Recent developments in in vitro and in vivo models for improved translation of preclinical pharmacokinetic and pharmacodynamics data

Abstract

Improved PK/PD prediction in early stages of drug development is essential to inform lead optimization strategies and reduce attrition rates. Recently, there have been significant advancements in the development of new in vitro and in vivo strategies to better characterize pharmacokinetic properties and efficacy of drug leads. Herein, we review advances in experimental and mathematical models for clearance predictions, advancements in developing novel tools to capture slowly metabolized drugs, in vivo model developments to capture human etiology for supporting drug development, limitations and gaps in these efforts, and a perspective on the future in the field.

Introduction

In drug metabolism and pharmacokinetics (PK), hepatic metabolism represents a primary focus of drug discovery programs (Zhang and Tang 2018) because the liver is the main site of metabolism for the majority of drugs. More than 90% of the FDA approved drugs are administered through systemic (intravenous and oral) routes (Saravanakumar et al. 2019). Orally administered drugs are absorbed into the portal vein and travel to the liver to undergo first-pass metabolism whereas drugs administered intravenously achieve direct systemic exposure. Understanding the extent of hepatic metabolism and predicting hepatic clearance are therefore pivotal in understanding the trajectory of a drug through the body.

Despite efforts to minimize metabolism during drug design, drug metabolism is still the major clearance pathway of small molecule drugs with cytochrome P450 being the major enzyme class (Cerny 2016; Saravanakumar et al. 2019; Bhutani et al. 2021). Of the 164 small molecule drugs approved by U.S. FDA from 2015 to mid 2020, CYP3A was found to the major eliminating enzyme system (Bhutani et al. 2021). In vitro approaches to determine metabolic clearance have become routine in drug discovery. Standard in vitro models to assess the potential for and extent of hepatic clearance (CL_h) early in drug development include liver microsomes, liver S9 fractions, and hepatocyte suspensions (Richardson et al. 2016; Sodhi and Benet 2021). In vitro intrinsic clearance (CL_{int,in vitro}) data is used to predict in vivo human CL_h and used for designing first-in-human studies. In vitro in vivo extrapolation (IVIVE) is extensively applied in drug discovery and development to understand the relationship between data obtained in vitro (i.e. clearance estimated from hepatocytes or liver fractions) and in vivo (i.e. clearance following an intravenous dose in a rodent) (Poulin P. et al. 2012; Cho et al. 2014).

Mathematical models ranging from simple to complex are often used for IVIVE of hepatic clearance. For example, the role of transporters in drug disposition is becoming increasingly recognized (Zamek-Gliszczynski et al. 2018; Benet et al. 2019). Therefore, estimation of transporter-mediated uptake intrinsic clearance is becoming more routine in drug optimization paradigms, although it should be noted that IVIVE of transporter-mediated clearance remains a challenge. Different in vitro methods for determination of metabolism CL_{int,in vitro} and uptake transporter CL_{int,in vitro} are discussed here with focus on phenomenon of albumin mediated uptake.

Despite a wealth of in vitro data, predicting in vivo metabolism clearance and transporter mediated clearance remains complex, often resulting in under-predictions of in vivo clearance (Wood Francesca Leanne et al. 2017; Chothe et al. 2018; Liang et al. 2020) (Choi et al. 2019). The reasons for this underprediction are still not clear. A number of factors have been suggested which include, but are not limited to, non-specific binding, diffusion barrier in the form of an unstirred water layer (Wood Francesca L et al. 2018), the lack of albumin in hepatocyte and microsomal incubations for both cytochrome and transporter mediated CL_{int,in vitro} estimations (Rowland et al. 2007; Rowland et al. 2008; Bi et al. 2020; Francis et al. 2020), violation of the fundamental assumption of hepatic elimination models (Benet and Sodhi 2020), and to the fact that current approaches do not account for the pharmacokinetic volume of distribution that can vary for each drug (Benet and Sodhi 2020; Sodhi and Benet 2021). For instance, use of albumin in suspended or plated hepatocytes has led to better IVIVE for compounds that are uptake transporter substrates (Kim S-J et al. 2019; Bi et al. 2020; Li N et al. 2021). Another factor that can hamper IVIVE is the inability to determine CL_{int,in vitro} for low clearance compounds. Traditional in vitro systems are limited by their rapid decline in enzyme activity, which renders these systems unsuitable for assays that require longer incubation times (greater than 1 hours for microsomal and 4 hour for hepatocyte suspension). Therefore, calculation of metabolic CL_{int,in vitro} becomes difficult for slowly metabolized compounds. For instance, the lower limit of CL_h estimated from either human liver microsome (HLM) or hepatocyte suspension would be around 6 to 10 mL/min/kg (corresponding to approximately 20% loss of parent compound in the assay), which is about one third of the human hepatic blood flow (Di and Obach 2015), thus making it difficult to accurately measure depletion for slowly metabolized drugs in a conventional in vitro assay. With slowly metabolized compounds on the rise in drug discovery, researchers have utilized different approaches such as calculation of CL_{int,in vitro} from metabolite formation, allometric scaling) or developing different in vitro methods (hepatocyte relay method and systems including co-culture, 3D and flow systems) (Dash et al. 2009; Di et al. 2012; Chan et al. 2013; Ohkura T. et al. 2014; Hultman et al. 2016; Da-Silva et al. 2018), which allow for longer incubation times leading to increased confidence in measuring CL_{int,in vitro} for low clearance compounds (Hultman et al. 2016; Chan et al. 2019; Docci et al. 2019). Herein, the benefits and caveats of applying more recent models such as adding cell culture additives to slow hepatocyte dedifferentiation and 3D hepatocyte cultures to predict hepatic clearance will be discussed.

For in vivo studies in drug development, key considerations include understanding IVIVE for both PK and pharmacodynamics (PD) (Amore et al. 2010), considering potential species-differences in PK (absorption, distribution, metabolism and excretion) (Lin JH 1995; Martignoni et al. 2006; He and Wan 2018) and PD, and the potential impact of disease state on PK and PD outcomes.. Furthermore, choosing the best models for PD can present many challenges (Felmlee et al. 2012). One of the key challenges during drug discovery is development of in vivo animal models which replicate disease conditions in humans accurately (Hansen et al. 2017; Carlessi et al. 2019; Ireson et al. 2019). For instance, development of animal models for drug induced liver injury (DILI) is challenging because of the wide range of mechanisms that can underlie the damage (McGill and Jaeschke 2019; Yokoi and Oda 2021). For example, although there is a not a complete overlap between human and rodent in non-alcoholic steatohepatitis (NASH) pathology, several mouse models are often used to reproduce human NASH pathology which results in lack of clinical translation (Hansen et al. 2017). Some of the recent disease models that were used to assess PK and PD for small molecules are discussed here.

Finally, although much emphasis is put on preclinical animal research, such studies are resource-intensive and by nature, these studies raise ethical concerns (Arora et al. 2011; Freires et al. 2017). Although, animal data remains a requirement for a compound to progress into clinical stages initiatives by policy makers to replace, reduce and refine (3R) the use of animals in drug discovery has led researchers to use alternate in vivo models to study PK/PD. For instance, zebrafish was found to be a good model to test for permeability of small molecules (Kim SS et al. 2017). Herein, we discuss the benefits and limitations of applying alternative models such as zebra fish, invertebrates Drosophila melanogaster (fruit fly) and Caenorhabditis elegans (roundworm) in studying ADME properties during drug development.

In vitro models for pharmacokinetic predictions

Traditionally, metabolism studies have been carried out in microsomes and hepatocyte suspension cultures (Shibata et al. 2002; Ito and Houston 2005; Riley et al. 2005; Brown et al. 2007; Hallifax David et al. 2010). The resulting in vitro measurements can serve for the calculation of the in vivo intrinsic clearance (CL_{int,in vivo}) and CL_h using established scaling factors, such as estimates about microsomal yield from tissue preparation or hepatocellularity (Ito and Houston 2004; Grime et al. 2013). With increasing appreciation of the important, often rate-limiting contributions of hepatic transporter-mediated uptake and excretion (Soars et al. 2007), as well as the considerably higher phase II metabolic activity of hepatocytes compared to microsomes (Stringer et al. 2008), live liver cell models are considered the gold standard rather than subcellular fractions (Chiba et al. 2009).

Importantly however, multiple limitations remain. Firstly, both microsomes and suspension cultures are short-lived in vitro, which complicates CL_{int,in vitro} estimations, particularly for low-turnover compounds. Secondly, while the main routes of biotransformation are essential in understanding drug disposition, quantitative profiles of the circulating and excretory metabolites of a new chemical entity in-vivo cannot be accurately predicted. Thirdly, most suspension experiments use hepatocyte pools from different donors, which are useful to predict population averages, but cannot identify metabolic outliers. In the following section, we will discuss recent advances in tackling these aforementioned issues using 3D human culture systems.

Clearance predictions for low-turnover compounds

An important limitation of both microsomes and suspension cultures remains their short in vitro lifetime, which reduces the confidence in determinations of CL_{int,in vitro} predictions for low-turnover compounds. To overcome this limitation, recent efforts have focused on the use of cultured hepatocytes as they are able to provide longevity of the hepatocytes. It was shown that near-confluent monolayers allow for better retention of hepatocyte cytoarchitecture and activity, cell-cell contacts induce higher levels of phenotypic functions in human hepatocytes than extracellular matrix configuration or composition (Khetani and Bhatia 2008). Early attempts comparing hepatocytes in suspension to cells cultured in 2D monolayers found only minor increases of CL_{int,in vitro} estimates for low turnover compounds in cultured hepatocytes (Griffin and Houston 2005). The low magnitude of effects is likely due to the rapid dedifferentiation of hepatocytes in 2D culture, which starts to become evident as early as 30 min after plating (Rowe et al. 2010; Lauschke, Vorrink, et al. 2016) . Thus, in recent years a variety of strategies have been presented to increase the functional lifespan of hepatocytes in culture.

