VARIDT 2.0: structural variability of drug transporter.

The structural variability data of drug transporter (DT) are key for research on precision medicine and rational drug use. However, these valuable data are not sufficiently covered by the available databases. In this study, a major update of VARIDT (a database previously constructed to provide DTs' variability data) was thus described. First, the experimentally resolved structures of all DTs reported in the original VARIDT were discovered from PubMed and Protein Data Bank. Second, the structural variability data of each DT were collected by literature review, which included: (a) mutation-induced spatial variations in folded state, (b) difference among DT structures of human and model organisms, (c) outward/inward-facing DT conformations and (d) xenobiotics-driven alterations in the 3D complexes. Third, for those DTs without experimentally resolved structural variabilities, homology modeling was further applied as well-established protocol to enrich such valuable data. As a result, 145 mutation-induced spatial variations of 42 DTs, 1622 inter-species structures originating from 292 DTs, 118 outward/inward-facing conformations belonging to 59 DTs, and 822 xenobiotics-regulated structures in complex with 57 DTs were updated to VARIDT (https://idrblab.org/varidt/ and http://varidt.idrblab.net/). All in all, the newly collected structural variabilities will be indispensable for explaining drug sensitivity/selectivity, bridging preclinical research with clinical trial, revealing the mechanism underlying drug-drug interaction, and so on.

INTRODUCTION

Drug transporter (DT) is known as the key determinant of drug absorption, distribution, clearance and elimination (1,2), and its variability is reported to be critical for balancing drug efficacy with safety (3), reversing drug resistance (4) and predicting drug-drug interaction (5–8). Although the variability data of DT are essential for preclinical and clinical studies (9–11), they are extensively dispersed in literatures (12). The VARIDT 1.0 database (12) was therefore developed to explicitly describe three aspects of DT variability: (i) genetic polymorphism, and epigenetic regulation of DT; (ii) species/tissue/disease-specific DT abundance and (iii) exogenous factors modulating DT activity or altering the disposition of transported drug. Meanwhile, VARIDT 1.0 also enabled the interplay analysis among multiple aspects of DT variability (12). Due to its unique data coverage, VARIDT has emerged to be the indispensable complements to other transporter-related databases (13–31) in assessing drug safety (32–35), reversing drug resistance (36) and so on (37–40).

Recently, there are increasing demands on the 3D structural variability data of DT, which are expected to provide key information for the studies on precision medicine and rational drug use (41). Such data include: (a) the mutation-induced spatial variations in folded state that are critical for the understanding of drug sensitivity and selectivity (42), (b) the difference among the DT structures of human and model organisms that are key for bridging preclinical study with clinical trials (43), (c) the outward-facing and inward-facing conformations that are essential for elucidating the dynamics and underlying steps of transporting cycle (44) and (d) the xenobiotics-driven alterations in 3D complexes of DT that are crucial for revealing the mechanism underlying drug-drug interaction (45–49). Furthermore, due to the lack of DTs’ structures and their structural variability data, homology modeling is adopted as well-established protocol to construct protein structure (50), and the resulting structure together with its variability have been successfully and widely adopted to study protein-protein interaction (51), discover novel drug and epitope (52,53), promote protein engineering and design (54), and so on (55–57).

Till now, several online resources have been constructed and are still active to provide DT-related structure data (13–16,58–63). Some contain transporter structures as part of the broader collection of biological or pharmacological data (e.g. UniProt (13), TTD (58) and IUPHAR/BPS Guide to Pharmacology (16)); some others (e.g. MemProtMD (59), EncoMPASS (60), and ABCMdb (61)) provide 3D conformation collected from PDB database (14); the remaining (e.g. TCDB (15), and iMusta4SLC (62)) offer the hyperlinks redirecting to PDB entries (14). A preliminary assessment of PDB’s data (14) reveals that only a small fraction (∼14%) of all DTs in VARIDT 1.0 (12) have experimentally resolved structures available. To cope with the lack of DT structures, the SWISS-MODEL (63) can thus be adopted to predict DT structure by homology modeling, and AlphaFold (64,65) can facilitate the determination of transporter's structure. However, all the existing online resources do not systematically describe the 3D structural variability data of DTs, which asks for a major update of VARIDT 1.0 to provide the comprehensive information on describing all four aspects (i–iv) of DT structural variabilities (42–45).

