Trivariate Linear Regression and Machine Learning Prediction of Possible Roles of Efflux Transporters in Estimated Intestinal Permeability Values of 301 Disparate Chemicals

Abstract

A system for predicting apparent bidirectional permeability (P_app) across Caco-2 cells of diverse chemicals has been reported. The present study aimed to investigate the relationship between in silico-generated P_app (from apical to basal side, P_{app A to B}) for 301 substances with diverse structures and a binary classification of the reported roles of efflux P-glycoprotein or breast cancer resistant protein. The in silico log(P_{app A to B}/P_{app B to A}) values of 70 substances with reported active efflux and 231 substances with no reported active efflux were significantly different (p < 0.01). The probabilities of active efflux transport estimated by trivariate analysis with log MW, log D_{pH 6.0}, and log D_{pH 7.4} for the 70 active-efflux-positive compounds were higher than those of the other 231 substances (p < 0.01); the area under the corresponding receiver operating characteristic (ROC) curve was 0.81. Further probability values estimated using a machine learning algorithm with 30 chemical descriptors as inputs yielded an area under the ROC curve of 0.79. Using a secondary set of 52 efflux-positive and 48 efflux-negative medicines, the final trivariate-generated probabilities resulted in no significant differences between these binary groups (p = 0.09); however, the final machine learning model demonstrated a good area under the ROC curve of 0.79. Consequently, a combination of the previously established system for generating the permeability coefficients across intestinal monolayers (a continuous variable) and the currently proposed system for predicting the roles of additional active efflux (a binary classification) could prove useful; high accuracy was achieved by applying machine learning using in silico-generated chemical descriptors.

INTRODUCTION

Oral substance absorption in the gastrointestinal tract is dependent on many factors, e.g., the local pH, drug-metabolizing enzymes, and influx or efflux transporters.¹⁾ Caco-2 cell monolayers with the pH gradient adjusted to that between the gastrointestinal lumen and plasma have been shown to well reflect the human oral absorption of medicines in the gut.^2,3) We successfully used trivariate regression analyses (based on simple physicochemical properties estimated in silico) to predict the apparent permeability (P_app) values obtained using bidirectional Caco-2 cell permeation experiments [i.e., influx from the apical side to the basal side (hereinafter referred to as A to B) and efflux from the basal side to the apical side (hereinafter referred to as B to A)] for 230 disparate industrial chemicals and drugs.⁴⁾ The observed influx and efflux log P_app values of these substances were both significantly correlated in trivariate analyses with molecular weights and octanol–water distribution coefficients (log D) at the apical pH of 6.0 and at the basal pH of 7.4; furthermore, improved accuracy was obtained by applying a light gradient boosting machine learning system (LightGBM) using up to 19 in silico chemical descriptors.⁴⁾

In the previous study, some predicted log P_app values of drugs in the primary set of test substances fell outside the threefold-error region, presumably because of the contributions of active efflux/influx pumps in the experimental Caco-2 membrane environment.⁴⁾ To estimate the oral absorption in vivo of a diverse range of general chemicals, food components, and drugs it is important to provide additional information on the net differences in the in silico estimated P_{app A to B} and P_{app B to A} values caused by active transport in addition to passive diffusion. The aim of the present study was to investigate the relationship between the in silico estimated P_{app A to B} and P_{app B to A} values for substances and the sparsely reported roles of efflux P-glycoprotein (PgP) or breast cancer resistant protein (BCRP) in their absorbability. Our focus was simple to understand apparent net differences in the in silico estimated P_{app A to B} and P_{app B to A} of a diverse range of general chemicals. Separated investigation for PgP and BCRP was not suitable under the limited active transport information for general chemicals. We report herein the estimation of possible contributions of efflux transporters in the intestinal permeability of chemicals (in binary classification: 1 = positive efflux transport, 0 = no efflux transport) using two methods: (1) trivariate analyses with molecular weights and log D at the apical pH of 6.0 and at the basal pH of 7.4 and (2) a new and reliable machine learning system using 30 in silico-derived chemical descriptors.

MATERIALS AND METHODS

Substances

The structures of the 301 substances (including industrial chemicals, food-derived substances, and pharmaceuticals) involved in the study were projected onto a two-dimensional plane in the chemical space (with unlabeled axes, Fig. 1) by applying the previously reported generative topographic mapping methods,^4–10) thereby allowing visualization of their chemical diversity. The data sets to which linear regression and machine learning were applied in this study covered the 301 substances (Table 1), which included 219 chemicals previously tested,⁴⁾ 11 compounds with reported permeability values,^11,12) and 71 substances for which we previously established physiologically based pharmacokinetic modeling for rats and/or humans.^10,13) A literature search was performed in PubMed with the keywords “‘chemical name’ and (PgP substrate transporter or BCRP substrate transporter)” to identify possible roles of efflux transporters to establish the binary classification. A secondary set of 100 medicines (52 efflux positive and 48 efflux negative) were taken from a new Japanese drug database containing information on the roles of efflux transports (Supplementary Table S1). Histograms and receiver operating characteristic (ROC) curves for the 301 primary chemicals and 100 secondary medicines showing estimates of the probable contributions of efflux pumps to intestinal permeability were prepared using Prism software (GraphPad Software, San Diego, CA, U.S.A.). Statistical analyses were also performed using Prism.

Fig. 1. Coordinate Values of the Primary Set of 301 Chemicals (A) and the Secondary Set of 100 Medicines (B) in a Two-Dimensional Plane with 25 Partitions Illustrating Variety in Their Chemical Structures

Chemicals with information in the literature on possible roles of efflux transporters (i.e., by P-glycoprotein or breast cancer resistant protein) contributing to the membrane permeability are shown in gray, representing binary classification scores of 1. Those shown in white represent a binary classification of 0, meaning no roles of efflux transporters or no relevant information. This proposed evaluation of variety in the chemical space^4–10) was also conducted for our studies on intestinal permeability^4,7) and for rat and human pharmacokinetic modeling studies.^9,10)

Table 1. Possible Contributions of Efflux Transporters in Intestinal Permeability of Primary 301 Chemicals in Binary Classification Estimated by Trivariate Analyses Using Physiological Properties and with a New Machine Learning System Using in Silico 30 Chemical Descriptors

