Cryo-EM structure of Slco2a1
To understand the structural basis for PG recognition, we determined the cryogenic electron microscopy (cryo-EM) structure of rat Slco2a1 bound to PGE2 (dinoprostone) at 3.2 Å (Fig. 1, Supplementary Fig. 2 and Supplementary Table 1) and PGF2α (dinoprost) at 3.4 Å (Supplementary Figs. 3-4a and Supplementary Table 1). Prostaglandin E2 has a ketone group at the C9 position of the cyclopentane ring, which characterises this molecule as a prostaglandin of the ‘E’ series. Prostaglandin F2α, in contrast, has a hydroxyl group at the C9 position of the cyclopentane ring, in addition to a hydroxyl at the C11 position. The designation ‘F’ and suffix ‘α’ indicate that the C9 hydroxyl group is in the alpha configuration, meaning it is on the same side of the ring as the C11 hydroxyl group.
a The Cryo-EM structure of Slco2a1 showing the location of the bound PGE2 ligand, along with the Kazal-domain, C-loop, and intracellular helix (ICH) domains. The major and minor sites within the substrate binding cavity and the lateral openings into the membrane bilayer are indicated. b A zoomed-in view of the binding site reveals the main interactions formed with PGE2. Key residues that interact with or are near the substrate are displayed as sticks. Cryo-EM volume density for PGE2 is shown (orange and threshold 0.2). c Schematic of PGE2 binding interactions. d Cell-based transport assays for wild type (WT) and mutant human SLCO2A1. n = 14 independent experiments for the WT, n = 11 for the mock-transfected control and n = 4 (R561A); n = 3 (R561K; R561L; R561Q; W565F; W565L; F557Y); n = 4 (W565A; F557A; M380A; S339A); n = 8 S346A; data shown are the mean and error bars are s.d. e A zoomed-in view of the PGE2 binding site, rotated 180° relative to panel b, with the surface coloured by hydrophobicity. The prominent hydrophobic band is indicated along with key interacting residues.
Rat Slco2a1 shares 83 % sequence identity (89 % similarity) with the human ortholog8. Slco2a1 adopts an outward-open conformation, with the canonical two six-transmembrane (TM) N- and C-terminal bundles of the MFS fold tightly packed on the intracellular side of the membrane. The two 6-TM bundles fan apart towards the extracellular side of the membrane, enclosing a large polar cavity that extends towards the cytoplasm. Consistent with previous structures of SLCO family members38,39,40,41,42, Slco2a1 contains a large extracellular domain (ECD) inserted between TM 9 and TM 10, resembling a Kazal-like domain consisting of an alpha-beta fold with three conserved disulfide bonds. The ECD features several conserved N-glycosylation sites, which are essential for correctly localizing SLCO2A1 to the plasma membrane43. Additional extracellular loops are situated on the N-terminal bundle, between TM 3 and TM 4, and between TM 5 and TM 6, which remain unresolved in our maps. On the C-terminal bundle, we observe another well-resolved extracellular loop between TM 11 and TM 12, which we term the C-loop, that forms an unstructured hairpin that interacts with the membrane-facing side of the Kazal-like ECD. Both structural features are conserved within the broader SLCO/OATP superfamily; however, their roles remain unclear. Finally, the intracellular side of the transporter is characterized by an extension of TM 12, which protrudes beyond the membrane plane and into the cytoplasm, along with an extension of TM 7 that wraps around the intracellular side of the transporter and is called the intracellular helix (ICH)39,44.
Recognition of prostaglandin E2 and F2α
PGE2 and PGF2α are located at the base of the central cavity and adopt near identical binding poses, consistent with their comparable affinities for rat Slco2a1 (Ki values of ~25–43 nM45,46,47). The binding pose of PGE2 is also observed in the recent structure of human SLCO2A141. The E-ring in PGE2 and PGF2α is positioned within ~4 Å of a conserved cluster of residues on TM 11, which consists of Arg561, Trp565 and Phe557 (Fig. 1b, c & Supplementary Fig. 4a). We termed this part of the transporter the major binding site, given that this is where the majority of the PG molecule sits within the transporter. Alanine substitutions of these residues cannot transport PGE2, indicating their importance for function in both rat and human SLCO2A1 (Fig. 1d & Supplementary Fig. 4b). Arginine 561 is strictly conserved within the wider SLCO family of organic anion transporters48,49,50, while Trp565 and Phe557 are only found in members that recognize PG substrates (SLCO2B1 & SLCO3A1) (Supplementary Fig. 1), indicating these residues may be necessary for PG selectivity. Substitution of Arg561 with alanine abolished transport. Replacement with either lysine, glutamine or leucine reduced activity in human SLCO2A1 to <20 % of WT levels, demonstrating the critical importance of the positively charged guanidino group for PG induced transport. Substitution of Trp565 with alanine also abolished transport, whereas a conservative substitution with phenylalanine retained 40 % of WT transport levels. Replacement with leucine, which is also hydrophobic, abolished transport to <20 % of WT levels. Mutation of Trp565 to glycine has also been observed in patients with pachydermoperiostosis (PDP)51, demonstrating the importance of this tryptophan to the physiological function of SLCO2A1. Similarly, substitution of Phe557 with alanine abolished activity, whereas substitution with tyrosine retained 60% WT levels. Combined, these results demonstrate the importance of Phe557, Arg561 and Trp565 to SLCO2A1 function and PG transport41,42.
