Formae speciales of cereal powdery mildew: close or distant relatives?

Summary

Powdery mildew is an important disease of cereals, affecting both grain yield and end‐use quality. The causal agent of powdery mildew on cereals, Blumeria graminis, has been classified into eight formae speciales (ff.spp.), infecting crops and wild grasses. Advances in research on host specificity and resistance, and on pathogen phylogeny and origins, have brought aspects of the subspecific classification system of B. graminis into ff.spp. into question, because it is based on adaptation to certain hosts rather than strict host specialization. Cereals therefore cannot be considered as typical non‐hosts to non‐adapted ff.spp. We introduce the term ‘non‐adapted resistance’ of cereals to inappropriate ff.spp. of B. graminis, which involves both pathogen‐associated molecular pattern‐triggered immunity (PTI) and effector‐triggered immunity (ETI). There is no clear distinction between the mechanisms of resistance to adapted and non‐adapted ff.spp. Molecular evolutionary data suggest that the taxonomic grouping of B. graminis into different ff.spp. is not consistent with the phylogeny of the fungus. Imprecise estimates of mutation rates and the lack of genetic variation in introduced populations may explain the uncertainty with regard to divergence times, in the Miocene or Holocene epochs, of ff.spp. of B. graminis which infect cereal crop species. We propose that most evidence favours divergence in the Holocene, during the course of early agriculture. We also propose that the forma specialis concept should be retained for B. graminis pathogenic on cultivated cereals to include clades of the fungus which are strongly specialized to these hosts, i.e. ff.spp. hordei, secalis and tritici, as well as avenae from cultivated A. sativa, and that the forma specialis concept should no longer be applied to B. graminis from most wild grasses.

Introduction

Many species of plant‐pathogenic fungi are classified below the species level into special forms (formae speciales, ff.spp.). This taxonomic grouping describes a situation in which each forma specialis (f.sp.) is classified as: (i) adapted to a specific host species; and (ii) shows no or minimal morphological difference from its closest relatives at the species level (Schulze‐Lefert and Panstruga, 2011). Although the ff.spp. of a fungus often, although not always, relate to the specialization to various crop species, their evolutionary origins are largely obscure, as are the mechanisms by which they have become specialized to different hosts.

We review the biology and evolution of ff.spp. in Blumeria graminis (DC.) Speer, the causal agent of powdery mildew on cereal crop species and grasses (subfamily Pooidae). Powdery mildew is a widespread disease of many monocotyledonous and dicotyledonous plants, caused by Ascomycetes of the order Erysiphales (Braun et al., 2002). Mildew pathogens are obligate biotrophic fungi, depending entirely on living plant cells for survival and reproduction. Blumeria graminis is a major problem for cereal production as it can reduce both quality and yield (Conner et al., 2003; Everts et al., 2001), leading to large economic losses either in reduced output or increased costs of production. Eight ff.spp. have been described for B. graminis, based on their host specialization, each having a capacity for infection limited to a single host genus (Marchal, 1902; Oku et al., 1985). These hosts are wild grasses from the genera Dactylis, Agropyron, Poa and Bromus and four genera which include cereal crops: Triticum (wheat), Hordeum (barley), Secale (rye) and Avena (oats). The corresponding ff.spp. are named dactylidis, agropyri, poae, bromi, tritici, hordei, secalis and avenae, respectively (Fig. 1). Agropyron is now split into several genera, e.g. A. repens (couch grass) is now Elymus repens (Dewey, 1983), but the mildew pathogen is still called f.sp. agropyri. This subspecific taxonomy of B. graminis has been brought into question by several studies which have revealed that individual isolates of B. graminis have host ranges extending to plants from more than one genus and even in different tribes. Moreover, B. graminis is a common pathogen of many grasses which are not hosts of named ff.spp., such as Lolium perenne or ryegrass (Carver et al., 1990).

Figure 1.

Schematic view of Blumeria graminis formae speciales (Marchal, 1902; Oku et al., 1985). ^aThe wheat–rye hybrid triticale is a host for B. graminis related to f.sp. tritici (Troch et al., 2012; Walker et al., 2011). ^bOther grass genera, such as Lolium, are also infected by powdery mildew (Carver et al., 1990; Eshed and Wahl, 1975; Hardison, 1945; Inuma et al., 2007), but Marchal (1902) and Oku et al. (1985) did not name the formae speciales of B. graminis that infect them.

We describe the state of current knowledge about the f.sp. concept in B. graminis. We discuss the host specificity of the ff.spp. and recent advances in the molecular and cellular basis of adaptation of the fungus to different hosts. We then address the genetic diversity of B. graminis and the phylogenetic relationships among the ff.spp. Finally, we review the co‐evolutionary history of the ff.spp. with their hosts, including their emergence and centre of origin, as well as possible dates of divergence of the ff.spp. The combination of current knowledge about pathogen evolution, host adaptation and genetic diversity indicates that, although groups of isolates within B. graminis can be recognized as being adapted to specific host species or genera, particularly the four main small‐grain cereal crops, the concept of the f.sp. as a taxonomic group within the B. graminis species may no longer be valid.

