Application of In Silico, In Vitro and Preclinical ... - ACS Publications


Application of In Silico, In Vitro and Preclinical...

0 downloads 122 Views 881KB Size

Article pubs.acs.org/molecularpharmaceutics

Application of In Silico, In Vitro and Preclinical Pharmacokinetic Data for the Effective and Efficient Prediction of Human Pharmacokinetics Kenneth H. Grime,* Patrick Barton, and Dermot F. McGinnity Respiratory & Inflammation DMPK, AstraZeneca R&D, Mölndal, SE 43183 Mölndal, Sweden ABSTRACT: In the present age of pharmaceutical research and development, focused delivery of decision making data is more imperative than ever before. Resulting from several years’ success, failure and consequential learning, this article also proffers advice and guidance on which in vitro and in vivo experiments to perform to facilitate efficient and cost-effective pursuit of candidate drugs with acceptable human pharmacokinetic profiles. Predictive in silico models are important in directing design toward compounds with the highest probability of having suitable DMPK properties rather than in predicting human pharmacokinetics per se, and the value and utility of such approaches are reviewed with the intention of providing direction to DMPK scientists. Relating to absorption, distribution, elimination and effective half-life, strategies are described to provide direction in commonly encountered scenarios. KEYWORDS: Drug Metabolism and Pharmacokinetics, Drug Discovery, in silico, in vitro−in vivo extrapolation

1. INTRODUCTION

be performed and conclusions drawn about the possibilities of drug therapy for a given therapeutic agent.15 In recent years the pharmaceutical industry has struggled with many internal and external pressures such as inaccessibility of druggable targets and changes in societal, economic and regulatory environments that are all ultimately linked to rising costs. These have played a part in precipitating strategic changes in R&D organizations while the advances in knowledge and understanding that have been gained within these internal R&D organizations have not yet translated into decreased attrition.4,16−18 In a desire to tackle some of the industry’s challenges and seek pharmacological and biological innovation, larger pharmaceutical companies have turned to collaborations with smaller biotechnology organizations or academic laboratories. Similarly externalization of synthetic chemistry and biological screening to contract research organizations (CROs) has been sought to decrease costs and increase flexibility.19,20 However, the skills and knowledge gained from years of learning in such functions as DMPK and medicinal chemistry more so than ever need to be retained and applied in the new delivery model. Moreover, large Pharma DMPK departments that have focused on process and efficient in vitro and in vivo data delivery at the expense of in cerebro comprehension, interpretation and influence may be competitively disadvantaged. Thus a great opportunity is presented: as centralized screening functions and CROs are used to provide much of the

High attrition rates in drug discovery and development coupled with rising costs have been the single biggest challenge to the pharmaceutical industry in the recent past. Suboptimal drug pharmacokinetics (PK) was identified as a major contributor to the failure of potential medicines some twenty-five years ago,1 but more recent reassessment suggests that the primary reasons for drug attrition have consistently been lack of efficacy and toxicology/clinical safety risk more so than suboptimal PK properties per se.2,3 Nevertheless, such evaluations coupled with the molecular biology revolution, leading to target based drug discovery and increased medicinal chemistry output, precipitated the realignment of Drug Metabolism and Pharmacokinetics (DMPK) departments within Drug Discovery and challenged industry scientists to develop enhanced throughput assays and an increased ability to more accurately predict and optimize clinical PK properties.4−7 Thus, the DMPK transformation that started in the 1990s and the comprehension that has emerged, mirrored by advancement in detailed knowledge of the physicochemical profiles of oral drugs and the impact of such properties on DMPK and interlinked safety aspects,8−14 should have a significant positive impact for pharmaceutical research. Indeed, after selection of a relevant drug target, maximizing the likelihood of a candidate drug becoming a successful therapy is achieved by combining adequate potency against the target protein and an acceptable safety profile with a balance of optimized PK parameters and minimized drug−drug interaction (DDI) potential. Innovation and progress in aligned areas are similarly important in the overall goal to create more successful outcomes in pharmaceutical development. For example, an integrated pharmacokinetic−pharmacodynamic (PKPD) approach is fundamental if apposite studies are to © 2012 American Chemical Society

Special Issue: Predictive DMPK: In Silico ADME Predictions in Drug Discovery Received: Revised: Accepted: Published: 1191

August 28, 2012 November 13, 2012 December 19, 2012 December 19, 2012 dx.doi.org/10.1021/mp300476z | Mol. Pharmaceutics 2013, 10, 1191−1206

Molecular Pharmaceutics

Article

Figure 1. Relationship between once daily oral dose (mg/kg/day), clearance (CL, mL/min/kg) and steady state volume of distribution (Vss, L/kg) for a drug that is 90% bound to plasma proteins and whose elimination half-life has been altered in the range 2.5 to 25 h through changing CL or Vss. In scenario A, primary pharmacological target potency has been fixed at 3 nM and CL has been fixed at 4.5 mL/min/kg. In scenario B, primary pharmacological target potency has been fixed at 3 nM and Vss has been fixed at 1 L/kg. In scenario C, CL and Vss have been fixed at 0.9 mL/min/kg and 1 L/kg (elimination half-life of 13 h) and potency has been changed from 0.03 to 30 nM.

constant (=CL/Vss, where CL is clearance). Although F is an observed pharmacokinetic parameter, it is a function of the fraction of dose absorbed (Fabs), the fraction of drug escaping intestinal metabolism (Fg) and the fraction of the drug escaping hepatic CL (Fh) such that F = FabsFgFh. The absorption rate constant is commonly rapid enough not to significantly impact the effective T1/2 (authors' unpublished observation based on an analysis of marketed oral drugs). Thus it is clear from the human dose equation that there are only four parameters available to optimize: potency against the target protein, CL, Vss and Fabs. Alteration of one of these parameters while fixing the others at a single set value demonstrates that potency and Fabs have a linear impact on dose while CL and Vss can have a much greater impact (Figure 1). For any given target, in the absence of a more thorough understanding of the human PK−PD relationship which will only be established in the clinic, this equation can set the foundation for the drug discovery optimization strategy. There are many DMPK experiments that can be performed during the drug discovery process, but the aim of the following sections is to give experiential based guidance on which are key and how to interpret the data with respect to optimizing new chemical entities (NCEs) toward drugs with acceptable DMPK properties. This should allow streamlined and focused delivery supported by decisions that steer the course for oral drug discovery projects and avoid confusion.

data from the now well established assays, DMPK scientists within Pharma R&D can focus on maximizing value to their portfolio through applying knowledge gained and focused innovation. The additional use of in silico modeling together with the use of automated model building methodologies offers the DMPK scientist even greater ability to input into the design of new compounds, bringing unique insight for rapid project progress. Against such a backdrop, the purpose of this manuscript is to provide advice and guidance with the intention of facilitating the efficient and cost-effective pursuit of candidate drugs with acceptable pharmacokinetic profiles through focusing on the right experiments and predictions in a field where so many experiments can be performed without actually yielding any decision making data. This article aims to offer experience based strategic insight and tactical solutions for specific issues that are routinely faced by the Drug Discovery scientist.

2. OPTIMIZING PHARMACOKINETICS IN DRUG DISCOVERY The prediction and optimization of human pharmacokinetics, efficacious clinical dose and associated exposure is a central component of a “drug hunting” strategy. In parallel, sufficient exposure in relevant preclinical species is required to allow suitable safety testing as well as the development of pharmacodynamic disease models. For oral drug delivery, a simplistic one compartment PK model can be derived if it assumed that the target plasma (unbound drug) concentration at steady state is directly related to efficacy and that the pharmacokinetic elimination half-life (T1/2) is the effective halflife (the phase of the drug concentration−time profile responsible for maintaining drug concentrations above the required level to achieve efficacy). With such assumptions, the following equation can be derived: dose (mg/kg/day) =

3. ABSORPTION As outlined in section 2, for orally administered drugs, it is important to characterize the oral bioavailability (F) of a compound during drug discovery but also to understand the relevance of the breakdown into first pass hepatic and intestinal CL and absorption. What is acceptable in terms of preclinical and clinical absorption is a question that in our experience can confound drug discovery teams and lead to time and effort being unnecessarily wasted. Fundamentally what is required is to allow sufficient exposure of a new chemical entity (NCE) for appropriate preclinical safety testing and to avoid excessive variability in systemic drug levels in patients. As interpatient variability in bioavailability is related to its extent,21 absolute bioavailability in human of greater than 30% is an appropriate target, and it is therefore judicious to target a minimum predicted human absorption of at least 50%, as F is a product of the fraction of absorbed drug, the fraction escaping intestinal

(24/T ) ·MEC·(ka − kel) 1 1 ⎤ − Fka⎡⎣ 1 − e−kelT 1 − e−kaT ⎦

In this the dose interval is T hours such that the plasma concentration of the drug required for efficacy will always be greater than or equal to the minimum effective concentration (MEC) and the number of doses given per day is 24/ T. The steady state volume of distribution is Vss, F is bioavailability, ka is the absorption rate constant and kel is the elimination rate 1192

dx.doi.org/10.1021/mp300476z | Mol. Pharmaceutics 2013, 10, 1191−1206

Molecular Pharmaceutics

Article

dependent efflux is more likely.38 In the absence of more sophisticated simulations incorporating solubility and dissolution rate measures, the intestinal drug concentration range sufficient for interacting with P-gp may be estimated from the maximum dose taken/250 mL (intestinal fluid volume) or, alternatively, the maximum concentration in the enterocyte can be estimated from (Fabska × dose/Qent) where Qent is enterocyte blood flow.29,39 It should be noted that this Qgut equation generally gives 100-fold lower estimations of concentration than using the intestinal fluid volume approach such that a 1 mg/kg oral dose may be differentially described as having a relevant intestinal concentration of approximately 6 or 600 μM.24 Caution should also be used when using Km values from in vitro assays such as the Caco-2 assay since they may overestimate the in vivo situation and are sensitive to the expression level of P-gp.38 Thus, while for compounds with moderate to low permeability an assessment of the role of P-gp can be made using the Qgut equation and P-gp Km estimated from basolateral to apical Caco-2 drug concentration data, it is advisable when in a “risk zone” to use simulation software, such as Gastroplus that has been validated with known clinical data for converting in vitro experimental values (from the particular laboratory where the novel data is being generated) to those that can be used for in silico prediction. Although when viewed from a certain level, variables such as transit time through the gastrointestinal tract and dissolution from a tablet may influence the extent of oral absorption, targeting compounds with appropriate physicochemical properties that ensure sufficient solid crystalline solubility and transmembrane permeability should maximize the oral absorptive potential. The apparent permeability measure (Papp) is typically obtained from assessment of drug flux across a monolayer of cells intended to mimic the intestinal barrier. The most commonly used are the Caco-2 and MDCK cell lines.29,40 Relationships between human absorption and in vitro Papp are required to put the in vitro data into context, but Papp data must first be transformed into effective permeability (Peff) which describes intestinal permeability per unit surface area. Effective permeability can be determined experimentally using perfused intestinal segments and then correlated to Caco-2 or MDCK Papp.29,41 It is important to contextualize both the key variables (measured in vitro permeability and solid crystalline (not amorphous) solubility data), and pharmacokinetic modeling/ simulation tools such as Gastroplus or Simcyp can be used to make human absorption predictions with these inputs.38,42,43 To facilitate rapid and effective decision making early in the life of a NCE (Table 1), our laboratory has used such modeling tools to a generate solubility−permeability heat map (Figure 2). Twenty-eight oral drugs with in-house measured aqueous crystalline solubility and Caco-2 Papp values were used as a basis for this, and an analysis of the predicted versus observed human absorption allowed the guidelines to be drawn that permit better early decisions than simply using a Papp to fraction absorbed plot. For example, from Figure 2 a crystalline solubility measure of 100 μM and a Caco-2 Papp of 5 × 10−6 cm/s will most likely result of a dose fraction absorbed in human of greater than 50%, with lower values on either parameter putting the NCE in a risk category for lower human Fabs. Figure 2 illustrates how to effectively use in vitro data in a powerful way to make robust decisions early in drug discovery. More data such as estimated absorption in preclinical species following pharmacokinetic experiments will emerge to support or question these early decisions (Table 1), and as the