The carboxylate groups of the α-chain of PGE2 and PGF2α, which impart a net negative charge to these molecules at physiological pH, curl beneath their respective E-rings and face the positively charged region created by the Arg561, Phe557, and Trp565 cluster. The position of the carboxylate group in PGE2 and PGF2α may also explain why the aromatic groups of Trp565 and Phe557 are important for PG recognition. The proximity of these two aromatic side chains to Arg561 amplifies the positive charge of this side chain within the binding site, which would facilitate the orientation of the E-ring and carboxylate group observed in the cryo-EM structures. As discussed below, Arg561 is also likely to play an essential role in the general alternating-access mechanism within the SLCO superfamily, as well as in its interaction with Glu78 on TM5.
The positioning of the α-chain beneath the E-ring is also aided by the proximity of Met379 on TM8 ( ~ 3.6 Å), which pushes the C1-carboxylate group under the E-ring in both substrates. In PGE2, the C1-carboxylate group forms hydrogen bonds with the NE1 nitrogen on Trp565 on TM11 and Ser338 on TM7, and it is located close (4 Å) to His533 on TM10. Alanine substitution of Met379 (Met380 in human SLCO2A1) also reduces PGE2 transport to ~30 % WT levels, demonstrating the importance of this side chain in PG recognition. However, replacement of Ser338 (Ser339 in human SLCO2A1) had little impact on SLCO2A1 activity (Fig. 1d and Supplementary Fig. 4b), suggesting this side chain does not play an important role in PG recognition. In contrast, mutation of His533 to glutamate abolishes the transport activity of human SLCO2A142.
In PGF2α the interaction profile with the transporter is very similar, except Arg561 now projects into the binding site and sits ~4 Å from the C1-carboxylate (Supplementary Fig. 4a, c), indicating this side chain plays an active role in ligand recognition. Indeed, during 1 µs molecular dynamics (MD) simulations of this structure, which are discussed further below, Arg561, along with Trp565 and Ser338, directly interact with the C1-carboxylate group (Supplementary Fig. 4d-e). In contrast, the ω-chain on PGE2 and PGF2α extends away from their respective E-rings and projects into a hydrophobic cavity formed by Val216 and Pro219 on TM5, Ile341, Leu344 and Ser345 on TM7, Leu372 and Ala375 on TM8, as well as Ala526 and Cys530 on TM 10. We labelled this the minor site. The hydroxyl group on the ω-chain in PGE2 is situated within hydrogen bonding distance to Ser345 (3.4 Å) and the E-ring hydroxyl at C11 (2.8 Å), which may further stabilize the lipid in this orientation (Fig. 1b, c). Substitution of Ser345 (Ser346 in human SLCO2A1) with alanine reduced PGE2 transport to 80% WT levels (Fig. 1d), indicating a subtle, but measurable, effect of PG recognition.
A notable feature of the binding poses for both PG molecules is their location relative to a prominent hydrophobic band in the binding pocket (Fig. 1e & Supplementary Fig. 4c). The E-rings are oriented towards the positive charge on Arg561 and are positioned adjacent to the conserved side chain Trp565, which forms the beginning of this hydrophobic band. The α- and ω-chains extend above and below the plane of this hydrophobic band, which also lies at the centre of the transporter. As discussed below, the position of this band serves as the pivot point in the structural rearrangement of the transporter between outward- and inward-facing states, thereby placing PG molecules in an optimal position for alternating-access transport across the membrane.
