFR overexpression, P53 mutation, and invasion. The third group (the reference group) included cells with a noninvading phenotype that had either EGFR overexpression or P53 mutation. The P values from the test that the regression model was significant (F-statistic, 2 degrees of freedom) were adjusted using the Benjamini-Hochberg technique; probe sets with a false discovery rate of <0.05 were selected for further investigation (12). Genes on this list were compared with annotated gene ontology (GO) biological processes, and a functional annotation cluster analysis was done using the online DAVID tool (http://niaid.abcc.ncifcrf.gov/). Annotations were derived from the DAVID databases. Supervised hierarchical clustering used in the heat maps was based on the Euclidean distance. The statistical analyses were done using the SAS/STAT software version 9.2 of the SAS System for personal computer.

The molecular pathways and networks underlying tumor invasion remain largely to be elucidated. To characterize a gene expression profile underlying the invasive phenotype of invading EPC2-hTERT-EGFR-p53^R175H cells, we profiled mRNA obtained from laser capture–microdissected EPC-hTERT cells overexpressing EGFR and with a p53 missense mutation (R175H) that was either invasive or noninvasive in the organotypic culture model, as well as noninvasive cells that either overexpressed EGFR alone or had p53 (R175H) mutation alone, using Affymetrix U133-2 plus microarrays.

Analysis of the gene expression microarray data revealed a tumor-invasive signature corresponding to a probe set list of 939 probe sets, in which there was significant differential expression of these probe sets in invasive versus noninvasive cells (Supplementary Table S1). A supervised hierarchical cluster analysis showed that this tumor-invasive signature could distinguish significant differences between cell lines in organotypic culture with and without an invasive phenotype as presented in a heat map (Fig. 1A).

To determine whether this invasive signature derived from invading EPC2-hTERT-EGFR-p53^R175H cells grown in organotypic culture can be extended to human ESCC, we analyzed gene expression profiles from five paired human ESCCs. We found that overall gene expression patterns of invasive cells grown in organotypic culture are highly similar to those from ESCC tumors, and this tumor-invasive signature was able to discriminate esophageal tumors from adjacent normal tissue and classified all tumors and 80% of the normal samples correctly (Fig. 1B). To validate this tumor-invasive signature, we performed a secondary analysis on an independent cohort of five paired human ESCC tumors using this tumor-invasive signature and observed that it was capable of classifying 80% of ESCC tumors and all adjacent normal tissues correctly (Fig. 1C). Using the t test statistic to rank the probe sets in this tumor-invasive signature, we identified periostin, a cell adhesion molecule, with the most significant difference in gene expression between invading transformed esophageal epithelial cells compared with noninvading normal or dysplastic epithelial cells (t test for two independent groups, P < 0.0001). Strikingly, periostin was also found to be significantly upregulated in both independent cohorts of human ESCC tumors compared with adjacent normal tissue; the fold change in mean periostin expression was found to be 5.0 and 5.7 in the respective cohorts.

To explore the biological processes associated with invasion of ESCC, we tested for enrichment of GO terms within this tumor-invasive signature. This tumor-invasive signature was found to be significantly enriched in nine biological processes (false discovery rate, P < 0.01; data not shown; Supplementary Table S1 and S2). To investigate the cellular processes altered within the tumor microenvironment, we selected three biological processes for further investigation and observed that probe sets clustered in each biological process (cell/biological adhesion, development-related and cellular component organization, and biogenesis) were able to classify invading transformed esophageal epithelial cells from noninvading cells (Supplementary Fig. S1A–C; Table 1). Interestingly, periostin was found to be the highest upregulated gene (6.54-fold) across the three biological processes (Table 1) and was unique in this feature among all the genes in the data sets. In aggregate, these results reveal a tumor-invasive signature derived from the organotypic culture model, which applies faithfully to human ESCC and highlights periostin as an important gene required for facilitating tumor cell invasion.

We next sought to validate whether the upregulation of periostin is concomitant with invading transformed esophageal cells as well as establishes the localization of periostin expression in these cells. Consistent with the results of the microarray analysis, increased periostin mRNA expression in invading EPC2-hTERT-EGFR-p53^R175H cells was confirmed by LCM isolation of invading and noninvading EPC2-hTERT-EGFR-p53^R175H cells in the organotypic culture model followed by RNA isolation, amplification, and quantitative PCR analysis (Fig. 2A). Furthermore, periostin protein accumulation in invading EPC2-hTERT-EGFR-p53^R175H cells grown in organotypic culture was observed by immunohistochemical staining (Fig. 2B, right, b) but not in control EPC2-hTERT-EGFR cells (Fig. 2B, left, a). Interestingly, periostin protein expression was also observed in the epithelial-stromal interface in tumors formed by EPC2-hTERT-EGFR-p53^R175H cells in a xenograft tumor model (Fig. 2C, white arrows). These results establish the preferential expression of periostin in invading EPC2-hTERT-EGFR-p53^R175H cells both in vitro and in vivo.

