Redefining the Histone Deacetylase Inhibitor Pharmacophore: High Potency with No Zinc Cofactor Interaction.

A novel series of histone deacetylase (HDAC) inhibitors lacking a zinc-binding moiety has been developed and described herein. HDAC isozyme profiling and kinetic studies indicate that these inhibitors display a selectivity preference for HDACs 1, 2, 3, 10, and 11 via a rapid equilibrium mechanism, and crystal structures with HDAC2 confirm that these inhibitors do not interact with the catalytic zinc. The compounds are nonmutagenic and devoid of electrophilic and mutagenic structural elements and exhibit off-target profiles that are promising for further optimization. The efficacy of this new class in biochemical and cell-based assays is comparable to the marketed HDAC inhibitors belinostat and vorinostat. These results demonstrate that the long-standing pharmacophore model of HDAC inhibitors requiring a metal binding motif should be revised and offers a distinct class of HDAC inhibitors.

Histone deacetylases (HDACs) are an important family of epigenetic regulatory enzymes that selectively remove ε-N-acetyl groups from lysine residues of post-translationally modified proteins. In the context of histones, HDACs serve as epigenetic erasers and effect changes in the transcription of many genes in response to a variety of stimuli.[1] Humans encode 18 HDAC isozymes, divided among five classes (I, IIa, IIb, III, and IV). Classes I, IIa, IIb, and IV are zinc metalloenzymes that share a common three-dimensional structure, while class III HDACs, also known as sirtuins, have distinct structures unrelated to the former and possess NAD-dependent enzymatic activity.

Therapeutic inhibition of HDACs has been pursued for decades broadly across the family and selectively among subsets of the human isozymes. To date, five HDAC inhibitors (HDACis) have been approved for treatment of cutaneous and peripheral T-cell lymphoma and multiple myeloma.[2,3] More recently, HDAC inhibition has been evaluated clinically as a mechanism to reactivate HIV gene expression in latently infected immune cells as part of a viral eradication strategy,[4−6] and it has been studied preclinically in a variety of cardiovascular and neurological disease settings.[7,8]

Pharmacokinetic properties, mutagenic potential, and drug-induced liver injury have been associated with HDAC inhibitors,[9−11] potentially limiting the clinical utility of marketed HDACis beyond oncology indications. These liabilities have been associated with the metal binding elements present in the HDACi pharmacophore. Whether natural product-derived or synthetically designed, HDAC inhibitors have a well-known pharmacophore with three elements that mimic natural peptide substrates (Figure ): a metal binding motif that coordinates to the catalytic zinc, a hydrophobic linker that mimics the N-alkyl side chain of lysine, and a protein surface interacting moiety that serves as a peptide substrate recognition element. A structure-based explanation for this pharmacophore has been provided by X-ray crystal structures of substrate- and inhibitor-bound complexes in which ligands interact with surface residues, occupy the lysine tunnel, and position a metal binding moiety in proximity to the catalytic zinc where, depending on the ligand, they make either one or two inner-sphere contacts to the metal.[12−14]

Figure 1

Panel of HDAC inhibitors. Natural products apicidin (1) and azumamide (2) that inspired the design of novel HDAC inhibitors that lack a zinc-binding moiety with enzymatic inhibitory activity (3–6) and apidicin-inspired ketone 7. Inhibitors are colored to indicate where they bind the HDAC active site: surface (blue), hydrophobic tunnel (black), and catalytic zinc (red).

Here, we describe the discovery and optimization of highly potent HDAC inhibitors that challenge the traditional pharmacophore model by demonstrating that the previously ubiquitous metal binding motif can be removed. Without a zinc-binding moiety, these compounds do not contain electrophilic substructures (i.e., ketones), are negative in the Ames mutagenicity assay, and demonstrate cellular activity comparable to clinically utilized HDAC inhibitors. We characterized the binding mode of this novel inhibitor class by X-ray crystallography and describe its HDAC isozyme subtype selectivity, inhibition kinetics, and activity in multiple cellular models of HIV latency reversal.

