Evolutionarily conserved partial gene duplication in the Triticeae tribe of grasses confers pathogen resistance

Origin and evolution of HvARM1

Transient single-cell silencing of Rnr5 significantly reduced NHR of barley to Bgt [39]. BlastX analysis revealed the homology of Rnr5 to PUBs. PUBs contain an N-terminal U-box domain and a C-terminal armadillo-(ARM) repeat domain [40, 41]. Although Rnr5 was most closely related to OsPUB15 in rice [42], it does not appear to be the barley ortholog because the encoded protein of 442 amino acids is considerably shorter than a regular PUB and contains only the C-terminal ARM-repeat region as a conserved domain (Fig. 1a). The barley genome also contains a gene for a full-length PUB protein of 831 amino acids with highest similarity to OsPUB15 that was therefore named HvPUB15, whereas Rnr5 was designated as HvARM1 (Table 1). Protein similarity between HvPUB15 and HvARM1 starts at positions L398 and L9 of HvPUB15 and HvARM1, respectively, between the conserved U-box and ARM-repeat regions of HvPUB15 (Additional file 1: Figures S2 and S3a). Sequence similarity between the two genes extends upstream from the HvARM1 initiating codon spanning the first intron of HvARM1, which corresponds to the exon 3 sequence of HvPUB15, until it abruptly ends within the U-box sequence of HvPUB15 (Fig. 1a and b). A further upstream sequence inside HvARM1 intron 1 as well as the untranslated exon1 sequence did not exhibit significant similarity to any annotated gene or repetitive DNA element in the barley genome. These results strongly suggest that HvARM1 originated as a partial gene duplicate of HvPUB15. An 8-bp deletion downstream from the first five N-terminal amino acids of HvARM1 restored the initial reading frame (Fig. 1c) because its initiating ATG corresponds to an out-of-frame codon of HvPUB15 (Fig. 1d). Whole genome shotgun (WGS) sequences of five additional species of the Triticeae tribe of grasses, wheat (Triticum aestivum) and its potential wild progenitors Aegilops speltoides, Ae. tauschii, and T. urartu plus rye (Secale cereale), also revealed the presence of HvARM1-like genes, suggesting a monophyletic origin of the partial gene-duplication event in a common Triticeae ancestor dating back at least 15 million (M) years [43] (Fig. 1e). To address the question of whether different evolutionary constraints act on PUB15 and ARM1 genes, we searched both orthologous gene groups for footprints of purifying or diversifying selection by calculating d_N/d_S ratios (ω) at the codon level in a phylogenetic context. Protein sequence conservation among the seven species was high in both the U-box-containing N-terminal and the ARM-repeat-containing C-terminal parts of PUB15 (Additional file 1: Figure S3b), the existing polymorphisms being in agreement with phylogenetic species distances. By contrast, sequence conservation was reduced among ARM1 proteins (Additional file 1: Figure S3c), most clearly evident when comparing the two wild wheat species. As shown in Fig. 1f and Additional file 1: Figure S3d, both genes are subjected to purifying selection at the very N-terminus of ARM1 and within the ARM-repeat region. Selection was neutral in ARM1 outside these regions, whereas PUB15 sequences remained under purifying selection along the entire ARM1-overlapping part of the gene. This suggests that the function of ARM1 proteins is restricted to the binding of one or a few protein ligand(s) via their ARM repeats, whereas structural constraints on full-length E3 ligases that have to bind to substrate proteins and mediate the interaction with the highly conserved UBC domain of E2s are probably more stringent.

Identifier	HvPUB15	HvARM1
cDNA clone ID	HO23D08	HO14H18
HarvEST assembly #35 unigene Nr.	3072	3071
Full-coding sequence cDNA Acc. Nr.	AK361754	AK371875
Morex WGS contig Acc. Nr.a	CAJW010005672	CAJX010121345
Barke WGS contig Acc. Nr.a	CAJV010187631	CAJV010188692
Bowman WGS contig Acc. Nr.a	CAJX010851782	CAJX010121345
High-confidence barley gene IDa	HORVU3Hr1G113910	HORVU3Hr1G081380
Chromosomeb	3HL	3HL
Position (Mbp)b	689.57	594.73
Syntenic to Brachypodium distachyon, Oryza sativa, Sorghum bicolor	No	No
Syntenic to Ae. tauschii c	Yes	No

^aMost significant BlastN result with 99–100% identity to genomic sequence of barley (http://webblast.ipk-gatersleben.de/barley/)

