Strategy for Identifying Rational Sensitivity Analysis Using PBPK Modeling for Precipitant Drug–Drug Interaction Predictions

Abstract

Physiology Based Pharmacokinetic (PBPK) modeling is an established essential tool for predicting and/or analyzing drug–drug interactions (DDI). Uncertainty and variability associated with in vitro determined DDI‐related parameters have often been considered a limitation for predicting PBPK‐DDIs. Sensitivity analysis (SA) around DDI input parameters using PBPK analysis is often applied for assessing the relevance of clinical DDI predictions/prioritization/study designs. This perspective aims to explore and advocate practical approaches for precipitant (inhibitor/inducer) PBPK‐DDI SA for optimal clinically relevant evaluations.

PBPK model outcomes depend on the reliability of model structure and input parameters. Specifically, for DDI predictions, in vitro determined IC₅₀/K _i/k _inact or E _max/EC₅₀ values for precipitant DDIs and appropriate estimates for fraction metabolized (f _m) and/or transported for object DDIs are crucial determinants. Regulatory recommendations for PBPK models have indicated less confidence in transporter‐mediated DDI precipitant applications than in CYP‐mediated ones, reflecting perceived uncertainty associated with in vitro transporter inhibition values. ¹ There is an expectation of conducting SA around in vitro DDI values for a precipitant drug, especially in the absence of a confirmatory clinical study. Regulatory agencies commonly refer to in vitro variability reported in various databases such as the drug interactions solutions database (DIDB®) to recommend a SA range, reflective of different assay systems or lab‐to‐lab differences in procedures. ² Since these values are generated through non‐harmonized controlled in vitro assay studies, they may not provide credible ranges for SA. Furthermore, there is no defined harmonized strategy across industry, regulatory and scientific communities. For example, the European Medical Agency (EMA) guideline on reporting of PBPK modeling mentions a 10‐fold SA for inhibitory parameters of CYP enzymes and 30‐fold for transporters, both indicated as “worst‐case” scenarios. ¹ Often arbitrarily assigned, a pertinent rationale for defining an appropriate SA range in PBPK‐DDI applications is essential.

PBPK‐DDI SA for enzyme inhibition

In vitro assays are used to generate IC₅₀/K _i values of drug metabolizing enzymes and transporters (DMETs) for DDI assessments. ³ In early development, a PBPK model validated using pharmacokinetic (PK) data can be applied for DDI assessment as a perpetrator. If a new chemical entity (NCE) is determined to be a CYP inhibitor, SA around the relevant in vitro determined K _i value is expected because this value may lead to over/underpredictions of the clinical DDI. The key question is what is an appropriate range to apply? We propose a strategy to determine an in vitro to in vivo correlation that may be established for calibrating and identifying a sensitivity range for the K _i value of an NCE.

For most of the clinically relevant DMETs, there is at least one identified index/sensitive clinically relevant substrate and strong inhibitor which can be used both for in vitro and clinical DDI assessments (Table 1). ³ Such substrate–inhibitor pairs are used in in vitro studies as positive controls, with the precipitant NCE, a test case, against the same substrate; for example, midazolam as a sensitive/index CYP3A substrate and itraconazole as a strong CYP3A inhibitor. The clinical DDI between midazolam and itraconazole has been thoroughly characterized. Combining in vitro DDI assays, clinical DDIs and verified PBPK models for both midazolam and itraconazole, an in vitro to clinical extrapolation for DDI (IVCE‐DDI) can be robustly estimated. The measured IC₅₀/K _i value for the positive control inhibitor determined in a routine industry standard assay can be compared to the clinically calibrated value. The fold‐difference (in vitro vs. calibrated) can be used to perform an appropriate SA around the NCE K _i value for the DDI assessment. This approach would overcome the issues associated with subjective selection of a SA range based on wide variability of values determined from in vitro assays across laboratories for the same positive control (DIDB®);for example, there is a 10,000‐fold difference in reported K _i values for itraconazole (0.0013–11 μM); yet, the lowest in vitro value can predict the observed clinical DDI. ² For cimetidine, a weak CYP3A inhibitor, the fold difference across reported values is small (K _i ≈ 35.8–370 μM), but the calibrated K _i value of 11 μM falls outside the reported in vitro range. ² For an OAT1 inhibitor, there is a ~36‐fold difference in the reported in vitro IC₅₀ range (3.6–130 μM) and the clinically calibrated value (3.5 μM) is similar to the most potent value. ²

Table 1.

