Keywords

1 Introduction

RNA-binding proteins (RBPs) are a diverse class of proteins that control every step of RNA processing and RNA function in the cell. They are characterized by dedicated domains involved in RNA binding and can have accessory domains engaged in protein-protein interactions or enzymatic activities.

In higher plants, RBP function so far has been best studied in the reference plant Arabidopsis thaliana. Among the RBPs present in the Arabidopsis genome are 197 proteins with an RNA recognition motif (RRM), the most abundant type of RNA-binding domain, and 28 K homology (KH) domain proteins first identified in mammalian heterogeneous nuclear protein hnRNP K (Silverman et al. 2013). In addition, 26 Pumilio (PUM) domain proteins, nine DEAD-box helicases as well as five proteins with cold shock domains (CSDs) have been identified (Silverman et al. 2013). Another 450 proteins harbor pentatricopeptide repeat (PPR) domains. PPR domains consist of multiple 35-amino acid repeats of which two are known to be engaged in specific RNA recognition (Barkan and Small 2014). These proteins are imported into mitochondria or chloroplasts and regulate all aspects of RNA metabolism, e.g., RNA editing, splicing, RNA cleavage, and translation in organelles (Schmitz-Linneweber and Small 2008; Barkan and Small 2014).

A suite of Arabidopsis RBPs have been experimentally characterized, mainly through loss-of-function mutants and transgenic plants ectopically overexpressing RBPs. These approaches revealed a crucial role for RBPs in development (Kalyna et al. 2003; Ripoll et al. 2006; Kupsch et al. 2012; Völz et al. 2012; Ferrari et al. 2017; Foley et al. 2017; Teubner et al. 2017), timing of plant reproduction (Macknight et al. 1997; Streitner et al. 2008; Hornyik et al. 2010), responses to abiotic stress (Kim et al. 2007b, c, 2008, 2010; Park et al. 2009), pathogen defense (Fu et al. 2007; Qi et al. 2010; Jeong et al. 2011; Lyons et al. 2013; Nicaise et al. 2013), responses to phytohormones (Lu and Fedoroff 2000; Hugouvieux et al. 2001; Riera et al. 2006; Carvalho et al. 2010; Hackmann et al. 2014; Löhr et al. 2014), and circadian timekeeping (Heintzen et al. 1994; Staiger 2001; Jones et al. 2012; Schmal et al. 2013; Perez-Santángelo et al. 2014). At the biochemical level, an impact of defined RBPs on RNA processing including pre-mRNA splicing, 3′ end processing, processing of microRNA precursors, and translation has been described (Lopato et al. 1999; Simpson et al. 2003; Vazquez et al. 2004; Dong et al. 2008; Stauffer et al. 2010; Ren et al. 2012; Rühl et al. 2012; Juntawong et al. 2013; Sorenson and Bailey-Serres 2014; Staiger 2015; Carvalho et al. 2016). Recent attempts to comprehensively identify RBPs, summarized in Sect. 2, provided experimental evidence for RNA binding for most of the previously identified Arabidopsis RBPs and identified a plethora of proteins with noncanonical RBDs.

Systems approaches to describe RNA–protein interactions globally come in two main flavors (Fig. 1). In RNA-centric approaches, proteins associated with mRNAs are recovered by RNA pull-down and identified by mass spectrometry, a technique referred to as mRNA interactome capture (Baltz et al. 2012; Castello et al. 2012) (Fig. 1a). In protein-centric approaches, the focus is laid on a particular RBP. The RNA complement associated with the RBP of interest, the ribonome, is identified via immunoprecipitation of the RBP from cell lysates and identification of the bound target RNAs, initially by microarrays (Tenenbaum et al. 2000; Galgano and Gerber 2011; Guerreiro et al. 2014) or more recently via high throughput sequencing (Licatalosi et al. 2008; König et al. 2010; Rossbach et al. 2014; Müller-McNicoll et al. 2016) (Fig. 1b).

