Strategies for Breeding Durable Resistance to Rice Blast Using pi21

Rice blast caused by Pyricularia oryzae (syn. Magnaporthe oryzae) is a destructive disease of rice worldwide that poses a serious threat to rice production. Strengthening blast resistance is an important objective in rice breeding programs. Race-specific resistance genes (R-genes) confer complete resistance to blast, but new races of blast pathogen can overcome it. After the first report of breakdown of resistance conferred by the R-gene Pik in 1963, this type of resistance has frequently been broken 1–7 years after the release of resistant varieties to farmers in Japan and other countries. To overcome this genetic vulnerability, Japanese rice breeders have focused on the use of race-nonspecific resistance in Japanese upland rice varieties whose resistance has been maintained for a long time. However, linkage drag between genes controlling this type of blast resistance and undesired traits has hindered its use. Therefore, researchers genetically dissected race-nonspecific resistance to rice blast. Among detected QTLs, a single recessive resistance gene, pi21, was identified by map-based cloning. The use of pi21 has improved durable resistance in rice breeding programs.


INTRODUCTION
Rice blast caused by the fungus Magnaporthe grisea (Pyricularia oryzae) is a major biotic constraint in rice cropping regions worldwide that threatens global rice production and productivity. Between 10% and 30% of the annual rice harvest is lost because of infection by rice blast fungus [1]. Genetic improvement of resistance against rice blast is a significant and primary target in rice breeding programs. Varietal resistance has been explored for over a century for sustainable production [2]. As in other plant-pathogen interactions, resistance to blast is categorized into two types, race-specific (complete, qualitative or true) and race-nonspecific (partial, quantitative or field) [3,4]. Race-specific resistance (1) is based on the hypersensitive reaction, (2) is often complete, and (3) is characterized by a resistant infection type. Race-nonspecific resistance is a susceptible infection type that allows effective control of a pathogen under natural field conditions and is considered to be durable when plants are exposed to new races of the pathogen and maintain their previous degree of resistance.
Race-specific blast resistance is achieved through many race-specific resistance genes (R-genes) identified in a broad range of the world's rice germplasms [5]. R-genes dramatically enhance blast resistance, resulting in stable rice production, but their extensive use poses a serious risk of the generation of new races of the blast pathogen and the quick breakdown of resistance. In Japan, the first breakdown of resistance occurred only 2 years after the release of the resistant "Kusabue" with the R-gene Pik introgressed from a Chinese variety in 1963; similarly, resistance conferred by several different R-genes was broken in the following years (Table 1). Similar trends were reported in Korea and Colombia [6]. Hence, rice breeders started paying attention to the use of race-nonspecific resistance. Table 1. Instances of breakdown of R-gene-mediated resistance to rice blast in Japan.
Japanese upland rice varieties are potential gene donors of race-nonspecific resistance [8]. Genetic studies indicate that their resistance is controlled by multiple genes or polygenes, two of which may be linked to the phenol reaction (Ph) locus on chromosome 4 or the lax Crop Breed Genet Genom. 2019;1:e190013. https://doi.org/10.20900/cbgg20190013 panicle (lax) locus on chromosome 11 [9,10]. However, conventional genetic approaches cannot determine the exact number of genes associated with the resistance or their chromosomal locations. Moreover, upland rice varieties have undesired characteristics, in particular poor grain and eating quality [8]. Before upland rice varieties can be used to improve durable resistance to rice blast, the genetic dissection of the resistance is required.
The Rice Genome Project was initiated in Japan in 1991 and has greatly contributed to rice genetics and breeding [11][12][13][14]. Molecular markers mapped over the 12 rice chromosomes and over the entire genome sequence are powerful tools to identify genes controlling quantitative traits. Many quantitative trait loci (QTLs) for agricultural traits, including race-nonspecific resistance of Japanese upland rice varieties, have been identified using these markers, and beneficial QTL alleles have been introduced into elite genetic backgrounds. This review focuses on the identification of pi21, a QTL allele conferring race-nonspecific resistance in a durably resistant variety, by genome-based analyses and its use in breeding programs. The current status of the identification of other genes found in durably resistant varieties, gene pyramiding and the use of multiline varieties are also discussed.

