Full article: Characterisation of intravenous pharmacokinetics in Göttingen minipig and clearance prediction using established in vitro to in vivo extrapolation methodologies

Formulae display: $MathJax Logo$ ?Mathematical formulae have been encoded as MathML and are displayed in this HTML version using MathJax in order to improve their display. Uncheck the box to turn MathJax off. This feature requires Javascript. Click on a formula to zoom.

Abstract

The use of the Göttingen minipig as an animal model for drug safety testing and prediction of human pharmacokinetics (PK) continues to gain momentum in pharmaceutical research and development. The aim of this study was to evaluate in vitro to in vivo extrapolation (IVIVE) methodologies for prediction of hepatic, metabolic clearance (CL_hep,met) in Göttingen minipig, using a comprehensive set of compounds.
In vivo clearance was determined in Göttingen minipig by intravenous cassette dosing and hepatocyte intrinsic clearance, plasma protein binding and non-specific incubation binding were determined in vitro. Prediction of CL_hep,met was performed by IVIVE using conventional and adapted formats of the well-stirred liver model.
The best prediction of in vivo CL_hep,met from scaled in vitro kinetic data was achieved using an empirical correction factor based on a ‘regression offset’ of the IVIV relationship.
In summary, these results expand the in vitro and in vivo PK knowledge in Göttingen minipig. We show regression corrected IVIVE provides superior prediction of in vivo CL_hep,met in minipig offering a practical, unified scaling approach to address systematic under-predictions. Finally, we propose a reference set for researchers to establish their own ‘lab-specific’ regression correction for IVIVE in minipig.

Keywords:

Introduction

Porcine, in particular the minipig, has received growing attention within pharmaceutical research and development as an alternative non-rodent in vivo model for toxicology assessment (Bode et al. Citation2010; Ganderup et al. Citation2012), formulation development (Henze et al. Citation2019) as well as pharmacokinetic (PK) modelling supporting human PK predictions (Yoshimatsu et al. Citation2016; Ding et al. Citation2021). This is due in part to similarities in their biochemistry, physiology and anatomy with humans (Achour et al. Citation2011; Swindle et al. Citation2012). Additionally, being relatively small in size, readily accessible and easy to handle makes them an attractive alternative to dogs and non-human primates (NHP), particularly considered in context of 3R’s and societal constraints (Forster et al. Citation2010; Singh et al. Citation2016). Furthermore, various aspects of drug disposition have been investigated in vitro and in vivo in various pig breeds, including the commonly used Göttingen minipig, revealing their potential for clinical translation (Dalgaard Citation2015; Lignet et al. Citation2016; Wilkinson et al. Citation2017; Poulin et al. Citation2019).

The human PK prediction is an important late-stage research activity, given its impact on dose projections, exposure profiles and therapeutic index, and so is widely used to assess the technical quality of novel drug candidates (McGinnity et al. Citation2007). Arguably, when verified across several pre-clinical PK species, physiologically based pharmacokinetic (PBPK) modelling in combination with in vitro to in vivo extrapolation (IVIVE), can provide reliable prediction of human small molecule PK (Jones et al. Citation2006; Jones et al. Citation2011). However, accuracy and robustness of prospective predictions depends on judicious extrapolation of human in vitro data and accompanying animal data (Jones et al. Citation2015; Miller et al. Citation2019). Hepatic, metabolic clearance (CL_hep,met) is considered a key parameter influencing both bioavailability and systemic half-life of oral drugs. Consequently, drug discovery programmes aim to design drug candidates with low predicted CL_hep,met and so considerable effort has been directed to establishing quantitative clearance (CL) predictions using various IVIVE methodologies. Several of these incorporate empirical scaling factors to account for the systematic under-prediction of CL often reported from scaled in vitro intrinsic clearance (CL_int) data (Houston Citation1994; Obach Citation1999; Ito and Houston Citation2004; Riley et al. Citation2005; Hallifax et al. Citation2010; Sohlenius-Sternbeck et al. Citation2010; Sohlenius-Sternbeck et al. Citation2012; Lin et al. Citation2016; Williamson et al. Citation2020; Francis et al. Citation2021; Jones et al. Citation2022). It is well established that demonstrating acceptable IVIVE of CL for candidate drugs in animal PK models improves prediction accuracy and reduces uncertainty in human CL predictions when applying the same scaling approach to human in vitro data (Jones et al. Citation2006; Sohlenius-Sternbeck et al. Citation2012). Typically rat, dog, and if justified NHP, are used for this purpose but minipig offers a compelling non-rodent alternative for reasons mentioned above and given reported similarities in metabolism between minipig and human for various cytochrome P450 (CYP450) and non-CYP450 enzymes such as CYP2C, CYP3A, N-acetyl transferases (NAT) and aldehyde oxidase (AO) (Thorn et al. Citation2011; Dalgaard Citation2015; Wilkinson et al. Citation2017).

To date, few have assessed the intravenous (IV) PK of small molecule drugs in minipig and fewer still have evaluated predictivity of CL in this species from in vitro data (Lignet et al. Citation2016; Yoshimatsu et al. Citation2016; Ding et al. Citation2021). The current study aims to expand upon the systemic PK characterised in female Göttingen minipigs (Siefert et al. Citation1999; Suenderhauf et al. Citation2014; Lignet et al. Citation2016; Patel et al. Citation2017; Ding et al. Citation2021) using a comprehensive reference set that includes central nervous system (CNS) and non-CNS acting drugs, spanning a broad range of physico-chemical properties and covering many of the key human drug metabolising enzymes. To the best of the author’s knowledge, this is the first time different IVIVE methodologies have been compared for prediction of CL in Göttingen minipig for a large dataset. Finally, a reference set is proposed allowing other researchers to establish their own ‘regression offset’ for practical and transparent correction of CL_hep,met predictions in minipig, as has been done for rat and humans (Sohlenius-Sternbeck et al. Citation2010, Citation2012). This in turn should facilitate application of a common, unified in vitro scaling approach across pre-clinical species and humans that also addresses the systematic under-prediction often seen from in vitro data.

Materials and methods

Chemicals

Reference compounds altanserin, antipyrine, atenolol, bupropion, buspirone, carbamazepine, cimetidine, citalopram, diazepam, N-desmethylclozapine, diphenhydramine, doxepin, fluoxetine, gabapentin, indomethacin, metoclopramide, propranolol, risperidone, verapamil, and Way-100635, were purchased from Sigma Aldrich (St Louis, MO, USA). All other chemicals were of analytical grade and obtained from commercial suppliers.

In silico predictive models

In silico prediction of log D_7.4 and microsomal non-specific binding (fu_inc) were made using ADMET™ Predictor software version 10.3 (Simulation-Plus, Lancaster, CA, USA).

In vitro studies

Determination of drug binding to plasma proteins and microsomes

The free fraction of drug in pooled mixed gender Göttingen minipig plasma and human liver microsomes (HLM) were determined by equilibrium dialysis using 96-well HTD-dialysis plates (HTD Dialysis LLS, Gales Ferry, CT, USA) with dialysis membranes (molecular weight cut off 12–14 KDa). Gender differences in plasma protein binding are not expected and so mixed pooled plasma was used for fu_p determination. Pooled plasma was obtained from BioIVT (Westbury, NY, USA, Product no. MPG00PLK2M2N, lot no. MPG24695) and HLM were purchased from Corning^® (Corning, NY, USA. Product no. 452117, lot no. 38296). One side of the HTD-dialysis plate was loaded with matrix (HLM or plasma) and the other side with buffer (100 mM sodium phosphate buffer, pH 7.4). Test compound was dissolved in DMSO, excepting gabapentin, which was dissolved in water, then spiked (5 µL of 0.2 mM) into blank (995 µL) plasma (10 or 100%) or HLM protein matrix (1 mg/mL) giving a final nominal concentration 1 µM (≤0.5% DMSO). The matrices were loaded into respective chambers and equilibrated against buffer for 5 h at 37 °C (in an air incubator with 5% CO₂ with shaking). Samples from both chambers (buffer and plasma or HLM) were aliquoted to fresh tubes then matrix matched using an equal volume of opposite blank matrix before extraction with cold solvent (3 volumes) containing an appropriate bioanalytical internal standard. After centrifugation (20 min, 3200 g, 4 °C) the supernatants were diluted with appropriate volumes of water then analysed by LC/MS-MS. Recovery of all compounds was >80%, unless otherwise stated.

The unbound fraction in plasma proteins (fu_p) and HLM (fu_inc) were calculated according to: (1) ${F u}_{p} = \frac{C_{buffer}}{C_{plasma}} where C is the test compound concentration$ (1) (2) ${HLM f u}_{inc} = \frac{C_{buffer}}{C_{microsomes}}$ (2)

The HLM fu_inc values were used as a first approximation of non-specific binding in the hepatocyte incubations given known challenges with measuring fu_inc in suspended hepatocytes. Others (Austin et al., Citation2005; Kilford et al. Citation2008) reported a strong correlation between HLM fu_inc (1 mg/mL) and hepatocyte fu_inc (1·10⁶ cells/mL) supporting that measured microsomal fu_inc approximates hepatocyte fu_inc at the hepatocyte incubation concentration used in the present study.

