**Investigation to Explain Bioequivalence Failure in Pravastatin Immediate-Release Products**

**Alejandro Ruiz-Picazo 1 , Sarin Colón-Useche 1,2,3 , Blanca Perez-Amorós 1 , Marta González-Álvarez 1 , Irene Molina-Martínez 2 , Isabel González-Álvarez 1, \* ,**† **, Alfredo García-Arieta <sup>4</sup> and Marival Bermejo 1**


Received: 13 November 2019; Accepted: 4 December 2019; Published: 9 December 2019

**Abstract:** The purpose of this work is to explore the predictive ability of the biopharmaceutics classification system (BCS) biowaiver based on the dissolution methods for two pravastatin test products, where one of them showed bioequivalence (BE) while the other test failed (non-bioequivalence, or NBE), and to explore the reasons for the BE failure. Experimental solubility and permeability data confirmed that pravastatin is a BCS class III compound. The permeability experiments confirmed that the NBE formulation significantly increased pravastatin permeability, and could explain its higher absorption rate and higher Cmax. This finding highlights the relevance of requiring similar excipients for BCS class III drugs. The BCS-based biowaiver dissolution tests at pH 1.2, 4.5, and 6.8, with the paddle apparatus at 50 rpm in 900 mL media, were not able to detect differences in pravastatin products, although the NBE formulation exhibited a more rapid dissolution at earlier sampling times. Dissolution tests conducted in 500 mL did not achieve complete dissolution, and both formulations were dissimilar because the amount dissolved at 15 min was less than 85%. The difference was less than 10% at pH 1.2 and 4.5, while at pH 6.8 *f* <sup>2</sup>, results reflected the Cmax rank order.

**Keywords:** bioequivalence; Biopharmaceutics Classification System; in vitro; dissolution test; pravastatin

#### **1. Introduction**

The scientific rationale for accepting biopharmaceutics classification system (BCS) biowaivers and in vitro demonstrations of bioequivalence (BE) is based on the assumption that drug permeability and solubility as classification parameters are the limiting factors for drug absorption, and that the excipients contained in the drug products do not affect intestinal permeability or motility (such as gastric emptying or intestinal transit time). For products containing class III (high-solubility, and low-permeability) drugs, which demonstrate very rapid (>85% in 15 min) and similar in vitro dissolutions to that of the reference product at all physiological pHs, it is assumed that test and reference products behave as drug solutions once emptied from the stomach into the duodenum, and

consequently their bioavailabilities in rate and extent must be similar if excipients do not alter the drug absorptions. In vitro BE or biowaivers based on the BCS are now included in the main regulatory guidances around the world, with some slight discrepancies on classification boundaries summarized and discussed by Lenic et al. [1], Zheng et al. [2], and Hoffsäss and Dressman [3].

Two recent reports estimated the probability of proving BE (or the risk of obtaining non-bioequivalent (NBE) or bio-inequivalent (BI) results) for products containing drugs from all BCS classes, and whether the quality control (QC) in vitro dissolution test could predict the in vivo bioequivalence outcome [4,5].

Ramirez et al. published in 2010 a survey of 124 bioequivalence studies of drugs that were classified according to the BCS in order to explore the probability of passing the BE standard for the different BCS classes and the predictive ability of the quality control dissolution test of the BE outcome [5]. In the survey, they found several drug products (including pravastatin) that failed the BE demonstration, in spite of the adequate power of the BE study design (>80%) and the fact that the drug products passed the QC dissolution test. In other words, even if the study had enough power to correctly conclude bioequivalence (H1 alternative hypothesis), the products were found to be inequivalent, and the null hypothesis (H0) of inequivalence could not be rejected [6,7]. The authors concluded that the usually employed QC dissolution tests were not adequate to allow a biowaiver of in vivo bioequivalence studies.

In the Cristofoletti et al. report [4], the authors surveyed a random sample of 500 BE studies from a database from the Brazilian medicines agency (ANVISA). In this study, the drugs were classified according to the BCS and Biopharmaceutical Drug Disposition and Classification System (BDDCS) to evaluate the outcome of bioequivalence studies. For Cristofoletti et al., the failure in pravastatin (a class III compound), in spite of the adequate power of the studies (>80%), remains unexplained.

In both surveys, the probability of obtaining a BE result when the dissolution profiles were similar was around 90% for class I and III drug products (post-test probability or positive predictive value), whereas for class II drug products, the post-test BE probability after a similar dissolution profile was 61%. On the other hand, the probability of false positive results (i.e., similar dissolution but NBE results) was almost 90% for class II drugs. These results point out the lack of an in vivo predictive value of the pharmacopeia dissolution tests [4,5].

The purpose of this work is to explore the predictive ability of BCS biowaiver-based dissolution methods for two test pravastatin products versus the innovator reference product, where one of the test products corresponds to the failing product from Ramirez et al. survey [5], and to explore the reasons for the BE failure. In addition, we have investigated the influence of the product excipients on intestinal permeability to assess the need of additional in vitro tests (apart from dissolution) to ensure the in vitro predictability of the in vivo bioequivalence outcome.

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

#### *2.1. Compounds*

Pravastatin (MW = 446.52 g/mol) was obtained from Sigma-Aldrich (Barcelona, Spain), and Acetonitrile (ACN) was obtained from VWR International (West Chester, PA, USA). Methanol (MeOH), hydrogen chloride (HCl), and trifluoroacetic acid (TFA) were purchased from Fisher Scientific (Pittsburgh, PA, USA). Sodium hydroxide (NaOH), sodium chloride (NaCl), and sodium dihydrogen phosphate monohydrate (NaH2PO4·H2O) were received from Sigma-Aldrich (Barcelona, Spain). Pravastatin is a weak acid with a p*K*<sup>a</sup> = 4.21 and log *P* = 1.65 [8]. For oral administration, it is used in the form of sodium salt. Pravastatin sodium is a white hygroscopic powder, easily soluble in water and methanol, and acetonitrile, and practically insoluble in chloroform [9].

Metoprolol, *n*-octanol, acetonitrile, and methanol were purchased from Sigma (Barcelona, Spain).

#### *2.2. Pravastatin Formulations*

Lipemol 40 mg tablets (Bristol-Myers, Squibb, S.A., London, UK) were used as reference products. Their excipients are croscarmellose sodium, magnesium stearate, magnesium oxide, microcrystalline cellulose, yellow iron oxide, anhydrous lactose, and povidone K30.

Pravastatin bioequivalent (BE) and non-bioequivalent (NBE) formulations were donated by a Spanish pharmaceutical company. The excipients in the NBE formulations are croscarmellose sodium, magnesium stearate, microcrystalline cellulose, yellow iron oxide, colloidal silica, magnesium carbonate, and anhydrous lactose.

In the BE formulation, the excipients are magnesium stearate, microcrystalline cellulose, yellow iron oxide, povidone K30, calcium phosphate monobasic anhydrous, sodium starch glycolate, trometamol, and sodium phosphate dibasic dehydrate.

The company provided samples from the batches used in the BE study for tests and reference products.

The results obtained in their corresponding 2 × 2 crossover BE studies are reported in Table 1. The NBE formulation failed to show bioequivalence in Cmax.


#### **Table 1.** In vivo bioequivalence results of the test formulations.

#### *2.3. Experimental Techniques*

#### 2.3.1. Solubility Assays: Saturation Shake-Flask Method

To estimate pravastatin solubility, an excess of solid drug was added in buffer solutions pH 1.2, 4.5, and 6.8 at 37 ◦C. The solubility assays were performed according to the World Health Organization (WHO) guidelines protocols [10]. The equilibrium was reached in 8 h. A sample concentration was determined using high-performance liquid chromatography (HPLC) with ultraviolet (UV) detection.

#### 2.3.2. Lipophilicity Indexes: Partition Coefficients

Bulk phase partition coefficients (P) between *n*-octanol (analytical grade, Sigma-Aldrich, Barcelona, Spain) and phosphate buffer pH 6.80, 50 mM were determined for metoprolol and pravastatin.

The partition coefficient was calculated as the ratio between octanol concentration and the aqueous concentration. Three replicates were done to determine the average value.

Partition coefficients can be used as an index to provisionally classify compounds in terms of permeability [11]. Metoprolol was chosen as the reference compound for high permeability because its oral fraction absorbed is higher than 95%. Thus, drugs that exhibit partition coefficients and human intestinal permeability values lower than the value for metoprolol are considered low-permeability drugs.

#### 2.3.3. Permeability Assays: Cell Culture and Transport Studies

Caco-2 cells were grown in a polycarbonate membrane. To reach the confluence, 250,000 cells/cm<sup>2</sup> were seeded in six Transwell plates and fasted for 19–22 days with Dubelcco's Modified Eagle's Media, with 1% l-glutamine, 10% fetal bovine serum, and 1% penicillin/streptomycin, at 37 ◦C temperature, 90% relative humidity, and 5% CO2.

The confluence of the cells was tested by measuring the transepithelial electrical resistance (TEER). Permeability studies were conducted with an orbital shaking (50 rpm) and with pH 7.0 in both chambers. Standard protocols were described and validated previously in our laboratory [12–16]. Four samples of 200 µL were taken and replaced with a fresh buffer from the receiver side at 15, 30, 45, and 90 min.

Pravastatin transport was studied in the solution at five concentrations (50, 100, 358 (highest single dose per tablet), 500, and 1000 µM). Permeability studies were performed in both directions: apical-to-basal (A-to-B) and basal-to-apical (B-to-A). The permeability value of pravastatin was compared with metoprolol, the high permeability reference compound.

The apparent permeability coefficient was calculated according the following equation:

$$\mathcal{C}\_{\text{rezrier}\,t} = \frac{Q\_{\text{total}}}{V\_{\text{rezrier}} + V\_{\text{donor}}} + \left( (\mathcal{C}\_{\text{rezrier}\,t-1} \cdot f) - \frac{Q\_{\text{total}}}{V\_{\text{rezrier}} + V\_{\text{donor}}} \right) e^{-\mathbf{P}\_{\text{eff}\,\mathbf{0},\mathbf{1}} \cdot \mathbf{S} \cdot \left( \frac{1}{V\_{\text{carrier}}} + \frac{1}{V\_{\text{donor}}} \right) \cdot \mathbf{A} t} \tag{1}$$

where *Creceiver,t* is the drug concentration in the receiver chamber at time *t*, *Qtotal* is the total amount of drug in both chambers, *Vreceiver* and *Vdonor* are the volumes of each chamber, *Creceiver*,*t*−<sup>1</sup> is the drug concentration in the receiver chamber at the previous time, *f* is the sample replacement dilution factor, *S* is the surface area of the monolayer, ∆*t* is the time interval, and *Pe*ff is the permeability coefficient as was described by Mangas-Sanjuan et al. [14].

The permeability value of pravastatin 358 µM (highest single dose per tablet) was compared with the permeability value of the reference and test formulations at the same concentration of pravastatin. Experiments in the presence of the formulation excipients were done by dissolving a formulation tablet in 250 mL of buffer and filtrating the obtained dispersion to eliminate nonsoluble excipients.

#### 2.3.4. Disintegration

These assays were performed using a tablet disintegration tester (Hanson Research, Chatsworth, CA., USA) to measure the tablet disintegration time. According to the United States Pharmacopeia (USP) and European Pharmacopeia (Ph. Eur.) guidelines, the experiments were carried out in 800 mL of the media at 37 ◦C (*n* = 3).

The disintegration studies were performed in different media, simulating pH in the gastrointestinal human tract at pH 1.2, 4.5, and 6.8, and unbuffered water.

#### 2.3.5. Dissolution Assays

Drug release experiments were performed in 900 mL and 500 mL of pharmacopeia media (hydrochloric acid buffer/acetate buffer/phosphate buffer) at pH 1.2, 4.5, and 6.8, respectively, at 37 ± 0.5 ◦C and 50 rpm with USP 2 (Pharma-Test PT-DT70) [17]. Samples were taken at 5, 10, 15, 20, 30, 45, and 60 min, and the sample volume was replaced by a fresh preheated medium. Samples were immediately centrifuged (at 10,000 rpm for 10 min) and diluted (1:1) in methanol.

The dissolution profiles were compared by *f* <sup>2</sup> (similarity factor) [18,19]. The *f* <sup>2</sup> calculations were performed in Microsoft Excel (2016) [20].

