Genetics and Genomics of ResistanceFree Access icon

Spatiotemporal Changes in Varietal Resistance to Wheat Yellow Rust in France Reveal an Increase in Field Resistance Level During the Period 1985–2018

    Affiliations
    Authors and Affiliations
    • Rémi Perronne1 2
    • Florence Dubs1
    • Claude de Vallavieille-Pope3
    • Marc Leconte3
    • Philippe du Cheyron4
    • Valérie Cadot5
    • Tiphaine Vidal3
    • Jérôme Enjalbert1
    1. 1Université Paris-Saclay, INRAE, CNRS, AgroParisTech, GQE-Le Moulon, 91190 Gif-sur-Yvette, France
    2. 2IGEPP, INRAE, Institut Agro, Université Rennes, 35653 Le Rheu, France
    3. 3Université Paris-Saclay, INRAE, AgroParisTech, UMR BIOGER, 78850 Thiverval-Grignon, France
    4. 4Arvalis-Institut du Végétal, 91190 Villiers-le-Bâcle, France
    5. 5GEVES Beaucouzé, 49071 Beaucouzé, France

    Published Online:https://doi.org/10.1094/PHYTO-05-20-0187-R

    Abstract

    Monitoring spatiotemporal changes in varietal resistance and understanding its drivers seem essential to managing plant diseases but require having access to the genetic basis of disease resistance and to its deployment. In this study, we focused on yellow rust (Puccinia striiformis f. sp. tritici) for three decades in France, by using field adult plant resistance levels, Yr race-specific resistance genes of varieties, presence of Puccinia striiformis f. sp. tritici pathotypes and their virulence profiles, and systematic surveys of the acreages of bread wheat varieties available at a yearly survey time and at a district level. Based on these data, we studied spatiotemporal changes in varietal resistance over the period from 1985 to 2018 in 54 French administrative districts (hereafter “departments”) by using a set of relevant indicators weighted by the relative acreage proportion of the varieties sown at the department level. Our analyses revealed an increase in varietal resistance over decades that would be due to the accumulation of both quantitative resistance and different race-specific resistance genes. We suggest that, beyond breeders, several actors, including examination offices, agricultural advisory services, and farmers, may have had a substantial influence on these spatiotemporal changes, promoting more resistant varieties and the rapid replacement of newly susceptible varieties by still resistant ones at the beginning of each epidemic.

    Agroecosystems have undergone profound changes since the mid-twentieth century. Among them, the trend toward genetic erosion, caused by the replacement of landraces by modern pure lines and hybrids, has led to changes in the levels of crop diversity according to the crop and the geographic area (Rauf et al. 2010; van de Wouw et al. 2010a, 2010b). Genetic erosion has progressed along with the optimization and simplification of cropping systems, the use of chemical inputs, and mechanization that have allowed us to homogenize the environmental conditions of crop growth and to achieve major yield gains. Because of current global changes, many crops show an interannual yield variability caused by climate instability (Lesk et al. 2016; Ray et al. 2015; Zhao et al. 2017) and experience a recurrence of pest pressures (Singh et al. 2016). Indeed, major past and recent epidemics have been exacerbated by a lack of crop diversity resulting in the rapid adaptation and spread of new virulent pathogen populations (McDonald and Stukenbrock 2016; Singh et al. 2015; Ullstrup 1972). In this context, increasing the genetic diversity of a given crop at a regional or national scale may be seen as a means of ensuring the interannual yield stability by promoting diverse responses to abiotic or biotic stresses.

    In most countries, varieties have been assessed for disease resistance for registration to the National Catalogue and often monitored after registration. In fewer countries, and for only some pathosystems, both precise annual pathogen race surveys and identification of race-specific resistance genes have been conducted for decades (Cowger and Brown 2019; de Vallavieille-Pope et al. 2012; Hovmøller et al. 2016; Liu et al. 2017; Su et al. 2003). This is particularly the case for wheat rusts, which are major diseases responsible for significant yield losses worldwide over the past decade (Beddow et al. 2015), especially through the emergence of more aggressive new races (Hovmøller et al. 2016; Singh et al. 2015). Indeed, race-specific resistance genes are major determinants of wheat–rust interactions, and their Mendelian inheritance was among the first described (Biffen 1905). The complementary monitoring of wheat varietal resistances and rust pathotypes has been essential for describing how breeders mobilize race-specific resistance genes over time (Bayles et al. 2000; de Vallavieille-Pope et al. 2012) and how the field resistance levels of particular varieties change (Papaïx et al. 2011), allowing a better assessment of epidemiological risks.

    Still, recommendations to limit rust outbreaks have usually focused on indicators based on varietal denomination only, such as the weighted average age of varieties (Brennan and Byerlee 1991; Pingali 1999), and their use has not been widespread in Western Europe. If these kinds of indicators are easy to quantify, they overlook available information on resistance characteristics of varieties. The surveys of spatial and temporal changes in field resistance levels, and of the deployment of resistance genes, appear more relevant to increase the durability of the sources of resistance (Hovmøller 2001), but they are much more complex to assess and have not yet been widely carried out.

