In order to engineer a CD25-independent version of IL-2, we displayed human IL-2 on the surface of yeast as a conjugate to Aga2p, and verified proper receptor binding properties with IL-2R β and γ_c ectodomain tetramers that were C-terminally biotinylated and coupled to phycoerythrin-conjugated streptavidin for use as a staining and sorting reagent^9,10. Yeast-displayed IL-2 bound to γ_c in the presence of IL-2Rβ, recapitulating the cooperative assembly of the heterodimeric receptor complex as seen with soluble IL-2 (Fig. 1a, Supplementary Fig. 1). We proceeded to carry out two generations of in vitro evolution (Fig. 1b, Supplementary Fig. 2). Our first generation in vitro evolution strategy was to create an error-prone PCR library of the entire IL-2 coding sequence (Supplementary Fig. 2), which resulted in selection of a predominant IL-2 variant containing an L85V mutation (Fig 1c, Supplementary Fig. 3).

From inspection of the wild-type (WT) IL-2 structure, we were surprised to find that position 85 was not a direct IL-2Rβ contact residue, but rather resided on the internal face of the IL-2 C-helix, within the hydrophobic core of the cytokine (Fig. 2a). Thus, we surmised that L85V may affect the structure of helix C in a way that enhances binding to IL-2Rβ. Therefore, we carried out a second generation selection where we made a biased library that contained F/I/L/V at amino acids L80, L85, I86, I89, I92, and V93, which are contained within the hydrophobic core and linker region on helix C (Fig. 1b,c). To rapidly select the most active variants, we used the yeast-displayed cytokines themselves to stimulate STAT5 phosphorylation in the human NK cell line YT-1 by coincubation at varying yeast:YT-1 cell ratios (Fig 1d, Supplementary Fig. 4). Several clones stimulated substantially more STAT5 phosphorylation at lower yeast:cell ratios than yeast-displayed WT IL-2 (Fig 1d, Supplementary Fig. 4). Sequencing of a selected panel of high-affinity IL-2 clones revealed a consensus set of mutations L80F/R81D/L85V/I86V/I92F (Fig 1c, Supplementary Fig. 3).

We expressed recombinant forms of several first and second-generation IL-2 clones in order to measure their binding affinities and kinetics for IL-2Rβ by surface plasmon resonance (SPR) (Fig. 1c, Supplementary Figs 3 and 5) and isothermal titration calorimetry (ITC) (Supplementary Fig. 6). By SPR, the affinity between IL-2 and IL-2Rβ was K_D = 280 nM. The IL-2 superkines, termed “super-2s”, clustered into low, medium, and high affinity classes. The highest affinity mutants had K_Ds of 1.2–1.7 nM (D10, H9). The affinity increases were uniformly manifested in reductions in off-rates (Fig 1c, Supplementary Figs 3 and 5).

In order to understand the structural consequences of the evolved mutations, we crystallized the D10 super-2 (Fig. 2a, Supplementary Fig. 7a, and Supplementary Table 1). In the structure of D10 alone, five of the six mutations clustered on the B-C loop and within the C-helix core, in positions that did not contact IL-2Rβ. Notably, the B-C-helix linker region was ordered in the electron density map (Supplementary Figs 7a and 8a), compared to other IL-2 structures where this region is often disordered (Supplementary Fig. 8a). Collectively, the F80, V85, and V86 substitutions appeared to collapse into a hydrophobic cluster that stabilized the loop by pinning the C-helix into the core of the molecule. Only one of the five consensus mutations, I92F, was at a position that contacted IL-2Rβ in the receptor complex (Fig. 2a), but it was deeply inserted between the C and A helices, contributing only an additional 10Ų of molecular surface buried by IL-2Rβ in the complex compared to Ile92. We also determined a low-resolution (3.8 Å) structure of the D10 ternary receptor complex to assess whether the mutations have perturbed the IL-2Rβ/γ_c receptor dimer geometry compared to the WT IL-2 complex (Fig. 2b, Supplementary Table 1). The overall IL-2Rβ/γ_c heterodimeric architecture and mode of cytokine/IL-2Rβ contact in the D10 ternary complex were essentially identical to the previously reported IL-2 quaternary assembly (RMSD = 0.43 Å) (Supplementary Fig. 7b).

