Anal. Chem. 1995,67, 48-57
Bioavailability Estimation by Reversed-Phase Liquid Chromatography: High Bonding Density C-18 Phases for Modeling Biopartitioning Processes Mei=MingHsieh and John 0. Dorsey* Deparfment of Chemistry, Universiiy of Cincinnati, Cincinnati, Ohio 45221-0172
There have been many attempts to estimate biological activity with either 1-octanoVwaterpartition coefficients or chromatographic retention parameters. Bulk phases may not be appropriate, however, for modeling a partitioning process in an interphase such as biological membranes. Chromatographic stationary phases can be argued as having structure similar to a membrane because of chain organization; however, the density of the grafted stationary-phasechains in commercially available stationary phases is much too low to provide a suitable model. We have previously developed a new scheme for derivatizing silica surfaces that produces stationa.~~ phases of significantly higher chain density than traditional methods. Investigation of the molecular mechanism and thermodynamics of solute partitioning into the different phases has shown that densely bonded reversed-phase stationary phases mimic partitioning to a biomembrane better than does bulk-phase octanol. Here we report chromatographic retention for pesticides, PAHs, and barbiturates using a C-18 column with high alkyl chain density, and in all cases, correlations of log with bioavailability are equivalent to or better than correlations of bioavailability with the octanoVwater partition coefficient, Prediction of toxicity, reactivity, and transport parameters is highly demanded in pharmaceutical and environmental areas for the design of new drugs and estimation of the environmental fate of synthetic chemicals, because the exact measurement is very expensive,technically difllcult, and time consuming. For example, in 1985, the cost of the laboratory procedure for estimating the environmental fate and transport properties of a single chemical was estimated to be as much as $100 000. Such measurements have been made for only about 1%of the approximately 70 000 compounds in the EPA’s inventory of manufactured chemicals.’ Quantitative structure activity relationships (QSARs) represent the entire area of such estimations. Among the structure/property descriptors used in QSAR correlations, the hydrophobic parameter, log P,the oil/water partition coefficient measured by a shake flask method, is the most popular and commonly encountered descriptor. The partition coefficient, P,for a compound distributing between immiscible organic and aqueous phases is defined * Resent address: Department of Chemistry, Florida State University, Tallahassee, FL 323063006. (1) Borman, S. Chem. Eng. News 1990,(Feb 19), 20-23.
48 Analytical Chemistry, Vol. 67, No. 1, January 1, 1995
by the equilibrium concentration of the analyte in the organic phase divided by the analyte concentration in an aqueous phase. An octanol/water system is most often used to represent the fatty and the aqueous phases. The introduction of high-performance liquid chromatography (HPLC) for lipophilicity determination in the early 1970s called attention to the advantages of the chromatographic method over the conventional “shake-flask” technique for QSAR studies. The shake-flask partition coefficient measurement is time consuming and requires very pure compounds, and the precision of the measurement for compounds with log P > 4 is usually poor because of the errors involved when determining very low concentrations in the aqueous phase. The chromatographic methods have the advantages of high speed of determination, less stringent purity requirements for the compounds selected, small amount of material needed, reproducibility, ease of use, and instrumentation available for automation. For these reasons, numerous papers have been published regarding the liquid chromatographic procedures for hydrophobicity determination and the correlation of the data obtained with the logarithm of the partitioning system, log P. The ultimate goal of all such efforts is the search for optimal conditions of both stationary phase and mobile phase, so the physicochemical and biological properties which are related to the hydrophobicity of a compound can be estimated. Two recent reviews of hydrophobicity estimations by reversed-phase liquid chromatography (RPLC) have been pub lished by Dorsey and Khalediz and by Lambert3 As long as log Pow (l-octanol/water partition coefficient) is accepted to represent hydrophobicity, several procedures have been reported based on dynamically coating a stationary phase with 1-octanol and using a 1-octanol-saturated aqueous eluent. Yet there is some question as to whether a bulk liquid such as l-octanol is really a good model for biological permeation barriers and adsorption sites. Other membranelike coating agents, such as oleyl alcohol, silicone oil, and liquid paraffin, have also been examined. A more recent relevant methodological publication is by Miyake et a1.4 in which phosphatidylcholine is recommended as the coating agent. Two common disadvantages in all of these octanol-like or membranelike chromatographic systems are that there is some question as to whether the column characteristics change during use and chromatography of more hydrophobic solutes (log Pow> 3) is practically impossible. (2) Dorsey, J. G.; Khaledi, M. G. j.Chromatogr. 1993,656,485-499. (3) Lambert, W. J. j . Chromatogr. 1993,656,469-484. (4) Miyake, K; Kitaura, F.; Mizuno, N.; Terada, H. j.Chromatogr. 1987,389, 47-56.
