Search
Close this search box.

i-shaped antibody engineering enables conformational tuning of biotherapeutic receptor agonists – Nature Communications

Molecular cloning

Antibody clones against each target were produced from various sources. anti-OX40, anti-4-1BB, anti-DR4, and anti-DR5 antibodies were discovered internally via mouse immunization campaigns. Sequences for all anti-CD40 antibodies used in this work were derived from publicly available databases and patent literature. The anti-IL-2Rβ and anti-IL-2Rγ antibodies used in this study were discovered in-house using a synthetic, fully human scFv yeast display library, as described below.

Gene fragments encoding all antibody constructs were synthesized as gBlocks or eBlocks (IDT) and cloned into the pRK mammalian expression vector using Gibson assembly (NEB, cat#E2611L). The pRK vector contains a cytomegalovirus (CMV) enhancer and promotor to control gene expression, an N-terminal secretion signal (MGWSCIILFLVATATGVHS), a C-terminal simian virus 40 (SV40) PolyA sequence, and an ampicillin resistance gene for bacterial selection. Unless otherwise stated, all Fc regions were human IgG1 with the effectorless mutations L234A/L234A/P329G (LALAPG; EU numbering). The contorsbodies were constructed by fusing the heavy chain and light chain Fab domains to the N- and C-termini of the Fc domain, respectively, as previously described21. In short, the C-terminus of the heavy chain (C220) was fused to the N-terminus of the Fc domain (D221), and the N-terminus of the light chain was fused to the C-terminus of the Fc domain (K447). Each fusion domain was separated by a flexible (G4S)2 linker.

To make hexameric 3C8, heavy chain variable regions were cloned into an hIgG1 backbone containing the previously described E345R/E430G/S440Y (RGY) mutations20,24,32. As the hexameric IgG exists in equilibrium, no additional purification measures were taken to isolate the desired species. To promote formation of the h2B isoform of IgG2 for the anti-CD40 antibodies, the C131S mutation was used (EU numbering, corresponds to the C127S mutation of White and colleagues)23. Fab constructs consisted of the light chain paired with a truncated heavy chain (1-225, EU numbering) and a C-terminal TEV protease-cleavage site, and a Flag tag. For all bispecific antibodies, knob-in-hole mutations were introduced into the Fc-domain to enable heterodimerization and prepared as described below65.

OX40 ECD (L29-D170) was cloned into the pRK mammalian expression vector with a TEV protease-cleavable N-terminal His tag. The IL-2Rβ ECD (A27-T240) and IL-2Rγ (L23-A262) extracellular domains (ECDs) were cloned into the pRK mammalian expression vector with a C-terminal His tag.

iAb engineering

In this work, the iAb conformation was induced in antibodies of interest through engraftment of specific sets of mutations (Fig. 1b and S1). The residue set used to induce domain exchange (iAbdx) was inspired by previous structural and mutational studies on the 2G12 antibody15,18, and the specific mutations with a representative example of the grafting approach can be found in Fig. 1b and S1. The affinity interface iAb mechanism utilizes a hydrophobic patch on the surface of the heavy chain variable domain to facilitate intramolecular Fab-Fab association. The residue sets used to generate these Fab-Fab interactions and facilitate iAb formation (iAbaff1 and iAbaff2) were inspired by lineages of broadly neutralizing anti-SHIV antibodies discovered in SHIV-infected macaques, specifically DH851 and DH89816. In order to graft each residue set onto “acceptor” antibody clones, we first aligned each antibody sequence and then substituted the amino acids at the given residues in Fig. 1b with the appropriate iAb inducing residue set. A representative example of two anti-OX40 antibody grafts is depicted in Fig. S1. Based on varying degrees of amino acid conservation at each of the residues, grafting the residue set resulted in between 4 to 8 mutations per antibody, with an average of 7 mutations across all antibodies studied in this work.

Protein expression and purification

With the exception of anti-DR4 and anti-DR5 antibodies, protein expression was performed by transfection of DNA into in-house HEK293 cells. Anti-DR4 and anti-DR5 antibodies drive apoptosis of HEK293 cells and were therefore produced with in-house CHO cells. For IgG and iAbs, co-transfection of heavy and light chain DNAs was performed. Since contorsbodies contain a genetic fusion of the light chain to the Fc region, only a single plasmid was required for monospecific formats. OX40 ECD was expressed with a baculovirus expression system in Tni insect cells in the presence of 10 mg/L kifunensine.

