Variation in Arabidopsis developmental responses to oomycete infection: resilience vs changes in life history traits


  • Lucie Salvaudon,

    Corresponding author
    1. CNRS, UMR 8079, F-91405 Orsay, France
    2. AgroParisTech, UMR 8079, F-91405 Orsay, France
    • Univ Paris-Sud, Laboratoire Ecologie Systématique et Evolution, UMR 8079, F 91405 Orsay, France
    Search for more papers by this author
  • Jacqui A. Shykoff

    1. Univ Paris-Sud, Laboratoire Ecologie Systématique et Evolution, UMR 8079, F 91405 Orsay, France
    2. CNRS, UMR 8079, F-91405 Orsay, France
    3. AgroParisTech, UMR 8079, F-91405 Orsay, France
    Search for more papers by this author

Author for correspondence:

Lucie Salvaudon

Tel: +33 1 69156529



  • Although plant resistance to aggressors has been well described, there is still little knowledge about the mechanisms underlying their tolerance to pathogens. Tolerance often appears to be mediated by changes in life history traits, shifting host resource investment from growth to reproduction, but whether host phenotype modifications induced after attack are adaptive is not always clear.
  • Here, we investigated the details of the impact of Hyaloperonospora arabidopsidis infection on several biomass, phenology and architectural traits of Arabidopsis thaliana, for three pathogen genotypes and three host plant genotypes that have been shown previously to differ greatly in fecundity and tolerance to infection.
  • We found that, although host genotype explains most of the variance in life history traits, these three lines differ critically in their response to infection, with delays and biomass losses at bolting, together with changes in inflorescence architecture, observed at one extreme host line, and an advantage at bolting for infected plants and no inflorescence alteration for the other.
  • These results suggest that the differences in tolerance observed previously in this pathosystem do not involve plasticity in inflorescence architecture, but may arise from induced changes at the vegetative stage, before plant transition to reproduction.


Defense against pathogens is generally classified into two categories: resistance strategies that limit infection or proliferation of a parasite, and tolerance strategies that reduce or compensate for the negative impact of infection on host fitness (i.e. virulence). These two aspects need not be exclusive and, although the costs of these defenses have been demonstrated (Simms & Triplett, 1994; Agrawal et al., 1999; Fornoni & Nunez-Farfan, 2000; Koskela et al., 2002), there is actually little evidence that they trade off against each other (Fornoni et al., 2004; Nunez-Farfan et al., 2007). Genetic variation in both strategies has been demonstrated in multiple species (Peever et al., 2000; Thrall et al., 2001; Kover & Schaal, 2002; Laine, 2004; Råberg et al., 2007) and they both play fundamental – although different – roles in the coevolution of hosts and their pathogens (Roy & Kirchner, 2000; Miller et al., 2006; Salvaudon et al., 2008a; Best et al., 2010). Mechanisms for resistance are numerous and have been studied extensively (Akira et al., 2006; Chisholm et al., 2006; Jones & Dangl, 2006), but the mechanisms underlying tolerance to pathogens are still not well understood (Baucom & de Roode, 2011) and tolerance is generally considered to be a form of phenotypic plasticity that alters life history traits depending on the presence of enemies. Because parasitism may alter optimal resource allocation among growth, survival and reproduction, infected hosts should benefit from increasing allocation to quicker reproduction at the expense of investments in the future (Forbes, 1993; Perrin et al., 1996). Such changes, including reduced age at maturity, increased reproductive effort, etc., have indeed been repeatedly observed in animal systems (Michalakis & Hochberg, 1994; Agnew et al., 2000). In plant systems, similar life history responses have mostly been studied in response to herbivory (McNaughton, 1983; Karban et al., 1999; Stowe et al., 2000), and comparatively fewer pathogen-induced changes have been reported (but see Shykoff & Kaltz, 1997, 1998; Korves & Bergelson, 2003; Pagán et al., 2008, 2009).

