A combined computational and structural model of the full-length human prolactin receptor

The prolactin receptor is an archetype member of the class I cytokine receptor family, comprising receptors with fundamental functions in biology as well as key drug targets. Structurally, each of these receptors represent an intriguing diversity, providing an exceptionally challenging target for structural biology. Here, we access the molecular architecture of the monomeric human prolactin receptor by combining experimental and computational efforts. We solve the NMR structure of its transmembrane domain in micelles and collect structural data on overlapping fragments of the receptor with small-angle X-ray scattering, native mass spectrometry and NMR spectroscopy. Along with previously published data, these are integrated by molecular modelling to generate a full receptor structure. The result provides the first full view of a class I cytokine receptor, exemplifying the architecture of more than 40 different receptor chains, and reveals that the extracellular domain is merely the tip of a molecular iceberg.

T he prolactin receptor (PRLR) and its primary ligand prolactin (PRL) constitute a complex receptor system, linked to more than 300 biological functions ranging from reproduction and cell differentiation to immune responses 1,2 . It is best known for its role in mammary gland development and lactation 3 as well as the pathology hyperprolactinemia 4 , but has also been linked to reproductive disorders 5 as well as breast 6,7 and prostate 8 tumorigenesis, and has therefore attracted significant pharmaceutical interest.
The PRLR belongs to the hematopoietic cytokine receptor superfamily, which consists of more than 40 members 9 . It is considered an archetype of the homodimeric group 1 of the family 9 , constituting the simplest cytokine receptors and including, for example, the growth hormone receptor (GHR), the erythropoietin receptor (EPOR) and the thrombopoietin receptor 9 . All these receptors lack intrinsic kinase activity, making them dependent on associated kinases such as the Janus kinases (JAKs) to mediate signalling 9 . They are single-pass transmembrane (TM) proteins with similar overall topologies: (1) a folded extracellular domain (ECD) responsible for ligand binding; (2) a TM domain (TMD) connecting the extracellular-and intracellular parts and (3) an intrinsically disordered intracellular domain (ICD) orchestrating downstream signalling 9,10 . Their ECDs fold into two fibronectin type III domains, named D1 (membrane-distal) and D2 (membrane-proximal). The latter contains a conserved WS-motif 11 that for the PRLR acts as a molecular switch during activation 12 . The ICDs have low-sequence conservation, except for two regions named Box1 and Box2. Box1 is a membrane-proximal proline-rich motif responsible for constitutive association of JAKs 13 , while the function of Box2 remains unclear, although some studies have suggested it to be involved in JAK2 association [13][14][15] . Recently, studies have revealed that homodimerization of group I cytokine receptors may occur in the absence of hormone, and is insufficient for receptor activation [16][17][18] . Binding of a hormone to the ECDs leads to the formation of an asymmetric ternary complex consisting of one hormone and two receptor chains [19][20][21] .
Parts of the PRL/PRLR receptor system have been structurally characterized including structures of the PRL 22 , the 1:1 and 1:2 complexes of PRL:PRLR-ECD [16][17][18] and the unliganded human (h) PRLR-ECD-D2 (ref. 12), as well as a recent characterization of the hPRLR-ICD as being intrinsically disordered throughout its length 10 . However, despite substantial efforts, important structural and functional aspects of the PRLR remain uncharacterized. As no structure is available of neither the unliganded PRLR-ECD nor of the TMD, the overall structure and mechanism of signal transfer remains elusive. Signal transduction across the membrane following hormone binding is suggested to occur by subunit rearrangements or propagation of local structural changes, providing the TMDs with a key role. Furthermore, it has been shown for several class 1 cytokine receptors that the ligand-independent homodimerization takes place in the TMDs [16][17][18] . Despite these significant roles, the TMDs remain largely understudied, although cellular reports on their characteristics and roles in signal transduction are now emerging. A recent study of the GHR proposed a model for signal transduction where the homodimer TMD conformation switches upon receptor activation from a parallel to a left-hand crossover structure, thereby separating the TMDs at the C-terminal end and eventually bringing the two JAK2s into productive positions 17 . However, analogous alanine-insertion studies of the GHR 23 and the PRLR 24 suggested that the activation mechanism of the two related receptors differs in a poorly understood manner.
Although high-resolution structures of multi-pass transmembrane proteins in connection with extra-membranous globular domains have been solved (for example, refs 25,26), no such structures are available of single-pass transmembrane proteins, which have fewer stabilizing contacts in their membrane-embedded region. The structural characterization of the full PRLR is further hampered by the challenges associated with studies of a protein consisting of three structurally diverse domains: a soluble, folded domain, a membrane-embedded domain and an intrinsically disordered domain. Currently, no single high-resolution method is capable of describing the structural characteristics of a protein of this size and complexity. To overcome these inherent limitations in sample preparation and individual structure-determination techniques, we here apply a multidisciplinary divide-and-conquer approach in which we combine data acquired with different techniques on overlapping domain variants of the hPRLR with molecular modelling. In the centre of this endeavour lies the determination of the hPRLR-TMD structure, which we here solve in 1,2-dihexanoyl-sn-glycero-phosphocholine (DHPC) micelles using nuclear magnetic resonance (NMR) spectroscopy. Combining this with previously published and new biophysical data of the soluble domains, we generate a structural model of the monomeric hPRLR, suggested to represent up to 70% of the PRLRs in the cell membrane 15 . This hPRLR structure provides the first full molecular architecture of a class I cytokine receptor, exemplifying more than 40 different receptor chains, and reveals that the extracellular domain is merely the tip of a molecular iceberg.

