Methylation of the nuclear poly(A)-binding protein by type I protein arginine methyltransferases – how and why

Introduction: the protein arginine methyltransferases

Arginine side chains in proteins can be posttranslationally modified by methylation (Bachand, 2007; Bedford, 2007; Bedford and Clarke, 2009; Di Lorenzo and Bedford, 2011). Four types of modification have been identified, and the enzymes catalyzing the methylation are classified according to these four products. Type I protein arginine methyltransferases (PRMTs) catalyze asymmetric dimethylation, i.e., transfer of two methyl groups to one of the terminal nitrogen atoms (ω-N^G,N^G-dimethylarginine), the most frequent modification (Figure 1). In humans, six type I enzymes are known, PRMT1, 2, 3, 4, 6, and 8 (Bedford, 2007; Bedford and Clarke, 2009; Di Lorenzo and Bedford, 2011). Among these, PRMT1 is the most abundant enzyme in those cells that have been analyzed (Tang et al., 2000; Fronz et al., 2008). Type II PRMTs catalyze the symmetric dimethylation of arginine side chains, i.e., transfer of one methyl group to each of the terminal nitrogen atoms (ω-N^G,N′^G-dimethylarginine) (Figure 1). PRMT5 is the only human enzyme of this type that has been identified with certainty (Bedford, 2007; Bedford and Clarke, 2009; Di Lorenzo and Bedford, 2011), whereas the activity of PRMT7 is controversial (Lee et al., 2005a; Jelinic et al., 2006; Migliori et al., 2012; Zurita-Lopez et al., 2012). Arginine monomethylated at one of the terminal nitrogen atoms (ω-N^G-methylarginine) is an obligatory intermediate in the reaction of both type I and type II PRMTs, but human PRMT7, classified as a type III enzyme, may generate monomethyl arginine as its final product (Zurita-Lopez et al., 2012) (but see the controversy referred to above). Interestingly, in yeast, the formation of ω-monomethylarginine can also be catalyzed by a member of the SPOUT family of methyltransferases, which are not related to the PRMTs mentioned so far and were previously only known to methylate RNA substrates (Young et al., 2012). Type IV methyltransferases, responsible for the monomethylation of the internal nitrogen atom (δ-N-methylarginine) have so far been identified only in fungi (Niewmierzycka and Clarke, 1999; McBride et al., 2007). S-Adenosylmethionine (SAM) is the methyl group donor for all known PRMTs and is converted to S-adenosylhomocysteine (SAH) in the course of the reaction.

Figure 1

Chemical structures of methylated arginine side chain in proteins.

Type I and others refer to the PRMTs catalyzing the modification. MR, methyl arginine. Peptide backbone is indicated with a wiggly line. Methyl groups transferred to the arginine side chain are marked in red.

This review will summarize our work concerning arginine methylation of the nuclear poly(A)-binding protein (PABPN1) by type I PRMTs and relate our results to the field of arginine methylation in general.

The nuclear poly(A)-binding protein

Poly(A) tails of eukaryotic mRNAs are bound by specific proteins in the cell (Mangus et al., 2003; Kühn and Wahle, 2004). The better-known variant is the cytoplasmic poly(A)-binding protein (PABPC in humans; Pab1p in Saccharomyces cerevisiae). PABPC, which is important in translation and the control of RNA stability, is composed of four RNA-binding domains of the RNA recognition motif (RRM) or ribonucleoprotein (RNP) type, followed by a linker and a conserved C-terminal α-helical domain. Although the nuclear poly(A)-binding protein (PABPN1 in humans) shares the RNA binding specificity of PABPC, its structure is quite different (Nemeth et al., 1995) (Figure 2): in the order N-terminus to C-terminus, it is composed of a polyalanine sequence, expansions of which cause the human genetic disease oculopharyngeal muscular dystrophy (OPMD) (Brais et al., 1998; Calado et al., 2000b; Winter et al., 2013); an unstructured acidic domain of unknown function; a coiled coil domain, predicted on the basis of its characteristic pattern of hydrophobic amino acid side chains and confirmed by CD spectroscopy (Kerwitz et al., 2003); a single central RRM domain; finally, an unstructured basic C-terminal domain. The C-terminal domain is 49 amino acids long and contains 13 arginine residues, all of which are asymmetrically dimethylated, as determined by a combination of mass spectrometry and Edman sequencing (Smith et al., 1999). With respect to this type of modification, certain aspects of its amino acid composition and its role in RNA binding and protein-protein interactions (see below), the C-terminal domain of PABPN1 is reminiscent of the RGG domain of other RNA-binding proteins (Kühn et al., 2003). Detailed structural information regarding PABPN1 is limited to the RRM domains of the human protein and a related protein, embryonic PABP2 from Xenopus (Ge et al., 2008; Song et al., 2008). The structures show a dimerization of the protein, and experiments in the solution confirm a tendency of the protein to oligomerize. However, at a concentration of ∼2 μm, the mammalian PABPN1 is predominantly monomeric, as determined by analytical ultracentrifugation (Nemeth et al., 1995; Meyer et al., 2002; Kerwitz et al., 2003). Various types of aggregation require higher protein concentrations (Winter et al., 2013). The RNA-binding unit is also a monomer (Meyer et al., 2002; Kühn et al., 2003; Song et al., 2008).

Figure 2

Domain structure of bovine PABPN1.

The N- and C-terminal amino acid sequences are depicted in single-letter codes. Arginine residues asymmetrically dimethylated in vivo are marked in red.

