Identification of Selective Inhibitors of NAD+-dependent Deacetylases Using Phenotypic Screens in Yeast*

Sir2 and Hst1 are NAD+-dependent deacetylases involved in transcriptional repression in yeast. The two enzymes are highly homologous yet have different sensitivity to the small-molecule inhibitor splitomicin (compound 1) (Bedalov, A., Gatbonton, T., Irvine, W. P., Gottschling, D. E., and Simon, J. A. (2001) Proc. Natl. Acad. Sci. U. S. A. 98, 15113–15118). We have now defined a critical amino acid residue within a small helical module of Hst1 that confers relative resistance to splitomicin. Parallel cell-based screens of 100 splitomicin analogues led to the identification of compounds that exhibit a higher degree of selectivity toward Sir2 or Hst1. A series of compounds based on a splitomicin derivative, dehydrosplitomicin (compound 2), effectively phenocopied a yeast strain that lacked Hst1 deacetylase while having no effect on the silencing activities of Sir2. In addition, we identified a compound with improved selectivity for Sir2. Selectivity was affirmed using whole-genome DNA microarray analysis. This study underscores the power of phenotypic screens in the development and characterization of selective inhibitors of enzyme functions.

Sir2-like enzymes constitute a family of NAD ϩ -dependent deacetylases found in diverse organisms ranging from bacteria to humans (1,2). These highly conserved enzymes catalyze a reaction that requires the consumption of NAD ϩ for the removal of the acetyl group from substrate lysine residues to generate nicotinamide, O-acetyl-ADP-ribose, and lysine (3)(4)(5)(6)(7)(8)(9)(10). Initially characterized as histone deacetylases, this family of enzymes was subsequently shown to have a broad range of substrates including p53, BCL6, and ␣-tubulin in mammalian cells and acetyl-CoA synthetase in bacteria (11)(12)(13)(14)(15)(16)(17). The yeast Saccharomyces cerevisiae has five Sir2-like proteins: Sir2p, Hst1p (homologue of Sir two), Hst2p, Hst3p, and Hst4p (homologues of Sir two) (1,18). Sir2p and its closest homologue, Hst1p, act as transcriptional repressors by promoting targeted histone deacetylation (11, 19 -21). These two enzymes, however, play distinct cellular roles, since they are directed to different chromatin regions by specificity factors. Sir2p, which is found in two separate multiprotein complexes, is critical for transcriptional silencing of large chromosomal domains at three loci: telomeres, the silent mating-type loci (HMR and HML), and the ribosomal RNA-encoding DNA (reviewed in Ref. 22). Hst1p-mediated repression, in contrast, is restricted to specific genes through the DNA-binding protein Sum1p and the tethering factor Rfm1p (23,24). Hst1p has previously been known to participate in repression of middle sporulation genes during mitotic growth (23) but has recently been shown to serve as a sensor and a regulator of cellular NAD ϩ levels through controlling the expression of genes involved in de novo NAD ϩ biosynthesis and the import of nicotinic acid (25). Little is known about cellular functions of other NAD ϩ -dependent deacetylases in yeast. Hst2p is a cytoplasmic enzyme that accounts for the majority of NAD ϩ -dependent deacetylase activity in a whole-cell lysate (5,26). Hst3p and Hst4p are highly related enzymes that perform overlapping functions in telomeric silencing and maintenance of genome integrity (1).
Small-molecule enzyme inhibitors are powerful tools for dissecting the functional roles of enzymes in vivo and represent a major class of drugs used in humans. However, since individual enzymes almost universally coexist with highly related members of enzyme families, with different members often having distinct biological functions, the utility of enzyme inhibitors as drugs or biological tools is critically dependent on specificity of the inhibitors. Whereas cell-free enzyme inhibition assays continue to be useful for assessing the specificity of inhibitors, we and others have employed DNA microarrays in whole-cell assays to determine the in vivo specificity of small-molecule inhibitors for their targets (27)(28)(29)(30)(31)(32). We have found that splitomicin (1; 1,2-dihydro-3H-naphtho [2,1-b]pyran-3-one) is active against Sir2p and to a lesser extent Hst1p, while having no inhibitory activity against Hst2p, Hst3p, or Hst4p (32).
