Pedunculate oak ( Quercus robur L.) decline has been observed in Europe during the last few decades, causing not only economic losses, but also losses of biodiversity and forest genetic resources ( Čater, 2015 ). Recently, a number of factors have been identified that contribute to Q. robur decline; e.g. Thomas et al. (2002) listed the most important stress elicitators in the oak decline complex, grouping them roughly into biotic (insect larvae, bark beetles, pathogenic fungi, and microorganisms) and abiotic (air pollutants, climatic extremes and site conditions). Although the causes of oak decline are complex and probably involve diverse direct and indirect factors ( Drobyshev et al., 2007 ), drought stress has been frequently reported as the main driver of oak dieback across Europe. Indeed, Andersson et al. (2011) found that the extreme drought in southern Sweden in 1992 was the triggering factor of the oak mortality in the following years. Cochard et al. (1992) reported serious oak dieback after the severe drought that occurred in France in 1976. Furthermore, Čater (2015) evidenced increased mortality rates of oak in Slovenia after remarkedly dry periods in 2003 and 2013. Similarly, the dieback of oak trees is widely present in Serbia, as well, as the consequence of the long-term decline of the Danube and the Sava rivers' water levels ( Stojnić et al., 2014 ). This issue is not only evident among mature stands, but it is also present at regeneration sites and is reflected in failed reforestation attempts, especially during dry years ( Čater & Batič, 2006 ; Stojanović et al., 2015 ). Furthermore, climate change is predicted to negatively influence pedunculate oak ecosystems and sustainable wood production ( Stojanović et al., 2014 ; Liović et al., 2019 ). Bearing this in mind, it appears that conventional breeding strategies based on selection of dominant stem quality and vigorous growth rate will have to be improved in order to address increasing needs in a rapidly changing environment ( Verryn, 2008 ).

The resistance of plants to low water availability is the consequence of four possible strategies: drought escape, drought avoidance, drought tolerance and drought recovery. Among these strategies drought avoidance and drought tolerance have been noted as major mechanisms, utilized by plant species grown under water deficit (arid) conditions ( Fang & Xiong, 2015 ). Regarding drought avoidance, water-use efficiency ( WUE ) is one of the key components responsible for maintaining favorable tissue water content thro-ugh morphological adjustments or physiological me-chanisms ( Farooq et al., 2009 ; Kooyers, 2015 ). The-refore, water-use efficiency might be considered as one of the principal components of drought adaptation in plant species. Water-use efficiency can be defined from different perspectives, depending on space scale of measurement and units of exchange being used for WUE determination ( Condon et al., 2004 ). Moreover, it might be observed at a given point of time or over a prolonged period ( Yul Yoo et al., 2009 ). From the physiological perspective particularly interesting is intrinsic water-use efficiency of leaf gas exchange ( WUE _i ), which can be defined as the ratio of net photosynthetic rate to stomatal conductance for water vapour. In general, as drought increases, simultaneous decline in net CO ₂ assimilation rate is more evident than the reduction of stomatal conductance, leading to an increase in WUE _i ( Pita et al., 2005 ; Bojović et al., 2017 ). Although the theoretical background behind this process is well-known, the effects of varying traits on WUE _i have not yet been experimentally tested in many tree species. Due to very tight relationship between leaf structure and function (i.e. leaf morphological and anatomical structure and physiological processes), the relationship between WUE _i and other traits is rather complex and requires a deeper analysis in order to identify those which determine WUE _i to the largest extent. For example, several studies have demonstrated a coupled response of net photosynthesis and stomatal conductance of CO ₂ to environmental signals ( Raftoyannis et al., 2006 ; Gallé & Feller, 2007 ). Along this line, stomatal morphology and density have also been shown to contribute to the diversity of stomatal conductance, which might influence WUE _i due to an aforementioned close co-rrelation between net CO ₂ assimilation rate and stomatal conductance ( Franks & Farquhar, 2007 ). Similarly, a study conducted on Fagus crenata Blume. provenances showed a significant correlation between leaf thickness on one side, and maximum photosynthetic rate and stomatal conductance on the other ( Bayramzadeh et al., 2008 ). Furthermore, under drought conditions the imbalance between the production of different reactive oxygen species (ROS) and scavenging capacity of antioxidant defense in the cell can lead to drought induced secondary stress, a phenomenon known as oxidative stress ( Avramova et al., 2017 ).

