Abstract 2778: Tumor target vs. tissue of tumor origin: cluster analysis of genomic profile of 42 human tumor in vitro and in vivo models.

Traditionally, drug development has relied upon testing cancer drug candidates in cell lines. Active drugs are then tested in human tumor xenograft models, usually selected based upon the cell lines in which the drug showed activity. The majority of drugs fail at this stage as they do not show activity in the xenograft models chosen. We performed Affymetrix genomic analysis on 42 human tumor xenograft models and the original cell lines from which they were established. The genomic profiles obtained underwent Unsupervised Hierarchical Cluster Analysis to ascertain which cell lines and xenograft models had similar genomic profiles and which did not. The analysis showed that only 24 of 42 human tumor xenograft models clustered side-by-side with the cell line from which they were established. All 6 human leukemia/lymphoma xenograft models clustered very well with the cell lines from which they were established, and they clustered perfectly according to histological class. Five out of six human colon tumor xenograft models clustered well with the cell lines from which they were established and according to histotype. Of the 18 xenograft/cell line pairs that did not cluster side-by-side, 10 pairs remained in the same general cluster, whereas the partners of 8 other pairs were dispersed across different major clusters. Ovarian, breast, melanoma, and pancreatic human tumor xenograft models did not cluster according to histotype. Our data may explain why some drugs that show in vitro activity in some cell lines are not active in other cell lines of the same histological type, and also why some drugs that show activity in vitro then fail in xenograft models. In our laboratory, the PANC-1 cell line is very often chosen as a model of pancreatic cancer. A drug showing activity in the PANC-1 cell line would next be tested in other in vitro models of pancreatic cancer (e.g., MIA PaCa-2, CFPAC-1, and BxPC-3). However, none of these other pancreatic models have a similar genetic profile to PANC-1. Based upon our data, the cell line showing most similarity to the PANC-1 cell line is the breast cancer cell line MDA-MB-231. It is our suggestion that a drug showing activity in the PANC-1 cell line should be tested in other cell lines showing similar genetic profiles, not in cell lines based on histotype. Another example from our analysis is the LOX-IMV1 melanoma cell line. Not only does this cell line not cluster with its corresponding LOX-IMV1 xenograft model, it clusters most closely with the NCI/ADR-RES ovarian cell line. In summary, the genomic profiles of approximately 57% of the tumor xenograft models analyzed closely associate with the cell line from which they were established. Some of the tumor xenograft models show very little similarity to the cell lines from which they were established. Additionally, many of the models (both xenografts and cell lines), do not cluster according to their tissue of origin. Citation Format: Michael J. Roberts, Michael S. Koratich, Murray Stackhouse, Richard D. May, Andrew D. Penman, Tommie A. Gamble, Kristy L. Berry, Joseph F. Murphy, Robert J. Rooney, Yulia Y. Maxuitenko. Tumor target vs. tissue of tumor origin: cluster analysis of genomic profile of 42 human tumor in vitro and in vivo 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 2778. doi:10.1158/1538-7445.AM2013-2778