SAM68-Specific Splicing Is Required for Proper Selection of Alternative 3' UTR Isoforms in the Nervous System.

Neuronal alternative splicing is a core mechanism for functional diversification. We previously found that STAR family proteins (SAM68, SLM1, SLM2) regulate spatiotemporal alternative splicing in the nervous system. However, the whole aspect of alternative splicing programs by STARs remains unclear. Here, we performed a transcriptomic analysis using SAM68 knockout and SAM68/SLM1 double-knockout midbrains. We revealed different alternative splicing activity between SAM68 and SLM1; SAM68 preferentially targets alternative 3' UTR exons. SAM68 knockout causes a long-to-short isoform switch of a number of neuronal targets through the alteration in alternative last exon (ALE) selection or alternative polyadenylation. The altered ALE usage of a novel target, interleukin 1 receptor accessory protein (Il1rap), results in remarkable conversion from a membrane-bound type to a secreted type in Sam68KO brains. Proper ALE selection is necessary for IL1RAP neuronal function. Thus the SAM68-specific splicing program provides a mechanism for neuronal selection of alternative 3' UTR isoforms.

Introduction

Alternative pre-mRNA splicing is a powerful mechanism that generates molecular diversity from a limited number of genes and is therefore thought to be essential for biological complexity and diversity in mammals. In particular, the regulation is highly dynamic and complex in the central nervous system (CNS) (Barbosa-Morais et al., 2012, Merkin et al., 2012). Alternative splicing decisions are known to be dynamically switched during neural development (Kalsotra and Cooper, 2011, Vuong et al., 2016) and show distinct patterns in a neuronal tissue- or cell type-specific manner (Iijima et al., 2016, Nguyen et al., 2016, Li et al., 2007, Raj and Blencowe, 2015). Furthermore, neuronal activity modulates alternative splicing of neural genes via Ca2+-dependent signaling pathways (Razanau and Xie, 2013). Thus, neuronal alternative splicing is dynamically controlled in a spatiotemporal manner, which likely contributes to brain function complexity and diversity (Li et al., 2007, Raj and Blencowe, 2015). However, the RNA regulatory mechanisms underlying spatiotemporal and dynamic alternative splicing in neurons are only now being uncovered.

Neuronal alternative splicing is dynamically exerted by regulatory activity and unique expression patterns of RNA-binding proteins (RBPs). We previously identified SAM68 (Src-associated in mitosis of 68-kDa protein, khdrbs1) as a critical regulator of neuronal activity-regulated alternative splicing (Iijima et al., 2011). Moreover, two related proteins, SLM1 and SLM2 (SAM-like molecule 1 and 2), have been implicated in neuronal cell-type-specific splicing (Ehrmann et al., 2013, Iijima et al., 2014, Nguyen et al., 2016). SAM68, SLM1, and SLM2 belong to the STAR (signal transduction and activation of RNA) family of proteins, which share 70%–80% of amino acid sequence identities in their KH-type RNA-binding domains (Di Fruscio et al., 1999). Important targets of SAM68, SLM1, and SLM2 are the mRNAs encoding Neurexin (Nrxn) proteins (Iijima et al., 2016). Neurexins are synaptic cell surface receptors extensively regulated at alternative splicing level (Missler and Sudhof, 1998). All three STAR family proteins induce skipping of exon 20 at the Nrxn alternatively spliced segment 4 (AS4). The splicing decision at AS4 is critical for differential interactions with several ligands that are essential mediators of synaptic properties, including neuroligins, leucine-rich repeat proteins, and the Cbln1-GluD2 complex (Baudouin and Scheiffele, 2010, Boucard et al., 2005, Ko et al., 2009, Krueger et al., 2012, Matsuda and Yuzaki, 2011, Uemura et al., 2010). Indeed, the Nrxn AS4 is particularly important for synaptic strength and plasticity regulation (Aoto et al., 2013, Traunmuller et al., 2016), which is dynamically controlled by STAR family proteins in neuronal activity- and cell-type-specific fashions (Ehrmann et al., 2013, Iijima et al., 2011, Iijima et al., 2014, Nguyen et al., 2016, Traunmuller et al., 2016).

Several groups have previously identified additional substrates for SAM68 and SLM2 (Chawla et al., 2009, Ehrmann et al., 2013, Ehrmann et al., 2016, Huot et al., 2012, La Rosa et al., 2016, Traunmuller et al., 2016). Knockout mice of SAM68, SLM1, and SLM2 exhibit several morphological and functional defects in adult brains (Ehrmann et al., 2016, Iijima et al., 2011, Iijima et al., 2014, Lukong and Richard, 2008, Traunmuller et al., 2016). We previously found that Sam68 and Slm1 KO mice particularly have cerebellar malformation and motor deficits (Iijima et al., 2011, Iijima et al., 2014). Nevertheless, most neuronal functions of STAR family proteins in the mature brain remain unresolved. However, given that SAM68 and SLM1 are widely expressed in the brain throughout life, spatiotemporal regulation of alternative splicing by SAM68/SLM1 could play a critical role in multiple aspects of neuronal development, differentiation, and function. Thus, the recent findings pave the way to uncover and characterize novel targets for spatiotemporal alternative splicing programs by SAM68/SLM in the nervous system. Here we reveal that SAM68 shapes neuronal diversity of alternative 3′ UTR isoforms and demonstrate the critical role of the SAM68 splicing program in the proper 3′ UTR selection.

