**Segmental-Dependent Solubility and Permeability as Key Factors Guiding Controlled Release Drug Product Development**

**Milica Markovic 1,**† **, Moran Zur 1,**† **, Noa Fine-Shamir 1 , Ester Haimov 1 , Isabel González-Álvarez <sup>2</sup> and Arik Dahan 1, \***


Received: 23 January 2020; Accepted: 20 March 2020; Published: 24 March 2020

**Abstract:** The main factors influencing the absorption of orally administered drugs are solubility and permeability, which are location-dependent and may vary along the gastrointestinal tract (GIT). The purpose of this work was to investigate segmental-dependent intestinal absorption and its role in controlled-release (CR) drug product development. The solubility/dissolution and permeability of carvedilol (vs. metoprolol) were thoroughly studied, in vitro/in vivo (Octanol-buffer distribution coefficients (Log D), parallel artificial membrane permeability assay (PAMPA), rat intestinal perfusion), focusing on location-dependent effects. Carvedilol exhibits changing solubility in different conditions throughout the GIT, attributable to its zwitterionic nature. A biorelevant pH-dilution dissolution study for carvedilol immediate release (IR) vs. CR scenario elucidates that while the IR dose (25 mg) may dissolve in the GIT luminal conditions, higher doses used in CR products would precipitate if administered at once, highlighting the advantage of CR from the solubility/dissolution point of view. Likewise, segmental-dependent permeability was evident, with higher permeability of carvedilol vs. the low/high Peff marker metoprolol throughout the GIT, confirming it as a biopharmaceutical classification system (BCS) class II drug. Theoretical analysis of relevant physicochemical properties confirmed these results as well. A CR product may shift the carvedilol's solubility behavior from class II to I since only a small dose portion needs to be solubilized at a given time point. The permeability of carvedilol surpasses the threshold of metoprolol jejunal permeability throughout the entire GIT, including the colon, establishing it as a suitable candidate for CR product development. Altogether, this work may serve as an analysis model in the decision process of CR formulation development and may increase our biopharmaceutical understanding of a successful CR drug product.

**Keywords:** controlled release drug product; biopharmaceutics classification system; drug solubility; drug permeability; location-dependent absorption

#### **1. Introduction**

Oral drug absorption depends on various parameters: physicochemical (e.g., ionization, pKa, solubility, physicochemical stability, lipophilic nature, polar surface area (PSA), molecular weight,), physiological (e.g., gastrointestinal pH, surface area available for absorption, transit time, expression of certain transporters, enzymes), and parameters associated with the dosage form [1–3]. However, keeping this complexity in mind, it was determined that the drug permeability and solubility/dissolution

in the gastrointestinal aqueous milieu are the two essential variables that guide absorption in the gastrointestinal tract (GIT) [4].

These two key factors, the solubility and the permeability, are location-dependent and can vary along the GIT. Change in pH or presence of bile salts can modify drug solubility/dissolution in a given intestinal segment; for a drug to be considered as a high solubility compound as per the biopharmaceutical classification system (BCS), it needs to be dissolved in an aqueous media (250 mL or less) with the different pH values relevant to the GIT lumen (1.0–6.8) [5–7]. Likewise, intestinal permeability is also location-dependent, and pertains in each region of the GIT [1,3,8,9]. Therefore, in different scenarios, e.g., drug discovery, drug and formulation development, and regulatory considerations, assigning the BSC class membership founded only on physicochemical drug features may lead to the incorrect decision [2,10–13]. Thus, regional-dependent permeability factors also need to be considered, for instance, expression of membrane transporters (influx/efflux) along the intestinal tract [11,14–16], luminal pH that influences the changes in the drug ionization [1,3,8,13], local water absorption [10], and others.

Segmental-dependent biopharmaceutical considerations are particularly important for controlled-release (CR) drug products; the drug is continuously released throughout the entire GI; therefore, it is not sufficient for a drug moiety to have suitable solubility/permeability in only one particular intestinal segment [17–19].

Carvedilol is a third-generation β-blocker, and is commonly used for treating hypertension, heart failure, and left ventricular dysfunction (LVD) [20,21]. The pharmacokinetics and pharmacodynamics of carvedilol from controlled release (CR) and immediate release (IR) products were compared in two clinical studies [22,23]. The data from these studies demonstrated that once-daily CR carvedilol is clinically correspondent to the IR carvedilol drug product administered two times a day, in patients with heart failure and asymptomatic post-myocardial infarction [23]. In addition, carvedilol CR maintains steady β1-adrenergic blockade with a dose administered once every 24 h [22]. Metoprolol is a passively transported drug which is not affected by the P-glycoprotein (P-gp) efflux transport [24], while carvedilol is a substrate of P-gp [25]. Carvedilol inhibits the activity of P-glycoprotein (P-gp) transporter [26,27]. It also undergoes extensive stereoselective first-pass metabolism; the main cytochrome P450 enzymes responsible for the metabolism of both R(+) and S(−)-carvedilol are CYP2D6 and CYP2C9, with some of the resulting metabolites having pharmacological activity. Despite its extensive first-pass metabolism, marketed carvedilol CR capsules have a bioavailability of 85% relative to IR tablets, with good clinical efficacy [23]. Hence, carvedilol was shown as a successful candidate for development as a controlled-release drug product, despite the fact that the pH variations along the GIT may significantly alter both the solubility and the permeability of this ionizable (basic) drug [28,29]. This raises the question of carvedilol's location-dependent intestinal solubility and permeability and its successful use as a CR product.

This work aimed to study the segmental-dependent biopharmaceutical consideration of carvedilol as a model basic drug, analyzed in view of CR scenario, allowing to pinpoint the rational for a successful CR drug product. Carvedilol solubility/dissolution and permeability were systematically investigated, through in vitro/in vivo (Octanol-buffer distribution coefficients (Log D), parallel artificial membrane permeability assay (PAMPA), rat intestinal perfusion) techniques, focusing on location-dependent variations, as well as theoretical physicochemical properties analysis of the drug. As a Food and Drug Administration (FDA) recommended standard for the low-high permeability class boundary, we used metoprolol, which also served as an accompanying model compound, since it is also marketed as a CR drug product. This study offers a deeper understanding of the factors that could influence segmental-dependent permeability and solubility in a controlled-release setting, and their contribution to a successful controlled-release drug product.

#### **2. Materials and Methods**

#### *2.1. Materials*

Carvedilol, metoprolol, sodium chloride, potassium phosphate monobasic, and sodium phosphate dibasic, hexadecane, octanol, and trifluoroacetic acid (TFA) were purchased from Sigma Chemical Co. (St. Louis, MO, USA). Water and acetonitrile (Merck KGaA, Darmstadt, Germany) were ultra-performance liquid chromatography (UPLC) purity grade. All other substances were of analytical reagent grade.

#### *2.2. Solubility*

The pH-dependent solubility, as well as carvedilol solubility BCS classification, was evaluated by the shake-flask method, as previously reported [8,30]. Briefly, the equilibrium solubility of carvedilol was studied at 37 ◦C, at pH 7.5 with phosphate buffer (potassium phosphate monobasic and sodium phosphate dibasic), pH 4.5 acetate buffer (sodium acetate and acetic acid), and pH 1.0 maleate buffer (maleic acid). Five hundred microliters of buffer was added to glass vials, and excess carvedilol quantities were added to buffer-containing glass vials, until the solution was no longer clear. Equilibrium was verified by comparison of 48- and 72-h samples. The pH of each solution was measured following the drug's addition to the buffer solution. The vial caps were firmly sealed, and the vials were placed in a shaking incubator (100 rpm, 37 ◦C). Before the drug concentration was analyzed, the vials were centrifuged at 10,000 rpm (10,621 rcf) for 10 min, and the supernatant was removed, followed by drug quantification with UPLC. For dose number (D0) calculations, the highest dose of carvedilol immediate-release (IR) oral drug product was taken to be 25 mg [22].

#### *2.3. Octanol-Bu*ff*er Distribution Coe*ffi*cients*

Octanol-buffer distribution coefficients (Log D) for carvedilol and metoprolol were determined at pH 6.5, 7.0, and 7.5 using the shake-flask method [12,30]. This pH range represents the physiological pH relevant for the intestinal tract (naturally, permeability from the stomach is considered not significant). Carvedilol and metoprolol solutions were prepared in a phosphate buffer saturated with octanol (pH 6.5, 7.0, and 7.5.), and consequently equilibrated at 37 ◦C, 48 h with an equal volume of buffer saturated with octanol of corresponding pH. The aqueous and octanol phase were parted by centrifugation, and the concentration of the drug in the aqueous phase was quantified by UPLC; the drug in the octanol phase was determined by mass balance.

#### *2.4. Biorelevant pH-Dilution Dissolution Studies*

An in vitro biorelevant pH-dilution dissolution study was performed (n = 5 each) as we have previously published [31–33], to simulate drug dose dissolution while traveling along the GIT, in two scenarios: carvedilol concentrations of 100 µg/mL vs. 320 µg/mL, simulating the highest IR dose (25 mg; COREGTM) and CR dose (80 mg; COREG CRTM) on the market, taken with 250 mL of water. An aqueous suspension of the drug dose was first diluted into HCl 0.1M to obtain a pH of 1.2 (dilution factor 1:0.66) and agitated for 15 min (100 rpm at 37 ◦C), to mimic the stomach compartment, as we have previously reported. Then, samples were further diluted with fasted state simulated intestinal fluid (FaSSIF) (Biorelevant.Com Ltd., London, UK) with a dilution factor (1:1) for 30 min, followed by a dilution factor of 1:1.5, agitated for 30 min, and consequently 2 other dilutions of 1:1 with agitation time of 1 h each, to closely mimic the conditions throughout the small intestinal travel; the complete time of the study was 3 hours and 15 min (with samples taken at time points 0, 15, 45, 75, 105, 135, 195 min). During the course of the study, samples were centrifuged, filtrated, and the drug concentration was instantly quantified by UPLC. The solubilized drug amount, quantified by UPLC, was compared to the total amount of drug, which was calculated using the initial drug dose and consequent dilutions. This comparison enabled evaluation of the fraction of dose dissolved vs. precipitated for the IR vs. CR simulated experiments. The pH gradient throughout the experiment was designed to mimic the physiological conditions along the GIT, with a final pH of 7.6.

#### *2.5. Parallel Artificial Membrane Permeability Assay Studies*

In vitro permeability studies through an artificial membrane were carried out in the hexadecane-based parallel artificial membrane permeability assay (PAMPA) using Millipore (Danvers, MA) 96-well MultiScreen-Permeability filter plates with 0.3 cm<sup>2</sup> polycarbonate filter support (0.45 µm). The filter supports in every well were impregnated with 15 µL of a 5% solution (*v*/*v*) of hexadecane in hexanes and were then permitted to dry for 1 h. This time frame allowed the hexanes to be entirely evaporated, producing a consistent hexadecane layer. The permeability studies using hexadecane layer were carried out according to the standard protocol, with minor modifications [13]. Briefly, both carvedilol and metoprolol solutions (n = 4) were prepared in phosphate buffer solution (pH 6.5, 7.0, and 7.5) with comparable ionic strength and osmolality (290 mOsm/L). PAMPA sandwich plates were composed of donor wells containing various drug solutions (200 µL), and the receiver wells containing blank buffers (300 µL). The plate was incubated at room temperature, and samples were taken from the receiver plates every hour for a total of 4 h. Apparent permeability coefficient (Papp) was calculated from the linear plot of drug collected in the acceptor side vs. time with the following equation:

$$\mathbf{P\_{app}} = \frac{\mathbf{dQ/dT}}{\mathbf{A} \times \mathbf{C\_0}}$$

where dQ/dt is the appearance rate in the steady-state of carvedilol/metoprolol from the receiver side, C<sup>0</sup> is the starting drug concentration in the donor side (0.02 mM for carvedilol, and 0.1 mM for metoprolol), and A is the membrane surface area (0.048 cm<sup>2</sup> ). Linear regression was used to acquire the steady-state appearance rate of the drug on the receiver side.

#### *2.6. Rat Single-Pass Intestinal Perfusion (SPIP)*

The rat effective permeability coefficient (Peff) of carvedilol and metoprolol in different intestinal regions was evaluated using the single-pass intestinal perfusion (SPIP) model. This experimental model was designed and validated to account for the complex physiological background of drug absorption along the GIT: the living animal, intact and viable GIT including tissue composition, membrane morphology, expression/distribution of functional transporters/enzymes, and the composition of the luminal milieu of the different segments, are all part of the high biorelevance of this model [34–36]. All animal experiments were performed according to the protocols accepted by the Ben-Gurion University of the Negev Animal Use and Care Committee (Protocol IL-07-01-2015). The animals (male Wistar rats weighing 230–260 g, Harlan, Israel) were housed and handled in agreement with the Ben-Gurion University of the Negev Unit for Laboratory Animal Medicine Guidelines.

The experimental procedure used for the in situ experiments in rats was previously described [3,12,13,30]. Prior to the experiment, the rats were fasted overnight. Namely, rats were anesthetized and positioned on a 37 ◦C surface (Harvard Apparatus Inc., Holliston, MA, USA), and a 3 cm midline abdominal incision was performed. Considering the complexity behind each of the intestinal segments, permeability was simultaneously measured through 3 separate intestinal regions (length of 10 cm each); a proximal segment of the jejunum at pH 6.5 (beginning at 2 cm under the ligament of the Treitz), a distal segment of the ileum at pH 7.5 (finishing 2 cm above the cecum), and the colonic segment at pH 6.5 (approximately 6 cm); the pH values through each region corresponded to the physiological pH of that region [1,3]. Each intestinal segment was cannulated at both sides and was rinsed with the relevant blank perfusion buffer. Phosphate buffers containing carvedilol and metoprolol were prepared at pH 6.5 and 7.5, while maintaining similar ionic strength and osmolality (290 mOsm/L) in all buffers. All solutions were incubated in a water bath at 37 ◦C. The steady-state conditions were established by perfusing the drug-containing buffer solution (0.02 mM) for 1 h, and an additional hour of perfusion followed, with sample collection every 10 min. The pH value was determined in

the outlet samples to ensure the there was no pH variation during the course of perfusion. At the end of the perfusion study, the drug concentration in the outlet samples was determined by UPLC, and the length of the intestinal segment used for perfusion was measured for further permeability calculations. The effective permeability (Peff; cm/s) through the gut wall was calculated through to the following equation:

$$\mathcal{P}\_{\rm eff} = \frac{-\text{Qln}\left(\mathcal{C}'\_{\rm out}/\mathcal{C}'\_{\rm in}\right)}{2\pi \text{RL}}$$

Q being the perfusion buffer flow rate (0.2 mL/min), C ′ out/C′ in is the ratio of the outlet and the inlet drug concentration that has been adjusted for water transport by the gravimetric method [37–39], R is the radius of the intestinal segment (conventionally used as 0.2 cm), and L is the length of the perfused intestinal segment.

