Study sites and microclimatic conditions

The study was carried out in five mature beech (Fagus sylvatica L.) stands of similar age and structure along a precipitation gradient in the Pleistocene lowlands of north-western Germany in 2011. The precipitation gradient covers a 130-km-long northwest–southeast transect from the east of the state of Lower Saxony to the western part of Saxony-Anhalt and includes a climatic gradient from a suboceanic to a subcontinental climate with a continuous long-term (1981–2010) mean annual precipitation (MAP) decrease from 855 to 594 mm yr⁻¹, a decrease in mean annual early summer growing season (April-June)precipitation (MSP) from 188 to 147 mm yr⁻¹ and a mean annual temperature (MAT) increase from 8.5 to 9.1°C (Table 1). Climatic data were obtained from a 1 × 1 km² grid data set (Deutscher Wetterdienst, Offenbach, Germany). We additionally calculated a simplified forest aridity index (FAI) according to Führer et al. (2011) as FAI = 100 × T_Jul–Aug/(P_May–Jul + P_Jul–Aug), where T is the temperature and P the precipitation of the associated interval. As the atmospheric evaporative demand in the growing season is highest in midsummer (July and August), the July precipitation was weighted by a factor of two in the denominator. This index has been developed for the comparison of different beech stands in southeast Europe; FAI values in the distribution range of beech are generally < 4.75 (Führer et al., 2011).

Table 1. Stand characteristics of the five investigated beech (Fagus sylvatica) forests along a precipitation gradient in northwest Germany

Site	Code	Symbol	MAP	MSP	MAT	FAI	DBH	H	n tree	n sample
Given are the site codes and symbol colours, mean annual precipitation (MAP; mm yr−1) and mean early summer growing season (April–June) precipitation (MSP; mm) for the period 1981–2010, mean annual temperature (MAT; °C), the forest aridity index (FAI), the average diameter at breast height (DBH; cm), tree height (H; m) and the numbers of tree individuals measured (ntree) and samples per site (including pseudoreplicates; nsample). Values are mean ± SE. The sample numbers for wood density and transmission electron microscopy measurements are not given here; see the corresponding Materials and Methods section.
Calvörde	Ca		593.91	146.75	9.3 ± 1.9	6.19	37.42 ± 2.47	25.82 ± 0.18	5	13–17
Klötze	Kl		654.76	157.30	9.1 ± 1.9	5.60	42.36 ± 3.65	29.71 ± 0.29	5	14–16
Göhrde	Go		707.11	164.57	9.0 ± 1.8	5.15	40.13 ± 2.00	24.20 ± 1.14	5	16–19
Unterlüß	Un		816.06	175.96	8.7 ± 1.8	4.86	40.83 ± 3.07	22.27 ± 1.10	5	16–18
Sellhorn	Se		855.37	188.14	8.7 ± 1.8	4.53	39.79 ± 2.43	28.93 ± 1.04	5	14–17

All five sites were situated on highly acidic and nutrient-poor sandy soils developed in fluvioglacial sands or moraine deposits of the penultimate Ice Age (Saalian) covered by periglacial drift sand. Mean (± SE) tree age was 104.6 ± 6.7 yr. For detailed information on stand structure and climatic parameters at the five study sites, see Müller-Haubold et al. (2013) and Hertel et al. (2013). Some structural data on the individual beech trees investigated are given in Table 1.

Tree selection and plant material

Five mature beech trees of similar size and canopy position within a particular stand were selected at the five sites and branch and twig samples collected from the uppermost sun-exposed crown with tree-climbing equipment in August 2011. Per tree, five twigs of c. 50 cm length were air-cut and immediately transferred to polyethylene tubes filled with water containing a sodium-silver-chloride complex (16 μg l⁻¹ silver (Ag) and 8 mg l⁻¹ NaCl; Micropur katadyn, Wallisellen, Switzerland) to prevent microbial growth and stored at 4°C. Across the gradient, all branches were more or less of comparable size and age, although a certain variation in sapwood area at a given branch age was unavoidable (Supporting Information Fig. S1). Additionally, all leaves distal to the sampled twig segments were harvested. For wood density determination, three large branch wood samples were collected from the uppermost canopy of a tree. A list of all traits measured, the corresponding acronyms and units is given in Table 2.

