We previously reported that the Arabidopsis PPR protein SOT5 (At1g30610), also named EMB2279, is required for intron splicing of the two chloroplast housekeeping genes, rpl2 and trnk (Huang et al., 2018). The sot5 (emb2279-3) mutant contains a G to A mutation at the first base (+1) of the seventh intron and displays a leaf virescent phenotype ( Figures 1A, C ). This point mutation leads to accumulation of the three alternative splicing variants in sot5, compared to the wild type (WT) ( Figure 1B ). DNA sequencing of these cloned PCR products indicated that the longest one retained the whole seventh intron with 84 nts ( Figure 1B and Figure S1 ), the middle transcript retained 9 nts more at the 3’ ss of exon 6 ( Figure 1B and Figure S1 ), and the shortest one lacked 22 nts at the 3’ end of exon 6 ( Figure 1B and Figure S1 ). These results indicate that the mutation of the first invariant G at the 5’ ss of intron 7 results in intron retention and activation of two cryptic splice sites named -22 ss and +10 ss ( Figure 1C ). Since the level of band d was much higher than that of band c, the cryptic splicing mainly occurred at -22 ss in sot5 ( Figure 1B ). Polypeptide prediction indicated that the -22 ss product encodes a truncated protein with only 6 PPR motifs that should be loss of function ( Figures 1D , E ); the intron-retained transcript encodes a probably partially functional protein with 10 PPR motifs (Chen et al., 2020); the +10 ss product encodes almost the same protein as the WT one except the seventh PPR motif is slightly changed ( Figures 1D , E , Figure S2 ). Thus, these results interpret well why sot5 displays a virescent phenotype, but the emb2279-1 knockout mutant is embryo lethal.

The weak allele sot5, in which an intron mutation leads to generation of three abnormal transcripts, provides a good material to study regulation of the cryptic splice site usage. To dissect the genetic pathway for the cryptic splice site selection of SOT5 intron 7, we screened for sot5 suppressors which exhibit a WT-like phenotype using the ethyl methanesulfonate (EMS)-mutagenized sot5 seeds. We obtained a number of WT-like suppressor lines for the sot5 mutant ( Figure S3 ). In this study, four suppressor lines, namely F23, F11, E1 and E25, were selected for further analysis based on their similar phenotype and alternative splicing pattern of SOT5 intron7 ( Figures 1A–C and Figure S3 ). Compared to sot5, all these suppressor lines produced a lower level of -22 ss transcripts (band d in Figure 1B ), but a higher level of +10 ss transcripts (band c in Figure 1B ), suggesting a positive correlation between the green leaf and the level of the +10 ss transcripts. In sot5 mutant, the splicing efficiency of plastid rpl2 and trnK intron is significantly decreased (Huang et al., 2018). Indeed, the suppressors of sot5 were able to recover the splicing defect of plastid rpl2 intron, which is consistent with their WT-like phenotype ( Figure 1B ). In addition, our genetic analyses showed that the suppressor phenotype was caused by a single gene mutation, and interestingly F23, F11 and E1 were allelic while E25 was mutated in another gene. Taken together, our data suggest that the sot5 phenotype is suppressed by second mutations that can modulate cryptic splice site selection and splicing efficiency.

To isolate the suppressor gene in F23, we produced a mapping population from the cross between F23 and Landsberg erecta (Ler). Through Mapping-By-Sequencing (MBS) technique, the gene responsible for phenotypic suppression of sot5 was mapped to the region with the length ~8 M bp on chromosome 5, which contains 7 point mutations ( Figures S4A, B ). Among these 7 mutations, one stop codon-gain mutation from C to T occurred at 187 bp of the CDS in the locus At5g19350, which encodes RBP45d, a member of the RBP45/47 family with triple RNA recognition motifs (RRM) ( Figures 2A and S4B ). Considering that sot5-activated cryptic splicing efficiency is altered in F23, we speculated that RBP45d was the most likely gene to suppress sot5. Thus, we sequenced the RBP45d gene in F11 and E1, which were genetically testified as F23 alleles. Indeed, F11 has a point mutation from G to A at the 481 bp position from the start codon of the RBP45d CDS, resulting in the substitution of glycine (G) by arginine (R) at the 161 amino acid residue in the conserved RRM2 domain, whereas E1 contains a premature stop codon mutation from G to A at the 80 bp position from the start codon of the RBP45d CDS ( Figure 2A ).