PARADIGM SHIFT IN RICE BLAST RESISTANCE
The breakdown of the resistance conferred by R-genes occurs 1-7 years after their release to farmers (Table 1). This so-called genetic vulnerability is explained by the emergence of new races of blast pathogen when varieties with the same resistance genotype are predominant in farmers' fields. Genetic studies on durably resistant varieties have accelerated a shift from race-specific to race-nonspecific resistance genes in rice breeding programs in Japan.

QTLs Underlying Race-Nonspecific Resistance to Rice Blast
QTLs for race-nonspecific resistance to rice blast were analyzed in progeny derived from crosses between Japanese upland and paddy rice varieties. In an "Owarihatamochi" (resistant, upland) × "Nipponbare" (moderately susceptible, paddy) cross, two resistance QTL alleles on chromosome 4 and one on chromosome 12 from "Owarihatamochi" were identified [15]. Each QTL explained from 13.7% to 45.7% of the total phenotypic variation. The results suggest that the resistance of "Owarihatamochi" is controlled by a small number of QTLs with different contributions. The QTL on chromosome 4 was inherited as a single recessive gene and was designated pi21 [15]. In addition, one region on chromosome 11 was significant at a lower probability threshold [15]; it was later designated Pi34 and was analyzed in lines derived from a Japanese upland variety [16,17]. Resistance QTL alleles from other upland varieties were detected in regions similar to those in "Owarihatamochi" on chromosomes 4, 11 and 12, although their relative contributions to decreasing disease severity differed among cross combinations [18,19]. Resistance QTL alleles from other varieties were located mainly in a 30-Mb region of chromosome 4 and on chromosomes 1, 3, 6 and 11 [16,[20][21][22][23][24]; these observations imply that genetic differentiation of disease resistance genes may cause variation in the magnitude of resistance conferred by respective QTLs. Accumulated evidence highlights the target regions for improving race-nonspecific resistance to rice blast.

Characterization of pi21 Using a Near-Isogenic Line
The effect of the pi21 resistance allele alone cannot be evaluated by inoculation tests using a donor variety that carries multiple resistance QTL alleles [15,25,26]. Near-isogenic lines (NILs; lines genetically identical except in one or a few loci) are useful for characterization of loci conferring complex agricultural traits owing to their homogeneous genetic background. A NIL for pi21 in the genetic background of the susceptible variety "Aichiasahi" was used to test the response to 16 widely distributed blast races [26,27]; the quantitative and consistent effect of pi21 against all of the races tested was found.
A transient increase in the expression of pathogenesis-related (PR) genes at 3-6 h after inoculation with a virulent race was observed in plants carrying pi21 but not in plants carrying R-genes [27]. Inoculation of plants carrying pi21 with elicitor solution mimicked this response, and removal of the elicitor from the inoculum decreased blast resistance in these plants [28]. These observations imply a role of pi21 in the pre-penetration plant-pathogen interaction through elicitor-triggered immunity. Unlike in plants lacking pi21, inoculation tests after application of an antagonist of ethylene biosynthesis did not decrease blast resistance in plants carrying pi21 in comparison with the corresponding untreated controls [26]. Since the inhibition of ethylene biosynthesis decreases resistance to a number of diseases [29], this distinctive response implies the involvement of ethylene signaling in pi21-mediated resistance. A recent study has suggested the complex control of signaling in pi21-mediated resistance [30].
Unlike other defense genes such as WRKY45 and BSR1, which alter resistance to multiple plant pathogens [31,32], pi21 does not affect resistance to the bacterial pathogen Xanthomonas oryzae pv. oryzae or the fungal pathogen Rhizoctonia solani [28]. Therefore, it might inhibit the hyphal growth of the blast pathogen, and identification of protein(s) interacting with Pi21 would provide further insight into mechanism of pi21-mediated resistance.