Determination of apparent intrinsic clearance (CL_int,app) in suspended hepatocytes

Pooled female minipig cryopreserved hepatocytes (BioIVT, Westbury, NY. Product no. F00615, lot no. OFM) were rapidly thawed in a 37 °C water bath (approximately 90 sec). Hepatocytes were gently re-suspended in 45 mL of pre-warmed Dulbecco’s Modified Eagle Medium (DMEM) (Gibco; Thermo Fisher Scientific, Inc., Waltham, MA, USA). The cell suspension was centrifuged (4 min, 60 g, room temperature), the supernatant removed, and the remaining cell pellet re-suspended in 45 mL pre-warmed Leibowitz-15 medium (Gibco; Thermo Fisher Scientific, Inc., Waltham, MA, USA) supplemented with 11 mM D(+)-glucose, 4 mM N_aHCO₃, and 25 mM sodium HEPES. The cell suspension was again centrifuged (4 min, 60 g, room temperature) and supernatant removed. The hepatocytes were re-suspended in 2–4 mL Leibowitz-15 medium (cell vial yield: >5 million viable cells). An equal volume (25 μL) was mixed with Leibowitz-15 medium and trypan blue solution and the total number of cells, cell concentration and viability were determined by Trypan blue exclusion using a haemocytometer under microscope. CL_int experiments were only performed if hepatocyte viability was ≥80%. The hepatocyte cell suspension was subsequently mixed with Leibowitz-15 medium to give a cell concentration of 2·10⁶ total cells/mL.

Compound stocks were prepared in DMSO, excepting gabapentin which was dissolved in water, then working solutions (1 µM) were spiked (175 μL) into the hepatocyte suspension (2·10⁶ cells/mL total cells, 175 µL) giving a final nominal incubation concentration of 0.5 µM (≤0.05% DMSO). The incubations were performed in an Inheco Incubator Shaker DWP on a Hamilton Microlab Star liquid handling robot (Hamilton Robotics, Bonaduz, Switzerland). The rate of parent compound disappearance was determined from duplicate incubations (37 °C). The reactions were initiated by addition of compound then sampled at time points: 1, 5, 10, 15, 30, 60, 90, and 120 min. The samples (25 µL) were extracted with cold acetonitrile (5 volumes) containing an appropriate bioanalytical internal standard. The plates were centrifuged (2700 g, 20 min, 4 °C) and supernatants analysed by LC-MS/MS.

The first order elimination rate constant (k) was derived from the mono-exponential decline of the parent compound with time: (3) $C_{t} = C_{0} \cdot e^{- k t}$ (3)

C_t is the concentration at time t and C₀ is the concentration at t = 0.

From the natural log of peak area ratios (test compound peak area/internal standard peak area) plotted against incubation time, the slope (k) of the linear regression was determined. (4) ${Hepatocyte C L}_{int, app} = \frac{k \cdot V}{C}$ (4)

C is the hepatocyte concentration (10⁶ cells/mL), k is the elimination rate constant and V is the incubation volume (mL). All linear regressions were evaluated according to their correlation coefficient (R²) with values typically >0.98.

Test compounds were assessed in each in vitro assay in duplicate or triplicate incubations on at least one test occasion. Drug probe substrates were incubated alongside test compounds for all in vitro assays to ensure reproducibility of the fu_p, HLM fu_inc and hepatocyte CL_int,app assays (data not presented but Landqvist Citation2014 approach adopted (Landqvist et al. Citation2014)).

In vivo studies

Fifteen female Göttingen minipigs were purchased from Ellegaard A/S (Dalmoes, Denmark) and used to determine key in vivo PK parameters for 20 selected compounds. This work was carried out in accordance with the Danish legislation regulating animal experiments; Law and Order on Animal experiments; Act No. 474 of 15/05/2014 and No. 2028 of 14/12/2020, and with the specific licence for this experiment issued by the National Authority.

Upon arrival, animals were acclimatised for at least two weeks prior to commencement of in vivo experiments. They were fed twice daily with standard pellet feed and had access to water ad libitum. Animals were given fresh straw, bedding and activity toys daily. During the acclimatisation period animals were housed 3 per pen, but post vascular access button implantation and on day of experimentation they were housed individually. Doses were calculated according to individual body weight. Their average weight was 17.8 ± 4.4 kg per animal, ranging between 11.3 and 23.2 kg. The majority of test compounds were cassette dosed and animals were reused following a 7-day washout phase to optimise animal usage in accordance with 3R’s philosophy implemented at H. Lundbeck A/S.

Dual ‘Vascular Access Buttons’ (VAB) were implanted in the jugular vein of each animal, with two catheters available for dosing and blood sampling. For PK evaluation, a cassette dosing approach was utilised for the majority of test compounds. This involved simultaneous administration of >1 test compound at relatively low doses to each animal (n = 3). The main advantages to this approach being reduction in animal numbers and also the number of samples for quantitative bioanalysis. The risk presented by potential drug–drug interactions was minimised by ensuring the number of co-administered compounds per cassette was <5; the compounds were administered at low but detectable dose levels; wherever possible cassette dosing compounds with similar pharmacology was avoided; cassettes were selected ensuring minimal chromatographic interference and potential for mass clashes arising from metabolites (such as oxidation, dealkylation, hydrolysis etc). The dosing solutions were thoroughly checked for compound precipitation and discrete dosing was employed for compounds with known risks of causing cytochrome P450 inhibition. Each test compound was administered to minipigs (n = 3) via IV bolus injection, cassette 1: antipyrine (1 mg/kg), atenolol (0.3 mg/kg), fluoxetine (0.3 mg/kg), metoclopramide (0.3 mg/kg), cassette 2: citalopram (0.5 mg/kg), N-desmethylclozapine (0.5 mg/kg), indomethacin (0.38 mg/kg), gabapentin (0.5 mg/kg), cassette 3: carbamazepine (0.5 mg/kg), diazepam (0.5 mg/kg) cassette 4: Way-100635 (0.1 mg/kg), diphenhydramine (0.2 mg/kg), doxepin (0.1 mg/kg), cassette 5: buspirone (0.5 mg/kg), cimetidine (0.5 mg/kg), propranolol (0.1 mg/kg), discrete dosing: altanserin (0.1 mg/kg), bupropion (0.1 mg/kg), fluoxetine (1.0 mg/kg), risperidone (0.1 mg/kg), verapamil (0.2 mg/kg). Antipyrine, fluoxetine, atenolol, metoclopramide, Way-100635, doxepin and diphenhydramine were dosed with saline; all other compounds were administered in 10% hydroxy propyl (HP)-beta-cyclodextrin (Roquette, Lestrem, France).

Following dose administration, blood samples (0.5 mL) were collected on ice into K₃EDTA tubes at time points: 0.033, 0.083, 0.25, 0.5, 1, 3, 6, 12 and 24 h. Blood (50 µL) was mixed with cold 0.1 M HEPES (100 µL, Sigma, St. Louis, MO, USA. Product no. H0887) and frozen (−80 °C) until analysis. Plasma was isolated from remaining blood via centrifugation (2500 g, 10 min, 4 °C) then frozen until analysis. Calibration standards (0.1–1000 ng/mL) were prepared in blank whole blood and plasma matrices in order to quantify compound concentration time profiles from the PK studies.

Pharmacokinetic parameters were calculated by non-compartmental analysis (NCA) of blood and plasma compound concentration–time profiles determined in each animal using Phoenix WinNonlin software (Version 8.3. Certara, Princeton, NJ, USA). The area under the plasma and blood concentration-time curves (AUC), elimination half-life (T½), volume of distribution at steady-state (V_ss), and total clearance (CL_total) were calculated. The trapezoidal method was used to calculate the AUC from time zero to the last measurable concentration (AUC_0–t) and AUC to infinity (AUC_0–∞) was extrapolated by <15%. The CL_total was calculated according to the following equation: (5) ${C L}_{total} = \frac{Dose}{{AUC}_{0 - \infty}}$ (5)

The in vivo blood-to-plasma concentration ratio (R_b) was determined using AUC_0–∞ according to EquationEquation (6)(6) $R_{b} = \frac{blood {AUC}_{0 - \infty}}{plasma {AUC}_{0 - \infty}}$ (6) and used to calculate the in vivo total blood clearance (blood CL_total) according to EquationEquation (7)(7) ${Blood C L}_{total} = \frac{{Plasma C L}_{total}}{R_{b}}$ (7) . For IVIVE of CL_hep,met the in vivo R_b value was applied where available; otherwise R_b was estimated according to ion class (1-haematocrit for acids, 1.00 for other ion classes). Haematocrit has been measured by Ellegarrd (https://minipigs.dk/about-gottingen-minipigs/background-data) and a mean value for female Göttingen minipigs aged 6–8 months (spanning standard body weight of 14.2 kg) is reported as 0.43 which aligns with published data (Bollen et al. Citation2005). Where literature data allowed, renal excretion (Fe) was subtracted from the CL_total giving CL_hep (assumed that compounds underwent negligible biliary excretion). Thus, the remaining CL is attributed to metabolism in the liver (e.g. CL_hep,met). (6) $R_{b} = \frac{blood {AUC}_{0 - \infty}}{plasma {AUC}_{0 - \infty}}$ (6) (7) ${Blood C L}_{total} = \frac{{Plasma C L}_{total}}{R_{b}}$ (7)

LC-MS/MS analysis

LC-MS/MS analyses for in vitro CL_int,app, fu_p and fu_inc assays and in vivo PK studies were performed using electrospray ionisation and multiple reaction monitoring (for more details see Supplementals, Worksheet ).

