Oral squamous cell carcinoma (OSCC) is one of the most prevalent malignancies worldwide, and the third most common cancer in developing nations.1 Effective management requires the ability to predict the natural history of individual lesions at molecular level,2 whereas molecular mechanisms responsible for the development of oral cancer are still largely unknown. Traditional gene-targeted studies that focus on one or a few genes at a time cannot provide insight into the changes in global gene expression associated with carcinogenesis. Recent developments of microarray techniques make it possible to compare relative expression levels of several thousand genes simultaneously in a single experiment.3 A large number of articles have been published on the use of microarray in the study of OSCC,4–8 which extracts differentially expressed genes from clinical OSCC specimens compared with normal tissues. However, the published microarray data, though analyzed through a variety of statistical strategies, were divergent from each other. One reason may be that it is difficult to control the heterozygosity between the clinical specimens, and oral cancers with different etiological backgrounds which can be distinguished by their different global gene expression patterns.9
Though the etiological factors of OSCC vary in different parts of the world, the putative high risk factors are environmental factors including continued use of tobacco.10 Benzo (a) pyrene (B (a) P) is the most important carcinogen among the components of tobacco. Studies on the function of B (a) P in carcinogenesis focused on cellular metabolism in response to B (a) P in the short term.11,12 However, the downstream events of B (a) P metabolites should be the main process in carcinogenesis, and only a few studies have considered this problem. Park et al13 have established a tumorigenic cell line by exposing human papillomavirus type 16 (HPV16)-immortalized oral keratinocytes to B (a) P. Rey et al14 examined the changes in gene expression of this malignant cell line compared with immortalized oral keratinocytes using representational difference analysis, whereas this study did not provide us with a global gene expression profile associated with tumorigenesity.
Recent development of microarray techniques has made it possible to provide insights into the changes in global gene expression associated with carcinogenesis induced by various etiological factors. The goal of the present study was to characterize the global expression profiling associated with the tumorigenesity of oral epithelial cells, and to search for the candidate genes which may play roles in the initiation of oral cancer related with HPV infection and B (a) P exposure.
Cell culture and tumorigenesity analysis
Human immortalized oral epithelial cells (HIOEC) were established from normal oral mucous epithelial cells through HPV16 E6 and E7 genes transfection,15 and this cell line was not tumorigenic in nude mice. However, tumorigenic cells were induced by B (a) P.16 Briefly, HIOEC cells were treated with B (a) P (Sigma, USA) dissolved in dimethyl sulfoxide for 6 months with an increasing concentration from 0.1 mg/L to 1.2 mg/L gradually, and the HIOEC cells treated with B (a) P underwent clone screening at the 18th passage. Then the cells were injected into nude mice subcutaneously to examine their tumorigenesity, and were named HIOEC-B (a) P (HB) cells. The tumor tissues were examined histopathologically. The 55th passage (HB-55p) formed tumors first, which contained mainly moderately differentiated keratinocytes; tumors formed from the 94th (HB-94p) passage were poorly differentiated and similar to typical squamous cell carcinoma in histopathology.
HIOEC cells and HB cells before the 21st passage (HB-21p) were cultured in serum-free medium (Gibco, USA), and the HB-21p cells were cultured in Dulbecco's modified Eagle's medium supplemented with 10% fetal bovine serum (Hyclone, USA).
Total RNA was extracted from HIOEC cells, HB-56p and HB-96p cells, using TRIzol reagent (Sigma, USA), followed by chloroform extraction and isopropyl alcohol precipitation. Ethanol washed total RNA pellets were resuspended in RNase-free water and passed through an RNeasy spin column (Qiagen, USA) for further purification. Eluted total RNA was quantified with a portion of the recovered total RNA adjusted to a final concentration of 1 μg/μl. Quality of all starting total RNA samples was assessed prior to target preparation and processing steps.
