Article

Top Abstract Article References

Significant recent interest has focused on predicting oral drug absorption from drug structure (Wessel et al., 1998; Clark, 1999; Ghuloum et al., 1999; Egan et al., 2000; Norinder and Österberg, 2001; Stenberg et al., 2001; Zhao et al., 2001). Several misperceptions regarding the nature of absorption seem common in many of these approaches. Among these is that intestinal absorption, permeability, fraction absorbed, and, in some cases, even bioavailability are all equivalent and can be used interchangeably. In reality, these are different processes and are related to drug structure in different ways. A consequence of failure to appreciate these differences is development of a model from one property, such as permeability, and applying it to predict a different property, such as bioavailability, which is dependent upon permeability, solubility, and first-pass metabolism, for example. A second common misperception is that absorption, permeability, fraction absorbed, etc. are fundamental properties of the molecule and can be predicted directly from some structural representation of the drug. In reality, drug absorption is a complex process dependent upon drug properties such as solubility and permeability, formulation factors, and physiological variables, including regional permeability differences, pH, luminal and mucosal enzymology, and intestinal motility, among others. In this article, we will explore the influences of these different variables on drug absorption and the implications with regards to current attempts to develop predictive drug absorption algorithms.

The bioavailability of a drug (F) after oral administration is described in general by the following relationship.
where f_a, f_g, f_h are the fractions of intact drug absorbed (f_a) that escape irreversible elimination as a drug passes sequentially from the gastrointestinal tract, across the gut wall (f_g), and traverses the liver (f_h) into the systemic circulation (Pond and Tozer, 1985). Thus, bioavailability can be equal to or less than the fraction absorbed depending upon the extent of metabolism and loss during the absorption process. Figure 1 shows these relationships schematically for a number of combinations of absorption and metabolism. Comparison of panels 2 and 3, for example, clearly show that poor blood levels of a drug can be a consequence of poor absorption or good absorption accompanied by extensive metabolism.

View larger version (18K): [in this window] [in a new window]   Fig. 1.   Schematic representation of impact of metabolism on drug absorption. In the first panel, absorption of the drug is high as reflected in plasma concentration (Cp) of total drug-related material (DRM) with time. DRM is defined as parent drug and any metabolites formed during the absorption process (Pond and Tozer, 1985). In this case, metabolism and other drug losses are minimal such that the concentration of parent drug approaches that of DRM. In the middle panel, absorption is high, resulting in high circulating levels of DRM, but since metabolism is significant, parent drug concentrations are low. The last panel illustrates the case of low absorption and correspondingly low metabolism. By comparing the last two panels, it can be seen that low circulating parent drug levels can be a consequence of either poor absorption or good absorption accompanied by extensive metabolism. Although historically metabolism during absorption was thought to occur primarily in the liver, increasingly the contribution of the intestinal mucosa to first-pass metabolism is being recognized (Wacher et al., 2001).

Both fraction absorbed and bioavailability are measures of the extent of absorption. In contrast, permeability is related to the rate of absorption.
where J, the absorptive flux, is equal to the permeability of the intestinal mucosa to the drug, the surface area available for absorption, and the concentration gradient across the mucosa (Ho et al., 1977). Factors that can influence permeability include structural characteristics, such as size, shape, charge, etc., in addition to position in the intestinal tract, drug formulation components, and intestinal pH. Factors affecting the concentration gradient include intrinsic solubility of the drug, formulation components, solid-state properties, etc.

It has been argued on theoretical grounds that a fundamental relationship exists between the rate measured as a permeability coefficient and the extent of absorption (Sinko et al., 1991). This has led to the increasing use of in vitro permeability models, such as the Caco-2 cell monolayer system, to serve as an experimental surrogate for predicting oral absorption potential of drug candidates in a drug discovery setting (Stewart et al., 2000; White, 2000). Although in some cases it has been possible to directly correlate absorption with permeability (Artursson and Karlsson, 1991), more often the very poor correlation shown in Fig. 2a for a series of synthetic antibacterial agents is observed in the drug discovery setting. In this particular example, rat bioavailability is compared with Caco-2 cell permeability coefficients for a series of oxazolidinone antibiotics (Hilgers et al., 2002). As we have discussed, although bioavailability is a function of both absorption and metabolism, for these compounds, metabolism is relatively minimal, and to a first approximation, the fraction absorbed and bioavailability are equivalent. Particularly striking is the range of absorption observed for the poorly permeable compounds. What is common among the poorly permeable, well absorbed solutes, however, is high aqueous solubility, generally exceeding 2.5 mg/ml (Hilgers et al., 2002), suggesting that this property may help to compensate for the poor in vitro permeability observed.

