Many plant-associated pathogens impede plant growth and reproduction. In response, plants defend themselves from pathogen attack through two layers of defense [1]. The first layer is the PAMP (pathogen-associated molecule pattern)-triggered immunity (PTI), that is, plant cell surface pattern-recognition receptors (PRRs) detect PAMP elicitors. PTI is often a non-host resistance to the non-adapted pathogens. The second layer of defense is that plant resistance (R) proteins recognize specific pathogen effectors and elicit an effector-triggered immunity (ETI). While usually occupying extracellular niches, the pathogens extract the nutrients for their growth and proliferation from host cells, and the host cytoplasm and organelles which serve as important sites of molecular host-pathogen interaction. Thus, in contrast to PTI, ETI is effective against the adapted pathogens. The ETI-based recognition is mediated by a class of R proteins or effector-recognition receptors with the nucleotide-binding site -leucine-rich repeat (NBS-LRR or NLR) domains. The NLR proteins are often involved in race-specific resistance under the ‘gene-for-gene’ framework in many crop plants [2]. The problem with the use of such race-specific NLR genes in crop cultivars is that they have quickly become ineffective when new, more virulent races appear in the adapted pathogens [3]. For this reason, plant breeders and pathologists have focused on discovery, characterization and use of race-nonspecific genes for durable resistance. However, decades of genetic and breeding research have only been able to identify a limited number of genes in crop plants with durable and broad-spectrum resistance to pathogens (particularly rust pathogens in cereal crops). These race-nonspecific genes include those encoding ATP-binding cassette (ABC) transporter and kinase-START (steroidogenic acute regulatory [StAR] protein-related lipid transfer) proteins [4, 5].

Stem (or black), leaf (or brown) and stripe (or yellow) rusts are among the most damaging fungal diseases of wheat and other cereal crops around the world. Leaf rust is the most common of the three diseases in the Great Plains of North America [6] and more recently stripe rust has occurred more frequently in the Canadian Prairies and other parts of the Great Plains [7, 8]. Since the rust pathogens, Puccinia graminis f. sp. tritici (stem rust), P. triticina (leaf rust) and P. striiformis f. sp. tritici (stripe rust), are widely distributed, capable of forming new races virulent to previously resistant cultivars, able to travel long distances and develop rapidly under optimal environmental conditions, effective control of the rust diseases has been challenging. Breeding for new resistant cultivars has been the most cost-effective means of controlling the rust diseases. Consequently a large number of genes conferring resistance to leaf rust (Lr), stem rust (Sr) and yellow rust (Yr) have been identified in wheat cultivars with 71 Lr, 57 Sr and 53 Yr genes being recently catalogued [9]. A majority of these rust resistance genes are race-specific, conferring the resistance to one or a few races of a rust pathogen and these genes are known or assumed to encode the NLR proteins. However, a few other resistance genes, such as Yr36 and Lr34 (=Lr34/Yr18/Sr57) encoding kinase-START and ABC proteins, respectively, are known to confer race-nonspecific resistance (i.e., resistance to most or all races of the same rust pathogen or resistance to multiple rust pathogens). There is little information about the proteins and their families beyond these few well-characterized rust resistance genes. For this reason, Peng and Yang [5] recently used primer sequences of non-SNP markers and flanking sequences of SNP markers for known Lr genes or QTLs for leaf rust resistance to predict candidate genes located at the same or adjacent genomic regions of wheat, but their prediction was only able to identify the candidate genes in the three gene families (ABC, NLR and START), likely due to the limited annotations of rust resistance genes across the large, complex wheat genome.

The genome of model plant Arabidopsis (Arabidopsis thaliana) was first sequenced in 2000 by the Arabidopsis Genome Initiative [10] and its latest version along with functional annotation of over 27,000 genes and 35,000 proteins is maintained in TAIR10 and Araport11 [11]. The Arabidopsis genome contains many polymorphic defense-related genes [12, 13], and several pathways of Arabidopsis in response to pathogen infection have been well characterized [14]. For example, some150 NLR genes were found in Arabidopsis and their characterization allowed for annotation of candidate disease resistance genes in humans and many flowering plants [15, 16]. It is estimated that approximately 70% of the genes associated with the development of cancers in humans have orthologs present in Arabidopsis [15].Therefore, Arabidopsis may serve as a model system for elucidating the spectrum of plant-pathogen interactions, and the knowledge of pathogen resistance mechanisms gained from this species is useful in crop-pathogen systems.

