Substrate and drug recognition mechanisms of SLC19A3

Dear Editor,

Vitamins B₁ and B₆ are two water-soluble vitamins. Their active forms thiamine pyrophosphate and pyridoxal 5′-phosphate serve as cofactors for numerous enzymes involved in multiple biochemical reactions that are essential to maintain the composition and energy metabolism of the human body.¹ Consequently, their deficiency leads to a variety of diseases, such as neurological abnormalities and cardiovascular diseases.² Humans cannot synthesize these vitamins de novo and must obtain them from the diet. Two specific transporters, thiamine transporter 1 (ThTr1 or SLC19A2) and thiamine transporter 2 (ThTr2 or SLC19A3), have been identified to be the major transmembrane transporters that are involved in the uptake of both vitamins thus far.^3,4,5,6 Interestingly, the clinical antineoplastic drug fedratinib, a Janus kinase 2 (JAK2) inhibitor used in treating myelofibrosis, induces Wernicke’s-like encephalopathy in some myelofibrosis patients due to thiamine scarcity.^7,8 This is caused by its off-target inhibitory activity against both SLC19A2 and SLC19A3.⁹ Here, we report the cryo-electron microscopy (cryo-EM) structures of human SLC19A3 in complex with vitamins B₁, B₆, or fedratinib. Remarkably, all these compounds bind to the same site on SLC19A3 via a closely related structural element. Mutagenesis studies further revealed the critical residues of SLC19A3 for substrate and drug recognition. In summary, our work provides a structural framework for understanding the substrate diversity of SLC19A3.

To assist in the structure determination, we attached a BRIL tag to the N-terminus of human SLC19A3 and exploited the BRIL/Fab/Nb module, similar to the method reported in our structural study of SLC19A1 (Fig. 1a; Supplementary information, Fig. S1a).¹⁰ We first confirmed that BRIL-tagged SLC19A3 can actively transport radiolabeled thiamine ([³H]-thiamine) into HEK293F cells, exhibiting a slightly lower K_m (0.7 vs 1.1 μM) and V_max (12.0 vs 19.0 pmol/mg/min) compared to the wild-type (WT) protein (Fig. 1b; Supplementary information, Fig. S1b). Then, we determined the cryo-EM structure of SLC19A3 in complex with thiamine at a 3.7 Å resolution by local refinement of only the transporter portion of the BRIL-SLC19A3/Fab/Nb complex (Fig. 1c; Supplementary information, Figs. S1c, S2 and Table S1). In this structure, SLC19A3 adopts a typical major facilitator superfamily fold in an inward-facing conformation, with the two-half transmembrane (TM) domains (TM1–6 and TM7–12) opening to the cytoplasmic side, similar to the reported SLC19A1 structures (Fig. 1c; Supplementary information, Fig. S1d).^10,11,12 The additional EM density inside SLC19A3 fits well with the structure of thiamine, whose binding pose was further validated by molecular dynamics (MD) simulations (Fig. 1c; Supplementary information, Fig. S3a). The surface electrostatic distribution of the binding site also nicely complemented the charge characteristics of thiamine (Supplementary information, Fig. S1e, f). Thiamine extensively interacted with multiple residues of SLC19A3 (Fig. 1d; Supplementary information, Fig. S1g, h). Specifically, Glu32, Glu110, and Asn297 form hydrogen bonds, salt bridges, or favorable electrostatic interactions with the nitrogen atoms in the pyrimidine or thiazolium ring. In addition, Tyr113 packs against the pyrimidine ring through π–π interactions. On the other end, the hydroxyl group of thiamine is stabilized by Tyr151 and Glu320 through hydrogen bonding. Moreover, a few hydrophobic residues, including Leu35, Trp59, and Leu296, complement the hydrophobic skeleton of thiamine.

Fig. 1: Molecular mechanisms of substrate recognition by SLC19A3.