#### *2.7. Physicochemical Analysis*

The theoretical fraction extracted into octanol (fe) was calculated using the following equation [40,41]:

$$\mathbf{f\_e} = \frac{\mathbf{f\_u}\mathbf{P}}{1 + \mathbf{f\_u}\mathbf{P}}$$

where P stands for the octanol–water distribution coefficient of the unionized drug form and f<sup>u</sup> is the drug fraction unionized at a certain pH. The f<sup>u</sup> vs. pH was plotted according to the Henderson–Hasselbalch equation, using the following literature pKa values: 9.7 for metoprolol [42] and 7.8 for carvedilol [28].

#### *2.8. Ultra-Performance Liquid Chromatography*

An ultra-performance liquid chromatography (UPLC) instrument Waters (Milford, MA, USA) Acquity UPLC H-Class was equipped with a photodiode array detector and Empower software. The instantaneous determination of carvedilol and metoprolol was accomplished using a Waters Acquity UPLC XTerra C<sup>18</sup> 3.5-µm 4.6 × 250 mm column. The gradient mobile phase consisted of 90:10 going to 30:70 (v/v) water:acetonitrile (containing 0.1% TFA) at a flow rate of 0.5 mL/min during 4 min. The wavelength of detection and retention times for carvedilol and metoprolol were 230 and 275 nm and 2.5, 3.1 min, respectively. UPLC injection volumes for all analyses were in a range from 2 to 50 µL. The limit of quantitation was termed as the lowest drug concentration that could be measured with an accuracy and precision of <20%, as per US Food and Drug Administration Guidelines. Precision was stated as the intra- and inter-day relative standard deviation (RSD). Intra-day accuracy and precision were determined by analyzing six replicates of control samples on the same day (samples of known concentration), while the inter-day accuracy and precision were evaluated by measuring six replicates of control samples on three different days. The carvedilol limit of quantification was 5 ng/mL, and for metoprolol 25 ng/mL, and the inter- and intra-day coefficients of variations were <1.0% and 0.5%, respectively.

#### *2.9. Statistical Analysis*

Log D studies and PAMPA assays were replicated with n = 6 and n = 4, respectively. Animal studies were replicated with n = 6. All values are stated as means ± standard deviation (SD). Statistically significant differences between the experimental groups were evaluated by the nonparametric Kruskal–Wallis test for multiple comparisons, and the two-tailed nonparametric Mann–Whitney *U* test for two-group comparison. A *p* < 0.05 was considered statistically significant.

#### **3. Results**

#### *3.1. Solubility*

The solubility data for carvedilol in the three pH values (1.0, 4.0, and 7.5) at 37 ◦C are presented in Table 1. The solubility data presents a complex picture: from pH 1.0 to 4.0, the solubility was rising, and again decreased towards pH 7.5, where the solubility was very low. The dose number was calculated using the subsequent equation: D<sup>0</sup> = M/V0/Cs, where M is the highest single-unit dose strength of carvedilol (25 mg) [22], V<sup>0</sup> is the initial volume of water (250 mL), and C<sup>s</sup> is the solubility at each pH; drug molecules with D<sup>0</sup> ≤ 1 are considered highly soluble. At a pH of 1.0 and 7.5, the dose number for carvedilol was higher than 1, indicating low BCS solubility class membership. The chemical structure of carvedilol is presented in Table 2. ≤ ≤

**Table 1.** Carvedilol solubility values (mg/mL) in the tree pH values 1.0, 4.0, and 7.5, at 37 ◦C, and the corresponding dose number (D<sup>0</sup> ) for a 25 mg dose. Data presented as mean ± SD; n = 6.



**Table 2.** Carvedilol and metoprolol molecular structures and relevant physicochemical parameters.

#### *3.2. Biorelevant pH-Dilution Dissolution Studies*

We have studied the ability of the two highest marketed dosages for both IR (25 mg) and CR (80 mg) carvedilol drug products to accomplish and maintain complete dissolution of the carvedilol dose in the dynamic GIT environment using the pH-dilution method we have previously developed [31]. The dissolution results are presented in Figure 1, where it can be observed that a significant difference between the dissolution behavior of the 25 mg and 80 mg drug product was detected. The results indicate what may happen if these doses were to be orally administered at once; while the 80 mg dose would quickly precipitate, the 25 mg dose was completely dissolved (with ~15 min delay) and maintained dissolved throughout the GIT travel.

**Figure 1.** Dissolution of the highest carvedilol dose for IR and CR drug products on the market (25 mg and 80 mg, respectively) in the dynamic GIT environment, using the pH-dilution dissolution method. Values are presented as means ± SD; \*\*\* *p* < 0.001; n = 5. The pH at each time point for IR: 1.4 at 15 min; 1.9 at 45 min; 5.8 at 75 min; 6.9 at 105 min; 7.2 at 135 min; 7.3 at 195 min; and for CR drug product: 1.6 at 15 min; 2.9 at 45 min; 7.0 at 75 min; 7.4 at 105 min; 7.6 at 195 min.

#### *3.3. Log D*

The octanol–water distribution coefficients (Log D) for carvedilol and metoprolol were measured at the three pH values of 6.5, 7.0, and 7.5, representative of the environment of the small intestine (Figure 2). It can be seen that both carvedilol and metoprolol have evident pH-dependent upward Log D in the investigated pH range (6.5–7.5), however, while the Log D of metoprolol ranged from 0.8 (pH 6.5) to −0.2 (pH 7.5), carvedilol Log D was positive and ranged from 2.7 (pH 6.5) to 3.7 (pH 7.5). −

**Figure 2.** The octanol-buffer distribution coefficients for carvedilol vs. metoprolol at pH values of 6.5, 7.0, and 7.5. Values are presented as means ± SD; n = 6.

#### *3.4. Physicochemical Analysis*

− The theoretical fraction unionized (fu) and fraction extracted into octanol (fe) as a function of pH for carvedilol vs. metoprolol are presented in Figure 3. The f<sup>u</sup> of the basic drugs carvedilol and metoprolol was negligible at low pH, and rose as the pH increased, producing a standard sigmoidal shape. It can be seen that the f<sup>e</sup> vs. pH plot of both drugs shows a similar pattern, but with a shift to the left at the lower pH values. The shift degree equals to Log (P − 1) at the midpoint of the f<sup>e</sup> and f<sup>u</sup> sigmoidal curves [40]. Experimental octanol-buffer distribution of the drugs at the three pH values of 6.5, 7.0, and 7.5 are also presented in Figure 3 and are in excellent correlation with the theoretical plots.

−

**Figure 3.** The theoretical fraction unionized (fu) and fraction extracted into octanol (fe) plots are presented as a function of pH for carvedilol and metoprolol. Log D values for both drugs are presented as circles; n = 6.

#### *3.5. PAMPA Assay*

The transported amounts vs. time in the PAMPA experiment for carvedilol and metoprolol are presented in Figure 4, with their matching Papp values. Compatibly to the log D results, the same pH-dependent upward permeability trend was found for both drugs; carvedilol showed considerably higher log D than metoprolol in the studied pH range, and the PAMPA permeability values confirmed this trend, as can be seen in Figure 4.

**Figure 4.** The hexadecane-based parallel artificial membrane permeability assay (PAMPA) permeability studies for carvedilol vs. metoprolol in the different pH conditions along the small intestine: amounts transported (mmol) as a function of time (left panel), and the corresponding Papp values (right panel; cm/s). Mean ± SD; n = 4.

#### *3.6. Rat Intestinal Perfusion Studies*

The values of carvedilol vs. metoprolol effective permeability coefficient (Peff) determined using the rat SPIP model, through the three intestinal segments: the proximal jejunum (pH 6.5), the distal ileum (pH 7.5), and the colon (pH 6.5), are presented in Figure 5. It can be observed that all of the permeability studies revealed a similar trend: higher pH led to higher permeability values, and as a result, the permeability of carvedilol and metoprolol in the ileum was significantly higher than in the jejunum. Furthermore, at any given intestinal segment/pH, the permeability of carvedilol was higher than that of metoprolol. In addition, when looking at the colon, the permeability value of carvedilol was higher than that of metoprolol in the jejunum (marked as a dashed line in Figure 5).

**Figure 5.** Effective permeability coefficient (Peff; cm/s) obtained for carvedilol vs. metoprolol in three rat intestinal segments, the upper jejunum (pH 6.5), terminal ileum (pH 7.5), and the colon (pH 6.5). The black dashed line represents the permeability of metoprolol in the jejunum (pH 6.5), which is the low/high Peff class boundary standard. Data are presented as means ± SD; \*\*\* *p* < 0.001 between jejunum and ileum for both carvedilol and metoprolol; n = 6.

#### **4. Discussion**

The variable physiological conditions throughout the GIT can greatly influence the rate and degree of oral drug absorption [10]. It was previously shown that there is a high level of correlation between the drug jejunal permeability and the fraction of dose absorbed from an IR drug product [11,47,48]. Conversely, for CR formulations, to obtain an optimal dissolution, intestinal permeability, and hence, acceptable bioavailability, a larger part of the GIT has to be accounted for in comparison to IR drug product, highlighting the crucial importance of regional variation among absorption factors. Both metoprolol and carvedilol are marketed as controlled-release products (metoprolol extended-release tablets of 25, 50, 100, or 200 mg; and carvedilol CR of 10, 20, 40, or 80 mg) which allows us to use them as model drugs in predicting important parameters that may dictate the development of a successful CR product [49,50].

Carvedilol is an alkaline drug that exhibits poor solubility in different conditions throughout the GIT [29]. However, the solubility studies revealed a high solubility at pH 4.0 (Table 1). In different studies, using simulated/aspirated media, it was shown that depending upon the experimental technique, some discrepancies are noticed [51,52], however, an apparent rise in solubility in the simulated intestinal fluid in the fed state (FeSSIF; pH = 5) is evident. This could be explained by the carvedilol chemical structure, where the aliphatic -NH group is more basic than the carbazole -NH group, which could lead to protonation, creating a soluble salt with the anionic form of the buffer, causing an increase in solubility [28,53]; in acidic media, the aliphatic -NH is ionized forming a cationic center, while in basic media, the carbazole -NH is ionized forming a anionic center. This zwitterionic nature is responsible for the unique solubility pattern presented in Table 1. At any rate, the low solubility values of carvedilol at acidic and neutral environment indicate a low-solubility BCS classification, as in the case of IR carvedilol product, a maximal single unit dose is 25 mg [23], leading to a dose number higher than 1 in different GIT locations. Alongside the high permeability values throughout the GIT, it was confirmed that carvedilol is indeed a BCS class II compound.

Under such solubility limitations, developing carvedilol as a CR drug product may, in a way, help to avoid solubility limitations. By definition, a CR product releases the drug gradually from the formulation while traveling along the GIT, and so, in place of requiring the solubilization the entire

dose at once, only a small fraction of the dose needs to be solubilized at a given time point, which may allow overcoming solubility limitations. On the other hand, at each point throughout the intestinal tract, the aqueous volume is lower than the 250 mL initially taken with the drug dose. In particular, it was reported that when simulating the fate of low-solubility drugs after oral administration, the small intestinal water volume that allowed the best fits with in vivo data was about 130 mL (ranging 10–150 mL in the fasted state), and 10 mL in the colon (with estimations as large as 125 mL in the fasted state) [54]. Nevertheless, if there is sufficient fluid in the lumen at each point, it may be possible to obtain adequate drug solubility throughout the digestive system.

The pH-dilution dissolution experimental method mimics the passage and fate of the drug dose through the different GIT segments over time and hence, allows revealing whether the drug can be solubilized, and remain dissolved, while in the GIT. Our results elucidate that generally the highest IR dose (25 mg) has the ability to be dissolved, and remain such throughout the GIT travel, which explains why this formulation is an efficient marketed drug product. On the other hand, if the CR dose of carvedilol (80 mg) were to be administered at once as a simple IR formulation, rapid precipitation would take place (Figure 1), preventing the success of such drug products. Formulating this carvedilol dose as a CR product allows overcoming this solubility/dissolution limitation by distributing small portions of the dose at each stage throughout the GIT. This model analysis illustrates the application of biopharmaceutical aspects in the decision process of successful CR formulation development.

A literature search showed that carvedilol in vitro absorption was studied in a model called the Boehringer–Mannheim ring model using porcine intestine [55]. According to this study, the main route of absorption for carvedilol was transcellular, and the optimal absorption was obtained in the neutral pH of 6.8. In situ intestinal perfusion with mesenteric blood sampling in rats using human intestinal fluids and biorelevant media was used to study the food effect on the intestinal solubility and permeability of carvedilol [52]; however, the use of biorelevant media that contain high lecithin concentration would also affect the solubility aspect of carvedilol in comparison to our SPIP study. Therefore, we used the buffers for the perfusion study instead. This study also did not account for the segmental-dependency of the solubility/permeability of carvedilol.