Targeting specific functional traits to improve WUE _i , as well as understanding the relative contribution of these characteristics to it should be one of the principal goals in the genetic improvement of Q. robur. Therefore, the objective of the present research was to assess the relevance of various functional traits as indicators of WUE _i . We hypothesized that application of various multivariate statistical methods (i.e. principal com-ponent analysis, stepwise regression analysis and path coefficient analysis) will enable the identification of leaf functional traits that could serve as indicators of Q. robur adaptation to drought, as well as for evaluation of the nature and magnitude of the relationship between physiological, biochemical and morpho-anatomical characteristics, on one side, and WUE _i , on the other.

We found significant difference in WUE _i between Q. robur half-sib lines, although the results evi-denced a narrow range of WUE _i variation (7.8 μmol mol ^-1 ). The highest mean value of WUE _i was observed in half-sib line B05, whereas the lowest was recorded in half-sib line B17. In addition, the analysis of variance showed that most of the studied traits varied significantly between Q. robur half-sib lines. The highest variability was observed for physiological traits (except for chlorophyll content), while among biochemical and morphoanatomical characteristics, significant differences were observed only for certain parameters ( TPC , TFC , TSP , RSC DPPH , LMA , SD and LS ) (Table 2). Based on the results of principal component analysis, the first five principal components (PCs), accounting for 95.7% of the total variance, were selected (Table 3).

The first three principal components explained 71% of total variation among the studied half-sib lines; e.g. PC1 accounted for 20.2% of total variance, whereas PC2 and PC3 accounted for 26% and 24.7%, respectively. The rest of the components (PC4 and PC5) varied to a lesser extent (together: 23.8% of total va-riance). Similar contribution of the selected principal components to the total variance suggests similar con-tribution of analogue groups of characteristics to discrimination of the exmined half-sib lines. The highest correlation with PC1 corresponded to WUE _i (0.920), and those variables related to leaf anatomical traits (i.e. stomatal size and density): SD (0.825), LS (-0.818) and WS (-0.641) (Table 3). Variables with the highest loadings with PC2 were mostly biochemical parameters: RSC NO (0.916), RSC ABTS (0.923), TFC (0.844), TSP (0.731) and TPC (0.679). A relatively high loading with PC2 was also determined for the FRAP (0.477). The variables with the highest scores on PC3 were the gas exchange parameters C _i (0.925) and E (0.915), as well as LMA (0.869). The second highest loading of WUE _i was with PC3 (0.299), therefore PC1 and PC3 were selected for the visual presentation of relationships of examined traits with WUE _i (Fig. 2). WUE _i and six other traits were the closest to the circle, meaning that most of their variation was explained by PC1 and PC3. Likewise, the angles between vectors of WUE _i and LMA , C _i and E were more orthogonal than the angles between WUE _i and SD , LS and WS , confirming the results of grouping based on loadings with principal components presented in Table 3.

Stepwise regression analysis was used to determine the traits contributing for the majority of total half-sib-lines-level WUE _i variability. The results revealed that almost the entire variation in leaf WUE _i was attri-buted to six variables ( SD , LMA , FRAP , Chl , WS and C _i ) (Table 4). Stepwise regression model with SD and LMA achieved high coefficient of determination (80.4%), which highlights the relevance of these traits in the explanation of total leaf WUE _i variation. Since these two traits belong to different PCA groups, it demonstrates the ability of this statistical method to select properties that best describe the dependent variable, with a minimum of collinearity between them.