#### *2.4. Analysis of the Samples*

The samples were analyzed by HPLC (Alliance-Waters 2695, Barcelona, Spain) using a Nova-Pak C18 column (4 µM, 3.9 × 150 mm) and UV detector (Waters 2487, Barcelona, Spain) at wavelengths of 238 nm. The flow-rate was 1.0 mL/min, and the mobile phase contained 50:40:10 methanol, water with 0.1% trifluoroacetic, and acetonitrile. The retention time of pravastatin was 3.2 min, and the limit of quantification was 0.02 µM. The analysis method fulfilled the linearity (*r* > 0.99), accuracy, and precision criteria (<5%).

The metoprolol HPLC method was published previously by our group [13,21].

#### *2.5. Statistical Analysis*

Results are shown as mean ± standard deviations. Statistical analysis of permeability values were two-tailed student t-tests, or analysis of variance (ANOVA) and Scheffé post hoc. The significance level was 0.05, and the software used was statistical package SPSS, V.20.00.

#### **3. Results**

The solubility pH profile for pravastatin at pH 1.2, 4.5, and 6.8 was 439.80 ± 17.42, 503.87 ± 24.20, and 479.58 ± 17.39 mg/mL (Figure 1). As expected considering its acidic nature, the lowest solubility was obtained at the lowest pH. The highest strength/dose of pravastatin (40 mg) would be soluble in 0.09 mL of water at pH 1.2. Dose number (Do) is 3.20 <sup>×</sup> <sup>10</sup>−<sup>4</sup> . Therefore, pravastatin is a highly soluble drug. ANOVA test and Scheffé post hoc comparison detected differences in the solubility values at pH 1.2 versus the higher pHs, while no differences were detected between solubility values at pH 4.5 versus 6.8. − −

**Figure 1.** The pH solubility profile of pravastatin determined by the shake-flask method.

The *n*-Octanol partition coefficient was *P* = 0.70 ± 0.07 for pravastatin. This value was obtained with relation to a 65:35 *n*-octanol/phosphate buffer, with a pH 6.80, 50 mM. For metoprolol, the partition coefficient was *P* = 0.23 ± 0.05. According to these results, pravastatin could be provisionally classified as a high-permeability compound, but as Takagi, et al. [11] described previously, the existence of an active transport mechanism (absorptive or secretive) would bias the classification based purely on lipophilicity.

The permeability value of pravastatin was compared with a metoprolol (reference compound) value in order to classify pravastatin as a high- or low-permeability compound. Different concentrations of pravastatin were studied in order to characterize the transport mechanism of the drug across the intestinal membrane (Figure 2).

**Figure 2.** In vitro permeability values of pravastatin in Caco-2 cell monolayers at different drug concentrations. The highest clinical dose of pravastatin is equal to 358 µM (358 µM = 0.152 mg/mL, 100 µM = 0.04 mg/mL, 500 µM = 0.21 mg/mL, and 1000 µM = 0.42 mg/mL).

The analysis of variance (ANOVA) detected statistically significant differences among the permeability values obtained at different pravastatin concentrations (*p* = 0.0003).

Permeability values in the different formulations of pravastatin were also compared. The results are shown in Figure 3. The ANOVA and post hoc test showed statistically significant differences between the reference and NBE formulation (*p* < 0.05).

**Figure 3.** Pravastatin permeability values of the application programming interface (API) (358 µM), reference, non-bioequivalent (NBE), and bioequivalent (BE) formulations. Error bars represent the standard deviation.

In Figure 4, disintegration times of pravastatin products are depicted with statistical differences between formulations. The disintegration endpoint of each sample is recorded as per the United States Pharmacopeia (USP) definition, in which no palpable form or outline of the sample is observed on the screen of the test apparatus or adhering to the lower surface of the disk.

**Figure 4.** Disintegration times of various pravastatin products in different disintegration media. \* Significant difference (*p* < 0.05).

Results of the dissolution tests in different buffers and volumes are summarized in Figure 5.

**Figure 5.** Dissolution profiles of three pravastatin formulations (reference product, non-bioequivalent formulation (NBE) and bioequivalent formulation (BE)) in the paddle apparatus at 50 rpm in buffered media at pH 1.2, 4.5, and 6.8 (*n* = 6).

The reference and NBE products showed complete dissolution (>85%) in 15 min in the USP 2 at 50 rpm in 900 mL of buffered media at pH 1.2, 4.5, and 6.8. However, the BE product showed complete dissolution at 30 min. The same tests were conducted at 500 mL to verify whether a volume lower than 900 mL could detect differences in BCS class III formulations. However, the amount dissolved in these reduced volumes did not reach complete dissolution. Therefore, these dissolution tests are difficult to interpret.

#### **4. Discussion**

− Based on the experimental results of solubility and permeability determinations, pravastatin is classified as a class III drug according with BCS, as its Dose number (Do) is less than 1 (≤3.20 <sup>×</sup> <sup>10</sup>−<sup>4</sup> ) at all the relevant pHs, and the permeability value of pravastatin was lower than that of metoprolol. Pravastatin is classified as a low-permeability drug, which is consistent with an oral bioavailability of 17% due to a low intestinal absorption (34% of dose administered) and a first-pass hepatic effect of 66% of the absorbed drug [22]. On the other hand, our experimental results are consistent with the existence of a secretion mechanism, as permeability increased at higher concentrations. The permeability value obtained in the presence of sodium azide that corresponded with the passive diffusion permeability confirms that not even at the highest pravastatin concentration tested was the efflux mechanism saturated. In addition, the permeability obtained in the presence of sodium azide, when any potential transporter contribution was nullified, was as high as metoprolol permeability, which is consistent with the experimentally estimated partition coefficient.

The observed changes in pravastatin permeability values in different formulations demonstrated that, in spite of the assumption of excipients being inert, actually some of them affect drug bioavailability, in particular for drugs with a carrier-mediated transport mechanism like pravastatin. Many excipients have shown its ability to inhibit the secretion activity of P-glycoprotein or MRP-2 transporters, increasing the permeability of the substrate drug [23–26]. Other authors have demonstrated that changes in paracellular route permeability are also affected by the presence of excipients [27–29]. The formulation with the largest permeability is the NBE formulation, which is consistent with the failed Cmax due to supra-bioavailability.

Disintegration differences in pravastatin products were in line with the in vivo outcome for the NBE formulation, but there was no rank order correlation in the cases of the reference and the BE formulation, which showed marked differences in disintegration times that are not reflected in vivo. This might be attributed to the differences in permeability that are compensated for by the differences in disintegration/dissolution, since the BE formulation also exhibited higher permeability than the reference formulation, but this difference was compensated for by its slower disintegration/dissolution. In 500 mL of pH 6.8 buffer, the dissolution of the formulations showed the same trend as the observed in the in vivo results, and the *f* <sup>2</sup> similarity factor indicated differences between the reference and NBE formulation. This may be because these dissolution conditions are predictive or simply a coincidence, since in these conditions complete dissolution was not achieved and the results are difficult to rely on.

Dissolution assays in the paddle apparatus at 50 rpm in 900 mL cannot detect differences in dissolution profiles of the reference and NBE formulations using classical buffers at pH 1.2, 4.5, and 6.8, because complete dissolution (>85%) was achieved in 15 min. However, in buffers of pH 1.2 and 6.8, it is evident that the dissolution was more rapid for the NBE formulation. This small difference, together with the large difference in permeability, seems to be the cause of the Cmax failure, which was borderline, but the point estimate of the ratio test/reference was slightly >10% different, and the confidence interval did not include the 100% value. This difference in permeability highlights the relevance of requiring similar excipients for BCS class III drugs, as the high solubility could allow a similar in vitro dissolution, and even a similar in vivo dissolution, while still other factors may affect their oral fraction from being absorbed. For the BE product, the BCS-based biowaiver dissolution test in 900 mL would have led to a false negative result [3], i.e., differences observed in vitro while BE was observed in vivo. This possibility is not a problem from a regulatory point of view, since the companies always have the possibility to conduct an in vivo BE study, whereas the regulatory problem is to approve the NBE formulation based on in vitro dissolution profiles when the Cmax is notably different and not able to show equivalence.

On the other hand, our dissolution results show that for these pravastatin products reduced volumes (500 mL) seem to be more discriminatory, i.e., both products are detected as nonsimilar at pH 1.2 and 4.5, and differences parallel or follow the same rank order correlation of the in vivo outcome at pH 6.8. However, these dissolution studies are difficult to interpret because complete dissolution is not reached and, therefore, the *f* <sup>2</sup> similarity factor lacks any meaning. In fact, at pH 1.2 and 4.5, the amounts dissolved at 15 min, which is an approximation of the median gastric emptying time in the reference and the NBE formulations, can be considered similar. Therefore, with the simplistic assumption of gastric emptying after 15 min of residence in the stomach, the same amount would be released to the duodenum. The amounts dissolved at 15 min at pH 6.8 are different, but that pH is not the expected pH when the drug is emptied from the stomach. Therefore, if the Cmax differences are due not only to the differences in permeability but also to differences in dissolution, it is evident that the usual volume of 900 mL is not adequate. Dissolution tests in 500 mL of buffer at 50 rpm were conducted to explore if volumes closer to the fluid volume in the gastrointestinal tract offer different results. However, the predictive power at pH 6.8 was not confirmed. For example, at pH 1.2 and 6.8, the *f* <sup>2</sup> similarity factor was not able to conclude similarity because dissolution did not reach 100% at the end of the study but an asymptote. However, in relative terms, the NBE formulation and the reference exhibit similar amounts dissolved at 15 min (less than 10%). At pH 4.5, even the *f* <sup>2</sup> similarity factors concludes with the NBE formulation and the reference.

#### **5. Conclusions**

The BCS-based biowaiver dissolution tests with the paddle apparatus at 50 rpm in pH 1.2, 4.5, and 6.8 in 900 mL media were not able to detect the in vivo Cmax differences for pravastatin products. The different Cmax seems to be the result of the combined effect of a higher permeability of the NBE formulation due to the excipients inhibition of the efflux transporters and a more rapid disintegration and dissolution. Those combined effects could not be detected with the current dissolution conditions

in a volume of 900 mL and the criteria of similarity at 15 min, but the difference of the NBE formulation was observed at earlier sampling times (e.g., 5 min) and/or when the dissolution tests were conducted in 500 mL of dissolution media. Nevertheless, at 500 mL volume and pH 1.2 and 4.5, the BE formulation was also detected as nonsimilar.

**Author Contributions:** Methodology, validation, and investigation, A.R.-P., S.C.-U. and B.P.-A.; formal analysis, M.G.-A. and I.M.-M.; conceptualization, writing—review and editing, A.G.-A.; writing—original draft preparation, visualization, supervision, I.G.-A.; project administration and funding acquisition M.B.

**Funding:** This research was funded by Agencia Estatal Investigación and European Union through FEDER (Fondo Europeo de Desarrollo Regional) by the project "Modelos in vitro de evaluacion biofarmaceutica", grant number SAF2016-78756(AEI/FEDER, EU). The author S.C.-U.APC was funded by DAP, University of Los Andes, Venezuela.

**Conflicts of Interest:** The authors declare no conflict of interest. Spanish Agency for Medicines and Health Care Products had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript; and in the decision to publish the results. This article represents the personal opinions of the authors, and does not necessarily represent the views or policies of their corresponding Regulatory Authorities.

#### **References**


© 2019 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*

### **A Mechanistic Physiologically-Based Biopharmaceutics Modeling (PBBM) Approach to Assess the In Vivo Performance of an Orally Administered Drug Product: From IVIVC to IVIVP**


Received: 14 December 2019; Accepted: 15 January 2020; Published: 17 January 2020

**Abstract:** The application of in silico modeling to predict the in vivo outcome of an oral drug product is gaining a lot of interest. Fully relying on these models as a surrogate tool requires continuous optimization and validation. To do so, intraluminal and systemic data are desirable to judge the predicted outcomes. The aim of this study was to predict the systemic concentrations of ibuprofen after oral administration of an 800 mg immediate-release (IR) tablet to healthy subjects in fasted-state conditions. A mechanistic oral absorption model coupled with a two-compartmental pharmacokinetic (PK) model was built in Phoenix WinNonlinWinNonlin ® software and in the GastroPlus™ simulator. It should be noted that all simulations were performed in an ideal framework as we were in possession of a plethora of in vivo data (e.g., motility, pH, luminal and systemic concentrations) in order to evaluate and optimize these models. All this work refers to the fact that important, yet crucial, gastrointestinal (GI) variables should be integrated into biopredictive dissolution testing (low buffer capacity media, considering phosphate versus bicarbonate buffer, hydrodynamics) to account for a valuable input for physiologically-based pharmacokinetic (PBPK) platform programs. While simulations can be performed and mechanistic insights can be gained from such simulations from current software, we need to move from correlations to predictions (IVIVC → IVIVP) and, moreover, we need to further determine the dynamics of the GI variables controlling the dosage form transit, disintegration, dissolution, absorption and metabolism along the human GI tract. Establishing the link between biopredictive in vitro dissolution testing and mechanistic oral absorption modeling (i.e., physiologically-based biopharmaceutics modeling (PBBM)) creates an opportunity to potentially request biowaivers in the near future for orally administered drug products, regardless of its classification according to the Biopharmaceutics Classification System (BCS).

