Gene network-based analysis identifies two potential subtypes of small intestinal neuroendocrine tumors

Abstract

Background: Tumor transcriptomes contain information of critical value to understanding the different capacities of a cell at both a physiological and pathological level. In terms of clinical relevance, they provide information regarding the cellular "toolbox" e.g., pathways associated with malignancy and metastasis or drug dependency. Exploration of this resource can therefore be leveraged as a translational tool to better manage and assess neoplastic behavior. The availability of public genome-wide expression datasets, provide an opportunity to reassess neuroendocrine tumors at a more fundamental level. We hypothesized that stringent analysis of expression profiles as well as regulatory networks of the neoplastic cell would provide novel information that facilitates further delineation of the genomic basis of small intestinal neuroendocrine tumors.

Results: We re-analyzed two publically available small intestinal tumor transcriptomes using stringent quality control parameters and network-based approaches and validated expression of core secretory regulatory elements e.g., CPE, PCSK1, secretogranins, including genes involved in depolarization e.g., SCN3A, as well as transcription factors associated with neurodevelopment (NKX2-2, NeuroD1, INSM1) and glucose homeostasis (APLP1). The candidate metastasis-associated transcription factor, ST18, was highly expressed (>14-fold, p < 0.004). Genes previously associated with neoplasia, CEBPA and SDHD, were decreased in expression (-1.5 - -2, p < 0.02). Genomic interrogation indicated that intestinal tumors may consist of two different subtypes, serotonin-producing neoplasms and serotonin/substance P/tachykinin lesions. QPCR validation in an independent dataset (n = 13 neuroendocrine tumors), confirmed up-regulated expression of 87% of genes (13/15).

Conclusions: An integrated cellular transcriptomic analysis of small intestinal neuroendocrine tumors identified that they are regulated at a developmental level, have key activation of hypoxic pathways (a known regulator of malignant stem cell phenotypes) as well as activation of genes involved in apoptosis and proliferation. Further refinement of these analyses by RNAseq studies of large-scale databases will enable definition of individual master regulators and facilitate the development of novel tissue and blood-based tools to better understand diagnose and treat tumors.

Figure 1

Re-analysis of two small intestinal NET sets ( details in methodology ). A, B. Principal component analysis and scatterplot of arrays along the first two principal components demonstrating spatial separation between control (normal mucosa) and tumor samples. C, D. Volcano plot of differentially expressed genes in Tumor compared to Normal for each of the sample sets. The most differentially expressed genes are labeled according to their fold changes.

Figure 2

Secretory interactome analysis of two small intestinal NET sets. A, B. BioGRID secretory protein-protein interaction subnetworks of small intestinal NET microarrays. Proteins involved in secretory function are shown in green, while their neighbors are shown in white. Key genes in these pathways were examined by qPCR in the independent set (see Figures 3 and 4). C. Subnetwork cluster similarity heatmap. Darker shades reflect greater extent of shared proteins across network clusters in the two small intestinal NET protein-protein interaction subnetworks.

Figure 3

Neurodevelopmental and COSMIC-based transcript expression in SI NET samples. A. Enteroendocrine-related transcription factors in each of the data sets identified expression of 3 and 12 murine ortholog TFs, respectively. Commonly expressed TFs, involved in the regulation of neurodevelopment, included INSM1, NKX2-2 and ST18. B. QPCR analysis of transcripts predicted by COSMIC analysis to be decreased in small intestinal NETs. Both CEBPA and SDHD expressed levels ~50% of normal mucosa consistent with a decreased expression and potentially a loss of function as has been noted in hematological cancers [71] and paragangliomas [39]. C. QPCR analysis of neurodevelopmental transcripts in the independent set confirmed elevated expression of INSM1, and NEUROD1 and elevated expression of BEX1 and NKX2-2 validating the transcriptome-based analyses. Mean ± SEM, *p < 0.05 vs. normal mucosa. Tumors n = 13, normal mucosa n = 8.

Figure 4

Co-analyses of the two small intestinal NET sets. A. Correlation profile of transcript alterations in each of the tumor sets. Both tissue databases were marginally correlated (R = 0.50). B. Commonly elevated transcripts in both datasets predominantly include genes involved in neuroendocrine secretion and regulation thereof. Error bars indicate the range of fold changes across the two datasets, while green points reflect average gene expression. C. Network analysis of the top ranked genes (see B) identified the most densely connected module to be related to secretion (interactome identified by multiple links). D. Gene-ontology and Reactome pathway demonstrating overlap between the two tumor sets; common pathways included secretion and xenobiotic responses (toxic environmental chemicals) as well as neurodevelopmental gene expression and alternative metabolic cycling (urea and TCA) consistent with a hypoxic phenotype (see Additional file 5: Table S1 and Additional file 6: Table S2). E. QPCR analysis of secretome-related transcripts in the independent set identified significant over-expression of all eight genes (ranging from APLP1 to SCN3A). *p <0.05 vs. normal mucosa. 3F. QPCR analysis of highly expressed transcripts in the independent set identified significant over-expression of ADCY2, AKT3 and ST18. Mean ± SEM, *p < 0.05 vs. normal mucosa. Tumors n = 13, normal mucosa n = 8.