PABPN1 was discovered by its in vitro effect on the synthesis of mammalian poly(A) tails: the poly(A) polymerase catalyzing poly(A) tail synthesis has a very low activity due to a weak affinity for RNA. PABPN1, together with the heterooligomeric cleavage and polyadenylation specificity factor (CPSF), increases the affinity of poly(A) polymerase for the RNA to be elongated, apparently by a direct contact with the enzyme. Increased RNA affinity of poly(A) polymerase results in increased processivity of polyadenylation and, consequently, a dramatically enhanced rate of elongation (Wahle, 1991; Bienroth et al., 1993; Kerwitz et al., 2003). After the polymerization of 200–250 nucleotides, processive elongation is terminated because the interaction between CPSF and poly(A) polymerase can no longer be maintained. The relatively homogeneous length of the poly(A) tails synthesized in this reconstituted in vitro reaction corresponds to the length of newly synthesized poly(A) tails in vivo. PABPN1 is essential for this length control (Wahle, 1995; Keller et al., 2000; Kühn et al., 2009). This model is supported by several in vivo observations, including a modest decrease in steady-state poly(A) tail length upon knockdown of PABPN1 in primary mouse myoblasts (Chen et al., 1999b; Benoit et al., 2005; Apponi et al., 2010). However, because the knockdown of PABPN1 had no effect on steady-state poly(A) tail length in two studies using other cells (Beaulieu et al., 2012; Bhattacharjee and Bag, 2012), a role of the protein in general mRNA polyadenylation has recently been challenged (Beaulieu et al., 2012). Possible explanations for the lack of an effect include compensatory roles of other poly(A)-binding proteins (Bhattacharjee and Bag, 2012) and, as with any RNAi experiment, insufficient depletion. Most importantly, the in vitro data demonstrate an effect of PABPN1 on the rate of poly(A) tail synthesis, which has not been examined in vivo under conditions of PABPN1 depletion; whether or not there should be an effect on the final length, and thus steady-state length distribution, cannot be predicted with certainty. Nevertheless, more evidence regarding the role of PABPN1 in mRNA polyadenylation in vivo is desirable. Experiments in Schizosaccharomyces pombe and mammalian cells (Lemay et al., 2010; Lemieux et al., 2011; Beaulieu et al., 2012) suggest that PABPN1 has an additional function in promoting the activity of the exosome, a protein complex with 3′ exonuclease activity involved in many RNA decay and processing reactions both in the nucleus and the cytoplasm. A role of PABPN1 in alternative polyadenylation has also recently been described (de Klerk et al., 2012; Jenal et al., 2012).

Saccharomyces cerevisiae does not have a true PABPN1 orthologue. The closest relative, Rbp29p, which possesses a very similar RRM but has little sequence homology outside this domain, is a cytoplasmic protein likely to be involved in translation (Winstall et al., 2000). Instead, a role in mRNA polyadenylation is played by the protein Nab2p. However, in contrast to PABPN1, Nab2p does not appear to increase the rate of poly(A) chain elongation and is involved only in restricting poly(A) tail growth to the normal length of ∼70–90 nucleotides, as shown both by in vitro experiments and the analysis of mutants. An additional function of Nab2p in the generation of export-competent mRNA-protein complexes (mRNPs) can be separated from its role in polyadenylation (Hector et al., 2002; Dheur et al., 2005; Kelly et al., 2010; Soucek et al., 2012). Despite the partial functional overlap between PABPN1 and Nab2p, the proteins are structurally quite distinct. For example, Nab2p has no RRM but binds poly(A) using zinc fingers (Brockmann et al., 2012; Soucek et al., 2012). Nab2p also contains an RGG domain, which is subject to arginine dimethylation (Green et al., 2002).

Structure and mechanism of type I PRMTs

Crystal structures are available for several type I enzymes, mammalian PRMT1 (Zhang and Cheng, 2003), PRMT3 (Zhang et al., 2000), and PRMT4 (the coactivator-associated arginine methyltransferase, or CARM1) (Troffer-Charlier et al., 2007; Yue et al., 2007) as well as for the yeast enzyme Hmt1p (=Prmt1p) (Weiss et al., 2000) and Arabidopsis thaliana PRMT10 (Cheng et al., 2011). Many of the structural features discussed below are also conserved in the type II enzyme PRMT5 (Sun et al., 2011). All structures show a conserved catalytic core composed of two domains, an N-terminal Rossman fold that binds SAM and a C-terminal β-barrel domain (Figure 3). All PRMTs that have been crystallized form homooligomers. A homodimer is the minimal functional unit as mutations disrupting the dimer abolish enzymatic activity (Weiss et al., 2000; Zhang and Cheng, 2003; Cheng et al., 2011). Some PRMTs, including PRMT1, also form larger oligomers (Weiss et al., 2000; Wang et al., 2001; Zhang and Cheng, 2003). Dimerization depends on a conserved α-helical dimerization arm that, in the primary structure, is inserted into the N-terminal end of the β-barrel domain and makes hydrophobic contacts with the SAM-binding domain of the other subunit. As a consequence, the dimer adopts a doughnut-like shape with a central cavity. The active site of the PRMTs, identified by the location of bound SAH and the methyl-accepting arginine, is located in a cleft between the β-barrel domain and the Rossman fold and is accessible from the central cavity of the dimer (Figure 3).

Figure 3

Cartoon representation of a PRMT1 dimer with SAH and substrate arginine (PDB 1OR8).

