Spinal muscular atrophy (SMA) is a muscular disease characterized by the death of motoneurons, and is a major genetic cause of infant mortality. Mutations in the SMN1 gene, which encodes the protein survival motor neuron (SMN), are responsible for the disease. SMN belongs to the Tudor domain protein family, whose members are known to interact with methylated arginine (R) or lysine (K) residues. SMN has well-defined roles in the metabolism of small non-coding ribonucleoproteins (snRNPs) and spliceosome activity. We previously showed that SMN relocated to damaged interphase centromeres, together with the Cajal-body-associated proteins coilin and fibrillarin, during the so-called interphase centromere damage response (iCDR). Here we reveal that SMN is a chromatin-binding protein that specifically interacts with methylated histone H3K79, a gene expression- and splicing-associated histone modification. SMN relocation to damaged centromeres requires its functional Tudor domain and activity of the H3K79 methyltransferase DOT1L. In vitro pulldown assays showed that SMN interacts with H3K79me1,2 at its functional Tudor domain. Chromatin immunoprecipitation confirmed that SMN binds to H3K79me1,2-containing chromatin in iCDR-induced cells. These data reveal a novel SMN property in the detection of specific chromatin modifications, and shed new light on the involvement of a putative epigenetic dimension to the occurrence of SMA.
The survival motor neuron (SMN) gene is present at an inverted duplicated locus on chromosome 5 in human cells (Lefebvre et al., 1995). The telomeric copy, called SMN1, differs from the centromeric copy, called SMN2, by five nucleotides. Both genes are transcribed, and one of the nucleotide changes in SMN2 favors the deletion of one exon (the exon 7), resulting in the synthesis of a truncated unstable protein called SMNΔ7 and very little full-length functional SMN (Kashima and Manley, 2003; Lorson and Androphy, 2000; Lorson et al., 1999; Monani et al., 1999; Singh et al., 2006; Singh and Singh, 2011; Singh et al., 2009)}. Homozygous deletions or mutations in SMN1 are responsible for the development of the autosomal recessive disorder proximal spinal muscular atrophy (SMA), the most common genetic cause of infant mortality (Burghes and Beattie, 2009; Coovert et al., 1997; Melki, 1997). The severity of the disease depends on the ability of SMN2 to compensate for the absence of SMN protein (Lefebvre et al., 1997). The absence of functional SMN protein primarily affects motoneurons, leading to muscle paralysis and atrophy (Monani, 2005).
SMN is expressed ubiquitously and forms a large multiprotein complex with Gemin proteins (2 to 8) and Unrip proteins (Baccon et al., 2002; Carissimi et al., 2005; Carissimi et al., 2006; Charroux et al., 1999; Charroux et al., 2000; Gubitz et al., 2002; Pellizzoni et al., 2002). This complex is implicated in the nucleo-cytoplasmic transport of small nuclear RNAs (snRNAs) and in their maturation to small nuclear ribonucleoproteins (snRNPs) as part of the spliceosomal complex (Akten et al., 2011; Deryusheva et al., 2012; Fischer et al., 1997; Leung et al., 2011; Liu et al., 1997; Makarov et al., 2012; Massenet et al., 2002; Matera and Shpargel, 2006; Sebbag-Sznajder et al., 2012). Studies using different models of SMA have suggested that the severity of the disease could correlate with the inefficiency of snRNP maturation, and as a consequence, to defective splicing of target pre-mRNAs (Chari et al., 2009; Gabanella et al., 2007; Imlach et al., 2012; Lotti et al., 2012; Pellizzoni, 2007; Winkler et al., 2005; Workman et al., 2009; Zhang et al., 2008). The implication of SMN in such a fundamental molecular pathway in cells of all origins is difficult to reconcile with the selective involvement of the neuromuscular system in SMA. To that extent, recent studies suggested that loss of other non-snRNP biogenesis-related functions of SMN is important in the development of SMA pathology (Bäumer et al., 2009; Praveen et al., 2012). Recently, SMN2 exon 7 splicing was shown to be dependent on functional SMN protein, with an increased requirement in motoneurons (Jodelka et al., 2010; Ruggiu et al., 2012). Other recent studies suggested that SMN loss-of-function preferentially affected the U12-dependent minor spliceosome, which could account for the misregulation of U12-dependent specifically processed genes (Boulisfane et al., 2011; Lotti et al., 2012), possibly in a non-cell-autonomous-dependent manner (Imlach et al., 2012; Lotti et al., 2012). This implies that SMN may contribute, directly or indirectly, to a specific machinery involved in the coupled transcription/splicing of RNAs that are essential for the survival of motoneurons.
SMN is a Tudor domain-containing protein. Tudor domains are part of the ‘royal’ domain superfamily and recognize methylated arginine (R) or lysine (K) residues (Botuyan et al., 2006; Chen et al., 2011; Huang et al., 2006; Huyen et al., 2004). The SMN Tudor domain binds preferentially to symmetric dimethyl-arginine (sDMA) residues in Arg–Gly (RG)-rich sequences of spliceosome-associated Sm family proteins and in coilin, the major structural component of the nuclear domains known as Cajal bodies (CBs) (Boisvert et al., 2002; Côté and Richard, 2005; Friesen et al., 2001; Liu et al., 2012; Raska et al., 1991; Selenko et al., 2001; Tripsianes et al., 2011). The localization of SMN to CBs depends on its interaction with methylated and phosphorylated coilin (Hearst et al., 2009; Hebert et al., 2002; Hebert et al., 2001; Toyota et al., 2010) and phosphorylation of SMN itself (Petri et al., 2007; Renvoisé et al., 2012). Following hypomethylation of coilin, SMN complexes form individual nuclear bodies called Gemini bodies or gems juxtaposed to CBs (Liu and Dreyfuss, 1996). Apart from these two nuclear domains, SMN does not accumulate elsewhere in the nucleus under normal conditions, although diffuse nuclear staining can be observed by immunofluorescence (IF) (Renvoisé et al., 2006). However, following destabilization of interphase centromeres by the viral E3 ubiquitin ligase protein ICP0 (from herpes simplex virus type 1), we observed centromeric accumulation of three CB proteins: coilin, fibrillarin and SMN (Morency et al., 2007). We named this response the interphase centromere damage response (iCDR). We showed that coilin interacted with the damaged centromeric chromatin, suggesting that epigenetic changes to centromeric chromatin induced by ICP0 might be responsible for triggering the iCDR. In this study we analyzed the interdependency between coilin and SMN for their accumulation at damaged centromeres, and found that coilin was necessary but not sufficient for the targeting of SMN to centromeres. The Tudor domain of SMN was required for its centromeric accumulation. We investigated the requirement for both R- and K-methyltransferase activities, and found that depletion of K-methyltransferase disruptor of telomeric silencing 1-like (DOT1L) protein significantly affected the targeting of SMN to damaged centromeres. DOT1L induces the mono- and di-methylation histone H3 at K79. We therefore analyzed the capacity of SMN to interact with methylated H3K79. We found that SMN interacted with mono- and di-methylated H3K79 through its Tudor domain. We then performed chromatin immunoprecipitation analysis of centromeric SMN in iCDR-induced cells, and found that SMN interacted with mono- and di-methylated H3K79 within nucleosomes in centromeric chromatin. Overall, these data demonstrate that SMN is a chromatin-binding protein that recognizes specific histone markers through its Tudor domain.
