Abstract

RESULTS

Comparative study of HDAC inhibitors

Recently, we have reported the optimization of a miniaturized kinetic assay for the biochemical profiling of HDAC1, 2, 3, 6 and 8 ¹⁷. However, implementation of this assay for Class IIa HDACs proved challenging due to the low catalytic turnover of the acetylated tripeptide substrate 3 as well as a recently reported Class IIa-specific substrate 4 (Fig 1a), both of which require a prohibitively large amount of enzyme^9,18,19. During assay development, we observed diminished turnover by Class I HDACs of Boc-functionalized substrate 5 compared to tripeptide substrate 3 (Supplementary Fig 1)¹⁸. We therefore devised a new tripeptide substrate 6, which, like substrate 4, features the relatively labile and sterically more demanding trifluoroacetyl group that is readily hydrolyzed by the catalytically less avid Class IIa HDACs (Fig 1a,b). With substrate 6, Class IIa HDACs exhibit markedly faster kinetics, further reducing the requisite enzyme concentration (0.002-0.03 ng/μL; Supplementary Fig 2) and allowing a first, precise profiling of HDACi against HDAC1-9. Enzymatic activity of HDAC10 and HDAC11 could not be determined with these or other prepared substrates (data not shown).

Open in a separate window

Figure 1

Development of a platform for biochemical profiling of human deacetylases. (a) Chemical structure of substrates 3 - 6. (b) Comparative enzymatic activity of HDAC1-9 with tripeptide substrate 3 and trifluoro acetyl-lysine tripeptide substrate 6, studied at equivalent substrate concentration (10 μM). Data represent mean values of three measurements ± s.d. Substrate 6 allows miniaturized, kinetic study of HDAC4, 5, 7, 8 and 9.

Collaboratively, we synthesized and assembled a panel of structurally-diverse small-molecule HDACi 1, 2, 7-20 that comprise most of the relevant literature-reported tool compounds and pharmaceutically developed clinical candidates (Supplementary Fig 3)^1,8. We next conducted a high-throughput, precise profiling of HDACi potency against all Class I and II enzymes, in a miniaturized dose-ranging format (Supplementary Table 1).

A chemical genetic phylogeny of human histone deacetylases

Using statistical methods validated for assessing evolutionary relatedness, we constructed a chemical genetic phylogeny of deacetylases derived from these kinetic data (Fig 2a). This approach was selected to prompt inferences into biochemical, pharmacologic and structural relationships. The analysis revealed a number of unexpected findings. First and foremost, the Class IIa enzymes are not targeted by most HDACi at pharmacologically-relevant concentrations. None of the inhibitors tested demonstrated a preference for Class IIa enzymes. In fact, significant inhibitory activity was observed only several orders of magnitude above the K_i for Class I/IIb enzymes. Consequently, none of the inhibitors tested is suitable for use as a tool compound to study Class IIa enzymes in settings where Class I/IIb enzymes are functionally present (i.e. in cells). Indeed, this class of structurally diverse probes exhibits high functional redundancy for inhibition of HDAC1, HDAC2 and HDAC3. Data for a representative benzamide (MS-275 19) and hydroxamate (SAHA 1) are presented schematically in Fig 2b, c.

Open in a separate window

Figure 2

Chemical phylogenetic analysis of HDACs identifies unexpected selectivity of HDAC inhibitors. (a) Hierarchical clustering of HDACs and a representative panel of structurally-diverse HDAC inhibitor tool and investigational compounds 1, 2, 7-20 weighted by inhibitory potency (K_i). Complete quantitative data are shown in Supplementary Table 1. Data based on mean values of triplicate measurements. (b,c) Chemical structure and enzymatic selectivity profile of (b) MS-275 19 and (c) SAHA 1, overlaying molecular phylogeny. HDAC dendrograms are adapted from Supplementary Figure 4. Circles are proportionate in size to K_i on a logarithmic scale, as shown.