Recently, there is a growing interest in the use of bioinformatics approaches to inferring about genes of agronomic and adaptive importance in crop plants from the model plant Arabidopsis, and such inference has included flowering-related genes in wheat and barley [17], garden pea (Pisum sativum) [18], soybean (Glycine max) [19, 20], mungbean (Vigna radiata) [21] and cotton (Gossypium hirsutum; [22], and rice [23, 24]. However, little work has been done with prediction of disease resistance genes in crop plants from Arabidopsis. The recent release of wheat genome sequences [25] along with subsequent efforts in more in-depth sequencing will allow for the use of bioinformatics approaches to identifying new defense-related genes that encode the proteins of different families in addition to ABC, NLR and kinase-START families described above through comparative analyses of wheat and Arabidopsis genomes. Thus, the objectives of this study are (i) to infer homology of defense-related genes in Arabidopsis and wheat and (ii) to classify these homologous genes into different gene families.

This study employed three bioinformatics and genomics approaches to predict up to 1790 defense-related candidate genes within 11 gene families (Table 1) in Arabidopsis and their homologs for race-specific and race-nonspecific resistance to leaf, stem and stripe rusts in wheat (Table 4). In addition to the three gene families (ABC, NLR and START) that are already known to confer race-specific and race-nonspecific resistance to wheat rusts and other pathogens [4, 5], eight new gene families for plant defense are now being inferred by the three approaches. It is somewhat surprising that Approach 1 (inference based on sequences of markers for wheat rust resistance) predicted more candidate genes in individual families over the Arabidopsis genome than did Approaches 2 and 3 except for the kinase family where Approach 3 predicted the most genes. Approach 2 did predict the most candidate genes overall as expected because it was based on a genome-wide scan of associations between 21 defense-related traits and 12.8 million SNPs across the Arabidopsis genome. However, the majority (>75%) of the predicted candidate genes by Approach 2 could not be assigned to any of the 11 gene families, comparing to the proportions of unassigned candidate genes being 20.3 and 50.2% by Approaches 1 and 3, respectively. The prediction was based on the latest Arabidopsis genome annotation file in Araport11 [11].

It should be of little surprise to observe partial overlaps of candidate genes for plant defense in Arabidopsis as predicted by the three approaches (Fig. 2). First of all, while all inferred candidate genes are related to plant defense, they are derived from different sources, probably representing different parts of genomic regions. The candidate genes by approaches 1 and 3 were inferred indirectly through homology between DNA sequences of Arabidopsis and wheat as theses inferences were based on the primer sequences of non-SNP (approach 1) or SNP (approach 3) markers related to rust resistance in wheat. In contrast, approach 2 inferred the candidate genes based directly on their physical positions (as shown in the Arabidopsis genome annotation file: Araport11_GFF3_genes_transposons.201606.gff [11]) relative to the chromosomal positions of the SNPs with significant associations with the 21 defense-related phenotypes. Second, the 21 phenotypes used in approach 2 covered a wide range of pests including infections of bacteria, fungi and insects whereas the phenotypes used in approaches 1 and 3 were limited to responses to three rust fungi. As genes known to be exclusively crucial for defense against bacterial intruders may not be the same for the resistance to fungal pathogens [50], the observed partial overlaps of identified candidate genes between the three approaches would likely be expected. Third, despite the evolutionary conservation of core (shared) component genes for fungal resistance in monocot (e.g., cereals including wheat) and dicot (Arabidopsis) plants since their split ~200 Mya ago, clade/species-specific innovation genes are also required to form a fully functional module in plant innate immunity or defense [50], again likely leading to the observed partial overlaps.

