The relationship of physico-chemical properties and structure to the differential antiplasmodial activity of the cinchona alkaloids

The accumulation of an ionizable drug in cells is determined (apart from special transport mechanisms) by two major parameters: the partition coefficient P between the aqueous and the lipid phase, usually expressed as log P (measured in the octanol-water system for the un-ionized compound), and its modification by the ionization constant(s) (pK_a) of the drug. For partially ionized compounds, the partition coefficient between 2 phases at any fixed pH is called the distribution coefficient D, usually expressed as log D; this assumes that only the un-ionized species partitions from the aqueous to the organic phase [20]. The difference between log P and log D is, therefore, determined by the percentage of the drug ionized at different pH values and is given by the standard equation for a base [21]. When the difference between pK_a and pH is more than 1.0 log unit, it has been shown [22] that one can merely subtract this difference from log P to obtain log D. This makes it possible to calculate log D [23] from a knowledge of log P, pK_a and pH. The availability of the ClogP programme, developed by C. Hansch and A. Leo at Pomona College, California, as part of the Pomona College Medicinal Chemistry Project, has now provided a useful method of computing approximate log P values for compounds of known structure. Values may sometimes be unreliable since differences due to conformational isomerism are not detected [25].

In the case of a drug possessing two basic centres, as quinoline antimalarial drugs all do, the correction needed in log P to obtain log D at a given pH involves the additive contribution of both ionized species, and the base equation [21] must be modified: [26, 28].

Equation (1) for a diacidic base

Log D at the physiological pH of 7.4 was obtained experimentally from the plot of log D versus pH. This was compared with the calculated value of Log D, obtained from Log P, using equation (1). Log D at the probable pH of the digestive vacuole or lysosome was calculated from log P by substituting pH 5.2 [27] for 7.4 in equation (1). The application of equation (1) provides a useful rapid method of checking measured values of log D, and of calculating these values from log P when unknown. It should be noted that for diacidic bases, pK_a1 > pK_a2.

To calculate the vacuolar accumulation ratios (VAR) that weakly basic lipophilic drugs might achieve by virtue of the pH difference between lysosome and external medium or plasma [5], the equation of Henderson & Hasselbach was used in the following form [29].

Equation 2

Dissociation constants and partition coefficients were determined by Robertson Microlit Laboratories (Madison, New Jersey, USA) at 25°C using the Sirius GLpK automated computerized potentiometric system [30]. Titration employed water containing 0.15 N KCl in an argon atmosphere. The pK_as were determined in triplicate with a S.E. of ± 0.20.

Partition coefficients between octan-1-ol and water were measured by dual-phase potentiometric titration using various amounts of water-saturated octanol. Titrant addition was carried out with vigorous stirring of the assay solution. Three different ratios of octanol/water were employed for each compound. The log P values were obtained from the difference between the aqueous pK_a of the species and the apparent pK_a determined from the dual phase titration [31]. Measurements were carried out in triplicate with a S.E. of ± 0.40. The potentiometric method was validated by comparison with results obtained by the standard shake-flask technique [32].

The drug lipophilicity values at pH 7.4 and 5.2 (log D) are the octanol/water distribution ratios, and were obtained from the plot of log P versus pH. Calculated log D values for pH 5.2, representing the conditions in the lysosome, and for log D at pH 7.4, were obtained using equation (1).

The method was slightly modified from Parapini et al. [33]. The 8 M sodium acetate/acetic acid buffer (pH 5.0) was prepared as follows:

Make up solid sodium acetate to 8 M with warm water (continue to heat at 37°C in a water bath while mixing: it will not dissolve completely) and then adjust the pH to 5.0 with 8 M acetic acid at 37°C, which leads to the remaining solid dissolving).

One hundred μL was dispensed into wells of a microtitre plate, and 50 μL of alkaloid hydrochloride or sulphate solution in water, at 80 mM, or water alone, was added. Verapamil (VE) and promethazine (PMZ) (Sigma Chemicals, Poole, UK) were used as hydrochlorides. CQ (Sigma, UK) was used as the diphosphate. Addition of 50 μL of an 8 mM α-chlorohemin (Sigma, UK) solution in DMSO to each well started the experiment.

