Sources of phenotypic variation of wood density and relationships with mean growth in two Eucalyptus species in Argentina

Aim of study: To describe the radial patterns of wood density, and to identify their main sources of variation, and the potential tradeoffs with mean tree growth, in two Eucalyptus species. Area of study: Mesopotamian (Corrientes and Entre Ríos provinces) and Pampean region (Buenos Aires province) of Argentina. Materials and methods: Eucalyptus grandis and Eucalyptus viminalis, growing in genetic trials installed in two sites per species were studied. X-ray wood microdensity profiles were developed from core samples. Each profile was proportionally divided in 10 sections. Mean, maximum, minimum and the standard deviation of wood density, for each section were computed. Mean annual growth was used to study the relationships with wood microdensity variables. A linear mixed-effects model computed the significance of different sources of phenotypic variation. Pearson ́s correlation computed the relationships between variables. Main results: The pattern of radial variation in E. grandis showed a decrease in wood density from pith to bark, mainly due to the decrease in minimum wood density, while in E. viminalis, wood density increased towards the outer wood. In both species, the standard deviation of the wood density increased along the radial profile from pith to bark. Significant variation in wood density was explained by site, provenance and clone/family effects. In E. grandis mean, maximum and minimum wood density were negatively correlated with mean growth, whereas in E. viminalis correlations were positive but close to zero. Research highlights: Both the pattern of radial variation of wood density and the relationship between wood density and mean growth were different in the studied Eucalyptus species, and they varied within species depending on the site they were growing and genetic provenance.


Introduction
Commercial forestry production tends to a scheme of multiple wood products. The first thinning produces principally raw material suited for pulp and boards while second or commercial harvests provide wood for solid uses (López & López, 2011). Thus, it is relevant to know the physical properties of wood produced in each growth phase of the trees since they affect the wood quality required by the different industries (Zobel & Jett, 1995;Castro da Silva, 2002;Apialoza et al., 2013;Burdon & Moore, 2018).
Within-tree wood variation, both longitudinal and radial, influences the quality of raw material (Downes et al., 1997). In juvenile-wood, located near the pith, wood properties largely vary yearly, resulting in a lower quality material for saw timber (Castro da Silva, 2002;López & López, 2011). In contrast, adult wood has more stable and appropriate properties for this industry (Fukasawa, 1984;Evans et al., 2000;Núñez, 2011). Properties that vary from juvenile to adult stage include wood density, fiber length, cell wall thickness, and chemical composition (Barnett & Jeronimidis, 2003), which depend on genetic, environmental factors and silvicultural management (Bao et al., 2001;Cobas, 2012).
Wood density is one of the most studied physical properties, probably associated with its relatively easy determination, becoming the most relevant physical wood property used in genetic improvement programs (Alves et al., 2020). It is an indicator of the quality and performance of wood, for both fiber and solid products (Apialoza et al., 2005). The sawmill industry prefers lower radial wood density variation, either for decorative or for structural purposes (López & López, 2011). In this regard, it is possible to predict wood behaviour under different efforts from its density (Cobas et al., 2014).
In addition to the within-tree variation of wood due to the cambial age, wood density shows a significant variation among trees, which is controlled by genetic effects (Zobel & Jett, 1995;Zobel & Sprague, 1998;Nabais et al., 2018), and by the environmental conditions during the cells wood formation (Schweingruber, 1996). In this regard, silvicultural management may influence wood density by affecting the resources availability for the trees (Zobel & van Buijtenen, 1989). Moreover, there may be different relationships -positive, negative or neutralbetween tree growth and wood density (Gonçalves et al., 2004). Thus, tree selection based on growth traits could have indirect consequences on wood density (Zobel & Jett, 1995), which demands determining the possible correlations between growth and wood density.