a Schematic diagram of BRIL-SLC19A3. b Concentration-dependent uptake of [³H]-thiamine by SLC19A3 (n = 3, mean ± SEM). c Structure model of the SLC19A3/thiamine complex shown in ribbon representation. The TM numbers are labeled, and the two discrete TM bundles are colored differently. The EM density of thiamine is shown on the side. d Details of the interaction between SLC19A3 and thiamine. e Impact of the binding site mutations on [³H]-thiamine uptake. f Functional verification of the potential proton-sensing residues. g Fluorescence-detection size-exclusion chromatography-based thermostability assay. PL pyridoxal, PN pyridoxine, PLP pyridoxal 5′-phosphate, PM pyridoxamine. h Quantitative determination of the inhibition of SLC19A3 thiamine transport activity by pyridoxamine. The IC₅₀ is calculated by fitting to a nonlinear regression model. i Structure model of SLC19A3/pyridoxamine complex. The EM density of pyridoxamine is shown on the side. j Comparison of the thiamine and pyridoxamine binding sites in SLC19A3. k Details of the interaction between SLC19A3 and pyridoxamine. l Impact of the binding site mutations on [³H]-pyridoxine uptake. m Inhibition of the thiamine transport activity of SLC19A3 and its mutants by fedratinib. The IC₅₀ of fedratinib is indicated. n Structure model of SLC19A3/fedratinib complex. The EM density of fedratinib is shown on the side. o Comparison of the thiamine and fedratinib binding sites in SLC19A3. p Details of the interaction between SLC19A3 and fedratinib. q Comparison of the fedratinib binding sites in SLC19A3 and JAK2 (PDB: 6VNE). The obvious distinct spaces observed in their binding sites are labeled with dashed circles. For all the uptake assays, data are normalized to the activity of WT SLC19A3 and all experiments were performed in triplicate or quadruplicate (n = 3 or 4, mean ± SD). The hydrogen bonds and salt bridges are indicated with dashed lines, and the corresponding distances are labeled in angstroms.

To investigate the functional significance of the residues involved in thiamine binding, we individually mutated them to alanine and examined the [³H]-thiamine transport activity of mutants in HEK293F cells. Our results showed that substitutions of most of the residues that participate in hydrophilic or π–π interactions reduced the transport activity of SLC19A3, especially E32A, E110A, and Y113A (Fig. 1e; Supplementary information, Fig. S4a), aligning with their higher free energy contributions to thiamine binding in our MD simulation analysis (Supplementary information, Fig. S3b). In addition, these mutants weakened the thiamine-induced thermostabilization of SLC19A3 accordingly (Supplementary information, Fig. S5a, e), further supporting that the functional effects were caused by the impaired substrate recognition. In contrast, the hydrophobic residues play trivial roles in substrate recognition, as the corresponding mutations barely affected thiamine binding and transport. Of note, nearly all the key residues are conserved in SLC19A2 (Supplementary information, Fig. S1i), suggesting that SLC19A2 likely recognizes thiamine in the same way as SLC19A3.

As previously reported, lower extracellular pH suppresses thiamine uptake by SLC19A3.^4,5 We then examined whether the three negatively charged residues around the thiamine binding site (Glu32, Glu110, and Glu320) (Fig. 1d) play a role in this process. We reasoned that mutating the key glutamate residue(s) to uncharged glutamine should disrupt the proton sensitivity of SLC19A3. The thiamine uptake assay showed that E110Q mutant was more sensitive to alteration of the extracellular pH than WT SLC19A3, while both E32Q and E320Q mutants were less pH dependent. Moreover, the E32Q/E320Q double mutation almost abolished the thiamine transport and proton sensing activity of SLC19A3 (Fig. 1f; Supplementary information, Fig. S4b). This impact might be partially caused by the reduced substrate binding to SLC19A3 upon the protonation of Glu32 and Glu320 (Supplementary information, Fig. S5b, f). Overall, these results suggest that Glu32 and Glu320 are crucial for the function of SLC19A3 and likely involved in the proton regulation. However, the underlying mechanism requires further studies.

In addition to thiamine (vitamin B₁), SLC19A3 can also transport vitamin B₆.⁵ To reveal the molecular mechanism for the recognition of various substrates, we further sought to determine the structure of SLC19A3 bound to vitamin B₆. Vitamin B₆ usually refers to the six common forms, including pyridoxal, pyridoxine, pyridoxamine, and their phosphorylated derivatives (Supplementary information, Fig. S6a). Our thermostability assay showed that among the four vitamin B₆ derivatives we tested, only pyridoxamine increased the stability of SLC19A3, similar to the stability following binding to the classic substrate thiamine (Fig. 1g; Supplementary information, Fig. S6b). Moreover, pyridoxamine efficiently inhibited the thiamine uptake by SLC19A3 (Fig. 1h). Therefore, we chose pyridoxamine for the following structural study.