To verify the mapping result, we overexpressed the native RBP45d CDS in F23. Our data showed that transgenic lines 35S:RBP45d/F23 not only displayed the sot5 phenotype ( Figure 2B ), but also had the same alternative splicing pattern as the sot5 mutant ( Figures 2C, D ). In addition, we generated a knockout allele of RBP45d (rbp45d-CR, with one base insertion at the first exon) in sot5 mutant background by CRISPR/CAS9 technique, which presented WT-like phenotype ( Figures 2A, B ). These results indicated that the phenotypic recovery of sot5 is attributed to the loss-of-function mutation in RBP45d. We then isolated the single mutant, named rbp45d-4, from the F₂ population of the cross between F23 and Col-0 and rbp45d-5 from the F₂ population of the cross between F11 and Col-0. RT-PCR analysis showed that RBP45d transcripts were almost undetectable in rbp45d-4 and F23 ( Figure 2C ), indicating that the point mutation in rbp45d-4 triggers the nonsense-mediated mRNA decay (NMD) pathway. Taken together, our results imply that RBP45d mutations are responsible for the recovery of sot5 leaf virescence.

It has been reported that the Arabidopsis genome has eight members in the RBP45/47 family (Wang and Brendel, 2004). RT-PCR analysis and eFP Browser (http://bar.utoronto.ca/efp_arabidopsis/cgi-bin/efpWeb.cgi) showed that they are ubiquitously expressed in all organs including roots, inflorescences and leaves ( Figure S5 ). Subcellular localization experiment showed that RBP45a, RBP45b, RBP45c, RBP45d and RBP47b were localized in the both nucleus and cytoplasm, while RBP47c’ mainly localized in the cytoplasm ( Figure S6 ). The mutant phenotype and physiological function of RBP45/47 family members were not well characterized yet. To know whether other members of the RBP45/47 family can function in the same manner as RBP45d to suppress the sot5 phenotype, we identified four T-DNA insertion mutants rbp45a, rbp45c, rbp47a and rbp47b from the ABRC stock ( Figures 3A, B ), and made their double mutants with sot5 by genetic crossing. All the four single mutants displayed the WT-like phenotype under normal growth conditions, while the double mutants with sot5 showed the sot5-like phenotype ( Figure 3C ). Furtherly, RT-PCR analysis showed that the SOT5 intron7 splicing pattern was not altered in these mutants when compared with that in sot5 and F23 ( Figure 3D ). Therefore, our data indicate that rbp45a, rbp45c, rbp47a and rbp47b cannot suppress the phenotype of sot5.

Recently, Chang et al. (2022) and Wang et al. (2021) reported that RBP45d regulates temperature-responsive flowering in Arabidopsis. In this study, we also observed the later flowering phenotype of rbp45d mutants including rbp45d-4, rbp45d-5 and rbp45d-CR (a knockout rbp45d mutant created by CRISPR/CAS9 technique) under 16h light/8h dark growth conditions ( Figure 4A upper panel and Figure S7 ). The number of rosette leaves for rbp45d mutants to flower were significantly bigger than that for WT ( Figure 4B the left panel and Figures S7B and S7C ). However, there were no significant difference in flowering time between WT and other rbp45/47 plants ( Figures 4A, B ). In addition, we found that loss-of-function mutations in RBP45d but not in other RBP45/47 genes led to the significant shorter primary root phenotype, compared with that of WT, when seedlings were vertically grown on 1/2 MS plates ( Figure 4A lower panel, Figure S7 ). For 5-day-old seedlings, primary root length of rbp45d-4 and rbp45d-5 was 64% and 80% of WT, respectively ( Figure 4A lower panel and Figure 4B ). Such a difference in primary root length became much larger between 10-day-old rbp45d and WT or other rbp45/47 seedlings. These results suggest that the biological function of RBP45d is different from that of other RBP45/47 paralogs.

