Computer-Aided Biopharmaceutical Characterization: Comprehensive Notes on Gastrointestinal Absorption Simulation

Introduction to Gastrointestinal Absorption Simulation in Silico

Biopharmaceutical assessment is a critical component across various stages of drug discovery and development. In the early phases, pharmaceutical profiling facilitates the identification of "drug-like" molecules suitable for preclinical and clinical progression. In later stages, extended biopharmaceutical evaluation guides formulation strategies and predicts the impact of food on drug absorption. The growing emphasis on characterizing drugs and pharmaceutical products has catalyzed interest in in silico tools. These tools are designed to identify critical factors, such as drug physicochemical properties and dosage form factors, that influence in vivo performance. They enable the prediction of drug absorption based on selected input data sets. While in silico pharmacokinetic (PK) models can confirm different administration routes (as noted by Gonda and Gipps, 1990; Grass and Vee, 1993), the primary focus remains on the prediction of pharmacokinetics for orally administered drugs. Drug absorption from the gastrointestinal (GI) tract involves a complex interplay of physicochemical properties, physiological factors, and formulation-related factors. Modeling this complexity has evolved from simple approaches like the pH-partition hypothesis to sophisticated dynamic models such as the Compartmental Absorption and Transit (CAT) model. Modern dynamic models represent GI physiology by accounting for transit, dissolution, and absorption. Prominent models include the Advanced Dissolution, Absorption and Metabolism (ADAM) model, the Grass model, the GI-Transit-Absorption (GITA) model, and the Advanced CAT (ACAT) model. Several of these are integrated into commercial software such as GastroPlus™, SimCYP, PK-Sim®, Cloe® PK, and INTELLIPHARM® PKCR.

Theoretical Background: The Advanced Compartmental Absorption and Transit (ACAT) Model

The ACAT model, implemented in GastroPlus™, is an advanced semi-physiological absorption model based on the Biopharmaceutics Classification System (BCS) and established knowledge of GI physiology. It utilizes a system of coupled linear and nonlinear rate equations to simulate physiological effects on drug absorption during transit through successive GI compartments. The human GI tract model consists of nine compartments linked in series: the stomach, duodenum, two jejunum compartments, three ileum compartments, the caecum, and the ascending colon. Each compartment is further divided into sub-compartments containing unreleased drug, undissolved drug, dissolved drug, and drug entered into the enterocytes. The rate of change of dissolved drug concentration in each compartment depends on ten distinct processes: (I) transit into the compartment, (II) transit out, (III) release from the formulation, (IV) dissolution of particles, (V) precipitation, (VI) lumenal degradation, (VII) absorption into enterocytes, (VIII) exsorption from enterocytes back to the lumen, (IX) absorption into the portal vein via the paracellular pathway, and (X) exsorption from the portal vein via the paracellular pathway. Each process is associated with a specific rate constant. The transfer rate constant ($k_t$) is determined by the mean transit time. The dissolution rate constant ($k_d$) is calculated based on local pH, drug concentration, fluid percentage, and bile salt concentration. The absorption rate constant ($k_a$) is the product of the effective permeability ($P_{eff}$) and an absorption scale factor (ASF). The ASF corrects for changes in physiological conditions along the GI tract, such as surface area, pH, and the expression of transport or efflux proteins. Default ASF values are often estimated using a logD model, which assumes that as the ionized fraction of a compound increases, the effective permeability decreases.

Physiologically Based Pharmacokinetics (PBPK) and Disposition

Once a drug passes the basolateral membrane of enterocytes, it reaches the portal vein and liver, potentially undergoing first-pass metabolism before entering systemic circulation. The ACAT model connects to either a conventional PK compartment model or a physiologically based PK (PBPK) disposition model. PBPK models describe drug distribution in major tissues, treating them as either perfusion-limited or permeability-limited. Tissues are represented as single compartments linked by blood circulation. By integrating input parameters such as partition coefficients, metabolic rate constants, elimination rate constants, and protein binding, researchers gain mechanistic insight and improved prediction accuracy for human pharmacokinetics. While PBPK traditionally required a high volume of input data, advances in predicting liver metabolism, tissue distribution, and absorption from in vitro and in silico data have made these models increasingly attractive for pharmaceutical research.