Establishment of IVIVE

Several methods for establishing IVIVE of CL_hep,met using the well-stirred liver model (WSM) in conventional (Method 1) and adapted formats (Methods 2–5) are compared. The details for each method are provided below.

Prediction of blood CL_hep,met using the WSM (Methods 1–3)

Clearance prediction using the WSM whereby all binding parameters (fu_b and fu_inc) have been incorporated according to the following equation: (8) $Method 1 : Predicted blood {C L}_{hep, met} = \frac{Q_{h} \cdot {C L}_{int, app} / {f u}_{inc} \cdot PSF \cdot {f u}_{b}}{Q_{h} + ({C L}_{int, app} / {f u}_{inc} \cdot PSF \cdot {f u}_{b})}$ (8) where fu_b =fu_p/R_b

Clearance prediction using the WSM whereby it is assumed that fu_inc = 1 which then cancels out accordingly: (9) $Method 2 : Predicted blood {C L}_{hep, met} = \frac{Q_{h} \cdot {C L}_{int, app} \cdot PSF \cdot {f u}_{b}}{Q_{h} + {C L}_{int, app} \cdot PSF \cdot {f u}_{b}}$ (9)

Clearance prediction using the WSM whereby it is assumed fu_b/fu_inc approximates 1 and so all binding parameters cancel out accordingly: (10) $Method 3 : Predicted blood {C L}_{hep, met} = \frac{Q_{h} \cdot {C L}_{int, app} \cdot PSF}{Q_{h} + {C L}_{int, app} \cdot PSF}$ (10)

Q represents liver blood flow in Göttingen minipigs (38.9 mL/min/kg, (Lignet et al. Citation2016)) and PSF is the Göttingen minipig physiological scaling factor accounting for published hepatocellularity (124 · 10⁶ cells/g liver) and liver weight (16.7 g liver/kg body weight) (Lignet et al. Citation2016). This is expressed as: Hepatocellularity · liver weight/kg body weight/1000 giving clearance in units of mL/min/kg body weight.

Prediction of in vivo CL_int and blood CL_hep,met using IVIVE with empirical correction based on a ‘regression offset’ approach (IVIVE Method 4)

As previously described the in vitro CL_int,app is scaled to a predicted in vivo CL_int and blood CL_hep,met according to EquationEquations (11)–(14) (Sohlenius-Sternbeck et al. Citation2012). Initially, hepatocyte CL_int,app values for the recommended model reference compounds (n = 24, see ) are transformed to a scaled CL_int according to EquationEquation (11)(11) ${Scaled i n vitro C L}_{int} = (\frac{{C L}_{int, app}}{{f u}_{inc}}) \cdot PSF \cdot {f u}_{b}$ (11) . The in vivo blood CL_int is derived from the observed blood CL_hep,met using the rearranged well-stirred model according to the following equation: (11) ${Scaled i n vitro C L}_{int} = (\frac{{C L}_{int, app}}{{f u}_{inc}}) \cdot PSF \cdot {f u}_{b}$ (11) (12) ${Derived i n vivo blood C L}_{int} = \frac{Q_{h} \cdot {blood C L}_{hep, met}}{Q_{h} - {blood C L}_{hep, met}}$ (12)

The regression offset equation is then established from a plot of log [derived in vivo blood CL_int] on Y-axis versus log [CL_int,app/fu_inc · PSF · fu_b] on the X-axis using simple linear regression (EquationEquation (13)(13) $Log [(\frac{{Q_{h} \cdot blood C L}_{hep, met}}{Q_{h} - {blood C L}_{hep, met}})] = a \times \log [(\frac{{C L}_{int, app}}{{f u}_{inc}}) \cdot PSF \cdot {f u}_{b}] + b$ (13) ). With this approach, all measured and predicted in vitro variables (including fu_b) are grouped together on the X-axis. The rationale for this approach has been discussed previously (Sohlenius-Sternbeck et al. Citation2010, Citation2012). (13) $Log [(\frac{{Q_{h} \cdot blood C L}_{hep, met}}{Q_{h} - {blood C L}_{hep, met}})] = a \times \log [(\frac{{C L}_{int, app}}{{f u}_{inc}}) \cdot PSF \cdot {f u}_{b}] + b$ (13)

Whereby a is the slope and b is the intercept.

The ‘regression offset’ equation is then used to empirically correct the scaled CL_int,app (EquationEquation (11)(11) ${Scaled i n vitro C L}_{int} = (\frac{{C L}_{int, app}}{{f u}_{inc}}) \cdot PSF \cdot {f u}_{b}$ (11) ) to a log predicted in vivo blood CL_int. The predicted blood CL_hep,met is finally calculated using the WSM according to the following equation: (14) ${Predicted blood C L}_{hep, met} = \frac{Q_{h} \cdot predicted {blood i n vivo CL}_{int}}{Q_{h} + predicted blood i n vivo {CL}_{int}}$ (14)

This differs to the traditional approach (described below as IVIVE Method 5) whereby unbound CL_int (CL_int,ub) is considered e.g. log transformed scaled in vitro and derived in vivo CL_int,ub values are correlated for the recommended model reference set.

Prediction of in vivo CL_int,ub and blood CL_hep,met using IVIVE with empirical correction based on a ‘regression offset’ approach (IVIVE Method 5)

The in vitro CL_int,app is scaled to a predicted in vivo CL_int,ub and blood CL_hep,met according to EquationEquations (15)–(18). Initially, hepatocyte CL_int,app values for recommended model reference compounds (n = 24) are transformed to a scaled in vitro CL_int,ub according to the following equation: (15) ${Scaled i n vitro C L}_{int, u b} = (\frac{{C L}_{int, app}}{{f u}_{inc}}) \cdot PSF$ (15)

The blood in vivo CL_int,ub is derived from the observed blood CL_hep,met using the rearranged well-stirred model according to the following equation: (16) $Derived i {n vivo blood C L}_{int, u b} = \frac{Q_{h} \cdot {blood C L}_{hep, met}}{{f u}_{b} \cdot (Q_{h} - {blood C L}_{hep, met})}$ (16)

The ‘regression offset’ equation is then established from a plot of log [derived in vivo blood CL_int,ub] on the Y-axis versus log [CL_int,app/fu_inc · PSF] on the X-axis using simple linear regression (EquationEquation (17)(17) $Log [\frac{Q_{h} \cdot {blood C L}_{hep, met}}{{f u}_{b} \cdot (Q_{h} - {blood C L}_{hep, met})}] = a \cdot \log [(\frac{{C L}_{int, app}}{{f u}_{inc}}) \cdot PSF] + b$ (17) ). With the conventional approach CL_int,ub is considered on both axes: (17) $Log [\frac{Q_{h} \cdot {blood C L}_{hep, met}}{{f u}_{b} \cdot (Q_{h} - {blood C L}_{hep, met})}] = a \cdot \log [(\frac{{C L}_{int, app}}{{f u}_{inc}}) \cdot PSF] + b$ (17) whereby a is the slope and b is the intercept.

The ‘regression offset’ equation is then used to empirically correct the scaled CL_int,app (EquationEquation (15)(15) ${Scaled i n vitro C L}_{int, u b} = (\frac{{C L}_{int, app}}{{f u}_{inc}}) \cdot PSF$ (15) ) to a log predicted in vivo blood CL_int,ub. The predicted blood CL_hep,met is then finally calculated using the WSM according to the following equation: (18) ${Predicted blood C L}_{hep, met} = \frac{Q_{h} \cdot predicted {CL}_{int, u b} \cdot {f u}_{b}}{Q_{h} + predicted {CL}_{int, u b} \cdot {f u}_{b}}$ (18)

Statistical analysis of predictive performance

The following statistical analyses were performed for each of the IVIVE plots presented in and in order to assess their quality of prediction.

Figure 1. Correlation between predicted blood CL_hep,met scaled from in vitro data using IVIVE Methods 1–3 and the derived in vivo blood CL_hep,met for reference compounds (n = 33). Panel A presents the IVIVE of CL from the WSM where all binding parameters are included (IVIVE Method 1); panel B where fu_inc is excluded based on the assumption fu_inc = 1 (IVIVE Method 2); and panel C where all binding parameters are excluded based on the assumption fu_b/fu_inc cancel out (IVIVE Method 3). The solid line represents line of unity and dotted line represents line of best fit from linear regression. The dot-dashed line represents the liver blood flow limitation set in minipig (38.9 mL/min/kg). Closed and open circles represent reference compounds included or excluded from linear regression analysis and statistical analysis () of the IVIVE of CL_hep,met.

Figure 2. Correlation between log transformed predicted in vivo blood CL_int scaled from in vitro data using Methods 4 and 5 and the derived in vivo blood CL_int for selected reference compounds (n = 24, listed in ). Panels A–C present the IVIVE of CL_int using the WSM with all measured in vitro variables grouped together on the X-axis (Sohlenius-Sternbeck et al. Citation2012); the corresponding regression corrected IVIVE plot of predicted and derived in vivo CL_int; the corresponding regression corrected IVIVE plot of predicted and derived in vivo blood CL_hep,met respectively (IVIVE Method 4). Panels D–F present the IVIVE of CL_int traditionally established with unbound in vitro CL_int correlated with in vivo CL_int corrected with fu_b; the corresponding regression corrected IVIVE plot of predicted and derived in vivo CL_int,ub; the corresponding regression corrected IVIVE plot of predicted and derived in vivo blood CL_hep,met, respectively (IVIVE Method 5).

