To understand the importance of DNA copy number aberrations in OSCC progression, we sought to identify DNA copy number aberrations unique to carcinoma cells found in metastatic lymph nodes that are associated with changes in the expression of encoded genes. Moreover, we reasoned that since metastatic tumor cells have de-regulated apoptotic pathways 
, we could leverage on this survival advantage to find those genes associated with the growth-promoting pathways inherent to these cells. Using a large-scale siRNA interference screen, we further prioritized functionally relevant amplified/over-expressed growth-promoting genes associated with OSCC metastasis. We showed that CNAs-associated transcripts differentially expressed in carcinoma cells with an aggressive phenotype (i.e., metastatic to lymph nodes) can be biomarkers with both prognostic information and functional relevance. In particular, we demonstrated the potential of G3BP1 as a novel therapeutic target in metastatic OSCC.
DNA copy number profiling of the 37 samples in the training cohort detected most of the previously reported CNA associated with poor outcome in patients with OSCC or head and neck squamous cell carcinoma 
. Interestingly, we detected more genomic alterations in carcinoma cells from non-metastatic primary OSCC than in those from metastatic OSCC, which is consistent with previous findings in other cancer types 
. It may be an indication that not all genetic alterations are related to metastasis, thus supporting the theory that metastasis is derived from a subclone in the primary tumor. The DNA copy number in the four chromosomal regions (3p25.3-21.1, 5q31-35.3, 9p24.3- 22.3, and 18q21-23), which harbor 59 of the 95 transcripts, is significantly higher in metastatic OSCC than in non-metastatic primary OSCC. Subsequent KD test showed that multiple genes involved in the survival of OSCC cell lines locate in these regions. Thus multiple gene expression changes that may be essential to OSCC cancer progression may be the result of the same CNA event in these regions.
Accumulating evidence supports the hypothesis that metastasis derives from a subclone of cancer cells in a late-stage primary tumor 
. In this study, we focused on metastasis by comparing carcinoma cells from non-metastatic primary carcinomas with carcinoma cells metastatic to lymph nodes, rather than a pair-wise comparison between carcinoma cells from primary OSCC and corresponding metastatic carcinoma cells isolated from lymph nodes. This distinction is important, because there is evidence that prior to metastasis, the carcinoma in the primary site already contains malignant cells with genetic changes associated with metastatic potential, as we and others have shown 
. Given that the comparison of genetic changes between carcinoma cells in lymph nodes vs. those in the primary site of non-metastatic carcinomas involves two distinct micro-environmental milieus, we used laser-capture microdissection to enrich for carcinoma cells. Because of this, we could not identify genes in the cancer-associated stroma that may be essential to the metastasis process. In addition, due to the relatively small sample size of the training set, the genetic variances among individual patients may reduce the power to identify true differences in CNAs or gene expression dysregulation between these two groups of tumor cells.
The 95 gene signature was strongly associated with OSCC survival in an independent patient cohort. We focused on survival as an outcome because understanding that metastasis is a strong clinical predictor of survival, we reasoned that those genomic aberrations, and associated changes in gene expression, unique to tumor cells from metastatic lymph nodes would likely be candidate biomarkers of patient outcomes. It is possible that the 95 gene signature associated with poor prognosis represents a correlative phenomenon of cell survival, rather than being causative of the metastatic process per se, which is complex and involves a number of phenotypes such as enhanced survival, migration, proliferation, anoikis, etc. In fact, one motivation for our study design was to identify genes associated with aggressive tumor, which is associated with these phenotypic attributes, and to identify genes that are involved in conferring cells a survival advantage (regardless of the mechanisms through which survival is affected).
It is noteworthy that our 95 gene set was not associated with differences in nodal metastasis in the independent patient cohort. There are multiple reasons for this finding. First, the types of tissue analyzed were different between the training and validation cohorts (i.e., the training set consisted of tumor tissue from lymph nodes and non-metastatic primaries whereas the validation set consisted of biopsy samples derived from the primary site only). Secondly, the training set consisted of tumor cells only whereas the validation set represented a mixture of tumor and stromal cells. Thirdly, we may have lacked the power to detect statistically significant differences in nodal status. Fourth, the primary sites of metastatic tumors are likely composed of mixtures of clonal populations – some with metastatic potential and some without. In contrast, we believe that tumor cells in metastatic lymph nodes represent a more homogeneous clonal pool, at least with respect to the metastatic phenotype.