Wood density

Wood density (WD; g cm⁻³) was determined for three branch wood samples per tree (mean diameter ± SE: 3.09 ± 0.07 cm; mean length ± SE: 15.15 ± 0.43 cm), yielding 64 samples in total. Fresh volume was gravimetrically measured at a precision of 10 mg after removing pith and bark by water displacement according to Archimedes' principle and branch samples subsequently oven-dried at 105°C for 72 h.

Hydraulic conductivity and vulnerability curves

Hydraulic properties were measured for three to five samples per tree, yielding 78 samples in total. Branch segments were shortened to c. 28 cm length (mean basipetal diameter ± SE: 8.04 ± 0.11 mm), lateral branches were cut off, scars were sealed with quick-drying superglue (Loctite 431; Henkel, Düsseldorf, Germany) and the segments were connected to the Xyl'em apparatus (Bronkhorst, Montigny les Cormeilles, France). Segments were flushed three times for 10 min at a pressure of 120 kPa with filtered (0.2 μm) and degased demineralized water (10 mM KCl and 1 mM CaCl₂) and maximum hydraulic conductivity (K_h; kg m MPa⁻¹ s⁻¹) was recorded along a 6-kPa pressure difference. The diameter of each segment was measured twice at the basipetal and distal ends, and at four positions along the segment. In order to calculate specific conductivity (K_S; kg m⁻¹ MPa⁻¹ s⁻¹) normalized by sapwood area, a regression analysis between total cross-sectional area (A_cross; mm²) and corresponding xylem cross-sectional area (A_xylem; mm²) was carried out assuming that all rings were still functional. From each segment, high-quality top-view images from the planed thick and thin ends were analysed for A_cross and A_xylem with the software ImageJ (v.1.44p; http://rsb.info.nih.gov/ij). The following regression coefficients were used to calculate sapwood area without pith and bark for a given segment diameter: A_xylem = −3.715 + 0.770 A_cross (P < 0.001; r² = 0.98; n = 238). According to linear regression analyses, K_h divided by the maximal basipetal sapwood area revealed the strongest relationships (Table S1). This pattern was documented in Hajek et al. (2014) and Hoeber et al. (2014). Leaf -specific conductivity (K_L; kg m⁻¹ MPa⁻¹ s⁻¹) was calculated by dividing K_h by the total supported leaf area distal to the branch segment (A_leaf; cm²).

Subsequently, branch segments were inserted into a honeycomb custom-made rotor (Delzon et al., 2010) of the Cavitron (Cochard et al., 2005) attached to a commercially available centrifuge (Sorvall RC-5C; Thermo Fisher Scientific, Waltham, MA, USA). Conductivity measurements started at 1.0 MPa and were repeated stepwise at intervals of 0.2–0.3 MPa until the percent loss of conductivity (PLC) reached at least 90%. Vulnerability curves were generated by plotting PLC against xylem pressure (Fig. 1), and the pressure causing 50% loss of conductivity (P₅₀) was calculated according to a sigmoidal function (Pammenter & Vander Willigen, 1998) as PLC = 100/(1 + exp(s/25 × (P_i − P₅₀))), where s (% MPa⁻¹) is the negative slope of the curve at the inflexion point and P_i the xylem pressure. The xylem pressures causing 12% (P₁₂; air entry point) and 88% (P₈₈) loss of conductivity were also calculated following Domec & Gartner (2001) as P₁₂ = 2/(s/25) + P₅₀ and P₈₈ = −2/(s/25) + P₅₀.