Coilin is necessary but not sufficient for SMN relocation to damaged centromeres
The iCDR implicates three proteins that are normally concentrated in CBs. To determine whether all three proteins are dependent on each other for their relocation to damaged centromeres, the iCDR was induced by expression of ICP0 in cells depleted of one of these proteins. The efficiency of siRNAs targeting coilin, fibrillarin and SMN was tested by WB and IF (supplementary material Fig. S1A,C; Fig. S2A). The absence of centromeric colocalization of all proteins within each depletion background was also verified in the absence of ICP0 expression (supplementary material Fig. S1B,D; Fig. S2B). The nuclear distribution of coilin could appear ‘spotty’ in a subset of SMN-depleted cells, as reported previously (Girard et al., 2006; Shpargel and Matera, 2005); however, coilin did not colocalize with centromeres (supplementary material Fig. S1Bv). Overall, our data showed that depletion of SMN did not prevent the relocation of coilin (Fig. 1Aiv) or fibrillarin (supplementary material Fig. S2Civ) to damaged centromeres. Similar results were obtained for the centromeric relocation of coilin and SMN in fibrillarin-depleted cells (Fig. 1Aiii,Biii, respectively). Depletion of coilin did not affect the behavior of fibrillarin at damaged centromeres (supplementary material Fig. S2Cii), whereas it abrogated SMN relocation (Fig. 1Bii). All these data were reproduced with a second siRNA targeting coilin, fibrillarin or SMN (data not shown). These data suggest that SMN, but not fibrillarin, is dependent on coilin for its relocation to damaged centromeres. To confirm these behaviors in another biological context, ICP0 was co-expressed with a coilin mutant depleted of its RG box (HA–CoΔRG; Fig. 2). HA–CoΔRG was dominant-negative towards endogenous coilin because it induced the disappearance of coilin from CBs (Fig. 2ii). Similarly, fibrillarin and SMN did not show characteristic CB labeling in cells expressing HA–CoΔRG (Fig. 2iv,vi). None of these observations were made in cells expressing HA-tagged unmodified coilin (HA–Co; Fig. 2i,iii,v). Co-expression of HA–Co and ICP0 did not prevent relocation of coilin, fibrillarin or SMN to damaged centromeres (Fig. 2vii,ix,xi). In contrast, co-expression of HA–CoΔRG and ICP0 prevented relocation of coilin and SMN, but not fibrillarin, to damaged centromeres (Fig. 2viii,x,xii). It is not technically possible to check for the presence of ICP0 and the HA-tagged protein in the same cell together with labeling the endogenous CB protein of interest and centromeres; however our experimental procedure was designed so that all ICP0-positive cells were also positive for expression of the HA-tagged protein (data not shown). These data, together with those obtained by the knockdown approach, confirm that coilin is necessary for the relocation of SMN to damaged centromeres.
HeLaPV cells are known to express hypomethylated coilin (Hebert et al., 2002), and to display a subset of gems that are separate from CBs, suggesting a weak interaction between coilin and SMN in these cells (Liu and Dreyfuss, 1996) (Fig. 3Ai). Expression of ICP0 in HeLaPV cells induced the relocation of coilin and fibrillarin, but not SMN, to damaged centromeres, (Fig. 3Aii–iv). It is known that the RG box of coilin is modified by methylation of R residues, and that this modification favors the binding of SMN to coilin (Hebert et al., 2001; Hebert et al., 2002). Therefore, the latter results suggest that the presence of SMN at damaged centromeres might simply be a consequence of its interaction with fully methylated coilin. To test the validity of this assumption, we used the SYM10 antibody, which directly recognizes coilin methylated in its RG box (Boisvert et al., 2002), to detect the presence of methylated coilin at centromeres. Although SYM10 clearly labeled CBs in cells not expressing ICP0 (Fig. 3Bi,iii), no colocalization between coilin and SYM10 or centromeres and SYM10 was observed in ICP0-expressing cells (Fig. 3Bii,iv, respectively). Of note, the expression of ICP0 in cells is easily identified by the specific multiple spot pattern adopted by coilin (Fig. 3Bii) (see Morency et al., 2007) (Figs 1, 2; supplementary material Figs S1, S2). Although we cannot exclude the possible lack of sensitivity of the SYM10 antibody towards low amounts of fully methylated centromeric coilin, these data, together with those obtained in HeLaPV cells, suggest that coilin does not need to be methylated in its RG box to accumulate at damaged centromeres. Altogether, these data suggest that: (i) relocation of SMN to damaged centromeres is unlikely to be only a consequence of its interaction with methylated coilin (i.e., that other activities are required), and (ii) although centromeric coilin is necessary to induce the centromeric targeting of SMN, it is not sufficient.
