Design and Synthesis of Dihydroxamic Acids as HDAC6/8/10 Inhibitors

Abstract We report the synthesis and evaluation of a class of selective multitarget agents for the inhibition of HDAC6, HDAC8, and HDAC10. The concept for this study grew out of a structural analysis of the two selective inhibitors Tubastatin A (HDAC6/10) and PCI‐34051 (HDAC8), which we recognized share the same N‐benzylindole core. Hybridization of the two inhibitor structures resulted in dihydroxamic acids with benzyl‐indole and ‐indazole core motifs. These substances exhibit potent activity against HDAC6, HDAC8, and HDAC10, while retaining selectivity over HDAC1, HDAC2, and HDAC3. The best substance inhibited the viability of the SK‐N‐BE(2)C neuroblastoma cell line with an IC50 value similar to a combination treatment with Tubastatin A and PCI‐34051. This compound class establishes a proof of concept for such hybrid molecules and could serve as a starting point for the further development of enhanced HDAC6/8/10 inhibitors.


Introduction
In the past decades, drug discovery efforts have focused intensely on the development of inhibitors with high target selectivity. At the same time it is well recognized that successful drugs typically exhibit polypharmacology, and that the "onetarget-one-disease" approach often oversimplifies the complex biology underlying most pathologies. [1] Combination therapy approaches against multiple targets are clinically successful, but there are advantages to developing a single drug that engages multiple targets, particularly when mono-targeted drugs are not already clinically available. Such advantages include guaranteed action against both targets in drug-exposed tissues, as well as simplified pharmacodynamics, manufacture, and regulatory approval. [2] Histone deacetylases (HDACs) regulate the acetylation state of lysine residues of histones as well as other protein substrates, and therefore play a pivotal role in many cellular processes. Modulation of HDAC activity with inhibitors is known to be effective in treating different pathologies, and four HDAC inhibitor (HDACi) drugs have been approved by the FDA with many more being evaluated in clinical studies. [3] Many HDACi, including the four approved drugs, inhibit most HDAC isozymes and are known to have severe side effects, particularly due to inhibition of HDACs 1, 2, and 3. [4] Isozymespecific HDACi are expected to alleviate these liabilities, and numerous selective HDACi have been described. [5] Specifically targeting two or more distinct HDACs can also be beneficial, but this presents a particular challenge when those two HDACs belong to different isozyme classes with different structural requirements for efficient binding (the Zn 2 + -dependent HDACs are grouped into Class I (HDACs 1,2, 3,8), Class IIa (HDACs 4,5,7, 9), Class IIb (HDACs 6,10) and Class IV (HDAC11)). Such a situation exists for late-stage neuroblastoma, where high HDAC8 (Class I) and HDAC10 (Class IIb) expression levels strongly correlate with poor outcomes, and the two enzymes are considered as targets for treatment. [6] On the one hand, inhibition and knock-down of HDAC8 favors cell-cycle arrest and differentiation, retards cell growth, and induces cell death in vitro and in vivo. [7] On the other hand, inhibition and knockdown of HDAC10 halts autophagic flux and impairs DNA damage repair mechanisms, leading to an increased sensitivity to chemotherapy. [8] Furthermore, simultaneous inhibition of HDAC8 and HDAC10 has been shown to be effective in killing neuroblastoma cells alone and in combination with retinoic acid treatment. [6] HDACi are usually described as containing three structural modules: a zinc-binding group (ZBG), a "linker" moiety, and a "cap group". The cap groups in most HDACi are solvent exposed and often tolerate a variety of chemical modifications. Arming of HDACi cap groups with other targeted scaffolds to make chimeras has been particularly successful, producing combination HDAC-IDO1, [9] -proteasome, [10] -PDE5, [11] -kinase, [12] -IMPDH, [13] -BET, [14] -SERM, [15] -topoisomerase, [16] and other inhibitors. [17] HDACi have also been incorporated into PROTACs. [18] We envisioned developing a new HDACi, with activity against HDAC8 and HDAC10 by combining two isozyme-specific and highly potent HDAC inhibitors into a chimeric inhibitor.
