Quantitative Structure-Activity Relationship Studies of a Series of Sulfa Drugs as Inhibitors of Pneumocystis carinii Dihydropteroate Synthetase

Abstract

Sulfone and sulfanilamide sulfa drugs have been shown to inhibit dihydropteroate synthetase (DHPS) isolated from Pneumocystis carinii. In order to develop a pharmacophoric model for this inhibition, quantitative structure-activity relationships (QSAR) for sulfa drugs active against DHPS have been studied. Accurate 50% inhibitory concentrations were collected for 44 analogs, and other parameters, such as partition coefficients and molar refractivity, were calculated. Conventional multiple regression analysis of these data did not provide acceptable QSAR. However, three-dimensional QSAR provided by comparative molecular field analysis did give excellent results. Upon removal of poorly correlated analogs, a data set of 36 analogs, all having a common NHSO₂ group, provided a cross-validated r² value of 0.699 and conventional r² value of 0.964. The resulting pharmacophore model should be useful for understanding and predicting the binding of DHPS by new sulfa drugs.

Pneumocystis carinii can be considered a protozoan organism, but it shares many characteristics with fungi (5). As one of the more common AIDS-defining illnesses, P. carinii pneumonia is often the ultimate cause of morbidity for immunocompromised individuals. As such, treatment and long-term prophylaxis of P. carinii pneumonia continue to be active areas of study. Currently, the most effective treatment is co-trimoxazole, a synergistic combination of the sulfanilamide sulfamethoxazole and the dihydrofolate reductase inhibitor trimethoprim. The use of co-trimoxazole, however, produces adverse side effects in 20 to 57% of patients, resulting in discontinuation of treatment (12). Other combination treatments, including pyrimethamine and either sulfadiazine or sulfadoxine, are also plagued with side effects which limit their use and give rise to the need for more effective treatments (13).

The antimetabolite sulfa drugs function by inhibiting dihydropteroate synthetase (DHPS), an enzyme catalyzing a crucial step in the biosynthesis of tetrahydrofolate and, ultimately, nucleotides. The lack of a mechanism for uptake of preformed folate in P. carinii and the absence of a DHPS pathway in mammals make this enzyme an ideal target for drug therapy. As noted in a recent paper by Hong et al. (7), of the many sulfa drugs that have been synthesized, few have been tested against P. carinii. Given that sulfa drugs differ in their ability to elicit adverse effects, an exploration of the structure-activity relationships (SAR) of a variety of sulfa drugs could lead to more effective treatments. To aid in this exploration as well as to gain further insight into the SAR of these compounds, we have performed quantitative SAR (QSAR) studies using a conventional QSAR methodology as well as comparative molecular field analysis (CoMFA) (4). CoMFA is a powerful three-dimensional (3-D) QSAR method based on the observation that receptor-ligand interactions are largely shape dependent and noncovalent. In a CoMFA analysis, each molecule is defined by a set of steric and electrostatic points in space (8, 9). These representations are stored and then mathematically correlated with activity data by partial least-squares (PLS) analysis. A pharmacophoric model can be derived in 3-D space by using cross-validation in this analysis, and the resulting extensive equation is subjected to statistical analysis to establish the predictive ability of the model as indicated by a cross-validated r² value of greater than 0.5. Additional output from the statistical analysis includes a measure of the reproducibility of the model (conventional r² value) as well as standard error and an F value. Visual output of the QSAR equation is provided by contour plots, which indicate to the viewer those areas about a template molecule that are predicted to improve or decrease potency. Hypothetical structures are designed on the basis of these contour plots, and their activities can be predicted, leading to a cycle of synthesis, testing, and refinement of the pharmacophoric model.

MATERIALS AND METHODS

Data set.

