Leaf hydraulic capacity and vulnerability to drought stress are key determinants of plant competitive ability and productivity. Yet, it is not clear how these traits correlate to climatic variables across biomes and whether a trade-off exists between them.
We collected leaf hydraulics data for 130 woody angiosperms from selected published articles. Species-specific values of leaf hydraulic capacity on a leaf area (Kleaf_area) and dry mass (Kleaf_mass) basis, leaf water potential inducing 50% loss of Kleaf (P50), as well as climatic variables (mean annual temperature, MAT and precipitation, MAP) for study sites were collected. Species were classified as belonging to three major biomes, that is dry sclerophyllous forests (DSF), temperate forests (TMF) and tropical forests (TRF).
Significant differences were observed between biomes, with DSF species displaying the lowest hydraulic efficiency (low Kleaf) and the highest resistance to drought stress (low P50). P50 was correlated with both MAP and MAT, with species from low precipitation habitats having the lowest P50 values. Both Kleaf_area and Kleaf_mass were positively correlated with MAP and MAT. Leaf gas exchange rates were positively correlated with both Kleaf_area and Kleaf_mass. Although no correlation was found between P50 and Kleaf_area, a weak trade-off between leaf hydraulic safety and capacity emerged when P50 was plotted versus Kleaf_mass.
Leaf hydraulics emerge as an important functional trait underlying plant adaptation to different habitats and contributing to shape vegetation features in different biomes.
Plant gas exchange, photosynthesis and terrestrial productivity are constrained by the root-to-leaf water transport capacity (Sperry 2000). In fact, CO2 uptake through stomata is unavoidably coupled to massive water loss from photosynthetic tissue to the atmosphere (Jones 1998), and only adequate hydraulic efficiency of plant water transport systems can warrant continuous replacement of water lost by transpiration and maintenance of leaf hydration.
The plant plumbing system can be conveniently described on the basis of a hydraulic analogue of the Ohm's law, as being composed by three main hydraulic resistances arranged in series to each other, that is the root system, trunk and lateral stems, and foliage (Tyree & Ewers 1991; Cruiziat, Cochard & Améglio 2002). Leaves, in particular, are known to make up a very large share (30–60%) of plant hydraulic resistance (Yang & Tyree 1994; Nardini 2001; Sack & Holbrook 2006), as a likely consequence of the intricate vein system and several cell layers that water has to cross while flowing from the petiole to the sites of evaporation (Cochard, Nardini & Coll 2004b; Sack, Streeter & Holbrook 2004). Hence, it is not surprising that leaf hydraulics has been reported to represent a major determinant of plant productivity (Gortan et al. 2009; Brodribb & Feild 2010). In particular, species-specific values of leaf hydraulic conductance (Kleaf) per unit surface area (Kleaf_area) have been reported to be positively correlated to maximum stomatal conductance to water vapour and net photosynthesis (Brodribb et al. 2005), underlying the important contribution of Kleaf_area variation across species and functional groups to plant persistence, reproduction and competitive success (Hao et al. 2008; Nardini, Pedà & La Rocca 2012a).
High Kleaf_area and photosynthetic productivity have been shown to be correlated with increased vein length per unit surface area (vein density), suggesting that shortening the extra-vascular pathway from bundle sheaths to mesophyll evaporation sites is a key structural modification to achieve high rates of water flow in the leaf (Brodribb, Feild & Jordan 2007), allowing higher evaporative water loss coupled to higher stomatal density/aperture and fast CO2 uptake (Franks, Drake & Beerling 2009). High major vein density has been also suggested to play a role in determining differential drought resistance of plant species (Scoffoni et al. 2011). In fact, the leaf hydraulic system is very sensitive to water stress, and declines in leaf water potential typically induce fast reduction in Kleaf (Brodribb & Holbrook 2006). Leaf hydraulic vulnerability is generally quantified in terms of P50, that is, the leaf water potential value inducing 50% loss of Kleaf with respect to values recorded in well-hydrated samples (Brodribb & Holbrook 2003). Kleaf drop under drought stress conditions is caused by reductions in water transport capacity of the vein xylem because of cavitation-induced embolism (Nardini, Salleo & Raimondo 2003) and xylem wall collapse (Cochard et al. 2004a), or impairment of the extra-vascular water pathway as due to blockage of aquaporins (Shatil-Cohen, Attia & Moshelion 2011) and cell shrinkage. However, most experimental evidences point to xylem embolism as the major determinant of Kleaf impairment under drought stress conditions (Trifilò et al. 2003; Johnson et al. 2012). Higher major vein density would confer greater tolerance to drought-induced hydraulic disruption to leaves, by providing alternative water pathways to by-pass embolized veins (Sack et al. 2008), thus significantly lowering the P50 and allowing leaves to supply water to photosynthetic cells even at relatively lower water potential values (Scoffoni et al. 2011; Nardini, Pedà & La Rocca 2012a).