The genus Eucalyptus is one of the main components of global forestry, with more than 20 million hectares planted in the world (Iglesias-Trabado & Wilstermann, 2009). They contribute to a large proportion of planted forests -and the industrial chains they feed-of southern South-American countries (Brazil, Chile, Uruguay and Argentina). In Argentina, approximately 27 Eucalyptus species have been introduced. The most relevant commercial species of this genus is Eucalyptus grandis (Hill. Ex Maiden) (SAGPyA, 2013), which is grown in subtropical areas. It was introduced mainly due to its rapid growth, and has reached a high degree of genetic improvement (Marco & White, 2002), with maximum growth rates around 50 m 3 ha -1 year -1 and a mean wood density around 0.4 g cm -3 . Wood from short rotations, as commonly developed in E. grandis for pulp production, is highly unfavourable as raw material for solid purposes due to high tensile growth stress, presence of knots and large variability of their physical and mechanical properties (Castro da Silva, 2002;Murphy et al., 2005;Souza 2006;Hernández et al., 2014;. Of lower current relevance in commercial terms but of high potentiality for its high abiotic stress tolerance is Eucalyptus viminalis. It combines relatively high growth and frost tolerance (Cappa et al., 2010), allowing to be planted in temperate areas. This species has a lower degree of genetic improvement in Argentina, reaching mean growth rates ranging from 15 to 40 m 3 ha -1 year -1 in Pampean region, where it is subjected to freezing temperatures in winter and drought conditions in summer. Its mean wood density is higher than in E. grandis, around 0.6 g cm -3 . This species is currently used for the cellulose industry. However, it has been identified as the best species for veneer and plywood in Brazil (Iwakiri et al., 2013) as well as for other solid uses.
Based on this background, the objectives of the present study were: a) to identify the significant sources of phenotypic variation of the wood density of E. grandis and E. viminalis, by using the non-destructive technique of densitometry by X-rays and; b) to estimate phenotypic correlations between mean growth and wood density traits, such as mean, minimum and maximum wood density, and a measure of wood density variation along the stem radius. To have a large variation in the wood density traits studied, and to disentangle the relative relevance of environmental (site) vs. genetics, different genetic origins growing in two sites per species were analysed. The major results are interpreted and compared from the point of view of the wood technological implications and the prospects of genetic improvement of both species.

Study material
Two increment cores per tree were mechanically collected with a Pressler increment borer, but only one 5.15 mm diameter-sample per tree was analysed for this study purpose. The samples were taken at breast height (1.30 m) perpendicular to the stem, in south-north 3 Phenotypic variation of wood density in Eucalyptus orientation, in two experimental trials of each E. grandis and E. viminalis (Table 1). In E. grandis, the samples were collected in 2016 from two clonal trials installed in 2008 and 2010 in the Mesopotamian region of Argentina. In E. viminalis two provenance/half-sibs family trials were sampled in 2015, corresponding to a network installed from 1998 to 2000 in the Pampa region. In both species, the trials were installed by the Instituto Nacional de Tecnología Agropecuaria (INTA) of Argentina as part of its tree genetic improvement program.
Five to 15 trees were sampled per clon/family in E. grandis and E. viminalis, respectively (Table S1 [suppl.]). In E. grandis, provenance effect was grouped in "Local" ("Loc"), considering clones selected in Concordia (Entre Ríos, Argentina) from a local landrace of unknown provenance, and "Introduced" ("Int") considering newly introduced materials from four provenances of Australia (Table S1 [suppl.]). Due to the low number of individuals of each Int provenance, they were analysed together ("Loc" vs "Int", with around 90 individuals in each group, Table S1 [suppl.]). In E. viminalis, five provenances were also studied (one local and four more recently introduced from Australia, Table S1 [suppl.]). In this species, since we had a balanced number of samples, we could analyse the effect of each individual provenance on the studied variables. In total, 184 and 582 increment cores were collected in E. grandis and E. viminalis, respectively.

X ray densitometry: wood microdensity
Increment cores were cut lengthwise, perpendicularly to the fiber. A 2 mm thickness lamina per core was dried to moisture equilibrium and subsequently analysed by indirect X-ray densitometry (Polge, 1966). The resulting X-ray films were scanned and the digital images were processed with the WinDENDRO® software (Guay et al., 1992), obtaining a spatial resolution of 25 μm. The variation of shades of grey in each lamina was contrasted with a scale (different thickness) radiographed with the samples for which it is known its density, allowing to transform the grey pattern of each lamina in density values.