Name	Cas No.	Role of efflux PgP or BCRPa)	log MW	log D pH 6.0	log D pH 7.4	In silico log Papp A to B, nm/s	log(Papp A to B/Papp B to A)	Split group No.	Probability of role for active efflux transporters
									Trivariate, cross validation	Tri-variate, final	Machine learning, cross validation	Machine learning, final
Abemaciclib	1231929-97-7	115)	2.71	1.64	2.92	1.80	−0.97	7	0.55	0.55	0.77	0.69
Acetaminophen	103-90-2	016)	2.18	0.43	0.43	2.47	−0.06	8	0.09	0.10	0.02	0.03
Acrylonitrile	107-13-1	0	1.73	0.59	0.59	2.03	−0.58	7	0.00	0.00	0.03	0.03
Acyclovir	59277-89-3	0	2.35	−1.37	−1.37	1.38	−0.05	10	0.30	0.30	0.85	0.29
AG-041R	199800-49-2	0	2.74	3.49	3.49	1.72	−0.84	8	0.42	0.42	0.67	0.49
Allopurinol	315-30-0	017)	2.13	−0.73	−0.74	1.79	−0.04	1	0.09	0.10	0.00	0.02
Alprazolam	28981-97-7	0	2.49	2.77	2.77	2.70	−0.04	5	0.27	0.25	0.08	0.07
Ambroxol	18683-91-5	0	2.58	−0.31	0.77	1.58	−1.30	10	0.52	0.51	0.04	0.11
Amlodipine	88150-42-9	0	2.61	0.31	1.29	1.04	−1.55	6	0.51	0.50	0.46	0.63
Aniline	62-53-3	0	1.97	1.23	1.25	2.73	−0.09	4	0.00	0.00	0.04	0.03
Apigenin	520-36-5	0	2.43	2.59	2.52	2.06	−0.30	4	0.20	0.21	0.07	0.10
Apixaban	503612-47-3	118)	2.66	0.99	0.99	2.32	0.03	8	0.45	0.45	0.82	0.75
Apomorphine	58-00-4	0	2.43	1.64	2.74	1.79	−0.36	8	0.30	0.32	0.50	0.30
Aspirin	50-78-2	0	2.26	−1.05	−2.25	1.74	0.19	3	0.15	0.13	0.01	0.03
Atenolol	29122-68-7	119)	2.42	−2.66	−1.77	0.87	0.18	10	0.47	0.46	0.32	0.32
Atomoxetine	83015-26-3	0	2.41	0.70	1.53	1.39	−1.39	6	0.32	0.32	0.04	0.05
Atorvastatin	134523-00-5	120)	2.75	2.76	1.40	1.53	0.00	5	0.37	0.36	0.31	0.61
Azamethiphos	35575-96-3	0	2.51	1.73	1.73	2.51	−0.20	2	0.30	0.31	0.17	0.12
Azithromycin	83905-01-5	121)	2.87	−1.59	0.13	1.07	−0.99	9	0.83	0.83	0.81	0.90
Balsalazide	80573-04-2	0	2.55	−1.81	−2.04	1.13	−0.49	3	0.48	0.46	0.60	0.23
Baricitinib	1187594-09-7	122)	2.57	1.20	1.20	2.03	−0.25	2	0.37	0.37	0.70	0.67
Benidipine	105979-17-7	0	2.70	3.75	4.82	1.66	−0.77	7	0.47	0.45	0.79	0.67
Bensulide	741-58-2	0	2.60	4.09	4.09	1.73	−0.44	1	0.31	0.29	0.05	0.11
Benzimidazole	51-17-2	0	2.07	1.43	1.57	2.92	0.17	2	0.00	0.00	0.02	0.02
Benzoic acid	65-85-0	023)	2.09	−0.03	−1.37	3.07	1.18	9	0.00	0.00	0.00	0.02
Benzydamine	642-72-8	0	2.49	1.17	2.36	1.13	−1.85	6	0.39	0.39	0.51	0.20
Beraprost	88430-50-6	124)	2.60	2.16	0.80	1.48	−0.31	3	0.28	0.27	0.91	0.87
Betaxolol	63659-18-7	0	2.49	−0.01	0.88	1.54	−1.36	4	0.41	0.41	0.36	0.33
Bisoprolol	66722-44-9	125)	2.51	−0.70	0.20	1.38	−1.33	1	0.45	0.46	0.18	0.38
Bisphenol A	80-05-7	126)	2.36	3.73	3.73	2.39	−0.06	1	0.12	0.11	0.03	0.08
Bisphenol F	620-92-8	0	2.30	2.95	2.95	2.55	−0.07	8	0.09	0.10	0.03	0.02
Bisphenol S	80-09-1	0	2.40	1.79	1.61	2.67	0.43	9	0.20	0.20	0.02	0.02
Bosentan	147536-97-8	0	2.74	3.07	1.79	1.37	−1.05	1	0.40	0.35	0.95	0.47
Caffeine	58-08-2	0	2.29	0.42	0.42	2.72	0.02	10	0.19	0.18	0.42	0.23
Canagliflozin	842133-18-0	127)	2.65	3.00	3.00	1.60	−0.54	8	0.36	0.37	0.61	0.66
Carbamazepine	298-46-4	0	2.37	2.17	2.17	2.53	−0.19	4	0.17	0.18	0.04	0.06
Ceritinib	1032900-25-6	028)	2.75	1.01	2.23	1.10	−1.46	1	0.60	0.60	0.94	0.56
Chloramphenicol	56-75-7	0	2.51	0.79	0.78	1.89	−0.15	5	0.35	0.34	0.12	0.25
Chlorogenic acid	327-97-9	0	2.55	−3.05	−4.14	0.87	−0.95	2	0.45	0.45	0.26	0.22
Chlorpromazine	50-53-3	0	2.50	2.58	3.66	0.94	−1.92	10	0.35	0.34	0.29	0.28
Chlorpyrifos	2921-88-2	029)	2.55	4.79	4.79	2.24	0.14	10	0.22	0.22	0.05	0.04
Chlorzoxazone	95-25-0	0	2.23	1.59	0.87	2.90	0.12	3	0.05	0.05	0.00	0.02
Chrysin	480-40-0	030)	2.41	3.24	3.18	2.01	−0.33	6	0.15	0.16	0.07	0.05
Cimetidine	51481-61-9	0	2.40	0.00	0.75	1.21	−0.40	4	0.34	0.34	0.40	0.17
Ciprofloxacin	85721-33-1	131,32)	2.52	−2.95	−2.67	1.39	−0.07	7	0.49	0.51	0.79	0.75
Cotinine	486-56-6	0	2.25	0.32	0.33	2.45	−0.13	5	0.15	0.15	0.02	0.08
Coumarin	91-64-5	0	2.16	2.06	2.06	2.87	−0.08	1	0.02	0.02	0.00	0.03
Curcumin	458-37-7	033)	2.57	2.39	2.38	2.11	0.05	1	0.34	0.32	0.09	0.28
Dabigatran	211914-51-1	134)	2.67	−1.20	−0.66	1.42	−0.10	4	0.57	0.58	0.67	0.78
Darolutamide	1297538-32-9	135)	2.60	1.89	1.89	1.70	−0.50	6	0.37	0.37	0.34	0.36
Dexamethasone	50-02-2	136)	2.59	1.92	1.92	1.93	−0.36	4	0.35	0.36	0.62	0.77
Dextromethorphan	125-71-3	037)	2.43	0.77	2.02	1.42	−1.64	7	0.38	0.36	0.71	0.30
Dichloromethane	75-09-2	0	1.93	1.36	1.36	1.89	−0.43	4	0.00	0.00	0.04	0.03
Diclofenac	15307-86-5	038)	2.47	2.95	1.60	2.79	1.40	2	0.16	0.14	0.05	0.04
Digoxin	20830-75-5	139)	2.89	1.33	1.33	1.16	−0.59	7	0.60	0.62	0.77	0.84
Dihydrocodeine	125-28-0	0	2.48	−0.54	0.75	1.37	−1.30	5	0.47	0.45	0.88	0.72
Diltiazem	42399-41-7	140)	2.62	1.57	2.82	1.51	−1.38	3	0.47	0.48	0.72	0.67
Diphenylamine	122-39-4	0	2.23	3.19	3.19	2.14	−0.22	1	0.03	0.03	0.00	0.02
Disopyramide	3737-09-5	0	2.53	−0.16	0.85	1.21	−1.14	8	0.45	0.46	0.53	0.26
Disulfoton-sulfone	2497-06-5	0	2.49	2.18	2.18	2.40	−0.08	3	0.27	0.27	0.01	0.07
Domitroban	112966-96-8	0	2.58	2.92	1.56	2.04	0.03	7	0.20	0.22	0.06	0.09
Doxorubicin	23214-92-8	141)	2.74	−1.36	−0.08	0.88	−0.63	4	0.67	0.68	0.70	0.90
Duloxetine	116539-59-4	042)	2.47	1.19	2.02	1.11	−1.79	7	0.36	0.35	0.14	0.07
Edoxaban	480449-70-5	143)	2.74	0.64	1.28	1.30	−0.92	2	0.56	0.57	0.70	0.71
Efonidipine	111011-63-3	0	2.80	5.40	5.57	1.63	−0.70	1	0.44	0.40	0.71	0.46
EGCG	989-51-5	044)	2.66	1.67	1.52	1.62	−0.24	1	0.44	0.42	0.83	0.43
Empagliflozin	864070-44-0	0	2.65	1.90	1.90	1.36	−0.95	5	0.43	0.41	0.87	0.42
Epicatechin gallate	1257-08-5	145)	2.65	2.15	2.01	1.64	−0.24	4	0.37	0.39	0.35	0.62
Eprosartan	133040-01-4	0	2.63	0.98	0.47	1.25	0.20	9	0.39	0.39	0.91	0.48
Erythromycin	114-07-8	146)	2.87	−0.08	1.19	1.06	−0.83	1	0.73	0.74	0.97	0.90
Esaxerenone	1632006-28-0	147)	2.67	2.55	2.55	2.20	0.14	2	0.40	0.40	0.65	0.69
Estradiol	50-28-2	0	2.43	3.37	3.37	2.12	−0.20	8	0.18	0.18	0.39	0.23
Fenbufen	36330-85-5	0	2.40	1.86	0.50	2.71	1.05	9	0.12	0.13	0.03	0.05
Fentanyl	437-38-7	148)	2.53	2.10	3.38	1.28	−1.58	9	0.38	0.39	0.25	0.39
Fexofenadine	83799-24-0	149)	2.70	2.49	2.49	1.52	−0.23	9	0.41	0.43	0.22	0.64
Flavone	525-82-6	0	2.35	3.38	3.38	2.26	−0.15	5	0.14	0.12	0.01	0.02
Florfenicol	73231-34-2	0	2.55	0.70	0.68	1.54	0.14	7	0.37	0.38	0.19	0.10
Flunitrazepam	1622-62-4	0	2.50	2.45	2.45	2.77	−0.06	10	0.27	0.27	0.26	0.16
Fluoro-loxoprofen	1241405-00-4	0	2.42	0.96	−0.37	2.75	1.28	1	0.21	0.18	0.16	0.10
Fluvastatin	93957-54-1	150)	2.61	1.98	0.63	2.24	0.91	10	0.28	0.29	0.88	0.79
Fluvoxamine	54739-18-3	0	2.50	0.63	1.89	1.45	−1.40	5	0.45	0.42	0.19	0.19
G004	865483-06-3	0	2.75	3.63	2.29	2.00	−0.07	3	0.33	0.33	0.37	0.16
GB-115	678996-63-9	0	2.64	2.57	2.57	1.70	−0.39	5	0.40	0.37	0.75	0.30
Glibenclamide	10238-21-8	151)	2.69	2.95	1.62	1.96	−0.07	10	0.31	0.32	0.29	0.62
Haloperidol	52-86-8	052)	2.58	1.07	2.31	1.49	−1.00	3	0.46	0.46	0.81	0.20
Hippuric acid	495-69-2	0	2.25	−1.98	−3.14	1.01	0.08	7	0.14	0.17	0.02	0.02
Ibrolipim	133208-93-2	0	2.65	2.27	2.27	1.88	−0.13	5	0.42	0.40	0.08	0.10
Ibuprofen	15687-27-1	0	2.31	1.50	0.15	3.04	1.30	4	0.06	0.07	0.07	0.05
Imipramine	50-49-7	142)	2.45	1.80	2.86	1.58	−1.32	1	0.31	0.32	0.04	0.32
Indomethacin	53-86-1	0	2.55	2.96	1.60	2.80	1.12	4	0.18	0.20	0.39	0.30
Ipragliflozin	761423-87-4	0	2.61	2.54	2.54	1.37	−0.82	5	0.37	0.35	0.76	0.36
Isophthalonitrile	626-17-5	0	2.11	1.06	1.06	2.71	−0.12	10	0.03	0.02	0.01	0.02
Isopsoralen	523-50-2	0	2.27	2.16	2.16	2.70	−0.04	2	0.09	0.10	0.04	0.03
Itopride	122898-67-3	0	2.55	−0.07	1.22	1.06	−1.64	1	0.48	0.49	0.35	0.44
Ivabradine	155974-00-8	153)	2.67	1.07	2.36	1.21	−1.72	5	0.58	0.54	0.95	0.72
Ketoprofen	22071-15-4	054)	2.40	1.33	0.00	3.04	1.01	1	0.18	0.15	0.06	0.06
Lamivudine	134678-17-4	155)	2.36	−1.04	−1.03	1.41	−0.22	5	0.29	0.29	0.07	0.20
Lemildipine	94739-29-4	0	2.66	4.33	4.33	1.75	0.08	2	0.33	0.32	0.58	0.47
Lenalidomide	191732-72-6	156)	2.41	0.03	−0.17	1.04	−0.20	5	0.28	0.28	0.03	0.29
Letermovir	917389-32-3	157)	2.76	1.99	1.99	1.78	0.11	1	0.51	0.49	0.76	0.74