Modeling Parameters and Simulation Strategy

GastroPlus™ ACAT modeling requires several categories of input parameters. Physiological parameters are provided as default population mean values for fasted and fed states, including transit times, pH, volumes, lengths, and radii of GI regions. Drug physicochemical properties include solubility, permeability, logP, $pK_a$, and diffusion coefficients. Pharmacokinetic parameters include clearance (CL), volume of distribution ($V_c$), and first-pass extraction percentages for the oral cavity, gut, or liver. Formulation characteristics involve particle size distribution, density, and release profiles for controlled-release (CR) formulations. Regional solubility is calculated using the Henderson–Hasselbalch relationship based on compartmental pH and known solubility at a single pH. Recent software versions also account for bile salt effects and include mean precipitation time to model poorly soluble weak bases moving from the acidic stomach to the small intestine. Effective permeability values refer specifically to human jejunal permeability. If measured values are unavailable, estimates from CaCo-2, PAMPA, or animal studies are used via a permeability converter that transforms inputs based on a training data set correlation model. Modeling generally follows a sequence of data collection, parameter optimization, and model validation. Verified models are used to understand how parameters affect PK profiles, identify target in vivo dissolution for in vitro-in vivo correlation (IVIVC), evaluate dosing regimens, predict food effects, and perform stochastic virtual trials.

Case Study: Nimesulide GI Simulation and Input Selection

A study on nimesulide oral absorption demonstrated how different assumptions regarding key factors influence model accuracy. Two independent models were constructed using the same in vivo data set but different presumptions. Model 1 assumed nimesulide was a substrate for intestinal influx transporters; consequently, ASFs were adjusted to match in vivo data, using an experimentally determined intrinsic solubility and an in silico predicted human jejunal permeability of $2.225 \times 10^{-4}\,cm/s$. Model 2 assumed nimesulide absorption was governed by a pH-surfactant induced increase in solubility in the GI milieu. In Model 2, ASFs remained at default values, while solubility and permeability ($2.002 \times 10^{-4}\,cm/s$) were optimized. Clinical data for a $100\,mg$ dose showed nimesulide $t_{max}$ varies between $1$ and $4\,h$. Model 1 predicted a $t_{max}$ of $3.15\,h$ (PE: $21.25\%$) and an AUC of $25.96\,μg\,h/mL$ (PE: $-0.70\%$). Model 2 predicted a $t_{max}$ of $3.40\,h$ (PE: $15.00\%$) and an AUC of $25.69\,μg\,h/mL$ (PE: $0.35\%$). Both models was considered accurate for $C_{max}$ and AUC, with prediction errors below $10\%$. However, the Model 2 assumption matched the Biopharmaceutics Drug Disposition Classification System (BDCCS) view that BCS class II drugs are rarely influx transporter substrates. Model 2 also suggested nimesulide dissolution was the rate-limiting factor for absorption, estimating approximately $70\%$ drug absorption from immediate-release (IR) tablets compared to the nearly $100\%$ estimated in Model 1.

Case Study: Gliclazide (GLK) Mechanistic Modeling

Gliclazide (GLK) is a BCS class II ampholyte with pH-dependent solubility. For a $80\,mg$ IR tablet dose, input parameters included a logP of $1.448$, $pK_a$ values of $2.9$, $5.8$, and $9.6$, and an experimental solubility of $0.025\,mg/mL$ at pH $4.37$. Initial simulations using default GastroPlus™ ASFs diverged from in vivo data. The Optimization module was used to adjust regional ASF values. For instance, the default duodenum ASF was $2.760$, while the optimized value was $1.289$. The resulting lower adjusted ASF values in the small intestine suggested the influence of efflux transporters like Mrp2 and Mrp3. Utilizing these adjusted parameters, the model predicted a fraction absorbed ($F_a$) of $99.94\%$, correlating well with the nearly $100\%$ bioavailability reported in literature. The predicted $C_{max}$ and AUC had prediction errors of less than $10\%$. Regional distribution analysis demonstrated that $69.9\%$ of the dose was absorbed in the proximal GI (duodenum and jejunum), with the remainder absorbed in distal regions.

Case Study: Carbamazepine (CBZ) Modeling and Regional Absorption

Carbamazepine (CBZ), a BCS class II compound, was modeled across four dosage forms: IR suspension ($200\,mg$), IR tablet ($400\,mg$), XR tablet ($400\,mg$), and XR capsule ($300\,mg$). Zhang et al. (2011) utilized two methods to obtain PK parameters and ASFs: deconvoluting PK data for the IR suspension under fasted conditions to find CL ($1.127\,L/h$) and $V_c$ ($63.06\,L$), and an optimization method fitting nine parameters simultaneously. Optimized ASFs were approximately $10$ times higher than default logD model values, indicating rapid small intestine absorption. A stomach transit time of $0.1\,h$ was used for the suspension and $0.25\,h$ for solid forms in the fasted state, whereas $1\,h$ was utilized for all forms in the fed state. Colon transit was set to $36\,h$. The model captured the absorption plateau observed in the IR tablet, with actual peak occupancy times ($POT_{20}$) ranging from $3.7$ to $41\,h$ compared to a predicted range of $2.9$ to $40\,h$. Regional absorption analysis showed CBZ IR formulations are absorbed mostly in the small intestine, while XR formulations are absorbed significantly in the caecum and colon. Kovacevic et al. (2009) also modeled CBZ using a single-compartment model ($V_c = 1.26\,L/kg$, $CL = 0.024\,L/h/kg$) and a particle radius of $25\,μm$ (vs. $75\,μm$ in the Zhang study), showing that absorption models depend heavily on the reference PK profile used for validation.