View larger version (23K): [in this window] [in a new window]   Fig. 2.   Oral absorption of oxazolidinone antibiotics in the rat. A, oral bioavailability is poorly correlated with Caco-2 cell permeability (adapted from Hilgers et al., 2002). Compound I (Zyvox) and II (Eperezolid) were the first two clinical candidates in this structural class. Common among the low-permeability well absorbed solutes is high solubility. Eperezolid, for example, has aqueous solubility of 4.2 mg/ml. In contrast, the low-permeability poorly absorbed solutes have solubilities generally less than 0.05 mg/ml. B, shows better correlation if both Caco-2 cell permeability and aqueous solubility are used to predict fraction dose absorbed using the maximum absorbable dose model (see text for details). Compounds identified with solid squares are the most significantly underpredicted members of the group. These oxazolidinones share characteristics of high Caco-2 cell permeability and very limited aqueous solubility (less than 0.01 mg/ml; Hilgers et al., 2002).

A simple model for interrelating permeability and solubility of a drug to estimate absorption potential is maximum absorbable dose (MAD) (Johnson and Swindell, 1996). Briefly, MAD calculates the total mass of a drug that could be theoretically absorbed if a saturated solution of the compound with solubility S in the small intestinal volume (SIV) were absorbed with a first-order rate constant, K_a, for a time equivalent to the small intestinal transit time (SITT):
Using this model parameterized for the rat and comparing the predicted MAD to dose actually administered, the relationship shown in Fig. 2b was found. Clearly, better prediction of absorption is achieved when solubility, permeability, and dose are taken into consideration compared with the situation obtained with a single parameter correlation.

Structure Permeability Relationships

Permeability is still an important, if not the sole, determinant of absorption, and it is informative to explore mechanisms contributing to permeability given the interest in development of structure-based computational models of this property. To reach the systemic circulation, a drug must move from the intestinal lumen through an unstirred water layer and mucus coat adjacent to the epithelial cell surface. Movement across the epithelial layer takes place by two independent routestranscellular flux (i.e., movement across the cell) and paracellular flux, or movement between adjacent epithelial cells, restricted by the presence of tight junctions between the cells (Diamond, 1977; Gumbiner, 1987; Jackson, 1987). The solute then encounters a basement membrane, interstitial space, and mesenteric capillary wall in accessing the mesenteric circulation. Any and all of these microenvironments can be considered a resistance to solute movement with an associated permeability coefficient. Thus, the overall process consists of a number of resistances (reciprocal permeabilities) in series (Ho et al., 1977). Furthermore, the influence of drug structure with permeability in these different domains will be different. For example, unstirred water layer permeability is inversely related to solute size, whereas paracellular permeability is dependent upon both size and charge. In the latter case, the characteristics of the paracellular "pore" result in size resisted diffusion as the size of the solute approaches that of the paracellular space. Furthermore, cations are more permeable than neutral species, which in turn are more permeable than anions, consistent with the negative charge characteristics of the paracellular space (Kottra and Frömter, 1983; Adson et al., 1995).

