Structural Requirements of the Extracellular to Transmembrane Domain Junction for Erythropoietin Receptor Function*

The erythropoietin receptor (EpoR) is crucial for erythrocyte formation. The x-ray crystal structures of the EpoR extracellular domain lack the juxtamembrane (JM) region and the junction to the transmembrane (TM) domain. Yet the JM-TM regions are important for transmitting the conformational change imposed on the receptor dimer by Epo binding. Cysteine-scanning mutagenesis of the JM-TM regions identified three novel constitutively active mutants, demonstrating close disulfide-bonded juxtapositioning of these residues in the JM (L223C) and N-terminal TM domain (L226C, I227C). Chemical cross-linking defined the interface of the active helical TM dimer and revealed that the JM-TM segment encompassing Leu226-Leu230 is non-helical. Molecular dynamics and NMR studies indicated that the TM-JM junction forms an N-terminal helix cap. This structure is important for EpoR function because replacement of this motif by consecutive leucines rendered the receptor constitutively active.

The erythropoietin receptor (EpoR) 1 belongs to the cytokine receptor family (1), and along with its ligand erythropoietin (Epo), is the primary regulator of mammalian erythropoiesis (2,3). Activation of the EpoR by Epo results in the activation of cytosolic Janus kinase 2 (JAK2) tyrosine kinase (4). Activated JAK2 phosphorylates tyrosine residues on itself, in various signaling molecules and in the EpoR cytoplasmic domain. These EpoR phosphotyrosines become docking sites for several signaling molecules and can activate the mitogen-activated protein kinase pathway. Phosphorylated EpoR recruits and activates STAT (signal transducers and activators of transcription) proteins, which are involved in the regulation of gene expression (5).
The EpoR has a large extracellular ligand binding domain, a single transmembrane (TM) helix, and a large cytoplasmic domain. The crystal structure of the extracellular ligand binding domain of the EpoR has been solved in complex with Epo (6), an agonist peptide (EMP1) (7), an antagonist peptide (EMP33) (8), and alone (9). These structures were dimeric and differed by the relative orientations of the extracellular domains. Receptor dimerization alone is not sufficient for activation as studies with dimerizing antagonists of Epo show that inhibitory dimeric conformations exist (8). In addition, we have recently shown that the relative dimeric orientation of the TM and cytoplasmic domains is also important for activation (10,11). The unliganded EpoR is assumed to adopt a different conformation from the ligand-activated receptor, but both orientations are thought to be stabilized by interactions between the EpoR TM domains (12,13).
The TM region of the EpoR stretches from roughly Leu 226 to Leu 247 . These 22 mostly hydrophobic residues are sufficient to span the cell membrane. The TM region must adopt helical secondary structure in order to satisfy backbone hydrogen bonds within the hydrophobic membrane interior. The juxtamembrane sequence on the cytosolic side of the EpoR appears to be ␣-helical and contains within its first eleven amino acids a hydrophobic motif (Leu 253 , Ile 257 , Trp 258 ), which is required for JAK2 activation (10). This motif is conformationally rigid, and its orientation is dictated by the register of the EpoR TM domain. Precise orientation of this motif is important for optimal signaling not only for the EpoR, but also for the gp130 subunit of the interleukin 6 receptor family (14).
In contrast, there are no data on the juxtamembrane region on the extracellular side of the membrane. It is not known where the transition between the extracellular domain and the TM ␣-helix occurs or whether this transition has a rigid structure. In this report, we define where that transition takes place by using several complementary approaches: single cysteine scanning mutagenesis and cross-linking, NMR spectroscopy, and molecular dynamics simulations. We show that three cysteine mutants in this region (L223C, L226C, I227C) result in constitutive activation of the receptor, we define the interface between the TM domains of an activated EpoR dimer, and we show that the junction between the JM and TM domains is not helical and its sequence is important for proper receptor function. 2

EXPERIMENTAL PROCEDURES
Site-directed Mutagenesis-cDNAs were cloned into the pMX-IRES-GFP bicistronic retroviral vector upstream of the IRES (15). In the wild-type HA-tagged murine EpoR cDNA Cys 180 and Cys 289 were mutated to Ser and Ala, respectively. By introducing a stop codon after amino acid 355 a thiol-free receptor backbone was created into which all single amino acid substitutions were cloned. cDNAs coding for the 28 single cysteine mutations in the EpoR juxtamembrane and transmembrane domains (A220C-L247C) as well as the Leu 6 and Leu 9 mutations were generated by overlap extension PCR with synthetic oligonucleotides. All constructs were verified by sequencing.
Cell Culture-Six-well plates of 293-derived human Bosc cells (16) were transiently transfected (Profection, Promega) with cDNAs encoding for the empty plasmid, wild-type EpoR, or mutant receptors in the bicistronic vector pMX-IRES-GFP as described (15). Cells were harvested 48 h after transfection for cross-linking experiments. To obtain BaF3 cell lines stably expressing the mutant receptors, retroviral supernatants were generated by co-transfection of the Bosc packaging cell line with receptor DNA and the ecotropic vector pCL eco (17). 5 ϫ 10 5 BaF3 cells were centrifuged for 120 min at 37°C and 900 ϫ g in the presence of 16 g of Polybrene and 750 l of virus supernatant. After 48 h the transfection efficiency was evaluated by analysis of GFP expression.
