An Arabidopsis PWI and RRM motif-containing protein is critical for pre-mRNA splicing and ABA responses

The phytohormone abscisic acid (ABA) is important for growth, development and stress responses in plants. Recent research has identified ABA receptors and signalling components that regulate seed germination and stomatal closure. However, proteins that regulate ABA signalling remain poorly understood. Here we use a forward-genetic screen to identify rbm25-1 and rbm25-2, two Arabidopsis mutants with increased sensitivity to growth inhibition by ABA. Using RNA-seq, we found that RBM25 controls the splicing of many pre-mRNAs. The protein phosphatase 2C HAB1, a critical component in ABA signalling, shows a dramatic defect in pre-mRNA splicing in rbm25 mutants. Ectopic expression of a HAB1 complementary DNA derived from wild-type mRNAs partially suppresses the rbm25-2 mutant phenotype. We suggest that RNA splicing is of particular importance for plant response to ABA and that the splicing factor RBM25 has a critical role in this response. The phytohormone ABA plays a critical role in plant stress responses. Here, using a forward-genetic screen, Zhan et al. discover a splicing factor that plays an important role in splicing HAB1 phosphatase and fine-tuning ABA sensitivity in Arabidopsis.

T he phytohormone abscisic acid (ABA) is critical for many processes in the life cycle of plants, such as the promotion of seed maturation, establishment and maintenance of seed dormancy, regulation of stomatal aperture and adaptation to environmental stresses [1][2][3] . On exposure to environmental stresses, plants rapidly increase their ABA content as a result of increased ABA synthesis, decreased ABA catabolism, decreased formation of inactive ABA conjugates, or a combination of these factors 4 . The de novo biosynthesis of ABA is through the cleavage of a C 40 carotenoid precursor, followed by the conversion of the intermediate xanthoxin to ABA via ABA aldehyde 2 . The cleavage step catalysed by a 9-cis epoxycarotenoid dioxygenase (NCED) has generally been considered the rate-limiting step in the pathway 4,5 . Wang et al. 6 found that CED1 (for NCED defective 1), which is a putative a/b hydrolase domain-containing protein and is allelic to the BODYGUARD gene that is essential for cuticle biogenesis, plays an important role in regulating the expression of NCED3 and genes encoding other signalling components downstream of ABA biosynthesis.
Following ABA biosynthesis, multiple signal transduction pathways, which amplify the primary signal produced when ABA binds to its receptors, are required to control the ABAregulated adaptive responses. These adaptive responses include stomatal closure (which allows the plant to conserve water), the accumulation of compatible osmolytes such as proline 7 , and the differential expression of a wide array of stress-responsive genes 8,9 . The core ABA signal transduction pathway is mediated by receptors in the PYR/PYL/RCAR family, which, on binding to ABA, inhibit the type 2C protein phosphatases (PP2Cs) such as ABI1, ABI2 and HAB1 (refs 3,10-12). The PP2Cs are negative regulators of the ABA signalling pathway 13,14 . Inhibition of the PP2Cs leads to the activation of protein kinases in the sucrose non-fermenting related kinase 2 (SnRK2) family. In the absence of ABA, the PP2Cs can directly inactivate SnRK2 kinases by dephosphorylating a critical amino acid residue in their kinase activation loop 15,16 . These kinases are required for the subsequent regulation of the activities of downstream effectors in ABA signalling.
Since the three SnRK2 kinases, SnRK2.2, SnRK2.3 and SnRK2.6, control virtually all of the ABA responses in plants, the substrate proteins of the SnRK2s function as the downstream effectors of ABA responses 17 . The best known of the substrates include several transcription factors and ion channels [18][19][20][21] . The basic leucine zipper family transcription factors such as ABI5 and other ABA response factors or ABA-responsive element-binding factors (AREBs) control the transcription of ABA-responsive genes 18,20 . The ion channels such as the slow anion channel SLAC1 and the potassium channel KAT1 are critical for guard cell regulation by ABA 19,[21][22][23][24] . Recent phosphoproteomics studies identified dozens of additional substrate proteins of the SnRK2s including several proteins with potential roles in RNA processing and proteins involved in chloroplast function 25,26 . These newly identified substrate proteins are potential effectors of ABA signalling, although their functions in ABA responses are largely unknown.
