In vitro characterization of the catalytic domain of human histone deacetylase 5

TL;DR

This study examines the activity and inhibition of HDAC5, a protein linked to disease, to help develop more targeted drugs with fewer side effects.

Contribution

The paper provides new insights into the catalytic behavior of HDAC5 and its inhibition by TFMO-based compounds.

Findings

The catalytic domain of HDAC5 was characterized in vitro.

FFK24 and NT160 inhibit HDAC5 with a TFMO zinc-binding group.

Findings support developing selective inhibitors for class IIa HDACs.

Abstract

Both histone acetyltransferases (HATs) and histone deacetylases (HDACs) control the acetylation state of conserved lysine residues in histone tails. Thereby, modulating chromatin structure and gene transcription. Disturbance of this precisely balanced acetylation state contributes to neuronal, cardiovascular, muscle degenerative, autoimmune diseases and cancer. To restore this delicate balance, HDAC inhibitors (HDACi) are employed. However, employing pan-HDAC inhibitors, that target a broad spectrum of HDACs, often leads to significant side effects. Therefore, the development of isoform specific inhibitors is urgently needed. Among the HDAC family, class IIa HDACs, particularly HDAC5, have emerged as promising drug targets due to their tissue-specific expression patterns and presence in large regulatory complexes. Recent progress in selective class IIa HDAC inhibition has led to the development of novel HDACi that contain a 5-(trifluoromethyl)-1,2,4-oxadiazole (TFMO) zinc-binding group (ZBG) to interact with the Zn²⁺ ion in the active site, avoiding the drawbacks associated with promiscuous hydroxamic acid–based HDACi. This study focuses on the in vitro characterization of the catalytic domain of human HDAC5 and its inhibition by the TFMO-based HDACi FFK24 and NT160. Our findings contribute to a better biochemical understanding of the kinetic parameters of HDAC5 activity and support the development of selective inhibitors targeting class IIa HDACs.

The online version contains supplementary material available at 10.1038/s41598-026-37633-5.

Figures

Purification and SAXS of HDAC5cd. (a) 15% SDS gel showing the different fraction samples from the purification of HDAC5cd WT and H1006Y mutant. Lane 1 shows the standard protein marker. Lanes 2–5 represent samples taken from the starting material after high spin centrifugation, IMAC flowthrough, IMAC wash fraction and IMAC elution. Lane 6 shows the IEX eluate, Lane 7 shows the SEC elution. (b) Scattering data of HDAC5cd. Experimental data are shown in black dots, with grey error bars. The GNOM *p(r)* fit is shown as red line and below is the residual plot of the data. The Guinier plot is added

Enzymatic characterization of HDAC5cd and the H1006Y mutant. (a) Salt dependency. The assay buffer was adjusted to the corresponding salt concentrations and after reaction at RT for 90 min, the reaction was stopped and fluorescence was recorded. Fluorescence was normalized to the 137 mM sample. (b) pH dependency was measured in either 50 mM Hepes (pH 6–6.75), Tris (pH 7–10) or phosphate (pH 11–12) buffer with the corresponding pH. Values were normalized to pH 8. (c) HDAC deacetylation assay of HDAC5cd and its H1006Y mutant (d) on both class I and II substrates Boc-Lys-(Ac)-AMC (blue) and Boc-L

Enzyme inhibition assays of NT160 and FFK24 with HDAC5cd. IC_50_ measurements of NT160 (a) and FFK24 (b) with HDAC5cd with a reaction time of 90 min. Kinetic measurements of increasing NT160 (c) and FFK24 (d) concentrations and a reaction time of 30 min. Lineweaver-Burk depiction of the kinetic measurements for NT160 (e) and FFK24 (f). Graphs were created with Graphpad prism Version 10.3.1 and fitted with the log(inhibitor) vs. normalized response – Variable slope, Michaelis Menten and simple linear regression fit.

Predicted binding modes of HDAC5 inhibitors FFK24 and NT160. (A) Surface representation of the HDAC5 (grey) AlphaFold model with the catalytic zinc ion shown in green at the bottom of the binding pocket. (B) Top four docking poses for each inhibitor: FFK24 in cyan (left) and NT160 in magenta (right) illustrate that the common zinc‑binding head and core align deeply within the active site, whereas the more flexible tail regions extend toward the solvent. (C) Cross-sectional views of the binding pocket highlighting protein residues (stick representation) that interact with the inhibitors and ill

Comparison of the surface architecture of HDAC4 and HDAC5. For HDAC4, both the WT crystal structure (a, 2VQM) and H976Y mutant (b, 2VQV) are bound to a hydroxamic acid inhibitor, which has been removed for clear representation of the active center^[26](#CR26)^. For HDAC5 AlphaFold models were generated for the WT (c) and H1006Y (d) mutant. The conserved histidine and mutated tyrosine residue are highlighted in orange and both sidechains are displayed as sticks, while the Zn-ion is colored in cyan. Further, the active site close to the Zn-ion is highlighted with a red asterisk.

