Fig. 8. Model outlining the evolutionary mechanism of Hox PG1 gene ‘function shuffling’. The cis-regulatory elements characterized for the mouse and human Hoxa1 and Hoxb1 genes [3' RAREs (blue), Hox/Pbx binding sites (red)] are assumed to be present in the ancestral, pre-’third’-duplication, condition. We also postulate the presence of a regulatory domain directing midbrain expression of Hoxa1 (purple), although no such domain has yet been characterized. The duplication event in the lineage leading to teleosts produced redundant copies of both Hoxa1 and Hoxb1 in an ancestor of the zebrafish. The hoxa1b duplicate was eventually lost by accumulation of deleterious mutations (‘non-functionalization’) as predicted by classical models. By contrast, the hoxb1a and hoxb1b genes accumulated complementary degenerative changes in their cis-regulatory elements, such that hoxb1a lost early RARE-mediated expression and hoxb1b lost autoregulation. This led to retention of the duplicate genes, as both were required to maintain the expression pattern and function of the single Hoxb1 ancestral gene (sub-functionalization), as predicted by the DDC model. As hoxa1a and hoxb1b shared similar coding sequences and expression patterns, these two genes were now functionally redundant with respect to a role during gastrulation in setting up segmental organization of the hindbrain. These non-orthologous genes were thus able to go through another ‘sub-functionalization’ event, such that hoxa1a lost its early RARE-mediated expression, which was retained by hoxb1b. Thus, hoxb1b became essential for proper hindbrain segmentation, the role played in the ancestral state by Hoxa1. Retention of the hoxa1a gene in the lineage leading to zebrafish was presumably dependent on a function that was not redundant with hoxb1b, possibly a role in midbrain patterning. We term this rearrangement of PG1 gene roles ‘function shuffling’.