Results

Characterization of SAM68/SLM1-Dependent Alternative Splicing Programs

To decipher alternative splicing programs encoded by SAM68 and SLM1 proteins, we attempted to locate new candidate RNA substrates by microarray-based screening using SAM68/SLM1 knockout mice. We utilized the exon array on the primary experiments, the dataset was validated by RT-qPCR, and the altered exons were further confirmed by RNA sequencing (RNA-seq). We previously showed that SLM1 protein acts as a heteromeric complex with SAM68 in co-expressing neurons (Iijima et al., 2014). Given that STARs share 70%–80% of amino acid sequence identity in their RNA-binding domains (Di Fruscio et al., 1999), it is expected that SAM68 would share a significant amount of RNA substrates with SLM1 with functional redundancy. Therefore, for the initial transcriptomic analysis, we attempted to identify candidate RNA substrates in the midbrain of both SAM68/SLM1 double-knockout (Sam68/Slm1 DKO) mice and SLM1 single-knockout (Slm1 KO) mice. We focused this analysis on the midbrain because this area is a site of prominent co-expression of SAM68 and SLM1. Initially, we compared the levels of gene expression between wild-type (WT), Slm1 KO, and Sam68/Slm1 DKO mice. A scatterplot showed that the gene expression profiles of Slm1 KO and Sam68/Slm1 DKO mice were highly similar to those of WT mice (correlated efficiency: 0.996–0.997) (Figure S1A), indicating that knockout of SAM68 and/or SLM1 did not significantly influence overall transcript levels. In fact, the volcano plots showed that there were only 10–12 genes that are significantly altered in both Sam68/Slm1 DKO and Slm1 KO mice compared with WT (corrected p values < 0.05; threshold set: fold change [FC] ≥ 2.0) (Figure S1B, and Table S1). Validation by RT-qPCR showed that the gene alterations are partially shared between both genotypes (Figure S1C), but others are unique for either SAM68 or SLM1 (Figure S1D). However, Slm1 transcripts were not listed in the altered genes on the exon array. Although we previously confirmed that SLM1 protein is completely lacking in Slm1 KO mice (Iijima et al., 2014), the RNA-seq data exhibited that the transcripts lacking exon 2 remain expressed (data not shown). That is why Slm1 transcripts were not listed in the downregulated genes. Reportedly, SAM68 has multiple functions on RNA metabolism, and a multitude of RNA substrates including non-coding RNAs have been identified using other approaches (Li et al., 2017, Sanchez-Jimenez and Sanchez-Margalet, 2013, Vogel and Richard, 2012). Therefore the very modest number of transcriptomic changes identified in our sample was surprising. The results were largely confirmed by RNA-seq analysis in Sam68/Slm1 DKO mice (cor. efficiency: 0.986) (Figure S2A). Nevertheless, our results suggest that, even in Slm1/Sam68 double-knocked mice, the effect on total transcript levels is likely to be only minor in the mouse midbrain.

We next examined exon alteration between WT, Slm1 KO, and Sam68/Slm1 DKO mice. We observed that 122 and 172 exons were altered by more than 2.4-fold in Slm1 KO and Sam68/Slm1 DKO mice, respectively (Figure 1A). Given that the whole gene expression profiles were almost unchanged (Figure S1), the majority of the exon alterations were likely due to the change in splicing events. We then compared the altered profiles at the exon level between Slm1 KO and Sam68/Slm1 DKO midbrains. The Venn diagram exhibited that 66 of 228 exons overlapped between the genotypes. We also found that 106 exons were altered only in Sam68/Slm1 DKO mice (Figure 1B). These exons are likely to contain SAM68-specific targets. We also observed that 56 exons were altered only in Slm1 KO mice. Indeed, given our previous finding that splicing activity of Nrxn3 exon20 is quite opposite between the two proteins (Iijima et al., 2014), these could also include exons that are regulated differentially between SAM68 and SLM1. Gene ontology (GO) analysis of the altered exons in each genotype showed that major subsets were enriched for similar terms, but those in Sam68/Slm1 DKO mice were much more enriched for the neuronal terms (Figure 1C, red terms). Therefore, these results imply that SAM68 and SLM1 encode overlapping but distinct alternative splicing programs.

Figure 1

Comprehensive Comparison of Altered Exon Profiles between Slm1KO and Sam68/Slm1DKO Mice

Total RNAs from midbrains of WT, Slm1KO, and Sam68/Slm1DKO mice were subjected to data analyses on exon array (Agilent, Sure Print G3 Mouse Exon Microarray 2x400 K) (n = 3 animals/genotype).

(A) Scatterplots showing fold change for exons (Slm1KO versus WT, Sam68/Slm1DKO versus WT) (total 122 and 172 exons, respectively; threshold set: FC ≥ 2.4, raw probe signal intensity ≥100 in either of the two genotypes, normalized gene expression > −3 in either of the two genotypes) (n = 3 per genotype) (red and blue dots).

(B) Venn diagram showing the numbers of altered exons (total 228 exons; threshold set: FC ≥ 2.4, raw probe signal intensity ≥100 in either of the two genotypes, normalized gene expression > −3) in both Slm1KO and Sam68/Slm1DKO mice.

(C) Comparison of altered exons by GO analysis between Slm1KO and Sam68/Slm1DKO mice. Genes that encode altered exons (FC ≥ 2.4) shown in (A) (Slm1KO: 89 genes; Sam68/Slm1DKO: 112 genes) were subjected to GO analysis. Enrichment was thresholded by p value (p < 0.05). Red represents the neuronal terms.

The SAM68-Specific Splicing Program Preferentially Regulates Alternative 3′ UTR Exons of Neuronal Genes

To further pursue the potential difference in the splicing program between SAM68 and SLM1, we then classified significantly altered exons into five categories (coding sequence [CDS], 5′ untranslated region [5′ UTR], 3′ untranslated region [3′ UTR], duplicated [containing both CDS and UTR], and unknown [not annotated in refseq] exons), and compared the relative percentage of each altered exon between Slm1 KO and Sam68/Slm1 DKO mice. Interestingly, we noticed that there was a remarkable difference in the pattern of the exon alteration between Slm1 KO and Sam68/Slm1 DKO mice; 3′ UTR exons were preferentially altered in Sam68/Slm1 DKO mice (Figure 2A), although RNA-seq data in Sam68/Slm1 DKO mice showed that these exon alterations largely included all alternative exon events (i.e., cassette exons, mutually exclusive exons, alternative 5′ splice site, alternative 3′ splice site, and retained introns) (Figure S2B). Indeed, 3′ UTR exons were frequently observed in the top lists of significantly altered exons in Sam68/Slm1 DKO mice (Figure 2B). We listed 35 genes whose 3′ UTR exons were significantly altered (threshold set: FC > 2.4, p < 0.05) (Table S2). Twenty of 35 genes were unique for Sam68/Slm1 DKO mice. Importantly, arranged scatterplots of all exons (251 exons) in 35 genes showed that the alteration in 3′ UTR exons likely did not follow the change in their neighboring coding exons within their encoding genes (Figure 2C), indicating a specific alteration in alternative 3′ UTR isoform choice of these genes.