Host Specificity

The classification of B. graminis into eight ff.spp. based on host specialization (Marchal, 1902; Oku et al., 1985) oversimplifies a complex situation (Brown et al., 2002). Several studies have shown that specialization to the host genus may not be as absolute as the nomenclature suggests. Hardison (1944, 1945) showed that B. graminis ff.spp. can infect a wider host range than the genus from which each isolate was obtained, and a study of wheat powdery mildew in northern China found that 286 wild plant accessions belonging to 40 species in seven genera were susceptible to infection by B. graminis f.sp. tritici (Sheng et al., 1993, 1995). Eshed and Wahl (1970) demonstrated that B. graminis ff.spp. hordei, tritici and avenae, isolated from barley, wheat and oat crops in Israel, were compatible with wild grasses belonging to different genera in several tribes. Significantly, none of the ff.spp. hordei and avenae isolates induced infection on their non‐adapted cereal hosts Avena or Triticum, and Hordeum or Triticum, respectively, although f.sp. tritici showed a compatible interaction with Avena and Hordeum plants. Wyand and Brown (2003) found that isolates of ff.spp. hordei, tritici, secalis and avenae sampled in the UK only infected species from which they had been sampled. Eshed and Wahl (1970) reported that isolates from Israel had a wider host range than those from elsewhere in the world, reflecting the broader genetic diversity of host plants in West Asia.

In research on the recent emergence of powdery mildew on triticale, cross‐infection tests were conducted with isolates sampled from triticale, wheat and rye. Isolates from triticale were also pathogenic on wheat, supporting the hypothesis that triticale mildew emerged through a host range expansion of B. graminis f.sp. tritici (Troch et al., 2012; Walker et al., 2011). Moreover, isolates from wheat and rye were occasionally able to infect triticale (Linde‐Laursen, 1977; Troch et al., 2012; Walker et al., 2011), underscoring the close relationship of this intergeneric hybrid with its parents wheat and rye.

In a study of the host specificity of B. graminis isolates sampled from wild grasses, such isolates were capable of causing infection on many other wild grass species, but were non‐pathogenic on cultivated barley, wheat and oats (Eshed and Wahl, 1975). This contrasts sharply with the ability of isolates sampled from cereal crops in West Asia to infect wild grasses (Eshed and Wahl, 1970).

Altogether, these results indicate that the classification of B. graminis from cereals into ff.spp. is based on adaptation to certain hosts, summarized in Table 1, rather than on strict host specialization as described by Marchal (1902). They also suggest the possibility that wild grasses may play a dual role in epidemics of powdery mildew on cereals. First, wild grasses may act as temporary hosts, allowing the persistence of cereal powdery mildew between seasons. Chasmothecia (the sexual structures formerly known as cleistothecia) preserved on the debris of wild grasses may enable the oversummering of B. graminis ff.spp. hordei, tritici and avenae (Eshed and Wahl, 1975). Second, wild grasses may act as common hosts for different ff.spp. of cereal powdery mildew, providing opportunities for hybridization between them.

Table 1.

General adaptation of Blumeria graminis formae speciales (ff.spp.) to cultivated cereal hosts*

Source hosts (origin of isolates)	Inoculated hosts†
	Oat	Barley	Wheat	Rye	Triticale
B. graminis f.sp. avenae	+++	−	−	−	−
B. graminis f.sp. hordei	−	+++	−	−	−
B. graminis f.sp. tritici	±	±	+++	−	+
B. graminis f.sp. secalis	−	−	±	+++	+
B. graminis f.sp. ‘triticale’	−	−	++	±	+++

*Based on Eshed and Wahl (1970), Troch et al. (2012), Walker et al. (2011) and Wyand and Brown (2003).

†+, infection; –, no infection; more + corresponds to a higher aggressiveness.

Sexual hybridization between B. graminis ff.spp. has been demonstrated experimentally (Hiura, 1965). In particular, crosses between ff.spp. tritici, secalis and agropyri produced fertile chasmothecia, crosses between f.sp. hordei and ff.spp. tritici, secalis or agropyri produced sterile chasmothecia, whereas no chasmothecia were produced when f.sp. avenae was crossed with tritici or hordei (Hiura, 1978). In principle, hybridization may increase the host range and hence disease potential through recombination of avirulence (AVR) genes from the parental ff.spp. However, several studies have indicated that hybrids between ff.spp. have lower pathogenicity than the parental isolates on either parent's host plant (Oku et al., 1986; Tosa, 1989a, b). An hypothesis for the emergence of powdery mildew on the wheat–rye hybrid triticale was that it was a hybrid of ff.spp. tritici and secalis. Combined pathology and molecular phylogenetic studies rejected this hypothesis, however, and revealed that this ‘new’ powdery mildew emerged through a host range expansion of B. graminis f.sp. tritici (Troch et al., 2012; Walker et al., 2011). It is possible that a few modifications in the effector repertoire of f.sp. tritici were sufficient to adapt to triticale, which is closely related to wheat. A similar breakdown of resistance was found for Magnaporthe oryzae f.sp. oryzae, which probably emerged from a host range expansion from a Setaria millet, possibly because the loss of a strain‐specific effector gene allowed the fungus to colonize rice (Couch and Kohn, 2002; Couch et al., 2005).

Host Resistance

The infection process of B. graminis and host resistance have been reviewed by Eichmann and Hückelhoven (2008), Green et al. (2002), Hückelhoven and Panstruga (2011) and Zhang et al. (2005), among others. Here, we focus on ‘non‐adapted’ interactions of cereals with B. graminis ff.spp. Non‐host resistance is usually defined as durable resistance of all known genotypes of a plant species to all known genotypes of a pathogen species (reviewed by Lipka et al., 2008; Mysore and Ryu, 2004; Niks and Marcel, 2009; Nürnberger and Lipka, 2005; Schulze‐Lefert and Panstruga, 2011; Schweizer, 2007). As these studies showed that the subspecific classification of B. graminis into ff.spp. is based on adaptation to certain hosts, rather than strict host specialization, cereals cannot be considered as typical non‐hosts to inappropriate ff.spp.; hence, we prefer to use the term ‘non‐adapted resistance’.

