Structure-based development of novel sirtuin inhibitors.

Sirtuins are NAD+-dependent protein deacetylases regulating metabolism, stress responses, and aging processes. Mammalia possess seven Sirtuin isoforms, Sirt1-7, which differ in their subcellular localization and in the substrate proteins they deacetylate. The physiological roles of Sirtuins and their potential use as therapeutic targets for metabolic and aging-related diseases have spurred interest in the development of small-molecule Sirtuin modulators. Here, we describe an approach exploiting the structures available for four human Sirtuins for the development of isoform-specific inhibitors. Virtual docking of a compound library into the peptide binding pockets of crystal structures of Sirt2, 3, 5 and 6 yielded compounds potentially discriminating between these isoforms. Further characterization in activity assays revealed several inhibitory compounds with little isoform specificity, but also two compounds with micromolar potency and high specificity for Sirt2. Structure comparison and the predicted, shared binding mode of the Sirt2-specific compounds indicate a pocket extending from the peptide-binding groove as target side enabling isoform specificity. Our family-wide structure-based approach thus identified potent, Sirt2-specific inhibitors as well as lead structures and a target site for the development of compounds specific for other Sirtuin isoform, constituting an important step toward the identification of a complete panel of isoform-specific Sirtuin inhibitors.

Sirtuin isoforms contribute to various key aspects of metabolic regulation, disease pathologies, and aging [1,21]. They are thus considered attractive therapeutic targets for diseases such as cancer and neurodegenerative disorders [22,23], which has spurred interest in the mechanisms of Sirtuin catalysis and regulation and in small-molecule regulators for in vivo studies and therapy [22]. Inhibition of Sirt1 was shown to sensitize cells for DNA-damaging cancer therapeutics [24], and inhibition of Sirt1 and Sirt2 can itself decrease tumor growth [25,26]. A variety of Sirtuin activating and inhibiting small molecules has thus been described [22,23]. However, most of these compounds show limited potency, and their isoform specificity is often low or has not been tested. The widely used inhibitor sirtinol (1; Figure 1), for example, has an IC 50 of 38 μM against Sirt2 in an in vitro assay, shows only ~3-fold weaker potency against Sirt1, and no data have been reported for its effect on other isoforms [23,27,28]. For Sirt1, EX-527 (2; Figure 1) was described as potent inhibitor with an IC 50 of ~0.1 μM, and about two orders of magnitude lower potency against Sirt2 and Sirt3 and no effect against Sirt5, whereas no data are available for Sirt4, 6, and 7 [29]. Several more Sirtuin inhibitors have been described, but most of them resemble sirtinol, with reported IC 50 in the higher μM range, comparable potencies against several isoforms, and no data for other isoforms [23,30].
Crystal structures of the catalytic cores of bacterial and yeast Sirtuins as well as of mammalian Sirt2, 3, 5, and 6 reveal a conserved overall structure [31]. They contain a large Rossmann fold domain and a small, structurally more variable Zn 2+ -binding domain. The substrates, NAD + and the acetyllysine side chain, enter the active site from opposite sides of a cleft between these do-mains, and the acetyl group then appears to be transferred via a 1'-O-alkylamidate reaction intermediate [4]. For several Sirtuin inhibitors, the lack of pronounced isoform specificity might be due to their potential binding to the pocket for the NAD + cosubstrate common to all Sirtuin isoforms. Sirtuins have different protein targets, however, even if they are colocalized, like Sirt3 and 5 in mitochondria [13]. Although they show no strict sequence specificity, Sirtuins display residue preferences around the deacetylation site [32][33][34], and the polypeptide binding pocket thus should enable isoforms-specific contacts for inhibition. A mechanism-based, peptide-derived inhibitor indeed showed an IC 50 of 4 μM for Sirt1, and ~17-fold and >77-fold lower potency against Sirt2 and Sirt3, respectively [35], indicating the peptide binding pocket as a promising target site. Interaction details with this and other inhibitors remain to be resolved, however, as the only inhibitor complex structure (other than complexes with non-specific NAD + analogues) is the Sirt5 complex with suramin, a non-specific Sirt1/2 inhibitor partially occupying the NAD + and peptide binding pockets [36]. (2) are known Sirtuin inhibitors. 1 shows low potency and limited discrimination between Sirt1 and Sirt2. 2 is a potent Sirt1 inhibitor, shows much lower potency against Sirt2 and Sirt3, and has no effect on Sirt5, but data for other isoforms are lacking. The novel compounds 3 and 4 are potent Sirt2 inhibitors and show only weak effects on Sirt1, 3, 5, and 6 (see text).