    Monitoring spatiotemporal changes in varietal resistance and understanding their drivers requires having access to the genetic basis of disease resistance and its deployment. This is why we focused on wheat yellow rust (caused by Puccinia striiformis Westend. f. sp. tritici Eriks.), an obligate biotrophic fungus characterized by a complex life cycle and a cereal as primary host (Hovmøller et al. 2011). Varietal resistance to yellow rust, as for many pathogens, is determined by the presence of genes causing either all-stage race-specific resistance or adult plant non-race-specific resistance. A major race-specific resistance depends on a specific genetic interaction between a Yr resistance gene of the variety and an Avr avirulence gene of the P. striiformis f. sp. tritici pathotype, following a gene-for-gene interaction (Flor 1956; Periyannan et al. 2017). Race-specific resistance genes, which confer complete resistance to specific P. striiformis f. sp. tritici pathotypes, are monogenically inherited and expressed since the seedling stage (i.e., corresponding to all-stage resistance genes). The deployment of one race-specific resistance gene over large areas of bread wheat creates a strong selective pressure on pathogen populations, leading to the rapid local evolution of a new virulent pathotype and thus the rapid breakdown of this race-specific resistance gene (Bayles et al. 2000). This varietal resistance, based partly on the use of Yr resistance genes, explains the recurrence of epidemics characterized by the successive breakdown of these genes by new pathotypes, also known as boom-and-bust cycles (Bayles et al. 2000; de Vallavieille-Pope et al. 2012). Currently, 83 Yr resistance genes have been reported and permanently named (McIntosh et al. 2017) and 67 temporarily designated (Wang and Chen 2017). By contrast, non-race-specific resistances can be quantitatively inherited (i.e., due to several genes with partial effects or quantitative trait loci) or can be due to major genes expressed only at later stages of plant development (i.e., adult plant resistance genes; Chen 2013; Rosewarne et al. 2013). Non-race-specific resistance usually confers long-term partial durable resistance against all pathotypes of the pathogen (Periyannan et al. 2017). Both race-specific and non-race-specific resistance genes can contribute to the high resistance level observed for certain varieties (Paillard et al. 2012), and overall the choice of the variety is considered the most effective, economical, and environmentally-friendly lever to manage yellow rust (Gladders et al. 2007).

    In France, the arms race between yellow rust and bread wheat has been thoroughly scrutinized (de Vallavieille-Pope et al. 2012). France is an important bread wheat production area, subject to the recurrence of yellow rust epidemics. Since the early 1980s, field adult plant resistance levels, Yr race-specific resistance genes of varieties, and presence of P. striiformis f. sp. tritici pathotypes and their virulence have been described. These surveys have shown that P. striiformis f. sp. tritici pathotypes can have a geographic structure (Bahri et al. 2009; Enjalbert et al. 2005) and that development of the disease depends largely on more favorable weather conditions in the northern half of the territory, especially along the northwest coastline. Moreover, a systematic survey of the acreages of bread wheat varieties is performed in France. These statistics have shown that the northern half of France is characterized by high varietal diversity and fast varietal turnover (Perronne and Goldringer 2018). Indeed, farmers rapidly withdrew the varieties affected by the breakdown of Yr race-specific resistance genes and replaced these varieties with other varieties available for the same end uses but still resistant to yellow rust (Perronne et al. 2017). However, it should be noted that characterization of the Yr race-specific resistance genes of wheat genotypes is not performed during the registration procedure; most postulations are performed after registration. This can be explained by a phenotypic characterization in field of the wheat genotypes during the registration procedure, without additional assessment based on tests using a particular set of P. striiformis f. sp. tritici pathotypes or molecular markers (which are not necessarily available for all Yr genes). Furthermore, knowledge of Yr race-specific seedling resistance genes at registration does not necessarily guarantee the field adult plant resistance level after registration.

    Based on the preceding rationale, we addressed a set of core questions about potential spatiotemporal changes in varietal resistance in France:

    • Q1. How has the acreage proportion of wheat, characterized by different numbers of Yr resistance genes, changed in recent decades, and were these changes spatially structured?

    • Q2. Did the weighted mean and the weighted variance of the field resistance level change in recent decades, and were these changes spatially structured?

    • Q3. Focusing on effective Yr resistance genes each year (i.e., taking into account the presence of the dominant P. striiformis f. sp. tritici pathotypes), does the acreage proportion of wheat characterized by at least one of these effective Yr genes have pattern comparable or different to those considering all (i.e., accounting for both effective and ineffective Yr genes)?

    • Q4. Are all or parts of these patterns based on disease resistance information comparable with patterns of the weighted average age of varieties?

    MATERIALS AND METHODS

    Data sources.

    Acreage of bread wheat varieties.

    The Wheat Board, currently FranceAgriMer, carried out systematic annual surveys of acreages of bread wheat varieties in France from 1980 to 2018 in 54 French administrative districts (hereafter “departments”). A questionnaire was sent to about 500 farmers in each department to obtain information about the varieties sown and their acreages at the farm level. First, the Wheat Board estimated the relative proportions of each bread wheat variety sown in the department with an accuracy of 0.5%. Second, we estimated acreages of varieties based on the total production area of bread wheat known each year in each department and provided by Agreste (Agreste, Statistique Agricole Annuelle). We excluded several years of data because of sampling biases from surveys (1980, 1984, 1991, and 1997).

    Characterization of field adult plant resistance level of bread wheat varieties to yellow rust at registration and after registration.

    During the registration procedure, the French Examination Office (GEVES) characterized the varieties according to their field adult plant resistance level to yellow rust through Value for Cultivation, Use and Sustainability studies. The same characterization procedure was then conducted during the postregistration period of commercial use of a variety by the agricultural advisory services (Arvalis-Institut du Végétal). These surveys were carried out annually over the period 1979 to 2018 at a national level, allowing the evaluation of the field adult plant resistance levels during each of the 2 years of registration and afterward for at least the main years of the commercial cycle of the variety. Throughout the period, the field adult plant resistance level was rated from 1 to 9 (i.e., highly susceptible to highly resistant). The registration date of each variety was also provided by GEVES. During the registration procedure, both Value for Cultivation, Use and Sustainability trials under conditions of natural contamination and inoculation trials with the dominant P. striiformis f. sp. tritici pathotypes from the previous year were carried out for 2 years. Some data from the postregistration field adult plant resistance levels were missing. Because these data were subject to only rare and low temporal variations, missing data imputation was based on the scorings of the year before and after the missing data. If the two scores were similar, this information was used to replace the missing value. If the two scores were different, a value was chosen depending on the situation. When the missing year corresponded to an epidemic period, the lowest value was used for replacement; that is, we assumed that the breakdown of the Yr race-specific resistance gene started in that year. Otherwise, the highest value was used; that is, we assumed that no particular cause could have influenced the value during this year.

    Characterization of P. striiformis f. sp. tritici pathotypes and their virulence profiles.

