We selected six potential biomarkers of OSCC for the current validation study by examining DNA microarray data both from our laboratory, as well as that published in the literature.5 SPARC, POSTN, TNC
, and TGM3
microarray expression differences were validated by both qRT-PCR and IHC of TMA sections. The qRT-PCR results for MMP3
did not reach statistical significance. In contrast with another IHC study reporting high levels of periostin expression within oral carcinoma epithelium,12
we noted periostin staining to be localized primarily to the stroma and did not see robust staining within tumor cells themselves (). The reason for this disparity in IHC staining is unclear. Non-overlapping antibody epitopes may partially explain the disparity of periostin IHC staining patterns. The antibody we used for periostin IHC was raised in rabbits against recombinant human periostin containing 648 amino acid residues (corresponding to amino acids 22–669 of full-length periostin) with an N-terminal HisTag fusion (http://biovendor.com/pdf/RD181045050.pdf
), whereas periostin IHC experiments by Siriwarden et al. utilized a polyclonal antibody generated by immunizing rabbits with a specific peptide (EGEPEFRLIKEGETC) corresponding to amino acids 679–692 of full-length periostin.12
Despite the strong stromal predominance of periostin expression we observed, the percentage of OSCC tumors positive for epithelial periostin in our study (58%) was comparable to that reported by Siriwarenda et al.12
(69%). A majority (65%) of the stage III/IV OSCC tumors, including 100% of T4 tumors we examined, were positive for epithelial periostin immunostaining, compared to only 25% of the stage I/II tumors. These findings suggest that epithelial expression of periostin may be associated with a more aggressive tumor phenotype in OSCC. This is supported by other studies, which show that subsets of HNSCC cells expressing periostin, or cells engineered to overexpress periostin, exhibit enhanced tumor growth and invasiveness, and tumors that express periostin have a more invasive phenotype.11,12
We found the proportion of metastatic lymph node tumors positive for epithelial periostin expression (23.5%) was less than half that of primary tumors (58.3%). However, each of the positively stained lymph node tumors was associated with a primary tumor that also had epithelial periostin expression, suggesting that presence of periostin in the epithelium of primary tumors may be necessary, but not sufficient, for its presence in metastatic tumors.
In oral and laryngeal squamous cell carcinoma, increased levels of tenascin-C immunostaining have been found to correlate with malignancy and invasion22–25
Abundant expression of tenascin-C in our OSCC TMA sections was localized primarily to the stroma, although some minor staining of tumor cells was also observed, particularly at tumor edges adjacent to desmoplastic stroma. This observation is consistent with reports in the literature showing tenascin-C localization along the invasive fronts of carcinomas of the lung, liver, bladder, and skin.26,27
Roepman, et al.28
and Schmalbach, et al.29
to be significantly down-regulated in metastatic HNSCC compared to both non-metastatic tumors and normal epithelium. O’Donnell et al.30
similarly found significant down-regulation of TGM3
gene expression in metastatic primary OSCC tumors compared to non-metastatic primaries. Our cross-sectional IHC data show that the levels of transglutaminase-3 protein expression were seen to decrease in a stepwise fashion from normal to premalignant to malignant specimens. This suggests that the loss of transglutaminase-3 activity might be associated with the progression of squamous cell carcinoma.
All of the up-regulated gene markers we identified by reviewing gene microarray reports, validated by qRT-PCR, and subsequently studied with IHC revealed protein expression to be localized primarily within the stroma, and modestly or not at all within tumor cells. This finding illustrates an important point regarding the interpretation of gene microarray data based on the methodology used for specimen processing. The methods employed by different laboratories for tumor specimen processing vary significantly.5
While some investigators isolated relatively homogeneous populations of tumor cells for microarray analysis via laser capture microdissection (LCM), others established arbitrary thresholds for minimum tumor cell content in surgical specimens, as assessed by histologic evaluation of adjacent tissue, prior to RNA extraction and microarray analysis. The latter method clearly results in varying amounts of stromal cells contributing to the final pool of extracted RNA, and thus the variability of the resultant microarray data. Notably, even LCM does not ensure isolation of a purely homogeneous population of tumor cells, as varying degrees of leukocytosis and neovascularization within tumors exist and correlate with survival, tumor stage, metastases, and presence of extracapsular spread in HNSCC.31–33
Up-regulation of SPARC
, or TNC
was not reported by any of the DNA microarray studies that examined expression of HNSCC cell lines34–36
or by others that employed LCM to isolate tumor cells from stroma.37–40
Presumably this is due to the relative absence of stromal cells within the analyzed specimens in these studies, although absence of one or more of these markers on the microarrays used by these studies may also contribute. These findings, together with the IHC data we report here, suggest that up-regulation of SPARC
, and TNC
is due to 1) up-regulation within stromal cell populations vs carcinoma cells and/or 2) stroma-induced transcriptional upregulation of these markers in cancer cells. In any case, these observations underscore the importance in examining both tumor and stroma in the pathogenesis of OSCC.
The “seed and soil” hypothesis of tumor-stromal interaction was originally proposed by Paget in 1889, but only recently have researchers examined how tumor microenvironments influence the growth and spread of cancers. Carcinoma-associated fibroblasts (CAFs), ECM macromolecules, neovascularization, and inflammatory and immune cell infiltration within the stroma adjacent to tumors can have profound effects on tumor progression in breast, prostate, and skin carcinomas.41
The situation in OSCC is less well understood, but studies of CAFs, ECM turnover and tumor cell motility have begun to delineate the role of desmoplastic stroma in OSCC carcinogenesis.42–44
Recently, Weber et al.45
performed genome-wide analysis of loss of heterozygosity (LOH) and allelic imbalance (AI) on LCM-isolated specimens of tumor stroma and tumor epithelium from over 120 OSCC patients with a history of smoking. They discovered over 40 hot spots of LOH/AI within the stroma, nearly twice as many as they found in the epithelium, and subsequently identified three stroma-specific loci that were significantly associated with tumor size and cervical lymph node metastasis.45
These findings again highlight the importance of examining both stromal as well as epithelial elements in OSCC, and suggest that stromal alterations play a crucial part in facilitating OSCC invasion and metastasis.