Oxazole-Bridged Combretastatin A-4 Derivatives with Tethered Hydroxamic Acids: Structure–Activity Relations of New Inhibitors of HDAC and/or Tubulin Function

New inhibitors of tubulin polymerization and/or histone deacetylase (HDAC) activity were synthesized by attaching alkyl tethered hydroxamic acid appendages of varying length to oxazole-bridged combretastatin A-4 analogous caps. While their antiproliferative and microtubule disrupting effect was most pronounced for derivatives with short spacers, HDAC inhibition was strongest for those with longer spacers. These findings were further supported by computational methods such as structure-based docking experiments exploring the target interactions of the derivatives with varying linkers. For instance, compounds featuring short four-atom spacers between cap and hydroxamic acid inhibited the growth of various cancer cell lines and human endothelial hybrid cells with IC50 values in the low nanomolar range. In line with their ability to inhibit the microtubule assembly, four- and five-atom spacered hydroxamic acids caused an accumulation of 518A2 melanoma cells in G2/M phase, whereas a compound featuring a six-atom spacer and performing best in HDAC inhibition, induced a G1 arrest in these cells. All these beneficial anticancer activities together with their selectivity for cancer cells over non-malignant cells, point out the great potential of these novel pleiotropic HDAC and tubulin inhibitors as drug candidates for cancer therapy.


Introduction
Histone deacetylases (HDAC) catalyze the deacetylation of ε-N-acetyl lysine residues of histones thus regulating the expression of genes which are important for crucial cellular processes such as chromatin condensation and decondensation (DNA replication, transcription, and repair). Certain HDAC enzymes also modify non-histone proteins such as signal transduction mediators, transcription factors and regulators, as well as structural proteins resulting in modulation of cell growth, differentiation, migration, and angiogenesis [1]. HDACs are overexpressed in various solid tumors, e.g., in gastric cancer, prostate

Chemistry
The 4,5-diaryloxazoles were synthesized via a Van Leusen reaction. The required starting benzaldehydes 1a-c and TosMIC reagents 2a-c were prepared according to literature procedures, i.e., the former via alkylation of isovanillin with the corresponding ethyl ω-bromoalkanoates, and the

Chemistry
The 4,5-diaryloxazoles were synthesized via a Van Leusen reaction. The required starting benzaldehydes 1a-c and TosMIC reagents 2a-c were prepared according to literature procedures, i.e., the former via alkylation of isovanillin with the corresponding ethyl ω-bromoalkanoates, and the latter via dehydration of their tosylmethyl formamide precursors (obtained from reaction of 3-bromo/chloro-4,5-dimethoxybenzaldehyde or 3,4,5-trimethoxybenzaldehyde with toluenesulfinic acid and formamide) [25][26][27][28]. The synthesis of the target hydroxamic acids 4a-i was carried out in two steps. Van Leusen reaction of 1a-c and 2a-c gave the oxazoles 3a-i in moderate yields (Scheme 1). Conversion of the ethyl esters 3a-i to the analogous hydroxamic acids 4a-i was accomplished in moderate to high yields. The target compounds 4a-i were obtained as colorless solids. In addition, carboxylic acid analog 4j was prepared for comparison purposes and obtained from hydrolysis of 3g under basic conditions (Scheme 1). latter via dehydration of their tosylmethyl formamide precursors (obtained from reaction of 3bromo/chloro-4,5-dimethoxybenzaldehyde or 3,4,5-trimethoxybenzaldehyde with toluenesulfinic acid and formamide) [25][26][27][28]. The synthesis of the target hydroxamic acids 4a-i was carried out in two steps. Van Leusen reaction of 1a-c and 2a-c gave the oxazoles 3a-i in moderate yields (Scheme 1). Conversion of the ethyl esters 3a-i to the analogous hydroxamic acids 4a-i was accomplished in moderate to high yields. The target compounds 4a-i were obtained as colorless solids. In addition, carboxylic acid analog 4j was prepared for comparison purposes and obtained from hydrolysis of 3g under basic conditions (Scheme 1).

Biological Evaluation
First, all compounds (4a-j) were tested for their growth inhibitory potential in cancer cell lines. The new derivatives 4a-i, the new carboxylic acid analog 4j and its ethyl-ester 3g were screened in MTT assays for anti-proliferative activity against a panel of six cancer cell lines of four entities as well as against the human endothelial hybrid cell line Ea.Hy926 (Table 1). The bromo derivatives 4d-f were also tested against the non-malignant human dermal fibroblasts HDFa. IC50 values of the known HDACi SAHA and the VDA CA-4 were taken from earlier studies for comparison. Compounds 4a-i led to dose-dependent growth inhibition of all cancer cell lines and the endothelial hybrid cells Ea.Hy926. Carboxylic acid 4j and its ester 3g did not affect the viability of 518A2 melanoma and HT-29 colon carcinoma cells even at concentrations as high as 50 µM, which suggests that the hydroxamate side chain is crucial for the anti-proliferative activity. Scheme 1. Reagents and conditions: (i) K 2 CO 3 , EtOH, reflux, 2 h, 40-53%; (ii) 50% H 2 N-OH in H 2 O, NaOH, CH 2 Cl 2 /MeOH (1:2), r.t., 1 h, 64-93%; (iii) 1 M aq. NaOH, MeOH, r.t., 24 h, 68%.

