Nucleotides in both donor and acceptor splice sites are responsible for choice in NAGNAG tandem splice sites

Abstract

Among alternative splicing events in the human transcriptome, tandem NAGNAG acceptor splice sites represent an appreciable proportion. Both proximal and distal NAG can be used to produce two splicing isoforms differing by three nucleotides. In some cases, the upstream exon can be alternatively spliced as well, which further increases the number of possible transcripts. In this study, we showed that NAG choice in tandem splice site depends considerably not only on the concerned acceptor, but also on the upstream donor splice site sequence. Using an extensive set of experiments with systematically modified two-exonic minigene systems of AFAP1L2 or CSTD gene, we recognized the third and fifth intronic upstream donor splice site position and the tandem acceptor splice site region spanning from −10 to +2, including NAGNAG itself, as the main drivers. In addition, competition between different branch points and their composition were also shown to play a significant role in NAG choice. All these nucleotide effects appeared almost additive, which explained the high variability in proximal versus distal NAG usage.

Supplementary Information

The online version contains supplementary material available at 10.1007/s00018-021-03943-2.

Introduction

Pre-mRNA splicing is a key step in gene expression in which non-coding regions (introns) are removed and coding regions (exons) are ligated to form an mRNA molecule. The exon is defined by the intron boundaries consisting of a conserved donor splice site (5′ss) and acceptor splice site (3′ss). In many genes, some splice sites can be recognized ambiguously, which leads to alternative splicing and increases protein diversity. Tandem acceptor splice sites cover 20% of alternative splicing events [1–7]. 3′ss located 3 bp from each other, referred to as NAGNAG splice sites (proximal NAG—upstream; distal NAG—downstream) [3, 8–10], occur in 30% of human genes and 5% of them are probably spliced in both of their NAGs [10, 11]. It results in two splicing isoforms differing in local amino acid composition being produced, which may have a functional relevance by changing local hydrophobicity and charge, distances between relevant sites in proteins or recognition sequences for post-translational modifications [12].

NAG choice (selecting NAG in tandem NAGNAG 3′ss) takes place after the first transesterification reaction by a “scanning mechanism” [13]. During the C-to-C* spliceosome complex transition, RNA is scanned to find AG sequence with a better preference to YAG [14]⁠, which is subsequently used in the second transesterification reaction as a recognition sequence marking the intron–exon junction. However, another downstream located YAG sequence may compete with the first one [13, 15]. Splice site competition is ruled mainly by the splice site region sequence itself [5, 16, 17], probably being mediated by constitutive splicing factors SLU7, SPF45 (RBM17) and others [13, 18–20]. The distance between the splice site and branchpoint (BP) sequence is an additional factor involved in NAG choice [1, 15, 21] and the 3′ss sequence is suspected to be another one. To our knowledge, the position associated with NAGNAG regulation experimentally confirmed so far is a nucleotide immediately preceding the NAGNAG motif [1]. The other positions’ roles remain unclear or not systematically verified. In Saccharomyces cerevisiae, the potential effect of donor position was postulated by Deirdre et al. [22], who observed higher usage of distal NAG, if G or C was located at the third intronic position. However, the nature of this effect remained unsolved.

All these findings indicate that our knowledge about factors driving NAG choice is incomplete. To test the hypothesis that not only NAGNAG itself stands behind different NAG usage, we systematically investigated the splice site’s regions possibly involved in NAG choice. We analyzed human transcriptomic data to identify the general drivers of NAG choice. Additionally, two different experimental minigene systems were chosen to confirm transcriptomic data—AFAP1L2 gene’s two-exonic system, where exon 14 can be naturally skipped, while exon 15 contains AAGCAG 3′ss motif—and the CTSD gene’s two-exonic system, with an artificially implemented NAGNAG 3′ss motif. This system enabled us to describe 3′ss and upstream 5′ss region features, which we hypothesized to regulate proximal/distal NAG usage.

Results

Transcriptomic analysis of NAG choice

To investigate the dependence of NAG choice on particular nucleotides in 3′ss, or upstream 5′ss, RNA-Seq data generated from 16 human tissues collected by Bradley et al. [1] were analyzed. To filter out the main source of variability, 3′ss were categorized according to the NAGNAG sequence and subsequently subgrouped according to the nucleotides in specific 3′ss and 5′ss positions. This approach reduced the item numbers in some subgroups to very low values (Supplementary Fig. S1) and sufficiently represented NAGNAGs were especially AAGCAG, CAGCAG and TAGCAG. Shifts in NAG choice were most noticeable in A−7, where distal NAG usage (dNu) median was significantly raised in A−7(C) or A−7(T) variants almost for all NAGNAGs (Supplementary Table T1). However, dNu diversified in dependence on nucleotide composition also in other positions. C or G in A−8 and G in A−9 and A−10 of 3′ss supported dNu. Similar shifts were observed for nucleotides C and G located in D+3 and nucleotide G in D+4 5′ss positions (Fig. 1, Supplementary Fig. S1).