Prior to evaluating intestinal permeability (Peff) results, the threshold for the low/high permeability class membership must be set, since it reflects the penetration degree that allows complete absorption. For this purpose, metoprolol is a widely used and accepted standard compound [56,57]. Metoprolol exhibits significant segmental-dependent intestinal permeability with increasing Peff towards the distal parts of the small intestine. Therefore, the question is raised, which permeability should be taken as the class boundary: jejunal (~5 <sup>×</sup> <sup>10</sup>−<sup>5</sup> cm/s) or the much higher ileal value (~1.2 <sup>×</sup> <sup>10</sup>−<sup>4</sup> cm/s), presented in Figure 5. Absorption data obtained from humans revealed that 80% of metoprolol dose from IR product occurs already in the upper 50 cm of the small intestine (duodenum and proximal jejunum), leaving no drug for absorption in the ileum [58]. This was later shown in rats as well [59]. Therefore, ileal permeability values of metoprolol are not physiologically relevant for an IR drug product; it can be claimed that from an IR metoprolol product, no drug arrives into the ileum, as the entire dose gets absorbed much earlier. Hence, the jejunum permeability of metoprolol allows its complete absorption, and this value should be taken as the low/high threshold for permeability classification. Similarly, carvedilol demonstrated segmental-dependent permeability that matched the trend of metoprolol (the permeability in the ileum was significantly higher than in the jejunum, as demonstrated in Figure 5). However, the Peff values of carvedilol were significantly higher than that of metoprolol in each intestinal segment. Importantly, colonic permeability values for both carvedilol and metoprolol were higher than that of metoprolol in the jejunum (illustrated as a dashed black line in Figure 5), validating the biopharmaceutical suitability of carvedilol, and metoprolol, to be developed as CR drug products. Typically, a CR drug product releases the drug continuously over 12–24, and since the transit time throughout the small intestine is 3-4 hours [60], the majority of the dose is released in the colon. This explains why high colonic drug permeability is a key biopharmaceutical factor in the decision process of CR drug product development. This permeability analysis of both model drugs

(carvedilol and metoprolol) demonstrated the decision process required for successful CR dosage form development.

Permeability studies both in vitro and in vivo (Log D, PAMPA, SPIP) resulted in higher permeability of carvedilol vs. metoprolol in all of the investigated segments/pHs. Both PAMPA and SPIP methods (Figures 4 and 5) showed the same upward trend. The in vitro permeability models used in this work account for simple passive diffusion, without taking into account intestinal transporters, however, the in vivo SPIP model accounts for all permeability mechanisms, including active transport. When looking at the in vivo results (Figure 5), it can be seen that throughout the entire intestinal tract, carvedilol's permeability was higher than that of metoprolol's in the jejunum. In the colon (Figure 5), carvedilol's permeability was higher than both metoprolol's high/low permeability benchmark and permeability of carvedilol in the jejunum. The permeability of carvedilol in the colon was lower than in the ileum, likely due to a shift in the ionization state. This correlation between artificial and animal permeability studies depicts the main mechanism of carvedilol's permeability as passive absorption. Furthermore, the octanol-buffer distribution coefficient of carvedilol was tremendously higher than that of metoprolol at different pH values (Figure 2). Even though Log P and Log D values are widely used as a replacement for passive intestinal permeability, relying solely on physicochemical drug properties when assessing drug permeation may lead to incorrect conclusions. For instance, the polar surface area (PSA) of carvedilol and metoprolol is 75.7 A<sup>2</sup> and 50.7 A<sup>2</sup> (Table 2), respectively [44,46]; lower PSA is usually associated with higher permeability, and hence, judging merely based on this characteristic would lead to the wrong conclusion. Therefore, prior to assigning a BCS classification, the many relevant aspects need to be thoroughly considered, to circumvent misconception in drug research, development, and regulation.

The solubility–permeability interplay is an important part of evaluating a novel drug formulation. By merely looking at the solubility improvement that the formulation allows can be ambiguous in terms of predicting the consequent oral drug absorption, and vice versa, this interplay must be accounted for when aiming to optimize the solubility–permeability balance, and the overall drug absorption. Carvedilol is a low solubility compound whose solubility enhancement when developing a CR drug product relied on using a phosphate salt in the CR formulation. This increase in solubility might be responsible for a slight decrease in overall bioavailability. However, in the case of this drug product, it did not affect the clinical efficacy of carvedilol.

Carvedilol is both a substrate [26,61] and inhibitor [62] of the efflux transporter P-glycoprotein (P-gp). The involvement of intestinal transporters in general, and specifically P-gp, in the absorption process following oral administration is more biorelevant for low-permeability drugs, and the regional-dependent expression of the relevant transporters should be considered in these cases [63,64]. However, for high-permeability compounds, neither active uptake nor efflux transporters are expected to be rate-limiting [65,66]. Given that carvedilol has very high passive intestinal permeability throughout the entire GIT (Figure 5), the fact that it is a P-gp substrate would not be significant in the in vivo conditions. In addition, as mentioned before, carvedilol undergoes extensive stereoselective first-pass (CYP2D6 and CYP2C9). Similarly to transporters, intestinal enzymes may also exhibit regional-dependent expression, which needs to be accounted for when developing oral CR formulation [67]. Extensive knowledge of intestinal/hepatic transport and enzymatic metabolism is essential in the development process of a CR product.

#### **5. Conclusions**

Altogether, the analysis of carvedilol/metoprolol presented in this work serves as a model for a suitable candidate for a CR product development, from both the permeability and solubility/dissolution point of view. This work may increase our biopharmaceutical understanding of a successful CR drug product.

Regional-dependent drug permeability and solubility/dissolution, and the effects of these factors on CR drug product development is often overlooked, and in this article, we aimed to emphasize these important issues; yet, additional data, including pharmacokinetics, metabolism, and pharmacotherapy considerations, are essential for the thorough prediction of a CR candidate.

**Author Contributions:** M.M., M.Z., N.F.-S., E.H., I.G.-Á., and A.D. worked on conceptualization, methodology, investigation, analyzed the data, and outlined the manuscript. A.D., M.M., and M.Z. wrote the skeleton of the paper, and all authors contributed to the writing, review, and editing of the full version. All authors have read and agreed to the published version of the manuscript.

**Funding:** This work received no external funding.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### *Article* **Cubic Microcontainers Improve In Situ Colonic Mucoadhesion and Absorption of Amoxicillin in Rats**

**Juliane Fjelrad Christfort 1, \* , Antonio José Guillot 2 , Ana Melero 2, \*, Lasse Højlund Eklund Thamdrup 1 , Teresa M. Garrigues 2 , Anja Boisen 1 , Kinga Zór <sup>1</sup> and Line Hagner Nielsen 1**


Received: 24 March 2020; Accepted: 10 April 2020; Published: 14 April 2020

**Abstract:** An increased interest in colonic drug delivery has led to a higher focus on the design of delivery devices targeting this part of the gastrointestinal tract. Microcontainers have previously facilitated an increase in oral bioavailability of drugs. The surface texture and shape of microcontainers have proven to influence the mucoadhesion ex vivo. In the present work, these findings were further investigated using an in situ closed-loop perfusion technique in the rat colon, which allowed for simultaneous evaluation of mucoadhesion of the microcontainers as well as drug absorption. Cylindrical, triangular and cubic microcontainers, with the same exterior surface area, were evaluated based on in vitro release, in situ mucoadhesion and in situ absorption of amoxicillin. Additionally, the mucoadhesion of empty cylindrical microcontainers with and without pillars on the top surface was investigated. From the microscopy analysis of the colon sections after the in situ study, it was evident that a significantly higher percentage of cubic microcontainers than cylindrical microcontainers adhered to the intestinal mucus. Furthermore, the absorption rate constants and blood samples indicated that amoxicillin in cubic microcontainers was absorbed more readily than when cylindrical or triangular microcontainers were dosed. This could be due to a higher degree of mucoadhesion for these particular microcontainers.

**Keywords:** in situ perfusion; microdevices; shape; mucoadhesion; colon absorption

#### **1. Introduction**

Oral drug delivery remains the preferred administration route due to ease of use, flexibility and patient compliance [1]. Despite many advances in oral delivery systems [2–4], the design of 'the ideal delivery device' is still widely discussed and depends on the application.

In the past decades, an increased interest in colon drug delivery has led to significant research with a focus on designing delivery devices that target this part of the gastrointestinal (GI) tract [5]. In addition to local delivery, the colon has also been suggested as an interesting site for systemic drug delivery due to increased oral bioavailability for some drugs and longer transit time compared to the small intestine [6–8]. The prolonged transit time allows timing of the treatment to periods with maximal disease activity; for example, in diseases where symptoms are more pronounced in the morning (hypertension, asthma and arthritis) [7].

Mucoadhesion is an important factor in relation to targeted delivery to any part of the GI tract, since it can prolong the residence time and facilitate drug release in close proximity to the epithelium. To understand mucoadhesion, six general theories have been proposed [9]. Amongst these are the wetting theory and the mechanical theory. The wetting theory is mainly applied to liquid or low-viscosity systems [9], while the mechanical theory can be applied to more rigid and adhesive materials. The mechanical theory explains mucoadhesion in terms of interlocking into irregularities on a rough surface [9,10]. Due to the complex nature of mucoadhesion, it is not likely that the phenomenon can be described by one of these theories alone [9]. Properties affecting mucoadhesion have been thoroughly investigated and it has been indicated that size and shape have a high impact on the mucoadhesive strength for micro- and nano-scale particles [11–13]. Advanced polymeric particles have paved the way for many new controlled drug delivery strategies [14]. However, as the field of drug delivery is moving towards more advanced microdevices, additional knowledge is needed to fully disclose and understand the influence of the multitude of parameters influencing mucoadhesion of drug delivery devices and the associated absorption of drugs.

Unidirectionally-releasing microdevices have been proposed for oral drug delivery to ensure a high local concentration of the active pharmaceutical ingredient (API) at the absorption site and to prolong the residence time [15,16]. The prolonged residence time is proposed to occur by decreased shear stress and increased retention [17]. For example, planar microdevices with a diameter of 200 µm were shown to enhance in vivo retention after oral dosing in mice [15]. These microdevices were found to adhere better than microspheres with the same surface area in the proximal and medial intestine. In the colon, however, the microspheres were retained approximately two times more than the microdevices [15]. The concept of planar microdevices was further developed with the inclusion of nanostraw structures on the surface, which was shown to enhance bioadhesion in a Caco-2 cell flow system when compared to similar microdevices without nanostraws [18]. Microcontainers, one type of unidirectionally-releasing microdevice, have previously been shown to improve oral bioavailability of drugs and provide inherent mucoadhesion [16,19]. Microcontainers are micrometer-sized devices (fabricated in polymers) with an inner cavity for storage of the API. Coating of the API-loaded cavity protects the content from the harsh environment of the stomach and provides unidirectional release at a relevant site in the GI tract [16]. Due to versatile fabrication processes, microcontainers have been fabricated in different materials, shapes and size ranges [20–22].

The influence of material composition, shape and size on the interaction with mucus has also been evaluated for microcontainers in an intestinal ex vivo perfusion model [22–24]. Here, triangular microcontainers adhered more readily in the mid-part of a porcine small intestinal section than cylindrical ones. Similarly, larger cone shaped microcontainers generally adhered better than cylindrical microcontainers with the same size [22,24]. When evaluating these microdevices fabricated in different materials in an ex vivo perfusion model, poly(lactic-*co*-glycolic acid) (PLGA) (50:50) microcontainers showed slightly higher mucoadhesion compared to polycaprolactone (PCL) and PLGA (75:25) microcontainers [23]. In a different study performed in the same model, there was a tendency that SU-8 microcontainers adhered better to the mucosa than PCL microcontainers with the same dimensions [22].

Techniques like the perfusion model described above, or atomic force microscopy (AFM), are commonly applied to study mucoadhesion ex vivo [25,26]. However, these techniques do not allow for simultaneous evaluation of the permeation across the intestinal barrier. In order to study API permeation, methods like in vitro studies with Caco-2 cells or ex vivo studies using an Ussing chamber setup are traditionally used [27,28].

A well-documented method to explore the performance of drug delivery systems in situ is the intestinal perfusion technique in rodents, introduced by Schanker in 1958 [29]. Several decades later, the technique is still widely applied in the field of intestinal absorption research due to its versatility [30–32]. Furthermore, the in situ technique provides the opportunity to investigate absorption and mucoadhesion simultaneously in a specific region of the intestine. These are important characteristics for evaluation of both local delivery and formulations aiming for prolonged release followed by absorption.

This in situ perfusion technique is applied in anesthetized rodents, where the relevant part of the intestine is isolated from nearby tissues, but not removed from the living organism. The perfusion can be performed as single-pass intestinal perfusion or with a closed-loop technique [32]. In both cases, the neural, endocrine, blood and lymphatic contributions are maintained during the experiment in order to simulate in vivo conditions. Both perfusion techniques have shown equally good correlations to literature values of the absorbed fraction of the oral dose (Fabs) in humans [32,33]. The closed-loop in situ perfusion technique based on Doluisio's method [34] has been widely applied for intestinal absorption of different drugs and in different regions of the intestine [35–37]. Earlier, the closed-loop intestinal perfusion technique has been applied to investigate the mechanisms governing absorption from drug suspensions [38] and drug-loaded microdevices [16] in the small intestine. The closed-loop perfusion technique has previously been validated to study colonic absorption [37], but has not yet been applied to study mucoadhesion and absorption simultaneously in this region of the GI tract.

Previously, we have seen that the shape and surface texture of microdevices influence mucoadhesion ex vivo [22,39]. More specifically, triangular microcontainers previously resulted in improved mucoadhesion in parts of a small intestinal section compared to cylindrical microcontainers [22]. However, these microcontainers were not normalized with respect to the surface area. Thus, it is unclear whether the shape or the difference in surface area was the reason for this observed effect. To elaborate on this, the present study aimed to evaluate differently shaped microcontainers, with the same exterior surface area, using a closed-loop in situ intestinal perfusion technique. This technique allowed us to simultaneously evaluate mucoadhesion of microcontainers and absorption of a model drug, amoxicillin, in the colon of rats. Cylindrical, triangular and cubic microcontainers were evaluated regarding in vitro release, in situ mucoadhesion and in situ absorption of amoxicillin. Additionally, the mucoadhesion of empty cylindrical microcontainers with pillars on the top surface were compared to a control without pillars.

