The biotrophic fungus Puccinia striiformis f.sp. tritici, a basidiomycete that causes yellow rust on wheat, is spread by wind-dispersed spores. Analysis of amplified fragment length polymorphism (AFLP) variation showed that the fungus frequently migrates between the UK, Germany, France and Denmark. There is no biological evidence for sexual or parasexual reproduction under natural conditions, and this was supported by the lack of recombination, as revealed by AFLP, over the time and area represented by the samples in this study. A phylogeographic analysis revealed that there was effectively a single, clonal population in the four countries, up to 1700 km apart, consistent with a ‘continent-island’ model in which Denmark is the recipient of migrants from other countries. In five cases, specific pathogen clones were dispersed between the UK and Denmark, and on at least two recent occasions clones were also spread from the UK to Germany and France, causing outbreaks of yellow rust on wheat cultivars that were previously resistant to the disease in these countries. The agronomic consequences of migration were enhanced because of the limited genetic diversity for yellow rust resistance in wheat cultivars in the area. These results demonstrate that long-distance migration of pathogen clones, coupled with low diversity in the host species, may cause previously useful resistance genes to become ineffective for disease control on a continental scale.
Spores of fungal plant pathogens may be spread by wind over long distances, either in a single step or in several successive steps, thereby enhancing disease epidemics across large areas, e.g. wheat yellow rust (Zadoks, 1961) and coffee leaf rust (Bowden et al., 1971), wheat stem rust (Roelfs, 1986), barley stripe rust (Dubin & Stubbs, 1986) and potato late blight (Goodwin et al., 1992). The emergence of new molecular techniques for fingerprinting individual organisms, such as fungal isolates, has greatly increased the possibility of detecting such migration in populations of plant pathogens (Brown, 1996; Milgroom, 1997).
Wheat yellow rust is caused by the basidiomycete fungus Puccinia striiformis f.sp. tritici. There is no evidence that the fungus, which is biotrophic, completes a sexual reproductive cycle, as no alternate host has been identified, unlike many other cereal rusts (Stubbs, 1985). It is spread by dikaryotic urediniospores, and there are many indications that the spores are dispersed over wide areas. In Europe, live spores of the fungi that cause powdery mildew and brown rust of barley and wheat were wind-transported for at least 600 km across the North Sea, between the UK and Denmark (Hermansen et al., 1978). Transport of spores that cause yellow rust of wheat was not demonstrated directly, but the epidemics of wheat yellow rust on wheat in Denmark in 1971 and 1972 may have been initiated by spores blown across the North Sea (Hermansen & Stapel, 1972).
The diversity of races or pathotypes, common concepts in studies involving gene-for-gene interactions between host and pathogen, is often low in P. striiformis f.sp. tritici populations, although it may vary considerably between region, year and host cultivar (Stubbs, 1985; Hovmøller, 2001). The mechanisms by which new variation is created in P. striiformis are not fully understood, but mutation from avirulence to virulence is generally considered to occur fairly frequently (Wellings & McIntosh, 1990). Formation of new pathotypes by reassortment of whole nuclei has been observed in experiments where host plants were inoculated with two different pathotypes (Little & Manners, 1969a; Wright & Lennard, 1980). In another experiment, fusion of germ tubes of urediniospores was observed in vitro, but no sign of parasexual recombination was observed (Little & Manners, 1969b). Neither process is known to occur in the natural P. striiformis f.sp. tritici population. Genetic diversity in P. striiformis f.sp. tritici at the isozyme level as well as the molecular level is also low, compared to that of other pathogenic fungi of agricultural crops (Newton & Caten, 1985; Dickinson et al., 1990; Justesen et al., 2002). However, the low genetic diversity does not prevent the fungus from rapidly evolving new pathotypes, which enable yellow rust to develop into epidemics on previously resistant wheat cultivars (Stubbs, 1988; Wellings & McIntosh, 1990; Bayles et al., 1997; Hovmøller, 2001).
In the work reported here, long-distance migration of yellow rust spores in north-west Europe was investigated. First, to test whether the population of P. striiformis f.sp. tritici was indeed fully asexual, evidence for recombination was examined in the pathogen population within the time and area represented by the samples. Having shown that P. striiformis is strictly clonal, the hypothesis tested was whether new pathotypes had evolved independently by mutation in existing clonal lineages within four north-west European countries: the UK, Denmark, France and Germany. Migration of clones between the four countries was investigated as an alternative hypothesis. The analyses were performed by statistical tests of phylogenies (Burt et al., 1996; Templeton, 1998) based on AFLP markers. The agronomic consequences of long-distance migration are illustrated by five specific cases where particular clones, which shared pathotype and AFLP phenotype, resulted in resistance genes or combinations thereof, becoming ineffective for the control of yellow rust in four north-west European countries in recent years.
