Selection between a natural and a cryptic 5′ splice site: a kinetic study of the effect of upstream exon sequences

Construction of mutants

All DNA fragments to be transcribed were inserted in the polylinker of plasmid pSP64 (Melton et al., 1984). The wild-type fragment was a Pst I–Sac I fragment of 846 bp from the proximal part of the adenovirus-2 (Ad2) early region 3 transcription unit (E3), as described previously (Domenjoud et al., 1991). It contains the E3 intron 1 (395 bp) flanked by 259 bp of exon 1 and 192 bp of exon 2.

5′ deletion mutants (labeled 5′dl) were prepared by the action of the exonuclease Bal 31. For NY mutants, a synthetic oligonucleotide harboring a Hind III recognition site at the 3′ end and a Pst I site at the 5′ end was made double-stranded using a primer hybridizing to the Hind III site and the Klenow enzyme. After digestion with Hind III and Pst I, the DNA fragment was inserted in the residual polylinker of a 5′dl50 mutant (a 5′ deletion mutant which contains 7 nucleotides of the natural exon 1), opened by the same enzymes.

The transcribed version of the newly created exon was self-complementary for NY54:

• 5′-gaauacaagcuuCGCGCGUACUGACGCGCGcugcaggucGGCUCCG-3′

and linear for NY22:

• 5′-gaauacaagcuuAAUAAUACUGACGCGCGcugcaggucGGCUCCG-3′

(lower case letters: polylinker; upper case letter: from 5′ to 3′: synthetic oligonucleotide and natural sequence; bold characters: nucleotides involved in base-pairing).

Mutants labeled pm modify the sequence of the donor site D1 or Dcr1 and were originally prepared from the wild-type transcript by site-directed mutagenesis, according to Inouye and Inouye (1987). To introduce these point mutations in 5′dl or NY mutants, we used an exchange method. Restriction fragments harboring either D1 (Sac II–Apa I) or Dcr1 (Apa I–EcoR I) were excised from 5′dl or NY clones and replaced by the homologous fragments including the mutated sites from other transcripts. The sequence of the mutants was determined by using the dideoxynucleotide sequencing procedure (Winship, 1989).

RNA synthesis and in vitro splicing

Methods were as previously described (Domenjoud et al., 1991). Briefly, the pSP64 plasmids containing the inserted E3 fragments were linearized by EcoR I and transcribed in the presence of [α-³²P]CTP. With 100,000 cpm Cerenkov of RNA precursor per assay, the specific activity of the transcripts was sufficient to detect splicing intermediates clearly.

For in vitro splicing, the conditions were as established previously, except that Mg⁺⁺ and polyvinyl alcohol concentrations were raised to 2.5 mM (instead of 2 mM) and 3.1% (instead of 2.6%), respectively. Three preparations of HeLa cell nuclear extract were used in these experiments. Splicing efficiency of the wild-type transcript was 70–75% in all three cases. Splicing products were analyzed in 5% polyacrylamide gels (acrylamide/bisacrylamide, 20/1, w/w) containing 8 M urea. The bands corresponding to the different molecules were cut out and their radioactivity counted. The radioactivity of blank gel fragments at the same level as the molecules was also counted and subtracted after normalization to the same amount of total radioactivity in the lane. In our hands, this method was more reliable than the densitometric analysis of the autoradiograms. The radioactivity of intermediates (i: IVS–exon 2 and exon 1), products (p: IVS and exon 1–exon 2), and remaining precursor (pr) could thus be determined. As exon 1 is very short in several transcripts and runs out of the gels, our determinations were derived from IVS–ex2 measurements, irrespective of the transcript. The value of i was then calculated, taking into account the number of cytosines in IVS–ex2 and exon 1. The proportion of intermediates, products, and remaining precursor at each time was (i × 100)/(i + p + pr), or (p × 100)/(i + p + pr), or (pr × 100)/(i + p + pr), respectively. All experimental values are the average of duplicates. For the kinetic studies, these proportions were determined after 0, 10, 15, 20, 25, 30, 40, 50, 60, 90, 120, 150, 180, and 240 minutes of incubation for the same transcript. The time course curves were drawn, and the time lags were estimated as the intersection of the maximum slope of the curves with the axis of abscissae. To optimize comparisons, transcripts to be compared were labeled using the same CTP batch; they were used fresh (within a week) and spliced in parallel with the same nuclear extract preparation.