To get a measure of bias, expressed as the average fold error (AFE), the geometric mean fold error was calculated according to the following equation (Tang et al. Citation2007): (19) $AFE = 10^{\frac{1}{N} \sum \log (\frac{observed}{predicted})}$ (19)

To get a measure of precision the average absolute fold error (AAFE) was calculated according to the following equation (Tang et al. Citation2007): (20) $AAFE = 10^{\frac{1}{N} \sum | \log (\frac{observed}{predicted}) |}$ (20)

The root mean square error (RMSE), also a measure of precision for the predictions, was calculated using the following equation (Sheiner and Beal Citation1981): (21) $RMSE = \sqrt{\frac{1}{N} \sum {(predicted - observed)}^{2}}$ (21)

The correlation coefficient (R²) was determined from linear regression analyses to quantify the strength of relationship between predicted and observed datasets. Finally the specific fold errors of deviation between the predicted and observed values were also calculated (% fold error ≤2-fold, 3-fold and 5-fold).

Results

In vivo pharmacokinetics

Mean plasma and blood concentration-time profiles (n = 3) following IV bolus dosing of compounds (n = 20) to female Göttingen minipigs are presented in . The profiles for both matrices show a high level of concordance allowing robust determination of in vivo R_b values. Despite being dosed in cassette format and administered at relatively low doses, rich data profiles were achieved for the majority of compounds with a high proportion revealing measurable concentrations to 12 or 24 h. The relatively low IV bolus doses meant they were generally well-tolerated. Mild sedative effects were reported for animals receiving the cassette containing diazepam and carbamazepine; mild ataxia was observed shortly after dosing in animals receiving other cassettes (cassettes 1, 2, 4, and 5) and was likely associated with C_max. These transient effects were classified as being mild in nature and so the pharmacokinetics were unlikely to have been negatively affected.

Figure 3. Total plasma and blood concentration-time profiles (mean ± stdev, n = 3) of reference compounds (n = 20) administered by discrete or cassette IV bolus injection to female Göttingen minipigs. *N-des.cloz is an abbreviation for N-desmethylclozapine.

The plasma PK parameters (CL_total, V_ss, T_1/₂) and in vivo R_b are summarised in . The CL_total ranged from 2.0 to 47 mL/min/kg indicating a comprehensive range had been profiled in vivo spanning low, moderate up to very high hepatic extraction ratios (ER) considered against published LBF (38.9 mL/min/kg). Lowest and highest CL_total values were reported for gabapentin and fluoxetine (2.6 and 47 mL/min/kg) which equated to 5% and 121% of LBF, respectively. Additionally, the compound set included all ion classes (acidic, zwitterionic, neutral and basic compounds) and so a wide range of plasma V_ss values were found, ranging from 0.6 to 13 L/kg for gabapentin and fluoxetine respectively, as well as plasma T_1/2 (ranging between 0.9 and 8.5 h for risperidone and metoclopramide, respectively). The inter-individual variability for CL_total and V_ss were relatively low for all compounds (ca. < 25% CV) excepting propranolol and N-desmethylclozapine (29% and 36% CV for CL_total and 45% and 38% CV for V_ss, respectively). The proportion of total AUC_0–∞ extrapolated for 17 compounds was <5%, and for indomethacin, metoclopramide, and fluoxetine it was <15%. The AUC_0–∞ values were used to calculate the in vivo R_b for 15 compounds giving ratios between 0.63 and 1.42. It was not possible to determine R_b for antipyrine, atenolol, fluoxetine and metoclopramide due to insufficient blood sample volumes. For these compounds, R_b is taken from the literature where available or estimated according to ion-class (see Method section). One compound, fluoxetine, was dosed in discrete and cassette PK format. The CL_total closely aligned between formats (42 versus 47 mL/min/kg, respectively) and V_ss was within 2-fold (8.9 versus 13.4 L/kg, respectively). Where possible, the Göttingen minipig IV PK parameters for compounds investigated in this study have been compared with published data from other breeds (). The CL_total and V_ss data found for antipyrine, atenolol, cimetidine, diazepam and verapamil in female Göttingen minipigs (current study) aligned comparatively well with published data in Göttingen minipig; all within 2-fold for CL_total and 2.5-fold for V_ss (Lignet et al. Citation2016; Patel et al. Citation2017; Ding et al. Citation2021). Considering all data, there was a trend towards lower CL_total in NIBS compared with Göttingen minipig as noted for propranolol, antipyrine, ketoprofen, acetaminophen, antipyrine and atenolol (ca. 1.3- to 4.3-fold). V_ss was more closely aligned across breeds and gender as noted for ketoprofen, antipyrine, diclofenac, propranolol and atenolol (ca. <1.5-fold difference); the exception being cimetidine (3.7-fold difference).

Table 1. In vivo PK parameters determined in female Göttingen minipigs for reference compounds (n = 20).

Download CSV Display Table

Table 2. Published minipig in vitro and in vivo IV PK data for reference compounds (n = 19).

Download CSV Display Table

Table 3. In silico predictions and in vitro properties determined in Göttingen minipig for the complete reference set (n = 33 compounds).

Download CSV Display Table

In vitro to in vivo extrapolation of CL using the WSM

To establish an IVIVE of CL_hep,met in Göttingen minipigs, the in vitro rate of metabolism, non-specific binding within the incubation and binding to plasma proteins were investigated for a broad range of compounds (n = 33) covering diverse physico-chemical property space according to ion class (acidic, zwitterionic, neutral, basic) and lipophilicity (logD_7.4 ranging from −2.0 to 2.6). Göttingen minipig hepatocyte CL_int,app, fu_p and HLM fu_inc values are summarised in and visualised alongside in vivo plasma CL_total in . The plots show CL_int,app ranging between 3.7 and 294 μL/min/10⁶ cells and fu_p ranging between 0.04 and 98.5% unbound. Microsomal and hepatocyte fu_inc have been reported to be similar (Austin et al. Citation2005; Kilford et al. Citation2008). The extent of non-specific binding in the hepatocyte preparations (1·10⁶ cells/mL) for each compound was therefore considered to be similar to the measured HLM fu_inc (1 mg/mL). The HLM fu_inc ranged between 0.57 and 100% unbound. The in vitro CL_int,app values of acetaminophen, antipyrine, carbamazepine, hydrochlorothiazide, theophylline, moxifloxacin, gabapentin, ketoprofen and atenolol could not be detected with any certainty as they all had CL_int,app values below the assay sensitivity range (e.g. CL_int,app values <3 μL/min/10⁶ cells). However, CL_int,app values of 1.9 and 2.2 μL/min/10⁶ cells were used for gabapentin and ketoprofen, respectively as measurable turnover could still be discerned in the parent disappearance versus time plots.

Figure 4. Comparison of the key parameters used in the WSM for prediction of CL_hep,met and in vivo CL_int for each reference compound (n = 33). The dynamic range and spread of values measured in vitro (fu_p, fu_inc, CL_int,app) and in vivo (plasma CL_total) are presented in panels A–D, respectively. Closed symbols represent reference compounds recommended for establishing the regression offset.

Several IVIVE methods (1–5) were employed to predict in vivo blood CL_hep,met in minipig for a diverse drug dataset (n = 33 compounds) given that in vivo CL parameter is considered most physiologically relevant in context of estimating the hepatic ER. Additionally, IVIVE Methods 4 and 5 were employed to predict in vivo blood CL_int and CL_int,ub given these parameters are considered more appropriate transformations of the data for linear regression analysis. Principally, they provide a greater dynamic range to evaluate IVIVE whilst minimising bias due to a convergence towards LBF for high intrinsic clearance drugs scaled using the WSM. In accordance with previous work (Sohlenius-Sternbeck et al. Citation2012), the refined reference set was identified on the basis it satisfied key inclusion criteria to establish a hepatocyte ‘regression offset’ that is proposed in the current manuscript for IVIVE of CL_hep,met in minipigs. Namely, that an appropriate dynamic range is covered across relevant in vitro assays and in vivo CL_total (, panels A–D); a range of CYP450 and UGT drug metabolising enzyme pathways are represented (); that CL_total does not exceed LBF; that hepatic metabolism is the major route of elimination (assumptions related to renal and biliary excretion and extra-hepatic metabolism are stated in the in vivo methods section); that compounds known to undergo enterohepatic recirculation, or whose cell permeability is dependent on hepatic uptake transporters, are excluded. Consequently, drugs such as fluoxetine, doxepin citalopram, diphenhydramine were excluded from this part of the analysis as their in vivo CL_total exceeded LBF (38.9 mL/min/kg); also indomethacin given significant biliary excretion and enterohepatic recirculation has been demonstrated in other species (Yesair et al. Citation1970). Other drugs were excluded based on their in vitro profile. For example, felodipine was removed due to very low and uncertain plasma protein binding (fu_p ∼0.04%); theophylline and propranolol because their hepatocyte CL_int,app values lay far outside the assay sensitivity limits (≪3 and ≫300 µL/min/10⁶ cells, respectively).