The relationship between life history trait changes and fitness compensation in the presence of enemies, however, might not always be straightforward. Pathogen-induced changes can also be attributed to the pathological effects disturbing a host's physiology, or even to adaptive manipulation by the parasite (Michalakis & Hochberg, 1994; Shykoff & Kaltz, 1998). The evolutionary history of host–pathogen associations might also lead to unexpected situations in which host fitness is actually lower in the absence of damage, as is sometimes observed with herbivores (de Mazancourt et al., 2001) or pathogens (Dedeine et al., 2001; Salvaudon et al., 2008b). This phenomenon has been attributed to ‘evolved dependence’, whereby host plants have adapted to an environment that includes recurrent damage – with the associated resource allocation optimum – and hence become maladapted when this predictable damage is removed (de Mazancourt et al., 2005). Disentangling the contributions of pathogen virulence, host re-allocation strategies and their evolutionary history is thus one of the challenges in understanding tolerance and its impact on host fitness.

Arabidopsis thaliana has been a model of choice for host interactions with a wide range of pathogens, including viruses (Pagán et al., 2010), bacteria (Buell, 2002; Jakob et al., 2002), fungi (Lipka et al., 2010) and oomycetes (Coates & Beynon, 2010). Its short life cycle is ideal for the study of tolerance and pathogen-induced changes in life history traits. Arabidopsis families exhibit large variation in resistance (Bakker et al., 2006; Goss & Bergelson, 2006; Nemri et al., 2010), but also in a variety of life history traits (Weinig et al., 2003a; Koornneef et al., 2004; Bergelson & Roux, 2010) as well as in their tolerance to pathogens (Goss & Bergelson, 2007; Pagán et al., 2008; Salvaudon et al., 2008b) and herbivory (Weinig et al., 2003b). Previous studies on pathogen virulence have also shown that some A. thaliana families have little or no fitness loss (Goss & Bergelson, 2007; Salvaudon et al., 2008b). In particular, Salvaudon et al. (2008b) demonstrated that, among a range of host families, the extent of fitness loss caused by infection with the oomycete pathogen Hyaloperonospora arabidopsidis (parasitica) was positively correlated with the fecundity of the host plant in the absence of disease. At the extremes of this range, the host line with the highest fecundity suffered from high pathogen virulence, whereas the least fecund line actually increased its seed production when infected, suggesting a potential role for evolved dependence in this pathosystem. In order to investigate whether this observed variation in tolerance to H. arabidopsidis was linked to variation in the plasticity of life history traits, we performed experimental infections on three A. thaliana lines among those previously investigated, using the two extremes and one intermediate in fecundity. We tested the impact of host genotype, infection status, pathogen genotype and their interactions on biomass, phenology and architectural traits of the host plants at different stages of the life cycle. We show here that the variation in tolerance observed previously is reflected in a variation in pathogen-induced changes in life history traits, but that, unexpectedly, the largest changes – especially a delay in bolting and increase in branches and fruit numbers – were observed for the host line with the lowest tolerance, whereas the other extreme appeared to be more resilient to such modifications. These results suggest that tolerance mechanisms might not necessarily involve plastic phenology or architectural traits.

Materials and Methods

Hyaloperonospora arabidopsidis (Gäum.) (formerly Hparasitica) is a pathogenic oomycete specific to Arabidopsis thaliana (L.) Heynh. (Brassicaceae). This biotrophic parasite produces both sexual stages – oospores that remain within leaves until host death – and asexual conidiospores on the surface of infected leaves, which constitute the main symptom of infection. Reproduction by oospores is critical for the survival of this pathogen between the active growing seasons of its host, but plays no role in within-season dynamics. Asexual reproduction via conidiospores, however, is responsible for dissemination and epidemic dynamics within populations and seasons. This pathogen is usually not lethal for its host and symptoms are quite subtle in natural populations, but it nonetheless constitutes a selective pressure on A. thaliana, as 27 resistance genes targeted against H. arabidopsidis have been reported (Slusarenko & Schlaich, 2003), some of which appear to be under positive or balancing selection (Bakker et al., 2006). Three different strains of H. arabidopsidis were used in this experiment: ‘Emwa’ originated from an initial collection in East Malling (UK) and was provided by the Sainsbury Laboratory (John Innes Center, Norwich, UK), ‘Fri3’ was collected as oospores in Fribourg (Switzerland) and ‘Ors3’ originated from conidiospores collected in Orsay (France). All strains were maintained as asexual populations for several generations on susceptible A. thaliana accessions by successive conidiospore transmissions, and thus are likely to be genetically homogeneous.

