Regulation of mRNA Abundance by Polypyrimidine Tract-Binding Protein-Controlled Alternate 5′ Splice Site Choice

Alternative splicing (AS) provides a potent mechanism for increasing protein diversity and modulating gene expression levels. How alternate splice sites are selected by the splicing machinery and how AS is integrated into gene regulation networks remain important questions of eukaryotic biology. Here we report that polypyrimidine tract-binding protein 1 (Ptbp1/PTB/hnRNP-I) controls alternate 5′ and 3′ splice site (5′ss and 3′ss) usage in a large set of mammalian transcripts. A top scoring event identified by our analysis was the choice between competing upstream and downstream 5′ss (u5′ss and d5′ss) in the exon 18 of the Hps1 gene. Hps1 is essential for proper biogenesis of lysosome-related organelles and loss of its function leads to a disease called type 1 Hermansky-Pudlak Syndrome (HPS). We show that Ptbp1 promotes preferential utilization of the u5′ss giving rise to stable mRNAs encoding a full-length Hps1 protein, whereas bias towards d5′ss triggered by Ptbp1 down-regulation generates transcripts susceptible to nonsense-mediated decay (NMD). We further demonstrate that Ptbp1 binds to pyrimidine-rich sequences between the u5′ss and d5′ss and activates the former site rather than repressing the latter. Consistent with this mechanism, u5′ss is intrinsically weaker than d5′ss, with a similar tendency observed for other genes with Ptbp1-induced u5′ss bias. Interestingly, the brain-enriched Ptbp1 paralog Ptbp2/nPTB/brPTB stimulated the u5′ss utilization but with a considerably lower efficiency than Ptbp1. This may account for the tight correlation between Hps1 with Ptbp1 expression levels observed across mammalian tissues. More generally, these data expand our understanding of AS regulation and uncover a post-transcriptional strategy ensuring co-expression of a subordinate gene with its master regulator through an AS-NMD tracking mechanism.


Introduction
Eukaryotes rely on post-transcriptional control of their gene expression programs to a remarkable extent. A compelling example of this trend is the ability of many mammalian transcripts to undergo alternative splicing (AS) regulated by interplay between RNA-encoded cis-elements and trans-acting factors [1][2][3][4]. Distinct AS patterns include singular and mutually exclusive cassette exons, alternative 59-and 39-terminal exons, intron retention events, and alternate 59 and 39 splice site (59ss and 39ss) choice [1,5]. Of these, the latter two categories (A5C and A3C), involve alternative utilization of exonic termini and constitute a major part of tissuespecific AS programs [6,7]. Many of these events are known to have biologically and medically important functions (e.g., [8,9] and references therein).
Several earlier studies have begun elucidating molecular mechanisms involved in A5C regulation. Important factors affecting recognition of 59ss include (1) intrinsic efficiencies, or strengths with which these elements interact with the U1 snRNP component of the spliceosome and (2) the presence of adjacent splicing silencer or enhancer sequences that can modulate AS outcomes by recruiting cognate trans-regulators [10]. A common A5C regulation strategy relies on a splicing silencer positioned between an upstream and a downstream 59ss alternatives (u59ss and d59ss) [11,12]. This often stimulates the u59ss though silencerdependent repression of the competing d59ss [11,12]. However, the situation is complicated by the fact that hnRNP family proteins interacting with classical splicing silencers may additionally activate splicing reaction when recruited downstream of a 59ss [13]. Thus, hnRNP binding between u59ss and d59ss alternatives could theoretically bias A5C towards the u59ss by either repressing the d59ss, stimulating the u59ss or both. It is generally unclear which of these three possibilities is realized in natural contexts since most published studies on A5C regulation mechanisms rely largely on recombinant or/and in vitro approaches [11][12][13].
Mounting evidence suggests that in addition to generating multiple protein isoforms from a single gene [2,14] AS is widely used to control gene expression levels [4,15,16]. A prevalent posttranscriptional mechanism regulating mRNA abundance involves coupling between AS and nonsense-mediated decay (NMD), a quality control mechanism targeting mRNAs containing premature translation termination codons (PTCs) for degradation [17,18]. AS-NMD plays important roles in diverse biological processes [19] including regulation of RNA-binding protein expression [20,21], granulocyte development [22], axonal guidance [23] and brain response to seizures [24].
An hnRNP family member called polypyrimidine tract-binding protein 1 (Ptbp1/PTB/hnRNP I; [25,26]) is known to control expression of several genes through AS-NMD. Ptbp1 homeostasis in proliferating cells is maintained by an AS-NMD-mediated autoregulation mechanism [27]. Dampening Ptbp1 levels in neurons by microRNA miR-124 triggers global changes in cellular AS patterns and leads to increased expression of at least three AS-NMD targets encoding brain-enriched Ptbp1 paralog Ptbp2/ nPTB/brPTB and post-synaptic proteins Gabbr1 and PSD-95/ Dlg4 [28][29][30]. These genes contain Ptbp1-repressible cassette exons essential for open reading frame (ORF) integrity. Skipping of these exons in the presence of Ptbp1 results in a frame-shift and triggers NMD. On the other hand, their inclusion upon Ptbp1 down-regulation leads to accumulation of translationally active mRNAs. It is currently unknown whether the repertoire of Ptbp1dependent AS-NMD targets is limited to brain-enriched mRNAs or if it could additionally include other types of transcripts, e.g. those undergoing down-regulation during nervous system development.
Here we carried out a systematic analysis of transcriptome-wide RNA sequencing (RNA-seq) data and uncovered a large repertoire of Ptbp1-regulated A5C and A3C targets. Strikingly, one of the newly identified A5C events participates in an unusual AS-NMD circuitry controlling the abundance of mRNA encoding Hps1, a subunit of the Rab32/38 guanine nucleotide exchange factor (GEF) essential for biogenesis of lysosome-related organelles and mutated in patients with Hermansky-Pudlak Syndrome (HPS; OMIM: 203300; [31][32][33][34]). We describe the mechanistic underpinning of this regulation and provide evidence that this posttranscriptional mechanism may play an important part in shaping Hps1 tissue-specific expression patterns.

