Engineering a Single-Agent Cytokine/Antibody Fusion That Selectively Expands Regulatory T Cells for Autoimmune Disease Therapy Free

This article is featured in In This Issue, p.1811

Interleukin-2 is a pleiotropic cytokine that orchestrates the proliferation, survival, and function of both immune effector cells and regulatory T (T_Reg) cells to maintain immune homeostasis. IL-2 signals through activation of either a high-affinity (∼100 pM) heterotrimeric receptor (composed of IL-2Rα, IL-2Rβ, and the shared common γ [γ_c]) or an intermediate-affinity (∼1 nM) heterodimeric receptor (composed of only the IL-2Rβ and γ_c chains) (1–3). Consequently, IL-2 sensitivity is dictated by the nonsignaling IL-2Rα-chain, which is abundantly expressed on the surface of T_Reg cells but virtually absent from naive immune effector cells (i.e., NK cells and memory phenotype [MP] CD8⁺ T cells) (1, 2, 4). Formation of the IL-2 cytokine/receptor complex leads to activation of intracellular JAK proteins, which are constitutively associated with IL-2Rβ and γ_c. JAK proteins phosphorylate key tyrosine residues in the receptor intracellular domains, leading to recruitment and activation of STAT-5 to effect immune-related gene expression and regulate functional outcomes (1, 5, 6).

Because of its essential role in the differentiation and growth of T_Reg cells, the IL-2 cytokine has been extensively characterized in preclinical models to treat a range of autoimmune diseases, including diabetes and multiple sclerosis. These models have underlined the need to administer low doses of the cytokine to take advantage of the enhanced IL-2 sensitivity of T_Reg over effector cells (7, 8). More recently, proof-of-concept clinical trials backed by mechanistic studies have demonstrated that low-dose IL-2 therapy specifically activates and expands T_Reg cells to alleviate pathologic autoimmune conditions (9–11). However, careful dose titration is required for these studies, and the off-target activation of effector cells (particularly activated cells with upregulated IL-2Rα expression) remains of concern.

Boyman and colleagues (12) demonstrated that treating mice with complexes of IL-2 and the anti–IL-2 Ab JES6-1 biases cytokine activity toward T_Reg cells to orchestrate an immunosuppressive response, offering an exciting opportunity for targeted autoimmune disease therapy (13). Subsequent work has demonstrated that IL-2/JES6-1 complexes prevent development of autoimmune diseases (14–17) and promote graft tolerance (18, 19) in mice. We recently determined the molecular structure of the IL-2/JES6-1 complex to elucidate the mechanistic basis for its selective stimulation of T_Reg over immune effector cells. JES6-1 sterically obstructs IL-2 interaction with the IL-2Rβ and γ_c subunits to block signaling on IL-2Rα^low effector cells but also undergoes a unique allosteric exchange mechanism with the IL-2Rα subunit, wherein surface-expressed IL-2Rα displaces the JES6-1 Ab and liberates the cytokine to signal through the high-affinity heterotrimeric receptor on IL-2Rα^high T_Reg cells (Fig. 1A). This phenomenon occurs because key residues in the IL-2 AB interhelical loop engage the JES6-1 Ab and the IL-2Rα subunit in distinct orientations; thus, IL-2/Ab and IL-2/IL-2Rα binding are mutually exclusive, leading to bidirectional exchange. Activation of the IL-2 signaling pathway on IL-2Rα^high cells further upregulates IL-2Rα expression to create a positive feedback loop that exquisitely favors T_Reg cell expansion (17).

The immunosuppressive effects of IL-2/JES6-1 complexes make them enticing candidates for autoimmune disease treatment in humans, but clinical administration of mixed IL-2/Ab complexes is complicated by logistical challenges in drug formulation, including optimization of the dosing ratio and instability of the cytokine/Ab complex. Previously, IL-2 has been covalently linked to an anti–IL-2 Ab to enhance its in vivo half-life and stability (20). However, this approach is incompatible with the allosteric exchange mechanism enacted by the IL-2/JES6-1 complex, as tethering IL-2 to the JES6-1 Ab greatly enhances the apparent Ab/cytokine affinity, obstructing the triggered release that is essential for T_Reg bias. To overcome this obstacle to therapeutic development, we used a structure-based engineering strategy to design a single-agent IL-2/JES6-1 fusion that preserves Ab/receptor exchange. Through modulation of the cytokine/Ab affinity, we successfully recapitulated the selective T_Reg potentiation elicited by mixed IL-2/JES6-1 complex treatment, and we demonstrated that our engineered cytokine/Ab fusion controlled autoimmune disease better than the mixed complex in an induced mouse model of colitis. Collectively, our biophysical and functional studies present a mechanism-driven biomolecular engineering approach that enables the therapeutic translation of a cytokine/Ab complex and that can readily be adapted to other systems for a range of immune disease applications.

The sequence encoding hexahistidine-tagged mouse IL-2 (mIL-2; aa 1–149) was cloned into the pMal vector with an N-terminal maltose-binding protein (MBP) followed by a 3C protease site. mIL-2 was expressed in the periplasm of BL21(DE3) Escherichia coli cells by 20 h induction at 22°C with 1 mM isopropyl β-d-thiogalactopyranoside (IPTG). Protein in the periplasmic compartment was isolated by osmotic shock and purified by Ni-NTA (Qiagen) affinity chromatography. Purity was improved via size-exclusion chromatography on a Superdex 75 column (GE Healthcare) in HEPES-buffered saline (HBS; 150 mM NaCl in 10 mM HEPES, pH 7.3).

mIL-2Rα (aa 1–213) ectodomain, mIL-2Rβ ectodomain (aa 1–215), and mγ_c ectodomain (aa 34–233) were secreted and purified using a baculovirus expression system, as previously described (3). 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.