The three Arabidopsis accessions were all susceptible to the three H. arabidopsidis strains and were chosen on the basis of their different fecundities and tolerance to infection observed previously (Salvaudon et al., 2008b). The host family ‘Pyr’ was originally sampled in the Pyrenees in the south of France. It was the least fecund line, but also the most tolerant, even experiencing higher fitness when exposed to H. arabidopsidis infection. By contrast, the line ‘Gb’, sampled in Great Britain, was the most fecund line and suffered fitness losses when infected. The third line, ‘Sue’, was sampled in Sweden and had intermediate fecundity with no detectable positive or negative impact of infection. All families were maintained by selfing.

In order to determine the respective effects of host genotype, infection, pathogen genotype and their interactions on the life history traits of A. thaliana, a full-factorial experiment with all possible combinations of all treatments was performed. The resulting 12 treatments combined the three host families (‘Pyr’, ‘Sue’, ‘Gb’) and four inoculation types (water-inoculated controls, ‘Emwa’, ‘Fri3’ and ‘Ors3’), with 40 replicates of each treatment. All plants were sown on the same day in 5 × 5 × 5 cm3 compost soil pots in a fully randomized pattern and placed in the dark at 5°C for 1 wk in order to synchronize germination. Seedlings were then grown for 16 d in the glasshouse (23°C : 15°C, day : night). The young plants were inoculated with an 8-μl drop of conidiospore suspension for the two largest leaves and a 4-μl drop on all smaller leaves of the corresponding parasite strain diluted at 5 × 104 spores ml−1, or of pure water for the control treatment. Each plant was placed in its own closed transparent plastic cylinder to prevent contamination and to maintain high hygrometry, and then randomly arranged in a growth chamber (10 h : 14 h, light : dark photoperiod, 16°C ± 3°C average temperature and 95–100% hygrometry). The number of sporulating leaves was recorded 12 d after inoculation. After 25 d in the chamber, the plants were moved again into the glasshouse (natural photoperiod, 29°C : 20°C, day : night), the covers of the plastic cylinders were removed and they were watered ad libitum until senescence. From the first day in the glasshouse, the plants were checked three times per week to record their dates of bolting (beginning of inflorescence development) and flowering (first opened flower). Among the 40 replicates of each treatment combination, 10 were chosen at random to be sacrificed at bolting and 10 others to be sacrificed at flowering for estimates of plant biomass. The number of rosette leaves was counted, and sacrificed plants were photographed, cut at the base of the rosette and immediately weighed to the precision of 1/1000 g (fresh mass). Dry mass was obtained by re-weighing the dried plants after 10 d in a 60°C drying oven. Plant surface areas at bolting and flowering were calculated from the photographs using Image J (version 1.36b, Wayne Rasband, National Institutes of Health, Bethesda, MD, USA) software. Measures at flowering also included inflorescence height. The remaining 20 replicates were kept until senescence, when architectural parameters were measured: number of inflorescences, height of the first branch, height of the first fruit, number of all flowering branches and number of fruits. Seed production was estimated as the total seed weight, collected during the whole fructification period.

Statistical analyses

All statistical and principal component analyses were performed with JMP software version 5.1.2 (SAS Institute, Cary, NC, USA). The number of sporulating leaves in the nine H. arabidopsidis inoculation treatments (excluding water-inoculated controls) was analyzed by an ordinal logistic test for the effect of host family, pathogen strain and their interaction. Because the life history traits of the plants are usually highly correlated, and in order to avoid bias caused by multiple ANOVAs, we analyzed all the measured traits in multivariate analyses of variance (MANOVAs) at the three different developmental stages: bolting, flowering (using the data from sacrificed plants at these two stages) and senescence (for the remaining plants).The MANOVA models tested the effects of host family, inoculation type (controls vs pathogen inoculation), pathogen strain nested within the inoculation type, interactions between host family and inoculation, and interactions between host family and pathogen strain (nested within inoculation type). For each variable, we ensured that residuals were correctly distributed among predicted values and that they followed a normal distribution. For this purpose, we log-transformed, with the function y = log(x + 1), the data for the date of bolting, the number of leaves at bolting, the surface at bolting and the inflorescence height at flowering. Fresh and dry masses at bolting and dry mass at flowering were analyzed using ranks of the values, as these variables were highly asymmetrical. We chose not to analyze date of flowering, number of inflorescences and height of the first branch because they did not satisfy assumptions of normality. Additional individual ANOVAs were also performed for each variable (Supporting Information Tables S1–S13) as an indication of how they contributed to the observed differences among treatments, and principal component analyses were carried out on each group of variables in order to provide a graphical description of this contribution. The principal components were calculated in JMP as independent linear combinations of the standardized original variables with the same transformation as in the MANOVAs, using their correlation matrix. Of the 480 total plants, we excluded the data for three control plants because of doubts about their possible contamination or errors during inoculation, and for 14 plants (five replicates for bolting measures, four for flowering and five for architectural measures) that died of unknown causes before measurements.


