Main

Iron is essential for life. Despite this central role in biology, free ferrous iron (Fe2+) is toxic. In excess, iron can catalyse the production of free radicals, leading to cellular damage. Iron levels are therefore tightly controlled, both at the cellular and organism level.

In mammals, iron levels are regulated by the action of hepcidin, a peptide hormone, on ferroportin (FPN), the only known iron efflux transporter2,3,4 (Fig. 1a). FPN mediates absorption of dietary iron by transporting ferrous iron across the basolateral surface of intestinal enterocytes. FPN also enables iron recycling from hepatocytes and macrophages5. The synthesis of FPN is transcriptionally regulated by cellular hypoxia, concentrations of iron and haem, and inflammatory signalling6. In settings of increased serum iron levels, liver-derived hepcidin negatively regulates cell surface FPN by acutely blocking iron transport7 and inducing FPN ubiquitination, internalization and degradation1,8,9,10. Therefore, hepcidin decreases serum levels of iron by suppressing FPN-mediated dietary iron absorption and the release of iron from cellular stores.

Fig. 1: Structures of human ferroportin.
figure 1

a, Ferroportin effluxes cellular iron (Fe2+) by an alternating access mechanism. Hepcidin binds to outward-open ferroportin and induces ubiquitination and degradation. b, Transport of Fe2+ and Co2+ by human ferroportin reconstituted into calcein-containing liposomes. Time-course experiments were performed by the addition of ions (arrow) and the steady-state kinetic analysis was performed by fitting initial rates, obtained from the linear phase of transport, to the Michaelis–Menten function. Time-course data are representative experiments, and data are mean ± s.e.m. from n = 3 independent experiments. c, Cryo-EM map of apo-FPN–Fab45D8 complex in lipid nanodisc. d, Cryo-EM density of apo-bound FPN (left) and Co2+–hepcidin-bound FPN (right). The N and C domains are coloured in different shades of blue for apo-FPN, and green for Co2+–hepcidin-bound FPN. Hepcidin (orange) binds to an extracellular facing cavity in FPN.

Full size image

Iron disorders in humans can result from dysregulation of hepcidin or FPN. Deficits in hepcidin-mediated regulation of FPN, often due to hereditary haemochromatoses, lead to iron overload and widespread tissue damage that affect the liver, heart, pancreas, and joints11,12,13. By contrast, inappropriate increases of hepcidin levels yields iron-restricted anaemia14,15. Although several approaches to restore aberrant FPN function have been evaluated in clinical trials16,17,18, none has succeeded so far.

The molecular mechanism of FPN regulation by hepcidin remains incompletely defined. A confluence of human genetics studies and structure–function evaluations have identified key regions of FPN important in hepcidin regulation7,19,20,21,22. To understand how FPN transports iron, and how this process is regulated by hepcidin, we used a combination of cryogenic electron-microscopy (cryo-EM), molecular dynamics simulations, and in vitro biochemical assays. These studies reveal the molecular recognition of iron and hepcidin by FPN, demonstrate acute inhibition of iron transport by hepcidin, and suggest a regulatory mechanism that enables hepcidin to selectively target actively transporting ferroportin molecules for degradation.

Structures of apo- and hepcidin-bound FPN

We screened the antigen-binding fragments (Fabs) of antibodies previously raised against FPN23 for use as a fiducial mark to guide image alignment of a small membrane protein embedded in a lipid nanodisc. This approach leverages a strategy previously proposed for structure determination by single particle cryo-EM24. Fab45D8 was previously shown to be non-competitive with hepcidin and does not induce FPN internalization23. Indeed, Fab45D8 did not alter the binding properties of hepcidin to nanodisc-reconstituted FPN (Extended Data Fig. 1).

We determined cryo-EM structures of FPN reconstituted in a NW9 nanodisc25 and bound to Fab45D8, both in the apo state (3.2 Å) (Fig. 1, Extended Data Table 1, Extended Data Fig. 3) and bound to hepcidin and Co2+ (2.5 Å) (Extended Data Fig. 4). We reconstituted FPN metal transport in a liposome-based calcein-quenching assay, which yielded a maximum velocity (Vmax) of −0.00269 ΔF s−1 and a Michaelis–Menten constant (Km) of 13.6 μM for FeCl2 (Fig. 1b, Extended Data Fig. 2). Unlike Fe2+, Co2+ is not readily oxidized and therefore provides a tractable surrogate divalent metal for FPN biochemical and structural studies. CoCl2 was also robustly transported by FPN with a Vmax of −0.00339 ΔF s−1 and Km of 33.5 μM, supporting its use as a substrate mimetic. Fab45D8 decreased both the Vmax and Km values of CoCl2 transport (Extended Data Fig. 2), which suggests that Fab45D8 prefers a subset of conformations within the FPN ensemble, but does not inhibit transitions required for metal transport.

The cryo-EM density map of FPN enabled building of an atomic model of FPN regions important for iron transport and hepcidin binding2,7, and a portion of the intracellular loop 3 (ICL3) important in hepcidin-induced FPN internalization8,9,10 (Extended Data Fig. 5). The FPN extracellular loop 5 (ECL5) remains unresolved, probably owing to conformational flexibility. We separately obtained an X-ray crystal structure of Fab45D8 at 2.1 Å to enable modelling (Extended Data Table 2).

Both cryo-EM structures reveal a monomeric FPN bound to a single Fab45D8 molecule, which binds FPN extracellular loop 2 (ECL2) (Fig. 1c, Extended Data Fig. 6). Similar to other MFS transporters, FPN contains 12 transmembrane (TM) helices arranged in two domains (Fig. 2a), with a large central cavity open to the extracellular side in both structures (Fig. 2b). Ferroportin shares structural similarity with a previously characterized bacterial homologue, the Bdellovibrio bacteriovorus FPN (BbFPN) transporter26 (Extended Data Fig. 6). Unlike most other MFS transporters, the FPN TM7 is interrupted by a short non-helical stretch of six residues. This unique feature, previously posited to be important in iron binding27, is shared between FPN and BbFPN. Similar to BbFPN, human FPN metal transport is stimulated by transmembrane pH gradients, but the addition of valinomycin had no notable effect (Extended Data Fig. 7).

Fig. 2: Structure of apo-FPN.
figure 2

a, Ribbon diagram of FPN reveals 12 transmembrane helices. The N and C domains are coloured in different shades of blue. Cutaway surface view (right) shows outward-open conformation. b, Intracellular gating residues are shown as sticks. c, TM10 and TM5 form an extensive network of interactions, further stabilizing the outward open conformation. Residues highlighted in red in b and c are known loss-of-function mutations that lead to ferroportin disease in humans.