Fig. 1.

NAG choice is dependent on 5′ss and 3′ss nucleotide composition. a Selection of several genes undergoing alternative splicing, where exon being endowed with NAGNAG 3′ss comes after another alternatively spliced/skipped exon. dNu is counted according to RNA-seq data. Significantly distinct dNu (p < 0.05) for transcript variant differing from first (Da) is marked with a star (*). b Splicing scheme showing difference between alternatively spliced isoforms used in A. c Nucleotide scheme applied for in silico analysis used in d and e. dNu was counted based on RNA-seq data and split according to NAGNAG and particular 3′ss (d) or 5′ss (e) nucleotide. Only AAGCAG and CAGCAG were selected to be presented here. dNu is presented as box plot with variable box width, where width represents number of analyzed genes (in log scale). Significantly different dNu values (Mann–Whitney p value < 0.01) are marked with a red arrow (if dNu differs in relation to all other variants) or with a red dot (if dNu differs in relation to all other variants without dot)

If the 5′ss sequence influences NAG choice, transcriptomic data can offer specific data about this phenomenon in the case of alternative splicing events, when different 5′ss are used (exon skipping, alternative 5′ss). Overall, 78 different genes had a sufficient number of reads in the RNAseq data and dNu in the interval from 9 to 90% in at least one transcript type (Supplementary Table T2). When different 5′ss were combined with a downstream NAGNAG exon, dNu significantly varied in some transcripts (Fisher test, p < 0.01, Fig. 1). This variation was not linked to the type of upstream exon alternative splicing (Supplementary Table T2), but only to the 5′ss nucleotide sequence bordering an upstream spliced intron, which seemed to be the key factor driving NAG choice.

AFAP1L2 and CTSD as model two-exonic systems for studying NAG choice

To study the modifying factors of NAG choice, the selected experimental model should meet the following criteria: (I) the upstream exon undergoing incomplete recognition, (II) the downstream exon producing both NAGNAG isoforms in an appropriate ratio different for full-length (FL) transcript and skipping of upstream exon (Δe2), (III) criteria I and II reproducible in the minigene system, where the exons surrounding the model exons are different from the endogenous ones and (IV) the inner intron relatively short, to enable easy manipulation with minigene plasmid (Fig. 2). This arrangement could generally result in four primary alternative splicing isoforms: FL using either proximal (FL-P) or distal splice site in exon 3 (FL-D), or transcripts with skipped upstream exon (Δe2-P, Δe2-D; Fig. 2a). Finally, two independent model systems were chosen. The first one (AFAP1L2—exons 14 and 15) represented fully natural system with AAGCAG 3′ss, and CTSD artificial system with introduced TAGTAG 3′ss [23], which should serve as an evolutionary non-adjusted splicing model. The inner introns were very short in both systems (82 nt and 92 nt in the AFAP1L2 and CTSD system, respectively). It enabled us to study the influence of intron length on NAG choice as well. However, a prolonged AFAP1L2 inner intron (170 nt) did not significantly alter dNu (data not shown), which indicated the weak effect of the intron length.

Fig. 2.

Experimental design and NAG choice in artificial minigenes. a Transcript nomenclature used in our study. Generally, a transcript including all exons is marked with FL (full length) and transcript skipping some of its exons with Δ. IR corresponds to middle intron retention. P is applied to proximal and D to distal NAG in exon 3. b Minigene construct schemes. c Distal NAG usage (dNu) differs depending on exon 2 inclusion in both studied minigenes (*p value < 0.05). d Exemplary electrophoretogram

Increased dNu in AFAP1L2 Δe2 transcript compared to FL was independently found in an RNA-seq dataset confirmed from a thyroid gland (PRJNA30709)—predominant AFAP1L2 expressing tissue. dNu reached 22% and 57% here for FL and Δe2, respectively (Supplementary Table T2). A similar trend was reproduced in our minigene (containing AFAP1L2 exon 14, intron 14 and exon 15), where dNu in Δe2 reached only slightly lower value than the endogenous one (~ 35%).

Similarly to AFAP1L2, NAGs usage significantly differed for FL and Δe2 transcript in CTSD minigene construct, where dNu was significantly higher among FLs (62%) than Δe2 isoforms (26%; Fig. 2). Higher dNu accompanied different transcripts (FL vs. Δe2) in both model systems, which indicates that exon skipping itself is not a cause of dNu shifts, but 5′ss nucleotide composition could play a dominant role as transcriptomic data indicates. Both minigenes contained the same exon 1, but model-specific exon 2 contained different 5′ss. CTSD had G at D+3 position in exon 2, which should support dNu according to the transcriptomic data.