Ptbp1 controls a number of alternate 59 and 39 splice site events
To uncover additional Ptbp1 targets, we adapted a previously described RNA-seq analysis algorithm relying on Fisher's exact test to identify significantly regulated A5C and A3C events (see e.g., [35]; Fig. 1A). After confirming functionality of this approach with training RNA-seq datasets from neuroblastoma CAD and fibrosarcoma L929 cells ( Fig. S1 and Tables S1 and S2) we repeated the analysis for our RNA-seq datasets obtained for CAD cells transfected with control siRNA, Ptbp1-specific siRNAs or a mixture of siRNAs against Ptbp1 and its brain-enriched paralog Ptbp2 with a largely overlapping AS regulation preferences [26,36,37] (siControl, siPtbp1 and siPtbp1/2, respectively; [29]; NCBI Gene Expression Omnibus accession number GSE37933). This identified 41 A5C and 52 A3C events consistently regulated in both siPtbp1 and siPtbp1/2 samples (Fisher's exact test p,0.05 and .5% difference in the percent spliced in statistic (y) [7]; Tables S3 and S4).
Of note, Ptbp1 had been previously proposed to regulate the A5C event in the Usp5 gene identified by our analysis [43]. This positive control and 11 examples of newly identified targets representing the four AS patterns (Ptbp1-induced bias towards u59ss, d59ss, u39ss or d39ss) were selected for RT-PCR validation. Satisfyingly, all 12 genes showed readily detectable AS changes upon Ptbp1 and Ptbp1/2 knockdown (Fig. 1C-D and Fig. S1B). Ptbp1 knockdown accounted for most of the effect in all targets except Usp5 where Ptbp2 contribution was substantial (Fig. S1B). We concluded that Ptbp1 was involved in large-scale regulation of A5Cs and A3Cs.
Ptbp1 regulates Hps1 expression through alternate 59ss choice coupled with NMD Consistently highest Dy values in the A5C category were detected for the exon 18 of the Hps1 gene that its homozygous loss-of-function form leads to type 1 HPS (Table S3 and Fig. 1D). Utilization of the u59ss was expected to generate a full-length Hps1 ORF, whereas splicing at the alternative d59ss was predicted to generate a PTC-containing version of exon 18 (18L) triggering NMD ( Fig. 2A). HPS is currently incurable condition associated with albinism, prolonged bleeding, ceroid storage and frequently, a progressive lung disease limiting patients' lifespan [32][33][34]. Similar symptoms are observed in the pale ear mouse model homozygous for a loss-of-function Hps1 allele [32,44].
To test if the expression of this medically important gene was indeed controlled by AS-NMD in a Ptbp1-dependent manner, CAD cells pre-treated with siControl, siPtbp1 or siPtbp1/2 were incubated in the presence of either cycloheximide (CHX), an inhibitor of protein synthesis also blocking NMD, or an equal

