Structure Activity Relationship Studies around DB18, a Potent and Selective Inhibitor of CLK Kinases

Three series of our lead CLK1 inhibitor DB18 have been designed, synthetized and tested against CLKs and DYRK1A kinases. Their cytotoxicity was subsequently measured on seven representative cancer cell lines. Guided by docking experiments, we focused on the less constrained part of the scaffold, and showed that drastically different substituents can be tolerated here. This work ended with the discovery of another promising derivative 12g, with IC50 = 0.004 µM in the inhibition of HsCLK1 and IC50 = 3.94 µM for the inhibition of HsDYRK1A. The SAR results are discussed in the light of extensive molecular modeling analyses. Finally, a kinome scan (463 human kinases) confirmed the outstanding selectivity of our lead compound DB18, suggesting that this scaffold is of prominent interest for selective CLK inhibitors. Altogether, these results pave the way for the development of inhibitors with novel selectivities in this family of kinases.


Introduction
The human kinome comprises 538 kinases playing essential functions by catalyzing protein (518 enzymes) or lipid (20) phosphorylation (encoded by almost 3% of the 19,000 human genes) [1]. Protein kinases (PKs) are involved in the regulation of numerous cellular processes, such as metabolism, cell cycle progression, cell adhesion, vascular function and angiogenesis, often in response to an external stimulus. Aberrant kinase activity (e.g., by hyperactivation, or genetic alterations such as mutations and translocations) plays an important role in the pathogenesis of many diseases including neurodegenerative, cardiovascular, autoimmune and inflammatory diseases as well as in numerous cancers. Over the past three decades, this family of enzymes has emerged as one of the most important suppliers of drug targets and it is estimated that one-quarter of the current drug discovery efforts worldwide are focused on the PKs [2]. There are already 71 American Food and Drug Administration (FDA)-approved small molecule PK Thus, the goals of this study are: 1. to describe the synthesis of these target molecules; 2. to report their activity on the CLK kinases and, for the most promising compounds, also their action on the DYRK1A kinase; 3. to present their cytotoxicities against a representative panel of seven cancer cell lines; and 4. for the most active inhibitors, to Thus, the goals of this study are: 1. to describe the synthesis of these target molecules; 2. to report their activity on the CLK kinases and, for the most promising compounds, also their action on the DYRK1A kinase; 3. to present their cytotoxicities against a representative panel of seven cancer cell lines; and 4. for the most active inhibitors, to rationalize their action on CLKs by extensive molecular modelling studies. In a final stage, the kinome selectivity profile of DB18 against a panel of 463 human kinases will be presented.

Chemical Syntheses
The synthesis of our first target molecules 7 uses the same route as for the compound DB18 (Scheme 1). It starts from the known 2-chloro-8-methoxyquinazoline 1 [14], which by reaction with aniline 2a, or the 3,5-dichloroaniline 2b, in the presence of a palladium catalyst gave the 2-substituted anilinoquinazolines 3a and 3b in fair yields. The 8-hydroxyquinazolines 4a and 4b, obtained from intermediates 3a and 3b by the demethylation of the methoxy group in the presence of BBr 3 , reacted with propargyl bromide in the presence of three equivalents of potassium carbonate in acetone under reflux for 8 h to give the key intermediates 5a and 5b. The presence of propargyl groups in compounds 5 was evidenced by the presence of a doublet at δ 5.01 and a triplet at δ 3.65 ppm in their 1 H-NMR spectrum. The 13 C-NMR spectrum also confirmed the presence of O-propargyl groups in 5 by showing the carbon peaks at δ 56.46, 78.91 and 78.58 ppm. Then, the target molecules 7a and 7b were obtained in good yields, using a click reaction [15][16][17][18] between the alkynes 5 and the azide 6 [19]. The triazoles were evidenced by all spectral data and in particular the disappearance of propargylic groups and appearance of the triazole protons at 8.79 ppm. If we take 7a as a representative example we observe the NH at 9.88 ppm, the triazole proton at 8.09 ppm, the CH 2 at 5.48 ppm, with the methyl of the tolyl group at 2.53 ppm. The characteristic proton in position 4 of the quinazoline appears at 9.29 ppm. The other signals belong to the aromatic/heteroaromatic protons. In 13 C NMR we identify the (C-H) carbon signal of triazole at 125.8 ppm, the methylene group at 62.1 ppm and the methyl of the tolyl group at 20.3 ppm. The signal of the carbon in position 4 of the quinazoline is at 161.8 ppm. The other signals belong to the aromatic/heteroaromatic systems. All these data have been confirmed by extensive 2D experiments and similar results have been obtained for the other quinazolines. Scheme 1. Synthesis of the first targets, the anilino-2-quinazoline 7. Reagents andconditions ; (i) aniline 2a or 3-chloroaniline 2b, Pd(OAc)2, BINAP, Cs2CO3, 1,4-dioxane, 100° C, 16h, 3a (52%), 3b (61%); (ii) BBr3, DCM, 16 h, RT, 4a (49%), 4b (54%); (iii) propargyl bromide, K2CO3, acetone, reflux, 6h, 5a (47%), 5b (68%); (iv) 6 (2 equiv), CuSO4. 5H2O, sodium ascorbate, RT, 16 h, 7a (58%);7b (54%).
The synthesis of the second series of molecules is presented in Scheme 2.

Kinase Inhibition Studies
All molecules prepared during this study have been first screened against a short panel of kinases to perform a primary evaluation of their bioactivity. Our compounds were tested at 10 µM and 1 µM on eight disease-related serine/threonine protein kinases including cyclin-dependent kinases (CDK5/p25, CDK9/cyclinT), proviral integration site for Moloney murine leukemia virus kinase (PIM1), glycogen synthase kinase-3 beta (GSK3β), CDC-like kinase 1 (CLK1), dual specificity tyrosine phosphorylation regulated kinase 1A (DYRK1A), casein kinase 1-ε (CK1ε) and the mitotic protein kinase Haspin. The results are reported in Table S1. For the compounds demonstrating significant activities, more in-depth studies have been performed on human CLKs and DYRK1A kinases. Corresponding results will be presented by a series of designed molecules and compared to the data obtained earlier with our lead molecule DB18 [14].