Previous efforts involving other epithelial cancers had indicated that periostin promotes invasion and anchorage-independent growth in several epithelial cancer cell lines and in tumors such as head and neck cell carcinoma, oral, breast, and ovarian cancers (13–16). Elevated levels of periostin have been detected in the sera of patients with breast, thymoma, and non–small cell lung cancer, suggesting periostin secretion during tumorigenesis (17–19). We sought to investigate directly whether periostin had a functional role in facilitating tumor cell migration and invasion in ESCC. An RNA interference approach with shRNA was used to induce stable knockdown of periostin expression in EPC2-hTERT-EGFR-p53^R175H cells (Fig. 3A). Knockdown of periostin in EPC2-hTERT-EGFR-p53^R175H cells led to significant decrease in cell migration and invasion compared with control shscrambled cells (Fig. 3B and C). Notably, reduced periostin expression in EPC2-hTERT-EGFR-p53^R175H cells also resulted in decreased invasion of these cells into the underlying matrix in organotypic culture (Fig. 3D).

In parallel studies, periostin was retrovirally overexpressed in two independent cell lines (EPC2-hTERT-EGFR-p53^R175H-1 and EPC2-hTERT-EGFR-p53^R175H-2; Fig. 4A), and although overexpression of periostin showed no effect on proliferation (data not shown), we observed that these cells displayed increased migration and invasion (Fig. 4B and C). Furthermore, when EPC2-hTERT-EGFR-p53^R175H cells overexpressing periostin were grown in organotypic culture, increased invasion into the underlying matrix was also observed (Fig. 4D). Collectively, these results show the direct functional role of periostin in promoting tumor cell migration and invasion, particularly within the context of the tumor microenvironment.

We next sought to determine if the upregulation of periostin is also found in human ESCC by assessing periostin expression in primary ESCC (grade III) tumors. Increased periostin mRNA expression was observed in ESCC tumors compared with their matched normal controls (Fig. 5A). A marked increase in periostin protein levels was observed in all ESCC tumors studied compared with their matched normal mucosa controls (Supplementary Fig. S2; Fig. 5B). To determine the localization of periostin in human ESCC, further immunohistochemical analysis of periostin expression was performed in a tissue microarray containing 73 ESCC tumors and adjacent normal tissue, and revealed periostin expression in invasive ESCC tumor cells (Fig. 5C, bottom, c, arrowheads) and consistent accumulation in tumor stroma (Fig. 5C, bottom, c, arrows) compared with normal esophageal tissue in which periostin expression was detected primarily in and around blood vessels (Fig. 5C, left, a, arrows). Periostin expression was also detected in high-grade esophageal intraepithelial neoplasia, which was found in a restricted number of cases, particularly in the epithelial-stromal interface (Fig. 5C, right, b, arrows). In addition, the stroma in all 73 ESCC cases were scored for periostin immunohistochemical staining intensity and it was shown to have higher levels of periostin staining compared with matched normal esophagus, which showed weak to no periostin staining. Specifically, the ESCC scored either 2 or 3, meaning moderate or intense staining, whereas the normal esophagus was 0 or 0.5 (Fig. 5C, d), which would suggest that induction of periostin could also arise in the stroma during ESCC progression.

It is believed, but not established, that periostin might enhance tumor cell invasion and metastasis through increased integrin signaling, augmenting cell survival through the phosphoinositide 3-kinase (PI3K)/Akt pathway or fostering epithelial-to-mesenchymal transition (16, 20, 21). Given that periostin was identified initially in invading EPC-hTERT-EGFR-p53^R175H cells through functional genomics and bioinformatic analysis, we hypothesized that periostin expression is dependent on both activated EGFR signaling and p53 mutation. Notably, periostin expression was upregulated in EPC2-hTERT-EGFR-p53^R175H cells compared with control cells overexpressing EGFR alone (EPC2-hTERT-EGFR) or mutant p53 alone (EPC2-hTERT-p53^R175H; Fig. 6A). In addition, periostin promoter reporter assays in EPC2-hTERT-EGFR-p53^R175H cells displayed the highest promoter activity compared with control cells with either EGFR overexpression or p53 mutation alone (Supplementary Fig. S3), suggesting that both genetic alterations may be required to activate periostin expression at a transcriptional level. This result was further corroborated when EPC2-hTERT-EGFR-p53^R175H cells were stimulated with EGF and increased periostin protein expression was observed (Fig. 6B). This induction of periostin upon EGFR stimulation led us to test whether the inhibition of EGFR signaling and/or restoration of wild-type p53 function could inhibit periostin by treating EPC2-hTERT-EGFR-p53^R175H cells. We used AG1478, a EGFR tyrosine kinase inhibitor, and 5-iminodaunorubicin, a small-molecule compound that restores wild-type p53 signaling by inducing apoptosis and cell cycle arrest, as illustrated by the induction of p21(22). Periostin protein expression was noted to be decreased when inhibited by AG1478 or 5-iminodaunorubicin or both (Fig. 6B). Taken together, these data support the notion that periostin expression is modulated mechanistically by activated EGFR signaling and p53 mutation.

