Two linked pairs of Arabidopsis TNL resistance genes independently confer recognition of bacterial effector AvrRps4

Plant immunity requires recognition of pathogen effectors by intracellular NB-LRR immune receptors encoded by Resistance (R) genes. Most R proteins recognize a specific effector, but some function in pairs that recognize multiple effectors. Arabidopsis thaliana TIR-NB-LRR proteins RRS1-R and RPS4 together recognize two bacterial effectors, AvrRps4 from Pseudomonas syringae and PopP2 from Ralstonia solanacearum. However, AvrRps4, but not PopP2, is recognized in rrs1/rps4 mutants. We reveal an R gene pair that resembles and is linked to RRS1/RPS4, designated as RRS1B/RPS4B, which confers recognition of AvrRps4 but not PopP2. Like RRS1/RPS4, RRS1B/RPS4B proteins associate and activate defence genes upon AvrRps4 recognition. Inappropriate combinations (RRS1/RPS4B or RRS1B/RPS4) are non-functional and this specificity is not TIR domain dependent. Distinct putative orthologues of both pairs are maintained in the genomes of Arabidopsis thaliana relatives and are likely derived from a common ancestor pair. Our results provide novel insights into paired R gene function and evolution. Plant immunity requires recognition of pathogen effector proteins by specific intracellular immune receptors. Here, Saucet et al. identify an additional pair of Arabidopsis receptors that act together to trigger defence responses upon recognition of the AvrRps4 effector from the bacterial pathogen Pseudomonas syringae.

P lant immunity is activated upon direct or indirect perception of pathogen molecules. Cell surface receptors perceive relatively invariant pathogen molecules such as flagellin or chitin, and intracellular receptors perceive the presence or action of pathogen effectors 1 . The intracellular immune receptors are called resistance (R) proteins and encoded by R genes. Effector recognition by R proteins leads to effectortriggered immunity, which often culminates in programmed cell death, also known as the hypersensitive response (HR) 2 . However, disease resistance can be achieved without HR 3 . Most plant R genes encode nucleotide-binding, leucine-rich repeat (NB-LRR) proteins that structurally and functionally resemble mammalian nucleotide-oligomerization domain-like (NOD-like) receptors 4 . NB-LRR proteins are modular and presumed to undergo intra-and inter-molecular reconfiguration upon effector recognition, activating defence 5 . However, mechanisms of recognition, activation and downstream signalling are poorly understood. R proteins of the TIR-NB-LRR (TNL) subclass carry a Toll/Interleukin-1 receptor/Resistance (TIR) protein domain at their N-termini, with homology to cytoplasmic domains of mammalian and Drosophila transmembrane immune receptors 6,7 . Examples include the tobacco N gene for virus resistance 8 , flax L and M genes for flax rust resistance 9 , several Arabidopsis RPP genes for downy mildew resistance 10 , Arabidopsis RPS4 (originally defined as conferring recognition of Pseudomonas syringae pathovar pisi effector AvrRps4) 11 and Arabidopsis RRS1-R that confers resistance to Ralstonia solanacearum via recognition of effector PopP2 (refs 12,13). The RRS1-R allele in accession Nd-1 confers PopP2 recognition, whereas the Col-0 RRS1-S allele does not.
RRS1-R (from Ws-2) and RRS1-S (from Col-0) are required for AvrRps4 recognition by RPS4, and reciprocally RPS4 is required for PopP2 recognition by RRS1-R in Ws-2 (refs 14,15). RRS1 and RPS4 are adjacent, divergently transcribed and function cooperatively for AvrRps4 and PopP2 recognition, and for resistance to the fungus Colletotrichum higginsianum 14,15 . NB-LRRs can require other NB-LRRs for function 16 . For example, RPP2A and RPP2B, two adjacent TNLs in Col-0, are both required for resistance to Hpa race Cala2 (ref. 17). Rice RGA4 and RGA5 form a NB-LRR pair for recognition of Avr-Pia and Avr1-CO39, two effectors from the fungal pathogen Magnaporthe oryzae 18 . Tobacco N requires the 'N-required gene' NRG1 (ref. 19), and multiple genes of the ADR1 class are required for function of RPS2, RPP4 and RPP2 (ref. 20).