At the stem level, increased resistance to embolism-induced hydraulic dysfunction often comes at the cost of reduced hydraulic efficiency (Bucci et al. 2006) and higher carbon costs (Jacobsen et al. 2007). The validity of such an efficiency/safety trade-off paradigm at the leaf level has been questioned on the basis of lack of correlation between P50 and Kleaf_area (Blackman, Brodribb & Jordan 2010; Scoffoni et al. 2012). However, other studies showed that scaling Kleaf by dry mass, to account for the efficiency of the leaf plumbing system to supply water to the bulk of leaf cells and to estimate the carbon cost of leaf hydraulic construction (Nardini, Pedà & Salleo 2012b; Simonin, Limm & Dawson 2012), does actually suggest the existence of relationships between Kleaf_mass, P50, vein density and leaf mass per unit surface area (LMA) (Nardini, Pedà & La Rocca 2012a). These findings lead to hypothesize that drought resistance could be achieved by plants by producing leaves of small size, with high major vein density (and with smaller, more numerous and thick-walled conduits, Blackman, Brodribb and Jordan 2010; Nardini, Pedà and La Rocca 2012a), that perform better in terms of resistance to cavitation events but might imply increased construction costs per unit surface area (high LMA) and relative to water use efficiency (low Kleaf_mass).
The biogeographic trend from small sized to large leaves when moving from xeric to mesic habitats (Givnish 1987) might indeed be related to the above-cited relationships between leaf size, major vein density and drought resistance (Sack & Scoffoni 2013). However, the correlations and trade-offs between leaf hydraulic capacity, drought resistance and plant species distribution in different biomes have been only occasionally investigated. It might be hypothesized that high Kleaf_area is adaptive in mesic/wet biomes, as increased photosynthetic productivity might allow plants to out-compete neighbours in such resource-rich habitats (Nardini & Tyree 1999). On the other hand, increased drought tolerance might be a better strategy in water-limited habitats, although this might imply decreased leaf hydraulic efficiency, possibly translating into lower potential and realized growth rates (Sinclair, Zwieniecki & Holbrook 2008). A recent study by Blackman, Brodribb and Jordan (2012) have highlighted the relationships between leaf hydraulic vulnerability and species’ bioclimatic limits, with special reference to average precipitation in different regions of Tasmania. However, to the best of our knowledge, a comprehensive analysis of relationships between leaf hydraulic efficiency, drought vulnerability of plant assemblages from different biomes and related climatic variables, is still missing. In this study, we present results of a literature survey and related data collection and analysis, aimed at elucidating the eventual correlations between leaf hydraulic traits and climate across three different biomes, that is dry sclerophyllous forests (DSF), temperate forests (TMF) and tropical forests (TRF).
Materials and methods
An extended literature survey of articles published between years 2000 and 2012 and dealing with leaf hydraulics was undertaken using the Scopus data base (www.scopus.com). Leaf hydraulic data were extracted and included in a data set. The criterion to include a study in the data set was the presence of at least one relevant parameter over those selected for further analysis (see below). Only studies reporting data for woody angiosperms were included. This survey resulted in the selection of 25 articles containing leaf hydraulics data for 130 woody angiosperm species, growing in different areas of the globe (see Table S1, Supporting information). Species’ names and relative families were gathered from each study, and species’ native habitat as well as description of sampling areas was used to classify different species as components of vegetation of one of three major biomes, that is DSF, TMF and TRF. Further subclassification within each of these groups was not considered as an option, because in this case, some sub-biomes would have been represented by <10 species.