Interrogating the bidirectional hierarchical clustering of small molecules and proteins, remarkable chemotype-deacetylase relationships emerge. Driving the striking clustering of HDACi are principally the linker-chelator motifs, as most clearly observed with the benzamide based inhibitors (i.e. o-aminoanilides MS-275 19, CI-994 20 and MGDC-0103 18, see Supplementary Information for structures). In the second dimension, a provocative correlation was observed when comparing this chemical phylogeny to the molecular phylogeny of HDAC1-11 (Supplementary Fig 4). In general, HDACs with known, high sequence identity exhibit correspondingly high degrees of relatedness in both analyses. Yet pharmacology appears to defy phylogeny for HDAC6 and HDAC8, between which class assignments are reversed. Here, the inhibitor sensitivity emulates the substrate preferences, as for all deacetylases studied (Fig 1b), rendering 6 also the preferred reagent for biochemical studies of HDAC8.

Inhibition of Class IIa deacetylases by acetylated lysine

Substrate 3 shows no appreciable turn-over with human Class IIa deacetylases, despite possessing the acetylated lysine feature present in post-translationally modified proteins. Having established a capacity to elicit sensitive measurements of Class IIa enzymatic activity using the non-natural trifluoroacetyl substrate 6, we asked whether substrate 3 and the related substrate 21 might inhibit HDAC4, 5, 7 and 9. Remarkably, dose-ranging studies of substrates 3 and 21 demonstrated potent inhibitory activity against Class IIa enzymes (IC₅₀ 0.6-2.3 μM; Fig 3). These results demonstrate that substrates 3 and 6 exhibit comparable affinity for Class IIa enzymes, suggesting that the trifluoroacetyl group is not a critical determinant of binding affinity. Additionally, Class IIa enzymes appear to bind ε–N-acetyl lysine peptides with comparable affinity to Class I and IIb HDACs, yet Class IIa enzymes fail to deacetylate acetyl-lysine in this context. These data importantly establish the plausibility that Class IIa deacetylase domains function as acetyl-lysine binding domains, rather than functional deacetylases.

Open in a separate window

Figure 3

Inhibition of Class IIa HDACs by acetylated lysine based substrates. Inhibition of trifluoroacetyl-lysine substrate 6 processing by (a) acetyl-lysine substrate 3 (b) acetyl-lysine substrate 21 is presented in dose-response format for Class IIa HDACs. Data represent mean values of triplicate measurements ± s.d. IC₅₀ curves were fit by logistic regression.

Discovery of a first non-selective HDAC inhibitor

Broad inhibitors of enzyme classes are of great use as chemical probes for cell biology, biochemistry and proteomics. The surprising observation that non-selective HDAC inhibitors do not exist prompted the desire to identify such an agent. As suggested by the chemical phylogenetics, the central clustering of cinnamic hydroxamates identifies this pharmacophore as most leveraged for non-selectivity. We and others have observed dramatic contributions to ligand potency and selectivity by the structure and conformation of HDACi capping features^7,20. Thus, we endeavored to expand a library of capped cinnamic hydroxamic acids, based on a high-throughput, parallel synthesis scheme we have used previously with success in targeting individual HDACs ^21,22. This approach involves the clean and efficient condensation of a hydrazide-based linker-chelator feature with a diverse collection of aldehydes to readily explore the chemical space of the capping group. Meta- and para-substituted hydrazide-functionalized cinnamic hydroxamic acids were prepared (see Synthetic Procedures in Supplementary Information) and condensed with a set of 160 aliphatic and aromatic aldehydes to yield a HDAC-biased library of 320 compounds (Fig 4a).

