|Home | About | Journals | Submit | Contact Us | Français|
Saccharomyces cerevisiae Msl5 (branchpoint binding protein) orchestrates spliceosome assembly by binding the branchpoint sequence 5′-UACUAAC and establishing cross intron-bridging interactions with other components of the splicing machinery. Reciprocal tandem affinity purifications verify that Msl5 exists in vivo as a heterodimer with Mud2 and that the Msl5–Mud2 complex is associated with the U1 snRNP. By gauging the ability of mutants of Msl5 to complement msl5Δ, we find that the Mud2-binding (amino acids 35–54) and putative Prp40-binding (PPxY100) elements of the Msl5 N-terminal domain are inessential, as are the C-terminal proline-rich domain (amino acids 382–476) and two zinc-binding CxxCxxxxHxxxxC motifs (amino acids 273–286 and 299–312). A subset of conserved branchpoint RNA-binding amino acids in the central KH-QUA2 domain (amino acids 146–269) are essential pairwise (Ile198–Arg190; Leu256–Leu259) or in trios (Leu169–Arg172–Leu176), whereas other pairs of RNA-binding residues are dispensable. We used our collection of viable Msl5 mutants to interrogate synthetic genetic interactions, in cis between the inessential structural elements of the Msl5 polypeptide and in trans between Msl5 and yeast splicing factors (Mud2, Nam8 and Tgs1) that are optional for vegetative growth. The results suggest a network of important but functionally buffered protein–protein and protein–RNA interactions between the Mud2–Msl5 complex at the branchpoint and the U1 snRNP at the 5′ splice site.
Regulated pre-mRNA splicing figures prominently in the control of eukaryal gene expression and defects in pre-mRNA splicing underlie many human diseases (1,2). The yeast Saccharomyces cerevisiae has been a powerful model system to elucidate the composition and function of the core splicing machinery and the dynamic changes in the spliceosome that orchestrate splicing chemistry and fidelity (3–5). The yeast spliceosome transits through multiple steps—assembly, activation, catalysis, and disassembly—mediated by five snRNPs (U1, U2, U4, U5, U6) and scores of proteins that interact with the snRNPs and the pre-mRNA. The spliceosomal snRNAs are all essential for yeast viability. By contrast, many of the yeast splicing proteins are inessential for vegetative growth, especially factors implicated in early steps of spliceosome assembly. Classical genetics and synthetic genetic array analyses (6–12) have highlighted networks of mutational synergies between the inessential yeast splicing factors, whereby combining two viable null mutations results in synthetic lethality or ‘sickness’. Such genetic interactions among individually dispensable players in a common pathway (splicing) meets an operational definition of redundancy (i.e. ‘the provision of additional or duplicate systems, equipment, etc., that function in case an operating part or system fails’). Genetic redundancy does not necessitate that the synthetic interactor proteins perform the identical task, but rather that the process (e.g. spliceosome assembly) can be accomplished with the aid of either protein, conceivably via different microscopic sub-pathways.
This scenario is exemplified by the genetics of Tgs1, the enzyme responsible for synthesis of the signature 2,2,7-trimethylguanosine (TMG) cap structure of the U1, U2, U4 and U5 snRNAs (13,14). Tgs1 catalyzes two successive methyltransfer reactions from AdoMet to the N2 atom of 7-methylguanosine nucleotides via a distributive mechanism (10,14–16). The TMG cap is not essential for viability of eukaryal cells, insofar as a tgs1Δ mutant of fission yeast grows normally (17). The tgs1Δ mutation of budding yeast causes a growth defect at cold temperatures, though tgs1Δ cells grow as well as TGS1 cells at 34–37°C (10,13). The tgs1Δ mutants of budding and fission yeast lack any detectable TMG caps on their U1, U2, U4 and U5 snRNAs, as gauged by IP-Northerns using an anti-TMG antibody (13,17), signifying that there is no Tgs1-independent route to generate TMG caps. tgs1Δ yeast cells have apparently normal steady-state levels of snRNAs, and they display no overt aberrations in the RNA or protein contents of their spliceosomal snRNPs, except for the acquisition of the nuclear m7G cap-binding complex as a stoichiometric component of the U1 snRNP (13,18).
That fungi grow in the absence of Tgs1 suggested there might be backup mechanisms to ensure the function of the U snRNAs when the TMG modification is missing. This idea was confirmed by synthetic genetic array analysis in budding yeast, which revealed that the effects of ablating the TMG cap are buffered by spliceosome assembly factors that are themselves inessential for vegetative growth (12,13). The strongest genetic interactor with Tgs1 is Mud2. Whereas ablation of Mud2 per se has no apparent effect on yeast cell growth, the tgs1Δ mud2Δ combination resulted in unconditional synthetic lethality (13). Mud2 was initially identified by the Rosbash lab in a ‘MUD screen’ (mutant U1 die) for yeast mutations that cause synthetic lethality with otherwise viable mutations in the U1 snRNA (19). Mud2, a 527-amino acid polypeptide composed of a hydrophilic N-terminal domain and a C-terminal RRM3 domain, is the yeast homolog of human splicing factor U2AF65. Mud2 exists as a heterodimer with Msl5, the essential yeast homolog of mammalian branchpoint binding protein SF1 (20). Mud2 interacts with the pre-mRNA/U1 snRNP complex, in a manner that depends on the branchpoint sequence of the intron and the association of Mud2 with Msl5. Msl5 then facilitates recruitment of the U2 snRNP, binding of which displaces Mud2/Msl5 from the branchpoint (19–23). The synthetic lethality of tgs1Δ and mud2Δ mutations implies that spliceosome assembly during vegetative growth depends on either Mud2 or a TMG cap on one or more of the U snRNAs.
Mud2 interacts genetically with many other spliceosome assembly factors by the criteria of synthetic lethality and synthetic sickness. These include the U1 snRNP subunits Mud1 and Nam8, the nuclear cap-binding complex (implicated in cap-dependent recruitment of U1 snRNP to the 5′ splice site), the U2 snRNP components Lea1 and Msl1, and splicing factors Cwc21, Swm2 and Swt21 (6,7,9,11,12,16,24). We recently exploited the synthetic lethal mud2Δ interactions to delineate functional domains and structure-activity relations in yeast Mud2 (16). We found that the C-terminal RRM3 domain is necessary but not sufficient for Mud2 function in complementing growth of tgs1Δ mud2Δ and mud1Δ mud2Δ strains. Other changes in Mud2 elicited distinct phenotypes in tgs1Δ versus mud1Δ backgrounds. For example: (i) alanine-cluster mutation of the RNP2 and RNP1 motifs (424LLLLNC and 478AGNIYIKF) of the RRM3 domain was lethal in tgs1Δ, but not in mud1Δ; (ii) the segment from amino acids 112–172 was critical for Mud2 function in tgs1Δ, but not in mud1Δ. The requirements for Mud2 function are apparently more stringent when yeast cells lack TMG caps on U snRNAs than when they lack Mud1 in the U1 snRNP.
