Histone deacetylase 10 structure and molecular function as a polyamine deacetylase

Cationic polyamines such as spermidine and spermine are critical in all forms of life, as they regulate the function of biological macromolecules. Intracellular polyamine metabolism is regulated by reversible acetylation and dysregulated polyamine metabolism is associated with neoplastic diseases such as colon cancer, prostate cancer and neuroblastoma. Here we report that histone deacetylase 10 (HDAC10) is a robust polyamine deacetylase, using recombinant enzymes from Homo sapiens (human) and Danio rerio (zebrafish). The 2.85 Å-resolution crystal structure of zebrafish HDAC10 complexed with a transition-state analogue inhibitor reveals that a glutamate gatekeeper and a sterically constricted active site confer specificity for N8-acetylspermidine hydrolysis and disfavour acetyllysine hydrolysis. Both HDAC10 and spermidine are known to promote cellular survival through autophagy. Accordingly, this work sets a foundation for studying the chemical biology of autophagy through the structure-based design of inhibitors that may also serve as new leads for cancer chemotherapy.

P olyamines such as putrescine, spermidine and spermine are ubiquitous and essential in all living systems [1][2][3] . Typically present at millimolar concentrations, these cellular polycations bind to nucleic acids and other anionic macromolecules to stabilize structure and regulate function in myriad biological processes. Polyamine metabolism is highly regulated and consists of an elaborately orchestrated balance of cellular uptake, intracellular transport, biosynthesis and catabolism (Fig. 1a) 4 . Dysregulation of polyamine metabolism is frequently associated with cancer and other hyperproliferative diseases, and enzymes of polyamine metabolism are emerging drug targets 5,6 .
A critical strategy for the regulation of polyamine function is reversible acetylation. This strategy is analogous to the regulation of protein function by reversible lysine acetylation through reactions catalysed by histone acetyltransferases and histone deacetylases (HDACs, also known more generally as lysine acetyltransferases and lysine deacetylases). In eukaryotes, this branch of polyamine metabolism is partially compartmentalized in the nucleus (Fig. 1a). Spermidine is acetylated at the N 8 position by an N-acetyltransferase in the cell nucleus 7 and then exported to the cytosol, where it is deacetylated by a metaldependent N 8 -acetylspermidine deacetylase, also known as polyamine deacetylase (PDAC) 8,9 . The substrate specificity of PDAC is very strict, as it does not deacetylate cytosolic N 1 -acetylspermidine or N 1 -acetylspermine 10 . Selective inhibition of PDAC activity in HeLa cells increases N 8 -acetylspermidine levels but not acetylated histone levels, so PDAC activity is distinct from HDAC activity 11 .
To date, the identity of the cytosolic PDAC in eukaryotes has remained elusive. However, selective inhibitors have been developed that block mammalian PDAC activity [11][12][13] . Our recent studies of a bacterial PDAC 14 belonging to the arginase-deacetylase superfamily 15,16 led us to suggest that the enigmatic PDAC might be one of the two cytosolic class IIb Zn 2 þ -dependent HDACs, HDAC6 or HDAC10.
We recently reported the X-ray crystal structures of both catalytic domains (CDs) of HDAC6, showing that a conserved active site lysine serves as a gatekeeper in CD1, in that it dictates specificity for carboxy-terminal acetyllysine substrates bearing a negatively charged a-carboxylate group 17 . In all other human HDACs, this residue is otherwise conserved as leucine or methionine, except for HDAC10 in which it is a glutamate ( Fig. 1b and Supplementary Fig. 1). Accordingly, we hypothesized that this strictly conserved glutamate could similarly serve as a gatekeeper to confer specificity for positively charged acetylpolyamine substrates. HDAC10 is highly expressed in the liver, spleen and kidney 18,19 , enriched in the cytoplasm 19 and resistant to inhibition by sodium butyrate 20 . These properties match the tissue distribution, subcellular localization and inhibitor sensitivity of mammalian PDAC [8][9][10][11][12][13] . Thus, we further hypothesized that HDAC10 might be a PDAC. Here we report enzymological studies conclusively demonstrating that HDAC10 is a robust PDAC and a poor lysine deacetylase. In addition, we report the 2.85-Å resolution crystal structure of HDAC10 complexed with a polyamine transition state analogue inhibitor. This structure reveals that the conserved glutamate gatekeeper and a sterically constricted active site confer specificity for N 8 -acetylspermidine hydrolysis and disfavour acetyllysine hydrolysis.