We recently showed that Tubastatin A, which is annotated as a selective HDAC6 inhibitor, is also a highly potent HDAC10 binder. [19] We additionally recognized that the selective HDAC8 inhibitor PCI-34051 [20] bears a structural similarity to Tubastatin A: both compounds share an N-benzylindole core. Where-as, in PCI-34051 (Figure 1, top left) the indole moiety (blue) functions as the linker with a ZBG at C6, it is part of the γcarboline cap group (orange) of Tubastatin A (Figure 1, top right). Similarly, the N-benzyl moiety in PCI-34051 (orange) is the cap group, while functioning as the linker (blue) for Tubastatin A. Interestingly, the two compounds present their ZBGs at opposing positions of this core, which is presumably responsible for their very different selectivity profiles. We postulated that these two known inhibitors could be merged, to make a hybrid HDACi (Figure 1, top middle). Because Tubastatin A inhibits both HDAC6 and HDAC10, these hybrids would likely be HDAC6/8/10 inhibitors. [6] While HDAC6 expression does not significantly correlate with prognosis in neuroblastoma, [7a,21] HDAC6 inhibitors have been found to be well tolerated in clinical studies [22] and selective HDAC6 inhibition has been shown to be non-cytotoxic in cancer settings. [23] We therefore allowed HDAC6 inhibition as a feature of our compounds.
In our strategy, the hybrid inhibitors would contain two ZBGs, each one responsible for selectively binding to different enzymes. Thus, the hydroxamic acid on the heterocycle would serve as the ZBG with respect to HDAC8 inhibition ( Figure 1, lower left), with the phenyl hydroxamate functioning as part of the cap group. The situation would be reversed in the case of HDAC6/10 inhibition (Figure 1, lower right).
We have previously demonstrated that a basic nitrogen in the cap group of Tubastatin A analogs is important for potent HDAC10 binding, probably via hydrogen bonding with gatekeeper residue E272. [19] Therefore, we expected that a hybrid inhibitor bearing a PCI-34051-like indole, with no substitution at and Tubastatin A (right) are depicted with their zinc binding groups (ZBG) in red, linkers in blue, and cap groups in orange. Recognition that both inhibitors share an N-benzylindole scaffold inspired the design of hybrid inhibitors (center) with two ZBGs (red) and a central core that would function as both cap group and linker (green). Lower left and right: Inhibition of HDAC8 and HDAC6/10 would result from engagement of one of the two ZBGs. Lower center: Synthesized mono-and hybrid dihydroxamic acids used in this study.

ChemMedChem
Full Papers doi.org/10.1002/cmdc.202000149 C2 or C3 of the indole, would be unlikely to give potent HDAC10 activity. Little information is available in the literature with respect to SAR around the PCI-34051 linker indole, but we postulated that the bulky γ-carboline cap group of Tubastatin A might be too large to serve as linker (linkers are usually relatively slender) for a hybrid inhibitor. We therefore synthesized a variety of phenyl-hydroxamic acids 1a-1g as HDAC6/10 inhibitors and indolyl/indazolyl-hydroxamic acids 2a-2g as HDAC8 inhibitors to serve as benchmark comparisons to the corresponding dihydroxamic acids 3a-3g, which should inhibit HDAC6/8/10 ( Figure 1, bottom middle). In this numbering scheme, PCI-34051 is labelled as 2a and Tubastatin A as 1c.