Cross-linking Studies on Live Cells-Cells were washed with phosphate-buffered saline and resuspended in 1 ml of phosphate-buffered saline containing 1 mM MgCl 2 and 0.1 mM CaCl 2 . If cysteines of interest were in the TM domain, extracellular cysteines were blocked by incubation with 100 M N-ethylmaleimide (NEM) for 30 min at room temperature. NEM was removed by spinning the cells and aspiration of the supernatant. Cross-linking was then performed for 10 min at room temperature with a final concentration of the cross-linker 1,2-dimaleimidylphenylene (o-PDM) dissolved in Me 2 SO of 100 M. Cells were lysed with 1ϫ Nonidet P-40 lysis buffer in the presence of 2% ␤-mercaptoethanol to quench excess cross-linker. Lysate samples were separated on 10% Tricine SDS-PAGE gels, transferred to nitrocellulose, and incubated with EpoR antibody directed against the receptor N terminus (UBI) at a dilution of 1:1000, respectively, followed by incubation with protein A-horseradish peroxidase (Amersham Biosciences) at a dilution of 1:10,000 and detection by enhanced chemiluminescence (Amersham Biosciences). Actin was detected by using a monoclonal actin antibody (Sigma) and incubation with a mouse-horseradish peroxidase antibody. For non-reducing gels, cells were lysed as described. Non-reducing Laemmli buffer was added to the lysates, and the proteins were separated by gel electrophoresis without prior boiling.
Assay for Growth Factor-independent Proliferation-Parental BaF3 cells and BaF3 cells expressing the wild-type EpoR, or the various constructs growing in IL-3 were washed extensively in RPMI medium, then placed in a 24-well plate with RPMI medium containing 10% fetal bovine serum without cytokines. Cell numbers were counted using a Coulter Counter after 1-10 days.
Dual Luciferase Assays-Functional STAT5 activation was assessed by measuring luciferase production of Bosc cells transfected with the pLHRE-Luc (lactogenic hormone responsive element), which contains six copies of the prolactin STAT5 binding site in the rat ␤-casein promoter (19). As an internal control the pRL-TK vector (Promega, Madison, WI) was used. Bosc cells were seeded in 96-well plates at 5 ϫ 10 5 cells/well and transfected with 50 ng of EpoR cDNA and 10 ng of the reporter cDNA using the Lipofectamine transfectant. After 12 h, cells were stimulated with 100 units/ml Epo. Four hours after transfection, the cells were lysed, and luciferase assays were performed using the dual luciferase reporter assay kit (Promega).
Peptide Synthesis and Purification-Two peptides corresponding to the Epo receptor transmembrane domain (Asp-Leu-Asp-Pro-Leu-Ile-Leu 7 -Thr-Leu 9 -Ser-Leu-Ile-Leu-Val-Leu-Ile-Ser-Leu-Leu-Leu-Thr-Val-Leu-Ala-Leu-Leu-Ser-His-Arg-Arg) were synthesized on an ABI 430A solid-phase peptide synthesizer (Applied Biosystems, Foster City, CA) using t-BOC-chemistry. [1-13 C]Leucine (Cambridge Isotope Laboratories, Andover, MA) was incorporated at either Leu 7 or Leu 9 of the TM peptide sequence. Leu 7 and Leu 9 correspond to Leu 228 and Leu 230 of the murine EpoR sequence, respectively. The N and C termini were protected. The crude peptide (5-15 mg) was purified by reverse-phase HPLC on a C4 column using gradient elution containing 0.1% triflu-oroacetic acid. The three-gradient elution starts with a largely aqueous solution of 70% distilled water, 12% acetonitrile, and 18% 2-propyl alcohol, and is changed to a more hydrophobic composition of 40% acetonitrile and 55% 2-propyl alcohol that elutes the peptides. The elution was monitored by the optical absorbance at 220 nm. The purity was confirmed by mass spectrometry.
Epo receptor transmembrane peptides were reconstituted into DMPC (dimyristoylphosphatidylcholine) bilayers at a 50:1 lipid-to-peptide ratio. Purified EpoR peptides were reconstituted by detergent dialysis by first dissolving DMPC, lyophilized peptide, and detergent (octyl-␤-glucoside) in trifluoroethanol. This mixture was incubated at 37°C for over 2 h, then trifluoroethanol was removed by evaporation using a stream of argon gas and then placing the sample under vacuum. The dry mixture was rehydrated with phosphate buffer (10 mM phosphate and 50 mM NaCl, pH 7), so that the final concentration of octyl-␤-glucoside was 5% (w/v). The rehydrated sample was stirred slowly for at least 6 h, and octyl-␤-glucoside was removed by dialysis using Spectra-Por dialysis tubing (3500 MW cutoff) for 24 h against phosphate buffer at 30°C. The membrane-reconstituted peptides adopted helical transmembrane orientations as assessed by polarized IR spectroscopy. The reconstituted membranes were then pelleted to form multilamellar dispersions and loaded into NMR rotors. Excess water was removed by spinning the sample in a tabletop rotor spinning unit.
