Abstract

© 2015 Collins et al. When we expressed a small nonprotein-coding transcript derived from the mitochondrial VS plasmid in the nucleus of Neurospora we found that it was efficiently spliced at one or more of eight 5′ splice sites and ten 3′ splice sites, which are present apparently by chance in the sequence. Further experimental and bioinformatic analyses of other mitochondrial plasmids, random sequences, and natural nuclear genes in Neurospora and other fungi indicate that fungal spliceosomes recognize a wide range of 5′ splice site and branchpoint sequences and predict introns to be present at high frequency in random sequence. In contrast, analysis of intronless fungal nuclear genes indicates that branchpoint, 5′ splice site and 3′ splice site consensus sequences are underrepresented compared with random sequences. This underrepresentation of splicing signals is sufficient to deplete the nuclear genome of splice sites at locations that do not comprise biologically relevant introns. Thus, the splicing machinery can recognize a wide range of splicing signal sequences, but splicing still occurs with great accuracy, not because the splicing machinery distinguishes correct from incorrect introns, but because incorrect introns are substantially depleted from the genome.