Figure 2

Altered 3′ UTR Exon Events Of Neuronal Genes in Sam68/Slm1DKO Brains

(A) Classification of exons altered in Slm1KO and Sam68/Slm1DKO mice on exon array datasets. Exons are classified into the following five categories: CDS, 5′ UTR, 3′ UTR, duplicated, and unknown exons. (threshold set: raw probe signal intensity ≥100 in either of the two genotypes, FC ≥ 2.0). The x axis represents the percentage of altered exons per classified exon. Annotation was referenced on Mouse July 2007 (NCBI37/mm9).

(B) The list of top 20 list exons that were significantly increased or decreased in Sam68/Slm1DKO midbrains (excluding unknown genes and genes including exons altered at the gene level) (threshold set: FC ≥ 2.4, raw probe signal intensity ≥100 in either of the two genotypes, normalized gene expression > −3 in either of the two genotypes, p < 0.05 [compared to WT]). SKO: Slm1KO; DKO: Sam68/Slm1DKO.

(C) Arranged scatterplots of all exons (total 251 exons) in 35 genes that include the significantly altered 3′ UTR exons in Sam68/Slm1DKO mice. CDS, 5′ UTR, 3′ UTR, duplicated, and unknown exons.

(D) GO analyses of genes that include altered 3′ UTR exons in Sam68/Slm1DKO mice; 78 genes that include significantly altered 3′ UTR exons (FC ≥ 2.0, p < 0.05) were subjected to GO analysis. Enrichment was thresholded by p value (p < 0.05). Keyword category (left). GO-enriched terms (right). Red represents the neuronal terms.

(E) Aberrant 3′ UTR exon selection of the representative genes, Il1rap and Cp, in Sam68/Slm1DKO brains shown by RNA-seq (Illumina Hiseq). The alignment of RNA-seq was based on the UCSC genome browser Mouse NCBI37/mm10 assembly.

Interestingly, GO analyses of the altered 3′ UTR exons in Sam68/Slm1 DKO mice predicted that significant numbers of these targets might include transcripts encoding transmembrane or secreted proteins with neuronal function (Figure 2D). Intriguing examples were exon 8b of Il1rap (interleukin 1 receptor accessory protein, synaptic adhesion protein), exon 26b of Pcdh15 (protocadherin-15, cell adhesion protein that plays an essential role in maintenance of normal retinal and cochlear function), exon 19 of Cp (ceruloplasmin, iron transporter), and exon 4b of Glra3 (glycine receptor alpha 3, glycinergic ion channel) (see Figure 2C). Indeed, RNA-seq analysis showed that the proximal 3′ UTR exons of these transcripts were markedly included in Sam68/Slm1 DKO mice, whereas these were almost excluded in WT mice (Figures 2E and S2C), resulting in a long-to-short isoform switch of several neuronal targets through alteration in alternative last exon (ALE) selection in Sam68/Slm1 DKO mice. In addition, because preferential alteration in 3′ UTR exons occurred in Sam68/Slm1 DKO, but not particularly in Slm1 KO mice (Figures 2A and 2B), we hypothesized that the aberrant choice of alternative 3′ UTR isoforms was largely caused by the single-knockout effect of SAM68. To clarify the possibility, the altered 3′ UTR exon events observed in the exon array were validated in Sam68 KO, Slm1 KO, and Slm2 mutant (Slm2 MT) brains separately by RT-qPCR analysis (Figure 3). Slm2 MT mice expressed SLM2 protein that lacks a first QUA domain (Figure S3), which results in a significant reduction in SLM2 activity toward alternative splicing of Nrxn AS4, a major SLM2 target in the brain (Figures S3E and S3F). In this analysis, we focused on eight genes (Il1rap, Cp, Pcdh15, Lrrcc1, Pcdh17, Dlgap1, Sema3a, and Fbxl3) observed only in Sam68/Slm1 DKO on the exon array. The RT-qPCR analyses revealed that these exon alterations did not occur in Slm1 KO and Slm2 MT mice, except for Fbxl3, and were specifically caused by single loss of Sam68 (Figure 3). The Sam68 KO-specific alternation included all three types of alternative 3′ UTR splicing events (ALE type [Figure 3A], ALE type with alternative 5′ splice site [Figure 3B], and alternative polyadenylation type [APA] [Figure 3C]). Thus, these data show that the SAM68-specific splicing program controls alternative 3′ UTR isoform selection.

Figure 3

SAM68-Specific Splicing of Alternative 3′ UTR Exons in the Nervous System

The usage of alternative last exon (ALE) or alternative polyadenylation (APA) of candidate RNA substrates was validated by RT-qPCR analysis using adult midbrains from WT, Sam68KO, Slm1KO, and Slm2MT mice. Fold change (FC) and significant differences were compared with WT mice. The threshold cycle (CT) value of total transcripts was normalized to that of Gapdh, whereas the relative quantification (RQ) value of each alternative isoform was normalized to that of each total mRNA (n = 3–6 animals per genotype).

(A) ALE: Three genes, Cp (Ceruloplasmin), Pcdh15 (Protocadherin 15), and Lrrcc1 (leucine-rich repeat and coiled-coil domain-containing protein 1).

(B) ALE with alternative 5′ splice sites: Three genes, Dlgap1 (disk large-associated protein 1), Il1rap (interleukin-1 receptor accessory protein), and Pcdh17 (protocadherin 17).

(C) Alternative polyadenylation type (APA): Two genes, Fbxl3 (F-box/LRR-repeat protein3) and Sema3e (semaphorin 3e). Data are presented as the mean ± SEM. Significance is indicated as follows: ***p < 0.001; **p < 0.01; *p < 0.05; Student's t test.