In the first series, we explored the role of the substituents on the aniline linked to the quinazoline (Table 1). The DB18 had a chlorine in meta position; 7a had no substituent, and 7b had two chlorine atoms in meta-meta' positions. The unsubstituted compound 7a proved to be 100 times less active than our lead compound on the human CLK1 (IC 50 : 1100 and 11 nM respectively). Therefore, it was not explored further on the other CLK kinases and on DYRK1A. On the other hand, the dichloro molecule 7b proved to be just slightly less active than DB18 on the MmCLK1 and thus its activity was studied on the other human CLK kinases. It showed a profile similar to our reference compound with a small decrease in affinity for the four kinases and was completely inactive on the human Molecules 2022, 27, 6149 7 of 24 DYRK1A. All together these results indicated that a meta-chloro substituent on the aniline group is required for activity but that there should be some flexibility in this binding pocket since the dichloro remained significantly active and selective. However, at this stage, the meta-chloroaniline group used for DB18 appeared as the best compromise. In the second series of molecules, we explored the possibilities of modifications in the upper part of the DB18 lead compound by replacement of the methyl group through polar atoms and/or functional groups. The results of the biological studies are reported in Table 2. The introduction of bromine led to a compound (12d) with low activities on the human CLK1 kinases. Similarly, the corresponding phenolic derivative 12c demonstrated poor IC 50 for human CLK1. Therefore, these two molecules were not explored on the other CLK kinases, as well as on DYRK1A. Next, functional groups were introduced in this para position of the aromatic group. The alcohol 12e showed low affinities (IC 50 at 150-200 nM) on the human CLK1 and CLK4 kinases. The affinity was slightly higher (50 nM) on CLK2 and very low (7.5 µM) on CLK3 but a significant activity was also observed on DYRK1A (2.3 µM). The last designed compounds were the ester 12f and the acid derivative 12g. The activity of the ester 12d was low on the human CLK1, in the 0.2-0.3 µM range and therefore it was not further characterized, neither on the other CLK kinases, nor on DYRK1A. On the contrary, the corresponding acid 12g proved to be very active (4 nM), on the human CLK1. For HsCLK2 the IC 50 was found at 28 nM and for HsCLK4 at 55 nM. As often found in these series, the IC 50 is higher for HsCLK3 with a 339 nM value. Finally, this molecule exhibited also a low, but significant, affinity for HsDYRK1A with an IC 50 of 3.94 µM. Altogether, these results demonstrated that some modifications were possible in the upper part of the reference compound. In particular, the acid 12g demonstrated IC 50 values very similar to DB18 on the human kinases CLK1 to CLK4. However, if we consider the selectivity CLK1 versus DYRK1A, 12g is less selective than DB18 which exhibited no action at 100 µM on this particular kinase.
In the third series, we explored the possibility to replace the central aromatic group by cyclohexyl substituents. Based on some preliminary molecular modelling studies it appeared useful to add a primary amine on the cyclohexyl in order to obtain some extra hydrogen bond interactions on the upper part of the kinases. The N-Boc protected intermediates 15 and 18 proved to be completely inactive in the preliminary screening assays on CLKs and other kinases (Table S1) and therefore were not studied further. Thus, only the molecules 16 and 19 have been considered and the results are given in Table 3. These molecules differ both by the cisand trans-stereochemistry of the cyclohexyl substituents and by the number of chlorine atoms (1 or 2) on the aniline group. The four derivatives 16a, 16b, 19a and 19b demonstrated a low, but significant activity on the human DYRK1A kinase, in the 2-5 µM range. Thus, two molecules 16a and 16b were selected for evaluation on the four human CLKs. Interestingly, their selectivity profiles were similar to DB18 (good affinities for CLK1, CLK2 and CLK4 with lower activity on CLK3) albeit with a slight loss in efficacy compared to our lead molecule. Altogether, these results indicated that there is some space in this central part of the CLK kinase active pockets to adjust for substituents like cyclohexyl groups, which are bulkier than an aromatic system. However, with the compounds 16 and 19, there was a considerable loss in selectivity regarding DYRK1A. For these compounds, the IC 50 values were in the low micromolar range althoughin the case of DB18, no inhibition was observed at 100 µM.
These results indicate that, in spite of significant structural changes, modifications in this central part of our lead compound were still possible while keeping a similar global profile of CLKs' inhibition. However, this occurred with a significant loss of selectivity for CLKs vs. DYRK1A.
The most significant results from these studies will be discussed by using molecular modelling methods in the next part of this paper.

Structural Investigation by Molecular Simulations
The first series constituted a brief exploration of the aniline part of our lead DB18. It showed that the meta-chloro position has a huge effect on CLK1, since DB18 is 100 times more potent than the non-substituted analogue 7a (Table 1). Interestingly, the meta-meta'dichloro 7b showed an intermediate activity. We performed thorough molecular modelling investigations in order to understand these results (See Supplementary Materials), and we present a summary of them in the following. The binding mode of 7a and 7b have significant similarities to that already described for DB18 [14]. The anilinoquinazoline is deeply buried in the active site, stacked with F241, and forming two key hydrogen bonds with the backbone amide of L244 ( Figure 2).
