Side chain effect in the modulation of αvβ3/α5β1 integrin activity via clickable isoxazoline-RGD-mimetics: development of molecular delivery systems

Construction of small molecule ligand (SML) based delivery systems has been performed starting from a polyfunctionalized isoxazoline scaffold, whose αvβ3 and α5β1 integrins’ potency has been already established. The synthesis of this novel class of ligands was obtained by conjugation of linkers to the heterocyclic core via Huisgen-click reaction, with the aim to use them as “shuttles” for selective delivery of diagnostic agents to cancer cells, exploring the effects of the side chains in the interaction with the target. Compounds 17b and 24 showed excellent potency towards α5β1 integrin acting as selective antagonist and agonist respectively. Further investigations confirmed their effects on target receptor through the analysis of fibronectin-induced ERK1/2 phosphorylation. In addition, confocal microscopy analysis allowed us to follow the fate of EGFP conjugated α5β1 integrin and 17b FITC-conjugated (compound 31) inside the cells. Moreover, the stability in water solution at different values of pH and in bovine serum confirmed the possible exploitation of these peptidomimetic molecules for pharmaceutical application.


Results and discussion
Chemistry. To introduce a functionalizable chain in position 3 of the isoxazoline scaffold, we thought that a terminal alkyne could be a versatile moiety to be exploited in 1,3-dipolar Huisgen cycloaddition with different azide-linkers. To follow our previously reported synthetic protocol, 5-hexynal 1 had to be synthesized from the corresponding commercially available alcohol, via Swern oxidation with oxalyl chloride and TEA in DMSO (99% yield, Scheme 1). The Knövenagel reaction between 1 and t-butyl-acetoacetate afforded in 40% yield the alkylidene acetoacetate 2 as a 1/4 mixture of Z/E isomers (Scheme 1, path A). The unsatisfactory yield, never observed previously for simpler aldehydes, even using a microwave assisted protocol 19 , was attributed to alkyne side reactions. For this reason, we protected the alkyne moiety with TMS group as reported by Cruciani and co-workers 20 . As a consequence, the Knövenagel reaction was performed using TiCl 4 /pyridine affording the alkyne-protected alkylidene acetoacetate 11 in 83% yield as 1/4 mixture of Z/E isomers as well and confirming the previous hypothesis of alkyne side reactivity (Scheme 1, path B). Reaction conditions for the addition of bis-(N,O)-trimethylsilylhydroxylamine to 2 and 11 were optimized on the basis of our previous experience 21,22 , in order to avoid the formation of oxime by-product as a result of the undesired 1,2-addition process 23 . According to these considerations, the reaction was performed in DCM in the presence of a catalytic amount (5%) of ytterbium triflate as Lewis acid. It should be noticed that the TMS protecting group of the hydroxylamine was removed during the usual work up procedure, inducing the fast conversion of the intermediate adduct to trans 5-hydroxyisoxazolidine-4-carboxylate 3 or 12, as a single trans epimer, via intramolecular hemiketalisation. Introduction of the malonic pharmacophore was performed at this stage by acylation at the nitrogen with methyl malonyl chloride, in the presence of TEA. The isoxazolidines 4 or 13 were isolated in 51% and 50% respectively overall yield after two steps. Dehydration to the unsaturated racemic isoxazolines 5 or 14 was accomplished by mesylation of the hydroxyl group followed by base induced elimination. The removal of TMS protecting group on alkyne moiety of 14 by tetrabutylammonium fluoride afforded the product 5 in 53% yield. Selective removal of the t-butyl ester was then accomplished by treatment of 5 with an excess of trifluoroacetic acid in dichloromethane. The arginine mimetic chain was introduced by reaction of the acid 6 with 4-aminobenzylamine, following standard coupling conditions (HBTU/TEA in DCM) to give 7 in 76% yield. Hydrolysis of the methyl ester required a particular effort, since the undesired removal of the malonic side chain easily occurred, favored by the following spontaneous aromatization of the heterocycle as confirmed by LC-MS analysis. After several trials under different conditions, excellent results were obtained with a 7·10 −3 M solution of LiOH·H 2 O in a 2:1 mixture of THF/water, following the reaction evolution by TLC, and stopping it at the disappearance of the starting ester. The acid 8 was obtained in quantitative yield.