Splice site strength in AFAP1L2 exons drives exon inclusion and a balance among splicing isoforms

The impact of NAGNAG type was described earlier [1, 5, 8] and this dependence was confirmed also in our minigene systems (Supplementary Fig. S2). To elucidate a potential role of other factors, several series of minigenes were prepared and checked for NAG choice. To verify the influence of all four ss in the AFAP1L2 two-exonic system on balance among the splicing isoforms and dNu, variants with modified strength in all these splice sites (A2, D2, A3, D3) were compared (Fig. 3). The overall balance among splicing isoforms was relatively complicated and ambiguous, but some trends appeared obvious. Intron retention was driven by the inner splice sites, D2 and A3, and skipping of both exons (Δe2e3) especially by D3. Generally, significant dNu changes were influenced only by D2 and A3 variations.

Fig. 3.

dNu is affected only by intron surrounding splice sites. a Scheme of AFAP1L2 minigene and mutations included in analysis. Numbers affiliated to particular splice sites behind underscore sign (_) correspond to their MaxEnt values. Changes in particular 3′ss (A2) or 5′ss (D2 or D3) were combined with three different NAGNAGs (AAGCAG, CAGCAG, TAGTAG). b Frequency of main splicing isoforms. c dNu for FL and Δe2, independently. Error bars represent standard deviations. Significantly distinct dNu from wt (p < 0.05) is marked with a star (*). While exon recognition is affected by nearly all splice site manipulations, dNu was significantly different only in D2_5.5 variant and between different NAGNAGs

To generalize the rules, how the splice site strength helps to recognize splicing modules and subsequent proportions of particular transcript isoforms, a total of 149 variants with a different combination of 5′ss and 3′ss (Supplementary Table T3) were used for regression analysis. Dependence of FL, Δe2, Δe3, Δe2e3 and IR frequency on splice site strength was tested in a linear model as the simplest approximation (Table 1). Based on our previous observation, the MaxEnt score calculated for the proximal 3′ss in the second exon was used. In agreement with an intuitive exon recognition understanding, including both exons was positively supported by all splice sites. IR required stronger outer sites (A2 and D3) and weaker inner sites. While plain exon 3 inclusion (Δe2) was positively driven by both A3 and D3 splice sites and negatively by A2 and D2, analogous requirements were not fulfilled for exon 2 inclusion alone (Δe3). Only decreasing A3 and D3 strength resulted in enhanced exon 2 inclusion (Δe3) frequency. A similar dependence was also observed in Δe2e3. Generally, the AFAP1L2 two-exonic system was more sensitive to any change in exon 3-surrounding splice site strength, while manipulating exon 2 splice site strength influenced only some splicing isoforms.

Table 1.

Dependence of AFAP1L2 splicing isoform frequency on splice site MaxEnt values in the linear regression model

	Splice site
	A2	D2	A3(NAGNAG)	D3
Splicing isoform
FL	0.03 ± 0.01 (p = 5.80e−5)	0.07 ± 8.01e−3 (p = 2.68e−16)	0.02 ± 3.01e−3 (p = 3.66e−10)	6.02e−3 ± 7.98e−3 (p = 0.45)
Δe2	− 0.07 ± 9.54e−3 (p = 2.39e−11)	− 0.04 ± 7.59e−3 (p = 2.16e−6)	0.02 ± 2.85e−3 (p = 8.37e−17)	0.06 ± 7.55e−3 (p = 8.93e−13)
Δe3	2.48e−3 ± 3.47e−3 (p = 0.48)	− 9.75e−5 ± 2.76e−3 (p = 0.97)	− 3.05e−3 ± 1.04e−3 (p = 3.49e−3)	− 0.04 ± 2.75e−3 (p = 3.31e−38)
Δe2Δe3	− 7.76e−5 ± 2.15e−3 (p = 0.97)	1.95e−5 ± 1.71e−3 (p = 0.99)	− 3.17e−3 ± 6.43e−4 (p = 1.27e−6)	− 0.03 ± 1.70e−3 (p = 3.31e−63)
IR	0.03 ± 7.89e−3 (p = 3.63e−4)	− 0.03 ± 6.28e−3 (p = 4.69e−7)	− 0.07 ± 2.36e−3 (p = 1.32e−44)	0.13 ± 6.25e−3 (p = 0.03)

Table depicts regression coefficients of particular MaxEnt values in final linear regression model equations, where ± values correspond to standard errors and (p =) to p values

Nucleotides in 3′ss regulate NAG choice

To dissect NAG choice modulators, we firstly investigated the sequence neighboring the NAGNAG 3′ss. Previously, only the A−7 position was studied by minigene assay [1]. Based on our in silico analysis, we extended the explored region to positions from A−11 to A+3 to embrace all potentially relevant nucleotides. All three substitutions generated in each position were combined with AAGCAG and CAGCAG 3′ss motifs (Fig. 4, Supplementary Fig. S3, Supplementary Table T4). The trends in both groups were very similar, indicating additive effect of these positions.

