Model system

Sessile oak is a broad-leaved deciduous tree species occurring over most of Europe and of high environmental, cultural and economic value. The timing of oak flushing depends mostly on temperature (Vitasse et al. 2009a; Laube et al. 2013), with only a secondary role of photoperiod (Basler & Körner 2012). Indeed, budburst follows a dormancy period encompassing an endodormancy phase in early winter which is generally released by chilling (cold) temperatures, and an ecodormancy phase in early spring which is sensitive to forcing (warm) temperatures (reviewed in Lang 1987). Powdery mildew is one of the most common fungal diseases of oaks in Europe (Mougou, Dutech & Desprez-Loustau 2008) and the main source of biotic damage in the study area (personal observations). Powdery mildew fungi are obligate biotroph parasites, infecting photosynthetic organs (Glawe 2008). In oaks, the disease is caused by a complex of several closely related species: Erysiphe alphitoides, E. hypophylla, E. quercicola and Phyllactinia guttata, with E. alphitoides currently highly predominant but probably of exotic origin (Mougou, Dutech & Desprez-Loustau 2008; Mougou-Hamdane et al. 2010). Oaks display ontogenic resistance to powdery mildew: only young expanding leaves are susceptible to the disease (Edwards & Ayres 1982). Like oaks, the fungus is affected by seasonality. Oak powdery mildew mostly overwinters as chasmothecia (sexual fruiting bodies acting as dormancy organs), which release the ascospores that form the primary inoculum in spring. The timing of ascospore release in spring was shown to be related to winter temperature, with earlier ascospore release after milder winter conditions (Marçais et al. 2009). Infection results in the production of a mycelium, which produces huge numbers of conidiospores (asexual spores) at the leaf surface. Conidiospores constitute the secondary inoculum, with subsequent infection cycles during the course of the season resulting in a build-up of fungal populations. The continuous availability of susceptible leaves in an oak stand during the growing season results both from differences in flushing time between trees and the occurrence of several growth flushes in a tree during one season. Foliar infection causes a decrease in net photosynthesis and a more general change in metabolism, which may affect tree growth and survival, especially for seedlings (Hewitt & Ayres 1975; Marçais & Desprez-Loustau 2014).

Study area

The study was carried out in two valleys of the Pyrenees in southern France (from 43°15′N, 00°44′W to 42°52′N, 00°07′E) (Vitasse et al. 2009b). A first transect was set up in the Gave valley (Hautes Pyrénées), and a second was established in the Ossau valley (Pyrénées Atlantiques) (Fig. 1) (Table 1). These two valleys, located 30 km apart, are parallel and display an increase in elevation from north to south. The region is characterized by a temperate oceanic climate (see Table 2 and Fig. S1 for meteorological details). Mean annual temperatures decrease linearly with elevation, by about 0.43 °C for every 100 m increase in elevation (mean from 2005 to 2011). In total, 14 natural established oak populations on south-facing slopes were monitored for the different experiments, from 131 to 1630 m above sea level, in hill and mountain vegetation belts (see Table 1). The range of elevation extended over 1499 m in the Gave valley and 1355 m in the Ossau valley. We used the terms ‘low elevation’ (< 500 m asl), ‘medium elevation’ (500–1000 m asl) and ‘high elevation’ (> 1000 m asl) to describe populations.

Meteorological measurements

Air temperature was recorded with data loggers (HOBO Pro RH/Temp, Onset Computer Corporation, Bourne, MA, USA) at each study site (14 sites). All sensors were intercalibrated in the laboratory before installation. At each site, sensors were installed 1.5 m above the ground in an open area close to the studied population (10–100 m away from the plot, at the same elevation). Sensors were protected by a white plastic shelter, to prevent exposure to rain or direct sunlight. Data were recorded hourly, from 1 January 2005 to 31 October 2012. Several weather stations suffered short periods of technical failure. Years for which spring data were missing were not considered in the spring frost analyses.