The last step of data processing used a computer routine written in R language (R Development CoreTeam, 2015). Since the studied Eucalyptus do not present well-defined growth rings, each microdensity profile was divided proportionally in 10 sections of equal length, from pith to bark. The section length of each individual sample depended on the whole sample length (Fig. S1 [suppl.]). The following microdensity variables were computed from each section: Mean wood density (Meds), maximum wood density (Maxs), minimum wood density (Mins), and standard deviation (Stds) representing the variability of wood density values for each section. Likewise, the corresponding values for the complete microdensity profile were estimated: Med r , Max r Min r and Std r . Finally, the diameter at breast height (DBH, cm) was measured to estimate the mean annual growth (MAG, cm year -1 ) of each tree by dividing the DBH by the tree age (DBH/age). variation. All sections of a tree were considered as repeated measurements of wood density upon the same tree, describing the radial wood density variation (Faraway, 2006;Bates, 2010). The maximum restricted likelihood method (REML) was computed to estimate values in the following lmer function:

Statistical analysis
Where, Y ijkl = ijkl observed wood microdensity value (mean, maximum, minimum and standard deviation) corresponding to the ijkl -section μ =overall mean α i = fixed effect of the i -section β j = fixed effect of the j -site т k = fixed effect of the k -provenance β j т k = interaction site by provenance a l(k) = random effect of the l -clon/family nested in the k -provenance b i(m(l)) = microdensity value of the i -section of the m -tree nested in the l -clon/family, of the k -provenance nested in the j -site e ijklm = random error In both species, provenance was considered a fixed factor. In the case E. grandis because only two levels were taken into account ("Loc" and "Int"), and in E. viminalis, even when all the provenances were considered, they corresponded to a second cycle of introduced genetic plant materials from a restricted region of the natural distribution area of the species, selected by their growth performance.
Significance of fixed levels was determined using the Satterthwaite Approximation method (Bates, 2010). Random levels significance was computed using the likelihood ratio test (P < 0.05), comparing the complete model with a reduced model without the factor.
Pearson's correlation coefficients (p< 0.05) were computed between MAG and wood density variables (Med r , Max r , Min r and Std r ) at tree level. The function corr.test of the R software was used for this purpose (R Development CoreTeam, 2015). The r values interpretations were done considering sign (positive or negative) and grade or force (low: 0.10< r 0.30; moderate: 0.30< r 0.50; and hight: > 0.50) (Vinuesa, 2016).

Descriptive statistics
Mean wood density in E. grandis and E. viminalis was 0.43 g cm -3 and 0.55 g cm -3 respectively, ranging from 0.31 g cm -3 to 0.61 g cm -3 and from 0.43 g cm -3 to 0.77 g cm -3 in E. grandis and E. viminalis, respectively (all sources of variation). MAG and wood density variables described for both species in each site (mean values and their standard error) are shown in Table 2. To see the mean values of each clone or family per site and species, please refer to the Figs. S2 and S3 [suppl.] provided as supplementary material.
Considering both sites, in E. grandis, Med r was similar in Loc and Int provenances (0.43 g cm -3 ± 0.04 g cm -3 ). Similar values for Max r , Min r and Std r were computed in CAAC for both provenances, while Med r and Max r were higher in CONC for Loc in relation to Int (0.45 g cm -3 vs 0.43 g cm -3 , and 0.542 g cm -3 vs 0.52 g cm -3 , respectively).
In both species, the coefficient of variation was higher for MAG (22% and 35% for E. grandis and E. viminalis, respectively) than for Med r , Max r and Min r , which were lower and similar between species and traits: 10% in Med r , 9% in Max r and 12 % Min r . In the case of Std r , the coefficient of variation was higher than the other density variables (24% and 17% in E. grandis and E. viminalis, respectively).

Phenotypic variation of wood density in Eucalyptus
Significant sources of phenotypic variation: longitudinal model with repeated measures The mean values estimated for the fixed effects and their statistical significance, as well as the components of variance for random effects expressed in terms of standard deviation, are presented in Tables 3 and 4 for E. grandis and E. viminalis, respectively.
