|Home | About | Journals | Submit | Contact Us | Français|
The multiprotein complex Set1/COMPASS is the founding member of the histone H3 lysine 4 (H3K4) methyltransferases, whose human homologs include the MLL and hSet1 complexes. COMPASS can mono-, di-, and trimethylate H3K4, but transitioning to di- and trimethylation requires prior H2B monoubiquitination followed by recruitment of the Cps35 (Swd2) subunit of COMPASS. Another subunit, Cps40 (Spp1), interacts directly with Set1 and is only required for transitioning to trimethylation. To investigate how the Set1 and COMPASS subunits establish the methylation states of H3K4, we generated a homology model of the catalytic domain of Saccharomyces cerevisiae yeast Set1 and identified several key residues within the Set1 catalytic pocket that are capable of regulating COMPASS's activity. We show that Tyr1052, a putative Phe/Tyr switch of Set1, plays an essential role in the regulation of H3K4 trimethylation by COMPASS and that the mutation to phenylalanine (Y1052F) suppresses the loss of Cps40 in H3K4 trimethylation levels, suggesting that Tyr1052 functions together with Cps40. However, the loss of H2B monoubiquitination is not suppressed by this mutation, while Cps40 is stably assembled in COMPASS on chromatin, demonstrating that Tyr1052- and Cps40-mediated H3K4 trimethylation takes place following and independently of H2B monoubiquitination. Our studies provide a molecular basis for the way in which H3K4 trimethylation is regulated by Tyr1052 and the Cps40 subunit of COMPASS.
The human MLL gene located on chromosome 11q23 undergoes frequent translocations resulting in the pathogenesis of infant hematological malignancies, including both acute lymphoid and acute myeloid leukemia (11, 22, 30). Although the first MLL translocation was identified more than 20 years ago, little was known about the molecular function(s) of MLL until its Saccharomyces cerevisiae yeast homolog, Set1, was identified in a macromolecular complex named COMPASS (complex of proteins associated with Set1) (16). COMPASS is a histone H3 lysine 4 (H3K4) methyltransferase capable of mono-, di-, and trimethylating its histone substrate (12, 16, 18, 21, 25). We now know that human MLL is found in a COMPASS-like complex capable of methylating H3K4 (10, 25, 26).
Histone H3K4 trimethylation is highly correlated with active transcription. This modification requires the presence of histone H2B monoubiquitination, a process known as histone cross talk (7, 29). In addition to H2B monoubiquitination, other factors, several of which are found within COMPASS, are also required for proper H3K4 trimethylation. This includes the Cps25 (Sdc1), Cps35 (Swd2), Cps40 (Spp1), and Cps60 (Bre2) subunits of COMPASS (14, 17, 23, 24). Cps40 (Spp1) and Cps60 (Bre2) were shown to be required for proper H3K4 trimethylation by COMPASS (24). The human homologs of these two proteins are also required for proper H3K4 trimethylation within the human COMPASS-like complexes (6, 13, 28). Based on these observations, it was suggested that Cps40 (Spp1)/Cps60 (Bre2) may directly interact with Set1 and that via this interaction it can alter Set1's catalytic pocket in a way that allows the trimethylation of H3K4 (28). To address this hypothesis, we have generated a homology model for the catalytic domain of Set1. From this model, we have identified several key residues within Set1's catalytic pocket that are capable of regulating the trimethylation activity of COMPASS. One such residue, Tyr1052 of Set1, appears to play an essential role in the regulation of the available space in the catalytic pocket of Set1 and thereby regulates the transition from mono- to di- to trimethylation. Tyr1052 is of particular interest because it corresponds to a site known as the Phe/Tyr switch that has been shown to intrinsically govern the product specificities of many histone methyltransferases. According to the Phe/Tyr switch model, SET domain monomethyltransferases, such as SET7/9 and SET8, harbor a Tyr in this position, whereas di- and trimethyltransferases, such as G9A and DIM-5, possess a phenylalanine or other hydrophobic residue in this site (3, 4, 33). Although this model is generally applicable to the SET domain family, Set1 and its homologs are exceptions because they possess a tyrosine in the Phe/Tyr switch but are able to catalyze mono-, di-, and trimethylation of H3K4. Indeed, we have demonstrated that a single point mutation of this residue (Y1052F) which removes the hydroxyl group within the catalytic pocket of Set1 can suppress the loss of Cps40 (Spp1) in the H3K4 trimethylation pattern.