Fig. 1
figure 1

Strategies to globally identify in vivo RNA–protein interaction in Arabidopsis. (a) RNA-centric strategies such as mRNA interactome capture employ oligo(dT) affinity capture. RNA and bound proteins are covalently linked in planta through UV irradiation. RNA–protein complexes are recovered by oligo(dT) pull-down. Proteins are released by RNase treatment, subjected to tryptic digest and identified via mass spectrometry. (b) Protein-centric methods focus on a particular RBP and aim at identifying its in vivo RNA targets. Based on the cross-linking agent, RIP using formaldehyde or CLIP technique using UV light are distinguished. Proteins are immunoprecipitated. In RIP-seq cross-links are reversed by heat treatment, RNA is isolated, subjected to reverse transcription and PCR amplification for HITS. Targets enriched upon RIP are determined relative to mock IP controls, e.g., Xing et al. 2015, or relative to polyadenylated RNA, e.g., Meyer et al. 2017. In iCLIP (König et al. 2010), RNA–protein complexes are subjected to RNase treatment. Bound proteins are digested with proteinase, leaving a polypeptide at the cross-link site. Reverse transcriptase stops there, allowing the detection of the cross-link site at the −1 position of the processed sequencing reads

Full size image

2 The Arabidopsis RBPome

Of all predicted RBPs in Arabidopsis, RNA binding has only been experimentally confirmed for a limited number of them. A first attempt to globally identify proteins based on their ability to interact with mRNAs in vivo was made for cultured Arabidopsis cells (Schmidt et al. 2010). In this study, mRNAs and interactors were recovered under native conditions by affinity chromatography on an oligo(dT) cellulose column followed by two-dimensional gel electrophoresis. The protein components were identified via Maldi-TOF. In the RNA-bound proteome were a suite of RRM proteins including members of the family of glycine-rich RNA-binding proteins like AtGRP2 (Arabidopsis thaliana glycine rich RNA-binding protein 2), AtGRP7 and AtGRP8 (Lewinski et al. 2016), the two oligouridylate-specific RBP45 and RBP47 proteins (Lorkovic et al. 2000), and CSD proteins.

In 2012, mRNA interactome capture was reported to comprehensively identify proteins interacting with mRNAs in mammalian cells (Baltz et al. 2012; Castello et al. 2012). This technique employs in vivo cross-linking of mRNA and bound proteins by UV light irradiation. The RNA–protein complexes are recovered by pull-down of polyadenylated RNAs using magnetic beads coated with oligo(dT). Proteins are released by RNase treatment, subjected to tryptic digest and identified via mass spectrometry (Fig. 1a). Following these pioneering studies, this technique was applied to a wide range of organisms including yeast, Drosophila melanogaster, Caenorhabditis elegans, Leishmania, trypanosomes, and Plasmodium (Mitchell et al. 2013; Beckmann et al. 2015; Matia-Gonzalez et al. 2015; Bunnik et al. 2016; Lueong et al. 2016; Sysoev et al. 2016; Wessels et al. 2016; Nandan et al. 2017). A minimal core mRNA bound proteome occurring in both human and yeast was defined by Beckmann and coworkers (Beckmann et al. 2015). Lately, mRNA interactome capture has also been successfully applied to Arabidopsis (Marondedze et al. 2016; Reichel et al. 2016; Zhang et al. 2016).

2.1 The mRNA Interactome of Arabidopsis Protoplasts

The first mRNA interactome capture experiments in Arabidopsis employed widely differing tissues to catalog RBPs. Gueten and coworkers chose protoplasts, cells without a cell wall, assuming that UV cross-linking should occur as efficiently as in mammalian cell monolayers (Zhang et al. 2016). Leaf mesophyll protoplasts are also widely used in transient assays to study the regulation of gene expression.

A mesophyll protoplast mRNA interactome was defined with a total of 325 proteins based on enrichment in cross-linked samples vs. non-cross-linked controls with a log2 fold change above 2 (Zhang et al. 2016). Of these, one class was represented by 123 ribosomal proteins of which 52 were also present in the core mRNA-bound proteome of human and yeast cells (Beckmann et al. 2015). The second class comprised 70 proteins with a known RBD. For 41 of them, a role in mRNA binding and RNA biology had already been described while the remaining proteins had a potential role in mRNA processing. Moreover, 12 of the RBPs in the second class overlapped with the RBPs identified in the native oligo(dT) affinity chromatography approach (Schmidt et al. 2010). The third class comprised 132 candidate RBPs. Of these, 49 were metabolic enzymes, mainly oxidoreductases. Moreover, numerous proteins related to photosynthesis were found. As these are generally strongly expressed, their RNA binding activity and the domains involved beg for an independent validation. One of the enzymes was the Arabidopsis ortholog of phosphoglycerate kinase whose RNA binding capacity has previously been validated in yeast and human cells (Beckmann et al. 2015).