Biological Evaluation
First, all compounds (4a-j) were tested for their growth inhibitory potential in cancer cell lines. The new derivatives 4a-i, the new carboxylic acid analog 4j and its ethyl-ester 3g were screened in MTT assays for anti-proliferative activity against a panel of six cancer cell lines of four entities as well as against the human endothelial hybrid cell line Ea.Hy926 (Table 1). The bromo derivatives 4d-f were also tested against the non-malignant human dermal fibroblasts HDFa. IC 50 values of the known HDACi SAHA and the VDA CA-4 were taken from earlier studies for comparison. Compounds 4a-i led to dose-dependent growth inhibition of all cancer cell lines and the endothelial hybrid cells Ea.Hy926. Carboxylic acid 4j and its ester 3g did not affect the viability of 518A2 melanoma and HT-29 colon carcinoma cells even at concentrations as high as 50 µM, which suggests that the hydroxamate side chain is crucial for the anti-proliferative activity.
On average, the CA-4 resistant HT-29 colon carcinoma and the multi-drug resistant MCF-7 Topo mamma carcinoma cells were least sensitive to 4a-i [29,30]. In contrast, 4a-i were most active against 518A2 melanoma, HCT-116 colon carcinoma, and endothelial hybrid cells Ea.Hy926. The low IC 50 values of the compounds 4a-i against Ea.Hy926 cells are indicative of a potential application as a vascular disruptive agent like the parent CA-4. Moreover, the bromo substituted derivatives 4d-f showed a distinct selectivity for cancer and endothelial cells over non-malignant human dermal fibroblasts HDFa. The chloro substituted compounds 4a-c and the bromo substituted compounds 4d-f showed an interesting structure-activity relationship (SAR). The anti-proliferative activity increased with decreasing linker length when going from caproic acid derivatives 4c and 4f over valeric acid derivatives 4b and 4e to butyric acid derivatives 4a and 4d. Interestingly, the trimethoxy derivatives 4g-i did not fit in this SAR since 4i was on average more cytotoxic than 4h. Several earlier studies had shown that the substitution of an m-methoxy group at the A-ring of CA-4 derivatives by halide increases their activity. We now observed a similar phenomenon since the chloro substituted derivatives 4a-c and the bromo substituted analogs 4d-f were superior to the trimethoxy derivatives 4g-i. Additionally, we determined the IC 50 values of the test compounds when applied to 518A2 melanoma cells for different incubation periods (24 and 72 h). In the case of the most strongly anti-proliferative compounds (4a, 4d, and 4g), the IC 50 values after 72 h were about eight-fold lower compared with those after 24 h. Since most of the investigated cell lines have division periods of 20-30 h, we assume that the test compounds exert their effect by blocking the cell division and by triggering apoptosis [31,32]. Earlier publications by our group disclosed a significant tumor selectivity of the control compounds SAHA and CA-4 (i.e., low activity against CHF/chicken heart fibroblast cells) which can explain the observed tumor selectivity of compounds 4d-f [33,34].
Next, we investigated how the length of the linker of the test compounds may influence cytoskeletal components, the inhibition of HDAC1 and HDAC6, and cell cycle progression. These analyses were performed only with the bromo substituted derivatives 4d-f which showed slightly higher anti-proliferative activity on average compared to their chloro or methoxy congeners. At first, the new hybrid compounds 4d-f were tested for their potential inhibition of the microtubule assembly, which is a typical feature of the parent compound CA-4. MTT assays already revealed some CA-4 characteristics such as selectivity for EA.Hy926 endothelial hybrid cells and a reduced efficacy against HT-29 colon carcinoma cells [33]. The effect of the test compounds on the polymerization of tubulin was determined in vitro using purified pig brain tubulin ( Figure 2). An amount of 10 µM of 4d inhibited the polymerization of pig brain tubulin nearly completely, while 4e exhibited a merely moderate inhibitory effect and 4f virtually none.  [31,32]. Earlier publications by our group disclosed a significant tumor selectivity of the control compounds SAHA and CA-4 (i.e., low activity against CHF/chicken heart fibroblast cells) which can explain the observed tumor selectivity of compounds 4d-f [33,34]. Next, we investigated how the length of the linker of the test compounds may influence cytoskeletal components, the inhibition of HDAC1 and HDAC6, and cell cycle progression. These analyses were performed only with the bromo substituted derivatives 4d-f which showed slightly higher anti-proliferative activity on average compared to their chloro or methoxy congeners. At first, the new hybrid compounds 4d-f were tested for their potential inhibition of the microtubule assembly, which is a typical feature of the parent compound CA-4. MTT assays already revealed some CA-4 characteristics such as selectivity for EA.Hy926 endothelial hybrid cells and a reduced efficacy against HT-29 colon carcinoma cells [33]. The effect of the test compounds on the polymerization of tubulin was determined in vitro using purified pig brain tubulin ( Figure 2). An amount of 10 µM of 4d inhibited the polymerization of pig brain tubulin nearly completely, while 4e exhibited a merely moderate inhibitory effect and 4f virtually none. These results are in line with the anti-proliferative activity pattern of the compounds and were additionally confirmed on a cellular level by immunostaining of alpha-tubulin in 518A2 melanoma cells ( Figure 3). Caproic acid derivative 4f, showing the highest IC50 values in MTT assays, did not affect the microtubule cytoskeleton even at concentrations as high as 4 µM. In contrast, 4e eroded the These results are in line with the anti-proliferative activity pattern of the compounds and were additionally confirmed on a cellular level by immunostaining of alpha-tubulin in 518A2 melanoma cells ( Figure 3). Caproic acid derivative 4f, showing the highest IC 50 values in MTT assays, did not affect the microtubule cytoskeleton even at concentrations as high as 4 µM. In contrast, 4e eroded the highly organized microtubule network, but left some intact clusters especially around the nuclei whereas 0.5 µM of 4d was enough to cause a complete disruption of the microtubule cytoskeleton. Similar alterations of the cytoskeleton of endothelial Ea.Hy926 cells were observed upon treatment with 0.2 µM of 4d for 24 h ( Figure S1, Supporting Information). The deacetylation of tubulin by compounds 4d and 4f is presented below. In addition, elevated levels of reactive oxygen species (ROS, Figure S2, Supporting Information), which are known to trigger apoptosis and reverse chemoresistance in tumors, were observed in 518A2 melanoma cells (4d: 241% ± 17; 4e: 230 ± 31; 4f: 198 ± 24). Again, as already observed for anti-proliferative activity the ability to elevate ROS levels decreases with increasing linker length.  We also investigated the bromo derivatives 4d-f with different linker lengths for their inhibitory effect on the deacetylation capacity of recombinant human HDAC1 and HDAC6 (Table 2). Contrary to the inhibition of tubulin polymerization and cell proliferation, which decreased with growing linker length, the HDAC inhibition increased with linker length. Compound 4d, the most cytotoxic compound in this row featuring a four-atom spacer, showed only moderate HDAC6 inhibition (IC50: 13.8 ± 0.2 µM). Compound 4e, carrying a five-atom linker, had a distinctly lower IC50 value (3.5 ± 0.1 µM), whereas 4f, the compound with a six-atom linker, had the lowest IC50 value of this triad (0.32 ± 0.02 µM), which was even slightly lower than that of the known HDAC6 selective inhibitor tubacin (0.38 ± 0.03 µM). Concerning HDAC1 inhibition, 4d and 4e showed similar IC50 values (4.0 ± 0.1 and 3.8 ± 0.1 µM) whereas 4f was again the most potent compound (0.49 ± 0.05 µM). Unlike HDAC1 which We also investigated the bromo derivatives 4d-f with different linker lengths for their inhibitory effect on the deacetylation capacity of recombinant human HDAC1 and HDAC6 (Table 2). Contrary to the inhibition of tubulin polymerization and cell proliferation, which decreased with growing linker length, the HDAC inhibition increased with