In the design of the ligand-linker library, molecules terminating with an amine were prepared in order to obtain carriers for molecules with carboxylates or thiocarboxylates as active functionalization, as for instance the cytotoxic agent fumagillin or the fluorescein isothiocyanate diagnostic derivatives. To this purpose N-Boc-2-azido-ethylamine 15a and N-Boc-3-azido-propylamine 15b were prepared from the corresponding 2-bromo-ethylamine and 3-bromo-propylamine by protection of the amino moiety followed by substitution of the bromide with NaN 3 . The two azides were then reacted with methyl ester 7 in the presence of 10% copper (0) powder and TEA·HCl salt at room temperature in a 1:1 mixture of t-BuOH and water 24 . The click reaction led to the exclusive formation of the desired 1,4-disubstituted [1,2,3]-triazoles 16a and 16b 25 in 57% and 67% yield respectively. Removal of the methyl ester under the above reported conditions afforded compounds 17a (80%) Scheme 1. Synthetic pathways for the synthesis of methyl ester 7 and the corresponding acid 8, using the route (A) or (B) to evaluate the effects of alkyne-side chain on the yield of some critical steps of reaction. and 17b (87%) (Scheme 2, pathway A). Initially, the N-Boc protection at the linker amine moiety was retained in order to avoid the possible interference of a second amine group in the ligand-receptor binding. Moreover, it may mimic a carbamate linked payload.
For the preparation of molecules terminating with a carboxylate, able to conjugate drugs or diagnostics possessing functionalizable amines or hydroxyl groups, as for instance paclitaxel, compound 7 was submitted to Huisgen reaction in the presence of t-butyl 2-azido-acetate 18a and t-butyl 3-azido-propionate 18b, easily prepared from the corresponding bromides. The cycloaddition afforded compounds 19a and 19b in low yield (20% and 34% respectively).
Complete regiocontrol was observed also in this case, affording exclusively the 1,4-disubstituted triazoles. Selective removal of the methyl ester was accomplished as reported above to afford acids 20a and 20b (Scheme 2, pathway B). Even in this case, the t-butyl group was retained to avoid interferences in the binding and to mimic the ester connection with a payload. In order to verify if elongation of the linker could turn into a lower interference in integrin-ligand binding, we planned to synthesize more complex systems. The assembly of these composite molecules could be faced through several protocols, differing for the order of formation of the strategic bonds. Therefore, the synthetic protocol was optimized for each specific substrate. To obtain compound 24, having the succinic moiety as a typical and stable spacer between the carrier and the payload 26 , 3-bromo-propylamine was coupled with mono t-butyl pentandioic acid and the bromide was substituted with NaN 3 , to afford compound 22 in good yield (90%). Then, the click reaction with methyl ester 7 was performed under the usual conditions (75% yield). Removal of the methyl ester moiety, afforded the acid 24 in 40% yield (Scheme 2, pathway C). The synthesis of the simpler compound 26 was accomplished in one step by performing the Huisgen reaction on the acid 8 with ethyl 5-azido-pentanoate. Due to the nature of substrate 8, the conditions of the reaction were modified and Cu(OAc) 2 was used in the presence of sodium ascorbate 27 . The click reaction afforded 26 in 20% yield, as consequence of the difficulties in the purification of the product from the copper salts (Scheme 3, pathway E).
Many examples in the literature report the use of PEG as linker for drug-ligand connection for its ability to increase the solubility and to decrease the immunogenicity of the products 28 . Moreover, PEG-drug conjugates often exhibit a favourable in vivo behaviour and are less prone to enzymatic digestion. These data prompted us to design further carriers, containing the PEG fragment. To this purpose, compound 7 was reacted with commercially available N-Boc-1-amino-3,6-dioxa-8-octaneazide (Boc-NH-PEG(2)-N 3 ) as reported in Scheme 2 (pathway D). Compound 27, isolated in 79% yield was then transformed into the corresponding acid 28 under the usual hydrolytic conditions (Scheme 2).