Results
HDAC10 is a PDAC. To determine steady-state kinetic parameters for the hydrolysis of a variety of acetylpolyamine and acetyllysine substrates, we prepared full-length recombinant HDAC10 from human (H. sapiens, hHDAC10) and zebrafish (D. rerio, zHDAC10). The CDs of these orthologues are highly similar based on 63%/79% sequence identity/similarity; only fulllength constructs yielded soluble proteins. We first assayed the lysine deacetylase activity of these constructs against two commercially available aminomethylcoumarin (AMC)-conjugated fluorogenic acetyllysine peptides, GAK(ac)-AMC and RGK(ac)-AMC (molecular structures of all substrates are shown in Supplementary Fig. 2). Both hHDAC10 and zHDAC10 exhibited very low activity with these peptides (Fig. 2 and Supplementary  Table 1). These measurements were consistent with earlier studies reporting that HDAC10 activity was not measurable using fluorogenic peptide substrates 21 . We also tested activity with 11 non-fluorogenic acetyllysine substrates using a liquid chromatography-mass spectrometry (LC-MS)-based discontinuous assay 17 to avoid potential steric clashes that might occur with the bulky AMC group. These substrates included K(ac)NL and GAK(ac)NLQ, which mimic the peptide segment containing K73 in the proposed HDAC10 substrate MutS homologue 2 (MSH2) 22 . Again, we observed no or very low catalytic activity (Fig. 2 and  Supplementary Table 1). In comparison with its closest relative, the cytosolic isozyme HDAC6 (ref. 17), HDAC10 was clearly not an efficient lysine deacetylase.
We next tested the PDAC activity of hHDAC10 and zHDAC10 against several polyamine substrates using the LC-MS assay (Fig. 2, Supplementary Fig. 3  enzymes exhibited optimal catalytic activity and specificity for the hydrolysis of N 8 -acetylspermidine. For hHDAC10 and zHDAC10, k cat /K M ¼ 2,900±200 and 4,600±300 M À 1 s À 1 , respectively; in comparison, N 1 -acetylspermidine was a poor substrate with k cat /K M ¼ 24±5 and 7±2 M À 1 s À 1 , respectively. Substrate specificity trends observed for zHDAC10 and hHDAC10 matched those of the previously reported mammalian N 8 -acetylspermidine deacetylase activity [9][10][11] . As a positive control, we measured the N 8 -acetylspermidine deacetylase activity of Mycoplana ramosa acetylpolyamine amidohydrolase (APAH), which exhibits a similar preference for PDAC activity over acetyllysine deacetylase activity but which also exhibits broader polyamine substrate specificity 23 . As a negative control, we examined the N 8 -acetylspermidine deacetylase activity of human and zebrafish HDAC6 (ref. 17). HDAC6 preferentially catalyses the deacetylation of the peptide substrate RGK(ac)-AMC with a catalytic efficiency 4100-fold greater than that for the deacetylation of N 8 -acetylspermidine. Therefore, HDAC6 is clearly an acetyllysine deacetylase and HDAC10 is an N 8 -acetylspermidine deacetylase. It is notable that the molecular function of HDAC10 coincides with the cellular function of the cytosolic PDAC indicated in Fig. 1a.