The physical association of the inessential Mud2 protein with the essential Msl5 (20) raises the question of whether mud2 phenotypes are mediated via Msl5. Msl5 is a 476-amino acid polypeptide composed of a central branchpoint RNA binding domain (amino acids 147–331, embracing KH, QUA2 and Zn knuckle modules; see Figure 1) flanked by an N-terminal domain that binds Mud2, and N-terminal and C-terminal proline-rich motifs that might contribute to the interaction of Msl5 with the Prp40 subunit of the U1 snRNP (21,22,25–27). Initial studies showed that a 2-hybrid clone expressing the Gal4 activation domain fused to Msl5-(10–362) was able to complement growth of an msl5Δ mutant, whereas a Gal4 fusion to Msl5-(10–183) could not (22). An N-terminal Msl5 deletion mutant lacking the first 56 amino acids was unable to bind stably to Mud2, but nonetheless complemented msl5Δ when the ‘bbpΔ56’ allele was integrated into the yeast chromosome (20). These results indicate that neither the binding of Mud2 to Msl5 nor the putative interactions of the Msl5 proline-rich C domain are essential per se for vegetative growth. Here we map at higher resolution the margins of the active Msl5 protein and we then show that the N and C domains of Msl5 are genetically redundant.
We examine the issue of whether Mud2 phenotypes are Msl5-mediated by surveying the effects of Msl5 mutations in a mud2Δ background. The rationale here is that mud2Δ synthetic phenotypes resulting from absence of Mud2 from the Msl5-Mud2 heterodimer ought to be phenocopied by Msl5 mutations that efface its Mud2 binding site. We also interrogate the effects of Msl5 mutations in the putative Prp40-binding motif and the KH, QUA2 and Zn knuckle modules on complementation of msl5Δ and msl5Δ tgs1Δ and msl5Δ nam8Δ yeast strains. Our results reveal a rich network of intramolecular and intermolecular genetic interactions of the yeast branchpoint binding protein.
Yeast MUD2-TAP and MSL5-TAP strains were purchased from Open Biosystems. Tandem-affinity purifications from whole cell extracts (derived from 6-l yeast cultures grown in YPD medium to A600 of 2.0–2.5) were performed as described (18). The polypeptides in the MUD2-TAP and MSL5-TAP preparations were resolved by SDS–PAGE; the lanes were sliced horizontally into 10 segments. In situ trypsin digestion of polypeptides in each gel slice, LC-MS/MS analysis and protein/peptide identifications from the LC-MS/MS data were performed as described in detail elsewhere (18). Total numbers of peptide spectra assigned to specific yeast polypeptides (obtained by merging information obtained from all the gel slices in the lane) are compiled in Table 1.
CEN LEU2 plasmids bearing wild-type or mutated MSL5 genes under the control of the native MSL5 promoter were constructed as follows. First, we made plasmid pRS415-MSL5(5′3′), a derivative of the yeast shuttle vector pRS415 (CEN LEU2) with a genomic DNA segment spanning the 509-bp immediately upstream of the MSL5 ORF cloned between the XhoI and BamHI sites and a genomic DNA segment comprising the 264-bp segment immediately downstream of the MSL5 ORF inserted between the NotI and SacI sites. The wild-type MSL5 ORF was PCR-amplified with primers that introduced a BamHI site preceding the start codon and a NotI site following the stop codon. N-terminal truncation mutants MSL5-(25–476), MSL5-(35–476), MSL5-(55–476), MSL5-(69–476), MSL5-(88–476), MSL5-(102–476) and MSL5-(140–476) were PCR-amplified with sense-strand primers that introduced a BamHI site and a methionine codon in place of the codons for Ala24, Leu34, Ile54, Val68, Pro87, Asp101 or Pro139. The C-terminal truncation mutants MSL5-(1–458), MSL5-(1–437), MSL5-(1–425), MSL5-(1–401), MSL5-(1–381), MSL5-(1–347), MSL5-(1–312), MSL5-(1–298), MSL5-(1–272), MSL5-(1–263) and MSL5-(1–254) were PCR-amplified with anti-sense strand primers that introduced a stop codon (and a flanking NotI site) in lieu of the codons for Gly459, Val438, Lys426, Ser402, Ser382, Ile348, Asn313, Cys299, Cys273, Gly264 or Gln255. Combined N- and C-terminal truncation mutants were PCR-amplified with the appropriate primers. Alanine mutations were introduced into MSL5 and MSL5-(55–476) by two-stage PCR overlap extension using mutagenic primers. The various PCR products were digested with BamHI and NotI and then inserted between the BamHI and NotI sites of pRS415-MSL5(5′3′). The inserts of the resulting pRS415-MSL5 plasmids were sequenced completely to exclude the presence of unwanted coding changes.
We developed plasmid shuffle assays to gauge mutational effects on Msl5 activity in msl5Δ, tgs1Δ msl5Δ, mud2Δ msl5Δ and nam8Δ msl5Δ strains that rely for viability on maintenance of MSL5 on a URA3 plasmid. To establish the complementation assay in the msl5Δ background, we transformed a heterozygous MSL5 msl5::kanMX diploid with p360-MSL5, a CEN URA3 plasmid in which expression of MSL5 is driven by a yeast TPI1 promoter. The diploid was sporulated, asci were dissected and haploid msl5Δ Ura+ progeny were recovered. The msl5Δ Ura+ cells were resistant to geneticin and unable to grow on medium containing FOA (5-fluoroorotic acid, a drug that selects against the URA3 plasmid).
