Structural basis for the activation and inhibition of Sirtuin 6 by quercetin and its derivatives

Mammalian Sirtuin 6 (Sirt6) is an NAD+-dependent protein deacylase regulating metabolism and chromatin homeostasis. Sirt6 activation protects against metabolic and aging-related diseases, and Sirt6 inhibition is considered a cancer therapy. Available Sirt6 modulators show insufficient potency and specificity, and even partially contradictory Sirt6 effects were reported for the plant flavone quercetin. To understand Sirt6 modulation by quercetin-based compounds, we analysed their binding and activity effects on Sirt6 and other Sirtuin isoforms and solved crystal structures of compound complexes with Sirt6 and Sirt2. We find that quercetin activates Sirt6 via the isoform-specific binding site for pyrrolo[1,2-a]quinoxalines. Its inhibitory effect on other isoforms is based on an alternative binding site at the active site entrance. Based on these insights, we identified isoquercetin as a ligand that can discriminate both sites and thus activates Sirt6 with increased specificity. Furthermore, we find that quercetin derivatives that inhibit rather than activate Sirt6 exploit the same general Sirt6 binding site as the activators, identifying it as a versatile allosteric site for Sirt6 modulation. Our results thus provide a structural basis for Sirtuin effects of quercetin-related compounds and helpful insights for Sirt6-targeted drug development.


Results
Quercetin and its derivative luteolin activate Sirt6-dependent deacetylation. To clarify the effects of quercetin and its derivatives on Sirt6 deacetylation activity, we tested them in several assays. In the FdL assay, quercetin appeared to inhibit Sirt6 but control reactions showed that it quenches FdL fluorescence, preventing reliable measurements ( Supplementary Fig. 1a). In a coupled enzymatic assay with acetylated H3K9 peptide (H3K9ac), which represents a physiological Sirt6 deacetylation site, quercetin also appeared to inhibit Sirt6 but control reactions revealed suppressive quercetin effects on the downstream enzymes of the assay ( Supplementary  Fig. 1b). To overcome the artifacts in these widely used assays, we tested the compounds in a robust MS-based assay 27 with the H3K9Ac substrate. Quercetin and luteolin (5; Fig. 1a) yielded a dose-dependent increase in Sirt6 activity, with a > 2-fold maximum stimulation and EC 50 values of ~1.2 mM (Fig. 1b). We did not observe any evidence for the reported Sirt6 inhibition by lower quercetin concentrations 15,22 and conclude that quercetin and luteolin act as low potency activators of Sirt6 peptide deacetylation activity.
Acetyl-peptides are poor Sirt6 substrates and may not fully mimic physiological deacetylations in cells 12,28 . We therefore examined quercetin effects on the deacetylation activity of Sirt6 on purified full-length histones and nucleosomes (Fig. 1c). Quercetin substantially activated Sirt6 in deacetylating nucleosomal H3K18ac (H3K9ac levels were below the detection limit). Moreover, quercetin also augmented the H3K18ac and H3K9ac deacetylase activities of Sirt6 on free full-length histones (Fig. 1c), which are otherwise poor substrates for Sirt6. Quercetin thus activates Sirt6-dependent deacetylation of its physiological histone substrates.
A Sirt6/quercetin crystal structure reveals binding site and interaction details. To identify Sirt6's quercetin binding site and interaction details, we determined a crystal structure of a Sirt6/ADP-ribose/quercetin complex at 1.84 Å resolution (Table 1; Fig. 1d,e). Well-defined electron density for the quercetin ligand revealed the distal end of Sirt6's extended acyl binding channel as binding site (Fig. 1d-f). The catechol moiety of quercetin (ring B) is buried in a protein pocket and its 4'-hydroxyl group forms a hydrogen bond to the Pro62 backbone oxygen, and the 4′-and 3'-hydroxyls form water-mediated interactions with the backbone of Ala53 and Ile61 and the side chain of Asp116 (Fig. 1e,f; Supplementary Fig. 1c). Asp116 normally forms a hydrogen bond to the NAM moiety of NAD + bound in the so-called "C-site", illustrating that catechol pocket and C-site partly overlap. The quercetin chromen-4-one system contributes to complex formation through hydrophobic contacts with funnel surface patches formed by Phe64/82/86 and Val70/115, and Met136/157 (Fig. 1e,f). Comparison of the Sirt6/ quercetin complex with a Sirt6 complex with the pyrrolo[1,2-a]quinoxaline-based activator UBCS039 shows that they share most of their binding sites (Fig. 1f). The catechol group of quercetin overlays well with the UBCS039 Twinning was detected through L-tests with CCP4 POINTLESS and twin fractions were determined during amplitude-based twin refinement with Refmac.