List of substrate–precipitant pairs and clinically calibrated/estimated IC₅₀ or K _i or EC₅₀/E _max values for the precipitant

Enzyme/Transporter	Substrate	Inhibitor/Inducer
CYP3A4/5	Midazolam	Itraconazole and hydroxy metabolite/Multiple dose Rifampicin
CYP1A2	Caffeine	Fluvoxamine
CYP2C8	Repaglinide	Gemfibrozil
CYP2C9	S‐warfarin	Fluconazole
CYP2C19	Omeprazole	Fluvoxamine
CYP2D6	Desipramine	Fluoxetine
P‐gp	Digoxin	Verapamil
BCRP	Rosuvastatin	Cyclosporine
OATP1B1/1B3	Rosuvastatin	Single dose Rifampicin
OAT1	Furosemide	Probenecid
OAT3	Furosemide	Probenecid
OCT2	Metformin	Dolutegravir
MATE1/2‐K	Metformin	Pyrimethamine

Reference: ICH M12 Guideline. ³

Depending on the deviation of the in vitro value from the clinically calibrated one, a suitable rationale can be applied. When the measured IC₅₀ for the positive control inhibitor is less potent than the clinically calibrated value, the fold difference can be used for the SA range on the IC₅₀/K _i of the NCE test drug. Alternatively, if the positive control inhibitor is more potent in vitro than in vivo, it may be more appropriate to assume that the in vitro IC₅₀ for the NCE is the most conservative value, thus not requiring SA.

Successful implementation of the IVCE‐DDI strategy depends on the use of a clinically relevant object‐precipitant pair in in vitro assays and robust clinical DDI characterization of this object‐precipitant pair. The availability of validated PBPK models for the object‐precipitant pair is essential to obtain a clinically calibrated IC₅₀/K _i value. This approach would help harmonize SA analysis around IC₅₀/K _i values generated across various laboratories and in vitro systems.

Metabolism‐ or time‐dependent inhibition complicates the application of the IVCE‐DDI approach for an NCE, but having clinically relevant object‐precipitant pairs can help.

PBPK‐DDI SA for transporter inhibition

The IVCE‐DDI approach can also be applied to transporter‐mediated PBPK DDI assessments for an NCE. Established transporter substrate‐inhibitor pairs are used routinely as positive controls in in vitro assessments (Table 1 ). It is acknowledged that there are limited examples for clinical DDIs for such substrate–inhibitor pairs and some drugs maybe substrates of multiple transporters. Rosuvastatin is often used as a BCRP clinical probe, but it is also considered to be a clinical probe for OATP1B1/1B3. As the fraction transported by the clinically relevant transporters (BCRP/OATP1B1/1B3) has been sufficiently characterized for rosuvastatin, even with the aforementioned limitations, the IVCE‐DDI would be a good starting point to identify a suitable SA. Indeed, a similar IVCE‐DDI approach was used to justify the SA range applied for a cabotegravir OAT1/3‐mediated inhibition DDI prediction using PBPK modeling and probenecid (positive control in the in vitro assays). ⁴ Recent use of endogenous transporter biomarkers combined with PBPK models and clinical data enables accurate assessment of transporter K _i/IC₅₀ values. ⁵

PBPK‐DDI SA for enzyme induction values

PBPK‐DDI assessments for enzyme induction use EC₅₀ and E _max values. Although it is possible to perform SA on both parameters, typically EC₅₀, which is affected by drug concentration, is the focus. Rifampicin is a well‐established clinical inducer of several DMETs as well as being routinely used as a positive control in vitro along with midazolam when assessing CYP3A induction potential. There is an established and verified rifampicin PBPK model with EC₅₀/E _max values calibrated from clinical DDI studies conducted with the sensitive CYP3A substrate midazolam. Hence, comparing the rifampicin EC₅₀ value from the in vitro induction assay with that of the clinically calibrated value may help to characterize and identify a practical SA range for the NCE test compound using PBPK analysis (Figure 1 ).

Figure 1.