The nitrophenyl ring is stacked between the F172 and K290 through π-π and π-cation network. The orientation of nitrophenyl and triazole ring are altered in the pocket due to absence of meta chlorine in 7a, because of the lack of intramolecular interactions ( Figure 2a). The anilinoquinazoline of 7b is also involved in two backbone hydrogen bonds with L244 amide (Figure 2b). The nitrophenyl core is accommodated in the pocket like DB18 and its 2-nitro group formed electrostatic interactions with the primary amine of K290 residue. Thus, from the molecular docking experiments, the binding behaviour of 7a and 7b in the CLK1 pocket are identified as slightly different. For both inhibitors, the 2-anilino quinazoline occupy exactly the same pocket, with a RMSD of 2.5 Å, while the triazolo phenyl rotated along the phenol axis. According to molecular simulations, the higher affinity of DB18 compared to 7a results from a combination of several features: (i) an increase in van der Waals interactions with the protein due to the size of the chloro substituent, (ii) a strong intramolecular halogen/π interaction which limits the entropic cost of ligand binding, (iii) a greater stabilization of the protein active site. The nitrophenyl ring is stacked between the F172 and K290 through π-π and π-cation network. The orientation of nitrophenyl and triazole ring are altered in the pocket due to absence of meta chlorine in 7a, because of the lack of intramolecular interactions ( Figure 2a). The anilinoquinazoline of 7b is also involved in two backbone hydrogen bonds with L244 amide (Figure 2b). The nitrophenyl core is accommodated in the pocket like DB18 and its 2-nitro group formed electrostatic interactions with the primary amine of K290 residue. Thus, from the molecular docking experiments, the binding behaviour of7a and 7bin the CLK1 pocket are identified as slightly different. For both inhibitors, the 2-anilino quinazoline occupy exactly the same pocket, with a RMSD of 2.5 Å, while the triazolo phenyl rotated along the phenol axis. According to molecular simulations, the higher affinity of DB18 compared to7a results from a combination of several features: (i) an increase in van der Waals interactions with the protein due to the size of the chloro substituent, (ii) a strong intramolecular halogen/π interaction which limits the entropic cost of ligand binding, (iii) a greater stabilization of the protein active site.
Docking, confirmed by molecular dynamics, revealed that the active conformation of DB18 is maintained by several intramolecular interactions. In addition to the strong halogen/π bond, a CH:O interaction restricts phenyltriazole movements (Table 4). This stabilization of the ligand allowed a stronger hydrogen bond with the key residue L244. The anilinoquinazoline forms two hydrogen bonds respectively during 54 and 60 % of the simulation for 7a, andthe same atoms form two hydrogen bonds respectively during 89 and 93 % of the simulation for DB18. a b Figure 2. (a) 7a in CLK1 cavity; (b) 7b in CLK1 cavity; the ligands 7a and 7b are shown as thick residues in color-faded magenta. Color coding: oxygen-red, nitrogen-blue, carbon-grey (in case of protein). Backbone, side chain hydrogen bonds are shown as yellow dotted lines, π-π stackings are given in light blue dotted lines and and π-cation showed in green dotted lines whereas, charged interactions are provided in pink dotted lines. To focus the interactions of active site residues with ligands 6.0, Å surface area of protein has been shown in the figures.
Docking, confirmed by molecular dynamics, revealed that the active conformation of DB18 is maintained by several intramolecular interactions. In addition to the strong halogen/π bond, a CH:O interaction restricts phenyltriazole movements (Table 4). This stabilization of the ligand allowed a stronger hydrogen bond with the key residue L244. The anilinoquinazoline forms two hydrogen bonds respectively during 54 and 60% of the simulation for 7a, andthe same atoms form two hydrogen bonds respectively during 89 and 93% of the simulation for DB18. Table 4. Geometric properties of inhibitors during molecular dynamics of the complexes with CLK1 (mean value). The last line represents bar plots of the dihedral (probability density of torsion during the simulation). Note the higher flexibility of 7a, as evidenced by the larger standard deviation of its dihedral, as compared with DB18. The halogen/π bond is symbolized by a dotted line on the chemical structure (right).

7a
DB18 Molecules 2022, 27, x FOR PEER REVIEW 10 of 26 Table 4. Geometric properties of inhibitors during molecular dynamics of the complexes with CLK1 (mean value). The last line represents bar plots of the dihedral (probability density of torsion during the simulation). Note the higher flexibility of 7a, as evidenced by the larger standard deviation of its dihedral, as compared with DB18. The halogen/π bond is symbolized by a dotted line on the chemical structure (right).
As far as the protein is concerned, the whole CLK1 is globally a little more stabilized by the binding of DB18 as compared to 7a (RMSD of 2.4 vs 2.6 Å). Particularly, the Nt-domain is slightly more structured ( Figure S1).
Since the anilino group is only partially buried in the active site, the dichloro analogue 7b is penalized by the solvent exposure of the hydrophobic meta'-chloro substituent, which makes no interaction with the protein (Figure 2b).
Because of its low affinity for CLK1, 7a was not examined further, but we determined the profile of DB18 and 7b for the other CLKs and DYRK1A kinases. As we already mentioned for DB18, [14] the binding mode of our inhibitors is very similar for the four CLKs. So, it was not unexpected that we observed a similar activity profile of DB18 and 7b for CLK1, CLK2 and CLK4. The huge drop of affinity for CLK3 and DYRK1A was already analyzed in details for DB18, and imputed to the inability of the enzymes to adapt their conformation to the inhibitors by an induced-fit mechanism, as CLK1, -2 and -4 do [14].
Because molecular docking suggested that the para-methyl of the first series is solvent-exposed ( Figure 2), we tested if a polar substituent could be more favorable, giving rise to the second series. Surprisingly, the para-hydroxy was not so good, as shown in Table 2. However, increasing the distance between the polar head and the phenyl permitted to recover part of the affinity, thanks to a strong hydrogen bond with D250 and/or Y249. Even more surprisingly, a carboxylic acid proved to be the best substituent on this position. Docking of 12g suggested that this extreme modification of the physico-chemical nature of the inhibitor forced a rotation of 90° along the phenol (Figure 3a).