pyridine moiety and reproduces its key interaction to Pro62 17 . The chromen-4-one substitutes for the hydrophobic pyrrolo[1,2-a]quinoxalines of UBCS039-related compounds, which can vary between UBCS039 derivatives due to the non-directed nature of its interactions and the wide and hydrophobic architecture of the Sirt6 funnel 17 . It thus appears that the buried catechol group of quercetin functions as an anchor in Sirt6 binding, while the chromen-4-one provides smaller and non-specific binding contributions and might be particularly amenable for modifications to increase compound affinity. Since quercetin occupies the distal end of the extended Sirt6 acyl channel it is compatible with binding of acetylated substrates but would overlap with longer substrate acylations such as myristoylations ( Supplementary  Fig. 1d). Testing quercetin on Sirt6-dependent demyristoylation indeed revealed a concentration-dependent inhibition (Fig. 1g), consistent with results from continuous Sirt6 assays with a quencher-containing long chain acyl substrate 29 . Thus, quercetin stimulates the deacetylation activity of Sirt6 and inhibits Sirt6-dependent demyristoylations, as observed for UBCS039 and consistent with their binding to the remote end of the Sirt6 acyl channel.
Sirt6 complex structures with activating and inhibitory quercetin derivatives reveal interaction differences and suggest modulation mechanisms. Cyanidin (6; Fig. 1a) is an quercetin derivative that was reported to activate Sirt6 with slightly increased potency and efficacy (EC 50 460 ± 20 µM; 55-fold stimulation) 23 . In our activity assays, cyanidin caused precipitations at higher concentrations, preventing reliable measurements. It showed already significant Sirt6 activation at lower concentrations (20-80 μM; Fig. 2a), however, indeed indicating a higher potency than for quercetin. To obtain insights in activation mechanism and compound features relevant for the improved potency, we solved a crystal structure of a Sirt6/ADP-ribose/cyanidin complex (Table 1; Fig. 2b; Supplementary Fig. 2a). The activator is well-defined by electron density for its B-ring catechol, but features weaker density for the A and C ring system, which could have functional relevance but might also be caused by the lower solubility (and resulting occupancy) of cyanidin. The A/C rings nevertheless can be positioned unequivocally, in the same sites as for quercetin (Fig. 2c), and the compounds thus show the same binding mode. www.nature.com/scientificreports www.nature.com/scientificreports/ Catechin gallate (CG) is the most potent quercetin derivative reported to inhibit rather than activate Sirt6-dependent deacetylation (IC 50 2.5 ± 0.1 μM) 23 . Testing CG in our MS assay confirmed the surprising inhibitory effect of this closely quercetin-related compound, albeit with slightly lower potency (IC 50 80 ± 15 μM; Fig. 2d). To rationalize the inhibitory effect, we determined a crystal structure of a Sirt6/CG complex at 2.0 Å resolution (Table 1; Fig. 2e). The bound inhibitor CG is well-defined by electron density and occupies the same binding region as the activator quercetin, with identical positions for their catechol moieties but a rotated chromen-4-one in GC to accommodate its additional, bulky trihydroxy benzoyl group (Fig. 2e,f; Supplementary  Fig. 2b). The chromen-4-one now interacts with Trp71 of the acyl channel exit, and the trihydroxy benzoyl group forms hydrophobic interactions with the other side of the exit funnel and a polar contact to the Gly155 backbone. The inhibitor does not overlap with the binding pocket for acetyl substrate but the catechol group assumes a position partly overlapping with the NAM moiety of the NAD + cosubstrate (Supplementary Fig. 2c). Testing CG inhibition at different NAD + concentrations did not result in significant changes of residual activity ( Supplementary  Fig. 2d), however, indicating that the inhibition is not based on NAD + competition. The fact that closely related compounds with identical catechol binding mode act as activators indeed suggests that the modulatory binding occurs after the NAM release step of the catalytic cycle.