The Tudor domain is necessary but insufficient for the accumulation of SMN at damaged centromeres
We then asked whether specific domains in SMN were required for its centromeric targeting. In SMN, several domains that are more or less important for its targeting to CBs have been described (Renvoisé et al., 2006). We focused our analysis mainly on two domains: (i) the Tudor domain, because it mediates the interaction of SMN with the sDMA isoforms of RG motif-containing proteins such as coilin and Sm core proteins (Boisvert et al., 2002; Hebert et al., 2002); and (ii) the ex7 domain, because its loss leads to the synthesis of an unstable SMN_Δex7 protein with a reduced capacity to self-oligomerize, which could partly explain the inability of SMN_Δex7 to compensate for the absence of full-length SMN in SMA patients (Lorson and Androphy, 2000; Lorson et al., 1998). Five EGFP–SMN fusion proteins expressing full-length or truncated SMN were evaluated by IF in terms of their ability to accumulate at centromeres in ICP0-co-expressing cells (Fig. 4). The proper expression and biological properties of all these proteins have been characterized in the presence and absence of endogenous SMN and in various cell types (Renvoisé et al., 2006). EGFP–SMN_FL expresses full-length SMN; EGFP–SMN_Δex7 (aa 1–278 plus the first four aa encoded by ex8) expresses SMN_Δex7 protein; EGFP–SMN_472Δ5 (aa 1–146) lacks the C-terminal half of SMN, but contains the Tudor domain; EGFP–SMN_N86 (aa 1–86) lacks the C-terminal half of SMN, including the Tudor domain; and EGFP–SMN_E143K is a mutant harboring a glutamic acid (E) to K substitution at position 134 in the Tudor domain, which has been identified in type I SMA patients (Lefebvre et al., 1998) (Fig. 4A). EGFP–SMN_FL was found in CBs in the absence of ICP0, and colocalized with centromeres in ICP0-expressing cells (70.6±17.9%; Fig. 4Bi,ii). EGFP–SMN_Δex7 and EGFP–SMN_472Δ5 showed ‘spotty’ nuclear patterns and colocalized with coilin in CBs in ICP0-negative cells, as reported previously (Renvoisé et al., 2006). Their re-location at centromeres in cells expressing ICP0 was significantly reduced compared to EGFP–SMN_FL [12.5±4.6% (P = 0.029) and 4.1±1.3% (P = 0.026) for EGFP–SMN_Δex7 and EGFP–SMN_472Δ5, respectively; Fig. 4Biii–vi]. Additional deletion of the Tudor domain induced the nucleoplasmic distribution of EGFP–SMN_N86, but no colocalization with centromeres in ICP0-expressing cells (Fig. 4Bvii,viii). Finally, EGFP–SMN_E143K, which could colocalize with coilin in a subset of CBs (Renvoisé et al., 2006), was also significantly affected for its accumulation at centromeres in ICP0-expressing cells (10.9±1.2%, P = 0.025) (Fig. 3Bix,x). In order to rule out any discrepancies in the EGFP–SMN proteins behavior due to variations in the levels of ICP0 rather than to the absence of specific SMN protein domains, we performed WB on co-transfected cells. No variations of ICP0 levels could be detected in the different samples (supplementary material Fig. S3). These data suggest that: (i) although a functional Tudor domain is required, its presence is insufficient for significant accumulation of SMN at damaged centromeres; (ii) the SMN_Δex7 protein, although able to colocalize with coilin in CBs, is significantly reduced in its capacity to relocate to centromeres, which confers to ex7, and therefore to the capacity of SMN to self-oligomerize, a particular importance for the interaction of SMN with destabilized centromeres.
Activity of type I and II PRMTs is not required to target SMN to damaged centromeres
The data concerning the probable hypomethylation of centromeric coilin and the requirement for the Tudor domain of SMN led us to hypothesize that, although coilin is required, other biological events are likely responsible for the relocation of SMN to damaged centromeres. Because Tudor domains are implicated in the interaction with methylated R and K, we decided to investigate the requirement for the activity of the following for the induction of centromeric SMN during the iCDR: type I PRMTs (PRMT-1, -3, -4 and -6), which induce asymmetric R di-methylation (aDMA); type II PRMT-5, which induces sDMA; and HKMT PR-Set7, which induces di-methylation of histone H4 at K20. We inactivated PRMTs and PR-Set7 using specific siRNAs and then induced the iCDR by expressing ICP0. siRNAs were first verified for their effect on their target protein (Fig. 5A). Decreases in PRMT5 and PR-Set7 substrates [histone H4 symmetrically di-methylated at R3 (H4R3me2s) and K20 (H4K20me2), respectively] were also assessed. Depletion of PRMT-1, -3, -4, -5, -6 or PR-Set7, and subsequent expression of ICP0, did not modify the behavior of SMN towards destabilized centromeres (Fig. 5Bi,ii; Table 1). Accordingly, addition of the methyltransferase inhibitor 5′-deoxy-5′-(methylthio)adenosine (MTA) prior to ICP0 expression did not prevent the iCDR, including centromeric SMN accumulation (data not shown). Overall, these data suggest that the absence of R methylation activity did not directly affect SMN redistribution to damaged centromeres, reinforcing the conclusion that centromeric accumulation of SMN is not dependent on fully methylated coilin.
, <50%; ++, 50–70%; +++, 70–90%; ++++, >90%.
Based on the centromere signal obtained with the human anti-centromere antibody (CREST) by immunofluorescence the ICP0 activity on centromeres in the absence of DOT1L seems much greater than in control cells.
When present, the number of centromeric SMN spots in ICP0-expressing cells is often fewer than in control cells.
The depletion of PRMT5 provokes the accumulation of cells without gems.
All ICP0-expressing cells show diffuse nucleoplasmic SMN.
N/A, not applicable.