Fig. 4.

dNu is affected by 3′ss sequence. a Scheme of AFAP1L2 minigene and positions designated for mutations. b Frequency of main splicing isoforms is divided into separated groups differing only in one position. Wt sequence is depicted in all groups and is highlighted in red. c dNu for FL and Δe2, independently. Error bars represent standard deviations. Significantly distinct dNu from all other variants in position (p < 0.05) is marked with red arrow indicating direction of change (↑↓). In A+2 position, variant A is excluded from the comparison, because of multitandem splice site formation. In position A−7, C and T variants are significantly different from A and G variants (indicated by red dot), but not from each other

All variations in 3′ss impacted on dNu in both FL and Δe2, reflecting shared 3′ss sequences. G nucleotides in polypyrimidine tract (PPT) positions A−10, A−9 and A−8 caused substantially increased dNu as well as C or T in A−7, as already shown in Bradley et al. [1]. All these results corresponded to transcriptomic data indicating that these nucleotides could support distal NAG choice (Fig. 1). In our minigene system, G in A−11 had negligible effect on dNu, while in transcriptomic data, this tendency was apparent only for highly abundant CAGCAG 3′ss. There is a possibility that AFAP1L2 itself or generally AAGCAG 3′ss represents a specific system that is more tolerant to nucleotide exchanges in more distant 3′ss positions. It could be connected to PPT composition. PPT mutations in positions from A−8 to A−10 had an important impact on exon recognition as apparent from IR frequency, and this effect decreased in A−11. It is not clear if this behavior was interconnected, or not, but at least in A−7, a position, where nucleotide exchange does not influence exon recognition too much, particular nucleotides influenced NAG choice substantially, which supports a hypothesis for independence of exon recognition and NAG choice.

Contrary to what was expected, the G nucleotide in the first exonic position following distal NAG (A+1) did not enhance dNu, while A or C had a stronger dNu supporting effect. This is in direct contrast to the nucleotide bias in this 3′ss position, where G and A are predominantly preferred [14]. There is a question, if these results are AFAP1L2 specific, or not. Transcriptomic data revealed only significantly higher dNu for A compared to G in A+1 for AAGCAG, while both G and A in A+1 supported distal NAG for CAGCAG motif (Fig. 1). As the absolute number of available AAGCAG 3′ss is not too high, it cannot be decided clearly, if the remaining differences in dNu are supported by transcriptomic data, or not, and if these differences are caused by epistatic interaction of AAGCAG and A+1 motif as can be deduced from insufficient transcriptomic data, or if AFAP1L2 represents other specific interaction causing shifted dNu balance in dependence on A+1. Similarly doubtful remains the position A+3.

A preference for T nucleotide among dNu supporting sequences prevailed in the A+2 position in both minigene and transcriptomic data. This dNu increase was also detected in the variant with A in A+2, but the interpretation of this variant is not so straightforward, because an A nucleotide in A+2 position formed a multitandem splice site containing 4 AGs in a row (aagcagCAGCAG). The observed dNu is thus more likely the effect of multitandem 3′ss composition than the direct effect of A in A+2. From this reason, this variant was not considered, although a minigene analysis confirmed weak support of distal NAG.

Nucleotides in 5′ss influence NAG choice

RNAseq data and minigene analysis confirmed that inclusion of different preceding exons containing different 5′ss sequences can impact on NAG choice. To test the effect of 5′ss composition on dNu, we generated minigene constructs with all possible substitutions in each position from D−2 to D+6 for both AAGCAG and CAGCAG variant (Fig. 5, Supplementary Fig. S4, Supplementary Table T4). However, these substitutions could not be done simply by replacing wild-type AFAP1L2 nucleotides, because this would substantially decrease 5′ss strength and subsequent exon 2 inclusion (not shown). To avoid these unintended attributes, 5′ss needed to be modified in D+6 or both D+5 and D+6 positions independently for the particular positions and compared within this group.

Fig. 5.

dNu is affected by the 5′ss sequence. a Scheme of AFAP1L2 minigene and positions designated for mutations. b Frequency of main splicing isoforms is divided into separated groups differing in second exon’s 5′ss only in one position. Wt is highlighted in red for each group separately. c dNu for FL and Δe2, independently. d, e Frequency and dNu for variants carrying variations in first minigene exon. While dNu in FL transcripts varies, when 5′ss of exon 2 is mutated, dNu in Δe2 responds to exon 1 mutations. Error bars represent standard deviations. Red arrow indicates direction of significantly distinct dNu change (↑↓) compared to all other variants in position (p < 0.05)

Contrary to variants in 3′ss, dNu differed only for FL splicing isoforms and not for Δe2. It clearly demonstrated the modifying factors to be located directly in exon 2, where they can be effective only in the case of both exon inclusion. Besides that, and in agreement with transcriptomic data, nucleotides in position D+3 had the most significant influence on NAG choice. The C and G nucleotides in this position resulted in a substantial shift toward dNu in FL splicing isoform, while dNu in Δe2 remained constant. Surprisingly, D−2 and D+5 positions also seemed to affect NAG choice in FL splicing isoform. A nucleotide in exonic D−2 position and G in intronic D+5 increased dNu, while T in D−2 or C in D+5 had the opposite effect. Comparison with transcriptomic data showed that the effect of T in D−2 could be specific for AAGCAG variants. However, G in D+5 is highly preferred, and the abundance of exons containing different nucleotides here is relatively low and dNu derived from transcriptomic data could be biased. In position D+4, only the C nucleotide shifted dNu mildly and similarly to other 5′ss positions; this effect can be specific for some NAGNAG variants, as TAGCAG group revealed the same dependence in the whole transcriptome data, while CAGCAG not.

