Alternative RNA splicing complexes containing the scaffold attachment factor SAFB2

Alternative splicing affects up to 70% of human genes, and is thought to be a key factor enabling metazoan complexity (Johnson et al., 2003). Patterns of RNA splicing are thought to be established by a cellular splicing `code', in which differences in the expression levels of RNA splicing regulators determine the cellular splicing response of specific target RNAs, dependent on their collection of cis-acting binding sequences (Fu, 2004; Matlin et al., 2005). Alternative splicing regulators include the SR proteins, SR-related proteins such as Tra2β and hnRNPs. These act to either repress or enhance use of particular splice sites, and splicing differences can result from fairly subtle differences in the expression of ubiquitous splicing regulators (Park and Graveley, 2005). Alternative splicing patterns can also be driven by the expression of cell type-specific splicing regulators such as NOVA which is expressed in neural tissues (Ule et al., 2005). A further cell type-specific splicing regulator is the T-STAR protein (a.k.a. SLM2 and KHDRBS3). T-STAR is specifically expressed in the developing brain and adult testis, and belongs to the STAR family of proteins which link Signal Transduction and RNA processing, and which also contains Sam68 and SLM1 proteins (Stoss et al., 2004; Stoss et al., 2001; Venables et al., 2004).

Most of the biochemical details of the splicing process have been established in vitro, in which case spliceosomes form by stepwise assembly of snRNPs and spliceosomal proteins. However, a number of constitutive splicing factors are added to spliceosomes as pre-formed units, including the PRP19 complex (Makarov et al., 2002), addition of which activates the spliceosome for catalytic activity. Spliceosomes are thought to assemble de novo on nascent transcription complexes, although there is also evidence to suggest the existence of pre-formed supraspliceosomes, which splice transcripts (Azubel et al., 2006; Gornemann et al., 2005; Lacadie and Rosbash, 2005). Alternative splicing regulators in the SR family of proteins are highly mobile in the cell, and frequently exit splicing factor compartments to roam the nucleus to initiate new protein and RNA interactions (Misteli, 2001). However, whether individual alternative splicing regulators spend most of their time as monomeric proteins or whether they are stably and quantitatively associated in pre-formed complexes is not known.

Alternative RNA splicing regulation can be linked to transcriptional elongation (Batsche et al., 2006), transcriptional termination (Soller and White, 2004) and to signal transduction pathways (Stamm, 2002). The scaffold attachment protein SAFB1 has been implicated in splicing, signalling and transcription, so might play a role in each of these cellular processes (summarised in Fig. 1). SAFB1 has an RNA recognition motif (RRM), and so probably also directly interacts with target RNAs. SAFB1 interacts in the yeast two hybrid system with a number of alternative RNA splicing regulators including Tra2β, SF2/ASF, 9G8, SRp30c, YT521B, SRp86, SLM1, T-STAR (also known as SLM2), hnRNP A1 and hnRNP D (Denegri et al., 2001; Hartmann et al., 1999; Li et al., 2003; Nayler et al., 1998b; Stoss et al., 2001; Weighardt et al., 1999) and with protein kinases which phosphorylate splicing regulators (SRPK1 and CLK2) (Nikolakaki et al., 2001). Besides these yeast two hybrid interactions based on SAFB1 protein, very little is known about the existence of real endogenous protein interactions or heterocomplexes. However, SAFB1 can alter splice site selection in an E1A minigene, and inhibit splicing of a Tra2β-regulated exon (Nayler et al., 1998b; Stoilov et al., 2004). SAFB1 has also been reported as a component of in vitro assembled spliceosomes (Rappsilber et al., 2002).

SAFB1 has been also implicated in transcription. SAFB1 has a homeobox-like DNA binding domain called a SAF box (Fig. 1A), and was originally independently purified by column chromatography using an assay based on binding to a radio-labelled scaffold attachment region (SAR) DNA, and by screening a bacterial expression library with a DNA probe containing the HSP70 promoter (Oesterreich et al., 1997; Renz and Fackelmayer, 1996). Protein interactions detected by yeast two hybrid and protein pull down experiments have also been detected between SAFB1 and the carboxyl-terminal domain (CTD) of RNA polymerase II, which acts as a nucleation site for groups of proteins involved in RNA processing (Nayler et al., 1998b). SAFB1 interacts with several nuclear hormone receptors including the oestrogen receptor ERα (a transcription factor) (Debril et al., 2005; Oesterreich, 2003; Oesterreich et al., 2000; Townson et al., 2004), the tight junction protein ZO2 [also implicated in nuclear signalling (Traweger et al., 2003)] and with the DNA binding protein CHD1 (Tai et al., 2003).

