Custom Antigen Ligands: Painting a Precise Portrait of 'Unruly' Targets

Targets don’t always just sit there and 'wait patiently.'

In classic drug discovery textbooks, targets are often portrayed like this: they quietly stay on the cell surface or inside the cell, waiting for scientists to 'hit the bullseye' with a drug molecule. This narrative is simple and inspiring—but compared to the real world, it's a bit idealized.

In reality, many of the most valuable targets aren’t so 'cooperative.' They frequently switch shapes, like a door that sometimes swings open and sometimes stays shut; their expression varies dramatically across different tissues; their binding pockets are tightly occupied by endogenous ligands, making it extremely hard for an outsider drug to 'cut in line'; and some targets only reveal critical attack sites when they 'shake hands' with another protein—when alone, those sites almost physically don’t exist.

This creates a tricky situation: we use recombinant proteins to 'mimic' targets for screening, but these mimics might not look much like the real target in its natural state. The molecules we screen end up binding to 'the version of the target you gave them,' not necessarily 'the target inside a patient’s body.'

Custom antigen ligands aim to solve this engineering problem: how to create a 'stand-in' for these tricky targets—one that presents the most critical epitopes to candidate drugs in the right conformation, with the right quaternary structure, and in an environment that closely resembles the natural one. This isn’t just simple protein purification; it’s a proactive design of the target.

Global drug development is facing the double pressure of target depletion and druggability bottlenecks. Known 'druggable' targets account for only a small fraction of potential therapeutic targets. In the GPCR superfamily, many orphan receptors and class B/adhesion receptors remain hard to drug due to structural and expression complexities; ion channel subfamilies like TRP and ASIC face similar technical barriers. The development of custom antigen ligand technology is essentially an attempt to knock on this tightly closed door.

antigen ligand

1. When you "extract" a target, what exactly are you losing?

The default process for traditional antigen preparation is: clone the extracellular domain of the target → recombinantly express it → purify it → use it for immunization or screening. Behind this process hides an often unverified assumption—"the recombinantly expressed extracellular domain can represent the true appearance of the natural target."

For many targets, this assumption may not hold.

Take multi-pass transmembrane proteins (like GPCRs or ion channels) as an example. The structure of these proteins is determined by the overall arrangement of the transmembrane helices, and the conformation of extracellular loops depends on structural constraints from the transmembrane regions, support from the lipid bilayer environment, and stabilization by endogenous ligands. When you express the extracellular domain alone, it's like tearing a page out of a 3D pop-up book—the page loses the structural support of the whole book and struggles to fold into the 3D shape it had in the full book. What you get is a protein fragment "doing its own thing" in solution, and its conformational ensemble might differ significantly from its natural state. Antibodies screened against this fragment are likely targeting the "recombinant protein" rather than the "native protein on the cell surface." At the same time, expressing the extracellular domain alone loses native post-translational modifications like glycosylation, which often contribute directly to epitope formation, further widening the gap between recombinant antigens and natural targets.

Targets with multiple conformations face another dilemma. Some kinases switch between active (DFG-in) and inactive (DFG-out) states, with the ATP binding pocket changing by more than 10 Å. If your antigen protein in solution predominantly adopts the inactive conformation—or is a mixture of multiple conformations—the binders you screen might target the wrong conformational state. This means that candidates screened over half a year could be weak or even inactive at the cellular level.

Even trickier are complex epitope targets. Some of the most therapeutically valuable epitopes only get exposed or formed when the target interacts with another protein. Using a monomeric target as the antigen makes it physically impossible to present this epitope—no matter how many rounds of screening, finding binders for it is very difficult.

It’s worth pointing out that these three situations aren’t mutually exclusive—a single target may have several of these characteristics, piling on the difficulty. However, they all lead to the same conclusion: the antigen may no longer be a "given material extracted from nature," but rather a "tool engineered according to screening purposes." Targets need to be "presented" in a specific form—ideally with the correct conformation, the correct quaternary structure, and the correct post-translational modifications.