The predominant methods for the improved maintenance of hepatic phenotypes are 3D culture methods. For a comprehensive evaluation of the available culture paradigms we refer the interested reader to recent systematic reviews on this topic (Hutzler et al. 2015; Lauschke, Hendriks, et al. 2016; Lin C and Khetani 2016; Underhill and Khetani 2018; Lauschke et al. 2019; Prior et al. 2019; Shen et al. 2019; Zhou et al. 2019). For the prediction of low-clearance compounds, micropatterned co-cultures (MPCCs), microfluidic co-culture setups and spheroid systems have been presented.

In MPCCs, hepatocytes are seeded on patches of spotted extracellular matrix, surrounded by supportive stromal fibroblasts (75% surface area as fibroblast and 25% as hepatocytes) (Khetani and Bhatia 2008). This culture configuration provides higher metabolic activity compared to suspended hepatocytes (Kratochwil et al. 2017; Da-Silva et al. 2018) and improves phenotypic stability for multiple weeks, which allows for the use of MPCCs in long-term experiments. Based on these advantages, MPCCs have been extensively used for intrinsic metabolic clearance predictions. Using the well stirred liver model, with and without correction for plasma protein binding, human in vivo clearance could be predicted for 72-76% within 3-fold confidence intervals, which further increased to 92% if correction for protein binding was only incorporated for compounds with reported clearance rates less than or equal to 1 ml/min per kilogram (Lin C et al. 2016; Chan et al. 2019). Notably, the prediction accuracy appeared to be clearance-dependent, with increasing underprediction observed for higher clearance compounds. For low clearance compounds prediction accuracy was 90%, whereas in vivo clearance was generally underpredicted for intermediate clearance compounds if corrected for protein binding (57% accuracy) (Chan et al. 2019). These results are not unique to MPCC in vitro systems, as clearance-dependent underprediction has been noted by multiple labs in traditional microsomal and hepatocyte incubations as well (Wood et al., 2017; “Clearance prediction methodology needs fundamental improvement”). However, predictions were substantially better compared to hepatocyte suspensions and conventional 2D monolayers from the same donor (with the exception of substrates for uptake transporters for which 2D cultures were most appropriate) (Lin C et al. 2016; Umehara K et al. 2020). While absolute clearance estimates for low clearance compounds can be similar between culture methods, the use of MPCC cultures considerably reduced the confidence intervals, which translates into increased confidence for in vivo pharmacokinetic predictions (Docci et al. 2019). MPCC estimates of CL_{int,in vitro} for compounds that are predominantly metabolized by UGTs were highly reproducible (coefficient of variation <30%) and on average significantly higher relative to hepatocyte suspensions. Furthermore, predicted hepatic CL_{int,in vivo} was found to be within 3-fold for 6 out of 13 drugs for suspended hepatocytes whereas it was 9 out of 13 for MPCC (Docci et al. 2020). Typically, a fibroblast-only control culture is run in parallel to determine any fibroblast clearance, which is then subtracted from the CL_{int,in vitro} estimated from MPCC.

Hepatic spheroids consisting of hepatocytes in monoculture or hepatocytes co-cultured with non-parenchymal liver cells constitute a second commonly used system for clearance predictions. While liver spheroids are well established and extensively characterized on transcriptomic, proteomic and metabolomic level (Bell et al. 2016; Bell et al. 2017; Vorrink et al. 2017; Bell et al. 2018; Messner et al. 2018), they have predominantly been used for hepatotoxicity analyses (Mizoi, Hosono, et al. 2020; Miranda et al. 2021), drug target validations, metabolic profiling (see sections below), and studies into their utility for metabolic clearance predictions only begin to emerge (Mizoi, Arakawa, et al. 2020). A pilot study showed that the CL_{int,in vitro} of four low to intermediate clearance compounds could be accurately predicted within 3-fold from in vitro data using as little as 6,000 hepatocytes per measurement (Kanebratt et al. 2021). In contrast, in vivo CL_{int,in vitro} of high clearance compounds was underpredicted. These overall promising data were corroborated in a separate study using a non-overlapping set of seven low and intermediate clearance compounds in which in vivo clearance of all drugs was accurately predicted with an average fold error of 0.7 (Riede J. et al. 2021).

In addition to MPCCs and spheroids, a few other strategies have been presented to improve low clearance evaluations. Using a microfluidic co-culture system of liver cells (HepaRG cells or human hepatocytes) and stromal cells (HμREL system) for clearance predictions of 10 low and intermediate clearance compounds, CL_{int,in vivo} for 70% of compounds could be estimated within 3-fold (average fold error = 2.3), which was comparable to suspension cultures, while having lower inter-experimental variability (Bonn et al. 2016). Hultman et.al. reported improved prediction of CL_int,in-vivo for 14 slowly metabolized drugs using HμREL as compared to hepatocyte suspension (50% and 71% of the predicted values were within 2-fold and 3- fold, respectively) (Hultman et al. 2016). A similar average metabolic activity was observed for 11 enzyme markers (CYP3A, CYP2D6, CYP2C9, CYP2B6, CYP1A2, FMO, AKR, AO, NAT2, UGT1A1, UGT/SULT) in HepaRG (at day 7), HμREL (at day 8), and MPCC (at day 8) as compared to the primary pooled cryopreserved hepatocytes (at day 0) in suspension using different lots. A >10-fold lower metabolic activity was found for both iPSC-derived hepatocyte-like cells (at day 4) and HepG2 (at day 4) (Kratochwil et al. 2017).

Furthermore, conceptually different strategies have been presented based on the supplementation of culture media with cocktails that delay hepatocyte dedifferentiation in 2D culture. HepExtend constitutes a supplement with non-disclosed composition that has been reported to extend the functional life span of hepatocytes in 2D sandwich culture, resulting in prediction of human hepatic in vivo clearance within 3-fold for 83% (10/12) of compounds tested with an average fold error = 2.2 (Lancett et al. 2018). However, more data is needed to make conclusions regarding the added value of this approach, particularly as the use of non-disclosed components hinders the mechanistic understanding.

Combined, the highlighted studies indicate that MPCCs and spheroids can improve clearance prediction of slowly metabolized compared to conventional systems, such as microsomes, hepatocyte suspensions and conventional 2D monolayer cultures. In contrast, average fold errors for microfluidic systems and approaches based on media supplementation appear considerably higher. Interestingly, microsomes, hepatocyte suspensions, as well as MPCCs and spheroids showed an overall tendency to underpredict human in vivo clearance (Wood Francesca Leanne et al. 2017). It should furthermore be noted that the available data for most systems is still limited and even experiments in the same culture system conducted by different groups often differ in methodological details regarding e.g. medium composition, hepatocyte concentrations and culture time, resulting in pronounced differences in predicted CL_{int,in vivo} values between studies (Louisse et al. 2020). Standardization and careful side-by-side benchmarking of the different 3D models to achieve acceptance as an industry standard for clearance predictions is thus imperative.

Elucidating the metabolic routes of new compounds

Besides clearance predictions, 3D liver models have found considerable use for the early identification of human key metabolites of novel compounds. Conventional models, such as liver microsomes, liver S9 fraction and hepatocyte suspension cultures, constitute overall reliable predictors of metabolic routes; however, metabolites formed by phase II conjugation or non-CYP-mediated cleavage reactions were commonly missed (Anderson et al. 2009). Using a test set of 27 drugs corresponding to 56 major metabolites that account for 10% or more of either circulating drug-related material or excreted dose in vivo, only 22 (39%), 26 (46%), and 31 (55%) of these metabolites could be detected in incubations of liver microsomes, liver S9 fraction, and human hepatocyte suspensions, respectively (Dalvie et al. 2009). In contrast, MPCCs identified 38 (68%) and 43 (77%) after 48h and 7 days in culture, respectively, providing a considerable advancement over more traditional methods (Wang WW et al. 2010). More recently, MPCC from different species used to characterize metabolism of TAK-041 accurately reproduced metabolites observed in vivo in rat, dog, monkey and human, although more metabolites were detected in vitro in MPCC incubations than were observed in vivo (add the reference here). For instance, only 1 out of 8 in vivo observed metabolites were observed in human hepatocyte suspension whereas all were observed in human MPCC (Kamel et al. 2020).

In liver spheroids, metabolite identification data is only available for small sets of candidate molecules. In one study, spheroids formed both phase II and phase II human-specific metabolites of acetaminophen, diclofenac, midazolam, propranolol, lamotrigine and salbutamol (Ohkura Takako et al. 2014). Results were particularly promising for salbutamol for which salbutamol-4-O-sulfate constituted the main metabolite in spheroids but was not detected in microsomes, S9 fraction, or hepatocyte suspension cultures, and was not predominant in in vivo rat studies. Furthermore, spheroids identified 14 out of 16 major metabolites across these six tested drugs (88%) after 72h of incubation, including demethylation, hydroxylation, glucuronidation and deethylation reactions (Kanebratt et al. 2021).

Thus, based on the available data, MPCCs constitute the most well characterized system for metabolite identification. While promising data for spheroids have been presented, further benchmarking data is needed to make conclusions about their utility for the identification of circulating metabolites. Furthermore, data for all systems remains qualitative in nature and accurate quantitative predictions using in vitro data have so far not been presented.

Modeling the effect of genetic variability on hepatic drug metabolism

Hepatocytes for metabolic studies are often pooled in order to increase the confidence that the obtained data approximates the average of the general population rather than a single donor. While this strategy is useful in early drug discovery to discern average metabolic differences between compounds, it can also be beneficial to identify and mechanistically understand outlier samples.

