Author for correspondence: Markus Löw Tel: +61 3 5321 4309 Email: email@example.com
•Increasing atmospheric concentrations of phytotoxic ozone (O3) can constrain growth and carbon sink strength of forest trees, potentially exacerbating global radiative forcing. Despite progress in the conceptual understanding of the impact of O3 on plants, it is still difficult to detect response patterns at the leaf level.
•Here, we employed principal component analysis (PCA) to analyse a database containing physiological leaf-level parameters of 60-yr-old Fagus sylvatica (European beech) trees. Data were collected over two climatically contrasting years under ambient and twice-ambient O3 regimes in a free-air forest environment.
•The first principal component (PC1) of the PCA was consistently responsive to O3 and crown position within the trees over both years. Only a few of the original parameters showed an O3 effect. PC1 was related to parameters indicative of oxidative stress signalling and changes in carbohydrate metabolism. PC1 correlated with cumulative O3 uptake over preceding days.
•PC1 represents an O3-responsive multivariate pattern detectable in the absence of consistently measurable O3 effects on individual leaf-level parameters. An underlying effect of O3 on physiological processes is indicated, providing experimental confirmation of theoretical O3 response patterns suggested previously.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
Direct assessment of O3 impact on growth and carbon sink strength of large, adult forest trees is difficult and involves long-term observations (Manning, 2003). For example, an effect of twice-ambient O3 concentrations (2 × O3) on the growth of adult trees was only detectable after eight growing seasons, and even then only on one of the two species involved (on Fagus sylvatica, Pretzsch et al., 2010). It is therefore important to identify stages of O3 impact, that is, effects at the molecular and biochemical levels that precede potential visible foliar symptoms and growth decline (Wild & Schmitt, 1995). Despite a large number of papers (for reviews see, e.g., Matyssek & Sandermann, 2003; Wittig et al., 2009) and some good theoretical understanding of aspects of the mode of action (Heath, 2008; Matyssek et al., 2008), identification of O3-specific effects on leaves of adult forest trees is surprisingly elusive under field conditions. A number of studies have summarized that many tree physiology-related parameters are affected by O3, and theoretical frameworks to interpret such multiple and often transient responses have been proposed (Karnosky et al., 2003; Nunn et al., 2005; Matyssek et al., 2007a; Wittig et al., 2009; Leisner & Ainsworth, 2011). Empirical and statistical evidence in support of such theoretical frameworks is often poor, because physiological and biochemical responses to O3 can apparently be inconsistent or conflicting (Bortier et al., 2000; Nunn et al., 2005; cf. Matyssek et al., 2007a). These studies provide an insight into the O3 response of individual parameters, but do not provide a comprehensive simultaneous multivariate analysis of the overall leaf-level response to O3.
Most reviews on the topic postulate consistent metabolic patterns characterizing the response to O3, linking responses within and across organizational levels from molecules to the whole plant (Matyssek & Sandermann, 2003; Sandermann & Matyssek, 2004; Heath, 2008; Matyssek et al., 2008; Heath et al., 2009). Such patterns would be detectable as concerted changes in single biochemical and physiological parameters. Despite the high variability in the response to O3 of most of the biochemical and physiological parameters in question (Nunn et al., 2005), underlying patterns composed from several measured parameters may become accessible using multivariate explorative statistical methods. One study by Nali et al. (2005) utilized a combination of multivariate variance analysis and canonical discriminant analysis characterizing the overall response of clover to O3, with the aim of analysing the biochemical and physiological bases of O3 tolerance and identifying potential biomarkers for O3. Kontunen-Soppela et al. (2007) used principal component analysis (PCA) on metabolomic profiles of two clones of O3-exposed birch to identify genotypic and O3 effects on leaf metabolites such as phenolics, and lipophilic and polar compounds. PCA, a common multivariate method, was successfully employed more generally to identify patterns in physiological datasets (García-Plazaola et al., 2000; Tausz et al., 2001; Wright et al., 2004; Warren et al., 2005). Such an approach allows one to test whether one or more response patterns identified by PCA are related to O3 exposure (Kontunen-Soppela et al., 2007).