Results
Structure of hPRLR-TMD in micelles. As the missing piece in the structural description of the hPRLR, we determined the structure of the TMD in a membrane-mimicking environment using solution state NMR spectroscopy. To ensure overlap with previously determined structures, we used a hPRLR-TMD construct with a five-residue overlap between hPRLR-D2 (ref. 12) and hPRLR-TMD in its N-terminus (F206-D210), and a five-residue overlap with hPRLR-ICD 10 in its C-terminus (G236-V240). The resulting 37-residue hPRLR-TMD harboured residues F206-V240, as well as an N-terminal G-S sequence.
hPRLR-TMD was expressed and fast-track purified to 495% purity ( Supplementary Fig. 1) using a novel method 27 and screened for suitable detergents and temperatures for the structural studies ( Supplementary Fig. 2). A high concentration of DHPC (TMD:DHPC 1:700) at 37°C provided narrow linewidths and the expected number of peaks and hence was selected for the structural studies. An SDS-PAGE of hPRLR-TMD ( Supplementary Fig. 1) revealed a clear monomer band below 14 kDa, with a faint dimer band just above 14 kDa. The deviation from the average molecular weight of hPRLR-TMD (3,959.78 Da) can be explained by the commonly encountered gel-shifting phenomenon observed for membrane proteins 28 . The oligomeric state of hPRLR-TMD reconstituted in DHPC (1:700) was further evaluated by native mass spectrometry, which is sensitive to oligomerization of membrane proteins 29 (Fig. 1a). The peaks identified at 1,980.56 m/z in the þ 2 charge state and at 3,958.16 m/z in the þ 1 charge state are consistent with the average molecular weight of one monomer (3,959.78 Da). No peaks representing higher oligomers of hPRLR-TMD were observed. Thus, hPRLR-TMD appeared monomeric under these conditions.
We proceeded to assign the chemical shifts of hPRLR-TMD in DHPC micelles by standard heteronuclear NMR methods. Manual assignments resulted in 99.0% completeness for backbone 15 N, 13 C and 1 H resonances, and 90.3% for side chain proton resonances. The secondary structure was evaluated by secondary chemical shift (SCS) values of C a , C b , C 0 and H a calculated from published random coil values 30 , and with the motif identification from chemical shifts (MICS) programme 31 using C a , C b , C 0 , N H , H N and H a chemical shifts. Together these analyses suggested a-helical conformation for residues D210-A233 (Supplementary Figs 3a and 4a), additionally supported by diagnostic a-helical nuclear Overhauser effects (NOEs) (Supplementary Fig. 3a). The SCSs further suggested that residues L234-G236 continued in a transient a-helical turn ( Supplementary Figs 3a and 4a). The SCSs and NOE patterns of residues F206-N209 and Y237-V240 did not support highly populated secondary structures for these residues. Analysis of possible N-and C-terminal a-helical capping (N-cap and C-cap) motifs using MICS ( Supplementary  Fig. 4c,d) suggested N209 or D210 (0.114 and 0.118, respectively) to potentially form transient N-cap structures.
The above results were supported by 15 N-T 2 -relaxation times measured on hPRLR-TMD in DHPC micelles ( Supplementary  Fig. 3c). Little variation was observed in backbone dynamics for residues T212-L234 with 15 N-T 2 -relaxation times just below 50 ms. Towards the N-and C-terminus, the relaxation times gradually increased, suggesting faster dynamics and correlating well with the lack of secondary structure in these regions. Similar relaxation measurements conducted on hPRLR-ICD G236-Q396 (including residues S238-V240 of hPRLR-TMD) (Fig. 2a-c) supported these findings and further showed fast backbone dynamics of the ICD.
The structure of hPRLR-TMD in DHPC micelles was determined using 754 NMR-derived restraints (Table 1), including short-and medium-range NOE-, dihedral angle-and hydrogen bonding restraints ( Supplementary Fig. 3a,b). Dihedral angle restraints were estimated by TALOS 32 and f-angles further refined from 3 J(H N -H a ) coupling constants 33 ( Supplementary  Fig. 3b). We used the presence of small coupling constants and characteristic a-helical NOE patterns as basis for including hydrogen bonds between H i (N) and (C)O i-4 as restraints ( Supplementary Fig. 3b, grey circles). A final set of 200 structures was calculated with Aria/crystallography & NMR system (CNS), 34,35 and the 10 structures with the lowest energy conformations and without significant violations were selected to represent the structure of the monomeric hPRLR-TMD (Fig. 1c,d and Table 1). The lowest energy structure formed a single a-helix from residues D210-A233 (Fig. 1e) with a length of 36 Å and two symmetrically placed tryptophans pointing in opposite directions on each side of the a-helix (Fig. 1g). A slight bend around S221 was observed, with a bending angle of 6°. The a-helix contained mainly hydrophobic residues, but harboured a few residues in the   N-terminal part with side chains capable of forming hydrogen bonds: S216, S221 and C225 (Fig. 1f). S221 and C225 were located on the same side of the helix, while S216 was located half a turn away (Fig. 1f).