The in vitro sulfa drug activities used in this study were previously determined by Hong et al. using P. carinii DHPS in a cell-free system (7). Fourteen compounds (no. 1 to 14) with originally reported 50% inhibitory concentrations (IC₅₀s) of >10 μM were reexamined in the laboratory by the previously reported procedure (7) (with exceptions noted below), and the resultant data are reported in Tables 1 and 2. The sulfa drugs used in the study by Hong et al. included sulfones and aryl sulfanilamides with structural variations as follows: (i) the nature of the amide aryl group, (ii) the substituent type and substitution pattern of the amide aryl group, and (iii) the substitution on the 4-aminoaryl ring.

TABLE 1.

Structures and activities of reexamined sulfones  

Compound no.	Compound	Substituent at position:				IC50 (μM)a
		R1	R2	R3	R4
1	Diformyldapsone	-NHCOH	-NHCOH			621.8b
2	Difluorodinitrophenylsulfone	-F	-F	-NO2	-NO2	946.4b

^aInhibition was tested at concentrations of 100 and 500 μM, and the resulting data were pooled with earlier inhibition data to generate IC₅₀s as reported previously (7). 

^bThe calculated IC₅₀ is outside of the experimental range.  

TABLE 2.

Structures and activities of various reexamined sulfanilamides  

Compoundf	Substituent at position:			IC50 (μM)
	R1	R2	R3

^aInhibition was tested at concentrations of 100 and 500 μM, and the resulting data were pooled with earlier inhibition data to generate IC₅₀s as reported previously (7). 

^bInhibition was tested at a concentration of 100 μM, and the resulting data were pooled with earlier inhibition data to generate IC₅₀s as reported previously (7). 

^cInhibition was tested at a concentration of 500 μM, and the resulting data were pooled with earlier inhibition data to generate IC₅₀s as reported previously (7). 

^dThe calculated IC₅₀ is outside of the experimental range. 

^e—, a negative correlation between the drug concentration and inhibition was noted. 

^fNumbers in boldface are compound numbers.  

Testing of sulfa compounds.

DHPS assays were carried out on compounds 1 to 14 as previously described by Hong et al. (7). The enzyme assay buffer contained 40 mM Tris-HCl (pH 8.2), 5 mM MgCl₂, 10 mM dithiothreitol, 66 nM p-aminobenzoic acid (PABA; made as a mixture of 16 nM [³H]PABA and 50 nM unlabelled PABA), and 100 μM freshly prepared reduced 6-hydroxymethylpterin pyrophosphate. The reaction was initiated by the addition of Spodoptera frugiperda 9 cell lysates containing 4 U of enzyme (1 U being the amount of enzyme required to catalyze the production of 1 pmol of 7,8-dihydropteroate per h at 37°C). After 1-h incubations, the reactions were stopped by adding 300 μl of 1 M citrate-phosphate buffer, pH 3.8. Using a modified ether extraction method, the radioactive 7,8-dihydropteroate formed was separated from unreacted [³H]PABA and the radioactivity was measured in a scintillation counter.

To determine the IC₅₀s, stock solutions of each sulfa drug were prepared in dimethyl sulfoxide (DMSO) and then diluted to 100 and 500 μM in water. As opposed to the previous assay conditions, the DMSO concentration was sometimes as high as 6%. These high concentrations of DMSO had no effect on enzyme activity. These data were pooled with earlier inhibition data and analyzed by linear regression to generate IC₅₀s as reported previously (7). Compound NSC74428-i (no. 35) was dropped from all analyses due to the observance of a negative correlation between the drug concentration and inhibition.

Computational approach.

Calculations were performed on a Silicon Graphics Indigo 2 workstation equipped with an Impact processor. CoMFA and structure generation were executed by the Tripos Associates SYBYL version 6.2 molecular modeling package with a QSAR module (15). Conformational searches were performed with the MacroModel program (3), and conventional QSAR was performed with Tsar software provided by the Oxford Molecular Modeling Group (11). The default SYBYL, MacroModel, and Tsar settings were used unless otherwise noted.

Conventional QSAR studies.

