Evolution of NLR repertoires

Nine papers here work at evolutionary scale, from a single allelic series to 56,280 sequences. What they share is a finding that NLR families do not expand uniformly, and that where they do not expand is as informative as where they do.

Expansion is lineage-specific and uneven

The NRC network is the case in point. Across 40 Solanales and 29 Asterales genomes, NRC-type receptors make up 50.3% of the Solanales NLR repertoire and 6.6% of the Asterales one, despite similar total NLR counts per genome (338.6 and 300). Asterales expanded other classes instead, TIR-NLRs in particular. In lettuce, NRCs are 17 sequences against 359 total, while TIR-NLRs take over half (pai-2026-lettuce-nrc).

Legumes show the same unevenness on a different axis. NLR counts range from 88 in Lupinus albus to 970 in Arachis hypogaea, correlating only moderately with proteome size, and three species carry up to three times as many as their proteome predicts (marques-2026-legume-tissue).

That smallness is what makes lettuce useful. Two helpers and nine functional sensors is a network you can hold in your head, and the corpus uses it twice — once to map evolutionary trajectories, once as a bioengineering substrate (pai-2026-lettuce-nrc, toghani-2026-sensor-helper-interface).

Subfunctionalisation leaves a signature

In lettuce the two helpers support different sensor groups, and the split shows up in selection analysis as well as in function. No positively selected sites were detected in either helper or in the group that signals through both. Sensors depending on the conserved NRC0 helper carry some; sensors depending on the newer Ast-LsNRC1 carry more, concentrated on the concave face of the LRR (pai-2026-lettuce-nrc).

Helper dependency also tracks genomic position, but only sometimes. Ast-LsatNRC1 sits in a physical cluster with the sensors it serves; LsatNRC0 is not clustered with anything. Across other Asterales the pattern is species-specific, present in chicory and absent in sunflower (pai-2026-lettuce-nrc).

Conservation over very long intervals

Two results push the timescale. The NB-centred sensor-helper binding mode holds across five asterid orders and a non-asterid one, so what is conserved is a binding geometry rather than a sequence (toghani-2026-sensor-helper-interface). And chloroplast targeting holds across five CCR-NLRs spanning roughly 360 million years, back to a fern, with a phylogeny of 56,280 CCR-NLR sequences behind it (ibrahim-2026-nrg1-chloroplast).

Structural properties travel too. The octameric architecture appears in receptors from wheat and Arabidopsis, so it survived the monocot-dicot split (guo-2026-ccg10-resistosome), and the undecamer is a property of a clade rather than of one protein (toghani-2026-sni-undecamer).

Diversification at close range

The MLA allelic series does the opposite: more than thirty alleles above 90% identity, of which one recognises a given effector, and three residues make the difference (gomezdelacruz-2026-mla3-mimicry). Effector families diversify to match — the NRC-binding surfaces of an L-shaped effector family are the least conserved surface residues in it (madhuprakash-2026-avrcap1b-tol9a), and a single ortholog swap of one domain transfers suppression activity (seager-2026-avrcap1b-nrc3).

Where expression sits in this

Tissue is part of the repertoire. Legume NLRs are 2.66-fold more tissue-specific than genes generally and lean towards roots where the transcriptome leans towards leaves. Classes differ: TNPs are 87% root-specific, CCG10-NLRs favour roots, CCR-NLRs favour leaves (marques-2026-legume-tissue).

A repertoire is not just how many receptors a genome has. It is which ones are switched on, and where.