DOT1L H3K79 methyltransferase activity is required for SMN accumulation at damaged centromeres
53BP1 is a double Tudor-domain-containing protein that was shown to interact with H4K20me2, and to a lesser extent H3K79me2 (Botuyan et al., 2006; Huyen et al., 2004). Single, double and triple methylation of H3K79 is dependent on DOT1L methyltransferase activity (Feng et al., 2002; Jones et al., 2008; Steger et al., 2008). We next performed experiments similar to those described above with inactivation of DOT1L. We were unable to validate an antibody that efficiently recognizes DOT1L by WB or IF; therefore, the efficiency of the DOT1L siRNA was confirmed by detection of a decrease in the amount of the DOT1L mRNA (Fig. 6A), and its substrate H3K79me2 (Fig. 6B). We initially confirmed that depletion of DOT1L did not affect the activity of ICP0 in centromeric protein degradation (data not shown). Depletion of DOT1L did not affect the location of SMN in CBs (Fig. 6Ci, ICP0− cell). Further expression of ICP0 in DOT1L-knockdown cells showed the classical ‘multi-dotted’ pattern typical of coilin accumulation at centromeres (Fig. 6Ci, ICP0+ cells). This confirmed that ICP0 was active when DOT1L was depleted and that the targeting of coilin to damaged centromeres was not affected by the absence of DOT1L. The same ICP0+ cells also showed that SMN did not colocalize with centromeric coilin. Co-staining of SMN and centromeres confirmed that SMN did not accumulate at ICP0-damaged centromeres in most DOT1L-depleted cells (Fig. 6Cii, arrow; Table 1), and that when it did so it was with a dramatically reduced signal (Fig. 6Cii, arrowhead; Table 1). To quantify this event precisely, DOT1L KD ICP0-transfected cells were counted (about 500 ICP0− or ICP0+ cells, in triplicate) and divided into three groups: (i) cells showing centromeric SMN; (ii) cells with SMN in gems/CBs; and (iii) cells with no visible gems/CBs (Fig. 6D). For control no-siRNA and siRNA control (siCtrl)-transfected cells, more than 80% of cells positive for ICP0 showed SMN at the centromeres. Very few ICP0+ cells (<5%) showed SMN in gems/CBs and very few ICP0− cells (<10%) did not contain gems/CBs. Cells depleted of DOT1L with siDOT1L showed a clear decrease in centromeric SMN (35±13%) and a concomitant increase in the proportion of ICP0+ cells with SMN in gems/CBs (50±14%). In addition, absence of DOT1L did not affect gem/CB formation as only 13±4% of cells did not contain gems/CBs, similar to the results of control and siCtrl-transfected cells. Overall, these data show that DOT1L activity is required for the targeting of SMN to damaged centromeres and suggest that SMN may relocate to damaged centromeres in a methylated H3K79-dependent manner.
SMN binds to mono- and di-methylated H3K79
Both a functional SMN Tudor domain and DOT1L are required for the targeting of SMN to damaged centromeres. This suggested that SMN binds to methylated H3K79 through its Tudor domain. GST pulldown experiments confirmed that SMN interacted with mono- and di-methylated, but not tri-methylated, H3K79. This interaction implied a functional Tudor domain because the SMN_E134K protein was unable to bind to methylated H3K79 (Fig. 7A). This interaction was specific for methylated H3K79 because di-methylated H4K20 did not interact with SMN. One should be cautious about the specificity of the antibodies for histone modifications. Therefore, we next performed pulldown assays using biotinylated peptides spanning the H3K79 region (aa 61–90) with unmodified, mono- or di-methylated K79 or a control peptide from histone H4 (aa 2–24). SMN from cellular extracts was captured by the mono- and di-methylated K79 peptides, although unmethylated H3K79 peptide also pulled down SMN to some extent, unlike H4 peptide (Fig. 7B). This confirmed that SMN could interact with methylated H3K79 and ruled out any artifact due to antibodies interacting non-specifically with modified H3K79. Together, these results demonstrate that SMN interacts with histone H3 molecules that are mono- or di-methylated at K79 and that the interaction occurs through its functional Tudor domain.
SMN is a chromatin-binding protein
To confirm that SMN could interact with methylated H3K79 in a chromatin context, we performed chromatin immunoprecipitation (ChIP) analysis of SMN in cells expressing ICP0 or its non-functional mutant FXE (ICP0mut) under tetracycline induction (Fig. 7C) (Gross et al., 2012). Expression of ICP0 induces the iCDR with centromeric SMN, whereas ICP0mut does not (Morency et al., 2007); therefore, binding of SMN to methylated H3K79 should increase in ICP0-expressing cells compared to controls. Cells were treated with tetracycline or not for 24 hours and then analyzed by ChIP (see Materials and Methods). We first verified that the amount of chromatin used for ChIP was similar for all samples, and that the sonicated chromatin in all samples was homogenous (supplementary material Figs S4, S5). ChIP analysis showed mono- and di-methylated H3K79 bound to SMN in ICP0-expressing cells (i.e. cells with centromeric SMN), confirming the binding of SMN to methylated H3K79 in the chromatin context (Fig. 7C). Overall, these data demonstrate that SMN is a chromatin-binding protein that specifically interacts with mono- and di-methylated H3K79 in damaged centromeric chromatin.
Our data demonstrate that the Tudor protein SMN, whose lack of function is responsible for the occurrence of the infant motoneuron decay-associated disease SMA, is a chromatin-binding protein that interacts with histone H3 that is mono- or di-methylated at K79 (H3K79me1,2). The chromatin-binding ability of SMN was elucidated by analysis of the iCDR, an enigmatic response to damage occurring at centromeres. Although its physiological significance remains elusive, study of the iCDR enables us to uncover unknown molecular features of the proteins involved. Centromeric accumulation of the iCDR-associated protein SMN is dependent on upstream events requiring the targeting of coilin to damaged centromeres. Although coilin is necessary, it is clearly insufficient to induce centromeric localization of SMN; activity of the H3K79 methyltransferase DOT1L is also required. The iCDR was discovered recently as a cellular response occurring after the destabilization of interphase centromeres by the viral protein ICP0 of HSV-1 (Morency et al., 2007). This response implies the relocation to damaged centromeres of three CB proteins: coilin, fibrillarin and SMN (see model in Fig. 8). SMN partners such as Gemin or Sm proteins, and U2 snRNAs, were not similarly detected at centromeres (Morency et al., 2007). This suggested that the molecular features that lead to the accumulation of the three proteins at centromeres are likely to be different from those that attract them to CBs. Coilin was shown to locate on metaphase chromosome from different Drosophila cells, possibly accumulating at centromeres (Liu et al., 2009). SMN was recently involved in proliferation and differentiation of stem cells in larval Drosophila CNS and male germline suggesting a role of SMN in cell division (Grice and Liu, 2011). Whether this SMN function could be related to its capacity to interact with centromeres in order to protect