To assay complementation in the double-mutant backgrounds, we generated a heterozygous TGS1 tgs1::natMX MSL5 msl5::kanMX diploid, a heterozygous MUD2 mud2::natMX MSL5 msl5::kanMX diploid, and a heterozygous NAM8 nam8::natMX MSL5 msl5::kanMX diploid by crossing the respective haploid deletion strains with the msl5Δ p360-MSL5 strain. The diploids were sporulated, asci were dissected, and haploid tgs1Δ msl5Δ Ura+, mud2Δ msl5Δ Ura+ and nam8Δ msl5Δ Ura+ progeny were recovered that were resistant to both geneticin and clonNAT and unable to grow on medium containing FOA.
For plasmid shuffle assays, the test strains were transformed with the pRS415-MSL5 plasmids. Individual Leu+ transformants were streaked on agar medium containing 1mg/ml FOA. Growth was scored after incubation for 7 days at 18, 25, 30 or 37°C. Lethal mutants were those that failed to form colonies at any temperature. Individual FOA-resistant colonies with viable MSL5 alleles were grown to mid-log phase in YPD broth and adjusted to A600 of 0.1. Aliquots (3µl) of serial 10-fold dilutions were spotted on YPD agar plates, which were then incubated at 18, 25, 30 and 37°C. (The tgs1Δ strains were tested for growth on YPD agar at 25, 30 and 37°C; tgs1Δ cells are inviable at 18°C.) Growth was assessed as follows: (+++) colony size was indistinguishable from strains bearing wild-type MSL5; (++) slightly reduced colony size; (+) only pinpoint macroscopic colonies were formed; (±) only microscopic colonies were formed; (−) no growth.
We used the tandem affinity purification (TAP) method to isolate Mud2- and Msl5-containing complexes from soluble whole-cell extracts of MUD2–TAP and MSL5–TAP yeast cells (18,28). The protein contents of the samples were analyzed in parallel by SDS–PAGE, then digested in situ with trypsin. The resulting peptides were analyzed by liquid chromatography-coupled tandem mass spectrometry and correlated to specific peptide fragments identified in the S. cerevisiae proteome. The total numbers of peptide spectra assigned to individual yeast proteins with >95% confidence were compiled and are presented in Table 1. (We omitted from this compilation any ribosomal proteins, heat-shock proteins, protein chaperones and metabolic enzymes.) Mud2 and Msl5 had the highest spectral counts in both TAP preparations; indeed the numbers of Mud2-derived spectra were similar in the Mud2-TAP and Msl5-TAP samples (407 and 467, respectively) as were the numbers of Msl5-derived spectra (289 and 229 in the Mud2-TAP and Msl5-TAP samples, respectively) (Table 1). These findings are in accord with the previously reported analyses of 35S-methionine-labeled Mud2-TAP and Msl5-TAP preparations (20) and provide further evidence that Mud2 and Msl5 comprise a stable complex in vivo. [Also in agreement with previous studies (20), we found that heat shock protein Ssb2 and ribosomal protein Rpl4 were the next most abundant sources of peptide spectra in both TAP samples: Mud2-TAP yielded 147 Ssb2 spectra and 69 Rpl4 spectra; Msl5-TAP yielded 132 Ssb2 spectra and 79 Rpl4 spectra.]
Both Mud2-TAP and Msl5-TAP were associated with all 10 of the U1-specific protein subunits of the U1 snRNP (Prp39, Prp40, Snu71, Snu56, Snp1, Mud1, Luc7, Prp42, Nam8 and Yhc1) and with several subunits of the Sm protein ring. By contrast, we detected no peptides in either TAP preparation derived from the U2 snRNP subunits Lea1, Rse1, Hsh155, Cus1 and Msl1 or from the U5 snRNP subunits Prp8 and Snu114. These findings suggest that affinity purified yeast Msl5-Mud2 heterodimer is specifically associated with the U1 snRNP, conceivably because the TAP purification recovered early intermediates in spliceosome assembly that contained U1 snRNP at the 5′ splice site and Msl5-Mud2 at the branchpoint.
Also compiled in Table 1 under the category of ‘Others’ are sundry non-ribosomal proteins implicated in RNA transactions or RNA binding that were detected in the Mud2-TAP and Msl5-TAP preparations. We include this information for the sake of transparency and for those who might be interested in these yeast proteins. Note that these ‘others’ were also present in TAP preparations of bulk Sm-containing U snRNPs, isolated U1 and U2 snRNPs, and C/D and H/ACA snoRNPs (18), often with spectral counts similar to those reported in Table 1, which might mean that they are associated generically with RNAs in yeast ribonucleoproteins.
The C-terminal RRM3 domains of human U2AF65 and yeast Mud2 are implicated in protein–protein interactions with branchpoint binding protein rather than, or in addition to, RNA binding (22,29,30). RRM3-like domains, renamed U2AF homologs motifs (UHMs), have been identified in many other proteins, where they are proposed to comprise a protein interaction module (31). We exploited the NMR structure of a binary complex of human U2AF65 RRM3 bound to a peptide ligand 13PSKKRKRSRWNQD25 derived from the N-terminal domain of SF1 (30) (Figure 2A) to interrogate the homologous Mud2–Msl5 interface in yeast. The RRM3 fold comprises a four-stranded β-sheet flanked on one side by two α-helices. The central β1 and β3 strands of the sheet (referred to as the RNP2 and RNP1 motifs, respectively) typically make atomic contacts with bound RNA in classical RRM–RNA complexes (32), but this surface of the β-sheet is occupied in the U2AF65-SF1 complex by a C-terminal α-helix (Figure 2B). SF1 binds along the opposite side of the RRM3 fold, principally by insertion of a conserved SF1 tryptophan side chain (Trp22 in human SF1; Trp35 in yeast Msl5) into a hydrophobic pocket of U2AF65 formed by the ‘back’ side of the β-sheet and the two α-helices (Figure 2A). The U2AF65 side chains Glu397, Glu400, Leu449, Phe454 and Val459 that form the SF1 binding site are conserved in Mud2 (as Glu443, Glu446, Leu498, Phe503 and Val508) and are numbered according to the Mud2 counterparts in Figure 2A. Here we tested the effects of two double-alanine mutations E443A-E446A and L498A-F503A, plus a quadruple-alanine change E443A-E446A-L498A-F503A on Mud2 activity in two synthetic lethal genetic backgrounds—tgs1Δ mud2Δ and mud1Δ mud2Δ—in which cell viability requires a functional Mud2 protein. The mutations were introduced into the fully active MUD2-(41–527) gene (16) carried on a CEN plasmid under the control of the native MUD2 promoter and the ‘wild-type’ and mutant alleles were tested for complementation by plasmid shuffle, after which the viable FOA-resistant strains were tested for growth on YPD agar at 18, 25, 30 and 37°C. The key findings were that neither of the double-alanine changes in the putative Msl5 interface affected Mud2 activity, as gauged by colony size compared to the ‘wild-type’ Mud2 control (Supplementary Table S1). Indeed, even a double charge inversion mutation—E443K-E446K—did not affect complementation (Supplementary Table S1). The quadruple-alanine mutant E443A-E446A-L498A-F503A retained complementation activity in both synthetic lethal strains, albeit with modest hypomorphic phenotypes, whereby E443A-E446A-L498A-F503A mud1Δ cells formed smaller colonies at 37°C (scored as ++ growth) and E443A-E446A-L498A-F503A tgs1Δ cells displayed ++ growth at 25, 30 and 37°C (Supplementary Table S1).