Author Summary
Mammalian gene expression is extensively controlled at the post-transcriptional level and understanding of the underlying mechanisms can provide important biomedical insights. Here we identified a number of novel alternate splicing (AS) events where the choice between competing splice sites (ss) is regulated by polypyrimidine tractbinding protein 1 (Ptbp1/PTB/hnRNP-I). A top-scoring event was the choice between alternate upstream and downstream 59ss (u59ss and d59ss) in the Hps1 gene mutated in patients with type 1 Hermansky-Pudlak Syndrome (HPS). Preferential utilization of the u59ss in the presence of Ptbp1 gives rise to stable mRNAs encoding a full-length Hps1 protein, whereas the d59ss bias triggered by Ptbp1 down-regulation generates RNA species cleared by nonsense-mediated decay (NMD). We show that Ptbp1 functions in this circuitry by activating the intrinsically weaker u59ss. Brain-enriched Ptbp1 paralog Ptbp2/nPTB/brPTB stimulated the u59ss utilization but with a considerably lower efficiency than Ptbp1. We propose that this mechanism accounts for a tight correlation between Hps1 with Ptbp1 expression levels observed in mammalian tissues. Overall, these data expand our understanding of AS regulation and uncover an AS-NMD-mediated tracking mechanism ensuring co-expression of master regulator and its subordinate gene.
amount of control solution (DMSO) (Fig. 2B-D and S2A). RT-PCR analysis of these samples using F1/R1 primers ( Fig. 2B-C) confirmed Ptbp1 dependence of the A5C switch and showed that utilization of the d59ss was significantly elevated in the presence of CHX (e.g., 2.45-fold increase for siPtbp1-treated samples; p = 8.4610 24 ), consistent with the sensitivity of the corresponding splice form to NMD. Importantly, RT-quantitative (q) PCR analyses of the above six samples using F2/R2 primers revealed significant down-regulation of the Hps1 mRNA steady-state levels upon Ptbp1 or combined Ptbp1 and Ptbp2 knockdown ( Fig. 2D; ttest p = 0.0017 for siControl vs. siPtbp1 and p = 0.025 for siControl vs. siPtbp1/2). This down-regulation effect was completely rescued by CHX treatment (Fig. 2D) indicating that a major fraction of Hps1 transcripts in the siPtbp1 and siPtbp1/2 samples was subjected to NMD. Similar changes in relative abundance of the two A5C forms and Hps1 mRNA expression were detected when we inhibited NMD with siRNA targeting its key component, Upf1 [17,18] (Fig. S2B-D).
We next wondered if the newly identified post-transcriptional mechanism could modulate Hps1 expression at the protein level. Since immunodetection of the endogenous Hps1 protein is complicated by its relatively low abundance [45], we constructed a CMV promoter-driven plasmid containing an EGFP ORF fused in frame with the relevant 39-terminal fragment of the Hps1 gene (Fig. 2E). CAD cells pre-treated with siControl, siPtbp1 or siPtbp1/2 siRNAs were transfected with this construct and analyzed by RT-qPCR and immunoblotting 72 hours posttransfection. RT-qPCR analysis using EGFP-specific primers (F3/R3) showed that, similar to the endogenous Hps1 mRNA, recombinant EGFP-Hps1 transcripts were significantly downregulated in the siPtbp1 and siPtbp1/siPtbp2 samples (Fig. 2E) and underwent corresponding A5C changes (Fig. S2E). On the other hand, expression levels of a similarly designed control construct containing EGFP ORF but lacking the Hps1 part were virtually unchanged upon Ptbp1 and Ptbp1/2 knockdown (Fig. 2E).
Immunoblotting analysis of the EGFP-Hps1-transfected samples with an EGFP-specific antibody detected a ,50 kDa EGFP-Hps1 fusion protein band that was absent in the mocktransfected sample (Fig. 2F). Importantly, the expression of EGFP-Hps1 protein decreased upon Ptbp1 and Ptbp1/2 knockdown ,4 and ,7 fold, respectively (ANOVA p = 1.  Taken together, these results strongly suggest that Hps1 expression levels are controlled by AS-NMD mediated by Ptbp1.