In this study, we evaluated 100 analogues of splitomicin (1) for their relative inhibitory properties against yeast Sir2p or Hst1p with the goal of developing specific inhibitors for either of these two related deacetylases. Similar to the original approach that led to the identification of splitomicin (1) (32), iterative phenotypic screens were carried out to identify inhibitors with desired selectivity for Sir2p or Hst1p, in which properties of compounds were selected based on their ability to phenocopy sir2 or hst1 deletion mutant strain in yeast. Using the isogenic yeast strains, we were able to identify structural features necessary for specificity by comparing activity against the two deacetylases. Our results demonstrate that it is feasi-ble to achieve high selectivity for inhibition of Sir2p or Hst1p and that cell-based assays are a useful approach for identification and evaluation of selective inhibitors of NAD ϩ -dependent deacetylases.

EXPERIMENTAL PROCEDURES
Molecular Modeling-Yeast Sir2p and Hst1p were modeled based on the human SIRT2p structure using the SwissModel server (33,34). Molecular docking of splitomicin (1) onto the human SIRT2p structure was performed manually using Xfit version 4.0 (35). Structure figures were made with PyMol version 0.80 (36).
Yeast Media and Strains-All strains were grown in synthetic complete (SC) 1 medium or selective drop-out medium containing 2% glucose. Compounds or the solvent Me 2 SO was added to media with the final Me 2 SO concentration of 0.5%. Genotypes of yeast strains are listed in Table I. Additional deletion mutant strains, sir2::kanMX, hst1::kanMX, hst2::kanMX, hst3::kanMX, and hst4::kanMX derived from BY4741 or BY4742 were obtained from a haploid deletion collection (Saccharomyces Genome Deletion Project, Research Genetics). YAB14053 and YAB14079 were generated by one-step PCR-mediated gene replacement using integrating pRS400 plasmid containing the kanMX gene (37) or pRS406 containing URA3 (38), respectively.
Silencing Assays-For spot dilution assays, individual colonies were pregrown overnight at 30°C in SC liquid medium lacking leucine (see Fig. 2) or SC (see Fig. 5, A and B). Cells were pelleted and resuspended in sterile water, and A 660 values of cell suspensions were normalized. Ten microliters of the cell mixture and of several 5-fold (see Fig. 2) or 10-fold (see Fig. 5, A and B) serial dilutions were spotted onto the appropriate tester plates (see legends to Figs. 2 and 5) with Me 2 SO or compounds. Cells were incubated at 30°C for 36 h before photography. Compound screening was performed as previously described (32). 2 Synthesis of Splitomicin Analogues-Splitomicin analogues were synthesized or purchased from commercial sources, of which compounds 1 to 45 were recently described. 2 Compounds 13,14,16,27,31,and 33 were not evaluated in this study. The chemical structures of compounds 46 -106 are shown in Fig. 9 of the supplemental material.
Northern Blot Analysis-Strains were inoculated into SC medium, grown overnight at 30°C, diluted to 2.5 ϫ 10 6 cells/ml (15 ml) in the medium with Me 2 SO or compounds, and grown for an additional 10 h. Total RNA from logarithmically grown cultures was isolated using the hot acid phenol method. 7.5 g of RNA was separated by electrophoresis through 4-morpholinepropanesulfonic acid/formaldehyde-agarose gels, transferred to GeneScreen Plus membranes (PerkinElmer Life Sciences), and probed with DNA labeled with [ 32 P]dCTP using the Ready-To-Go DNA Labeling Beads (-dCTP) kit (Amersham Biosciences). Blots were exposed to a PhosphorImager screen (Molecular Dynamics) and quantified using ImageQuant software (Amersham Biosciences). DNA for probes was made as previously described (TNA1, PDA1, and ␣2) (25,32).
DNA Microarrays-cDNA microarray experiments were performed as previously described (25,32). The complete data sets containing the mean log 2 of the expression ratios and p values are available in the Microsoft Excel spreadsheet format on the World Wide Web at www.fhcrc.org/labs/simon/bedalov/jbc.html.