Magic Angle Spinning NMR Spectroscopy-Solid-state MAS NMR measurements were performed on an Avance 360 MHz spectrometer using a 4-mm double resonance probe (Bruker Instruments, Billerica, MA). 13 C spectra were acquired with ramped amplitude cross-polarization (20) and TPPM proton decoupling (21). The 1 H pulse length was 3.6 s. The total contact time for cross polarization was 2 ms, and the recycle delay was 2 s. The proton decoupling field strength was 86 kHz. The 13 C frequencies were referenced to the carbonyl resonance of solid glycine at 176.02 ppm from TMS. The MAS frequency was 10 kHz Ϯ 3 Hz and the temperature was maintained at Ϫ30°C. The spectra represent the average of ϳ60,000 scans and were processed without line broadening. 13 C line shapes were analyzed using the deconvolution module of Xedplot (Bruker, Billerica, MA). The carbonyl peak was fit with three Gaussian line shapes, two accounting for the natural abundance 13 C and peptide background and one for the 13 C label in the peptide.
Molecular Dynamics Simulations-Molecular dynamics simulations were performed using the same approach and force field that we previously employed to successfully predict the conformation of a miniprotein (22). All simulations were unrestrained and initiated with a fully extended initial conformation built by the LEaP module of AMBER (version 7, Ref. 23). Solvation effects were incorporated using the Generalized Born model, as implemented in AMBER (24,25).

Cloning and Expression of Receptors with Single Cysteine
Mutations-Cysteine-scanning analysis is a valuable and established method for predicting the topology of membrane proteins and to obtain information on parameters such as distance and conformation of introduced cysteine residues (26 -28). This approach has been shown to be especially useful in the case of TM segments, where structural data are not easily obtained.
Introduction of cysteine residues at sequential positions allows characterization of interfaces, either by detection of spontaneously formed disulfide bonds or by chemical cross-linking. Formation of intermolecular disulfide bonds may lead to constitutively active receptors. An example is the EpoR R129C mutant, where replacement of an extracellular arginine with a cysteine leads to disulfide bond formation and constitutive activation (29,30).
We constructed an HA-tagged EpoR in the retroviral pMX-IRES-GFP vector, containing only the four conserved cysteine residues of the EpoR extracellular domain, that are indispensable for protein folding and ligand binding (Fig. 1a). These cysteines are thought to form two intramolecular disulfide bonds between residues Cys 28 and Cys 38 and Cys 66 and Cys 82 , respectively (30). A fifth cysteine residue in the extracellular domain, Cys 180 , was replaced by the related amino acid serine. In the intracellular sequence, Cys 289 was mutated to alanine and the cytoplasmic tail was truncated by introduction of a stop codon after Pro 355 . As a result, this truncated receptor, trEpoR4Cys, does not contain cysteines that should be available for sulfhydryl cross-linking, but includes Tyr 343 , that is known to be one of the two STAT5 binding sites (31,32). Into this thiol-free background, as well as into the wild-type EpoR, single amino acid substitutions to cysteine were introduced consecutively from Ala 220 to Leu 247 ( Fig. 1, a and b).
Transiently transfected human 293-derived Bosc cells were tested for expression of the receptor mutants. Immunoblotting of cell lysates with an EpoR antibody revealed that all receptors were expressed equally well with the expected size of 41 kDa for truncated receptors (Fig. 3b) and 68 kDa for full-length receptors, respectively, and were able to respond to Epo by activation of STAT5, as verified by luciferase assays (data not shown).
Cysteine Scanning of the Juxtamembrane Sequence and Identification of Constitutively Active Receptors-The mutant receptors were then examined for the induction of biologic effects in murine BaF3 cells. This proB cell line requires IL3 for survival and proliferation but can be rendered Epo-dependent upon transduction of the EpoR (33). Retroviral supernatants were prepared as described previously (3,16) for infection of parental BaF3 cells. Stable cell lines expressing the mutant receptors were established and tested for infection efficiency, i.e. GFP expression, and expression of the receptors at the cell surface by HA staining and subsequent FACS analysis. For all receptors the ratio between receptors expressed at the cell surface and the infection efficiency, as determined by GFP expression, was at levels comparable to the wild-type receptor (Fig. 2c).
BaF3 cells expressing the different EpoR cysteine point mutants were tested for their ability to grow in medium containing only 10% fetal serum. As a control, the mutant EpoR R129C was used, which is known to allow factor-independent cell growth (29,30,34). Of the 28 cysteine mutants introduced into the wild-type receptor, only L223C, L226C, and I227C were able to grow in the absence of IL3 or Epo. Fig. 2a shows that the kinetics of ligand-independent growth after selection of cells for 1 week without growth factor is the same for R129C-expressing cells and cells harboring a mutation at L223C, L226C, or I227C. The selection for constitutive activation can be detected as an enrichment for the percentage of cells that express GFP (Fig. 2b).
The constitutive activity of EpoR L223C, L226C, and L227C is likely caused by formation of intermolecular disulfide bonds that fix the EpoR in a conformation that is productive for signaling. To test this hypothesis we assayed their migration in non-reducing ( Fig. 3a) and reducing ( Fig. 3b) Tricine gels and by immunoblotting. As shown in Fig. 3a, strong and equivalent disulfide bonds were formed by A220C to D224C and L226C mutants. P225C, I227C, and L228C formed disulfide bonds with reduced efficiency. None of the other mutants from T229C to L247C formed dimers. Thus, there is no correlation between disulfide bond formation and activity, as the adopted conformation of inactive dimers may not be productive for signaling. This was also observed in studies involving the Neu (Erb-2) receptor, where only a subset of the receptors showing constitutive dimerization because of an engineered cysteine mutation had transforming activity (35).