To identify cellular factors important for ABA responses, particularly ABA responses in leaf development, we conducted a genetic screen in Arabidopsis thaliana for mutants with defective leaf greening response to ABA treatment. We report here that roa1 (regulator of ABA response) mutants are hypersensitive to ABA (HAB) as indicated by impaired leaf greening in response to ABA. The ROA1/RBM25 locus defines a splicing factor required for the splicing of transcripts of many genes including those involved in ABA signal transduction pathways. Our work has revealed a critical role of pre-mRNA splicing in ABA responses in plants.
Results roa1 mutants are hypersensitive to ABA. Identification of a large number of mutants defective in seed germination or guard cell movements has led to the discovery of genes involved in ABA biosynthesis and its downstream signalling events 3 . High levels of ABA inhibit plant growth and development and cause leaves to become pale or yellow 27,28 . The cellular factors controlling these processes are poorly understood. To identify critical regulators for ABA responses at the seedling growth stage, we screened an A. thaliana transfer DNA (T-DNA) insertion population with known integration sites (the SALK collection from ABRC) for mutants with increased sensitivity to exogenous ABA in the growth medium in terms of root growth and leaf yellowing. A similar screen was performed with an in-house generated T-DNA insertional population in a TOUCH 4 promoter:luciferase reporter line 29 (pTCH4:LUC). These mutants were designated as roa. The roa1-1 (SALK_064472 from the SALK collection) and roa1-2 (from the pTCH4:LUC population) mutants were chosen for in-depth characterization ( Supplementary Fig. 1a). To compare roa1-1 and roa1-2 side-byside, roa1-2 was backcrossed with Col-0 and roa1-2 allele without the pTCH4:LUC transgene was obtained. In the absence of exogenous ABA, roa1-1 and roa1-2 seedlings do not exhibit obvious growth defects (Fig. 1a). However, after being transferred to media supplemented with 5 mM ABA, the newly emerged leaves in roa1-1 and roa1-2 seedlings are yellow, whereas all of the leaves in the wild-type seedlings are green (Fig. 1a). The mutants also had reduced primary root elongation in response to 5 mM ABA (Fig. 1a,b).There is substantially less accumulation of chlorophylls in ABA-treated roa1-1 and roa1-2 leaves (Fig. 1a,c). Genetic analysis showed that both roa1-1 and roa1-2 mutations are recessive and are caused by mutations in a single nuclear gene (Supplementary Data 1). Pair-wise crosses between roa1-1 and roa1-2 suggested that the roa1-1 and roa1-2 mutations are allelic to each other (Fig. 1d).
To investigate whether the roa1-1 and roa1-2 mutations alter plant responses to other plant hormones or abiotic stress, we examined the responses of roa1-1 and roa1-2 plants to heat stress and two additional hormones. Sensitivity of the soil-grown roa1-1 and roa1-2 plants at both vegetative and reproductive stages to heat stress is not altered when they are compared with the wildtype plants ( Supplementary Fig. 1b). We used the precursor of ethylene biosynthesis, 1-amino-cyclopropane-1-carboxylic acid (ACC), to examine the effect of ethylene on hypocotyl elongation of dark grown roa1-1 and roa1-2 seedlings. Hypocotyl elongation of roa1-1, roa1-2 and wild-type seedlings was equally inhibited by exogenous application of ACC ( Fig. 1e; Supplementary Fig. 1c). We further examined effects of auxin on hypocotyl and root elongation of roa1-1 and roa1-2 seedlings. roa1-1 and roa1-2 seedlings essentially responded to the same extent to indole-3-acetic acid in the growth media ( Fig. 1f; Supplementary Fig. 1d,e). Therefore, while ROA1 may be involved in the response to stresses and hormones other than ABA, we find no evidence of its involvement in the response to auxin, ethylene or heat.
ROA1 encodes a PWI and RRM motif-containing protein.
