Evaluation of Solubility-Limited Absorption as a Surrogate to Predicting Positive Food Effect of BCS II/IV Drugs

The effect of food on oral drug absorption is a complex phenomenon that has long captivated the attention of pharmaceutical researchers and clinicians [1]. Food can significantly alter the pharmacokinetics (PK) of orally administered drugs by modifying the rate and extent of drug absorption, thereby impacting both their therapeutic efficacy and safety [2]. Food-induced physiological changes affecting absorption include changes in gastric emptying, gastric pH, luminal drug concentrations, bile secretion, and splanchnic blood flow [3]. Understanding the effect of food on the systemic exposure of an orally administered drug is a critical aspect of drug development, regulatory requirements, and clinical practice. During drug development, a pilot food effect (FE) study is often conducted as part of a phase I clinical trial for drugs that are expected to have low bioavailability to inform dosing with or without food in subsequent trials. Later, a pivotal FE study must be conducted with the final dose and formulation of the drug to inform drug labeling in support of new drug application (NDA) submission [4, 5], unless the investigational drug is a biopharmaceutical classification system (BCS) class I drug with an oral bioavailability ≥ 85%. Predictions of FE can significantly streamline drug development by guiding formulation strategies, optimizing clinical trial designs, and informing dosing recommendations. In early clinical development, decisions on the need to conduct a time- and resource-consuming FE study in phase I require a binary (yes/no) prediction with high confidence rather than a quantitative prediction. A reliable, conservative prediction of a lack of FE for the drug of interest can help avert pilot FE studies, as well as delay the pivotal FE study until later in the drug development process when there is more certainty about the drug’s market potential.

Current methods to predict human FE during drug development include in vitro and in silico tools [6]. In vitro methods include simple solubility- and permeability-based models such as BCS classification [7], predictions based on physicochemical properties such as dose number and maximum absorbable dose [8, 9], and compendial dissolution methods using fasted and fed state simulated intestinal fluid (FaSSIF/FeSSIF), as well as complex in vitro tools such as TIM and dynamic gastric model (DGM) systems [10]. There is a growing interest in the application of in silico physiologically based pharmacokinetic (PBPK) models (e.g., SimCYP, GastroPlus, OSP Suite) that incorporate various factors influencing oral drug absorption for quantitative prediction of FE [11‐18]. However, both in vitro and in silico tools have some limitations in predicting FE on drug absorption. In vitro studies often simplify gastrointestinal environments and operate under static environments leading to in vitro–in vivo (IVIV) disconnect and resulting in a systematic underprediction of in vivo drug absorption. Methods employing dose number assume a luminal fluid volume of 250 ml [19], whose relevance in vivo is not verifiable. PBPK models describe oral drug absorption with processes such as dissolution, precipitation, solubilization, etc., each with its own set of parameters [20, 21]. The downside of this complexity is that model verification and parameterization are often hindered by the IVIV disconnect and parameter nonidentifiability in “middle out” approaches—too many parameters with uncertainty and too few observed data (concentration–time data) adversely affecting confidence in model predictions [22‐25].

Nonidentifiability issues in a PBPK model may be overcome through model simplification/lumping [26]. In an accompanying article in this issue [22], we proposed a simplified PBPK model in which drug solubility represents all solubility-driven processes (dissolution, precipitation, solubilization, etc.). For a poorly soluble BCS II drug that is not a substrate of cytochrome P450 3A (CYP3A) or intestinal transporters, the in vitro solubility employed in the model then becomes the sole parameter associated with uncertainty, which can be readily optimized against observed plasma drug concentrations. The aim of this work is to explore a novel approach to predict positive FE of BCS II/IV drugs using the proposed simplified PBPK model, recognizing solubility limited absorption as the sole driver of positive FE for BCS II/IV drugs. Through a comprehensive analysis, we test the hypothesis that a binary FE prediction using solubility-limited absorption (SLA) identified by a simplified PBPK model in a conservative setting can reliably predict positive FE for BCS II drugs that are not extensively metabolized or effluxed in the gut.

