Caco-2 permeability of weakly basic drugs predicted with the Double-Sink PAMPA pKaflux method

https://doi.org/10.1016/j.ejps.2004.11.011Get rights and content

Abstract

The aim of this study was to analyze pH-dependent permeability of cationic drugs in Caco-2 cell monolayers using the pKaflux method and to correlate the results with those obtained in PAMPA (parallel artificial membrane permeability assay). The pH-dependent permeability of verapamil and propranolol was studied in Caco-2 cell monolayers. The data were subsequently processed using software developed for the PAMPA pKaflux method. Literature values for an additional nine cationic drugs were also analyzed. Double-Sink PAMPA data were also obtained for the same cationic drugs, to compare with the Caco-2 data. The Algorithm Builder program was then used to develop a predictive model of Caco-2 permeability based on PAMPA permeability and calculated Abraham molecular descriptors. From the relationship between permeability and pH it was shown that in PAMPA only the uncharged form of the drugs permeated across the membrane barrier, while charged and ionized forms of the drugs were significantly permeable in Caco-2. The charged-form permeability, Pi, was therefore determined and subsequently subtracted from all permeability coefficients in Caco-2 prior to the comparison with PAMPA. The resulting intrinsic permeability coefficients (Po) obtained in Caco-2 were successfully related to those derived from the PAMPA model. In this study we have shown that permeability coefficients obtained in PAMPA can predict the passive transcellular permeability in Caco-2.

Introduction

The established in vitro permeability assay in pharmaceutical research is based on cellular models, such Caco-2 and MDCK (Karlsson and Artursson, 1991, Adson et al., 1995, Yamashita et al., 2000, Ho et al., 2000, Neuhoff et al., 2003). One aim of such models is to predict passive human oral absorption of test compounds. More recently, the parallel artificial membrane permeability assay, i.e., PAMPA, has started to serve as an adjunct to the prevailing cellular methods (Kansy et al., 1998, Avdeef, 2003). Even as such, its worth had to be demonstrated by comparisons to the established cellular models. But such comparisons have proved to be illusive, largely due to inconsistencies in how (1) assay pH, (2) the aqueous boundary layer (ABL), and (3) incomplete mass balance (%R) are treated in cell assays by the different laboratories.

Over 30 PAMPA model variants have been documented in considerable detail (Avdeef, 2003), only a few of which had been described in the prior literature. The “Double-Sink” PAMPA model, highly refined to predict the published human jejunal permeability values (Avdeef, 2003), appears very promising. The so-called second generation PAMPA consists of (a) pH gradient “sink” for acids, (b) lipophilic gradient “sink” for bases, simulating the effect of serum proteins in the receiver compartment, (c) 20% (w/v) phospholipid mixture (some of which are negatively charged constituents) in dodecane, and (e) very efficient individual-well magnetic stirring, reducing the permeation time to 10–20 min for highly permeable compounds.

The first serious attempt to establish a reliable numerical comparison between Caco-2 and PAMPA, based on a comparative study of rat in situ intestinal permeability of 17 fluoroquinolones (including three congeneric series), resulted in a very promising outlook: both PAMPA and Caco-2 (when the ABL effect was properly treated) predicted rat in vivo data equally well (Bermejo et al., 2004).