To investigate whether 5′ss influence on NAG choice is driven directly by its nucleotides and not by other factors, such as the distance to 3′ss, 5′ss of the pET exon 1 was modified in both the AAGCAG and CAGCAG variant. Similar to exon 2, the 5′ss was modified to keep sufficient splice site strength and contain A (P1) or G nucleotide in D+3 (P2) (Fig. 5d, e). Analogously, the dNu increased significantly in Δe2 transcript in P2, while remaining unchanged in FL. Thus, dNu responded to exon 1 mutations only if exon 2 was skipped, further emphasizing the importance of upstream exon’s 5′ss for NAG choice.

Besides, additive 5′ss and 3′ss effects were tested (Fig. 6). D+3 and A−7 variants were combined. While D+3A/A−7T (ADD1) and D+3G/A−7T (ADD2) variants showed a significant dNu increase in both Δe2 and FL transcripts compared to D+3A and D+3G, respectively, the difference between ADD1 and ADD2 resembled the difference between D+3A and D+3G in FL only. These results clearly showed that 5′ss and 3′ss effects could be independently combined.

Fig. 6.

3′ss and 5′ss effects on dNu are independently combinable. a Scheme of AFAP1L2 minigene and variants used to evaluate additive effect of 3′ss and 5′ss on dNu. b Frequency of main splicing isoforms. c dNu for FL and Δe2, independently. Error bars represent standard deviations. Significantly distinct dNu of ADD variants from corresponding D+3 variant (p < 0.05) is marked with a star (*)

Nucleotides in branch point and its distance to 3′ss influence NAG choice

One of the previously described factors influencing 3′ss choice is the BP distance from 3′ss [1, 15, 21]. We searched for the BP sequence up to 50 nucleotides upstream from 3′ss using Human Splicing Finder 3.1 (HSF) [24]. There are four As in this region that could serve to form a lariat structure during splicing. According to HSF, only those located in positions −23, −27 and −36 could be used. We used A>T mutations in the predicted BP to validate their use and influence on 3′ss choice (Fig. 7). BP in −27 was predicted to be the strongest. BP in −32 seemed to be too weak to be used and is in good agreement with the observed expression pattern in the variant BP-32.

Fig. 7.

BP mutations result in the use of different BP and subsequent shifts of dNu. a Scheme of AFAP1L2 minigene and used branch point variants. Strength of particular BP is expressed as HSF score. b, c Frequency of main splicing isoforms and dNu determined for single, double and triple BP mutants in variants of D+3A and D+3G. d, e Frequency of main splicing isoforms and dNu determined for variants modifying main BP-27. Error bars represent standard deviations. Significantly distinct dNu from BP variant from wt (p < 0.05) is marked with a star (*)

From all single-nucleotide BP mutations, only the one in position −27 caused a significant increase in IR, indicating it to be predominant BP and that other sequences cannot completely substitute this BP. Additionally, it led to increased dNu. As double-mutant BP-27-23 showed enhanced IR and almost restored wt dNu, BP-23 is probably the main alternative BP preferring distal NAG.

It seems that the main BP is located in −27, and those in other positions (especially −23) are used only if the main BP recognition fails or can be substituted much more easily and are not required for 3′ss recognition. Anyway, it cannot be excluded that a small proportion of transcripts used other BP.

Subsequently, the possibility that BP choice can be driven by 5′ss sequence was tested. Surprisingly, BP-36 variant showed increased dNu only if combined with D+3A in FL, a donor sequence normally more supporting proximal 3′ss, while the combination with D+3G did not result in this shift. If BP-36 is used, its disruption should boost the main BP-27 and dNu should reach wt levels. However, this explanation can be rejected because dNu fell again in double-mutant BP-23-36 preserving BP in −27 intact. In the triple mutant, disrupting all three possible As serving as BP did not block FL production and the IR frequency did not differ from double-mutant BP-23-27 too much. It indicates that very unusual or weak sites can be used as BP and balance among them is important for dNu.

Finally, nucleotides around BP in position −27 was changed to find out if BP sequence composition has a direct effect on dNu (Fig. 7d, e). Mutation BP-27_76 displays decreased BP strength according to the HSF and dNu level gets closer to a variant with disrupted BP (BP-27_0). However, BP recognition was not disrupted by this mutation as evident from the IR level, and only the dNu was changed. It indicates that not only a BP distance to the 3′ss, but that the BP nucleotide composition also plays an important role in 3′ss choice.

