RESEARCH

◥ control the relative orientation of the ECDs in RESEARCH ARTICLE SUMMARY the dimeric complex. The “angle” series varied the scissor angle between the two ECDs, whereas the “distance” series varied their relative proxim- ENGINEERING ity. The designed DARPin ligands were validated by means of x-ray crystallography for representa- tive complexes. The systematic variation of angular Topological control of and distance parameters generated a range of full, biased, and partial agonism of EpoR signaling signaling induces differential ◥ in the human erythroid cell ON OUR WEBSITE line UT7/EPO, as shown with flow-cytometry and effects in hematopoiesis Read the full article at http://dx.doi. immunoblotting for phos- Kritika Mohan*, George Ueda*, Ah Ram Kim†, Kevin M. Jude†, Jorge A. Fallas†, org/10.1126/ phorylated downstream science.aav7532 effectors. In general, in- Yu Guo†, Maximillian Hafer, Yi Miao, Robert A. Saxton, Jacob Piehler, ...... Vijay G. Sankaran, David Baker‡, K. Christopher Garcia‡ creasing the angle or dis- tance between the receptor ECDs resulted in a progressive partial agonism, as measured with INTRODUCTION: Receptor dimerization is a system, a well-characterized dimeric cytokine changes in maximum response achieved (Emax) fundamental mechanism by which most cyto- receptor system. and median effective concentration (EC50). Biased kines and growth factors activate Type-I trans- signal transducer and activator of transcription Downloaded from membrane receptors. Although previous studies RESULTS: We used the DARPin (designed (STAT) activation was elicited by some of the sur- have shown that ligand-induced topological ankyrin repeat protein) scaffold because of rogate DARPin ligands. We also evaluated the changes in the extracellular domains effects of these DARPin agonists on dif- (ECDs) of dimeric receptors can affect ferentiation and proliferation of hemato- signaling output, the physiological poietic stem and progenitor cells (HSPCs) relevance is not well understood. This that were maturing into the erythroid http://science.sciencemag.org/ is a difficult problem to study because lineage. The partial agonists displayed an engineered ligand system does not stage-selective effects on HSPCs, whereas exist that would enable a systematic the biased agonists more selectively pro- exploration of the relationship between moted signaling at either the early or late ligand-receptor dimer geometry and stages of differentiation. signaling. This question, as well as the capability of exploiting these structure- CONCLUSION: We have designed a activity relationships for drug discovery, series of surrogate ligands to system- are important because most atically alter receptor dimer topology and modulate signaling outputs. An exert pleiotropic effects that limit their on May 28, 2019 therapeutic utility. New approaches are important feature of these ligands is needed to modulate cytokine signaling that the dimerized modules are rigidly and identify clinically efficacious var- connected, thus allowing us to deter- iants. By contrast, small-molecule ligands mine structures of the ligand-receptor for –coupled receptors have complexes, or model them accurately, been successfully discovered through and correlate precise geometries with medicinal chemistry and used to in- signaling output. Some variants induce duce conformational changes that lead stage-selective differential signaling in to physiologically relevant biased primary cells. This “topological tuning” signaling outputs. A comparable may be attributed to altered intracellular approach for dimeric receptors could Three different EpoR/DARPin complex topologies are shown orientations of 2 (JAK2) open a path to new pharmacological that elicit full, partial, and null agonism. Full agonism, red; relative to its substrates or mechanical parameters for cytokines and growth partial, green; and null, blue. The EpoR/DARPin structure shown distortion that leads to changes in com- factors. on the green cell is from a crystal structure, whereas the plex stability and receptor internalization. complexes shown on the red and blue cells are models. Although our experiments do not reveal a RATIONALE: In order to better un- predictive framework to relate topology to derstand how the extracellular structure its modular nature. We isolated a high-affinity signaling, this approach can be used to empirically of cytokine-receptor complexes affects down- DARPin to EpoR using yeast display and in vitro identify specific dimer topologies that correlate stream signaling events, we developed an evolution and determined the crystal structure with therapeutically desirable signaling outputs engineered ligand system to precisely con- of the DARPin/EpoR complex. We then con- for any dimeric receptor system.▪ trol the orientation and proximity of dimeric verted these monomeric DARPin binding mod- receptor complexes that would enable the ules into C2 symmetric homodimeric agonists The list of author affiliations is available in the full article online. measurement of structure-activity relation- by incorporating in silico designed dimerization *These authors contributed equally to this work. ships between receptor dimer geometry, interfaces. This rigidly connected dimeric DARPin †These authors contributed equally to this work. signaling, and function. We applied this scaffold then enabled us to design a series of ‡Corresponding author. Email: [email protected] (D.B.); [email protected] (K.C.G.) approach to design geometrically controlled extended ligands through sequential insertion Cite this article as K. Mohan et al., Science 364, eaav7532 “ ” ligands to the receptor (EpoR) of ankyrin repeat spacers to systematically (2019). DOI: 10.1126/science.aav7532 IMAGE: K. C. GARCIA AND E. SMITH

Mohan et al., Science 364, 750 (2019) 24 May 2019 1of1 RESEARCH

◥ chemistry (21). Finding small-molecule drugs is RESEARCH ARTICLE not currently feasible for Type-I transmembrane receptors that bind to protein ligands, such as cytokines, through large extracellular domains PROTEIN ENGINEERING (ECDs). Nevertheless, ligand engineering strat- egies that take advantage of the potential for biased signaling behavior by dimeric cytokine Topological control of cytokine receptors could be used as probes to better un- derstand cytokine receptor signaling biology but also may have therapeutic potential. receptor signaling induces differential The overall geometry of the cytokine receptor dimer can influence the strength of the down- effects in hematopoiesis stream signal (6, 22–24); however, approaches do not exist to precisely control the orientation and Kritika Mohan1*, George Ueda2,3*, Ah Ram Kim4,5,6†, Kevin M. Jude7†, proximity of ligand-receptor complexes. The ability Jorge A. Fallas2,3†, Yu Guo8†, Maximillian Hafer9, Yi Miao1, Robert A. Saxton1,7, to topologically control receptor dimerization Jacob Piehler9,10, Vijay G. Sankaran4,5,6, David Baker2,3,11‡, K. Christopher Garcia1,7,12‡ could open access to a new pharmacologic metric for assessing structure-activity relationships Although tunable signaling by G protein–coupled receptors can be exploited through medicinal of cytokine and growth factor receptor signal- chemistry, a comparable pharmacological approach has been lacking for the modulation of ing. Previous studies have shown that intact signaling through dimeric receptors, such as those for cytokines. We present a strategy to (25, 26) and diabodies can act as sur- modulate cytokine receptor signaling output by use of a series of designed C2-symmetric rogate cytokine agonists capable of modulating Downloaded from cytokine mimetics, based on the designed ankyrin repeat protein (DARPin) scaffold, that can signaling outputs through receptor dimerization systematically control (EpoR) dimerization orientation and distance (24, 27), but their segmental flexibility prevents between monomers. We sampled a range of EpoR geometries by varying intermonomer angle control of the precise geometry or distance be- and distance, corroborated by several ligand-EpoR complex crystal structures. Across the range, tween monomers. we observed full, partial, and biased agonism as well as stage-selective effects on We integrated a previous protein-binding

hematopoiesis. This surrogate ligand strategy opens access to pharmacological modulation of scaffold technology, termed DARPin (designed http://science.sciencemag.org/ therapeutically important cytokine and growth factor receptor systems. ankyrin repeat protein) (28), with protein design and engineering approaches to create a self- assembling, rigidly connected dimeric scaffold ytokines are secreted that medi- propagation of a phosphorylation cascade lead- that enables precise control of the relative orien- ateawiderangeoffunctions,principallyin ing to signal transducer and activator of tran- tations of the receptor ECDs in a dimeric complex the , but they also exert scription (STAT) activation and induction of (Fig. 1A). We focused on the erythropoietin recep- C many other homeostatic functions that expression programs (3, 5). Structural and func- tor (EpoR), which plays a critical role in erythroid affect mammalian physiology (1–3). As one tional studies of cytokine-receptor complexes lineage commitment and differentiation to pro- example, hematopoiesis, the process by which together with similar studies on growth factor duce mature red blood cells (RBCs) (20). We found blood cells are formed, is influenced at both early dimeric receptor systems, including receptor that systematic variation of angular and distance

and late stages by a milieu of different cytokines tyrosine and serine-threonine kinases, show parameters of the DARPin-based agonists leads to on May 28, 2019 that engage and activate specific cell-surface re- that a wide range of receptor dimer geometries partial and biased agonism of the EpoR signal. Fur- ceptors expressed on different lineages of cells are compatible with signaling (3, 5–9). thermore, this variation in geometry has allowed (4). When cytokines bind to their receptors, re- Given their powerful actions, many natural us to selectively activate EpoR signaling at specific ceptor dimerization triggers transphosphorylation cytokines have been investigated as therapeutic stages of hematopoiesis. Collectively, these studies of Janus (JAK)/TYK kinases, and subsequent agents. Whereas select cytokines are clinically offer a general strategy for the regulation of cyto- safe and effective, such as erythropoietin (EPO) kine receptor signaling through topological con- (10), most natural cytokines, especially immune trol of the receptor dimer that can be applied to 1Department of Molecular and Cellular Physiology, Stanford cytokines, have pleiotropic effects on many dif- other dimeric receptor systems. University School of Medicine, Stanford, CA 94305, USA. ferent cells and stimulate multiple downstream 2Department of Biochemistry, University of Washington, Results 3 signaling responses, resulting in toxicity or lack Seattle, WA 98195, USA. Institute for Protein Design, Selection of monomeric DARPins University of Washington, Seattle, WA 98195, USA. 4Division of efficacy (11, 12). Nevertheless, cytokines have of Hematology/Oncology, The Manton Center for Orphan potential as therapeutics if their actions can be selective for EpoR Disease Research, Boston Children’s Hospital, Harvard harnessed, as exemplified by recent advances Our overall strategy was to engineer a protein Medical School, Boston, MA 02115, USA. 5Department of Pediatric Oncology, Dana-Farber Cancer Institute, Harvard for cytokines such as interleukin-2 (IL-2) and module that binds with high affinity to a cyto- Medical School, Boston, MA 02115, USA. 6Broad Institute of IL-10 (13, 14). Recent studies, using natural and kine receptor ECD and subsequently design the MIT and Harvard, Cambridge, MA 02142, USA. 7Howard engineered ligands to both cytokine and recep- module to form a range of dimeric geometries Hughes Medical Institute, Stanford University School of tor (RTK) class receptors, show that would in turn constrain the dimerization Medicine, Stanford, CA 94305, USA. 8State Key Laboratory of Medicinal Chemical Biology and College of Pharmacy, that signal strength can be modulated down- geometry of the bound receptor (Fig. 1A). We Nankai University, Tianjin, People’s Republic of China. stream of dimeric receptors (15–18). Further- focused on a well-characterized system, human 9Division of Biophysics, Department of Biology, University of more, select natural and synthetic mutants of EpoR, as our cytokine receptor target for topo- 10 Osnabrück, 49076 Osnabrück, Germany. Center for Cellular cytokines such as IL-2, human growth hormone logical tuning. We chose DARPins as the bind- Nanoanalytics, University of Osnabrück, 49076 Osnabrück, Germany. 11Howard Hughes Medical Institute, University of (hGH), (SCF), and EPO that ex- ing scaffold because they are modular, which Washington, Seattle, WA 98195, USA. 12Department of hibit partial and biased signaling properties have enables size variation (28), and have been used to Structural Biology, Stanford University School of Medicine, been identified (17, 19, 20). In G protein–coupled design self-assembling oligomers (29). A DARPin Stanford, CA 94305, USA. receptors (GPCRs), biased signaling is physiolog- molecule typically consists of “X” number of *These authors contributed equally to this work. †These authors contributed equally to this work. ically relevant and generally depends on binding ankyrin repeats flanked by N- and C-terminal “ ” ‡Corresponding author. Email: [email protected] (D.B.); of small-molecule ligands, making this exploitable capping regions, resulting in NXC DARPins. [email protected] (K.C.G.) as a drug discovery strategy by using medicinal Each ankyrin repeat is formed by 33 amino

Mohan et al., Science 364, eaav7532 (2019) 24 May 2019 1of15 RESEARCH | RESEARCH ARTICLE

Fig. 1. Engineering and characterization of high-affinity DARPin binding to EpoR. (A) An overview of our topological control strategy, using EpoR as a model system to investigate modulation of signaling responses. (B) Schematic representation of the yeast-displayed N3C DARPin construct used for library generation. The randomized positions are shown in blue. (C)Histograms of yeast-displayed naïve N3C library, pool of DARPin clones after in vitro evolution, and pool of shuffled library clones after in vitro evolution binding to EpoR. MFI, mean fluorescence intensity. (D) Representative surface plasmon resonance sensograms for pre- and post-shuffled library

N3C DARPin clones binding to EpoR. Kd, dissociation constant. (E) Structure of N3C E2 binding to EpoR. (F) The structure of E2/EpoR (PDB 6MOE) is aligned with