One molecule of the dimer is shown in transparent colors. Domains are color-coded as follows: β-barrel, yellow; SAM-binding domain, green; dimerization arm, blue. The N-terminal α-helices, which only form upon binding of SAH (or SAM), are labeled red. SAH (orange) and arginine (black) are in stick representation and colored by element (except carbon atoms).

A mechanism of catalysis has been proposed, whereby two glutamate residues, located in the double E loop of the SAM-binding domain, contact the guanidino group of the arginine side chain to induce a redistribution of electrons that facilitates the attack of one nitrogen atom on the methyl group of SAM (Zhang et al., 2000; Zhang and Cheng, 2003). Although the substitution of either glutamate strongly reduces the activity of PRMT1, a simultaneous substitution of both is relatively well tolerated (Rust et al., 2011). Evidence has been presented that, in contrast to earlier proposals, the deprotonation of the guanidinium group is not an essential part of the mechanism and methyl transfer may be promoted predominantly by the proximity between the guanidinium group and the methyl group of SAM in the active site (Rust et al., 2011). A pair of methionine residues has been proposed to be important in orienting the arginine side chain (Gui et al., 2011).

The methyl donor SAM is buried by α-helices at the N-terminus of the conserved core; thus, these have to move or unfold to release the SAH product (Zhang et al., 2000; Yue et al., 2007) (Figure 3). As the same α-helices form part of the peptide-binding site, an ordered mechanism has been suggested in which SAM has to bind before the protein substrate and the protein has to dissociate to allow SAH release (Yue et al., 2007). However, two groups performing steady-state kinetic studies of PRMT1 and 6 have come to conflicting conclusions, one favoring an ordered mechanism, in which SAM is bound first and the peptide second, and the products are released in the reverse order (Lakowski and Frankel, 2008, 2009) and the other supporting a random mechanism (Obianyo et al., 2008; Obianyo and Thompson, 2012). Conceivably, the structural data could be reconciled with a random mechanism by the involvement of additional peptide-binding sites.

An ordered mechanism would entail a strictly distributive mechanism of methylation, i.e., obligatory peptide release after each methylation event and rebinding for a second methyl transfer. A random mechanism would be compatible with either a distributive or a processive mechanism. Processive methylation, i.e., more than one methyl transfer without intermittent peptide release, can be imagined for two methyl transfers to the same arginine side chain and/or for the methylation of multiple side chains in a single polypeptide. Experimentally, the processivity can be directly determined from the number of methylation events per substrate molecule at early time points under conditions of substrate excess: With a distributive enzyme, monomethylated product will be dominant, whereas a processive enzyme will generate dimethylated or even multiply methylated products. We have examined the processivity of PRMT1 and, less extensively, PRMT3 with a number of substrates, including several peptides and full-length PABPN1 (Kölbel et al., 2009). The amounts of singly and multiply methylated products were determined by mass spectrometry and were found to correspond well to a random distribution, arguing that each methyl transfer is an independent event. Also supporting a distributive mechanism, the methylation of one substrate was immediately blocked by the addition of a large excess of a competing substrate. Finally, the dissociation rate constants of peptide and protein substrates, in the presence or absence of SAM or SAH, were similar to the k_cat of the methylation reaction, which is inconsistent with multiple methyl transfers per binding event. Also supporting a distributive reaction, Feng et al. (2011) determined peptide dissociation from PRMT1 to be much faster than methyl transfer. However, the discrepancy in dissociation rate constants is puzzling. Wooderchak et al. (2008) also found a distributive behavior of PRMT1 with a peptide containing a single arginine residue, and a distributive mechanism has been described for PRTM6 as well (Lakowski and Frankel, 2008). In contrast, Osborne et al. (2007) reported PRMT1 to be partially processive in the methylation of a single arginine residue, i.e., monomethyl arginine and dimethyl arginine were found in comparable quantities. One caveat is that the oligomeric state of the enzymes might be relevant but was not explicitly considered in any study of PRMT processivity. Whereas Kölbel et al. (2009) found a distributive behavior of PRMT1 at a low concentration, at which the enzyme probably exists as a dimer (Feng et al., 2011), others (Osborne et al., 2007; Wooderchak et al., 2008) used concentrations favoring higher oligomers (Feng et al., 2011) but came to conflicting conclusions regarding the processivity. Thus, the oligomeric state of the enzyme is not the most likely explanation for the discrepancies. Alternatively, the processivity of the enzyme could be different for different substrates, and this in fact seems to be the case in one study (Lakowski and Frankel, 2009).

PRMTs are relatively slow enzymes. PRMT1 has k_cat values on the order of 1/min for good peptide or protein substrates; a higher rate of 17/min has only been reported for the PRMT5-pICLn complex (Pesiridis et al., 2009). The affinities for peptide/protein substrates are fairly high, with K_M values in the low micromolar to submicromolar range. The methylation of the PABPN1 by PRMT1 proceeds with an apparent K_M of 0.07 μm (Osborne et al., 2007; Fronz et al., 2008; Feng et al., 2011).