Full size image

Several interacting residues define an intracellular gate that keeps the N and C domains of FPN in an outward open conformation. Similar to an interaction in BbFPN26, R489 in TM11 of the C domain forms an ionic interaction with D157 in TM4 of the N domain (Fig. 2b), which is further supported by an extended ionic and hydrogen-bonding network including residues E486 (TM11) and R88 (TM3). An additional cluster of ionic and hydrogen bonding interactions between TM5 in the N domain and TM10 in the C domain of FPN further stabilizes the outward open conformation (Fig. 2c). Mutation of several residues within the intracellular gate leads to FPN loss of function in ferroportin disease, which highlights the importance of the gate in coordinating the conformational steps necessary for iron efflux19,28.

Iron binds to the N and C domains of FPN

Cryo-EM maps from a Co2+–hepcidin–FPN complex revealed two new densities in the central cavity of FPN, corresponding to single metal-binding sites in the N and C domain, respectively (Fig. 3a–c). Within the C domain, Co2+ directly interacts with C326 in TM7b and H507 in TM11 while making a water-mediated contact with D325 and the backbone carbonyl of T320 (Fig. 3b). Notably, the tetrahedral coordination geometry for Co2+ is fulfilled by the carboxy terminus of hepcidin. Within the N domain, we observe density for Co2+ coordinated directly by TM1 residues D39 and H43 (Fig. 3c). Mutation of D39 to alanine or glutamate residues abolishes iron transport22,27, which was previously ascribed to disruption of a putative calcium-binding site within the N domain of FPN27. Our structure, by contrast, suggests that D39 directly binds the substrate. Transport assays with reconstituted FPN, however, revealed clear modulatory effects of calcium (Extended Data Fig. 7). We observed a modest decrease in both Km and Vmax values for Co2+ with intra-liposomal CaCl2, whereas the rate of transport was increased by approximately 40% for Fe2+.

Fig. 3: Iron binds to the N and C domains of FPN.
figure 3

a, Ribbon diagram of the FPN–Co2+–hepcidin complex. b, c, Close-up views of cryo-EM density for the Co2+ ion (pink) in the FPN C domain (b) and N domain (c). d, Top, in molecular dynamics simulations with Fe2+ initially positioned randomly in bulk water surrounding FPN, the Fe2+ ions spontaneously bind to a region near H507, D325 and D504. The aggregated position of Fe2+ ions from six simulations, each 1.2 μs in length, is shown superimposed with apo-FPN. Bottom, in one representative simulation, an Fe2+ ion binds within 200 ns and remains localized at this site for more than 1,000 ns. Distance shown is from the ion to the nearest oxygen atom of D325. Thick trace is a 15-ns sliding mean. e, Comparison of TM7b conformation in apo-FPN and FPN bound to Co2+ and hepcidin. In simulations without Fe2+, TM7b is dynamic, with notable fluctuation of D325. D325 coordinates Fe2+ in simulations and leads to decreased TM7b motion.

Full size image

The identification of two metal-binding sites within FPN raises the question of whether both sites are required for iron efflux. Diseases that cause mutation and experimental mutagenesis studies suggest that the N domain metal site is crucial for iron efflux, whereas the C domain site may not be required for transport. Several mutations in FPN that cause hereditary haemochromatosis directly coordinate Co2+ in the C domain, including C326S/F/Y13,29,30,31 and H507R32. These mutations are predicted to disrupt the precise tetrahedral coordination geometry that is required for metal binding within the C domain. However, these mutants are fully capable of iron transport, and lead to iron overload even in the presence of hepcidin. Although mutation of the C domain residue D325 leads to decreased iron efflux22, recent modelling studies of an inward-open conformation of FPN suggests that D325 may interact with the N domain in the outward-open state; loss of iron efflux in D325 mutants may therefore be caused by disruption of the extracellular gate33. By contrast, mutation of D39 to alanine in the N domain metal-binding site of FPN completely abolishes iron efflux from HEK293 cells22, which suggests that the N domain may be a primary site for effluxed iron. Because the C domain metal-binding site is important for hepcidin binding, it may primarily serve an important iron-dependent regulatory function in hepcidin control of FPN activity.

Hepcidin binding may influence interactions between FPN and the metal. To understand divalent metal binding to FPN in the absence of hepcidin, we used all-atom molecular dynamics simulations. We first performed six simulations of apo-FPN in a hydrated lipid bilayer with Fe2+ ions initially positioned randomly in bulk solvent. In all six independent simulations, Fe2+ ions bound spontaneously to the C domain, localizing to the unwound region of TM7 near residues D325, D504 and H507 within hundreds of nanoseconds (Fig. 3d, Extended Data Fig. 8). The Fe2+ ion also occasionally moves closer to TM1 to interact with D39 in the N domain. In parallel simulations of apo-FPN without Fe2+ ions, we observed mobility of TM7b with considerable fluctuations of D325 (Fig. 3e), consistent with comparatively weaker cryo-EM density for TM7b as compared to other transmembrane helices in apo-FPN (Extended Data Fig. 8). By contrast, both D325 and TM7b are less mobile in simulations with iron bound at the C domain site (Fig. 3e). Divalent metal binding to the C domain may therefore stabilize an otherwise dynamic TM7b in a conformation that favours hepcidin binding.

Hepcidin occludes outward-open FPN

Hepcidin binds FPN in a central cavity between the N and C domains, acting as a molecular cork to occlude the iron efflux pathway (Fig. 4a). Consistent with acute inhibition of iron transport by hepcidin7, we observed a decrease in Co2+ transport by reconstituted FPN when hepcidin is included within the liposome. By contrast, hepcidin added to the external solution had minimal effect on metal transport (Fig. 4b), which suggests that the calcein assay primarily measures transport in the forward direction (cellular efflux). Constitutively inhibited transport of a liposome-reconstituted hepcidin–FPN complex could be rescued by the addition of membrane-permeable β-mercaptoethanol, which probably denatures internally bound hepcidin by reducing its disulfide bonds (Fig. 4c).

