Anal. Chem. 1998, 70, 2092-2099
Estimation of the Partitioning Characteristics of Drugs: A Comparison of a Large and Diverse Drug Series Utilizing Chromatographic and Electrophoretic Methodology Melissa Hanna,† Vern de Biasi,‡ Brian Bond,‡ Colin Salter,‡ Andrew J. Hutt,† and Patrick Camilleri*,‡
Department of Pharmacy, King’s College London, Manresa Road, London, SW3 6LX, U.K., and SmithKline Beecham Pharmaceuticals, New Frontiers Science Park, Third Avenue, Harlow, Essex CM19 5AW, U.K.
1-Octanol-water log P values for a large number of standards and bioactive molecules have been correlated to the logarithm of the corresponding capacity factors determined by reversed-phase high-performance liquid chromatography, using a novel dynamically coated phase, containing phosphatidylcholine. Similarly a correlation was also obtained for log P and capacity factors determined by micellar electrokinetic capillary chromatography (MECC), involving the use of phosphatidylcholine-bile acid mixed micelles in the separation buffer. Statistical analysis of data obtained via both methods has shown that either method will give reliable log P predictions, although MECC is generally more useful for neutral and basic compounds. It is recommended that, as both methods can easily be set up in an analytical laboratory, their combined use provides rapid methodology for the confident estimation of hydrophobicity, as measured by log P for the widest diversity of chemical structures. Hydrophobicity is very often given considerable attention in the design of bioactive molecules. This physicochemical parameter is normally measured as the ratio of the concentration of a drug between the two phases, 1-octanol and water, and is defined by the logarithm of the thermodynamic partition coefficient, log P. Hansch and co-workers have collected log P values for a large number of compounds, and extensive compilations of these may be found in the literature.1,2 The work of these authors has been very beneficial in rationalizing drug design. Accurate log P data can be generated by the “shake-flask” method. However, this method is relatively slow, and unless radiolabeled compounds are used, the range of log P values that can be directly measured is limited from about -3.0 to +3.0. Thus, the introduction of analytical methods for the indirect determination of the hydrophobicity of solutes has aroused much interest over a number of years.3-6 More recently, the need for methodology for the rapid estimation of log P has increased dramatically due to a consider†
King’s College London. SmithKline Beecham Pharmaceuticals. (1) Hansch, C.; Leo, A. In Substituent Constants For Correlation Analysis in Chemistry and Biology; Wiley: New York, 1979. (2) Leo, A.; Hansch, C.; Elkins, D. Chem. Rev. 1971, 71, 525. (3) Braumann, T. J. Chromatogr. 1986, 373, 191. ‡
2092 Analytical Chemistry, Vol. 70, No. 10, May 15, 1998
able increase in the rate of synthesis of molecules as potential drugs; the “combinatorial approach” for the simultaneous synthesis of a large set of compounds has added pressure for the development of analytical methods for the estimation of log P, preferably avoiding shake-flask methodology. Reversed-phase high-performance liquid chromatography (RPHPLC) has been the most popular technique for the indirect estimation of log P.3-6 This analytical method is most useful when the relative hydrophobicity of structurally related molecules is being assessed. In our experience, for chemically unrelated solutes, determination of log P by HPLC can often lead to unreliable estimates of this parameter, and this is in part due to interaction of molecules with the incompletely derivatized silanol groups of the stationary phase. Drug partitioning through biological membranes has also been modeled by RP-HPLC using an immobilized artificial membrane (IAM) stationary phase, where phosphatidylcholine is covalently immobilized on a silica surface via a “linker” group.7,8 This naturally occurring lipid and other related detergents, such as phosphatidylethanolamine and phospatidylserine, are present in biological membranes, and these molecules are thought to be good models for lipid-mediated