Characteristics of the transcripts

As in previous work, we have studied the splicing of the first intron of the early region 3 (E3) from adenovirus-2. The characteristics of the wild-type transcript used for our in vitro studies have already been described (Domenjoud et al., 1991). Briefly, with the exception of vector polylinker sequences, it contains only natural contiguous E3 sequences and all the cis-acting elements required for an efficient splicing at the natural donor site D1 and for the silencing of a cryptic donor site Dcr1 located 74 nucleotides downstream. The hybrids that can be formed between the 5′ extremity of U1 RNA and D1 or Dcr1 are 7 or 6 bp long, respectively; this rules out the possibility that the site’s strength is the only parameter of selection of D1. Dcr1 can be induced either by suppression of D1 or by modifications of the upstream exon sequence. The same acceptor site is used with both donor sites.

To establish the kinetics of splicing independently at the two sites, D1 or Dcr1 were suppressed by mutation of their first intronic G (Fig. 1B). To test the influence of exon sequences, 5′ deletion or 5′ deletion-insertion mutants were constructed so that the upstream exon has different primary or secondary structures (Fig. 1A). The presence of a potential hairpin structure HP1 in the 120 nucleotide exonic segment, shown to be necessary for the exclusive selection of the natural site, has been demonstrated previously (Domenjoud et al., 1991). HP1 is conserved in the 5′dl41 exon that lacks the exonic upstream sequences. A further truncation of exon 1 disrupts HP1, but a small artificial hairpin, HP59, can be constructed from the residual sequence, and an enzymatic study showed that HP59 exists in the 5′dl59 transcript in solution (H. Gallinaro, unpublished data). The NY54 and NY22 exons are artificial exons (see Materials and Methods); NY54 can be folded in a small hairpin structure, HP54. The transcripts are named first by the exon designation (Fig. 1A), followed by the D1–Dcr1 combination (Fig. 1B).

Splicing at the natural donor site in the absence of the cryptic site

The pm1 series of transcripts, where the cryptic site was suppressed by a point mutation, was used to study splicing at the natural site. A typical time course of appearance of intermediates and products of transcript WT-pm1 is shown in Figure 2. The other data are summarized graphically in Figure 3A and B. Note that the scale of ordinates is expanded 10 times for intermediates relative to products.

The curves for the WT-pm1, 5′dl41-pm1, 5′dl59-pm1, NY54-pm1, and NY22-pm1 transcripts are superimposable, that is, they do not differ from each other more than do those from the same transcript in different gels or experiments; therefore, only the composite curves are presented (Fig. 3A). The intermediates are detected after a time lag of about 12 minutes (see also arrow in Fig. 2). Thereafter, a linear accumulation rising to a maximum at about 30 minutes is observed, followed by an exponential decrease. After 3–4 hours of incubation, a plateau is reached with 1–2% of the transcripts remaining in the form of intermediates. The products of the reaction become detectable 5 minutes later than the intermediates. Thereafter, they accumulate to reach a plateau representing the amount of fully spliced transcripts, which at 3–4 hours is equivalent to 79–85% of the initial transcript.

The transcript 5′dl94-pm1 behaves differently (Fig. 3B). The lag periods are longer (+2 minutes), and the peak of intermediates is lower and appears flattened, while the plateau of intermediates remains unchanged at 1–2%. The amount of products formed at 3–4 hours is lower and reaches only ∼70% of the initial transcript. It is possible that the reactions are not entirely terminated at 4 hours.