Oak phenology monitoring

The timing of flushing was monitored along the two elevation gradients, for 7 years (2005–2007 and 2009–2012) on adult trees and 2 years (2011–2012) on seedlings (Table 1). In each of the 14 populations, 10–34 dominant adult trees of similar height (about 15 m) were monitored. The mean distance between studied trees within a population was 109 m in the Gave valley and 50 m in the Ossau valley. Trees were also selected for their phenology (first assessed in 2005), with the aim of capturing the widest possible variation (i.e. earliest and latest trees) for each population. We monitored 60–80 natural seedlings (20–30 cm high) protected in enclosed plots of 20 seedlings in the two low-elevation populations in the Gave valley. Bud development was monitored at 10-day intervals, in each population, from March to June. Observations were made by one or two trained observers with binoculars (magnifying power: ×10), about 15 m away from adult trees. At each date, the proportion of buds at a given phenological stage was assessed for each adult tree (Vitasse et al. 2009a). We considered two successive stages of interest in terms of frost and disease damage: budburst and leaf unfolding. The date of budburst was defined as the day on which 50% of the tree buds had burst and the first leaves had started to emerge (first stage of low freezing resistance) (Lenz et al. 2013). The date of leaf unfolding was defined as the date on which the first leaves of 50% of the buds had unfolded (beginning of the period of high susceptibility to powdery mildew) (Edwards & Ayres 1982). For seedlings, we considered only the stage of the apical bud (i.e. budburst was the date on which the apical bud burst and its first leaves began to emerge, and leaf unfolding was the date on which the first leaf of the apical bud was unfolded). The dates of the two stages were estimated, for each selected adult tree and seedling, by linear regression between two measurement campaigns. Finally, for each population, budburst and leaf unfolding dates were calculated as the means for the adult trees sampled. We selected the first and last quartiles as representative of the early- and late-flushing trees, respectively, per population and per year, for both budburst and leaf unfolding dates. Finally, the date of leaf senescence (i.e. the time point at which 50% of the leaves of the tree had changed colour or fallen) was monitored similarly along the two gradients, from September to December. The canopy duration was thus estimated as the difference in days between the date of leaf senescence and the date of leaf unfolding.