metabolism and liver extraction as it passes from the portal vein to the systemic circulation (F = FabsFgFh).22 Of course the 30% target for F should not be viewed as a hard cutoff, since several marketed oral drugs sit in this category, but simply as an area where intersubject variability may hamper drug development. An assessment of the role of intestinal drug extraction is a component of the human bioavailability prediction. The human and preclinical animal intestinal drug metabolizing enzymes are well characterized with CYP3A, CYP2C9 and UGT dominating.23−28 Mathematical models enabling in vitro data to be used in the prediction of Fg have also been described.29 Despite intestinal CYP content being extremely low compared to that of the liver30,31 and intestinal intrinsic CL values being similar to hepatic once corrected for expression levels,32 extraction by the gut can in some cases be similar to or exceed hepatic extraction.29,33 The reasons for this include positioning of the drug metabolizing enzymes and the P-glycoprotein (P-gp) drug efflux transporter (which often shares substrate specificity with CYP3A4) in the villus tip of the enterocytes facilitating cycling of drugs and prolonged exposure and facilitating intestinal metabolism.29,30,34 Significant intestinal extraction is commonly associated with relatively metabolically unstable drugs,29,35 so for a drug to be efficiently extracted by intestinal drug metabolizing enzymes, it would need not only to have sufficient exposure to those enzymes in the intestine but also to be relatively rapidly metabolized. However, oral drug discovery programs typically optimize toward compounds with moderate to high permeability and solubility, high intrinsic metabolic stability and low involvement of intestinal efflux transporters. Such NCEs are unlikely to carry a significant risk of intestinal metabolic extraction. Nevertheless, consideration should be made for substrates of UGT and CYP3A that do not fall into the categories described above, but observations such as low rat oral bioavailability should not in isolation lead one down such a path of investigation (see below). For compounds with moderate permeability or higher and good solubility (see Figure 2), the intestinal absorption is unlikely to be limited by intestinal drug efflux transporters such as P-gp.36,37 However, if a drug falls outside this category and the Km describing the affinity of the substrate−transporter interaction is relatively high (tens of micromolar or above) and/or the dose is considerably less than 1 mg/kg, in vivo P-gp

Figure 2. A “heat map” for providing rapid prediction of human absorption from solid crystalline solubility and in vitro permeability data. Twenty-eight oral drugs with measured solubility and permeability data and known human absorption were used as a basis for prediction tool. 1193

dx.doi.org/10.1021/mp300476z | Mol. Pharmaceutics 2013, 10, 1191−1206

Molecular Pharmaceutics

Article

Table 1. A Suggested Generic DMPK Screening Cascade, with Screening Rounds 1−4 assays/experiments LogD7.4 solubility rat hepatocyte CLint human liver microsomal (HLM) CLint human plasma protein binding (PPB)

human hepatocyte CLint and PPB (rat) Caco-2 AB Papp (pH 6.5/7.4) rat PK

dog PK iv/po with urine collection reversible CYP inhibition time dependent CYP inhibition assay biliary clearance assessment DDI assays (enzyme identification, drug transporter inhibition assays) Caco-2 AB/BA efflux and drill down reactive metabolite assays (e.g., cyanide and reduced glutathione trapping following metabolic activation)

decisions/comments Screening Round 1 Is the compound in the correct property space with respect to CLint, PPB, permeability, CYP inhibition? with measured in vitro Papp (or predicted once correlation of observed data with relevant physicochemical properties established) human absorption can be estimated rat hepatic metabolic clearance can be predicted with assumed blood/plasma ratio and PPBa initial estimate of human hepatic metabolic clearance with PPBb allows prediction of human clearance and calculation of relevant potency on which to base human dose prediction Screening Round 2 estimate of human hepatic metabolic clearance with assumed or measured blood/plasma ratio and PPBc for predicting human absorptiond can in vivo CLint be predicted accurately to within 2-foldgood in vitro−in vivo extrapolation (IVIVE)?e Screening Round 3 confirm two species IVIVE, human absorption and renal CL prediction drug−drug interaction (DDI) predictionf DDI prediction Screening Round 4 considered as candidate drug profiling assays

a

For a streamlined screening approach, assaying for human plasma protein binding (PPB) only in round 1 is recommended. PPB is often predictable to a degree of accuracy (particularly when one knows the value of human PPB) that allows acceptable decisions on whether to go to rat PK experiment or not in round 2rat clearance predictions can be made using estimations of rat PPB from human PPB or logD based predictions, and in round 2 when rat PPB is measured an accurate understanding of whether rat clearance is well predicted or not is gained. bHuman liver microsomal (HLM) CLint has a larger dynamic range than human hepatocyte CLint due to the scaling factors (e.g., 120 million hepatocytes/g of liver compared to 45 mg of microsomal protein/g of liver60) and offers an understanding of the oxidative metabolism liability for the compound in the absence of complications such as membrane permeability and active transport considerations associated with hepatocytes. In the authors' experience, obtaining HLM and human hepatocyte CLint values has on many occasions proved valuable in identifying issues with the latter that require investigation. Obviously if phase II routes of metabolism (directly on the parent molecule) are dominant, then HLM CLint may be of limited use. cFrom Round 1 (including target potency data) an early prediction of human PK and dose can be made from just five DMPK assays and by the end of Round 2 more confidence is gained on the validity of the predictions. dEarly in a project it may be necessary to obtain this data in round 1, but since apical to basolateral Caco-2 permeability data is highly predictable from such parameters as logD, it can quickly be pushed further down the screening cascade to round 2 or 3. It is not necessary to generate Caco-2 Papp efflux ratio data early in the screening cascade unless there is a genuine concern over lack of penetration to the central nervous system. eEarly in a project’s life it is necessary to obtain iv rat clearance data to understand if there is a good IVIVE. Oral rat absorption data is also useful early in a project’s life to understand if there will be genuine concerns over rat bioavailability limiting safety or PD studies. However, once IVIVE understanding is established, obtaining rat iv data can be relegated down the screening cascade and oral data need only be generated in planning for PD studies. fFor a given chemical series an understanding of how this relates to logD7.4 can be gained very early, and therefore screening in round 3 is appropriate once an understanding of whether the compound is likely to have acceptable human PK properties has been achieved.

(Figure 3, adapted from ref 9). The authors of this manuscript are aware that other researchers have published data indicating a slightly different understanding of how rat and dog absorption relate to the clinical situation and the arguments used to support these assertions. This has been discussed previously9 and we reiterate here that in our experience over many years that for drugs absorbed by the trans-cellular route, not only does dog pharmacokinetic data indicate a systematically greater absorption than in rat, but also that dog absorption better represents the clinical situation. The synopsis is therefore that in vitro crystalline solubility and Papp data allow an early estimation of human absorption which can then be supported by in vivo dog absorption data. Rat oral absorption data should be treated with caution as our experience suggests this may be an underestimate of absorption in higher species including human and poor oral absorption in the rat should not per se preclude progression of compounds

compound progresses more effort can be put into understanding the biopharmaceutical properties of the drug as required.38,43 Rat oral absorption is, in the main, not a good predictor of human absorption.9 This is most likely explained by the fact that, once normalized for body surface area, the rat small intestine has a four times lower surface area compared to human.44 Nonetheless, the rat absorption data is important for determining if safety assessment studies can be adequately performed. It is therefore necessary to identify whether it is possible to achieve high enough exposures in the rat to allow suitable margins over the predicted human exposure to be attained. The human exposure can be predicted using the available data on the compound or using the generic target profile for the proposed drug. Predicting human absorption from preclinical in vivo data is more achievable using dog as a model, since the dog to human absorption correlation appears strong when absorption is not by the paracellular route oral 1194

dx.doi.org/10.1021/mp300476z | Mol. Pharmaceutics 2013, 10, 1191−1206

Molecular Pharmaceutics

Article

Figure 3. Relationship between human and dog oral absorption for eighteen marketed drugs with molecular weight ranging from 325 to 646. Reprinted with permission from ref 9. Copyright 2007 Bentham Science Publishers.

Figure 4. Relationship between human steady state volume of distribution (Vss) and rat, dog and mouse Vss corrected for human/ preclinical animal differences in plasma protein binding. Red spots are acidic drugs, blue are bases, yellow are neutrals and green are zwitterionic drugs. Adapted from ref 9.

into development assuming the requisite oral exposure margins in safety studies are achievable.

Basic drugs tend to have similar plasma protein binding to neutral drugs, in contrast to acids which show higher binding even for a given logD7.4.11 Being positively charged at physiological pH, bases have favorable interactions with acidic phospholipid head groups leading to higher tissue affinity, and therefore for the same plasma protein binding as a neutral compound, a base will tend to have a higher tissue distribution.48 Ion trapping in such subcellular acidic organelles as lysosomes also has a large effect, particularly in lysosome rich organs such as the liver and lung.49 Even a weakly basic drug may have an apparent volume of distribution three times the physical blood volume, and dibasic drugs can have volume of distribution values greatly in excess of its monobasic analogues.11,50 In summary, Vss is reasonably predictable from in vitro data11,51 and physicochemical properties.52,53 Once preclinical in vivo PK data is available, it can be considered the most robust and predictable of PK parameters with the Oie and Tozer method superior for making human predictions, as demonstrated by several laboratories.54−56 Awareness of this can be profitable when forming human Drug Discovery strategies. One confounding factor can be enterohepatic recirculation, following biliary elimination of parent drug or of conjugates of the parent drug that are subsequently hydrolyzed back to the parent in the intestine. Absorption of such biliary eliminated drug material leads to an extended terminal half-life and a larger than expected volume of distribution. Clearance is relatively unaffected as a calculated parameter from the PK profile, but caution needs to be applied when using non-bile duct canulated animal (rat and dog for example) data in making predictions of human Vss, since the same efficiency of the many and various processes involved may not translate equally well across the species and, of course, with so many processes involved, the terminal phase in human is likely to result in high intersubject variability.

4. VOLUME OF DISTRIBUTION The volume of distribution (V) is governed by the equation V = [fup/fuT] × VT + VP, where fup is the unbound drug fraction in the plasma; fuT is the unbound drug fraction in the tissue; VT is the volume of the tissue and VP is the volume of the plasma. As VT is usually considerably greater than VP, changes in fup will directly affect V.22 In essence, interspecies differences in tissue binding are assumed to be minimal and therefore differences in plasma protein binding have the biggest influence on disparity in steady state volume of distribution (Vss) across species.9,45 Considering Vss,u to be conserved across preclinical species provides a realistic prediction of human Vss9 and is significantly superior to allometric scaling techniques.46,47 For different chemotypes one should expect characteristically low (acids), moderate to low (neutrals) or moderate to high (bases) steady state volumes of distribution,9 where the low, moderate and high labels can be assigned as less than 1 L/kg, 1−3 L/kg and greater than 3 L/kg. In a Drug Discovery setting, knowledge of the expected boundaries for a given chemical class can be translated into a strategy for obtaining the necessary human elimination T1/2 through an understanding of the extent to which clearance will need to be reduced and what parameters are available in order to make such a change. The overriding influence on the distribution of acidic drugs is that of extensive binding to plasma albumin and low tissue affinity due to unfavorable charge interactions with negatively charged phospholipids of tissue membranes.48 Apparent distribution volumes thus approach that of albumin, approximately 0.1 L/kg,22 and in our experience do not exceed 0.3 L/ kg unless active hepatic uptake is a determining factor. As such, an acidic NCE of interest with a Vss greater than this value merits closer attention, starting with scrutiny of the PK profile itself. The volume of distribution for neutral drugs is governed by hydrophobic interactions with plasma proteins and tissue membranes. Increasing lipophilicity raises tissue affinity but has the opposite effect of restricting tissue distribution due to an increase in plasma protein binding. Consequently, distribution volume tends to be confined to the range 1 to 3 L/kg for a high proportion of neutral compounds (Figure 4, adapted from ref 9).

