Binding Is the Easy Part

An antibody–drug conjugate is not a targeted drug. It is a delivery problem with an antibody bolted to the front of it. Binding is the easy part, and it is almost never the part that fails. What decides whether a conjugate works is what happens in the hour or two after binding: how quickly the receptor is pulled into a coated pit, whether the complex is recycled back to the surface or committed to the lysosome, whether the lumen acidifies enough to activate the proteases, and whether the released payload can cross a membrane once it is free. Every one of those is a property of the receptor rather than of the antibody — which is why TROP2 supports two approved conjugates that behave nothing like each other, and why HER2-low tumours respond to one HER2 conjugate and not to another.

Antibody–drug conjugate internalisation, trafficking and the bystander effect — 37 clickable nodes from the antigen layer through endocytosis, endosomal sorting, lysosomal release, payload action and efflux. Open the interactive version to click any protein through to its ELISA kit, biosimilar or In Vivo antibody.

The antigen only gets you to the door

Target selection in this field is usually discussed as though copy number were the whole question. It is not even the most important part of it. %s works at modest surface density because it internalises constitutively and quickly; abundant, slowly internalising antigens have disappointed repeatedly. %s sits at the other extreme — the transferrin receptor takes up cargo continuously and at high rate, which makes it the reference point for internalisation-driven delivery and also the standing warning, because continuous recycling returns much of what it takes in.

Three other properties belong in the same conversation and are routinely left out of it. CEACAM5 is shed into serum in quantity, so circulating antigen absorbs conjugate before it reaches tissue. Claudin-18.2 is a tight-junction protein, accessible to antibody only once transformation has disrupted the junction — its selectivity comes from accessibility rather than from expression. EpCAM is abundant on carcinoma and abundant on normal epithelium, and the programmes built on it were limited by the normal tissue rather than by the tumour. Nectin-4 makes the same point in the clinic: the skin and ocular toxicity of enfortumab vedotin tracks where the antigen is, not where the disease is.

CD30 is the best-behaved antigen of the group — near-absent from normal tissue and rapidly internalising — and it is where the field learned what a cleavable linker plus a permeable payload buys you. CD33 is the cautionary one: the first approved conjugate, withdrawn, then reinstated on fractionated dosing. That reversal was a statement about saturation and delivery kinetics, not about the antigen. FOLR1 and EGFR round out the layer and both have companion-diagnostic stories attached — mirvetuximab soravtansine failed in a broad ovarian population and succeeded in a receptor-high one.

The sorting decision is where dose is silently lost

Most conjugates enter through clathrin-coated pits: clathrin recruits the receptor, dynamin-2 pinches the vesicle off, and the complex arrives at an EEA1-positive early endosome with nothing yet having happened. Dynamin inhibition is the standard control for showing that an ADC's activity depends on being taken up at all rather than on surface signalling.

From there the routes diverge, and one of them is a loss. Rab11 and FcRn return the conjugate to the surface intact. The FcRn mechanism is the awkward one, because the same pH-dependent binding that gives an ADC its long circulating half-life also rescues conjugate out of the tumour cell that just internalised it. That is a design tension rather than a defect, and it is invisible in any binding or uptake assay that does not distinguish arrival from retention.

The alternative is commitment: Rab5-to-Rab7 conversion sends the complex towards degradation, which on this pathway is the wanted outcome rather than a failure. caveolin-1 adds a third possibility that is easy to misread. Caveolar uptake can deliver a conjugate to a compartment that never acidifies properly, giving high measured uptake and low potency — a result that looks like a payload problem and is a routing problem. The way to separate them is a V-ATPase block, which abolishes potency without touching binding.

Release chemistry decides the bystander effect

Once the conjugate reaches an LAMP1-positive lysosome, the linker decides what happens next. A cleavable valine-citrulline linker cut by cathepsin B liberates membrane-permeable MMAE, which diffuses out of the cell that took it up and into neighbours whether or not those neighbours carry the antigen. That is the bystander effect, and it is why brentuximab vedotin and enfortumab vedotin work in tumours with heterogeneous antigen expression. Worth noting that cathepsin B knockout does not abolish release — the step is more redundant than the textbook diagram implies. legumain is where the next generation of linkers is going, for the straightforward reason that it is lysosomal and tumour-enriched.

