Deciphering the plant splicing code: experimental and computational approaches for predicting alternative splicing and splicing regulatory elements.

Extensive alternative splicing (AS) of precursor mRNAs (pre-mRNAs) in multicellular eukaryotes increases the protein-coding capacity of a genome and allows novel ways to regulate gene expression. In flowering plants, up to 48% of intron-containing genes exhibit AS. However, the full extent of AS in plants is not yet known, as only a few high-throughput RNA-Seq studies have been performed. As the cost of obtaining RNA-Seq reads continues to fall, it is anticipated that huge amounts of plant sequence data will accumulate and help in obtaining a more complete picture of AS in plants. Although it is not an onerous task to obtain hundreds of millions of reads using high-throughput sequencing technologies, computational tools to accurately predict and visualize AS are still being developed and refined. This review will discuss the tools to predict and visualize transcriptome-wide AS in plants using short-reads and highlight their limitations. Comparative studies of AS events between plants and animals have revealed that there are major differences in the most prevalent types of AS events, suggesting that plants and animals differ in the way they recognize exons and introns. Extensive studies have been performed in animals to identify cis-elements involved in regulating AS, especially in exon skipping. However, few such studies have been carried out in plants. Here, we review the current state of research on splicing regulatory elements (SREs) and briefly discuss emerging experimental and computational tools to identify cis-elements involved in regulation of AS in plants. The availability of curated alternative splice forms in plants makes it possible to use computational tools to predict SREs involved in AS regulation, which can then be verified experimentally. Such studies will permit identification of plant-specific features involved in AS regulation and contribute to deciphering the splicing code in plants.

Introduction

Seminal discoveries in RNA biology in recent years have established a central role for RNAs in gene regulation at the transcriptional, post-transcriptional, and translational level in eukaryotes (reviewed in Chen, 2009; Sharp, 2009; Voinnet, 2009; Licatalosi and Darnell, 2010; Kalsotra and Cooper, 2011; Staiger and Green, 2011). In photosynthetic eukaryotes a vast majority of protein-coding genes (up to 90%) contain non-coding intronic sequences, hence the primary transcripts must undergo splicing to generate mature functional mRNAs (Reddy, 2007; Barbazuk et al., 2008; Labadorf et al., 2010). Pre-mRNA splicing is carried out by the spliceosome, a large ribonucleoprotein complex. In plants, as in animals, there are two types of spliceosomes. The major type is called U2 type, which performs splicing of U2-dependent introns, whereas the minor U12 type is involved in splicing of rare U12-dependent introns (Simpson and Brown, 2008). Both spliceosomes consist of five snRNAs (U1, U2, U4, U5, U6 in the major spliceosome and U11, U12, U4atac, U5, and U6atac in the minor spliceosome). The protein composition of the major spliceosome has been extensively studied in animals, revealing that it contains close to 200 proteins (Wahl et al., 2009; Valadkhan and Jaladat, 2010). Computational analysis has revealed that plants have RNA and protein components of both spliceosomes (Ru et al., 2008; Simpson and Brown, 2008).

Primary transcripts from intron-containing genes can be alternatively spliced by differential selection of splice sites, leading to production of multiple mature mRNAs from a single gene, which is considered a major source for proteome diversity (Black, 2003; Reddy, 2007; Pan et al., 2008; Ru et al., 2008; Kalsotra and Cooper, 2011). Protein isoforms produced by splice variants may have altered functions (Black, 2003; Stamm et al., 2005). In addition, AS plays an important role in gene regulation through regulated production of splice variants with a premature termination codon, which are degraded through nonsense-mediated decay and other RNA surveillance mechanisms (Chang et al., 2007; Kurihara et al., 2009; Barbazuk, 2010; Palusa and Reddy, 2010; Staiger and Green, 2011) or contain target sequences for miRNA so that they are either degraded or not translated (Tan et al., 2007; Chen, 2009). Hence, post-transcriptional regulation of gene expression by pre-mRNA splicing plays a crucial role in generating transcriptome and proteome diversity and provides novel ways to fine-tune gene regulation.

AS in plants was under-appreciated until recently, and it was considered rare as pre-mRNAs of only a few genes were known to undergo AS. For instance, in 2001 pre-mRNAs from only three dozen genes in plants were known to undergo AS (Reddy, 2001). However, the completion of the Arabidopsis genome little over a decade ago, and other plant genomes more recently, as well as the availability of a massive amounts of transcribed sequence data in the form expressed sequence tags (ESTs)/cDNAs and limited RNA sequence data generated with next generation sequencing (NGS) technologies, have allowed the analysis of transcriptome-wide AS in several plants including Arabidopsis, rice, grape, and cucumber (Campbell et al., 2006; Wang and Brendel, 2006; Reddy, 2007; Baek et al., 2008; Barbazuk et al., 2008; Filichkin et al., 2010; Guo et al., 2010; Lu et al., 2010; Sanchez et al., 2010; Zenoni et al., 2010). A variety of splicing-sensitive microarrays such as splice junction arrays and tiling arrays, that are used extensively in animals (Hallegger et al., 2010) have not been widely used in plants to analyze AS globally (Love et al., 2010; Rehrauer et al., 2010; Zenoni et al., 2010). RNA sequencing (RNA-Seq) using NGS platforms has allowed the detection of rare transcripts, precise quantification of transcript levels and global analysis of AS (Pan et al., 2008; Wang et al., 2009a). A recent analysis of AS in plants using RNA-Seq has revealed that over 40% of intron-containing genes in Arabidopsis (Filichkin et al., 2010) and about 48% in rice (Lu et al., 2010) undergo AS, although it is not known how much of AS is due to noise in the splicing process (Melamud and Moult, 2009) and how much is regulated AS with biological consequences. In humans, pre-mRNAs from almost every multi-exon gene are alternatively spliced and misregulation of splicing results in developmental abnormalities and disease (Pan et al., 2008; Ru et al., 2008; Sanford et al., 2009). As more RNA-Seq data from different cell types, tissues, developmental stages and under different biotic and abiotic stresses become available, the known repertoire of AS in plants is likely to increase. In addition, many splice variants are differentially expressed in a tissue- or development-specific manner or in response to developmental cues and stresses (Yoshimura et al., 2002; Iida et al., 2004; Palusa et al., 2007; Reddy, 2007; Schindler et al., 2008; Simpson and Brown, 2008; Filichkin et al., 2010; Simpson et al., 2010; Staiger and Green, 2011).