Materials and methods
Danish samples of P. striiformis f.sp. tritici, chosen based on the results of Justesen et al. (2002), are listed in Table 1(a). In order to investigate migration of P. striiformis within north-west Europe, additional samples from the UK, France and Germany were obtained. Each sample had been virulence-assayed according to Bayles et al. (2000) to determine the pathotype (Table 1(b); Fig. 1). The 12 isolates from the UK represented eight pathotypes that were detected in the previous Danish study (Justesen et al., 2002) and one unique pathotype. The six French and six German isolates represented all known Yr17-virulent pathotypes from these countries in 1997–98.
Table 1. Origin, pathotype and amplified fragment-length polymorphism (AFLP) group of isolates of Puccinia striiformis f.sp. tritici. (a) Danish isolates of Puccinia striiformis f.sp. tritici selected from Justesen et al. (2002); (b) British, French and German isolates collected between 1988 and 1998.
All isolates possessed virulence for Yr3, the resistance in Strubes Dickkopf (SD); and avirulence for Yr5, Yr7, Yr8, Yr10, the resistance in Spaldings Prolific (SP) (Bayles & Stigwood, 1996; Bayles et al., 1997; Justesen et al., 2001). Polymorphic loci are indicated by the following pathotype codes: virulence for Yr1 = 1; for Yr2 = 2; for Yr3b +Yr4b= 4; for Yr6 = 6; for Yr9 = 9; for Yr17 = ; for Carstens V = [CV]. Avirulence for these resistances is shown by ‘–’.
A DNA-extraction procedure appropriate for fungal spore samples of small size was used (Justesen et al., 2002). Polymorphic AFLP fragments were scored as binary characters for each isolate, whereas monomorphic fragments across all isolates were not included. Throughout the text, AFLP phenotypes are equivalent to marker phenotypes based on these binary characters.
Analysis of AFLP data
An unrooted tree of the isolates was estimated from AFLP data. The set of most parsimonious trees (where the criterion was to find the tree with the minimum number of changes) was found using the MIX program in phylip (Phylogeny Inference Package version 3·5c., J. Felsenstein, University of Washington, Seattle, USA). The majority-rule consensus tree, which consists of all groups that occur in more than 50% of the trees investigated, was determined by the CONSENSE program. Branch lengths for this tree were then estimated by the least-squares method, using the FITCH program.
A randomization method (Burt et al., 1996) was used to test whether or not recombination had occurred in the P. striiformis f.sp. tritici population. Isolates representing each combination of pathotype and AFLP phenotype observed were selected to avoid duplication of apparently identical phenotypes. For each of the observed polymorphic AFLP markers, data were randomized among AFLP phenotypes using the phylip program SEQBOOT. The length of the most parsimonious tree was then found for each of 1000 randomized data sets. The lengths of trees constructed from data randomized in this way are estimates of the lengths of trees of individuals sampled from a freely recombining population. In the randomization test, the null hypothesis is that the population sampled has undergone recombination. If the length of the observed tree is significantly shorter than the mean of those estimated from randomized data sets, the null hypothesis can be rejected in favour of the alternative hypothesis – that the population is clonal and that the observed variation is the result of mutation.
A phylogeographic analysis, using a randomization test, was used to test the divergence of the P. striiformis populations in the UK, Denmark, France and Germany. The test was a simplified form of one described by Templeton (1998), which tests the correlation between phylogenetic distance and geographic distance. The lengths of the most parsimonious trees were found for each country separately, using MIX. The sum of the lengths of these four trees was Ya. The data set used in the randomization test included all AFLP phenotypes observed in the four countries; AFLP phenotypes found in both the UK and Denmark, or in all four countries, were represented twice or four times, as appropriate. All AFLP phenotypes in this set were randomized among the four countries. Again, lengths of the most parsimonious trees were found for each country separately; the sum of the lengths of these trees was Yr. This was done for 1000 randomizations. The Yr values were estimates of the summed tree lengths under the null hypothesis that there is a single population of P. striiformis f.sp. tritici in the four countries, with free migration between these countries. If migration were restricted, so that isolates from each country tended to occupy different parts of the tree, the sum of the actual tree lengths (Ya) for the four countries would be shorter than the sum of the lengths for the randomized data sets (Yr). If Ya was significantly shorter than Yr, the null hypothesis that there is effectively a single population in the four countries could be rejected.