    Seven major epidemics occurred over the period 1984 to 2018 in the north of France, whereas only one epidemic occurred specifically in the south of France. The P. striiformis f. sp. tritici pathotypes detected over the period 1984 to 2009 in France are described in detail in de Vallavieille-Pope et al. (2012), and data on their virulence profiles were extracted from this study (see Supplementary Table S1.1 and S1.2 for details). These European pathotypes were then replaced by two major new P. striiformis f. sp. tritici pathotypes from the near-Himalayan region over the period 2010 to 2018 (du Cheyron et al. 2016; Hovmøller et al. 2016; Lefèvre et al. 2015; see Supplementary Table S1.1 for details).

    Postulation of Yr race-specific resistance genes.

    The Yr race-specific resistance genes were postulated at the seedling stage based on a set of 12 French P. striiformis f. sp. tritici pathotypes with complementary virulence profiles. This set varied over the period from 1984 to 2018 to account for new P. striiformis f. sp. tritici pathotypes occurring in each successive epidemic period. The postulation was assumed to conform to a gene-for-gene interaction; that is, the resistance depends on a specific interaction between a Yr resistance gene of the variety and an Avr avirulence gene of the P. striiformis f. sp. tritici pathotype (Flor 1956). The method of postulation is described in detail in de Vallavieille-Pope et al. (1990). Responses of varieties at the seedling stage after inoculation were scored based on a 0 to 9 scale proposed by McNeal et al. (1971).

    Indicators.

    To follow the spatiotemporal changes in the number of Yr race-specific resistance genes (Q1), we characterized the relative acreage proportion of four categories of bread wheat varieties carrying either no resistance gene (0), at least one gene (1+), at least two genes (2+), or at least three genes (3+). No variety presented more than three known Yr race-specific resistance genes in our study, among the genes that can be characterized by the postulation method. These indicators, based on proportions, are always calculated in terms of relative acreages, that is, based only on varieties for which data on both acreage proportion and postulation of Yr genes are available, whether or not this postulation has identified one or several Yr genes. The set of P. striiformis f. sp. tritici pathotypes used to postulate the Yr race-specific resistance genes does not necessarily allow us to determine the absence or the presence of all Yr genes for each variety. In most cases, the set of P. striiformis f. sp. tritici pathotypes made it possible to identify the absence of any Yr genes (susceptible variety for all pathotypes used) or the presence of one or more Yr genes while other Yr genes were absent (resistant variety with one or several Yr race-specific resistance genes identified considering the set of pathotypes used). However, in numerous cases the set of pathotypes used yielded a nonspecific result (i.e., a Yr gene is assumed to be present because of the response observed, but it cannot be associated with one particular named Yr gene). We may therefore only hypothesize that at least one Yr gene is present among a set of potential genes while excluding certain Yr genes because of the susceptible reactions observed. Moreover, in some cases the variety appeared resistant to all the P. striiformis f. sp. tritici pathotypes tested, which allows us to hypothesize that at least one unknown Yr gene is present. To compute the relative acreage proportion of bread wheat associated with at least one Yr gene, we included cases with intermediate results and resistant varieties to all pathotypes because, in both cases, we cannot conclude that more than one race-specific Yr gene was present. The relative acreage proportion of bread wheat resistant to all the P. striiformis f. sp. tritici pathotypes included in the set has been used as additional information for further interpretation.

    To study the field adult plant resistance level to yellow rust (Q2), we calculated the weighted mean (FRLwm) and the weighted variance of the field adult-plant resistance level to yellow rust (FRLwv). Both indicators are weighted by the relative acreage proportion of the varieties sown, as follows:

    FRLwm=k=1Kαkrelfk
    FRLwv=k=1Kαkrel(fkk=1Kαkrelfk)2

    with

    αkrel=αk/k=1Kαk

    and

    k=1Kαkrel=1

    where K is the number of varieties sown in a department in a year, fk is the field adult plant resistance level to yellow rust of variety k (either at registration only or accounting for postregistration changes), αk is the acreage proportion of variety k in that department that year, and αkrel is the relative acreage proportion of variety k. We need to compute αkrel because the cumulative acreage proportion of varieties is generally less than one because of either sampling biases in surveys or varieties we discarded when computing the indicator because of missing information about their resistance levels. We distinguished two options to calculate each indicator. The first option is based solely on the field adult plant resistance level defined at registration. The second option also took into account the changes in field resistance levels over time. Concerning the second option, we assumed that after the end of the postregistration evaluation of a variety, the field adult plant resistance level was stable. The end of the postregistration evaluation was usually caused by an important decrease in the acreage of the variety, being indeed at the end of its commercial life cycle (Perronne et al. 2017). This second option for calculation can be considered more representative of the field situation each year. To account for divergence in values between the two calculation options, we computed the difference between the indicators of field resistance levels to yellow rust at registration and after registration.

    To focus on the effective Yr genes (Q3), we calculated the same FRL indicators as above but considered only the Yr genes controlling the dominant P. striiformis f. sp. tritici pathotypes during the year and in the area studied. We therefore defined areas characterized by distinct epidemics associated with different P. striiformis f. sp. tritici pathotypes and then, in each area, a set of dominant P. striiformis f. sp. tritici pathotypes representing 96% of all the isolates received over the period from 1985 to 2018 (Supplementary Table S1.1). For each dominant P. striiformis f. sp. tritici pathotype, we defined a temporal period delineating each epidemic associated with this pathotype, and during this temporal period, this pathotype was considered to render some particular Yr race-specific resistance genes ineffective, depending on its virulence profile (see Supplementary Table S1.1 for details). In this context, we computed the relative acreage proportion of bread wheat characterized by three categories: resistant, intermediate, and susceptible responses to yellow rust. The relative acreage proportion of resistant (or intermediate) varieties corresponded to varieties resistant (or at least intermediate) to all dominant P. striiformis f. sp. tritici pathotypes occurring during a year in a department, with this resistance (or at least sometimes intermediate response) being due to at least one Yr race-specific resistance gene. The relative acreage proportion of susceptible varieties corresponded to varieties susceptible for at least one of the dominant P. striiformis f. sp. tritici pathotypes occurring in a year in a department. As indicated previously, responses of varieties at the seedling stage after inoculation were scored based on a 0 to 9 scale proposed by McNeal et al. (1971) (see Supplementary Material 1 for details).