EPO/EpoR to indicate relative binding sites Downloaded from of the two ligands. EPO is depicted as green cylinders. (G) The intimate shape complementarity of E2 binding to EpoR is depicted with the side chains shown for key interacting E2 residues. http://science.sciencemag.org/

on May 28, 2019 acids and consists of a b turn followed by two ometry, at a site different but overlapping with merizedEpoRECDs(Fig.2Bandfig.S5).The antiparallel a helices. Twenty-seven of these po- the EPO binding site, using three ankyrin re- formation of the C_R3 DARPin dimer and a sitions are highly conserved in naturally occur- peat domains. The b turn loop regions of the 2:2 C_R3/EpoR complex was confirmed with ring ankyrin repeats and crucial to maintaining ankyrin domains engage the top of the D1 do- size exclusion chromatography, and crystal struc- the repeat protein framework (30, 31), whereas main, whereas the helical regions pack against tures were determined to 1.2 and 3.2 Å resolu- the remaining six positions can support random- thesideoftheD1domainwithhighshape tion, respectively. We assessed the ability of the ization for library generation (Fig. 1B, blue resi- complementarity and extensive buried sur- DARPin dimers to dimerize EpoR on the surface dues). We developed two yeast-displayed DARPin face area. of live cells using single-molecule total internal libraries,N2CandN3C(Fig.1Bandfig.S1),which fluorescence (TIRF) microscopy (Fig. 2C and fig. were screened against EpoR ECD, resulting in a Generation of agonist DARPin dimers S6). For this purpose, EpoR fused to an N-terminal set of low-affinity ligands with dissociation con- In order to endow the monomeric DARPin monomeric XFP (mXFP)–tag was expressed in stants (Kd) in the micromolar range (Fig. 1C and NEEEC with the ability to dimerize EpoR and HeLa cells at physiologically relevant densities fig. S2). A combination of random mutagenesis act as an agonist, it was converted into two (~0.5 to 1 copies/mm2; ~1000 to 2000 copies/cell) by means of error-prone polymerase chain re- different homodimeric geometries by incorporat- and labeled by means of specific nanobodies. action (PCR) and DNA-shuffling techniques ing designed dimer interfaces at the nonbinding EpoR dimerization was quantified through dual- resulted in a >104-fold increase in affinity of indi- face of the DARPin (29). The interface mutations color cotracking of individual receptor dimers vidual clones (Fig. 1, C and D). The highest-affinity provide hydrophobic side chain interactions at (fig. S6 and movie S1). Like EPO, both ligands clone, N3C-E2, was expressed in Escherichia coli, nonsolvent accessible positions and salt bridge showed ligand-dependent dimerization of EpoR and its affinity was measured with surface plas- interactions at solvent-exposed positions, which at the cell surface, although C_R3 was less ef- mon resonance (SPR). E2 binds to EpoR ECD with results in the formation of noncovalent self- ficient than A_R3 (Fig. 2C, fig. S6, and movie S2). a1.89nMKd, providing a >1000-fold increase in assembling homodimers, A_R3 and C_R3, where Dimerization of EpoR initiates a sequential binding affinity relative to 8A9 (Fig. 1D). We de- Rx indicates the number of ankyrin repeats (Fig. 2, cascade of phosphorylation events, the initial note this DARPin as NEEEC, comprising three A and B, and figs. S3 and S4). The A and C dimers membrane proximal step being phosphorylation EpoR binding ankyrin repeats, EEE, and the N- and differ by a shift in the relative orientation of the of JAK2, followed by tyrosines on the EpoR intra- C-terminal capping regions. monomers across the C2 axis. Dimer C_R3 has an cellular domain (ICD), and then principally activa- The crystal structure of NEEEC (Fig. 1, E to G, offset of one repeat at the dimerizing interface tion of STAT5 and extracellular signal–regulated and table S1) shows that it binds to the EpoR compared with dimer A_R3, providing a twist of kinase (ERK), and to a lesser extent STAT1 membrane-distal D1 domain with a 1:1 stoichi- 30° in the relative orientation of the two di- and STAT3. Other pathways not surveyed here,

Mohan et al., Science 364, eaav7532 (2019) 24 May 2019 2of15 RESEARCH | RESEARCH ARTICLE

Fig. 2. Dimerization scaffolds for E2 resulting in agonism. (A and B) Point mutations (shown in green) inserted on the rear face of the DARPin lead to the formation of noncovalent homodimers with two different geometries. A schematic representation is shown for formation of the 2:2 DARPin:EpoR complex. An N3C DARPin containing three ankyrin repeats is represented by three spheres. The EpoR binding face of the ankyrin repeat is shown as “E” (pink), and the dimerizing face is shown as “D” (green), whereas the two EpoR domains D1 and D2 are shown

in teal. The dimer interface Downloaded from for C_R3 (PDB 6MOG) is offset by one repeat providing additional twist in the EpoR geometry in C_R3/EpoR (PDB 6MOH) (B). (C) Ligand- induced dimerization of EpoR on the surface of cells, as http://science.sciencemag.org/ observed with dual-color single-molecule TIRF micros- copy. (Inset) The method of single-molecule cotracking analyses. (D) Dose-response curves for STAT1/3/5 and ERK activation performed in UT7/ EPO cells stimulated with EPO, A_R3, or C_R3 for 15 min. Data are mean ± SD for two independent replicates. A sig- on May 28, 2019 moidal dose-response analysis was preferred over Gaussian distribution because only some ligands demonstrate a decrease in MFI values at high concentrations. including mitogen-activated protein kinase varying the scissor angle (rotation) and proximity nature of the DARPin molecule (33), this results (MAPK) and phosphatidylinositol 3-kinase (PI3K), between EpoR ECD monomers (distance). We in dimers in which the two EpoR binding inter- are also activated by EPO. We analyzed the designed three different DARPin dimer extension faces assume different separations and orienta- phosphorylation levels of various downstream series to interrogate signaling as a function of tions (Fig. 3). To engineer the “angle” series, one effectors on the human erythroid cell line, UT7/ these parameters in a systematic manner (Fig. 3). or more P repeats are inserted at the C-terminal EPO (32), using phospho-flow cytometry (Fig. 2D). The “angle” series is intended to pivot the EpoR end to provide DARPins such as NEEEP..C, (Fig. 3, Monomeric DARPins do not dimerize EpoR, gen- dimer in a scissorlike manner (Fig. 3, top). The top, and fig. S7). The dimerizing residues (in- erating no pSTAT5 signal. Both DARPin dimers “distance” series is intended to separate the dicated with underline) are engineered on the inducecomparableefficacy(Emax; the maximum EpoR subunits in the dimer while not substan- same terminus as that of the P repeat insertions. response achieved) compared with EPO for pSTAT5 tially changing the dimer angle (Fig. 3, middle). As the added repeats grow along a helical path, and pERK. Dimer A_R3 exhibits a higher Emax The “mono-extended” series differs from the angle this results in the generation of a series of dimers for pSTAT1 and pSTAT3 than EPO, whereas dimer and distance series in that both EpoR subunits are with variation in the angle between the two re- C_R3 exhibits a lower Emax (Fig. 2D). Thus, the bound to one extended DARPin (monomeric) that ceptor ECD monomers. The resultant extended relatively small orientational differences between contains two EpoR binding sites (Fig. 3, bottom). dimer is termed A_angle_Rx, where A indicates dimers A_R3 and C_R3 result in apparent sig- Systematic variation in each of these parame- the designed dimer interface, and “x” indicates naling differences. ters can be achieved by inserting nonbinding “P” the total number of ankyrin repeats. For exam- repeats derived from the consensus ankyrin repeat ple, NEEEPPC is termed A_angle_R5. Topological control of EpoR sequence, which changes the relative postions of To engineer the “distance” series, the nonbinding dimerization geometry the EpoR binding repeats across the designed P repeats are inserted at the N terminus—for To move toward topological control, we designed dimerization interface in predictable ways that example, NPP…EEEC (Fig. 3, middle, and figs. a series of NEEEC homodimers, systematically can be modeled. Because of the nonplanar, helical S8 and S9). The dimerizing residues are also

Mohan et al., Science 364, eaav7532 (2019) 24 May 2019 3of15 RESEARCH | RESEARCH ARTICLE placed at the N terminus so that this results in but with the bound EpoRs in an inverted rotation- where “x” indicates the total number of ankyrin two helical DARPins growing in opposite di- al orientation relative tothedimerizedDARPins repeats. rections with a twofold symmetry. With the ad- (“face-to-back” versus “face-to-face”). The two Each of the dimers was expressed in E. coli, dition of more P repeats, the binding repeats move sets of EpoR binding repeats are positioned purified through size-exclusion chromatography, further apart, leading to a distance-based series, at the two termini, separated by a series of P re- and bound to EpoR with similar affinities, as termed A_dist_Rx. peat insertions that can be varied in number to measured with SPR (figs. S7 to S9). To validate Last, we designed a monomeric, extended control the distance between the two receptors— the designs and also to determine whether any DARPin with two binding sites included on the for example, NEEEPPPP…EEEC. These mono- clashes were apparent between the EpoR recep- same DARPin chain (Fig. 3, bottom, and fig. S9), meric extended DARPins are termed M_Rx, tors, we determined six crystal structures of Downloaded from http://science.sciencemag.org/

on May 28, 2019

Fig. 3. Topological control of EpoR geometry.Three different series were ofthetworeceptorswiththeinsertionofPrepeatsisindicatedbypinkarrows. designed to systematically modify the different components of receptor The series are generated through stepwise insertion of one P repeat, and dimerization—specifically, angle (A_angle_Rx), distance (A_dist_Rx), the nomenclature “Rx” indicates the total numberofankyrinrepeats.The and inversion of geometry (M_Rx). The insertion (indicated by blue arrows) cartoon representations are shown along with the crystal structures for one of nonbinding “P” ankyrin repeats (blue) at different termini of nonplanar representative member per series: A_angle_R5 (PDB 6MOJ), A_dist_R7 DARPins leads to different receptor dimer topologies. The relative movement (PDB 6MOK), and M_R12 (PDB 6MOL).

Mohan et al., Science 364, eaav7532 (2019) 24 May 2019 4of15 RESEARCH | RESEARCH ARTICLE representative members of the series: the apo intended to convey qualitative relative signal distance (Fig. 5F) by comparing the signaling C_R3 dimer (1.2 Å resolution) as well as EpoR strengths, we more precisely confirmed with response for A_dist_R6 (face-to-face dimer, dis- complexes with C_R3 (3.46 by 3.16 Å resolu- immunoblotting the appearance of signal bias tance 49 Å) and M_R11 (face-to-back dimer, dis- tion), A_angle_R5 (3.39 by 2.45 Å resolution), (Fig. 4D). A similar STAT1 bias was observed tance 53 Å), where M_R11 is a mono-extended C_angle_R5 (3.0 by 2.0 Å resolution), A_dist_R7 for A_angle_R5 and C_angle_R4, with extremely DARPin containing two EpoR binding sites (5.1 Å resolution), and M_R12 (4.58 by 3.14 Å high pSTAT1 signal at highest concentrations separated by five ankyrin repeats, NEEEPPPP- resolution) (Fig. 3, fig. S10, and tables S1 and S2). (fig. S13). Immunoblot analysis of cell lysates PEEEC. M_R11 delivers an extremely weak The structures validate the DARPin dimer de- further verified the STAT1 bias observed for A_R3, pSTAT5 signal (Fig. 5G), suggesting that the signs, albeit with minor rigid body variations A_angle_R5, and C_angle_R4 compared with EPO face-to-face orientation appears to be favorable because of lever arm effects centered at the (Fig. 4D and figs. S13C and S14). for the intracellular JAK2 to transphosphorylate DARPin dimer interface (fig. S11), allowing us within an EpoR dimer, thus more effectively to model the full geometry series based on the EpoR distance series initiating signaling. These experiments show known effects of P repeat insertion. In all of the The basic building block of the distance series that rotational reorientation of the receptor structures determined, whereas the D1 domain is A_R3 with zero P repeats (NEEEC), and sub- ECDs can profoundly affect signaling output, of EpoR maintains a rigid contact with the sequent members incorporate one or more P which is consistent with previous studies in DARPin, the D2 domain exhibits some seg- repeats: A_dist_R4 (NPEEEC) through A_dist_R9 other systems using chimeric receptors (23) mental flexibility and adapts variable positions (NPPPPPPEEEC) (Fig. 5A). Contrary to the angle but now provides a ligand-mediated approach to relative to D1 that are likely influenced by crystal series in which a reduction in pSTAT5 and pERK exploit this parameter on natural, unmodified packing. We measured a range of 28° variation in Emax was observed, the first few members of receptors. D2 position (fig. S12), which is consistent with A_dist exhibit the same Emax despite the increased that observed in previously published structures distance between the receptors (Fig. 5B). In- Effects of designed ligands of EpoR. stead, the effects of the distance are reflected in on erythropoiesis Downloaded from arightshiftintheEC50 values. Furthermore, Erythropoiesis begins with lineage commit- EpoR angle series a bias toward higher pSTAT1 and pSTAT3 Emax ment of hematopoietic stem and progenitor cells The basic building block of the angle series is than EPO or other distances was observed for (HSPCs) into the erythroid lineage, and cells A_R3 with zero P repeats (NEEEC), and sub- A_dist_R5 (figs. S14 to S16). Analysis of a panel of progressivelydifferentiateandmaturetobecome sequent members with one or more P repeats downstream phosphorylated proteins by means of RBCs (Fig. 6A). This process of RBC development