Envelope structure of the unliganded hPRLR-ECD. To generate a full structural model of the monomeric hPRLR, we needed a representation of the unliganded ECD. However, all known structures of the PRLR-ECD are crystal structures solved in complex with PRL, placental lactogen or PRL-based antagonists 19,20,[36][37][38] . We previously attempted to solve the structure of the unliganded hPRLR-ECD in solution, but found it incompatible with acquisition of high-quality NMR data 12 . Instead, the structure of hPRLR-ECD-D2 was solved, revealing that the overall conformation of the unliganded ECD-D2 was similar to that of the liganded state, but with significant local differences in strand length and in the WS-motif 12 . Thus, to obtain structural information on the full ECD in the unliganded state, we measured solution small-angle X-ray scattering (SAXS) profiles of the unliganded hPRLR-ECD, shown to be monomeric by size-exclusion chromatography 22 (Figs 3b and 4b). The resulting SAXS curve was fitted to the coordinates of a liganded hPRLR-ECD structure 37 using CRYSOL 39 . The theoretical scattering curve of this ECD model fits in detail with the experimentally determined envelope (w 2 ¼ 1.2) and a pseudo-atom model was constructed using DAMMIN (Fig. 3a,b). The angle between D1 and D2 in the ECD model was, within the resolution measurable by SAXS analysis, similar to both the liganded and unliganded GHR and EPOR structures (PDB entries 2AEW, 1A22, 1ERN and 4Y5Y). Together these findings suggested that no substantial structural rearrangement occurs upon ligand binding to hPRLR-ECD, and provided a model of the unliganded hPRLR-ECD.
Ensemble description of the hPRLR-ICD. We recently showed the long-form hPRLR-ICD to be intrinsically disordered with five transiently populated a-helices 10 . Here, by NMR, we further investigated the intrinsic propensity of the ICD to homodimerize and found no evidence to support dimerization of the unmodified ICD chain (Fig. 2d)   the hydrodynamic radius (R H ) of hPRLR-ICD to 74 Å±1 Å (Fig. 4e). The calculated average R H of the FM ensemble was 58±6 Å, suggesting it to be slightly more compact than proposed by the experimentally determined R H (discussed below).
Membrane embedment and interactions. Although the TMD is presumed to be the only truly membrane-embedded part of the hPRLR, the ECD and the ICD are tethered by the TMD to the outer or inner leaflets of the lipid bilayer, respectively. Potential interactions between the hPRLR and the membrane bilayer should therefore not be ignored. We recently established that hPRLR-ICD interacts with lipids characteristic of the inner membrane leaflet 10 , but it is unknown if-and to what extentthe ECD interacts with the membrane as suggested from theoretical considerations 41 .
Using an analogous approach as for hPRLR-ICD 10 , putative interactions between hPRLR-D2 and 1-palmitoyl-2-oleoylsn-glycero-3-phosphocholine (POPC), the most abundant lipid of the mammalian outer membrane 42 , were probed using NMR by titrating small-unilamellar vesicles (SUVs) into a solution of 15 N-hPRLR-ECD-D2. However, in contrast to the ICD, we observed no effect on the chemical shifts or peak intensities on addition of 16 Â molar excess of POPC (Fig. 4a), suggesting no significant affinity between the ECD-D2 and the major lipid constituent of the membrane bilayer.
Embedment of hPRLR-TMD in DHPC was examined with a series of experiments. First, the proximity and interaction between water and backbone amides were investigated using amide-water NOEs ( Supplementary Fig. 5a). Not surprisingly, NOEs were readily detected between water and the amides in the flexible N-and C-terminal regions (F206-D210, K235-V240), whereas no detectable NOEs were identified from the majority of the amides in the TMD a-helix. However, we did observe NOEs between water and the amides of T212, V213, S216, V217 and S221, some of which are positioned close to the centre of the TMD. However, these NOE peaks could originate from the proximate hydroxyls of T212, S216 and S221. To further understand the TMD embedment, we therefore performed hydrogen-deuterium (H-D) exchange experiments at different levels of D 2 O following establishment of a quasi-stationary state ( Supplementary Figs 5b and 6). These data supported the pattern revealed from the amide-water NOE data, suggesting some degree of water contact at the N-terminal part of the a-helix, potentially explained by this being more polar than the C-terminal ( Supplementary Fig. 5c). The chemical shifts, NOEs, coupling constants and 15 N-T 2 -relaxation times clearly established structure in the N-terminal end of the a-helix. Thus, the non-uniform exchange pattern and the water NOEs are likely not a result of extreme dynamics, but instead of the N-terminal polarity in combination with the properties of the detergent.