Wood anatomy, potential conductivity and branch growth rate

Branch segments from the basipetal end were used for wood anatomical investigation, yielding 77 samples in total. Transverse sections cut with a sliding microtome (G.S.L.1; Schenkung Dapples, Zurich, Switzerland) were digitalized at ×100 magnification using a stereomicroscope equipped with a digital camera (SteREOV20; Carl Zeiss MicroImaging GmbH, Göttingen, Germany). Image processing was performed with the software Adobe Photoshop CS6 (v.13.0.1; Adobe Systems Inc., San Jose, CA, USA) and ImageJ using the particle analysis function to estimate single and cumulative vessel lumen area (A_lumen; m²), vessel density (VD; n mm⁻²) and vessel diameter (D; μm) from major (a) and minor (b) vessel radii according to the equation given by Lewis & Boose (1995) as D = ((32 × (a × b)³)/(a² + b²))^¼ and used to calculate the hydraulically weighted diameter (D_h; μm) according to Sperry et al. (1994) as D_h = ƩD⁵/ƩD⁴. Relative vessel lumen area (A_lumen : A_xylem; %) was calculated by dividing cumulative vessel lumen area (A_lumen; m²) by the corresponding sapwood area (A_xylem; m²). Carlquist's vulnerability index (VI) (Carlquist, 1977), which is commonly used to indicate a species' adaptation to xeric or mesic conditions (De Micco et al., 2008), was calculated as VI = (D/1000)/(VD/1000 000). The vessel grouping index (V_G) was estimated by dividing the total number of vessels by the number of grouped vessels, and the solitary vessel index (V_S) by dividing the number of solitary vessels by the total number of vessels; for these calculations, a subsample of 216.4 ± 5.6 vessels (mean ± SE) were measured per branch sample, with both solitary vessels and grouped vessels considered as a vessel group (Scholz et al., 2013). For all calculations (except for V_G and V_S), the complete cross-section was analysed, yielding 3537–30 989 measured vessels per branch sample. For V_G and V_S, pie slices from the cross-section were used, yielding 110–379 measured vessels per sample. Potential conductivity (K_P; kg m⁻¹ MPa⁻¹ s⁻¹) was calculated according to the Hagen–Poiseuille equation as K_P = (((π × ƩD⁴)/128 η) × ρ)/A_xylem, where η is the viscosity (1.002 10⁻⁹ MPa s) and ρ the density of water (998.2 kg m⁻³), both at 20°C, and A_xylem (m²) is the analysed sapwood area. Additionally, the relative abundance of vessels in five vessels size classes was calculated by dividing the number of vessels in a class by the total vessel number.

We further calculated branch growth rate (G_dw; mg m⁻¹ yr⁻¹) according to Sterck et al. (2012) as G_dw = BAI × WD × 10⁻⁶, where BAI is the average annual basal area increment (m² yr⁻¹) and WD the corresponding wood density (kg m⁻³). BAI was calculated by analysing the area of each growth ring separately, yielding 451 analysed growth rings in total.

Transmission electron microscopy

A subsample of three to four branch segments from each of three trees per site was used for transmission electron microscopy (TEM) observations of intervessel pit membrane thickness (T_m; nm) and intervessel wall thickness (T_w; nm), yielding 18 samples in total. Samples were stored in 70% ethanol, and were from the same branches that were used to construct vulnerability curves and for wood anatomical measurements. Samples were prefixed in a standard solution (2.5% glutaraldehyde, 0.1 mol phosphate and 1% saccharose, pH 7.3) and prepared according to a standard TEM protocol (Jansen et al., 2009b). Ultrathin, transverse sections were observed with a JEOL 1400 transmission electron microscope (JEOL Ltd, Tokyo, Japan). The conduit size and general morphology of bordered pits were used to distinguish intervessel pit membranes from vessel–tracheid or tracheid–tracheid pit membranes. Given the morphological continuum between vessel elements and tracheids, the exact nature of conduits could not always be distinguished. Therefore, TEM measurements were only included in our analyses when the vessel identity was clear.

Leaf morphology

All leaves distal to the twig segments used for hydraulic conductivity measurements were stripped off and scanned for determination of single and cumulative leaf areas (WinFolia 2005; Régent Instruments, Quebec City, QC, Canada). Per branch segment, eight to 132 leaves were scanned, yielding 5464 leaves in total. The average leaf size (LS; cm²) was determined by dividing the total leaf area by the number of leaves per branch, and the Huber value, that is, the sapwood-to-leaf-area ratio (A_xylem : A_leaf; 10⁻⁴ m² m⁻²), was calculated by dividing the sapwood area determined according to the linear regression analysis mentioned under hydraulic conductivity and vulnerability curves by the corresponding total leaf area (A_leaf; cm²). Subsequently, leaves were oven-dried at 70°C for 48 h in order to determine the specific leaf area (SLA; cm² g⁻¹). The carbon isotope signature (δ¹³C) of the leaf dry mass was analysed by mass ratio spectroscopy (Deltaplus; ThermoFinnigan, Bremen, Germany) at the Centre for Stable Isotope Research and Analysis (KOSI), University of Göttingen.