A novel method has been proposed capable of establishing a quantitative framework for the early prediction of FE by estimating a SR range from the interplay of solubility of the drug in biorelevant medium and the dose. This method aims to establish a conservative prediction, with zero uncertainty for both positive FE and non-FE conditions, which enables confident and timely decisions during drug development. Furthermore, it precisely establishes the scope of uncertainty about the impact of food on drug exposure (SR), allowing for simple but informative risk assessment. Although FE impacts the time to reach the maximum blood/plasma concentration (T_max), C_max, and AUC, compound selection was based on the C_max ratio. This is because, C_max is likely to be more sensitive to food-induced changes in exposure compared to AUC, while FE changes in T_max are likely to be confounded by gastric emptying. The characterization of in vitro solubility in physiological pH or FaSSIF and FeSSIF medium are often considered to adequately mimic in vivo conditions of orally administered drugs. However, these measured solubilities may still be conservative as in vitro settings may not adequately capture the in vivo sink conditions and transit kinetics [10]. Thus, not all BCS II/IV drugs classified based on measured solubility exhibit positive FE. According to a recent publication [18], only 36% of the 111 BCS II/IV approved drugs (25/68 BCS II, 6/25 BCS IV, 9/18 BCS II/IV) exhibit a positive FE, while 60% had no FE and 6% had negative FE. Out of the BCS II drugs exhibiting positive FE, only 15 had FE with significantly increased exposure (AUC ratio ≥ 2.0) [27]. All 15 were drugs with logP > 3 for which enhanced bile solubilization can play an important role in increasing in vivo solubility and therefore absorption, supporting that poor solubility is a plausible predictor of positive FE for BCS II drugs. The lack of FE for 60% of BCS II drugs can be explained by a possible misclassification of BCS I drugs as BCS II, based on conservative in vitro measures of solubility disconnected from the real in vivo solubility. In the previous research article by Owens et al. [27], the authors also showed that negative FEs are mostly seen in BCS III drugs (28% of the 28 drugs) and only 3 of these showed an AUC ratio of < 2.0-fold. Given the weak correlation of FE with BCS classes, the authors concluded that multiple physiological mechanisms impact FE [27]. However, since none of the 25 BCS I drugs in their study showed a positive FE (as expected from this class), it follows that high absorption resulting from high solubility and permeability does guarantee a lack of FE. This implies that mechanisms of FE not mediated by solubility and permeability (food–drug complexation, lower luminal drug concentration in fed state leading to increased susceptibility of substrates to intestinal efflux and inhibition of CYP3A and/or efflux transporters in gastrointestinal tract) are rare, even if theoretically possible. The absence of positive FE for BCS I drugs also suggests that it is very unlikely for measured solubility to over-predict in vivo solubility and misclassify a BCS II as BCS I.

Dose-adjusted solubility provides a drug-independent determinant of oral drug absorption in vivo. As dose-adjusted solubility is decreased from a hypothetically high value, the corresponding simulated oral drug absorption is constant at 100% and high, until a critical threshold, below which oral drug absorption starts to decrease. A drug with a dose-adjusted solubility below this critical threshold is said to have SLA (Supplementary Fig. S3). The in vivo solubility of such a drug may be enhanced by food via its influence on one or more factors (gastric pH and gastric emptying rate, as well as drug and bile salt concentrations). The information provided in Table 1 can aid in early clinical development, if the compound FaSSIF/D values are less than 3 × 10⁻⁵ 1/ml (below SR), it is likely to present FE. If FaSSIF/D is greater than 1 × 10⁻³ 1/ml (above SR), it is likely to exhibit no FE. If the FaSSIF/D value is within the SR area, then SLA determination with PBPK simulations can help to predict if the drug is likely to have FE or not, if the drug is not extensively metabolized in the gut. As can be appreciated, FaSSIF/D can indeed discriminate between drugs with or without FE, under the term that drugs are outside the SR area (Fig. 4). A conservative setting for FE prediction was ensured by choosing an upper and lower limit of conservative SR based on a determination of FaSSIF in our laboratory for 26 compounds. The FaSSIF values found in the literature are derived from different laboratories and unreported conditions of measurements. However, despite these limitations, the inclusion of the additional compounds with literature-reported FaSSIF data provided comparable SR limits to those derived from in-house data (Fig. 4).

