Dual recognition of structurally unrelated mildew effectors underlies the broad-spectrum resistance of Pm3e in wheat Authors: Lukas Kunz, Zoe Bernasconi, Matthias Heuberger, Jonatan Isaksson, Alexandros G. Sotiropoulos, Ursin Stirnemann, Jigisha Jigisha, Fabrizio Menardo, Thomas Wicker, Marion C. Müller, Beat Keller Published: 2026 ABSTRACT Broad-spectrum resistance genes are highly valuable for sustainable crop protection, yet the molecular basis of their activity is often unknown. The Pm3 allelic series in wheat encodes NLR receptors that recognize avirulence (AVR) effectors of wheat powdery mildew. Here, we show that near-identical Pm3 alleles vary greatly in resistance efficacy and broadness against a global mildew isolate collection and subsequently use this model system to study the mechanisms underlying broad-spectrum resistance. We demonstrate that two alleles, Pm3d and Pm3e, provide resistance against most isolates worldwide, by each recognizing two AVR genes, thereby lowering the risk of resistance breakdown. Pm3d recognizes two highly similar RNase-like AVRs, encoded by gene paralogs. In contrast, Pm3e detects two structurally diverse AVRs: one shared with Pm3d, the other originating from a large, uncharacterized protein family with a discrete structural fold. Using chimeric Pm3 NLRs, we identify specificity-defining polymorphisms of Pm3d and Pm3e against their diverse effector targets. Lastly, we demonstrate that Pm3d and Pm3e activities can be combined in engineered Pm3 NLRs, thereby further extending their recognition spectrum. Our findings highlight the potential of Pm3 immune receptors for long-lasting wheat protection by demonstrating their versatility in recognizing structurally diverse effectors and their amenability to NLR engineering. Introduction Crop protection through breeding of genetically resistant cultivars represents a cornerstone of sustainable agriculture1. The underlying resistance genes frequently encode nucleotide-binding leucine-rich repeat (NLR) receptors recognizing specific pathogen effectors (avirulence effectors, AVRs)2. AVR recognition by NLRs typically triggers a localized cell death response (hypersensitive response, HR), thereby restricting pathogen proliferation3,4. Wheat is among the most widely cultivated crops worldwide5. However, its production is threatened by various pathogens, including wheat powdery mildew (Blumeria graminis f. sp. tritici, hereafter Bgt), which is considered one of the 10 biggest challenges among fungal crop pathogens6. Wheat breeding efforts have defined several dozen resistance genes against powdery mildew, among them several that encode NLR receptors capable of recognizing parts of the impressive effector diversity of Bgt7,8. To date, the AVRs recognized by wheat NLRs Pm1, Pm2, Pm3, Pm8, Pm17 and Pm60 have been identified9–16. Interestingly, all known AVR effectors in Bgt and the closely related barley powdery mildew (Blumeria hordei, Bh), exhibit an RNase-like structure and belong to the RNase-Like Proteins associated with Haustoria (RALPH) effector superfamily17,18. RALPH effectors comprise more than half of the total effector repertoire of the Bgt and Bh pathogens17. Despite the prevalence of RALPH effector genes in the genome, Bgt also encodes several hundred candidate effector proteins that do not belong to this superfamily12. Whether some of these effectors can also be recognized by wheat immune receptors and hence represent AVRs is currently unknown. AVR effectors continuously evolve and diversify to evade NLR recognition, a phenomenon that is particularly pronounced in fast-evolving fungal pathogens. Consequently, many NLRs are overcome quickly, resulting in race-specific resistance activity that protects crops from only a subset of pathogen races19. Therefore, it is of great importance to identify and molecularly characterize new resistances with broad recognition spectra that can more durably withstand pathogen evolution. Strategies to identify broadly active resistance genes include the introgression of new resistance from closely related crop species or wild crop relatives. In recent years, NLR engineering aiming at expanding the recognition spectrum through the introduction of naturally occurring, or new, structure-inferred polymorphisms shows great promise to broaden NLR-based resistance20,21. Additionally, broadly active NLRs may exist in nature as standing genetic diversity of previously described race-specific genes. In particular, extensive allelic series of NLRs hold great promise for the identification of variants with broader recognition spectra. The Pm3 resistance gene in wheat comprises at least 17 functional alleles (Pm3a-Pm3t), each conferring race-specific resistance against Bgt. These alleles encode highly similar NLR receptors with >97% sequence identity, making Pm3 one of the most extensive allelic series in crops22,23. Several allele pairs, such as Pm3a/Pm3f and Pm3b/Pm3c, were found to recognize overlapping Bgt race spectra by recognizing the same AVR effectors yet differ in recognition sensitivity and HR amplitude9,11,24. Up to now, three AVRs of the Pm3 allelic series are known, the RALPH effectors AvrPm3a/f, AvrPm3b/c and AvrPm3d, recognized by the allelic pairs Pm3a/Pm3f, Pm3b/Pm3c and the Pm3d allele, respectively9,11. Investigations into the evolutionary origins of Pm3 alleles have identified Pm3CS as the ancestral allele, which has however been completely overcome by modern Bgt isolates25. Some functional alleles (Pm3a/Pm3b/Pm3c/Pm3f) arose from gene conversion events with other NLR genes, resulting in highly polymorphic sequence blocks, whereas others (Pm3d/Pm3e/Pm3g) are nearly identical with the ancestral Pm3CS allele, differing by only two to five amino acids (Fig. 1a)25,26. The detailed study of these Pm3CS-like alleles has so far been hampered by the lack of near-isogenic lines that would allow to define recognition spectra without confounding genetic background effects. Moreover, previous studies tested Pm3 alleles against local Bgt populations, but not against a global Bgt diversity panel27, which would be key to identifying alleles with broad-spectrum resistance.Fig. 1Pm3e and Pm3d provide broad-spectrum resistance against a global Bgt population.A Schematic representation of NLR variants encoded by allelic Pm3 genes. Black lines indicate amino acid polymorphisms compared to the ancestral, non-functional Pm3CS variant serving as a reference. AlphaFold models of the known AvrPm3a/f, AvrPm3b/c, and AvrPm3d avirulence proteins are included. Black arrows indicate recognition of the AvrPm3 proteins by their corresponding Pm3 NLR variants. B Summary of the resistance response of transgenic lines expressing individual Pm3 alleles in the wheat cultivar ‘Bobwhite’ against a global collection of 92 Bgt isolates. Bgt leaf coverage was assessed 8-9 days after infection using a scale with five categories ranging from 0 (0% leaf coverage) to 1 (100% leaf coverage). Pm3e and Pm3d provide broad-spectrum resistance against a global Bgt population. A Schematic representation of NLR variants encoded by allelic Pm3 genes. Black lines indicate amino acid polymorphisms compared to the ancestral, non-functional Pm3CS variant serving as a reference. AlphaFold models of the known AvrPm3a/f, AvrPm3b/c, and AvrPm3d avirulence proteins are included. Black arrows indicate recognition of the AvrPm3 proteins by their corresponding Pm3 NLR variants. B Summary of the resistance response of transgenic lines expressing individual Pm3 alleles in the wheat cultivar ‘Bobwhite’ against a global collection of 92 Bgt isolates. Bgt leaf coverage was assessed 8-9 days after infection using a scale with five categories ranging from 0 (0% leaf coverage) to 1 (100% leaf coverage). Here, we use transgenic wheat lines to systematically compare recognition spectra of various Pm3 alleles against 92 Bgt isolates collected worldwide. We show that Pm3d and Pm3e provide broad-spectrum resistance and identify their corresponding AVR effectors. We reveal that the Pm3e immune receptor recognizes two structurally divergent AVRs, whereas Pm3d resistance is based on the recognition of two nearly identical RALPH effectors, encoded by gene paralogs. Finally, we demonstrate that Pm3d and Pm3e resistance can be combined through NLR engineering, yielding immune receptors with extended recognition spectra. Results Pm3d and Pm3e confer broad-spectrum resistance against Bgt To compare the resistance spectra of the Pm3a-g alleles in the absence of potential genetic background effects in Pm3 donor lines, we used previously generated transgenic wheat lines expressing individual Pm3 alleles in the susceptible wheat cultivar ‘Bobwhite’28–31. We investigated the resistance provided by Pm3a, Pm3b, Pm3d, Pm3e, Pm3g and the ancestral allele Pm3CS, while omitting Pm3c and Pm3f since their resistance spectra are fully covered by Pm3b and Pm3a, respectively24 (Fig. 1A). For each allele, we chose a single transgenic event that was previously found to exhibit high transgene expression with no pleiotropic effects28–31. We infected these transgenic lines with 92 Bgt isolates selected from a previously described global Bgt panel15,27. Individual Pm3 alleles exhibited markedly different resistance spectra against the global Bgt population ranging from no resistance to broad-spectrum resistance against nearly all tested isolates (Fig. 1B). Our data confirm that the ancestral allele Pm3CS has been completely overcome by Bgt. Similarly, Pm3g, which differs from Pm3CS by five amino acids, did not confer significant resistance (Bgt leaf coverage ≤0.5) against most isolates. The two highly divergent alleles, Pm3a and Pm3b, provided resistance against 12% and 67% of the global Bgt population, respectively. Importantly, Pm3d and Pm3e provided broad-spectrum resistance against nearly all tested Bgt isolates worldwide, with only two and one virulent isolate, respectively (Fig. 1B). This finding is remarkable, given that Pm3d and Pm3e differ by only three and two amino acids from the non-functional Pm3CS (Fig. 1A). We conclude that the group of Pm3CS-like alleles, namely Pm3CS, Pm3d, and Pm3e, in combination with their corresponding AVR factors, provides a highly interesting model system to study the molecular basis of broadly active resistance alleles. However, although a previous study identified the RALPH effector gene BgtE-20069b as AvrPm3d11, the identity of the AvrPm3e gene remains unknown. Avirulence on the Pm3e resistance allele is controlled by two AVR loci in Bgt To identify AvrPm3e, we applied the Bgt avirulence depletion assay (AD assay), which relies on the generation of haploid F1 populations between parental isolates of opposite phenotype, followed by R gene-mediated selection and bulk sequencing to identify genomic regions exhibiting avirulence allele depletion (described in ref. 16). We performed three crosses between isolates exhibiting opposite phenotypes on the Pm3e#2 transgenic line. Namely, we generated crosses between the Pm3e-avirulent isolates CHN_52-27 or IRN_GOR_2 and the isolate CHN_17-40, the only Pm3e-virulent isolate in our survey (Fig. 2). Additionally, we crossed CHN_52-27 to CHVD_042201, a Swiss isolate that was recently discovered to break Pm3e resistance in an independent study (Fig. 2)32. The three F1 progeny populations were then grown on the Pm3e#2 transgenic line, resulting in a Pm3e-selected (i.e. AvrPm3e-depleted) bulk, or on the susceptible wheat cultivar ‘Kanzler’, resulting in a non-selected control bulk.Fig. 2Avirulence depletion mapping in Bgt reveals two genetic loci associated with avirulence on Pm3e.Avirulence depletion mapping of AvrPm3e loci in four independent crosses between Bgt isolates. Virulence phenotypes of the parental isolates on the Pm3e#2 transgenic wheat line are shown on the left. Statistical analysis of sequenced F1 bulks grown on ‘Kanzler’ (non-selected control) or selected on the Pm3e#2 transgenic line (Pm3e-selected) reveal regions with deviations from an expected 1:1 parental genotype ratio, indicating avirulence allele depletion. Datapoints represent individual SNP markers along the 11 chromosomes of Bgt isolate CHE_96224 (x-axis) with their -log10 transformed p-values (y-axis) when testing for deviation from expected 1:1 segregation (G-test). The two genomic regions exhibiting Pm3e-specific depletion signals are highlighted and labeled as AvrPm3e_1 (common to all four crosses) and AvrPm3e_2 (only found in the IRN_GOR_2 x CHN_17-40 and the USA_2 x CHE_94202 crosses). Avirulence depletion mapping in Bgt reveals two genetic loci associated with avirulence on Pm3e. Avirulence depletion mapping of AvrPm3e loci in four independent crosses between Bgt isolates. Virulence phenotypes of the parental isolates on the Pm3e#2 transgenic wheat line are shown on the left. Statistical analysis of sequenced F1 bulks grown on ‘Kanzler’ (non-selected control) or selected on the Pm3e#2 transgenic line (Pm3e-selected) reveal regions with deviations from an expected 1:1 parental genotype ratio, indicating avirulence allele depletion. Datapoints represent individual SNP markers along the 11 chromosomes of Bgt isolate CHE_96224 (x-axis) with their -log10 transformed p-values (y-axis) when testing for deviation from expected 1:1 segregation (G-test). The two genomic regions exhibiting Pm3e-specific depletion signals are highlighted and labeled as AvrPm3e_1 (common to all four crosses) and AvrPm3e_2 (only found in the IRN_GOR_2 x CHN_17-40 and the USA_2 x CHE_94202 crosses). Additionally, we generated Pm3e-selected and non-selected bulks from a fourth cross between the isolates USA_2 and CHE_94202, which are both avirulent on the Pm3e#2 line (Fig. 2). This cross was initially designed for an independent project; however, we recovered viable progeny on the Pm3e#2 line, suggesting that the parental isolates USA_2 and CHE_94202 carry genetically unlinked AvrPm3e components. After sequencing Pm3e-selected and unselected bulks of each cross, we identified regions with strong deviations from an expected 1:1 segregation ratio of parental genotypes to define regions under Pm3e-mediated selection, likely harboring AvrPm3e components (see Methods for details). For the crosses IRN_GOR_2 x CHN_17-40 and USA_2 x CHE_94202, our non-selected control bulks revealed one region in each cross, that deviated from the expected 1:1 ratio of parental genotypes even in the absence of Pm3e-mediated selection (Fig. 2). Such effects are likely caused by the presence of specific virulence factors or due to genetic incompatibility between parental isolates in the corresponding genomic regions (for details see Supplementary Note 1). To ensure the identification of Pm3e-specific selection signals, we excluded these regions from further analysis. Focusing on Pm3e-specific signals of selection, we identified a candidate locus on Chr-04 for all four crosses, which we defined as the AvrPm3e_1 locus. Additionally, we identified a second Pm3e-specific peak on Chr-09 in two of the four crosses which we designated as the AvrPm3e_2 locus (Fig. 2). In the USA_2 × CHE_94202 cross, the Pm3e-selected bulk showed depletion of CHE_94202 at the AvrPm3e_1 locus and depletion of USA_2 at the AvrPm3e_2 locus, indicating that each parental isolate carries a genetically distinct AvrPm3e component. Taken together, our data revealed that avirulence towards Pm3e in Bgt is likely controlled by two independently segregating AVR loci. AvrPm3e_1 is a non-RALPH avirulence effector To functionally validate AvrPm3e candidate effectors, we first focused on the AvrPm3e_1 locus on Chr-04, common to all four crosses. We defined the genomic interval exhibiting the strongest selection signal for each cross (see methods) and subsequently defined the overlap of these intervals based on the Bgt reference assembly of Pm3e-avirulent isolate CHE_96224 (Bgt_genome_v3_1633 (Supplementary Table 1). The overlapping interval had a size of 3.2 kb harboring a single gene, BgtE-5754, encoding an effector protein with 367 amino acids, which is significantly larger than previously identified AVRs in the Blumeria genus (Fig. 3A). Importantly, the BgtE-5754 gene exhibits non-synonymous polymorphisms between all respective parents of the four genetic crosses, making it a prime AvrPm3e_1 candidate (Fig. 3B, Supplementary Fig. 3A, D, G).Fig. 3AvrPm3e_1 (BgtE-5754) belongs to a structurally defined family of Bgt effectors, distinct from RALPH effector proteins.A A single genomic region on chromosome 04 exhibits a Pm3e-specific depletion signal in all four crosses. Regions with the strongest depletion signal (see Fig. 2) for each cross are shown in the CHE_96224 reference genome. The AvrPm3e_1 candidate locus (overlap between all four crosses) contains a single effector gene, BgtE-5754, which exhibits non-synonymous polymorphisms between the parental isolates of all four crosses. B Structural model of the AvrPm3e_1 candidate effector BgtE-5754 in the Pm3e-avirulent isolate CHN_52-27 and the virulent isolate CHN_17-40, with non-synonymous polymorphisms (red lines). C, D Agrobacterium-mediated co-expression of BgtE-5754_CHN_52-27 (BgtE-5754-A1) or BgtE-5754_CHN_17-40 (BgtE-5754-V1) with Pm3e-HA or a GUS negative control in N. benthamiana. Leaves were imaged with a standard camera (C) or a Fusion FX imager (D). The experiment was repeated three times with n = 6 leaves per experiment (n = 18). E Agrobacterium-mediated co-expression of Pm3e-HA with various BgtE-5754 (AvrPm3e_1) variants or GUS. Boxplots show HR intensities from 18 infiltration spots across three experiments (symbols indicate experiments). Boxes represent the interquartile range (IQR; 25–75%) with the thick horizontal line indicating the median, and whiskers extending to the most extreme values within 1.5×IQR. Letters above boxplots denote statistically significant differences at p < 0.05 based on Dunn’s test. Details and exact p-values for statistical tests are reported in Supplementary Dataset 4. Representative pictures for each combination are shown. F Schematic representation of the BgtE-5754 effector protein showing all naturally occurring amino acid polymorphisms found in Bgt compared to the BgtE-5754-A1 variant. AlphaFold3 structural prediction of BgtE-5754 is shown below. Model quality scores are provided in Supplementary Fig. 5. Polymorphisms in black have no effect on Pm3e recognition; red ones disrupt recognition. G, J Comparison between AvrPm3e_1-like genes/proteins and those belonging to the RALPH effector family. Datapoints represent individual effectors. G Effector expression levels in reference isolate CHE_96224 at two days post-inoculation. H Number of exons per gene. I Protein length of all family members. J Number of proteins predicted to have a signal peptide (SignalP6.0). AvrPm3e_1 (BgtE-5754) belongs to a structurally defined family of Bgt effectors, distinct from RALPH effector proteins. A A single genomic region on chromosome 04 exhibits a Pm3e-specific depletion signal in all four crosses. Regions with the strongest depletion signal (see Fig. 2) for each cross are shown in the CHE_96224 reference genome. The AvrPm3e_1 candidate locus (overlap between all four crosses) contains a single effector gene, BgtE-5754, which exhibits non-synonymous polymorphisms between the parental isolates of all four crosses. B Structural model of the AvrPm3e_1 candidate effector BgtE-5754 in the Pm3e-avirulent isolate CHN_52-27 and the virulent isolate CHN_17-40, with non-synonymous polymorphisms (red lines). C, D Agrobacterium-mediated co-expression of BgtE-5754_CHN_52-27 (BgtE-5754-A1) or BgtE-5754_CHN_17-40 (BgtE-5754-V1) with Pm3e-HA or a GUS negative control in N. benthamiana. Leaves were imaged with a standard camera (C) or a Fusion FX imager (D). The experiment was repeated three times with n = 6 leaves per experiment (n = 18). E Agrobacterium-mediated co-expression of Pm3e-HA with various BgtE-5754 (AvrPm3e_1) variants or GUS. Boxplots show HR intensities from 18 infiltration spots across three experiments (symbols indicate experiments). Boxes represent the interquartile range (IQR; 25–75%) with the thick horizontal line indicating the median, and whiskers extending to the most extreme values within 1.5×IQR. Letters above boxplots denote statistically significant differences at p < 0.05 based on Dunn’s test. Details and exact p-values for statistical tests are reported in Supplementary Dataset 4. Representative pictures for each combination are shown. F Schematic representation of the BgtE-5754 effector protein showing all naturally occurring amino acid polymorphisms found in Bgt compared to the BgtE-5754-A1 variant. AlphaFold3 structural prediction of BgtE-5754 is shown below. Model quality scores are provided in Supplementary Fig. 5. Polymorphisms in black have no effect on Pm3e recognition; red ones disrupt recognition. G, J Comparison between AvrPm3e_1-like genes/proteins and those belonging to the RALPH effector family. Datapoints represent individual effectors. G Effector expression levels in reference isolate CHE_96224 at two days post-inoculation. H Number of exons per gene. I Protein length of all family members. J Number of proteins predicted to have a signal peptide (SignalP6.0). Signal peptide (SP) prediction using different software was inconsistent for BgtE-5754 (see Supplementary Fig. 1, Supplementary Note 2). Therefore, we co-expressed the entire coding sequence of BgtE-5754 with Pm3e in Nicotiana benthamiana and assessed its ability to elicit a Pm3e-specific hypersensitive response (HR). First, we focused on the haplovariants identified in the parental isolates CHN_52-27 (Pm3e-avirulent) and CHN_17-40 (virulent). Indeed, the BgtE-5754_CHN_52-27 