Taken together, these results suggest that SAFB1 plays an integral role in coordinating several aspects of RNA synthesis, and its regulation via signalling pathways. More recently, a homologous SAFB2 gene has been identified, which is divergently transcribed from an adjacent gene on chromosome 19. SAFB2 encodes a protein which is 70.5% identical and 81% similar at the amino acid level to SAFB1 (Townson et al., 2003). This raises the question of whether SAFB1 and SAFB2 have identical or distinct cellular functions. To address this, we have raised monospecific antisera which specifically distinguish between the endogenous proteins. Our data are consistent with distinct properties for these proteins, with SAFB2 uniquely detected in stable higher molecular mass nuclear complexes, and having a distinct nuclear organisation. Surprisingly, other splicing regulators in the SR and SR-related protein families were also present in parallel but SAFB-independent protein complexes. These stable endogenous protein complexes might have important consequences for the in vivo selection of splice sites.

Whereas SAFB1 is a nuclear protein, SAFB2 has been reported to be in part cytoplasmic, where it interacts with vinexin (Townson et al., 2003). To test this we directly visualised the distribution of SAFB1-GFP and SAFB2-GFP fusion proteins and found both to be exclusively nuclear (data not shown). To enable us to analyse endogenous proteins we raised mono-specific antisera to both. We expressed unique regions as fusion proteins in E. coli (Fig. 1A), then used these to raise and affinity purify new specific antisera against SAFB1 (in rabbits) and SAFB2 (in sheep). Both antibodies recognised a single protein of approximately 175 kDa on western blots (Fig. 2) which was specifically and completely pre-absorbed by the corresponding immunising protein (supplementary material, Fig. S1).

To confirm the antisera are monospecific we expressed SAFB1-GFP, SAFB2-GFP and GFP alone separately in HEK293 cells, and probed western blots of whole cell lysates from these cells using our αSAFB1 and αSAFB2 antisera (Fig. 2). Both antisera recognised their corresponding endogenous proteins in each cell lysate (this corresponds to the lower band present in each lane). In addition, αSAFB1 recognised another slightly slower migrating protein corresponding to SAFB1-GFP only in the cells transfected with SAFB1-GFP (lane 1), and not in cells transfected with either SAFB2-GFP or GFP (lanes 2 and 3). Similarly αSAFB2 recognised a slightly slower migrating protein only in the cells transfected with SAFB2-GFP (lane 5), but not in cells transfected with either SAFB1-GFP or GFP (lanes 4 and 6 respectively). Hence we conclude our antisera are mono-specific, and do not cross react between SAFB1 and SAFB2.

We next compared the subnuclear localisation of endogenous SAFB1 and SAFB2 using confocal microscopy in HeLa cells, where we found that SAFB1 and SAFB2 had distinct nuclear distributions (typical results are shown in Fig. 3A,B). SAFB1 protein was primarily concentrated in discrete intra-nuclear foci in HeLa cells. Although SAFB2 was also present in these intranuclear regions (see arrows), it was more abundantly distributed within the general nucleoplasmic population. The cellular distribution of both SAFB1 and SAFB2 was cell type dependent: in HEK293 cells both proteins had a more general distribution throughout the nucleoplasm (typical data is shown in Fig. 3C).

The 175 kDa protein bands detected by our new monospecific antisera co-migrated with the protein detected by the pan-specific monoclonal antibody 6F7 (supplementary material, Fig. S1), which was raised to a peptide sequence (PEARDSKEDGRKF) conserved in both SAFB1 and SAFB2 (Townson et al., 2003). However, the sub-nuclear distribution pattern detected by indirect immunofluorescence and confocal microscopy in HeLa cells using monoclonal antibody 6F7 precisely co-localised with the distribution of SAFB1 but not SAFB2 (Fig. 3A,B). This is consistent with the 6F7 antibody primarily recognising SAFB1 in vivo, either because SAFB1 is more abundant, or perhaps because the 6F7 epitope is masked in SAFB2 by other protein interaction partners (SAFB2 is present in higher molecular mass complexes in vivo; see below).

In order to determine if the expression patterns of SAFB1 and SAFB2 can be quantitatively regulated during development, we carried out indirect immunofluoresence in adult human testis which contains multiple cell types that can be individually recognised (Fig. 3D). Whereas both SAFB1 and SAFB2 proteins were detected in the nuclei of germ cells, only SAFB2 protein was highly expressed in the nuclei of the somatic Sertoli cells.

Rat SAFB1 protein has been shown to inhibit splicing of a variable exon from the tra2β gene (Stoilov et al., 2004). To test if human SAFB2 might have similar splicing activity, we co-transfected the tra2β variable minigene with plasmids encoding human SAFB1-GFP and SAFB2-GFP fusion proteins, or GFP alone (Fig. 4). Both SAFB1 and SAFB2 strongly inhibited inclusion of this variable exon (lanes 1 and 2). In contrast to the effect of SAFB1 and SAFB2, Tra2β protein positively autoregulated splicing of this variable exon (compare the splicing pattern in lane 3 with Tra2β-GFP, with the effect of GFP alone in lane 4, and minigene only in lane 5) (Stoilov et al., 2004). Quantitative results from three independent minigene splicing assays are shown in Fig. 4B.