In this study, a major update of VARIDT (structural variability of drug transporter) was therefore conducted, which systematically described all four aspects of DT structural variability. As shown in Figure 1, these availability data included: (a) the mutation-induced spatial variations in folded state, (b) the difference among the DT structures of human and model organisms, (c) the outward-facing and inward-facing conformations of the transporting cycle and (d) the xenobiotics-driven alterations in 3D complex. With the significant advances in the field of precision medicine, these data updated to the VARIDT 2.0 version (https://idrblab.org/varidt/ and http://varidt.idrblab.net/) will be indispensable for understanding drug sensitivity/selectivity, bridging preclinical research with clinical trial, elucidating the dynamics of drug transporting cycle, revealing the mechanisms underlying a variety of drug–drug interactions, and so on so forth.

Figure 1.

The structural variability data of drug transporter (DT) updated to the VARIDT 2.0. (a) the mutation-induced spatial variations in folded state; (b) the difference among the DT structures of human and model organisms; (c) the outward/inward-facing conformations of the transporting cycle; (d) the xenobiotics-driven alterations in 3D complex.

FACTUAL CONTENT AND DATA RETRIEVAL

Collection of structure data for drug transporters

The entire structures (covering the whole protein sequence) of wild-type DTs were first collected by searching the names or synonyms of all DTs in PubMed (66) and PDB (14) using the keyword combination of ‘[DT name] + structure’, ‘[DT name] + conformation’, ‘[DT name] + [substrate name] + structure’, ‘[DT name] + [substrate name] + complex’, ‘[DT name] + complexed with’, ‘[DT name] + in complex with’, and so on. Additionally, such entire structures of wild-type DTs could also be identified by blasting the full-length sequence of DT against all those sequences in PDB (14). Only the PubMed literature or PDB entry that explicitly described the entire structures of the wild-type DTs were collected and recorded in VARIDT 2.0. Second, for those DTs without their entire structures, the structures of their transmembrane domain or region were identified by searching or blasting against PubMed (66) and PDB (14), and the transmembrane structures were collected to represent the structures of the corresponding DTs. As a result, a total of 73 DTs were identified with experimentally resolved structure available. These structures were determined by diverse techniques (such as NMR spectroscopy, X-ray crystallography, and electron microscopy) based on various systems (such as Escherichia coli BL21, HEK293 cells, and SF9 cells), and the resolutions of most DT structures were within the range of 1.0 Å to 4.0 Å.

Apart from the structures of wild-type DT, many structure variants of DT were also described in PubMed (66) and PDB (14). Most of the variants (∼160 structures) were in complex with various endogenous ionic or small molecular substrates, and the remaining variants (∼50 structures) bond with different exogenous ligands (such as drugs, environmental chemicals, etc.).

Moreover, due to the lack of DTs’ structures and the structural variability data, a well-established protocol of homology modeling (50) was adopted to provide the explicit data on illustrating DTs’ structural variability (detailed information on homology modeling was provided in the following section entitled ‘Structural variability data generated by homology modeling’). Meanwhile, a variety of DT structures predicted by AlphaFold (64,65), a popular structure prediction tool based on Artificial Intelligence (AI), were collected for referencing DTs’ structural variability.

Structural variability data provided in VARIDT 2.0

Mutation-induced spatial variations in folded state

Mutation-induced spatial variations in the folded state of DT are essential for understanding drug sensitivity & selectivity (42,67), because of the resulting variations of interaction pattern, charge environment, hydrophobic property, and volume of each residue in drug binding pocket (68–74). Figure 2 showed the spatial variations in the folded state of human monocarboxylate transporter 1 (hMCT1) induced by a point mutation (D309N), which were explicitly recorded and described in VARIDT 2.0. hMCT1 was reported to drive the transmembrane transportation of several drugs that treated metabolic diseases, such as atorvastatin (75), and was known to be a promising target for immunosuppression (76). As described in the online VARIDT 2.0 and Figure 2A, the carboxyl group of D309 in wild-type hMCT1 electron microscopy structure interacted with the guanidine group of R313 through hydrogen bonds (68). In contrast, the hydrogen bonds between the above residues vanished, as shown in Figure 2B, due to the introduction of N309′s amino group in the mutated (D309N) structure of hMCT1. In other words, the point mutation of D309N released the key R313 residue, which broke its hydrogen bond interaction with the carbonyl oxygen of M151 in transmembrane domain 5 (TM5) and formed a new critical interaction with Y34 in TM1.