Molecules 2022, 27, x FOR PEER REVIEW 10 of 26 Table 4. Geometric properties of inhibitors during molecular dynamics of the complexes with CLK1 (mean value). The last line represents bar plots of the dihedral (probability density of torsion during the simulation). Note the higher flexibility of 7a, as evidenced by the larger standard deviation of its dihedral, as compared with DB18. The halogen/π bond is symbolized by a dotted line on the chemical structure (right).
As far as the protein is concerned, the whole CLK1 is globally a little more stabilized by the binding of DB18 as compared to 7a (RMSD of 2.4 vs 2.6 Å). Particularly, the Nt-domain is slightly more structured ( Figure S1).
Since the anilino group is only partially buried in the active site, the dichloro analogue 7b is penalized by the solvent exposure of the hydrophobic meta'-chloro substituent, which makes no interaction with the protein (Figure 2b).
Because of its low affinity for CLK1, 7a was not examined further, but we determined the profile of DB18 and 7b for the other CLKs and DYRK1A kinases. As we already mentioned for DB18, [14] the binding mode of our inhibitors is very similar for the four CLKs. So, it was not unexpected that we observed a similar activity profile of DB18 and 7b for CLK1, CLK2 and CLK4. The huge drop of affinity for CLK3 and DYRK1A was As far as the protein is concerned, the whole CLK1 is globally a little more stabilized by the binding of DB18 as compared to 7a (RMSD of 2.4 vs. 2.6 Å). Particularly, the Nt-domain is slightly more structured ( Figure S1).
Since the anilino group is only partially buried in the active site, the dichloro analogue 7b is penalized by the solvent exposure of the hydrophobic meta'-chloro substituent, which makes no interaction with the protein (Figure 2b).
Because of its low affinity for CLK1, 7a was not examined further, but we determined the profile of DB18 and 7b for the other CLKs and DYRK1A kinases. As we already mentioned for DB18, [14] the binding mode of our inhibitors is very similar for the four CLKs. So, it was not unexpected that we observed a similar activity profile of DB18 and 7b for CLK1, CLK2 and CLK4. The huge drop of affinity for CLK3 and DYRK1A was already analyzed in details for DB18, and imputed to the inability of the enzymes to adapt their conformation to the inhibitors by an induced-fit mechanism, as CLK1, -2 and -4 do [14].
Because molecular docking suggested that the para-methyl of the first series is solventexposed (Figure 2), we tested if a polar substituent could be more favorable, giving rise to the second series. Surprisingly, the para-hydroxy was not so good, as shown in Table 2. However, increasing the distance between the polar head and the phenyl permitted to recover part of the affinity, thanks to a strong hydrogen bond with D250 and/or Y249. Even more surprisingly, a carboxylic acid proved to be the best substituent on this position. Docking of 12g suggested that this extreme modification of the physico-chemical nature of the inhibitor forced a rotation of 90 • along the phenol (Figure 3a). In this conformation, the nitrophenyl forms a strong π-stacking with F172, and the carboxylic acid forms a salt bridge with K290, in addition to a hydrogen bond with the backbone of A171. This strong link between the bottom of the active site and its lid formed by the β-sheet domain G154-N195/Q226-E242 further stabilized the active site and explains why this compound is the best in this second series. The overlap between 12g and DB18 in CLK1 is presented in Figure 3b. 12g is also very potent on other CLKs, and still relatively selective towards CLK3 and DYRK1A. The docking of 12g in Dyrk1A is represented in Figure 3c and its overlap with DB18 in Figure 3d.
The very high affinity of the carboxylic acid derivative was a surprise. We were first astonished that the active site could tolerate an electronegative substituent in this region, because it is surrounded by many acidic amino-acids: E169, E206, D288, E292 and D325. Of course, several basic residues are also present, but they are few (K191, K194 and K290), and can only partially equilibrate the charges. The resulting electrostatic clash explains probably why docking experiments suggested a reorientation of the benzoic acid part of the inhibitor, as compared with DB18. We checked this aspect through molecular dynamic simulation, by replacing the para-methyl of DB18 by a carboxylic acid, in the exact binding mode of DB18. After 300 ns, we observed an abrupt shift of the inhibitor position. It seems that the electronegative environment generated by acidic residues, particularly D250 and E292, forced an expulsion of the benzoic acid via a rotation of the phenol dihedral ( Figure S2). We realized thereafter that this acidic region of the active site is precisely the domain encompassing the phosphate groups of the natural substrate ATP of kinases as evidenced by several crystal structures (For example, Figure 4 shows a {DYRK1A:ATP} complex (PDB:7A4O_B)} [21]. Since ATP is often bound to Mg 2+ (see for example PDB:4IIR for a {kinase:ANP:Mg 2+ } complex [22], it is probable that CLKs also bind Mg 2+ in this position, even if the cation has not been seen in X-ray structures.