Open in a separate window

Figure 4

Synthesis and testing of an HDAC-biased chemical library and identification of a non-selective HDAC inhibitor. (a) Library design of meta- and para-substituted hydroxamic acid HDAC inhibitors, utilizing parallel condensation of aldehydes, efficiently samples chemical diversity at the capping feature. (b) Biochemical profiling data for the para-substituted sub-library (n=160 compounds), presented in dose-response format for inhibition of HDAC5. Structural variation in the capping feature was observed to confer a broad range of potency, as illustrated with the most (IC₅₀= 18 nM) and least (IC₅₀ = 55 μM) potent small molecules tested. (c) Comparative biochemical profiling of meta- (red) and para-substituted (blue) sub-libraries for relative inhibition of HDAC2 and HDAC3. The complete library was studied and is displayed at a range of concentrations (0.03, 0.3, 3.0 and 30.0 μM). Compounds of this structural class do not discriminate between HDAC2 and HDAC3. (d) Comparative biochemical profiling of meta- (red) and para-substituted (blue) sub-libraries for relative inhibition of HDAC5 and HDAC7. The complete library was studied and is displayed at a range of concentrations (0.03, 0.3, 3.0 and 30.0 μM). Para-substituted cinnamic hydroxamic acids exhibit increased potency for HDAC5, relative to meta-substituted regioisomers. (e) Specificity profile of pandacostat 22 overlaying molecular phylogeny. HDAC dendrograms are adapted from Supplementary Figure 4. Circles are proportionate in size to K_i on a logarithmic scale, as shown. (f) Immunoblot of Jurkat cells treated with pandacostat for 24 hours and stained for acetylated histones (AcH3K18), acetylated alpha-tubulin (AcTub) or GAPDH. (g) Chemical structure of pandacostat 22.

The entire library was profiled against Class I and IIa HDACs in dose-ranging format to provide a richly annotated data set. The capping feature was confirmed to confer a dramatic effect on target potency, as shown in Figure 4b. Pair-wise comparison of potency for individual deacetylases revealed a substantial impact of linker substitution and geometry on target selectivity, particularly evident between HDAC5 and other Class IIa enzymes (Fig 4c,d; Supplementary Fig 5). Based on these profiling data, we selected four compounds with high potency against Class IIa HDACs relative to Class I inhibition. These compounds were resynthesized on 30 mg scale, purified by reversed phase HPLC and assayed in dense dose-response format for the accurate determination of potency and selectivity.

One compound was identified, which uniformly inhibited all profiled HDAC isoforms, in contrast to control compounds MS-275 19, SAHA 1 and trichostatin A 8 (Fig 4e; Supplementary Fig 6, 7). We term this compound pandacostat 22 (Fig 4g). Assessment of cellular permeability and non-selectivity were assessed by immunoblotting for changes in protein acetylation in treated cancer cell lines. Indeed, pandacostat confers hyperacetylation of Class I deacetylase targets (bulk histones) and the prominent HDAC6 target α-tubulin in a time- and dose-dependent manner (Fig 4f). Both biomarkers unambiguously demonstrate cytosolic and nuclear activity of Pandacostat. Intracellular on-target activity for Class IIa HDACs cannot as of yet be probed due the lack of bona fide markers for these isoforms.

DISCUSSION

Lysine acetylation has emerged as a regulatory mechanism for diverse cellular processes in developmental and disease biology. Recent global proteomic analyses have identified more than 1700 protein substrates of histone deacetylases³. Ongoing, detailed, mechanistic studies of substrate protein structure and function require well-annotated small molecule probes. Here, we report an effort to characterize the target selectivity profile of pharmaceutical and tool HDAC inhibitors. The initial motivation for this research was to ascertain subtle differences in enzyme potency which might avail research or therapeutic opportunities. Given the extensive characterization of HDAC inhibitors by academic scientists and pharmaceutical manufacturers, we were surprised to observe unexpected selectivity and significant target redundancy among this class of structurally-diverse compounds.

Most striking was the general lack of inhibitory activity against Class IIa enzymes. The lack of potency of ortho-aminoanilides for Class IIa HDACs was not surprising based on prior studies of HDAC6 and HDAC8, which suggested extraordinary selectivity for HDAC1, 2 and 3. However, the overall lack of potency of hydroxamic acid-based inhibitors was highly unexpected. We interpret this observation based on the available crystal structures of HDAC4 (2VQM) and HDAC7 (3C0Z, 3C10) bound to hydroxamate inhibitors. None of the ligand-protein complexes shows the expected bidentate chelation geometry of the central zinc cation, as observed in the structures of ligand-bound human HDAC8 (1T64, 1T69) and bacterial homologs (e.g. 1ZZ1). According to DFT calculations, the tight bidentate complexation is a result of the deprotonation of the hydroxamic acid upon ligand binding²³. The observed geometry in the published structures, however, is more in line with weaker monodentate binding mode of the neutral form of the hydroxamic acid (Supplementary Fig 8)²³.