Fig. 4: Hepcidin binding to FPN requires iron.
figure 4

a, Surface representation of the FPN–Co2+–hepcidin complex. b, Hepcidin (hep.) was added internally (int.) to calcein-loaded liposomes inhibits FPN transport. External (ext.) hepcidin has no effect on transport. c, Reconstitution of a FPN–hepcidin complex shows inhibited transport. The addition of 1 mM β-mercaptoethanol (β-ME) rescues transport activity. All transport traces are mean values from n = 3 independent experiments. d, Ribbon diagrams of apo-FPN (blue) aligned to FPN–Co2+–hepcidin (green, orange and pink spheres). Red arrows highlight structural differences. e, Close-up views of the hepcidin-binding site. Residues in red are known hepcidin-resistance mutations. Rhodamine green (RhoG)–hepcidin labelling site on position 17 is highlighted. f, Close-up of C domain metal-binding site. g, Fluorescence polarization increase in RhoG-labelled hepcidin (RhoG–hepcidin) as nanodisc-reconstituted FPN is titrated with a Kd value of 210 nM. The addition of 10 μM FeCl2 or CoCl2 increases the affinity of hepcidin to 2.5 nM and 7.7 nM (pKd = −8.11 ± 0.16), respectively. The hepcidin concentration range in healthy human adults is shown in orange. Data are mean ± s.e.m. from n = 3 independent experiments. h, C domain metal-binding site mutants decrease RhoG–hepcidin binding affinity at FPN, even in the presence of 50 μM CoCl2. Data are mean ± s.e.m. from n = 3 independent experiments. i, Model for iron-coupled hepcidin regulation of FPN function. In settings of iron efflux, TM7b is conformationally stabilized by iron coordination in the C-domain regulatory site. High-affinity hepcidin binding to outward-open FPN depends on the direct coordination of iron in the C domain.

Full size image

Although the conformations of apo-bound and Co2+–hepcidin-bound FPN are highly similar (root mean square deviation (r.m.s.d.) value of 0.73 Å), the Co2+–hepcidin-bound structure shows an approximately 3 Å rigid body separation of the N and C domains on the extracellular side of the N domain, hepcidin binding leads to displacement of TM2 from the central cavity, driven in part by a specific contact between I6 of hepcidin and T61 and Y64. Within the C domain, the largest conformational changes occur around the Co2+-binding site, leading to changes in the conformation of TM7b and the extracellular side of TM11.

Hepcidin makes extensive polar and hydrophobic contacts with FPN with a total buried surface area of around 1,300 Å2. Our structure of hepcidin bound to FPN provides insight into disease-causing mutations that are associated with FPN gain-of-function in hereditary haemochromatoses. Several FPN mutations decrease hepcidin binding to FPN, including N144H/D/T34,35,36, C326S/F/Y13,29,30,31, Y333H37, Y501C7,38, D504N7,39 and H507R32. Of these, C326 and H507 directly coordinate the cobalt ion and mutations therefore indirectly affect hepcidin affinity or alter the atomic basis of binding specificity (Fig. 4e). Other interactions between FPN and hepcidin are either hydrophobic or depend on hepcidin amide backbone atoms, which is consistent with the relatively high tolerance of amino acid substitutions within hepcidin40. For example, Y333 fits into a hydrophobic cavity in hepcidin and hydrogen bonds with the backbone carbonyl of hepcidin residue M21 (Fig. 4e). D504 in FPN coordinates the backbone amide of hepcidin H3, whereas Y501 π-stacks with the imidazole side chain of H3. A further hydrogen bond between N144 and Y501 further stabilizes this interaction network. In contrast to these mutations, the Y64N/H mutants retain hepcidin binding7, but are completely resistant to hepcidin-induced FPN ubiquitination41,42. The outward displacement of TM2 near Y64 induced by hepcidin may therefore be important for FPN ubiquitination (Fig. 4d).

Hepcidin is coupled to iron binding

The direct interaction between hepcidin and Co2+ in our structure is unexpected (Fig. 4f), and suggests that divalent metals may be important for hepcidin binding to FPN. We therefore directly tested the effect of Fe2+ and Co2+ on hepcidin affinity at FPN. A fluorescently tagged version of hepcidin labelled at position 17 (rhodamine green–hepcidin23, RhoG–Hep) (Fig. 4e) bound to nanodisc-reconstituted ferroportin with an apparent dissociation constant (Kd) value of 210 nM (pKd = −6.67 ± 0.02; mean ± s.e.m.) (Fig. 4g). In the presence of 10 μM FeCl2, we observed a markedly increased affinity of 2.5 nM (pKd = −8.61 ± 0.21), an almost 80-fold change in the potency of hepcidin at FPN. The addition of CoCl2 also increased hepcidin affinity. By contrast, CaCl2 did not support high affinity hepcidin binding (Extended Data Fig. 7), potentially due to an inability to form a direct interaction bridging FPN and hepcidin. Consistent with a cooperative effect, the effect of CoCl2 on hepcidin binding was saturable (Extended Data Fig. 7).

The reference range for hepcidin concentration in healthy adults is around 1 to 30 nM43. Our in vitro binding experiment with purified FPN indicates minimal hepcidin binding to FPN in the absence of a divalent metal, which suggests that metal binding to FPN may regulate hepcidin activity in vivo. Our structure of FPN bound to Co2+ and hepcidin revealed conformational changes in TM7b associated with hepcidin binding (Fig. 3e) and a direct contact between hepcidin and Co2+(Fig. 3b); both could be important for the observed effect of Fe2+ and Co2+ on hepcidin binding. We therefore tested whether disruption of the C domain iron-binding site influences hepcidin binding. Even in the presence of 50 μM CoCl2, the D325N, C326S and H507R mutants bound hepcidin weakly, titrating in a micromolar range similar to wild-type FPN in the absence of divalent metals (Fig. 4h). These results highlight the crucial role of the C domain metal site in potent hepcidin binding to FPN, which may be important in homeostatic control of iron levels in a physiological setting.

Discussion

Ferroportin is a central regulator of iron homeostasis in humans. Our structural and biochemical studies show that hepcidin binding to FPN is greatly potentiated by iron itself, potentially owing to the stabilizing effect iron has on the hepcidin-binding site of FPN (Fig. 4i). With iron, the binding affinity of hepcidin falls in a range concordant with the concentration of hepcidin observed in healthy human adults. In normal iron homeostasis, this may enable hepcidin to selectively bind and regulate FPN molecules actively transporting iron and loaded with Fe2+, while sparing FPN molecules located on cells with low transport activity. Hepcidin binding to FPN would both trap the transporter in an outward open state and limit iron egress; both actions acutely decrease iron efflux, as has been recently reported7. Increased levels of hepcidin levels may inappropriately overcome this regulatory strategy and degrade FPN even in the absence of active iron efflux. The potentiation of hepcidin activity by iron may therefore have immediate consequences for the development of hepcidin mimetics currently in clinical trials16. Furthermore, hepcidin antagonism by direct targeting of FPN may require molecules with high potency to overcome the nanomolar effect of the hormone in the presence of iron. The structural and functional insights into FPN function therefore provide crucial foundations for the discovery of therapeutic agents for human disorders of iron homeostasis.

