Mouse-human z-scores of motifs with experimentally verified ESS activity, which is deleterious for exons. Four of four of the Wang et al. [24] experimentally verified ESSs have negative z-scores, consistent with negative selection acting to remove them from coding regions. For the Zhang and Chasin [23] experimentally verified ESSs, nine of ten have negative z-scores (GTATTGTT, z = -0.006).

Comparison of COMIT z-scores to maximum likelihood K_s scores. There is a clear correlation between mouse-human z-scores and mouse-human K_s based on (a) naïve codon completion or (b) nucleotide-by-nucleotide K_s. These correlations indicate that the qualitative conservation of many motifs is not method-dependent.

Motif conservation is robust across the rat-mouse, human-dog, and elephant-tenrec lineages. This is visible when using nucleotide-by-nucleotide K_s values, or when using COMIT z-scores. Strong Spearman rank correlations (ρ) are observed in comparisons of both (human-dog)-(rat-mouse) and (elephant-tenrec)-(rat-mouse). Spearman correlations are considerably stronger for the COMIT z-scores, indicating the superiority of the method to K_s. For each of these comparisons the Spearman correlation is highly significant, with permutation test P-value < 0.00001.

Distribution of mouse-human COMIT z-scores. Motifs containing CpGs have systematically lower conservation scores (dotted red curve, mean = -12.8), while the remaining motifs have higher conservation scores (dashed green curve, mean = 5.2), consistent with hypermutability of CpG dinucleotides in mammalian genomes. The non-CpG distribution has an excess of motifs with high z-scores, as can be seen from the long rightward tail. This suggests that selection has acted to maintain sequence conservation of these motifs across species.

Conservation of hexamer submotifs of the yeast ACS DNA replication origin motif. Of the 24 hexamers, 21 are consistent with the ACS consensus TKTTTATRTTTWGTT and have positive z-scores in comparisons of S. cerevisiae and S. bayanus coding sequences. These results support the hypothesis that ACS motifs in coding regions are functionally active, and also indicate that COMIT is capable of detecting coding motifs functional at the DNA level.

Schematic of the COMIT algorithm for identifying unusually conserved motifs in coding regions. The example illustrates how the score would be calculated for the motif ACAAAG, using genome-wide coding sequence alignments for two species. Each instance of the motif is identified in species 1, and the observed conservation - that is, whether all bases are identical among the two species - is calculated. The expected conservation at each instance is modeled from genome-wide frequencies of nucleotide-level conservation patterns conditional on the aligned amino acids. For each instance, the expected conservation is calculated from all possible ways in which the motif could be conserved at that location given the amino acids in each species, using values from Table 1 (typically some of these quantities, such as (H, Y)₁₁₁, will be zero). The observed and expected conservation levels are compared and normalized to yield a conservation z-score for each motif.

Experimentally verified ESEs are preferentially conserved by natural selection. (a) Motif z-scores (greater z indicates greater conservation), based on mouse-human comparisons, show a strong quantitative correlation (Spearman ρ = 0.422, permutation test P-value = 0.045) with splicing inclusion rates as measured experimentally in [4]. (b) K_s values, also based on mouse-human comparisons, show a far weaker correlation (ρ = -0.0725, P-value = 0.606). Black lines indicate regression fits. While the two motifs with the highest splicing inclusion rates do exhibit below-average K_s values, this is the only apparent effect, indicating that COMIT scores are better for assessing functional motifs.