Parameter Sensitivity Analysis (PSA)

Parameter Sensitivity Analysis (PSA) is a feature in GastroPlus™ used to explore how changes in one or two parameters affect predicted PK profiles. In PSA, a parameter is varied gradually across a predetermined range while others are held constant. This tool is valuable when input values are rough estimates or when modeling highly variable drugs. For GLK, PSA showed that $F_a$ was insensitive to variations in drug particle density and effective particle radius, and even a $10$-fold decrease in solubility would maintain F_a > 85\%. However, larger particles and lower solubility increased the $t_{max}$. For CBZ, Kovacevic et al. found that complete absorption (F_a > 85\%) could be achieved with solubility as low as $0.05\,mg/mL$ (down from $0.12\,mg/mL$) and particle sizes up to $90\,μm$. Zhang et al. observed that CBZ absorption was dissolution rate-limited rather than solubility-limited when solubility was above $0.2\,mg/mL$. PSA is also a critical pre-formulation tool. Kuentz et al. (2006) used PSA on a poorly soluble drug, finding that bioavailability was insensitive to particle size reduction and solubility enhancement within the studied range ($160\,mg$ dose). This led to the selection of a simple capsule formulation over a complex delivery system, saving significant resources. Conversely, Dannenfelser et al. (2004) used PSA to identify that solubility and particle size significantly influenced absorption, leading to the development of a solid dispersion formulation.

Virtual Trials and Stochastic Simulation

Virtual Trials involve stochastic simulations on a specified number of subjects to account for inter-subject variability. Parameters are randomly sampled from distributions (means with coefficients of variation, CV%). Results include means, CV%, confidence intervals, and probability contours for $F_a$, bioavailability, $t_{max}$, $C_{max}$, and AUC. Tubic et al. (2006) applied this to talinolol (a P-gp substrate), incorporating variability in GI transit, pH, protein binding, and renal clearance. Using $12$ subjects, they found that all observed clinical data fell within the minimal and maximal individual simulation profiles, confirming the choice of log-normal distribution CV% values. Tsume and Amidon (2010) and Zhang et al. (2011) utilized Virtual Trials for bioequivalence (BE) studies. Zhang et al. simulated a $2 imes 2$ crossover study on $25$ subjects for a hypothetical XR CBZ tablet. They tested two virtual formulations: Test 1 (dissolution $f_2 = 67.4$) and Test 2 (dissolution $f_2 = 38.2$). Despite different in vitro dissolution, Test 2 was bioequivalent within the $80$ to $125\%$ criteria. The authors noted that virtual BE studies might under-predict variability if intra-subject variability is omitted.

Comparative Analysis of Fasted and Fed States

Food affects drug absorption through changes in gastric emptying, pH, fluid composition, hepatic blood flow, and bile salt concentrations. Lipophilic drugs often show increased exposure due to improved solubilization, while hydrophilic drugs may show negative food effects due to impeded permeation. GastroPlus™ provides default physiology changes: stomach pH rises from $1.3$ (fasted) to $4.9$ (fed); stomach transit time increases from $0.25\,h$ to $1.00\,h$; stomach volume increases from $50\,mL$ to $1000\,mL$; and hepatic blood flow increases from $1.5\,L/min$ to $2.0\,L/min$. Jones et al. (2006b) effectively predicted food effects for six compounds by inserting biorelevant solubility data for gastric, intestinal, and colonic fluids into ACAT compartments. Zhang et al. found that food lowered the $C_{max}$ and prolonged $t_{max}$ for CBZ IR suspension due to gastric emptying delays, but increased $C_{max}$ for IR tablets and XR capsules due to bile salt-enhanced dissolution. Parrott and Lave (2008) proposed a strategy where absorption models are first refined in preclinical species (e.g., beagle dogs). For BCS class I drugs like theophylline, food effects were easily simulated using default models. For BCS class II drugs like aprepitant, changes to diffusion coefficients and regional solubility were required, demonstrating that modeling challenging drugs often requires preclinical animal cross-verification.