Finally, the acid 8 was treated with the PEG-azide 29, easily prepared from the corresponding commercially available acid. As previously observed, the nature of the starting isoxazoline suggested the use of Cu(OAc) 2 and sodium ascorbate as catalysts (Scheme 3, pathway F). Under these conditions compound 30 was isolated in 40% yield given the difficult purification step. Fluorescent labelled molecules are useful for localization of proteins, visualization of intracellular processes and study of interactions between ligand and receptors. In particular, imaging by fluorescence provides many advantages in terms of selectivity and sensitivity of the detection. For this purpose, compound 17b, showing good selectivity and potency toward α 5 β 1 integrins, was subjected to further functionalization with a diagnostic dye. To this purpose, its precursor (16b) was deprotected at the N-terminal position of the side chain by using TFA in DCM, and conjugated with FITC (fluorescein isothiocyanate) in presence of an Pharmacological studies. Cell adhesion assay. In our previous experience, isoxazoline-based integrin ligands showed a strong potency towards α v β 3 and α 5 β 1 integrins, not affected by changing the alkyl group linked to position 3 18 . In designing this novel small library of peptidomimetics, alkyl chains were substituted by polyfunctionalized linkers including the triazole ring and amide, carbamate or ester moieties, that may create further interactions with the binding pocket. The ability of the synthesized racemic ligands to inhibit the adhesion of K562 cells (human erythroleukemic cells, expressing α 5 β 1 integrin) or SK-MEL-24 cells (human malignant melanoma cells, expressing α v β 3 integrin) to immobilized fibronectin was evaluated 29 . These cell models are widely used to investigate potential antagonists/agonists of α 5 β 1 or α v β 3 integrin-mediated cell adhesion [30][31][32][33][34][35] . In these experiments, the cells were seeded onto plates coated with fibronectin and allowed to adhere before quantitation of the number of adherent cells, in presence of increasing concentrations of the compounds. As a negative control, under these conditions, no significant cell adhesion was observed for BSA-coated plates or nonspecific substrate-coated plates (i.e., collagen I for SK-MEL-24 expressing α v β 3 and poly-L-lysine for K562 expressing α 5 β 1 integrin, data not shown). The ability of the new compounds to inhibit the adhesion of SK-MEL-24 and K562 cells to fibronectin was compared with that of the standard antagonist compounds Ac-Asp-Arg-Leu-Asp-Ser-OH (Ac-DRLDS) and H-Gly-Arg-Gly-Asp-Thr-Pro-OH (GRGDTP), known to be potent inhibitors of cell adhesion mediated by α v β 3 and α 5 β 1 integrins respectively 36 , and standard agonist Ref E (entry 15, Table 1), a potent and selective α 5 β 1 ligand 37 . The obtained results are summarized in Table 1. From the results, it may be observed that the potency towards integrin α v β 3 is generally maintained, while few compounds show high potency towards α 5 β 1 . Compounds terminating with a carbamate moiety (17a, 17b and 28) behave as antagonists to integrin receptors, displaying anyway different selectivity. In particular, 17a displayed similar potency towards both receptors (entry 2), while the longer 17b is an excellent selective ligand for α 5 β 1 (entry 3), displaying IC 50 in the low nanomolar range. The opposite preference could be observed for compound 28 that showed a good potency for α v β 3 receptor (entry 8). On the other hand, compounds having a terminal ester group (20a, 20b, 24, 26) in the side chain, behaved generally as agonists, inducing an increase in cell adhesion. All the molecules having a linear chain linked to the triazole ring (20a, 20b, 26) displayed high affinity for α v β 3 integrin (entries 4, 5 and 7) in the sub-micromolar range. By introducing an amide moiety in the central part of an elongated linker, as in compound 24, the opposite selectivity was observed, still maintaining an agonist effect (entry 6). The introduction of a short PEG fragment induced complete loss of the activity (compound 30, entry 9). To complete the study, we also performed cell adhesion assay with the alkyne intermediate 8, which showed a good potency and selectivity for α v β 3 . These data seem to suggest a possible influence of the terminal moiety on the agonist/antagonist role. A structure-activity relationship rationalization is still elusive, because agonist/antagonist behaviour 38 for integrin ligands is known to depend also on the concentration 39,40 . Anyway, the good to excellent potency of the members of this class of peptidomimetics confirms their possible use as shuttle for drugs or diagnostic selective delivery to cells overexpressing these two classes of integrins.
Analysis of α 5 β 1 integrin-mediated ERK phosphorylation. To gain further information about the agonist/antagonist role of our peptidomimetic integrin ligands and to verify the effect on intracellular signalling, we investigated the effect of most active compounds 17b and 24 on fibronectin-induced phosphorylation of ERK1/2 in K562 cells, which express α 5 β 1 integrin. The mechanism by which components of extracellular matrix generate intracellular signalling through integrins requires indeed an increased phosphorylation of cytoplasmatic second messengers. Phosphorylation of ERK1/2 plays a central role in fibronectin-mediated survival signalling through integrins: in fact, cell adhesion activates ERK1/2 by binding of α 5 β 1 integrins at the cell surface to extracellular matrix proteins such as fibronectin. The experiment was performed by serum-starving K562 cells in RPMI-1640 containing 1% FBS for 16 h; thereafter, they were preincubated with compound 17b, 24 (10 −7 -10 −9 M) or the vehicle for 60 minutes and then plated for 60 minutes on fibronectin.
When K562 cells were exposed to fibronectin for 60 minutes a much stronger signal was observed for phosphorylated ERK1/2, in comparison to vehicle-treated cells (Fig. 2).
Preincubation with compound 17b (10 −7 -10 −9 M) caused a significant concentration-dependent reduction in the amount of fibronectin-induced ERK1/2 phosphorylation levels in K562 cells (Fig. 2), thus confirming a significant effect of the ligand binding on intracellular signalling cascade. On the contrary, when K562 cells, not preincubated with fibronectin, were exposed to compound 24, a significant concentration dependent increase in ERK1/2-phosphorylation was recorded thus confirming its agonistic behaviour (Fig. 2).