Removal of the HDACi zinc-binding moiety has been explored previously, and despite significant effort, there have only been a few compounds reported to date with measurable biochemical activity.[15−18] The starting points for these studies have been the natural products apicidin (1) and azumamide (2, Figure ), which possess large, cyclic peptides that primarily interact with the protein surface. Removal of the zinc-binding elements from 1 and 2 led to inhibitors 3–5. In addition, diversity-oriented synthesis approaches identified a novel macrocyclic scaffold lacking a zinc-binding moiety (6). While each of these inhibitors exhibit significantly reduced activity in both biochemical and cellular assays, they provide inspiration that HDAC inhibition can be achieved with inhibitors that do not engage the catalytic zinc (Table S1).

To better understand the interaction network of the large cyclic peptide family of apicidins, we determined a crystal structure of apicidin (1) bound to human HDAC2 (Figure ) and analyzed the noncovalent interactions qualitatively and quantitatively using Scorpion (Desert Scientific Software).[19] As observed previously in crystal structures of HDACs bound to inhibitors containing a ketone zinc-binding element,[20,21] the ketone of 1 is coordinated to the catalytic zinc of HDAC2 as the hydrated gem-diol(ate) making two inner-sphere contacts (Figure A). Although the electron density maps cannot distinguish OH from O–, we observe asymmetry in the coordination distances between the zinc and the two oxygen atoms of the gem-diol(ate) (1.9 and 2.4 Å), which is consistent with previous observations for trapoxin A, HC toxin, substrate analogs, and ketone-containing inhibitors[20] and suggests that nucleophilic attack by the zinc-activated water leads to a transition state mimic in which the hydroxyl (O1) is stabilized by interactions with Zn2+, His141, and His142, while the oxyanion (O2) interacts with Zn2+ and Tyr304. Consistent with the network of interactions involving the zinc and nearby catalytic residues, the Scorpion scores reveal a significant contribution of the gem-diol(ate) to the overall binding of 1 (Figure S2). Moving away from the zinc-binding moiety, the methylene linker of 1 occupies the HDAC2 tunnel with most of the hydrophobic interactions involving the side chains of Phe151 and Phe206, which are conserved residues across HDAC isozymes 1, 2, and 3. The Scorpion scores for the methylene linker indicate its importance as a major contributor of favorable noncovalent interactions that drive potency. Beyond the HDAC2 tunnel, the cyclic peptide of 1 sits on the protein surface. Despite its relatively large size, the most noteworthy interactions come from just two parts of the macrocyclic peptide. First, the three amide NHs of the cyclic peptide are all within the hydrogen bonding distance of the side chain of Asp100 (Figure A), and the Scorpion scores for these atoms reflect these favorable contacts (Figure S2). Second, the indole of 1 is stacked on top of Gly28, His29, and Pro30. In contrast, the iso-leucyl and piperidinyl side chains of the cyclic peptide are solvent exposed with little direct contribution to binding. This helps explain the moderate ligand binding efficiency (LBE) of 1 (0.27), despite its interaction with the catalytic zinc.

Figure 2

HDAC2-bound structures of apicidin (1) and compound 7. (A) Crystal structure of apicidin (1) (gray) bound to HDAC2 (green). Section of the 2Fo–Fc electron density map contoured at 1σ is overlaid (blue mesh). (B) Crystal structure of compound 7 (orange, PDB 6WBW) bound to HDAC2 (green). Apicidin 1 (gray) is overlaid for comparison. Hydrogen bonds are indicated by yellow dashes; water molecules are red spheres, and zinc is a gray sphere.

In addition to apicidin, we considered the acyclic synthetic inhibitor 7 (Figure ) as a starting point to evaluate the removal of the zinc-binding motif. Compound 7 is the result of the optimization of a screening hit related to apicidin and has been previously described.[22−24] Compound 7 exhibits a similar profile when compared to 1, including biochemical inhibitory potency and HDAC isozyme selectivity, and both contain a ketone zinc-binding moiety. However, 7 is smaller than 1 (478 Da vs 624 Da) and is more ligand efficient (HDAC3 ligand efficiency (LE) of 0.36 for 7, compared to 0.27 for apicidin 1).