Analysis of prostaglandin recognition using molecular dynamics
To gain further insight into the mechanism of PG recognition, we undertook unbiased MD simulations of rat Slco2a1 in complex with PGE2 and PGF2α using the cryo-EM structures as starting poses. We conducted three 1 µs simulations to assess the stability of PGE2 and PGF2α in the binding site (Fig. 2a & Supplementary Fig. 4f-g). Clustering analysis of the MD trajectories demonstrates that PGE2 and PGF2α remain reasonably stable in their respective binding poses, with ligand root mean square deviation (r.m.s.d) values averaged over all trajectories of 3.26 Å ± 0.96 for PGE2 and 4.23 Å ± 1.10 for PGF2α, respectively. The ligand r.m.s.d values compare favourably with those calculated from the protein backbone, which were in the range of 2.80–6.97 Å, indicating the ligand flexibility is consistent with the overall flexibility of the protein. The modelled positions of the ligands are further supported by the low number of clusters (12 for PGE2 and 17 for PGF2α) and by the fact that the three largest clusters accounted for 95% and 77% of the simulation frames, respectively. A striking result from the cluster analysis is the stability of both the C1-carboxylate and the E-ring in the simulations (Fig. 2b), suggesting that these are important features determining ligand recognition and selection within SLCO2A1.
a A 2D plot showing backbone-aligned ligand r.m.s.d values of the PGE2 position in three replicates of the 1μs simulation. b A zoomed-in view of the binding site shows the three most representative binding structures obtained by clustering the trajectory pooled across three microsecond-long unbiased molecular dynamics simulation replicates. The starting pose was the cryo-EM structure of Slco2a1 bound to PGE2 (inset left). The key interacting residues within <6 Å of the bound PG molecule are highlighted. c An interaction map of the residues within 10 Å of the PGE2 molecule from the three replicates of the simulation. Only interactions present in more than 50 % of the simulation frames are shown. d A 2D plot showing the frequency of different types of interactions observed for residues that interacted with the ligand in more than 50 % of the simulation frames. Results were calculated from the trajectory pooled across all three replicates.
We next analyzed which residues interacted most frequently with PGE2 and PGF2α. Residues that interacted > 50 % of the time with PGE2 are located on TM 5, TM 7 and TM 8 and cluster around the ω-chain (Fig. 2c). In the PGF2α simulations, the majority of interactions are made to the α-chain, with far fewer to the ω-chain (Supplementary Fig. 4e). This may explain the higher r.m.s.d. values for this PGF2α compared to PGE2 and why the ω-chain may play a less important role in ligand selection. Tryptophan 565 is unique to SLCO family members that are dedicated PG transporters and interacts with the C1-carboxylate on the α-chain in both PGE2 and PGF2α. In the PGF2α bound structure, Arg561 and Ser338 also interact with the C1-carboxylate group (Supplementary Fig. 4f). Further analysis of MD simulations reveals a dominant role for van der Waals contacts and hydrophobic interactions along with hydrogen bonds to Ser345, Ser338, His533, Arg561 and Trp565 (Supplementary Fig. 4e-f) in binding the PG molecules.
Overall, the MD simulations align with our mutagenesis data (Fig. 1d, Supplementary Fig. 4b) and highlight the role of Trp565 and hydrophobic interactions in orienting the PG ligands.
Structural basis for drug recognition by SLCO2A1
SLCO2A1 is noted for its interactions with several commonly prescribed drug classes, such as NSAIDs, antihypertensives, anti-PD medications, and asthma treatments33. As direct uptake is often not measured, whether these drugs function as inhibitors or are transported remains unclear for many members of the SLCO superfamily. However, using mass spectrometry, we measured the ability of human SLCO2A1 to transport representative drugs from these four drug families. Our results show that Zafirlukast, the anti-leukotriene receptor antagonist, and Losartan, an angiotensin receptor blocker used to treat hypertension, are transported (Fig. 3a). In contrast, Fentiazac, an NSAID, and Tolcapone, used to treat PD, showed no detectable uptake. The calculated IC50 values for Zafirlukast, Losartan, Fentiazac and Tolcapone are 1.12, 2.12, 45.2 and 24.1 μM, respectively (Fig. 3b). The values for Fentiazac and Tolcapone are considerably higher compared to Zafirlukast and Losartan, suggesting these drugs might interact differently with SLCO2A1, which may explain why these drugs are inhibitors rather than substrates.
a Cell-based transport assays of human SLCO2A1 with four clinically prescribed drug molecules. n = 3 independent experiments (Fentiazac and Tolcapone) and n = 9 independent experiments (Zafirlukast and Losartan); data shown are the mean and error bars are s.d. b IC50 data for the four drug molecules; errors are s.d. c A zoomed-in view of the binding site shows the position of Zafirlukast. Cryo-EM volume density is shown (orange and threshold 0.2). Inset—Structural overlay with the bound PGE2 molecule (PDB: 9R0M). d A zoomed-in view of the binding site shows the position of Losartan. Cryo-EM volume density is shown (orange and threshold 0.2). Inset—Structural overlay with the bound PGE2 molecule (PDB:9R0M). e An interaction map of the residues within 10 Å of Zafirlukast from the three replicates of the simulation. f An interaction map of the residues within 10 Å of Losartan from the three replicates of the simulation. g Cell-based transport assays of human SLCO2A1 with Losartan (n = 13 independent experiments (mock); n = 13 (WT); n = 7 (R561A); n = 13 (W565A) and Zafirlukast (n = 12 (mock); n = 11 (WT); n = 6 (R561A); n = 12 (W565A); data shown are the mean and error bars are s.d.