The estimated mean wood density in E. grandis was 0.45 g cm -3 . Wood density significantly decreased along the radius of the tree (section effect), both for mean and minimum wood density: -0.004 g cm -3 and -0.006 g cm -3 per section, respectively, from pith to bark. However, the variability of wood density through the section increased in the same direction. The radial decrease of the maximum density was not significant, so it seems that the decrease in mean wood density along the radius in E. grandis could be associated with the decrease in the minimum wood density (see Table 3 and Fig. 1).
Significant differences could also be established in E. grandis (Table 3) between sites for mean, maximum and standard deviation of wood density, while the effect of provenance (Loc vs Int, indicated "Loc" as reference value in Table 3) was not significant in any of the microdensity variables. However, the interaction site x provenance was significant for maximum wood density. This is the consequence of an increase in maximum wood density of Loc in CONC, even when this increase was not associated to a significant difference among provenances.
Regarding random effects, even when the major contribution to the total phenotypic variance was related to the individual trees, only clone level was statistically significant for all studied wood density traits (Med r , Max r and Min r ), except for standard deviation (Table 3).
In E. viminalis (Table 4), the radial variation was significant for all wood density traits analysed. Unlike the patterns observed in E. grandis, wood density traits (mean, maximum, minimum) increased throughout the radius of E. viminalis stems (Fig. 1). For each 10% of radial section, mean density increased 0.02 g cm -3 , and the same was observed for maximum (0.03 g cm -3 per section) and minimum wood density (0.01 g cm -3 per section). The radial variability of wood density values, measured as the standard deviation, increased from pith to bark similarly to E. grandis (Fig. 1). However, in E. viminalis increased fourfold (Fig. 1).
Significant differences were also found between sites in E. viminalis, but not in the mean wood density ( Table  4). In Del Valle site (VALL) the maximum wood density and the variability of wood density were lower, while the minimum wood density was higher, than in Guaminí site (GUAM). In relation to the provenances effect, mean and minimum wood density were significantly lower for Bald Hills (Bal), Federation Road (Fed) and Bonang (Bon) provenances than for Argentina (Arg) (reference value in the model), but no differences were observed between Argentina and Errinundra Road (Err). For maximum wood density, only significant differences were found between Bonang or Errinundra Road and Argentina. All introduced provenances showed higher standard deviation than Argentina, although these results need to be interpreted site by site given that all provenances, but Errinundra Road, showed lower standard deviation in Del Valle than Guaminí.
Statistically significant contributions were established for families and tree levels in E. viminalis for the four variables studied (mean, maximum and minimum wood density and the standard deviation). A major contribution to the variance was computed for maximum wood density, both at family (0.02 g cm -3 ) and tree levels (0.07 g cm -3 ). When the data were disaggregated by site (Table S3 [suppl.]), the correlations at individual level between MAG and the different wood density variables (Med r , Max r and Min r ) in E. grandis were low to moderate and negative, while a correlation with positive sign was found between MAG and Std r . The correlations between MAG and the wood density variables in CAAC and CONC, respectively, were as follows: Med r (r: -0.37 and -0.25), Max r (r: -0.22 and -0.24), Min r (r: -0.45 and -0.26) and Std r (r: 0.18 and -0.13). However, at the clone level, the correlations between MAG and the wood density variables increased in CAAC in relation to the pooled data of both sites. For example, negative relationships were found (r: -0.56, -0.43 and -0.65, for Med r , Max r and Min r , respectively), not being significant the relations between MAG and Std r . In the case of CONC, these correlations, although negative, were not significant.
Regarding E. viminalis results analysed by site ( Table  S3 [suppl.]), the correlations at individual level between MAG and the wood density traits (Med r , Max r and Min r ), differently to E. grandis, were positive in both sites, while the correlation between MAG and Std r was negative in GUAM (r: -0.09) but significant only in VALL (r: 0.17). No significant correlations were established at the family level in each site between MAG and the four wood density variables (Table S3 [ suppl.]).