Integrin internalization. Integrin trafficking is an important mechanism employed by cells to regulate integrin-extracellular matrix interactions, and thus cellular signalling, during processes such as cell migration and invasion 42 . α 5 β 1 integrin is internalized, trafficked to recycling endosomes and then recycled to the plasma membrane 43,44 . Membrane trafficking pathways influence α 5 β 1 's ability to promote invasion and metastasis 45,46 . In order to investigate the effects of peptidomimetics on integrin trafficking, α 5 β 1 integrin internalization was observed by confocal microscopy on HEK293 cells transfected with α 5 -enhanced green fluorescent protein (EGFP) plasmid (Fig. 3), as these cells endogenously express β 1 subunit 47,48 . HEK293 + α 5 -EGFP cells were treated with fibronectin (10 µg/mL) alone or in combination with the antagonist 17b (1 µM) or with the agonist 24 (1 µM) alone. As  shown in Fig. 3, after 15 minutes of treatment with the endogenous agonist fibronectin, α 5 β 1 integrin was mainly localized in the cytoplasm (Fig. 3, panels d-f), if compared with vehicle-treated cells in which the integrin is quite completely located on the plasma membrane (Fig. 3, panels a-c). Moreover, when HEK293 + α 5 -EGFP cells were pre-treated with the antagonist 17b before the addition of fibronectin, it prevented integrin internalization, that remained mainly located on the membrane (Fig. 3, panels g-i). On the contrary, compound 24, mimicking the agonist behaviour of fibronectin, induced α 5 β 1 internalization (Fig. 3, panels l-n): in fact, the integrin was mostly localized in the cytoplasm.
Stability assays. The ligand 17b, which showed best results in the pharmacological evaluation, was subjected to stability assays at neutral (pH=7), basic (pH=10) and acidic (pH=3) conditions, as representative of the class of compounds. LC-MS injections were performed after 30 min from the preparation of the samples and then after each 1.5 hours, following the experiments for 14 h (Fig. 4). During the last synthetic step, the methyl ester deprotection, the amidic bond between the heterocycle and malonyl group showed high sensitivity to the basic environment and, as expected, the hydrolysed A was detected in traces as the unique degradation product. As a consequence of the spontaneous aromatization of A, also an increasing amount of B was observed after a short period. The aromatised form B was tested to evaluate its effect on the activity of selected integrins but, as consequence of the absence of a crucial pharmacophore, it was not active. This behaviour doesn't represent a big issue since strong basic conditions are quite unusual in physiological environment. In fact under stress conditions with a large excess of base, the decomposition of 17b to A occurred in few minutes. On the contrary, during the acidic treatment and under neutral conditions, 17b resulted quite stable and after several hours only a 5% loss of product was detected, in favour of the formation of A and B. Finally, the stability of 17b was also checked in bovine serum, confirming that no significant degradation occurs in this physiological medium.

Confocal microscopic observation of isooxazoline FITC-conjugated distribution.
In order to visualize the localization of isoxazolines inside the cells and to explore the possibility to use them as shuttles for selective delivery of therapeutic payloads and diagnostics, compound 31 was synthesized conjugating 17b with FITC, as described above. This FITC conjugated isoxazoline maintained the ability to reduce K562 cells adhesion in a concentration-dependent manner, similar to compound 17b, but with a lower potency (IC 50 : 0.13 ± 0.01 μM). To study intracellular localization of compound 31, HEK293 cells were transfected with a plasmid coding for α 5 integrin subunit (HEK293 + α 5 ), as they do not express this subunit endogenously 47,48 but they express β 1 . When HEK293 + α 5 cells were treated with 31 (0.1-1 μM) we observed that the compound was localized in the cell cytoplasm, as it could be internalized (Fig. 5). In addition, the internalization of compound 31 was concentration-dependent: at a higher concentration (1 μM) it accumulated inside the cells in a greater extent. Moreover, we observed that the internalization of compound 31 is α 5 integrin-mediated because it was not able to enter inside HEK293 cells that do not express α 5 integrin endogenously (Fig. 5a). In order to establish the possibility for compound 31 to be internalized also in cancer cells, endogenously expressing α 5 β 1 , K562 cells, derived from chronic myelogenous leukemia, were exposed to compound 31 (0.1-1 μM), as reported in Methods section. As shown in Fig. 5b, compound 31 was internalized in K562 cells in a concentration-dependent way.
