Molecular signaling network complexity is correlated with cancer patient survivability

Abstract

The 5-y survival for cancer patients after diagnosis and treatment is strongly dependent on tumor type. Prostate cancer patients have a >99% chance of survival past 5 y after diagnosis, and pancreatic patients have <6% chance of survival past 5 y. Because each cancer type has its own molecular signaling network, we asked if there are "signatures" embedded in these networks that inform us as to the 5-y survival. In other words, are there statistical metrics of the network that correlate with survival? Furthermore, if there are, can such signatures provide clues to selecting new therapeutic targets? From the Kyoto Encyclopedia of Genes and Genomes Cancer Pathway database we computed several conventional and some less conventional network statistics. In particular we found a correlation (R(2) = 0.7) between degree-entropy and 5-y survival based on the Surveillance Epidemiology and End Results database. This correlation suggests that cancers that have a more complex molecular pathway are more refractory than those with less complex molecular pathway. We also found potential new molecular targets for drugs by computing the betweenness--a statistical metric of the centrality of a node--for the molecular networks.

Fig. 1.

Protein–protein interaction network for basal cell carcinoma as constructed from the KEGG pathway processed by KEGGgraph. Node labels are HGNC gene symbols. Yellow nodes are the top three highest in betweenness centrality.

Fig. 2.

Scatter plot showing the correlation between cancer 5-y survival probability and degree-entropy. Datapoints are shown for 13 cancer sites, including all of the cancer sites with KEGG pathways, except prostate cancer. The x axis is the 5-y survival probability for the cancer site, from the SEER database. The y axis is the degree-entropy (H) for the cancer site, calculated from the KEGG pathway. The line is a linear regression fit, with R² = 0.7. The error bars are set at fixed estimates of the error, with the x axis error bars fixed at ±10% and the y axis error bars fixed at ±0.1.