Overall variability between different donors can be considerable; for instance comparison of the CL_{int,in vitro} in spheroids and MPCCs of three different donors showed inter-donor variability of up to 6-fold for selected compounds (Chan et al. 2013; Kanebratt et al. 2021). These observations mirror extensive inter-individual differences in vivo where metabolic activity of relevant drug metabolizing enzymes, such as CYP3A4 and CYP2D6, can differ up to 100-fold in large biobanks (Westlind-Johnsson et al. 2003; Frederiksen et al. 2021). These differences can be caused by physiological, environmental and genetic factors, which can be mimicked in 3D culture models of primary human liver cells, thus allowing the study the of inter-individual differences between patients’ livers and their effects on drug pharmacokinetics (Ingelman-Sundberg and Lauschke 2018). For instance, dextromethorphan is primarily metabolized by CYP2D6 (Km = 3.7μM) with only minor contributions by the CYP3A enzyme system (Km = 223 μM). Indeed, in spheroids established from cells of normal CYP2D6 metabolizers, the CYP2D6 specific metabolite dextrorphan was most abundant (Vorrink et al. 2017). However, in spheroids from poor CYP2D6 metabolizers, the CYP3A4 metabolite 3-methoxymorphinan was most abundant, thus demonstrating the utility of the system to quantitatively model metabolic fluxes as a function of genotype. While similar findings have to the best of our knowledge not been presented in other 3D liver models of primary human cells, it can be assumed that those might be capable of recapitulating genetically encoded differences in hepatic metabolism.

In vivo models

This section will serve to appreciate the recent advancements in animal model experimentation for PK/PD studies in various therapeutic areas as well as discuss some opportunities and challenges related to the design of in vivo studies.

Drug metabolism

The cytochrome P450 superfamily is involved in numerous oxidative reactions and plays a key role in the metabolism of drugs and other xenobiotics (Bogaards et al. 2000). One major challenge in drug development is to determine which animal species best resembles the drug metabolism capabilities of humans necessary for the design of translational PK/PD studies (Turpeinen et al. 2007).

Neyshaburinezhad et al. recently studied the changes in liver enzymatic activity of CYP2D1 and its related hepatic clearance in a type I and II diabetes rat model after treatment with insulin and metformin. The post-treatment effects of insulin and metformin administration on CYP2D1 activity and hepatic clearance were examined by measuring changes in the ratio of dextrorphan to dextromethorphan in a type I and II diabetic rat perfused liver model. For induction of type I diabetes, 65 mg/kg streptozotocin (STZ) was injected intraperitoneally overnight to fasted male Sprague-Dawley rats (Neyshaburinezhad et al. 2020). STZ damages the Langerhans islets β cells, resulting in hypoinsulinemia and hyperglycemia (Shah and Smith 2015). Type II diabetes was induced by intraperitoneal injection of 110 mg/kg nicotinamide followed by intraperitoneal injection of streptozotocin. Nicotinamide is methylated in vivo by nicotinamide N-methyltransferase into N¹-methylnicotinamide which is associated with oxidative stress and insulin resistance at high concentrations (Országhová et al. 2012). Treatment was initiated one week after disease induction to maximize the potential effects of diabetes on increasing levels of proinflammatory cytokines, and therefore, CYP2D1 activity (Neyshaburinezhad et al. 2020). It was found that type I diabetes significantly decreased the activity of CYP2D1 (i.e, significant decrease in mean metabolic ratios of the untreated rats compared to the control group). Although mean metabolic ratios decreased in type II diabetic rats, the difference was not statistically significant. The study also demonstrated a decrease in hepatic clearance of dextromethorphan in type I and II diabetic rats.

Lu et al. recently generated a cytochrome P450 2J3/10 CRISPR-Cas9 KO rat model to study its function in vivo (Lu et al. 2020). The CYP2J3/10 KO rat model was developed using the sequence fragments of CYP2J3 and CYP2J10, the orthologous genes of CYP2J2 in humans to get target sites. The deficiency of metabolic function of CYP2J was confirmed by evaluating astemizole metabolism both in vitro and in vivo. The rat liver microsome incubations and in vivo studies indicated that the metabolic function of CYP2J3/10 was decreased in KO rats. There was no effect on serum protein, alkaline phosphatase, transaminase, bilirubin, triglyceride, and cholesterol levels. However, myocardial enzyme creatinine kinase (CK), creatine kinase–muscle brain type (CK-MB), and their ratio (CK-MB/CK) was significantly increased by 140%, 80%, and 60%, respectively suggesting that there may be myocardial injury in CYP2J3/10 KO rats (Lu et al. 2020). These results are not surprising as CYP2J2 has been recognized for its endogenous cardioprotective role of bioactivating the pro-inflammatory arachidonic acid into anti-inflammatory EETs (Askari et al. 2013; Aliwarga et al. 2018).

Miura et al. recently investigated the respective roles of cytochromes P450 2C9 and 3A in the oxidation of (S)-warfarin and diclofenac in a chimeric mice model transplanted with human hepatocytes (Miura et al. 2019). The targeted expression of the herpes simplex virus type 1 thymidine kinase in the liver of severely immunodeficient mice enabled the mice livers to be replaced with mature and functional human liver tissue (Hasegawa et al. 2011). The humanized-liver mice were orally pretreated with 15 mg/kg tienilic acid to metabolically inactivate CYP2C9 and then treated with a single dose of 0.50 mg/kg (S)-warfarin or 10 mg/kg diclofenac. (S)-warfarin 7’-hydroxylation was significantly reduced in the CYP2C9-inactivated mice as demonstrated by a decrease in 7-hydroxywarfarin C_max and AUC_infinity of 22 and 16%, respectively. This study suggests that humanized-liver mice pretreated with tienilic acid could serve as a potential in vivo model for metabolically inactivated CYP2C9. Such models could be useful to determine f_m (fraction metabolized by each CYP isoform) of victim drugs in vivo (Miura et al. 2019).

Oncology

Rodents have been extensively used to facilitate the pharmacological evaluation of existing and potential new medicines in oncology (Ireson et al. 2019). However, despite numerous successes in drug development, the likelihood of approval for oncology drugs is lower than for those in any other therapeutic area because these models lack the predictive power required to translate preclinical efficacy into clinical activity (Gould et al. 2015).

Tada et al. recently assessed the PK of 5-fluorouracil after hepatectomy in a colorectal liver metastasis rat model to evaluate correlation between liver dihydropyridine dehydrogenase, which is main enzyme catabolizing 5-fluorouracil (5-FU), and 5-FU toxicity (Tada et al. 2020). The model consists of male Wistar rats undergoing a hepatectomy by which the median and left lateral hepatic lobes were removed, as described previously (Martins et al. 2008; Komori et al. 2014). Following this procedure, a central venous catheter insertion was performed. Then, the neck was surgically dissected, and a polyurethane catheter was inserted in the right external jugular vein and advanced into the superior vena cava. A PinPort™ was set to the edge of the catheter to allow access to the vein. The administration of 5-fluorouracil was performed four days after the hepatectomy. It was concluded that 5-FU dose should be reduced for patients undergoing major hepatectomy, because of the possibility of increased 5-FU toxicity due to reduction of dihydropyrimidine dehydrogenase (Tada et al. 2020).

Impaired renal and hepatic function

Alkharfy et al. recently studied the effects of compromised liver function on the PK of thymoquinone in a Wistar rat model (Alkharfy et al. 2020). Liver impairment was induced with a single intraperitoneal injection of 800 mg/kg d-galactosamine. Galactosamine is a potent hepatotoxic substance that induces both hepatocyte necrosis and apoptosis by inhibiting hepatic RNA synthesis via the production of uridine diphosphate hexosamines, limiting DNA transcription (Apte 2014; Saracyn et al. 2015). Treatments with D-galactosamine and gentamicin were able to produce significant hepatic and kidney injury respectively which was confirmed by a significant increase in systemic biomarkers (AST and ALT levels for liver and Scr and BUN for kidney) as compared to control rats (Alkharfy et al. 2020).

Li et al. recently studied the PK of clozapine and norclozapine in a non-alcoholic fatty liver disease (NAFLD) rat model (Li Z et al. 2021). To evaluate the possible PK changes of clozapine and its metabolite in the early stages of NAFLD, a rat model of the disease was induced by a diet containing 1% orotic acid. In vivo experiments with rats reveal that the mechanisms by which orotic acid-supplemented diet induces liver steatosis include the activation of the sterol regulatory element binding protein-1c (SREBP1c) (Matilainen et al. 2020). SREBP-1c, a member of the family of SREBP membrane-bound transcription factors, has been established as the principal regulator of hepatic fatty acid biosynthesis. Thus, an increase in SREBP-1c expression increases the rate of lipogenesis in the liver (Wang Y-M et al. 2011). The study showed a reduction of CYP1A1/2 activity leading to a significantly slower in vitro hepatic microsomal CL_int of clozapine. As a result, the AUC of norclozapine and the AUC_norclozapine/AUC_clozapine significantly increased. However, systemic exposures to clozapine after intravenous and oral administration were comparable with the control group. Furthermore, steady-state brain concentrations of the parent drug and metabolite were significantly higher in NAFLD rats compared to the control group.

Toth et al. recently studied the interaction of OATP1B2 expression and NASH on pravastatin plasma clearance (Toth et al. 2020). Four months of age male C57BL/6 wild-type Oatp1b2^+/−, and Oatp1b2^−/− were fed either a methionine and choline sufficient diet as a control or a methionine and choline-deficient diet to induce NASH. After 6 weeks of feeding, 10 mg/kg/10 mL pravastatin was injected into the carotid artery of the study animals. The study demonstrated the synergetic interaction between diet-induced NASH and the genetic knock-out of Oatp1b2, which results in a 4.4-fold increase in AUC values compared to the C57BL/6 wild-type and Oatp1b2^+/− mice. This suggests that only a single allele of the Oatp1b2 gene is needed to avoid supratherapeutic plasma concentrations of pravastatin.

Fang et al. recently studied the PK/PD of beinaglutide, a recombinant human glucagon-like peptide-1 (GLP-1) acid, in a NASH mouse model (Fang et al. 2021). Seven-week old male mice homozygous for the obese spontaneous mutation (ob/ob) were fed the Gubra-Amylin NASH diet which is comprised of high fat (40% total fat kcal), high fructose (22% by weight), and high cholesterol (2% by weight) for up to 9 weeks. The disease was confirmed by biopsy and display hallmarks of NASH. Body weight, liver mass, plasma ALT and AST levels were significantly reduced following treatment with Beinaglutide. Beinaglutide doses ranging from 0.6 mg/kg to 2.4 mg/kg were subcutaneously administered thrice daily for a four-week treatment period. Beinaglutide was shown to dose-dependently reduce blood glucose and stimulate insulin secretion. It also significantly reduced body weight, liver mass, as well as intrahepatic lipid as opposed to the control group.