Recombinant JES6-1 Ab and immunocytokine (IC) and mutants thereof were coexpressed using the previously described BacMam technique (21) adapted for 293F human embryonic kidney cells (Life Technologies). For JES6-1 Ab and variants thereof, the JES6-1 V_H chain followed by the rat IgG2a H chain constant domains was cloned into the pVLAD6 vector with a C-terminal hexahistidine tag. The JES6-1 V_L chain followed by the rat κ L chain constant domain was separately cloned into the pVLAD6 vector with a C-terminal hexahistidine tag. For IC constructs, mIL-2 was cloned N-terminal to the V_L domain of the L chain construct, spaced by a (Gly₄Ser)₂ linker. H and L chain constructs were separately transfected into Spodoptera frugiperda insect (SF9) cells as previously described (21), and the resulting viral supernatants were used to infect 293F cells (Life Technologies) in the presence of 10 mM sodium butyrate (Sigma-Aldrich). H and L chain viruses were cotitrated to determine optimal infection ratios for equivalent expression of the two chains. Infected 293F cells were harvested after 72 h, and secreted protein was captured from the supernatant via Ni-NTA (Qiagen) affinity chromatography. Proteins were further purified to >98% homogeneity with a Superdex 200 sizing column (GE Healthcare) equilibrated in HBS, and purity was confirmed by SDS-PAGE analysis. Rat IgG2a isotype control Ab (Clone eBR2a) was purchased commercially (eBioscience).

For expression of biotinylated mIL-2 and mIL-2 receptor subunits, proteins containing a C-terminal biotin acceptor peptide (BAP)-LNDIFEAQKIEWHE were expressed and purified 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-Aldrich). Excess biotin was removed by size exclusion chromatography on a Superdex 200 column equilibrated in HBS. Abs and ICs were biotinylated using the EZ-Link Sulfo-NHS-LC-Biotinylation kit (Pierce) according to the manufacturer’s protocol.

Unmodified YT-1 (22) and IL-2Rα⁺ YT-1 human NK cells (23) were cultured in RPMI complete medium (RPMI 1640 medium supplemented with 10% FBS, 2 mM l-glutamine, minimum nonessential amino acids, sodium pyruvate, 25 mM HEPES, and penicillin/streptomycin [Life Technologies]) and 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 previously (24). Ten million unsorted IL-2Rα⁺ YT-1 cells were washed with FACS buffer (PBS, pH 7.2, containing 0.1% BSA) and incubated in FACS buffer with PE-conjugated anti-human IL-2Rα Ab (clone BC96; BioLegend) for 2 h at 4°C. PE-labeled IL-2Rα⁺ cells were then incubated with paramagnetic microbeads coated with an anti-PE IgG for 20 min at 4°C, washed once with cold FACS buffer, and sorted on an LS MACS separation column (Miltenyi Biotec) according to the manufacturer’s protocol. Purified eluted cells were resuspended and grown in RPMI complete medium. Enrichment of IL-2Rα⁺ cells was evaluated using an Accuri C6 flow cytometer (BD Biosciences), and persistence of IL-2Rα expression was monitored by PE-conjugated anti-human IL-2Rα Ab labeling and flow cytometric analysis of sorted IL-2Rα⁺ YT-1 cells.

For cytokine labeling, hexahistidine-tagged mIL-2 was exchanged into labeling buffer (50 mM HEPES, pH 8.2, 150 mM NaCl) with a 0.5 ml, 7000 m.w. cut-off Zeba Spin Desalting Column (Thermo Fisher Scientific). A 5-fold molar excess of lysine-reactive (succinimidyl ester chemistry) second-harmonic-active dye SHG1-SE (25) (Biodesy) was added to the protein. The reaction proceeded on ice for 5 min and was then stopped by removal of unreacted SHG1-SE by buffer exchange into HBS with a 0.5 ml, 7000 m.w. cut-off Zeba Spin Desalting Column (Thermo Fisher Scientific). The reacted protein was centrifuged at 16,000 × g at 4°C for 20 min to pellet any precipitate, and the supernatant was analyzed by UV-Vis spectroscopy. The degree of labeling was determined to be 1 (dye/protein stoichiometry).

For target immobilization, Ni-NTA lipid-containing small unilamellar vesicles were generated in TBS (pH 7.6) from the Ni-NTA Bilayer Surface reagent (Biodesy) according to the manufacturer’s protocol. Supported lipid bilayers were formed by fusion of the Ni-NTA small unilamellar vesicles on the glass surface of a 384-well Biodesy Δ plate. The formed bilayer was washed into HBS. SHG1-SE–labeled hexahistidine-tagged mIL-2 was added to each well at 500 nM final concentration (20 μl well volume). The protein was allowed to attach to the surface through the Ni-NTA/hexahistidine-tag interaction overnight at 4°C. The plate was then equilibrated to room temperature for 30 min, and unbound protein was washed out. Following 20 min incubation at room temperature, the plate was transferred to the Biodesy Δ for data collection.

To assess JES6-1 exchange, 500 nM IL-2Rα was injected at time t = 0, and second harmonic generation (SHG) signal was monitored for 10 min. Two-fold serial dilutions of JES6-1 Ab ranging from 2 μM to 31 nM in assay buffer containing 500 nM IL-2Rα (to keep the receptor concentration constant) were injected, and percentage of SHG signal change (ΔSHG) was tracked for 10 min. To assess IL-2Rα exchange, 500 nM JES6-1 was injected at time t = 0, and SHG signal was monitored for 10 min. Two-fold serial dilutions of IL-2Rα ranging from 32 to 0.5 μM in assay buffer containing 500 nM JES6-1 (to keep the Ab concentration constant) were injected, and ΔSHG was tracked for 10 min. The percentage change in SHG intensity was calculated as (SHG_F − SHG_B)/SHG_B × 100, where SHG_B is the SHG signal at t = 0 and SHG_F is the SHG signal at each time point postinjection.