Thus, the shortening of exon 1 from 274 to 45 nucleotides does not affect the rate or efficiency of splicing; a fortiori, there is no influence of exon structure. In contrast, there is a dramatic transition between the transcripts harboring the 45 nucleotide NY22 exon and the 31 nucleotide 5′dl94 exon. Since these two exons differ only by the presence or absence of an artificial sequence that is not likely to play a specific role, we infer that a 31 nucleotide exon is not sufficient to insure an optimal rate of splicing, while a 45 nucleotide exon is sufficient. The minimum upstream exon length required for full splicing efficiency of other transcripts has been proposed to be 30–50 nucleotides (Parent et al., 1987; Adami et al., 1989; Mayeda and Oshima, 1990). It seems plausible that a minimum exon size is required for the optimal binding of splicing factors at the donor site, and that this size is between 31 and 45 nucleotides in the case of our transcript. Thus, provided that exon length is sufficient, the data demonstrate that the previously observed positive influence of exon 1 on the selection of D1 (Domenjoud et al., 1991) is not due to an enhancing effect on the kinetics or efficiency of splicing at this site.

Splicing at the cryptic donor site in the absence of the natural site

The effect of exon 1 on splicing at Dcr1 was studied with transcripts WT-pm2, 5′dl41-pm2, 5′dl59-pm2, and 5′dl94-pm2, in which D1 was suppressed by a point mutation at G1 (Fig. 1). The time course of appearance of intermediates and products was similar, and the composite curves are shown in Figure 3C. The curve of pm2 intermediates is clearly different from that of pm1. The time lag is 16 minutes instead of 12 (compare also the position of arrows in Fig. 2). No sharp peak is formed, partly because the amount of intermediates remaining after 3–4 hours is 2–3 times higher for Dcr1 than for Dl. The appearance of products is also retarded, and their final plateau is lower (55–65% instead of 79–85%). The similarity of the kinetics of splicing irrespective of exon 1 structure indicates that this exon does not favor selection of D1 by inhibiting splicing at Dcr1. It is worth noting that, in contrast to the pm1 series, there is no significant reduction of splicing efficiency in the 5′dl94-pm2 transcript relative to the other transcripts of the pm2 series. This may be related to the size of the 5′dl94 exons: 105 nucleotides for the cryptic exon (pm2) against 31 nucleotides for the natural exon (pm1).

A comparison of the D1 and Dcr1 data shows that they are much more dispersed for Dcr1 than for D1 (compare standard errors in Fig. 3A and C). As conditions were the same and the experiments often performed in parallel, the dispersion of the results is likely to be a particular feature of Dcr1. We assume that Dcr1, which is not an authentic donor site and has a different sequence context than D1, might be more sensitive to small differences of the in vitro system. It is evident, however, that splicing at Dcr1 is less rapid and efficient than splicing at D1.

Cis-competition: splicing efficiency at the natural and cryptic 5′ splice sites

The time course of appearance of intermediates and products of splicing at D1 and Dcr1 was established for the various transcripts harboring the two sites (pm0). In Figure 4, we compare the appearance of fully spliced products from the various pm0 transcripts to that from the homologous pm1 transcripts in which Dcr1 was suppressed. With the WT and 5′dl41 exons, splicing efficiency at D1 is reduced when the cryptic site is present (from 83 to 75% of the initial transcript, after 3–4 hours of incubation). A difference is also observed in the appearance of intermediates (Fig. 5); in particular, the time lag is lengthened, while the height of the peak and rate of the reactions are reduced. As these differences are small, we performed an experiment in which transcripts WT-pm1 and WT-pm0 were studied in parallel and confirmed that the decrease was not due solely to experimental variations (Fig. 2). This difference between the pm1 and pm0 transcripts was further confirmed in several other experiments in which only splicing efficiencies were determined. We conclude that the difference is significant, that splicing at D1 is impeded by the presence of Dcr1, and therefore that Dcr1 is recognized in our in vitro system.