Methods

No statistical methods were used to predetermine sample size. The experiments were not randomized and the investigators were not blinded to allocation during experiments and outcome assessment.

Expression and purification of human ferroportin

The wild-type human FPN1 (also known as SLC40A1) gene was cloned into a pVL1392 vector containing an expression cassette comprised of a C-terminal human rhinovirus 3C (HRV-3C) protease recognition sequence followed by a human protein C epitope tag (EDQVDPRLIDGK) and an 8× polyhistidine tag. Baculovirus was generated using Spodoptera frugiperda Sf9 insect cells (unauthenticated and untested for mycoplasma contamination, Expression Systems 94-001F) and the construct was expressed in S. frugiperda Sf9 insect cells. Cells were collected 48 h after transduction and stored at −80 °C until further use. Frozen cell pellets were thawed and washed with a hypotonic buffer (20 mM HEPES pH 7.50, 1 mM EDTA, supplemented with 20 μg ml−1 leupeptin, and 160 μg ml−1 benzamidine) before solubilizing with 50 mM HEPES pH 7.5, 300 mM NaCl, 1% (w/v) N-dodecyl-β-d-maltopyranoside (DDM, Anatrace), 0.1% (w/v) cholesteryl hemisuccinate (CHS, Steraloids), 1 mM EDTA, and supplemented with 20 μg ml−1 leupeptin, and 160 μg ml−1 benzamidine for 1 h at 4 °C. After centrifugation, the resulting supernatant was loaded on homemade anti-protein C antibody Sepharose beads and washed extensively in 50 mM HEPES pH 7.50, 300 mM NaCl, 2 mM CaCl2, 0.1% (w/v) DDM, 0.01% (w/v) CHS. FPN was eluted with 50 mM HEPES pH 7.50, 300 mM NaCl, 0.1% (w/v) DDM, 0.01% (w/v) CHS, 0.2 mg ml−1 protein C peptide (Genscript), and 5 mM EDTA. The protein was concentrated with a Vivaspin 100-kDa MWCO concentrator and the monomeric FPN fraction was collected after size-exclusion chromatography (SEC) over a Superdex S200 Increase 10/300 GL column (GE Healthcare) equilibrated with 20 mM HEPES pH 7.50, 100 mM NaCl, 0.1% (w/v) DDM, and 0.01% (w/v) CHS.

FPN mutants used in binding studies were generated using the Q5 site-directed mutagenesis method (NEB) and cloned, along the WT FPN construct, into a pEZT vector containing an expression cassette comprised of a C-terminal human rhinovirus 3C (HRV-3C) protease recognition sequence followed by an 8× polyhistidine tag and a 1D4 epitope tag (TETQSVAPA). The resulting vectors were transfected into Expi293F Human Embryonic Kidney cells (Life Technologies) with the Expifectamine transfection kit (Life Technologies) using 1 μg DNA per ml culture as per the manufacturer’s instructions. Cultures were collected 48 h after transfection and stored at −80 °C until further use. FPN mutants were purified as described for the wild type, except the cleared solubilized supernatant was loaded on Ni-NTA Sepharose beads and washed extensively in 50 mM HEPES pH 7.50, 300 mM NaCl, 0.1% (w/v) DDM, 0.01% (w/v) CHS, 30 mM imidazole. FPN was eluted with 50 mM HEPES pH 7.50, 300 mM NaCl, 0.1% (w/v) DDM, 0.01% (w/v) CHS, 250 ml imidazole. The protein was concentrated and purified by SEC as described above.

Expression and purification of MSP

Constructs encoding MSP-NW9 or MSP-NW1125 in the pET28b vector (Addgene 133442) were transformed into BL21(DE3) Rosetta Escherichia coli, and grown in terrific broth medium supplemented with 2 mM MgCl2 and 0.1% (w/v) glucose at 37 °C. At OD600 of around 0.6, expression was induced by addition of 400 μM isopropyl β-d-1-thiogalactopyranoside (IPTG) and lowering the temperature to 20 °C. Cells were collected after 16 h and resuspended into 5 ml lysis buffer (200 mM Tris pH 8.0, 500 mM NaCl, 1% (v/v) Triton X-100 (Sigma), 0.02 mg ml−1 leupeptin, 0.16 mg ml−1 benzamidine and benzonase) per gram pellet. After stirring for 30 min at 4 °C, cells were lysed by pulsed sonication on ice. The lysate was cleared by centrifugation at 15,000g for 25 min at 4 °C and loaded on Ni-NTA Sepharose. Ni-NTA beads were washed with 50 mM Tris pH 8.0, 500 mM NaCl, 1% (v/v) Triton, then 50 mM Tris pH 8.0, 500 mM NaCl, 50 mM sodium cholate, then 50 mM Tris pH 8.0, 500 mM NaCl, and finally with 50 mM Tris pH 8.0, 500 mM NaCl, 30 mM imidazole. MSP was eluted with 50 mM Tris-HCl pH 8.0, 500 mM NaCl, 400 mM imidazole and dialysed into 50 mM Tris-HCl, pH 8.0, 100 mM NaCl, 1 mM EDTA, 0.1 mM TCEP at 4 °C. The next day, MSP was concentrated on a Vivaspin 10-kDa MWCO concentrator, aliquots were fash frozen in liquid nitrogen and stored at −80 °C for reconstitution.