2.2 The mRNA Interactome of Etiolated Arabidopsis Seedlings

Another mRNA interactome capture experiment employed 4-days-old etiolated Arabidopsis seedlings (Reichel et al. 2016). This was based on the rationale that UV-absorbing pigments present in green plant tissue may interfere with UV cross-linking in planta and their absence in etiolated tissue may allow more efficient UV cross-linking.

Around 300 of the 746 proteins identified altogether were significantly enriched in UV cross-linked samples vs. non-cross-linked controls with a false discovery rate below 1% and designated the “At-RBP set.” Eighty percent of these have a known RBD, and 75% have been linked to RNA biology. More than 400 additional proteins did not meet the significance criteria applied for the “At-RBP set” and were classified as “candidate RBPs.”

Notably, of the 197 computationally predicted RRM proteins in Arabidopsis 160 were detected in the input fraction in etiolated seedlings (Silverman et al. 2013). Half of these were recovered in the “At-RBD set” and another 50 were present among the “candidate RBPs.” Similarly, seven of the predicted KH proteins were present in the “At-RBD set” and 12 were among the “candidate RBPs.” Of the predicted 450 members of the PPR protein family only 60 were detected in the input fraction, likely due to low abundance (Schmitz-Linneweber and Small 2008; Reichel et al. 2016). Only six PPR proteins were found in the “At-RBP set” and another twelve in the “candidate RBPs,” likely because most RNAs in the organelles lack poly(A) tails. A comparison of the identified proteins to the mRNA interactome in other model organisms revealed that 52 were present in the interactomes of humans (Baltz et al. 2012; Beckmann et al. 2015), mice (Kwon et al. 2013; Liao et al. 2016), and yeast (Beckmann et al. 2015) and were assigned to basic functions in RNA metabolism such as translation, splicing, and RNA unwinding.

In addition to RBPs with known RBDs many Arabidopsis proteins emerged that have not been linked to RNA binding so far. Among novel RBPs were proteins harboring a YT521-B homology (YTH) domain (Li et al. 2014). YTH domain proteins have been shown to bind N6 methyladenosine and thus serve as readers of the m6A mark in mammals (Wang et al. 2014). In addition, Alba domain containing proteins have been identified. Alba domain proteins are well characterized in archaebacteria where they act as transcriptional repressors and in other eukaryotes where they control translation (Goyal et al. 2016). In plants, they have not yet been functionally characterized. The only observation pointing to RNA binding is the recovery of an Arabidopsis Alba domain protein by RNA-affinity chromatography (Gosai et al. 2015). WHIRLY domain containing proteins have been characterized as single-stranded DNA binding proteins in organelles (Krause et al. 2009) and in maize, association of a WHIRLY protein with chloroplast transcripts has been observed (Prikryl et al. 2008). The identification of three WHIRLY proteins in the etiolated seedling interactome (Reichel et al. 2016) and of WHIRLY1 upon oligo(dT) affinity chromatography in Arabidopsis cells (Schmidt et al. 2010) now provides evidence for global in vivo RNA binding.

In addition, a plethora of proteins with potential RNA binding activity have been detected. To substantiate their RNA-binding properties, independent replication is desirable. Among those are proteins with the Domain of unknown function 1296, cytoskeletal proteins, and photoreceptors. The identification of plasma membrane intrinsic proteins has led to the speculation that aquaporins may be involved in transport of RNAs between cells (Reichel et al. 2016).

