Predicting Drug Extraction in the Human Gut Wall: Assessing Contributions from Drug Metabolizing Enzymes and Transporter Proteins using Preclinical Models

Gut-wall metabolism has been studied in several animal models [96‐102]. By varying sites of drug administration (oral, intraportal and intravenous routes are common but intraperitoneal is also used) and PK sampling, the available fractions in liver and intestine can be calculated from comparison of AUCs under first-order conditions [103‐107]; however, comparison of intravenous and oral AUCs after sampling at one site is more straightforward and routinely applied [108‐110]. As such, similar issues described with the indirect approaches are to be expected. Despite differences in the CYP450 isoforms expressed in the rat compared with human (see the following paragraphs), a good correlation [root mean square error (RMSE) = 0.19] between rat and human F _G has been reported using a set of 11 CYP3A-metabolized compounds that had both intravenous and oral PK data [111]. Ten of these 11 compounds studied had human F _a of 0.8 or higher. As rat is a good model for human oral drug absorption, it follows that any differences in the intestinal availability should arise from intestinal efflux or metabolism. For BCS class I compounds, good solubility ensures sufficient concentrations for the saturation of efflux transporters. A high permeability rate ensures a low residence time within enterocytes, resulting in a rate of metabolic extraction that is limited by the rate of drug permeation through the enterocytes, rather than by the intrinsic ability of the intestinal DMEs. With permeability-limited intestinal extraction, any differences in enzyme isoform, abundance or activity have limited impact on the metabolic extraction, especially since CYP450s with broad substrate specificity abound in both human and rat, and their regional distributions are similar in both species. However, for hepatic extraction, lower hepatic concentrations compared with the gut lumen implies a greater role for transporters, while species differences in plasma protein binding and blood flow rate contribute to varying exposures to uptake and efflux transporters, as well as to DMEs, heightening the impact of differences in transporter and enzyme isoform, abundance and activities on hepatic extraction. This may explain why the rat is a good model for the prediction of human F _G but is a poor predictor of human oral bioavailability [112]. A strong correlation was also reported between cynomolgus monkey and human F _G for human CYP3A [113]. However, the cost and ethical concerns limit the availability of monkey PK studies. Animal models may be useful in the prediction of human F _G for CYP3A-metabolized compounds, but not much is known about their utility for other intestinal DMEs.

In situ approaches include the perfused gut loop, single-pass and recirculating intestinal perfusions, portal vein cannulations, portacaval shunts and transpositions [114‐121]. These techniques open the possibility of studying route-dependent intestinal metabolism following systemic and luminal drug presentation. This has been shown with acetaminophen, morphine and enalapril [96, 98, 106, 118, 122, 123]. Naturally-occurring tissue structure and physiology are retained, notwithstanding surgical manipulation and effect of anaesthetics [114, 124, 125]. Minimal interference of intestinal function and architecture means optimal tissue viability is maintained [106]. A full complement of endogenous DMEs and transporter proteins are present. Discrete segments of the small intestine can be evaluated to assess the impact of regional differences in DMEs or transporter expression and gut physiology. Intestinal metabolism can be more rigorously evaluated if drug concentrations are measured after sampling from mesenteric or portal veins [126]. Detailed methodologies are provided elsewhere [114, 127‐130]. In situ models have several advantages over in vivo models. For instance, bypassing the stomach means acidic compounds are unlikely to precipitate, and therefore dissolution rates do not confound intestinal drug concentrations and resultant plasma levels [94]. These models can be exploited to investigate metabolism–transporter interplay, as exemplified with midazolam, indinavir and UK-343,664 [48, 131, 132]. Various aspects need to be controlled [133], including luminal flow rate and thickness of the unstirred water layer (UWL), which can be rate-limiting for rapidly absorbed compounds [106]. In spite of their utility, isolating perfused organs from, or within, the laboratory animals typically requires specialized surgical procedures. As such, these approaches are arguably more labour-intensive, time-consuming and costly compared with other in vivo and in vitro approaches.

