De novo design of potent and selective mimics of IL-2 and IL-15

Computational design of de novo cytokine mimetics:

The design of de novo cytokine mimetics began by defining the structure of hIL-2 in the quaternary complex with the IL-2RβƔ_c receptor as the template for the design. After inspection, the residues composing the binding-site were defined as hotspots using Rosetta’s metadata (PDBInfoLabels). The structure was fed into the new mimetic design protocol that is programmed in PyRosetta, which can automatically detect the core-secondary structure elements that compose the target-template and produce the resulting de novo mimetic backbones with full RosettaScripts compatible information for design. Briefly, the mimetic building algorithm works as follows. For the first generation of designs, each of the core-elements was idealized by reconstruction using loops from a clustered database of highly-ideal fragments (fragment-size 4 amino acids, see Data availability). After idealization, the mimetic building protocol aims to reconnect the idealized elements by pairs in all possible combinations. To do this it uses combinatorial fragment assembly of sequence-agnostic fragments from the database, followed by cartesian-constrained backbone minimization for potential solutions (i.e. where the N- and C- ends of the built fragment are close enough to link the two secondary structures). After minimization, the solutions are verified to contain highly ideal fragments (i.e. that every overlapping fragment that composes the two connected elements is also contained within the database) and that no backbone clashes with the target (context) receptor. Successful solutions were then profiled using the same database of fragments in order to determine the most probable amino acids at each position (this information was encoded as metadata on each design). Next, solutions for pairs of connected secondary structures were combinatorially recombined (by using graph theory connected components) to produce fully connected backbones. Since the number of solutions grows exponentially with each pair of elements, at each fragment combination step we ranked the designs to favor those with shorter interconnections between pairs of secondary-structure core elements (i.e. effectively with shorter loops), and kept only the top solutions. Fully connected backbone solutions were profiled by layer (interface, core, non-core-surface, surface) in order to restrict the identities of the possible amino acids to be layer-compatible. Finally, all the information on hotspots, compatible built-fragment amino acids and layers were combined (hotspot has precedence to amino acid probability, and amino acid probability took precedence to layer). These fully profiled backbones were then passed to RosettaScripts for flexible backbone design and filtering (see SI Appendix A). For the second generation of designs, we followed two approaches. In the first approach, we just simply executed Rosetta sequence redesigns of our best first generation optimized design (G1_neo2_40_1F, SI Appendix B). In the second approach, we engineered new mimetics using G1_neo2_40_1F as the target template. The mimetic design protocol in this second generation was similar to the one described for the first generation, but with two key differences. Firstly, the core-elements (i.e. those that are secondary structures) were no longer built from fragments, but instead by discovering parametric equations of repetitive phi and psi angles (omega fixed to 180°) that result in secondary structures that recapitulated each of the target helices as close as possible, a “pitch” on the phi and psi angles was allowed every 3rd residue in order to allow the helices the possibility to have curvature (final angle parameters: H1: phi=−60.4, psi=−45.8, phi_pitch=−1.0, psi_pitch=2.0; H2: phi=−64.5, psi=−38.4, phi_pitch=4.0, psi_pitch=−8.0; H3: phi=−64.6, psi=−40.6, phi_pitch=0.0, psi_pitch=0.0; H4: phi=−64.3, psi=−41.7, phi_pitch=0.0, psi_pitch=0.0). By using these parametric equations, the algorithm can variate the length of each of the core-elements up to ∓8.a.a. (compared to input the template). Reductions in the size of the core elements were not allowed to remove hotspots from the binding site. All length variations of the core-elements were reconnected with loops from a clustered database of highly ideal loops (fragment-size of 7 amino acids). The rest of the design algorithm was in essence similar to the one followed in the generation one (SI Appendix C). The Rosetta energy functions used for sequence design were “talaris2013” and “talaris2014”, for the first and second generation of designs, respectively.

The databases of highly ideal fragments used for the design of the backbones for the de novo mimetics (see Data availability) were constructed with the Rosetta application “kcenters_clustering_of_fragments” using an extensive database of non-redundant (publicly available) protein structures from the RCSB protein data bank, which was comprised of 16767 PDBs for the 4-mer database (first generation of designs), and 7062 PDBs for the 7-mer database used for the second generation designs (see Data Availability). The computational algorithms for designing de novo mimics are also provided (see Data Availability).