In non-reducing Tricine gels, besides the 100-kDa complex which was detected for trL223C, trL226C, and trI227C, an additional 90-kDa double band was observed for all receptors including the negative control trEpoR4Cys. This double band corresponds to receptor dimers formed via the four extracellular cysteines in the unglycosylated and glycosylated forms of the receptor (see below and Fig. 4c). Indeed, under non-reducing conditions, these conserved cysteines are able to participate in the formation of higher oligomeric complexes of unknown stoichiometry (36). The 100-kDa complex detected for the trL223C, trL226C and trI227C mutants is a specific product of receptors whose TM helices are in close proximity (app. 3 Å). The differences in apparent molecular weight between the EpoR mutants and the wild-type receptor are presumably caused by differences in the shape of the cross-linked dimer and depend on the location of disulfide bond formation. In this regard, the trR129C mutant can be detected under nonreducing conditions as an ϳ95 kDa dimer (data not shown). Addition of ␤-mercaptoethanol to the samples reduced all oligomeric complexes to the monomeric forms that appear as one single band under the experimental conditions used (Fig. 3b).
Cross-linking of Receptors in Cells with a 9-Å Spacer-Truncated receptors containing single cysteine mutations in the background of trEpoR4Cys were transiently transfected into Bosc cells and subjected to cross-linking with the sulfhydrylspecific, membrane-permeable, homobifunctional cross-linker (1, 2)-dimaleimidylphenylene (o-PDM) on living cells (Fig. 4, a  and b). As the two maleimide groups of o-PDM are coupled to an aromatic ring, the distance spanned by the cross-linker does not vary much in solution, and allows the linkage of sulfhydryl groups separated by a distance of 9.4 Ϯ 0.5 Å (37). o-PDM has the shortest spacer of the chemical cross-linkers tested and resulted in a more distinct pattern of cross-linking than crosslinkers with a longer spacer arm (data not shown).
We assayed the sequence from Ala 220 to Leu 247 , which includes both the extracellular juxtamembrane region and the TM domain of the EpoR. Most secondary structure prediction algorithms predict that Pro 225 is the last amino acid in the juxtamembrane region so that the membrane-spanning ␣-helix stretches from Leu 226 to Leu 247 . Only the algorithm developed by Rost et al. (38) predicts that the juxtamembrane region has an extended structure through Pro 225 to Thr 229 and that the TM helix begins at Leu 230 (10).
To inhibit any reaction of the extracellular cysteines with the cross-linker, receptors harboring cysteine mutations in the transmembrane domain (L226C-L247C, Fig. 4a) were first incubated with the water-soluble reagent N-ethylmaleimide (NEM), that blocks free cysteines. All cells were then crosslinked with o-PDM, lysed with Nonidet P-40 buffer and an aliquot of the lysate was used for the analysis on reducing 10% Tricine gels (Fig. 4, a and b).
Immunoblotting with an EpoR antibody shows that mutants L226C-L230C cross-link without any periodicity (Fig. 4a), while Fig. 4b shows that indeed the entire region L220C to L230C cross-links without interruption, which is incompatible with the prediction that the TM domain extends through to Pro 225 . Besides their monomer and dimer bands, receptors that were not treated with NEM reveal an additional o-PDM crosslinked double band of 90 kDa (Fig. 4b). Again, this double band represents cross-linking via the four conserved cysteines in the extracellular part of the receptor (see above non-reducing gels,   3a) and exhibits a different mobility from the band that results from cross-linking of receptor mutants containing TM cysteine residues (Fig. 4c). As a loading control the blot was stripped and re-probed with an actin-antibody. Addition of NEM (like in Fig. 4a) precluded detection of o-PDM mediated cross-linking of residues 220C-225C (not shown) and decreased but did not abolish cross-linking of L226C to L228C (compare Fig. 4, a and b for L226C), suggesting that these residues are at the junction between the JM and TM domains as generally predicted, and still partially accessible to NEM. Because of this partial inhibition of the subsequent cross-linking reaction with o-PDM it is not possible to quantify the extent of dimer formation from this blot. To be able to cross-link extracellular cysteines, NEM was omitted in the experiment shown in Fig. 4b, and all residues reacted with the chemical in a comparable manner and no preferred dimerization of distinct residues is detectable.
From our cross-linking results we conclude that the first five amino acids of the predicted TM domain (Leu 226 -Leu 230 ) are not ␣-helical and rigid. In addition to residues belonging to the non-helical motif (Leu 224 -Leu 230 ), only residues L232C, L241C, L244C, and to a lesser extent A245C, and very weakly L246C could be cross-linked with o-PMD (Fig. 4a). This result was obtained both when cells were lysed in buffer containing detergent (1% Nonidet P-40), and also for lysates that were obtained by directly boiling cells in reducing Laemmli buffer (not shown). As a result, cross-linking must have occurred in native membranes and was not an artifact of detergent solubilization.
