Molecular basis for hierarchical histone de-β-hydroxybutyrylation by SIRT3

Chemical modifications on histones constitute a key mechanism for gene regulation in chromatin context. Recently, histone lysine β-hydroxybutyrylation (Kbhb) was identified as a new form of histone acylation that connects starvation-responsive metabolism to epigenetic regulation. Sirtuins are a family of NAD+-dependent deacetylases. Through systematic profiling studies, we show that human SIRT3 displays class-selective histone de-β-hydroxybutyrylase activities with preference for H3 K4, K9, K18, K23, K27, and H4K16, but not for H4 K5, K8, K12, which distinguishes it from the Zn-dependent HDACs. Structural studies revealed a hydrogen bond-lined hydrophobic pocket favored for the S-form Kbhb recognition and catalysis. β-backbone but not side chain-mediated interactions around Kbhb dominate sequence motif recognition, explaining the broad site-specificity of SIRT3. The observed class-selectivity of SIRT3 is due to an entropically unfavorable barrier associated with the glycine-flanking motif that the histone Kbhb resides in. Collectively, we reveal the molecular basis for class-selective histone de-β-hydroxybutyrylation by SIRT3, shedding lights on the function of sirtuins in Kbhb biology through hierarchical deacylation.

Histone lysine β-hydroxybutyrylation (Kbhb) has been detected in yeast, fly, mouse and human, and 44 Kbhb sites have been identified in human and mouse cells 15 . β-hydroxybutyrylation modified on the ε-group of lysine distinguishes itself from acetylation by its branch, chiral, and four-carbon length properties. The levels of histone Kbhb are significantly elevated under the conditions of starvation or streptozotocin-induced diabetic ketosis. It has been proposed that histone Kbhb directly connects ketone body metabolism to gene regulation, given the high concentration of β-hydroxybutyrate in blood during fasting, starvation, or prolonged intense exercise 15,18 . Importantly, histone H3K9bhb is associated with gene upregulation in a starvation-responsive manner, and distinguishes an amount of these genes from others that are marked by H3K9ac and H3K4me3, suggesting a unique role in connecting epigenetic regulation and starvationresponsive metabolism 15 . To further elucidate Kbhb functions in gene regulation, its cognate eraser(s) awaits to be characterized.
The sirtuin family proteins (e.g., human SIRT1-7), a class of NAD + -dependent deacetylases that remove acetyl marks from various cellular proteins including histones, function in a wide range of biological pathways in responses to nutritional and environmental perturbations 19 . Activation of sirtuins increases the lifespan of several model organisms, while mutation or inhibition of sirtuins leads to the onset of aging phenotypes 20,21 . Recently, emerging evidence suggested that sirtuins could remove acyl groups other than acetyl from lysine residues 22 . For example, SIRT1-3 can remove crotonyl marks from lysine 23 , while SIRT4 is reported to have deglutarylation and dehydroxymethyl-glutarylation avtivities 24,25 . SIRT5 removes the malonyl, succinyl, and glutaryl marks 13,26,27 . SIRT6 preferentially hydrolyzes long alkyl acyl marks from lysines, such as myristoyl mark 28 . In summary, sirtuins play an important role in removing various acyl marks from lysines of histone as well as nonhistone substrates.
SIRT3 is a predominant mitochondria matrix protein that modulates the activity of key metabolic enzymes via protein deacetylation 29 . Besides, several studies reported that SIRT3 could act as a histone deacylase both in vitro and in vivo 23,[30][31][32] , suggesting a moonlighting function of SIRT3 in nucleus 33 . Through systematic profiling studies, here we reported that human SIRT3 functions as a histone de-β-hydroxybutyrylase at sites H3 K4, K9, K18, K23, K27, and H4K16. Interestingly, SIRT3 is incapable of removing bhb at sites H4 K5, K8, K12 that have flanking glycine residues both in vitro and in cells. Such classselectivity is not observed in HDAC3, a key member of the Zn-dependent histone deacetylases. Co-crystal structural analyses of SIRT3 bound to H3K9bhb, H3K4bhb, and H4K16bhb peptides as well as structural-based mutagenesis studies revealed the molecular basis underlying class-selective recognition and erasure of histone Kbhb. Hence, our work on hierarchical histone deacylation by SIRT3 suggests a potential regulatory mechanism that links acylation dynamics to gene regulation under metabolic alternations.