Hps1 is co-expressed with Ptbp1 in vivo
We wondered if the newly identified AS-NMD regulation could account for Hps1 expression patterns in vivo. In line with published reports [28,[46][47][48], our RT-qPCR analyses showed that Ptbp1 was expressed across a wide range of adult and embryonic tissues with the lowest levels observed in brain, heart, skeletal muscle and testis (Fig. 3A). When the same set of tissues was assayed for Hps1 mRNA, we detected a striking positive correlation between Ptbp1 and Hps1 expression levels (Pearson's correlation coefficient r = 0.951, p = 3.2610 216 ; Fig. 3B and Fig. 3D). In addition, both the u59ss and the d59ss Hps1 isoforms were detected in brain, heart, skeletal muscle and testis whereas only the u59ss isoform was present elsewhere (Fig. 3C). Overall, there was a strong negative correlation between Ptbp1 levels and the d59ss utilization efficiency (Pearson's r = 20.626, p = 6.7610 24 ; Fig. 3E). Similar relationships between Ptbp1 expression and incidence of corresponding splice forms were detected for other newly identified A5C and A3C genes (Fig. S3). Of note, expression patterns between Hps1 and Ptbp2 mRNA levels correlated in a negative fashion (Fig. S4). This argued against a major role of Ptbp2 in shaping Hps1 expression in vivo and likely reflected the reciprocal relationship between Ptbp1 and Ptbp2 [28,47,49]. We concluded that Ptbp1 but not Ptbp2 may control Hps1 abundance across mouse tissues.
Polypyrimidine elements between u59ss and d59ss of Hps1 exon 18 are necessary for the A5C regulation To gain insights into the molecular mechanism underlying Hps1 regulation, we prepared a minigene cassette containing Hps1 exon 18, exon 19 and the intervening intron under control of a doxycycline-inducible promoter [TRE-mini-1819(WT); Fig. 4A]. CAD cells pre-treated with siControl, siPtbp1 or siPtbp1/2 were transfected with this construct and the minigene-specific splicing patterns were analyzed by RT-PCR (Fig. 4B). Similar to the endogenous Hps1 mRNA, minigene-derived transcripts used preferentially u59ss in the siControl sample and d59ss in the siPtbp1 and siPtbp1/2 samples ( Fig. 4B-C). This indicated that cis-elements responsible for the dependence of Hps1 splicing pattern on Ptbp1/2 were located in a vicinity of the regulated exon 18.
Ptbp1 is known to form high-affinity complexes with repeated UCUC, UCUU, CUCU or UUCU motifs [26,48,50]. Two pyrimidine-rich stretches (Py1 and Py2) containing consensus tetramers in pyrimidine-rich contexts occur in Hps1 between the Ptbp1-regulated u59ss and d59ss (Fig. 4A) and this arrangement is conserved across mammalian species (Fig. S5). We addressed possible functional significance of these elements by mutating either Py1 or Py2 in the TRE-mini-1819 context (Fig. 4A) and repeating the CAD transfection experiment with the resultant TRE-mini-1819(Py1-mut) and TRE-mini-1819(Py2-mut) constructs. Strikingly, mutation of either of the two Py sequences was sufficient to completely abolish the A5C regulation with the splicing pattern shifting towards d59ss in siControl, siPtbp1 and siPtbp1/2 samples (Fig. 4B-C).
Overall, these experiments suggest that both Py sites are required to orchestrate Hps1 A5C regulation under physiological conditions. Of the two sites, Py2 plays a more decisive role than Py1 and Ptbp1 is a noticeably stronger regulator than Ptbp2.

Ptbp1 binding to Py1 and Py2 elements regulates the A5C
To test if Ptbp1 directly interacted with Py1 and Py2 sequences, we carried out a biotinylated RNA pull-down assay (Fig. 4D). Interaction between Ptbp1 and a wild-type Hps1 probe comprising both Py1 and Py2 sites was readily detectable by this approach  (Fig. 4D). However, mutation of either Py1 or Py2 noticeably reduced this interaction, with a greater reduction in binding affinity observed upon inactivation of Py2 ( Fig. 4D and Fig. S6D).
To examine whether Ptbp1 recruitment to the Py sites was responsible for biasing the Hps1 A5C towards u59ss, we prepared a synthetic RNA comprising the mini-1819(WT) cassette and analyzed splicing of this substrate in vitro using HeLa S3 nuclear extract (NE; Fig. 4E). After a 60-min incubation at 30uC, two splicing products were detected by RT-PCR using F1/R1 primers corresponding to splicing at the u59ss (,65%) and the d59ss (,35%) (Fig. 4E). Notably, when we immunodepleted Ptbp1 from the NE and repeated the experiment, the d59ss utilization increased to ,90% (t-test, p = 5.2610 23 ; Fig. 4E-F). Ptbp1 depletion had no effect on the efficiency of constitutive splicing of a control adenovirus-derived RNA substrate (AdV) (Fig. 4E). To further ensure that the change in the Hps1 splicing upon Ptbp1 withdrawal was a specific effect, we supplemented immunodepleted NE with purified recombinant Ptbp1 protein and repeated the analysis. Notably, the addition of increasing Ptbp1 amounts led to a progressive decline in the d59ss utilization and a corresponding increase in the u59ss utilization ( Fig. 4G-H). Less efficient u59ss rescue was observed when we used purified recombinant Ptbp2 instead of Ptbp1 (Fig. S7).
Thus, Ptbp1 binds to the Py1 and Py2 sequences within exon 18/L and directly biases the choice between the two alternate 59 splice sites towards u59ss. Similar to our above results, Ptbp2 is less efficient than Ptbp1 in promoting u59ss utilization in vitro.