In summary, we show that the same EpoR mutants (L223C, L226C, and I227C) that exhibit constitutive activation also cross-link with o-PDM. In addition, residues 220 -222, 224, 225, and 228C-230C cross-link with o-PDM, but do not exhibit constitutive activation. It is interesting that none of the other cross-linked cysteine mutants at the C-terminal end (e.g. L241C, L244C, A245C) of the TM domain are able to grow in the absence of ligand, although cells expressing the EpoR L244C mutant are significantly protected from apoptosis (not shown) upon withdrawal of cytokines. Dimerization at the level of the TM domain may for example not allow sufficient flexibility of the cytosolic domains, which are required for assembly of signaling complexes. This was reported for the epidermal growth factor receptor, where activation involves a rotation of the TM domains due to the flexibility of the extracellular juxtamembrane region (39). For the EpoR, previous reports (11), as well as our cross-linking studies, pinpointed to Leu 241 and Leu 244 as a critical region for receptor dimerization. Replacement of Leu 241 by charged residues (40) or introduction of a glycine-proline mutation (13) at Leu 240 and Leu 241 negatively affects receptor function.
Helix Capping at Thr 229 -Leu 230 -Our disulfide cross-linking results indicate that the first five residues of the predicted TM domain do not form an ␣-helix. In agreement with this observation, the amino acid sequence at the N terminus of the EpoR TM domain (i.e. 228 LTLSLIL 234 ) is consistent with the hp-XpXhX sequence for N-helix caps as described by Aurora and Rose (41) where h and p represent hydrophobic and polar amino acids, respectively. In this sequence, Leu 230 is the N-cap amino acid, i.e. the first amino acid in the helix whose carbonyl is fully hydrogen bonded to an iϩ4 backbone N-H.
Molecular dynamics simulations were performed on a peptide corresponding to the TM and juxtamembrane regions of the EpoR. At 300 and 325 K, simulations initiated from com- FIG. 3. Expression of receptor mutants. a, detection of EpoR dimers under non-reducing conditions. Bosc cells were transfected with DNA coding for the empty vector, trEpoR4Cys or the various single cysteine mutant receptors (trA220C-trL247C), lysed, and run on a 10% Tricine gel before immunoblotting with an EpoR antibody. Arrows indicate the size of the monomer and the dimeric proteins. Formation of unspecific intermolecular disulfide bonds to the four extracellular cysteines are marked by a bracket and can be distinguished from transmembrane domain disulfide bond formation by a difference in size (90 versus 100 kDa, respectively). b, for control of expression levels by Western blotting, lysates were prepared as described and run on a 10% Tricine gel under reducing conditions before immunoblotting with an EpoR antibody to detect EpoR monomers.
pletely extended conformations folded to partially helical structures with Leu 230 as a helical cap. Fig. 5a illustrates the representative structure obtained from cluster analysis of the simulation at 300 K. As Aurora and Rose (41) point out, helix capping typically involves both hydrophobic and hydrogen bonding interactions. This is seen in the hydrophobic contact that Leu 228 has with Leu 232 and Ile 233 , and hydrogen bonding of the Thr 229 side chain back to the amide NH of Leu 232 . This hydrogen bond is present in Ͼ60% of the structures sampled during the simulation (Fig. 5a). In Fig. 5b, the average fractional helical content for each residue in the sequence is shown for simulation at 300 K. The N-terminal region shows individual residues that sample helical conformations, but no stretches of more than two helical residues are evident. In contrast, all residues after Leu 230 have 50 -80% average helical content, with fraying of the helix evident at the C-terminal residues. This pattern is similar at 325 K (data not shown), with the transition between unstructured and helical regions again occurring at Leu 230 .
Additional support for the N-cap structure comes from solidstate magic angle spinning (MAS) NMR measurements of peptides corresponding to the TM and juxtamembrane regions of the EpoR. Specifically, we compared the NMR spectra of peptides 13 C-labeled at the backbone carbonyl of Leu 228 or Leu 230 . The 13 C chemical shift of the backbone carbonyl is sensitive to hydrogen bonding and the conformation around the carbon-yl-C␣ bond (42,43). Amino acids in helical secondary structure exhibit 13 CϭO chemical shifts in the range of 175-176 ppm, while extended ␤-strand and ␤-sheet secondary structure have chemical shifts below ϳ172 ppm. Leu 230 is the last amino acid in helical secondary structure in the N-cap structure shown in Fig. 5c.
The MAS NMR spectrum of the EpoR TM peptides 13 Clabeled at Leu 230 is shown in Fig. 5d (top). Only the carbonyl region of the spectrum is shown, along with a deconvolution into Gaussian components. The experimental spectrum (black) exhibits two resonances. The resonance at 172 ppm corresponds predominantly to the natural abundance 13 C signal from the lipid carbonyls. The natural abundance 13 C background was independently fit in MAS spectra of lipid and unlabeled peptide to two Gaussian components (blue). The resonance at 175.3 ppm corresponds to the [1-13 C]carbonyl carbon of Leu 230 . The deconvoluted [1-13 C]carbonyl resonance is shown in red. The chemical shift is characteristic of ␣-helical secondary structure. The NMR spectrum of L228 (Fig. 5d,  bottom) on the other hand is dramatically different, as no sharp resonance corresponding to an ␣-helix is observed. Using the same 13 C natural abundance background for the fit of the line shape, the [1-13 C]carbonyl resonance (red) is instead broadened and shifted to ϳ172 ppm, consistent with the loss of defined helical secondary structure.