The inhibition of variable exon inclusion of the tra2β minigene implicates both the SAFB proteins in splicing regulation. In the case of SAFB1, this activity was maintained even after deletion of the RNA recognition motif, suggesting that it is mediated through protein-protein interactions depleting nuclear splicing factors, rather than direct RNA-protein interactions (Stoilov et al., 2004).

We used our monospecific antisera to test if SAFB1 or SAFB2 were quantitatively associated with similar or distinct protein complexes within the nucleus. Nuclear protein extracts from HEK293 cells, treated (Fig. 5B) or not (Fig. 5A) with micrococcal nuclease (MN), were fractionated on sucrose gradients under high salt conditions to analyse stable nuclear complexes, then we tested the distribution of endogenous proteins within fractions by SDS-PAGE and immunoblotting. SAFB1 protein was consistently detected in small complexes towards the top of the gradient most probably corresponding to monomeric protein (expected size around 170 kDa). By contrast, SAFB2 was present in much larger complexes (between fractions 8 and 14, corresponding to a molecular mass between that of the monomeric protein and 670 kDa). Prior MN treatment of the nuclear extracts removed the SAFB2 protein in fractions above around 700 kDa, showing that this fraction of nuclear SAFB2 protein is associated with nucleic acids (Fig. 5B). However, the bulk of SAFB2-containing complexes were not shifted by MN treatment, showing that these protein complexes are held together by intermolecular protein-protein interactions rather than through a common association with bridging nucleic acid molecules. By contrast, the distribution of SAFB1 protein on the sucrose gradients was not detectably affected by MN treatment, so most of this protein is not quantitatively associated with nucleic acids. Similarly to SAFB1, the transcription factor TAF_II15 [also a reported SAFB1-interacting protein (Townson et al., 2004)] also migrated as a somewhat smaller complex.

The above experiments suggested that SAFB1 is found in cells either as a monomer or as a component of a small nuclear complex, whereas SAFB2 is part of higher molecular mass nuclear protein complexes. We attempted to identify stable interacting components which might be part of these complexes by immunoblotting of potential protein partners previously identified from protein interaction screens and over-expression studies (see Fig. 1B and Introduction). Both α-SAFB1 and α-SAFB2 antisera efficiently immunoprecipitated their cognate proteins, and co-immunoprecipitated each other (Fig. 5C,D). This implies that although the bulk of SAFB1 is monomeric (where it is detected on the sucrose gradients), a small fraction must be associated with the SAFB2 protein. α-SAFB1 and α-SAFB2 also both efficiently co-immunoprecipitated endogenous Sam68 and hnRNP G proteins from HEK293 cells, whereas no immunoprecipitation was observed with IgGs prepared from normal sera (Fig. 5E,F). All protein-protein interactions detected were independent of RNA, since immunoprecipitations were carried out in the presence of Benzonase which digests nucleic acids.

We similarly used immunoblotting to test the distribution of Sam68 and hnRNPG proteins on the sucrose gradients, both of which we found to be stably and quantitatively associated with nuclear protein complexes as they migrated in excess of their molecular masses (Fig. 5A,B, lower two panels). hnRNPG has a monomeric molecular mass of 43 kDa, but migrated between 43 and up to 440 kDa; Sam68 has a monomeric molecular mass of 68 kDa, but migrated right across the gradient, although much of the protein was in complexes of less than 158 kDa. Both proteins were associated with nucleic acids, as micrococcal nuclease significantly shifted their distributions within the gradient.

We used immunoprecipitation to test for possible interactions of SAFB2 with components of the active spliceosome and transcriptional machinery (discussed in Fig. 1 and Introduction). The active spliceosome is characterised by the PRP19 complex, which contains a number of proteins including 116K, SKIP and PRP19 itself (Makarov et al., 2002). None of these proteins was immunoprecipitated by α-SAFB1 or α-SAFB2 (Fig. 5G and data not shown). Hence although a fraction of SAFB2 protein is associated with nuclear RNA, the SAFB2 protein complex is either not stably or quantitatively associated with active splicing complexes within the nucleus. Similarly, neither α-SAFB1 nor α-SAFB2 co-immunoprecipitated the large subunit of RNA polymerase II (Fig. 5H), and hence neither are stably associated with active (or inactive) transcription complexes.

Rat Slm2 is a Sam68-related protein which has also been reported as a SAFB1-interacting protein (Stoss et al., 2001). Quantitative expression of the human Slm2 orthologue T-STAR is normally restricted to the testis, where it may form a component of a cell type-specific splicing regulatory complex which includes the germ cell-specific protein RBM (Venables et al., 1999). To test whether human SAFB2 also interacts with T-STAR, we transiently expressed a T-STAR-FLAG fusion protein, and immunoprecipitated endogenous SAFB1 and SAFB2. In both cases T-STAR-FLAG was co-immunoprecipitated with SAFB1 and SAFB2, but not with a pre-immune antiserum (Fig. 6A).