linker length. Compound 4d, the most cytotoxic compound in this row featuring a four-atom spacer, showed only moderate HDAC6 inhibition (IC 50 : 13.8 ± 0.2 µM). Compound 4e, carrying a five-atom linker, had a distinctly lower IC 50 value (3.5 ± 0.1 µM), whereas 4f, the compound with a six-atom linker, had the lowest IC 50 value of this triad (0.32 ± 0.02 µM), which was even slightly lower than that of the known HDAC6 selective inhibitor tubacin (0.38 ± 0.03 µM). Concerning HDAC1 inhibition, 4d and 4e showed similar IC 50 values (4.0 ± 0.1 and 3.8 ± 0.1 µM) whereas 4f was again the most potent compound (0.49 ± 0.05 µM). Unlike HDAC1 which is found in the nucleus of cells where it is responsible for the eponymous deacetylation of histones, HDAC6 locates predominantly in the cytoplasm and has several targets including α-tubulin, HSP90, cortactin, and β-catenin [35,36]. The inhibition of HDAC6 induces hyperacetylation of these molecules resulting in a reduction of cell motility, and proliferation, and eventually induces cell death [37]. The ability of compound 4f to inhibit HDAC6 was confirmed by western blot analyses ( Figure 4) as well as by immunofluorescence staining of acetyl-alpha-tubulin in 518A2 melanoma cells ( Figures  S3 and S4, Supporting Information). In both experiments, treatment of the cells with 4f caused a distinct increase of acetyl-alpha-tubulin. Thus, a distinct difference between 4d and 4f concerning their effects on the microtubule cytoskeleton became visible. While 4d destroys the microtubules in line with its high tubulin polymerization inhibitory activity, 4f did not destroy the microtubules yet enhanced the acetylation grade of microtubules due to its strong HDAC6 inhibition.   Imidazole-bridged CA-4 derivatives carrying hydroxamic acid appendages had previously been found to induce alterations of the actin cytoskeleton, such as augmented formation of stress fibers to the effect of an impaired cell motility [34,38]. Such alterations are typical reactions to microtubule destabilization and hyperacetylation of cortactin as a consequence of HDAC6 inhibition [39]. Thus, we investigated the bromo derivatives 4d-f for their effect on the actin cytoskeleton of 518A2 melanoma cells ( Figure 5). Even though their effects on the microtubules of these cells differed, all of them induced the formation of actin stress fibers which traversed the whole cell body, while the control cells showed only filamentous actin in the periphery. The associated anti-migratory effects of compounds 4d-f were then tested in so-called wound healing assays. In this assay a strip of cells is scratched off a confluent grown cell monolayer of 518A2 cells, followed by monitoring the gapclosing process operating not by proliferation but by active migration (Figures S5 and S6, Supporting Information). The re-closure of the scratch area was significantly retarded in samples treated with compounds 4d-f for 24 h (38-45% wound closure) compared to vehicle treated control cells (63% wound closure). Imidazole-bridged CA-4 derivatives carrying hydroxamic acid appendages had previously been found to induce alterations of the actin cytoskeleton, such as augmented formation of stress fibers to the effect of an impaired cell motility [34,38]. Such alterations are typical reactions to microtubule destabilization and hyperacetylation of cortactin as a consequence of HDAC6 inhibition [39]. Thus, we investigated the bromo derivatives 4d-f for their effect on the actin cytoskeleton of 518A2 melanoma cells ( Figure 5). Even though their effects on the microtubules of these cells differed, all of them induced the formation of actin stress fibers which traversed the whole cell body, while the control cells showed only filamentous actin in the periphery. The associated anti-migratory effects of compounds 4d-f were then tested in so-called wound healing assays. In this assay a strip of cells is scratched off a confluent grown cell monolayer of 518A2 cells, followed by monitoring the gap-closing process operating not by proliferation but by active migration (Figures S5 and S6, Supporting Information). The re-closure of the scratch area was significantly retarded in samples treated with compounds 4d-f for 24 h (38-45% wound closure) compared to vehicle treated control cells (63% wound closure). Stress fiber formation in combination with microtubule destruction frequently leads to the arrest of cells in the G2/M phase of the cell cycle. By contrast, HDAC inhibition typically induces an arrest of cells in the G1 phase [40][41][42][43][44][45]. To investigate whether microtubule destabilization or HDAC inhibition of the test compounds 4d-f prevails on cell cycle regulation, cell cycle arresting effects were tested in 518A2 melanoma cells by FACS analysis (Figure 6, Table 3). As expected, the strongly microtubule-destabilizing compounds 4d and 4e led to an accumulation of 518A2 cells in G2/M phase. In contrast, the stronger HDAC inhibitor 4f, which lacks microtubule destabilizing activity, induced a G1 phase arrest in the investigated melanoma cells.
The effects of 4d on the cell cycle regulatory proteins p21, p27, and cyclin D1 in 518A2 melanoma cells were investigated (Figure 7). At higher doses 4d increased the level of cyclin D1 which was comparable with the effect on cyclin D1 by SAHA. It seems that the HDAC inhibitory properties of 4d at higher concentrations caused the induction of cyclin D1. In addition, 4d distinctly suppressed p21 expression while SAHA showed no effects on p21. It is conceivable that the observed p21 suppression is associated with the cytoskeleton targeting of 4d which is also in line with the G2/M arrest caused by 4d in 518A2 cells. In contrast to that, 4f seemingly upregulated p21 expression in a dose-dependent way in congruence to SAHA.
Finally, compound 4d was tested concerning in vivo applicability and the toxicity of compound 4d to mice was investigated ( Figure S7, Supporting Information). High doses of 4d (1 × 100 mg/kg i.p., 1 × 200 mg/kg orally) were tolerated well by the treated mice and they showed no signs of toxicity (i.e., no weight loss, weight changes of maximal 5%, normal behavior). Thus, a more thorough in vivo Stress fiber formation in combination with microtubule destruction frequently leads to the arrest of cells in the G2/M phase of the cell cycle. By contrast, HDAC inhibition typically induces an arrest of cells in the G1 phase [40][41][42][43][44][45]. To investigate whether microtubule destabilization or HDAC inhibition of the test compounds 4d-f prevails on cell cycle regulation, cell cycle arresting effects were tested in 518A2 melanoma cells by FACS analysis ( Figure 6, Table 3). As expected, the strongly microtubule-destabilizing compounds 4d and 4e led to an accumulation of 518A2 cells in G2/M phase. In contrast, the stronger HDAC inhibitor 4f, which lacks microtubule destabilizing activity, induced a G1 phase arrest in the investigated melanoma cells.
The effects of 4d on the cell cycle regulatory proteins p21, p27, and cyclin D1 in 518A2 melanoma cells were investigated (Figure 7). At higher doses 4d increased the level of cyclin D1 which was comparable with the effect on cyclin D1 by SAHA. It seems that the HDAC inhibitory properties of 4d at higher concentrations caused the induction of cyclin D1. In addition, 4d distinctly suppressed p21 expression while SAHA showed no effects on p21. It is conceivable that the observed p21 suppression is associated with the cytoskeleton targeting of 4d which is also in line with the G2/M arrest caused by 4d in 518A2 cells. In contrast to that, 4f seemingly upregulated p21 expression in a dose-dependent way in congruence to SAHA.
Finally, compound 4d was tested concerning in vivo applicability and the toxicity of compound 4d to mice was investigated ( Figure S7, Supporting Information). High doses of 4d (1 × 100 mg/kg i.p., 1 × 200 mg/kg orally) were tolerated well by the treated mice and they showed no signs of toxicity (i.e., no weight loss, weight changes of maximal 5%, normal behavior). Thus, a more thorough in vivo investigation of 4d in suitable tumor xenograft models is recommended due to the manageable toxicity profile of this compound. investigation of 4d in suitable tumor xenograft models is recommended due to the manageable toxicity profile of this compound.