Atomistic molecular dynamics simulations of mIL-2 were performed using the Gromacs 5.0.4 package (26) with the Amber99SB-ILDN force field (27) and the TIP3P water model (28). Initial conformations were obtained from the crystal structures of the mIL-2/JES6-1 single-chain variable fragment (scFv) (Protein Data Bank identifier [PDB ID] 4YQX), mIL-2/S4B6 Fab (PDB ID 4YUE), and human IL-2 (hIL-2)/hIL-2Rα (PDB ID 1Z92) complexes, with the binding partner omitted. The Modeler 9.14 package (29) was used to mutate the hIL-2 conformation to the mouse sequence and to build crystallographically unresolved loops, with five predicted configurations for each crystal structure. Starting conformations of mIL-2 were positioned in a dodecahedron box with a minimum of 10 Å between the protein’s surface and the edge of the box. The protein was solvated with ∼8400 water molecules; five sodium ions were added to neutralize the charge. Energy minimization was performed using a steepest descent algorithm, and solvent was equilibrated for 500 ps at 300 K and 1 bar with the positions of the protein atoms held fixed. Production simulations were performed using a 2 fs timestep, with trajectory lengths of 150 ns and an aggregate simulation time of 19.4 μs. Constant temperature and pressure were maintained by applying a velocity-rescaling thermostat (30) and a Parrinello–Rahman (31) barostat. Bonds were constrained using the linear constraint solver for molecular simulations (LINCS) algorithm (32), and electrostatic interactions were treated with the particle mesh Ewald method (33). Adaptive sampling was performed by reseeding simulations from poorly sampled regions of the mIL-2 conformational landscape; conformations in these regions were identified by time-structure independent component analysis (34), an unbiased method to determine the slowest degrees of conformational freedom.

A Markov state model (MSM) with a lag-time of 5 ns was constructed from the simulation data using the MSMBuilder 2.8.2 package (35). Similar to prior analysis of IL-2 simulations (24), conformations were clustered into 50 states using a hybrid k–centers k–medoids algorithm, with distances between all pairs of conformations determined from the root mean square deviation (RMSD) of the backbone atom positions. A representative trajectory of the predicted equilibrium dynamics was constructed using kinetic Monte Carlo sampling of the MSM transition probability matrix. Interresidue distances, dihedral angles, and RMSD were computed using MDTraj 1.7.2 (36).

For IL-2 affinity titration studies, biotinylated mIL-2Rα and IL-2Rβ receptor chains, biotinylated JES6-1 Ab mutants, or biotinylated JES6-1 IC mutants 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 nonspecific binding. Less than 100 response units of each ligand was immobilized to minimize mass transfer effects. Three-fold serial dilutions of mIL-2 were flowed over the immobilized ligands for 60 s, and dissociation was measured for 240 s. Surface regeneration for all interactions was conducted using 15 s exposure to 1 M MgCl₂ in 10 mM sodium acetate (pH 5.5).

For IC receptor exchange studies, biotinylated mIL-2Rα or mIL-2Rβ 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 nonspecific binding. Less than 100 response units of each ligand were immobilized to minimize mass transfer effects. Three-fold serial dilutions of mIL-2 or JES6-1 IC mutants were flowed over the immobilized ligands for 60 s, and dissociation was measured for 240 s. Surface regeneration for all interactions was conducted using 15 s exposure to 1 M MgCl₂ in 10 mM sodium acetate (pH 5.5).

Surface plasmon resonance (SPR) experiments were carried out in HBS-P+ Buffer (GE Healthcare) supplemented with 0.2% BSA at 25°C, and all binding studies were performed at a flow rate of 50 μl/min to prevent analyte rebinding. Data were visualized and processed using the Biacore T100 evaluation software version 2.0 (GE Healthcare). Equilibrium titration curve fitting and equilibrium binding dissociation (K_D) value determination were implemented using GraphPad Prism assuming all binding interactions to be first order.

Approximately 2 × 10⁵ YT-1 or IL-2Rα⁺ YT-1 cells were plated in each well of a 96-well plate and resuspended in RPMI complete medium containing serial dilutions of mIL-2, mIL-2/Ab complexes, JES6-1 IC, or JES6-1 IC mutants. Complexes were formed by incubating a 1:1 M ratio of Ab or Ab fragment to mIL-2 for 30 min at room temperature. 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 (PBS, pH 7.2, containing 0.1% BSA) and incubated with Alexa Fluor 647–conjugated anti-STAT5 pY694 (BD Biosciences) diluted in FACS buffer for 2 h at room temperature. Cells were then washed twice in FACS buffer, and mean fluorescence intensity (MFI) was determined on a CytoFLEX flow cytometer (Beckman Coulter). Dose-response curves were fitted to a logistic model, and EC₅₀s were calculated using GraphPad Prism data analysis software after subtraction of the MFI of unstimulated cells and normalization to the maximum signal intensity. Experiments were conducted in triplicate and performed three times with similar results.

For relative T_Reg to CD8⁺/CD4⁺ T cell proliferation studies, 12-wk-old C57BL/6 mice (three per cohort) or NOD mice (four per cohort) were injected i.p. with PBS, mixed IL-2/JES6-1 complex (prepared by preincubating 1.5 μg mIL-2 [eBioscience] with 6 μg JES6-1 [2:1 cytokine/Ab molar ratio] in PBS for 30 min), or 6 μg of the indicated JES6-1 IC mutants on days 1, 2, 3, and 4. Mice were sacrificed on day 5 by cervical dislocation, and spleens were harvested. Single-cell suspensions were prepared by mechanical homogenization, and absolute count of splenocytes was assessed for each spleen by automated cell counter (Vi-CELL; Beckman Coulter). Cells were resuspended in PBS and subsequently stained for 30 min on ice with fluorophore-conjugated anti-mouse mAbs for phenotyping of T_Reg (CD4⁺IL-2Rα⁺Foxp3⁺) or CD8⁺ effector T cells (CD8⁺) using BV605-conjugated anti-CD4 (clone RM4-5; BioLegend), PeCy7-conjugated anti–IL-2Rα (clone PC61.5; eBioscience), PerCP-Cy5.5–conjugated anti-CD8 (clone 53-6.72; eBioscience), and mAbs. Fixable Blue Dead Cell Stain Kit (Life Technologies) was used to assess live cells. Cells were then washed twice with FACS buffer (1% BSA, 1% sodium azide) and fixed in Foxp3 Transcription Factor Fixation/Permeabilization Buffer (eBioscience) for 30 min on ice. After two washes in Permeabilization Buffer (eBioscience), T_Reg cells were stained with FITC-conjugated anti-mouse/rat Foxp3 mAb (clone FJK-16s; eBioscience) for 1 h on ice. Two final washes were conducted in Permeabilization Buffer, and cells were resuspended in FACS buffer for flow cytometric analysis on an LSR II (BD Biosciences). Data were analyzed using FlowJo X software (Tree Star). Ratios of the absolute numbers of T_Reg cells to either CD8⁺ effector T cells or total CD4⁺ T cells are presented. Statistical significance was determined by two-tailed unpaired Student t test. Experiments were performed three times with similar results.