CLONING AND CHARACTERIZATION OF GENES FOR RACE-NONSPECIFIC RESISTANCE
More than 100 loci for resistance to blast have been identified and more than 30 of them have been cloned [33][34][35]. Most of them encode nucleotide-binding site (NBS) leucine-rich repeat (LRR) proteins that interact with pathogen effectors and trigger defense reactions according to the gene-for-gene model of recognition [36,37]; the exceptions are Pid2, which encodes a receptor-like kinase [38], and Ptr, which encodes an Armadillo repeat protein [39]. Recent studies in rice and Arabidopsis showed that different genes for NBS-LRR proteins, such as a pair of tightly linked NBS-LRR genes, cooperate in pathogen recognition and resistance [40][41][42][43][44][45]. In contrast, the number of genes that have been cloned for race-nonspecific resistance is limited.

Map-Based Cloning of pi21
Linkage analysis and progeny testing narrowed down the Pi21 locus to a small region carrying a single gene locus, Os04g0401000 [27]. This gene encodes a protein with a putative heavy-metal-binding domain and a proline-rich region. Comparison of sequences between resistant and susceptible varieties identified 21-and 48-bp deletions in the resistant variety, suggesting that one or both of these deletions confer resistance. Transgenic complementation testing confirmed that a loss-of-function mutation in the gene improves resistance to blast. Suppression of the expression of the gene in a susceptible variety resulted in a level of resistance similar to that of plants carrying pi21 [27]. Conversely, transgenic plants with higher expression levels of the susceptibility-conferring Pi21 allele were more susceptible to blast than those with lower expression. Increased expression of Pi21 did not alter the strength of the Pia-dependent hypersensitive reaction against an avirulent race. These results demonstrated that Pi21 is a negative component of defense that belongs to a pathway different from R-gene-mediated resistance.
Information on the variation of QTL alleles allows the use of a wide range of germplasm. In the case of the pi21 gene, Asian cultivated rice has 12 haplotypes determined by insertion/deletion variations at three sites in the proline-rich region, which is presumed to be involved in protein-protein interactions in multicellular organisms [46,47]. Each of the 12 haplotypes carries one of the two deletions or two smaller deletions compared with "Owarihatamochi" haplotype, but it is difficult to predict the resistance/susceptibility phenotype from DNA sequences. Inoculation testing using a series of backcrossed lines carrying each of the Pi21 haplotypes in the same genetic background indicated that only the line carrying the "Owarihatamochi" haplotype showed improved resistance to blast; the rest were susceptible, similar to the recipient variety [27]. The results suggest that the two deletions in the resistance pi21 allele are optimal to cause the loss of function, which increases resistance to blast. Rice varieties carry susceptibility alleles that can be replaced with the resistance pi21 allele.
Genes associated with disease susceptibility are considered essential for plant growth, and their loss frequently has deleterious effects [48]. Hence, for practical use, resistance alleles that show a partial loss of function and are mildly pleiotropic are desirable. In the xa13 allele for resistance to X. oryzae pv. oryzae, a mutation prevents pathogen propagation, but the mutated protein retains certain functions in normal pollen development [49]. DNA variations that cause amino acid changes in Pi21 in Asian cultivated rice were found only in the proline-rich motif sequences encoded in the middle of the gene, whereas the C-terminal region and the putative heavy-metal-binding domain in the N-terminal region are free from variations [27]. Such observations imply that the resistance pi21 allele maintains certain functions important for plant growth, as in the case of xa13. The lower survival rate of plants whose Pi21 expression is strongly suppressed by RNAi is in line with this hypothesis (S. Fukuoka, unpublished data). A recent study reported increased blast resistance conferred by a CRISPR/Cas9-edited Pi21 gene [50]. Characterization of the agronomic traits of lines having diverse Pi21 variants will provide further evidence on this topic.