Lisinopril	76547-98-3	0	2.61	0.70	0.52	1.10	−0.08	10	0.41	0.41	0.76	0.41
Loflazepate	71735-10-9	0	2.52	−1.01	−1.66	1.91	0.29	10	0.37	0.38	0.12	0.10
Lomitapide	182431-12-5	0	2.84	4.28	5.13	1.68	−0.87	6	0.51	0.52	0.48	0.39
Losartan	114798-26-4	158)	2.63	1.44	0.53	1.52	0.03	10	0.34	0.35	0.55	0.54
Lovastatin	75330-75-5	159)	2.61	4.16	4.16	1.36	−0.59	4	0.27	0.29	0.66	0.78
Loxoprofen	68767-14-6	060)	2.39	1.03	−0.32	2.86	1.13	5	0.14	0.15	0.03	0.08
L-Thyroxine	51-48-9	061)	2.89	4.21	3.57	1.25	0.21	10	0.46	0.47	0.37	0.10
Lucifer yellow CH	67769-47-5	1	2.65	−8.73	−9.58	0.79	−0.18	5	0.69	0.76	0.02	0.44
Macitentan	441798-33-0	062)	2.73	3.03	2.39	1.67	−0.89	3	0.38	0.38	0.60	0.28
Maleic acid	108-31-6	0	1.99	−0.08	−0.08	1.32	0.17	7	0.00	0.00	0.08	0.04
Mangiferin	4773-96-0	063)	2.63	0.58	0.45	1.41	−0.97	9	0.43	0.43	0.73	0.34
m-Cresol	108-39-4	0	2.03	2.13	2.13	2.82	−0.09	6	0.00	0.00	0.10	0.06
Mefenamic acid	61-68-7	0	2.38	3.20	1.85	3.00	1.27	1	0.10	0.06	0.03	0.06
Melengestrol acetate	2919-66-6	0	2.60	4.03	4.03	2.10	−0.20	9	0.27	0.29	0.81	0.33
Menthofuran	494-90-6	0	2.18	3.97	3.97	2.61	−0.23	7	0.00	0.00	0.04	0.06
Metformin	657-24-9	064)	2.11	−4.08	−4.06	0.99	−0.44	10	0.22	0.22	0.33	0.09
Methotrexate	59-05-2	164)	2.66	−4.23	−5.53	1.02	0.42	7	0.50	0.56	0.60	0.87
Methoxsalen	298-81-7	0	2.33	1.99	1.99	2.49	−0.09	2	0.15	0.16	0.07	0.05
Metoprolol	51384-51-1	065)	2.43	−1.13	−0.24	1.47	−1.48	6	0.42	0.41	0.46	0.33
Midazolam	59467-70-8	0	2.51	2.89	3.22	2.49	−0.49	5	0.31	0.29	0.09	0.11
Mirtazapine	85650-52-8	042)	2.42	1.95	2.93	1.64	−1.41	2	0.28	0.29	0.56	0.45
Mofezolac	78967-07-4	0	2.53	1.23	−0.08	2.54	1.27	9	0.25	0.25	0.03	0.24
Molinate	2212-67-1	0	2.27	2.29	2.29	2.22	−0.20	2	0.09	0.10	0.04	0.06
Mono(2-ethylhexyl) phthalate	4376-20-9	0	2.44	2.74	1.51	2.65	1.31	1	0.17	0.14	0.02	0.07
Monobutyl phthalate	131-70-4	0	2.35	0.56	−0.67	2.43	1.15	8	0.14	0.14	0.06	0.04
Morphine	57-27-2	166)	2.45	−1.37	−0.09	1.29	−0.93	6	0.48	0.46	0.47	0.78
m-Toluic acid	99-04-7	0	2.13	0.45	−0.89	3.10	1.20	3	0.00	0.00	0.01	0.03
N,2-Dimethylaniline	611-21-2	0	2.08	2.08	2.13	2.80	−0.17	6	0.00	0.00	0.04	0.04
N,N-Diethyl-3-methylbenzamide	134-62-3	0	2.28	2.32	2.32	2.74	0.00	10	0.11	0.10	0.08	0.09
N,N-Dimethylaniline	121-69-7	0	2.08	2.36	2.41	2.89	−0.10	2	0.00	0.00	0.02	0.03
Nadolol	42200-33-9	167)	2.49	−1.76	−0.92	1.22	0.05	8	0.46	0.48	0.31	0.54
N-Ethylaniline	103-69-5	0	2.08	2.20	2.25	2.78	−0.09	3	0.00	0.00	0.00	0.03
Nicotine	54-11-5	0	2.21	−1.38	−0.09	1.73	−1.47	6	0.29	0.27	0.06	0.05
Nifedipine	21829-25-4	168,69)	2.54	3.58	3.58	2.60	−0.20	6	0.25	0.26	0.13	0.51
Nilvadipine	75530-68-6	0	2.59	3.80	3.80	1.94	−0.45	2	0.29	0.29	0.54	0.36
N-Methylaniline	100-61-8	0	2.03	1.93	1.98	2.66	−0.13	4	0.00	0.00	0.03	0.04
Norfloxacin	70458-96-7	170)	2.50	−3.26	−2.98	1.36	−0.14	1	0.50	0.51	0.80	0.75
Novaluron	116714-46-6	0	2.69	4.96	4.92	2.55	1.05	6	0.30	0.32	0.20	0.10
N-Phenylglycine	103-01-5	0	2.18	−1.67	−2.97	2.52	1.12	4	0.09	0.09	0.02	0.02
o-Cresol	95-48-7	0	2.03	2.13	2.13	2.86	−0.03	4	0.00	0.00	0.04	0.05
Ofloxacin	82419-36-1	131)	2.56	−2.23	−1.99	1.87	−0.29	10	0.51	0.51	0.93	0.78
Olanzapine	132539-06-1	171)	2.49	0.95	2.20	1.44	−1.46	1	0.39	0.40	0.11	0.61
Olopatadine	113806-05-6	172)	2.53	1.88	1.87	1.35	−0.65	8	0.31	0.31	0.48	0.51
Omeprazole	73590-58-6	173)	2.54	2.06	2.05	2.68	0.48	2	0.31	0.31	0.39	0.41
Opicapone	923287-50-7	174)	2.62	2.14	1.41	2.75	0.20	4	0.30	0.33	0.27	0.42
Oseltamivir	196618-13-0	175)	2.49	−0.90	0.39	1.43	−0.87	8	0.46	0.48	0.37	0.74
p-Aminobenzoic acid	150-13-0	0	2.14	−0.15	−1.50	2.70	1.11	9	0.01	0.00	0.00	0.02
Paraacetaldehyde	123-63-7	0	2.12	0.53	0.53	2.37	0.20	8	0.04	0.05	0.08	0.04
p-Coumaric acid	7400-08-0	076)	2.21	0.14	−1.22	2.35	0.96	8	0.05	0.05	0.02	0.02
p-Cresol	106-44-5	1	2.03	2.13	2.13	2.76	0.03	8	0.00	0.00	0.01	0.06
Pemafibrate	848259-27-8	177)	2.69	1.74	0.79	1.65	0.44	9	0.37	0.38	0.44	0.60
Perfluorodecanoic acid	335-76-2	078)	2.71	1.95	1.38	2.20	0.38	3	0.42	0.42	0.02	0.05
Perfluorooctanoic acid	335-67-1	079)	2.62	0.71	0.14	2.14	0.58	3	0.40	0.39	0.02	0.12
PF-04937319	1245603-92-2	0	2.64	1.92	1.92	1.73	−1.17	1	0.41	0.40	0.95	0.61
Phenacetin	62-44-2	0	2.25	1.74	1.74	2.71	0.05	7	0.11	0.10	0.02	0.04
Phenobarbital	50-06-6	180)	2.37	1.29	1.05	2.65	0.18	6	0.20	0.19	0.09	0.25
Phthalazine	253-52-1	0	2.11	1.23	1.23	3.09	0.06	6	0.01	0.02	0.04	0.02
Phthalazone	119-39-1	0	2.16	0.96	0.96	3.04	0.24	2	0.05	0.06	0.02	0.02
Phthalimide	85-41-6	0	2.17	0.93	0.92	2.84	−0.25	10	0.08	0.07	0.05	0.03
Phthalonitrile	91-15-6	0	2.11	1.06	1.06	2.82	−0.13	2	0.00	0.02	0.02	0.02
p-Hydroxybenzaldehyde	123-08-0	0	2.09	1.38	1.24	2.83	0.13	9	0.00	0.00	0.00	0.02
p-Hydroxybenzoic acid	99-96-7	0	2.14	0.25	−1.11	2.82	1.25	5	0.00	0.00	0.01	0.02
Pitavastatin	147511-69-1	150)	2.62	1.78	0.48	2.17	0.57	3	0.32	0.31	0.59	0.62
p-Nitrophenol	100-02-7	0	2.14	1.71	1.33	3.06	0.49	1	0.00	0.00	0.00	0.02
Pomalidomide	19171-19-8	181)	2.44	0.21	0.03	2.44	−0.18	3	0.30	0.29	0.17	0.27
p-Phenetidine	156-43-4	0	2.14	1.11	1.21	2.71	0.16	8	0.04	0.04	0.02	0.03
Pranlukast	103177-37-3	082)	2.68	0.63	0.63	1.96	−0.24	9	0.48	0.48	0.20	0.16
Pravastatin	81093-37-0	183)	2.63	0.88	−0.47	1.00	0.30	9	0.34	0.34	0.93	0.82
Progesterone	57-83-0	141)	2.50	3.58	3.58	2.08	−0.50	6	0.21	0.23	0.38	0.40
Propranolol	525-66-6	184)	2.41	0.37	1.26	1.35	−1.56	2	0.32	0.34	0.09	0.18
Psoralen	66-97-7	0	2.27	2.15	2.15	2.79	0.03	3	0.10	0.10	0.01	0.04
p-Toluic acid	99-94-5	0	2.13	0.45	−0.89	3.07	1.25	1	0.00	0.00	0.01	0.03
Pyrithiobac	123342-93-8	0	2.51	0.28	−0.61	2.26	0.78	5	0.30	0.31	0.07	0.21
Quetiapine	111974-69-7	185)	2.58	2.73	2.84	1.48	−1.22	3	0.33	0.33	0.53	0.71
Quinine	130-95-0	186)	2.51	0.35	1.66	1.32	−1.63	10	0.46	0.44	0.57	0.53
Quinotolast	101193-40-2	0	2.49	0.31	−0.83	2.01	0.28	6	0.28	0.27	0.12	0.08
Quizartinib	950769-58-1	187)	2.75	3.32	4.30	1.29	−1.11	10	0.50	0.50	0.80	0.66
Rabeprazole	117976-89-3	0	2.56	1.79	1.78	1.69	−0.69	7	0.33	0.34	0.86	0.51
Ranitidine	66357-35-5	188)	2.50	−1.92	−0.63	1.33	−0.04	8	0.50	0.52	0.37	0.43
Resorcinol	108-46-3	089)	2.04	0.91	0.91	2.53	−0.13	9	0.00	0.00	0.00	0.02
Rhodamine123	62669-70-9	190)	2.54	1.57	2.08	1.04	−1.09	6	0.37	0.37	0.18	0.44
Risperidone	106266-06-2	185)	2.61	0.37	1.66	1.50	−1.37	5	0.55	0.52	0.86	0.67
Rivaroxaban	366789-02-8	191)	2.64	1.82	1.82	2.30	−0.04	4	0.39	0.40	0.22	0.35
Rosuvastatin	287714-41-4	192)	2.68	0.23	−1.13	1.52	0.28	7	0.36	0.41	0.65	0.67
Roxadustat	808118-40-3	0	2.55	−0.29	−1.43	1.85	0.70	3	0.35	0.34	0.68	0.25
Sacubitril	149709-62-6	193)	2.61	2.45	1.08	2.17	0.75	7	0.24	0.27	0.25	0.50
Safinamide	133865-89-1	0	2.48	0.50	1.68	1.78	−0.88	10	0.42	0.41	0.03	0.07
Salicylamide	65-45-2	0	2.14	1.09	1.04	2.89	0.22	1	0.03	0.03	0.00	0.02
Salicylic acid	69-72-7	0	2.14	−0.50	−1.45	3.01	1.11	5	0.02	0.04	0.01	0.02
Sarpogrelate	125926-17-2	0	2.63	1.50	1.36	1.33	−0.33	6	0.41	0.40	0.80	0.36
Scopoletin	92-61-5	0	2.28	1.75	1.61	2.94	0.24	8	0.11	0.12	0.06	0.04
Sepimostat	103926-64-3	0	2.57	−2.48	−2.46	1.02	−0.61	3	0.54	0.52	0.10	0.15
Sesamin	607-80-7	094)	2.55	2.76	2.76	2.10	−0.65	8	0.29	0.30	0.34	0.19
Silmitasertib	1009820-21-6	0	2.54	3.03	2.93	2.12	0.63	5	0.30	0.28	0.02	0.14
Sparfloxacin	110871-86-8	160)	2.59	−2.54	−2.30	1.79	−0.56	7	0.52	0.55	0.91	0.76
Styrene	100-42-5	0	2.02	3.25	3.25	2.69	0.01	9	0.00	0.00	0.01	0.03
Sulfasalazine	599-79-1	195)	2.60	−0.91	−2.22	1.76	−0.50	5	0.37	0.39	0.00	0.13
Sulindac	38194-50-2	096)	2.55	1.53	0.18	2.59	0.81	6	0.26	0.26	0.22	0.22
Sumatriptan	103628-46-2	097)	2.47	−0.99	0.26	1.61	−1.23	4	0.46	0.46	0.32	0.31
Suvorexant	1030377-33-3	0	2.65	3.27	3.27	2.02	−0.38	1	0.38	0.36	0.98	0.55
TA-510	133118-88-4	0	2.52	1.99	1.99	2.70	0.16	2	0.30	0.30	0.41	0.39
Tacrolimus	104987-11-3	198)	2.91	4.24	4.24	1.32	−0.52	9	0.50	0.52	0.81	0.87
Tadalafil	171596-29-5	199,100)	2.59	2.69	2.69	2.02	−0.44	9	0.32	0.33	0.74	0.72
Tedizolid	856866-72-3	0	2.57	1.66	1.66	2.62	0.06	1	0.37	0.35	0.83	0.28