General workflow to identify a suitable Sensitivity Analysis (SA) and a representative example for identification of SA range using IVCE‐DDI approach for PBPK‐DDI for an NCE as a CYP3A inducer with rifampicin‐midazolam used as inducer‐substrate positive control from various runs of in vitro assay done on multiple occasions from an individual lab. ^#Please note that these data values are examples from an internal dataset and will vary between labs. Sensitivity analysis must be tailored to specific in vitro assay conditions and the PBPK platform used. They require re‐optimization and justification if assay conditions or test systems change. *Reddy et al. IQ paper 2024. ⁷

PBPK‐DDI SA around physiology‐related parameters

Apart from the aforementioned DDI input parameters, it is important to consider other values which may have perceived uncertainty, that is, those related to in vitro systems, physicochemical properties of a drug or clinical pathology/physiology of a disease. Drug properties which influence the PK of the drug include plasma protein binding (PPB), lipophilicity or permeability amongst others. While there may be uncertainty associated with each of these parameters, a certain level of confidence (in the values used) is achieved when the PBPK model PK predictions of the precipitant NCE are verified using clinical data. Hence, only parameters with significant uncertainty in in vitro assessments, for example, PPB for very highly bound drugs (f _u ≤ 0.001) which may prove to be highly sensitive for DDI outcomes when combined with the in vitro DDI values should be considered for SA.

Physiological factors with uncertainty that can influence DDI outcomes include but are not limited to the disease impact on DMETs expression or activity, disease severity, polymorphisms, gender, diet, lifestyles, ethnicity, and age. If critical, any relevant parameter with significant uncertainty or variability, should be considered for SA when assessing the impact on DDIs using PBPK modeling; for example, using diverse CYP3A ontogenies from two independent studies for pediatric DDI assessments. ⁶

Local and global sensitivity analysis for DDI relevant parameters

In the context of PBPK‐DDI SA for a precipitant, local sensitivity analysis considers a single DDI mechanism. However, there may be multiple parameters that contribute to the magnitude of the DDI, for example, a precipitant may have more than one mechanism for a DDI pathway (e.g., both induction and inhibition of CYP3A). Furthermore, the dose may change depending on the clinical indication of the precipitant drug. In such cases, an ideal approach is to carry out PBPK‐DDI SA around each parameter/mechanism separately, followed by a global SA to capture various clinical scenarios. ⁴ For drugs that are inhibitors of more than one transporter, this IVCE‐DDI approach may be considered in combination with a suitable endogenous biomarker approach. For drugs that are both OATP1B and BCRP inhibitors, the endogenous biomarker approach may be used to tease out the OATP1B inhibition value from clinic, and for BCRP, the IVCE approach may be applied.

PBPK‐DDI SA for precipitant exposure outcomes

PBPK models for a NCE, in terms of exposure, are validated when data become available from clinical studies. In early clinical development until the efficacious dose is optimized, SA around the clinically relevant dose of the precipitant NCE to test the impact on the extent of the DDI with an index object‐drug PBPK model is a good practice. Critical parameters for a precipitant are the AUC and Cmax when the systemic DDI with enzyme or transporter is relevant or it is the dose when the precipitant DDI is expected due to alteration of gut enzyme/transporters. The PBPK‐DDI sensitivity range around the dose should cover the extreme dose levels that would be tested for the precipitant NCE and supratherapeutic levels when no DDI is predicted at the therapeutic dose. The clinically established therapeutic dose should be considered for any decisive DDI‐related extrapolations in the clinic. The top dose for sensitivity analysis can be dictated by the maximum tolerated dose, if established. Formulation‐related changes can also be considered especially when the precipitant has no or weak risk predicted by a PBPK model and any changes in formulation would be sensitive to the precipitant exposure.

Concluding remarks

A practical approach for identifying a suitable SA range when using PBPK‐DDI assessments is essential and requires consortium/industry/regulatory‐agencies‐wide collaboration and alignment. Development of best‐practice documents required to standardize the IVCE‐DDI approach could be considered relevant in suitable collaborations. IVCE‐DDI is suggested as a starting point to establish a suitable SA range for PBPK‐DDI analysis and depending on the identified uncertain or variable input parameters, a local or global SA may be considered.

CONFLICT OF INTEREST

K.S.T. is an employee of and may hold stock in GSK, P.S. is an employee of and may hold stock in AstraZeneca, and K.R.Y. is an employee of and may hold stock in Certara UK Ltd. As Editor‐in‐Chief of Clinical Pharmacology & Therapeutics, K.R.Y. was not involved in the review or decision process for this paper.