We sampled 86 individuals developing on three locations. In the case of E. grandis , we analyzed 4 clones growing in Concordia (Entre Ríos province, Argentina (31°22’ S; 58° 07’ W; 43 m asl) that belong to the network of genetic improvement field trials of INTA (National Institute for Agricultural Technology). More details about these clones can be found in Tesón et al . (2011) and Tesón et al . (2012) . Data from E. viminalis also belong to individuals from INTA’s network of genetic improvement field trials, growing in Guaminí (35° 52'S; 60° 43'W; 265 m.a.s.l.) and Del Valle (37° 6'S; 62° 26'W; 297 m.a.s.l.), Buenos Aires province, Argentina. The E. globulus trees are individuals with two seed origins, local and Valdivian, grown in a plantation in Necochea (Buenos Aires province, 38º 57´S; 58º 57´W; 17 m.a.s.l.), which were sampled as part of another research project whose results have only been partially published ( Igartúa & Monteoliva, 2010 ).

We collected one sample per tree from 30 E. viminalis, 36 E. grandis and 20 E. globulus . Sapwood from E. grandis and E. viminalis was sampled with an increment borer at 1.3 m above ground level along the north-south radius, while samples from E. globulus were obtained by destructive sampling (slices) at the same height. E. grandis and E. viminalis individuals were 14 years old, whereas E. globulus were 11 years old (n=14) and 15 years old (n=6).

Cross sections (20 µm) were cut from the last two growing periods using a sliding microtome. Sections were then stained for 5 minutes in safranin (1%), dehydrated in a graded alcohol series, and mounted in Entellan® for microscopic analysis. Images of the stained cross-sections were captured with a digital camera (Olympus DP71, Japan) mounted on a research microscope (Olympus BX50, Japan), using either 4x or 40x objectives.

Image analysis software (ImagePro Plus, v. 6.3, Media Cybernetics, USA) was used to analyze captured images for the following parameters : fiber wall thickness (Fwt, µm, n = 100 per sample), fiber diameter (Fd, µm, n = 100 per sample), fiber lumen diameter (Fld, µm, n = 100 per sample), vessel lumen diameter (Vd, µm, n = 100 per sample), mean vessel lumen area (A, mm ² , n = 100 per sample), and vessel density (N, mm ^-2 , n = 5 captured images per sample). Double fiber wall thickness was measured from radial and tangential walls and their average divided by 2. Fiber lumen diameter was measured in radial and tangential directions. Vessel lumen diameter and area was measured in the transverse section and the number of vessels in each 1 mm ² field was counted to determine vessel density. Basic density (D, kg·m ^-3 ) was obtained as the ratio of dry weight to saturated volume.

From the initial variables mentioned above, the following were calculated:

- Lumen fraction (F, unitless), which is a measure of the relative amount of sapwood area available for water transport. It was calculated from mean vessel lumen area (A) and vessel density (N) for each individual sample.

- Vessel composition (also known as “vulnerability index”, S, mm ⁴ ), proposed as a measure of vessel size distribution within the full space occupied by vessels, and calculated as the ratio between mean vessel lumen area (A) and vessel density (N).

- Theoretical xylem-specific hydraulic conductivity ( Ks , kg·m ^-1 ·MPa ^-1 ·s ^-1 ), which was estimated by the HagenPoiseuille law, as follows:

One-way ANOVA was used to test for differences in selected variables between species. Differences were further analyzed with Tukey multiple pairwise comparison test (a=0.05). Contingency tables were used to establish differences among species in vessel size distribution. Shifts in vessel characteristics were analyzed through variance of the four selected vessel variables (A, N, F and S), at a general inter-specific level as well as for each species. Based on the results, we examined the level of variation that S and F presented with respect to the variation showed by the other two variables, as follows:

From a functional standpoint, not only does the average size of wood characters matter, but also its variation. For example, a few large vessels (with small relative weight in the arithmetic mean) may be responsible for a significant increase in xylem water transport capacity, as was observed in E. grandis ( Table 1 , Fig. 2 ). At the genus level, S was the feature that showed the greatest variation (coefficient of variation (CV): 69.6%), followed by N, A, and F (CV: 55.6%, 48.1% and 33.8%, respectively). The same trend was observed for E. grandis , while in E. globulus and E. viminalis the order changed, with S remaining as the most variable feature, followed by F, A and N (data not shown).

In general (in all three species), vessel composition (S) explained 87.3% of total variation described by A and N, compared to only 12.7% accounted for by changes in lumen fraction (F). In turn, percentages also varied between species although similar trends were observed. At one extreme, in E. grandis, S explained 95.5% of total variation, while in E. globulus it only accounted for 65.6% of total variation.