Jiang et al. (Jiang et al. 2020) recently studied the PK/PD of diclofenac and its metabolites in a drug-induced liver injury mice model. TgCYP3A4/hPXR-humanized mice were used to mimic the clinical metabolism of diclofenac and explore the possible mechanism for acute hepatotoxicity. To induce liver injury, 80 mg/kg diclofenac, 80 mg/kg 4′-hydroxy-diclofenac, 80 mg/kg 5-hydroxy-diclofenac, and 80 mg/kg diclofenac glucuronide were intraperitoneally administrated into the study rodents. Diclofenac, 5-hydoxy-diclofenac, and diclofenac glucuronide significantly elevated serum alanine aminotransferase concentrations while the effect of 4’-hydroxy-diclofenac was not significant. Diclofenac and its metabolites were also shown to increase serum cytokine levels including IL-1β, IL-6, IL-12, and IL-17, with diclofenac glucuronide being the most involved in the activation of the hepatic immune system and the pathogenesis of diclofenac-induced liver injury.

Opportunities and challenges

Dramatically rising costs and extremely high failure rates in drug development have led many to re-evaluate the value of animal studies (Van Norman 2019). The core of the problem may be rooted in animal modeling itself. Unlike in human clinical trials, no best-practice standards exist for animal testing (Mak et al. 2014; Johnson 2020). There are several opportunities in which preclinical studies could improve to increase translational success rates. A critical step in preclinical development is the conduct of well-designed studies to establish the PK/PD relationship in the animal model so it can be scaled to humans (Gabrielsson Johan et al. 2009). Understanding the relationships between PK and PD, as well as the potential lack of concordance between the two, will improve the interpretation of resulting data in the successful prediction of the human relationship (Gabrielsson J. and Green 2009). Additionally, studies underline the importance of biomarker quality to predict therapeutic success and translatability (Wendler and Wehling 2012). As science advances, newer cutting-edge biomarkers are emerging and being developed for drug development. These include gene expression profiling, imaging biomarkers, and proteomic biomarkers (Bai et al. 2011). Cohen et al. (Cohen 2008) also noted that the concentrations of an active compound and its metabolites are an underused translational biomarker in animal studies that deal with the mechanism of action of the drug. This represents a missed opportunity to relate the results in other species and to plan the dosage in humans (Gabrielsson Johan et al. 2010). Failure to integrate sex as a variable in preclinical studies may contribute to failing clinical trials. Indeed, preclinical studies are often biased by conducting the study only with male models to avoid the complications and variability associated with the estrous cycle, while clinical trials include both men and women (Lee et al. 2018). Strict measures are needed to correct the sex bias seen in animal research. For instance, journal editors and reviewers should enforce authors to highlight the use of only one sex in the title of the article as this would encourage researchers to balance the sex distribution of animals to be used (Zucker and Beery 2010).

Alternative models

The past decades have seen dramatic advances in in vitro preclinical models, in vitro to in vivo modeling, and in silico predictors, but rodent models remain the gold standard for pharmaceutical discovery and development. However, the use of rodents in biomedical research has been the subject of public scrutiny and objections from within the scientific community for years. Partly, this is due to the lack of concordance of rodent studies with clinical trial outcomes (Pound et al. 2004; Perel et al. 2007; van der Worp et al. 2010), calling into question to contribution of experimental design flaws (Hackam 2007; Kilkenny et al. 2009) and/or a lack of biological translation from rodents to humans (Couzin-Frankel 2013; Seok et al. 2013).

The concept of replacement (from the 3Rs) includes partial replacement, which is based on the idea that some animals, based on current scientific thinking, are not considered capable of experiencing distress/suffering (Tannenbaum and Bennett 2015). This includes invertebrates such as Drosophila melanogaster and Caenorhabditis elegans and immature forms of vertebrates such as Danio rerio embryos (Kendall et al. 2018). These models may provide opportunities to identify toxicological liabilities before testing in rodents and have also been considered for PK applications.

Use of Zebrafish (Danio rerio) in Drug Discovery

Zebrafish have been steadily gaining popularity as a model organism, and have several advantages for drug discovery (Rennekamp and Peterson 2015). Much of the biology is conserved with higher vertebrates, including homology to many human drug-metabolizing enzymes (van Wijk Rob C. et al. 2016). It is estimated that 70% of human genes have orthologous zebrafish genes (while 83% of human genes have mouse orthologs) (Howe Clark, et al. 2013). Furthermore, the ease and low cost of husbandry, high fecundity, small size, rapid development, and transparent embryos make screening in these models highly feasible. Embryos are not regulated as animal studies up to 120 hours postfertilization (hpf) and have fully developed liver and gut organs by 76 hpf (Strähle et al. 2012; van Wijk Rob C. et al. 2016); for these reasons, many chemical screens utilize embryos.

In spite of these advantages, several limitations must be overcome to fully integrate zebrafish into the PK aspect of the drug discovery pipeline. First and perhaps most importantly, almost all zebrafish pharmacological and toxicological screens have failed to measure internal exposure concentrations, which are key to deriving the pharmacokinetic parameters needed to translate findings from zebrafish to humans. This has been and continues to be problematic due to technical difficulties in measuring drug and metabolite concentrations in such small organisms (Berghmans et al. 2008; Rennekamp and Peterson 2015). Second, there is a lack of essential parameters needed for refinement of physiologically-based pharmacokinetic (PBPK) models (Khazaee and Ng 2018). Third, although efforts have been made to identify zebrafish orthologs of human drug-metabolizing enzymes (McGrath and Li 2008), especially cytochromes P450 (Goldstone et al. 2010), many ortholog identifications (particularly for phase II conjugative enzymes) have been limited to sequence alignments, which may fail to identify overlapping substrate specificities (van Wijk Rob C. et al. 2016).

However, despite these challenges, the first zebrafish PK model was created for acetaminophen in 2016 and scaled to 12 higher vertebrate species (Kantae et al. 2016), with a recent update to include acetaminophen and major metabolites measurements in nanoscale blood sampling (Van Wijk R. C. et al. 2019). Newer studies have built upon the acetaminophen work to include PK models for isoniazid (van Wijk R. C. et al. 2020) and valproic acid analogs (Siméon et al. 2020). To evaluate the differences in the CYP-mediated drug metabolism between zebrafish and human, in vitro assays were performed using the adult male and female zebrafish liver microsomes (ZLM), whole embryo microsomes at different developmental stages and HLMs. Four model substrates (dextromethorphan, diclofenac, testosterone and midazolam) were evaluated (Saad et al. 2017). Diclofenac was found to showed similar production of main metabolites at similar ratios in both ZLMs and HLMs, whereas dextromethorphan showed similar metabolites but different metabolic ratios. For testosterone, completely different metabolites were formed whereas midazolam was not found to be metabolized in ZLMs (Saad et al. 2017). This result was similar to previously reported study where it was found that testosterone metabolism was different in ZLMs and HLMs (Chng et al. 2012).

The differences in metabolism in ZLM and HLM prevent extrapolation of in vitro data from zebra fish to human. To circumvent this limitation, ‘humanized zebrafish’ expressing human CYP3A4 were developed. Although the ‘humanized zebrafish’ showed increased metabolism as compared to wild type, comparative analysis with microsomes from ‘humanized zebrafish’ was not compared (Poon et al. 2017). As efforts continue in this direction, the quality of the models and translation to higher mammals is expected to continue to improve.

Another emerging area of application of the zebrafish model in drug discovery is in microbiome-drug interactions. Zebrafish can be maintained in axenic states and the desired microbiome introduced (Bertotto et al. 2020). The model has already been applied in the context of environmental toxicology, including for the pesticide chlorpyrifos (Wang X et al. 2019), the antibacterial/antifungal agent triclosan (Gaulke et al. 2016), and for microplastics (Jin et al. 2018; Qiao et al. 2019). Drug-microbiota effects have also begun to be investigated in the model, including for sinomenine/morphine (Chen et al. 2020) and for antibiotics (Almeida et al. 2019).

Invertebrate models for PK studies

The concept of partial replacement also includes invertebrates at all life stages. Among the invertebrate models, Drosophila melanogaster (fruit fly) and Caenorhabditis elegans (roundworm) have received the most attention for drug discovery applications (Willoughby et al. 2013; Ali et al. 2018; Maitra and Ciesla 2019). This is due to significant advantages in these models (reviewed in (Pandey U. B. and Nichols C. D. 2011; Apfeld and Alper 2018)), including a high degree of genetic conservation (estimated ~75% genetic similarity with humans), fast and inexpensive culture in the laboratory, rapid propagation, short lifespan, the availability of many genetic mutants and reporter strains, and the ability to easily make new genetic modifications using CRISPR-Cas9.

In addition to significant advantages, there are also several limitations that make translation of invertebrate studies to humans challenging. Most obviously, the physiology is quite different between vertebrate (particularly mammalian) species and invertebrates. Both C. elegans and D. melanogaster lack many of organs present in mammals, such as the liver, where drug metabolism primarily takes place in mammals. Instead, C. elegans has liver-like functions mainly in the intestine (Dimov and Maduro 2019), while D. melanogaster liver-like functions are carried out by cells in the fat body and oenocytes (Gutierrez et al. 2007). Both invertebrate species also lack an adaptive immune response, which does limit immunological studies but does allow for assessing the conserved innate immune response in the absence of the adaptive response (Ermolaeva and Schumacher 2014; Maitra and Ciesla 2019). In addition to these physiological differences, simply the small size of both C. elegans and D. melanogaster makes measuring internal concentrations of drugs and metabolites difficult; therefore, the vast majority of studies in both species report external doses of the parent drug only. This is a significant limitation, perhaps especially in C. elegans, which has significant barriers to uptake and internal concentrations are often only 1% or less of the external dose (Hartman J. H. et al. 2021). Finally, although orthologs of many mammalian xenobiotic metabolizing enzymes have been identified in both species, including cytochromes P450 (Menzel et al. 2001; Chung et al. 2009) and UDP-glucuronosyltransferases (Luque and O’Reilly 2002; Fontaine and Choe 2018), the vast majority of xenobiotic-metabolizing enzymes from both species have not been biochemically characterized for substrate specificity, which makes their relevance to mammalian drug metabolism unclear.

