To date, there have been few publications analyzing gene expression in primary SBNETs. One study by Drozdov et al. examined 9 SBNETs and normal small bowel mucosa samples using Affymetrix U133A arrays, with probesets corresponding to 14,500 genes. They performed a gene network analysis and ultimately reported that of 3470 genes in 10 ontology pathways, 27% were differentially expressed. They then focused upon 2 GPCR and 7 cAMP response-element binding genes, and hypothesized that overexpression of these 2 classes of genes may result in neural activation of secretory genes, providing a possible explanation of the hormonal behavior of these tumors12
. In our study there was no significant upregulation of these 2 GPCR targets (ADCY2
). Four of the postulated CREB targets showed upregulation >5 fold in SBNETs samples; CHGB was upregulated 17.5 fold, BEX1 was upregulated 15.1 fold, and SCG3 was upregulated 7.2 fold. In PNETs, CHGB was also upregulated 7.4 fold. SCG2 and 3 were upregulated in PNETs a well, to 8.0 and 5.8 fold, respectively ().
Exon Array Results for Selected Genes of Interest from Other Studies
The largest study of gene expression profiling in PNETs was by Missiaglia et al., who evaluated 72 PNETs, 7 metastases, and 10 normal pancreatic controls (5 whole pancreas, 5 islet cell preparations) using custom 18.5K arrays11
. Four categories of tissues emerged from unsupervised cluster analysis: normal pancreas, normal islets, insulinomas, and non-functional PNETs. They found that there were 113 upregulated and 25 downregulated genes in insulinomas compared to normal samples, and 198 upregulated and 55 downregulated genes in non-functional PNETs. The tuberous sclerosis 2 gene (TSC2
), an inhibitor of the Akt-mTOR pathway, was downregulated in both of these tumor sub-types, and patients with low TSC2
expression had decreased survival. They also determined that SSTR2
expression was significantly upregulated in 25 well-differentiated, non-functional PNETs as compared to 14 insulinomas. Fibroblast growth factor 13 (FGF13
) was found to be overexpressed in metastases versus their primary tumors, and was significantly associated with liver metastasis at diagnosis, as well as decreased survival in well-differentiated PNETs. The clinical implications of this study were validation of a mechanism for mTOR inhibitors’ efficacy in treatment of PNETs, a mechanism supporting the use of somatostatin analogues stabilizing disease progression in non-functional PNETs, and that FGF13
may be a useful marker for progression. From our ST array data, we found that TSC2
, and FGF13
showed no significant change in expression between tumor and normal tissue in both SBNETs and PNETs ().
Previous work by Duerr et al.10
analyzed 24 PNETs (16 well-differentiated endocrine tumors [WDET WHO classification] and 8 well-differentiated endocrine carcinomas [WDEC]) and 6 malignant GI-NETs (3 primary ileal, 1 colon, 2 liver metastases) using Affymetrix U133A arrays. When examining PNETs, they found that the genes FEV
, and GADD45
β were significantly overexpressed in WDECs as compared to WDETs; they also found that microarrays underestimated the degree of upregulation as compared to qPCR. They reported that previous genes of interest in PNETs (MEN1
, cyclin D1, retinoic acid receptor β, p16INK4A/p14ARD
, and p27Kip1
) were not significantly different between these WHO subtypes. Comparison of 19 PNETs to 6 GI-NETs revealed 385 differentially-expressed genes with at least 1.5 fold change and p-values of <0.05; 157 were upregulated and 228 downregulated in GI-NETs. The most differentially overexpressed genes in GI-NETS included ECM1
(28-fold by microarray, 39-fold by qPCR), VMAT1
(25-fold, 523-fold by qPCR), LGALS4
(24-fold, 43-fold by qPCR). The implications from this study were that 4 genes were found which could help distinguish between PNETs of different WHO classes, and between GI-NETs and PNETs. The shortcoming of this study was that it only looked at a small number of GI-NETs, and only 3 of these were small bowel primaries. In our evaluation by ST arrays, we found that for these 3 genes, there was statistically significant upregulation of VMAT1
(22 fold) in SBNETs. LGALS4
showed no change between tumor and normal tissue ().
Couvelard et al.19
examined 12 benign PNETs (WHO-1 and WH0-2 categories) and 12 malignant PNETs (WHO-3) using custom microarrays representing 9932 transcripts, and found that a cluster of 123 genes could differentiate between these 2 groups. The fold changes observed were relatively modest, ranging from 0.47 (downregulated in WHO-3) to 2.26 (upregulated in WHO-3 group), and did not overlap significantly with genes discussed above by Missiaglia or Duerr et al. Since these studies and others20, 21, 22
using varying genomic expression array platforms have identified different, non-overlapping genes thought to be of importance, we decided to focus our efforts on one important class of genes known to play a role in these tumors using a qPCR based strategy in order to develop a reliable and predictive model. Other reasons for choosing this strategy include the fact that studies comparing qPCR to microarray results have shown and increased sensitivity of the former, that members of the GPCR group emerged as candidates with differential expression in our exon arrays and in Drozdov’s study, and this class of genes is deemed as one of the most promising for development of new therapeutic agents by the pharmaceutical industry23
There have been just a few studies that have attempted to use gene expression profiling to determine the site of origin from tissue derived from liver metastases. Posorski et al. evaluated 17 NET metastases and 6 primary tumors from 17 patients by both comparative genomic hybridization (CGH) using Agilent 105K CGH microarrays, and genome-wide expression using Agilent 44K expression microarrays. Multiple techniques were used to analyze this data set, including hierarchical clustering of 41,000 genes, which revealed 1,760 differentially expressed genes segregating into 3 clusters (primaries in ileum, pancreas, and stomach). They then attempted to formulate the simplest expression profile they could find that would discriminate between these primary sites. This began by determining whether there was upregulation in the metastasis of CD302
>13 times that of the primary site, which indicated that the tumor was ileal in origin. If this was not the case, if PPWD1
was downregulated >3-fold, the tumors were of pancreatic origin, and if not this, but >4-fold upregulated for ABHD14B
, then the stomach was the site of the primary tumor14
. One thing that is not clear from this study is how the fold changes for each were calculated, but it appeared to require using results from the primary as well, which would not make diagnosis from only liver biopsy tissue possible. Furthermore, the overall number of tumors examined here was small. It is difficult to assess whether this profile would be validated in our pancreatic and small bowel samples, as CD302
was not assessed in our ST arrays. However, our pancreatic samples did not show the same >3-fold downregulation in PPWD1
as seen in Posorski’s study ().
