AG‐exclusion zone revisited: Lessons to learn from 91 intronic NF1 3′ splice site mutations outside the canonical AG‐dinucleotides

Abstract Uncovering frequent motives of action by which variants impair 3′ splice site (3′ss) recognition and selection is essential to improve our understanding of this complex process. Through several mini‐gene experiments, we demonstrate that the pyrimidine (Y) to purine (R) transversion NM_000267.3(NF1):c.1722‐11T>G, although expected to weaken the polypyrimidine tract, causes exon skipping primarily by introducing a novel AG in the AG‐exclusion zone (AGEZ) between the authentic 3′ss AG and the branch point. Evaluation of 90 additional noncanonical intronic NF1 3′ss mutations confirmed that 63% of all mutations and 89% (49/55) of the single‐nucleotide variants upstream of positions ‐3 interrupt the AGEZ. Of these AGEZ‐interrupting mutations, 24/49 lead to exon skipping suggesting that absence of AG in this region is necessary for accurate 3′ss selection already in the initial steps of splicing. The analysis of 91 noncanonical NF1 3′ss mutations also shows that 90% either introduce a novel AG in the AGEZ, cause a Y>R transversion at position ‐3 or remove ≥2 Ys in the AGEZ. We confirm in a validation cohort that these three motives distinguish spliceogenic from splice‐neutral variants with 85% accuracy and, therefore, are generally applicable to select among variants of unknown significance those likely to affect splicing.

Several studies from our and other groups have shown that for the NF1 gene, which is mutated in individuals with neurofibromatosis type 1 (NF1), the proportion of pathogenic variants that alter splicing, for simplicity named splice mutations throughout the paper, is among the highest found in human disease genes (Ars et al., 2000;Messiaen et al., 2000;Wimmer et al., 2007). About two-thirds of these splice mutations are located outside the canonical GT and AG nucleotides (Messiaen & Wimmer, 2008). Several reasons may in concert account for these observations. First, a high proportion of splice mutations may be intrinsic to the structure of this large gene which is composed of 57 constitutively spliced exons and, hence, has a high number of splice sites that may be altered by mutations. Second, even "leaky" NF1 splice mutations, that is splice mutations leading to a splice effect only in a proportion of, but not in all, transcripts from this allele, may result in an attenuated but still clinically recognizable NF1 phenotype. This has been observed in a number of patients with typical cutaneous NF1 features such as multiple café au lait maculae (CALM) and cutaneous neurofibromas (see e.g., Fernandez-Rodriguez et al., 2011), including a case that will be discussed in this report. This phenomenon may differentiate NF1 from other genes associated with less specific symptoms, for example an increased cancer risk, that may not be recognized when present in an attenuated/mild form. In other words, the high frequency (compared to other genes) of noncanonical NF1 splice mutations observed may in part result from ascertainment bias, since mild NF1 phenotypes resulting from "leaky" splice mutations may come more frequently to clinical attention than attenuated phenotypes in other syndromes. Finally, and probably most importantly, splice defects are more effectively uncovered in this gene than in other genes, since our and a few others laboratories systematically apply RNA-based protocols as the first line mutation analysis assay. In our laboratories, direct complementary DNA (cDNA) sequencing of the entire coding sequence starting with RNA extracted from shortterm lymphocyte cultures treated with puromycin to prevent nonsensemediated RNA decay, proved to be highly sensitive also for the detection of noncanonical splice mutations (Messiaen & Wimmer, 2008).
Having applied this mutation analysis strategy for roughly two decades, we have identified a large number of noncanonical NF1 splice mutations fully characterized at the transcript level. They provide now a unique data set to gain insights into the mechanisms of action of noncanonical splice mutations, which in turn will help to establish the processes by which splice sites are recognized and selected. Furthermore, a better understanding of the mechanisms by which noncanonical splice mutations affect splicing will have immediate implications for molecular genetic diagnostics of NF1 and beyond, as it may lead to improved algorithms to stratify which gene variants found by sequencing of genomic DNA (gDNA) are likely to have an impact on splicing and should, therefore, be further analyzed at the RNA level. This will, in turn, reduce the number of variants of unknown significance (VUS) and improve the diagnostic data return from massive parallel gDNA sequencing.