Our finding that 18 of the 26 gene targets tested in the KD screen showed an effect on the viability of OSCC cells indicates that this strategy, if validated in larger studies, could prove successful in identifying biomarkers of OSCC metastasis with therapeutic potential. It is possible that with the criteria chosen for siRNA validation (growth suppression >30% in at least one cell line), other genes unrelated to OSCC cell survival might also meet this threshold. However, the effects on OSCC cell survival upon siRNA KD of an additional 708 human kinases showed that our 26 target genes had a significantly higher proportion of genes involved in OSCC cell survival. We selected genes from the human kinome for this comparison because kinases are known to regulate signaling pathways which could affect growth and survival. Thus, this represented a comparable pool that could also be enriched with genes relevant for the same phenotype for which the 26 target genes were being tested. Our results suggest that our approach was successful at identifying an enriched gene set associated with OSCC progression.
In this study, we focused on G3BP1, because among the 18 genes whose KD induced growth suppression, G3BP1 KD suppressed growth in 4 of the 5 cell lines, and, moreover, the effect was stronger in the lines derived from late-stage OSCC. This feature, if validated further, could prove attractive in the search for targeted therapies against head and neck carcinomas. G3BP1 was first identified as a Ras-GTPase-activating protein (GAP) binding protein as it co-immunoprecipited with GAP 
. Over-expression of G3BP1 has been detected in a number of human cancers including breast, lung, prostate, colon, thyroid, esophageal, and head and neck cancers 
, and over-expression has also been associated with lymph node metastasis and worse survival in esophageal squamous cell carcinoma 
. G3BP1 has been implicated in several cancer-associated survival pathways including Ras
. Recently however, a study by Annibaldi et al. put into question the interaction between G3BP1 and RasGAP 
. The fact that G3BP1 KD did not affect ERK1/2 phosphorylation status or the active form of Ras protein in our cell lines implies that growth suppression and apoptosis was not through inhibition of this pathway. G3BP1 also exhibits cell-cycle dependent RNase activity that affects stability of c-myc
mRNA. G3BP1 has been shown to participate in stress granule (SG) assembly, and the stress granule is the site where some mRNAs are stored during stress conditions and where the ultimate fate of these mRNAs is decided once stress has subsided 
. Formation of SGs in cells may inhibit apoptosis Arimoto K 
, and it is possible that this is one mechanism by which G3BP1 inhibition induced apoptosis in our cell lines.
Another question to address is why induction of apoptosis seemed to be more pronounced in cell lines derived from late-stage OSCC after G3PB1 KD. Since G3BP1 KD induced apoptosis preferentially in the metastatic cell lines with disruptive mutations in TP53
gene, it is possible that this effect might be p53-dependent. Loss of TP53
function is one of the most common genetic alterations in squamous cell carcinoma (SCC) of the head and neck 
and head and neck tumors with disruptive TP53
mutations have been found to be more likely metastatic and to have worse survival 
. This latter point is relevant to this study, in that oncogenesis-mediated proliferation has been shown to be p53 pathway-dependent 
. Indeed, a recent study implicated G3BP1 in the p53 pathway, as KD of G3BP1 led to up-regulation of p53 levels and activity in breast cancer cell line MCF7 
. In lung cancer cell line H1299, KD of G3BP1 suppressed anchorage-independent cell growth 
. Thus, our results regarding the effects of G3BP1 inhibition appear consistent with those found in other studies. However, given the diversity of functions with which G3BP1 has been associated, further studies are warranted to elucidate the mechanism of action.
In conclusion, we have identified a 95-gene set based on CNA-associated gene dysregulation unique to OSCC metastasis which is prognostic in an independent dataset. In addition, we identified 18 amplified/over-expressed genes which appear to promote cell viability and growth. In particular, we identified G3BP1 as a potential target against metastatic OSCC. Further investigation is warranted to examine the biologic role of these CNA-associated differentially expressed genes and to determine the mechanism by which G3BP1 inhibition induces apoptosis in metastatic OSCC cells.