Telithromycin	191114-48-4	1101)	2.91	1.56	3.07	1.17	−1.09	3	0.72	0.72	0.93	0.87
Tepotinib	1100598-32-0	0102)	2.74	0.24	1.00	1.69	−0.97	5	0.61	0.59	0.42	0.29
Terephthalonitrile	623-26-7	0	2.11	1.06	1.06	2.78	−0.02	10	0.03	0.02	0.01	0.02
Tetrabromobisphenol A	79-94-7	0	2.74	6.39	6.22	1.66	0.03	10	0.29	0.29	0.03	0.08
Thalidomide	50-35-1	0103)	2.41	0.45	0.38	2.35	0.08	8	0.27	0.27	0.50	0.21
Theophylline	58-55-9	0	2.26	0.25	0.23	2.43	−0.10	4	0.16	0.16	0.13	0.07
Timolol	26839-75-8	0	2.50	−1.25	−0.55	1.44	−1.16	8	0.44	0.46	0.28	0.33
Tolbutamide	64-77-7	0	2.43	1.24	−0.10	3.01	1.18	6	0.18	0.18	0.11	0.05
Toluene	108-88-3	0	1.96	2.58	2.58	2.73	−0.01	10	0.00	0.00	0.01	0.06
Tranilast	53902-12-8	0	2.51	0.20	−0.98	2.62	1.09	6	0.31	0.29	0.18	0.15
trans-Ferulic acid	537-98-4	0104)	2.29	0.17	−1.19	2.56	1.18	9	0.11	0.10	0.01	0.03
trans-Resveratrol	501-36-0	0	2.36	2.71	2.71	2.36	−0.14	7	0.16	0.15	0.02	0.02
Trazodone	19794-93-5	0	2.57	2.40	3.12	1.77	−1.14	5	0.41	0.37	0.06	0.42
Triazolam	28911-01-5	0	2.54	3.48	3.48	2.56	−0.22	10	0.26	0.26	0.20	0.11
Trichloroethylene	79-01-6	0	2.12	2.71	2.71	1.82	−0.47	8	0.00	0.00	0.02	0.04
Trimethylamine	75-50-3	0	1.77	−2.85	−2.15	1.29	−1.11	5	0.00	0.00	0.10	0.03
Trimethylamine N-oxide	1184-78-7	1105)	1.88	−1.11	−1.08	0.92	0.05	2	0.00	0.00	0.02	0.08
Vadadustat	1000025-07-9	0	2.49	−1.13	−2.27	1.35	−0.21	5	0.30	0.32	0.06	0.14
Vatanidipine	116308-55-5	0	2.84	6.50	6.85	1.76	−0.77	5	0.46	0.40	0.84	0.44
Verapamil	52-53-9	141)	2.69	1.47	2.71	1.36	−1.37	6	0.54	0.54	0.56	0.72
Warfarin	81-81-2	0106)	2.54	1.82	0.46	3.00	1.28	2	0.25	0.24	0.12	0.09
Wyeth-14643	50892-23-4	0	2.51	1.74	0.52	2.38	0.79	10	0.22	0.23	0.06	0.06
YJC-10592	1226894-87-6	0	2.75	0.62	1.84	1.42	−0.98	9	0.61	0.62	0.39	0.44
Zolpidem	82626-48-0	0	2.49	2.81	3.12	2.21	−0.60	3	0.26	0.27	0.35	0.29
Zonisamide	68291-97-4	1107)	2.33	0.64	0.64	2.58	−0.11	3	0.21	0.20	0.00	0.08
1,1,2,2-Tetrahydroperfluoro-1-decanol	678-39-7	0	2.67	5.47	5.47	2.02	−0.36	1	0.32	0.29	0.04	0.09
1,2,3-Trimethylbenzene	526-73-8	0	2.08	3.90	3.90	2.53	−0.11	7	0.00	0.00	0.02	0.03
1,2,4-Tribromobenzene	615-54-3	0	2.50	4.50	4.50	2.78	0.64	2	0.19	0.19	0.02	0.02
1,2,4-Trimethylbenzene	95-63-6	0	2.08	3.90	3.90	2.63	0.07	6	0.00	0.00	0.04	0.03
1,2-Dibromobenzene	583-53-9	0	2.37	3.64	3.64	2.58	−0.10	2	0.12	0.13	0.02	0.02
1,2-Dichloroethane	107-06-2	0	2.00	1.81	1.81	1.97	−0.33	4	0.00	0.00	0.05	0.04
1,2-Phenylenediamine	95-54-5	0	2.03	0.46	0.49	2.68	−0.11	2	0.00	0.00	0.02	0.01
1,3-Dinitrobenzene	99-65-0	0	2.23	1.54	1.54	2.67	−0.07	7	0.10	0.09	0.05	0.03
1,3-Di-o-tolylguanidine	97-39-2	0	2.38	0.49	0.69	1.59	−1.08	9	0.27	0.26	0.03	0.06
1,3-Diphenylguanidine	102-06-7	0	2.32	−0.34	−0.14	1.53	−1.27	9	0.26	0.25	0.00	0.02
1,3-Phenylenediamine	108-45-2	0	2.03	−0.07	−0.02	2.77	0.04	8	0.00	0.00	0.01	0.02
1,4-Dibromobenzene	106-37-6	0	2.37	3.63	3.63	2.66	0.23	1	0.14	0.13	0.05	0.02
1,4-Dioxane	123-91-1	0	1.94	−0.30	−0.30	2.22	−0.12	4	0.00	0.00	0.08	0.05
1,4-Phenylenediamine	106-50-3	0	2.03	−0.54	−0.10	2.45	−0.31	9	0.06	0.05	0.01	0.02
1,7-Dimethylxanthine	611-59-6	0	2.26	−0.19	−0.23	2.42	−0.22	6	0.19	0.18	0.10	0.09
1-Methylxanthine	6136-37-4	0	2.22	−0.43	−0.60	1.47	−0.46	10	0.16	0.15	0.02	0.03
1-Naphthaleneacetic acid	86-87-3	0	2.27	1.32	−0.03	3.17	1.31	3	0.05	0.05	0.00	0.04
2-(1H-Imidazol-2-yl)pyridine	18653-75-3	0	2.16	0.99	1.10	2.80	−0.05	7	0.08	0.07	0.02	0.02
2,3,5,6-Tetrafluorobenzoic acid	652-18-6	0	2.29	−0.97	−1.54	2.12	0.70	3	0.21	0.20	0.01	0.02
2,3,5,6-Tetrafluorobenzylalcohol	4084-38-2	0	2.26	1.57	1.57	2.89	0.13	4	0.11	0.11	0.04	0.04
2,3-Dimethylaniline	87-59-2	0	2.08	1.86	1.88	2.74	−0.06	4	0.00	0.00	0.04	0.04
2,4,6-Tribromophenol	118-79-6	0	2.52	3.79	2.78	2.89	1.18	10	0.16	0.17	0.02	0.02
2,4-Dibromophenol	615-58-7	0	2.40	3.25	3.11	2.77	0.53	4	0.14	0.15	0.02	0.02
2,4-Dimethylaniline	95-68-1	0	2.08	1.85	1.88	2.81	0.07	8	0.00	0.00	0.03	0.04
2,4-Dinitrophenol	51-28-5	0	2.26	−0.17	−1.25	3.03	0.22	9	0.12	0.12	0.01	0.02
2,5-Dimethylaniline	95-78-3	0	2.08	1.86	1.88	2.82	0.08	7	0.00	0.00	0.02	0.04
2,6-Dimethylaniline	87-62-7	0	2.08	1.53	1.53	2.75	−0.13	10	0.00	0.00	0.03	0.04
2-Aminobiphenyl	90-41-5	0	2.23	2.69	2.69	2.71	−0.01	9	0.05	0.05	0.00	0.02
2-Chloroaniline	95-51-2	0	2.11	2.11	2.11	2.85	−0.09	7	0.00	0.00	0.01	0.02
2-Chlorobenzoic acid	118-91-2	0	2.20	−1.16	−2.05	2.56	0.89	9	0.12	0.11	0.00	0.02
2-Chlorophenol	95-57-8	0	2.11	2.32	2.29	2.79	0.12	4	0.00	0.00	0.02	0.02
2-Hydroxybenzimidazole	615-16-7	0	2.13	1.36	1.36	2.78	0.04	8	0.01	0.02	0.02	0.02
2-Hydroxybiphenyl	90-43-7	0	2.23	3.12	3.12	2.43	−0.02	1	0.04	0.03	0.00	0.02
2-Hydroxyphenethyl alcohol	7768-28-7	0	2.14	1.13	1.13	2.72	−0.13	4	0.05	0.04	0.02	0.02
2-Hydroxyphenylacetic acid	614-75-5	0	2.18	−0.91	−2.25	2.06	0.50	3	0.08	0.06	0.00	0.02
2-Mercaptobenzimidazole	583-39-1	0	2.18	1.80	1.80	2.94	0.11	7	0.05	0.04	0.02	0.03
2-Mercaptoimidazole	872-35-5	0	2.00	−0.62	−0.62	1.93	0.05	8	0.00	0.00	0.02	0.05
2-Methoxy-4-nitroaniline	97-52-9	0	2.23	1.39	1.39	2.69	0.06	2	0.08	0.10	0.04	0.03
2-Methyl-1,4-naphthoquinone	58-27-5	1108)	2.24	1.90	1.90	2.39	−0.14	2	0.07	0.08	0.03	0.09
2-Nitrotoluene	88-72-2	0	2.14	2.46	2.46	2.75	−0.06	3	0.00	0.00	0.00	0.03
2-tert-Butylphenol	88-18-6	0	2.18	3.40	3.40	2.10	−0.42	2	0.00	0.00	0.03	0.04
3,4-Dimethylaniline	95-64-7	0	2.08	1.85	1.88	2.76	0.00	8	0.00	0.00	0.03	0.04
3,5-Dimethylaniline	108-69-0	0	2.08	1.86	1.88	2.69	−0.15	9	0.00	0.00	0.01	0.04
3-Aminobenzenesulfonic acid	121-47-1	0	2.24	−4.54	−4.65	1.30	0.01	6	0.36	0.32	0.01	0.02
3-Aminophenol	591-27-5	0	2.04	0.53	0.54	2.64	−0.05	7	0.00	0.00	0.01	0.02
3-Cyanopyridine	100-54-9	0	2.02	0.53	0.53	2.73	0.01	5	0.00	0.00	0.01	0.02
3-Ethylphenol	620-17-7	0	2.09	2.62	2.62	2.68	−0.13	7	0.00	0.00	0.02	0.03
3′-Hydroxyacetanilide	621-42-1	0	2.18	0.43	0.43	2.55	0.02	3	0.10	0.10	0.01	0.03
3-Hydroxybiphenyl	580-51-8	0	2.23	3.12	3.12	2.48	−0.09	6	0.02	0.03	0.01	0.01
3-Hydroxycoumarin	939-19-5	0	2.21	1.78	1.78	2.94	0.45	4	0.07	0.07	0.02	0.02
3-Hydroxyflavone	577-85-5	0	2.38	2.39	2.39	1.89	0.04	6	0.17	0.18	0.05	0.03
3-Nitroaniline	99-09-2	0	2.14	1.39	1.39	2.70	−0.07	7	0.04	0.03	0.03	0.03
3-Nitrophthalic acid	603-11-2	0	2.32	−3.39	−3.55	1.35	0.08	3	0.37	0.34	0.04	0.03
4,4'-Dihydroxybiphenyl	92-88-6	0	2.27	2.22	2.21	2.61	0.03	8	0.09	0.10	0.03	0.02
4-Acetamidobenzenesulfonyl chloride	121-60-8	0	2.37	1.02	1.02	1.42	0.04	6	0.23	0.22	0.03	0.05
4-Aminotoluene-3-sulfonic acid	88-44-8	0	2.27	−5.00	−5.09	1.24	0.08	6	0.41	0.37	0.02	0.02
4-Chloro-o-cresol	1570-64-5	0	2.16	2.82	2.82	2.82	0.05	8	0.00	0.00	0.01	0.03
4-Chlorophenol	106-48-9	0	2.11	2.51	2.51	2.71	0.07	7	0.00	0.00	0.01	0.02
4-Ethylphenol	123-07-9	0	2.09	2.62	2.62	2.62	−0.03	2	0.00	0.00	0.02	0.03
4-Hydroxy-2,6-dimethylaniline	3096-70-6	0	2.14	0.65	0.75	2.62	−0.15	3	0.06	0.06	0.02	0.04
4-Hydroxybiphenyl	92-69-3	0	2.23	3.12	3.11	2.65	0.09	9	0.03	0.03	0.00	0.01
4-Nitrotoluene-2-sulfonic acid	121-03-9	0	2.34	−4.42	−4.43	1.38	0.11	10	0.41	0.40	0.24	0.04
4-Nonylphenol	84852-15-3	0109)	2.34	6.09	6.09	1.95	−0.59	9	0.00	0.01	0.08	0.05
4-sec-Butylphenol	99-71-8	0	2.18	3.29	3.29	2.55	−0.10	2	0.00	0.00	0.02	0.04
4-α-Cumylphenol	599-64-4	0	2.33	4.39	4.39	2.34	0.00	2	0.06	0.06	0.04	0.08
5-(p-Tolyl)-1H-tetrazole	24994-04-5	0	2.20	0.04	−1.17	3.07	1.03	3	0.06	0.05	0.00	0.03
5-Amino-2-chlorotoluene-4-sulfonic acid	88-53-9	0	2.35	−4.51	−4.56	1.20	−0.05	7	0.38	0.41	0.02	0.02
6-Amino-1-naphthol-3-sulfonic acid	87-02-5	0	2.38	−4.10	−4.48	0.99	−0.05	5	0.37	0.40	0.01	0.03
6-Amino-2-naphthalenesulfonic acid	93-00-5	0	2.35	−4.45	−4.56	1.15	0.03	1	0.39	0.41	0.03	0.02
7-Ethoxycoumarin	31005-02-4	0	2.28	2.19	2.19	2.92	0.07	8	0.10	0.11	0.06	0.05
7-Hydroxycoumarin	93-35-6	0	2.21	1.58	1.44	2.93	0.15	4	0.07	0.07	0.03	0.03
7-Hydroxyflavone	6665-86-7	0	2.38	3.28	3.25	1.65	−0.09	4	0.13	0.14	0.03	0.03