The tolerance of CLKs' active site towards very diverse substituents of the phenyl encourage us to test another extreme, with a basic group. To this end, we synthetized the third series by replacing the nitrophenyl of the first series by a cyclohexylamine. The resulting compounds still showed some activity (Table 3), even if it was weaker than the other series, and they are still weaker on CLK3 and DYRK1A than other kinases. Interestingly, the affinity is higher for DYRK1A than the first series. Since the DYRK1A active site is more polar than that of other kinases [14], it is not surprising that a polar inhibitor is better on this enzyme. Docking of 16b suggested that its binding mode is closer to DB18's than 12g's ( Figure S10). The triazole forms a stronger π-stacking with F172, however, and the primary amine forms a salt bridge with D250. Because of this interaction, the inhibitor curls itself up, allowing the meta'-chloro substituent of the phenyl to be more buried in the active site than for DB18. This explains why it has a figure of the same order on CLK1 as the mono chloro 16a.  The very high affinity of the carboxylic acid derivative was a surprise. We were first astonished that the active site could tolerate an electronegative substituent in this region, because it is surrounded by many acidic amino-acids: E169, E206, D288, E292 and D325. Of course, several basic residues are also present, but they are few (K191, K194 and K290), and can only partially equilibrate the charges. The resulting electrostatic clash explains probably why docking experiments suggested a reorientation of the benzoic acid part of the inhibitor, as compared with DB18. We checked this aspect through molecular dynamic simulation, by replacing the para-methyl of DB18 by a carboxylic acid, in the exact binding mode of DB18. After 300 ns, we observed an abrupt shift of the inhibitor position. It seems that the electronegative environment generated by acidic residues, particularly D250 and E292, forced an expulsion of the benzoic acid via a rotation of the phenol dihedral ( Figure S2). We realized thereafter that this acidic region of the active site is precisely the domain encompassing the phosphate groups of the natural substrate ATP of kinases as evidenced by several crystal structures (For example, Figure 4 shows a {DYRK1A:ATP} complex (PDB:7A4O_B)} [21]. Since ATP is often bound to Mg 2+ (see for Overlap of 12g andDB18in CLK1 Overlap of 12g andDB18in DYRK1A  [22], it is probable th bind Mg 2+ in this position, even if the cation has not been seen in X-ray struc The tolerance of CLKs' active site towards very diverse substituents encourage us to test another extreme, with a basic group. To this end, we sy third series by replacing the nitrophenyl of the first series by a cyclohexyla To further evaluate our CLK inhibitors, it will be necessary to use in vivo models, most likely in mouse. So, we also tested the thirdseries on mouse CLK1 orthologue (MmCLK1) and the results are reported in Table 5. The results showed that the inhibitors are also active, or very active, on this species, but that their IC 50 's showed variable differences with the human protein. In the first series, the best inhibitor is 10 times more potent on human than mouse CLK1. In the second and third series, they are in the same range but, interestingly, 12g is the best inhibitor for both species in these new series of analogues.
It was surprising to see such differences between human and mouse IC 50 's in the first series, since the two orthologues share a high-sequence homology (90% sequence identity). In the active site region, the only difference is a H187R mutation. These residues are not in direct contact with inhibitors, and they even face the opposite direction, outside the active site. Therefore, we hypothesized that this sequence variation could have an indirect effect on the ligand affinity. To investigate this hypothesis, we ran molecular dynamics experiments for both human and mouse CLK1. A deep analysis of the Apo conformation of the enzymes showed that the 187 position is more flexible in mouse than in human enzyme, in agreement with the chemical nature of corresponding amino-acids. Consequently, the secondary structures around H/R187 are globally more flexible in the mouse orthologue ( Figure 5).  This flexibility drives a slight instability of the R186-V193 β-strand of the lid domain (β-3), which propagates to the whole active site and limits the affinity of the inhibitors. On the contrary, movements of the loop and the attached β-strand seems to stabilize the upper R160-A171 β-strand in MmCLK1 (β-1), which covers the active site. Taken together, these differential movements simultaneously favoured the induced-fit mechanism in HsCLK1 vsMmCLK1, while perturbing inhibitor/protein contacts in MmCLK1. Interestingly, we already stressed the importance of this domain movement for explaining CLK3 and DYRK1A selectivity of DB18 [14].
To summarize our SAR/molecular modelling analyses, several conclusions can be This flexibility drives a slight instability of the R186-V193 β-strand of the lid domain (β-3), which propagates to the whole active site and limits the affinity of the inhibitors. On the contrary, movements of the loop and the attached β-strand seems to stabilize the upper R160-A171 β-strand in MmCLK1 (β-1), which covers the active site. Taken together, these differential movements simultaneously favoured the induced-fit mechanism in HsCLK1 vs. MmCLK1, while perturbing inhibitor/protein contacts in MmCLK1. Interestingly, we already stressed the importance of this domain movement for explaining CLK3 and DYRK1A selectivity of DB18 [14].
To summarize our SAR/molecular modelling analyses, several conclusions can be drawn. First, the meta substituent of the aniline part of the inhibitors is crucial for its activity, because of three features: (i) an increase in interactions with the protein, (ii) an intramolecular interaction that favours the active conformation of ligands, and (iii) an increased structural stabilization of the active site. Second, the triazole-bearing group (phenyl of cyclohexyl) is very versatile, since the protein tolerate here both hydrophobic (toluene), and hydrophilic (benzyl alcool) portions, and even ionic groups of opposite charge (benzoic acid, cyclohexylamine). Third, the intrinsic flexibility of CLKs' active site has a pronounced effect on inhibitors selectivity, as well as orthologue affinity (human/mouse), and appeared as more important than simple amino acid conservation/mutation.

Cytotoxicity Studies
To complete our study, we performed a cytotoxicity screening of the 15 molecules, investigating 7 representative cancer cell lines. The compound's effect on cell viability was evaluated on Huh-7 (hepatocellular carcinoma), CaCo 2 , HCT-116 (colorectal adenocarcinoma), MCF7, MDA-MB231, MDA-MB468 (breast carcinoma), PC3 (prostate carcinoma) and NCI-H727 (lung carcinoid). Human skin fibroblasts were used as reference for noncancerous cells. Roscovitine, Doxorubicin and Taxol were used as positive controls. The cells were treated for 48 h with 25 µM compounds and fixed in cooled 90% ethanol/5% acetic acid for 20 min. The nuclei were stained with Hoechst 33342 and counted using HCS technology. Compounds that induced more than 30% cell viability decrease when compared to DMSO control treatment (set at 100%), for at least one cell line, were retained for the determination of the IC 50 over a range of 6 concentrations. The results are presented in Table S2.