Ptbp1 stimulates u59ss usage
Two alternative models could account for the above results: (1) direct activation of u59ss by Ptbp1 or (2) repression of d59ss indirectly biasing the choice towards u59ss. To distinguish between these possibilities, we prepared three TRE-mini-1719 minigenes comprising Hps1 exons 17, 18 and 19 along with the intervening introns (Fig. 5A). Of these, TRE-mini-1719(WT) contained intact exon 18 u59ss and d59ss and therefore was expected to be regulated similarly to the TRE-mini-1819(WT) minigene above. In the other two constructs, TRE-mini-1719(u59ss-mut) and TREmini-1719(d59ss-mut), the corresponding sites were inactivated by mutations thus allowing us to test whether Ptbp1/2 had an effect on utilization of the only remaining 59ss (Fig. 5A).
We introduced these constructs into CAD cells pre-treated with siControl, siPtbp1 or siPtbp1/2 and analyzed the samples 72 hours post-transfection by multiplex RT-PCR using a combination of two primer pairs (F1/R5 and F4/R4) designed to measure the ratio between expression levels of spliced and total minigene-specific transcripts (Fig. 5B). Three distinct RT-PCR products were detectable in TRE-mini-1719(WT) samples with the F4/R4 ''normalizer'' band at the bottom and the u59ss-and d59ssspliced variants of the F1/R5 amplicon at the top (Fig. 5B). As expected, the ratio between the two top bands changed upon Ptbp1 or Ptbp1/2 knockdown indicating an increased usage of the d59ss (Fig. 5B). TRE-mini-1719(u59ss-mut) and TRE-mini-1719(d59ss-mut) samples gave rise to two products: the F4/R4 normalizer and either the d59ss-or u59ss-spliced variant of the F1/ R5 amplicon, respectively (Fig. 5B). Importantly, down-regulation of Ptbp1 alone or in combination with Ptbp2 had no detectable effect on the d59ss-spliced product/normalizer ratio for TREmini-1719(u59ss-mut) but significantly reduced the u59ss-spliced product/normalized ratio in the TRE-mini-1719(d59ss-mut) samples ( Fig. 5B-C).
Similar results were obtained when we re-analyzed the above samples by RT-qPCR using splice junction-specific primers designed to distinguish between u59ss and d59ss use (Fig. S8A-D).
Indeed, siPtbp1 and siPtbp1/2 up-regulated the d59ss-spliced products and diminished the abundance of the u59ss-spliced ones in the WT, whereas d59ss utilization was not affected by siPtbp1 and siPtbp1/2 in the u59ss-mut transcripts. On the other hand, u59ss was clearly repressed by siPtbp1 and siPtbp1/2 in the d59ssmut transcripts. Thus, Ptbp1 appeared to activate the u59ss rather than repress the d59ss.
To test whether this could be a direct effect, we carried out an in vitro splicing assay with a synthetic mini-1819 RNA substrate mutated at the d59ss position [mini-1819(d59ss-mut)] and analyzed the reaction products by RT-PCR (Fig. 5D). As expected, a single u59ss-derived splice form was detected after a 60-min incubation with HeLa S3 NE (Fig. 5E). The mini-1819(d59ss-mut) splicing efficiency was dramatically diminished when we immunodepleted Ptbp1 from the NE (,10-fold down-regulation; t-test p = 1.1610 26 ; Fig. 5E-F). Analysis of the reaction products by RT-qPCR confirmed that this reduction in splicing efficiency (t-test p = 1.4610 26 ; Fig. S8H) is accompanied by a reciprocal increase in the pool of unspliced RNA (Fig. S8F). Importantly, supplementing immunodepleted NE with purified recombinant Ptbp1 rescued mini-1819(d59ss-mut) splicing in a dose-dependent manner (Fig. 5G-H and Fig. S8I). We concluded that Ptbp1 regulates Hps1 A5C by stimulating the u59ss.

Regulation of Hps1 AS depends on difference between u59ss and d59ss strengths
Our data so far suggested that Ptbp1 interacts with Py1 and Py2 sequences within exon 18 and enhances u59ss utilization. Interestingly, predicted splicing strength of u59ss was lower than that of d59ss (scores S u59ss = 76.1 vs. S d59ss = 94.1 obtained using Analyzer Splice Tool server http://ibis.tau.ac.il/ssat/ SpliceSiteFrame.htm; [51,52]) and similar differences were detected in other mammalian species (Table S5). To test if this feature was important for the regulation, we generated a series of modified TRE-mini-1819 minigenes where the natural u59ss was substituted with the d59ss or/and the d59ss was substituted with the u59ss [TRE-mini-1819(d59ss/d59ss), TRE-mini-1819(u59ss/u59ss) and TRE-mini-1819(d59ss/u59ss); Fig. 6A]. All of these permutations lowered the DS = S d59ss -S u59ss difference between the two 59ss strengths. Notably, when we transfected CAD cells with the corresponding minigenes, the upstream 59 splice position was constitutively selected in all siRNA-treated samples ( Fig. 6B-C). Similar effects were observed when we weakened the u59ss or strengthened the d59ss by substituting them with synthetic 59ss sequences (Fig. S9). These results are consistent with the model that Hps1 A5C regulation requires u59ss to be weaker than d59ss.