In Silico Evaluation
Structure-based docking was used to explore the interactions of the different compounds with the respective target proteins on a molecular level. Molecular docking is an efficient technique for calculating the binding modes of a compound and estimating their binding affinities. Docking in general is a selection and optimization process, trying to find the best fit of a molecule in the binding site of a protein according to a scoring function [46].
To evaluate the role of the varying linker-length, the CA-4 derivative (Cap) and the hydroxamic acid group (ZBG) for the binding affinity to tubulin, docking studies were performed for the structures 4d-f. In the tubulin structure 5LYJ, used for the docking studies, the microtubuledestabilizing agent CA-4 binds to the colchicine site of the tubulin β-chain close to the interface of the neighboring tubulin α-chain, which restrains this binding site. The CA-4 site is a buried hydrophobic pocket shaped by residues Val238, Cys241, Leu242, Leu248, Ala250, Leu255, Ala316, Ile318, Ala354, and Ile378 [47].
The docking studies for tubulin showed almost identical binding positions of the CA-4 cap of the molecules 4d-f in the hydrophobic colchicine binding pocket similar to the original ligand CA-4 ( Figure 8A). The linker extended towards the tubulin α-chain cap and the hydroxamic acid was predicted to form hydrogen bonds with Asn349 from the β-chain (4d-f) and potentially with Ser178 from the α-chain (4f). As summarized in Table 4, the estimated affinities slightly decreased with increasing linker length, a finding that is in agreement with the wet-lab results. Another observation was that the longer the linker, the more twisted it needed to be to fit into the capped cavity. Thus, the chain had to adopt torsions which are only seldom observed in crystal structures and which are considered as unfavorable; see Figure 8B [48,49].