transport of drugs. Several IAM chromatographic phases have been synthesized and are now commercially available. In this study, we have adopted a novel and inexpensive alternative to the bonded-phase approach by preparing a modified stationary phase using dynamic coating methodology. Such a technique has the potential advantage of versatility, in terms of both the nature of lipid used and the possibility of facile preparation of mixed-component phases. Using this method, log P estimates were obtained for a large set of structurally diverse compounds. Values of log P obtained by this chromatographic method were compared to those obtained by another approach involving micellar electrokinetic capillary chromatography (4) Kaliszan, R. Quantitative Structure Chromatographic Retention Relationships; Wiley: New York, 1987. (5) Lambert, W. J. J. Chromatogr. 1993, 656, 469. (6) Valko, K.; Bevan, C.; Reynolds, D. Anal. Chem. 1997, 69, 2022. (7) Yang, C. Y.; Cai, S. J.; Liu, H.; Pidgeon, C. Adv. Drug. Del. Rev. 1996, 23, 229. (8) Pidgeon, C.; Marcus, C.; Alvarez, F. In Applications of Enzyme Biotechnology; Kelly, J. W., Baldwin, T. O., Eds.; Plenum Press: New York, 1991; p 201220. S0003-2700(97)01122-0 CCC: $15.00
© 1998 American Chemical Society Published on Web 04/11/1998
(MECC),9-11 one of the modes of separation of capillary electrophoresis. In MECC, a surfactant is introduced in the separation buffer at a concentration that is above the critical micelle concentration (cmc). Unlike free-zone capillary electrophoresis, MECC is primarily used to resolve neutral analytes, and migration time is largely dependent on the corresponding hydrophobicity characteristics. The observed migration is related to the free energy of partitioning of analytes between the aqueous buffers and the interior lipid phase of the micelles: highly water soluble molecules are normally excluded from the micelle and migrate rapidly through the capillary, whereas more lipophilic solutes exhibit longer retention times. As in the case of chromatography, the thermodynamic parameter k′ (capacity factor) is related to the amount of solute distributed between the micelle (cm) and the aqueous phase (caq) and is given by eq 1. The k′ for an analyte is
k′ ) cm/caq
(1)
related to its migration time (tr) by eq 2, where to and tm are the
k′ )
tr - to to(1 - tr/tm)
(2)
migration times of an analyte that does not interact with the micelle and one that migrates with the micelle, respectively. In practice, to is measured from the retention of water and tm from the retention time of an analyte that is completely incorporated in the micelle. For tm approaching a value of infinity, eq 2 becomes identical to the equation used to determine capacity factors in HPLC. Partitioning only occurs in the migration time window between to and tm. The “pseudostationary” phase used in the present MECC study consisted of mixed bile acid-phosphatidylcholine (PC) micelles. The inclusion of PC in the mobile phase permitted the direct comparison with log P estimates obtained by the HPLC method, where phosphatidylcholine was physically adsorbed on to a C8 stationary-phase support. It is frequently the case in studies attempting to relate chromatographic parameters and log P that data are reported for large sets of homologous and/or structurally related agents or for relatively small numbers of structurally unrelated compounds. In the present investigation, we have statistically evaluated both HPLC and MECC methodologies for their ability to estimate hydrophobicity, as measured by octanol-water partition coefficients, of a large number, (n ) 106) of structurally unrelated compounds including acids, bases, and neutrals, with a range of log P octanol/water values between -0.52 and 7.63. EXPERIMENTAL SECTION Materials. The following standard compounds and drugs were purchased from Sigma (Poole, Dorset): deoxycholic acid (sodium salt), taurodeoxycholic acid (sodium salt), caffeine, (9) Adlard, M.; Okafo, G.; Meenan, E.; Camilleri, P. J. Chem. Soc., Chem. Commun. 1995, 2241. (10) Ishihama, Y.; Oda, Y.; Uchikawa, K.; Asakawa, N. Anal. Chem. 1995, 67, 1588. (11) Herbert, B. J.; Dorsey, J. G. Anal. Chem. 1995, 67, 744.