Although the PDAC activity of zHDAC10 and hHDAC10 was highest with N 8 -acetylspermidine, substantial catalytic activity was also observed with acetylcadaverine and acetylputrescine, as well as N-(8-aminooctyl)acetamide; in contrast, the shorter substrates N-butylacetamide and N-(3-aminopropyl)acetamide exhibited minimal or no activity, respectively (Fig. 2). These results suggested that the ideal length of a PDAC substrate ranged 8-12 atoms long. The N4 amino group of N 8 -acetylspermidine was required for optimal activity in comparison with N-(8-aminooctyl)acetamide, which lacks the secondary amino group. Even so, the amino group could be located at N4 or N5, based on the substantial catalytic activity measured with acetylcadaverine and acetylputrescine. However, the amino group could not be at N3, based on the lack of substantial catalytic activity with N 1 -acetylspermidine and N-(3-aminopropyl)acetamide. In addition, given the attenuated catalytic activity measured for N 1 ,N 8 -diacetylspermidine, we concluded that the positive charge of the N 1 -amino group of N 8 -acetylspermidine contributed to, but was not critical for, enzyme-substrate recognition. In comparison with HDAC10, the bacterial PDAC APAH exhibited broad substrate specificity with all polyamine substrates except for N-butylacetamide; it also exhibited weak lysine deacetylase activity (Supplementary Table 1 and Supplementary Fig. 3).
To test our hypothesis that E274 is a gatekeeper conferring substrate specificity for N 8 -acetylspermidine, we prepared E274L zHDAC10. In accord with our hypothesis, this substitution diminished N 8 -acetylspermidine deacetylase activity by 20-fold and enhanced acetyllysine deacetylase activity by B100-fold (Supplementary Fig. 3 and Supplementary Table 1). Thus, the single-point mutation E274L completely reversed the substrate specificity of zHDAC10, converting it from a PDAC into an HDAC (Fig. 2c).
The crystal structure of the Y307F zHDAC10D-AAT complex ( Fig. 3a) revealed an overall butterfly-like architecture with a pseudo-two-fold symmetry axis perpendicular to several a-helices defining the PDAC-CDAC domain interface. Each domain adopted the a/b-fold first observed in the crystal structure of arginase 25 and subsequently observed in the HDAC family 26 . However, the CDAC domain lacked helices ZA1-A2, aA, aB2-aB3 and active site loops L1-L5 (Fig. 3b). Interdomain interactions were mediated by a-helices aF and aG, loop LH (which connects aH1 and aH2) and the C-terminal tail of each domain, burying B1,700 Å 2 surface area. The 48-residue linker was missing in the electron density, presumably due to  AcPUT, acetylputrescine; AcPAD, N-(3-aminopropyl)acetamide; BTA, N-butylacetamide; N 1 -AcSPD, N 1 -acetylspermidine; N 1 -AcSPM, N 1 -acetylspermine;  (Fig. 3c). A close-up view of the active site tunnel revealed a canonical HDAC active site at the base of the tunnel, with AAT binding in an extended conformation (Fig. 3d). The trifluoroketone moiety of AAT formed a gem-diolate, making asymmetric coordination interactions with Zn 2 þ (O1-Zn 2 þ and O2-Zn 2 þ coordination distances of 2.1 and 2.5 Å, respectively); hydroxyl group O2 also hydrogen bonded with H136 and H137. The binding of AAT thus mimicked the tetrahedral intermediate and its flanking transition states for the hydrolysis of N 8 -acetylspermidine. This binding mode was similar to that observed in the APAH-AAT and zHDAC6 CD2-AAT complexes 17,29 .
The structure of zHDAC10 allowed us to study substrate specificity determinants: in particular, the active site was much more constricted compared with other isozymes such as HDAC6. On one side of the active site, gatekeeper E274 was close (3.7 Å) to the N4 atom of AAT (Fig. 3d). Although this interatomic separation was too long for a hydrogen bond interaction, it was sufficiently close for a favourable electrostatic interaction. The conversion of HDAC10 from a PDAC into an HDAC through the E274L mutation (Fig. 2) demonstrated that this electrostatic interaction is important for PDAC specificity.
Opposite E274, there was an additional constriction contributing to PDAC specificity: a unique one-turn 3 10 -helix ZA2 (P 23 ACE) resulting from a two-residue insertion plus a two-residue mutation in the L1 loop relative to zHDAC6, effectively lengthening the narrow active site tunnel by B5 Å (Fig. 3c,d). This four-residue segment is universally conserved in HDAC10 orthologues with only minor variations, appearing predominantly as PECE ( Supplementary Fig. 1). The steric protrusion formed by this segment favours the binding of long acetylpolyamine substrates (Fig. 3c) and disfavours the binding of shorter acetyllysine substrates (Fig. 4). Removal of the 3 10 -helix ZA2 insertion yielded construct zHDAC10 DZA2, which exhibited 15-fold decreased PDAC activity and 16-fold increased HDAC activity (Supplementary Table 1, Fig. 2 and Supplementary Fig. 3). Therefore, both gatekeeper E274 and the 3 10 -helix ZA2 insertion promote PDAC activity and suppress HDAC activity.