Figure 2.

Spatial variations in the folded state of human monocarboxylate transporter 1 (hMCT1) induced by a point mutation (D309N), which were described in VARIDT. (A) the carboxyl group of D309 in wild-type hMCT1 electron microscopy structure interacted with the guanidine group of R313 through hydrogen bonds (key residues were shown as sticks, and hydrogen bonds were highlighted as red dashed lines); (B) the point mutation of D309N released the key R313 residue, which broke its hydrogen bond interaction with the carbonyl oxygen of M151 in transmembrane domain 5 (TM5) and formed new critical interaction with Y34 in TM1 (key residues were provided as stick, and hydrogen bonds were highlighted as red dashed line); (C) the structure of wild-type hMCT1 folded to inward-facing state (D309 was shown in stick, and the distances between key TMs were labeled); (D) the structure of hMCT1 mutant folded to outward-facing state (N309 was shown in stick, and the distances between key TMs were labeled).

Moreover, as described in Figure 2C (68), the structure of wild-type hMCT1 folded to an inward-facing state. Because of the variations between interaction patterns in Figure 2A and B, the restrictions between TM8 and TM5 were loosened, while a new constraint between TM8 and TM1 was established, making the mutated structure of hMCT1 an outward-facing conformation (Figure 2D). Particularly, the distance between G19 in TM1 and M379 in TM10 was significantly enlarged from 18.4 Å in wild-type hMCT1 (Figure 2C) to 25.6 Å in the mutated structure (Figure 2D), and the distance between P144 in TM5 and N321 in TM8 was also expanded from 5.8 Å in wild-type hMCT1 to 11.3 Å in the mutated one (Figure 2C and D). All in all, the spatial variations between the folded states of wild-type and mutated DTs that were provided in VARIDT 2.0 could alter the affinity between drug and DT, and thus affect drug sensitivity/selectivity (68,77,78).

Difference among the DT structures of human and model organisms

Various model organisms were adopted in biomedical research with their distinct advantages and limitations, and the information obtained from these model organisms may not be able to reflect human physiology (79–81). The differences among DT structures of human and model organisms were therefore essential for bridging preclinical study with clinical trial (43,82). Figure 3 showed the difference between the crystal structures of human urea transporter 1 (hUT1) and bovine urea transporter 1 (UT1b), which were systematically recorded and described in VARIDT. As reported, the sequence identity between hUT1 and UT1b equaled to 80.99%, which guaranteed a similarly rapid urea diffusion following concentration gradient in both species (83). As shown in VARIDT 2.0 website and Figure 3A and B, although over 40 residues were non-conserved between hUT1 and UT1b (highlighted in red on the ribbon plot), their conformations looked highly similar, and it was reasonable to deduce that their functions in drug transportation might be similar.

Figure 3.

The differences between the crystal structures of human urea transporter 1 (hUT1) and bovine urea transporter 1 (UT1b), which were provided in VARIDT. Crystal structures (described in ribbon diagrams) of and non-conserved residues (highlighted in red) between hUT1 and UT1b were shown in (A) and (B), respectively. Drug transporting tunnel (grey ribbon diagram) and key residues (colored in green and orange for the non-polar and polar residues, respectively) of hUT1 and UT1b were illustrated in (C) and (D), respectively. Key residues were represented in sticks.