RBP45/47 proteins are homologous with yeast Nam8p protein and human TIA-1 protein. It was reported that TIA-1 can directly interact with U1C, the core protein of U1 snRNP (Forch et al., 2002). And the cryo-electron microscopy structure of the yeast Saccharomyces cerevisiae prespliceosome revealed that yeast Nam8 protein physically interacts with the U1C. Thus, we proposed that RBP45d may function in a similar manner with TIA-1 and Nam8. We then used yeast two hybrid (Y2H) and BiFC techniques to detect whether RBP45d interacts directly with U1C, U1 70K and U1A of U1 snRNP in Arabidopsis. Y2H results showed that RBP45d directly interacted with U1C, but not with U1A and U1 70K ( Figure 5A ). Consistently, BiFC analysis also revealed that RBP45d was bound to U1C in both the nucleus and cytoplasm in Arabidopsis protoplasts ( Figure 5B ), which is consistent with their subcellular localization ( Figure S6 ). Taken together, these results indicate that RBP45d interacts with U1C in vitro and in vivo.

Meanwhile, Y2H analysis showed that except for RBP47C, the RBP45/47 members interacted with U1C ( Figure 5A ), whereas BiFC assays indicated that all the tested RBP45/47 paralogs interacted with U1C mainly in the cytoplasm ( Figure 5 ). Thus, our results indicate that the biological significance of U1C interaction with RBP45d may differs from that with other RBP45/47 members.

The above results suggest that RBP45d is recruited to the U1 snRNP via interacting with U1C, which is consistent with the working model of yeast Nam8 and human TIA1. It indicates the molecular mechanism of RBP45d regulating 5’ss selection is also conserved in plants. Then we asked whether RBP45d can interact with U-rich sequences. In the sot5 mutant, the G to A mutation at the 5’ ss of intron 7 results in activation of the two cryptic splice sites (-22 ss and +10 ss) and preferential utilization of the -22 ss ( Figures 1B, C ). In contrast, the absence of RBP45d in the suppressor lines F23, F11 and E1 leads to preferentially utilize +10 ss ( Figure 1B ), implying that RBP45d plays an important role in controlling the the cryptic splice sites usage. Search for U-rich elements within sot5 intron 7 indeed revealed the presence of the AUAAAUUUU sequence downstream of the -22 ss ( Figure 6A ). We proposed that RBP45d can bind to this U-rich sequence and subsequently promote the upstream -22 ss selection.

To test this hypothesis, we carried out RNA electrophoretic mobility shift assay (REMSA) with a series of synthesized 26 nt RNA oligo probes including the U-rich element (SOT5-IN7) and the mutated variants (U-poor or U-less, SOT5-IN7-M1~4), and purified His-tagged RBP45d protein from E. coli ( Figures 6B, C ). REMSA results showed that RBP45d bound to the Cy5-SOT5-IN7 probe in a dose-dependent manner, and the unlabeled SOT5-IN7-NL probe was able to well compete with the labeled probe in RBP45d binding ( Figure 6D ). Further analysis showed the unlabeled SOT5-IN7-M1-NL and SOT-IN7-M2-NL probes, in which the first A and AU of intron 7 were mutated into G and GC, respectively, still efficiently blocked RBP45d binding to the labeled Cy5-SOT5-IN7 probe ( Figures 6E, F ), whereas the unlabeled SOT5-IN7-M3-NL probe with mutations at the first AU and 4 continuous U of intron 7 significantly reduced the binding affinity to RBP45d ( Figure 6E ). Meanwhile, the unlabeled SOT5-IN7-M4-NL probe completely lost the binding affinity to RBP45d when all the AU-rich element was mutated into the CG-rich one ( Figure 6F ). Quantification of the signal intensity of these protein-RNA bands revealed that the affinity of RBP45d binding to RNA increased with the increase in U and/or A levels of the probes ( Figure 6G ). Thus, our results suggest that RBP45d can bind to the U-rich sequence element, and binding affinity between RBP45d and RNA is dependent on the content of U and/or A.