In addition to AS events that include or delete long sequences in the pre-mRNA, other subtle AS events due to tandem acceptors (NAGNAG, N being any nucleotide) that result in gain or loss of three nucleotides in the spliced mRNA are common in land plants and animals (Iida et al., 2008; Schindler et al., 2008; Sinha et al., 2010). Primary transcripts of miRNAs also undergo AS (Hirsch et al., 2006; Szarzynska et al., 2009). In addition to cis-splicing, occurrence of trans-splicing, which produces chimeric transcripts by joining transcripts derived from two different nuclear protein-coding genes on the same or different chromosomes, has been reported in plants (Kawasaki et al., 1999; He et al., 2008; Guo et al., 2010). In rice, over 200 chimeric transcripts derived from trans-splicing were predicted from short-read sequence data and some of these were verified by RT-PCR (Zhang et al., 2010).

Comparative analysis of prevalence of different types of AS events in plants and animals has revealed that there are some fundamental differences between them. In plants a vast majority of splice variants (up to 56%) are due to intron retention, whereas it is not that prevalent in metazoans (5% in humans; Iida et al., 2004; Ner-Gaon et al., 2004; Wang and Brendel, 2006; Baek et al., 2008; Filichkin et al., 2010; Labadorf et al., 2010). In animals, exon skipping is the most common form of AS (58% in humans) and it is less prevalent in plants (8% in Arabidopsis). The differences in the frequencies of different types of AS events between plants and metazoans are thought to reflect the differences in how plant and animal cells recognize exons and introns. However, it is not known why intron retention is prevalent in plants and what mechanisms contribute to it. Interestingly, transcriptome analysis of 18 accessions of Arabidopsis thaliana has revealed that intron retention events differed between accessions (Gan et al., 2011).

The high rate of occurrence of AS in plants and its regulation by stresses and developmental cues (Kalyna et al., 2006; Palusa et al., 2007; Reddy, 2007; Barbazuk et al., 2008; Simpson et al., 2008, 2010; Barbazuk, 2010; Filichkin et al., 2010; Zenoni et al., 2010; Reddy and Ali, 2011) has sparked a growing interest and led to further studies focused on revealing the full extent of AS in plants by deep sequencing. Such studies will aid in understanding the biological functions of splice variants and the mechanisms by which plant cells regulate AS. This review focuses on computational tools used in predicting AS and splicing regulatory elements (SREs) involved in the regulation of AS. We also briefly discuss some recent experimental approaches that have been used in animals to identify targets of RNA binding proteins (RBPs), which can be applied to plants to discover RNA sequences that bind splicing regulators.

Transcriptome-Wide Detection and Visualization of AS

Before the advent of NGS, large-scale studies of AS in plant and mammalian systems were carried out mostly using sequences of Expressed Sequence Tags (ESTs) and full-length cDNAs (Haas et al., 2003; Campbell et al., 2006; Wang and Brendel, 2006; Chen et al., 2007; Gu and Guo, 2007; Ner-Gaon et al., 2007; Wang et al., 2008a; Sablok et al., 2011). These studies have increased dramatically the estimated number of plant genes that exhibit AS, and identified intron retention events as the most common AS event (Reddy, 2007; Wang et al., 2008a). The decline in the cost of sequencing using NGS platforms has made large-scale sequencing readily available, and transcriptome profiling using RNA-Seq has already been carried out in several plant species (Filichkin et al., 2010; Lu et al., 2010; Zenoni et al., 2010; Gan et al., 2011). However, analysis of these massive amounts of sequence data and the short length of sequence reads require novel computational tools, which has become a major bottleneck in mining the data to extract biologically relevant conclusions, especially for accurately predicting AS (Liang et al., 2009; Chodavarapu et al., 2010; Fiume et al., 2010; Marguerat and Bahler, 2010).

Transcripts vs splice graphs

Full-length cDNAs provide the best evidence for a gene’s splice forms – aligning such a sequence to the reference genome provides evidence for the exact exon-intron structure of a transcript. ESTs are shorter, but still usually cover several exons. NGS reads on the other hand, are short (around 100 bp in today’s technology), and only provide local evidence for transcript structure. This makes prediction of splice forms from RNA-Seq difficult, and most of the methods for transcriptome assembly first construct an object called a splice graph (Heber et al., 2002). A splice graph is a compact graphical representation of a gene’s exon-intron structure that captures all the ways in which exons for a given gene may be assembled into a transcript (Heber et al., 2002; Xing et al., 2004; Harrington and Bork, 2008; Sammeth et al., 2008; Bonizzoni et al., 2009; Labadorf et al., 2010; Richardson et al., 2011; Rogers et al., 2012). Figure 1 illustrates the concept. The compact structure allows researchers to visualize a gene’s AS easily (Harrington and Bork, 2008; Rogers et al., 2012), facilitates integration of ESTs into coherent models (Heber et al., 2002; Xing et al., 2004; Bonizzoni et al., 2009), aids statistical analysis of AS across a genome (Labadorf et al., 2010; Rogers et al., 2012), and facilitates comparisons between gene families (Richardson et al., 2011).

Figure 1

Splice forms for the ING2 gene from . A splice graph is a compact representation that shows all the ways in which a gene’s exons may be combined. These plots, generated by SpliceGrapher, use color coding to highlight AS events.

AS prediction using RNA-Seq

Transcriptional activity and AS can be studied using RNA-Seq with and without a reference genome (see Figure 2 for pipeline overview, and Table 1 for a list of tools). We begin our discussion with methods that require a reference genome. In this scenario, the first step is aligning the reads to the genome; we distinguish between two types of short-read mapping methods: those that allow only a limited number of gaps (usually a few bp at the most), and those that are able to map reads across splice junctions. Until a few years ago reads were short (32–36 bp) and most read mapping algorithms such as the Bowtie program (Trapnell and Salzberg, 2009) performed ungapped alignment. As read length continues to increase (100 bp and higher using today’s technology), the number of reads that span splice junctions increases as well, and with it the number of programs that perform spliced alignment (see, e.g., Trapnell et al., 2009; Au et al., 2010; Jean et al., 2010; Wang et al., 2010). Once mapped, read coverage (the distribution of reads that align within a region of interest) then provides evidence for exons and splice junctions recapitulated in the RNA-Seq data (see Figure 3).

Figure 2

Pipelines for prediction and quantification of splice forms from RNA-Seq data. Methods such as trans-ABySS and Trinity perform de novo prediction, and do not require a reference genome (left); these methods do not rely on alignment programs to map reads to a genome. When a reference genome is available, methods such as Cufflinks first map the reads to the genome, followed by a step of assembly of splice forms or their quantification (right). Some of these methods require, or can use annotated isoforms to guide the process.