2.3 The mRNA Interactome of Arabidopsis Cultured Cells and Leaves of Adult Plants

Another mRNA interactome capture experiment was performed on cell suspension cultures generated from roots of the Arabidopsis accessions Col-0 and Landsberg erecta. In parallel, leaves of four-weeks-old Arabidopsis Col-0 plants were investigated (Marondedze et al. 2016). Of 1145 proteins identified altogether in these three samples, 914 appeared only in UV cross-linked samples, and 233 proteins were significantly enriched upon UV cross-linking relative to non-cross-linked samples. More than 350 proteins were known RBDs whereas 736 were novel candidate RBPs not previously assigned an RNA-related function or known RBD, including many enzymes of intermediary metabolism, and thus await further experimental proof (Marondedze et al. 2016).

The discovery of many novel RBPs begs for further investigation of the RNA-binding properties of these proteins. Accordingly, methods to define RNA targets of candidate RBPs genome wide using protein-centric methods have recently been adapted for the use in Arabidopsis, as discussed below.

3 Toward Arabidopsis Ribonomes

Approaches to globally identify in vivo targets of an RBP in Arabidopsis mostly rely on transgenic plants expressing an epitope-tagged version of the RBP. Immunopurification is performed via an antibody directed against the epitope tag. To mirror-image the endogenous expression pattern, authentic promoters are used and the constructs are introduced into a loss-of-function mutant (Köster and Staiger 2014). Alternatively, endogenous RBPs can be recovered with dedicated antibodies.

To freeze the in vivo RNA–protein interactions before cell lysis, cross-linking is performed by exposing plants to formaldehyde in RNA immunoprecipitation (RIP) or by UV irradiation in UV cross-linking and immunoprecipitation (CLIP) (Fig. 1b). Formaldehyde efficiently cross-links nucleic acids and proteins in vivo but also cross-links proteins. Thus, not only direct targets are recovered. This is circumvented by using 254 nm UV light that cross-links proteins directly binding to nucleic acids in the neighborhood of the excited nucleobase but does not cross-link proteins.

To date, a comprehensive determination of in vivo targets, the ribonome, has been performed for only a few Arabidopsis RBPs, both nucleocytoplasmic proteins and chloroplast-localized proteins with different tasks in posttranscriptional regulation. In the subsequent sections, selected examples are presented.

3.1 HLP1, An hnRNP A/B-Like Protein Involved in Alternative Polyadenylation

HLP1 is an Arabidopsis RBP resembling mammalian hnRNP A/B-like proteins (Zhang et al. 2015). High throughput sequencing (HITS)-CLIP of HLP1 fused to GFP and expressed under control of the strong, constitutive Cauliflower Mosaic Virus 35S RNA promoter identified above 5500 transcripts bound in vivo (Zhang et al. 2015). When endogenous HLP1 protein was precipitated by a specific antibody, 6850 transcripts bound in vivo were detected with an overlap of above 3000 transcripts to the HLP1-GFP precipitation. The prevalence of cross-linked regions near polyadenylation sites provoked the hypothesis that HLP1 may control polyadenylation. Indeed, in more than 2000 transcripts the distal polyadenylation site was preferred over the proximal polyadenylation site in hlp1 mutant plants. Around 19% of these transcripts were also recovered by HLP1 HITS-CLIP, pointing to a role for HLP1 in the control of alternative polyadenylation, at least partly by direct binding. In line with this, MEME motifs overrepresented in the cross-link regions, namely A-rich (5′-AGAAAA-3′) and U-rich (5′-UUUUCU-3′) motifs, resembled motifs enriched in the vicinity of the poly(A) site, 5′-AAAGAAAA-3′ and 5′-UGUUUC-3′. The presence of cross-link regions in other parts of the transcripts apart from the 3′ untranslated region (UTR) suggests that HLP1 may also affect other aspects of pre-mRNA processing in addition to polyadenylation.