B. graminis and the multilayered plant immune system

In the zigzag model of plant defence, plants recognize the presence of pathogens at various levels (Dodds and Rathjen, 2010; Jones and Dangl, 2006). Initially, the recognition of conserved pathogen‐associated molecular patterns (PAMPs) leads to PAMP‐triggered immunity (PTI). In powdery mildew, early defences, which may be a form of PTI, are expressed in both susceptible and R‐gene‐resistant cultivars, and are associated with the induction of defence‐related genes (Boyd et al., 1996) and the formation of papillae (Zeyen et al., 2002), involving the generation of NO (Prats et al., 2005), H₂O₂ (Thordal‐Christensen et al., 1997), phenolics (Carver et al., 1994) and the deposition of callose below the attempted site of penetration (Zeyen et al., 2002). A papilla may either stop penetration or, if penetration succeeds, become a ‘collar’ for the neck of the haustorium (Zeyen et al., 2002).

In a second layer of defence, the host recognizes specific effector (AVR) proteins. In several pathogens, these proteins suppress PTI and facilitate infection, leading to effector‐triggered immunity (ETI) (Dodds and Rathjen, 2010; Jones and Dangl, 2006). In diseases caused by biotrophs, ETI involves a hypersensitive response (HR), including cell death, preventing the establishment of a biotrophic interaction (Schulze‐Lefert and Panstruga, 2003). Most R genes, including those in cereals for mildew resistance, act against a subset of pathogen races. Two AVR genes with effector activity have been isolated from B. graminis (Nowara et al., 2010; Ridout et al., 2006), but the fungus has many homologues of these genes (Sacristan et al., 2009). This family and other candidate effectors evolve rapidly and their diversity may contribute to host range and parasite speciation (Spanu et al., 2010).

In addition, in barley, recessive loss‐of‐function mlo alleles mediate durable, broad‐spectrum resistance towards otherwise virulent B. graminis f.sp. hordei (Jørgensen, 1994). This resistance is associated with papilla formation and H₂O₂ accumulation and stops the fungus at the pre‐penetration stage (Hückelhoven et al., 1999). Wild‐type Mlo genes have been identified as negative regulators of PTI, required for susceptibility (Buschges et al., 1997; Consonni et al., 2006). Homologues of Mlo in wheat have a similar function to the barley gene (Elliott et al., 2002; Konishi et al., 2010). Several lines of evidence point to a functional link between mlo‐mediated resistance in barley and non‐host resistance (Humphry et al., 2006; Schweizer, 2007; Trujillo et al., 2004; Zellerhoff et al., 2010).

Interactions of cereals with non‐adapted ff.spp.

Inoculation of non‐adapted B. graminis ff.spp. on different cereals has been employed to discover which resistance mechanisms form the basis of non‐adapted resistance. Much of the research has investigated f.sp. tritici on barley.

Tosa et al. (1990) observed the incidence of papillae and cell wall penetration to infer compatibility between ff.spp. and non‐adapted hosts. On the basis of the low incidence of papillae, ff.spp. tritici, secalis and agropyri were found to be more compatible with any accession of wheat, rye or wheatgrass than with barley. However, hordei induced a low incidence of papillae on each of the four hosts. These four ff.spp. had low compatibility with oat, although hordei was the least incompatible, as indicated by the lower incidence of papillae. These responses suggest involvement of a form of PTI in non‐adapted resistance.

Tosa et al. (1990) found that ff.spp. agropyri and secalis induced similar cellular responses in wheat leaves as the appropriate f.sp. tritici. The resistance of wheat to these ff.spp. was attributable to HR, a response conditioned by gene‐for‐gene interactions in cultivar specificity. Moreover, Tosa and Tada (1990) and Matsumura and Tosa (1995), who crossed isolates of these ff.spp. on common hosts and analysed the segregation of virulent and avirulent progeny, concluded that ff.spp. agropyri and secalis carry AVR genes corresponding to wheat genes for resistance to f.sp. tritici.

Non‐adapted resistance of cereals to B. graminis ff.spp. is associated with penetration resistance caused by papilla formation, in some cases supplemented by single‐cell HR. Blumeria graminis f.sp. tritici develops more slowly than f.sp. hordei, and this difference in timing is reflected in the expression of barley genes involved in defence (Boyd et al., 1996). Olesen et al. (2003) found that, when barley, wheat or oat leaf epidermal cells were attacked by non‐adapted ff.spp., penetration attempts failed in association with papilla deposition by epidermal cells. Attacked cells died or, if visible haustoria were formed, the plant cell died soon afterwards. In addition, when epidermal cells were first attacked by the adapted f.sp., as inducer, and later by the non‐adapted ff.spp., as challenger, there was a dramatic suppression of defence responses within epidermal cells containing the inducer haustorium. This allowed most challenger attacks to form a haustorium which functioned to support colony development of the non‐adapted f.sp. This implies that non‐adapted isolates are capable of establishing a fully compatible infection if PTI and ETI are suppressed through previous infection by an adapted isolate of the appropriate f.sp.