OT-I Ly 5.1⁺ mice (three per cohort) were sacrificed by cervical dislocation and lymph nodes (head, axillary, inguinal, and mesenteric) were harvested. Single-cell suspensions were prepared by homogenization (gentleMACS Dissociator; Miltenyi Biotec), and CD8⁺ effector T cells were negatively sorted on autoMACS (Miltenyi Biotec). Isolated CD8⁺ T cells were labeled with CFSE and injected i.v. into C57BL/6 mice (Ly 5.2) (1 × 10⁶ cells/mouse). Mice were then treated i.p. on day 1 with PBS or with 2 nmol SIINFEKL peptide alone or in combination with 75 μg polyI:C7, 7.5 μg mIL-2 plus 30 μg Rat IgG2a isotype control Ab (Clone 2A3; BioXcell), mixed mIL-2/JES6-1 complex (formed by preincubating 7.5 μg mIL-2 [PeproTech] with 30 μg JES6-1 [2:1 cytokine/Ab molar ratio] in PBS for 15 min), or 30 μg JY3 IC. On days 2, 3, and 4, mice were treated with the same doses of mIL-2 plus Rat IgG2a isotype control Ab, mixed mIL-2/JES6-1 complex, or JY3 IC. Mice were sacrificed on day 5 by cervical dislocation, and spleens were harvested, homogenized, and analyzed by flow cytometry as described for immune cell subset proliferation studies. Four immune populations were distinguished and profiled for IL-2Rα expression using the following fluorophore-conjugated anti-mouse mAbs: adoptively transferred CD8⁺ T cells: V500 Horizon–conjugated anti-CD3 (clone 500A2; BD Biosciences), PerCP-Cy5.5–conjugated anti-CD8 (clone 53-6.72; eBioscience), allophycocyanin-conjugated anti-CD45.1 (clone A20; eBioscience), and eFluor 450–conjugated anti–IL-2Rα; T_Reg cells (CD3⁺CD4⁺Foxp3⁺): V500 Horizon–conjugated anti-CD3 (clone 500A2; BD Biosciences), PerCP-conjugated anti-CD4 (clone RM4-5; BD Biosciences), PE-conjugated anti-mouse/rat Foxp3 (clone FJK-16s; eBioscience), and APC-conjugated anti–IL-2Rα mAbs (clone PC61.5; eBioscience); MP CD8⁺ T cells (CD3⁺CD8⁺CD44⁺IL-2Rβ⁺): V500 Horizon–conjugated anti-CD3, PerCP-Cy5.5–conjugated anti-CD8 (clone 53-6.72; eBioscience), allophycocyanin-conjugated anti-CD44 (clone IM7; eBioscience), PE-conjugated anti–IL-2Rβ (clone 5H4; eBioscience), and eFluor 450–conjugated anti–IL-2Rα (clone PC61.5; eBioscience) mAbs; and NK cells (CD3⁻CD49b⁺CD161): V500 Horizon–conjugated anti-CD3, PE-conjugated anti-CD49b (clone DX5; eBioscience), APC-conjugated anti-CD161 (clone PK136; eBioscience), and eFluor 450–conjugated anti–IL-2Rα mAbs. Fixable Viability Dye eFluor 780 (eBioscience) was used to assess live cells. Data were analyzed using FlowJo X software (Tree Star). Relative number of cells and MFI of IL-2Rα are presented for each cohort in all four immune cell subsets. Statistical significance was determined by two-tailed unpaired Student t test. The experiment was performed three times with similar results.

BALB/c mice (six per cohort) were injected i.p. daily for 7 d with PBS, 7.5 μg mIL-2 plus 30 μg isotype control Ab (clone 2A3; BioXcell), mixed mIL-2/JES6-1 complex (formed by preincubating 7.5 μg mIL-2 [PeproTech] with 30 μg JES6-1 [2:1 cytokine/Ab molar ratio] in PBS for 15 min), or 30 μg JY3 IC. Beginning on day 8, mice were administered 3% dextran sodium sulfate (DSS) (m.w. = 40,000; MP Biomedicals) in their drinking water to induce colitis. On day 15, body weight was recorded, and disease severity was assessed using clinical disease activity index, as described previously (4, 37). On day 16, mice were sacrificed and entire colons were removed (from cecum to anus). Colon length was measured, and shortening was used as an indirect marker of pathological inflammation. Statistical significance was determined by one-way ANOVA plus Dunnett multiple comparison posttest. The experiment was performed two times with similar results.

The aforementioned allosteric exchange mechanism allows for displacement of JES6-1 in the cytokine/Ab complex by the surface-bound IL-2Rα receptor subunit (Fig. 1A). This mechanism was supported by structural and SPR-based studies (17). To demonstrate the bidirectionality of the Ab/receptor exchange mechanism, we interrogated the capacity of both Ab and receptor to engage bound mIL-2 complexes. To this end, we used a SHG detection platform, which was previously used to detect conformational changes in proteins in time and space (25, 38, 39). IL-2 was labeled with a second-harmonic-active dye and immobilized to a surface. The tethered cytokine was then saturated with either IL-2Rα (Fig. 1B, top) or JES6-1 (Fig. 1B, bottom). Subsequently, various concentrations of soluble JES6-1 (Fig. 1B, top) or IL-2Rα (Fig. 1B, bottom) were added, and changes in SHG signal, indicative of modulations in average tilt angles of the dye particles conjugated to IL-2, were quantified. In both topologies, dose-dependent conformational changes were observed in IL-2 upon adding soluble protein to the immobilized complex, demonstrating the bidirectional exchange between Ab and receptor engagement of the cytokine.