With respect to transcellular permeability, the relationship of solute structure with permeability depends upon the mechanism involved. Historically, a passive diffusion pathway has been assumed for most solutes. Nevertheless, an increasing number of active absorptive and secretory processes in intestinal epithelial cells are being identified for which many common drugs are substrates (Tsuji and Tamai, 1996). Although active transport involves specific interactions between solute and transporter, passive diffusion is dependent upon solute partitioning into the cellular plasma membrane and diffusion coefficient within the membrane (Jackson, 1987). Both these processes are influenced by the physicochemical and structural characteristics of the drug. Factors influencing plasma membrane partitioning are solute size, lipophilicity, hydrogen bonding potential, and charge characteristics, whereas diffusion is dependent upon size or total molecular surface area properties (Conradi et al., 1996). In general, nonpolar surface area favors partitioning, whereas polar hydrogen bonding functionality do not favor partitioning. With respect to diffusion, an inverse relationship with size is found, similar to the situation with paracellular permeability.

These multiple influences on permeability are manifested in a number of different ways. If intestinal permeability of a number of homologous nonactively transported solutes is measured as a function of membrane partitioning or, more commonly, an organic solvent partition coefficient as a surrogate, a sigmoidal relationship is frequently observed (Ho et al., 1977; Camenisch et al., 1996). For solutes with little or no membrane affinity, permeability is low, resulting primarily from paracellular diffusion of the solute. As the propensity of the solute to partition into the cell membrane increases, permeability also increases as a result of the significant increase in surface area of the transcellular pathway relative to the paracellular route. This increase in permeability will approach a plateau value beyond which further increases in partition coefficient do not result in increased permeability. This is the so-called aqueous boundary layer-limited situation where diffusion across the cell is very rapid relative to diffusion of the solute through the unstirred water/mucus layer adjacent to the cell (Westergaard and Dietschy, 1974). Perturbing hydrodynamics that shift the plateau to a new limiting permeability can modify the dimensions and resistance of this layer.

In the case of ionizable solutes, permeability is also pH-dependent. The neutral uncharged species is capable of transcellular passive diffusion, whereas the charged species is restricted to the paracellular pathway. Thus, the observed permeability of such molecules is dependent upon the relative concentrations of charged and neutral species. In the case of a weak acid, such as salicylic acid, at pH less than about 5.5, rat intestinal permeability is aqueous boundary controlled. Increasing pH results in progressively lower permeability coefficients. At pH greater than 9, a limiting small permeability is achieved that is independent of further pH increases. This limiting permeability represents the paracellular diffusion of the charged anion (Ho et al., 1983).

For solutes that are substrates for active uptake or efflux transporters, the relative contribution of the active pathway will depend upon the concentration of the solute in the lumen. For example, lisinopril is absorbed by a peptide transporter in the mucosa and shows progressively lower absorptive permeability in rat intestine as the concentration of drug is increased (Swaan et al., 1995). Other drugs that have been shown to be substrates for absorptive transporters include L-DOPA, baclofen, and melphalan (large neutral amino acid transporters), cephalasporin and -lactam antibiotics (oligopeptide transporters), pravastatin (monocarboxylic acid transporters), forscarnet and fosfomycin (phosphate transporters) (Tamai and Tsuji, 1996).

More recently, the role of efflux transporters in influencing the permeability and overall bioavailability of drugs has gained considerable attention. Among these transporters is P-glycoprotein (P-gp) expressed on the apical surface of normal intestinal mucosa. In contrast to absorptive transporters that increase uptake of substrates from the intestinal lumen, P-gp impedes uptake by returning a portion of drug entering the mucosa back to the lumen in a concentration-dependent manner. An increasing number of drugs have been shown to be substrates for P-gp, including human immunodeficiency virus-protease inhibitors (Li et al., 2002) and verapamil (Sandstrom et al., 1998), where intestinal permeability is increased as lumen concentration increases. The potential for such P-gp-mediated intestinal efflux to provide a source of drug-drug interactions affecting oral bioavailability is also increasingly being described (Matheny et al., 2001). A recent example described significantly increased bioavailability of fexofenadine in the presence of St. John's wort (Wang et al., 2002). Given the importance of P-gp in drug absorption, attempts have been made to develop structure-transport relationships with varying degrees of success (Stouch and Gudmundsson, 2002). One of the more promising models argues for the importance of hydrogen bonding acceptor groups in the substrate with a specific spatial orientation (Seelig, 1998). It must be noted that P-gp is only one of an emerging number of efflux transporters, including the multidrug resistance-associated proteins and breast cancer-resistance protein present in the intestinal mucosa and thought to influence drug permeability and bioavailability.