According to path coefficient analysis performed on traits selected by stepwise regression analysis, stan-dardized path coefficients for all selected traits were significantly different from zero. Thus, it could be considered that direct effect of every selected variable on WUE _i was statistically significant (Table 5). Among these traits, SD and LMA achieved the highest direct effects on WUE _i , but no significant indirect effect through any other selected trait. Other parameters, selected by stepwise regression analysis, had both significant direct and indirect effects through SD and LMA . The complex linkage on dominant direct and indirect effects of the examined traits on leaf-level WUE _i of Q. robur half-sib lines is given in the Table 6.

Considering the negative effects of drought on Q. robur forests in Europe, as well as the projected scenarios concerning the impact of climate change on this species, it might be concluded that the plant's economy of water consumption will be of primary importance for oak survival. For this reason, there is a widespread opinion that the adaptive capacity of trees toward drought should take priority over high timber quality in order to preserve the diversity of forest ge-netic resources ( Kelleher et al., 2015 ). Since water-use efficiency has been seen as the most important com-ponent of drought adaptation in plants ( de Almeida Silva et al., 2012 ), certain authors believe that breeding of plants characterized by high water-use efficiency is one of the key objectives of sustainable, climate-smart forestry ( Flexas et al., 2013 ). Indeed, the trees with higher water-use efficiency show a specific water-saving strategy that allows them to avoid large water losses and/or maintain higher assimilation rates than trees with lower water-use efficiency, which could have an impact on their competition for growth and survival, especially in drought prone environments ( Mészáros et al., 2007 ; Brendel et al., 2008 ).

The presence of statistically significant differences between Q. robur half-sib lines in terms of majority of examined leaf functional traits is presumably the result of their divergent genetic make-up. In general, leaf physiological traits showed the highest variability be-tween half-sib lines, with only Chl showing no statistical differences between them. A possible explanation of this phenomenon might be a low heritability of leaf chlorophyll content, as already documented by Roussel et al. (2009) for Q. robur genotypes.

The results further showed that although the majority of examined characteristics did not achieve important direct effect on WUE _i , the valuable information of the relationship between morpho-anatomical traits and WUE _i was observed by PCA. We evidenced that all of the examined leaf stomatal traits were in the same PCA group (PC3) with WUE _i , suggesting that these characteristics can be good indicators to be used in the selection for water-use efficient Q. robur genotypes. Moreover, according to stepwise regression analysis and path coefficient analysis, stomatal density achieved the strongest direct effect on variation of WUE _i . Stomata size and density have a key role in the regulation of gas exchange between the leaves and the atmosphere. Controlling water flux inside plants, stomata also mo-dulate a balance between water loss and CO ₂ uptake, thus determining water-use efficiency ( Holland & Richardson, 2009 ). Previous studies have demonstrated that stomatal conductance largely depends on stomatal aperture characteristics, which in turn determine ma-ximum theoretical conductance, as well as the speed of stomata response to environmental signals ( Dow et al., 2014 ). Aasamaa et al. (2001) believe that length of the stomatal pore plays the most important role in the determination of stomatal conductance in trees grown under different water supplies. Similar results (i.e. sto-matal length was positively correlated with g _s ) were also reported by Ohsumi et al. (2007) . These results have been confirmed by our study as well, since a positive correlation between g _s and LS (data not shown) was observed. However, guard cell size also affects the speed of stomata response to environmental factors. Hetherington & Woodward (2003) stated that larger stomata often exhibit slower responses in comparison to the smaller ones, showing a greater potential for hydraulic dysfunction under water stress. Likewise, plants characterized with higher g _s generally exhibit lo-wer water-use efficiency ( Lawson & Blatt, 2014 ). In the present study we observed a negative correlation between WUE _i and LS , and positive between WUE _i and SD , suggesting that selection for water-use efficient half-sib lines should be oriented toward those with more numerous and smaller sized stomata. Such approach would be in accordance with the results of Aasamaa et al. (2001) , who demonstrated a clear negative correlation pattern between the length of the stomatal pore and sensitivity to increasing drought in six forest tree species.