Non‐adapted resistance of barley to f.sp. tritici is associated with penetration resistance by H₂O₂ accumulation in papilla and HR of attacked cells (Hückelhoven et al., 2001). Enhanced expression of barley BAX inhibitor‐1 suppresses penetration resistance to the non‐adapted f.sp. tritici (Eichmann et al., 2004). This links barley non‐adapted resistance to the induction of cell death because BAX inhibitor‐1 has a role in modulating cell wall‐associated defence and establishing full compatibility of f.sp. hordei with barley (Eichmann et al., 2010).

In a microscopic analysis of barley defences against B. graminis f.sp. tritici, non‐adapted resistance was associated with several mechanisms of cellular defence expressed at distinct stages of fungal development (Trujillo et al., 2004). Non‐adapted resistance of barley to f.sp. tritici is established not only by penetration resistance caused by papillae and early HR, but also by post‐penetration resistance, leading to late hypersensitive cell death or restricted fungal development. Using transient‐induced gene silencing (TIGS), a role of the t‐SNARE protein HvSNAP34 in race‐nonspecific resistance of barley, mediated by mlo, was observed, with a proposed function in papilla formation (Douchkov et al., 2005). ETI, however, was demonstrated to be independent of HvSNAP34 function and associated with HR instead of papilla formation. Miklis et al. (2007) showed a critical role for actin cytoskeleton function in mlo‐mediated and non‐adapted resistance of barley, which was highly effective against adapted and non‐adapted powdery mildew species, respectively. These data show that barley defence against the non‐adapted f.sp. tritici is determined by mechanisms similar to those involved in race‐specific and race‐nonspecific resistance to the adapted f.sp. hordei.

Aghnoum and Niks (2010) developed barley lines with substantial susceptibility to B. graminis f.sp. tritici at the seedling stage. Analysis of the infection of these lines by non‐adapted ff.spp. showed that: (i) this resistance is quantitative, as indicated by differences in haustorium formation and colony development among accessions; (ii) genes for basal resistance have effects specific to pathogen developmental stages; and (iii) these genes are also specific to different B. graminis ff.spp. Zellerhoff et al. (2010) described the transcriptional responses of a near‐isogenic pair of barley lines differing in the allelic status of the Mlo gene to three pairs of host/non‐host pathogens causing powdery mildew, rust and blast on cereals. Results suggested that non‐adapted and mlo‐mediated defences of barley are functionally related. Moreover, non‐host resistance to different fungi is associated with more robust regulation of complex, but largely non‐overlapping, sets of pathogen‐responsive genes involved in similar metabolic or signalling pathways.

Altogether, these studies indicate that the resistance of cereals to non‐adapted ff.spp. of B. graminis involves both PTI and ETI, and that there is no clear distinction between mechanisms of resistance to adapted and non‐adapted ff.spp. Schweizer (2007) concluded that non‐adapted resistance in barley depends on non‐compromised, PAMP‐triggered basal defence, which implies that fungal effectors are largely ineffective against non‐adapted hosts. Niks and Marcel (2009) argued that non‐adapted resistance and basal resistance to adapted pathogens may rest on the same or similar mechanisms, and that effective suppression of PTI requires specific effectors. This second argument underscores the relevance of studies which found that B. graminis ff.spp. agropyri and secalis carry AVR genes corresponding to wheat genes for resistance to the adapted f.sp. tritici (Matsumura and Tosa, 1995; Tosa and Tada, 1990; Tosa et al., 1990). Schulze‐Lefert and Panstruga (2011) proposed that both PTI and ETI contribute to non‐host resistance such that, with increasing phylogenetic divergence time between two plant species, the relative effectiveness of PTI increases, whereas the relative contribution of ETI decreases. This hypothesis is supported by the studies discussed above because the resistance of cereals to infection by closely related ff.spp., such as wheat to ff.spp. agropyri or secalis, is based largely on HR, so that the interactions are compatible if AVR activity is lost, whereas infection by more distantly related ff.spp., such as f.sp. tritici on barley, is stopped by papilla formation. More diverse cytological and molecular studies, based on interactions other than B. graminis f.sp. tritici on barley, should be carried out to understand non‐adapted resistance in more depth.

Phylogeny

Several studies have tried to elucidate the phylogenetic relationships among the ff.spp. of B. graminis. Nevertheless, the subspecific phylogeny of B. graminis is not yet resolved, particularly the relationship of ff.spp. from crops to isolates from wild grasses. Phylogenetic studies of rDNA internal transcribed spacer (ITS), β‐tubulin, chitin synthase 1 and 28S rDNA (Inuma et al., 2007; Wyand, 2001; Wyand and Brown, 2003) and of AVR‐effector gene homologues (Sacristan et al., 2009), analysis of the incidence of cell wall penetration (Tosa et al., 1990) and data on reproductive isolation (Hiura, 1978) have all concluded that ff.spp. tritici, secalis and agropyri are more closely related to each other than to hordei or avenae (Fig.  2). Tosa et al. (1990) suggested that hordei is closer than the ‘tritici clade’ to both the ancestral form of B. graminis and avenae (Fig. 2A). The latter proposal was supported by a phylogeny of β‐tubulin genes (Wyand and Brown, 2003). This partially reflects host classification, because Triticum, Secale and Agropyron are more closely related to each other than to Hordeum within the tribe Triticeae. Avena, however, is in a different supertribe of Poaceae, the Poodae rather than Triticodae.