With the WT exon, there is no detectable splicing at Dcr1, although we estimate that we would have detected its products if 0.5% of the precursor had been spliced at Dcr1. However, the 5′dl41 exon, which lacks 139 E3 nucleotides upstream of HP1, gives rise to a low amount of splicing at Dcr1 (1–2%) relative to the WT exon. Thus, these 139 nucleotides are likely to contain an element partially responsible for the silencing of Dcr1.

The differences between the pm1 and pm0 transcripts are accentuated when exon 1 is further truncated. In 5′dl59, the rates of the reactions are reduced relative to WT-pm0 and 5′dl41-pm0 (Fig. 4), and the 3–4 hour splicing efficiency is 66% as opposed to 74–76%. Simultaneously, the transcripts become more permissive for splicing at Dcr1 (5%). A more marked effect is observed with transcripts NY: splicing efficiencies at D1 and Dcr1 are 59% and 13%, respectively (Table 1). The presence or absence of the small stem-loop structure HP54 does not influence splicing of the NY transcripts. A decrease in the formation of intermediates is also observed, more marked for NY54 and NY22 than for 5′dl59 (Fig. 5). As the time course of splicing of the pm1 and pm2 transcripts is not modified with the same exon 1 changes (Fig. 3), we conclude that the 120 nucleotide segment, containing HP1 and located at the 3′ extremity of exon 1, influences splicing by interfering during cis-competition.

The 5′dl94 truncated exon is too short to allow optimal splicing efficiency in the pm1 transcript (Fig. 3B). The presence of Dcr1 in the 5′dl94-pm0 transcript promotes a further decrease in splicing at D1 (from 72 to 41%) and a marked induction of splicing at Dcr1 (25%). Thus, when splicing at D1 is impaired due to the suboptimal size of exon 1, the transcript becomes more permissive to splicing at Dcr1. The effects of suboptimal exon size and of cis-competition appear additive.

In summary, the data show that upstream exon sequences, which have no influence on splicing at D1 in the absence of Dcr1, clearly affect this splicing in the presence of the cryptic site Thus, exon 1 — in particular, the sequence harboring the potential stem-loop structure HP1 — confers resistance to competition to the natural site D1 and therefore acts as a selector for this site.

Variations in the kinetic curves and their interpretation

Comparison of the gels (Fig. 2) and kinetic curves (Figs. 3, 4, and 5) of the various transcripts reveals marked differences in the appearance and utilization of splicing intermediates, as well as in the appearance and yield of fully spliced products. This repercussion of the structure of the transcripts on the kinetics of splicing is likely to derive from its influence on one or the other of the various steps in the assembly of splicing complexes. In Figure 6, we present a simplified scheme of mammalian splicing complex assembly that should be helpful for interpreting the kinetic study.

The precursor is first coated with proteins (reaction a). Then, these early complexes are committed to splicing (Legrain et al., 1988; Michaud and Reed, 1991; reaction b), giving rise to commitment complexes (pr1 and pr2, for D1 and Dcr1, respectively). Thereafter, active type 1 spliceosomes are made from commitment complexes (reaction c) in which splicing reaction 1 proceeds. This period, rich in crucial events as far as spliceosome assembly is concerned, appears as the time lag of the curve of intermediates in our kinetic studies. The length of this time lag differs according to the transcripts, as illustrated for instance in Figure 2. There is clearly a delay in the appearance of D1 intermediates in WT-pm0 or of Dcr1 intermediates in WTpm2 relative to D1 intermediates in WT-pm1 (arrows). The time lags were estimated for the various transcripts (Table 1). The data show that the time lag of D1 intermediates is lengthened, relative to the pm1 series, by about 2–3 minutes when exon 1 is truncated (5′dl94 exon) or when Dcr1 competes with D1 (pm0 transcripts). The time lag of the Dcr1 intermediates is longer than that of the D1 intermediates by about 4 minutes (pm2 series) and is further lengthened in the presence of D1. Thus, spliceosome assembly is slowed down when exon 1 is too short, or in the conditions of cis-competition. Furthermore, spliceosomes may assemble more or less rapidly, according to the donor site. The individual reactions that are affected by the various structural changes of the precursor remain to be determined.

