CAR-T cell therapy

A T cell normally needs two things to commit: its receptor must engage a peptide displayed on MHC, and a separate costimulatory receptor must confirm the encounter. A chimeric antigen receptor replaces both with one polypeptide. An antibody-derived single-chain variable fragment sits outside the membrane, joined through a hinge and transmembrane segment to an intracellular tail carrying a costimulatory domain — CD28 or 4-1BB — spliced directly onto the CD3ζ signalling chain. Antigen binding clusters the receptor, ζ chains are phosphorylated, and the cell activates as if it had received both signals at once.

The consequence is a change of recognition rules. Because the scFv binds antigen directly, the tumour cannot escape by downregulating HLA or by defects in antigen processing, which are among the commonest immune-evasion routes. The price is that the target must be on the outside of the cell. Everything intracellular — the great majority of the tumour-specific proteome — is invisible to a CAR.

Why CD19 and BCMA, and almost nothing else

No surface protein is exclusive to a cancer. The approved targets were chosen because their normal-tissue expression is survivable: CD19 is on the whole B lineage, so CD19-directed therapy causes B-cell aplasia, which is managed with immunoglobulin replacement; BCMA is on plasma cells, with the same logic. That is a tissue argument, not a tumour argument, and it does not generalise. A CAR against a solid-tumour antigen shared with lung, gut or vascular endothelium has no equivalent escape hatch.

The costimulatory choice tunes behaviour rather than specificity. CD28-containing constructs signal strongly and expand fast; 4-1BB constructs expand more slowly, favour oxidative metabolism and mitochondrial biogenesis, and persist longer. Neither choice fixes the deeper failure modes.

Toxicity is on-mechanism

Cytokine release syndrome is not an impurity effect. Activated CAR-T cells recruit host myeloid cells, and the resulting IL-6 — largely macrophage-derived — drives the fever, hypotension and capillary leak; blocking the IL-6 receptor with tocilizumab works because the pathway is the mechanism. Immune effector cell–associated neurotoxicity involves endothelial activation and blood–brain barrier disruption and is less cleanly reversed. Both scale with tumour burden and with the vigour of expansion, which is to say with efficacy.

The unresolved problems

Relapse after CD19 therapy is frequently antigen loss — including alternatively spliced CD19 transcripts missing the epitope — so the escape route is the same single-epitope dependence that gave the specificity. In solid tumours the cells must first traffic and extravasate, then function inside a microenvironment that is hypoxic, adenosine-rich and TGF-β–suppressed, against heterogeneous antigen; none of these is addressed by receptor design alone.

Manufacturing imposes a further constraint that is biological rather than industrial: the starting material is the patient’s own lymphocytes after cytotoxic therapy, and their differentiation state predicts persistence. Allogeneic “off-the-shelf” products remove that variability but create two new requirements at once — the donor TCR must be disrupted to prevent graft-versus-host disease, and the resulting cells must then evade host rejection, where deleting HLA class I invites NK-cell killing. That trade is still open.