Cellular methods are not entirely standardized, and the basis for comparison can be inconsistent. For example, cellular measurements are often done at just one pH, usually 6.5 or 7.4, not withstanding that the apparent permeability coefficient is pH-dependent for ionizable test compounds. The luminal pH can span from pH 5.0 to 8.0 between the proximal and distal portions of the small intestine. Since the surface adjacent to the enterocytes of the intestinal tract is believed to have an acid microclimate, ranging from pH 5.2 to 6.0 in the proximal jejunum to 6.9 in the distal ileum (Shiau et al., 1985, Said et al., 1986, Lucas and Whitehead, 1994), picking the “wrong” pH for the in vitro assay, in the worst-case circumstance, could lead to as much as a 100-fold (two pH units) misprediction of the oral absorption. Examples of Caco-2 measurements done at two different values of pH may be found (Adson et al., 1995, Pade and Stavchansky, 1997, Yamashita et al., 2000) and in four cases, seven to nine different pH values, ranging from 4.8 to 8.5, were employed to study the pH-dependent transport of cationic drugs, the moderately lipophilic alfentanil (Palm et al., 1999, Nagahara et al., 2004), metoprolol (Neuhoff et al., 2003), and the hydrophilic cimetidine (Palm et al., 1999, Nagahara et al., 2004) and atenolol (Neuhoff et al., 2003).

In addition to the pH effect issues, further uncertainties arise due to the different stirring speeds employed in cellular assays. Even though some test compounds may be expected to show pH-dependent permeability, the effect of the aqueous boundary layer (ABL) adjacent to the in vitro cellular barrier can obscure the pH-dependence when lipophilic compounds are examined (Karlsson and Artursson, 1991, Adson et al., 1995, Palm et al., 1999, Ho et al., 2000, Nagahara et al., 2004, Youdim et al., 2003). For example, even though the octanol–water log D at pH 7.4 of propranolol is about 30 times greater than that of metoprolol, Papp values (10−6 cm/s units), respectively, have been measured to be nearly the same: 59 and 51 (Karlsson and Artursson, 1991) 75 and 68 (Caldwell et al., 1998). Similar observations have been reported in PAMPA assays, where sometimes the reverse order between propranolol and metoprolol is observed (Wohnsland and Faller, 2001).

In blood–brain barrier (BBB) applications (Mahar Doan et al., 2002), the calculated “efflux” ratio, basolateral-to-apical (B  A) divided by the apical-to-basolateral (A  B) permeability, can be greatly underestimated if one or both directions of transport is ABL-limited. Also, the efflux ratio can be miscalculated for ionizable molecules when gradient-pH conditions are employed (Neuhoff et al., 2003). The consequences of these two important effects may not be widely understood.

Because of the efficient mixing near the surface of the intestinal epithelium, the in vivo ABL is estimated to be 30–100 μm thick (Rechkemmer, 1991, Winne et al., 1987, Anderson et al., 1988, Levitt et al., 1990). The ABL in the endothelial microcapillaries of the BBB is probably <1 μm, given that the diameter of the capillaries is about 6 μm and given the segmented-flow mixing effect of circulating erythrocytes (Krämer et al., 2001). But, in unstirred in vitro permeation epithelial cell models and in PAMPA, the ABL values can be 1000–4000 μm thick, depending on permeation cell geometry and dimensions, and efficiency of the stirring device (Karlsson and Artursson, 1991, Wohnsland and Faller, 2001, Ruell et al., 2003, Avdeef et al., 2004). These big differences between in vivo and in vitro ABL thicknesses have practical consequences.

If unstirred or mildly stirred (<100 rpm) cellular assays ignore the ABL effect with lipophilic test compounds, the measured permeability values will not correctly indicate the in vivo conditions of transport, and will merely reveal properties of water-limited, rather than membrane-limited transport (Ho et al., 2000, Avdeef, 2003). In early discovery projects, where binning decisions are made, an apparent permeability need only exceed a certain upper limit, e.g., (1–5) × 10−6 cm/s, for the compound to be accepted for further study. However, in lead optimization projects involving lipophilic molecules, the leveling effect of the ABL can produce a ‘flat’ structure–activity relationship, where structural changes in the molecules are made, but no significant changes in the measured permeability are observed. Four examples (Caldwell et al., 1998, Yazdanian et al., 1998, Irvine et al., 1999, Hilgers et al., 2003) of this are illustrated in Fig. 1, where log Papp is plotted against calculated or observed pH 7.4 octanol–water apparent partition coefficient, log D, values. In each case, the log Papp values appear to reach a maximum limit, suggesting ABL-limited transport. This “flatness” may mislead medicinal chemists concerning further structural modifications. Also, the effects of physiological pH on the tested compounds may not be correctly interpreted, due to the flatness. To overcome such potential limitation, it often is necessary to correct the apparent permeability coefficient for the effect of the ABL.