Statistical analyses

The trait variables investigated are summarized in Table 2. Our approach and the labour-intensive process of gaining access to the uppermost canopy using tree-climbing equipment forced us to apply a nested design, that is, three to five samples were taken per tree, but only five trees were analysed per site.

All statistical analyses were performed with the software package R (R Development Core Team 2013, version 3.0.0) except for linear regression analyses which were executed with the software Xact 8.03 (SciLab, Hamburg, Germany). All variables were tested for normality with a Shapiro–Wilk normality test and log-transformed if required. Linear mixed effect (LME) models with FAI, MAP or MSP as a fixed variable were used to test for significant differences in all trait variables along the rainfall gradient. We assumed nonindependence of different samples within a tree and of different trees within a plot in the models by adding plot and tree nested in plot as random effects. For investigating relationships between trait variables, Pearson correlation analyses were carried out on the tree level.

To search for patterns of interrelationships among the examined 23 traits, a principal components analysis (PCA) was conducted with plot-level means using the software package Canoco (v.5.02; Biometris, Wageningen, the Netherlands) with all functional traits centred and standardized before analysis. Although our number of variables exceeded the number of plots by far, the first few eigenvectors are little affected when the matrix is not of full rank and do not lead to incorrect interpretations of ordinations in reduced space (Legendre & Legendre, 1998).

Variation in functional traits along the precipitation gradient

Across our gradient, MAP declined by 261 mm yr⁻¹ while MSP (from April to June) declined by 41 mm. Changes in precipitation and the increase in MAT by 0.4°C from the wet to the dry end were additionally mirrored in the FAI, which increased by 1.66 (Table 1).

According to the LME models, only five of the 23 structural, hydraulic, wood anatomical and foliar traits were significantly related to the three measures of water availability, MAP, MSP and FAI (Table 3). A key result is that xylem safety increased linearly with an increase in climatic aridity (Fig. 2a). The three measures of embolism resistance (P₁₂, P₅₀ and P₈₈) differed in their responses, with P₁₂ showing the strongest decrease with a reduction in MAP (by 15% or 0.44 MPa), followed by P₅₀ (by 10% or 0.33 MPa), while P₈₈ was the least modified trait (by 5% or 0.23 MPa; Fig. 1; Table S2). The increase in hydraulic safety towards the drier sites was not associated with a decline in xylem efficiency measured as specific (K_S; Fig. 2b) or potential conductivity (K_P; Table 3), even though the highest conductivities were observed at the wettest site, Sellhorn (Se; Table S2). In contrast to K_S, K_L was strongly influenced by changes in MAP and declined by 40% towards the dry end of the gradient (Fig. 2c).

WD did not vary significantly along the gradient (Fig. 2d), and of all wood properties examined only D was significantly related to both FAI and MSP (and at marginal significance to MAP; P < 0.10; Table 3). From the wet to the dry end of the gradient, average D declined by 7% (i.e. by 1.3 μm) (Figs 2e, 3a) as a result of changes in the abundance of vessels in the two size classes 10–20 μm (D_10–20 μm; increase by 10%) and 20–30 μm (D_20–30 μm; decline by 10%; Fig. 3b), while the abundance of vessels in the other three size classes was unaffected by the precipitation level (Table S3). As a consequence, D_h did not vary across the gradient (Tables 3, S2), which contrasts with the decrease in D.

None of the parameters related to vessel distribution, that is, VD (Fig. 2f), V_G (Fig. 2g), V_S and VI, was altered in response to the precipitation decrease (Table 3). By contrast, T_m increased by 15% with increasing climatic aridity (Fig. 2h), while T_w remained unchanged (Table 3). The strong negative relationship between MAP and T_m according to the linear regression analysis could, however, not be confirmed by the LME models (Table 3), presumably because of the low sample number and thus high variation within two of the five sites (Table S2). Neither branch growth rate (G_dw) nor the lumen to sapwood area ratio (A_lumen : A_xylem) or A_xylem : A_leaf was significantly related to the precipitation decrease.