5. CLEARANCE Optimization of clearance (CL) is typically one of the more significant challenges for a drug discovery project. Identification of the elimination route and rate in preclinical species and optimization in human are major goals in most projects for oral drug therapy. The major drug elimination routes in humans and preclinical species are metabolism, renal and biliary. As there is 1195

dx.doi.org/10.1021/mp300476z | Mol. Pharmaceutics 2013, 10, 1191−1206

Molecular Pharmaceutics

Article

currently no reliable way to predict human elimination pathways from purely in silico or in vitro methods, a combination of establishing CL routes in preclinical species and use of in vitro human tools is required to predict human CL. Until relatively recently it has been common practice to predict human CL by cross-species allometric scaling irrespective of the elimination route.57−59 However, for hepatic metabolic CL, it is now widely accepted that the use of in vitro data (in vitro−in vivo extrapolation, IVIVE) can and should be relied on to make accurate predictions.10,60−62,64 This science came to the fore almost 20 years ago following the seminal publication of an IVIVE strategy showing how existing models for describing hepatic metabolic CL (CLH), incorporating terms for liver blood flow (Qh), hepatic intrinsic clearance (CLint) and blood binding (fub) could be used.63 Over the subsequent years, a dramatic increase in studies and awareness in this area led to gains in understanding and refinement.10,60−62,64 Isolated hepatocytes are regarded as the most useful in vitro system for predictive studies since they contain the full complement of enzymes a compound is likely to encounter during first pass metabolism and transporter proteins, which can be key determinants of hepatic CL10,65,66 and should therefore form the basis of IVIVE for CL. Perhaps the most discussed and controversial parts in the general understanding have been the inclusion, or not, of in vivo and in vitro drug binding terms and the fact that often the direct use of standard biological scaling factors to scale in vitro intrinsic CL yields a systematic underprediction of in vivo CLint. Detailed reviews have already been published on this subject, and it is now established that incorporation of drug binding terms is pivotal in making successful predictions.10,60,62,67,68 Figures 5

Figure 6. Relationship between in vivo and scaled in vitro human hepatic metabolic intrinsic clearance (CLint) when all drug binding terms are included. In vivo CLint was calculated from observed blood clearance (CLH) after removal of nonhepatic metabolic clearance and deconvoluted from the Well Stirred Model: CLint = [CLH/(fub × (1 − CLH/Q))], where Q is hepatic blood flow and fub is the drug fraction unbound in blood. In vitro CLint was calculated from human hepatocyte CLint scaled up to whole organ clearance using scaling factors60 and divided by measured fuinc (drug fraction unbound in the in vitro incubation). Red spots are acidic drugs, blue are bases, yellow are neutrals and green are zwitterionic drugs.

much more detailed understanding of predictive accuracy.10,62,69 Hepatocyte CLint is measured and used with the incubational binding term, fuinc (fraction unbound in the incubation), to calculate unbound in vitro CLint which is calculated up to the whole liver unbound CLint using scaled scaling factors.63 Unbound in vivo CLint is back calculated from in vivo hepatic metabolic CL (using total CL and subtracting nonhepatic metabolic CL values) adjusted for in vivo blood binding (using plasma protein binding and the relevant blood to plasma distribution coefficients). The unbound in vitro and in vivo CLint values form a line of prediction from which future predictions of unbound in vivo CLint values for NCEs can be made once the in vitro CLint is determined.10,60,62,67 A high degree of confidence in the human metabolic CLH prediction can be gained if the unbound in vivo CLint has been accurately predicted to within 2-fold in two preclinical species (typically rat and dog). Interestingly a recent publication has reconsidered how drug binding in vitro and in vivo can affect clearance predictions, with the group involved following up questions they had over how well clearance can be predicted for drugs highly bound to plasma proteins with some detailed thinking and investigation.70 However, in our laboratory, significant research effort has focused on acidic drug chemistry over the past decade, and our findings are that highly bound drugs do not suffer any more bias or uncertainty in prediction than less highly bound drugs. Other researchers also indicate this to be their observation,71 and our conclusion is that the “regression corrected method” detailed above and previously10,60 gives a robust and effective prediction of human hepatic metabolic clearance across all drug classes and, as we state above, an unbound in vivo/in vitro CLint ratio in rat and dog of less than 2 gives confidence in human predictions. There are many possible reasons for the offset between in vitro and in vivo CLint giving rise to the need for the “regression corrected method”, as detailed previously,10 and it is the belief of the authors that the fact that the method appears empirical is not important, since it works consistently. This is demonstrated not

Figure 5. Relationship between in vivo and scaled in vitro human hepatic metabolic intrinsic clearance (CLint) when all drug binding terms are ignored. In vivo CLint was calculated from observed blood clearance (CLH) after removal of nonhepatic metabolic clearance and deconvoluted from the Well Stirred Model: CLint = [CLH/(1 − CLH/ Q)], where Q is hepatic blood flow. In vitro CLint was calculated from human hepatocyte CLint scaled up to whole organ clearance using scaling factors.60 Red spots are acidic drugs, blue are bases, yellow are neutrals and green are zwitterionic drugs.

and 6 demonstrate the impact of the inclusion of drug binding terms. The prediction method for in vivo hepatic metabolic CL involves initially building in vivo−in vitro CLint models for each species using test sets of drugs (acid, basic and neutral) with known in vivo hepatic metabolic CL values. Prediction of unbound in vivo CLint rather than CL affords the investigator a 1196

dx.doi.org/10.1021/mp300476z | Mol. Pharmaceutics 2013, 10, 1191−1206

Molecular Pharmaceutics

Article

only in training sets of drugs60 but also for test drugs from the author’s laboratory. There is of course the possibility that there will be compound to compound differences that may alter the unbound in vivo/in vitro CLint ratio or offset, but that is the reason for the strictness around the rat and dog in vivo/in vitro CLint ratios. One really key point should not be overlooked: considerable rigor and attention to detail is required to ensure that as much understanding of the in vitro data and its limitations and inconsistencies from the in vivo situation are accounted for. In the authors’ view it is misguided to simply claim that IVIVE “does not work” based on initial data, and it is our experience that the answers are often simple artifacts of in vitro and in vivo test conditions (see below) and in fact considerable effort can be wasted searching for complex solutions when the basics are overlooked. The lack of an IVIVE in rat and/or dog for a NCE or chemical series can lead directly to an experimental artifact or an interesting biological phenomenon being identified and the relevance to the human situation established. Even when unbound in vivo hepatic intrinsic CL (and therefore hepatic CL) is well predicted in two preclinical species, in vitro assay details that may be humanspecific should be considered. Experimental artifacts are plentiful, and examples come across in the authors' laboratory over the past few years include the following: the substrate specific inhibitory effects of different solvents;7,10 in vitro substrate concentrations being inappropriate for determination of in vivo CLint;10,72 inconsistent maintenance of buffer pH during incubations (which can occur when cell culture media buffered with bicarbonate rather than a more artificial buffer such as HEPES are used outside a CO2 incubator or indeed vice versa10); active hepatic uptake being the rate limiting step in the clearance process, effectively resulting in one-way “distribution” from the blood into the liver73 and nonlinearity of plasma protein binding with respect to drug concentration. These have all been factors in IVIVE “not working” in our laboratory, but concerted scientific investigation and attention to detail has yielded information to aid understanding and allow confident predictions. The process of investigation for poor IVIVE should begin with the clearance prediction equation, typically the Well Stirred Model.62,63 Inspection of this shows that the in vivo PK profile itself is a reasonable starting point. For example, knowledge of the expected compound concentrations for a particular chemotype can indicate if the plasma concentrations may have been misquantified or indeed if the compound concentrations in the plasma very early after dosing have not been accounted for in the sampling regimen. Once satisfied with the quality and relevance of existing data, investigation of the in vitro measured terms should follow: blood to plasma ratio, plasma protein binding, CLint and fuinc. Incubational binding is often predicted from simple lipophilicity terms,74−76 but when in vivo CLint is not well predicted, incubational binding can be easily measured. This typically involves equilibrium dialysis analogous to plasma protein binding experiments74 but for hepatocytes can also be performed simply by an incubation period short enough to avoid significant metabolism of parent drug followed by a low speed (100g) centrifugation step. Sampling of the buffer and comparison to the starting concentration facilitates a rapid and informative fuinc analysis. Moreover, an estimation of the unbound drug concentration available for metabolism in the actual CLint experiment is facilitated by determining CLint at different hepatocyte or microsomal protein concentrations.10 While ultimately the most relevant measure, this technique is

the most difficult since measurement of CLint at very low enzyme concentrations, particularly for low turnover compounds, is challenging. It can be considered if other investigations fail to yield the reasons for poor IVIVE. The method has helped in this regard in the authors' laboratory and underlines again that detail should not be overlooked. One other point of note is that although the hepatic metabolic clearance prediction method operated by AstraZeneca10,60 uses human hepatocytes rather than human liver microsomes (for reasons described above including the membrane permeability of the drug being challenged and complete transporter and enzymology activity), there is a definite place for human liver microsomal data and clearance predictions utilizing this data (Table 1). Human liver microsomal (HLM) CLint has a larger dynamic range than human hepatocyte CLint due to the scaling factors (e.g., 120 million hepatocytes/g of liver compared to 45 mg of microsomal protein/g of liver60) and offers an understanding of the oxidative metabolism liability for the compound in the absence of complications such as membrane permeability and active transport considerations associated with hepatocytes. In the authors' experience, obtaining HLM and human hepatocyte CLint values has on many occasions proved valuable in identifying issues with the latter that require investigation. Besides being eliminated from the body by metabolism, drugs can be eliminated directly into the urine or excreted into the bile. If a drug is cleared from the systemic circulation into the renal proximal tubule only by the process of filtration through the glomerulus, then renal CL is the product of the fraction unbound in blood and glomerular filtration rate (GFR).22 This process occurs for all drugs. However, since human GFR is only approximately 1.7 mL/min/kg, CL by this mechanism is limited by GFR. Moreover, passive reabsorption back into the kidney is highly correlated with physicochemical properties of the NCE so that typically only compounds with negative logD7.4 values are passively renally cleared to any significant extent.77 Active renal excretion of drugs is complex and involves several processes including passive and active tubular secretion and passive and active reabsorption. Active renal secretion can be considered a two-step process consisting of drug uptake of across the basolateral membrane of the proximal tubule followed by exit across the apical membrane. Different sets of transporters polarized to either the apical or basolateral membrane are involved: In human the organic anion transporters OAT1 and OAT3 and the organic cation transporter OCT2 are the predominant transporters in the basolateral membrane while the apical step can involve MDR1 (P-glycoprotein), MRP2, MRP4 or BCRP along with organic cation transporters including OCTN1, OCTN2 and MATE-1 and organic anion transporters such as OAT4 or URAT1.78 Obviously predicting human renal CL when drug transporters are involved is open to more risk than when only filtration and passive reuptake of drug into the blood occurs. Even so, an effective prediction method has emerged based on data from a set of 36 diverse drugs having active secretion. It appears that human renal CL in healthy subjects can be predicted with a high degree of accuracy directly from dog renal CL after correction for differences in plasma protein binding and kidney blood flow79 (Figure 7). Male rat renal clearance correlates less well with human possibly due to poor species crossover of OAT substrates or male/female differences for rat oatp substrates.79−81 Although the method depicted in Figure 7 is 1197

