Picking the wrong target is the most expensive mistake in drug discovery. The recent DLBCL data are a reminder that picking the right one is not always enough.
Epcoritamab (Epkinly, from Genmab and AbbVie) is a CD20xCD3 bispecific antibody. One arm binds CD20 on the malignant B cell, the other binds CD3 on a T cell. It forces the two together and redirects the T cell to kill the tumour.
In the Phase 3 EPCORE DLBCL-1 trial, epcoritamab as a single agent improved progression-free survival over chemotherapy but missed its overall-survival endpoint. In EPCORE DLBCL-4, epcoritamab plus lenalidomide cut the risk of progression or death by about 60 percent against a standard chemotherapy regimen.
Same target. Same disease. The variable that changed was the state of the T cells doing the work.
The single-agent miss was on overall survival, not on tumour control. Progression-free survival still improved. The survival comparison was blunted because many patients in the control arm went on to receive effective salvage after progression, including the same drug class.
So the monotherapy result is not evidence that CD20xCD3 is the wrong mechanism. The mechanism worked. The trial simply could not convert that into a survival separation. The interesting scientific question is the ceiling on the mechanism, not its validity.
A bispecific engager does not bring its own immune cells. It recruits the patient’s T cells. In heavily pretreated relapsed lymphoma those T cells are exhausted, reduced in number, and sitting in a suppressive tumour environment.
A perfect engager still underperforms if the effector pool is spent. Raising the ceiling means improving the T cells, not re-engineering the antibody.
Lenalidomide is a molecular glue. It binds cereblon, the substrate-recognition part of a ubiquitin ligase, and reshapes it to recruit and destroy two transcription factors: Ikaros (IKZF1) and Aiolos (IKZF3). In T cells those two factors hold down the activation program. Removing them lowers the activation threshold and switches on the effector genes.
The two mechanisms meet at a shared effector engine. Lenalidomide (left) degrades Ikaros and Aiolos, lifting their repression of IL-2; epcoritamab (right) forms the synapse that puts a T cell on the tumour. The IL-2 axis in the centre is what the engager needs and cannot generate on its own. Edges are typed by evidence source.
The single most important consequence is IL-2. Ikaros and Aiolos are direct repressors of the IL2 gene, so degrading them releases IL-2 production.
IL-2 is the main growth and survival signal for activated T cells. It drives clonal expansion, sustains the activated state, and pushes cells toward a cytotoxic phenotype with more granzyme and perforin. It also supports NK cells.
This is exactly what a T-cell engager needs. The engager supplies the first signal by clamping CD3 onto the tumour cell. On its own, in exhausted cells, that activation produces only a weak and short-lived IL-2 response. Lenalidomide restores the IL-2 that amplifies and sustains what the engager starts. One drug provides the contact, the other provides the fuel.
IL-2 is a secreted cytokine, not an internal signal. The T cell exports it, and it then acts on IL-2 receptors on the cell surface in two modes: autocrine, where the same T cell responds to its own IL-2 and expands, and paracrine, where it recruits and sustains neighbouring T cells and NK cells in the same tumour region. So more IL-2 is not just self-help. It is a local amplifier for the whole effector pool the engager is trying to mobilise.
A useful contrast: giving IL-2 as a drug is toxic and tends to expand suppressive regulatory T cells. Lenalidomide raises IL-2 output from the T cell and, at the same time, destabilises the regulatory-T-cell program by degrading the same factors that support FOXP3, so the balance tips toward effectors rather than suppression. This is not a clean one-way switch: IL-2 is itself a growth signal for regulatory T cells, which carry the high-affinity receptor, so the extra IL-2 pushes the other way. The prevailing view is that the direct loss of FOXP3 support dominates, and the net effect favours effectors, but the tension is real.
Cereblon is not a T-cell protein. It is expressed across essentially every tissue. In our own reference data it is present, and highly expressed, in all 51 tissues measured.
So why is the effect selective? Because the ligase is only the machine. What matters is which factor gets degraded in a given cell and whether that cell depends on it. Same cereblon, different output: Ikaros and Aiolos in lymphocytes, CK1-alpha in del(5q) myelodysplastic syndrome, and SALL4 in the developing limb.
CRBN sits at similar levels in the T and NK cells the drug acts on and in unrelated cell types such as neurons, mast cells and kidney cells. Expression breadth does not predict where the effect lands. Values are mean expression from integrated scRNA-seq.
That last one is where expression breadth and toxicity finally meet. SALL4 degradation in the developing fetus is why thalidomide and lenalidomide are severe teratogens, controlled through strict pregnancy-prevention programs. The rest of the safety profile follows the same logic: myelosuppression (neutropenia and low platelets) because the drug acts on the blood and lymphoid compartment, venous thrombosis as a class effect, and, with long maintenance, a raised risk of second cancers. Added to a T-cell engager, the shared immune activation also raises the risk of cytokine release and tumour flare.
The lesson for anyone assessing a degrader: the expression map of the target protein is a poor guide to safety. A naive reading of cereblon, present in liver, heart, brain, and everywhere else, would predict broad organ toxicity that does not happen. The real risk lives in the substrate-dependency map, tissue by tissue, not in where the ligase sits. Safety has to be read at the level of what the drug does in each cell type.
None of this is hidden. Put the lenalidomide axis (cereblon, Ikaros, Aiolos) next to the engager’s effector axis (CD3, IL-2, interferon-gamma, granzyme) and the connecting path runs straight through the causal and regulatory biology, resolved to the cell types where it acts.
This is what Disease Atlas by Euretos is built to make legible: reading disease biology from primary data at cell-type resolution, so a combination rationale, and its safety logic, can be traced through mechanism rather than inferred from how often a gene appears in the literature.
The combination result was not luck, and it was not a trial-design trick. It was a predictable consequence of a mechanism with a known ceiling.
When a mechanism is capped by the fitness of the cells it depends on, the highest-value question is not “is this the right target” but “what raises the ceiling”. Read the biology cell type by cell type, and the second drug often picks itself.