Quantitative Structure-Activity Relationship (QSAR) of indoloacetamides as inhibitors of human isoprenylcysteine carboxyl methyltransferase

Abstract

A QSAR is developed for the isoprenylcysteine carboxyl methyltransferase (ICMT) inhibitory activities of a series of indoloacetamides (n = 71) that are structurally related to cysmethynil, a selective ICMT inhibitor. Multivariate analytical tools (principal component analysis and projection to latent structures), multi-linear regression and comparative molecular field analysis (CoMFA) are used to develop a suitably predictive model for the purpose of optimizing and identifying members with more potent inhibitory activity. The resulting model shows that good activity is determined largely by the characteristics of the substituent attached to the indole nitrogen, which should be a lipophilic residue with fairly wide dimensions. In contrast, the substituted phenyl ring attached to the indole ring must be of limited dimensions and lipophilicity.

A large number of mammalian proteins contain the CaaX box motif, where C is cysteine, a is generally an aliphatic residue, and X is one of several different amino acids. These proteins undergo a sequential three-step post-translational modification involving (i) prenylation of the cysteine residue mediated by one of two soluble isoprenyltransferases (protein farnesyltransferase FTase or protein geranylgeranyltransferase type I (GGTase-1), ¹ (ii) proteolytic removal of the aaX residues by the endoprotease Ras-converting enzyme 1 (Rce1), ² and (iii) methylation of the newly exposed prenylcysteine by the enzyme isoprenylcysteine carboxyl methyltransferase (ICMT). ^2,3,4 These changes result in a more hydrophobic protein with a unique structure at the C-terminus that serves as a specific recognition motif in protein-protein interactions.^5,6,7 Several CaaX proteins including members of the Ras family of GTPases have been implicated in oncogenesis and tumour progression. Processing by the prenylation pathway is widely thought to contribute to these roles. ^{6, 8} For this reason, the enzymes involved in this pathway have received considerable attention as targets in drug discovery programs. ⁹.¹⁰ A number of FTase inhibitors (FTIs) with high efficacy and low toxicity have been evaluated in clinical trials, but their efficacies in patients have turn out to be less than initially expected. ¹¹ As a consequence, attention has shifted to the post-prenylation enzymes Rce1 and ICMT as potential alternative targets to FTase, in particular developing inhibitors to ICMT which catalyzes the final step of methylation of the prenylcysteine. ^12-18 Recently, an indole-based selective inhibitor of ICMT was identified through the screening of a diverse chemical library comprising of over 70 subfamilies derived from unique scaffolds. ¹⁵ This compound called cysmethynil [compound 1D in Table 1] had an in vitro IC₅₀ of 2.4 μM in the initial screen for enzyme inhibitory activity.

Table 1.