In the present study, we exploited an extensive database of leaf-level physiological and biochemical parameters of c. 60-yr-old Fagus sylvatica (European beech) trees exposed to experimental 2 × O3 under open field conditions (Matyssek et al., 2007a). The database integrates data collected during the years 2003 and 2004. The parameters in the database cover multiple aspects of leaf-level physiology and biological pathways (including photosynthesis, C metabolism, phytohormones, antioxidants, pigments, and stomatal O3 uptake) and were selected to represent biochemical and physiological leaf functions that are known to be affected by O3 (Matyssek & Sandermann, 2003). Owing to the inclusion of the extraordinary European drought season of 2003 (Ciais et al., 2005), two climatically contrasting growing seasons were captured in the database. This allows the analysis of O3 impact on tree functioning in interaction with drought conditions. A range of significant leaf-level responses to O3 treatment have been reported previously for F. sylvatica, but they were highly variable in extent and time (Nunn et al., 2005; Matyssek et al., 2007a, 2010b). As in many other studies, few of the selected leaf parameters available in the database were consistently responsive to O3 stress. Nevertheless, after 8 yr of 2 × O3 exposure, growth of F. sylvatica was significantly impaired (Pretzsch et al., 2010).
We applied PCA to explore higher-order variables derived from the original database with the aim of testing the hypothesis that one or several of these extracted variables (principal components, PCs) represent multivariate metabolic processes that are responsive to O3, that is, they respond significantly and with consistent patterns to 2 × O3. Our second aim was to examine further whether a relationship exists between such PCs (if any) and the widely used O3 exposure index AOT40 (accumulated O3 exposure over a threshold of 40 nl O3 l−1) and/or the physiologically relevant O3 dose (i.e. the cumulative stomatal O3 uptake, COU), which has been identified as the preferable metric for correlating O3 impact with plant responses (Matyssek et al., 2007b, 2008).
Materials and Methods
Field experiment and database
The data analysed in this study were collected from Fagus sylvatica L. trees of the ‘Kranzberger Forst’ free-air O3 exposure facility. In this mixed beech and spruce stand in southern Germany (48°25′N, 11°39′ E, 485 m asl), five large (c. 60 yr old and 28 m tall) F. sylvatica trees were exposed to twice-ambient O3 (2 × O3; capped at 150 nl O3 l−1 to prevent acute O3 injury) using a free-air canopy exposure system (Werner & Fabian, 2002; Matyssek et al., 2007a). Five similar trees served as ambient control (1 × O3). The experimental O3 exposure was operated from 2000 to 2007. O3-induced growth reduction of F. sylvatica was found when growth data from the 8 yr of the study were analysed (Pretzsch et al., 2010). No such reductions were found in earlier studies, for example, when growth data were analysed after 6 yr of exposure (albeit with a different approach; Wipfler et al., 2009).
In this paper, we exploited a large database of leaf-level parameters established by the CASIROZ project (‘The carbon sink strength of beech in a changing environment: Experimental risk assessment by mitigation of chronic ozone impact’; details in Matyssek et al., 2007a, 2010b), resulting from samples of F. sylvatica leaves during the growing seasons 2003 and 2004 at ‘Kranzberger Forst’, that is year 4 and 5 of the experimental O3 exposure period. Simultaneous coordinated sampling campaigns for parameters used in this study were conducted five times each year (in May, June, July, September, and October), and a total of 82 parameters were assessed on leaves or leaf samples collected from the same standardized positions of selected sample branches in the sun and the shade crown of each tree. This resulted in a multivariate metabolic ‘snapshot’ of leaf samples for each tree (n =5 trees per O3 treatment), crown position (sun or shade), month (five sampling campaigns yr–1), and year (2003 or 2004), that is a total of up to 200 potential samples per variable. After quality control, data were transferred into a database (Access, several versions used, Microsoft, Unterschleißheim, Germany).
Table 1 lists O3 indices for the growing seasons covered by the database (2003 and 2004) and Fig. 1 shows climatic data for all study years. In general, O3 regimes and weather data in 2004 were close to the long-term meteorological averages, whereas 2003 was an extraordinarily dry and hot year (Ciais et al., 2005) with high external O3 exposure (SUM0 and AOT40, Table 1), although stomatal O3 uptake was not high (COU, Table 1; Löw et al., 2006).
Table 1. Growing season O3 indices in the ‘Kranzberger Forst’ for the years 2003 and 2004
SUM0, the integral of mean hourly O3 concentrations; AOT40, the accumulated O3 exposure over a threshold of 40 nl O3 l−1; COU, the accumulated stomatal O3 flux based on leaf area. O3 indices are calculated for the time between leaf expansion in spring and leaf fall in autumn. Sun crown data after Löw et al. (2006).