Previously, we reported the G to A mutation at the 5’ ss (+1G-to-A) of ClpR4 intron 2 in the clpR4-3 mutant, which displays virescent leaves (Wu et al., 2013). Similarly, the mutation of the authentic splice site leads to retention of intron 2 and activation of the two cryptic splice sites (+16 ss and +49 ss) ( Figure 7A ), producing three different transcripts, named band a, b, and c that are not detected in WT ( Figure 7C ). To examine whether RBP45d is involved in regulation of alternative splicing of ClpR4 intron 2 in clpR4-3, we constructed the rbp45d-4 clpR4-3 double mutant by genetic crossing. Although rbp45d-4 was unable to suppress the virescent phenotype of clpR4-3 ( Figure 7B ), the splice sites usage and splicing efficiency of ClpR4 mRNA were clearly altered in the double mutant, where band b products significantly decreased while band d products appeared and band c products increased slightly, compared with those in the single clpR4-3 mutant ( Figures 7C, D ). Sequencing of the cloned PCR products showed that band b, c and d products were derived from the cryptic splice sites +49 ss, +16 ss, and -96 ss, respectively ( Figures 7A, C and Figure S8 ). These results suggest that the absence of RBP45d in clpR4-3 allows the U1 snRNP complex to preferentially recognize the cryptic splice sites -96 ss and +16 ss, instead of the cryptic splice site +49 ss. We found a five consecutive U (U-rich) element downstream of the cryptic splice site +49 ss (the underlined letters in Figure 7A ). It is possible that RBP45d binds to this U-rich element in ClpR4 intron 2, and subsequently promotes the cryptic +49 ss selection in clpR4-3.

To clone the suppressor gene in the E25 line, which displays a similar phenotype to F23, F11 and E1 but was not allelic with them, we used the backcrossed F₂ population between E25 and sot5 to clone the suppressor gene via the MutMap approach. Our data showed that the suppressor gene was located in a region (~1.3 M) containing 15 candidate genes on Chromosome 1 ( Figure S9 ). Interestingly, we found a G to A mutation at the 5’ ss of the eleventh intron in the PRP39a gene, which encodes a U1 snRNP auxiliary protein PRP39a ( Figure 8A ). To verify the result, we identified a T-DNA insertion mutant prp39a (prp39a-1, salk_133733) from the ABRC stock, and made the sot5 prp39a-1 double mutant by genetic crossing. The double mutant exhibited the same phenotype as prp39a-1 ( Figure 8B ). Likewise, PCR analysis showed that the pattern of splicing products in sot5 prp39a was similar to that in E25 ( Figures 8C, D ). In addition, we overexpressed the PRP39a genomic DNA in E25, and found that the phenotype and the splicing pattern of the transgenic lines were the same as those of sot5 ( Figures 8C, D ). Taken together, our results suggest that PRP39a mutations contribute to the suppression of sot5 in E25.

The yeast Prp39 and it paralog Prp42 form the scaffold connecting auxiliary proteins associated with U1 snRNP to the core (Li et al., 2017). RT-PCR analysis and eFP browser showed that Arabidopsis PRP39a is ubiquitously expressed ( Figure S10A ). However, PRP39a protein specifically localized in nucleus, which is different from that of RBP45d ( Figure S10B ). Our genetic screening of sot5 suppressors indicated that RBP45d and PRP39a act in the same genetic pathway with regard to SOT5 intron 7 alternative splicing. In addition, the prp39a-1 mutant exhibited the same shorter primary root phenotype as rbp45d, compared with that of WT ( Figure S11 ). We then examined whether PRP39a and RBP45d physically interact each other in vitro and in vivo. Y2H results showed that PRP39a directly interacted with RBP45d, but not with other RBP45/47 members ( Figure 9A ). Consistently, BiFC analysis showed that PRP39a and RBP45d interacted specifically in the nucleus ( Figure 9B ), which is consistent with the nucleus localization of PRP39a ( Figure S10 ). However, we did not detect the fluorescence signal between PRP39a and RBP47b ( Figure 9B ), indicating that they are not associated each other in vivo. These results are consistent with previously reported that the role of RBP45d in alternative splicing relies on its specific interaction with PRP39a (Chang et al., 2022). Meanwhile both Y2H and BiFC results showed that PRP39a was able to interact with U1C in the nucleus ( Figure 9B ). Taken together, our results suggest that Arabidopsis RBP45d, PRP39a and U1C function in a similar way to that in yeast with regard to alternative splicing regulation.