www.impactaging.com Despite of the limited structural information for Sirtuin/inhibitor complexes, more and more structures of different Sirtuin isoforms reveal their subtle differences. Here, we describe a structure-based approach for identifying novel, isoform-specific inhibitors for human Sirtuins. Using crystal structures of human Sirt2, 3, 5, and 6, we identified potential ligands for the peptide binding grove through a docking screen with a small molecule library. Characterization of the docking hits in in vitro assays reveal two potent, Sirt2specific compounds as well as a target site apparently enabling isoform specificity and additional compound scaffolds for further development, demonstrating the power of this approach for the development of specific Sirtuin inhibitors.

Identification of candidate compounds through a docking screen
Despite the physiological and therapeutic importance of Sirtuins [22], there is a paucity of potent, isoformspecific inhibitors [23,30]. For the identification of novel Sirtuin inhibitor classes, we used the available crystal structures of human Sirt2 (PDB entry 1J8F) [37], Sirt3 (3GLS) [38], Sirt5 (2NYR) [36], and Sirt6 (3K35)[39] in docking screens. To avoid compounds blocking the NAD + binding site, which is similar in different Sirtuin isoforms, we used complexes of the four Sirtuins with the NAD + -fragment ADP-ribose as receptor structures. Complexes were generated by transferring ADP-ribose from the experimental Sirt6/ADP-ribose structure to the other structures based on a superposition of the enzymes. The vacant space in the docking target site, defined as a cube centered around C1' of the ADP-ribose, thus corresponds to the pocket for recognition of the isoform-specific protein substrates and thus should allow isoforms-specific contacts.
For identifying potential ligands, we then docked the 1990 structurally diverse compounds of the National Cancer Institute (NCI) diversity set (http://www.dtp.nci.nih.gov) to the target sites of the four Sirtuins/NAD + complexes. Despite some overlap between the hit lists, most of the compounds on top of the list for each isoform differed from the other isoforms (Table 1, 2; Supplementary Table 1). This result suggests that the chosen receptor sites indeed posses isoforms-specific structural features pronounced enough to be distinguishable for a docking approach, and thus for the identification of isoform-discriminating ligands.
In vitro testing of docking hits reveals non-specific, semi-specific, and Sirt2-specific inhibitors To evaluate the inhibitory potency and specificity of the identified potential ligands, we selected the top ten to seventeen hits of all four docking runs, resulting in 40 different compounds after removing NCS compounds 13987, 95609, 371878, and 343227, which we previously found to be incompatible with our assay. These compounds were then tested in in vitro activity assays for their effects against each of the four Sirtuin isoforms, Sirt2, 3, 5, and 6 ( Figure 2a, Table 2). The experiments showed that two compounds were incompatible with the assay, and yielded varying types of results for the remaining 38 compounds. In total, 20 compounds showed significant inhibitory effects (>25 % loss of activity) on one or several isoforms at the 100 μM compound concentration used, whereas 18 compounds had no such effect ( Table 3). The high hit rate of 53 % for the general ability to inhibit Sirtuins is comparable to hit rates in other library docking screens [40,41] and indicates a successful enrichment of true ligands on top of the hit lists. Of the 20 inhibitory compounds, six (30 % of the inhibitory ones) inhibited more than one isoform; five of the six (25 %) were "semi-specific", i.e. inhibited more than one but not all isoforms tested, whereas one compound (CSC1) behaved as a "broad-band" Sirtuin inhibitor that inhibited all isoforms. Surprisingly, all 14 compounds with a significant inhibitory effect against only one isoform were selective for Sirt2, yielding a panel of potential lead structures for the development of Sirt2specific inhibitors. For Sirt3, 5, and 6, for which no specific modulators have yet been described, the five "semi-specific" inhibitors could still be interesting as leads (see below), especially the three compounds that just affect two isoforms.