In this study, we focused on mechanisms of action of noncanonical intronic 3′ss mutations. During the initial steps of splicing, degenerated sequence motives at the splice sites are bound by small nuclear ribonucleoproteins snRNPs. While the 5′ss is primarily defined by a 9-nuclotides (nts) sequence motive that base-pairs with the RNA moiety of the U1 snRNP, the sequence elements defining the 3′ end of an intron are more complex. They consist of three highly degenerate sequence motives: the branch point (BP), the polypyrimidine tract (PPT), and the 3′ss invariant AG-dinucleotide which is most frequently (96% of 3′ss) preceded by a pyrimidine residue (Y).
In the initial steps of 3′ss recognition (assembly of the complex E) the branchpoint sequence (BPS) is bound to the branch point binding protein SF1/BBP which later is replaced by the U2 snRNP (complex A). The 3′ss AG and the PPT are recognized and bound by the 35-and 65-kDa subunits of the U2 snRNP Auxiliary Factor (U2AF), respectively. U2AF65 forms a stable heterodimer with U2AF35 interacts with the BPS bound SF1/BBP and helps to recruit the SF1/BBP-replacing U2 snRNP to the BPS (Wahl, Will, & Luhrmann, 2009 and references cited therein).
An active spliceosome is formed after the recruitment of further snRNPs and several conformational rearrangements. The first step of splicing involves a hydrophilic attack by the 2′-OH of the BP adenosine on the 5′ss, leading to the formation of the 5′-exon and the intron lariat intermediate. In the second step, the 3′-OH of the 5′exon attacks the 3′ss, leading to exon joining and intron excision (Wahl et al., 2009). For accurate splicing, the spliceosome must locate the AG-dinucleotide that defines the 3′ss in this second catalytic step. Several lines of evidence show that the spliceosome has a strong preference for use of the first AG-dinucleotide downstream of the BP (Anderson & Moore, 1997;Chua & Reed, 2001;Krainer & Maniatis, 1985;Umen & Guthrie, 1995). As a consequence the sequence between the BP and the proper 3′ss AG is devoid of AGs and, therefore, is called an AG-exclusion zone (AGEZ; Gooding et al., 2006). The AGEZ mainly found appreciation as a tool to search for potential (distant) BP (Gooding et al., 2006).
Here we show that interruption of the AGEZ by a mutationcreated AG is the most frequent motive of action of noncanonical intronic 3′ss mutations. In two-thirds of the mutations, this leads to using the mutation-created AG instead of the authentic 3′ss, but in nearly half of the mutations, this leads also to exon skipping in a proportion or in all transcripts of the mutated allele. This suggests that the absence of AG dinucleotides in this region is not only necessary to ensure the selection of the authentic 3′ss during catalytic Step II of splicing, but also for accurate recognition and selection of the authentic 3′ss in the initial steps of splicing.
We further show that nearly 90% of all bona fide splice mutations follow this or two additional action motives. These three motives will be useful to select for further transcript analysis that intronic VUS found at a gDNA level that are most likely to have a splice effect.

| Patients
Between 1998 and 2018, in 8,690 unrelated patients sent for clinical NF1 testing to two centers, that is the Medical Genomics Laboratory at the University of Alabama in Birmingham (UAB) and the Institute of Human Genetics, Medical University Innsbruck (MUI), and NF1 reportable variant was identified. All were analyzed by an RNA-based protocol as the first line mutation analysis assay. The analyses were performed with informed consent from all patients or their parents.
The sporadic patient F8519 analyzed at MUI is a 28-year-old female. She fulfills the diagnostic criteria of NF1 with around 20-30 CALMs, bilateral Lisch nodules, at least five intradermal neurofibromas, a number of tiny nodules interpreted as cutaneous neurofibromas and an asymptomatic optic pathway glioma.
