Your Agonist Antibody Is Not an Agonist

An agonist antibody does not activate its receptor. It gets held still by something else, and the receptors underneath cluster because they are crowded. That sentence sounds like a technicality and is in fact the whole pharmacology of the class. The TNF receptor superfamily has no enzymatic activity to switch on and no conformational lever to pull; its members signal when enough of them are brought close enough together for adaptors to be recruited to adjacent tails. A natural ligand such as CD40L or 4-1BBL arrives as a membrane-anchored trimer and is clustered further into higher-order arrays by the cell presenting it. A soluble, bivalent IgG cannot reproduce that on its own, however tightly it binds — so the clustering has to be supplied by a third party.

Agonist antibodies and FcγR crosslinking — 38 clickable nodes from the natural ligands and their TNFRSF receptors, through the Fcγ receptors that supply the clustering, to both NF-κB arms and the dose-limiting output. Open the interactive version to click any protein through to its ELISA kit, biosimilar or In Vivo antibody.

Proximity, not occupancy

4-1BBL makes the point more bluntly than any other member of the family: it is essentially inert in soluble form and active only when membrane-bound. It is not even a conventional trimer — it assembles as a three-bladed propeller — and the fact that the same protein works or does not work depending on whether it is anchored tells you that the receptor is counting neighbours rather than measuring occupancy. OX40L behaves the same way, and GITRL is unusual in being constitutively present on endothelium and many antigen-presenting cells, so its receptor carries a resting tone that CD40 and 4-1BB do not.

The consequence for antibody design is uncomfortable. Affinity maturation does not help, because the limiting quantity is geometry rather than binding energy. A bivalent IgG can bridge two receptor trimers, and two is not enough to recruit TRAF2 and TRAF5 to adjacent tails in a stable way. Nor does epitope choice rescue it on its own. What actually makes an antibody agonistic in vivo is whatever holds several of them in a dense lattice.

The inhibitory Fc receptor is the best crosslinker

The clustering is supplied by an Fcγ receptor on a neighbouring cell, and the identity of that receptor changes the outcome completely. The counter-intuitive winner is FcγRIIb — the inhibitory Fcγ receptor — precisely because it signals into nothing. It is a passive scaffold. It holds antibody molecules densely enough for the receptors underneath to cluster and contributes no signalling of its own to confound the result.

FcγRIIa and FcγRIIIa crosslink too, and both bring complications. FcγRIIa activates the myeloid cell doing the holding, which is where much of the cytokine release comes from. FcγRIIIa triggers ADCC, so an IgG1 agonist can delete the very cells it was meant to stimulate. On a regulatory-T-cell-rich target such as GITR that deletion may be the therapeutic mechanism; on an effector T cell it is the opposite of one. This is why isotype and Fc engineering change an agonist's behaviour more than its variable region does, and why potency measured on an FcγR-free cell line predicts remarkably little about what happens in tissue.

It also means the agonism belongs to the tissue rather than to the molecule. The same antibody in a FcγR-rich compartment and an FcγR-poor one is, functionally, two different drugs. ICOS is instructive by contrast: it is a CD28-family member that signals as a dimer through PI3K and does not require higher-order clustering, which is why an ICOS agonist can work without any FcγR contribution at all.

Read the slow arm, not the fast one

Clustering feeds both NF-κB arms, but only one of them reports on it. The canonical arm through IKKβ to RelA is fast, transient and shared with essentially every inflammatory receptor on the same cell — including TLR4. That overlap is not academic. An agonist antibody preparation carrying endotoxin produces a convincing canonical NF-κB signal, a convincing dendritic-cell activation phenotype and a convincing cytokine profile, all for entirely the wrong reason.

The non-canonical arm is different in kind and that difference is what makes it useful. TRAF3 and cIAP1 hold NIK at vanishing concentrations by continuous ubiquitination and degradation rather than by keeping it inactive. Receptor clustering destroys that complex, and NIK then accumulates because destruction stopped, not because anything positive was switched on. Processing of NF-κB2 from p100 to p52, and the appearance of RelB in the nucleus, then takes hours and requires the lattice to hold throughout.