^bBased on high-confidence (HC)-gene mapping of the barley reference sequence [72]

^cBased on the Ae. tauschii genome https://doi.org/10.1073/pnas.1219082110

Allelic variants of HvARM1

The phylogenetic and functional (see below) data of ARM1 suggest that the gene is under selection for maintaining a quantitative level of resistance among Triticeae species to powdery mildew infection. We therefore analyzed gene variants (alleles) in a diverse collection of barley genotypes (Additional file 2: Tables S1–S3) for significant association with the severity of powdery mildew infection. Table 2 shows significantly associated single-nucleotide polymorphisms (SNPs) as well as gene-haplotype polymorphisms in two diverse, worldwide collections of barley landraces and cultivars. No association of HvPUB15 gene variants with the same trait was found in these populations. This result supports the view that HvARM1 — despite its partial nature — represents a functional gene protecting barley from powdery mildew attack, whereas the cellular functions of HvPUB15 may be more complex.

Function of HvARM1 during powdery mildew attack

To validate the transient-induced gene silencing (TIGS) effect of HvARM1 in the nonhost interaction with Bgt [39] and to further assess its role in the interaction with the adapted Bgh, we generated transgenic plants with silenced HvARM1. In rice, a detrimental effect of the knock-out mutation of OsPUB15 was described that included severe growth retardation and seedling lethality [44]. A similar phenotype was observed in approximately 25% of transgenic barley events at the T1 generation (Additional file 1: Figure S4a). These individuals died after a few weeks, suggesting lethality caused by the homozygous transgene that may result in stronger silencing, in line with the failure to identify homozygous T2 or T3 lines. Homozygous lethality could reflect off-target silencing of the potentially housekeeping HvPUB15 gene or an additional, more basal role of HvARM1 in barley vegetative growth and development. The possible off-targeting of HvPUB15 by the RNAi construct was addressed by software-based prediction and by off-target transcript quantification (Additional file 1: Figure S4b and Fig. 2a). Using default settings including end-stability difference and target-site accessibility thresholds, and the HarvEST:Barley assembly #35 sequence dataset [45], the software si-Fi21 (https://doi.org/10.5447/ipk/2017/9) predicted 33 and 6 efficient 21-nt small interfering RNAs (siRNAs) for HvARM1 and HvPUB15, respectively. However, the T3 transgenic lines consistently exhibited silencing of HvARM1, whereas no reduction of HvPUB15 mRNA levels was found, suggesting that the transgene expression levels of the selected lines were not sufficient to silence the off-target.

Because HvARM1 was discovered in a TIGS screen for attenuated NHR [39], we first tested T3 progeny of three selected events for susceptibility to Bgt (Additional file 1: Figure S4c). Although there was a considerable variability between individuals per line, two lines exhibited higher susceptibility to the non-adapted fungus as compared to the control group of azygous segregant plants. In general, azygous plants are considered as better controls since they have undergone the same transformation procedures and lost the transgenic construct by segregation. Indeed, the transformation procedure had an impact on the Bgt interaction because the azygous control group was on average more susceptible than the Golden Promise wild type. Figure 2b additionally shows that the three selected events were also more susceptible to Bgh, compared to a population of control plants consisting of azygous segregants plus progeny from three azygous individuals identified in the T2 generation.

Bombardment with gene-specific RNAi and with over-expression (OEX) constructs for a direct comparison of altered HvARM1 versus HvPUB15 expression levels revealed that gene-specific HvARM1 silencing increased the relative susceptibility index (SI) to Bgh (Table 3), in line with the enhanced susceptibility observed in stable transgenic barley T3 plants. On the other hand, we found no significant effects of altering HvPUB15 mRNA levels on the interaction of transformed cells with Bgh, again indicating more complex, homoeostatic rather than defense-related functions of the encoded protein. Following powdery mildew inoculation, endogenous transcript levels of HvARM1 in peeled leaf epidermis were more strongly up-regulated above a basal level of expression compared to HvPUB15 (Additional file 1: Figure S5), which also suggests a defense-related role of HvARM1.