μl O3 l−1 h
1 × O3
μl O3 l−1 h
2 × O3
μl O3 l−1 h
1 × O3
μl O3 l−1 h
2 × O3
mmol O3 m−2
1 × O3
mmol O3 m−2
2 × O3
μl O3 l−1 h
1 × O3
μl O3 l−1 h
2 × O3
μl O3 l−1 h
1 × O3
μl O3 l−1 h
2 × O3
mmol O3 m−2
1 × O3
mmol O3 m−2
2 × O3
Cumulative O3 uptake
The daily COU was calculated with a multiplicative stomatal conductance O3 uptake algorithm, as part of the ANAFORE forest model (Deckmyn et al., 2007; Op de Beeck et al., 2007). This process-based model was parameterized for each tree and crown position individually and the performance of the gas exchange calculation of the model was validated using long-term cuvette measurements. The measured meteorological data, soil moisture, and O3 concentrations from the research site were used to drive the model for the years 2003 and 2004. Tree-specific budbreak and leaf-fall data for both years were used to determine the length of growing season for each tree and crown position individually. COU was calculated on a half-hourly basis for the growing season (from approx. mid-May to mid-October of each year). COU and AOT40 integrated over the week before each sampling date were then used in the analysis (Table 1).
Analyses were performed using the statistical software R (R version 2.13.1 pc-linux-gnu 32 bit; R Development Core Team, 2011). A total of 82 leaf-related parameters were in the database, with n =196 leaf samples for each parameter (out of a potential 200, because four missing samples precluded gap-filling by imputations; see next paragraph). Parameters reported on leaf area basis were converted to DW basis using specific leaf area (SLA) at each date, tree, and crown position. To ensure normal distributions, the Shapiro–Wilk test (shapiro.test from R package ‘stats’) was performed and data were transformed accordingly via Box-Cox transformation (bcPower/yjPower from R package ‘car’). The ‘lambda’ setting of the Box-Cox algorithm was automatically optimized for each parameter individually to achieve normally distributed data.
Missing data were gap-filled via multiple imputation by chained equations (using the ‘mice’ package in R with m = 25 imputations; Van Buuren & Groothuis-Oudshoorn, 2011), assuming ‘data is missing at random’ (at each sampling date, O3 treatment, and crown position) and multivariate normality. As accuracy and reliability of gap-filling via the multiple imputation process benefits from a high number of parameters (Enders, 2010), all available leaf-level data (82 parameters) were used up to this step.
Input parameters for the subsequent PCA (prcomp from R package ‘stats’), were selected from the imputed data according to the following criteria:
•The input parameters should represent broad functional leaf-level categories putatively affected by O3 (C metabolism, isotope ratios, antioxidants/defence compounds/pigments, phytohormones, enzyme activities, and photosynthesis; Matyssek et al., 2007a). Weighting towards any one category was avoided by choosing similar numbers of parameters for each category.
•Logically or mathematically strongly autocorrelated parameters were omitted following correlation analysis (Pearson’s product moment correlation coefficient, cor.test from R package ‘stats’).
•Only parameters with the fewest missing values for each category were selected for analysis to avoid bias as a result of gap-filling.
In total, 27 input parameters were used in the PCA run on n =196 leaf samples (Table 2).
Table 2. Loadings of the selected originally measured parameters in the first five principal components (PC1–5) of the PCA
Only loadings ≥ |0.19| (significant at P <0.005) are shown. All parameters have been measured on the leaf level and are expressed on a leaf DW basis, except for dimensionless ratios, where indicated. References indicate original studies. For loading plots of PCs 1–3, see Fig. S1.
Principal components were accepted based on visual examination of the Scree plot and in line with the Kaiser criterion (all eigenvalues > 1). PC scores and loadings were determined after Varimax axis rotation (Johnson & Wichern, 1992). Five PCs were accepted, explaining a cumulative 60% of the total data variance.