Experiments were allowed to run at 35°C for 24 hr., done in triplicate and repeated on three occasions. The final DMSO concentration in this system was 25% w/v, while final haematin concentration (0.4 μmol/well) was 2 mM, and the final drug concentration (4.0 μmol/well) was 20 mM. Drug/haematin molar ratio was 10:1.

Then insoluble β-haematin and haematin aggregate were pelleted by centrifugation at 3000 g for 20 min at room temperature (22°C) and the supernatant discarded. Two hundred μL of DMSO were added to each pellet to dissolve un-reacted haematin and centrifugation was repeated. After discarding the supernatant, 200 μL of 0.2 M NaOH were added to convert the β-haematin pellet to soluble alkaline haematin. For each well the alkaline haematin was quantified by serial twofold dilution in water and OD measurement at 414 nm. The effect of each agent on β-haematin production was recorded compared with the control (water + DMSO + acetate buffer alone).

Each mono-protonated cinchona alkaloid structure was built in HyperChem Release 7 for Windows (Hypercube Inc. Gainesville, Florida) using a molecular mechanics procedure under MM+ [34] with bond-length restraints for N1-O12 and N1 to C5' distance [15] (bond dipoles, no cut offs employed). The geometry was optimised to an rms (root mean square) gradient of 0.001 in vacuo (Polak-Ribière method). A periodic box, 15 × 15 × 15Å around the drug was then set up, containing 112 water molecules. The system was optimised in MM+ using switched cut-offs (outer 7.5 and inner 3.5 Å), to an rms gradient of 0.5. Then a molecular dynamics programme was run for 1 ps, with 0.001 ps steps, relaxation time 0.1 ps, to a simulation temperature of 300 K. This was followed by MM+ geometry optimisation to an rms gradient of 0.2. The molecular dynamics run was repeated and a further MM+ protocol was carried out to a gradient of rms 0.004 on the selected drug.

Angles and bond-lengths were measured on the models, and the dipole moments were determined after removal of the N-1 proton, using the semi-empirical PM3 programme [35, 36] in singly-excited configuration interaction. (RHF [Restricted Hartree-Fock], charge 0, spin multiplicity 1, lowest state, orbital criterion, five occupied and five unoccupied orbitals.)

Models of the Q-haematin and EQ-haematin complexes were prepared in MM+ using the structures of Q and EQ optimised in water (see above), with a co-ordination link between haematin iron and drug O-12. The crystal structure of α-chlorohaemin [37] was used for the initial haematin monomer model.

The semi-empirical quantum mechanics program ZINDO/S [38] in singly excited configuration interaction was used to calculate the expected electronic spectrum of the optimised 1: 1 quinine/haematin complex in benzene. One molecule of singly protonated quinine, co-ordinated to one molecule of haematin, with carboxylic acid groups undissociated, was given an arbitrary charge of +1 and spin multiplicity of four. Fifteen occupied and 15 unoccupied orbitals were included in the calculation.

(EQ and EQD were kindly supplied by Dr. Reinecke of Buchler GmbH, D-38110 Braunschweig, Germany.)

Q and QD, C and CD are reported to have very similar pK_a1 values (8.4 ± 0.4) for the quinuclidine aliphatic N (table 1: footnotes). A value of 8.5 ± 0.15 (table 1) was found, with a slight increase in HQ and HQD due to the conversion of the electron-attracting vinyl group to an ethyl group. All the cinchona alkaloids in table 1 had very similar pK_a2 values (4.25 ± 0.15) for the more weakly basic quinoline aromatic N. The only previous determinations [41] for the threo compounds EQ and EQD were carried out in 80% methyl-cellosolve / 20% water and showed both 9-epimers to have higher pK_a values than Q and QD in the same solvent. Results obtained in water (Table 1) confirmed this finding and revealed a consistent ΔpK_a1 of 1.1 units between the two solvent systems, and of 1pK_a unit between the erythro and the threo isomers. Experimental values for log P are shown in table 1 as are the figures calculated using the ClogP programme. The two sets are in generally good agreement, and show the expected slight increase (+0.4) with the dihydro compounds (HQ and HQD) and decrease (-0.3) with the loss of the 6-methoxy group (C, CD). Experimental measurement of log D vs pH over the range 3–11 permitted the determination of log D at pH 7.4. In addition knowledge of log P and pK_a allowed the calculation of log D at pH 7.4 using equation (1). Measured and calculated figures (table 1) are in good agreement, while reported values (table 1: footnotes) show a wider variation. Results for log D at pH 7.4 for the erythro compounds fall in the range 1.50–1.97 (mean 1.74 ± 0.24) while the threo isomers (range 0.55–0.76) show a mean of 0.66 ± 0.11. In view of the acidic milieu (pH 5.2) existing in the vacuole within the lysosomal membrane, log D at pH 5.2 was also measured and calculated and the figures are shown in table 1. The comparison of calculated results for these agents at pH 5.2 with those obtained for chloroquine (CQ) is of interest, since because of CQ's two high pK_a values [43, 44], the drop in pH had a much larger influence than on Q, QD, EQ or EQD. The drop in log D between pH 7.4 and pH 5.2 for Q, QD, EQ and EQD was ~2 logs while for CQ it was > 4 logs.

