Dnr
In silico and in vitro determination of substrate
specificity for Breast Cancer Resistance Protein
(BCRP) transporter at the blood-brain barrier
Fen Wang
Degree Project in Pharmaceutical Modelling within Pharmaceutics and
Biopharmaceutics, 45.0 hp, 3FG001, 2020 - 2021
Supervisor: Maria Karlgren, Per Larsson Examiner: Per Artursson
Table of Contents
Abstract ...1
Introduction ...2
Aim ...6
Materials and Methods ...6
Cell culturing ... 6
Permeability/Transport experiments ... 6
LC-MS/MS analysis ... 7
Statistics ... 7
In silico model building ... 7
Parametrization for selected substrate and non-substrate molecules ... 8
Simulation and assessment of transporter substrate specificity ... 9
Results ...9
In vitro ... 9
CG model of BCRP ... 13
CG-BCRP Transporter imbedded in POPC bilayer ... 14
Parametrization of drug molecules ... 15
Discussion and Conclusions...20
Evaluation of the in vitro results ... 20
Some notes on the marker (enalaprilat)... 21
Novelty of the in silico model ... 22
Interpretation of the retention of Dantrolene in TMD... 22
Future improvement of the model system ... 23
Conclusions ... 23
References ...24
Appendix. Supplement tables and figures...26
A1. Additional information on BCRP transporter models ...26
Supplement Notes: BCRP transporter model building and optimization ... 26
A2. Validation of the parameterization of drug CG models...33
Abstract
Background
The Breast Cancer Resistance Protein (BCRP) drug transporter is important for drug disposition and plays a critical role in regulating drug entry into the brain. Its substrate spectrum overlaps with substrates of Multi Drug Resistance Protein 1 (MDR1, P-gp), which influences and complicates the interpretation of data on drug distribution into tissues (e.g. brain). Distinguishing BCRP mediated transport from the transport by the MDR1 is often problematic. However, with new in vitro tools, this is now possible. In this project, two drug compounds, i.e. Dantrolene and Ritonavir, were investigated using these new in vitro models.The results from the experimental in vitro assay were matched with molecular dynamics (MD) simulations. Using coarse-grained (CG) simulations, a model of the BCRP transporter in a lipid bilayer was built, this model is based on the human BCRP structure revealed by Taylor et al (2017). Simulations were run for Dantrolene (a known substrate of BCRP) independently three times, and another with Ritonavir (a non-substrate) three times.
Aim
To determine substrate specificity for the BCRP transporter for two compounds, and to construct a CG model of BCRP transporter to see whether in silico methods can be used as an alternative for assessing substrate specificity.
Methods
Madin-Darby canine kidney (MDCK) II cell line with no endogenous canine MDR1 (cMDR1) expression (MDCKcMDR1-KO), overexpressing human MDR1 (hMDR1) (MDCK-hMDR1cMDR1-KO) and stable expression of human BCRP (hBCRP) (MDCK-hBCRPcMDR1-KO) cells were cultured and used in Transwell experiments. Samples were analyzed using LC-MS/MS to determine the substrate concentrations. Apparent permeability and efflux ratio was calculated and evaluated.
and non-substrate molecules. And visual inspection was done with the visual molecular dynamics (VMD) program and PyMOL.
Results
In vitro transport experiment confirmed that Dantrolene is a BCRP specific substrate, and
Ritonavir is MDR1 specific substrate. Following simulations of these two compounds, Dantrolene is observed to stay in the transmembrane domains (TMD) for a certain period (on average several hundreds of nanoseconds), while Ritonavir is not found to bind in the TMD, which provides a proof of concept for future studies.
Introduction
The blood-brain barrier (BBB), which consists of the cerebral capillary endothelial cells connected via tight junctions present throughout the brain parenchyma, is an active organ and has important functions for brain homeostasis and protection (Hammarlund-Udenaes
et al., 2008; Abbott, 2014). Drug compounds need to pass the luminal/apical and
abluminal/basolateral membranes of the endothelial cells to transverse the BBB (Hammarlund-Udenaes et al., 2008). There are numerous transporters governing influx and efflux of endogenous and exogenous compounds across BBB: solute carriers (SLCs) mediating the entry of major nutrients and efflux from the brain of some metabolites; ATP-binding cassette (ABC) transporters restrict the brain distribution of xenobiotics such as drugs by pumping them back to circulating blood. Among the ABC efflux transporters, Multidrug Resistance Protein 1 (MDR1, P-glycoprotein, P-gp; gene name:
ABCB1) and Breast Cancer Resistance Protein (BCRP; gene name: ABCG2) are the
dominant efflux transporters on the apical membrane, especially MDR1 in rodents and BCRP in primates. And the absolute expression level of BCRP is higher than that of MDR1 at the human BBB (Shawahna et al., 2011; Uchida et al., 2011), it is noteworthy that BCRP functions as a homodimer (Abbott, 2014).
central nervous system (CNS) drugs with knowledge that could explain clinical consequences, such as increased toxicity or altered efficacy. This also meets the U.S. Food and Drug Administration (FDA, 2020) guidance for industry: investigational drugs should be evaluated in vitro to determine whether they are substrates of MDR1 and/or BCRP, highly permeable and highly soluble drugs could be exempt unless there are potential safety concerns of drug distributing into tissues e.g. brain.