Data relative to leaf hydraulic capacity and drought vulnerability, as well as to leaf morphological traits and gas exchange rates (when available), were extracted from each study. These included leaf hydraulic conductance scaled by leaf surface area (Kleaf_area, mmol s−1 MPa−1 m−2), leaf water potential inducing 50% loss of Kleaf (P50, MPa) and maximum leaf conductance to water vapour (gL, mmol m−2 s−1). When available, leaf water potential at turgor loss point (TLP, MPa) and leaf mass per area (LMA, g m−2) were gathered from the same studies. In some cases, TLP, LMA and gL values were obtained from independent studies carried out on the same species. On the basis of the above data, leaf hydraulic conductance per unit leaf dry mass (Kleaf_mass, mmol s−1 MPa−1 kg−1) was calculated as Kleaf_mass = Kleaf_area/LMA. The data set was analysed using a Rosner's Extreme Studentized test for multiple outliers, which identified four Kleaf_area values as out of range. The related species were excluded from further elaboration and analysis.
Experimental methods used in different studies to estimate leaf hydraulic capacity and vulnerability were rather heterogeneous, and included the evaporative flux method (Sack et al. 2002), the high-pressure and vacuum-pressure techniques (Salleo et al. 2003), the rehydration kinetics technique (Brodribb & Holbrook 2003) and finally the acoustic emission recordings from dehydrating leaves (Kikuta et al. 1997). The different hydraulic methods listed above have been reported to yield consistent and comparable results in terms of maximum leaf hydraulic capacity (Sack et al. 2002; Salleo et al. 2003; Scoffoni et al. 2008; Nardini et al. 2010). However, the consistency of leaf hydraulic vulnerability parameters estimated using different techniques has never been tested in depth, to the best of our knowledge, and this might represent a limitation of our data set. Another limitation of our analysis might arise from the fact that Kleaf data in different studies are sometimes referred only to lamina hydraulic conductance, while in other cases, petiole hydraulic conductance is included as well. However, petioles generally account for 10% or less of total leaf hydraulic resistance (Sack, Tyree & Holbrook 2005) so that related errors in analyses and correlations are expected to be small.
Coordinates for each study and sampling site were gathered from the articles, when available. When not available, coordinates were derived from Google Earth (http://www.google.com/earth) using names of localities indicated in each study. If these were not clearly specified, coordinates of the closest major town were acquired and attributed to the species’ entry. Geographical coordinates were then used to extract mean annual temperature (MAT) and mean annual precipitation (MAP) for each species/site using the CRU-TS Climate Data base, provided by the CGIAR Consortium for Spatial Information (http://www.cgiar-csi.org). All climatic data were based on the 1981–2000 reference period.
Mean values and SEM of each of the above-listed parameters were calculated for each biome. Because not all the selected parameters were available for each species, the n for different parameters was different as specified in each figure (see 'Results'). Statistical differences between biomes for different physiological variables were assessed using one-way anova, followed by post hoc pairwise comparisons based on the Tukey's test. Eventual significant correlations between different physiological and climatic parameters were tested using Pearson product-moment correlation.
The analysis of leaf hydraulic data of 130 woody angiosperms from three major biomes revealed important differences both in terms of water transport efficiency and drought vulnerability. Kleaf_area data were available for 113 species and showed a 10-fold variation from the lowest value of 2·3 mmol s−1 MPa−1 m−2 in Swartzia simplex (Fabaceae) to the highest record of 28·5 mmol s−1 MPa−1 m−2 reported for Hymenaea martiana (Fabaceae). Kleaf_mass values could be calculated for 83 woody angiosperms and showed an even more impressive variation, ranging from a minimum of 8·0 mmol s−1 MPa−1 kg−1 in Olearia pinifolia (Asteraceae) up to 277·5 mmol s−1 MPa−1 kg−1 in Aegiphila sellowiana (Verbenaceae). Large variability was also observed in terms of leaf hydraulic vulnerability across 84 woody angiosperms, with some very vulnerable species undergoing 50% loss of Kleaf at water potential values as high as −0·4 MPa (Magnolia grandiflora, Magnoliaceae), and extremely drought resistant ones maintaining more than 50% residual Kleaf down to water potential values of −5·2 MPa (Arbutus menziesii, Ericaceae).