The mechanism regulating Hps1 A5C may recur in other genes
We finally asked whether other A5C events uncovered in our bioinformatics screen featured pyrimidine-rich sequences between u59ss and d59ss and a weaker u59ss. To this end, we measured density of putative Ptbp1-binding tetramers (UCUC, UCUU, UUCU, CUCU) between u59ss and d59ss in three classes of A5C events: (1) biased towards u59ss in the presence of Ptbp1, (2) biased towards d59ss in the presence of Ptbp1 and (3) 100 randomly selected instances of Ptbp1-insensitive A5C (Fig. 7A and Table  S6). This analysis showed that the incidence of Ptbp1 motifs was significantly higher in the class 1 events compared to the class 3 control (KS test, p = 0.0041) whereas the class 2 events did not significantly differ from the control (KS test, p = 0.25).

Discussion
Mammalian gene expression is extensively controlled at the post-transcriptional level and understanding molecular mechanisms underlying this regulation can generate valuable biomedical insights. In this study, we interrogated transcriptome-wide RNAseq data and uncovered a number of functionally diverse Ptbp1dependent A5C and A3C events. We demonstrated that Ptbp1 directly controls the choice between the u59ss and the d59ss in Hps1 exon 18 (Fig. 7C). Both Ptbp1 binding motifs (Py1 and Py2) were required for the regulation at physiological Ptbp1 concentrations ( Fig. 4A-C) as well as for optimal binding of Ptbp1 to corresponding RNA probes ( Fig. 4D and Fig. S6D). Moreover, the splicing switch could be recapitulated by altering Ptbp1 concentration in vitro (Fig. 4E-H and Fig. S7). The Ptbp1 paralog Ptbp2 contributed little to the Hps1 A5C control (Fig. S4, Fig. S6A-C and Fig. S7).
Previous work on A5C mechanisms has suggested that recruitment of an hnRNP protein between two alternate 59ss may bias the AS choice towards the u59ss by either activating this site directly, repressing its d59ss competitor or a combination of the two effects [11][12][13]. In the case of Hps1, Ptbp1 appears to achieve this effect by directly stimulating a relatively weak u59ss rather than repressing its intrinsically stronger d59ss competitor (Fig. 5,  Fig. 6 and Fig. S9). Since A5C targets with Ptbp1-induced u59ss bias show enrichment of pyrimidine-containing motifs between u59ss and d59ss and their u59ss tends to be weaker than and the d59ss (Fig. 7A-B), Hps1-like A5C regulation may recur in other genes.
What could be the mechanism allowing Ptbp1 to activate the Hps1 u59ss? One possibility might involve stimulation of U1 snRNP recruitment to the u59ss by the Ptbp1 complex assembled at the Py1 and Py2 sequences. Similar strategies have been proposed to mediate activation of an upstream 59ss by the hnRNPlike protein TIA-1 [13,53] and other RNA-binding proteins interacting with downstream intronic splicing enhancers [9,54]. Moreover, Ptbp1 complex assembled on the c-Src pre-mRNA in the vicinity of the AS exon N1 has been shown to form contacts with U1 recruited to the N1 59ss [55]. Although this leads to repression of N1 splicing, interaction between Ptbp1 and U1 might result in opposite effects in other AS contexts with distinct structures of the ternary complex between Ptbp1, pre-mRNA and U1. Similar to position-specific effects on AS observed for other RNA-binding proteins [56][57][58], Ptbp1 tends to function as a splicing repressor when recruited upstream or/and within AS exons and as an activator when bound to downstream sequences [48,59]. Additional characterization of the Hps1 A5C may shed new light on this poorly understood phenomenon.
Intriguingly, functionality of the Hps1 A5C appears to rely on a finely tuned balance between the u59ss and d59ss strengths, since all mutations strengthening the relatively weak u59ss or weakening the relatively strong d59ss lead to constitutive utilization of the u59ss ( Fig. 6 and Fig. S9). It is somewhat surprising that recombinant Hps1 transcripts containing two equally strong or equally weak 59ss fail to generate a mixture of the two splice isoforms upon Ptbp1 and Ptbp2 withdrawal. This might hint at the existence of additional factors biasing the A5C towards the u59ss. Interestingly, u59ss rescue by purified Ptbp1 in Ptbp1-depleted in vitro splicing reactions was incomplete in a subset of our assays (e.g., compare Fig. 5E and Fig. 5G). This would be consistent with direct interaction of a hypothetical u59ss-stimulating factor with Ptbp1 protein. We plan to address this interesting prediction in our future studies.
The AS-NMD circuitry identified in our work (Fig. 7C) may account for tissue-specific Hps1 expression. We show that splicing at the u59ss gives rise to functional Hps1 ORF whereas utilization of the d59ss generates NMD-susceptible transcripts (Fig. 2). Since Ptbp1 is required for selecting the u59ss alternative, this mechanism likely ensures a strong positive correlation between Hps1 and Ptbp1 expression levels across tissues (Fig. 3). Type 1 HPS caused by homozygous loss-of-function mutations in Hps1 is typically manifested by reduced pigmentation, prolonged bleeding and lysosomal storage defects in many tissues. Further complications include inflammatory bowel disease and life-limiting pulmonary fibrosis [32][33][34]44]. Despite the multi-organ nature of this syndrome, HPS patients and pale ear mice do not typically develop neurological, cardiac or muscular problems  [32][33][34]44,60]. This would be consistent with the naturally low Hps1 levels in tissues expressing little Ptbp1.
Ptbp1 has been previously shown to regulate expression levels of several genes through AS coupled with NMD or nuclear retention and elimination (NRE) of aberrantly spliced transcripts [28][29][30]. However, in all of these cases Ptbp1 down-regulation increased steady-state levels of the corresponding mRNAs in the neuronal lineage. Thus, Hps1 provides a remarkable example of AS-NMD circuitry enabling tight co-expression of a target gene and its posttranscriptional master regulator. One possible advantage of this strategy could be ''de-noising'' of the Hps1 expression outputs in the presence of Ptbp1, since Ptbp1 own expression is stabilized by an auto-regulatory AS-NMD feedback loop [27]. On the other hand, this may allow developmental dynamics of Hps1 to be synchronized with expression changes in other Ptbp1 targets thus maximizing the overall coordination of cellular differentiation process.
In conclusion, our work uncovers a large set of Ptbp1-controlled A5C and A3C events and provides molecular insights into mechanism regulating expression output of the disease-related Hps1 gene. We predict that further examples of the master regulator tracking strategy described here for Hps1 will be identified in the future.

