**4.2.7 Permeation mechanism study** *in vivo*

During drug development the detailed information about the mechanism of permeation and possible efflux or metabolic instability are needed to design the structure of the desired drug and its delivery system. To get detailed information researchers have been used different methods such as: knockout or gene deficient animals for studying the effect of a specific transporter, special enzyme or transporter inhibitors (e.g. efflux inhibitors) or receptor antagonists to eliminate the desired transport effect from the study.

In order to study passive diffusion of drug candidates without interfering of other permeation mechanisms, a number of methods have been developed. For example, it is possible to use excess molar of unlabelled compound in order to saturate the transporters, enzymes or facilitated mechanisms. Also it is possible to use efflux transporters' inhibitors (e.g verapamil for P-gp). Beside these, by studying the Michaelis-Menten behaviour of drugs, it is possible to ensure that the permeation mechanism is passive diffusion (unsaturable) or not.

#### **4.2.8** *Ex vivo*

E*x vivo* experiments are developed to study drug candidates more reliably out of the body in the simulated physiologic condition (pH, temperature, buffer, nutrients, oxygen) which have the advantage of being applicable in post mortem human samples obtained by autopsy. The resulted data from these experiments have been shown acceptable correlation with *in vivo* experiments. Although in this method impossible experiments and studies in living organism can be conducted, but the differences between the living organism and the slices obtained by autopsy according to the degradation of some proteins should be take into account (Cardoso et al., 2010).

#### **4.3** *In vitro*

In order to do more rigorous investigations on the complex mechanisms occurred in endothelial cell membranes and in intracellular compartments (e.g. active and passive efflux and influx) in the BBB of a living organism, *in vitro* methods can be used. *In vitro* models of BBB should be simple, reproducible and mimic the *in vivo* conditions (both normal and pathologic). Most of the *in vitro* models of BBB are based on endothelial cells as the foundation of BBB and different animals are used to prepare cell cultures. The results should be interpret carefully because of the differentiations (the lower tightness of the developed cell lines, the phenotype modification and the absence of intercellular contact and in vivo signallings occur during the cell isolation). But it is a reliable method for high throughput screening experiments, in order to compare the penetration ability of a set of compounds (Cardoso et al., 2010). The main categories of *in vitro* models include

parameters of CNS drug candidates including half-life, Cmax, Tmax, total exposure, volume of distribution, clearance, BBB influx and efflux rates for different brain regions and most importantly the Kp,uu at steady state can be obtained and calculated using microdialysis driven data. These data can be used for pharmacodynamic studies and dosing regimens

The methods reviewed in sections 4.2.1 to 4.2.6 give information about the overall exposure

During drug development the detailed information about the mechanism of permeation and possible efflux or metabolic instability are needed to design the structure of the desired drug and its delivery system. To get detailed information researchers have been used different methods such as: knockout or gene deficient animals for studying the effect of a specific transporter, special enzyme or transporter inhibitors (e.g. efflux inhibitors) or receptor

In order to study passive diffusion of drug candidates without interfering of other permeation mechanisms, a number of methods have been developed. For example, it is possible to use excess molar of unlabelled compound in order to saturate the transporters, enzymes or facilitated mechanisms. Also it is possible to use efflux transporters' inhibitors (e.g verapamil for P-gp). Beside these, by studying the Michaelis-Menten behaviour of drugs, it is possible to ensure that the permeation mechanism is passive diffusion

E*x vivo* experiments are developed to study drug candidates more reliably out of the body in the simulated physiologic condition (pH, temperature, buffer, nutrients, oxygen) which have the advantage of being applicable in post mortem human samples obtained by autopsy. The resulted data from these experiments have been shown acceptable correlation with *in vivo* experiments. Although in this method impossible experiments and studies in living organism can be conducted, but the differences between the living organism and the slices obtained by autopsy according to the degradation of some proteins should be take

In order to do more rigorous investigations on the complex mechanisms occurred in endothelial cell membranes and in intracellular compartments (e.g. active and passive efflux and influx) in the BBB of a living organism, *in vitro* methods can be used. *In vitro* models of BBB should be simple, reproducible and mimic the *in vivo* conditions (both normal and pathologic). Most of the *in vitro* models of BBB are based on endothelial cells as the foundation of BBB and different animals are used to prepare cell cultures. The results should be interpret carefully because of the differentiations (the lower tightness of the developed cell lines, the phenotype modification and the absence of intercellular contact and in vivo signallings occur during the cell isolation). But it is a reliable method for high throughput screening experiments, in order to compare the penetration ability of a set of compounds (Cardoso et al., 2010). The main categories of *in vitro* models include

resulted from different passive or active influx and efflux systems.

antagonists to eliminate the desired transport effect from the study.

(Alivajeh & Palmer, 2010).

(unsaturable) or not.

into account (Cardoso et al., 2010).