To further investigate the ALE choice by SAM68, we focused on alternative splicing of Il1rap (ALE with alternative 5′ splice site), Pcdh15 (ALE), Cp (ALE), and Glra3 (ALE). The RT-qPCR analyses revealed that short-form (SF) variants of Il1rap, Pcdh15, Cp, and Glra3 including proximal 3′ UTR exons were dramatically increased in the midbrain of Sam68 KO and Sam68/Slm1 DKO mice, whereas the long-form (LF) variant was reciprocally reduced (Figures 4A–4D, S4, and S5A). Notably, whereas >90% of Il1rap transcripts account for an LF variant in WT mice, >50% of these transcripts were occupied by the atypical SF variant in Sam68 KO and Sam68/Slm1 DKO mice (Figure 4C). By contrast, knockout of Slm1 did not affect any isoform levels of these transcripts and did not have additive effects with loss of SAM68 (Figures 4A–4D and S4). In addition, the isoform alteration in other analyzed transcripts as shown in Figure 3 (Lrrcc1, Pcdh17, Dlgap1, sema3e, and Fbxl3) also had no additive effects with the double knockout (Figures S5B–S5D). Thus, we confirmed that these ALE selections are specifically controlled by SAM68. Interestingly, at the protein level, inclusion of Il1rap exon 8b, Pcdh15 exon 26b, Cp exon 17, and Glra3 exon 4b results in production of soluble forms, lacking transmembrane domains or glycosylphosphatidylinositol anchor (Figure S6A). Indeed, when these soluble-form variants were expressed in HEK293T cells, significant amounts of protein products were detected in the cultured medium (Figure S6B). The majority of Il1rap, Pcdh15, Cp, and Glra3 transcripts are LF variants encoding transmembrane proteins in WT brains (Figures 4A–4D and S4). There are three ALEs in Il1rap, which produce two transmembrane (isoforms 1 and 3) and one soluble isoform (isoform 2) (Figure 4E, illustration). Consistent with the altered ALE selection at the transcript level, protein analysis by parallel reaction monitoring exhibits significant reduction in transmembrane protein isoform 1 in Sam68 KO brains relative to overall Il1rap protein levels (Isoform 1, 2, 3 [total]) (Figure 4E). These results indicate that aberrant ALE selection of these transcripts in Sam68 KO causes marked conversion into atypical secreted type of proteins in the nervous system.

Figure 4

Sam68KO Causes Atypical Long-to-Short Isoform Conversion of Il1rap and Cp via Aberrant Usage of ALEs

(A and B) Schematic illustration of alternative exon choice at Il1rap exon 8 and Cp at exon 13 (top panel) and the representative gel images of semi-quantitative RT-PCR with these 3′ UTR exon choices in midbrains from WT, Slm1KO, and Sam68/Slm1DKO mice (bottom panel). (A) Exon 8b on Il1rap and (B) exon 13 on Cp.

(C and D) Relative levels of total mRNA and two alternative isoforms (LF and SF variants) and abundance ratio of SF (red) to LF (blue) between midbrains from WT, Sam68KO, Slm1KO, and Sam68/Slm1DKO mice by RT-qPCR. The RQ value of total transcripts was normalized to that of Gapdh, whereas the RQ value of each alternative isoform was normalized to that of the total transcripts. For the abundance ratio of SF to LF, the percentage of the SF variant was largely estimated from the CT value (CT) directly compared with that of LF (CT) at the same threshold set for the CT value. RQ + RQ values for the total transcript level were set to 100%. RQ value of two transcripts was normalized to that of Gapdh; (C) Il1rap (D) Cp (n = 3–6 animals per genotype).

(E) Quantification of IL1RAP protein isoforms by parallel reaction monitoring (PRM). To quantify low-abundant protein isoforms, heavy reference peptides for Isoform 1/2/3 (total), Isoform 1, and Isoform 3 of IL1RAP were used in PRM-liquid chromatography-mass spectrometry. Plots show normalized endogenous (light) to reference (heavy) peak intensities of WT and Sam68KO hippocampal samples (n = 5 per genotype) or average changes between genotypes for Isoform 1 and 3 (Isoform 1/2/3 [total] set as reference). Data are presented as the mean ± SEM. Significance is indicated as follows: ***p < 0.001; **p < 0.01; *p < 0.05. One-way ANOVA followed by Bonferroni's test.

Soluble IL1RAP Influences Synaptogenic Signaling through Transsynaptic IL1RAP-PTPδ Interaction