**Keywords:** oral absorption; in silico modeling; GastroPlus; Phoenix WinNonlin; pharmacokinetics; clinical studies; ibuprofen; manometry; gastrointestinal; mechanistic modeling; PBPK; PBBM

#### **1. Introduction**

Although advances have been made and insights have improved throughout the years, there is still a lot of gastrointestinal (GI) variables that are poorly understood that should be investigated for their influence on drug release and systemic exposure after oral intake of a drug product [1,2]. The knowledge has improved about the intestinal behavior of an active pharmaceutical ingredient (API) in terms of solubility, dissolution, permeation, supersaturation, and precipitation, as demonstrated in different clinical aspiration studies performed over the last ten years [3–7]. In these studies, drug concentrations were measured in healthy volunteers after aspiration of GI fluids after oral administration. Subsequently, drug concentrations were determined in these aspirates in parallel with collecting blood samples to assess systemic exposure. As these studies contributed to formulation behavior in the GI tract, it was not always straightforward to correlate the measured drug concentrations in the upper part of the small intestine with concentrations appearing in blood. The knowledge about the impact of the surrounding dynamic GI environment on drug- and formulation behavior remains rather scarce and requires further investigation. The constantly changing climate of GI pH and motility patterns can alter drug behavior along the GI tract in such a way that it is necessary to investigate these mechanisms and, in a next step, to take these variables into account in in vitro and in silico predictive models to facilitate oral drug development [8–11]. GI motility is defined by the different contractile phases of the migrating motor complex (MMC): phase I is an inert period with little activity; phase II features sporadic contractions gradually ascending in magnitude; and phase III is characterized by powerful, high-frequency contractile bursts that promote emptying of contents where peak flow rates are observed [12]. In recent work, a clinical aspiration study was performed that aimed to measure the impact of physiological variables on the systemic exposure of orally-administered ibuprofen (immediate-release tablets, 800 mg) [13,14]. The outcome of this study demonstrated how phase III contractions and fluctuating pH (caused by the low buffer capacity) in the human intestinal tract had a major impact on ibuprofen's dissolution and, consequently, absorption in fasted (*n* = 20) and fed state (*n* = 17).

Based on these new insights, it has become clear that working in a biorelevant setting (i.e., simulated GI media, multi-compartmental in vitro models, solubility/permeability interplay) will result in more accurate predictions. From that perspective, the OrBiTo community took the initiative to design a decision tree which makes it handy for formulation scientists to select the most appropriate biopredictive dissolution test depending on the biopharmaceutical properties of the drug compound and the type of formulation [15]. This decision tree clearly focuses on some biorelevant aspects of the GI tract that play a pivotal role in and have a significant impact on the luminal behavior of a drug product; these variables should not be neglected in a biopredictive dissolution test. For instance, in the case of weakly basic compounds, the implementation of a GI transfer should be included in order to capture the supersaturated state of the drug after transfer from the stomach compartment to the intestinal compartment. Besides the optimization of in vitro tools, mechanism-based in silico models should be optimized and validated at the same time. The outcome of the in vitro dissolution tests can serve, in a second step, as input for physiologically-based pharmacokinetic (PBPK) platforms to simulate the systemic exposure of the drug. While a lot of progress has been made by mechanism-based in silico models to identify key issues in the development of new oral drug products [11,16–19], there are still many aspects that are poorly understood that need to be optimized/integrated to maximize the utility of these models towards predicting the systemic outcome of novel and generic drug candidates. Commercially available software packages such as the Simcyp® simulator, GastroPlus™, and PK-Sim® are just a few programs that are frequently used in the non-clinical stage of drug product development to get an idea about the in vivo performance of the drug product when administered to patients. The underlying syntax/algorithm of these packages describes the mass transport of the drug throughout the different built-in compartments and should be adequately reflecting the physiological processes

of the human body. From an academic perspective, it is our mission to see (i) if the underlying mathematical equations are making any sense and (ii) if they are representing the physiological variables in a proper and biorelevant manner (physiological range). For instance, all these programs describe the stomach compartment as a single, well-stirred compartment, assuming that a drug will be homogeneously distributed along the entire stomach after oral administration. Based on measured gastric concentrations of the non-absorbable markers, phenol red and paromomycin, it was clearly shown that these markers were not homogeneously distributed among the different regions of the stomach (i.e., fundus, body and antrum). Therefore, we developed a mechanistic oral absorption model in the Berkeley-Madonna® software package (Version 8.3.18) that could only explain the observed luminal data when the stomach was handled as a two-compartmental model that was connected with a bypass flow to reflect the immediate fast transfer of liquid from stomach to small intestine after drinking a solution of these markers [20].

For this study, we aimed to reflect the luminal and systemic concentrations of ibuprofen under fasting state conditions starting with the simplest model, assuming a first-order kinetic process for dissolution, gastric emptying and absorption. In a second step, the model was revised, and dissolution was handled as pH dependent and gastric emptying was treated as a first-order process until the time of appearance of phase 3 contractions post-dose—after which, the remaining dose was directly transferred to the duodenal compartment. The mechanistic model focused on the integration of phase III contractions to simulate a house-keeper wave that is responsible for the direct release of ibuprofen particles from the stomach into the small intestine. In the different compartments of the small intestine, the dissolution of ibuprofen is driven by the regional pH, determining the fraction dissolved and undissolved. Afterward, a statistical analysis was performed to see how both scenarios matched with the observed luminal and systemic concentrations. In addition to this model, an advanced compartmental absorption and transit (ACAT™) model was developed in GastroPlus™ to assess the impact of dynamic pH, fluid volumes and gastric emptying on the systemic performance of ibuprofen. A comparison of these simulations was made with simulations performed by default settings.

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

#### *2.1. Reference Intraluminal and Systemic Data of Ibuprofen*

#### 2.1.1. Intraluminal and Systemic Profiling of Ibuprofen in Healthy Volunteers

The study was held at the University of Michigan Hospital after receiving approval by the internal review board (IRB) at both University of Michigan and FDA (HUM00085066) under the project (HHSF223201310144C (Sun D. and Amidon G.L., Principle Investigators)—09/30/15–12/31/18 "Modernization of in vivo-in vitro Oral Bioperformance Prediction and Assessment: A research study to evaluate the performance of an ibuprofen oral dosage form in the gastrointestinal tract of healthy adult volunteers") [13,14]. Briefly, 13 healthy volunteers (men and women) were recruited; 7 out of 13 subjects participated in the study twice to generate intra-subject variability data. All volunteers provided written informed consent to participate in this study. After a fasting period, a multi-lumen GI tube from MUI Scientific (Mississauga, ON, Canada) was introduced via the mouth to the small intestine. Abdominal fluoroscopy was performed to ensure the GI tube was properly positioned in the different regions of the GI tract (i.e., stomach, duodenum, proximal and distal jejunum). The subject was asked to remain in bed while the GI tube was equilibrated by performing a baseline GI motility test for approximately 3–5 h (Medical Measurement Systems (MMS), Williston, VT, USA). Prior to the administration of the ibuprofen tablet, an intravenous catheter was introduced in the antecubital area of the subject for blood collection. The catheter was kept open with a heparin and saline solution. The subjects were asked to empty his/her bladder prior to the start of the study. At approximately 4:00 AM, the subject was given a single oral dose of ibuprofen (800 mg tablet). The study drug was administered with 250 mL of water containing USP grade phenol red (0.1 mg/mL). The actual amount of water consumed was measured and recorded. Volunteers were not obliged to drink the total amount of administered water to avoid any feeling of nausea at the start of the study. GI samples (stomach, duodenum and jejunum) were collected at 0, 0.25, 0.5, 0.75, 1, 1.5, 2, 2.5, 3, 4, 5, 6, and 7 h. Blood samples (4 mL/time point) were collected at 0, 0.167, 0.33, 0.5, 0.75, 1, 1.5, 2, 2.5, 3, 4, 5, 6, 7, 8, 12 and 28 h. Plasma was separated from blood samples by centrifugation and stored at −80 ◦C until analysis. The pH of GI fluid samples was immediately measured and recorded. The GI fluid samples were centrifuged at a speed of 17,000× *g* for 10 min and the supernatant was placed in the new tube for drug concentration analysis.

#### 2.1.2. Recording of Post-Dose Phase III Contractions

MMC phase III motility periods were identified from the water-perfused manometric measurements using spectral density estimation and penalized logistic regression as described in detail by Hens and co-workers and will be briefly discussed here [13]. After positioning, the catheter was connected to a computer console that generated real-time manometry recordings in the different segments of the GI tract (Medical Measurement Systems, Dover, NH, USA). The manometric channels attached to the catheter were perfused with water at a rate of 2 mL/min and served as intestinal pressure recording ports to assess intestinal motility. Each segment contained four motility channels to monitor pressure events. Baseline intestinal motility was evaluated for 3–5 h prior to study drug administration of the tablet. Subsequently, GI motility was measured continuously for 7 h. Powerful antral phase III contractions were defined as the occurrence of regular 2–3 contractions per minute for at least 2 min with an average amplitude of 75 mmHg. Duodenal phase III contractions were characterized by 11–12 contractions per minute with an average amplitude of 33 mmHg which can last for at least 3 min. As the contractile activity propagates, it becomes less spatiotemporally organized resulting in slower propulsion rates in the distal small bowel. The corresponding spectral density estimate of a phase III period will have high energy levels in the 10–12 cycles/min components, leading to a concentrated spectrum. During non-phase III motility, the spectral density will have a more diffuse spectrum. Using penalized logistic regression, it was clearly observed that the proportion of energy in the 9–12 cycle/min frequencies is an important predictor of phase III motility.

2.1.3. Thermodynamic Equilibrium Solubility of Ibuprofen in Fasted-State Human Gastric and Intestinal Fluids (Fahgf/Fahif)

The thermodynamic solubility of ibuprofen was determined by the shake-flask method (25 RPM), incubating gastrointestinal fluids for 24 h with an excess amount of ibuprofen (Acros Organics, Morris Plains, NJ, USA) at 37 ◦C. The fluids that were used for measuring the thermodynamic solubility of ibuprofen were aspirated gastric, duodenal and jejunal fluids of three different time points of subject B005-F2. Following the 24 h incubation, samples were centrifuged for 15 min at 17,000 *g* (AccuSpin Micro 17, Fisher Scientific, Pittsburgh, PA, USA). The supernatant was diluted 10-fold with methanol (Fisher Scientific, Pittsburgh, PA, USA) and again centrifuged for 5 min in order to discard any proteins that could interfere with the HPLC analysis (see below). Solubility measurements were performed in triplicate.

#### 2.1.4. Bioanalysis of Ibuprofen by HPLC

Solubility samples were analyzed by HPLC–UV (Hewlett Packard series 1100 HPLC Pump, Santa Clara, CA, USA), combined with Agilent Technologies 1200 Series Autosampler (Santa Clara, CA, USA). A volume of 5 µL was injected into the HPLC system connected to a UV lamp that was able to detect ibuprofen at a wavelength of 220 nm (Agilent 1100 Series UV lamp, Santa Clara, CA, USA). An isocratic run containing 70% acetonitrile (VWR International, West Chester, PA, USA) and 30% purified water (both containing 0.1% TFA) was used to detect ibuprofen at a retention time of 2.9 min using a reversed-phase C-18 column (Eclipse Plus C18, 4.6 × 150 mm, 5.5 µm, Agilent Technologies) and a 1 mL/min flow rate. The calibration curve was made in methanol based on a stock solution of ibuprofen in methanol (1 mM). Linearity was observed between 10.32 µg/mL and 0.32 µg/mL. The observed peaks were integrated using ChemStation software (Agilent Technologies, B.04.03 version). The developed analytical method met the FDA requirements for bioanalytical method validation [21].