variant (identical to CHE_96224 reference) elicited strong Pm3e-dependent HR, whereas the variant from the virulent parent CHN_17-40, differing by four amino acids (R166Q, L171P, R256Q, N313K), did not (Fig. 3B–D). Importantly, neither of the two BgtE-5754 variants, nor Pm3e, induced an HR response upon co-expression with a GUS negative control (Fig. 3C, D). We therefore concluded that BgtE-5754 is indeed AvrPm3e_1 and designated the avirulent BgtE-5754_CHN_52-27 variant as BgtE-5754-A1 and the virulent BgtE-5754_CHN_17_40 variant as BgtE-5754-V1. Of note, expression of N-terminally truncated versions of BgtE-5754-A1, corresponding to different putative SP predictions, did not result in Pm3e-mediated HR, suggesting that full-length BgtE-5754 acts as AvrPm3e_1 (Supplementary Fig. 1, Supplementary Note 2). For the remaining crosses, the genetic analysis suggested that IRN_GOR_2 and CHE_94202 would carry recognized variants of BgtE-5754. BgtE-5754_IRN_GOR_2 (hereafter BgtE-5754-A2) and BgtE-5754_CHE_94202 (BgtE-5754-A3) differ by three and one amino acids from BgtE-5754-A1, respectively (R256Q, Q335H, E367G in IRN_GOR_2; R256Q in CHE_94202). Both variants triggered strong HR upon co-expression with Pm3e (Supplementary Fig. 2). Conversely, the parental isolates CHVD_042201 and USA_2 are predicted to contain non-recognized variants of BgtE-5754. In both isolates, BgtE-5754 differs by two amino acid polymorphisms from BgtE-5754-A1 (R256Q, R309K in CHVD_042201; R256Q, R312M in USA_2), consequently named BgtE-5754-V2 and V3. In agreement with our genetic analysis, neither BgtE-5754-V2 nor BgtE-5754-V3 elicited a Pm3e-mediated HR response (Supplementary Fig. 2). We therefore concluded that BgtE-5754 represents the AvrPm3e component responsible for the mapped AvrPm3e_1 region in all four crosses. To ensure that the lack of recognition of virulent BgtE-5754 variants was not due to an absence of protein production in N. benthamiana, we tagged representative avirulent and virulent variants BgtE-5754-A1, A2, V1 and V2 with C-terminal mRFP-FLAG epitopes. Western blot analysis revealed that all four tested variants were efficiently produced in N. benthamiana, indicating that amino acid polymorphisms in the virulent variants BgtE-5754-V1 and V2 do not negatively affect protein amounts, but are rather involved in Pm3e evasion (Supplementary Fig. 3). Interestingly, the western blot analysis also revealed that BgtE-5754 might undergo posttranslational processing in N. benthamiana, although it is currently unclear whether this phenomenon is biologically relevant (see Supplementary Note 3). In addition to the six BgtE-5754 variants identified in the parental isolate used for mapping, we detected five additional variants among 220 global Bgt isolates27. Of these, three triggered a Pm3e-dependent HR (BgtE-5754-A4, A5, A6), whereas the remaining two did not (BgtE-5754-V4, V5) (Fig. 3E, Supplementary Note 4). To identify the polymorphisms responsible for Pm3e evasion, we tested consequences of individual amino acid substitutions in the BgtE-5754-A1 backbone, revealing that all naturally occurring polymorphisms that evade Pm3e-mediated recognition are located in a short C-terminal stretch of the BgtE-5754 protein (Fig. 3F, Supplementary Note 4, Supplementary Fig. 4). In the absence of structural information for BgtE-5754 or any closely related effectors, we used AlphaFold to predict its structure (Supplementary Fig. 5). This revealed that BgtE-5754 is structurally unrelated to the RNase-like fold found in all previously identified AVR effectors in Bgt, belonging to the RALPH effector superfamily (Fig. 3F). Pfam analysis did not reveal any strong homology to known proteins. Interestingly, four of the immune-evasive amino acid polymorphisms affect three positions within the same predicted alpha helix in the BgtE-5754 structure, while the fifth affects the neighboring alpha helix (Fig. 3F), indicating this C-terminal domain is involved in BgtE-5754 recognition. The observation that BgtE-5754 is significantly larger than previously identified AvrPm3 effectors and appears structurally unrelated to RALPH effectors prompted us to investigate this effector structure in more detail. Based on Alphafold2 structural modeling and TM-align to infer structural similarity, we found 112 proteins exhibiting structural similarity with BgtE-5754 (AvrPm3e_1) in Bgt, which we consequentialy refer to as the AvrPm3e_1-like superfamily (for details see Supplementary Note 5). This indicates that AvrPm3e_1-like effectors, comparable to the RALPH superfamily, constitute an expanded and diversified effector superfamily that constitutes ~13% of the effectors in Bgt (Supplementary Note 5). RALPH effector genes have been previously described to exhibit high average expression levels at early infection stages and to consist of two protein-coding exons resulting in small effector proteins (100-150 residues) that predominantly contain N-terminal SPs for export via the secretory pathway (Fig. 3G–J,17,18). In contrast, we found AvrPm3e_1-like effector genes to exhibit lower average expression during haustorial stages and to predominantly consist of a single large exon that gives rise to much larger effector proteins ( ~ 300–400 residues) (Fig. 3G–I). Additionally, we found that AvrPm3e_1-like effectors would often lack a classic SP (Fig. 3J), as exemplified by AvrPm3e_1 (BgtE-5754) itself, indicating they might be secreted by non-conventional pathways (Supplementary Fig. 1, Supplementary Note 2). Based on structural predictions from a previous study34 and publicly available databases, we investigated the effector repertoire of Bgt sister species Bh and other more divergent powdery mildew species. This analysis identified 82 AvrPm3e_1-like effector proteins in Bh, whereas none were identified in other powdery mildew species, suggesting that this effector group is specific to the Blumeria genus (see Supplementary Note 5). Pm3e recognizes a second AVR effector, AvrPm3e_2, with an RNase-like structure In the next step, we investigated the AvrPm3e_2 locus on Chr-09 for the presence of an additional AVR factor (Fig. 2). This locus was identified in two out of the four performed genetic crosses (IRN_GOR_2 x CHN_17-40 and USA2 x CHE_94202). We defined the AvrPm3e_2 locus based on the overlapping genomic regions with the strongest selection signals for both crosses based on the reference assembly of CHE_96224 (Bgt_genome_v3_16) (Fig. 4A). The identified interval of 312 kb contains 13 high-quality annotated genes including three effector candidates. Importantly, a single effector candidate (BgtE-20069b in CHE_96224) is polymorphic between parental isolates of both crosses and was therefore considered the best AvrPm3e_2 candidate (Fig. 4A, B). Between IRN_GOR_2 and CHN_17-40, the BgtE-20069b effector differs by two amino acid polymorphisms, BgtE-20069b_T86I in IRN_GOR_2 (using BgtE-20069b in CHE_96224 as a reference) and BgtE-20069b_S74T in CHN_17-40 (Fig. 4B). In the second cross, the BgtE-20069b sequence found in USA_2 is identical with the reference assembly of CHE_96224, whereas the effector gene is deleted in CHE_94202 (Supplementary Fig. 6A).Fig. 4AvrPm3e_2 (BgtE-20069b) is a RALPH effector.A Genomic region on chromosome 09 exhibiting a Pm3e-specific depletion signal in two out of four crosses (IRN_GOR_2 x CHN_17-40 and USA_2 x CHE_94202). Regions with the strongest depletion signal for both crosses are shown and aligned to the CHE_96224 reference genome. The AvrPm3e_2 candidate locus, defined by the overlap between both crosses, in the reference genome CHE_96224 is shown as a zoom-in. Effector genes (red) and non-effector genes (gray) are represented by arrows. Gene models are not drawn to scale. Effectors exhibiting non-synonymous polymorphisms in the parental isolates of individual crosses are marked by orange (IRN_GOR-2 x CHN_17-40) and dark red (USA_2 x CHE_94202) arrowheads. B Schematic protein model of the AvrPm3e_2 candidate effector BgtE-20069b in the Pm3e-avirulent isolate IRN_GOR-2 and virulent isolate CHN_17-40. Sequence polymorphisms are defined in comparison with BgtE-20069b sequence in the CHE_96224 reference isolate and are highlighted by black lines. C, D Agrobacterium-mediated co-expression of BgtE-20069b_T86I-mRFP or BgtE-20069b_S74T-mRFP with Pm3e-HA or a GUS negative control in N. benthamiana. Leaves were imaged with a camera (C) or a Fusion FX imager (D). The experiment was performed three times with n = 6 leaves per experiment (total n = 18). E Predicted 3D structure of BgtE-20069b_T86I according to Alphafold3 modeling. Coloring represents predicted local distance difference test (pLDDT) local confidence according to Alphafold (100 (dark red) – 0 (dark blue)). AvrPm3e_2 (BgtE-20069b) is a RALPH effector. A Genomic region on chromosome 09 exhibiting a Pm3e-specific depletion signal in two out of four crosses (IRN_GOR_2 x CHN_17-40 and USA_2 x CHE_94202). Regions with the strongest depletion signal for both crosses are shown and aligned to the CHE_96224 reference genome. The AvrPm3e_2 candidate locus, defined by the overlap between both crosses, in the reference genome CHE_96224 is shown as a zoom-in. Effector genes (red) and non-effector genes (gray) are represented by arrows. Gene models are not drawn to scale. Effectors exhibiting non-synonymous polymorphisms in the parental isolates of individual crosses are marked by orange (IRN_GOR-2 x CHN_17-40) and dark red (USA_2 x CHE_94202) arrowheads. B Schematic protein model of the AvrPm3e_2 candidate effector BgtE-20069b in the Pm3e-avirulent isolate IRN_GOR-2 and virulent isolate CHN_17-40. Sequence polymorphisms are defined in comparison with BgtE-20069b sequence in the CHE_96224 reference isolate and are highlighted by black lines. C, D Agrobacterium-mediated co-expression of BgtE-20069b_T86I-mRFP or BgtE-20069b_S74T-mRFP with Pm3e-HA or a GUS negative control in N. benthamiana. Leaves were imaged with a camera (C) or a Fusion FX imager (D). The experiment was performed three times with n = 6 leaves per experiment (total n = 18). E Predicted 3D structure of BgtE-20069b_T86I according to Alphafold3 modeling. Coloring represents predicted local distance difference test (pLDDT) local confidence according to Alphafold (100 (dark red) – 0 (dark blue)). A previous study has identified BgtE-20069b as AvrPm3d and showed that it triggers a strong Pm3d-dependent HR in N. benthamiana11. The study also investigated whether BgtE-20069b is recognized by other Pm3 NLRs and concluded that it is not recognized by Pm3e11. Based on our genetic indications, we tested BgtE-20069b_T86I present in the avirulent parental isolate IRN_GOR_2 and BgtE-20069b_S74T from the Pm3e-virulent isolate CHN_17-40 for Pm3e-mediated HR in N. benthamiana. Consistent with the previous study11, we did not observe an HR response with any