Arginine methyltransferases modifying PABPN1

Although enzymatic activities generating dimethyl arginine had been known for a while, the first PRMT sequences were discovered by genetic and two-hybrid interaction screens (Henry and Silver, 1996; Lin et al., 1996; Chen et al., 1999a), and additional homologues were then identified by sequence comparisons. As PRMTs were not purified to homogeneity from tissue (for an exception, see Wang et al., 2001), it is not clear a priori whether they function by themselves or merely act as the catalytic subunits of larger protein complexes. For example, the type II enzyme PRMT5, which symmetrically methylates some of the core proteins of spliceosomal small nuclear ribonucleoprotein (snRNP) complexes, is known to do so as part of a heterotrimeric ‘methylosome’. One subunit of this complex, pICLn, contributes to the substrate specificity by interacting with the core domain of the snRNP proteins, and PRMT5 modifies an unstructured C-terminal extension of this domain (Friesen et al., 2001; Meister et al., 2001, 2002; Pesiridis et al., 2009). Histones are also substrates for PRMTs. As methylation is limited to specific chromatin regions, the PRMTs are likely to interact with other proteins that direct them to these domains, as shown for PRMT1 and PRMT4/CARM1 (Chen et al., 1999a; Koh et al., 2001; Xu et al., 2004).

PABPN1 contains exclusively asymmetrically dimethylated arginine residues; thus, it must be a substrate for type I PRMTs. As it was unknown whether recombinant PRMTs produced in Escherichia coli would display the full activity and substrate specificity of the ‘authentic’ enzymes, an unbiased approach was chosen to find the PRMTs acting on PABPN1. On the basis of an in vitro methylation assay, the enzymes were partially purified from cells and separated from each other. These experiments identified PRMT1, 3, and 6 as enzymes methylating PABPN1, whereas fractions containing PRMT2 and 4 were inactive (Fronz et al., 2008). PRMT8 was not considered because it is a neuron-specific, membrane-associated enzyme (Lee et al., 2005b), whereas PABPN1 is ubiquitously expressed. Assays with PRMTs purified from E. coli confirmed the assignment of PRMT1, 3, and 6. Moreover, each of the three active PRMTs appears to methylate PABPN1 on its own, without the help of additional proteins. The evidence is as follows: the concentration of the respective PRMT in each partially purified fraction was estimated by Western blotting and comparison to a standard provided by recombinant protein. Thus, a specific activity of the ‘authentic’ protein could be calculated and compared with the specific activity of the corresponding recombinant enzyme. As the numbers matched, the partially purified enzyme contained no proteins facilitating methyl transfer. Because the recoveries in each purification step were high, no such stimulatory factors had been lost during the partial purification. Thus, PRMT1, 3, and 6 methylate PABPN1 on their own (Fronz et al., 2008). PRMT1 and 3 were also found in PABPN1 aggregates in tissue from patients with OPMD, consistent with the in vitro data (Tavanez et al., 2009).

The problem of PRMT substrate specificity

PRMT1 is about 20-fold more efficient than PRMT3 and 6 in the methylation of PABPN1, as measured by k_cat/K_m. In extracts from PRMT1 knockout mouse embryonic stem (ES) cells, the methylation activity directed against PABPN1 was strongly reduced but easily detectable, and in the same cell line, the protein was almost fully methylated at steady state in vivo (Fronz et al., 2008). Thus, PRMT3 and 6 can substitute for PRMT1. However, the methylation of a single protein by several PRMTs is not a universal phenomenon. For example, the protein hnRNP K is methylated almost exclusively by PRMT1, which is 3000- and 200-fold more efficient than PRMT3 and 6, respectively, in vitro. In the PRMT1^-/- ES cells, methylation of hnRNP K was not detectable (Ostareck-Lederer et al., 2006; Fronz et al., 2008). Overlapping but nonidentical substrate spectra of PRMT1, 3, and 6 can also be seen in the survey of Pahlich et al. (2006), in a comparison of several peptide substrates (Fronz et al., 2008) and in the methylation of the N-terminal tails of histones (Di Lorenzo and Bedford, 2011). In contrast, the substrates of PRMT4/CARM1 do not appear to overlap with those of the other PRMTs (Lee and Bedford, 2002; Pahlich et al., 2006; Di Lorenzo and Bedford, 2011).

Type I PRMTs recognize local amino acid sequences rather than larger protein structures. For example, taking protein domains containing methyl-accepting arginine residues out of context and appending them to a host protein generates PRMT substrates as good as the original protein (Fronz et al., 2008; Kölbel et al., 2012). Also, synthetic peptides can be methylated with efficiencies similar to the proteins from which their sequences were derived (Osborne et al., 2007; Fronz et al., 2008). In some cases, a peptide sequence was a better substrate in isolation than in its natural context, suggesting that accessibility and/or conformation may be important for methylation, as discussed below (Fronz et al., 2008).

The rules by which PRMTs recognize the arginine residues to be methylated remain poorly characterized. Thus far, crystallography has not been very informative. Only PRMT1 has been co-crystallized with a peptide substrate. Unfortunately, the structure reveals, in addition to the arginine in the active site, several ill-defined electron densities, suggesting peptide binding in three different sites. These sites are acidic grooves on the surface of PRMT1, obviously corresponding to the clustered arginine residues found in typical methylation substrates (see below). Although the presence of more than one methylatable arginine in the peptide may have contributed to the heterogeneity of substrate binding, the structure clearly shows that the peptides can be bound in more than one way on the surface of PRMT1. In fact, there are acidic grooves on the surface of PRMT1 in addition to those occupied in the crystal (Zhang and Cheng, 2003). Similar negatively charged potential peptide-binding sites can also be found on the surface of PRMT3 (Zhang et al., 2000), yeast Rmt1/Hmt1 (Weiss et al., 2000), and Arabidopsis PRMT10 (Cheng et al., 2011) but not PRMT4 (Troffer-Charlier et al., 2007; Yue et al., 2007).