fenoterol, ketoprofen, acetanilide, metoprolol, 2-naphthol, haloperidol, benzophenone, biphenyl, flurbiprofen, phenobarbitone, chlorpheniramine, carbamazepine, acetylsalicylic acid, valproic acid, ranitidine, verapamil, thioridazine, carprofen, indomethacin, tolfenamic acid, promazine, chlorpromazine, trifluoperazine, quinidine, sulfamethoxazole, dipyridamole, ibuprofen, melphalan, nadolol, chlorambucil, theophylline, captopril, azathioprine, trimethoprim, amoxicillin, ephedrine, and megesterol acetate. The following were from Aldrich (Gillingham, Dorset): 1-nitronaphthalene, paracetamol, 1-naphthoic acid, salicylic acid, and sodium phosphate (dibasic, heptahydrate). The following were from Astra (Watford): lignocaine and boric acid. Phenol was from Fisons, (Loughborough, Leicestershire) naphthalene and sodium nitrite were from and B. D. H. Lab Chemicals Division (Poole, Dorset). Testosterone, propranolol, flufenamic acid, tiaprofenic acid, antipyrine, codeine phosphate, diphenydramine, pyrilamine, hydroxyzine, and desipramine were obtained in-house (King’s College London). Soybean lecithin (minimum 95% pure) (Epikuron, Lucas Meyer & Co.) was kindly supplied by Dr. M. J. Lawrence (Department of Pharmacy, King’s College London). Cimetidine, clonidine, imipramine, nabumetone, ropinirole, 4-chloroaniline, 4-nitroaniline, 3-methyl-4-nitroanisole, 6-thioguanine, and halofantrine were available in-house (SmithKline Beecham Pharmaceuticals, Harlow, Essex). HPLC grade acetonitrile and methanol were purchased from Aldrich (Gillingham, Dorset). The structures and code numbers of the compounds used in this investigation are presented in Figure.1. The remaining compounds, indicated by a code number in Table 1, were recently synthesized materials from SmithKline Beecham Pharmaceuticals drug discovery programs. These compounds included heterocyclic derivatives, tertiary amines, ethers, thioethers, amides, guanidines, aromatic amines, phenols, ketones, sulfonamides, and carboxylic acids. The entire compound set included acidic, basic, and neutral compounds with a range of log P values between -0.52 and +7.63. Instrumentation. Reversed-phase HPLC was carried out with an HP1050 system, consisting of a solvent delivery system, degasser, and UV detector (operated at 254 nm), all connected to a Gilson autosampler fitted with a 20-µL loop. Data collection and integration were performed using a Waters 860 data aquisition system. MECC was carried out using a Beckman P/ACE 5000 system fitted with a UV detector set at 214 nm. Methods. (a) RP-HPLC. A Hypersil C8 MOS 100A (10 cm × 4.6 mm i.d. 5 µm) column was used. Dynamic coating of the stationary phase was successfully achieved using a 1 mM solution of lecithin (phosphatidylcholine) in methanol/water (80:20 v/v) and recycling it through the column via an HPLC pump for ∼24 h. UV spectrophotometry was utilized to quantify the amount phosphatidylcholine coated onto the stationary phase during this period. The retention characteristics of a set of compounds were determined before and after the coating experiment and the data compared. For the chromatographic analysis, the UV detection wavelength was set at 254 nm, injection volume was 20 µL, and flow rate varied between 0.5 and 2.0 mL min-1 at ambient temperature depending upon the mobile-phase composition. Mobile phases consisted of 20:80, 40:60, 50:50, 60:40 (v/v) acetonitrile/sodium phosphate buffer (35 mM, pH 7.4) and 100% sodium phosphate Analytical Chemistry, Vol. 70, No. 10, May 15, 1998
2093
2094 Analytical Chemistry, Vol. 70, No. 10, May 15, 1998
Figure 1. Representative structures from compound library analyzed.