Table 1

Tools for predicting isoforms, their expression, and alternative splicing from RNA-Seq data.

Method	Task	Input data	Notes
Trans-ABySS (Robertson et al., 2010)	IP, IE	De novo	Requires Abyss contigs
Trinity (Grabherr et al., 2011)	IP, IE	De novo
Rnnotator (Martin et al., 2010)	IP	De novo
Scripture (Guttman et al., 2010)	IP	G
IsoLasso (Li et al., 2011)	IP, IE	G	Improved version of IsoInfer (Feng et al., 2011)
NSMAP (Xia et al., 2011)	IP, IE	G
Cufflinks (Trapnell et al., 2010)	IP, IE	G, A	Annotated isoforms are optional
TAU (Filichkin et al., 2010)	IP	G,A	Annotated isoforms are optional; does not scale well with read length
SpliceGrapher (Rogers et al., 2012)	SG	G, A
IsoEM (Nicolae et al., 2010)	IE	G, A
IsoformEX (Kim et al., 2011)	IE	G, A
SpliceTrap (Wu et al., 2011)	IE	G, A	Only handles exon skipping
NEUMA (Lee et al., 2011)	IE	G, A
Solas (Richard et al., 2010)	IE	G, A
rSeq (Jiang and Wong, 2009)	IE	G, A
RSEM (Li et al., 2010; Li and Dewey, 2011)	IE	De novo	Requires a transcriptome assembler

The tools vary in the specific task they address; we distinguish between several tasks: isoform prediction (IP), isoform expression (IE) and splice graph prediction (SG). The tools also vary in the input data they require: de novo (no input required except for the RNA-Seq data), a reference genome (G) or annotated isoforms (A).

Figure 3

Read coverage for the gene SCL33 in . The top panel shows the annotated gene model; the middle panel shows reads that map across splice junctions, with labels showing the number of reads that aligned across each junction and novel splice junctions highlighted in green. The bottom panel shows the distribution of reads across the gene. Here the read depth ranges from 1 to over 300, demonstrating that read coverage can be highly variable even across known exons (shaded regions on the graphs). This variability can make it difficult to distinguish between weakly expressed splice forms and background noise. This figure was generated by SpliceGrapher.

The first studies that predicted AS from RNA-Seq data were performed in mammals and focused on detection of exon skipping, the most common form of AS in these systems (see, e.g., Mortazavi et al., 2008; Pan et al., 2008; Sultan et al., 2008; Wang et al., 2008b; Tang et al., 2009); detection of exon skipping is relatively simple, requiring the detection of reads that span a splice junction that skips a known exon. Several programs are now available for assembling transcripts using ungapped and spliced alignments among them Cufflinks (Trapnell et al., 2009), and Scripture (Guttman et al., 2010). Whereas these programs were designed for mammalian genomes, TAU (Filichkin et al., 2010) was designed to predict transcripts for the model plant A. thaliana. The program predicted an assortment of non-canonical splice junctions associated with splice forms that were later validated via RT-PCR (Filichkin et al., 2010). However, its spliced alignment component has memory requirements that make it prohibitive with increasing read length.

Some methods simultaneously predict transcripts and their abundance, using read depth as information that can be used to untangle transcripts from each other. They are based on the idea that read depth can be expressed as a weighted sum across the transcripts that are represented in a sample. These include IsoLasso (Li et al., 2011) and NSMAP (Xia et al., 2011).

RNA-Seq data alone may not be sufficient to resolve splice forms unambiguously (Lacroix et al., 2008; Rogers et al., 2012). Accordingly, Cufflinks was recently enhanced to incorporate information from gene models into its transcript prediction method (Roberts et al., 2011); the TAU program is also able to leverage gene annotations. We developed the SpliceGrapher tool, which uses gene models to establish a context for interpreting evidence from EST or RNA-Seq alignments, and to predict novel AS events only when the evidence for them is strong (Rogers et al., 2012). Furthermore, our tool leaves the predictions in the form of splice graphs when transcripts cannot be unambiguously resolved. Our work shows the strength of this approach and the tradeoffs associated with using gene annotations to augment TAU or Cufflinks predictions.

Accurate splice junction identification is crucial for making accurate AS predictions, but the short length of NGS reads makes spliced alignment especially challenging. A splice junction may occur anywhere within a read, so the read may have just a few bases on one side of a junction. Therefore, methods that use simple heuristics such as the existence of canonical splice site dimers and acceptable intron lengths can lead to many false-positive splice junctions (Rogers et al., 2012). This highlights the importance of modeling splice junction sequence characteristics as implemented in MapSplice (Wang et al., 2010), for example, or the sophisticated sequence-based splice site models used by SpliceGrapher (Rogers et al., 2012) and PALMapper (Jean et al., 2010).

Not all plants have a reference genome available. In its absence, de novo transcriptome assembly packages can be used. These methods construct transcripts based on overlapping k-mers. One such program is ABySS (Simpson et al., 2009), a de novo genome assembler that has been used to predict AS patterns such as exon skipping, intron retention, and alternative 5′ splice sites (Birol et al., 2009). The recently developed Trinity suite (Grabherr et al., 2011) assembles reads into splice graphs, and predicts splice forms by tracing paths through the graph.

Tiling and exon-junction arrays are an alternative platform for studying AS (Clark et al., 2002; Johnson et al., 2003; Mockler and Ecker, 2005; Cuperlovic-Culf et al., 2006; Hallegger et al., 2010). Tiling arrays provide an unbiased view of a genome using probes evenly spaced across its entire length, thus permitting discovery of novel AS events (Ner-Gaon and Fluhr, 2006; Hazen et al., 2009). Ner-Gaon and Fluhr (2006) found that whole-genome tiling arrays identified as much AS activity in A. thaliana as the most comprehensive EST-based studies. However, microarrays suffer from several limitations such as cross-hybridization and varying probe binding affinities that increase hybridization signal noise, and poor sensitivity for detecting isoforms that differ only in a few nucleotides, and also in cases where minor isoforms are expressed at low levels (Mockler and Ecker, 2005; Hallegger et al., 2010). These factors, combined with the success of RNA-Seq, have impeded adoption of this technology by plant researchers.