Materials and Methods
Plasmids pGEM3Zf(+) and pEGFP-C1 vectors were from Promega and Clontech, respectively. AdML-M3 construct encoding an adenovirus-specific splicing substrate (Addgene #11244) and pEM275 and pEM288 plasmids encoding FLAG-tagged Ptbp1 and Ptbp2, respectively, were described previously [28,61]. New constructs were generated using standard molecular cloning techniques and enzymes from NEB as outlined in Table S7. Site-specific mutagenesis was done using KAPA HiFi DNA polymerase (KAPA Biosystems) and corresponding mutagenic primers (Table S8). All plasmid maps and sequences are available on request.

Cell cultures
CAD cells (Cath.a-derived mouse neuroblastoma) [62] were cultured in Dulbecco's Modified Eagle Medium/High Glucose (DMEM; GIBCO, USA), supplemented with 11% FetalClone III Serum (Hyclone, USA), 1 mM sodium pyruvate (GIBCO, USA), 100 IU/ml penicillin and 100 mg/ml streptomycin, at 37uC in the presence of 5% CO 2 . For transfection experiments, cells were plated in the CAD medium without antibiotics at a density of 4610 5 cells per well of a tissue culture 6-well plate. Twelve hours post-plating, cells were transfected with corresponding siRNAs (ThermoScientific Dharmacon, USA) using Lipofectamine RNAi-MAX (Invitrogen, USA). Following 36-hour incubation, cell cultures were typically re-transfected with 1 mg of a minigene plasmid using Lipofectamine 2000 and incubated for another 36 hours prior to RNA harvest. In some experiments, cells were treated with either 100 mg/ml of CHX dissolved in DMSO or DMSO control for 8 hours. In the FLAG-Ptbp1 and FLAG-Ptbp2 over-expression experiments, 35 ng of pEM275 or 90 ng of pEM288 was co-transfected with 100 ng of Hps1 TRE-mini-1819 minigene and the total DNA amount was adjusted to 1 mg with an EGFP-encoding control plasmid (pCIG) and incubated for 48 hours.