When values for the selected parameters were averaged for group of species from different regions of the globe, clear differences emerged between biomes (Fig. 1). In particular, Kleaf_area was similar in DSF and TMF (7·2 ± 1·0 and 8·4 ± 0·7 mmol s−1 MPa−1 m−2, respectively) but significantly higher in TRF (14·3 ± 1·3 mmol s−1 MPa−1 m−2). A similar trend was observed for Kleaf_mass, although in this case, the magnitude of differences was larger, with DSF and TMF biomes averaging values of 61·3 ± 10·2 and 70·1 ± 11·1 mmol s−1 MPa−1 kg−1. Again, the highest values for leaf hydraulic conductance, even on a dry mass basis, were recorded for TRF (160·1 ±17·3 mmol s−1 MPa−1 kg−1). The lowest P50 values, indicating higher resistance to drought-induced hydraulic failure, were recorded for DSF plants (−2·44 ± 0·20 MPa). Progressively higher P50 values were recorded for TMF and TRF (−1·98 ± 0·18 and −1·54 ± 0·08 MPa, respectively), all these differences between biomes being statistically significant.
The assemblage of species arising from our literature survey apparently comprised species thriving under very different climatic conditions, with MAP ranging from <400 mm up to about 3000 mm, and MAT spanning from 7 to 28 °C. Significant correlations were observed between leaf hydraulic parameters and climatic variables describing the habitat of origin/sampling of different species (Fig. 2). In particular, P50 was correlated to both MAP (although P50 values were not available for species growing over MAP >1600 mm) and MAT, with more negative values occurring in species from dry areas, and relatively more drought vulnerable species prevailing under mesic/wet as well as warmer conditions. Both Kleaf_area and Kleaf_mass were positively correlated to both MAP and MAT so that leaves with highest hydraulic efficiency occurred in mesic/wet and warm sites, while dry and/or cool regions displayed species assemblages characterized by lower leaf hydraulic capacity. Significant relationships emerged between LMA and MAP (r2 = 0·07, P < 0·05) as well as between TLP and MAP (r2 = 0·13, P < 0·01), while neither LMA nor TLP were correlated to MAT.
Data of maximum leaf conductance to water vapour (gL) were available only for <50% of the species selected in the literature survey. Nonetheless, significant positive correlations were observed between gL and both Kleaf_area (Fig. 3a) and Kleaf_mass (r2 = 0·34, P < 0·01, data not shown). P50 was found to be significantly correlated to TLP (Fig. 3b), but not to LMA (P = 0·08). In turn, LMA was correlated to Kleaf_mass (r2 = 0·28, P < 0·001) but not to Kleaf_area (P = 0·08).
Although no correlation was found between P50 and leaf hydraulic capacity on a leaf area basis (Fig. 4a), a weak but significant relationship emerged when P50 was plotted versus Kleaf_mass (Fig. 4b).
Leaf hydraulic capacity and drought vulnerability have been frequently reported to represent key physiological traits driving plant adaptation to habitats characterized by different levels of water and light availability (Sack, Tyree & Holbrook 2005; Nardini, Pedà & La Rocca 2012a). Despite substantial species-specific heterogeneity within each biome, probably due to the important role played by niche-specific micro-climatic conditions as drivers of leaf hydraulic adaptation/acclimation (Lo Gullo et al. 2010), the results of our analysis revealed important differences in terms of leaf hydraulic capacity and vulnerability between three major biomes.
Tropical forest plants displayed the highest leaf hydraulic capacity, both on a leaf area and dry mass basis. Hence, not only the leaf photosynthetic surface unit is best supplied with water in plants inhabiting the tropical biome, but also carbon costs incurred by these plants to obtain high hydraulic efficiency are relatively low when compared to plants from temperate and semi-arid biomes. On the contrary, P50 was relatively less negative in temperate and tropical forest species when compared to DSF, which also displayed the lowest Kleaf_mass.