Since it has been proposed that Cps40 (Spp1) links H2B monoubiquitination to H3K4 trimethylation (31), we tested whether the loss of H2B monoubiquitination is suppressed by the Y1052F mutation. Our results indicate that the regulation of H3K4 trimethylation by Cps40 (Spp1) is subsequent to and independent of H2B monoubiquitination. We have also demonstrated that the association of Cps40 (Spp1) with COMPASS and chromatin does not require H2B monoubiquitination or association with Cps35 (Swd2), as previously proposed (31). Overall, our study implies the existence of a Phe/Tyr switch by the catalytic domain of Set1, providing a molecular basis for how H3K4 trimethylation is regulated by the Cps40 (Spp1) subunits of COMPASS and demonstrating the diverse regulatory mechanisms for setting H3K4 methylation states by COMPASS.
The homology model of the catalytic domain of Saccharomyces cerevisiae Set1 was generated by using Swiss-PdbViewer (9) and, with the crystal structure of the human MLL1 catalytic domain (Protein Data Bank structure accession no. 2w5z) as a template. The model was subsequently submitted to the Swiss-Model server (http://swissmodel.expasy.org/) for refinement. A histone H3 substrate peptide (residues 2 to 8); the product, S-adenosyl-l-homocysteine (AdoHcy); and the Zn(II) ion in the PostSET motif were modeled into the Set1 model through a structural alignment with the coordinates of the MLL1/AdoHcy/histone H3 peptide complex (Protein Data Bank structure accession no. 2w5z) (27). The Set1 ternary complex model was then energy minimized by using Crystallography and NMR System, version 1.21 (1), and manual adjustments to the model were performed by using Coot (8). The final model encompasses residues 924 to 1080 of Set1, the 7-residue H3 peptide, AdoHcy, and the PostSET Zn(II) ion. The stereochemistry of the Set1 model was validated by using PROCHECK, which verified that nonglycine residues are absent from the disallowed regions of the Ramachandran plot. Structural figures were rendered in PyMOL (http://pymol.sourceforge.net/).
Plasmids carrying wild-type or mutant Set1 genes were constructed by cloning each Set1 gene into pRS315. Set1Δ Cps40Δ (cps40::kanMX6), Set1Δ Cps60Δ (cps60::kanMX6), and Set1Δ Rad6Δ (rad6::kanMX6) strains were created by PCR-mediated gene disruption of each gene.
Purification of a tandem affinity purification (TAP)-tagged protein was performed essentially as previously described (16). Six liters of yeast were grown to an optical density at 600 nm (OD600) of 2.0 to 3.0. After the cells were harvested with centrifugation, they were lysed with glass beads in ~40 ml of a lysis buffer (10 mM Tris [pH 8.0], 150 mM NaCl, 0.1% NP-40, protease inhibitor cocktail). After clarification of the lysate with centrifugation, TAP-tagged proteins were bound to immunoglobulin G-Sepharose beads overnight at 4°C. Following cleavage of the TAP tag with tobacco etch virus protease, the calcium concentration of the eluate was adjusted to 2 mM CaCl2, and the eluate was incubated with calmodulin-Sepharose 4B. Following several washes with a calmodulin binding buffer (10 mM Tris [pH 8.0], 150 mM NaCl, 0.1% NP-40, 5 mM β-mercaptoethanol, 2 mM CaCl2), the bound proteins were eluted by the addition of an EGTA elution buffer (10 mM Tris [pH 8.0], 150 mM NaCl, 0.1% NP-40, 5 mM β-mercaptoethanol, 10 mM EGTA). The purity of each fraction was then analyzed by sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) and silver staining of the protein complexes.
The yeast cells grown overnight were collected, washed, and resuspended in a lysis buffer (20 mM Tris [pH 7.5], 50 mM KCl, 1 mM EDTA, 0.1% NP-40, 1 mM dithiothreitol). Cells were disrupted by being vortexed with glass beads for 30 min at 4°C. The lysate was clarified with centrifugation and subjected to SDS-PAGE and Western analysis with antibodies specific for mono-, di-, and trimethylated H3K4 (H3K4me1, -2, and -3); unmodified histone H3 was used as a loading control.
