CAR-T Cell Signalling Pathway: Costimulation, Output and Assays

A chimeric antigen receptor is a synthetic protein wired into a real signalling pathway. The extracellular half is an antibody fragment; everything below the membrane is borrowed from the T-cell receptor complex and its costimulatory partners. That hybrid design is why CAR-T therapy works, and why its two defining problems — cytokine release syndrome and loss of persistence — are signalling problems rather than manufacturing ones. Reading the pathway from CD19 engagement to granzyme B release shows where each failure is generated.

Recognition without MHC: the CAR module itself

The targeting half of a second-generation CAR is a single-chain variable fragment against CD19, joined through a hinge and transmembrane segment to a costimulatory tail and then to CD3ζ. Recognition is antibody-mediated, so it is independent of peptide–MHC — which removes the two commonest escape routes, HLA loss and defective antigen processing. CD19 is canonical not because it is tumour-specific but because the on-target toxicity, B-cell aplasia, is survivable.

The synapse a CAR forms is not the classical one. There is no bull's-eye of concentric supramolecular clusters; the contact is disordered, assembles in minutes and disassembles quickly. That geometry is why CAR-T cells serially kill faster than conventional cytotoxic T cells at the same effector-to-target ratio, and why the receptor is comparatively insensitive to low antigen density: below roughly a thousand molecules per target cell, most CD19 CARs stop triggering.

The subtler failure mode is tonic signalling. scFv domains that self-associate cluster in the membrane and let LCK phosphorylate CD3ζ with no antigen present. The product then reaches the patient already partly differentiated and carrying an exhaustion programme. Antigen-independent CD3ζ phosphorylation measured during manufacture is one of the more predictive quality attributes a CAR has — and it is a property of the binder, not of the signalling tail.

The proximal module: LCK, ZAP-70 and the LAT signalosome

Once the receptor clusters, LCK phosphorylates the paired tyrosines of the CD3ζ ITAMs, creating the bidentate site read by the tandem SH2 domains of ZAP-70. From that point a CAR and a TCR are indistinguishable to the cell, with one exception: a CAR carries three ITAMs against the intact CD3 complex's ten. Amplification is correspondingly lower, and deliberately inactivating two of the three has been shown to improve persistence rather than harm it.

Active ZAP-70 phosphorylates LAT, a transmembrane adaptor with no catalytic activity that works purely as a scaffold. It recruits SLP-76, and together they position PLCγ1 to hydrolyse PIP₂ into IP₃ and diacylglycerol. Every branch below derives from those two messengers, so LAT occupancy is a better proxy for functional signal strength than receptor expression is. CAR signalling is also faster and more intense than TCR signalling but decays sooner, which makes kinetics more informative than endpoint measurements.

Costimulation: CD28 versus 4-1BB

Signal one alone produces a T cell that kills once and then dies. CD28 supplies signal two by recruiting PI3K through its YMNM motif and activating AKT and then mTOR: rapid aerobic glycolysis, fast expansion, a strong early effector burst, and typically a shorter-lived product with a steeper cytokine curve in the first week.

4-1BB (CD137) does not use PI3K at all. Its tail binds TRAF2, which nucleates a ubiquitin-dependent complex that activates IKK and drives NF-κB, pushing cells towards mitochondrial biogenesis, fatty-acid oxidation and central memory differentiation. Because the two endodomains feed different transcription factors rather than the same one at different strengths, AKT and mTOR read out a CD28 construct while TRAF2 and nuclear NF-κB p65 read out a 4-1BB one. Comparing both on a single panel is the commonest reason costimulation experiments produce uninterpretable data.

Three transcriptional branches: NFAT, NF-κB and AP-1

IP₃ empties the ER calcium store and triggers store-operated entry; sustained Ca²⁺ activates calcineurin, which dephosphorylates NFAT and exposes its nuclear localisation signal. What NFAT then does depends on its company. Partnered with AP-1 it drives the effector programme; unpartnered, as happens under chronic antigen exposure, it induces the exhaustion-associated transcription factors instead. Nuclear NFAT is therefore not a straightforward activation marker.