Another important finding from this study is that the distributions of predicted SNPs for individual rust resistance genes were highly skewed towards specific gene families (Table 4). At the extreme, the predicted SNPs for a rust resistance gene appeared only in one gene family with eight such unique association pairs being Lr21-NLR, Lr34-ABC, Lr37-START, Sr2-Cupin, Yr24-Transcription factor, Yr26-Transporter, Yr36-Kinase and Yr53-Kinase. Two of these pairs, Lr21-NLR and Lr34-ABC, are expected because Lr21 and Lr34 are well known to confer race-specific and race-nonspecific resistance to leaf rust and they encode NLR and ABC proteins, respectively [4, 5]. Yr36 encodes a protein with a kinase domain fused to a putative START lipid-binding domain [49] and thus the Yr36-Kinase pair is expected though Yr36 is sometimes cited as a gene encoding the START protein in the past [5]. The Sr2-Cupin is probably expected as well because the Sr2 region consists of 10 cupin-domain-containing GLP (Germin-Like Protein) genes [51]. The information is yet available on molecular and functional characterizations of the remaining four unique pairs, Lr37-START, Yr24-Transcription factor, Yr26-Transporter and Yr53-Kinase, and this is certainly an area for future research.

Functional (rust resistance) diversity of a gene family may be indicated by the predicted numbers of SNPs residing around the rust resistance genes. At one end of the diversity spectrum, three gene families were specific to single rust resistance genes (Table 4): the Peroxidase is specific to Sr36, the protease specific to Sr26 and the transporter specific to Yr26. At the other end of the diversity spectrum, the ABC and NLR were functionally diverse as the abundant SNPs in these families were distributed over many rust resistance genes. This pattern of functional diversity was confirmed by the estimates of Gini-Simpson diversity index [52] for individual gene families: the Peroxidase, protease and transporter families were the least diverse with the estimates of zero while the ABC and NLR families the most diverse with the estimate of 0.698 and 0.640, respectively. The moderate-sized estimate of functional diversity (0.450) for the cupin family is hardly surprising as the GLPs in the cupin family are a group of small (~220 amino acid residues), functionally and taxonomically diverse proteins with two of them known to hydrogen peroxide, a plant defense signal [53]. Thus, the richness and evenness of the predicted SNPs within individual gene families may serve as a useful indicator of functional diversity of genes for rust resistance and other agronomic traits in future studies.

Our study is unique in several ways. First, our identification of the candidate genes and SNPs in coding and noncoding regions over the 11 gene families contributes significantly to current understanding of functional diversity with rust resistance genes in wheat. It was recently reported [54] that only six race-specific wheat rust resistance genes (Lr1, Lr10, Lr21, Sr33, Sr35 and Yr10) were cloned, all encoding the same class of proteins with NLR domains. Even fewer (only three) race-nonspecific rust resistance genes were cloned so far [4], with Lr34 encoding a protein in the ABC family, Lr67 encoding a protein in the STP (sugar transporter protein) family and Yr36 encoding a protein in the kinase family. In this study, we were able to identify up to 1790 candidate genes encoding proteins belonging to the 11 families that are in physical proximity to the rust resistance genes distributed over different wheat chromosomes. While the race-specific resistance is often due to a single NLR gene, there are cases (e.g., Lr10) where such resistance is due to two or more adjacent genes [55]. Second, our comparative genomic analysis between Arabidopsis and wheat largely avoids several problems, such as tediousness, low marker density and limited recombination rate, that often arise from the conventional map-based approaches [56, 57], thereby accelerating the discovery of rust resistance genes in wheat and other cereals. Despite the ongoing international efforts, the full genome sequence of allohexaploid wheat remains difficult to obtain because of (i) its colossal size (17.1 Gb vs. 0.135 Gb of the Arabidopsis thaliana genome, a 126-fold difference); (ii) high sequence identity of homologous genes of three highly related subgenomes (A, B and D); (iii) genomic complications [~24% of the genes undergoing intrachromosomal duplications and ~81% of the genome consisting of repetitive DNA, primarily LTR (long terminal repeat) retrotransposons] [58]. On the other hand, our comparative genomics approach allows for leveraging the rich genomic resources from the deeply-sequenced and well-annotated Arabidopsis thaliana genome [35] for identification of wheat rust resistance gene homologs. Third, our genome-wide approach broadens the scope of previous studies focusing only on functional and molecular characterizations of protein-coding genes in the ABC and NLR families for rust resistance [4, 56]. Our analysis largely confirms such characterizations, and more importantly the new candidate genes and families serve as an important basis for future research towards their complete characterization and their use for wheat breeding for rust resistance.