#### **2. Materials and Methods**

#### *2.1. Materials*

Amoxicillin trihydrate was bought from TCI (Tokyo, Japan) and Eudragit® L100 was acquired from Evonik Industries (Essen, Germany). Dibutyl sebacate, isopropanol and potassium phosphate monobasic for the HPLC mobile phase were purchased from Sigma Aldrich (St. Louis, MO, USA). Methanol was bought from VWR International (Radnor, PA, USA).

The negative epoxy based photoresist SU-8 was used for production of microcontainers. Formulations with two different viscosities were used (i.e., SU-8 2035 and 2075) and the cross-linked structures were developed in Developer mr-Dev 600. Resist and developer were purchased from Micro Resist Technology GmbH (Berlin, Germany). Single-side polished ø100 mm Si substrates with a thickness of 525 µm were acquired from Topsil Globalwafers A/S (Frederikssund, Denmark).

For the phosphate buffered saline (PBS) used in the in situ perfusion studies, sodium chloride, potassium chloride, sodium phosphate dibasic and potassium phosphate monobasic were purchased from Scharlab (Barcelona, Spain). Animals were from Charles River Laboratories (Quebec, QC, Canada). Sylgaard 184 silicone elastomer kit was purchased from Dow Chemical (Midland, MI, USA). Ultrapure water used throughout the studies was obtained from a Q-POD® dispenser (Merck Millipore, Burlington, MA, USA).

#### *2.2. Fabrication of SU-8 Microcontainers*

Microcontainers with five different designs were used in these experiments. Amoxicillin-loaded microcontainers enteric coated with Eudragit® L100, were produced as 3D structures in three different shapes; cylindrical, equilateral triangular prism and quadrangular prism (Table 1). The central cylindrical compartment for drug loading was designed with a constant volume for all shapes. Furthermore, the three shapes were designed to have a constant outer surface area when neglecting

the top surface (which was coated). By maintaining a constant interaction area, the studies aim to isolate the influence of geometry on the colonic mucoadhesion. In addition to the above mentioned microcontainers, cylindrical microcontainers with and without 35 µm pillars (in diameter) on the sidewalls were produced (Table 1). λ λ


**Table 1.** Design parameters of the microcontainers evaluated in the present study. Data represents mean ± SD and *n* = 5 unless otherwise specified.

a the diameter for all cylindrical microcontainers and the side length for the cubic and triangular microcontainers. b corresponding to 309243.3 µm<sup>2</sup> . <sup>c</sup> *<sup>n</sup>* <sup>=</sup> 3. <sup>d</sup> *<sup>n</sup>* <sup>=</sup> 6. <sup>e</sup> pillar dimensions: 41 <sup>µ</sup>m high with a diameter of 35 <sup>±</sup> 2.2 <sup>µ</sup>m.

All microcontainers were produced following an approach first introduced for drug delivery devices in [40], then adapted and modified in [41]. Starting out with clean Si substrates, a release layer consisting of 5 nm Ti and 20 nm Au was deposited (Temescal FC-2000, Ferrotec Corporation, Santa Clara, CA, USA) using electron-beam evaporation. The release layer ensures adequate adhesion during drug loading and lid coating, but allowed for harvesting the microcontainers without damaging the SU-8 structures. All microcontainers featured an approximately 35 µm thick bottom layer formed by spin coating SU-8 2035 (RCD8 manual spin coater, Süss MicroTec, Garching, Germany). This was followed by performing a soft bake at 50 ◦C for 2 h (ramping rate 2 ◦C/min, used for all baking steps), conducting UV exposure using doses in excess of 200 mJ/cm<sup>2</sup> and then carrying out a post exposure bake (PEB) at 50 ◦C for 6 h using the aforementioned ramping rate. The UV exposure was conducted using a maskless aligner (MLA100 Tabletop Maskless Aligner, Heidelberg Instruments, Heidelberg, Germany) or a conventional mask aligner (Karl Süss Mask Aligner MA6, Süss MicroTec, Garching, Germany) operated in soft contact mode. The sidewalls of the microcontainers were defined in SU-8 2075, which was spin coated and subject to a soft bake at 50 ◦C for 10 h. The UV exposure was conducted using a dose of 500 mJ/cm<sup>2</sup> . For defining the cylindrical empty reference microcontainers with no pillars on the sidewalls, the mask aligner was operated in proximity exposure mode to avoid direct contact between the soda-lime glass mask and the SU-8. The remaining four microcontainer types were UV exposed using the maskless aligner. After UV exposure, a PEB of 10 h at 50 ◦C was carried out. To form the pillars on the sidewalls of one of the empty control microcontainer groups, a final layer of SU-8 2035 was spin coated, soft baked at 50 ◦C for 2 h, UV exposed using a dose of 200 mJ/cm<sup>2</sup> and finally subjected to a PEB at 50 ◦C for 6 h. The substrates, carrying the multitude of different drug delivery devices, were developed by immersion in two separate baths for 2 × 20 min and flushed with copious amounts of isopropanol before leaving the substrates to dry. Prior to drug loading

and lid coating, the Si substrates were diced (Automatic Dicing Saw DAD 321, DISCO, Tokyo, Japan) into 12.8 <sup>×</sup> 12.8 mm<sup>2</sup> chips, each containing 625 microcontainers arranged in a 25 × 25 array.

The microcontainers were characterized using both conventional bright-field microscopy (Nikon Eclipse L200, Nikon Metrology, Tokyo, Japan) for extracting data on the horizontal dimensions and vertical scanning interferometry (PLu Neox 3D Optical Profiler, Sensofar Metrology, Barcelona, Spain) aimed at characterizing the vertical dimensions (i.e., inner and outer heights and pillar height). The results of the topology characterization are summarized in Table 1. For the horizontal measurements (i.e., diameters and side lengths), a value corresponding to the optical resolution multiplied by a factor of 3 was used. The optical resolution, R, of the used objective (20×, NA = 0.45) was evaluated as R = (0.61λ)/NA where the illumination wavelength was λ = 550 nm. This resulted in the stated measurement uncertainty of ±2.2 µm. When considering the height measurements, we used the standard deviation based on an ensemble of measurements. Generally the horizontal dimensions are subject to small variations and the spread in the vertical dimensions are governed by the homogeneity of the spin coated SU-8 layers.

#### *2.3. Loading of Amoxicillin into Microcontainers and Spray Coating with Eudragit*® *L100*

For the in situ perfusion study, the microcontainer chips were loaded with amoxicillin as previously described [16,42]. Briefly, amoxicillin was manually distributed on the microcontainer chip and excess drug between the microcontainers was subsequently removed with an air gun. For the in vitro release studies, the microcontainers were loaded using a PDMS shadow mask to cover the chip area between the microcontainers, as previously described [43]. In both cases, the chip with microcontainers was weighed before and after loading to determine the loaded amount of amoxicillin. Visualization of the loading was carried out by scanning electron microscopy (SEM) using a Hitachi TM3030Plus tabletop microscope (Hitachi High-Technologies Europe, Krefeld, Germany).

A lid of Eudragit® L100 was deposited over the microcontainers as previously described [19], by spray coating an isopropanol solution with 1% *w*/*v* Eudragit® L100 and 5% *w*/*w* (in relation to the polymer) dibutyl sebacate. Briefly, the solution was sprayed over the amoxicillin-loaded microcontainers one chip at a time using an ultrasonic spray coater (Exactacoat system, Sono-Tek, Milton, NY, USA) with an Accumist nozzle operating at 120 kHz. Each chip was coated with 30 loops of two alternating spray paths having an offset of 2 mm, resulting in a total of 60 passages. Visualization of the coated microcontainers was carried out by SEM.

#### *2.4. In Vitro Release Studies*

The release of amoxicillin from the differently shaped microcontainers coated with Eudragit® L100 was measured using a µDISS ProfilerTM (Pion Inc., Billerica, MA, USA), as previously described in the literature [16,19,42]. Initially, a calibration curve was prepared by addition of different volumes of amoxicillin in PBS stock solution to 10 mL PBS (pH 7.4) followed by measurements of UV absorbance in the range of 270–280 nm. For the release study, a microcontainer chip was placed on top of a cylindrical magnetic stirring bar with double-sided carbon tape, transferred to a sample vial and covered with 10 mL PBS at the same time as the experiment was started. Studies were performed at 37 ◦C with a stirring rate of 100 rpm, and the path length of the UV probes was 5 mm. UV measurements were carried out every 10 s until an amoxicillin release of 100% was observed after approximately 60 min.

#### *2.5. Closed-Loop Colon Perfusion Study in Rats*

Male Wistar rats were used in accordance with 2010/63/EU directive of 22 September 2010 regarding the protection of animals used for scientific experimentation. The Ethics Committee for Animal Experimentation of the University of Valencia approved the experimental protocols (code A1544541996825). Male Wistar rats weighing 240 ± 12 g were fasted for 3 h before the experiments with ad libitum access to water. The rats were anesthetized by intraperitoneal injection of pentobarbital sodium (40 mg/kg) prior to the surgical procedure.

The drug absorption rate constant of amoxicillin in the colon and the mucoadhesion of microcontainers were evaluated by the in situ closed-loop perfusion method based on Doluisio's technique [34]. Briefly, the animals were placed under a heating lamp and a midline abdominal incision was made to expose the intestine. The colon section was isolated by making two incisions; one after the caecum and the other just before the rectum. Two glass syringes connected to three-way stopcock valves were introduced in the incisions with the help of two cannulas, creating an isolated compartment as depicted in Figure 1. Procedures were performed with care to avoid disturbance of the intestinal blood supply and intestinal bleeding. In order to remove all intestinal content and wash the colon, the intestinal section was thoroughly flushed with PBS pre-heated to 37 ◦C. The colon was carefully placed back into the peritoneal cavity and the abdomen was covered with cotton wool pads to prevent peritoneal liquid evaporation and heat loss [16,32,35].

**Figure 1.** Schematic overview of the in situ colonic perfusion study. The different microcontainers were dosed through two cannulas connected to glass syringes creating a closed colon compartment. To investigate the absorption of amoxicillin, intestinal samples were collected from the two glass syringes every 5 min throughout the experiment and a blood sample was collected 30 min after the experiment. The plasma and intestinal samples were analyzed with high performance liquid chromatography (HPLC) to determine the concentration of amoxicillin. The mucoadhesion of the microcontainers was evaluated as the percentage adhering to the colonic section after the study.

A number of microcontainers, corresponding to 0.4 mg amoxicillin (120 cylindrical, 278 cubic or 136 triangular microcontainers) were dispersed in 5 mL pre-heated PBS and introduced through the cannulas into the isolated section. For the empty reference cylinders without and with pillars, 204 ± 25 and 223 ± 45 microcontainers, respectively, were dosed in the same manner. Samples of 150 µL were collected every 5 min up to a period of 30 min (Figure 1). Sample withdrawal was performed by pushing the luminal content from one syringe to the other, alternatively from the proximal syringe to the distal one and the other way around [44]. Intestinal samples were stored at −20 ◦C until further analysis.

Water flux absorption processed during the experiment could be significant, and hence it must be considered [45]. To do so, the volume of the intestinal content was measured in every animal after the whole procedure (Vt) and compared to the initial volume (V0) of 5 mL. The drug concentration in the samples was corrected according to Equation (1):

$$\mathbf{C}\_{t} = \mathbf{C}\_{\mathbf{e}} \text{ (V}\_{t}/\text{V}\_{0})\_{\prime} \tag{1}$$

where C<sup>t</sup> represents the concentration in the absence of water reabsorption at time t, and C<sup>e</sup> is the experimental value. The corrected concentration, C<sup>t</sup> , was then used to calculate the actual absorption rate coefficient in relation to the initial concentration (C0) [45]. The absorption rate coefficients (ka) of the compounds were determined by nonlinear regression analysis of the remaining concentrations in the intestinal lumen (Ct) versus time according to Equation (2):

$$\mathbf{C}\_{\mathbf{t}} = \mathbf{C}\_{0} \,\mathrm{e}^{-\mathrm{k}}\,\mathrm{t} \tag{2}$$

After 30 min, a cardiac puncture was performed under anesthesia to collect the blood from the rat (Figure 1). The blood was collected in heparinized tubes and centrifuged at 10 ◦C and 6100× *g* for 8 min. The plasma was stored at −20 ◦C until further analysis. After the experiment, the isolated colon section was cut and placed onto a glass slide with the luminal side upwards to determine the number of microcontainers. A light microscope (Nikon Eclipse 50i, Nikon, Tokyo, Japan) with camera (Nikon digital camera, DXM1200C, Nikon, Tokyo, Japan) was used to visualize the microcontainers on the colon section.

#### *2.6. High Performance Liquid Chromatography Analysis of Intestinal and Plasma Samples*

The concentration of amoxicillin in the intestinal fluid and plasma was determined by high performance liquid chromatography (HPLC). HPLC analyses were performed on a Shimadzu HPLC system (Shimadzu, Kyoto, Japan). The system consisted of a CBM-20A system controller, SIL-20AC HT auto sampler, LC-20AD pump, DGU-20A5R degassing unit, CTO-20AC column oven, RID-20A refractive index detector, and SPD-30A photodiode array detector. The mobile phases consisted of A: phosphate buffer (6.8 g KH2PO4/L, pH 5) and B: acetonitrile, and the samples were run with an isocratic method with an A:B mobile phase ratio of 95:5 v/v. A Luna 5.0 µm C18 100 Å, 250 × 4.6 mm column (Phenomenex ApS, Værløse, Denmark) was used for the analyses and samples were run at 25 ◦C.

The intestinal samples were vortexed and centrifuged at 10,600× *g* for 10 min and the supernatant was transferred to HPLC vials with 300 µL flat bottom inserts (Frisenette, Knebel, Denmark). For the plasma samples, 60 µL plasma was mixed with 100 µL methanol and otherwise treated as described above for the intestinal samples. Calibration curves were prepared from stock solutions of amoxicillin in water. For the plasma sample analysis, accurate volumes of the stock solutions were mixed with plasma and methanol (same ratio as the samples) and treated the same way as the samples. A volume of 20 µL was injected and the flow rate was 0.8 mL/min with a run time of 10 min for each sample. The absorbance was measured at 230 nm.