It is noteworthy that 3 and 4 share considerable portions of their scaffolds ( Figure 1). They both feature a steroid moiety, but modified at opposite ends; 3 is an estradiol with a bulky substituent at the 17-hydroxyl group, whereas 4 contains a steroid scaffold with a 2-benzyl pyrimidine fused to the A ring. Thus, the steroid scaffold is an attractive lead structure for the development at least of Sirt2-specific compounds, and even larger modifications are possible, which should enable to avoid signifycant side activity against steroid receptors (see below).

Proposed binding orientation and molecular determinants for specificity
In order to gain insights into potential molecular determinants of compound specificity and starting points for compound variations in future development efforts, we analyzed the docking positions and orientations of 3 and 4 and differences in the respective binding sites between the Sirtuin isoforms ( Figure  3a,b,c). In the model of the Sirt2 complex with 3 ( Figure 3a), hydrophobic interactions are observed between the estradiol ring system and several side chains (Phe96, Leu107, Phe119 and Ile169) of a large, hydrophobic cavity (formed by the interacting residues and Ala85, Tyr104, and Ile118) extending into the small Sirtuin Zn 2+ -domain. In addition, the hydroxyl-group of 3 forms hydrogen bonds to the side chain of Asn168, the backbone of Gln167, and the α-phosphate-group of ADP-ribose.
For the Sirt2 complex with 4 ( Figure 3b), two orientations were predicted with similar frequencies and binding energies. Orientation 1 shows only hydrophobic interactions between the sterol ring system and the hydrophobic pocket around Phe119 described above, whereas orientation 2 covers the same site, allowing less optimized hydrophobic interactions but in addition hydrogen bridges to Asn168 and Gln167 as observed in the model of the Sirt2 complex with 3.
A comparison of the corresponding pockets in the four structurally characterized Sirtuin isoforms reveals that Sirt3, 5, and 6 have narrower pockets than Sirt2 ( Figure  3c). In Sirt3, the hydrophobic residues are moved toward each other, closing the binding pocket compared to Sirt2. In Sirt5 and 6, there are bulkier residues (in both cases two Trp) lining the hydrophobic pocket, again making it smaller, and the protein backbone is partly moved into the pocket either from top-left (Sirt5) or bottom-right (Sirt6). These differences likely contribute to the observation in our screen, as well as from analysis of available compounds, that specific inhibition seems easier to achieve for Sirt2 than for Sirt3, 5, and 6. It suggests that for specific inhibition of the latter isoforms, smaller scaffolds should be favored (see below). For improvement of Sirt2 inhibitors, the docking model suggests that the hydrophobic substituent at position 17 of the sterol scaffold of 3 should be further extended. Such additional groups should contain polar functions, which could interact with polar protein groups or solvent at the pocket entrance and which would improve the solubility and possibly other properties of this compound.   (8) is an established inhibitor for yeast Sir2, and HR73 (9) and the tetracyclic pyrimidinedione 10 for mammalian Sirt1. 11, 12, and 13 are identified here as Sirt2-specific inhibitors, which form a series of structurally related compounds yet show varying inhibition potencies.
www.impactaging.com 2-hydroxy -2-((3-oxoestr-4-en-17-yl) oxy) -1H-indene-1,3(2H)-dione (CSC16, 5) is also a steroid-based compound ( Table 2), but despite of its similarity, especially to 3, it had no effect on any of the tested Sirtuin enzymes. According to the structures and docking models, the binding pocket is too narrow for the compound in Sirt3 and 5, so that it was docked with low binding energy to the surface of the respective Sirtuin. In Sirt2 and 6, 5 could be docked in the hydrophobic pocket, but in both cases in a disadvantageous position, with carbonyl-and hydroxyloxygens of the compound near hydrophobic patches.

