Due to its superior Zn²⁺ binding affinity, hydroxamic acids are the most common ZBG in HDACi, as demonstrated by their abundance among clinical and approved inhibitors. As a polar moiety with increased capacity for both inter- and intra- molecular hydrogen bonds, hydroxamates exhibit excellent solubility and in vitro stability [37]. Despite their exquisite potency against HDACs, installing a hydroxamate on any molecule with limited optimization is strongly biased for generating pan-HDACi, among many other possible interactions with metalloproteins (aminopeptidases, carbonic anhydrase, etc.) that can be chelated in a similar manner [38]. Hence, hydroxamates are often faced with selectivity barriers, which are mitigated via optimization of the linker and cap groups [39]. Another major drawback to hydroxamates is their poor PKs due to rapid elimination and metabolite cytotoxicity, thus decreasing their potential for clinical use [40]. Cytotoxicity mainly occurs through a Lossen’s rearrangement (Figure 5) where a reactive isocyanate intermediate forms and can interact with DNA, inducing mutagenesis [41].

The majority of hydroxamate HDACi have a propensity to act in a pan-inhibitory fashion (Figure 6A) [42–49]. For example, trichostatin A displays picomolar potency towards class I HDACs but also inhibits class IIB with nanomolar potency [43]. This phenomenon is also observed with abexinostat (PCI-24781), a preclinical HDACi developed for the treatment of B-cell lymphoma, and the more recently developed HDAC-IN-30 [47, 49, 50]. Notably, hydroxamates inhibit HDAC6 preferentially over other HDACs, whereas class I inhibition is a result of extensive optimization of the inhibitor cap region to achieve higher affinity for HDACs 1–3 and HDAC8 [8, 51–54].

Notably, linker length and composition are not directly correlated with regard to class or isozyme selectivity. For example, trichostatin A contains a chain of sp²-hybridized carbons for a linker while fimepinostat and quisinostat contain pyrimidines as linkers, yet all 3 molecules inhibit HDACs 1–3 in the picomolar range (Figure 6A) [43, 44, 48]. Similarly, pracinostat and abexinostat display nanomolar potencies towards class I and class IIB HDACs in spite of bearing different linkers—cinnamyl and phenyl respectively [42, 43].

To further confirm the lack of linker significance in HDAC isozyme selectivity, two HDAC8-selective inhibitors, MMH409 and PCI-34051 (Figure 6B) show comparable IC₅₀ values (23.4 nmol/L and 10 nmol/L respectively) while containing an anilide and an indole as linkers. A unique feature of these HDAC8 selective inhibitors is their ability to go through an intramolecular π–π stack (Figure 6C). In the case of MMH409, density functional theory (DFT) calculations have shown that the molecule adopts a cis:trans ratio of 93%:7% in solution, which when docked computationally, was shown to be the most stable binding orientation to HDAC8 [43]. While not confirmed for PCI-34051, the molecule can adopt a similar conformation which could explain its selectivity towards HDAC8.

Interestingly, the largest diversities between HDACi are depicted within the cap group where differences in the surface residues of HDAC isozymes are optimized for substrate recognition, unlike the conserved residues around the catalytic tunnel. An interesting observation is the presence of free-amine and N-heterocycles in the cap region across multiple inhibitors (e.g., fimepinostat, quisinostat, HDAC-IN-30, and pracinostat), which may be contributing to increased affinity towards HDAC. This is because the presence of several aspartate residues around the rim of the catalytic tunnels (such as adjacent Asp92 and Asp93 in HDAC3) allows for hydrogen bonding opportunities and potential salt bridges formation depending on the ionization state of heterocycles and amines in the local pH environment.

Due to the advantage of a bigger catalytic tunnel of class I HDACs, 2-amino anilides are often exploited for class I HDAC selectivity as they are larger in comparison to hydroxamates (Figure 4). Therefore, hydroxamate toxicity [via isocyanate formation and eventual mutagenicity (Figure 5)] and persistent off-target effects are eliminated. Surprisingly, despite being selective for class I, 2-amino anilides have a very limited affinity towards HDAC8 (Figure 7). An additional benefit of employing 2-amino anilides is the presence of a phenyl ring which can potentially engage in π–π interactions within the tunnel residues [e.g., phenylalanine (Phe) as demonstrated in Figure 8] further engaging the HDAC protein. Additionally, the ring can be both sterically and electronically tuned for a stronger interaction with individual HDAC family members. For example, the addition of an FT substituent (usually a heterocycle at the 5-position of the anilide), improves the selectivity profile for HDAC2 due to occupation of the lower periphery of the HDAC2 pocket (Figure 9). Thus, inhibitors bearing a FP substituent will not inhibit HDAC3 but will selectively inhibit HDAC1 and HDAC2 (Figure 10). Another notable feature of HDAC3-selective O-amino anilides is the presence of fluorine at the 4- or 5- position of the phenyl ring [e.g., tucidinostat (Table 2), RGFP966, and RGFP109 (Figure 11)]. While the size of the HDAC3 catalytic tunnel is not amenable to anilide substituents larger than fluorine atoms, this trend has been replicated through multiple structure-activity relationship (SAR) studies.