Principal component scores were then used to examine the effects of the experimental factors in a mixed-model ANOVA with ‘tree’ as a random factor (subject), months (May, June, July, September, October within each year, i.e. seasonal effect), year (dry year 2003/year with average moisture 2004) and crown position (shade crown/sun crown) as within-subject factors and O3 treatment (1 × O3/2 × O3) was the between-subject factor (aov from R package ‘stats’). Homogeneity of variances was ensured via Levene’s test before the ANOVA (leveneTest from R package ‘car’). Because the experimental increase in O3 concentration imposed over a broad range of environmental conditions (as represented by factors sampling date and year) was used as one factor in the ANOVA, results for the factor O3 are not biased by potential correlations between ambient O3 and environmental conditions. Validity and stability of the PCA results were tested by removing or adding physiological parameters to/from the selected parameters. The number of parameters that represent each leaf-internal process or category was kept constant. Additionally, gap-filling and ANOVA analysis were repeated 25 times. The resulting PC scores, loadings, and ANOVA outcomes from different configurations were compared manually. These comparisons did not focus on the numeric values for scores, loadings and P levels, but on the resulting statistically significant patterns. The PC loading patterns and ANOVA results reported in this study were similar for at least 95% of all tested configurations.
After the ANOVA indicated a significant effect of experimental O3 exposure on one or more resulting PCs, this relationship was further explored with a general linear model (GLM) of PC scores vs crown position and year with selected O3 uptake (COU) or external exposure (AOT40) indices as covariates. Relationships were considered significant at P <0.001. Because periods between sampling dates were 1–2 months, both O3 indices were integrated over a period as representative as possible (in terms of O3 exposure and uptake) of each sampling date. Statistical considerations (goodness-of-fit, relative importance of each regressor within the R2 based on the sum of squares of the GLM, R package ‘relaimpo’, lmg-method, Groemping, 2006) gave no clear preference to any integration period from 1 d up to 4 wk before sampling (data not shown). We chose an integration period of 1 wk before each sampling date.
The first five PCs together accounted for 57% of the variance in the original data set (Table 2). Only PC1 (accounting for 19% of the variance) was statistically significantly affected by O3 (Table 3). Parameter loadings indicate the individual relationship of an input parameter to the newly identified components. They can be read like correlation coefficients, with the sign indicating the direction of the relationship, and values closer to |1| indicating a stronger contribution to the PC in question. Loadings with an absolute value ≥ |0.19| (significant relationship with the PC at P <0.005) were considered important (Table 2, Supporting Information, Fig. S1). Positive loadings on PC1 were found for the parameters cellulose, Chla + b, and the redox state of glutathione (GSSG, the oxidized proportion of total glutathione), an oxidizing scavenger protecting against oxidative stress. The de-epoxidation state of the xanthophyll cycle (DEEPS), indicating the activation threshold of light protection in the leaf, as well as total ascorbate, a direct scavenger against O3 and reactive oxygen species (ROS) that is part of the ascorbate-glutathione cycle, showed a negative loading on PC1. Additionally, δ13C, starch and sucrose, along with the carboxylation enzymes phosphoenolpyruvate carboxylase (PEPc) and ribulose-1,5-diphosphate carboxylase/oxygenase (Rubisco) contributed negative loadings to PC1 (Table 2). Out of 12 original parameters with significant loadings on PC1 (Table 2), only two were affected by O3 if tested individually, all were affected by crown position, nine were affected by month, and four were affected by year (Table S2).
Table 3. Mixed-model ANOVA on principal component (PC) scores with ‘tree’ as a random factor (subject), month, year and crown level as within-subject factors and O3 as the between-subject factor (n =196 samples)
Detailed ANOVA results are given in Table S1.
Significance: ***, P <0.001; **, P <0.01; *, P <0.05. Four-way interactions were not significant at P <0.05.
Symbols indicate the general direction of significant effects for the individual experimental parameters (not shown for interactions):
↑ in O3 treatment row indicates PC scores are higher in 2 × O3 than in 1 × O3.
↓ in crown position row indicates PC scores are lower in the sun crown than in the shade crown.
↑ in year row indicates PC is higher in the year 2004 than in the dry year 2003, ↓ indicates scores are lower in 2004.
∼ in the month row indicates variable scores for this PC during the growing season; \ indicates decreasing PC scores over the season; U indicates scores are lowest in the mid-season; ∩ indicates scores are highest in the mid-season.
For corresponding mean values of PC1 scores, see Fig. 2(a). For corresponding ANOVAs on all single (input) parameters see Table S1.