#### *2.7. Data Analysis*

All data processing was performed in Microsoft Excel 2016 (Redmond, WA, USA), GraphPad Prism version 6.0 (GraphPad software, San Diego, CA, USA) and SPSS version 22.0 (IBM Corp, Armonk, NY, USA). The data is expressed as the mean ± standard deviation (SD) unless otherwise stated. Statistical differences were determined using one-way ANOVA followed by Games–Howell post-hoc analysis, where *p*-values below 5% were considered significant.

#### **3. Results and Discussion**

#### *3.1. Microcontainer Characterization and Preparation*

For the present study, which addresses the impact of microcontainer geometry on the colonic mucoadhesion and absorption during in situ experiments in rats, three 3D container designs were investigated: cylindrical, cubic and equilateral triangular (Table 1). The microcontainers all had a cylindrical center compartment for drug storage and the outer surface area (neglecting the top surface) was kept constant to ensure an identical interaction area between microcontainers with different shapes and the mucosal layer in the colon.

The microcontainers were loaded with 1.50 ± 0.27 mg, 0.89 ± 0.19 mg and 1.78 ± 0.18 mg amoxicillin per chip for the cylindrical, cubic and triangular microcontainers, respectively (Figure 2a–c). Despite having the same inner cavity volume for drug loading, the cubic microcontainers were loaded with significantly less amoxicillin than the two other shapes. This is ascribed to the manual loading process, where the additional corners can affect the loading efficiency. After drug loading, the microcontainers were coated with Eudragit® L100. Inspection with SEM showed uniform coatings covering the cavities of the drug-loaded microcontainers for all three shapes (Figure 2d–f).

**Figure 2.** SEM images of a (**a**) cylindrical, (**b**) cubic and (**c**) triangular microcontainer loaded with amoxicillin and a (**d**) cylindrical, (**e**) cubic and (**f**) triangular microcontainer loaded with amoxicillin and coated with Eudragit® L100. All scale bars represent 100 µm.

Besides the three shapes described above, cylindrical reference microcontainers with pillars on the top surface of the sidewalls were also fabricated (but not loaded with drug) and the mucoadhesion of these were compared to the mucoadhesion of traditional empty cylinders of similar dimensions as the control (Table 1).

#### *3.2. In Vitro Release Studies*

To investigate the release rate, the in vitro release of amoxicillin from the microcontainers was evaluated on a µDISS ProfilerTM in PBS at pH 7.4 (Figure 3).

For all three formulations, the loaded amount of amoxicillin was released after 60 min (98 ± 1%, 98 ± 3% and 94 ± 3% for the cylindrical, cubic and triangular microcontainers, respectively). The observed in vitro release of amoxicillin from the Eudragit® L100-coated microcontainers was expected since the coating dissolves at pH values above 6.0. After 30 min, a release of approximately 60% was observed for the cylindrical and triangular microcontainers, whereas there was a trend towards slower release of amoxicillin from the cubic ones (44 ± 10% after 30 min).

After 45 min, at least 80% of the initial amoxicillin dose was released, which categorizes the formulation as an immediate release formulation according to the European Pharmacopoeia [46]. Comparable pH-dependent release profiles have previously been observed for drugs loaded in microcontainers and coated with Eudragit® L100 [16,47].

**Figure 3.** Amoxicillin in vitro release profiles as a function of time from cycylindrical, lindrical, cucubic and triang and triangular microcontainers in PBS (pH 7.4). All microcontainers were coated with Eudragit® L100. Data represent mean ± SD, *n* = 4.

#### *3.3. In Situ Closed-Loop Colon Perfusion Study in Rats*

The in situ closed-loop perfusion technique was applied to study the interaction between microcontainers and the colonic mucus layer, and whether this interaction affected the absorption of amoxicillin from the microcontainers compared to a control solution.

#### 3.3.1. Mucoadhesion of Microcontainers

After 30 min of the perfusion study, the microcontainers were manually localized and counted by inspecting the colon sections under a light microscope (Figure 4). When qualitatively investigating the microcontainers retained in the colonic mucus, it was observed that the microcontainers were mainly found in clusters that were partly or completely covered by mucus (Figure 4). Similar clustering trends have previously been observed for other microdevices evaluated on a cell monolayer under flow [17].

The mucoadhesion of the microcontainers was quantified as the percentage of microcontainers adhering to the colonic mucus after 30 min compared to the total amount of microcontainers dosed (Figure 5). It was found that 12 ± 7% of the loaded cylindrical microcontainers were retained in the colonic mucus. In contrast, a significantly higher (*p* = 0.019) number of cubic microcontainers (33 ± 12%) were found to be retained in the colon sections after the same period of time (Figure 5). The percentage of the triangular microcontainers in the colonic mucus was 28 ± 26% and a higher inter-individual variation was observed for these rats compared to the rats in the other groups (Figure 5).

Based on the data presented in Figure 5, the only significant difference was between the cubic and cylindrical microcontainers loaded with amoxicillin. The absence of significant differences between the other groups can be explained through the rather large data distribution in the group dosed with triangular microcontainers, which varied between 3 and 81% (Figure 5). As expected, the shape with the least pronounced mucoadhesion was the cylindrical one, since this shape did not provide any corners or edges to allow for interaction with the mucus. The most mucoadhesive shape seemed to be the cube, although the differences were non-significant due to the large variations observed for triangular microcontainers. If the cubic and triangular microcontainers are analyzed based on geometry/topology, there are obvious differences between the two shapes. The cubic structures have 6 surfaces with approximately the same area and shape, 12 edges and 8 corners, whereas the triangles have 5 surfaces, 9 edges and 6 corners. We believe that the number of surfaces, corners and edges

strongly influence the way the microcontainers are retained in the mucus. These shape differences would also result in different contact surfaces between the microcontainers and the mucus, and thus, differences in mucoadhesion according to the wetting theory [9]. The cylinders will for example have a much smaller contact surface with the mucus if they land on their curved side.

**Figure 4.** Microscopy images of microcontainers in colonic mucus following in situ perfusion studies. (**a**,**b**) cylindrical microcontainers, (**c**) pillared reference microcontainers, (**d**) cubic and (**e**) triangular microcontainers. All scale bars represent 100 µm.

**Figure 5.** Mucoadhesion of the five microcontainer formulations expressed as percentage of dosed microcontainers still adhering to the mucus after the closed-loop intestinal perfusion study (mean ± SD, *n* = 4–7). \* indicates a significant difference with a *p*-value below 5%.

A previous study investigated the mucoadhesion of cylindrical and triangular microcontainers in an ex vivo perfusion model and found triangular microcontainers to be significantly more mucoadhesive in the mid-part of the intestinal section [22]. This finding, in relation to the results in the present study about cubic microcontainers being more adhesive than cylindrical microcontainers, suggest that the presence of corners or edges can influence the mucoadhesion properties. However, the changing properties of the mucosa along the GI tract makes it difficult to directly extrapolate findings from one intestinal region to another.

We did not expect the material SU-8 to have any significant influence on mucoadhesion in itself. Even if SU-8 interacted with the mucus layer, the effect of this interaction would be similar for all the shapes since they have been normalized to have the same exterior surface area. Hence, we would not expect the differently shaped microcontainers to adhere differently as a result of interfacial interactions relating to material properties. Thus, we expect that the observed interaction with the intestinal mucosa is mostly due to mechanical detainment due to the differences in shape rather than chemical interactions.

In addition to the influence of the shape itself, the mucoadhesive effect of pillars applied to the top surface of cylinders was investigated (Figure 5). The amount of reference cylindrical microcontainers with pillars adhering to the colonic mucus was found to be 16 ± 13% after 30 min, which was very similar to the percentage of conventional reference and loaded cylinders (13 ± 9% and 12 ± 7%, respectively). In accordance with the mechanical theory of mucoadhesion described in the introduction, the pillars were introduced on the reference cylinders as bioinspired structures to increase adhesion by filling the imperfections of a rough surface. However, unexpectedly, no differences in mucoadhesion were observed by adding these small structures on top of the reference microcontainers. This could be attributed to the size or number of pillars, which might have been insufficient in order to interact with the mucus in a significant way. Finally, it was observed that handling of the reference microcontainers with pillars resulted in loss of some pillars when the microcontainers were evaluated with SEM before dosing, which could outweigh the potential adhesion effect of the pillars. To fully investigate the concept of surface structures further, additional studies are needed. Nevertheless, the microcontainers themselves seem to present adequate geometrical forms which can be detained in the mucus layer without help from smaller structures on the surface.

Surface modified microdevices have previously been found to increase adhesion in vitro and ex vivo [18]. However, these microdevices were not investigated in situ or in vivo and the surface structures on these devices were remarkably smaller (60–160 nm) than the pillars on the surface of the microcontainers in the present study (41 µm in height and 35 µm in diameter). In a different study, the impact of larger and more complex surface structures was investigated ex vivo and these were found to have a large impact on the adhesive properties of the microdevices [39].

When comparing the loaded and coated microcontainers to the empty reference microcontainers with and without pillars, it is important to consider the possible effect of the lid coating. Eudragit® polymers have previously been found to possess mucoadhesive properties when applied on nanocapsules [48]. Thus, the coating itself could influence the adhesion of the microcontainers even if Eudragit® L100 is expected to dissolve quickly at pH 7.4. However, all three types of cylindrical microcontainers appeared to result in similarly low mucoadhesion (Figure 5), which indicates that the shape is the most important factor for mucoadhesion in the present study.

In two euthanized rats dosed with cylindrical reference microcontainers, a remarkably smaller number of microcontainers were found to adhere after 30 min than for other rats dosed with reference cylinders. These findings indicate that the microcontainers interact differently with the colonic mucus in the presence of peristalsis, irrigation and water-resorption processes, which emphasizes the importance of evaluating mucoadhesion in situ as well as ex vivo.

#### 3.3.2. Absorption of Amoxicillin

To address whether the mucus retention affected the absorption of amoxicillin from the microcontainers, blood and intestinal samples were collected. Based on the remaining concentrations in the intestinal lumen, the absorption rate constant (ka) was calculated for amoxicillin in solution and amoxicillin dosed via cylindrical, cubic and triangular microcontainers (Figure 6). The values for k<sup>a</sup> are relevant in order to evaluate how mucoadhesion affects the absorption of amoxicillin.

**Figure 6.** First-order absorption rate constants (ka) calculated from the closed-loop intestinal perfusion studies for Eudragit® L100 coated microcontainers loaded with amoxicillin and for an amoxicillin control solution (mean ± SD, *n* = 6).

− − − − For amoxicillin in cubic microcontainers, k<sup>a</sup> was calculated to be 2.5 <sup>±</sup> 0.6 h−<sup>1</sup> , which is not statistically different to the value obtained for the solution (2.6 <sup>±</sup> 0.4 h−<sup>1</sup> ) (Figure 6). For cylinders and triangular prisms, k<sup>a</sup> of amoxicillin was calculated to be 0.0 <sup>±</sup> 0.7 h−<sup>1</sup> and 0.0 <sup>±</sup> 0.5 h−<sup>1</sup> , respectively (Figure 6). In the case of the cubic microcontainers, absorption seems to be faster than the release, resulting in a positive ka. This could be related to the slower in vitro release observed from this shape (Figure 3). On the contrary, the concentrations of amoxicillin measured in the lumen after dosing with cylindrical and triangular microcontainers appeared to be constant during the whole experiment. This could indicate that the absorption and release occurred with the same rate, and, thus, the absorbed amount of amoxicillin was continuously replaced by the released amount.

After 30 min, a blood sample was collected to compare k<sup>a</sup> to the amount of amoxicillin absorbed from the colon during the experiment. In plasma, amoxicillin was mainly detected after dosing in solution and cubic microcontainers (0.26 ± 0.03 and 0.08 ± 0.02 µg/mL, respectively). On the contrary, amoxicillin could only be detected in plasma from one of the six rats dosed with triangular microcontainers (resulting in 0.02 ± 0.02 µg/mL for the group on average). Systemic uptake of amoxicillin was not detected in any of the blood samples from the rats dosed with cylindrical microcontainers. The absorption of amoxicillin has previously been shown to vary in different regions of the GI tract with limited absorption in the colon [49]. These region-dependent differences in absorption are believed to be caused by decreased levels of the uptake transport responsible for the absorption of amoxicillin [49,50].

In summary, the control solution and the cubic microcontainers were the formulations with the highest ka, which also resulted in the highest concentration of amoxicillin in the plasma after 30 min. Amoxicillin dosed in the control solution had the obvious advantage that it was already in solution and available for absorption, whereas the amoxicillin powder inside the microcontainers needed more time to be released, solubilized and then absorbed. Based on the in vitro release profiles (Figure 3), only approximately 40% of the dose was expected to be released in the intestinal medium after 30 min, which could explain the observed difference in plasma concentrations. The different preconditions, but yet comparable performances, for the solution and the cubic microcontainers, suggested that

the microcontainers must hold a different advantage, which might be related to the mucoadhesion. A high degree of mucoadhesion as observed for the cubic microcontainers would result in a high local concentration of amoxicillin facilitating the absorption.

Cylindrical microcontainers have previously been evaluated in an in situ closed loop perfusion model in the small intestine in order to investigate mucoadhesion and absorption of furosemide [16]. The microcontainers were found to adhere to the intestinal mucus and result in a higher absorption rate constant for furosemide when compared to a control solution [16]. The differences between this work and the present one can be attributed to the properties of the API and the intestinal section in which the absorption takes place.

#### **4. Conclusions**

In the present study, we investigated the influence of microdevice shape on colonic mucoadhesion and drug absorption by applying an in situ closed-loop intestinal perfusion technique. Cylindrical, triangular and cubic microcontainers were loaded with amoxicillin as a model drug and subsequently coated with Eudragit® L100. The amoxicillin release was evaluated in vitro and the absorption of amoxicillin and adhesion of microcontainers was evaluated in a closed-loop intestinal perfusion model in anesthetized rats.