2.2 | Identification of intronic NF1 3′ splice site mutations outside the canonical AG-dinucleotides Direct cDNA sequencing of NF1 transcripts isolated from puromycintreated short-term lymphocyte cultures is the core mutation detection method of comprehensive NF1 mutation analysis as described by Messiaen and Wimmer (Messiaen & Wimmer, 2008;Messiaen & Wimmer, 2012;Messiaen et al., 2000). This assay reliably and effectively reveals mutation-induced splice alterations of the NF1 transcripts. The underlying genomic alteration is subsequently uncovered by PCR amplification and Sanger sequencing of the aberrantly spliced exon/intron. Using this approach a total of 78 different intronic 3′ss mutations outside the canonical AG-dinucleotides were identified in 137/8250 (1.7%) unrelated and molecularly confirmed NF1 patients subjected to mutation analysis at UAB and 13/440 (2.9%) index patients analyzed at MUI (Tables S1 and S2). Center; https://grenada.lumc.nl/LOVD2/mendelian_genes/) for NF1 mutations located at positions -50 to -3 of the 3′ss. Thirteen mutations with sufficient information on the associated splice effect in the original literature or deposited in the database had not been identified in our own cohort. These published mutations were also included in our analysis (Table S1). The mutations are described in accordance with the Human Genome Variation Society (http:// www.hgvs.org/mutnomen) guidelines using the NF1 mRNA sequence RefSeq NM_000267.3 as a reference with the A of the ATG start codon as position c.1. Exons/introns are numbered according to the reference sequence LRG_214 with in addition the widely known legacy numbering given in parenthesis (Messiaen & Wimmer, 2012).

| Minigene experiments
The RTB hybrid-minigene plasmid (Ryan & Cooper, 1996) previously provided by Dr. Thomas Cooper (Baylor College of Medicine) was used to construct minigenes containing the NF1 exons 15 (11) and 16 (12a) with flanking intronic sequences. Wild-type and mutant sequences were first PCR-amplified from the genomic DNA of patient F8519 (mutation c.1722-11A>G) and from a patient carrying mutation c.1722-2A>G. The PCR primer pair (Table S3) was designed to amplify a 2150-bp fragment containing the entire genomic sequence from intron 14 (10c; 230 bp upstream of exon 15) to intron 16 (12a; 212 bp downstream of exon 16). The PCR-products from the patients' genomic DNA contained primerintroduced SalI and KpnI sites at the 5´and 3´ends, respectively, which were used to clone the PCR fragments after digestion with these restriction enzymes into the SalI and KpnI sites located between the internal exon 2 and the last, exon 4, of the RTB minigene. Wildtype (C1) and mutant (C2, C7) minigene constructs were selected by colony PCR and subsequent sequencing.  Table S3). The PCR products were separated on 2.0% TAE-agarose gels. Amplification products were sequenced with primer RTBP4F.

| Splice site prediction
Five in silico splice site prediction programs that can be interrogated simultaneously by the commercial Alamut Visual software (vs 10.2; Interactive Biosoftware, Rouen, France) were used to predict the effect of the NF1 mutations and variants tested in the minigene constructs on the "strength" of splice sites, which is given by a score that is computed in each program by a different algorithm. In brief: WIMMER ET AL.

| 1147
Position weight matrices originally developed by Shapiro and Senapathy (1987) are used by the programs SpliceSiteFinder-like and Human Splicing Finder (Desmet et al., 2009). Position weight matrices are also used by the Human Splicing Finder to score potential branch points (Desmet et al., 2009). The program MaxEntScan is based on MaxEnt, an algorithm for derivation and scoring of constrained marginal maximum entropy distributions (Yeo & Burge, 2004). The program NNSPLICE uses the machine-learning approach of a neural networks to score splice sites (Reese, Eeckman, Kulp, & Haussler, 1997). GeneSplicer combines several splice site detection techniques, among which Markov models (Pertea, Lin, & Salzberg, 2001). The parameters were set in the Alamut Visual software (vs 10.2) to give numeric out-put data in the range from 70 to 100 for SpliceSiteFinder-like, 0-16 for MaxEntScan, 0.01-1 for NNSPLICE, 0-21 for GeneSplicer and 65-100 for Human Splicing Finder.
All NF1 mutations were analyzed with the bioinformatics program Splicing Prediction in Consensus Elements (SPiCE) that combines predictions from SSF-like and MES to generate sores that are used to predict the spliceogenicity of variants (Leman et al., 2018).
Given the severe limitations of predicting weakly conserved branch point sequences, predicted branch points were not used to define the beginning of an AGEZ. For simplicity, a mutation was considered to fall into the AGEZ when it was downstream (in 3′ direction) of the AG dinucleotide that is the first AG upstream (in 5′ direction) of the genuine 3′splice site AG. served as a control. In this latter construct (C5 in Figure 2a) the purine A of c.1722-12_1722-11delATinsTA is replaced by the purine G so that the interchange of nucleotides -11 and -12 does not create an AG-dinucleotide. In agreement with the notion that creation of an AG-dinucleotide within the AGEZ is a major mechanism by which the mutation c.1722-11T>G leads to loss of the use of the authentic 3′ss, transcripts from construct C6 showed skipping of exon 12a to the same extent as transcripts from constructing C2 while transcripts from construct C5 retained exon 12a just as the wildtype construct C1 (Figure 2c).