The methyltransferase activity of purified COMPASS toward recombinant histone H3 was tested as described previously (12). Purified COMPASS was incubated with recombinant histone H3 in the methyltransferase reaction buffer (50 mM Tris [pH 8.8], 20 mM KCl, 5 mM MgCl2, 0.5 mM dithiothreitol) along with cold S-adenosylmethionine. The methylation of histone H3 was tested by the application of the reaction mixture to SDS-PAGE gels and by Western analysis with antibodies to differently modified or unmodified forms of H3K4.
Immunofluorescence microscopy was performed essentially as described previously (2). Briefly, cells were grown to mid-log phase in yeast extract-peptone-dextrose (YPD) medium and fixed with 4% paraformaldehyde, followed by incubation with zymolase. Spheroplasted cells were placed on a slide and incubated with mouse anti-H3K4me3 primary antibody (Abcam) at a 1:500 dilution. For a secondary antibody, Alexa 488-conjugated goat anti-mouse immunoglobulin G (Invitrogen) was used at a 1:500 dilution. 4′,6′-Diamidino-2-phenylindole (DAPI) staining was performed by using 1 μg/ml for 5 min immediately before mounting with Citifluor (Ted Pella, Inc.). Cells were examined at room temperature using an imaging system (DM 5000; Leica Microsystems, Inc.) with a 100× HCX PL FLUOTAR objective (Leica Microsystems, Inc.). The images were captured with a digital camera (DFC 340 FX Alpha; Leica Microsystems, Inc.) and processed using AF6000 (Leica Microsystems, Inc.). Images were taken from multiple fields, with the same gain and exposure time for all samples. The images shown in this report are representative of what we observed in all of our experiments.
The cells were grown in raffinose-containing growth medium up to an OD600 of 0.9 and then switched to a galactose-containing growth medium for 90 min prior to formaldehyde-based in vivo cross-linking. Immunoprecipitations were performed following a modified chromatin immunoprecipitation (ChIP) protocol described previously (14). After immunoprecipitation, quantitative PCRs (qPCRs) were performed using primers for a GAL1 segment (PCR product is between 494 and 665 bp from the ATG of GAL1; 5′-CAGTGGATTGTCTTCTTCGGCCGC-3′ and 5′GGCAGCCTGATCCATACCGCCATT3′). The control primer pairs for a chromosome V (ChrV) segment (5′-GGCTGTCAGAATATGGGGCCGTAGTA-3′ and 5′-CACCCCGAAGCTTTCACAATAC-3′) and a DNA polymerase I (POL1) segment (5′-GGGTGGTTTAGTTTTTGAACCTGA-3′ and 5′-TAATAACCTTGGTAAAACACCCTG-3′) were used for background determination. For ChIP assay of PMA1, cells were grown in a dextrose-containing medium up to an OD600 of 1.0, followed by formaldehyde-based in vivo cross-linking. Immunoprecipitations and qPCR were performed as described above. A control primer pair for a POL1 segment was used for background determination. The primer pairs are as follows. GAL1-1 is the same as for the 494-to-665-bp PCR product described above, GAL1-2, 5′-CAGAGGGCTAAGCATGTGTATTCT-3′ and 5′-GTCAATCTCTGGACAAGAACATTC-3′; PMA1-1, 5′-ACAAACTGACCCATCTTACGGTTT-3′ and 5′-GACCGACGAAAAACATAACGAACT-3′; and PMA1-2, 5′-ACTGTTATCCCAACTGATGGTCGT-3′ and 5′-GAAAGTTTGGTCACCGTAATGTTTG-3′.
The coordinates for the Set1 homology model have been deposited in the Protein Model Data Base under structure accession number PM0075786.
Because Set1/COMPASS has proven to be challenging to characterize structurally, given its size and subunit complexity, we generated a homology model of the Set1 catalytic domain (Fig. (Fig.1a)1a) to gain a better insight into the molecular mechanism of H3K4 methylation by Set1/COMPASS. This model allowed us to identify residues within the active site of Set1 that may contribute to its product specificity or the degree of lysine methylation. Utilizing this model, we identified several active-site residues that may participate in substrate binding or catalysis. One such residue is Tyr1052 within the lysine binding channel (Fig. (Fig.1b1b).