To dissect AvrRps4-triggered immunity, we positionally cloned the RRIR locus. We found another pair of R genes, At5g45050 and At5g45060, highly similar to RRS1/RPS4, which confers RRIR, and named the locus as RRS1B/RPS4B. We show using Agrobacterium-mediated transient assays in Nicotiana tabacum that co-expression of RRS1B and RPS4B confers recognition of AvrRps4 but not PopP2. The RRS1B and RPS4B proteins associate with each other, as previously reported for RRS1 and RPS4 (ref. 24). Non-authentic R protein associations can also occur between RRS1B and RPS4, and RRS1 and RPS4B, but do not enable effector recognition, suggesting that these R proteins must pair with their appropriate respective partner for function. Although RRS1B/RPS4B does not recognize PopP2, PopP2 still associates in planta with RRS1B. TIR domain exchanges between RRS1 and RRS1B, or between RPS4 and RPS4B, do not compromise function. A chimera in which the C-terminal 3 exons of RRS1-R are exchanged for those of RRS1B confers effector-independent HR in combination with RPS4, but not with RPS4B, suggesting integrity of this region is important to prevent defence signalling before effector perception. We finally compare RRS1/RPS4-like gene pairs in several Brassicaceae. These data contribute to understanding TNL-mediated immunity in plants and provide new insights into paired plant immune receptor function and evolution.
Co-expression of avrRps4 or popP2, together with RRS1 and RPS4, by agro-infiltration in N. tabacum leaves results in HR 24 . Using a similar method, we were able to show AvrRps4 recognition by RRS1B/RPS4B in N. tabacum. Co-expression of avrRps4, but not GFP, together with RRS1B Ws-2 and RPS4B Ws-2 triggered HR at 5 d.p.i. (Fig. 3b). We then tested AvrRps4 recognition by RRS1B RLD , RPS4B RLD and RPS4B (D3829)RLD using this assay. Consistent with phenotypes observed in Arabidopsis RLD and Ws-2 rps4-21/rps4b-1 complemented lines, coexpression of RPS4B RLD with either RRS1B RLD or RRS1B Ws-2 in N. tabacum did not show HR to AvrRps4 ( Fig. 3b and Supplementary Fig. 6). On the other hand, an AvrRps4triggered HR was observed when RRS1B RLD was co-expressed with either RPS4B (D3829)RLD or RPS4B Ws-2 ( Fig. 3b and Supplementary Fig. 6). The stability of each fusion protein used in the transient assay was verified by western blot analysis ( Supplementary Fig. 7). These data suggest that RRS1B RLD is functional and that the insertion causing the premature stop codon in RPS4B RLD is responsible for non-functionality and loss of RRIR in the RLD accession.
AvrRps4-induced defence genes require either RPS4 or RPS4B. Given two paired R genes that provide resistance to one effector, we investigated the quantitative contributions of RRS1B/RPS4B to AvrRps4-triggered immunity compared with RRS1/RPS4 by measuring defence gene induction using quantitative RT-PCR. Based on previous studies, we selected defence marker genes (SARD1, SID2, PAD4 and EDS5) that are specifically regulated by Figure 3 | RPS4B RLD carries a mutation responsible for loss of AvrRps4 recognition. (a) HR assay using leaf infiltration with Pf Pf0-1 secreting AvrRps4 KRVY AAAA or AvrRps4 in Ws-2 rps4-21, RLD, Ws-2 rps4-21/ rps4b-1 and stable transgenic lines of Ws-2 rps4-21/rps4b-1-expressing RPS4B Ws-2 , RPS4B RLD or RPS4B (D3829)RLD (under native promoter and terminator). RPS4B Ws-2 and RPS4B RLD were cloned from accession Ws-2 and RLD, respectively. RPS4B (D3829)RLD corresponds to RPS4B RLD with the WT inserted nucleotide at position 3,829 bp removed, restoring similar coding sequence to Ws-2. Pictures were taken 24 h.p.i. Magenta arrows indicate leaves showing HR. (b) HR assay in N. tabacum leaves using transient A. tumefaciens transformation. Each leaf section was co-infiltrated to express a combination of different R genes (shown on the left) together with GFP or avrRps4-GFP. RRS1, RPS4, RRS1B and RPS4B were cloned from Ws-2 or RLD gDNA. Cell death pictures were taken 5 d.p.i. These experiments were repeated at least three times with similar results.