#### *2.2. Mechanistic Oral Absorption Modeling in Phoenix WinNonlin*®

WinNonlin User-Customized Mechanistic Model to Stress the Pivotal Underlying GI Variables: InVivo\_GIS versus InVivo\_GISPlus

A compartmental model including stomach, duodenum and jejunum with first-order transit and absorption rates was designed to describe the time evolution of ibuprofen mass and concentrations in duodenum, jejunum and plasma. The following assumptions were made with respect to the 'in vivo Gastrointestinal System model (InVivo\_GIS)':


This basic model was further extended to explore the influence of intestinal pH and motility. Therefore, we developed a model which we will refer to as the 'InVivo\_GISPlus':


Model schemes are represented in Figure 1. The system of differential equations was written as the American Standard Code for Information Interchange (ASCII) code and run in Phoenix WinNonlin® V8 (Certara, Princeton, NJ, USA).

Six differential equations were used, describing the amount of solid ibuprofen as a function of time in the stomach (*Msolid stomach*), in the duodenum (*Msolid duodenum*), and in the jejunum (*Msolid jejunum*), as well as the amount of dissolved ibuprofen as a function of time in the duodenum (*Mdissolved duodenum*), in the jejunum (*Mdissolved jejunum*) and in plasma (*Mplasma*). Integrated model parameters were:


In each subject, the elimination rate coefficient (*K*el) and the distribution volume in plasma (V3) were fixed to the value obtained after performing a non-compartmental PK analysis. Time to the

next phase III wave post-dose (TMMC) was fixed to the experimentally determined value [22]. The individual pH values in duodenum and jejunum at each time point were used in InVivo\_GISPlus model to recalculate ibuprofen's solubility at each time point. The absorption rate coefficient was fixed to a high value of 12 h−<sup>1</sup> based knowing that ibuprofen does not have any permeability-related issues (fraction absorbed ~1) [22]. This value was based on ibuprofen permeability in rat small intestine [23] that was scaled up to human *P*eff value with the human-rat correlation described by Zakeri-Milani et al. [24]. The 13 mathematical equations to describe the mass transport of ibuprofen in the InVivo\_GISPlus model are summarized in the Supplemental Information. Both models InVivo\_GIS and InVivo\_GISplus were fitted simultaneously to duodenal, jejunal and plasma concentrations.

**Figure 1.** Mass transport analysis scheme and assumptions of InVivo\_GIS and InVivo\_GISPLUS models. *K*empt: first-order emptying rate coefficient; *K*a: first-order absorption rate coefficient; K\_TD and K\_TJ: first order transit coefficients from duodenum to jejunum and from jejunum to distal segments, respectively; TMMC: time to the next Phase III wave post-dose; K\_Diss: first-order dissolution rate constants.

#### *2.3. Mechanistic Oral Absorption Modeling in GastroPlus*™

#### ௦௧ ௦ௗ ௗ௨ௗ௨ ௦ௗ ௨௨ ௦ௗ ௗ௨ௗ௨ ௗ௦௦௩ௗ 2.3.1. GastroPlus™ Advanced Compartmental Absorption Transit (ACAT™) Mechanistic Absorption Model

௨௨ ௗ௦௦௩ௗ ௦ Simulations were performed by the commercially available PBPK modeling platform GastroPlus™ 9.6 (Simulations Plus, Inc., Lancaster, CA, USA) and all simulations were judged based on and compared with the observed luminal and systemic concentrations of ibuprofen after oral administration of 800 mg ibuprofen (Shreveport, LA, USA; IBU™—Ibuprofen Tablets, USP, 800 mg) to twenty healthy subjects in fasted state.

The advanced compartmental and absorption transit (ACAT™) was applied with slight adjustments related to pH, gastric emptying and fluid dynamics. This model is described in detail by Hens and Bolger [25]. To implement a dynamic pH and fluid volume as a function of time, a mixed-multiple dosage form was selected. The mixed-multiple dosage form consisted of 13 different .cat files, personalized by a different pH and volume value to simulate a dynamic fluid and pH model over time. The implemented pH values were the same average values according to the values as measured during the clinical aspiration study. The implemented values for the fluid volume were extracted from Mudie and co-workers [26]. The gastric transit time was set at 2.04 h, which conforms with the average time to phase III contractions post-dose, as observed in the clinical aspiration study of ibuprofen.

− MedChem Designer 5.0 (Simulations Plus, Inc., Lancaster, CA, USA) was applied to draw the molecular structure of ibuprofen. Data describing the drug's physicochemical and biopharmaceutical properties were obtained from literature or from estimates calculated by ADMET predictor 9.0 (Simulations Plus, Inc., Lancaster, CA, USA). All physicochemical and biopharmaceutics parameters that were used to perform the simulations are described in Table 1.


**Table 1.**Physicochemical, biopharmaceutical and pharmacokinetic disposition properties to perform simulations in GastroPlus™for ibuprofen.

A two-compartmental PK model was used to describe the distribution and clearance of ibuprofen. Values for rate constants (K10, K12, K21) were optimized based on literature data that reported systemic data of ibuprofen after intravenous (IV) administration of 800 mg ibuprofen with an infusion rate of approximately 6 min (between 5 and 7 min) [31]. Estimations of these rate constants were performed by the PKPlus™ module.

#### 2.3.2. Data Presentation

The observed intraluminal and systemic concentration–time profiles are presented as the mean ± standard deviation (SD) for all participating subjects and are extracted from previous work [13,14]. Pharmacokinetic and intraluminal parameters are reported and compared with the simulated outcomes.

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

#### *3.1. Mechanistic Oral Absorption Modeling in WinNonlin*

Figure 2 shows the simulated outcomes (model-fitted values) when applying the InVivo\_GIS (purple lines) versus the InVivo\_GISPlus (red lines).

When comparing both simulated profiles derived from the two models, it is clear that including luminal pH values and the 'house-keeper' phase III wave contractions provide better predicted (i.e., model-fitted) values. This is not only the case for the simulated plasma concentrations—capturing plasma *C*max—but also for the improved predictions with respect to luminal levels reflecting the oscillations associated with pH changes. Applying dynamic pH values and, therefore, showing an improved reflection of the intraluminal behavior was also observed when using the InVivo\_GISPlus model to simulate the average concentration–time profiles in the duodenum, jejunum and plasma concentrations (Figure 3).

To quantitatively assess the improved predictions with the InVivo\_GISPlus model, a comparison between the outcomes derived from the InVivo\_GIS and InVivo\_GISPlus was made and the simulated results were compared with the observed data and the prediction error was expressed as an absolute percentage deviation. Table 2 summarizes the mean absolute percentage deviation between predicted and experimental values in the main pharmacokinetic parameters (*C*max, *T*max, and *AUC*) in the three compartments, namely the duodenum, jejunum, and plasma.


**Table 2.** Average absolute deviation percentage between predicted and experimental values of *C*max, *T*max and *AUC* in the duodenum, jejunum and plasma.

**Figure 2.** *Cont*.

**Figure 2.** (**A**) and (**B**) describe the experimental and model-fitted concentration values of ibuprofen for all subjects. Experimental (black dots) and model-fitted concentration values of ibuprofen in the duodenal, jejunal and systemic compartments for each and every subject in fasting state conditions. Each row includes the observed and simulated concentration–time profiles in the duodenal, jejunal and plasma segments in three separate columns. The purple line reflects the InVivo\_GIS model-predicted values, whereas the red line corresponds with the simulated (model-fitted) values derived from the InVivo\_GISPlus model. The experimental values are reflected by the dark grey lines and dots.

**Figure 3.** Simulated outcomes of the average concentrations using the InVivo\_GISPlus model. Average experimental ibuprofen concentrations are shown by the black dots in the duodenum (**A**), jejunum (**B**) and plasma (**C**) across all the subjects. Simulated values of the InVivo\_GISPlus model are represented with red lines. The experimental values are reflected by the dark black lines and dots.

Table 3 shows the applied settings of each model parameter to adequately simulate the corresponding intraluminal and systemic concentration–time profiles for each individual.

**Table 3.** InVivo\_GISPlus individual and average parameter values. Median values and the fitted values for the average subjects are also shown for comparison. SD: standard deviation; CV: coefficient of variation. *K*empt is the gastric emptying rate constant; K\_TD and K\_TJ represent the transit rate constants in the duodenal and jejunal compartment, respectively; K\_Diss represent the intestinal dissolution rate constant; V1 and V2 represent the duodenal and jejunal residual volumes, respectively.


The estimated volumes are higher than the reported values in duodenum and jejunum measured by magnetic resonance imaging (MRI) [26]. Nevertheless, it should be noted that the applied average value could be interpreted as the volume of fluid directly in contact with the solid particles during that specific period. Considering that continuous secretion and absorption of water in the small intestine occurs, the volume of fluid flowing through the segment can be high. A potential interpretation to translate the obtained parameters to the in vitro GIS device could be that the average fluid volume in contact with solid ibuprofen in duodenum and jejunum is in total 220 mL over an 8 h period to complete absorption. That would correspond to a 0.5 mL/min volumetric flow to be implemented in the in vitro GIS system.

As ibuprofen will be heavily dependent on the pH along the GI tract to dissolve, the dissolution rate was plotted against the residual pH values. Figure 4 depicts the intestinal dissolution rate (estimated from the InVivo\_GISPlus model) versus the measured average pH (duodenum and jejunum) for each individual as a function of time.

**Figure 4.** Average intestinal pH values in each subject (purple line and squares) and the intestinal dissolution rate (black line and dots) estimated from InVivo\_GISPLUS model.

The in vivo dissolution rates in each subject can be estimated from the differential equations derived from the InVivo\_GISPlus model. As Figure 4 shows, the pH fluctuations sometimes dictate directly the dissolution rate. The overlap is not perfect as other variables (such as the fluid volume) also affect the dissolution rate. However, in this model, a static volume was considered. If the dissolution rate increases, the amount of ibuprofen entering the systemic circulation will increase as well. Therefore, the intestinal dissolution rate was plotted against the absorption rate, deconvoluted from the plasma concentration–time profiles. This was done for each individual (Figure 5).

**Figure 5.** The intestinal in vivo ibuprofen dissolution rate estimated from InVivo\_GISPLUS model and the absorption rate obtained from Wagner–Nelson deconvolution profiles in each individual. Absorption rates were derived from previous work by Bermejo and co-workers [22].

This work also aimed to demonstrate the impact of gastric emptying on the systemic exposure of the drug. In this case, and as shown by Hens et al., the time of phase III contractions post-dose will determine the arrival of ibuprofen in the intestinal tract. The faster these contractions will be initiated, the higher the plasma *C*max will be. It is hypothesized that a fast onset of this house-keeper wave will remove more drug content directly from the stomach into the small intestine, resulting in high amount of drug that will be available for absorption (assuming pH > pKa). Whenever these phase III contractions are rather postponed (e.g., due to the intake of food), drug release from the stomach to the small intestine will be rather pulsatile than instantaneous, resulting in a lower driving force for intestinal absorption which ultimately leads to a lower plasma *C*max. The variability in gastric emptying of solid particles was also observed by Locatelli and colleagues when visualizing the gastric emptying process of pellets by scintigraphy studies [32]. The variability in emptying as a function of time was compared with the variability in emptying that was simulated by the InVivo\_GISPlus model for each subject and demonstrated a similar trend (Figure 6).

#### *3.2. Mechanistic Oral Absorption Modeling in GastroPlus*™

The first commercial software program to attempt a comprehensive description of the GI tract in the context of a PBPK model was GastroPlus™ (Simulations Plus, Inc., Lancaster, CA, USA). The first version of GastroPlus™ was established in August 1998 and was based on the work of Lawrence X. Yu and Gordon L. Amidon [33]. The model consisted of "continuous stirring tank reactor" compartments to describe the transit of a drug from one segment in the GI tract to the other, with simple estimations of (i) dissolution based on aqueous solubility and (ii) absorption rate coefficients based on existing pharmacokinetic data. In 2001, an advanced compartmental absorption and transit (ACAT™) model was developed and defined each compartment's volume and transit by mass balance approximations. Later on, in 2018, Hens and Bolger aimed to convert the static settings of the ACAT™ model to more

dynamic settings [25]. They developed a dynamic fluid and pH model in the GastroPlus™ simulator to reflect the dynamic alternations of fluid volumes and pH values in the different compartments of the GI tract. Especially in the case of BCS class 2 compounds, suffering from poor aqueous solubility and high permeability, these dynamic settings will result in improved predictions towards the in vivo outcome of the drug product when comparing these simulations with the simulations obtained when static, default settings were applied. The benefit of these dynamic settings has already been shown for posaconazole, a weakly basic drug [25].