We previously reported the crystal structure of 7 bound to HDAC2,[25] which allowed us to directly compare its binding mode to that of 1 (Figure B). The gem-diol(ate) zinc-binding motif and methylene linker of 7 are nearly indistinguishable from 1, in terms of both conformation and noncovalent interactions with HDAC2. However, noteworthy differences are observed at the protein surface region due to chemical divergence of the two compounds. The quinoline of 7 occupies a similar space as the indole of 1, making interactions with Gly28, His29, and Pro30 but with subtle differences (i.e., the methoxyethers of the two do not overlap). The imidazole and amide NHs of 7 also preserve bidentate hydrogen bonds to Asp100. Finally, the azetidine follows the same path as that of the cyclic peptide 1, partially overlapping with the piperidinyl side chain. As judged by the extent of their noncovalent interactions, 1 and 7 are quite similar (overall Scorpion scores of 24.5 and 23.1 for 1 and 7, respectively) with similar distributions of key contacts heavily weighted toward the zinc-binding group and linker (Figure S2). As such, we viewed 7 as an attractive lead for identifying small molecule HDAC inhibitors lacking a zinc-binding moiety.

The exploration of structure activity relationships (SARs) of the lysine-mimicking alkyl chain was examined by removing the ketone of 7 and varying the length of the alkyl chain from zero to eight carbons (Table ). The deletion of the carbonyl of 7 without truncation of the alkyl chain provided compound 8 and resulted in a 3–4 log loss in biochemical inhibitory potency against HDACs 1, 2, and 3. This large reduction in activity can be explained through the loss of inner-sphere zinc coordination contacts of the inhibitor and the introduction of unfavorable contacts between the nonpolar linker and the catalytic zinc. On the basis of this rationale, compounds with a shortened alkyl chain containing six or seven carbons were expected to have similarly unfavorable interactions with the catalytic zinc and were not prepared (i.e., 9 and 10). Compounds with alkyl chains of five carbons or fewer (compounds 11–16) were synthesized, and their inhibitory activity was measured. The five-carbon chain inhibitor 11 had a similar HDAC inhibition profile as compound 8. In contrast, a significant improvement in biochemical activity was observed (nanomolar potency range) when the alkyl chain was shortened to either four or three carbon atoms, 12 and 13, respectively. Further truncation to two, one, or no carbons (i.e., 14–16) led to a gradual weakening of HDAC inhibitory activity, illustrating the importance of retaining the favorable nonpolar interactions of the alkyl chain.

Table 1

Structure Activity Relationships of Alkyl Chain Modifications to HDAC InhibitorsA

Compounds tested are >95% pure as determined by LCMS and 1H NMR spectra. All reported values are the mean of at least two assay measurements with standard deviation within 3-fold of the reported value.

NHS = normal human serum.

% reversal and activation are comparable across all compounds tested.

Both 12 and 13 had the highest HDAC3-derived ligand binding efficiency in this set and demonstrated measurable, albeit weak, activity in a cellular model of HIV latency reversal.[26] However, the relative potency of compound 12 across HDACs 1, 2, and 3 revealed a more balanced profile when compared to 13 that was similar to its progenitors 1 and 7. These data demonstrated that a similar degree of pan-inhibition can be achieved against the class I HDAC isozymes in the absence of a zinc-binding moiety. Importantly, 12 was found to be Ames-negative in an exploratory non-GLP 3-strain Ames assay. The selectivity profile of inhibitor 12 against HDACs 1–11 was examined in a panel of enzymatic assays, revealing that 12 was quite selective for HDACs 1, 2, 3, and 10 (Table S3).