Using a directed yeast two hybrid assay we mapped two adjacent regions in SAFB1 and SAFB2 that mediated the observed interactions with Sam68 and T-STAR (Fig. 6B). The glutamate and arginine (ER)-rich region of SAFB1 and SAFB2 interacted strongly with both Sam68 and T-STAR, whereas the downstream glycine (G)-rich region of SAFB1 and SAFB2 interacted with Sam68 and T-STAR only weakly. The Sam68 and T-STAR protein interactions were mediated by their C-termini including the RG region (rich in arginine and glycine), and the tyrosine rich C-terminal domain (YD region), but the RG-rich region alone was not sufficient for the interaction to take place. The tyrosine-rich C-terminal alone caused auto-activation in the yeast two hybrid system so could not be tested (data not shown).

The data above suggests that SAFB2 forms a core set of stable protein interactions with an exclusive subset of proteins involved in RNA processing. Other potentially interacting proteins involved in transcriptional control and signalling were not detected, so might be more transient interaction partners. We directly visualised a set of these proteins within the nucleus (Fig. 7). Consistent with the protein interaction data described above, SAFB2 had a similar distribution to Sam68 protein (Fig. 7A), but had a quite distinct nuclear distribution from the CTD of RNA polymerase II (Fig. 7B). Similarly, SAFB1 had a different nuclear distribution from the ERα (Fig. 7C; note that this image is of MCF7 cells, which express ERα, whereas 7A and 7B are of HeLa cells).

We next tested whether these nucleic acid-independent nuclear complexes were specific to SAFB2, Sam68 and hnRNPG, or might also apply to other splicing regulators. We immunoblotted fractions from the sucrose gradients shown in Fig. 5 to examine the mobility of endogenous proteins in the SR/SR-related family of alternative splicing regulators, individual members of which have also been reported to interact with SAFB1 (see Fig. 1B). Similarly to SAFB2, quantitative amounts of each of these proteins migrated more rapidly on the sucrose gradients than their expected molecular masses, and so were also associated with larger endogenous nuclear complexes (Fig. 8). These were also of a similar size to those detected for SAFB2. The mobility of these SR/SR-related protein complexes on sucrose gradients were largely MN insensitive, suggesting that they too are either formed or maintained independently of nucleic acids. An interesting exception was the case of Tra2β, which actually associated with slightly higher molecular mass complexes after MN treatment (compare the Tra2β panels in Fig. 8A and 8B). The higher molecular mass fractions of Tra2β, 9G8 and SRp75 each had a slightly reduced mobility in SDS-PAGE compared with the monomeric fractions, suggesting that these proteins are also possibly subject to post-translational modification, e.g. phosphorylation, which could control their entry into higher order nuclear protein complexes. Although Sam68 also appears in higher molecular mass fractions after MN treatment, this is probably a result of the release of proteins from the large nucleic acid-mediated complexes, which would otherwise be pelleted at the bottom of the gradient: consistent with this, the relative amount of Sam68 protein present in fractions 15-20 also increases after MN treatment.

We performed immunoprecipitations to test whether these other splicing regulators were stably associated with SAFB1 or SAFB2. However, under identical conditions to those of Fig. 5, monospecific α-SAFB1 or α-SAFB2 antisera did not co-immunoprecipitate Tra2β (Fig. 8C), global (detected using mAb 10H3) or specific members of the SR protein family (detected using antibodies specific for ASF/SF2 or 9G8) (Fig. 8F-H). These are true negative results since, (1) each of these proteins were present in the cell lysate and so available for immunoprecipitation (see input lane for each IP); and (2) in each of these experiments, as a positive control, we confirmed that both SAFB1 and SAFB2 had efficiently immunoprecipitated and observed their reciprocal co-immunoprecipitation (data not shown).

Although absence of co-immunoprecipitation can be caused by factors such as excessive ionic strength of the extraction buffers or masking of the interaction surfaces preventing access of antibodies, these results suggest that Tra2β and SR proteins are not part of the core stable protein complex which includes SAFB2. Although we were unable to detect a stable association with either SAFB1 or SAFB2, antibodies to Tra2β efficiently immunoprecipitated Sam68 protein. Hence, Sam68 is a shared stable component of physically separate Tra2β and SAFB protein complexes (Fig. 8D).