Full text

Introduction

Next to phosphorylation or methylation, acetylation and deacetylation of histone tails represent a key mechanism by which eukaryotes regulate gene transcription^1–5^. In this process histone acetyltransferases (HATs) and histone deacetylases (HDACs) control the acetylation state of conserved lysine residues in histone tails, thereby modulating chromatin structure between open (acetylated) and condensed (deacetylated) chromatin conformations^6,7^. Disturbance of this tightly regulated process causes many diseases including cancer, by disruption of vital developmental and cell cycle steps^8,9^. Both HAT and HDAC protein families are well known to form large regulatory complexes^10,11^. Especially HDACs act together with other transcriptional repressors like YY1, Mad/Max or NCoR/SMRT both in vitro and in vivo^12–15^.

Human HDACs are classified based on their sequence homology to the HDACs of yeast^16,17^. Class I is formed by HDACs 1, 2, 3 and 8, class II by HDACs 4, 5, 6, 7, 9 and 10, class III by HDACs Sirt 1–7 and class IV by HDAC11^16,18^.

In contrast to all other classes, class II HDACs, are divided into subgroups IIa and IIb based on their domain architecture^16^. Members of class IIa (HDAC4, 5, 7, 9) contain N-terminal myocyte enhancer factor-2 (MEF2) and 14-3-3 binding motifs and are roughly double in size compared to those of class I^19–21^. Due to the 14-3-3 binding motifs class IIa HDAC proteins can shuttle between cytosol and nucleus^22,23^. HDACs 6 and 10 form class IIB, as both contain a duplicated histone deacetylase domain^16,24,25^. Due to the complexity of class IIa, a typical approach is the expression, purification and analysis of isolated domains rather than the full-length protein. As a direct result, crystal structures of the catalytic domain are only available for HDAC4 and HDAC7^26,27^. Another difference of class IIa to class I is the variation of a conserved tyrosine residue which is replaced by a histidine in the deacetylase domain in class IIa^28^. Experimentally, the two classes are discriminated by two class specific substrates, Boc-Lys-(Tfa)-AMC for class IIa and Boc-Lys-(Ac)-AMC for class I. Aside from the sterically demanding character of the trifluoroacetyl (Tfa) group, it presents as a better leaving group due to its electron-withdrawing properties compared to the acetyl group, enabling measurement of the weak deacetylase activity of class IIa^29–32^.

Class IIa HDACs have been identified as a favorable target for selective inhibitors due to their participation in large regulatory repressor complexes and tissue specific expression patterns^33–35^. Class IIa HDACs are highly expressed, for example in brain, heart, and skeletal muscle, along with roles in cancer, neurodegeneration, cardiac hypertrophy, and immune disorders^36,37^. Whereas it is commonly accepted that class IIa HDACs act through multi-protein complexes, their pleiotropic particular effects in physiological and pathological conditions are not fully understood. Class IIa HDACs, particularly HDAC4 and HDAC5, act as a necessary linker between MEF2 transcription factors in the nucleus and the SMRT/NCoR corepressor complex containing enzymatically active HDAC3. Thus, enzymatically weak HDACs like class IIa HDACs form crucial complexes to address enzymatically highly active HDACs to the chromatin environment. This eventually leads to histone deacetylation by HDAC3, here for example at the MEF2 target gene promoters, chromatin condensation and gene repression. Important physiological and pathophysiological roles are known for all class IIa HDACs^38,39^. Currently promising applications of class IIa HDAC inhibitors are in the cardiovascular field (with a certain focus on HDAC9), in neurodegeneration, differentiation, immunology and cancer. Crystal structures are not yet available for HDAC5 and HDAC9. Recent advances in inhibitor research lead to the development of a new class of non-hydroxamate HDACi, eliminating the selectivity, mutagenicity and metabolic stability concerns and severe side effects connected with hydroxamate HDACi^40,41^. These TFMO-based, class II selective HDACi such as NT160^42^ and FFK24^43^ (both first described in patent WO 2013/008162), provide promising tools for probing HDAC5 function and offer a basis for the development of more selective therapeutic agents^44,45^. However, there is still a lack of subtype selectivity within the class IIa HDAC class. One approach to tackle this problem is to consistently compare the structures of these enzymes for common and different features. However, despite its biological importance (of course next to the other class IIa HDACs), both high-resolution crystal structure and a detailed enzymatic characterization of HDAC5 are lacking, hindering structure-based inhibitor design and functional understanding. Thus, this study focusses on HDAC5.

We expressed and purified the catalytic domain of HDAC5 (Residues 656–1122) and performed its biochemical characterization. We mutated the class IIa specific histidine residue (H1006) to tyrosine which enabled conversion of both substrate classes, enabling us to determine kinetic parameters of both WT and H1006Y mutant for either substrate. Additionally, we report nanomolar K_i_ values for both the NT160 and FFK24 inhibitors.