NAD ϩ Measurements-Total cellular NAD ϩ levels were measured as previously described (25,40). Three separate cultures were grown in SC medium lacking tryptophan overnight at 30°C, diluted to 3.5 ϫ 10 6 cells/ml (50 ml) in the medium with Me 2 SO or compounds, and grown for an additional 7-11 h until reaching a density of 2.5 ϫ 10 7 cells/ml.

Exchange of a Small Helical Module of Sir2p for the Equivalent Segment of Hst1p Results in Resistance to Splitomicin
(1)-A screen for the alleles of SIR2 that are resistant to splitomicin (1) has identified a small helical module of Sir2p that creates a putative substrate recognition site (33,41,42) as a probable site where splitomicin (1) acts (32) (Fig. 1A). Although Hst1p is highly homologous to Sir2p in this 50-amino acid region (92% similarity, 86% identity) (Fig. 1B), it displays relative resistance to splitomicin (1) (32). To better understand the basis for this resistance, we prepared a model of Sir2p or Hst1p bound to the inhibitor based on the crystal structure of human SIRT2p (33) and found seven nonconserved amino acid residues surrounding the binding site (Fig. 1C). We hypothesized that engineering these seven residues of Hst1p in the Sir2p scaffold structure would give rise to resistance similar to that of Hst1p. The Sir2-Hst1 chimeric protein was initially tested for its ability to complement a telomeric silencing defect of a sir2 deletion mutant strain (sir2⌬) marked at the left arm of chromosome VII with a URA3 reporter gene (43). Silencing of the telomeric URA3 reporter was judged by the ability of cells to grow on medium containing 5-fluoroorotic acid (5-FOA). Silencing in the sir2⌬ strain was fully complemented by the expression of the chimeric protein (SIR2-HST1 chimera) ( Fig.  2A). We next compared the antisilencing effects of splitomicin (1) in strains containing the wild-type (SIR2) and the SIR2- HST1 chimera. Splitomicin (1) impaired the growth of the SIR2 strain at 5 M on medium containing 5-FOA, indicating a loss of silencing (Fig. 2B). However, the SIR2-HST1 chimera strain grew 5-25-fold better than the SIR2 strain on 5-FOA with 5 M splitomicin (1). This result suggests that unique residues within the small helical module of Hst1p confer resistance to the antisilencing effects of splitomicin (1). To identify the responsible residues, a series of SIR2 mutants were constructed in which two residues at a time were changed to the cognate residues in HST1 (SIR2-K285R/D290E, SIR2-Y298L/N299D and SIR2-M302L/H303Q) (Fig. 1, B and C). Silencing was at the wild-type levels with the three mutant alleles in the absence of splitomicin (1) ( Fig. 2A). However, only SIR2-Y298L/ N299D conferred resistance to the inhibitory effect of splitomicin (1). The SIR2-Y298L/N299D allele showed an equivalent degree of resistance to SIR2-Y298N, which had been independently identified as a splitomicin (1)-resistant allele in our prior study (32) (Fig. 2B). Interestingly, SIR2-Y298L/N299D and SIR2-Y298N conferred about 5-fold resistance to splitomicin (1) relative to the SIR2-HST1 chimera, suggesting that other mutations in the chimera (i.e. K285R, D290E, M302L, and H303Q) counteract the splitomicin (1) resistance effect of the Y298L mutation. Indeed, SIR2-K285R/D290E and SIR2-M302/H303Q conferred slight sensitivity to splitomicin (1) relative to the wild-type SIR2 ( Fig. 2B and data not shown). These results suggest that a probable key residue for the relative resistance of Hst1p toward splitomicin (1) is leucine 244, which corresponds to tyrosine 298 of Sir2p. (1) is still active against both Sir2p and Hst1p, although it has relative selectivity for Sir2p (32). The ability of splitomicin (1) to discriminate between the small helical modules in Sir2p and Hst1p provided a strong incentive to improve selectivity through the modification of splitomicin (1). To identify compounds that exhibit a higher degree of selectivity toward either Sir2p or Hst1p, we synthesized a library of 100 splitomicin analogues. We have recently