Understanding tumor initiation is critical for developing early detection and cancer prevention strategies. However, this problem has been virtually impossible to study because pre-malignant mutant cells are indistinguishable from their normal counterparts. To overcome this hurdle, our lab uses a mouse genetic system termed Mosaic Analysis of Double Markers (MADM) that, from a non-labeled mouse heterozygous for a tumor suppressor gene (TSG), generates GFP-labeled TSG-null cells and RFP-labeled sibling wildtype cells. The sparseness of mutant cells due to the low frequency of inter-chromosomal recombination closely mimics the clonal origin of human cancer, and the definitive correlation between color and genotype greatly facilitate the analysis of the earliest tumorigenic events. In particular, increased ratio of green to red cell numbers (G/R ratio) serves as the clear indication of initial expansion of mutant cells. Malignant glioma is the deadliest type of cancer in the brain. One devastating fact is the inevitable progression of benign grade II tumors into full malignancy within a few years. Previously our lab identified oligodendrocyte precursor cells (OPCs) as its cell-of-origin because MADM revealed a massive expansion of pre-malignant mutant OPCs (G/R ratio > 100) long before malignant transformation. Surprisingly, the overall OPC density barely changed in the mutant brain, suggesting that the pre-malignant expansion of mutant OPCs was achieved at the cost of WT OPCs. Using in vivo time course analysis and in vitro co-culture experiments, we determined that the competition is mediated via active killing of WT OPCs by mutant ones. These observations well explain the inevitable progression of glioma since any individual OPC gaining an advantageous mutation would outcompete less fit ones. By repeating this process, a mutant OPC would eventually gather all necessary mutations for the full-blown transformation. This finding also predicts the unavoidable failures with cytotoxic treatment methods because they merely provide selective pressure on the evolutionary process of the tumor mass, which eventually lead to the enrichment of most resistant tumor cells. Therefore, an effectiveness treatment must deal with cell competition between mutant and WT OPCs. In a proof-of-principle experiment, we used a genetic model to remove the competitiveness of mutant OPCs, and successfully blocked glioma formation. Different from conventional studies that are mainly focused on cell intrinsic gene functions during tumorigenesis, MADM enabled our discovery of non-cell-autonomous mechanism for glioma progression, which should lead to completely novel therapeutic strategies.