0003-2700/95/0367-0048$9.00/0 0 1994 American Chemical Society
The alternative to the stationary phases dynamically coated with a lipophilic substance has been the physically stable hydrophobic stationary phases. The stable reversed-phase HPLC systems, used for hydrophobicity assessment since 1975, employ alkyl (usually octadecyl) ligands chemically bonded to the silica support surface. Generally, the correlations of log P and the HPLC data obtained on hydrocarbonaceous silica materials are good only as long as the solutes analyzed are closely related (congeneric). M e r years of study, however, there is no universally accepted method of performing these estimations. The linear correlation coefficients range from 0.5 to 0.999, depending on the choice of the columns and compounds tested. For example, Thus and Kraak5 reported that a phenyl bonded column gave signiscantly better correlations than an octadecyl (C-18) bonded phase. Minick et al.6 using the same type of phenyl column, but a different octadecyl column, reported significantly better correlations with the C-18 column. The approach used to verify the validity of the HPLC method of hydrophobicity assessment is correlation of retention parameters with the 1-octanol/water partition coefficient, log P. However, there is no reason to assume log P i s the best possible descriptor in QSAR The process that either the 1-octanol/water system or HPLC are designed to model is a partitioning process. The partitioning of solute molecules into lipid bilayers and biological membranes is the basis for drug and metabolite uptake, passive transport across membranes, and bioaccumulation. Bulk phases like 1-octanol/water may not be appropriate for modeling a partitioning process in an interphase such as biological membranes, and the bonding density of almost all of the commercial C-18 columns may be too low to provide a suitable model. Comparison of solute partitioning in bulk phases, the stationary phase/mobile phase in RPLC, and membrane/water systems has been extensively studied by Dill et aL7-l0 From the structure point of view, 1-octanol/water is a bulk phase while lipid bilayer membranes are interfacial phases. Interfacial phases have high surface to volume ratios, and their physical properties vary with distance from the interface. In contrast, physical properties in bulk phases are uniform throughout. For example, there is a gradient of chain disorder in the hydrocarbon core of the bilayer; the surfactant chains are most highly aligned near the head groups, and the order diminishes with distance toward midlayer." Moreover, the chain ordering of the bilayer phospholipids increases with surface density. Properties of interfacial phases depend on surface density whereas properties of bulk phases do not. These structural differences between bilayers and other interfacial and bulk phases such as oil or octanol should be manifested as differences in the nature of solute partitioning into them. Dill's theory7,*has predicted the following: (i) There will be an equilibrium gradient of solute concentration in the bilayer in contrast to the uniform distribution expected in a bulk phase; this prediction is consistent with neutron scattering experiments.12 (5) Thus, J. L. G.; Kraak, J. C. ]. Chromatogr. 1985, 320, 271-309. (6) Minick, D. J.; Sabatka,J. J.; Brent, D. A]. Liq. Chromatogr. 1987,10,25652589. (7) Marqusee, J. A; Dill, K. A ]. Chem. Phys. 1986,85, 434-444. (8) Dill, K A; Naghizadeh, J.; Marqusee, J. A Annu. Rev. Phys. Chem. 1988, 39, 425-461. (9) DeYoung, L. R; Dill, K A Biochemistry 1988,27, 5281-5289. (10)DeYoung, L. R; Dill, K A J. Phys. Chem. 1990, 94,801-809. (11) Dill, K A; Flory, P. J. Proc. Nutl. Acad. Sci. U.SA. 1980, 77, 3115-3119. (12) White, S. H.; King, G. I.; Cain, J. E. Nature 1982, 290,161-163.