Powdery mildew phenology monitoring

Phenological observations focused on spore release in spring in five Gave valley populations, assessed by passive trapping (see Table 1). First data were acquired previously (Desprez-Loustau et al. 2010) by using biological traps and infection counts data. In that case, young seedlings grown in a pathogen-free environment until susceptible stage were exposed to natural inoculum in each of the five stands along the altitudinal gradient in 2006 and 2007. For practical reasons, only one trap composed of 13 seedlings (approximately 200 cm² of susceptible area), changed every 10 days, could be used at each date and altitude from early April to the end of June (see complete description in Desprez-Loustau et al. 2010). In order to obtain data in more years and to tackle variation in fungal phenology within stands, this previous study was extended in 2011 and 2012 with a modified trapping method. Here, we used a simple trap device constituted of Whatman^® filter paper (24 cm in diameter), sprayed with Tris-EDTA buffer (dilution 4X, pH = 8) before exposure to improve trapping capacity and spore preservation, that was fixed on a horizontal wooden support, in combination with molecular specific detection of the fungus (Schweigkofler, O'Donnell & Garbelotto 2004; Garbelotto, Smith & Schweigkofler 2008). This allowed us to use several traps in each site at each exposition period. In each oak stand (see Table 1), eight traps were hanged in the canopy of trees, at a height of about 10 m above ground level. In order to take into account the possible variation in fungal phenology linked to tree phenology, four early- and four late-flushing trees, distributed throughout the stand, were selected (based on 2005–2007 and 2009–2010 assessments). One trap, fixed to a 50-cm-tall wooden stake, was also placed in each enclosed plot of seedlings used for phenology monitoring. Filters were changed at regular intervals (15 days on average), from March to September in 2011, and from March to July in 2012. At the end of the exposure period, filters were individually placed in paper bags and brought back to the laboratory. Upon arrival, each filter was dried at ambient air temperature and then individually placed in a plastic bag and stored at −80 °C until further processing. The spores were washed off the filters, by dispensing 40 ml of Tris-EDTA buffer warmed at 65 °C into each plastic bag (Schweigkofler, O'Donnell & Garbelotto 2004; Garbelotto, Smith & Schweigkofler 2008). A spore suspension was generated by backward and forward movements and friction. This suspension was poured into a Falcon^® 50-ml tube (washings from 1 filter per tube). Tubes were stored at −20 °C. We concentrated the spores on a support that could be used for DNA extraction, by filtering them with a filtration pump (Combisart^® Systems from Sartorius Stedim Biotech). For each tube, the spores were filtered on a 47-mm-diameter sterile nitrocellulose membrane with 0.45-μm-diameter pores (Sartorius Stedim Biotech), a diameter smaller than that of the spores. The membranes were cut into two equal halves with a scalpel, and each half was placed in a 2-ml Eppendorf^® tube with two 2-mm-diameter steel beads. The samples were lyophilized and shaken for 2 min, at 30 Hz, in a Qiagen Tissue Lyser^®. DNA was then extracted with a kit, used according to the manufacturer's instructions (Invisorb^® Spin Plant Mini Kit from Invitek). For the amplification of specific fragments targeting E. alphitoides, E. quercicola and E. hypophylla, the oligonucleotide primers ery-F (5′ CTC AGT CGT GGC ATC TGC T 3′) (Mougou-Hamdane et al. 2010) and o-micro-rev (5′ CGC CTC CAG CCG AAA CC 3′) (Heuser & Zimmer 2002), binding to the ITS1 region of rDNA, were used for PCR. For each sample, we used 1.5 μl of pure and 1:10-diluted DNA extracts (to dilute out potential PCR inhibitors) for PCR. A mixture of the following components was added to each DNA extract, for a final volume of 20 μl: 13.55 μl of distilled water, 2 μl of 10 ×  PCR buffer (Life Technologies, Karlsruhe, Germany), 1.2 μl of 25 mM MgCl₂, 0.8 μl of dNTP, 0.4 μl of each primer and 0.15 μl of Taq DNA polymerase (Life Technologies, Karlsruhe, Germany). PCR was performed over 45 cycles (30 s at 94 °C, 40 s at 55 °C, 50 s at 72 °C) preceded by a primary denaturation phase (3 min at 94 °C) and followed by a terminal elongation phase (7 min at 72 °C). The presence or absence of DNA was determined by agarose gel electrophoresis (2% polyacrylamide gel, migration for 25 minutes at 100 V). In order to determine the Erysiphe species involved, two samples of PCR products were selected per tree, from 2011 and 2012, which were sequenced with the same oligonucleotide primers, by the Sanger method (dye terminators), on an Applied Biosystems^® 3730 sequencer (Life Technologies, Karlsruhe, Germany).

Powdery mildew spores were considered to be present in a given trap filter for a given exposure period if at least one of the four PCR tests (2 membranes × 2 dilutions) was positive for the exposed filter. For each positive filter, the date of trapping was defined as the date in the middle of the filter exposure period. In each tree canopy, the beginning of the sporulation period was defined as the date corresponding to the first positive filter. For several traps, we were unable to determine this date because some filters were absent or destroyed due to poor weather conditions.

Safety margins

The budburst date of a tree is not informative per se of the tree ability to escape the impact of biotic and abiotic hazards in its local environment, since this also depends on the dates of hazard occurrence. In order to obtain a tree-centred relevant variable, we calculated ‘safety margins’ at tree level in each population and year as the time separating the budburst date from the date of hazard occurrence (spore emission, spring frost). The safety margin quantifies the stress escape, with low or negative safety margins indicating a high probability of exposure to the biotic or abiotic stress concerned and positive safety margins indicating a high probability of escaping stress by the trees.