In Silico Evaluation
Structure-based docking was used to explore the interactions of the different compounds with the respective target proteins on a molecular level. Molecular docking is an efficient technique for calculating the binding modes of a compound and estimating their binding affinities. Docking in general is a selection and optimization process, trying to find the best fit of a molecule in the binding site of a protein according to a scoring function [46].
To evaluate the role of the varying linker-length, the CA-4 derivative (Cap) and the hydroxamic acid group (ZBG) for the binding affinity to tubulin, docking studies were performed for the structures 4d-f. In the tubulin structure 5LYJ, used for the docking studies, the microtubule-destabilizing agent CA-4 binds to the colchicine site of the tubulin β-chain close to the interface of the neighboring tubulin α-chain, which restrains this binding site. The CA-4 site is a buried hydrophobic pocket shaped by residues Val238, Cys241, Leu242, Leu248, Ala250, Leu255, Ala316, Ile318, Ala354, and Ile378 [47].
The docking studies for tubulin showed almost identical binding positions of the CA-4 cap of the molecules 4d-f in the hydrophobic colchicine binding pocket similar to the original ligand CA-4 ( Figure 8A). The linker extended towards the tubulin α-chain cap and the hydroxamic acid was predicted to form hydrogen bonds with Asn349 from the β-chain (4d-f) and potentially with Ser178 from the α-chain (4f). As summarized in Table 4, the estimated affinities slightly decreased with increasing linker length, a finding that is in agreement with the wet-lab results. Another observation was that the longer the linker, the more twisted it needed to be to fit into the capped cavity. Thus, the chain had to adopt torsions which are only seldom observed in crystal structures and which are considered as unfavorable; see Figure 8B [48, 49].  To explore reasons for the differences in the HDAC inhibitory potency of the compounds on a molecular level, computational docking studies were performed. The calculations were based on the X-ray structures 5ICN (HDAC1) and 5EDU (HDAC6) and were carried out for the compounds 4d-f as well as for vorinostat and the respective co-crystallized ligands (see methods section). Generally, HDACs feature an active site with a relatively narrow tunnel pointing towards the buried catalytic zinc(II) cation to which the hydroxamic acid of the natural substrates and inhibitors such as vorinostat binds. The two HDAC structures used in this study exhibit very similar binding sites and can be superimposed with a low backbone RMSD of 1.9Å (superposition calculated with PyMol, see Figure  9A).  To explore reasons for the differences in the HDAC inhibitory potency of the compounds on a molecular level, computational docking studies were performed. The calculations were based on the X-ray structures 5ICN (HDAC1) and 5EDU (HDAC6) and were carried out for the compounds 4d-f as well as for vorinostat and the respective co-crystallized ligands (see methods section). Generally, HDACs feature an active site with a relatively narrow tunnel pointing towards the buried catalytic zinc(II) cation to which the hydroxamic acid of the natural substrates and inhibitors such as vorinostat binds. The two HDAC structures used in this study exhibit very similar binding sites and can be superimposed with a low backbone RMSD of 1.9Å (superposition calculated with PyMol, see Figure 9A). The best poses calculated with SeeSAR for all compounds (including re-docking of the cocrystallized ligands and vorinostat) showed a similar orientation of the linker threading through the narrow tunnel and the hydroxamic acid chelating the zinc(II) cation ( Figure 9B). While the estimated affinities are all in a similar range, the values suggest that a longer linker is more favorable (Tables 5  and 6), which is in accordance with the experimental results ( Table 2). This could be due to the better fit through the long (e.g., ~10 Å in HDAC6) and narrow active site tunnel, which would allow the hydroxamic acid group with longer linker length to reach the zinc ion more easily [50]. In contrast, the distance between the hydroxamic acid terminus and the first benzene attached to the linker in 4d is only ~8 Å, which might make it difficult to find a good hydroxamic acid fit without causing a clash of the bulky CA-4 derived capping group with the protein. In all docked compounds, the hydroxamic acid group formed hydrogen bonds with the buried amino acids His610 and Gly619 of the HDAC6 structure 5EDU (co-crystallized to trichostatin A) as well as a bond to the zinc ion [50]. Generally, the hydroxamic acid and the more buried part of the linker of all three compounds 4d-f overlap also with the respective part of the co-crystallized The best poses calculated with SeeSAR for all compounds (including re-docking of the co-crystallized ligands and vorinostat) showed a similar orientation of the linker threading through the narrow tunnel and the hydroxamic acid chelating the zinc(II) cation ( Figure 9B). While the estimated affinities are all in a similar range, the values suggest that a longer linker is more favorable (Tables 5  and 6), which is in accordance with the experimental results ( Table 2). This could be due to the better fit through the long (e.g.,~10 Å in HDAC6) and narrow active site tunnel, which would allow the hydroxamic acid group with longer linker length to reach the zinc ion more easily [50]. In contrast, the distance between the hydroxamic acid terminus and the first benzene attached to the linker in 4d is only~8 Å, which might make it difficult to find a good hydroxamic acid fit without causing a clash of the bulky CA-4 derived capping group with the protein. In all docked compounds, the hydroxamic acid group formed hydrogen bonds with the buried amino acids His610 and Gly619 of the HDAC6 structure 5EDU (co-crystallized to trichostatin A) as well as a bond to the zinc ion [50]. Generally, the hydroxamic acid and the more buried part of the linker of all three compounds 4d-f overlap also with the respective part of the co-crystallized trichostatin A ( Figure 9C, left panel). The capping group, which is rather solvent exposed, was more variable in its position, but may form hydrogen bonds with His651 and Phe680.
The docking results for HDAC1 structure 5ICN were more diverse. This could be due to the fact that the structure is co-crystallized with a peptide inhibitor (H4K16Hx) and the loop around residue Asp99 undergoes a significant rearrangement to allow binding of the peptide [51]. Note that there are only two HDAC1 structures available to date (5ICN and 4BKX), of which 4BKX is an apo structure, so 5ICN was the only available complex structure. Reasonable poses for all compounds could be generated with LeadIT, whereas the optimization in SeeSAR was successful for all compounds but 4d ( Figure 9C, right panel). Furthermore, the binding poses of the three compounds 4d-f and vorinostat differed more in the hydrogen and metal bonds formed by the hydroxamic acid. While the hydroxamic acid group of 4f was predicted to form hydrogen bonds with Tyr303 and His141, it was predicted to only form one hydrogen bond with Gly301 in 4e. The capping group might form H-Bonds with Asn95 and Gly149. While the results were more distinct for HDAC6 (where a good protein-ligand complex structure was available), generally, the docking analysis showed that the molecules with longer linker length tend to have a higher affinity to HDAC1 and HDAC6.