Our data indicated that single loss of SAM68 caused aberrant ALE selection of Il1rap, Pcdh15, and Glra3 in Sam68 KO, resulting in marked conversion into atypical secreted type of proteins in the nervous system. Therefore, we then tested the influence of short/secreted isoforms on neuronal functions. A previous study revealed that IL1RAP and the paralog IL1RAP-like 1 (IL1RAP-L1) organize excitatory synapses through transsynaptic interaction with the protein tyrosine phosphatase δ (PTPδ), a member of the presynaptic cell adhesion molecule, in the nervous system (Yoshida et al., 2011, Yoshida et al., 2012) (Figure 5A). We examined the mRNA expression of Il1rap and of the related molecules in various brain regions and the developing cortex. The transcripts were ubiquitously expressed in whole brain tissues and throughout development (Figures S7A and S7B). In addition, ALE choice of Il1rap in Sam68 KO and Sam68/Slm1 DKO mice was altered at the same level between the cortex, midbrain, and cerebellum (Figure S7C). Here, we tested the effect of soluble IL1RAP (sIL1RAP) on IL1RAP-induced presynaptic organization and PTPδ-induced postsynaptic organization. To this end, we employed a co-culture system wherein primary cerebellar neurons are combined with non-neuronal cells expressing a single synaptogenic molecule (Scheiffele et al., 2000) (Figures 5B and S7D). Cerebellar culture is a highly homogeneous neuron culture, which is appropriate for this assay. IL1RAP-hemagglutinin (HA)-expressing HEK293T cells triggered robust levels of presynaptic differentiation, as measured by recruitment of the presynaptic marker synaptobrevin (VAMP2) (Figure S7E left). By contrast, co-expression with sIL1RAP-HA in HEK293T cells or introduction of sIL1RAP-HA into the cultured neurons with lentivirus significantly reduced the recruitment of the presynaptic marker, demonstrating the competitive effect of sIL1RAP on synapse organization mediated by IL1RAP-PTPδ interaction. The paralog IL1RAP-L1-expressing HEK293T cells also triggered presynaptic differentiation (Figure S7E middle). Similar to IL1RAP, co-expression with sIL1RAP-HA in HEK293T cells significantly reduced IL1RAP-L1-induced recruitment of the presynaptic marker (Figure S7E middle) but did not affect neuroligin-1-induced recruitment (Figure S7E right), confirming the competitive effect of sIL1RAP on other PTPδ-mediated synapse organization. We next examined the influence of sIL1RAP on postsynaptic recruitment onto PTPδ-expressing HEK293T cells when sIL1RAP is co-expressed. Co-culture assay showed that PTPδ-expressing HEK293T cells induced postsynaptic differentiation, as measured by recruitment of the postsynaptic marker PSD95 (Figure 5C). Co-expression with sIL1RAP-HA in HEK293T cells also significantly affected PTPδ-induced recruitment of the postsynaptic marker, but did not influence NRX1β-induced recruitment. Synaptogenic activity was severely reduced when co-cultured with Il1rap knockdown cerebellar neurons (Figure S7F), confirming that PTPδ-induced post-synaptogenic activity of cerebellar neurons might be largely dependent on transsynaptic interaction with IL1RAP rather than the other partner (e.g., IL1RAP-L1) in cerebellar neurons. We then tested PTPδ-induced synaptogenic activity on a co-culture system combined with Sam68 KO neurons. We found that PTPδ-induced postsynaptic assembly in Sam68 KO neurons was significantly lower than in WT neurons, whereas NRX1β-induced assembly was comparable between WT and Sam68 KO neurons (Figure 5D). Therefore, these results suggest that proper ALE selection of Il1rap by SAM68 is required for synaptogenic signaling through transsynaptic IL1RAcP/IL1RAP-L1-PTPδ interaction.

Figure 5

Soluble IL1RAcP Disturbs PTPδ-Induced Synaptogenic Signaling and IL-1-Mediated NMDAR Function in the Nervous System

(A) Illustration of excitatory synapse organization through synaptic interaction of IL1RAP and the related-protein IL1RAP-L1 with PTPδ, and IL-1-induced potentiation of NMDAR-mediated calcium influx through interaction with IL-1 receptor (IL1R) in the CNS.

(B) Schematic illustration of neuron-HEK293T cell co-culture assay. To examine IL1RAP-mediated postsynaptic assembly, HEK293T cells expressing PTPδ or neurexin-1β (NRX1β)-HA were co-cultured with cerebellar neurons (DIV10-14).

(C and D) Soluble IL1RAP (sIL1RAP) disturbs PTPδ-induced synaptogenic signaling. Postsynaptic assembly on HEK293T cells was detected by immunostaining with postsynaptic marker, PSD-95. (C) HEK293T cells expressing PTPδ or NRX1β-HA with or without sIL1RAP-HA (ratio 1:1). (D) HEK293T cells expressing PTPδ or NRX1β were co-cultured with cerebellar granule neurons from WT or Sam68KO cerebella (n = 23–34 cells/each group in >10 separated fields [see the number on each graph column]) Scale bar, 5 μm.

(E and F) Calcium imaging with Fluo-4 AM in cultured hippocampal neurons. Soluble IL1RAP disturbs IL-1-induced potentiation of calcium influx mediated via NMDARs. Intracellular calcium levels were measured by Fluo-4 intensity. Quantification of intracellular calcium level at 1 min before NMDA stimulation (Pre) and at 0, 2, and 10 min after stimulation. (E) The traces (left) and quantification (right) of the relative intracellular calcium level in control neurons, Il1rap knockdown neurons, and sIL1RAP-HA-expressing neurons with lentiviral infection (control, n = 130 fields; Il1rap knockdown, n = 50 fields; sIL1RAP-HA expressing, n = 30 fields, in three independent experiments). (F) The traces (left) and quantification (right) of the relative intracellular calcium level in wild-type, Sam68KO, and Slm1KO neurons (wild-type, n = 50 fields; Sam68KO, n = 50 fields; Slm1KO expressing, n = 40 fields, in three independent experiments). Data are presented as the mean ± SEM. Significance is indicated as follows: **p < 0.01; *p < 0.05. Student's t test in (C and D); one-way ANOVA followed by Dunnett's test in (E and F).

Soluble IL1RAP Disturbs IL-1-Induced Ca2+ Influx Mediated through NMDAR Activation

Reportedly, interleukin (IL)-1 mediates not only inflammatory activity in pathological conditions but also long-term potentiation and memory formation in physiological situations by interaction with the IL-1 receptor (IL1R1) (Yirmiya et al., 2002). Such effects of IL-1 are mediated by IL1RAP. IL1RAP governs IL-1β-mediated N-methyl-D-aspartic acid (NMDA) receptor (NMDAR) activation through NR2A phosphorylation by Src family kinases in the hippocampal neurons (Figure 5A) (Huang et al., 2011). Actually, we confirmed that IL-1 and IL-1R1 transcripts were expressed in cortical and hippocampal regions (Figures S7A and S7B). Therefore, to test whether aberrant usage of Il1rap ALE could influence NMDAR-dependent plasticity, we examined the effect of sIL1RAP on NMDAR-dependent Ca2+ influx mediated through IL-1 signaling in the cultured hippocampal neurons. As NMDA (20 μM)-induced Ca2+ influx is potentiated at a low concentration of IL-1β (0.01 ng/mL) (Huang et al., 2011), we performed intracellular Ca2+ imaging using Fluo-4 AM in cultured hippocampal neurons under similar experimental conditions, as previously reported (Huang et al., 2011). We observed elevation of Ca2+ level, as measured by fluorescence of Fluo-4 in control neurons for a few minutes after NMDA application in the presence of 0.01 ng/mL of IL-1β (Figure 5E). Consistent with the previous report, Ca2+ elevation was significantly reduced in Il1rap knockdown neurons. In line with the knockdown effect, the Ca2+ elevation was significantly lower in sIL1RAP-HA-expressing neurons (Figure 5E). We confirmed the neuronal secretion of sIL1RAP-HA from cultured hippocampal neurons (Figure S7G). We further tested the NMDAR-dependent plasticity in Sam68 KO hippocampal neurons. Notably, Ca2+ elevation was significantly lower in Sam68 KO neurons, but not in Slm1 KO ones (Figure 5F). These results indicate that sIL1RAP impairs IL-1-induced Ca2+ influx mediated through NMDAR activation by antagonizing neuronal IL-1 signaling.