A statistical summary is provided in detailing the prediction performance achieved with IVIVE Methods 1–5 for the refined compound set and the parameter CL_hep,met. The associated visualisations of predicted versus observed CL_hep,met are presented in panels A–C and panels C and F, respectively. In terms of predicting CL_hep,met, method 4 (empirical correction applied when all in vitro parameters are on the X-axis) performed best in terms of prediction bias (AFE = 0.9), precision (AAFE = 1.4 and RMSE = 1.5), strength of relationship (R² = 0.70) and percent within 2-fold observed (96%). Method 3 (where all binding terms are excluded e.g. fu_b/fu_inc =1) achieved a similar level of performance but with a slightly lower R² (0.65). With Method 5 (empirical correction applied to traditional CL_int,ub axes) slightly poorer values were seen across all statistical parameters. All three of these prediction methods performed significantly better than method 1 (standard approach including all binding parameters) and Method 2 (assumed fu_inc=1) which achieved much lower percentages of predicted CL_hep,met within 2-fold of observed (58 and 38%, respectively). Additionally, the bias (AFE = 2.5 and 4.0) and precision (AAFE = 2.7 and 4.1) with these methods proved much inferior. However, the R² for method 1 (0.69) outperformed all other IVIVE methods excepting Method 4. Low numbers of acidic (n = 2) and zwitterionic (n = 2) compounds in the current reference set prevented a thorough evaluation of IVIVE performance according to individual ion class.

Table 4. Statistical analyses for IVIVE methodologies 1–5 used to predict minipig in vivo blood CL_hep,met via the WSM.

Download CSV Display Table

The statistical summary in details the prediction performance achieved with IVIVE Methods 4 and 5 for the refined compound set and the parameter in vivo blood CL_int and in vivo blood CL_int,ub before and after the empirical regression offset corrections were applied. The associated visualisations for Method 4 of log transformed [CL_int,app · PSF · fu_b/fu_inc] and log transformed [derived in vivo blood CL_int] and the regression corrected, log transformed predicted versus observed in vivo CL_int are presented in panels A and B, respectively. With Method 5 the regression line is established with in vitro CL_int,ub correlated with in vivo CL_int,ub i.e. having been corrected with fu_b. The visualisation of log transformed [CL_int,app/fu_inc · PSF] and log transformed [derived in vivo blood CL_int,ub] and the regression corrected, log transformed predicted versus observed in vivo CL_int,ub are presented in panels D and E, respectively. In terms of IVIVE performance, method 4 (empirical correction applied when all in vitro parameters are on the X-axis) resulted in superior predictions compared to method 5 (traditional IVIVE approach using CL_int,ub axis), before and after regression correction. With the regression corrected predictions, the bias was equivalent between methods (AFE = 1.0) but the precision was higher with method 4 compared to Method 5 (AAFE = 1.9 versus 2.9, RMSE = 2.3 versus 3.9). The agreement was also better for Method 4 (R² = 0.61 versus 0.52) as was the per cent predicted within 2-fold observed (67 versus 46% respectively).

Table 5. Statistical analyses for IVIVE methodologies 4–5 used to predict minipig in vivo blood CL_int via the WSM.

Download CSV Display Table

Discussion

The minipig is emerging as an established non-rodent PK model and may in time replace dog and/or NHP in DMPK as well as in broader drug research and development activities. Several well-known breeds of minipig are used globally within the pharmaceutical industry including Yucatan, Hanford, and Sinclair in the United States and NIBS in Japan (Nunoya et al. Citation2007). Whereas in Europe the most common breed is Göttingen which is viewed as a particularly well managed breed showing minimal genetic drift (Bollen and Ellegaard Citation1997; Simianer and Kohn Citation2010). In this work, we sought to expand the PK dataset published on human reference drugs administered to Göttingen minipig (Siefert et al. Citation1999; Suenderhauf et al. Citation2014; Lignet et al. Citation2016; Patel et al. Citation2017; Ding et al. Citation2021) in order to support future verification and refinement of PBPK models in minipig and to assess IVIVE methodologies for prediction of in vivo CL_hep,met.

The IV PK was evaluated in female Göttingen minipigs for 20 well-studied compounds. To the best of our knowledge, this is the first-time in vitro ADME and in vivo PK data have been reported for the majority of these compounds in Göttingen minipig ( and ). Typically, PK studies are undertaken in male animals. However, due to their ease of handling, female minipigs are routinely used for PK studies at H. Lundbeck A/S and so form the basis of the data presented in this manuscript.

A comprehensive, systematic evaluation has not yet been undertaken in minipig to robustly qualify PK differences due to gender or breed across the various CYP450 and non-CYP450 enzyme substrates. The in vivo PK parameters have been summarised for reference drugs studied in minipig (). Whilst there is no obvious gender difference for V_ss it is difficult to conclude on in vivo drug CL. Of the seven drug substrates previously assessed in male and female Göttingen minipigs, limited differences were noted in CL_total for CYP450 substrates: diazepam, cimetidine and theophylline, or renally excreted hydrochlorothiazide. Whereas higher CL_total (ca. 2-fold) was observed with CYP450 substrates antipyrine and atenolol in females and conversely midazolam (ca. 2-fold) in males. Others have reported gender differences in vitro, based on protein abundance or enzymatic activities, for various human CYP1A, CYP2A and CYP2E substrates with higher metabolic activity in female than male (Skaanild Citation2006; Buyssens et al. Citation2021).

The cassette PK reported in female Göttingen minipigs in this study for antipyrine, atenolol, cimetidine, diazepam, and verapamil compared favourably with published data (Lignet et al. Citation2016; Patel et al. Citation2017; Ding et al. Citation2021) with values falling within 2-fold for CL_total and V_ss excepting cimetidine (2.3-fold difference for CL_total) and diazepam (2.5-fold difference for V_ss). These data indicate good reproducibility can be achieved across laboratories as well as PK formats (discrete versus cassette). Indeed, in the current study the PK of the human CYP2D6 substrate fluoxetine was tested in both discrete and cassette PK formats with reported CL_total and V_ss values showing close alignment (). This supports the previously held notion that cassette dosing can be an efficient means to assess PK of multiple compounds if previous recommendations are followed (Manitpisitkul and White Citation2004; Smith et al. Citation2007; Nagilla et al. Citation2011; Liu et al. Citation2012; Bridges et al. Citation2014).

Relatively little is known about PK differences across porcine and minipig breeds. Others have shown from in vitro and in vivo studies that hepatic CYP3A activity is higher in Göttingen minipig than in conventional pigs or other minipig breeds (Skaanild Citation2006; Helke et al. Citation2016; Patel et al. Citation2017). Acknowledging the relatively sparse dataset (), differences in published IV PK across minipig breeds were evaluated for five drug substrates dosed in male Göttingen and male NIBS minipig (Lignet et al. Citation2016; Yoshimatsu et al. Citation2016). Interestingly, whilst V_ss was very similar, there was a tendency towards high CL_total in Göttingen compared with NIBS. The greatest differences were 4.3-fold and 1.9-fold for the human CYP2C9 substrates diclofenac and ketoprofen respectively. Published PK data in minipig is somewhat limited compared with other species. It is hoped that provision of cassette PK data for a broader set of human reference drugs in female Göttingen minipigs, in a format that minimises animal usage, will encourage others to complete comparative studies across gender/breeds.

In vitro to in vivo extrapolation of CL

Prediction of CL from in vitro systems using IVIVE plays an essential role in triaging compounds for in vivo assessment as well as informing selection of drug candidates with optimised PK for clinical development (McGinnity et al. Citation2007). The premise being that hepatic, metabolic clearance (CL_hep,met) remains the major route of elimination for the majority of drugs (Cerny Citation2016) and so optimisation to low in vitro CL_int,app will translate to low in vivo systemic CL improving likelihood of good oral bioavailability and half-life.

Often the WSM is selected for the purpose of IVIVE given its relative simplicity and comparable performance to other models (Pang and Rowland Citation1977; Ito and Houston Citation2004). Over time, several IVIVE methodologies have been developed for this model to predict in vivo CL_hep,met in pre-clinical species and humans. These include incorporation of correction factors, both empirical and pseudo-mechanistic in nature, to address systematic underprediction typically observed from scaled in vitro data (Obach Citation1999; Grime and Riley Citation2006; Hallifax et al. Citation2010; Sohlenius-Sternbeck et al. Citation2010; Hallifax and Houston Citation2012; Sohlenius-Sternbeck et al. Citation2012; Grime et al. Citation2013; Poulin Citation2013; Williamson et al. Citation2020; Francis et al. Citation2021; Jones et al. Citation2022). Retrospective analyses are an important feature to improving IVIVE approaches. Another important consideration for prospective CL predictions on novel compounds, is the degree of prediction uncertainty. An often held tenet for IVIVE is that if prediction cannot be established in pre-clinical species how can it then be rationalised that it will be predictable if the same scaling approach is applied to human in vitro data (Sohlenius-Sternbeck et al. Citation2012; Grime et al. Citation2013; Miller et al. Citation2019). Therefore, demonstrating acceptable IVIVE of CL_hep,met in animal models becomes an important endeavour and so we look to minipig as an alternate, non-rodent, second species on which to judge the quality of IVIVE of CL_hep,met; in turn building confidence for prospective human CL prediction.