To further probe the interface, we tested the effect of an alanine mutation at Trp35 of yeast Msl5 on Msl5's function in vivo, as gauged by complementation of the otherwise inviable msl5Δ mutant strain. The W35A allele (and all other MSL5 mutants tested in this study) was installed on a CEN plasmid under the control of the native MSL5 promoter and transformed into a msl5Δ p(CEN URA3 MSL5) strain. Viable FOA-resistant W35A cells were recovered after plasmid shuffle and grew as well as wild-type MSL5 cells on YPD agar at 18, 25, 30 and 37°C, as gauged by colony size (Table 2). If Trp35 is essential for Msl5 interaction with Mud2, and Mud2 activity is contingent on this interaction, then we might expect the MSL5-W35A allele to mimic mud2Δ with respect to its synthetically lethality with tgs1Δ. However, this was not the case (Table 2).
Taken together, these results suggest that either the interaction of Mud2 with Msl5 is not significantly affected by the structure-guided mutations (e.g. because the interface of the yeast proteins in vivo is different or more complex than that revealed by the NMR structure of the human RRM3 domain and the SF1-derived peptide) or that the binding of Mud2 to Msl5 is not essential for Mud2 activity in the mud1Δ and tgs1Δ backgrounds.
We used a primary structure alignment of S. cerevisiae and Aspergillus fumigatus branchpoint binding proteins (Figure 1) to guide a deletional analysis of the Msl5 polypeptide. A series of N-terminal truncation alleles were tested for msl5Δ complementation. We found that deleting 24, 34 or 54 amino acids from the N-terminus had no apparent impact on yeast growth. [This agrees with the viability of the ‘bbp-Δ56’ mutant reported previously (20).] Truncations of 68 or 87 amino acids resulted in moderately slower growth (scored as ++ in Table 2). Extending the deletions to 101 and 139 amino acids (i.e. to the proximal border of the Msl5 KH domain) resulted in severe conditional lethality, whereby msl5-(102–476) and msl5-(140–476) cells grew slowly at 37°C (++) and were unable to form colonies at 18, 25 or 30°C (Table 2). Thus, the segment between amino acids 88 and 101 includes constituents that are either essential per se or are required for Msl5 activity when the proximal N-terminal domain is missing.
The Msl5-(55–476) mutant should be unable to form a heterodimeric complex with Mud2 (20). If Mud2 acts via the Msl5-Mud2 heterodimer, then we expect the MSL5-(55–476) allele to phenocopy mud2Δ with respect to its synthetically lethality with tgs1Δ. Indeed, this was the case, insofar as MSL5-(55–476) did not complement a msl5Δ tgs1Δ strain (Table 2). As one might expect, the even shorter mutants MSL5-(69–476) and MSL5-(88–476) that complemented msl5Δ also displayed synthetic lethality with tgs1Δ (Table 2). By contrast, the deletion alleles MSL5-(35–476) and MSL5-(25–476) were as active as wild-type MSL5 in the tgs1Δ background (Table 2). These findings, and those above, implicate the segment from amino acids 35–54 (but not Trp35 per se) as necessary for Msl5 interaction with Mud2.
We constructed a series of 11 C-terminal truncation alleles of MSL5 (depicted in Figure 1) and tested them for msl5Δ complementation. MSL5-(1–298) and all of the longer versions sustained normal growth at all temperatures tested (Table 2). This signifies that the C-terminal 178-amino acid segment—which includes the proline-rich domain, the distal zinc knuckle and the linker domain connecting them—is dispensable for Msl5 activity. Truncating the C-terminus to amino acid 272, which deleted the proximal zinc knuckle, resulted in slowed growth of the MSL5-(1–272) strain at all temperatures tested (Table 2). Further deletions into the QUA2 domain (placing the C-terminus at position 263 or 254) abolished Msl5 activity (Table 2).
The effects of incremental C-terminal deletions were significantly different in a tgs1Δ background, whereby truncations to or beyond amino acid 401 were unconditionally lethal (Table 2). Thus, the proline-rich C-terminal segment from amino acids 402–476 becomes essential in the absence of TMG capping of the U snRNAs. Contained within the proline-rich domain are multiple copies of the peptide motif PPG[M/L/F] that mediates protein–protein interactions with GYF-domain proteins such as yeast Smy2 (20,33,34). Lesser C-terminal deletions of Msl5 progressively regained function in the tgs1Δ background. For example: MSL5-(1–425) tgs1Δ was slow growing at 37°C and inviable at 25 or 30°C; MSL5-(1–437) tgs1Δ grew as well as MSL5 tgs1Δ at 37°C, was slow growing at 30°C (++), and was inviable at 25°C; and MSL5-(1–458) tgs1Δ displayed +++ growth at 30 and 37°C and ++ growth at 25°C (Table 2). The severity of the growth defects in tgs1Δ correlated with the number of C-terminal PPG[M/L/F] motifs, with six copies in full-length Msl5, four copies in Msl5-(1–458), two copies in Msl5-(1–437) and just one motif in the synthetic lethal Msl5-(1–401) variant (Figure 1). We considered the prospect that the synthetic lethal and sick phenotypes of C-terminal Msl5 deletions in a tgs1Δ background might reflect the loss of a putative Msl5-Smy2 interaction, in which case we would expect a synthetic genetic interaction between Smy2 and Tgs1. Accordingly, we tested the effects of ablating SMY2 (an inessential yeast gene) in a tgs1Δ background, but observed no synthetic growth phenotype for smy2Δ tgs1Δ cells (data not shown). We surmise that the need for the Msl5 C-terminus in the absence of TMG caps is not exerted via Smy2.