Pathogen infection success

The intensity of H. arabidopsidis infection was estimated as the number of host leaves with conidiospores 12 d after inoculation. Because this pathogen has a latency of 5 d on average under the conditions of this experiment, this estimate thus combines the success of the initial inoculation and spread of infection, both systemically within the host and via self re-inoculation by the first conidiospores. Most (97.2%) of the 360 inoculated plants became infected, and all combinations of host families and pathogen strain were compatible. However, host families varied significantly for quantitative success of infection (ordinal logistic test, Wald χ22df = 123.73, P < 0.0001), with a smaller number of infected leaves on the host ‘Sue’ compared with ‘Pyr’ and ‘Gb’. The H. arabidopsidis genotypes also differed significantly, as the strain ‘Fri3’ had greater infection intensity than ‘Emwa’ and ‘Ors3’ (Wald χ22df = 19.76, P = 0.0001; Fig. 1). Specific combinations of host and parasite genotypes also differed significantly for the intensity of infection (host–parasite genotype interaction; Wald χ24df = 14.55, P = 0.0057).

Figure 1.

Average number of Arabidopsis thaliana leaves (± SE) with conidiospore sporulation 12 d after inoculation for the 360 plants of the Hyaloperonospora arabidopsidis infection treatments and the three host plant families. Black circles, H. arabidopsidis strain ‘Fri3’; white circles, strain ‘Ors3’; gray circles, strain ‘Emwa’.

Life history traits at bolting and flowering

The date of bolting was determined as the beginning of inflorescence development, which corresponds to the transition between the vegetative and reproductive phases. At this time point, the rosette biomass was estimated as the number of leaves, surface, and fresh and dry masses. The combination of these traits, as well as the bolting date, was analyzed with a MANOVA (Table 1). The host family explained most of the variance in these traits (MANOVA contrasts on host line effects, ‘Gb’ vs ‘Pyr’, ‘Gb’ vs ‘Sue’, ‘Pyr’ vs ‘Sue’: F5,99 = 239.1, F5,99 = 20.7 and F5,99 = 145.9, respectively; all P < 0.0001), and there was an overall significant difference in these traits between controls and infected plants. There was, nonetheless, a significant interaction between host line and inoculation type, which was also found in separate ANOVAs for each individual variable (Tables S1–S5), demonstrating that the infection-induced changes were different among host families. This is visualized with principal component descriptive statistics in Fig. 2, where the first principal component separates the three host lines, and the second component (representing a smaller portion of the total variance) separates the control and infected treatments. Indeed, life history traits at bolting were different between control and inoculated plants for all three families (MANOVA contrasts on host line × inoculation type effects, ‘Gb’ controls vs inoculated, ‘Sue’ controls vs inoculated, ‘Pyr’ controls vs inoculated: F5,99 = 10.9, F5,99 = 7.3 and F5,99 = 3.6, respectively; P < 0.0001 for ‘Gb’ and ‘Sue’, P = 0.006 for ‘Pyr’), but these changes corresponded to a tendency for biomass reduction and bolting delay in ‘Gb’- and ‘Sue’-infected plants, whereas an opposite trend was observed in the host family ‘Pyr’ (Fig. 2). There was little difference among genotypes of infecting H. arabidopsidis strains, this effect being slightly below the significance threshold. For those plants analyzed at first flower opening, biomass traits and inflorescence height varied among host family and inoculation type only (see MANOVA in Table 2).