The keto group of quercetin at C-ring position 4 is missing in the more potent activator cyanidin and the potent inhibitor CG, which removes an unfavourable interaction -with Met157 in case of cyanidin -and might be a key for their increased potency. Comparing Sirt6 complexes of activating and inhibiting compounds further suggests modulation mechanisms and differences relevant for their differing effects (Fig. 2g). The inhibitor CG differs from activators in a slight rotation of the catechol moiety, and more drastically in a tilted position of the chroman and the presence of its additional bulky substituent, orienting them toward N-terminus and Val154, respectively. The Val154/Gly155 peptide bond flips to provide space, but this and other differences in protein conformation are subtle. The bulky group pointing toward the N-terminus is CG's dihydrochroman, and its strong tilt compared to quercetin/cyanidin is due to the saturated C-ring in CG. Interestingly, activating quercetin derivatives appear to feature an unsaturated, planar C-ring, while inhibiting family members tend to have saturated C-rings (our data and 15 ) and thus likely bind in the tilted, CG-like orientation. The Sirt6-activating compound UBCS039 17 orients its pyrrolo[1,2-a]quinoxaline similar to the chromene of the activating quercetin derivatives (Fig. 2g), which supports the relevance of this moiety position for activation. Both inhibiting and activating quercetin derivatives (and pyrrolo[1,2-a]quinoxalines) overlap with the C-site that accommodates the NAM moiety of NAD + until NAM is released in the first catalytic step 10 . This binding mode implies for activators that they bind during or after NAM release. Such a mechanism was characterized for the potent Sirtuin inhibitor Ex-527 and also suggested for trichostatin A 24,30 , and it is indicated for the inhibitory quercetin derivatives by the lack of NAD + competition (see above). It thus appears that the quercetin-based activators and inhibitors share this mechanistic feature, consistent with their similarity in chemical structure and binding mode. Different effects on the stability or conformational details of the acyl channel of Sirt6/product complexes, possibly due to the inhibitor's chroman group pointing toward the N-terminus (Fig. 2g), might cause their differing effects on Sirt6 deacetylase activity, but details remain to be established.
Quercetin inhibits other Sirtuin isoforms by exploiting an alternative binding site. Due to the isoform-specific features of the Sirt6 acyl channel, in particular its wide and hydrophobic architecture 10,31 , its ligands should be Sirt6 specific. An overlay of the Sirt6/quercetin complex with Sirt1,2,3, and 5 indeed shows that the binding site is blocked in these other isoforms by cofactor loop and a helix bundle ( Supplementary Fig. 3a). However, effects of quercetin and related polyphenols were also reported for Sirt1 20 . We thus tested the quercetin isoform specificity in activity assays with human Sirt1, Sirt2, Sirt3, Sirt6 (all with acetylated substrate), and Sirt5 (with succinylated substrate). Quercetin caused a concentration-dependent inhibition of all Sirtuin isoforms except for Sirt6, which was instead activated as described above (Fig. 3a). Due to the lack of a Sirt6-like binding site in the other isoforms, quercetin has to exploit an alternative binding site to cause this opposite, inhibitory effect.
To identify the alternative quercetin binding site and modulation mechanism, we determined a crystal structure of a Sirt2/quercetin complex at 2.2 Å resolution (Table 1; Fig. 3b,c). Well-defined electron density revealed that quercetin binds at the Sirt2 active site entrance, through π-π stacking with Tyr-114 and Phe-235 and anion-π interactions with Glu-116 and Glu-120 (Table 1; Fig. 3c; Supplementary Fig. 3b). Interestingly, quercetin is located between two symmetry-related Sirt2 monomers and interacts with the same surface region of the second monomer, albeit with rotated orientation and through fewer contacts ( Supplementary Fig. 3c), suggesting the interaction with the first monomer to be the one relevant in solution. An overlay with a Sirt2/substrate peptide complex reveals that quercetin partly occupies the pocket accommodating substrate moieties C-terminal from the acetyl-Lys and thus sterically prevents substrate binding (Fig. 3d). Such a peptide competitive mechanism is supported by Sirt2 inhibition tests that show that increasing substrate concentrations lead to weaker inhibitory effects of a fixed inhibitor concentration (Fig. 3e).
Comparing the Sirt2/quercetin complex to other Sirtuin isoforms reveals that its binding site is also accessible in Sirt1, 3, and 5, whereas it is occupied in Sirt6 by the N-terminus ( Fig. 3f; Supplementary Fig. 3d). Since the quercetin binding site of Sirt6 is blocked in these other isoforms, our structures define two mutually exclusive modulator binding sites for Sirt6 and for Sirt1,2,3,5, respectively.