Diacylglycerol takes the other route, recruiting PKCθ to the synapse to phosphorylate CARMA1 and nucleate the CBM complex with BCL10 and MALT1. That complex activates IKK, degrades IκB and releases NF-κB onto survival genes and IL-2. This is the branch 4-1BB reinforces from the side, so the CBM components belong in persistence experiments rather than in cytotoxicity ones.

The third branch runs from diacylglycerol through Ras, Raf, MEK1/2 and ERK1/2 to AP-1. Since AP-1 is the partner NFAT needs, the ratio of ERK output to calcium flux sets whether a stimulated cell commits to effector function or drifts towards exhaustion. Constructs over-expressing AP-1 components restore function in chronically stimulated cells — the clearest evidence that this is a stoichiometric problem, not a signal-strength one.

Output: killing, cytokines and the inhibitory brakes

The cytotoxic output is deliberately redundant. Granule exocytosis delivers perforin and granzyme B into the target, while FasL engages Fas for a slower, granule-independent route. Losing either arm alone rarely abolishes clearance in vivo, so a construct that fails to kill has almost always failed upstream, at synapse formation or proximal signalling.

The cytokine output is where the therapy becomes dangerous. CAR-T cells secrete IL-2, IFN-γ and TNF-α; those cytokines activate monocytes and macrophages, which produce the bulk of the circulating IL-6 that defines cytokine release syndrome. IFN-γ and TNF-α report CAR-T activity, IL-6 reports the myeloid response to it, and the two dissociate in time.

Chronic exposure then upregulates PD-1, LAG-3 and TIM-3. PD-1 recruits SHP-2, which preferentially dephosphorylates CD28 rather than CD3ζ, so its dominant effect is to remove costimulation rather than signal one. CD28-based CARs are therefore intrinsically more sensitive to PD-1 blockade than 4-1BB-based ones, and combination experiments should be planned accordingly.

Key targets and matching reagents

Studying CAR-T signalling in vivo

Most tractable interventions in this pathway are receptor blockade, costimulatory engagement and cytokine neutralisation. In syngeneic and humanised models, experiments fall into three groups.

1. Manipulating the target and the costimulatory input

Anti-CD19 defines the antigen axis and is the reagent of choice for confirming that a phenotype is target-dependent rather than construct-intrinsic. Anti-CD28 and anti-4-1BB let you engage or block the two costimulatory arms separately, which is the cleanest way to test whether a persistence difference really tracks with the endodomain. Agonist antibodies to costimulatory receptors are exquisitely sensitive to endotoxin-driven background activation, so low-endotoxin functional-grade material is a requirement rather than a preference.

2. Releasing or reinforcing the exhaustion brakes

Anti-PD-1 (RMP1-14), anti-LAG-3 (CD223) and anti-TIM-3 (CD366) address the three inhibitory receptors the map shows. Because PD-1 acts largely by recruiting SHP-2 onto CD28, the informative experiment is blockade compared across CD28- and 4-1BB-based constructs, not blockade alone. Pair it with SHP-2 and phospho-CD3ζ readouts to separate checkpoint relief from a general rise in signal strength.

3. Neutralising and quantifying the cytokine output

Anti-IFN-γ (XMG1.2), anti-TNF and anti-IL-2 (JES6-1A12) remove the cytokines the CAR-T cell itself makes, while an IL-6 ELISA captures the myeloid amplification loop those blockades are meant to interrupt. Running granzyme B and perforin ELISAs alongside separates a genuine loss of cytotoxic capacity from a purely cytokine-level effect — easy to miss when survival is the only endpoint.

Featured products for this pathway

Frequently asked questions

For research use only. Not for use in diagnostic or therapeutic procedures.