Discussion
Recently, we presented conjugates of imidazoles and hydroxamic acids which combine HDAC inhibition with cytoskeletal modulation [34,38,52]. However, conjugates with imidazoles derived from CA-4 had lost crucial CA-4 typical properties such as inhibition of the polymerization of tubulin. Herein, we introduced a new series of conjugates of oxazole-bridged CA-4 and hydroxamic acids that show dual tubulin and HDAC interference which mainly depends on the length of the linker connecting these two fragments. In terms of cytotoxicity, inhibition of tubulin polymerization, and generation of ROS, the potency of these conjugates grows with decreasing linker length. The strong anti-proliferative effect especially of the derivatives with the shortest linkers 4a, 4d, and 4g is probably due to their interference with formation of the spindle apparatus during mitosis. The derivatives with longer tethers increasingly lose this property resulting in a distinctly reduced anti-proliferative effect which was confirmed by in silico studies and was attributed to the increasingly tensed fitting of the linkers in the capped cavity.
By destabilizing the microtubules, the new conjugates affect the cell division and probably the integrity of endothelial cell layers. Moreover, they showed a distinct specificity for endothelial hybrid cells Ea.Hy926 over non-malignant cells HDFa. Both properties underline the great potential of these derivatives as VDAs which trigger the collapse of the leaky and low-quality blood vessels of tumors by destabilizing their endothelial lining, initiating necrosis in the core of the tumor [19]. An assessment of the vascular-disrupting activity of the new compounds is currently underway.
The HDAC inhibitory effect of the new conjugates showed the opposite dependency on linker length. The IC 50 values for inhibition of both HDAC1 and HDAC6 decreased with increasing linker length reflecting the fact that the active site of the zinc-dependent HDACs is accessible for HDACi only by a long narrow hydrophobic channel. We rationalized this trend by in silico experiments.
The observed G1-arrest of 518A2 cells by 4f matches with similar observations for other HDACi. The MDA-typical G2/M arrest by derivatives 4d and 4e is in line with their pronounced microtubule disrupting effect which probably overrides their HDACi activity. The stress fiber induction as well as the anti-migratory activity observed for 4d-f was independent of linker length. Together with the lack of toxicity in mice, compound 4d appears to be an especially promising drug candidate.
In this work we were able to demonstrate the anti-proliferative, cell-cycle arresting, microtubule-destabilizing and actin stress fiber-inducing effects of novel CA-4 analogues with the potential to inhibit histone deacetylases. The results are indicative of the great potential of this new class of compounds which probably affect the tumor vasculature either by inhibiting angiogenesis or by disruption of already established tumor blood vessels. Investigations of these anti-vascular effects as well as their in vivo activity are currently underway.

