Here, we have employed a functional SELEX strategy in vitro
to identify intronic and exonic sequence elements that silence nearby 5′ splice sites. Importantly, the silencers we characterized function both in vitro
and in vivo
. The data support several conclusions. First, effective silencing can be achieved without occluding or inactivating the affected site. Second, two of the six motifs (B and D) recovered from the selection may function idiosyncratically since they are not enriched in or near pseudoexons compared to real exons, yet still act as silencers in the construct used for selection. This result suggests the possibility that many splicing control elements may be relevant only in their specific pre-mRNAs and thus invisible to informatic techniques which rely on statistical analyses of motif frequencies. Third, informatic analyses indicate that four of the identified motifs (A, C, E and F) play a role in suppressing the use of both pseudo 5′ splice sites and pseudoexons (see Supplementary Material
). Fourth, the results strongly suggest that intrinsic features of RNA sequences (e.g. motif D and perhaps B, C and F) themselves (without the participation of ancillary proteins) can influence the ability of 5′ splice sites to compete with each other. Fifth, we demonstrate that silencers can alter the U1 snRNP/5′ splice site complex in a manner that renders the affected site to be at a competitive disadvantage with respect to 5′ splice sites occupied by an “unaltered” U1 snRNP/5′ splice site complex (). Sixth, and perhaps most importantly, we show that effects on non-rate limiting kinetic steps in splicing can lead to dramatic shifts in splice site choice.
In our experimental design, we focused on one construct under one set of conditions in extracts prepared from one cell type and analyzed only those motifs that gave the strongest phenotypes. It is important to note that we ignored sequences that produced substantive but less dramatic splice site switching. While we do not know how many sequence motifs would yield intermediate phenotypes, it seems likely that it is much larger than six. Furthermore, only one intronic and one exonic position were analyzed. It will be of interest to determine which sequences would emerge if the selections were carried out with the randomized sequence inserted at different positions or in different contexts. It will also be of interest to determine if sequences that could overcome the silencing effects could be selected in silencer containing constructs. Despite the inherent limitations of our current study, the results suggest that the regulation of splicing may be remarkably subtle and the “splicing code” (e.g. Black, 2003
; Fu, 2004
; Hertel, 2008
; Matlin et al., 2005
) correspondingly complex.
Intuitively, it would be expected that the most effective splicing silencers would sequester or otherwise inactivate the affected splice site. The silencers we have selected clearly do not function in this manner since the affected site remains fully functional but is not used when a competing site is present. When interpreted from a kinetic perspective, our observations rule out any scenario in which the silencers affect a slow or rate limiting step. If this were the case, presence of the silencers would cause reductions in the ability of the affected site to function proportional to the extent of activation of the competing site.
To account for the observed data, the silencers must affect a fast kinetic step to allow action of the silencer without substantive effects on the overall reaction rate. The fast reaction step has to be more than one order of magnitude faster than the slow reaction step when slowed by the silencer, and, consequently, several orders of magnitude faster than the rate limiting step when unaffected by the silencer. Introduction of the silencers at both splice sites would slow the fast reaction step at both sites and re-establish the original pattern of splice site selection, as we have observed (see and ).
We suggest that such a fast step could be the joining of 5′ and 3′ splice sites in a complex committed to splice site choice. This interpretation is consistent with the findings of Lim and Hertel (2004)
who showed that commitment to the general splicing pathway and commitment to specific splice site pairing are kinetically separable. Specifically, they demonstrated that commitment to splice site pairing became irreversible only upon formation of A complex, when U2 snRNP is locked onto the pre-mRNA. We suggest that the silencers function by altering the way in which U1 snRNP engages the 5′ splice site in such a way that it is less able to efficiently engage U2 snRNP, and thus the unaffected site obtains a competitive advantage. This interpretation is consistent with our nuclease protection data (see ) and with recent studies using site-directed hydroxyl-radical footprinting (Dönmez et al., 2007
) which demonstrated that U1 and U2 snRNPs are in close proximity in early spliceosomal complexes. These studies suggested that the two snRNPs engage in a direct spatially-fixed interaction wherein the 5′ end of U2 snRNP is located on a specific “side” of U1 snRNP. Any distortion or misorientation of U1 snRNP caused by a silencer could negatively affect its ability to establish proper contact with U2 snRNP. The notion that appropriate alignment of U1 and U2 snRNPs early in spliceosome assembly is crucial for commitment to splice site choice provides a possible explanation for the observation that some of the silencers affect “weak” 5′ splice sites more than “strong” ones. We suggest that U1 snRNP bound to “weak” sites, i.e. those with fewer base pairs, are already somewhat “misaligned” and thus more susceptible to further distortion by the silencers.
In addition, it seems likely that hnRNP A1 may exert its silencing function in a similar manner; i.e. by affecting the alignment of U1 and U2 snRNPs. There have been several distinct models proposed for the mechanism by which hnRNP A1 can silence (reviewed in Black, 2003
; Black and Graveley, 2006
; Matlin et al., 2005
). While all of these mechanisms may be valid in certain contexts, our finding that the protein binds simultaneously with U1 snRNP suggests that it may directly interfere with the ability of U1 snRNP to interact correctly with U2 snRNP either by altering the conformation of the U1 snRNP particle or perhaps by shielding important surfaces on the snRNP necessary for establishing correct alignment with U2 snRNP.
In summary, we suspect that a large fraction of examples of alternative and regulated splicing events will be dictated by kinetic parameters similar to the ones we have described. In this regard, there are several examples where the function of splicing control elements is only evident in the presence of competing splice sites (e.g. Cheah et al., 2007
; Lam et al., 2003
; Reed and Maniatis, 1986
) and additional poorly understood examples where sequence context determines the use of competing sites (e.g. Chen and Helfman, 1999
; Khelil et al., 2008
; Krawczak et al., 2007
; Manabe, et al. 2007
; Mayeda and Ohshima, 1988
; Nelson and Green, 1988
; O’Neill et al., 1998
, Ule et al., 2006
In any multi-intronic pre-mRNA, each internal 5′ splice site is in competition with multiple 5′ splice sites as is each 3′ splice site. Myriad cross exon and cross intron interactions ensure that correct sites are paired. However, it is not hard to imagine that minor perturbations of any of these interactions [e.g. by tissue specific or developmental stage specific changes in the levels of splicing factors (Karni et al., 2007
; Olson et al., 2007
; Park et al., 2004
), rates of transcription and/or chromatin structure] (reviewed in House and Lynch, 2008
; Kornblihtt, 2006
; Maniatis and Reed, 2002
) could alter a delicate kinetic balance and result in the very complex patterns of splice site choice that are observed. Such a view makes sense from an evolutionary perspective in that small advantageous changes in intronic or exonic sequence could rapidly lead to the expansion of proteomic diversity seen in higher eukaryotes.