of the tested BgtE-20069b variants in combination with Pm3e, whereas BgtE-20069b_T86I was readily recognized by Pm3d (Supplementary Fig. 7). In the same study, it was observed that the BgtE-20069b protein in N. benthamiana was undetectable on Western blots, indicating that the effector protein might be unstable in planta, despite triggering HR upon co-expression with Pm3d11. Based on this observation, we attempted to stabilize BgtE-20069b by a C-terminal fusion of the effector with a highly soluble mRFP fluorophore. Indeed, co-expression of stabilized BgtE-20069b_T86I-mRFP with Pm3e resulted in an HR response, whereas BgtE-20069b_S74T-mRFP found in the virulent parental isolate CHN_17-40 did not (Fig. 4C, D). Similarly, BgtE-20069b-mRFP from the avirulent isolate USA_2 of the second cross induced a Pm3e-mediated HR response (Supplementary Fig. 6B, C). Therefore, we concluded that BgtE-20069b, in addition to representing AvrPm3d, is indeed AvrPm3e_2 and that the lack of previous recognition of this effector by Pm3e in the N. benthamiana experimental system can likely be attributed to a lack of protein stability. Thus, BgtE-20069b is an AVR for both Pm3d and Pm3e, but the two NLRs exhibit differences in sensitivity towards this effector. Owing to this complexity, we will refer to this AVR as BgtE-20069b instead of AvrPm3d/AvrPm3e_2 and specify the exact protein variants of BgtE-20069b throughout the manuscript. Importantly, BgtE-20069b is a RALPH effector with a predicted RNase-like structure (Fig. 4E,11,17). We conclude that Pm3e recognizes two structurally unrelated avirulence proteins, thereby explaining the broad-spectrum resistance activity of this Pm3 allele against the global Bgt diversity. Pm3d recognizes two nearly identical RNase-like AVRs, encoded by gene paralogs In our survey, Pm3d also exhibited broad resistance activity, comparable to Pm3e (Fig. 1B). This raised the question of whether Pm3d also recognizes multiple AVRs. Bourras et al. (2019) showed that BgtE-20069b in Bgt reference isolate CHE_96224 is recognized by Pm3d and that Pm3d-virulent isolates, like CHE_94202, lack BgtE-20069b due to a gene deletion11. However, a UV mutant of the Pm3d-avirulent isolate CHE_96224, termed Bgt_FAB, which was generated in our lab as part of an independent project35 coincidentally contains a deletion of the entire coding sequence of BgtE-20069b but exhibits an avirulent phenotype on the Pm3d#1 transgenic line (Fig. 5A, Supplementary Table 2). This strongly supports the existence of a second AvrPm3d component (hereafter AvrPm3d_2).Fig. 5Avirulence on Pm3d is controlled by two neighboring paralogous AVR genes encoding highly similar RALPH effector proteins.A The UV mutant Bgt_FAB exhibits an avirulent phenotype on the Pm3d#1 wheat line, despite carrying a deletion of the BgtE-20069b (AvrPm3d_1/AvrPm3e_2) effector gene. The phenotype of wild type Bgt isolate CHE_96224, used for UV mutagenesis, is shown as comparison. B Avirulence depletion mapping on Pm3d in two Bgt crosses. Virulence phenotypes of the parental isolates on the Pm3d#1 line are shown. Statistical analysis of sequenced F1 bulks selected on ‘Kanzler’ (non-selected control) or the Pm3d#1 line (Pm3d-selected) are shown on the right. ‘Kanzler’ bulks are drawn with gray lines to represent that these panels were reprinted from Fig. 2. Datapoints represent individual SNP markers along the 11 chromosomes of Bgt isolate CHE_96224 (x-axis) with their -log10 transformed p-values (y-axis) when testing for deviation from expected 1:1 segregation (G-test). Pm3d-specific depletion signal is indicated by a red arrowhead. C The AvrPm3d candidate locus, defined by the overlap between both crosses, in the reference genome CHE_96224. Effector genes (red) and non-effector genes (gray) are represented by arrows. Gene models are not drawn to scale. Effectors exhibiting non-synonymous polymorphisms in the parental isolates of individual crosses are marked by yellow (CHN_52-27 x CHVD_042001) and dark red (USA_2 x CHE_94202) arrowheads. D Gene models of the neighboring effector paralogs BgtE-20069a and BgtE-20069b in the Pm3d-avirulent isolate USA_2 and virulent isolate CHE_94202. The non-synonymous mutation (leading to a R75K substitution in BgtE-20069a) is highlighted by a red vertical line, the deletion of BgtE-20069b in CHE_94202 is indicated by crossed black lines. E, F Agrobacterium-mediated co-expression of BgtE-20069a-mRFP (AvrPm3d_2) or BgtE-20069a_R75K-mRFP with Pm3d-HA or a GUS negative control in N. benthamiana. Leaves were imaged with a camera (E) or a Fusion FX imager (F). The experiment was performed three times with six leaves (total n = 18). G Representative phenotyping pictures of isolates that carry a BgtE-20069b deletion on the Pm3#d1 transgenic line and the corresponding sister line Sd#1. For each isolate, a schematic representation of the two genes is depicted. A black cross indicates BgtE-20069b deletion. Avirulence on Pm3d is controlled by two neighboring paralogous AVR genes encoding highly similar RALPH effector proteins. A The UV mutant Bgt_FAB exhibits an avirulent phenotype on the Pm3d#1 wheat line, despite carrying a deletion of the BgtE-20069b (AvrPm3d_1/AvrPm3e_2) effector gene. The phenotype of wild type Bgt isolate CHE_96224, used for UV mutagenesis, is shown as comparison. B Avirulence depletion mapping on Pm3d in two Bgt crosses. Virulence phenotypes of the parental isolates on the Pm3d#1 line are shown. Statistical analysis of sequenced F1 bulks selected on ‘Kanzler’ (non-selected control) or the Pm3d#1 line (Pm3d-selected) are shown on the right. ‘Kanzler’ bulks are drawn with gray lines to represent that these panels were reprinted from Fig. 2. Datapoints represent individual SNP markers along the 11 chromosomes of Bgt isolate CHE_96224 (x-axis) with their -log10 transformed p-values (y-axis) when testing for deviation from expected 1:1 segregation (G-test). Pm3d-specific depletion signal is indicated by a red arrowhead. C The AvrPm3d candidate locus, defined by the overlap between both crosses, in the reference genome CHE_96224. Effector genes (red) and non-effector genes (gray) are represented by arrows. Gene models are not drawn to scale. Effectors exhibiting non-synonymous polymorphisms in the parental isolates of individual crosses are marked by yellow (CHN_52-27 x CHVD_042001) and dark red (USA_2 x CHE_94202) arrowheads. D Gene models of the neighboring effector paralogs BgtE-20069a and BgtE-20069b in the Pm3d-avirulent isolate USA_2 and virulent isolate CHE_94202. The non-synonymous mutation (leading to a R75K substitution in BgtE-20069a) is highlighted by a red vertical line, the deletion of BgtE-20069b in CHE_94202 is indicated by crossed black lines. E, F Agrobacterium-mediated co-expression of BgtE-20069a-mRFP (AvrPm3d_2) or BgtE-20069a_R75K-mRFP with Pm3d-HA or a GUS negative control in N. benthamiana. Leaves were imaged with a camera (E) or a Fusion FX imager (F). The experiment was performed three times with six leaves (total n = 18). G Representative phenotyping pictures of isolates that carry a BgtE-20069b deletion on the Pm3#d1 transgenic line and the corresponding sister line Sd#1. For each isolate, a schematic representation of the two genes is depicted. A black cross indicates BgtE-20069b deletion. Since BgtE-5754 (AvrPm3e_1) does not trigger Pm3d-mediated HR in N. benthamiana (Supplementary Fig. 8), we hypothesized that AvrPm3d_2 encodes a third, unknown effector. To identify AvrPm3d_2, we performed the avirulence depletion assay using two of the four Bgt crosses, namely USA_2 x CHE_94202 and CHN_52-27 x CHVD_042201 for which parental isolates exhibit opposite phenotypes on the Pm3d#1 transgenic line (Fig. 5B). We identified a single Pm3d-specific locus on Chr-09 for both crosses (Fig. 5C), which corresponded to the AvrPm3d_1/AvrPm3e_2 locus (Fig. 2). These results confirm that AvrPm3e_1 (BgtE-5754), residing on Chr-04, is a Pm3e-specific avirulence gene and suggest that the unknown AvrPm3d_2 component is genetically closely linked to AvrPm3d_1/AvrPm3e_2 (BgtE-20069b). We therefore defined the overlapping intervals from the two crosses in the Bgt reference assembly of CHE_96224. Within the resulting 453 kb interval, only two effector genes were polymorphic between the parental isolates of both crosses: BgtE-20069b, which is deleted in the virulent parental isolates CHVD_042201 and CHE_94202, and its directly adjacent paralogous gene BgtE-20069a which exhibits amino acid polymorphisms between the parents of both crosses (Fig. 5C, D, Supplementary Fig. 9). In a first step, we confirmed that both BgtE-20069b variants found in both avirulent isolates USA_2 (BgtE-20069b) and CHN_52-27 (BgtE-20069b_H96N) induced a Pm3d-mediated HR therefore replicating the findings of Bourras at al. (2019)11 and confirming BgtE-20069b as AvrPm3d_1 in both crosses (Supplementary Fig. 9). Next, we investigated if BgtE-20069a is AvrPm3d_2. In the reference isolate CHE_96224, and in the avirulent parental isolate USA_2, the two paralogs BgtE-20069b and BgtE-20069a differ by two polymorphisms (F53S and T86I). Due to their similarity, BgtE-20069a has previously been tested for recognition by Pm3d in N. benthamiana, however, no HR response was observed11. Given our observation that the BgtE-20069b effector can be stabilized by fusion to an mRFP fluorophore we followed a similar strategy for BgtE-20069a. Co-expression of BgtE-20069a-mRFP from avirulent isolate USA_2 with Pm3d in N. benthamiana triggered HR, whereas the R75K variant from virulent parent CHE_94202 (BgtE-20069a_R75K-mRFP) did not (Fig. 5E, F). Similarly, BgtE-20069a_V108G-mRFP from avirulent isolate CHN_52-27 was recognized by Pm3d, while the virulent parent CHVD_042201 also carried the immune evasive R75K variant (Supplementary Fig. 9D). Thus, we concluded that BgtE-20069a is AvrPm3d_2 and that R75K represents a gain-of-virulence mutation. We verified protein production of the Pm3d-immune evasive BgtE-20069a_R75K, along the recognized variants BgtE-20069b_T86I, BgtE-20069b_S74T and BgtE-20069a, via C-terminal mRFP-FLAG epitope tagging, followed by Western blot analysis. All four variants were efficiently produced to similar levels in N. benthamiana, indicating that the R75K gain-of-virulence mutation does not affect effector stability, but rather interferes with the Pm3d recognition (Supplementary Fig. 10). Based on mapping coverage33, we identified 12 among 220 analyzed global isolates, that carry a BgtE-20069b deletion (Supplementary Table 3). Analysis of the BgtE-20069a sequence in these isolates revealed five isolates with the reference BgtE-20069a sequence as either a single copy gene (three isolates) or as a duplication (two isolates) as well as seven isolates with the R75K gain-of-virulence mutation (Supplementary Table 3). Phenotyping of two representative isolates