dx.doi.org/10.1021/mp300476z | Mol. Pharmaceutics 2013, 10, 1191−1206

Molecular Pharmaceutics

Article

scenario is that a poor prediction of rat clearance from in vitro hepatic metabolic data would precipitate a renal and biliary excretion study, assuming the compound to be of sufficient interest as a potential human therapeutic agent. From such data accurate decisions must be made about the suitability of the compound to progress further, and therefore, understanding rat to human differences in biliary clearance is important. In the authors’ laboratory, a comprehensive analysis of twenty-two drugs of all charge types and several different therapeutic classes has been compiled in order to compare rat and human biliary clearance data.91 For nine of the drugs it was possible to include dog data in the analysis. Biliary clearance values were normalized for body weight and cross-species differences in plasma protein binding. For 86% of the drugs, rat unbound biliary clearance values, when normalized for body weight, exceeded those for human by factors ranging from 9 to over 2500-fold. Hepatic uptake and efflux transporter involvement was defined for many of the drugs, and the findings suggested that, regardless of the biliary efflux transporters implicated, when drugs do not require active hepatic uptake to access the liver, there may be fairly insignificant differences in rat, dog and human biliary clearance. Conversely, when the organic aniontransporting polypeptide drug transporters are involved, one may expect at least a 10-fold discrepancy in rat to human biliary clearance but very little dog/human discrepancy in biliary clearance regardless of the processes involved. Perhaps such observations are not surprising given that access of drugs across the sinusoidal membrane regulates the clearance of drugs removed by biliary excretion92 and functionally the rat hepatic uptake transporters appear more efficient than their human and dog counterparts.73,93 A recent study of 123 NCEs showing significant overlap in physicochemical space between rat biliary excretion data and human OATP/rat oatp substrate definition appears to support this hypothesis,94 but more studies are warranted in this space. Sandwich-cultured hepatocytes maintain liver-specific functions for several days and exhibit the formation of bile canaliculi and the localization of efflux transporters on the canalicular membrane.95 Understanding the drug concentration inside the hepatocyte is key, and by using the intracellular drug concentrations to calculate an in vitro biliary intrinsic clearance (CLbile,int), the prediction of in vivo CLbile,int is greatly improved.82 The robustness of the method is still to be extensively tested, but within a particular laboratory, the rat hepatocyte sandwich culture system could be “in house validated” using the drugs used by Nakakariya et al.82 and NCEs with rat biliary clearance data and used in conjunction with the human hepatocyte sandwich culture system to support human biliary clearance predictions. Extrahepatic metabolic clearance can also be important. A starting point for consideration of such should include a combination of in cerebro identification of metabolically labile functional groups and subsequent experimental determination of the major metabolites and the enzymes responsible in order to elucidate potential mechanisms via, for example, amidases/ esterases, amine oxidases and transferases. Followup experiments using different subcellular fractions such as plasma, cytosol and nonhepatic microsomes can be illuminatory. Plasma hydrolysis of drugs can easily be scaled to whole body clearance from in vitro data by simply multiplying the measured in vitro elimination rate constant (ln(2)/T1/2) by the volume of blood. Of course unless the rate is quite rapid, drug hydrolysis in the plasma alone is unlikely to be a major

Figure 7. Relationship between human renal clearance and dog renal clearance corrected for human/dog differences in plasma protein binding and kidney blood flow. Red spots are acidic drugs, blue are bases, yellow are neutrals and green are zwitterionic drugs. Reprinted with permission from ref 79. Copyright 2011 The American Society for Pharmacology and Experimental Therapeutics.

not mechanistic in origin and inaccurate predictions may result when there are significant human−dog differences at the level of the transporter−drug interaction, this is a step forward and helpful in making predictions to human. Analogous to the in vitro hepatocyte sandwich culture model used to predict biliary CL,82 there remains the possibility that in vitro tools such as the renal proximal tubule monolayer78 can supply data that support the prediction of human renal CL, but that remains an aspiration at this time. Biliary excretion of parent drugs is also a two-step process involving the hepatic uptake transporters organic anion transporting polypeptide (OATP), organic anion transporter (OAT) or organic cation transporter (OCT) and bile canalicular efflux transporters breast cancer resistance protein (BCRP), multidrug resistance protein 1 (MRP-1 or Pglycoprotein, P-gp) or multidrug resistance protein 2 (MRP2).83,84 Biliary excretion can be an important route,85 but there has not been a wealth of literature on the subject of predicting human biliary CL of drug candidates, perhaps because of the scarcity of relevant clinical data.61 A wide variety of interspecies allometric scaling approaches have been assessed,85−87 but given the low number of drugs used in the analyses, the fact that some of the examples used involved total drug related material excreted rather than parent drug88 and that allometry underpredicts human biliary clearance for some drugs but not others,89,87 a more extensive analysis has been required. Morris and co-workers recently demonstrated, from a database of eighteen drugs with known rat and human biliary clearance, that, when unbound clearance is considered, simple allometry using an exponent of 0.66 gave the best predictions. For only 31% of the drugs the human predictions fell within 2fold of observed, and for a further 31% the error was between 2-fold and 3-fold. In agreement with the previous studies, however, some drugs were shown to have human biliary clearance overestimated by 1 to 2 orders of magnitude. Multiple species allometry using biliary clearance data corrected for plasma protein binding gave much better predictions.90 However, multiple-species allometry to an extent affords better predictions because of the smoothing out of data from species where there is a large discrepancy with the human data. Additionally, in a drug discovery setting, the most likely 1198

dx.doi.org/10.1021/mp300476z | Mol. Pharmaceutics 2013, 10, 1191−1206

Molecular Pharmaceutics

Article

clearance pathway. For example, an in vitro measured plasma hydrolysis half-life of 5 min results in a rat and human clearance of approximately 7 mL/min/kg since blood volumes are approximately 50 mL/kg. More important perhaps is the indication that hydrolysis can occur at a number of sites throughout the body and the cumulative clearance can thus be both challenging to predict and high. One other point worthy of note is that if a drug is unstable in plasma, instability post sampling from the in vivo PK experiment or in the prepared analytical standards increases the risk of inaccurate measurements and predictions. Because of the large numbers of measured input parameters for human CL predictions (hepatocyte CLint, extrahepatic CLint, plasma protein binding, incubational binding, blood to plasma ratio, renal and biliary CL in rat and dog), CL predictions are open to more uncertainty than absorption or steady state volume of distribution. Nevertheless, the strategy presented provides an effective set of experiments to facilitate drug optimization and lowers the risk of an incorrect prediction of human CL as far as currently possible. In summation, total human CL can be predicted as follows: hepatic metabolic CL from IVIVE involving human hepatocyte CLint determination (Figure 6), renal CL predicted from dog renal CL (Figure 7) and biliary CL predicted from rat biliary CL (Figure 8)

human CL prediction to within 2-fold) in approximately onethird to one-half of occasions when using rat data alone, onehalf when using dog data and best (approximately two-thirds of occasions) when using monkey data alone. Clearly such relationships rely on simple blood flow driven CL with no significant interspecies understanding at the molecular level (for example in enzyme−drug or transporter−drug relationships or plasma protein binding). However, they offer a sanity check on human CL and T1/2 predictions from the detailed approach outlined in this publication. When there are differences in predictions between methods, it should prompt further investigations to provide, if at all possible, a mechanistic explanation. As such, it is the belief of the authors that monkey data need only be sought when it is relevant either as a pharmacological or safety species of relevance (that is, on rare occasions when dog is not useful) or on extremely rare occasions when there is a specific mechanistic reason for studying monkey PK. An example of this, as described previously,10 involved potential candidate drug from a series of moderately lipophilic bases. The compounds were shown to be metabolized primarily by human CYP3A4 in vitro but were shown to be unusually high affinity substrates in human hepatocytes and human liver microsomes (Km values of approximately 0.5 μM, determined according to the method described by Obach and Reed-Hagen72) but not in rat or dog hepatocytes. Fortuitously, the compounds were found to display similar kinetic characteristics with hepatocytes isolated from cynomolgus monkey. An in vivo pharmacokinetic study demonstrated that, despite the very low substrate concentration relative to the enzyme concentration in vitro and concerns over maintenance of steady state conditions with respect to the concentration of the enzyme−substrate complex, incubations conducted at 50 nM yielded a good prediction of monkey clearance when an intravenous dose of 0.1 mg/kg was used (maximum unbound systemic concentration was less than 50 nM), but overpredicted the clearance when a dose of 1 mg/kg was used. Similarly, incubations using 1 μM substrate underpredicted the clearance at 0.1 mg/kg. Thus, under appropriate conditions, the in vitro data predicted in vivo clearance well, but because rat and dog metabolism, at least kinetically, did not well represent the human situation, monkey data was extremely useful in adding confidence to the IVIVE. This is, we believe, quite different from using nonhuman primates simply because clearance as a percentage of liver blood flow may be closer to that predicted in human. Good IVIVE (based on unbound in vivo and in vitro intrinsic clearance ratios, as described earlier) in rat and dog should under normal circumstances be sufficient to give confidence in human predictions, and simple allometric relationships only strengthen the case, nothing more, since they may only be expected in onethird to one-half of occasions for rat and approximately 50% of the time for dog. Therefore, if it is assumed that rat and dog need to have a similar hepatic extraction ratio as humans in order to progress a compound toward clinical studies, compounds with acceptable human PK will be discarded unnecessarily and time and effort may be spent searching for species other than rat and dog in which to perform PK studies.98 In our experience the only requirement is that sufficient exposure in rat and dog can be attained for safety and pharmacodynamic studies and that the clearance in both species is very well understood (in vivo/in vitro unbound CLint ratio of less than 2). Then a compound can satisfactorily progress.

Figure 8. Relationship between unbound human biliary clearance and unbound rat (solid circles) and unbound dog (open squares) biliary clearance. Reprinted with permission from ref 91. Copyright 2012 The American Society for Pharmacology and Experimental Therapeutics.

supported by data from hepatocyte sandwich culture incubations. Although standard allometric scaling approaches offer fairly poor predictivity of human CL,58,96 a method based simply on CL as a fraction of liver blood flow observed in preclinical species may afford the DMPK scientist some additional reassurance if the analysis is consistent across two preclinical species and supports the human CL estimation toward a consensus. Examination of Ward and Smith's data set58 and that of Caldwell and co-workers96 indicated that such a method had more predictive power than simple allometry. In the Caldwell analysis of 144 drugs, human CL (expressed in L/ h) was found to approximate to 40 × rat CL, human Vss (expressed in L) approximated to 200 × rat Vss and human elimination T1/2 approximated to 4 × rat T1/2 with an average fold error of less than 3 for 80% of the drugs. However, Ward et al.58 did find monkey to be the best single species predictor of human clearance, and separate analyses by Tang et al.97 and Lombardo and coauthors56 indicate that the simple (CL as a fraction of liver blood flow) approach may only be true (giving 1199