In vitro, a complete amoxicillin release was observed after 60 min from the three types of microcontainers. From the microscopy analysis of the colon sections after the in situ perfusion study, it was evident that a significantly higher percentage of cubic microcontainers than cylindrical microcontainers (33 ± 12% and 12 ± 7%, respectively) was detained in the mucus. Additionally, the absorption rate constants and the blood samples indicated that amoxicillin in cubic microcontainers was absorbed more readily (2.5 <sup>±</sup> 0.6 h−<sup>1</sup> and 0.08 <sup>±</sup> 0.02 <sup>µ</sup>g/mL, respectively) than when cylindrical microcontainers (0.0 <sup>±</sup> 0.7 h−<sup>1</sup> and no absorption detected) or triangular microcontainers (0.0 <sup>±</sup> 0.5 h−<sup>1</sup> and 0.02 ± 0.02 µg/mL) were dosed. This could be due to a higher degree of mucoadhesion for these particular microcontainers.

With the present study, we have demonstrated that the in situ closed-loop intestinal perfusion model is a promising tool to evaluate overall performance of microdevices in a confined region of a rat intestine. Based on the presented results, the use of more complex microcontainer shapes including more edges and corners, such as star shapes, should be investigated in the future.

**Author Contributions:** Conceptualization, J.F.C., A.M., L.H.E.T., T.M.G., A.B., K.Z. and L.H.N.; methodology and supervision, J.F.C., A.M., A.B., K.Z. and L.H.N.; formal analysis, J.F.C., A.J.G., A.M., T.M.G., K.Z., L.H.N.; investigation and visualization, J.F.C., A.J.G., A.M., L.H.E.T.; resources, A.M. and A.B.; writing—original draft preparation, J.F.C., A.J.G. and L.H.E.T.; writing—review and editing, all authors; funding acquisition, T.M.G. and A.B. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by the Novo Nordisk Foundation (NNF17OC0026910), MIMIO–Microstructures, microbiota and oral delivery and by the Danish National Research Foundation (DNRF122) and Villum Foundation (Grant No. 9301), Center for Intelligent Drug Delivery and Sensing Using Microcontainers and Nanomechanics (IDUN). Furthermore, the research was conducted with support from the University of Valencia precompetitive project UV-INV-PRECOMP12-80750 and GV/2013/086 project of the Valencian Government (Generalitat Valenciana).

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

*Review*

### **The Segregated Intestinal Flow Model (SFM) for Drug Absorption and Drug Metabolism: Implications on Intestinal and Liver Metabolism and Drug–Drug Interactions**

### **K. Sandy Pang \*, H. Benson Peng and Keumhan Noh**

Leslie Dan Faculty of Pharmacy, University of Toronto, Toronto, ON M5S 3M2, Canada; hao.peng@mail.utoronto.ca (H.B.P.); keumhan.noh@utoronto.ca (K.N.) **\*** Correspondence: ks.pang@utoronto.ca; Tel.: +1-416-978-6164

Received: 12 March 2020; Accepted: 27 March 2020; Published: 1 April 2020

**Abstract:** The properties of the segregated flow model (SFM), which considers split intestinal flow patterns perfusing an active enterocyte region that houses enzymes and transporters (<20% of the total intestinal blood flow) and an inactive serosal region (>80%), were compared to those of the traditional model (TM), wherein 100% of the flow perfuses the non-segregated intestine tissue. The appropriateness of the SFM model is important in terms of drug absorption and intestinal and liver drug metabolism. Model behaviors were examined with respect to intestinally (M1) versus hepatically (M2) formed metabolites and the availabilities in the intestine (F<sup>I</sup> ) and liver (FH) and the route of drug administration. The %contribution of the intestine to total first-pass metabolism bears a reciprocal relation to that for the liver, since the intestine, a gateway tissue, regulates the flow of substrate to the liver. The SFM predicts the highest and lowest M1 formed with oral (po) and intravenous (iv) dosing, respectively, whereas the extent of M1 formation is similar for the drug administered po or iv according to the TM, and these values sit intermediate those of the SFM. The SFM is significant, as this drug metabolism model explains route-dependent intestinal metabolism, describing a higher extent of intestinal metabolism with po versus the much reduced or absence of intestinal metabolism with iv dosing. A similar pattern exists for drug–drug interactions (DDIs). The inhibitor or inducer exerts its greatest effect on victim drugs when both inhibitor/inducer and drug are given po. With po dosing, more drug or inhibitor/inducer is brought into the intestine for DDIs. The bypass of flow and drug to the enterocyte region of the intestine after intravenous administration adds complications to in vitro–in vivo extrapolations (IVIVE).

**Keywords:** segregated flow intestinal model (SFM); traditional model (TM); route-dependent intestinal metabolism; first-pass effect; drug-drug interactions; DDI; in vitro in vivo extrapolations; IVIVE

#### **1. The Intestine–Liver Unit**

The extent of the absorption of orally administered drugs is controlled by the intestine and liver, which are anatomically linked as a serial unit that is sequentially perfused by the circulation (Figure 1). The intestine is the gateway tissue to the liver and is important for drug absorption and first-pass removal. The superior mesenteric artery (SMA) supplies blood to the small intestine and its venous drainage, together with venous returns from the spleen, pancreas, gallbladder and gastrointestinal tract (GIT) including the stomach, constitute the hepatic portal vein flow (QPV), which is approximately 75% of the total liver blood flow, QH. Together with the hepatic artery (QHA), the remaining 25% of QH, the dual flows collectively perfuse the liver.

The intestine is endowed with absorptive transmembrane transporters in simple columnar, epithelial cells known as enterocytes that line the inner surfaces of the small intestine. These cells contain numerous protrusions known as the villi and microvilli that increase the surface area multiple-fold to absorb drug molecules or nutrients from the gut lumen. Intestinal absorption models have been classically linked to drug properties and the dosage form (pKa, logP, and solubility), as well as the physiology of the gastrointestinal tract (pH, gastrointestinal transit time, gastric emptying time, surface area, and microbiota) that control the fraction of dose absorbed (Fa) [1–11]. In addition to passive diffusion, absorptive transporters known as the apical solute carrier transporters (SLC), as exemplified by the PEPT1 (oligopeptide transporter 1), OATP1A2, OATP2B1 (the organic anion transporting polypeptide 1A1 and 2B1), MCT1 (the monocarboxylic acid transporter 1), ASBT (apical sodium dependent bile acid transporter) that reclaims bile acids, and OCT (organic cation transporter), facilitate the entry of weak acids and weak bases [12–18]. Counterbalancing drug entry are the efflux transporters—the P-gp (P-glycoprotein), BCRP (breast cancer resistance protein) and MRP2 (multidrug resistance-associated protein 2) that mediate drug or metabolite secretion back to the intestinal lumen [19,20], and this backward flux tends to reduce the net absorption of solutes. The OSTα and OSTβ (organic solute transporter α and β, half-transporters) transport bile acids out of the enterocytes [21]. It is well recognized that P-gp is capable of secreting highly lipophilic drugs [22,23]. Since lipophilic drugs with high solubility and permeability (Biopharmaceutical Classification System or BCS, Class I) are readily reabsorbed, the excretory function of P-gp is readily nullified [24]. The significance of P-gp, being more abundant distally in the ileum is, therefore, reduced for drugs that are readily reabsorbed [20,23,25,26]. However, for highly soluble but poorly permeable Class III BCS drugs, P-gp is more effective in reducing intestinal drug absorption [7]. It is also notable that drug permeability can be influenced by the pH of the intestinal lumen that becomes more and more basic and in turn, influence the extent of drug absorbed [3,8]. Segment-dependent decline in membrane permeability, reduced surface area from the duodenum to ileum [27] and pH changes along the intestine [8,28] are noted. These variables will modulate the extent of passive drug absorption. α β α β

− − − **Figure 1.** The intestine as a gateway tissue to the liver. Because of intestinal removal [extraction ratio, E<sup>I</sup> or (1 − F<sup>I</sup> )], the drug entering the liver is reduced, and the liver may further remove the drug with a liver extraction ratio (EH) to effect first-pass metabolism. The fraction absorbed, F<sup>a</sup> and F<sup>I</sup> or (1 − EI ), and F<sup>H</sup> or (1 − EH) influence the systemic bioavailability, Fsys. This figure was reproduced with permission from Noh and Pang [18], Wiley, 2019.

After crossing the intestinal membrane, the drug is met with metabolizing enzymes such as the cytochromes P450 3A (CYP3A) and UDP-glucuronosyltransferases, UGTs [29–32]. The most abundant CYP isoform is CYP3A4, which exceeds other isoforms such as 2C9, 2C19 > 2J2 > 2D6 that are present in lower quantities [31,33–35]. UGT 1A (1A1, 1A6, 1A5, 1A8, and 1A10) and 2B (2B7, 2B15, and 2B17) subfamilies are present to mediate the glucuronidation of morphine, raloxifene, mycophenolate, bisphenol A and gemfibrozil [36–40]. The intestinal metabolic activities for CYP3A4 and some of

the UGTs are comparable to, or higher than, those in the liver [31,41,42]. Cytosolic glutathione S-transferases [43,44] are found abundantly, whereas epoxide hydrolases [43] and sulfotransferases (SULT) [45] are present at much lower quantities in the intestine.

The availability of the intestine (F<sup>I</sup> ) after intestinal metabolism or secretion is defined as (1 − E<sup>I</sup> ) [where E<sup>I</sup> is the intestinal extraction ratio], and hepatic availability, FH, is given by (1 − EH) [where E<sup>H</sup> is the hepatic extraction ratio]. The overall systemic availability, Fsys, is given by FaFIFH. Following oral (po) drug dosing, the fraction of the dose absorbed (Fa) is attributed to dosage forms and/or solubility properties, intestinal removal via metabolism or secretion (defined by the intestinal extraction ratio, EI ), and liver removal (defined as the hepatic extraction ratio, EH), respectively. The product of the availabilities, FaFIFH, constitute the net fraction, the systemic availability, Fsys. For this reason, the intestine and liver are both capable of removing a significant proportion of the orally administered dose, a phenomenon known as the first-pass effect [46]. The extent of intestinal versus liver removal of drugs is therefore intimately related [47–50].

#### **2. Reason or Need for Intestinal Flow Models**

Although the development of clearance concepts for the intestine has lagged behind that for the liver [51–53], there have been some activities trending towards the fabrication of a useful and meaningful intestine clearance model to predict the extent of removal and examine how the intestine influences the rate of liver removal according to the route of drug administration. The correct intestinal model will exert serious implications in terms of drug–drug interactions (DDIs) with inducers or inhibitors, or in terms of in vitro–in vivo extrapolation (IVIVE).

#### **3. Route-Dependent Intestinal Metabolism**

Midazolam is a prototypic probe substrate of CYP3A4 metabolism that is often utilized for the screening of CYP3A4 and CYP3A5 activities in inhibition or induction studies [42,54–58]. Midazolam is metabolized by both the intestine and liver [42,59]. For the completely absorbed drug (F<sup>a</sup> ~ 1), there was a dramatically lower intestinal extraction ratio (E<sup>I</sup> = 0.08), measured across the arterial and hepatic portal venous blood for midazolam after its intravenous administration among anhepatic patients whose livers were removed during transplantation surgery [59]. In comparison, the mean fraction metabolized across the intestinal mucosa when given intraduodenally was much higher (E<sup>I</sup> = 0.43). This first, direct evidence uniquely shows route-dependent metabolism of the small intestine. Clinically, the erythromycin breath test relates well to the midazolam unbound liver clearance and not correlated to the intestinal clearance [60]. For radiolabeled (-)morphine that forms morphine 3-glucuronide (M3G) in both the intestine and liver, M3G was absent and undetectable in the vascularly perfused rat intestine preparation when morphine from the reservoir recirculated the rat intestine, a scenario akin to the systemic administration of morphine. This contrasts the copious presence of the radiolabeled M3G metabolite in both the intestinal lumen and reservoir after the intraduodenal administration of morphine into the gut lumen [61]. Additional animal and human studies attest to the same trend of a higher extent of intestinal metabolism after oral (po) than after intravenous (iv) drug administration (Table 1). These examples serve as direct evidence that display route-dependent metabolism of the small intestine. There will be a corresponding route-dependent change in the proportion of liver metabolites formed as well, since the unmetabolized drug leaving the intestine now enters the liver for further processing.



#### **4. Intestinal Flow Models: Segregated Flow (SFM), QGut, and Traditional (TM) Models**

Compartmental models are ill equipped to examine the extent of drug metabolism among metabolizing tissues or organs that are arranged serially. Hence, physiologically based pharmacokinetic (PBPK) modeling of the intestine and liver works a lot better. The approach has been used to appraise the extent of intestine vs. liver removal of drugs [48,49,73–78]. Here, the view is that the intestine is perfused 100% by superior mesenteric arterial flow (QSMA), which drains into the portal venous blood (QPV) for the traditional intestinal model (TM), and, upon combining with QHA, these flows in turn perfuse the liver. However, the TM would not explain route-dependent intestinal metabolism on midazolam [59] and morphine [61], which propelled us to develop useful intestinal flow models that can describe this phenomenon. The segregated flow model (SFM) describes a split flow pattern, as proposed by Klippert and Noordhoek [79], with a lower flow rate perfusing the active, enterocyte region (f<sup>Q</sup> or fraction of the total intestinal flow, <20%) that houses the enzymes and absorptive/efflux transporters, and the remainder flow (>80%) perfusing the non-active, serosal region has since surfaced [80]. With oral administration, the entire dose amount needs to cross into the enterocyte region—the volume of which is conveniently viewed as (fQ´Vint), where Vint (or V<sup>I</sup> ) is the volume of the total intestine—whereas, for intravenous dosing, <20% of the drug in the circulation reaches the enterocyte region, and this will effectively reduce the rate of drug removal by the intestine. The

segregated flow behavior of the intestine is found to explain route-dependent intestinal removal observed for many drugs.

