One of the main issues of the mammalian genome evolution is the “isochore:” a queer spatial structure of the genome in GC content. So far, no flawless explanations were achieved due to contradictory observations against the proposed models: e.g., selection, biased gene conversion, and mutation bias models. This indicates that the conventional frameworks of the molecular evolution may be inadequate to understand the fundamental mechanisms harbored in the isochore evolution. The typical approach to study molecular evolution is the “backward” analysis by using extant molecular data: we infer the past by extrapolating relatively recent evolutionary patterns based on observable data. In the isochore evolutionary study, however, this conventional approach has at least two kinds of issues: (1) critical evolutionary signal for the isochore evolution might be easily eroded during the evolutionary process; (2) to directly analyze the ishcore evolution, we need to handle non-coding regions, for which conventional evolutionary models may be inappropriate. To overcome those problems, we took advantage of ENU mutagenesis as a tool for the experimental evolution, accelerating evolutionary rate to make it possible to observe ongoing evolution through a large number of de novo mutations. In other words, we attempted “forward” analysis on the genome evolution. We found bidirectional (diverged) mutation pressures that support the legendary Sueoka’s mutation bias theory. Our finding has potential to explain the enigmatic isochore evolution.