To assess the potential role of Tyr1052 in Set1's methyltransferase activity and product specificity, we developed yeast strains in which Set1 can be mutated and incorporated into COMPASS for both in vivo and in vitro studies (Fig. (Fig.2A).2A). Using this system, we mutated the Phe/Tyr switch residue Tyr1052 to phenylalanine in Set1 (Fig. (Fig.2B).2B). The Y1052F mutant yields stable COMPASS, as demonstrated by the stability of Set1 (Fig. (Fig.2B).2B). Set1 carrying the Y1052F mutation exhibits greater H3K4 trimethyltransferase activity than the wild-type complex in vivo, in agreement with the Phe/Tyr switch model (Fig. 2C and D). An analysis of H3K4 methylation in the Tyr1052 mutant strain reveals an increase in the ratio of H3K4me3 to H3K4me2 relative to their levels in the wild type (Fig. (Fig.2D2D).
This phenomenon is also true for the in vitro methyltransferase activity of COMPASS. Purified COMPASS either from the wild-type strain or from the strain harboring a Set1 Y1052F mutant demonstrates similar stability within Set1 (Fig. 3A and B). However, complexes carrying Set1 Y1052F are much more active in H3K4 trimethylation than the wild-type enzyme (Fig. (Fig.3A).3A). Similar to the results of the in vivo studies (Fig. (Fig.2D),2D), the ratio of H3K4me3 to H3K4me2 is also increased in the in vitro studies, in contrast to wild-type COMPASS in which H3K4me2 is the most-abundant species. Furthermore, our analysis of the molecular composition of the wild-type and the Y1052F-mutated COMPASS indicated that the subunit composition of the complex is not altered by this single point mutation within Set1's catalytic pocket (Fig. (Fig.3B3B).
In previous studies, we demonstrated that Cps40 (Spp1) and Cps60 (Bre2) can regulate the pattern of H3K4 trimethylation and postulated that this may occur through direct regulation of the catalytic pocket of Set1 (24). Since we have identified a single point mutation within the catalytic pocket of Set1(Y1052F) that can enhance the H3K4 trimethylation activity of the enzyme, we wanted to determine whether this mutation suppresses the trimethylation loss in a cps40 (spp1) null strain. To this end, cps40 (spp1) and set1 double null cells were transformed with either the vector, the vector containing wild-type Set1, or the vector containing the Y1052F mutant. The Set1 levels were similar in the wild-type and mutant Set1-transformed cells, as assayed by Western analyses of whole-cell extracts (Fig. (Fig.4A)4A) and purified COMPASS (Fig. (Fig.4C).4C). In contrast, the H3K4 trimethylation levels are greatly increased in the Y1052F mutant relative to the levels with wild-type Set1, with no observable change in the H3K4 dimethylation level, demonstrating that the cps40 deletion is suppressed by Y1052F in vivo (Fig. (Fig.4B).4B). This observation indicates that Cps40 (Spp1)'s function is to regulate the H3K4 trimethylation activity of Set1/COMPASS through this residue. Similarly, COMPASS purified from a cps40 (spp1) null strain is defective in H3K4 trimethylation; however, the Y1052F mutation within Set1 suppresses this deficiency in an in vitro histone methyltransferase assay (Fig. (Fig.4D).4D). This observation is further supported by a previously published study indicating that Cps40 (Spp1) and Set1 can interact directly with each other in yeast (5a).