The parasite lysosome, which has been proposed as the main site of drug uptake, has an acidic pH. Assuming drug distribution reaches an equilibrium state, and only uncharged drug can pass through membranes, it is argued that the ratio between concentration of drug outside the erythrocyte and the concentration in the lysosome depends on the difference between [H⁺] of the medium in vitro or plasma in vivo (pH 7.4), and the lysosomal compartment, according to the Henderson-Hasselbach equation [5]. Both the water/lipid partition (log P) and the acid dissociation constant(s) (pK_a) of a drug influence the passage of drug into the lysosome and its trapping by protonation. Trapping of a drug in the acidic vacuole is enhanced if the base, like chloroquine (CQ), can accept more than one proton at the lysosomal pH [28]. Calculated CQ concentrations within the infected erythrocyte approximate to measured uptake values, but the more lipophilic agents like amodiaquine (AQ) achieve significantly greater levels than calculated, indicating that other factors are involved in the concentrative uptake of the drug [29]. Calculated lysosomal uptake ratios (vacuolar accumulation ratio: VAR, the ratio of calculated drug concentration in the lysosomal water to concentration in the external medium water at equilibrium) for cinchona alkaloids, for resistance-reversers VE and PMZ, and for CQ are given in table 1. The most interesting finding here was the small variation in calculated VAR seen among the eight alkaloids examined (mean: 167.8 ± 7.4 [SD]). The minor variation reflects slight differences in the protonation of the aromatic N at pH 5.2, while the aliphatic N is 100% protonated in all the alkaloids examined.

A 10 fold molar excess of Q, QD, HQ, HQD, C, CD, or CQ inhibited the BHIA haematin dimerization reaction. Significant inhibition was not seen with EQ, EQD, VE or PMZ. (Table 1, figure 2). These results fit in with the reported high activity of CQ [73] and the erythro cinchona alkaloids on P. falciparum and the low activity of EQ, EQD [15, 46], PMZ [52] and VE [53].

As pointed out by Karle and co-workers, [15, 46], the N1-H1 bond in 8, 9 threo-alkaloids consistently points in the same direction as the C9-O12 bond, while in the erythro-alkaloids they are at right angles (see the last column of table 2). The calculated dipole moment and proton affinity of each active alkaloid are lower than that of its 9-epimer, as first demonstrated by Karle & Bhattacharjee [46]. This is confirmed by our finding of higher pK_a1 values for the 9-epimers (table 1). Dipole moments and proton affinities calculated here consistently differ from values in ref. [46], probably because we used a semi-empirical configuration interaction calculation rather than an ab initio approach.

Space-filling models of the cinchona alkaloids orientated to face a putative receptor are shown in figure 3.

A Q-haematin co-ordinated model was calculated using MM+, in comparison with EQ-haematin. The results of the ZINDO/S calculation of the electronic absorption spectrum of Q-haematin are compared with experimental observations in table 3 below. Differences in the orientation of the N1 proton with respect to the quinoline and porphyrin rings are shown in figure 4 below.

