Since ASEs in one species may be CSEs in the other (i.e. lineage-specific ASE/CSEs), it is necessary to specify the splicing pattern of the exons studied. Therefore, we classify ASEs into three major groups according to splicing pattern conservation (see Materials and Methods). Each group is subsequently divided into four subsets (Table ). We then compare the frequencies of reading frame preservation (designated as "FRFP", i.e., the proportions of exons of which the lengths are divisible by 3) between simple and complex ASEs (Fig. ). For Group A, the FRFPs for human (mouse) simple and complex ASEs are 43.0% (45.7%) and 37.9% (35.5%), respectively. In comparison, the FRFPs of CSEs approximate 40% (39.7% in human and 39.5% in mouse [22
]). It has been well recognized that CSEs have lower FRFPs than ASEs. However, we find that although this is true for simple ASEs (P
-values < 0.01 in both human and mouse; all statistical tests used in this section are the Fisher's exact test), it does not seem to hold for complex ASEs. The FRFPs are higher in CSEs than in complex ASEs in both species, though the differences are not highly significant (both P
-values > 0.01). Overall, our results indicate that simple and com1plex ASEs are under opposite selection pressure for protein reading frame preservation. Particularly, complex ASEs differ significantly in FRFP from commonly regarded ASEs, which are dominated in number by simple ASEs. We then extract conserved ASEs (Group C) from Group A. Note that "conservation" here refers to the conservation of the ASE/CSE splicing pattern between human and mouse, rather than the simple/complex pattern. We find that the FRFP of Group C simple ASEs increases to 49.8% in human and 53.4% in mouse (Fig. ). Meanwhile, for Group C complex ASEs, the FRFPs decrease to <35% (34.3% for human; 33.3% for mouse) (Fig. ). It is obvious that simple ASEs in Group C have higher FRFPs than in Group A, whereas the reverse is true for complex ASEs in both human and mouse. We then compare the FRFPs of simple and complex ASEs with those of CSEs. For simple ASEs, Group A has lower FRFPs than Group C, while both groups have higher FRFPs than those of CSEs (Fig. ). However, for complex ASEs, the trend is reversed. Even if the expected FRFP of CSEs is set as 45% [32
], the trends still hold well in conserved ASEs. Therefore, simple and complex ASEs seem to cause FRFP changes to the opposite ends when compared with CSEs. Note that the "CSEs" stated above are those with unspecified splicing pattern conservation. We therefore retrieve 21,669 pairs of conserved CSEs for comparison. The FRFPs of conserved CSEs are 38.4% in human and 38.3% in mouse, respectively. These figures further confirm our observations that CSEs tend to have higher FRFP than complex ASEs but lower FRFP than simple ones. Overall, our result supports Chen et al's suggestion that simple and complex ASEs cause evolutionary changes to the contrary ends with CSEs in-between [31
The retrieved human-mouse orthologous exon pairs
Comparisons of frame-preserving frequencies of simple ASEs and complex ASEs in ASE conservation unspecified group (Group A), lineage-specific ASE group (Group B), and conserved ASE group (Group C).
Comparisons of frame-preserving frequencies of simple ASEs, complex ASEs, and complex+ flanking exons in Groups A and C.
To further probe the effects of splicing pattern conservation on frame preservation, we compare the FRFPs between conserved and lineage-specific ASEs (Groups C and B). As shown in Figure , for simple ASEs, conservation of ASE/CSE splicing pattern results in an increase in FRFP. In contrast, splicing pattern conservation causes the FRFP to drop in complex ASEs, such observation disobeys the previous view [18
] that conserved ASEs have a higher probability to be frame-preserving than lineage-specific ones.
On the other hand, also see Table , we find that >70% of the ASEs (either simple or complex) have CSE counterparts in the other species, indicating that AS patterns tend not to be evolutionarily conserved in human and mouse. If only conserved ASEs are considered, the simple splicing pattern has a much higher probability of being conserved between human and mouse than the complex splicing pattern (Table ). The result indicates that most complex ASEs are lineage-specific.
Classification of conserved ASEs in terms of simple/complex splicing pattern
Another issue of interest is that, since a complex ASE looks like a simple event plus one (or two) exon extension/truncation event(s) (see Fig. ), the FRFPs of complex ASEs may in fact reflect the effects of exon extension/truncation events. However, as shown in Table , we find that the FRFPs in the lineage-specific exon extension/truncation events are around 50%, whereas in conserved events, the FRFPs significantly increase to over 60% (both P-values < 0.001; Table ). Such an increase in FRFP towards conserved ASEs is similar to what is observed in simple ASEs. Therefore, exon extension/truncation events and complex ASEs may be under different selection pressures for reading frame preservation. We speculate that a complex splicing event is rather an integrated "module" that requires synchronized changes in neighboring exons, than merely a simple ASE accidentally coupled with exon extension/truncation events. To find support for this hypothesis, we further analyze whether the length changes caused by complex ASEs and their flanking exons can offset the frame-shifting effects of each other and retain the upstream reading frame. We find that the FRFPs of complex ASEs coupled with flanking exons (complex+flanking exons) are close to those of simple ASEs (Fig. ). In Group C, the FRFPs of complex+flanking exons (49.2% in human and 47.8% in mouse) are significantly higher than those of conserved CSEs (dashed lines in Fig. ; both P-values < 0.01). Therefore, the selection pressure for frame preservation may apply to transcripts as a whole, but not to complex ASEs per se. Furthermore, our results imply that in an alternatively spliced transcript, neighboring exons of an ASE may evolve in a coordinated way to avoid protein dysfunction.
The frame-preserving frequencies of exon extension/truncation AS events