New Study Brings Mouse Cancer Models Closer to Humans
Before a new drug, device, or surgical procedure can be tested in humans, each must first be tested in animals, the usual preclinical biosystem being mice. Genetic engineering has produced hundreds of strains of mice for this purpose, with cancers that look the same and develop in the same organs or tissues as tumors seen in people. But the question remains: How similar are these tumors in terms of their signaling pathways and gene transcription? How well do they mimic the human cancer biosystem?
"Before now, the criteria that have been used to select mouse models for cancer research were most often based on a single gene, one that is mutated or lost in a particular human cancer," explained Dr. Snorri Thorgeirsson of NCI's Center for Cancer Research (CCR). "But extrapolating findings from mouse models to the human situation is complicated because of the molecular heterogeneity seen in humans."
To better match mouse cancer models to human cancers and take this heterogeneity into account, NCI scientists including Dr. Thorgeirsson and researchers from Northwestern University's Feinberg School of Medicine in Chicago focused on a subset of genes in human hepatocellular carcinoma (liver cancer) that are associated with survival, and compared their expression patterns with those seen in mouse hepatocellular carcinoma - a process they aptly named "comparative functional genomics."
Dr. Thorgeirsson and colleagues worked with seven different mouse models induced to develop liver cancer: two through the chemicals ciprofibrate and diethylnitrosamine; four through transgenic overexpression of the genes Myc, E2fl, and Tgfa; and one strain, a knockout, missing the Acoxl-|- gene. After collecting tissue samples from 68 mouse tumors, the researchers used microarray analysis and grouped the samples according to gene expression prevalence.
Gene expression profiles from these 68 tumor samples fell into three clusters: One included samples from mice that were exposed to diethylnitrosamine, a carcinogen, and those that overexpressed Myc and Tgfa; the second included mice that overexpressed Myc and E2fl; and the third included those exposed to ciprofibrate, a peroxisome proliferator, as well as those that overexpressed Acoxl-|-.
The researchers then matched these clusters with the gene expression profiles from 91 human hepatocellular carcinoma samples that had been preclassified according to survival-correlated gene expression patterns. They found that cluster 1 mouse tumors matched profiles from humans who had poor survival rates (categorized earlier, in the September 2004 issue of Hepatology, as subclass A), the cluster 2 tumors matched up with those from humans who had better survival rates (categorized as subclass B), and cluster 3 had low correlation with human liver cancers. Other aspects that were similar between the mouse and human tumors in each cluster included proliferation and apoptosis rates, as well as the degree of ubiquitination - a process of marking proteins for degradation - seen between the two.
In their article based on this research, published in the December 2004 issue of Nature Genetics, Dr. Thorgeirsson and colleagues wrote, "Taken together, these results support the notion that better- or best-fit mouse models for human studies can be identified by applying genome-scale comparison of gene expression patterns." They also pointed out that the lack of correlation between human tumors and mouse tumors induced by ciprofibrate support previous studies that showed the human liver may be insensitive to peroxisome proliferators.
While the tissue in this study came from hepatocellular carcinomas, Dr. Thorgeirsson said that the findings hold promise for preclinical research on a wide range of human cancers.
In an editorial related to this research, published in the January 2005 issue of Nature Genetics, Drs. Thomas Graeber and Charles Sawyers of the UCLA School of Medicine commend the study's findings, but note that this type of research could miss crucial events in cancer development that might not be reflected in gene expression. "Although we expect gene expression-guided mouse modeling to advance cancer biology," they wrote, "we wonder whether the mouse modeling and computational biology communities should consider a more comprehensive approach." Dr. Thorgeirsson agrees with their assessment and, to that end, says that his future research plans include complementing comparative functional genomics with both computational and proteomic approaches.
By Brittany Moya del Pino