AvrRps4 and PopP2 in a RRS1/RPS4-dependent manner at early stages of immunity 27 . Genotypes Ws-2, Ws-2 rps4-21, Ws-2 rps4b-1 and rps4-21/rps4b-1 were infiltrated with either H 2 O, Pf Pf0-1-carrying AvrRps4 or AvrRps4 KRVY AAAA mutant. Six hours post infiltration, the fold change of all selected defence marker genes compared with H 2 O treatment was consistently more induced in Ws-2, rps4-21 or rps4b-1 after infiltration with Pf Pf0-1 (AvrRps4) than with Pf Pf0-1 (AvrRps4 KRVY AAAA ; Fig. 4 and Supplementary Table 4). Assuming that each pair functions independently, this indicates that RRS1B/RPS4B (in rps4-21) and RRS1/RPS4 (in rps4b-1) were able to activate a similar set of defence genes upon AvrRps4 recognition, and are therefore likely to share downstream signalling mechanisms. In the single knockout mutants tested, there was no consistent pattern of quantitative differences between RRS1/RPS4-and RRS1B/RPS4Bdependent defence genes fold induction triggered by AvrRps4 ( Fig. 4 and Supplementary Fig. 8). We also could not observe greater gene induction in Ws-2 compared with the either single mutants. We infer that RRS1/RPS4 and RRS1B/RPS4B independently activate defence genes to a level adequate for resistance in response to AvrRps4. In the rps4-21/rps4b-1 double mutant fold induction of defence genes triggered by Pf Pf0-1 (AvrRps4) is not different to Pf Pf0-1 (AvrRps4 KRVY AAAA ), but is significantly lower than in AvrRps4-treated Ws-2 or single mutants (Fig. 4). This means AvrRps4-triggered defence genes induction is fully dependent on functional RRS1B/RPS4B and/or RRS1/RPS4, which is consistent with the loss of resistance to Pst DC3000 (AvrRps4) observed in the double mutants ( Fig. 2d and Supplementary Fig. 1).
RRS1B and RPS4B associate and function together. We have established that there are two R gene pairs with B60% identity that function independently to recognize AvrRps4 ( Supplementary Figs 9 and 10). Therefore, we wanted to understand how these R proteins work with their specific pair partners. Genetic evidence of pair specificity was shown in Fig. 2c where the RRS1B þ RPS4 combination in Ws-2 rrs1-1/rps4b-1 cannot give HR to Pf Pf0-1 (AvrRps4). Similarly, the RRS1 þ RPS4B combination in Col-0 rps4-2/rrs1b-1 does not give resistance to Pst DC3000 (AvrRps4; Fig. 2d). In an N. tabacum transient assay, co-expression of RRS1-R þ RPS4B or RRS1B þ RPS4 combinations did not give HR to either AvrRps4 or PopP2 (Fig. 5a,b and Supplementary Fig. 11). In contrast, popP2 only triggered HR when co-expressed with RRS1-R þ RPS4, and avrRps4-triggered HR with both RRS1-R þ RPS4 and RRS1B þ RPS4B (Fig. 5a). This confirms that each R protein requires its cognate pair partner for function.
RRS1 and RPS4 physically associate to form a functional recognition complex in planta 24 . Therefore, we tested by coimmunoprecipitation (co-IP) whether RRS1B and RPS4B also associate in planta, and if this association is specific to pair partners. Similar to RRS1-GFP and RPS4-FLAG, RRS1B-GFP co-IP with RPS4B-FLAG when transiently co-expressed in Nicotiana benthamiana leaves (Fig. 5b lanes 2 and 6, and Supplementary  Fig. 12). Interestingly, we also found RRS1B-GFP co-IP with RPS4-FLAG, and RRS1-GFP co-IP with RPS4B-FLAG (Fig. 5b lanes 3 and 5 and Supplementary Fig. 12). However, the intensity of the co-IP signals was notably higher with the appropriate pair partners than with the inappropriate partners. We infer that the complexes formed of inappropriate pairing (for example, RRS1 þ RPS4B or RRS1B þ RPS4) are less stable, which could explain their non-functionality. Collectively, similar to RRS1/ RPS4, RRS1B/RPS4B associate in planta to form a complex before any effector perception, and R protein pair complexes are only functional when formed of the appropriate partners.