Using the Tsar suite of programs, QSAR studies were performed on the original data set of 44 molecules. The dependent variable was defined as the inverse log of the IC₅₀ calculated to three significant figures. Two independent variables were incorporated into this QSAR study. The first was the partition coefficient (log P), a quantitative measurement of the hydrophobicity of a molecule calculated by summing the log P contributions of the individual fragments of a compound. These standard fragment values came from the Tsar fragment database and are based on a library of compounds whose log P values had been previously measured by the partitioning of the molecule between a nonpolar and a polar solvent (most commonly octanol and water) (6).

Molar refractivity, the second independent variable, provides a measure of the inherent steric properties of a molecule and is also calculated by a summation of the individual-substituent contributions retrieved from the Tsar database. The substituent values were derived from a library of compounds whose molar refractivities were experimentally calculated from their corresponding refractive indices, molecular weights, and densities. Both independent-variable values were generated by Tsar, and regression analysis was performed to furnish the correlation coefficient, r².

Molecular conformational analysis.

Due to the lack of structural data available for a sulfa drug inhibitor-DHPS complex and the inherent flexibility of the sulfa drugs, conformational analysis of sulfamethoxypyridazine (7) was performed using MacroModel. The geometry of the sulfanilamides was first optimized with the MM3 force field, using the dielectric constant of water to simulate aqueous solvation and an energy gradient convergence criterion of 0.005 kcal/mol. A Monte Carlo sampling method conformational search was then performed using 1,000 steps. All compounds within 100 kcal/mol of the global minimum were reported. The global minimum and the 10 lowest-energy conformations were then imported into Sybyl and compared by root–mean-square (RMS) fit analysis to each other and to the crystal structure of sulfamethoxypyridazine (2) to determine the reasonability of the structures generated. Least-squares fit analysis of all non-H atoms of the global minima with each of the 10 most similar conformations resulted in RMS values of less than 0.20 Å, indicating that these lowest-energy conformations were all highly similar. The crystal structure data presented two different conformations of sulfamethoxypyridazine per unit cell. One of the conformations (molecule A of reference 2) was similar to within 0.463 Å (RMS value) of the global minimum structure determined by MacroModel when a least-squares fit analysis of all non-H atoms was performed.

The remaining compounds were then constructed, using the global minimum of sulfamethoxypyridazine as a template. Gasteiger-Hückel charges were calculated for each compound. Any remaining portion of the sulfamethoxypyridazine template for each compound was defined as an aggregate, and the newly formed fragments of the molecular geometry were optimized by using the Tripos force field with a 0.001-kcal/mol energy gradient convergence termination point. The aggregate was then deleted, and the entire molecular geometry was optimized by using a 0.05-kcal/mol energy gradient convergence.

As noted by Cramer et al. (4), the alignment rule is crucial to the development of a reliable CoMFA model. To obtain a good alignment, Multifit, a flexible-fit protocol executed in SYBYL, was then performed between each molecule and the structure of sulfamethoxypyridazine at the eight points labeled in Fig. 1. To minimize the possibility of molecular distortions, the Multifit spring constant was reduced to 10 kcal/(mol)(Å)².

FIG. 1.

Points used in alignment by the Multifit protocol. rot 1, rotation 1.

The use of conformational searches in conjunction with flexible data sets which lack active-conformation data and the use of multiple conformations have been previously documented by Nicklaus et al. (10) and Tong et al. (14). The use of multiple conformations of an ambiguous molecule allows the less-active conformer to be eliminated from the data set, thereby improving the predictive correlation coefficient. Thus, for the compounds difluorodinitrophenylsulfone (no. 2), NSC355394-h (no. 13), NSC403439-f, and sulfaquinoxoline (7), for which different conformations can result from a 180° rotation about the aryl-amide bond (rotation 1 in Fig. 1), two conformations (A and B) were placed in the data set. The less-active conformation was then deleted from the data set after the initial analysis had been performed.

CoMFA generation and analysis.

