Improved survival in metastatic melanoma is associated with immune genes expressed at the site of disease Ricardo D Lardone, Seema B Plaisier, Peter A Sieling, Delphine J Lee* Dirks/Dougherty Laboratory for Cancer Research, Department of Translational Immunology, John Wayne Cancer Institute (Santa Monica, CA 90404). *
[email protected]
INTRODUCTION Melanoma incidence has been increasing for the past 30 years. One of its most dangerous features is a high inherent metastatic potential compared to many other cancers. Metastatic melanoma (MM) patients have poor median survival rates, although some patients survive for many years after their diagnosis of metastatic disease. These MM outcome differences can be due to a combination of diverse biological factors involving the tumor cells and their relationship to host cells, including immune system cells. A better knowledge of these factors linked to MM long‐term survival is important for more effective therapy development and survival prediction. Factors and pathways relevant for survival in MM might be identified using high‐throughput gene expression analysis of tumors of patients with distinct survival characteristics. Even more powerful would be the comparison of multiple independent gene‐expression analyses. Often, the available methods to compare gene‐expression profiles test for correlation, overlap or enrichment between sets of genes. However, they consider only a small fraction of differentially expressed genes by applying strict thresholds for differential expression, while ignoring genes with small, yet reproducible changes. Instead, rank‐rank hypergeometric overlap (RRHO) analysis (Plaisier et al., 2010) uses the full range of consistent gene expression patterns between biological classes. This threshold‐free algorithm can identify an overlapping gene set with greatest statistical significance when comparing two independent high‐throughput gene expression profiles. Thus, weak but conserved gene expression changes are not excluded from downstream pathway analysis.
Figure 2: Immune‐related biological processes and pathways are enriched in GOS
Figure 4: B cells are enriched in LS MM group A: Cell type profiling of significantly overlapping good outcome genes dataset using “Gene Enrichment Profiler” database. Cell/tissue types were ranked according to the number of probes with enrichment score values >700. The top‐10 cell/tissue types from the database are shown. B: Representative examples of CD20‐IHC labeled FFPE sections from an independent group of MM showing more B cells infiltrates in LS patients compared to SS patients.
HYPOTHESIS Gene expression programs defining MM survival can be identified from coincident gene signatures obtained through multiple RRHO pairwise comparisons of independent high‐throughput gene‐ expression MM experiments. While defining relevant factors in the tumor microenvironment, these common signatures could also have outcome predictive potential and suggest novel therapeutic targets.
Figure 1: RRHO comparisons of three independent MM gene expression profile datasets identify a “good outcome signature” (GOS)
Figure 5: Predicted interactions network from GOS connect cell types in LS MM
A: Most highly enriched Gene Ontology (GO) terms by Gene Ontology Consortium. B: Top 20 canonical pathways identified by Ingenuity Pathway Analysis. C: Top 20 KEGG pathways identified by DAVID Bioinformatics Resource. Categories in each analysis are ranked by multiple hypothesis corrected p‐values.
Figure 3: Expression of GOS predicts outcome in an independent MM dataset Unsupervised hierarchical clustering analysis of the GOS in an independent Stage III dataset correctly separated most of the samples in accordance to their clinical outcome.
A Circos plot depicting a network of experimentally observed or high‐level‐of‐confidence predicted interactions was built with integrative information from STRING 9.1 (for interaction evidence) and the Gene Enrichment Profiler (showing cells expressing those nodes/genes) databases. Gene symbols were colored based on cell types showing the highest expression for that gene.
CONCLUSIONS A: Three complete MM gene expression profile datasets (“Set A”, “Set B” and “Set C”) divided into long‐term survival (LS) and short‐term survival (SS) groups were pairwise compared using the threshold‐free RRHO algorithm. RRHO analysis shows statistically significant overlap between the genes with increased expression in the LS class. RRHO of “Set A” and “Set B” datasets showed 394 overlapping genes (max log hypergeometric p‐value = 19.1, p‐value