DHPC is, as a short-tail version of POPC, in theory only capable of assembling a B16 Å wide hydrocarbon bilayer 43 , compared to the 29 Å of a POPC bilayer 44 , rendering a spherical micelle-embedment model unlikely. Thus, in light of the above data, it seems likely that DHPC patches the TMD surface with a prolate ellipsoid monolayer ( Supplementary Fig. 5d) as suggested also for the OmpX 43 . Despite this, the hPRLR-TMD region embedded in DHPC fits remarkably well with the region predicted to be within the native bilayer (T211-L234) 45 , possibly caused by anchoring at two charged residues; D210 and K235. Hence, even with the shortcomings of DHPC, the data collectively suggest that the membrane embedment range was well-simulated and constituted T211-L234.
In conclusion, the ICD has previously been shown to interact with membrane bilayer constituents, whereas the ECD-D2 appears intrinsically void of significant affinity for POPC. The data collectively supported that T211-L234 of the TMD were embedded in the membrane mimetics, while the extracellular F206-D210 and intracellular K235-V240 protruded at each end.
Data integration to generate a full structural model. We combined the experimentally based ensemble of the ICD and the unliganded ECD with the structure of the TMD to generate a complete structural model of the hPRLR (Fig. 4). The overlapping region between the hPRLR-ECD and hPRLR-TMD (F206-D210) formed little regular structure, as evidenced by the SCSs from the solution structure of the ECD-D2 (ref. 12), and the coupling constants, relaxation rates and SCSs of the TMD (Supplementary Fig. 3a-c). Thus, residues P203-N209 most likely constitute a linker region without regular secondary structure between b-strand G of the ECD-D2 and the TMD a-helix, which we term juxtamembrane-linker 1 (JML1). Proceeding the TMD helix, L234-G236 continued in a transient helical turn, while we observed no regular secondary structure in the overlap region between the TMD and the ICD (G236-V240) neither in the data from this study ( Supplementary Fig. 3a-c and Fig. 2a) nor from the previous characterization of the ICD 10 . We thus refer to residues Y237-C242 as juxtamembrane-linker 2 (JML2). Hence, we find that the overlap regions between the three isolated domains lacked regular secondary structure and therefore were suitable as assembly sites. We oriented the ECD perpendicular to the membrane surface, as no interactions between hPRLR-ECD-D2 and POPC were observed (Fig. 4a), while the TMD was oriented in accordance with the DHPC-embedment data presented above.
The resulting structure of the monomeric hPRLR provides the first view of the molecular architecture of a full class I cytokine receptor (Figs 4 and 5). Moving from N-to C-terminus, it consists of the ECD (Q1-I202), JML1 (P203-N209), TMD (D210-G236), JML2 (Y237-C242) and ICD (I243-H598). The hPRLR structure is B345 Å on the vertical axis from N to C terminus, of which the ECD constitute B20%, the TMD B10% and the ICD B70%.

Discussion
In this work we have presented a structural model of the unliganded, monomeric hPRLR-ECD generated from SAXS data. The structure of the liganded ECD 37 fits in detail into the unliganded ECD SAXS envelope, revealing that the relative orientation of D1 and D2 is preserved (Fig. 3a). Thus, in line with findings of time-resolved Förster resonance energy transfer (FRET) studies 46 , no substantial conformational changes appear to take place in the PRLR-ECD upon hormone binding. This observation is consistent with findings for the GHR; the crystal structure of the unliganded GHR-ECD 23 shows only minor changes compared to the ligand bound state, pointing towards a signal transduction mechanism that rely on subunit reorientation, as suggested for the GHR 17 . However, there are also studies pointing at differences in the signal transduction mechanism between the GHR and the PRLR. For example, although the ICDs of the receptors share properties such as intrinsic disorder and conserved motifs (Box1 and Box2), they also display interesting differences in their pattern of transient structures and number and position of lipid interaction domains (LIDs) 10 . Further, in alanine-insertion studies of the GHR 23 and the PRLR 24 , insertions in the JML2 of the GHR affected receptor activity 23 , while analogous insertions in the PRLR had no effect 24 . For the JMLs to function as hinges in the subunit reorientation mechanism they need to have some degree of rigidity. The lack of effect on insertions in the hPRLR-JML2 may be understood from the structural model presented here and previously published data 10 , showing that the hPRLR-JML2 is without rigid structure, whereas the corresponding region in the GHR showed propensity for transient a-helical structures. Thus, if JML2 in the PRLR should function as a hinge during activation, the required rigidity must be imposed by other components than the intrinsic structure alone. Overall, these findings suggest that the PRLR and the GHR may share similar activation mechanisms on the extracellular side of the receptor, but vary intracellularly in line with their different biological functions. The structure of the monomeric hPRLR-TMD presented in this work revealed the extent of the a-helical secondary structure and a characteristic tryptophan symmetry. Tryptophans are preferably positioned towards the ends of bitropic TMD a-helices 47 , receive the greatest energy reward for partitioning into the lipid head group region 48 , and may act as interfacial anchors that regulate the helix tilt angle 49 . The placement of W214 and W230 in the second helix turn from the N-and Ctermini of the TMD, respectively, fits well with a role of these in membrane anchoring. If the PRLR exerts its functions through a similar mechanism as suggested for the GHR 17 , the hPRLR-TMD should be able to switch between two different dimer conformations corresponding to the inactive and active states. With this in mind, W214 and W230 may be important not only in terms of controlling the crossing angle of the lowest energy monomer and dimer, but also in rendering a second dimerization interface, with a different crossing angle, less favourable.