Although the more-active sulfa drugs have a sulfonamide N-H which is ionized at physiological pH, CoMFA was performed on the data set of 47 aligned compounds (43 plus 4 conformation isomers) in their nonionized forms. A lattice spacing of 2.0 Å in the x, y, and z directions that extended 5.0 Å beyond the extremities of all of the compounds was used. An sp³ carbon atom with a +1 charge was used as a probe atom in the calculation of interaction energies at each of the lattice points. The lattice point energies were then used in the QSAR analysis to generate the steric and electrostatic fields.

A PLS analysis was performed on the data set with cross-validation using 5 components and 10 groups. The dependent variable for the PLS analysis was the inverse log of the IC₅₀ calculated to three significant figures. The process of deleting each compound once from the analysis, recalculating the model, and predicting the excluded compound leads to the generation of residual values. A residual is thus the actual activity minus the predicted activity, and those compounds with large residuals are identified as outliers.

The less-active conformations of the four compounds with A and B conformations were deleted to give a final data set of 43 compounds. The full data set less the four unfavorable conformations was defined as D1 (see Table 3).

TABLE 3.

CoMFA PLS summary results

Data set	Statistical data
	r2		SE	F value	No. of components	No. of compounds
	Cross-validated	Conventional
D1	0.488	0.952	0.404	147.9	5	43
D2	0.699	0.964	0.340	159.5	5	36

A PLS analysis was then performed once again, using the leave-one-out cross-validation technique to determine the optimal number of components. A final PLS analysis was performed with the optimal number of components without validation to give the conventional r² value as well as generate the steric and electrostatic fields.

As noted above, the active form of the sulfa drug involves ionization at the amide position. To best model this effect, the N-sulfonamide-substituted compounds Ro55615 (no. 3), Ro72844 (7), and Ro52928 (7) and the sulfone compounds (no. 1, no. 2, and dapsone [7]) were deleted from the data set. Although Ro211182 (no. 4) is predicted to have activity similar to experimental values (see Table 4), it was nonetheless deleted because the quaternary center adjacent to the amide would have had an abnormal effect on the ionization process. A PLS analysis was also performed on this data set of 39 (36 plus 3 conformational isomer) compounds, with the 3 less-active conformations being deleted as before to give a total data set of 36 compounds. This data set was called D2 (see Table 3).

TABLE 4.

Predicted IC₅₀s and residual values for compounds removed from data set D1 to yield data set D2

Compound no.	Compound (reference)	Predicted −log IC50	Residual valuea
	Dapsone (7)	−0.0128	0.833
1	Diformyldapsone	−2.45	−0.341
2	Difluorodinitrophenylsulfone	−2.61	−0.371
	Ro72844 (7)	1.60	0.0419
	Ro52928 (7)	1.03	0.668
3	Ro55615	−1.12	−0.405
4	Ro211182	−2.73	0.0341

^aResidual value = actual property − predicted property. 

RESULTS

Conventional QSAR studies.

Regression analysis of DHPS IC₅₀s versus Tsar-calculated log P and molar refractivity data resulted in an r² value of only 0.142, indicative of a poor correlation between the lipophilicity of the compounds and their activity. This lack of correlation has been noted previously for sulfanilamide data sets. However, a strong correlation of activity with the ionization constant (pK_a) of sulfanilamides is widely accepted, and several QSAR studies have been developed (1). Since pK_a data for our compounds were not available, no further conventional QSAR studies were performed. Calculation of pK_a values remains a possibility for future studies.

CoMFA.

The results for the two data sets of the CoMFA studies are summarized in Table 3. For the first data set (D1) of 43 compounds, a cross-validated r² value of 0.488 and a conventional r² value of 0.952 were obtained. Deletion of the sulfone and N-disubstituted compounds gave data set D2 and resulted in increases in the cross-validated and a conventional r² values to 0.699 and of 0.964, respectively. A comparison of the actual and calculated IC₅₀s for the compounds deleted from the D1 data set can be seen in Table 4.