In summary, intestinal permeation of a solute is a complex process containing contributions from a number of pathways, the relative importance of which depend upon the nature of the solute. Each of these pathways will be dependent upon the structure of the solute, but in different ways. The complexity of this property will make it very challenging to develop simple, global, structure-based predictive models of intestinal permeability.

Solubility Factors

Both rate and extent of absorption can be significantly influenced by the intestinal solubility characteristics of the drug in the intestinal lumen. Frequently, these solubility characteristics may be optimized by the formulation scientist in ways that can have profound influences on the absorption of the drug in vivo. In the case of highly crystalline drugs, for example, aqueous solubility of the most stable crystal form generally is fairly low, resulting in poor absorption characteristics. Formation of higher energy, lower melting crystal forms or amorphous solids frequently yields more rapidly dissolving and higher solubility forms of the drug (Hörter and Dressman, 2001). The experimental human immunodeficiency virus-protease inhibitor PNU-103017, for example, has intrinsic solubility of about 1 to 3 µg/ml. Dosed as a conventional suspension to dogs, absorption was slow, giving very low blood levels. PNU-103017 is weakly acidic (pK_a values about 6 and 9) and dosing as a pH 10 solution gives significantly more rapid and extensive absorption. When an amorphous suspension of the drug was dosed, absorption characteristics were similar to the solution, consistent with much more rapid dissolution and higher solubility of the amorphous material compared with the crystalline solid (Wald et al., 1995).

Similar enhancements in absorption of high-permeability low-solubility drugs can be achieved through particle size reduction. The resulting increase in surface area enhances dissolution characteristics and can frequently result in significantly improved absorption. Danazol is a poorly water-soluble drug (10 µg/ml) that shows poor absorption in human. Administration of a conventional suspension of danazol to dogs results in similar poor absorption (C_max, 0.2 ± 0.06 µg/ml; 5.1 ± 1.9% bioavailability). Reducing particle size to an average of 85 nm (nanoparticle dispersion) profoundly increases absorption to give a C_max of 3.01 ± 0.8 µg/ml and bioavailability of 82.3 ± 10.1% (Liversidge and Cundy, 1995).

Ionizable drugs can exhibit significant differences in absorption properties dependent upon the pH characteristics of the intestinal environment. Cinnarizine is a vasodilator drug originally developed in Japan. It is a very insoluble drug, intrinsic solubility 15 ng/ml, with two basic groups with pK_a values of 1.94 and 7.47, respectively. Consequently, it is very soluble in acidic solutions. In the clinic, the absorption of cinnarizine was found to depend on the gastric acidity of patients. In individuals with high gastric acid content, cinnarizine rapidly dissolves giving good absorption characteristics. In those individuals showing low-gastric acid content, C_max and AUC were reduced by approximately 75 to 85% (Ogata et al., 1986).

Physiological and Environmental Influences on Absorption

We have already seen how intestinal pH variations can affect the permeability characteristics of ionizable drugs. A more complex environmental phenomenon potentially impacting drug absorption arises from the presence or absence of food in the gastrointestinal tract. The presence of food can increase, decrease, or have no effect on absorption, depending upon the characteristics of the drug and the food. Oltipraz, for example, is a highly lipid-soluble antischistosomal agent that is practically insoluble in aqueous medium. Administration of 500-mg tablets to fasting humans resulted in barely detectable blood levels (Ali et al., 1984). When administered in conjunction with either a low- or high-fat meal, the drug was rapidly absorbed and reached very high blood levels. The mechanism underlying this significantly increased absorption is not known but is speculated to arise from increased solubilization of the drug by bile acids secreted in response to the meal or effects on stomach emptying. Other drugs exhibiting such food-enhanced absorption include griseofulvin, hydralazine, and felodipine (Fleisher et al., 1999).