for each category on Pm3d#1 transgenic wheat revealed that isolates with two copies of BgtE-20069a exhibit an avirulent phenotype, whereas isolates with a single copy of BgtE-20069a showed an intermediate phenotype. Importantly, only isolates carrying BgtE-20069_R75K are fully virulent on Pm3d#1 (Fig. 5g). Taken together, our results show that Pm3d-mediated resistance is indeed based on the recognition of the two paralogous effectors BgtE-20069b (AvrPm3d_1/AvrPm3e_2) and BgtE-20069a (AvrPm3d_2) and that virulence on Pm3d wheat requires the disruption of both AVRs, by gene deletion of BgtE-20069b and the simultaneous presence of the R75K mutation in BgtE-20069a. Pm3d and Pm3e specificity is defined by single amino acids in the LRR domain Given that both Pm3d and Pm3e recognize the RALPH effector BgtE-20069b (AvrPm3d_1/AvrPm3e_2), we compared the activity of both Pm3 alleles towards various mRFP-stabilized BgtE-20069b and BgtE-20069a (AvrPm3d_2) variants upon co-expression in N. benthamiana. We selected four variants of BgtE-20069b and three variants of BgtE-20069a found in the parental isolates of our four genetic crosses. Pm3d showed the strongest immune response against BgtE-20069b and BgtE-20069b_T86I, an intermediate response to BgtE-20069b_S74T and BgtE-20069b_H96N and a somewhat weaker response against BgtE-20069a and BgtE-20069a_V108G (Fig. 6a, Supplementary Fig. 11). Only BgtE-20069a_R75K completely evaded recognition by Pm3d (Fig. 6a, Supplementary Fig. 11). In contrast, Pm3e only mounted an immune response against BgtE-20069b and BgtE-20069b_T86I but did not recognize other BgtE-20069b or BgtE-20069a variants (Fig. 6a, Supplementary Fig. 11). We therefore conclude that Pm3d clearly excels in recognizing the BgtE-20069b/BgtE-20069a RALPH effector variants by mounting a stronger and broader immune response (Fig. 6a). In contrast, Pm3e recognizes fewer BgtE-20069b variants and none of the BgtE-20069a variants but has the unique ability to recognize the structurally unrelated BgtE-5754 (AvrPm3e_1) effector.Fig. 6Single amino acid polymorphisms in the LRR domain of Pm3d and Pm3e determine recognition specificities toward their AVR effectors.A Recognition strength of Pm3d-HA and Pm3e-HA towards variants of BgtE-20069b-mRFP (AvrPm3d_1/AvrPm3e_2) and BgtE-20069a-mRFP (AvrPm3d_2) in N. benthamiana, shown as a heat map. Schematic representations of amino acid polymorphisms defining Pm3d and Pm3e are shown on the left. Schematic representations of BgtE-20069b/BgtE-20069a variants with defining amino acid polymorphisms relative to the BgtE-20069b variant in reference isolate CHE_96224 (left) are shown above the heat map. The heatmap shows the mean HR strength from n = 18 individual replicates across three independent experiments (full dataset is shown in Supplementary Fig. 11). B Heatmap depicting recognition strength of 16 C-terminally HA-tagged Pm3CS/Pm3d/Pm3e NLR chimeras towards BgtE-5754-A1 (AvrPm3e_1) and selected C-terminally mRFP-tagged BgtE-20069b/BgtE-20069a variants. Schematic representations of amino acid polymorphisms defining the individual chimeras are shown on the right. Amino acid polymorphism combinations are illustrated in the drawing relative to the Pm3CS reference. The heatmap shows the mean HR strength from n = 18 individual replicates across three independent experiments (full dataset is shown in Supplementary Fig. 14). C AlphaFold3 structural prediction of the LRR domain of the Pm3_Ch16 construct combining all Pm3d-/Pm3e-specific polymorphisms (highlighted in orange). D Schematic summary of datasets depicted in (B) and Supplementary Fig. 14, highlighting Pm3d/Pm3e-specific polymorphisms that positively or negatively influence recognition of BgtE-5754 (top) or BgtE-20069b/BgtE-20069a (bottom). Polymorphisms highlighted by a yellow or red line are required for recognition of BgtE-5754 and BgtE-20069b/BgtE-20069a, respectively. Polymorphisms that quantitatively influence recognition strength are indicated by arrows (upward arrow = increase HR; downward arrow = decrease HR). Single amino acid polymorphisms in the LRR domain of Pm3d and Pm3e determine recognition specificities toward their AVR effectors. A Recognition strength of Pm3d-HA and Pm3e-HA towards variants of BgtE-20069b-mRFP (AvrPm3d_1/AvrPm3e_2) and BgtE-20069a-mRFP (AvrPm3d_2) in N. benthamiana, shown as a heat map. Schematic representations of amino acid polymorphisms defining Pm3d and Pm3e are shown on the left. Schematic representations of BgtE-20069b/BgtE-20069a variants with defining amino acid polymorphisms relative to the BgtE-20069b variant in reference isolate CHE_96224 (left) are shown above the heat map. The heatmap shows the mean HR strength from n = 18 individual replicates across three independent experiments (full dataset is shown in Supplementary Fig. 11). B Heatmap depicting recognition strength of 16 C-terminally HA-tagged Pm3CS/Pm3d/Pm3e NLR chimeras towards BgtE-5754-A1 (AvrPm3e_1) and selected C-terminally mRFP-tagged BgtE-20069b/BgtE-20069a variants. Schematic representations of amino acid polymorphisms defining the individual chimeras are shown on the right. Amino acid polymorphism combinations are illustrated in the drawing relative to the Pm3CS reference. The heatmap shows the mean HR strength from n = 18 individual replicates across three independent experiments (full dataset is shown in Supplementary Fig. 14). C AlphaFold3 structural prediction of the LRR domain of the Pm3_Ch16 construct combining all Pm3d-/Pm3e-specific polymorphisms (highlighted in orange). D Schematic summary of datasets depicted in (B) and Supplementary Fig. 14, highlighting Pm3d/Pm3e-specific polymorphisms that positively or negatively influence recognition of BgtE-5754 (top) or BgtE-20069b/BgtE-20069a (bottom). Polymorphisms highlighted by a yellow or red line are required for recognition of BgtE-5754 and BgtE-20069b/BgtE-20069a, respectively. Polymorphisms that quantitatively influence recognition strength are indicated by arrows (upward arrow = increase HR; downward arrow = decrease HR). Thus, Pm3d and Pm3e exclusively recognize BgtE-20069b_S74T and BgtE-20069a, or BgtE-5754, respectively. This provided the opportunity to study the molecular basis of NLR specificity among closely related, allelic NLR variants. Importantly, the Pm3d and Pm3e NLRs and the non-functional Pm3CS differ at only four amino acid positions, all located within the LRR repeats (Fig. 6B, C). Both Pm3d and Pm3e share the W657 polymorphism (R657 in Pm3CS) in the LRR repeat 4. Additionally, Pm3d contains two polymorphisms, namely R1153 (G1153) and R1356 (T1356) in the C-terminal LRR repeats 22 and 28. Pm3e exhibits a single additional polymorphism, V1332 (E1332) in the C-terminal LRR repeat 27 (Fig. 6B, C). We introduced each polymorphism either alone or in combinations into the Pm3CS backbone, resulting in 16 chimeric constructs (Pm3_Ch1-16), covering the three natural alleles (Pm3CS, Pm3d and Pm3e) and 13 artificial chimeras (Fig. 6B). We verified protein stability of all chimeras and the absence of autoactivity upon expression in N. benthamiana (Supplementary Fig. 12,13). Next, we co-expressed the 16 NLR variants with BgtE-20069b_S74T and BgtE-20069a (recognized only by Pm3d), and BgtE-5754 (recognized only by Pm3e). Additionally, we tested the 16 chimeras against BgtE-20069b_T86I, which induces strong HR with Pm3d and weaker HR with Pm3e (Fig. 6A). This analysis revealed that the W657 polymorphism is necessary for the recognition of the RALPH effector BgtE-20069b and its variants. While W657 alone is sufficient to induce a HR response against BgtE-20069b_T86I, it must be combined with other polymorphisms to strongly recognize BgtE-20069b_T86I or to mount an HR response against the more immune-evasive effector variants BgtE-20069b_S74T and BgtE-20069a. In particular, R1356 (found in Pm3d) acts strongly synergistically with W657. The combination of the three amino acids present in Pm3d appears to be optimal as none of the other chimeras recognized the BgtE-20069a variant (Fig. 6B, Supplementary Fig. 14A–C, for details see Supplementary Note 6). In contrast, the V1332 polymorphism, naturally present in Pm3e, is crucial for recognizing the effector BgtE-5754, as none of the chimeras lacking it showed an HR response. Although V1332 alone is sufficient, its effect is influenced by other SNPs: the R1356 polymorphism, present in Pm3d but not in Pm3e, acts antagonistically with V1332 in BgtE-5754 recognition, an effect that can be compensated by the simultaneous presence of W657 (Fig. 6B, Supplementary Fig. 14A–C, for details see Supplementary Note 6). Taken together, our findings show that the unique abilities of Pm3d and Pm3e rely on only a few polymorphisms in the LRR domain of the NLR protein. Consequently, the non-functional Pm3CS can be rendered active against BgtE-20069b or BgtE-5754 by the introduction of single polymorphisms, W657 and V1332, respectively. Furthermore, we show that additional polymorphisms can act either synergistically or antagonistically, and that the final combination balances these effects to generate a functional NLR (Fig. 6B–D). Pm3e and Pm3d specificities can be combined in chimeric NLRs Stacking of Pm3 alleles using transgenic methods has shown great promise to enhance Bgt resistance in wheat36. However, several Pm3 alleles were found to be incompatible due to interallelic suppression effects37. We therefore tested for interallelic suppression between Pm3d and Pm3e in wheat and N. benthamiana. In both experimental systems, we observed suppression of Pm3e activity by Pm3d and vice versa and concluded that combining the two R genes in the same wheat line would be of limited use (Fig. 7A, B, Supplementary Note 7).Fig. 7Pm3d and Pm3e recognition specificity can be combined in chimeric Pm3 NLRs.A,B Virulence phenotypes of isolates CHN_17_40 (A) and CHE_94202 (B) on various wheat lines. “BW” refers to the cultivar ‘Bobwhite’, the background for the Pm3d#1 and Pm3e#2 transgenic lines. “Pm3d#1 / Pm3e#2” denotes F1 plants from a Pm3d#1 × Pm3e#2 cross. Sd#1 and Se#2 refer to transgene-free sister lines of Pm3d#1 and Pm3e#2. Boxplots display infection scores (10-step scale, 0% to 100% leaf coverage) at 11dpi. Eighteen leaf segments across three independent plates were scored (symbols indicate plates). Statistical significance between the Pm3d#1 x Pm3e#2 F1 individuals and the respective parental lines was assessed using the Wilcoxon rank-sum test; *** indicates p < 0.001. C Results of allelic suppression testing in N. benthamiana using the effector variants BgtE-20069b_S74T-mRFP, BgtE-20069a-mRFP (both exclusively recognized by Pm3d-HA) and BgtE-5754-A1 (exclusively recognized by Pm3e-HA). Boxplots show quantified HR intensities from 18 infiltration spots per combination, obtained from three independent experiments (denoted by distinct