Many lines of evidence suggest TIR-TIR interactions are important for TNL function 24,37 . The TIR domains of RRS1 and RPS4 interact in a yeast two hybrid assay and associate in planta after co-IP 24 . TIR domains of these proteins are essential for effector recognition and defence activation 24 . In the preactivation state, the RRS1/RPS4 heterodimer is proposed to be inactive. This correlates with the RRS1 TIR suppression of RPS4 TIR -triggered cell death when the TIR domains of these two paired R proteins associate (Fig. 6a) 24 . Interestingly, we found that RRS1B TIR can also suppress RPS4 TIR -triggered cell death (Fig. 6a), suggesting that TIR domains from different pairs can interact. Therefore, we investigated whether RRS1B TIR and RPS4B TIR can associate with each other in planta, and whether they can associate with RPS4 TIR and RRS1 TIR , respectively, by co-IP. After agro-infiltration in N. benthamiana, RRS1B TIR -FLAG co-IP with both RPS4B TIR -GFP and RPS4 TIR -GFP; and RPS4B TIR -FLAG co-IP with both RRS1B TIR -GFP and RRS1 TIR -GFP ( Fig. 6b and Supplementary Fig. 13). This suggests that, similar to full-length proteins, RRS1B TIR and RPS4B TIR can heterodimerize and also associate with RPS4 TIR and RRS1 TIR , respectively, in planta.
TIR swaps between R protein pairs retain function. We next assessed if, despite association of TIR domains between nonpaired R proteins, the TIR domains contribute to the specificity of R protein function with each respective pair partner. To answer this question, we constructed chimeras in which RRS1 Ws-2 and RRS1B Ws-2 TIR domains were exchanged, and similarly with RPS4 Ws-2 and RPS4B Ws-2 TIR domains. We designated the four domains of RRS1 and RPS4 (TIR, NB, LRR and C-Terminal Domain) as 'AAAA' and of RRS1B and RPS4B as 'BBBB', defining TIR domain swaps as RRS1 BAAA and RRS1 ABBB fulllength chimeric proteins (Fig. 6c). These chimeras were tested with WT R proteins for AvrRps4 and PopP2 recognition in N. tabacum transient assay. Similar to RRS1 AAAA and RPS4 AAAA , RRS1 BAAA þ RPS4 AAAA and RRS1 AAAA þ RPS4 BAAA combinations recognized both AvrRps4 and PopP2 (Fig. 6d). On the other hand, similar to RRS1 BBBB and RPS4 BBBB , RRS1 ABBB þ RPS4 BBBB and RRS1 BBBB þ RPS4 ABBB recognized AvrRps4 only (Fig. 6d). Accumulation of chimeric proteins was confirmed by immunoblot ( Supplementary Fig. 14a,b). These results show that exchange of TIR domains from paralogous R genes does not compromise AvrRps4 or PopP2 recognition. In addition, the TIR domains do not drive the pair partner specificity for function, therefore other domain-domain interactions must account for the pairing specificity.
We also characterized additional domain swaps between RRS1 and RRS1B, with the breakpoint between the end of exon 4 and the beginning of exon 5. Exons 5, 6 and 7 encode the WRKY domain of these proteins and B260 amino acids between the LRR domain and the WRKY domain. These swaps were designated as 'AAAB' and 'BBBA', and tested for recognition of AvrRps4 and PopP2 in the presence of either RPS4 or RPS4B. Accumulation of chimeric proteins was confirmed by immunoblot ( Supplementary Fig. 14c, d). Neither of the RRS1 AAAB and RRS1 BBBA chimeras conferred recognition of AvrRps4 or PopP2 (Supplementary Fig. 15). However, RRS1 AAAB in combination with RPS4 showed constitutive activity and triggered cell death in the absence of effector, but not in combination with RPS4B ( Supplementary Fig. 15). This suggests that integrity and appropriate interactions in the C-terminal regions of these proteins might be required to prevent effector-independent defence activation.