Drugs that exhibit decreased or delayed absorption when administered with food include alendronate, furosemide, ketoprofen, theophylline, and many others. Potential mechanisms contributing to these effects are thought to include possible sequestration of free drug with food components or bile acids, thus reducing the free concentration available for absorption (Fleisher et al., 1999). Food increases gastric emptying, which could help to improve absorption, but it also decreases small intestinal transit time that will decrease absorption for incompletely or poorly absorbed drugs.

Along with food, drug formulation components or excipients can affect small intestinal transit and influence absorption (Wilding, 2000). This is particularly pronounced for incompletely absorbed drugs for which extent of absorption is dependent upon the residence time of the drug in contact with the small intestinal epithelium. In the case of ranitidine formulated with sodium acid pyrophosphate (SAPP) as an effervescent tablet, absorption was significantly reduced in humans compared with ranitidine administered in the absence of SAPP. The rate of absorption was decreased and extent of absorption reduced by about 50% in the presence of SAPP, concomitant with a reduction in small intestinal transit time of 50% (Koch et al., 1993). Similarly, in a study with cimetidine, mannitol reduced small intestinal transit time by about 23%, accompanied by a decreased C_max and AUC_0-24 (Adkin et al., 1995). Polyethylene glycol 400 has shown similar effects on small intestinal transit and bioavailability of ranitidine in human studies.

Another potential consequence of increased transit has been proposed for digoxin and possibly other drugs which are substrates for P-glycoprotein in the small intestine (Siegmund, 2001). Since intestinal permeability of such compounds may depend upon the relative activity of the P-glycoprotein in the intestine, factors affecting this activity may also affect absorption. One determinant is drug concentration, which will influence the degree of saturation of the transporter. Another consideration is the specific activity of the transporter within the intestine itself. Evidence suggests that P-glycoprotein is not homogenously distributed throughout the intestinal tract but rather increases in abundance from proximal to distal small intestine (Fricker et al., 1996). Therefore, drugs that may be substrates for P-gp but are fairly permeable may be well absorbed in the duodenum and proximal jejunum, which has little P-glycoprotein. If absorption is shifted to more distal parts of the small intestine by decreased transit time, however, P-glycoprotein may play a more significant role in absorption. Although definitive proof for such a mechanism has not been presented yet, it seems attractive and would contribute to the concept of an "absorption window" favoring the proximal small intestine for certain drugs which are substrates for P-gp. On the other hand, for drugs that are primarily metabolized by CYP3A4, more rapid transit would move the drug into the distal small intestine where metabolic activity is decreased (Paine et al., 1997), potentially resulting in more intact drug absorbed than for the proximal situation.

A further confounding issue in relating absorption to structure is the possibility of saturable absorption or deviation from dose proportionality not caused by metabolism processes. Shown in Fig. 3 are blood levels achieved as a function of dose for three clinical candidates from the oxazolidinone antibiotic class. Compound I, Zyvox, was eventually approved for treatment of gram-positive infections. It is a relatively highly soluble (3.2 mg/ml) highly permeable (P_e = 24 × 10⁶ cm/s in rat ileum) drug and consequently shows good absorption and dose proportionality (D. H. Batts and D. J. Stalker, unpublished data). Compound II (Eperezolid) is similarly highly soluble (4.2 mg/ml) but has much lower permeability (P_e = 6.2 × 10⁶ cm/s in rat ileum) and also exhibits a linear dose-absorption response. Compound III, however, is much less soluble (0.4 mg/ml) as well as poorly permeable (P_e = 6.6 × 10⁶ cm/s in rat ileum). This situation leads to significant nonproportionality in absorption and decreasing fraction absorbed as the dose is increased. Clearly, for such a drug, the fraction dose absorbed depends upon the dose administered and is only very indirectly related to structure.

View larger version (17K): [in this window] [in a new window]   Fig. 3.   Dose-dependent oral absorption of oxazolidinones in clinical studies. Linezolid (compound I) has characteristics of high aqueous solubility and high permeability in both Caco-2 cell and perfused rat intestine and shows good proportionality as the dose is increased. Eperezolid (compound II) has similiarly high solubility but significantly lower permeability than I. It shows good absorption and linearity over the dosing range studied. In contrast, compound III is a drug with both low aqueous solubility and low permeability, resulting in poor absorption and substantially decreased fraction absorbed as the dose is escalated.