High plant hydraulic conductance has been frequently interpreted as an adaptive functional trait in habitats with good water availability (Nardini & Tyree 1999; Zhang & Cao 2009; Manzoni et al. 2013). Because leaves represent a major limitation of plant hydraulics, it is not surprising that such a relationship was also confirmed when comparing Kleaf of species from habitats with contrasting precipitation (Blackman, Brodribb & Jordan 2012). High Kleaf is thought to confer competitive advantage to plants by allowing larger gas exchange rates and, hence, CO2 fixation, without increasing the risk of fatal desiccation of tissues (Meinzer 2002; Brodribb, Feild & Jordan 2007). The frequently reported relationship between maximum leaf conductance to water vapour and Kleaf_area is an expression of this functional advantage and was also confirmed by our analysis across 52 angiosperms from different habitats (Fig. 3). On the other hand, the ability to survive water stress by maintaining the functionality of the leaf hydraulic system might be more important for plants thriving in drought-prone biomes, than the achievement of high gas exchange rates, unless drought deciduous strategies are adopted. This view is supported by relatively lower P50 and Kleaf values in plant species native to DSF. The analysis of correlations between leaf hydraulic parameters and climatic data confirm the above hypotheses (Fig. 2), as significant relationships were observed between P50, Kleaf and two major descriptors of habitats’ climatic traits, that is MAP and MAT.
Overall, data of leaf hydraulic vulnerability at the biome level are consistent with those reported for stems by Maherali, Pockman and Jackson (2004) and Choat et al. (2012). At the stem level, Mediterranean woody plants were found to be significantly more resistant to xylem cavitation than both temperate and tropical forest woody species, with P50 values averaging −5·0 MPa in DSF and about −2·5 to −3·0 MPa in the other two biomes (Maherali, Pockman & Jackson 2004). The comparison between the leaf hydraulics data set presented in this study and previous meta-analyses carried out for stems confirms, at a biome level, two important facts. First of all, leaves tend to be more vulnerable to drought-induced hydraulic dysfunction than stems and hence may act as safety hydraulic fuses to prevent catastrophic hydraulic failure at the stem level during intense and or prolonged drought (Bucci et al. 2012). Secondly, the parallel variation of P50 values across biomes at the stem and leaf level confirm that leaf and stem hydraulic traits are generally coordinated to each other (Meinzer et al. 2008). Moreover, the close correlation between P50 and TLP found in our analysis suggests that apoplastic and symplastic drought resistance are strictly coordinated to each other, as recently shown by other studies (Blackman, Brodribb & Jordan 2010; Vilagrosa et al. 2010). The correlation between these parameters suggests that biomes characterized by low MAP are inhabited by species with low values of both P50 and TLP, in accordance with a recent meta-analysis by Bartlett, Scoffoni and Sack (2012).
Previous studies have revealed the existence of an efficiency/safety trade-off at the stem level, in that high stem specific hydraulic conductivity was frequently found to be associated with relatively higher vulnerability to xylem cavitation and vice versa (e.g. Bucci et al. 2006; Jacobsen et al. 2007; Markesteijn et al. 2011). Increased resistance to xylem cavitation at the stem level has been frequently reported to be associated with high wood density (Meinzer et al. 2009; Hoffmann et al. 2011; Nardini, Battistuzzo & Savi 2013), as a likely consequence of thick vessel walls and/or presence of abundant mechanical tissues, which are both thought to protect the xylem from the risk of implosion under very negative pressures (Hacke et al. 2001; Jacobsen et al. 2005). Because the construction of high-density wood is metabolically expensive, the above-cited correlations and trade-offs between stem hydraulic efficiency, safety and density further suggest that enhanced drought resistance implies substantial carbon investment by plants that might significantly reduce the amount of carbon available for growth. This observation might indeed provide a functional explanation for the frequently reported trade-off between plant drought resistance and relative growth rate (Polley et al. 2002; Wikberg & Ögren 2004).
Similar relationships have been sometimes suggested to hold true also at the leaf level. As an example, low P50 values have been associated to thick xylem conduit walls (Blackman, Brodribb & Jordan 2010), high major vein density (Scoffoni et al. 2011) and high xylem conduit density at the vein level (Nardini, Pedà & La Rocca 2012a). However, attempts at relating Kleaf_area to leaf hydraulic vulnerability have never revealed clear trade-offs between these traits (Blackman, Brodribb & Jordan 2010; Bucci et al. 2012). Similarly, in our extended survey, we found only a weak, not significant correlation between P50 and Kleaf_area (Fig. 4a).