Yeast display:

Yeast were transformed with genes encoding the proteins to be displayed together with a linearized pETcon3 vector. The vector was linearized by 100 fold overdigestion by NdeI and XhoI (New England Biolabs) and then purified by gel extraction (Qiagen). The genes included 50 bases of overlap with the vector on both the 5’ and 3’ end such that homologous recombination would place the genes in-frame between the AGA2 gene and the Myc tag on the vector. Yeast was grown in C-Trp-Ura media prior to induction in SGCAA media as previously described^34,35,51. After induction for 12–18 hours, cells were washed in chilled display buffer (50mM NaPO₄ pH 8, 20mM NaCl, 0.5% BSA) and incubated with varying concentrations of biotinylated receptor (either human or murine IL-2Rα, IL-2Rβ or Ɣ_c) while being agitated at 4°C. After approximately 30 minutes, cells were washed again in a chilled buffer and then incubated on ice for 5 minutes with a FITC-conjugated anti-c-Myc antibody (1 uL per 3×10⁶ cells) and streptavidin-phycoerythrin (1 uL per 100 uL volume of yeast). Yeast was then washed and counted by flow cytometry (Accuri C6) or sorted by FACS (Sony SH800).

Mutagenesis and affinity maturation:

Site-saturation mutagenesis (SSM) libraries were constructed from synthetic DNA from Genscript. For each amino acid on each design template, forward primers and reverse primers were designed such that PCR amplification would result in a 5’ PCR product with a degenerate NNK codon and a 3’ PCR product, respectively. Amplification of “left” and “right” products by COF and COR primers yielded a series of template products each consisting of a degenerate NNK codon at a different residue position. For each design, these products were pooled to yield the SSM library. SSM libraries were transformed by electroporation into conditioned Saccharomyces cerevisiae strain EBY100 cells, along with linearized pETcon3 vector, using the protocol previously described by Benatuil et al. For details of the primers used in the creation of SSM libraries SI Tables S5–6.

Combinatorial libraries were constructed from synthetic DNA from Genscript containing ambiguous nucleotides and similarly transformed into a linearized pETcon3 vector. For details of the primers used in the creation of combinatorial libraries see SI Tables S7–8.

Protein expression:

Genes encoding the designed protein sequences were synthesized and cloned into pET-28b(+) E. coli plasmid expression vectors (GenScript, N-terminal 6xHis-tagged followed by a thrombin cleavage site. For all the designed proteins, the sequence of the N-terminal tag used is MGSSHHHHHHSSGLVPRGSHM (unless otherwise noted), which is followed immediately by the sequence of the designed protein. Plasmids were then transformed into chemically competent E. coli Lemo21 cells (NEB). Protein expression was performed using Terrific Broth and M salts, cultures were grown at 37°C until OD₆₀₀ reached approximately 0.8, then expression was induced with 1 mM of isopropyl β-D-thiogalactopyranoside (IPTG), and the temperature was lowered to 18°C. After expression for approximately 18 hours, cells were harvested and lysed with a Microfluidics M110P microfluidizer at 18,000 psi, then the soluble fraction was clarified by centrifugation at 24,000 g for 20 minutes. The soluble fraction was purified by Immobilized Metal Affinity Chromatography (Qiagen) followed by FPLC size-exclusion chromatography (Superdex 75 10/300 GL, GE Healthcare). The purified Neo-2/15 was characterized by Mass Spectrum (MS) verification of the molecular weight of the species in solution (Thermo Scientific ), Size Exclusion - MultiAngle Laser Light Scattering (SEC-MALLS) in order to verify monomeric state and molecular weight (Agilent, Wyatt), SDS-PAGE, and endotoxin levels (Charles River).

Human and mouse IL-2 complex components including hIL-2 (a.a. 1–133), hIL-2Rα (a.a. 1–217), hIL-2Rβ (a.a. 1–214) hIL-2RƔ_c (a.a. 1–232), mIL-2 (a.a. 1–149), mIL-2Rα ectodomain (a.a. 1–213), mIL-2Rβ ectodomain (a.a. 1–215), and mƔ_c ectodomain (a.a. 1–233) were secreted and purified using a baculovirus expression system, as previously described^39,43. For the zippered hIL-2RβƔ_c heterodimer, the aforementioned extracellular domain residues for the human/mouse IL-2Rβ and human/mouse IL-2RƔ_c were separately cloned into baculovirus expression constructs containing 3C protease-cleavable basic and acidic leucine zippers, respectively, for a high-fidelity pairing of the receptor subunits, as described previously⁵². The IL-2Rβ and IL-2RƔ_c constructs were transfected independently and their corresponding viruses were co-titrated to determine optimal infection ratios for equivalent expression of the two chains. Insect cell secretion and purification proceeded as described for IL-2 cytokine and receptor subunits. All proteins were purified to >98% homogeneity with a Superdex 200 sizing column (GE Healthcare) equilibrated in HBS. Purity was verified by SDS-PAGE analysis. For expression of biotinylated human IL-2 and mouse IL-2 receptor subunits, proteins containing a C-terminal biotin acceptor peptide (BAP)-LNDIFEAQKIEWHE were expressed and purified as described via Ni-NTA affinity chromatography and then biotinylated with the soluble BirA ligase enzyme in 0.5 mM Bicine pH 8.3, 100 mM ATP, 100 mM magnesium acetate, and 500 mM biotin (Sigma). Excess biotin was removed by size exclusion chromatography on a Superdex 200 column equilibrated in HBS.