described the synthesis and in vivo evaluation of a selection of FIG. 1. Model of splitomicin (compound 1) with yeast Sir2p or Hst1p. A, the crystal structure of human SIRT2p (33) with splitomicin (1; red) docked in the small helical module. Side chains of residues that, when mutated in Sir2p, confer resistance to splitomicin (1) (H286Q, L287M, or Y298N) (32) are mapped in blue (space-filling representation). B, sequence alignment of yeast Sir2p, Hst1p, and human SIRT2p in the small helical module. The secondary structure elements are shown aligned above the sequences. C, an expanded view of the potential splitomicin (1)-binding site in the superimposed models of yeast Sir2p and Hst1p rotated by ϳ90°C relative to A. Splitomicin (1) atom colors are as follows. Green, carbon; red, oxygen; gray, hydrogen. Side chains of the resistance-conferring residues His 286 , Leu 287 , and Tyr 298 in Sir2p are shown in red. The corresponding residues in Hst1p, His 232 , Leu 234 , and Leu 244 , respectively, are shown in blue. ␣-Carbons of the seven nonconserved residues between Sir2p and Hst1p that are mutated in this study besides Y298L are marked in black (space-filling representation). Side chains of conserved putative splitomicin (1)-contacting residues are shown in yellow.

Cell-based Chemical Screen of 100 Splitomicin Analogues for Sir2p Inhibitors-Splitomicin
these analogues for Sir2p-inhibitory activity. 2 Inhibition of Sir2p by the library of splitomicin-like small molecules was assessed in a cell-based assay through the ability of the compounds to induce growth of the strain with the URA3 telomeric reporter in medium lacking uracil. Because the screen for Sir2p inhibitors makes use of functional readout (i.e. cell growth), it integrates target inhibition with cytotoxicity of compounds when evaluating the activity at high concentration (Fig. 3A). For example, the cytotoxicity of splitomicin (1) becomes apparent at 5 M, with an IC 50 value of 82 M. 2 Therefore, to accurately assess the Sir2p inhibition, the activity of the compounds was expressed as the minimal inhibitory concentration (MIC), a concentration that resulted in 10% of the maximum growth enhancement observed using a potent, low cytotoxic analogue, 41 (Fig. 3A). 2 We identified 18 Sir2-active compounds that had a MIC of 0.49 -48 M, with splitomicin (1) having the lowest MIC. 2 The remaining 82 analogues had a MIC of Ͼ150 M and were classified as Sir2p-inactive compounds ( Fig. 9 of the supplemental material).
Primary Screen for Hst1p Inhibitors Based on the Derepression of an MSE-lacZ Reporter-To evaluate the ability of splitomicin analogues to inhibit Hst1p, we used a phenotypic assay in which inhibition of Hst1p activity permitted lacZ expression. It has been shown that the promoters of many middle sporulation genes contain a cis-acting element called middle sporulation element (MSE) that is repressed in an Hst1p-dependent fashion (23). Thus, to assess Hst1p inhibition, yeast strains were transformed with a plasmid that contained a lacZ reporter gene inserted downstream of the NDT80 (Ϫ78) MSE (23). In a ␤-galactosidase filter assay, a wild-type (HST1) strain showed strong repression of lacZ, whereas the same strain treated with 50 M splitomicin (1) showed comparable lacZ expression to an hst1⌬ strain (Fig. 3B). We screened the compound library using a single 50 M concentration for those that perturbed repression. All of the 18 Sir2p-active compounds, except for compound 26, induced lacZ expression ( Fig. 3B and data not shown). Interestingly, 26, which had the 5-benzyloxy group on splitomicin (1) (Fig. 4C), showed no derepression, suggesting that it might be selective for Sir2p (Fig. 3B). Of 82 Sir2p-inactive analogues, we identified 48 hits with weak to moderate levels of Hst1p-dependent derepression, with the most potent compounds shown in Fig. 3C. Potency was assayed by examining the duration and intensity of lacZ expression as well as by repeating the screen at lower concentrations (data not shown).