However, an in-depth investigation of those non-conserved residues in UT1′s transporting tunnel may tell you otherwise (Figure 3C and D). As shown, compared with the non-conserved residues (C87, A278, I292, T295 and G333) in UT1b's drug transporting tunnel, the corresponding amino acids (W92, F283, M297, M300 and C338) in hUT1 presented larger side chains, which resulted in strong steric hindrance and hindered drug transports. Moreover, three non-polar residues (W92, F136 and M300) in hUT1 were replaced by the polar ones (C87, Y131 and T295) in UT1b, which completely re-defined the physicochemical profiles of the drug transporting tunnel (shifting from a non-polar tunnel primarily colored in green to a polar one mainly colored in orange). All in all, those above structural difference shown in VARIDT were essential for bridging preclinical study with clinical trial and understanding the difference among species in drug transportation.

Outward-facing and inward-facing conformations

The transportation of a drug through DTs was a multi-step process, and the DT structure of high-resolution was the basis for understanding the structure dynamics of this complex process (68,84). Among these structures, the outward- and inward-facing conformations of DTs were reported as essential for elucidating the underlying steps of a transporting cycle (44,84–86). Figure 4 showed the outward-facing and inward-facing conformations of two typical DTs recoded in VARIDT 2.0: human monocarboxylate transporter 1 (hMCT1) and human P-glycoprotein (hMDR1). Both DTs were represented using ribbon plots and cylindrical helices on the left and right side, respectively. As described in Figure 4A, the outward-facing and inward-facing conformations of hMCT1 were viewed from three different perspectives: (i) parallel to the membrane, (ii) from the intra-cellular space and (iii) from the extra-cellular space. Clearly, there were distinct conformation variations between these two critical snaps of a drug transporting process, which could provide the valuable information to reconstruct the dynamic process of drug absorption, distribution or exclusion, and to design novel ligands that interfered with the hMCT1-affiliated signaling pathways.

Figure 4.

Outward-facing and inward-facing conformations of two DTs recoded in VARIDT 2.0, human monocarboxylate transporter 1 (hMCT1) & human P-glycoprotein (hMDR1). (A) hMCT1 conformations were viewed from different perspectives: parallel to the membrane (top), from the intra-cellular space (middle), and from extra-cellular space (bottom); (B) hMDR1 conformations were viewed parallel to the membrane. These structural conformations were shown using ribbon (on the left side) and cylindrical helices (on the right side).

Similarly, those outward-facing and inward-facing conformations of hMDR1 were illustrated in Figure 4B (viewed parallel to cell membrane). hMDR1 was one of the most famous DTs in drug exclusion and resistance, which was able to exclude both drug and toxic molecule from the cells, thereby protecting tissues from toxic substances (87). Compared with hMDR1′s outward-facing conformation, some transmembrane domains (TMs), in the inward-facing conformation, moved as a rigid body, while the remaining TMs underwent major rearrangement (85,88). All in all, the outward/inward-facing conformations of all DTs shown in VARIDT laid the solid foundation for depicting the mechanisms and dynamic processes underlying drug transportation.

Xenobiotics-driven alterations in 3D complexes

The affinities of DTs in transporting drugs could be extensively modulated by xenobiotics, which, in turn, affected the pharmacokinetics, efficacy and safety of the transported drugs (89). In other words, the xenobiotics-driven alterations in DT-included complexes were essential for revealing mechanism underlying drug-drug interactions (45,90–93). Figure 5 provided xenobiotics-driven alterations in the 3D complexes of two representative DTs recoded in VARIDT 2.0: human breast cancer resistance protein (hABCG2) and human serotonin transporter (hSERT). hABCG2 could exclude a variety of chemotherapeutic drugs from cancer cells, which resulted in a resistance to these anti-cancer drugs (84,94,95). As shown in Figure 5A, two typical conformational states of hABCG2 were explicitly described in VARIDT 2.0: (i) apo-closed conformation on the left side and (ii) inhibitor trapped inward-facing conformation on the right side (84). Particularly, in apo-closed state, TM2 and TM5 helixes form a tightly packed helical bundle near the cytosolic region, sealing off the crevice at the dimer interface. Meanwhile, the xenobiotics mitoxantrone (DHAQ) drove the conformation alterations in TM2 and TM5, which resulted in the shifts of hABCG2 to its inward facing conformation. In other words, DHAQ stabilized hABCG2 in its inward-facing conformation, and could thus affect not only the transporting cycle but also the drug efficacy.