It is well known that U1 snRNP plays a critical role in initiation of 5’ ss recognition by base pairing of U1 snRNA and 5’ ss consensus sequence. And several U1 snRNP auxiliary proteins such as Nam8, Luc7 and RBM25 can also influence 5’ ss selection, especially when the 5’ss sequence are not conserved (Fortes et al., 1999; Puig et al., 1999; Forch et al., 2002; Puig et al., 2007; Zhou et al., 2008). Based on the data reported previously, the Arabidopsis U1 snRNP component RBP45d is required for alternative splicing for a set of genes (Kanno et al., 2020; Chang et al., 2022). Consistently, the two RBP45d homologs, yeast Nam8 and human TIA-1, are implicated in alternative splicing via binding to the U-rich region downstream of a weak 5’ ss (Puig et al., 1999; Le Guiner et al., 2001; Aznarez et al., 2008; Qiu et al., 2011). The similar molecular function of Nam8, TIA-1, and RBP45d is attributed to their highly conserved domains including the three RRM domains and the C-terminal Gln-rich tail. Cryo electron microscopy (cryo-EM) analysis of prespliceosome structure revealed that Nam8 or TIA-1 binds to U-rich sequences through the RRM2 domain to define the use of weak 5’ ss, and interacts with the U1C-terminal region near the U1-5’ ss helix via the RRM3 domain (Plaschka et al., 2018). Recently, Cryo-EM structure of the yeast U1 snRNP showed that the alternative splicing factors Prp39 and Prp42 form heterodimer that acts as a central scaffold connecting all auxiliary proteins associated with U1 snRNP to the core (Li et al., 2017). In human, there is no Prp42 homolog, but the Prp39 homolog PrpF39 forms a homodimer that acts as the Prp39/Prp42 heterodimer in U1 snRNP. In addition, Nam8/TIA-1 interacts with the Prp39/Prp42 heterodimer by its RRM3 domain and C-terminal tail (Plaschka et al., 2018). The structural and biochemical data provide insight into alternative splicing mediated by Nam8/TIA-1 and Prp39/prp42. Although the U1 snRNP complex and structure remain to be dissected in plants, most of the counterparts of yeast U1 snRNP core subunits and auxiliary proteins are present in Arabidopsis and encoded by multiple genes (Wang and Brendel, 2004). Considering that the basic mechanism of pre-mRNA splicing is highly conserved in all eukaryotic organisms, we propose that RBP45d functions as a U1 snRNP auxiliary factor in the same way as Nam8/TIA-1. As expectedly, our data showed that RBP45d can not only bind to U1C (the core protein of U1 snRNP) and PRP39a (the scaffold protein connecting the auxiliary proteins to the core), but also to the downstream U-rich RNA sequence near 5’ ss of SOT5 intron 7. Accordingly, we suggest a working model that RBP45d can recruit spliceosome to specific 5’ ss and enhance the stability of spliceosome and pre-mRNA complex by directly binding to U-rich RNA sequences downstream of 5’ ss ( Figure 10 ).