Structures and experimental pIC₅₀^a values of compounds in database

No	R 1	R 2	pIC 50	No	R 1	R 2	pIC 50
1A1	3-methylphenyl	isobutyl	4.80	5A1	2-methoxyphenyl	isobutyl	4.66
1B1	3-methylphenyl	cyclopropylmethyl	5.04	5B1	2-methoxyphenyl	cyclopropylmethyl	4.79
1C1	3-methylphenyl	n-hexyl	5.13	5C1	2-methoxyphenyl	n-hexyl	5.09
1Db	3-methylphenyl	n-octyl	5.68	5D	2-methoxyphenyl	n-octyl	5.22
1E	3-methylphenyl	benzyl	5.19	5E	2-methoxyphenyl	benzyl	4.88
1G1	3-methylphenyl	3-trifluoromethylbenzyl	5.10	5G1	2-methoxyphenyl	3-trifluoromethylbenzyl	5.21
1H1	3-methylphenyl	2-naphthylmethyl	4.76	5H1	2-methoxyphenyl	2-naphthylmethyl	4.53
1I1	3-methylphenyl	3-phenoxypropyl	4.00	5I1	2-methoxyphenyl	3-phenoxypropyl	4.00
1J1	3-methylphenyl	H	4.00	5J1	2-methoxyphenyl	H	4.00
2A1	3-fluorophenyl	isobutyl	5.43	6A1	3-chloro-4-fluorophenyl	isobutyl	5.04
2B1	3-fluorophenyl	cyclopropylmethyl	4.83	6B1	3-chloro-4-fluorophenyl	cyclopropylmethyl	5.24
2C1	3-fluorophenyl	n-hexyl	4.89	6C1	3-chloro-4-fluorophenyl	n-hexyl	5.03
2D	3-fluorophenyl	n-octyl	5.62	6D	3-chloro-4-fluorophenyl	n-octyl	5.15
2E	3-fluorophenyl	benzyl	4.97	6E	3-chloro-4-fluorophenyl	benzyl	4.98
2F 1	3-fluorophenyl	4-tert-butylbenzyl	5.30	6F1	3-chloro-4-fluorophenyl	4-tert-butylbenzyl	5.12
2G1	3-fluorophenyl	3-trifluoromethylbenzyl	5.24	6G1	3-chloro-4-fluorophenyl	3-trifluoromethylbenzyl	5.21
2I1	3-fluorophenyl	3-phenoxypropyl	4.00	6I1	3-chloro-4-fluorophenyl	3-phenoxypropyl	4.00
2J1	3-fluorophenyl	H	4.40	6J1	3-chloro-4-fluorophenyl	H	4.00
3A1	4-methylphenyl	isobutyl	5.01	7A1	3,5-bis(trifluoromethyl)phenyl	isobutyl	4.59
3B1	4-methylphenyl	cyclopropylmethyl	4.95	7B1	3,5-bis(trifluoromethyl)phenyl	cyclopropylmethyl	4.00
3C1	4-methylphenyl	n-hexyl	5.13	7D	3,5-bis(trifluoromethyl)phenyl	n-octyl	4.43
3D	4-methylphenyl	n-octyl	4.97	7E	3,5-bis(trifluoromethyl)phenyl	benzyl	4.48
3G1	4-methylphenyl	3-trifluoromethylbenzyl	5.19	7F1	3,5-bis(trifluoromethyl)phenyl	4-tert-butylbenzyl	4.53
3H1	4-methylphenyl	2-naphthylmethyl	4.71	7G1	3,5-bis(trifluoromethyl)phenyl	3-trifluoromethylbenzyl	4.00
3J1	4-methylphenyl	H	4.00	7H1	3,5-bis(trifluoromethyl)phenyl	2-naphthylmethyl	4.00
4A1	3-ethoxyphenyl	isobutyl	5.21	7I1	3,5-bis(trifluoromethyl)phenyl	3-phenoxypropyl	4.00
4B1	3-ethoxyphenyl	cyclopropylmethyl	5.19	7J1	3,5-bis(trifluoromethyl)phenyl	H	4.00
4C1	3-ethoxyphenyl	n-hexyl	5.18	8A1	4-phenoxyphenyl	isobutyl	5.23
4D	3-ethoxyphenyl	n-octyl	5.66	8B1	4-phenoxyphenyl	cyclopropylmethyl	4.00
4E	3-ethoxyphenyl	benzyl	5.11	8C1	4-phenoxyphenyl	n-hexyl	4.00
4F1	3-ethoxyphenyl	4-tert-butylbenzyl	4.90	8E	4-phenoxyphenyl	benzyl	5.03
4G1	3-ethoxyphenyl	3-trifluoromethylbenzyl	5.31	8G1	4-phenoxyphenyl	3-trifluoromethylbenzyl	4.00
4H1	3-ethoxyphenyl	2-naphthylmethyl	5.08	8H1	4-phenoxyphenyl	2-naphthylmethyl	4.00
4I1	3-ethoxyphenyl	3-phenoxypropyl	4.00	8I1	4-phenoxyphenyl	3-phenoxypropyl	4.00
4J1	3-ethoxyphenyl	H	4.00	8J1	4-phenoxyphenyl	H	4.00

^aDetermination of pIC₅₀ is given in Reference ¹⁹

^bCysmethynil

The indole-based library from which cysmethynil was identified consists of some 70 analogues which differ in the substitution at the phenyl ring and the indole nitrogen (Table 1). These compounds were screened for ICMT inhibitory activities ¹⁹ and had IC₅₀ values ranging from 2 μM to more than 50 μM. ²⁰ Thus, they constitute a valuable database from which quantitative structure-activity relationships (QSAR) can be deduced for the purpose of optimizing and identifying indoloacetamides with more potent inhibitory activity. The objective of this work is to develop suitably predictive QSAR models that can achieve these aims.

To develop the QSAR, several descriptors representative of size, electronic and lipophilic characteristics were used to characterize the compounds. Area, volume (including polar surface area and polar volume), and the Sterimol parameters (L1, B1, B5) ²¹ of the substituted phenyl ring and the group attached to the indole nitrogen were chosen to characterize the size component. Lipophilicity was represented by ClogP and Hansch π values of the substituted phenyl ring (π_Ph) and the N-substituent (π_N). The electronic character of the compounds were captured by HOMO, LUMO, molar refractivity, dipole moment, Hammett constant (σ_aromatic) of the phenyl substituents and the inductive constant (σ*) of the group attached to the indole nitrogen. The parameters were determined from force field minimized structures of the compounds using commercially available software ²² and are listed in Supplementary Information (Table 1).