SAM68 Directly Binds to the Cryptic Polyadenylation Signal Sequence on Intron 8 of Il1rap

To address the molecular mechanism by which SAM68 targets the significant number of ALEs, we attempted to identify the recognition element of SAM68 for ALE splicing. Reportedly, the canonical poly(A) signal (PAS) sequences (AAUAAA) are optimal binding sites for SAM68 (Feracci et al., 2016, Ray et al., 2013). A recent study suggested that SAM68 masks this intronic PAS to prevent premature termination of the transcript through aberrant alternative polyadenylation (La Rosa et al., 2016). Therefore, to identify the cryptic PAS at the intronic sequence of Il1rap, we performed 3′ rapid amplification of cDNA ends analysis from Sam68 KO brains and detected two major transcripts of Il1rap exon 8b (Figure S8A, arrows). The sequence analyses confirmed that the two transcripts were the full-length of exon 8b (exon 8b LF) and shorter ones (exon 8b SF) (Figures 6A and S8B). Actually, we found that the 3′ UTR of exon 8b contains two putative PAS sites (PAS1 and PAS2) (Figure 6A, blue boxes). Here, RNA-seq showed that most of the transcript reads were terminated around PAS1 in Sam68/Slm1 DKO brains (Figure 6B). Although an RT-qPCR study detected the transcripts of exon 8b LF in Sam68 KO brains by using LF-unique primer set (primer 3) when compared with those of WT, it appeared that the amount was very small (only 2-fold higher compared with that of WT) (Figure S8C). These data indicate that the major transcripts in Sam68 KO brains are exon 8b SF. Indeed, the sequences of PAS1 were completely conserved between humans and mice (Figure S8D). Therefore PAS1 is possibly the most actionable in the absence of SAM68.

Figure 6

SAM68 Directly Binds to Cryptic PAS in the Intron 8 of Il1rap

(A) The full-length cDNA sequence of Il1rap exon 8 (left), and the schematic illustration of the two major transcripts in Sam68KO brains (exon 8b SF and exon 8b LF) (right). Green indicates the coding exon region, blue shows putative PAS sites on the 3′ UTR.

(B) RNA-seq on Il1rap exon 8b in wild-type and Sam68/Slm1DKO brains. Arrowheads represent two putative PAS sites (PAS1 and PAS2).

(C–F) Mapping of SAM68 recognition elements in Il1rap. (C and D) UV cross-linked RNA immunoprecipitation (CLIP) assay. (C) Positions of three primer sets used for the assay. (D) The representative gel loading images of the CLIP assay using anti-SAM68 antibody and the quantification by RT-qPCR analysis (n = 3 brains). (E and F) Biotinylated RNA pull-down experiments. (E) Biotinylated RNA oligonucleotide probes covering the 3′ UTR sequence of Il1rap exon 8b used in pull-down experiments. The PAS a/c mut probe contains two nucleotide changes (red). (F) The pull-down experiments with mouse adult brain extracts. Bound proteins were detected by western blot analysis with anti-SAM68 and anti-Rbfox1 antibodies. SAM68 binding was quantified by densitometric scanning of western blot signals (n = 5).

Data are presented as the mean ± SEM. Significance is indicated as follows: ***p < 0.001; **p < 0.01; *p < 0.05. Student's t test.

Expectedly, UV cross-linked RNA immunoprecipitation with SAM68 antibody in WT brains showed the assembly of SAM68 near PAS1, whereas binding in other regions was much weaker (Figures 6C and 6D). To further test the direct binding to PAS1, we examined the binding of SAM68 to synthetic RNA oligonucleotides spanning 30 bases of the PAS1 region (Figure 6E). The Il1rap 8b UTR WT probe (PAS1 WT) yielded efficient binding of endogenous SAM68 from brain extracts in the pull-down assays (Figure 6F). Furthermore, mutation of two nucleotides in a presumptive SAM68-binding PAS site (PAS1 a/c mut) significantly reduced the recovery of SAM68 (Figures 6E and 6F), demonstrating that endogenous SAM68 can directly recognize the PAS1 sequence. Under the same conditions, the other RBP, Rbfox1, was not recovered in the precipitates. Therefore, these data suggest that SAM68 regulates ALE selection through direct binding to the cryptic PAS in intron 8 of WT brains.

Tissue-Specific SAM68 Expression Determines ALE Selection in Spatial Fashion

Given that a significant number of SAM68-targeted transcripts could be expressed in tissues other than the brain, it would be of interest to explore how SAM68-dependent ALE selection is controlled in those other tissues. Therefore, we examined the expression profiles of Il1rap, Cp, Pcdh15, and Lrrcc1 in various tissues. We observed that transcripts of Il1rap, Cp, and Lrrcc1 were detected ubiquitously (Figure S9A). On the other hand, expression of SAM68 exhibited a tissue-specific pattern (Figure S9B). In particular, SAM68 expression appears to be very subtle in the liver. RT-qPCR also showed low expression of not only Sam68 but also Slm1 in the liver at the transcript level (Figure S9C). We then examined the ratio of Il1rap and Cp splicing isoforms (LF versus SF) in several tissues. The ratio was highly variable among tissues (Figure 7A). In contrast to the brain, both major transcripts were the SF variant in the liver, in which SAM68 expression is very low. Indeed, we found that the amount of SAM68 is inversely correlated with the abundance of the SF variant (Il1rap, R2 = 0.85, p = 0.008; Cp, R2 = 0.86, p = 0.04, Figure 7B). We also observed that the amount of the Il1rap SF variant was significantly increased in the Sam68 KO lung and brain compared with the WTs (Figure 7C), whereas ectopic expression of SAM68 in primary liver cell culture significantly reduced the SF variant (Figure 7D). These results showed that ALE selection of these SAM68 targets is highly dependent on the expression dose of SAM68.