To date, C. elegans has been applied to drug screening efforts, and in two studies of metal salts, C. elegans mortality correlated with rat and mouse oral LD₅₀ (Williams and Dusenbery 1988). Furthermore, in wider organic chemical screens, C. elegans larval growth predicted rabbit or rat developmental toxicity with an accuracy around ~53% (whereas the concordance between rat and rabbit is 58%) (Boyd et al. 2010; Boyd et al. 2016). It has been suggested that this predictivity could be improved by combining multiple endpoints in C. elegans (Hunt 2017). The worm has also been used to investigate metabolism of fluoropyrimidine cancer drugs by the gut microbiome (Hunt et al. 2012; Scott et al. 2017). The C. elegans model has been gaining popularity in microbiome research (Hartman Jessica H et al. 2021) due to the ease of producing axenic animals and introducing the desired microbiome by bacterial feeding. Recently a toolkit has been published to study microbiome effects quickly and reproducibly in the model (Dirksen et al. 2020).

Some labs have taken advantage of the lack of some orthologous xenobiotic metabolizing enzymes in C. elegans (e.g. CYP1A and CYP2E1) to engineer humanized worm strains that express vertebrate cytochrome P450 enzymes (Harris et al. 2020). This approach could allow for assessment of metabolic liabilities in a medium throughput manner as part of a drug development pipeline, but so far this approach has only been applied for a limited number of P450s and xenobiotic compounds.

D. melanogaster have also been used for drug development efforts, and recently have been proposed as a new model to study solute carrier membrane transporters (Wang Y et al. 2018). D. melanogaster was used to identify genes associated with carboplatin toxicity (King et al. 2014), although its relevance to human chemotherapeutic efficacy has not been established. This model has also been applied to drug screening efforts, as reviewed elsewhere (Pandey Udai Bhan and Nichols Charles D 2011; Strange 2016; Wang Y et al. 2018). Drosophila melanogaster has also been proposed as a model for microbial drug metabolism, and relatively straightforward protocols have been established for experimentally manipulating the microbiome of the flies (Douglas 2018), although there has not been widespread use of the model for this purpose to date. Overall, the use of invertebrate models in drug discovery is an emerging area that holds promise but requires more detailed studies to assess translatability to mammalian species.

IVIVE

Methods for Metabolic CL_{int,in vitro} Determination

Hepatic clearance is routinely predicted in drug discovery efforts by experimentally determining an in vitro intrinsic clearance (CL_{int,in vitro}), converting that measurement to a prediction of in vivo intrinsic clearance (CL_{int,in vivo}), then applying a model of hepatic disposition to predict hepatic clearance (CL_h, and this process is commonly referred to as in vitro to in vivo extrapolation (IVIVE) (Sodhi and Benet 2021). CL_{int,in vitro} is typically obtained by using the in vitro half-life (T_1/2) method in which drug is incubated with the preferred enzyme source (typically recombinant enzymes, microsomes, cytosol, S9 or hepatocytes) at a low substrate concentration (less than K_m) and the depletion of substrate concentration is measured over time (Obach et al., 1997). By incorporating fraction unbound in the incubation (f_u,inc), the first order rate constant for unbound substrate depletion (k_inc,u) is measured as negative of the slope of natural log percent remaining versus incubation time plot. With consideration of amount of enzymes or cells in the incubation and the volume of the incubation (V_inc), CL_{int,in vitro} is calculated as follows:

$$C{L}_{int,invitro}=k_{inc,u}\cdot \frac{V_{inc}}{amountenzymesorcells-invitroincubation}$$ (1)

The CL_{int,in vitro} can then be scaled to an estimate of CL_{int,in vivo} by accounting for the differences in amount of enzymes or cells in the in vitro incubation to that of an average liver:

$$C{L}_{int,invivo}=\frac{amountenzymeorcells-wholeliver}{amountenzymesorcells-invitroincubation}\cdot k_{inc,u}\cdot V_{inc}$$ (2)

This is achieved by using physiological scaling factors (microsomal protein per gram liver (MPPGL), cytosolic protein per gram liver (CPPGL), S9 protein per gram liver (S9PPGL) or hepatocellularity per gram liver (HPGL)) depending on which in vitro system is used, as well as the liver weight per kg body weight. The values for all these physiological scaling factors can be found in the literature. For human CL_{int,in vivo} predictions, values used for MPPGL range from 32 to 77 mg microsomal protein per gram liver (Wilson et al. 2003; Hakooz et al. 2006; Barter et al. 2007; Doerksen et al. 2020) (Naritomi et al., 2001), and values for HPGL range from 99 to 139 million hepatocytes per gram liver (Wilson et al. 2003; Sohlenius-Sternbeck 2006; Barter et al. 2007), while the commonly used value of human liver weight per kg body weight ranges from 21.4 to 25.7 g liver/kg body weight (Davis et al., 1993) (Pelkonen and Turpeinen 2007).

To enable predictions of total hepatic clearance, the physiological limitations of organ blood flow (Q_H) and fraction unbound in the blood (f_u,B) are considered in tandem with predictions of CL_{int,in vivo} by applying a model of hepatic disposition. Since hepatic clearance measurements must be conducted using systemic concentrations (without knowing the unmeasurable intrahepatic concentrations driving elimination), the field of pharmacokinetics relied on chemical engineering reactor models (for which only entering and exiting reactant amounts are known without any measurements from within the reactor) to estimate hepatic clearance. The most commonly used hepatic disposition models include the well-stirred model (WSM), the parallel tube model (PTM), and the dispersion model. Each hepatic disposition model assumes a different degree of hepatic mixing of drug, and thus a different driving force unbound concentration for hepatic elimination. The simplest model, the WSM, assumes that drug is homogeneously distributed throughout the liver due to infinite mixing, while the PTM assumes first-order incremental metabolism throughout the liver (zero mixing), and there are an infinite number of dispersion models between these two boundary models that are defined by different dispersion numbers ranging from zero to infinity. Although the WSM is not physiologic, it has been extensively used because of its simple mathematical manipulation (Pang et al., 2019) and was first derived in the 1970s (Rowland et al., 1972) (Wilkinson and Shand, 1975) as the following relationship:

$$C{L}_{H,WSM}=\frac{Q_H\cdot f_{u,B}\cdot C{L}_{int}}{Q_H+f_{u,B}\cdot C{L}_{int}}$$ (3)

Of note, it has recently been noted that the unphysiologic WSM can adequately describe all experimental isolated perfused rat liver studies with reasonable experimental conditions (Sodhi, Wang, et al. 2020).

Although the process of predicting in vivo hepatic clearance from in vitro measures of drug metabolism in liver tissue appears straightforward, the accurate prediction of clearance using IVIVE continues to be a significant challenge (Chiba et al. 2009; Wood Francesca Leanne et al. 2017; Benet and Sodhi 2020; Sodhi and Benet 2021). It has been consistently observed throughout the field that scaling CL_{int,in vitro} to CL_{int,in vivo} using physiological scaling factors often results in a marked underprediction for compounds that are primarily metabolized, in both preclinical species (rat) and humans with both microsomes and hepatocytes (Shibata et al. 2002; Hallifax David et al. 2005; Ito and Houston 2005; Chiba et al. 2009; Hallifax D and Houston 2009; Hallifax David et al. 2010; Wood Francesca Leanne et al. 2017; Yamagata et al. 2017). Moreover, the underprediction has also been observed to be clearance-dependent, with negative bias increasing with increasing CL_{int,in vitro} and/or total CL_h (Hallifax David et al. 2005; Hallifax D and Houston 2009; Hallifax David et al. 2010; Wood Francesca Leanne et al. 2017; Hallifax David and Houston 2019; Benet and Sodhi 2020). It has been reported that the human microsomal underprediction of CL_{int,in vivo} was 2.8-fold and in hepatocytes was 4.2-fold, with similar findings in rats (Wood Francesca Leanne et al. 2017). Recently, it has been noted that the differences in CL_{int,in vitro} between microsomes and hepatocytes is more pronounced for drugs that display low permeability characteristics, with microsomal CL_{int,in vitro} values being larger than that of hepatocytes (Keefer et al. 2020). These authors also noted that there was no correlation between the extent of difference in CL_{int,in vitro} of microsomes versus hepatocytes and P-glycoprotein (P-gp) efflux ratio, concluding that P-gp may not be active in hepatocyte suspension assays (Keefer et al. 2020). In another recent study, the disconnect between human microsomal and hepatocyte CL_{int,in vitro} values was found to be more profound for CYP3A4 substrates (Williamson et al. 2020), suggesting the potential for enzyme-transporter interplay in hepatocyte incubations due to the overlapping substrate specificities between CYP3A4 and P-gp. These results emphasize the need to better understand the limitations of in vitro systems utilized for clearance predictions, as well as the critical importance of considering membrane passage intrinsic clearances in clearance prediction determinations. Furthermore, systematic differences were observed in donor-matched samples of HLM and hepatocytes suggesting inter-individual differences might not be a sole reason for difference between microsomal and hepatic CL_{int,in vitro} (Wegler et al. 2021).

As current IVIVE methodologies cannot reliably quantitatively predict in vivo clearance, IVIVE provides a practical approach for drug discovery scientists to rank-order metabolic stability of compounds in lead optimization efforts. Often, empirical scaling factors (ESFs) are applied to in vitro data in order to more accurately predict clearance (Ito et al., 2005), (Sohlenius-Sternbeck et al., 2012), however, such approaches are often biased as the ESF is highly dependent on dataset used to generate the scaling factor. Therefore, such approaches may be useful when predicting clearance within a series of compounds that are structurally similar in lead optimization efforts, however, prospectively predicting in vivo clearance for new chemical entities of diverse chemical compositions poses challenge.