Distinguishing BCRP mediated transport from transport by the MDR1 transporter is often problematic, there is a general problem for the current frequently used in vitro models including colon adenocarcinoma cell line Caco-2, Madin-Darby canine kidney II (MDCK) and pig kidney cell line LLC-PK: endogenous expression of other ABC transporters, especially MDR1, interferes the interpretation of BCRP-mediated transport (Karlgren et
al., 2017; Wegler et al., 2020). However, novel in vitro models with advantage of absence
of interfering canine MDR1 (cMDR1) are now available. Firstly, MDCKcMDR1-KO is obtained by knocking out cMDR1 in an MDCK wildtype cell line with a gene-editing technology called CRISPR-Cas9 (Simoff et al., 2016). Then, MDCK-hMDR1cMDR1-KO is obtained by knocking out cMDR1 in an MDCK cell line already overexpressing human MDR1 (hMDR1) using CRISPR-Cas9 as well (Karlgren et al., 2017). Very recently, a stable MDCK cell line overexpressing human BCRP (hBCRP) with no endogenous cMDR1 (MDCK-hBCRPcMDR1-KO) is available, too (Wegler et al., 2020).With these three improved in vitro models, drug compound would more easily be identified to be substrates of BCRP and/or MDR1. In this project, two drug compounds, i.e. Dantrolene and Ritonavir, are investigated using these new models, the results are compared to the previously reported data (Wegler et al., 2020).
Molecular dynamics simulations
al., 2018). Even though obtaining the timescale of the entire transport process is still a
major challenge, MD simulations have been successfully used to describe the transition steps in various transport cycles (Loschwitz et al., 2020).
Among the different MD simulation techniques, coarse-grained (CG) is bridging the traditional all-atom (AA) models to the continuum scale. In contrast to AA simulations, for CG-based simulations, several atoms have been merged into beads, so atomistic resolution is lost and conformational landscapes of molecular system can be more effectively sampled. Besides, CG simulations have a much greater integration steps, e.g. Martini force field (FF) can use much larger time steps (e.g. 20~30 fs) than AA which typically use 1~2 fs. Due to the versatility and acceptable accuracy of Martini model at reproducing experimental and atomistic MD data, it has rapidly become the most popular CG lipid FF (Loschwitz et al., 2020). The open beta of Martini 3 released in 2018 has overcome the shortcomings of the previous martini 2 (e.g. certain molecules tend to interact too strongly), now with new bead types and sizes, it can cover broader chemical space and has an improved interaction balance, e.g. more hydrogen bonding capabilities, smoother transitions between the beads (Souza et al., 2021).
model of the TMDs. The NBDs of BCRP were resolved at lower resolution, thus a homology model of BCRP NBD based on structure of ABCG5/G8 was built, it was then docked into the EM density map and modifications to the secondary structure elements were fitted (Taylor et al., 2017). The antigen-binding fragments of 5D3 (5D3-Fab) was added to facilitate the determination of high-resolution structure and it is found not altering the interaction between BCRP and substrate estrone-3-sulfate (Taylor et al., 2017; Manolaridis et al., 2018). Without conformation specific antibody fragments or ligands, BCRP transporter adopts a closed conformation with substantial structural rearrangement of two critical transmembrane helices, which is highly unanticipated. Cyro-EM and biochemical studies demonstrated that binding of three chemotherapeutic compounds, respectively, opened the closed conformation but with different effects. Nevertheless, BCRP adopts an apo-closed conformation in the rest state which contrasts with MDR1 and MRP1 while these two ABC transporters adopt rather open conformations in the TMDs and NBDs at rest state. It is proposed that BCRP selects its substrate by sensing whether it can effectively shift from the apo-closed conformation to inward-facing state (Orlando and Liao, 2020). To the best of my knowledge, the available BCRP models are manually adjusted (including shift, rotation, modification and refinement) in COOT (Crystallographic Object -Oriented Toolkit) to fit the cyro-EM map (Taylor et al., 2017; Manolaridis et al., 2018; Orlando and Liao, 2020) , or coarse-grain MD of known inhibitor bound to BCRP to reveal the conformational dynamics of TMD and dynamics of the inward- to outward-facing switch (Khunweeraphong et al., 2019), it is the first attempt to use Martini 3 (Souza et al., 2021) to marinized substrate or non-substrate and simulate in a CG BCRP transporter in apo-closed conformation imbedded in a POPC model membrane.
et al., 2020).Here the retention in the TMD for Dantrolene and Ritonavir was simulated and compared to the obtained in vitro data.
Aim
To determine substrate specificity for the BCRP transporter for two of model compounds, and to construct a CG model of BCRP transporter to see whether in silico methods can be used as an alternative for assessing substrate specificity.