Visualization of NGS data is now supported by several genome browsers, including Stein et al. (2002), the Integrated Genome Browser (IGB; Nicol et al., 2009), the Integrative Genomics Viewer (Robinson et al., 2011), and GenomeView (Abeel, 2011), that can display short-read coverage graphs and transcripts. SpliceGrapher (Rogers et al., 2012), designed specifically for AS analysis, includes a plotting tool that displays splice graphs, gene models, and short-read coverage graphs, and highlights AS events. It generates static plots that cannot be manipulated the way a genome browser visualization can, but we have found them to be very useful for viewing of plant transcriptome data, because of the relatively short intron length.

Differential AS

Detection of differentially expressed genes is perhaps the most common analysis task performed on microarray data, and many methods are available (Grant et al., 2007; Durinck, 2008). NGS technology and the availability of microarrays that allow a distinction between splice forms (e.g., exon-junction and tiling arrays) are opening the possibility of extending this idea to detection of differential AS: genes that show differences in the patterns of AS under different conditions or developmental stages. Several studies have revealed regulated AS in plants (Yoshimura et al., 2002; Iida et al., 2004; Palusa et al., 2007; Reddy, 2007; Schindler et al., 2008; Simpson and Brown, 2008; Filichkin et al., 2010; Simpson et al., 2010; Kumar et al., 2011; Staiger and Green, 2011), and detection of differential AS by high-throughput methods will further enhance our understanding of the role of AS, and help in developing a plant condition-dependent splicing code.

A number of studies have used microarrays to estimate isoform expression levels (Mockler and Ecker, 2005; Blencowe et al., 2009; Hallegger et al., 2010), and methods have been proposed to detect splice form expression levels from exon arrays (Purdom et al., 2008; Xing et al., 2008). These methods have been used to study AS in mammals (Xing and Lee, 2008; Warzecha et al., 2009), but have not been widely adopted in plants (Love et al., 2010; Rehrauer et al., 2010). Microarrays have several limitations in this context including a large number of false-positive AS predictions (Gaidatzis et al., 2009).

NGS data address many of the limitations of microarrays – they are more sensitive to weakly expressed isoforms and have a broader dynamic range (Roy et al., 2011) – but they also introduce considerable challenges. For example, it may not be possible to find a unique mapping between reads and the reference genome (Lacroix et al., 2008). Additionally, read coverage can be highly variable due to a variety of factors such as library preparation methods, flow cell characteristics and reads that align to multiple locations (Mortazavi et al., 2008; Fang and Cui, 2011).

Detection of differential AS requires establishing accurate models for the splice forms represented in the data (discussed earlier), quantifying splice form expression in a way that allows detection of weakly expressed splice forms, and performing statistical tests to differentiate between the relative expression of splice forms across samples. Several measures are used to report expression levels based on short-read coverage. A common metric is RPKM (reads per kilobase per million reads), developed in (Mortazavi et al., 2008). Many methods report expression levels in RPKM (Mortazavi et al., 2008; Jiang and Wong, 2009; Guttman et al., 2010; Feng et al., 2011; Kim et al., 2011; Xia et al., 2011) or an equivalent measure such as FPKM (fragments per kilobase per million reads; Nicolae et al., 2010; Trapnell et al., 2010). When comparing expression levels between genes or isoforms, RPKM may bias estimates in favor of longer sequences (Costa et al., 2010), so other measures have been proposed (see, e.g., Lee et al., 2011). Comparisons with microarrays have confirmed that RNA-Seq data can provide at least as much sensitivity as microarrays (Mortazavi et al., 2008), but without the limitations previously mentioned. Several studies have validated RNA-Seq expression estimates using qRT-PCR and these have demonstrated that the two methods produce estimates that are consistent with one another (see, e.g., Wang et al., 2008b; Nicolae et al., 2010; Richard et al., 2010; Lee et al., 2011).

Recently software has been developed to measure differential AS from RNA-Seq. For example, Cuffdiff, an extension of the Cufflinks package, compares transcript expression levels on the basis of read coverage in two experiments (Trapnell et al., 2010). The transcripts may be generated from Cufflinks or may come from annotated gene models. Cuffdiff generates statistical test results for fold changes at either the transcript or gene level. Studies in mammals have used these tools to compare differentially expressed splice forms between different conditions. The authors of the Cufflinks software used it to study transcriptional changes in mouse myoblast cell lines. They identified 70 genes in which the prevalent splice forms changed as cells transitioned from myocyte production to myotube fusion (Trapnell et al., 2010). In another study, Twine et al. (2011) found statistically significant differences in the expression of alternative splice forms between normal and diseased brain cells that yielded insights into the progression of Alzheimer’s disease.

Few plant studies have applied differential AS tools. Differential AS in response to stresses has been reported in A. thaliana (Filichkin et al., 2010). In addition, Jiao and Meyerowitz (2010) investigated flower development in A. thaliana, in part using the program rSeq (Jiang and Wong, 2009) to estimate isoform expression levels. Their analysis revealed differentially expressed splice forms for three genes (APETALA1, APETALA3, and AGAMOUS) between two flowering stages. Analysis of intron retention events showed that they occurred more frequently with the U12 spliceosomes than with the U2 spliceosomes (Jiao and Meyerowitz, 2010).

These results underscore the potential for using RNA-Seq to probe AS under changing conditions such as organism development or stresses. With NGS, researchers are no longer restricted by the cost of ESTs or a dearth of plant-specific microarrays. In addition, a growing number of software tools are helping to automate the analysis of transcript expression from NGS data. The availability of some NGS analysis pipelines through the iPlant cyber infrastructure is going to help plant researchers who do not have the expertise or the required computational infrastructure to analyze their RNA-Seq data (Goff et al., 2011). NGS data analysis is a focus area for iPlant, and several methods are available, including Cufflinks, and several read mapping algorithms; efforts are underway to add the SpliceGrapher package to the iPlant infrastructure. These factors combined, provide unprecedented opportunity for researchers to explore the role of AS across all plant species.

Regulation of Splicing

An important question in AS is regulation of splice site choice. The splicing code, i.e., the set of biological rules for determining the splicing outcomes in both constitutive and alternative splicing, is only beginning to be addressed in animals (Barash et al., 2010a) and very little is known about it in plants.