Overall, our results reveal a novel tumor-invasive signature derived from invading transformed cells compared with control noninvading normal and dysplastic cells in organotypic (three-dimensional) culture. This signature is one that annotates primary invasive human esophageal squamous cell cancer as distinctive from adjacent normal human esophageal mucosa in two independent cohorts of tumors, underscoring the fidelity and utility of this tumor-invasive signature. It is conceivable that this molecular signature might be informative in the future for other squamous cell cancers arising in different tissues.

GO analysis (DAVID databases) of specific biological processes that are believed to be involved in the tumor microenvironment reveals the consistent and unique upregulation of periostin, suggesting its critical functional role. First, periostin was found to be upregulated in invading transformed cells and in primary esophageal squamous cell cancers based on immunohistochemical and Western blot analysis. Second, the direct functional role of periostin is underscored by overexpression and knockdown experiments in which genetic manipulation of periostin dramatically influences the degree of tumor invasion in organotypic culture. Third, EGFR signaling and p53 mutation, both canonical genetic alterations in ESCC, seem to converge upon periostin based on luciferase reporter gene assays as well as inhibition of EGFR signaling and restoration of wild-type p53 function. In aggregate, these novel results underscore the utility of the organotypic culture model for the discovery of direct biological effectors of tumor invasion into the microenvironment, which has been largely elusive to date. Local tumor invasion in the mesenchymal stromal compartment is important given that it temporally precedes tumor dissemination in the lymphatic and blood vessels for tumor metastasis.

Various avenues of investigation have highlighted the important role of the microenvironment in enhancing the initial dissemination of malignant tumor cells. Dynamic interactions between the epithelium and mesenchymal stroma contribute to boosting the invasive phenotype of tumor cells by activating a variety of genes facilitating cell proliferation, dedifferentiation, migration, and invasion (23, 24). Significant changes such as loss of cell-cell contacts, disruptions in cell adhesion junctions, and altered cell ECM interactions within the tumor microenvironment converge to increase the ultimate metastatic potential of tumor cells (25). Therefore, identification of gene expression pattern changes during initial stages of tumor progression within the tumor microenvironment is crucial to understanding the causes of tumor invasion and, ultimately, tumor metastasis to distant organ sites. Thus, our results provide new platforms into the investigation of tumor invasion.

Periostin has been shown also to have a role in bone, tooth, and heart formation during development (26, 27) and is only reexpressed and upregulated in adult tissue after vascular, skeletal, or bone injuries (28, 29). Periostin is similarly overexpressed in human cancers (14, 30). Our results highlight that a tumor-invasive signature defines genetically engineered ESCC that is reproduced in human ESCC. Our studies also suggest that induction of periostin, a secreted protein with a long half-life (data not shown; Supplementary Fig. S4), may alter the tumor microenvironment by accumulating in the stroma and facilitating invasion through matrix remodeling. This may be achieved through the regulation of collagen I fibrillogenesis (31), serving as a bridge between tenascin C and the ECM, as well as through interaction with α_Vβ₃ integrins. Furthermore, periostin may promote tumor cell survival in the matrix of the microenvironment by activating the Akt/PI3K pathway. Lack of periostin may lead to the suppression of Notch1 signaling (32); conversely, it is conceivable that overexpression of periostin could activate Notch1 signaling in cancers. Indeed, we have evidence of activated Notch signaling in invasive tumor cells grown in organotypic culture (data not shown). Future translational and clinical investigations might seek to exploit periostin as an attractive therapeutic target, especially in a combinatorial fashion, for example, with inhibitors of receptor tryosine kinases.

No potential conflicts of interest were disclosed.

We thank the Morphology Core (S. Mitchell, D. Budo, and G. Swain), the Molecular Biology/Gene Expression Core (G. Wu and S. Keilbaugh), and the other members of the Rustgi laboratory for the helpful discussions.

Grant Support: NIH/National Cancer Institute P01-CA098101 (C.Z. Michaylira, G.S. Wong, A.J. Klein-Szanto, J.A. Diehl, M. Herlyn, H. Nakagawa, R. Hammond, P. Gimotty, and A.K. Rustgi), NIH/NRSA F32DK-075230 (C.Z. Michaylira), NIH T32-CA115299 (G.S. Wong), and NIH/National Institutes of Diabetes, Digestive and Kidney Diseases Center for Molecular Studies in Digestive and Liver Diseases (P30-DK050306).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.