Chemistry
The synthesis of N-benzylated indole derivatives was performed in two to three steps starting with alkylation of indole building blocks 4, 5, or 8 with 4-methoxybenzyl chloride (PMBCl) or 4carbomethoxybenzyl bromide to give 6, 7, 9, and 10 (Scheme 1, top and middle). Indoles 1a and 3a were obtained by treatment of methyl esters 6 and 7 with hydroxylamine, respectively. Gramine derivative 1b [24] was made from 1a in a Mannich reaction by using formaldehyde and dimethylamine. Gramines 2b and 3b were obtained from formylindoles 9 and 10, respectively, via reductive amination with dimethylamine to give 11 and 12, followed by hydroxamic acid formation. [25] The γ-carboline scaffold of Tubastatin A analogues 2c and 3c was prepared by Fischer indole synthesis using hydrazine 13 and 1methylpiperidin-4-one (14), with subsequent BOC protection to give 15 (Scheme 1, bottom). Bromide 15 was converted to ester 16 via carbonylation, CoreyÀ GilmanÀ Ganem oxidation, and then BOC removal. [26] Lastly, alkylation of 16 with PMBCl or 4carbomethoxybenzyl bromide before hydroxamic acid formation provided 2c and 3c, respectively.
In analogy to the gramine derivatives 2b/3b, an additional step for the synthesis of the hydroxamic acids 1g, 2g and 3g was performed (Scheme 3, bottom). Starting from indazoles 31 and 32, [31] benzylation followed by reductive amination gave 33-35, which were converted to the corresponding hydroxamic acids 1g, 2g, and 3g as before.
Within these three scaffolds, inhibition of HDAC8 was slightly diminished by the introduction of a second hydroxamic acid moiety when compared to its parent monohydroxamic acid inhibitor (e. g. 3a compared to 2a). We were intrigued to find that HDAC10 activity was strongly increased by the addition of a second hydroxamic acid group in all three cases, with 3f showing the largest increase in potency by a factor of 70 over 1f. The high potency of these substances, despite their lack of a basic amine group, was surprising in light of our previous findings with Tubastatin A derivatives. [19] We also examined the γ-carboline series 1c/2c/3c and found the HDAC10 activity of 3c is similar to 1c (Tubastatin A). The bulky γ-carboline as a linker group (2c) produced the weakest HDAC8 inhibitor from the six monohydroxamic acids 2a-2g. Furthermore, the corresponding dihydroxamic acid derivative 3c showed even further diminished inhibition values toward HDAC8 (pIC 50 = 6.29). This was consistent with our hypothesis that γ-carboline derivatives would be too bulky.
Having shown that our design plan succeeded with a variety of scaffolds, we tested the dihydroxamic acids (3a-3g) against all the remaining Class I and Class IIb HDACs to establish selectivity profiles (Table 2). HDACs 1, 2, 3, and 6 were assayed with the HDAC-Glo TM I/II system. As expected, all of the dihydroxamic acids are excellent HDAC6 inhibitors, with pIC 50 values of 7.12-7.82. They also show some increased activity against HDAC1-3, when compared to PCI-34051 and Tubastatin A. Compounds 3b and 3c have relatively weak HDAC8 and moderate HDAC1 activity, disqualifying them for further biological investigation along with 3a and 3f, which have the highest HDAC1-3 activities of all the inhibitors. Substances 3d, 3e, and 3g have the best selectivity profiles overall with low HDAC2/HDAC3 and moderate HDAC1 activity, and were selected for further biological testing (vide infra).
Parallel to our biochemical evaluation, we attempted to cocrystallize the dual inhibitors with Danio rerio (zebrafish) HDAC10, [32] where we had introduced A24E and D94A substitutions to more closely resemble the human HDAC10 active site. The crystal structure of the "humanized" zebrafish HDAC10-3a complex was determined at 2.05 Å resolution (PDB 6VNQ), whereas other inhibitors either gave no crystals, or their resulting crystals diffracted poorly. The overall protein structure is quite similar to that of wild-type zebrafish HDAC10 in its complex with a slender trifluoroketone inhibitor, [33] with a rootmean-square deviation (rmsd) of 0.24 Å for 514 Cα atoms (Table S2). However, due to the rigidity and bulk of 3a there are significant local structural changes in the active site. Specifically, the 3 10 helix containing the P 23 (E,A)CE motif that protrudes into the active site shifts, on average, by 1.4 Å (maximum shift = 1.9 Å). In other HDAC10 structures, the P 23 (E,A)CE motif sterically constricts the active site, presumably to favor the binding of long slender polyamine substrates. However, the current structure reveals that this motif can shift to accommodate the binding of certain bulky inhibitors.