The traditional way to do the correction has been to perform the Caco-2 assay at several different stirring speeds, in the range 100–1000 rpm, and use a hydrodynamic equation to extract cellular from apparent permeability coefficients (Karlsson and Artursson, 1991, Adson et al., 1995, Palm et al., 1999, Ho et al., 2000). Another technique, first applied to black lipid membrane permeability measurements of ionizable molecules in the 1970s by Gutknecht and coworkers (Gutknecht and Tosteson, 1973, Gutknecht et al., 1977, Walter and Gutknecht, 1984) does not require stirring. A new extension of the Gutknecht procedure, called the ‘pKaflux method,’ has been applied to PAMPA models (Ruell et al., 2003, Bermejo et al., 2004). Calibration procedures have extended the method to non-ionizable molecules.

The third area of inconsistency of data treatment in cellular studies, that of mass balance, becomes problematic with highly lipophilic compounds (Sawada et al., 1994, Sawada et al., 1999, Wils et al., 1994, Ho et al., 2000, Krishna et al., 2001, Mahar Doan et al., 2002). During solute transport in the cellular assay, lipophilic compounds can rapidly reach equilibrium distribution between the aqueous and the cellular compartments. In one study, nearly 90% of a compound disappeared from the aqueous phase (Sawada et al., 1999). If the effect of mass balance is ignored, the apparent permeability can be significantly underestimated (Avdeef, 2003).

This paper considers the three areas of inconsistencies in data treatment in cellular studies, and focuses on establishing a quantitative relationship between PAMPA and Caco-2 passive permeability of weakly basic drugs, as a continuation of the effort started with the fluoroquinolone ampholytes series (Bermejo et al., 2004). To do this, reliable literature Caco-2 measurements of basic compounds, done at two or more pH points, were identified. For these molecules, PAMPA permeability coefficients were measured. The intrinsic permeability values, Po, were computed for both sets of data. The relationship between the two intrinsic permeability scales was explored using in silico methods based on molecular property descriptors. New Caco-2 data for verapamil and propranolol were collected in the pH 4.8–8.0 interval, with accounting for mass balance, to evaluate the ABL-correction using the so-called pKaflux model in Caco-2 cells, using PAMPA membranes as positive controls.

Section snippets

Materials

Alprenolol, atenolol, cimetidine, diltiazem, metoprolol, nadolol, pindolol, propranolol, terbutaline, and verapamil used in this study were purchased from Sigma-Aldrich (St. Louis, MO, USA), and used as received. [3H]-S-propranolol (703 GBq/mmol) and [14C]-mannitol (1.9 GBq/mmol) were purchased from NEN™ Life Science Products Inc. (Boston, MA, USA). Alfentanil was obtained as a 0.5 mg/mL saline solution (Rapifen IV) from Janssen-Cilag through Apoteket AB, Stockholm, Sweden. PAMPA lipid was

Results and discussion

Table 1 lists the physicochemical parameters of the bases, including the AB-calculated molecular descriptors. The lipophilicity of the molecules covers five orders of magnitude in the apparent octanol–water partition coefficient at pH 7.4.

Conclusion

New multi-pH Caco-2 permeability measurements of verapamil and propranolol, as well as published Caco-2 values of nine other molecules, were successfully correlated to those measured by the Double-Sink PAMPA model (Fig. 7). It was shown for the first time that two different methods (Eqs. (3), (7)) for correcting Caco-2 apparent permeability coefficients for the ABL effect produce the same cellular permeability coefficients. From the union of the pKaflux and the hydrodynamic models, it was

Nomenclature

All permeability coefficients are in units of cm/s.