Following expression, affinity chromatography was performed using MabSelect SuRe resin (Cytiva, cat#17543803) for Fc-containing proteins, CaptureSelect CH1-XL resin (ThermoFisher, cat#194346201 L) for Fabs, and NiNTA agarose resin (Qiagen, cat#30210) for the receptor ECDs. Elution buffers consisted of 50 mM sodium citrate at pH 3.0 and 150 mM NaCl for the MabSelect SuRe and CaptureSelect CH1-XL resins, and 50 mM sodium phosphate at pH 7.4, 200 mM NaCl, and 400 mM imidazole for the NiNTA resin. Size exclusion chromatography was used as the final purification step using a HiLoad 16/600 Superdex 200 column. Protein quality was determined by analytical SEC using a Waters xBridge BEH200A SEC 3.5 um (7.8 × 300 mm) column (Waters, cat#176003596) and by SDS-PAGE. All antibody formats were stored in a buffer consisting of 20 mM histidine acetate and 150 mM NaCl at pH 5.5, while the receptor ECDs were stored in 25 mM tris pH 7.5 and 150 mM NaCl.

Bispecific IgG and iAb production was performed, as previously described66. In brief, half antibodies containing either the knob (T366W) or hole (T366S, L368A, Y407V) mutations were first expressed in separate cell cultures and purified as described above. Two half antibodies were assembled into a single bispecific antibody through annealing, reduction, and oxidation steps. Annealing of the 1:1 half antibody mass mixtures was performed at 37°C for 25 min followed by 24°C for 30 min. Reduction of the disulfides was performed with the addition of 2 mM dithiothreitol for 2 h. After a 30 min oxidation step using 5 mM dehydroascorbic acid, the desired heterodimer species was separated from unwanted homodimers using size exclusion chromatography. Due to the genetic fusion of the light chain, bispecific contorsbodies were produced in a single cell culture as described above without any in vitro assembly steps.

Negative stain electron microscopy

Antibody samples for negative stain EM analysis were exchanged into a buffer consisting of 25 mM tris and 150 mM NaCl, concentrated to 1 mg/ml, and filtered through 0.22 μm membranes (Costar, cat#8160). Samples were then diluted to 0.01 mg/ml, and 4 μl of the diluted sample was immediately deposited on a glow-discharged (Solarus plasma cleaner, Gatan) ultra-thin carbon coated 400-mesh copper grid (Electron Microscopy Sciences). After incubation for 30 s, the remaining liquid was blotted away with filter paper (Whatman, cat#WHA1001090), and the grid was washed 5× with 30 μl of filtered 2% uranyl acetate (Electron Microscopy Sciences). The excess uranyl acetate stain was blotted away with filter paper after 30 s. The grids were imaged on a Talos 200 C equipped with a 4 K Ceta CMOS camera (ThermoFisher) at 73,000× magnification (2 Å per pixel). SerialEM was used for all data collection, and image processing was performed with cisTEM analysis software to generate 2D class averages. Percentages of i- and Y-shaped antibodies for a given sample were calculated using the number of particles in each 2D class.

Generation of F(ab’)2

The F(ab’)2 construct of 3C8 iAbaff1 was generated through proteolytic cleavage of the lower hinge using a modified matrix metalloproteinase 3 (MMP3) as described previously67. In brief, the MMP3 protease was fused to the N-terminus of an in-house affinity matured anti-human Fc antibody based on the rheumatoid factor RF6168. Additionally, the MMP3 protease was engineered for more efficient activation through the addition of an enterokinase cleavage site within the pro-domain. Pro-domain cleavage and subsequent activation of the MMP3-antibody fusion construct was achieved by incubating 16 units of enterokinase (NEB, P8070L) for every 25 μg protein at room temperature for 16 hours in a buffer containing 25 mM tris at pH 7.5, 150 mM NaCl, and 10 mM CaCl2. To inactivate the enterokinase, 0.1 mg/ml soybean trypsin inhibitor (Sigma, 17075029) was added to the protein solution. The activated MMP3-antibody fusion was mixed with the 3C8 iAbaff1 construct at a 1:10 molar ratio and incubated overnight at 37 °C. MabSelect SuRe resin was used to remove all cleaved Fc, unreacted IgG, and MMP3-antibody fusion protein, then the supernatant was purified with size exclusion chromatography and analyzed via SDS-PAGE.