The studies presented here demonstrate that Tyr1052 of Set1 is inhibitory to H3K4 trimethylation, as wild-type COMPASS null for Cps40 (Spp1) is incapable of properly trimethylating H3K4 both in vivo and in vitro (Fig. 4B and D). However, the presence of Cps40 (Spp1) or the mutation of Tyr1052 to phenylalanine that results in the deletion of the hydroxyl group from the phenol side chain promotes H3K4 trimethylation by Set1 (Fig. 4B and D). Since our previous studies demonstrated that both Cps40 (Spp1) and Cps60 (Bre2) are required for proper H3K4 trimethylation, we also tested the role of Y1052F mutation in H3K4 trimethylation in the absence of Cps60 (Bre2). Indeed, we find that the mutation of Tyr1052 to phenylalanine promotes H3K4 trimethylation by Set1 in the absence of Cps60 (Bre2) (Fig. 5A and B). Interestingly, H3K4 dimethylation in a cps60 (Bre2) null strain is also complemented by a Y1052F mutation of Set1 (Fig. (Fig.5B).5B). Monoubiquitination of histone H2B by Rad6/Bre1 is required for proper H3K4 methylation by COMPASS (7, 25). We therefore tested whether the Set1 Y1052F mutant can compensate for the loss of Rad6. As a control for our in vivo studies, we tested for the Set1 levels in these cells and have confirmed that the deletion of rad6 or the point mutation of the Set1 catalytic domain did not alter Set1's stability in cells (Fig. (Fig.5C).5C). As shown by the results in Fig. Fig.5E,5E, the in vivo methylation activity of COMPASS in a rad6Δ background is not suppressed by the Set1 Y1052F mutation. This is also true for the in vitro studies (Fig. (Fig.5F).5F). When similar levels of Set1 within COMPASS purified from either the wild-type or rad6 null background were tested for H3K4 methylation, Y1052F did not suppress the loss of Rad6's effect on H3K4 methylation (Fig. (Fig.5F5F).
We previously demonstrated that H2B monoubiquitination/H3K4 methylation cross talk is regulated via the Cps35 (Swd2) subunit of COMPASS, as this subunit interacts with the chromatin of COMPASS-regulated genes in an H2B monoubiquitination-dependent manner and is required for proper H3K4 methylation (14). Furthermore, the Cps35 (Swd2) ortholog in mammals, Wdr82, also interacts with chromatin in a H2B monoubiquitination-dependent manner and is a specific component of the mammalian Set1/COMPASS and not of the MLL complexes (mammalian COMPASS-like complexes) (32). Recently, it has been proposed that Cps35 (Swd2) is ubiquitinated by the Rad6/Bre1 complex when both Cps35 (Swd2) and ubiquitin are overexpressed in yeast cells (31). Based on the observation that Cps40's association with chromatin is reduced in a strain carrying point mutations in the proposed ubiquitination sites of Cps35 (Swd2), it has been postulated that this histone cross talk in yeast is further regulated via recruitment of Cps40 (Spp1) through Cps35 (Swd2) (31). However, our data presented in this report indicate that Cps40 (Spp1) regulates the pattern of H3K4 trimethylation independently of H2B and/or Cps35 (Swd2) monoubiquitination. Cps40 (Spp1) regulates H3K4 trimethylation through a Phe/Tyr switch within Set1's catalytic pocket (Fig. (Fig.4),4), and the loss of H3K4 methylation as a result of the loss of H2B monoubiquitination is not suppressed by the Y1052F mutation (Fig. 5C to F). Therefore, it appears that these two pathways, Cps40 (Spp1)-dependent H3K4 methylation and monoubiquitination-dependent H3K4 methylation, are functioning via different mechanisms.
To further investigate the relative roles of Cps35 (Swd2) and Cps40 (Spp1) in histone cross talk, we compared the chromatin recruitment of each protein in the presence and absence of Rad6. ChIP was performed with myc-tagged Cps35 in wild-type and either rad6Δ or H2B(K123R) strains. As previously shown, Cps35's association with chromatin is Rad6- and H2B monoubiquitination dependent (Fig. 6A and B). To test the proposed model (31) that Cps40 (Spp1) interacts with chromatin in a Cps35 (Swd2)-dependent manner, we have utilized a cps35 deletion strain, recently generated by Nagy and colleagues (19). Cps60 was TAP-tagged in cps35Δ strains, and COMPASS was purified from this background. The resulting complex was analyzed by employing multidimensional protein identification technology (MudPIT). In this method, we use the spectral counts, i.e., the total number of tandem mass spectrum-matching peptides from a protein. Spectral counts have been defined as excellent markers of protein abundance in shotgun proteomics analyses (15). Dividing the spectral count by molecular weight defines a spectral abundance factor (20), and hence, proteins of different sizes can be compared. Previously, we demonstrated that wild-type COMPASS contains one copy of the Set1, Cps50 (Swd1), Cps40 (Spp1), and Cps30 (Swd3) subunits; less than one copy of Cps35 (Swd2); two copies of Cps60 (Bre2); and at least three copies of Cps25 (Sdc1). Since the loss of Cps35 (Swd2) results in some loss of COMPASS stability, we normalized the levels of each COMPASS subunit from both wild-type and cps35 (swd2) null backgrounds to the level of Set1. Based on our MudPIT analysis of COMPASS purified from a cps35 (swd2) null background, the level of Cps40 (Spp1) appears to be unchanged relative to the level of Set1 (Fig. (Fig.6C).6C). We see an increase in the levels of Cps60 and Cps25 in COMPASS from the cps35 null background. This observation is to be expected, as we have tagged Cps60 in this background and Cps60 and Cps25 form a heteromeric complex. Since the Set1 level is reduced in a cps35 null background, we therefore observe an increase in the level of the Cps60/25 heteromeric complex. Overall, contrary to a previously published study (31), our data indicate that the association of Cps40 (Spp1) with COMPASS does not require Cps35 (Swd2), as we are able to purify the same ratio of Cps40 (Spp1) with COMPASS in the absence of Cps35 (Swd2) as in its presence (Fig. (Fig.6C).6C). Our data also indicate that the loss of H2B monoubiquitination does not alter the association of Cps40 with chromatin of COMPASS-regulated genes, contrary to a previously published study (31) (Fig. 6D and E). We have also previously demonstrated that Cps60's association with COMPASS and COMPASS-regulated genes is unchanged in strains defective for H2B monoubiquitination (14).