Our first approach was to employ multivariate tools (principal component analysis PCA, and projection to latent structures PLS) to analyze the structure-activity relationship. ²³ PCA, a pattern recognition technique, serves to summarize information in a form that can reveal relationships between compounds and the parameters used to characterize them. ²⁴ It is particularly useful in situations where many of the parameters are correlated, as in this case. PCA will then transform the correlated variables to a new set of uncorrelated variables called principal components. When the data set (72 compounds, 20 parameters) was analyzed by PCA, a significant 3 component model (r² =0.72, q² = 0.54) was obtained, which implies that the properties captured by these components can account for 72% of the observed variation, at a predictability level of 54%. An examination of the loading plot of this model (Figure 1a) shows that the 1^st component receives input from size (volume, area, molar refractivity, Sterimol parameters of N) and lipophilicity (ClogP, π_N). The parameters contributing to the 2^nd component are HOMO, lipophilicity (π_Ph) of the substituted phenyl ring, and the Sterimol length (PL1) and width (PB1, PB5) parameters of the substituted phenyl ring. Interestingly, the 1^st component receives steric and lipophilicity inputs from the N substituent in contrast to the less important 2^nd component which is characterized by the same parameters of the phenyl ring. The score plot which depicts the distribution of the 72 compounds based on the properties captured by the 1^st and 2^nd principal components is shown in Figure 1b. Most of the compounds are distributed uniformly within the ellipse, with two exceptions, namely compounds with no N substituent (lower left quadrant) and compounds with m,m-bis(trifluromethyl) substitution on the phenyl ring (lower right quadrant). The isolation of these two subsets, which incidentally are among the weakest inhibitors of the series, can be attributed to their less than optimal steric and lipophilic properties which are encoded in the 1^st and 2^nd principal components.

Figure 1.

Figure 1a Loading plot of 1^st and 2^nd principal components (p[1], p[2]) of 72 compounds and 20 descriptors.

Figure 1b Score plot of principal components t₁ vs t₂ for compounds (n = 72, 20 descriptors). Compounds are identified in Supplementary Information (Table 1). The ellipse corresponds to the confidence region based on Hotelling T² (0.05). Compounds identified by “ 7” have m.m-bistrifluromethyl groups on the phenyl ring and compounds identified by “J” have no substituent on the indole N. Both sets of compounds are isolated from the rest.

Figure 1c Coefficent Plot for PLS model A derived from 70 compounds and 14 descriptors, based on the 1^st component. Descriptors with positive coefficients are directly related to activity while those with negative coefficients are inversely related. Magnitude of parameter indicates its relative contribution to the model.

PSA = polar surface area; PV = polar volume; piPh2 = Hansch constant π of substituted phenyl ring; pi N = Hansch constant π of N substituent; CMR = molar refractivity; PL1,PB1, PB5 = Sterimol parameters of phenyl ring; NL1, NB1, NB5 = Sterimol parameters of N substituent; HOMO = highest occupied molecular orbital; LUMO = lowest unoccupied molecular orbital.

Next, we went on to develop a projection model for predicting biological activity from the principal components. This is achieved with PLS, a regression extension of PCA. The 1^st PLS model was obtained with all compounds and parameters. This model accounted for 68% of the variation in biological activity (r² = 0.68) at a predictability level of 49% (q² =0.49). It was improved by omitting 2 outliers which exhibited large differences between the observed and predicted values (compounds 3E and 8D) and removing parameters that did not make significant contributions to activity (dipole moment, σ_N, σ_Ph) or are duplicated by existing parameters (volume, surface area, π_Ph). The resulting model (A) had an r² of 0.72 and q² of 0.60. A test set of 12 compounds was arbitrarily selected from the 70 compounds and used to predict the activities of the remaining 58 compounds. The level of predictability, as measured by the root mean square of prediction (RMSEP), was just satisfactory at 0.48. Figure 1c is the coefficient plot of the final PLS model A. This plot identifies the parameters that contributed most to activity (as reflected by the length of the bar), and the nature of the correlation (direct = positive coefficient or inverse = negative coefficient). Thus, the most significant parameters (in order of decreasing importance) are polar surface area and polar volume > lipophilicity of substituted phenyl ring (π_Ph) > width parameters B1, B5 of the N substituent and π_N > Sterimol L1 of N substituent. The coefficient plot nicely illustrates the contrasting requirements of the N and phenyl ring substituents. For good activity, the N group must be lipophilic and have large width dimensions. In contrast, the substituted phenyl ring should be less lipophilic and have sterically smaller dimensions.