Another study by Edfeldt et al.13
analyzed 18 primary tumors, 17 lymph node metastases, and 7 liver metastases from 19 patients with SBNETs. They hybridized RNA to QArray2 microarrays (containing 24,650 genes), then performed cluster analysis, which separated the tumors into 3 groups: 11 primaries, 5 nodes, and 7 liver metastases formed one group; 7 nodes another; and the final group consisted of 5 other primaries. They concluded that the expression clusters did predict clinical outcomes, and that expression patterns were different between primary tumors and their lymph nodes. They reported 8 genes that were differentially expressed between clusters, which included ACTG2
, and CDH6
. Although this study did suggest several gene targets, we only found 3 of these genes (TPH1, REG3A,
) to be differentially expressed between primary tumor and normal samples in our study, and only in SBNETs ().
In this study, RNA was extracted from whole tumor tissues, which often contain a fibrous reaction surrounding the primary tumors. We noted a statistically significant change in expression profile of many genes with this technique, which might have been slightly different if we had extracted from a purer tumor population. Posorski et al. used laser capture microdissection to enrich for the tumor cell population, which they felt was important to obtain useful results14
. One advantage we did have in this study however, was having normal tissue, primary tumor, and metastases from the same patients. We also did not make full use of the genes identified as being significantly differentially expressed from our ST exon arrays, and instead chose to focus on genes in the GPCR pathway. The higher fold-changes seen with some of these GPCR genes was due in part to the more sensitive qPCR technique used, in contrast to microarray hybridization.
Comparison of the gene expression patterns between normal and tumor tissues allowed us to establish profiles to test in metastases. Of these GPCR genes, OXTR and GPR113 were both >5-fold upregulated in all 11 SBNETs versus normal tissues, and both were not upregulated in all 15 PNETs. Furthermore, 11 of 15 PNETs also had >5-fold downregulation of both ADORA1 and SCTR, which was only seen in 1 poorly-differentiated SBNET (). When the criteria of upregulation of both OXTR and GPR113 (relative to normal) were used for distinguishing the site of origin in 10 liver metastases, it correctly predicted 4 of 5 SBNETs and 5 of 5 PNETs. However, the practicality of these results are limited in that they were compared to normal tissues, which would not be relevant to a core biopsy sample of a liver metastasis. We reevaluated the expression profiles of OXTR, GPR113, ADORA1, and SCTR between primary SBNETs and PNETs relative to our control gene GAPDH without normalization to corresponding normal tissues, and found that OXTR and SCTR expression to be the most useful for distinguishing between the two tumors. OXTR expression was >20-fold greater than SCTR in 8 primary SBNETs (3 were indeterminate), while there was <5-fold expression difference in 14 PNETs (1 was indeterminate). When blinded to 10 liver metastases’ primary site, these differences in OXTR and SCTR expression correctly identified 4 of the 5 small bowel metastases, 4 of 5 pancreatic metastases, and 1 metastasis from each site was indeterminate. Since this model correctly predicted the primary site in 80% of these samples, with no incorrect predictions (and 20% indeterminate), this test could have practical clinic value in the evaluation of core biopsy specimens of liver metastases. Validation with a larger number of samples will be needed to confirm the value of this profile, as will evaluation of metastases from other GI sites, such as the stomach, duodenum, and colorectum. Systematic examination of a larger subset of differentially expressed genes from both the exon and GPCR arrays may also prove useful for making these finer discriminations and to improve diagnostic accuracy.
Knowledge that the expression pattern of a metastasis is consistent with a specific primary site could lead to improved surgical exploration. In a study of 123 patients with metastatic NETs, Wang et al. found that only 35% of GI-NETs (small bowel, colorectal, and stomach) were seen on CT scan (and even less on Octreoscan), whereas all PNETs had a pancreatic mass visible on CT4
. It is our experience that many small bowel tumors are too small to be seen on conventional imaging, and knowledge that a liver metastasis was likely of small bowel origin by virtue of its gene expression pattern would make the decision to explore that patient easier. At exploration, careful palpation of the entire small bowel will uncover primary tumors even a few mm in size, and patients with SBNETs can have relatively small primaries in the face of bulky metastatic disease. In these patients, an aggressive surgical approach is warranted5
, since resection of the primary site in patients with liver metastases leads to both improved progression free survival as well as overall survival24