For abiotic hazard related to spring frost, safety margins (SM_sf) were calculated as follows:

SMsf=Dbb−Dlsf

where D_bb is the budburst date of a given tree (Julian days) and D_lsf is the date (in Julian days) of the last spring frost event of the tree site. Choosing a temperature threshold to define a frost event is not straightforward. A standard and conservative definition of a frost event by a daily minimum temperature below 0 °C was used in several studies (Alberto et al. 2011; Anandhi et al. 2013). However, when assessing risks of frost damage to plants, a more relevant threshold should refer to the temperature actually experienced by leaves and causing physiological damage. Slightly negative temperatures (around −3 °C) were defined as LT50 (lethal temperatures at which 50% of the samples died) for oak leaves during flushing under controlled conditions (Liepe 1993; Lenz et al. 2013; Vitasse et al. 2014b). However, in our study, in situ leaf temperature is variable in different parts of the crown and for seedlings compared to trees. In particular, crown temperature may be slightly colder than the temperature recorded by the on-site meteorological sensors put at 1.50 m above soil level and below a shelter (Kollas et al. 2014). We therefore calculated safety margins with two values: the conservative standard value of 0 °C and the −3 °C value. The same were used for all trees, whatever the site and the phenological ranking (early versus late flushing), in the absence of any data showing within-species genetic differentiation for frost hardiness during flushing in oaks. The spring frost safety margins were calculated for 13 populations along the Gave and Ossau gradients, from 2005 to 2012 (see Table 1 for details).

For biotic hazard related to powdery mildew, safety margins (SM_pm) were calculated as follows:

SMpm=Dsp−(Dlu+20)

where D_sp is the date on which powdery mildew sporulation began (Julian days), defined for each site and year as the date for which spores were trapped in at least 75% of the traps placed in adult trees (as determined by observation or interpolation between two exposure campaigns), and D_lu is the leaf unfolding date (Julian days) plus 20 days, to account for the period of susceptibility of young leaves to powdery mildew (Edwards & Ayres 1982 and personal observations). Powdery mildew safety margins were calculated for six populations along the Gave gradient, in 2006–2007 and 2011–2012 (see Table 1 for details and Desprez-Loustau et al. (2010) for additional data for powdery mildew sporulation dates in 2006–2007).

Finally, a safety window was estimated as the difference in days between the date of the beginning of powdery mildew sporulation (D_sp) and the date of the last spring frost event (D_lsf).

Assessment of powdery mildew infection

Powdery mildew infection was assessed at the end of July, in five populations in the Gave valley (2011 and 2012) and five populations in the Ossau valley (2012) (see Table 1). For the Gave valley populations, we monitored the same trees and seedlings as were used for spore trapping. For the Ossau valley populations, we monitored four early- and four late-flushing adult trees (based on 2005–2007 and 2009–2011 assessments). For each studied oak, we collected a single twig from each of the four cardinal points (north, south, east and west) in the canopy, at a height of about 10 m, with a pole pruner, when possible. Two leaves were randomly chosen from the first growth cycle on each shoot (second flushes were quite rare). The infection of each leaf was visually assessed, by trained observers, as the percentage of the leaf area covered by mycelium. A magnifying glass was used to facilitate this assessment if necessary. The infection percentage per tree was calculated as the mean infection level of the eight leaves. For the seedlings, infection was estimated as the percentage of the leaf area covered by mycelium, for the whole plant.

Statistical analyses

We used a general linear model to test the effects of elevation, phenological phenotype (early vs. late flushing) and their interaction on frost and powdery mildew safety margins (for both seedlings and adult trees) (PROC GLM, SAS). For adult trees, infection was also analysed, with a general linear model, after data log transformation, with phenological phenotype, elevation, valley x year interaction, valley x year x elevation interaction and phenological phenotype x elevation interaction as effects. We carried out Tukey's test to compare mean infection values between sites. For seedlings, infection was analysed as a binary variable (0: uninfected; 1: infected, due to the small number of seedlings infected, with low levels of damage), with a logistic model, with phenological phenotype, site x year interaction and site x year x phenological phenotype interaction as effects (PROC LOGISTIC, SAS). If interaction with phenological phenotype was found to be significant in infection and safety margin models, separate analyses were carried out by site. We investigated whether safety margin was a good predictor of damage (i.e. infection), by carrying out a general linear model analysis, by site, for both seedlings and adult trees. The distribution of residuals was analysed by Q-Q plots, the plotting of a distribution histogram, skewness, kurtosis and Shapiro–Wilk tests (PROC UNIVARIATE, SAS) and by comparisons of predicted vs. observed data plots, to check the validity of the models. The correlation between leaf unfolding and the start of powdery mildew sporulation dates within populations was analysed by calculating a Pearson's correlation coefficient from the centred-reduced values per tree (mean and standard deviation calculated for each site and year) (PROC CORR, SAS). For comparisons of sporulation start dates between seedlings and trees, we used a general linear model per year and per elevation. All statistical analyses were performed with sas 9.2 software (SAS Institute Inc. 2011).