Results

In vitro enzymatic characterization of HDAC5cd and the H1006Y mutant

To characterize the catalytic domain (cd) of HDAC5 in vitro, we expressed and purified the catalytic domain (HDAC5cd) spanning amino acids 656–1122 (Q9UQL6). After successful immobilized-metal-affinity chromatography (IMAC), ion-exchange chromatography (IEX) and size-exclusion chromatography (SEC), the purification was monitored via sodium dodecyl sulfate poly-acrylamide gel-electrophoresis (SDS-PAGE) and reveals a band at 55 kDa for HDAC5cd and states a high degree of purity (Fig. 1a). Subjecting HDAC5cd to small angle X-ray scattering (SAXS) revealed that HDAC5cd is folded, a monomer and contains no aggregates (Fig. 1b-d; Table 1). Similarly, the H1006Y mutant was purified yielding a similar amount of protein (7 mg per 6 L cell culture) and purity.

Fig. 1. Purification and SAXS of HDAC5cd. (a) 15% SDS gel showing the different fraction samples from the purification of HDAC5cd WT and H1006Y mutant. Lane 1 shows the standard protein marker. Lanes 2–5 represent samples taken from the starting material after high spin centrifugation, IMAC flowthrough, IMAC wash fraction and IMAC elution. Lane 6 shows the IEX eluate, Lane 7 shows the SEC elution. (b) Scattering data of HDAC5cd. Experimental data are shown in black dots, with grey error bars. The GNOM p(r) fit is shown as red line and below is the residual plot of the data. The Guinier plot is added in the right corner. (c) p(r) function of HDAC5cd indicate an elongated molecule. (d) Dimensionless Kratky plot of HDAC5cd shows a compact elongated molecule.

Table 1. Overall SAXS data HDAC5cd.Data collection parametersSAXS DeviceXenocs Xeuss 2.0 with Q-XoomDetectorPILATUS 3 R 300 K windowlessDetector distance (m)0.55Beam size0.8 mm x 0.8 mmWavelength (nm)0.154Sample environmentLow Noise Flow Cell, 1 mm øAbsolute scaling methodComparison with scattering from pure H_2_ONormalizationTo transmitted intensity by direct beam-Scattering intensity scaleAbsolute scale, cm^− 1^s range (nm^− 1^)0.05–5.5 Sample

HDAC5cd Organism H. sapiens UniProt IDQ9UQL6Mode of measurementbatchTemperature (°C)15Exposure time (# frames)600 s (24 frames)Protein buffer25 mM HEPES pH 7.5, 150 mM KCl, 1 mM DTTProtein concentration (mg/ml)2.15 Structural parameters

Guinier Analysis (PRIMUS) I(0) ± σ (cm^− 1^)0.0362 ± 0.0002Rg  ± σ (nm)2.81 ± 0.02s-range (nm^− 1^)0.129–0.456 min < sRg < max limit 0.361–1.281Data point range1–57Linear fit assessment (R^2^)0.978 PDDF/P(r) Analysis (GNOM) I(0) ± σ (cm^− 1^)0.0362 ± 0.0002Rg  ± σ (nm)2.89 ± 0.02Dmax (nm)10.10Porod volume (nm^3^)82.47s-range (nm^− 1^)0.129–4.795χ2 / CorMap P-value1.035 / 0.542 Molecular mass (kDa) From I(0)49.85From Qp (Porod 1951)45.62From Size & Shape (Franke et al. 2018)47.40From GNNOM (Molodenskiy et al. 2022)52.90From sequence51.06 (monomer)SASBDB accession codes (Kikhney et al. 2020)SASDXL6 Software ATSAS Software Version (Manalastas-Cantos et al. 2021)3.0.5Primary data reductionPRIMUS (Konarev et al. 2003)Data processingGNOM (Svergun 1992)Statistic goodness-of-fit testχ 2, CorMap (Franke et al. 2015)

With the purified HDAC5cd in hand, we examined its enzymatic activity under varying salt concentrations and pH conditions within a fluorogenic HDAC assay. The activity towards the model substrate Boc-Lys-(Tfa)-AMC remained stable up to 600 mM NaCl, but decreased slightly at higher concentrations (Fig. 2a). For the pH screening a clear optimum between pH 7–10 is visible, while the activity is gradually decreasing in buffer with a pH below 7 and above 10 (Fig. 2b).

Fig. 2. Enzymatic characterization of HDAC5cd and the H1006Y mutant. (a) Salt dependency. The assay buffer was adjusted to the corresponding salt concentrations and after reaction at RT for 90 min, the reaction was stopped and fluorescence was recorded. Fluorescence was normalized to the 137 mM sample. (b) pH dependency was measured in either 50 mM Hepes (pH 6–6.75), Tris (pH 7–10) or phosphate (pH 11–12) buffer with the corresponding pH. Values were normalized to pH 8. (c) HDAC deacetylation assay of HDAC5cd and its H1006Y mutant (d) on both class I and II substrates Boc-Lys-(Ac)-AMC (blue) and Boc-Lys-(Tfa)-AMC (black). The reaction was stopped after 30 min. Graphs were created with Graphpad prism Version 10.3.1. All measurements were done in triplicates.