Neo-2/15 crystal and co-crystal structures:

C-terminally 6xHis-tagged endoglycosidase H (endoH) and murine IL-2Rβ and IL-2RƔ_c were expressed separately in Hi-five cells using a baculovirus system as previously described. IL-2RƔ_c was grown in the presence of 5 μM kifunensine. After approximately 72 hours, the secreted proteins were purified from the media by passing over a Ni-NTA agarose column and eluted with 200 mM imidazole in HBS buffer (150 mM NaCl, 10 mM HEPES pH 7.3). EndoH was exchanged into HBS buffer by diafiltration. mIL-2RƔ_c was deglycosylated by overnight incubation with 1:75 (w/w) endoH. mIL-2Rβand mIL-2RƔ_c were further purified and buffer exchanged by FPLC using an S200 column (GE Life Sciences).

Monomeric Neo-2/15 was concentrated to 12 mg/ml and crystallized by vapor diffusion from 2.4 M sodium malonate pH 7.0, and crystals were harvested and flash frozen without further cryoprotection. Crystals diffracted to 2.0 Å resolution at Stanford Synchrotron Radiation Laboratory beamline 12–2 and were indexed and integrated using XDS (Kabsch, 2010). The space group was assigned with Pointless (Evans, 2006), and scaling was performed with Aimless (Evans and Murshudov, 2013) from the CCP4 suite (Winn et al., 2013). Our predicted model was used as a search ensemble to solve the structure by molecular replacement in Phaser (McCoy et al., 2007), with six protomers located in the asymmetric unit. After initial rebuilding with Autobuild (Terwilliger et al., 2008), iterative cycles of manual rebuilding and refinement were performed using Coot (Emsley et al., 2010) and PHENIX (Adams et al., 2010).

To crystallize the ternary Neo-2/15:mIL-2Rβ:mIL-2RƔ_c complex, the three proteins were combined in equimolar ratios, digested overnight with 1:100 (w/w) carboxypeptidases A and B to remove purification tags, and purified by FPLC using an S200 column; fractions containing all three proteins were pooled and concentrated to 20 mg/ml. Initial needlelike microcrystals were formed by vapor diffusion from 0.1 M imidazole pH 8.0, 1 M sodium citrate and used to prepare a microseed stock for subsequent use in microseed matrix screening (MMS, (D’Arcy et al., 2014)). After a single iteration of MMS, crystals grown in the same precipitant were cryoprotected with 30% ethylene glycol, harvested and diffracted anisotropically to 3.4 Å × 3.8 Å × 4.1 Å resolution at Advanced Photon Source beamline 23ID-B. The structure was solved by molecular replacement in Phaser using the human IL-2Rβ and IL-2RƔ_c structures (PDB ID: 2B5I) as search ensembles. This produced an electron density map into which two poly-alanine alpha helices could be manually built. Following rigid body refinement in Phenix, electron density for the two unmodeled alpha helices, along with the BC loop and some aromatic side chains, became visible, allowing docking of the monomeric Neo-2/15. Two further iterations of MMS and use of an additive screen (Hampton Research) produced crystals grown by vapor diffusion using 150 nl of protein, 125 nl of well solution containing 0.1 M Tris pH 7.5, 5% dextran sulfate, 2.1 M ammonium sulfate and 25 nl of microseed stock containing 1.3 M ammonium sulfate, 50 mM Tris pH 7.5, 50 mM imidazole pH 8.0, 300 mM sodium citrate. Crystals cryoprotected with 3 M sodium malonate were flash frozen and diffracted anisotropically to 2.5 Å × 3.7 Å × 3.8 Å at Advanced Light Source beamline 5.0.1. After processing the data with XDS, an elliptical resolution limit was applied using the STARANISO server (Bruhn et al., 2017). Rapid convergence of the model was obtained by refinement against these reflections using TLS and target restraints to the higher resolution human receptor (PDB ID: 2B5I) and Neo-2/15 structures in Buster (Smart et al., 2012; Bricogne et al., 2016), with manual rebuilding in Coot, followed by a final round of refinement in PHENIX with no target restraints. Structure figures were prepared with PyMol (Schrodinger, LLC. 2010. The PyMOL Molecular Graphics System, Version 2.1.0). Software used in this project was installed and configured by SBGrid (Morin et al., 2013).