Secondary Assay for Hst1p Inhibition by the Derepression of the Nicotinic Acid Permease Gene TNA1-Hst1p not only represses middle sporulation genes, but it also represses genes encoding enzymes in the de novo NAD ϩ synthetic pathway and high affinity nicotinic acid permease, TNA1 (25). As a secondary assay for Hst1p inhibition, we prepared RNAs from wild-type strains treated with 50 M Hst1p-active compounds identified in the primary screen and examined the expression levels of TNA1 by Northern blot analysis. Splitomicin (1) induced TNA1 expression to 80% of the level produced in an hst1⌬ strain (Fig. 4A). The expression levels of TNA1 for splitomicin (1) as well as compounds 24, 25, 20, and 40 correlated well with their MIC for Sir2p, suggesting that these compounds inhibit both Hst1p and Sir2p (Fig. 4, A and C). However, TNA1 derepression level was relatively low (35%) for compound 26 despite its significant inhibition of Sir2p (MIC of 0.97 M). Relatively weak activity of 26 against Hst1p was consistent with the lack of activation of the MSE-lacZ reporter. Fig. 4B shows the TNA1 derepression of the eight Sir2p-inactive analogues that induced high lacZ derepression. Of these analogues, compound 2, dehydrosplitomicin (3Hnaphtho[2,1-b]pyran-3-one), displayed the highest TNA1 derepression (50%) (Fig. 4, B and C). Dehydrosplitomicin (2) showed a dose-dependent derepression of TNA1 transcript with a maximal activity at 150 M (Fig. 4D). The other 40 Sir2p-inactive compounds that displayed weak lacZ derepression showed less than 34% of TNA1 derepression (Fig. 4C).
Dehydrosplitomicin (2) Has No Effect on the Sir2p-silencing Functions at Telomeres and the Silent Mating-type Loci-To verify that dehydrosplitomicin (2) is selective for Hst1p, its inhibitory activity toward Sir2p was further evaluated in cellbased assays. We first reassessed telomeric silencing using a sensitive spot dilution assay. Wild-type (SIR2) cells containing the telomeric URA3 reporter were unable to grow on medium lacking uracil but were able to grow on medium containing 5-FOA (Fig. 5A). Disruption of silencing in a sir2⌬ strain dere- pressed the URA3 reporter, which enabled cells to grow on medium lacking uracil and killed cells on medium containing 5-FOA. Splitomicin (1) with doses as low as 6.25 M derepressed the URA3 reporter, phenocopying the sir2⌬ mutant. In contrast, even at 100 M, dehydrosplitomicin (2) failed to perturb silencing at the telomere, although it exhibited cytotoxicity. The inability of dehydrosplitomicin (2) to inhibit Sir2p function in silencing was not limited to telomeric genes. Dehydrosplitomicin (2) was also unable to perturb silencing of an HMR TRP1 reporter (44) (Fig. 5B). Additionally, SIR2 silencing measured by transcription of the ␣2 gene at the HML␣ locus in a MATa strain was also not affected by dehydrosplitomicin (2) (Fig. 5C). The failure of dehydrosplitomicin (2) to inhibit Sir2p activity in cell-based assays was not due to the lack of its permeability, because dehydrosplitomicin (2) derepressed the MSE-lacZ reporter (Fig. 3C) and TNA1 at 50 M (Fig. 4B) and exhibited toxicity at 100 M (Fig. 5, A and B). Taken together, our results show that dehydrosplitomicin (2) does not perturb the silencing functions of Sir2p, and thus its inhibitory activity is selective for Hst1p.