Target labeling and microarray processing
The isolated total RNA samples were processed as recommended by the Affymetrix Gene Chip Expression Analysis Technical Manual. In brief, single-stranded and then double-stranded (ds) cDNA was synthesized from the poly(A)+ mRNA present in the isolated total RNA (10 μg total RNA starting material for each sample reaction) using the Affymetrix one-cycle cDNA Synthesis Kit (Affymetrix, USA) and poly (T)-nucleotide primers that contained a sequence recognized by T7 RNA polymerase. A portion of the resulting ds cDNA was used as a template to generate biotin-tagged cRNA from an in vitro transcription reaction, using the GeneChip IVT Labeling Kit (Affymetrix, USA). Fifteen micrograms of the resulting biotin-tagged cRNA was fragmented to strands of 35–200 bases in length following prescribed protocols. Subsequently, 10 μg of this fragmented target cRNA was hybridized at 45°C with rotation for 16 hours to probes present on an Affymetrix U133 plus 2.0 arrays. The GeneChip arrays were washed and then stained in an Affymetrix Fluidics Station 400, followed by scanning on a Hewlett-Packard GeneArray scanner. The hybridizations were replicated in triplicate for each kind of cell. The scanned results were quantified and normalized from DAT files to CEL files using GCOS software (Affymetrix, USA).
For interpretation of microarray data, it was assumed that genes with fold-changes <2 were of negligible biological significance. Genes with fold-changes >2 between the HIOEC, HB-56p and HB-96p cells were considered to be differentially expressed. Scatter plot analysis was an important visual algorithm and was used to display the distribution of differentially expressed genes. Clustering analysis was performed to collect genes that showed similar expression changes.
Gene ontology (GO) annotation
Genetools is a web interface (http://www.genetools.microarray.ntnu.no/egon) which carries out simple data mining using GO for DNA microarray data. The data mining consists of the assignation of the most characteristic GO term to each gene. GO terms are related to human genes and proteins. A gene product can involve one or more biological processes and may be associated with one or more cellular components. When we searched for the distribution of genes in a specific ontology (e.g. cellular component), we selected the ontology and also the GO term's level (=5).
Clinical OSCC specimen treatment
Oral squamous cell carcinoma of the tongue, gingival and buccal regions and the corresponding normal wound marginal tissues were acquired from surgical operations with the approval of the concerned patients. Cancerous tissue was cut into halves, and total RNA was extracted from one half, the counterpart of which was observed under a microscope to assure at least 80% malignant component. The normal epithelial tissues were obtained from wound marginal tissues of the same patient through Dispase II (Gibco, USA) treatment. The total RNA was extracted using TRIzol (Sigma, USA) as described in RNA extraction. A total of 25 pairs of specimen were included.
Quantitative reverse transcription polymerase chain reaction (qRT-PCR)
Samples of total RNA isolated from cells used in microarray hybridization and clinical samples were set aside for verification using SYBR Green-based qRT-PCR. Briefly, 1 μg of RNA was primed by Oligo(dT)16 (Promega, USA) and reverse transcribed using AMV reverse transcriptase in a 20-μl reaction system. From a 2-fold dilution of this cDNA solution, 0.5 μl was used in a 20-μl PCR reaction containing 10 μl SYBR premix Ex Taq, 0.4 μl Rox (TaKaRa, Japan), 10 μmol/L forward and reverse primer. All primers were designed by submitting RefSeq sequences to Primer Premier 5.0 software; the amplicons were approximately 100–200 bp (primer sequences are shown in Table 1). PCR amplification was conducted in 96-well MicroAmp Optical plates (Applied Biosystems, USA) on an Applied Biosystems PRISM 7300 Sequence Detection System under the following conditions: 10 seconds denaturation and enzyme activation at 95°C, followed by 40 cycles of denaturation (95°C, 5 seconds), annealing (specific to each gene, 31 seconds) and extension and fluorescence testing (72°C, 45 seconds). Results were normalized to beta-actin to control for differences in RNA loading, quality and cDNA synthesis. Amplicon size and reaction specificity was confirmed by agarose gel eletrophoresis. Each sample was assayed in triplicate and the median threshold cycle values were used to calculate the fold change between treated and control samples.17 Standard deviations were also calculated.