**4.2.8** *Ex vivo*

**4.3** *In vitro*

**4.2.7 Permeation mechanism study** *in vivo*

cell based and non cell based methods. Cell based models are simplification of *in vivo* system in which the brain and non brain derived cell cultures are used to study the permeation and transport of drug candidates. The brain derived cell cultures (primary endothelial cultures) show closest phenotype to the *in vivo* brain while their preparation and handling are more difficult than non-brain derived cell lines. Primary endothelial cultures prepared by isolating animal brain micro vessels and *seeding* in culture medium where the endothelial cells grow out and make suitable mono layers for experiments. In order to mimic the *in vivo* system more closely co-cultures included astrocytes have been developed which provide more physical and physiological features in comparison with primary cell cultures (Cardoso et al., 2010). Non brain derived models use the epithelial cell cultures (e.g. Caco 2) and modified epithelial cell cultures which are used for drug absorption studies in order to rank the permeability of CNS drug candidates. Non cell based *in vitro* models include the parallel artificial membrane permeability assay (PAMPA) and immobilized artificial membranes (IAMs) which used as HPLC columns and mimic the properties of biological membrane (Abbott, 2004). PAMPA models initially developed for study passive oral absorption and successfully applied in the pharmaceutical industry. Recently, it has been modified for using in BBB permeation studies and showed good correlation with *in vivo* findings (Mensch et al., 2010).

#### **4.4 BBB permeation prediction methods (***in silico* **methods)**

*In vivo*, *ex vivo* and *in vitro* methods of assessing brain drug penetration leads to high quality data resemble most of the permeation mechanisms in BBB, but they are highly cost and time demanding and are not suitable for screening of large compound libraries. As soon as BBB studies have begun, attempts to predict the BBB permeation properties of drug candidates lead to primary structure activity relationships which later accepted as essential rules of CNS drug development. These structural features later used to develop quantitative relationships to predict the pharmacokinetic properties of CNS drugs. During years and improving the knowledge about the effect of different passive and active mechanisms of brain drug penetration, the prediction models improved and specific models to predict different aspects of BBB permeation have been developed. In order to develop a model first the prediction endpoint (dependent variable or experimental value) should be measured or obtained from the literature. The quality of these data is deterministic for developed model certainty. After selection of the data set, the inclusion of each point in data set should be evaluated and possible outliers should be determined. The next step is to split data set in training and test sets and measure or calculate the desired independent descriptors. The significant descriptors should be selected and the relationship between the dependent and independent variables should be developed using appropriate modelling method. While the model has been developed, its predictive ability along with other validation parameters should be calculated and the effect of selected descriptors on the experimental value should be defined. The details of each step are provided in following sections. Some commercial software have been developed to predict the brain drug penetration which can be used to get primary estimations about the CNS activity of a compound.

#### **4.5 Prediction endpoints (Experimental data)**

In order to get initial information about the BBB permeation of new drug entities, studying the existing information using different methods is more interesting than experimental

Blood Brain Barrier Permeation 19

Molecular weight < 400-500 Da

ClogP\* <7

logD7.4 1-3

Polar surface area < 60-70 A°2

Flexibility 1.27

pKa 7.5-10.5

\* The studies showed that logPoct/water have poorer correlation with permeation data in comparision

physiologic condition should be defined and the models should be developed accordingly (Lavenskij et

Table 3. Descriptors used in rules of five methods and their cut off points (Di, 2008; Palmer,

After preparing a number of descriptors, the best descriptor or a combination of descriptors which are able to describe the desired dependent variable (prediction end point) should be

In this method, the studied property (e.g. BBB permeation) affecting parameters should be extracted from theoretical findings (several processes include in the overall result) and convert to mathematical representations. The provided descriptors depend on their effects (positive or negative, direct or inverse) on desired property should be correlated to the prediction end point and the resulted equation could be used for prediction purposes

It is so important to exclude insignificant descriptors to prevent over fitting and biased results using a descriptor selection method. The number of descriptors depends on the modelling method. For simple multivariate regression methods, the number of descriptors depends on the number of data points, while for partial least square and principal component analyses methods it is not limited. In addition to the number of the descriptors

N+O <6

with ΔlogP or logD7.4. Recent studies showed that the ionization state of drug candidates in

selected. There are two approaches for descriptor selection:

al., 2009, 2010; Shayanfar et al., 2011).

**4.7 Model development** 

**4.7.1 Mechanistic approach** 

(Lavenskij et al., 2010).