To further validate PLS model A, the same database was used to derive a stepwise multiple linear regression (MLR) equation. ²⁵ The best equation (1) has four descriptors (polar surface area PSA, polar volume PV, Sterimol parameter B1 of the substituted phenyl ring and lipophilicity contribution of the substituted phenyl ring π_Ph). When the coefficients in equation 1 are standardized, the relative contribution of each parameter is of the order PSA > PV > PB1 > π_Ph. The parameters are inversely related to activity, with the 1^st three parameters significant at p < 0.001 and π_Ph significant at p = 0.001.

$$\begin{array}{c}{\mathrm{pIC}}_{50}=10.23(\pm 0.55)-0.016{(\pm 0.003)}^{\ast }\mathrm{PSA}-0.010{(\pm 0.002)}^{\ast }\mathrm{PV}-0.835(\pm \hfill \\ 0.195)\ast \mathrm{PB}1-0.190{(\pm 0.056)}^{\ast }\pi \mathrm{Ph}.\hfill \\ \mathrm{n}=70,\mathrm{r}=0.811,{{\mathrm{r}}^2}_{\text{adjusted}}=0.636,\text{standard error of estimate}=0.32,\mathrm{F}=30.69.\hfill \end{array}$$ Equation 1:

The Sterimol parameters and lipophilicity contribution of the N substituent are not found in Equation 1, but are encoded in polar surface area, to which they are significantly correlated (Supporting Information, Table 2). The inclusion of PB1 in equation 1 is notable because PB1 did not contribute significantly to the PLS model. PB1 is also not strongly correlated to the other Sterimol parameters (PB5, PL1) but received a significant contribution from HOMO.

Table 2.

Summary of CoMFA Analysis of Training Set (n = 56)

Cross validated correlation coefficient q2	0.646
Standard error of prediction (SEP)	0.342
Number of Componentsa	7
Non cross-validated correlation coefficient r2	0.868
Standard error of estimate (SEE)	0.209
F value	44.91
Field Contributions
Steric	0.68
Electrostatic	0.32
r2 predb	0.601

^aOptimum number of components obtained from cross-validated PLS analysis. This number of components is also used in the final non cross-validated analysis.

^bPredictive r² is based only on the 13 compounds not included in the training set. It is determined from r² _pred = [SD – PRESS] / SD where SD = sum of the squared deviations between IC₅₀ of compounds in test set and mean IC₅₀ of training set molecules, and PRESS is sum of the squared deviations between predicted and actual IC₅₀ for every member in the test set.

At this juncture, some deductions can be made for the structural requirements for inhibitory activity of the indoloacetamides. It is evident that there is a requirement for compounds with small polar surface areas, a feature that is largely determined by the dimensions and lipophilicity of the N-substituent. The relationship is inverse, so good activity is associated with lipophilic N substituents. The bulk of the N substitutent is also critical and in this context, the width parameters of the group (B1, B5) are just as important as the length of the group (L1). Thus, the poor activity of compounds with N-3-phenoxypropyl substituent may be traced to its lack of bulk (long and narrow). This group is almost as long as the N-octyl substituent found in cysmethynil but the N-octyl group is more flexible and thus has a larger width (NB5) than the N-4-phenoxypropyl group. Another significant observation is that the substituted phenyl ring had a lesser impact on activity compared to the N group. In contrast to the preference for more lipophilic and bulkier N substituents, the substituted phenyl ring must be kept small (width and length) and less lipophilic. Thus, the poor activities of compounds with 3,5-bis (trifluoromethyl)phenyl and 4-phenoxyphenyl substituents can be traced to the unfavourable width and length of these groups respectively.

Further confirmation of the QSAR developed so far was sought from comparative molecular field analysis (CoMFA). ²⁶ Based on a preliminary analysis, three compounds (8A1, 3E and 8D) were omitted because of large residual values. The remaining 69 compounds were divided into a training set (n = 56) and a test set (n = 13). Compounds in both sets had the same range of biological activity and were well represented in terms of the type of substituents on the indole nitrogen and phenyl rings. The best CoMFA model for the training set had a cross validated q² of 0.646 (7 principal components), a non cross-validated r² of 0.868 and standard error of estimate (SEE) of 0.209 (Table 2). The steric and electrostatic field contributions were 68% and 32% respectively, indicating a greater steric influence on activity. The training set was able to predict the activity of the test set compounds with an r² _pred of 0.601. A plot of predicted versus observed values is given in Figure 2.