Variability and Bias in Human Absorption Literature and Impact on Structure-based Models

A general procedure for deriving structure-based models uses a training set of data permeability, fraction absorbed, etc., which is used to develop an empirical correlation with some descriptor-based representation of the drugs. The resulting model is then validated against a different collection of data and, if the validation is successful, the model is assumed to be predictive. Since human data are derived for drugs, which in general are reasonably well behaved in vivo (otherwise they would not be drugs), the models are biased toward such compounds. For example, human permeability measurements are a fairly recent development, originating from one or two laboratories with the capability for such determinations. Of the 20 to 30 drugs that have been measured, approximately 65% are high permeability (Winiwarter et al., 1998). High-permeability drugs are absorbed to the extent of 90% or greater in human (Amidon et al., 1995). Clearly, this results in pretty sparse coverage of the more poorly absorbed drugs, and hence, prediction for these compounds is suspect.

A second consideration in the modeling of permeability data is the error associated with the measurements. This can be substantial given the difficulty of conducting the experiments and the inherent variability for reasons discussed earlier in this article. Naproxen, for example, has a measured permeability of 8.4 ± 3.3 × 10⁴ cm/s, whereas piroxicam is reported to be 7.8 ± 7.5 × 10⁴ cm/s (Winiwarter et al., 1998). It is possible to develop highly correlative models using the mean permeability values. Given the variability of the data, however, the ability to confidently discriminate between even moderately dissimilar compounds may be difficult.

Similar considerations apply to the absorption data. Generally, what is reported in the literature is bioavailability. Given that bioavailability is a function of absorption and metabolism, however, attempts have been made to calculate fraction dose absorbed as the dependent parameter since this property is presumed to be more directly related to compound chemical and structural characteristics. In a recent and fairly comprehensive review of such literature, data for 241 drugs were described (Zhao et al., 2001). Criteria were presented for estimating fraction absorbed from the bioavailability, which were then used to support the estimations. These estimations are probably reasonable, again given the diversity of the data and methodology used to quantitate the in vivo results. Nevertheless, of the 241 drugs, 184 or 76% were estimated to be absorbed to the extent of 50% or greater. Of these, 108 (45% of the total) are greater than 90% absorbed. Clearly, these data are highly skewed toward well absorbed drugs, as is to be expected. From the perspective of being representative of structural and/or property diversity, however, these compounds are unlikely to adequately represent diversity of medically relevant chemistry space.

As in the situation with permeability data, a rather large variability also accompanies the human absorption data. In a recent article, Andrews et al. (2000) discussed variability in bioavailability values for a group of 282 compounds taken from literature reports and FDA files. Although the mean experimental variability in %F was 12, this was not uniformly distributed since the error increased with increasing %F. Expressed as a coefficient of variation, more highly bioavailable drugs have smaller coefficient of variation, and this increases significantly as bioavailability decreases (Yu et al., 2002). Again, as in the case for permeability, this is to be expected given the complex interrelationships among properties and processes contributing to bioavailability.

Concluding Comments

Drug absorption is a complex process and, although related to drug structure, it is related in fairly complex ways. Failure to appreciate this complexity in attempting to build models has at least two potential consequences. First, using questionable data in constructing a predictive model will result in predictions of low confidence (i.e., is the predicted FA for a compound 30% or could it really be 60%). Secondly, even if the 30% value is accepted, is that the best that can be achieved given the occasionally profound influences of formulation optimization on performance? An alternative approach to modeling absorption is to develop structure-based models for the properties contributing to the absorption process, such as solubility and permeability. These can then be used to identify opportunities for optimization. For example, if a potential drug is expected to have poor absorption as a consequence of low-intrinsic aqueous solubility, then this is a property amenable to manipulation by the formulation scientist. On the other hand, if the compound is both poorly permeable and soluble, along with a significant metabolic liability, optimization may be difficult if not impossible. Such candidates present high risks to successful development and should be identified as such early in the discovery/preclinical development process. Judicious development and use of computational models will clearly aid in these processes.