Localization and protein interactions of HvPUB15 and HvARM1

Fusion proteins of HvARM1 and HvPUB15 with yellow fluorescent protein (YFP) showed a similar fluorescence pattern as non-fused YFP, suggesting nucleo-cytoplasmic localization (Additional file 1: Figure S6, panels a–c), in agreement with the localization of the proposed rice ortholog OsPUB15 [44]. Because the presence of the conserved ARM protein-protein interaction domain in HvARM1 suggests binding to other barley protein(s), we carried out a yeast two-hybrid screening in a prey library from Bgh-attacked barley leaves using HvARM1 as bait. This led to the identification of six barley proteins that interacted strongly and reproducibly with HvARM1 (Fig. 3a and Additional file 1: Table S8). Out of these six candidates the homologs of Clp-protease adaptor protein ClpS1 and Thylakoid formation 1 protein THF1 of Arabidopsis thaliana [46, 47] also strongly interacted with HvPUB15 (Fig. 3b) and may therefore be ubiquitination substrates of the E3 ligase. By using an in vitro ubiquitination assay we could show that HvPUB15 has ubiquitin ligase activity (Additional file 1: Figure S7). Because HvPUB15 catalyzed the polymerization of ubiquitin chains rather than auto-ubiquitination, it might be an E4 rather than an E3 ligase [48]. The possibility that HvThf1 and HvClpS1 are ubiquitination substrates for HvPUB15 was tested by transient OEX of the HvPUB15 gene together with either HvThf1:YFP or HvClpS1:YFP, followed by the quantification of YFP-fluorescing cells 24 h after the bombardment. As shown in Table 4, co-expression with HvPUB15 significantly reduced the number of HvThf1:YFP-fluorescing cells, suggesting that HvThf1 is an in vivo substrate to HvPUB15. By contrast, no indication of HvPUB15-mediated degradation of HvClpS1 was found. The possible involvement of these two HvARM1-interacting proteins in the interaction with Bgh was tested by TIGS and transient OEX. We observed a trend for reduced susceptibility by silencing and significantly enhanced susceptibility by OEX of HvThf1 (Table 3), which may indicate that the proposed HvPUB15 substrate protein functions as a host susceptibility factor. Transient OEX of HvClpS1 also enhanced susceptibility to Bgh, although there was no indication of the opposite effect by HvClpS1 silencing. Co-localization experiments of HvThf1-YFP and HvClpS1-YFP C-terminal fusion proteins with the plastid marker Rubisco small subunit [49] confirmed their expected plastid localization (Additional file 1: Figure S6, panels d–k).

In vivo interaction of HvARM1, HvPUB15, and HvPUB15^ARM (the ARM domain of HvPUB15) with HvThf1 and HvClpS1 was assessed by split-YFP bimolecular functional complementation (BiFC) assays in Agrobacterium-infiltrated Nicotiana benthamiana leaves and by co-immunoprecipitation in A. thaliana protoplasts. Figure 4 shows that the transient co-expression of HvPUB15 or HvPUB15^ARM with either HvClpS1 or HvThf1 gave rise to BiFC (YFP) signals primarily in epidermal cells. The localization patterns of the fluorescence signals indicated that the proteins interacted in the cytoplasm, which was confirmed by the absence of co-localization with the plastid marker protein 35S:SSU_1–79–mCherry. The BiFC signals were abolished or strongly reduced by using the U-box mutant HvPUB15^P245A as an interaction partner, suggesting specificity of the interaction. HvARM1 also interacted with HvClpS1 and HvThf1. Moreover, interactions were observed between HvPUB15 and HvARM1, and here the HvPUB15 U-box mutation increased BiFC signals instead of reducing them. The specificity of the interaction was further confirmed using additional controls and quantitative fluorescence measurement (Additional file 1: Figures S8 and S9). Co-immunoprecipitation experiments in A. thaliana protoplasts of cMyc-tagged HvARM1 and HvPUB15^ARM together with either HvThf1 or HvClpS1 confirmed in vivo interaction of HvPUB15^ARM with HvThf1, HvARM1 with HvThf1, and HvARM1 with HvClpS1 (Fig. 5). Taken together, the results suggested in vivo interaction of HvPUB15 with HvClpS1 and HvThf1, whereby the presence of an intact U-box was required for the interaction. In addition, HvARM1 as well as the ARM -domain of HvPUB15 interacted with HvClpS1 and HvThf1. One of the HvPUB15-interacting proteins, HvThf1, was identified as a potential ubiquitination substrate and as a host susceptibility factor to Bgh.