In addition, to confirm that compound 31 is internalized through an active mechanism of endocytosis, K562 cells were exposed for 2 h to monensin (2 µM), which blocks the trafficking of α 5 integrin from the Golgi stack to the trans Golgi network 49 . Inhibition of integrin trafficking prevented compound 31 internalization in K562 cells (Fig. 5b); these results demonstrate that compound 31 enters cells relying on internalization rather than on passive permeability. www.nature.com/scientificreports www.nature.com/scientificreports/ As compound 31 derived from the conjugation of FITC to compound 17b that was a specific α 5 β 1 integrin ligand, to confirm that the selectivity was maintained, we treated HEK293 cells transfected with plasmids coding for α v and β 3 subunits (HEK293 + α v β 3 ) with compound 31 (0.1-1 μM) (Fig. 5c). In cells not expressing α 5 β 1 integrin compound 31 was not able to be internalized. These results confirm from one hand the α 5 β 1 -dependent internalization mechanism of compound 31 and from the other hand displayed that the specificity towards α 5 β 1 of compound 17b was maintained although conjugation with FITC molecule.
Computational studies. To rationalize the pharmacological activity observed for the new isoxazoline-RGD-mimetics, compounds 17b, 20b and 24 were selected to investigate the effects of the side chain in position 3 of the isoxazoline core on the interaction with the target receptors. In the family of the selective α 5 β 1 ligands, we chose the most active compound 17b which inhibits α 5 β 1 integrin-mediated cell adhesion at low nM concentrations, and compound 24 acting as an agonist of α 5 β 1 receptor at µM levels. On the other hand, compound 20b was chosen as selective agonist ligand of α v β 3 integrin. Automated docking calculations were carried out with the Glide software package (version 7.0) by using the crystal structure of the extracellular segment of integrin α 5 β 1 in complex with a disulfide-bonded cyclic RGD peptide (PDB code 4WK4) and the crystal structure of the extracellular domain of integrin α v β 3 in complex with the cyclic pentapeptide Cilengitide (PDB code 1L5G), according to the procedures reported in the Experimental Section. As the docking approaches were successful in reproducing the crystallographic binding mode of the cyclic peptides at the interface of the α and β subunits by means of electrostatic and H-bond interactions, we applied the same docking protocols to both enantiomers of compounds 17b, 20b and 24 to generate computational models for the interaction of these ligands with α 5 β 1 and α v β 3 integrins and evaluate their ability to properly fit the receptor site. In all the calculations, the experimentally observed binding modes of the cyclic peptides with α 5 β 1 and α v β 3 integrins were used as reference models for the analysis of the docking results in terms of protein-ligand interactions. Docking results pointed out that the (R) and (S) enantiomers of the new functionalized isoxazolines show different behavior, with better performances exhibited by the (R)-isomer, especially in the α 5 β 1 integrin. Based on the number of docking poses reproducing the key interactions of the X-ray complex, (R)-17b was the best α 5 β 1 ligand confirming the pharmacological results. In the best pose of (R)-17b (as well as in ten other poses) into the α 5 β 1 binding pocket, the carboxylate group of the ligand is coordinated to the metal cation in the MIDAS region of the β 1 subunit, while the aniline moiety forms H-bond interactions with the negatively charged side chain of Asp227 in the α 5 subunit (Fig. 6, left). Further stabilizing interactions involve the formation of H-bonds between the ligand carboxylate group and the backbone amide hydrogen of Asn224 and Ser134 (and Tyr133 in some poses) in the β 1 unit, and π-stacking interactions between the ligand aromatic group and the α 5 -Phe187 side chain. A H-bond between the Boc carbonyl moiety of ligand and the amino group of β 1 -Lys182 side chain is also observed. The long-chain substituent (bearing the triazole ring) at the position 3 of the isoxazoline is well accommodated at the interface between the α and β subunit; in particular, the triazole ring establishes stabilizing contacts with the side chains of residues α 5 -Trp157, β 1 -Tyr133, β 1 -Ser177 and β 1 -Lys182 (Fig. 6, left). On the contrary, docking results show that most poses of compound 17b (both R and S enantiomers) into α v β 3 lack the H-bond interaction between the ligand aniline moiety and the side chain of α v -Asp218, and display an unfavorable arrangement of the long-chain substituent at the position 3 of the isoxazoline at the α/β integrin interface. Due to residue differences between the binding sites, the long chain bearing the triazole ring of compound 17b can fit unhindered only into the pocket available at the α 5 β 1 interface, thus confirming the pharmacological results in terms of selectivity. A comparison www.nature.com/scientificreports www.nature.com/scientificreports/ of α 5 β 1 and α v β 3 integrins highlights that the mutations of α 5 -Phe187 into α v -Tyr178 and of α 5 -Asp227 into α v -Asp218 might allow the aniline moiety of the ligand to maintain the same H-bond and π-stacking interactions in the two binding sites. Other mutations at the α/β interface modify size and shape of the pocket accessible to