Both hnRNPA1 (monomeric size ∼34 kDa) and hnRNPH (monomeric size ∼49 kDa) also migrated on sucrose gradients in excess of their expected molecular mass, but in the main part in slightly smaller complexes than the SR proteins (Fig. 8). However, a minor fraction of hnRNPH was present in extremely large nuclear complexes (between the pellet and fraction 5 of the gradient). Similarly to the SR family of splicing regulators, a comparison of mobility of these proteins with and without micrococcal nuclease treatment indicated that most of the hnRNPA1 and hnRNPH proteins are maintained in complexes that are independent of nucleic acids. By immunoprecipitation we found that neither hnRNPA1 or PTB (also known as hnRNPI) were stably associated with either SAFB1 or SAFB2 in vivo (Fig. 8E,I).

SAFB1 has been described as a nuclear scaffold protein, splicing is frequently co-transcriptional (Listerman et al., 2006) and some splicing factors have been associated with chromatin in chromatin immunoprecipitation (CHIP) experiments (Swinburne et al., 2006). Therefore we considered the possibility that both the SAFB proteins and other splicing regulators might be largely attached to chromatin rather than being released into soluble nuclear extracts. In this case the splicing regulator complexes we detected in the above experiments might contain only a fraction of the total cellular amount of these proteins. To test this we analysed fractions from each stage of a nuclear extract preparation (Dignam et al., 1983) by immunoblotting: these are, a cytoplasmic fraction (C), a chromatin and nuclear matrix-associated fraction (P1), and proteins that precipitate on dialysis (P2) to produce the final soluble nuclear extract (NE), as described in the Materials and Methods (Fig. 9). In order to analyse stable complexes that exist in the nuclear extract (NE) at a higher ionic strength (0.3 M salt) than in the final extract (0.1 M), the glycerol gradient fractionation experiments were carried out using a HEK293 nuclear extract which had not been dialysed against buffer D. Hence the relative amount of soluble nuclear proteins compared to the proteins extracted with chromatin components (P1) corresponds to the addition of the `P2' and `NE' fractions. SAFB1 and SAFB2 were distributed between the chromatin (P1) and soluble nuclear fractions (NE+P2), but with more in the soluble nuclear extract (Fig. 9). We also found a large fraction of Sam68, 9G8 and to a lesser extent Tra2β and hnRNPG were also present in the soluble fraction. Hence we conclude that the proteins we are visualising in our immunoprecipitation and fractionation experiments represent a significant quantity of the total cellular protein levels.

SAFB1 expression is required for normal development in the mouse, since the effects of a genetic SAFB1 knockout are not compensated by normal physiological expression of SAFB2 (Ivanova et al., 2005). The data in this paper show that although these proteins are 70.5% identical at the protein level, they have distinct molecular properties: (1) SAFB2 is associated with much bigger nuclear protein complexes than SAFB1; (2) SAFB2 and not SAFB1 is detectably associated with nucleic acids, although the integrity of the core SAFB2 protein complexes was micrococcal nuclease-independent; (3) SAFB1 and SAFB2 have distinct subnuclear organisations. This was particularly evident in HeLa cells, in which SAFB2 has a much more general distribution throughout the nucleoplasm, where most of the active co-transcriptional splicing is likely to take place (Zeng et al., 1997); (4) although SAFB1 and SAFB2 are divergently transcribed from the same promoter region (Townson et al., 2003), they can be differentially regulated to give cell type-specific ratios of expression. In the testis, SAFB2 was expressed at high levels in Sertoli cells, whereas SAFB1 was poorly expressed in these cells. Hence the short promoter region between the two genes must contain cell type-specific regulatory elements. It is possible that SAFB2 has a specific splicing function in Sertoli cells. Despite these differences, a direct and stable physical association between SAFB1 and SAFB2 could be detected by immunoprecipitation. This is most easily explained by a model in which SAFB1 protein is primarily monomeric in HeLa or HEK293 cells, whereas most of the SAFB2 protein is recruited into multi-protein nuclear complexes with stable core components, including a fraction of the SAFB1. Alternatively, SAFB1 might be more weakly associated with these higher molecular mass complexes, and so more readily released under the conditions used for gradient fractionation.

