Abstract
A large-scale analysis of human polyadenylation signals was carried out in silico. The most canonical AAUAAA hexamer and its 11 single-nucleotide variants that are most frequent in human genes were used to search for polyadenylation signals in the terminal sequences. Out of 18,277 poly(A) sites that were identified from 26,414 human genes, 82.5% of the sites were found to contain at least one of these 12 hexamers as a polyadenylation signal within 40 nucleotides upstream of the poly(A) site. The rest (17.5%) did not contain any of these hexamers, which suggests the existence of yet unknown signals. A total of 20,347 terminal sequences in close proximity to 12 polyadenylation signals were collected using modified EST clustering technique to establish a large-scale database of polyadenylation signals. To characterize the 12 hexamers, the locations of polyadenylation signals that were identified as "authentic" and the uracil contents of the downstream region of the signal were examined. Based on this analysis, the 11 variants of the canonical AAUAAA were identified as possibly forming "functional" signals as AAUAAA. Moreover, the observed frequency of 41.9% for AAUAAA was significantly lower than those of other reports, suggesting that the non-canonical variants are more important in the polyadenylation process than frequently recognized. Since the poly(A) sites processed by those non-canonical variants have not been generally annotated in major gene databases, it is important to determine whether the variant hexamers could work as polyadenylation signals that may be responsible for generating heterogeneity of mRNAs by alternative polyadenylation.