Materials and Methods
Cell culturingMDCKcMDR1-KO, MDCK-hMDR1cMDR1-KO and MDCK-hBCRPcMDR1-KO cells were cultured (by Dr Ivailo Simoff, Department of Pharmacy) as previously described (Wegler
et al., 2020). Cell culture media and supplements were purchased from ThermoFisher Scientific (Waltham, MA, USA) or Sigma-Aldrich (St. Louis,MO). The cells were cultured at 37°C, 95% humidity and 5% CO2 and sub-cultured twice a week1.
Permeability/Transport experiments
For Transwell transport experiments, 5 x 105 cells were seeded on 12 mm, 0.4 μm Transwell membrane inserts (Corning, Amsterdam, Netherlands). For cells pre-treated with sodium butyrate, in this case, specifically BCRP cells, culturing medium supplemented with 10 mM sterile-filtered sodium butyrate were added to the cells 24 h prior to the transport experiment (Wegler et al., 2020). Four days after seeding, Transwell transport experiments were performed as previously described (Simoff et al., 2016) and MDCKcMDR1-KO was used as control.
The apparent permeability coefficient (Papp, unit: cm s-1) was calculated according to Eq.1:
𝑃𝑎𝑝𝑝 = (𝑑𝑄/𝑑𝑡)/(𝐴 × 𝐶0) 𝐸𝑞. 1
1 Cell culture by myself was the original plan but it didn’t work out due to Covid-19 pandemic, high appreciation to
Where dQ/dt is the steady-state flux (μmol s-1), A is the surface area of the filter (cm2) and C0 is the initial concentration in the donor chamber (μM).
The efflux ratio (ER), defined as the ratio of the secretory permeability and the absorptive permeability (see Eq. 2), is the simplest approach for identifying efflux substrates in various tissue barriers including BBB. Transepithelial electrical resistance (TEER) were also measured before and after the transport experiments to verify the integrity of the cell monolayers (Hubatsch, Ragnarsson and Artursson, 2007).
𝐸𝑅 = 𝑃𝑎𝑝𝑝,𝑏𝑎/𝑃𝑎𝑝𝑝,𝑎𝑏 𝐸𝑞. 2 LC-MS/MS analysis
Substrate concentrations were analyzed by LC-MS/MS using an Acquity UPLC coupled to a XEVO TQ triple-quadrupole (both from Waters Corp, Milford, MA). The separation was done with a C18 BEH 1.7 μm column (from Waters Corp., Milford MA) and with a mobile phase that consisted of acetonitrile and formic acid in purified water. The flow rate was set on 0.5 mL/min and the injection volume of the samples was 5 μL. Warfarin was used as internal standard.
Statistics
Two independent transport experiments have been run for MDCK-hBCRPcMDR1-KO, one for MDCKcMDR1-KO and one for MDCK-hMDR1cMDR1-KO. All experiments were performed in both
directions in triplicate, mean and standard deviation (SD) of apparent permeability were calculated. Mean of efflux ratio were calculated by taking the average of three efflux ratios obtained from the pairwise secretory permeability and the absorptive permeability. Student t-test was used for statistical comparisons.
In silico model building
lipid bilayer, as POPC is a well-studied phospholipid that is often used to represent a major component of eukaryotic cell membranes.
Structure of ligand free and apo-closed BCRP (PDB: 6VXF) was used as an initial template for model building and refinement (see Fig. 1). MODELLER was used to fill the missing residues by treating the original structure as a template and building a comparative model using the full sequence (Webster, 2000). Then transporter was converted to a CG-representation with secondary structure added by martinize.py (De Jong et al., 2013).
Fig. 1.Cartoon representation of BCRP transporter (a homodimer, PDB: 6vxf).
The BCRP transporter was then inserted into the membrane bilayer by insane.py script (Wassenaar et al., 2015), but the initial output system was not ideal, thus the transporter was shifted with respect to membrane for a couple of different distances and visualized in PyMOL (version 2.4.1), the purpose was to ensure that its presence inside the membrane fit the assembly elucidated by Taylor et al (2017). The periodic boundary condition (pbc) box size is 17.5×15.5×19.5 (nm3).
Parametrization for selected substrate and non-substrate molecules
Tirado-Rives, 2005; Dodda, De Vaca, et al., 2017; Dodda, Vilseck, et al., 2017). Initial AA-to-CG index files were obtained with the AA-to-CGbuilder tool (J. Barnoud, https://jbarnoud.github.io/cgbuilder/). Then all parameters of bonded interactions (bonds, constraints, angles and etc.) were set as 0, an automatic parameterization software (Swarm-CG) was used to optimize bonded parameters, the optimization was undergone three successive cycles: 1) optimizing constraints/bonds & angles; 2) optimizing angles & dihedrals; 3) all parameters are refined altogether (Empereur-Mot et al., 2020). Simulation and assessment of transporter substrate specificity
For in silico experiment, initially one simulation was run for Model01 with different compounds, and same for Model02 (see details of the models in appendix: A1. Additional information on BCRP transporter models and A3. Summary of the simulations). Upon model being optimized and acceptable, three separate simulations were run for Model03 with Dantrolene and same as for Ritonavir (see details in appendixA3. Summary of the simulations). After the simulations, inspection and analysis were performed by VMD, PyMOL etc., root mean square deviation (RMSD) and root mean square fluctuation (RMSF, i.e. standard deviation) of the backbone CG beads’ positions in the trajectory were calculated.