Benzohydrazides have shown immense potential in targeting class I HDACs. Hydrazides hold a remarkable advantage over hydroxamates—avoidance of glucuronidation in phase II metabolism [75, 76]. Typically, UDP-glucuronosyl transferases (UGTs) transfer a glucuronide unit onto a hydroxyl unit to tag it for excretion in phase II metabolism (Figure 13), which was observed with the FDA-approved drug belinostat in clinical trials [75, 77]. This limits the utility of the drug due to faster elimination and deteriorates its PK profile.

The use of benzohydrazides has often been linked with moderate HDAC3 selectivity. Most hydrazides shown in Figure 14 bear an alkyl chain “tail” extending from the β-nitrogen of the hydrazide [78–82]. A propyl chain captures the ideal size of the alkyl chain to be installed as the linker. A butyl chain is also tolerated, although potency loss is observed across HDACs 1–3 with larger chains such as pentyl and larger groups [78–82].

UF010 (Figure 14) is a novel hydrazide inhibitor with a 4-bromophenyl as a linker with exquisite potency for class I HDACs, specifically HDAC2 and HDAC3 [81]. While UF010 has no cap group, the presence of the bromine atom, which is relatively larger in size, can act as a “plug” to the tunnel. The hydrazide moiety of UF010 bears a butyl chain (C4) tail. In a similar manner, SR-4370, a derivative of UF010 with a 2,3-difluorobenzene substituent in place of the bromine atom, shows a similar trend in potency, with improved selectivity towards HDAC3 (21.6–96.6-fold selectivity over HDAC2 and HDAC3) [82]. Compound 4, shares a similar structure to chidamide with the exception of the ZBG being a hydrazide instead of a fluoro-anilide, and the cap group containing a phenyl instead of a picolyl unit [56, 80]. This compound also bears a propyl chain extending from the hydrazides and offers higher potency and selectivity (12–100-fold) towards HDAC3 over HDAC2 and HDAC3 compared to chidamide (pan-class I HDACi) [56, 80]. Compound 5 and compound 6 display the importance of the cap group and its influence on potency [78]. For example, compound 6 shows an increase in inhibition of ~10-fold towards HDAC2, ~8-fold towards HDAC3, and ~7-fold increase towards HDAC1 when compared to compound 5, where the only difference between the two compounds is a methylene unit that makes the cap group of compound 5 an amine while compound 6’s cap group is a benzamide [78]. Finally, compound 7 is an example of a molecule that combines multiple moieties that yield selectivity for HDAC3—the presence of a hydrazide, cinnamyl linker, secondary amine at the end of the linker, an N-heterocycle (indole) in the cap group [79]. Compound 7 exhibits a 280 pmol/L potency towards HDAC3 with 16-fold selectivity over HDAC1 and > 160-fold selectivity over HDAC2 [79].

Thiol inhibitors are often generated from disulfide bridges or thioesters that act as prodrugs, releasing the thiols as ZBGs upon metabolism. For example, romidepsin and psammaplin A (Figure 15A) contain disulfide bridges that are reduced in the intracellular environment to generate free thiols that are capable of Zn²⁺-chelation (Figure 15B) [84, 85]. Similarly, largazole (Figure 15A), a thioester, is recognized as a thioesterase substrate in cells and is hydrolyzed to release a free thiol as a ZBG (Figure 15B) [86, 87]. While offering superior potency towards HDACs, thiols tend to lack selectivity as seen in the example of romidepsin. However, in cellular contexts, the IC₅₀ values of thiol HDACi decrease significantly, due to reduction/hydrolysis of prodrugs. Whereas in vitro experiments that lack reductants or hydrolases show increased IC₅₀ values of such inhibitors [85, 86]. For example, largazole becomes a picomolar inhibitor upon hydrolysis, while psammaplin A also converts to a picomolar inhibitor upon reduction of the disulfide and allows for increased selectivity for HDAC1 (Figure 15A) [58–87].