Cryptic splice sites are repressed by complicated mechanisms in eukaryotes. Study of many model organisms revealed that when the authentic splice site is mutated, the cryptic splice sites close to the authentic splice site will be activated (Roca et al., 2003). Usually, there are multiple cryptic splice sites nearby. The mechanism of which cryptic splice site is preferentially selected is not very clear. It is reported that according to the conservation degree of the sequence near 5’ ss, the strong or weak 5’ ss can be predicted. The more conservative the motif is (i.e. it can be perfectly base paired with U1 snRNA), the stronger the splice site is. And the strong splice site will be preferentially and efficiently spliced. On the contrary, the less conservative the motif, the weaker the splice site is (Roca et al., 2003; Dawes et al., 2022). However, when there are AU-rich or U-rich sequences in the intron, this prediction is not accurate, because the U-rich sequence downstream of the weak splice site will be bound by specific splicing factors, such as Nam8p or TIA-1, and the weak splice site will be activated in this scenario. It is reported the AU-rich sequences in plant introns are more abundant and the sequence of 5’ ss is more variable. Therefore, the selection of plant splice sites may be more dependent on the location and abundance of AU- rich or U-rich sequence of introns (McCullough et al., 1993). But what trans-acting factor that acting on the AU- rich or U-rich sequence in plants remains unclear. Our study showed that RBP45d is a such splicing factor for U-rich introns splicing in plants. When the 5’ ss of SOT5 intron7 is mutated, two cryptic splice sites at position - 22 and + 10 are activated, but the + 10 ss (splice site) is much weaker than the - 22 ss. RBP45d or PRP39a mutation reverse the usage of two cryptic splice sites ( Figure 10 ). By carefully analyzing the sequences downstream of these two cryptic splicing sites, we found that there is a U- rich sequence that can be bound by RBP45d downstream of - 22 ss. In addition, we demonstrated that RBP45d physically interacted with PRP39a, consistent with the results that they act on the same genetic pathway.

CryoEM revealed that yeast U1 snRNP contains seven additional U1 auxiliary proteins including Prp39, Prp40, Snu71, Snu56, Luc7, Prp42, and Nam8 (Li et al., 2017). Yeast Luc7, a Zinic-finger RNA binding protein, is also involved in 5 ‘ss recognition (Fortes et al., 1999). It is suggested that interaction of yeast Luc7 with the upstream exon stabilizes the pre-mRNA–U1 snRNP interaction. Similarly, human hLuc7A affects 5’ss selection possibly via the formation of a network of protein-RNA interactions that stabilize pre-mRNA–U1 snRNP interaction and thus formation of the commitment complex. (Puig et al., 2007). There are three LUC7 paralogs (LUC7A, LUC7B and LUC7RL) in Arabidopsis genome and they specifically promote splicing of a subset of terminal introns and affects plant development and adaptation to stress (de Francisco Amorim et al., 2018). Whether the LUC7 family members acted at the same pathway with RBP45d and PRP39a remains unclear. Arabidopsis RBM25 (homolog of yeast Snu71) was aslo an U1 auxiliary protein, but it acts differently with RBP45d to regulate alternative splicing (Kanno et al., 2017). Human RBM25 was reported to specifically bind to Bcl-x RNA through a CGGGCA sequence located within exon 2 and enhance Bcl-x_s 5’ ss selection and thus regulate apoptotic cell death (Zhou et al., 2008). In addition, human RBM25 was found to associate with a U1 snRNP associated factor, hLuc7A by Co-IP (Zhou et al., 2008). These results indicate that the different U1 snRNP auxiliary protein may bind different cis-element in exons or introns to stabilize U1 snRNP and pre-mRNA complex and differentially regulate the alternative splicing of the pre-mRNA. The relationship among RBP45d, RBM25, and LUC7 family members remains to be elucidated.

It is noted that in Arabidopsis, although some of the U1 snRNP auxiliary proteins are encoded by two or more genes, they generally do not function redundantly in pre-mRNA splicing. RBP45d is a good example. Besides, Kanno et al., 2017 reported that Arabidopsis PRP39a mutation changes the alternative splicing of GFP reporter gene, while PRP39b cannot, indicating that they have different functions in splicing regulation. Probably, LUC7A, LUC7B and LUC7RL may also preserved their unique function in regulating pre-mRNA splicing. How these Arabidopsis U1 snRNP auxiliary proteins regulate 5’ss selection remains to be explored.