Sedimentation velocity analytical ultracentrifugation (SV-AUC)

A concentration series of 3C8 Fab with the iAbaff1 residue set graft was analyzed by sedimentation velocity analytical ultracentrifugation (SV-AUC) to demonstrate the presence of dimer in solution and determine the affinity of the homodimer. AUC is a well-established method for quantitative analysis of macromolecule interactions in solution. Sedimentation velocity analytical ultracentrifugation (SV-AUC) experiments were performed in an Optima XL-I analytical ultracentrifuge (Beckman-Coulter, Indianapolis, IN) at 20 °C and 50,000 RPM (201,600 g). 3C8 iAbaff1 Fab was prepared at 0.6, 0.2, and 0.07 mg/ml in a buffer containing 25 mM histidine acetate at pH 5.5 and 150 mM NaCl and loaded into the sample sector of 2-sector 3 mm charcoal filled EPON centerpieces (Spin Analytical, Berwick, ME) with the diluent buffer in the reference sector. Samples were equilibrated to 20 °C for 2.5 h before the run was started. Sedimentation was monitored at 280 nm using the UV/vis absorbance system on the centrifuge in continuous mode with a radial step size of 0.003 mm. 150 scans were collected for each sample over approximately 8 h. The distributions of the apparent sedimentation coefficient, g(s*), taken at the same reduced sedimentation time for each concentration were calculated using SedAnal v6.8069, and the increase in the weight average sedimentation coefficient with increasing concentration is characteristic of an associating system (Fig. S7A).

Subsequently concentration difference data for each protein concentration were globally fit to a monomer—dimer equilibrium model in SedAnal v6.80 (Fig. S7B)70. The molecular weight of the monomer, the sedimentation coefficients of the monomer and the dimer, as well as the KD for the monomer—dimer equilibrium were floated as variables in the fit. The molecular weight of the dimer is fixed in the model to be twice the molecular weight of the monomer. The results are collected in Table S1. These results confirm that 3C8 iAbaff1 Fab forms reversible dimers in solution with KD ~ 6.8 μM.

Affinity measurements

Solution affinity constants for all antibodies were determined on a Biacore 8k+ or T200. Antibodies were diluted to 1 μg/ml in HBS-P+ buffer (Cytiva, cat#BR100671) and captured on a Series S Protein A chip (Cytiva, cat#29127555) according to the manufacturer’s protocols. Serial dilutions of the appropriate receptor ECDs (recombinantly produced OX40, CD40, 4-1BB, DR4, DR5, IL-2Rβ, and IL-2Rγ, as described above) were prepared in HBS-P + . The dilutions were injected for 3 min, followed by a 5 min dissociation step. Affinity constants were determined from kinetic fits to the sensograms using the Biacore Evaluation Software.

Cell binding analysis

A 0.6 µM solution of each anti-IL-2Rβ or anti-IL-2Rγ antibody was incubated overnight at 4 °C with 2.4 µM of Alexa Fluor 488 labeled anti-human IgG goat affiniPure Fab fragment (Jackson, cat#109-547-008). Serial dilutions of each 4:1 molar ratio Fab:antibody mixture were prepared in clear 384-well FACS plates in 20 µL FACS buffer (1x PBS with 1% BSA). 20 µL of FACS buffer containing 80,000 IL-2Rβ and IL-2Rγ overexpressing Jurkat cells was added to each well and incubated for 4 h at 4 °C. Cells were pelleted and washed 2 times, resuspended in 40 µL of FACS buffer, and analyzed on an iQue3 cytometer (Sartorius).

Receptor-mediated internalization assay

WT IgG, iAbaff1, or hexameric versions of the anti-OX40 clone, 3C8, were labeled with pHAb amine reactive dye according to the manufacture’s protocols (Promega, cat#G9841). OX40 expressing Jurkat cells were treated with the indicated concentration of each pHAb labeled format for 1 h at 37 °C and 5% CO2 in RPMI media containing 10% FBS and 2 mM L-glutamine (cRPMI). Cells were then washed twice with PBS containing 1% BSA and fluorescence was measured using the BL2 channel of a Sartorius iQue3.