Direct binding of the U1 snRNP protein Prp40 to Msl5 is thought to bridge the 5′ splice site and branchpoint segments of the intron and stabilize the commitment complex, an intermediate in spliceosome assembly (19,34). Tandem WW domains in yeast Prp40 bind to the isolated 94PSPPPVYDA102 nonapeptide of yeast Msl5, which contains a PPxY100 motif recognized by WW domain proteins (27). To probe the role of the Prp40-binding site, we introduced a triple-alanine mutation P97A-P98A-Y100A and tested its effect on Msl5 activity in vivo. The finding that MSL5-(P97A-P98A-Y100A) cells grew as well as MSL5 cells at all temperatures (Table 3) implies that a Prp40-Msl5 interaction via the PPxY100 motif is not essential per se in budding yeast under standard laboratory growth conditions, perhaps because there are redundant systems in place to provide the bridging interactions between U1 snRNP and the branchpoint. The inessentiality of the PPxY100 motif was also noteworthy in light of the fact that an incremental deletion of amino acids 88–101 containing the motif was lethal at 18–30°C (Table 2), which raised the prospect that the loss of the Prp40-binding motif might be buffered by contributions of the upstream Msl5 N-terminal domain. To evaluate this scenario, we introduced the P97A-P98A-Y100A mutation into the otherwise fully active truncated allele MSL5-(55–476) that lacks the Mud2-interaction module. Testing msl5Δ complementation by plasmid shuffle yielded microscopic colonies of MSL5-(55–476)-(P97A-P98A-Y100A) cells that, after amplification on rich medium, grew very poorly on YPD agar at all temperatures (forming only microscopic colonies after ≥7 days; scored as ± growth in Table 4). A similar synthetic genetic interaction was observed when we tested the msl5Δ complementation activity of full-length MSL5-(P97A-P98A-Y100A) in a mud2Δ background (Table 5). MSL5-(P97A-P98A-Y100A) mud2Δ cells were viable but slow growing at 37°C (++), formed only microscopic colonies after prolonged incubations at 25 and 30°C (±growth), and failed to grow at 18°C (Table 5). These results suggest functional redundancy in the bridging interactions of Msl5 with the U1 snRNP: either via Prp40 binding to the Msl5 PPxY100 motif or via interactions of the Msl5(Nterm)-Mud2 heterodimer with one or more U1 snRNP components (34).
Recognition of the intron branchpoint by branchpoint binding protein is accomplished by an autonomous RNA-binding module composed of a proximal KH domain fused to a distal QUA2 (Quaking homology 2) domain. The NMR structure of human SF1 bound to an RNA (5′-AUACUAACAA) containing the consensus yeast branchpoint sequence 5′-UACUAAC (35) reveals an extensive network of atomic contacts between conserved amino acid side chains (many of them hydrophobic) and the RNA nucleobases and sugars, while the RNA phosphates are surface-exposed and make relatively few protein contacts (Figure 3). Here we used the SF1–RNA structure to guide a mutational analysis of the KH and QUA2 modules of yeast Msl5.
The KH domain (colored green in Figure 3) engages the CUAACAA-3′ segment of the RNA ligand (the branchpoint adenosine is colored yellow in Figure 3). Here we introduced double-alanine or triple-alanine mutations in clusters of conserved KH residues that line the RNA binding groove, these being: N163A-V165A; L169A-R172A-L176A; I189A-R190A and V195A-K196A. The mutant alleles were tested for msl5Δ complementation (Table 3). Given that these side chains are identical in SF1 and Msl5 (Figure 3), and that the RNA–protein contacts observed for these side chains in the SF1 structure are to a canonical yeast branchpoint RNA recognition element, it is our presumption that the same RNA contacts are very likely to be made by the equivalent Msl5 side chains, allowing us to interpret mutational outcomes accordingly.
The results provide new insights to which RNA contacts are most important. MSL5-(N163A-V165A) and MSL5-(V195A-K196A) cells grew as well as MSL5 cells at all temperatures. Thus, the contacts of Asn163 with the cytosine base, of Val165 and Lys196 with the adenine base preceding the branchpoint adenosine, and of Val195 with the branchpoint adenine and the 3′-terminal adenine of the RNA ligand are all inessential for Msl5 function. By contrast, the L169A-R172A-L176A and I189A-R190A mutations were lethal (Table 3). Leu169 is imputed to intercalate between and make van der Waals contacts to the branchpoint adenine and the upstream flanking adenine (Figure 3). Arg172 contacts the 3′-phosphate of the branchpoint adenosine. Leu176 contacts both the adenine N3 atom and the ribose 2′OH of the branchpoint adenosine. Ile189 contacts the N1 atom of the branchpoint adenine. Arg190 contacts the 3′-terminal adenine base (Figure 3).
The QUA2 domain (colored magenta in Figure 3) binds to the proximal 5′-AUACU segment of the RNA. We tested two double-alanine mutations in the Msl5 QUA2 module, L256A-L259A and T265A-R267A, targeted to side chains that are identical in SF1 and comprise the SF1 QUA2-RNA interface (Figure 3). MSL5-(T265A-R267A) cells grew as well as MSL5 cells at 25–37°C, but were slower growing (++) at 18°C (Table 3). We surmise that the RNA interactions of Thr265 (a van der Waals contact with the ApC phosphate) and Arg267 (with the ribose 2′OH and 3′ phosphate of the 5′-terminal adenosine and with the N6, C5 and N7 atoms of the downstream adenine) are not essential for Msl5 function in vivo. By contrast, the MSL5-(L256A–L259A) allele was unconditionally lethal (Table 3). Leu256 makes multiple close van der Waals contacts (2.9–3.1Å) to the ApC dinucleotide, specifically to the adenosine ribose and the cytidine base and ribose (Figure 3). Leu259 makes close van der Waals contacts (3.0Å) to the uracil base and ribose of the flanking uridine nucleoside (Figure 3).