Table 1. MANOVA on life history traits at bolting
EffectTestValue or F DF(Num,Den)P-value
Host familyWilks' λ0.05510,198 < 0.0001
Inoculation type F 11.47 5,99< 0.0001
Parasite strain (Inoculation)Wilks' λ0.8410,1980.061
Host × InoculationWilks' λ0.6310,198< 0.0001
Host × Parasite (Inoculation)Wilks' λ0.8620,329.30.76
Table 2. MANOVA on life history traits at the first open flower
EffectTestValue or FDF(Num,Den)P-value
Host familyWilks' λ0.07510,200< 0.0001
Inoculation type F 7.505,100< 0.0001
Parasite strain(Inoculation)Wilks' λ0.9710,2000.99
Host × InoculationWilks' λ0.8910,2000.29
Host × Parasite(Inoculation)Wilks' λ0.8320,332.610.55
Figure 2.

Principal components representation of the variables included in the MANOVA on Arabidopsis thaliana host life history traits at bolting (log-transformed date of bolting, number of rosette leaves and surface at bolting, rank order of fresh and dry rosette weights). Average values (± SE) for each host line and inoculation status on the first two principal components (representing 91.36% and 4.87% of the total variance, respectively; Supporting Information Table S14). Open symbols, mock-inoculated controls; closed symbols, Hyaloperonospora arabidopsidis-inoculated treatments. Circles, Arabidopsis host families ‘Gb’; squares, ‘Sue’; triangles, ‘Pyr’.

Plant architecture and seed production

For the analysis of inflorescence architecture and seed production, we combined the data for total seed weight (collected progressively as they fell from mature fruits) and final number of branches, fruits and height of the lowest fruit when the plants reached senescence (Table 3), which all had correlation values above 0.5. Again, most of the variance was caused by host families, with the host line ‘Gb’ being the most fecund and presenting the greatest number of branches and fruits, whereas ‘Pyr’ was lowest for these traits and ‘Sue’ was intermediate (Fig. 3, Table 3, MANOVA contrasts on host line effects, ‘Gb’ vs ‘Pyr’, ‘Gb’ vs ‘Sue’, ‘Pyr’ vs ‘Sue’: F4,218 = 117.8, F4,218 = 17.8 and F4,218 = 55.7, respectively; all P < 0.0001). The infection by H. arabidopsidis also had a significant impact on the inflorescences, with a tendency to reduce their height and increase the number of branches and fruits (Fig. 3). These infection-induced changes were significantly different among the host lines, being larger for ‘Gb’ than ‘Sue’, and nonsignificant for ‘Pyr’ (Fig. 3, Table 3, MANOVA contrasts on host line by inoculation type effects, ‘Gb’ controls vs inoculated: F4,218 = 20.9, P < 0.0001; ‘Sue’ controls vs inoculated: F4,218 = 3.5, P = 0.009; ‘Pyr’ controls vs inoculated: F4,218 = 3.6, P = 0.69). However, as opposed to the other architectural variables, seed production was not affected by the infection and a separate ANOVA revealed no significant effects of inoculation type or its interaction with host line (Tables 4, S11–S13). Indeed, although the host line by parasite strain interaction was significant in the separate ANOVA for seed production (Table 4), this was caused by a single switch in rank order among the three host lines in the ‘Emwa’ strain treatment, whereas there was no significant difference in seed production among the controls and the three H. arabidopsidis strain treatments within each host line. Finally, there was no overall significant effect of the H. arabidopsidis strain or host line by parasite strain interaction in the MANOVA (Table 3).

Table 3. MANOVA on life history traits at the end of life cycle
EffectTestValue or FDF(Num,Den)P-value
Host familyWilks' λ0.288,436< 0.0001
Inoculation typeF 12.944,218< 0.0001
Parasite strain(Inoculation)Wilks' λ0.968,4360.35
Host × InoculationWilks' λ0.818,436< 0.0001
Host × Parasite(Inoculation)Wilks' λ0.8916,666.640.08
Table 4. ANOVA on seed production
EffectDFSum of squaresF ratio P
Host family254577.5474.18< 0.0001
Inoculation type1229.300.620.43
Parasite strain (inoculation)22202.592.990.052
Host × inoculation2165.940.230.80
Host × parasite (inoculation)45958.454.050.0034
Figure 3.