Early alignments of the amino acid sequences surrounding methylated arginines revealed repeats of the consensus sequence (F/G)GGRGG(G/F) (Kim et al., 1997). This motif is a hallmark of the RGG domain, which is both an RNA-binding domain (Cobianchi et al., 1988; Kumar et al., 1990; Kiledjian and Dreyfuss, 1992; Casas-Finet et al., 1993) and involved in protein-protein interactions, including self-association, which leads to the cooperativity of RNA binding (Cobianchi et al., 1988; Kumar et al., 1990; Casas-Finet et al., 1993; Cartegni et al., 1996; Bouvet et al., 1998; Dormann et al., 2010). However, the environment of dimethylarginine in other proteins, e.g., histones or PABPN1, does not correspond to the RGG consensus. In fact, synthetic peptides corresponding to the RGG consensus are not even particularly good substrates for PRMT1 (Osborne et al., 2007; Fronz et al., 2008). It seems likely that the RGG consensus reflects not only determinants for PRMT-dependent arginine methylation but also the sequences required by the functional role of the protein domain.

More recent sequence alignments have revealed a general conservation of three amino acids in the positions between -7 and +7 with respect to the methylated arginine: These are, in the order of their abundance, glycine, in particular in the position +1; arginine, except in positions -1 and +1; and proline. A specific consensus pattern did not emerge (Chen et al., 2006; Shien et al., 2009; Shi et al., 2012). The enrichment of glycine and proline, which are ill-suited for side-chain interactions, will be discussed below. One obvious problem of sequence alignments performed so far is that the particular enzyme catalyzing the methylation was not considered; thus, differing sequence specificities of the various PRMTs may blur the result. Also, the relatively large number of methylated arginine residues identified in RGG sequences dominate the data set on which the analysis is based.

Only a limited number of attempts have been made to evaluate the importance of particular residues for PRMT-dependent methylation by systematic variation of peptide substrates. Glycine was found to be important in the +1 position in an early study using an unidentified methyltransferase activity purified from tissue (Rawal et al., 1995). More recent experiments examining a single arginine embedded in an RGG consensus peptide showed that although glycine was among the preferred residues in positions +1 and +2, several other amino acids also functioned well for methylation by PRMT1 (Wooderchak et al., 2008). Interestingly, although alanine and serine were not tolerated in the +1 position in this peptide and in a protein that was examined in parallel, both amino acids are found in this position in the C-terminal domain of PABPN1 (Smith et al., 1999). The same is true for tyrosine in position +2. Thus, the preference for particular amino acids in a certain position may depend on the sequence of the entire peptide. Note, however, that although PRMT1 is the most efficient methyltransferase for PABPN1 overall, the specificities of PRMTs for the particular arginines in question have not been tested. Osborne et al. (2007) examined a series of peptides derived from the N-terminus of histone H4. Although the methylation by PRMT1 was directed toward Arg3, a patch of basic residues at positions 16–20 was important for substrate recognition. However, we have described several synthetic peptides that are excellent PRMT1 substrates, yet lack a basic patch in a comparable position (Fronz et al., 2008). Again, this suggests that sequence preferences are context-dependent and flexible. In general, however, amino acids distant from the methylation site seem to be important because shorter peptides have repeatedly been found to be poor substrates (Rawal et al., 1995; Osborne et al., 2007; Kölbel et al., 2012).

It is apparent that, in addition to the sequences of peptide substrates, their conformation must also be important. In the first description of the three-dimensional structure of a PRMT, it was already noted that the location of the active sites in the central cavity of a PRMT3 dimer was likely to limit the range of potential substrates (Zhang et al., 2000). In most PRMTs, the central cavities are approximately 8×12 Å wide (Cheng et al., 2011). We have modeled an arginine side chain on the surface of spheres of different sizes and found that the maximum sphere diameter allowing access of the arginine to the active site of a PRMT1 dimer is 8.9 Å (Kölbel et al., 2012). Because this is smaller than the smallest globular protein fold, it seems that arginine methylation must be limited to either relatively large loops emanating from globular folds or sections of the peptide chain lacking tertiary structure. In fact, RGG domains and other sequences containing methylated arginines have the attributes of intrinsically disordered or natively unfolded regions of proteins, for example, high net charge and the enrichment of arginine, glycine, and proline, all of which are considered ‘disorder-promoting’ amino acids (Dunker et al., 2001). The N-terminal ‘tails’ of histones are known to be disordered from crystal structures (Luger and Richmond, 1998). The 61-amino-acid-long C-terminal domain of nucleolin, a prototype RGG domain, lacks stable secondary or tertiary structure, at least in isolation (Zahariev et al., 2005). The C-terminal domain of PABPN1 can safely be assumed to lack stable structure, as it is a binding site for transportin (Calado et al., 2000a; Fronz et al., 2011), which is known to bind its cargo in an unfolded or mostly unfolded conformation (Lee et al., 2006; Cansizoglu et al., 2007; Imasaki et al., 2007; Zhang and Chook, 2012).