buffer (35 mM, pH 7.4). Analytes were prepared in approximate concentrations between 0.1 and 0.5 mg mL-1 in filtered mobilephase components and sonicated until completely dissolved. Sodium nitrite was used for a completely unretained marker in all analyses. (b) Capillary Electrophoresis. A 57 cm × 50 µm fused-silica capillary was used (effective length 50 cm), with UV detection at 214 nm. Each individual analysis consisted of three consecutive 1-s high-pressure injections (20 psi) into each of (1) a solution of the analyte under investigation in the run buffer, (2) an aqueous solution of methanol (5% v/v), and (3) a solution of halofantrine in run buffer (fully retained marker analyte). Separation buffers consisted of 40 mM deoxycholate or taurodeoxycholate in 50 mM borate buffer at pH 8.0. For studies incorporating phosphatidylcholine, the run buffer was prepared by dissolving 25 mM phosphatidylcholine in the buffer. This was achieved after stirring the solution for approximately 3-6 h or until all cloudiness had disappeared. Analyte solutions for capillary electrophoresis were prepared at concentrations of approximately 0.5-1.0 mg mL-1 in the solution of 40 mM taurodeoxycholate/ 50 mM borate pH 8.0 run buffer, sonicated until dissolved, and then filtered through a Whatman 0.2-µm nylon membrane filter before introduction onto the capillary. (c) Log P. Measured 1-octanol-water partition coefficient values used in this study were obtained from the literature12 (all values taken being the measured value (M)) except in a few cases where not available; these compounds therefore have the literature calculated values (C)12 and are depicted in Table 1 by an asterisk) and in-house data at SmithKline Beecham. These were determined by using the traditional shake-flask method, where a solute is distributed between 1-octanol and aqueous buffer which have been mutually saturated. (d) Statistical Analysis. Ordinary least-squares regression was performed in relating single independent variables to the (12) Craig, P. N. In Cumulative Subject Index and Drug Compendium; Drayton, C. I., Ed.; Hansch, C., Sammes, P. G., Taylor, J. B., Eds., Comprehensive Medicinal Chemistry; Pergamon Press: Oxford, U.K., 1990; pp 237-991.
dependent variable of interest, log P. The maximum number of compounds having complete data for the model fitted was used in each regression. Separate regression analyses were performed for acid and nonacid compounds. In building a multiple regression model, ordinary least squares was again used but both acid and nonacid compounds were included. However, a term to take account of the differences between acids and nonacids was included in the model (i.e., parallel regressions for the two groups of compounds). Other variables were assessed by forward and backward elimination in a stepwise procedure. In all of these regressions, residual plots were examined to identify outliers and plots of Cook’s distance to access influential compounds along with the usual checks for equality of variances and normality of the data.13-15 Statistical analyses and graphical presentations were performed using STATISTICA for Windows V5.1 produced by Statsoft Inc. RESULTS AND DISCUSSION Reversed-Phase HPLC. The amount of lecithin (PC) coated on the column was estimated by UV analysis of the coating solution prior to and postcoating. This solution was estimated to have ∼35 mg in total “missing” postcoating by observation of a decrease in the absorption at the λmax, indicating successful coating on the column surface. Comparison of the chromatographic retention characteristics of a series of standard compounds of diverse structure including acids, bases, and neutrals (for representative structures, see Figure 1), both prior to and postcoating indicated an increase in analyte retention for all compounds with no change in elution order. An assessment was also made of the stability of the column under typical experimental conditions with varying amounts of acetonitrile in the mobile-phase buffer. It was estimated by an (13) Cook, R. D. Technometrics 1977, 19, 15. (14) Cook. R. D. J. Am. Stat. Assoc. 1979, 74, 169. (15) Draper. N. R.; Smith. H. Applied Regression Analysis, 2nd ed.; Wiley: New York, 1981; pp 69-71.