added in, A_angle_R4 (NEEEPC) through phospho-barcoding reflected the above trends, in can be faithfully recapitulated in vitro by cultur- http://science.sciencemag.org/ A_angle_R7 (NEEEPPPPC), progressively tilt which the strong bias was observed for STAT1 for ing human primary HSPCs with various cyto- each EpoR monomer toward the membrane the A_dist_R5 ligand (Fig. 5, C to E; figs. S14 and kines over 2 to 3 weeks (Fig. 6A) (36, 37). As a (Fig. 4A). As measured with phospho-flow cy- S15; and table S5). The biased and prolonged main driver of RBC development, EPO plays a tometry in UT7/EPO cells, the dimer A_angle pSTAT1 response for A_dist_R5 was further role in all aspects of this process, from lineage series exhibits a progressive reduction in Emax verified through immunoblotting with 15- and commitment to terminal maturation (Fig. 6A) levels of STAT5 phosphorylation as the EpoR 60-min ligand treatments (Fig. 5D) and may be (38, 39). Therefore, distinct stages of erythroid dimer scissor half-angle increases from ~38° attributable in part to the differential intracel- maturation may have differences in EPO signaling. (A_R3) to 72° (A_angle_R9) (Fig. 4, A and B). lular processing rate for the internalized recep- Using DARPins that either increased the dis- Dimers A_angle_R4 and A_angle_R5 act as tors (Fig. S17). In general, ligands with faster tance or altered interaction angles between EpoR partial agonists, whereas A_angle_R7 does not internalization of cell-surface EpoR upon ligand dimers, we found a graded reduction in over-

elicit a signal detectable with flow cytometry. treatment showed a more potent response. Fur- all erythroid differentiation and cellular expan- on May 28, 2019 Similarly, the C_angle series shows by means thermore, with an increase in angle or distance, a sion (Fig. 6B and fig. S18). Looking at overall of flow cytometry a stepwise decline in activa- decrease in activity was observed, which corre- erythroid differentiation by using flow cytom- tion that is confirmed with immunoblotting (fig. lates with the slower rate of internalization of the etry allowed us to assess the kinetics of HSPC S13). In addition to the reduction in Emax,the receptor. Recycling of EpoR back to the surface is commitment into the erythroid lineage and effects of increase in angle are also reflected by slower for A_dist_R5 than other DARPin ligands. the continuation of erythroid terminal matu- changesinmedianeffectiveconcentration(EC50), For the distance series, the right shift of the EC50 ration throughout the culture. In addition, pro- which is defined as the ligand concentration that was surprising because we would not imagine liferation is one of the important hallmarks in provides 50% of maximal response (Emax). A strain on the dimer through pure translation of erythropoiesis that is coupled with the overall dif- progressive right shift was observed for the EC50 the receptors to wider intermonomer distances ferentiation efficacy. Therefore, these analyses values, implying the need for a higher ligand con- unless there was receptor preassociation that together in a time-dependent manner can give centration to generate the same levels of signal, could present an energy barrier to separation. us further insights into stage-selective depend- despite all of the dimers having the same affinity Although we did not see EpoR predimerization encies of specific EpoR activities in erythropoiesis. for EpoR, as measured with SPR by using soluble with single-molecule TIRF microscopy, increasing This range of erythroid differentiation efficacy EpoR ECD (fig. S7). We speculate that as the the distance may interfere with weak synergistic stimulated by various DARPin series was as- scissor angle between EpoRs becomes more ex- interactions between the receptor subunits, pos- sessed by using two markers: CD71, a marker of treme, strain is induced as a result of cell mem- sibly mediated by the transmembrane domains early erythroid commitment, and CD235a or brane constraints, and this is reflected in less (34, 35). glycophorin A, a marker of terminal erythroid efficient bivalent complex formation. maturation (Fig. 6C). The majority of cells cul- We also analyzed a broad panel of down- Inverted EpoR geometry tured with EPO, A_R3, A_angle_R4, C_R3, and stream phosphorylated proteins using phospho- We assessed the impact on signaling of relative A_dist_R5 differentiated into CD235a+ erythroid rylation barcoding, which provides a broad EpoR ECD rotation with respect to one another cells, whereas the later members of the series “footprint” of relative signal strengths. This data while maintaining an approximately similar dis- (wider angle or farther distance of the EpoR showed a similar trend of reduced Emax with tance. In the native EPO/EpoR dimer structure, dimer) impaired or failed to effectively promote increasing scissor angle (Fig. 4C, fig. S13B, and the receptors are face-to-face. Having validated erythroid commitment and differentiation (Fig. tables S3 and S4). At higher concentrations, a the design of a face-to-back orientation with the 6C). Consistent with the pattern we observed bias was observed for STAT1 and STAT3 by A_R3 crystal structure of the DARPin M_R12/EpoR with flow cytometry for cell surface markers, compared with EPO by means of phospho-flow. complex(Fig.3),weassessed face-to-back orien- treatment of cells with agonists that resulted Because the phosphorylation barcoding data are tation while maintaining a close intermonomer in delayed or impaired erythroid differentiation

Mohan et al., Science 364, eaav7532 (2019) 24 May 2019 5of15 RESEARCH | RESEARCH ARTICLE also resulted in reduced cellular proliferation exhibited 1.7 ± 0.1% CD235a+CD71– cells, whereas the later stages. In addition, C_R3 promoted (Fig. 6D). The cells cultured with EPO, A_R3, or cells with the recombinant EPO and A_R3 showed early erythropoiesis poorly but showed increased A_dist_R5 showed at maximally potent concen- 28.7 ± 2.0% and 26.8 ± 1.6% in this population, differentiation efficacy in the later stages (fig. S18), trations very similar erythroid differentiation respectively, indicating the impairment of ter- suggesting that this ligand selectively promotes efficacy on differentiation day 5 (early phase of minal erythropoiesis in the late stage with the later stages of erythropoiesis while having re- erythropoiesis), with CD235a+CD71+ of 31.7 ± A_dist_R5 (Fig. 6E). Furthermore, cells cultured duced stimulation earlier. Collectively, these results 1.0%, 28.0 ± 1.6%, and 30.9 ± 0.6%, respectively with A_dist_R5 demonstrated reduced prolifer- demonstrate how topological alterations in EpoR (Fig. 6E). However, on differentiation day 12, ation, relative to EPO or A_R3, in the later stages dimerization can selectively promote effective cells cultured with A_dist_R5 showed lower of the erythroid differentiation culture (Fig. 6D), signaling at specific stages of human erythropoiesis CD235a+CD71––positive cells compared with demonstrating that A_dist_R5 specifically pro- but fail to effectively enable differentiation and EPO and A_R3. Cells cultured with A_dist_R5 motes early erythropoiesis with impairments at proliferation at other stages. Downloaded from http://science.sciencemag.org/

on May 28, 2019

Fig. 4. Signaling responses induced through variation in EpoR dimer angle. (A) Addition of P repeats at the C terminus leads to increase in the angle between the EpoR ECDs, as shown. (B) Signaling bias observed for the dose-response curves for the angle series. Shown are percentage of pSTAT1/3/5 and pERK induced by the various DARPin ligands relative to EPO in UT7/EPO cells. Data are mean ± SD for two independent replicates. A sigmoidal dose-response analysis was preferred over Gaussian distribution because only some ligands demonstrate a decrease in MFI values at high concentrations. (C) Bubble plot representation of the various downstream pathways activated by EPO or the DARPin ligands in UT7/EPO cells with 15, 60, and 120 min of stimulation. The size of the bubble correlates to the signal strength determined as the log2 value of the ratio of the signal generated by the ligand to unstimulated cells. (D) The levels of phosphorylated downstream effectors were analyzed in the cell lysates by immunoblotting after 15 and 60 min of ligand treatment. EpoR-immunoprecipitates were analyzed by immunoblotting for phosphorylation of specific Tyr residues in the EpoR ICD. The dashed lines indicate the sites where the immunoblots were cropped for clarity. The uncropped immunoblots are included in fig. S14.

Mohan et al., Science 364, eaav7532 (2019) 24 May 2019 6of15 RESEARCH | RESEARCH ARTICLE

EpoR EpoR D1 D1

D2 D2 Downloaded from http://science.sciencemag.org/

on May 28, 2019

Fig. 5. Signaling responses induced by variation in EpoR ECD proximity. (A) Addition of P repeats at the N terminus leads to increase in the distance between the EpoR ECDs, as shown. (B) Signaling bias observed for the dose-response curves for the distance series. Percentage of pSTAT5 and pERK induced by the various DARPin ligands relative to EPO in UT7/EPO cells. Data are mean ± SD for two independent replicates. A sigmoidal dose-response analysis was preferred over Gaussian distribution because only some ligands demonstrate a decrease in MFI values at high concentrations. (C) Bubble plot representation of the various downstream pathways activated by EPO or the DARPin ligands in UT7/EPO cells with 15, 60, and 120 min of stimulation. The size of the bubble correlates to the signal strength. (D) Immunoblot analysis of phosphorylated downstream effectors in cell lysates after 15 and 60 min of ligand treatment, and analysis of EpoR-immunoprecipitates for phosphorylation of specific Tyr residues in the EpoR ICD. The dashed lines indicate the sites where the immunoblots were cropped for clarity. The uncropped immunoblots are included in fig. S14. (E) The log2 value of the ratio of the signal generated by the ligand to unstimulated cells at 60 min. (F) The model for M_R11 binding to EpoR. (G) Inversion of the relative geometry of the EpoR ECDs for the mono-extended M_R11, which possesses a similar inter-ECD distance to A_dist_R6, leads to a complete loss of pSTAT5 activity.

Mohan et al., Science 364, eaav7532 (2019) 24 May 2019 7of15 RESEARCH | RESEARCH ARTICLE Downloaded from http://science.sciencemag.org/

on May 28, 2019

Fig. 6. Effects on hematopoiesis of topologically controllable EpoR cytometry analysis of erythroid differentiation by EPO, A_R3, or A_dist_R5 ligands. (A) Overview of erythropoiesis and in vitro differentiation of at different time points are shown with flow cytometry analysis. Means ± human primary hematopoietic cells. (B) Erythroid differentiation from SEM for three independent experiments are shown. (F) Experimental CD34+ HSPC with surrogate EPO ligands. Differentiation has been scheme of analyzing ligand-stimulated downstream signaling stimulated assessed with flow cytometry with the markers CD71 and CD235a at day by EPO, A_R3, A_dist_R5, and C_R3 during erythropoiesis. (G) Total MFI 12 of differentiation. Means ± SEM for three independent experiments are of pSTAT1/3/5 stimulated with EPO, A_R3, A_dist_R5, or C_R3 at early shown. (C) Total CD235a+ cells and (D) overall cell proliferation during (day 5 of differentiation) and late (day 12 of differentiation) erythropoiesis. erythroid differentiation. Data are means ± SEM for three independent Data are (mean ± SEM) for three independent experiments [(B), (E), experiments (n = 2 independent experiments for A_dist_R5). (E) Flow and (G)]. **P <0.01, *P <0.05.

Mohan et al., Science 364, eaav7532 (2019) 24 May 2019 8of15 RESEARCH | RESEARCH ARTICLE