With this knowledge we addressed the kinetics of HDAC5cd with 137 mM NaCl at pH 8.0. The resulting kinetics show that the WT can only utilize Boc-Lys-(Tfa)-AMC (Fig. 2c, black) with a reaction velocity of 1.85 ± 0.14 µmol/min/mg and a K_M_ of 88.1 ± 10.9 µM, while no activity is observed for the class I Boc-Lys-(Ac)-AMC substrate (Fig. 2c, blue). As previously described by Bottomley et al.^26^ for HDAC4cd, reversal of the class IIa specific tyrosine to histidine mutation leads to activity of class IIa HDACs towards the class I substrate. To test if this is also observable for HDAC5cd, we created the corresponding H1006Y mutant and repeated the previous kinetic measurements. In these measurements, the H1006Y mutant exhibited an activity of 1.09 ± 0.03 µmol/min/mg towards the class I substrate (Fig. 2d, blue) as well as an activity of 1.06 ± 0.006 µmol/min/mg towards the class II substrate (Fig. 2d, black). Resulting in a k_cat_ values of 97.1 ± 7.4 min^− 1^ (WT, Class IIa), 57.2 ± 1.6 min^− 1^ (H1006Y, Class I) and 55.7 ± 3.2 min^− 1^. In unison with the 1.7-fold decrease in v_max_ compared to the WT, the K_M_ of the H1006Y mutant is lowered by 2-fold to 38.9 ± 0.3 µM (Class I) and 46.7 ± 8.4 µM (Class IIa) for both substrates (Table 2). Thereby, increasing the catalytic efficiency of the WT (1.10 ± 0.16 µM^− 1^*min^− 1^) to 1.47 ± 0.04 µM^− 1^*min^− 1^ (Class I) and 1.19 ± 0.22 µM^− 1^*min^− 1^ (Class IIa) for the H1006Y mutant.

Table 2 Kinetic parameters of both HDAC5cd and H1006Y mutant towards the class I Boc-Lys-(Ac)-AMC and class II Boc-Lys-(Tfa)-AMC substrate. K_M_ (µM)WT, class IIH1006Y, class IH1006Y, class II88.1 ± 10.938.9 ± 0.346.7 ± 8.4v_max_ (µmol/min/mg)1.85 ± 0.141.09 ± 0.031.06 ± 0.06k_cat_ (min^− 1^)97.1 ± 7.457.2 ± 1.655.7 ± 3.2k_cat_/K_M_(µM^− 1^*min^− 1^)1.10 ± 0.161.47± 0.041.19 ± 0.22

In vitro characterization of the inhibitor NT160 and FFK24

With the purified protein available and the kinetic activity assay for HDAC5cd established, we aimed to characterize the interaction with two literature-known class IIa selective inhibitors NT160^42^ and FFK24^43^ in regard to IC_50_ and K_i_. As a first step towards the kinetic inhibition of HDAC5cd we confirmed the IC_50_ for NT160 (Fig. 3a) and FFK24 (Fig. 3b) as 5.0 ± 1.1 nM and 32 ± 5.1 nM, respectively (Table 3). Adapting these concentrations to the kinetic activity assay yielded multiple inhibition curves (Fig. 3c, d, e, f) which enabled us to calculate nanomolar K_i_ values of 8.85 ± 1.17 nM (NT160) and 95.7 ± 16.7 nM for both inhibitors (Table 3). The resulting K_i_ are 1.7- (NT160) and 3.0-fold (FFK24) larger than the determined IC_50_ values, while the 10-fold difference in inhibition potential reported in the literature is maintained.

Fig. 3. Enzyme inhibition assays of NT160 and FFK24 with HDAC5cd. IC_50_ measurements of NT160 (a) and FFK24 (b) with HDAC5cd with a reaction time of 90 min. Kinetic measurements of increasing NT160 (c) and FFK24 (d) concentrations and a reaction time of 30 min. Lineweaver-Burk depiction of the kinetic measurements for NT160 (e) and FFK24 (f). Graphs were created with Graphpad prism Version 10.3.1 and fitted with the log(inhibitor) vs. normalized response – Variable slope, Michaelis Menten and simple linear regression fit.

Table 3. Comparison of IC_50_ values from the literature, calculated IC_50_ and K_i_ values.IC_50_ literatureIC_50_ this studyK_i_ this studyNT1601.2 ± 0.17 nM^#^5.0 ± 1.1 nM8.85 ± 1.17 nMFFK2412 ± 0.1 nM*32 ± 5.1 nM95.7 ± 16.7 nMRatio FKK24/NT160106.410.8^#^ (Turkman et al.^42^); * (Asfaha et al.^43^) for HDAC4cd.

To elucidate the structural basis of the different inhibition potencies, both NT160 and FFK24 were docked into the active site of the AlphaFold model of HDAC5cd using AutoDock Vina (version1.2.3) and cross-checked with Gnina. The receptor grid box was centered on the catalytic site of HDAC5cd and spanned all residues lining the active site, with box dimensions of 62 × 56 × 50 Å (center coordinates: 18.3, -10.6, 0.1), to ensure full coverage of the catalytic channel and lower pocket. Docking parameters were set to an exhaustiveness of 40, an energy range of 4 kcal/mol, and a maximum of 10 output poses per ligand. Vina’s scoring terms approximate van‑der‑Waals, hydrogen‑bonding, hydrophobic and torsional contributions under an implicit‑solvent, mostly rigid‑receptor model; therefore, absolute binding energies are interpreted qualitatively. In all low‑energy poses (FFK24 -8.2 - -7.4 kcal/mol, NT160 -8.6 - -7.2 kcal/mol), the trifluoromethyl-oxadiazole (TFMO) zinc‑binding group and the adjacent linker occupy the catalytic channel of HDAC5cd and coordinate the metal ion, whereas the distal cap regions adopt multiple orientations consistent with their higher conformational flexibility (Fig. 4).