RRS1B associates with AvrRps4 and PopP2 in planta. Recently, it was shown that the RRS1-R/RPS4 R protein complex associates with AvrRps4 and PopP2 in planta 24 . RRS1 and PopP2 were also shown to interact in a yeast split-ubiquitin assay 13 . However, this interaction is not sufficient for R protein activation as PopP2 also interacts with RRS1-S Col-5 but does not trigger an immune response in Col-5 (ref. 13). We examined whether RRS1B/RPS4B can also associate with AvrRps4 and also PopP2 in planta even if RRS1B/RPS4B cannot recognize PopP2. We coexpressed RRS1 Ws-2 -FLAG þ RPS4 Ws-2 -FLAG and RRS1B Ws-2 -FLAG þ RPS4B Ws-2 -FLAG, with either GFP, avrRps4-GFP or popP2-GFP in N. benthamiana. The GFP protein alone did not co-IP with the RRS1/RPS4 and RRS1B/RPS4B complexes ( Supplementary Figs 16 and 17). Similar to RRS1/RPS4 complex, RRS1B/RPS4B complex co-IP with both AvrRps4-GFP and PopP2-GFP ( Supplementary Figs 16 and 17). Interestingly, the signal intensity for RRS1/RPS4 was much stronger compared with RRS1B/RPS4B, indicating the RRS1/RPS4 complex has stronger affinity for both effectors than the RRS1B/RPS4B complex. Despite this observation, this result suggests that RRS1B/RPS4B associates in planta with AvrRps4 but also with PopP2. However, we cannot exclude that AvrRps4 and PopP2 break the pre-formed R protein heterodimer to associate with each of the partner separately. Considering the high intensity of RRS1 protein bands compared with RPS4 after AvrRps4 and PopP2 IP, we then tested whether RRS1 and RRS1B alone could associate with these effectors. Consistent with Williams et al. 24 , RRS1 co-IP with both AvrRps4 and PopP2 (Fig. 7 lanes 3 and 5, and Supplementary Fig. 18) 24 . Interestingly, RRS1B also co-IP with both AvrRps4 and PopP2 (Fig. 7 lanes 4 and 6, and Supplementary Fig. 18). Like in complex with their partner, RRS1 showed a stronger association with AvrRps4 and PopP2 than RRS1B. Altogether, this suggests that, similar to RRS1-S/RPS4 (ref. 13), protein-protein association of PopP2 with RRS1B/ RPS4B can be detected but this is not sufficient for defence activation.

Pair duplication and gain of WRKY domain in the Brassicaceae.
To shed some light on the evolution of these tandem paired R genes, we examined the genomes of several species related to   Each leaf section was co-infiltrated to express a different combination of WT and chimera proteins with TIR domains exchanged between 'A' pair and 'B' pair proteins (shown on the left) together with GFP, avrRps4-GFP or popP2-GFP. RRS1 BAAA and RRS1 ABBB were generated by swapping sequences coding for TIR domains between RRS1 and RRS1B. RPS4 BAAA and RPS4 ABBB were generated similarly. Cell death pictures were taken 5 d.p.i. These experiments were repeated at least three times with similar results. NATURE COMMUNICATIONS | DOI: 10.1038/ncomms7338 ARTICLE A. thaliana. Using the nucleotide sequences of each gene from each pair in megablast searches 38 , we found that in the genomes of the sister species Arabidopsis lyrata 39 , and the closely related species Capsella rubella 40 , there are distinct putatively orthologous pairs matching both the 'A' pair (RRS1/RPS4) and 'B' pair (RRS1B/RPS4B). In A. lyrata, we also found an apparent duplication of the 'B' pair, resulting in a paralogous 'C' pair with a high level of identity to the 'B' pair (Fig. 8b,c). Genome to genome comparison revealed that the regions of the A. thaliana, C. rubella and A. lyrata genomes harbouring the linked pairs were syntenic, with broad conservation of DNA sequence similarity and directionality ( Supplementary Fig. 19). Although highly syntenic, the region in A. lyrata is expanded in comparison to A. thaliana and C. rubella (Supplementary Fig. 19). Additionally in A. lyrata, we found a divergently transcribed pair of TNLencoding genes (gene names: 9301337 and 9301336) with similar identity to both the 'A' and 'B' pairs, which we named AlRPS4like and AlRRS1-like, respectively (Fig. 8a-c). The RRS1-like gene of this pair, however, lacks the WRKY protein-encoding sequence, which is fused to RRS1 and RRS1B. Inspection of the DNA sequence directly following this RRS1-like gene confirmed that this was not due to the presence of an early stop codon. We also examined the genome of the more distant Arabidopsis relative Brassica rapa 41 . In the B. rapa genome, we found no WRKY-fused NB-LRR-encoding genes. A single B. rapa gene pair (gene names: Bra027598 and Bra027599) had the highest identity to the A. thaliana 'A' and 'B' pairs and is located in a region syntenic to the 'A' and 'B' region on A. thaliana chromosome 5 (Fig. 8a-c and Supplementary Fig. 20). This pair lacks a WRKY fusion, and shows the highest nucleotide and protein