Consequently we have considered five established IVIVE methodologies to predict in vivo CL_hep,met in Göttingen minipigs for a diverse reference set (n = 24 compounds), where in vitro parameters (hepatocyte CL_int,app, fu_p, HLM fu_inc) are measured in one laboratory (rather than compilation of published datasets). The ‘regression correction’ also referred to as ‘regression offset’ approach was included as it offers several advantages (Riley et al. Citation2005; Sohlenius-Sternbeck et al. Citation2010; Sohlenius-Sternbeck et al. Citation2012; Grime et al. Citation2013; Williamson et al. Citation2020). It can be readily implemented for different in vitro matrices (hepatic microsomes, S9 fractions, hepatocytes), performs well for ‘training sets’ but also proprietary candidate drugs (data from author’s laboratory not presented) and provides a means to normalise CL_int,app data e.g. that may vary over time due to changes in assay conditions (Sohlenius-Sternbeck et al. Citation2012). Its main limitation has been a lack of suitable reference compounds from which to establish a lab-specific ‘regression offset’ equation. This is something we address in the current manuscript for Göttingen minipig (). Whilst attempts to develop more mechanistic scaling factors are welcomed, other IVIVE approaches reliant on different empirical or pseudo-mechanistic scalers (Poulin Citation2013; Francis et al. Citation2021; Jones et al. Citation2022) are excluded from this analysis. It is the authors opinion that until these are further substantiated across research groups and multiple species, it is preferable to use the regression offset given it is practical to implement across PK species and can be uniformly applied to all drug classes.

Figure 5. Prediction of CL_hep,met using recommended IVIVE Method 4 (note for method 5 the IVIVE process is established in the same way but using CL_int,ub and EquationEquations (14)–(17)). Step 1: the lab-specific ‘regression offset’ is established using the recommended reference set highlighted in Panel 2. Step 2: to assess IVIVE of CL_hep.met for a new compound, in vitro data is generated in the same lab-specific assays used in step 1 then scaled using the lab-specific regression offset (slope and intercept) for comparison with the in vivo CL_hep,met.

A reference set to establish the ‘regression offset’ equations and to evaluate IVIVE performance for the 5 IVIVE methodologies, was carefully selected according to previous recommendations (Sohlenius-Sternbeck et al. Citation2012) as outlined in the results section and visualised in and . In total, 24 compounds were chosen ( and and ) but predictions for the full compound set (n = 33) are provided in Supplementals (Worksheet Tab 4). To the best of the authors knowledge, this is the first time IVIVE of CL_hep,met has been performed on such a large dataset in minipig, extending the initial work of Lignet et al. (Citation2016). Utilising in-house and published in vivo data, the refined set covered relevant dynamic ranges for all parameters (); a breadth of physico-chemical properties including all ion classes and spanned a number of the most important drug-metabolising enzymes (). As such, it is the authors opinion this refined reference set can be used by others to establish their own laboratory specific ‘regression offset’ for IVIVE in Göttingen minipig. The predicted versus observed CL_hep,met are visualised for the five IVIVE methodologies in panels A–C and panels C and F and their predictive performance summarised in . In other species, IVIVE using the WSM where all binding parameters are included (Method 1), whilst mechanistically sound and showing good correlation between predicted and observed, has generally resulted in significant under-prediction bias and lack of precision. Using Method 1, we also confirm a substantial under-prediction bias in minipig with R², AFE, AAFE and percent within 2-fold of 0.69, 2.5, 2.7, and 58%, respectively, in-line with findings for rat (Sohlenius-Sternbeck et al. Citation2012; Poulin Citation2013; Francis et al. Citation2021) and human (Obach Citation1999; Ito and Houston Citation2004; Riley et al. Citation2005; Sohlenius-Sternbeck et al. Citation2010; Poulin Citation2013; Francis et al. Citation2021; Jones et al. Citation2022). If non-specific incubation binding is assumed to be negligible (Method 2, fu_inc = 1) the magnitude of under-prediction bias increases and both accuracy and correlation decrease (R², AFE, AAFE and percent within 2-fold = 0.51, 4.0, 4.1, and 38%, respectively). Others (Obach Citation1999; Sohlenius-Sternbeck et al. Citation2010; Sohlenius-Sternbeck et al. Citation2012) have shown exclusion of all binding parameters (Method 3, fu_b/fu_inc = 1) can lead to significant improvements in IVIVE for certain chemotypes (typically neutral or basic compounds at physiological pH). We also show in minipig exclusion of all binding parameters improves the IVIVE (R², AFE, AAFE and percent within 2-fold = 0.65, 1.1, 1.4, and 83%, respectively) but conclude mechanistically this is unlikely a consequence of binding parameters cancelling out since fu_b and fu_inc were poorly correlated (Supplementals, Worksheet Tab 5). This adaption to the WSM has not been widely accepted and analysis on larger data sets in other species suggests inclusion of these drug binding terms is important for IVIVE and prediction of CL_hep,met (Riley Citation2005; Grime and Riley Citation2006). It should also be noted Method 3 is less suited for highly protein bound compounds and cannot be uniformly applied to all ion classes given significant over-prediction bias observed with acidic and, to a lesser extent, zwitterionic compounds (Sohlenius-Sternbeck et al. Citation2010). We note in the current study only 16% of the refined reference set for minipig were acidic/zwitterionic.

With Methods 4 and 5 the IVIVE performance issues commonly observed with Methods 1–3 are resolved using an empirical ‘regression offset’ correction factor established through linear regression analysis on a well-defined reference set (Sohlenius-Sternbeck et al. Citation2010, Citation2012). Both these methods afforded inspection of IVIVE on logarithmic transformed CL_int axes as well as CL_hep,met axes (); the former recognised as the more appropriate parameter for quantitative regression analysis (Ito and Houston Citation2004; Riley et al. Citation2005). With Method 4, the CL_int,app and binding terms fu_b and fu_inc were incorporated on the X-axis so that only the observed in vivo CL_int (obtained after rearranging the well-stirred model) was on the Y-axis. This method gave the best IVIVE performance in minipig ( and and panels B and C) with good correlation, minimal under-prediction bias and superior precision and accuracy for the regression corrected predicted in vivo CL_int (R², AFE, AAFE and percent within 2-fold = 0.61, 1.0, 1.9, and 67% respectively) and predicted CL_hep,met (R², AFE, AAFE and percent within 2-fold = 0.70, 0.9, 1.4, and 96%, respectively). Whilst an independent test set would provide valuable corroboration of the method, encouragingly, others have shown in rat and human that a comparable level of performance was achieved for regression reference sets and independent test sets (Sohlenius-Sternbeck et al. Citation2012; Grime et al. Citation2013). With Method 5, the linear regression analysis was established using the more traditional approach based on logarithmic transformed unbound CL_int axes ( panels D and E). In minipig, regression corrected predictions from Method 5 performed better than IVIVE Methods 1 and 3 (R², AFE, AAFE and percent within 2-fold = 0.52, 1.0, 2.9, and 46%, respectively, for predicted in vivo CL_int and R², AFE, AAFE and percent within 2-fold = 0.61, 1.1, 1.9, and 83%, respectively, for predicted CL_hep,met) but inferior to Methods 2 and 4. The improvement in IVIVE with Method 4 over Method 5 in minipig has been reported previously in human (Sohlenius-Sternbeck et al. Citation2012) and also observed in rat according to unpublished data on proprietary compounds generated within our laboratory. However, in real terms, CL_hep,met predicted from the two regression models are somewhat similar; as shown by the strong correlation in . Consequently whilst IVIVE using Method 4 is the recommended approach, those preferring to establish IVIVE on the unbound CL_int axes can expect to achieve broadly similar results for prediction of CL_hep,met using Method 5.

Figure 6. A comparison of the predicted blood CL_hep,met from IVIVE Methods 4 and 5 is presented. The solid lines represent line of unity and dotted lines represent line of best fit from linear regression. Closed circles represent recommended reference compounds to establish the empirical regression correction and subsequently used in the statistical analyses () of the IVIVE of predicted in vivo CL_int and predicted CL_hep,met.

In order to provide a complete in vivo dataset in the same breed and gender future work will be directed to generating additional in vivo PK in female Göttingen minipigs for those reference compounds proposed in where IV PK currently exists only in male NIBS (buprenorphine, lidocaine, raloxifene, and fexofenadine) or male Göttingen (diclofenac and ketoprofen) minipigs. Work is also ongoing in our laboratory to bring forward recommended reference compounds to establish equivalent ‘regression offsets’ across species including male Sprague–Dawley rat, male beagle dog, male cynomolgus monkey and human, such that the approach can be readily adopted in other laboratories without necessitating repeat of in vivo PK; the only requirement being to generate the in vitro datasets using their assays to establish their own lab-specific regression offset equations.