them in fast growing cells becomes a sensible question in light of the conclusion of the Grice and Liu study (Grice and Liu, 2011). Coilin is the main structural component of CBs and its absence provokes the disappearance of CBs (Hebert et al., 2001; Tucker et al., 2001). However, coilin deficiency in mouse and Drosophila models does not affect their viability at least in normal, unstressed conditions (Liu et al., 2009; Walker et al., 2009). CB formation depends on the post-translational modification of coilin by methylation and phosphorylation (Hebert et al., 2002; Toyota et al., 2010). We show in this study that the presence of coilin at damaged centromeres is necessary for the centromeric recruitment of SMN, but not of fibrillarin. Data obtained both in HeLaPV cells and by SYM10 antibody labeling suggest that centromeric coilin is hypomethylated, unlike CB-associated coilin; however, the data concerning the dependency of centromeric SMN localization on coilin methylation seem contradictory. On the one hand, hypomethylated centromeric coilin (in HeLaPV cells) is not sufficient to induce centromeric SMN, which suggests that centromeric coilin needs to be fully methylated to induce centromeric SMN; and on the other hand, data obtained by SYM10 labeling of ATCC HeLa cells, which showed centromeric SMN, suggest that centromeric coilin is indeed hypomethylated. Accordingly, the Y12 antibody, which recognizes Sm proteins and more generally sDMA in RG-rich repeats (Brahms et al., 2000), including those present in the coilin RG box (Hebert et al., 2002), does not label centromeres during the iCDR (Morency et al., 2007). To exclude any potential artifact due to lack of ICP0 activity in HeLaPV cells, which would prevent SMN from being targeted to damaged centromeres, we confirmed that ICP0 induces the proteasomal-dependent degradation of CENPs and their disappearance from centromeres. We did not observe any differences compared to ATCC HeLa cells (data not shown). A mutation of SMN in HeLaPV cells that would explain its absence at damaged centromeres can be ruled out because sequencing of SMN cDNA obtained from a pool of HeLaPV mRNAs did not reveal any amino acid modification compared to the original SMN sequence (data not shown). These results suggest that HeLaPV cells are likely to be defective in other activities that could explain the lack of targeting of SMN to damaged centromeres in spite of the presence of centromeric coilin. In contrast, the absence of certain post-translational modifications of coilin in HeLaPV cells could affect its activity at centromeres, which would indirectly affect centromeric SMN accumulation. For instance, the phosphorylation status of coilin might be different in HeLaPV cells, weakening its interaction with SMN. It would be interesting to assess the requirement of coilin phosphorylation for the targeting of both coilin and SMN at centromeres. SMN is a Tudor protein that interacts preferentially with sDMA of non-histone proteins, including coilin (Boisvert et al., 2002; Côté and Richard, 2005; Liu et al., 2012; Tripsianes et al., 2011). In mammalian cells, Tudor domains are found in about 30 proteins, which differ in terms of their capacity to bind either methyl-K or methyl-R residues (Chen et al., 2011; Taverna et al., 2007; Yap and Zhou, 2010). So far, none of these proteins has been shown to interact with both methylated K and R. Our study shows that SMN, in addition to its previously described capacity to interact with methylated R, also interacts with H3K79me1,2 through its functional Tudor domain. A recent in vitro study described the inability of the SMN Tudor domain to bind to methyl-K residues in synthetic peptides, as measured by isothermal titration calorimetry (ITC) binding assays (Tripsianes et al., 2011). In that study, a truncated version of SMN spanning the Tudor domain (aa 84 to 147) was used to perform ITC assays with peptides harboring methyl-R or -K residues. In accordance with these results, our data show that the inactivation of the H4K20 methyltransferase PR-Set7, unlike DOT1L inactivation, does not prevent relocation of SMN to damaged centromeres, and in fact no binding of SMN to H4K20me2 was detected. This suggests that the amino acid environment of the recognized methyl-K is important. It is also likely that the SMN Tudor domain, when contained within the functional full-length protein, possesses additional binding properties towards methyl-K. We showed that SMN_Δex7, although it contains an intact Tudor domain, was significantly affected for its accumulation at damaged centromeres. SMN_Δex7 has a reduced ability to self-oligomerize, which confers different biological functions on the protein (Lorson et al., 1998). Because basic Tudor domains, such as that in SMN, are likely to function in a protein homodimer to recognize ligands harboring methyl-R residues (Tripsianes et al., 2011), it is possible that SMN dimerization is required for its Tudor domain to be fully compatible with methyl-K residues. In addition, we showed that an SMN protein mutated in its Tudor domain (SMN_E134K), which matches an SMN mutant protein found in SMA patients (Lefebvre et al., 1998), did not bind to H3K79me1,2, which suggests that the SMN Tudor domain has specificity towards methylated H3K79. We also showed that a functional Tudor domain was required for SMN to efficiently accumulate at damaged centromeres. This suggests that the SMN Tudor domain is essential for centromeric SMN activity during iCDR.
Other important data supporting the link between H3K79me and SMN show a significant decrease in the number of ICP0-expressing cells with centromeric SMN upon DOT1L knockdown. The yeast protein DOT1 and its human homolog, DOT1L, are the only known methyltransferases with specificity for H3K79 (Nguyen and Zhang, 2011; Steger et al., 2008). Its absence in knockout yeast, flies, and mice results in the complete loss of all forms of methylated H3K79 (Jones et al., 2008; Shanower et al., 2005; van Leeuwen et al., 2002). The methylation of H3K79 by DOT1/DOT1L results from trans-histone cross-talk that was initially found in combination with mono-ubiquitylation of H2B at K123 and K120 in yeast and mammals, respectively (Briggs et al., 2002; Krogan et al., 2003; Ng et al., 2003b; Ng et al., 2002; Sun and Allis, 2002; Wood et al., 2003). However, recent reports showed that several other types of H2B K ubiquitylation could lead to H3K79 methylation (Chatterjee et al., 2010; Whitcomb et al., 2012; Wu et al., 2011). Therefore, although exclusively catalyzed by DOT1/DOT1L, the methylation of H3K79 is a complex event in which several protein complexes are involved. H3K79 methylation is a hallmark of actively transcribed chromatin, but is also involved in DNA damage repair (DDR) of double-strand breaks (DSB) and ultraviolet radiation-associated DNA lesions (Giannattasio et al., 2005; Huyen et al., 2004; Ng et al., 2003a; Schübeler et al., 2004; Steger et al., 2008). SMN is unlikely to be recruited to damaged centromeres due to NER or a DSB-associated H3K79me-dependent DDR mechanism, as occurs with 53BP1, another Tudor domain protein (Giannattasio et al., 2005; Huyen et al., 2004). Indeed, proteins implicated in various