Although the TMDs of class I cytokine receptors have been shown to be responsible for receptor homodimerization [16][17][18] , no classical dimerization motifs 50 are present in the hPRLR-TMD and the monomeric form was readily obtained in this work even at high concentrations. Thus, hPRLR-TMD 206-240 appears to have a weak inherent propensity to homodimerize. In fact, the lack of classical dimerization motifs is shared by the related hEPOR-TMD, and its structures have also only been solved in monomeric forms 51,52 . This weak inherent dimerization propensity suggests that other components than those of the hPRLR-TMD 206-240 /DHPC system may be important to drive TMD dimerization. Potential factors include the adjacent domains, membrane interaction partners such as specific lipid components (for example, phosphoinositides, cholesterols etc.) or accessory bound proteins (for example, JAK2). The polar residues in the core of the hPRLR-TMD a-helix (S216, S221 and C225) may be of possible relevance to homodimerization. In TMDs, such side chains are often involved in inter-monomeric hydrogen bonds, improving the stability and specificity of a-helical associations with one of the most common participants being serine 53 . Interestingly, the least common amino acid type in TM regions is cysteine 54 , suggesting that C225, placed deep within the TMD, might play a special role. A cellular mutagenesis study on the hPRLR has shown that substituting C225 with a serine decreased ligand-independent dimerization by B30% (ref. 18), suggesting that C225 takes part in TMD dimerization, without being vital. In hPRLR-TMD, S221 and C225 are positioned in two adjacent a-helical turns, while S216 is positioned half a turn away (Fig. 1f), suggesting that the hPRLR has the potential to form two different dimerization interfaces involving serines. These two could be interchangeable through rotation of the TMDs during receptor activation, perhaps with C225 as pivot point.
The structures of the monomeric TMDs from the related human and mouse (m) EPOR were recently solved in n-dodecylphosphocholine (DPC) micelles on the basis of dihedral angles obtained from TALOS using C a chemical shifts only, a modest number of NOEs and backbone hydrogen bond restraints 51,52 . Due to a lack of restraints on the side chains of these structures, we regard them mainly as backbone structures, represented by a straight a-helix spanning residues L226-W258 for both species. Surprisingly, these a-helices continue nine residues into the JM-region without any apparent flexibility, having important implications for the mode of hormone-bindinginduced signal propagation from the TMD to Box1. We note that no non-sequential NOEs for H249-L253 (hEPOR-TMD) or around H249 (mEPOR-TMD) or other data appear to support this a-helix extension 51,52 . In contrast, relaxation data for both EPOR-TMDs reveal increased internal dynamics from H249-P254, supporting a less-structured conformation not captured by the structures 51,52 . Thus, if these regions instead are interpreted as forming transient helical structures, these related structures have a similar overall topology as hPRLR-TMD.
Inherent limitations in sample preparation and individual techniques for structure determination make atomic-resolution studies of the structurally diverse hPRLR challenging. We therefore applied an approach in which data were integrated from multiple structural disciplines combining new SAXS, MS and NMR data with previously published data and molecular modelling to acquire the molecular architecture of the full hPRLR (Figs 4 and 5). This model provides the first molecular architecture of a full-length class I cytokine receptor, revealing the relative sizes of the individual domains. The hPRLR model consists of the soluble, globular ECD (Q1-I202), the unstructured JML1 (P203-N209), the membrane-embedded TMD (D210-G236), the unstructured JML2 (Y237-C242) and the intrinsically disordered ICD (I243-H598). The ICD was in a previous study shown to contain five transient a-helices (Fig. 4, red stretches) and three non-cooperative LIDs along the chain 10 , which have not been directly included in the model, but are shown in Fig. 5 (green). The full hPRLR structure reveals that the ECD, which until recently was the only structurally characterized part of the class I cytokine receptors, merely constitutes the tip of an extensive molecular iceberg. Previously, visualizations of this receptor family have typically shown a disproportionally dominating ECD compared to the TMD and ICD, likely reflecting the overall information content available for each domain. From our structure it is now evident that the ICD is more than twice as extended in the direction of the membrane normal compared to the ECD and the TMD combined (Fig. 4). The structural flexibility and large capture radius of the ICD allows it to reach and interact with a variety of interaction partners, possibly also other receptors' ICDs. Interaction sites in disordered regions typically only constitute a few residues, so-called small linear motifs (SLiMs), and hence the ICD is geared to simultaneously interact with many kinases, phosphatases and other proteins. Box1 is an example of such a SLiM in the PRLR-ICD, shown to interact with JAK2 (ref. 13), most likely through its FERM (4.1, Ezrin, Radixin, Moesin) domain 55 . To illustrate the relative sizes of these proteins, Box1 along with the FERM-SH2 domains from TYK2 (ref. 56), a JAK2 homologue, are highlighted in Fig. 5.