symbols). Statistical significance was assessed using a paired Wilcoxon signed-rank test; *** indicates p < 0.001. D–F Agrobacterium-mediated co-expression of D BgtE-5754-A1, E BgtE-20069b_T86I-mRFP, F BgtE-20069b_S74T-mRFP, and G BgtE-20069a-mRFP with various Pm3-HA variants. Boxplots show HR intensities from 18 infiltration spots across three experiments (symbols denote experiments). Letters above boxplots denote statistically significant differences at p < 0.05 based on Dunn’s test. Representative HR images are shown above the boxplots. H, I Haustorial index in epidermal cells of wheat cultivar ‘Bobwhite’ transformed via particle bombardment with the reporter pUbi:GUS and the indicated Pm3 constructs. Haustorial index represents the percentage of spores forming haustoria in GUS-positive cells, assessed at 44hpi for H Pm3d-avirulent, Pm3e-virulent isolate CHN_17_40 and for I Pm3e-avirulent, Pm3d-virulent isolate CHE_94202. Black dots represent individual experiments; colored lines indicate the mean across three experiments. Different letters indicate statistically significant differences according to ANOVA followed by Tukey HSD test (p < 0.05). A–G Boxes represent the interquartile range (IQR; 25–75%) with the thick horizontal line indicating the median, and whiskers extending to the most extreme values within 1.5×IQR. Details and exact p-values for all statistical tests shown in A-I) are reported in Supplementary Dataset 4. Pm3d and Pm3e recognition specificity can be combined in chimeric Pm3 NLRs. A,B Virulence phenotypes of isolates CHN_17_40 (A) and CHE_94202 (B) on various wheat lines. “BW” refers to the cultivar ‘Bobwhite’, the background for the Pm3d#1 and Pm3e#2 transgenic lines. “Pm3d#1 / Pm3e#2” denotes F1 plants from a Pm3d#1 × Pm3e#2 cross. Sd#1 and Se#2 refer to transgene-free sister lines of Pm3d#1 and Pm3e#2. Boxplots display infection scores (10-step scale, 0% to 100% leaf coverage) at 11dpi. Eighteen leaf segments across three independent plates were scored (symbols indicate plates). Statistical significance between the Pm3d#1 x Pm3e#2 F1 individuals and the respective parental lines was assessed using the Wilcoxon rank-sum test; *** indicates p < 0.001. C Results of allelic suppression testing in N. benthamiana using the effector variants BgtE-20069b_S74T-mRFP, BgtE-20069a-mRFP (both exclusively recognized by Pm3d-HA) and BgtE-5754-A1 (exclusively recognized by Pm3e-HA). Boxplots show quantified HR intensities from 18 infiltration spots per combination, obtained from three independent experiments (denoted by distinct symbols). Statistical significance was assessed using a paired Wilcoxon signed-rank test; *** indicates p < 0.001. D–F Agrobacterium-mediated co-expression of D BgtE-5754-A1, E BgtE-20069b_T86I-mRFP, F BgtE-20069b_S74T-mRFP, and G BgtE-20069a-mRFP with various Pm3-HA variants. Boxplots show HR intensities from 18 infiltration spots across three experiments (symbols denote experiments). Letters above boxplots denote statistically significant differences at p < 0.05 based on Dunn’s test. Representative HR images are shown above the boxplots. H, I Haustorial index in epidermal cells of wheat cultivar ‘Bobwhite’ transformed via particle bombardment with the reporter pUbi:GUS and the indicated Pm3 constructs. Haustorial index represents the percentage of spores forming haustoria in GUS-positive cells, assessed at 44hpi for H Pm3d-avirulent, Pm3e-virulent isolate CHN_17_40 and for I Pm3e-avirulent, Pm3d-virulent isolate CHE_94202. Black dots represent individual experiments; colored lines indicate the mean across three experiments. Different letters indicate statistically significant differences according to ANOVA followed by Tukey HSD test (p < 0.05). A–G Boxes represent the interquartile range (IQR; 25–75%) with the thick horizontal line indicating the median, and whiskers extending to the most extreme values within 1.5×IQR. Details and exact p-values for all statistical tests shown in A-I) are reported in Supplementary Dataset 4. Instead, we aimed at combining Pm3d and Pm3e specificity in a single Pm3 NLR. As a basis, we used the identified Pm3d/e chimeras Pm3_Ch12 and Pm3_Ch16 that exhibited Pm3d- and Pm3e-like specificity (Fig. 6B, Supplementary Fig. 14). Both chimeras showed an HR response against BgtE-5754 (AvrPm3e_1) which was comparable to Pm3e, suggesting that the chimeras fully incorporate Pm3e activity. Furthermore, both chimeras mounted an immune response towards BgtE-20069b_S74T (AvrPm3d_1 variant) which is exclusively recognized by Pm3d. However, the observed HR response was significantly weaker than for Pm3d, in particular for Pm3_Ch12. In addition, both chimeras failed to recognize the BgtE-20069a (AvrPm3d_2) effector variant (Fig. 6B, Supplementary Fig. 14). This indicates that Pm3_Ch12 and Pm3_Ch16, although combining Pm3d- and Pm3e-like specificities, do not fully incorporate the Pm3d activity (Fig. 6B–D). Previously, Stirnweis and colleagues identified two amino acid polymorphisms in the NB-ARC domain (L456P, Y458H) that enhance HR output of Pm3 allelic variants and occasionally increase their recognition spectrum without causing NLR autoactivity38. To test if the weak or absent HR response of Pm3_Ch12 and Pm3_Ch16 towards BgtE-20069b_S74T and BgtE-20069a could be improved, we introduced the enhancing polymorphisms L456P and Y458H into the NB-ARC domain of Pm3_Ch12 and Pm3_Ch16, resulting in “enhanced” chimeras, enhPm3_Ch12 and enhPm3_Ch16. We subsequently confirmed efficient protein expression and the absence of NLR autoactivity of the engineered NLR constructs (Supplementary Fig. 15,16). When tested for recognition of BgtE-5754 (AvrPm3e_1), both enhanced chimeras mounted an HR response comparable to Pm3e and the original Pm3_Ch12 and Pm3_Ch16 chimeras (Fig. 7D). This suggests that recognition of BgtE-5754 is not drastically affected by the two polymorphisms in the NB-ARC. In contrast, both enhanced chimeras performed significantly better against variants of the RALPH AVR effectors BgtE-20069b and BgtE-20069a (Fig. 7E–G). Importantly, the weak HR response of Pm3_Ch12 and Pm3_Ch16 against BgtE-20069b_S74T was significantly increased in the enhanced chimeras, now matching the HR levels mounted by Pm3d (Fig. 7F). Furthermore, the improved chimeras, like Pm3d, also mounted an HR response towards BgtE-20069a (AvrPm3d_2), which evaded recognition by the original chimeras Pm3_Ch12 and Pm3_Ch16 (Fig. 7G). Next, we tested whether the combined Pm3d- and Pm3e-activity of the enhPm3_Ch12 and enhPm3_Ch16 chimeras can also be observed in wheat under natural AVR expression levels. We used particle bombardment to deliver Pm3 constructs into the epidermis of the susceptible wheat cultivar ‘Bobwhite’ and assessed the haustorium formation index of diagnostic Bgt isolates that distinguish Pm3d and Pm3e resistance, namely Pm3d-avirulent/Pm3e-virulent isolate CHN_17-40 and Pm3d-virulent/Pm3e-avirulent isolate CHE_94202 (Fig. 7A, B). As expected, Pm3d significantly reduced the haustorium formation rate of isolate CHN_17-40, while it was unaffected by Pm3e (Fig. 7H). Congruently, Pm3e expression significantly reduced haustorium formation of isolate CHE_94202 while Pm3d did not (Fig. 7I). Importantly, both enhanced chimeras, enhPm3_Ch12 and enhPm3_Ch16, reduced haustorium formation against both isolates, demonstrating that they are functional in wheat and indeed combine Pm3d and Pm3e resistance activity (Fig. 7H-I). These results show that combining Pm3d- and Pm3e-specific polymorphisms in the LRR domain with NLR-enhancing polymorphisms in the NB-ARC domain allows for the incorporation of the broad-spectrum resistance activities of Pm3d and Pm3e in a single NLR. Discussion Identifying broad-spectrum resistance genes against pathogens remains an important objective of resistance breeding. We identified Pm3d and Pm3e as broadly active resistance alleles against globally occurring Bgt. Our approach, using transgenic lines expressing individual Pm3 alleles, enabled evaluation of resistance without confounding effects arising from non-uniform wheat backgrounds. The broad activity of Pm3d and Pm3e is surprising, given that they are nearly identical to the non-functional Pm3CS. Importantly, this broad activity of Pm3d and Pm3e does not result from constitutive resistance caused by pleiotropic effects due to transgene overexpression, as we identified fully virulent isolates for both alleles. We show that both Pm3d and Pm3e recognize two Bgt AVR genes, which likely contribute to their broad resistance spectrum, as successful resistance evasion by Bgt requires simultaneous modifications in both encoded effectors. Indeed, we found that gain-of-virulence mutations in all three avirulence genes occur in Bgt, but they do not readily co-occur in the same isolates, thereby providing the basis for the broad activity of Pm3d and Pm3e worldwide. Nevertheless, the combination of several virulence mutations through sexual recombination in the Bgt pathogen might occur as the Pm3d and Pm3e resistance genes are increasingly deployed. The ability of Pm3d and Pm3e to recognize multiple AVR genes aligns with growing evidence that NLR/AVR interactions form complex networks that might involve several host and pathogen components39–41. For instance, the NLR pair RGA5/RGA4 recognizes two structurally related MAX AVRs (Avr-Pia and Avr-CO39) from Magnaporthe oryzae42,43. Furthermore, genetic evidence suggests that certain NLRs can also recognize structurally unrelated effectors. For instance, several Mla alleles in barley provide resistance against powdery mildew, wheat blast, and stem rust through recognition of, in part, still unidentified effectors44,45. Given the divergent effector repertoires of these fungi, the recognized effectors are likely to be structurally unrelated. Therefore, our discovery that Pm3e can recognize two structurally unrelated effectors is of particular significance, as it shows that dual recognition specificity against structurally unrelated fungal effectors can be encoded within a single NLR protein. The two AvrPm3e effectors belong to two different, structurally defined effector superfamilies. The RALPH superfamily (AvrPm3e_2) contains many characterized avirulence and virulence effectors from Bgt and Bh18. In contrast, the AvrPm3e_1-like superfamily, first described based on sequence similarity over a decade ago46 remains functionally uncharacterized. This is surprising given its ~13% contribution to the Bgt effectome. An independent study by Seong et al. identified a structural effector superfamily in Bh, termed „cluster 40“, that corresponds to our definition of AvrPm3e_1-like effectors34. Of note, the authors did not find evidence for the existence of this superfamily in other phytopathogenic fungi, therefore supporting our findings that the AvrPm3e_1-like effectors are an innovation of