"In silico" prediction of blood-brain barrier permeation and P-glycoprotein activity

P-glycoprotein is an ATP-dependent efflux transport protein which is highly
expressed in many human tissues such as the intestinal epithelium and the
blood-brain barrier, and is over-expressed in many cancer cells.1 This transporter
carries a wide variety of chemically unrelated compounds. It binds
them within the cell lipid membrane, and flips them to the outer leaflet or
exports them to the extracellular medium.2 Since P-glycoprotein affects the
distribution of many drugs, assessing the interactions between drugs and Pglycoprotein
at an early stage of drug development is important.
It has been shown that the binding of a drug to the transporter occurs in a
two-step process.3{5 (i) The drug partitions from the extracellular environment
to the lipid membrane, and after diffusion to the inner cytosolic leaflet of
the bilayer, (ii) it binds to P-glycoprotein most likely via
hydrogen bond formation.
Different methods have been used to assess the lipid-water partition coefficient, such as isothermal titration calorimetry, and lipid monolayer insertion
measurements. However, the lipid-water partition coefficient depends on the
lipid used, and in turn on the lateral packing density of the lipid layer. Therefore
an approach based on surface activity measurements was developed, which
allows the prediction of the lipid-water partition coe�cient for membranes of
different lateral packing densities.7 Measurements of the surface pressure of
the drug in buffer solution as a function of concentration (Gibbs adsorption
isotherm) yields the air-water partition coefficient (Kaw), the critical micellar
concentration (CMC), and the cross-sectional area of the compound (AD),
provided experiments are performed under conditions of minimal electrostatic
repulsion. Since air has a dielectric constant close to that of the lipid core
region of a membrane, there is a direct relationship between the partition
of a drug into the air-water interface, and the partition into the lipid-water
interface.8 The cross-sectional area, as well as the lipid-water partition coefficient (and by extension the air-water partition coefficient), are thus crucial
parameters to assess the binding and diffusion of a drug into a lipid bilayer.
In a first part of the thesis, I focused on the membrane binding step. Since
the cross-sectional area of a compound is a crucial parameter for drug partitioning
into the lipid bilayer, the quality of the data obtained by mean of surface
activity measurements are most important. For this purpose, in a first step,
I improved the calibration of the experimental settings, by assessing several
factors like the evaporation or the solvent effect. In a second step, I developed
computer routines for unbiased evaluation of these measurements. In a third
step, I developed an algorithm to calculate the cross-sectional area of a compound
oriented at a hydrophilic-hydrophobic interface; this algorithm has been
calibrated on a set of measured data, in order to find from a conformational
ensemble the conformation of the membrane-bound drug.
In a second part of the thesis, I focused on the binding of a drug to
P-glycoprotein. P-glycoprotein is monitored essentially by three types of assays,
(i) the measurement of ATP hydrolysis activity of the transporter, (ii)
a competition assay against calcein-AM, and (iii) a transcellular transport assay
through polarized P-glycoprotein over-expressing cell monolayer. Based
on a modular binding approach to assess the two-step binding of a drug to
P-glycoprotein (Figure 1),5 I developed several rules to predict the outcome
of these experimental assays. Each rule, predicting one particular assay, has
been tested on experimental datasets.
In a third part of the thesis, I developed a working interface to handle
multiple structures of compounds, to calculate the new descriptors involved in
the two-step binding of drugs to P-glycoprotein (membrane partitioning, and
binding to the transporter), and to calculate the outcome of the prediction
rules. Moreover the working interface has been designed in a way the user can
easily define new rules, or even introduce a new multidrug transporter (e.g.
the multidrug transporter MRP1).
Starting from well characterized physical-chemical parameters, I developed
a coherent ensemble of descriptors to assess by a rule-based approach the
thermodynamics and kinetics of P-glycoprotein activation. This ensemble has
been embedded in a customizable working interface, allowing easy evaluation
of the in silico predictions.