The precipitation gradient had a surprisingly small effect on the sun canopy leaves of the beech trees: neither SLA nor the foliar carbon isotope signature (δ¹³C) changed along the gradient, suggesting that higher climatic aridity did not lead to stomatal closure. However, we found a trend to larger leaf sizes at the drier sites, contrary to expectation (P < 0.10; Fig. 2i).

Determinants of hydraulic efficiency and branch growth

At the tree level, the variation in K_L was caused by alteration of the A_xylem : A_leaf ratio (P < 0.001; r = 0.60) and, to a lesser extent, with modification of D (P < 0.05; r = 0.41), but K_L showed no relationship to VD, D_h or K_S (Table 4). Differences in K_S, by contrast, were mainly driven by changes in D (Fig. 4a) and not by VD (Fig. 4d). In contrast to K_S, wood density was unrelated to D (Fig. 4b) but closely associated with VD (Fig. 4e). We further could confirm the commonly assumed trade-off between hydraulic efficiency and growth; fast-growing branches possessed a more efficient hydraulic system composed of larger (Fig. 4c) but fewer vessels (Fig. 4f) as compared with slower growing branches, and a significant positive relationship existed between G_dw and both K_S and K_P (P < 0.05; r = 0.44), but there was also a significant negative relationship with WD (P < 0.05; r = 0.40; Table 4). Furthermore, G_dw was negatively related to branch age (BA; P < 0.001; r = 0.64) and LS (P < 0.01; r = 0.48), and positively to δ¹³C (P < 0.05; r = 0.35), indicating that younger branches grew faster at the cost of more frequent stomatal closure. This finding is further supported by the negative relationship between BA and K_S (P < 0.01; r = 0.52), and the absence of any relationship between BA and WD (Table 4).

Determinants of embolism resistance

At the tree level, only two of eight examined traits commonly associated with embolism resistance (K_S or K_L, WD, D, VD, V_G, T_m, A_xylem : A_leaf and SLA) were found to be related to embolism resistance in our study (Table 4). Contrary to expectation, neither hydraulic efficiency (K_S and K_L) nor D decreased with increasing embolism resistance, indicating that a clear trade-off between hydraulic efficiency and safety does not exist in European beech. However, both VD and T_m were negatively related to embolism resistance, with VD showing a closer relationship to the air-entry point at high pressure (P₁₂; P < 0.01; r = 0.46) than to the P₅₀ value (P < 0.05; r = 0.34), while T_m was most strongly associated with P₈₈ (P < 0.05; r = 0.56), that is, the full embolism point at low pressure, and less strongly to P₅₀ (P < 0.10; r = 0.40). Increased embolism resistance is achieved by the beech trees along the gradient by producing more vessels arranged in distinct groups but not necessarily by reducing vessel diameter (Fig. 5a); this is indicated by the positive relationship between VD and V_G (Fig. 5b). This finding is supported by a negative relationship between V_G and P₁₂ at marginal significance (P < 0.10; r = 0.28). These adjustments in the vascular system apparently allowed trees to maintain a favourable leaf water status at the drier sites, as indicated by the negative relationship between VD and δ¹³C (Fig. 5c), that is, a higher vessel density decreased the need to reduce leaf conductance.

The Pearson correlation analysis revealed several relationships between embolism resistance and morphological and growth-related parameters that were to some extent unexpected: BA was positively related to P₅₀ (P < 0.05; r = 0.34; Table 4, Fig. S2b) and P₈₈ (P < 0.01; r = 0.49), and branch size (A_xylem) was positively associated with P₁₂ (P < 0.05; r = 0.37), P₅₀ (P < 0.01; r = 0.47) and P₈₈ (P < 0.01; r = 0.49; Table 4). G_dw as a measure of productivity was negatively related to P₈₈ (P < 0.05; r = 0.34) but not P₁₂ and P₅₀.