**4.7.2 Statistical approach** 

2010)

Rotatable bonds <8

H bond donor <3

H bond acceptor <7

Property The cutoff for BBB permeation

measurement. There are different (*in vivo* or *in vitro*) indicators which are able to evaluate the rate or extent of drug permeation to the BBB (see section 4.1.3). Among them logBB values have been used extensively for *in silico* methods in order to predict the extent of drug penetration to the brain and the related data sets can be found in the literature. Unbound drug fraction, logPS and BUI% have been used to develop the prediction methods, while some researchers used *in vitro* data (e.g. PAMPA derived P-gp binding affinity) for their studies (Dagenais et al., 2009). Beside these BBB+/- and CNS+/- data which have been extracted from logBB experiments and implications of brain disorders or targets about primary site of action of compounds respectively, were utilized for classification purposes (Klon, 2009). It seems that using the combined information derived from different indicators will be more useful than individual ones. The quality of selected data set should be considered according to the experimental method which used to obtain it (data set homogenesity). The homogenesity of logBB data sets have been questioned, but the studies showed that these combined data sets are applicable. Also the outliers should be determined using statistical methods or according to the experimental method. One of the most common statistical methods is to compute deviations of a single data point from mean dependent or independent variables or both of them and exclude highly deviated datum. In fact an applicability domain for each prediction method should be defined and the compounds out of this domain should be excluded from analyses. For experimental procedures it should be kept in mind that if special efflux inhibitors are used or not. In some methods, scientists are used unlabeled substrate to saturate the desired enzyme or transporter or receptor and the resulted data from these experiments should not be combined with others (Lavnevskij et al., 2010). The third point which should be kept in mind is that the number of the data points should be enough for developing statistical properties (e.g. regression coefficients) of the developed model and also for excluding a part of data as test set. If it is not possible the prediction capability of developed model cannot be evaluated and it will be applicable for the entire data set.

#### **4.6 Descriptors**

The structural features and physicochemical properties (Table 2) of the studied compounds should be extracted using the available experimental and computational methods (commercial software, fragment based methods, …). The most studied and evaluated descriptors to define the BBB permeation are those related with passive diffusion. Table 3 contains the details of most frequently used descriptors as well as their effects on BBB permeation. As can be seen from the table, the overall findings about the structural features (also known as the rule of five) of the CNS drug candidates are:


It should be noted that these rules should be used cautiously during drug design procedure. For example, although high lipophilicity increase the permeation rate but it causes the poor solubility, metabolic instability and higher membrane bounding which are not suitable properties for a drug candidate.


Table 2. Frequently used descriptors and software.

measurement. There are different (*in vivo* or *in vitro*) indicators which are able to evaluate the rate or extent of drug permeation to the BBB (see section 4.1.3). Among them logBB values have been used extensively for *in silico* methods in order to predict the extent of drug penetration to the brain and the related data sets can be found in the literature. Unbound drug fraction, logPS and BUI% have been used to develop the prediction methods, while some researchers used *in vitro* data (e.g. PAMPA derived P-gp binding affinity) for their studies (Dagenais et al., 2009). Beside these BBB+/- and CNS+/- data which have been extracted from logBB experiments and implications of brain disorders or targets about primary site of action of compounds respectively, were utilized for classification purposes (Klon, 2009). It seems that using the combined information derived from different indicators will be more useful than individual ones. The quality of selected data set should be considered according to the experimental method which used to obtain it (data set homogenesity). The homogenesity of logBB data sets have been questioned, but the studies showed that these combined data sets are applicable. Also the outliers should be determined using statistical methods or according to the experimental method. One of the most common statistical methods is to compute deviations of a single data point from mean dependent or independent variables or both of them and exclude highly deviated datum. In fact an applicability domain for each prediction method should be defined and the compounds out of this domain should be excluded from analyses. For experimental procedures it should be kept in mind that if special efflux inhibitors are used or not. In some methods, scientists are used unlabeled substrate to saturate the desired enzyme or transporter or receptor and the resulted data from these experiments should not be combined with others (Lavnevskij et al., 2010). The third point which should be kept in mind is that the number of the data points should be enough for developing statistical properties (e.g. regression coefficients) of the developed model and also for excluding a part of data as test set. If it is not possible the prediction capability of developed model cannot be

The structural features and physicochemical properties (Table 2) of the studied compounds should be extracted using the available experimental and computational methods (commercial software, fragment based methods, …). The most studied and evaluated descriptors to define the BBB permeation are those related with passive diffusion. Table 3 contains the details of most frequently used descriptors as well as their effects on BBB permeation. As can be seen from the table, the overall findings about the structural features

It should be noted that these rules should be used cautiously during drug design procedure. For example, although high lipophilicity increase the permeation rate but it causes the poor solubility, metabolic instability and higher membrane bounding which are not suitable

Descriptor Topological descriptors Constitutional, Molecular properties, Quantum

chemical, ACDLabs, free aqueous solubility energy Software Absolve, Dragon, Hyperchem, Volsurf, MOE, Cerius package

evaluated and it will be applicable for the entire data set.

(also known as the rule of five) of the CNS drug candidates are:

**4.6 Descriptors** 


properties for a drug candidate.

Table 2. Frequently used descriptors and software.


\* The studies showed that logPoct/water have poorer correlation with permeation data in comparision with ΔlogP or logD7.4. Recent studies showed that the ionization state of drug candidates in physiologic condition should be defined and the models should be developed accordingly (Lavenskij et al., 2009, 2010; Shayanfar et al., 2011).

Table 3. Descriptors used in rules of five methods and their cut off points (Di, 2008; Palmer, 2010)