Another indicator expressing the great variation of S compared to F was the maximum-minimum ratio. At general level, S varied 50-fold while F showed a 6-fold variation. While this trend was also recorded within species, it is noteworthy that E. grandis differed from the other two species, with a 50-fold and 3-fold variation in S and F, respectively. On the other hand, both E. globulus and E. viminalis exhibited an 8-fold variation in S, while F varied 5-fold and 3-fold, respectively.

Differences between species were obtained in parameters ß and a of the model relating theoretical Ks to F and S ( Table 2 ). This model showed an adequate fit for all analyzed species, and the parameters used were highly significant. The model applied to the pooled data of the three species resulted in the following parameters values: ß= 1.5 and a=0.5.

Differences found in ß for E. viminalis, E. globulus and E. grandis represent approximately a 21- and 41-fold variation in Ks for each 10-fold increase in F, respectively ( Fig. 3a ). The d values found are similar for E. globulus and E. grandis , and a little lower for E. viminalis . They produce a 2.7 and 2.6-fold variation in Ks for a 10-fold increase in S, respectively ( Fig. 3b ). This indicates that for the three species under consideration, the hydraulic conductivity is more sensitive to changes in lumen fraction (F) than to changes in vessel composition (S).

This can also be seen in Fig. 4 , where F and S were plotted against Ks showing that the variation in F cuts the conductivity isolines more perpendicularly, while S does so more obliquely. The greatest general variance of S compared to F results in a greater contribution of the former to the variation between A (x axis, Fig. 4 ) and N (y axis, Fig. 4 ).

By adjusting the model that relates theoretical Ks (estimated through Hagen-Poiuseulle Law) to F and S, using the three species dataset, we obtained values for ß and d similar to those estimated by Zanne et al . (2010) for all angiosperm species, that is, 1.5 and 0.5, respectively. As shown in Zanne et al . (2010) , we also observed a lack of orthogonality between S and F, and a greater influence of F on Ks than that of S. However, in contrast to their results, the variation in Ks was more influenced by F than by S. In addition, contrary to what we expected, the relative weight of S on Ks (d coefficient) was slightly lower in each of the studied species (or similar in the pooled analysis of the three species) than in the general survey by Zanne et al . (2010) . This means that vessel composition, estimated by S, has a low influence on Ks , in contrast to previous findings showing a high correlation between measures of vessel diameter distribution and measured maximum Ks ( Barotto et al ., 2016 ). The relative influence of S on Ks , which was low compared to F, is also a counterintuitive result considering the high relative weight of large vessels on Ks , since this is directly proportional to the 4th power of vessel diameter (Hagen Poiseulle Law). However, despite its lower relative weight, S has a significant influence on Ks , as its greater variability indicates that the plant can alter its conductive system to achieve greater conductive efficiency without modifying substantially the lumen fraction involved, which was a feature of great stability in all studied species. At the same time, and because conductivity increases approximately with the square of A, but only with the first power of N, altering vessel composition could have major influence on potential Ks . Nonetheless, in quantitative terms, the model applied to the three Eucalyptus species studied suggests that the great variation of S only produced a change in theoretical conductivity of 17 times, whereas F produced a change of 32 times.

The results of this study were in agreement with the second proposed hypothesis. Based on the average growth observed in the areas where the studied species are planted in Argentina, and because of the functional relationship between water transport capacity and growth ( Tyree & Ewers, 1991 ; Vander Willigen & Pammenter, 1998 ; Kondoh et al ., 2006 ), it was expected that E. viminalis would present a similar Ks than E. globulus, and both species would present a lower Ks than E. grandis . As expected, Ks of E. viminalis did not significantly differ from Ks of E. globulus . However, the mean Ks was relatively high, with an intermediate value between that of E. grandis and E. globulus . This relatively high Ks suggests that, compared to E. globulus , E. viminalis is a species with a high capacity for water transport, at least at stem level, which could imply that the growth levels it typically achieves are below its potential. This is probably due to the fact that, being a species with high resistance to adverse environmental conditions, E. viminalis is usually grown in areas with some kind of growth-limiting factor (drought, low temperatures). This is in agreement with the results obtained by Barotto et al . (2016) in a study of these two species planted in a common place in Argentina (analysis without environmental constraints) that compare hydraulic conductivity and vulnerability to cavitation (at branch level). Their results showed that, compared with E. globulus, E. viminalis had higher specific hydraulic conductivity (both measured and theoretically estimated from anatomy) and higher cavitation resistance.