Although roa1-1/SALK_064472 was annotated as having a T-DNA insertion in the At1g60200 gene, we could not find any T-DNA insertion in this gene. We carried out thermal asymmetric interlaced PCR (TAIL-PCR) analysis and identified a T-DNA insertion downstream of At1g60200 in roa1-1 (Fig. 2a). Although roa1-2 also came from a T-DNA mutant population, there is no T-DNA insertion in roa1-2. To identify the gene responsible for the roa1-1 and roa1-2 mutant phenotypes, we prepared mapping populations by crossing roa1-1 and roa1-2 mutants with Landsberg erecta wild-type plants and performed map-based cloning for the ROA1 locus. Genetic mapping pinpointed ROA1 to a genomic region that contains the At1g60200 gene. In the roa1-1 mutant, we identified a single nucleotide substitution in At1g60200, and this substitution changes alanine at position 899 to valine in the deduced polypeptide ( Fig. 2a,b). The roa1-2 mutation also has a single nucleotide substitution in At1g60200, which results in the change of glutamine at position 570 to a premature stop codon in the deduced polypeptide ( Fig. 2a,b). We confirmed the ROA1 gene is At1g60200 by functional complementation analysis. The wild-type At1g60200 gene driven by its native promoter complemented the roa1-1 and roa1-2 mutant phenotype (Fig. 2d,e). ROA1 expression is severely disrupted in roa1-1, possibly due to the T-DNA insertion downstream of ROA1 (Fig. 2f).
ROA1 encodes a putative PWI and RRM motif-containing protein with similarities to the human RNA-binding protein 25 (HsRBM25) (Fig. 2b,c). Hereafter, we refer to ROA1 as Arabidopsis RBM25 and refer to roa1-1 and roa1-2 mutant alleles as rbm25-1 and rbm25-2, respectively. Arabidopsis RBM25 shares substantial sequence similarity (50% over the RRM motif and 71% over the PWI motif) with the HsRBM25, which is known as a splicing factor 30 . Database searches revealed that RBM25 orthologues are present in other plant species including monocots (such as rice and maize) and dicots (such as soybean and tomato) ( Fig. 2c; Supplementary Fig. 2). The last amino acid of RBM25 (changed to valine in rbm25-1) is conserved in nearly half of the plant RBM25 orthologues ( Supplementary Fig. 2). As expected of a putative splicing factor, the RBM25-green fluorescent protein (GFP) fusion protein driven by the RBM25 native promoter is localized in the nucleus of Arabidopsis root cells (Fig. 2g).
RBM25 negatively controls ABA responses in early development. Seed germination is sensitive to ABA and we were interested whether RBM25 plays a role in this process. rbm25-1 is slightly hypersensitive to exogenous ABA at the stage of seed germination (Fig. 3a) and is substantially hypersensitive to ABA in post-germination seedling development (Fig. 3b,c). This hypersensitivity to ABA was suppressed in three independent complementation lines of rbm25-1 (Fig. 3a,b,c), indicating that loss-of-function of RBM25 results in this phenotype. ABA is also known to induce stomatal closure on drought stress. We performed water loss assays with detached shoots of soil-grown rbm25 mutant plants to determine whether rbm25 mutations alter plant transpirational water loss. Transpirational water loss rates are the same between the wild-type and rbm25 mutants with or without exogenous application of ABA (Fig. 3d). These results suggest that rbm25 mutations do not alter stomatal responses to ABA.
The rbm25-1 mutation alters transcript accumulation profiles. We performed RNA-seq experiments to determine whether the rbm25-1 mutation affects transcript accumulation profiles and whether altered transcript accumulation profiles may help explain the increased sensitivity of rbm25-1 mutant to ABA treatment. We used the Illumina HiSeq 2500 System to sequence mRNA-seq libraries prepared from wild-type and rbm25-1 seedlings with three biological replicates and obtained a minimum of 14.7 million of pair-end clean sequence reads (Supplementary Data 2). Compared with those in the wild-type, 106 genes displayed higher transcript levels (by at least 2-fold and with Po0.05), while 97 genes showed lower transcript levels in rbm25-1 under control conditions (Fig. 4a,b; Supplementary Data 2). The RNA-seq analysis also revealed that 185 genes displayed at least a 2-fold  transcription-PCR (qRT-PCR) analysis, we confirmed that the expression of At5g65080 and At1g53490 is substantially reduced in rbm25 mutants, while expression of At4g33720 is increased in rbm25 mutants (Fig. 4d). At5g65080 encodes a MADS-box transcription factor. At1g53490 encodes a RING/U-box superfamily protein. At4g33720 encodes a CAP (cysteine-rich secretory proteins, Antigen 5 and Pathogenesis-related 1 protein) superfamily protein.