We also targeted mutations to the 252KR253 dipeptide in the QUA2 domain of yeast Msl5, which is the counterpart of the 240RK241 dipeptide in human SF1 (Figure 3). Domain swaps and local sequence swaps have implicated the sequence of this simple dipeptide as the determining factor in the specificity of yeast Msl5 for the conserved yeast branchpoint motif 5′-UACUAAC (26). The structure of the complex of human SF1 with the yeast branchpoint sequence does not illuminate how this might be, insofar as Arg240 is located 7.2Å away from the closest atom in the RNA. However, Lys241 in SF1 does make multiple van der Waals contacts with the C2, N1 and N6 atoms of the first adenine base in the UACUAAC motif (35) and this adenosine is a key determinant of the binding affinity of yeast Msl5 for the branchpoint (26). Thus, one might expect that subtracting the equivalent basic residue in yeast Msl5 (Arg253), in tandem with the vicinal Lys252 residue, would affect Msl5 activity in vivo. To the contrary, the MSL5-(K252A–R253A) allele was as active as wild-type MSL5 in complementation of msl5Δ at 25–37°C (+++ growth) and sustained ++ growth at 18°C (Table 3). We also performed a yeast-to-human dipeptide swap [the reciprocal of the human-to-yeast swap (26)], along with single amino acid swaps, and found that MSL5-(K252R–R253K), MSL5-(K252R) and MSL5-(R253K) cells grew as well as wild-type MSL5 cells at all temperatures (Table 3). Thus, our findings do not support an important role for the 252KR253 dipeptide for Msl5 function in an otherwise wild-type background.
The dispensability of the N-terminal 54-amino acid and C-terminal 213-amino acid domains raised the issue of whether they might make genetically redundant contributions to Msl5 activity in vivo. To address this question, we deleted the N-terminal 54 amino acids from our series of otherwise viable C-terminal truncation alleles and tested them for msl5Δ complementation (Table 4). MSL5-(55–458) and MSL5-(55–437) cells grew as well as MSL5 at all temperatures, while MSL5-(55–425) cells displayed +++ growth at 18–30°C and ++ at 37°C. Thus, we surmise that the distal segment of the proline-rich domain that includes five of the PPG[M/L/F] motifs is not functionally redundant with the N-terminal Mud2-interaction domain. However, further truncation of the C-terminus to amino acid 401 results in + growth at 37°C and ++ growth at 18 and 30°C (Table 4). The growth defect was exacerbated by deleting amino acids 313–347; such that MSL5-(55–312) cells did not grow at 37°C and displayed + growth at 18 and 30°C and ++ growth at 25°C (Table 4). Finally, the msl5-(55–298) allele was unconditionally lethal (Table 4). These results imply functional overlap between the N-terminal module and an internal region that includes the distal zinc knuckle. At this point, the basis for the functional overlap is undefined; indeed, because we have not directly measured the intracellular localization and steady state levels of the biologically inactive Msl5-(55–298) protein in vivo (such measurements would have to be performed in a yeast strain coexpressing a functional Msl5 protein to sustain cell growth), we cannot rule out that the functional overlap of the N- and C-terminal domains is at the level of Msl5 folding, stability or localization, versus an overlap in directing spliceosome assembly.
If the mutational synergies between the N-terminal domain deletion and the C-terminal deletion series reflects the loss of Mud2 binding to the N-terminus, then we might expect a similar pattern of mutational synergies between the absence of Mud2 and the C-terminal deletion series. To interrogate this scenario, we tested the C-terminal deletions for complementation in an msl5Δ mud2Δ background (Table 5). As an initial control, we verified that the MSL5-(55–476) allele was as active as wild-type MSL5 in msl5Δ mud2Δ complementation (Table 5), i.e. that loss of the Mud2-binding domain of Msl5 does not synergize with mud2Δ (which itself elicits no obvious growth phenotype). We found that MSL5-(1–425) mud2Δ cells displayed +++ growth, signifying a lack of functional overlap between Mud2 and the segment of the proline-rich domain that includes five of the PPG[M/L/F] motifs. The effects of further Msl5 truncations in the mud2Δ background echoed, but did not exactly replicate, the hierarchy of effects seen in the context of Msl5-(55–476) (compare Tables 4 and and5).5). For example, the C-terminal truncations to amino acids 401, 381 and 347 elicited growth defects in mud2Δ cells that were more severe at 18 and 25°C than at 30 and 37°C (Table 5). The MSL5-(1–312) allele, which retains the distal zinc knuckle and is active in msl5Δ complementation, was synthetically lethal with mud2Δ (Table 5).
To focus on the potential roles and genetic interactions of the zinc knuckles, we introduced alanine mutations pairwise in lieu of the putative zinc-binding cysteine and histidine residues (Figure 1). In the context of full-length Msl5, the C273A–C276A and H281A–C286A mutations that disrupt the proximal knuckle and the C299A–C302A and H307A–C312A mutations that disrupt the distal knuckle had no impact on msl5Δ complementation (Table 3), in keeping with our earlier results that deletion of the distal zinc knuckle in MSL5-(1–298) did not affect msl5Δ complementation and that MSL5-(1–272) cells lacking both zinc knuckles were viable albeit slow-glowing.
We then proceeded to introduce the zinc knuckle alanine mutations into MSL5-(55–476) and test them for msl5Δ complementation (Table 4). In this context, the knuckle mutations C273A–C276A and C299A–C302A supported +++ growth at all temperatures, while H281A–C286A sustained +++ growth at 18–30°C and ++ growth at 37°C. However, the most distal mutation H307A–C312A in the second knuckle displayed severe mutational synergy with the N-terminal deletion, whereby MSL5-(55–476)-(H307A–C312A) cells displayed + growth at all temperatures (Table 4). Because the C299A–C302A and H307A–C312A changes are expected to exert synonymous effects on zinc-binding by the second knuckle, we suspect that the deleterious effects of MSL5-(55–476)-(H307A-C312A) are caused by a zinc-independent structural perturbation of this region of the protein.
We tested our collection of Msl5 alanine-mutants for their ability to complement growth in an msl5Δ tgs1Δ background. The MSL5-(P97A-P98A-Y100A) allele, which was fully active in msl5Δ complementation, was extremely feeble in complementing msl5Δ tgs1Δ, such that MSL5-(P97A-P98A-Y100A) tgs1Δ cells formed only microscopic colonies after prolonged incubations at 25, 30 and 37°C (±growth; Table 3). Thus, mutation of the Prp40-binding PPxY100 motif of Msl5 elicited strong synthetic phenotypes in three different backgrounds: (i) tgs1Δ; (ii) mud2Δ and (iii) MSL5-(55–476), wherein the Mud2–Msl5 interaction is ablated. These results fortify the already substantial genetic connections between the Msl5(N-term)-Mud2 heterodimer, the protein components of the U1 snRNP (i.e. Prp40, Nam8 and Mud1) and the snRNA TMG cap structure.