3.2 | Sixty-three percent of 91 intronic NF1 3′ splice site mutations outside the canonical AGdinucleotides create an AG-dinucleotide within the AGEZ and 90% fall into one of three simply defined categories that can be used to select possibly spliceogenic unclassified variants Prompted by these findings we investigated whether the creation of an AG-dinucleotide within an AGEZ could be a frequent mechanism of noncanonical intronic 3′ss mutations. To this end, we collected 77 additional NF1 mutations of this type identified by direct cDNA sequencing in our laboratories and 13 previously reported NF1 According to their splice effect, they fall into three categories: (a) mutations that cause exon skipping of the downstream exon (type 1 splice effect according to (Wimmer et al., 2007)), (b) mutations that lead to utilization of a pre-existing cryptic 3′ss (type 4 splice effect) and (c) mutations that create a novel AG-dinucleotide which is used as a novel 3′ss (type 3 splice effect; Figure 3b). Analyzing mutations affecting nucleotides upstream of position -3 (this is at positions -33 to -4), we noticed that 22 single nucleotide variants (SNVs) including one single nucleotide insertion, and 11 deletions, insertions or delins of more than one nucleotide had a type 1 or 4 splice effect (Table S1 F I G U R E 2 Minigene experiments confirm the splice effect of NF1 mutation c.1722-11T>G and reveal that the mutation acts mainly by creation of an AG-dinucleotide within the AG exclusion zone.  and S2). Nineteen of the 24 SNVs, including two single-nucleotide deletions, at position -3 led to a type 1 and/or type 4 splice effect and five to a type 3 splice effect (listed in part C and part D, respectively, of Tables S1 and S2).
Overall, 58 of 91 mutations created an AG-dinucleotide of which 57/91 (63%) are located within the AGEZ. These 57 mutations include 52/75 SNVs (green bars in Figure 3a) and 4/14 deletions/ delins (green lines in Figure 3) and 1/2 single nucleotide insertions (green arrowhead in Figure 3a). As can be deduced from Figure  within the AGEZ replace a pyrimidine by a purine (Table S2).
Taken together, these results suggest that the creation of an AG within the AGEZ accounts at least partly for the splice effect of 89% (49/55) of SNV mutations, including single-nucleotide deletions and insertions, upstream of position -3 ( Table 2, the first type of VUS). Other mechanisms may be responsible for the observed splice effects of mutations at position -3 and deletions/insertions of more than one nucleotide. Only 5/24 SNVs, including two single-nucleotide deletions, at position -3 create a tandem AG-AG of which the newly created AG, which is more proximal to the BP, is used (Tables S1 and S2, part D).
However, all but one SNV at this position replace a pyrimidine by a purine ( Table 2, the second type of VUS). This is in line with the notion that a pyrimidine (C or T) at position -3, which is found in 96% of all mammalian 3′ss (Shapiro & Senapathy, 1987;Zhang, 1998), is crucial for their recognition and/or use by the splicing machinery.
Eight of the 11 deletions and delins of two or more nucleotides lead to loss of at least two pyrimidines from the PPT, suggesting that they exert their splice effect by affecting the "strength" of the PPT (  Table 2, third line).
In conclusion, 81 (89%) of the 91 intronic 3′ss mutations outside the canonical AG-dinucleotide fall into one or more of the following F I G U R E 3 (a) Schematic presentation of 91 NF1 3′ splice-site mutations upstream of the canonical AG-dinucleotide. Intronic positions -33 to -1 and the first two nucleotides (1 and 2) of the following exon are indicated on the x-axis. The noncanonical single nucleotide substitutions are shown in the bar graph. The left bar shows the total number of substitutions per nucleotide position. Mutations that lead to skipping of the downstream exon (type 1) and/or usage of a cryptic 3´ss (type 4) are displayed in red and mutations that create an AG-dinucleotide which is used as a novel 3´ss (type 3) are displayed in blue. The green bars to the right represent the number of these substitutions that generate an AGdinucleotide in the AG exclusion zone (AGEZ). Deletions (horizontal lines), delins (horizontal lines and arrowheads) and insertions (arrowheads) are indicated below the bar graph. Lines and arrowheads in green indicate that the mutation generates an AG-dinucleotide in the AGEZ. The deleted and/or inserted nucleotides are given below the lines and arrowheads. Red letters indicate that the mutation leads to a type 1 or type 4 splice effect and blue letters indicate a type 3 splice effect. show here that creation of a novel AG within the AGEZ, that is the intronic sequence between the BP and the AG of the authentic 3′ss, is a frequent action motive of intronic sequence variants with a splice effect.