Some features of the modelled drug-haematin complexes are of interest. The Fe-O12 bond length is 1.846 Å in Q-haematin and 1.842 Å in EQ-haematin which can be compared with 1.847 Å for the O-Fe distance in the high-spin co-ordination complex with a phenoxide [49]. The minimum distance between the carbon atoms of the quinoline ring in the drug, and those in the porphyrin ring was 3.1 Å in both models. This would allow Van der Waals interactions between the conjugated rings for both Q-haematin and EQ-haematin.

The H1-Fe distance in Q-haematin is 4.6 Å, while in EQ-haematin it is 3.9 Å, Proximity of positive charges of N1 proton and iron in EQ-haematin, and the higher value of the proton charge [46] might make co-ordination complex formation more difficult than for Q-haematin. The more solvated, more positive, epimer N-1 proton may interfere with the initial hydrophobic interactions of the quinoline and porphyrin rings, affecting co-ordination of O-12 to haematin iron. The importance of solvent water interaction with protonated N1 in inhibiting hydrophobic binding to haematin is suggested by the observation that in 40% DMSO both Q and EQ interact similarly with haematin [50]. The direction of the positive electric field of protonated N1 determines the higher pK_a and solvent interaction and is likely to be, as suggested by Karle and colleagues [15, 48], the crucial difference between the inactive and active cinchona alkaloids, since this difference in direction, shown in figure 4, alters both the stereochemistry of interactions between Q or EQ and haematin and increases the solvation and charge of N1-H1 in EQ. These considerations are supported by the lesser effect of the threo/erythro distinction on the activity of mefloquine congeners and phenanthrene methanols (WR 122,455) where the piperidine ring is more easily able to rotate close to the quinoline ring.

The peak wavelengths and strengths of the calculated electronic absorption spectrum compare well with the observed spectrum of the aqueous complex, extracted into benzene, reported earlier [12] (table 3). Despite the fact that, in view of the discrepancy in wavelength and strength of the α-peak, the structure is not yet completely correct, the α- and β-peaks support the formation of a hydrophobic quinine-haematin co-ordination complex while the strong γ-peak at 408 nm indicates haematin iron in a high-spin state, as expected from the putative structure. This is similar to α-chlorohemin [37], where Fe^III is 0.45 Å above the plane of the porphyrin ring. It appears unlikely that the quinoline ring interacts with the other side of the haematin ring structure, as has been proposed for the quinoline ring of CQ [63] since a high spin co-ordination complex could not then be formed.

In summary, positive correlations between antiplasmodial activity and drug lipophilicity have been reported [43, 45, 47] for a number of 4-aminoquinoline drugs related to CQ and amodiaquine (although not for the cinchona alkaloids) as well as for some aminothiol [74] and hydroxamate [75] chelators. In addition, a proportionality between activity and inhibition of formation of β-haematin has been recognized for 4-aminoquinolines [76] and also for Q and QD [19], The present report, however, suggests that the differences in the antiplasmodial potency of the erythro and threo cinchona alkaloids on CQ-sensitive isolates of P. falciparum may be accounted for as the cumulative consequences of their log D (pH7.4) and BHIA values. The availability of these two critical parameters within this highly homogeneous group of drugs (table 1) makes it possible to apply a simple test of their sequential and combined effects: The ratio of the mean log D (pH 7.4) for the erythro compounds (1.74 ± 0.24) to that of the threo compounds (0.66 ± 0.11) is 12:1. The BHIA values for the 2 series are 90 ± 7% and 7 ± 4%, with a similar ratio of 12.5:1. Since these effects are cumulative, their combined result would raise the mean IC50 value for the threo series to 150 (ie 12 × 12.5) times that of the erythro group. With an experimental mean IC50 of the erythro alkaloids of 26.7 nM, the calculated reduction of activity of the threo alkaloids leads to an expected mean IC50 of (26.7 × 150) ie 4.0 μM, while the observed experimental value is 3.1 ± 0.4 μM. Alternatively the enhancement in activity of the erythro series would lead to an IC50 of 3085/150 ie 20.6 nM; the experimental mean for the 6 erythro compounds is 26.7 nM.

The agreement between the calculated and experimental values of IC50 is noteworthy.

These observations may provide a useful platform for the further investigation of antiplasmodial agents, particularly in the climate of increased drug-resistance.