Previously, we found that the C-helix of IL-2 appears to undergo subtle repositioning upon binding to IL-2Rα¹¹ (Fig. 2c, Supplementary Fig. 8a). Inspection of three WT unliganded IL-2 structures revealed conformational variability in the C-helix position, consistent with higher crystallographic B-factors in this helix relative to the rest of the molecule (Supplementary Fig. 8b). We compared the structure of our D10 super-2 to that of an unliganded structure of IL-2, and IL-2 in the receptor complexes. We found that the C-helix in D10 was more similar to that seen in the two receptor-bound conformations of IL-2 than the free forms, having undergone a relatively small shift towards the helical core as a consequence of the stabilizing mutations (Fig. 2c).

We used molecular dynamics (MD) simulations of IL-2 and D10 to further interrogate the mechanism responsible for higher binding affinity to IL-2Rβ by super-2 (Fig. 2d,e). We constructed an atomically detailed Markov state model (MSM) in order to directly probe the relative conformational flexibility of IL-2 versus D10. Analysis of the MSM clearly demonstrated that D10 was more stable than IL-2, and that IL-2 visited nearly twice as many clusters as D10. For example, D10’s most populated state had an equilibrium probability of ~0.20, compared to ~0.05 for IL-2, demonstrating that D10’s equilibrium population was far more localized than IL-2. Helix B, the B-C loop, and helix C appeared rigidified in D10 compared to IL-2 as evidenced by reduced RMSD from the starting conformations (Fig. 2d, Supplementary Movies 1, 2). F92 appeared to act as a molecular wedge between helix C and helix A, stabilizing the more C-terminal end of the helix (Fig. 2a). We also simulated both D10 and IL-2 starting in a receptor-bound-like structure and monitored the divergence in RMSD of the B-C loop and helix C from the actual receptor-bound structure. IL-2 (Fig. 2e-left, Supplementary Fig. 8c) quickly “wandered” from the receptor conformation and experienced drastic fluctuations compared to D10 (Fig. 2e-right, Supplementary Fig. 8c, Supplementary Movies 1, 2). Based on these observations, we propose a mechanism whereby the reduced flexibility of helix C in super-2, as a result of its improved core packing with helix B, results in a superior receptor-binding poise that increases its affinity for IL-2Rβ, and consequently mimics a functional role of CD25. The structural and MD results suggest that the evolved mutations in super-2 cause a conformational stabilization of the cytokine, that reduces the energetic penalties for binding to IL-2Rβ.

We asked if the super-2s demonstrated signaling potencies on cells in accordance with their IL-2Rβ binding affinities, and whether their activities depended on cell surface expression of CD25. We determined the dose-response relationships of WT IL-2 versus the super-2s 6-6, D10, and H9 on both CD25⁻ and CD25⁺ human YT-1 NK cells by assaying STAT5 phosphorylation with flow cytometry (Fig. 3a–d, Supplementary Fig. 9). On CD25⁻ YT-1 cells, the EC₅₀ of H9 and D10 were decreased over 10-fold (EC₅₀ = 2.5 and 1.8 ng/mL, respectively) compared to IL-2 (EC₅₀ = 39 ng/mL), with the 6-6 mutein yielding an EC₅₀ intermediate between IL-2 and H9/D10 (EC₅₀ = 15 ng/mL), consistent with the improved affinity of the super-2s for IL-2Rβ (Fig. 3a). On CD25⁺ YT-1 cells, the EC₅₀ of IL-2 decreased over 50-fold relative to CD25⁻ YT-1 cells, from 39 to 0.66 ng/mL (Fig. 3b). In contrast, the EC₅₀ of H9 and D10 improved only modestly in the presence of CD25 (EC₅₀ of 0.47 and 0.52 compared to 2.5 and 1.8 ng/mL, respectively) (Fig. 3b).