Figure 5.

Xenobiotics-driven alterations in the 3D complexes of two representative DTs recoded in VARIDT: human breast cancer resistance protein (hABCG2) and human serotonin transporter (hSERT). (A) two typical conformational states of hABCG2 were described in VARIDT 2.0: the apo-closed conformation on the left side & the inhibitor trapped inward-facing conformation by mitoxantrone (DHAQ, colored in violet) on the right side (two key domains TM2 and TM5 were colored in green and orange, respectively); (B) the experimental hSERT structures in complexed with two antidepressants (paroxetine & fluvoxamine) were illustrated, and the S1 site was shown by violet surface and both drugs were represented by yellow spheres.

Moreover, the structural information of DT complexes collected in VARIDT 2.0 varied according to their complexed ligands (xenobiotics). Particularly, hSERT was popular target of the marketed antidepressants, and Figure 5B showed its experimentally resolved structures in complex with two drugs (paroxetine & fluvoxamine, both were the selective serotonin reuptake inhibitors). The binding of different drugs could induce alteration in the S1 binding pocket (shown in violet). In VARIDT, the alteration data driven by different xenobiotics in the 3D complexes of DTs could thus give key information for revealing the mechanism underlying drug–drug interaction.

Structural variability data generated by homology modeling

A preliminary assessment of PDB (14) revealed that only a small fraction of the DTs in VARIDT 1.0 (12) have experimentally resolved structure available. These results were consistent with the previous reports (50,96), which highlighted the challenges in the expression or crystallization of membrane proteins (97). Therefore, homology modeling had emerged as a valuable approach for generating transporter structures (98). So far, it had been applied as well-established protocol to build protein structure (50), and the resulting structures and their variability data had been widely and successfully adopted to study protein-protein interactions (51), discover new drug or epitope (52,53), promote protein engineering or design (54,99), identify thermostabilizing mutations (55), and predict the molecular mechanisms underlying drug transportation (100).

In this study, the well-established protocol RosettaCM was adopted to generate structural models and variabilities for each DT (50). As shown in Figure 6, the adopted protocol consisted of four key steps. (S1) identifying suitable structure templates through sequence alignment against PDB (14) & threading DT sequence over single/multiple top-ranked templates; (S2) parsing sequence to 3mer and 9mer fragments to generate fragment file by Rosetta server (101); (S3) forming five model candidates for comparatively modeling each DT using RosettaCM Hybridize module; (S4) optimizing the generated structures by energy minimization, and evaluating the structure quality by assessing from two different perspectives. To describe the reliability of the generated structure, the resulting energy scores of each computationally modelled structure and its quality evaluation outcomes were fully provided in VARIDT, which were further described in the following section entitled ‘Evaluation of the modelled structural variability data’.

Figure 6.

The protocol of comparative modeling used in this study to generate structural models of DTs together with their structural variability data. (S1) identifying suitable structure templates through sequence alignment against PDB and threading DT sequence over top-ranked templates; (S2) parsing sequence into 3mer and 9mer fragments to generate fragment file; (S3) forming five model candidates for comparatively modeling each DT; (S4) optimizing the generated structures by energy minimization, and evaluating the quality of the generated structures.

For modeling the mutation-induced spatial variations, the nucleotide polymorphism data within the coding region of DTs affecting drugs’ transportation were first identified from VARIDT (12). Then, these polymorphisms were mapped to DT sequence for template discovery. For modeling, the species-specific structure differences, DT sequences of various model organisms (such as rat, mouse, zebrafish, bovine and rabbit) and human were first collected. These sequences were then used to identify their structural templates. For modeling outward & inward-facing conformations, the special attentions were paid to the conformation orientation of the modelled DTs during their identifications of structural template. For modeling xenobiotics-driven alteration, the xenobiotics capable of altering DT activity were first identified from VARIDT (12). Then, the widely applied flexible docking strategy, induced fit docking (102), was employed to generate the complex.