These phenotypes led us to ask whether there new strategy allows tunable manipulation of ferentiation. This approach could also be used may be biased signaling responses downstream geometric parameters, but there currently is to survey cytokine receptor dimer geometries of EpoR that correlate with such effects (Fig. 6F). no predictive framework to relate these geometric in order to identify signaling outputs of clinical Erythroid cells stimulated with A_dist_R5 showed changes to signaling outcome. As such, the topo- interest in other cytokine receptor system, such higher activation of STAT1/3/5 than with the logical variation this system affords through as in the immune system. The scaffold we de- recombinant EPO or A_R3 in the early stages proteinligandengineeringcanbeanalogizedto scribe here could be used as a preclinical phar- (day 5). However, erythroid cells stimulated with exploration of small-molecule pharmacology by macological tool to determine whether different A_dist_R5 showed higher activation of STAT1 using medicinal chemistry. degrees of agonism are optimal for a given and STAT3, but lowered pSTAT5 was observed Although the strategy itself is mechanistically cytokine in a particular therapeutic indication. in the late stage of erythropoiesis (day 12) relative agnostic, the signaling effects we observed in to EPO or A_R3 in the late stage of erythropoiesis response to the topologically varied ligands are Materials and methods (Fig. 6G). Furthermore, cells stimulated with likely due to two factors. First, the orientational EpoR protein expression and purification C_R3 showed low activation of STAT5 on dif- effects of the ECD could propagate to the ICDs The DNA encoding for EpoR ECD was cloned ferentiation day 5 (Fig. 6G). However, although that associate with the JAK2 kinases, affecting into pAcGP67-A, an insect cell protein expression the level of activation of STAT5 was still lower their relative proximities and orientations, which vector. The vector includes a C-terminal 3C pro- with C_R3 relative to EPO or A_R3, the level of alters the efficiency by which JAK2 acts on its tease site, followed by a biotin-acceptor STAT5 activation between A_dist_R5 and C_R3 subsequent substrates on the EpoR ICD and tag (BAP tag, GLNDIFEAQKIEW), and a hexa- was not significantly different (P = 0.40) on STAT1, -3, and -5 (Figs. 4D and 5D and fig. S14). Histidine tag for affinity purification. The DNA differentiation day 12, a time point when C_R3 In different receptor topologies, the JAK2/ICD encoding for the glycomutant of EpoR ECD-N46Q, began to perform better than A_dist_R5 (Fig. 6, complexes are likely repositioned into orienta- N157Q (25) was also cloned into pAcGP67-A with BandG).Thisdatasuggeststhatbiasedsignal- tions that are more or less conducive for trans- a 6-His tag for affinity purification. The proteins ing is affected by changes of EpoR dimer geom- phosphorylation of the opposing JAK2 in the were expressed using the baculovirus expression Downloaded from etry in primary human erythroid cells. However, dimer, as well as tyrosine sites on the EpoR ICD. system. The baculovirus stocks were prepared by we have only surveyed a limited range of down- This could subsequently result in altered recruit- cotransfection of the BaculoSapphire DNA (Orbi- stream effectors and thus may be missing specific ment and/or phosphorylation of STATs and gen) and the pAcGP67-A DNA into Spodoptera signals that mediate the observed phenotypic other adaptors that directly engage the JAK2 frugiperda (Sf9). Next, the viruses were used to effects. Our data suggests that this biased sig- and ICD phospho-tyrosines, and these mem- infect the Trichoplusia ni (High Five) cells. After

naling is also dependent on the differentiation brane-proximal phosphorylation differences 72hours,theproteinswereharvestedfromthe http://science.sciencemag.org/ stage–specific context of the cells on which they will subsequently propagate to supernatant and purified using nickel affinity act. The same agonists can result in different programs in the nucleus (41, 42). In general, chromatography (Nickel-NTA, Qiagen), followed downstream signaling depending on cell state. the more extreme angles and distances result by size-exclusion chromatography with a Super- The ability to modulate signaling through a in less efficient downstream phosphorylation dex S200 column (GE Healthcare). The proteins single cytokine pathway at various stages of of all downstream adaptors to similar extents. were maintained in HEPES buffered saline (HBS, hematopoietic differentiation has broad appli- However, uncoupling in the efficiency of JAK2 20 mM HEPES pH 7.4, 150 mM sodium chloride). cations in tuning desired responses for specific and STAT phosphorylation for some of the EpoR ECD was site-specifically biotinylated at ligands. ligands—for example, A_R3, A_angle_R5, and the C-terminal BAP tag using BirA ligase and A_dist_R5—results in the emergence of biased re-purified by size exclusion chromatography. Discussion STAT activation. Protein concentrations were determined by UV

Although tunable signaling is a foundation A second explanation of the observed effects absorbance at 280 nm using a Nanodrop2000 on May 28, 2019 of GPCR pharmacology that can be exploited likely relates to mechanical distortion. As the spectrometer (Thermo Scientific), and then stored through small-molecule medicinal chemistry, topologiesareforcedintomoreextremeranges, at –80°C. similar pharmacological principles have not been strain is induced in the system, resulting in fewer deeply explored for dimeric receptors that engage numbers of signaling complexes formed, with DARPin expression and purification cytokines and growth factors because of the dis- subsequent reduced receptor dimerization effi- The DNA encoding for the N2C and N3C DARPin tinct structural challenges presented by the dif- ciency as observed with single-molecule assays, proteins was cloned into pET-28 vector, a bacterial ferent systems. For GPCRs, signal modulation is leading to lower Emax and right-shifted EC50s expression vector with a Kanamycin antibiotic achieved through small-molecule ligands that and disproportionately affecting some second resistance marker. The vector includes a C-terminal bind within pockets and induce different con- messengers more than others. Supporting this hexa-Histidine tag for affinity purification. The formations of the GPCR helices within the mem- interpretation, there is a systematic pattern of plasmids containing the gene of interest were brane, resulting in differential activation of G reduced Emax that correlates with increasing used to transform Rosetta DE3 competent cells. proteins and arrestins (40). Structure-activity dimer angle. For the distance series, there is no Thecellsweregrownat37°Cin2YTmediasup- relationships for GPCR agonists can be obtained change in Emax until a sharp threshold, at which plemented with Kanamycin (40 mg/mL) until the through medicinal chemistry. However, a com- point it decreases abruptly. This could reflect a culture reached log-phase growth. Next, IPTG parable approach does not exist to explore signal “rupture” distance at which point dimerized was added to the culture to induce protein ex- tuning by the large class of Type-I transmem- JAK2 molecules can no longer transphospho- pression at a final concentration of 1 mM. The brane receptors, which require protein ligands to rylate efficiently. This rupture happens earlier culture was shaken at 37°C for 3 hours. Next, the bindtolargeECDsandinducelargeconfor- with the angle series than with the distance proteins were harvested from the cells by sonica- mational changes that lead to dimerization. series because the contortions needed to recover tion, and purified using nickel affinity chromatog- Although there have been hints that conforma- are greater. Possibly, both the angle and distance raphy, followed by size-exclusion chromatography tional changes that affect dimer topology may series interfere with weak, cooperative interac- with a Superdex S75 column (GE Healthcare). The bias signaling (6, 15, 18, 24), the functional rel- tions between the transmembrane domains of proteins were maintained in phosphate-buffered evance has not been well characterized and is EpoR that have been reported (34, 35). saline(PBS,135mMNaCl,2.5mMKCl,8.0mM not well understood. We describe a general As we have demonstrated for RBC biogenesis, Na2HPO4,30mMKH2 PO4, pH 7.2). Protein strategy to probe the effects of topological this approach is suitable for examining how concentrations were determined by UV absorb- variation of receptor dimerization on signaling specific properties of cytokine signaling path- ance at 280 nm using a Nanodrop2000 spec- and function for cytokines and, in principle, any ways may have differentiation-stage specificity trometer (Thermo Scientific), and then stored Type-I class receptor that signals as a dimer. This in hematopoiesis or other aspects of cell dif- at –80°C.

Mohan et al., Science 364, eaav7532 (2019) 24 May 2019 9of15 RESEARCH | RESEARCH ARTICLE

The constructs for DARPin dimers based on TCCGACCTGGGTAAGAAACTGCTGGAAG- EpoR at a final concentration of 1 mM monomer, the dimer C_R3 scaffold include an N-terminal CAGCTCGTGCTGGTCAGGACGACGAAGTTC- in 1 mL PBE, with shaking for 2 hours at 4°C. The hexa-Histidine tag for affinity purification. All of GTATCCTGATGGCTAACGGAGCAGATGTCA- yeast cells were washed with PBE and then in- the extended DARPin proteins were maintained ACGCGASWGACSWCWMCGGTKYCACA- cubated with Alexa Fluor 488-labeled anti-cMyc in Tris-buffered saline (50 mM Tris pH 8, 150 mM CCTCTGCACTTAGCTGCAMWARHCGGTC- () and streptavidin- NaCl). The DARPin dimers A_dist_R5 through ATTTGGAAATCGTCGAAGTTCTACTGAAG- conjugated Alexa Fluor 647. The cells were then A_dist_R10, all of the mono-extended M_Rx CACGGTGCTGACGTTAATGCTASWGAT- sorted using FACS to select for clones that ex- DARPins, A_angle_R7, A_angle_R9, C_angle_R6, MWSWMCGGADYGACTCCGTTACATC- hibited high levels of EpoR binding (MFI of Alexa and C_angle_R8 were induced at 16°C for 16 to TAGCGGCTANTADSGGACATCTCGAGA- Fluor 647) relative to DARPin display levels (MFI 18 hours. TTGTAGAGGTGTTACTCAAATACGGCGC- of Alexa Fluor 488) on yeast. AGATGTAAACGCATMCGATSWCWMC- Similarly, for the N2C library, the subsequent Cell culture GGCRYGACGCCATTGCACCTCGCAGCT- rounds of selections were performed at 333 nM, The human UT7/EPO cells (32) were cultured AVWAWSGGCCACTTAGAGATAGTTGAAGT- 40 nM and 4 nM EpoR concentrations for rounds in DMEM (Dulbecco’s Modified Eagle Medium) GCTCTTAAAGTATGGAGCCGACGTGAATGC- 5, 6, and 7, respectively. Whereas, for the N3C supplemented with 10% v/v fetal bovine serum, CCAGGACAAATTCGGTAAGACCGCTTTCGA- library, the subsequent rounds of selections penicillin-streptomycin, 1 mM L-glutamine, HEPES CATCTCCATCGACAACGGTAACGAGGACCTG- were performed at 333 nM, 10 nM and 4 nM buffer, and 2 U/mL recombinant EPO (Amgen) GCTGAAATCCTGCAA EpoR concentrations for rounds 5, 6, and 7, (20). The cells were maintained at 37°C with Theoretical diversity: 5.2 × 1011 respectively. 5% CO2. The cultures were passaged every 2 Diversity of library generated through elec- to 3 days. troporation: 1.0 × 109 Second-generation library: Evolution of high-affinity DARPins

Yeast display of DARPins Evolution of EpoR binding DARPins The two second generation DARPin libraries Downloaded from DARPins were displayed on the surface of yeast The two yeast-displayed DARPin libraries (N2C (N2C and N3C) containing DNA-shuffled clones S. cerevisiae strain EBY100 by fusion to the andN3C)werepannedagainsttheEpoRECDto were panned against biotinylated-EpoR ECD to C terminus of the Aga2 protein. The DARPin select for high affinity clones using a combina- select for high affinity clones, as described above. insert, along with an N-terminal 3C protease tion of magnetic-activated cell sorting (MACS) Diversities of the second-generation libraries cleavage site and a C-terminal cMyc epitope and fluorescence-activated cell sorting (FACS). generated through electroporation: 9

were cloned into the PCT302 vector. Briefly, Biotinylated EpoR was used for all the selection N2C library: 3.2 × 10 http://science.sciencemag.org/ competent yeast cells were electroporated with rounds unless otherwise noted. The first round N3C library: 4.6 × 109 plasmids and recovered in SDCAA selection of selection was performed with 5.4 × 109 N2C The first round of selections was performed media at 30°C (43). Next, grown cultures were and 5 × 109 N3C yeast cells to ensure coverage with 1 × 1010 yeast, to ensure coverage of library passaged in SDCAA media before inducing ex- of library diversity. First, EpoR was mixed with diversity. For round 1, EpoR was mixed with pression in SGCAA media. The cultures were 500 ml of magnetic streptavidin microbeads 1000 ml of magnetic streptavidin microbeads induced at 30°C for 24 hours, followed by 20°C. (Militenyi) to a final concentration of 400 nM. (Militenyi) to a final concentration of 400 nM, The display of full-length DARPins on the sur- The mixture was then incubated with DARPin- and EpoR binding clones were isolated using face of yeast was confirmed by staining the cells displaying yeast cells at a final volume of 20 mL MACS LS columns. For round 2, the library pools with an Alexa Fluor 488-labeled anti-cMyc anti- in PBE (PBS, 0.5% bovine serum albumin, 2 mM were incubated with 50 nM and 10 nM EpoR- body (1:100 dilution; Cell Signaling) and moni- EDTA). The cells were incubated with shaking tetramers for N2C and N3C libraries, respectively.