Fig. 4. Predicted binding modes of HDAC5 inhibitors FFK24 and NT160. (A) Surface representation of the HDAC5 (grey) AlphaFold model with the catalytic zinc ion shown in green at the bottom of the binding pocket. (B) Top four docking poses for each inhibitor: FFK24 in cyan (left) and NT160 in magenta (right) illustrate that the common zinc‑binding head and core align deeply within the active site, whereas the more flexible tail regions extend toward the solvent. (C) Cross-sectional views of the binding pocket highlighting protein residues (stick representation) that interact with the inhibitors and illustrating coordination of the lower inhibitor moiety to the zinc ion and tight accommodation of the shared head/core region, in contrast to the more exposed, mobile tail. FFK24 in cyan (left) and NT160 in magenta (right), hydrogen bonds are shown as yellow dashed lines. (D) Schematic 2D interaction diagrams of FFK24 (left) and NT160 (right) showing conserved contacts formed by the identical head and core (black protein residues and inhibitor atoms) and distinct interaction patterns of the inhibitor tails, depicted in various shades of blue for FFK24 and pink for NT160, respectively; dashed lines indicate different possible interaction types such as hydrogen bonds, polar contacts, and van der Waals interactions.

For FFK24, the head group forms reproducible contacts with Ser788 and Asp789 in the lower pocket, supporting a defined anchoring of the inhibitor tail. In contrast, NT160 not only engages Ser788 and Asp789 but additionally establishes stable hydrophobic and CH/π contacts to Leu973 and at least one proline residue in three of four top‑ranked poses, in some cases involving several proline atoms as interaction partners, indicative of a particularly strong local anchoring of the cap group. Proline residues are known to participate in remarkably stabilizing non‑covalent interactions with aromatic and aliphatic groups, with reported stabilization energies in the range of several tens of kJ/mol, such that neglecting these contacts can lead to an underestimation of ligand binding strength in docking studies^46,47^.

Discussion

Due to the large size of class IIa HDACs, a common approach is to express only the catalytic domain for activity assays and inhibitor screening. Thereby, making it possible to express these domains in E. coli. Similar to HDAC4^26^ and HDAC7^27^, we expressed and purified the catalytic domain of HDAC5 (Amino acids 656–1122) and the H1006Y mutant. Our salt and pH screens of the monomeric catalytic domain show that the enzymatic activity is not significantly impaired at salt concentrations as high as 1 M (Fig. 2a). For the pH optimum, maximum activity is observed between pH 7–10, with the activity reducing significantly at higher and lower pH (Fig. 2b). This is in line with the theoretical pI of 6.01 and the pKs values of the catalytically active site chains. Similar to HDAC4cd^26^, HDAC7cd^27^ and HDAC8^48^, the catalytic Zn^2+^ ion is coordinated by aspartates and one histidine residue in HDAC5cd. During the reaction, an additional histidine and aspartate residue form a proton donor/acceptor chain. Moving from pH 7 towards pH 6 leads to protonation of the imidazole ring of either histidine, disrupting the ability of HDAC5cd to catalyze the deacetylation reaction. Additionally, the decreased activity above pH 10 is in line with findings of Hildmann, et al.^29^ for HDAC8, rat liver HDAC and FB188 histone deacetylase-like amidohydrolase (HDAH).

In the kinetic deacetylation assays the HDAC5cd WT can only deacetylate the class II substrate (Fig. 2c, black), while the H1006Y mutant also shows activity towards the class I substrate (Fig. 2d, blue). This is in line with the study of Bottomley, et al.^26^ which showed that introducing the same mutation in HDAC4cd resulted in activity on the class I substrate. Both the HDAC5cd WT and H1006Y mutant show high activities of 97.1 ± 7.4 min^− 1^ (WT, Class II substrate), 55.7 ± 3.2 min^− 1^ (H1006Y mutant, Class II substrate) and 57.2 ± 1.6 min^− 1^ (H1006Y mutant, Class I substrate) and two digit micromolar K_M_ values for either substrate, comparable to those of HDAC7cd^27^.