dx.doi.org/10.1021/mp300476z | Mol. Pharmaceutics 2013, 10, 1191−1206

Molecular Pharmaceutics

Article

profiling rat bioavailability rather than absorption in a higher species, investigated the bioavailability for 553 compounds and Caco-2 permeability for a further 617. The physicochemical properties governing permeability and bioavailability were shown to be dependent on charge type with 85% of anions having a polar surface area (PSA) of less than 75 Å2 giving a high probability of good permeability and bioavailability, whereas for other charge types the “rule of 5” had sufficient predictivity.101 Thus, although it is not true to suggest that there are holistic physicochemical property ranges that allow differentiation of druglike from nondruglike,102 regions of physicochemical space can be identified that maximize the chances of acceptable gastrointestinal absorption, and the “rule of 5” remains of fundamental importance. The use of computational methods to predict gastrointestinal absorption has generated considerable interest in the literature although generally they have lower utility than ensuring the appropriate physical properties as described above. In general the approaches taken are to model human intestinal absorption (HIA) directly or else to model the individual properties known to govern the absorption process, permeability and solubility. Typically the permeability is estimated from Caco-2 or MDCK cells together with artificial membranes such as PAMPA, and solubility is determined using either kinetic or thermodynamic measurements. There are a plethora of in silico models described in the literature to predict in vitro measured permeability, and these have been well reviewed.54,103 It is sufficient to understand that, in general, the physical properties that relate to good permeability also give rise to good gastrointestinal absorption, vide supra. The importance of effective solubility has been confirmed by the fact that six years ago over 55% of the Top 200 orally administered immediate release drugs in the U.K., Japan, US and Spain were classed as high solubility.104 Hansch et al.105 reported the not unexpected quantitative structure property relationship (QSPR) showing an inverse linear relationship between the logarithm of the aqueous thermodynamic solubility and logP such that logS = 0.978 − 1.33 × logP, while others have illustrated that the strength of interaction formed between molecules within the solid crystal lattice as estimated by the melting point (MP) of the compound and the logP determine the solubility of a molecule in the solid form: LogS0solid = 0.5 − 0.01(MP − 25) − logP.106 Linear free energy relationships (LFER) for the estimations of aqueous solubility have been proposed107 where a compound’s aqueous solubility is related to the molar refractivity, polarizability, hydrogen bond acidity, hydrogen bond basicity and McGowan molecular volume. These LFER are widely accepted in their use and have gained acceptance, in many ways, due to the level of molecular understanding they possess. Nevertheless, the prediction of solubility completely from structure is notoriously difficult, perhaps best exemplified by the relatively recent solubility challenge, with over 100 entrants generating a wide range of different in silico models on 100 compounds for which intrinsic solubility had been measured in a standard assay. Thirty-two compounds whose intrinsic solubility was also measured but not disclosed were used to assess the predictive ability of each of the models. Post analysis of the data using a variety of statistical methods and descriptors showed prediction errors clustered around ∼1 log unit on average.108 Furthermore, and somewhat surprisingly, they showed that simple linear regression approaches proved superior to the more complex modeling methods involving nonlinear methods and complex descriptors. There are a

In summary, the clearance prediction methods outlined here we have developed over several years through detailed investigations and we find them consistently fruitful in providing understanding and ultimately well founded confidence in human pharmacokinetic predictions. Therefore, if the predicted human pharmacokinetics are not satisfactory, there is, we argue, very good reason to be cautious. However, if sufficient value is placed on the pharmacological target, then strategic thinking on how to overcome the shortcomings needs to take place. These may include, for example, decisions on dosing regimen and controlled release formulations if the permeability and solubility of the proposed drug allow it.

6. IN SILICO APPROACHES IN DIRECTING DRUG DESIGN The above sections have described a focused strategy for modern drug discovery DMPK scientists based on a simple approach of attempting to understand and influence absorption, distribution, clearance and therefore PK T1/2. The approach involves relatively few assays and experiments but insists on detailed comprehension of the data and accuracy of prediction. Design of the right molecules is also a vital part of an efficient and successful drug discovery program. Predictive in silico models are crucial in focusing design toward compounds with the highest probability of having suitable DMPK. To that end, current knowledge is reviewed below. Absorption. Over the past decade or so there has been significant progress in both the understanding of how drug absorption behavior relates to chemical structure and physicochemical properties. Many of the publications in this area aim at the identification of the optimum physicochemical space for oral drugs and derive empirical rules based on probability arguments and hence suggest “rules of thumb” that maximize the probability of success in achieving good absorption. The most widely adopted work in this area is the “rule of five”,13 which came out of an analysis of candidate drugs that had reached phase II clinical trials and were generally considered to have good absorption properties. The “rule” suggested compounds with a molecular weight greater than 500 and a clogP greater than 5 are more likely to have poor absorption due to poor permeation. A separate analysis of a large Caco-2 permeability and gastrointestinal absorption data set from structurally diverse compounds suggested that lipophilicity and molecular weight are the most important properties relating to permeability, since as a compound’s molecular weight increases so must its hydrophobicity in order to improve the permeability.99 In general the single property that can best be used to optimize gastrointestinal absorption is the lipophilicity at pH 7.4 as described by Comer et al.:100 A logD7.4 of less than 1 generally leads to good solubility but potentially low absorption due to low passive permeability, although there may be paracellular absorption if molecular weight is less than 350; logD7.4 values between 1 and 3 is the ideal range for good intestinal absorption, with compounds tending to have a good balance of solubility and passive permeability; for compounds with logD7.4 values between 3 and 5, permeability is likely to be good but absorption is compromised due to reduced solubility. A logD7.4 of greater than 5 leads to generally poor absorption and bioavailability due to poor solubility and high first pass metabolism. Our experience supports this analysis: working in the logD7.4 range of 1−3 for this and a variety of other PK and toxicity reasons maximizes the chance of success. Another analysis, albeit 1200

dx.doi.org/10.1021/mp300476z | Mol. Pharmaceutics 2013, 10, 1191−1206

Molecular Pharmaceutics

Article

models were also built for rat and mouse liver microsomal data, and structural features useful for one species but not for others were apparent.121 A conceptually simple in silico model for the prediction of total human clearance, based on the premise that similar structures exhibit similar pharmacokinetic properties and using the k-nearest neighbor (kNN) technique, showed that for acids, bases and zwitterions predictions could be made to within 2-fold of the observed clearance.122 Recently a linear PLS model, built using physicochemical descriptors, structural fragments and a data set of 754 compounds, has been shown to describe the biotransformation process better than physicochemical properties.123 The model yielded a geometric mean fold error of 2.1 and a percentage of compounds with predicted total human clearance within 2-fold and 3-fold of 59% and 80% respectively. It is true to say, therefore, that despite the inherent difficulties associated with in silico clearance predictions, models and general principles exist that can be used to impact the compound design stage. For renal clearance, PLS or RF models that can predict human renal clearance for acidic/zwitterionic, basic and neutral drugs with approximate average fold errors of 3, 3 and 4 respectively have been published.124 Advantageously, a classification tree generated using the classification and regression trees (CART) method allowed a simple set of renal clearance rules to be defined that can be applied to aid drug design. The rules are influenced by lipophilicity and ion class and can correctly predict 60% of an independent test set. These percentages increase to 71% and 79% for drugs with renal clearances of 1 mL/min/kg, respectively. Similarly, a computational method for the prediction of biliary excretion in rats has been described.125 The model was built using 50 diverse structures from 14 different discovery programs and fitted the data with an r2 = 0.89. The properties of the compounds covered a range of physicochemical space including molecular weights (278−739), number of rotational bonds (1−16), PSA (46.9−183.4 Å2) and ΔGsolv, aq (−53.0 to 15.7 kcal/mol). The analysis used a simple multiple linear regression technique with three physicochemical properties: polar surface area (PSA), free energy of aqueous solvation (ΔGsolv,aq) and an indicator variable for the presence or absence of a carboxylic acid. A test set of 25 literature compounds with known biliary clearance was used to validate the model. Predictions were good with an overall r2 = 0.86.

multitude of QSPR models reported in the literature and several reviews that offer a comprehensive summary of the different methodologies and assessment of commercially available software,109−113 and the top five commercially available predictive software packages when tested against a 122 compound test set were those provided by SimulationPlus, Admensa, Pharma Algorithms ADME Boxes, ChemSilico, ACDLabs.114 In a drug discovery context, where the molecular changes to compounds are often small and the material is of relatively consistent solid characteristics (crystalline or amorphous), the greatest probability of successful predictions is given by the approach described by Rogers et al.115,116 Volume of Distribution. The in silico prediction of human and preclinical species Vss should always start with a consideration of chemotype. Gleeson et al. first highlighted this by categorizing the mean and standard deviation of literature human Vss values.117 For acids the mean Vss was found to be 0.2 L/kg; for basic drugs, 4.1 L/kg; for zwitterions, 0.8 L/kg; and for neutrals, 1.1 L/kg. Three different statistical methodologies were used in prediction models based on this data set: Bayesian neural networks (BNN), classification and regression trees (CART) and partial least-squares (PLS). The results in prediction of an independent test set showed the human model to have an r2 of 0.60 and an rms (root-meansquare) error in prediction of 0.48. The corresponding rat model had an r2 of 0.53 and an rms error in prediction of 0.37, indicating both models to be very useful in the compound design stage of the drug discovery process. Recently a data set of 669 drugs46 was analyzed using random forest (RF) and PLS methods to represent linear and nonlinear statistical techniques. The models were validated with an external test set of 29 compounds, and it was shown that these models were able to predict human Vss within geometric mean 2-fold error, comparable to predictions using in vivo pharmacokinetic data. Clearance. The in silico estimation of clearance is one of the most difficult of the ADME properties to predict due to the strong structural dependency of metabolism.118 A number of different methods have been employed, and these are described below. Quantitative structure−pharmacokinetic relationships (QSPkR), developed from physicochemical and topological properties and employing a number of different statistical methods, have indicated that the most successful methods are the general regression neural networks (GRNN), support vector regression (SVR) and k-nearest neighbor clustering (kNN), where the percentage of compounds with predicted total clearance within 2-fold error of actual values was in the range of 62−74%, a level of prediction accuracy offering high value in prioritizing lead compounds. 117 An in silico classification model for the prediction of human liver microsomal (HLM) stability using the apparent intrinsic stability (CLint,app) as the end point has also been used successfully.119 The data set comprised 14557 compounds and was partitioned into a training set consisting of 11646 and a test set of 2911 together with a validation set of a further 276 compounds. Random forest and Bayesian classification methods with MOE, E-state descriptors, ADME keys, and ECFP_6 fingerprints were used to generate QSAR models, with the best showing 80 and 75% prediction accuracy for the test and validation sets respectively. Another classification model for HLM metabolic stability data, using a data set of 24081 the naive Bayesian classification method available in Pipeline Pilot using FCFP6 fingerprints, was shown to have a prediction accuracy of 78%.120 Interestingly, classification

7. STRATEGIC USE OF PK PARAMETERS As stated above, there are four key parameters to optimize in a Drug Discovery environment, with CL and Vss having the biggest impact on dose and avoiding excessively high maximal plasma concentrations. A basic tenet of clinical pharmacokinetics is that the magnitude of both the desired response and toxicity are functions of the drug concentration. Accordingly, for a relevant drug target, therapeutic failure results when either the concentration is too low, giving ineffective therapy, or the concentration is too high, producing unacceptable toxicity. Between these limits lies a concentration range associated with therapeutic success: the “therapeutic window”. This fluctuation in the drug concentration depends on both frequency of dosing and the effective half-life.22 For once daily dosing our laboratory targets a human half-life prediction of between 16 and 20 h, since being 2-fold out in either direction (e.g., 16 to 8 h or 32 h) elevates the dose and maximum drug concentration (Cmax) by only a factor of 2 or results in an accumulation from 1201

dx.doi.org/10.1021/mp300476z | Mol. Pharmaceutics 2013, 10, 1191−1206

Molecular Pharmaceutics

Article

then targeting a minimum free drug concentration (Cmin) at the end of the dosing interval is what is important in practice and the period prior to the drug concentration reaching that value can give a greater receptor occupancy for a much less than linear increase in efficacy (hence the targeting of a given Cmin for required effect). In this very common scenario, for acidic drugs with distribution volumes fixed at a lower limiting value, attenuation of the half-life through plasma protein binding (impacting the CL but not Vss) lowers the Cmax (and therefore dose) for a required Cmin and efficacy. Neutral drugs have moderate Vss values, and therefore, as with acidic drugs above, a simple look through the math informs that if Vss is 1 L/kg and we require a T1/2 of 20 h, CL must be 0.55 mL/min/kg, and if plasma protein binding is 90% (typical for a neutral drug of moderate lipophilicity), human hepatocyte CLint will be approximately 1 μL/min/million cells. Since raising the plasma protein binding for neutral compounds may involve raising lipophilicity, a more appropriate strategy is to focus on lowering lipophilicity and metabolic blocking to control CLint. A similar strategy can be considered for basic drugs, although the fact that bases have high affinity for phospholipid membranes owing to interactions with acidic head groups can be used as an advantage if other issues such as safety considerations allow raising the pKa to facilitate increased half-life through higher Vss. In summary, the purpose of this manuscript has been to demonstrate that relatively little experimental data is required to gain a large amount of confidence in predicted human PK. Table 1 covers these key experiments. Additionally, in an age when centralized screening functions provide much routine DMPK data, scientists can focus on maximizing the chances of success through performing bespoke experiments and understanding how each PK parameter can be used to form a successful strategy.