RBM25 regulates the splicing of pre-mRNAs. RBM25 has similar functional domains as the human splicing factor HsRBM25. To determine whether RBM25 functions in pre-mRNA splicing, we examined our RNA-seq data sets to look for potential effects of rbm25-1 mutation on pre-mRNA splicing. The analysis revealed that 359 genes have splicing defects in untreated rbm25-1, while 416 genes show splicing defects in ABA-treated rbm25-1 ( Fig. 5a Relative mRNA level  We designed primers unique to the intron of genes that are retained in the rbm25-1 plants and carried out semi-qRT-PCR assays to validate the IR results from the RNA-seq analysis. For RT-PCR analysis, we selected At4g16143, At5g09330, At2g42010, At3g53340, At4g32040 and At4g35800. At4g16143 encodes importin alpha isoform 2. At5g09330 encodes NAC domaincontaining protein 82. At4g42010 encodes phospholipase D beta 1. At3g53340 encodes nuclear factor Y, subunit B10. At4g32040 encodes KNOTTED1-like homeobox gene 5. At4g35800 encodes the unique largest subunit of nuclear DNA-dependent RNA II. The RT-PCR assays found that transcripts of At4g16143, At5g09330, At2g42010, At3g53340, At4g32040 and At4g35800 contain an intron in rbm25-1 and rbm25-2 with or without ABA treatment (Fig. 5c,e; Supplementary Fig. 3). Thus, the RT-PCR results confirmed the RNA-seq data of IR events in rbm25-1.
RT-PCR analysis confirmed that exon 4 of At2g48120 is skipped in rbm25 mutants under both control and ABA treatment conditions ( Fig. 5c; Supplementary Figs 3 and 4). At2g48120 encodes the pale cress protein involved in chloroplast mRNA maturation 31 . The genes carrying defective alternative splicing patterns in rbm25-1 plants under control or ABA treatment conditions encode proteins with diverse functions in many biological processes, and the predicted roles of a good portion of these genes involve responses to abiotic or biotic stresses (Supplementary Data 3). It is possible that altered activities of mis-spliced genes in unstressed rbm25-1 mutant account for differentially expressed genes in this mutant under control conditions as determined by RNA-seq analyses. There are 31 genes with a predicted role in DNA-dependent transcription in each category of abnormal splicing in unstressed rbm25-1 mutant (Supplementary Data 3). These genes include promoter-specific transcription factors, general transcription factors and RNA polymerase II subunits.
We reasoned that disrupted function of the mis-spliced genes in rbm25-1 plants should at least partly explain the impaired responses of rbm25 mutants to ABA treatment including alterations in gene expression and increased sensitivity of rbm25 mutants to ABA in terms of root growth and leaf yellowing. Thirty genes involved in transcription were found to have splicing defects in their transcripts in rbm25-1 mutant under ABA treatment (Supplementary Data 3). These genes include the largest subunit of DNA-dependent RNA polymerase II, TATA-binding proteins, a nuclear factor, promoter-specific transcription factors and the subunits of the mediator complex. The mediators are involved in regulation of transcription from RNA polymerase II 32 . Loss-of-function or reduced activities of these genes will certainly lead to altered expression of genes in rbm25-1 mutant under ABA treatment. Although genes carrying splicing defects in rbm25-1 mutant under ABA treatment encode proteins involved in diverse biological processes, a substantial portion of the genes encode proteins that are predicted to function in biotic or abiotic stress-response pathways (Supplementary Data 3). In addition, the calcium- WT rbm25-1 #2 #3  NATURE COMMUNICATIONS | DOI: 10.1038/ncomms9139 ARTICLE sensing receptor and HAB1 are known regulators of ABA responses [33][34][35] . There are additional PP2Cs whose transcripts are mis-spliced in ABA-treated rbm25-1 mutant. Like HAB1, these PP2Cs may function as negative regulators of ABA signalling. Finally, impaired functions of four splicing factors (SC35-like splicing factor 33 encoded by At1g55310, CC1-like splicing factor encoded by At2g16940, U2 snRNP splicing factor encoded by At1g60900 and splicing endonuclease 1 encoded by At3g45590) may contribute to defects of alternative splicing of gene transcripts in rbm25-1. ABA