repair mechanisms (non-homologous end joining, nucleotide excision repair, homologous recombination) and phosphorylation of H2Ax at S139 (a hallmark of DDR) do not accumulate at centromeres during the iCDR (Morency et al., 2007). One study reported that H3K79me2 associates with centromeric minor satellite chromatin in mouse cells (Jones et al., 2008). Together with other heterochromatin marks (H3K9me2 and H4K20me3), levels of this histone mark are significantly reduced in ES cells from DOT1L-knockout mice (Jones et al., 2008). This results in the accumulation of aneuploid cells during mouse embryogenesis and in death of the embryos at around 10 days post-coitum (Jones et al., 2008). No such data are available for centromeric chromatin in human cells. However, our ChIP experiments using SMN obtained from ICP0-expressing cells as the target confirmed that SMN interacts with H3K79me1,2 in the centromeric nucleosome, at least during the iCDR. At this stage, we cannot tell whether methylated H3K79 is associated with centromeric nucleosomes before ICP0 is expressed (its accessibility is increased after ICP0 chromatin destabilization (Gross et al., 2012), or whether destabilization of centromeres by ICP0 activates DOT1L, which in turn induces H3K79 methylation in centromeric nucleosomes. ICP0 is a RING finger (RF) protein whose RF-associated E3 ubiquitin ligase activity has been described in vivo and in vitro (Boutell and Everett, 2003; Boutell et al., 2002; Hagglund and Roizman, 2004). Many ICP0 substrates have been identified, but given the many interactions of ICP0 with cellular proteins, it is highly unlikely that all its partners and substrates have been discovered. One can hypothesize that ICP0 directly or indirectly induces H2B K ubiquitylation, which in turn activates DOT1L to induce methylation of H3K79. During the iCDR, coilin acts upstream of SMN and is required for the presence of SMN at centromeres. This suggests that whatever the effect of ICP0 on histone modifications, other epigenetic features, probably associated with the presence of coilin, induce trans-histone cross-talk leading to the activation of DOT1L, the methylation of centromeric H3K79, and finally the targeting of SMN to damaged centromeres. Multiple attempts to detect a direct interaction between in vitro synthesized SMN and biotinylated/methylated H3K79 peptides failed. Therefore, we cannot rule out an interaction between SMN and DOT1L or one of its partners that would connect SMN to methylated H3K79. However, (i) DOT1L does not contain an RG box and is not known to be methylated at R residues, which would account for a possible interaction with the SMN Tudor domain; (ii) no studies conducted to detect DOT1/DOT1L partners revealed the presence of SMN in DOT1/DOT1L protein complexes (Bitoun et al., 2007; Mohan et al., 2010; Mueller et al., 2007; Mueller et al., 2009; Park et al., 2010); (iii) SMN does not bind to H3K79me3 whose formation is also catalyzed by DOT1/DOT1L; and (iv) SMN functional Tudor domain is required for its interaction with methylated H3K79. All these data highly suggest that SMN directly bind to methylated H3K79. Although SMA is one of the most studied genetic disorders, the mechanism that leads to motoneuron loss is poorly understood. Because SMN is involved in the biogenesis of spliceosomal snRNPs, many studies of molecular aspects of SMA are focused on deciphering the impact of SMN loss on snRNP biogenesis and activities (Chari et al., 2009; Gabanella et al., 2007; Imlach et al., 2012; Lotti et al., 2012; Pellizzoni, 2007; Winkler et al., 2005; Workman et al., 2009; Zhang et al., 2008). SMN makes a crucial contribution to the activity of the splicing machinery by interacting with spliceosome-associated proteins, whereas methylation of H3K79 is positively correlated with the transcription process. It is now widely accepted that maturation of nascent primary transcripts, including splicing, could occur co-transcriptionally (Kornblihtt, 2007; Manley, 2002; Neugebauer, 2002; Pandit et al., 2008; Perales and Bentley, 2009). Recently, another level in the cross-talk between transcription and splicing has been identified: histone modifications of chromatin (Alexander and Beggs, 2010; Brown et al., 2012; Hnilicová and Staněk, 2011; Luco et al., 2011). Notably, H3K79me1 was found to be enriched, together with other histone modifications, in nucleosomes positioned over exons, suggesting that H3K79 methylation has a role in splicing (Andersson et al., 2009; Dhami et al., 2010; Ernst and Kellis, 2010; Schwartz et al., 2009). Although the hypothesis that widespread splicing abnormalities cause SMA is still a matter of debate (Bäumer et al., 2009; Praveen et al., 2012), the requirement for functional SMN during early postnatal development for the splicing of specific transcripts involved in motoneuron survival has been suggested (Bäumer et al., 2009; Zhang et al., 2008). Accordingly, it has recently been shown that SMN loss-of-function affects the U12 splicing pathway in Drosophila, mammalian cells and human lymphoblasts from SMA patients (Boulisfane et al., 2011; Lotti et al., 2012). U12 splicing pathway is related to the so called minor spliceosome involved in the processing of a small proportion (about 1%) of introns (Patel and Steitz, 2003). Therefore, one of the ideas currently proposed is that the absence of functional SMN may selectively target genes processed through specific pathways; U12-dependent splicing being one of them (Imlach et al., 2012; Lotti et al., 2012). Even so, not all U12 intron-containing genes are affected by the absence of SMN (Lotti et al., 2012), and other studies in Drosophila suggested that viability defects in Smn null mutants are unlikely due to minor spliceosome level perturbations (Praveen et al., 2012). By demonstrating for the first time the capacity of SMN to bind to H3K79me1,2-containing chromatin, our study suggests an additional molecular level concerning the role of SMN in SMA. It is indeed tempting to speculate that beside regulating a specific splicing pathway, SMN plays an essential role in motoneuron and/or muscle cell homeostasis by connecting H3K79 methylation-dependent gene expression and splicing of particular transcripts, some of which potentially involved in motoneuron and/or muscle cell survival.
Materials and Methods
Cells and plasmids
HeLa cells were grown at 37°C in Glasgow's modified Eagle's medium. HeLaPV, HeLa TREX (TR, Invitrogen), TREX-ICP0 (TR-ICP0), and TREX-FXE cells were grown at 37°C in DMEM. All media were supplemented with 10% fetal bovine serum, L-glutamine (1% v/v), 10 U/ml penicillin, and streptomycin. For T-REX cells, blasticidin (5 µg/ml) was added to the medium; for TR-ICP0 and TR-FXE cells, blasticidin and zeocin (100 µg/ml) were added. TR-ICP0 and TR-FXE cell lines, which stably and inducibly express ICP0 and FXE, respectively, were constructed by transfecting T-REX cells with a pcDNA 4/TO plasmid (Invitrogen), which was designed to express ICP0 or FXE. Cell clones were then isolated under Zeocin selection. Expression of ICP0 or FXE was induced by addition of tetracycline (1 µg/ml) to the medium and was confirmed by western blotting and immunocytochemistry (Gross et al., 2012).