A novel AG in the AGEZ was created by 57/91 (63%) of all mutations and 49/55 (89%) of the SNV (including single-nucleotide deletions and insertions) upstream of position -3. Overall, this number is comparable to the 86% (62/72) of intronic mutations outside the 3′YAG that were found by Vorechovsky (Vorechovsky, 2006) to create AGs. However, in contrast to Vorechovsky's findings, according to which 93% (58/62) of these AGs were used as novel 3′ss in vivo (Vorechovsky, 2006), only two thirds (32/49) of the novel AGs analyzed in our study are used in all (n = 25) or at least in a proportion (n = 7) of the aberrantly spliced transcripts as novel 3′ss instead of the authentic 3′ss (Type 3 splice effect). This difference results from ascertainment bias in Vorechovsky's mutation cohort which consisted only of variants that either activated a cryptic 3′ss or were used as a "de novo" 3′ss but did not include variants leading to exon skipping. Overall, the view that the novel 3′ss appear to out-compete the authentic 3′ss in the mutated sequence context is reflected when comparing the scores of the novel 3′ss with the scores of the authentic 3′ss in the mutated sequence context using the programs Max-EntScan (MES; Yeo & Burge, 2004) and splice site finder (SSF) (Shapiro & Senapathy, 1987;Zhang, 1998). The mean scores of the novel 3′ss are substantially higher than the mean score of the authentic 3′ss in the mutated sequence context where it is substantially lower than in the wildtype context (see Table 3 and Table S2). However, there are exceptions to this general rule, especially when the newly generated AG-dinucleotide is located upstream of position -12 in the intron, that is upstream of the splice consensus sites described by Cartegni, Chew, and Krainer (2002) (see e.g., 2410-15A>G, 2410-14A>G in Table S2). Furthermore, the mean score of the novel 3′ss is lower than the mean score of the authentic 3′ss in the wildtype context. The main question arising from this finding is therefore by which molecular mechanisms the newly generated AG-dinucleotide interferes with the use of the authentic 3′ss which leads than either to the use of the variant-generated AG-dinucleotide as novel 3′ss (type 3 effect) or in other instances, to exon skipping or use of a pre-existing cryptic 3′ss that is different from the newly generated AG-dinucleotide (type 1 or type 4 effect, Figure 3b).
The model of a linear scanning mechanism that selects the first AG downstream of the BP in the second catalytic step of mRNA splicing (Umen & Guthrie, 1995) may be applicable to a type 3 splice effect. However, it is not applicable to type 1 and type 4 effects resulting from almost 50% of the 49 mutations either in all (n = 17) or a proportion of the aberrantly spliced transcripts (n = 7). As most (22/24) newly created AGs that lead to a type 1 or 4 splice effect are Y>R changes, they may act through decreasing the affinity of the PPT to U2AF65. However, using mutation c.1722-11T>G as an example our minigene experiments clearly show that the creation of a novel AG in the AGEZ is a strong action motive for these intronic 3′ splice sites, as well.
One possible explanation applicable to both types of splice effects arises from a study of Chen et al. (2017) which was aimed at defining the pathways by which the key components involved in the early steps of splicing recognize and select candidate 3′ss sites. Their data suggest that there is no intrinsic process that leads to a single U2AF binding to one pre-defined authentic 3′ss, instead they suggest that in the early stages of splicing (complex E) a number of candidate  (Bruun et al., 2016), it may also be speculated that the AGcreating mutations in the AGEZ could exert their effect on splicing by creating an hnRNPA1 binding motif. The observation that the majority of NF1 mutations (40/53) generating an AG in the AGEZ upstream of position -3 are preceded by a C or T and, therefore, fit with the 3′ss motive YAG also contained in the hnRNP A1 recognition motif, may be taken as a support for these hypotheses. Nevertheless, it should be noted that five of the NF1 SNVs with a type 1 and six with a type 3 splice effect create an AAG or GAG in the AGEZ upstream of position -3 (see Table S2). Furthermore, a minigene construct containing the NF1 variant c.1722-11T>A creating an AAG in the AGEZ exerts a strong splice effect similar to the splice effect of mutation c.1722-11T>G, which creates a TAG ( Figure S2).