Total internal reflection fluorescence (TIRF) microscopy and single particle tracking

Jurkat T cells were transfected (Amaxa) with 0.3 µg of an OX40-mNeongreen plasmid in a pRK vector backbone 48–72 h before live cell imaging. 48 well glass bottom plates were coated with 100 µg/mL poly-L-lysine for 30 min at 37 °C, washed, and allowed to dry overnight before addition of anti-CD3ε antibodies (OKT3 at 10 µg/mL) to stimulate T cells and enhance spreading. All imaging was performed on a Nikon TIRF system with a 100× 1.49 NA objective, Hamatsu Orca FusionBT SCMOS camera, and iLas2 laser system for ellipse illumination to flatten the field at an imaging depth of 75–100 nm. After cell adhesion to surfaces the pre-treatment datasets were acquired at 20 Hz for 12.5 to 25 s (250–500 frames). The indicated anti-OX40 antibody formats were then added to the imaging wells at 2 μg/mL (13.3 nM) and the same pre-treated cells plus additional cells were acquired at the same frame-rate for the next 10 min. At least 6 cells per condition were analyzed from two independent experiments resulting in over 20,000 OX40 trajectories per condition. Tracking was performed with a DiaTrack 3.0 MatLab runtime application and mean square displacement plots were generated with custom written Igor track analysis code71. In brief, image stacks were background subtracted in DiaTack and processed with a gaussian filter based on a 1.2 pixel half-width half-max value for the point spread function. Particle-identification thresholds were user determined to ensure proper identification versus background after previewing 50-100 frames. Tracks with a max displacement of 3 pixels (390 nm) and minimum lifetime of 3 frames (150 ms) were used to generate MSD curves that average the square displacement of all molecules over the given time frame. Representative max projection images were created with imageJ and track insets were created with Igor and registered to the representative fields in Adobe Illustrator.

Yeast display

An in-house derived, S. cerevisiae yeast displayed scFv library with a diversity of 1.6×109 unique sequences was used to discover binders against IL-2Rβ and IL-2Rγ using a combination of magnetic-activated cell sorting (MACS) and fluorescence-activated cell sorting (FACS) as depicted in Fig. S11. Briefly, yeast were electroporated with plasmid encoding the scFv library and grown to log phase at 30 °C in SD-CAA media. To ensure sampling of the entire library, the number of yeast cells used for each round of selection was 10-fold over the theoretical or measured library diversity. scFv display was controlled via a galactose-inducible promoter. Yeast were grown at 20 °C in SG-CAA media (containing galactose) for 24-48 hours at a starting OD600 of 1.0 to induce scFv display. Before each round of selection, scFv expression on the yeast surface was confirmed using an Alexa Fluor 488-labeled anti-cMyc antibody (1/50 dilution, Cell Signaling, cat# 2279). Binding to IL-2Rγ or IL-2Rβ was determined by the addition of the indicated concentration of in-house derived IL-2Rβ containing a C-terminal Avi tag for site specific biotinylation or commercially sourced IL-2Rγ-biotin (Acro biosystems, cat#ILG-H85E8) and streptavidin (SA) beads (Miltenyi, cat#130-048-101) or Alexa Fluor 647-labeled SA tetramers (ThermoFisher, cat#S21374). For initial rounds of selection using magnetic SA beads, biotinylated antigen (500 nM final concentration) was mixed with 1 mL of SA beads prior to addition of yeast to enhance avidity. Subsequent rounds of selection were conducting with increasing stringency using tetrameric SA-antigen (500 nM SA) followed by decreasing concentrations of monomer. For monomer selections, yeast were first incubated with biotinylated antigen, washed with PBS containing 1% BSA and then stained with a 1/1000 dilution of SA. Each round was checked for enrichment of a binding population by staining yeast with a titration of antigen and analyzing fluorescence using an iQue3 (Sartorius). The results for 37 nM are shown in Fig. S11. DNA was extracted from each the final rounds of selection via ZymoPrep (Zymo Research, cat#D2004), transfomed into E. coli, and individual colonies were selected for sanger sequencing.