The pci110 plasmid expressing ICP0 protein, has been described previously (Everett et al., 1993). The plasmid pHA-coilin was constructed by in-frame cloning of the coilin cDNA into a pcDNA3.1 plasmid containing the HA tag sequence. pHA-coilinΔRG expresses a coilin protein in which amino acids (aa) 388–419 (including its RG box) are deleted. Plasmids expressing various EGFP-tagged SMN proteins (pEGFP-SMN, pEGFP-SMNΔex7, pEGFP-SMN_472Δ5, pEGFP-SMN_N86) have been described previously (Renvoisé et al., 2006) and were a generous gift from S. Lefèbvre (Institut Jacques Monod, Paris). The plasmid pGFP-SMN_E134K was constructed from pGFP-SMN by introduction of a single mutation that replaces the glutamine residue at position 134 with a lysine residue using the QuickChange Site-Directed Mutagenesis Kit (Stratagene). Plasmids expressing GST-SMN and GST-SMN_E134K were constructed by in-frame cloning of SMN and SMN_E134K cDNA into pGEX vectors (Clontech). All new constructs were checked for their sequences and in-frame cloning by sequencing.
Cellular and histone extracts, GST pulldown, peptide pulldown, and western blotting
To obtain cellular extracts, cells were washed once with phosphate-buffered saline (PBS), scraped off the flask and centrifuged at 1000 g for 5 minutes. The resulting cell pellet was resuspended in 200 µl of lysis buffer [15 mM Tris-HCl (pH 7.5), 2 mM EDTA, 0.25 mM EGTA, 15 mM NaCl, 0.3 mM sucrose, 0.5% Triton X-100, 300 mM KCl, protease inhibitor cocktail; Roche]. Samples were incubated on ice for 30 minutes, and centrifuged at 6000 g for 5 minutes at 4°C to remove all debris. Histone extraction was performed by scraping cells off the plates, resuspending them in PBS-phenylmethylsulfonyl fluoride (PMSF) and sonicating them in an ultrasonic water bath twice for 15 seconds. Nuclei and debris were centrifuged at 12,000 g for 10 minutes at 4°C. Pellets were resuspended in HCl (250 mM) and incubated on ice for 30 minutes before centrifugation at 12,000 g for 20 minutes at 4°C. Trichloroacetate was added to the supernatant at a final volume of 20%, and incubation was carried out for a further 30 minutes on ice before centrifugation at 12,000 g for 20 minutes at 4°C. Pellets were washed once with 500 µl of a mixed acetone/HCl (10 mM) solution and once with acetone alone, and were then dried in a SpeedVac concentrator (Savant) for 5 minutes before resuspension in water. For GST pulldown assays, 5 µg of GST fusion protein preparation was mixed with 10 µg of histone extract in Bis-Tris-propane (BTP) buffer [25 mM BTP (pH 6.8), 0.5% Triton X-100, 1 M KCl, protease inhibitor cocktail]. All extracts were first incubated for 1 hour at 4°C with continuous mixing with beads linked to the GST protein alone in order to reduce the background signal. The pre-cleared extracts were then incubated overnight at 4°C with continuous mixing with the appropriate GST fusion protein beads or a negative GST bead control. The beads were harvested by brief centrifugation, and were then washed five times with BTP buffer [25 mM BTP (pH 6.8), 1 M KCl]. Protein complexes were directly eluted from the beads by adding Laemmli buffer and were then boiled.
For peptide pulldown, 5 µg of biotinylated/methylated peptides corresponding to histone H3K79, H3K79me1 or H3K79me2 (aa 61–90; biot-lirklpfqrlvreiaqdfk79tdlrfqssavm; ProteoGenix) or histone H4 [aa 2–24; Millipore (12-372)] were first incubated with streptavidin–agarose beads (Pierce) for 2 hours at 4°C in PBS. The beads were then extensively washed with PBS to remove unbound peptides. Cellular extracts were then added to the beads/peptides and incubated in BTP buffer [25 mM BTP (pH 6.8) 0.5 M KCl, protease inhibitor cocktail] for 18 hours at 4°C. Samples were washed six times with BTP buffer and then Laemmli buffer was added. For western blotting (WB), samples were loaded on SDS-polyacrylamide gels before electrophoresis, transfer, and detection using appropriate antibodies.
Chromatin immunoprecipitation assays
Cells were seeded on Petri dishes (5×106 per dish). The following day, ICP0 expression was induced by adding tetracycline (1 µg/ml) to the medium for 24 hours. Formaldehyde was added to the medium at a final concentration of 0.5% for 5 minutes at room temperature (RT). Cells were washed twice with cold PBS, scraped off the dish, resuspended in 1 ml of buffer A [100 mM Tris-HCl (pH 9.4), 10 mM DTT] at a concentration of 5×106 cells per ml, and incubated at room temperature (RT) for 15 minutes. They were then incubated at 30°C for a further 15 minutes and centrifuged at 1000 g. Cells were lysed in 1 ml of buffer B [10 mM HEPES (pH 6.5), 10 mM EDTA, 0.5 mM EGTA, 0.25% Triton X-100] at 4°C for 5 minutes and centrifuged at 3000 g. They were then resuspended in 1 ml of buffer C [10 mM HEPES (pH 6.5), 10 mM EDTA, 0.5 mM EGTA, 200 mM EGTA] and incubated at 4°C for 5 minutes, before centrifugation at 3000 g. Nuclei were then resuspended in 300 µl of buffer D [50 mM Tris-HCl (pH 8), 10 mM EDTA, 1% SDS, protease inhibitor cocktail; Roche] and incubated at RT for 10 minutes, before sonication with a Bioruptor device (Diagenode). The sample volume was adjusted to 3 ml with buffer D′ [50 mM Tris-HCl (pH 8), 10 mM EDTA; protease inhibitor cocktail; Roche] and an antibody specific for SMN (2 µg/ChIP) was added before incubation overnight at 4°C. Salmon sperm DNA/protein G agarose (200 µl; 50% slurry; Millipore) was added to each sample at 4°C for 2 hours. Then, the beads were washed five times with buffer D, and Laemmli buffer was added. De-crosslinking was performed by heating the samples at 95°C for 90 minutes. Samples were then loaded on SDS-PAGE gels for WB.