Our study focuses on inference of gene families for plant defense in a model plant (Arabidopsis) and an agriculturally important crop (wheat). Similar inferences can be found in other model or crop plants though they often focus on specific disease loci rather than a genome-wide approach used in our study. For example, Li et al. [59] recently conducted a GWAS analysis (similar to our approach 2) to identify a natural allele of a C₂H₂-type transcription factor with race-nonspecific resistance to blast pathogen (Magnaporthe oryzae) in rice. The resistance allele (SNP33-G) differs from its susceptibility counterpart (SNP33-A) in just a single nucleotide in the promoter of the broad-spectrum resistance Digu 1 (bsr-d1) gene (LOC_Os03 g32230), 618 base pair before the coding region. This point mutation causes reduced gene expression through the binding of the repressive MYB transcription factor, thereby inhibiting H₂O₂ degradation and enhancing rice blast resistance. An added novelty of this newly discovered allele is that its broad-spectrum race-nonspecific resistance does not link to yield penalty in resistant cultivars as found, for example, in Lr34-wheat yield tradeoff [60] and Pigm-rice yield tradeoff [61]. Many other studies have used an analysis (similar to our approach 2) to reveal resistance genes largely from the NLR or ABC family in different cereals (see Table 1 of Krattinger and Keller [62] for specific examples), but it remains to be seen the interspecies homology analysis similar to our approach 1 or 3.

The knowledge on a wide array of resistance candidate genes in different gene families as acquired in our and other studies will have important implications for new breeding strategies of developing durable resistance to multiple races of the same pathogens or multiple pathogens in wheat and other cereals [56, 63]. The key to success of these new breeding strategies is the ability to stack race-specific resistance genes (mostly known in the NLR gene family) along with race-nonspecific genes in the other gene families (e.g., those in the ABC family) to produce the durable resistance to wheat rusts and other cereal pathogens. The gene stacking is preferred over the conventional breeding methods of singularly deploying single race-specific resistance genes as it will avoid lack of durability of race-specific resistance due to rapid mutation or loss of recognized pathogen effectors. The candidate resistance genes identified in our and other studies need to be further tested and validated before they can be effectively used for the gene stacking. However, the current process of gene testing and validating remains slow (one gene at a time) and costly as it often uses the gene knockout approach. For most durable resistances with complex, polygenic inheritance (e.g., a cluster of 13 genes in the NLR family uncovered at the Pigm locus for rice blast resistance [61]), it will be practically feasible in near future to have a genome-wide manipulation of the relevant genes through some latest genome editing technologies [64], thereby capturing the joint contributions of all stacked genes towards the improved durable resistance.

Our study was able to identify the candidate genes in three gene families known to confer rust resistance in wheat and eight new families, thereby enhancing current understanding of functional diversity with defense-related genes in genomes of model plant Arabidopsis and cereal crop plant wheat. Our sequence-based annotation and comparative genomic analysis of Arabidopsis and wheat genomes allows for genome-wide identification of candidate genes for plant defense or other traits. In contrast, the commonly used map-based or marker-based approaches only have the limited ability to discover new defense-related genes (e.g., rust resistance genes in wheat and other cereals) due to usual challenges such as tedious fine mapping, low marker density and limited recombination rate. This is particularly true for large, complex genomes of wheat and other cereal crops with incomplete and imperfect sequencing and annotation information despite the ongoing international efforts on genome sequencing in these crops. The acquisition for a wide array of resistance candidate genes in different gene families in our and other studies is an important first step towards implementing a new gene-stacking strategy that combines race-specific resistance genes (mostly known in the NLR family) with race-nonspecific genes in the other gene families to breed for the durable resistance to wheat rusts and other cereal pathogens.