Genome-wide search for expressed partial gene duplicates

Does the partially duplicated gene pair of PUB15 and ARM1 represent a unique case in Triticeae genome evolution, or could we find indications for additional partial gene duplicates with putative functions? To address this question we conducted a genome-wide search for pairs of full-length complementary DNA (cDNA) sequences with high sequence similarity but clearly different lengths of their longest open reading frames. The rationale behind this approach was to select for partially duplicated or rearranged genes that are expressed and therefore more likely to fulfill a biological role. Starting with a library of 23,614 full-length cDNA sequences (http://barleyflc.dna.affrc.go.jp/bexdb/), we found 1154 matching cDNA pairs with a sequence identity of 80–99%. A subsequent tBlastx analysis of these pairs revealed 205 pairs with a length difference of matching open reading frames of > 25%. After further filtering steps to exclude non-spliced transcripts and chimeric as well as partial clones, we identified eight expressed pairs of putative, partially duplicated genes including HvPUB15/HvARM1 (Additional file 2: Tables S4 and S5 and Additional file 1: Figure S13). A majority of these are localized at non-tandem positions in the barley genome (five or more gene models apart from each other, or on different chromosomes). Three genes encode proteins from families known to be involved in plant-pathogen interactions: PUB15 plus two receptor-like kinases. Besides the pair of ARM1 and PUB15 we found two more gene pairs giving rise to full-length and truncated proteins, respectively, which are conserved across Triticeae species (Fig. 6a). The truncated open reading frames of one of these pairs encoding a receptor-like kinase are caused by a 19-bp deletion in barley reconstituting a stop codon, or by stop codon mutations approximately 200 bp further downstream in T. aestivum and S. cereale, which mark the insertion of a non-coding genomic sequence. The truncated open reading frames of the second pairs are caused by the insertion of approximately 2 kb of a non-coding sequence. Finally, although this was not a criterion for their selection, several transcripts encoding truncated proteins were up-regulated by powdery mildew attack (Fig. 6b and Additional file 2: Table S6). Therefore, these rearranged or partial gene duplicates in addition to HvARM1 might have evolved to fulfill new functions in Triticeae species.

Plant and fungal material

TIGS and transient OEX experiments were done in 7-day-old seedlings of spring barley Golden Promise, except for OEX of site-directed mutagenesis (SDM) and TIGS of HvThf1 and HvClpS1 where the closely related genotype Maythorpe was used. Stable transgenic barley plants of cv. Golden Promise were generated as described [73]. Bombarded leaf segments or transgenic plants were inoculated with Swiss Bgt field isolate FAL 92315, or Swiss Bgh field isolate CH4.8 throughout the study.

Sequence alignment, phylogenetic analysis, and estimation of pressures of selection

PUB15 and ARM1 orthologs were obtained from five Triticeae species, besides barley, including Secale cereale [21, 74], Triticum aestivum [75], and its three potential wild progenitors T. urartu (A genome), Aegilops speltoides (B genome), and Ae. tauschii (D genome). The sequences were retrieved by BlastN search of the H. vulgare sequences (AK361754 and AK371875, respectively) against the whole genome assembly databases (http://webblast.ipk-gatersleben.de/ryeselect/ for S. cereale and https://urgi.versailles.inra.fr/blast/blast.php for T. aestivum and its relatives). The rice ortholog, OsPUB15 (XM_015795011), was downloaded as outgroup sequence. All 17 coding sequences could be fully acquired but for the Ae. speltoides ARM1 sequence lacking the last exon.

The sequences were aligned with MAFFT v7.308 [76] using the default settings within Geneious 10.0.9 (https://www.geneious.com) [77] followed by manual adjustment. The model of sequence evolution was determined with jModeltest 2.1.10 [78]. The best-fit model, identified with the Akaike information criterion AIC [79, 80], was the general time-reversible (GTR) [81] with rates variation according to a gamma distribution [82]. The maximum likelihood (ML) phylogenetic tree was calculated with RAxML v8.2.7 [83] using the GTRGAMMA model and 100 bootstrap replicates (options –f a and –x).

Pressures of selection were investigated at the codon level throughout the phylogeny for both proteins separately. The pressure of selection can be estimated by ω (non-synonymous substitution rate divided by synonymous substitution rate, d_N/d_S). An ω < 1 suggests that the site is under negative, or purifying, selection, while an ω > 1 indicates that positive selection is occurring, and ω = 1 for neutral changes. Values of ω were estimated for each non-ambiguous codon using the codeml program within the package PAML4.9 h [84]. Following Jeffares et al. [85], codon site models M0 [86, 87], allowing only one class of ω, M1a [88, 89], allowing two categories of sites (0 < ω₀ < 1 and ω₁ = 1 with proportions p₀ and p₁ = 1- p₀), and M2a [88, 89], which includes a proportion of sites under positive selection, were tested. Likelihood ratio tests (LRTs) were performed to compare models M0 and M1a, as a test for variation of ω among sites, and models M1a and M2a, as a test for positive selection, against a χ² distribution (with one and two degrees of freedom, respectively). Both proteins showed variations of ω among sites with M1a versus M0 LRTs significant. But with nearly identical likelihoods for models M1a and M2a, PUB15 and ARM1, though to a lesser extent, proved to be under purifying selection with a small portion of sites under neutral selection. Sequence information is available at https://doi.org/10.6084/m9.figshare.c.4092686.v1.

