Abstract
Segmentation is a common feature of disparate clades of metazoans, and its evolution is a central problem of evolutionary developmental biology. We evolved in silico regulatory networks by a mutation/selection process that just rewards the number of segment boundaries. For segmentation controlled by a static gradient, as in long-germ band insects, a cascade of adjacent repressors reminiscent of gap genes evolves. For sequential segmentation controlled by a moving gradient, similar to vertebrate somitogenesis, we invariably observe a very constrained evolutionary path or funnel. The evolved state is a cell autonomous 'clock and wavefront' model, with the new attribute of a separate bistable system driven by an autonomous clock. Early stages in the evolution of both modes of segmentation are functionally similar, and simulations suggest a possible path for their interconversion. Our computation illustrates how complex traits can evolve by the incremental addition of new functions on top of pre-existing traits.