Zinc coordination by the ionized hydroxamate group of 3a is achieved by a mixture of two different monodentate binding modes (Figure 2A). The hydroxamate of the major conformer (67 % occupancy) coordinates to zinc through the NÀ O À group (OÀ Zn 2 + separation = 2.1 Å; Figure 2B). The phenolic hydroxyl group of Y307 is within hydrogen bonding distance to both the hydroxamate NH and NÀ O À groups (OÀ N and OÀ O separations = 2.6 and 2.7 Å, respectively). A Zn 2 + -bound water molecule is also observed (OÀ Zn 2 + separation = 2.2 Å), which donates a hydrogen bond to the hydroxamate C=O group (OÀ O separation = 3.1 Å) and forms hydrogen bonds with H136 and H137 (OÀ N separations = 2.3 and 2.7 Å, respectively).
At first glance, the hydroxamate group of the minor conformer of 3a (33 % occupancy) appears to coordinate to Zn 2 + in a manner similar to that observed for bidentate hydroxamate-zinc interactions as observed in other HDAC10inhibitor complexes ( Figure 2C). [33] However, the C=OÀ Zn 2 + separation of 2.7 Å is not consistent with inner sphere metal  bond with N93, which in turn forms a hydrogen bond with W205. Of note, N93 of zebrafish HDAC10 aligns with D91 of human HDAC10, so these hydrogen bond interactions may be somewhat altered upon inhibitor binding to human HDAC10. Although no dihydroxamic acid in this study provided crystals of sufficient quality with HDAC8, the crystal structure of 3a complexed with D. rerio HDAC6 was determined at 1.94 Å resolution (Table S3, PDB 6VNR). The structure of this complex reveals that 3a binds to HDAC6 as a single conformer (Figure 3A). The catalytic Zn 2 + ion is coordinated in monodentate fashion by the hydroxamate NÀ O À group of 3a (average OÀ Zn 2 + separation = 2.1 Å), and the hydroxamate C=O group accepts a hydrogen bond from the Zn 2 + -bound water molecule (average OÀ O separation = 2.4 Å). The Zn 2 + coordination geometry is similar to that observed for the major conformer of 3a bound to HDAC10 ( Figure 2B). Also similar to the HDAC10-3a complex, the aromatic ring of the phenylhydroxamate nestles in an aromatic crevice, here defined by F583 and F643.
Interestingly, superposition of the two enzyme-inhibitor complexes reveals that the capping group conformation of 3a differs between the HDAC6 and HDAC10 complexes ( Figure 3B). As observed in the HDAC10-3a complex, the capping group is clamped down by E24 of the P 23 (E,A)CE motif; however, the capping group of 3a in the HDAC6 complex is oriented toward solution; the indole hydroxamate makes hydrogen bond interactions with N645 as well as R788 of a symmetry mate in the crystal lattice. Since the inhibitor capping group has a specific binding site in a pocket in the active site of HDAC10 that is not conserved in the active site of HDAC6, this feature must contribute to selectivity for binding to HDAC10. We next measured viability of the HDAC8/10 sensitive SK-N-BE(2)C neuroblastoma cell line [6] after treatment with our hybrid inhibitors. On the basis of our biochemical profiling and previous experience with Tubastatin A derivatives, [19] we began with the two amine-containing inhibitors 3e and 3g. We were surprised to find that both hybrid molecules showed little effect up to 100 μM, although a 1 : 1 molar ratio of PCI-34051 (2a) and Tubastatin A (1c) gave an IC 50 value of 7.3 μM (6.5-8.1 μM 95 % C.I.; Figures 4A and S1). In order to explain this discrepancy, we measured cellular markers/phenotypes which are indicative of target engagement. Whereas PCI-34051 (2a) and HDAC8 inhibitor 2g increased acetylation of the HDAC8 substrate SMC3 in a dose-dependent manner, [6] hybrid inhibitor 3g showed no effect relative to solvent control ( Figure 4B). Furthermore, dihydroxamates 3e and 3g failed to produce an HDAC10 knockdown phenotype, i. e. increased lysosomal staining with the acidotropic LysoTracker DND-99 dye ( Figure 4C). [8b] Tubastatin A (1c) and monohydroxamate 1g were effective as positive controls.