Although the structural basis of HDAC10 selectivity for N 8 -acetylspermidine over acetyllysine-containing peptide substrates appeared straightforward, the structural basis of selectivity for N 8 -acetylspermidine over N 1 -acetylspermidine was more subtle to discern. The active site tunnel in the PDAC domain was surrounded by a vast surface of negative electrostatic potential (Fig. 5a,b and Supplementary Fig. 6), resulting in large part from conserved aspartate and glutamate residues ( Supplementary Fig. 1). This electrostatic surface feature is unique to HDAC10 orthologues and contrasts with that of the lysine deacetylase HDAC6 CD2 (Fig. 5c). However, the surface of the HDAC10 PDAC domain became less anionic and ultimately cationic, at the base of the active site tunnel (Fig. 5b). The binding of N 1 -acetylspermidine would place the positively charged N4 amino group closer to this cationic surface, which would disfavour N 1 -acetylspermidine binding relative to N 8 -acetylspermidine binding. The strict substrate regioselectivity of HDAC10 for N 8 -acetylspermidine contrasts with the broad substrate specificity of bacterial APAH, which has a glutamate residue (E117) to stabilize the positively charged N4 amino group through a salt bridge (Supplementary Fig. 7).
The crystal structure of the Y307F zHDAC10D-AAT complex revealed that the primary amino group at the end of AAT is disordered (Fig. 3d), but it would be located near the side chains of N93, D94 and A24. Interestingly, D94 corresponds to acetyllysine binding residues D101 of hHDAC8 and S568 of hHDAC6 CD2 (refs 17,30,31). To study the influence of N93 and D94 on substrate binding and catalysis in zHDAC10, we prepared the N93A and D94A mutants. Neither mutant exhibited substantial differences in steady-state kinetic parameters (Supplementary Table 1 and Supplementary Fig. 3). Thus, neither N93 nor D94 are required for enzyme-substrate recognition and catalysis.
The CDAC domain of HDAC10 has no known catalytic function, although it may play a role in directing cytoplasmic enrichment 19 . Neither the Zn 2 þ ligands nor catalytic residues were conserved in the CDAC domain (Fig. 3b). The overall fold of the CDAC domain was most similar to that of the PDAC domain, as well as the CDs of HDAC6 (B2.1 Å r.m.s.d. for B213 Ca atoms in each). In contrast with the PDAC domain, the CDAC domain exhibited low amino acid sequence identity and greater divergence across species (for example, 28% sequence identity between zebrafish and human proteins) ( Fig. 3a and Supplementary  Fig. 1); however, a-helices aG and aH at the interdomain interface were highly conserved ( Supplementary Fig. 5).

Discussion
Our enzymological and X-ray crystallographic results suggest that E274 of zHDAC10 is conserved in HDAC10 orthologues as a gatekeeper to establish specificity for a cationic substrate, just as a b Figure 4 | Constriction of the zHDAC10 active site by gatekeeper E274 and the gA2 helix. (a) Crystal structure of zHDAC10 with a bound substrate based on superposition of the structure of the H574A zHDAC6 CD2-RGK(ac)-AMC complex (PDB 5EFN). Atoms of zHDAC6 CD2 are omitted for clarity; note that the arginine side chain of substrate RGK(ac)-AMC is disordered and not shown. The docked substrate suggests that unfavourable steric clashes will occur between the backbone carbonyl of acetyllysine and E274 (red) and between the AMC group (which mimics the residue at the þ 1 position adjacent to the scissile acetyllysine residue) and the ZA2 helix (purple). (b) Same structure as a, except that the E274L mutation is introduced (teal) as described in Methods. The E274L mutation widens the active site, alleviating the steric clash with the backbone carbonyl of the acetyllysine substrate, which in turn introduces HDAC activity.