Gene architecture and composition in pre-mRNA splicing

Comparative genomics studies on gene structure have revealed major differences in the architecture of plant and animal genes (Reddy, 2007). Plant genes are generally shorter than animal genes with fewer exons and significantly shorter introns. In animals, the average size of exons is about 140 nucleotides, whereas introns are several thousands of nucleotides long (average 3000 nucleotides in humans). In contrast, plant exons and introns are about the same length (in Arabidopsis for example, the average length of exons, and introns is 173 and 172 nucleotides, respectively) or about twice the size of exons in rice (e.g., exons 193 and introns 433 nucleotides; Reddy, 2007; Baek et al., 2008). Based on the smaller size of plant introns relative to animal introns and the high number of intron retention events, it is thought that the primary mode of splicing in plants is accomplished via an intron definition mechanism. However, there are plants whose introns are longer, but still have a high rate of intron retention (e.g., grape, with intron length of around 970, computed using the current genome annotations), and animals with short introns and exon skipping rates characteristic of mammalian systems (e.g., Tetraodon nigroviridis (puffer fish) which has introns of length 600 bp on average).

Plant introns are rich in U and UA nucleotides and exons are G-rich (Goodall and Filipowicz, 1989). Several studies indicate that cis-elements involved in intron recognition in plants are likely to be different from those in yeast and animals (Reviewed in Reddy, 2001; Schuler, 2008). For instance, animal introns are not accurately spliced in plants and vice versa. High U or UA content in introns was found to be important for splice site recognition and efficient splicing of U2 type and U12 type introns (Goodall and Filipowicz, 1989; Lambermon et al., 2000; Lorkovic et al., 2000; Reddy, 2001; Lewandowska et al., 2004; Simpson et al., 2004), suggesting the presence of proteins that interact with U or UA rich elements. In tobacco, three proteins (UBP1, RBP45, RBP47) that bind to intronic sequences have been characterized. UBP1 has been shown to be necessary for efficient splicing of pre-mRNA as well as mRNA accumulation (Goodall and Filipowicz, 1989; Lambermon et al., 2000). An analysis of AS in Medicago, poplar, Arabidopsis and rice, revealed that except for rice, AS was most prevalent for introns with decreased UA content (Baek et al., 2008). In addition to intronic sequences, exonic purine-rich sequences have been shown to influence splice site choice (Egoavil et al., 1997; McCullough and Schuler, 1997; Lewandowska et al., 2004; Schuler, 2008). However, aside from broad surveys of gene architecture, composition, and mutational analysis of splice sites, there has been little work performed to uncover the putative cis-elements that would provide substantial computational evidence for the intron definition model as the primary means by which intron-retaining transcripts are generated.

Cis-elements in pre-mRNAs that control splicing

In metazoans, four core sequence elements at the exon/intron boundaries and the intron/exon boundaries are necessary for splice site recognition by the spliceosome. These include (i) a motif at the 5′ splice site (SS) or donor site with a conserved GU dinucleotide, (ii) another motif at the 3′ SS or acceptor site with a conserved AG dinucleotide, (iii) a stretch of pyrimidines (polypyrimidine tract) upstream of the 3′ SS and (iv) a branch point 17–40 nucleotides upstream of the polypyrimidine tract. Components of the spliceosome recognize these core signals. U1 snRNP recognizes the 5′ SS, U2AF35 and U2AF65 recognize the 3′ SS and polypyrimidine tract, respectively and U2 snRNP recognizes the branch point. The 5′ and 3′ SSs are very similar between plants and animals, and the polypyrimidine tract in plants is rich in Us (Reddy, 2007). Several mutants in which splice sites are affected have been isolated and consequences of these mutations have been reviewed extensively (Brown, 1996; Schuler, 2008). Although these core elements are conserved across species, they are very short and they alone, are not sufficient to define exons and introns and recruit the splicing machinery. Additional sequences in exons and introns, which are collectively referred to as splicing regulatory elements (SREs), are important for constitutive splicing as well as AS. The SREs function as either splicing enhancers or suppressors and affect splice site choice. The efficiency by which the spliceosome recognizes exons and introns is in part determined by numerous protein factors that recognize these SREs. Depending on the location of SREs and their effect on splicing, they are grouped into four classes: exonic splicing enhancers (ESEs), exonic splicing silencers (ESSs), intronic splicing enhancers (ISEs), and intronic splicing silencers (ISSs; Black, 2003; Chasin, 2007; Wang and Burge, 2008; Wang et al., 2009b).

Computational studies to predict SREs

In animals systems, many studies have been performed to predict SREs. Most of these studies focused on exon skipping as this is the most prevalent AS event (Fairbrother et al., 2002; Sorek et al., 2004; Dror et al., 2005; Ohler et al., 2005; Ratsch et al., 2005; Yeo et al., 2005; Barash et al., 2010a,b) and some on tandem 3′ splice sites (Akerman and Mandel-Gutfreund, 2006; Xia et al., 2006). Computational tools to predict ESEs such as RESCUE-ESE (Relative Enhancer and Silencer Classification by Unanimous Enrichment) have been used to predict hexameric ESEs in animals (Fairbrother et al., 2002). Several other programs have been developed to identify splicing elements or to simulate the splicing of primary transcripts (Schwartz et al., 2009a). These methods are based on the rationale that alternatively spliced exons may contain sequence elements that are absent in constitutively spliced exons and introns or vice versa. Analysis of sequences that are over or under represented in alternatively spliced exons compared to constitutively spliced exons has led to prediction of sequences that could contribute to AS. Experimental validation of some of these has led to identification of many cis regulatory elements involved in regulating splice site choice either by enhancing or silencing the AS of a particular exon. These studies have shown that cis-elements involved in splicing can exist predominantly in exons and also in introns in some cases (Blencowe, 2000; Chasin, 2007). The theme that emerged from these analyses is that most SREs often share common features: they are short (about 6–10 nucleotides) and loosely conserved between different targets of the same protein, suggesting a flexible nature of the RNA-protein interaction. These motifs, that are individually weak but present in multiple copies, act as binding sites for competing trans factors (Ladd and Cooper, 2002). The most commonly studied elements are ESEs and the SR family of splicing regulators that bind them. It is now clear that many, if not all exons, constitutively or alternatively spliced, harbor ESEs that bind to SR proteins (Black, 2003; Chasin, 2007; Barash et al., 2010a). By contrast, ESS elements are usually bound by hnRNP proteins and can repress interactions across exons and introns that participate in spliceosome assembly (Blencowe, 2006; Chasin, 2007).