single dose to steady state of only 4-fold. On the other hand to be wrong by a factor of 2 in the prediction of a half-life of only 8 h (e.g., 8 to 4 h) for a once daily drug would lead to a dose and Cmax elevation of 8-fold. Although a somewhat empirical assessment, it is a constructive way to view the risks. Of course one may take the view that appropriate compounds are potentially being discarded and that drug projects (in particular medicinal chemistry colleagues) are asked to chase even harder goals. Our view is based on personal experience gained from somewhat bitter experience of being correct to within 2-fold on the tough to meet composite parameter of half-life, but then leaving considerably higher hurdles in clinical development due to the elevation in clinical dose as the prediction was 2-fold in the wrong direction. Even then it should be noted that the predictive methodologies outlined in this article do not offer a panacea for successful PK prediction (subjectively characterized by correctly estimating a parameter within 2-fold), since successful predictions of individual PK parameters may not equate to a successful prediction of clinical dose and exposure due to the cumulative impact of a 2-fold error in CL and Vss predictions on T1/2, exposure and dose.9 Understanding the interplay between the PK parameters and dose allows strategies to be formed that can differentiate between speed and success in a drug discovery project and the converse. For example, an awareness of the likely limited range of Vss values for a given chemotype (Figure 4) allows an approach for obtaining acceptable half-lives in human through an appreciation of what an acceptable CL value is and what parameters are available to work on to manipulate CL. If optimizing acidic compounds, the likely upper limit for Vss is likely to be 0.3 L/kg in the majority of cases (Figure 4). Using the equation elimination T1/2 = ln 2 × Vss/CL, it is clear that a CL value of 0.15 mL/min/kg must be obtained if a 16 h halflife is to be achieved (assuming a rapid enough absorption that the elimination half-life is essentially dependent on CL and Vss). Using AstraZeneca prediction methods60 and assuming that the acidic drug in question is 99% bound to plasma proteins and has a blood to plasma ratio of 0.6, a human hepatocyte CLint of 1 μL/min/million cells not only would be required but also would need to be robustly and confidently measured. Indeed if the plasma protein binding was only 90%, a human hepatocyte CLint of 0.1 μL/min/million cells would be required, which would be challenging to experimentally determine with confidence. A postulated strategy for optimizing acidic drugs is to manipulate the CL of such compounds through increasing plasma protein binding, provided that free blood levels can be maintained to provide efficacy at the target receptor.10 Typically plasma protein binding is not a parameter for optimization in drug discovery projects, but this example provides a good illustration that a quantitative understanding of the interplay between the different PK and PD parameters can facilitate the right strategy to identification of an acceptable clinical candidate within a defined area of chemical space. Previous approaches may have perhaps involved a little more serendipity, albeit guided by in vivo pharmacology results, but the end point is the same. We note for example that only 3% of 60 marketed oral acidic drugs have plasma protein binding of less than 99% with half-lives of more than 8 h.126,127 Reservations over such a strategy have been raised since the pharmacological target exposure to unbound drug remains constant regardless of the extent of plasma protein binding.48 However, if one aims to “cover” a set fraction of pharmacological receptors to achieve the required efficacy,



AUTHOR INFORMATION

Corresponding Author

*AstraZeneca, Respiratory & Inflammation DMPK, AstraZeneca R&D Mölndal, SE 43183 Mölndal, Sweden. Tel: +46 (0) 317761815. Fax: +46 (0)317762800. E-mail: ken.grime@ astrazeneca.com. Notes

The authors declare no competing financial interest.



ACKNOWLEDGMENTS We acknowledge Rob Riley, Douglas Fergusson, Peter Webborn, Stuart Paine, Richard Weaver, Anne Cooper, Matt Soars, Rupert Austin and Andy Davis for many high quality scientific discussions over several years.



REFERENCES

(1) Prentis, R. A.; Lis, Y.; Walker, S. R. Pharmaceutical innovation by the seven UK-owned pharmaceutical companies (1964−1985). Br. J. Clin. Pharmacol. 1988, 25, 387−396. (2) Schuster, D.; Laggner, C.; Langer, T. Why drugs fail–a study on side effects in new chemical entities. Curr. Pharm. Des. 2005, 11, 3545−3559. (3) Kubinyi, H. Drug research: myths, hype and reality. Nat. Rev. Drug Discovery 2003, 2, 665−668. (4) Sams-Dodd, F. Optimizing the discovery organization for innovation. Drug Discovery Today 2005, 10, 1049−1056. (5) Macarron, R. Critical review of the role of HTS in drug discovery. Drug Discovery Today 2006, 11, 277−279. 1202

dx.doi.org/10.1021/mp300476z | Mol. Pharmaceutics 2013, 10, 1191−1206

Molecular Pharmaceutics

Article

(6) Riley, R. J.; Kenna, J. G. Cellular models for ADMET predictions and evaluation of drug-drug interactions. Curr. Opin. Drug Discovery Dev. 2004, 7, 86−99. (7) Riley, R. J.; Grime, K. Metabolic screening in vitro: metabolic stability, CYP inhibition and induction. Drug Discovery Today 2004, 1, 365−372. (8) Leeson, P. D.; Springthorpe, B. The influence of drug-like concepts on decision-making in medicinal chemistry. Nat. Rev. Drug Discovery 2007, 6, 881−890. (9) McGinnity, D. F.; Collington, J.; Austin, R. P.; Riley, R. J. Evaluation of human pharmacokinetics, therapeutic dose and exposure predictions using marketed oral drugs. Curr. Drug Metab. 2007, 8, 463−479. (10) Grime, K.; Riley, R. J. The Impact of In Vitro Binding on In Vitro - In Vivo Extrapolations, Projections of Metabolic Clearance and Clinical Drug-Drug Interactions. Curr. Drug Metab. 2006, 7, 251−264. (11) Riley, R. J.; Martin, I. J.; Cooper, A. E. The influence of DMPK as an integrated partner in modern drug discovery. Curr. Drug Metab. 2002, 3, 527−550. (12) van de Waterbeemd, H.; Gifford, E. ADMET in silico modelling: towards prediction paradise? Nat. Rev. Drug Discovery 2003, 2, 192−204. (13) Lipinski, C. A. Drug-like properties and the causes of poor solubility and poor permeability. J. Pharmacol. Toxicol. Methods 2000, 44, 235−249. (14) Lipinski, C. A.; Lombardo, F.; Dominy, B. W.; Feeney, P. J. Experimental and computational approaches to estimate solubility and permeability in drug discovery and development settings. Adv. Drug Delivery Rev. 1997, 23, 3−25. (15) Gabrielsson., J.; Dolgos, H.; Gillberg, P. G.; Bredberg, U.; Benthem, B.; Duker, G. Early integration of pharmacokinetic and dynamic reasoning is essential for optimal development of lead compounds: strategic considerations. Drug Discovery Today 2009, 14, 358−372. (16) Malik, N. N. Drug discovery: past, present and future. Drug Discovery Today 2008, 13, 909−912. (17) Petrillo, E. W. Lean thinking for drug discoverybetter productivity for pharma. Drug Discovery World 2007, No. Spring, 9− 14. (18) Schmid, E. F.; Smith, D. A. Keynote review: Is declining innovation in the pharmaceutical industry a myth? Drug Discovery Today 2005, 1, 1031−1039. (19) Sinha, G. Pharma bankrolls academic institutes. Nat. Biotechnol. 2012, 30, 474. (20) Hughes, B. China spurs pharma innovation. Nat. Rev. Drug Discovery 2008, 7, 631−632. (21) Hellriegel, E. T.; Bjornsson, T. D.; Hauck, W. W. Interpatient variability in bioavailability is related to the extent of absorption: implications for bioavailability and bioequivalence studies. Clin. Pharmacol. Ther. 1996, 60, 601−607. (22) Rowland, M.; Tozer, T. N. Clinical PharmacokineticsConcepts and Applications, 3rd ed.; Williams and Wilkins: 1989. (23) Komura, H.; Iwaki, M. In vitro and in vivo small intestinal metabolism of CYP3A and UGT substrates in preclinical animals species and humans: species differences. Drug Metab. Rev. 2011, 43, 476−498. (24) Galetin, A.; Gertz, M.; Houston, J. B. Contribution of intestinal cytochrome p450-mediated metabolism to drug-drug inhibition and induction interactions. Drug Metab. Pharmacokinet. 2010, 25, 28−47. (25) Thelen, K.; Dressman, J. B. Cytochrome P450-mediated metabolism in the human gut wall. J. Pharm. Pharmacol. 2009, 61, 541−558. (26) Paine, M. F.; Hart, H. L.; Ludington, S. S.; Haining, R. L.; Rettie, A. E.; Zeldin, D. C. The human intestinal cytochrome P450 “pie”. Drug Metab. Dispos. 2006, 34, 880−886. (27) Zhang, Q. Y.; Dunbar, D.; Ostrowska, A.; Zeisloft, S.; Yang, J.; Kaminsky, L. S. Characterization of human small intestinal cytochromes P-450. Drug Metab. Dispos. 1999, 27, 804−809.