A total of 28 polymorphic AFLP bands were detected. Twenty-four of these were phylogenetically informative in that they were shared by two or more isolates and lacking from two or more other isolates. The 28 markers revealed 20 AFLP phenotypes (groups A–T in Table 1). The relationships of the groups to one another, measured by the number of band differences, were visualized by an unrooted tree (Fig. 2). Isolates representing 16 of the groups were observed in Denmark, isolates representing seven of the groups were observed in the UK, and isolates representing three groups were observed in both France and Germany. This discrepancy in the number of groups in each country probably reflected the selection of isolates rather than differences in genetic diversity in the natural P. striiformis f.sp. tritici populations in the four countries.
Isolates from the four countries were interspersed throughout the tree. Five of the seven AFLP phenotypes from the UK were observed in Denmark; two of the three phenotypes observed in France were observed in Denmark, as were two of the three German phenotypes. This is an indication that the rate of migration may be sufficiently high to restrict independent divergence of the P. striiformis f.sp. tritici populations in the four countries. Where more than one isolate was present in an AFLP group, they either had the same pathotype (groups E and J in Fig. 2) or closely related sets of pathotypes (groups K, O and A).
The distribution of AFLP bands among P. striiformis f.sp. tritici genotypes was analysed to determine whether there was any evidence that recombination may have affected the population structure over the time and area represented by the samples. The tree estimated from the observed AFLP data set had a length of 37 steps, which was very significantly shorter than the mean length of 1000 trees from randomized data sets (81·7 ± 2·3 steps; Fig. 3). This allowed the null hypothesis that the fungus has a population structure consistent with recombination to be rejected in favour of the alternative hypothesis that the fungus reproduces asexually.
Having shown that there was no evidence for recombination in the P. striiformis f.sp. tritici population represented by the samples investigated, a phylogeographic method was used to analyse migration between countries in north-west Europe. The sum of the lengths of the separate trees for the four countries was 82 steps (35 for Denmark, 25 for the UK, and 11 each for France and Germany). This fell clearly within the range of the summed tree lengths constructed when AFLP phenotype data sets were randomized among the four countries, 83·1 ± 7·5 steps (Fig. 4). The hypothesis that the populations of P. striiformis f.sp. tritici in these countries have evolved separately, that is, without migration, was therefore rejected. The data are consistent with the hypothesis that there is effectively a single population of P. striiformis f.sp. tritici in the four countries, and that AFLP phenotypes found in more than one country either migrated between these countries, over a distance of up to 1700 km, or had a common origin in yet another country.
It is striking that in five cases, isolates from different countries were present in the same groups (note that each group represents isolates of indistinguishable AFLP phenotypes): A, E, J, K and O (Fig. 2). Given the conclusions that P. striiformis f.sp. tritici reproduces asexually, and that its urediniospores migrate between the UK and Denmark, whether directly or otherwise, these five groups of isolates may represent instances where the same pathogen clone may have caused rust on wheat cultivars with the same resistance gene in different countries. The case with isolate 94/519, collected from cv. Brigadier in Cambridgeshire in 1994, and isolate 11/97, collected from cv. Brigadier in Skælskør in 1997 (Table 1), is the most striking (group K in Fig. 2). These two isolates were sampled at the first detected outbreak of yellow rust on wheat with the Yr17-resistance gene in each country. The two isolates were indistinguishable in pathotype and in AFLP phenotype. Group K also contained French and German isolates of the same pathotype collected in 1997 and 1998.
The second case comprised another group of isolates with virulence for Yr17 (group O, Fig. 2). This group contained isolates of pathotype 124 – 9 –, which was also found in all four countries. Other isolates of the same pathotype were located in the groups N, P and Q, which all differed from group O by one AFLP marker. Isolates of group P were observed only in Denmark, and isolates of groups N and Q were observed only in Germany and France, respectively.