Exome capture sequencing

Genomic DNA was extracted from barley leaf material from a single plant for each accession and used for the hybridization with the barley SeqCap Ez oligo pool (Design Name: 120426_Barley_BEC_D04, [54]. Quality-trimmed reads were mapped to the reference genome (http://webblast.ipk-gatersleben.de/barley_ibsc/downloads/) with Burrows-Wheeler Aligner (BWA) v0.7.5a using the mem algorithm with default parameters [90] and retaining only properly paired reads. Variant calling and realignment around indels were performed with Genome Analysis Toolkit (GATK), version 2.7.4 (https://software.broadinstitute.org/gatk/). Variant calls were filtered for high quality and ≥ 80% of samples being represented at each locus, and a dataset of 449,585 SNPs was produced, suitable for genetic association analysis of the two genes under investigation (full information about genome-wide variants from this dataset will be published elsewhere).

Association genetic analysis

Association of SNP and gene haplotypes (marker) of HvARM1 and HvPUB15 with the severity of Bgh infection (trait) was calculated based on genetic and phenotypic data of two diverse collections of cultivated barley (H. vulgare ssp. vulgare). Bgh infection values were determined in a detached leaf assay using second leaves of approximately 12-day-old seedlings, as described [91]. First, a worldwide collection of 76 landraces (WHEALBI_LRC) was inoculated either with isolate JKI-75 or JKI-242, which exhibit a complex and complementing virulence spectra [92]. Second, a worldwide collection of 127 cultivars (WHEALBI_CULT) was inoculated with the same two Bgh isolates. Both populations consisted of single seed-derived lines, and an average of 5 parallel plants per line was used in each inoculation assay. For passport data of all lines see Additional file 2: Table S1. Seven days after inoculation, disease was scored by estimating the percentage of leaf area covered by fungal mycelium. Because disease scores were variable between different inoculation experiments, they were normalized to internal standards cv. Roland or Morex, as indicated. Phenotypic data of all isolate-genotype combinations are based on two independent inoculation series. SNP calls were derived from exome capture resequencing, and haplotypes were calculated based on the combination of SNP calls per gene. SNP-trait and haplotype-trait associations were calculated in TASSEL v4.1 using a mixed linear model with kinship as random effect. Marker data for kinship calculations were derived from 4032 polymorphic genotyping by sequencing (GBS) markers. Marker-trait associations were assumed significant if the Holm’s-corrected p value was < 0.05 (number or SNP or haplotypes/gene = number of tests).

TIGS and transient over-expression

TIGS constructs were generated and transferred by particle bombardment into leaf epidermal cells of 7-day-old barley seedlings as described [93]. Leaf segments were inoculated 3 days after the bombardment with Bgh at a density of 140–180 conidia mm^–2. Transformed GUS-stained epidermal cells as well as haustoria-containing transformed (susceptible) cells were counted 48 h after inoculation, and TIGS effects on the susceptibility index (SI) were statistically analyzed [91].

For verification of transgene effects, HvARM1 was excised from a subclone of the bacterial artificial chromosome (BAC) HVVMRXALLhA0581d24 (Acc. Nr. KM979563) as a StuI/SphI fragment and inserted into SmaI/SphI sites of pIPKTA09 [94]. For transient OEX of candidate genes, full-coding sequences were polymerase chain reaction (PCR) amplified from cDNA and inserted as an XbaI fragment into the multiple cloning site of pIPKTA09. Mutations (HvARM1^-L286H and HvARM1^-L308K) were introduced by SDM using the QuikChange Kit (Stratagene, San Diego, CA, USA). The resulting sequence-verified constructs were bombarded into barley as described for BAC clones. For PCR primers used in this study see Additional file 1: Table S9.