Evaluation of the modelled structural variability data

Two popular criteria were used to evaluate the structure generated by homology modeling, which included the (a) percentage of the modelled residues within the favored region of Ramachandran plot (103) and (b) QMEANBrane score specifically designed to measure modeling qualities for membrane protein (104). Particularly, the Ramachandran plot described the GLOBAL geometric attributes of the modelled structure by visualizing the dihedral angles of all residues in a structure, and the QMEANBrane score illustrated the LOCAL qualities of an alpha-helical transmembrane protein model by combining statistical potential with a per-residue weighting scheme (103,104). Due to the innate independence between these two criteria, they were collectively considered in VARIDT 2.0 to evaluate those modelled structural variabilities. For criterion (a), the percentage of modelled residues within the favored Ramachandran region (≥98%, 95–98% and <95%) were accepted in VARIDT 2.0 database as: Excellent, Medium and Poor, respectively (Figure 7A). For criterion (b), the QMEANBrane score of ≥0.8, 0.6–0.8 and <0.6 was considered in VARIDT 2.0 as High, Medium and Poor, respectively (Figure 7A). When the evaluation result of any criterion was classified to ‘Poor’, the corresponding modelled structure would be considered as unreliable. As a result, the evaluation outcome based on both criteria for each computationally modelled DT structure were fully described in VARIDT 2.0 to indicate the level of precision and reliability of the generated structural variability data.

Figure 7.

Multiple criteria for model assessment. (A) criteria used for assessing generated models: Ramachandran plot (Ca, left) and QMEANBrane score (Cb, right). The percentages of modelled residues in the favored regions of Ramachandran plot (≥98%, 95–98% and <95%) were adopted in VARIDT as Excellent (green), Medium (yellow), and Poor (red), respectively. QMEANBrane scores of ≥0.8, 0.6–0.8 and <0.6 were considered in VARIDT as High (green), Medium (yellow) and Poor (red), respectively. (B) superposition between hSERT structure (grey) and the structures modelled based on Drosophila dopamine transporter (dDAT; blue), and Aquifex aeolicus leucine transporter (LeuT; red). The percentage of the residues of modelled blue structure within favored Ramachandran regions was much higher than that of the red one (96.3% and 90.8% for blue and red structures, respectively), and QMEANBrane score of the blue structure was also much higher than that of the red one (0.82 and 0.75 for blue and red structures, respectively). S1 residues were shown in lines.

Taking serotonin transporter as an example, three experimentally resolved structures determined by X-ray crystallography or electron microscopy were collected in the first place, which included human serotonin transporter (hSERT; PDB ID: 6VRH), Drosophila dopamine transporter (dDAT; PDB ID: 4XPT), Aquifex aeolicus leucine transporter (LeuT; PDB ID: 3F3A). Second, the crystal structures of dDAT and LeuT were adopted to model two hSERT structures, and the experimental structure of hSERT (6VRH) was utilized to assess the quality of the two modelled outcomes. As provided in Figure 7B, the structure in grey represented the experimentally resolved structure of hSERT (6VRH), and the structures in blue and red indicated those structures modelled using the templates of dDAT (4XPT) and LeuT (3F3A), respectively. Compared with the red structure, the blue one presented the much better consistency with the experimental structure in grey, since its RMSD of the backbone atoms of all helix residues (1.19 Å, blue structure) was much lower than that (3.51 Å) of the red one, and its RMSD of the backbone atoms of all S1 pocket residues (0.84 Å, blue structure) was much lower than that (1.81 Å) of the red. Particularly, only two residues (F341 & V501) in the S1 pocket of blue structure showed a substantial shift from their original positions in the experimentally resolved structure, while six resides (I172, F341, Y95, S336, I168 & F334) in red structure presented major shift from their original positions in the experimentally resolved structure. These variations above in the qualities of modelled structures were consistent with the evaluation outcomes, as the percentage of the residues of modelled blue structure within favored Ramachandran regions was much higher than that of the red one (96.3% and 90.8% for blue and red structures in Figure 7B, respectively), and the QMEANBrane score of the blue structure was also much higher than that of the red one (0.82 and 0.75 for blue and red structures in Figure 7B, respectively). All in all, this example provided clear illustration on the effectiveness of those two independent criteria in evaluating the computationally modelled structural variability data.