**Figure 6.** (Upper plot) Data from 19 individual scintigraphy studies of gastric emptying of pellets collected by Locatelli et al. [32]. (Bottom plot) Individual solid ibuprofen particle gastric emptying kinetics predicted by InVivo\_GISPLUS model.

#### 3.2.1. Solubility versus pH: pH-Driven Dissolution

Thermodynamic solubility of ibuprofen was determined in three different aspirated fluids (i.e., gastric, duodenal and jejunal) of subject B005-F2. The ADMET Predictor 9.0 was used to predict the acidic pKa and how solubility would be defined in the physiological range. The observed versus simulated solubility values closely matched as depicted in Figure 7.

The solubility factor is equal to the ratio of the maximum solubility to the intrinsic solubility for this specific acidic pKa. This demonstrates the ability of ibuprofen to easily dissolve at pH levels above its pKa, converting to its ionized form which is more soluble than its non-ionized form.

**Figure 7.** Solubility versus pH profile for ibuprofen. The green line represents the predicted solubility versus pH curve, whereas the blue dots represent the measured solubility values in fasted-state human GI fluids. The upper blue dot is a solubility value of ibuprofen measured in fasted-state duodenal fluid by Heikkilä and co-workers [28].

It should be noted that in this clinical study, authors monitored the residual bulk pH in the GI tract (stomach, duodenum, and jejunum) and measured the solution concentrations of ibuprofen in these different regions. The pH values of the aspirates were extremely fluctuating as a consequence of the low buffer capacity [13]. As research scientists in pharmaceutical industry don't have any access to these values and mostly make use of high buffer capacity media to explore drug dissolution, overestimations in predicting the plasma *C*max will be made (see below: Advanced compartmental absorption and transit simulations: static simulations with default settings). However, there is more evidence that biopredictive dissolution setting should focus on integrating relevant aspects that have a major impact on the fraction dissolved of a drug. For instance, lowering the buffer capacity of the media and highlighting the interplay between surface and bulk pH is a critical aspect that should not be neglected. This has been observed for modified-release, but also for immediate-release formulations [34–37]. Pepin and co-workers modeled the dissolution profiles of acalabrutinib (weak base, pKas 3.54 (B), 5.77 (B), 12.1 (A)) using an in-house Excel® tool and concluded that making use of the bulk apparent drug solubility will lead to an over-estimation of the drug dissolution rate at all pH values below the highest drug pKa [38]. By taking into account product particle size distribution (P-PSD) and the surface pH of drug particles, accurate simulated dissolution rates were simulated. In addition, in the case of ionizable compounds such as ibuprofen, the pH at the surface of dissolving particles (pH0) is a complex function between buffer- and drug-related properties. In the case of ibuprofen, in vitro results demonstrated that the surface pH is lower than the bulk pH (ranging from pH 4.8–5.8) in 5 mM bicarbonate depending upon hydrodynamics which can hamper the dissolution process as such. This was observed by Al-Gousous and co-workers, who observed differences in bulk and surface pH (and thus bulk and surface solubility) resulting in slower dissolution kinetics of ibuprofen when using low bicarbonate-buffered, dissolution media [39]. Therefore, the use of the human buffer bicarbonate will be more favorable to adequately reflect the luminal dissolution kinetics and to represent the relevant interactions between buffer species and drug molecules [39–41]. Biopredictive dissolution tests performed by Cristofoletti et al. demonstrated that ibuprofen dissolution in a lower buffer capacity medium (i.e., 5 mM phosphate buffer) affected bulk pH and its own dissolution kinetics [42,43]. Considering P-PSD and the self-buffering capacity of ibuprofen resulted in the best simulation with respect to plasma *C*max and AUC for two Nurofen® tablets of 200 mg orally administered to healthy

adults. In another clinical study by Hofmann and colleagues [44], ibuprofen suspensions (varying in particle size radius) were intraduodenally administered in healthy subjects. After administration, duodenal pH was monitored in parallel with systemic exposure of ibuprofen. The administration of small particles led to a more pronounced pH drop than for large particles under the same infusion conditions. Still, absorption rates were higher for these smaller particles compared to the larger particles, but no significant differences in plasma *C*max were observed, suggesting that variability in the systemic outcome of the drug is more related to the rate of gastric emptying (motility-driven) and/or intestinal transit times. Besides the bulk/surface pH, the hydrodynamics of the GI tract also have an enormous impact on the dissolution rate. Performing dissolution experiments for 200 mg of ibuprofen in 5 mM phosphate buffer at 75, 50 and 30 rotations per minute (RPM) in the USP II apparatus, demonstrated maximum cumulative fractions dissolved of 0.83, 0.84 and 0.26, respectively. Based on computational fluid dynamics (CFD), the presented shear rates in the GI tract are more in line with the shear rates that are reproduced when lower rotation speeds are applied [6].

In conclusion, some important, yet crucial, GI variables should be integrated into biopredictive dissolution testing (low buffer capacity media, phosphate versus bicarbonate buffer, hydrodynamics) to account for a valuable input for PBPK platform programs.

3.2.2. Simulation of Distribution and Clearance of Ibuprofen: A Two-Compartmental Pharmacokinetic (PK) Approach

Simulation of distribution and clearance of ibuprofen was performed in the PKPlus™ module based on literature data that showed the clearance of ibuprofen after an intravenous administration of an 800 mg dose of ibuprofen with a perfusion rate of approximately 6 min. For this study, 12 healthy subjects (aged between 18 and 65 years old) were recruited [31]. One-, two- and three-compartmental models were compared, and the best fit was observed for a two-compartmental PK model based on the Akaike Information Criterion (AIC) and *R* 2 . As no extensive first-pass metabolism is observed for ibuprofen [45], a two-compartmental approach is definitely sufficient in describing the distribution and clearance of ibuprofen. Observed versus simulated data are depicted in Figure 8.

**Figure 8.** Observed (green squares) versus predicted (green line) concentrations of ibuprofen after intravenous administration of an 800 mg dose to 12 healthy subjects [31]. Simulations were performed by using a two-compartmental PK model.

3.2.3. Advanced Compartmental Absorption and Transit Simulations: Static Simulations with Default Settings

In the first set of modeling experiments, the default settings of GastroPlus™ were applied in order to assess predictions when a static volume and pH is applied in each compartment of the ACAT™ model (Table 4).


**Table 4.** Default setting values for volume (mL) and pH in the GastroPlus™ simulator.

Regarding the fact that (i) there is no dissolution-limiting step at pH > 6 and that (ii) large volumes are always present during the simulation time, fast onset of dissolution was observed in the intestinal compartments (Figure 9).

#### **Figure 9.** Simulated concentration–time profiles of ibuprofen in the different GI compartments of the GastroPlus™ simulator.

Solution concentrations in the stomach are negligible due to the integrated acidic pH (pH 1.3) that will prevent almost any dissolution of ibuprofen in the gastric compartment. After transfer, ibuprofen will rapidly dissolve in the more neutral pH environment of the intestinal tract. There is a trend that the dissolution is the highest in the duodenum followed by the first part of the jejunum and the second part of the jejunum (diluting effect). When the 800 mg dose appears in the first compartment of the GI tract, dissolution will be enhanced due to the large volume (44.57 mL) and high pH (pH 6). As the solubility of ibuprofen is approximately 2 mg/mL at pH 6.2 (Figure 7), an enormous amount of ibuprofen will immediately dissolve and be available for absorption without any limitations. As the concentration at the surface of the ibuprofen particles (*C*s) drives the dissolution rate under sink conditions (pH < 6.2), whereas the bulk concentration (*C*b) comes into play in the case when there are no sink conditions [39].

This is an ongoing process until it hits the transit time to move forward for the remaining undissolved particles towards the next compartment. Obviously, this luminal behavior will be reflected in the systemic compartment, as depicted in Figure 10.

**Figure 10.** Observed (blue squares) and simulated (blue line) plasma concentrations of ibuprofen after oral administration of an 800 mg dose.

Pharmacokinetic parameters are shown in Table 5.

**Table 5.** Pharmacokinetic disposition parameters between observed and simulated data applying the default settings.


Using the default settings, simulations performed by GastroPlus™ are overestimating the observed pharmacokinetic parameters with respect to plasma *C*max and plasma *T*max. Although predictions are in the same concentration range, there is still a 33% higher predicted plasma *C*max compared to the observed plasma *C*max. Moreover, the simulated plasma *T*max is appearing tremendously earlier related to the fast dissolution in the intestinal compartments. The clinical aspiration study clearly demonstrated the presence of ibuprofen in the GI tract up to 7 h. This simulation, however, informs us that the amount absorbed of ibuprofen is 100% in approximately 2 h (Figure 11).

In conclusion, there is a mismatch between the simulations and the observed data, highly related to physiological variables (i.e., pH, fluid volumes and gastric emptying) that were considered as pivotal covariates explaining intersubject variability in oral and systemic drug behavior of ibuprofen. Therefore, in the next set of simulations, these variables will be optimized.

3.2.4. Advanced Compartmental Absorption and Transit Simulations: Dynamic Simulations with Adjusted Settings

In the second set of simulations experiments, the goal was to better (in terms of accuracy) reflect the intraluminal behavior taking into account some important physiological aspects that were not considered during the simulations when default settings were applied. All data derived from the clinical aspiration study were analyzed in previous work and revealed how GI motility and pH have a major impact on plasma *C*max and *T*max, respectively. As intraluminal pH is not static at all in the human GI tract, the ACAT™ model was adjusted using a constantly changing pH as a function of time, in line with the values observed in the clinical aspiration study. This was applied to all segments of the

GI tract (i.e., stomach, duodenum and jejunum). Moreover, the gastric transit time was delayed from 0.25 h to 2.04 h. As the observed house-keeper wave was on average 2.04 h [22], the gastric transit time was postponed to this specific value. The strong burst of phase III contractions is a surrogate for the rapid emptying of gastric content into the upper small intestine and was demonstrated to have an important impact on the plasma *C*max of ibuprofen for this specific study.

**Figure 11.** Amount absorbed of ibuprofen under the default settings in the GastroPlus™ simulator.

Application of mixed-multiple doses as dosage form in GastroPlus™ makes it possible to implement different .cat files for each time point (every 15 min) all containing a different fluid volume and pH as observed by MRI data and pH values of aspirated fluids, respectively. All these separate .cat files were uploaded in the GastroPlus™ simulator and were used in chronological order. Figure 12 demonstrates the simulated and observed plasma data.

**Figure 12.** Simulated and observed plasma concentrations of ibuprofen applying the dynamic ACAT™ model.

Pharmacokinetic disposition parameters are shown in Table 6.

∞

The application of the dynamic settings resulted in a predicted plasma *C*max differing 22% compared to the observed plasma *C*max. The predicted plasma *T*max was similar to the observed

∞

plasma *T*max. The exposure (expressed as the area under the curve (*AUC*)) is the same as observed for the simulations performed with the default settings, highly likely because the amount of dose that is absorbed is 100%. However, the dissolution and absorption of ibuprofen were more sustained using dynamic settings (Figure 13).

**Table 6.** Pharmacokinetic disposition parameters between observed and simulated data applying the default settings.


**Figure 13.** The amount dissolved, absorbed, reaching the portal vein and reaching the systemic circulation after mechanistically modeling of ibuprofen concentrations in GastroPlus™.

When comparing the fraction absorbed from the observed date (by deconvolution) versus the fraction observed that was simulated in GastroPlus™ using the dynamic settings, a positive overlap was observed (Figure 14).

Figure 15 represents the simulated and observed intraluminal profiles in the different regions of the GI tract.

**Figure 14.** The observed fraction absorbed (based on Wagner–Nelson deconvolution) versus the simulated fraction absorbed from the GastroPlus™ simulator.

**Figure 15.** Simulated and observed intraluminal concentrations of ibuprofen in the different segments of the GI tract. Observed data are presented as the mean±SD.