We have examined the importance of peptide backbone conformation for PRMT1-dependent methylation in PABPN1. Among the 13 methylated arginines, one particular residue, Arg289, is strongly preferred by PRMT1 in vitro. This preference is maintained in a synthetic peptide covering the region and depends on a proline residue in the -1 position. A glycine is present in the +1 position with respect to the preferred arginine. Proline and glycine residues are both strongly enriched in certain positions of reverse turns. A proline residue in the -1 position and a glycine residue in the +1 position would tend to put the arginine residue into an apical position of a reverse turn, in which the side chain points away from the peptide backbone forming the turn (Hutchinson and Thornton, 1994). We have proposed that this would be favorable for the arginine side chain to reach into the active site of a PRMT1 dimer. As a test of this hypothesis, a pentapeptide containing a Pro-Arg sequence was tested as a PRMT1 substrate in a linear and in a cyclic form. The cyclization of short proline-containing peptides is known to stabilize reverse turns with the proline in the i+1 position, thus putting the neighboring arginine into the i+2 position (Pease and Watson, 1978; Hollósi et al., 1987; Stradley et al., 1990; Kumaki et al., 2001). In fact, the cyclization of the pentapeptide led to a 15-fold increase in k_cat and a 6-fold increase in k_cat/K_m for PRMT1-dependent methylation. Thus, the data confirm that backbone conformation is important for substrate recognition by PRMTs. More specifically, they support the idea that a reverse turn facilitates the methylation of an arginine side chain, presumably either in the i+1 or i+2 position. Interestingly, based on modeling of complexes of PRMT4/CARM1 with substrate Yue et al. (2007) proposed that the peptide would have to adopt a β-turn structure for the arginine guanidino group to reach into the active site.

Many proteins contain their methylated arginines in either N- or C-terminal regions. Examples include yeast Npl3p (Siebel and Guthrie, 1996), PABPN1 (Smith et al., 1999), fibroblast growth factor (Klein et al., 2000), nucleolin (Lapeyre et al., 1986), hnRNPA1 (Kim et al., 1997), FUS (Rappsilber et al., 2003) and the related EWS protein (Belyanskaya et al., 2001), fibrillarin (Lischwe et al., 1985), and histones (Di Lorenzo and Bedford, 2011). Although this might facilitate the entry of the arginine side chains into the central hole and the active sites of PRMTs, methylated arginines also occur at internal sequences. For example, hnRNP K is methylated between two structured domains, KH domains 2 and 3 (Ostareck-Lederer et al., 2006), FUS and EWS are also methylated internally (Belyanskaya et al., 2001; Rappsilber et al., 2003), and the same is true for Sam68 (Côté et al., 2003; Rappsilber et al., 2003). The methylated RGG domain of Nab2p is found between an N-terminal α-helical domain and C-terminally adjacent zinc fingers (Soucek et al., 2012). Threading of the polypeptide chain into the PRMT dimer is not possible in these cases. Obviously, transient opening of the PRMT dimer and closure over the polypeptide chain is one possibility. However, looping of the substrate into the PRMT cavity is an attractive alternative. It may be relevant that glycine, the amino acid side chain most strongly enriched around arginine methylation sites, accelerates loop formation in model peptides (Krieger et al., 2005). Yet, the absence of steric obstacles that might prevent an arginine side chain from reaching the active site is clearly not the only criterion for substrate selection by PRMTs. For example, we have found that free arginine is not methylated to any detectable extent (Knut Kölbel and Elmar Wahle, unpublished data) and a single arginine embedded in a polyglycine environment is a poor substrate (Robert Puschmann, Bodo Moritz, and Elmar Wahle, unpublished data).

Functional consequences of PABPN1 methylation

In vitro methylation of PABPN1 and other proteins is not only slow but is also inefficient in the sense that we have been unable to introduce the full set of 26 methyl groups into recombinant PABPN1 in vitro. However, milligram quantities of fully methylated PABPN1 have been purified to homogeneity from tissue, and thorough analysis by mass spectrometry showed that, except for the acetylation of the N-terminus, arginine methylation was the only detectable modification of the protein. Thus, a comparison of the in vitro behavior of recombinant, unmethylated PABPN1 from E. coli and the protein purified from tissue has been the major approach to reveal functional consequences of arginine methylation.

There is currently no reason to believe that modified arginine side chains in PABPN1 could be demethylated; our attempts to detect such an activity in vitro were not successful. Thus, it is unlikely that PABPN1 undergoes cycles of methylation/demethylation as a means of a reversible modification of its activity.

Two functions of PABPN1 have been characterized in some detail in vitro. The first is the association with RNA: binding is specific for poly(A) or poly(G) and an isolated site is bound with a K_d in the low nanomolar range. RNA binding is mediated by the RRM and by the C-terminal domain, the latter contributing affinity but not specificity for A residues (Wahle et al., 1993; Meyer et al., 2002; Kühn et al., 2003). As described above, the functional unit of PABPN1 in RNA binding is a monomer. Nevertheless, functionally relevant oligomerization is indicated by a modest degree of cooperativity in RNA binding (Meyer et al., 2002; Kühn et al., 2003), and the formation of a higher order structure of the poly(A)-PABPN1 complex may be important for polyadenylation and length control (Keller et al., 2000; Kühn et al., 2009). Although the RRM of PABPN1 can dimerize (see above), the C-terminal domain also contributes to homotypic interactions and is responsible for the cooperativity of RNA binding (Fan et al., 2001; Kühn et al., 2003). Although arginine methylation occurs in a domain contributing to both RNA binding and cooperativity, the modification has a marginal effect at best, maybe a 2-fold increase in affinity and no detectable change in specificity or cooperativity (Kühn et al., 2003; Fronz et al., 2011). Likewise, although the related RGG domain of other RNA-binding proteins contributes to RNA binding, its methylation generally does not appear to affect this activity (Valentini et al., 1999; Raman et al., 2001).

The second function of PABPN1 studied in vitro is its role in mRNA polyadenylation, both the stimulation of poly(A) polymerase and the contribution to poly(A) tail length control. The stimulation of poly(A) polymerase requires the coiled coil domain, the RRM, and the C-terminal domain (Kerwitz et al., 2003). The requirement for the C-terminal domain is not explained by its role in RNA binding. However, like its binding to RNA, the function of PABPN1 in polyadenylation is not detectably affected by arginine methylation (Nemeth et al., 1995).