Cell Lines:

Unmodified YT-1⁵³ and IL-2Rα⁺ YT-1 human NK cells⁵⁴ were cultured in RPMI complete medium (RPMI 1640 medium supplemented with 10% fetal bovine serum, 2 mM L-glutamine, minimum non-essential amino acids, sodium pyruvate, 25 mM HEPES, and penicillin-streptomycin [Gibco]). CTLL-2 cells purchased from ATCC were cultured in RPMI complete with 10% T-STIM culture supplement with ConA (Corning). 24 hours prior to signaling studies, CTLL-2 cells were resuspended in RPMI lacking T-STIM culture supplement for IL-2 starvation. Viability (>95%) of CTLL-2 cells was verified by trypan blue exclusion (counted in a hemocytometer) immediately before performing the signaling assays. All cells were maintained at 37°C in a humidified atmosphere with 5% CO₂. The subpopulation of YT-1 cells expressing IL-2Rα was purified via magnetic selection as described previously³⁹. Enrichment and persistence of IL-2Rα expression were monitored by analysis of PE-conjugated anti-human IL-2Rα (Biolegend) antibody binding on an Accuri C6 flow cytometer (BD Biosciences).

Circular dichroism (CD):

Far-ultraviolet CD measurements were carried out with an AVIV spectrometer model 420 in PBS buffer (pH 7.4) in a 1 mm path-length cuvette with a protein concentration of ~0.20 mg/ml (unless otherwise mentioned in the text). Temperature melts where from 25 to 95 °C and monitored absorption signal at 222 nm (steps of 2 °C/min, 30 s of equilibration by step). Wavelength scans (195–260 nm) were collected at 25°C and 95°C, and again at 25°C after fast refolding (~5 min).

Binding studies:

Surface plasmon resonance (SPR): For IL-2 receptor affinity titration studies, biotinylated human or mouse IL-2Rα, IL-2Rβ, and IL-2RƔ_c receptors were immobilized to streptavidin-coated chips for analysis on a Biacore T100 instrument (GE Healthcare). An irrelevant biotinylated protein was immobilized in the reference channel to subtract non-specific binding. Less than 100 response units (RU) of each ligand was immobilized to minimize mass transfer effects. Three-fold serial dilutions of hIL-2, mIL-2, Super-2, or engineered IL-2 mimetics, were flowed over the immobilized ligands for 60 s and dissociation was measured for 240 s. For IL-2RβƔ_c binding studies, saturating concentrations of hIL-2Rβ (3 uM) or mIL-2Rβ (5 uM) were added to the indicated concentrations of hIL-2 or mIL-2, respectively. Surface regeneration for all interactions was conducted using 15 s exposure to 1 M MgCl2 in 10 mM sodium acetate pH 5.5. SPR experiments were carried out in HBS-P+ buffer (GE Healthcare) supplemented with 0.2% bovine serum albumin (BSA) at 25°C and all binding studies were performed at a flow rate of 50 L/min to prevent analyte rebinding. Data was visualized and processed using the Biacore T100 evaluation software version 2.0 (GE Healthcare). Equilibrium titration curve fitting and equilibrium binding dissociation (KD) value determination was implemented using GraphPad Prism assuming all binding interactions to be first order. SPR experiments were reproduced three times with similar results. Biolayer interferometry: binding data were collected in a Octet RED96 (ForteBio, Menlo Park, CA) and processed using the instrument’s integrated software using a 1:1 binding model. Biotinylated target receptors, either human or murine IL-2Rα, IL-2Rβ or Ɣ_c were functionalized to streptavidin-coated biosensors (SA ForteBio) at 1μg/ml in binding buffer (10 mM HEPES [pH 7.4], 150 mM NaCl, 3 mM EDTA, 0.05% surfactant P20, 0.5% non-fat dry milk) for 300 seconds. Analyte proteins were diluted from concentrated stocks into binding buffer. After baseline measurement in binding buffer alone, the binding kinetics were monitored by dipping the biosensors in wells containing the target protein at the indicated concentration (association step) and then dipping the sensors back into baseline/buffer (dissociation). For heterodimeric receptor binding experiments for IL-2RβƔ_c, Ɣ_c was bound to the sensor while IL-2Rβ was in solution at saturating concentrations(i.e. at least ~2.5 fold molar excess over the K_d).