To further understand how these drugs interact with the transporter, we determined the cryo-EM structure of rat Slco2a1 bound to Zafirlukast at 3.2 Å and Losartan at 3.1 Å (Table 1 & Supplementary Figs. 5-6). The drug-bound structures adopt similar outward-open conformations as the apo structure. The r.m.s.d. between Slco2a1-apo and each of the ligand-bound states is 0.38 Å (apo vs. Zafirlukast) and 0.182 Å (apo vs. Losartan). Zafirlukast adopts an L-shaped configuration in the binding pocket, occupying a position similar to that of the endogenous PG ligands (Fig. 3c). The most striking observation is the lack of polar interactions between the drug and the binding site, consistent with our observations for the two PG molecules. The sulfonyl group, which is located at one end of the drug molecule, is positioned equidistant between the E-ring oxygens and C-carboxylate of the PG molecules discussed above, and it sits within 4 Å of Arg561, with the 2-methylphenyl group, which is stabilized by Phe212 on TM 5. In the modelled position, the drug does not interact with Trp565; however, a 90° rotation of the carbamoyl moiety would bring the 2-methylphenyl group within ~ 3.0 Å of Trp565, which might easily occur following a conformational change in the transporter or the drug, and was observed in our MD simulations (Supplementary Fig. 7c). The methoxy group sits near the C15-hydroxyl group observed in PGE2, whilst the remainder of the drug extends along the hydrophobic surface of the binding site. Specifically, the drug packs against Val216 and Pro219 on TM5, Ile341 and Leu344 on TM 7 and Ile367, Asn371, Leu372 and Ala375 on TM 8, while the cyclopentyl ester, which is located at one end of the drug, packs against Tyr223 at the cytoplasmic end of the binding pocket. The phenyl and 3-methyl indole groups, which are present in the middle of the drug molecule, extend behind TM 5 and into the minor site, which is occupied by the α-chain of the PG molecules. Somewhat counterintuitively, therefore, the much larger molecule of Zafirlukast can mimic the key interactions observed for the endogenous PG ligands.
In contrast, Losartan adopts a C-shaped configuration in the binding site (Fig. 3d), with the biphenyl group oriented vertically relative to the membrane, positioning the tetrazole group near Trp565 (~ 4 Å) and occupying a position similar to the E-ring of the PG molecules. The remainder of the biphenyl group packs against Val216 and Pro219 in the N-terminal bundle, as well as the hydrophobic band near Ile341 in the C-terminal bundle. The 4-chloro-imidazol-5-methanol group occupies a position comparable to the C1-carboxylate observed in the PG molecules and sits near Phe212, Met379, His533, and Met560. The 2-butyl group extends into the C-terminal bundle, occupying a newly formed cryptic hydrophobic cavity created following a rotamer rearrangement in Phe337 in TM 7. Several similarities can be observed in how Losartan and Zafirlukast interact with Slco2a1 when compared to PGE2 and PGF2α (Fig. 3c, f). The most notable aspects include interactions with Trp565, the presence of a negative charge near the C1-carboxylate binding position, adjacent to Arg561 and engagement with the prominent hydrophobic band. Indeed, mutation of Arg561 or Trp565 to alanine reduced the ability of SLCO2A1 to transport both drug molecules (Fig. 3g). The mutations had a greater effect on Zafirlukast uptake, possibly due to the more extensive interactions observed between this drug and the transporter.
To gain further insight into how these drugs interact with Slco2a1, we conducted three 1 µs simulations using the same methodology applied to the PG molecules discussed above (Supplementary Fig. 7). Clustering analysis of the MD trajectories reveals that both Zafirlukast and Losartan are relatively stable within the binding site, with ligand r.m.s.d values averaged over all trajectories of 4.30 Å ± 0.45 for Zafirlukast and 5.0 Å ± 0.99 for Losartan. However, these values compare favourably with the r.m.s.d values for PGE2 and PGF2α. Within the simulation frames, the top three clusters accounted for 99.2 % of the positions sampled by Zafirlukast, whereas for Losartan, this figure dropped to 75.9 %, aligning with the values obtained for PGF2α. Our analysis, therefore, indicates that Zafirlukast has fewer degrees of freedom within the binding site than Losartan, consistent with the higher-quality volume observed for this ligand.