Unlike splitomicin (1), which displayed only 3-fold selectivity, compound 26 displayed 13-fold selectivity in favor of Sir2p over Hst1p (19%/1.5%) (Fig. 7B). Despite the enhanced selectivity for Sir2p, 26 was about 2-fold less potent in inhibiting Sir2p than splitomicin (1), as judged by the number of SIR2affected transcripts (19% for 26 and 41% for splitomicin (1)), which was expected from the MIC (0.97 M for 26 and 0.45 M for splitomicin (1)). Compound 26 up-regulated two of the HST1-repressed genes (1.5%) with only one of the two overlapping transcripts specific to HST1. Conversely, only 6.3% of transcriptional up-regulation by compound 26 was mediated through HST1, whereas close to half was mediated through SIR2. As a result, off-targets accounted for more than half of observed up-regulation, since there was no overlap between the genes up-regulated by compound 26 and in hst2⌬, hst3⌬, or hst4⌬ mutant. 3 Since dehydrosplitomicin (2) was less potent than splitomicin (1) or compound 26, the wild-type strain was treated with higher concentrations (50 M) of dehydrosplitomicin (2). Remarkably, the selectivity of dehydrosplitomicin (2) was reversed for Hst1p, since it exhibited a 5-fold preference for Hst1p over Sir2p (6.9%/1.4%) (Fig. 7C). Whereas dehydrosplitomicin (2) up-regulated only one of the SIR2-repressed genes (1.4%), it also up-regulated nine HST1-repressed genes (6.9%) with genes involved in de novo NAD ϩ synthesis such as BNA4 and BNA2 and the middle stages of sporulation such as SPS4 among the most highly regulated transcripts. 3 Dehydrosplitomicin (2) up-regulated a total of 26 genes, where 35% were mediated through Hst1p, 3.8% through Sir2p, 7.7% through Hst3p, and 58% through other targets. Of the two HST3-responsive genes regulated by dehydrosplitomicin (2), one was also controlled by SIR2, leaving only one specific to HST3. Taken together, the analysis of the transcriptional profiles confirms distinct Sir2p and Hst1p selectivity profiles of splitomicin (1) and newly described compounds, with 26 exhibiting higher selectivity toward Sir2p and dehydrosplitomicin (2) having preferential activity toward Hst1p.
Compound 25 and Dehydrosplitomicin (2) Increase Steadystate NAD ϩ Levels-We have recently shown that elimination of HST1-mediated repression of genes involved in de novo NAD ϩ synthesis leads to increased cellular NAD ϩ levels (25). The elevated NAD ϩ levels in an hst1⌬ mutant have been shown to be entirely due to up-regulation of the de novo pathway and not due to decreased NAD ϩ consumption (25). To assess whether this effect can be phenocopied with smallmolecule inhibitors of Hst1p, we measured the steady-state NAD ϩ levels in wild-type cells treated with these compounds at 50 M. Two compounds were chosen for the analysis: compound 25, which was the most potent in derepressing the TNA1 gene but was not selective (Fig. 4, A and C), and dehydrosplitomicin (2), which was highly selective toward Hst1p. Compound 25 increased NAD ϩ levels 57% (p ϭ 0.0005) in wild-type cells, which was about 80% of the level observed in hst1⌬ cells (Fig. 8). However, compound 25 also showed a 15% increase in NAD ϩ in hst1⌬ cells (p ϭ 0.0009), indicating that the compound is not acting solely through the HST1 pathway. The Hst1p-independent rise in NAD ϩ levels with relatively high doses of compound 25 (50 M) may be due to inhibition of other NAD ϩ -dependent enzymes, which could result in decreased NAD ϩ consumption. We have observed that deletion of HST2 or HST3 results in a small increase in NAD ϩ levels probably through reduced NAD ϩ consumption (25). In contrast, the Hst1p-selective inhibitor dehydrosplitomicin (2) increased NAD ϩ levels entirely through the inhibition of Hst1p, since the treatment of the hst1⌬ cells did not lead to further increase in NAD ϩ levels (Fig. 8). In the wildtype cells, dehydrosplitomicin (2) increased NAD ϩ levels 25% (p ϭ 0.02). The lower level of increase in NAD ϩ reflected lower potency of dehydrosplitomicin (2) compared with compound 25. These data demonstrate that pharmacologic inhibition of Hst1p by compound 25 and dehydrosplitomicin (2) phenocopies up-regulation of the de novo NAD ϩ synthetic pathway and the increased cellular NAD ϩ levels of an hst1⌬ strain. DISCUSSION The traditional approach to the development of specific inhibitors primarily relies on in vitro screens. Since this method tracks one target at a time, identification of exquisitely specific inhibitors is a daunting task. Primary screens yield agents with undetermined specificity and need to be followed by counterscreens against related enzymes that allow identification of inhibitors with a higher degree of selectivity. For example, the specificities of numerous compounds have been examined against a large panel of protein kinases in enzyme inhibition assays (45,46). However, the fact that a compound does not inhibit unintended targets in vitro cannot conclusively prove its specificity, because the compound might bind to a conformational epitope in vivo that is not found in vitro, to a protein-protein interface, or simply to a target that is not represented in the in vitro panel. It would be ideal to ensure that all relevant targets are examined simultaneously in a physiological context, where they are present in their native conformations. Here, we use cellbased screens for the identification of inhibitors that demonstrate a high degree of selectivity against two members of yeast NAD ϩ -dependent deacetylases, Sir2p and Hst1p. The phenotypic assays take advantage of distinct cellular roles of these two highly related nuclear enzymes.
