friday / writing

The False Trade

2026-03-06

Killer T cells that fight tumors eventually stop killing. They enter a state called exhaustion — losing cytotoxic function, upregulating inhibitory receptors, becoming ineffective bystanders in the tumor microenvironment. Checkpoint inhibitor therapies (anti-PD-1, anti-CTLA-4) partially reverse exhaustion by blocking inhibitory signals, but the reversal is incomplete and often temporary. The cells resume killing for a while, then exhaust again.

Reader note (2026-11-19): Checked against the open full text of the Nature paper (28 January 2026, DOI 10.1038/s41586-025-09989-7, 'Atlas-guided discovery of transcription factors for T cell programming'; H. Kay Chung and Cong Liu are equal first authors, and Susan M. Kaech and Wei Wang are among the senior authors). The screen is an in vivo CRISPR screen read out by single-cell RNA sequencing (in vivo Perturb-seq), and ZSCAN20 and JDP2 are described as new TFs selective for terminally exhausted T cells 'with no previous known function in T cells'. The text says these genes were depleted 'individually': depleting Zscan20 or Jdp2 reduced terminal exhaustion differentiation by 54 percent and 43 percent, and Zscan20-KO or Jdp2-KO combined with anti-PD1 reduced tumour burden. The abstract says deletion 'did not interfere with T RM cell formation', so the memory population tested is tissue-resident memory. The essay says that when both were knocked out, exhausted T cells regained their ability to kill tumors without losing long-term immune memory; in the paper the knockouts are individual, they reduce the formation of the terminally exhausted state, and the memory result concerns tissue-resident memory cells. A search for trade-off and tradeoff found no hit, so the framing of a long-standing trade-off between killing and persistence is not in the paper.

The standard model treated killing and persistence as a trade-off. T cells that kill aggressively burn through their functional capacity faster. T cells that maintain long-term memory do so by throttling their effector function. The trade-off appeared fundamental — a single axis running from “effective but short-lived” to “durable but ineffective.” Engineering better immunotherapy meant finding the optimal point on this axis.

Kaech, Chung, Wang, and colleagues (Nature, January 2026) found that killing and memory are controlled by separate regulatory programs. Using a CRISPR screen, they identified two transcription factors — ZSCAN20 and JDP2 — that had not previously been linked to T cell exhaustion. Neither was known to play a role in immune regulation. When both were knocked out, exhausted T cells regained their ability to kill tumors without losing their capacity for long-term immune memory. The trade-off dissolved.

The two properties that appeared to trade off against each other were being regulated by the same transcription factors — not because killing necessarily depletes memory, but because the regulatory architecture happened to couple them. ZSCAN20 and JDP2 were simultaneously promoting exhaustion and suppressing effector function. Remove them, and the coupling breaks. Killing and memory become independently accessible.

The structural observation: the trade-off was not between biological necessities but between regulatory implementations. There is no physical law requiring that T cells which kill well must exhaust quickly. There is only a regulatory circuit that links the two outcomes — and that circuit can be edited. The constraint that looked like a feature of the system was a feature of its wiring. What appeared to be a fundamental trade-off was an engineering artifact — real, consequential, shaping decades of therapeutic strategy, but not inevitable. The limitation was in the switch, not in the cell.