Results
Systematic profiling histone deacylase activities of sirtuins By means of click chemistry followed by affinity purification and competition assays, Bao et al revealed that SIRT3 is a histone decrotonylase and binds to H3K4 crotonylation (H3K4cr) peptide substrate at an affinity of 25.1 μM in the absence of NAD + . Taking histone H3K9 acylation as a probe, we synthesized a panel of histone H3 peptides encompassing residues 1-15 (H3 1-15 ) with lysine 9 bearing acetyl and thirteen other non-acetyl acyl marks ( Fig. 1a). Next, we expressed and purified NAD + -dependent deacetylase domains of all human sirtuins as well as full-length bacteria CobB that serves as a prototype sirtuin control ( Fig. 1b and Supplementary Fig. S1). We then performed isothermal titration calorimetry (ITC) to profile interactions between the well-behaved sirtuin proteins (SIRT1-3, 5, 6, and CobB) and histone H3K9 acylation peptides in the absence of NAD + . The yields of SIRT4 and SIRT7 were low and thereby were not used for ITC titration.
On the binding affinity heatmap of ninety interaction pairs, each sirtuin family member displays unique H3K9 acylation binding feature ( Fig. 1c and Supplementary Fig.  S2; Table S1). Previously reported specific deacylase activities of sirtuins are all corroborated by micromolar binding affinities in this heatmap, with K D values of 24.9 µM for SIRT3-Kcr, of 5.4 and 4.4 µM for SIRT5-Ksucc and SIRT5-Kglu, and of 12.6 µM for SIRT6-Kmyr ( Fig. 1c) 13,23,26,28 . Most sirtuins bind well to straight-chain acyl marks from 1-carbon Kfo to 14-carbon Kmyr, except that SIRT5 does not bind Kbu/Kcr, and SIRT6 does not bind Kfo/Kcr. The two acidic acyl marks, namely Kglu and Ksucc, are recognized by SIRT5 and CobB; interestingly, Ksucc but not Kglu is accommodated by SIRT2. The hydroxyl-replaced Khib and Kbhb marks are both recognized by SIRT3, SIRT5, and CobB; however, only Kbhb but not Khib can be recognized by SIRT1 and SIRT2 (Fig.  1c). The diversified acyllysine selectivity reflects delicate design of the active center over a common sirtuin scaffold. It is interesting to note that most sirtuins displayed optimal binding towards non-acetyl acyl marks, notably those with longer alkyl chain ( Fig. 1c and Supplementary  Fig. S2; Table S1).
Next, we measured deacylase activities of all sirtuins including SIRT4 and SIRT7 by mass spectrometry (MS) against eight H3K9 acylation marks: Kac, Kbu, Kcr, Ksucc, Kglu, Khib, Kbhb, and Kmyr. In general, all binding events mentioned above were well translated into deacylase activities ( Fig. 1d and Supplementary Fig. S3). This is consistent with the notion that sirtuin enzymes preferentially bind to acylated peptide first before NAD + loading 34 , and efficient peptide substrate engagement is required for deacylation to occur. Collectively, bacterial CobB is able to deacylate eight H3K9 acylation marks tested but SIRT7 does none of them. SIRT4 and SIRT6 are able to hydrolyze Kac, Kbu, and Kmyr but not the acidic Ksucc/Kglu, the hydroxyl-replaced Khib/Kbhb, and the rigidified Kcr marks. In contrast, SIRT5 is able to remove most marks except for Kcr. SIRT1-3 behave similarly and are sensitive to acidic Ksucc/Kglu, except for SIRT2-Ksucc; additionally, SIRT1 and SIRT2 cannot hydrolyze Khib, while SIRT3 displays an deacylase activity towards both Khib and Kbhb (Fig. 1d).