Figure 2.

Plot of the predicted pIC₅₀versus observed pIC₅₀ values of n= 69 compounds (compounds omitted are 3E, 8A1 and 8D) based on best CoMFA model. r² _pred for Training set = 0.868, r² _pred for Test set = 0.601.

Visualization of the steric contours of this model shows a swathe of green distributed at the proximal and distal ends of the N substituent. A yellow patch is interspersed between these green zones (Figure 3a). The inference is that there are two optimal lengths of the N substituent corresponding to the green contours - a short group that is within reach of the proximal green patch and a longer group that coincides with the distal green patch. The shorter N-3-trifluromethylbenzyl and longer N-octyl groups, both of which are associated with good activity, fall within these green zones. On the other hand, groups linked to poor activity like N- (2-naphthyl)methylene and N-(4-tert-butylbenzyl) protruded into the disfavoured yellow zone. Interestingly, only CoMFA revealed the presence of two optimal lengths of the N-substituent. The other models emphasized the preference for a bulky N-group that presumably extended into the green zones. Earlier, the poor activity of the N-3-phenoxypropyl group was attributed to its narrow dimensions. The CoMFA steric contour map shows that the phenyl ring of this substituent extended into the sterically disfavoured yellow zone. Only yellow steric contours were found in the vicinity of the phenyl ring, but these were sparse compared to the yellow zones around the N-group. The inference is that the steric characteristics of the phenyl ring contributed less to activity and there is a preference for less bulk at this position.

Figure 3.

Figure 3a Steric map from the CoMFA model showing the alignment based on the indole ring. Green contours (contribution level of 80%) represent areas where steric bulk will enhance activity, and yellow contours (contribution level of 20%) highlight areas which should be kept unoccupied for increased activity.

Figure 3b Electrostatic map from the CoMFA model showing the same alignment as in Figure 3a. Blue contours (contribution level of 85%) represent regions where an increase in positive charge will enhance activity, and red contours (contribution level of 15%) highlight areas where more negative charge is favoured.

None of the models investigated so far has identified electrostatic characteristics to be important contributors to activity, and this was further confirmed by the CoMFA model. The main feature in Figure 3b that can be correlated with the findings from PLS and MLR were the red contours around the phenyl ring, which implied that the ring should be electron rich for activity. As shown in Equation 1, the width parameter PB1 of the phenyl ring is inversely correlated to activity. PB1 is significantly correlated to HOMO (Pearson correlation coefficient - 0.616, p < 0.0001). Molecules with large HOMO imply electron rich structures. Thus, the poor activity associated with the 3,5-bis(trifluromethyl) substituted phenyl ring may be attributed to its steric and electrostatic features. Sterically, it is too wide (B1) and the electron withdrawing trifluromethyl groups render the phenyl ring electron deficient (low HOMO).

To put our models to the test, several compounds with different substituents on the indole nitrogen and phenyl ring were considered and their activities predicted from Equation 1 and the CoMFA model. For example, 2-(1-octyl-5-m-methoxy-1H-indol-3-yl)acetamide has a predicted pIC₅₀ of 5.57 based on Equation 1 and 5.52 based on CoMFA. Noting that the predicted pIC₅₀ values of cysmethynil are 5.72 (Equation 1) and 5.32 (CoMFA), the synthesis of this compound is worth considering.

In conclusion, we have used different approaches to establish the QSAR of indoloacetamides as inhibitors of ICMT. The approaches are complementary and serve to give a more definitive picture of the structural requirements for activity. These are (i) the presence of a lipophilic residue on the indole nitrogen, in contrast to the requirement for a less lipophilic substituted phenyl ring, (ii) the presence of two optimal lengths for the N group which should also have fairly wide dimensions, and (iii) the phenyl ring, which unlike the N-substituent, should have not have bulky substituents and must be electron-rich. These findings also serve to delineate the pharmacophore for activity but as it stands, remains incomplete because the changes made so far have focused on the N-substituent and the phenyl ring. The primary amide side chain has not been structurally altered and a more comprehensive picture will emerge when concurrent changes are made at this location. In any case, the QSAR discussed in this report will provide directions for future structural modifications of this class of compounds and prioritizing candidates for synthesis.