As previously mentioned, this approach is not valid for compounds with high intestinal metabolism nor substrates of intestinal transporters; both felodipine and clopidogrel are high clearance drugs (see Table 2) with extensive CYP3A4-mediated intestinal metabolism. Consequently, the bioavailability of these drugs is limited by metabolism to a much larger extent than limited by solubility. Extensive gut metabolism is likely for very high clearance CYP3A4 or uridine diphosphate lucuronosyltransferase (UGT) substrates and can be identified in the PBPK model when the simulated C_max with FaSSIF solubility is greater than the observed C_max (since it is unlikely that FaSSIF solubility is greater than in vivo solubility). However, even when the C_max with FaSSIF solubility is lesser than the observed C_max, gut metabolism cannot be excluded for CYP3A4 and UGT substrates. The individual contributions of solubility and gut metabolism in limiting drug absorption are nonidentifiable since model optimization against observed plasma drug concentrations can resolve uncertainty only in one parameter. Thus, for drugs with gut metabolism, the resulting masking will provide conservative estimates of FE prediction, if it is possible to apply the proposed approach (C_max simulated with FaSSIF < C_max observed). Unfortunately, a reliable quantitative prediction of gut metabolism is not easy even for CYP3A substrates [28‐30]. However, since most high-clearance drugs are screened out during lead optimization, and low-clearance drugs are not likely to be impacted by gut metabolism, the proportion of drugs for which FE cannot be reliably predicted by the method proposed by our work is expected to be low. It is noteworthy that felodipine has no FE despite extensive metabolism by CYP3A4 in the gut, which seems to suggest that the standard high-fat food recommended for FE studies does not impact CYP3A4-driven gut metabolism. This is also consistent with the absence of FE for BCS I drugs [27], many of which are CYP3A4 substrates. However, certain other foods such as grapefruit juice have been shown to selectively inhibit intestinal metabolism [31].

Conventional PBPK models [13, 15‐17] that are used for FE predictions include multiple unverifiable processes such as dissolution, precipitation, and solubilization, thus introducing a large array of parameters in the models. Difficulty in verifying underlying mechanisms leads to a model with many assumptions that may fit the observed data but cannot reliably predict an untested scenario (e.g., FE). The parameterization of these models relies on in vitro data that are generally under-predictive and cannot be optimized against observed concentration–time profile due to nonidentifiability. Therefore, FE predictions with PBPK models in the traditional setting tend to be conservative and uncertain resulting in unnecessary clinical FE studies [32]. This implies that pilot FE studies would be conducted even for drugs that may not have SLA or FE in vivo (e.g., a drug such as mefenamic acid). We propose here a basic framework to assess FE prediction based on a simplified model, where SLA is the surrogate for FE. This allows the only parameter that impacts FE (solubility) to be optimized against observed data for drugs that are not extensively metabolized/effluxed in the gut, thereby resulting in a more reliable binary FE prediction. However, this is only a retrospective study. In the future, this approach can be applied for prospective predictions. A reliable binary prediction of FE (yes/no) in a conservative setting (no false negatives) using the method proposed in this work is more valuable for making timely decisions on the need for a pilot FE study and timing of a pivotal FE study compared with quantitative prediction by a PBPK in the traditional setting based on several assumptions and uncertain parameters.

The results from this work have demonstrated that dose-adjusted FaSSIF solubility can be used to discriminate drugs with positive FE from those with no FE outside SR. Within SR, drugs with SLA identified by PBPK are likely to have FE. Comparable SR limits, with or without the addition of drugs with literature-reported FaSSIF data to drugs with FaSSIF solubility measured in-house, despite interlaboratory differences in the FaSSIF values, show that the SR limits are not too sensitive to these differences and that the SR limits established in this work can be applied to drugs with FaSSIF solubility measured elsewhere. The selection of SR based on conservative, in-house FaSSIF measurements on 26 drugs, and identification of SLA by PBPK allows for reliable prediction of FE to enable decisions on the need for pilot FE study and timing of pivotal FE study. It is important to note that this approach cannot be applied to drugs with extensive gut metabolism or transporter-mediated elimination. The reliability of positive FE prediction using SLA was tested with six compounds within SR, for which PK-Sim models were available. The simplified PBPK model proposed here combines all nonidentifiable parameters into a single identifiable parameter for the purpose of FE prediction, although this may compromise quantitative prediction accuracy. Our work shows that the binary prediction accuracy, which is critical for decision-making in clinical development, is not compromised. Extension of this work in the future to cover all compounds within SR will serve to further enhance confidence in the use of SLA to identify drugs that are likely to exhibit positive FE.