Analytical Chemistry, Vol. 70, No. 10, May 15, 1998
2095
Table 1. Data Corresponding to log k′ HPLC (40% Organic in Mobile Phase) and MECC (TDCA/PC)a log k′ compd
log P
HPLC 40%
1 2 3a 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26a 27 28 29 30 31 32 33 34a 35 36 37 38a 39a 40 41 42 43 44 45a 46 47 48 49 50 51 52 53
-0.07 0.40 0.83 1.16 1.83 3.19 1.39 1.49 1.88 2.26 2.80 3.35 3.36 3.40 4.10 2.32 0.51 3.32 3.56 3.13 4.16 5.25 1.47 2.51 3.39 3.93 2.45 3.12 1.19 2.75 0.38 2.26 0.27 0.27 3.27 4.27 3.27 4.16 5.70 4.90 3.79 4.55 5.35 5.03 5.90 0.80 2.58 2.33 2.85 1.59 3.30 4.05 4.60
-0.76 -0.66 -0.53 -0.19 0.34 0.91 0.13 0.06 -0.13 0.73 0.48 1.00 0.95 1.24 1.25 0.76 -0.62 0.52 0.38 -0.55 -0.51 0.13 -0.12 -1.04 0.32 -0.35 0.24 -0.76 -2.08 -1.27 0.23 -1.31 -0.34 0.06 0.57 -0.24 0.34 0.92 0.52 0.58 1.04 0.90 1.01
a
1.35 -0.55 -0.23 0.17 0.27 0.33 0.49 0.79 0.74
log k′
ΜΕCC
compd
log P
HPLC 40%
-0.67 -0.30 0.71 -0.27 0.42 1.51 0.19 0.18 0.48 0.34 1.19 1.64 1.10 1.19 2.12 0.80 -0.50 1.06 1.72 0.44 0.51 1.30 0.16 0.23 1.56 1.37 0.60 0.18 0.30 0.30 -0.77 0.60 -0.15 0.08 1.35 0.82 1.20 1.89 1.43 2.14 1.75 2.30 2.16 2.18 2.31
54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100a 101a 102 103a 104 105 106a
4.25 5.41 3.71 0.93 4.52 3.90 2.64 4.42 2.73 2.78 4.57 1.19 3.08 3.28 0.16 2.86 2.83 2.78 3.59 3.10 3.69 3.97 2.34 2.84 7.63 5.23 3.15 3.40 2.73 6.08 5.84 5.84 5.84 5.44 5.75 4.66 5.74 5.65 -0.07 1.70 -0.02 0.34 3.44 0.89 0.10 0.91 2.13 0.33 3.50 -0.52 0.71 1.18 3.90
0.34 0.82 0.46 -0.54 0.79 0.77 0.64 0.80 0.02 -1.13
0.62 0.78 1.13 0.31 1.19 1.72 1.49
-0.44 0.88 0.51 0.37 0.37 0.73 0.32 0.22 0.83 0.60 0.95 0.10 -0.02 1.43 1.06 0.41 1.22 0.27 1.24 1.47 1.44 1.25 1.18 0.86 1.43 -1.13 -0.32 -0.94 0.59 -0.97 -1.24 -0.39 0.40 -0.07 -0.35 -0.29 -0.13 1.11
ΜΕCC 1.48 1.88 1.41 0.45 1.39 2.14 1.36 1.63 0.92 0.32 2.23 0.70 1.44 1.19 1.09 0.30 1.45 1.25 1.08 0.85 0.37 1.80 1.32 1.46 0.94 2.24 2.04 2.12 2.16 2.16 2.25 1.93 1.26 2.23 0.48 -0.37 0.19 0.20 -0.08 0.11 -0.16 0.30 -0.75 0.33 0.05
Literature log P value obtained via calculation.12
examination of collected eluent from the column over a timed period of flushing that leaching of the PC from the column was not significant even at 60% acetonitrile (that is, less than 1% over 24 h with no apparent change in analyte retention) and much less at 20% organic (less than 0.15% over 24 h). Coating of the stationary phase with PC also proved to be highly reproducible. In fact, the coating procedure in our laboratory has now been carried out successfully on five separate occasions. For all these experiments, the same set of five compounds was injected and corresponding capacity factors were 2096 Analytical Chemistry, Vol. 70, No. 10, May 15, 1998
calculated on each occasion. Regression analysis of the relationship between log k′ versus log P were found to be essentially identical in terms of slope, intercept, and correlation parameters on each occasion. Log k′ values were obtained using the same dynamically coated stationary phase, but varying the percentage of acetonitrile in the mobile phase. Experiments were carried out using 20, 40, and 60% organic content, and equations were derived by relating log k′ values of a variety of analytes from each mobile phase to the corresponding octanol-water log P values. Log k′ values extrapolated to 0% organic content in the mobile phase were also determined (data for 40% organic content given in Table 1). Of all the