STAT5 phosphorylation studies:

In vitro studies: Approximately 2×10⁵ YT-1, IL-2Rα⁺ YT-1, or starved CTLL-2 cells were plated in each well of a 96-well plate and re-suspended in RPMI complete medium containing serial dilutions of hIL-2, mIL-2, Super-2, or engineered IL-2 mimetics. Cells were stimulated for 15 min at 37°C and immediately fixed by addition of formaldehyde to 1.5% and 10 min incubation at room temperature. Permeabilization of cells was achieved by resuspension in ice-cold 100% methanol for 30 min at 4°C. Fixed and permeabilized cells were washed twice with FACS buffer (phosphate-buffered saline [PBS] pH 7.2 containing 0.1% bovine serum albumin) and incubated with Alexa Fluor® 647-conjugated anti-STAT5 pY694 (BD Biosciences) diluted 1:50 in FACS buffer for 2 hr at room temperature. Cells were then washed twice in FACS buffer and MFI was determined on a CytoFLEX flow cytometer (Beckman-Coulter). Dose-response curves were fitted to a logistic model and half-maximal effective concentration (EC₅₀ values) and corresponding 95% confidence intervals were calculated using GraphPad Prism data analysis software after subtraction of the mean fluorescence intensity (MFI) of unstimulated cells and normalization to the maximum signal intensity. Experiments were conducted in triplicate and performed three times with similar results. Ex vivo studies: Spleens and lymph nodes were harvested from wild-type C57BL/6J or B6;129S4-Il2ra^tm1Dw (CD25KO) mice purchased from The Jackson Laboratory and made into a single cell suspension in sort buffer (2% Fetal Calf Serum in pH 7.2 phosphate-buffered saline). CD4+ T cells were enriched through negative selection by staining the cell suspension with biotin-conjugated anti-B220, CD8, NK1.1, CD11b, CD11c, Ter119, and CD19 antibodies at 1:100 for 30 min on ice. Following a wash with sort buffer, anti-biotin MicroBeads (Miltenyi Biotec) were added to the cell suspension at 20 μL per 10⁷ total cells and incubated on ice for 20 minutes. Cells were washed and then resuspended. Negative selection was then performed using EasySep Magnets (STEMCELL Technologies). Approximately 1 ×10⁵ enriched cells were added to each well of a 96-well plate in RPMI complete medium with 5% FCS with 10-fold serial dilutions of mIL-2, Super-2, or Neo-2/15. Cells were stimulated for 20 min at 37°C in 5% CO_2, fixed with 4% PFA and incubated for 30 minutes at 4°C. Following fixation, cells were harvested and washed twice with sort buffer and again fixed in 500 μL 90% ice-cold methanol in dH₂O for 30 min on ice for permeabilization. Cells were washed twice with Perm/Wash Buffer (BD Biosciences) and stained with anti-CD4-PerCP in Perm/Wash buffer (1:300), anti-CD44-Alexa Fluor 700 (1:200), anti-CD25-PE-Cy7 (1:200), and 5 μL per sample of anti-pSTAT5-PE pY694 for 45 min at room temperature in the dark. Cells were washed with Perm/Wash and re-suspended in sort buffer for analysis on a BD LSR II flow cytometer (BD Biosciences). Dose-response curves were fitted to a logistic model and EC50 values and corresponding 95% confidence intervals were determined using GraphPad Prism data analysis software after subtraction of the MFI of untreated cells and normalization to the maximum signal intensity. Experiments were performed in triplicate and repeated three times with similar results.

In vivo murine airway inflammation experiments:

Mice (C57BL/6J, purchased from The Jackson Laboratory) were inoculated intranasally with 20μL of whole house dust mite antigen (Greer) resuspended in PBS to a total of 23μg Derp1 per mouse. From Days 1–7, mice were given a daily intraperitoneal injection of 20μg mIL-2 in sterile PBS (pH 7.2), a molar equivalent of Neo-2/15 in sterile PBS, or no injection. On Day 8, circulating T cells were intravascularly labeled and tetramer-positive cells were enriched from lymph nodes and spleen or lung as previously described (Hondowicz, Immunity, 2016). Both the column flow-through and bound fractions were saved for flow cytometry analysis. Cells were surface stained with antibodies and analyzed on a BD LSR II flow cytometer with BD FACSDiva software (BD Biosciences). Antibodies used: FITC anti-Ki67, clone SolA15, PerCP-Cy5.5 anti-CD25, clone PC61, eFluor 450 anti-Foxp3, clone FJK-16S, BV510 anti-CD8, clone 53–6.7, BV605 anti-PD-1, clone J43, BV711 anti-CD4, clone RM4–5, BV786 anti-CD62L, clone MEL-14, PE anti-CD69, clone H1.2F3, PE-CF594 anti-B220, clone RA3–6B2, PE-Cy7 anti-CXCR5, clone 2G8 and BUV395 anti-Thy1.2, clone 53–2.1. All flow cytometry files were analyzed using FlowJo 9.9.4 and statistical analysis was performed using Prism 7. All experiments were performed in accordance with the University of Washington Institutional Care and Use Committee guidelines.

Colorectal carcinoma in vivo mice experiments:

CT26 cells were sourced from Jocelyne Demengeot’s research group at IGC (Instituto Gulbenkian de Ciência), Portugal. On day 0, 5 × 10^5 cells were injected subcutaneously (s.c.) into the flanks of BALB/c mice purchased from Charles River with 50 μL of a 1:1 mixture of Dulbecco’s modified Eagle medium (Gibco) with Matrigel (Corning). Starting on day 6, when tumor volume reached around 100mm3, Neo-2/15 and mIL-2 (Peprotech) were administered daily by intraperitoneal (i.p.) injection in 50 μL of PBS (Gibco). Mice were sacrificed when tumor volume reached 1,300 mm3. BALB/c mice were purchased from Charles River. Flow cytometry: All reagents were purchased from Gibco by Life Technologies (Thermo Fisher Scientific) unless stated otherwise. Excised tumors were minced and digested using a mix of collagenase I, collagenase IV (Worthington) and DNase I (Roche) in a shaker for 20 minutes, 250 rpm at 37°C. After digestion, samples were passed through a 100μm cell strainer, and resuspended in cold complete RPMI 1640 medium, supplemented with 10 mM of HEPES buffer, 1 mM of sodium pyruvate, 50μM of 2-mercaptoethanol, 100 U/mL of penicillin and 100 μg/mL of streptomycin and complemented with 1% non-essential amino acids (NEAA), 1% GlutaMAX supplement and 10% heat-inactivated fetal bovine serum (HI FBS). The cell suspensions from the spleens and the inguinal lymph nodes were obtained through the smashing of the tissues against the filter of a 100μm cell strainer. Cells were resuspended in PBS with 2% FBS and 1mM EDTA and stained for extracellular markers for 45 min at 4°C. Cell suspensions were then fixed, permeabilized and stained for intracellular markers using the eBioscience™ Foxp3 / Transcription Factor Staining Buffer Set from ThermoFisher Scientific. Samples were analyzed in a BD LSRFortessa™ flow cytometer equipped with a BD FACSDiva software™ and data were analyzed in FlowJo V10 software and the statistical analysis performed using Prism 5. Antibodies (BioLegend) used in colon carcinoma experiments were: CD45-BV510 (30-F11), CD3-BV711 (17A2), CD49b-FITC (DX5), CD4-BV605 (RM4–5), CD8-PECy7 (53–6.7), and Foxp3-APC (FJK-16s; eBioscience). Fixable Viability Dye eFluor 780 (eBioscience) was used to exclude dead cells. Animals were maintained according to protocols approved by the Direção Geral de Veterinária and iMM Lisboa ethical committee.

Melanoma in vivo experiments:

B16F10 cells were purchased from ATCC. On day 0, 5×10⁵ cells were inoculated into the mice (C57BL/6J purchased from Jackson) by s.c. injection in 500 μL of Hank’s Balanced Salt Solution (Gibco). Starting on the specified day, Neo-2/15 or mIL-2 (Peprotech) treatments were administered daily by intraperitoneal (i.p.) injection in 200 μL of LPS-free PBS (Teknova). Treatment with TA99 (a gift from Noor Momin and Dane Wittrup, Massachusetts Institute of Technology) at 150 μg/mouse was added later at the (as indicated). Mice were sacrificed when tumor volume reached 2,000 mm3. Flow cytometry: Excised tumors were minced, enzymatically digested (Miltenyi Biotec), and passed through a 40-μm filter. Cells from spleens and tumor-draining lymph nodes were dispersed into PBS through a 40-μm cell strainer using the back of a 1-mL syringe plunger. All cell suspensions were washed once with PBS, and the cell pellet was resuspended in 2% inactivated fetal calf serum containing fluorophore-conjugated antibodies. Cells were incubated for 15 minutes at 4°C then fixed, permeabilized, and stained using a BioLegend FoxP3 staining kit. Samples were analyzed on a BD Fortessa flow cytometer. Antibodies (BioLegend) used in melanoma experiments were: CD45-BV711 (clone 30-F11), CD8-BV650 (53–6.7), CD4-BV421 (GK1.5), TCRβ-BV510 (H57–597), CD25-AF488 (PC61), FoxP3-PE (MF-14). Animals were maintained according to protocols approved by Dana–Farber Cancer Institute (DFCI) Institutional Animal Care and Use Committee.