Tyree, Velez and Dalling (1998) were apparently among the first to suggest the validity and informative value of normalization of shoot and root hydraulic conductance by respective organ dry weight, to obtain estimates of ‘carbon efficiency’ of plant hydraulic systems. This alternative scaling procedure allowed Tyree, Velez and Dalling (1998) to highlight the fact that pioneer tropical species spend less carbon to build efficient hydraulic pathways than late-successional species, making them more competitive in forest gaps. However, examples of normalization of leaf hydraulic data by dry mass have not been available in the literature until very recently. Simonin, Limm and Dawson (2012) have shown that LMA and leaf life span correlate to Kleaf_mass but not to Kleaf_area, leading to the suggestion that Kleaf_mass is an important index of the carbon cost associated with contrasting water use strategies across different species. In a comparison of sun versus shade leaves of Quercus ilex L., Nardini, Pedà and Salleo (2012b) have shown that Kleaf_mass was about 40% higher in the former leaf type than in the latter, while Kleaf_area was found to change by more than twofold. These results suggest that hydraulic acclimation of leaves to contrasting light intensity within a crown enhances water transport per unit leaf surface area while maintaining relatively constant the carbon costs of the leaf water transport system. Scaling Kleaf by leaf dry mass also allowed Nardini, Pedà and La Rocca (2012a) to highlight significant trade-offs between P50 and carbon efficiency of the leaf hydraulic system across six species of the genera Acer and Quercus, raising questions about the eventual general validity of this relationship across larger species assemblages.
The validity of the use of mass-based metrics to describe leaf functional traits and their correlations has been recently questioned (Lloyd et al. 2013), starting from the assumption that the leaf is a photosynthetic organ and that photosynthesis is naturally an area-based process. This argument is valid if the leaf is considered as a purely planar object intercepting light and losing water, as a consequence of the need to absorb CO2. In reality, the leaf hydraulic system is designed to supply water to the bulk of mesophyll photosynthetic cells and to deliver a profitable return to the plant when compared to the construction costs (Westboy, Reich & Wright 2013). In this view, while area-based metrics are certainly meaningful in a physiological context, mass-based normalization offers additional opportunities and insights into an ecological context, as in the present study.
In the present analysis, a significant correlation was found between P50 and Kleaf_mass, with species displaying the most drought-resistant foliage also tendentially incurring in greater costs per unit hydraulic efficiency, in contrast with species characterized by higher carbon efficiency of their hydraulic systems and less negative P50 values. While it must be stressed that the very low r2 value (0·10) of this relationship reveals that only about 10% of the variation of P50 might be actually clearly linked to Kleaf_mass, our finding suggests the existence of a possible, although weak, ‘global’ trade-off between hydraulic capacity and safety of leaves across biomes, that might indeed represent one of the mechanistic links between ecological and physiological traits recognized as distinctive of species inhabiting contrasting climatic zones.
Assuming that lower Kleaf_mass is associated with high LMA (Simonin, Limm & Dawson 2012) and that high LMA is also driven by some leaf vein traits (Castro-Díez, Puyravaud & Cornelissen 2000; Méndez-Alonzo, Ewers & Sack 2013), it might be speculated that low values of both P50 and Kleaf_mass arise as a common result of morpho-anatomical traits known to confer drought tolerance to the leaves, with special reference to leaf size, major vein density and vein conduit size and density. In fact, high major vein density contributes, at least partially, to increasing LMA (Méndez-Alonzo, Ewers & Sack 2013). However, a detailed analysis and discussion of the drivers of LMA across biomes is outside the scope of this study, and readers are referred to Sack et al. (2013) for a recent and comprehensive review of this important topic. Leaf size is known to be correlated to several climatic variables, and the biogeographical trend towards smaller leaves from high-rainfall to drought-prone climatic zones is probably one of the earliest recognized and described patterns in vegetation science (McDonald et al. 2003; and references therein). On the contrary, venation density is known to increase from mesic/wet to xeric habitats (Uhl & Mosbrugger 1999; Sack & Scoffoni 2013), and recent studies linking water availability, leaf expansion rates, cell size and final venation density suggest the existence of adaptive links between leaf size and vein density (Zhang et al. 2011; Carins Murphy, Jordan & Brodribb 2012; Sack et al. 2012; Brodribb, Jordan & Carpenter 2013).
Further studies are called to confirm and clarify the structural/functional relationships underlying the relationship between leaf drought vulnerability and hydraulic capacity on a dry mass basis, as well as the ecological role played by variation of these traits in shaping vegetation at different scales, from micro-climatic niches up to extended biomes.
We are grateful to the Associate Editor and anonymous reviewers for very helpful comments to previous versions of the manuscript.