gatekeeper K330 is conserved in the corresponding position of zHDAC6 CD1 to establish specificity for an anionic substrate 17 . This gatekeeper is necessary but not sufficient for full activity and selectivity: both zHDAC10 and zHDAC6 CD1 have additional steric bulk opposite the gatekeeper that further constricts the active site. In zHDAC10, the 3 10 -helix ZA2 provides this steric bulk; in hHDAC6 CD1 (ref. 17), F105 and Y225 provide this steric bulk. Regardless, the gatekeeper appears to be a 'hot spot' for dictating substrate specificity.
With regard to the substrate specificity of HDAC10, a sizeable array of steady-state kinetic measurements clearly demonstrated that HDAC10 is a PDAC (Fig. 2 and Supplementary Table 1). Even so, HDAC10 is proposed to deacetylate the cytosolic proteins Hsc70/Hsp70 and MSH2 (refs 22,32). The deacetylation of Hsc70/ Hsp70 by HDAC10 is not characterized at the molecular level, so we cannot comment on this possibility. Although the deacetylation of K73 of MSH2 correlates with HDAC10 levels, enzyme catalysis has not been directly demonstrated 22 ; moreover, our steady-state kinetic measurements showed that HDAC10 does not catalyse the deacetylation of MSH2-based peptide substrates K(ac)NL and GAK(ac)NLQ (Supplementary Table 1). Intriguingly, however, we demonstrated that the MSH2-based substrate GAK(ac)NLQ is efficiently deacetylated by HDAC6 (Supplementary Table 1 and Supplementary Fig. 3), which is also known to interact with and deacetylate MSH2 in vivo 33 . Our in vitro results are more consistent with HDAC6 being the MSH2 deacetylase. As a point of speculation, perhaps the very low levels of lysine deacetylase activity measured for some but not all peptide substrates (Supplementary Table 1) are responsible for the effect of HDAC10 on MSH2 acetylation status.
As previously mentioned, HDACs adopt the same a/b-fold first observed in the crystal structure of arginase, a binuclear manganese metalloenzyme that catalyses the hydrolysis of arginine to form ornithine plus urea 15,25 . This evolutionary relationship was unexpected, as there is very low amino acid sequence identity between HDACs and arginases. However, identical a/b-folds (b-strand order 21387456) and the conservation of metal binding sites (the Mn 2 þ B site of arginases is conserved as the Zn 2 þ site of HDACs) suggest that HDACs and arginases divergently evolved from a common primordial ancestor. As one biological function of arginase is to provide ornithine for polyamine biosynthesis, it is striking that the arginase-deacetylase fold is also recruited for a PDAC function in polyamine metabolism. Our phylogenetic analysis ( Fig. 6) indicated that the closest relationship between the HDAC and arginase families is between the CDAC domain of Cavia porcellus HDAC10 and H. sapiens agmatinase (sequence identity ¼ 19%). Interestingly, this analysis also suggested that the evolution of PDAC activity in vertebrate HDAC10 and the bacterial deacetylase APAH occurred convergently.
Recently, it has been demonstrated that HDAC10 protects cancer cells from chemotherapeutic drugs by mediating autophagy, a survival response to the cellular damage and metabolic stress induced by cytotoxic drugs; indeed, the upregulation of HDAC10 is a marker of poor outcome for advanced stage neuroblastoma patients 32 . However, the knockdown or inhibition of HDAC10 blocks autophagy in a panel of neuroblastoma cells lines, thereby sensitizing these highly malignant cells to the cytotoxic drug doxorubicin 32 . As the suppression of autophagy to sustain the cytotoxicity of chemotherapeutic drugs is a novel strategy for cancer chemotherapy 34,35 , HDAC10 is an emerging target for the treatment of advanced-stage neuroblastoma 32 .
The polyamine spermidine is also a key factor in autophagy and increased levels of endogenous or exogenous spermidine induce autophagy and extend lifespan in a variety of cell types, including human immune cells 36,37 . Recent studies show that the inhibition of ornithine decarboxylase, which utilizes arginasederived ornithine to generate putrescine, reduces cellular polyamine levels and suppresses autophagy in eukaryotic cells, thereby attenuating infection by Trypanosoma cruzi 38 . Thus, polyamine metabolism is closely linked to autophagy.