Simulated results are in line with the observed data, although slightly overestimated with respect to the jejunal concentrations. The reason for this phenomenon can be attributed to the fact that GastroPlus™ handles the small intestine as different compartments, all characterized by specific transit times. As the human jejunum and ileum are quite large (2.5 m and 3.5 m, respectively), the presence of fluid pockets may play a significant role in the amount of drug that will dissolve. It is estimated that the jejunum consists of 13 mL of fluids based on the MRI data of Mudie and co-workers [26], which is more than enough for ibuprofen to dissolve. However, in vivo, these volumes are presented as small, separate water pockets that may hamper the drug to reach its equilibrium solubility. The impact of pockets is less concerned for the duodenal compartment as this segment is only 25 cm long and, based on these simulations, the absence of water pockets seems to have less impact on drug dissolution as the predictions are in line with the in vivo duodenal concentrations. Therefore, these simulations may open public debate to discuss the relevance of simulating these pockets in computational modeling as recently done by Yu and co-workers [46], who developed a dynamic fluid compartmental and absorption transit (DFCAT) model. Related to the compound characteristics (e.g., solubility and permeability), stochastic modeling of these fluid pockets could influence the predicted outcome of the drug product. Figure 16 represents the amounts of ibuprofen dissolved versus undissolved when applying a dynamic versus static fluid model (pH values of the different compartments were the same for both simulations).

We hypothesize that the present fluids will impact the amount of drug dissolved, even though the drug has no limited capacity with respect to dissolution (BCS class 1/2a/3) and regardless of favorable pH to initiate dissolution of the drug in the intestine (BCS class 2a) [47]. Future studies should shed light on the relevance and importance of this topic.

The rate of gastric emptying will alter systemic exposure in such a way that it is favorable to have a fast release of ibuprofen into the small intestine which will generate a high driving force for intestinal absorption. In contrast, when gastric emptying is delayed and ibuprofen will be sustained released from the stomach, the driving force will not be as big as for a direct onset of gastric emptying. Figure 17 demonstrates the parameter sensitivity analysis (PSA) with respect to the 'gastric transit time'. These data are in line with the observed data from the clinical study where a fast onset of post-dose phase III contractions resulted in a higher plasma *C*max.

### Dynamic Volumes

### Static Volumes

**Figure 16.** Simulated GI profiles when applying a static (i.e., constant) versus a dynamic fluid model.

**Figure 17.** Impact of stomach transit time on the plasma *C*max for ibuprofen after oral administration of an 800 mg dose in fasted state. PSA: parameter sensitivity analysis.

#### **4. Conclusions and Future Directions: Requesting Biowaivers?**

In conclusion, this work demonstrated a mechanistic modeling approach to explain the intraluminal and systemic performance of an orally administered ibuprofen drug product (RLD) in fasted-state conditions. The simulations that were performed highlighted the importance of considering gastric emptying, fluid volumes, motility, and pH as indispensable covariates that should be included in the models to ensure realistic predictions of systemic plasma concentrations of ibuprofen. Both these simulations were done in a user-customized model (Phoenix WinNonlin®) and in a commercially available software package (GastroPlus™). In both cases, we demonstrated the importance of integrating physiological variables to mechanistically understand and observe the impact of these parameters on ibuprofen intestinal absorption. With respect to fluid volumes, simulations in GastroPlus™ demonstrated the impact of dynamic fluids, as measured by an MRI study, on the dissolved amounts of ibuprofen throughout the different GI compartments. There should be further developed with respect to the 'complexity' of in silico models to predict the in vivo outcome plasma levels of a drug. Clear guidelines should assist formulation scientists in whether specific physiological variables should be considered and should be integrated into computational models, highly depending on the purpose of the simulation. Parameter and sensitivity analyses can serve as useful tools to assess the sensitivity of physiological variables predicting the in vivo performance of an oral drug product. Even today, the dynamics of these GI processes are relatively poorly described quantitatively and do not adequately reflect the in vivo GI conditions and thus the plasma performance of the orally administered ibuprofen. While simulations can be performed and mechanistic insight gained from such simulations from current software, we need to further determine the dynamics of the GI variables controlling the dosage form transit, disintegration, dissolution absorption and metabolism along the GI tract in order to move from correlation to prediction (IVIVC → IVIVP). The obtained underlying cumulative dissolution profile derived from the GastroPlus™ simulator (Figure 13) and the simulated concentration–time profiles from the Phoenix WinNonlin® platform (Figure 2) will serve as a reference to optimize our in-house biopredictive dissolution device, the Gastrointestinal Simulator (GIS). Establishing the link between biopredictive in vitro dissolution testing and mechanistic oral absorption modeling (i.e., physiologically-based biopharmaceutics modeling (PBBM)) opens an opportunity to potentially request biowaivers in the near future for orally administered drug products, regardless of its classification according to the BCS [47–50].

**Supplementary Materials:** Both mathematical equations for simulations performed in Phoenix WinNonlin® as well as the dynamic .cat files for simulations in GastroPlus™ are provided for the reader. These .cat files are compatible with all versions of GastroPlus™. All these files can be found at the following link for free: https://zenodo.org/record/3562430#.XfSmtOhKg-U.

**Author Contributions:** Conceptualization, M.B., B.H. and G.L.A.; Data curation, M.B., J.D., D.M., P.P., Y.T., K.S. and G.L.A.; Formal analysis, B.H., J.D., D.M., Y.T., K.S. and G.L.A.; Funding acquisition, G.L.A.; Investigation, P.P.; Software, M.B.; Writing-original draft, B.H. All authors have read and agreed to the published version of the manuscript.

**Funding:** This work was partially supported by grant # HHSF223201510157C and grant # HHSF223201310144C by the U.S. Food and Drug Administration (FDA). B.H. acknowledges the financial support from the Flemish Research Council (FWO—applicant number postdoctoral researcher: 12R2119N).

**Acknowledgments:** These data were presented at the 2019 annual American Association of Pharmaceutical Scientists (AAPS) meeting on 3 November 2019 in San Antonio, TX.

**Conflicts of Interest:** The authors declare no conflict of interest. Yasuhiro Tsume is from Merck & Co., Inc., the company had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, and in the decision to publish the results.

#### **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* **BCS Class IV Oral Drugs and Absorption Windows: Regional-Dependent Intestinal Permeability of Furosemide**

#### **Milica Markovic 1 , Moran Zur 1 , Inna Ragatsky 1 , Sandra Cviji´c <sup>2</sup> and Arik Dahan 1, \***


Received: 16 November 2020; Accepted: 30 November 2020; Published: 2 December 2020

**Abstract:** Biopharmaceutical classification system (BCS) class IV drugs (low-solubility low-permeability) are generally poor drug candidates, yet, ~5% of oral drugs on the market belong to this class. While solubility is often predictable, intestinal permeability is rather complicated and highly dependent on many biochemical/physiological parameters. In this work, we investigated the solubility/permeability of BCS class IV drug, furosemide, considering the complexity of the entire small intestine (SI). Furosemide solubility, physicochemical properties, and intestinal permeability were thoroughly investigated in-vitro and in-vivo throughout the SI. In addition, advanced in-silico simulations (GastroPlus ®) were used to elucidate furosemide regional-dependent absorption pattern. Metoprolol was used as the low/high permeability class boundary. Furosemide was found to be a low-solubility compound. Log D of furosemide at the three pH values 6.5, 7.0, and 7.5 (representing the conditions throughout the SI) showed a downward trend. Similarly, segmental-dependent in-vivo intestinal permeability was revealed; as the intestinal region becomes progressively distal, and the pH gradually increases, the permeability of furosemide significantly decreased. The opposite trend was evident for metoprolol. Theoretical physicochemical analysis based on ionization, pKa, and partitioning predicted the same trend and confirmed the experimental results. Computational simulations clearly showed the effect of furosemide's regional-dependent permeability on its absorption, as well as the critical role of the drug's absorption window on the overall bioavailability. The data reveals the absorption window of furosemide in the proximal SI, allowing adequate absorption and consequent effect, despite its class IV characteristics. Nevertheless, this absorption window so early on in the SI rules out the suitability of controlled-release furosemide formulations, as confirmed by the in-silico results. The potential link between segmental-dependent intestinal permeability and adequate oral absorption of BCS Class IV drugs may aid to develop challenging drugs as successful oral products.

**Keywords:** BCS class IV drugs; segmental-dependent intestinal permeability; intestinal absorption; oral drug delivery; biopharmaceutics; physiologically-based pharmacokinetic (PBPK) modeling; furosemide

#### **1. Introduction**

The biopharmaceutical classification system (BCS) developed by Amidon et al. revealed that the solubility/dissolution of the drug and its intestinal permeability are the two key factors that dictate drug absorption following oral administration [1,2]. Drug solubility in the gastrointestinal milieu may change in different intestinal segments, e.g., due to pH changes, in a fairly predictable manner; depending on the pKa, the solubility of acidic drugs may increase as the luminal pH rises in more distal regions of the small intestine, and vice versa for basic drugs [3–5]. On the other hand, time- and segmental-dependent intestinal permeability is more complicated and harder to predict [1]. Mechanisms contributing to segmental-dependent permeability throughout the gastrointestinal tract (GIT) include different morphology along the GIT, variable intestinal mucosal cell differentiation, changes in the drug concentration (in case of carrier-mediated transport), modulation of tight junction permeability, and luminal contents and properties, e.g., pH, transporter expression, variability in the structure/composition of the intestinal membrane itself, and more [6–11].

The four BCS classes highlight the limiting factors of the absorption process: (1) Class I, high-solubility high-permeability drugs, indicate the easier and straightforward development process, and complete absorption is expected; (2) Class II, low-solubility high-permeability drugs, indicate that a solubility/dissolution limitation is expected; (3) Class III, high-solubility low-permeability drugs, indicate that the intestinal absorption of this class of drugs will be limited by the permeability rate; and (4) Class IV, low-solubility low-permeability drugs [12]. Since Class IV drugs suffer from inadequate solubility and permeability, they have very poor oral bioavailability and are inclined to exhibit very large inter- and intrasubject variability. Therefore, unless the drug dose is very low, they are generally poor oral drug candidates. Yet, according to some estimates, ~5% of the world's top oral drugs belong to this class [13–15]. In some cases, this is due to the absorption window, which is often critical for the success or failure of a certain drug. In order to gather information about the drug absorption window, extensive work and thorough analysis of luminal conditions and drug absorption is needed, within different locations throughout the GIT. Here, we present such analysis for BCS class IV drug, furosemide [16].

Furosemide is a powerful loop diuretic and is indicated for treating edematous conditions associated with heart, renal, and hepatic failure, as well as for the treatment of hypertension [17,18]. Drug therapy with furosemide is often complex, due to apparent erratic oral systemic availability and unpredictable responses to an administered dose [19]. Even though furosemide is a class IV drug, it is a very common and widely prescribed drug on the market.

In this work, we aimed to investigate the reason for apparent success of furosemide as a marketed product, despite its poor biopharmaceutical properties, and classification as BCS class IV drug, in order to allow development of future class IV compounds. We posit that segmental-dependent permeability of furosemide may contribute to its absorption complexity and provide a certain absorption window in which the drug has suitable permeability and, hence, gets absorbed. For this reason, we investigated the in-vivo intestinal permeability of furosemide throughout different segments of the small intestine. Solubility studies, as well as theoretical physicochemical analysis of furosemide and advanced modern in-silico GastroPlus® simulations, were performed, in order to elucidate the mechanistic reasons behind the experimental results. Furosemide data were compared to the β-blocker metoprolol, the Food and Drug Administration (FDA) reference drug for the low/high permeability class boundary. Overall, this experimental setup allowed us to reveal important insights on the performance of furosemide, despite its unfavorable drug-like properties, and discuss extrapolation of these insights to other BCS class IV drug candidates.

#### **2. Methods**

#### *2.1. Materials*

Furosemide, metoprolol, phenol red, potassium chloride, potassium phosphate monobasic, potassium phosphate dibasic, sodium chloride, acetic acid, maleic acid, *n*-octanol, and trifluoroacetic acid (TFA) were all purchased from Sigma Chemical Co. (St. Louis, MO, USA). Acetonitrile and water, ultra-performance liquid chromatography (UPLC) grade were purchased from Merck KGaA, Darmstadt, Germany. Remaining chemicals were of analytical reagent grade.