Once splicing reaction 1 is completed, some conformational changes are necessary to allow the onset of the second splicing reaction (reactions d and e, Fig. 6). This is suggested by the difference in the time required for the appearance of intermediates and products, which is about 5 minutes in the WT-pm1 transcripts (Table 1, lag p − i). This time is not significantly modified upon truncation of exon 1 (5′dl94-pm1) and slightly increased in the presence of the cryptic site (pm0 transcripts). For splicing at Dcr1, lag p − i is 2 minutes longer than at D1 and is further lengthened in the pm0 transcripts. Though small, these differences were reproducible and indicate that splicing can be influenced not only by defective early spliceosome assembly, but also by some defects in later reactions that can be detected by the analysis of the time course curves.

The curves of intermediates never tend to zero, and a plateau is reached in all cases. This shows that intermediates persist under a form different from spliceosomes of type 1 and 2 proper — a form we propose to designate “intermediate complexes.” The plateau of the curve indicates that equilibrium is reached between formation and utilization of the intermediates. The level of the plateau is 2–3 times higher for Dcr1 than for D1 (Table 1, i2 and i1), which suggests that utilization of the Dcr1 intermediates is slower relative to formation in the pm2 transcripts than that of D1 in the pm1 transcripts, in agreement with the lengthening of the time lag p − i. The level of the plateau does not significantly change as long as the formation of intermediates is active enough (compare i1 of the pm1 and pm0 transcripts in Table 1) and is reduced when the rate of formation is low, as expected (compare i2 of the pm2 and pm0 transcripts). The persistance of intermediate complexes when no more products are made may be due to conformational changes of these complexes or to an inhibition of reaction e, whatever the cause.

As not all the precursor is engaged in splicing, we postulate that “inactive complexes” (pr0) can be assembled in addition to the commitment complexes. We observe a correlation between the increase of the proportion of pr0 and that of the time lag of intermediates (Table 1). Therefore, we propose that the assembly of pr0 follows kinetic rules of its own (reaction b0 in Fig. 6), and that competition for the precursor occurs during early assembly. It cannot be ruled out that the assembly of pr0 is partially reversible, which may explain the low increase of the plateau of the curves of products in some transcripts.

Splicing efficiency can be estimated as the proportion of precursor engaged in splicing, that is, the sum of intermediates and products formed at the end of the reaction. These proportions, after 3–4 hours of incubation, are indicated in Table 1 (pr1 for D1 and pr2 for Dcr1). They are representative of the proportion of precursor that was committed to splicing at D1 and Dcr1 during early spliceosome assembly (reaction b). The exact values may be somewhat different, as when the reactions are not entirely terminated at 3–4 hours; however, such values are indicative of earlier events. The maximal slope of the curve of products (which is expressed as the percentage of fully spliced transcripts made in 10 minutes) can be considered an index of the overall rate of the splicing reaction and varies, as do pr1 and pr2. As already stated, the rate and efficiency of splicing at D1 are dependent on the structure of exon 1 in the pm0 transcripts and independent of it in the pm1 transcripts.

Thus, the analysis of the time course curves allows the specification of the period at which a given structural feature may modify the use of a 5′ splice site and may thereby orient further investigation. It also allows a relatively easy comparison of transcripts harboring several splice sites and various modifications.