Influence of branch age on embolism resistance across the gradient

Although we carefully sampled branches of similar size across the gradient, a certain variation in BA per given A_xylem was observed (Fig. S1). This variation was unrelated to changes in MAP and thus not site-specific (Table 3). According to an LME model with BA added as a covariable or as a third random effect, MAP was still significantly related to all three measures of embolism resistance (covariable: P < 0.01; random effect: P < 0.05). We further removed the youngest and oldest branches from the data set, that is, the 15% extreme ages younger than 4 years and older than 9 years; this linear regression analysis supported the LME model with BA included as a covariable or random effect (Fig. S2a). Consequently, BA can be excluded as the principal driver of embolism resistance along the gradient.

Mutual interrelationships between traits

The interrelationships between hydraulic, wood anatomical and foliar traits revealed by the tree-level Pearson correlation analysis (Table 4) principally match the results of a PCA conducted with plot-level means. The first two principal components together explained 72% of the variance (Table 5). Axis 1 (eigenvalue 0.44) was strongly associated with MAP and all hydraulic and wood anatomical traits including foliar δ¹³C, but showed no relationship to K_L, P₈₈ and the traits depending on sapwood area. These parameters and the foliar traits SLA and LS as well as G_dw were associated with axis 2. The plot-level analysis thus reveals relationships between climatic aridity, branch wood anatomical properties, hydraulic efficiency and embolism resistance that are closer than those found in the tree-level analysis shown in Table 4: neither hydraulic efficiency nor vessel diameter was related to embolism resistance at the tree level but such relationships became visible at the plot level (Fig. S3). Our differential analyses at the branch, tree and plot (stand) levels revealed that, for most functional traits, the largest variation existed among the branches within a tree, followed by the variance between the trees of a plot, while a much lower variation was detected between the five plots along the precipitation gradient (see Table S4). This suggested selecting the tree as the focal object of analysis and pooling over the replicate branches, while examining differences between plots in a second step.

Climatic influences on wood anatomical, hydraulic and foliar traits

We observed a close relationship between MSP and D in similarly sized branches from the uppermost canopy of mature beech trees, partly supporting our first hypothesis which postulated that vessel diameter and hydraulic efficiency decrease with a lasting reduction in precipitation. The decline in D by 7% from the wet to the dry end of the gradient was nearly exclusively caused by shifts in the relative abundance of vessels with a diameter between 10 and 30 μm. In contrast, the three larger vessel diameter classes, which primarily consist of early-wood vessels and are of great importance for the calculation of D_h, were unaffected. As a consequence, D_h did not change across our precipitation gradient. The altered diameter pattern is also mirrored in the absence of any relationship between either MAP or MSP and K_P or K_S, because hydraulic conductivity increases with diameter raised to the fourth power according to the Hagen–Poiseuille law, and the diameter alteration in the 10–30 μm class is of low importance for K_P.

The functionality of the vascular system is best analysed in conjunction with the dependent foliage for understanding wood anatomical modifications in response to the environment (Carlquist, 2012). Both tissues related to water transport might have coevolved with respect to traits governing xylem vulnerability to embolism (Brodribb et al., 2003; Nolf et al., 2015). Although K_P and K_S remained more or less constant, K_L decreased unexpectedly by 40% with a MAP decrease of 260 mm yr⁻¹. Generally, K_L is supposed to increase in plants that are frequently exposed to drought in order to maintain a favourable leaf water status (Mencuccini & Grace, 1995; Sterck et al., 2008). This is achieved either by an increase in K_S or more likely by adjustment of A_xylem : A_leaf. In Scots pine (Pinus sylvestris), for example, both K_L and A_xylem : A_leaf increased with climate dryness across Europe, while K_S and D remained unchanged (Martinez-Vilalta et al., 2009; Sterck et al., 2012). This indicates that this coniferous species responds to increasing drought exposure mainly by reducing the supported leaf area per unit sapwood area. In beech, the decrease in K_L with a reduction in MAP along the gradient was caused by an increase in dependent leaf area towards the drier stands. By contrast, most other studies found an increase in A_xylem : A_leaf with increasing climatic aridity, both within and between species (Poyatos et al., 2007; Gleason et al., 2013). Along our gradient, however, mean leaf size tended to increase, and not decrease, towards the drier eastern stands as also observed by Meier & Leuschner (2008) in a precipitation gradient of beech forests on sandstone in Central Germany.