Discussion
Mean wood density values of E. grandis reported in this study are in the range of values informed by other  Phenotypic variation of wood density in Eucalyptus authors. Arango & Tamayo (2008) reported mean wood density values ranging from 0.38 g cm -3 to 0.55 g cm -3 in clones of E. grandis (8 year-old), similar to those indicated by Castro da Silva (2002) (0.31 g cm -3 -0.59 g cm -3 ). In both cases, the studies were carried out in Brazil.
Values reported for different genetic trials and commercial plantations of this species in Argentina were in the range of 0.39 to 0.49 g cm -3 (Harrand & López, 2007;López & López, 2011;Alarcón et al., 2018). However, Monteoliva et al. (2017) obtained mean density values higher than those reported here (0.52 g cm -3 ) in clones of E. grandis planted in Entre Ríos, Argentina (southwards from CONC). The estimated values are in the range considered suitable for pulp and paper production (0.40 g cm -3 to 0.60 g cm -3 , Downes et al., 1997), wood density values that can be obtained from juvenile wood in E. grandis (Núñez, 2011).
In E. viminalis, average wood density was 0.55 g cm -3 with no significant differences between sites. These values are similar than the higher values reported for E. viminalis by Otegbeye & Kellison (1980), Pathauer (2005 (0.40 g cm -3 -0. 57 g cm -3 ) and Alarcón et al. (2018) for the basic wood density (0.48 g cm -3 -0.52 g cm -3 ). However, our values are lower than those reported by Iwakiri et al. (2013) of 0.61 g cm -3 and Monteoliva et al. (2017) of 0.67 g cm -3 for the basic wood density, in the same plantations of this study. These differences highlight the importance of the methodological approaches to determine the basic wood density, which can lead to different results on the same materials (Alarcón et al., 2018).

Sources of phenotypic variation of wood density
Significant differences among sites, provenances, clones or families have been widely reported for wood characters both in coniferous species (Zobel & Sprague, 1998;Larson et al., 2001;Rigling et al. 2002;Martínez-Meier et al., 2011;Salaya-Domínguez et al., 2012;George et al., 2015;Klisz et al., 2016), and in broad-leaves (Arango & Tamayo, 2008;Harrand et al., 2009;Cappa et al., 2010;López & López, 2011;Moreno & Igartua, 2015). In E. grandis, López & López (2011) showed significant differences involving 12 sites for both mean wood density and radial wood density variation. In the present study, the fixed effect of site showed significant differences in E. grandis for mean, maximum and standard deviation of wood density. Higher values were found in CONC, representing a two-year older plantation than CAAC. CONC is a site with higher average annual precipitation and lower average annual temperature (i.e. a more favourable water balance) (Instituto de Clima y Agua, Instituto Nacional de Tecnologia Agropecuaria, Argentina) than CAAC. In E. viminalis, the site level was also statistically significant for maximum, minimum and standard deviation wood density, but not for mean wood density, showing GUAM the higher values of maximum wood density but the lower minimum wood density, and a larger standard deviation than in VALL.
Significant differences among sites could be attributed partially to slight differences in age, since wood density normally increases with age (Bermúdez Alvite et al., 2002;Resquin et al., 2012;Moreno & Igartúa, 2015), but also to differences in environmental conditions, not only local but also as evolutionary legacies of the climate of the origin of the plant material . Dendroecology studies have shown the sensibility of wood density and growth to climatic conditions at inter-annual level (Schweingruber FH, 1996;Rozas et al., 2016) as well as intra-annual level (e.g. Rozenberg et al., 2001;Rozenberg & Paques, 2004;Martinez-Meier et al., 2015). High rainfall, which ensures high soil water availability, stimulates the biomass production (e.g. Le Quéré, 2015), whereas soil water deficit would stimulate the production of high wood density (Bouriaud et al., 2005). In our study, however, higher wood density was produced in the site with higher precipitation and lower mean temperature (e.g. better water balance, CONC site) in E. grandis. In the case of E. viminalis the differences in rainfall between sites are somewhat compensated by differences in temperature and soil conditions making difficult to know a priori which site has a better water balance for the plants (see Table 1). In this regard, the mean wood density did not differ between sites in this species, and maximum and minimum densities did differ in opposite direction.