Overall structure of SIRT3 bound to H3K9bhb peptide To decipher the structural basis underlying H3K9bhb recognition by SIRT3, we determined the binary crystal structure of SIRT3 118-399 bound to the H3 6-15 K9bhb peptide in an NAD + free state at 1.95 Å resolution (Supplementary Table S2). There is one SIRT3 monomer in one crystallographic asymmetric unit. Based on the electron density, we could model residues 122-395 of SIRT3 and trace the " 7 ARKSTGG 13 " segment of the H3 peptide (Fig. 3a). SIRT3 adopts a classical sirtuin fold that consists of a zinc-binding small lobe and a NAD + -binding large lobe arranged in a Rossmann fold (Fig. 3b). The histone peptide (and NAD + ) binds to a cleft formed at the interface of the two lobes. Structural alignment of the binary complex with an apo form SIRT3 (PDB: 3GLS) revealed induced bending of the small lobe upon H3K9bhb peptide insertion (Fig. 3c), consistent with previous report 35 . We calculated a rotation angle of 18.6°b ased on the DynDom3D webserver analysis 36 . The peptide-binding surface is largely negatively charged, being electrostatically favorable for the basic histone peptide targeting (Fig. 3d, left). There is a snug fitting of the H3K9bhb peptide into the substrate-binding pocket with Kbhb side chain inserted into an elongated tunnel (Fig. 3d, right). Consistent with the observed preference of SIRT3 for long-chain acylations, extra space exists next to the bhb group, being well positioned to accommodate long-chain acyl marks like Kdod and Kdec (Fig. 1c).
The Kbhb mark is inserted into an elongated pocket formed by residues F294, V292, F180, I230, Q228, H248, and V324 (Fig. 4c). Besides an effect of charge neutralization of lysine, the bhb moiety of Kbhb affords extra hydrophobicity and hydrogen-bond forming capacity. Compliantly, the Kbhb mark is recognized by both hydrophobic contacts involving residues F294, V324, F180, I230, and hydrogen bonding interactions involving residues Q228, H248, V292 and a molecule of water (Fig.  4c). SIRT3 can deacylate both Kbhb and Kcr marks. As revealed in previous structural studies 23 , the π-π stacking between F180 and Kcr plane as well as hydrophobic contacts contribute to Kcr recognition (Fig. 4dii). The saturated bhb moiety cannot form π-π stacking with F180. As an adaptation, hydrophobic contacts with F180 and I230, as well as direct hydrogen bonds of the β-hydroxyl group with H248 and Q228 contribute to bhbspecific mode of recognition (Fig. 4di).
It has not escaped our attention that both electron density (Fig. 4c) and interaction environment (Fig. 4di) clearly suggest that the S-form enantiomer of bhb is captured in the crystal structure despite the fact that the Kbhb peptide was synthesized as the R/S racemic mixture. In support, ITC titration using newly synthesized stereospecific H3K9bhb peptides revealed a~6.5-fold binding preference for S-bhb (K D = 39.2 μM) over R-bhb (K D = 253.8 μM) (Fig. 4e). Consistently, enzymatic kinetic assays revealed a K m of 39.03 μM for S-bhb and a K m of 90.11 μM for R-bhb, and in total~2-fold catalytic efficiency preference for the S-form (k cat /K m = 66.62 s −1 M −1 ) over the R-form (k cat /K m = 36.62 s −1 M −1 ) enantiomers (Fig. 4f). Our structural analyses revealed that the S-form bhb is concurrently stabilized by hydrogen bonding and hydrophobic contacts. By contrast, in the case of R-form Kbhb, the substrate engagement mode is imperfect since the abovementioned hydrogen bonding interactions and the hydrophobic contacts are not compatible ( Supplementary Fig. S4a).