We sought to further probe the CD25-independence of the super-2s by taking advantage of a previously characterized mutation in IL-2, Phe42 to Ala (F42A), which showed reduced binding to CD25 by ~220-fold for H9 (K_D 6.6 nM versus 1.4 μM) and ~120-fold for IL-2 (K_D 6.6 nM versus 0.8 μM) (Supplementary Fig. 10)^12,13. The F42A mutation is an alternative diagnostic probe of the relative CD25 dependence of IL-2 and super-2. The F42A mutation right-shifted the dose-response curve of WT IL-2 on CD25⁺ cells by ~1 log, but had no effect on CD25⁻ cells (Fig. 3c). In contrast, H9 was far less sensitive to the F42A mutation, with the dose-response curves of H9 versus H9 F42A being very similar on both CD25⁺ and CD25⁻ cells (Fig. 3d).

We assessed the activity of several super-2s on T cells that were either deficient in, or expressed CD25. For the former experiment, CD4⁺ T cells were isolated from CD25-knockout mice, followed by stimulation by either WT IL-2 or six super-2s and assaying for STAT5 phosphorylation at a range of cytokine concentrations (Fig 3e, Supplementary Fig. 11). CD25^−/− CD4⁺ T cells responded poorly to exogenous WT IL-2 stimulation, but super-2s induced STAT5 phosphorylation in these cells proportional to their affinity for IL-2Rβ.

The principle functional effect of IL-2 is to promote T cell proliferation, particularly for naïve T cells. Human naïve CD4⁺ T cells were isolated and left either unstimulated or stimulated with plate-bound anti-CD3 antibody with or without the different IL-2 variants (Fig 3f, Supplementary Fig. 12). Increased proliferation effects on naïve human T cells correlated with increased affinity for IL-2Rβ and STAT5 phosphorylation shown earlier, as the rank order of potency was D10 = H9 > 6-6 > WT IL-2 (see Supplementary Fig. 12 for the complete titration).

We next tested the IL-2 variants for their ability to induce STAT5 phosphorylation on experienced human CD4⁺ T cells (Supplementary Fig. 13), which highly express the trimeric IL-2Rαβγ complex. Human CD4⁺ T cells were in vitro activated by T cell receptor (TCR) stimulation and rested to generate “experienced” human CD4⁺ CD25⁺ T cells. As for the CD25⁺ YT-1 cells, we observed a much smaller difference between IL-2 and the super-2s.

We assessed the potency of the super-2 H9 on expansion of CD25^low versus CD25^high T cells, in comparison to WT IL-2 and IL-2/anti-IL-2 monoclonal antibody (mAb) complexes, which have been shown to exert reduced pulmonary edema yet very potent anti-tumor responses in vivo^14–16. On antigen-experienced (memory-phenotype, MP) CD8⁺ T cells, expressing only low levels of CD25 but high levels of IL-2Rβγ, H9 induced more than three times the rate of proliferation and expansion as WT IL-2 (Fig 4a, Supplementary Fig. 14a). However, on CD4⁺ CD25^high T regulatory (Treg) cells, we found that the CD25-competent WT IL-2 and H9 achieved comparable maximal expansion, demonstrating again that expression of CD25 mitigates the difference between super-2 and WT IL-2 (Fig. 4a, S14b). Thus, the H9 exhibits the desired property that it shows enhanced stimulation towards CD8⁺ T cells, but not towards Treg cells compared to WT IL-2.

As previously reported, administration of high-dose WT IL-2 for five days induced substantial pulmonary edema, which is known to be CD25-dependent¹⁵ (Fig. 4b). Although significantly more stimulatory for cytotoxic CD8⁺ T cells (Fig. 4a), H9 super-2 caused substantially less pulmonary edema (Fig. 4b).