Our arguments so far have concentrated on the probable role of hydrophobicity (logD) and inhibition of haematin dimerization as the determinants of antiplasmodial activity among eight cinchona alkaloids. Clearly these factors are also important in activity against CQ-resistant isolates. However, a further parameter, interaction with the CQ-resistance mechanism, must also be considered. Entry to the cell, concentration in the acidic lysosomal vacuole, and interaction with haematin have all been postulated as the site of the resistance mechanism. Arguments [61, 77, 78] and evidence [51, 79, 85] for and against drug efflux have not yet reached a consensus. Nevertheless, the most important mutations associated with CQ-resistance are in the genes coding for two intrinsic proteins in the membrane of the lysosome [80, 56] where quinoline drugs are believed to be concentrated.

One of these, PGH-1, is a protein of the multidrug-resistance type [81], where over-expression is associated with resistance to highly hydrophobic drugs such as mefloquine, halofantrine and artemisinin [82‐84]. The other, is PfCRT, where a Lys76Thr mutation is 100% associated with CQ-resistance in vitro[56]. In clinical studies, inability to detect the mutant allele predicts treatment success [57]. Replacement of the wild type pfcrt gene in sensitive clones by the resistance-associated allele from strains of Asian or South American origin confers CQ-resistance, accompanied by reduced uptake of the drug, and the acquisition of the characteristic chemosensitizing effect of VE [58]. The detailed structure of the PfCRT protein is still under investigation. In its topology the sequence resembles the aqueous chloride channel of Salmonella typhimurium and a range of other organisms [59], though 3 of the 4 specific ion-binding motifs are modified. Modelling (work in progress) reveals that the side chain of residue 76 may project into an access pore leading from the lysosomal content. In the wild type (Lys76) this is positively charged, which would attract anions and repel cations. Mutated to neutral Thr76, cations would find access easier. PfCRT has recently been expressed efficiently in yeast Pichia pastoris by Roepe's research group [60], and functionally appears to be a membrane channel which, when mutated, may mediate chloride-dependent, VE-inhibitable drug transport. Thus PfCRT could indeed be an anion channel, and by varying the entrance or exit of charge-balancing chloride anion may modulate lysosomal pH [87, 88]. This might affect the trapping of basic drugs [5], or interfere with drug-haematin binding [86]. Doubt has been cast [89] on the validity of pH determinations underlying the argument in refs 86–88. These pH effects are perhaps rendered unlikely by the stereospecific interactions of the physico-chemically similar diastereomers Q and QD with mutated PfCRT [58, 90]. It is more likely that CQ itself passes through modified PfCRT, since the high drug and proton concentration in the lysosome present a formidable pressure for efflux, given a suitable channel.

For 4-aminoquinolines, the positive correlation between hydrophobicity and relative activity in CQ-resistant isolates, where the more hydrophobic AQ and its metabolite DAQ show significant activity [54] is an important finding. This observation is paralleled by the CQ-sensitising effect, on CQ-resistant isolates, of lipophilic weak bases like VE and PM [52, 53]. We are investigating a link between polymorphism in PfCRT, resistance reversal by VE, and the clinical efficacy of AQ (in press).

VE and PM show no significant interaction with the haematin polymerisation process (table 1, figure 2) and they have little intrinsic antiplasmodial action. Combining our recent data on 4-aminoquinolines [73] and the erythro cinchona alkaloids studied here there is a significant correlation between hydrophobicity and log relative activity in CQ-resistant parasites, extending the observations of Bray et al.[54] (figure 5).

It has been proposed [61, 73] that 4-aminoquinolines which are more polar at lysosomal pH (such as CQ) are more likely to pass through the mutated PfCRT channel and escape from the lysosome into the parasite cytoplasm, while those which are more hydrophobic (AQ and its active metabolite DAQ) are likely to bind to the hydrophobic channel lining and stay inside. Their positive charges repel the further access of 4-aminoquinoline to the channel. The high hydrophobicity in acid of VE and PMZ (see table 1) would allow them to bind in a mutated PfCRT channel and their positive charge would prevent efflux of CQ and related drugs. Molecular computations on CQ-resistance reversing agents have indeed determined that their common features are hydrophobicity and electronegativity (i.e. proton attraction) [55].