Abstract 59: Identifying tumor driving signaling pathways for companion diagnostics using computational pathway models.

Proceedings: AACR 104th Annual Meeting 2013; Apr 6-10, 2013; Washington, DC Introduction Targeted drug treatment requires reliable companion diagnostics for therapy selection. Genomic and transcriptomic data can provide input for this, provided tools exist to convert this complex data into meaningful clinical information. We develop computational models of oncogenic pathways, to assess which one drives tumor growth in an individual patient and what is the causing (epi)genetic defect. Computational pathway models Based on a selection of experimentally validated direct target genes, we built initial models of the Wnt, ER, AR and Hedgehog pathways, covering their transcriptional program. We have modeled each pathway by a Bayesian network, which interprets the target genes’ mRNA levels (Affymetrix U133Plus2.0), and infers a probability that the respective pathway is active in a certain sample. Model parameters are based on literature insights and experimental data. Results A first Wnt model, calibrated on cell line data, validated perfectly on 32 normal colon samples and 32 colon adenomas from patients ([GSE8671][1]). A second Wnt model, calibrated on these 64 patient samples, correctly predicted no Wnt activity in all 44 normal colon samples, and Wnt activity in 97 of 101 colon cancer samples from [GSE20916][2]. Next, we tested the second Wnt model on other cancer types. On 25 breast cancer cell lines from [GSE12777][3] with known Wnt status, the model correctly identified the two samples with an active pathway. On two patient data sets ([GSE12276][4], n=204; [GSE21653][5], n=266) Wnt activity was predicted in a higher number of basal samples compared to other subtypes (p=0.021 and p=2.7e-5, respectively), in line with increasing evidence for Wnt activity in this subtype. Finally, tests on liver ([GSE9843][6], [GSE6764][7]) and medulloblastoma sets ([GSE10327][8]) confirm the power of these models to predict Wnt pathway activity. A first ER model was calibrated on estrogen-deprived and -stimulated MCF7 cell lines ([GSE8697][9]). Applied on breast cancer cell line data from [GSE21618][10], increased incidence of ER pathway activity was found in tamoxifen-sensitive cell lines compared to resistant ones. On breast cancer patient data ([GSE12276][4], [GSE9195][11], [GSE6532][12]) the model showed no pathway activity in ER- samples, and an active ER pathway in 26-38% of the ER+ samples. Within the latter group, model-predicted ER activity correlated with improved survival. Clinical utility studies to correlate ER activity to hormone therapy response are in progress. Finally, the AR model showed promising results on prostate cancer cell lines ([GSE34211][13], [GSE36133][14]), as did the Hedgehog model on medulloblastoma samples ([GSE10327][8]). Conclusion Our computational pathway models predict functional activity of oncogenic pathways for an individual patient based on mRNA data, complementary to existing molecular and histopathology staining tests. Clinical utility for therapy response prediction is currently being validated with clinical partners. Citation Format: Wim Verhaegh, Henk van Ooijen, Marcia Alves de Inda, Kalyan Dulla, Ralf Hoffmann, Dianne van Strijp, Pantelis Hatzis, Hans Clevers, Anja van de Stolpe. Identifying tumor driving signaling pathways for companion diagnostics using computational pathway models. [abstract]. In: Proceedings of the 104th Annual Meeting of the American Association for Cancer Research; 2013 Apr 6-10; Washington, DC. Philadelphia (PA): AACR; Cancer Res 2013;73(8 Suppl):Abstract nr 59. doi:10.1158/1538-7445.AM2013-59 [1]: /lookup/external-ref?link_type=NCBIGEO&access_num=GSE8671&atom=%2Fcanres%2F73%2F8_Supplement%2F59.atom [2]: /lookup/external-ref?link_type=NCBIGEO&access_num=GSE20916&atom=%2Fcanres%2F73%2F8_Supplement%2F59.atom [3]: /lookup/external-ref?link_type=NCBIGEO&access_num=GSE12777&atom=%2Fcanres%2F73%2F8_Supplement%2F59.atom [4]: /lookup/external-ref?link_type=NCBIGEO&access_num=GSE12276&atom=%2Fcanres%2F73%2F8_Supplement%2F59.atom [5]: /lookup/external-ref?link_type=NCBIGEO&access_num=GSE21653&atom=%2Fcanres%2F73%2F8_Supplement%2F59.atom [6]: /lookup/external-ref?link_type=NCBIGEO&access_num=GSE9843&atom=%2Fcanres%2F73%2F8_Supplement%2F59.atom [7]: /lookup/external-ref?link_type=NCBIGEO&access_num=GSE6764&atom=%2Fcanres%2F73%2F8_Supplement%2F59.atom [8]: /lookup/external-ref?link_type=NCBIGEO&access_num=GSE10327&atom=%2Fcanres%2F73%2F8_Supplement%2F59.atom [9]: /lookup/external-ref?link_type=NCBIGEO&access_num=GSE8697&atom=%2Fcanres%2F73%2F8_Supplement%2F59.atom [10]: /lookup/external-ref?link_type=NCBIGEO&access_num=GSE21618&atom=%2Fcanres%2F73%2F8_Supplement%2F59.atom [11]: /lookup/external-ref?link_type=NCBIGEO&access_num=GSE9195&atom=%2Fcanres%2F73%2F8_Supplement%2F59.atom [12]: /lookup/external-ref?link_type=NCBIGEO&access_num=GSE6532&atom=%2Fcanres%2F73%2F8_Supplement%2F59.atom [13]: /lookup/external-ref?link_type=NCBIGEO&access_num=GSE34211&atom=%2Fcanres%2F73%2F8_Supplement%2F59.atom [14]: /lookup/external-ref?link_type=NCBIGEO&access_num=GSE36133&atom=%2Fcanres%2F73%2F8_Supplement%2F59.atom