Epitope mapping

Epitope mapping of anti-IL-2Rβ and anti-IL-2Rγ clones was performed using a Carterra LSA. Alanine substitutions were first introduced at each residue of 6x histidine tagged IL-2Rβ and IL-2Rγ ECDs using PCR based mutagenesis (n = 206 and 203, respectively). If an alanine was already present the residue was mutated to glycine. Cystines were not mutated. All mutants were expressed in 293 cells and purified using NiNTA agarose resin as described above. Purified IL-2Rβ mutants were then arrayed and captured on a NiHC200M sensor chip for 5 min. A bispecific format where only one arm was specific for IL-2Rβ was flowed over the chip for 5 min and buffer was flowed for 5 min to allow for dissociation. The chip was then regenerated with 350 mM EDTA twice for 5 min and prepped with 5 mM NiCl for 5 min. This process was repeated for each of the anti-IL-2Rβ lead clones, and with the IL-2Rγ mutants combined with the anti-IL-2Rγ lead clones. Mutant receptor capture levels were calculated for each mutant at each cycle and response unit measurements were taken at the end of the association phase of each antibody. Ligand levels were plotted against antibody binding to identify alanine mutations that impacted antibody binding and these positions were highlighted on the previously reported crystal structure of the corresponding receptor41.

Bridging ELISA

Recombinantly expressed human IL-2Rβ was coated onto a Maxisorp 96-well plate (ThermoFisher, cat#44-2404-21) overnight at 4 °C using a 1 µg/ml solution in PBS. The wells were then blocked with a solution of 0.5% BSA and 2 mM EDTA in PBS for 1 hour at room temperature. Three-fold dilutions of the lead antibodies were prepared in PBS with a top concentration of 60 µg/ml, and 100 µl of the antibody dilutions were added to the wells. The plates were incubated for 1 hour at room temperature and then washed 3 times with PBS. During the antibody incubation, a solution of 10 µg/ml biotinylated human IL-2Rγ (Acro Biosystems, cat#ILG-H85E8) and 100 µg/ml streptavidin-HRP (SouthernBiotech, cat#7100-05) was prepared in PBS and incubated at 37 °C for 30 minutes. The IL-2Rγ and streptavidin-HRP solution was diluted ten-fold, and 100 µl was added to the each washed Maxisorp well. After incubation for 30 minutes at 37 °C, the plate was washed 3 times with PBS. 100 µl of TMB substrate (ThermoFisher, cat#N301) was added to each well, and the absorbance at 650 nm was measured after 10 minutes. The absorbance signal was reported as fold-change over a control well without any added antibody.

Cell-based reporter assays

OX40, 4-1BB, DR4, and DR5 assays were performed as previously described20,24. In short, both OX40 and 4-1BB assays used overexpressing Jurkat-NFκB-luciferase reporter cells (from an in-house source and Promega cat#JA2351, respectively) seeded at 80,000 and 40,000 cells/well, respectively, in 20 µl RPMI containing 1% glutamine and 10% heat inactivated fetal bovine serum in 384-well tissue culture plates (Corning cat#3764). Antibody dilutions in 20 µl of the same RPMI medium were added to each well. Plates were incubated overnight at 37 °C and 5% CO2. The level of receptor agonism was determined by quantification of luciferase expression using 40 µl Bright-Glo reagent per well after a 5 min room temperature incubation (Promega, cat#E2650) and a Perkin-Elmer Envision plate reader.

For DR4 and DR5, Colo-205 cells (ATCC cat#CCL-222), which endogenously over-express both DR4 and DR5, were plated at 100,000 cells/well in a white 96-well plate (Corning cat#3917) in 50 µl RPMI supplemented with 1% glutamine and 10% heat inactivated fetal bovine serum and incubated overnight at 37 °C and 5% CO2. The next day, the cells were treated with 50 µl of the indicated antibody at various dilutions. After incubation for 24 hrs at 37 °C and 5% CO2, cell death was quantified by CellTiter-Glo 2.0 (Promega, cat#G9242) using a Perkin-Elmer Envision plate reader.

For the CD40 bioassay, reporter cells were purchased from Promega (cat#JA2151) and used to assess CD40 agonist activity as follows. Cells were thawed and 10,000 cells were plated in each well of a black walled 384-well tissue culture treated plate (Corning, cat#3764) in 20 µL of cRPMI. Cells were allowed to adhere for 6 hours at 37 °C and 5% CO2. 20 µL of a serial dilution of the indicated antibody in cRPMI was then added to the cells and incubated overnight under the same conditions. The following day, 40 µL of Bright-Glo was added to each well and luciferase signal was read on a Perkin Elmer Envision plate reader.