All statistical analysis was performed using Spring 7.0 software. Data were expressed as mean ± standard deviation (SD). The microarray data were analyzed statistically using one-way analysis of variance (ANOVA) between HIOEC, HB-56p and HB-96p. P values of <0.05 were considered statistically significant.
Characteristics of transcript during the malignant transformation by B (a) P
The transformation process could be divided into 2 stages. Stage 1 displayed the process from HIOEC to HB-56p cells, which contained the mechanism how cells acquire tumorigenesity in nude mice. Stage 2 showed the process during which the tumor transformed gradually into being malignant. The differentially expressed genes were analyzed using a scatter plot (Figure 1), which showed that there were more genes up- or down-regulated in stage 1 (HB-56p versus HIOEC, Figure 1A). Cluster analysis showed the difference of transcript signatures between the three cell lines and the concordance of the hybridization experiment (Figure 2).
Fold change was an important index which depicted the change of gene expression. When comparing the distributions of fold changes, we found that a smaller part of the genes had changed in the whole 38 500 represented genes in the microarray if we set 2-fold as the criterion (Table 2). The stylized Venn diagram in Figure 3 depicts the patterns of changes in gene expression levels during the transformation process of HIOEC cells to HB-96p cells (Table 3). Given that genes with consistent expression patterns have significant effects on malignant phenotypes and are potentially valuable for malignant transformation, sets A-D were those of primary interest in analyzing these data:
(a) Set A contained those genes that showed a consistent and continued change throughout both stages of the malignant transformation examined (i.e. HB-56p versus HIOEC and HB-96p versus HB-56p, as well as HB-96p versus HB-56p).
(b) Set B contained genes for which expression changes in the early stages of the transformation process (HB-56p versus HIOEC), with persistence of the change at completion of malignant transformation (HB-96p versus HB-56p) but without additional change between HB-56p and HB-96p.
(c) Set C contained genes showing changes in expression during the later stages of the transformation process (HB-96p versus HB-56p), without change between HIOEC and HB-56p, producing an overall change in expression levels between HB-96p and HIOEC.
(d) Set D contained genes showing changes in expression during the transformation overall (HB-96p versus HIOEC), but without demonstrable changes in the early or later stages of the transformation process.
(e) Sets E and G, respectively, represented genes showing changes in expression in HB-56p versus HIOEC and HB-96p versus HB-56p. However, these genes were not detected as showing significant changes overall during the process of the malignant transformation (HB-96p versus HIOEC) and were, therefore, not considered further by us. We predicted that set F would be empty (because any gene falling into set E and set G should fall into set A), and this was indeed the case.
We observed more changes in gene expression in the early stages of the transformation process (between HB-56p and HIOEC, when cells could initially form tumors in nude mice) than in the later stages of the transformation process (between HB-96p and HB-56p, when the tumors contained more and more malignant component in histopathology). Sets A-D contained 883 genes, of which 497 showed an increase in expression levels and 386 showed a decrease. These 883 genes were divided according to the sets of the Venn diagram in Figure 3 and then ranked by mean fold change between HB-96p and HIOEC.
Figure 4 shows the GO classification of genes differentially expressed between HB-96p and HIOEC. We observed that these genes mainly involved in the biological process such as macromolecule metabolism, signal transduction, establishment of localization, regulation of cellular physiological process, transport, and so on; genes involved in the immune response and response to pathogens were radically down-regulated instead of up-regulated. Molecular function of these genes focused on transition metal ion binding, adenyl nucleotide binding, kinase activity, phosphotransferase activity, transcription factor and cofactor activity, the down-regulated genes involved in transition metal ion binding and calcium ion binding were more than the up-regulated genes with the same functions. The protein products of these differentially expressed genes were mainly integral to membranes, localized in the nucleus or related to the cytoskeleton.