Phenological variations within and between populations

The dates of budburst and leaf unfolding were strongly correlated (slope = 1.009, r² = 0.99, P < 0.0001), with a mean of 4 days between the two dates. The mean dates of leaf unfolding and sporulation at population level were significantly and positively correlated with elevation (Table 3; see Fig. 2, for example). However, the pattern of change in phenology with elevation differed considerably between oak and powdery mildew: oak had a much stronger response to elevation (leaf unfolding delayed by 3.3 days per 100 m increase in elevation) than powdery mildew (0.6 days.100 m⁻¹) (Table 3). The canopy duration was significantly negatively correlated with elevation (−4.5 days.100 m⁻¹) (Table 3).

In the study sites, leaf unfolding occurred 5 days later on average in seedlings than in adult trees. The variability of leaf unfolding phenology for seedlings was almost as high as that for trees (mean of 26 days between the seedlings unfolding their leaves earliest and last). For adult trees, the timing of leaf unfolding showed greater year-to-year fluctuations at high elevation than at low elevation (individual standard deviation = 9.2 and 5.8 on average, respectively; y = 0.87×^0.33, r² = 0.42, P < 0.0001) (see Fig. S2). This difference was not related to greater differences in winter/spring temperatures at high elevation than at low elevation (mean standard deviation = 1.0 ± 0.06, from the beginning of February to the end of May in 2004–2012).

Frost and powdery mildew exposure along the elevation gradient

Safety margins for both spring frost and powdery mildew decreased significantly with increasing elevation (Table 3) (Fig. 3). The same trend was observed for spring frost safety margins whatever the temperature threshold considered for a frost event. The altitudinal trend was slightly more pronounced for powdery mildew safety margins. In adult trees, budburst generally occurred well after the last spring frost event, whatever the elevation (see Fig. 2 as an example). The spring frost safety margin was positive for 72% of the tree-per-year observations with the 0 °C threshold and 98% with −3 °C during the study period. By contrast, the powdery mildew safety margin was mostly negative, in all populations (Fig. 3). The powdery mildew safety margin was positive for only 8% of the tree-per-year observations, these trees being located at low elevation. Safety margins varied considerably between trees within a population, for both powdery mildew and spring frost hazards (P < 0.0001 and P = 0.015 for phenology × elevation interaction, respectively) (Fig. 3).

The duration of the safety window decreased significantly with increasing elevation by around 40 days from 100 m to 1600 m asl (see Fig. S3).

Frost and powdery mildew exposure within populations

A large variation in the date of first spore detection was observed between the traps on different adult trees within sites in a single year, particularly at low elevations in 2012 (Fig. 4). However, powdery mildew phenology was not correlated with oak phenology within populations (Pearson correlation = 0.023, P = 0.85) (Fig. 4). Sporulation was detected earlier (about 1 month) in seedling traps than in adult tree traps at 150 m asl in 2012 (P = 0.0051), but no difference was observed in 2011 (P = 0.85) neither at 400 m asl in 2012 (P = 0.27). We therefore considered that differences between traps only represented within-site variation, and this justified that the same date of inoculum exposure (calculated from data over all traps in a site) could be used for all trees. Only E. alphitoides was detected in all the samples studied, on the basis of its ITS sequence.

Reflecting differences in flushing dates, the powdery mildew and spring frost safety margins differed by 11–15 days on average between early- and late-flushing individuals within sites, showing a decreasing trend with elevation (P < 0.0001) (Fig. 3). However, difference in safety margins between early- and late-flushing individuals was positive for powdery mildew and negative for spring frost. Only early-flushing trees at low elevation had a positive or weakly negative SM_pm, enabling them to avoid powdery mildew spores (Figs 2 and 3). Overall, the powdery mildew safety margin was positive for 15% of the tree-per-year observations for early-flushing trees vs. only 2% for late-flushing trees, always at low elevation. The spring frost safety margin was positive for 66% of the tree-per-year observations for early-flushing trees vs. 78% for late-flushing trees with a 0 °C threshold (98% for both with a −3 °C threshold).