O3 treatment × crown position
O3 treatment × month
Crown position × month
O3 treatment × year
Crown position × year
Month × year
O3 treatment × month × year
Crown position × month × year
Principal component 2 had strong positive loadings from α-tocopherol, free 1-aminocyclopropane-1-carboxylic acid (ACC) and PEPc activity. Δ18O, photosynthesis parameters (Asat, the light-saturated rate of CO2 uptake), light-saturated photosystem II electron transfer rate (ETR), and stomatal conductance (gs) were loaded negatively on PC2 (Table 2). From eight original parameters with significant loadings on PC2, one showed an O3 effect, six showed an effect of crown position, eight indicated a month effect, and seven parameters showed a year effect when tested individually.
δ13C and GSSG contributed positively to PC3, whereas cellulose concentration, glucose concentration, concentration of total glutathione (GSH) and concentration of lignin displayed negative loadings. In total, PC3 was composed of eight original parameters, two of which showed an O3 effect, seven a crown position effect, all of them a month effect, and five a year effect when tested individually.
Principal component 4 was composed of carbohydrate contributions (cellulose, fructose, and glucose), and total ascorbate with positive loadings, while abscisic acid (ABA), Chla + b content, photosynthetic efficiency (Photeff), Asat, and ETR contributed negatively. The individual effects of the experimental factors on the original parameters were as follows: one out of 11 was affected by O3, seven were affected by crown position, all were affected by month, and eight were affected by year.
The main parameters that positively contributed to PC5 were ABA, concentration of the hormone free salicylic acid (SA, a stress-signalling compound), Photeff, sucrose and ribulose-1,5-bisphosphate turnover limited rate of photosynthesis (Jmax). Antagonists in PC5 were cellulose, total active aromatic cytokinin content (Taacc), Asat, ETR, and gs. Of the 12 original parameters contributing to PC5, one was affected by O3, seven were affected by crown position, 10 were affected by month, and eight were affected by year.
PC scores related to environmental factors
ANOVA results showed that PC1, but none of the other PCs, was significantly affected by the experimental O3 exposure (P <0.05, Table 2, S1). PC1 scores were consistently higher in 2 × O3 than in 1 × O3 (Fig. 2a; Table 3). Irrespective of O3, PC1 scores were also affected by year (higher scores in the drought year 2003 than in the average year 2004, Table 3, Fig. 2a), month, crown position (higher scores in the shade than in the sun crown), and by an interaction between month and crown position (higher in shade than in sun leaves, with pronounced mid-seasonal maximum in the shade; Table 3, Fig. 2a), and crown position with year.
Scores for PC2 were significantly affected by month, crown position and year (all P <0.001; Table 3), and interactions of crown position × month and month × year. PC2 scores were decreased in the sun crown, and lowered during the dry year 2003, but displayed a general trend towards negative loading as the growing season progressed (Table 3), a trend that is modified between the two years (significant interaction month × year, Table 3).
Principal component 3 scores were significantly lower in the sun crown than in the shade crown, and higher during the dry 2003 than during the humid 2004 season (Table 3). A significant trend towards lower PC3 scores was detectable with progressing growing season (month), being more pronounced during the dry year 2003 (see significant month × year and O3 treatment × year interactions, Table 3).
Principal component 4 was affected by seasonality (significant month effect, Table 3), crown position and year, with crown position × month, O3 treatment × year and month × year interactions. PC4 scores indicated a minimum during the mid-growing season (Table 3).
Similarly, PC5 scores were affected by month but peaked during mid-summer (as opposed to PC4, Table 3). PC5 was also affected by year, did not differ between crown positions, but showed crown position × year, month × year and crown position × month × year interactions (Table 3).
Relationship of PC1 scores to cumulative O3 uptake and external O3 exposure
Principal component 1 was significantly related to both COU and AOT40 integrated over the week before each sampling date (Fig. 2b,c; P <0.001 for each of the effects of year, crown position, and for either AOT40 or COU as covariates in the GLM). However, the relative contribution of the O3 indices to the R2 of the statistical model was different between AOT40 and COU. AOT40 contributed 3% to the R2 of the GLM (the factor crown position explained 84% of R2, while the factor year explained 13%). In comparison, COU had a higher relative importance, with 13% (crown position explained 74% of R2, year 13%).