Validation of microarray using qRT-PCR
Expression levels and patterns of IGFBP3, S100A8, KRT6B, MAP2K1 and MET observed using microarray were reproduced almost exactly using qRT-PCR (Figure 5), while levels of IGFBP3 and GDF15 in both HB-56p and HB-96p, and levels of TGFBR2 and TP73L in HB-96p tested using qRT-PCR were much higher than the microarray data. In general, the magnitudes of gene expression differences detected through the validation were demonstrated to be within 1-fold of the microarray data. However, larger fold-differences were elucidated using qRT-PCR, as microarray data has been reported to compress gene expression changes.18 Expression levels of IGFBP3 were validated further in the 25 OSCC specimens (Figure 6), which was up-regulated in 19 out of 25 specimens (76%), and up-regulated more than 2-fold in 17 out of 25 (68%).
As one high risk etiological factor of OSCC, HPV type 16 was previously used to immortalize normal oral epithelial cells, while this immortalized cell line could not form tumors in nude mice.15 However, the cells acquired tumorigenesity through subsequent exposure to B (a) P followed by continued passaging in vitro.16 So, the biological events occurring during the malignant transformation process are crucial in oncogenic research.
In the present study, hybridization of mRNA-derived probes to cDNA microarray allowed us to generate expression profiles for thousands of genes simultaneously and to identify the molecular mechanisms underlying a continuous transformation process during which an immortalized cell line acquired increasing tumorigenesity through B (a) P exposure. We distinguished genes that played roles in different stages of the transformation process successfully through a Venn diagram. Most parts of the differentially expressed genes focused on the stage of HB-56p versus HIOEC, during which cells possess tumorigenesity initially, though the tumor contained mostly benign contents. This finding indicates that genes associated with this process may play a crucial role in the transformation process.
The differentially expressed genes were then annotated using Gene Ontology. Most of these genes involved in macromolecule metabolism, signal transduction and regulation of cellular physiological process, which may be the common physiological process in transformed cells. These changed genes mainly possessed molecular functions of transition metal ion binding, adenyl nucleotide binding, kinase activity, and transcription factor or cofactor activity, suggesting that genes with these functions might be crucial in the transformation process. Their protein products were mainly integral to membranes, localized in the nucleus and cytoskeleton, which depicted the crucial sites of change in transformed cells and the cytoskeleton associated genes partly explained the changes in cell shape of HB cells compared with HIOEC cells.
B (a) P first required metabolic activation by phase I enzymes and then were subjected to detoxification by phase II enzymes. Phase I enzymes were encoded by the cytochrome P450 (CYP450s) gene family, whereas phase II enzymes included the glutathione-S-transferases (GSTs).19 The metabolism formed a proposed ultimate carcinogen, this diol epoxide metabolite of B (a) P was capable of forming stable DNA-adducts leading to mutations and oxidative damage in target organs.20 In addition, genes involved in the metabolism of B (a) P did not change significantly in this study as it did in other documents.11,12,19,20 In this study, CYP24A1, a member of CYPs, was up-regulated in set B with a fold change of 3.071 and GSTT2, a member of GSTs, was up-regulated in set D with a fold change of 2.367, whereas genes of GSTA4, MGST2, CYP1B1 and CYP27B1, were down-regulated in set B. One reason may be that the metabolism of B (a) P was only an initial factor of carcinogenesis, with the deprivation of B (a) P and the continuous passaging in vitro, these metabolism associated genes would lose their influence on the downstream event.
We did not find genes which were up or down-regulated in neoplastic cells compared with normal cells and immortalized cells in the study of Rey,14 such as cyclophilin A, tumor necrosis factor receptor associated protein p67, c-myc promoter binding protein (MBP1), antileukoproteinase SKALP and heat shock protein 90 alpha, which were differentially expressed in this study.