Figure 2.

Existing subspecific phylogenies of Blumeria graminis. (A) Diagram summarizing the conclusions of Tosa et al. (1990) on the relationships between formae speciales of B. graminis based on cytological evidence. The close relationship of ff.spp. tritici, secalis and agropyri is supported by molecular phylogenetic evidence (Inuma et al., 2007; Wyand, 2001; Wyand and Brown, 2003). (B) Summary of the maximum parsimony phylogeny of Inuma et al. (2007), estimated from sequences of four genomic regions. Host genera of isolates are indicated. Note that isolates from wild Avena are not closely related to those from crops. (C) Maximum parsimony phylogeny of B. graminis isolates, estimated from rDNA internal transcribed spacer (ITS) sequences and rooted with reference to Podosphaera pannosa and Cystotheca lanestris (Wyand, 2001). Numbers of isolates sampled from each host species are in parentheses; all isolates from the same host species had identical sequences (GenBank accession numbers: B. graminis, AJ313137–AJ313148; C. lanestris, AF011289 and AB000933; P. pannosa, AF011322). Bootstrap frequencies are shown for relevant clades. Note that isolates from Avena sativa lie within the clade containing isolates from other crops. Host classification in (B) and (C). All except Diarrhena: subfamily Pooideae. Black type, supertribe Triticodae; Roman, tribe Triticeae; italics, tribe Bromeae. White‐on‐black, supertribe Poodae; Roman, tribe Aveneae; italics, tribe Poeae. Diarrhena: subfamily Bambusoideae, supertribe Oryzodae, tribe Diarrheneae.

Contrasting conclusions have been published about the relationship of B. graminis on crops and wild grasses. A multilocus phylogenetic analysis revealed nine clades, generally reflecting the phylogeny of host genera (Inuma et al., 2007; Fig. 2B). In most cases, isolates from each host genus were monophyletic, but f.sp. bromi and isolates from Lolium were polyphyletic. Isolates from Avena were not closely related to those from crops, ff.spp. tritici, secalis and hordei; one of these isolates was from wild A. fatua and one from an unknown Avena species. An rDNA‐ITS phylogeny placed an isolate from cultivated A. sativa in Switzerland in the same clade as the two other Avena isolates (fig. 2 in Inuma et al., 2007). By contrast, another ITS phylogeny placed f.sp. avenae from A. sativa in the UK within a clade of isolates from crops, including wheat, barley, rye, L. perenne and the wild grass E. repens; ff.spp. bromi and poae were polyphyletic (Wyand, 2001; Fig. 2C). In another study, f.sp. avenae from the UK and Finland had identical ITS and β‐tubulin sequences and were related to hordei, secalis and tritici (Wyand and Brown, 2003). Neither of the subspecific phylogenies of B. graminis (Inuma et al., 2007; Wyand, 2001) reflected the classification of host genera or even tribes (Fig. 2). These results imply that isolates of B. graminis from cereal crops fall into a single clade representing a small fraction of the phylogenetic diversity of the species as a whole, that ff.spp. of B. graminis from most wild grasses are invalid because they are polyphyletic, and that f.sp. agropyri may be an exception to this situation as a group within the clade of B. graminis from crops.

To clarify these relationships, a phylogenetic analysis of 24 isolates belonging to the eight recognized B. graminis ff.spp. was conducted using β‐tubulin gene sequences retrieved from GenBank (Inuma et al., 2007; Troch et al., 2012; Wyand and Brown, 2003). The question of particular interest is the relationship of ff.spp. on crops to isolates from wild grasses. Although only one gene was analysed, it contained 379 characters, 98 of which were parsimony informative. The 50% majority rule consensus tree (Fig. 3) underscores the close relatedness of powdery mildew on wheat, rye, triticale and Agropyron, all belonging to one clade. A second clade containing f.sp. hordei was closely related to this ‘Triticum clade’. An isolate from A. sativa (cultivated oats) formed a sister group of the hordei clade, whereas isolates from other Avena spp. (wild oats) clustered in a separate clade. This implies that the classification of isolates from all Avena spp. into f.sp. avenae is invalid. It would be desirable to sample B. graminis isolates from more Avena species, cultivated and wild, to clarify the phylogeny of isolates classified in f.sp. avenae. Isolates from Bromus were placed in three clades, whereas isolates from Poa and Dactylis each belonged to one clade, as in Inuma et al. (2007).

Figure 3.

Phylogeny of Blumeria graminis formae speciales inferred by Bayesian analysis from the β‐tubulin gene. The 50% majority rule consensus tree is shown. Maximum likelihood analysis recovered the same topology. Numbers above the branches are posterior probabilities. Branch lengths are proportional to the number of character changes. Labels on the phylogeny are hosts from which isolates were sampled and GenBank accession numbers. Coloured areas indicate genera from which isolates were obtained: red, Dactylis; dark blue, Avena; green, Bromus; yellow, Triticum; light blue, Secale; white, Triticale; orange, Agropyron; purple, Hordeum; grey, Poa.

Altogether, these molecular evolutionary data suggest that the taxonomic classification of Blumeria graminis into ff.spp. is not consistent with the phylogeny of the fungus. In the cladistic approach to biological classification, only monophyletic groups are recognized (Hörandl, 2006), and the nomenclature of ff.spp. is not governed by the International Code of Nomenclature for algae, fungi and plants (McNeil et al., 2012). More comprehensive sequencing of genes with greater phylogenetic signal of isolates of all ff.spp., and from wild grasses, is required to clarify the subspecific phylogenetic relationships within B. graminis. This will lead to a better understanding of the origin and epidemiology of B. graminis, with implications for disease control.