Discussion

Initially, splicing of tandem acceptor or donor splice sites used to be considered a simple noise tolerated by cells [3, 25]. However, the functional diversification of proteins encoded by both splicing isoforms was confirmed later [6]. This finding was also supported by these motifs’ high conservancy among vertebrates, especially for acceptor splice site tandems and, surprisingly, this conservancy also holds true for flanking intronic sequences [1, 26]. This indicates that specific regulatory mechanisms assure the stability of NAG choice during evolution and the correct balance between both transcript isoforms. Although many studies focused on deciphering the NAGNAG splicing regulation, the exact NAG choice determinants and its corresponding molecular mechanism remain unclear. In this work, we have experimentally shown that these determinants are located over both acceptor and donor splice sites (Fig. 8) and are responsible for dNu differences in transcripts containing different upstream 5′ss.

Fig. 8.

Schematic representation of nucleotides supporting usage of proximal/distal 3′ss. Nucleotides in black correspond to conservative dinucleotides, while colored letters account for nucleotides supporting distal (upper panel) or proximal NAG (lower panel). Only nucleotides supported by both minigene and transcriptomic data are depicted. In the position A+1 (*), G nucleotide supports dNu according to transcriptomic data, but this dependence was not proved by AFAP1L2 minigene

Specific sequences in/around tandem 3′ss can influence not only dNu, but also exon recognition itself, as reported for example in patients carrying a 2623G>T mutation in the CFTR gene [7]. In that case, CAGGAG to CAGTAG conversion caused significant exon skipping accompanied by a 3′ss strength decrease. In our minigenes, some NAGNAGs also showed disrupted 3′ss recognition as we could deduce from enhanced IR and Δe2e3 transcript levels (Supplementary Fig. S2). In general, it seems that 3′ss recognition is driven by proximal 3′ss strength, but some details do not fit this paradigm completely, e.g., AAGAAG has a MaxEnt score comparable with AAGYAG in the AFAP1L2, but the IR level was significantly higher in the case of AAGAAG. In previous years, multiple evidence on the involvement of splicing regulators binding across the intron–exon border was gathered [27–29], which could explain the observed discrepancies. Indeed, some NAGNAG can theoretically bind splicing repressors such as hnRNP A1, A2/B1 or H1 [30], or splicing activators like SRSF1 [31], or SC35 [32]. However, their exact influence on 3′ss recognition still remains to be elucidated, also considering their competition with U2AF35—a protein naturally involved in intron–exon border recognition [29].

The N nucleotide preceding both proximal and distal AG has a predominant role in NAG choice. In most exons, only one NAG is used which can be also applied to tandem 3′ss that have the potential to support both of them (e.g., CAGCAG). On the other hand, sequences strongly preferring only one of them (e.g., HAGGAG) were shown to enable the choice of non-optimal NAG in specific cases. It indicates that NAGNAG preference must be modulated by other factors. In both our studied genes, we confirmed that the RAGYAG sequence strongly supported distal 3′ss use, while the HAGGAG sequence had the opposite effect, as already described in previous studies [1, 21]. We thus also confirmed that N nucleotides are important determinants of NAG choice, which also applied for NAGTAG 3′ss—a sequence resulting in an incorporated amber stop codon if both RNA are spliced in proximal 3′ss and the TAG sequence is in the reading frame. In that case, RNA could be degraded by nonsense-mediated decay, which could cause a bias in the observed dNu in trascriptomic data. However, the minigen was designed to keep the TAG sequence out of the reading frame and, thus, any similarity between the transcriptomic and minigene data rather indicated minor dNu bias caused by NMD. Additionally, NAGTAG 3′ss is a very rare tandem 3′ss [23], which could be a result of selective pressure eliminating premature stop codons.

NAGNAG splicing variability observed in both minigenes implies that features outside the NAGNAG sequence must also be involved in determining a particular isoform formation. For example, a specific role of the canonically unimportant A−7 position (four nucleotides upstream of the proximal 3′ss) in NAG choice has previously been described and supported by transcriptome analysis [1]. Reanalyzing the RNAseq data from 16 tissues [1] together with our systematic AFAP1L2 3′ss screening showed that not only the A−7 position, but the whole region extending from A−10 to A+2 helps to regulate NAG choice. Indeed, Gs from A−10 to A−8 support dNu even more strongly than C or T in A−7. Cis-regulatory elements’ role was suggested earlier [8], as was the splicing factors’ role in interfering with the scanning process by searching for AG and stabilizing the hairpin loops on the pre-mRNA’s branchpoint—splice site region [33]. However, our analysis ascertained that no specific nucleotide motif was responsible for this process, but an overall RNA composition of this area drove it.