level identity with AlRPS4-like and AlRRS1-like. We therefore speculate that these are the ancestors of the 'A', 'B' and 'C' pairs lacking the WRKY fusion. The B. rapa RRS1-like gene (Bra027598) is fused to another, distantly related, TIR-NB-LRRencoding gene at its 3 0 end, although this may be a genome assembly or gene prediction artefact. Only the RRS1-like half was used for the analyses presented. Although their amino-acid sequences have diverged, we found that the DNA sequences of each species' RRS1 or RRS1B/C WRKY domain had the highest nucleotide identity and longest alignment to AtWRKY35 (selected alignments in Supplementary Fig. 21). These results suggest that a duplication of the ancestral pair fused with the sequence of a WRKY35-like gene in the common ancestor of A. thaliana, C. rubella and A. lyrata. Later in this common ancestor, a duplication event produced the 'A' and 'B' pairs. Specifically in the A. lyrata lineage, a further duplication of the 'B' pair occurred, resulting in the 'C' pair.

Discussion
This study revealed that RRS1B/RPS4B, an R gene pair paralogous and closely linked to RRS1/RPS4, also recognizes AvrRps4, but not PopP2. In this system, multiple paired R proteins cooperate in a specific manner for the recognition of several effectors from unrelated plant pathogens (Supplementary Fig. 22).
The evolutionary advantage for the plant to carry two R gene pairs recognizing the same effector, AvrRps4, is unclear. Conceivably, RRS1B/RPS4B, which recognizes AvrRps4, duplicated to create RRS1-R/RPS4, recognizing AvrRps4 and PopP2. Although our analyses of the genomes of various relatives of A. thaliana could not reveal which of the two pairs is ancestral, we did uncover the probable pre-WRKY fusion ancestor of both pairs, present in B. rapa and A. lyrata. We also found that distinct orthologous pairs are conserved in the A. lyrata and C. rubella genomes, suggesting that there is a selective advantage to maintaining both pairs. Indeed, in A. lyrata the 'B' pair has further duplicated to produce a closely related 'C' pair. We speculate that the duplication of such R gene pairs might reduce the effect of purifying selection and increase the potential for one of the pairs to evolve new or expanded effector recognition capacity 42 . In addition, developing two or more similar recognition systems might enable the plant to maximize protection of an important cellular complex generally targeted by pathogen effectors. However, accession RLD lost both RRS1/ RPS4 and RRS1B/RPS4B functions. Conceivably, RLD evolved in an environment not exposed to pathogens recognized by RRS1/ RPS4 and/or RRS1B/RPS4B, and mutating these two R gene pairs could enhance fitness if there is a cost to resistance. Another hypothesis is that mutating these R gene pairs would have avoided a hybrid incompatibility 43 . Indeed, mutations in RRS1 can cause auto-activity of the RRS1/RPS4 complex leading to growth arrest 44 and this could also be the case for RRS1B/RPS4B.
The appropriate partners of each of RRS1/RPS4 and RRS1B/ RPS4B are required for effector-triggered immunity. These proteins form heteromeric complexes in a resting state before effector perception (Supplementary Fig. 22a). In addition, heteromeric complexes can be formed between partners from different pairs before effector recognition. However, such complexes are not functional for AvrRps4 and PopP2 recognition and/or downstream signalling activation. Altogether, this indicates that despite the similarity in motif prediction, TNLs evolved particular inter-molecular specificity for function. The TIR domains are not responsible for this phenomenon, and this specificity must be conferred by domains elsewhere in the R protein. However, we cannot exclude that association between WT proteins and TIR domains from different pairs is the result of overexpression in N. benthamiana, which might not occur under native expression. Integrity of NB-LRR protein is an important factor for function. Swapping parts of NB or LRR domains between the Rx and Gpa2 protein disrupt specific intra-molecular interactions and generate auto-active phenotype 45,46 . Similarly, we attempted to define the domains of RRS1-R responsible for PopP2 recognition by domain swaps with RRS1B. Although we did not identify this domain in this study, we were able to show that a combination of the RRS1 AAAB chimera with RPS4, but not with RPS4B, resulted in effector-independent HR. We infer that appropriate interactions between multiple domains in the complex are required to create a functional but not auto-active complex. Additional domain swap experiments between the two pairs will provide an opportunity for further mechanistic investigation.