Results
In vitroTable 1. Efflux Ratio of drugs in different cell lines
Drug Item MDCKcMDR1-KO
MDCK-hMDR1cMDR1-KO MDCK-hBCRPcMDR1-KO* Dantrolene 1 0.6 0.6 8.1 2 0.6 0.7 8.2 3 0.6 0.7 7.9 Mean 0.6 0.7 8.1 SD1 0.03 0.05 0.60 Literature2 0.5 0.6 9.4 Ritonavir 1 1.5 107.9 0.2 2 1.5 96.8 0.2 3 1.9 53.0 0.2 Mean 1.6 85.9 0.2 SD1 0.18 26.50 0.01 Literature2 0.8 362.4 0.5
* Only one representative experiment is shown. 1 SD: standard deviation
2 Source data from M. Karlgren; corresponding ERs published in previous studies (Karlgren et al., 2017; Wegler et al., 2020).
Fig. 2 Efflux ratios (ERs) of Dantrolene and Ritonavir in different cell lines: MDCK-hBCRPcMDR1-KO (BCRP), MDCKcMDR1-KO (CL2); MDCK-hMDR1cMDR1-KO (CL10) with comparison to literature (Wegler et al., 2020) reported results. Dotted line indicates an efflux ratio of 1. Data are presented as mean for one representative experiment performed in triplicate. Significant differences from the MDCK-hBCRPcMDR1-KO control were determined using Student’s t-test. Differences between MDCK-hBCRPcMDR1-KO and MDCKcMDR1-KO are indicated with *, p < 0.05. Differences between MDCK-hMDR1cMDR1-KO and MDCKcMDR1-KO are indicated with +, p < 0.05.
*
Cumulative fraction transported versus time for individual filters with linear regression lines are shown in Fig. 3.
Fig. 3 Cumulative fraction transported versus time for individual filters with linear regression lines.
MDCKcMDR1-KO (CL2); MDCK-hMDR1cMDR1-KO (CL10); MDCK-hBCRPcMDR1-KO (BCRP). g, h, i sub-figures are the marker enalaprilat corresponding to a, b, c of test compound Dantrolene. j, k, l sub-figures are the marker enalaprilat corresponding to figures d, e, f of test compound Ritonavir.
The apparent permeability of test compounds is summarized in Table 2. Since enalaprilat is a new marker and an in-house threshold of acceptable apparent permeability (Papp) in MDCK cell lines on 12-well filters is not available yet, the Papp values and the standard deviation (SD) of triplicate runs of this marker are also presented in Table 2.
Table 2. The apparent permeability of test compounds and the marker (enalaprilat) Compound Item BCRP
1 (10-6 cm/s) CL101 (10-6 cm/s) CL21 (10-6 cm/s)
Papp AB2 Papp BA2 Papp AB Papp BA Papp AB Papp BA
Dantrolene 1 18.94 154.33 117.24 72.40 110.95 72.00 2 17.07 139.64 116.26 78.95 115.92 66.99 3 18.62 147.54 103.71 77.48 111.82 68.79 Mean 18.21 147.17 112.41 76.27 112.90 69.26 SD 1.00 7.35 7.54 3.44 2.66 2.53 Literature3 22.74 210.74 142.11 91.00 196.65 90.69 Ritonavir 1 321.74 64.66 1.48 159.40 61.84 92.43 2 289.85 64.92 1.40 135.89 66.22 99.54 3 293.12 65.30 2.37 125.45 53.45 99.07 Mean 301.57 64.96 1.75 140.25 60.50 97.01 SD 17.50 0.32 0.54 17.40 6.49 3.98 Literature3 131.20 65.03 0.38 139.25 110.14 92.11 Enalaprilat (Dantrolene) 1 2.54 1.39 7.76 0.56 1.82 3.44 2 1.89 0.83 1.66 0.79 2.89 0.60 3 2.41 0.97 1.55 0.60 1.90 1.64 Mean 2.28 1.06 3.66 0.65 2.21 1.89 SD 0.35 0.29 3.55 0.12 0.60 1.44 Enalaprilat (Ritonavir) 1 12.26 3.01 10.86* 3.58* 10.94* 3.73* 2 4.64 2.26 10.91* 3.59* 10.95* 3.57* 3 3.92 2.09 11.91* 7.52* 10.86* 3.67* Mean 6.94 2.45 11.23 4.90 10.91 3.66 SD 4.62 0.49 0.59 2.27 0.05 0.12
Note: 1 MDCK-hBCRPcMDR1-KO (BCRP); MDCK-hMDR1cMDR1-KO (CL10); MDCKcMDR1-KO (CL2). 2 P
app AB: the apparent absorptive permeability; Papp BA: the apparent secretory permeability. 3 Source data from M. Karlgren; corresponding ERs published in previous studies (Karlgren et al., 2017; Wegler et al., 2020).