the isoxazoline substituent, thus affecting the ligand binding mode (see Supplementary Figure S41). Molecular dynamics simulations allowing partial receptor flexibility showed that (R)-17b maintains stable interactions with α 5 β 1 , similar to those observed in the docking poses. In particular, the long-chain substituent is firmly placed at the α/β integrin interface, with the triazole and the Boc carbonyl moiety engaged in interactions with residues α 5 -Trp157, β 1 -Tyr133 and β 1 -Lys182. Instead, the docking poses of compound 24 reveal some difficulties in establishing the key interactions with the α 5 β 1 pocket, mainly due to the lack of a simultaneous favorable fitting of the long-chain substituent at the α/β interface. Compared to 17b a reduced number of docking poses reproducing the X-ray interactions is achieved and MD simulations confirm high mobility of (R)-24 in the binding pocket. The binding determinants of 20b were finally investigated. The docking poses of (R)-20b into α v β 3 integrin are characterized by the interaction of the carboxylate group with the metal cation in the MIDAS region and the Asn215 and Tyr122 residues in the β 3 subunit, and by the interaction of the aniline moiety with the side chains of α v -Asp218 or α v -Asp150, and of α v -Tyr178. Moreover, the Boc carbonyl moiety of the substituent at the position 3 of the isoxazoline can create H-bonds with the backbone amide hydrogen of α v -Ala149 or the guanidinium group of β 3 -Arg216, while the triazole ring can form H-bond or cation-π interactions with the side chain of β 3 -Arg214 (Fig. 6, right). Molecular dynamics simulations allowing partial receptor flexibility showed that (R)-20b maintains stable interactions with α v β 3 , especially those involving the acid pharmacophoric group and the triazole ring. In short, according to these results, only the subtle fitting of suitable features of the isoxazoline substituent with appropriate receptor traits at the α/β integrin interface seems to allow the optimal interaction www.nature.com/scientificreports www.nature.com/scientificreports/ of both the pharmacophoric moieties and the functionalizable chain of the isoxazoline-RGD-mimetics with the integrin binding site.

Conclusions
The synthesis of a small library of ligands obtained by conjugation of polyfunctionalized linkers to isoxazoline ring via Huisgen-click reaction, allowed to verify the possibility to use these scaffolds as α v β 3 and α 5 β 1 targeting motifs, with the aim to use them as "shuttles" for selective delivery of therapeutics and diagnostics to cancer cells. The different behavior of the members of the library seems to suggest a correlation between the terminal moiety of the linker chain and the cell adhesion inhibition or activation, even if a rationale in structure-activity relationship is not predictable and needs further investigation. Compound 17b, that showed excellent potency towards α 5 β 1 integrin in the nanomolar range as antagonist, was selected for further investigation to establish the effect on fibronectin induced ERK phosphorylation. It was able to prevent fibronectin-induced α 5 -integrin-mediated ERK1/2 intracellular signaling activation and α 5 -integrin internalization.
Moreover, compound 31, isoxazoline FITC-conjugated derived from compound 17b, confirmed the possibility to exploit these integrin ligands as "shuttles" for the selective delivery of therapeutics and diagnostics as consequence of its internalization only inside integrin expressing cells.
Finally, stability in water solution at different values of pH and in bovine serum was verified in order to confirm the potential exploitation of these peptidomimetic molecules for pharmaceutical applications.

Methods
All chemicals were purchased from commercial suppliers and used without further purification. Anhydrous solvents were purchased in sure seal bottles over molecular sieves and used without further drying. Flash chromatography was performed on silica gel (230-400 mesh). NMR Spectra were recorded with Varian Mercury Plus 400 or Unity Inova 600 MHz spectrometers. Chemical shifts were reported as δ values (ppm) relative to the solvent peak of CDCl 3 set at δ = 7.27 ppm ( 1 H-NMR) or δ = 77.0 ppm ( 13 C-NMR), CD 3 OD set at δ = 3.31 ppm ( 1 H-NMR) or δ = 49.0 ppm ( 13 C-NMR), D 2 O set at δ = 4.79 ppm ( 1 H-NMR), CD 3 CN set at δ = 1.93 ppm ( 1 H-NMR) or δ = 117.7 ppm ( 13 C-NMR), (CD 3 ) 2 CO set at δ = 2.04 ppm ( 1 H-NMR) or δ = 29.8 ppm ( 13 C-NMR). Coupling constants are given in Hertz. LC-MS analyses were performed on an HP1100 liquid chromatograph coupled with an electrospray ionization-mass spectrometer using a Phenomenex Gemini C18-3 µ -110 Å column, H 2 O/CH 3 CN as neutral solvent at 25 °C or H 2 O/CH 3 CN with 0.2% formic acid as acid solvents (positive scan 100-500 m/z, fragmentor 70 eV). Another set of experiments has been performed on an HP1100 liquid chromatograph coupled with an electrospray ionization-ion trap mass spectrometer MSD1100 using a Phenomenex Zorbax C18-3.5 µ -80 Å column, H 2 O/CH 3 CN with 0.08% trifluoroacetic acid as acid solvents (positive scan 100-500 m/z, fragmentor 70 eV). Compounds 9 and 10 were synthesized following a known procedure and analytical data are in agreement with the literature. Analyses indicated by the symbols of the elements or functions were within ±0.4% of the theoretical values.