#### *2.2. Solubility Studies*

The pH-dependent solubility studies were performed using the shake flask method, as previously reported [20–23]. The equilibrium solubility of furosemide was determined at both 37 ◦C and at room temperature (25 ◦C), in phosphate buffer pH 7.5, acetate buffer pH 4.0, and maleate buffer pH 1.0. Surplus quantity of furosemide was introduced to glass vials holding buffer solutions with different pH; the pH of those solutions was measured following drug addition to the buffers and, consequently, placed in the shaking incubator (100 rpm) at 37 ◦C. The vials were centrifuged (10,000 rpm, 10 min), and the supernatant was instantly analyzed by UPLC. The dose number for furosemide was calculated using the established equation: *D*<sup>0</sup> = *M*/*V*0/*C*s; *M* being the highest single-unit dose strength of furosemide (taken as 80 mg [24]), *V*<sup>0</sup> is the initial volume of water (250 mL), and *C*<sup>s</sup> is the solubility at each pH; the drug is considered highly soluble if the *D*<sup>0</sup> < 1.

#### *2.3. Evaluation of Octanol-Bu*ff*er Partition Coe*ffi*cients (Log D)*

Furosemide and metoprolol experimental octanol-buffer partition coefficients (Log D) were studied at pH 6.5, 7.0, and 7.5 using the shake-flask method [8,11]. Drug solutions in octanol-saturated phosphate buffer (pH 6.5, 7.0, 7.5) were equilibrated at 37 ◦C for 48 h. The octanol and water phase were divided via centrifugation, and the drug content in the water phase was quantified using UPLC; the furosemide/metoprolol concentration in the octanol phase was determined by mass balance.

#### *2.4. Physicochemical Analysis*

The theoretical fraction extracted into octanol (fe) was calculated using the following equation [25,26].

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

in which P represents the octanol-water partition coefficient of the unionized drug form, and f<sup>u</sup> is the fraction unionized of the drug at a certain pH. Experimental Log P values were taken from the literature for both furosemide (2.29) [27] and metoprolol (2.19) [28]. The f<sup>u</sup> versus pH was plotted according to the Henderson-Hasselbalch equation, using the pK<sup>a</sup> literature values: 9.68 for metoprolol [29] and 3.8 for furosemide [24].

#### *2.5. Rat Single-Pass Intestinal Perfusion*

Effective permeability coefficient (Peff) of furosemide versus metoprolol in various intestinal segments was assessed using the single-pass rat intestinal perfusion (SPIP) in-vivo model. The murine studies were completed according to the approved protocol by Ben-Gurion University of the Negev Animal Use and Care Committee (Protocol IL-08-01-2015). The animals (male Wistar rats weighing 230–260 g, Harlan, Israel) were housed and handled according to Ben-Gurion University of the Negev Unit for Laboratory Animal Medicine Guidelines. All animals were fasted overnight (12–18 h) with free access to water; rats were randomly allocated to different experimental groups. The intestinal perfusion study was performed according to the previous reports [7,9,30–32]. Animals were anesthetized via intramuscular injection of 1 mL/kg ketamine-xylazine solution (9%:1%) and placed on a heated (37 ◦C) surface (Harvard Apparatus Inc., Holliston, MA, USA); the rat abdomen was uncovered via a midline incision (~3 cm). Permeability (Peff) was measured in proximal jejunum (starting 2 cm lower from the ligament of Treitz), mid-small intestine (SI) segment (isolated between the end of the upper and the beginning of the lower segments), and distal segment of the ileum (ending 2 cm above the cecum) accounting for the complexity of the entire SI [7]. Intestinal segments were cannulated on both ends and perfused with drug-free buffer. Working solutions containing furosemide (320 µg/mL), metoprolol (400 µg/mL), and phenol red (a non-absorbable marker for water flux measurements) were prepared with potassium phosphate monobasic and sodium phosphate dibasic, to achieve pH of 6.5, 7.0 and 7.5; osmolarity (290 mOsm/L) and ionic strength in all buffers was maintained throughout the study. Drug solutions were incubated in a 37 ◦C water bath. Steady-state environment was ensured by perfusing the drug-containing buffer for 1 h, followed by additional 1 h of perfusion, during which sampling was done every 10 min. The pH of the collected samples was measured in the outlet sample to verify that there was no pH change throughout the perfusion. All samples were assayed by UPLC. The length of each perfused intestinal segment was measured in the end of the experiment. The effective permeability (Peff; cm/s) through the rat SI wall was calculated according to the following equation:

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

in which Q is the perfusion buffer flow rate (0.2 mL/min); C ′ out/C′ in is the ratio of the outlet/inlet drug concentration adjusted for water transport; R is the radius of the intestinal segment (conventionally used as 0.2 cm); and L is the exact length of the perfused SI segment as was measured at the experiment endpoint [7,33,34].

#### *2.6. Analytical Methods*

Concentration of furosemide and metoprolol was evaluated using an UPLC instrument Waters Acquity UPLC H-Class (Milford, MA, USA), with a photodiode array detector and Empower software. Furosemide and metoprolol were separated on Acquity UPLC XTerra C18 3.5 µm 4.6 mm × 250 mm column (Waters Co., Milford, MA, USA). Gradient mobile phase, going from 70:30% to 90:10% *v*/*v* 0.1% trifluoroacetic acid in water/acetonitrile, respectively, on a flow rate of 1 mL/min (25 ◦C). The inter- and intraday coefficients of variation were < 1.0% and 0.5%, respectively.

#### *2.7. Statistics*

Solubility studies were performed in four replicates; Log D studies were performed in six replicates, whereas animal perfusion studies were *n* = 4. Values are expressed as means ± standard deviation (SD). To determine statistically significant differences among the experimental groups, a 2-tailed nonparametric Mann–Whitney U test for 2-group comparison was used; *p* < 0.05 was termed significant.

#### *2.8. In-Silico Simulations*

Computer simulations of furosemide absorption and concomitant plasma concentrations following oral administration in humans were conducted using GastroPlusTM software package (v. 9.7.0009, 2019, Simulations Plus Inc., Lancaster, CA, USA). The required input data regarding drug physicochemical and pharmacokinetic properties were experimentally determined, taken from literature or in-silico predicted. Human permeability values throughout the SI were calculated from the experimental rat single-pass intestinal perfusion data, using the software integrated "permeability converter". Drug disposition was best described by three-compartmental pharmacokinetic model, whereas the relevant parameters (clearance (*CL*), volume of distribution (*V*d) and distribution constants between central and peripheral compartments) were estimated using PKPlus software module, based on the in-vivo plasma concentration data for an intravenous (i.v.) bolus dose [35]. The application of three-compartmental model to describe furosemide pharmacokinetics has already been reported in literature [36,37]. Graphical data from literature were digitized using DigIt™ program (version 1.0.4, 2001–2008, Simulations Plus, Inc., Lancaster, CA, USA). Physiological parameters were the software default values representing fasted state physiology of a healthy human representative.

The software simulates drug absorption from the GIT using the integrated Advanced Compartment Absorption and Transit (ACAT) GIT model that consists of nine compartments (stomach, duodenum, two segments of jejunum, three segments of ileum, caecum, and ascendant colon). These compartments are linked in series, and the amount of drug dissolved and absorbed from each compartment is calculated by the system of differential equations. More details on the ACAT model can be found

in the literature [38,39]. Regarding the fact that furosemide is a poorly-soluble drug, the model accounted for the effect of bile salt on drug solubility and diffusion coefficient. Drug dissolution rate under physiological conditions was predicted using the software default Johnson dissolution equation (based on modified Nernst-Bruner equation) [40].

The validity of the model (i.e., the selection of input values) was validated by comparison of the prediction results (bioavailability (F), maximum plasma concentration (Cmax), time to reach Cmax (tmax), and area under the plasma concentration-time curve (AUC0–∞)) with published data from the in-vivo studies for peroral (p.o.) drug administration. Percent prediction error (%PE) between the predicted and mean in-vivo observed data from a clinical study was calculated using the following equation:

$$\% \text{PE} = \frac{(\text{Observed value} - \text{Predicted value}) \times 100}{\text{Observed value}}.$$

In the next step, the generated model was used to mechanistically interpret furosemide regional absorption pattern, and to estimate the outcomes for various hypothetical drug dissolution scenarios (illustrating drug dissolution from immediate-release (IR) and controlled-release (CR) oral formulations). In the last case, hypothetical dissolution profiles were used as additional inputs to describe drug release rate in-vivo, and the selected dosage form was "CR dispersed" to allow input of the tabulated dissolution data.

#### **3. Results**

The solubility values obtained for furosemide at 37 ◦C and at room temperature (25 ◦C) are summarized in Table 1, as well as the corresponding dose number (D0). Furosemide showed pH-dependent solubility, in accordance with its acidic nature. It can be seen that, while, at pH 7.5, furosemide has suitable solubility (as evident by D<sup>0</sup> lower than 1), at the lower pH values, 1.0 and 4.0, it is poorly soluble. When taking 80 mg as the highest dose strength, although D<sup>0</sup> < 1 was obtained at pH 7.5, at pH 1.0 and 4.0, the D<sup>0</sup> is higher than 1; hence, furosemide was found to be a low-solubility compound according to the BCS.

**Table 1.** Furosemide solubility values (µg/mL) at the tree pH values 1.0, 4.0, and 7.5, at 37 ◦C (upper panel), and at room temperature (25 ◦C; lower panel), as well as the corresponding dose number (D<sup>0</sup> ) calculated for an 80-mg dose. Data presented as mean ± SD; *n* = 6.


Octanol-buffer partition coefficient values of furosemide and metoprolol at the three pH values 6.5, 7.0, and 7.5 (representing the conditions throughout the small intestine) are presented in Figure 1. Both drugs presented a clear pH-dependent Log D values across the studied pH range, with opposite trends; while furosemide's partitioning decreases as the pH rises, metoprolol shows higher partitioning into octanol at higher pH (metoprolol is the acceptable reference drug for the low/high permeability class boundary). In addition, furosemide's Log D at pH 6.5 was higher than that of metoprolol at the same pH; this is a surprising finding since Log D may sometimes be used as a surrogate for passive

permeability. Indeed, at higher pH values (7.0 and 7.5), metoprolol Log D increases, while furosemide decreases, and metoprolol Log D becomes higher than furosemide. *Pharmaceutics* **2020**, *12*, x 6 of 17

*Pharmaceutics* **2020**, *12*, x 6 of 17

**Figure 1.** The octanol-buffer partition coefficients, Log D, for furosemide and metoprolol at the three pH values 6.5, 7.0, and 7.5. Data are presented as the mean ± S.D.; *n* = 6 in each experimental group.

− Furosemide and metoprolol physicochemical properties are presented in Table 2. Figure 2 presents furosemide versus metoprolol theoretical fraction unionized (fu) and fraction extracted into octanol (fe) as a function of pH. The plots have a standard sigmoidal shape, with opposite trends for furosemide vs. metoprolol. The f<sup>e</sup> vs. pH plot follows the same pattern to the f<sup>u</sup> plot, only with a shift to the right (higher pH values) for acidic drug (furosemide), and to the left (lower pH values) for basic (metoprolol) drugs. The shift magnitude in both cases equals Log(P − 1) at the midpoint of the f<sup>e</sup> and f<sup>u</sup> curves [25,26]. The experimental drug octanol-buffer partitioning at the three pH values (6.5, 7.0, and 7.5) are illustrated in Figure 2, as well, and it can be seen that they were in excellent agreement with the theoretical plots. − −

**Table 2.** Physicochemical parameters and chemical structure of furosemide and metoprolol.

**Figure 2.** The theoretical fraction unionized (fu) and fraction extracted into octanol (fe) plots as a function of pH for furosemide and metoprolol, as well as experimental buffer-octanol partitioning of the drugs in the three pH values 6.5, 7.0, and 7.5 (*n* = 5).

The effective permeability coefficient (Peff, cm/sec) values of furosemide and metoprolol determined using the single-pass intestinal perfusion (SPIP) rat model, in three intestinal segments, namely proximal jejunum (pH 6.5), mid small intestine (pH 7.0), and distal ileum (pH 7.5), are presented in Figure 3. It can be seen that significant regional-dependent permeability of furosemide throughout the small intestine was evident: the permeability of furosemide gradually decreases, while the permeability of metoprolol gradually increases, as the SI segments become more distal.

**Figure 3.** Effective permeability values (Peff; cm/s) obtained for furosemide and metoprolol after in-situ single pass perfusion to the rat proximal jejunum at pH 6.5, mid-small intestine at pH 7.0, and to the distal ileum at pH 7.5. Mean ± S.D.; *n* = 4 in each experimental group; \*\* *p* < 0.01, \*\*\* *p* < 0.001.