Growth differentiation factor 15 (GDF15) was consistently up-regulated both in the transformation process of HB-56p versus HIOEC and HB-96p versus HB-56p. This gene is a member of the TGFβ superfamily, involves the regulation of tissue differentiation and maintenance, and acts as cytokine and growth factor. GDF15 could increase the basal ERK1 phosphorylation and prolong the estrogen-stimulated ERK1 phosphorylation; its expression levels are up-regulated in breast cancer and may serve as a surrogate marker for AKT activation in breast cancer.21
Five genes were consistently down-regulated during the transformation process of both HB-56p versus HIOEC and HB-96p versus HB-56p. Gap junction protein alpha 1 (GJA1, also named connexin 43), plays a role in cell communication. Overexpression of GJA1 through retroviral delivery in breast tumor cells did result in a dramatic suppression of tumor growth when the cells were implanted into the mammary fat pad of nude mice. 22 Loss of GJA1 in human tumors might play an important role in the dysregulation of normal growth control. 23 Plasma membrane localization and formation of channels were not required for growth inhibition by GJA1, whereas nuclear localization of C-terminal portion of GJA1 may exert effects on gene expression and growth.24 Snail homolog 2 (SNAI2, also named SLUG) was a negative regulator of transcription from RNA polymerase II promoter and a ces-1-related zinc finger transcription factor gene with antiapoptotic activity.25 Over expression of SNAI2 may promote tumorigenesis through increased resistance to programmed cell death in cell lines from human breast carcinoma and melanoma.26 One possible explanation for this discrepancy may be related to the different origin of the epithelial cells. Interferon induced transmembrane protein 1 (IFITM1) attended in negative regulation of cell proliferation, which associates with other proteins at the cell surface, formed complex relaying growth inhibitory and aggregation signals.27
As one of the most differentially expressed genes in HB-56p versus HIOEC, IGFBP3 was a mediator of growth suppression signals and a putative tumor suppressor gene,28,29 whereas it was reported to be up-regulated in neoplastic tissues compared with normal counterparts in several reports.30,31 Its expression was higher in less aggressive tumors, but was not associated with disease progression.32 Increased breast epithelial IGFBP3 expression was a feature of tumorigenesis with cytoplasmic immunoreactivity in the absence of significant nuclear localization.33 The interaction between IGFBP3 and the EGFR system might be central to whether IGFBP3 acted as a growth stimulator or inhibitor in breast cancer cells.34S100A8 is a member of the S100 family of proteins containing 2 EF-hand calcium-binding motifs, involved in the regulation of a number of cellular processes such as cell cycle progression and differentiation. The decreased expression of S100A8 might play an important role in the pathogenesis of human esophageal squamous cell carcinoma, being particularly associated with poor differentiation of tumor cells.35
In summary, we have identified genes differentially expressed in oral epithelial cells transformed by B (a) P. These gene subsets can potentially help in better understanding the development mechanisms associated with OSCC derived from B (a) P exposure and IGFBP3 may play a potential role in the initiation of oral cancer related with B (a) P exposure.
1. Greenlee RT, Hill-Harmon MB, Murray T, Thun M. Cancer statistics. CA Cancer J Clin 2001; 51: 15-36.
2. Mendez E, Cheng C, Farwell DG, Ricks S, Agoff SN, Futran ND, et al. Transcriptional expression profiles of oral squamous cell carcinomas. Cancer 2002; 95: 1482-1494.
3. Vigneswaran N, Wu J, Sacks P, Gilcrease M, Zacharias W. Microarray gene expression profiling of cell lines from primary and metastatic tongue squamous cell carcinoma: possible insights from emerging technology. J Oral Pathol Med 2005; 34: 77-86.
4. Kuriakose MA, Chen WT, He ZM, Sikora AG, Zhang P, Zhang ZY, et al. Selection and validation of differentially expressed genes in head and neck cancer. Cell Mol Life Sci 2004; 61: 1372-1383.
5. Whipple ME, Mendez E, Farwell DG, Agoff SN, Chen C. A genomic predictor of oral squamous cell carcinoma. Laryngoscope 2004; 114: 1346-1354.