The third case comprised the group of isolates which represented the most frequent pathotype in Denmark between 1993 and 1995, – 24 – 9– –. Many of these isolates originated from cultivar Sleipner (approved as Slejpner in the UK) with Yr9 resistance. Isolates of pathotype – 24 – 9 – – – were present in three AFLP groups (groups E, G and C in Fig. 2), which were clearly different from the AFLP patterns of other pathotypes. All but two of these isolates were in group E, where two British isolates collected in Suffolk in 1988 and in Sussex in 1990 were also present. Following the ineffectiveness of Yr9, cultivars in which Yr9 was combined with Yr6 became widely grown. Five pathotypes were detected that included the matching virulence combination, of which – 2469– – was the most frequent. This pathotype was present in AFLP groups A, D and R (all present on the same main branch in Fig. 2), and in the groups F and I on the same main branch as the ‘Sleipner isolates’. In group A, a Danish isolate collected in 1994 at Roskilde and a UK isolate collected in Roxburghshire in 1988 were indistinguishable in both AFLP type and pathotype.
The last case comprised isolates with the pathotype 12 – – – – [CV] (group J in Fig. 2), where an isolate collected in the UK was indistinguishable from a Danish isolate. These isolates were virulent on cultivars Hereward (widely grown in the UK in the 1990s) and Kraka (up to 80% of the wheat grown in Denmark in the 1980s).
The fungal samples in this study were collected in connection with national virulence survey programmes in four north-west European countries (Bayles et al., 2000), and represented all combinations of pathotype and AFLP phenotype in a previous Danish study (Justesen et al., 2002) and a fair proportion of the known pathotypes present in the UK from 1988–95 (Bayles et al., 1997). In addition, all known Yr17-virulent pathotypes from Germany and France in 1997 and 1998 were included, whereas other pathotypes in these countries were not considered (Bayles et al., 2000). As the sampling scheme in the virulence surveys in the UK and Denmark favoured the detection of new pathotypes, most of the actual genetic diversity in the natural P. striiformis f.sp. tritici populations in these two countries was probably sampled. Nevertheless, the level of AFLP diversity was very low, as in previous molecular research on P. striiformis (Newton & Caten, 1985; Dickinson et al., 1990). The basis for the samples was slightly different in the four countries. In Denmark, individual yellow rust lesions were collected, generally at low disease severity, whereas the samples from Germany, France and the UK were based on single leaf cultures. To avoid a potential risk of using contaminated samples, each isolate should ideally have been re-established on the basis of a single urediniospore. However, as this technique is difficult to establish as a routine with sufficient confidence in Puccinia striiformis f.sp. tritici, the purity of samples was ensured on the basis of infection types on differential cultivars and intensity of polymorphic AFLP fragments. Only very few samples were discarded as a result of mixed infection types on single plants of a differential cultivar, or AFLP fragments of low intensity for bands that were otherwise clearly present or absent in other isolates.
Analysis of molecular variation is particularly useful in studies of migration if it can be assumed that the DNA markers used are selectively neutral (Brown, 1996; Milgroom, 1997). It is reasonable to assume that AFLP bands are indeed neutral markers (Ridout & Donini, 1999), in contrast to resistance-specific virulence, which is subject to strong selection by resistance genes in host cultivars (Stubbs, 1985; Munk et al., 1991; Hovmøller et al., 1993; Brown, 1995). Hence, only AFLP markers were used in the phylogenetic analysis. As usual in studies of molecular phylogenetics, it was assumed that marker identity reflected common descent and that identities due to convergence, parallelism and reversals were minimal. It has been argued that DNA sequence data would be most informative in this respect, but due to the very low amount of molecular diversity, it may be difficult to achieve sufficient sequence data in Puccinia striiformis f.sp. tritici. Furthermore, as AFLP-derived phylogenetic relationships within a species or closely related species may give similar results to sequence-based data, as in a recent study of Ustilaginomycetes (Bakkeren et al., 2000), it appears that AFLP was the most appropriate choice in the present case. The analysis gave no indication of any kind of recombination (Fig. 3), in agreement with the lack of biological evidence for sexual or parasexual behaviour under natural conditions (Stubbs, 1985; Manners, 1988). The sensitivity of the test used to detect recombination is not known precisely, but it is likely that the null hypothesis will not be rejected even if only a few individuals in each generation originate by sexual reproduction in the population under investigation (Burt et al., 1996). In practice, therefore, these results indicate that the rate of recombination between loci is of the same order of magnitude as the rate of mutation of AFLP markers, or even lower. Despite the low level of genetic variation, and the absence of recombination, the fungus is notorious for rapid development of virulent pathotypes which are able to infect wheat cultivars with new sources of yellow rust resistance (Stubbs, 1988; Wellings & McIntosh, 1990; Bayles et al., 2000; Hovmøller, 2001).