Map-Based Cloning of Other QTLs for Race-Nonspecific Resistance
Pi35 is a major resistance QTL on chromosome 1; its resistance allele was found in a Japanese breeding line, "Hokkai 188", that has maintained resistance under natural field conditions since 1961 [51]. Linkage analysis and complementation testing revealed that Pi35 is allelic to Pish, a typical R-gene that encodes an NBS-LRR protein [52]. Among 6 differences in the deduced amino acid sequences between Pi35 and Pish, one of the four residues in the LRR region is significantly associated with race-nonspecific resistance [53]. However, the analysis of chimeras between Pish and Pi35 confirmed that the three other residues in the LRR region and two residues in the NBS domain are also associated with the resistance, suggesting that a combination of multiple functional polymorphisms in the gene confers race-nonspecific resistance [53]. Plants with Pish are completely resistant to a single blast isolate but susceptible to other isolates tested under natural field conditions in Japan. Plants with Pi35 were less resistant to the isolate avirulent to Pish plants; but were consistently resistant against the other blast isolates tested [53]. This example implies that the quantitative nature of resistance governed by an NBS-LRR protein gene may decrease selection pressure against the pathogen.
Pi63 is a major resistance QTL in a 30-Mb region of chromosome 4; its resistance alleles were found in a Japanese upland rice variety, "Kahei" [19]. Linkage analysis and complementation testing demonstrated that this gene encodes an NBS-LRR protein and is located within an R-gene cluster [54,55]. Not only the difference in amino acid sequences, but also different expression levels of Pi63 and its counterpart allele in a susceptible variety could lead to the resistant phenotype [54]. Interestingly, resistance conferred by Pi63 is isolate-specific, as demonstrated using a NIL for Pi63. Such characteristic has not been identified in the genetic background of the donor variety because of the effect of the race-nonspecific pi21 allele and alleles of other resistance QTLs. Increased expression of Pi63 in transgenic lines led to moderate resistance against pathogen isolates that produce a highly susceptible phenotype in the NIL for Pi63. Therefore, variations of the expression levels of genes for NBS-LRR proteins could be part of the genetic mechanism of race-nonspecific resistance in rice.
Panicle blast 1 (Pb1) on chromosome 11 is a gene derived from the indica variety "Modan" [56]. Plants carrying this gene are blast susceptible during young vegetative stages, but the resistance level increases as the plants grow, and persists even after heading [57]. This gene is useful for conferring resistance to panicle blast because the varieties that have this gene maintain resistance over several decades [57]. Map-based cloning of Pb1 revealed that it encodes an atypical NBS-LRR protein that has no Ploop and some motifs in the NBS domain are degenerated [58]. Pb1 transcript levels have increased during the development of Pb1-resistant varieties and effectively control panicle blast [58]. Pb1-mediated resistance seems to be mediated by a signaling pathway distinct from that involving typical NBS-LRR proteins [59].

PRE-BREEDING AND BREEDING OF RICE RESISTANT TO BLAST USING pi21
Most Japanese upland rice varieties are donors of pi21; their morphological and physiological characteristics are distinct from those of elite genotypes. Breeding efforts to introduce resistance alleles of major QTLs from upland rice started in the 1920s. But the trials were unsuccessful because of the co-introduction of undesirable characteristics from the donors [7,60]. This co-introduction could be explained by tight linkage of genes controlling independent traits (linkage drag) and/or by the effect of the target gene on other traits (pleiotropic effect). Therefore, the development of NILs for pi21 in a desirable genetic background is a possible strategy for determination of the cause of this association and for enhancing the use of resistance QTL alleles from unimproved genetic resources.

Development of NILs for pi21
Agronomic traits of NILs for pi21 in the genetic background of an elite rice variety ("Mineasahi") were evaluated [27]. Despite the presence of less than 5% of donor chromosome sequences, plants carrying pi21 had poor grain and eating quality, which were not observed in reference lines carrying Pi21 [27]. The results strongly support the idea that pi21 or a gene(s) tightly linked with it controls grain characteristics. During NIL development, DNA markers tightly linked with the target QTL are used for foreground selection, and background selection around that QTL is not intensive. When the precise map position of the target is not determined, the size of the selected introgression will be larger so as not to miss the gene. Such situation could be the reason for the difficulty in the use of beneficial traits of unimproved genetic resources. The two cases, linkage drag and pleiotropy, cannot be discriminated unless the linkage can be broken.