Figure 7

The Distinct Amount of SAM68 Is Responsible for Proper 3′ UTR Isoform Selection of Il1rap and Cp in the Nervous and Non-nervous Systems

(A) Abundance ratio of SF to LF in the brain, lung, intestine, spleen, and liver of WT mice. For the abundance ratio of SF to LF, the percentage of the SF variant was largely estimated from the CT value (CT) directly compared with that of LF (CT) at the same threshold set for the CT value. RQ + RQ values for total Il1rap were set to 100%. RQ value of two transcripts was normalized to that of Gapdh (n = 3–4 animals per group).

(B) Reciprocal correlation between SAM68 level and production of Il1rap and Cp SFs. SAM68 was quantified by western blot analysis. The value for the cerebellum was set to 1.0. Correlation coefficients between SAM68 and the SF transcript of Il1rap and Cp (right) were determined in the scatterplot analysis. The gray lines in the scatterplot are the 95% confidence limit of the best fit line.

(C) Quantification of the SF variant of Il1rap between brain and non-neuronal tissues (lung, intestine, liver, and spleen) from WT and Sam68KO mice by RT-qPCR. The RQ value of total transcripts was normalized to that of total Il1rap (n = 3 animals per genotype). RQ values for wild-type brain (Cb) were set to 1.0.

(D) Quantification of the SF variant of Il1rap between the primary liver cell cultures and the ones in which SAM68 was ectopically expressed with lentiviral infection. The RQ value of total transcripts was normalized to that of Gapdh. The RQ value of SF transcripts was normalized to that of total Il1rap (n = 3 cultures per group).

(E) Representative images of western blot analysis with the α-SAM68 antibody. Human hepatoma cell line, HepG2, cells aberrantly express SAM68 at high level.

(F) Low production of the Il1rap SF variant in HepG2 cells. The abundance ratio of SF to LF was compared between the normal mouse liver and HepG2 cells (n = 3–4 cultures per group).

(G) Restoration of aberrant ALE selection in HepG2 cells by knockdown of SAM68. Knockdown of aberrantly expressing SAM68 partially but significantly increased the level of the Il1rap SF isoform in HepG2 cells The RQ value of SF transcripts was normalized to that of total Il1rap (n = 3–4 cultures per group).

(H and I) Model of tissue-specific isoform selection of Il1rap and Cp through usage of ALEs between the brain and liver by physiologically expressed SAM68. (H) The neurons strongly express SAM68, so that they dominantly produces the membrane forms. In contrast to the brain, secreted forms lacking transmembrane domain or glycosylphosphatidylinositol anchor are abundantly produced in the liver. (I) SAM68-specific ALE selection is required for the organization of IL1RAP-dependent excitatory synapses through transsynaptic interaction with PTPδ in the nervous system. On the other hand, absence of SAM68 causes the release of IL1RAP into the plasma, which could be necessary for homeostatic control of IL1-mediated inflammation.

Data are presented as the mean ± SEM. Significance is indicated as follows: ***p≪0.001; *p < 0.05; Student's t test.

Furthermore, we observed that although SAM68 is not expressed in the normal mouse liver, it was strongly expressed in a human hepatocarcinoma cell line, i.e., HepG2 cells (Figure 7E). In association with the strong expression of SAM68, we found that the ratio of the Il1rap SF variant in HepG2 cells was markedly lower (<40%), compared with that in the normal mouse liver (Figure 7F). To verify whether the low amount of the Il1rap SF variant in HepG2 cells is due to the aberrant expression of SAM68 in carcinoma cells, we examined the knockdown effect of human SAM68 (hSAM68) on the ratio of Il1rap splicing variants in HepG2 cells. We found that knockdown of hSAM68 partially, but significantly, increased the Il1rap SF variant (Figure 7G). These results further suggest that SAM68 is a dominant regulator for ALE selection of Il1rap throughout the whole tissue. Thus, the absence of SAM68 causes a long-to-short isoform switch of the neuronal targets in non-neuronal tissues (Figure 7H), indicating that the SAM68 expression level is critical for the tissue-specific selection of alternative 3′ UTR isoforms through ALE choice. Indeed, whereas atypical sIL1RAP could impair PTPδ-mediated synapse organization in the nervous system (Figure 5), physiological sIL1RAP in plasma plays a homeostatic role in IL-1 signaling by antagonizing the interaction with IL-1R1 in the immune system (Jensen et al., 2000, Smeets et al., 2005). Therefore, SAM68-dependent ALE selection could be necessary to exert distinct functions of ubiquitously expressed molecules between the nervous and the non-nervous systems (Figure 7I).

Discussion

Distinct Alternative Splicing Activity between SAM68 and the Related Proteins SLMs

We showed that neuronal alternative splicing by STAR family proteins is an important mechanism for functional diversification. Here, we conducted transcriptomic analyses using Slm1 KO and Sam68/Slm1 DKO brains and showed a different splicing activity between SAM68 and SLM1. This study focused on the neuronal isoform selection in 3′ UTR by SAM68 and demonstrated their functional aspects through the identification of a novel target IL1RAP in neurons (Figure 5). Very recently, two articles also elucidated the interaction between SAM68 and U1 small nuclear ribonucleoprotein particle (snRNP) as a global mechanism underlying ALE regulation by SAM68 (Naro et al., 2019, Subramania et al., 2019), supporting our findings in the CNS. U1 snRNP prevents premature transcript termination by inhibition of cryptic PASs (Berg et al., 2012, Kaida et al., 2010). Therefore, our findings on ALE selection in CNS also might be largely explained by the interaction with U1 snRNP. However, given that the U1 binding-like sequences were not observed around cryptic PAS1 on Il1rap (see Figure S8), additional mechanism also could be possible. Considering the direct binding to cryptic PAS in WT brains (Figure 6), another possibility is that SAM68 may block the recruitment of such 3′ end machineries on the PAS as the CPSF and CstF to prevent the proximal termination of Il1rap pre-mRNA. ALE selection is related to alternative polyadenylation; such 3′ end formation factors have been shown to play a role in alternative splicing (Misra and Green, 2016).