Isoquercetin is an activating ligand for the quercetin site with improved specificity. Since the quercetin sites of Sirt6 and Sirt1,2,3,5 are mutually exclusive, suitable derivatives should allow to exploit specific features of either site and thus show improved selectivity. We thus tested the isoform selectivity of the quercetin derivative isoquercetin (Fig. 1a), which features a bulky sugar moiety that might be accommodated in the acyl channel of Sirt6 but not in the alternative site. Isoquercetin indeed retained the Sirt6 stimulating activity, www.nature.com/scientificreports www.nature.com/scientificreports/ albeit with even lower potency, and it showed no significant effects on Sirt1-3-dependent deacetylation and Sirt5-dependent desuccinylation (Fig. 4a,b). To confirm the molecular basis of the improved Sirt6 selectivity of isoquercetin, we determined a crystal structure of a Sirt6/ADP-ribose/isoquercetin complex. Initial soaking experiments failed due to a PEG molecule bound to the acyl channel, but substituting PEG with ethylene glycol enabled us to solve the Sirt6/isoquercetin complex structure at 1.8 Å resolution (Table 1; Fig. 4c). The quercetin moiety of isoquercetin occupies the activator site identical to the parent compound ( Fig. 4d; Supplementary  Fig. 3e), and its additional sugar moiety is thereby placed in the acyl channel. The density for the isoquercetin sugar is weaker than for its quercetin moiety, likely due to flexibility, and it indeed forms no significant positive interactions (Fig. 4c; Supplementary Fig. 3e). Overlaying isoquercetin with the ligand of our Sirt2/quercetin structure illustrates that the sugar moity would clash with Tyr-114, rationalizing the improved selectivity for Sirt6 (Fig. 4e). The sugar at this compound position thus acts as a negative selector, i.e. does not disturb Sirt6 binding significantly but hinders accommodation in the alternative binding site of the other isoforms, indicating that other moieties at this position should result in further improvements in selectivity and -with more appropriate moieties for beneficial interactions -in better potency.

Discussion
Sirt6 is considered as therapeutic target for aging-related diseases 1,3 , but the few available Sirt6 modulators require improvements concerning potency, selectivity, and bioavailability 2,24 . Partly contradictory effects on Sirt6 activity were previously reported for the plant flavone quercetin 15,21,22,26 , and we find that the compound interferes with two popular Sirtuin activity assays, which is a common problem for optical assays 32 . Our robust MS assay confirms quercetin as a low potency Sirt6 activator. Previously reported inhibitory effects at lower quercetin www.nature.com/scientificreports www.nature.com/scientificreports/ concentrations could not be observed and likely are assay artifacts caused by the optical effects of the compound and/or the use of Sirt6 tagged with GST, which is known to bind quercetin tightly 33 . We could confirm, however, that some quercetin derivatives act as Sirt6 inhibitors rather than activators, and the slight differences in potency might again be attributed to the previous use of a GST tag 23 . The more potent quercetin-based Sirt6 modulators 23 , together with our results, suggest that potency and specificity of this scaffold can be significantly increased and improved compounds might eventually even become suitable for pharmacological applications. They further indicate that Sirt6 modulation might contribute to physiological effects of such more potent compound family members. Indeed, treating U2OS cancer cells with the potent inhibitor CG led to significant histone H3 hyperacetylation, an established marker for Sirt6 inhibition, and this effect was abolished in SIRT6-deficient cells (Fig. 4f) 34 . The Sirt6 activating quercetin derivatives showed low potency, and quercetin itself affects more potently a variety of other cellular targets, such as kinases 19 . Sirt6 might still contribute to the beneficial health effects of quercetin, but potentially indirectly, through its activation by increased NAD + levels from quercetin inhibition of CD38 35 . Nonetheless, we found that treatment of U2OS cells with 25 to 100 μM quercetin induced a dose-dependent decrease of H3 acetylation that was not observed in SIRT6-deficient cells, demonstrating clear Sirt6-dependent cellular effects of higher quercetin concentrations (Supplementary Fig. 3f). Direct Sirt6 activation thus might contribute to physiological effects observed at higher quercetin concentrations. Importantly, the more potent quercetin derivatives, such as CG, indicate that Sirt6-activating and -inhibiting natural quercetin derivatives with even further increased potency might exist and that Sirt6 modulation might contribute to effects of compound family members and thus of flavone-containing foods or extracts. Due to the wide Sirt6 binding side and few specific interactions, even other natural compound families with a catechol group or a related moiety could exploit this binding mode. They might be the explanation why the deacetylase activity of Sirt6 is weak in vitro yet significant in vivo 7 . Furthermore, quercetin inhibits other Sirtuin isoforms with low potency and more potent derivatives appear to exist also