Interview: Inside the IL-12 switch campaign

Kothai Parthiban, PhD, co-authored 'Conditional activation of IL-12 through a Fibronectin-EDB dependent switch gate', published in mAbs. The study describes a dual-specificity Fab that holds IL-12 inactive until it meets fibronectin extra-domain B (FN-EDB), a tumour-associated matrix antigen. The binding parameters were set by quantitative systems pharmacology (QSP) modelling before a candidate existed.

Thirty years of IL-12 attempts have run into the same toxicity ceiling: intratumoural dosing, half-life extension, protease-cleavable pro-drugs. Why is this approach different?

The historical challenge with IL-12, as with other potent cytokines, has been systemic toxicity, which significantly narrows the therapeutic window. Intratumoural dosing, half-life extension and protease-cleavable pro-drugs each address part of the challenge, but they leave a fundamental question of how to achieve sufficient and sustained IL-12 signalling at the tumour site without exceeding the threshold for systemic toxicity. The key difference with this approach is that IL-12 remains functionally masked in circulation and is only released upon binding to its target in the tumour microenvironment. This provides a more direct mechanism for restricting IL-12 activity to the tumour while minimising systemic exposure to active cytokine.

Protease-cleavable designs unmask irreversibly. Once cleaved, the molecule stays cleaved. This switch can close again. What does reversibility give you, and what does it cost?

The key advantage of reversibility is that it gives much greater control over where and when IL-12 is active. With an irreversible protease-cleavable design, once IL-12 is unmasked, it remains active even if the molecule subsequently leaves the tumour. So, if activation occurs in the wrong place or the molecule escapes into the circulation, there is no way to switch that activity back off.

With a reversible switch, IL-12 remains functionally inactive in circulation and is activated only upon engagement with the tumour antigen. Importantly, if it moves away from the target and the antigen is no longer present, the switch turns off again. In simple terms, the cytokine is active where the target is present and inactive where it is not.

The trade-off is that reversibility requires sufficiently strong and sustained target engagement to maintain IL-12 activity within the tumour. If the target affinity or tumour retention is insufficient, the molecule may switch off too frequently, reducing effective IL-12 exposure. The engineering challenge is therefore to balance target affinity, residence time and reversibility to achieve sustained tumour activity while minimising exposure to active IL-12 in peripheral tissues.

 In short, irreversible cleavage offers simplicity but no way back, whereas reversibility provides greater control. The challenge is to engineer that control carefully enough to maintain sufficient activity where it is needed while limiting systemic activity.

Why anchor the gate in the extracellular matrix rather than on the tumour cell surface?

Tumour cells are heterogeneous, and cell-surface antigens may be expressed at different levels. In contrast, the extracellular matrix is a more abundant and stable component of the tumour microenvironment. By anchoring the gate in the extracellular matrix, the approach is less dependent on consistent antigen expression across individual tumour cells. It also allows the molecule to remain positioned within the tumour microenvironment, where it can switch on IL-12 locally.

The binding parameters were fixed by modelling before there was a candidate to test them against. That puts a lot of weight on the model. What happens to the campaign if the window turns out to be unreachable?

The fact that the binding parameters were defined upfront, requiring high affinity for one target and moderate but meaningful affinity for the other, does place significant weight on the model. However, by combining smart antibody engineering approach with the sponsor’s machine-learning model, we were able to translate those predictions into actual molecules and experimentally achieve the predicted binding affinities.

 Importantly, our phage-display selection strategies, together with the CDR-shuffled library approach, were key to achieving the predefined benchmarks. This gives us confidence that QSP modelling can be translated into experimentally achievable antibody properties.

 That said, we would not treat the predicted window as a fixed assumption. The model defines the Mode of Action hypothesis, while the experimental data determine whether the desired window is actually achievable. If the initial modelled window proves too narrow, the engineering approach used allows us to adjust the binding parameters and iterate the design. In that sense, QSP guided engineering as used here is an effective approach for challenges such as this one.   

Two unrelated specificities in the same Fab, held in a defined ratio to each other. Affinity maturation is built to improve one thing at a time. What breaks when you ask it to do this?

The main challenge is that we are not optimising a single property in isolation. We are engineering two independent specificities within the same Fab while maintaining a defined functional ratio between them. At the same time, the molecule needs to recognise two unrelated targets in a competitive manner, retain neutralising activity, minimise polyreactivity, and maintain a precisely tuned affinity differential between the two targets.

Our library design and selection strategy are therefore critical. They allow us to control the sequence space and preserve the desired relationship between the two binding activities throughout the engineering process. The objective is not simply to maximise affinity for each target independently, but to identify molecules that achieve the right balance of affinity, specificity and function simultaneously. 

What breaks with a conventional affinity-maturation approach is the assumption that you can optimise one specificity first and then optimise the second without affecting the first. Here, the two activities are inherently coupled within the same molecule. The real engineering challenge is to optimise both functions while maintaining the defined affinity ratio and the required functional properties. That is why the library design and selection strategy are particularly important for this approach.

The switch behaviour is established in vitro. What are the next experiments that would test whether it holds in a tumour?

The next step would be to test the mechanism in tumour-relevant 3D models and, subsequently, in murine in vivo models. The 3D models would ideally incorporate the extracellular matrix together with relevant tumour and immune-cell populations to confirm that target engagement triggers IL-12 activation while the molecule remains functionally masked in the absence of the target.

In vivo models will be important for understanding where the molecule distributes, where IL-12 becomes pharmacodynamically active, and whether that activity translates into anti-tumour efficacy. Together, these experiments will help determine whether the switch behaviour observed in vitro translates into selective activation within the tumour microenvironment.

Under what conditions would this switch logic hold for another cytokine or another matrix antigen? And where would it fail?

The switch logic should be transferable to other cytokines or matrix antigens, provided the underlying biology supports similar requirements. For the cytokine, there needs to be a meaningful therapeutic benefit from local activation, with systemic activity contributing disproportionately to toxicity. For the matrix antigen, there needs to be sufficient and selective expression within the tumour microenvironment, together with adequate accessibility and retention to support sustained target engagement. The key is that the switch preserves the same fundamental principle, enabling conditional switching between antigen-bound and unbound states.

Where it could fail is at both the molecular and cellular levels. At the molecular level, the challenge is the engineering required to introduce two unrelated binding specificities within the same Fab while maintaining the desired affinity balance, specificity and function. At the cellular level, the approach could be limited if the matrix antigen is too heterogeneous, poorly accessible, or also expressed at significant levels in healthy tissues.

It could also be challenging if the cytokine requires continuous receptor engagement and the reversible switch does not provide sufficient residence time to sustain the desired pharmacology. Finally, if the cytokine has significant systemic activity even at very low exposure, there may simply not be enough therapeutic window for the switch to create meaningful separation between tumour activity and systemic toxicity.

So, the platform is potentially generalisable, but it is not cytokine- or antigen-agnostic. For each new application, the key questions are whether the target provides sufficient tumour selectivity and whether the cytokine biology is well suited to conditional, reversible activation.