The small helical module in Sir2p and Hst1p, which is the postulated substrate recognition domain of the enzymes, is highly conserved (33,41,42), raising the question of why splitomicin (1), which probably alters or blocks substrate access by interacting with this module, can be relatively selective toward Sir2p (32). To answer this question, we carried out a grafting analysis, which replaced the small helical module in Sir2p with Hst1p sequences to form a chimeric protein. The chimera was fully functional in silenc-ing and conferred resistance to splitomicin (1) (Fig. 2). The molecular basis for the selectivity can be attributed largely to tyrosine 298 in Sir2p, which corresponds to leucine 244 in Hst1p, based on the analysis of point mutants in Sir2p (Figs. 1C and 2B). The model of yeast Sir2p or Hst1p based on the human SIRT2p structure (33) suggests that splitomicin (1) can fit into the hydrophobic cleft of the small helical module (13 Å long and 10 Å wide) by contacting conserved solventexposed hydrophobic residues lining the cleft, which are probably used for substrate recognition (Fig. 1C in yellow). Among these residues, which are identical in Sir2p and Hst1p, phenylalanine 296 has been shown to be required for Sir2p activity because the SIR2-F296L allele is completely defective in silencing and correlates with a decrease in enzymatic activity (47). Although splitomicin (1)-contacting residues are identical in Sir2p and Hst1p, there must be sufficient structural differences in the binding clefts of these two enzymes to explain the inhibitor selectivity. Positioned at the hinge between the two short helices ␣ 4 and ␣ 5 , the critical selectivity-determining residue tyrosine 298 is distant from the proposed splitomicin (1)-binding site (14 Å away) (Fig.  1C). The bulk of tyrosine 298 probably allows a favorable conformation of the cleft into which splitomicin (1) can fit. Indeed, substitution with a smaller residue such as leucine (in Hst1p) and asparagine (in the SIR2-Y298N splitomicin (1)-resistant mutant) results in reduced inhibition (Fig. 2B). This substitution possibly leads to more flexibility in the cleft and alters the orientation and conformation of the ␣ 4 helix, which presumably contacts with splitomicin (1) through the conserved hydrophobic residues (i.e. proline 292, valine 295, and phenylalanine 296) (Fig. 1C). Thus, Hst1p may possess a wobbly site interaction with splitomicin (1), causing relative resistance to splitomicin (1).
Whole-genome DNA microarray analysis examines the effects of splitomicin (1), compound 26, and dehydrosplitomicin (2) against all related targets (i.e. Sir2p, Hst1p, Hst2p, Hst3p, and Hst4p) as well as unintended targets to assess their specificities. The three compounds do not interfere with the Hst2p, Hst3p, or Hst4p pathway. Compared with splitomicin (1), dehydrosplitomicin (2) and compound 26 elicit a higher proportion of transcriptional responses that are not related to NAD ϩdependent deacetylases. The off-target activity of dehydrosplitomicin (2) is a likely explanation of cytotoxicity exhibited at high concentration (Fig. 5, A and B). The evaluation of Sir2p and Hst1p reveals that splitomicin (1) is only 3-fold selective for Sir2p over Hst1p (Fig. 7A). Compound 26, however, achieves 13-fold selectivity in favor of Sir2p over Hst1p (Fig.  7B). The 5-benzyloxy substituent of compound 26 (Fig. 4C) ought to be tolerated well in the binding cleft of Sir2p, since it presumably directs toward the empty space near the L1 loop ( Fig. 1C). A possible explanation for the ineffective inhibition of Hst1p by compound 26 is that the nonconserved residues proximal to the binding site may influence the shape and/or dynamic behavior of the binding cleft, blocking access to the 5-benzyloxy group.