mobile-phase compositions, that is 0, 20, 40, and 60% organic content, 40% gave the best correlation with log P for acids and nonacids considered separately (Table 2) (data corresponding to other organic compositions not shown). A number of the more hydrophobic analytes were not eluted using 20% acetonitrile in the mobile phase so that the data set is smaller than that for the other mobile phases. For highly hydrophobic compounds with only two results, for 60 and 40% organic component in the mobile phase, 50% organic was also run in order to obtain the extrapolated aqueous value. Similarily, for some highly polar compounds, analyses at 0% organic component in the mobile phase were carried out. In Table 2 we have correlated log k′ values for basic and neutral compounds separately from those analytes that contained an acidic functionality. Attempts to correlate log k′ with log P for the entire compound set yielded a poorer linear relationship. Removal of the acids from the correlation analysis resulted in marked improvement of regression coefficients for all mobile phases containing acetonitrile. Under the experimental conditions used (pH 7.4), carboxylic acids and the aliphatic amines are expected to be ionized and negatively and positively charged, respectively. Examination of the data obtained has indicated that the acids clearly, unlike the amines, formed a distinct subgroup within the compound set. Austin et al.16 have shown that the charged form of certain amines is able to partition significantly into a phospholipid phase unlike carboxylate ions. However, this observation may be dependent on the structure of the acid as there are indications in the literature that arylalkyl carboxylic acid derivatives may interact with phospholipids differently from aromatic acids.17 A recent investigation18 concerned with the relationship between log P and log k′ employing an IAM stationary phase, with PC chemically bonded to the support, similarly found that carboxylic acids formed a distinct subset compared with neutral and basic compounds, the data being rationalized on the basis of the work by Austin et al.16 The implication of this observation with respect to the present investigation are 2-fold: (1) that the coated phase used is chromatographically comparable to a bonded phase and (2) that the charged form of basic amines interacts with the stationary phase in a manner similar to phospholipid vesicles. As can be seen from the equations given in Table 2, the confidence was greatly improved in estimating log P when (16) Austin, R. P.; Davis, A. M.; Manners, C. N. C. J. Pharm. Sci. 1995, 84, 1180. (17) Barbato, F.; Rotonda, M.; Quaglia, F. J. Pharm. Sci. 1997, 86, 225. (18) Salminen, T.; Pulli, A.; Taskinen, J. J. Pharm. Biomed. Anal. 1997, 15, 469.
Table 2. Parameters and Statistical Data for Equations Relating to log P vs log k′ from RP-HPLCa method (% organic)
compd group
N
R
R2
F
intercept
pb (intercept)
slope
pb (slope)
coated phase (60%)
all nonacids acids all nonacids acids all nonacids acids all nonacids acids all nonacids acids
99 86 13 100 85 15 90 73 17 102 86 16 74 62 12
0.52 0.83 0.52 0.69 0.86 0.80 0.74 0.81 0.71 0.78 0.79 0.79 0.57 0.77 0.69
0.27 0.68 0.27 0.48 0.74 0.63 0.55 0.65 0.54 0.61 0.62 0.62 0.33 0.60 0.48
35.87 180.27 3.96 90.60 235.88 22.31 108.33 131.85 17.57 156.16 134.78 23.08 35.40 88.40 9.21
3.29 2.97 5.42 2.56 1.93 4.30 1.15 0.62 2.06 0.52 0.38 1.00 3.02 2.80 6.59