A recent compendium of cis-elements was used in the development of a mammalian “splicing code.” It included 171 known motifs and 326 new motifs that were used to predict splicing patterns of exons (Barash et al., 2010a). While most research has focused on discovering these elements within mammalian species, some progress has been made in identifying ESE motifs in Arabidopsis (Pertea et al., 2007). The goal of the study was to improve splice site prediction accuracy in Arabidopsis by incorporating potential ESE motifs into splice site recognition programs. The authors constructed a dataset (ESEAra) comprised of around 4000 high quality Arabidopsis gene models containing close to 17,500 coding exons. They extracted 50 bp regions flanking the boundaries of internal and terminal exons in the ESEAra dataset, established frequency distributions of hexamers near weak and strong splice sites, and used a scoring threshold to identify 84 putative ESE hexamers in these regions. Of these 84 ESE hexamers, 35 (12 at the 5′ end, 6 at the 3′ end, and 17 at both ends) had experimental evidence for ESE activity in Arabidopsis (Pertea et al., 2007). Several of the detected motifs exhibit the GAAGAA hexamer, which is a part of a recognized human ESE.

Aside from the above study, our catalog of plant cis regulatory elements involved in AS is limited. Over two decades ago, it was shown that AU-rich sequence elements in plant introns are required for plant pre-mRNA splicing (Goodall and Filipowicz, 1989; Filipowicz et al., 1995). Since then, our knowledge of trans factor binding sites has been constrained to putative or few experimentally defined regulatory motifs. Yoshimura et al. (2002) experimentally identified an AU-rich splicing regulatory element (GU[G|C|A]UUGC[C|U]UAUUUGAAUUGCAG) located in an exon/intron boundary responsible for tissue-specific AS of tobacco chloroplast ascorbate peroxidase (chlAPX) pre-mRNA. Four isoforms are produced via AS of the 3′-terminal region of chlAPX pre-mRNA (tAPX-I, sAPX-I, -II, and -III) due to alternative excision of intron 11 and intron 12. Subsequent site directed mutagenesis of the cis regulatory element and RNA gel mobility shift assays confirmed interaction with a particular trans factor expressed in leaves but not roots. The identity of this trans factor remains unknown. Using two 9G8 (one of the mammalian SR proteins) binding sequences in gel shift assays, it was shown that AtRSZ22 also binds to those sequences (Lopato et al., 1999). AtGRP7 (Arabidopsis thaliana glycine-rich RNA binding protein) regulates AS of its own pre-mRNA as well as its paralog, AtGRP8, pre-mRNA by producing a PTC-containing splice variant when its levels increase. Similarly, AtGRP8 autoregulates its AS and AtGRP7 pre-mRNA AS (Schoning et al., 2007, 2008). This regulation involves binding of AtGRP7 to its pre-mRNA and the pre-mRNA of AtGRP8 and vice versa (Schoning et al., 2007, 2008). The sequences in these pre-mRNAs that bind to AtGRP7 and AtGRP8 have been identified (Schoning et al., 2007, 2008). Although there are several other instances of auto- and cross-regulation of AS of several spliceosomal proteins, cis-elements involved in regulation of AS are not known (Kalyna et al., 2003; Reddy, 2007; Barta et al., 2008; Stauffer et al., 2010; Reddy and Ali, 2011).

Because intron retention is so pervasive in plants, uncovering the sequence elements that lead to this splice form would be a valuable breakthrough, as there are fundamental implications for post-transcriptional regulation of gene expression either by NMD/RUST (Lewis et al., 2003), or by generating new target sites for miRNA binding. There is evidence that generation of premature termination codon in plants by AS leads to degradation of mRNAs (Arciga-Reyes et al., 2006; Schoning et al., 2008; Kurihara et al., 2009; Barbazuk, 2010; Palusa and Reddy, 2010; Staiger and Green, 2011). In humans, it has been predicted that retained introns in the 3′ UTR can create and/or increase putative miRNA targets (Tan et al., 2007). Analysis of all known retained introns and their flanking exons can be performed computationally to determine what sequence elements contribute to intron retention. Building a set of potential regulatory sequences would lend predictive power to discern the likelihood that a given gene will produce an intron-retaining transcript.

Validation of computationally predicted SREs

In animals, in vitro and in vivo splicing assays with pre-mRNAs containing wild type and mutated putative SREs have been used to validate predictions. Unfortunately, in vitro methods that employ S100 or nuclear extracts for splicing assays for identifying SREs cannot be applied to plants, as there is no plant-derived in vitro splicing assay system. However, the validity of computationally predicted plant cis-elements in splicing regulation can be tested using two different approaches. In one approach, one can generate mutations in the predicted cis-element and compare the splicing of wild type to mutated gene in a transient or stable expression system, which will allow analysis of splicing in its natural exon/intron context. Alteration of predicted cis-elements can be done in a high-throughput manner using the strategy described in Figure 4. Splicing of the wild type and mutated gene can be first tested in protoplasts by RT-PCR using a forward primer corresponding to the tag and a reverse gene-specific primer. If the candidate gene is not expressed in mesophyll cells then it can be analyzed in transgenic lines. If there is an SRE then one can test the importance of individual bases in that element by site directed mutagenesis. If there are two cis-elements in a gene that are complementary to each other and have the potential to form a stem/stem-loop structure, then a change in the sequence of one of the two elements or both to disrupt base pairing should affect splicing. To confirm that the base pairing in the two cis-elements, not the sequence itself, is necessary, one can mutate both elements in such a way that the sequence is changed but the two elements are complementary to each other. If this still shows regulated splicing then it is likely that the base pairing is involved in regulated splicing.

Figure 4

Approach to mutate a predicted . In the first PCR the target gene with two primers sets (F1/R1 and F2/R2). The F1 and R1 primer set amplifies the gene from the initiation codon to the predicted cis-element and F2 and R2 will amplify from the predicted cis-element to the stop codon. Primers F1 and R2, in addition to gene-specific sequence (shown in green), will be tailed with sequences complementary to Gateway vector primers (shown in dark yellow). Similarly, primers R1 and F2, in addition to gene-specific sequence, will be tailed with the changed sequence in the predicted cis-element (shown in blue). In the second PCR, the two gene fragments from the first PCR will be mixed. This overlapping template will be amplified using primers complementary to primers F1 and R2 tailed with the attB1 and attB2 Gateway sequences, which can then be cloned into a Gateway donor vector and into a plant transformation vector with a tag as a fusion to the N-terminus. The wild type gene will be cloned in a similar fashion except that only one PCR will be done with the F1/R2 primer set containing the entire attB1 and attB2 Gateway sequences.

A second approach for validating cis-elements in an intron is to use a reporter gene (GFP) that is interrupted by a test intron that contains predicted cis-elements. By following the gene’s splicing pattern, one can determine if the signals are present exclusively in the inserted part of the gene. A similar approach could be used to study cis-elements in other regions of a gene. Identified SREs can then be further analyzed by using site directed mutagenesis.