The importance of exonic nucleotides was supported by mutations in positions A+1 and A+2. Contrary to the non-tandem 3′ss, nucleotides in an exonic position directly neighboring one with distal NAG are not too conserved [9], which might indicate that they are not crucial for exon recognition but can still influence 3′ss choice significantly. In fact, the nucleotide changes in these exonic positions caused relatively strong dNu shifts in agreement with previously described dependences [21]. Notably, G nucleotide in A+1 position weakened dNu, which was in direct contrast to the consensus 3′ss, where G in A+1 is preferred [14]. While purines in A+1 enhance 3′ss strength, which is typically accompanied by better exon recognition and subsequent inclusion, the same preference is probably not required to detect the AG/N boundary before the C* complex formation. U5 snRNP, which determines the 3′ exon positioning prior to the second transesterification reaction through its components U5 snRNA and Prp8 [34], might be responsible for this preference. However, current knowledge is different. Although U5 snRNA contributes to 3′ exon binding using loop 1, partial mutations showed that RNA–RNA contacts are not the main binding specificity modulators [35, 36].

How exactly the AG is selected is not known. A scanning mechanism searching for the first AG during the C-to-C* complex transition was suggested [19, 37]. In the scanning mechanism, AG selection depends on both BP–AG and AG–AG distance [19]. It seems evident that nucleotide composition of the scanned sequence can also modify the first AG recognition’s efficiency. Proteins known to be involved in AG selections are Slu7 [38], Prp18 [39, 40], RBM17 [20] or Prp22 [41]. Their function consists of relocating Prp8 from the BP to the 3′ss during C* complex formation [42], but their direct impact on NAG choice is disputable. While Slu7 function was shown to be required to recognize distant AGs (relative to BP) and not for close ones by Brys and Schwer [43], Slu7 knockdown did not affect NAG choice as proved in Tsai et al. [15]. Similarly, we did not detect any changes in dNu after Slu7 silencing (not shown). If the nucleotides preceding AG have the potential to regulate its recognition, direct interactions between spliceosomal components and intronic pre-mRNA sequences should be applied here. These interactions were described for both Slu7 and Prp22, which can interact with the intron 3′ tail [42], but sequence specificity has not been confirmed yet [38]. Prp8 displays specificity for YAG preceding the 3′ss [44], which could be responsible for higher dNu in RAGYAGs and lower dNu in YAGRAG. In previous works [1, 8], the influence of 3′ss strength (represented by the MaxEnt score) on NAG choice was hypothesized. Indeed, in CTSD and AFAP1L2 minigenes, dNu correlated with the proximal position MaxEnt score, but the correlation was not very strong (Supplementary Fig. S5). 3′ss strength expresses the efficiency of 3′ss recognition likely mediated by U2AF, which is higher for YAG containing 3′ss. Therefore, there is a possibility that this correlation is a coincidence of binding specificities of different proteins. In fact, the Prp8’s binding specificity could explain the observed dNu much better, because it takes place just before the second transesterification step, when NAGs are physically chosen.

All the above-mentioned aspects are assigned to the 3′ss. In this work, we firstly described that 3′ss choice is also driven by the 5′ss sequence. If an alternatively spliced exon is located upstream of a NAGNAG containing exon, the 5′ss are responsible for differences in dNu among alternatively spliced transcripts. The main 5′ss determinants were represented especially by the third and fifth intronic positions. There are at least two possible mechanisms for the 5′ss to influence 3′ss choice: it can be done either indirectly by BP selection before the first or directly before the second splicing reaction.

Regarding the direct mechanism, the 5′ss should participate in AG selection after the C complex is formed. Before C-to-C* spliceosome complex transition, the intronic part of 5′ss is located in the catalytic centre, forming the lariat’s central part. This whole structure must be remodeled to release the lariat which is subsequently replaced by the second exon. The presence of non-optimal BP can result in a second reaction blockage that can be suppressed by PPT prolongation and/or its modification [45]. It shows that the second step’s efficiency is dependent on real nucleotide composition of lariat structure including the 5′ss intronic part and 3′ss choice might be driven by the same mechanism. Thus, nucleotides around the branch site could also influence the AG search dynamics and subsequently modify dNu, as shown in our experiments.

Selecting the branch point could be a conceivable mechanism enabling dNu diversification before the first splicing reaction indirectly. In AFAP1L2, using BP in position −23 supported distal 3′ss at a higher frequency than BP in −27. It corresponds to our knowledge about the BP distance to the 3′ss as an important 3′ss choice driver [1, 15]. Although competition among different BP can be modulated by 5′ss, an AG search performed by a spliceosome is a final arbiter driving 3′ss usage.

In the search for the functional relevance of NAG choice, some studies concluded that isoform ratios in different tissues varied [1, 3, 46–49]. However, variations across different tissues are rather weak, which suggests that tissue-specific splicing factors have only a minor influence on NAGNAG splicing [50, 51]. It seems that NAG choice is driven mainly by factors which apply irrespective of tissue type, which is highly consistent with the scanning mechanism performed by the spliceosome itself. Despite this preview, Szafranski et al. [47] discovered a novel regulatory mechanism termed “physiologically triggered, a concerted shift in alternative splicing” (PCAS). PCAS is an alternative splicing shift occurring as a reversible response to specific physiological conditions.