DISCUSSION

The compounds deleted to give the D2 data set included the N-methoxy-substituted compound Ro55615 (no. 3). As seen in Table 2, this compound was completely inactive in the assay system used, although it closely resembles the highly active compound Ro43476 (7). In contrast, the N-amide acetyl-substituted compounds Ro72844 and Ro52928 (7) were some of the most potent compounds in this study. This discrepancy can be explained by commonly accepted SAR for sulfanilamides (1). Only compounds which can form an ionized species at the amide position have good activity as sulfa drugs. Thus, while compound no. 3 cannot undergo ionization, Ro72844 and Ro52928 are readily hydrolyzed to the ionizable -HN-SO₂-species under even slightly basic conditions. To test the possibility that hydrolysis of Ro72844 and Ro52928 occurs during the enzyme binding assay, the latter compound was exposed to weakly basic solutions (pH, ∼9). A rapid disappearance of Ro52928 was observed, and sulfamethoxazole was noted as the only organic product. However, when placed in the buffered enzyme-free assay system at pH 8.2, hydrolysis of Ro52928 and Ro72844 was not observed. It seems highly likely, nonetheless, that these compounds are converted to their corresponding sulfanilamides under the conditions of the assay, and it may even be possible that an allosteric site on DHPS is effecting this hydrolysis.

Since it seemed likely that the active compounds Ro72844 and Ro52928 would be better represented in the CoMFA as their corresponding monosubstituted sulfanilamides, and since at least in one case the hydrolysis product was a compound already included in the study (sulfamethoxazole), they were deemed redundant and deleted from the data set. In addition, the sulfone compounds, which lack the sulfamide functionality, were deleted due to their inability to ionize as sulfanilamides.

An examination of the residual values for the nonionizable compounds provides further evidence supporting the removal of these compounds to give the D2 data set. As can be seen in Table 4, the predicted activities of the sulfone compounds differ from their observed activities by as much as 0.8 U. Because the original IC₅₀ data was input as −log IC₅₀s, this would result in a predicted activity that was incorrect by as much as 1 order of magnitude. A similar result was also seen with the disubstituted sulfanilamides.

The improvement in the cross-validated correlation coefficient of the QSAR resulting from deletion of these six compounds from the data set is consistent with the ionization theory and indicates that the D2 model designed from N-ionizable compounds describes the pharmacophore of sulfanilamides more accurately than does the D1 model.

CoMFA fields.

To better visualize the CoMFA fields, the CoMFA contours were generated for the D1 and D2 PLS analyses; they are shown in Fig. 2. In the electrostatic contour plots (Fig. 2A and C), the red polyhedra represent regions on the drug molecule at which a negative charge is predicted to result in lower IC₅₀s (i.e., greater drug potency). The red regions themselves might be imagined as electropositive groups within the receptor. The blue polyhedra represent regions in which a positive charge on the drug would have a beneficial effect on IC₅₀s. The blue contours are, of course, complementary to this and may be treated loosely as electron-rich groups within the active site of the receptor.

FIG. 2.

CoMFA electrostatic and steric contour plots from the D1 (A and B) and D2 (C and D) PLS analyses. Sulfamethoxypyridazine is shown within the contours for reference. The electrostatic contour plots (A and C) show contours (color coded in red) corresponding to regions where electron-rich groupings are predicted to improve activity, while electron-poor groups within the blue electrostatic contours are predicted to improve activity. The steric contour plots (B and D) are shown separately in yellow and green. The green steric contours show where added steric bulk is predicted to be beneficial for activity, while the yellow contour indicates regions where steric bulk is negatively correlated with activity.

The steric contour plots (Fig. 2B and D) likewise show regions in which increased steric bulk would likely enhance activity by lowering the IC₅₀ (green polyhedra). The yellow polyhedra indicate areas in which steric bulk is predicted to decrease binding. The structure of sulfamethoxypyridazine is shown for reference in the views. Thus, green polyhedra might represent hydrophobic cavities within the receptor in which there is room for hydrophobic groups on a ligand, and the yellow polyhedra could be interpreted as being areas that are already occupied by the receptor and would thus prohibit effective drug binding.