Conclusion

In summary, the purpose of this manuscript has been to share a substantial new PK dataset generated in female Göttingen minipig that can be used across the broader research community interested in validating and refining PBPK models for minipig; in context of growing momentum for PK applications such as IVIVE and human PK. Furthermore, this work highlights for the first time that an empirical, regression corrected, IVIVE can be established in minipig providing the basis for an alternate non-rodent species for assessment of IVIVE in support of robust human CL predictions. Finally, a refined reference set is proposed that other researchers can use to establish their own regression offset for improved IVIVE of CL_hep,met in minipig.

Supplemental material

IXEN_2115425_SuppMat.xlsx

Download MS Excel (285.5 KB)

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This work was supported jointly by H. Lundbeck A/S and Innovation Fund Denmark under [Grant number 9065-00189B].

References

Achour B, Barber J, Rostami-Hodjegan A. 2011. Cytochrome P450 Pig liver pie: determination of individual cytochrome P450 isoform contents in microsomes from two pig livers using liquid chromatography in conjunction with mass spectrometry. Drug Metab Dispos. 39(11):2130–2134.
PubMed Web of Science ®Google Scholar
Austin RP, Barton P, Cockroft SL, Wenlock MC, Riley RJ. 2002. The influence of nonspecific microsomal binding on apparent intrinsic clearance, and its prediction from physicochemical properties. Drug Metab Dispos. 30(12):1497–1503.
PubMed Web of Science ®Google Scholar
Austin RP, Barton P, Mohmed S, Riley RJ. 2005. The binding of drugs to hepatocytes and its relationship to physicochemical properties. Drug Metab Dispos. 33(3):419–425.
PubMed Web of Science ®Google Scholar
Bode G, Clausing P, Gervais F, Loegsted J, Luft J, Nogues V, Sims J, Steering Group of the RP. 2010. The utility of the minipig as an animal model in regulatory toxicology. J Pharmacol Toxicol Methods. 62(3):196–220.
PubMed Web of Science ®Google Scholar
Bollen P, Ellegaard L. 1997. The Gottingen minipig in pharmacology and toxicology. Pharmacol Toxicol. 80( Suppl 2):3–4.
PubMedGoogle Scholar
Bollen PJA, Madsen LW, Meyer O, Ritskes-Hoitinga J. 2005. Growth differences of male and female Göttingen minipigs during ad libitum feeding: a pilot study. Lab Anim. 39(1):80–93.
PubMed Web of Science ®Google Scholar
Bridges TM, Morrison RD, Byers FW, Luo S, Scott Daniels J. 2014. Use of a novel rapid and resource-efficient cassette dosing approach to determine the pharmacokinetics and CNS distribution of small molecule 7-transmembrane receptor allosteric modulators in rat. Pharmacol Res Perspect. 2(6):e00077.
PubMedGoogle Scholar
Buyssens L, De Clerck L, Schelstraete W, Dhaenens M, Deforce D, Ayuso M, Van Ginneken C, Van Cruchten S. 2021. Hepatic cytochrome P450 abundance and activity in the developing and adult Gottingen minipig: pivotal data for PBPK modeling. Front Pharmacol. 12:665644.
PubMed Web of Science ®Google Scholar
Cerny MA. 2016. Prevalence of non-cytochrome P450-mediated metabolism in food and drug administration-approved oral and intravenous drugs: 2006–2015. Drug Metab Dispos. 44(8):1246–1252.
PubMed Web of Science ®Google Scholar
Dalgaard L. 2015. Comparison of minipig, dog, monkey and human drug metabolism and disposition. J Pharmacol Toxicol Methods. 74:80–92.
PubMed Web of Science ®Google Scholar
Ding N, Yamamoto S, Chisaki I, Nakayama M, Matsumoto SI, Hirabayashi H. 2021. Utility of Gottingen minipigs for the prediction of human pharmacokinetic profiles after intravenous drug administration. Drug Metab Pharmacokinet. 41:100408.
PubMed Web of Science ®Google Scholar
Forster R, Bode G, Ellegaard L, van der Laan JW. 2010. The RETHINK project on minipigs in the toxicity testing of new medicines and chemicals: conclusions and recommendations. J Pharmacol Toxicol Methods. 62(3):236–242.
PubMed Web of Science ®Google Scholar
Francis LJ, Houston JB, Hallifax D. 2021. Impact of plasma protein binding in drug clearance prediction: a data base analysis of published studies and implications for in vitro-in vivo extrapolation. Drug Metab Dispos. 49(3):188–201.
PubMed Web of Science ®Google Scholar
Ganderup NC, Harvey W, Mortensen JT, Harrouk W. 2012. The minipig as nonrodent species in toxicology–where are we now? Int J Toxicol. 31(6):507–528.
PubMed Web of Science ®Google Scholar
Grime K, Barton P, McGinnity D. 2013. Application of in silico, in vitro and preclinical pharmacokinetic data for the effective and efficient prediction of human pharmacokinetics. Mol Pharm. 10(4):1191–1206.
PubMed Web of Science ®Google Scholar
Grime K, Riley RJ. 2006. The impact of in vitro binding on in vitro–in vivo extrapolations, projections of metabolic clearance and clinical drug-drug interactions. Curr Drug Metab. 7(3):251–264.
PubMed Web of Science ®Google Scholar
Hallifax D, Foster JA, Houston JB. 2010. Prediction of human metabolic clearance from in vitro systems: retrospective analysis and prospective view. Pharm Res. 27(10):2150–2161.
PubMed Web of Science ®Google Scholar
Hallifax D, Houston JB. 2012. 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. 101(8):2645–2652.
PubMed Web of Science ®Google Scholar
Helke KL, Nelson KN, Sargeant AM, Jacob B, McKeag S, Haruna J, Vemireddi V, Greeley M, Brocksmith D, Navratil N, et al. 2016. Pigs in toxicology: breed differences in metabolism and background findings. Toxicol Pathol. 44(4):575–590.
PubMed Web of Science ®Google Scholar
Henze LJ, Koehl NJ, O'Shea JP, Kostewicz ES, Holm R, Griffin BT. 2019. The pig as a preclinical model for predicting oral bioavailability and in vivo performance of pharmaceutical oral dosage forms: a PEARRL review. J Pharm Pharmacol. 71(4):581–602.
PubMed Web of Science ®Google Scholar
Houston JB. 1994. Utility of in vitro drug metabolism data in predicting in vivo metabolic clearance. Biochem Pharmacol. 47(9):1469–1479.
PubMed Web of Science ®Google Scholar
Ito K, Houston JB. 2004. Comparison of the use of liver models for predicting drug clearance using in vitro kinetic data from hepatic microsomes and isolated hepatocytes. Pharm Res. 21(5):785–792.
PubMed Web of Science ®Google Scholar
Jones HM, Chen Y, Gibson C, Heimbach T, Parrott N, Peters SA, Snoeys J, Upreti VV, Zheng M, Hall SD. 2015. Physiologically based pharmacokinetic modeling in drug discovery and development: a pharmaceutical industry perspective. Clin Pharmacol Ther. 97(3):247–262.
PubMed Web of Science ®Google Scholar
Jones HM, Gardner IB, Collard WT, Stanley PJ, Oxley P, Hosea NA, Plowchalk D, Gernhardt S, Lin J, Dickins M, et al. 2011. Simulation of human intravenous and oral pharmacokinetics of 21 diverse compounds using physiologically based pharmacokinetic modelling. Clin Pharmacokinet. 50(5):331–347.
PubMed Web of Science ®Google Scholar
Jones RS, Leung C, Chang JH, Brown S, Liu N, Yan Z, Kenny JR, Broccatelli F. 2022. Application of empirical scalars to enable early prediction of human hepatic clearance using IVIVE in drug discovery: an evaluation of 173 drugs. Drug Metabo Dispos. 50(8):1053–1063.
Web of Science ®Google Scholar
Jones HM, Parrott N, Jorga K, Lave T. 2006. A novel strategy for physiologically based predictions of human pharmacokinetics. Clin Pharmacokinet. 45(5):511–542.
PubMed Web of Science ®Google Scholar
Kilford PJ, Gertz M, Houston JB, Galetin A. 2008. Hepatocellular binding of drugs: correction for unbound fraction in hepatocyte incubations using microsomal binding or drug lipophilicity data. Drug Metab Dispos. 36(7):1194–1197.
PubMed Web of Science ®Google Scholar
Landqvist C, Middleton B, Jones B, O'Donnell C. 2014. A novel global approach taken by AstraZeneca to monitor primary DMPK assay performance and understand the inter and intra site assay variability. Drug Discov World. 15:59–63.
Google Scholar
Lignet F, Sherbetjian E, Kratochwil N, Jones R, Suenderhauf C, Otteneder MB, Singer T, Parrott N. 2016. Characterization of pharmacokinetics in the Gottingen minipig with reference human drugs: an in vitro and in vivo approach. Pharm Res. 33(10):2565–2579.
PubMed Web of Science ®Google Scholar
Lin C, Shi J, Moore A, Khetani SR. 2016. Prediction of drug clearance and drug-drug interactions in microscale cultures of human hepatocytes. Drug Metab Dispos. 44(1):127–136.