The X-ray crystal structure of 12 bound to HDAC2 reveals clear electron density for the inhibitor in a binding pose that does not interact with the catalytic zinc (Figure ). Shared structural elements of 7 and 12 show indistinguishable ligand conformations in their respective HDAC2-bound states, and the binding pocket residues adopt the same conformation. Rather than contacting inhibitor 12, the catalytic zinc is bound instead to an acetate buffer ion through bidentate inner-sphere contacts; the interaction of a zinc-bound acetate has been observed in other inhibitor-free crystal structures of HDAC isozymes.[27,20,28] The acetate is also within the van der Waals interaction distance of the terminal methyl of 12 in the HDAC2 tunnel. A key observation made from the HDAC2 cocrystal structure with 12 is that the linker length is sufficient to capture available hydrophobic contacts with the side chains of Phe151 and Phe206. Therefore, the >100-fold loss of potency observed for inhibitor 12 is ascribed to the loss of the interaction made by the hydrated ethyl ketone with the catalytic zinc.

Figure 3

Crystal structure of compound 12 (magenta) bound to HDAC2 (green). The 2Fo–Fc electron density is shown (blue mesh), contoured at 1σ.

With the four-carbon alkyl chain of 12 providing submicromolar inhibitory potency and desirable ligand efficiency, we turned next to optimization of the ligand–protein surface interactions. The replacement of the azetidinylamide was prioritized because few protein–ligand interactions were observed with the azetidine (Figures B and 3), providing little contribution to the Scorpion scores of compounds 7 and 12 (Figure S4). An unbiased library approach was taken, which led to the identification of the indole acetamide 17 (Table ). While simple indoles have been shown to be suitable azetidine replacements for inhibitors that contain zinc-binding moieties,[22,23] the efforts described herein focused on improving activity in the absence of the zinc-binding moiety. This led to divergent SAR with additional elaboration of the indole moiety proving to be key to unlocking highly potent HDAC inhibitors (Table ). Initial libraries revealed that indole 17 and 5-methoxyindole 18 were able to provide similar relative potency profiles as 12 with balanced HDACs 1, 2, and 3 inhibition. However, replacement of the azetidine resulted in inhibitors with lower ligand efficiency that were significantly more lipophilic (i.e., ACD LogP), leading to increased off target activities including PXR activation, reversible CYP450 inhibition, and time-dependent CYP3A4 inhibition (TDI) (Table ). Indole substitution at the 2-position and the introduction of a methyl to the benzylic methylene led to compounds 19 and 20, which showed improvement in inhibitory potency while maintaining a HDAC3 LBE of 0.25. HDAC panel selectivity was examined for 19, revealing a similar profile to 12 with selectivity for HDACs 1, 2, 3, 10, and 11 (Table S3). Both 12 and 19 possessed overall HDAC selectivity profiles similar to vorinostat. Compound 19 exhibited good selectivity in a lead profiling panel of 115 biochemical and binding assays at Eurofins Panlabs, Inc., designed to study its selectivity beyond the HDAC enzyme family. The panel consisted of a wide variety of receptors, enzymes, and transporters. The compound was tested in duplicate at 10 μM. Significant hits, defined as showing >50% effect at 10 μM, were subsequently titrated. Compound 19 was found to have one significant hit, which titrated to <1 μM (binding to human serotonin 5-HT2B receptor), and 9 hits titrating between 1 and 10 μM. This represents selectivity of 10–1000-fold for HDAC inhibition and suggests that 19 is a promising lead for further optimization (Table S5). Additionally, compound 19 was negative in the exploratory Ames assay and exhibited reasonable pharmacokinetic properties (Clp 25 mL/min/kg, t1/2 1.2 h, F 26%, Wistar Han rats).

Table 2

Structure Activity Relationships of Indole Acetamide HDAC InhibitorsA

Compounds tested are >95% pure as determined by LCMS and 1H NMR spectra. All reported values are the mean of at least two assay measurements with standard deviation within 3-fold of the reported value.

NHS = normal human serum.

% reversal and activation are comparable across all compounds tested.