Principal components representation of the variables included in the MANOVA on Arabidopsis thaliana host life history traits at the end of the life cycle (total seed weight, number of branches with fruits, number of fruits and height of the lowest fruit). Average values (± SE) for each host line and inoculation status on the first two principal components (representing 77.17% and 14.37% of the total variance, respectively; Table S15). Open symbols, mock-inoculated controls; closed symbols, Hyaloperonospora arabidopsidis-inoculated treatments. Circles, Arabidopsis host families ‘Gb’; squares, ‘Sue’; triangles, ‘Pyr’.


In a previous study (Salvaudon et al., 2008b), we reported a large variation in the virulence of H. arabidopsidis on A. thaliana, with a larger reduction in seed production in the most fecund host line, an increase in the least fecund line and no difference in intermediate lines, which might underlie differences in their tolerance to this pathogen. Here, we found that these three host families also differ in their plasticity of other life history traits in the presence of the disease. Indeed, for combinations of traits related to biomass, phenology, architecture and fitness at the time of bolting and at the end of the life cycle, we observed significant interactions between host families and their infection status, with large phenotypic modifications in the host line ‘Gb’, smaller modifications to the line ‘Sue’ and little to no modifications in the host line ‘Pyr’. For both ‘Gb’ and ‘Sue’, the H. arabidopsidis-induced changes corresponded to a delayed bolting with a reduction in rosette biomass (leaf number, surface and weight, Fig. 2), and a more ‘bushy’-like appearance of the inflorescence at the end of the life cycle, as infected plants presented more but lower branches and fruits (Fig. 3). The intensity of these changes was greater for ‘Gb’. By contrast, the host line ‘Pyr’ presented opposite effects of infection at the time of bolting, although with little amplitude – towards earlier bolting date with higher biomass – and no significant changes in architecture (Figs 2, 3). These discrepancies in individual host responses to infection were significant, despite the fact that host genotypes explained most of the variance in all traits. More specifically, the rank order in fecundity of the three tested host lines (independent of their infection status) was the same as observed in Salvaudon et al. (2008b).

Abiotic stresses are known to influence host response to infection (Wolinska & King, 2009) and herbivore attack (Wise & Abrahamson, 2007); therefore, in this experiment, we wanted to test the plasticity response of A. thaliana to infection with H. arabidopsidis in the absence of other sources of stress for the plant. Therefore, plants enjoyed optimal conditions for the whole life cycle, including ab libitum watering, unlike in our previous experiment in which we imposed water stress as soon as fruiting began to hasten completion of the life cycle. Here, we observed no large differences in virulence expression among host lines, probably because the long fructification period allowed even relatively small and tardy plants to produce their full complement of seeds, whereas such plants would have had a reduced window for seed production in the previous experiment (Salvaudon et al., 2008b). This discrepancy emphasizes the fact that optimal glasshouse conditions might yield a different picture of tolerance traits than that which really happens under natural conditions. However, observations early in the life cycle are consistent with these previous results. Indeed, here we observed that infected ‘Pyr’ plants were younger and larger at bolting, whereas ‘Sue’ and ‘Gb’ were smaller and older than their uninfected counterparts. Hence, although variation in tolerance based on final fitness could not be confirmed here, the similar ranking in biomass at bolting compared with the previous results suggests that the ‘Pyr’ line is indeed more tolerant to H. arabidopsidis damage than the other lines and performs better in the presence of the pathogen. Furthermore, although host lines also varied in their resistance to the three H. arabidopsidis strains tested, ‘Pyr’ and ‘Gb’ had a similar infection intensity, and therefore it seems unlikely that their different phenotype responses resulted from differential pathogen damage, unless as yet unknown resistance mechanisms acting on other aspects of the pathogen interaction – at the level of sexual oospore production, for instance – played a role.