Hence, although the exact molecular mechanisms of action remain to be understood, our results suggest that absence of AG-dinucleotides in the region between the BP and the AG of an authentic 3′ss may not only be necessary to ensure accurate AG selection during the catalytic Step II of splicing but also for appropriate assembly of the splicing components in the initial steps of 3′ss recognition and selection. Our findings may render new avenues to further study this model.
Two other frequent action motives that were identified in this study are removal or replacement by purines of >1 pyrimidine from the PPT/AGEZ and replacement of a pyrimidine at position -3 of the 3′ss by a purine. The most conceivable molecular mechanisms by which mutations following these motives interfere with the selection and use of the authentic 3′ss is by decreasing the affinity of the PPT and the 3′ss YAG to the U2AF65 and U2AF35 subunits, respectively.
Impact of our findings on mutation analysis in clinical and research settings: Independent from the question by which molecular mechanisms the analyzed intronic splice mutations impair the recognition and selection of the authentic 3′ss and lead to aberrant splicing, the identification of three motives of action of noncanonical 3′ss mutations will be clinically highly useful.  Shapiro & Senapathy, 1987;Zhang, 1998) as made available through the Alamut Visual software to predict whether a VUS is likely or unlikely to affect splicing (Houdayer et al., 2012). The suggested rule is to select any VUS with a mutant score of the authentic splice site that is 15% lower than the wildtype score of this authentic splice site when calculated with the program MES and also 5% lower than the wildtype score when calculated with SSF. This rule works with high sensitivity and specificity for VUS located at the splice site consensus sites described by Cartegni et al. (2002), but is not reliable outside these consensus sites. For the 3′ss this consensus sequence spans position -12 in the intron to +2 in the exon. Using logistic regression analysis on a larger data-set this strategy was further developed into a bioinformatics program named Splicing Prediction in Consensus Elements (SPiCE) that generates scores (Leman et al., 2018). Optimal sensitivity and specificity of splice effect prediction are reached on the analyzed variants when the thresholds were set on SPiCE scores of 0.115 and 0.749, respectively, with scores <0.115 predicted not to affect splicing and scores >0.749 predicted to affect splicing. About 10% of the analyzed variants fall into the "gray area" in-between these thresholds.
Calculating the SPiCE scores for 60 of our 91 noncanonical NF1 splice mutations, which are located within the defined 3′ss consensus site, two are false negatives (score <0.115) and three fall into the "gray area" while the remaining 55 have scores >0.749. Hence, depending on whether the threshold for further analyses is set at 0.115 or 0.749, the SPiCE program reaches a sensitivity of 97% (58/60) or 92% (55/60). Of note, four of the five NF1 splice mutations with a score <0.749, that is c.4515-3delT, c.6642-5T>G, c.6757-10T>G and c.6757-9T>G, also belong to the 5/60 mutations in this cohort that do not fall into one of the here proposed "actionable" motives/selection criteria for noncanonical 3′ ss mutations. The fifth mutation with a SPiCE score <0.115, c.2851-6_2851-3delCTTT, falls into the category of variants removing two or more pyrimidines from the PPT. Vice versa, the SPiCE score of 1.00 for mutation c.6642-7_6642-6insAAAA clearly predicts a splice effect of this mutation inserting four purines into the PPT, which is a rare mutation type that we did not define as a separate selection criteria.
Taken together, the three here defined action motives by which noncanonical 3'ss mutations exert their splice effect are well recognized by the splice site prediction tools MES and SSF, and consequently the SPiCE program. Hence, the selection of potentially splicogenic VUS will be highly sensitive and specific when applying for the SPiCE program as well as when applying the here defined selection criteria, but both approaches will have difficulties with variants that do not fall into one of the here identified action motives.
The major advantage of using the here defined selection criteria is their applicability also to variants located upstream of the position -12 where the SPiCE program is not applicable. Roughly a third (31/91) of all bona fide NF1 splice mutations affect nucleotides in this intronic region. 26 of these 31 mutations (84%) fall into one of the three action motives we identified and would, therefore, have been selected as potentially spliceogenic by the here defined selection criteria.

CONFLICT OF INTERESTS
The authors declare that there are no conflict of interests.

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available from the corresponding author upon reasonable request.