A number of potential protein interaction partners for SAFB1 have been described in the literature from yeast two hybrid and transient over-expression studies (Fig. 1B), but much less is known about its endogenous partners. Under our immunoprecipitation conditions, we found endogenous SAFB1 and SAFB2 formed stable interactions with an exclusive subset of proteins amongst those previously described (Fig. 1B). Each of these core proteins are RNA binding proteins. Consistent with this, the cellular location of endogenous SAFB1 has been reported to be dependent on the integrity of cellular RNA rather than DNA (Chiodi et al., 2000). At the subnuclear level, the distribution of SAFB1 and SAFB2 largely overlapped with Sam68, whereas the CTD of RNA polymerase II and the ERα had distinct distributions. The stable protein interactions with Sam68 and T-STAR are mediated primarily by the ER-rich regions of SAFB1 and SAFB2 (amino acids 610-772 for SAFB1, and 619-791 for SAFB2). Glutamate/arginine-rich regions are seen in a number of nuclear proteins, and so might represent a common stable protein interaction domain (Hartmann et al., 1999). This region is different from those that interact with the oestrogen receptor ERα (the central region of the protein, incorporating amino acids 426-600), and TAF_II15 (amino acids 720-915, comprising the C terminus of the protein and including the glycine-rich region). ERα and TAF_II15 might interact more transiently with SAFB1/2, since they were only detectable by immunoprecipitation after protein cross-linking (Townson et al., 2003; Townson et al., 2004; Traweger et al., 2003). The other proteins shown in Fig. 1B which were not detected as stable partners in our assays might similarly fall into this group of transient interacting proteins. Although they might be either less stable or shorter in duration, these transient protein interactions are likely to be still physiologically important in mediating the interaction of SAFB and its stable partners with other proteins in the cellular milieu. Consistent with this, in our assays, although Tra2β was not detected as a stable interacting protein, both SAFB1 and SAFB2-GFP fusion proteins inhibited a Tra2β-dependent alternative splice (although this was based on over-expressed protein).

The stable complexes of alternative splicing regulators we have identified in this study are nucleic acid independent. There is much current interest in the role of non-snRNP protein subcomplexes in pre-mRNA splicing (Ajuh et al., 2000; Makarov et al., 2002). Indeed it has been argued that the spliceosome itself may be preassembled and added together as a pre-assembled unit, although there is equally evidence in favour of co-transcriptional spliceosome assembly. Endogenous SR protein complexes were physically independent of SAFB1 and SAFB2 as assayed by immunoprecipitation, but contained at least some shared components (Sam68 protein stably interacted with Tra2β). The existence of pre-assembled complexes of alternative splicing regulators might have important implications for the selection of alternative splice sites because, first, groups of interacting RNA binding proteins held together by strong protein interactions might have an increased avidity for target RNAs which contain binding sites for more than one member of the complex. This may collectively stabilise weaker interactions with alternatively regulated RNAs, and so assist in the selection of true alternative exons within pre-mRNAs (Singh and Valcarcel, 2005). Second, these complexes may represent molecular storage compartments. These models are not mutually exclusive and might apply differently to different splicing regulators. The SAFB2 complexes are associated with nucleic acids, suggesting a more active role in splicing, whereas the SR proteins are not, which might be more consistent with a role in storage.

The stoichiometry and repertoire of splicing regulator complexes may be affected by cell-type-specific modifications such as phosphorylation of ubiquitously expressed proteins affecting protein interactions. A number of the SR proteins associated with the higher molecular mass complexes migrated more slowly in SDS-PAGE, suggesting they are post-translationally modified in these complexes, possibly by phosphorylation. In very specific tissues such as the testis, SAFB probably interacts with cell-type specific protein homologues, including RBMY, hnRNPGT and T-STAR. In this case the different RNA binding capabilities of these proteins might act to stabilise the binding of different regulated transcripts in different cells, and might act to repress or activate splicing. Hence qualitative and quantitative adjustments to these splicing regulator complexes might affect splicing decisions within the cell.

SAFB1 (amino acids 0-758) in pBS (a kind gift from G. Biamonti, Pavia, Italy) was sub-cloned from SAFB1 in pEGFP-C1. Full-length SAFB1 was generated by combining SAFB1 from pEGFP-C1 and SAFB1 from pACTII (isolated in a yeast two hybrid screen with human T-STAR protein) into pGBKT7 (BD Biosciences/Clontech, CA, USA). SAFB2 (KIAA0138) in pBS was provided by Kazusa DNA Research Institute, Japan and was subsequently cloned into pGBKT7. Portions of SAFB1 and SAFB2 were cloned in frame with a (His)₆ tag in pET32a for antibody generation (Novagen, Merck Bioscience, Darmstadt, Germany) by PCR cloning with the regions described below. The T-STAR and Sam68 constructs in yeast two hybrid vectors pGBKT7 and pGADT7 were described previously (Venables et al., 2004). SAFB1 and SAFB2 yeast two hybrid constructs were generated by PCR cloning into pGADT7 (BD Biosciences, Clontech). The SAFB1 RRM region is defined as amino acids 406-609, the ER-rich region is amino acids 610-772, and the G-rich region is amino acids 773-915. SAFB2 RRM is amino acids 406-618, ER-rich is 619-791 and G-rich is 792-953. T-STAR FLAG was generated by PCR cloning full-length T-STAR into p3XFLAG-CMV-10 (Sigma). The FLAG immunoprecipitation control ESCO2 FLAG construct was provided by A. Trainer, University of Newcastle, UK. SAFB1 and SAFB2 GFP fusion constructs were generated in pGFP3 (Venables et al., 2004).