In vivo analysis of pre-mRNA splicing by expressing splicing reporters in transient assays (e.g., protoplasts or leaf transfections) have not been used widely (Gniadkowski et al., 1996; Lambermon et al., 2002; Lewandowska et al., 2004; Isshiki et al., 2006; Schuler, 2008; Stauffer et al., 2010). The lack of a plant-derived in vitro splicing system makes transient assay systems very attractive for studying the constitutive splicing and AS of pre-mRNAs for genes that are expressed in leaves.

Experimental approaches for transcriptome-wide identification of SREs

RNA binding proteins (RBPs) with characteristic RNA binding motifs such as the RNA recognition motif (RRM) and the K-homology (KH) domain that interact with specific RNAs, profoundly impact gene expression at various levels (transcription, capping, splicing, polyadenylation, biogenesis of miRNAs and siRNAs, RNA transport, localization and degradation, small RNA regulated gene expression) and have been shown to play important roles in development and disease in animals (Licatalosi and Darnell, 2010). In plants, a large repertoire of RNA binding proteins have been identified using bioinformatic tools. In Arabidopsis there are more than two hundred RNA binding proteins (RBPs), including 200 RRM containing RBPs and 30 KH domain proteins (Lorkovic, 2009). Many plant RBPs are unique to plants, suggesting that they are likely to have novel RNA targets and perform plant-specific functions. A large fraction of RBPs are implicated in pre-mRNA splicing (Reddy, 2007; Simpson et al., 2008; Lorkovic, 2009; Reddy and Ali, 2011; Staiger and Green, 2011). Interestingly, many pre-mRNAs encodings RBPs (e.g., pre-mRNAs of SR and SR-like proteins, glycine-rich RNA binding proteins, and other spliceosomal proteins) are extensively alternatively spliced in plants and there is tight regulation of AS of these genes (Isshiki et al., 2006; Palusa et al., 2007; Barta et al., 2008; Simpson et al., 2010; Stauffer et al., 2010; Staiger and Green, 2011). Protein-RNA interactions are crucial in many aspects of RNA metabolism. These interactions are dependent on RNA sequence elements in RNAs that interact specifically with proteins. Although the in vivo RNA targets of most plant RBPs are unknown, genetic studies with some of the RBPs indicate a crucial role for these in different developmental processes (flowering time, flower development, circadian responses) and various biotic and abiotic stress-signaling pathways (Reddy, 2007; Simpson et al., 2008; Lorkovic, 2009; Staiger and Green, 2011).

RNA sequences that bind to RBPs are generally analyzed using various methods. These include RNA immunoprecipitation (RIP), which involves co-immunoprecipitation of RNAs with RBPs without prior cross-linking to the RNA using an antibody to a specific RBP followed by sequencing of the precipitated RNAs (RIP-seq) or probing microarrays with precipitated RNA (RIP-chip; Brown et al., 2001; Tenenbaum et al., 2002; Barkan, 2009; Wang et al., 2009c). The caveats of this approach include loss of true interactions because the transient and dynamic nature of RNA-protein interactions, thereby precluding co-IP of some RNAs, and at the same time, high noise levels due to the sticky nature of RNAs (Mili and Steitz, 2004). Moreover, the binding RNA sequences represent both direct and indirect targets as most RBPs interact with other proteins and form complexes (Reddy, 2007). Advances in efficient ways to UV crosslink RNAs with RBPs in vivo followed by immunoprecipitation of complexes with an RBP specific antibody (CLIP, Cross-Linking ImmunoPrecipitation) under stringent conditions and sequencing of RNA using high-throughput RNA sequencing (HITS-CLIP) have paved the way to map, in an unbiased manner, transcriptome-wide direct RNA targets of an RBP (Ule et al., 2003, 2005, 2006; Wang et al., 2009b; Darnell, 2010; Licatalosi and Darnell, 2010). In HITS-CLIP, cells/tissues are exposed to ultraviolet (UV) radiation, which generates covalent bonds between RNA and proteins that are in close contact (Ule et al., 2003, 2005). Because of the covalent bonds, RNA-protein complexes can be purified under stringent conditions. A modified version of HITS-CLIP called Photoactivatable-Ribonucleoside-Enhanced Cross-linking and Immunoprecipitation (PAR-CLIP) enhances cross-linking efficiency and allows identification of the locations in the RNA that are involved in interacting with RBP (Hafner et al., 2010a). In PAR-CLIP, a photoreactive ribonucleoside analog, 4-thiouridine, is incorporated into nascent RNAs, which are then cross-linked to RBPs by exposing cells to UV radiation. RNAs that are complexed with a particular RBP are then immunoprecipitated using an antibody to that RBP. RNAs in immunoprecipitated samples are then converted into cDNA and sequenced using NGS platforms. Since the cross-linked RNA region shows T to C mutations in the sequenced cDNA, the binding region in RNA can be precisely identified (Hafner et al., 2010a).

An in-depth discussion on advantages and disadvantages of these methods is available in recent reviews (Barkan, 2009; Darnell, 2010; Hafner et al., 2010b). These methods could be applied to plants to create a transcriptome-wide map of RNA binding sequences that bind to a given RBP. These tools, developed with animal systems, are fairly recent and have the potential to uncover plant SREs, especially those involved in AS events that are prevalent in plants as compared to animals. These procedures can identify the binding site landscape or “RNA map” for plant SR proteins, one of the key regulators of pre-mRNA splicing and other RBPs. Incorporating the results from these studies into computational predictions should lead to a better understanding of the splicing code in plants.

RNA secondary structure and other properties that affect AS

In addition to the presence of SREs, there are several properties of pre-mRNA transcripts that were shown to affect AS. These include intron length, GC-content, splice site strength, and pre-mRNA secondary structure (Ladd and Cooper, 2002; Chasin, 2007; Shepard and Hertel, 2008; Schwartz et al., 2009b; Labadorf et al., 2010). A comparison of retained and constitutively spliced introns identified using Arabidopsis tiling arrays revealed that retained introns tended to be shorter and to have higher GC-content than constitutively spliced introns (Ner-Gaon and Fluhr, 2006). Similar characteristics were also observed for introns in rice (Zhang et al., 2010).