Generation of anti-Neo-2/15 polyclonal antibody:

Mice (C57BL/6J purchased from Jackson) were injected i.p . with 500 μg of K.O. Neo-2/15 in 200 μL of a 1:1 emulsion of PBS and Complete Freund’s Adjuvant. Mice were boosted on days 7 and 15 with 500 μg of K.O. Neo-2/15 in 200 μL of a 1:1 emulsion of PBS and Incomplete Freund’s Adjuvant. On day 20, serum was collected and recognition of Neo-2/15 was confirmed by ELISA. For the ELISA, plates were coated with Neo-2/15, K.O. Neo-2/15, or mIL-2 mixed with ovalbumin for a total of 100 ng/well in carbonate buffer. Coated plates were incubated with murine serum diluted 1:1000 in PBS. Binding was detected using anti-mouse IgG conjugated to HRP and developed with TMB. Results were quantified using absorption at 450 nm.

Enzyme-linked immunosorbent assay (ELISA):

High-binding 96-well plates (Corning) were coated overnight at 4°C with 100 ng/mL of Neo-2/15, mIL-2 (Peprotech), hIL-2 (Peprotech), or ovalbumin (Sigma-Aldrich) in carbonate buffer. Antibody binding to target proteins was detected using HRP-conjugated sheep anti-mouse IgG (GE Healthcare) at 75 ng/mL. Plates were developed with tetramethylbenzidine and HCl. Absorbance was measured at 450 nm with an EnVision Multimode Plate Reader (PerkinElmer).

T cell proliferation assay:

Cells were isolated from mice (C57BL/6J purchased from Jackson) spleens using the EasySep T Cell Isolation Kit (Stemcell Technologies). Cells were plated in RPMI in 96-well culture plates at a density of 10,000 cells/well. Media were supplemented with regular or heat-treated Neo-2/15, rmIL-2, or Super-2 (as indicated). After 5 days of incubation at 37°C, cell survival and proliferation were measured by CellTiter-Glo Luminescent Cell Viability Assay (Promega).

In vivo experiments:

For Treg expansion experiments (Figure 4a), naive C57BL/6 mice were treated daily with Neo-2/15 or mIL-2 at the indicated concentrations (n=2–3 per group). After 14 days, spleens were harvested and analyzed by flow cytometry using the indicated markers; For immunogenicity experiments (Figure 4c), C57BL/6 mice were inoculated with 5×105 B16F10 cells by subcutaneous injection. Starting on day 1, mice were treated daily with Neo-2/15 (10 μg) or equimolar mIL-2 by intraperitoneal (i.p.) injection (n=10 for each group). After 14 days, serum (antiserum) was collected and IgG was detected by ELISA in plates coated with fetal bovine serum (FBS 10%, negative control), Neo-2/15, mIL-2, hIL-2, or Ovalbumin (OVA) as a negative control. Polyclonal mouse IgG against Neo-2/15 (Anti-Neo-2/15 pAb) was generated using complete Freund’s adjuvant in conjunction with a knockout of Neo-2/15 (“K.O. Neo-2/15”, which is an inactive double point mutant of Neo-2/15: Y14D, F99D); For Colorectal cancer experiments (Figure 4d), BALB/C mice were inoculated with CT26 tumors. Starting on day 6, mice were treated daily with i.p. injection of mIL-2 or Neo-2/15 (10 μg), or left untreated (n = 5 per group); For Melanoma experiments (Figure 4e), C57BL/6 mice were inoculated with B16 tumors as in “a)”. Starting on day 1, mice were treated daily with i.p. injection of Neo-2/15 (10 μg) or equimolar mIL-2 (n = 10 per group). Twice-weekly treatment with TA99 was added on day 3. Mice were euthanized when weight loss exceeded 10% of initial weight or when tumor size reached 2,000 mm3; For CD8⁺ : Treg ratio in Melanoma experiments (Figure 4f), C57BL/6 mice were inoculated with B16 tumors and treated by daily i.p. injection as indicated. Treatment with TA99 (bottom plot) was started on day 5 and continued twice-weekly. Tumors were harvested from mice when they reached 2,000 mm3 and analyzed by flow cytometry. The CD8⁺ : Treg cell ratio was calculated by dividing the percentage CD45+ TCRβ+ cells that were CD8+ by the percentage that were CD4+ CD25+ FoxP3+.