Isolation, expression and purification of Fab45D8

The heavy and light chain sequences of monoclonal antibody 45D823 were separately cloned into pcDNA3.4 and the resulting vectors were transfected into Expi293F human embryonic kidney cells (unauthenticated and untested for mycoplasma contamination, Life Technologies) using a 1:2 mass ratio of light and heavy chain DNA with the Expifectamine transfection kit (Life Technologies) as per the manufacturer’s instructions. Supernatant containing monoclonal antibody 45D8 was collected 136 h after transfection and loaded on homemade protein G sepharose beads and extensively washed with a buffer containing 20 mM HEPES pH 7.50, and 100 mM NaCl. Monoclonal antibody 45D8 was eluted with 100 mM glycine (pH 3.0) and fractions were immediately neutralized with 200 mM HEPES pH 7.50. To generate the Fab fragment, 10 mg of purified antibody 45D8.1 was diluted into 9.5 ml freshly prepared cleavage buffer (20 mM sodium phosphate pH 7.00, 10 mM EDTA, and 10 mM cysteine) and treated with 0.5 ml agarose immobilized papain (Thermo Scientific) at 37 °C. After 16 h, the cleaved Fab45D8 fragment was purified by reverse Protein A affinity chromatography, followed by SEC into buffer containing 20 mM HEPES pH 7.5 and 100 mM NaCl. Fab45D8 was concentrated on a Vivaspin 10-kDa MWCO concentrator, and aliquots were flash-frozen in liquid nitrogen and stored at −80 °C for later use.

Reconstitution of FPN into lipidic nanodisc

Purified FPN (0.2–0.5 mg) was mixed with purified MSP and a lipid mixture containing a 2:3 weight ratio of 1-palmitoyl-2-oleoylphosphatidylcholine (POPC, Avanti) and 1-palmitoyl-2-oleoyl-sn-glycero-3-phospho-(1′-rac-glycerol) (POPG, Avanti). For reconstitution into NW9 nanodiscs, an FPN:MSP:lipid molar ratio of 1:20:1,100 was used. For reconstitution into NW11 nanodiscs, an FPN:MSP:lipid molar ratio of 1:20:800 was used. The reconstitution sample was nutated for 1 h at 4 °C before addition of 0.2 g ml−1 SM2-BioBeads (BioRad), and the reconstitution sample was further nutated overnight at 4 °C before removal of the biobeads. FPN containing nanodiscs for cryo-EM studies were purified by loading the reconstitution sample on anti-protein C antibody Sepharose beads and washing extensively with 20 mM HEPES pH 7.5, 100 mM NaCl, and 1 mM CaCl2 to remove empty nanodiscs. FPN containing nanodiscs were eluted with 20 mM HEPES pH 7.50, 100 mM NaCl, 0.25 mM EDTA, and 0.2 mg ml−1 protein C peptide (Genscript) and concentrated on a Vivaspin 50-kDa concentrator. FPN containing nanodiscs for binding studies were purified by SEC over a Superdex S200 Increase 10/300 GL column (GE Healthcare) equilibrated with 20 mM HEPES, pH 7.5, 100 mM NaCl.

Crystallization and structure determination of Fab45D8

Purified Fab45D8 was diluted to 13.0 mg ml−1 in 20 mM HEPES pH 7.5, 100 mM NaCl. Fab45D8 crystals were obtained in 0.3 M trimethylamine-N-oxide (TMAO), 0.1 M Tris pH 8.5, and 30% (w/v) PEG 2000 MME at 20 °C. Individual crystals were flash frozen in liquid nitrogen after a 30 s soak in 0.3 M TMAO, 0.1 M Tris pH 8.5, and 30% (w/v) PEG 2000, and 20% (v/v) ethylene glycol. A full diffraction dataset was collected at the Advanced Photon Source GM/CA-CAT beamline 23ID-B, and processed using xia2dials44 implementation of XDS45. The structure of the Fab was solved by molecular replacement using Phaser46, with a search model of a closely related germline mouse monoclonal antibody (PDB code 6BZV47) lacking complementarity determining regions (CDRs). The model was iteratively improved by refinement in Coot48 and Phenix49. Data collection and refinement statistics are summarized in Extended Data Table 2. The final model contained 96.77%, 2.23% and 0% in the favoured, allowed and outlier regions of the Ramachandran plot, respectively as assessed by MolProbity50.

Calcein transport assay for divalent cations

FPN was reconstituted into liposomes for divalent cation transport assays. Empty liposomes were prepared as a 3:1 mass ratio of 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphoethanolamine (POPE, Avanti) to POPG dissolved in chloroform, followed by gentle evaporation of the chloroform under a stream of nitrogen gas, and overnight desiccation. The lipids were dissolved in 20 mM HEPES pH 7.40, 100 mM KCl to a final concentration of 12.5 mg ml−1, sonicated until homogenous, subjected to multiple freeze-thaw cycles, and extruded through a 400 nm polycarbonate filter (Avestin) to generate unilamellar vesicles. Subsequently 0.13% (w/v) Triton-X100 (Sigma) was added to destabilize liposomes, corresponding to the onset of solubilization as determined by turbidity measurements at 540 nm. Purified FPN was added at a 1:50 protein to lipid mass ratio and incubated for 15 min at 4 °C. The FPN–hepcidin complex was reconstituted by incubating 1 μM purified FPN with 3 μM hepcidin (Bachem) and 3 μM CoCl2 for 30 min on ice before addition of liposomes. The FPN–Fab complex was reconstituted by incubating 1 μM purified FPN with 2 μM purified Fab45D8 for 30 min on ice before addition of liposomes. Control liposomes devoid of FPN were prepared in parallel using the same concentration of DDM. To remove excess detergent, 0.05 g ml−1 of SM2-BioBeads were added to the sample and nutated for 1 h at 4 °C, then 0.05 g ml−1 SM2-BioBeads were added followed by incubation overnight at 4 °C, and finally addition of 0.08 g ml−1 SM2-BioBeads followed by incubation for 2 h at 4 °C. Liposomes were collected by ultracentrifugation at 264,500g for 30 min and resuspended at a concentration of 4.0 mg ml−1 lipids in internal buffer comprised of 100 mM KCl and 20 mM HEPES pH 7.40 or pH 7.50, before flash freezing in liquid nitrogen and storage at −80 °C. On the day of the transport assay, liposomes were thawed and diluted to 2 mg ml−1 with internal buffer containing 250 μM calcein (Sigma), and where indicated 5 μM hepcidin (Bachem) or 1.25 mM CaCl2 was added. The samples were then subjected to three freeze–thaw cycles, and extruded through a 400-nm polycarbonate filter. The calcein-containing liposomes were concentrated and washed four times by repeated ultracentrifugation and resuspension in external buffer containing 100 mM NaCl and (i) 20 mM HEPES pH 6.80, pH 7.40, pH 7.50, or pH 8.20, or (ii) 20 mM Tris/MES pH 6.00, pH 7.50 or pH 9.00. Reconstitution yields determined by the Micro BCA protein assay kit (Pierce) were 205–222 pmol mg−1 lipid for FPN samples, and 275 pmol mg−1 lipid for the FPN–Fab45D8 complex assuming 1:1 stoichiometric incorporation. Presence of FPN and Fab were further verified by SDS–PAGE. Immediately before the assay, liposomes were diluted to 0.125–0.25 mg ml−1 lipid in external buffer. Time-course fluorescence traces were recorded as 1–2 s integrations using a FluoroMax-4 (Horiba) or a FluoroLog (Spex) with λex of 490 nm and λem of 520 nm. Steady-state fluorescence was recorded for at least 2 min, before addition of small aliquots of freshly prepared stocks of FeCl2, CoCl2, MnCl2, NiCl2, CuCl2, or ZnCl2. To stabilize the ferrous (Fe2+) state, we prepared iron as a 1:10 ratio of sodium ascorbate:FeCl2 immediately before the experiment. A negative-inside membrane potential was generated by incubating liposomes with 100 nM of the K+-selective ionophore valinomycin (Sigma) for 1 min before the experiment. To determine the full extent of the calcein-quenching response, 10 μM of the divalent cation ionophore calcimycin (Sigma) was added at the end of the experiment. Transport data were normalized to the mean baseline fluorescence intensity before addition of ion, and initial rates of transport were obtained from linear regression and fit to the Michaelis–Menten function in GraphPad Prism 8 (GraphPad Software). Statistical significance was determined using Student’s t-test. For clarity, mixing artefacts ± 5 s from the time of addition have been removed from representative fluorescence traces.