Catalytic center and mutagenesis-based enzymatic studies
In order to validate the functional importance of the pocket residues, we generated corresponding mutants of SIRT3 118-399 and measured their deacylase activity by RP-HPLC. As shown in Fig. 4g, mutation of the catalytic residue H248 (H248A and H248F) almost completely abrogated the activity, attesting to its essential role in catalysis and/or substrate recognition. F180A, I230A, V292A, F294A, and V324A are also detrimental to the hydrolysis of Kbhb by SIRT3 to various degrees. We designed another mutant, residue V324M, intended for blocking insertion of the Kbhb moiety into its binding pocket. This mutant, retaining only about 15% activity of the wild-type enzyme, is most severe in inhibiting the hydrolysis of Kbhb by SIRT3 besides the H248 mutants (Fig. 4g). The V324M mutant is about 14-fold lower in binding affinity with K9bhb than wild-type SIRT3 (SIRT3 WT -H3K9 bhb ,  Fig. S4b). These results demonstrate the important roles of these residues in recognition and hydrolysis of Kbhb by SIRT3.
Consistently, we observed broad deβ-hydroxybutyrylase activities of SIRT3 on tested histone sites other than H4 K5, K8, and K12 in RP-HPLC based deacylation assays ( In support, H3K14bhb that has two flanking glycine residues from the Nterminus (GGKA) displayed weakest binding affinity ( Fig. 5b and Supplementary Fig. S5a). In comparison to SIRT3, the class I Zn-dependent histone deacetylase HDAC3 showed no class-selectivity on histone H3 and H4 tails (Fig. 5d nd Supplementary Figs. S5b, S7), highlighting a molecular functional distinction between the two subfamilies of histone deacylases.
To test if SIRT3 could deacylate Kbhb at nucleosomal level, we prepared Kbhb modified nucleosomes from bhbtreated HEK293 cell and subjected them for deacylation by SIRT3. Dot blot assays verified the site-selectivity and type-selectivity of corresponding antibodies (Supplementary Fig. S8). After SIRT3 treatment, we immunoblotted Kbhb of the SDS-PAGE resolved histone samples (Fig.   5e). We were able to detect clear deacylation activity of SIRT3 over H3K4bhb, H3K9bhb, H3K14bhb, H3K18bhb, H3K23bhb, and H4K16bhb, but not H4K5bhb, H4K8bhb, and H4K12bhb (Fig. 5f). Importantly, an active site mutant of SIRT3, H248Y, displayed no activity toward Kbhb-modified nucleosomes, suggesting a direct role of SIRT3 in nucleosomal deacylation of Kbhb.
Collectively, the above data demonstrate a classselective de-β-hydroxybutyrylation activity of SIRT3 that distinguishes it from HDACs at both peptide and nucleosome levels in vitro.

SIRT3 regulates histone Kbhb levels in nucleus
SIRT3 is mainly a mitochondria matrix protein that regulates metabolism biochemically via protein deacetylation 29 . To confirm its nuclear presence, we performed fluorescence-based co-localization and cell fractionation studies in HEK293 cells. Z-stack 3D imaging of Cterminally EGFP-labeled SIRT3 revealed clear fluorescence signals in nucleus despite much brighter signals in mitochondria (Fig. 6a, b and Supplementary Fig. S9, Movies S1, S2). Subcellular fractionation assays showed that overexpressed or endogenous SIRT3 could be detected in both mitochondria and nuclear fractions (Fig.  6c-f). Notably,~10% of total SIRT3 was observed in the chromatin-bound extraction. Immunoblotting assays using bhbNa treated cells showed that SIRT3 can be recruited onto chromatin in a dosage-dependent manner, possibly induced by upregulated histone Kbhb levels (Fig.  6g). Interestingly, while the Kbhb signal is enriched on histone H3 under lower and more physiological bhbNa concentrations (10 and 20 mM), the Kbhb levels of H2A/ H2B are preferentially upregulated at a supplemented bhbNa concentration of 50 mM (Fig. 6g), suggesting mechanisms of histone type-specific Kbhb regulation in response to metabolic alternations. Collectively, all the above results suggest a direct association of SIRT3 with chromatin.