That slowness is the assay. It is the step a non-crosslinked antibody cannot fake, and it is the reason a late time point tells you more here than an early one. In myeloid cells specifically, TRAF6 carries the CD40 signal, which is why CD40 agonism licenses dendritic cells and macrophages rather than acting on T cells directly — the therapeutic effect of an anti-CD40 antibody is almost entirely indirect.

What a licensed cell and a rescued T cell actually do

Downstream, the licensed antigen-presenting cell raises CD80, CD86, MHC class II and IL-12. Those four are what convert agonism into antigen presentation rather than into undirected inflammation, and CD80 and CD86 are blockable, so the contribution of presentation can be tested rather than assumed.

On the T-cell side the signature is survival rather than proliferation. Bcl-xL and survivin are induced by 4-1BB and OX40 signalling, and what this receptor family principally does is stop effector T cells dying. That is a different pharmacology from co-stimulation through CD28, with a different time course, and it is why the effect of an agonist antibody outlives its own clearance. IL-2 neutralisation separates survival-driven accumulation from genuine expansion, and IFN-γ reports that the type 1 programme was achieved at both ends of the interaction.

The toxicity is the same mechanism in the wrong place

The dose-limiting effects of this class are not off-target. 4-1BB agonism produces the strongest effector expansion of the group and the most hepatotoxicity, and the two have never been cleanly separated: granzyme B raised in the liver looks exactly like granzyme B raised in the tumour. Cytokine release comes from the same FcγR-bearing myeloid cells that were doing the useful crosslinking, which is why IL-6 and TNF rise alongside the intended effect rather than independently of it.

CD8 depletion is the standard test of whether an anti-tumour effect is cell-mediated at all, and on CD40 agonism the answer is sometimes no — macrophage-dependent responses exist and are indistinguishable from outside. CD27 is the member of this group where agonism has been relatively well tolerated clinically, and the contrast with CD40 is worth sitting with: identical crosslinking requirement, very different therapeutic index.

The one control that settles it

Everything above turns on a single question — is this antibody intrinsically agonistic, or is it borrowing agonism from a bystander cell? There is one reagent pair that answers it outright. Oxelumab, the anti-OX40L antibody, is stocked both as a standard In Vivo antibody and as an Fc-reduced variant of the same clone. Same clone, same epitope, same affinity, FcγR engagement removed. Run the pair: activity that survives the swap was intrinsic, and activity that disappears was borrowed.

The in vivo counterpart is the mouse CD16/CD32 blocking antibody, which removes the scaffold in the animal rather than from the molecule. Together they give the positive and negative arms of the same experiment, and they are the reason this pathway is unusually tractable for a question that is normally argued from structural inference.

Which reagents this map is made of

Fourteen of the thirty-eight nodes carry a functional-grade In Vivo antibody, and on this pathway they sit where the decisions are. The agonist receptors themselves are covered: CD40, OX40, 4-1BB, ICOS and CD70. The CD40 antibody is clone FGK45 — the one the field standardised on — and it is stocked ultra-low endotoxin, which here is a requirement rather than a refinement, for the reason given above. TNF is stocked ultra-low endotoxin for the same reason: LPS induces it directly and the induction would be scored as the agonist's effect.

For reading out and dissecting the consequence: CD80, CD86, IL-2, IFN-γ and CD8α. Said plainly, what is not blockable in this range: the Fcγ receptors themselves, every TRAF, NIK, both NF-κB subunits, IL-12, IL-6, Bcl-xL and survivin. GITR and CD27 are served by the ragifilimab and varlilumab biosimilars rather than by the functional-grade range, and carry no In Vivo dot for that reason. The workable design is therefore to manipulate the Fc and read the p100-to-p52 step — which is exactly what this range supports.

Two adjacent maps in this library are worth reading alongside this one and are deliberately not redrawn here: T-cell co-stimulation covers what these receptors do, and the TNF receptor superfamily map covers TNFR1 against TNFR2. This one is about something narrower and more practical — why an antibody against any of them only works when something else is holding it still.

For research use only; not for use in diagnostic or therapeutic procedures. Explore the full library of interactive pathway diagrams.