toring the fluorescence by flow cytometry. for 2 hours at 4°C. Next, the yeast cells were The EpoR tetramers were generated by mixing on May 28, 2019 washed with PBE and flowed over MACS LS biotinylated EpoR and streptavidin-conjugated Yeast displayed DARPin libraries separation columns (Miltenyi) to isolate the Alexa Fluor 647, at a molar ratio of 4.5:1. Next, the Two site-directed DARPin libraries (N2C and EpoR binders. The enriched library pool was yeast cells were incubated with EpoR-tetramers for N3C) were created by assembly PCR using the regrowninSDCAAmediaandtheninducedin 2 hours at 4°C. The yeast cells were then washed following degenerate codons. The gel-purified SGCAA. In subsequent rounds, the total number with PBE and incubated with anti-Alexa Fluor 647 assembled product was combined with linearized of yeast cells used was ten times the number of microbeads. The EpoR binders were isolated using PCT302 vector and electroporated into EBY100 yeast cells eluted in the previous round or 1 × MACS LS separation columns. yeast cells (43). The electroporated yeast were 108 cells, whichever is larger. To further enrich the library pool, FACS recovered with SDCAA selection media and in- For round 2, the on-bead selection with high method was employed in subsequent rounds. duced with SGCAA media, as described above. avidity effects was repeated to select for more For both the libraries, the subsequent rounds N2C Library: EpoR binders without increasing the stringency. of selections were performed at 10 nM, 1 nM, TCCGACCTGGGTAAGAAACTGCTGGAAG- EpoR was mixed with 250 ml of magnetic strep- and 0.1 nM EpoR concentrations for rounds 3, CAGCTCGTGCTGGTCAGGACGACGAAGTTCGT- tavidin microbeads (Miltenyi) to a final concentra- 4, and 5, respectively. For round 6, the library ATCCTGATGGCTAATGGAGCAGATGTCAACGC- tion of 400 nM and incubated with yeast-displayed pool was subjected to a kinetic sort; a method GASWGACNWTDBGGGTDYGACTCCGCTG- DARPin cells in 10 mL final volume. In round 3, the to select for clones with slower off-rates. First, the CACCTGGCTGCTMDSRHMGGTCACCTGGA- library pool (1 × 108 yeast cells) was incubated library pools were incubated with 5 and 2.5 nM AATCGTTGAAGTTCTACTGAAGAACGGTGCTG- with Alexa Fluor 647-labeled anti-cMyc antibody, EpoR for N2C and N3C libraries, respectively. ACGTTAACGCTASWGATNWTWMCGGA- to enrich for transformants displaying full length The yeast cells were washed and then dual-stained DYGACACCACTTCATCTAGCAGCGAVW- DARPin clones. The mixture was incubated with with Alexa Fluor 488-labeled anti-cMyc and MDSGGACATCTCGAGATTGTCGAGGTCT- shaking for 2 hours at 4°C, followed by PBE streptavidin-conjugated Alexa Fluor 647. Next, a TACTCAAACACGGCGCAGACGTAAATGCA- washes. The yeast cells were then incubated 100-fold excess of unbiotinylated EpoR, 500 nM CAGGACAAATTCGGTAAGACCGCTTTCGATA- with anti-Alexa Fluor 647 microbeads (Miltenyi) for N2C and 250 nM for N3C in 1 mL was added TCTCCATCGACAACGGTAACGAGGACCTGGC- andflowedoverMACSLSseparationcolumnsto to the cells. The mixture was shaken at room TGAAATCCTGCAA isolate the full-length clones. temperature, and the levels of bound protein Theoretical diversity: 2.1 × 1010 In subsequent rounds, FACS method was em- were measured at regular intervals using flow Diversity of library generated through elec- ployed to selectively enrich for EpoR binding cytometry. The levels did not change signifi- troporation: 1.8 × 109 clones. For round 4, the propagated and induced cantly over 21 hours. The cells were washed, N3C Library: yeast cells from round 3 were incubated with restained, and then subjected to FACS.

Mohan et al., Science 364, eaav7532 (2019) 24 May 2019 10 of 15 RESEARCH | RESEARCH ARTICLE

Analysis of individual clones DNA shuffling P repeat sequence: DAAGMTPLHLAAANGH- To isolate individual clones, yeast cultures from The EpoR binding clones with random muta- LEIVEVLLKYGADVNAK the last selection rounds were plated on SDCAA tions inserted through error-prone PCR were To engineer the “distance” series, the non- agarose plates. Individual colonies were trans- subjected to DNA shuffling (44)toenhancebind- binding P repeats were inserted at the N terminus, ferred to SDCAA media in culture tubes, passaged ing affinity. The amplified and purified DNA for example NPP…EEEC. Insertion of repeats at and then induced. Next, 93 clones were screened product from various clones was combined, and the N terminus proved challenging, as illustrated against a fixed EpoR concentration. For each then used as the template for DNA shuffling. in fig. S9. A straightforward design based on the clone, ~5 × 105 cells were transferred to a V- Briefly, 2 mg of combined DNA was subjected to angle series such as NPEEEC displayed a sig- bottom plate and incubated overnight with bio- DNase treatment (DNase I, Invitrogen) by in- nificant loss in binding affinity toward EpoR. tinylated EpoR at 4°C. The cells were washed cubating at 15°C for three mins, followed by 80°C Three possible reasons were identified. First, and stained with streptavidin-conjugated Alexa for 10 min to denature the enzyme. The DNA frag- an analysis of the models revealed a clash be- Fluor 647. Next, the clones were analyzed for ments were subjected to PCR clean-up (Qiagen) tween residues of the inner helices from the first binding using flow cytometry. The highest af- to remove extremely small fragments. Next, the binding repeat and the preceding P repeat (fig. finity clones were identified, and their plasmids fragments were allowed to recombine at regions S7A). No such clashes were observed in the E2 were isolated using the Zymoprep Yeast Plasmid of moderate homology. The PCR was performed DARPin where the N-terminal capping region Miniprep II Kit (Zymo Research), and sequenced. using Taq (New England Biolabs) and PfuUltra helix precedes the first binding repeat (fig. S7B). Yeast-displayed DARPin ligands were also titrated (Agilent) polymerases with progressive hybrid- Second, an analysis of the sequence of the E2 against EpoR at different concentrations to obtain ization. The reaction product was further ampli- N-terminal capping region revealed a Glu to Lys 5 EC50 values. As described above, ~5 × 10 cells fied using PfuUltra polymerase. The product was mutation in the inner helix, which resulted from were transferred to a V-bottom plate and in- purified using gel electrophoresis and the bands the error-prone mutagenesis performed during cubated with 100 mL of serially diluted EpoR corresponding to N2C and N3C sizes were spe- affinity maturation. The Lys residue is well posi- solution. Next, the cells were washed, stained, cifically extracted and purified. The gel-purified tioned to interact with EpoR. Third, insertion of Downloaded from andthenanalyzedforEpoRbindingbyflow product was combined with linearized PCT302 the P repeat implies existence of b-turn loops cytometry. vector and electroporated into EBY100 yeast cells preceding the loop region from the first binding to create the second-generation libraries. Diver- repeat. In E2, A_R3 (NEEEC) and the angle Surface plasmon resonance sity of library generated through electroporation: series, no b-turn loops precede the loop region 9 Dissociation constants (Kd) for EpoR-binding N2C library: 3.2 × 10 for the first ankyrin repeat. The newly inserted 9

DARPin ligands were determined by surface N3C library: 4.6 × 10 loops could introduce destabilizing interactions. http://science.sciencemag.org/ plasmon resonance (SPR) using the BIAcore To address these issues, a modified repeat (P′) T100 instrument (GE Healthcare). First, bio- Engineering the extended DARPins was used as the repeat preceding the first E tinylated EpoR was captured on a streptavidin- The modular nature of the DARPin was utilized repeat such as NP..P′EEEC. In the P′ repeat, the coated (SA) sensor chip (GE Healthcare) with toward the generation of the extended DARPins. sequence of the inner helix was modified from immobilization density in the range of 100 reso- The DARPin molecules are non-planar, and each HLAAANG to HKAARAG. The “KAARA” se- nance units (RU). Similarly, a control flow cell repeat provides a 2-3° counter-clockwise rotation quence is derived from the inner helix of the was also prepared with an off-target protein such relative to the preceding repeat (33). Thus, it is N-terminal capping region and installed in the P′ as biotinylated-IFNAR1 or biotinylated-Fzd8 for crucial to maintain the overall curvature of the repeat based on structural symmetry of the two reference subtraction. The binding kinetics were DARPin to avoid a loss in binding affinity while helices. This eliminates the clashes between the performed at 25°C with a flow rate of 50 ml/min. insertingmoreankyrinrepeats.Thesequenceof helices and also re-positions the Lys residue to

EpoR binding DARPin ligands were serially di- the P repeat was carefully designed to maintain interact with EpoR. Furthermore, the loop region on May 28, 2019 luted in HBS buffer supplemented with 0.005% the curvature observed in the N3C E2 monomer, with the sequence DAAGM was shortened to Gly, P20 surfactant (GE Healthcare) and injected over and the A_R3 and C_R3 dimers. The framework to facilitate the turn while eliminating any de- the SA chip. Dissociation kinetics were monitored residues in the P repeat were based on the se- stabilizing interactions, (fig. S7C). Increase in for 200-500 s, as needed, based on the ligand quence of the consensus repeat. The positions the loop length for DARPins has been previously affinity. For high affinity ligands, the chip was that were randomized in the library design (de- published (45). Here, we show that the loops can regenerated with 4 M MgCl2 after each ligand noted as X) (fig. S1) were assigned residues from also be shortened without affecting the structure injection. The binding kinetics and dissociation a Frizzled8-binding DARPin. As a negative con- for A_dist_R7 (fig. S7D). The dimerizing residues constants were determined using the BIAcore trol, a DARPin dimer comprising of only non- are now maintained at the N-terminal repeats. T100 evaluation software. The sensograms ob- binding P repeats was designed (NPPPPC) to To generate A_dist_R4, P′ repeat and the dim- tained were either fit to the 1:1 kinetic binding eliminate the possibility of off-target (Frizzled8) erizing residues were inserted at the N-terminal model or to the steady-state affinity model. The effects. The control dimer shows no affinity to- end (NP′EEEC). Whereas, to generate A_dist_R5, kinetic binding curves for each ligand were gen- ward EpoR or Frizzled8 (fig. S6). one P followed by one P′ repeat was inserted, erated by plotting the time-dependent response As described above, the E2 DARPin is com- NPP′EEEC. Similarly, the design for A_dist_R6 units in Prism 7 (GraphPad). prised of three EpoR binding repeats, denoted as can summarized as NPPP′EEEC. Also, the se- “EEE,” and the N- and C-terminal capping re- quence of the P repeat used for generating the Error-prone PCR gions. To engineer the “angle” series, one or distance series is different from the P repeat used The EpoR binding individual clones obtained more P repeats was inserted at the C terminus for the angle series (C-terminal insertions). But from the first set of selections were subjected to provide DARPins such as NEEEP..C. Further- both the sequences are referred to as “Prepeats” to error prone PCR to introduce random muta- more, the dimerizing repeats (indicated by un- for the sake of simplicity. tions using the GeneMorph II Random Muta- derline) are placed on the same terminus as the P repeat sequence: DAAGGTPLHEAARAGH- genesis (Agilent). Various amounts of DNA P repeat insertions for maximum impact. In LEIVEVLLKYGADVNAV template were tested to quantify the rate of dimer A_R3/C_R3 (NEEEC), the dimerizing res- P′ repeat sequence: DGTPLHKAARAGHLEI- mutagenesis. The GeneMorph kit was used at idues span over the three ankyrin repeats and VEVLLKYGADVNAV a concentration of 0.1 ng DNA/reaction, to the two capping regions. To generate A_angle_R4, To engineer the “mono-extended” DARPin achieve an average mutation rate of 2.5 to 3 onePrepeatwasinsertedattheC-terminalend series, two sets of EpoR binding repeats were mutations/gene. The PCR product was further (NEEEPC) and the set of dimerizing residues were included on the same DARPin molecule. This amplified and purified by gel electrophoresis for also shifted by one repeat. Similarly, the design for series is based on monomeric extended DARPins, DNA shuffling. A_angle_R5 can summarized as NEEEPPC. and thus named “mono-extended.” To generate

Mohan et al., Science 364, eaav7532 (2019) 24 May 2019 11 of 15 RESEARCH | RESEARCH ARTICLE a series, the two sets of binding repeats were Cell lysis and immunoprecipitation cells. The antibodies used are as follows: anti- positioned at either termini, and the P repeats UT7/EPO cells were starved overnight and stim- EpoR (VP2E8, Abcam), Alexa Fluor 647 Goat are inserted in the middle, for example NEEEPP… ulated with saturating concentrations of EPO anti-mouse (405322, Biolegend), and Alexa Fluor EEEC. The second set of EpoR binding repeats (111 nM) and DARPin ligands (1 mM) for 5 and 647 Goat anti-mouse (A21235, Invitrogen). are positioned similar to the distance series. 15 min at 37°C. Next, the cells were washed Thus, the P’ approach was adopted for the gen- with ice-cold PBS and resuspended in 100 ml/ Crystallization, data collection, eration of the mono-extended series, as well. The 5E6 cells lysis buffer and incubated with rotation and refinement insertion of one to four repeats in the middle for 30 min at 4°C. The Pierce IP lysis buffer Protein complexes were formed by mixing 1:2 showed severe clashes in the models gene- (Thermo Scientific) was supplemented with molar ratios of DARPin and EpoR ECD (or 1:1 rated. As a result, the first member of the series phosphatase inhibitor cocktail (Abcam, 1/100 v/v), DARPin/EpoR in the case of monomeric E2) along consisted of five non-binding repeats, M_R11, Halt phosphatase inhibitor cocktail (Thermo with Carboxypeptidases A and B at 1:100 (w/w), NEEEPPPPP’EEEC. Similarly, the design for Scientific, 1/100 v/v),Sodiumorthovanadate and phenylmethane sulfonyl fluoride at 0.5 mM M_R12 can be summarized as NEEEPPPPPP’EEEC (NEB, 0.5 mM final concentration), and EDTA- final concentration. After overnight incubation with a total of six non-binding repeats inserted. free protease inhibitor cocktail (Roche, 1 tablet in at 4°C, complexes were purified by FPLC on an P repeat sequence: DAAGGTPLHEAARAGH- 10 mL of buffer). The cell lysates were cleared by S200 column, eluted in HBS buffer, and con- LEIVEVLLKYGADVNAV centrifugation at 10 krpm for 15 min at 4°C. The centrated. Crystallization was performed by sitting P’ repeat sequence: DGTPLHKAARAGHLE- samples were denatured by the addition of 30 ml drop vapor diffusion using a Mosquito nanoliter IVEVLLKYGADVNAV sample buffer (0.28 M Tris, pH 6.8; 30% v/v pipetting robot. Monomeric E2-EpoR complex crys- glycerol; 1% w/v SDS; 0.55 M b-mercaptoethanol; tals were grown at 9 mg/mL from 1 M NaH2PO4/ Phospho-flow signaling assays 0.0024% w/v bromophenol blue) and boiling for K2HPO4 pH 5.6. Crystals were cryoprotected by UT7/EPO cells were starved overnight for 5 min, resolved by 12.5% SDS-PAGE, and analyzed addition of ethylene glycol to 25% before har- ~18 hours in base media without EPO. The cells by immunoblotting. vesting for data collection at APS beamline 23 Downloaded from were stimulated with EPO or the DARPin lig- For anti-EpoR immunoprecipitations, anti- ID-B. C_R3 crystals were grown at 17 mg/mL ands for 15 min at 37°C, followed by fixation EpoR antibody VP2E8 (Abcam, 4.6 mg/18E6 cells) from 0.1 M Tris pH 8.5, 25% PEG 3350, 1 mM with paraformaldehyde (Electron Microscopy was added to the clarified cell lysates and in- reduced glutathione, and 1 mM oxidized gluta- Sciences) for 10 min at room temperature. Next, cubated with rotation for 1 hour at 4°C. Next, thione. Crystals were cryoprotected by the addi- the cells were permeabilized for intracellular 30 ml of Pierce protein G magnetic beads (Thermo tion of ethylene glycol to 25% before harvesting