Given the more favorable properties of H9 in comparison to IL-2, we assessed its ability to stimulate effector functions of cytotoxic T cells in four different tumor models in vivo, where high-dose IL-2 administration has been previously shown to result in tumor regression^15,17. To this end, C57BL/6 mice were injected subcutaneously with B16F10 melanoma cells, followed by administration of either high-dose IL-2, IL-2/anti-IL-2 mAb complexes, or H9 super-2, once tumor nodules became visible and palpable. PBS-treated control mice rapidly developed large subcutaneous tumors reaching a volume of about 1500 mm³ on day 18 (Fig. 4c). As previously shown, high-dose IL-2 treatment was able to delay tumor growth by as much as 39% on day 18 (p<0.05), while IL-2/anti-IL-2 mAb complexes exerted very effective tumor control, reducing tumor growth by >80% on day 18 (p<0.005) (Fig. 4c). Significantly, similar to IL-2/anti-IL-2 mAb complexes, mice receiving high-dose H9 showed a dramatic decrease of tumor load on day 18, which was reduced by >80% compared to PBS (p<0.005) and by >70% compared to WT IL-2 (p<0.005) (Fig. 4c). Similar results were obtained using three other tumor models, including murine colon carcinoma and Lewis lung carcinoma injected subcutaneously (Fig. 4d,e) and B16F10 cells administered intravenously (Fig. 4f, Supplementary Fig. 15). Collectively, these data show that the H9 super-2 is very effective against different tumors, albeit inducing reduced pulmonary edema.

The practical implications are that this conformational nuance in IL-2 can be exploited for therapy. Super-2 robustly activates cytotoxic CD8⁺ T cells and NK cells for potent anti-tumor immune responses, yet it elicits minimal toxicity, suggesting that super-2 could warrant reconsideration for clinical applications of IL-2.

Error-prone and site-directed libraries of IL-2 were displayed on yeast as previously described¹⁸ and stained with biotinylated IL-2Rβ at successively decreasing concentrations. Staining was detected with streptavidin-PE and yeast were separated using paramagnetic anti-PE microbeads (Miltenyi; MACS). Enrichment of positively-staining yeast was monitored by flow-cytometry.

Human IL-2 variants and the ectodomains of IL-2Rβ, γ_c, and CD25 were expressed in Hi5 cells and purified as previously described¹¹. Proteins were concentrated to 8–20 mg/mL and crystallized by vapor diffusion in sitting drops. Diffraction studies were performed at the Stanford Synchrotron Radiation Laboratory and the Advanced Light Source. Crystal structures were solved by molecular replacement with PHASER¹⁹ and refined using PHENIX²⁰ and COOT²¹.

C57BL/6 and Thy1.1-congenic mice on a C57BL/6 background were maintained under specific pathogen-free conditions and used at 3–6 months of age. Experiments were performed in accordance with the Swiss Federal Veterinary Office guidelines and approved by the Cantonal Veterinary Office.

2–3×10⁶ CFSE-labeled CD44^high CD8⁺ T cells from Thy1.1-congenic mice were injected intravenously to Thy1.2-congenic animals. Mice received daily intraperitoneal (i.p.) injections of PBS, 20 μg IL-2, 1.5 μg IL-2/anti-IL-2 mAb complexes, or 20 μg H9 for 5 days. On the sixth day, spleens were removed and analyzed by flow cytometry.

Pulmonary edema was determined by measurement of pulmonary wet weight on the sixth day after five daily i.p. injections of PBS, 20 μg IL-2, 1.5 μg IL-2/anti-IL-2 mAb complexes, or 20 μg H9 as previously described¹⁵.

B16F10 melanoma cells, Lewis lung carcinoma, or murine colon carcinoma 38 cells were injected into mice (3–4 mice/group), as previously reported^15,17. Treatment consisted of five daily i.p. injections of PBS, 20 μg IL-2, 1.5 μg IL-2/anti-IL-2 mAb complexes, or 20 μg H9.