(i) The partial chain ordering should disfavor solute retention in the bilayer relative to amorphous bulk phases. (ii) Solute uptake should decrease significantly with increased surface density of the chains. There is also evidence in favor of (ii) and (iii). Partition coefficients of many anesthetic agents into membranes are 2-15 times lower than their coefficients of partitioning into olive oil.l3 More recent experimentsgJOhave measured the membrane/water partition coefficients of benzene into lipid bilayers as a function of the surface density of the phospholipid chains and have shown that partitioning into the bilayer is dependent not only on the partitioning chemistry but also on the surface density of the bilayer chains. Increasing surface density leads to solute expulsion; benzene partitioning decreases by 1 order of magnitude as the surface density increases from 50 to 90% of its maximum value. This theory then predicts that partitioning methods based on bulk phases cannot be an accurate representation of an interfacial partitioning process. On the other hand, there are a number of similaritiesthat exist between the mobilephase/stationary-phaseinterface in RPLC and the membrane/water interface. The chemically bonded phase does not behave as a liquid but resembles much more the ordered array of the membraneous hydrocarbon chains.14 The residual silanol groups, some of them charged at neutral pH, and the adsorbed layer of organic modifier and coextracted water molecules may be expected to resemble the polar, outer membrane regions. However, the grafted chain density of all the commercially available columns is much too low to provide a suitable model for a biomembrane.2 Sentell et al.15 developed a method of synthesizing reversed-phase stationary phase with bonding density as high as 4.1 pmol/m2 for C-18 ligands. The partitioning processes for some small solutes were studied on columns with a bonding density range from 1.6to 4.1 pmol/m2, and a maximum in partition coefficient was found at approximately 3.0 pmol/m2.16 At low densities the partition coefficient increases with bonding density, as the silica surface is becoming depolarized and chain interactions are negligible. Beyond the critical stationary-phase bonding density, the partition coefficient begins to decrease as chain interactions begin to become the dominant force, and entropic expulsion of solutes from this interphase occurs. The newly synthesized stationary phases with highly bonded alkyl chain densities should behave more like biological membranes, and partition coefficients measured using these phases should give highly relevant values for modeling biological partitioning processes. The failure of the l-octanol/water modeling for bioaccumulation processes has been elegantly demonstrated by Opperhuizen et They investigated the thermodynamic properties of the partitioning of chlorobenzenes between fish lipids and water and showed that bioconcentration is accompanied by positive enthalpy and entropy changes. In contrast, the partitioning of these compounds between octanol and water is accompaniedby negative enthalpy and by small negative or positive entropy changes. They concluded that this is because of the different structures of fish lipids and octanol and that only under very specific conditions (13) Seeman, P.W. Pharmacol. Rev. 1972, 10, 353-418. (14) Dorsey, J. G.; Dill, K. A Chem. Rev. 1989,89,331-346. (15) Sentell, K B.; Barnes, K W.; Dorsey, J. G.J. Chromatogr. 1988,455,95104. (16)Sentell, K B.; Dorsey, J. G. Anal. Chem. 1989, 61,930-934. (17) Opperhuizen,A; Seme, P.; Van der Steen, J. M. D. Enuiron. Sei. Technol. 1988,22,286-292.
Analytical Chemistry, Vol. 67, No. 1, January 1, 1995
49
and only for structurally similar compounds can a relationship between 1-octanol/water partitioning and bioaccumulation be expected. Even the more comprehensive fugacity-based models of bioavailability are generally based on an 1-octanol/water relationship.l8Jg The thermodynamicsof transfer of alkanes and monofunctional saturated organics between water and octanol was also studied by Cabani et al?O By using gas-phase transfers for both water and octanol, they were able to calculate enthalpies and entropies for the transfer of solutes between octanol/water mixtures. They found that the values of both the enthalpy and entropy are positive. Their results are in disagreement with the results of Opperhuizen et al.17 This may indicate that the thermodynamic driving force is dependent on the type of solute partitioning from water to octanol. The thermodynamics of retention of small molecules on a wide range of stationary phases in RPLC have been investigated by Cole et a1.21322 Using various mobile phases with columns of bonding density from 2.39 to 4.07 pmol/m2, they found that as the stationary-phase bonding density increases, the retention process is governed more by the entropic difference of the solute between the two phases. The thermodynamics in RPLC is in agreement with the thermodynamic driving force for partitioning of pesticides between water and fish lipids found by Opperhuizen et Cole and Dorsey also compared enthalpies and entropies of transfer in RPLC with previously reported values for the partitioning of a nonpolar solute