the Blumeria lineage. Recently, 9o9, an AvrPm3e_1-like effector of Bh, was identified as an interaction partner of the barley Bh-susceptibility factor RACB, a small monomeric GTPase47. Reminiscent of our finding for BgtE-5754 (AvrPm3e_1), the authors failed to predict a clear SP for the 9o9 effector and overexpression of the full-length 9o9 effector increased the penetration rate of Bh in barley. Similarly, the full-length BgtE-5754, but not any of the tested N-terminal truncation variants corresponding to putative SP prediction, triggered a Pm3e-mediated HR in N. benthamiana. This suggests that the N-terminal region is indeed part of the mature AvrPm3e_1 protein and might not function as a classical SP. Consequently, the mechanism by which effectors, such as BgtE-5754 or 9o9, exit the fungal cell and reach the host cytoplasm remains unclear48. Given increasing evidence from other phytopathogens for alternative secretion pathways, it is likely that similar mechanisms also exist in Blumeria49,50. The identification of both an avirulence and a virulence factor among the AvrPm3e_1-like effectors suggests that, like RALPH effectors, they represent important components in the effectome of Bgt and Bh. Therefore, it is likely that the AvrPm3e_1-like superfamily contains additional AVRs that have remained undetected, for instance, because previously performed effector screens have focused on the RALPH effectors exclusively11. We found that the specificity of Pm3d and Pm3e towards BgtE-5754 and BgtE-20069a/b effectors is encoded within single amino acid changes compared to the completely inactive Pm3CS. Similarly, in the NLR Sr50, a single substitution was necessary and sufficient to restore recognition towards a non-recognized AvrSr50 variant. Reminiscent of our discovery that all other polymorphisms modulate rather than define the HR response, additional mutation in Sr50 also acted synergistically on HR amplitude51. The fact that specificity towards structurally unrelated effectors can be encoded by single amino acid changes in an otherwise inactive NLR backbone, such as Pm3CS, indicates that recognition of additional structural folds, both from Bgt and beyond, might be engineerable into existing NLR receptors. In addition, it suggests that NLRs are more versatile than previously assumed and that other NLRs might also contain the intrinsic capacity to recognize structurally unrelated effectors. Finally, we engineered chimeric NLRs with dual Pm3d/Pm3e specificity by combining natural polymorphisms, thereby overcoming interallelic suppression between the two Pm3 alleles. These findings highlight the potential of NLR engineering when pyramiding of natural R genes is hampered by antagonistic effects. Furthermore, the successful combination of naturally occurring polymorphisms in the NB-ARC and LRR domain resulting in our enhanced Pm3 chimeras (enhPm3_Ch12 and enhPm3_Ch16), exemplifies how allelic diversity can inform NLR engineering to alter recognition specificity or modulate the HR output. Such approaches, informed by naturally occurring diversity within allelic series, might be particularly useful for the engineering of NLRs where high-resolution structural data is missing, as is the case for the majority of NLR receptors52,53. In addition, the observation that only five or six polymorphisms are sufficient to render the non-functional Pm3CS NLR into the highly functional enhPm3_Ch12 or enhPm3_Ch16 chimeras, that incorporate the broad-spectrum activities of both Pm3d and Pm3e, opens the possibility for direct engineering of these new extended specificities in elite breeding material. This could for instance be achieved via targeted base editing of the naturally occurring Pm3CS allele that is found in many wheat breeding lines25,26,54. Previous studies have reported that NLR engineering efforts based on HR output measured in N. benthamiana did not consistently translate into resistance in the corresponding homologous host55,56. Therefore, our finding that the chimeric receptors are functional in wheat and restrict pathogen penetration is reassuring and renders them promising candidates for stable expression in wheat. In summary, our study demonstrates that allelic series are a valuable source of diversity that can be exploited in disease-resistance breeding. First, allelic series may contain so far undiscovered broadly active alleles that can be directly incorporated into breeding programs. Secondly, polymorphisms within allelic series provide valuable information to guide NLR engineering. For both approaches, highly diverse allelic series rich in unexplored diversity, such as the Pm3 locus, are particularly well suited. Furthermore, the ability of Pm3 NLR to recognize structurally unrelated effector folds, combined with their amenability to NLR engineering, makes them particularly valuable for breeding. We propose that the identification of naturally occurring or the creation of engineered NLRs with multiple recognition specificities represent a promising strategy for resistance breeding. Such NLRs might act similarly as R gene stacks by reducing the likelihood of resistance breakdown, thereby contributing to more durable wheat protection. Methods Plant material All transgenic wheat lines used in this study have been previously generated and express individual Pm3 alleles under a maize ubiquitin promoter in the susceptible wheat cultivar ‘Bobwhite SH 98 26’. These transgenic lines (alongside respective sister lines) were selected as best performing lines exhibiting high transgene expression and race-specific Bgt resistance in the absence of pleiotropic effects28–31,36. Specifically, the transgenic lines Pm3a#1, Pm3b#1, Pm3d#1, Pm3e#2, Pm3g#1 and Pm3CS#19 were used in this study. For Pm3d#1 and Pm3e#2 the corresponding sister lines Sd#1 and Se#2 were included as controls. To generate Pm3d#1 x Pm3e#2 F1 hybrids, homozygous Pm3d#1 and Pm3e#2 parental lines were crossed and the resulting F1 progeny were phenotypically assessed for Pm3d- and Pm3e-specific resistance. Fungal isolates and virulence phenotyping All Bgt isolates used in this study represent twice single spore isolated, genetically uniform cultures that have been described in previous studies15,16,27,32. Isolates were maintained on detached leaves of the susceptible wheat cultivar ‘Kanzler’ placed on 0.5% agar plates (PanReac AppliChem), supplied with 4.23 mM benzimidazole57. Virulence phenotypes were assessed on detached leaf segments cut from primary leaves of 10 day old seedlings that were placed on 0.5% agar plates before infection (0.5% food grade agar (PanReac AppliChem), 4.23 mM benzimidazole) and incubated at 20 °C57. Virulence scoring was performed 8-11 days post infection, depending on the growth speed of individual Bgt isolates on the susceptible wheat cultivar ‘Kanzler’, serving as an infection control. Virulence phenotypes reported or depicted in Fig. 1b, Fig. 2, Fig. 5a, b, g were determined on three to six biological replicates based on the observed mildew leaf coverage (MLC) and classified in five qualitative categories: 1 (100% MLC), 0.75 (75% MLC), 0.5 (50% MLC), 0.25 (25% MLC) and 0 (0% MLC). Phenotyping results are summarized in Supplementary Dataset 1. To assess the more quantitative interallelic suppression phenotype in the Pm3d#1 x Pm3e#2 F1 hybrids, 18 biological replicates were assessed and individually scored based on observed MLC in 11 categories ranging from 0 to 100%. Representative images were selected to depict virulence phenotypes throughout the manuscript. Avirulence depletion assay Avirulence depletion mapping was performed as previously described16. In brief, the AD assay relies on the generation of a haploid F1 progeny population by crossing two Bgt isolates with opposite virulence phenotypes on the R gene of interest, followed by selection of the F1 population on a resistant host cultivar, thereby resulting in depletion of F1 progenies carrying the parental avirulent genotypes. AVR-depleted F1 populations are then bulk-sequenced and compared with a non-selected control bulk to identify genomic regions with strong deviations from 1:1 segregation of parental markers, indicating the presence of an AVR factor. Sexual crosses were performed between the isolates CHN_52-27 (MAT1-1) x CHN_17-40 (MAT1-2), IRN_GOR_2 (MAT1-1) x CHN_17-40 (MAT1-2) and CHE_94202 (MAT1-1) x USA_2 (MAT1-2) by co-infection of the susceptible wheat cultivar ‘Kanzler’ in a greenhouse chamber58. The cross between CHN_52-27 (MAT1-1) and CHVD_042201 (MAT1-2) was performed for a previous study following the same protocol16. Arising chasmothecia were harvested, dried and ascospores ejected as previously described16. Arising F1 progeny were then subjected to either selection on Pm3e#2 or Pm3d#1 transgenic lines (Pm3-selected bulks) or grown on ‘Kanzler’ (non-selected bulk). Conidiospores of individual bulks were harvested 10 days post infection and fungal DNA was extracted using a previously described CTAB/phenol chloroform extraction protocol9. Bulk DNA was sequenced on a NovaSeq X Plus platform with our commercial partner Novogene (https://www.novogene.com) at a coverage of approximately 120x. To perform the AD bioinformatics pipeline, Illumina reads of the parental isolates of each cross were aligned to the reference genome of the respective avirulent parent of each cross and high-quality markers defined. The respective genome assemblies have been described previously59. Mapping of Illumina data, SNP calling and high-quality marker definition were performed as described previously16 using bowtie2 v2.4.4. and FreeBayes v1.3.6. The corresponding Python script is available from: https://github.com/MarionCMueller/AD-assay/blob/main/scripts/Script_BSA_SNPcall.v8.py. Following marker definition, sequencing data of Pm3-selected and non-selected bulks were mapped to the respective avirulent parental genome assembly, followed by the determination of parental genotype ratios. To test for deviations from the expected 1:1 ratio indicating avirulence depletion, a G-test of goodness-of-fit was performed using the R package RVAideMemoire and the G.test() command as described in ref. 16. The full code is available under https://github.com/MarionCMueller/AD-assay/blob/main/scripts/BSA_Blumeria_functions.R. To define genomic intervals of AVR loci, the percentage of alternative alleles over five consecutive polymorphic sites were averaged and the largest interval that surpassed 90%, 95%, and 98% of reads, respectively, was extracted. For each cross, the interval with the highest selection was retained. To define the overlapping interval between all crosses, 400 bp of the flanking regions of each interval were extracted and subsequently aligned against the reference genome Bgt_genome_v3_1633 using blastn v2.12.060. AvrPm3 haplovariant mining To assess haplovariant diversity of