The generated ensemble model of the ICD represents its unbound intrinsic structure, represented by an ensemble of 25 conformers in Fig. 4. However, in the cell the structural ensemble of the ICD may be different in several ways. First, of all the three LIDs 10 have not been restricted to interact with the membrane in this model. Since experimental measurements of the R H in the presence of SUVs would be dominated by signals from the SUVs, this was not attempted, but it is likely that the LID-mediated membrane contacts would result in a smaller R H of the ICD. Thus, we regard the experimentally determined R H obtained in the absence of SUVs as an upper limit capture radius for the ICD. Second, constitutively bound kinases have deliberately been omitted, primarily because their mutual binding sites have not been adequately described, if at all known. Last, post-translational modifications of the ICD, in particular phosphorylations, may have the potential to change the structural ensemble of the ICD 57 and thus its compactness. Importantly, there are no indications to suggest the ICD to be folded, and the ICD does not intrinsically dimerize (Fig. 2d). Hence, other proteins or modifications such as acetylation and phosphorylations would be needed to promote ICD dimerization, in essence completely analogous to the hormone-induced dimerization of the ECD-D2 domains.
The presented structural model of the hPRLR provides important new insights on the full structure of class I cytokine receptors, and provides a framework for understanding the mechanisms related to these receptors, as for example, derived from cellular studies. However, being based on a divideand-conquer approach, it still poses unresolved questions. First, due to the methodological restrictions on structural characterization of a PRLR variant that includes both the ECD and TMD, the exact relative orientation of these domains remains unestablished. However, in the alanine-insertion study by Liu and Brooks 24 , insertions in PRLR-JML1 did not affect receptor functionality, suggesting that the relative orientation between the ECD and the TMD is not essential. Second, it calls for reservations that the TMD structure was solved in detergent rather than a bilayer. Consequently, the angle between the TMD and the bilayer plane (helix tilt angle) remains speculative. Furthermore, although the structural envelope was obtained in the present work and the unliganded ECD-D2 structure is available 12 , a high-resolution structure of the unliganded ECD is missing. Last, the model does not include any potential interactions with the glycosaminoglycan layer, suggested to implicate the WS-motif 11 , or effects from post-translational modifications.
In conclusion, our head to toe structural model of the hPRLR provides a starting point for future refinements and may help design strategies for novel structural and functional studies. Importantly, it exemplifies the architecture of the many biologically fundamental receptors of the class 1 cytokine receptor family, and specifies a scaffold onto which a new view on cellular signalling can be built.
Methods 25% (w/v) sucrose, 5 mM EDTA, 1 Â PBS buffer (pH 7.4), 1 mM PMSF), and sonicated on ice. Subsequently the IBs were collected by centrifugation (20,000g, 25 min, 4°C). This cycle of resuspension, sonication and centrifugation was repeated three times. The resulting IBs were resuspended in 50 mM Tris-HCl buffer, collected by centrifugation (20,000g, 20 min, 4°C), solubilized in 12 ml (per l culture) 1.5% (w/ v) sarkosyl, 100 mM DTT, 20 mM Tris-HCl buffer (pH 7.4) and incubated at room temperature with gentle agitation for 3 h. Insoluble material was removed by centrifugation (12,000g, 20 min, 4°C). The supernatant was dialyzed against 0.5% (w/ v) sarkosyl, 10 mM NaCl, 50 mM Tris-HCl buffer (pH 7.4) to remove DTT and cleaved with thrombin to release the glutathione S-transferase carrier protein. After cleavage the solution was lyophilized, followed by resuspension in milliQ water (200 ml ml À 1 of original solution). This solution was divided into batches of 50 ml, each of which was added to 750 ml of a 1:2 chloroform:methanol solution and mixed well. The solution was centrifuged (14,000g, 2 min, 4°C), resulting in separation in three layers. The top aqueous layer was carefully removed. Subsequently, 500 ml of MeOH was added to the remaining solution, followed by thorough mixing. The mixture was incubated on ice for 20 min, followed by centrifugation (16,000g, 40 min, 4°C). The supernatant containing the target protein was transferred to a glass vial, and the organic solvent evaporated under a stream of N 2 .