Finally, we analyzed the interactions made by the drugs within the binding site (Fig. 3e, f). There is a notable difference between the two drugs: Zafirlukast makes extensive interactions with the binding site, whereas Losartan forms very few interactions with nearby aliphatic side chains. The most notable hydrogen bond interactions with Zafirlukast involved Gln42 and Ser345, whereas in Losartan, we observed hydrogen bonds only with Trp565 in 10% of the frames. Overall, the binding modes are dominated by hydrophobic, van der Waals and aromatic interactions, similar to our observations for the PG molecules.
Structural basis for inhibitor recognition
Our results demonstrate that while Zafirlukast and Losartan are transported by human SLCO2A1, Fentiazac and Tolcapone are not under our assay conditions. Therefore, Fentiazac and Tolcapone are likely to act as inhibitors of PG transport (Fig. 3a). To gain insight into why these drugs are not transported, we determined the cryo-EM structure of rat Slco2a1 bound to either Fentiazac or Tolcapone at 3.1 Å (Supplementary Table 1 & Supplementary Figs. 8–9). As with the two structures bound to Zafirlukast and Losartan, the structures of Slco2a1 bound to Fentiazac and Tolcapone also adopt outward-open conformations. The r.m.s.d. between Slco2a1-PGE2 was 0.37 Å (PGE2 vs. Fentiazac) and 0.51 Å (PGE2 vs. Tolcapone). Fentiazac binds in the same pocket as the PG molecules, adopting a similar vertical orientation to Losartan (Fig. 4a). The pseudo-symmetrical structure of the compound made it challenging to model its orientation, and we leave open the possibility that it might also engage in the opposite orientation. However, using our prior observations from the PG-bound structures, we modelled the drug with the chloride atom occupying a position similar to that of the carboxylates, which we termed the negative cluster.
a A close-up view of the binding site reveals the position of Fentiazac. Cryo-EM volume density is depicted (orange and threshold 0.7). Inset–A structural overlay with the bound PGE2 molecule. b An interaction map of the residues within 10 Å of Fentiazac from the three replicates of the MD simulation. c 2D plot showing backbone-aligned ligand r.m.s.d values of the Fentiazac position in three replicates of the 1μs MD simulation. d A close-up view of the binding site reveals the position of Tolcapone. Cryo-EM volume density is depicted (orange and threshold 0.7). Inset–A structural overlay with the bound PGE2 molecule. e 2D plot showing backbone-aligned ligand r.m.s.d values of the Tolcapone position in three replicates of the 1μs MD simulation. f An interaction map of the residues within 10 Å of Tolcapone from the three replicates of the MD simulation. Only interactions present in more than 50 % of the simulation frames are shown.
Nevertheless, in either orientation, the carboxylate group in Fentiazac occupies the same position as the E-ring in PGE2 and PGF2α. Although we don’t observe a direct interaction between the carboxylate and Arg561 in the binding site, the cryo-EM volume around Arg561 extends towards the drug molecule (Supplementary Fig. 11a). When we analyzed the stability of the drug in the binding site using MD, we observed that Arg561 moves to form a salt bridge interaction with the carboxylate (Fig. 4b & Supplementary Fig. 11b-c) and interacts with the drug with a frequency > 60 %. The remaining dominant interactions with the drug are van der Waals and hydrophobic, with a minor Pi-stacking interaction with Phe212, Trp565, and Phe557, which are similar to those observed in the PG-bound structures.
However, the r.m.s.d. values averaged over all trajectories are 6.29 ± 1.05 (Fig. 4c), which is notably higher than for either PG-molecule or the transported drugs, indicating the drug is not held as stably in the binding site. The phenyl-thiazole group extends outward toward the cytoplasmic side of the binding pocket and packs against Pro219, Val216, Ser345, and Leu372, as observed in the structures of Zafirlukast and Losartan. In this orientation, Fentiazac sits almost 90° relative to the position of the ω-chain in the PG structures and does not engage the minor pocket. Although the chlorophenyl group occupies the same position as the α-chain in PG-bound structures, the chloride atom does not occupy the same position as the C1-carboxylate. Instead, the chloride atom is located closer ( ~ 3.2 Å) to Met379 on TM 8 and ( ~ 4.0 Å) to His533 on TM10.
We also observed strong density in the binding site adjacent to Fentiazac (Supplementary Fig. 11d), which allowed us to model a Lauryl Maltoside Neopentyl Glycol (LMNG) detergent molecule into the structure. The LMNG molecule enters the binding site through the lateral opening noted in the PG-bound structures. One of the two maltopyranoside groups wedges between TM 5 and TM 8, whereas one of the didecylpropane chains extends down into the binding site, adjacent to the phenyl-thiazole group of Fentiazac and close to Leu348 and Ile367. The detergent density in Fentiazac partially overlaps with that of cholesterol hemisuccinate (CHS), which is observed in our PGE2 and apo structures. As discussed below, the location of these lipophilic molecules within the lateral opening in the transporter suggests that ligands gain access to SLCO2A1 via the membrane.