We further checked the deacylation activity in cells with overexpressed SIRT3. Consistently, overexpression of wild type but not the H248Y mutant SIRT3 reduced bhb levels at sites H3K4, H3K9, H3K18 and H4K16, but not H4K5, H4K8, H4K12 (Fig. 6h). Interestingly, the levels of H3K14bhb and H3K23bhb are not influenced at cellular level upon SIRT3 overexpression (Fig. 6h), which contradicts with a clear nucleosomal substrate activity in vitro (Fig. 5f), and suggests possible mechanisms of active establishment of H3K14bhb and H3K23bhb patterns in HEK293 cells. Of note, a "reading-and-writing" cross-talk between H3K14 and H3K23 acylations have been implicated in the case of MOZ/MORF (a.k.a. KAT6A/6B) complexes that are responsible for maintaining global histone H3K23ac level in a BRPF1-dependent manner 37 . We have previously shown that the DPF domains of MOZ and DPF2 are specific histone H3K14 acylation readers,

Molecular basis underlying class-selectivity of SIRT3
Next, we aligned all three binary SIRT3 structures reported in this study, and found that four pairs of backbone interactions involving residues E296, G295, and E325 are highly conserved, which serves as a common binding mode and dominates Kbhb sequence motif recognition (Fig. 8a). Dihedral angle analysis revealed that main-chain geometries of the central Kbhb motif (X -i K i X i +1 ) fall in the core β-strand region with proper φ/ψ dihedral angle distribution (Fig. 8a) 39 . Interestingly, those sites that SIRT3 is incapable or inefficient to catalyze are characteristic of a glycine-flanking feature (Fig. 5a). Although the poor tolerance cannot be due to steric clash since glycine is side-chain free, the high degree of rotational freedom endowed by glycine could create an entropically unfavorable barrier for the abovementioned backbone interactions, thereby explaining the observed class-selectivity of SIRT3. In support, calorimetric titrations revealed a high entropy cost of −28.1 cal/mol/deg for H3K14bhb, which is in sharp contrast with an entropy change of −4.2 cal/mol/deg for H3K4bhb and −7.8 cal/ mol/deg for H4K16bhb (Supplementary Table S1). Conceivably, the bending conformations of peptide substrate are pre-stabilized in part by intra-chain hydrogen bonding interactions in the cases of H3K4bhb and H4K16bhb (Fig.  7d, f).
Previous structural studies of HDAC1 (a close paralog of HDAC3) bound to an H4K16ac peptide analog suggested that class I HDACs adopt a deep and narrow pocket for acetyllysine insertion, and the histone peptide is organized in a "turn-staple" conformation around an acidic residue (D99 of HDAC1) for backbone engagement (Fig. 8b) 40 . Both SIRT3 and class I HDACs primarily rely on the substrate backbone for recognition, accounting for their relatively broad site-specificity. In the meantime, the backbone-binding mode of class I HDACs is not β-conformation selective, which is distinct from SIRT3 and therefore explains the observed non class-selectivity of HDAC3.
To further verify our analysis, we synthesized a series of mutant histone peptides by introducing or removing Kbhb-flanking glycine residues around representative sites including H3K4, H3K9, H3K14, H3K18, H4K8, H4K16 and subjected them for binding and enzymatic assays using SIRT3 (Fig. 8c). As summarized in Fig. 8d, single G mutation around H3K4bhb caused reduced binding by 15-fold for T3G (K D = 205.3 μM) and by fivefold for Q5G (K D = 79.4 μM), while "T3GQ5G" double mutation completely disrupted binding. Similarly, the existence of two Kbhb-flanking glycine residues at sites H3K9 (R8G/S10G), H3K14 (A15G), H3K18 (R17G/ Q19G), and H4K16 (A15G/R17G) abolish interaction completely. In further support, reverse "G-to-A" mutation in the case of H4(G7A/G9A)K8bhb restored binding to a K D of 177.6 μM (Fig. 8d). The results of enzymatic assays are consistent with our ITC binding assay results. All "GG" mutant peptides cannot be catalyzed by SIRT3, while "G-to-A" mutation in H4(G7A/G9A)K8bhb peptide rescues its deacylation ability by SIRT3 (Fig. 8e). Moreover, successful restoration of binding and catalysis by "Gto-A" mutation suggests that lack of charge/polarity in the "GG" motifs is unlikely an alternative reason for their being weak SIRT3 substrates.