Replacement of the Helix Cap by an Oligoleucine
Segment-To assess the importance of the helix cap motif for the activation and signaling of the EpoR, we replaced the first six or nine amino acids of the TM domain by leucines (Leu 6 , Leu 9 ). This was accomplished by mutating to leucine residues Ile 227 , Thr 229 , Ser 231 , and Ile 233 as the other four residues of the cap structure are originally leucine (see Fig. 6a). Structural and functional assays had shown that oligoleucine motifs are helical and capable of efficient dimerization within an artificial transmembrane because of the packing of ␣-helices in a leucine zipper type of interaction (44). We hypothesized that replacement of the helix cap by leucines would elongate the helix up to Leu 226 and result in a different mode of activation because of a different pitch of the downstream helix and possibly enforced self-assembly. Thus we infected BaF3 cells with viruses for either the wild-type receptor, Leu 6 or Leu 9 and the R129C mutant as a control. Cells were selected for their ability to grow in the absence of ligand or in a dose-dependent response to Epo. No difference could be seen for Leu 6 or Leu 9 in the response to increasing amounts of Epo after 48 h of growth, when compared with the wild-type receptor (data not shown). Nevertheless, during selection of cells in 10% fetal serum for 10 days, replacement of the first six residues (Leu 226 -Leu 230 ) resulted in a significant number of cells that survived longer than cells expressing wild-type EpoR, but proliferated poorly and eventually died, so that no difference in growth could be detected by counting. On the other hand, cells expressing the receptor, where the proximal region of the predicted TM domain was replaced by an ␣-helical structure composed of nine leucines (Leu 226 -Leu 234 ), i.e. the complete helix cap motif, the receptor showed ligand-independent activation (Fig. 6b). The sole difference between Leu 6 and Leu 9 is residue Ile 233 , which is an isoleucine in Leu 6 and leucine in Leu 9 . Molecular dynamics simulations indicate that Ile 233 forms hydrophobic interactions with the cap residue Leu 228 . These results suggest that the ␤-branched isoleucine residue plays an important role in stabilizing the cap structure and the unliganded inactive dimeric conformation of the receptor.
In summary these data give convincing evidence that the first nine amino acids of the EpoR TM domain are involved in the formation of a helix cap, and that this motif allows the EpoR to be activated in a ligand-dependent manner, whereas the TM helices presumably stabilize the dimeric interface of the active receptor. Deletion of the cap motif by extension of the TM ␣-helix leads to constitutive activation and uncontrolled cell proliferation.
Amino Acid Insertions into the Juxtamembrane Region of the EpoR-To test whether the structure near Pro 225 is flexible we inserted one glycine residue between Asp 224 and Pro 225 , either in the context of the wild-type receptor (EpoR-GP) or the constitutively active receptor mutant R129C (R129C-GP). The combination of these two residues should only be tolerated in the context of a flexible structure. Introduction of a GP mutation in the TM ␣-helix disrupts its correct structure and de-FIG. 5. Leu 228 -Leu 232 form a helix cap on the N terminus of the transmembrane domain of the EpoR. a, histogram (bars) and cumulative histogram (line) for the distance between Thr 229 OH to Leu 232 HN. The large peak under 3 Å (ϳ60% of the structures) represents the helix-capping hydrogen bond. b, fraction of structures with ␣-helical conformation of each residue, averaged over MD simulation at 325 K. The segment from Leu 230 to the C terminus shows significant helical content. In contrast, the N-terminal segment shows little tendency to forms ␣-helix. c, structure of the junction of the TM and juxtamembrane regions of the EpoR from molecular dynamics simulations. The simulations were undertaken on the full TM and juxtamembrane sequence from Pro 239 to Leu 272 , which includes the region shown in Fig. 1 (numbering from the mature, wild-type murine receptor is used). d, MAS NMR spectroscopy of peptides corresponding to the TM-juxtamembrane region of the Epo receptor containing [1-13 C]Leu 230 (top) and [1-13 C]Leu 228 (bottom). The experimental spectra (black) are deconvoluted into the natural abundance 13 C background from lipid and peptide (blue) and the specific [1-13 C]Leu resonance from the peptide (red). The chemical shift of the 1-13 C resonance of Leu 230 at 175 ppm is characteristic of ␣-helical secondary structure. This resonance shifts to lower frequency and becomes broader which is characteristic of extended ␤-structure.
creases tyrosine phosphorylation of JAK2 and the receptor. In the background of the R129C mutant, this mutation abrogates ligand-independent signaling (13).