There are eight paralogs in the Arabidopsis RBP45/47 family. Whether they function redundantly or not remains unclear. Here we showed that RBP45d functions differently with other RBP45/47 members. The specificity of RBP45d in pre-mRNA splicing is supported by the following evidence. The sot5 phenotype can be rescued by mutations in RBP45d but not in other RBP45/47 members. Only RBP45d is interacted with PRP39a, since other RBP45/47 members lack the long C-terminal tail that was demonstrated to play an important role in protein-protein interaction between RBP45d and PRP39a (Chang et al., 2022). Furthermore, we observed that rbp45d mutants have short primary root and late flowering phenotypes, which are not present in other rbp45/47 mutants ( Figure 4 and Figure S7 ). However, we found some common features in all the RBP45/47 members. For example, they are localized in both the nucleus and cytoplasm ( Figure S5 ), and they all interact with the U1 snRNP core subunit U1C ( Figure S6 ). Thus, it is likely that other RBP45/47 members are also involved in splicing regulation through the mechanisms different from those of RBP45d. It is noted that most of the RBP45/47 members including RBP45d are localized both in nucleus and cytoplasm, indicating that they may also play roles in mRNA regulation in cytoplasm. It will be interesting to investigate biological functions of other RBP45/47 members as well as the molecular mechanisms by which they act in plants.

It is interestingly that the Tabaco NtRBP45 was recently reported to facilitate post-transcriptional gene silencing (PTGS) by promoting siRNA production (Zhang et al., 2020). While the Aarabidopsis RBP45d was reported to promote RNA-directed DNA methylation (RdDM), similar to the known components of RdDM (NRPD2A, RDR2, AGO4 and AGO6). It is suggested that RBP45d facilitates siRNA production by stabilizing either the precursor RNA or the slicer protein (Wang et al., 2021). Several research groups have also found mutations of some mRNA splicing factors reduced the level of siRNA dependent on POL IV, but it is unclear how these splicing factors affect the level of siRNA (Ausin et al., 2012; Dou et al., 2013; Huang et al., 2013; Zhang et al., 2013). It will be interesting to explore how the splicing factors function in RNA-directed DNA methylation in the future.

The Arabidopsis ecotype Columbia-0 (Col-0) was used as the wild type (WT) in this study. The mutants sot5 and clpR4-3 were previously described (Huang et al., 2018, Wu et al., 2013). The prp39a-1 (salk_133733), rbp45a (salk_140650), rbp45c (salk_063484), rbp47a (salk_142402), and rbp47b (GK-626C01) mutants were obtained from the ABRC stock center (https://abrc.osu.edu/) and were genotyped by PCR using gene specific primers ( Table S1 ). Double mutants prp39a-1 sot5, rbp45a sot5, rbp45c sot5 rbp47a sot5 rbp47b sot5, and clpR4-3 rbp45d-4 were identified from F₂ generations derived from crosses between single mutants by the PCR-based genotyping procedure. Seeds were surface-sterilized by 75% ethanol and stratified at 4°C for 3 days, and then sown onto half-strength Murashige and Skoog (MS) medium with 0.5% sucrose. Seeds were also directly sown in soil and grown in a phytotron with long-day conditions (16 h light/8 h dark) and light intensity (100 μmol photons m^-2 s^-1) at 22°C.

In order to obtain sot5 suppressors, sot5 seeds were ethyl methanesulfonate (EMS) mutagenized (0.1%). About 30000 M₂ plants from 5000 M₁ lines were screened. Several suppressor lines that displayed WT-like green leaves were obtained from the M₂ generation. To clone F23, we produced an F₂ population from the cross between F23 and Landsberg erecta (Ler). In the F₂ population, about eighty F23 plants were pooled for Whole Genome Resequencing and Mapping-By-Sequencing (MBS) analysis (OE Biotech Co.,Ltd). To clone the E25 gene, we crossed the E25 with sot5 and generated the BC₁ F₂ population. About thirty E25 plants were pooled for Whole Genome Resequencing and Mutmap analysis (Biomarker Technologies).

The RBP45d coding DNA sequence (CDS) and PRP39a genomic DNA were cloned into the pENTR SD/D-TOPO entry vector (Invitrogen), and then recombined into the pGWB2 destination vector, respectively. The destination vector containing RBP45d was transformed into F23 mutant and the destination vector containing PRP39a was transformed into E25 mutant according to the method as our previously described (Huang et al., 2018). To obtain recombinant RBP45d protein, the RBP45d CDS was transferred into the pet 51b vector and then was fused with His tag at its N-terminal.