Whereas certain processes, such as passive diffusion, can be relatively well-predicted from animals [94, 106, 134‐139], pronounced species differences in expression of DME isoforms, substrate selectivity and abundance along the gastrointestinal tract [30, 31, 113, 140‐142] implies that human F _G prediction from preclinical species is not always feasible. The differences in relative protein expression of individual CYP450 isoforms in rat, dog, monkey and human small intestines are presented in Fig. 1. These species differences have led to substantial differences in apparent F _a × F _G as well as F _G [30, 143‐145]. Metabolism studies, using drug substrates for human CYP450s [141, 143, 146] and UGT enzymes [142, 147], have generally reported poor correlation between human and animal F _G. Certain DMEs appear to be selectively expressed in human intestines, including UGT1A8, UGT1A10 and SULT1A3 [140, 148, 149]. To the best of our knowledge, there is no known animal orthologue of SULT1A3 or UGT1A10 [147]. Additionally, there are known species differences for other enzyme classes expressed in the gut, such as the carboxylesterases [3, 4, 150, 151]. If a drug is shown to be a substrate for one of these enzymes then predicting human F _G using animal data would be questionable.

Species differences in regional enzymatic profiles exist within the intestine (Table 2). The regional decrease in CYP3A in humans, as a function of distance along the small intestine from the duodenum to the ileum [9, 85], is similar to the rat small intestine [152‐154]. Interestingly, in the rat intestine Cyp2b1 is highly expressed, whereas the equivalent isoform in humans (CYP2B6) is not expressed [9]. Rat Cyp2b1 is present at much higher protein levels in the upper parts of the intestine, whereas in humans the equivalent isoform (CYP2D6) is not [154]. The trend in the regional expression of rat Cyp2c isoforms (such as Cyp2c6) is opposite to that of Cyp3a, with higher expression towards the lower bowel [154] (Table 2). In contrast to being absent in the human small intestine, the extrahepatic enzyme Cyp1a1 has been reported by some authors to be the predominant CYP450 isoform expressed in rat small intestine, together with Cyp3a [9, 154‐156]. The glucuronidation/sulphation ratio can differ between regions according to individual species. In rat colon and proximal jejunum, the ratio was 16 and 23, respectively [157], showing a clear species difference in conjugation activity and regional difference compared with humans (see Sect. 3 and van de Kerkhof et al. [90]).

Regional expression of Mdr1b in rat intestine is similar to regional expression of MDR1 in the human small intestine and colon [74‐76, 78, 158] (Table 2). Contrary to this, canine intestinal Mdr1 expression is highest in the jejunum and very low in the lower parts of the ileum and colon [159]. For the MRP2 transporter, humans are similar to rat, with high expression in the proximal small intestine and very low levels in the lower bowel [74, 78, 83]. Abcc3 (Mrp3) is highly expressed in the colon of rats [160], and mice, at both the mRNA and protein levels [161] (Table 2). Interestingly, MRP3 (the corresponding transporter gene in humans) has the highest level of all the efflux proteins expressed along the intestinal tract [74]. In mouse, the expression of bcrp is higher in the mid jejunum compared with the rest of the intestinal tract [162]. A similar cellular location (basolateral in enterocytes) found in mice, rats and humans further supports a high degree of conservation for ABC transporters amongst eukaryotes.

In vivo rodent models with intestinal or hepatic enzyme gene knockdown or replacement have demonstrated the impact of intestinal versus hepatic elimination [163‐168]. These knockout (KO) and/or genetically modified (GM), transgenic (TG) mouse models have been established to create more reliable in vivo systems to study and predict human response to novel chemical entities (NCEs) [163, 169‐173]. For example, the importance of CYP3A metabolism in the intestine and liver has been illustrated with docetaxel in KO mice lacking all Cyp3a genes [Cyp3a (−/−)] [174]. When the CYP3A anticancer drug was administered intravenously to Cyp3a (−/−) mice, a sevenfold increase in systemic exposure was observed compared with wild-type. After oral dosing, an 18-fold higher systemic exposure was reported. Similar findings were reported for lopinavir and triazolam (CYP3A) [175, 176], debrisoquine (CYP2D6) [177] and tolbutamide (CYP2C9) [178]. This highlights the critical role intestinal CYP3A plays in human first-pass oral clearance and bioavailability.