3.2 The Glycine-Rich RBP AtGRP7

AtGRP7 (Arabidopsis thaliana glycine rich RNA-binding protein 7) is another hnRNP-like protein with an N-terminal RRM and a C-terminus enriched in contiguous glycine residues. AtGRP7 is regulated by the circadian clock and negatively autoregulates its own oscillations by alternative splicing and Nonsense-mediated decay (Staiger et al. 2003; Schmal et al. 2013). Additionally, it is involved in several steps of posttranscriptional regulation including alternative splicing, nucleic acid chaperone function, and pri-miRNA processing (Kim et al. 2007a; Streitner et al. 2012; Köster et al. 2014). To gain insights into the breadth of its in vivo targets, individual nucleotide resolution cross-linking and immunoprecipitation (iCLIP) and RIP-seq were performed (Meyer et al. 2017). AtGRP7 fused to GFP was expressed from its own promoter including all regulatory elements (5′ UTR, intron, and 3′ UTR) in the atgrp7-1 loss-of-function mutant. In parallel, transgenic plants expressing GFP alone or an RNA-binding dead variant of AtGRP7 with a single conserved arginine in the RRM mutated to glutamine (AtGRP7 R49Q) were used as negative controls.

iCLIP identified 858 transcripts with significant iCLIP hits in four out of five biological replicates for AtGRP7-GFP that were not present in the controls. RIP-seq identified 2453 transcripts enriched by AtGRP7-GFP relative to total polyadenylated RNA. The higher number may be due to the higher cross-linking efficiency of formaldehyde compared to UV light, and the recovery of many indirect targets. 452 transcripts were common in both data sets, suggesting that they represent a set of high confidence binders. The iCLIP cross-link sites were observed in all transcript regions, the UTRs, coding sequence and introns. After correcting for the length of the feature in the genome, cross-link sites in the 3′ UTR prevailed. Conserved motifs in the vicinity of the cross-link sites generally were U/C rich.

To determine how AtGRP7 may impact its downstream targets, the binding targets were cross-referenced against transcriptome data from AtGRP7 overexpressing plants or loss-of-function mutants. In both, the AtGRP7 overexpressors or the mutant, a similar number of transcripts was expressed at elevated or reduced levels compared to wild-type plants. Notably, significantly more differentially expressed iCLIP targets were downregulated in AtGRP7-overexpressors than upregulated. In turn, more of the differentially expressed AtGRP7 iCLIP targets were expressed at elevated in the mutant than at reduced levels. This indicates a predominantly negative effect of AtGRP7 on its targets. Among the targets were more circadianly regulated transcripts than expected. In particular, elevated AtGRP7 levels lead to damping of circadian oscillations of target transcripts including DORMANCY/AUXIN ASSOCIATED FAMILY PROTEIN2 and CCR-LIKE. This conforms with the idea that the circadian clock regulated AtGRP7 functions as a molecular slave oscillator, conveying temporal information from the core circadian clock within the cell (Rudolf et al. 2004). In addition, changes in splicing patterns were observed for iCLIP and RIP-seq targets upon misexpression of AtGRP7, confirming a role for AtGRP7 in the control of alternative splicing.

3.3 The Splicing Regulator SR45

Arabidopsis thaliana serine/arginine rich (SR)-like protein SR45, the counterpart of metazoan RNPS1, is an SR-like protein with two RS domains, flanking either side of the RRM (Badolato et al. 1995; Golovkin and Reddy 1999). Notably, recombinant Arabidopsis SR45 can activate splicing of a β-globin splicing reporter in HeLa cell S100 extracts (Ali et al. 2007). SR45 occurs in two splice isoforms that arise through differential usage of a 3′ splice site in intron 6. This leads to two protein isoforms that differ by seven amino acid residues and in their function: SR45.1 is involved in petal development in flowers, whereas SR45.2 is important for root growth (Zhang and Mount 2009). Genome-wide targets for SR45.1 were determined during early seedling development (Xing et al. 2015) and in inflorescences (Zhang et al. 2017), respectively.

In seedlings, RIP-seq identified 4361 transcripts from 4262 genes that were enriched upon precipitation of SR45.1-GFP from nuclei of transgenic plants compared to mock precipitation from wild type plants (Xing et al. 2015). These were designated SARs, for SR45 associated RNAs. A Gene Ontology term analysis showed that 43 of 147 abscisic acid (ABA) signaling genes (30%) were among the SARs, in line with a function for SR45 in the ABA signaling pathway (Carvalho et al. 2010). Hundred and forty-eight of the SARs had an altered expression in the sr45-1 mutant, suggesting that binding of SR45 has functional consequences.

