Antifungal benzimidazoles disrupt vasculature by targeting one of nine β-tubulins

Thiabendazole (TBZ) is an FDA-approved benzimidazole widely used for its antifungal and antihelminthic properties. We showed previously that TBZ is also a potent vascular disrupting agent and inhibits angiogenesis at the tissue level by dissociating vascular endothelial cells in newly formed blood vessels. Here, we uncover TBZ’s molecular target and mechanism of action. Using human cell culture, molecular modeling, and humanized yeast, we find that TBZ selectively targets only 1 of 9 human β-tubulin isotypes (TUBB8) to specifically disrupt endothelial cell microtubules. By leveraging epidemiological pesticide resistance data and mining chemical features of commercially used benzimidazoles, we discover that a broader class of benzimidazole compounds, in extensive use for 50 years, also potently disrupt immature blood vessels and inhibit angiogenesis. Thus, besides identifying the molecular mechanism of benzimidazole-mediated vascular disruption, this study presents evidence relevant to the widespread use of these compounds while offering potential new clinical applications.


INTRODUCTION
The vascular system is built by the combination of de novo formation of blood vessels by 15 vasculogenesis and the sprouting of new vessels from existing vessels via angiogenesis 1,2 . 16 Imbalances in angiogenesis underlie a variety of physiological and pathological defects, including 17 ischemic, inflammatory, and immune disorders 1,3,4 . Indeed, angiogenesis is central to tumor 18 malignancy and cancer progression, as new blood vessels must be established to supply oxygen 19 and nutrients to the growing tumor. Accordingly, inhibition of angiogenesis is now a well-20 recognized therapeutic avenue [1][2][3][4][5] . Defined angiogenesis inhibitors such as Avastin (FDA approved   Here, we experimentally determined TBZ's specific molecular target and cellular mechanism of 58 vascular disrupting activity. We find that TBZ disrupts microtubule growth, with increased 59 potency in endothelial cells. Using predictive molecular modeling, human cell culture, and 60 humanized yeast, we find TBZ predominantly targets only one of nine human β-tubulins, 61 suggesting an explanation for its cell-type specificity. Finally, based on epidemiological data 62 mining and chemical structures, we discovered that a larger family of benzimidazoles-in clinical

73
Thiabendazole disrupts microtubule plus ends in endothelial cells 74 Thiabendazole exhibits broad-spectrum activity against fungal and nematode crop pests 25 , but 75 prior to demonstration of its VDA activity, it was generally thought to lack activity in tetrapods 24 . HUVECs, TBZ significantly reduced the accumulation of EB1 at microtubule plus ends ( Fig. 1A',

89
B') as compared to its control (Fig. 1A, B). Importantly, and consistent with the overall normal 90 morphology and patterning of TBZ-treated embryos 20 , we found that TBZ had a substantially less 91 robust effect on EB1 accumulation at microtubule plus ends in fibroblasts as compared to 92 endothelial cells (Fig. 1C-E). These data are consistent with, and provide new insights into,  Thiabendazole selectively targets TUBB8 among human β-tubulins 96 Three commonly observed mutations in fungal and nematode β-tubulins (F200Y, E198A, and 97 F167Y) confer resistance to TBZ ( Fig. 2A, B), suggesting that its binding site is in the vicinity of 98 these residues 22,27,31-43 (File S1). Based on the previously observed benzimidazole suppressor 99 6 mutations, we used 3D structural modeling to evaluate TBZ's potential binding sites in a fungal 100 β-tubulin. We first constructed 3D homology models of the Schizosaccharomyces pombe (fission 101 yeast) wild-type and TBZ-resistant F200Y β-tubulins, based on the previously determined Ovis 102 aries β-tubulin crystal structures (PDB: 3UT5 46 and 3N2G 47 ) as templates. We computationally 103 refined the structures and then evaluated potential binding modes of TBZ, as detailed in the 104 Methods, using computational docking algorithms to localize TBZ's potential binding sites within 105 the fungal β-tubulin structures (Fig. S1). We identified a binding site around F200 to be the most 106 probable (File S2). We found that the preferred binding conformations of TBZ in both models