A major finding of this study is that embolism resistance increased in parallel with declining D with increasing climatic aridity across our gradient, supporting our second hypothesis. To our knowledge, this is the first study demonstrating a linear intraspecific increase in embolism resistance with increasing drought stress. This finding provides additional support for the assumption that F. sylvatica is highly plastic with respect to embolism resistance (Herbette et al., 2010), and that embolism resistance is under environmental rather than genetic control in this angiosperm tree species (Wortemann et al., 2011). In P. sylvestris and Pinus pinaster, by contrast, P₅₀ was unrelated to climatic aridity (Martinez-Vilalta et al., 2009; Lamy et al., 2014). These contrasting results might be explained by the observation that populations of angiosperm tree species generally possess a higher intraspecific variation in hydraulic traits than conifers (Anderegg, 2015). However, one has to keep in mind that vessel diameter, hydraulic properties and embolism resistance are influenced not only by the moisture regime but certainly also by other biotic and environmental factors, which complicates the interpretation of results obtained from climatic gradients. It has been shown that tree size and sampling height (Lintunen & Kalliokoski, 2010; Anfodillo et al., 2013), as well as soil properties, in particular nutrient availability (Goldstein et al., 2013) and water storage capacity (Tokumoto et al., 2014), strongly influence the hydraulic architecture of plants. In cross-regional comparisons, it is therefore mandatory to keep tree size (or age) and soil properties as constant as possible in order to detect effects of environmental factors.

The observed modification of the hydraulic system across our precipitation gradient indicates that trees growing at the dry end tolerate more negative stem water potentials before water transport to the canopy is impeded. That this hydraulic adaptation was effective with respect to the maintenance of leaf gas exchange under drier conditions is suggested by the invariance of leaf δ¹³C across the precipitation gradient.

Relationships between wood properties and embolism resistance

One of the principal hypotheses of plant hydraulics postulates a general trade-off between hydraulic safety and efficiency. This hypothesis is not confirmed by our data, which is in line with the majority of studies recently summarized by Gleason et al. (2016). Although both D and the three measures of embolism vulnerability decreased with climatic aridity and a close relationship between D and hydraulic efficiency was observed, neither D nor K_S was related to xylem safety. Tyree & Sperry (1989) argued that D is not directly related to the mechanism of embolism formation but rather results from a correlation with pit membrane pore size. Nevertheless, several studies have demonstrated this close, albeit presumably indirect, relationship among and across species (Wheeler et al., 2005; Maherali et al., 2006; Domec et al., 2010; Hajek et al., 2014). We assume that the range of variation in D and P₅₀ was too small to confirm this relationship across the trees in our intraspecific study. Instead, VD was closely related to the air-entry point of embolism formation (P₁₂) but not to hydraulic efficiency. The V_G was likewise associated with P₁₂ (although at marginal significance only) and closely related to VD, as observed by Lens et al. (2011). These results are in line with Carlquist's (1966) assumption that vessels should group to a larger degree in species adapted to water-limited environments with a xylem tissue composed of nonconductive fibres and no vasicentric tracheids; this would enable the maintenance of the conductive flow path by pathway redundancy if the largest vessel of the group embolizes. Through such vascular modification with more and to a higher degree grouped vessels in order to lower the xylem pressure at the onset of embolism formation, trees are presumably able to tolerate more negative leaf water potentials before stomata have to be closed. This assumption is supported by our negative relationship between VD and foliar δ¹³C. It has further been shown that stomatal closure correlates with the entry point of embolism formation in angiosperms, that is, the P₁₂ value (Nardini et al., 2001; Cochard et al., 2002).

In contrast to VD and V_G, T_m correlated best with the xylem pressure inducing 88% loss of hydraulic conductance (P₈₈), although a marginally significant relationship also existed with P₅₀. It has been shown that the thickness of the membrane in angiosperm xylem is directly related to its maximum pore diameter (Jansen et al., 2009a), making air-seeding more likely for thin membranes (Wheeler et al., 2005). However, this does not explain why in our sample the closest relationship existed to the most negative pressure at P₈₈. It is likely that there are still serious gaps in our mechanistic understanding of air-seeding in plants (Schenk et al., 2015).