a) 1, roles of efflux PgP or BCRP have been reported; 0, no roles or no info.

Multi Regression and Machine Learning Analyses

Univariate, bivariate, and trivariate linear regression analyses and trivariate logistic regression analysis with simple physicochemical properties were performed to generate the binary classification (Table 2) using Microsoft Excel and Prism software, respectively.¹⁴⁾ To evaluate the generalizability of the predictions, tenfold cross-validations were carried out for the trivariate linear regressions⁹⁾ using the split groups 1–10 indicated in Table 1. To verify the trivariate prediction system, estimated probabilities of active efflux for a secondary set of 100 medicines were generated in silico using the final above-established equations.

Table 2. Correlation Coefficients and Factors Obtained Using Univariate, Bivariate, and Trivariate Analyses of Physiological Properties for Estimating the Probability of Active Efflux for the 301 Chemicals to Achieve Binary Classification

	Property	Correlation coefficient	p-Value	Factor	95% confidence interval
Univariate analysis
	log MW	0.42	<0.01**	0.76	0.57 to 0.92
	log DpH 6.0	0.12	0.03*	−0.026	−0.050 to −0.002
	log DpH 7.4	0.06	0.27	−0.013	−0.035 to 0.010
	log Papp(AtoB)/(BtoA)	0.21	<0.01**	−0.13	−0.20 to −0.06
Bivariate analysis
log MW and log DpH 6.0		0.46	<0.01**
	log MW	-	<0.01**	0.80	0.62 to 0.99
	log DpH 6.0	-	<0.01**	−0.039	−0.060 to −0.017
	Intercept	-	<0.01**	−1.6	−2.0 to −1.2
log MW and log DpH 7.4		0.44	<0.01**
	log MW	-	<0.01**	0.80	0.61 to 0.99
	log DpH 7.4	-	<0.01**	−0.028	−0.048 to −0.007
	Intercept	-	<0.01**	−1.6	−2.1 to −1.2
log MW and log Papp (AtoB/BtoA)		0.43	<0.01**
	log MW	-	<0.01**	0.76	0.57 to 0.92
	log Papp(AtoB/BtoA)	-	0.057	−0.064	−0.13 to 0.002
	Intercept	-	<0.01**	−1.5	−1.9 to −1.02
Trivariate linear analysis
log MW, log DpH 6.0, and log DpH 7.4		0.47	<0.01**
	log MW	-	<0.01**	0.78	0.60 to 0.97
	log DpH 6.0	-	<0.01**	−0.10	−0.17 to −0.04
	log DpH 7.4	-	0.03*	0.065	−0.005 to −0.13
	Intercept	-	<0.01**	−1.6	−2.0 to −1.2
Trivariate logistic regression
log MW, log DpH 6.0, and log DpH 7.4		0.50	<0.01**
	log MW	-	-	6.3	4.5 to 8.4
	log DpH 6.0	-	-	−0.55	−0.96 to −0.15
	log DpH 7.4	-	-	0.21	−0.69 to −0.098
	Intercept	-	-	−17	−22 to −12

Predicted probability in linear regression = 0.78 × (log MW) − 0.10 × (log D_{pH 6.0}) + 0.065 × (log D_{pH 7.4}) −1.6. * p < 0.05: ** p < 0.01.

In an attempt to improve the accuracy of the predicted probabilities of active efflux, machine learning algorithms using LightGBM were adopted as described previously⁹⁾ with slight modifications. For each chemical substance, descriptor sets containing 1710 items related to chemical structural and physicochemical properties were obtained using open-source in silico programs (such as RDKit and Mordred), as described previously.⁹⁾ Estimation models were extensively validated with a modeling approach designed to integrate cross-validation⁹⁾ using the split groups 1–10 (indicated in Table 1). We thereby determined the optimum set of hyperparameters to obtain the highest correlation coefficients for the models and to perform initial evaluation of model performance. The estimated probabilities of active efflux for a secondary set of 100 medicines were also generated using the final above-established model in silico. There was no over fitting evaluated by performing Y-scrambling.

RESULTS AND DISCUSSION

To confirm that the evaluated substances exhibited adequate structural diversity, the chemical structures of the primary 301 and secondary 100 test substances were projected onto a two-dimensional chemical space with 25 subdivisions (Fig. 1). By means of the keyword search mentioned in Materials and Methods, the substances were divided by binary classification into two groups: a score of 1 for possible roles of PgP or BCRP transporters mentioned in the literature or a score of 0 for no apparent roles of transporters or no information (Table 1). The physicochemical properties of the test substances were estimated using in silico methods; the molecular weight (MW), log D_{pH 6.0}, log D_{pH 7.4}, and in silico-estimated P_{app A to B} and log(P_{app A to B}/P_{app B to A}) are given in Table 1.

The relevance was examined of in silico log(P_{app A to B}/P_{app B to A}) values generated by the previously established machine learning system for pH-dependent Caco-2 systems⁴⁾ to the binary classification values based on the reported roles of PgP or BCRP efflux (Fig. 2). A histogram of the apparent differences in intestinal influx and efflux log P_app values for the primary 301 chemicals is shown in Fig. 2A. The mean and median in silico-generated log(P_{app A to B}/P_{app B to A}) values of 70 substances with active efflux transport (i.e., with binary efflux scores of 1) were −0.42 and −0.25, respectively; these log(P_{app A to B}/P_{app B to A}) values were lower than those of −0.08 and −0.06 for the 231 substances with no roles for active efflux or no information (i.e., with binary scores of 0). The two binary groups were significantly different with p < 0.01 by the unpaired t-test. An ROC curve is shown in Fig. 2B; the area under the ROC curve was 0.64 [95% confidence interval (CI) 0.57–0.72] for in silico log(P_{app A to B}/P_{app B to A}).

Fig. 2. Histogram of Estimated Intestinal Influx and Efflux LogP_app Values for the Primary Set of 301 Chemicals (A) and the Receiver Operating Characteristic (ROC) Curve (B)

Estimates were generated using a light gradient boosting machine learning system (LightGBM) that incorporated 17 and 19 in silico chemical descriptors for influx and efflux, respectively.⁴⁾ Reported roles of efflux P-glycoprotein or breast cancer resistant protein are shown as 1 in the binary classification (gray bars), whereas no roles or no information are shown as 0 (white bars).

Univariate linear regression analyses for the 70 substances with active efflux transport and the 231 substances without apparent active efflux transport revealed that the observed binary classification of the 301 compounds was correlated significantly (but with relatively low correlation coefficients) with log MW (r = 0.42, p < 0.01), with log(P_{app A to B}/P_{app B to A}) (r = 0.21, p < 0.01), and with log D_{pH 6.0} (r = 0.12, p = 0.03) under the present conditions (Table 2). Bivariate analyses established that the observed binary classification of the 301 substances was correlated with the log MW and log D_{pH 6.0} (r = 0.46, p < 0.01), log MW and log D_{pH 7.4} (r = 0.44, p < 0.01), and with log MW and log(P_{app A to B}/P_{app B to A}) (r = 0.43, p < 0.01); for the latter pair, the calculated factor for log(P_{app A to B}/P_{app B to A}) was not significant, whereas all other factors were significant. (Table 2). Trivariate analysis revealed that the binary classification of the 301 compounds was significantly correlated with the log MW, log D_{pH 6.0}, and log D_{pH 7.4} (r = 0.47, p < 0.01) (Table 2). Although further multivariate analyses were performed with other combinations of three chemical parameters from among log MW, log D_{pH 6.0}, log D_{pH 7.4}, and log(P_{app A to B}/P_{app B to A}), no further improvement in correlation coefficient was evident (data not shown). Although solubility is another important factor influencing the drug absorption, in our previous study,⁴⁾ the in silico solubility (log S) values of substances did not contribute to the estimation for experimentally observed the log P_{app A to B} or log P_{app B to A}. In the similar way, the current liner regression system was not improved with the in silico solubility. This procedure led to the following equation: predicted probability of active efflux = 0.78 × (log MW) − 0.10 × (log D_{pH 6.0}) + 0.065 × (log D_{pH 7.4}) − 1.6 (p < 0.01). To confirm this model, trivariate logistic regression analysis was also performed (Table 2), resulting in another following equation: predicted probability (in logistic regression) = 6.3 × (log MW) − 0.55 × (log D_{pH 6.0}) + 0.29 × (log D_{pH 7.4}) − 17 (p < 0.01). There were a good correlation coefficient of 0.95 and a slope of 1.00 between the 301 predicted probability values in the linear and logistic regression.

To optimize the generalization capabilities of the predictions of probability of active efflux, 10-fold cross-validation processes were carried out for the trivariate regression. A histogram of the estimated probabilities for the primary 301 substances is shown in Fig. 3A. The mean and median estimated probabilities of the 70 substances with active efflux transport (i.e., with binary classification scores of 1) were 0.41 and 0.39, respectively; these values were higher than those of 0.19 and 0.15 for the 231 substances with no role for active efflux or no relevant information (i.e., with binary scores of 0). The difference between the two groups was significant by unpaired t-test (p < 0.01). Under the current conditions, the balanced accuracy was 0.77 with a threshold probability of 0.27. The corresponding ROC curve is shown in Fig. 3B; the area under the ROC curve was 0.81 (95%CI 0.76–0.87). The trivariate predictions were verified after cross-validation processes and finalized. Using the secondary set of 100 medicines shown in Supplementary Table S1, trivariate probability predictions were performed (Fig. 3C). The mean probability of 0.45 estimated using the trivariate equation for the 52 substances with active efflux transport was apparently higher than that of 0.38 for the 48 substances without active efflux transport; however, the difference between these binary groups was not significant by unpaired t-test (p = 0.09), and the accuracy was 0.66. The ROC curve is given in Fig. 3D; the area under the ROC curve was 0.66 (95%CI 0.51–0.80). Although the simple trivariate regression system had good potential for predicting the internal 301 data set after cross-validation, when it was applied to the outer 100 new medicines, its potential in predicting the binary classification was not good under the present conditions. These findings should be the apparent limitation that the simple trivariate regression system on general chemicals, especially substances not actively transported, might affect the predictability for application to a new medicine group.

Fig. 3. Histograms (A, C) and Receiver Operating Characteristic (ROC) Curves (B, D) for the Primary Set of 301 Chemicals (A, B) and the Secondary Set of 100 Medicines (C, D) for Which Possible Contributions of Efflux Transporters to Intestinal Permeability Were Estimated by Trivariate Analyses with the Following Input Parameters: Molecular Weight and Octanol–Water Distribution Coefficients at the Apical pH of 6.0 and at the Basal pH of 7.4⁴⁾

Reported roles of efflux P-glycoprotein or breast cancer resistant protein are shown in binary classification as 1 (grey bars); no roles or no information are shown as 0 (white bars).