In the Tolcapone bound structure, we again suffered from limited resolution and pseudo-symmetry in the molecule. Using the same rationale as we applied to the Fentiazac compound, we modelled the nitro group near the α-chain of the PG substrates, and the carbonyl group occupies the same position as the C15-hydroxyl group on the ω-chain of PGE2 (Fig. 4d). In this orientation, tolcapone does not interact with Trp565 nor position a negatively charged group near the E-ring or C1-carboxylate binding sites in the transporter. We also observe that Arg561 adopts two rotamer positions in the cryo-EM volume (Supplementary Fig. 11e), indicating that the drug fails to engage Arg561 and that Arg561 is capable of adopting two rotamer positions within the transporter. Tolcapone also fails to interact with the hydrophobic band in this orientation. Instead, it binds near Pro219, Val216, Ala375, Met379, Ile341, and Ser345, similar to Fentiazac. However, Tolcapone extends several polar hydroxyl groups into the minor site, which may account for its orientation in the binding site. In the associated MD analysis of this binding pose, we observed considerable movement of the drug in the binding site (Fig. 4e & Supplementary Fig. 11f-g). The r.m.s.d values averaged over all trajectories are 6.73 ± 1.26, which is similar to that of Fentiazac and indicates this drug is also unable to adopt a stable binding pose. Or at least, a pose that is as stable as those observed for the PG molecules. The main interactions between the drug and the transporter are dominated by hydrophobic interactions made to Pro219, Val216, Ile341 and Leu372 (Fig. 4f & Supplementary Fig.11e). However, we did observe some pi-stacking interactions between the drug and Trp565 (Supplementary Fig. 11g), which may explain why this drug has some affinity for the binding site. Further free-energy calculations will be needed to follow up on these preliminary simulations.
In summary, the two non-transported drug molecules exhibit characteristics of substrate mimicry. Both Fentiazac and Tolcapone contain hydrophobic groups that interact with the same regions of the PG binding site that accommodate the aliphatic side chains, while their negatively charged polar groups fit into pockets that recognise the E-ring and the C1-carboxylate groups. However, unlike PGE2, PGF2α, Zafirlukast, and Losartan, neither Fentiazac nor Tolcapone appears capable of adopting a stable binding pose in the transporter and effectively engaging Arg561, which, as discussed below, we propose is essential for triggering transport across the membrane.
Alternating access transport in SLCO2A1
In the MFS, alternating access transport is generally understood to occur through the concerted movement of four pairs of helices comprising the extracellular and intracellular gates, respectively44,52. In the outward-open state, as observed for rat Slco2a1 (Fig. 1a), the intracellular gate is formed by TM 4 and TM 5 packing against TM 10 and TM 11. The extracellular gate, formed by TM 1 and TM 2 and TM 7 and TM 8, is splayed apart in the open state. During the transition from outward-open to inward-open states, the helices of the extracellular gate must close to seal the binding site from the extracellular space. The closure of the extracellular gate, coupled with ligand binding in the central pocket, subsequently triggers the opening of the intracellular gate and the release of the ligand into the cytoplasm through a symporter mechanism53.
To understand how this mechanism operates in human SLCO2A1, we compared our cryo-EM structure of PGE2-bound rat Slco2A1 (PDB:9QZO) with the AlphaFold-predicted structure (Uniprot: Q00910)54, which adopts an inward-facing state. The structural overlay reveals that the primary movement during transition from outward to inward-facing states occurs in the N-terminal bundle rather than the symmetrical movement of the N- and C-terminal bundles observed in some members of the MFS28. In our model, the N-terminal bundle rocks against the PGE2 molecule, which is positioned horizontally via the interactions with Arg561, Trp565 and the hydrophobic band in the binding pocket (Fig. 5a). In transitioning from the outward-open to inward-open state, TM 1 and TM 2 rotate ~40° towards TM 7 and TM 8 and are stabilized in the closed state by two salt bridges between Glu60 and Lys61 on TM 1 with Lys350 (Lys351 in human SLCO2A1) and Glu353 (Glu354 in human SLCO2A1) on TM7 (Fig. 5b). While Lys61Ala and Glu354Ala variants had little impact on transport, substituting Glu60 or Lys351 with alanine abolished transport, indicating the importance of the Glu60-Lys351 salt bridge in the transport mechanism (Fig. 5c). On the cytoplasmic side of the transporter, the intracellular gate helices, TMs 4 and 5, move away from TMs 10 and 11 along an axis originating from Pro219, executing a ~ 28° rotation and breaking the salt bridge between Asp196 and Asp197 on TM4 and Arg542 and Lys549 on TMs 10 and 11, respectively. Substituting Asp196 or Asp197 with alanine abolished transport, while Arg542Ala retained approximately 50% activity (Fig. 5c). A similar set of salt bridge interactions was noted in the SLCO1B1 structure39, suggesting these are conserved features of the SLCO superfamily.