from bulk organic liquid to water and found that the chromatographic process, even with low bonding density stationary phases, is not well modeled by bulkphase oil/water partitioning processes.21 In this paper, chromatographic retention of pesticides, barbiturates, and PAHs was measured on a high bonding density stationary phase (4.07 pmol/m2) and a low bonding density stationary phase (2.39 pmol/m2) using several compositions of methanol/water mixtures. Log k’, values, the capacity factor with a mobile phase of 100%water, were calculated from extrapolating a plot of log k‘ vs volume percent of MeOH and a plot of log k’ vs E~(30)scale for each compound. Calculated log k’, values were then correlated with diverse biological actions and estimations of bioavailability from lipophilicity determined by RPLC and octanol/ water were compared. EXPERIMENTAL SECTION Materials. HPLC grade methanol (Fisher Scientific, Fair Lawn, NJ) was used without further puri6cation. Water was obtained from a Barnstead Nanopure I1 water purification system (Barnstead Co., Boston, MA) fitted with a 0.45 pm filter. Pure solutes were used as received, and stock solutions were made in HPLC grade methanol. All pesticides were standards from the U.S. Environmental Protection Agency. All barbiturates, phenanthrene, anthracene, and benz[alanthracene were from Sigma Chemical Co. (St. Louis, MO). Naphthalene was from Eastman Kodak (Rochester, NY), pyrene was from Chem Service (West Chester, PA), and benzene was from Aldrich (Milwaukee, MI). (18)Clark, K E.;Gobas, F. A P.; Mackay, D. A M. Enuiron. Sci. Technol. 1990, 24,1203-1213. (19)Trapp, S.; Mathies, M.; Scheunert, I.; Topp, E. M. Enuiron. Sei. Technol. 1990,24,1246-1252. (20)Cabani, S.;Conti, G.; Mollica, V.; Bemazzani, J.J. Chem. Soc., Furuduy Truns. 1991,87,2433-2442. (21)Cole, L.A;Dorsey, J. G. Anal. Chem. 1992,64,1317-1323. (22)Cole, L. A;Dorsey, J. G.; Dill, K. A Anal. Chem. 1992,64,1324-1327.
50 Analytical Chemistry, Vol. 67, No. 1, January 1, 1995
The chemicals were dissolved in methanol to give 10-1000 ppm solutions. A sodium phosphate buffered aqueous phase (PH 7.00) was used for barbiturates prior to mixing with methanol. Retention Measurements. All retention measurements were made using an AB1 Analytical Kratos Division (Ramsey, NJ) Spectroflow 400 pump, with UV detection at 254 nm using a Kratos Spectroflow 757 variablewavelengthabsorbance detector. Sample injection was performed using a Valco (Houston, TX) C6W injector fitted with a 10 pL sample loop, and detector output was recorded on a Scientific Products Quantigraph chart recorder (Houston Instrument, Austin, nr).Two “homemade” C-18 columns, 5 cm x 4.6 mm id., with alkyl chain densities of 4.07 and 2.39 pmol/ m2,respectively, were used. The base silica material was from a single lot of 10 pm particle diameter Davisil (W. R Grace, Baltimore, MD) with a surface area (&ET) of 300 m2/g and a pore size of 150 A. The synthesis, packing, and bonding density determination may be found elsewhere.15 Each column was thermostated at 30 “C by use of a water jacket and a Brinkmann Lauda (Westbury, MI) Model MT heater/circulator. The eluent flow rate was varied from 1.0 to 4.5 mL/min, depending on the mobilephase composition. Retention times were determined manually based on the peak maximum position. The breakthrough time (0used to calculate capacity factors was determined at each composition by the elution of an injection of DzO (Sigma, St. Louis, MO). RESULTS AND DISCUSSION
Chromatographic Retention Measurement. Most often the logarithm of the capacity factor, log k’, is chosen as the experimental variable to correlate with bioavailability. However, according to the study of Miyake et linear correlation coefficients for plots of log Powvs log k’ using certain compositions of methanol/water mixtures increase as the water content in the mobile phase increases, indicating that log k’,, the capacity factor with a mobile phase of 100% water, is a better descriptor of lipophilicity than log k’. The advantages of using log k’, are that it is independent of any organic modifier effects, it reflects polar/ nonpolar partitioning in a manner similar to shakeflask measurements, and it is dependent on the solute’s structure and polar fun~tionalities.”-~6 The log k’, parameter is difticult to measure directly, because of prohibitively long retention times. Thus, log k’, is most often estimated by extrapolatinga linear plot of log k’ vs volume percent organic modifier to the intercept with 0% organic modifier, representing retention in a 100%aqueous phase: log k’ = log k’, - S(Vd %)
(1)
An alternative method to estimate the log k’, value uses the ET (30) solvatochromic scale, an empirical measure of solvent p0iarity:27
log k’ = a
+ s(k(30))
(2)
Here log k’, is obtained by extrapolating a linear plot of log k’ vs &(30) polarity to the intercept with the E~(30)value of 63.11 kcal/ (23)Miyake, K;Miauno, N.; Terada, H. J. Chmmufogr. 1988,439,227-235. (24)Braumann, T.; Weber, M.; Grimme, L. H.J. Chromatogr. 1983,261,329343. (25)Schantz, M. M.; Martire, D. E. J. Chromatogr. 1987,391,35-51. (26)Snyder, L. R;Dolan, J. W.; Gant, J. R 1.Chromatogr. 1979, 165,3-30. (27)Johnson, B.P.;Khaledi, M. G.; Dorsey, J. G. Anal. Chem. 1986,58,23542365.