These data pointed toward poor cell permeability of the highly polar dihydroxamic acids 3e and 3g (Table S4). We examined this further by testing target engagement of selected inhibitors in a cellular BRET HDAC10 target engagement assay. [34] Substances 1d and 1g had very similar FRET and BRET pIC 50 values, indicating good cell permeability for monohydroxamic acids with or without a basic amine side chain ( Figure 4D). Dihydroxamic acid 3g, on the other hand, had a pIC 50 value in the BRET assay which is more than 60 fold weaker than in the FRET assay, consistent with poor cell permeability. Compound 3d, which lacks a basic amino side chain and is slightly less polar, showed only a moderate~5-fold loss of potency between the FRET and BRET assays, pointing toward improved cell permeability.
As discussed above, previous data from our group showed that only those Tubastatin A derivatives which contain a basic nitrogen in the cap group are potent (pIC 50 �~8.0) HDAC10 binders. [19] At the outset of this study, we assumed that our hybrid molecules would also require a basic nitrogen in their cap/linker group. Even though all the compounds in this study which contain such a nitrogen are indeed potent HDAC10 binders (i. e. 3b, 3c, 3e, and 3g), we were surprised to find that this functionality was not required (i. e. 3a, 3d, and 3f). In the latter case, the addition of a hydroxamic acid at C6 of the indole ring was sufficient to produce highly potent HDAC10 binders (e. g. 3f versus 1f). The hydrogen bond found between the cap group hydroxamic acid in 3a with N93 in zebrafish HDAC10 may also be formed with D91 in human HDAC10, potentially explaining the tight HDAC10 binding of the dihydroxamic acids. Hydrogen bonding to D91 may offer another opportunity, in addition to binding with the gatekeeper E272 residue, for the development of potent and selective HDAC10 binding.

Conclusion
In summary, the similar central scaffolds of PCI-34051 (2a) and Tubastatin A (1c) served as the basis to develop proof of concept hybrid molecules which target HDAC8 and HDAC10 (and HDAC6). We synthesized potent and selective inhibitors and gained structural insight into the binding of "bulky" inhibitors to HDAC10, but not all substances showed cellular activity. We attribute this discrepancy to poor cell permeability,  Table 1. BRET values are determined from an experiment run in triplicate. E) Western blots of Ac-SMC-3, Ac-tubulin, and Ac-histone 3. The pan-HDACi panobinostat was used as a positive control for histone 3 acetylation. F) Dose-response data for 3d and a 1 : 1 molar ratio of PCI-34051 (2a) and Tubastatin A (1c). The cell proliferation assays were run in triplicate. Error bars represent S.D. Data are normalized to a vehicle control, which was also performed in triplicate. potentially due to the high polarity of the substances. Dihydroxamic acid 3d is sufficiently cell permeable to inhibit the growth of SK-N-BE(2)C cells, and showed activity similar to a combination treatment of 2a and 1c. Future work could aim to improve efficiency of such hybrids using a pro-drug strategy that masked one or both of the hydroxamic acid groups, or via the replacement of at least one of the hydroxamic acids with a different ZBG.