To further corroborate the allosteric exchange mechanism, we performed molecular dynamics simulations to study the distinct conformational states of IL-2 when bound to the JES6-1 Ab versus the IL-2Rα receptor, as well as the transition between these states. IL-2, and in particular the IL-2Rα–binding epitope of the cytokine, is known to exhibit extensive conformational flexibility (40–42). We constructed an atomically-detailed MSM of the conformational landscape of free IL-2. The equilibrium dynamics captured by the MSM predicted that IL-2 stably adopts a JES6-1–bound conformation even in the absence of Ab but occasionally relaxes to a distinct metastable state that resembles the IL-2Rα–bound conformation (Fig. 1C, Supplemental Video 1). The Ab-bound and receptor-bound states of the cytokine diverge significantly with respect to RMSD, interresidue distances, and residue-specific dihedral angles in all three interhelical loops (Fig. 1C). The transition from the JES6-1–bound to the IL-2Rα–bound states involves significant conformational rearrangements and, in particular, destabilization of a salt bridge and a hydrogen bond in the AB and BC loops, respectively, that appear to rigidify these regions (Fig. 1C, red and orange). These changes coincide with the loss of a cation–π interaction between the B helix and the β strand of the CD region, accompanied by increased flexibility of the latter (Fig. 1C, green). Inspection of the primary transition path with higher temporal resolution suggests that loss of loop rigidity occurs in sequential fashion. Deformation of the BC loop, which interacts with IL-2Rα, is predicted to precede destabilization of the AB loop, which engages IL-2Rα at its C-terminal end and JES6-1 at its N-terminal end (Fig. 1D). Such a stepwise transition may facilitate allosteric exchange between the JES6-1 Ab and IL-2Rα subunit (17). Taken together, our biophysical and computational studies offer mechanistic insight into the Ab/receptor exchange that drives the T_Reg cell bias induced by stimulation with mixed IL-2/JES6-1 complex.

To stabilize the IL-2/JES6-1 complex with an eye toward translation, we fused the IL-2 cytokine to the full-length JES6-1 Ab, tethering IL-2 to the N-terminal end of the L chain via a flexible (Gly₄Ser)₂ linker (Fig. 2A). Based on the IL-2/JES6-1 complex structure (17), the C terminus of IL-2 is predicted to be 19.9 Å from the N terminus of the JES6-1 L chain (Supplemental Fig. 1A). Our cytokine/Ab construction (hereafter denoted the JES6-1 IC) was designed to allow for intramolecular cytokine engagement. Interaction between IL-2 and JES6-1 within the IC was confirmed by SPR-based titrations of the IL-2Rα subunit. Whereas untethered IL-2 binds the IL-2Rα subunit with an equilibrium K_D of 9.8 nM, JES6-1 IC has a 30-fold weaker IL-2Rα affinity (K_D = 290 nM) (Fig. 2B), reflective of cytokine sequestering by the tethered Ab.

The IL-2/JES6-1 affinity (K_D = 5.6 nM) is similar to the IL-2/IL-2Rα affinity (K_D = 9.8 nM) and significantly stronger than the IL-2/IL-2Rβ affinity (K_D = 7.4 μM) (Supplemental Fig. 1B). Thus, effective exchange of the IL-2 cytokine between the JES6-1 Ab and the IL-2Rα subunit is observed when the affinities are closely matched. We hypothesized that tethering IL-2 to the JES6-1 Ab would enhance the apparent cytokine/Ab affinity due to avidity effects, and this increased cytokine/Ab affinity would in turn weaken IL-2/IL-2Rα interaction in the context of the IC. We speculated that changes in the Ab/cytokine affinity and, by consequence, the IC/IL-2Rα affinity, would impact on the exchange mechanism in a biphasic manner. If the affinity of the IL-2/JES6-1 complex was greatly reduced, the Ab would fall off constitutively, leading to the cytokine to behave the same as the naked IL-2 and activate both IL-2Rα^high T_Reg and IL-2Rα^low effector cells, thus erasing the robust T_Reg cell IL-2 signaling bias conferred by JES6-1 (Supplemental Fig. 1C). Conversely, if the affinity of the IL-2/JES6-1 complex was significantly increased to the limit of an irreversible interaction, the Ab would never be displaced by IL-2Rα, ablating the exchange mechanism and precluding cytokine activity on both T_Reg and effector cells (Supplemental Fig. 1C). Consequently, there exists an optimal IL-2/Ab affinity to maximize T_Reg over effector cell expansion, and substantial enhancement of the IL-2/Ab affinity through IC construction could push this affinity outside of the optimal range.

To examine the functional consequences of Ab/cytokine tethering on cell subset-selective activity, we tested activation of two genotypically matched cell lines based on the YT-1 human NK lineage that differ only in their expression of the IL-2Rα subunit (23), as a surrogate for stimulation of IL-2Rα^high T_Reg cells compared with IL-2Rα^low effector cells. IL-2 exhibited over 30-fold more potent activation (as measured by STAT5 phosphorylation) on IL-2Rα⁺ cells compared with IL-2Rα⁻ cells, as was expected owing to the higher affinity of the heterotrimeric versus the heterodimeric IL-2 receptor complex. The mixed IL-2/JES6-1 complex induced weaker activation of both cell lines but, importantly, showed more pronounced obstruction of signaling on IL-2Rα⁻ compared with IL-2Rα⁺ cells, rationalizing the complex’s IL-2Rα^high T_Reg bias. JES6-1 IC did not activate IL-2Rα⁻ cells and induced much weaker activation of IL-2Rα⁺ cells compared with the mixed complex, consistent with its impaired interaction with the IL-2Rα subunit (Fig. 2C). We explored how this differential in vitro signaling would translate into in vivo immune cell subset bias. Administration of IL-2 alone to NOD mice did not induce an increase in T_Reg relative to CD8⁺ effector T cell abundance, but treatment with IL-2/JES6-1 complex doubled the T_Reg/CD8⁺ T cell ratio. However, this increase was completely absent for JES6-1 IC, indicating that stabilization of the IL-2/Ab affinity had disrupted the exchange mechanism, ablating cytokine activity on both T_Reg and effector cells (Fig. 2D, Supplemental Fig. 1C). Accordingly, enrichment of T_Reg in the total CD4⁺ T cell population was observed following treatment with IL-2/JES6-1 complex but not JES6-1 IC (Supplemental Fig. 1D).