On the basis of our understanding of the binding conformation of compound 12, we hypothesized that the indolyl–methyl of 19 provided a favorable conformational bias, positioning the indole to interact with the protein surface. While the methyl does contribute to increasing potency, it does so at the expense of projecting a lipophilic substituent into bulk solvent, thus simultaneously imposing an unfavorable entropic penalty. We determined a crystal structure of 19 bound to HDAC2 (Figure ), which confirmed both the conformation of the substituted indole and the solvent exposure of the methyl substituent. In order to mitigate the entropic penalty, an aminomethylene was incorporated in place of the methyl group, resulting in further potency enhancement while maintaining ligand efficiency and simultaneously improving the physicochemical properties (compound 21). When the aminomethylacetamide was combined with a hydroxyl at the 5-position of the indole, inhibitor 22 was discovered. Despite lacking a zinc-binding moiety, 22 achieves comparable ligand efficiency and biochemical potency against HDACs 1, 2, and 3 as progenitors 7 and 1, which contain zinc-binding elements that contribute significantly to their HDAC potencies. Moreover, compound 22 and related molecules in this series have lowered ACD LogD and reduced off-target activities, providing the first proof-of-concept that HDAC inhibitors with low nanomolar potency are possible without interacting with the catalytic zinc.

Figure 4

(A) Crystal structure and electron density map (blue mesh, contoured at 1σ) for compound 19 containing a hydroxy-methyl-indole. Edge–Face π electron interactions and an extensive water network mediating protein–ligand interaction are observed. (B) Overlay of 19 (cyan) and 12 (magenta). Water molecules are depicted as spheres and the protein, as a semitransparent surface.

To understand the kinetic properties of inhibitors in this series, compounds 7, 12, and 19 were selected because they represent a range of IC50’s and allow a kinetic comparison of molecules with and without a zinc-binding moiety. The three compounds were examined for the inhibition of HDAC3 using progress curve analysis. Consistent with ketone 7 and previously reported HDAC inhibitors lacking a zinc-binding moiety,[16−18,29] compounds 12 and 19 both inhibit HDAC3 with linear progress curves indicating fast-on/fast-off kinetics (Figure S6). Jump dilution experiments performed for both 12 and 19 bound to HDAC3 corroborated a rapid dissociation mechanism for the inhibitors as the enzymatic rate rapidly recovered upon compound dilution (data not shown).

HDAC inhibition has been shown to disrupt HIV latency with several published clinical studies of marketed HDAC inhibitors (developed for oncology) showing promising results.[4,6] We evaluated a panel of compounds, alongside the marketed drugs belinostat and vorinostat, in HIV latency reversal assays. Two model systems were evaluated in these studies. In the first system, a latently infected Jurkat cell line[26,30] was used that contains a single integrated lentiviral vector expressing the Tat H13L mutant gene and a partially attenuated TAT variant as well as TAR RNA elements derived from HIV-1NL4-3. For detection, a luciferase reporter protein was inserted in place of the Nef gene. Activation of the latent HIV provirus is monitored by an increase in luminescence, resulting from the expression of the luciferase reporter. This is reported in Tables and 2. In a second model system, selected inhibitors were evaluated for their effects on HIV latency in a primary cell model.[30,31] Naive CD4+ T cells isolated from peripheral blood mononuclear cells from healthy donors were polarized and infected with VSV-G pseudotyped HIV-1 lacking GagPol but containing Tat H13L, Env, CD8a, and Nef. As a reporter of viral expression, Nanoluciferase P (Nluc) was inserted downstream of IRES. Post-infection, CD8a positive cells were selected and then driven to quiescence by reducing cytokines in cell media. After 28 days post-isolation, cells were monitored for quiescence by reduction in Nluc activity and Ki-67. Quiescent cells were treated with compounds for 24 h and then monitored for reactivation from latency by monitoring the expression of the nanoluciferase reporter via the addition of the NanoGlo luminescent substrate.