Although stomatal size and density define stomatal conductance, stomatal density is negatively correlated with stomatal length, in general ( Pearce et al., 2006 ; Pyakurel & Wang, 2014 ), that has also been confirmed by our research. Concerning relationship between SD and g _s , the opposite results have been reported. While Pearce et al., (2006) reported that stomatal density is positively correlated with g _s in riparian poplar species, a number of experiments showed that the relationship between stomatal density and stomatal conductance does not seem to be straightforward ( Jones, 1997 ; Ilgin & Caglar, 2009 ; Lawson & Blatt, 2014 ). Likewise, in contrast to stomata length, recent studies have de-monstrated high sensitivity of stomatal density to different climate and environmental conditions ( Stojnić et al., 2015a , b ). According to Aasamaa et al., (2001) smaller stomata respond faster to environmental signals than larger stomata, so the combination of smaller sto-mata and high stomatal density might provide greater plant capacity to rapidly adjust stomatal conductance and optimize gas exchange.

Besides the notable influence of SD on WUE _i , the second strongest direct effect on the variation of WUE _i was achieved by LMA , as demonstrated by the path coefficient analysis. In contrast, LA was grouped in PC5 group, apart from others, indicating poor effect on WUE _i variability. Recent studies have evidenced that LMA is linked with WUE _i mainly through the tight correlation with both A and g _s ( Hassiotou et al., 2010 ). Indeed, structural leaf changes related to variation in LMA usually correspond to alterations in photosynthetic capacity, thereby affecting WUE _i . High LMA is related to the formation of thicker leaves and dense mesophyll tissues as the consequence of large number of tightly packed cells with more lignified cell walls ( Niinemets, 2001 ). Such leaf structure implies changes in maximum photosynthetic rate as the result of longer paths for water diffusion from stomata to chloroplast, as well as higher chloroplast shading in the interior of the leaf ( Novriyanti et al., 2012 ; Read et al., 2014 ). Due to increased water diffusion resistance in mesophyll as the consequence of enhanced water path through the intercellular spaces to stomata, high LMA is frequently associated with reduced CO ₂ diffusion, and therefore, a lower g _s ( Muir et al., 2017 ). Nevertheless, despite low stomatal conductance values, a high LMA might positively affect both water-use efficiency and photosynthetic capacity in plants growing under stress conditions, due to the increased number of chloroplasts in the mesophyll tissues ( Bresson et al., 2011 ; Stojnić et al., 2015b ). In addition, compact mesophyll tissue (i.e. high LMA ) is also shown to cause greater leaf resistance to the water diffusion in the process of transpiration ( Novriyanti et al., 2012 ). This study also confirms these results, since we found that LMA and E were highly negatively correlated, having their highest loadings on PC3 (0.869 and -0.915, respectively).

Since most of the studies were focused to explain the effects of drought on either metabolic and tran-scriptomic profiles or specific enzyme activities in different species ( Mittler et al., 2006 ), this paper kept track on the relationship between WUE _i and radical scavenger activity assessed by the biochemical assays. Recently, several studies have evidenced that bio-chemical biomarkers might be efficiently used as the indicators of environmental stress effects on the physiology of seedlings ( Popović M et al., 2016 ; Schiop et al., 2015 ). Likewise, Popović B et al. (2016) reported increased biosynthesis of specific phenolic compounds, such as myricetin, chrysin, kaempferol and isoferulic acid etc., as well as increased activity of biosynthetic enzyme (PAL-phenylalanin ammonium lyase) under induced drought stress in poplar clones. Finally, Štajner et al. (2013) demonstrated increased percentage of ra-dical scavenger activity against DPPH radical and increased FRAP value under drought conditions during summer months in beech species, as well as in different melliferous tree species, such as Fraxinus sp. and Ro-binia pseudoacacia ( Štajner et al., 2011 ). However, our results evidenced that the examined biochemical characteristics, presumably related to tolerance of oxi-dative stress, did not have considerable significance in conditions of moderate drought. Indeed, only FRAP was selected by stepwise regression analysis and, acco-rding to the results of path coefficient analysis, showed significant effect on the WUE _i , both directly and in-directly through SD and LMA .