As a result of these findings, we aimed to achieve higher accuracy for predictions of the binary classification of the 301 test compounds by using a machine learning algorithm that followed the approach reported previously, with slight modifications.⁴⁾ To estimate the probability of the role of efflux transport, we applied machine learning algorithms that used the 30 physicochemical descriptors shown in Supplementary Table S2 as input values. A histogram of estimated probabilities using the machine learning algorithm for the 301 test compounds is given in Fig. 4A. The estimated mean and median probabilities for the 70 test substances known to have active efflux transport were 0.50 and 0.47, respectively; these values were higher than those of 0.26 and 0.04 for the 231 test substances with no active efflux transport or no information. The difference between the two groups was significant, with p < 0.01 by unpaired t-test. Under the current conditions, the balanced accuracy was 0.77 with a threshold value of 0.20. The corresponding ROC curve is shown in Fig. 4B; the area under the ROC curve was 0.79 (95%CI 0.73–0.85). The predictions were verified after cross-validation processes and were finalized. Using the secondary set of 100 medicines shown in Supplementary Table S1, finalized machine learning predictions were performed (Fig. 4C). The mean and median probabilities of the 52 medicines with active efflux transport were 0.57 and 0.61, respectively; these values were higher than those of 0.32 and 0.25 for the 48 medicines with no active efflux transport or no information. The difference between the two groups was significant by unpaired t-test (p < 0.01) and the accuracy was 0.67. The corresponding ROC curve is given in Fig. 4D; the area under the ROC curve was 0.76 (95%CI 0.67–0.86).

Fig. 4. Histograms (A, C) and Receiver Operating Characteristic (ROC) Curves (B, D) for the Primary Set of 301 Chemicals (A, B) and the Secondary Set of 100 Medicines (C, D)

The histograms show the estimated probability of contributions of efflux transporters to intestinal permeability predicted using a new machine learning system with 30 in silico-generated input parameters consisting of the physicochemical descriptors shown in Supplementary Table S2. Reported roles of efflux P-glycoprotein or breast cancer resistant protein are shown in binary classification as 1 (grey bars); no roles or no information are shown as 0 (white bars).

The above-described computational methods represent a new alternative approach that could contribute to chemical safety screening by giving context to the results of physiologically based pharmacokinetic modeling after virtual oral administrations. In constructing detailed physiologically based pharmacokinetic models, active efflux clearance and passive permeation clearance should be separated to describe the concentration-dependent cellular membrane permeation of a compound. However, this newly developed approach to the estimation of the probability of contributions of efflux transporters to low apparent intestinal permeabilities of chemicals will contribute to the effectiveness of computational toxicology for assessing the potential risk of industrial chemicals and/or food components. The simplified models described here could be easily applied to predict a variety of drug exposures and have potential for use by a wide range of industry researchers and regulatory authorities. The current system based on influx and efflux log P_app values of substances may be also a possible survey tool for the asymmetrical transport of non-transporter substrates in future. Anyway, a combination of the previously established system⁴⁾ for estimating the permeability coefficients across intestinal cell monolayers (a continuous variable) and the currently proposed in silico estimation system for the roles of active efflux in addition to passive diffusion in membrane permeabilities (a binary classification) could be useful in drug research and toxicological assessments. In conclusion, high accuracy in predicting the binary classification was achieved by applying machine learning based on in silico-derived physicochemical descriptors, and the generalizability of the results was ensured by choosing a set of test substances with diverse chemical structures.

Acknowledgments

The authors thank Drs. Kimito Funatsu, Fumiaki Shono, Masato Kitajima, Junya Ohori, Kentaro Handa, Hiroshi Yano, Rie Saito, Izumi Sano, Masaya Fujii, Jun Tomizawa, Wataru Kobari, Airi Kato, and Norie Murayama for their assistance and David Smallbones for copyediting a draft of this article. This work was supported in part by the Japan Chemical Industry Association Long-range Research Initiative Program.

Conflict of Interest

The authors declare no conflict of interest.

Supplementary Materials

This article contains supplementary materials.

