Therapeutics & platforms
TCR-engineered T-cell therapy
Peptide–MHC recognition and its extreme sensitivity, why HLA restriction fragments the patient population, and the cross-reactivity that killed patients when affinity was engineered upward.
A chimeric antigen receptor can only see the cell surface. A T-cell receptor sees what the cell is doing inside. Every nucleated cell continuously degrades a sample of its own proteome in the proteasome, transports the fragments into the endoplasmic reticulum and displays eight- to eleven-residue peptides on MHC class I. A TCR reads that display. Cancer-testis antigens such as MAGE-A4 and NY-ESO-1, mutated intracellular drivers, viral oncoproteins — none of which reach the membrane — become addressable. That is roughly an order of magnitude more of the proteome than surface-restricted approaches can reach, and it is the entire reason to engineer TCRs rather than CARs.
Sensitivity, and why it has to be extreme
The trade is antigen density. A CAR target may be present at tens of thousands of copies per cell; a specific peptide–MHC complex is typically present at tens to a few hundred. Natural T-cell signalling copes with this through serial triggering — one complex engaging many receptors in succession — and through kinetic proofreading, where the signalling cascade only completes if binding persists long enough, which discriminates against brief non-specific contacts. A handful of complexes can be enough to activate a cell. The machinery is remarkable, and it is also the source of the field’s danger: a receptor this sensitive will respond to a peptide that merely resembles its target.
HLA restriction
The peptide is only visible in the groove of a particular allele. A TCR raised against MAGE-A4 on HLA-A02:01 is useless in a patient with a different haplotype, so each product serves a fraction of the population — roughly half of Europeans for A02:01, and considerably less in most other ancestries. This is a structural inequity of the modality, not an oversight. The tumour can also simply stop presenting: loss of β2-microglobulin or downregulation of HLA class I removes the entire display, an escape route that CARs are immune to and TCRs are not.
The affinity trap
Natural TCRs bind peptide–MHC weakly, with dissociation constants in the micromolar range. That is not a design flaw. Thymic negative selection deletes clones that bind self-peptides strongly, so low affinity is the residue of a safety filter. Engineering affinity upward removes the filter, and the consequences have been demonstrated in patients. An affinity-enhanced MAGE-A3 TCR caused fatal cardiac toxicity in a 2013 trial by cross-reacting with an unrelated peptide from titin presented on beating cardiomyocytes — a target invisible to every preclinical screen used at the time. A different MAGE-A3 receptor caused fatal neurotoxicity by recognising MAGE-A12 in brain. Systematic cross-reactivity screening — alanine and X-scan mapping of the recognition motif, then searching the human proteome for matches — became mandatory because of those deaths, and it remains the field’s most consequential safety step.
Two engineering problems follow the receptor itself. An introduced α/β pair can mispair with the endogenous chains, lowering expression and creating receptors of unknown specificity; murinised constant domains, added disulfide bonds or knockout of the endogenous TCR address it. And a soluble alternative exists: fusing an affinity-matured TCR to an anti-CD3 fragment, as in tebentafusp for HLA-A*02:01–positive uveal melanoma, redirects any passing T cell without engineering the patient’s cells at all.