Statistics, data standardization, access and retrieval

As described in Table 1, four types of DTs’ structural variabilities were collected, which included both the experimentally resolved and the computationally modelled structures. (a) for mutation-induced spatial variations in the folded state of DT, a total of 145 structures originating from 42 DTs were collected, (b) for differences among the DT structures of human and model organisms, a total of 1622 inter-species structures covering 292 DTs of 17 species (human, rat, mouse, dog, zebrafish, rabbit, bovine, pig, chicken, sheep, fruit fly, frog, orangutan, monkey, hamster, guinea pig and horse) were described, (c) for outward- and inward-facing conformations, a total of 118 conformations belonging to 59 DTs were provided and (d) for xenobiotics-driven alterations in 3D DT complexes, a total of 822 xenobiotics-regulated structures of 506 xenobiotics in complex with 57 DTs were collected and described. As a result, 292 DTs with at least one type of structure variability data were shown in VARIDT, which were closely related to the transportation of 844 drugs (including 570 approved & 274 clinical/preclinical drugs) for the treatment of 271 disease classes (such as coronavirus infections, lung cancer, diabetes mellitus, depression, parkinsonism, hypertension and asthma) as defined by the latest WHO International Classification of Diseases (105). Moreover, as one popular structure prediction software tool using AI technique, AlphaFold (64,65) has been used to predict the structure of protein, and a hyperlink to the structure predicted by AlphaFold was also provided in the corresponding DT page of VARIDT 2.0, which could be used as a valuable reference when describing DTs’ structural variability.

Table 1.

The statistics of all data updated to VARIDT 2.0. DT: drug transporter; TCDB: Transporter Classification Database; ICD11: the latest WHO International Classification of Diseases. TCDB-defined DT families were directly adopted from the third-level of TCDB families

		No. of DTs (no. of structures)	No. of TCDB-defined DT families	No. of drugs transported by DTs (approved/in clinical trial)	No. of ICD11-defined disease classes	Detailed description on DT structural variability provided in VARIDT 2.0
All DTs and their various structure data available in VARIDT 2.0		416 (2498)	54	885 (585/132)	414	Four types of DT structural variability data were updated as illustrated in Figure 1
Structural variabilities in VARIDT 2.0	(a) Mutation-induced spatial variations in folded state of DT	42 (145)	16	655 (456/92)	233	A total of 266 pharmacokinetic variations measured by drug clearance, response, etc.
	(b) Difference among the DT structures of multiple species	292 (1622)	44	841 (569/119)	267	A total of 17 DT’s species origins such as human, rat, mouse, dog, zebrafish, etc.
	(c) Outward-facing and inward-facing DT conformations	59 (118)	6	653 (450/98)	235	A total of 118 paired outward- and inward-facing conformations of 59 human DTs
	(d) Xenobiotics-driven alterations in 3D complex of DT	57 (822)	15	709 (495/104)	239	A total of 409 regulating activity data from 214 xenobiotics against 18 human DTs

To make the access and analysis of VARIDT data convenient for all users, the collected raw data were carefully cleaned up and then systematically standardized. These standardizations included: (i) all DTs were standardized by and crosslinked to several available databases, and the extended data of each DT could be identified by the diverse hyperlinks to UniProt (13), ClinicalTrials.gov (106), TTD (58), PDB (14), INTEDE (107), PubChem (108), TCDB (15), ICD-11 (105), ChEBI (109), NCBI Gene (66), dbSNP (110), Drugs@FDA (111) and CAS Registry Number (112). (ii) The web page of each DT was re-organized by categorizing all drugs and endogenous metabolites into single super-class entitled ‘Molecular Transporting Profiles of This DT’, and all transported molecules were grouped according to their clinical statuses (such as: approved, clinical trial and preclinical). (iii) The web page that described the epigenetic regulations data for each DT was also re-organized by categorizing the epigenetic regulations according to their corresponding diseases. All structure variability data can be viewed, assessed, and downloaded from VARIDT 2.0, which is freely assessable without login requirement by all users at: https://idrblab.org/varidt/.