PubMed Web of Science ®Google Scholar
Liu X, Ding X, Deshmukh G, Liederer BM, Hop CE. 2012. Use of the cassette-dosing approach to assess brain penetration in drug discovery. Drug Metab Dispos. 40(5):963–969.
PubMed Web of Science ®Google Scholar
Lombardo F, Shalaeva MY, Tupper KA, Gao F. 2001. ElogDoct: a tool for lipophilicity determination in drug discovery. 2. Basic and neutral compounds. J Med Chem. 44(15):2490–2497.
PubMed Web of Science ®Google Scholar
Manitpisitkul P, White RE. 2004. Whatever happened to cassette-dosing pharmacokinetics? Drug Discov Today. 9(15):652–658.
PubMed Web of Science ®Google Scholar
McGinnity DF, Collington J, Austin RP, Riley RJ. 2007. Evaluation of human pharmacokinetics, therapeutic dose and exposure predictions using marketed oral drugs. Curr Drug Metab. 8(5):463–479.
PubMed Web of Science ®Google Scholar
Miller NA, Reddy MB, Heikkinen AT, Lukacova V, Parrott N. 2019. Physiologically based pharmacokinetic modelling for first-in-human predictions: an updated model building strategy illustrated with challenging industry case studies. Clin Pharmacokinet. 58(6):727–746.
PubMed Web of Science ®Google Scholar
Nagilla R, Nord M, McAtee JJ, Jolivette LJ. 2011. Cassette dosing for pharmacokinetic screening in drug discovery: comparison of clearance, volume of distribution, half-life, mean residence time, and oral bioavailability obtained by cassette and discrete dosing in rats. J Pharm Sci. 100(9):3862–3874.
PubMed Web of Science ®Google Scholar
Nunoya T, Shibuya K, Saitoh T, Yazawa H, Nakamura K, Baba Y, Hirai T. 2007. Use of miniature pig for biomedical research, with reference to toxicologic studies. J Toxicol Pathol. 20(3):125–132.
Google Scholar
Obach RS. 1999. Prediction of human clearance of twenty-nine drugs from hepatic microsomal intrinsic clearance data: an examination of in vitro half-life approach and nonspecific binding to microsomes. Drug Metab Dispos. 27(11):1350–1359.
PubMed Web of Science ®Google Scholar
Obach RS, Lombardo F, Waters NJ. 2008. Trend analysis of a database of intravenous pharmacokinetic parameters in humans for 670 drug compounds. Drug Metab Dispos. 36(7):1385–1405.
PubMed Web of Science ®Google Scholar
Pang KS, Rowland M. 1977. Hepatic clearance of drugs. I. Theoretical considerations of a "well-stirred" model and a "parallel tube" model. Influence of hepatic blood flow, plasma and blood cell binding, and the hepatocellular enzymatic activity on hepatic drug clearance. J Pharmacokinet Biopharm. 5(6):625–653.
PubMedGoogle Scholar
Patel NJ, Yumibe N, Ruterbories K, Huang N, Burns L, Tan J, White D, Liu J, Brocksmith D, Bouchard G, et al. 2017. Pharmacokinetics of intravenous and oral metformin and r,s-verapamil in Sinclair, Hanford, Yucatan and Göttingen minipigs. Int J Pharmacokinet. 2(2):81–91.
Google Scholar
Poulin P. 2013. Prediction of total hepatic clearance by combining metabolism, transport, and permeability data in the in vitro-in vivo extrapolation methods: emphasis on an apparent fraction unbound in liver for drugs. J Pharm Sci. 102(7):2085–2095.
PubMed Web of Science ®Google Scholar
Poulin P, Collet SH, Atrux-Tallau N, Linget JM, Hennequin L, Wilson CE. 2019. Application of the tissue composition-based model to minipig for predicting the volume of distribution at steady state and dermis-to-plasma partition coefficients of drugs used in the physiologically based pharmacokinetics model in dermatology. J Pharm Sci. 108(1):603–619.
PubMed Web of Science ®Google Scholar
Riley RJ, McGinnity DF, Austin RP. 2005. A unified model for predicting human hepatic, metabolic clearance from in vitro intrinsic clearance data in hepatocytes and microsomes. Drug Metab Dispos. 33(9):1304–1311.
PubMed Web of Science ®Google Scholar
Sheiner LB, Beal SL. 1981. Some suggestions for measuring predictive performance. J Pharmacokinet Biopharm. 9(4):503–512.
PubMedGoogle Scholar
Siefert HM, Domdey-Bette A, Henninger K, Hucke F, Kohlsdorfer C, Stass HH. 1999. Pharmacokinetics of the 8-methoxyquinolone, moxifloxacin: a comparison in humans and other mammalian species. J Antimicrob Chemother. 43( Suppl B)):69–76.
PubMed Web of Science ®Google Scholar
Simianer H, Kohn F. 2010. Genetic management of the Gottingen minipig population. J Pharmacol Toxicol Methods. 62(3):221–226.
PubMed Web of Science ®Google Scholar
Singh VK, Thrall KD, Hauer-Jensen M. 2016. Minipigs as models in drug discovery. Expert Opin Drug Discov. 11(12):1131–1134.
PubMed Web of Science ®Google Scholar
Skaanild MT. 2006. Porcine cytochrome P450 and metabolism. Curr Pharm Des. 12(11):1421–1427.
PubMed Web of Science ®Google Scholar
Smith NF, Raynaud FI, Workman P. 2007. The application of cassette dosing for pharmacokinetic screening in small-molecule cancer drug discovery. Mol Cancer Ther. 6(2):428–440.
PubMed Web of Science ®Google Scholar
Sohlenius-Sternbeck A-K, Afzelius L, Prusis P, Neelissen J, Hoogstraate J, Johansson J, Floby E, Bengtsson A, Gissberg O, Sternbeck J, et al. 2010. Evaluation of the human prediction of clearance from hepatocyte and microsome intrinsic clearance for 52 drug compounds. Xenobiotica. 40(9):637–649.
PubMed Web of Science ®Google Scholar
Sohlenius-Sternbeck A-K, Jones C, Ferguson D, Middleton BJ, Projean D, Floby E, Bylund J, Afzelius L. 2012. Practical use of the regression offset approach for the prediction of in vivo intrinsic clearance from hepatocytes. Xenobiotica. 42(9):841–853.
PubMed Web of Science ®Google Scholar
Suenderhauf C, Parrott N. 2013. A physiologically based pharmacokinetic model of the minipig: data compilation and model implementation. Pharm Res. 30(1):1–15.
PubMed Web of Science ®Google Scholar
Suenderhauf C, Tuffin G, Lorentsen H, Grimm H-P, Flament C, Parrott N. 2014. Pharmacokinetics of paracetamol in Göttingen minipigs: in vivo studies and modeling to elucidate physiological determinants of absorption. Pharm Res. 31(10):2696–2707.
PubMed Web of Science ®Google Scholar
Swindle MM, Makin A, Herron AJ, Clubb FJ Jr, Frazier KS. 2012. Swine as models in biomedical research and toxicology testing. Vet Pathol. 49(2):344–356.
PubMed Web of Science ®Google Scholar
Tang H, Hussain A, Leal M, Mayersohn M, Fluhler E. 2007. Interspecies prediction of human drug clearance based on scaling data from one or two animal species. Drug Metab Dispos. 35(10):1886–1893.
PubMed Web of Science ®Google Scholar
Thorn HA, Lundahl A, Schrickx JA, Dickinson PA, Lennernas H. 2011. Drug metabolism of CYP3A4, CYP2C9 and CYP2D6 substrates in pigs and humans. Eur J Pharm Sci. 43(3):89–98.
PubMed Web of Science ®Google Scholar
Unger SH, Chiang GH. 1981. Octanol-physiological buffer distribution coefficients of lipophilic amines by reversed-phase high-performance liquid chromatography and their correlation with biological activity. J Med Chem. 24(3):262–270.
PubMed Web of Science ®Google Scholar
Wilkinson DJ, Southall RL, Li M, Wright LM, Corfield LJ, Heeley TA, Bratby B, Mannu R, Johnson SL, Shaw V, et al. 2017. Minipig and human metabolism of aldehyde oxidase substrates: in vitro-in vivo comparisons. AAPS J. 19(4):1163–1174.
PubMed Web of Science ®Google Scholar
Williamson B, Harlfinger S, McGinnity DF. 2020. Evaluation of the disconnect between hepatocyte and microsome intrinsic clearance and in vitro in vivo extrapolation performance. Drug Metab Dispos. 48(11):1137–1146.
PubMed Web of Science ®Google Scholar
Yesair DW, Callahan M, Remington L, Kensler CJ. 1970. Role of the entero-hepatic cycle of indomethacin on its metabolism, distribution in tissues and its excretion by rats, dogs and monkeys. Biochem Pharmacol. 19(5):1579–1590.
PubMed Web of Science ®Google Scholar
Yoshimatsu H, Konno Y, Ishii K, Satsukawa M, Yamashita S. 2016. Usefulness of minipigs for predicting human pharmacokinetics: prediction of distribution volume and plasma clearance. Drug Metab Pharmacokinet. 31(1):73–81.
PubMed Web of Science ®Google Scholar
Zhu C, Jiang L, Chen TM, Hwang KK. 2002. A comparative study of artificial membrane permeability assay for high throughput profiling of drug absorption potential. Eur J Med Chem. 37(5):399–407.
PubMed Web of Science ®Google Scholar

Characterisation of intravenous pharmacokinetics in Göttingen minipig and clearance prediction using established in vitro to in vivo extrapolation methodologies

Abstract

Introduction