Alloy’s perspective on the multispecific engineering problem
Most protein engineers designing multispecifics today learned the craft on monoclonals, and some of the discovery and development infrastructure followed the same path. That’s a reasonable starting point. It’s also where a lot of the late-stage surprises come from.
Multispecific antibodies aren’t just harder monoclonals. They’re a fundamentally different class of molecule, and programs that treat them as an incremental engineering problem tend to find that out the hard way.
What Makes Multispecifics Different
Conventional antibody therapeutics leverage millions of years of evolution to their advantage. Immunize the right animal, screen broadly, and nature has done most of the work. The molecules that come out are often already optimized for binding, expression, stability, and manufacturability.
Multispecific antibodies don’t have that. No animal produces a molecule that engages two distinct targets through a single scaffold (IgG4 half-molecule exchange being a caveated exception), because there’s never been an evolutionary reason to. Every multispecific format in use today is a human invention, and they all share one property: solving a protein engineering problem in one place tends to create a new problem somewhere else.
There are several conventional approaches: common light chain, heavy and light chain heterodimerization mutations and domain swaps, pairing-free scFv and VHH-based designs, among others. None of these work universally across a given pair of antibody clones. They work often, somewhere in the range of 50-80%, but not always. That tells you something important: multispecific engineering is still fundamentally an empirical problem, not settled science. Common light chain makes downstream engineering more tractable, but constraining the light chain during discovery can sometimes make hit identification more challenging. scFvs reduce pairing complexity but introduce biophysical liabilities. VHH domains sidestep the pairing problem but carry immunogenicity and developability considerations that need active management. Every solution trades a pound of flesh somewhere.
The practical consequence is that format selection is a deliberate choice. It has to be grounded in the specific target biology, affinity and avidity requirements, route of administration, and what the molecule will encounter in vivo. Choosing by default or by analogy to a previous experience with different biology is one of the more costly mistakes a multispecific program can make early on.
How to Navigate the Combinatorial Complexity
The toolset has gotten significantly better as the field compounds its experience. Alloy’s unique mAbForge™ Multi can screen thousands of format variants in parallel, with results in four weeks. That matters because multispecific activity and manufacturability remain difficult to predict computationally. The experiments have to be run to determine how a given format and VH/VL pair will express, purify, and bind in the pharmacologically relevant orientation. Generating that data across many variants at once, early in the program, is a different kind of de-risking than committing to a format and waiting months to find out if it works.
Discovery infrastructure has evolved in parallel. Common light chain transgenic animals, phage- and yeast-displayed antibody libraries, and VHH-specific platforms have expanded what’s accessible for constrained discovery campaigns. Extensive characterization of both the building blocks and the final bispecific for function and drug-like properties creates early visibility into manufacturability and developability fitness, including for subcutaneous delivery where problems tend to surface at the worst possible time.
What the toolset doesn’t replace is accumulated judgment. For teams without deep prior exposure to multispecifics, the decisions at the front of a program can look familiar. They’ve made calls on format, affinity, and developability before. The difference is that in multispecific programs those calls carry more downstream consequence and the feedback loop is longer. The ability to design multiple robust engineering streams, move with high velocity to generate data, select the winning format, and iterate again if necessary, is how Alloy approaches these programs.
What the Field Still Needs
Where the field is genuinely behind isn’t engineering. The tools for making complex molecules have advanced considerably. Understanding biology and having translationally relevant models remains the bigger challenge.
A bispecific targeting two pathways needs a disease model where both human targets are expressed and active in a disease-relevant context. A trispecific needs three. In many cases, that animal doesn’t exist. Programs end up validating one mechanism at a time, which is a real limitation when the scientific rationale depends on the combination. The most complex multispecifics being designed today are ahead of the model systems available to test them in any translationally meaningful way.
Advances in genetic engineering and CRISPR-based humanization are helping close that gap. And the data generated from the programs Alloy runs across formats, targets, and indications is part of how the field is building its way there. Every program teaches something the next one benefits from. The validation playbook for multispecifics is being written right now, one partner program at a time and getting it right requires connecting early discovery and engineering decisions to in vivo outcomes from the start. That’s a thread that runs through every multispecific program we take on.
If you’re navigating any of this, from format selection to building a full multispecific discovery campaign, we’re sharing what we’re learning at our upcoming webinar. Come see the thinking in practice.
Sign up for our upcoming webinar
Register for Alloy’s June 23rd multispecifics webinar. Three vignettes on what’s next in multispecific antibody engineering.