It has been noted that higher ESF values were observed for drugs with higher CL_{int in vivo} values both in for human and rat hepatocytes and microsomes. The average ESF ranged from 0.61 (for observed CL_{int,in vivo} of <10ml/min/kg) to 1200 (for observed CL_{int,in vivo} of >10,000 ml/min/kg) for human hepatocytes, whereas for HLM the average ESF ranged from 0.7 (for observed CL_{int,in vivo} of <10ml/min/kg) to 58 (for observed CL_{int,in vivo} of >10,000 ml/min/kg). Similarly, the average ESF ranged from 0.13 (for observed CL_{int,in vivo} of <10ml/min/kg) to 180 (for observed CL_{int,in vivo} of >10,000 ml/min/kg) for rat hepatocytes whereas for rat microsomes the average ESF ranged from 0.086 (for observed CL_{int,in vivo} of <10ml/min/kg) to 230 (for observed CL_{int,in vivo} of >10,000 ml/min/kg) (Wood Francesca Leanne et al. 2017). This suggest that a single ESFs cannot be utilized to ‘correct’ the IVIVE underprediction bias, and these observations confirm the clearance-dependent underprediction trends observed by the field (as mentioned above). Since these quantitative predictions did not show any bias between cryopreserved and fresh hepatocytes (McGinnity et al. 2004; Floby et al. 2009) and the systemic underprediction of CL was observed in both species, the underprediction of clearance was considered to be species and system independent. Segregated ESFs (ESFs based on prediction data segregated according to a particular range of CL_{int,in vitro} values) were proposed to remove clearance dependence (Hallifax David and Houston 2019). A two-step scaling approach to predict human clearance from rat was also proposed wherein in first step each human segregated CL_{int,in vitro} level was corrected with the corresponding rat segregated ESF. Then in second step each rat segregated ESF was multiplied by the ratio of log average human ESF to log average rat ESF. These two approaches (use of human segregated ESF approach and two step scaling approach) gave comparable predictions (Hallifax David and Houston 2019).

Conventional models have also been modified to account for phenomena such as pH partitioning and albumin-facilitated uptake (Table 1). Berezhkovskiy (Berezhkovskiy 2011) suggested that due to pH differences between plasma (or extracellular water) and intracellular water, the unbound drug concentration in plasma and intracellular liver are not equal for ionizable compounds. Berezhkovskiy proposed a modification in the conventional model to account for the fraction ionized (ionized correction model in table 1). Although, the average fold error is comparable to the traditional equation (average fold error was 0.88 and 0.90 for conventional equation and modified equation respectively) a considerable change of human plasma and human tissue concentration time profiles for 24 drugs was predicted (Berezhkovskiy 2011).

Table 1.

Clearance prediction models

Prediction Method	Model	References
Conventional IVIVE	$CL=\frac{Q_{liver}{R}_{BP}C{L}_{int}\frac{f_{up}}{f_{uinc}}}{Q_{liver}+R_{BP}C{L}_{int}\frac{f_{up}}{f_{uinc}}}$	(Obach 1999)
Ionization correction model	$CL=\frac{Q_{liver}{R}_{BP}C{L}_{int}\frac{f_{up}{F}_I}{f_{uinc}}}{Q_{liver}+R_{BP}C{L}_{int}\frac{f_{up}{F}_I}{f_{uinc}}}$ Where $F_I=\frac{f_{uion,plasma}}{f_{uion,IW}}$ and $f_{uion}=\frac{1}{1+{10}^{(pH-pKa)}}$ (for acidic compounds) OR FI = 1 (for neutral compounds) OR $f_{uion}=\frac{1}{1+{10}^{(pKa-pH)}}$ (for basic compounds)	(Berezhkovskiy 2011)
fuliv method	$CL=\frac{Q_{liver}{R}_{BP}C{L}_{int}\frac{f_{u,liver}}{f_{uinc}}}{Q_{liver}+R_{BP}C{L}_{int}\frac{f_{u,liver}}{f_{uinc}}}$ Where $f_{u,liver}=\frac{PLR{f}_{up}{F}_I}{1+(PLR-1)f_{up}{F}_I}$ Where PLR is plasma to whole liver ration of either albumin or alpha acid glycoprotein	(Poulin Patrick, Hop Cornelis ECA, et al. 2012; Poulin Patrick, Kenny Jane R, et al. 2012)
Extended clearance model	$CL=\frac{Q_{liver}{R}_{BP}C{L}_{int,u,all}{f}_{up}}{Q_{liver}+R_{BP}C{L}_{int,u}{f}_{up}}$ where $C{L}_{int,u,all}=\frac{(C{L}_{int,u,uptake}+C{L}_{passive})*(C{L}_{int,u,met}+C{L}_{int,u,bile})}{(C{L}_{int,u,sinusoidalefflux}+C{L}_{passive}+C{L}_{int,u,met}+C{L}_{int,u,bile})}$	(Sirianni and Pang 1997; Umehara K-i and Camenisch 2012)
Conventional model with Kpuu	$CL=\frac{Q_{liver}{R}_{BP}C{L}_{int,u,all}{f}_{up}}{Q_{liver}+R_{BP}C{L}_{int,u}{f}_{up}}$ where      ${CL}_{int,u,all}={CL}_{int,u,met}{K}_{puu}$	(Izumi et al. 2017; Li N et al. 2021)
Hepatic uptake intrinsic clearance method	$CL=\frac{Q_{liver}{R}_{BP}C{L}_{int,u,all}{f}_{up}}{Q_{liver}+R_{BP}C{L}_{int,u}{f}_{up}}$ where      ${CL}_{int,all}=\left({CL}_{int,u,uptake}+{CL}_{int,u,passive}\right)$	(Izumi et al. 2017)

The free drug theory states that equilibrium between the free and protein-bound drug is instantaneous. However, some studies reported greater drug uptake into liver than that predicted based upon the free fraction of drug in blood (Forker and Luxon 1981; Weisiger et al. 1981; Blanchard et al. 2004; Blanchard et al. 2006). These studies predicted a higher clearance in presence of serum as compared to clearance in absence of serum. The various purported mechanisms for this observation has been summarized in some of the recent publications (Poulin et al. 2016; Bowman C and Benet 2018; Bteich et al. 2019). One hypothesis regarding increased metabolic clearance in the presence of protein is that ionic interactions between the protein–drug complex and hepatocyte cell surface would supply more unbound drug to the cell membrane, therefore resulting in a higher intracellular concentration than would be expected based on the free concentration in plasma. Based on this hypothesis Poulin et.al, proposed that the unbound fraction in whole liver (f_u,liver) would be higher than in plasma (f_up) under in vivo conditions. It was proposed that instead of f_up, f_u,liver should be used, which accounts for the differential amount of albumin present in plasma and in liver (plasma to whole-liver concentration ratio, PLR) (Table 1). Intracellular concentrations of albumin are negligible. Hence, PLR essentially is obtained by converting the extracellular fluid-to-plasma concentration ratio into whole-liver–plasma ratio. This method led to improved accuracy in clearance prediction as compared to conventional method and Berezhkovskiy’s method. However, it was argued that Poulin’s method did not offer significant improvements over simple empirical correction of average bias. (Hallifax David and Houston 2012; Poulin Patrick, Hop Cornelis ECA, et al. 2012; Poulin Patrick, Kenny Jane R, et al. 2012). Moreover, it was also argued that as per the basic thermodynamic principle of microscopic reversibility, an increase in concentrations near the membrane at equilibrium cannot increase intracellular concentrations in the absence of an energy drive process (active transport or pH) (Korzekwa and Nagar 2017).

CL_int,u when Transporters are Involved

According to the free drug hypothesis, the unbound concentrations of drug in the blood must be equal to the unbound concentration inside of cells. However, for drugs that violate the free drug hypothesis (i.e. transporter substrates), the concentration of intracellular and extracellular unbound drug has the potential to differ. The degree of difference is based on the relative contributions of unbound active and passive intrinsic uptake clearance (PS_int,uptake), unbound active and passive intrinsic sinusoidal efflux clearance (PS_int,efflux), unbound metabolic intrinsic clearance (CL_int,met), and unbound biliary intrinsic clearance (CL_int,bile). The Extended Clearance Model has been developed, which integrates these membrane passage intrinsic clearances with metabolism and biliary intrinsic clearances towards improved clearance predictions for transporter substrates (Sirianni and Pang 1997; Umehara K-i and Camenisch 2012; Benet et al. 2018), and is governed by the following relationship:

$$C{L}_{int,u}=\frac{(C{L}_{int,met}+C{L}_{int,bile})\cdot P{S}_{int,uptake}}{C{L}_{int,met}+C{L}_{int,bile}+P{S}_{int,efflux}}$$ (4)

Evaluation of the potential rate-determining steps is helpful in further simplifying this relationship. For example, for compounds with high passive permeability that overwhelm any active transport, passive membrane passage intrinsic clearances will be equal to one another and will be much larger than the sum of metabolic and biliary intrinsic clearances, i.e. PS_int,uptake and PS_int,efflux >> (CL_int,met + CL_int,bile), and therefore the total CL_int,u relationship simplifies to the sum of biliary and metabolic intrinsic clearances, i.e. CL_int,u = CL_int,met + CL_int,bile. The various scenarios have been extensively interrogated by the field (Sirianni and Pang 1997; Kunze et al. 2015; Umehara K et al. 2020), and of note it is widely recognized by these authors that the extended clearance model is a WSM concept. That is, when transporter-mediated processes are incorporated, the only analysis possible is the WSM since the differing concentrations driving the various metabolic and membrane passage clearances are assumed to be the same concentration, thus the utilization of alternate models of hepatic disposition in such clearance predictions offers no advantage (Benet et al. 2018; Sodhi, Liu, et al. 2020). More recently, the extended clearance model derivation has been presented following oral and IV dosing, suggesting that evaluation of AUC ratios are more valuable approach to evaluate drug-drug and pharmacogenomic interactions than evaluating rate-determining steps, since ultimately exposure changes are the relevant outcome and the emphasis on rate-determining steps can be misleading (Benet et al. 2018). For compounds that are predominantly eliminated by liver, liver exposure will be determined primarily by biliary excretion and metabolic clearance. Hence inhibition of metabolism or biliary clearance could have a significant effect on liver exposure (Watanabe et al. 2009; Chu X et al. 2013; Patilea-Vrana and Unadkat 2018; Kulkarni et al. 2020; Sodhi, Liu, et al. 2020). In contrast, inhibition of an uptake transporter has the potential to alter the liver C_max and half-life, but not the intraorgan liver exposure. Incorporation of the extended clearance model in clearance predictions is evident in a number of recent investigations (Watanabe et al. 2010; De Bruyn et al. 2016; Patilea-Vrana and Unadkat 2018; Umehara K et al. 2020).