A MEME search for SR45 binding motifs revealed four overrepresented motifs within SAR genes. Two G/A rich motifs are largely positioned within exons and show strong similarity to the binding motifs of two metazoan splicing regulators Transformer 2 (Tra2) and serine/arginine-rich splicing factor 10 (SRSF10). Furthermore, one G/A rich motif closely resembles the GAAG motif, a known cis-regulatory element in regulating alternative splicing in plants. In contrast, two U/C rich motifs peak within intronic regions near 5′ and 3′ splice sites, in line with the observation that the majority of SARs were from intron-containing genes and the known role as a splicing regulator (Xing et al. 2015).

To gain insights into a potential role of SR45 in flower development, RIP-seq was performed for SR45.1-GFP in inflorescence tissue (Zhang et al. 2017). The resulting reads were analyzed by two different bioinformatics pipelines, one based on mapping reads to the genome and one directly quantifying annotated transcripts. SARs in inflorescence were defined based on a twofold enrichment compared to GFP only controls and the identification by both pipelines. Of 1812 SARs in inflorescence, 677 overlapped with the SARs in seedlings.

Notably, 19 transcripts encoding splicing factors were among the SARs including SR45 itself, the three SR proteins SR30, SR34, and SCL35, the pre-mRNA processing factors PRP39, PRP40A, PRP40B, and PRP2, and the RNA helicase RH42, pointing to a hierarchical regulation of posttranscriptional regulators (Keene 2007). Genes upregulated in the sr45-1 mutant are enriched for defense response genes. Indeed, the sr45-1 mutant was more resistant to bacterial and fungal pathogens. Of 68 upregulated defense response genes in sr45-1, 10 were SARs. Thus, SR45 has an additional role as a negative regulator of plant immunity.

Furthermore, 81 of the inflorescence SARs were aberrantly spliced in the sr45-1 mutant. Determination of potential SR45 binding sites in inflorescence SARs uncovered an overrepresentation of the purine-rich motifs GGNGG, GNGGA, and GNGGNNG. Importantly, GGNGG and related motifs are enriched in introns and exons that are alternatively spliced in the sr45-1 mutant, irrespective of the splicing event is favored or suppressed by SR45. This led to the suggestion that SR45 identifies regions for alternative splicing and acts as a facilitator for other splicing factors. However, the identified binding motifs for SR45 in inflorescences differ from that in seedlings, which might be in part due to the different bioinformatic tools used for motif determination. Both RIP-seq data sets nevertheless strengthen SR45’s key role as an important splicing factor in Arabidopsis. However, in both RIP-seq experiments intron-less transcripts were identified in addition to intron-containing transcripts, pointing to functions of SR45 beyond its known role in pre-mRNA splicing.

Interestingly, a comparison between the U/C-rich motifs of AtGRP7 and the U/C-rich motifs of SR45 identified by MEME in seedlings revealed a high degree of similarity (Meyer et al. 2017). The functional significance remains to be tested.

3.4 Cold Shock Protein 1

In bacteria, CSPs are upregulated upon cold stress and destabilize RNA secondary structure at low temperatures (Sommerville 1999). To elucidate a potential involvement of Arabidopsis CSPs in the regulation of cold responsive genes, RIP followed by gene chip analysis was performed for CSP1 (Juntawong et al. 2013).

More than 6000 mRNAs were identified. Comparison of these CSP1-associated transcripts in total RNA and RNA loaded onto polysomes revealed an enrichment of mRNAs associated with ribosome biogenesis in the pool of actively translating RNAs. The high GC content in 5′ UTRs of these mRNAs suggested that CSP1 is involved in removing secondary structures in the 5′ UTR to facilitate their translation. Accordingly, these mRNAs were less efficiently loaded onto polysomes at low temperature in the atcsp1-1 mutant compared to wild type plants or CSP1 overexpressing plants (Juntawong et al. 2013).