Hepcidin-binding assays

Fluorescence polarization measurements were performed using rhodamine-green labelled hepcidin labelled at position 17 (RhoG–hepcidin)23. For FPN saturation binding experiments, samples were prepared in a black 384-well plate (Greiner) containing 0–1 μM of nanodisc-reconstituted NW11-FPN and 5 nM RhoG–hepcidin in sample buffer comprising 20 mM HEPES pH 7.50, 100 mM NaCl, and supplemented with FeCl2 or CoCl2 as indicated. For ion stimulation experiments, 100 nM NW11-FPN and 5 nM RhoG–hepcidin was mixed with 0–600 μM of CoCl2 or 0–1 mM CaCl2. For Fab-binding experiments, 100 nM NW11-FPN and 5 nM RhoG–hepcidin was mixed with 0–3 μM of Fab45D8 in sample buffer containing 10 μM CoCl2. Binding reactions were equilibrated for 60 min at room temperature, and fluorescence polarization was recorded on a Biotek Synergy H4 (Agilent) in polarization mode using fixed bandpass filters with λex of 484 nm and λem of 520 nm. Binding data were processed in GraphPad Prism 8 (GraphPad Software) using 3- or 4-parameter curve fitting.

Analytical fluorescence size-exclusion chromatography was performed by mixing 25 μg of NW11-FPN with 2× fold molar excess of RhoG–hepcidin in sample buffer consisting of 20 mM HEPES (pH 7.5), 100 mM NaCl and 10 μM CoCl2. Samples were incubated for 20 min on ice and 1.5× molar excess of Fab45D8, or sample buffer, was added followed by incubation for 30 min on ice. For homologous competition, 1 μM NW11-FPN was mixed with 30 μM unlabelled hepcidin (Bachem) in sample buffer containing 20 mM HEPES (pH 7.5), 100 mM NaCl and 10 μM CoCl, and incubated for 30 min on ice. Then, 2 μM RhoG–hepcidin was added and the sample incubated for 30 min on ice. Samples were injected on a Superdex 200 Increase 10/300 GL column (GE Life Sciences) pre-equilibrated in 20 mM HEPES (pH 7.5), 100 mM NaCl, and 10 μM CoCl2. RhoG–hepcidin fluorescence was recorded using an FP-1520 Intelligent Fluorescence Detector (Jasco) with λex of 493 nm and λem of 524 nm.

Cryo-EM sample preparation and data collection

Nanodisc-reconstituted apo-FPN was mixed with 1.15 molar excess of Fab45D8 and incubated on ice for 30 min. The complex was purified by size-exclusion chromatography over a Superdex S200 Increase 10/300 GL column (GE Healthcare) equilibrated with 20 mM HEPES (pH 7.5), 100 mM NaCl. For Co2+ or hepcidin samples, 600 μM CoCl2 and 30 μM hepcidin (Bachem) were added to nanodisc-reconstituted FPN and incubated for 20 min on ice before addition of Fab45D8. The resulting complex was purified over SEC as for the apo sample but with the addition of 100 μM CoCl2 in the chromatography buffer. Collected fractions were supplemented with fresh hepcidin to 30 μM. For both preparations, fractions containing the nanodisc–FPN–Fab45D8 complex were concentrated to approximately 3 mg ml−1 on a Vivaspin 50-kDa MWCO concentrator and freshly used for electron microscopy.

For high-resolution cryo-EM, the apo-FPN–Fab45D8 complex was diluted to 0.0375 mg ml−1 in 20 mM HEPES pH 7.5, 100 mM NaCl directly before vitrification, and 2 μl sample was applied to glow-discharged gold holey carbon 1.2/1.3 300-mesh grids (Quantifoil) coated in-house with graphene oxide51,52,53. Grids were blotted for 2–4 s at 0 force and 10 s wait time before being plunge vitrified in liquid ethane using a MarkIV Vitrobot (ThermoFisher). The blotting chamber was maintained at 22 °C and 100% humidity during freezing.

Co2+ or hepcidin samples were diluted to 1.5 mg ml−1 in gel filtration buffer (20 mM HEPES (pH 7.5), 100 mM NaCl, 100 μM CoCl2) before vitrification on UltrAufoil 1.2/1.3 300 mesh grids. Grids were blotted for 3 s at 0 force and 5 s wait time before plunge vitrification in liquid ethane using a MarkIV Vitrobot (ThermoFisher). The blotting chamber was maintained at 22 °C and 100% humidity during freezing.