Sequence signals surrounding the 5′ and 3′ splice site junctions, polypyrimidine tract (PPT), and branch sites also impact splice site selection. For instance, in humans and plants it has been shown that splice sites flanking retained introns contain weaker signals than those flanking constitutively spliced introns (Kurmangaliyev and Gelfand, 2008; Labadorf et al., 2010). Studies on PPT and branch sites (Brown, 1996; Tolstrup et al., 1997) in plants suggest that their metazoan equivalents are more pronounced. However, their importance in correctly splicing pre-mRNA was shown by Simpson et al. (2002). This evidence suggests a need to generate a catalog for all AS variants centered at these signal-rich regions.

Cis regulatory elements in pre-mRNA secondary structure are known to regulate AS by sometimes affecting the recruitment of SR proteins (Buratti and Baralle, 2004; Buratti et al., 2004). Formation of secondary structure between complementary repeat-pair sequences in two introns flanking a cassette exon is proposed to mask splice sites and cause exon skipping events (Lian and Garner, 2005). A similar mechanism where complementary sequences in two exons can base pair across an intron may lead to intron retention. There are only a few examples in plants where RNA secondary structure has been shown to regulate splicing. In photosynthetic eukaryotes, AS of pre-mRNAs from genes encoding proteins involved in thiamine metabolism (THIC and THI4 in Chlamydomonas and THIC in Arabidopsis) is regulated by a thiamin pyrophosphate (TPP) binding riboswitch (Bocobza et al., 2007; Croft et al., 2007; Wachter et al., 2007). High levels of TPP result in its binding to an RNA aptamer in the pre-mRNA, which leads to production of splice variants with a PTC or long 3′UTR that are unstable (Bocobza and Aharoni, 2008; Wachter, 2010). In plants, the transcription factor IIIA (TFIIIA), which is necessary for transcription of 5S RNA, is alternatively spliced by inclusion or exclusion of its third exon; the splice variant with the included exon contains a PTC and is a target of NMD (Barbazuk, 2010). This exon is thought to be derived by exonization of 5S RNA and is conserved from moss to flowering plants, but not in green algae and metazoans (Fu et al., 2009; Hammond et al., 2009). The 5S RNA-like element assumes a 5S RNA structure, which regulates AS of its own pre-mRNA. The 5S RNA-like structure binds ribosomal protein L5, which in turn promotes exon skipping and generation of a functional transcript, whereas in the absence of L5, this exon is included, and results in the production of a PTC-containing unstable transcript (Hammond et al., 2009), suggesting a tight regulation of functional TFIIIA mRNA by the L5 protein.

Regulation of Alternative Splicing by Chromatin Organization

Apart from the features mentioned above, several recent studies point to epigenetic regulation of AS. Access to DNA may be affected by chromatin organization or methylation, which could impact the rate at which a gene is transcribed thereby affecting AS (Luco et al., 2011). In animals, it is now well established that the recruitment of splicing factors to pre-mRNAs occurs co-transcriptionally by the RNA Pol II carboxy-terminal domain, but the completion of splicing may occur co-transcriptionally or post-transcriptionally (Pandya-Jones and Black, 2009; Luco et al., 2011). Furthermore, the rate of elongation of transcription regulates AS, where rapid elongation favors recruitment of the splicing machinery to strong splice sites, thereby including exons with strong splice sites; whereas slow elongation allows recruitment of spliceosome to weak sites (Luco et al., 2011).

Some studies implicate a more direct role for chromatin remodeling enzymes. For instance, histone deacetylases in yeast and humans interact directly with U2 snRNP (Gunderson and Johnson, 2009) and the histone methyltransferase with U1 snRNP-specific protein U1C (Ohkura et al., 2005) and alter splicing patterns. Furthermore, chromatin remodelers SWI/SNF in yeast and humans (e.g., Braham) also regulate AS by interacting with spliceosomal proteins and recruiting snRNPs (Batsche et al., 2006; Tyagi et al., 2009). In Arabidopsis, mutations in protein arginine methyl transferase 5 (PRMT5), which methylates arginine residues in histones and Sm spliceosomal proteins, impair the circadian rhythm (Sanchez et al., 2010). It has been shown that AS of core-clock regulated genes (e.g., pseudo response regulator 9) and other pre-mRNAs is altered in the prmt5 mutants (Sanchez et al., 2010). Whether the observed effects in prmt5 mutants are due to altered methylation of Sm proteins and/or due to epigenetic defects remains to be studied.

Genome-wide nucleosome positioning and methylation studies in Arabidopsis and humans revealed that DNA associated with nucleosomes is more highly methylated than the flanking DNA (Nahkuri et al., 2009; Chodavarapu et al., 2010). Furthermore, nucleosomes are enriched in exons, particularly at exon-intron and intron-exon boundaries. Included exons and exons with weak splice sites are highly enriched in nucleosomes (Schwartz et al., 2009c; Spies et al., 2009; Tilgner et al., 2009). These studies suggest a role for nucleosome positioning in defining exons as well as regulating splicing (Nahkuri et al., 2009; Schwartz et al., 2009c; Spies et al., 2009; Tilgner et al., 2009; Chodavarapu et al., 2010). Small interfering RNAs (siRNAs) trigger transcriptional gene silencing by inducing heterochromatin formation in a sequence specific manner. Targeting of siRNA to intronic or exonic sequences that are close to an alternative exon has been shown to regulate the splicing of that exon (Allo et al., 2009), suggesting a role for siRNAs in AS.

Conclusion

Elucidation of multiple layers of gene regulation is critical to understand how plants grow, differentiate and respond appropriately to their environment. Regulated pre-mRNA splicing is emerging as an important layer in gene regulation. Extensive studies aimed at identifying features that control splicing in animal cells suggest that the combination of multiple characteristics in pre-mRNAs, including loosely conserved cis-elements and/or secondary structure(s) in transcripts, chromatin modification, and the rate of transcription regulate splicing (Barash et al., 2010a; Schor et al., 2010; Luco et al., 2011). Thus far, our knowledge of AS regulation in plants is limited to a few experimentally determined ESE motifs in plants. A prerequisite for elucidating the splicing code for plants is to identify candidate SREs. New technological advances such as deep sequencing of RNA, together with in vivo RNA-RBP cross-linking methods (HITS-CLIP, PAR-CLIP) have paved the way to find SREs globally in animals. Although these methods are applicable to plants, they have yet to be employed. Determining the splicing code in plants requires analysis of entire transcriptomes across tissues and conditions, and relating the observed patterns of AS to a variety of sequence signals, including SREs, nucleosome positioning signals, and possibly other as-yet unknown factors that affect splice site choice. Addressing these challenges and integrating the results to formulate the plant splicing code is a daunting task that will require a significant effort from the plant research community.

Conflict of Interest Statement

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.