The ability of the H1006Y mutant to deacetylate the class I substrate originates from the role of the conserved tyrosine in class I HDACs. Here, the conserved tyrosine residue facilitates direct interactions with the acetylated lysine tail via its hydroxy group, thereby stabilizing the formed oxyanion and tetrahedral intermediate^48,49^. Another effect of this mutation in HDAC4cd^26^ and HDAC7cd^27^, is a 1,000-fold increased activity. We however discovered no such effect for the H1006Y mutant of HDAC5cd. Compared to the activity of HDAC7cd WT of 0.012 ± 0.004 min^− 1^ and the H843Y mutant (in HDAC7cd) of 66 ± 9 min^[− 1 [27^, our WT already shows the 1,000-fold increased activity with 97.1 ± 7.4 min^− 1^. To explain this difference, we compared the AlphaFold models of HDAC5cd WT and H1006Y with the crystal structures of HDAC4cd WT and H976Y bound to hydroxamic acid inhibitor^26^ in Fig. 5. Here we found, that both histidine’s in HDAC4 and 5 (Fig. 5a and c) are rotated away from the active site coordinating the Zn-ion, while both tyrosine substitutions (Fig. 5b and d) are rotated towards the active site. With the distance between both hydroxy groups of the exchanged tyrosine and the coordinated Zn-ion reaching 4.4 Å. One striking difference between HDAC4cd and HDAC5cd however, is the accessibility of the active site. While the active site is less obstructed in HDAC4cd WT, the mutation to tyrosine decreases the accessibility significantly. In contrast, the accessibility of the active site does not change in the AlphaFold model for HDAC5 WT and H1006Y mutant and always resembles a conformation like the HDAC4cd H976Y mutant. Therefore, it is likely that the more optimal and confined configuration of the active site results in 1,000-fold higher activity in HDAC4cd H976Y mutant and HDAC5cd WT. This also highlights, that the tyrosine residue is only necessary to successfully deacetylate the class I substrate^29–32^.

Fig. 5. Comparison of the surface architecture of HDAC4 and HDAC5. For HDAC4, both the WT crystal structure (a, 2VQM) and H976Y mutant (b, 2VQV) are bound to a hydroxamic acid inhibitor, which has been removed for clear representation of the active center^26^. For HDAC5 AlphaFold models were generated for the WT (c) and H1006Y (d) mutant. The conserved histidine and mutated tyrosine residue are highlighted in orange and both sidechains are displayed as sticks, while the Zn-ion is colored in cyan. Further, the active site close to the Zn-ion is highlighted with a red asterisk.

In the initial IC_50_ assays with NT160 and FFK24, we were able to reproduce the reported values of NT160 and FFK24 (Table 3) to similar extend, considering that FFK24 was only tested against HDAC4cd and that NT160 was tested with a commercially assay kit. We further could show that both inhibitors inhibited HDAC5cd in a kinetic assay with substrate concentrations 40-fold higher than in the IC_50_ assays. The results of the kinetic inhibition assays suggest that the inhibition mechanism of both inhibitors does not follow a simple competitive inhibition model, as v_max_ is reduced at high inhibitor concentrations. While both inhibitors are designed to bind within the active center of HDAC5, it cannot be ruled out, that the inhibitors bind to other parts of HDAC5cd in an allosteric fashion. However, it is more likely that the high binding affinities of 8.85 ± 1.17 nM (NT160) and 95.7 ± 16.7 nM (FFK24) compared to the low substrate affinity of 88.1 ± 10.9 µM leads to enzyme mainly bound to inhibitor, resulting in reduced v_max_ since the substrate is not able to displace the inhibitor. Both K_i_ values are 1.7- and 3.0-fold higher than the IC_50_ values, with the initial 6.4-fold difference in calculated IC_50_ between NT160 and FFK24 increasing to 10-fold in K_i_. These results are in accordance with the 10-fold difference in IC_50_ reported by Asfaha, et al.^43^ and Turkman, et al.^42^. This difference in inhibition strength can be explained with our docking analysis. Which suggests that the conserved TFMO - metal chelation and linker positioning are shared between NT160 and FFK24, while the additional network of proline‑ and leucine‑mediated contacts available to the NT160 head group provides a plausible explanation for its approximately ten‑fold higher inhibitory potency compared with FFK24 (Fig. 4).

While studying the catalytic domain of HDAC5 provides a first step towards understanding the function and activity of HDAC5, full-size HDAC5 might display a different catalytic activity and conformation, especially when bound to co-repressors or within large regulatory complexes. However, our characterization platform for HDAC5 obtained the same rank order of inhibition at isolated enzymes and in a cellular context as previously published for FFK24^43^. FFK24 gave an IC_50_ at HDAC4 of 12 nM whereas it obtained an IC_50_ in the cellular HDAC assay specific for class IIa HDACs of 77 nM, whereas vorinostat, a broad-spectrum inhibitor with almost no HDAC4 inhibition (IC_50_ of 42 µM) also showed low activity in the cellular class IIa selective HDAC assay. This confirms that the potency of class IIa HDAC inhibitors can be estimated at isolated enzymes as a surrogate measurement for in vivo characterization. Thus, driving the development of isoform specific inhibitors, which in the larger scope can be used in combination therapy, as shown for Osimertinib and LMK235^50^, to unlock more potent formulations specifically targeting HDAC isoforms.

Apart from inhibitor development, our study presents as the first in depth characterization of the activity of HDAC5. As earlier characterizations by Downes, et al. ^51^ utilizing HDAC5 containing immuno-precipitates and ^3^H-labelled histones focused on functional mutants of the Zn-coordination and proton donor chains.

Taken together we could establish expression, purification and kinetic enzyme assays for HDAC5cd. We prove that HDAC5cd is a monomer in solution and identified the optimal pH and salt concentration for HDAC5cd assays. Further, we compared IC_50_ assays with kinetic measurements and report nanomolar inhibitory activities for both NT160 and FFK24 to HDAC5cd.