As mentioned above, PABPN1 also appears to stimulate the activity of the exosome. This function has not been examined in vitro, and the effect of arginine methylation remains to be examined.

The S. pombe orthologue of PABPN1 (Pab2) is also arginine-methylated in its C-terminal domain. In the absence of methylation, the protein has an increased tendency to self-associate or aggregate (Perreault et al., 2007). This appears to be a more general phenomenon. For example, the self-association of the yeast protein Npl3p is also increased in the absence of arginine methylation (Yu et al., 2004; McBride et al., 2007), and we have noticed a strongly increased background in immunoprecipitations from PRMT1^-/- cells, consistent with increased protein aggregation in the absence of methylation (Ostareck-Lederer et al., 2006). As the formation of fibrillar aggregates of PABPN1 is important in the etiology of OPMD (Brais et al., 1998; Calado et al., 2000b; Winter et al., 2013), an effect of methylation on the aggregation of the protein would be of particular interest. However, although several assays demonstrated a higher tendency of recombinant PABPN1 purified from E. coli to oligomerize or aggregate compared with the fully methylated protein purified from calf thymus, this was traced to the presence of an N-terminal His-tag, not the lack of methylation: the recombinant protein lacking the tag showed no more self-association than the methylated protein from calf thymus in pull-down experiments and did not display any aggregation in gel-shift experiments for binding to poly(A) (Fronz et al., 2008). Thus, at the concentrations used in such assays, the unmethylated protein behaves normally. The effects of methylation on aggregation at higher concentrations cannot be excluded. The recombinant protein aggregates in an erratic manner when present at high concentrations during purification, so the effect of methylation could not be evaluated more systematically for technical reasons (Scheuermann et al., 2003; Fronz et al., 2008).

As a nuclear protein, PABPN1 has to associate with nuclear import receptors. The import receptors of the importin β or karyopherin family associate with import cargo in the cytoplasm and mediate passage through the nuclear pore. In the nucleus, the small G protein Ran, in its GTP-bound form, binds the import receptor and displaces the cargo molecule. The directionality of import depends on the cytoplasmic Ran pool being kept in the inactive GDP form by cytoplasmic Ran-specific GTPase activating proteins and on nuclear Ran being maintained in the GTP form by nuclear guanine nucleotide exchange factors (Görlich and Kutay, 1999; Fried and Kutay, 2003; Pemberton and Paschal, 2005). A receptor mediating PABPN1 import is transportin or karyopherin β2, based on the following evidence: PABPN1 binds to transportin, but not importin α/β, in a RanGTP-sensitive manner (Calado et al., 2000a; Fronz et al., 2011). In in vitro import assays, PABPN1 is imported by transportin but not by importin α/β (Fronz et al., 2011). The canonical binding site for transportin has been described as a basic or hydrophobic cluster followed by an R/H/KX_2–5PY motif (PY-NLS) within an overall basic, unstructured region of the protein (Suzuki et al., 2005; Iijima et al., 2006; Lee et al., 2006; Cansizoglu et al., 2007; Imasaki et al., 2007). The C-terminus of PABPN1 provides a perfect match to this consensus, except that the distance between the PY motif and the upstream arginine is six amino acids (Figure 2). Given that the consensus in this respect is vague and the transportin-binding site overall is flexible (Cansizoglu et al., 2007; Süel et al., 2008), this deviation can apparently be tolerated. Binding to transportin in vitro is sensitive to a deletion of the last eight amino acids of PABPN1 and competed by a peptide comprising the last 25 amino acids (NLS peptide) (Fronz et al., 2011). The same peptide competes with transportin-dependent nuclear import in vitro (Fronz et al., 2011). The C-terminal domain is also sufficient for nuclear import in vivo (Calado et al., 2000a). Genetic data confirm that S. pombe Pab2 is indeed imported by the orthologue of transportin (Mallet and Bachand, 2013). However, for mammalian PABPN1, the existence of an alternative import pathway has recently been suggested (Mallet and Bachand, 2013). The association of PABPN1 with transportin is sensitive to arginine methylation, as initially shown by a comparison of recombinant PABPN1 with ‘authentic’ protein purified from tissue. In vitro methylation of recombinant PABPN1 by PRMT1 was sufficient to interfere with binding to recombinant transportin (Fronz et al., 2011). This is in agreement with crystal structures displaying contacts between transportin and arginine side chains in PY-NLSs from other cargo proteins (Lee et al., 2006; Cansizoglu et al., 2007; Imasaki et al., 2007; Zhang and Chook, 2012). Competition assays with the methylated and unmethylated NLS peptide showed that methylation reduces the affinity for transportin about 10-fold. More efficient competition by unmethylated, as compared with methylated peptides, was also observed in in vitro nuclear import assays. Interestingly, transportin and poly(A) compete for binding to PABPN1; thus, methylation favors the transfer of PABPN1 to poly(A). Transportin binding, but not poly(A) binding, also prevents the in vitro methylation of PABPN1 by PRMT1. The presence of methylated arginine residues in a number of known or predicted transportin cargoes suggests that the results obtained with PABPN1 may have general significance (Fronz et al., 2011).