Several TNLs have been shown to recognize an effector directly and this direct interaction is presumed to trigger R protein activation 13,[47][48][49] . The exact mechanism by which RRS1/RPS4 and RRS1B/RPS4B recognize AvrRps4 and PopP2 has not yet been reported. Although we only showed protein association by co-IP, we can speculate that the potential direct interaction of AvrRps4 with RRS1/RPS4 and RRS1B/RPS4B triggers R protein intra-molecular reconfigurations (and perhaps changes in oligomerization) leading to defence activation ( Supplementary  Fig. 22b). Such mechanisms could also be suggested for PopP2 recognition but, even if RRS1B/RPS4B associate with PopP2, this association does not trigger immunity. Therefore, an additional molecular event, occurring in the RRS1/RPS4/PopP2 but not in RRS1B/RPS4B/PopP2 complex, might be required ( Supplementary Fig. 22c). Both RRS1 and RRS1B carry WRKY domains in their C-termini. WRKY domain-containing proteins play crucial roles in regulating plant defence 50 . If AvrRps4 and PopP2 virulence activities are to target WRKY domaincontaining proteins, we could hypothesize that RRS1 and RRS1B, in association with RPS4 and RPS4B, respectively, could behave as decoys for effectors 44,51 . Intriguingly, at the nucleotide level, the WRKY domain of all the RRS1 orthologues and paralogues shows the highest identity to WRKY35, suggesting a common origin. Despite this, the amino-acid sequence of the WRKY domain of RRS1B (WRKY16) is phylogenetically in group IIe of WRKY transcription factors, whereas that of RRS1 (WRKY52) lies in Group III (ref. 52). Conceivably, following duplication, differential adaptive changes occurred in the WRKY domains. If these domains are involved in effector recognition, the two R gene pairs may confer recognition of different subsets of WRKY-targeting effectors or effector alleles. Therefore, an important challenge is now to determine the precise role of each RRS1/RPS4 and RRS1B/RPS4B domain, and particularly to identify which of these domain(s) interact with AvrRps4 and PopP2. Also, identification of R protein complex components before and after effector recognition will help us to understand how TNL proteins function to activate plant immunity ( Supplementary Fig. 22d,e).
In summary, our data reveal an answer to a long-standing puzzle of how an Arabidopsis Col-0 or Ws-2 rps4 mutant can still activate defence in response to AvrRps4. We show that the RRS1B/RPS4B resembles structurally and functionally the RRS1/ RPS4 pair. The RRS1/RPS4 and RRS1B/RPS4B gene pairs will be   ChromoTek). Antibody-coupled agarose beads were collected and washed three times in GTEN buffer, resuspended in 3 Â SDS-loading buffer and denatured at 96°C for 10 min. Proteins were separated by SDS-PAGE and analysed by immunoblotting using anti-FLAG M2-Peroxidase (HRP) (A8592 Sigma) or anti-GFP(B-2) HRP (sc-9996 Santa Cruz Biotechnology). We used 1/10,000 dilution of 200 mg ml À 1 anti-FLAG and anti-GFP antibodies in 5% non-fat milk.