In Vitro–In Vivo Correlation (IVIVC) and Deconvolution

Two approaches link in vitro and in vivo data using the ACAT model: convolution (predicting plasma profiles from in vitro inputs) and deconvolution (estimating the in vivo dissolution profile). For GLK, Five virtual in vitro profiles were tested: incomplete dissolution (profile a), and >85\% dissolution in $60$, $45$, $30$, and $15$ minutes (profiles b through e). Profiles b, c, d, and e exhibited a high level A IVIVC using both convolution and deconvolution approaches (rescaling time by $12$-fold linear factor for profile e), suggesting a biorelevant dissolution specification of >85\% in $60$ minutes. For CBZ, Kovacevic et al. tested various media ($0.1$, $0.25$, $0.5$, and $1\%$ sodium lauryl sulfate (SLS)) for IR and CR tablets. They found that $1\%$ SLS served as a "bioperformance" medium for both, with regression analysis showing high Level A IVIVC. Zhang et al. reviewed FDA-submitted CBZ data and found that in vitro dissolution in water was slower than in vivo fed dissolution but faster than fasted dissolution. Etoricoxib studies by Okumu et al. compared SGF, USP-SIF, and FaSSIF ($500\,mL$ and $900\,mL$) media. The simulated profiles from SGF (PE AUC: $-6.77\%$, PE $C_{max}$: $11.59\%$) and $900\,mL$ FaSSIF correlated better with in vivo data than $500\,mL$ FaSSIF (PE $C_{max}$: $28.04\%$).

Biowaiver Considerations

A biowaiver allows the substitution of in vivo BE studies with in vitro data to reduce costs and ethical burdens. Traditionally limited to BCS class I (highly soluble, highly permeable), recent criteria from the EMA and WHO have expanded eligibility to some BCS class III (if very rapidly dissolving) and BCS class II drugs (if dose-to-solubility ratio at pH $6.8$ is $≤ 250\,mL$ and absorption is $≥ 85\%$). In the GLK study, simulations showed that dissolution rates between $15$ and $60$ minutes resulted in similar PK profiles, suggesting biowaiver extension is rational for this BCS II drug. For etoricoxib, fast and complete absorption similar to an oral solution suggested biowaiver suitability. For CBZ, although insensitive to input kinetics, its narrow therapeutic index remains a regulatory barrier. Tubic-Grozdanis et al. identified ibuprofen, ketoprofen, diclofenac, piroxicam, and terbinafine as BCS II biowaiver candidates, while mefenamic acid and miconazole were excluded due to dissolution and solubility-limited absorption. Tsume et al. showed for BCS class III drugs (cimetidine, atenolol, amoxicillin) that permeability, not dissolution, was the rate-limiting step, as BE was maintained for release rates up to $T_{85\%} = 45$ or $60\,min$. Crison et al. (2012) justified a biowaiver for metformin hydrochloride ($500\,mg$ dose), demonstrating that release rates between $5\,min$ and $2\,h$ had no statistically significant effect on $C_{max}$ and AUC, regardless of $f_2$ test results, advocating for mechanistic modeling over simple statistical profile comparisons.

Conclusions

Computational modeling of GI absorption provides a mechanistic framework to interpret the interplay between physicochemical drug properties, formulation parameters, and human physiology. Tools like PSA allow for the identification of critical absorption barriers and help define better drug delivery strategies. Virtual Trials and PBPK modeling enhance the reliability of predictions by accounting for individual variability. Despite the complexity and requirement for high-quality data, validated in silico models facilitate Quality by Design (QbD) in pharmaceutical development. While some industry hesitate due to the lack of confidence in predictions or missing empirical data, the continued collection of biopharmaceutical data and the success of published examples promise a wider acceptance of in silico techniques as indispensable, cost-effective tools for assessing drug bioperformance.

Questions & Discussion

Question: Why did Model 2 for nimesulide provide a more realistic interpretation than Model 1? Response: While both models accurately predicted average plasma profiles (prediction error < 10\%), Model 2 relied on the pH-surfactant induced increase in solubility, matching the BDCCS classification for BCS II drugs which typically aren't substrates for influx transporters. It also revealed incomplete drug absorption from IR tablets ($∼70\%$), suggesting dissolution is the limiting factor.

Question: How does GastroPlus™ model the effect of bile salts on solubility? Response: Recent versions of the software allow for the direct calculation of regional solubility changes based on compartmental bile salt concentrations and pH, reflecting the in vivo environment more accurately than simple aqueous models.

Question: What are the default GastroPlus™ values for human stomach transit under fasted vs. fed conditions? Response: According to the provided table, the default stomach transit time is $0.25\,h$ for the fasted state and $1.00\,h$ for the fed state.

Question: What was the significance of the virtual trial for talinolol? Response: The virtual trial incorporated variability in physiological parameters (pH, transit, radii) and PK parameters (protein binding, renal CL). The simulation encompassed all observed clinical data, proving that presumed log-normal distributions can effectively model real-world variability.