represses the expression of genes involved in photosynthesis including the small subunit of ribulose-1,5bisphosphate carboxylase/oxygenase (Rubisco) (rbcS genes) and chlorophyll a/b-binding proteins 36,37 . Related to this aspect of cellular function, we found mis-splicing of a group of genes that are involved in biogenesis and functionality of chloroplasts in ABA-treated rbm25-1 mutant. These genes include pale cress protein (encoded by At2g48120; the mutant of this gene has reduced chlorophyll content 31 ), the PsbQ subunit of the oxygenevolving complex of photosystem II (encoded by At4g21280) 38 , starch branching enzyme 2.2 (encoded by At5g03650; it is transcriptionally regulated by the ABA-insensitive 4 (ABI4) 39 ), UDP-glycosyltransferase superfamily protein involved in flavonoid biosynthetic process (encoded by At4g09500) and plastid transcriptionally active 5 (PTAC5, encoded by At4g13670 (ref. 40)) (Supplementary Data 3). Mis-splicing for the transcripts of these proteins in rbm25-1 mutant may be associated or even contribute to the leaf yellowing phenotype of the mutant. Finally, we observed that a cyclin-dependent protein kinase (CDKC2 encoded by At5g64960) is mis-spliced in rbm25-1 under ABA treatment (Supplementary Data 3). CDKC2 colocalizes with spliceosomal components in a manner dependent on the transcriptional status of the cells and on CDKC2-kinase activity. Expression of CDKC2 modifies the location of spliceosomal components 41 . Thus, disruption of CDKC2 may exacerbate the splicing defects caused by the rbm25-1 mutation.
RBM25 controls the alternative splicing of HAB1 transcripts. Our RNA-seq analysis revealed that the last intron of At1g72770 (which encodes HAB1) is retained in rbm25-1 plants (Supplementary Data 3). HAB1 is one of the clade A PP2Cs that are co-receptors of ABA and important negative regulators of ABA signalling 14,33,42,43 . HAB1 transcripts are known to exist in three alternatively spliced forms (The Arabidopsis Information Resource, http://www.arabidopsis.org). In our efforts to confirm intron-retention event in rbm25-1 and rbm25-2 with RT-PCR and qRT-PCR analyses, we found additional splicing defects of HAB1 in the rbm25 mutants. The rbm25 mutations cause altered accumulation of four alternatively spliced transcripts of HAB1 under control and ABA treatment conditions (Fig. 6a,b). Because HAB1 is upregulated by ABA, the defects of HAB1 are more   pronounced in these two mutants under ABA treatment.
Ectopic expression of HAB1 suppresses rbm25-1. Loss-offunction mutations in HAB1 cause ABA hypersensitivity in early seedling growth and development 14,33,44 . Consistent with previous reports, the hab1-1 mutant is hypersensitive to ABA during seed germination and early seedling development (Fig. 3b). We observed that in response to ABA treatment hab1-1 mutant seedlings exhibit a leaf yellowing phenotype similar to that of rbm25 mutants, although the yellowing is not as severe as in rbm25 (Fig. 2d)  reduction of the HAB1.1 form contributes significantly to the leaf yellowing phenotype of rbm25 mutant plants.
ABA affects splicing of transcripts. We also examined the effect of ABA on the splicing pattern of transcripts in wild-type plants.
ABA induces alternative splicing events in the transcripts of 27 genes in wild-type plants (Supplementary Data 3). Among these alternatively spliced genes in ABA-treated wild-type plants, two of them (At3g28670 and At3g24170) showed more severe splicing defects in ABA-treated rbm25-1 mutant. These results show that the ABA treatment causes defects in alternative splicing of pre-mRNAs in wild-type plants, and the splicing defects become much more severe when the splicing factor RBM25 is impaired.
ABA regulates RBM25 transcript level and post-translationally. ROA1 expression is upregulated by ABA treatment (Fig. 7a). This is consistent with publicly available gene expression profiling data on ROA1 under ABA treatment obtained using whole genome tiling arrays 45 ( Supplementary Fig. 5). This is also consistent with the ABA-induced mutant phenotypes and more severe splicing defects in rbm25 mutants under ABA treatment (Supplementary Data 3), and suggests a need for more RBM25 to strengthen pre-mRNA splicing under ABA treatment.