The main aim of this study was to identify associations between changes in leaf-level physiological processes of adult F. sylvatica upon exposure to O3 under field conditions. O3 is known to affect many leaf-internal processes (Matyssek & Sandermann, 2003; Matyssek et al., 2007a; Heath, 2008; Heath et al., 2009). Because each effect on individual parameters can be minor, variable, or only transient (Nunn et al., 2005), it is difficult to assess effects of O3 on trees before growth impairment can be measured. This is especially relevant for adult trees where O3 impact might be masked by large C storage available to repair and defence processes (Matyssek et al., 2008).
We employed PCA on a large multivariate physiological data set with the aim of identifying underlying response patterns related to O3. We extracted five PCs explaining c. 60% of the original variance in the dataset. In some other studies, fewer PCs were sufficient to explain a similar proportion of data variance in physiological or morphological leaf parameters (Tausz et al., 2001; Wright et al., 2004). In contrast to those studies, our data covered seasonal courses, strong interannual climate variations, sun and shade crown positions, as well as the experimental O3 regimes, which can all be expected to affect any response variable or metabolic pattern (Karnosky et al., 2005; Matyssek et al., 2010b). Furthermore, we used more input variables, covering a wider range of physiological leaf functions than the cited studies. Such studies, using a greater number of input parameters, commonly report smaller proportions of variance explained by relevant PCs (e.g. Kontunen-Soppela et al., 2007). Therefore PCs extracted in our analyses, for example, PC1 covering 19% of the overall variation (Table 2), are meaningful representations of underlying response patterns.
Only PC1 was significantly affected by the experimental O3 regime and appears suitable for identifying physiological leaf-level response patterns to O3 (Tables 2, 3). As only five of the originally measured 27 parameters were affected by O3, and only two of these five are represented in PC1, this indicates an otherwise unrecognized multivariate pattern.
In our ANOVA (Table 3), the responses of the PCs to the experimentally imposed increase in O3 over ambient O3 were tested under simultaneous consideration of sampling date and sampling year, factors that incorporate variability in environmental factors such as temperature or irradiance, which are often correlated to ambient O3. As this study focused on O3 effects, we will henceforth only discuss PC1, as the only PC significantly affected by O3, in detail. Scores of PC1 were significantly higher under 2 × O3 than under 1 × O3 (Table 3, Fig. 2), even under extraordinarily dry conditions, when most O3 effects on physiological parameters appeared to be overruled by drought (Löw et al., 2006). The physiological interpretation of PC1 can be derived from the loadings contributed by the original variables (Table 2). PC1 is mainly indicative of changes in defence mechanisms and carbon metabolism. High scores, as observed under 2 × O3, are related to high cellulose concentration but low sucrose (Blumenröther et al., 2007) and glucose concentrations – that is, structural rather than soluble carbohydrates. High scores of PC1 are also related to low activities of the carboxylating enzymes Rubisco and PEPc, consistent with negative O3 effects on carbon fixation reactions (Dizengremel et al., 1994), even though the instantaneous measurement of carbon fixation, Asat or Jmax, was not relevant for this PC (Table 2). Enzyme activity and external N sources and concentration can also be related to N metabolism, as indicated by δ15N (Högberg, 1997; Kolb & Evans, 2003). According to Haberer et al. (2007a), δ15N was decreased in leaves under 2 × O3, a response reflected in PC1 as well.
High scores of PC1 were furthermore associated with more negative δ13C (cf. Gessler et al., 2009). Being a measure of C isotope discrimination of assimilated C, δ13C is directly dependent on ci/ca ([CO2] in the sub-stomatal cavity ci / [CO2] in ambient air ca), with high ci related to more negative δ13C (Cernusak et al., 2005). Declining δ13C (i.e. towards ‘more negative’ concentrations) may be caused by a decrease in photosynthetic carboxylation, a reduced proportion of carbon fixed via PEPc, or high stomatal conductance (reducing resistance to CO2 influx), or any combination of these factors (Cernusak et al., 2005). In adult beech trees, stomatal narrowing (decrease in gs) rather than opening was observed under 2 × O3 (Löw et al., 2006). In addition, stomatal opening would lead to greater transpiration, which may be reflected by Δ18O signatures (Adams & Grierson, 2001) – an effect apparently not related to PC1 (Table 2). Therefore, the association of δ13C with PC1 seems to reflect changes in carboxylation, also indicated by enzyme activities (Table 2), rather than increased transpiration.