3.5 The cpRNP Family

The highly abundant chloroplast ribonucleoproteins (cpRNPs) have been well characterized for their role in regulating chloroplast transcripts (Ohta et al. 1995). The cpRNPs comprise an acidic domain and two RRMs. They are encoded in the nucleus and imported into chloroplasts. Mutants in distinct cpRNPs are widely affected in processing of transcripts in the chloroplast, leading to defects in chloroplast development and, consequently, plant performance owing to the essential role of the chloroplast in photosynthetic energy (Ruwe et al. 2011). For example, mutants deficient in CP29A (29 kDa chloroplast protein A) and CP31A (31 kDa chloroplast protein A) showed gross defects at low ambient temperature. RIP performed with antibodies against the endogenous proteins and subsequent hybridization of coprecipitated RNAs on tiling arrays covering the Arabidopsis chloroplast genome (RIP-Chip) showed that CP29A and CP31A associate with large overlapping sets of chloroplast transcripts including strong enrichment for psbB, psbD, psaA/B, atpB, ndhB and intermediate enrichment for almost all chloroplast mRNAs (Kupsch et al. 2012). Both CP29A and CP31A are required for accumulation of chloroplast mRNAs under cold stress. Furthermore, binding of CP31A to 3′ ends of certain transcripts serves to protect these transcripts against 3′ exonuclease activity (Kupsch et al. 2012). Together with the known role of CP31A in RNA (Tillich et al. 2009) this points to multiple functions in posttranscriptional regulation in chloroplasts.

For CP33A (33 kDa chloroplast protein A), RIP-chip revealed an association with a large body of chloroplast mRNAs (Teubner et al. 2017). A global reduction in mRNAs and proteins making up the photosynthetic apparatus was found in the cp33a mutant. In line with a crucial role for CP33A in the development of the photosynthetic apparatus, cp33a null mutants have an albino phenotype and are not able to survive without external sucrose supply (Teubner et al. 2017).

3.6 The PPR Protein AtCPR1

In contrast to the broad substrate specificity of the cpRNPs, a very narrow substrate specificity was found for a representative of the PPR class of nuclear-encoded RBPs that are imported into organelles. AtCPR1 (Arabidopsis thaliana CHLOROPLAST RNA PROCESSING 1) is important for the production of subunits of the thylakoid protein complexes (Ferrari et al. 2017). Atcpr1 mutants are yellow-white because the subunits of the photosynthetic apparatus do not accumulate.

RIP-chip was performed for AtCPR1 under native conditions. Hybridization of bound targets to chloroplast tiling arrays revealed specific binding of AtCPR1 to only few transcripts, the psaC transcript encoding a photosystem I subunit, petB-petD encoding Cytochrome b6 and the subunit IV of the cytochrome b6/f complex. Because during RIP RNase was used to digest unprotected RNA, it was possible to delineate the binding regions. Binding to the petB-petD intergenic region correlated with a requirement for processing of the polycistronic transcript comprising petB and petD (Ferrari et al. 2017), thus providing proof for the functional relevance of the observed in vivo binding.

4 Combined Analysis of RNA–Protein Interaction and RNA Secondary Structure Landscapes

In addition to RNA sequence, RNA secondary structure also strongly influences the interaction of RBPs with their cognate RNA binding motifs (Cruz and Westhof 2009; Vandivier et al. 2016). RNA structure may facilitate binding of RBDs with a preference for double-stranded RNA or inhibit binding of RBPs with a preference for single-stranded RNA. Protein interaction profile sequencing (PIP-seq) allows simultaneous delineation of in vivo RNA secondary structure and protein-protected sites (PPSs) (Fig. 2) (Gosai et al. 2015). To identify PPSs, samples are treated with a single-strand specific or double-strand specific RNase. Proteins are then denatured before library preparation. To determine the RNA secondary structure, proteins are denatured by SDS and removed by protease digestion to make sites protected by proteins in vivo accessible for RNases. Collectively, motifs that are enriched in the samples used to determine protein protected sites compared to the samples used for structure determination are in vivo target sites of RBDs.