Protein phosphorylation is central to ABA signalling 18,20,23 . ABA increases the phosphorylation levels of several splicing factors including arginine/serine-rich splicing factor 41, a PWI domain-containing protein, splicing factor U2af small subunit A and serine/arginine-rich protein splicing factor RSZ32, while ABA also represses the phosphorylation levels of several splicing factors 26 . Because RBM25 is upregulated by ABA, we examined whether phosphorylation status of RBM25 may also be regulated by ABA. We used a novel mass-spectrometry-based label-free quantitation method that facilitates systematic profiling of plant phosphoproteome changes with high efficiency and accuracy 46 . This method employs synthetic peptide libraries tailored specifically as internal standards for complex phosphopeptide samples and accordingly, a local normalization algorithm, LAXIC, which calculates phosphopeptide abundance normalized locally with co-eluting library peptides. Our mass-spectrometry studies revealed that RBM25 is phosphorylated and ABA represses the phosphorylation level of RBM25 (Fig. 7b,c). Based on this data, we suggest that dephosphorylation of RBM25 may be important for its splicing function under ABA treatment. Alternatively, dephosphorylation of RBM25 may be important for its protein stability because phosphorylation often leads to protein degradation through ubiquitin-mediated proteolysis 47 4). (b,c) Quantitative phosphoproteomic profiles of total proteins were performed as described 46 . Proteins were extracted from 12-day-old WT plants subjected to 50 mM ABA treatment for 0 or 0.5 h, and the enrichment of the phosphoproteins and label-free quantification were performed as described 46 . This experiment was repeated three times with similar results. (b) Mass spectrometric spectra of a phosphopeptide (aa 407-432) in RBM25 identified in vivo. (c) Relative abundance of the RBM25 phosphopeptide.

Discussion
In our unique genetic screen for regulators of ABA responses at seedling developmental stage, we identified a splicing factor, RBM25. The rbm25 mutant plants do not exhibit obvious growth and developmental defects under normal growth conditions, and the mutants are not altered in their responses to heat stress, ethylene or auxin. We established the identity of RBM25 through a map-based cloning strategy followed by a gene complementation analysis. Our genetic evidence demonstrated that RBM25 is required for proper ABA responses in plants by maintaining proper patterns of splicing of gene transcripts. Although splicing is important for all cellular processes involving gene expression, the splicing factor RBM25 is essential for root growth and leaf development in response to ABA but appears not to be required for normal growth and development. The rbm25 mutants are hypersensitive to ABA in the growth medium, indicating that RBM25 may be involved in mediating signal transduction events downstream of ABA. Indeed, our RNA-seq analyses revealed that RBM25 controls the splicing of one of the central ABA signalling molecules, HAB1 (Fig. 6). Ectopic expression of HAB1.1 alone can partially suppress the hypersensitive phenotype of rbm25-2 to ABA in growth medium (Fig. 6). These results indicate that RBM25 is critical for ABA signalling.
Because gene splicing is an essential cellular process, it is difficult to obtain and study null alleles of splicing factor genes due to their lethality. The rbm25-1 and rbm25-2 alleles appear to be null alleles since RBM25 is not expressed in rbm25-1 and the PWI domain, which is presumably important for splicing function, is truncated in rbm25-2 (Fig. 2b,f). To our knowledge, this is the first time null alleles of a conserved splicing factor are found and characterized in plants. The conditional phenotype (that is, under ABA treatment only) of rbm25 mutants implies that RBM25 is not essential for splicing for all or most plant genes. However, there are some splicing defects in rbm25 mutant plants even under control conditions, but the splicing defects become more severe under ABA treatment such that growth and developmental phenotypes are manifested by the mutants under ABA treatment. Because rbm25 mutants lack obvious growth and developmental defects in the absence of ABA or in responses to heat stress and two additional hormones (ethylene and auxin), but have severe phenotypic defects in the presence of ABA ( Fig. 1 and Supplementary Fig. 1), we propose that plants may have a particular requirement for splicing in the presence of ABA. Indeed, this splicing factor (RBM25) shows increased expression under ABA treatment, which is consistent with the notion of an exaggerated requirement for splicing in the presence of ABA. On the other hand, we speculate that ABA signalling also regulates splicing in at least two ways: to increase the level of RBM25 transcript; and regulate the phosphorylation status of RBM25 ( Fig. 7; Supplementary Fig. 5) and other splicing factors 26 . The phosphorylation levels of several splicing factors are regulated by ABA in wild-type plants 26 , which perhaps strengthen splicing so as to help plant cope with ABA and related environmental stress. In this study, we showed that the phosphorylation level of RBM25 is downregulated by ABA and we propose that reduced phosphorylation of RBM25 might be important for its activity as a splicing factor or its protein stability under ABA treatment although the physiological significance of RBM25 phosphorylation remains to be determined. Because orthologues of RBM25 exist in various plant species (Fig. 2c; Supplementary Fig. 2), the molecular function of RBM25 may be conserved in other plant species including crops.