A new and instructive finding was that four double-alanine mutations in the KH-QUA2 domain that had no apparent effect on msl5Δ complementation were synthetically lethal in the tgs1Δ background: these were N163A-V165A (lethal at 25, 30 and 37°C), V195A-K196A (lethal at 25 and 30°C; ++ growth at 37°C), T265A-R267A (lethal at 25 and 30°C; + growth at 37°C) and K252A-R253A (lethal at 25, 30 and 37°C) (Table 3). These results indicate that the essentiality or dispensability of many constituents of the imputed interface of Msl5 with the branchpoint RNA is context-dependent and can be influenced by the presence or absence of other components of the splicing machinery (in this case, the TMG cap). The simplest interpretation is that proper spliceosome assembly in vivo relies on either the contacts of Asn163, Val165, Val195, Lys196, Thr265 and Arg267 with the conserved yeast branchpoint RNA sequence or the presence of the TMG cap.
Further insights to the mutational synergy of MSL5-(K252A-R253A) and tgs1Δ came from the finding that the swap-allele MSL5-(K252R-R253K) was lethal in tgs1Δ cells at 25 and 30°C and sufficient for only + growth at 37°C (Table 3). Moreover, whereas the single-mutant MSL5-R253K was fully functional in the tgs1Δ background, the MSL5-K252R tgs1Δ strain was inviable at 25 and 30°C and displayed feeble growth (+) at 37°C (Table 3). We surmise that the mutational synergy is caused by the replacement of Lys252 with either alanine or arginine. This is notable given that Lys252 in yeast Msl5 is predicted not to interact directly with the branchpoint RNA (35).
Finally, we found that the double-alanine mutations of the zinc-binding residues of the knuckle modules displayed little or no mutational synergy with tgs1Δ (Table 3). In sum, these studies reveal new genetic interactions of the essential Msl5 protein with the TMG cap that could not have been appreciated by a synthetic genetic array of tgs1Δ against the yeast knockout collection (10).
Bridging interactions between constituents of the U1 snRNP engaged at the 5′ splice site and Msl5 bound to the branchpoint RNA sequence are thought to be critical for spliceosome assembly. Emerging evidence, here and elsewhere (36), suggests that the cross-intron bridging contacts are multiple and functionally buffered, such that elimination of a single pairwise interaction (e.g. between branchpoint binding protein and U1 snRNP, U1 and the 5′ splice site, or branchpoint binding protein and the branchpoint RNA) might not elicit an overt phenotype unless other cross-intron contacts are also impaired. To explore this idea, we surveyed several of our Msl5 mutants for their ability to complement growth in an msl5Δ nam8Δ background (Table 6). Nam8 is a stoichiometric protein component of the yeast U1 snRNP that is inessential for vegetative yeast growth. The 523-amino acid Nam8 protein contains three tandem RRM domains (RRMs 1, 2 and 3) flanked by N-terminal leader and C-terminal tail segments. Whereas the leader, tail and RRM1 are dispensable for vegetative Nam8 function in multiple synthetic lethal genetic backgrounds (including tgs1Δ), Nam8 activity in synthetic lethal situations is compromised by mutations in the putative RNA binding sites of the RRM2 and RRM3 domains (24). We thought that absence of Nam8 might synergize with a subset of our viable Msl5 mutants, especially ones with mutations in the RNA-binding KH-QUA2 domain and the Prp40-binding motif PPxY100.
Indeed, we found that nam8Δ was unconditionally synthetic lethal with MSL5-(P97A-P98A-Y100A) (Table 6), suggesting that the otherwise dispensable contacts of the PPxY100 motif with U1 component Prp40 are rendered essential for spliceosome assembly in the absence of U1 component Nam8. Also, nam8Δ was synthetically lethal with several otherwise benign mutations of the RNA binding site in the KH-QUA2 module, these being N163A-V165A and K252A-R253A (unconditionally lethal), V195A-K196A (lethal at 18 and 25°C; + and +++ growth at 30 and 37°C, respectively), and T265A-R267A (lethal at 18, 25 and 30°C; + growth at 37°C) (Table 6). These results show how constituents of the protein–protein and protein–RNA interfaces of Msl5 are either essential or not, depending on the composition of the U1 snRNP. It is notable that the synergies of KH-QUA2 and PPxY mutations in the absence of Nam8 track closely with those seen in the absence of TMG caps (compare Tables 3 and and6),6), hinting that an altered state of the U1 snRNP in tgs1Δ cells (18) might account for their reliance on otherwise optional features of Msl5.
Ablation of Nam8 also caused a severe synthetic growth defect with a C-terminal truncation of Msl5 to amino acid 401, resulting in lethality at 18, 25 and 30°C and + growth at 37°C (Table 6). A partial gain-of-function in the nam8Δ background was seen when the C-terminus was extended to amino acid 425, with +++ growth at 37°C, but only + growth at ≤30°C (Table 6), while MSL5-(1–437) nam8Δ cells displayed +++ growth at all temperatures tested. Finally, it was noteworthy that MSL5-(55–476) nam8Δ cells were viable and grew normally at 18, 25 and 30°C, though they failed to grow at 37°C (Table 6). By contrast, the mud2Δ nam8Δ combination is unconditionally lethal (16). This is an instance in which elimination of the Mud2-interaction domain of Msl5 did not phenocopy the ablation of Mud2. Because the former maneuver is less severe in nam8Δ than the latter, we suggest that Mud2 may have a context-dependent role in splicing that does not require its heterodimerization with the Msl5 N-terminal module.