In addition to the difference in splicing activities between SAM68 and SLM1, this study also suggested a difference between SLM1 and another family protein, SLM2. We newly mapped the entire SLM1-dependent program and revealed that a significant number of exons seemed to be altered in Slm1 KO brains (Figure 1 and Table S2), whereas SLM2 encodes a highly selective alternative splicing program that regulates only a few synaptic molecules (Traunmuller et al., 2016). Regardless of the high structural homology between SLM1 and SLM2 (Di Fruscio et al., 1999), the large functional difference between the two closely related proteins is very surprising. We previously showed that SAM68 can heteromerize with SLM1, but not with SLM2 (Iijima et al., 2014), which suggests that endogenous SLM2 ordinarily exists as a homodimer. Thus dimer formation is intrinsically different between SLM1 and SLM2. Increased RNA affinity through dimer formation is a critical parameter enabling SLM proteins to select their functional targets with the transcriptome (Feracci et al., 2016). Therefore, one possibility is that the structural difference in dimer interface between SLM1 and SLM2 complexes results in distinct splicing programs. However, numerous questions on the functional difference between STARs remains to be addressed in future studies.

Critical Role of Proper ALE Selection of Il1rap between the Nervous and Other Systems by Distinct SAM68 Expression Level

This study revealed that SAM68 is a dominant factor for ALE selection of Il1rap in the nervous system. mIL1RAP is necessary for organizing excitatory synapses through transsynaptic interaction with PTPδ in the CNS (Yoshida et al., 2012). This study demonstrated that, in addition to the significant reduction in mIL1RAP (Figure 4), the competitive effect of sIL1RAP could accelerate the impairment in PTPδ-mediated synapse organization (Figures 5A–5D). Reasonably, the competitive effect is supported by the X-ray structural analysis showing that the Ig domains of IL1RAP and PTPδ are the elements responsible for the heterophilic interaction (Yamagata et al., 2015). Both IL1RAP and PTPδ have several transsynaptic binding partners. PTPδ organizes synapses through interaction with IL1RAP-L1 and Slitrk3 (Takahashi et al., 2012, Yoshida et al., 2011). In addition to IL1RAP, because we demonstrated the competitive effect of sIL1RAP on presynapse assembly onto HEK293T cells expressing another paralog, IL1RAP-L1, on co-culture assays (Figure S7E), sIL1RAP may influence several related transsynaptic types of synaptogenic signaling in the CNS. We also revealed that sIL1RAP significantly affects IL-1β-induced NMDAR activation in hippocampal neurons (Figures 5E and 5F). Thus, this study suggests that proper ALE usage by SAM68-specific splicing is critical for both aspects of synaptic organization and plasticity in the CNS.

In contrast to the brain, the liver is thought to be a major source of sIL1RAP, which is suggested to play an important role in the homeostasis of IL-1 signaling by antagonizing the interaction of IL1RAP with IL-1R1 in the immune system (Jensen et al., 2000, Smeets et al., 2005). The reduced level of physiological sIL1RAP in the plasma is in fact implicated in several diseases (Bozaoglu et al., 2014, Michaud et al., 2014). Thus, tissue-specific SAM68 expression could play a critical role in distinct functions of ubiquitously expressed proteins between the nervous and non-nervous systems through ALE selection (Figures 7H and 7I).

Regulatory Functions of the SAM68 Splicing Program Dedicated to Alternative 3′ UTR Isoform Diversity

Thousands of mammalian genes encode alternatively spliced isoforms in their 3′ UTR (Miura et al., 2013, Tian et al., 2005). Here, we demonstrated that SAM68 is required for the spatial control of alternative 3′ UTR isoforms between the nervous and the other systems by identification of new SAM68 targets (Figure 7). Importantly, GO analyses implied that SAM68 targets the 3′ UTR exons of multiple transcripts that encode neuronal membrane or secreted proteins (Figures 1E and 2D). The biochemical studies indeed found drastic shift to short isoforms by aberrant ALE selection in Sam68 KO brains, which could result in membrane-to-secreted isoform conversion at the protein level. Thus the findings strongly suggest that SAM68 is a key regulator for shaping the diversity of neuronal 3′ UTR isoforms in the nervous system.

The other intriguing point regarding alternative 3′ UTR selection is the molecular control at the transcript level. This study also found that SAM68 regulates not only ALE but also APA (Figures 2 and 3C), which alters the length of the 3′ UTR itself. Such alternative 3′ UTR diversity by APA and ALE contributes to the posttranscriptional processes such as translation, mRNA stability, and subcellular localization during development (Di Giammartino et al., 2011, Taliaferro et al., 2016) and dendritic localization and the local translation in the nervous system (Tushev et al., 2018). Therefore, it would be of interest to explore how the ALE/APA-mediated mechanism by SAM68 contributes to molecular functions at the transcript level in a future study. Overall, although the mechanism by which a specific subset of 3′ UTR exons is controlled by SAM68-specific splicing should be examined, our findings could provide a general principle underlying the control of alternatively spliced 3′ UTR isoforms.

Limitations of Study

In this study, we performed transcriptomic analysis using SAM68 knockout and SAM68/SLM1 double-knockout midbrains and revealed a different alternative splicing activity between SAM68 and SLM1; we characterized alternative 3′ UTR selection by SAM68-specific splicing in the nervous system. However, the open questions on the mechanism underlying the differential splicing activity between SAM68 and the related family proteins remains to be addressed in future studies.

Our findings extend the understanding on the neuronal function of SAM68, in particular through the identification of IL1RAP as a new SAM68 target. However, the physiological consequences were mainly obtained by neuronal culture system. Further studies are needed to confirm the functional relevance in vivo.

Methods

All methods can be found in the accompanying Transparent Methods supplemental file.