    The weighted average age of varieties WAt (Brennan and Byerlee 1991) was computed with varietal denomination only, to compare its spatiotemporal pattern to patterns observed on previous indicators focusing on varietal resistance (Q4), as follows:

    k=1KαkrelRk

    where Rk is the number of years since the registration of variety k.

    Statistical analyses.

    Before statistical analyses, we must specify that our statistical procedure aims not to control for spatial or temporal autocorrelations in our datasets but to describe the spatiotemporal structure. The different indicators considered previously are mainly descriptive and designed to capture the spatiotemporal trends found in previous studies (Perronne et al. 2017; Perronne and Goldringer 2018).

    Because our analyses are based on the deployment of bread wheat varieties, we estimated the wheat acreage proportion corresponding to the varieties included in the calculation of each indicator for each department × year combination. We considered that below a threshold of 80% of the wheat acreage characterized by varietal denomination, the estimated value of the indicator was not sufficiently reliable to be used for subsequent analyses. When a significant number of years for a department, usually 5 consecutive years, was below this threshold, this department was not used for subsequent analyses to avoid misinterpretation. This criterion led to the withdrawal of eight departments, representing 3.6 ± 0.9% of the national wheat acreage, and the remaining departments surveyed represented 85.4 ± 1.8% of the national wheat acreage over the period from 1985 to 2018. Over this period at the department level, several indicators were retained in these 46 departments for 32 years, representing 1,472 department × year combinations. Because of 0.5% missing data, corresponding to cases where the survey was not made in some departments in some years, 8 of the 1,472 department × year combinations had no information about acreage proportion. The threshold did not account for acreages of varietal mixtures because we assumed that varietal mixtures were composed of the same varieties and in the same proportions as the varieties sown in pure stands, which corresponds to the current use in France at the department level.

    Because of missing data from the various data sources, 0.8 to 3.1% of missing data were found in the various indicators after calculation (i.e., >0.5% as indicated previously due to missing data needed to calculate the indicators from the acreage proportion). We applied a robust locally weighted regression, the LOWESS procedure, to estimate the unknown values of these indicators (Cleveland 1979; Supplementary Material S2). The LOWESS procedure limits the influence of deviant points, mainly due to sampling biases affecting the systematic surveys, and it needs only a few parameter estimates and is therefore able to take into account even short-term trends. For some departments, one or two years were below the threshold of 80% of the wheat acreage characterized for some indicators over the period from 1985 to 2018 (concerning Q1, Q2, and Q4). In these cases, the values of indicators characterized by a threshold of 80% were removed before we applied the LOWESS procedure for reestimation. Particularly concerning Q3, taking into account the data needed, corresponding to information on the responses of a variety to all dominant P. striiformis f. sp. tritici pathotypes occurring in a given year, the threshold of 80% was more often reached. In this case, we chose to remove and reestimate the same values as those concerning Q1 to avoid withdrawing too much data and therefore affecting the subsequent statistical analysis.

    Second, based on raw data when available and estimated data when necessary, a two-step procedure aiming at identifying groups of departments with similar temporal trends was applied to the dataset of each indicator separately (always with a dimension of 46 departments over 32 years). We first applied a principal component analysis (PCA) and retained the minimum number of PCA axes representing ≥90% of the total variation. We applied a PCA to capture the major axes of variations of the dataset for each indicator (i.e., both the main spatial and temporal trends, on an orthogonal basis, each principal component axis thus giving different information than the other axes, while partially removing noise variations). We then applied a K-means clustering iterative algorithm to the retained PCA axes to obtain groups of departments with similar temporal trends and the Caliński-Harabasz criterion to select the optimal number of groups (Caliński and Harabasz 1974). However, when several Caliński-Harabasz criterion values were very close, we chose the smaller number of groups. The optimal number of groups has been restricted between 2 and 10 to avoid groups containing only one department. When we compared the weighted mean (or variance) based on adult plant resistance level at registration and after registration, we defined groups based on the level of resistance at registration only. Moreover, five departments are located in the south area, where yellow rust was present during only two short periods, 1996 to 2002 and 2012 to 2018. In this case, the PCA could not be applied because the indicators based on the effective Yr genes could not be calculated for all years over the whole period (Q3). However, the values of indicators were similar in these five departments, so we arbitrarily constituted a complementary group with these five departments, which is only presented graphically.

    Statistical analyses were performed in R version 3.5.1 (R Development Core Team 2019), with its packages ‘maps’ for mapping of indicators (Becker et al. 2018), ‘stats’ for Pearson correlations and the LOWESS procedure, and ‘vegan’ for PCA and clustering (Oksanen et al. 2019). Datasets and R scripts are available from the authors upon request and agreement of the various teams that produced the raw data (see the Acknowledgments for details).

    RESULTS

    Spatiotemporal structure of acreage proportion characterized by major Yr resistance genes.

    The use of complementary indicators reveals large temporal changes in the deployment of major Yr race-specific resistance genes, as well as a spatial structure with agricultural regions characterized by a comparable dynamic. First, from 1985 to 2018, nearly all wheat acreages were characterized by varieties with at least one major resistance gene, except for some departments in the south that have shown substantial acreages of a few varieties without any major resistance gene known, especially the varieties Manital and Victo (Figs. 1A and 2A). Then, the relative acreage proportion of wheat characterized by varieties with at least two major resistance genes often represented >80% of the total acreage proportion of wheat characterized in a department (Figs. 1B and 2B). The decrease observed for this latter indicator could be partly explained by some leading varieties characterized by only one major resistance gene postulated, especially in the 2000s (e.g., the varieties Isengrain and Trémie). However, this trend of a decrease of the proportion characterized by varieties with at least two major resistance genes could also be caused by the increase in the proportion of wheat resistant to all the P. striiformis f. sp. tritici pathotypes used to postulate genes at the seedling stage. This proportion of resistant varieties especially includes varieties with potentially two or more resistance genes (Figs. 1D and 2D). The deployment of varieties with at least three major resistance genes had a much more marked spatiotemporal structure compared with other indicators. Almost no variety appeared to have three major genes postulated in the 1980s. After that, the relative acreage proportion increased until the mid-2000s and then decreased, with a temporal pattern more or less smoothed depending on the group of departments considered (Figs. 1C and 2C). Finally, in the 1980s and 2010s, the relative acreage proportion of wheat resistant to all the P. striiformis f. sp. tritici pathotypes tested was higher, especially in the departments of the northwest coastline (Figs. 1D and 2D). In this region, this trend was due mostly to two leading varieties, Arminda in the 1980s and Altria in the 2000s (Figs. 1D and 2D). In contrast, the increase observed in all departments during the 2010s is associated with numerous varieties resistant to all the P. striiformis f. sp. tritici pathotypes considered during that period.