Our RNA-seq experiments detected many genes whose splicing is controlled by RBM25. We observed that the rbm25-1 mutant plants carry splicing defects in alternative splicing of pre-mRNAs (Supplementary Data 3). Our observation that there are more genes with defective splicing patterns when rbm25-1 mutant was treated with ABA is consistent with ABA upregulation of ROA1 ( Fig. 7a; Supplementary Fig. 5; Supplementary Data 3). Intron-retained transcripts in rbm25-1 mutant tend to include premature stop codons and therefore produce truncated or even inactive forms of proteins. In addition, many IR events may lead to the generation of nonsense-mediated decay substrates so that the overall level of the transcript would decrease and no translation would take place. Other abnormal, alternatively spliced transcripts in rbm25-1 mutant probably produce nonfunctional proteins, and such splicing events were seldom detected in wild-type plants in our RNA-seq experiments. Several previous studies have shown that mutants with altered sensitivity to ABA are affected in pre-mRNA splicing under normal developmental conditions [49][50][51][52][53][54][55] or under abiotic stress conditions including drought and salinity [56][57][58][59][60] . In the previous studies, the various mutants were defective in ABA-regulated seed germination and/or stomatal regulation, but not in seedling leaf greening. In addition, the previously published mutants are defective in not only splicing but also other RNA processing events, and the published work did not link the mis-splicing of a particular gene(s) to the mutant phenotypes. In contrast, the rbm25 mutants are primarily defective in leaf greening in response to ABA, and we have identified the mis-splicing of HAB1 as a critical mis-spliced gene contributing to the mutant phenotype (Fig. 6), although the defective splicing in many other genes may also contribute to the mutant phenotypes of rbm25.

Methods
Plant materials. T-DNA insertional mutant pool with known T-DNA insertion sites including rbm25-1 (SALK_064472) were obtained from Arabidopsis Biological Resource Center (ABRC; Columbus, OH). rbm25-2 (containing pTCH4:LUC reporter gene 29 ; seeds of pTCH4:LUC line was kindly provided by Dr Janet Braam) was isolated from a T-DNA-mutagenized Arabidopsis T 2 population. Seeds were surface sterilized and sown in germination medium (1 Â Murashige-Skoog (MS) salts, 2% sucrose, 1.2% agar, pH 5.7). For ABA-sensitivity screening, 5-day-old seedlings were transferred to new MS agar plates supplemented with 5 mM ABA and were allowed to grow vertically for an additional 15-20 days. Seeds of hab1-1 (SALK_002104) were obtained from the ABRC. Plants at all developmental stages were grown in a growth chamber at 22 ± 1°C under cool, white light (B100-120 mmol m À 2 s À 1 ) with a long-day photoperiod (16-h light/8-h dark).
Map-based cloning and gene complementation. The rbm25-1 mutant was crossed with the Landsberg erecta accession, and a total of 896 plants homozygous for the rbm25-1 phenotype in response to ABA in MS medium were selected from the F 2 population. A separate F 2 population was also generated from a cross between rbm25-2 and Landsberg erecta, and 896 plants homozygous for the rbm25-2 phenotype in response to ABA in MS medium were selected. Simple sequence length polymorphism markers were designed according to the information in the Cereon Arabidopsis Polymorphism Collection and were used to analyse recombination events. Mapping indicated that rbm25-1 and rbm25-2 mutations were located on the BAC clones T30E16 and F8A5. Candidate genes within this region were sequenced from the rbm25 mutants and compared with those in GenBank to find the rbm25-1 and rbm25-2 mutations. For gene complementation of roa1 mutants, a 7.4-Kb genomic fragment of At1g60200 including 2.2 Kb upstream of the translation initiation codon and 0.4 Kb downstream of the translation initiation codon was amplified by PCR with T13D8 as a template (primers were listed in Supplementary Data 1). The amplified fragment was first cloned via Gateway technology (Invitrogen) into the pENTR1A Dual Selection Vector. The RBM25 gene was then introduced into the pGWB501 vector, resulting in plasmid pGWB501-ROA1. The pGWB501-ROA1 construct was transferred into Agrobacterium tumefaciens (strain GV3101), and rbm25-1 and rbm25-2 plants were transformed by the floral dip method. Transgenic plants were selected on MS agar plates containing 50 mg l À 1 hygromycin.