For three other traits often associated with embolism resistance, namely WD, A_xylem : A_leaf and specific leaf area, we were not able to confirm such a relationship in our data set. Such relationships may become visible only in data sets spanning different species or in intraspecific studies with much greater site and trait variation than is encountered in our precipitation gradient. Other biological properties may also influence embolism resistance, notably branch age, as we found older branches to be most vulnerable to cavitation, in agreement with Domec et al. (2009). However, we believe that this effect is primarily caused by cavitation fatigue, that is, the reduction of cavitation pressure as a consequence of previous cavitation in older xylem elements (Hacke et al., 2001), even though we corrected for the potential bias of fatigue by applying a certain pressure before the onset of conductivity measurements (Jacobsen et al., 2007). Further research is needed to show how many growth rings of older branches remain conductive in angiosperm trees (but see Christensen-Dalsgaard & Tyree, 2014) and to clarify whether sample flushing before the construction of vulnerability curves introduces measuring artefacts.

Relationship between hydraulic efficiency and branch growth

Similar to Sterck et al. (2012), we use sun-canopy G_dw as a measure of productivity because a dependence of growth on branch hydraulics is more likely at this level than for stem wood or total aboveground productivity. In our study, faster growing younger branches of similar size to older ones possessed a more efficient hydraulic system and supported larger leaves, which subsequently caused more frequent stomatal closure. This finding is in agreement with our third hypothesis assuming a close relationship between hydraulic efficiency and growth, which might trade off with embolism resistance (Cochard et al., 2007).

Unexpected is our finding that older slower growing branches in a given diameter class were more vulnerable than younger faster growing branches, as is indicated by the negative relationship between G_dw and P₈₈. Generally, fast growth is associated with larger vessels that cause higher hydraulic efficiency, which should result in a relatively vulnerable xylem (Hajek et al., 2014). Although all branches investigated originated from the uppermost sun-exposed canopy, we assume that the older and the slower growing branches of similar size have not been exposed to full sunlight and thus to high atmospheric demand in the past, presumably because they were shaded by younger fast-growing branches. A similar pattern was observed by Herbette et al. (2010) when comparing sun-exposed and deeply shaded beech branches; the latter were on average 50% more vulnerable. This high variability in xylem embolism resistance, even at the same canopy position, is strong support for the frequently claimed high phenotypic plasticity in the morphology and physiology of European beech (Herbette et al., 2010; Bresson et al., 2011; Wortemann et al., 2011). Shade-tolerant beech trees produce leaves that tolerate either full sunlight or deep shade and it can be assumed that this great plasticity in leaf function is also reflected in the hydraulic system of branches and petioles. However, given that G_dw was not related to any other parameter of embolism resistance in our study, it is premature to conclude that a trade-off between embolism resistance and branch growth does exist in the beech trees of our study. In agreement with this assumption, several other studies failed to detect relationships between vulnerability and growth (Sterck et al., 2012; Hajek et al., 2014; Guet et al., 2015), indicating that embolism resistance is partly decoupled from hydraulic efficiency and biomass production (Fichot et al., 2015).

Conclusions

In the sun canopy of beech trees, morphological adjustment to permanently reduced precipitation seems to occur primarily through the development of more cavitation-resistant branch wood, and not through a reduction of the supported leaf area. As a result of this modification, trees at the dry end of the gradient are capable of tolerating 15% more negative stem water potentials before the onset of cavitation. Our results show that the branch hydraulic system of beech has a considerable adaptive potential to respond to the local climatic conditions. This finding is strong evidence in support of the assumption that embolism resistance is largely under environmental control in this angiosperm tree species. It remains unclear, however, whether the observed hydraulic variability at the branch level can fully buffer against constraints imposed on growth and vitality by a warming and drying climate. Over the last three to four decades, a radial growth decline has been observed in those beech stands of our study region that are exposed to less than c. 635 mm yr⁻¹ precipitation. This casts doubt on the capability of F. sylvatica to withstand increased drought and warming stress in the future solely through its flexible hydraulic response.