GPCR signaling pathways mediated by Gαprotein subunits

2. Make the Target 'Freeze' at the Most Worthwhile Moment to Aim For

Since the target moves, the idea is to find a way to make it 'stand still'—and ideally, stand still at the moment that matters most. This is the core logic of conformational locking. For GPCRs, the conformation when bound to an agonist and the conformation when bound to an antagonist may involve significant rearrangements of the transmembrane helices, and the positions of the extracellular loops differ noticeably between the two states. If you want an agonist-like activating antibody, you need to stabilize the antigen in the active conformation; if you want an antagonist, you need to lock it in the inactive conformation. Molecules screened with the wrong antigen conformation could go completely off-track—just like the conformationally diverse target dilemma mentioned in the previous section.

There are mainly two approaches to conformational locking. One is ligand-assisted locking—using a known pharmacological ligand (agonist, antagonist, or allosteric modulator) to stabilize the target in a specific conformation. This approach requires having a high-affinity, highly selective ligand available. The other is protein engineering locking—introducing stabilizing mutations (such as designing disulfide bonds to 'lock' cysteines that are close in space together, or adding capping sequences at helix ends) to reduce conformational flexibility. This requires sufficient knowledge of the protein structure to know exactly where to 'nail it' to lock the correct conformation, not accidentally lock the wrong conformation as well. In the GPCR field, conformational thermostabilization strategies, exemplified by StaR® technology (systematic scanning and combinatorial stabilizing mutations), have been successfully applied to solve the structures and discover drugs for dozens of GPCRs.

It must be said honestly that conformational locking can increase the proportion of the target in the desired conformation, but it’s usually hard to achieve a 100% single-state population. This adds extra requirements for downstream screening—for example, including competitive elution steps during screening to try to exclude molecules that bind to non-target conformations.

3. Give the Target a 'Simulated Stage'

Different formats for presenting membrane proteins for affinity selection experiments

Beyond the issue of conformation, there’s another dimension of deviation that needs to be addressed: the microenvironment of the target.

Multipass transmembrane proteins in their natural state are embedded in the lipid bilayer of the cell membrane. The lipid environment isn’t just a ‘background’; it helps shape the protein’s conformation, affects the orientation of extracellular loops, and in some cases even directly forms part of the epitope. Antibodies screened using just the extracellular domain as the antigen may have trouble recognizing the native target in its lipid environment—because once removed from the membrane, epitope orientation might change, or certain regions that are exposed only on the membrane surface may not be presented at all.

One mature approach for mimicking the membrane environment is the nanodisc: it wraps a small patch of lipid bilayer with membrane scaffold proteins to form a nanoscale disc, into which a single transmembrane protein can be embedded. This is like giving the target a ‘mini version’ of the cell membrane in vitro. Another approach is virus-like particles, which use the self-assembly property of viral structural proteins to densely display the target protein on the particle surface, particularly useful for enriching and screening low-abundance membrane proteins.

Aside from the previously mentioned complex epitope scenarios formed by interactions of different proteins, there’s another often-overlooked dimension: the quaternary structure of the target protein itself, meaning the target’s ‘assembly style.’ Many membrane proteins function naturally as dimers or higher-order oligomers. If you screen using a monomeric antigen, the antibodies obtained might fail to recognize or block the target in its native state—because the functional epitope is formed jointly by two subunits. Tools like Fc fusion (dimerization) and Foldon trimerization domains can achieve controlled oligomerization—making sure to match the target’s natural oligomeric state; for more complex assemblies like tetrameric ion channels, co-expression strategies or virus-like particle display may be needed to approximate the native state.

These strategies come with trade-offs: constructing nanodiscs or virus-like particles is much more complex than just coating recombinant protein, and oligomerization designs increase expression and purification difficulties. But for 'hard-to-drug' targets, it’s often worth the investment—after all, even if the screening goes smoothly, molecules selected with an inaccurate antigen are less meaningful.