Semi-specific compounds as inhibitor scaffolds and as lead compounds
It is noticeable that for all compounds of our screen with an inhibitory effect, Sirt2 is either the only isoform affected or belongs to the group of isoforms that is inhibited. Most likely, this result is due to the larger hydrophobic binding pocket identified from analysis of the docking models (see above). However, compounds that inhibit Sirt2 and another isoform might still be developable into compounds specific for the other isoforms. The very large, hydrophobic compounds CSC1, CSC6, CSC14, and CSC27 (Table 2) likely cover non-specifically many large cavities and are unlikely to yield good pharmacological compounds, but CSC10 (6) and CSC21 (7; Table 2) appear promising. Due to its size, howeer, the essential parts of 7 should be determined in a structure-activity study before further development. 6, in contrast, resembles compounds such as splitomicin (8), HR73 (9), and 10 ( Figure 4), which were reported as micromolar inhibitors for a Sirtuin (8 for yeast Sir2 [42], 9 and 10 for human Sirt1 [26,43]), but which were not tested against most mammalian isoforms. No structural data are available for Sirtuin complexes with any of these compounds, but our structure comparison (see above) suggests that smaller scaffolds might enable these compounds to exploit the site where 3 was docked, which appears not to be accessible for bulkier compounds in Sirt3, 5, and 6. These scaffolds should thus be considered for the development of the first inhibitors specific for these Sirtuin isoforms. Considering that 6 already shows different potencies against the isoforms tested here (Figure 2), its phenanthrene and isochinoline moieties as well as similar groups should be further evaluated for specific Sirtuin inhibition.

Activity of the novel Sirt2 inhibitors in a physiological system
Our novel compounds 3 and 4 are potent and specific Sirt2 inhibitors in vitro, and we next tested the effect of 4 (compound 3 is likely to be sensitive to esterase activity in lysates) on the Sirt2-dependent deacetylation of an in vivo substrate in a physiological environment. The acetylation levels of the Sirt2-deacetylation site Lys40 in α-tubulin [12] were analyzed in HEK cell lysates by using an acetyl-Lys40 specific antibody (Figure 4a). Adding Sirt2 strongly decreased the level of tubulin Lys40 acetylation in the lysate when no inhibitor was present, and the deacetylation effect was decreased in presence of 100 μM compound 4. Thus, compound 4 is an inhibitor for the physiological αtubulin deacetylation activity of Sirt2.
We further examined the likely suitability of 3 and 4 for in vivo use by applying analyses comparable to Lipinski's rule of five (using the molinspiration server at www.molinspiration.com). Both compounds appear generally suitable as pharmacological compounds, except for their lack of sufficient polar groups, leading to an unfavourable partitition coefficient. Compound 3 is further predicted to likely act as a ligand for nuclear receptors. In fact, the sterol ester 3 might be hydrolyzed in a physiological environment and thus generate a generic steroid receptor ligand. Compound 4, instead, was not predicted to bind to nuclear receptors, most likely due to the larger deviation from the steroid scaffold. Testing on HEK cells revealed very high cytotoxicity for 3 (compound 4 showed a smaller effect but possibly due to partial precipitation under these conditions), but it remains to be clarified whether the corresponding mechanism is based on Sirt2 inhibition. However, 3 and 4 are potent and isoforms-specific Sirt2 inhibitors, and especially 4 appears promising for the development of side effect-free Sirt2 inhibitors, with adding suitable polar groups to this scaffold being a first and obvious optimization step.