In the first series, the molecule 7a showed a good cytotoxicity against most of the cancer cell lines and was non-toxic towards fibroblasts. On the contrary, the dichloro molecule 7b exhibited no cytotoxicity, except a very low one with PC3 cells. It was also non-toxic to fibroblasts and therefore its profile was very close to our lead molecule DB18.
In the second series, the results were also very structure-dependent: 12d showed that it was highly cytotoxic for HuH7 cells and not for the others. Further, it had a moderate toxicity for fibroblasts. The two next molecules 12c and 12e demonstrated potent cytotoxicity for all cancer cells and a moderate toxicity towards fibroblasts. Finally, the last two compounds 12f and 12g were non-active against all cancer cells and also towards fibroblasts. Thus, their profile was very similar to our previous lead molecule DB18.
For the third series, the results were sensitive to the structures. All N-Boc protected compounds were non-toxic to fibroblasts. On the other hand, 18a, and to a lesser extent 15a, demonstrated significant cytotoxicities against a good number of the selected cancer cell lines, and 15b and 18a had cytotoxicities more focused on a few (MDA-MB-231 or HuH7). On the contrary, the derivatives with primary amines (16 and 19) showed potent cytotoxicities against the complete series of cancer cells. However, they demonstrated also moderate to high toxicities against the fibroblasts.

Study of DB18 Activity/Selectivity against a Large Panel of 468 Kinases
Since DB18 proved to have the best compromise between a high activity on CLKs and a very high selectivity vs. DYRK1A, it was selected for assay against a large panel of 468 kinases (KINOMEscan SM Assay) at 1 µM. Very interestingly, the compound was found to be highly selective for CLKs and affected to a lower extent HIPK1, 4 and TRKA.
As an example, only 1.4% of CLK2 activity remained after incubation with DB18 (the full list of results is reported in Table S5). The results are shown on a TREEspot™ Kinase dendrogram (DiscoverX) as an illustrative representation of the human kinome phylogenetic tree ( Figure 6). The codes reported on this figure indicate the subclasses of protein kinases: CMGC for CDKs, MAP kinases, GSK and CDK-like kinases; AGC for protein kinase A, G, and C families (PKA, PKC, PKG); CAMK for Ca 2+ /calmodulin-dependent protein kinases; CK1, cell kinases 1 (originally known as Casein Kinase 1); STE, STE kinases (homologs of yeast STErile kinases); TKL, tyrosine kinases-like; TK, tyrosine kinases. Each kinase tested in the assay panel was marked with a green dot. The hit kinases reported were marked with a red circle, except CLK2 which was marked with a blue circle. 0% represents the higher affinity, whereas small green dots indicate that at 1 µM, DB18 cannot inhibit significantly the kinase (as over 25% of the tested kinase are still on the affinity matrix after competition with the tested compound). S-Score (25) = (number of non-mutant kinases with %Ctrl < 25)/(number of non-mutant kinases tested = 403).

Chemical Syntheses
General information For this, see corresponding data in ref [14]. All triazole compounds were purified by flash column chromatography on silica gel using 20% ethyl acetate in hexane, unless otherwise noted. For all final compounds (7, 12, 16 and 19) a purity >95% was established by LC-MS.

Chemical Syntheses
General information: For this, see corresponding data in ref [14]. All triazole compounds were purified by flash column chromatography on silica gel using 20% ethyl acetate in hexane, unless otherwise noted. For all final compounds (7, 12, 16 and 19) a purity >95% was established by LC-MS.
Preparation of compounds 3, 4, 7a, 7b, 13: Compounds 3, 4, 5, 6, 7a, 7b, 13 were prepared as already described for DB18 [14].   After cooling down to rt, water (30 mL) was added and the resulting mixture was stirred at 200 rpm. It was filtered using a Büchner and the yellow precipitated solid was washed several times with water. It was transferred into a bigger flask and dried under vacuum with a pump. It afforded ethyl 4-amino-3-nitrobenzoate (9) (718 mg) whose NMR showed that it was pure. Rinsing by hot methanol the Büchner and the filter paper afforded additional 9 (60 mg) so the overall yield of 9 was 91%; 1 H NMR (300 MHz, acetone-d 6 44 mL) and an olive-shaped magnetic stirring bar were successively introduced. This mixture was stirred at rt until LiCl was dissolved and sodium borohydride (735 mg, 19 mmol) was added. A red color appeared. The reaction flask was flushed under nitrogen, tightly stoppered and stirred at rt for 40 min. At this time, thickening was observed and LiBH(OMe) 3 was formedin situas the reducing species. An excess of pressure (due to hydrogen formation) was evacuated by brief opening before tight stoppering again. This resulting mixture was stirred at 45 • C for 12 h upon which it changed from red to orange. TLC showed a complete reaction with a more polar spot of the expected alcohol (R f = 0.22 with petroleum ether/acetone 70:30 vs. 0.48 for the starting ethyl ester). Ethyl acetate (60 mL) and water (15 mL) were added and stirring at 500 pm was continued for 20-30 min. The remaining mineral mass was washed more than 10 times with stirring at rt with ethyl acetate. Ethyl acetate was then removed in vacuo followed by ethanol and water under high vacuum. The remaining residue and intermediate impure fractions were subjected to 4 successive chromatographies on silica gel (8-9 g) columns with gradient elution with petroleum ether/acetone 85:15 to 75:25 (deposit on silica gel was made with toluene + ethanol). It afforded 4-amino-3-nitrobenzyl