Origin of the ff.spp.

Co‐evolutionary history

Genetic and archaeological evidence indicates that the transformation of cereals from wild forms into domesticated crops originated in the Fertile Crescent at 10 000–12 000 bp (Balter, 2007; Feuillet et al., 2008; Salamini et al., 2002). Despite the lack of fossil records of most fungal plant pathogens, their evolutionary history can be reconstructed using tools based on population genetics, coalescence methods and phylogeography (Grünwald and Goss, 2011; Stukenbrock and McDonald, 2008). Blumeria graminis is an obligate biotroph and, as such, the evolution of the pathogen is influenced by the dispersal of its various host species. Several studies have addressed the co‐evolutionary history of B. graminis and its hosts. Two main categories of event may explain the highly specialized associations between Blumeria graminis ff.spp. and their hosts: (i) co‐speciation, in which two ff.spp. have separated in response to the speciation of the host, with which the fungus remains closely associated; and (ii) separation through host shifts or jumps, where the f.sp. switches from the ancestral host to a new species, closely or distantly related, respectively. Uncertainties in the subspecific classification of B. graminis focus on the time of divergence of ff.spp. and the specificity of adaptation to host genera.

Wyand and Brown (2003) sequenced the rDNA ITS region and the β‐tubulin gene of ff.spp. tritici, secalis, hordei and avenae, resulting in dendrograms with different topologies. As no topology was congruent with the phylogeny of host genera, they rejected the hypothesis that the ff.spp. co‐evolved with their host. Instead, they suggested that the divergence between ff.spp. of B. graminis might reflect rapid evolution in agriculture, possibly beginning in the early Holocene epoch.

Matsuda and Takamatsu (2003) sequenced the 28S rDNA gene of ff.spp. hordei and tritici. They estimated that these ff.spp. diverged approximately 11 million years ago, in sharp contrast with the proposal of Wyand and Brown (2003). This divergence time was calculated in three steps: (i) based on the molecular clock of the rbcL (ribulose‐bisphosphate carboxylase) gene in the sunflower alliance of plant families (Bremer and Gustafsson, 1997), they calculated the divergence time of the tribe Cardueae of the Asteraceae; (ii) from this divergence time, the nucleotide substitution rates of the D1 and D2 regions of 28S rDNA of Golovinomyces species (Erysiphaceae: Golovinomyceteae), a parasite that co‐speciates with host tribes in the Asteraceae, were estimated; and (iii) from this molecular clock, the divergence times of different nodes of the Erysiphales were calculated. As the hosts wheat (Triticum aestivum) and barley (Hordeum vulgare) diverged about 11.6 million years ago from their last common ancestor (Chalupska et al., 2008), this divergence time indicates co‐evolution of the ff.spp. with their hosts. This type of co‐evolution is co‐speciation, where host and pathogen diverge simultaneously, in contrast with host tracking, where the pathogen is likely to be younger than the host (Stukenbrock and McDonald, 2008).

Inuma et al. (2007) conducted a phylogenetic analysis of B. graminis based on the sequences of four DNA regions (ITS, 28S rDNA, chitin synthase 1 and β‐tubulin). The topology of the pathogen phylogeny was shown not to be congruent with that of their host plants (Fig. 2), and the hypothesis of co‐speciation between ff.spp. and their hosts was therefore not supported. Instead, the authors suggested that B. graminis expanded its host range together with host taxa, and that host jumping to phylogenetically distinct hosts occurred several times during the evolution of the fungus. In addition, they estimated the split between the Hordeum and Triticum clade to be 4.6 million years ago, applying Takamatsu and Matsuda's (2004) mutation rate of rDNA ITS. Again, this mutation rate was calculated from the molecular clock of rbcL in the sunflower alliance (Bremer and Gustafsson, 1997).

Oberhaensli et al. (2011) sequenced 273.3 kb of B. graminis f.sp. tritici BAC sequences and compared them with orthologous regions in the B .graminis f.sp. hordei genome. Protein‐coding regions were collinear and well conserved, but the intergenic regions showed very low conservation, mostly because of different integration patterns of transposable elements. They estimated a divergence time between tritici and hordei of 10 million years ago, based on a total of 13 kb of conserved intergenic sequence, including orthologous transposable elements and non‐coding sequences, using the same substitution rate as that of long terminal repeat (LTR)‐retrotransposons in rice nuclear genomes (Ma and Bennetzen, 2004). This divergence time, approximately two million years after the hosts wheat and barley, suggested co‐evolution of the ff.spp. with their hosts, possibly after a short phase of host expansion when the same ancestral pathogen species could grow on both hosts.

Divergence time of B. graminis ff.spp.