The present study enhances our understanding of the domain organization and essential structural features of Msl5, a critical organizer of spliceosome assembly that recognizes the intron branchpoint and makes manifold interactions with other components of the splicing machinery, especially Mud2 and the U1 snRNP. By characterizing the ability of deletion mutants and missense mutants to complement msl5Δ growth, we showed that: (i) ablation of the Mud2-binding (amino acids 35–54) and Prp40-binding (PPxY100) elements of the Msl5 N-terminal domain does not affect yeast growth under laboratory conditions; (ii) the C-terminal proline-rich domain (amino acids 382–476), the distal zinc knuckle (amino acids 299–312) and the intervening linker are also dispensable for Msl5 function in an otherwise wild-type background; (iii) a subset of conserved branchpoint RNA-binding amino acids in the KH-QUA2 domain are essential pairwise (Ile198–Arg190; Leu256–Leu259) or in trios (Leu169–Arg172–Leu176) for Msl5 activity, whereas others are dispensable (Asn163–Val165; Val195–Lys196; Thr265–Arg267) and (iv) the QUA2 domain 252KR253 dipeptide, purported to act as a governor of yeast versus human branchpoint sequence recognition (based on effects of QUA2 domain and dipeptide swaps on branchpoint RNA binding in vitro; 26), is also dispensable. Indeed, the structural and in vivo mutational data point away from the idea that this dipeptide is decisive for yeast branchpoint recognition in vivo. We find that neither of the two zinc-binding CxxCxxxxHxxxxC motifs (amino acids 273–286 and 299–312) flanking the KH-QUA2 domain is important per se for msl5Δ complementation. Whereas, the distal knuckle can be deleted without overt consequence for growth, extending the deletion to include the proximal knuckle is lethal. These results are consistent with previous inferences from RNA binding studies in vitro (25) that the proximal knuckle contributes to intron RNA binding, independent of its ability to coordinate zinc via the knuckle cysteines.
Because Msl5 is required for growth of S. cerevisiae, and relatively few mutants of Msl5 had been studied previously in vivo (20,37), it has escaped analysis by synthetic array approaches that have proven so fruitful in illuminating the genetic connections among the many inessential yeast splicing factors (8–10,12). Here we took advantage of our collection of viable Msl5 mutants to interrogate synthetic genetic interactions, between the inessential structural elements of the Msl5 polypeptide and between Msl5 and several yeast splicing factors that are optional for vegetative growth. Both approaches highlighted multiple redundancies of the dispensable components of the Msl5 protein, as summarized in Figure 4 and discussed briefly below.
Redundancies within the N-terminal segment of Msl5 are evinced by the lethality of simultaneously ablating the Mud2-binding module and the Prp40-binding PPxY100 motif (see the ‘intra-Msl5’ mutational synergies denoted by red lines in Figure 4). Moreover, mutating the PPxY100 motif is synthetically lethal with mud2Δ. (Mud2 mutational synergies are depicted as black lines in Figure 4). The simplest inference here is that parallel functionally buffered contacts between the Mud2–Msl5 complex at the branchpoint and the U1 snRNP at the 5′ splice site aid in spliceosome assembly in vivo, entailing Msl5 interaction with Prp40 while Mud2 interacts with another U1 protein, e.g. the essential U1 component Snu56 (38). Our finding that mutation of the PPxY100 motif is lethal in a nam8Δ background suggests that the Msl5–Prp40 interaction is itself buffered by Nam8, an otherwise inessential U1 snRNP subunit. (Nam8 synthetic interactions are shown as gold lines in Figure 4.) Otherwise benign mutations of the Mud2-binding and Prp40-binding sites of Msl5 are lethal in a tgs1Δ background in which U1 snRNA lacks the TMG cap structure. (Tgs1 synthetic interactions are denoted by green lines in Figure 4.) These results, together with prior findings of synthetic lethality between tgs1Δ, nam8Δ, mud2Δ and mud1Δ (9,10,16,24), underscore the plentiful overlapping genetic connections between the Msl5-Mud2 heterodimer and the U1 snRNP.
The redundancies of the C-terminal segment of Msl5 that we observe suggest the presence of two distinct functional units. Deletion of the proline-rich domain from amino acids 402–437 elicits synthetic lethality in tgs1Δ and nam8Δ backgrounds (Figure 4), but does not synergize comparably with mud2Δ or with intra-Msl5 deletion of the N-terminal Mud2-binding site. It is conceivable that this segment of the proline-rich domain mediates additional interactions with the U1 snRNP that are buffered by Nam8 and TMG caps. Extending the Msl5 C-terminal deletions inward to amino acid 312 and amino acid 298 imparts synthetic lethality with mud2Δ and the Msl5-NΔ54 deletion, respectively (Figure 4). To our knowledge, no specific physical interactions (protein–protein or protein–RNA) have been ascribed to amino acids 402–437 and amino acids 299–401 of Msl5 that could easily explain the genetic synergies. Absent any structural information for the Msl5 C-terminal domain, there is little impetus for speculation.
By contrast, there is an NMR structure available of the KH-QUA2 module of human SF1 bound to the consensus yeast intron branchpoint RNA element (35) that provided a blueprint for the mutagenesis of RNA-binding residues of yeast Msl5. It was the collection of viable alanine-cluster mutants in the KH and QUA2 domains that yielded novel insights to genetic redundancies between Msl5's RNA–protein contacts and its interaction with the U1 snRNP, as inferred from the many synthetic sick and synthetic phenotypes of otherwise benign KH-QUA2 mutations in the nam8Δ (and tgs1Δ) backgrounds (Figure 4). It is worth noting that the original identification of Msl5 by Abovich and Rosbash (21), entailing a genetic screen for synthetic lethality with an N-terminal deletion mutant of Mud2, yielded an allele with a single Gly230Ser mutation in the third α-helix of the KH domain. The equivalent residue in the human SF1 structure (Ala218) is buried in the hydrophobic core of the KH fold, quite far from the branchpoint RNA. Rutz and Séraphin (37) reported a cold-sensitive msl5 phenotype associated with coding changes V195D in the KH domain and E258V in the QUA2 domain. As noted earlier, the conserved Val195 side chain interacts directly with the branchpoint RNA, but is clearly inessential for function, as gauged by the fact that alanine substitution has no effect on msl5Δ complementation, even at cold temperature (Table 3). Thus, we surmise that the contribution of the V195D change to the cs defect reflects the deleterious impact of ectopic negative charge, rather than the loss of the van der Waals contacts of the valine to the vicinal RNA nucleobases. The Glu258 side chain is located on the face of the QUA2 α-helix opposite that which binds the branchpoint RNA. The upshot is that structurally guided missense mutations of amino acids that directly bind the branchpoint are well-suited to dissect the genetic redundancies of the Msl5-branchpoint interface, whether by screening candidate mutants for synergies (as done here) or by genome-wide approaches (12).
Supplementary Data are available at NAR Online: Supplementary Table 1.
NIH grants (GM52470, to S.S.) and (GM50288, to B.S.). Funding for open access charge: National Institutes of Health grant (GM52470, to S.S.).
Conflict of interest statement. None declared.
We are grateful to Dr Hediye Erjument-Bromage and the MSKCC microchemistry core laboratory for proteomics analysis. S.S. is an American Cancer Society Research Professor.