Materials and methods

Cloning of the catalytic domain of HDAC5 and the H1006Y mutant

The gene encoding the catalytic domain of HDAC5 (Q9UQL6, Amino acids 656–1122) cloned into pET24a with a C-terminal TEV-8xHis-tag was ordered from GenScript (Rijswijk, Netherlands). The H1006Y mutation was introduced with the primer pair 5’-GGAAGGCGGTTATGACCTGACCG-3’ and 5’-AGCGCCAGAACCACA-3’ by site directed mutagenesis.

Expression and purification

The catalytic domain of HDAC5 (Amino acids 656–1122) termed HDAC5cd, and the H1006Y point mutation were expressed in E. coli BL21 (DE3) in 2YT medium supplemented with 30 µg/mL kanamycin with pET24a-HDAC5cd-TEV-8xHis vector. A main culture of 2 L in baffled flasks was inoculated to OD_600_ = 0.05 from a preculture. Cells were grown at 37 °C and 180 rpm to an OD_600_ = 0.8–1. Next, protein expression was induced by addition of 40 µM ZnCl_2_ and 1 mM IPTG and the temperature was lowered to 20 °C. After expression overnight, cells were harvested by centrifugation at 5,000 xg, and the pellet was resuspended in resuspension buffer (300 mM KCl, 25 mM Hepes pH 7.5, 1 mM DTT) supplemented with 1,000 U DNAse and one protease inhibitor tablet (Roche). Afterwards, cells were disrupted at 1.5 kbar with a cell disruptor (Microfluidics). Cell debris and membranes were removed by another centrifugation at 150,000 xg at 4 °C for 1 h. Next, the supernatant was supplemented with 10 mM imidazole and used for IMAC. Here, the 5 mL HiTrap IMAC HP column (Cytiva) loaded with Ni^2+^ was equilibrated with low IMAC buffer (300 mM KCl, 25 mM Hepes pH 7.5, 1 mM DTT, 10 mM imidazole) before loading the supernatant. After loading, the column was washed with low IMAC buffer till the UV baseline was reached. Next, a wash with wash buffer (300 mM KCl, 25 mM Hepes pH 7.5, 1 mM DTT, 82.5 mM imidazole) was applied before HDAC5cd was eluted with elution buffer (300 mM KCl, 25 mM Hepes pH 7.5, 1 mM DTT, 300 mM imidazole). Elution fractions were concentrated with a centrifugal filter unit (10 kDa cut off, Amicon) and buffer exchanged to low IEX buffer (20 mM KCl, 25 mM Hepes pH 7.5, 1 mM DTT) using a PD10 column (Cytiva) according to manufacturer’s instructions. The resulting protein load was applied onto a 5 mL HiTrap Q HP column (Cytiva) equilibrated with low IEX buffer. After an initial wash at 125.4 mM KCl (18% high IEX buffer (500 mM KCl, 25 mM Hepes pH 7.5, 1 mM DTT)), HDAC5cd was eluted with a linear gradient from 18 to 100% high IEX buffer (500 mM KCl, 25 mM Hepes pH 7.5, 1 mM DTT). The fractions containing HDAC5cd were concentrated with a centrifugal filter unit (10 kDa cut off, Amicron) and subjected to SEC with a Superdex 75 increase 10/300 column (Cytiva) equilibrated in SEC buffer (150 mM KCl, 25 mM Hepes pH 7.5, 1 mM DTT). Prior to injection, centrifugation at 100,000 xg at 4 °C for 15 min was performed.

Smal angle X-ray scattering (SAXS)

The chosen sample to detector distance for the experiment was 0.55 m, results in an achievable q-range of 0.05–5.5 nm^− 1^. The measurement was performed at 15 °C with a protein concentration range of 2.15 mg/ml. The samples were injected in the Low Noise Flow Cell (Xenocs, Grenoble France) via autosampler. We collected 24 frames with an exposure time of ten minutes/frame and scaled the data to absolute intensity against water.

All used programs for data processing were part of the ATSAS Software package (Version 3.0.5)^52^. Primary data reduction was performed with the program PRIMUS^53^. With the Guinier approximation^54^, we determine the forward scattering I(0) and the radius of gyration (Rg). The program GNOM^55^ was used to estimate the maximum particle dimension (Dmax) with the pair-distribution function p(r). All parameters of the SAXS measurements are displayed in Table 1.

HDAC5 IC50 enzyme assays

For the enzyme assay, HDAC5cd (Amino acids 656–1122) was diluted to 0.02 mg/mL with assay buffer (137 mM NaCl, 50 mM Tris pH 8, 1 mM MgCl_2_, 1 mg/mL BSA). Next, in a 96 well format, a dilution series with an inhibitor assay buffer mix was performed (70 µL assay buffer + 10 µL inhibitor). After the addition of 10 µL protein mix (200 ng per reaction), the mix was incubated for 5 min at RT prior to addition of 10 µL 100 µM Boc-Lys-(Tfa)-AMC (Bachem) (final concentration 10 µM). Following an incubation for 90 min at RT, the reaction was stopped with 100 µL stop buffer (100 mM NaCl, 50 mM Tris pH 8, 16 mg/mL Trypsin). Fluorescence (Ex = 355 nm, Em = 460 nm) was measured after incubation at RT for 15 min. The fluorescence from the substrate + inhibitor control was subtracted from all fluorescence measurements. For calculation of the IC_50_, the normalized background subtracted fluorescence was plotted against the logarithmic inhibitor concentration and the Graphpad Prism (V10.3.1) fit “log(inhibitor) vs. normalized response – Variable slope” was used. All measurements were done in triplicates.