The wide range of estimated times of divergence between the different ff.spp. (Table 2) reflects two major challenges in the accurate dating of divergence times in plant pathogens (Grünwald and Goss, 2011). First, dating events in years is dependent on the knowledge of fairly accurate estimates of mutation rates per year or per generation, and also of the generation time, but these are not well understood for many plant pathogens. Inuma et al. (2007), Oberhaensli et al. (2011) and Takamatsu and Matsuda (2004) all calculated divergence times on the basis of substitution rates in plant DNA. A major objection to the use of these rates for powdery mildew fungi is that, most unusually for a eukaryote, all reproductive cells of almost all of these species are directly exposed to sunlight. Although their haustoria are embedded within the leaf (but, even then, only within epidermal cells, which lack chlorophyll), the mycelium from which conidiophores and thence conidia are formed, as well as sexual structures, are entirely superficial on the leaf surface. All nuclei which undergo mitosis or meiosis are therefore directly exposed to mutagens, including ultraviolet light, cosmic rays, ionizing radiation and chemicals, to an extent perhaps unmatched by any other group of eukaryotes. It is possible, therefore, that the rate of mutation in these fungi might be unusually high; this could be tested, in principle, by high‐throughput sequencing of mildew colonies exposed to daylight in natural conditions. If this hypothesis is correct, the use of plant mutation rates to calibrate phylogenies of powdery mildew fungi, including ff.spp. of B. graminis, will overestimate divergence times considerably.

Table 2.

Overview of divergence time estimates between B lumeria graminis formae speciales (ff.spp.) hordei and tritici, including the DNA sequences on which they were based and the calculation method

Estimated divergence time	DNA sequences	Calculation method	Reference
14 000 years ago	rDNA ITS, tub2	Suggestion, no calculation	Wyand and Brown (2003)
11 million years ago	28S rDNA	Based on the molecular clock of the rbcL gene for the sunflower alliance (Bremer and Gustafsson, 1997)	Matsuda and Takamatsu (2003)
4.6 million years ago	rDNA ITS	Based on the molecular clock of the rbcL gene for the sunflower alliance (Bremer and Gustafsson, 1997)	Inuma et al. (2007)
10 million years ago	Conserved intergenic sequences including orthologous transposable elements	Based on the substitution rate of LTR‐retrotransposons in rice nuclear genomes (Ma and Bennetzen, 2004)	Oberhaensli et al. (2011)

ITS, internal transcribed spacer; LTR, long terminal repeat; rbcL, ribulose‐bisphosphate carboxylase gene.

Moreover, mutation rates can vary greatly in different parts of the genome. The rbcL gene used to calibrate rates of DNA sequence evolution (Bremer and Gustafsson, 1997) encodes an important part of the metabolic mechanism of plants and thus may be expected to evolve slowly. By contrast, the intergenic sequences studied by Oberhaensli et al. (2011) are non‐coding and are therefore expected to evolve much more rapidly than sequences of metabolic genes.

The generation time of B. graminis is not unusual for a fungus and is unlikely to have greatly affected its rate of molecular evolution. Blumeria graminis undergoes one sexual and several asexual generations each year (Koltin and Kenneth, 1970).

The second challenge is that plant pathogens may lack genetic variation in introduced populations, which may be the only known populations (Grünwald and Goss, 2011). With little polymorphism, population genetic parameters can be difficult to estimate. Indeed, accurate reconstruction of the timing and order of speciation depends critically on obtaining suitable characters for phylogenetic analyses (Brito and Edwards, 2009). This was not the case in the studies mentioned above, which all used sequences with imprecise phylogenetic signals, such as the ITS region, to estimate a divergence time for B. graminis f.sp. hordei and tritici. When sufficiently polymorphic population data for these ff.spp. become available, the accuracy of the estimated divergence times could be improved.

For all of these reasons, it is possible that the divergence times of B. graminis ff.spp. in the Miocene epoch (Inuma et al., 2007; Matsuda and Takamatsu, 2003; Oberhaensli et al., 2011) may have been considerably overestimated, and may not date back millions of years to when their hosts diverged. The existing evidence does not conclusively disprove the contention of Wyand and Brown (2003) that the ff.spp. of B. graminis, which attack cultivated cereal species, may have diverged early in the pre‐history of agriculture in the Holocene. More data are required to reach a firm conclusion one way or the other. The genome of f.sp. hordei has been sequenced and annotated (Spanu et al., 2010), and the complete f.sp. tritici genome will soon be available (Parlange et al., 2011). These data may go some way to resolving the question of the divergence of ff.spp. of B. graminis.

Centre of origin

If a parasite evolves by host tracking, it is expected that the centre of origin of the pathogen will correspond to that of the host (Stukenbrock and McDonald, 2008). It was further suggested that the centre of diversity of B. graminis may coincide with the centre of origin of the hosts in West Asia (Wyand and Brown, 2003). Phylogeographical analysis of β‐tubulin and translation elongation factor 1‐α of B. graminis revealed that Israeli isolates formed diverse haplotypes, mostly near the centre of the evolutionary network, consistent with the hypothesis that this region contains the ancestral population and the centre of diversity of B. graminis (Troch et al., 2012). Parks et al. (2009) investigated the global evolutionary history of B. graminis f.sp. tritici by comparing 12 single‐nucleotide polymorphisms in the USA, UK and Israeli populations. Although they concluded that USA isolates are direct descendants of the ‘Old World’ populations from the UK and Israel, it was not possible to test whether the centre of diversity corresponded to the centre of origin of the pathogen. It may be possible to infer the evolutionary history of B. graminis by coalescence analysis of isolates from wild grasses and from cultivated cereals collected both in the Middle East, the proposed centre of origin, and elsewhere. The possibility that host shifts and jumps may have confused the evolutionary relationship between B. graminis and its host species must be considered in any analysis.