111
On measuring the polar contacts and clashing energies of TBZ with tubulin, we found that the 112 wild-type β-tubulin bound to TBZ more favorably with contact energy (-9.9 kcal/mol) as compared 113 to its F200Y counterpart, which showed unfavorable repulsions (+27.6 kcal/mol)(File S2, S3). For 114 the wild-type protein, TBZ's polar contacts included E198 and Q134 (Fig. 2C, S1). Arene-115 hydrogen interactions between the drug and protein included contributions from F200, L250, and 116 L253. However, for our F200Y mutant, repulsion was observed in our fixed ligand experiments 117 predominantly caused by unfavorable contacts made with Y200, F240, L250, and L253 (Fig. S1). 118 Our analyses suggest F200Y likely forms a hydrogen bond to E198 in the TBZ-resistant mutant, 119 thus constricting the pocket and occluding binding. are not yet fully understood, but recent studies indicate that their sequence diversity modulates 126 binding affinity to tubulin-binding drugs and influences microtubule dynamics through distinct 127 interactions with molecular motors 50,51 . The recurrence of TBZ resistance mutations at the same 128 three loci across diverse fungi and nematodes (File S1) led us to hypothesize that human β-tubulin 129 7 isotypes might have differential sensitivities to TBZ by virtue of incorporating resistant residues 130 at positions 167, 198, and 200, potentially explaining both its tissue-specific effects and generally 131 low toxicity in humans.

132
Indeed, multiple sequence alignment of human and yeast β-tubulin genes indicated that while F167 133 remained conserved across all the human isotypes, positions 198 and 200 were variable (Fig. 2B). 134 Moreover, all human β-tubulin isotypes except TUBB1 and TUBB8 contain the F200Y resistance 135 mutation. Because TUBB1 also harbors the other commonly observed E198A suppressor ( Fig.   136 2B), TUBB8 is the only human β-tubulin isotype predicted by sequence to be TBZ-sensitive.

137
Given this variability across isotypes, we next asked how the E198A and F200Y mutations would 138 be expected to affect TBZ's ability to bind at its predicted site in human isotypes. 139 We first evaluated this hypothesis computationally, by constructing 3D homology models for each 140 of the human β-tubulin isotypes in the same manner as for the fungal model (see Methods). We 141 then performed induced-fit docking with TBZ across our human β-tubulin models. Using a TBZ- Xenopus embryos 20 , we wished to test directly if TUBB8-specific binding could explain the 8 supplying human β-tubulin isotypes predicted to be resistant. We tested this by two independent 159 assays: (i) by overexpressing specific sensitive or resistant human β-tubulin isotypes in human 160 endothelial cells and (ii) by humanizing Baker's yeast's β-tubulin TUB2 to enable assays of 161 individual human β-tubulin isotypes.