(1) Generation of mono-specific antisera to SAFB1 and SAFB2. Antisera against SAFB1 was raised in rabbits against His-tagged fusions of amino acids (A) 487-542 and (B) 595-654. SAFB2 antisera was raised in sheep against a His-tagged fusion of amino acids 907-957. These regions were chosen because of the low similarity between the two proteins. Immunisation was carried out by Scottish Diagnostics, Midlothian. Affinity purification was achieved by binding sera to a SulfoLink (Pierce) column with the appropriate His-tagged peptide attached and elution by glycine pH2.5 (Venables et al., 2004).

(2) Generation of affinity purified Tra2β antibody used for immunoprecipitation: a peptide corresponding to amino acids 24-37 of Tra2β1 protein (HGSGKSARHTPARS +C for the coupling) was coupled to ovalbumin and used to immunise rabbits. The serum was precipitated with 50% ammonium sulphate, resuspended and dialysed against PBS (phosphate-buffered saline). Polyclonal Tra2β1-specific antibodies were purified by affinity chromatography using the immunogenic peptide bound to Sulfolink (Pierce), following the protocol given by the manufacturer. The specificity of the antiserum was controlled by western blots using HeLa nuclear extract, and extracts from 293-EBNA cells over-expressing GFP-Tra2β.

The assay was used as previously described (Venables et al., 2004).

The preparation of splicing competent HeLa or HEK293 cell nuclear extract was directly derived from the protocol described previously (Bourgeois et al., 1999; Dignam et al., 1983). Cells were grown in 8 l of medium to a density of 300,000 cells/ml, centrifuged and washed with 5 volumes of PBS. The cells were then resuspended in the hypotonic buffer A (10 mM Hepes pH 7.9, 10 mM KCl, 1.5 mM MgCl₂, 0.5 mM DTT), kept on ice for 8 minutes, centrifuged for 5 minutes at 2000 g and resuspended in 2 volumes of buffer A. Cell lysis was performed using a Dounce homogeniser (B pestle) and the suspension was centrifuged for 7 minutes at 2000 g to separate the nuclei from the cytoplasmic fraction. The nuclei were centrifuged at 25,000 g for 15 minutes (rotor Swing 41, Beckman) and resuspended in 2.5 volumes of high ionic strength buffer C (20 mM Hepes pH 7.9, 0.42 M NaCl, 1.5 mM MgCl₂, 0.2 mM EDTA, 0.5 mM DTT, 0.5 mM PMSF and 25% glycerol), giving a final NaCl concentration of about 0.3 M. The extraction was performed again using a Dounce homogeniser (B pestle), and followed by an incubation on ice for 30 minutes under gentle mixing. The insoluble fraction, containing mostly chromatin and nuclear matrix components, was eliminated by centrifugation at 25,000 g for 20 minutes (pellet 1). The supernatant was used directly for the characterisation of SAFB1/B2 complexes on sucrose gradients (see Fig. 5), or dialysed for 5 hours at 4°C against buffer D (20 mM Hepes pH 7.9, 100 mM KCl, 0.2 mM EDTA, 0.5 mM DTT and 17.5% glycerol) to obtain splicing-competent extracts. The components that precipitated during dialysis were eliminated by a final centrifugation at 17,000 minutes for 15 minutes (pellet 2), and the final nuclear extract was divided into aliquots and frozen in liquid nitrogen.

1/5000 of each cellular fraction (cytoplasm, pellet 1, pellet 2, final nuclear extract) was analysed by SDS-PAGE and immunoblotted with the appropriate antibodies, as indicated in the legend of Fig. 9.

Three 10-25% sucrose gradients were run in parallel at 180,000 g for 14 hours, in a buffer containing 20 mM Hepes pH 7.9, 300 mM NaCl, 0.2 mM EDTA and 0.5 mM DTT. The first gradient contained a mixture of high molecular mass proteins that were used to calibrate the migration of protein complexes on the gradient. Note that the different sizes indicated in the figures represent the peak fraction for each protein marker and cannot be considered as accurate measurements of the size of the complexes. The two other gradients corresponded to parallel experiments with 600 μl of the same HEK293 nuclear extract, pre-treated or not with micrococcal nuclease (MN). The MN treatment was as follows: the extract was incubated with 1 mM CaCl₂ and 2000 IU/ml MN, for 20 minutes at 30°C, and the reaction was stopped by addition of 1 mM EGTA. For each experiment, 20 equivalent fractions were collected and precipitated with 10% trichloroacetic acid and 0.4 mg/ml sodium deoxycholate. After centrifugation, the pellets were washed with acetone, resuspended in Laemmli loading buffer and denatured by boiling, for SDS-PAGE and western-blotting analysis.