Common to all Class IIa HDACs is the substitution of a tyrosine residue in the active site, which is conserved in Class I enzymes, as a histidine. Furthermore, it has been shown for HDAC4 and HDAC7 that the mutation of the respective histidine to tyrosine markedly increases the biochemical activity of both enzymes^19,24,25. Interestingly, in the Class I structures, this tyrosine forms a hydrogen bond to the hydroxamic acid carbonyl, which will increase binding affinity through hydrogen bond formation and as we speculate, sufficiently lower the pKa of the bound chelator facilitating deprotonation and consequently tighter binding. Consistent with this model is the 100-fold increased affinity observed with the hydroxamate LAQ-824 for the H976Y HDAC4 gain of function mutant ^9,24,25. These observations may explain, in part, the differential potency of hydroxamate-based HDAC inhibitors and provide useful guidance for Class IIa-selective inhibitor design.

The data presented herein underscore a particularly pressing need for Class IIa-selective inhibitors. Reports in the literature have identified putative Class IIa-selective chemotypes, though many of these data must be revisited with the new knowledge of Class IIa enzyme biochemistry. In this manner, substrate 6 may prove a useful research tool for high-throughput screening and follow-up chemistry. In addition, highly potent, selective inhibitors of HDAC6 are equally underrepresented in this set of chemical probes. Clinically, the apparent redundancy of pharmaceutical compounds for HDAC1, HDAC2 and HDAC3 is troubling, but resonates with the class-associated toxicities shared commonly by structurally dissimilar compounds. A historical explanation likely accounts for this redundancy. Current pharmaceutical HDAC inhibitors matured in anti-cancer medicinal chemistry programs, where driving biomarkers of histone hyperacetylation and single-agent cytotoxicity evolved lead candidates for clinical development. Consequently, the Class I targets linked to bulk chromatin acetylation and transcriptional networks are over-represented in the target profile of these agents.

From an evolutionary perspective, Class IIa HDACs predate Class I HDACs and histone proteins⁴, thus establishing the rationale that Class IIa enzymes process cytosolic protein or non-protein substrates^9,19. Our data suggest another plausible function of Class IIa enzymes. Given the observed enzymatic inhibition of the Class IIa enzymes by the simple acetyl-lysine tripeptide 3, we propose that a major function of Class IIa HDACs may be the recognition of acetyl-lysine in a sequence-dependent context. That is, Class IIa HDACs may function as receptors rather than enzymes, in some settings. The dissociation constant of substrate 3 for Class IIa HDACs is 3.7-12 μM, which is within the range of binding affinities reported for bromodomains (10-100uM)²⁶. Like Class IIa deacetylases, bromodomains are found in complex with chromatin-modifying enzymes and regulate transcription. This preliminary observation is the focus of ongoing research in our laboratories.

We present, for the first time, the kinetic study of the biochemically active HDACs and a comprehensive library of tool and pharmaceutical deacetylase inhibitors. These data are derived from robust assays and a novel substrate, which allow for the rapid and efficient study of Class IIa HDACs. Our studies have revealed the unexpected selectivity of previously perceived “non-selective” HDAC inhibitors. From literature-reported crystallographic data and ab initio calculations, we provide a rationale for the diminished potency that will guide future ligand development for Class IIa HDACs. Recognizing the broad, potential utility of a non-selective HDACi, we synthesized a library of Class IIa-biased inhibitors and identified the first pan-HDACi reported, to date with activity in biochemical and cellular assay systems. This tool compound is expected to have great utility to the research community. In studying the chemical phylogenetics of HDACs, we illustrate how a structurally-diverse set of small-molecule probes may be used for the functional classification of a protein family.

METHODS

Supplementary Material

Acknowledgments

Footnotes

References