    Fig. 1.

    Fig. 1. Groups of French departments characterized by different temporal trends over the period from 1985 to 2018 for the relative acreage proportion of bread wheat associated with at least A, one, B, two, C, or three race-specific resistance genes postulated at the seedling stage; D, the relative acreage proportion of bread wheat resistant to all Puccinia striiformis f. sp. tritici pathotypes included in the set used at the seedling stage; E, the relative acreage proportion of varieties resistant to all dominant P. striiformis f. sp. tritici pathotypes occurring a given year in a department; and F, the weighted average age of varieties WAt. The departments in white, characterized by >5 consecutive years with missing data, were not studied but represent <15% of the national wheat acreage over the period from 1985 to 2018. The shades of gray do not indicate a particular interpretation. See the Materials and Methods section and the Supplementary Materials for more details on the statistical analyses.

    Download as PowerPoint
    Fig. 2.

    Fig. 2. Mean temporal evolution, for each group of departments identified, of the relative acreage proportion of bread wheat associated with at least A, one, B, two, or C, three race-specific resistance genes postulated at the seedling stage; D, the relative acreage proportion of bread wheat resistant to all Puccinia striiformis f. sp. tritici pathotypes included in the set used at the seedling stage; E, the relative acreage proportion of varieties resistant to all dominant P. striiformis f. sp. tritici pathotypes occurring a given year in a department; and F, the weighted average age of varieties WAt. Error bars represent the standard deviations. Vertical dashed lines indicate the year of the epidemic peak at the beginning of all major epidemics occurring over the period from 1985 to 2018 with the area concerned, indicated by N, north; S, south; or N+S, both.

    Download as PowerPoint

    Spatiotemporal structure of the weighted mean and variance of the field resistance level.

    Based on the field adult plant resistance level, defined either at registration only or including postregistration changes, we found an increase in the weighted mean resistance level in most departments over the whole period from 1985 to 2018 (Figs. 3 and 4). However, this increase appears to be less substantial since the 2000s when postregistration changes are accounted for, showing a stabilized pattern or even a decrease in the 2010s (Fig. 5). The increase in the resistance level appeared to follow a spatial gradient, with the Brittany region already presenting a high resistance level in 1985, and this increase was subsequently observed on the rest of the northwest coastline, the east, the Paris Basin and the center, and finally the south of France (Figs. 3A and B and 4A and B). Since 2015, the weighted mean of the resistance level has appeared similar in all agricultural regions, being on average 7.3 ± 0.3 over the period from 2015 to 2018 at the department level when defined at registration only (Fig. 4A) and on average 6.2 ± 0.3 when postregistration changes are included (Fig. 4B).

    Fig. 3.

    Fig. 3. Groups of French departments characterized by different temporal trends over the period from 1985 to 2018 for A, the weighted mean (FRLwm) and C, the weighted variance (FRLwv) of the field adult plant resistance level to yellow rust defined at registration only; and B, the FRLwm and D, the FRLwv including postregistration changes.

    Download as PowerPoint
    Fig. 4.

    Fig. 4. Mean temporal evolution, for each group of departments identified, of A, the weighted mean (FRLwm) and C, the weighted variance (FRLwv) of the field adult plant resistance level to yellow rust defined at registration only; and B, the FRLwm and D, the FRLwv including postregistration changes. Error bars represent the standard deviations.

    Download as PowerPoint
    Fig. 5.

    Fig. 5. Groups of French departments characterized by the difference in values between the weighted mean of the field adult plant resistance level to yellow rust (FRLwm) computed from the level at registration only minus the level including postregistration changes (left), and mean temporal evolution of this indicator, for each group of departments identified (right). Error bars represent the standard deviations.

    Download as PowerPoint

    Considering the variance in the adult plant resistance level, defined either at registration only or including postregistration changes, we observed high variance levels in most areas and during most of the period from 1985 to 2018, except for recent years showing a decrease in variance (Figs. 3 and 4). In some groups of departments, a higher weighted variance of the resistance level was observed in some periods and was associated with the co-occurrence of leading varieties characterized by very different resistance levels (e.g., in the early 1990s in the north, susceptible varieties Duck and Thésée co-occurred with resistant varieties Arminda and Camp Rémy) (Fig. 4D). Over the entire period from 1985 to 2018, the weighted variance appeared slightly lower when defined at registration only rather than with postregistration changes included, although this difference was low, on average −0.2 ± 0.8 (Supplementary Fig. S3).

    Based on the difference in values between the weighted mean of the field adult plant resistance level computed from the level at registration only minus the level including postregistration changes, we found a marked trend structured in at least two periods (Fig. 5). In the 1980s and 1990s, the difference in values computed based on the level at registration only or including postregistration changes appeared low, suggesting that yellow rust pressure had little influence on the overall field adult plant resistance level within a department. In the 2000s, and then more markedly since 2012, we observed a higher difference in values, up to a mean difference of 1.0 ± 0.4 over the period from 2012 to 2018 (Fig. 5). This difference may be partly explained by a higher number of reevaluations of the resistance level after registration compared with previous decades because of the successive breakdowns of several major resistance genes since 1998 (Supplementary Fig. S4).

    Spatiotemporal structure of acreage proportion characterized by effective major Yr resistance genes.