The effect is most likely mediated by a diffusible molecule such as glucose, metal ions, oxygen, or accumulating metabolic products and a direct PCAS signaling target is probably the spliceosomal core. However, the PCAS factor’s identity and the functional role of the mechanism is still unknown. Altogether, there is still a long way to fully understanding the exact NAGNAG splicing regulation’s mechanism.

Materials and methods

Minigene construction

The analyzed exons and flanking introns’ genomic fragments (exons 7 and 8 of CTSD gene (NM_001909.5), and exons 14 and 15 of AFAP1L2 gene (NM_001001936.3)) were amplified from the healthy control’s genomic DNA, using Platinum Pfx DNA polymerase (Thermo Fisher Scientific) (Supplementary Table T5) and subsequently cloned into the pET01 vector (MoBiTec) using ApaI and SmaI restriction enzymes [23]. The final expression cassette is depicted in Fig. 1. The minigene variants were prepared by site-directed mutagenesis using specific primers, and PrimeSTAR HS DNA Polymerase (Takara) (Supplementary Table T5) and their nucleotide sequence were verified by Sanger sequencing. If necessary, particular NAGNAGs are abbreviated with double-letter combination corresponding to particular Ns (e.g., AAGAAG ~ AA). Nucleotide abbreviations follow IUPAC recommendations.

Cell culture

HeLa cells were obtained from Sigma-Aldrich and maintained in RPMI 1640 medium (Sigma-Aldrich) supplemented with 10% fetal bovine serum (Sigma-Aldrich). For cell transfection, they were seeded 1 day before transfection in a well of 12-well plate (1.2 × 10⁵ cells per transfection) and subsequently transfected by XtremeGene 9 (Roche) and 800 ng of plasmid DNA. RNA was extracted 24 h after transfection using a Quick-RNA MiniPrep Kit (Zymo Research).

RT-PCR analysis

RNA was reverse-transcribed with a Transcriptor First Strand cDNA Synthesis Kit (Roche) using random hexamers. Subsequent PCR was performed with Taq DNA polymerase (Thermofisher), forward primer pET_forward (pure or FAM-labeled) and reverse primer pET_reverse (Supplementary Table T5). Amplicons were visualized on 2% agarose gels and characterized by sequencing and/or by capillary electrophoresis (SEQme; Czech Republic) and analyzed by GeneMapper 4.1 (Thermo Fisher Scientific). The identity of particular transcripts was determined based on its fragment length and quantity as the peak area. If not mentioned otherwise, the frequency is expressed as a percent of all corresponding transcripts. In some specific cases, some transcripts were excluded from the analysis. For example, substitutions in D−1 to D+6 of AFAP1L2 affected cryptic 3′ss strength (MaxEnt score 5.12), which is located 19 nucleotides downstream of 5′ss. Subsequently, this cryptic 3′ss could be recognized and included in the transcript lacking exon 14 and having prolonged exon 15.

In silico and statistical analysis

Splice site strength was calculated using the in silico maximum entropy (MaxEnt) prediction model implemented in MaxEntScan [52].

To analyze the RNAseq data collected by Bradley et al. [1], the corresponding datasets were downloaded from the NCBI’s GEO database (accession number GSE30017) and analyzed by our scripts. Additionally, dNu for AFAP1L2 was also calculated from the RNAseq data of the thyroid gland collected in PRJNA30709. In short, the frequency of all transcripts was counted based on their genomic coordinates for all tissues collectively and subsequently sorted according to their real NAGNAG motif and nucleotides in the chosen 5′ss and 3′ss positions. Corresponding boxplots were created in R, and the difference in distal 3′ss usage between transcripts with different upstream 5′ss was evaluated using Fisher’s exact test.

The difference in distal 3′ss usage (dNu) values was evaluated using Mann–Whitney test for RNAseq data and t test for experimental data. The dependence between different isoform frequency and MaxEnt values was counted using regression analysis.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

PH, PS and TF contributed to the study conception and design. Material preparation, data collection and analysis were performed by PH, PS, TK and MK. Statistical analysis was performed by LR and the manuscript was written by PH and PS. All authors read and approved the final manuscript.

Funding

This work was supported by the Ministry of Education, Youth and Sports, Grant No. MUNI/A/1099/2019 and by the Centre for Cardiovascular Surgery and Transplantation, Grant No. 2020001.

Data availability

All data generated or analyzed during this study are included in this published article and its supplementary information files.

Code availability

Not applicable.

Declarations

Conflict of interest

The authors declare that they have no conflict of interest.

Ethical approval

Not applicable—research did not involve human participants.

Consent to participate

Not applicable.

Consent to publish

Not applicable.