CAR-T cell in vivo experiments:

In vitro T cell proliferation assay. Primary human T cells were obtained from healthy donors, who provided written informed consent for research protocols approved by the Institutional Review Board of the FHCRC. Peripheral blood mononuclear cells (PBMC) were isolated by centrifugation over Ficoll-Hypaque (Sigma). T cells were isolated using EasySep™ CD8 or CD4 negative isolation kits (STEMCELL Technologies). To stimulate T cells, T cells were thawed and incubated with anti-CD3/CD28 Dynabeads (Gibco) at 1:1 ratio in media supplemented with 50 IU/ml (3.1ng/ml) of IL2. Beads were removed after four days of incubation. Stimulated or freshly thawed unstimulated T cells were plated at 30000 or 50000 cells/well, respectively, in 96 well format and cultured in indicated concentrations of IL-2 or Neo-2/15 in triplicate. Three days later, proliferation was measured using CellTiter-Glo 2.0. (Promega). In vivo RAJI experiment: The FHCRC Institutional Animal Care and Use Committee approved all mouse experiments. Six- to eight-week-old NSG mice were obtained from the Jackson Laboratory. 0.5*10^6 RAJI tumor cells transduced with (ffLuc)-eGFP were tail vein injected into the NSG mice. Seven days post tumor inject, lentiviral transduced anti-CD19 CAR T cells (0.4*10^6 CD4, 0.4*10^6 CD8) prepared as described in (Liu et al, 2016) were infused i.v. into mice. hIL2 or Neo-2/15 at 20μg/mouse were given i.p. from day 8 to 16 post tumor injection.

Statistical and power analyses:

For statistical test a P-value of less than 0.05 considered significant, unless otherwise noted. For comparison of fitted curves in cellular phospho-STAT5 signaling assays, differences in EC₅₀ values were considered statistically significant if their 95% confidence intervals did not overlap. In vivo airway inflammation experiments: comparison of cell populations were performed using a two-tailed t-test. In vivo murine Colon cancer experiments: comparisons of the survival of tumor-bearing mice were performed using the log-rank Mantel-cox test (95% confidence interval). Comparisons of weight loss in tumor-bearing mice were performed using a two-tailed t-test. In vivo murine Melanoma experiments: comparisons of the survival of tumor-bearing mice were performed using the log-rank Mantel-cox test (95% confidence interval). Comparisons of weight loss in tumor-bearing mice were performed using a two-tailed t-test. The minimum group size was determined using G*Power for an expected large effect size (Cohen’s d = 1.75). For all the bar-plots, the whiskers represent ∓1-standard deviation and individual data points are shown (as dots) for experiments where the n<5. Unless otherwise noted, results were analyzed by one-way ANOVA, if significant (95% confidence interval), post-hoc t-tests were performed comparing groups, and P-values adjusted for multiple comparisons are reported.

Software:

The design of de novo protein mimics was performed using custom code (i.e. “Protein Mimic Designer”) programmed in Python⁶⁷, IPython⁶⁸, and using the scientific high-performance modules: PyRosetta⁴², numpy and scipy^69,70, matplotlib⁷¹, sklearn⁷², cython⁷³ and pandas⁷⁴. Data analyses were performed with custom code in Python and IPhyton. Protein sequence design was performed with Rosetta^40,41 and RosettaScripts⁴⁰. Protein visualization was performed using PyMOL⁷⁵. Simple protein-protein sequence alignments were performed using BLASTP, and structural based sequence comparisons were performed using MICAN⁷⁶. Sequence logos were generated with WebLogo⁷⁷.

Data availability:

PDB structures for Neo-2/15 monomer and its ternary complex with mIL-2RβƔ_c are available in the RCSB Protein Data Bank (PDB IDs: 6DG6 and 6DG5, respectively), diffraction images have been deposited in the SBGrid Data Bank (IDs: 587 and 588, respectively) and validation reports are part of the supplementary information. The de novo “Protein Mimic Designer” algorithm and databases used for designing the protein mimetics are available in the online repository Zenodo (doi: “to be provided with the final manuscript”). Other data and materials are available upon request to the corresponding authors.

Extended Data