What could be the purpose of weakening the PABPN1-transportin interaction by methylation? We have speculated (Fronz et al., 2011) that the answer to this question may be related to the fact that nuclear import is not irreversible: the directionality of nuclear import relies only on the dissociation of the import receptor-cargo complex by nuclear RanGTP; the passage of the complex through the nuclear pore itself is not directional. Thus, to the extent than an import receptor can associate with cargo in the nucleus and escape RanGTP, it should be able to export its cargo. Based on the sluggish activity of PRMTs and the inhibitory effect of transportin on arginine methylation, it is conceivable that newly synthesized PABPN1 is rapidly imported into the nucleus in an unmethylated state. Upon RanGTP-dependent dissociation in the nucleus, arginine methylation could then take place and favor the association of PABPN1 with poly(A) as opposed to a reassociation with transportin and return to the cytoplasm. However, PABPN1 is a shuttling protein (Calado et al., 2000a); shuttling may reflect the escape from nuclear RanGTP described above rather than the existence of a specific export mechanism. What effect might arginine methylation have on the result of shuttling, i.e., the steady-state distribution, as opposed to the (hypothetic) kinetic effect on the initial nuclear import of PABPN1? Conceivably, methylation might favor the long-term (steady-state) localization of the protein in the nucleus: In the cytoplasm, the concentration of free import receptors is high due to the low concentration of RanGTP and might suffice for the complex formation with cargo even if the affinity has been reduced by methylation. Thus, PABPN1 that has left the nucleus would be reimported despite its methylation. In the nucleus, the concentration of free import receptor is low due to its association with RanGTP, presumably below the K_d of the receptor-cargo interaction. Thus, a methylation-dependent drop in import receptor affinity would lead to reduced complex formation and, therefore, a lower rate of cargo escape to the cytoplasm.

Support for a role of arginine methylation in nuclear import and for one particular aspect of the above hypothesis has come recently from studies of S.pombe Pab2: this protein also contains a C-terminal PY-NLS and, as mentioned above, its nuclear import is mediated by the S. pombe orthologue of transportin. Although the inhibition of arginine methylation by the deletion of the responsible arginine methyltransferase had little effect on the nuclear import of the wild-type protein, it did rescue the import defect caused by a point mutation in the PY-NLS (Mallet and Bachand, 2013). This confirms that arginine methylation affects nuclear import in vivo. As the effect became visible only in the context of a point mutation weakening the association with transportin, the data specifically support an inhibitory effect of arginine methylation on the transportin-cargo interaction. Finally, the data agree with the idea that the cytoplasmic concentration of transportin is high enough to provide efficient nuclear import even when arginine methylation is functional. Only when the transportin-cargo interaction is weakened does the effect of methylation become visible.

An interesting evidence for a role of arginine methylation in fine-tuning the transportin-cargo interaction in vivo has been provided by investigations of a different transportin cargo, the protein FUS: mutations in the human FUS gene are responsible for a fraction of cases of familial amyotrophic lateral sclerosis (ALS), a fatal neurodegenerative disease of late onset. The FUS protein normally functions in the cell nucleus. However, in ALS patients carrying the mutation, the FUS protein forms cytoplasmic aggregates often associated with a reduced nuclear abundance. FUS is imported into the nucleus in a transportin-dependent manner. Interestingly, many of the known ALS-causing point mutations lie in the transportin-binding site, and some of them have indeed been demonstrated to reduce the efficiency of nuclear import. In a panel of mutations, the stronger effects on protein localization were correlated with an earlier onset of disease in patients (Dormann et al., 2010; Dormann and Haass, 2011). FUS is known to carry asymmetrically dimethylated arginine residues; in contrast to PABPN1, these are not in the canonical motifs of the transportin-binding site but in an adjacent RGG domain (Rappsilber et al., 2003). The inhibition of arginine methylation by the use of the general inhibitor adenosine dialdehyde, knockdown of PRMT1, or deletion of the PRMT1 gene rescued the nuclear import of FUS when it was compromised by mutations in its PY-NLS (Dormann et al., 2012; Tradewell et al., 2012). As in the case of PABPN1, the binding of FUS to transportin was reduced by arginine methylation (Fronz et al., 2011; Dormann et al., 2012). The inhibition of methylation had no detectable effect on the nuclear import of wild-type FUS (Dormann et al., 2012; Tradewell et al., 2012). Thus, like the data on S. pombe Pab2, the results suggest that the concentration of transportin in the cytoplasm is sufficiently high that arginine methylation has little effect on nuclear import unless a cargo mutation weakens the affinity for transportin.

Although there are many observations of arginine methylation affecting the nuclear-cytoplasmic distribution of various proteins, the effects have been very variable – sometimes cytoplasmic, sometimes nuclear localization being favored by methylation (discussed in Fronz et al., 2011). For example, Nab2p is imported by Kap104p, the yeast orthologue of transportin, binding to its RGG domain (Lee and Aitchison, 1999). Preventing the arginine methylation of Nab2p stops the nuclear export of this protein (Green et al., 2002). Nab2p appears to be exported not on its own but as part of an mRNP. For this and other reasons, it is unknown whether the effect of methylation is direct or indirect. In this context, it should be pointed out that even the data on the effect of arginine methylation on the localization of FUS may not be as clear-cut as it may seem: although there is agreement that the introduction of FUS mutated in its PY-NLS into cells incapable of arginine methylation rescues the protein’s nuclear localization, one study reported that the introduction of mutant FUS into cells simultaneously with an shRNA-depleting PRMT1 had the opposite effect, leading to enhanced cytoplasmic localization (Tradewell et al., 2012). This puzzling observation has not been further explored and remains unexplained. Clearly, more work is needed to understand the role of arginine methylation in nucleocytoplasmic transport of proteins.