Previous studies of the evolution of genome complexity have primarily relied on correlations between the abundances of various genomic elements (introns, transposons, gene size, etc.) and the product of the effective population size and mutation rate [8–13]. However, claims of causality based on such correlative analyses are always inconclusive, because other potentially important factors can never be excluded. In an attempt to circumvent some of the limitations of the correlative approach, we explored the evolution of genomic complexity in a more direct manner, by tracing the evolutionary dynamics of nested pairs of protein-coding genes in animals. This study covers only one, perhaps not even the most common, class of interleaved gene arrangements because we left out the numerous intron-contained small RNA genes [14]. Nevertheless, even this limited analysis clearly reveals the ongoing increase of the organizational complexity of animal genomes and suggests that this process occurs via a nonselective route.

By examining gene annotations and constructing sequence alignments, we identified the closest species with a completely sequenced genome in which each nested gene structure was absent. Absence of the nested structure in an appropriate outgroup species indicates its emergence (gain) in the respective lineage, whereas presence of the nested structure in the outgroup indicates its loss (Figure 1). Gains were found in all three taxa, with the emergence of 55 internal genes in at least 40 independent events in vertebrates, 52 internal genes in at least 48 events in Drosophila and 22 internal genes in as many events in Caenorhabditis. The rate of these acquisitions was approximately uniform throughout the course of evolution (Figure 2). By contrast, losses of nested gene structures were much rarer, with none detected in vertebrates, 17 in Drosophila and 2 in Caenorhabditis.

By comparing the gene structures and encoded protein sequences of internal and external genes to complete gene sets from the respective species, we deduced the mechanisms of formation of nested gene structures in vertebrates (Table 1). Nearly all nested gene structures seem to have emerged by insertion of a DNA sequence, which arose by gene duplication or retrotransposition, into an intron of a pre-existing gene. The origin of an internal gene was classified as a retrotransposition when it was intronless in a given species, whereas its non-nested ortholog in a sister species contained introns. A duplication at the DNA level was inferred when both the internal gene and a non-nested ortholog in a sister species had introns. In cases where the internal gene and a non-nested ortholog were both intronless, retrotransposition and duplication at the DNA level could not be discriminated. Five internal genes in humans are candidates for de novo origin from intron sequences (see Supplementary Data), including one gene with no sequence similarity beyond apes [placenta-specific 4 (PLAC4)] and another with no similarity beyond old world monkeys [saitohin (STH)] (Table S1). Analysis of the 12 recently sequenced Drosophila genomes showed that the majority of de novo genes originate in introns [15]. Consistent with this observation, we found 11 internal genes in D. melanogaster with no sequence similarity to any genes in the genome of the closely related D. yakuba. We did not identify any nested gene structures that evolved via the remaining two scenarios.

To discriminate between these three hypotheses, we analyzed gene expression data from human and D. melanogaster genomes (see Supplementary Material). We compared correlations of gene expression in 109 and 752 nested gene pairs in humans and D. melanogaster, respectively, to 1000 random sets of 109 and 752 adjacent gene pairs from corresponding genomes. There was no significant difference in mean correlation coefficients of gene expression levels between nested and adjacent genes in either human (0.33 ± 0.03 for nested and 0.33 ± 0.0008 for adjacent pairs) or D. melanogaster (0.041 ± 0.014 for nested and 0.030 ± 0.00046 for adjacent gene pairs), which is consistent with the neutral hypothesis. The observation that external genes have substantially more and longer introns than average in the respective species (Ref. [7] and Supplementary Material) is also compatible with the neutral hypothesis. Furthermore, examination of the available functional information for nested gene pairs (Table S1) did not reveal any obvious connections [7]. Fixation of originally neutral or even slightly deleterious sequence segments, such as introns and transposable elements, through genetic drift acting in relatively small populations is a common phenomenon in eukaryotic evolution that might be partially responsible for the evolution of complex phenotypes [8–13]. The increase in organizational complexity of intron-rich genomes via emergence of nested gene structures seems to be another facet of this process.