Tolerance mechanisms are not easily demonstrated, because many confounding trade-offs, environmental and evolutionary history factors all contribute to the realized fitness; however, most proposed mechanisms involve compensation strategies that are induced following damage, such as changes in photosynthetic rates for plants, use of stored resources, phenology changes, etc. (Stowe et al., 2000; Tiffin, 2000), and, more generally, adaptive plasticity in resource allocation, which allows for increased investment to reproduction when subjected to damage (Agnew et al., 2000). Indeed, in Arabidopsis, it has been shown that tolerance to herbivory is associated with increased branching and a longer reproductive period (Weinig et al., 2003b), or to an increase in relative inflorescence growth and a reduced reproductive period following Cucumber mosaic virus infection (Pagán et al., 2008). By contrast, we found that the most tolerant host line to H. arabidopsidis was also the most resilient to phenotypic changes during inflorescence development, which suggests that phenotypic plasticity after the transition to the reproductive stage is not linked with tolerance to this pathogen. Alterations in host life history traits can also result from pathogenic byproduct effects of the infection, or even an adaptive manipulation of host traits by the parasite in favor of its own dissemination (Michalakis & Hochberg, 1994). Although the latter hypothesis cannot be totally excluded, there is little evidence that H. arabidopsidis would benefit from the inflorescence architectural changes we observed, as its transmission stages – sexual oospores within host tissues and asexual conidiospores on the leaf surface – are produced in symptomatic rosette leaves and rarely observed on inflorescences. The changes in host life history traits on the least tolerant hosts are thus probably a result of other pathological damage caused by the parasite.

Our results suggest that A. thaliana tolerance to H. arabidopsidis is expressed mainly at the vegetative stage and compensates for rosette biomass losses. The absence of inflorescence alterations in the tolerant host ‘Pyr’ also suggests the intriguing possibility that resilience to pathogen-induced alterations of the phenotype could, in itself, reflect a defense strategy against H. arabidopsidis damage, although a larger range of tolerant host families would be needed to test this hypothesis, as we cannot, at this stage, draw general conclusions from a single Arabidopsis genotype. The genotype of H. arabidopsidis and the interaction of host and pathogen genotypes had no significant effect on life history trait MANOVAs; there was thus no evidence that tolerance is specific to particular pathogen genotypes in this pathosystem. However, it is interesting to note that, among the pathogens investigated in A. thaliana tolerance studies, H. arabidopsidis is the only specialist, and hence probably has a stronger co-evolution history with this host. Specialist and generalist herbivores can have different impacts on plant resistance (Lankau, 2007), and different types of pathogen, for instance biotrophic vs necrotrophic pathogens, trigger different resistance pathways (Glazebrook, 2005; Kliebenstein & Rowe, 2008). Whether they could also trigger different tolerance mechanisms would be an interesting hypothesis that could explain why the tolerance traits we observed differed from those of other studies with generalist pathogens or herbivores. The role of tolerance strategies in the co-evolution of hosts and their enemies is still not well known, but could be more important than previously assumed (Stinchcombe, 2002).

In Arabidopsis, local adaptation to environmental conditions shapes a large variation in many life history traits (Koornneef et al., 2004; Atwell et al., 2010; Brachi et al., 2010), including resistance (Goss & Bergelson, 2006; Kover & Cheverud, 2007; Nemri et al., 2010) and tolerance to pathogens (Kover & Schaal, 2002; Gao et al., 2009). The phenotypic responses to enemies thus results from an interplay between the evolutionary history of the host, which determines reproductive and defense strategies, the pathological effects of the enemy encountered and all other environmental factors – for instance if conditions are limiting – that may alter the efficiency of this response. The array of A. thaliana hosts investigated by us illustrates well how this interplay can result in drastic differential responses to H. arabidopsidis infection. As opposed to resistance, whose efficiency is determined jointly by the host and parasite genotype (Allen et al., 2004; Salvaudon et al., 2005, 2007; Coates & Beynon, 2010), tolerance to H. arabidopsidis appears to be determined mainly by the host plant, but to involve different mechanisms, with no architectural compensations at the reproductive stage, than those described for other pathogens or herbivores.


We thank L. Saunois for technical assistance, and C. Manificat and J.-S. Finger for help in data collection. We are also grateful to V. Héraudet for maintaining the H. arabidopsidis strain collection and for her useful help and advice in experimental design. L. Salvaudon is funded by the European Communities 7th Framework Programme (FP7/2007-2013) under Grant Agreement PIOF-GA-2009-236011.