Removal of Linkage Drag and Development of Varieties Carrying pi21
To remove donor chromosome segments around the Pi21 locus and elsewhere in the genome, a line carrying the pi21 allele was backcrossed with an elite paddy rice variety ("Koshihikari"), and progeny carrying a single 1.8-Mb fragment around the Pi21 locus from the donor was selected. Plants with recombination events within a 40-kb interval containing the Pi21 locus were selected from approximately 6000 progeny. The eating quality of a progeny line carrying the "Koshihikari" chromosomal sequence from a point less than 2.4 kb downstream of the Pi21 locus was equivalent to that of the elite variety, and the line was highly resistant to blast. In contrast, a progeny line carrying the donor chromosomal sequence up to 37 kb downstream of the Pi21 locus showed inferior eating quality [27]. These results clearly show that the resistance pi21 allele does not penalize agronomic traits, and the cause of the association is tight linkage with genes causing undesirable traits. The recessive nature of the resistance allele also made it difficult to select this locus by conventional procedures.
The promising line with improved blast resistance and desirable grain characteristics was released as "Tomohonami" ("Chubu 125") in 2009 [61]. "Tomohonami" has been used as an excellent donor of pi21 at more than 15 prefectural breeding stations and 6 research centers of the National Agriculture and Food Research Organization, Japan. More than 15 breeding lines carrying pi21 have been developed. Recently, "Fufufu", a line derived from "Tomohonami" carrying pi21 and Apq1 for high-temperature tolerance during ripening, was released, and its area of cultivation is increasing [62,63].
Mutant allele of negative regulators of defense such as pi21 may reduce yield because of constitutive activation of defense responses and have other secondary effects, as barley Mlo does [64,65]. However, slow induction of defense by pi21 contributes to pathogen control without penalty on yield, as confirmed by field tests at several locations [66]. The pi21 alleles are effective against diverse fungus races, so the use of pi21 might not be a strong driving force for changes in the structure of pathogen populations. The durability of resistance conferred by a gene needs to be proved by prolonged resistance of varieties carrying that gene alone under natural field conditions [67]. Monitoring of newly released varieties carrying pi21 will provide further evidence to confirm or disprove the durability of resistance conferred by pi21.

BREEDING STRATEGIES FOR DURABILITY OF BLAST RESISTANCE
Developing varieties that are resistant in a disease-prone area is a challenge in crop breeding. Despite the largest effect of the pi21 allele in comparison with other resistance alleles in "Owarihatamochi", a durably resistant variety, this allele alone may not be sufficient to control the disease under high disease pressure. Two breeding approaches are proposed to increase the durability of resistance to pathogens in crop plants [68]: (i) the use of multiline varieties carrying different resistance genes and (ii) combining multiple resistance genes in the same genotype. This section overviews these two approaches and explains technical issues that need to be considered for sustainable use of durable resistance to blast in rice.