CG model of BCRP
Fig. 4. Different representations of the BCRP transporter (Model03): a. Cartoon representation with
missing residues added by Modeller. b. Coarse-grained BCRP by Martini 3, Gromacs 2020.4.
CG-BCRP Transporter imbedded in POPC bilayer
Model01, Model02 and Model03 were all inserted in POPC bilayer, and the model
BCRP’s presence inside the membrane fit the assembly elucidated by Taylor et al (2017), a comparison of literature and the Model03 is illustrated in Fig. 5.
Fig. 5 Comparison of BCRP & membrane assembly reported and model built in this project: a. Ribbon
diagram of BCRP–5D3 (Fab) complex (picture is from Taylor et al (2017)). b. CG-BCRP (Model03, colored in orange) imbedded in POPC bilayer (beads in blue, white and green) in box of non-polarized water (marine blue sticks).
a b
Membrane Extracellular
Cytoplasm
Parametrization of drug molecules
To capture the symmetry of the molecule and the best shape, normal size beads (N-beads) for linear 4-1 arrangements, small beads (S-beads) for aliphatic rings, tiny beads (T-beads) for aromatic rings. The atom-to-bead mapping of Dantrolene is shown in Fig. 6, the mapping of Ritonavir is shown in Fig. 7.
Fig. 6 Atom-to-bead mapping of Dantrolene
Fig. 7 Atom-to-bead mapping of Ritonavir
Simulations of drug molecules in the BCRP and lipid bilayer model system
membrane Extracellular Cytoplasm Ritonavir BCRP a 19.5 nm 15.5nm 17.5nm
Fig. 8 Side view of BCRP in CG model with Ritonavir (yellow): a. Simulation setup of the BCRP (blue beads) into a POPC bilayer (Green for the PO4 head groups and grey for the lipophilic tails) in water (light blue). b. to f. Five snapshots of simulation as time goes (from 1μs till 5μs), for clarity, only Ritonavir
b c
d e f
at 0 ns
at 1000 ns at 2000 ns
17.5nm 15.5nm 19.5nm BCRP membrane Extracellular Cytoplasm Dantrolene a b c d e f at 2646 ns at 2716 ns at 2786 ns at 2842 ns at 2850 ns at 2858 ns
Fig. 9 Side view of BCRP in CG model with the retention of Dantrolene (orange) in TMD or movement to extracellular: a. Simulation setup of the BCRP (blue
Motion/fluctuations of the transporter during simulations
Both time series of RMSD of model system with Dantrolene and with Ritonavir indicate that the transporter structure is stable during MD simulation. And fluctuations appear to be greater for some extent when comparing the simulation with Dantrolene and simulation with Ritonavir (see Fig. 10). From principal component analysis (PCA), when projecting the motion (eigenvectors) of the ritonavir simulation onto the corresponding eigenvectors of the dantrolene simulation, there appears to be some differences in how the channel is sampling along eigenvectors for the first two PCs (see 2D projection of the PC1 & PC2 in Fig.11 a for chain A & Fig.11 b for chain B). Based on further investigation, the spectrum of RMSF of the transporter backbone shows that more blue beads (moving less) are in the ritonavir simulation than that with dantrolene (see Fig. 11 c & d).
Fig. 10 a. RMSD of backbone of Chain A of Model03 vs time, a comparison between simulations
(Dantrolene in red vs Ritonavir in cyan). b. RMSD of backbone of Chain B of Model03 vs time, a comparison between simulations (Dantrolene in red vs Ritonavir in cyan).
a
Fig. 11 a. 2D Projection of trajectory of Chain A of Model03 simulations, a comparison between
Dantrolene simulation (in red) and Ritonavir simulation (in cyan). b. 2D Projection of trajectory of Chain B of Model03 simulations. c. Spectrum of RMSF of BCRP backbone in Model03 with Dantrolene (minimum RMSF is in dark blue, maximum RMSF is in dark red). d. Spectrum of RMSF of BCRP backbone in Model03 with Ritonavir.
Discussion and Conclusions
The primary aim of in vitro experiment has been realized and progress has also been made for the in silico model, following are the detailed discussions from both perspectives. Evaluation of the in vitro results
The overall apparent permeability and efflux ratios obtained are comparable to the previous studies (Karlgren et al., 2017; Wegler et al., 2020) (see Table 1 and Table 2 for results of this project and source data of previous studies). For Dantrolene, very close ER is observed in MDCK-hBCRPcMDR1-KO. For Ritonavir, though ER differs to some extent
in MDCK-hMDR1cMDR1-KO, which is due to the fact that when the divisor is very small, the
quotient (ER) could differ a lot, significant difference could still be found between MDCK-hMDR1cMDR1-KO and MDCKcMDR1-KO (see Fig. 2).