staining by treatment with methanol (Fisher) for Scientific) were added, and incubated with rota- fordatacollectionatSSRLbeamline12-2. http://science.sciencemag.org/ 30 min at -20°C. The cells were then incubated tion for 1 hour at 4°C. The beads were then TheC_R3/EpoRcomplexcrystalsweregrown with the desired antibodies at a 1:50 dilution for washed three time with the supplemented lysis at 6 mg/mL from 0.1 M HEPES pH 7.3 and 1.4 M 1 hour at room temperature. The levels of the buffer. Next, the proteins were eluted by the NaKHPO4. Crystals were cryoprotected by the various phosphorylated proteins correspond to addition of 160 ml sample buffer and boiling for addition of glycerol to 24% before harvesting the measured values of mean fluorescence in- 5 min. The beads were removed by magnetic sep- for data collection at ALS beamline 8.2.2. The tensity (MFI). The background fluorescence of aration. The samples were resolved by 12.5% A_angle_R5/EpoR complex was crystallized at the unstimulated samples was subtracted from SDS-PAGEandanalyzedbyimmunoblotting. 4.5 mg/mL from 0.2 M K/Na tartrate, 0.1 M Bis all the readouts. Data was acquired using CytoFlex, Theantibodiesusedareasfollows:Purified Tris propane pH 8.5, and 20% PEG 3350 and flow cytometer instrument (Beckman Coulter). Mouse Anti-Stat1 (pY701, BD Biosciences), Rabbit cryoprotected by the addition of glycerol to 30% TheMFIvalueswerenormalizedtotheMFI anti-STAT1 42H3 (Cell Signaling Technology), before harvesting for data collection at ALS

of EPO at 37.04 nM and plotted in Prism 7 Purified Mouse Anti-Stat5 (pY694, BD Biosciences), beamline 5.0.1. Crystals of the C_angle_R5/EpoR on May 28, 2019 (GraphPad). The dose-response curves were gen- Rabbit anti-STAT5 3H7 (Cell Signaling Tech- complexweregrownat6mg/mLfrom0.3M erated using the “sigmoidal dose-response” analy- nology), anti-EpoR (VP2E8, Abcam), EpoR (D-5) MgSO4, 0.1 M Bis-Tris Propane pH 7.0, and 22.5% sis. A Gaussian distribution analysis was excluded HRP (sc365662, Santa Cruz Biotechnology), anti- PurePEGS cocktail (Anatrace) and cryoprotected as only some ligands demonstrate a decrease in JAK2 (EPR108[2], Abcam), anti-phospho(Y1007 + by the addition of ethylene glycol to 30% before MFI values at high concentrations. All the anti- Y1008) JAK2 (E132, Abcam), phosphor-EpoR Y426 harvesting for data collection at ALS beamline bodies used were purchased from BD Biosciences: (PA5-40305, Thermo Fisher Scientific), phospho- 5.0.1. The A_dist_R7/EpoR complex was crystal- Alexa Fluor 488 Mouse Anti-Stat1 (pY701), Alexa EpoR Y368 (PA5-38483, Thermo Fisher Scientific), lized at 8 mg/mL from 0.2 M NaCl, 0.1 M Tris pH Fluor 647 Mouse Anti-Stat3 (pY705), Alexa Fluor phospho-EpoR Y456 (PA5-64795, Thermo Fisher 8.5, and 25% PEG 3350 and harvested without 488 Mouse Anti-Stat3 (pY705), Alexa Fluor 647 Scientific), swine anti-rabbit immunoglobulins/ further cryoprotection for data collection at ALS Mouse Anti-Stat5 (pY694), and Alexa Fluor 488 HRP (P0399, Dako), and rabbit anti-mouse im- beamline 8.2.1. Initial crystals of the M_R12/ Mouse Anti-ERK1/2 (pT202/pY204). munoglobulins/HRP (P0161, Dako). EpoR complex were grown at 10 mg/mL from 0.2 M NaCl, 0.1 M CAPS pH 10.5, and 20% Phosphorylated protein analysis Receptor internalization PEG 8000. These crystals were pulverized to UT7/EPO cells were starved overnight and stim- UT7/EPO cells were starved overnight and then form a seed stock for subsequent seeding into ulated with saturating concentrations of EPO treated with 15 mg/mL cycloheximide (Sigma) for 0.2 M ammonium tartrate and 20% PEG 3350. and DARPin ligands for 15, 60, and 120 min at 4 hours. Next, the cells were stimulated with The resulting crystals were dehydrated for 1 month 37°C, followed by fixationwithparaformaldehyde EPO (111 nM) or the DARPin ligands (1 mM) for by increasing the reservoir concentration to for 10 min at room temperature. Next, the cells 15, 60, or 120 min at 37°C. The ligand solutions 25% PEG 3350. The crystals were cryoprotected were washed and resuspended in 0.5% bovine were also prepared in media containing cyclo- by addition of 10% glycerol and 10% ethylene serumalbumininPBS,andstoredat–80°C. The heximide. The cells were then washed with ice- glycol before harvesting for data collection at cells were then prepared for staining according cold PBS and incubated with anti-EpoR antibody APS beamline 23ID-B. All datasets were integ- to standard protocols. The cells were stained with for 1 hour at room temperature. Next, the cells rated and scaled using XDS (46)beforemer- a panel of antibodies (Primity Bio Pathway Pheno- were incubated with Alexa Fluor 647 goat anti- ging symmetry-related reflections with aimless typing service) and analyzed on LSRII Flow mouse antibody at a 1:100 dilution for 1 hour at (47, 48).DuetoanisotropyoftheC_R3/EpoR, cytometer (Becton Dickinson). The levels of phos- room temperature. The levels of cell-surface A_angle_R5/EpoR, C_angle_R5/EpoR, and M_R12/ phorylated protein were calculated as the Log EpoR correspond to the measured values of EpoR datasets, final ellipsoidal truncation was (base2)ratiooftheMFIofstimulatedsamplesto mean fluorescence intensity (MFI). The MFI performed using STARaniso before merging unstimulated samples. values were normalized to the MFI of untreated reflections (49).

Mohan et al., Science 364, eaav7532 (2019) 24 May 2019 12 of 15 RESEARCH | RESEARCH ARTICLE

Structures were solved by molecular replace- was preserved from M_R12. For A_dist, the rectly correspond to the relationship between ment with Phaser (50) using the crystal structure C-terminal EpoR binding mode was preserved the transmembrane or intracellular portions of of EpoR [ (PDB) ID 1ERN], which included the P′ shortened loop modifica- EpoR in the cellular context. split into D1 and D2 domains, and the predicted tion which was impactful for binding. Each DARPin structures as search models, or in the DARPin/EpoR complex was aligned to chain Single molecule fluorescence imaging case of the C_R3/EpoR complex using the re- A of dimers A (pdb id 5HRY) and C (pdb id 5KBA). Sample preparation fined C_R3 crystal structure. With the exceptions The synthetic signaling complex was generated HeLa cells were transiently transfected by cal- noted below, crystallographic models were pre- in the Rosetta macromolecular modeling suite cium chloride precipitation with 10 mgofpSems pared by iterative rounds of manual rebuilding using a symmetry definition file to generate the neo leader mXFP EpoR. In principle, a fluorescent- and reciprocal space refinement with COOT C2 symmetry starting from each aligned DARPin/ dead Green Fluorescent Protein (meGFP-Y66F/ (51) and Phenix (52–54). For the C_R3/EpoR, EpoR complex. The experimentally tested se- mXFP) was N-terminally fused to EpoR. After C_angle_R5/EpoR, and M_R12/EpoR complexes, quences were threaded to each backbone using transfection, cells were seeded on a PLL-PEG-RGD additional refinement with Buster (55, 56)was the Rosetta packing algorithm with the default coated glass cover slide (61). Imaging experiments performed to aid convergence of the structure talaris2014 score function followed by side-chain were typically carried out 24h post transfection. before final refinement in Phenix. TLS refine- energy minimization. Labeling, ligand incubation and subsequent imag- ment was performed for all structures except ing were performed in a custom-made incubation for the C_R3 dimer, for which anisotropic atomic Relax of EpoR clashing models chamber with a volume of 500 ml, which was displacement parameters (ADPs) were refined, Models that presented clashes between the mounted on the microscope. The experiments and A_dist_R7/EpoR. Individual ADPs were re- membrane-proximal regions of the EpoR ECD presented here were performed at room temper- fined for all structures except that grouped ADPs were identified by visual inspection. An analysis ature in the presence of media without phenol wererefinedforM_R12/EpoRandADPswere oftheEpoRconformationinallthecrystalstruc- red. Additionally, each sample was supplemented not refined for A_dist_R7/EpoR. For the low- tures obtained singled residues 113-120 as a hinge with an oxygen scavenger and a redox-active Downloaded from resolution A_dist_R7/EpoR structure, after molec- region which allows for conformational flexibility photo protectant to minimize photo bleaching ular replacement the structure was manually betweenthebindingdomainandthemembrane (62). For cell surface labeling of mXFP-tagged edited in COOT to correct the se- proximal domain. A Rosetta relax trajectory in EpoR DY647- and Rho11-labeled Nanobodies quence and to delete loop 4 of A_dist_R7, which search of a state with an energy minimum al- (NBs) were added in equal concentrations (2 nM was present in the search model but absent in lowing backbone torsion angles to be modeled each) and incubated for at least 5 min. High equi-

the purified protein. Model refinement was then flexibly in this region while keeping the rest of librium binding of the labeled NBs was ensured http://science.sciencemag.org/ completed by rigid body refinement in Phenix the construct fixed using coordinate constraints by keeping these in bulk solution. Co-locomotion of and jelly body refinement in REFMAC (57–59). was carried out for each clashing model. In the EpoRwasprobedbeforeandafterincubatinglig- Software used in this project was installed and case of dimer A_R3, the first attempt with Rosetta ands at indicated concentrations for at least 10 min. configured by SBGrid (60). Crystallographic data relax to relieve the severe clashes between EpoR collection and refinement statistics along with subunits was unsuccessful without having to Single molecule microscopy PDB deposition codes are reported in tables S1 make significant moves to the head region or Single molecule imaging was performed by TIRF and S2. rearranging the designed dimer interface orienta- microscopy. As described previously (20, 24)an tion, so the EpoR conformation from A_angle_R5 inverted microscope (Olympus IX71) equipped Generation of models was used instead for the relax trajectory. The with a triple-line total internal reflection (TIR) The synthetic EpoR agonists described in this relaxed output EpoR conformation from dimer condenser (Olympus) and a back-illuminated