After sonication the supernatant containing hPRLR-ICD G236-H598 was heated for 5 min at 95°C, incubated on ice for 10 min, and centrifuged to remove precipitate (10 min, 20,000g, 4°C). The supernatant containing hPRLR-ICD G236-H598 was added 10 mM DTT, precipitated to 35% with AMS on ice, gently stirred and incubated on ice for 2 h before centrifugation (20 min, 20, In case of hPRLR-ICD G236-Q396 , DNA was precipitated with a final concentration of 0.1 % (v/v) protamine sulphate added on ice, gently stirred and incubated on ice for 10 min before centrifugation (50 min, 37,000g, 4°C). The supernatant was heated for 5 min at 95°C, incubated on ice for 10 min, and centrifuged to remove the precipitate (10 min, 20,000g, 4°C). The supernatant containing hPRLR-ICD G236-Q396 was precipitated to 35% with AMS on ice, gently stirred and incubated on ice for 2 h before centrifugation (20 min, 20,000g, 4°C). The pellet was resuspended in 20 ml 20 mM Tris-HCl (pH 8.0) and dialyzed against 1 l 20 mM Tris-HCl (pH 8.0) at 4°C before applying the sample to a 5 ml HiTrap QFF column using a protocol similar to that of hPRLR-ICD G236-H598 .
Native mass spectrometry. A sample containing 25 mM hPRLR-TMD, 700 Â molar excess DHPC and 200 mM ammonium acetate buffer was prepared for native MS by extensive buffer exchange in À 3 kDa spinfilter with 200 mM ammonium acetate buffer (pH 7). Samples were loaded into gold coated nano-electrospray emitters prepared in-house as previously described 58 . A range of dilutions (using milliQ water) were prepared for MS analysis, ranging from 10 to 100 mM, maintaining the 1:700 hPRLR-TMD:DHPC ratio.
For assignments of backbone nuclei, heteronuclear NMR spectra were recorded on a sample containing 0.8 mM 13 C-, 15 N-hPRLR-TMD in 560 mM DHPC. Backbone assignments were performed manually from the analyses of 1 H-15 N-HSQC, HNCACB, CBCA(CO)NH and HNCO spectra acquired with non-uniform sampling 61 . The backbone chemical shifts were used in TALOS 32 to estimate the dihedral angle restraints, and in the motif identification from chemical shifts (MICS) programme 31 to identify possible motifs. NOE assignments were performed manually from analysis of a 15 N-NOESY-HSQC spectrum (mixing time of 100 ms) and 13 C-NOESY-HSQC spectra of the aliphatic region (mixing time of 150 ms) and the aromatic region (mixing time of 150 ms), acquired on a sample containing 1 mM 13 C-, 15 N-hPRLR-TMD in 700 mM d 9 DHPC.
A 3D HNHA spectrum 33 was recorded on 0.7 mM 15 N-hPRLR-TMD in 490 mM DHPC and 3 J(H N -H a ) couplings constants were extracted from the relative intensity of H a and H N peaks using the CcpNmr Analysis software 60 . The 3 J(H N -H a ) coupling constants were utilized to estimate the backbone dihedral f-angles from the Karplus relationship with coefficient values of 6.51, À 1.75 and 1.60 for A, B, and C, respectively 33 . For amides with coupling constants below o5 Hz along with NOE patterns characteristic of a-helical conformation, hydrogen bond restraints were created between H i (N) and (C)O i-4 .
A series of 1 H-15 N-HSQC spectra were recorded on a sample containing 0.7 mM 15 N-hPRLR-TMD in 490 mM DHPC to analyze of the decay of the transverse relaxation (T 2 ). T 2 relaxation times were calculated from standard HSQC spectra recorded at 800 MHz using seven different relaxation delays between 10 and 130 ms. The relaxation decays were fitted to single exponentials and relaxation times calculated using the CcpNmr Analysis software 60 .
Hydrogen-to-deuterium (H-D) exchange experiments were performed on samples containing 0.4 mM 15 N-hPRLR-TMD in 300 mM DHPC on the basis of the principles of Veglia et al. 62 Samples were lyophilized followed by resolubilization in 10, 30, 40, 50, 70 or 90% (v/v) D 2 O. For each sample, a 1 H-15 N-HSQC spectrum was acquired after an incubation period of 1 h. Since the peak intensities did not change significantly between 1 and 5 h, a 1 -h incubation period was deemed sufficient for reaching a quasi-stationary state.
Structure calculations of hPRLR-TMD. The assigned NOE peaks, dihedral angles and hydrogen bonding restraints obtained for hPRLR-TMD were applied in a standard simulated annealing protocol using Aria2 (version 2.3.2) 34 and CNS 35 . NOE peak intensities were calibrated and converted to inter-proton distances by Aria2 during each iteration step using a distance cutoff of 6 Å. Each run consisted of eight iterations, with 20 structures calculated in each of the first seven. The seven structures with lowest global energy were used as starting structures in the subsequent iteration. The structure calculations were evaluated using the CcpNmr Analysis software 60 and the CING suite 63 . Iteratively, assignments were checked manually, modified if needed, and structures recalculated. In the final iteration, 200 structures were generated of which the 10 lowest energy structures without significant violations (Table 1) were selected to represent the monomeric structure of hPRLR-TMD in DHPC micelles. The structures were visualized in PyMOL (DeLano Scientific). Ramachandran-plot statistics for the structure ensemble (residues 209-235) were calculated with PROCHECK 64 and are as follows: most favored (95.2%), additionally allowed (4.4%), generously allowed (0.4%) and disallowed (0.0%).