DISCUSSION
The roles of Sirtuins in central physiological processes and as drug targets have led to great demand for specific inhibitors for research and therapy [22,23]. Available compounds often feature limited or unknown specificities and mostly high micromolar potencies, and surprisingly little structural information is available for Sirtuin/inhibitor complexes, which could be used for rational improvement. However, we show here that the increasing number of Sirtuin isoform structures in noninhibited state [31] allow the structure-based identification of novel, isoform-specific inhibitor classes. A previous docking study on Sirt2 [44], the only structurally characterized mammalian isoform at that time, yielded only two Sirt2 inhibitors with IC 50 values below 100 μM (57 and 74 μM, respectively), and their isoform specificities were not evaluated. The lower hit rate and affinities than in our study might have www.impactaging.com several reasons. Tervo et al. used a structure as docking receptor that had undergone a molecular dynamics (MD) simulation, resulting in major deviations from the experimental structure, which might not well represent the major conformations of the protein. To take protein flexibility into account, using several structuresrepresenting different conformers -simultaneously, or simulating flexible side-chains during docking are now often used approaches [45]. The first option is not yet possible for most Sirtuins due to the lack of such multiple structures, and it appears that ignoring receptor flexibility is in fact a viable strategy for identifying Sirtuin inhibitors, at least for Sirt2. To take side chain flexibility into account, however, might be an approach for identifying compounds specific for the other isoforms studied here, Sirt3, 5, and 6 [40]. A further difference to our approach was that several potential inhibitors were not considered for experimental testing due to missing features assumed to be important for inhibition, which might have removed potent compounds. Also, a different docking software (GOLD) and compound database were used (Maybridge). The Mabridge database, being much larger than the NCI diversity set used here, is unlikely to be a bottleneck, but it has been observed repeatedly that docking programs differ in their performance depending on the interaction type, e.g. small versus large and hydrophobicversus hydrophilic ligands [40]. Binding of the most potent inhibitors identified appears to be dominated by hydrophobic interactions with a large cavity, which was previously observed not to give best results with GOLD [40]. Finally, we used a receptor with partially occupied NAD + binding site. Thereby, we tried to avoid to obtain compounds binding to the NAD + binding site present in similar form in all Sirtuins, as well as other NAD-dependent enzymes, likely resulting in little specificity. This approach further takes into account that interactions with the bound NAD + can contribute to inhibitor affinity. The docking models obtained indeed indicate that NAD + forms part of the bottom of the pocket likely occupied by the inhibitors. However, instead of mainly occupying the peptide binding cleft, the compounds were preferentially docked perpendicular to this cleft, extending into a hydrophobic pocket in the Zn 2+ domain. If this binding orientation is confirmed by structural studies, then fusing to compounds such as 6 smaller substituents exploiting the isoforms specific features of the peptide binding grooves might be an attractive approach for further improvement of inhibitor potency and specificity. Even though the NCI diversity set is supposed to represent a varying scaffold repertoire, it contains a number of series of structurally related compounds.
Interestingly, a series of related combinations of benzyl and naphtyl systems connected through a methylen and a methanimine bridge was identified in our screen (CSC22-24, Figure 4), and showed differences in the effects on the four Sirtuins tested (Figure 2a). All compounds inhibited only Sirt2, but CSC23 (11) had a larger effect than CSC22 (12) or CSC24 (13). They show a certain similarity to 1 (Sirtinol [28]; Figure 1) and the closely related Salermide [46], established Sirtuin inhibitors with some preference for Sirt2 over Sirt1 [23]. The observed differences suggest that further evaluating such combinations of three linked ring systems might have potential for further development of specific and potent inhibitors. Most promising for Sirt2 inhibition, however, appear the steroid-based compounds 3 and 4, which should be further developed to overcome some disadvantages, such as limited polarity and solubility. Further, 3 should be sensitive to cellular esterase activities and is likely to act as a ligand for nuclear receptors, whereas 4 shows larger deviation from the steroid scaffold and thus might be the more promising lead for the development of side effect-free Sirtuin inhibitors. Several known Sirtuin inhibitors also comprise polycylic scaffolds, such as the Sirt1-over-Sirt2-selective pyrimidinedione 6 [26] and Splitomicin (7; [42]), which slightly resemble the A-C ring system of steroids and thus might exploit the same binding site. Attractive approaches for further development of these known and our novel inhibitors would be their rational modification, but the lack of structural information for these and most other Sirtuin inhibitors hinders such approaches. The putative binding pocket identified here, however, might already allow identifying improved Sirtuin inhibitors by screening of larger virtual libraries with more drug-like compounds, even before the anxiously awaited structural information on Sirtuin inhibitor complexes become available.