Overall, our structural, binding and enzymatic studies demonstrate the molecular basis for anti-selective deacylation of glycine-flanking Kbhb sites of histones by SIRT3.

Discussion
A large collection of non-acetyl acylations has been identified on histone lysines, posing new challenges to characterize the cognate regulators and regulatory mechanisms 16,41 . Lysine β-hydroxybutyrylation is unique as it possesses a hydroxyl-group and is chiral and structurally branched in nature (Fig. 1a). Through profiling studies, here we demonstrate that four human sirtuins (SIRT1-3, SIRT5), bacterial sirtuin CobB, and HDAC3 are able to catalyze the hydrolysis of histone Kbhb. Such newly identified de-β-hydroxybutyrylase activities broaden the landscape of histone PTMs that are regulated by sirtuins as well as class I HDACs. Importantly, the observed class-selectivity of SIRT3 but not HDAC3 provides new insights into a regulatory mechanism that histone Kbhb as well as other histone acylations likely involves through hierarchical histone deacylation by sirtuins.
Our structural studies revealed a hydrogen bond-lined hydrophobic pocket of SIRT3 for Kbhb recognition and catalysis. The recognition mode of Kbhb is different from that of Kcr (Fig. 4d). Besides a common hydrogen bond between V292 and the amide NH of Kbhb/Kcr, the bhb group is uniquely stabilized by hydrogen bonding of its β-hydroxyl with Q228 and H248 of SIRT3; additionally, F180 and I230 contribute to critical hydrophobic contacts with the hydrocarbon portion of bhb. Intriguingly, what is captured in the crystal structure is the S-form bhb, whose stereo configuration best permits the observed hydrophobic contact (Fig. 4c). Such a stereo-selectivity may have interesting functional implications given the fact that the R-β-hydroxybutyrate is the primary form of ketone body, while the S-form bhb-CoA is mainly converted from trans-cr-CoA via a hydration reaction especially during the short-chain fatty acids (SCFA) metabolism 42,43 . The prevalence of Kbhb stereo-specific modification and its Backbone engagement for H4K16 recognition by HDAC1. Coordinates are taken from PDB entry 5ICN. The H4K16 was replaced by a hydroxamic acid functionality to mimic K16ac. Note that an acidic residue D99 dominates peptide backbone recognition in a non-β manner. c List of peptide sequences used for ITC titration. The Kbhb sites are highlighted red and its flanking glycine residues are highlighted blue. d ITC fitting curves of SIRT3 titrated with wild-type and glycine/ alanine-mutant Kbhb peptides around histone sites H3K4 (i), H3K9 (ii), H3K14 (iii), H3K18 (iv), H4K8 (v) and H4K16 (vi). e In vitro deacylation assay results of all wild-type (WT) and mutant histone peptides listed in panel C catalyzed by SIRT3. Error bars represent standard error of mean of three repeats regulation in cell remains to be explored in future studies. Our profiling studies also showed that another hydroxylsubstituted acyllysine, Khib, could be recognized and hydrolyzed by SIRT3, SIRT5, and CobB, but not SIRT1 and SIRT2 (Fig. 1a, c, d). The conservation and divergence of Khib versus Kbhb recognition by different sirtuin family members awaits further structural elucidations.