Experimental Section
Expression and purification of TwinStrepII-GST-HDAC10: A synthetic gene encoding TwinStrepII-GST-HDAC10 (human) was ordered from GeneArt (Thermo Fischer Scientific) and subcloned into the pFastBac1 vector. The resulting construct was used for transposition in E. coli DH10EMBacY cells. The isolated bacmid DNA was then used to generate the recombinant baculovirus. For protein expression, 10 mL of baculovirus was added to 1 L of Sf21 cells at a density of 1 × 10 6 cells/mL. The infected Sf21 cells were grown for 72 h in Sf-900 III SFM medium (Thermo Fischer Scientific) at 27°C. Cells were harvested by centrifugation and re-suspended in running buffer (100 mM Tris pH 8.0, 150 mM NaCl, 1 mM EDTA and 1 mM DTT) supplemented with 10 mM MgCl 2 , benzonase and cOmplete protease inhibitors (Merck). The cells were lysed using a Dounce homogenizer and the resulting lysate was centrifuged for 30 min at 4°C at 125 000 g in an ultracentrifuge. The clarified lysate was then loaded onto a 5 mL Strep-Tactin Superflow high capacity column (IBA) pre-equilibrated in running buffer. After sample loading and washing, the TwinStrepII-GST-HDAC10 protein was eluted in running buffer supplemented with 5 mM desthiobiotin (IBA). The elution fractions containing TwinStrepII-GST-HDAC10 were pooled and concentrated before being injected onto a HiLoad 16/600 Superdex 200 pg size exclusion chromatography column (GE Healthcare) pre-equilibrated with 25 mM HEPES/NaCl pH 7.5, 150 mM NaCl, 0.5 mM EDTA, 1 mM DTT and 10 % glycerol. Samples were eluted from the size exclusion chromatography column in the same buffer, flash-frozen in liquid N 2 and stored at À 80°C.

Note on the TR-FRET assay:
We have made slight modifications to the TR-FRET assay since the original publication where we described it. [19] Control experiments indicate that pIC 50 values for a given inhibitor tested in both assay formats are not statistically different. Therefore, data from the two assay formats can be reliably compared. The TR-FRET measurements of the monohydroxamic acids in this manuscript were measured in the original assay format, which is described directly below this paragraph. The TR-FRET measurements of the dihydroxamic acids in this manuscript were measured in the modified format, which is described two paragraphs below this one.

TR-FRET assay (used with the monohydroxamic acids):
All TR-FRET experiments were performed in white 384-well ProxiPlates (Perki-nElmer) using 50 mM HEPES pH 8.0, 150 mM NaCl, 10 mM MgCl 2 , 1 mM EGTA and 0.01 % Brij-35 (+ 2 % DMSO) as buffer. The concentrations of reagents in 10 μL final assay volume were 3 nM GST-HDAC10, 30 nM "Tubastatin-Alexa647-Tracer" and 0.5 nM Euanti-GST. GST-HDAC10 and LanthaScreen™ Eu-anti-GST Antibody were purchased from Life Technologies, and the Tubastatin-Alexa647-Tracer was synthesized in-house as previously described. [19] An 11-fold 1 : 3-serial dilution of compounds starting at 2 mM was prepared in 384 well pp-plates (Greiner) from 10 mM stocks in DMSO and 1 μL was transferred to assay plates. Nine μL of the reagent-mix was added and the plate was incubated for 1 h at RT before TR-FRET was measured in an EnVision™ plate reader equipped with a TR-FRET Laser module. Sample wells were exited with 3 flashes of the TRF-Europium Laser, and emission was measured at 620 nm and 665 nm to get the 665 nm/620 nm ratio. Percent inhibition was calculated for each well from negative control wells containing 2 % DMSO and positive control wells containing 20 μM SAHA. The resulting dose-response curves were fitted in ActivityBase (IDBS) using a four-parameter logistic model and pIC 50 -values were calculated.