work have a unique property that differentiate A_R3 was used to rescue a non-clashing model electron multiplied (EM) CCD camera (iXon on May 28, 2019 them from previous efforts at generating syn- for the A_angle_R4 complex. Although the gen- DU897D, Andor Technology) was used. The thetic type I cytokine agonists, namely they have erated non-clashing models do not represent sample was TIR illuminated via a 150x magnifi- fairly well-defined and rigid conformations that unique solutions, they demonstrate the steric cation objective with a numerical aperture of 1.45 provide a structural basis for each observed plausibility of a given complex. (UAPO 150 x /1.45 TIRFM Olympus). Rho11 was physiological output. This advantage was lever- excited by a 561 nm laser (CrystaLaser) at 0.95 mW aged by generating a crystallographic structure- Structural analysis of complex models (~32 W/cm2) and DY647 by a 642 nm laser informed set of signaling complex models for each Although the synthetic agonists themselves are (Omicron) at 0.65 mW (~22W/cm2). For fluores- of the A_dist_Rx, A_angle_Rx, and C_angle_Rx rather rigid, the EpoR ECD has some intrinsic cent detection a spectral image splitter (Dual- series. To best recapitulate both the backbone flexibility between the D1 and D2 domains giving View, Optical Insight) with a 640 DCXR dichroic topology of the P repeat and the N-/C-terminal rise to uncertainty in measurements derived from beam splitter (Chroma) combined with a 580/40 mode of EpoR binding, the backbone coordinates models. Indeed, multiple EpoR conformers were (Semrock, Rho11 detection) and a 690/70 (Chroma, from the M_R12/EpoR crystal structure were used observed in the crystal structures obtained for this DY647 detection) bandpass filters has been used to construct each experimentally tested member work (fig. S10). Because the DARPin-EpoR D1 in- dividing each emission channel into 512 × 256 of the extension series using the Rosetta macro- teraction is rigid, we made all distance and angle pixel. For each cell image stacks of 150 frames molecular modeling suite. While most presented measurements using Val119 of EpoR, at the hinge were recorded with temporal resolution of ap- non-clashing complex models upon superposi- between D1 and D2. Distance measurements be- proximately 32 ms/frame. tion of the E2-EpoR structure with the extended tween EpoR ECDs were made between the Cas backbone, several cases showed significant clashes of the residue Val119. Scissor half angle measure- Single molecule analyses between EpoR subunits that could not be physi- ments were taken using Val119 of EpoR as the The workflow of single molecule analyses is cally feasible. We used a constrained minimization vertex with rays extending through Asp2 at the outlined in fig. S6A. Single molecule localization protocol on each of the clashing complex models N-terminal cap of the DARPins and through was assessed by using multiple-target tracing to generate a complete set of plausible models. an alanine at the start of the C-terminal cap of (MTT) algorithm (63) and tracking was per- the DARPin (due to varying insertion lengths formed, as described previously (64). Because Backbone and symmetric complex generation in DARPins with X repeats, this was residue num- immobile molecules (typically ~10% of the entire Backbone coordinates for each DARPin extension ber 39 + 33X). Although the measurements dataset) were mostly caused by either non-specific length were derived from the M_R12/EpoR struc- presented (Figs. 3 and 4) represent the general binding of labeled NBs to the coverslip or by ture while keeping capping modules intact. For trend of complex geometries within the ex- endosomal uptake of receptors, these were iden- A/C_angle, the N-terminal EpoR binding mode tension series, these measurements do not di- tified by spatiotemporal cluster analysis (65)and

Mohan et al., Science 364, eaav7532 (2019) 24 May 2019 13 of 15 RESEARCH | RESEARCH ARTICLE

removed from the dataset. Dual-color co-tracking Anti-STAT3 (pY705; 1:20 dilution), or Alexa Fluor- 22. M. D. Ballinger, J. A. Wells, Will any dimer do? Nat. Struct. Biol. analysis was carried out as described recently 647 Mouse Anti-STAT1 (pY701; 1:20 dilution). 5, 938–940 (1998). doi: 10.1038/2911; pmid: 9808034 23. J. Staerk ., Orientation-specific signalling by (20). Prior to each experiment, Rho11 and DY647 Data were acquired using either BD Accuri C6 et al receptor dimers. EMBO J. 30, 4398–4413 channels were aligned with sub-pixel precision Cytometer or Canto instruments (Becton, Dick- (2011). doi: 10.1038/emboj.2011.315; pmid: 21892137 based on a calibration with multicolor fluores- inson and Company) and all analysis was per- 24. I. Moraga et al., Tuning cytokine receptor signaling by re-orienting cent beads (TetraSpeck microspheres 0.1 mm, formed using FlowJo. dimer geometry with surrogate ligands. Cell 160, 1196–1208 (2015). doi: 10.1016/j.cell.2015.02.011;pmid:25728669 Invitrogen). Individual molecules detected in both 25. M. Kai et al., Switching constant domains enhances agonist spectral channels within the same frame and activities of antibodies to a . within a distance threshold of 100 nm were con- REFERENCES AND NOTES Nat. Biotechnol. 26, 209–211 (2008). doi: 10.1038/nbt1376; sidered as co-localized. In order to reconstruct 1. M. Akdis et al., Interleukins (from IL-1 to IL-38), , pmid: 18157117 transforming growth factor , and TNF- : Receptors, functions, 26. H. Zhang ., Selecting agonists from single cells co-locomotion trajectories (co-trajectories) from b a et al and roles in diseases. J. Allergy Clin. Immunol. 138, 984–1010 infected with combinatorial antibody libraries. Chem. Biol. the identified population of co-localizing par- (2016). doi: 10.1016/j.jaci.2016.06.033; pmid: 27577879 20,734–741 (2013). doi: 10.1016/j.chembiol.2013.04.012; ticles, the tracking algorithm was applied to 2. G. M. Delgoffe, P. J. Murray, D. A. Vignali, Interpreting mixed pmid: 23706638 the co-localization dataset. Co-trajectories with a signals: The cell’s cytokine conundrum. Curr. Opin. Immunol. 27. K. Nakano et al., Effective screening method of agonistic , 632–638 (2011). doi: 10.1016/j.coi.2011.07.013; diabodies based on autocrine growth. minimum length of 10 consecutive steps (320 ms) 23 J. Immunol. Methods pmid: 21852079 347,31–35 (2009). doi: 10.1016/j.jim.2009.05.012; were considered as receptor dimers. The relative 3. J. B. Spangler, I. Moraga, J. L. Mendoza, K. C. Garcia, pmid: 19508873 fraction of co-locomotion was calculated with the Insights into cytokine-receptor interactions from cytokine 28. A. Plückthun, Designed ankyrin repeat proteins (DARPins): absolute number of trajectories from both chan- engineering. Annu. Rev. Immunol. 33, 139–167 (2015). Binding proteins for research, diagnostics, and therapy. , 489–511 (2015). doi: 10.1146/ nels and was corrected for stochastically double doi: 10.1146/annurev-immunol-032713-120211; pmid: 25493332 Annu. Rev. Pharmacol. Toxicol. 55 4. K. Kaushansky, Lineage-specific hematopoietic growth factors. annurev-pharmtox-010611-134654; pmid: 25562645 labeling with the same fluorophore within a N. Engl. J. Med. 354, 2034–2045 (2006). doi: 10.1056/ 29. J. A. Fallas et al., Computational design of self-assembling dimer.Therelativefractionofco-locomoting NEJMra052706; pmid: 16687716 cyclic protein homo-oligomers. Nat. Chem. 9, 353–360 (2017). particles was assessed with respect to the ab- 5. R. M. Stroud, J. A. Wells, Mechanistic diversity of cytokine doi: 10.1038/nchem.2673; pmid: 28338692 Downloaded from 30. H. K. Binz, M. T. Stumpp, P. Forrer, P. Amstutz, A. Plückthun, solute number of tracked particles in both chan- receptor signaling across cell membranes. Sci. STKE 2004, re7 (2004). pmid: 15126678 Designing repeat proteins: Well-expressed, soluble and stable nels and corrected for stochastically double-labeled 6. R. S. Syed et al., Efficiency of signalling through cytokine proteins from combinatorial libraries of consensus ankyrin dimers by the same fluorophore species: receptors depends critically on receptor orientation. Nature repeat proteins. J. Mol. Biol. 332, 489–503 (2003). 395, 511–516 (1998). doi: 10.1038/26773; pmid: 9774108 doi: 10.1016/S0022-2836(03)00896-9; pmid: 12948497 rel. co – locomotion = 7. M. Socolovsky, A. E. Fallon, H. F. Lodish, The receptor 31. H. K. Binz et al., High-affinity binders selected rescues EpoR-/- erythroid progenitors and replaces EpoR from designed ankyrin repeat protein libraries. 2 AB 1 in a synergistic interaction with c-kit. Blood 92, 1491–1496 Nat. Biotechnol. 22,575–582 (2004). doi: 10.1038/nbt962; http://science.sciencemag.org/ hi (1998). pmid: 9716574 pmid: 15097997 A þ B A B 8. O. Livnah ., Functional mimicry of a protein hormone by a 32. N. Komatsu ., Establishment and characterization of a 2 A þ B A þ B et al et al peptide agonist: The EPO receptor complex at 2.8 A. Science human leukemic cell line with megakaryocytic features: 273, 464–471 (1996). doi: 10.1126/science.273.5274.464; Dependency on granulocyte-macrophage colony-stimulating Where A, B, and AB are the absolute number pmid: 8662530 factor, interleukin 3, or erythropoietin for growth and survival. of trajectories observed for Rho11 and DY647 as 9. M. A. Lemmon, J. Schlessinger, Cell signaling by receptor Cancer Res. 51, 341–348 (1991). pmid: 1824823 well as co-trajectories, respectively (first part of tyrosine kinases. Cell 141, 1117–1134 (2010). doi: 10.1016/ 33. J. Li, A. Mahajan, M.-D. Tsai, Ankyrin repeat: A unique motif the product determines the relative co-locomotion j.cell.2010.06.011; pmid: 20602996 mediating protein-protein interactions. Biochemistry 45, 10. J. W. Adamson, : Into the new millennium. 15168–15178 (2006). doi: 10.1021/bi062188q; pmid: 17176038 while the second part determines the correction Semin. Oncol. 25,76–79 (1998). pmid: 9671336 34. A. Z. Ebie, K. G. Fleming, Dimerization of the erythropoietin factor for stochastically double labeling). 11. I. Moraga, J. Spangler, J. L. Mendoza, K. C. Garcia, Multifarious receptor transmembrane domain in micelles. J. Mol. Biol. determinants of cytokine receptor signaling specificity. 366, 517 –524 (2007). doi: 10.1016/j.jmb.2006.11.035; Human primary cell culture and analysis Adv. Immunol. 121,1–39 (2014). doi: 10.1016/ pmid: 17173930

on May 28, 2019 + B978-0-12-800100-4.00001-5; pmid: 24388212 35. S. N. Constantinescu et al., The erythropoietin receptor CD34 hematopoietic stem and progenitor cells 12. K. Ozaki, W. J. Leonard, Cytokine and cytokine receptor transmembrane domain mediates complex formation from mobilized peripheralbloodwerepurchased pleiotropy and redundancy. J. Biol. Chem. 277, 29355–29358 with viral anemic and polycythemic gp55 proteins. from the Fred Hutchinson Cancer Research (2002). doi: 10.1074/jbc.R200003200; pmid: 12072446 J. Biol. Chem. 278, 43755–43763 (2003). doi: 10.1074/ Center. Primary cells were cultured as previously 13. J. T. Sockolosky et al., Selective targeting of engineered jbc.M302974200; pmid: 12930840 T cells using orthogonal IL-2 cytokine-receptor complexes. 36. F. C. Giani ., Targeted application of human genetic described (18). Cells were cultured with one of et al Science 359, 1037–1042 (2018). doi: 10.1126/science.aar3246; variation can improve red blood cell production from stem following ligands: EPO (3 U/mL), A_R3 (10 nM), pmid: 29496879 cells. Cell Stem Cell 18,73–78 (2016). doi: 10.1016/ A_angle_R4 (1 mM), A_angle_R5 (1 mM), A_ 14. J. B. Mumm et al., IL-10 elicits IFNg-dependent tumor immune j.stem.2015.09.015; pmid: 26607381 angle_R6 (1 mM), C_R3 (10 nM), C_angle_R4 surveillance. Cancer Cell 20, 781–796 (2011). doi: 10.1016/ 37. J. Hu et al., Isolation and functional characterization of human j.ccr.2011.11.003; pmid: 22172723 erythroblasts at distinct stages: Implications for understanding (100 nM), C_angle_R5 (100 nM), C_angle_R6 15. C. C. M. Ho et al., Decoupling the functional pleiotropy of stem of normal and disordered erythropoiesis in vivo. Blood (100 nM), A_dist_R5 (333 nM), and A_dist_R7 cell factor by tuning c-Kit signaling. Cell 168, 1041–1052.e18 121, 3246–3253 (2013). doi: 10.1182/blood-2013-01-476390; (1 mM). Cell concentration was measured using (2017). doi: 10.1016/j.cell.2017.02.011; pmid: 28283060 pmid: 23422750 an automated cell counter (Beckman Coulter). 16. C. Thomas et al., Structural linkage between ligand 38. D. Metcalf, Hematopoietic cytokines. Blood 111, 485–491 discrimination and receptor activation by type I interferons. (2008). doi: 10.1182/blood-2007-03-079681; To evaluate erythroid differentiation, cells were Cell 146, 621–632 (2011). doi: 10.1016/j.cell.2011.06.048; pmid: 18182579 stained with CD71-Fluorescein Isothiocyanate pmid: 21854986 39. U. Testa, Apoptotic mechanisms in the control of (FITC), CD11b/ CD41a-Phycoerythrin (PE), and 17. S. Mitra et al., Interleukin-2 activity can be fine tuned with erythropoiesis. Leukemia 18, 1176–1199 (2004). doi: 10.1038/ CD235a-Allophycocyanin (APC). Propidium Iodide engineered receptor signaling clamps. Immunity 42, 826–838 sj.leu.2403383; pmid: 15208642 (2015). doi: 10.1016/j.immuni.2015.04.018; pmid: 25992859 (PI; 1:1,000) was used to discriminate live and 40. D. Winpenny, M. Clark, D. Cawkill, Biased ligand quantification 18. D. M. Freed et al., EGFR ligands differentially stabilize receptor in drug discovery: From theory to high throughput screening dead cells. All antibodies were used at a 1:20 di- dimers to specify signaling kinetics. Cell 171, 683–695.e18 to identify new biased m agonists. lution, unless otherwise noted. (2017). doi: 10.1016/j.cell.2017.09.017; pmid: 28988771 Br. J. Pharmacol. 173, 1393–1403 (2016). doi: 10.1111/ For phospho-STAT staining, differentiated 19. S. W. Rowlinson et al., An agonist-induced conformational bph.13441; pmid: 26791140 primary erythroid cells were starved in cytokine- change in the growth determines the choice 41. M. J. Waters, A. J. Brooks, JAK2 activation by growth hormone of signalling pathway. Nat. Cell Biol. 10, 740–747 (2008). and other cytokines. Biochem. J. 466,1–11 (2015). free media (1% BSA in IMDM) for 4 hours and doi: 10.1038/ncb1737; pmid: 18488018 doi: 10.1042/BJ20141293; pmid: 25656053 stimulated either with EPO (3U/mL), A_R3 20. A. R. Kim et al., Functional selectivity in cytokine signaling 42. R. D. Ferrao, H. J. Wallweber, P. J. Lupardus, Receptor- (10nM),orA_dist_R5(333nM)for1hour. revealed through a pathogenic EPO mutation. Cell mediated dimerization of JAK2 FERM domains is required for Treated cells were fixed and permeabilized as 168, 1053–1064.e15 (2017). doi: 10.1016/j.cell.2017.02.026; JAK2 activation. eLife 7, e38089 (2018). doi: 10.7554/ pmid: 28283061 eLife.38089; pmid: 30044226 previously described (18). Then, cells were 21. J. W. Wisler, K. Xiao, A. R. Thomsen, R. J. Lefkowitz, Recent 43. G. Chao et al., Isolating and engineering human antibodies stained with Alexa Fluor-647 Mouse Anti-STAT5 developments in biased agonism. Curr. Opin. Cell Biol. 27, using yeast surface display. Nat. Protoc. 1, 755–768 (2006). (pY694; 1:20 dilution), Alexa Fluor-647 Mouse 18–24 (2014). doi: 10.1016/j.ceb.2013.10.008; pmid: 24680426 doi: 10.1038/nprot.2006.94; pmid: 17406305