SIRT3 displayed both broad site-specificity (recognizing multiple acylation sites) and strict class-selectivity (unable to recognize a group of acylation sites) in histone deβ-hydroxybutyrylation. At the molecular level, these seemingly conflicting properties are well explained by a consensus β-staple selectivity of SIRT3 44 . As revealed by our parallel structural analyses, we showed that SIRT3 selects against glycine-flanking motifs due to their intrinsic flexibility. In addition to sites of H4K5, H4K8, H4K12 investigated here, other Kbhb sites of similar motifs are expected resistant to SIRT3 deacylation as well. As a key regulator in mitochondria, SIRT3-senstive mitochondrial acetylome has been profiled 45,46 . Remarkably, the "β-staple rule" is also conserved, in which flanking glycine or proline residues that are detrimental to the β-structure are disfavored 47,48 . Given the sequence and structural similarities of human sirtuins (Supplementary Fig. S10), the observed backbone selectivity is likely conserved among different sirtuin members or of different acylation types. This is supported by previous biochemical profiling studies using either acetyl-peptide microarray 46 or pre-acylated (Kbhb, Khib not included) designer nucleosome library 49,50 , in which glycineflanking Kac peptides or acylated H4 tail were shown to be poor substrates for all human sirtuins. HDAC3 also possesses histone de-β-hydroxybutyrylation activities but displays no clear class selectivity, likely because the Zndependent HDACs do not require a β-staple conformation of peptidyl substrate for recognition 51 .
Histone acetyl as well as non-acetyl acylations are often linked to active gene transcription. Intriguingly, the establishment of histone acetyl/acylation states are likely hierarchical, which is reflected by the fact that many histone acetyl/acyl-transferases are often organized into large complexes that contain histone acylation reader modules 52 . For example, our recent work on YEATS2containting ATAC acetyltransferase complex revealed a "read-write" Kac signal amplification mechanism in which recognition of H3K27ac by YEATS2 is required for ATAC-dependent maintenance of H3K9ac, thus ensuring full activation of ribosomal genes through establishment of histone hyperacetylation 53 . Here we revealed that downregulation of histone acylation levels by sirtuin family members are also hierarchical, indicative of a mechanism of hierarchical gene repression. Conceivably, those acylation sites that survive sirtuin treatment may serve as cis-acting marks to avoid unwanted chromatin compaction; on the other hand, they may act as seeds for re-establishment of a hyper-acylation pattern for gene reactivation by recruiting downstream effector complexes. Such a mechanism is supported by the existence of glycine-flanking motif-specific readers, such as the Brdt bromodomain that recognizes H4 K5acK8ac acetylation patterns 54,55 .
In summary, we demonstrate that the sirtuin family member SIRT3 is an eraser of Kbhb with class-selectivity against non-β (e.g., glycine-flanking) sequence motifs. Our work on hierarchical histone deacylation by SIRT3 suggests a potential regulatory mechanism that connects metabolism to gene regulation through dynamic acylations in sirtuin and modification biology.

Protein and peptide preparation
Human sirtuin genes are kind gifts from Dr. Jiahuai Han at Xiamen University. The NAD + -dependent deacetylase domain encompassing residues 118-399 of SIRT3 was subcloned into a pSUMOH10 vector (modified based on pET28b) containing an N-terminal 10xHIS-SUMO tag. All the mutants of SIRT3 were generated by the Quik-Change site-directed mutagenesis strategy and verified by sequencing. Wild-type SIRT3 118-399 was overexpressed in the Escherichia coli BL21 (DE3) strain (Novagen). After induction with 0.2 mM isopropyl β-D-thiogalactoside (IPTG) at 16°C in LB medium supplemented with 0.1 mM ZnCl 2 overnight, cells were harvested by centrifugation and suspended in buffer A (500 mM NaCl, 20 mM Tris-HCl, pH 7.5, 5% glycerol, 20 mM imidazole), then disrupted by an EmulsiFlex-C3 homogenizer (Avestin). The lysate was further cleared by centrifugation, and the supernatant was loaded onto a Histrap affinity column. After buffer A washing, bound proteins were subjected to on-column cleavage by ULP1 SUMO protease. The tag-free SIRT3 118-399 protein was collected as flow-through, centrifuge-concentrated, and then purified by an anion-exchange QHP column followed by size exclusion chromatography on a Superdex 75 column (GE Healthcare). Purified peak fractions were pooled, concentrated, aliquoted, and stored at −80°C for future use. All mutant SIRT3 and other sirtuin family proteins were purified using essentially the same procedure as described above.