REFERENCES

• 1) Dahan A, González-Álvarez I. Regional intestinal drug absorption: biopharmaceutics and drug formulation. Pharmaceutics, 13, 272 (2021).
• 2) Neuhoff S, Ungell AL, Zamora I, Artursson P. pH-Dependent passive and active transport of acidic drugs across Caco-2 cell monolayers. Eur. J. Pharm. Sci., 25, 211–220 (2005).
• 3) Neuhoff S, Ungell AL, Zamora I, Artursson P. pH-dependent bidirectional transport of weakly basic drugs across Caco-2 monolayers: implications for drug-drug interactions. Pharm. Res., 20, 1141–1148 (2003).
• 4) Kamiya Y, Omura A, Hayasaka R, Saito R, Sano I, Handa K, Ohori J, Kitajima M, Shono F, Funatsu K, Yamazaki H. Prediction of permeability across intestinal cell monolayers for 219 disparate chemicals using in vitro experimental coefficients in a pH gradient system and in silico analyses by trivariate linear regressions and machine learning. Biochem. Pharmacol., 192, 114749 (2021).
• 5) Kamiya Y, Otsuka S, Miura T, Takaku H, Yamada R, Nakazato M, Nakamura H, Mizuno S, Shono F, Funatsu K, Yamazaki H. Plasma and hepatic concentrations of chemicals after virtual oral administrations extrapolated using rat plasma data and simple physiologically based pharmacokinetic models. Chem. Res. Toxicol., 32, 211–218 (2019).
• 6) Yamazaki H, Kamiya Y. Extrapolation of hepatic concentrations of industrial chemicals using pharmacokinetic models to predict hepatotoxicity. Toxicol. Res., 35, 295–301 (2019).
• 7) Kamiya Y, Takaku H, Yamada R, Akase C, Abe Y, Sekiguchi Y, Murayama N, Shimizu M, Kitajima M, Shono F, Funatsu K, Yamazaki H. Determination and prediction of permeability across intestinal epithelial cell monolayer of a diverse range of industrial chemicals/drugs for estimation of oral absorption as a putative marker of hepatotoxicity. Toxicol. Rep., 7, 149–154 (2020).
• 8) Kamiya Y, Otsuka S, Miura T, Yoshizawa M, Nakano A, Iwasaki M, Kobayashi Y, Shimizu M, Kitajima M, Shono F, Funatsu K, Yamazaki H. Physiologically based pharmacokinetic models predicting renal and hepatic concentrations of industrial chemicals after virtual oral doses in rats. Chem. Res. Toxicol., 33, 1736–1751 (2020).
• 9) Kamiya Y, Handa K, Miura T, Yanagi M, Shigeta K, Hina S, Shimizu M, Kitajima M, Shono F, Funatsu K, Yamazaki H. In silico prediction of input parameters for simplified physiologically based pharmacokinetic models for estimating plasma, liver, and kidney exposures in rats after oral doses of 246 disparate chemicals. Chem. Res. Toxicol., 34, 507–513 (2021).
• 10) Kamiya Y, Handa K, Miura T, Ohori J, Kato A, Shimizu M, Kitajima M, Yamazaki H. Machine learning prediction of the three main input parameters of a simplified physiologically based pharmacokinetic model subsequently used to generate time-dependent plasma concentration data in humans after oral doses of 212 disparate chemicals. Biol. Pharm. Bull., 45, 124–128 (2022).
• 11) Asano T, Ishihara K, Morota T, Takeda S, Aburada M. Permeability of the flavonoids liquiritigenin and its glycosides in licorice roots and davidigenin, a hydrogenated metabolite of liquiritigenin, using human intestinal cell line Caco-2. J. Ethnopharmacol., 89, 285–289 (2003).
• 12) Bowles SL, Ntamo Y, Malherbe CJ, Kappo AM, Louw J, Muller CJ. Intestinal transport and absorption of bioactive phenolic compounds from a chemically characterized aqueous extract of Athrixia phylicoides. J. Ethnopharmacol., 200, 45–50 (2017).
• 13) Kamiya Y, Handa K, Miura T, Ohori J, Shimizu M, Kitajima M, Shono F, Funatsu K, Yamazaki H. An Updated in silico prediction method for volumes of systemic circulation of 323 disparate chemicals for use in physiologically based pharmacokinetic models to estimate plasma and tissue concentrations after oral doses in rats. Chem. Res. Toxicol., 34, 2180–2183 (2021).
• 14) Shimura K, Murayama N, Tanaka S, Onozeki S, Yamazaki H. Suitable albumin concentrations for enhanced drug oxidation activities mediated by human liver microsomal cytochrome P450 2C9 and other forms predicted with unbound fractions and partition/distribution coefficients of model substrates. Xenobiotica, 49, 557–562 (2019).
• 15) Raub TJ, Wishart GN, Kulanthaivel P, Staton BA, Ajamie RT, Sawada GA, Gelbert LM, Shannon HE, Sanchez-Martinez C, De Dios A. Brain exposure of two selective dual CDK4 and CDK6 inhibitors and the antitumor activity of CDK4 and CDK6 inhibition in combination with temozolomide in an intracranial glioblastoma xenograft. Drug Metab. Dispos., 43, 1360–1371 (2015).
• 16) Koenderink JB, van den Heuvel J, Bilos A, Vredenburg G, Vermeulen NPE, Russel FGM. Human multidrug resistance protein 4 (MRP4) is a cellular efflux transporter for paracetamol glutathione and cysteine conjugates. Arch. Toxicol., 94, 3027–3032 (2020).
• 17) Nakamura M, Fujita K, Toyoda Y, Takada T, Hasegawa H, Ichida K. Investigation of the transport of xanthine dehydrogenase inhibitors by the urate transporter ABCG2. Drug Metab. Pharmacokinet., 33, 77–81 (2018).
• 18) Hodin S, Basset T, Jacqueroux E, Delezay O, Clotagatide A, Perek N, Mismetti P, Delavenne X. In vitro comparison of the role of P-glycoprotein and breast cancer resistance protein on direct oral anticoagulants disposition. Eur. J. Drug Metab. Pharmacokinet., 43, 183–191 (2018).
• 19) Saaby L, Helms HC, Brodin B. IPEC-J2 MDR1, a novel high-resistance cell line with functional expression of human P-glycoprotein (ABCB1) for drug screening studies. Mol. Pharm., 13, 640–652 (2016).
• 20) Hochman JH, Pudvah N, Qiu J, Yamazaki M, Tang C, Lin JH, Prueksaritanont T. Interactions of human P-glycoprotein with simvastatin, simvastatin acid, and atorvastatin. Pharm. Res., 21, 1686–1691 (2004).
• 21) Sugie M, Asakura E, Zhao YL, Torita S, Nadai M, Baba K, Kitaichi K, Takagi K, Takagi K, Hasegawa T. Possible involvement of the drug transporters P glycoprotein and multidrug resistance-associated protein Mrp2 in disposition of azithromycin. Antimicrob. Agents Chemother., 48, 809–814 (2004).
• 22) Posada MM, Cannady EA, Payne CD, Zhang X, Bacon JA, Pak YA, Higgins JW, Shahri N, Hall SD, Hillgren KM. Prediction of transporter-mediated drug–drug interactions for baricitinib. Clin. Transl. Sci., 10, 509–519 (2017).
• 23) Tamai I, Takanaga H, Maeda H, Sai Y, Ogihara T, Higashida H, Tsuji A. Participation of a proton-cotransporter, MCT1, in the intestinal transport of monocarboxylic acids. Biochem. Biophys. Res. Commun., 214, 482–489 (1995).
• 24) Oshida K, Shimamura M, Seya K, Ando A, Miyamoto Y. Identification of transporters involved in beraprost sodium transport in vitro. Eur. J. Drug Metab. Pharmacokinet., 42, 117–128 (2017).
• 25) Bachmakov I, Werner U, Endress B, Auge D, Fromm MF. Characterization of beta-adrenoceptor antagonists as substrates and inhibitors of the drug transporter P-glycoprotein. Fundam. Clin. Pharmacol., 20, 273–282 (2006).
• 26) Sieppi E, Vahakangas K, Rautio A, Ietta F, Paulesu L, Myllynen P. The xenoestrogens, bisphenol A and para-nonylphenol, decrease the expression of the ABCG2 transporter protein in human term placental explant cultures. Mol. Cell. Endocrinol., 429, 41–49 (2016).
• 27) Mamidi RNVS, Dallas S, Sensenhauser C, Lim HK, Scheers E, Verboven P, Cuyckens F, Leclercq L, Evans DC, Kelley MF, Johnson MD, Snoeys J. In vitro and physiologically-based pharmacokinetic based assessment of drug–drug interaction potential of canagliflozin. Br. J. Clin. Pharmacol., 83, 1082–1096 (2017).
• 28) Yang L, Li M, Wang F, Zhen C, Luo M, Fang X, Zhang H, Zhang J, Li Q, Fu L. Ceritinib enhances the efficacy of substrate chemotherapeutic agent in human ABCB1-overexpressing leukemia cells in vitro, in vivo and ex-vivo. Cell. Physiol. Biochem., 46, 2487–2499 (2018).
• 29) Hemmer MJ, Salinas KA, Harris PS. Application of protein expression profiling to screen chemicals for androgenic activity. Aquat. Toxicol., 103, 71–78 (2011).
• 30) Fang Y, Cao W, Liang F, Xia M, Pan S, Xu X. Structure affinity relationship and docking studies of flavonoids as substrates of multidrug-resistant associated protein 2 (MRP2) in MDCK/MRP2 cells. Food Chem., 291, 101–109 (2019).
• 31) Merino G, Alvarez AI, Pulido MM, Molina AJ, Schinkel AH, Prieto JG. Breast cancer resistance protein (BCRP/ABCG2) transports fluoroquinolone antibiotics and affects their oral availability, pharmacokinetics, and milk secretion. Drug Metab. Dispos., 34, 690–695 (2006).
• 32) Zimmermann ES, de Miranda Silva C, Neris C, Torres B, Schmidt S, Dalla Costa T. Population pharmacokinetic modeling to establish the role of P-glycoprotein on ciprofloxacin distribution to lung and prostate following intravenous and intratracheal administration to Wistar rats. Eur. J. Pharm. Sci., 127, 319–329 (2019).
• 33) Sun X, Li J, Guo C, Xing H, Xu J, Wen Y, Qiu Z, Zhang Q, Zheng Y, Chen X, Zhao D. Pharmacokinetic effects of curcumin on docetaxel mediated by OATP1B1, OATP1B3 and CYP450s. Drug Metab. Pharmacokinet., 31, 269–275 (2016).
• 34) Kosloski MP, Bow DAJ, Kikuchi R, Wang H, Kim EJ, Marsh K, Mensa F, Kort J, Liu W. Translation of in vitro transport inhibition studies to clinical drug–drug interactions for glecaprevir and pibrentasvir. J. Pharmacol. Exp. Ther., 370, 278–287 (2019).
• 35) Zurth C, Koskinen M, Fricke R, Prien O, Korjamo T, Graudenz K, Denner K, Bairlein M, von Buhler CJ, Wilkinson G, Gieschen H. Drug–drug interaction potential of darolutamide: in vitro and clinical studies. Eur. J. Drug Metab. Pharmacokinet., 44, 747–759 (2019).
• 36) Hashimoto N, Nakamichi N, Yamazaki E, Oikawa M, Masuo Y, Schinkel AH, Kato Y. P-Glycoprotein in skin contributes to transdermal absorption of topical corticosteroids. Int. J. Pharm., 521, 365–373 (2017).
• 37) Kanaan M, Daali Y, Dayer P, Desmeules J. Lack of interaction of the NMDA receptor antagonists dextromethorphan and dextrorphan with P-glycoprotein. Curr. Drug Metab., 9, 144–151 (2008).
• 38) Zhang Y, Han YH, Putluru SP, Matta MK, Kole P, Mandlekar S, Furlong MT, Liu T, Iyer RA, Marathe P, Yang Z, Lai Y, Rodrigues AD. Diclofenac and its acyl glucuronide: determination of in vivo exposure in human subjects and characterization as human drug transporter substrates in vitro. Drug Metab. Dispos., 44, 320–328 (2016).
• 39) Brück S, Strohmeier J, Busch D, Drozdzik M, Oswald S. Caco-2 cells - expression, regulation and function of drug transporters compared with human jejunal tissue. Biopharm. Drug Dispos., 38, 115–126 (2017).
• 40) Qiang F, Kang KW, Han HK. Repeated dosing of piperine induced gene expression of P-glycoprotein via stimulated pregnane-X-receptor activity and altered pharmacokinetics of diltiazem in rats. Biopharm. Drug Dispos., 33, 446–454 (2012).
• 41) Lo YL, Huang JD. Effects of sodium deoxycholate and sodium caprate on the transport of epirubicin in human intestinal epithelial Caco-2 cell layers and everted gut sacs of rats. Biochem. Pharmacol., 59, 665–672 (2000).
• 42) O’Brien FE, Clarke G, Dinan TG, Cryan JF, Griffin BT. Human P-glycoprotein differentially affects antidepressant drug transport: relevance to blood-brain barrier permeability. Int. J. Neuropsychopharmacol., 16, 2259–2272 (2013).
• 43) Parasrampuria DA, Mendell J, Shi M, Matsushima N, Zahir H, Truitt K. Edoxaban drug-drug interactions with ketoconazole, erythromycin, and cyclosporine. Br. J. Clin. Pharmacol., 82, 1591–1600 (2016).
• 44) Jodoin J, Demeule M, Beliveau R. Inhibition of the multidrug resistance P-glycoprotein activity by green tea polyphenols. Biochim. Biophys. Acta, 1542, 149–159 (2002).
• 45) Vaidyanathan JB, Walle T. Cellular uptake and efflux of the tea flavonoid (-)epicatechin-3-gallate in the human intestinal cell line Caco-2. J. Pharmacol. Exp. Ther., 307, 745–752 (2003).
• 46) Hariharan S, Gunda S, Mishra GP, Pal D, Mitra AK. Enhanced corneal absorption of erythromycin by modulating P-glycoprotein and MRP mediated efflux with corticosteroids. Pharm. Res., 26, 1270–1282 (2009).
• 47) Yamada M, Mendell J, Takakusa H, Shimizu T, Ando O. Pharmacokinetics, metabolism, and excretion of [(14)C]esaxerenone, a novel mineralocorticoid receptor blocker in humans. Drug Metab. Dispos., 47, 340–349 (2019).
• 48) Henthorn TK, Liu Y, Mahapatro M, Ng KY. Active transport of fentanyl by the blood-brain barrier. J. Pharmacol. Exp. Ther., 289, 1084–1089 (1999).
• 49) Kim RB, Leake BF, Choo EF, Dresser GK, Kubba SV, Schwarz UI, Taylor A, Xie HG, McKinsey J, Zhou S, Lan LB, Schuetz JD, Schuetz EG, Wilkinson GR. Identification of functionally variant MDR1 alleles among European Americans and African Americans. Clin. Pharmacol. Ther., 70, 189–199 (2001).
• 50) Generaux GT, Bonomo FM, Johnson M, Mahar Doan KM. Impact of SLCO1B1 (OATP1B1) and ABCG2 (BCRP) genetic polymorphisms and inhibition on LDL-C lowering and myopathy of statins. Xenobiotica, 41, 639–651 (2011).
• 51) Golstein PE, Boom A, van Geffel J, Jacobs P, Masereel B, Beauwens R. P-glycoprotein inhibition by glibenclamide and related compounds. Pflugers Arch., 437, 652–660 (1999).
• 52) Iwaki K, Sakaeda T, Kakumoto M, Nakamura T, Komoto C, Okamura N, Nishiguchi K, Shiraki T, Horinouchi M, Okumura K. Haloperidol is an inhibitor but not substrate for MDR1/P-glycoprotein. J. Pharm. Pharmacol., 58, 1617–1622 (2010).