4. Shifting from a 'trial-and-error' approach to a 'calculate first, test later' mindset

By the time you’ve read this far, you’ve probably already felt how "grueling" the traditional development model for custom antigen ligands can be: you have to try different truncation boundaries, different mutation sites, different multimerization schemes—every combination of variables requires construction, expression, purification, and validation. A full cycle can take weeks or even months, and the failure rate isn’t low.

This is exactly where computational tools bring a change.

Structural prediction can provide more precise guidance for truncation design—before experiments even start, it can indicate potential boundaries for transmembrane regions, predict which hydrophobic surfaces might be incorrectly exposed after truncation, assess the area of exposed hydrophobic surfaces, and anticipate solubility and thermal stability risks. Structural prediction can also somewhat analyze the stabilizing effect of ligand binding on target conformation—whether the flexibility of surrounding residues is constrained after the ligand occupies the binding site. It’s worth noting that structural prediction still has inherent limitations in predicting large-scale conformational changes induced by small molecule ligands, so the results can’t replace experimental validation.

Furthermore, protein function prediction models can evaluate the feasibility of a design before construction: which truncation plan is more likely to yield a soluble, correctly folded protein? Will the introduced stabilizing mutations effectively reduce conformational flexibility, or might they disrupt key domains? This is like giving antigen design a "pre-screening" system—it’s not about ruling out failed plans after experiments, but locking in the candidates more likely to succeed before any lab work begins.

Take the MatwingsVenus™ (Xiaowu™) agent as an example. The capabilities of this kind of AI platform are built in exactly this direction. Its starting point is precise antigen truncation design—for all kinds of target proteins, the accuracy of truncation boundaries directly determines the quality of the recombinant antigen: truncating too long may include unnoticed hydrophobic sequences, causing expression failure; truncating too short may lose key epitopes. The MatwingsVenus™ (Xiaowu™) agent can mark transmembrane boundaries, identify potentially incorrectly exposed hydrophobic surfaces, and anticipate conformational changes after truncation during the design stage.

On top of this, the platform’s protein function prediction modules (VenusX/VenusG) elevate antigen optimization from "excluding failed plans" to "pre-selecting the right ones." They can output feasibility indicators for truncation designs before expression constructs are made—including solubility levels, thermal stability predictions, and hydrophobic patch distribution on the surface—helping researchers anticipate which truncation scheme is most likely to produce a soluble, correctly folded recombinant protein. For designing stabilizing mutations (like disulfide locking or replacing surface hydrophobic residues with hydrophilic ones), function prediction can also evaluate their potential impact on conformational stability and expression levels.

It’s worth mentioning that the computing power of these platforms isn’t limited to antigen design. Any demand that requires customized protein engineering solutions can be addressed within this unified structure reasoning framework. Of course, the role of these tools is to assist, not to replace—final decisions still need to be validated experimentally. But at least, the starting point for experiments can change from a 'needle in a haystack' approach to more like 'setting sail with a map.'

In conclusion: The era of target engineering is arriving.

Custom antigen ligands are essentially a form of target engineering—it’s no longer 'we use whatever the target naturally is,' but 'we aim to make the target present itself in the way we need.' This isn’t going against natural antigens, but rather building a methodological bridge between natural targets and engineered screening. When conformations can be locked, when quaternary structures can be reconstructed, when membrane environments can be simulated—those targets once set aside as 'too difficult' might finally get back on the agenda.

Targets aren’t always as 'lying flat' and waiting to be attacked as textbooks make them out to be. They’re dynamic, variable, with their own structural 'moods.' That’s precisely why tools that can handle this complexity are so valuable. Next time you hear about a GPCR or ion channel antibody drug successfully hitting the market, chances are there’s a group of carefully designed custom antigen ligands behind it—they’re silent and precise, marking the very beginning of that long journey in drug development.