FPN–Fab45D8 and Co2+/hepcidin–FPN–Fab45D8 movies were collected using a Titan Krios (ThermoFisher) outfitted with a K3 camera and Bioquantum energy filter (Gatan). The K3 detector was operated in super-resolution mode and the energy filter slit width was set to 20 eV. Movies were collected at a nominal magnification of 105,000×, physical pixel size 0.834 Å pixel−1, with a 70-μm C2 aperture and 100-μm objective aperture at a dose rate of 8 e− pixel−1 s−1. A total dose of 66 e− Å−2 was collected as a 120-frame movie, resulting in a 6-s movie with 0.55 e− per frame. Data were collected using semi-automated imaging scripts in SerialEM54. For FPN—Fab45D8, 5,009 movies were collected using a 3 × 3 image shift pattern at 0° tilt and 406 movies were collected on-axis with a 30° stage tilt in two separate data collection sessions. For Co2+/hepcidin–FPN–Fab45D8, 4,395 movies were collected at 0° tilt in a single data collection session.

Cryo-EM image processing

For FPN–Fab45D8, data were motion corrected and 2x binned on-the-fly using MotionCor255 in the SCIPION pipeline56. Motion corrected micrographs were imported into cryoSPARC57 and RELION58 and contrast transfer function parameters were calculated using CTFFIND459. CTF information for tilted images was estimated using patch CTF estimation in cryoSPARC. A total of 138,314 particles were selected from 497 micrographs using the Blob picker in cryoSPARC. 2D class averages were generated after extracting the putative particles with a 300-pixel box and binning to 64 pixels. Six of these averages were used as templates for further particle picking. Template picking yielded 4,737,795 particles. These were split into 6 groups to increase speed of processing, extracted in a 300-pixel box, and binned to 64 pixels. 2D classification was run in cryoSPARC with default settings except: number of 2D classes 200, initial classification uncertainty factor 4, number of online-EM iterations 40, batch size per class 300. Objectively ‘good’ (showing clear Fab and receptor density) class averages were selected and exported to RELION format using csparc.py60. Class averages that were not classified as ‘good’, but were not clearly ice contamination or graphene oxide edges, were run through a second round of 2D classification with default settings except: number of 2D classes 200, number of online-EM iterations 40, batch size per class 300. All ‘saved’ class averages from the second rounds of 2D classification in cryoSPARC were selected and exported to RELION format using csparc.py. Particles were extracted from CTF-corrected images in RELION at a box size of 300 pixels and binned to 128 pixels. In total, 1,326,130 particles, in three groups (detailed in Extended Data Fig. 3) were classified in 3D with image alignment in RELION using: an initial model generated in cryoSPARC from 80,000 particles collected on a Talos Arctica filtered to 40 Å, C1 symmetry, a regularization parameter of 4, for 30–35 iterations with no mask. Particles from classes with resolved transmembrane helices were selected, extracted in a 300-pixel box, and imported back into cryoSPARC. Non-uniform refinement was run with default settings and no resolution limit, resulting in angle and shift assignments for 850,000 particles. These particles were subsequently exported to RELION format using csparc.py and run through 3D classification without image alignment. Four of the twelve classes were selected, imported into cryoSPARC, run through non-uniform alignment with an automatically generated mask, and refined to a reported global resolution of 3.2 Å. The resulting map showed signs of mild preferred orientation (Extended Data Fig. 3). The particles were exported into RELION format using csparc.py, converted into an image stack, and imported into cisTEM61 as a refinement package. The particles were reconstructed and half-maps were generated using the ‘generate 3D’ command. These half-maps, as well as the half-maps from cryoSPARC were run through our lab’s directional Fourier shell correlation (dFSC) program62 clearly showing a more evenly weighted range of views in the map generated by cisTEM. Maps were sharpened in RELION. Average resolutions are reported using the FSC = 0.143 cut-off63 and were estimated in cryoSPARC and cisTEM. Local resolution estimates were performed through cryoSPARC, using a similar algorithm to blocres64, with an automatically generated kernel size of 24 voxels.

For Co2+/hepcidin–FPN–Fab45D8, data were motion corrected and x binned on-the-fly using MotionCor255 in the SCIPION pipeline56. Motion corrected micrographs were imported into cryoSPARC57 and contrast transfer function parameters were calculated using patch CTF estimation through cryoSPARC59. A total of 3,753,516 particles were selected from 4,395 micrographs using the template picker in cryoSPARC with 2D averages from the apo-FPN data set as templates. These particles were extracted in a 360-pixel box, and binned to 64 pixels. 2D classification was run in cryoSPARC with default settings except: number of 2D classes 200, initial classification uncertainty factor 4, number of online-EM iterations 40, batch size per class 300. Objectively ‘good’ (showing clear Fab and receptor density) class averages were selected for 3D classification. Class averages that were not classified as ‘good’, but were clearly not ice contamination, were run through a second round of 2D classification with default settings except: number of 2D classes 200, number of online-EM iterations 40, batch size per class 300. All ‘saved’ class averages from the second rounds of 2D classification in cryoSPARC were sent to 3D classification. These particles were subjected to two rounds of heterogeneous refinement in cryoSPARC that serves as a ‘trash collector’. Four initial models were used, three generated from an early round of ab initio model generation and our final apo-FPN–Fab45D8 structure. All initial models were filtered to 30 Å before refinement. Particles were unbinned and a final heterogeneous refinement was performed with three good initial models of apo-FPN–Fab45D8. Non-uniform refinement was run with default settings and no resolution limit, on the most populated class, showing clear density for hepcidin, resulting in angle and shift assignments for 310,647 particles. The particles were exported into RELION format using csparc.py, converted into an image stack, and imported into cisTEM61 as a refinement package. The particles were reconstructed and half-maps were generated using the ‘generate 3D’ command. These half-maps, as well as the half-maps from cryoSPARC were run through our laboratory’s directional Fourier shell correlation (dFSC) program62 clearly showing a more distributed range of views in the map generated by cisTEM. Maps were sharpened in cisTEM. Average resolutions are reported using the FSC = 0.143 cut-off63 and were estimated in cryoSPARC and cisTEM. Local resolution estimates were performed through cryoSPARC with an automatically generated kernel size of 18 voxels. Another job was run with a kernel size of 20 voxels and results were similar.

Model building and refinement

For apo-FPN, a homology model of human FPN in the outward open state was built using Modeller65, with a previously determined X-ray crystal structure of outward open BbFPN (PDB code 5AYN)26 as a template. After truncating putatively flexible regions (N terminus, ICL3, ECL5, and C terminus), the resulting model was fit into the 3.2 Å cryo-EM map of FPN–Fab45D8 using Chimera66. The initial template was manually rebuilt in Coot48 and iteratively refined with real space refinement implemented in Phenix49. Model geometry was assessed using MolProbity50. Modelling of FPN–Co2+–hepcidin complex used the apo-FPN structure as a starting point, with iterative improvements in Coot and Phenix. We used a previously determined X-ray crystal structure of hepcidin bound to a neutralizing Fab as a starting model21 for the peptide. Further validation was performed with EMRinger67 to compare the map and final model. Map-to-model FSCs were calculated within Phenix. Figures were prepared in Chimera66 and PyMol.