Materials
The known starting compounds 1a and 2a-c were prepared according to literature procedures [25][26][27]. The new compounds 1b and 1c were synthesized analogously to 1a (see below). All other starting compounds and reagents were purchased from Sigma-Aldrich.

General Procedure for the Synthesis of Intermediates 1b and 1c
Isovanillin (152 mg, 1.0 mmol) was dissolved in MeCN (5 mL) and cesium carbonate (652 mg, 2.0 mmol) was added. After stirring at 90 • C for 0.5 h, ethyl 5-bromovalerate (328 µL, 2.0 mmol) or ethyl 6-bromohexanoate (356 µL, 2.0 mmol), respectively, was added and the reaction mixture was stirred at 90 • C for 2 h. After filtration through celite, the solvent was removed under reduced pressure and the residue was washed twice with n-hexane. The resulting oil was dried in vacuum and used for the next step without further purification.

General Procedure for the Synthesis of Compounds 4a-i
The corresponding compound 3 (0.20 mmol) was dissolved in CH 2 Cl 2 /MeOH (9 mL, 1:2), hydroxylamine (50% in water, 0.5 mL, 15 mmol) and NaOH (200 mg, 5 mmol) were added and the reaction mixture was stirred at room temperature for 1 h. The solvent was removed, the residue was dissolved in water and adjusted to pH 7-8 with acetic acid. The aqueous phase was extracted with ethyl acetate (2 × 50 mL), dried over Na 2 SO 4 and concentrated in vacuum. The solid residue was recrystallized from CH 2 Cl 2 /n-hexane.

Tubulin Polymerization Assay
An amount of 50 µL of Brinkley's buffer 80 (BRB80) supplemented with 20% glycerol and 3 mM GTP was given in a black 96-well half-area plate with clear bottom. Then, test compounds 4d-f (final concentration: 5 µM, or 10 µM) or vehicle (DMSO) were added. After adding 50 µL tubulin in BRB80 (10 mg/mL) was pipetted in the wells and immediately placed in a pre-heated microplate reader (Tecan). The polymerization was determined turbidimetrically at 37 • C by measuring the absorption at 340 nm for 120 min in intervals of 5 min. Values were normally distributed for each group (Shapiro-Wilk test, p > 0.05) and there was homogeneity of variance (Levene's test, p > 0.05). The one-tailed Dunnett post-hoc test revealed significant inhibition of tubulin polymerization (p ≤ 0.001) for CA-4 and 4d compared to controls.

Immunofluorescence Staining of Microtubule Cytoskeleton
518A2 melanoma cells (5 × 10 4 cells/mL, 500 µL/well) or Ea.Hy926 endothelial hybrid cells (7.5 × 10 4 cells/mL, 500 µL/well) were seeded in 24-well plates on small glass coverslips and grown for 24 h. Then, the cells were exposed to the test compounds for 24 h. After washing the cells once with PBS, they were fixed in 3.7% formaldehyde in PBS (20 min, rt), and permeabilized and blocked in 1% BSA and 0.1% triton X-100 in PBS (30 min, rt). Then, the cells were incubated with monoclonal mouse anti-alpha-tubulin antibody (1 h, 37 • C). After washing the cells for three times with PBS, the cells were exposed to the secondary anti-mouse 488 antibody conjugate (1 h, rt, in the dark). Then, the cells were washed for three times with PBS and once with water. The glass coverslips were mounted in 4-88 based mounting medium supplemented with 1 µg/mL DAPI for counterstaining the nuclei and 2.5% DABCO. Alterations of the microtubule were documented by a Zeiss Imager A1 AX10 fluorescence microscope (400× magnification).

HDAC Inhibition
The HDAC inhibitory potential of the novel compounds was determined by utilizing the deacetylase activity of recombinant human HDAC1 (HDAC1 Inhibitor Screening Assay Kit; Cayman Chemicals) or HDAC6 (HDAC6 Inhibitor Screening Kit (Fluorometric); Biovision) towards the corresponding synthetic acetylated-peptide substrates resulting in the release of a fluorescent product. The assays were performed according to manufacturer's description of the in commercially available assay kits. The fluorescence intensity (HDAC1: λ ex = 352 nm, λ em = 452 nm; HDAC6: λ ex = 380 nm, λ em = 510 nm) as a measure of enzyme activity was measured at 37 • C with a microplate reader (Tecan). The IC 50 values were derived from dose-response curves and are expressed as the means ± SD of two independent experiments. A two-tailed t-test was performed, revealing significant differences (p < 0.0001) in HDAC6-inhibition for 4e-f compared with 4d as well as for 4d and 4f compared with 4e and in HDAC1-inhibition for 4d-e compared with 4f. The difference in HDAC1 IC 50 values of 4d compared with 4e were not significant.
For the cell cycle protein blots, 518A2 melanoma cells (2 × 106 cells/mL, 10 mL/well) were grown in 10 cm dishes for 24 h and then incubated with 4d (0.15-4 µM) or SAHA (5 µM) for 24 h. After harvesting and lysing of the cells, western blotting was performed as described before [54]. For subsequent analysis, the protein-loaded membranes were incubated with antibodies directed against p21 (Abcam ab109199), p27 (Abcam ab109199), cyclin D1 (Cell Signaling 29225) and GADPH (Santa Cruz sc25778) in 1:1000 dilutions. After incubation with horseradish peroxidase-coupled anti-IgG antibodies (1:10,000; Amersham), the blot was developed using the Celvin-S developer (Biostep) and the software SnapAndGo 1.8.1. 4.7.7. Immunofluorescence Staining of F-Actin 518A2 melanoma cells (5 × 10 4 cells/mL, 500 µL/well) were seeded in 24-well plates on small glass coverslips and grown for 24 h at 37 • C. Then, the cells were treated with 4d (0.5 µM), 4e (1.5 µM), 4f (4 µM), and vehicle (DMSO) for 24 h. After washing the cells once with PBS at 37 • C, they were fixed with 3.7% formaldehyde solution in PBS (pH 7.0) for 10 min at room temperature. Then, the cells were washed with PBS and permeabilized with 0.5% Triton X-100 in PBS for 5 min at room temperature. After washing with PBS, the cells were incubated with 200 µL of a 0.1 µM Acti-stain TM 488 phalloidin solution in PBS for 30 min at room temperature in the dark. Then, the cells were washed three times with PBS and once with water, and the glass coverslips mounted in 4-88-based mounting medium containing 1 µg/mL DAPI for counterstaining the nuclei and 2.5% DABCO. The effects on the actin cytoskeleton were documented by a Zeiss Imager A1 AX10 fluorescence microscope (400× magnification).