(28) Prueksaritanont, T.; Gorham, L. M.; Hochman, J. H.; Tran, L. O.; Vyas, K. P. Comparative studies of drug-metabolizing enzymes in dog, monkey, and human small intestines, and in Caco-2 cells. Drug Metab. Dispos. 1996, 24, 634−642. (29) Yang, J.; Jamei, M.; Yeo, K. R.; Tucker, G. T.; RostamiHodjegan, A. Prediction of intestinal first-pass drug metabolism. Curr. Drug Metab. 2007, 8, 676−684. (30) Doherty, M. M.; Charman, W. N. The mucosa of the small intestine: how clinically relevant as an organ of drug metabolism? Clin. Pharmacokinet. 2002, 41, 235−253. (31) Paine, M. F.; Khalighi, M.; Fisher, J. M.; Shen, D. D.; Kunze, K. L.; Marsh, C. L.; Perkins, J. D.; Thummel, K. E. Characterization of inter- and intra-intestinal differences in human CYP3A-dependent metabolism. J. Pharmacol. Exp. Ther. 1997, 283, 1552−1562. (32) Galetin, A.; Houston, J. B. Intestinal and hepatic metabolic activity of five cytochrome P450 enzymes: impact on prediction of first-pass metabolism. J. Pharmacol. Exp. Ther. 2006, 318, 1220−1229. (33) Galetin, A. Intestinal first-pass metabolism: bridging the gap between in vitro and in vivo. Curr. Drug Metab. 2007, 8, 643−644. (34) Zhang, Y.; Benet, L. Z. The gut as a barrier to drug absorption: combined role of cytochrome P450 3A and P-glycoprotein. Clin. Pharmacokinet. 2001, 40, 159−168. (35) Galetin, A.; Gert, M.; Houston, J. B. Potential role of intestinal firstpass metabolism in the prediction of drug−drug interactions. Expert Opin. Drug Metab. Toxicol. 2008, 4, 909−922. (36) Lin, X.; Skolnik, S.; Chen, X.; Wang, J. Attenuation of intestinal absorption by major efflux transporters: quantitative tools and strategies using a Caco-2 model. Drug Metab. Dispos. 2011, 39, 265−274. (37) Murakami, T.; Takano, M. Intestinal efflux transporters and drug absorption. Expert Opin. Drug Metab. Toxicol. 2008, 4, 923−939. (38) Bolger, M. B.; Lukacova, V.; Woltosz, W. S. Simulations of the nonlinear dose dependence for substrates of influx and efflux transporters in the human intestine. AAPS J. 2009, 11, 353−363. (39) Guidance for Industry Drug Interaction Studies - Study Design, Data Analysis, Implications for Dosing and Labeling Recommendations, U.S. Department of Health and Human Services Food and Drug Administration Center for Drug Evaluation and Research (CDER). 2012. (40) Hidalgo, I. J.; Raub, T. J.; Borchardt, R. T. Characterization of the human colon carcinoma cell line (Caco-2) as a model system for intestinal epithelial permeability. Gastroenterology 1989, 96, 736−749. (41) Sun, D.; Lennernas, H.; Welage, L. S.; Barnett, J. L.; Landowski, C. P.; Foster, D.; Fleisher, D.; Lee, K. D.; Amidon, G. L. Comparison of human duodenum and Caco-2 gene expression profiles for 12,000 gene sequences tags and correlation with permeability of 26 drugs. Pharmacol. Res. 2002, 19, 1400−1416. (42) Jamei, M.; Marciniak, S.; Feng, K.; Barnett, A.; Tucker, G.; Rostami-Hodjegan, A. The Simcyp population-based ADME simulator. Expert Opin. Drug Metab. Toxicol. 2009, 5, 211−223. (43) Sinha, V. K.; Snoeys, J.; Osselaer, N. V.; Peer, A. V.; Mackie, C.; Heald, D. From preclinical to human–prediction of oral absorption and drug-drug interaction potential using physiologically based pharmacokinetic (PBPK) modeling approach in an industrial setting: a workflow by using case example. Biopharm. Drug Dispos. 2012, 33, 111−121. (44) DeSesso, J. M.; Jacobson, C. F. Anatomical and physiological parameters affecting gastrointestinal absorption in humans and rats. Food Chem. Toxicol. 2001, 39, 209−228. (45) Øie, S.; Tozer, T. N. Effect of altered plasma protein binding on apparent volume of distribution. J. Pharm. Sci. 1979, 68, 1203−1205. (46) Lombardo, F.; Waters, N. J.; Argikar, U. A.; Dennehy, M. K.; Zhan, J.; Gunduz, M.; Harriman, S. P.; Berellini, G.; Liric Rajlic, I.; Obach, R. S. Comprehensive Assessment of Human Pharmacokinetic Prediction Based on In Vivo Animal Pharmacokinetic Data, Part 1: Volume of Distribution at Steady State. J. Clin. Pharmacol. 2012, DOI: 10.1177/0091270012440281. 1203

dx.doi.org/10.1021/mp300476z | Mol. Pharmaceutics 2013, 10, 1191−1206

Molecular Pharmaceutics

Article

17beta-D-glucuronide in cryopreserved human hepatocytes. Drug Metab. Pharmacokinet. 2003, 18, 33−41. (67) Riley, R. J.; McGinnity, D. F.; Austin, R. P. A unified model for predicting human hepatic, metabolic clearance from in vitro intrinsic clearance data in hepatocytes and microsomes. Drug Metab. Dispos. 2005, 33, 1304−1311. (68) Rawden, H. C.; Carlile, D. J.; Tindall, A.; Hallifax, D.; Galetin, A.; Ito, K.; Houston, J. B. Microsomal prediction of in vivo clearance and associated interindividual variability of six benzodiazepines in humans. Xenobiotica 2005, 35, 603−625. (69) Naritomi, Y.; Terashita, S.; Kagayama, A.; Sugiyama, Y. Utility of hepatocytes in predicting drug metabolism: comparison of hepatic intrinsic clearance in rats and humans in vivo and in vitro. Drug Metab. Dispos. 2003, 31, 580−588. (70) Poulin, P.; Kenny, J. R.; Hop, C. E.; Haddad, S. In vitro-in vivo extrapolation of clearance: modeling hepatic metabolic clearance of highly bound drugs and comparative assessment with existing calculation methods. J. Pharm. Sci. 2012, 101, 838−851. (71) Hallifax, D.; Houston, J. B. Evaluation of Hepatic Clearance Prediction Using In Vitro Data: Emphasis on fraction Unbound in Plasma and Drug Ionisation Using a database of 107 Drugs. J. Pharm. Sci. 2012, 101, 2645−2652. (72) Obach, R. S.; Reed-Hagen, A. E. Measurement of Michaelis constants for cytochrome P450-mediated biotransformation reactions using a substrate depletion approach. Drug Metab. Dispos. 2002, 30, 831−837. (73) Soars, M. G.; Grime, K.; Sproston, J. L.; Webborn, P. J.; Riley, R. J. Use of hepatocytes to assess the contribution of hepatic uptake to clearance in vivo. Drug Metab. Dispos. 2007b, 36, 859−865. (74) Austin, R. P.; Barton, P.; Cockfroft, S. L.; Wenlock, M. P.; Riley, R. J. The influence of nonspecific microsomal binding on apparent intrinsic clearance, and its prediction from physicochemical properties. Drug Metab. Dispos. 2002, 30, 1497−1503. (75) Austin, R. P.; Barton, P.; Mohmed, S.; Riley, R. J. The binding of drugs to hepatocytes and its relationship to physicochemical properties. Drug Metab. Dispos. 2005, 33, 419−425. (76) Hallifax, D.; Houston, J. B. Binding of drugs to hepatic microsomes: comment and assessment of current prediction methodology with recommendation for improvement. Drug Metab. Dispos. 2006, 34, 1078. (77) Smith, D. A.; Jones, B. C.; Walker, D. K. Design of Drugs Involving the Concepts and Theories of Drug Metabolism and Pharmacokinetics. Med. Res. Rev. 1996, 16, 243−266. (78) Brown, C. D. A.; Sayer, R.; Windass, A. S.; Haslam, I. S.; De Broe, M. E.; D’Haese, P. C.; Verhulst, A. Characterisation of human tubular cell monolayers as a model of proximal tubular xenobiotic handling. Toxicol. Appl. Pharmacol. 2008, 233, 428−438. (79) Paine, S. W.; Menochet, K.; Denton, R.; McGinnity, D. F.; Riley, R. J. Prediction of Human Renal Clearance from Preclinical Species for a Diverse Set of Drugs That Exhibit Both Active Secretion and Net Reabsorption. Drug Metab. Dispos. 2011, 39, 1008−1013. (80) Tahara, H.; Shono, M.; Kusuhara, H.; Kinoshita, H.; Fuse, E.; Takadate, A.; Otagiri, M.; Sugiyama, Y. Molecular cloning and functional analyses of OAT1 and OAT3 from cynomolgus monkey kidney. Pharm. Res. 2005, 22, 647−660. (81) Kato, Y.; Kuge, K.; Kusuhara, H.; Meier, P. J.; Sugiyama, Y. Gender difference in the urinary excretion of organic anions in rats. J. Pharmacol. Exp. Ther. 2002, 302, 483−489. (82) Nakakariya, M.; Shimada, T.; Irokawa, M.; Koibuchi, H.; Iwanaga, T.; Yabuuchi, H.; Maeda, T.; Tamai, I. Predominant Contribution of Rat Organic Anion Transporting Polypeptide-2 (Oatp2) to Hepatic Uptake of β-Lactam Antibiotics. Pharm. Res. 2010, 25, 578−585. (83) Kusuhara, H.; Sugiyama, Y. Pharmacokinetic modeling of the hepatobiliary transport mediated by cooperation of uptake and efflux transporters. Drug Metab Rev. 2010, 42, 539−550. (84) Shitara, Y.; Horie, T.; Sugiyama, Y. Transporters as a determinant of drug clearance and tissue distribution. Eur. J. Pharm. Sci. 2006, 27, 425−446.

(47) Berry, L. M.; Li, D.; Zhao, Z. Species differences in distribution and prediction of human V(ss) from preclinical data. Drug Metab. Dispos. 2011, 39, 2103−2116. (48) Smith, D. A.; Kerns, E. H. The effect of plasma protein binding on in vivo efficacy: misconceptions in drug discovery. Nat. Rev. Drug Discovery 2010, 9, 929−939. (49) MacIntyre, A. C.; Cutler, D. J. The potential role of lysosomes in tissue distribution of weak bases. Biopharm. Drug Dispos. 1988, 9, 513−526. (50) Cooper, A. E.; Ferguson, D.; Grime, K. Optimisation of DMPK by the inhaled route: challenges and approaches. Curr. Drug Metab. 2012, 13, 457−473. (51) Berry, L. M.; Roberts, J.; Be, X.; Zhao, Z.; Lin, M. H. Prediction of V(ss) from in vitro tissue-binding studies. Drug Metab. Dispos. 2010, 38, 115−121. (52) Gleeson, M. P.; Waters, N. J.; Paine, S. W.; Davis, A. M. In silico human and rat Vss quantitative structure-activity relationship models. J. Med. Chem. 2006, 49, 1953−1963. (53) Berellini, G.; Springer, C.; Waters, N. J.; Lombardo, F. In silico prediction of volume of distribution in human using linear and nonlinear models on a 669 compound data set. J. Med. Chem. 2009, 52, 4488−4495. (54) Zou, P.; Zheng, N.; Yang, Y.; Yu, L. X.; Sun, D. Prediction of volume of distribution at steady state in humans: comparison of different approaches. Expert Opin. Drug Metab. Toxicol. 2012, 8, 855− 872. (55) Jones, R. D.; Jones, H. M.; Rowland, M.; Gibson, C. R.; Yates, J. W.; Chien, J. Y.; Ring, B. J.; Adkison, K. K.; Ku, M. S.; He, H.; Vuppugalla, R.; Marathe, P.; Fischer, V.; Dutta, S.; Sinha, V. K.; Björnsson, T.; Lavé, T.; Poulin, P. PhRMA CPCDC initiative on predictive models of human pharmacokinetics, part 2: Comparative assessment of prediction methods of human volume of distribution. J. Pharm. Sci. 2011, 100, 4074−4089. (56) Lombardo, F.; Waters, N. J.; Argikar, U. A.; Dennehy, M. K.; Zhan, J.; Gunduz, M.; Harriman, S. P.; Berellini, G.; Liric Rajlic, I.; Obach, R. S. Comprehensive Assessment of Human Pharmacokinetic Prediction Based on In Vivo Animal Pharmacokinetic Data, Part 2: Clearance. J . C l i n . P h a rm a co l . 201 2, DOI: 10.117 7/ 0091270012440282. (57) Mahmood, I. Theoretical versus empirical allometry: Facts behind theories and application to pharmacokinetics. J. Pharm. Sci. 2010, 99, 2927−2933. (58) Ward, K. W.; Smith, B. R. A comprehensive quantitative and qualitative evaluation of extrapolation of intravenous pharmacokinetic parameters from rat, dog, and monkey to humans. I. Clearance. Drug Metab. Dispos. 2004, 32, 603−611. (59) Campbell, D. B. Extrapolation from animals to man. The integration of pharmacokinetics and pharmacodynamics. Ann. N.Y. Acad. Sci. 1996, 801, 116−135. (60) Sohlenius-Sternbeck, A.-K.; Jones, C.; Ferguson, D.; Middleton, B. J.; Projean, D.; Floby, E.; Bylund, J.; Afzelius, L. Practical use of the regression offset approach for the prediction of in vivo intrinsic clearance from hepatocytes. Xenobiotica 2012, 42, 841−853. (61) Lavé, T.; Chapman, K.; Goldsmith, P.; Rowland, M. Human clearance prediction: shifting the paradigm. Expert Opin. Drug Metab. Toxicol. 2009, 9, 1039−1048. (62) Ito, K.; Houston, J. B. Prediction of human drug clearance from in vitro and preclinical data using physiologically based and empirical approaches. Pharm. Res. 2005, 22, 103−112. (63) Houston, J. B. Utility of in vitro drug metabolism data in predicting in vivo metabolic clearance. Biochem. Pharmacol. 1994, 47, 1469−1479. (64) Obach, R. S. Predicting clearance in humans from in vitro data. Curr. Top. Med. Chem. 2011, 11, 334−339. (65) Soars, M,G.; McGinnity, D. F.; Grime, K.; Riley, R. J. The pivotal role of hepatocytes in drug discovery. Chem.-Biol. Interact. 2007a, 168, 2−15. (66) Shitara, Y.; Li, A. P.; Kato, Y.; Lu, C.; Ito, K.; Itoh, T.; Sugiyama, Y. Function of uptake transporters for taurocholate and estradiol 1204