162
To test if microtubule dynamics in human cells could be significantly restored by supplying 163 resistant β-tubulin isotypes, we singly transfected HUVEC cells with plasmids overexpressing 164 either TUBB4 or TUBB8 and assayed microtubule dynamics by measuring the comet lengths of 165 end-binding protein EB3 (Fig. 3A). Compared to untransfected HUVECs, we saw that 166 overexpressing TUBB4 significantly rescued the decrease in comet length observed in TBZ-167 treated cells (Fig. 3B). Transfection with TUBB8, by contrast, had no effect (Fig. 3B). The 168 differences became very significant after 30 minutes of exposure (Fig. 3B). 169 As an independent assay of TBZ action on human tubulins, we turned to humanized yeast, as our 170 previous work showed that of the nine human β-tubulins, only TUBB4 and TUBB8 could 171 functionally replace TUB2 in Saccharomyces cerevisiae 52 . From our modeling and docking data, 172 we hypothesized that yeast strains humanized with TUBB8 would be susceptible to TBZ while 173 humanizing with TUBB4 would confer TBZ resistance. Saccharomyces cerevisiae possesses 2 α-174 tubulins (TUB1 and TUB3) that interact with TUB2 to form tubulin heterodimers, which in turn 175 oligomerize to form microtubules. Wild-type BY4741 haploid strains are TBZ-resistant. However, 176 previous studies have shown that on deleting TUB3, yeast strains become susceptible to 177 benzimidazoles 53 likely due to reduced overall α-tubulin stoichiometry or possibly by TBZ 178 occluding TUB2's dimerization with TUB1 but not TUB3. Therefore, we performed all our yeast 179 replacement assays in a tub3Δ background, which yielded a clear growth defect in the presence of 180 TBZ (Fig. S3). In order to test the effect of TBZ on human β-tubulin isotypes TUBB4 and TUBB8, 181 we used CRISPR/Cas9 to construct yeast strains with these human isotypes in place of the 182 endogenous TUB2 and tested them in the presence or the absence of the drug (Fig. 3A). We found 183 that strains possessing wild-type TUB2 and human TUBB8 exhibited slow growth in the presence 184 of TBZ (at conc. 20 µg/ml). By contrast, the strain humanized with TUBB4, which is predicted to 185 be resistant to TBZ, grew normally in the presence of TBZ (Fig. 3C, S3A). 186 9 Together with our in silico docking data, our results in HUVECs and humanized yeast indicate 187 that TUBB8 is uniquely TBZ-sensitive, suggesting in turn that vascular endothelial cells are 188 selectively sensitive to its loss. As a complement to the epidemiological data, we also considered chemical properties by asking if 206 pesticide benzimidazoles shared similar chemical feature profiles relative to other benzimidazoles. 207 We curated >80 commercially available compounds in the benzimidazole class spanning a diverse 208 range including pesticides, fungicides, therapeutics, and preservatives. Upon hierarchical 209 clustering of these benzimidazoles based on their chemical properties computed from JOELib's 210 features matrix 72,73 (File S5), we found that pesticide benzimidazoles generally shared similar 211 chemical properties and clustered together (Fig. 4B).

212
Numerous commercially used benzimidazoles also function as vascular disrupting agents 213 We next tested if pesticides exhibiting the epidemiological signature and clustering in the same 214 clades by virtue of their chemical features would also specifically inhibit TUBB8 and function as 215 VDAs. We selected 12 commercially used benzimidazole compounds across 2 clusters (Fig. 4B). 216 Our list included 2 anthelmintics, both World Health Organization essential medicines 217 (albendazole and mebendazole) prescribed to treat broad-spectrum human intestinal nematode 218 infections; fenbendazole, an anthelmintic prescribed specifically for animals against 219 gastrointestinal nematode parasites; 2 currently banned pesticides, benomyl and carbendazim, 220 formerly used in agriculture; triclabendazole, specifically used to treat liver fluke infections; and 221 5 proton-pump inhibitors (esomeprazole, lansoprazole, omeprazole, pantoprazole, and 222 rabeprazole) used to treat gastrointestinal and stomach acid disorders. The latter set were from a 223 different clade and did not exhibit the epidemiological signature, serving as negative controls. 224 We first took advantage of our humanized yeast strains to rapidly discriminate TUBB8-specific 225 inhibition from general β-tubulin inhibition. We found that 5 of the 12 compounds tested 226 selectively inhibited TUBB8, as evidenced by the growth profiles observed for the humanized 227 strains when cultured in the presence of the drugs (Fig. 5, S4). Notably, none of the 5 proton pump 228 inhibitors or colchicine exhibited any tubulin inhibition (Fig. S4, S5A), confirming the specificity 229 of the epidemiological signature as a predictor of TUBB8 inhibition. In contrast, triclabendazole 230 was generally toxic, behaving as a pan-isotype inhibitor (Fig. S5B). 231 Testing the 5 positive TUBB8-inhibiting compounds in Xenopus laevis embryos showed strong 232 vascular disrupting activity for all 5 compounds (Fig. 5). As we observed previously for TBZ 20 , 233 the gross morphology of the treated embryos was largely normal (Fig. 5). Thus, this broader class 234 of benzimidazoles do in fact generally act as vascular disrupting agents in vertebrates.