The degree of unbound liver-to-blood partitioning coefficient (Kp_uu) is represented by the following expression (Sodhi, Liu, et al. 2020):

$$K{p}_{uu}=\frac{P{S}_{int,uptake}}{C{L}_{int,met}+C{L}_{int,bile}+P{S}_{int,efflux}}$$ (5)

When high passive permeation prevails, the expression simplifies to the value 1, however, when there is significant contribution of active transporter processes to the membrane passage intrinsic clearance values, the value of Kp_uu will not be unity and will reflect the relative degree of intraorgan unbound drug accumulation. Incorporation of the Kp_uu relationship into the Eq. 4 extended clearance model results in the following relationship for CL_int:

$$C{L}_{int}=K{p}_{uu}\cdot (C{L}_{int,met}+C{L}_{int,bile})$$ (6)

Kp_uu can be experimentally determined using several in vitro approaches, such as by experimentally determining the individual intrinsic clearances and using Eq. 5 (Umehara K-i and Camenisch 2012; Riede Julia et al. 2017), or by determining total drug partitioning (Kp) in hepatocytes and correcting this measurement with the fraction unbound in the hepatocyte (f_u,hepatocyte), i.e. Kp_uu = Kp · f_u,hepatocyte (Yabe et al. 2011; Chu Xiaoyan et al. 2013; Kulkarni et al. 2016; Riccardi et al. 2017; Riede Julia et al. 2017; Li et al. 2020). Determinations of f_u,hepatocyte are commonly evaluated via equilibrium dialysis of drug with liver homogenates (f_u,liver), equilibrium dialysis with hepatocytes at 4 °C (f_u,cell) (when transporters are inactive) (Yoshikado et al. 2017; Riccardi et al. 2020), or estimated by an established relationship with logD (Yabe et al. 2011; Riccardi et al. 2020). It has been shown that f_u,liver and f_u,cell values were independent of the species and cell types (Riccardi et al. 2018). Although both the methods (f_u,liver and f_u,cell) corelate well, f_u,liver was recommended over f_u,cell since f_u,liver data is more reliable with less experimental variability, experimentally simple, more amenable to high throughput format, does not suffer from residual metabolic activity, cytotoxicity or cell loss, and is less expensive. Similar to the extended clearance model, Kp_uu is also a WSM concept in that the driving force unbound intracellular concentrations are assumed to be the same for the various membrane passage and metabolic processes, a scenario that is unlikely, thus utilization of alternate hepatic disposition models to scale up Kp_uu-corrected intrinsic clearance determinations offer no advantage (Sodhi, Liu, et al. 2020).

Large ESFs are typically required for predicting in vivo clearance for uptake transporters using hepatocytes (Jones et al. 2012). Improved human hepatic clearance prediction was observed when ESFs derived from plated hepatocytes in cynomolgus monkey were applied to plated human hepatocyte data (De Bruyn et al. 2018). Different in vitro systems have been used to determine uptake clearance, including transporter overexpressing cells (eg. HEK OATP1B1), suspended hepatocytes, plated hepatocytes and sandwich culture hepatocytes. It was speculated that differences in these in vitro systems (differences in activity and expression levels of uptake transporters) lead to variable parameters from different systems (Cantrill and Houston 2017) which could contribute to the observed challenges in IVIVE success for transporter substrates. For instance, a comparison of PS_int,uptake for 8 drugs showed that PS_int,uptake in SCH < plated hepatocytea < suspended hepatocytes (Yabe et al. 2011; Ménochet et al. 2012; Cantrill and Houston 2017). The Lack of IVIVE of transporter-mediated drug disposition has often been attributed to in vitro in vivo differences in transporter expression or activity. Several studies have shown differences in expression levels of transporters in the hepatocyte systems as compared to the human liver (Lundquist et al. 2014; Vildhede et al. 2015; Vildhede et al. 2018), whereas other studies have shown no significant difference in levels of transporter expression between hepatocyte system and human liver tissues (Kimoto et al. 2012; Prasad et al. 2014; Badée et al. 2015; Kumar et al. 2019). It was found that expression of OATP1B1 is comparable in liver tissue and cryopreserved hepatocytes in both suspension and sandwich culture hepatocytes (Badée et al. 2015). For OATP1B3 and OATP2B1 the expression in cryopreserved hepatocytes in suspension were comparable to that observed in human liver tissue samples whereas in sandwich culture human hepatocytes (SCHH) it was up to 2.5-fold lower than liver tissue (Badée et al. 2015). More recently, it was shown that unbound uptake intrinsic clearance for rosuvastatin was similar in three in vitro systems (suspended, plated and sandwich culture hepatocytes) (Kumar et al. 2020). Furthermore, interspecies differences in hepatic uptake clearance have been reported previously (Menochet et al. 2012; Chu Xiaoyan et al. 2013; Liao et al. 2019), which might be a result of differences in protein homology and differences in absolute abundances of OATPs (Wang L et al. 2015; Kimoto et al. 2017).

The influence of albumin has also been observed on the hepatic uptake transport of xenobiotics in rat (Li et al. 2020), monkey (Liang et al. 2020; Li N et al. 2021), dog and human hepatocytes (Miyauchi et al. 2018; Bi et al. 2020; Kumar et al. 2020; Liang et al. 2020; Li N et al. 2021), transporter over expressing cells (Bi et al. 2020; Bowman CM et al. 2020; Kumar et al. 2020). The 4% bovine serum albumin (BSA) or human serum albumin (HSA) is often used to mimic the physiologic amount of albumin in plasma and has been applied to various species. Although, the underlying mechanism is not clear, in vitro hepatocyte systems supplemented with proteins (BSA, HSA or serum) appear to result in improved clearance predictions. This has led to research groups adding proteins in the incubation medium including 2% human serum albumin (Fukuchi et al. 2017), 10% fetal bovine serum (Oh et al. 2018) or 10% fetal calf serum (Riley et al. 2005), 4% BSA (Li N et al. 2021), 5% human serum (Yamagata et al. 2017), 5% human serum albumin (Kumar et al. 2020), 10% human or monkey serum (Liang et al. 2020), 100% human serum (Shibata et al. 2002), 100% plasma (Bowman CM et al. 2019; Bowman CM et al. 2020; Kumar et al. 2020).

Some studies have reported no correlation between albumin-mediated uptake and unbound fraction in plasma (Li N et al. 2021) whereas some have reported trends suggesting higher the binding more the effect of albumin mediated uptake (Bowman CM et al. 2019; Bowman CM et al. 2020; Francis et al. 2020). This was observed in monkey and human hepatocytes and in transporter overexpressing cells (Bowman CM et al. 2020; Liang et al. 2020). This suggests that highly bound drugs may be more significantly affected by proteins than drugs with low binding. Moreover, this was found to be independent of species-, assay format (suspension and monolayer assays), type of plasma protein (BSA, HAS, human plasma, human serum and rat serum), ionization state (acids, bases and neutrals) and ECCS classification (Francis et al. 2020). Recently, to quantify the identified relationship between f_up and fold-change in CL_{int,u in vitro}, linear regression analysis was performed on literature data (Francis et al. 2020). An equation was proposed to predict the fold-change in CL_{int,u in vitro} caused by the addition of proteins. The proposed equation is

$$Log10(Fold-changein{\text{CL}}_{\text{int,u in vitro}})=-0.3774(Log10{\text{f}}_{\text{up}})+0.03253$$ (7)

and can be utilized to predict fold-change for both rat and human datasets. Although a generic relationship between fold-change in CL_{int,u in vitro} and f_up was developed for both uptake transport and metabolically cleared drugs, the relationship is empirical since the underlying mechanisms are clearly not known. The advantage of using simple empirical equations to correct for protein-mediated uptake is that it can be applied to conventional assays without the complexity of adding proteins in incubations, which might decrease analytical sensitivity. The apparent unbound K_m (K_m,u) of the substrates was found to decrease in presence of plasma. Moreover, the decrease in the K_m,u was inversely dependent on protein binding. Compounds having higher protein binding (lower f_up) showed more decrease in K_m,u (Bowman CM et al. 2019; Bowman CM et al. 2020; Francis et al. 2020). Pitavastatin with an f_up of 0.01 showed a fold change of 107 whereas Pravastatin with an f_up of 0.5 showed a fold change of 1.71 in K_m,u in incubations with plasma as compared to buffer. Although the V_max values are expected to remain the same across the f_up ranges, V_max was also found to show similar trends but less marked. The mechanism for the apparent decrease in V_max needs further exploration. CL_{int,u in vitro} also showed the similar trends with higher difference (plasma/buffer) found with higher protein bound drugs (Bowman CM et al. 2019; Bowman CM et al. 2020; Francis et al. 2020). The passive diffusion was also found to be slightly higher in incubations with plasma as compared to buffer (Neuhoff et al. 2006; Bowman CM et al. 2019; Bi et al. 2020; Bowman CM et al. 2020). However, there was no clear trend observed. This suggests that plasma proteins may play a role in both active uptake process and passive diffusion although the mechanism is yet to discovered. Similar results have been observed in hepatocytes, HEK cells and Caco-2 cells (Bi et al. 2020; Bowman CM et al. 2020). There are several mechanisms proposed for this phenomenon which has been summarized recently (Bowman C and Benet 2018). More studies will be needed to know if there is a single mechanism or multiple mechanisms taking result in improved prediction. One of the mechanisms proposed is the “facilitated-dissociation” model, in which the interaction of the albumin–drug complex with the cell surface or a receptor enhances dissociation of the complex to provide more unbound drug molecules to be transported (Forker and Luxon 1981; Weisiger et al. 1981; Tsao et al. 1988). More recently, it was shown that the correlation between the albumin mediated uptake and unbound fraction in plasma is well captured by facilitated diffusion model. Furthermore, it was also shown that the Poulin’s proposed ‘f_u,p-adjusted model’ might overpredict the impact of albumin mediated uptake (Bi et al. 2020).