TR-FRET assay (used with dihydroxamic acids): TR-FRET assays were performed in white 384-well plates (4512, Corning) using 50 mM HEPES pH 8.0, 150 mM NaCl, 10 mM MgCl 2 , 1 mM EGTA and 0.01 % Brij-35 as buffer. The concentrations of reagent in 15 μL final assay volume were 5 nM TwinStrep-GST-HDAC10 (preparation described above), 25 nM "Tubastatin-AF647-Tracer" (S15, synthesis in Supporting Information) and 0.1 nM DTBTA-Eu 3 + -labeled Streptactin (synthesis in the Supporting Information). Inhibitors were tested at eight serial dilutions in triplicates ranging from 50 μM-86.7 pM and dosed from 10 mM and 0.1 mM DMSO stock solutions with a D300e Digital Dispenser (Tecan). After drug dosing to the premixed assay reagents in buffer, plates were shaken (800 rpm orbital shaker, 30 s), centrifuged (300 g, 1 min) and incubated at room temperature in the dark for 60 min. TR-FRET was measured with a CLARIOstar (BMG Labtech) plate reader, equipped with TR-FRET filters. Sample wells were excited with 100 flashes and fluorescence emission detected at 665 nm and 620 nm. FRET ratios were calculated from 665 nm/620 nm ratio and normalized for each plate using 50 μM SAHA treated negative controls and uninhibited positive controls. pIC 50 -values were calculated using nonlinear regression log(inhibitor) four parameters least squares fit in Graph-Pad Prism version 7.04 for Windows (GraphPad Software, La Jolla, CA, USA, www.graphpad.com).

Production of mono-clones stably expressing HDAC-nanoBRET proteins:
Plasmids expressing a fusion of HDAC10 with nanoluciferase were obtained from Promega (N2170). HeLa cells (0.75 × 10 6 ) were seeded in a 6 cm dish and after 24 h were transfected with a mix of 10 μg plasmid and 3 μL Fugene in 200 μL OptiMEM. In detail, cells were washed with pre-warmed OptiMEM and subsequently overlaid with 2.3 mL of OptiMEM. After addition of 200 μL transfection mix, cells were incubated for 24 h at 37°C. Cells were than trypsinized and 0.2 × 10 5 cells were seeded into both 10 cm and 15 cm dishes. Transformants were selected with 1 mg/ mL G-418 for 6 days with a media change after 3 days. Clones which formed colonies were picked by rinsing plates with 3 mL Trypsin/EDTA (Sigma T3924) followed by a 2 min incubation with 300 μL Trypsin/EDTA at 37°C. Colonies were then loosened and aspirated with a 10 μL filter tip and transferred to 24-well plates containing selection medium. Clones exhibiting a range of nanoluciferase activities were expanded and selected according to the highest BRET ratio.
BRET assay: The intracellular target engagement assay on HDAC10 was performed as described by the kit manufacturer in a 96-well plate (3600, Corning) format with 1.9 × 10 4 cells per well and a tracer concentration of 0.3 μM. Inhibitors were tested at ten 1 : 4 serial dilutions in triplicates ranging from 129 pM to 40 μM. Drug dosing was performed from 10 mM and 1 mM DMSO stock solutions with a D300e Digital Dispenser (Tecan), DMSO concen-ChemMedChem Full Papers doi.org/10.1002/cmdc.202000149 trations were normalized to 0.5 % for all wells. Assay plates were incubated at 37°C for 2 h followed by measurement of 450 nm and 650 nm luminescence (80 nm bandwidth) at room temperature with a CLARIOstar (BMG Labtech) plate reader 2 min after NanoLuc substrate addition.