RT-PCR and RT-qPCR
Total RNA was harvested from adherent cells using Trizol (Invitrogen). RNA was subsequently treated with 50 units/ml of RQ1 DNase (Promega) at 37uC for 1 hour to eliminate traces of genomic DNA. First-strand cDNA synthesis (RT) was typically performed in 10 ml reactions containing 2.5 mg of total RNA, 50 pmol of a random decamer primer (N10), 40 units of rRNAsin (Promega) and 100 units of SuperScript III reverse transcriptase (Invitrogen) at 50uC for 1 hour. Regular PCRs were carried out using Taq DNA polymerase (KAPA Biosystems) and amplification products were resolved by gel electrophoresis in 2% or 3.5% agarose gels. Quantitative PCR (qPCR) assays were done in triplicate using SYBR FAST qPCR Master Mix (KAPA Biosystems) and a StepOnePlus real-time PCR system (Applied Biosystems) and the signals were normalized to Gapdh mRNA levels. Relevant primer sequences are provided in Table S7.
Biotinylated RNA pull-down assays RNA probes were generated by transcribing linearized plasmid DNA in vitro with T7 polymerase (Promega) and biotin RNA labeling mix (Roche) for 2 hours at 37uC. Reactions were stopped by adding 1 unit of RQ1 DNase per 1 mg of template DNA and incubating the mixtures at 37uC for 15 min. Biotinylated RNAs were then extracted using phenol-chloroform (1:1) mixture, precipitated with ethanol and resuspended in DEPC-treated water. Pull-down assays were carried out by incubating 2 mg of purified RNA probes in 20 ml of buffer D (20 mM HEPES, pH 7.9, 100 mM KCl, 20% Glycerol, 0.5 mM DTT and 0.2 mM EDTA) supplemented with 80 ng yeast tRNA, 2.5 mg heparin, 40 units of rRNAsin (Promega) and 50% HeLa S3 NE (vol/vol; dialyzed against buffer D; ,100 mg protein in total) for 30 min at room temperature. RNA-protein complexes were then incubated with 20 ml of Streptavidin Sepharose Beads (Sigma) pre-washed in buffer D for 1 hour at 4uC. The beads were then washed thrice with buffer D and the RNA-associated proteins were eluted by boiling the beads for 10 min in 30 ml of 16 SDS PAGE sample buffer (0.0625 M Tris-HCl pH 6.8, 2% SDS, 5% b-mercapthoethanol, 10% glycerol and 0.01% bromophenol blue) and subsequently analyzed by immunoblotting.

Bioinformatics
To identify A5C and A3C events, fastq RNA-seq files for CAD cells treated with siControl, siPtbp1 or siPtbp1/2 ( [29]; NCBI Gene Expression Omnibus accession number GSE37933) were analyzed using TopHat aligner [63] and mm9 mouse genome assembly. The aligned junction read files were then processed using in-house Perl scripts (Dataset S1) designed to identify all possible pairs for A5C (u59ss-c39ss and d59ss-c39ss) and A3C (c59ss-u39ss and c59ss-d39ss) junctions across experimental samples. Junction reads corresponding to cassette exons were depleted by requesting that u59ss in A5C and d39ss in A3C pairs map to a known exon present in the UCSC gene, RefSeq gene or mRNA libraries (http://genome.ucsc.edu/). A5C and A3C pairs undergoing significant changes were identified by Fisher's exact test using R (http://CRAN.R-project.org/doc/FAQ/R-FAQ.html) (Dataset S1).

Ethics statement
All mouse work was conducted according to protocol approved by the Institutional Animal Care and Use Committee (IACUC) of Nanyang Technological University, Singapore. No surviving procedures were used. Mice were euthanized using isoflurane overdose procedure as recommended by IACUC. Figure S1 Optimization of the A5C and A3C discovery pipeline and validation of newly identified events. (A) To make sure our A5C/A3C analysis pipeline performed adequately, we first analyzed training RNA-seq data from CAD and L929 cells expected to exhibit markedly different AS patterns. This uncovered 195 A5C and 171 A3C significant cell line-specific events (p,0.05, Fisher's exact test) with apparent differences in the isoform-specific percent spliced in statistic (y [7]) exceeding 5% (Tables S1 and S2) Fig. 5B. Note that siPtbp1 and siPtbp2 reduce u59ss utilization in TRE-mini-1719(d59ss-mut) samples but have no detectable effect on d59ss utilization in TRE-mini-1719(d59ss-mut) samples. (E-F) RT-qPCR quantitation of residual unspliced mini-1819(d59ss-mut) RNA substrate after incubating it for 60 minutes with control-or Ptbp1-depleted NEs. Note that significantly more mini-1819(d59ss-mut) RNA substrate remains unspliced in the Ptbp1-depleted samples. (G-I) RT-qPCR quantitation of spliced products for experiments described in Fig. 5E        Dataset S1 RNA-seq data analysis pipeline for identifying significantly regulated A5C and A3C events. (ZIP)