mol for pure water. The E~(30)scale is based on the charge transfer absorption of 2,&diphenyl-4(2,4,&triphenyl-n-pyridinio) phenolate (ET-30), which is sensitive to both solvent dipolarity/ polarizability and solvent hydrogen bond donor ability. Using an extensive base of 332 sets of retention data as a function of mobile phase composition, Johnson et aLZ7compared the E~(30)scale with the commonly used volume percent plots and showed that log k’ is more linearly related to the E ~ ( 3 0 scale ) than volume percent for most of the data sets. It was further found that, for a solute/column pair, extrapolation of log k’ vs volume percentage of organic modifier gives signiiicantly different intercepts for different organic modifiers, while the extrapolation of log k’ vs E ~ ( 3 0leads ) to a common intersection point at approximately the E~(30)values of pure water.28 Using over 200 sets of chromatographic retention data, which were entirely different data from those in ref 27, Michels and Dorseyz9compared the estimation of log k‘, by the volume percentage approach and by the E ~ ( 3 0 ) approach and found that the &(30) approach gave a more reliable estimation of this lipophilicity parameter. This evaluation was based on the relative value of the 95%confidence interval about log k’,, the point of intersection of log k‘ vs solvent strength plots for different modifiers, the scatter of estimations with different modifiers, and the goodness of fit of the data to the linear model. The E~(30)approach was also found to be equivalent to or better than the quadratic approach to estimating k’,, where a quadratic dependence on volume percent organic modifier is assumed.B A thorough comparison of measured vs extrapolated log k’, value has been published by Hsieh and Dorsey30for compounds with a variety of functionalities. Although no extrapolated value gives a true estimate of the measured value for every solute, the E ~ ( 3 0 ) extrapolated value gives the most useful and reliable measure.
Estimation of Biotransfer Factors of Pesticides for Beef and Milk. Bioconcentration factor is a measure of the potential of a chemical to accumulate in an organism, and measured or predicted bioconcentration factors are very important for environmental risk assessment. The biotransfer factors for beef (BI,) and milk (Bd,defined as follows, for a number of pesticides were measured and correlated to the 1-octanol/water partition coefficients by Travis and Arms.31
Bb = concn in beef (mg/kg)/ daily intake of organic (mg/d) (3)
B, = concn in milk (mg/kg)/ daily intake of organic (mg/d) (4) The linear correlations between the 1-octanol/water partition coefficients and biotransfer factors for beef and milk were reported as31 log Bb = -7.735 log B, = -8.056
+ 1.033 log Pow + 0.992 log Pow
n = 36, r = 0.81 (5) n = 28, r = 0.74 (6)
The regression plots for the above equations are highly scattered as shown in Figures 1 and 2 of ref 31. In this study, capacity factors of selected pesticides were measured on two C-18 columns with bonding densities of 2.39 (28) Michels, J. J.; Dorsey, J. G.J. Chromafogr. 1988,457, 85-98. (29) Michels, J. J.; Dorsey, J. G. J. Chromatogr. 1990,499, 435-451. (30) Hsieh, M. M.; Dorsey, J. G. J. Chromatogr. 1993,631, 63-78. (31) Travis, C. C.; Arms, A D.Environ. Sci. Technol. 1988,22, 271-274.
la’
O
I a
a
-6
4