    Focusing on effective major Yr resistance genes each year (i.e., taking into account the presence of all dominant P. striiformis f. sp. tritici pathotypes the same year in the same area), we found that the relative acreage proportion of wheat characterized by resistant varieties due to at least one major resistance gene was usually low and spatiotemporally structured (Figs. 1E and 2E). Although this calculation method is conservative, these results appeared largely consistent and negatively correlated with the relative acreage proportion of wheat varieties characterized by a susceptible response (Supplementary Figs. S5.1 and S5.2). The proportion of varieties resistant to all dominant P. striiformis f. sp. tritici pathotypes generally represented <20% of the total acreage proportion of wheat characterized in a department, and indeed in 61% of the department × year combinations over the period from 1985 to 2018 (Fig. 2E). This proportion tended to 0% at the beginning of some epidemics in the 1980s and the 2010s (Fig. 2E), indicating that almost all varieties characterized for all P. striiformis f. sp. tritici pathotypes did not have effective major resistance genes against all these dominant P. striiformis f. sp. tritici pathotypes occurring in these years. On the contrary, this proportion generally increased after the year of the epidemic peak, especially when the period between two epidemics was sufficiently long (Fig. 2E).

    Spatiotemporal structure of the weighted average age of varieties.

    Based on the varietal denomination only, we found that the weighted average age of varieties WAt has been generally quite close between most of the departments since the early 1990s (Figs. 1F and 2F). If the departments of the northeast of France had lower WAt values than the departments of the center-west in the 1980s, the average WAt was 6.2 ± 0.8 and 7.3 ± 1.3 years in these two groups, respectively, over the period from 1992 to 2018 and appeared quite stable over the last three decades in these groups (Fig. 2F). In contrast, the departments characterized by higher WAt values over the period from 1985 to 2018 accounted only for 5.7 ± 0.5% of the national wheat acreage over this period.

    DISCUSSION

    The dynamic of yellow rust varietal resistance in France from 1985 to 2018 has been described by a set of indicators allowing us to synthetize information on postulated major Yr resistance genes at the seedling stage, the presence and virulence profiles of P. striiformis f. sp. tritici pathotypes, and the level of field adult plant resistance at registration and after registration. To more fully account for deployment of the various sources of resistance to yellow rust, all indicators have been weighted by the relative acreage proportion of the varieties sown at the department level. We found a spatiotemporal structure for all indicators based on varietal resistance information, with this structure depending on the information used. However, the weighted average age of varieties based on the varietal denomination only appeared to be stable and close across most of the departments studied over the last three decades in France, indicating a stable varietal turnover.

    Increase in the level of varietal resistance to yellow rust over decades.

    Overall, the level of varietal resistance to yellow rust increased in France over the period from 1985 to 2018. Indeed, the weighted mean of the field adult plant resistance level increased in most departments over the period from 1985 to 2011 (from 0.3 in Brittany to 3.3 in the south of France when the level at registration is considered), followed by a recent stabilization over the period from 2012 to 2018 (on average 7.2 ± 0.4 when defined at registration and 6.2 ± 0.3 when postregistration changes are included). This stabilization is associated with a decrease in the weighted variance over the last decade caused by greater similarity in resistance level of the leading varieties within a department. In the same way, the relative acreage proportion of wheat characterized by several postulated major Yr resistance genes, either effective or ineffective, also increased up to the mid-2000s. Finally, the relative acreage proportion of wheat characterized by at least one effective major Yr resistance gene was usually low and characterized by several spatiotemporal fluctuations over the whole period, reflecting the boom-and-bust cycles due to large areas of wheat characterized by a sole major resistance gene rapidly broken down by one or several new P. striiformis f. sp. tritici pathotypes.

    In the rest of the Discussion section, we first present the changes observed for each group of indicators associated with each question and consider the drivers that may be responsible for these changes. Then, we discuss the main factors that may have had an influence on these different patterns globally.

    Spatiotemporal structure of acreage proportion characterized by major Yr resistance genes.

    Considering the postulated major Yr resistance genes, we found an increase in the relative acreage proportion of wheat combining three Yr genes up to the mid-2000s, followed by a decrease. This pyramiding of Yr genes has been observed in previous studies for certain genes, such as Yr6, Yr9, Yr17, and Yr32 (Bayles et al. 2000; de Vallavieille-Pope et al. 2012), and could be explained by at least two complementary causes. On one hand, these changes could be caused by a fast varietal turnover, with newly released varieties characterized by new effective genes given the new epidemic context, but also characterized by other genes that became inefficient but remained present in the genetic background, such as Yr9 and Yr17. On the other hand, the decrease in varieties combining three Yr genes observed over the last decade could be caused by a higher number of varieties identified as resistant to all the P. striiformis f. sp. tritici pathotypes used to postulate the set of major resistance genes considered, especially the P. striiformis f. sp. tritici pathotypes Warrior and Warrior(-), corresponding to PstS7 and PstS10, respectively. For these varieties, the postulation of Yr race-specific resistance genes appeared impossible, and their number of Yr genes was arbitrarily set to one in our study, corresponding to a conservative choice to avoid overestimating the number of Yr genes.

    Spatiotemporal structure of the weighted mean and variance of the field resistance level.