Multiline Varieties for Blast Control in Rice
A multiline variety is a mixture of pure lines carrying different resistance genes. This concept was originally proposed in 1952 for controlling disease in oat [69], and its usefulness was confirmed [70,71]. Mixing varieties with different characteristics contributes to disease control in rice [72], but NILs in elite genetic backgrounds are more desirable components of multiline varieties to ensure the uniformity of agricultural traits. Case studies of various crop-pathogen combinations have shown differences in resistance among multiline varieties [73]. Hence, guidelines for the management of multiline varieties should be based on the evidence for particular crop-pathogen combinations.
As discussed above, the resistance conferred by a single R-gene is vulnerable, while in a field of heterogeneous plants with different R-genes, the damage by the pathogen is decreased. This phenomenon may be explained by (i) the dilution of inoculum owing to a decrease in the density of infected plants [73,74], (ii) barrier effect of resistant plants [75,76], and (iii) induced resistance because of pre-inoculation with avirulent pathogen isolates [77,78]. Under appropriate management, even the R-genes whose resistance has been overcome by the pathogen in the past can be used as components. Thus, this approach allows a sustainable use of R-genes in breeding programs.
Because of the limited number of available resistance QTL alleles, only R-genes have been used for practical breeding of multiline varieties in rice [75,79]. Multiline varieties that rely on 15 recurrent parents have been developed or are under development in Japan [75]. Of 15 R-genes used in the Japanese breeding programs, 13 were incorporated into these varieties (on average, 6.2 per recurrent parent). Despite breeding efforts since the 1980s, only five multiline varieties that rely on four recurrent parents have been released [80,81].
Commercial cultivation of the multiline variety "Sasanishiki BL" started in 1995 in Miyagi Prefecture in Japan [82]. The initial ratio of three NILs having each one of the Pik, Pik-m and Piz genes was 4:4:3. After the increase in the incidence of a pathogen race virulent to the lines carrying Pik and Pik-m in the fields of "Sasanishiki BL", the ratio was changed to 3:4:4, followed by the addition of a line carrying Piz-t, at a final ratio of Pik:Pik-m:Piz:Piz-t = 1:1:4:4. Although this multiline variety has maintained resistance for more than 9 years, pathogen races virulent to each of the NILs have been observed in the field [80,82]. This fact implies that "Sasanishiki BL" may not stabilize the race composition of the pathogen, and its small area of cultivation might instead explain its continued low disease incidence [80].
Multiline varieties of "Koshihikari", the leading variety in Japan, were used in Niigata and Toyama prefectures, with different gene components [79,83]. In Niigata Prefecture, the area for the multiline variety "Koshihikari BL" is larger than that for "Sasanishiki BL". Four out of 11 NILs were used every year, and the choice of the lines and their ratio were based on monitoring temporal race dynamics of the blast pathogen in the prefecture. A theoretical model to slow changes of the estimated pathogen population determined the proportion of resistant plants as 70% [84]. Accordingly, the proportions of the areas of occurrence of leaf blast and panicle blast in the prefecture decreased after the replacement of "Koshihikari" with "Koshihikari BL", and this trend has been maintained for more than 12 years since 2005 (http://www.pref.niigata.lg.jp/ nosanengei/1215712857692.html). This example shows that a multiline variety effectively controls blast damage under appropriate management.
These two cases show that multiline varieties of rice do not stabilize pathogen populations. Therefore, the number of genes in a multiline variety and determination of the components based on the monitoring of temporal changes of pathogen populations are key factors to ensure the durability of resistance. Developing a single multiline variety requires at least 3 (ideally 8 to 10) NILs and their seed production. Sampling of the pathogen and estimation of its population structure are required every year to choose the lines and their relative proportion for the next year. Hence, seed supply requires considerable cost and labor; software that helps seed management has been developed on the basis of simulation studies of temporal pathogen population dynamics in rice [80]. Another aspect of NILs to be considered is the value of their recurrent parent. Because of the high sensitivity of "Sasanishiki" to cold stress at booting stage and because its taste has lost favor among consumers, the cropping area of "Sasanishiki BL" has decreased correspondingly. The life of multiline varieties has become shorter and their market share has decreased because of climate change and diversification in consumers' requirements. These points suggest that the use of multiline varieties is beneficial for leading varieties but not for varieties grown for diverse purposes.