individual AvrPm3 effectors, Illumina resequencing datasets were mapped to the Bgt reference assembly of CHE_96224 as previously described27. For each isolate, mapped reads were visualized using integrative genomics viewer IGV (v2.8.6)61 and haplovariants defined manually. For isolates with ambiguous results due to gene duplications or sequence insertions (i.e., 9 bp insertion in AvrPm3e_1), determined haplovariants were verified by PCR amplification and subsequent in-house Sanger sequencing, using the primers listed in Supplementary Table 4. Gene cloning To produce Gateway-compatible entry clones of individual effectors and effector variants, their coding sequence was codon-optimized for N. benthamiana with the codon-optimization tool of Integrated DNA technologies (https://eu.idtdna.com/pages/tools/codon-optimization-tool) and subsequently produced with flanking attL1/attL2 sites by gene-synthesis (commercial partners: BioCat GmbH, Thermo Fisher Scientific, Integrated DNA technologies), PCR-based site-directed mutagenesis (single polymorphisms) and/or in-Fusion cloning (mRFP fusions). BgtE-20069a and BgtE-20069b variants were cloned without their predicted SP (SignalP4.0)62, whereas BgtE-5754 variants were cloned with their full-length coding sequence unless indicated otherwise. The DNA sequences of all effector constructs used in this study are compiled in Supplementary Dataset 2. Pm3CS-HA, Pm3d-HA, Pm3e-HA and Pm3g-HA genomic DNA sequences11 were cloned into the Gateway compatible pENTR plasmid using in-Fusion cloning (Takara Bio). Subsequently, PCR based site-directed mutagenesis with primers listed in Supplementary Table 4 was used to introduce individual amino acid polymorphisms resulting in the chimeric constructs Pm3_Ch1-16, enhPm3_Ch12 and enhPm3_Ch16. All entry clones described above were subsequently mobilized into the binary vector pIPKb00463 using Gateway LR clonase II (Invitrogen). Resulting expression constructs were sequence verified by in-house Sanger sequencing (effectors) or full plasmid sequencing (R genes) with our commercial partner (Microsynth AG) and transformed into Agrobacterium tumefaciens GV3101 using freeze-thaw transformation64. For bombardments of wheat epidermal cells, entry clones containing Pm3CS-HA, Pm3d-HA, Pm3e-HA, enhPm3_Ch12-HA and enhPm3_Ch16-HA (see above) were mobilized into a Gateway-compatible pGY1 expression vector65 using Gateway LR Clonase II (Invitrogen). The expression constructs for Pm3a-HA, Pm3b-HA, Pm3c-HA and Pm3f-HA in the binary vector pIPKb004 have been previously described9,11. The RNA silencing inhibitor p19 has been previously described66. AVR validation and HR quantification Agrobacterium-mediated expression in N. benthamiana was achieved as previously described11. For AVR-R co-expression, Agrobacteria at an OD600 of 1.2, carrying effector constructs, R gene constructs or the silencing inhibitor p19 were mixed in a 4:1:1 ratio prior to infiltration. The HR response was imaged and quantified at 4 days post-infiltration (unless indicated otherwise) using a conventional camera and/or a Fusion FX imaging System (Vilber Lourmat, Eberhardzell, Germany) and subsequently quantified using Fiji11,67. All co-expression experiments were performed three times using n = 6 biological replicates per assay, resulting in a total of n = 18 HR measurements for quantification and statistical analysis. Representative images were chosen to depict the HR phenotypes throughout the manuscript. Western blotting Effective production of effector or NLR proteins in N. benthamiana was tested using the Agrobacterium infiltration protocol described above. Single constructs were mixed with p19 silencing inhibitor in 4:1 ratio prior to infiltration. Infiltrated plant tissue was harvested at 2 days post infiltration and proteins were extracted in from 4 ×2 leaf disks (0.5 cm diameter) in 150 μl 2x Lämmli buffer (62 mM Tris–HCl pH 6.8, 10% glycerol, 2% SDS, 0.02% bromophenol blue, 1% β-mercaptoethanol) and boiled for 5 min at 95 °C followed by brief centrifugation (14,000 x g, 10 min) to remove cell debris. Protein extracts were separated on either 8% (NLR detection) or 4–20% gradient (effector detection) SDS polyacrylamide gels, respectively and subsequently blotted to a nitrocellulose membrane (Amersham Protran 0.2 μm NC). Detection of HA-tagged Pm3 proteins was achieved by using 1:3,000 diluted anti-HA-HRP antibody (rat monoclonal, clone 3F10, Roche). Detection of FLAG- or 3xFLAG-tagged effectors was achieved by using 1:3,000 diluted anti-FLAG-HRP antibody (mouse monoclonal, clone M2, Sigma Aldrich). HRP chemiluminescence was induced with WesternBright ECL HRP substrate (Advansta) and imaged on a Fusion FX imaging System (Vilber Lourmat, Eberhardzell, Germany). Bombardment of wheat epidermal cells The pGY1 plasmids carrying Pm3CS-HA, Pm3d-HA, Pm3e-HA, enhPm3_Chr12-HA and enhPm3_Chr16-HA (see above) were extracted using the NucleoBond Xtra Midi Kit (Macherey-Nagel) according to manufacturer's instructions. Particle bombardments were performed as described in ref. 24. To do so, 1.25 µg of Pm3 plasmids DNA as well as 1.25 µg pUbi-GUS plasmid DNA were attached to gold particles (11 μl of 25 mg/ml gold particles with 1 μm diameter, Bio-Rad, Hercules, USA) and shot onto 9-day-old first leaves of the susceptible wheat cultivar ‘Bobwhite’ using the PDS-1000/HETM (Bio-Rad, Hercules, USA) particle gun. Per experimental replication, two times five leaf segments were bombarded for each construct and isolates combination. One hour post bombardment leaf segments were infected with conidiospores of Bgt isolates CHN_17-40 or CHE_94202. Leaf segments were vacuum infiltrated in GUS-staining solution (0.1 M Na₂HPO₄/NaH₂PO₄ pH 7.0, 0.01 M EDTA, 0.005 M potassium hexacyanoferrate (II), 0.005 M potassium hexacyanoferrate (III), 0.1% (v/v) Triton X-100, 20% (v/v) methanol, 0.5 mg/mL 5-bromo-4-chloro-3-indoxyl-β-D-glucuronic acid) 44 h post inoculation and subsequently incubated at 37 °C for 3–5 h before destaining in 70% ethanol. For microscopic evaluation, leaves were submerged in Calcofluor White solution47. Haustorium formation rates were determined in GUS-positive cells attacked by a single Bgt spore. For each construct, at least 25 interactions were counted. The experiment was performed three times. Bioinformatics analysis Pfam domains of full-length BgtE-5754 were predicted using the InterPro webserver (https://www.ebi.ac.uk/interpro/)68. SPs for protein sequences of BgtE-5754 and BgtE-20069 were predicted using SignalP 4.1, 5.0, 6.0 and DeepTMHMM webservers with default parameters62,69–71. To define effectors with structural similarity to AvrPm3e_1 in Bgt, the structure of the 844 candidate Bgt effectors33 was predicted using AlphaFold272. Subsequently, the best-scoring model for each effector was compared pairwise for structural similarity using TM-align (v20220412)73. In a first step, hits with a pairwise TM-align score > 0.6 when compared to BgtE-5754 were considered as members of the AvrPm3e_1-like superfamily. To identify additional effectors with structural similarity, iterative searches against all previously defined AvrPm3e_1-like family members were performed (TM-align score > 0.6) until saturation. The resulting AvrPm3e_1-like superfamily contained 112 members belonging to the previously defined effector families E001 (including BgtE-5754), E027, E040 and E061 (nomenclature of ref. 33). The same procedure was applied using the protein structural model of BgtE-20069b to identify RALPH effectors. To search for AvrPm3e_1-like effectors in Bh, the 112 AvrPm3e_1-like proteins in Bgt were compared with previously predicted AlphaFold structures of the Bh effectome34 using TM-align (score >0.6). The Foldseek server74 was used to search for AvrPm3e_1-like effectors in powdery mildew fungi outside the Blumeria genus. To do so, BgtE-5754 and randomly selected five additional members of family E001 and two members of family E027, E040, E061 were searched against the AlphaFold/UniProt50_v4 database with search results restricted to Erysiphales. The resulting hits, irrespective of significance score, were retrieved and compared in pairwise similarity searches against the 112 AvrPm3e_1-like effectors in Bgt using TM-align (score >0.6). Accession numbers of analyzed protein structures can be found in Supplementary Dataset 3. SP prediction was performed for all proteins of Bgt reference isolate CHE_96224 (Bgt_CDS_v4_23, available at: 10.5281/zenodo.701850175) using SignalP 6.070 in the mode --organism eukaryote --mode slow-sequential with default thresholds for SP definition. Expression analysis of CHE_96224 effectors was performed using publicly available RNA-seq data from isolate CHE_96224 at two days post inoculation on the susceptible wheat cultivar “Chinese Spring” (SRA accession numbers: SRX7003606- SRX7003608). The coding sequence of CHE_96224 annotation Bgt_CDS_v4_23 (see above) was indexed using the Salmon v1.4.076 index command, using the chromosomes as decoys and with the option –keepDuplicates. Subsequently, the transcripts were quantified using salmon quant with the -l A option and RPKM values calculated with the edgeR package v3.40.277 using the command rpkm(). Finally, the average of the three biological replicates was calculated. Structural prediction of AVR proteins was performed using the AlphaFold3 server (https://alphafoldserver.com/)78. For BgtE-5754, the full-length protein sequence was used, whereas structural prediction of BgtE-20069b (AvrPm3d_1), BgtE-20002 (AvrPm3b/AvrPm3c), and Bgt_avrF2_7 (AvrPm3a/AvrPm3f) was performed on their respective mature proteins (SignalP 6.0)70. Predictions for all proteins were of sufficient quality (pTM score >0.8), except for the BgtE-20002 protein that had a pTM score of 0.33. Inspection of the multiple sequence alignment (MSA) for BgtE-20002 underlying the prediction revealed a low number of proteins in the MSA, likely due to the UniProt database being outdated for Bgt. High-quality prediction of the BgtE-20002 structure (pTM score >0.8 was achieved by using the ColabFold_AlphaFold2 pipeline79 with a custom MSA generated with the HHblits Toolkit80,81, containing all high-quality protein sequences of the BgtE-20002 effector family E018. All resulting structural predictions were visualized with ChimeraX v1.882. Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Supplementary information Supplementary Information Descriptions of Additional Supplementary Files Supplementary Dataset 1 Supplementary Dataset 2 Supplementary Dataset 3 Supplementary Dataset 4 Reporting Summary Transparent Peer Review file Supplementary Information Descriptions of Additional Supplementary Files Supplementary Dataset 1 Supplementary Dataset 2 Supplementary Dataset 3 Supplementary Dataset 4 Reporting Summary Transparent Peer Review file Source data Source Data Source Data