Mathematical modeling of radiation response Free

3804

Background/Purpose: To study radiation response from a systems biology perspective, we applied three mathematical models to the analysis of gene expression (GE) profiles and radioresponse (Survival Fraction at 2 Gy, SF2) in a 48 cell line database.

Methods: We developed two mathematical models in addition to the successful model of GE previously tested, to describe the correlation of SF2 and the biological variables: GE, ras mut/wt, tissue of origin (TO) and p53 mut/wt. We evaluated each model by determining the linear fit of the predicted and measured SF2 curve. Analysis of Variance (ANOVA) is used to determine the significant biological interactions. The models can be represented by the following equations:

GE model (GEM):SF2 = k₀ + k₁(y₁)

Additive model (AM):SF2 = k₀ + k₁(y_x) + k₂(tissue type) + k₃(ras status) + k₄(p53 status)

Interactive model (IM): SF2 = AM model + k₅(y_x)(tissue type) + k₆(y_x)(ras status) + ¼ where y represents GE and k_x are constants determined by fitting the model.

Results: Both models improved the ability of GE to fit the radioresponse linear regression model (R² = 0.3, 0.4 and 0.8 for the GEM, AM and IM models respectively). ANOVA identified TO and ras mut/wt but not p53 as the key interactions in the AM and IM model. We clustered the top 500 genes in the IM based on the significant biological interaction, identified by ANOVA. This identified three major clusters: 1. ras -mut cell lines, 2. ovarian and lung cancer (ras-wt) cluster and 3. TO cluster. GeneGo analysis identified pathways in each cluster that have been previously implicated in radiation signaling. Importantly, the only common pathway represented in all three clusters was regulated by c-jun/AP-1, suggesting that it might be a common key element in radioresponse. ANOVA identified TO as the main biological interaction of c-jun gene expression. Analysis by TO revealed three different correlations (direct, indirect and none) between c-jun expression and SF2. We tested the hypothesis suggested by the model that c-jun influences radioresponse in a tissue-dependent manner. The model predicted that downregulation of c-jun would induce radioresistance in lung cancer, no change in colon cancer and radiosensitization in breast cancer cell lines. siRNA downregulation of c-jun in 8 cell lines resulted in an overall induction of radiation resistance in the cell lines when analyzed as a group (p=0.045). The model was experimentally validated with two of three predictions confirmed as correct: Lung cancer cell lines (radioresitance, SF2 siRNA-cjun vs. siRNA-control, 0.59 vs. 0.47, p=0.005) and colon cancer cell lines (no change). Cell survival curves confirmed the induction of radioresistance in lung cancer cell lines by c-jun downregulation.

Conclusion: Mathematical modeling of radiation response allows the identification of biologically relevant data from genomic analysis. This approach to functional genomics may prove key in the understanding of the radiation response network