As can be seen in Fig. 2, the electrostatic CoMFA fields (contours A and C) for D1 and D2 are closely related. A large, blue, positive-charge-favored region is observed on one face of the 4-aminoaryl portion of the molecule. This region corresponds to the R₁ group of the 4-aminoaryl-substituted compounds, such as no. 6 to 14 in Table 2, which contain electron-rich groups at the R₁ position and demonstrate a relative lack of activity. The presence of electronegative R₁ groups in a blue contour region, which prefers to have electropositive groups, is consistent with the observed loss of activity. Conversely, a red, negative-charge-favored region over the opposite face of the sulfonamide aromatic ring of the molecules could be interpreted as a π interaction with an electropositive group in the receptor, such as a π-stacking interaction. The other red contour, occurring about the nonsulfanilamide aromatic ring, corresponds well with the electronegative N’s of the methoxypyridazine structure (contour C).

The steric CoMFA fields of D1 and D2 appear to be slightly different. Although a yellow, sterically unfavorable contour is localized around the 4-aminoaryl group in both D1 and D2, two much larger, green, sterically favorable contours encompass both ends of the molecule in the D1 model and appear to extend down most of the length of the molecule. This could be due to the sulfone compounds, which did not overlay properly with the sulfanilamides and were thus deleted from the D2 data set. In addition, a small, green, sterically favorable contour appears over the N of the sulfanilamide in the D1 model. This would directly coincide with the disubstituted sulfanilamides, which were determined to most likely undergo ionization in the assay system and, as such, were not included in the D2 dataset. As with the electrostatic fields, the yellow, sterically unfavorable region corresponds with the R₁ substitution of the p-amino-substituted sulfanilamides in Table 2 as well as NSC52105-s (7). It is generally accepted that p-amino-substituted sulfanilamides are only active if the R₁ group can be readily hydrolyzed in vivo or in cell culture (1). While the use of a cell-free system allows for greater ease of testing of a large number of sulfa drugs, hydrolysis of the p-amino sulfanilamides in vitro is unlikely, and therefore analogs which would otherwise be active in whole-cell assays would not be detected.

In conclusion, the steric and electrostatic features of a series of 43 sulfa drugs have been correlated with DHPS IC₅₀ data by a 3-D QSAR approach, CoMFA. CoMFA is a technique which has gained widespread acceptance in recent years as a tool for novel-drug development (8, 9). Two sulfa drug pharmacophore models were examined; the first contained analogs bearing substituents on the sulfanilamide N atom, while the second did not incorporate these analogs. The CoMFA model containing the N-substituted sulfanilamides was less accurate due to an underprediction of activity of several of these sulfanilamides. Deletion of these N-substituted sulfanilamides resulted in higher cross-validation values, and the reasonableness of their exclusion was further supported by the likelihood of their being hydrolyzed to active compounds already included in the model. This model is consistent with accepted SAR of sulfanilamides in that only acidic drugs with an ionizable sulfanilamide proton are expected to be active.

Bulky N substituents at the para position of the anilide ring detract substantially from DHPS binding, a finding that is reflected in the CoMFA contour plots in this region. According to reported SAR, these types of compounds would be expected to be active in vivo only if they were capable of hydolysis to the parent anilide (1). The lack of intrinsic activity (receptor binding) of R₁-substituted anilides (e.g., NSC52105-s and compounds 6 to 12) indicates that they do not undergo hydrolysis under the conditions of the assay. For the design of potent inhibitors of P. carinii DHPS by CoMFA, no bulky modifications should be attempted at the para position of the anilide ring.

With the many sulfa drugs already synthesized and described in the literature, this CoMFA study can be used as a preliminary screen for drugs with potential for improved activity toward P. carinii DHPS. Future studies will be directed toward the refinement of this model by the screening of additional compounds for inclusion in the QSAR analysis.