All the histone peptides used in this study were synthesized at >95% purity by Beijing SciLight Biotechnology Ltd. Co.

Isothermal titration calorimetry
Experiments were performed at 25°C on a MicroCal PEAQ-ITC instrument (Malvern Instruments). The sample cell containing 200 μL of 50 μM protein was titrated with 17 successive injections of 750 μM peptide. Acquired titration curves were fitted with the Origin 7.0 program using the "one set of binding sites" binding model. Protein concentrations were measured based on the UV absorption at 280 nm. Peptide concentrations were measured by weighing in large quantity.

In vitro deacylation assays
For deacylation activity profiling studies, the reaction system was prepared by mixing 5 μg of each sirtuin, 1 μg of each acylated lysine peptide and 5 mM NAD + in a buffer solution containing 20 mM Tris-HCl, pH 7.5 and 1 mM DTT. For SIRT3 catalyzed deacylation assays, 1 μM SIRT3 and 100 μM of each histone Kbhb peptide were added into the same reaction solution. For enzymatic kinetics studies, 1 μM of SIRT3 was incubated with different concentrations of histone H3 1-15 K9bhb-S/R (20, 40, 60, 80, 100, 200, 300, 400, and 500 μM) in the same reaction buffer at 37°C for a certain period of time within the initial linear range. After incubation, the reactions were stopped by adding trifluoroacetic acid (TFA) to a final concentration of 5% (v/v) followed by immediate frozen in liquid nitrogen. Enzyme-free reaction systems were used as controls. The resultant reaction mixtures were then analyzed either RP-HPLC or MALDI-TOF MS.
For RP-HPLC analysis, samples were analyzed by a Dionex/Thermo UltiMate 3000 HPLC system with an Acclaim TM RSLC 120 C18 column (2.1 mm × 100 mm, 2.2 μm). After loading, the reaction mixtures were washed by buffer A (0.1% TFA in water) for 10 min, and then eluted by a gradient of 1-20% buffer B (0.1% TFA in acetonitrile) over 20 min. The flow rate was 0.4 mL/min, and the wavelength for UV detection was 215 nm. For mass spectrometry, samples were analyzed by a 4800 plus MALDI TOF/TOF Analyzer (Applied Biosystems/MDS SCIEX) and operated in the Reflector Positive mode using 4000 Series Explorer Software. Laser intensity was set up to 3500 V.
For nucleosome-based de-β-hydroxybutyrylation assays, Kbhb-modified nucleosomes were prepared using bhbtreated HEK293T cells, and then are subjected for SIRT3 treatment. The reaction system was prepared by mixing 1 μg/μL wild type SIRT3 or its catalytic mutant H248Y, 1 μg/μL extracted nucleosomes and 5 mM NAD + in a buffer solution containing 20 mM Tris-HCl, pH 7.5, 350 mM NaCl and 1 mM DTT. After incubation at 37°C for 1 h, the reaction was stopped by adding 5× loading buffer, then boiled at 100°C for 10 min. Enzyme-free reaction systems were used as controls. The resultant reaction mixtures were then analyzed by immunoblotting assays.

Immunoblotting
HEK293T cells were collected and boiled in RIPA lysis buffer (ThermoFisher) in the presence of NAM, TSA, and protease inhibitor cocktail (Selleck). Proteins in the lysate were then separated by SDS-PAGE and transferred onto a NC membrane for blotting using annotated anti-Kbhb or anti-histone antibodies (PTM Biolabs).