6. Chen YJ, Lin SC, Kao T, Chang CS, Hong PS, Shieh TM, et al. Genome-wide profiling of oral squamous cell carcinoma. J Pathol 2004; 204: 326-332.
7. Hwang D, Alevizos I, Schmitt WA, Misra J, Ohyama H, Todd R, et al. Genomic dissection for characterization of cancerous oral epithelium tissues using transcription profiling. Oral Oncol 2003; 39: 259-268.
8. Toruner GA, Ulger C, Alkan M, Galante AT, Rinaggio J, Wilk R, et al. Association between gene expression profile and tumor invasion in oral squamous cell carcinoma. Cancer Genet Cytogenet 2004; 154: 27-35.
9. Ruutu M, Johansson B, Grenman R, Syrjanen S. Two different global gene expression profiles in cancer cell lines established from etiologically different oral carcinomas. Oncol Rep 2005; 14: 1511-1517.
10. Maier H, Dietz A, Gewelke U, Heller WD, Weidauer H. Tobacco and alcohol and the risk of head and neck cancer. Clin Invest 1992; 70: 320-327.
11. Wen X, Walle T. Preferential induction of CYP1B1 by benzo(a)pyrene in human oral epithelial cells: impact on DNA adduct formation and prevention by polyphenols. Carcinogenesis 2005; 26: 1774-1781.
12. Keshava C, Whipkey D, Weston A. Transcriptional signatures of environmentally relevant exposures in normal human mammary epithelial cells: benzo(a)pyrene. Cancer Lett 2005; 221: 201-211.
13. Park NH, Gujuluva CN, Baek JH, Cherrick HM, Shin KH, Min BM. Combined oral carcinogenicity of HPV-16 and benzo(a)pyrene: an in vitro
multistep carcinogenesis model. Oncogene 1995; 10: 2145-2153.
14. Rey O, Baluda MA, Park NH. Differential gene expression in neoplastic and human papillomavirus-immortalized oral keratinocytes. Oncogene 1999; 18: 827-831.
15. Zhang ZY, Sdek P, Cao J, Chen WT. Human papillomavirus type 16 and 18 DNA in oral squamous cell carcinoma and normal mucosa. Int J Oral Maxillofac Surg 2004; 33: 71-74.
16. Zhong LP, Pan HY, Zhou XJ, Ye DX, Zhang L, Yang X, et al. Characteristics of a cancerous cell line, HIOEC-B(a)P-96, induced by benzo(a)pyrene from human immortalized oral epithelial cell line. Arch Oral Biol 2008; 53: 443-452.
17. Livak KJ, Schmittgen TD. Analysis of relative gene expression data using real-time quantitative PCR and the 2(-Delta Delta C(T)) Method. Methods 2001; 25: 402-408.
18. Yuen T, Wurmbach E, Pfeffer RL, Ebersole BJ, Sealfon SC. Accuracy and calibration of commercial oligonucleotide and custom cDNA microarrays. Nucleic Acids Res 2002; 30: e48.
19. Nimura Y, Yokoyama S, Fujimori M, Aoki T, Adachi W, Nasu T, et al. Genotyping of the CYP1A1 and GSTM1 genes in esophageal carcinoma patients with special reference to smoking. Cancer 1997; 80: 852-857..
20. Dertinger SD, Nazarenko DA, Silverstone AE, Gasiewicz TA. Aryl hydrocarbon receptor signaling plays a significant role in mediating benzo(a)pyrene- and cigarette smoke condensate induced cytogenetic damage in vivo.
Carcinogenesis 2001; 22: 171-177.
21. Goulet RJ, Mehrotra S, Morimiya A, Badve S, Nakshatri H. The macrophage inhibitory cytokine integrates AKT/PKB and MAP kinase signaling pathways in breast cancer cells. Carcinogenesis 2005; 26: 900-907.