Given that HDAC10 is a PDAC with catalytic specificity for the hydrolysis of N 8 -acetylspermidine to yield spermidine and acetate, it is possible that the functions of HDAC10 and spermidine in autophagy are linked. With the crystal structure of HDAC10 now in hand, the structure-based design of isozymespecific inhibitors promises to open new avenues in studying the chemical biology of autophagy as well as the treatment of advanced-stage cancers in which this PDAC is implicated.

Methods
General. No statistical methods were used to predetermine sample size. PfuUltra High-Fidelity DNA polymerase (Agilent Technologies) was used for PCR analysis. Restriction enzymes (New England Biolabs) were used according to the manufacturer's specifications. Primers were synthesized by Integrated DNA Technologies. Escherichia coli strain NEB5a (New England Biolabs) was used for cloning procedures. Peptides were synthesized by Genscript. N-(3-aminopropyl) acetamide was purchased from Chem-Implex International, Inc. Acetylcadaverine a b c was purchased from TCI. N 1 -acetylspermidine dihydrochloride was purchased from Cayman Chemicals, Inc. Acetylputrescine, N 8 -acetylspermidine dihydrochloride, N 1 -acetylspermine trihydrochloride, N-butylacetamide and N-(aminooctyl)acetamide were purchased form Sigma-Aldrich. N 1 ,N 8 -diacetylspermidine was purchased from Toronto Research Chemicals. Inhibitors trichostatin A (TSA) and suberoylanilide hydroxamic acid (SAHA) were purchased from ApexBio. All substrates and ligands are used without further purification (495% purity). The transition state analogue inhibitor AAT was synthesized as previously described 24 .
For human HDAC10, the MBP-tag was not cleaved off. The MBP-fusion hHDAC10 was eluted from Amylose resin by buffer B (20 mM Tris (pH 8.0), 100 mM NaCl, 1 mM TCEP, 5% glycerol) plus 10 mM maltose (Sigma-Aldrich) and further purified by anion-exchange chromatography. For zebrafish HDAC10, the proteins were digested on-column using TEV protease in buffer B to remove the His-MBP tag and further purified by anion-exchange chromatography (HiTrap Q, GE Healthcare). Both zHDAC10 and hHDAC10 were finally loaded onto a HiLoad Superdex 200 column equilibrated in buffer C (50 mM HEPES (pH 7.5), 300 mM KCl, 1 mM TCEP, 5% glycerol). Proteins were concentrated to 410 mg ml À 1 and flash-cooled in liquid nitrogen and stored at À 80°C for further use.
For the preparation of zHDAC10D, 10 mg ml À 1 of zHDAC10 in buffer C was incubated with trypsin (Sigma-Aldrich) with a 1,000:1 molar ratio of zHDAC10:trypsin overnight at 4°C and then reloaded onto a HiLoad Superdex 200 column equilibrated in buffer C. Peak fractions corresponding to the co-elution of the PDAC-CDAC complex were pooled and concentrated to 30 mg ml À 1 (the PDAC-CDAC domains did not dissociate upon proteolytic nicking of the interdomain linker). The PDAC domain (residues 2-370) remained intact after proteolytic nicking as determined by LC-MS/MS. zHDAC10D proteins were flashcooled in liquid nitrogen and stored at À 80°C for further use.
Single-point mutants were generated using the Quickchange kit (Stratagene); all primers used for PCR mutagenesis are listed in Supplementary Table 4. The zHDAC10 DZA2 mutant was prepared by by swapping the helix ZA2 region (D 24 PACEI 29 ) with a tetrapeptide segment (S 461 HHP 464 ) from the corresponding zHDAC6 CD2 L1 loop. All mutants were expressed and purified using the same procedure as described above for wild-type zHDAC10. Y307F zHDAC10D was prepared in the same manner as described above for zHDAC10D.