This was carried out as described previously (Venables et al., 2004). The pre-absorption assay was carried out by pre-incubating the appropriate antibody with His-tagged fusions of SAFB1 and SAFB2 immunising peptides (see Generation of antibodies) and mock pre-absorption was with the equivalent peptide from the other protein, i.e. SAFB2 amino acids 604-672 is the equivalent region to SAFB1 immunising peptide B (peptide B was the more immunogenic peptide injected for α-SAFB2 antiserum generation from initial pre-absorption assays, not shown), and SAFB1 amino acids 869-910 is equivalent to the SAFB2 immunising peptide. Primary antibodies were used at the following concentrations: αSAFB1 1:1000, αSAFB2 1:200, αSAFB1 and 2 (mAb 6F7) (AbCam) 1:100, αSam68 (Santa Cruz) 1:1000, αhnRNP G (Soulard et al., 1993), 1:1000, αTra2β (Nayler et al., 1998a) 1:1000, αhnRNP A1 (mAb 9H10) (Pinol-Roma and Dreyfuss, 1992), αSR (mAb 10H3) (Venables et al., 2005), αASF/SF2 (mAb 96) (Soulard et al., 1993), α9G8 (mAb 1C6), 1:10, αPTB (mAb BB7) (Chou et al., 2000), αPRP19 (Makarov et al., 2002), αRNA pol II CTD (mAb 4H8) (UpState) and αFLAG M2 (Sigma) 1:1000. For the analysis of the sucrose gradients, the following other antibodies were used: hnRNP H, 1:2000 (a kind gift from D. Black, UCLA, CA); Tra2β, N-terminal purified polyclonal antibody 1:2000; 9G8, N-terminal purified polyclonal, 1:5000; TAF_II15, monoclonal 8TA2B10, 1:1000 (a kind gift from L. Tora, IGBMC, Illkirch, France).

Immunoprecipitation (IP) for detection of endogenous proteins was carried out in a similar manner to immunoprecipitation of FLAG fusion proteins, except that untransfected cells were used. For IP of the T-STAR-FLAG fusion protein, HEK293 cells were grown in DMEM with 10% FBS. Transfections were performed with GeneJammer (Stratagene) at 70% confluency in 75 cm² culture flasks with 12 μg T-STAR FLAG or ESCO2 FLAG as a control. 24 hour after transfection (or untransfected cells in the case of the detection of endogenous protein interactions), cells were lysed with IP lysis buffer (Li et al., 2003) for 20 minutes on ice. Soluble material was separated by centrifugation and subsequently pre-cleared with protein A. Pre-cleared lysate was incubated with αSAFB1, αSAFB2, normal rabbit IgG, normal sheep IgG, αSam68 or αTra2β for 1 hour on ice and then overnight with BSA-blocked protein A. The protein A was washed five times with IP wash buffer and boiled with Laemmli buffer to elute for subsequent analysis by western blotting.

150 ng of TRA2-BETA [MG Tra (Stoilov et al., 2004)] was transfected into HEK293 cells using GeneJammer (Stratagene), with either SAFB1-GFP, SAFB2-GFP, Tra2-GFP (1 μg of each) or GFP alone (100 ng). Patterns of splicing were monitored using RT-PCR as described previously (Stoilov et al., 2004).

HEK293, MCF7 or HeLa cells were grown on glass coverslips in DMEM with 10% FBS, fixed in methanol at –20°C, permeabilised in 0.1% Triton X-100 and blocked with 10% horse serum in PBS. Primary antibodies were as described for western blotting. Fluorescent secondary antibodies were from Molecular Probes (Invitrogen). DNA was counterstained with DAPI (VectaShield mounting medium, Vector Laboratories). Fluorescent microscopy was carried out using a Zeiss AxioPlan 2 fluorescence microscope and images were captured with a Zeiss AxioCam HRm camera and processed using Zeiss AxioVision software. Laser scanning confocal microscopy was performed using a Zeiss LSM 510 and associated software. All figures were prepared using Adobe PhotoShop 7.0 software. Since the SAFB1 antibody was raised in a rabbit, we were unable to co-localise this directly with Sam68 and other primary antibodies raised in rabbits. In these cases we localised these proteins with SAFB2 only (antisera raised in a sheep).

We thank Ingrid Ehrmann and Sushma Nagaraja Grellscheid for comments on the paper; Giuseppe Biamonti (Pavia, Italy) for the SAFB1 clone; Stefan Stamm (Erlangen, Germany) for the pan anti-tra2β antiserum and for the tra2β minigene; Evgeny Makarov and Rheinardt Luhrmann (Göttingen, Germany) for the antisera to components of the PRP19 complex; Laure Jobert and Laszlo Tora (IGBMC, Illkirch, France) for the anti-TAFII15 antibody. Kate Sergeant was supported by a CRUK William Ross PhD studentship. This work was supported by Cancer Research UK, the Wellcome Trust and the Newcastle Healthcare Charity (through project grants to D.J.E.) and the Institut National de la Santé et de la Recherche Médicale, the Centre National de la Recherche Scientifique and the Association pour la Recherche contre le Cancer (C.F.B. and J.S.).