A similar flow model, the QGut model [81–83], was coined as a minimal model based on the well-stirred model equation for the liver, namely, F<sup>I</sup> = QGut QGut+fuBCL<sup>I</sup> int [49], after the equation of Yang et al. [83] was corrected upon substitution of fu<sup>B</sup> for the unbound fraction to intestinal tissue, fu<sup>I</sup> . Since the villous flow (Qvilli) is 6% of the cardiac output as 19 L/h, the ratio of the Qvilli/QPV or f<sup>Q</sup> value for the QGut model is as high as 0.484 for a lipophilic drug such as midazolam [81–83]. Notably, f<sup>Q</sup> is different among these flow models: the SFM (f<sup>Q</sup> < 0.2), QGut model (f<sup>Q</sup> = 0.484) and TM (f<sup>Q</sup> =1). The f<sup>Q</sup> value is expected to affect the extent of intestine and liver removal (E<sup>I</sup> and EH) in the intestine–liver unit with respect to the route of drug administration.

#### **5. Equations for Prediction of Route-Dependent Intestinal Removal**

There are major differences in drug distribution and therefore intestinal drug clearance when the drug is entering from gut lumen into the villous tip or from the circulation (drug given intravenously) (Figure 2). For the TM, whereby the total intestinal flow perfuses the entire intestine (f<sup>Q</sup> = 1), there is no difference in the distribution and clearance of drug between oral and intravenous administration when the enterocyte and serosal regions are meshed together (Figure 2A). After po administration, the drug is absorbed into the enterocyte (yellow arrow) and is well distributed in the enterocyte (right graph); the distribution of drug into the enterocyte is also similar after intravenous administration, and the drug is again well-distributed into the enterocyte (f<sup>Q</sup> = 1). For the SFM (Figure 2B), the extent of distribution after po dosing for a rapidly absorbed drug is similar to that as for TM. Since the enterocyte region is perfused with a lower flow rate (fQ´QPV) according to the SFM, its drug extraction ratio for EI,po,SFM is therefore slightly higher than that for the TM, EI,po,TM, as the drug is associated with a longer transit time in the tissue [18]. However for iv dosing, there is a reduced distribution of drug reaching the enterocyte due to the reduced intestinal flow (f<sup>Q</sup> < 0.2), and there will be a smaller intestinal clearance pursuant to intravenous dosing (Figure 2B). Thus EI,po,SFM > EI,iv,SFM or FI,iv,SFM > FI,po,SFM (Figure 2B) when the drug is shunted away from the enterocyte region, especially for highly permeable drugs entering the intestinal tissue from the circulation than from the gut lumen [18,80].

The explicit solutions for both the TM and SFM (and QGut model) are provided by Sun and Pang [84], who placed the intestine and liver into simple or semi-physiologically based pharmacokinetic (PBPK) models upon viewing both metabolic as well as transport (basolateral influx and efflux) pathways in the intestine and liver (Figure 3). The only difference between the TM and SFM (or QGut model) is the presence of an additional intestinal compartment, since the intestine is now denoted as two subcompartments, the enterocyte and serosa, for the SFM and QGut model. For simplistic assignment of the volume and flow, f<sup>Q</sup> x volume or flow are used to designate the enterocyte volume and flow, respectively, and (1 − fQ) x volume or flow are used to denote the serosal volume and flow, respectively. A common solution ([Equation (1)] now surfaces to represent the systemic bioavailability with oral administration [84]. This common equation may be used to describe bioavailability, Fsys, when f<sup>Q</sup> = 1, 0.484 and <0.2, respectively, for the TM, QGut model, and the SFM.

$$\begin{array}{l} \text{AUC}\_{\text{P}}^{\text{I}\text{O}}/\text{Dose}\_{\text{W}} = \text{F}\_{\text{SYS}} = \text{F}\_{\text{a}} \text{F}\_{\text{I}} \text{F}\_{\text{H}}\\ \text{F}\_{\text{a}} \left[ \frac{\text{t}\_{\text{Q}} \text{Q} \text{v} \text{CL}^{1}\_{\text{d2}}}{\text{t}\_{\text{Q}} \text{Q} \text{v} \text{CL}^{1}\_{\text{d2}} + (\text{f}\_{\text{Q}} \text{Q} \text{v} + \text{fu}\_{\text{B}} \text{CL}^{1}\_{\text{d1}}) [\text{CL}^{1}\_{\text{int,met1}} + \text{CL}^{1}\_{\text{int,met2}} + \text{CL}^{1}\_{\text{int,sc}} (\text{I} - \text{F}\_{\text{s}})]} \right] \frac{\text{Q}\_{\text{I}l} (\text{CL}^{\text{H}}\_{\text{d2}} + \text{CL}^{\text{H}}\_{\text{int,il}})}{\text{Q}\_{\text{I}l} (\text{CL}^{1}\_{\text{d2}} + \text{CL}^{1}\_{\text{int,il}}) + \text{f}\_{\text{B}} \text{Cl}^{\text{H}}\_{\text{int,ft}} \text{I}} \end{array} (1)$$

where CL<sup>I</sup> d1 is the influx transport clearance and CL<sup>I</sup> d2 is the efflux transport clearance. CL<sup>I</sup> int,met is the intestinal intrinsic metabolic clearance (for pathways 1 or 2) and CL<sup>I</sup> int,sec is the secretory intestinal intrinsic clearance. In the liver, the sum of CL<sup>H</sup> int,sec and CL<sup>H</sup> int,met is CL<sup>H</sup> int; fu<sup>B</sup> is the unbound fraction in blood, and QPV and Q<sup>H</sup> are the portal venous flow and total liver blood flow, respectively. The superscripts I and H denote the intestine and liver, respectively. Notably, the unbound fractions of drug in intestine and liver tissue (fu<sup>I</sup> and fuH) are canceled out in the manipulation.

**Figure 2.** Schematic of drug molecules (D) traversing the intestinal membrane and entering the enterocyte for the tradtional model (TM) (**A**) and segregated flow model (SFM) (**B**). After po admininstration, the drug is absorbed into the enterocyte (yellow arrow) and distributed abundantly in the epithelisum (adjacent) for both the TM and SFM. After intravenous administration, the drug is distributed to the same extent in the epithelium according to the TM (f<sup>Q</sup> = 1) while the SFM (f<sup>Q</sup> < 0.2) predicts a much lower distribution of drug in enterocytes. This figure was reproduced with permission from Noh and Pang [18], Wiley, 2019.

**Figure 3.** Physiologically based pharmacokinetic (PBPK) models depicting the intestine as a single tissue or compartment for the TM (left) or as the two subcompartments, the enterocyte and serosal subcompartments for the SFM (right), perfused by segregated flows. This figure was reproduced with permission from Sun and Pang [84], Springer, 2010.

For a drug in the circulation entering the intestine, the rate of drug removal by the enterocyte is fQ.QPV (1 − F<sup>I</sup> )·CA, but there is no removal by the serosal region (Figure 4). The split flow pattern for the SFM or QGut model results in a flow-averaged outflow, portal venous concentration, CPV [49].

$$\overline{\mathbf{C}}\_{\text{PV}} = \frac{\mathbf{f}\_{\text{Q}} \mathbf{Q}\_{\text{PV}} \mathbf{F}\_{\text{I}} \mathbf{C}\_{\text{A}} + (1 - \mathbf{f}\_{\text{Q}}) \mathbf{Q}\_{\text{PV}} \mathbf{C}\_{\text{A}}}{\mathbf{Q}\_{\text{PV}}} = \mathbf{C}\_{\text{A}} [\mathbf{f}\_{\text{Q}} \mathbf{F}\_{\text{I}} + (1 - \mathbf{f}\_{\text{Q}})] \tag{2}$$

CPV

CPV **Figure 4.** Drug removal by the intestine–liver unit: the intestine controls the substrate flux to the liver. The contributions of intestinal (**A**) and liver (**B**) removal are given by Equations (3) and (4). The drug in the circulation enters two subcompartments of the intestine—the enterocyte and serosal compartments. Removal by the enterocyte but not seroal compartment results in a flow-averaged portal venous concentration,CPV. If intestinal removal is high, the contribution by the liver is opposite and will be low. This figure was modified with permission from Pang and Chow [49], ASPET, 2012.

This flow-averaged portal venous concentration is then combined with the arterial concentration (CA) to perfuse the liver. Along the same line of reasoning, the rates of removal of drug by the intestine and liver or the fractional contributions are given by,

$$\frac{\mathbf{v\_I}}{\mathbf{v\_I + v\_H}} = \frac{\mathbf{f\_Q} \mathbf{Q\_{PV}} (1 - \mathbf{F\_I})}{\mathbf{f\_Q} \mathbf{Q\_{PV}} (1 - \mathbf{F\_I}) + \mathbf{E\_H} \left\{ \mathbf{Q\_{PV}} [\mathbf{f\_Q} \mathbf{F\_I} + (1 - \mathbf{f\_Q})] + \mathbf{Q\_{HA}} \right\}} \tag{3}$$

and

$$\frac{\mathbf{v\_{H}}}{\mathbf{v\_{I}}+\mathbf{v\_{H}}}=\frac{\mathbf{E\_{H}}\left\{\mathbf{Q\_{PV}}[\mathbf{f\_{Q}}\mathbf{F\_{I}}+(1-\mathbf{f\_{Q}})]+\mathbf{Q\_{HA}}\right\}}{\mathbf{f\_{Q}}\mathbf{Q\_{PV}}(1-\mathbf{F\_{I}})+\mathbf{E\_{H}}\left\{\mathbf{Q\_{PV}}[\mathbf{f\_{Q}}\mathbf{F\_{I}}+(1-\mathbf{f\_{Q}})]+\mathbf{Q\_{HA}}\right\}}\tag{4}$$

The contributions of the intestine (v<sup>I</sup> ) and liver (vH) in first-pass removal are hence described by Equations (3) and (4). With f<sup>Q</sup> values = 1 (left) (TM), = 0.1 (SFM), or = 0.484 (QGut model) and with the assumption that QPV is approximated by QSMA, simulations show that, for a drug entering the intestine from the circulation, the TM predicts the highest intestinal contribution by the intestine–liver unit, whereas the SFM predicts the least; the QGut model predicts values somewhere in the middle (Figure 4A). The importance of the intestine increases when the liver possesses a low enzymatic removal capacity (high FH). Under the same scenario, results for the %contribution by the liver are the exact opposites, since there is a reciprocal relation to the intestine (Figure 4B). For the SFM, which suggests a lower contribution of metabolism by the intestine for drugs entering from the circulation, the contribution by the liver to first-pass removal is higher than those predicted for the TM and QGut model, since there is a greater substrate flux entering the liver that will result in a greater %contribution by the liver, especially for high E<sup>H</sup> drugs.

#### **6. Is the SFM the Better Intestinal Flow Model Compared to the TM?**

Theoretical development of the SFM readily explains the observed higher E<sup>I</sup> for midazolam and morphine given orally versus intravenously (also Table 1), as do many other drug examples or substrates. When different sets of in vivo or intestinal perfusion data were fitted to the TM versus the SFM, fits to the SFM were all superior over those for the TM. The fitted values of f<sup>Q</sup> were all <0.2, and the SFM was shown to better the TM statistically among all examples (Table 2). The villous flow pattern to the enterocyte region [85], being a low fraction (<0.2), has also been suggested by Granger et al. [86]. A better discrimination between the TM and SFM occurs when metabolite data are present, as provided by the example of morphine, which forms morphine-3-glucuronide (M3G) by the intestine and liver in the rat in vivo. The discriminatory power for the morphine study was further provided by the biliary versus urinary excretion ratio of the metabolite, M3G, which is unable to cross the liver membrane due to its polarity [87]. The M3G presence in bile suggests that the origin of the metabolite is from the liver. The urinary morphine 3-glucuronide originates from both intestinal and liver metabolism, and the observed ratio of M3G in urine/bile associated with intraduodenal morphine dosing was 2.55-fold that with intravenous morphine administration, as predicted for the SFM [76]. The observations for morphine and morphine 3-glucuronide correlated much better with the predictions from the SFM than from TM.


**Table 2.** Fitted values of f<sup>Q</sup> in rodents in vivo and in perfusion preparations.

By contrast, there is practically no difference in the fitted results between the SFM and TM for codeine, the inactive precursor that is *N*-demethylated to form morphine [77]. At first glance, the similarity of both the SFM and TM fits is unique, suggesting that the drug is not subject to intestinal metabolism. For codeine, rat Cyp2d1 (human CYP2D6) is of very low abundance in the intestine, and intestinal metabolism of codeine is very low. For that reason, the agreement of the TM and SFM fits to the codeine data infer a lack of intestine metabolism for codeine. We also recently observed the same pattern for the pan-inhibitor, ketoconazole, after oral and intravenous administration to the rat (unpublished information, Keumhan Noh, Lilly Xu, and K. Sandy Pang).

#### *6.1. Implications on Formation of Intestinal and Liver Metabolites*

Noh and Pang [18] examined the formation of the metabolites: M1 from intestine and M2 from liver, as well as extraction ratios of the intestine with the route of drug administration. For TM, the simulations verified that FI,po,TM = FI,iv,TM for highly permeable drugs, but FI,po,SFM < FI,iv,SFM for SFM and FI,po,SFM < FI,po,TM = FI,iv,TM < FI,iv,SFM. The SFM predicts the highest formation of the M1 metabolite with oral dosing but the lowest formation of M1 with intravenous administration; the converse should occur for M2 formation from liver. From M1/M2, the ratio would further unveil that there is more M2 formation arising via the iv route because of direct delivery of drug via the hepatic artery to the liver. Additionally, M1 is less formed according to the SFM for drugs administered iv than po. For this reason, the ratio M1/M2 would always be smaller after intravenous administration according to the SFM as well as TM (Figure 5).

**Figure 5.** Formation of M1 and M2, specific metabolites formed by the intestine and liver, respecitvely, as simulated by Noh and Pang [18]. The hepatic arterial flow (QHA), normally 25% of total liver blood flow (shown where red line is), delivers the drug directly into the liver, and this contributes M2 formation. Additionally, M2 formation is highest according to the SFM for iv drug administration wherein M1 formation is low due to the low fQ. For the TM, the extent of M2 formation is identical for a drug given orally and intravenously, when there is no QHA flow; the extent increases with increasing QHA. This figure was reproduced with permission from Noh and Pang [18], Wiley, 2019.