alcohol (10e) as a red crystallized solid (57.0 mg, 35%); 1 H NMR (300 MHz, acetone-d 6  37% Aqueous hydrochloric acid (3.2 equiv) was added to a stirred solution of 4-amino-3-nitrophenol (157.6 mg, 1 mmol) in water (2.2 mL). After cooling by an ice bath, an aqueous solution of sodium nitrite (91.7 mg, 1.31 mmol, 1.31 equiv) in water (0.41 mL) was added dropwise. Transfer of NaNO 2 was completed by rinsing with water (0.12 mL). After 1 h upon which the bath temperature had risen from 0 to 2 • C, an aqueous solution of sodium azide (104.7 mg, 1.6 mmol, 1.6 equiv) in water (0.53 mL) was added dropwise under stirring. Transfer of NaN 3 was completed by rinsing with water (0.15 mL). An abundant precipitation of yellow crystals immediately occurred. After 45 min further stirring at 0 • C, the ice bath was removed and the reaction mixture was left aside to stir at rt overnight. The aspect of the reaction mixture had not at all changed so the reaction of N 3 − was probably very fast. Extraction with ethyl acetate readily dissolved the precipitate and the organic extract was washed with water until neutral (3 times). Combined aqueous layers were reextracted with ethyl acetate and this second organic extract was washed with water until neutral (2 times). Combined organic extracts were dried (Na 2 SO 4 ), concentrated and put under vacuum to afford a brown solid. TLC spot of the azide 11c (0.42 with hexane/acetone 7:3) was yellow at the start and became brownish after less than one hour contact with air. Chromatography of the brown crude product on silica gel (2. 1-Azido-4-bromo-2-nitrobenzene (11d): To a stirred solution of 1-amino-4-bromo-2-nitrobenzene (11d) (500 mg, 2.30 mmol) in 6N HCl (10 mL) was added NaNO 2 (317 mg, 4.60 mmol) in 2 mL water at 0 • C. Then stirred for 2 h at 0 • C, after then add NaN 3 (299 mg, 4.60 mmol) lot wise at 0 • C, and stirred for overnight at rt. After completion of the reaction by TLC, the reaction mixture was poured into water and EtOAc. The phases were separated, and the organic layer was washed with water and concentrated to obtain the azide 11d as a yellow solid. Compound directly used for the next step without purification; 1 H NMR (400 MHz, CDCl 3  In a 25 mL round-bottomed flask, 4-amino-3-nitrobenzyl alcohol (19.7 mg, 0.117 mmol) was dissolved under stirring by adding water (1.25 mL), 37% aqueous hydrochloric acid (156 mg, 6 drops) and 95% ethanol (2.35 mL). The resulting orange solution was cooled at 0 • C. An aqueous solution of sodium nitrite (41.4 mg NaNO 2 , 0.594 mmol, 5.08 équiv + 462.4 mg water) was added dropwise under stirring. After 3 h further stirring at 0 • C, an aqueous solution of sodium azide (31.3 mg NaN 3 , 0.479 mmol, 4.1 équiv + 226.4 mg water) was added dropwise. Transfer of NaN 3 was completed by rinsing with water (77.1 mg). Upon addition of NaN 3 , the color of the raction mixture changed from orange to yellow and a precipitation appeared. Stirring was then continued for 100 min at 0 • C and the resulting mixture was then left aside in a freezer overnight. After warming up to rt, TLC showed a less polar spot of azide (R f = 0.35 with petroleum ether/acetone 70:30) and two minor more polar spots, the most polar seemed to be that of starting amine. Water (about 5 mL) and sodium bicarbonate (90 mg) were added. The resulting mixture was extracted 4 times with ethyl acetate. After drying over sodium sulfate and concentration, the remaining orange crude wax (22.4 mg) was subjected to a chromatography on a column of silica gel (5 g) with gradient elution with petroleum ether/acetone 95:5 to 80:20 (deposit on silica gel was made with toluene + a small amount of ethanol with heating). It Methyl 4-azido-3-nitrobenzoate (11f): To a stirred solution of methyl 4-amino-3-nitrobenzoate (10f) (500 mg, 2.55 mmol) in 6N HCl (10 mL) was added NaNO 2 (352 mg, 5.1 mmol) in 2 mL water at 0 • C. Then stirred at 0 • C for 2 h, after then added NaN 3 (331 mg, 5.1 mmol) lot wise at 0 • C, stirred for overnight at rt. After completion of the reaction by TLC, the reaction mixture was poured into water and EtOAc. The phases were separated, and the organic layer was washed with water and concentrated to obtain the azide 11f as a yellow solid. This compound was directly used for the next step without purification. In a separating funnel, water (0.42 mL), brine (saturated aq. NaCl: 1.57 mL) and 28-30% aqueous ammonium hydroxide (3 drops) were introduced to which was added the reaction mixture which was transferred by ethyl acetate and THF followed by diluted brine in order to transfer residual minerals which were remaining in the reaction flask. After partitioning, the aqueous phase was reextracted with ethyl acetate plus THF. Combined organic extracts were washed once with brine, dried (Na 2 SO 4 ), concentrated and the remaining residue was put under vacuum. Chromatography on silica gel (2.1 g) column, which was loaded with dichloromethane (deposit of the crude product on the silica gel by dissolving in THF plus a little bit of methanol with smooth heating) eluting with 2% MeOH-chloroform, afforded triazole 12c as a light orange powder (65.  