Gene Pyramiding for Sustainable Control of Blast
Gene pyramiding (combining multiple resistance alleles in the one genetic background) is another way to enhance durable resistance in crop plants. If a single genotype confers durable resistance, this approach is more desirable for breeders because breeding procedures and seed management are simpler than those with multiline varieties.
R-gene pyramids improve resistance to diverse pathogen isolates [85][86][87][88]. A comprehensive survey of a series of gene pyramids detected interaction among genes. Among combinations between one of the Pigm, Pi2, Pi9, Pi40 and Piz genes and one of the Pi1, Pi33 and Pi54 genes, Pigm/Pi1, Pigm/Pi54 and Pigm/Pi33 provided the best resistance at both seedling and heading stages [87]. These results highlight the importance of screening for favorable gene combinations to maximize resistance.
Broadening the spectrum of resistance by pyramiding R-genes may prompt the counter-evolution of the pathogen; for example, the resistance of a variety with four R-genes was overcome one year after its release [89]. The emergence of super-races that overcome the resistance of R-gene pyramids might increase over time when a single variety carrying an R-gene pyramid is cultivated in a large area. An epidemiological survey and simulation study on pathogen race dynamics suggest that replacement by varieties with different R-genes leads to drastic changes in the pathogen population structure that increase the risk of disease outbreak [90]. Hence, R-gene pyramids, each resulting in a strong selection pressure against its pathogen, may not improve the durability of resistance. Further study of the effect of R-gene pyramids on pathogen population dynamics in the field is necessary to develop the guidelines for their use.
Combining multiple resistance QTL alleles is considered to additively enhance race-nonspecific resistance. However, breeders and researchers know that disease resistance sometimes interacts with genetic backgrounds and/or environmental factors [91][92][93][94]. The data on resistance to blast over two decades support this idea in the context of race-specificity and temperature-dependent resistance, and indicate the existence of genetic loci that modulate the resistance or its mode of action [26,54,95,96]. To understand how resistance QTL alleles interact with such factors, it is important to determine the appropriate number and combinations of resistance genes. However, knowledge of the impact of QTL pyramiding on the robustness of plant defense in rice is limited [26,97].
In the genetic background of the susceptible "Aichiasahi", the average reduction of lesion area by pi21 in eight field trials was 87% compared with the recurrent parent, whereas that by the minor QTL alleles was 39% by Pi34, 45% by qBR4-2 and 22% by qBR12-1 [26]. Although the effects of these minor QTL alleles were sometimes undetectable, their combinations dramatically reduced lesion area both in field tests and in glasshouse inoculation tests. The line with four resistance alleles had a lesion area of ≤1%, which was similar to that in the donor and was only 6% of that in the line carrying pi21 only, suggesting that the QTL pyramid conferred robust resistance. Similar results were obtained for a series of lines with one to four resistance QTL alleles, despite the presence of background noise (effect of unidentified QTLs) from donors [97]. A more important observation is that a QTL pyramid improves the stability of resistance; the coefficient of variation of lesion area across field tests in the line carrying four resistance QTL alleles was smaller than those in lines with only one or two [26]. That study demonstrated the importance of minor QTL alleles for improving the stability of resistance, even if the effect of each of them is sensitive to the environment.
Histological study and expression analysis supported the idea that the hypersensitive reaction was not induced in the four-QTL pyramid line, unlike in R-gene-mediated resistance, but the defense response was greater than in no-QTL or pi21-only plants [26]. Therefore, the use of QTL pyramids may maintain an optimal balance between the effective control of the pathogen and selection pressure against it, and thus it may confer durable resistance to blast in rice.
One of the concerns in the use of QTL pyramids is linkage drag. Most resistance QTL alleles have not been used in commercial varieties by conventional breeding programs, possibly because of linkage drag, as in the case of pi21 [27]. The use of breeding lines with remaining undesirable traits decreases the efficiency of breeding proportionally to the number of resistance QTL alleles, as seen in conventional breeding. Therefore, the resistance alleles should be precisely mapped and the breeding program should start from linkage drag elimination. The fitness cost of resistance is another issue that should be evaluated in the future. The costs or penalties associated with the activation of defense responses in the absence of a pathogen attack may decrease yield [64,98,99]. Unlike the barley Mlo mutant [64,65], NILs for pi21 only appear to have no penalty on yield [28,68]. However, the penalty on plants that carry multiple resistance alleles has not been well clarified, although at least their growth does not appear to be affected (Fukuoka, unpublished data). However, the pleiotropic effects of resistance QTLs may be small and detectable only in large-scale field tests. Further evaluation in multiple environments is required to answer this question.

CONCLUSIONS
Progress in understanding the genetic control of race-nonspecific resistance in Japanese upland rice has led to a breakthrough in rice breeding, and marker-assisted pyramiding of relevant genes guarantees enhancement of the trait. Varieties carrying pi21 will provide further evidence of the durability of resistance in large cultivation areas. We recommend introducing other resistance QTL alleles into pi21-only varieties for robust disease control in disease-prone areas. Removal of undesirable agricultural traits that are tightly linked with the resistance QTL alleles needs to be considered. The cost of enhanced defense response in QTL pyramids has not yet been evaluated and should be optimized according to the risk of disease.

AUTHOR CONTRIBUTIONS
SF and KO wrote the manuscript, and reviewed, corrected, and proofed the final version.