To rescue the T_Reg cell–biased activity of the IC, we used crystallographic insights to rationally design a panel of eight single-point alanine mutants of the JES6-1 Ab. We selected four V_H and four V_L chain residues at the cytokine/Ab interface, intentionally avoiding residues proximal to the IL-2Rα chain to circumvent disruption of the allosteric exchange mechanism (Fig. 3A). We formatted each alanine mutant as a full-length Ab and characterized binding to the IL-2 cytokine via SPR titrations. All mutants with the exception of R62A decreased the Ab/cytokine affinity, with a maximum affinity impairment of 89-fold relative to the parent JES6-1 Ab (Fig. 3B, Table I).

To probe the effects of reduced cytokine/Ab affinity on IL-2Rα exchange, we reformatted each of the alanine mutants as IC fusions and measured the binding of these soluble IC variants to immobilized IL-2Rα using SPR. Each of the eight IC mutants tested exhibited increased receptor affinity compared with the parent JES6-1 IC, indicative of improved cytokine exchange due to enhanced Ab displacement. The most pronounced affinity improvement was observed for the Y41A IC mutant, which had a 2.2-fold tighter affinity for IL-2Rα than JES6-1 IC (Fig. 3C, top and Table I). Notably, neither the parent JES6-1 IC nor any of the mutant IC constructs bound the immobilized IL-2Rβ subunit, indicating that JES6-1–mediated blockade of the IL-2Rβ subunit remained intact for our engineered IC mutants (Fig. 3C, bottom).

We predicted that improved Ab displacement by the IL-2Rα subunit would potentiate mutant JES6-1 IC activity on IL-2Rα⁺ cells, and indeed, we observed that many of our IC mutants enhanced STAT5 signaling on IL-2Rα⁺ YT-1 cells relative to the parent JES6-1 IC. Three IC mutants (S34A, Y41A, and Y101A) recovered the extent of STAT5 signaling induced by the mixed IL-2/JES6-1 complex (Fig. 3D, top). None of the engineered IC mutants activated IL-2Rα⁻ YT-1 cells, consistent with the behavior of the mixed complex (Fig. 3D, bottom). Our JES6-1 IC mutant cellular activation assays also offered insight into the relationships between affinity, exchange, and functional response. We hypothesized that activity would correlate with affinity in a biphasic manner (Supplemental Fig. 1C), and our data support this postulate (Fig. 3E). However, additional structural factors appear to contribute to signaling output, as is evidenced by the much greater potency of activation induced by JES6-1 IC mutants with alanine substitutions in the V_L versus the V_H domain. Because the cytokine was linked to the JES6-1 L chain, H chain mutants disrupt the cytokine/Ab interface further from the tether and we would thus expect function to be more dramatically affected for V_H versus V_L mutants. Accordingly, the V_H mutants generally impair affinity to a greater extent than do V_L mutants (Fig. 3B, Table I). The discrepancy between V_H and V_L JES6-1 IC mutant constructs is even more apparent when comparing IL-2Rα⁺ cell activity to exchange (as determined by IL-2Rα affinity) (Table I). Although activity correlates with exchange within the JES6-1 IC V_L mutants, the JES6-1 IC V_H mutants all elicit weak stimulation of IL-2Rα⁺ cells, independent of their IL-2Rα exchange propensities (Fig. 3F).

Based on our three most active single-point mutant JES6-1 IC constructs (the V_L domain mutants S34A, Y41A, and Y101A) (Fig. 3D), we designed three double-alanine mutant IC constructs and one triple-alanine JES6-1 IC mutant construct to enhance IL-2Rα exchanging capacity and IL-2Rα⁺ cell-selective signaling. All multiresidue mutant IC constructs potentiated IL-2Rα exchange compared with the parent JES6-1 IC, with the most actively exchanging mutants (Y41A+Y101A and S34A+Y101A) exhibiting a 2.6-fold IL-2Rα affinity enhancement relative to the parent JES6-1 IC (Fig. 4A, top, Table II). Several of the multiresidue mutants (including Y41A+Y101A) also bound to immobilized IL-2Rβ due to the weakened cytokine/Ab interaction, but all IC mutants bound significantly weaker to IL-2Rβ than to IL-2Rα and had lower IL-2Rβ affinities than the free IL-2 cytokine (Fig. 4A, bottom). Signaling activity of the multiresidue IC mutants on IL-2Rα⁺ YT-1 cells correlated directly with IL-2Rα exchange; all mutants activated IL-2Rα⁺ cells more potently than the parent JES6-1 IC, and three of the four constructs had greater activity than the mixed IL-2/JES6-1 complex (Fig. 4B, top). The Y41A+Y101A and S34+Y101A mutants were again the most active and none of the mutants triggered appreciable activation of IL-2Rα⁻ YT-1 cells (Fig. 4B, bottom).