An evolutionary tree shorter than that expected from random mating could also arise if the sample studied were drawn from two or more populations in which there was sexual reproduction, but which did not interbreed (Wolfe & Knott, 1982). This would give rise to the appearance of linkage disequilibrium in the sample. However, such a sampling effect could be excluded in this study, where there was clearly a single population of P. striiformis f.sp. tritici covering both the UK and Denmark (Fig. 2), even to the extent that isolates from the two countries had indistinguishable AFLP fingerprints and pathotypes. Instead, linkage disequilibrium in this population is likely to have arisen from mutation and further evolved through host-induced selection (data not shown), according to the mechanisms described by Østergård & Hovmøller (1991).
The existence of a single population of P. striiformis f.sp. tritici in the UK and Denmark, and possibly in other countries in north-west Europe (Figs 1 and 2) does not necessarily imply that the rates of migration between different countries are equal in both directions. Rather, the pattern of evolution of pathotypes, and the complete absence of wheat yellow rust from Denmark in certain years (1985, 1986, 1996; Hovmøller, 2001) suggests that a continent–island model (Hedrick, 1985), in which migration is predominantly from the UK to Denmark, is the most appropriate description of the population biology of P. striiformis f.sp. tritici in this region. This is also consistent with a relatively large population size of the yellow rust fungus in the UK in April, May and June (the time of year when yellow rust is often re-established in Denmark) and the prevailing winds from the south and west in this period (Cappelen, 2000). Due to mild winters, the UK wheat crops tend to recover from winter injuries in spring 4–6 weeks earlier than in Denmark. It is also possible that indistinguishable isolates found in both countries may have a common origin somewhere else (Bayles et al., 2000; Hovmøller, 2001; Justesen et al., 2002), but this hypothesis could not be tested using the available data.
Migration predominantly from the UK to Denmark is consistent with the presence of the 12 – – 9 – pathotype (AFLP phenotype K in Fig. 2) in Cambridgeshire, UK in 1994, the first known site in Europe where Yr17 resistance was observed to be ineffective (Bayles et al., 2000), and with the appearance of an indistinguishable isolate in Denmark in 1997 (Justesen et al., 2002). Similarly, a second Yr17-virulent pathotype, 124 – 9 –, which was first discovered in Lincolnshire, UK in 1995, was observed in Denmark in 1997. Migration from the UK to Denmark is also consistent with the existence of the – 24 – 9 – – pathotype with AFLP phenotype E in the UK in the 1980s (Bayles et al., 1997) and the presence of indistinguishable isolates in Denmark in the 1990s, following a period when Yr9 cultivars in Denmark were completely resistant to wheat yellow rust. Similar migration events may have caused the appearance of indistinguishable isolates with combined Yr6 + Yr9 virulence (– 2469 – – pathotype, AFLP phenotypes A, D, I and R) and Carstens V virulence (12 – – – – [CV] pathotype, AFLP phenotype J) in the two countries.
The continent–island model may apply to other cases, where a disease regularly reappears in areas where the pathogen does not survive during part of the year, either because host plants are absent or the environment is too severe for pathogen survival, for example the ‘Puccinia pathway’ in North America (Roelfs, 1986) or the spread of yellow rust from Gansu province to other parts of northern China (Yang & Zeng, 1992). A similar situation was present in Denmark in the 1960s and 1970s, where barley powdery mildew was reintroduced each year from external sources. In this period, winter barley growing was forbidden in order to reduce the risk of powdery mildew and rust epidemics in spring barley crops (Stapel & Hermansen, 1968).