22. Qin H, Shao Q, Curtis H, Galipeau J, Belliveau DJ, Wang T, et al. Retroviral delivery of connexin genes to human breast tumor cells inhibits in vivo
tumor growth by a mechanism that is independent of significant gap junctional intercellular communication. J Biol Chem 2002; 277: 29132-29138.
23. King TJ, Fukushima LH, Yasui Y, Lampe PD, Bertram JS. Inducible expression of the gap junction protein connexin43 decreases the neoplastic potential of HT-1080 human fibrosarcoma cells in vitro
and in vivo.
Mol Carcinog 2002; 35: 29-41.
24. Dang X, Doble BW, Kardami E. The carboxy-tail of connexin-43 localizes to the nucleus and inhibits cell growth. Mol Cell Biochem 2003; 242: 35-38.
25. Inukai T, Inoue A, Kurosawa H, Goi K, Shinjyo T, Ozawa K, et al. SLUG, a ces-1-related zinc finger transcription factor gene with antiapoptotic activity, is a downstream target of the E2A-HLF oncoproteins. Mol Cell 1999; 4: 343-352.
26. Kajita M, McClinic KN, Wade PA. Aberrant expression of the transcription factors snail and slug alters the response to genotoxic stress. Mol Cell Biol 2004; 24: 7559-7566.
27. Deblandre GA, Marinx OP, Evans SS, Majjaj S, Leo O, Caput D, et al. Expression cloning of an interferon-inducible 17-kDa membrane protein implicated in the control of cell growth. J Biol Chem 1995; 270: 23860-23866.
28. Jenkins PJ, Khalaf S, Ogunkolade W, McCarthy K, David T, Hands RE, et al. Differential expression of IGF-binding protein-3 in normal and malignant colon and its influence on apoptosis. Endocr Relat Cancer 2005; 12: 891-901.
29. Yumoto E, Nakatsukasa H, Hanafusa T, Yumoto Y, Nouso K, Matsumoto E, et al. IGFBP-3 expression in hepatocellular carcinoma involves abnormalities in TGF-beta and/or Rb signaling pathways. Int J Oncol 2005; 27: 1223-1230.
30. Sohda M, Kato H, Miyazaki T, Nakajima M, Fukuchi M, Manda R, et al. The role of insulin-like growth factor 1 and insulin-like growth factor binding protein 3 in human esophageal cancer. Anticancer Res 2004; 24: 3029-3034.
31. Chan JM, Stampfer MJ, Ma J, Gann P, Gaziano JM, Pollak M, et al. Insulin-like growth factor-I (IGF-I) and IGF binding protein-3 as predictors of advanced-stage prostate cancer. J Natl Cancer Inst 2002; 94: 1099-1106.
32. Lu L, Katsaros D, Wiley A, Rigault de la Longrais IA, Risch HA, Puopolo M, et al. The relationship of insulin-like growth factor-II, insulin-like growth factor binding protein-3, and estrogen receptor-alpha expression to disease progression in epithelial ovarian cancer. Clin Cancer Res 2006; 12: 1208-1214.
33. Vestey SB, Perks CM, Sen C, Calder CJ, Holly JM, Winters ZE. Immunohistochemical expression of insulin-like growth factor binding protein-3 in invasive breast cancers and ductal carcinoma in situ:
implications for clinicopathology and patient outcome. Breast Cancer Res 2005; 7: R119-R129.
34. Butt AJ, Martin JL, Dickson KA, McDougall F, Firth SM, Baxter RC. Insulin-like growth factor binding protein-3 expression is associated with growth stimulation of T47D human breast cancer cells: the role of altered epidermal growth factor signaling. J Clin Endocrinol Metab 2004; 89: 1950-1956.
35. Kong JP, Ding F, Zhou CN, Wang XQ, Miao XP, Wu M, et al. Loss of myeloid-related proteins 8 and myeloid-related proteins 14 expression in human esophageal squamous cell carcinoma correlates with poor differentiation. World J Gastroenterol 2004; 10: 1093-1097.