Lipid interaction studies of hPRLR-D2. 1 H-15 N-HSQC spectra were acquired on samples containing 50 mM 15 N-hPRLR-ECD-D2 in 10 mM Na 2 HPO 4 , 10 mM TCEP, 2 mM DSS and 10% D 2 O (pH 7.4) and no POPC SUVs or POPC SUVs at a final concentration of 8 mM. Chemical shift differences and intensity ratios were compared using the CcpNmr Analysis software 60 to investigate possible interactions between hPRLR-ECD-D2 and POPC.
SAXS and envelope generation. SAXS data on the monomeric hPRLR-ECD Q1-D210 were recorded at the HESYLAB synchrotron in Hamburg, beam line X33. Scattering was recorded at three different protein concentrations (1.15, 2.24 and 4.38 mg ml À 1 in 10 mM Na 2 HPO 4 (pH 7.4)). The three scattering curves were recorded in succession flanked by recordings of the buffer background. A high-quality scattering curve was constructed by merging the low-concentration data for low-scattering angles, intermediate concentration for intermediatescattering angles and data from the high concentration experiment for the highest angle part of the scattering curve. The background sample consisted of the last (pure protein sample) dialysis buffer after dialysis was completed. Data were processed using the ATSAS package 65 . The theoretical scattering curve of the ECD model (on the basis of PDB entry 3D48 (ref. 37)) was fitted to the experimentally determined envelope (w 2 ¼ 1.22) (superposition performed in CRYSOL 39 , part of the ATSAS package 65 ) and the structure docked as a rigid body into the 3D density map by using the fit-in-map function from the UCSF CHIMERA 66 .
Relaxation measurements on hPRLR-ICD G236-Q396 . Two series of 1 H-15 N-HSQC spectra were recorded on 15 N-hPRLR-ICD G236-Q396 to analyze the T 1 and T 2 relaxation times. 1 H-15 N-HSQC spectra were recorded at 750 MHz ( 1 H) and 4°C with delay times between 10 and 1,000 ms (T 1 ) and 10-250 ms (T 2 ) with two triplicate measurements for each series. The relaxation decays were fitted to single exponentials and relaxation times determined using the CcpNmr Analysis software 60 .
The hydrodynamic radius of hPRLR-ICD G236-H598 was determined by PGSLED NMR diffusion experiments using the pulse sequence of ref. 67. The experiments were performed on 800 mM 15 N-hPRLR-ICD in 20 mM Na 2 HPO 4 /NaH 2 PO 4 , 8 mM TCEP, 0.5 mM DSS in 90% (v/v) D 2 O (pH 7). As reference, 1.5 mM a-cyclodextrin under identical buffer and experimental conditions was used. All spectra were recorded at 5°C on a Varian Inova 750 MHz ( 1 H) spectrometer. R H of hPRLR-ICD was calculated from the relative diffusion decays of hPRLR-ICD and a-cyclodextrin 67 , which has a R H of 7.52 Å (ref. 68).
Generation of FM ensemble. An ensemble of 1,000 models of the intrinsically disordered hPRLR-ICD region (G236-H598) was generated with Flexible Meccano 40 using default options and without any restraints. We used the HYDROPRO 10 (ref. 69) to predict the hydrodynamic properties (in particular the hydrodynamic radius R H ) of each of the members of the ensemble. These values were averaged (asoR H À 1 4 À 1 ) to estimate the value of R H our hPRLR-ICD ensemble. The software was used with default options, and in short describes the hydrodynamic properties of proteins by modelling the protein as a set of overlapping spheres that in turn results in a shell-model of the protein 69 .
Structural model of the full hPRLR. The full structural model was assembled through joining the individual domains at the overlapping sequences of the structures. The overlapping regions were aligned and 1,000 model templates were constructed from the three domains by building the longest model through multiple independent cycles of refinement in Modeller 9.15 (ref. 70). As starting structures we used (1) a model of the unliganded ECD, which we built from the X-ray structure of hPRLR-ECD in complex with prolactin (PDB entry 3MZG 37 , removing the prolactin before the run) using Modeller 9.15., and which was validated by comparing calculated and experimental SAXS curves using CRYSOL 39 , (2) the NMR structure of hPRLR-TMD in micelles and (3) the FM ensemble of hPRLR-ICD, which was validated by comparing calculated and experimental R H values as described above. Each of the 1,000 assembled models were subsequently refined by a Modeller routine through multiple cycles of conjugate gradient optimization (up to 100 steps each) optimizing the model with the variable target function method, followed by molecular dynamics with simulated annealing and a final optimization with conjugate gradients using the 'refine.slow' option of Modeller 70 . The ECD was oriented perpendicular to the membrane surface, as no affinity for POPC lipids was observed, while the TMD was embedded in the sketch membrane in accordance with the DHPC-embedment data. Finally, we discarded all conformations of the full-length hPRLR ensemble model where the ICD folded back into regions occupied by the bilayer.