Collectively, these studies indicate that while Cps35 (Swd2) is a key mediator of cross talk between H2B monoubiquitination and H3K4 methylation, as initially proposed (14), Cps40's (Spp1's) association with COMPASS does not depend on the presence of Cps35 (Swd2) within the complex or its prior monoubiquitination. Thus, while both Cps35 and Cps40 regulate the transitioning of methyl states of H3K4 by COMPASS, each does so by a distinct mechanism, demonstrating the complexity of H3K4 methylation regulation by COMPASS.
In histone methyltransferases that are functional as single polypeptides, the Phe/Tyr switch controls product specificity by modulating the binding of the active-site water molecule (5). In di- and trimethyltransferases, the presence of a phenylalanine or other hydrophobic residue in the Phe/Tyr switch site modulates the affinity for the active-site water, allowing multiple methylation reactions (s). In contrast, the methyltransferase activity and product specificity of Set1 are dependent on the subunits constituting COMPASS. Although SET1 would be predicted to be an H3K4 monomethyltransferase due to Phe/Tyr switch residue Tyr1052 (Fig. (Fig.1b),1b), Cps40 (Spp1) and Cps60 (Bre2) have a predominant role in determining its product specificity in the context of COMPASS. From the results of this study, we propose that these subunits could associate directly with the catalytic domain of Set1, possibly through interactions with the histone substrate binding cleft or the PostSET motif (Fig. (Fig.1a1a and 7A and B). Indeed, a recent structural study of the human MLL1 catalytic domain supports this model (27). In the crystal structure of an MLL1 ternary complex, the histone H3 peptide is bound in a deep channel formed by the inserted SET motif (iSET) helix and the PostSET domain, as illustrated in our homology model of the Set1 catalytic domain (27) (Fig. (Fig.1a).1a). The shifting of the iSET region away from the histone binding cleft results in an active site that is relatively open and solvent exposed when compared to the structure of the H3K9-specific methyltransferase Dim-5 and is believed to contribute to the weak methyltransferase activity of the free enzyme. The activity of MLL1 is greatly stimulated in the presence of the RbBP5 and Ash2L core subunits, which have been proposed to facilitate the closure of the iSET helix over the active site. Correlatively, we have shown that the association of Cps40 and Cps60, the functional homologs of Ash2L in yeast, with Set1 promote H3K4 di- and trimethylation, presumably through an analogous mechanism (24). Future studies are necessary to elucidate the nature of the interactions among Set1, Cps40, and Cps60, as well as the mechanism by which these interactions alter the activity and product specificity of COMPASS.
We thank Jon Wilson for providing the coordinates for the crystal structures of the MLL1 catalytic domain, Peter Nagy for sharing the cps35Δ strain, Jessica Jackson for technical assistance, and Edwin Smith for conversations and suggestions during the course of this study. We are also thankful to Laura Shilatifard for editorial assistance.
This work was supported in part by National Institutes of Health grants GM073839 to R. C. Trievel and GM069905 to A. Shilatifard.
Published ahead of print on 27 April 2009.