Crystallization. The Y307F zHDAC10D-AAT complex was crystallized using the sitting drop vapour diffusion method at 4°C. Protein (30 mg ml À 1 ) was first incubated with 5 mM AAT ligand in buffer C on ice for 30 min and then 0.4 ml of  Crystal structure determination. X-ray diffraction data were recorded at the Stanford Synchrotron Radiation Lightsource, beamline 14-1. Data reduction and integration was achieved with X-ray Detector Software 39 ; data collection and reduction statistics are recorded in Table 1. The structure of zHDAC10D (Y307F)-AAT complex was solved by molecular replacement using the programme Phaser 40 and a model of the zHDAC6 CD1-TSA complex (PDB entry 5EEF) 17 less ligands was used as the search probe. Initial phases from the Phaser solution were judged to be correct based on the initial electron density map, in which much of the omitted CDAC domain was clearly visible. The programme phenix.autobuild 41  To study steady-state kinetics with non-fluorogenic peptide substrates and acetylpolyamines, a discontinuous LC-MS was employed based on the protocol of assaying nonfluorogenic acetyllysine peptide substrates for HDAC6 (ref. 17). Assays were performed in triplicate. Briefly, 5 ml of enzyme (0.050-20 mM) in HEPES buffer (20 mM HEPES (pH 7.5), 100 mM NaCl, 5 mM KCl, 1 mM MgCl 2 ) was added to 35 ml of substrate solution to initiate the reaction. After incubation for 15-60 min at room temperature, the reaction was quenched by addition of 50 ml of 20 mM dansyl chloride dissolved in acetonitrile followed by 10 ml of NaHCO 3 (1.6 M, pH 10.0). The deacetylation products were derivatized with dansyl chloride at 50°C for 2 hrs. The deacetylation products were detected by LC-MS using a Waters SQD equipped with an Acquity UPLC (Waters, Milford, MA, USA) and quantified by using the standard curves generated from the mass signals of the corresponding deacetylated polyamines and peptides. Inhibition of HDAC10 activity was measured by using 0.49 mM zHDAC10 with 192 mM N 8 -acetylspermidine. Data were analysed by logistic regression for IC 50 determination and the inhibition constant K i was calculated based on the Cheng-Prusoff equation, K i ¼ IC 50 / (1 þ [S]/K M ) 45 . Assays were performed in triplicate at room temperature.
Electrostatic surface potential and residue conservation. The electrostatic potential of the protein surface was calculated using the Adaptive Poisson-Boltzmann Solver implemented in PyMol 46 . The following parameters were used for calculations: T ¼ 298.15 K, pH 7.5, AMBER forcefield, linearized Poisson-Boltzmann equation. The PQR file was generated using the web server PDB2PQR (http://nbcr-222.ucsd.edu/pdb2pqr_2.0.0/) 47 . Metal ions were manually added into the PQR file before the electrostatic potential calculation. The charge and radius of metal ions were set to þ 2 and 0.74 Å for Zn 2 þ ; þ 1 and 1.33 Å for K þ . Residue conservation of 250 HDAC10 orthologues (National Center for Biotechnology Information (NCBI) Protein Reference Sequences) was analysed by ConSurf 48 (http://consurf.tau.ac.il) and displayed in Pymol.
Phylogenetic tree. Protein sequences and their annotations were retrieved from the UniProt database 49 ; some unannotated sequences were obtained from the NCBI. An unrooted phylogenetic tree of the arginase-deacetylase family was generated based on sequence alignments calculated with ClustalOmega 50,51 and visualized using publicly available tools at the Interactive Tree of Life to generate Fig. 6 (refs 52-54).
Model of the zHDAC10-substrate complex. A model of the zHDAC10-substrate complex was prepared by superimposing the crystal structure of the zHDAC6 CD2-RGK(ac)-AMC complex (PDB 5EFN) on the crystal structure of the closely related deacetylase domain of zHDAC10. To generate Fig. 4a, atoms of zHDAC6 CD2 were omitted for clarity and the molecular surface of zHDAC10 was calculated with Pymol. The E274L mutation was then introduced using Coot and an energetically favourable leucine side chain rotamer was selected that coincided with the side chain torsion angles of the original glutamate side chain. The molecular surface was then calculated with Pymol to generate Fig. 4b.