    PAMPA

    parallel artificial membrane permeability assay

    PABL

    aqueous boundary layer (ABL) permeability coefficient; pH-independent

    Papp

    measured Caco-2 apparent permeability coefficient: 1/Papp = 1/PABL + 1/Pf + 1/Pc, usually not corrected for %R (with the exception of verapamil and propranolol); pH-dependent for ionizable molecules

    Pc

    cellular (Caco-2/MDCK) permeability coefficient; pH-dependent for ionizable molecules; Pc = Ptrans + Ppara + other mechanisms

    Pe

Acknowledgments

We thank Oksana Tsinman and Jeffrey Ruell for expertly conducting the PAMPA experiments. Part of this work was supported by AstraZeneca (Mölndal, Sweden).

References (50)

  • M.L. Lucas et al.

    A re-evaluation of the properties of the three-compartment model of intestinal weak-electrolyte absorption

    J. Theor. Biol.

    (1994)
  • N. Nagahara et al.

    The contribution of the paracellular route to pH dependent permeability of ionizable drugs

    J. Pharm. Sci.

    (2004)
  • J.A. Ruell et al.

    PAMPA – a drug absorption in vitro model. 5. Unstirred water layer in iso-pH mapping assays and pKaflux – optimized design (pOD-PAMPA)

    Eur. J. Pharm. Sci.

    (2003)
  • S. Yamashita et al.

    Optimized conditions for prediction of intestinal drug permeability using Caco-2 cells

    Eur. J. Pharm. Sci.

    (2000)
  • K.A. Youdim et al.

    in vitro trans-monolayer permeability calculations: often forgotten assumptions

    Drug Discov. Today

    (2003)
  • D. Zmuidinavicius et al.

    Classification structure–activity relations (C-SAR) in prediction of human intestinal absorption

    J. Pharm. Sci.

    (2003)
  • B.W. Anderson et al.

    Physiological measurement of luminal stirring in perfused rat jejunum

    Am. J. Physiol.

    (1988)
  • A. Avdeef

    Absorption and Drug Development

    (2003)
  • G.W. Caldwell et al.

    in vitro permeability of eight β-blockers through Caco-2 monolayers utilizing liquid chromatography/electrospray ionization mass spectrometry

    J. Mass Spectrom.

    (1998)
  • E.L. Cussler

    Diffusion Mass Transfer in Fluid Systems

    (1997)
  • F. Delie et al.

    A human colonic cell line sharing similarities with enterocytes as a model to examine oral absorption: advantages and limitations of the Caco-2 model

    Crit. Rev. Ther. Drug Carrier Syst.

    (1997)
  • I. Elimrani et al.

    Expression and localization of organic cation/carnitine transporter OCTN2 in Caco-2 cells

    Am. J. Physiol. Gastrointest. Liver Physiol.

    (2003)
  • J. Gutknecht et al.

    Diffusion of weak acids across lipid membranes: effects of chemical reactions in the unstirred layers

    Science

    (1973)
  • J. Gutknecht et al.

    Diffusion of carbon dioxide through lipid bilayer membranes. Effects of carbonic anhydrase, bicarbonate, and unstirred layers

    J. Gen. Physiol.

    (1977)
  • A.R. Hilgers et al.

    Predicting oral absorption of drugs: a case study of a novel class of antimicrobial agents

    Pharm. Res.

    (2003)
  • Cited by (196)

    • Impact of cholesterol and sphingomyelin on intrinsic membrane permeability

      2022, Biochimica et Biophysica Acta - Biomembranes
    View all citing articles on Scopus

    Contribution No. 9 from pION in the series: PAMPA – a Drug Absorption in vitro Model. Avdeef et al. (2004) is part 11 of the series.

    1

    Present address: Department of Chemistry (Biophysical Chemistry), Umeå University, SE-901 87 Umeå, Sweden.

    View full text