The spring frost safety margin was positive for both early- and late-flushing seedlings (SM_sf = 27 and 49 days, on average, at 150 and 400 m, respectively, with 0 °C threshold), whereas the powdery mildew safety margin was negative (SM_pm = −9 days and −18 days on average, at 150 m and 400 m asl, respectively). As for adult trees, powdery mildew safety margins were higher for early-flushing seedlings than late-flushing seedlings (P < 0.0001), whereas the opposite pattern was observed for spring frosts (P < 0.0001) (see Fig. S4).

Powdery mildew infection

Infection was lower at 200 m asl than at higher elevations, with 3% instead of 13% of the leaf area infected on average (Fig. 5; significant difference between 200 m and higher elevations in Tukey's tests). A year × valley interaction effect on infection was found (P < 0.0001), but there was no effect of a valley × year × elevation interaction (P = 0.17). At low elevations, late-flushing trees were much more infected than early-flushing trees (P = 0.004 and P = 0.0001 at 200 and 400 m asl, respectively), whereas infection did not vary with oak phenology at medium and high elevations (P = 0.43, P = 0.88 and P = 0.22 at 800, 1200 and 1600 m asl, respectively) (Fig. 5). Infection was strongly related to safety margin at low elevations (P = 0.0001 and P = 0.0095 at 150 and 400 m asl, respectively), whereas no correlation was demonstrated at medium and high elevations (P = 0.70, P = 0.52 and P = 0.77 at 800, 1250 and 1650 m, respectively).

Infection levels were much lower on seedlings than on adult trees (1.3% and 11% of the leaf area infected, respectively). At both sites, the infection rate was much higher among late-flushing seedlings than among early-flushing seedlings (55% versus 17% at 150 m, 33% versus 17% at 400 m), but this difference was significant only at 150 m (P = 0.013 and P = 0.38 at 150 and 400 m, respectively) (see Fig. S5). No effect of site × year and site × year × phenology interactions on seedling infection was detected (P = 0.68 and P = 0.53, respectively). Seedling infection levels were correlated with safety margin at 150 m, but not at 400 m (P = 0.0039 and P = 0.30, respectively).

Variation in abiotic and biotic stress escape along the elevation gradient

The observed phenological patterns may be interpreted from an evolutionary perspective. The positive spring frost safety margins for most trees along the elevation gradients are consistent with an effect of past natural selection favouring phenotypes able to escape spring frost damage. Indeed, frost damage can incur a high fitness cost, especially on seedlings, and is thus expected to exert a strong selective pressure (Agrawal, Conner & Stinchcombe 2004; Alberto et al. 2011; Augspurger 2011; Pau et al. 2011). However, selection to escape late spring frosts would have been less effective at higher elevations, as expressed by smaller safety margins and observations of leaf frost damage in some years (2 out of the 9 years of monitoring, see Fig. S6). It has to be noted that an increased risk of leaf frost damage at high elevation is not always observed. Lenz et al. (2013) showed a constant safety margin against damage by spring freeze events in tree populations along elevational gradients. Constraints affecting the timing of flushing are likely to vary along geographic gradients and to produce different evolutionary outputs (Pau et al. 2011). At high latitude or elevation, fitness is more strongly correlated with the ability of fruit to ripen fully rather than with the probability of survival (Morin, Augspurger & Chuine 2007; Morin, Viner & Chuine 2008). This favours early flowering and, thus, early budburst (Franjic et al. 2011; S. Delzon personal observations), possibly at the expense of spring frost damage. The oak populations at 1600 m were almost at the tree line for this species (Aubert 2001; Belet 2001; Vitasse et al. 2012), that is the edge of the habitat in which these trees can thrive. The growing season was found to be short at such elevations. The presence of strong constraints on the timing of flushing at high elevation is consistent with the low variation in early spring phenology between trees in a given year (expressing genetic variation). However, trees displayed a high degree of plasticity, as demonstrated by the high year-to-year fluctuations in the timing of flushing (Fig. S2). This is consistent with the predictions of Pau et al. (2011) that trees at high elevations, which put out leaves in harsh physical conditions, are likely to be more sensitive to climate (temperature) cues.