Several studies have evidenced that under moderate drought conditions carbon assimilation in plants de-cline mostly due to stomatal closure ( Flexas & Medrano 2002 ; Stojnić et al., 2016 ). In contrast, under conditions of low water availability photosynthetic rate is also co-limited by non-stomatal features including electron transport rate and photochemistry ( Petridis et al., 2012 ). Severe drought may lead to imbalance between antioxidant defenses and the increased level of reactive oxygen species (ROS), therefore causing an oxidative stress in plants ( Štajner et al., 2011 ). Namely, the decline in the internal CO ₂ ( C _i ) caused by ABA-induced stomatal closure during severe drought, and therefore decreasing CO ₂ availability for photosynthesis ( Chaves et al., 2003 ; Sánchez-Rodríguez et al., 2010 ), prevent regeneration of NADP ⁺ (i.e. main electron acceptor) by the Calvin cycle leading to enhanced electron leakage to molecular oxygen in chloroplasts, which, as a consequence, has an increased generation of different ROS ( Nahar et al. 2018 ). Under stress conditions, the absorbed light energy may not be entirely used for photosynthesis, therefore causing its repression - photoinhibition ( Petridis et al., 2012 ). To avoid such damage, plants tend to increase their antioxidant capacity by producing increased amo-unts of various compounds with antioxidant properties in order to suppress oxidative stress. Increased antioxidant and reducing capacity ( FRAP value) of extracts in sense of increased concentration of compounds with high antioxidant properties could have preventing and mi-tigating effects towards these oxidative injuries of PSII.

On the other hand, grouping of the examined traits by principal component analysis suggested that none of the examined biochemical characteristics, as well as chlorophyll content were grouped by PC1 or PC3, which had the highest loadings with WUE _i . Most of them were allocated in PC2 group, while FRAP and RSC DPPH were together with Chl in the PC4 group. Similar results have been reported by Ghasemzadeh & Jaafar (2011) , who observed lack of correlation between water-use efficiency, on one side, and total phenolics and flavonoids, on the other. It seems that drought conditions at the timepoint of sampling were still mild enough to trigger oxidative stress and cause major metabolic disturbances, which could affect anti-oxidant properties of the whole leaf extract, so these correlations were neglectable.

The results of our study demonstrated that under moderate water stress WUE _i was mainly the result of the plants' structural acclimation to surrounding en-vironmental conditions. We found that in such con-ditions stomatal density and leaf dry mass per unit of leaf area have achieved significant and dominant direct effect on WUE _i in Q. robur half-sib lines. Stomatal density achieved the highest score on PC1, in which WUE _i had the highest loading. Likewise, SD was the first to be included in the stepwise regression model, having the highest direct effect on WUE _i and, therefore, indicating that stomatal regulation could be considered as a dominant contributor to the regulation of WUE _i in the conditions of moderate drought. In addition, significant direct effect of LMA on WUE _i suggests that leaf structure strongly affected the water diffusion inside mesophyll tissues and related physiological processes of leaf gas exchange.

On the other hand, none of the examined biochemical traits were grouped by any of the principal components that had the highest loadings with WUE _i , suggesting their poor effect on WUE _i in conditions of moderate drought. Only FRAP was selected by stepwise regre-ssion analysis, achieving poor direct effect on variation of WUE _i .

Taking into account the foregoing discussion, we be-lieve that selection based on identification and quan-tification of the genetic correlations between WUE _i and structural leaf traits should have a vital place in future breeding programmes oriented towards improvement of WUE _i , especially in the conditions of moderate water deficit. Therefore, high intrinsic water-use efficiency of Q. robur species can possibly be achieved by indirect selection of breeding material via SD and LMA , the traits that have the highest direct effect on WUE _i . Similar approaches have already been recommended by other authors for certain crop species ( Hui et al., 2008 ; Golparvar & Karimi, 2012 ; Wu & Bao, 2012 ).

We sincerely appreciate Mr. Ed Bauer from USDA Forest Service Northern Research Station, Rhinelander, WI, for editing the manuscript for English language and style.