4-(4-(((2-((3-Chlorophenyl)amino)quinazolin-8-yl)oxy)methyl)-1H-1,2,3-triazol-1-yl)-3nitrobenzyl alcohol (12e): In a 25 mL round-bottomed flask containing 4-azido-3-nitrobenzyl alcohol (11e) (10.7 mg, 0,055 mmol) was added alkyne 5 (17.1 mg, 0.055 mmol), copper(II) sulfate pentahydrate (2.1 mg, 0.00824 mmol, 0.15 equiv), sodium ascorbate (5.2 mg, 0.026 mmol, 0.47 equiv), urea (1.8 mg, 0.0294 mmol, 0.53 equiv), water (53.1 mg, 2 drops), DMSO (105.0 mg, 6 drops), THF (0.553 mL) and a small olive-shaped magnetic stirring bar. The reaction flask was flushed under nitrogen, tightly stoppered and left under moderate stirring at rt for 4.5 h. TLC showed a complete reaction with the expected triazole as a very polar compound (R f~0 .04 with petroleum ether/acetone 70:30 vs. 0.45 for the starting alkyne and 0.35 for the starting azide. In a separating funnel, water (1 mL) and 25% aqueous ammonium hydroxide (4 drops) were introduced to which was added the reaction mixture which was transferred by ethyl acetate (8.3 mL overall). A fine brick red precipitate appeared (presumably Cu 2 O). After partitioning, the aqueous phase was reextracted with ethyl acetate (3-4 mL). Combined yellow organic extracts were dried (Na 2 SO 4 ), concentrated and the brown oily residue was put under vacuum. Chromatography on silica gel (6. To a stirred solution of alkyne 13 (100 mg, 0.32 mmol) in t-BuOH: water (1:1) (2 mL) was added azide 11f (134 mg, 0.64 mmol), sodium ascorbate (128 mg, 0.64 mmol), CuSO 4 .5H 2 O (161 mg, 0.64 mmol) in 4 mL water at rt. Then stirred at rt for overnight. After completion of the reaction by TLC, the reaction mixture was poured into water and EtOAc. The phases were separated, and the organic layer was washed with brine solution, dried over sodium sulphate and concentrated. Crude purified by column chromatography (60-120 silica gel, plane ethyl acetate) to afford triazole 12f (40 mg, 23%) as a brown solid. 1 H NMR (300 MHz, DMSO-d 6 2-((3-Chlorophenyl)amino)quinazolin-8-yl)oxy)methyl)-1H-1,2,3-triazol-1-yl)-3nitrobenzoic acid (12g): To a stirred solution of methyl ester 12f (30 mg, 0.0564 mol) in MeOH/THF/water (3:2:1) (3 mL) was added LiOH.H 2 O (7.0 mg, 0.169 mol) and stirring was continued at rt for 16 h. After completion of the reaction by TLC, the reaction mixture was concentrated, diluted with water and acidified with 2N aqueous HCl. Then the solid which was formed was filtered and dried to obtain carboxylic acid 12g as a yellow solid (15 mg, 52%). 1 (14): Step 1. Synthesis of trans-4-(tert-butoxycarbonylamino)cyclohexyl methanesulfonate: To a stirred solution of trans-4-Boc-aminocyclohexanol (500 mg, 2.32 mmol) in DCM (5 mL) was added TEA (1 mL, 6.97 mmol) at rt and stirred for 15 min. Then added Ms-Cl (0.3 mL, 3.48 mmol) drop wisely, stirred at RT for 6 h. After completion of the reaction by TLC, the reaction mixture was poured into water (5 mL). The phases were separated, and the organic layer was washed with a sat aq. NaHCO 3 solution (5 mL), dried over sodium sulfate and concentrated and dried to obtain the trans-4-(tert-butoxycarbonylamino)cyclohexyl methanesulfonate as a colorless liquid.
Step 2. Synthesis of tert-butyl (cis-4-azidocyclohexyl)carbamate (14): To a stirred solution of trans-4-(tert-butoxycarbonylamino)cyclohexyl methanesulfonate (500 mg, 1.70 mmol) in DMF (3 mL) was added NaN 3 (1.1 g, 17.06 mmol) at rt. Then heated to 80 • C for 24 h. After completion of the reaction by TLC, the reaction mixture was poured into ice water, solid was formed, filtered and dried to obtain the azide 14 as an off-white solid. This compound was directly used for the next step without purification.

Conclusions
The compound DB18 was established recently as a very potent inhibitor of the human CLKs kinases, with a remarkable selectivity vs. human DYRK1A. Starting from this lead molecule, we designed three series of analogues in order to explore the roles of the different parts of this compound: -the substituents on the aniline, -the tolyl group in the top part of DB18 anchored into CLK1 and,-the central part of this lead.
Fifteen new analogues have been designed, prepared and analysed for their inhibitory potency on various CLKs plus DYRK1A kinases, as well as for cytotoxicity studies. SAR data were analyzed through molecular docking experiments which afforded useful rationales for the observed activities. Furthermore, molecular dynamic studies highlighted the importance of protein kinases flexibility to explain inhibitors selectivity.
Although it had a slightly different binding mode compared to DB18, 12 gemerged as the most promising compound for this new series of molecules. It exhibited CLK inhibitory properties similar to those obtained with DB18, albeit with a slight drop in selectivity vs. DYRK1A. Based on these results, a kinome scan assay was performed on DB18 on 463 human kinases. It fully demonstrated the high selectivity of this molecule on CLKs, with limited extra activity only on HIPKs and TRKA.
Altogether, these results confirmed the qualities of the 2-anilinoquinazoline scaffold in the design of CLK inhibitors and the potentialities of DB18 and 12g as potent inhibitors of human CLK-1, -2 and -4 kinases, with low to no action on DYRK1A. Therefore, these compounds should be useful tools, for example, to explore the possible roles of CLKs' kinases in various aspects of human health.  Table S1: Primary evaluation of the inhibition of synthetized quinazolines against a short panel of mammalian kinases; Table S2: Cytotoxic studies (IC50, in µM) of the synthetized quinazolines; Figure S1: Secondary structure composition of the HsCLK1/DB18 (up) and HsCLK1/7a (down) complex during 1 µs molecular dynamics; Figure Table S3: Prime-based MM-GBSA energies in CLK1; Table S4: Prime-based MM-GBSA energies in DYRK1A; Table S5: Full data of kinome scan study for DB18: Inhibition of compounds against mammalian kinases; Cytotoxicity studies; Molecular docking/dynamics experiments; Full data of kinome scan study of DB18; Part 2: Views of NMR spectra. supported by the Conseil Régional de Bretagne) for supporting the KISSf screening facility (FR2424, CNRS and Sorbonne Université), Roscoff, France.