We determined the cryo-EM structure of SLC19A3/pyridoxamine complex at a 3.7 Å resolution (Fig. 1i; Supplementary information, Figs. S6c, S7 and Table S1), and settled the optimal pose of pyridoxamine in combination with MD simulations (Supplementary information, Fig. S3c). In this structure, SLC19A3 adopts a similar inward-facing conformation as it does when it binds thiamine (Fig. 1i; Supplementary information, Fig. S1j). Additionally, pyridoxamine occupies a similar pocket to thiamine and interacts with almost the same cluster of residues (Fig. 1j, k). For example, Glu32, Glu110, Tyr113, and Asn297 clamp the pyridine ring via hydrophilic or electrostatic interactions, while Leu35, Phe56, Trp59, Tyr113, and Leu296 create a hydrophobic environment to accommodate the carbon framework of pyridoxamine (Fig. 1k; Supplementary information, Fig. S6d, e). Strikingly, the electrostatic attraction between the amino group of pyridoxamine and the side-chain carboxyl group of Glu32 (Fig. 1k) could well explain the substrate preference of SLC19A3 for different vitamin B₆ variants (Fig. 1g; Supplementary information, Fig. S6a).⁵ Specifically, substitution of the amino group of pyridoxamine with a hydroxyl group in pyridoxine partially hindered its interaction with Glu32, while replacing the aminomethyl moiety with an aldehyde group in pyridoxal would eliminate this electrostatic interaction.

We then mutated the residues that contact pyridoxamine to alanine or structurally similar residues and verified their functional effects. In agreement with our structural observations and MD simulations (Supplementary information, Fig. S3d), substitution of most of the residues led to a drastic reduction in pyridoxamine binding and [³H]-pyridoxine uptake (Fig. 1l; Supplementary information, Figs. S4c, S5c, g, and S6f), highlighting the importance of these interactions in vitamin B₆ recognition.

The clinical development of fedratinib stalled for 4 years due to its unexpected off-target effects on SLC19A2 and SLC19A3.^7,8,9 To elucidate the underlying molecular mechanism and aid in further drug development, we proceeded to solve the cryo-EM structure of SLC19A3 in complex with fedratinib. We first confirmed that fedratinib could inhibit the thiamine transport activity of SLC19A3 in HEK293F cells (Fig. 1m). Moreover, the increased thermostability of SLC19A3 by fedratinib further indicated that fedratinib directly binds to SLC19A3 (Supplementary information, Fig. S8a). Thus, we determined the structure of SLC19A3/fedratinib complex at a 3.8 Å resolution (Fig. 1n; Supplementary information, Figs. S8b, S9 and Table S1). The binding of fedratinib does not induce global conformational changes in SLC19A3 (Fig. 1n; Supplementary information, Fig. S1j). In this structure, fedratinib adopts a bent conformation with the 2,4-diaminopyrimidine group as the hinge. The aminopyrimidine ring of fedratinib inserts into the same position and strikes a similar pose to thiamine (Fig. 1o), consistent with the notion that the aminopyrimidine group of fedratinib mimics that of thiamine and mainly contributes to its interaction with thiamine transporters.⁸ In particular, Glu32 and Glu110 flank the aminopyrimidine ring by hydrogen bonding with the two amine nitrogens. Meanwhile, Leu35 and Tyr113 further stabilize the pyrimidine ring through hydrophobic and π–π interactions. The two arms of fedratinib fit well with the adjacent TMs, thereby extensively interacting with the surrounding residues (Fig. 1p; Supplementary information, Fig. S8c, d). Owing to the larger size of fedratinib, it competes with thiamine and pyridoxamine and thus prevents their binding to SLC19A3. To verify the fedratinib binding site, we mutated the four pivotal residues (Glu32, Trp59, Glu110, and Tyr113) to alanine and examined their influence on the function of fedratinib. Our results showed that all of these mutations dampen the binding of fedratinib to SLC19A3 and its inhibition on thiamine transport (Fig. 1m; Supplementary information, Fig. S5d, h), thus confirming our structure model.

Although the aminopyrimidine ring of fedratinib is crucial for its binding to both SLC19A3 and JAK2, the concrete interaction modes between fedratinib and these two proteins are still remarkably different (Fig. 1q).¹³ Therefore, it should be possible to reduce the adverse effects of fedratinib by further modification to decrease its affinity to SLC19A3 without affecting its interaction with JAK2. For example, the amine group connecting the pyrimidine and benzenesulfonamide groups of fedratinib binds tightly to Glu32 of SLC19A3; however, it is not involved in vital interactions with JAK2 (Fig. 1p, q). More importantly, there is extra space around this amine in its JAK2 binding pocket (Fig. 1q). Hence, adding an appropriate chemical group in this position is a potential direction for further optimization.

According to our structural and functional data, it would be tempting to suggest that the nitrogenous hexatomic ring of thiamine and pyridoxamine are the main contributors to their specific interaction with SLC19A3. In support of this notion, fedratinib also binds SLC19A3 primarily through its central aminopyrimidine group even though its molecular weight is 2–3-fold larger than those of thiamine and pyridoxamine. Taken together, these results suggest that the binding pocket of SLC19A3 recognizes certain central elements, such as the aminopyrimidine group, but also exhibits reasonable plasticity to accommodate small molecules of varying properties and sizes. This plasticity contributes to its substrate diversity (Supplementary information, Fig. S8e).