Unfortunately, compensatory mechanisms arising from expression of host (murine) DMEs may confound data interpretation from KO models. For instance, clearance of midazolam was expected to be severely reduced in Cyp3a KO mice. In spite of this, metabolism was only marginally altered versus wild-type [179]. The revelation that several murine Cyp2c isoforms were significantly upregulated in Cyp3a KO mice, and could catalyze formation of midazolam 1′- and 4′-hydroxymidazolam, helped rationalize these results [179]. In contrast, no such effect was seen with triazolam, apparently a more selective Cyp3a substrate in mouse [176]. These compensatory mechanisms are likely to be drug- and species-dependent. In vitro studies characterizing background metabolism may prevent assessment of drugs susceptible to elevated host DMEs. Alternatively, exciting development of a viable mouse model in which all murine CYP450 genes have been deleted could avoid this issue altogether [180]. Interestingly, TG mice have been generated that are capable of expressing human CYP3A4 in the intestine and/or the liver on top of a mouse Cyp3a KO background [174]. As demonstrated with docetaxel and lopinavir, these humanized mice can provide mechanistic insight into the separate and combined roles of intestinal and hepatic human CYP3A, and the interplay with transporters such as P-gp in vivo [163, 175].

‘Bottom-up’ quantitative predictions that use in vitro data in conjunction with mathematical models describing processes within the gastrointestinal tract have been implemented [181‐185]. Several of the in vitro assays highlighted in Table 3 are simplistic in nature, take less time to complete, and are a fraction of the cost of animal experiments. This makes them eminently suited for screening NCEs and designing out presystemic metabolic liabilities. Models vary in complexity, from simple membrane preparations of individually recombinantly expressed enzymes [186‐190] to subcellular fractions [12, 43, 45, 85, 153, 191‐195], intact cells [196‐201] to tissue explants. The latter retain their in vivo tissue architecture, albeit lacking physiological surroundings. Although relatively labour-intensive and lower throughput (see Table 3), working with whole tissue preparations such as precision cut tissue slices (PCTS) [157, 202‐206], Ussing chamber [43, 69, 90, 93, 207], everted sac [126, 208‐212] and isolated tissue perfusions [94, 197, 213‐216] brings with it several advantages. The cell–cell contacts remain intact, all cell types are present and the DME systems, co-factors and transporters are available at physiologically relevant concentrations that more closely mimic the in vivo situation. With the isolated perfusion technique, variables such as temperature, pH, osmolality, blood pressure and flow can be controlled [217]. Metabolism on both sides of the intestine (luminal and vascular) can be studied. However, maintaining tissue viability is the major issue with this approach [94]. As such, its application is generally limited to animal tissue, and short-term incubations (approximately 2 h post-excision) impeding assessment of slowly metabolized drugs or enzyme induction [43]. This can be circumvented in situ but is not without complication. Loss of tissue viability caused by insufficient oxygenation is also problematic for everted sacs (1–2 h) and Ussing chambers preparations (2–4 h) [218‐220]. Nevertheless, everted sacs provide a fast and relatively inexpensive model for measuring regional differences in metabolism [106, 126]. The small volume inside the sacs offers analytical advantages over the isolated tissue perfusion and Ussing chamber, which due to sample dilution require sensitive bioanalysis to detect drug and/or metabolites [119, 126].

The Ussing chamber can be used with animal or human tissue providing a good model of drug absorption, transporter interactions, as well as metabolism, during passage across the gastrointestinal membrane [42, 69, 90, 219, 221‐223]. Drug can be added to luminal or serosal sides, allowing bidirectional transport and metabolism kinetics to be studied in different sections of the intestine [106]. Detailed mechanistic interpretation from studies such as these may require additional insight from experimentation with enzyme inhibitors, radiolabelled drug, or separate consideration of metabolism and permeability, e.g. using intestinal microsomes and Caco-2 monolayers. Indeed the extent of drug extraction in human intestine (E _g) has been successfully predicted for testosterone, midazolam and ropivacaine using human in vitro Ussing experiments [69, 93]. PCTS have received growing attention now that reproducible production of very thin slices (between 250 and 450 μm) is possible. These maintain better viability [203] and retain a high drug biotransformation capacity [43]. PCTS are also suitable for studying regional differences in intestinal metabolism, as well as regulation of enzymes and transporters involved in drug disposition [206, 224]. Applicable to all species, investigations have been reported in mouse [225], rat [157] and human [90]. Additionally, other tissues can be examined, allowing the extent of metabolism in different organs to be compared [204]. Potential drawbacks include poor penetration of highly metabolized drugs into the slices inner cell layers, lag time in phase II metabolism, and nonspecific binding to the slices [202, 226].