    Considering the weighted mean of the field adult plant resistance level, we found an increase in the level of resistance in most departments over the whole period. This dynamic is associated with a spatially structured temporal gradient, that is, a faster increase on the northwest coastline, reaching a similar resistance level in all agricultural regions in recent years. Moreover, this increase appeared steeper in the 2000s when we considered the resistance level defined at registration only. This pattern confirms an overall increase in the level of resistance at registration in France (Cadot 2017), due to both race-specific and non-race-specific resistances. However, when we accounted for postregistration changes, this increase appeared lower, and a decrease was even observed in some agricultural regions in the last decade. Such difference is caused primarily by the high number of reevaluations of the postregistration resistance level associated with the increase in the number of successive breakdowns of major resistance genes due to the occurrence of a set of new P. striiformis f. sp. tritici pathotypes (du Cheyron et al. 2016). The last decade has been indeed characterized by at least three major epidemics associated with P. striiformis f. sp. tritici pathotypes characterized by new virulence profiles (Hovmøller et al. 2016; Lefèvre et al. 2015). Moreover, the increase in the resistance level followed a spatially structured temporal gradient from the northwest coastline to the south of France. This geographic pattern refers to distinct contexts of pathogen pressures, ranging from the northwest coastline, subject to the recurrent succession of epidemics for decades (de Vallavieille-Pope et al. 2012), to the south, where until 2012 P. striiformis f. sp. tritici pathotypes were characterized by simple virulence profiles that have caused only one significant yellow rust epidemic, largely due to one susceptible variety (Enjalbert et al. 2005). However, in recent years the weighted mean of the field adult plant resistance level has become similar between all agricultural regions, these regions all being affected by PstS7 and PstS10. This trend toward homogenization could be caused by both the preference of farmers for more resistant varieties, supported by varietal recommendations, and the registration of more resistant varieties, with breeders being encouraged by demand and by changes in registration criteria promoting resistant varieties. Indeed, because of the economic importance of several wheat diseases, the registration of new varieties in the official National List includes penalties for very susceptible varieties. In this way, since 1985, the yield obtained in field trials without fungicide has accounted for half of the evaluation of yield performance of varieties, and since 1988, a bonus–penalty system has been applied to promote varieties with low yield differences when compared in yield trials with and without fungicides (CTPS 1989).

    Considering the weighted variance of the field adult plant resistance level, we found a high variance during most of the period from 1985 to 2018, with a noticeable decrease in recent years, either when we account for resistance level defined at registration only or when we include postregistration changes. These high variances during most of the period are usually related to the co-occurrence of leading varieties characterized by different resistance levels. Some susceptible varieties with superior baking quality have actually been maintained over the long term by actors in the wheat sector in the absence of other varieties available for this specific end use (Perronne et al. 2017). The decrease observed in recent years is caused by greater similarity in resistance levels of the leading varieties within a department. This similarity could be caused by the absence of highly susceptible varieties, either not registered in the National List or not chosen by farmers, and by the preferential use of varieties with a high resistance level (Cadot 2017). For instance, of the 20 leading varieties over the period from 2012 to 2018, only two had a resistance level at registration <5, and 16 varieties had a resistance level from 7 to 9.

    Spatiotemporal structure of acreage proportion characterized by effective major Yr resistance genes.

    Considering postulated effective Yr resistance genes, we found a usually low relative acreage proportion of varieties resistant to all dominant P. striiformis f. sp. tritici pathotypes and several spatiotemporal fluctuations over the whole period from 1985 to 2018. These fluctuations reflect the boom-and-bust cycles of yellow rust. Indeed, the sequential and massive release of new major Yr resistance genes in France and the rest of Europe exerts a major selective pressure on pathogen populations, which recurrently breaks down these newly introduced Yr resistance genes (Bayles et al. 2000; de Vallavieille-Pope et al. 2012). During these boom-and-bust cycles, the proportion of resistant varieties characterized by these effective major resistance genes may tend to zero before increasing because of the introduction of new Yr genes. However, even when only a low proportion of wheat was resistant because of effective major genes, most of the national wheat acreage did not appear highly affected by yellow rust. This could be explained by the sources of quantitative resistance due to many genes with partial effects (Chen 2013; Rosewarne et al. 2013) that confer long-term resistance against all pathotypes (Cowger and Brown 2019) and, when accumulated in the same variety, can induce a high level of durable resistance (Dedryver et al. 2009; Paillard et al. 2012). Moreover, the marked increase in the relative acreage proportion of wheat characterized by effective major resistance genes after an epidemic peak in most cases could be explained primarily by the choice of farmers, supported by varietal recommendations of advisory services. Indeed, from the beginning of an epidemic, it has been shown that farmers rapidly replaced newly susceptible varieties with still resistant varieties available for the same end uses and with comparable agronomic characteristics (Perronne et al. 2017). For example, the particular response observed over the period from 1987 to 1991 could be explained by the breakdown of two major genes, Yr6 and Yr9, and by the difficulty of substituting some varieties with superior baking quality (Perronne et al. 2017). The difference in responses observed between agricultural regions could be caused by differences in the level of pathogen pressure, leading to a faster and more marked varietal replacement in the regions most affected by the epidemic. So far, only one exception to the usually low relative acreage proportion of varieties resistant to all dominant P. striiformis f. sp. tritici pathotypes was observed in the south of France at the end of the 1990s. This particular case, where the relative acreage proportion of wheat characterized by effective major resistance genes was high despite the epidemic context, can be explained by the dominance of one P. striiformis f. sp. tritici pathotype affecting only one leading variety, the highly susceptible variety Victo (de Vallavieille-Pope et al. 2012).

    Spatiotemporal structure of the weighted average age of varieties.

    Considering the weighted average age of varieties, WAt, we found stable values, about 6 to 7 years, over the last three decades in most departments. These values are close to 5 to 6 years of varietal longevity for the different wheat rusts estimated by Kilpatrick (1975). However, despite a fast varietal turnover, higher than in some other countries where rusts are considered major diseases (Brennan and Byerlee 1991; Pingali 1999), the northern half of France has been subject to a recurrence of yellow rust epidemics in recent decades (de Vallavieille-Pope et al. 2012; du Cheyron et al. 2016). This result suggests that WAt is of limited interest to guide varietal replacement in order to better manage yellow rust epidemics in Western Europe. In particular, WAt does not take into account several potential causes that increase or decrease its value, such as the rapid breakdown of major resistance genes, the long-distance dispersal of spores of new virulent P. striiformis f. sp. tritici pathotypes by wind, and the durable resistance of some leading varieties.

    Main actors having a substantial influence on the level of varietal resistance in France.

    In view of the contrasting spatiotemporal changes in varietal resistance and of the multiple potential drivers proposed previously, we suggest that several actors may have had a substantial influence on these changes.