for these isoforms 36 and might contribute to physiological effects of the compound family. www.nature.com/scientificreports www.nature.com/scientificreports/ For exploiting insights from studying quercetin-based Sirtuin modulators for drug development, Sirtuin/compound interaction details are required. A pharmacophore model based on binding competition suggested four relevant hydrogen bonds 26 . Docking models indicated binding in active site and acyl channel, with inhibitors disturbing substrate binding and a more external binding of activators 23 . Both modelings were based on limited and partially incorrect data (see above). Our crystal structures in fact identify a deviating Sirt6 binding mode and provide interaction details and modulation mechanisms. Activators and inhibitors bind almost identical, with the catechol moiety acting as an anchor. It binds at the bottom of the Sirt6-specific acyl channel, similar to the pyridine moiety of activating pyrrolo[1,2-a]quinoxalines 17 . The resulting overlap with the C-site indicates that the modulatory binding takes place only after NAM release 32 . The chroman/chromene moieties contribute mostly non-directional hydrophobic contacts and thus can assume different orientations within the Sirt6 acyl funnel. They are structurally distinct from the corresponding pyrrolo[1,2-a]quinoxalines groups and more polar, revealing that these parts of the two compound families should be amenable to significant modifications that might improve affinity and/or solubility. Within the quercetin family, variations would mainly be restricted by the observation that saturation state and substituents of the chroman/chromene decide its orientation and thereby seem to determine whether a ligand activates or inhibits. It will be interesting to see which other scaffolds, besides pyrrolo[1,2-a]quinoxalines, can imitate either binding mode and possibly facilitate the development of highly potent and soluble allosteric Sirt6 activators and inhibitors, respectively. Furthermore, C-ring modifications in quercetin derivatives, respective corresponding positions in pyrrolo[1,2-a]quinoxalines or potential other ligand classes, appear in particular suitable for increasing Sirt6 selectivity due to their incompatibility with the newly identified quercetin binding site in Sirt1,2,3,5.
Our structures reveal alternative binding sites in Sirt6 respective Sirt1,2,3,5, and thereby explain quercetin's Sirt6-specific activation effect and its weak inhibition of other isoforms. The comparison of quercetin in both sites reveals C-ring modifications, as in isoquercetin, as a feature that can increase Sirt6 selectivity and possibly also affinity, and the Sirt6/CG complex reveals how quercetin derivatives can also exploit this site for potent Sirt6 inhibition. In principle, our structures also provide information for improving the inhibitory binding to other Sirtuin isoforms, but strong and specific binding might be difficult to achieve due to their related, rather flat and exposed sites. In summary, our results hence provide structural and mechanistic insights in Sirtuin modulation by quercetin-based compounds and will support for further development of Sirt6 activators and inhibitors.
Protein production and purification. A comparison of the protein constructs we used is shown in Supplementary Fig. 4. The recombinant N-terminal his-tagged human Sirt6 proteins were produced as previously described 16 . Briefly, Sirt6(1-355) in pQE80L.1 was fermented in E. coli M15[pREP4]; Sirt6(13-308) in pET151-D-TOPO was expressed in E. coli Rosetta2 (DE3) pLysS. Human Sirt2(55-356) was expressed from a pET-SUMO vector in E. coli BL21 (DE3) codon + . The proteins were purified by affinity chromatography with Talon resin (Clontech), followed by tag cleavage with Tobacco Etch Virus (TEV) protease. Tag and protease were removed through a second Talon affinity chromatography, and the proteins were further purified using cation exchange and gel filtration chromatography. Purified protein was concentrated to 10 mg/ml for Sirt6 and 36.5 mg/ ml for Sirt2, flash frozen in liquid nitrogen, and stored at −80 °C. Full length human Sirt1, human Sirt3 residues 118-399, and human Sirt5 residues 34-302 were prepared as described before 16,37 . Peptide deacylation assays. For coupled enzymatic peptide deacylation assays, reactions were run in a total volume of 100 µl containing 50 mM Na-phosphate pH 7.50, 5% DMSO, 0.6 mM DTT, 0.1% (v/v) Tween 20, 200 µM acetylated histone H3K9 peptide, 500 µM NAD + and 10 µM Sirt6. The reactions were monitored in an Epoch 2 plate reader (BioTek) at 340 nm wavelength. Control reactions to check for compound effects on downstream enzymes contained no Sirt6 and were spiked with 40 µM nicotinamide.
For FdL assays, reactions were run in a total volume of 50 µl containing 50 mM Tris-HCl pH 7.50, 100 mM NaCl, 5% DMSO, 100 µM acetylated FdL1-peptide, 500 µM NAD + and 10 µM Sirt6. After incubation at 37 °C for 1 h, reactions were stopped by adding 2 mM NAM and 10 mg/ml trypsin, incubated 20 min, and measured in a FluoDia T70 (Photon Technology) at wavelength 460 nm. Control reactions for fluorescence quenching effects of the compounds were run by adding compound at different concentrations after the deacetylation and development steps.