Some parameters related to the photosynthetic light reactions, defence and the antioxidative systems (DEEPS, GSSG, ascorbate, and Chla + b) also showed considerable loadings on PC1, indicating some oxidative stress signalling upon O3 exposure (an increase in GSSG has been ascribed to initial stress response; Tausz et al., 2004), maybe related to a weakening of antioxidative defence (indicated by negative loadings of ascorbate). Oxidative stress, however, did not seem to originate within the chloroplast (‘photo-oxidative stress’). Under photo-oxidative stress, DEEPS, as an indicator of protective thermal energy dissipation, is expected to increase (Demmig-Adams & Adams, 2006), and Chl content should decline. By contrast, in our study, high scores on PC1 are related to a decrease in DEEPS and an increase in Chl; hence the pattern reflected by PC1 appears consistent with oxidative stress initiated by O3 in the apoplast or at the plasmalemma (Matyssek & Sandermann, 2003).
High PC1 scores during the dry 2003 season (Fig. 2) may reflect stress from extraordinary drought on the leaf, eliciting responses similar to O3 impact. Some previous studies suggested that O3 effects were less severe under drought conditions (see summaries by Matyssek et al., 2007a, 2010b). However, exacerbation of stress under high O3 impact and drought as a result of synergetic effects has also been suggested, because both stress factors impose pressure on defence systems, thus perhaps shifting the ‘effective O3 dose’ towards increased responsiveness (Löw et al., 2006; Matyssek et al., 2006, 2008; Musselman et al., 2006; Tausz et al., 2007). The dependence of PC1 scores on month are in agreement with the seasonal variability of many leaf-level parameters (e.g. Löw et al., 2006; Table S2, cf. references in Table 2). PC1 scores were also greater in shade crown leaves, suggesting potentially lowered protection against stress (as suggested for drought stress effects, e.g., by Valladares & Pearcy, 2002), maybe because shade leaves have fewer defence compounds per unit COU (Wieser et al., 2002; Wieser & Matyssek, 2007). Inherently weaker defence in shade than in sun leaves coincides with an apparently stronger response of PC1 in the shade relative to the sun crown. Although the interaction between O3 and crown position was not statistically significant, there is a trend towards differences in slopes between sun and shade crown when PC1 scores are plotted against quantitative exposure (AOT40) and uptake (COU) data (Fig. 2). Low COU in the shade crown (as compared with the sun crown) may already trigger O3-driven responses, perhaps because of light limitation on photosynthate supply for repair and detoxification (Fig. 2).
The assessment of O3 impact on plants has moved towards O3 uptake-based thresholds and indicators to replace O3 exposure-based indices such as AOT40 (Ashmore et al., 2004; Matyssek et al., 2007b, 2008). The relationship of PC1 with AOT40 and COU was similar in the dry and the average year, and was similar in the sun and shade crown (Fig. 2b,c). In our study, the covariates AOT40 and COU both had a statistically significant effect on PC1 at P <0.001. However, COU contributed relatively more to the R2 of the GLM (comparable to the factor year), whereas the relative contribution of AOT40 was smaller. This result supports COU as a better index to potentially predict multivariate plant responses to O3.
Our study shows that multivariate analysis is able to detect O3-related changes in leaf physiology (PC1) in the absence of consistent O3 effects on individual parameters. As PC1 showed a consistent O3-related effect over two growing seasons, which was even found during the severe drought of 2003, it underlines the susceptibility of adult trees growing in a forest environment to elevated O3.
M.L. was supported by the Pownall, Irving, Davies Award, Department of Forest and Ecosystem Science, University of Melbourne. The CASIROZ project was supported by the European Commission – Research Directorate General, Environment Programme, Natural Resources Management and Services (contract no. EVK2 – 2002 – 00165, Ecosystem Vulnerability). Further research at Kranzberger Forst was supported by the integrated research centre ‘SFB 607: Growth and parasite defence – competition for resources in economic plants from agronomy and forestry’, funded by the Deutsche Forschungsgemeinschaft (DFG) and by Bayerische Staatsforsten AöR, Regensburg, the owner of the site. During his sojourn at the University of Antwerp, M.L. was supported by the Methusalem funding of the Research Center of Excellence ECO. We are grateful to L. Zimmermann, Bayerische Landesanstalt für Wald und Forstwirtschaft, Freising, for providing precipitation and soil moisture data. We would like to thank three anonymous reviewers and the communicating editor for their valuable input.