The input data regarding drug physicochemical and pharmacokinetic properties, used for in-silico simulations, are presented in Table 3. The simulated furosemide plasma concentration profile following p.o. administration is depicted in Figure 4, along with the mean profiles observed in the in-vivo studies. In addition, the observed and model predicted pharmacokinetic parameters are compared in Table 4. The presented data demonstrate that the generated model adequately describes furosemide absorption and disposition. The course of the predicted plasma profile fairly resembles the observed data. However, certain variations are observed between the mean in-vivo data from different studies referring to the same drug dose (Figure 4, Table 4). Indeed, it has been reported that furosemide oral absorption is highly variable between individuals, e.g., Cmax varied three-fold, and tmax varied five-fold [36,37,41]; moreover, individual AUC values for 40 mg furosemide oral dose varied between 1.57 and 3.76 µg·h/mL (more than two-fold) [36,37,41], and even larger AUC values were observed in another study with the same drug dose (2.23–6.10 µg·h/mL) [42], indicating that, regardless of the high PE(%) values in Table 4, the model predicted value of 3.66 µg·h/mL is not an overestimate of the extent of drug absorption. In addition, extensive intrasubject variability was observed for orally dosed furosemide, and these variations were attributed to the absorption process (i.e., day to day variations in physiological factors) since the repeated i.v. doses showed only marginal intrasubject variability [36,37,41]. Considering pronounced inter- and intraindividual variability in furosemide oral absorption, the simulated profile can be seen as a reasonable estimate (Figure 4). Moreover, the predicted fraction of oral drug absorption (cc. 52%) is in accordance with the values reported in the literature [36,37].


**Table 3.** The selected input parameters for furosemide absorption GastroPlus® simulation.

*Pharmaceutics* **2020**, *12*, x 9 of 17

**Figure 4.** GastroPlus® simulated (line) versus mean observed (markers) plasma concentration profiles following p.o. administration of furosemide. Mean observed values represent 40 mg immediate-release (IR) tablet profile I [43] and 40 mg IR tablet profile II [37].

**Table 4.** Comparison between GastroPlus® simulated and in-vivo observed furosemide pharmacokinetic parameters following p.o. drug administration.


<sup>a</sup> Refers to the mean plasma profile from [43] (40 mg IR tablet); <sup>b</sup> refers to the mean plasma profile from [37] (40 mg IR tablet); NA, not available/not applicable.

The predicted furosemide dissolution and absorption profiles following an IR oral formulation (IR tablet) are illustrated in Figure 5. The generated profiles clearly indicate that drug permeability is the limiting factor for absorption under fasted state GIT conditions. Namely, although furosemide is a low-solubility drug, due to ionization at the elevated pH conditions in the proximal SI, drug dissolution from an IR formulation is expected to be fast (>85% in 30 min). Therefore, furosemide absorption from an IR formulation is mainly governed by poor permeability. The predicted regional-dependent absorption distribution (Figure 6) further highlights the role of furosemide segmental absorption on the overall drug bioavailability. As implied by the regional-dependent permeability data, but also considering the surface area available for absorption, furosemide absorption predominantly happens in the proximal parts of the SI (76.6% of the total amount absorbed into the enterocytes), and only a minor fraction of drug (23.2% of the total amount absorbed into the enterocytes) passes into systemic circulation through mid and distal GIT regions.

*Pharmaceutics* **2020**, *12*, x 10 of 17

*Pharmaceutics* **2020**, *12*, x 10 of 17

**Figure 5.** GastroPlus® simulated dissolution and absorption profiles following p.o. administration of 40 mg furosemide dose (dissolution profile was simulated using the software default Johnson equation).

**Figure 6.** GastroPlus® simulated regional absorption of furosemide following p.o. administration of 40 mg drug dose (the simulated values refer to the fraction of drug dose that entered into the enterocytes).

The prediction results corresponding to various dissolution scenarios are presented in Figure 7b–d and Table 5. According to the simulated data, furosemide release rate from an oral formulation highly impacts the concomitant absorption process, whereas prolonged drug release rate leads to marked delay in the rate and extent of drug absorption. The estimated pharmacokinetic parameters (Table 5) indicate that furosemide bioavailability would show more than a 10-fold decrease in case the complete drug dissolution is achieved within 24 h in comparison to 15 min. A similar trend is observed for Cmax and AUC values (17.75- and 17.38-fold decrease, respectively), while tmax would be prolonged (about two-fold). It is interesting to note that tmax increases with decrease in drug dissolution up to some point, but further decrease in drug dissolution (e.g., 85% in more than 6 h) would not cause additional delay in peak plasma concentration. This is because, after cc. 2 h, the drug leaves proximal parts of the intestine, where majority of furosemide absorption takes place, and, later on, in mid and especially distal intestine, only a small fraction of drug can be absorbed, as illustrated in Figure 7d.

*Pharmaceutics* **2020**, *12*, x 11 of 17

**<sup>→</sup><sup>∞</sup> ∙**

**Figure 7.** GastroPlus® simulated furosemide dissolution profiles (**a**); and (**b**) the corresponding simulated plasma profiles; (**c**) absorption profiles; and (**d**) regional absorption distribution.



#### **4. Discussion**

BCS class IV drugs (e.g., sulfamethoxazole, ritonavir, paclitaxel, and furosemide) exhibit numerous unfavorable characteristics (low solubility and permeability, high presystemic metabolism, efflux transport), which make their oral drug delivery challenging. In addition to this, class IV drugs often demonstrate inter/intra-subject variability. Indeed, following oral administration, the absorption and bioavailability of furosemide are highly variable (37–51%) [35,41]. It has been suggested that this variability is highly depend on the absorption process [41], which in turn is dependent on drug aqueous solubility and intestinal permeability following oral administration [1,44]. It has also been hypothesized that variable gastric/intestinal first-pass metabolism can be a factor in causing incomplete and irregular furosemide absorption in humans [45]. Despite the unfavorable class IV drug characteristics, furosemide was shown to be exceptionally useful and successful marketed drug product for the treatment of edema [17]. For this reason, we decided to investigate furosemide's solubility and in-vivo regional-dependent permeability throughout the GIT, as main parameters that guide absorption of oral drugs.

It was shown that a correlation between human Peff in the jejunum and physicochemical parameters advocates that there is a high pH-dependent influence on the passive intestinal permeability in-vivo [46]. Indeed, furosemide in-vivo permeability data demonstrate a downward trend towards the distal intestinal segments as the pH gradually increases, a trend that can be expected for acidic drugs, since the pH in the intestinal lumen gradually increases towards distal SI regions (Figure 3). Many BCS class IV drugs are substrates for efflux transporters [47]. There is some evidence that furosemide might be a substrate for efflux transporters [48,49]; thus, such permeability trend could also be influenced by the P-glycoprotein (P-gp) transporter in which expression levels are increased from proximal to distal SI segments [6,50–52]. Since metoprolol's intestinal permeability is passive and does not involve carrier-mediated absorption, it exhibited pH-dependent intestinal permeability, with reverse tendency compared to furosemide; as a basic drug, metoprolol showed upward increase in permeability towards distal SI segments with rising pH values (Figure 3). At any point throughout the SI, furosemide exhibited significantly lower permeability than the benchmark (metoprolol's jejunum permeability), which confirms its BCS low-permeability classification and incomplete absorption. Despite the fact that furosemide is a low-permeability drug, the higher permeability in the proximal intestinal regions provides a window for furosemide absorption, and we posit that this is one of the main reasons for furosemide's sufficient bioavailability and success as a marketed drug. Theoretical f<sup>u</sup> and f<sup>e</sup> as a function of pH were found to be in excellent correlation to these in-vivo data. In addition, in-silico modeling indicated that furosemide dissolution from an IR formulation would be fairly complete before the drug leaves proximal SI (Figure 5), although the drug is generally classified as low-soluble, enabling timely delivery of the dissolved drug to the distinct absorption site. Complete furosemide dissolution under physiological conditions is also confirmed by the experimental solubility results (Table 1).

Furosemide Log D studies showed higher partition coefficient in comparison to metoprolol at pH 6.5, whereas, in the in-vivo intestinal perfusion experiment, furosemide showed significantly lower jejunum permeability than metoprolol (Figure 1). A possible reason for this difference in the partitioning and in-vivo permeability can be the polar surface area (PSA) of both drugs [53]. A sigmoidal relationship between the fraction absorbed following oral administration and the dynamic polar surface area was reported in the past [54–56]. It was shown that orally administered drugs with large PSA (>120) are hardly absorbed by the passive transcellular route, while drugs with a small PSA (<60) are almost completely absorbed [55,56]. This is in agreement with our results, as furosemide has much higher PSA (127.7) than metoprolol (53.2) [54,55]. Another reason for the difference in the partitioning and in-vivo permeability may be the presence of active efflux transport involved in the intestinal permeability. The influence of efflux transport at pH 6.5 (proximal intestinal segments) could decrease furosemide's permeability in-vivo, which was not accounted for in the octanol partitioning studies.

The Log P value of furosemide (2.3) is in the close proximity to that of metoprolol (2.2), pointing to high permeability (Table 2). However, the Log P calculation is based on the unionized drug fraction, and, since furosemide has acidic nature it is likely that, once it passes the acidic stomach environment, it will mostly be in ionized form (the pH throughout the GIT varies from 5.9–6.3 in the proximal SI to 7.4–7.8 in distal SI segments; pH in the colon is fluctuating between pH 5–8 [57]); therefore the high furosemide Log P is not in correspondence with permeability in-vivo. Thus, we posit that no single parameter can be used for measuring the drug absorption process, but rather, a combination of physicochemical parameters and in-vitro and in-vivo findings, as well as careful consideration of inclusion criteria prior to making decisions. Despite the high Log P value for furosemide, it was indeed confirmed that furosemide is a BCS class IV drug, based on both the solubility data (Table 1) and the intestinal permeability (Figure 3).

Suitable formulation is the main approach to create an efficacious drug product for the administration of BCS class IV drugs [47]. Absorption windows in the proximal intestinal segments can restrict the oral drug bioavailability and can be a significant limitation for the development of CR drug formulation. The underlying reasons are mechanistically explained by our in-silico results (Figure 7). As mentioned, furosemide permeability results revealed acceptable permeability in the proximal segments of the SI, which is presumably the reason why furosemide has appropriate drug bioavailability, despite being a BCS class IV drug. However, since CR products release the drug over 12–24 h, mostly in the colon, (transit time throughout the small intestine is 3–4 h [58]), the fact that furosemide is mainly absorbed from proximal SI segments, (with decreased permeability at distant GIT segments) prevents the formulation of furosemide as a CR product, as shown previously [21,59,60]. However, we believe that formulations based on gastro-retentive dosage forms (GRDF) can be shown as prosperous for furosemide [61]. There are several similar examples in the literature where absorption window occurs in the upper GI, and this has been used to create GDRF formulations to improve the drug absorption, such as riboflavin [62] and levodopa [59,63].

Several types of bariatric surgeries (specifically Roux-en-Y gastric bypass and mini bypass) result in bypassing the upper SI. In cases where the absorption window is indeed in this upper SI region, the absorption following the bariatric surgery can be hampered vastly, since the actual segment responsible for the majority of absorption is bypassed [64–66].

#### **5. Conclusions**

Regional-dependent permeability throughout the small intestine was evident for furosemide. The permeability of furosemide gradually decreases throughout the small intestine as a function of the pH change in the intestinal lumen. However, at any point throughout the small intestine, furosemide exhibited significantly lower permeability than the benchmark of metoprolol′ s permeability in the jejunum, which may explain the incomplete absorption of the drug. We propose that, for a drug to be classified as BCS low-permeability, its intestinal permeability should not match/exceed the low/high class benchmark anywhere throughout the intestinal tract, as well as is not restricted necessarily to the jejunum. Nevertheless, low-permeable drugs should not be treated as 'unfavorable' by default; instead, therapeutic potential and suitable formulation strategies should be considered on a case-by-case basis, taking into account the overall results of in-vitro, in-vivo, and in-silico testing, throughout the entire gastrointestinal tract.

**Author Contributions:** M.M., M.Z., I.R., S.C. and A.D. worked on conceptualization, methodology, investigation, analyzed the data, and outlined the manuscript. S.C. worked on software investigation. Writing: S.C., A.D. and M.M. prepared the original draft of the article, and M.Z. and I.R. 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**


**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

© 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*