Mohan et al., Science 364, eaav7532 (2019) 24 May 2019 14 of 15 RESEARCH | RESEARCH ARTICLE

44. A. J. Meyer, J. W. Ellefson, A. D. Ellington, Curr. Protoc. 58. G. N. Murshudov et al., REFMAC5 for the refinement of Source is a DOE Office of Science User Facility under contract Mol. Biol. 105, 15.21.1–15.12.7 (2014). macromolecular crystal structures. Acta Crystallogr. D Biol. DE-AC02-05CH11231. The Pilatus detector on 5.0.1 was funded 45. J. Schilling, J. Schöppe, A. Plückthun, From DARPins to Crystallogr. 67, 355–367 (2011). doi: 10.1107/ under NIH grant S10OD021832. Use of the Stanford Synchrotron LoopDARPins: Novel LoopDARPin design allows the selection S0907444911001314; pmid: 21460454 Radiation Lightsource (SSRL), SLAC National Accelerator of low picomolar binders in a single round of ribosome display. 59. O. Kovalevskiy, R. A. Nicholls, G. N. Murshudov, Automated Laboratory, is supported by the DOE Office of Science, Office of J. Mol. Biol. 426, 691–721 (2014). doi: 10.1016/ refinement of macromolecular structures at low resolution Basic Energy Sciences under contract DE-AC02-76SF00515. j.jmb.2013.10.026; pmid: 24513107 using prior information. Acta Crystallogr. D Struct. Biol. The SSRL Structural Molecular Biology Program is supported 46. W. Kabsch, XDS. Acta Crystallogr. D Biol. Crystallogr. 72,1149–1161 (2016). doi: 10.1107/S2059798316014534; by the DOE Office of Biological and Environmental Research 66, 125–132 (2010). doi: 10.1107/S0907444909047337; pmid: 27710936 and by the NIH and NIGMS (including P41GM103393). Author pmid: 20124692 60. A. Morin et al., Cutting edge: Collaboration gets the most out of contributions: K.C.G. conceived of the project. K.M., G.U., J.A.F., 47. P. R. Evans, An introduction to data reduction: Space-group software. eLife 2, e01456 (2013). doi: 10.7554/eLife.01456; D.B., and K.C.G. jointly developed the approach for tuning receptor determination, scaling and intensity statistics. Acta Crystallogr. pmid: 24040512 engagement geometry with extendable scaffolds through an D Biol. Crystallogr. 67, 282–292 (2011). doi: 10.1107/ 61. S. Wilmes et al., Receptor dimerization dynamics as a iteration of computational design (G.U., J.A.F., and D.B.) and S090744491003982X; pmid: 21460446 regulatory valve for plasticity of type I signaling. protein engineering (K.M. and K.C.G.). K.M. designed and constructed the DARPin libraries, performed yeast-display 48. M. D. Winn et al., Overview of the CCP4 suite and current J. Cell Biol. 209, 579–593 (2015). doi: 10.1083/jcb.201412049; selections and affinity maturation, designed and generated the developments. Acta Crystallogr. D Biol. Crystallogr. pmid: 26008745 sequences for the extended DARPins, expressed recombinant 67, 235–242 (2011). doi: 10.1107/S0907444910045749; 62. J. Vogelsang et al., A reducing and oxidizing system pmid: 21460441 minimizes photobleaching and blinking of fluorescent dyes. proteins, biophysically and functionally characterized all the DARPins, and performed immunoprecipitation and protein 49. J. F. Bruhn et al., Crystal structure of the Marburg virus VP35 Angew.Chem.Int.Ed.47, 5465–5469 (2008). doi: 10.1002/ immunoblotting experiments. G.U. and J.A.F. designed the initial oligomerization domain. J. Virol. 91, e01085-16 (2017). anie.200801518;pmid:18601270 doi: 10.1128/JVI.01085-16; pmid: 27847355 63. A. Sergé, N. Bertaux, H. Rigneault, D. Marguet, Dynamic two dimer interfaces. G.U. worked with K.M. to design sequences for the extended DARPin/EpoR complexes and generated three- 50. A. J. McCoy et al., Phaser crystallographic software. multiple-target tracing to probe spatiotemporal cartography of dimensional models of these complexes. A.R.K. performed the J. Appl. Crystallogr. 40, 658–674 (2007). doi: 10.1107/ cell membranes. Nat. Methods 5, 687–694 (2008). S0021889807021206; pmid: 19461840 doi: 10.1038/nmeth.1233; pmid: 18604216 erythropoiesis experiments. K.M., K.M.J., and Y.G. crystallized the protein complexes. K.M.J. and Y.G. collected the crystallography 51. P. Emsley, B. Lohkamp, W. G. Scott, K. Cowtan, Features 64. K. Jaqaman et al., Robust single-particle tracking in live-cell data and determined structures. M.H. performed the TIRF and development of Coot. Acta Crystallogr. D Biol. Crystallogr. time-lapse sequences. Nat. Methods 5, 695–702 (2008). microscopy experiments. Y.M. provided the reference proteins Downloaded from 66, 486–501 (2010). doi: 10.1107/S0907444910007493; doi: 10.1038/nmeth.1237; pmid: 18641657 for SPR. R.A.S. assisted with immunoblot analysis. J.P. supervised pmid: 20383002 65. F. Roder, S. Wilmes, C. P. Richter, J. Piehler, Rapid transfer the TIRF microscopy experiments. V.G.S. supervised the 52. P. D. Adams et al., PHENIX: A comprehensive Python-based of transmembrane proteins for single molecule dimerization erythropoiesis experiments. K.C.G. and D.B. supervised the project system for macromolecular structure solution. Acta Crystallogr. assays in polymer-supported membranes. ACS Chem. Biol. and interpreted the data. K.M. and K.C.G. wrote the manuscript D Biol. Crystallogr. 66, 213–221 (2010). doi: 10.1107/ 9,2479–2484 (2014). doi: 10.1021/cb5005806; with inputs from all authors. Competing interests: K.C.G., D.B., S0907444909052925; pmid: 20124702 pmid: 25203456 K.M., G.U., K.M.J., and J.A.F. are inventors on U.S. patent 53. T. C. Terwilliger ., Iterative model building, structure et al application 62/819,645 titled “topological control of receptor refinement and density modification with the PHENIX ACKNOWLEDGMENTS

signaling using synthetic homo- and hetero-dimeric cytokine http://science.sciencemag.org/ AutoBuild wizard. Acta Crystallogr. D Biol. Crystallogr. We thank A. Velasco for assistance. We thank I. Tickle for mimetics.” Data and materials availability: Coordinates and ,61–69 (2008). doi: 10.1107/S090744490705024X; 64 assistance with x-ray data processing. Funding: This work was structure factors of the complex have been deposited in the pmid: 18094468 supported by NIH R01-AI51321 (K.C.G.), NIH T32HL066987 Protein Data Bank with accession codes 6MOE, 6MOG, 6MOH, 54. P. V. Afonine ., Towards automated crystallographic et al (A.R.K.), NIH R01 DK103794, and R33 HL120791 (V.G.S.); Deutsche 6MOI, 6MOJ, 6MOK, and 6MOL. Diffraction images have been structure refinement with phenix.refine. Acta Crystallogr. Forschungsgemeinschaft (DFG) SFB 944 (J.P.); a gift from the deposited in the SBGrid Data Bank. All other data are available in , 352–367 (2012). doi: 10.1107/ D Biol. Crystallogr. 68 Mathers Fund (K.C.G); and the Howard Hughes Medical Institute the manuscript or the supplementary materials. S0907444912001308; pmid: 22505256 (HHMI) (D.B. and K.C.G.). GM/CA@APS has been funded in whole 55. G. Bricogne et al., BUSTER version 2.10.3. Cambridge, United or in part with federal funds from the National Cancer Institute Kingdom Glob. Phasing Ltd (2016). (ACB-12002) and the National Institute of General Medical SUPPLEMENTARY MATERIALS 56. O. S. Smart et al., Exploiting structure similarity in refinement: Sciences (NIGMS) (AGM-12006). This research used resources of science.sciencemag.org/content/364/6442/eaav7532/suppl/DC1 Automated NCS and target-structure restraints in BUSTER. the Advanced Photon Source, a U.S. Department of Energy (DOE) Figs. S1 to S18 Acta Crystallogr. D Biol. Crystallogr. 68, 368–380 (2012). Office of Science User Facility operated for the DOE Office of Tables S1 to S5 doi: 10.1107/S0907444911056058; pmid: 22505257 Science by Argonne National Laboratory under contract DE-AC02- Reference (66)

57. R. A. Nicholls, F. Long, G. N. Murshudov, Low-resolution 06CH11357. The Eiger 16M detector was funded by an NIH–Office Movies S1 and S2 on May 28, 2019 refinement tools in REFMAC5. Acta Crystallogr. D Biol. of Research Infrastructure Programs, High-End Instrumentation Crystallogr. 68, 404–417 (2012). doi: 10.1107/ Grant (1S10OD012289-01A1). The Berkeley Center for Structural 16 October 2018; accepted 8 April 2019 S090744491105606X; pmid: 22505260 Biology is supported in part by HHMI. The Advanced Light 10.1126/science.aav7532

Mohan et al., Science 364, eaav7532 (2019) 24 May 2019 15 of 15 Topological control of cytokine receptor signaling induces differential effects in hematopoiesis Kritika Mohan, George Ueda, Ah Ram Kim, Kevin M. Jude, Jorge A. Fallas, Yu Guo, Maximillian Hafer, Yi Miao, Robert A. Saxton, Jacob Piehler, Vijay G. Sankaran, David Baker and K. Christopher Garcia

Science 364 (6442), eaav7532. DOI: 10.1126/science.aav7532

Exploring a range of signaling Cytokines are small proteins that bind to the extracellular domains of transmembrane receptors to activate

signaling pathways inside the cell. They often act by dimerizing their receptors, and changes in dimer orientation of the Downloaded from extracellular domains can change the signaling output. Mohan et al. systematically explored this tuning effect by designing a series of dimer ligands for the erythropoietin receptor in which they varied the distance and angle between monomers. The topology affected the strength of activation and differentially affected different pathways, which raises the potential for exploiting such ligands in medicinal chemistry. Science, this issue p. eaav7532 http://science.sciencemag.org/

ARTICLE TOOLS http://science.sciencemag.org/content/364/6442/eaav7532

SUPPLEMENTARY http://science.sciencemag.org/content/suppl/2019/05/22/364.6442.eaav7532.DC1 MATERIALS

on May 28, 2019 REFERENCES This article cites 65 articles, 13 of which you can access for free http://science.sciencemag.org/content/364/6442/eaav7532#BIBL

PERMISSIONS http://www.sciencemag.org/help/reprints-and-permissions

Use of this article is subject to the Terms of Service

Science (print ISSN 0036-8075; online ISSN 1095-9203) is published by the American Association for the Advancement of Science, 1200 New York Avenue NW, Washington, DC 20005. 2017 © The Authors, some rights reserved; exclusive licensee American Association for the Advancement of Science. No claim to original U.S. Government Works. The title Science is a registered trademark of AAAS.