One exception from the literature data is the absorptive permeability of Ritonavir tested in MDCK-hBCRPcMDR1-KO (see Table 2), here having a Papp of 301 ×10-6 cm/s, whereas in Wegler et al. the Papp was instead 131 ×10-6 cm/s. For example, AB_filter 1 with Ritonavir’s permeability of 321.74 ×10-6 cm/s, it may be leaky as indicated by the marker enalaprilat (Fig. 3j), while the Ritonavir data for AB_filter 1 overlaps with filter 2 and 3 (Fig. 3d). Furthermore, all Ritonavir filters at 45 min also seems leaky. Hence, those data points were here considered not reliable and were excluded from the analysis. By excluding the 45 min samples, we obtained the Papp shown in Table 2. While excluding these values resulted in a reduced Papp, it still cannot lower the Papp to the same level of the source data used by Wegler et al. It should be noted that the standard curve for ritonavir was problematic, and if time allowed, the ritonavir samples would have been rerun with a new standard curve to improve quantification. This may have affected the Papp calculations and may provide one explanation to the high value obtained as compared to the literature (Karlgren et al., 2017; Wegler et al., 2020).
Some notes on the marker (enalaprilat)
is overlapping with the other two filters (see Fig. 3b), thus it appears not affect permeability of Dantrolene. Similar situation for the BA_filter 3 of Ritonavir in MDCK-hMDR1cMDR1-KO enalaprilat seems leaky at 45min (with value of 0.06 which is out of the scale of plot, thus not shown in the Fig. 3k), but Ritonavir’s linear regression line is very close to the other two filters (see Fig. 3e). As these observations are from very limited number of experiments, thus it needs to be further studied and confirmed in additional experiments in MDCK cells.
Novelty of the in silico model
As obtaining the timescale of the entire transport process is still a major challenge (Loschwitz et al., 2020), we managed to observe part of the transporting. Compared to the existing manually adjusted (including shift, rotation, modification, and refinement) models or CG models to study the conformation change from inward-facing to outward-facing (Taylor et al., 2017; Manolaridis et al., 2018; Khunweeraphong et al., 2019; Orlando and Liao, 2020), it is the first attempted CG BCRP model in apo-closed conformation using the latest martini force field. The Martini 3 has overcome the shortcomings of the previous martini 2 (e.g. certain molecules tend to interact too strongly), now with new bead types and sizes, it can cover broader chemical space and has an improved interaction balance, e.g. more hydrogen bonding capabilities, smoother transitions between the beads (Souza et al., 2021).”
Interpretation of the retention of Dantrolene in TMD
As one of the limitations of martini force field is that friction from atomistic degrees of freedom is missing, thus this preliminary observation of the retention may be more qualitatively a correct transition pathway, timescale should be interpreted with care. Future improvement of the model system
The model (Model03) built in this project has the advantage of using the latest Gromacs and martini 3 force field to study the BCRP in an apo-closed conformation that is what the transporter is at the resting state (Orlando and Liao, 2020). In comparison to actual complex biological system where ATP is necessary for ABC transporters such as BCRP to translocate substrates, ATP is missing in this in silico model, it could be improved for later studies. The membrane could also be improved to include cholesterol as well. MD is statistical mechanics, three independent simulations were run, it still may be individual event, more simulations with an improved model which contains ATP and extends to more compounds would be necessary for future studies to show significance of the ability to discriminate between substrates and non-substrates. Besides, similarly like in vitro models, in silico control models would be needed as well to screen BCRP specific substrate or shared substrate with MDR1. However, the attempt in silico model in this project has set well enough starting point for further exploration of BCRP substrate specificity.
Conclusions
In summary, for future studies, it is worth to explore more shared or specific substrates of BCRP and MDR1 by the new in vitro models used in this project, especially for CNS drugs previously reported or suspected to be BCRP substrates but cannot be discriminated by animal models (e.g. MDR1 knockout, BCRP knockout and combined MDR1/BCRP knockout mice) or other in vitro models (e.g. ZFN-mediated knockout of endogenous cMDR1 expression in an MDCK cell line overexpressing hBCRP) (Gartzke et al., 2015). And then use these compounds to perform cross validation for the future improved in
References
Abbott, N. J. (2014) Physiology and Basic Principles for Drug Handling by the Brain Anatomy and Physiology of the Blood – Brain Barriers. Edited by M. Hammarlund-Udenaes. doi: 10.1007/978-1-4614-9105-7.
Agarwal, S. and Elmquist, W. F. (2012) ‘Insight into the cooperation of P-glycoprotein (ABCB1) and breast cancer resistance protein (ABCG2) at the blood-brain barrier: A case study examining sorafenib efflux clearance’, Molecular Pharmaceutics, 9(3), pp. 678–684. doi: 10.1021/mp200465c.