The nature of the lactone ring in splitomicin analogues is critical for the reversed selectivity for Hst1p. The Sir2p-inhibitory activity is abolished by unsaturation (dehydrosplitomicin (2), 107, and 108) or bulky substitution (e.g. phenyl group in 23, 109, and 110) of the saturated six-membered lactone ring. Analogues with these modifications, however, remain active against Hst1p at 39 -62% of the level observed in splitomicin (1), with dehydrosplitomicin (2) being the most potent (Fig. 6C). The origin of the selectivity for Hst1p can be attributed to subtle differences in molecular shapes (e.g. a planar conformation in unsaturated analogues versus a slightly puckered conformation in splitomicin (1)) or in electronics. The anticipated flexibility of the Hst1p cleft architecture may lead to the subtle adjustment of specific residue positions to accommodate for these analogues, contributing to the observed selectivity profiles. We have recently described that the Sir2p inhibition activity of a splitomicin analogue is dependent on the reactivity of the carbonyl group in the lactone ring. 2 Thus, the lower potency against Hst1p in the Hst1p-selective analogues as compared with splitomicin (1) might result from its lower reactivity of the lactone ring toward nucleophiles present in the cleft (e.g. tyrosine 281 in Sir2p or tyrosine 227 in Hst1p) (Fig.  1, B and C). However, a detailed understanding of how compounds in this class of NAD ϩ -dependent deacetylase inhibitors interact with enzyme active site will probably require a structural determination of an inhibitor-enzyme complex.
A cell-based approach largely bypasses biochemical evaluation of inhibitory activity, which in the case of NAD ϩ -dependent deacetylases, correlates poorly with in vivo effects of inhibitors. Whereas splitomicin (1) exhibits Sir2p inhibition in vivo at submicromolar concentrations (MIC ϭ 0.49 M), its IC 50 as measured using yeast extract is only 50 -80 M (32). The large difference between the in vivo and the in vitro activity is not limited to the inhibitory efficacy of splitomicin (1). The in vivo selectivity of compounds also does not reflect the in vitro selectivity, since dehydrosplitomicin (2), which does not inhibit Sir2p in vivo, is a more potent inhibitor of Sir2p than Hst1p histone deacetylase activity in vitro (data not shown). These discrepancies may reflect a different ability of the inhibitors to interfere with Sir2-like enzymes when deacetylating the in vitro substrates as opposed to nucleosomes in the full chromatin context. Similar reasoning has been used to explain the different in vitro versus in vivo properties of the SIR2-R275A mutant (48). This SIR2 allele contains a point mutation within the small helical module, where splitomicin is proposed to act, and is fully active as a histone deacetylase in vitro. Although this protein is well expressed and is appropriately recruited to chromatin, it is incapable of establishing or maintaining a silenced state at target loci (48).
In summary, our study demonstrates the identification of splitomicin analogues with improved selectivity using a process that relies entirely on phenotypic assays. Traditionally, structure-activity relationships have been constructed using in vitro inhibition of a single target. We now show that cell-based assays can be used to construct structural features necessary for specificity by comparing activity of compounds against two independent but highly related targets. Splitomicin (1), which is highly active against Sir2p and to a lesser extent Hst1p, served as a lead compound for the identification of compound 26 and dehydrosplitomicin (2), which are highly selective for Sir2p or Hst1p, respectively. This result serves as a proof of concept in the development of highly selective inhibitors using a chemical genetic approach.