Synthesis.
General procedure for the Swern oxidation: (1) and (10).  www.nature.com/scientificreports www.nature.com/scientificreports/ in 5 min and the solution was then stirred for 15 min. TEA dry (10.1 ml, 72.5 mmol) was added dropwise and the mixture was stirred further at room temperature for 10 min. The mixture was diluted with 60 ml of Et 2 O and 30 ml of water. The organic layer, after dilution with Et 2 O, was washed with NH 4 Cl sat. (20 ml x 3). It was then dried over Na 2 SO 4 and evaporated under reduced pressure to give the product as a yellow oil and used without further purifications.

Synthesis of FITC-conjugated ligand
Cell adhesion assays. Plates (96 wells) (Corning Costar, Celbio, Milan,Italy) were coated by passive adsorption with fibronectin (10 µg/mL) or poly-L-lysine (0.002%) (Sigma-Aldrich SRL) overnight at 4 °C. Cells were counted with a haemocytometer and pre-incubated with different concentrations of each compound or with the vehicle (methanol) for 30 min at room temperature to reach a ligand-receptor equilibrium. Stock solutions (10 -2 M) of the assayed compounds were prepared in phosphate-buffered saline (PBS). At the end of the incubation time, the cells were plated (50000 cells/well) and incubated at room temperature for 1 h. Then, all the wells were washed with PBS to remove nonadherent cells, and 50 µL of hexosaminidase [4-nitrophenylN-acetyl-β-D-glucosaminide dissolved at a concentration of 7.5 mM in 0.09 M citrate buffer solution (pH 5) and mixed with an equal volume of 0.5% Triton X-100 in water] was added. This product is a chromogenic substrate for β-N-acetylglucosaminidase that is transformed in 4-nitrophenol whose absorbance can be measured at 405 nm. As previously described 51 , there is a linear correlation between absorbance and enzymatic activity. Therefore, it is possible to identify the number of adherent cells among treated wells, interpolating the absorbance values of unknowns in a calibration curve. Western blotting analysis. K562 cells were incubated in RPMI-1640 with 1% FBS for 16 h. Plates were coated with 10 µg/ml of fibronectin and blocked with 1% BSA (Sigma-Aldrich SRL). Subsequently, 4 × 106 cells were pre-incubated with different concentrations of compounds for 30 minutes. Cells were allowed to adhere for 1 hour on fibronectin in RPMI-1640 with 1% FBS. Cells treated with agonists were not incubated with fibronectin. Then, the cells were lysed in M-PER Mammalian Protein Extraction Reagent; (M-PER Pierce, Rockford, IL, USA) supplemented with phosphatase inhibitor (Sigma-Aldrich SRL) for 10 min at 4 °C by gently shaking. Cell debris were removed by centrifugation (14,000 × g for 15 minutes at 4 °C) and protein concentrations were estimated by BCA assay (Pierce, Rockford, IL, USA). Protein extracts (100 µg) were denatured at 95 °C for 3 min before being loaded and separated in 12% SDS-PAGE gels. The membranes were blocked in 1% BSA and incubated for 2 hours with anti-phospho-ERK 1/2 (extracellular signal-regulated kinase 1/2) (1:2500) or anti-total ERK 1/2 antibodies (1:5000) (Promega, Madison, WI, USA) and, thereafter with anti-rabbit peroxidase-conjugated secondary antibody. Digital images were acquired and analyzed following previously reported methods 52 .