Subcellular localization. The B2-Kb genomic DNA fragment upstream of the start codon of RBM25 (At1g60200) was amplified by PCR with T13D8 as a template using the primer pair At1g60200P-F5 and At1g60200P-R2, resulting in DNA fragment I. The coding region of At1g60200 (from the start codon to the last codon without stop codon or any intron sequences) was amplified by PCR with the primer pairs At1g60200CDS-F2 and At1g60200CDS-R2, resulting in DNA fragment II. The DNA fragments I and II were joined together by PCR amplification using the primer pairs At1g60200P-F5 and At1g60200CDS-R2 with DNA fragments of I and II as a template. The resulting DNA fragment III was cloned into a vector pENTR1A initially and subsequently into the binary vector pGWB504, resulting in the plasmid RBM25:RBM25-GFP. The RBM25:RBM25-GFP construct was then introduced into Arabidopsis wild-type (Col-0) plants by floral dip transformation with A. tumefaciens strain GV3101. The subcellular localization of RBM25:RBM25-GFP protein in roots of transgenic plants (T 2 generation) was determined with a Leica SP5X confocal microscope (Leica Microsystems).
RNA-seq data analyses and validation experiments. Fourteen-day-old wildtype and rbm25-1 seedlings grown on MS medium (1 Â MS salts, 2% sucrose, 0.6% agar, pH 5.7) were treated with 0 or 100 mM ABA for 6 h and were used for total RNA extraction. Total RNA was isolated with the Universal Plant Total RNA Extraction Kit (BioTeke) and treated with TURBO DNA-free Kit (Ambion) to remove any genomic DNA contaminants. mRNA-seq libraries were constructed following the standard Illumina protocols. There are three biological replicates per genotype. Illumina sequencing was performed in the Shanghai Center for Plant Stress Biology with an Illumina HiSeq 2500 System.
According to RNA-seq-mapped reads and the reference-annotated transcripts, transcriptomes were reconstructed for each sample by Cufflinks v2.2.1 (ref. 64). Given variable efficiency of mRNA enrichments and rRNA depletion kits in samples, these transcripts were masked in transcriptome constructions to improve the overall robustness of transcript abundance estimates. To obtain a high confidence of transcriptomes, the novel constructed transcript was filtered out when the abundance was o20% (default is 10%) of the most abundant isoform for each gene. Then all of the constructed transcriptomes were merged with the reference-annotated transcripts using Cuffmerge 64 to yield comprehensive reannotated gene transcripts including known and novel annotated transcripts in our RNA-seq samples. Subsequently, significantly differentially expressed genes were predicted by Cuffdiff 64 between the controls and tested samples, using twofold change and multiple test P valueo0.05.
Given that IR is the most frequent alternative splicing (AS) type in plants 65,66 and juncBASE especially improves the reliability of the detection for such AS type 67 , we adopted juncBASE v0.6 to identify AS events within genes for our samples. First, two different databases were constructed, one is based on comprehensive re-annotated gene transcripts (as described above) to identify all internal AS events (not alternative first or last exon events), and the other is derived from just the reference-annotated transcripts to define alternative first and last exons. Then, a series of python scripts from juncBASE were employed to identify exon-exon junctions, exon-intron junctions and AS events, qualify the events and give significantly differentially splicing events between samples. Accordingly, juncBASE-detected AS events ranged from B15,500 to B20,500, and among the events 358 to 1,464 ones showed significantly differentially AS events (P valueo0.05 and abs (delta_val)45) between pair-wise samples. Moreover, a module from CASH (comprehensive alternative splicing hunting, the renewed version of ASD (alternative splicing detector 68 ) was used to rescue 19 to 50 differential IR events. Only genes with alternative splicing patterns in all three biological replicates of rbm25-1 mutant plants are listed as differentially alternatively spliced genes.
For RT-PCR and qRT-PCR analyses, 5 mg of total RNA was used for synthesis of the first-strand cDNA. Each experiment had three to four biological replicates and was repeated at least three times. The comparative cycle threshold (ct) method was applied for calculating gene expression levels, and ACT2 was used as a reference gene.