To investigate the receptors' capability to activate the JAK-STAT signaling pathway, we transfected these receptors into Bosc cells. Cells were stimulated with 100 units/ml Epo, and the transcriptional activity was assessed by quantification of luciferase activity. As shown in Fig. 7 there was no major difference induced in the wild-type EpoR or the R129C mutant by this insertion. EpoR and EpoR-GP show comparable STAT5 reporter activity. Both R129C and R129C-GP show strong activation of STAT5 in the absence of ligand that can be further increased by addition of Epo. We infected BaF3 cells with virus coding for wild-type receptor EpoR, the constitutive receptor R129C, and the GP mutants EpoR-GP and R129C-GP and investigated their proliferation in the absence of Epo or with various Epo concentrations. No major difference was detectable in the ability to proliferate in the presence of Epo, and cells expressing R129C-GP could still grow in the absence of Epo (not shown). This result supports the hypothesis that the extracellular juxtamembrane region is not ␣-helical, but must be part of a rather flexible structure.
Replacement of the EpoR Extracellular Domain by the Coiled-coil (cc) of Put3-Recently we have investigated the ability of distinct dimeric conformations of the EpoR to signal. The extracellular domain of the EpoR was replaced by the Put3 dimeric cc domain, which has an amphipathic structure stabilized by hydrogen bonding and hydrophobic interactions (45,46). It imposes its coiled-coil geometry on a downstream TM helix whose dimer interface can then be predicted. Because the register of the coiled coil is known, fusion with downstream helices with different junction points will impose all seven relative orientations to the EpoR TM. Two of these fusion proteins (cc-EpoR-III and cc-EpoR-VI) define an active dimeric interface and mimic signaling by Epo-activated EpoR (11). Residue L244 is predicted to be in the interface of both active constructs, cc-EpoR-III and cc-EpoR-VI, (an a or d position of the coiled-coil a-g heptad repeat) and in cc-EpoR-III, Leu 241 and Leu 244 are predicted to be in the interface (11). Since both L241C and L244C appear as cross-linked bands in the present study (Fig. 4a), we suggest that this is representative of the activated EpoR dimeric interface.
Thus we used the cc-EpoR fusion protein system to ask whether cysteine-mediated cross-linking could be used to show that Leu 244 is in the interface of only the active cc-EpoR-III and cc-EpoR-VI fusion proteins and not of any of the others, cc-EpoR-0, I, II, IV, and V. Second, we tested whether the Nterminal five amino acid residues Leu 226 -Leu 230 maintain their non-helical structure found in wild-type EpoR, or adopt a helical structure because of the upstream coiled-coil. We constructed fusion proteins that resemble the seven described cc-EpoR constructs (cc-EpoR-0-VI), but have residue Leu 244 of the TM domain mutated to cysteine (L244C) and which contain the truncated thiol-free cytosolic domain as in cysteine-dependent oligomerization studies (Fig. 1a). Because of its close juxtaposition in the dimeric interface the presence of the L244C mutation specifically induced cross-linker-independent disulfide bond formation and dimerization in the wild-type EpoR and presents therefore a very sensitive assay to test the distance of receptor subunits in the dimeric interface (data not shown).
The receptors were transiently transfected into Bosc cells, lysed, and analyzed on 16.5% Tricine gels. All fusion proteins were expressed with the expected size of 18 kDa. Consistent with our hypothesis that the two active fusion proteins cc-EpoR-III and cc-EpoR-VI should have L244C in the dimeric interface, we were able to detect formation of disulfide-bonded dimers in the absence of cross-linker for these two receptors (Fig. 8a). As non-reducing conditions lead to unspecific dimerization due to the coiled-coil domain, reducing conditions were used without prior boiling that allowed to detect the stronger interaction between the disulfide-bonded cysteines at residues 244C. The agreement between the results obtained with engineered fusion proteins and those obtained with cysteine mutants of the wild-type EpoR argue strongly that Leu 244 is in the interface of the activated EpoR dimer and that the cc-EpoR fusion proteins indeed induce the predicted interface of the downstream helices.
To answer the second question, we mutated each of the five residues, Leu 226 , Ile 227 , Leu 228 , Thr 229 , and Leu 230 to cysteine and compared their o-PDM-induced cross-linking in the context of the wild-type EpoR and in the context of the Put3-cc-EpoR fusion protein cc-EpoR-0, which contains all TM residues. All residues in this motif cross-link with o-PDM in the context of the wild-type EpoR (Fig. 4a), but only one residue (L226C) cross-links in the context of the coiled-coil fusion protein cc-EpoR-0 (Fig. 8b). This is typical of an extended structure for the Leu 226 -Leu 230 region in the wild-type EpoR, but the latter results are characteristic for an ␣-helical structure of the same region in cc-EpoR-0 context. Because the last residue of Put3 is in a b position, only one interface (a or d) residue can be predicted to exist among the following five residues in the cc-EpoR-0 context. Indeed, Fig. 8b shows that only one residue, L226C, is able to cross-link. Based on prediction, the best candidate for cross-linking would have been residue 227. However, it is possible that introduction of a cysteine rotates the ␣-helix when it is fused to the coiled-coil so that cross-linking occurs at the preceding position, or that the intrinsic tendency to form an extended structure for the sequence Leu 226 -Leu 230 distorts the helix and brings Leu 226 in the interface. Nevertheless, the patterns of cross-linking are strikingly different in the wild-type EpoR and Put3 proteins for the same sequence. This is reminiscent of studies, which showed that identical sequences adopt different structures as a function of context (47). The rotation/distortion effect of Leu 226 -Leu 230 sequence is rel-evant only for cc-EpoR-0, since the other cc-EpoR fusion proteins contain N-terminal truncations of the EpoR TM domain, such as the active fusion proteins cc-EpoR-III and cc-EpoR-VI, which lack the first three or six TM domain residues, respectively. In addition, this effect may be responsible for the unusual properties of cc-EpoR-0, which induces cell survival and mitogen-activated protein kinase activation but fails to induce cell proliferation and STAT activation (11). DISCUSSION Our main observations are that introduction of cysteine residues in the EpoR juxtamembrane region (L223C), and transmembrane region (L226C, I227C) results in constitutive activation of the receptor and that the junction between the JM and TM domains is not helical, forms a helix cap instead and that its precise sequence is important for proper receptor function.