Dodda, L. S., Vilseck, J. Z., et al. (2017) ‘1.14∗CM1A-LBCC: Localized Bond-Charge Corrected CM1A Charges for Condensed-Phase Simulations’, Journal of Physical Chemistry B, 121(15), pp. 3864–3870. doi: 10.1021/acs.jpcb.7b00272.
Dodda, L. S., De Vaca, I. C., et al. (2017) ‘LigParGen web server: An automatic OPLS-AA parameter generator for organic ligands’, Nucleic Acids Research, 45(W1), pp. W331–W336. doi:
10.1093/nar/gkx312.
Empereur-Mot, C. et al. (2020) ‘Swarm-CG: Automatic Parametrization of Bonded Terms in MARTINI-Based Coarse-Grained Models of Simple to Complex Molecules via Fuzzy Self-Tuning Particle Swarm Optimization’, ACS Omega, 5(50), pp. 32823–32843. doi: 10.1021/acsomega.0c05469.
Gartzke, D. et al. (2015) ‘Genomic knockout of endogenous canine P-glycoprotein in wild-type, human P-glycoprotein and human BCRP transfected MDCKII cell lines by zinc finger nucleases’,
Pharmaceutical Research, 32(6), pp. 2060–2071. doi: 10.1007/s11095-014-1599-5.
Hammarlund-Udenaes, M. et al. (2008) ‘On the rate and extent of drug delivery to the brain’, Pharmaceutical Research, 25(8), pp. 1737–1750. doi: 10.1007/s11095-007-9502-2.
Hubatsch, I., Ragnarsson, E. G. E. and Artursson, P. (2007) ‘Determination of drug permeability and prediction of drug absorption in Caco-2 monolayers’, Nature Protocols, 2(9), pp. 2111–2119. doi: 10.1038/nprot.2007.303.
De Jong, D. H. et al. (2013) ‘Improved parameters for the martini coarse-grained protein force field’, Journal of Chemical Theory and Computation, 9(1), pp. 687–697. doi: 10.1021/ct300646g.
Jorgensen, W. L. and Tirado-Rives, J. (2005) ‘Potential energy functions for atomic-level simulations of water and organic and biomolecular systems’, Proceedings of the National Academy of Sciences of the United States of America, 102(19), pp. 6665–6670. doi: 10.1073/pnas.0408037102.
Karlgren, M. et al. (2017) ‘A CRISPR-Cas9 Generated MDCK Cell Line Expressing Human MDR1 Without Endogenous Canine MDR1 (cABCB1): An Improved Tool for Drug Efflux Studies’, Journal of Pharmaceutical Sciences, 106(9), pp. 2909–2913. doi: 10.1016/j.xphs.2017.04.018.
Khelashvili, G. et al. (2015) ‘Spontaneous Inward Opening of the Dopamine Transporter Is Triggered by PIP2-Regulated Dynamics of the N-Terminus’, ACS Chemical Neuroscience, 6(11), pp. 1825–1837. doi: 10.1021/acschemneuro.5b00179.
Khunweeraphong, N. et al. (2019) ‘The ABCG2 multidrug transporter is a pump gated by a valve and an extracellular lid’, Nature Communications, 10(1), pp. 1–14. doi: 10.1038/s41467-019-13302-2.
Kodaira, H. et al. (2010) ‘Kinetic analysis of the cooperation of P-glycoprotein (P-gp/Abcb1) and breast cancer resistance protein (Bcrp/Abcg2) in limiting the brain and testis penetration of erlotinib,
flavopiridol, and mitoxantrone’, Journal of Pharmacology and Experimental Therapeutics, 333(3), pp. 788–796. doi: 10.1124/jpet.109.162321.
Lee, J. Y. et al. (2016) ‘Crystal structure of the human sterol transporter ABCG5/ABCG8’, Nature, 533(7604), pp. 561–564. doi: 10.1038/nature17666.
Loschwitz, J. et al. (2020) ‘Computer simulations of protein – membrane systems’, (January).
Martens, C. et al. (2018) ‘Direct protein-lipid interactions shape the conformational landscape of secondary transporters’, Nature Communications, 9(1), pp. 1–12. doi: 10.1038/s41467-018-06704-1. Orlando, B. J. and Liao, M. (2020) ‘ABCG2 transports anticancer drugs via a closed-to-open switch’, Nature Communications, 11(1), pp. 1–11. doi: 10.1038/s41467-020-16155-2.
Shawahna, R. et al. (2011) ‘Transcriptomic and quantitative proteomic analysis of transporters and drug metabolizing enzymes in freshly isolated human brain microvessels’, Molecular Pharmaceutics, 8(4), pp. 1332–1341. doi: 10.1021/mp200129p.
Simoff, I. et al. (2016) ‘Complete Knockout of Endogenous Mdr1 (Abcb1) in MDCK Cells by CRISPR-Cas9’, Journal of Pharmaceutical Sciences, 105(2), pp. 1017–1021. doi: 10.1016/S0022-3549(15)00171-9.
Souza, P. C. T. et al. (2021) ‘coarse-grained molecular dynamics’, Nature methods. doi: 10.1038/s41592-021-01098-3.