HDAC5cd salt and pH enzyme assays

For the salt and pH dependency assays, the assay buffer was altered to contain the indicated salt concentration or pH value/ buffering agent (pH 6–6.75 Hepes, pH 7–10 Tris, pH 11–12 phosphate). In a 96 well-plate format, 80 µL assay buffer, 10 µL protein mix (0.02 mg/mL in assay buffer, 200 ng per reaction) and 10 µL 500 µM Boc-Lys-(Tfa)-AMC (Bachem) (final concentration 50 µM) were mixed and the reaction was stopped with 100 µL stop buffer (100 mM NaCl, 50 mM Tris pH 8, 16 mg/mL Trypsin) after incubation for 30 min at RT. Fluorescence (Ex = 355 nm, Em = 460 nm) was measured after incubation at RT for 15 min. The fluorescence from the substrate control was subtracted from all fluorescence measurements. All measurements were done in triplicates.

HDAC5cd kinetic enzyme assays

For the kinetic enzyme assay, HDAC5cd (Amino acids 656–1122) and the H1006Y mutant were diluted to 0.02 mg/mL with assay buffer (137 mM NaCl, 50 mM Tris pH 8, 1 mM MgCl_2_, 1 mg/mL BSA). Next, in a 96 well format, 70 µL assay buffer mix was mixed with 10 µL inhibitor (diluted in assay buffer, 1% DMSO final). After the addition of 10 µL protein mix (200 ng per reaction), the mixture was incubated for 5 min at RT prior to addition of 10 µL Boc-Lys-(Tfa)-AMC (Bachem) or Boc-Lys-(Ac)-AMC (Bachem) and the reaction was incubated for 30 min at RT. Afterwards, the reaction was stopped with stop buffer (100 mM NaCl, 50 mM Tris pH 8, 16 mg/mL Trypsin) and after incubation at RT for 15 min, fluorescence was measured with a Tecan reader with 355 nm excitation and 460 nm emission wavelengths. To calculate the reaction velocity, a calibration curve with AMC was utilized. To calculate the K_i_ from those measurements, we applied the mixed inhibition model and plotted the inverse reaction velocity against the inverse substrate concentration after Lineweaver-Burk. We further calculated the slopes for the linear relationship of the experimental values and further calculated the α factor and later the K_i_ with Eqs. (1) and (2).

\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\mathrm{m}\:=\alpha\:\mathrm{*}\frac{{v}_{max}}{{K}_{M}}$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\:\alpha=1+\:\frac{\left[I\right]}{{K}_{i}}$$\end{document}

Docking

A molecular docking study was performed to investigate the binding of various inhibitors to HDAC5cd. The AlphaFold3 model of HDAC5 (AF-Q9UQL6-F1) was prepared using AutoDock Tools (version 1.5.7)^56^. During preparation, all water molecules were removed. Kollman charges were added to the protein, which are essential for representing the electrostatic potential of atoms and are commonly used in molecular docking with AutoDock-based protocols. In addition, polar hydrogen atoms were added to the protein structure.

A grid box was defined using AutoDock Tools to encompass the active site of HDAC5, specifically designed to include the coordinates of all residues forming the catalytic site.

Docking simulations were performed using AutoDock Vina (version 1.2.3)^57^. Ligand structures were prepared in PDBQT format, and the following docking parameters were used: maximum number of binding poses = 10, maximum allowed energy difference = 4 kcal/mol, and depth of search algorithm (exhaustiveness) = 40.

The resulting docked complexes were analyzed using AutoDock Tools, and binding poses were ranked primarily based on their predicted binding affinities and interactions with the divalent ion located in the active site, which plays a critical role in HDAC5 function. Validation docking was performed with Gnina (v1.3.2), using the same grid and ligand inputs as in Vina. To partially account for induced fit, selected active‑site side chains were treated as flexible during docking, while the remainder of the receptor was kept rigid^58^. Further analysis of protein-ligand interactions was carried out using LigPlot+ (version 2.2.8)^59^ to visualize hydrogen bonds and hydrophobic contacts. Final visualization and structural interpretation of binding modes were conducted using PyMOL (The PyMOL Molecular Graphics System, Version 2.5.4., Schrödinger, LLC).

Generation of alphafold models for HDAC5cd WT and H1006Y mutant

HDAC5cd sequences were submitted to the AlphaFold server^60,61^ to generate models for both HDAC5cd WT and H1006Y mutant. pLDTT scores for both models including two Zn-ions are found in Supplementary Fig. 1.

Supplementary Information

Below is the link to the electronic supplementary material.

Supplementary Material 1