Guided by our cellular activation results, we chose to further characterize the Y41A+Y101A mutant, which had a 5.6-fold weaker cytokine/Ab affinity than the unmodified JES6-1 (Fig. 4C), and the S34A+Y101A mutant, which had a 1.9-fold weaker cytokine/Ab affinity than the parent JES6-1 (Supplemental Fig. 2A). We assessed the ability of the Y41A+Y101A IC mutant to expand immune cell subset populations in C57BL/6 mice and found that, in contrast with JES6-1 IC, Y41A+Y101A IC induced preferential T_Reg versus CD8⁺ effector T cell expansion to the same extent as the mixed IL-2/JES6-1 complex (Fig. 4D, Supplemental Fig. 3A), indicating that reducing cytokine/Ab affinity had restored biased activity on IL-2Rα^high cells. Consistent results were observed for the T_Reg versus CD4⁺ T cell ratio (Supplemental Fig. 3B), confirming specific expansion of T_Reg by the Y41A+Y101A IC variant. IC behavior in vivo was found to be highly sensitive to affinity and structural modifications, as the S34A+Y101A IC mutant did not elicit biased T_Reg cell expansion (Supplemental Figs. 2B, 3A), despite its similar exchange (Fig. 4A) and signaling (Fig. 4B) properties to the Y41A+Y101A IC mutant in vitro. Further studies demonstrated that the Y41A+Y101A IC mutant (hereafter denoted JY3 IC) effected preferential T_Reg over CD8⁺ effector T cell growth in a dose-dependent manner in C57BL/6 (Supplemental Fig. 2C) and NOD mice (Supplemental Fig. 2E). Moreover, JY3 treatment upregulated IL-2Rα expression on T_Reg cells to a much greater extent than the mixed IL-2/JES6-1 complex (Supplemental Fig. 2D, 2F). IC formulation also had benefits in enhancing maximum tolerated dose of the IL-2 cytokine compared with the mixed complex, as administration of a 7.4 μg dose of IL-2 in mixed complex format was lethal to NOD mice (three of four mice died), but an equivalent dose of the JY3 IC (30 μg) was well tolerated (zero of four mice died). This suggests that tethering the cytokine to the Ab may mitigate toxicity relative to the mixed complex, presumably through increased complex stability, which reduces off-target effects elicited by free IL-2.

Given that JY3 IC selectively expanded IL-2Rα^high cells in a mixed immune population, we wondered whether our construct would be able to precisely control immune cell subset populations in an adoptive T cell transfer model. We purified and CFSE-labeled CD8⁺ T cells from OT-I mice, transferred the cells into congenic C57BL/6 mice, and treated recipient mice with a low dose of the SIINFEKL peptide plus various IL-2/Ab regimens for analysis of recipient immune cell expansion (Fig. 5A). Peptide stimulation with or without poly-l-lysine and IL-2 coadministration with an isotype control Ab failed to expand adoptively transferred activated CD8⁺ T cells, whereas the mixed IL-2/JES6-1 complex induced robust proliferation of transferred cells, and JY3 IC further enhanced this expansion at equivalent doses (Fig. 5B). An identical trend was observed for IL-2Rα expression on transferred CD8⁺ T cells, with higher surface levels of IL-2Rα elicited by JY3 IC versus IL-2/JES6-1 complex treatment (Supplemental Fig. 4A).

Characterization of recipient mouse T cell subsets supported our hypothesis that JY3 IC biases IL-2 activity to IL-2Rα^high T_Reg cells over IL-2Rα^low naive effector cells. No T_Reg cell expansion was induced by IL-2/isotype control Ab treatment, but we observed an increase in T_Reg cell number following IL-2/JES6-1 complex treatment and an even more profound increase following JY3 IC treatment (Fig. 5C). In contrast, IL-2/isotype control Ab treatment expanded both MP CD8⁺ T cells and NK cells, whereas neither IL-2/JES6-1 complex nor JY3 IC promoted proliferation of these effector subsets (Fig. 5D, 5E). IL-2/JES6-1 complex also upregulated IL-2Rα expression on T_Reg and MP CD8⁺ T cells, although not on NK cells, and JY3 IC increased surface IL-2Rα levels on T_Reg and MP CD8⁺ T cells to a greater extent than the complex and also robustly upregulated IL-2Rα on NK cells (Supplemental Fig. 4B–D). Overall, our adoptive transfer studies establish that JY3 IC specifically targets IL-2 activity to IL-2Rα^high immune cell subsets and that it promotes more robust expansion and receptor upregulation on these subsets compared with the mixed IL-2/JES6-1 complex.

To explore the therapeutic potential of our engineered IC mutant, we compared the efficacy of JY3 IC to the mixed IL-2/JES6-1 complex in a DSS-induced colitis model. Mice were pretreated with PBS, IL-2 with an isotype control Ab, IL-2/JES6-1 complex, or JY3 IC for seven consecutive days, and disease was induced beginning on day 8 (Fig. 6A). One week after colitis induction, as compared with IL-2/isotype control Ab-treated mice, IL-2/JES6-1 complex–treated mice exhibited significant reductions in disease severity, including attenuated weight loss, increased colon length, and lower disease activity index (Fig. 6B–D), consistent with previous findings (17). JY3 IC further enhanced autoimmune disease prevention, with more pronounced improvements in weight loss, colon length, and disease activity score compared with the IL-2/JES6-1 complex (Fig. 6B–D). These results suggest that the JY3 IC could have therapeutic advantages to the mixed cytokine/Ab complex beyond the logistical considerations of stability and ease of formulation.

There is a growing interest in the development of Ab/cytokine fusions (ICs) to empower cytokines as drugs (43). Whereas cytokines have short in vivo half-lives (often <5 min) and thus require frequent dosing (44–46), Abs benefit from prolonged serum persistence due to neonatal Fc receptor–mediated recycling (47, 48). In addition, fusion to surface Ag-binding Abs can allow for targeted cytokine delivery tailored to particular disease indications (43, 49–54). However, clinical development of targeted ICs is hampered by the high potency of cytokines, which nullifies the effect of the targeting Ab and leads to toxicity through indiscriminate activation of all cytokine-response immune cell subsets (55–57). To circumvent the issue of potency, recent efforts have focused on reducing cytokine/receptor affinity through directed mutagenesis of the cytokine (58, 59), but cytokine modification may alter functional activity and also raises concerns about immunogenicity.