LysoTracker assay: SK-N-BE(2)-C cells were seeded into 6-well dishes at a density of 1.5 × 10 5 cells per well. Cells were treated with inhibitor over night at concentrations indicated in the figure and stained the following day for 1 h with 50 nM LysoTracker® Red DND-99 in medium under standard cell culture conditions. Cells were washed with ice-cold RPMI without phenol-red and trypsinized for 3 min at 37°C. Detached cells were centrifuged for 3 min at 8600 g and re-suspended in ice-cold RPMI without phenol red. Mean LysoTracker fluorescence was quantified on a BD FACSCanto II platform using the PE filter setting. Data were normalized to mean LysoTracker fluorescence of solvent (DMSO) treated cells.
Cell viability assay: SK-N-BE(2)C cells were cultured in Dulbecco's modified Eagle medium (DMEM, Lonza, Basel, Switzerland) supplemented with 10 % fetal calf serum (FCS, Sigma-Aldrich) and 1 % non-essential amino acids (NEAA, Lonza). SK-N-BE(2)C cells were seeded from a confluent flask into 96-well plates in 100 μL full growth medium at a density of 5 × 10 3 cells per well the day before treatment. Cells were treated in triplicates with drug concentrations as indicated in the respective figure. Drugs were dosed from 50 mM and 200 mM stock solutions with a D300e Digital Dispenser (Tecan) and DMSO concentrations were normalized to 0.2 % for all wells. Plates were shaken (1000 rpm orbital shaker, 30 s) and incubated under standard culture conditions for 72 h. To each well 20 μL of CellTiter-Blue® reagent (G8081, Promega) was added, plates were shaken (600 rpm orbital shaker, 20 s) and incubated at 37°C overnight. Fluorescence was measured (extinction/emission: 570 nm/590 nm) on a FluoStar Optima (BMG Labtech) plate reader. Cell viability was normalized with untreated positive controls and cell-free negative controls. IC 50 -values were calculated as described in the BRET assay.
Expression, purification, and crystallization of HDAC10. A "humanized" version of HDAC10 was designed by making the A24E and D94A substitutions in D. rerio HDAC10 so as to more closely resemble the active site of human HDAC10. The preparation of this HDAC10 construct using standard PCR mutagenesis techniques will be described separately; purification was achieved as described for the wild-type enzyme. [32][33] For crystallization, the protein solution [10 mg/mL HDAC10, 50 mM HEPES pH 7.5, 300 mM KCl, 5 % glycerol (v/v), and 1 mM tris-(2-carboxyethyl)phosphine (TCEP) was augmented with 2 mM 3a and incubated for 1 h on ice. Trypsin was added (1 : 1000 trypsin:HDAC10) and the mixture was allowed to digest at ambient temperature for 1 h and then filtered using a 0.22 μm centrifuge filter. Utilizing a Mosquito crystallization robot (TTP Labtech), a 100 nL drop of protein solution was added to a 100 nL drop of precipitant solution [0.168 M KH 2 PO 4 , 0.032 M K 2 HPO 4 , and 20 % PEG 3350] and microseeded with crystals of the HDAC10-Tubastatin A complex. The 200 nL sitting drop was equilibrated against 80 μL of precipitant buffer in the well reservoir at 4°C. Crystals appeared within one day.
Crystal structure determination of the HDAC10-3a complex: X-ray diffraction data for the HDAC10-3a complex were collected on NE-CAT beamline 24-ID-C at the Advanced Photon Source (APS), Argonne National Laboratory. Data were indexed by using iMosflm [36] and scaled with Aimless [37] in the CCP4 program suite. [38] The initial electron density map was phased by molecular replacement using Phaser; [39] the structure of Y307F HDAC10 (PDB 5TD7) [32] with solvent and ligand atoms removed was used as a search model. An iterative process of model building using Coot [40] and crystallographic refinement with Phenix [41] yielded the final model of the HDAC10-3a complex. The inhibitor was built into the electron density map during the final stages of refinement. MolProbity [42] was used to validate the final refined structure, which was deposited in the Protein Data Bank (PDB 6VNQ). All data reduction and refinement statistics are recorded in Table S2.