Activities in the Jurkat cell latency reversal, primary T-cell model, and biochemical HDAC3 inhibition assays are shown in Table . Human serum (NHS) concentrations were varied in the media for the Jurkat latency reversal assay (0.1% and 5% NHS), and the primary T-cell model was conducted in the presence of 10% fetal bovine serum (FBS). Marketed hydroxamic acid-based HDAC inhibitors vorinostat and belinostat, both of which have been evaluated for HIV latency reversal in a clinical setting, show similar activity across these cellular models of HIV reversal. Their efficacy was unaffected by assay serum concentrations due to their large unbound fractions in rat plasma. Ketone-based HDAC inhibitors apicidin (1) and 7 were also effective in these cellular models of HIV reactivation with apicidin demonstrating significant shifts between 0.1%, 5%, and 10% serum concentrations, attributable to its high plasma protein binding. Gratifyingly, the novel inhibitors 19, 21, and 22, which lack a Zn-binding group, all demonstrated cellular activity in both models of HIV reactivation. In the T-cell model, 19, 21, and 22 showed comparable activity (2–5 μM) to clinically utilized HDAC inhibitors vorinostat and belinostat, apicidin (1), and compound 7. Consistent with cellular activity being impacted by serum levels, the IC50 of 19 shifts 10-fold higher between 0.1% and 5% serum in the Jurkat model and has a similar potency in the primary T-cell assay. Similarly, inhibitors 21 and 22 have large shifts attributable to the presence of serum and their high plasma protein binding. This provides a plausible rationale for the observed biochemical to cellular shifts across the HDAC inhibitors evaluated herein, suggesting that further optimization of cellular activity could be achieved through improvements in the physicochemical properties of leads 19, 21, and 22. Taken together, the activity of the non-zinc binding HDAC inhibitors described herein represents a significant advance to the field, redefining the basic understanding of the requirements needed for a potent small molecule HDACi. These serve as attractive leads for further evaluation and as different tools for continued evaluation of HDACis as latency reversal agents.

Table 3

Biochemical and Cellular Activity of Selected Compounds in Cellular Models of HIV Latency ReversalA

compound	HDAC3 IC50 (nM)	fraction unbound in rat plasma	Jurkat HIV latency reversal IC50 (μM) 0.1% NHSB,D,E	Jurkat HIV latency reversal IC50 (μM) 5% NHSB,D,E	primary T-cell model IC50 (μM) 10% FBSC,D,E
vorinostat	29 ± 13	0.75	1.6 ± 0.2	1.9 ± 0.3	2.0 ± 0.7
belinostat	7.3 ± 2	0.138	0.47 ± 0.001	0.62 ± 0.02	1.3 ± 0.2
apicidin (1)	1.7F	0.0045	0.12F	0.59F	5.4D,F
7	0.6 ± 0.03	0.0574	0.30 ± 0.02	1.0 ± 0.02	0.71 ± 0.3
19	55.2 ± 20.5	0.0006	0.24 ± 0.07	3.2 ± 0.4	2.1 ± 0.03
21	18F	<0.0001	1.0 ± 0.2	4.8	2.3 ± 0.08
22	4.7 ± 2.0	<0.0001	1.3F	7.4	4.5 ± 0.2

Compounds tested are >95% pure as determined by LCMS and 1H NMR spectra. All reported values are the mean of at least two assay measurements with standard deviation within 3-fold of the reported value.

NHS = normal human serum.

FBS = fetal bovine serum.

% reversal and activation is comparable across all compounds tested.

IC50 determination was determined in the presence of an EC10 of vorinostat.

n = 1.

HDAC inhibitors have proven their utility in an oncology setting and recently have shown utility in HIV latency reversal. As such, there is a need to address the limitations of this important class of biologically active molecules. A novel series of HDAC inhibitors lacking a zinc-binding moiety has been developed and described herein. The class of inhibitors described herein do not interact with the catalytic zinc and are nonmutagenic, devoid of electrophilic and mutagenic structural elements. This newly disclosed HDAC inhibitor class exhibits promising selectivity profiles with biochemical and cell-based efficacy comparable to the marketed HDAC inhibitors belinostat and vorinostat.