For the HvThf1 and HvClpS1 protein degradation assay, 4 μg each of plasmid DNA encoding HvPUB15 or HvARM1 plus Thf1:YFP or ClpS1:YFP plus pUbiGUS [93] were co-bombarded into 7-day-old barley cv. Golden Promise. The numbers of YFP-fluorescing cells with plastid-localized signals were counted 24 h after particle bombardment, followed by GUS-staining [93]. The numbers of GUS-expressing cells were used for normalization of the YFP signal.

Inoculation and evaluation of transgenic plants

Phenotypic evaluation of Bgh and Bgt interactions was done microscopically on second, detached leaves of 12–14 day-old plants placed on phytoagar plates (23,2 cm × 23,2 cm) inoculated at a spore density of 30–40 conidia mm^− 2. Inoculated leaf segments were incubated for 48 h (Bgh) or 72 h (Bgt) followed by staining with Coomassie brilliant blue R 250 [95]. The number of growing colonies/leaf area was counted under a standard bright field microscope at 100× magnification.

Yeast two-hybrid experiments

Yeast two-hybrid screening was performed according to the Yeast Handbook and manual of Matchmaker™ Library Construction and Screening Kits (Takara/Clontech Laboratories, Saint-Germain-en-Laye, France). The full-length coding sequence of HvARM1 (1–442 AA) was used to screen a library of 7 × 10⁶ mating events according to [96]. For targeted Y2H assays, the coding region (1–831 AA) of HvPUB15 was used to test positive prey clones of the HvARM1 screening.

Bimolecular fluorescence complementation and co-immunoprecipitation

For bimolecular fluorescence complementation (BiFC) of HvARM1 and HvPUB15 proteins with potential plastid interactors, Nicotiana benthamiana plants were grown and agro-infiltrated as described in detail in Additional file 3: Methods S1. For BiFC with HvThf1 and HvClpS1, the wild-type full-length sequences of HvPUB15 or HvARM1, U-box mutants of HvPUB15, the ARM domain (351 to 831 AA) only of HvPUB15, or HvThf1 without N-terminal plastid import signal (-SP) were cloned into 35S::^GWVYNE-pBar and 35S::^GWVYCE-pBar GATEWAY destination vectors containing the N- and C-terminal split parts of the enhanced YFP protein Venus, respectively [97]. BiFC constructs were transiently co-expressed by infiltration of Agrobacterium tumefaciens transformed with the corresponding binary vectors, and examined by confocal laser scanning microscopy (CLSM) 48 h after infiltration. For the development of U-box mutants, a DNA fragment between 709 and 739 bp (from ATG) on the U-box domain of HvPUB15 was excised using BsaXI and replaced by ligating synthetic oligos carrying the respective U-box mutation.

For co-immunoprecipitation (Co-IP), YFP-tagged HvARM1 and HvPUB15^ARM under the control of the 35S promoter were generated by cloning the full-coding sequence of HvARM1 (1–442 AA) or the ARM-repeat region of HvPUB15 (351–831 AA) into pEARLEYGATE104 (Earley et al., 2006). cMyc-Tagged HvThf1 (1–294 AA) and HvClpS1 (1–161 AA) under the control of the CaMV 35S promoter were generated by cloning into pGWB418 [98]. Mesophyll-protoplast transformation and co-immunoprecipitation was done as described [99].

Subcellular localization of fluorescent proteins

For subcellular localization, full-length sequences of HvPUB15, HvARM1, HvThf1, and HvClpS1 were N- and C-terminally fused in-frame to YFP in pIPKTA48 and pIPKTA49 vectors (Additional file 1: Figures S11 and S12). Resulting YFP-fusion constructs were transiently expressed in 7-day-old barley leaf segments by particle bombardment and examined after 12–24 h of incubation with or without B. graminis inoculation using CLSM.

Genome-wide search for expressed partial gene duplicates

To identify expressed partial gene duplicates, a local database was generated using 23,614 full-length cDNA sequences of barley [100], and a Blast search was carried out against itself using the megablast tool in the Galaxy platform [101] with the recommended settings, except the percent identity cutoff was set to 80–99%. To exclude pairs of genes sharing only functional domains, an alignment-to-shorter-gene length ratio of at least 0.8 was set. To identify the open reading frames, a tBlastx analysis of the pairs was carried out using the Galaxy platform. Manual curation was done to exclude non-spliced transcripts and chimeric as well as partial clones. Identified full-length and partial genes were aligned using MAUVE and the MUSCLE algorithm (MegAlign Pro, Version 15.1.0 (155) DNASTAR).