A computational strategy has delivered a redesigned, more stable version of a cytokine protein that mimics the natural protein’s interactions with receptors, opening the way for designer cytokine-based therapeutics.

Messenger proteins called cytokines are secreted from cells and travel throughout the body. In the immune system, a cytokine called interleukin-2 (IL-2) delivers signals to receptors that regulate the activities of white blood cells. Cytokines such as IL-2 could therefore be used as therapeutics to modulate the strength of immune responses. However, the clinical uses of cytokines are often severely limited by properties that are inherent in their structures. Writing in Nature, Silva et al.1 describe how they have engineered a protein structure from scratch to replicate the beneficial receptor-binding properties of IL-2 without the drawbacks of the original cytokine.

IL-2 delivers its message by simultaneously binding to two receptor subunits known as IL-2 receptor β and IL-2 receptor γ (IL-2Rβ and IL-2Rγ). This binding pairs up the subunits to form a heterodimeric signalling protein called IL-2Rβγc. A third, non-signalling, receptor called IL-2Rα (also known as CD25) contributes to the formation of the signalling complex, strengthening the binding between IL-2 and IL-2Rβγcroughly 100-fold. IL-2 is used to treat some people with cancer, in whom it stimulates the cells of the immune system to work more effectively to eliminate tumour cells2. However, IL-2 has toxic side effects that limit its use3,4. The mechanisms responsible for the toxicity are not fully understood, but studies5 in mice indicate that the adverse effects are associated with the binding of CD25.

Several groups6 have therefore sought to engineer a variant of IL-2 that no longer binds CD25 but still promotes the formation of IL-2Rβγc. The standard engineering approach is to start with the natural protein and find a combination of mutations that results in a variant protein with the desired binding properties. But this approach has foundered when applied to IL-2. This is partly because the unmutated protein is not very stable, and, as Silva et al. demonstrate, the mutated proteins are typically even less stable, which is problematic for the manufacture and storage of a therapeutic agent. In addition, the mutated cytokines have reduced potency and exhibit residual binding to CD25.

Silva et al. therefore began afresh, and set out to design a protein structure from scratch that would provide a stable scaffold onto which they could add the structural elements required to produce the specific protein surfaces that bind to IL-2Rβ and IL-2Rγ. Crucially, these binding surfaces must be correctly positioned relative to each other in space to ensure that the designer cytokine engages the IL-2Rβγc heterodimer and triggers signalling.

The authors obtained information about the structural and spatial requirements of their designer cytokine by analysing the crystal structures of naturally occurring cytokine–receptor complexes7,8. IL-2 is one of a large family of cytokines that have at their core a bundle of four structural elements termed α-helices (Fig. 1a). These four α-helices are linked, in a defined order, by a series of short or long connecting loops. Instead of keeping this particular arrangement of α-helices and re-engineering the binding surfaces, Silva et al. reversed the process. They started by defining the positions of the all-important binding surfaces, and then used computational methods to design an arrangement of α-helices that not only links these surfaces but is also predicted to be stable.

The cytokine protein IL-2 has potential therapeutic applications, but its clinical effectiveness is limited in part by its instability. The protein has a core of four α-helices (shown as cylinders) connected by loops (not shown), and exerts its biological effects by forming interactions with target receptors at two regions (shaded areas) on its surface. b, Silva et al.1 used a computational strategy to redesign IL-2. They fixed the relative positions of the receptor-binding surfaces and then designed a stable four-helix bundle to present these surfaces to receptors. The designer protein is radically different in structure from the natural cytokine. For example, the connectivity of the four helices in the natural cytokine (left) is completely different from that of the analogous helices in the designer protein (right).

The proof of the pudding is in the eating, however. When the authors prepared and characterized the best candidates from the first round of design, the proteins showed promise in terms of IL-2Rβγc binding, but had fairly poor thermal stability. Clearly, the recipe required some improvement. Silva et al. went back to the drawing board, taking the best arrangement of α-helices from the first round and substantially extending the computational search for optimal loops to link them together. This second round of design-generated candidates had improved stability and exhibited excellent binding to IL-2Rβγc.

Silva et al. then carried out an additional, experimentally driven round of mutagenesis — a fine-tuning process in which single amino-acid residues are changed — to enhance the binding properties of the best candidate proteins, and then fully characterized the cytokine that had the highest overall binding affinity for IL-2Rβγc. The results are impressive. The final designer cytokine is highly stable and binds strongly to IL-2Rβγc, but not at all to CD25. Excitingly, this new protein is effective as a therapy in mouse models of skin and colon cancer, delivering the immunotherapeutic effects characteristic of natural IL-2, but with lower toxicity. The authors named their designer protein Neoleukin-2/15 (Neo-2/15), because it is a new cytokine that mimics natural interleukins 2 and 15.

The researchers then determined the crystal structure of Neo-2/15 in complex with IL-2Rβγc. Gratifyingly, the binding surfaces are positioned as designed, and the four-helix bundle matches the computational blueprint with almost pinpoint accuracy. The redesign of the interleukin’s four-helix bundle achieved by Silva et al. is remarkably radical: the order in which the α-helices are linked has been rearranged (Fig. 1b), and the amino-acid sequence of the resulting 100-residue protein is very different from that of either mouse or human IL-2.

It remains to be seen whether Neo-2/15 will deliver on its initial promise in the clinic. Moreover, perhaps the four-helix bundle is a particularly favourable case for re-engineering — other cytokine families that have more-complex architectures might be harder to redesign. Nevertheless, Neo-2/15 excitingly demonstrates that bold de novodesign, when combined with a deep knowledge of the structural determinants of receptor binding, can deliver designer cytokines that have bespoke binding properties. More broadly, Silva and colleagues’ approach to protein design has the potential for re-engineering any of the myriad biological systems that involve interactions between multiple proteins. In the meantime, the authors have opened up uncharted territory for therapeutics based on four-helix bundles, and there is plenty still to explore.