Taylor, N. M. I. et al. (2017) ‘Structure of the human multidrug transporter ABCG2’, Nature, 546(7659), pp. 504–509. doi: 10.1038/nature22345.
Uchida, Y. et al. (2011) ‘Quantitative targeted absolute proteomics of human blood-brain barrier transporters and receptors’, Journal of Neurochemistry, 117(2), pp. 333–345. doi: 10.1111/j.1471-4159.2011.07208.x.
Wassenaar, T. A. et al. (2015) ‘Computational lipidomics with insane: A versatile tool for generating custom membranes for molecular simulations’, Journal of Chemical Theory and Computation, 11(5), pp. 2144–2155. doi: 10.1021/acs.jctc.5b00209.
Webster, D. (2000) Protein Structure Prediction, Protein Structure Prediction. doi: 10.1385/1592593682. Wegler, C. et al. (2020) ‘Expanding the Efflux In Vitro Assay Toolbox: A CRISPR-Cas9 Edited MDCK Cell Line with Human BCRP and Completely Lacking Canine MDR1’, Journal of Pharmaceutical Sciences, pp. 1–9. doi: 10.1016/j.xphs.2020.09.039.
Weng, J. W., Fan, K. N. and Wang, W. N. (2010) ‘The conformational transition pathway of ATP binding cassette transporter MsbA revealed by atomistic simulations’, Journal of Biological Chemistry, 285(5), pp. 3053–3063. doi: 10.1074/jbc.M109.056432.
FDA (2020) In Vitro Drug Interaction Studies — Cytochrome P450 Enzyme- and Transporter-Mediated Drug Interactions, Guidance for Industry [Online]. Available at:
Appendix. Supplement tables and figures
A1. Additional information on BCRP transporter models
Supplement Notes: BCRP transporter model building and optimizationInitial model (Model01) is obtained with a combination of the best chain A (6vxf.A99990004.pdb) and the best chain B (6vxf.B99990005.pdb), which is concatenated manually and then converted to CG-representation by Martini 3, Gromacs 2020.4 (see Figure S1 and Table S1).
Figure S1. Different representations of BCRP Model01 (chain A: 6vxf.A99990004.pdb, chain B:
6vxf.B99990005.pdb): a. Cartoon representation with missing residues added by Modeller. b. Coarse-grained model (by Martini 3, Gromacs 2020.4).
Model02 is an improved model based on Model 1, as the long loops between the first and
the second β-strand of the core domain of the NBD pointed out and differs from what is observed from the cryo-EM structure, Khunweeraphong et al (2019) argued that introduction of such long loops while no suitable template structure decreases the reliablibity of the models, therefore the long loops are removed by PyMOL and this model is named Model02 which is abbreviated for model_no long loops (see Figure S2) .
Figure S2. Different representations of BCRP Model02 (chain A: 6vxf.A99990004.pdb, chain B:
6vxf.B99990005.pdb, while long loops are removed by PyMOL): a. Cartoon representation of superposition of 6vxf vs Model02 . b. Coarse-grained Model02 (by Martini 3, elastic, Gromacs 2020.4).
Summary of BCRP transporter models with missing residues added by Modeller and the respective overall quality factor evaluated by ERRAT server (https://saves.mbi.ucla.edu/) is shown in Table S1. The error values versus the residues of the chain(s) chosen to build the models are shown in Figure S3 to Figure S6.
Table S1. Summary of BCRP transporter models and its overall quality factor.
BCRP Transporter models Overall Quality Factor1
Chain A 6vxf.A99990001.pdb 59.872 6vxf.A99990002.pdb 61.564 6vxf.A99990003.pdb 60.325 6vxf.A99990004.pdb2 62.540 6vxf.A99990005.pdb 59.740 Chain B 6vxf.B99990001.pdb 58.400 6vxf.B99990002.pdb 64.658 6vxf.B99990003.pdb 62.480 6vxf.B99990004.pdb 62.237 6vxf.B99990005.pdb2 68.130 6vxf.AB99990001.pdb 66.279
Chain A&B 6vxf.AB99990002.pdb 68.817
6vxf.AB99990003.pdb2 69.4103
1: Analyzes the statistics of non-bonded interactions between different atom types and plots the value of the error function versus position of a 9-residue sliding window, calculated by a comparison with statistics from highly refined structures.
2: Chain selected to build Model01 and Model02.
A2. Validation of the parameterization of drug CG models
Figure. S7 Model scores during optimization, red line is Dantrolene, cyan line is Ritonavir, the best
model is in blue circle.
A3. Summary of the simulations
Table S2. Table of BCRP transporter models & simulations
Code Description Substrate/
non-substrate Run # sim_length
Model01 With long loops stick
out
Dantrolene md_d_2.tpr 5μs
Ritonavir md_r_2.tpr 5μs
Model02 No loops, missing
some residues
Dantrolene md_d_5.tpr 5μs
Ritonavir md_r_6.tpr 5μs
Model03 Intelligent model
with all residues