A non-cleavable linker behaves completely differently. What it yields is a charged lysine–linker catabolite that cannot cross a membrane unaided and needs the lysosomal transporter SLC46A3 even to reach the cytosol of the cell that internalised it. Lose that transporter and trastuzumab emtansine stops working while trastuzumab deruxtecan carries on. It is worth sitting with what that means for a resistance work-up: antigen expression is unchanged, internalisation is unchanged, lysosomal delivery is unchanged, and the conjugate is inert.

Two payload classes, one execution step

Free payload in the cytosol acts on one of two targets. MMAE, MMAF and the maytansinoids bind β-tubulin, and because microtubule poisons only kill dividing cells, conjugates built on them underperform in slowly proliferating tumours. cyclin B1 is the earliest quantitative sign that a tubulin-directed payload arrived, accumulating as the spindle checkpoint holds cells at the G2/M boundary.

Deruxtecan and SN-38 instead trap topoisomerase I on DNA, converting a normal repair intermediate into a double-strand break at the replication fork. The cell-cycle requirement is far less strict, which is a large part of why deruxtecan conjugates succeeded where maytansinoids did not. γH2AX is the pharmacodynamic read-out, and its appearance in antigen-negative neighbours is the most direct available evidence that a bystander effect occurred. From there the damage signal runs through p53 to arrest or death, PARP1 is engaged for repair and then cleaved during apoptosis, and both arms converge on caspase-3 — which is why any assay downstream of caspase-3 cannot tell you which payload class did the work.

Permeability cuts both ways

P-glycoprotein is the price of a permeable payload. A molecule that can diffuse into a neighbouring cell can equally be pumped straight back out, and P-glycoprotein efflux is the commonest acquired resistance mechanism in this class. It is also the reason bystander killing and efflux resistance are not independent design choices: they are two consequences of the same property.

There is a second effect that most conjugates were not designed for and many of them have. Several payload classes trigger immunogenic cell death, exposing calreticulin on the dying cell as an eat-me signal and releasing HMGB1 as a danger signal. And because nearly every clinical conjugate retains a functional IgG1 Fc, CD16 on NK cells contributes ADCC alongside the payload. Part of the measured activity of an ADC is therefore not payload delivery at all, and blocking FcγRIIIa is how the two contributions are separated. In mouse work the equivalent control is the CD16/CD32 blocking antibody — without it, uptake by host myeloid cells is scored as antigen-specific delivery.

What to measure, and in what order

The practical consequence of everything above is that the informative experiments sit between binding and killing, and most programmes measure only the two ends.

Which reagents this map is actually made of

It is worth being exact here rather than implying coverage that is not there. The antigen layer on this pathway is served by the biosimilar range, not by the functional-grade In Vivo range: the naked antibody backbones of ten named clinical conjugates — trastuzumab, sacituzumab govitecan, datopotamab, mirvetuximab, zolbetuximab, tusamitamab, polatuzumab, brentuximab vedotin, gemtuzumab ozogamicin and enfortumab — supplied as research-grade biosimilars of the antibody, without the payload. They carry their clinical identity in the tooltip and deliberately do not carry the blue In Vivo dot, which is reserved for the functional-grade range.

Five of the thirty-seven nodes are blockable in vivo: EGFR, CD71, EpCAM, CD16 and mouse CD16/CD32. That last one earns its place on a map about delivery, because without Fcγ receptor blockade the uptake of a conjugate by host myeloid cells in a xenograft is indistinguishable from the antigen-specific delivery the experiment was designed to measure.

Everything between the door and the cytosol — clathrin, dynamin-2, the Rab switches, FcRn, LAMP1, V-ATPase, cathepsin B, legumain and SLC46A3 — is measurable here rather than blockable. On a trafficking pathway that is the correct division of labour: the experiment is to follow the conjugate, not to neutralise the machinery carrying it.

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