For the IL-2 reporter assay, Jurkat cells were engineered in-house to express IL-2Rβ, IL-2Rγ, and a STAT5-luciferase reporter (Jurkatβγ-STAT5-Luc). 20,000 cells in 20 µL of cRPMI were added to 20 µL of cRPMI containing serial dilutions of antibodies or recombinant IL-2 and incubated overnight at 37 °C and 5% CO2. The following day, 40 µL of Bright-Glo reagent was added to each well and luciferase signal was quantified using a Perkin Elmer Envision plate reader. For IL-2 blocking experiments, cells were first coated with 1 µM of each monospecific anti-IL-2Rβ or anti-IL-2Rγ clone for 1 h prior to the addition of an IL-2 serial dilution. Schematics of each assay can be found in Fig. S4A-C.

Primary cell assays

Purified primary human CD8 + T cells (cat#70027) or NK cells (cat#70036) were purchased from STEMCELL technologies. Each cell type was obtained from a different individual donor (n = 1) and isolated to >85% purity by STEMCELL technologies. CD8 + T cells were pre-stimulated with a 1:1 ratio of CD3/CD28 T-Activator Dynabeads (Gibco, cat#11131D) at 1 × 106 cells/mL in cRPMI at 37 °C and 5% CO2. After 48 hours the Dynabeads were magnetically separated from the cells and cells were allowed to rest overnight in cRPMI at 37 °C and 5% CO2. 50 µl or cRPMI containing 25,000 cells was then added to 50 µL of cRPMI containing a serial dilution of the indicated mimetic antibody format or recombinant IL-2 and incubated at 37 °C and 5% CO2 in white 96-well plates (Corning, cat#3917). After 48 hours, 100 µL of CellTiter-Glo 2.0 (Promega, cat#G9242) was added to each well and luciferase signal was read on a Perkin Elmer Envision plate reader. The same protocol was followed for NK cells with the exception of the pre-stimulation step.

RNA-seq

Purified primary human CD8 + T cells (STEMCELL technologies, cat#70027) were pre-stimulated and rested as described above. 2 × 106 cells were plated in 2 mL of cRPMI in 6 well plates. Each well was treated in triplicate with 100 nM of the indicated mimetic antibody format or recombinant IL-2 and incubated at 37 °C and 5% CO2 for 24 h. Cells were pelleted and RNA was extracted using an RNeasy mini kit (Qiagen, cat#74014).

Total RNA was quantified with Qubit RNA HS Assay Kit (ThermoFisher) and quality was assessed using RNA ScreenTape on 4200 TapeStation (Agilent Technologies). For sequencing library generation, the Truseq Stranded mRNA kit (Illumina) was used with an input of 100 ng of total RNA. Libraries were quantified with Qubit dsDNA HS Assay Kit (ThermoFisher) and the average library size was determined using D1000 ScreenTape on 4200 TapeStation (Agilent Technologies). Libraries were pooled and sequenced on NovaSeq 6000 (Illumina) to generate 30 million single-end 50-base pair reads for each sample.

RNA-sequencing data were analyzed using HTSeqGenie72 in BioConductor73 as follows: first, reads with low nucleotide qualities (70% of bases with quality <23) or matches to rRNA and adapter sequences were removed. The remaining reads were aligned to the human reference genome (human: GRCh38.p10) using GSNAP74,75 version ‘2013-10-10-v2’, allowing maximum of two mismatches per 75 base sequence (parameters: ‘-M 2 -n 10 -B 2 -i 1 -N 1 -w 200000 -E 1 –pairmax-rna=200000 –clip-overlap’). Transcript annotation was based on the Gencode genes data base (human: GENCODE 27). To quantify gene expression levels, the number of reads mapping unambiguously to the exons of each gene was calculated.

Differential expression was calculated using EdgeR76 (version 3.40.1), grouping on replicates of each condition relative to the negative control condition (gD). Differentially expressed genes were determined using a Benjamini-Hochberg False Discovery Rate of 1%. The log2 fold-change of differentially expressed genes were normalized within each condition using sklearn.preprocessing.normalize77 (scikit-learn version 1.0.1) before hierarchical clustering was performed using scipy.cluster.hierarchy78 (SciPy version 1.7.3, parameters: method = ‘ward’).

Statistics and reproducibility

Statistical analysis for specific experiments can be found within the relevant Methods subsection. Sample size was guided by assay throughput. Cell-based agonism assays had at least n = 2 independent wells along with wide-spanning titration curves. Each assay was repeated 2-4 times with robust reproducibility. All other experiments were performed once, guided by throughput as well as time and resource requirements. No statistical method was used to predetermine sample size. No data were excluded from the analysis. No experiments were randomized or blinded.

Reporting summary

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