Unlike late-frost safety margins, safety margins for powdery mildew were mostly negative, indicating an absence of escape. It can be hypothesized that selective pressure by powdery mildew on oak phenology has been lower than by frost. Fitness costs associated with foliar pathogens may be lower than with frost damage because complete leaf loss is less likely. Only high powdery mildew infection rates observed in dense oak regenerations under favourable environment and in mature trees following insect or frost defoliation can lead to mortality (Marçais & Desprez-Loustau 2014). In addition, fitness costs of phenological maladaptation are higher for the pathogen (which needs to find susceptible hosts to complete its life cycle) than for the host, and generation times are shorter for the fungus. It is thus expected that the pathogen will always achieve a certain degree of phenological synchrony with its host, giving no chance to oaks to flush at a time when they would be safe from powdery mildew infection. The same reasoning can apply to other biotic antagonists (herbivore insects, other foliar pathogens), which exert their effects after flushing has occurred. Although safety margins for powdery mildew were increasingly negative with increasing elevation, infection did not increase above 400 m asl. Both tree characteristics (lower foliage densities) and environmental conditions (shorter growing season, lower temperatures, longer leaf-wetness duration) less favourable to powdery mildew development (Jarvis, Gubler & Grove 2002) could limit disease severity at higher elevations even with good matching of oak and fungus phenologies.

Variation of hazard escape within populations

At low elevations, phenological differences between trees resulted in contrasting exposures to biotic and abiotic hazards. Although both early- and late-flushing trees showed mostly positive frost safety margins during the study period, it cannot be excluded that in rare late-frost events, late-flushing trees have an advantage over early-flushing trees due to their larger safety margin. Strong selection caused by rare catastrophic events has been suggested to play an important role in the evolution of natural populations (Agrawal, Conner & Stinchcombe 2004). On the other hand, only early-flushing trees escaped powdery mildew exposure. Our observations suggest that the ‘spore cloud’ reached all the trees at a given site at about the same time. No spatial effects could be detected on the first spore trappings within a site (spatial analysis, performed with arcgis 10.2 software; data not shown). Moreover, the timing of spore trapping was not related to host tree phenology. This suggests that no local phenological adaptation of powdery mildew populations occurred in response to variation within oak populations, by contrast to reports of the local adaptation of powdery mildew strains, at tree level, in terms of their ability to infect trees (Roslin, Laine & Gripenberg 2007; Desprez-Loustau et al. 2011). A lack of local adaptation of powdery mildew to achieve phenological synchrony with its host has already been reported over a larger spatial scale (Desprez-Loustau et al. 2010). This contrasts with findings for other antagonistic interactions with oaks (e.g. for caterpillars), for which adaptive deme formations have been observed at very fine spatial scales (van Asch & Visser 2007). Early- and late-flushing trees were thus exposed to different powdery mildew pressures (spore concentrations) during the flushing period, in populations growing at low elevation (< 500 m). This resulted in higher levels of infection in late-flushing trees than in early-flushing trees. At low elevation, escape of biotic (powdery mildew) and abiotic (spring frost) hazards therefore corresponds to opposite phenological phenotypes. Such opposite selective forces may have contributed to the maintenance of local variability for the ‘time of flushing’ trait at low elevation (Alberto et al. 2011). The large difference between the last spring frost date and the sporulation date (i.e. the safety window) at low elevation makes it possible for oak populations to maintain this variability, with early-flushing trees escaping powdery mildew disease (and possibly damage due to other defoliators) and having a longer growing season, and late-flushing trees being more likely to survive late spring frost events. Seedlings in the low elevations sites displayed the same trends as adult trees, with early-flushing seedlings being at an advantage with respect to powdery mildew and late-flushing seedlings being at an advantage with respect to spring frosts. This was important to check because selective forces on tree species are strongest during the juvenile stage.