• 53) Young GT, Emery EC, Mooney ER, Tsantoulas C, McNaughton PA. Inflammatory and neuropathic pain are rapidly suppressed by peripheral block of hyperpolarisation-activated cyclic nucleotide-gated ion channels. Pain, 155, 1708–1719 (2014).
• 54) Fujii S, Hayashi H, Itoh K, Yamada S, Deguchi Y, Kawazu K. Characterization of the carrier-mediated transport of ketoprofen, a nonsteroidal anti-inflammatory drug, in rabbit corneal epithelium cells. J. Pharm. Pharmacol., 65, 171–180 (2013).
• 55) de Souza J, Benet LZ, Huang Y, Storpirtis S. Comparison of bidirectional lamivudine and zidovudine transport using MDCK, MDCK-MDR1, and Caco-2 cell monolayers. J. Pharm. Sci., 98, 4413–4419 (2009).
• 56) Takahashi N, Miura M, Kameoka Y, Abumiya M, Sawada K. Drug interaction between lenalidomide and itraconazole. Am. J. Hematol., 87, 338–339 (2012).
• 57) Menzel K, Kothare P, McCrea JB, Chu X, Kropeit D. Absorption, metabolism, distribution, and excretion of letermovir. Curr. Drug Metab., 22, 784–794 (2021).
• 58) Shin HB, Jung EH, Kang P, Lim CW, Oh KY, Cho CK, Lee YJ, Choi CI, Jang CG, Lee SY, Bae JW. ABCB1 c.2677G > T/c.3435C > T diplotype increases the early-phase oral absorption of losartan. Arch. Pharm. Res., 43, 1187–1196 (2020).
• 59) Schleibinger H, Leberl C, Ruden H. Nitrated polycyclic aromatic hydrocarbons (nitro-PAH) in the suspended substances of the atmosphere. 2. Comparison of the mutagenicity of nitro-PAH and dust extracts of the air in the Ames, SOS repair induction and SCE test. Zentralbl. Hyg. Umweltmed., 188, 421–438 (1989).
• 60) Narumi K, Kobayashi M, Kondo A, Furugen A, Yamada T, Takahashi N, Iseki K. Characterization of loxoprofen transport in Caco-2 cells: the involvement of a proton-dependent transport system in the intestinal transport of loxoprofen. Biopharm. Drug Dispos., 37, 447–455 (2016).
• 61) Hagenbuch B. Cellular entry of thyroid hormones by organic anion transporting polypeptides. Best Pract. Res. Clin. Endocrinol. Metab., 21, 209–221 (2007).
• 62) Atsmon J, Dingemanse J, Shaikevich D, Volokhov I, Sidharta PN. Investigation of the effects of ketoconazole on the pharmacokinetics of macitentan, a novel dual endothelin receptor antagonist, in healthy subjects. Clin. Pharmacokinet., 52, 685–692 (2013).
• 63) Pan LL, Wang AY, Huang YQ, Luo Y, Ling M. Mangiferin induces apoptosis by regulating Bcl-2 and Bax expression in the CNE2 nasopharyngeal carcinoma cell line. Asian Pac. J. Cancer Prev., 15, 7065–7068 (2014).
• 64) Yamazaki T, Desai A, Goldwater R, Han D, Lasseter KC, Howieson C, Akhtar S, Kowalski D, Lademacher C, Rammelsberg D, Townsend R. Pharmacokinetic interactions between isavuconazole and the drug transporter substrates atorvastatin, digoxin, metformin, and methotrexate in healthy subjects. Clin. Pharmacol. Drug Dev., 6, 66–75 (2017).
• 65) Smalley J, Kadiyala P, Xin B, Balimane P, Olah T. Development of an on-line extraction turbulent flow chromatography tandem mass spectrometry method for cassette analysis of Caco-2 cell based bi-directional assay samples. J. Chromatogr. B Analyt. Technol. Biomed. Life Sci., 830, 270–277 (2006).
• 66) Wandel C, Kim R, Wood M, Wood A. Interaction of morphine, fentanyl, sufentanil, alfentanil, and loperamide with the efflux drug transporter P-glycoprotein. Anesthesiology, 96, 913–920 (2002).
• 67) Misaka S, Knop J, Singer K, Hoier E, Keiser M, Muller F, Glaeser H, Konig J, Fromm MF. The nonmetabolized beta-blocker nadolol is a substrate of OCT1, OCT2, MATE1, MATE2-K, and P-glycoprotein, but not of OATP1B1 and OATP1B3. Mol. Pharm., 13, 512–519 (2016).
• 68) Dorababu M, Nishimura A, Prabha T, Naruhashi K, Sugioka N, Takada K, Shibata N. Effect of cyclosporine on drug transport and pharmacokinetics of nifedipine. Biomed. Pharmacother., 63, 697–702 (2009).
• 69) Malfará BN, Benzi JRL, de Oliveira Filgueira GC, Zanelli CF, Duarte G, de Carvalho Cavalli R, de Moraes NV. ABCG2 c.421C > A polymorphism alters nifedipine transport to breast milk in hypertensive breastfeeding women. Reprod. Toxicol., 85, 1–5 (2019).
• 70) Mendes C, Meirelles GC, Silva MAS, Ponchel G. Intestinal permeability determinants of norfloxacin in Ussing chamber model. Eur. J. Pharm. Sci., 121, 236–242 (2018).
• 71) Brandl EJ, Tiwari AK, Lett TA, Shaikh SA, Lieberman JA, Meltzer HY, Kennedy JL, Muller DJ. Exploratory study on association of genetic variation in TBC1D1 with antipsychotic-induced weight gain. Hum. Psychopharmacol., 28, 183–187 (2013).
• 72) Mimura N, Nagata Y, Kuwabara T, Kubo N, Fuse E. P-glycoprotein limits the brain penetration of olopatadine hydrochloride, H1-receptor antagonist. Drug Metab. Pharmacokinet., 23, 106–114 (2008).
• 73) Pauli-Magnus C, Rekersbrink S, Klotz U, Fromm MF. Interaction of omeprazole, lansoprazole and pantoprazole with P-glycoprotein. Naunyn Schmiedebergs Arch. Pharmacol., 364, 551–557 (2001).
• 74) Bicker J, Fortuna A, Alves G, Soares-da-Silva P, Falcao A. Elucidation of the impact of P-glycoprotein and breast cancer resistance protein on the brain distribution of catechol-O-methyltransferase inhibitors. Drug Metab. Dispos., 45, 1282–1291 (2017).
• 75) Morimoto K, Nakakariya M, Shirasaka Y, Kakinuma C, Fujita T, Tamai I, Ogihara T. Oseltamivir (Tamiflu) efflux transport at the blood-brain barrier via P-glycoprotein. Drug Metab. Dispos., 36, 6–9 (2008).
• 76) Konishi Y, Hitomi Y, Yoshioka E. Intestinal absorption of p-coumaric and gallic acids in rats after oral administration. J. Agric. Food Chem., 52, 2527–2532 (2004).
• 77) Ogawa SI, Shimizu M, Kamiya Y, Uehara S, Suemizu H, Yamazaki H. Increased plasma concentrations of an antidyslipidemic drug pemafibrate co-administered with rifampicin or cyclosporine A in cynomolgus monkeys genotyped for the organic anion transporting polypeptide 1B1. Drug Metab. Pharmacokinet., 35, 354–360 (2020).
• 78) Kimura O, Fujii Y, Haraguchi K, Kato Y, Ohta C, Koga N, Endo T. Effects of perfluoroalkyl carboxylic acids on the uptake of sulfobromophthalein via organic anion transporting polypeptides in human intestinal Caco-2 cells. Biochem Biophys Rep, 24, 100807 (2020).
• 79) Kudo N, Katakura M, Sato Y, Kawashima Y. Sex hormone-regulated renal transport of perfluorooctanoic acid. Chem. Biol. Interact., 139, 301–316 (2002).
• 80) Zhang C, Kwan P, Zuo Z, Baum L. In vitro concentration dependent transport of phenytoin and phenobarbital, but not ethosuximide, by human P-glycoprotein. Life Sci., 86, 899–905 (2010).
• 81) Kasserra C, Assaf M, Hoffmann M, Li Y, Liu L, Wang X, Kumar G, Palmisano M. Pomalidomide: evaluation of cytochrome P450 and transporter-mediated drug-drug interaction potential in vitro and in healthy subjects. J. Clin. Pharmacol., 55, 168–178 (2015).
• 82) Nakano R, Oka M, Nakamura T, Fukuda M, Kawabata S, Terashi K, Tsukamoto K, Noguchi Y, Soda H, Kohno S. A leukotriene receptor antagonist, ONO-1078, modulates drug sensitivity and leukotriene C4 efflux in lung cancer cells expressing multidrug resistance protein. Biochem. Biophys. Res. Commun., 251, 307–312 (1998).
• 83) Elsby R, Smith V, Fox L, Stresser D, Butters C, Sharma P, Surry DD. Validation of membrane vesicle-based breast cancer resistance protein and multidrug resistance protein 2 assays to assess drug transport and the potential for drug-drug interaction to support regulatory submissions. Xenobiotica, 41, 764–783 (2011).
• 84) D’Emanuele A, Jevprasesphant R, Penny J, Attwood D. The use of a dendrimer-propranolol prodrug to bypass efflux transporters and enhance oral bioavailability. J. Control. Release, 95, 447–453 (2004).
• 85) Boulton DW, DeVane CL, Liston HL, Markowitz JS. In vitro P-glycoprotein affinity for atypical and conventional antipsychotics. Life Sci., 71, 163–169 (2002).
• 86) Pussard E, Merzouk M, Barennes H. Increased uptake of quinine into the brain by inhibition of P-glycoprotein. Eur. J. Pharm. Sci., 32, 123–127 (2007).
• 87) Wang J, Gan C, Retmana IA, Sparidans RW, Li W, Lebre MC, Beijnen JH, Schinkel AH. P-Glycoprotein (MDR1/ABCB1) and breast cancer resistance protein (BCRP/ABCG2) limit brain accumulation of the FLT3 inhibitor quizartinib in mice. Int. J. Pharm., 556, 172–180 (2019).
• 88) Dou L, Mai Y, Madla CM, Orlu M, Basit AW. P-Glycoprotein expression in the gastrointestinal tract of male and female rats is influenced differently by food. Eur. J. Pharm. Sci., 123, 569–575 (2018).
• 89) Muhrez K, Largeau B, Emond P, Montigny F, Halimi JM, Trouillas P, Barin-Le Guellec C. Single nucleotide polymorphisms of ABCC2 modulate renal secretion of endogenous organic anions. Biochem. Pharmacol., 140, 124–138 (2017).
• 90) Martel F, Keating E, Azevedo I. Effect of P-glycoprotein modulators on the human extraneuronal monoamine transporter. Eur. J. Pharmacol., 422, 31–37 (2001).
• 91) Gong IY, Mansell SE, Kim RB. Absence of both MDR1 (ABCB1) and breast cancer resistance protein (ABCG2) transporters significantly alters rivaroxaban disposition and central nervous system entry. Basic Clin. Pharmacol. Toxicol., 112, 164–170 (2013).
• 92) Keskitalo JE, Zolk O, Fromm MF, Kurkinen KJ, Neuvonen PJ, Niemi M. ABCG2 polymorphism markedly affects the pharmacokinetics of atorvastatin and rosuvastatin. Clin. Pharmacol. Ther., 86, 197–203 (2009).
• 93) Hanna I, Alexander N, Crouthamel MH, Davis J, Natrillo A, Tran P, Vapurcuyan A, Zhu B. Transport properties of valsartan, sacubitril and its active metabolite (LBQ657) as determinants of disposition. Xenobiotica, 48, 300–313 (2018).
• 94) Nabekura T, Yamaki T, Ueno K, Kitagawa S. Inhibition of P-glycoprotein and multidrug resistance protein 1 by dietary phytochemicals. Cancer Chemother. Pharmacol., 62, 867–873 (2008).
• 95) Urquhart BL, Ware JA, Tirona RG, Ho RH, Leake BF, Schwarz UI, Zaher H, Palandra J, Gregor JC, Dresser GK, Kim RB. Breast cancer resistance protein (ABCG2) and drug disposition: intestinal expression, polymorphisms and sulfasalazine as an in vivo probe. Pharmacogenet. Genomics, 18, 439–448 (2008).
• 96) O’Connor R, O’Leary M, Ballot J, Collins CD, Kinsella P, Mager DE, Arnold RD, O’Driscoll L, Larkin A, Kennedy S, Fennelly D, Clynes M, Crown J. A phase I clinical and pharmacokinetic study of the multi-drug resistance protein-1 (MRP-1) inhibitor sulindac, in combination with epirubicin in patients with advanced cancer. Cancer Chemother. Pharmacol., 59, 79–87 (2007).
• 97) Kashihara Y, Ieiri I, Yoshikado T, Maeda K, Fukae M, Kimura M, Hirota T, Matsuki S, Irie S, Izumi N, Kusuhara H, Sugiyama Y. Small-dosing clinical study: pharmacokinetic, pharmacogenomic (SLCO2B1 and ABCG2), and Interaction (atorvastatin and grapefruit juice) profiles of 5 probes for OATP2B1 and BCRP. J. Pharm. Sci., 106, 2688–2694 (2017).
• 98) Tamura S, Ohike A, Ibuki R, Amidon GL, Yamashita S. Tacrolimus is a class II low-solubility high-permeability drug: the effect of P-glycoprotein efflux on regional permeability of tacrolimus in rats. J. Pharm. Sci., 91, 719–729 (2002).
• 99) Higashi H, Watanabe N, Tamura R, Taguchi M. In vitro P-glycoprotein-mediated transport of tadalafil: a comparison with sildenafil. Biol. Pharm. Bull., 40, 1314–1319 (2017).
• 100) Ding PR, Tiwari AK, Ohnuma S, Lee JW, An X, Dai CL, Lu QS, Singh S, Yang DH, Talele TT, Ambudkar SV, Chen ZS. The phosphodiesterase-5 inhibitor vardenafil is a potent inhibitor of ABCB1/P-glycoprotein transporter. PLoS ONE, 6, e19329 (2011).
• 101) Pachot JI, Botham RP, Haegele KD, Hwang K. Experimental estimation of the role of P-glycoprotein in the pharmacokinetic behaviour of telithromycin, a novel ketolide, in comparison with roxithromycin and other macrolides using the Caco-2 cell model. J. Pharm. Pharm. Sci., 6, 1–12 (2003).
• 102) Wu ZX, Teng QX, Cai CY, Wang JQ, Lei ZN, Yang Y, Fan YF, Zhang JY, Li J, Chen ZS. Tepotinib reverses ABCB1-mediated multidrug resistance in cancer cells. Biochem. Pharmacol., 166, 120–127 (2019).
• 103) Zimmermann C, Gutmann H, Drewe J. Thalidomide does not interact with P-glycoprotein. Cancer Chemother. Pharmacol., 57, 599–606 (2006).
• 104) Konishi Y, Shimizu M. Transepithelial transport of ferulic acid by monocarboxylic acid transporter in Caco-2 cell monolayers. Biosci. Biotechnol. Biochem., 67, 856–862 (2003).
• 105) Teft WA, Morse BL, Leake BF, Wilson A, Mansell SE, Hegele RA, Ho RH, Kim RB. Identification and characterization of trimethylamine-N-oxide uptake and efflux transporters. Mol. Pharm., 14, 310–318 (2017).
• 106) Gschwind L, Rollason V, Daali Y, Bonnabry P, Dayer P, Desmeules JA. Role of P-glycoprotein in the uptake/efflux transport of oral vitamin K antagonists and rivaroxaban through the Caco-2 cell model. Basic Clin. Pharmacol. Toxicol., 113, 259–265 (2013).
• 107) Gonçalves J, Silva S, Gouveia F, Bicker J, Falcao A, Alves G, Fortuna A. A combo-strategy to improve brain delivery of antiepileptic drugs: focus on BCRP and intranasal administration. Int. J. Pharm., 593, 120161 (2021).
• 108) Oh SJ, Han HK, Kang KW, Lee YJ, Lee MY. Menadione serves as a substrate for P-glycoprotein: implication in chemosensitizing activity. Arch. Pharm. Res., 36, 509–516 (2013).
• 109) Loo TW, Clarke DM. Nonylphenol ethoxylates, but not nonylphenol, are substrates of the human multidrug resistance P-glycoprotein. Biochem. Biophys. Res. Commun., 247, 478–480 (1998).