In this article, we describe a novel use of ICs to “shield” a cytokine from nonspecifically engaging immune cells and instead target it preferentially to T_Reg cells based on surface receptor expression levels. Our approach relies entirely on Ab engineering, thus obviating the need for cytokine manipulation and keeping both cytokine/receptor affinity and cytokine activity intact. Furthermore, our strategy completely eliminates off-target effects by fully sequestering the cytokine rather than simply lowering its receptor interaction affinity. Although the allosteric receptor/Ab exchange mechanism we describe is specific to the IL-2/JES6-1 system, the structure-based design principles we used to engineer an effective single-agent cytokine/Ab fusion can be extended to other ligand/Ab interactions for exclusive targeting of soluble factors to specific cell subsets of interest.

Our engineering workflow offered new insight into the relationship between Ab/cytokine affinity and signaling activity in the context of allosteric exchange. Previous studies have elucidated correlations between cytokine/receptor affinity and signaling activity (60–65). A systematic study of the interplay between cytokine/receptor complex stability and membrane-proximal signaling for the IL-13/IL-13Rα1/IL-4Rα complex revealed that cytokine activity correlated directly with cytokine/receptor affinity only outside of a buffering region, an affinity regimen within 100-fold (in either direction) of the wild-type interaction affinity wherein cytokine activity was insensitive to affinity changes (65). The complexities of activity–affinity relationships in other systems led us to speculate that the Ab/cytokine affinity in the IL-2/JES6-1 single-agent fusion could also exhibit nonlinear behavior. Furthermore, because the exchange mechanism depends on the relative strengths between cytokine interaction with the IL-2Rα receptor subunit versus the JES6-1 Ab, we would expect the activity of the IC on various cell subsets to depend strongly on binding parameters.

We predicted biphasic behavior of the IL-2Rα^high cell (i.e., T_Reg cell) activity bias of our engineered JES6-1 mutants with respect to their IL-2 affinities; at very low affinities, the Ab would constitutively dissociate, resulting in unbiased activation of all IL-2–responsive cells, and at very high affinities, the Ab would never dissociate, blocking activity on all cell subsets (Supplemental Fig. 1C). By modulating the affinity of the IL-2/Ab interaction over nearly two orders of magnitude (Fig. 3B, Table I), we aimed to probe the window of Ab/cytokine affinity “tunability” for optimization of preferential T_Reg cell expansion. As illustrated in Fig. 3E, we indeed observed a biphasic activity curve as IL-2 affinity was varied, although the tuning range was found to be surprisingly narrow. JES6-1 mutants with up to a 2.2-fold decrease in IL-2 affinity compared with the parent Ab exhibited improved T_Reg selectivity, but no improvements were observed for mutants with IL-2 affinities that were reduced by 10-fold or more (Fig. 3E, Table I).

Other factors such as structural considerations also apparently contribute to the activity of IC mutants. For instance, the E60A and H100A mutants have similar affinities for IL-2 but diverge significantly in their activation potencies (Fig. 3E). The E60A mutation is in the V_H domain, whereas the H100A mutation is in the V_L domain, suggesting that the location of the interface disruption with respect to chain affects IC activity. Consistent with this observation, V_L mutants exhibit uniformly stronger signaling activity on IL-2Rα⁺ cells than do V_H mutants. This phenomenon could also be impacted by the topology of the fusion itself, as the greater proximity to tethered IL-2 for V_L compared with V_H may render the V_L interface more robust against affinity disruption. The dramatic (>10-fold) affinity losses observed with three of four V_H mutants compared with only one of four V_L mutants support the influence of topological factors on IC mutant activity (Table I). Further complicating the affinity–activity relationship is the lack of correlation between biased signal activation in vitro and selective cell subset expansion in vivo. Although the S34A+Y101A and Y41A+Y101A (JY3) IC mutants behaved similarly with respect to STAT5 signaling on IL-2Rα⁺ and IL-2Rα⁻ cells (Fig. 4B), there was a clear discrepancy in their in vivo promotion of IL-2Rα^high versus IL-2Rα^low cell growth (Fig. 4D, Supplemental Fig. 2B).

From a therapeutic development standpoint, the IC format has clear advantages over mixed complex administration as it eliminates dosing ratio considerations and concerns about the free cytokine inducing off-target effects and toxicities or undergoing rapid clearance from the bloodstream (43, 46). However, we unexpectedly found that our engineered IC elicited greater IL-2Rα^high cell expansion in an adoptive T cell transfer model (Fig. 5) and prevented DSS-induced colitis more effectively than the mixed complex (Fig. 6), even though the two formats induced similar T_Reg to effector cell expansion ratios (Fig. 4D, Supplemental Figs. 2C, 2E, 3A). One possibility for the superior phenotypic behavior of the engineered IC could be that it is positioned more optimally on the biphasic T_Reg to effector cell activity curve based on its altered Ab/cytokine affinity (Supplemental Fig. 1C). Alternatively, the more extensive IL-2Rα upregulation induced by JY3 IC versus the mixed complex (Supplemental Fig. 2D, 2F) may present an advantage for the IC by fueling the transcriptional feedback loop that perpetuates IL-2 signaling (17). Regardless of rationale, the enhanced behavior of JY3 IC over the mixed complex provides an immediately useful reagent for expanding T_Reg cells to combat autoimmune disease, and the structure-guided engineering strategy we used to develop this construct will inform the design of other mechanism-driven therapeutic ICs. With a growing number of anti-cytokine Abs in development, including those against IL-2, -4, -6, -7, and -15 (66), our novel approach can be readily extended to a broad range of cytokine/receptor systems for disease-relevant applications.

We thank members of the Garcia, Bluestone, Kovar, and Pande laboratories for helpful advice and discussions and A. Velasco and D. Waghray for technical assistance.

Provisional patents concerning the technology described in this work have been filed. V.S.P. is a consultant and SAB member of Schrodinger, LLC, and Globavir, sits on the Board of Directors of Apeel Inc., Freenome Inc., Omada Health, Patient Ping, and Rigetti Computing, and is a General Partner at Andreessen Horowitz. The other authors have no financial conflicts of interest.