The rbp45d-CR mutant line was generated by CRISPR/Cas9-mediated genome editing. The sgRNA sequences were designed by the online software CRISPR-P v2.0 (Liu et al., 2017) and CRISPR-GE (Xie et al., 2017) to target exon1 of the RBP45d gene. The binary vector pYAO-Cas9-1300 (Feng et al., 2018) was linearized using BsaI (ThermoFisher Scientific, UK) endonuclease. Two complementary sgRNA nucleotides were synthesized and annealed to generate double-strand DNA with appropriate overhangs on both ends and then inserted into the pYAO-Cas9-1300 vector. The vector was transformed into the Agrobacterium tumefaciens strain GV3101 competence cells. And the GV3101 with the vector was transformed into the Col-0 plants by the floral dip method. The transgenic plants of T₁ generation were screened by Hygromycin B resistance, DNA from positive lines was extracted and PCR and sequencing were performed to identify mutations in sgRNA target sequences. Finally, we screened the Cas9-free background and homozygous RBP45d gene editing lines in T₂ generation plants to obtain rbp45d-CR mutant lines.

For RT-PCR, total RNAs were extracted from seedlings or different plant tissues by RNA Easy Fast Plant Tissue Kit (TIANGEN, DP452) according to the manufacturer’s instructions. DNase I treatment and RT-PCR analysis were conducted as previously described (Huang et al., 2018). Semi-quantitative RT-PCR was carried out using the gene-specific primers ( Table S1 ). To identify the splicing variants, the PCR products amplified by the primers spanning the introns were purified and cloned into pMD19-T vector (TaKaRa, D102A). About 10~20 clones of each PCR products were sequenced and aligned with the corresponding genes by SnapGene software.

To express YFP-fused protein, the full length CDS of the RBP45a, RBP45b, RBP45c, RBP45d, RBP47a, RBP47b, RBP47c, RBP47c’, PRP39a and U1C genes were amplified from the cDNA of WT seedlings using the primers ( Table S1 ). The sequences were cloned into the pENTR SD/D-TOPO entry vector (Invitrogen) and then recombined into the p2YWG7 vector (http://gateway.psb.ugent.be/vector/show/p2GWY7/search/index/) or BiFC vctors pE3130 and pE3136 (http://www.bio.purdue.edu/people/faculty/gelvin/nsf/protocols_vectors.htm). The purified plasmid was transformed into the Arabidopsis protoplasts according to the method (Wu et al., 2009). The BiFC vector combinations are indicated in the Figures 5 and 10 .

The CDS of RBP45a, RBP45b, RBP45c, RBP45d, RBP47a, RBP47b, RBP47c, PRP39a U1A, U1C and U1-70K were clone into the pENTR vector and then recombined into pDEST22 and pDEST32 destination vectors, respectively. For yeast two-hybrid assays, plasmids were transformed into yeast strain AH109 by the lithium chloride–polyethylene glycol method according to the manufacturer’s manual (Clontech). The transformants were selected on SD-Leu-Trp plates. The protein–protein interactions were tested on SD-Trp-Leu-His plates with or without 5mM 3-amino-1,2,4-triazole (3AT).

The RBP45d CDS was cloned into the expression vector pet51b, and protein expression was induced using 0.5 mM isopropyl-b–D–thiogalactopyranoside at 16°C for 12 h in E. coli strain Rosetta. The protein was purified as described by Zhu et al. (2020). The RNA probes were chemically synthesized, and their 5’-end were labeled by Cy5 (Sangon Biotech (Shanghai) Co., Ltd). The sequences of the probes were as follows:

Cy5-SOT5-IN7: 5’-(CY5) CUCAAUAUGAUAAAUUUUGUAAAUC-3’;

The corresponding non-labeled and mutated probes were also chemically synthesized for competition assays.

For EMSA assay, RBP45d protein and RNAs were incubated at room temperature for 30 min in the reaction solution (10 mM HEPES pH 7.3, 20 mM KCl, 1 mM MgCl2, 1 mM dithiothreitol, 5% glycerol (v/v) and 0.1 μg tRNA). The reactants were analyzed using a native polyacrylamide gel, and signals were detected by Azure Biosystems C600.