[i] All mouse, rat, and human RefSeq genes for which an alignment was possible were analyzed for the presence of ATG codons, as follows. Following alignment of genes for the three species, the segment 1-kb upstream of the promoter (as defined by RefSeq) was analyzed in two segments, from − 1 to − 600 and − 601 to − 1000 bp. Within each segment, 100-bp windows were scanned for the presence of an ATG triplet. An ATG desert was scored if the scan met the criteria that (1) the − 1 to − 700-bp window had a median ATG frequency of ≤1.4 in at least two of the three species, and that the − 601 to − 1000-bp window had a median ATG frequency of ≥1.6 and that this pattern was observed in at least two of these species. P-value <2.2 × 10⁻¹⁶.