Intestinal subcellular fractions (S9 homogenates or microsomes) are one of the more established in vitro approaches used in drug discovery. More information has been published on the physiological scalars (Table 4) and several investigators have explored quantitative prediction of F _G [49, 111, 227‐231]. Commercial availability of animal and human samples, ease of storage and automation, make them an attractive option in terms of assay speed, capacity and cost. The fractions contain multiple enriched enzymes in their native configuration for assessment of phase I and selective phase II metabolism [45, 113, 195, 231, 232]; however, they do not contain a full complement of DMEs and lack potentially important interactions with uptake and efflux transporters. Incubations require the addition of expensive co-factors for optimal DME activity, often at higher nonphysiological concentrations. Others have suggested metabolic rates in S9 and microsomal fractions can be much lower compared with matrix such as PCTS [43]. This may be attributed to poor recovery of enzymes through suboptimal preparation procedures [195]. Presented in Table 5 are 6β-hydroxy testosterone rat data (normalized to units of pmol/min/mg rat intestinal protein) including intrinsic clearance (CL_int) from in-house intestinal microsomes prepared under optimal conditions [233, 234]. Interestingly, in-house microsomes achieved broadly similar rates compared with PCTS and biopsies [43, 205, 225], and were much higher than previous microsomal preparations [212, 235]. This highlights progress made with DME extraction procedures, which, until recently, have limited the scalability of intestinal microsomes [192, 195]. Given the profound effect enzyme extraction procedure and incubation conditions can have on enzyme activity, consensus is needed on best practise before we can expect significant improvements to the accuracy and reproducibility of predictions.

In vitro models allow the function of metabolic enzymes, transporters and absorption processes to be studied in the gut wall. However, to apply this data to retrospectively explain, or prospectively predict, human oral PK ultimately requires insight into the mechanisms influencing drug behaviour in vivo [132, 236‐238]. As such, several mechanistic approaches of varying complexity have been described for in vitro to in vivo extrapolation (IVIVE), some of which are available commercially, e.g. GastroPlus™ and Simcyp^® [17, 92, 132, 238‐241]. These mathematical translations (Fig. 2) can be relatively straightforward and ‘minimal’ models, such as Q _Gut, require only in vitro metabolic CL_int and cell permeability data to estimate F _G. Several groups have reported successful prediction of F _G in animals [242, 243] and humans [49, 227, 228, 244, 245] with this approach. Drugs with high in vivo extractions (F _G values <0.5) were less accurately predicted and may reflect inability of the Q _Gut model to account for changes in enterocyte drug concentration and therefore saturation of DME and efflux transporter processes [246]. However, to the authors’ knowledge, no critical assessment of possible systematic underprediction of IVIVE similar to that reported for hepatic metabolism [247‐249] is available in the literature. Improved F _G and oral clearance prediction was noted when the same set of drugs were evaluated using a physiological-based PK (PBPK) model. This was partly attributed to the model’s ability to account for saturation of intestinal metabolism by using maximum velocity (V _max) and Michaelis-Menten constant K _m, the substrate concentration at which the reaction rate is half of V _max rather than CL_int [246].

Sophisticated PBPK models have been published, such as the segmental segregated flow model (SSFM), which encompass all salient variables, e.g. absorption, gastrointestinal transit, metabolism, transport and efflux [95, 98]. This allows route-dependent intestinal metabolism and zonal distribution of DMEs and transporters to be considered, as has been exemplified with morphine [98]. This in turn provides insight into the likely interplay between luminal transit, metabolism and active transport [250], and can be used in experimental design and to explore ‘what if’ scenarios [232, 236, 237, 251]. Comprehensive review of these dynamic, integrated modelling approaches have been provided elsewhere [238, 240, 252, 253]. Often, disadvantages perceived with these models relate to their inherent complexity and the level of detail required on parameters used in predictions. For example, transporter kinetic data (K _m and V _max) are often lacking, and abundance data for individual DMEs and transporters for the in vitro models and intact tissues are limited and hence restrictive to IVIVE [16, 246]. In response, proteomic- and mass spectrometry-based methods for protein quantification and establishment of scaling factors have started to emerge [32, 149, 254‐256].