Confocal laser scanning microscopy. HEK293 cells (not expressing α 5 but endogenously expressing β 1 integrin 47,48 were plated in 6 well plates and at 50-60% confluence were transiently transfected with α 5 -EGFP plasmid using Lipofectamine2000 transfection reagent (LifeTechnologies). After 48 h from transfection, HEK293 + α 5 -EGFP cells were assessed to verify integrins expression by flow cytometry (data not shown). α 5 -EGFP integrin plasmid was a kind gift from Michael Davidson (Addgene plasmid #56423). HEK293 + α 5 -EGFP cells were treated at 4 °C with fibronectin (10 µg/ml) or the agonist 24 (1 µM) in MEM for 1 hour. Cells exposed to the antagonist 17b (1 µM), before the addition of fibronectin, were pre-incubated with the compound for 20 minutes at 4 °C. After 1 h incubation with fibronectin or compound 24, cells were moved to 37 °C for 15 minutes. Then, the cells were washed twice with PBS and fixed with paraformaldehyde (3% in PBS, pH 7.4, 10 min). Thereafter, the cover slips were washed twice with 0.1 M glycine in PBS and twice with 1% BSA (bovine serum albumin) in PBS.
Specimens were embedded in Mowiol and analyzed using a Nikon C1s confocal laser-scanning microscope, equipped with a Nikon PlanApo 60×, 1.4-NA oil immersion lens. The images have elaborated using NIS-Elements C Software.
For internalization analysis of ligand 31, the mean fluorescence intensity was related to the background; the relative fluorescence intensity is reported in the graph in Fig. 5.
Computational studies. All  Ligand preparation. Ionized carboxylate and neutral aniline are suggested by the Epik module [Epik version 3.5, Schrödinger, LLC, New York, NY, 2016] as the relevant protonation states at pH = 7 for the acid and basic pharmacophoric groups of isoxazoline derivatives according to predicted pK a values of 3.3 (carboxylic acid) and 4.6 (phenyl anilinium derivative). These protonation states were considered for computational studies of the isoxazoline-containing compounds. Ionized carboxylate and protonated guanidinium groups have been employed for the cyclic RGD integrin ligands.
Protein preparation. The crystal structure of the extracellular domain of the integrin α 5 β 1 in complex with the disulfide-bonded cyclic peptide ACRGDGWC (PDB code 4WK4) and the crystal structure of the extracellular domain of the integrin α v β 3 in complex with the cyclic pentapeptide RGDf(NMe)V Cilengitide (PDB code 1L5G) were used for docking studies. The α 5 β 1 integrin structure was set up for docking as previously reported (residues 40-351 for chain α 5 and 121-358 for chain β 1 , Mg 2+ ion at MIDAS) 55 . The α v β 3 integrin structure was truncated to residue sequences 1-438 for chain α v and 107-354 for chain β 3 , and all the bivalent cations were modeled as Mn 2+ ions. Then, the Protein Preparation Wizard using the OPLS2005 force field was run to get the final structures.
Molecular docking. Docking calculations were performed using Glide version 7.0 [Glide version 7.0, Schrödinger, LLC, New York, NY, 2016] in the SP (Standard Precision) mode. Receptor grids were generated on the extracellular fragments of α 5 β 1 and α v β 3 integrin prepared as described in Protein Preparation. The settings of the docking step were defined as previously reported 32,52 . The GlideScore function was used to select 20 poses for each ligand after a post-minimization of the ligand structure within the binding site. The flexible-ligand docking method was selected and the SP mode was used with the option for amide bonds set to 'Penalize non planar conformation' . No Epik state penalty was added to the docking score.
Each docking protocol was initially tested for its ability to reproduce the X-ray binding mode of the cyclic RGD ligand in the receptor crystal structure. Glide was successful in reproducing the experimentally determined binding mode of the cyclic peptide ACRGDGWC in α 5 β 1 integrin and of Cilegitide in α v β 3 integrin, as they correspond to the best-scored poses in the two docking runs.
Molecular dynamics simulations. Short MD simulations were run for selected derivatives to investigate the reliability of docking poses while allowing partial receptor flexibility. MD simulations were performed starting from the docking best poses and using MacroModel version 11.1 [MacroModel version 11.1, Schrödinger, LLC, New York, NY, 2016]. The following protocol was employed: OPLS2005 force field, implicit GB/SA water model, temperature 300 K, 1.0 fs integration step, 10 ps equilibration time and 10 ns simulation time. Variable degrees of freedom were assigned to the different moieties of the ligand-receptor complex, constraining the position of the atoms that are farthest from the receptor binding site and removing the interactions which are likely to have a negligible influence on the results. The ligand-integrin complex employed in the docking calculations is divided into four substructures according to the following scheme: the ligand is allowed to freely move, residues within 10 Å from the ligand are constrained at the crystal positions with a force constant K = 100 kJ mol −1 Å −2 , residues within 5 Å from the second shell are constrained at the crystal positions with a force constant K = 200 kJ mol −1 Å −2 , residues within 10 Å from the third shell are frozen in their respective crystal positions. The remaining residues are not considered in the calculation. Each structure has undergone a TNCG minimization step and, after the equilibration at 300 K, 2.000 structures from each simulation were saved for the analysis.