In accord with the hypothesis, E. grandis had the highest theoretical Ks , which could explain, at least in part, the high growth rates of the clones analyzed in this study (between 45 and 55 m ³ ha ^-1 yr ^-1 at the age of 14 years, in years with high water availability, Tesón and Licata, EEA INTA Concordia, unpublished data). The high theoretical Ks for E. grandis was the consequence of high A and high N, which together caused F to be also the highest of the three species. These anatomical features are consistent with a high water transport capacity (high hydraulic efficiency), but probably with high vulnerability to drought and frost-induced cavitation. Vulnerability to cavitation curves in E. grandis (same clones as those analyzed here) indicate that this species has high vulnerability to drought-induced xylem cavitation, with a xylem water potential corresponding to 50% loss of hydraulic conductivity ( ₅₀ ) of -1.5 MPa ( Tesón et al ., 2012 ). These P50 values are also consistent with values reported for E. grandis planted in SouthAfrica ( Vander Willigen & Pammenter, 1998 ). These water potential values, and even lower ones, are achieved at leaf level under normal conditions during the growing season in the sites where E. grandis is cultivated in Argentina. First, it is important to point out that this high vulnerability to cavitation is not related to large vessels in E. grandis branches. On the contrary, this species has the smallest vessels at branch level ( Monteoliva et al ., 2017 ), compared to E. viminalis and E. globulus . In the three species, it seems that, at least at branch level, there is no trade-off between xylem safety and efficiency, but the opposite. Species with large vessels and high ks (such as E. viminalis ) are the safest ( Barotto et al ., 2016 ). There is no information about vulnerability to cavitation at stem level, and we cannot extrapolate results from branches to stems. So, in the case of E. grandis branches, it seems that the functional meaning of the high vulnerability to cavitation and resistance to water flow is to cavitate relatively soon during a drought acting as a fuse to protect the stem. Post-drought recovery in E. grandis has been observed in variables such as stem sap flux density and growth ( Tesón et al. , 2011 ). However, it is yet unknown whether the cavitated branches recover their Ks or if the plant resilience relies on the formation of new tissues. It has been argued that, in angiosperms with high vulnerability to cavitation, high wood capacitance and efficient repair mechanisms would be essential parts of water transport strategy ( Meinzer & McCulloh, 2013 ), which would be associated with low wood densities and small or negative safety margins (difference between the minimum daily water potential and cavitation threshold), leading to daily losses of Ks . This theory does not apply to E. grandis branches since they present high vulnerability to cavitation but high wood density, higher than the other studied Eucalyptus ( Monteoliva et al ., 2017 ). However, the low wood density of E. grandis stems, coupled with its high vessel lumen fraction, but also with its high fiber lumen fraction, suggest that it may have high capacitance to buffer internal water deficits and thus contribute to efficient water transport under high evaporative demand conditions (drought avoidance strategy). In contrast to the highly vulnerable branch xylem of E. grandis , ₅₀ of E. viminalis and E. globulus growing in Argentina was -3.76 MPa and -2.44 MPa respectively ( Barotto et al ., 2016 ), indicating quite high resistance to cavitation in the branches of these species and suggesting a different hydraulic strategy than that of E. grandis . In this regard, these species present a lower stem capacity than E. grandis for providing water to the leaves, resulting from their lower theoretical Ks and higher wood density (probably related to lower capacitance), but at the same time, presenting branches with higher Ks , lower vulnerability to cavitation and lower wood density. The combination of traits observed in Eucalyptus species challenges the well-established hypothesis about hydraulic functioning in woody species, as suggested by Meinzer & McCulloh (2013) , and deserves future research combining studies at stem, roots and branch level.

In the present study, when analyzing stem density, anatomy and theoretical Ks , a general trade-off did arise between wood density and Ks , which is in agreement with results reported by Pfautsch et al. (2016) for Eucalyptus growing in Australia. However, as in the case of Eucalyptus branches ( Barotto et al ., 2016 ), this should not necessarily imply a trade-off between hydraulic efficiency and safety. The particular and complex wood anatomy of Eucalyptus , composed of solitary vessels surrounded by vasicentric tracheids and parenchyma, allows for the possibility of avoiding this trade-off ( Barotto et al ., 2016 ). Moreover, there is no information available about embolism repair capacity in any Eucalyptus species, nor has their capacitance been quantified, so many questions about wood functional role in this genus remain unanswered.

Furthermore, E. grandis was the species that showed the greatest variation in vessel characteristics, suggesting that it would possess high genetic variability in these traits and/or a relatively high plasticity (greater than in the other studied species) to adjust its conductive system to accommodate to environmental conditions (e.g. climatic differences during the sampling period). Phenotypic plasticity in wood (dendroplasticity) is a fundamental mechanism for the adaptability of tree species to changing environmental conditions (e.g. Martinez-Meier et al ., 2008 , 2009 ). Its measurement in relation to the climatic conditions of the study sites will determine whether E. grandis is actually more plastic than E. viminalis and E. globulus , or if the environmental conditions where individuals grew had a greater range of variation, which resulted in a greater variability of wood characters. Beyond this, this study has quantified the effect of changes in F and S on theoretical water transport capacity ( Table 2 , Figs. 2 and 3), which could help explain the functional role of the anatomical changes observed both between and within individuals, for the three Eucalyptus species studied.