Recent emergence of B. graminis on triticale

Triticale (× Triticosecale Wittmack) is an intergeneric hybrid between wheat (Triticum ssp.) and rye (Secale cereale). In contrast with wheat and rye, this artificial cereal is of very recent origin and was commercialized at the end of the 1960s (Ammar et al., 2004; Oettler, 2005). During the last decade, triticale has gained considerable importance in Europe, as its production area has nearly doubled since 2000, to almost 3.3 Mha in 2010 (FAOSTAT). With the expansion of the growing area, however, powdery mildew has emerged and become a significant disease of triticale. Combined pathology and phylogenetic research demonstrated that this pathogen emerged through host range expansion of B. graminis f.sp. tritici from wheat (Troch et al., 2012; Walker et al., 2011). This implies that B. graminis has evolved the capacity to colonize a new host, triticale, which is closely related to its long‐standing host, wheat. Multilocus phylogeographical analysis revealed that this expansion occurred recently and at several times at different locations in Europe (Troch et al., 2012). Other recent examples of host range expansions of pathogens to triticale include the spread of rice blast disease to wheat and triticale in Brazil (Mehta, 1998) and of scald (Rhynchosporium secalis) from rye to triticale (Zaffarano et al., 2008). This host range expansion may eventually result in host shift speciation (Troch et al., 2012), so that the population specialized to triticale will become incapable of infecting wheat, with the cessation of gene flow between the populations infecting the two crop species (Giraud et al., 2010). A case of disease emergence consistent with ecological speciation by host shifts in prehistoric times was detected by molecular evolutionary analysis of the wheat fungal pathogen Zymoseptoria tritici (formerly Mycosphaerella graminicola; Stukenbrock et al., 2007).

Conclusions

Current knowledge about the f.sp. concept in B. graminis points to the need for research on the divergence times of the ff.spp., host specificity, host resistance and phylogeny. Imprecise estimates of mutation rates may account for much of the uncertainty about the divergence times of the subspecific taxa of B. graminis.

If the Miocene divergence hypothesis of Inuma et al. (2007), Matsuda and Takamatsu (2003) and Oberhaensli et al. (2011) is correct, in contrast with the Holocene divergence hypothesis (Wyand and Brown, 2003), an explanation is required for the existence of four related clades of B. graminis which are highly specialized to the four main cereal crops which originated in the Fertile Crescent (rye, barley, wheat and oats), recognized as ff.spp. secalis, hordei, tritici and avenae. This contrasts with a lack of similar specialization of B. graminis on wild grasses (Eshed and Wahl, 1975; Inuma et al., 2007; Wyand, 2001). Several observations are consistent with the Holocene hypothesis, but are harder to interpret with the Miocene hypothesis.

1. There is one clade of B. graminis pathogenic on cereal crop species. In the Holocene hypothesis, this could have originated from a regional population of the fungus in an area in which early agriculture was practised.

2. Blumeria graminis ff.spp. from cereals are highly specialized to their host species, but this is not the case for isolates from wild grasses, or even for isolates sampled from wild plants but classified into one of the ff.spp. pathogenic on crops. This specialization could have evolved during the course of agriculture, whereas, if it evolved during the previous 10 Myr, an explanation is required for why it evolved in clades that infected the grass species which, at a much later time, humans cultivated for food, but not for those that infected other species.

3. There are at least two clades of f.sp. avenae, one pathogenic only on cultivated oats and another on wild oats, even though A. sativa is cross‐fertile with A. fatua and is sometimes classified as a subspecies of it. The ‘cultivated’ f.sp. avenae could have evolved from a local population at an early agricultural site, as in the Holocene hypothesis, whereas the Miocene hypothesis requires the evolution of two clades, one a weakly specialized clade pathogenic on wild A. fatua and the other a much more strongly specialized clade pathogenic on genotypes of A. fatua, which became cultivated by humans as A. sativa, and was the sister group of clades pathogenic on cultivated (but not wild) grasses in the Triticeae.

4. The ff.spp. bromi and poae are polyphyletic, implying that the classification of B. graminis into ff.spp. is not consistent with the phylogeny of the fungus, and that these ff.spp. are therefore not valid taxa.

5. Similar defences operate against isolates of adapted and non‐adapted ff.spp., with little divergence in PTI mechanisms and effective ETI against non‐adapted ff.spp. It is reasonable to suggest that this is consistent with fairly recent divergence of the ff.spp.

6. Finally, against this, the Holocene hypothesis is not consistent with the calculated divergence times of ff.spp. pathogenic on crop species, although these time points in the Miocene were calculated from rates of variation of functional genes in plants, rather than non‐functional sequences in a fungus, which is exposed to high levels of mutagenic radiation throughout its life cycle.

To conclude, we propose that the f.sp. concept should be retained for B. graminis pathogenic on cultivated cereals, including clades which are strongly specialized to these hosts, namely ff.spp. hordei, secalis and tritici, as well as avenae from cultivated A. sativa; f.sp. agropyri may be valid as a distinct subclade related to ff.spp. tritici and secalis. We propose that the f.sp. concept should no longer be applied to B. graminis from other wild grasses, including wild Avena spp.; instead, the host species of origin can be mentioned when required when referring to the source of an isolate.

Recent advances in DNA analysis, including whole genome sequencing of B. graminis f.sp. hordei and tritici and other isolates, coalescence methods and cell biology of pathogenicity and host resistance, promise to advance our understanding of the evolution of host–pathogen specialization in cereal powdery mildew. This will hopefully resolve some of the uncertainties in the subspecific taxonomy of B. graminis.