dx.doi.org/10.1021/mp300476z | Mol. Pharmaceutics 2013, 10, 1191−1206

Molecular Pharmaceutics

Article

M. D. Multiparameter optimization: identifying high quality compounds with a balance of properties. Curr. Pharm. Des. 2012, 18, 1292−1310. (106) Jain, N.; Yalkowsky, S. H. Estimation of the aqueous solubility I: Application to organic nonelectrolytes. J. Pharm. Sci. 2001, 90, 234− 252. (107) Abraham, M. H.; Le, J. The correlation and prediction of the solubility of molecules in water using an amended solvation energy relationship. J. Pharm. Sci. 1999, 88, 868−880. (108) Llinas, A.; Glen, R. C.; Goodman, J. M. Solubility challenge: can you predict solubilities of 32 molecules using a database of 100 measurements? J. Chem. Inf. Model. 2008, 48, 1289−1303. (109) Hewitt, M.; Cronin, M. T. D.; Enoch, S. J.; Madden, J. C.; Roberts, D. W.; Dearden, C. In silico prediction of aqueous solubility: The solubility challenge. J. Chem. Inf. Model. 2009, 49, 2572−2587. (110) Bergström, C. A. S. Computational Models to Predict Aqueous Drug Solubility, Permeability and Intestinal Absorption. Expert Opin. Drug Metab. Toxicol. 2005, 1, 613−627. (111) Delaney, J. S. Predicting Aqueous Solubility From Structure. Drug Discovery Today 2005, 10, 289−295. (112) Dearden, J. C. In Silico Prediction of Aqueous Solubility. Expert Opin. Drug Discovery 2006, 1, 31−52. (113) Davis, A. M.; Bruneau, P. In Silico Prediction of Solubility. In Drug Bioavailabilty (40, Methods and Principles in Medicinal Chemistry), 2nd ed.; van de Waterbeemd, H., Testa, B., Eds; Wiley-VCH Verlag GmbH & Co. KGaA: Weinheim, Germany, 2009; pp 53−68. (114) Gleeson, M. P.; Hersey, A.; Hannongbua, S. In-Silico ADME Models: A General Assessment of their Utility in Drug Discovery Applications. Curr. Top. Med. Chem. 2011, 11, 358−381. (115) Rogers, S, L.; Davis, A. M.; Tomkinson, N. P.; van de Waterbeemd, H. QSAR modeling using automated updating correction libraries: application to a human plasma binding model. J. Chem. Inf. Model. 2007, 47, 2401−2407. (116) Rogers, S, L.; Davis, A. M.; Tomkinson, N. P.; van de Waterbeemd, H. Predictivity of simulated ADME AutoQSAR models over time. Mol. Inf. 2011, 30, 256−266. (117) Gleeson, M. P.; Waters, N. J.; Paine, S. W.; Davis, A. M. In silico human and rat vss quantitative structure-activity relationship models. J. Med. Chem. 2006, 49, 1953−1963. (118) Madden, J. C.; Cronin, T. D. Structure-based methods for the prediction of drug metabolism. Expert Opin. Drug Metab. Toxicol. 2006, 2, 545−557. (119) Yap, C. W.; Li, Z. R.; Chen, Y. Z. Quantitative structurepharmacokinetic relationships for drug clearance by using statistical learning methods. J. Mol. Graphics Modell. 2006, 24, 383− 395. (120) Lee, P. H.; Cucurull-Sanchez, L.; Lu, J.; Du, Y. J. Development of in silico models for human liver microsomal stability. J. Comput.Aided. Mol. Des. 2007, 21, 665−673. (121) Hu, Y.; Unwalla, R.; Denny, R. A.; Bikker, J.; Di, L.; Humblet, C. Development of QSAR models for microsomal stability: identification of good and bad structural features for rat, human and mouse microsomal stability. J. Comput.-Aided Mol. Des. 2010, 24, 23− 35. (122) Yu, M. J. Predicting total clearance in humans from chemical structure. J. Chem. Inf. Model 2010, 50, 1284−1295. (123) Berellini, G.; Waters, N. J.; Lombardo, F. In silico prediction of total plasma clearance. J. Chem. Inf. Model. 2012, 52, 2069−2078. (124) Paine, S. W.; Barton, P.; Bird, J.; Denton, R.; Menochet, K.; Smith, A.; Tomkinson, N. P.; Chohan, K. K. A rapid computational filter for predicting the rate of human renal clearance. J. Mol. Graphics Modell. 2010, 29, 529−537. (125) Lou, L.; Johnson, J.; Hsueh, M.-H.; Zheng, J.; Cai, H.; Xin, B.; Chong, S.; He, K.; Harper, T. W. In silico prediction of biliary excretion of drugs in rat based on physicochemical properties. Drug Metab. Dispos. 2010, 38, 422−430. (126) Taeschner, W.; Vozeh, S. Pharmacokinetic Drug Data. In Clinical Pharmacokinetics. Drug Data Handbook, 3rd ed.; Holford, N.,

(85) Mahmood, I. Interspecies scaling of biliary excreted drugs: a comparison of several methods. J. Pharm. Sci. 2005, 94, 883−892. (86) Mahmood, I.; Sahajwalla, C. Interspecies scaling of biliary excreted drugs. J. Pharm. Sci. 2002, 91, 1908−1014. (87) Sawada, Y.; Hanano, M.; Sugiyama, Y.; Iga, T. Prediction of the disposition of β-lactam antibiotics in humans from pharmacokinetic parameters in animals. J. Pharmacokinet. Biopharm. 1984, 12, 241−260. (88) Scatina, J. A.; Hicks, D. R.; Kraml, M.; Cayen, M. N. Disposition of a new tetrahydrocarbazole analgesic drug in laboratory animals and man. Xenobiotica 1989, 19, 991−1002. (89) Påhlman, I.; Edholm, M.; Kankaanranta, S.; Odell, M.-L. Pharmacokinetics of Susalimod, a highly biliary excreted sulphasalazine analogue, in various species. Non precictable human clearance by allometric scaling. Pharm. Pharmacol. Commum. 1998, 4, 493−498. (90) Morris, M.; Yang, Y.; Gandhi, Y. A.; Bhansalia, S. G.; Benincosa, L. J. Interspecies scaling: prediction of human biliary clearance and comparison with QSPKR. Biopharm. Drug Dispos. 2012, 33, 1−14. (91) Grime, K. H.; Paine, S. W. Species Differences in Biliary Clearance and Possible Relevance of Hepatic Uptake and Efflux Transporters Involvement. Drug Metab. Dispos. 2012, DOI: dmd.112.049312. (92) Yamazaki, M.; Akiyama, S.; Nishigaki, R.; Sugiyama, Y. Uptake is the rate-limiting step in the overall hepatic elimination of pravastatin at steady-state in rats. Pharm. Res. 1996, 13, 1559−1564. (93) Gardiner, P.; Paine, S. W. The impact of hepatic uptake on the pharmacokinetics of organic anions. Drug Metab. Dispos. 2011, 39, 1930−1938. (94) Varma, M. V.; Chang, G.; Lai, Y.; Feng, B.; El-Kattan, A. F.; Litchfield, J.; Goosen, T. C. Physicochemical property space of hepatobiliary transport and computational models for predicting rat biliary excretion. Drug Metab. Dispos. 2012, 40, 1527−1537. (95) LeCluyse, E. L.; Audus, K. L.; Hochman, J. H. Formation of extensive canalicular networks by rat hepatocytes cultured in collagensandwich configuration. Am. J. Physiol. 1994, 266 (6 Part 1), C1764− C1774. (96) Caldwell, G. W.; Masucci, J. A.; Yan, Z.; Hageman, W. Allometric scaling of pharmacokinetic parameters in drug discovery: can human CL, Vss and t1/2 be predicted from in-vivo rat data? Allometric scaling of pharmacokinetic parameters in drug discovery: can human CL, Vss and t1/2 be predicted from in-vivo rat data? Eur. J. Drug Metab. Pharmacokinet. 2004, 29, 133−143. (97) Tang, H.; Hussain, A.; Leal, M.; Mayersohn, M.; Fluhler, E. Interspecies Prediction of Human Drug Clearance Based on Scaling Data from One or Two Animal Species. Drug Metab. Dispos. 2007, 35, 1886−1893. (98) Wong, H.; Lewin-Koh, S. C.; Theil, F. P.; Hop, C. E. Influence of the compound selection process on the performance of human clearance prediction methods. J. Pharm. Sci. 2012, 101, 509−515. (99) Waring, M. J. Defining optimum lipophilicity and molecular weight ranges for drug candidatesMolecular weight dependent lower log D based on permeability. Bioorg. Med. Chem. Lett. 2009, 19, 2844−2851. (100) Comer, J. E. A. High throughput measurements of log D and pKa. Methods Princ. Med. Chem. 2003, 18, 55−59. (101) Martin, Y. A bioavailability score. J. Med. Chem. 2005, 48, 3164−3170. (102) Wenlock, M.; Tomkinson, N. T.; Leeson, P. Importance of physicochemical properties in drug discovery. Med. Chem. Drug Discovery 2012, in press. (103) Geerts, T.; Vander Hyden, Y. In silico predictions of ADME_Tox Predictions: Drug Absorption. Comb. Chem. High Throughput Screening 2011, 14, 339−361. (104) Takagi, T.; Ramachandran, C.; Bermejo, M.; Yamashita, S.; Yu, L. X.; Amidon, G. L. A Provisional biopharmaceutical classification of the top 200 oral drug products in the United States, Great Britain, Spain, and Japan. Mol. Pharmaceutics 2006, 6, 631−643. (105) Hansch, C.; Quinlan, J. E.; Lawrence, G. L. The linear free energy relationship between partition coefficients and the aqueous solubility of organic liquids. J. Org. Chem. 1968, 33, 347−350. Segall, 1205

dx.doi.org/10.1021/mp300476z | Mol. Pharmaceutics 2013, 10, 1191−1206

Molecular Pharmaceutics

Article

Ed.; Adis International Limited: Auckland, New Zealand, 1998; pp 1− 48. (127) Benet, L. Z.; Williams, R. L. Design and optimization of dosage regimens; Pharmacokinetic data. In The Pharmacological Basis of Therapeutics, 8th ed.; Gilman, A. G., Goodman, L. S., Rall, T. W., Nies, A. S., Taylor, P., Eds.; Macmillan Publishing Co.: New York, 1990; pp 1650−1735.

1206

dx.doi.org/10.1021/mp300476z | Mol. Pharmaceutics 2013, 10, 1191−1206