Molecular dynamics simulations

The structure of the outward-open apo conformation of human FPN was used as the starting coordinates for all simulations. Three different conditions were simulated (Supplementary Table 1): (1) the iron-absent condition, where no iron was added; (2) the iron-bound condition, where an Fe2+ ion was placed in the proposed iron binding site 5.6 Å from D325 α-carbon, 6.9 Å from D504 α-carbon, and 7.8 Å from H507 α-carbon; (3) the iron-in-bulk-solvent condition, where 15 Fe2+ ions were placed randomly in the water box outside the protein using Dabble68.

Simulation coordinates were prepared by removing non-FPN molecules from the initial structure. Prime (Schrödinger) was used to model missing side chains, and neutral acetyl and methylamide groups were added to cap protein termini. The unresolved loops between TM6-TM7 and TM9–TM10 (residues 239–290 and 394–451 respectively, inclusively) were not modelled. The termini surrounding these loops were capped. PropKa was used to determine the dominant protonation state of all titratable residues at pH 769,70. The structure was internally hydrated using Dowser71. Dabble was used to additionally fill the extracellular cavity68. The structure was aligned using the Orientation of Proteins in Membranes (OPM) server72.

Using Dabble, the protein was inserted into a pre-equilibrated POPC membrane bilayer. For all simulations except condition three (iron in bulk solvent), sodium and chloride ions were added at 150 mM to neutralize the system. For condition three, chloride ions were added to neutralize the system resulting in a concentration of 108 mM. A periodic box was used with dimensions 90 × 90 Å in the x–y plane and a water buffer of 10 Å above and below the protein to the periodic boundary. We used the CHARMM36m parameters for lipids, proteins, sodium and chloride ions, and the TIP3P model for waters73,74,75. The Fe2+ Lennard–Jones parameters were obtained from a compromise model previously described76.

All simulations were run on a single Graphical Processing Unit (GPU) using the Amber18 Compute Unified Device Architecture (CUDA) version of particle-mesh Ewald molecular dynamics (PMEMD)77,78. For each condition, six independent simulations were run. For each independent simulation, the system was minimized with 500 steps of steepest descent followed by 500 steps of conjugate gradient descent three times. Harmonic restraints of 10 and 5 kcal mol−1 Å−2 were used on the protein, lipid and Fe2+ ions for the first and second minimization, respectively. Harmonic restraints of 1 kcal mol−1 Å−2 were used on the protein and Fe2+ ions for the final minimization. The system was then heated from 0 K to 100 K over 12.5 ps in the NVP ensemble with a Langevin thermostat and harmonically restraining the protein heavy atoms and Fe2+ ions with a restraint of 10 kcal mol−1 Å−2. The system was further heated with the same restraints from 100 K to 310 K in the NPT ensemble over 125 ps. The system was equilibrated with harmonic restraints on protein heavy atoms and Fe2+ ions for 30 ns. The restraint strength started at 5 kcal mol−1 Å−2 and was reduced by 1 kcal mol−1 Å−2 every 2 ns for the first 10 ns and then by 0.1 kcal mol−1 Å−2 every 2 ns for the final 20 ns. Production simulations were performed at 310 K and 1 bar using the NPT ensemble, a Langevin thermostat and a Monte Carlo barostat. Every 200 ps snapshots were saved. All simulations were run for at least 2.2 μs. These simulations used a 4-fs time step with hydrogen mass repartitioning79. Bond lengths to hydrogen atoms were constrained using SHAKE79,80. Non-bonded interactions were cut off at 9 Å.

Simulation analysis methods

Molecular dynamics simulation snapshots were reimaged every 1 ns and centred using CPPTRAJ package in AmberTools1881. Simulations were visualized using Visual Molecular Dynamics and figures prepared in PyMOL82. Time traces from simulation were smoothed using a moving average with a window size of 15 ns unless otherwise indicated and visualized with the PyPlot package from Matplotlib. For all analysis in the manuscript that required structural alignment, we aligned to the initial ferroportin structure using the backbone atoms of residues 26–116, 127–228, 308–483, and 492–543.

To investigate the localization of Fe2+ ions, the iron-in-bulk-solvent simulations (condition 3) were analysed. To visualize the density of Fe2+ ions under a given simulation condition, the position of Fe2+ ions was recorded every 10 ns for each of the 6 simulations, each 2 μs in length. Each Fe2+ ion position was then drawn as a point superimposed on the starting structure (Fig. 3d). To quantify the binding events, the distance between iron and the closest side chain oxygen atom on D325 was measured. This distance was graphed over 1.2 μs, including the equilibration time (Fig. 3d, Extended Data Fig. 8b).

To investigate the dynamics and conformation of the TM7b region, the iron-absent (condition 1) and iron-bound (condition 2) simulations were compared. The TM1–TM7b distance was measured using distance between the Cα of V51 and the Cα of Y333. For each simulation, we calculated the average of the distance over 2.2 μs, excluding equilibration. The average over the simulations for each condition was plotted with error bars representing the s.e.m. (Extended Data Fig. 8g). The dynamics of D325 were also investigated. The dynamics were visualized by overlaying representative frames showing the movement of D325, the binding site, and TM7b. For iron-absent simulations, frames from a single replicate at 200, 350, 500 and 550 ns were overlaid (Extended Data Fig. 8d). For iron-bound simulations, frames from a single simulation at 200, 500, 725 and 1,000 ns were overlaid (Extended Data Fig. 8e). The conformational range of D325 was quantified by measuring the distance between Cγ of D325 and Cβ of S47. This was visualized for one simulation for each condition over a time of 1 μs inclusive of equilibration (Extended Data Fig. 8g). For each independent simulation, the mean of the Cγ D325–Cβ S47 distance was calculated over 2.2 μs. For each condition, the average over the simulations was plotted with error bars representing the s.e.m. (Extended Data Fig. 8g). The flexibility of D325 was quantified by calculating the root-mean-square fluctuation (RMSF) of the side-chain atoms of D325 using an in-house script (Extended Data Fig. 8g). Statistical significance was determined using the Mann–Whitney U-test (two-tailed).

Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this paper.