Limitations in the Predictability of IVIVE Approaches

Although extensive effort has been expended by the field to elucidate the reasons for IVIVE underprediction, there has not yet been a consensus for the mechanistic reasons underlying unsuccessful clearance predictions. There have been a number of attempts to understand the shortcomings of IVIVE, for instance, by investigating issues surrounding non-specific or protein binding (Obach 1997, 1999; Shibata et al. 2002; Riley et al. 2005; Blanchard et al. 2006; Hallifax David and Houston 2006))(Austin et al. 2005), the impact of donor variability in cryopreserved hepatocytes (Hallifax D and Houston 2009), the impact of drug ionization within the liver (Berezhkovskiy 2011; Poulin Patrick, Hop Cornelis ECA, et al. 2012; Poulin Patrick, Kenny Jane R, et al. 2012), the inability of conventional incubations to accurately capture clearance for low turnover compounds (Di et al. 2012; Bonn et al. 2016; Hultman et al. 2016), the barrier for diffusion of low permeability compounds through the unstirred water layer in hepatocyte incubations (Wood Francesca L et al. 2018), the potential involvement of xenobiotic transporters for primarily metabolized drugs (Bowman CM and Benet 2016), the integration of transporter-mediated membrane passage with metabolic elimination with the development of the Extended Clearance Model (Umehara K-i and Camenisch 2012), incorporation of liver-to-plasma partitioning or fraction unbound in the liver for transporter substrates (Poulin 2013; Izumi et al. 2017; Riccardi et al. 2017; Riccardi et al. 2018), inhibition by endogenous fatty acids (Kilford et al. 2009; Gill et al. 2012) well as the evaluation of various hepatic disposition models (Ito and Houston 2004; Hallifax David et al. 2010).

Despite these numerous investigations, there is still no consensus on the types of drugs for which IVIVE can be trusted to quantitatively predict clearance. It has been recently suggested that the theoretical basis of IVIVE clearance predictions should be re-examined (Benet and Sodhi 2020) (Sodhi and Benet, 2021, J Med Chem). In particular, with respect to f_u,B calculations from f_up and blood-to-plasma partitioning determinations for transporter substrates (due to the presence of transporters in erythrocyte membranes, as such calculations rely on the assumption that unbound drug concentrations in both plasma and blood are equal), the potentially underpredicted hepatic blood flow value utilized in clearance predictions, as well as the irrelevance of a fixed-volume incubation to capture the nuances of the in vivo pharmacokinetic hepatic volume of distribution in conventional clearance prediction approaches. The distribution of each drug will depend on its unique physicochemical properties; therefore a universal biological scaling factor alone is not considered appropriate for IVIVE, which many in the field presently believe will succeed.

Conclusion

Validity of hepatic clearance in vitro is bridged to the in vivo situation by evaluating the clearance of a compound in a rodent after oral and IV administration. Good IVIVE is imperative to predict a safe and efficacious dose in human, and models are continually being updated to incorporate experimental data along with mathematical formulae to determine the best fit for in vitro to in vivo correlations.

Relating pharmacokinetics to pharmacodynamics in vivo is crucial to understanding efficacy at given concentration, however it can be challenging to recapitulate a human disease in animal models for adequate characterization of drug efficacy. This is not only owing to a variety of interspecies differences in proteins such as drug metabolizing enzymes and transporters, but also since some leading disease models may need to be chemically induced. Well-designed studies to establish the PK/PD relationship in the appropriate animal model must be implemented such that reasonable scaling to the human condition can be applied (Gabrielsson Johan et al. 2009). In animal models, well-characterized biomarkers appear to be underutilized and could be valuable in better understanding compound efficacy and adverse effects (Bai et al. 2011; Wendler and Wehling 2012). Additionally, current pitfalls in in-vivo models is the continual employment of a single sex in evaluating PK and PD. Integration of female and male animals would be ideal given most drugs will ultimately be used in both sexes, however clear barriers include a doubling in the number of animals used to account for inter-sex variability.

Although it is not appropriate to completely replace in vivo experiments using alternative models, there are useful applications for lower invertebrate models such as zebrafish, as well as invertebrates C. elegans and Drosophila in drug discovery. There is general agreement that these models experience less stress and suffering compared to traditional models including rodents, and there are additional benefits to employing these models such as reduced cost, small size, and rapid development. Zebrafish have 70% genetic homology to humans compared to 83% homology observed between humans and mice (Howe, Clark, et al. 2013), thus there is sufficient genetic overlap to assess certain liabilities including toxicity prior to testing in rodents. However, evaluation of pharmacokinetics in these models is limited; zebrafish are quite small, and quantification of internal drug concentrations is rarely achieved, which is key to understanding the PK of a molecule. Drosophila have ~75% genetic similarity to humans and their rapid propagation makes them easy targets for genetic knockout using technologies including CRISPR/Cas9. Clear caveats of using Drosophila, however, include their differing physiology relative to humans. Interesting studies including the creation of zebrafish liver microsomes that express human CYP3A4 (humanized zebrafish, (Poon et al. 2017) are working toward better recapitulating metabolism using alternative models, and we look forward to upcoming research into improved PK and PD assays using alternative models.

In addition to in vivo and alternative models, in vitro models will remain vital to drug discovery, particularly in early discovery when high-throughput screening is necessitated. Novel technologies to bridge the gaps of traditional microsomal and hepatocyte incubation techniques for hepatic clearance prediction include 3D cultures such as micro-patterned co-cultures, microfluidic co-cultures, and spheroids. Use of MPCCs can improve the phenotypic stability of hepatocytes for multiple weeks compared to hepatocytes in suspension and improves CL_int prediction accuracy, especially for low clearance compounds. Similar to traditional microsomal and hepatocyte incubations, however, these systems tend to underpredict the clearance for high turnover compounds (Wood Francesca Leanne et al. 2017; Chan et al. 2019). One major advantage of these assays is that variability in clearance measurements was improved, resulting in better confidence in the predicted clearance (Docci et al. 2019). Spheroids can consist of monocultures or cocultures including nonparenchymal hepatic cells, and their application is mainly used in hepatotoxicity assessment (Mizoi, Hosono, et al. 2020; Miranda et al. 2021) despite having well-characterized transcriptomes, proteomes, and metabolomes (Bell et al. 2016; Bell et al. 2017; Vorrink et al. 2017; Bell et al. 2018; Messner et al. 2018). Microfluidic co-cultures have been noted to have lower interexperimental variability in terms of clearance assays (Bonn et al. 2016). The application micro-patterned co-cultures, spheroids, and microfluidic systems to hepatic clearance assays and predictions is only just emerging, thus we look forward to expanded datasets for validating the utility of these systems in drug discovery.

In summary, recent efforts have led to improved understanding of different in vitro and mathematical models which has further led to improved predictability of in vivo clearance. Slowly metabolized compounds are on the rise which has pushed the field to develop new in vitro tools to evaluate low turnover compounds that are more routinely used during the drug discovery process. Simultaneously, traditional in vivo models and alternative in vivo models have also been developed to better capture human etiology and develop PK-PD relationships to support drug development. These ever-evolving and interconnected experimental and mathematical approaches that harness in vitro and in vivo data to evaluate areas of agreement and areas of divergence. Critically evaluating in vitro to in vivo discrepancies followed by the creative design novel techniques and models will be required to bridge current and future-identified gaps. To optimize these efforts, continued integration of expertise from scientists in industry and academia will be required. Looking back at the past 50 years, we have come a long way in our understanding of mechanisms underlying PK and PD, from in vitro to in vivo. Going forward, we are excited about the continual and rapidly advancing technologies in our era of drug discovery and we highly anticipate harnessing these for improved discovery and development of highly efficacious and safe medications for those in need.

Abbreviations

3Rs

replacement, reduction, refinement

AUC

area under the concentration-time profile

CL

clearance

CL_h

hepatic clearance

CL_int

intrinsic clearance

C_max

maximal concentration

CPPGL

cytosolic protein per gram liver

CYP P450

cytochrome P450 enzyme

ESF

empirical scaling factor

F_m

fraction metabolized

f_u

fraction unbound (b: blood, h: hepatic, inc: incubation, p: plasma)

FDA

Food and Drug Administration

HLM

human liver microsomes

HPF

hours post-fertilization

HPGL

hepatocellularity per gram liver

iPSC

induced pluripotent stem cell

IVIVE

in vitro to in vivo extrapolation

k_inc,u

1^st order rate constant for unbound substrate depletion

K_p

drug partitioning coefficient

K_puu

liver-to-blood partition coefficient

LD₅₀

lethal dose that kills 50% of a test sample

MPCC

micro-patterned co-culture

MPPGL

microsomal protein per gram liver

NASH

non-alcoholic steatohepatitis

NAFLD

non-alcoholic fatty liver disease

PBPK

physiologically-based pharmacokinetic

PD

pharmacodynamics

PK

pharmacokinetics

PLR

plasma-to-whole-liver ratio

PS

total (active + passive) intrinsic uptake clearance

PTM

parallel-tube model

Q_h

hepatic blood flow

S9PPGL

S9 protein per gram liver

SREBP

sterol regulatory-element binding protein

UGT

UDP-glucuronosyltransferase

V_inc

volume of incubation

WSM

well-stirred model

ZLM

zebrafish liver microsomes