Improved compounds, derived from our Sirt2 inhibitors or identified through docking more drug-like compounds to their putative binding site, would be helpful for functional studies and promising for therapy. Acetylation of tubulin, a known Sirt2 deacetylation substrate [12], increases the stability of microtubule, and a reduction of tubulin acetylation in neurons has been reported for individuals with Alzheimer's disease [47]. Sirt2 inhibition could thus be an approach for therapy of this and other neurodegenerative diseases, such as Parkinson's disease, where Sirt2 inhibition was shown to prevent death of dopaminergic cells [48]. Sirt2 further contributes to regulation of cell cycle progression [49], and although details of its function www.impactaging.com seem to vary in different cancers, inhibition of Sirt2 seems to be a viable approach for destroying tumor cells [25,26]. The other isoforms studied here might also be interesting drug targets, such as the mitochondrial forms Sirt3 and 5 for metabolic and aging-related diseases due to their role in regulating metabolic enzymes and stressresponses [21,50]. Our data indicate that compounds with smaller ring systems, such as 6, might better allow exploiting the tighter pockets of these isoforms corresponding to the predicted binding site for 3 and 4. The binding site, however, is so far speculative, and the tighter conformation crystallized might just be one of several in solution, so that more information, such as Sirtuin/inhibitor complex structures or extensive structure-activity-relationship studies will be necessary for clear conclusions. Table 3. Effects of the highest docking hits on the activity of Sirt2, 3, 5, and 6  Sirt3 was expressed in E. coli and purified via Ni-NTA and size exclusion chromatography as described elsewhere [18], and Sirt5 according to the protocol of Schuetz et al. [36]. Sirt1 was expressed and purified as described for Sirt3, except that after affinity purification the protein was passed through a Superdex200 gel filtration column (GE Healthcare, New Jersy, USA) equilibrated in Buffer A (25 mM HEPES, pH 7.5, 100 mM KCl, 2 mM DTT), followed by HiTrapQ (GE Healthcare, Waukesha, WI, USA) anion exchange chromatography. The protein was bound to the HiTrapQ column in buffer A and eluted in a gradient to buffer A supplemented with 400 mM KCl. Fractions were analyzed by SDS-PAGE, and the Sirt1 fractions were pooled and transferred into buffer A using a NAP column (GE Healthcare). supplemented with protease inhibitors cocktail. Cell debris was removed by centrifugation (10,000g, 10 minutes) at 4°C, and 50µl reaction were prepared with identical amounts of cell lysate and either 100 μM compound 4, 1 μg Sirt2, or both. All samples further contained the same amount of DMSO (to account for the solvent of the stock of 4) and 25 mM HEPES (pH 7.5; to compensate for the solvent of the NAD + stock). 20 µg total protein was then loaded on a gel for blotting with the monoclonal anti-acetylated tubulin antibody (Sigma), and 10 µg total protein for the blot with anti-tubulin antibody (Epitomics, Burlingame, CA, USA). Proteins were blotted to polyvinylidenfluorid membrane, incubated with primary (2 h) and then secondary antibodies at room temperature(1h; anti-Mouse IR Dye 680LT and goat anti-Rabbit IR Dye 680LT (Li-Cor Biosciences, Lincoln, NE, USA), respectively), and the blots were scanned in the 700 nm channel using an Odyssey imaging system (Li-Cor).