The alternative hypothesis, that migration was restricted and the new Yr17-virulent clones observed in Denmark in 1997 and 1998 had developed by mutation in existing clones in the Danish P. striiformis f.sp. tritici population, is not supported by the data presented here, nor by the supplementary data on yellow rust epidemiology in Denmark in these years (Hovmøller, 2001; Justesen et al., 2002). No yellow rust was detected in Denmark in 1996, despite extensive searching – the population either survived undetected, which is not very likely due to the very different composition of the yellow rust population before and after 1996 (Justesen et al., 2002), or was re-established from external sources. If the population had survived undetected from 1995 to 1997, the most likely ‘ancestors’ of isolates in AFLP group K (pathotype 12 – – 9 –), which were the most common in Denmark in 1997–98, were isolates in AFLP group L or M (different from K in one and two AFLP fragments, respectively); Fig. 2. This possibility would require at least one mutation in an AFLP fragment and one mutation from avirulence to virulence in the same clone. While the mutation from avirulence to virulence for Yr17, which was subject to strong direct selection due to the Yr17-resistant cultivars, may have occurred in Denmark, it appears less likely that an additional mutation at a selective neutral locus occurred in the same clonal lineage during the same period. For isolates of AFLP group O (pathotype 124 – 9 –), which was the other frequent type in 1997–98, an isolate with a similar AFLP phenotype, but possessing virulence for Yr6, was observed in 1994 (Hovmøller, 2001). Despite the fact that the loss of virulence for Yr6 is quite possible, the fact that the same mutation should have taken place in the UK population in 1995, at a time when the 12469 – pathotype had not yet been detected in this country (Bayles & Stigwood, 1996), makes this scenario less probable.
In conclusion, the present results support the hypotheses of Zadoks (1961) and Hermansen & Stapel (1972) who studied the spread of disease; Hermansen et al. (1978) who studied the spread of spores; and Stubbs (1985) who studied the spread of pathotypes, that long-distance dispersal of yellow rust spores is highly significant for the spread of the yellow rust disease from one country to another in north-western Europe. In the present case, the spread of yellow rust across large geographical areas may lead to substantial costs for European farmers, in the form of either fungicide purchases or yield losses (Bayles et al., 1997; Bayles et al., 2000; Hovmøller, 2001). The agronomic consequences of long-distance pathogen migration are enhanced by the limited diversity in resistance genes used in European wheat-breeding programmes. At any one time, relatively few yellow rust resistance genes, which are still effective for yellow rust control in Europe, are available to breeders (McIntosh et al., 1995). When a wheat cultivar with previously effective resistance selects new, virulent pathotypes of P. striiformis f.sp. tritici, virulent clones multiply rapidly and are wind-dispersed (Zadoks, 1961; Stubbs, 1985). They may then establish epidemics in other fields of the same cultivar or of other cultivars with the same resistance specificity. This pattern of selection and dispersal causes the P. striiformis f.sp. tritici population rapidly to become dominated by a small number of newly selected, virulent clones over very large areas. The use of the same cultivars in several European countries may therefore periodically expose the whole of the wheat crop in northern Europe to the risk of new, virulent pathotypes.
The points made above are especially relevant at present because of commercial developments in wheat breeding in Europe, where companies are consolidating and breeding programmes are increasingly aimed at developing cultivars suitable for cultivation in several countries (Anon., 1997). Increasing internationalization of the plant breeding industry could exacerbate the tendency to genetic uniformity by further increasing the extent to which the same genes for resistance, to yellow rust as well as other diseases, are deployed in different countries. The severity of yellow rust has been restricted in recent years by the use of fungicides, but this option is now becoming limited by environmental restrictions (Anon., 1999) and by the increasing cost of discovering and registering fungicides (Urech, 1996; 1999). The long-term future of disease control therefore rests on the effective deployment of resistant cultivars. The risk of periodic outbreaks of pan-European epidemics of yellow rust would be lessened if wheat breeders had access to more diverse sources of resistance; if greater use were made of durable resistance (Johnson, 1992); and if different sources of resistance were used by different companies, preferably in different countries.
We thank R. A. Bayles, P. L. Stigwood (NIAB, Cambridge, UK), C. de Vallavieille-Pope (INRA, Thiverval-Grignon, France) and K. Flath (BBA, Kleinmachnow, Germany) for providing isolates; A. Drabek, E. M. Foster, E. Frederiksen and V. Mikkelsen for expert technical assistance; and our colleagues Lesley Boyd, Hanne Østergård and David Yohalem for critical comments on different versions of this manuscript. M.S.H. was supported by a European Union TMR Category 40 Fellowship. Additional support was provided by the former Ministry of Food, Agriculture and Fisheries (Denmark), the Biotechnology and Biological Sciences Research Council and the Ministry of Agriculture, Fisheries and Food (UK), and by travel grants from European COST Action 817.