Cell Cycle Analysis
The 518A2 melanoma cells (3 mL/well; 5 × 10 4 cells/mL) were grown on 6-well tissue culture plates for 24 h. After treatment with 4d (150, and 200 nM), 4e (0.8, and 1 µM), 4f (2, and 3 µM), or DMSO (control) for another 24 h, cells were fixed (70% EtOH, 24 h, 4 • C), washed with PBS and incubated with propidium iodide (PI; Carl Roth) staining solution (50 µg/mL PI, 0.1% sodium citrate, 50 µg/mL RNase A in PBS) for 30 min at 37 • C. The fluorescence intensity of 10,000 single cells was measured at λ em = 570 nm (λ ex = 488 nm laser source) with a Beckmann Coulter Cytomics FC 500 flow cytometer. The percentages of cells in the different phases of the cell cycle (G1, S and G2/M phase) were determined using the CXP Analysis software (Beckmann Coulter). The percentage of apoptotic cells was derived from sub-G1 peaks. Data was normally distributed for each group (Shapiro-Wilk test, p > 0.05). Since homogeneity of variance (Levene's test, p < 0.05) was violated, a correction was calculated for Anova (Welch test). The two-tailed Games-Howell's post-hoc test revealed the significant alteration (p ≤ 0.05) of the population of cells in G1 in 518A2 melanoma cells treated with 4d-f compared with vehicle-treated controls.

In Vivo Toxicity
In vivo toxicity of 4d was studied in nude mice (Charles River Laboratories, Sulzfeld, Germany). These experiments were carried out following the institutional guidelines. 4d was formulated in 10% Tween80/10% ethanol/80% saline for administration. One mouse (32 g) was treated once with 1 × 100 mg/kg body weight (i.p.), another mouse (30 g) was treated once with 1 × 200 mg/kg body weight (orally) of 4d and then both mice were observed for two weeks. The body weight of the mice was assessed daily under therapy.

Proteins and Compound Structures
The protein structures for tubulin, HDAC1, and HDAC6 used in the docking studies were obtained from the Protein Data Bank (Table 7, PDB, www.rcsb.org, accessed 6 March 2018) [55]. One representative per protein was selected for the docking studies based on the following criteria: The protein should be co-crystallized with a ligand most similar to the ligands from which the studied hybrid compounds 4d-f were derived (combretastatin A-4 for tubulin and hydroxamic acids for HDACs) [56]. Among these, the structures with the best resolutions were selected. To generate initial predictions of binding positions, called docking poses, FlexX (included in LeadIT 2.2.0, which was kindly provided by BioSolveIT, www.biosolveit.de/LeadIT), was used [57]. The PDB structures were prepared as follows: For tubulin, chains A and B of 5LYJ, for HDAC1, chain B of 5ICN, and for HDAC6, chain A of 5EDU were chosen [47,50,51]. The binding pockets were defined with the co-crystallized ligands as a reference ligand, including all amino acids within a radius of 10 Å for the HDAC structures and 6.5 Å for tubulin. The metal coordination of the zinc ion in both HDAC structures was set to 'spherical' and zinc was defined as an essential pharmacophore for guided docking. All other settings for the receptor definition were used as default. The docking library contained compounds 4d-f as well as the crystal structure's original ligand. For HDAC1 and HDAC6 vorinostat, a known HDAC inhibitor, was added to the library. The docking strategy, scoring and chemical parameters were kept as default. Only the maximum number of solutions per iteration and the maximum number of solutions per fragmentation were increased to 1000 steps each.

Pose Optimization with SeeSAR
Final pose optimization and affinity estimation was performed with SeeSAR 6.1, also kindly provided by BioSolveIT (www.biosolveit.de/SeeSAR). The best poses from the LeadIT docking were imported and were used for binding site definition. For each compound, 10 new poses were generated, and all were evaluated with the built in HYDE scoring function [58]. The poses with the best estimated affinities were chosen for further analysis.

Additional Software Used
PyMOL was used to analyze, compare, and visualize the binding pose predictions, as well as to create the 3D images [59].