Multidomain proteins are characterized by a mosaic of sequence fragments that encode structural or functional modules, called domains.  A protein's domain architecture (i.e., its constituent domains in N-terminal to C-terminal order) is integral to its function, because different constituent domains are responsible for different subfunctions, for example, recognition and transmission of an environmental signal..  Multidomain architectures are the product of mutational processes that generate new combinations of domains, which in turn depend on gene and genome organization. Multidomain architectures vary substantially across different taxonomic lineages, as do genomic and cellular properties.  This project examined how gene architecture shapes multidomain protein evolution and whether it does so in a lineage-dependent manner. 

The exon shuffling hypothesis states that new domain combinations arise through intronic recombination. More generally, this hypothesis posits a link between gene and protein architecture.  If the exon shuffling hypothesis is true, then introns should co-occur with domain boundaries more often than expected by chance.  Prior studies tested this hypothesis on a broad range of eukaryotic genomes and concluded that gene structure contributes to the formation of new domain architectures and does so throughout the eukaryotes.  However, those studies did not examine the relationship between gene architecture and domain architecture in different eukaryotic lineages. 

We investigated domain-intron co-occurrence in 16 fungal and 5 metazoan genomes. Fungal genomes possess a broad range of intron sizes and frequencies, making this a good comparative system.  We assessed the evidence for exon shuffling in the data set as a whole, and in the animal and fungal genomes separately. 

Prior studies estimated the expected number of intron-domain boundary co-occurrences under the assumption that introns are uniformly distributed across genes.  While computationally tractable, this null model is biologically unrealistic.  The positions of introns in genuine genes are not even approximately uniformly distributed. For this project, we designed and implemented two strategies for randomizing intron positions that provide good approximations of the genuine intron positioning without excessive computational costs. We carried out an explicit comparison of the uniform null model and our two new models with respect to the implied exon length distributions and the propensity to reject the null hypothesis.  We observed that uniformly distributed intron positioning results in a skewed exon length distribution with an excess of short exons. In contrast, exon lengths obtained with our randomization strategies closely approximate the genuine exon length distribution. 

We further observe that the three models differ substantially in their ability to reject the null hypothesis.  Relative to our methods, the uniform model dramatically underestimates the expected frequency of agreement between introns and domain boundaries.  This, in turn, exaggerates the significance of domain intron co-occurrence in the genuine data.  In animal genomes, the association between domains and introns was significant with all null models, suggesting that this association is so strong that it exceeds chance expectation even when the null model of intron positioning is inaccurate.  In contrast, only half of the fungal genomes tested displayed a significant association with all three models.  Moreover, even when this association is significant, the magnitude of the association is small. In the vertebrate genomes tested, 21% of all domain instances are flanked by introns at both ends. In contrast, only 7% of invertebrate and 3% of fungal domains satisfy this criterion.

In summary, the results of this project challenge the idea that exon shuffling has played an important role throughout the eukaryotic tree.  More accurate statistical tests developed for this project reveal that the genome-scale statistical association between gene architecture and domain architecture in fungi is weak and the effect size is small, suggesting that the contribution of exon shuffling to multidomain protein evolution in fungi is minor. This suggests that the evolutionary processes that shape multidomain protein evolution differ in different eukaryotic lineages.  The evidence for exon shuffling outside of metazoa should be revisited using more sensitive statistical models.

This project contributed to national scientific resources in the form of new, publicly available software, including a tree-based method for reconstruction of domain shuffling events and randomization heuristics that support realistic, conservative statistical tests without excessive computational costs.  This project has provided research training for undergraduate and doctoral students and postdoctoral fellows. 

Last Modified: 11/30/2022
Modified by: Marie Dannie Durand