Escape versus other strategies of defence against biotic and abiotic hazards

Our study focused on phenological variations as related to escape of abiotic and biotic hazards. Escape, that is processes and mechanisms preventing encounter with the deleterious agent, is the first line of defence (Combes 2001), in particular temporal escape for sessile organisms such as trees (Vitasse, Lenz & Körner 2014). However, organisms can use a second line of defence composed of two types of mechanisms, those which prevent or limit the damage caused by the deleterious agent and those which reduce the negative consequences of damage for fitness. The former are referred to avoidance or resistance according to authors and research fields, while tolerance corresponds to the latter (see e.g. Burdon 1987; Agrawal, Conner & Stinchcombe 2004; Baucom & de Roode 2011; Vitasse, Lenz & Körner 2014, for terminology issues and definitions).

During winter, leaf tissues enclosed within buds remain frost hardy, but their vulnerability to freezing temperatures gradually increases prior to and just after budburst (Kalberer, Wisniewski & Arora 2006; Lenz et al. 2013). Young leaves are particularly vulnerable to frost and many pathogens and herbivores, due to their anatomical, physiological and biochemical properties (Coleman 1986). The escape of spring frost by late flushing may be a major adaptive process along the elevation gradient (Gömöry & Paule 2011). Trees have optimized the timing of budburst in relation to the probability of frost in spring, by developing complex mechanisms of chilling/thermal time requirement and photoperiod (Cannell 1997; Basler & Körner 2012; Laube et al. 2013). By delaying flushing, trees maintain their winter frost hardiness for as long as possible, relative to adjustments of growing season length. A frost escape strategy may prevail in Q. petraea since this species was shown to be among the less freezing-tolerant species during bud burst and leaf unfolding (Lenz et al. 2013). However, a possible genetic differentiation for frost avoidance or tolerance between populations of Q. petraea along elevational gradients would deserve investigation (Vitasse, Lenz & Körner 2014).

Disease escape as a defence strategy in plants (trees) has been little documented (but see Ghelardini & Santini 2009), especially compared to the large number of studies related to insect herbivory (Aide 1992; Tikkanen & Julkunen-Tiitto 2003; Lamarre et al. 2014). Conversely, resistance and tolerance mechanisms are thought to play a critical role in the adaptation of plants to pathogens (Bell 1981; Burdon 1987; Combes 2001) and have been well documented for powdery mildew diseases (Glawe 2008; Desprez-Loustau et al. 2014). In particular, plants can rely on induced resistance (Verhagen, Van Loon & Pieterse 2006; Kempel et al. 2011), which is expressed only in response to the biotic stressor and is thus a valuable strategy in terms of costs, against non-permanent stress factors such as diseases or pests (Agrawal 1999; Agrawal & Karban 1999). Thus, at low elevation where environmental conditions are favourable to the fungus, oak defence against powdery mildew may be based on both phenological escape and resistance. At higher elevations, oak resistance to powdery mildew would be sufficient to ensure that infection rates remain moderate, due to less favourable environmental conditions for pathogen populations.

Implications in a context of climate change

Local patterns of adaptation resulting from past selection processes may not be suitable in the face of rapid changes in environmental conditions. Climate change is causing many trees to flush earlier in spring, exposing them to new hazards. Several studies have found that this major advance in the date of leaf unfolding may increase the frequency of spring frost damage (Kramer 1995; Hänninen 2006; Schwartz, Ahas & Aasa 2006; in Northern and Eastern Europe). However, some studies have indicated that spring frost risk is declining with climate change, because the date of the last frost is advancing more rapidly than flushing dates (Schwartz, Ahas & Aasa 2006; in Western Europe; Anandhi et al. 2013; Ge, Wang & Dai 2013). This pattern may also be accentuated if the timing of flushing does not advance because chilling requirements remain unmet (Fu et al. 2012; Laube et al. 2013; Dantec et al. 2014). Climate change will also alter the phenological relationships between interacting species, because responses to climate change may differ between species (Both et al. 2009; Singer & Parmesan 2010), as shown here for powdery mildew and oak. Given the complexity of climatic effects on the exposure of trees to biotic and abiotic hazards, process-based models of tree and fungus phenology (including chilling and thermal requirements) will be required to explore outputs in the future, using local climate scenarios.