a Structural overlay of the PGE2-bound Slco2a1 with the inward-facing AlphaFold2 model (AF-Q00910-F1). The main structural differences are evident in the N-terminal bundle, which rotates around the bound PGE2 substrate, represented by spheres. The axis of rotation traverses the hydrophobic band formed and runs through Trp565 and Arg561, resulting in TMs 1, 2, and 5 packing against TMs 7, 8, and 11 in the inward-facing state. b Conserved pairs of salt bridge interactions coordinate the arrangement of the extracellular and intracellular gate helices, located above and below the central PG binding site. c Cell-based transport assays for wild type (WT) and mutant human SLCO2A1; n = 12 independent experiments for the WT and the mock-transfected control; n = 4 for the mutants; data shown are the mean and error bars are s.d. d A zoomed-in view of the binding site illustrating the structural overlay between the Apo, PG-bound, and AF models of Slco2a1. The direction of movement observed in Arg561 and TM5 and TM2 are indicated.
An important question for all transport mechanisms is how ligand binding is coupled to the conformational changes that lead to substrate transport. To understand the structural implications for substrate binding, we determined the ligand-free structure of rat Slco2a1 to 2.9 Å (Supplementary Table 1 and Supplementary Fig. 10). As we observed in the previous ligand-bound structures, Slco2a1 adopts the same outward-facing state, with an r.m.s.d of 0.32 Å vs. the PGE2-bound structure. An overlay of the structures obtained in the presence of either PGE2 or PGF2α, compared to the Apo structure, illustrates that, within the binding pocket, the only significant movement following PG binding occurs in the rotamer position of Arg561 (Fig. 5d). In the Apo structure, Arg561 on TM11 forms a salt bridge interaction with Glu78 on TM2. In the PGE2-bound state, Arg561 and Glu78 change their rotamers but remain close enough to maintain the salt bridge. However, in the PGF2α structure, Arg561 points down towards the C1-carboxylate of the substrate, while Glu78 points away towards the extracellular gate and interacts with Lys53 on TM1, breaking their interaction. The density around Glu78, however, is weak, suggesting that Arg561 could move away without Glu78 needing to adopt the alternate rotamer. Additionally, the dual rotamer position for Arg561 in the Tolcapone-bound structure confirms the ability of Arg561 to change rotamer positions in response to ligand binding (Supplementary Fig. 11e). For Slco2a1 to transition to the inward-facing state, the 40° rotation on TMs 1 and 2 brings Glu78 into the position formally occupied by Arg561 in the outward open Apo state. At the same time, Arg561 moves further down into the pocket, which may help pull the PG towards the cytoplasmic exit tunnel, which opens as TMs 4 and 5 move away from TMs 10 and 11.
Arginine 561 and Glu78 are strictly conserved within the SLCO superfamily and are essential for PG transport9,49,55, with even a conservative substitution of Glu78 to aspartate reducing function by > 60 % (Fig. 5c). The equivalent arginine in SLCO1B1 and SLCO1B3, Arg580 (Supplementary Fig. 1), is also essential for function56,57, similar to our observation in SLCO2A1 (Fig. 1d). Our model therefore suggests that ligand recognition and transport may be separate events within the OATP family. Each member of the OATP family has specific side chains that recognise their respective ligands; in the case of SLCO2A1, this would be principally Trp565. Once successfully bound, the ligands would then disrupt the Arg-Glu interaction, triggering transport. This mechanism is an elegant way for the SLCO superfamily to evolve specific ligand recognition mechanisms whilst retaining a conserved mechanism to trigger alternating access transport. The proposed alternating access mechanism may also explain why Fentiazac and Tolcapone are not substrates, as these drugs are unable to adopt a stable binding pose within the transporter (Fig. 4c–e), restricting their ability to disrupt or otherwise modulate the salt bridge between Glu78 and Arg561.