RRIR mapping. To positionally clone the RRIR locus, a segregating population was generated by crossing Ws-2 rps4-21 with RLD. The Ws-2 rps4-21 Â RLD F1 progeny showed HR to Pf Pf0-1 (AvrRps4), although variation in HR intensity was observed compared with the Ws-2 rps4-21 parent (Fig. 1). This observation suggested a semi-dominance of the RRIR-conferring locus from Ws-2; this weak phenotype was designated 'hr' in contrast to the 'HR' of Ws-2 rps4-21. The F2 progeny derived from F1 was then phenotyped (for the presence or absence of HR to Pf Pf0-1 (AvrRps4)) and genotyped to map the RRIR locus ( Fig. 2a and Supplementary Table 1). After infiltration of 48 F2 plants with Pf Pf0-1 (AvrRps4), an approximate 1:2:1 ratio (8 (HR):28 (hr):12 (no HR); w 2 ¼ 2, P ¼ 0.572) was observed. This segregation suggested that the RRIR was associated with a single semi-dominant locus. To confirm the ratio observed in F2, we tested the HR phenotype segregation in 8 F3 plants derived from each of the 48 individual F2. By classifying the F3 families (F2:3) into those that showed HR, and those nonsegregating for the absence of HR, a 3:1 ratio was observed (37 F2:3 with (HR): 11 F2:3 (no HR); w 2 ¼ 0.111, P ¼ 0.945). This confirmed the presence of a single locus for the RRIR. To map the RRIR locus, we obtained and tested a total of 48 F2 individuals that did not segregate for the absence of Pf Pf0-1 (AvrRps4)-triggered HR in F3. As AvrRps4 recognition segregates 3:1 in a Ws-2 Â RLD cross 11 , RRIR is likely to be conferred by a locus that is closely linked to RRS1/RPS4. Therefore, we focused the mapping on the lower arm of the chromosome 5 (Fig. 2a), around RPS4, where all tested loci were homozygous for an RLD genotype in the 48 F2 individuals. Consistent with our hypothesis, markers SS007 (designed on RPS4) and SS017 mapped 2.1 cM (2 recombinants out of 48 F2 plants tested) from the RRIR locus confirming that RRIR and RPS4 loci were linked but distinct. The markers DFR.1 and N5-20408832 mapped, respectively, 5.7 cM (5 recombinants out of 44 F2 plants tested) and 14.1 cM (13 recombinants out of 46 F2 plants tested) from RRIR. The two recombinants (that is, two chromosomes heterozygous for Ws-2 genotype at the marker position) identified with SS017 were also recombinants at the N5-20408832 marker. Interestingly, no recombinants were similarly identified by DFR.1 and SS017 (Fig. 2a). Thus, we concluded that the RRIR locus resided between DFR.1 and SS017. In this region, only four TNLencoding genes are predicted in Col-0 (TAO1, LAZ5, At5g45050 and At5g45060; Fig. 2a) 55 . We focused on TNL-encoding genes for RRIR as AvrRps4-dependent recognition in Ws-2 is EDS1 dependent 25 . The SS117 marker was designed onto At5g45060, a TNL divergently transcribed from At5g45050. SS117 co-segregated with the RRIR locus (0 recombinants out of 48 F2 plants tested; Fig. 2a). TAO1 contributes to disease resistance against Pst DC3000 carrying AvrB in Arabidopsis 56 and LAZ5 is required for the lesion mimic mutant acd11 (ref. 57).
Analysis of RRS1/RPS4-like pairs in Brassicaceae. We used nucleotidenucleotide BLAST 2.2.29 þ 58 , specifically the megablast algorithm 38 , to search ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms7338 the predicted coding sequence databases from A. lyrata 39 , C. rubella 40 and B. rapa 41 with the A. thaliana Col-0-coding sequences of RRS1, RPS4, RRS1B and RPS4B. Following the identification of putatively orthologous genes, MUSCLE 59 and CLUSTAL W 60 nucleotide and protein sequence alignments were used to generate phylogenetic trees using the Neighbour-Joining and Maximum Likelihood (Tamura-Nei model) methods, each with 1,000 bootstraps. These analyses were carried out within the MEGA 6 suite 61 . Note that the trees presented in Fig. 8 were constructed with CLUSTALW protein alignments and the Neighbour-Joining method and were consistent with the trees generated using the other methods. FigTree v1.4 (http://tree.bio.ed.ac.uk/software/figtree/) was used to prepare the trees for publication. To determine the closest WRKY domain nucleotide sequence to the various RRS1 WRKY domain sequences, a database of all WRKY family genes in A. thaliana was searched using nucleotide-nucleotide megablast. Alignment of the nucleotide sequences of AtWRKY35 and RRS1 and RRS1B was accomplished using the EMBOSS MATCHER online tool 62 . To establish genomegenome synteny, the nucleotide sequences of the genomic regions harbouring the genes of interest were extracted and queried against each other using megablast. The tabular (output format 6) results were then loaded into the Artemis Genome Comparison Tool 63 , which was used to visualize and check for synteny across the various Brassicaceae RRS1/RPS4-encoding regions.