The function of the upstream exon

One of the major results of the kinetic study was to show that two 5′ splice sites, whose capacity to hybridize to U1 RNA is similar, may have very different kinetics of splicing when studied in the absence of each other. The difference is due both to a slower early spliceosome assembly and to a slower transition from spliceosomes carrying out splicing reaction 1 to those carrying out reaction 2 (Fig. 3 and Table 1). These characteristics of the time course curves lead us to predict that, when the two sites are present in a same transcript (pm0), D1 will be used much more rapidly than Dcr1. The 4 minute difference in the time lag of appearance of intermediates allows the assembly of a relatively large number of complexes committed to splicing at D1 (pr1) and a concomitant increase in the number of inactive complexes (pr0) in which the precursor is sequestered. Thus, the proportion of precursor available for splicing at Dcr1 will be lowered and, in turn, will lead to the formation of a relatively small number of complexes committed to splicing at Dcr1 (pr2). As the transition between spliceosomes carrying out reactions 1 and 2 is also longer, the final splicing efficiency at Dcr1 is expected to be low under conditions of cis-competition between the sites. Comparison of splicing at D1 in the pm0 and pm1transcripts and at Dcr1 in the pm0 and pm2 transcripts indeed indicates that Dcr1 is more affected by competition than D1 (Table 1).

To determine the effect of exon sequences on selection, the curves of appearance of intermediates and products of splicing in the pm0 transcripts with the various exons 1 can be compared (Figs. 4 and 5, and Table 1). We first observe that, compared to the pm1 or pm2 transcripts, as appropriate, the time lags are lengthened, irrespective of the exon or the site. This lengthening is accompanied by an increase in the proportion of the inactive complexes pr0, relative to that found in the pm1 transcripts, but smaller than that found in the pm2 transcripts. These observations are in agreement with the proposal that the assembly of pr0 complexes competes with that of pr1 and pr2, and that splicing at Dcr1 in the pm0 transcripts starts after a large amount of precursor has already been committed to splicing at D1 (or to inactivity). The similarity between the time lag changes, irrespective of the exon, also implies that their lengthening is not the consequence of modification of exon structure, but that it is related to the presence of both sites in a same transcript.

In contrast to the time lags, the overall rate of the reactions (i.e., the maximal slope of the curve of products, used as an index) is affected by exon 1 in the pm0 transcripts, as is the splicing efficiency at each site. The rate of splicing at D1 is reduced in the pm0 relative to the pm1 transcripts and gradually decreases when exon 1 becomes shorter. Conversely, splicing at Dcr1, which is not detectable in the WT-pm0 transcript, is induced when exon 1 is shortened. However, the effect of exon size on splicing rate and distribution of the precursor between the two sites is not linear. For instance, the 139 nucleotide shortening of the WT to the 5′dl41 exon has only a small effect, while further truncations (of 61, then 38 nucleotides) dramatically affect the distribution of the precursor. Thus, D1 appears to be “protected” against cis-competition in the transcripts WT-pm0 and 5′dl41-pm0, and this protection correlates with the presence of the 120 nucleotide sequence that can be folded into the stem-loop structure HP1. Transcript 5′dl59-pm0, which is permissive for Dcr1 (pr2: ∼5%), harbors HP59, a hairpin structure shorter than HP1, which might also have some protective effect. In contrast, HP54 has no influence. The two NY transcripts, whose exons are made of artificial sequences, are more permissive for Dcr1 (pr2: ∼13%) than 5′dl59-pm0, and, conversely, D1 is less protected. It is conceivable that there is no protection of D1 by exon 1 in these transcripts, and that their splicing pattern is the result of competition alone. If so, this would indicate that about 4.5 times less precursor is used for splicing at Dcr1 than for splicing at D1 in the conditions of cis-competition.

The data underline the importance of the structure of exon 1 in the selection of the natural site D1. As it does not exert its influence by modifying the kinetics of splicing directly, we conclude that the observed kinetic changes are due to its selector effect during cis-competition between the sites, and that the secondary structure of the exon is involved in the selection.