Fig. 2
figure 2

Protein interaction profile sequencing (PIP-seq). (a) To identify protein binding sites, i.e., sites that are protected from RNase digestion by interacting proteins (PPS), samples are treated with an RNase specific for double-stranded RNA (left) or for single-stranded RNA (right). Subsequently, proteins are denatured, leaving either target sites for proteins with a preference for single-stranded regions (left), or target sites for proteins with a preference for double-stranded regions (right). These sequences are used to generate libraries for HITS. (b) To determine the RNA secondary structure, proteins are denatured in a first step. Subsequently, samples are treated with RNase specific for double-stranded RNA (left) or for single-stranded RNA (right). Again, libraries for HITS are prepared. Collectively, motifs that are enriched in the samples used to determine protein binding sites compared to the samples used for structure determination are in vivo target sites of RBDs

Full size image

Gregory and coworkers applied PIP-seq to the nuclei of two specific cell types in the Arabidopsis roots that derive from epidermal cells through distinct differentiation, those cells bearing root hairs and those that do not (Foley et al. 2017). Distinct protein binding patterns were detected, and binding motifs either specific to hair cells, non-hair cells or common to both cell types were determined. To identify candidate proteins, RNA affinity chromatography was performed on immobilized oligonucleotides derived from enriched motifs. A GGN repeat motif enriched in sites protected in both hair cells and non-hair cells recovered SERRATE (SE) from root lysates, a zinc finger containing RBP involved in processing of miRNA precursors. A TG rich motif enriched in hair cell-specific protected sites identified AtGRP2, AtGRP7 and AtGRP8. Subsequently, AtGRP8 was shown to regulate root hair development at the posttranscriptional level.

An advantage of PIP-seq is that it does not rely on an antibody to identify target sites within bound transcripts. In contrast, subsequent identification of the cognate binding proteins requires in vitro binding techniques. Thus, binding in vivo has to be confirmed by independent means.

5 Achievements and Limitations of Arabidopsis In Vivo RNA–Protein Interaction

The recent mRNA interactome capture studies are very valuable in having established UV cross-linking and oligo(dT) affinity capture to determine the mRNA binding proteome also in Arabidopsis. A large number of previously predicted RBPs in Arabidopsis were now identified experimentally and many novel proteins without a previous assignment to RNA biology unearthed. Reichel and colleagues noticed a bias toward proteins with higher abundance in the interactome compared to the input (Reichel et al. 2016), suggesting that additional proteins with lower expression level may still be identified in the future. Only few of the mRNA interacting proteins were present in all three interactomes (Köster et al. 2017). This may partly be attributed to the widely differing developmental stages investigated. Among the commonly identified proteins are numerous cytoplasmic ribosomal proteins from the small and large ribosomal subunits, likely due to their high abundance, as well as the ubiquitously expressed glycine-rich RBPs AtGRP7 and AtGRP8 (Köster et al. 2017).

Future applications are the dynamics of posttranscriptional networks in response to endogenous and exogenous stimuli cues by describing changes in the mRNA bound proteomes. Furthermore, as proteins binding to nonpolyadenylated RNAs obviously remain elusive in these approaches, transcript-specific approaches have to be developed.

Transcriptome-wide identification of target transcripts bound by selected RBPs in vivo has overcome a major limitation in research on plant RNA-based regulation. Nevertheless, except for the PPR proteins, we are still far from understanding the exact binding specificity of most proteins and the consequences in vivo binding has for the targets. To correlate in vivo binding with function, the impact of mutated candidate binding motifs on RBP binding and target gene expression has to be determined.

Most bioinformatics pipelines today discussing motif discovery are limited to sequence data. Current efforts focus on developing bioinformatics pipelines for identifying conserved motifs taking RNA structure context into consideration (Maticzka et al. 2014). Molecular dynamics of RNA molecules are still compute intensive but can shed light on possible interaction sites and three dimensional structures (Tuszynska et al. 2015; Boniecki et al. 2016). Finally, heterogeneous datasets and analyses, fusing several kinds of sources, can improve meta-analysis with in silico and in vivo datasets. This is yet limited in Arabidopsis but will improve the information quality in the near future. Additionally, it will be important to have comprehensive databases on RBP target sites linked to the Arabidopsis information portal (The International Arabidopsis Informatics Consortium 2012). Such resources will be of great value to improve a systems understanding of RNA–protein interaction.