Pre-mRNA splicing of eukaryotes requires three basic signals (splicing motifs) for the recognition of introns. The splicing motifs are the 5' intron end (donor) and the 3' intron end (acceptor), and the branch site. The splicing motifs at the 5' and 3' splice sites, known as "ag|GTragt" ("|" is the splice junction; "r" is a or g) and "(y)12-17nAG|g" ("y" is c or t; "n" is a, t, g or c; and subscript indicates the repeat number) [1, 2]. A human expressed sequence tag-based study showed that 99.24% and 0.69% of introns are flanked by GT-AG and GC-AG dinucleotides (splice dinucleotides), respectively . Other types of splice dinucleotides are also found in the human genome; these are AT-AC(0.05%) and others (0.02%) . Irrespectively of these variations at the splice dinucleotides, there are two well-studied splicing mechanisms . One mechanism utilizes a major spliceosome, an assembly of five small nuclear ribonucleoprotein particles (U1, U2, U4, U5, and U6 snRNP). The other mechanism uses the minor spliceosome, consisting of U11, U12, U4atac, U5, and U6atac, instead. Thus, most of exons are flanked by the virtually "invariant" GT and AG dinucleotides (splice dinucleotides) . Other additional splicing motifs, such as enhancers and silencers located in exons and introns have vast variety in motif signals and locations, but contribute to splicing fidelity.
Some single base-pair substitutions occurring at the "invariant" splice dinucleotides cause alteration of splice patterns and are associated with serious diseases, for example, NF1 , GSTM4 , cyclin D1 , NUDT1(MTH1) , and LDLR , (for review, see [10, 11]). The Human Genome Mutation Database (HGMD) at the Institute of Medical Genetics in Cardiff [12, 13] has annotated a total of 9267 entries for mutations in the vicinity of splice sites, which include 2362, 756, 1199, and 1355 entries for mutations at splice dinucleotides at sites '+1(G)', '+2(T)', '-2(A)', and '-1(G)', respectively . The databases DBASS5  and DBASS3  contain 431 and 283 details of aberrant splice sites, respectively , which are generated as a result of disease-causing mutations in humans.
While other single base-pair substitutions at splicing dinucleotides are known to be maintained as single nucleotide polymorphisms (SNPs) in human populations [17, 18], the question is: "what determines whether a single base-pair substitution at a splice dinucleotide will be maintained as a SNP or eliminated from the population?"
To address this question, we evaluated SNPs at splice dinucleotides (sdSNPs) in the context of selective pressure in the course of evolution. Generally, functional constraints on exons differ between alternatively spliced exons (ASEs) and constitutively spliced exons (CSEs). ASEs are subject to relaxed negative selective pressure, which is suggested by their significantly higher Ka/Ks values compared with other exons . This relaxation is the most fundamental conceptual constituent of exon creation via alternative splicing (AS) and was first proposed at the time of the discovery of the exon/intron structure in the 1970s . Recent studies revealed that AS is an important mechanism for creating new exons [21–24] and that accelerated accumulation of SNPs at additional splicing motifs after gene duplication enhances exon generation .
Current advances in genome informatics and comparative genomics demonstrate that ASEs can be sub-divided into two contrasting categories. When ASEs are classified as conserved or non-conserved in exon structure, low synonymous rates are characteristic of conserved ASEs but not of those with non-conserved exonic structure . Moreover, when they are classified as boundary-shifting (complex) ASEs or non-boundary shifting (simple) ASEs (those of the former type change the exon/intron boundaries of the flanking exons whereas those of the latter type do not), complex ASEs are under stronger selection pressure at the amino acid level but less pressure at the RNA level than CSEs, while reverse trends were observed in simple ASEs [27, 28]. These opposite evolutionary effects between different AS patterns have been discussed as a key role of AS in the 'switch-like' regulation of gene expression .
If it is supposed that sdSNPs are related to variation within populations in the regulation of gene expression through alterations in splicing patterns, the evolutionary profiles of sdSNP may differ between sdSNPs flanking ASEs (sdSNPs at AE dinucleotides) and those flanking CSEs (sdSNPs at CE dinucleotides).
Here, we extracted the sdSNPs from all human genes, and evaluated them by comparing flanking exon properties between ASEs or CSEs. Each group was subsequently divided into three subgroups according to exon conservation status between human and mouse, namely transcript-conserved, genome-conserved, and non-conserved (See Methods for details of the criteria used). We found that sdSNPs exist in the human genome with high allele frequencies, and with no significant difference in flanking exon properties between ASEs and CSEs. Moreover, we also found that these sdSNPs are prone to be maintained at the splice dinucleotides of newly generated exons. These results suggest that sdSNPs are associated with relaxed selection pressure for newly generated exons.