The approach typically taken for high resolution structural studies of cytokine receptors and receptor tyrosine kinases is to independently determine the structures of their extracellular, transmembrane and intracellular domains. The juxtamembrane regions flanking the TM domain are generally not tar- FIG. 7. Mutational analysis of the juxtamembrane region connecting extracellular and transmembrane domain. Quantification of STAT5 transcriptional activity by the luciferase assay. Bosc cells were transfected with receptor DNA, JAK2, STAT5, pLHRE-Luc, and pRL-TK in order to allow normalization of luciferase activity. Cells were stimulated for 4 h with 100 units/ml Epo, lysed, and assayed for luciferase activity. Results were obtained from one representative experiment performed in triplicates Ϯ S.D.
FIG. 8. Replacement of the extracellular domain by the Put3 coiled-coil. a, Bosc cells were transiently transfected with DNA coding for chimeric receptors where the tr244C receptor extracellular domain and hinge region was replaced by the Put3-EpoR deletion mutants cc-EpoR-0 to cc-EpoR-VI. Lysates were analyzed for their dimerizing capacity on 16.5% Tricine gels by subsequent immunoblotting with an HA antibody. b, Bosc cells were transiently transfected with DNA coding for chimeric receptors where the Put3-EpoR mutant cc-EpoR-0 was fused to the cytoplasmic domains of tr226C-tr230C. Lysates were analyzed on 16.5% Tricine gels for their dimerizing capacity by cross-linking with o-PDM and immunoblotting with an HA antibody.
geted, yet are of intense interest because they couple ligand binding within the extracellular domain to conformational changes in the transmembrane and cytoplasmic domains.
Here, we show that the sequence linking the extracellular domain of the EpoR to the TM domain is extended and that the first amino acid residues of the predicted TM domain form a helix cap. Moreover, we describe three EpoR cysteine point mutants in the extracellular juxtamembrane region, L223C, L226C, and I227C, which induce constitutive activity by formation of disulfide-linked dimers. We show that insertion of glycine before Pro 225 is well tolerated and that replacement of the first nine residues of the EpoR TM domain with consecutive leucines leads to constitutive receptor activation. These observations are in agreement with proposed activation mechanisms based solely on close proximity of EpoR monomers, and would be consistent with the ability to form active EpoR dimers in a number of very different ways, i.e. with mimetic antibodies (48), the R129C mutation (29), agonist peptides (49) and coexpression of the gp55 envelope protein of the Spleen Focus Forming Virus (50 -52).
However, there is increasing evidence that specific orientations of EpoR monomers in the dimer complex are required for activation. For example, peptide antagonists have been described that hold the receptor extracellular domains in an inactive orientation (8). Moreover, insertion of a GP motif in the TM domain is detrimental for receptor activation. These results indicate that the TM domain must have a specific orientation for signaling.
The extended JM structure and the ability to cross-link the sequence from positions 220 to 230 suggest that the JM region has some flexibility. Intermolecular contacts between JM residues may be responsible for stabilizing inactive or active conformations in dimers of the wild-type receptor. The extracellular juxtamembrane sequences may be accessible to small molecule drugs that would target this region and activate receptors via different sites than those utilized by ligands.
By replacing the extracellular domain of the EpoR with the coiled-coil dimerization domain of the yeast transcription factor Put3, we previously engineered EpoR dimers with active and inactive transmembrane orientations. These studies showed that Leu 241 and Leu 244 , which are in the transmembrane interface of the active coiled-coil EpoR dimers are also crosslinked when replaced by cysteine residues.
Using o-PDM, not only residues 241 and 244, but also to a lesser extent residue 245 can also be cross-linked, although this amino acid cannot be in the same interface of a symmetric dimer. Molecular dynamics simulations suggest that this residue is the last helical residue before fraying of the TM helix (Fig. 5b) and might thus be more flexible. In our proposed model of the inactive dimer interface (11), the TM helices in the inactive dimer are rotated 100°counterclockwise compared with the active dimer. This inactive dimer conformation has Leu 245 in the interface. Consequently, our cross-linking profile may reflect a combination of inactive and active dimer conformations or an asymmetric dimer interface for the activated receptor (6). Ongoing experiments are attempting to distinguish between these possibilities.
In summary our data on the structure and function of the EpoR JM and TM domains support the concept that precise TM sequences are necessary for proper receptor activation and demonstrate that close apposition between residues bordering the TM domain is possible in the context of the cysteine containing mutants, leading to disulfide-bonded EpoR-activated dimers.