    First, breeders have increased both the field adult plant resistance level and the number of postulated major Yr resistance genes in their bread wheat genotypes. Indeed, they often use leading varieties of the previous breeding cycle as parents in their programs, obtaining new varieties by frequently intermating these leading varieties with some new sources of resistance, possibly from landraces and wild relatives (Mondal et al. 2016). In France, a substantial set of these leading varieties of recent decades were characterized by durable resistance due to both major resistance genes and genes associated with a quantitative resistance (Dedryver et al. 2009; Mallard et al. 2005; Paillard et al. 2012; Perronne et al. 2018). Consequently, breeding efforts led to both the accumulation of quantitative resistance due to several complementary genes with partial effects from different sources and the accumulation of a few major, mostly ineffective resistance genes. In addition, the phenotypic selection for yellow rust resistance promotes, intentionally or not, the sequential and massive release of major Yr race-specific resistance genes over time in several Western European countries, and breeders participated in the rapid breakdown of major resistance genes and the boom-and-bust cycles observed (Bayles et al. 2000).

    Second, registration criteria encouraged breeders to release resistant varieties (CTPS 1989). Indeed, GEVES considers several criteria related to the resistance to fungi to evaluate a variety. First, the yield obtained in field trials without fungicide accounts for half of the evaluation of yield performance of varieties since 1985. Secondly, a bonus–penalty system is applied to promote varieties with low yield differences when comparing yield trials with and without fungicide since 1988. In this context, breeders have potentially avoided proposing to register the most susceptible varieties that have several penalties and, furthermore, could not find their market with agricultural cooperatives and farmers, especially in the current context of recurrence of yellow rust epidemics. Third, when varieties are placed on the market, agricultural cooperatives can promote resistant varieties to their farmers when these varieties are otherwise equivalent in other agronomic performances, because disease resistance appears as an important criterion of varietal choice (Perronne et al. 2016). Moreover, in an epidemic context, especially at the beginning of the epidemic, farmers rapidly replace newly susceptible varieties with still resistant varieties available for the same end uses and having comparable agronomic characteristics (Perronne et al. 2017). Indeed, these newly susceptible varieties are rapidly identified by the agricultural advisory services in charge of reevaluating the resistance level after registration each year (Arvalis, Institut du Végétal), explaining the number of decreases in field resistance level at the beginning of each epidemic (du Cheyron et al. 2016).

    Conclusion and perspectives.

    The level of varietal resistance to yellow rust increased in France over the period from 1985 to 2018 because of both the accumulation of quantitative resistances due to many complementary genes with partial effects from different sources and the accumulation of several major, mostly ineffective, race-specific resistance genes. In the current context of more aggressive P. striiformis f. sp. tritici pathotypes, the proportion of wheat varieties resistant due only to known effective major Yr race-specific resistance genes has tended to zero, while rapidly, many varieties were identified as resistant to all P. striiformis f. sp. tritici pathotypes tested at the seedling stage, especially to pathotypes PstS7 and PstS10, because of unknown race-specific resistance genes or a combination of known and unknown resistance genes. Moreover, most of the national wheat acreage did not appear to be affected by the strong effect of major Yr resistance gene breakdowns observed on specific resistance effectiveness, demonstrating the major improvement performed by breeders through the accumulation of sources of quantitative resistance present in the French wheat germplasm. However, these dynamics do not allow us to quantify precisely the relative importance of the varietal improvement by breeders and the varietal turnover, that is, the tendency of farmers to replace susceptible varieties at the beginning of an epidemic. These results emphasize the importance of associating several Yr race-specific resistance genes and several genes inducing quantitative resistance in order to increase the long-term durability of yellow rust management (Brown 2015; Cowger and Brown 2019; Lof et al. 2017). This main trend of increasing yellow rust resistance can also be explained by the influence of the different actors that have led to the promotion of more resistant varieties for several decades and the rapid replacement of newly susceptible varieties with still resistant ones. However, some improvement clues exist, including the development of an early warning system for rust prevention and management at the European level based on a multistakeholder and multinetwork approach and a better sharing of communication and research infrastructures. Further increases in resistance level could also be expected in coming years in response to the Certificat d’Economie de Produits Phytosanitaires program, applicable to varieties resistant to pests since 2017 and applied since 2018, thus encouraging the cultivation of resistant varieties listed by an official service (Lorgeou et al. 2018).

    ACKNOWLEDGMENTS

    We thank FranceAgriMer for the provision of acreages of bread wheat varieties and Agreste for the total production area of bread wheat. We thank GEVES; Arvalis-Institut du Végétal; Service Régional de la Protection des Végétaux; and the wheat breeders who participated in the sampling of P. striiformis f. sp. tritici isolates. We thank L. Gérard and N. Retout for their technical assistance in the multiplication of P. striiformis f. sp. tritici isolates and pathotype tests and for plant cultivation in the greenhouse for resistance gene postulation, at INRAE BIOGER; and the editor and the two anonymous reviewers who helped to improve the manuscript.

    The author(s) declare no conflict of interest.

    LITERATURE CITED

    Funding: This analysis was supported by the RustWatch project, through two postdoctoral fellowships (to R. Perronne and F. Dubs), and funded by the European Union’s Horizon 2020 Research and Innovation Programme under grant agreement 773311. The data used were produced and formatted with support of the European Framework Programme 6 (grant 513959), BIOEXPLOIT (2006 to 2011), the European Framework Programme 6 Rex (grant 031499), Eurowheat ENDURE (2008 to 2010), the European Framework Programme 7 (grant 265865), PURE (2011 to 2014), the Innovation Fund Denmark, the Ministry of Higher Education and Science (grant 11-116241), RUSTFIGHT (2012 to 2015), the FSOV 2012-O project (2012 to 2015), the FSOV 2016-F project (2016 to 2019), the CASDAR C-2012-05 Seeds project (2012 to 2015), the Agence Nationale de la Recherche (ANR) as part of the “Investments d’Avenir” Programme (LabEx BASC; ANR-11-LABX-0034) (2012 to 2016), and the IDEEV project DivWheat (2017 to 2018).

    The author(s) declare no conflict of interest.