Cells were seeded at a very low density before siRNAs transfection. Two rounds of siRNA transfection with Effectene Transfection Reagent (Qiagen) were performed at 48-hour intervals, according to the manufacturer's protocol. The amount of siRNA was adjusted to the number of cells. At the end of the experiment, cells were harvested and protein levels were quantified. The proteins were then resuspended in Laemmli buffer and analyzed by WB.
The following siRNAs were used: control, 5′-UACAGCUCUCUCUCGACCC-3′; coilin_1, 5′-CGACUGCCUCAGAGUUAAA-3′; coilin_2, 5′-CCAACUGAGUUAUCAAAGG-3′; fibrillarin_1, 5′-UGAGGGUGUCUUCAUUUGU-3′; fibrillarin_2, 5′-GGACCAACAUCAUUCCUGU-3′; SMN_1, 5′-GCAUGCUCUAAAGAAUGGU-3′; SMN_2, 5′-AGACGGUUGCAUUUACCCA-3′; DOT1L, 5′-GCUCGCUAUGGAGAAUUAC-3′; PR-Set7, 5′-AAUCGCCUAGGAAGACUGAUC-3′; and PRMT5, 5′-GGCCAUCUAUAAAUGUCUG-3′. The siRNAs targeting PRMT1 (L-010102), PRMT3 (L-026786), PRMT4 (L-004130) and PRMT6 (L-007773) we used were ON-TARGET plus SMARTpool siRNAs (Thermo Scientific).
RNA extraction and RT-qPCR
HeLa cells not-transfected or transfected with control siRNA (siCtrl) or siDOT1L following the above described procedure were harvested, and total RNA extracted using Nucleosin RNA II columns (Macherey-Nagel). Reverse transcription (RT) was performed with 1 µg of RNA from each sample and using random primer [0.2 µm] and 200 units of Revert-Aid H M-MulV Reverse Transcriptase (Fermentas). Quantitative PCR was performed on cDNAs obtained from the RT, using light cycler 480 SYBR green I Master (Roche), and ran on a MX 3000 P apparatus (Stratagene). Data were normalized on human beta actin.
Primers used for the qPCR were as follow: DOT1L_fwd: 5′-CATCCGATGGGTCTGTGA-3′, DOT1L_rev: 5′-TGGTGTCATAGTCAATTAAAACGTAATTC-3′, β-actin_fwd: 5′-CGGGAAATCGTGCGTGACATTAAG-3′, β-actin_rev: 5′-GAACCGCTCATTGCCAATGGTGAT-3′.
Immunofluorescence and confocal microscopy
IF was performed as described previously (Morency et al., 2007). Cells were seeded at 1.5×105 per well in 24-well plates containing a round coverslip prior to IF analysis. All samples were examined under a meta-confocal microscope (LSM 510; Carl Zeiss, Wetzlar, Germany) at a resolution of 512×512 pixels using 0.8–1.0-µm-thick optical slices. A Zeiss Axiovert 200 M microscope was used at 63× (NA 1.25) or 100× (NA 1.3) magnification under oil-immersion objectives. Datasets were processed using LSM 510 software, and then ImageJ software (Rasband, W.S., ImageJ, US, National Institutes of Health, Bethesda, Maryland, USA. http://imageJ.nih.gov/ij/) and Adobe Photoshop.
The following antibodies were used: mouse monoclonal antibodies (mAbs) against coilin (Sigma), fibrillarin [38F3] (Abcam) and [72B9] (a kind gift from J. Cavaillé, LBME, Toulouse), SMN (BD Biosciences), ICP0  (a kind gift from R.D. Everett, CVR-Glasgow), and H4 (clone 62-141-13; Millipore); rabbit polyclonal antibodies against actin (Sigma), coilin (coilin 20410; Dundee Cell Products), GFP (Roche), PRMT1 (Millipore), PRMT3 (ab3765; Abcam), PRMT4/CARM1 (Cell Signaling, kindly provided by Muriel Leromancer, CRCL, Lyon), PRMT5 (Millipore), PRMT6 (ab47244; Abcam); PR-Set7 (kindly provided by E. Julien, IGMM, Montpellier), H3 (Millipore), H3K79me1 (pAb-082; Diagenode), anti-H3K79me2 (ab3594, Abcam), anti-H3K79me3 (pAb-068, Diagenode), H4K20me2 (Millipore), H4R3me2s (ab5823; Abcam), and dimethyl-arginine symmetric (SYM10; Millipore); and a human autoimmune serum (huACA) against centromeric proteins (Antibodies Incorporated). For IF, the secondary antibodies used were goat anti-rabbit, anti-mouse, and anti-human antibodies coupled to Alexa Fluor 488, 555, or 647 (Molecular probes). For WB, the secondary antibodies used were rabbit anti-mouse and swine anti-rabbit antibodies coupled to horseradish peroxidase (Dako).
We thank Suzie Lefèbvre of Institut Jacques Monod, Paris for providing the EGFP-SMN-expressing plasmids; Gregory Matera of the University of North Carolina, Chapel Hill for providing the HeLaPV cells; and Jérome Cavaillé of LBME, Toulouse, Muriel Leromancer of CRCL, Lyon, and Eric Julien of IGMM, Montpellier for providing antibodies. The Zeiss LSM 510 confocal microscope images were captured at the Centre Technologique des Microstructures, Plateforme de l'Université Claude Bernard Lyon 1. P.L. is a CNRS researcher.
M.S., P.T. and P.L. conceived and designed experiments; M.S., P.T., J.E.M. and P.L. performed the experiments; M.S., P.T., J.E.M. and P.L. analyzed the data; M.S., P.T., J.E.M. and P.L. contributed reagents/materials/analysis tool; and P.L. wrote the paper.
This work was supported by the Association Française contre les Myopathies [Trampoline grant number 15033 to P.L.]; the FINOVI Foundation [grant number 004 to P.L.]; the Association pour la Recherche contre le Cancer [grant number ARC-7979 and ARC-4910 to P.L.]; the Ligue Nationale Contre le Cancer; the LabEX DEVweCAN of the Université de Lyon, within the program ‘Investissements d'Avenir’ operated by the French National Research Agency [grants numbers: ANR-10-LABX-61 and ANR-11-IDEX-0007]; the Centre National de la Recherche Scientifique; and a studentship from the Ligue National Contre le Cancer (Comité du Rhône) to M.S.