#### *6.2. Implications of the SFM on Drug–Drug Interactions (DDIs)*

′ . Another reason for properly selecting the intestine flow model is on the prediction of DDI with an inducer or inhibitor. Because >80% intestinal flow bypasses the enterocytes according to the SFM, the route of administration of the inhibitor/inducer, if oral, should be much more effective than the intravenous route, with the underlying reason that the inhibitor/inducer concentrations would be higher in the enterocyte region. Hence, the extent of DDIs is dependent on how the victim drug or inhibitor/inducer is administered and which intestinal flow model, TM or SFM, prevails (Table 3). For midazolam given intravenously (2 mg) or orally (6 mg) to humans, its AUCiv increased 5-fold, whereas AUCpo increased 16-fold upon pretreatment with 3 po doses of 200 mg ketoconazole orally at 12 h prior to midazolam dosing, and twice at every 12 h thereafter [57]. For digoxin (1 mg), the inducer rifampin (600 mg daily po for 15 days) produced a dramatic lowering of AUCpo but not AUCiv of digoxin due to a 3.5-fold induction of intestinal P-gp protein [20]. In monkeys, ketoconazole inhibited the metabolism of simvastatin, a typical Cyp3a substrate, when given orally and increased the AUCpo 5 to 10x, without changing AUCiv for simvastatin given intravenously [90]. For midazolam, oral treatment (50 mg/kg/day for 4 days) of dexamethasone increased the Vmax values for 1′ -hydroxylation and 4-hydroxylation of midazolam in rat intestinal microsomes much more than that with iv dexamethasone [91]. For digoxin given to Wistar rats, purple grape juice (inhibitor of transporter or enzymes) increased the AUCpo (73%) but not AUCiv for digoxin [92]. There exist many other examples attesting to this interesting DDI pattern for orally but not intravenously administered victim drugs in the presence of inhibitors or inducers, also given orally (Table 3). These examples confirm the observation that inhibitors or inducers of intestinal enzymes act best after oral administration, since the concentration attained will be highest within the intestine, and the same goes for the victim drug. The inhibition expected for the SFM should be the greatest, and hence this would also create opposite changes in liver metabolism, since inhibition of the intestine leads to a greater flux of substrate towards liver metabolism.


**Table 3.** Greater inhibitory or inductive effects after oral administration than iv administration for drug–drug interactions (DDIs) of the intestine.

Noh and Pang [18] recently explored the properties of the SFM and TM models with respect to inhibitors via simulations. Within the assigned, limited parameter space set forth for the drug example, the reduction in M1 formation is highest when both inhibitor (intestine inhibition constant, K<sup>i</sup> = 2 µM) and drug are both given orally, and least or almost unaltered at all when the drug is given intravenously (Figure 6A). Inhibition of metabolism is revealed by the higher drug AUC in the presence of the inhibitor. Often, changes in metabolite patterns are able to reveal inhibition of enzymes within the tissue. For TM, the same extent of M1 formation occurs for both intravenous and oral drug

μ

administration, and inhibition of M1 formation is the same after iv or po drug administration. For SFM, a greater extent of inhibition exists for the drug given orally and least when given intravenously. Liver metabolism is in turn affected upon inhibition of the intestinal metabolism, and an inverse relation to that for the intestine is found.

**Figure 6.** Simulation of intestinally (M1) and hepatically (M2) formed metabolites. For simulation, M1 and M2 were assumed to be inhibited within the intestine only (**A**), and both the intestine and liver (**B**) for a drug example ([18]; data in Table 6 of the reference). The simulation showed that the SFM predicted the highest and lowest M1 formation after oral and intravenous drug admintration, respectively, and the TM predicts a similar extent. The inhibition on intestinal metabolism is the greatest when both the inhibitor and drug are given orally, as predicted by the SFM (**A**). When both intestine and liver metabolism is inhibited, the pattern of change is not readily predictable (**B**). A greater liver inhibition exists after iv drug administration, and the extent of inhibition within the liver can exceed that in the intestine (**B**). This figure was reproduced with permission from data in Table 6 of Noh and Pang [18], Wiley, 2019.

The patterns of intestinal and liver metabolites formed upon inhibition of both the intestine and liver are less revealing as to which tissue is being inhibited, since the proportions of M1 to M2 formed do not always change in the same direction. When inhibition occurring for both the intestine and liver (same K<sup>i</sup> = 2 µM for M1 and M2 formation), the fluctuations for M1 and M2 are small for the TM and SFM for oral drug administration when inhibition of the intestine is highest. Although inhibition is noted for the victim drug, the extent of M1 formation may even increase due to inhibition of liver metabolism to a greater extent for the drug given intravenously due to the higher input with QHA, with inhibition of the liver being more severe than for the intestine (Figure 6B). It is surmised that the extent of change here depends very much on the parameter space and susceptibility of the intestine versus the liver to the inhibitor and route of administration. But a higher AUC of the drug is strong evidence for the presence of the inhibitor on intestinal and liver metabolism.

#### *6.3. Changes in Intestinal and Liver Metabolism with Respect to Flow to Intestine and Liver*

Different flow rates to the enterocyte region in the intestine–liver unit would affect intestinal and liver drug processing differentially. An increase in QPV decreases the EI,po (increased FI,po), allowing for more substrate flow to the liver for both the TM and SFM. With the greater substrate flux but faster transit in the liver, the rate of liver metabolism may remain the same although the increase in liver blood flow increases the CL<sup>H</sup> [47,50]. The converse is also true, with a lower QPV or QSMA, an increase in E<sup>I</sup> and a lower flux to the liver will result.

#### *6.4. Implications of the SFM on IVIVE*

The IVIVE of transporter function is difficult to deduce when different transit times in GIT, gastric emptying rates, varying pH, and microenvironment exist [116]. The permeability, apical absorptive transporters, and split flow pattern of the intestine to the enterocyte and serosal regions, and efflux transporters complicate the IVIVE picture in the prediction of F<sup>a</sup> and F<sup>I</sup> . In terms of IVIVE, Kadono et al. [117] employed permeability measurements in artificial membranes to obtain F<sup>a</sup> from the apparent permeability (Papp) with the parallel artificial membrane permeability assay (PAMPA) and obtain F<sup>a</sup> and F<sup>I</sup> from a scaling factor against a standard such as midazolam using the Yang equation [83]. In addition, IVIVE may be poor for the SFM due to the split flow behavior of the intestinal models, when there is incomplete accessibility of the substrate in circulation to reach enterocytes to fully recruit the intestinal metabolic activity, and this translates to poor IVIVE for the liver. Moreover, methods for identification of intestinal enzymatic activities vary. There are differences in the intestinal functional activity with the mucosal scraping and buffer isolation methods [70,118]. Paine et al. [70] found CYP3A content in each intestinal segment as 30.6, 22.6 and 16.6 pmol/mg mucosal microsomal protein, with similar K<sup>m</sup> towards midazolam but varying Vmax values. von Richter et al. [119] showed that the CYP3A4 in isolated enterocytes (76 pmol/mg homogenate protein corresponded to 210 pmol/mg microsomal protein) and was 3.2-fold higher than that in corresponding liver samples, whereas the P-gp content was 7.2-fold higher in enterocyte homogenate than in liver. The CYP3A4 content from the isolated cell method is higher than that from mucosal scraping. Moreover, intestinal metabolism may occur within cells that are shed into the gut lumen that possess copious metabolic activities in the lumen [118]. Nishimuta et al. [120] employed human intestinal and human microsomes to predict the CYP3A intrinsic metabolic clearance for human intestinal microsomes (HIM) versus human liver microsomes (HLM) (CLint,HIM and CLint,HLM, corrected by the ratio of CLint,HIM to CLint,HLM), and alamethicin-activated HIM for the clearance of UGT substrates. The CYP3A intestinal intrinsic clearance (CLint,I,CYP3A) was highly correlated to hepatic intrinsic clearance (CLint,L,CYP3A), being 2.2-fold higher in liver, although the correlation was poorer for UGTs. Ito and Houston [34] scaled up the CLint,H with an empirical scaling factor (SF) of 6.2 g protein/kg weight to compensate for the extent of underprediction for IVIVE in rats. Allometric scaling shows that in vitro microsomal data consistently underestimate CLint,met,I and CLint,met,H. Hence, scaling and IVIVE remain somewhat empirical approaches.

#### **7. Other Intestinal Models**

Our laboratory has extended the SFM to the segmental, segregated flow model (SSFM) to accommodate transporter and enzyme heterogeneity [121]. However, we have oversimplified the segments as a 1/3 of the total volume, flow and permeability characteristics (Figure 7), even knowing that the surface area, permeability, and lengths of the segments of the digestive tract differed [27]. We found higher abundance P450 activity in the proximal region but higher localization of P-gp in the distal region; this pattern produced the lowest availability in drug absorption (Figure 8). This same trend was confirmed by Watanabe et al. [122] years later in a simulation study. The transporter distributions and functions along the intestinal segments reveal similar transporter and drug metabolizing enzyme distribution patterns along the small intestine for rodents and humans (Table 4). Therefore, the rat may be used to predict drug transport across the small intestine in humans. The same extrapolation, however, is not recommended for drug metabolizing enzymes due to the known species differences observed among animal species [123]. The TM- or SFM-PBPK models have been developed to encompass heterogeneity of transporters and enzymes for improved prediction of PK, including polymorphism and sex differences in enzymes, and tease out contributions of intestine and liver in first-pass metabolism (Table 4). Other factors on the physiology of the GIT may also be considered. It is known that the duodenum is the shortest segment and is approximately 1/5 and 1/7 the lengths of the jejunum and ileum, respectively [28]. As shown by the transport of substrates in segments using chamber or single-pass segmental perfusion, drug permeability, revealed with use of a deconvolution-permeability model, is higher in the jejunum [124,125]. Moreover, the pH and transit times in the duodenum, jejunum and ileum differed [28]. Dressman et al. [3] described, in the continuous absorption model, that the GIT is a continuous tube with varying spatial properties on permeability and solubility and pH, surface area, lengths, diameters, gastric emptying [4], highlighting the importance of gastric emptying time, small intestinal transit time, and effective surface area for absorption [5]. There are other models that accommodate variation in villi surface area, in drug permeability along the intestinal segment. Wu [126] applied the SSFM to examine enterohepatic circulation of glucuronides and found that the processes is affected by segmental distribution of enzymes. With accountability of segmental CYP and P-gp activities, reasonable absorption, efflux, and metabolism are observed for midazolam and compound S [25].

**Figure 7.** An expanded intestinal flow model—the segmental segregated flow PBPK model depicting the intestine as three different segmental regions with segregated flows to the enterocyte and serosal subcompartments. This figure was reproduced with permission from [121], ASPET, 2003.

α **Figure 8.** Heterogeneous distribution of Cyp3a and P-gp in the rat intestine, and changes accompanying the inducer, pregnenolone 16α-carbonitrile (PCN) on intestinal bioavailablity. Both P-gp and Cyp3a relative protein expressions were determined by Western blotting (see referecne 26). The scale on the y-axes of the left panel represents an arbitray scale. Segments 1, 2, 7, and 8 are the duodenal, proximal jejunal, distal jejunal and ileal segments, respectively. The symbols, duo and jej of the left panel denote the duodenum and jejunum, respectively. This figure was reproduced with permission from [26], ASPET, 2006.

α β α β Commercially available softwares on drug absorption include Simcyp® (advanced dissolution absorption metabolism (ADAM) model is implemented in Simcyp®), GastroPlus and GI-Sim [127], and GUT framework [128], which tackle the subject of drug absorption. Although the same input parameters may be used, the software show different F<sup>a</sup> prediction characteristics depending on the rate-limiting steps of oral drug absorption [127]. The advanced compartmental absorption transit model or ACAT model [9], first conceived by Yu and Amidon [2] as the compartment absorption model [1], has evolved to include permeability (in silico properties derived from chemical structure), logP, pKa, particle size and dose. Dissolution that is based on the Nernst–Brunner modification of the Noyes–Whitney equation is implemented. The influx and efflux transporters [129], pH and pKa, and heterogeneous enzyme distribution are recognized as important processes of the software [2,10,11]. Other considerations include the microbiota and composition. It appears that most of these models deal with dosage form and drug properties and may not have considered the segregated flow behavior of the intestine. A suggestion is for these software developers to consider first finalizing their software based on the absorption of a drug solution while incorporating flow and enzyme/transporter heterogeneity, then combining this to another model with the drug and intestine properties (logP, pKa, particle size, pH, surface area) to consider drug absorption.



**Table 4.** *Cont*.

#### **8. Conclusions**

This review has highlighted that metabolite formation and DDIs of the intestine are not well predicted by the traditional intestinal flow model (TM) with respect to the routes of administration of drug and inhibitor. Instead, we recognize the importance of the segregated flow model (SFM) as the premier model to examine intestinal drug metabolism. The evidence in the literature is compelling in support of the SFM based on route-dependent intestinal metabolism. The higher propensity of inhibition with oral and not intravenous dosing is indisputable. Implementation of the SFM is just an additional intestinal compartment away, and this PBPK segregated intestinal flow model (SFM) should be expanded to encompass heterogeneity of transporters and enzymes (SSFM) for improved prediction of PK, including polymorphism and sex differences in enzymes to tease out contributions of intestine and liver in first-pass metabolism. This type of metabolism model could now be coupled with an absorption model to fully investigate the different aspects of Fa, F<sup>I</sup> and FH. We encourage the use of the more "bottom–up" approach in PBPK modeling to provide mechanistic insight into intestinal metabolism/transport [148] by incorporating the SFM into the model. Another improvement could be made is when the QSMA is not assumed to equal QPV. The difference in flow (QPV-QSMA) is due to the venous returns from the coeliac and splenic arteries, and stomach and mesenteries. These venous returns would join that from the small intestine (QSMA) and the hepatic arterial flow to perfuse the liver [149,150].

**Funding:** This research received no external funding.

**Acknowledgments:** We thank Qi Joy Yang for discussion.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


from in vitro data: Reaching a consensus on values of human microsomal protein and hepatocellularity per gram of liver. *Curr. Drug Metab.* **2007**, *8*, 33–45. [CrossRef]


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

*Review*