Lipid densities suggest ligand entry via the membrane
Prostaglandins have high logP values and are highly soluble in lipid environments, prompting us to question how these molecules access the binding site from the extracellular space. As noted above, the two six-helix bundles in SLCO2A1 splay apart sufficiently to open the binding site to the lipid bilayer via two lateral openings (Fig. 1a). In our cryo-EM maps for the apo and ligand-bound structures, we observe a concentration of lipid-like densities clustering within this opening, specifically between TM5 from the N-terminal bundle and TM8 from the C-terminal bundle. The position of these lipids is in a very similar position to the density for LMNG observed in the Fentiazac complex (Supplementary Fig. 11d), although they do not fully enter the binding site. In the PG-bound and apo structures, we modelled this density as CHS due to the planar profile (Fig. 6a). Similar lipid-like densities were observed in the organic cation transporter OCT3 (SLC22A3)58 and SLCO1B139.
a A CHS molecule can be modelled into the cryo-EM volume density observed in the vestibule that opens into the membrane between TMs 5 and 8. Cryo-EM volume density is shown (orange and threshold 0.2). b A zoomed-in view of the lateral opening and the cryo-EM volume density (orange and threshold 0.2). Residues within 5 Å of the bound CHS are shown. Note, Gly368 in rat SLCO2A1 is equivalent to Gly369 in human SLCO2A1. c Cell-based transport assays for wild type (WT) and mutant human SLCO2A1. n = 8 independent experiments for the WT and the mock-transfected control, and n = 4 for the mutants; data shown are the mean and error bars are s.d. d Schematic model of the transport mechanism in Slco2a1, highlighting key structural features and their role in ligand binding and conformational changes in the transporter. (i) opening observed between TM 5 and TM 8 in the outward-open state (ii) transition to the inward-open state via the movement of TM 1 and TM 2, (iii) AF2 model of the inward-open state, (iv) PG exit from the binding pocket into the inner leaflet of the membrane.
The location of the CHS density, which sits above the ω-chain of PGE2 and PGF2α, prompted us to speculate that PG molecules may enter the transporter from the membrane. Two mutations in SLCO2A1, which are located near the CHS molecule on TM5 (Gly222Arg) and TM8 (Gly369Asp), were previously identified in patients diagnosed with PHO (Fig. 6b). These mutations result in a severe reduction of activity (Fig. 6c). The CHS also sits near Tyr223 and Ala220 on TM5. Substituting Tyr223 with Ala also reduced activity by approximately 85 %, while a conservative phenylalanine replacement retained activity, indicating the importance of a hydrophobic group at this position. Similarly, substituting Ala220 with valine reduced activity by approximately 40 %, whereas a glycine replacement retained activity, indicating that maintaining the space between TMs 5 and 8 is also essential. Our results, combined with the location of the PHO disease mutations, suggest that this region of SLCO2A1 is crucial for function and support the hypothesis that some PG and drug molecules may enter the transporter from the membrane.
Prostaglandin and drug transport via SLCO2A1
Taken together, our structural, functional and modelling data suggest a mechanism for PG and drug transport via SLCO2A1 (Fig. 6d). In the outward-open state, ligands access the binding pocket from the outer leaflet of the plasma membrane, likely through the opening observed between TM 5 and TM 8 (i). The presence of several basic side chains in combination with the hydrophobic band favours the recognition of negatively charged, lipophilic molecules, such as PGs, and the chemically similar drug molecules discussed above. The interaction with Trp565, along with the conserved hydrophobic band, positions the PG in a productive binding pose. In this position, the E-ring aligns with the rotation axis necessary to transition to the inward-open state via the movement of TM 1 and TM 2 to close the extracellular gate (ii). The E-ring and C1-carboxylate groups disrupt the salt bridge between Arg561 and Glu78, weakening the interaction between the extracellular gate helices TM 5 and TM 8. The specific role of the Arg561-Glu78 salt bridge in the mechanism warrants further investigation, as the conformational change between outward and inward-facing states also involves the pairs of salt bridges in the extracellular and intracellular gates mentioned above. However, Arg561 and Glu78 are strictly conserved within the SLCO superfamily, suggesting this interaction may form part of a conserved mechanism.
Comparison with the inward-open AF2 model indicates that when the transporter transitions to the inward-open state (iii), TM 11 rotates inwards, pulling Arg561 away from the C1-carboxylate and restoring its interaction with Glu78. Reforming the Arg561-Glu78 bond would eliminate the favourable interaction with the C1-carboxylate and could be described as a substrate hand-off model, with Arg561 serving as the trigger for the conformational change. Finally, in the inward-open state, a similar cavity opens into the inner leaflet of the membrane between TMs 5 and 8, directly adjacent to the C1-carboxylate. The PG could then exit the binding pocket into the inner leaflet of the membrane, and subsequently into the cytosol (iv). An interesting aspect of our mechanism is the potential role of the membrane as a selectivity filter. Only sufficiently lipophilic molecules would gain access to the transporter binding site, wherein only those that can orient correctly and stably to engage Arg561 trigger the transport cycle.