Regarding the effect of mean temperature, Thomas et al. (2004) found a plastic response of wood density in E. camaldulensis, both variables being positively related. This contrasts to our results in E. grandis, which presented higher wood density in the site with lower temperature (CONC). In the case of E. viminalis, both sites largely differed in mean temperature (around 3°C), but that did not result in significant differences in mean wood density. The different responses to mean climatic conditions of the studied species -among them and compared to other studies-could be associated to the seasonal variability of climatic conditions and to the particular strategies of response to abiotic stressor agents at species level (Olivar et al., 2003;Wimmer & Downes, 2003;Bouriaud et al., 2005). In this sense, it has been proposed that the studied species respond differently to water deficit because of different hydraulic architectures .
Another source of phenotypic variation in wood density in our studied species was the genetic provenance. According to Nabais et al. (2018), species with a large geographic range may present high variability between provenances for traits such as wood density. In the case of E. viminalis, although the provenances studied in this work represent only a small portion of the natural range of the species in Australia/Oceania (a range of 15 degrees of latitude, and from the sea level to 1,500 m altitude), they correspond to a second introduction of plant material selected for their growing performance (Cappa et al., 2010). This may imply that they would maintain also significant differences for wood characteristics that can be explored further for the genetic improvement of the species. It is relevant to indicate that in the present study, a trend towards higher wood microdensity values in lower altitude provenances were observed, coincident with the plant material indicated by Cappa et al. (2010) as with the best growth performance in Argentina. No differences were found between provenances in E. grandis, probably because the genetic material was pooled at this level. However, the genetic effect at clone level showed a significant component of variance for all wood density variables, except for Std r .
Variation in wood density was observed among individuals (in E. viminalis), provenances and sites within each species, but also along the radial section of the stems. The segmentation of the radial profile describing radial wood density variation, both in E. grandis and in E. viminalis, resulted from the impossibility of delimiting annual growth rings in the studied species. The 10 established sections allowed describing their variation by means of a mixed linear model. This variation was significant for the mean and minimum wood density and its standard deviation in E. grandis and for all the variables studied in E. viminalis. These results lead to conceptual patterns of radial variation showing marked differences inherent to the species. It can be attributable to the presence of only juvenile wood in E. grandis and juvenile-mature wood in E. viminalis considering the age of the sampled trees and references of wood stage for the same species (Núñez, 2011;López & López, 2011;Iwakiri et al., 2013). For E. grandis, the pattern of variation was coincident with the type 3 modelled by Panshin & Zeeuw (1980), described for species with circular porosity of vessels, but also for E. grandis of diffuse porosity (Fukazawa, 1984;Arango & Tamayo, 2008;Núñez, 2011). In E. viminalis the significant increase in the mean, maximum and minimum wood density along the radius, from pith to bark, was coincident with the pattern 1 described by the same authors (Panshin & Zeeuw, 1980), which is described for species with diffuse porosity (Fukasawa, 1984). These changes in wood density along the radial direction, from pith to bark, are influenced by the cambial age as well as by the annual growth (Cobas, 2012), and the shape of the curves describing the radial pattern may differ when are analysed based on the growth-ring widths than with fixed sections. The type 1 form of radial variation is described by a positive exponential form with a rapid growth in the juvenile portion (near to the pith) followed by a stabilization of density values towards the bark. That general form was observed for the pooled data of E. viminalis. In this sense, the linear model here proposed implies a simplification of the observed pattern. However, even with this simplification, the model represents a good tool to describe the general trends and their sources of variation, capturing satisfactorily most of the radial variability pattern (Fig. 1). Within-tree source of variation is indicated as the major relevance source of wood density variation for broad-leaf species with diffuse porosity (Downes et al., 1997). On the other hand, although the authors do not propose it as a causal effect, the radial variation in wood density is related to the fibers-length variation in Eucalyptus, which increases from pith to bark (Igartúa & Monteoliva, 2010).
Given that the presence of growth stresses is a phenomenon in Eucalyptus genus, varying at intra and interspecific level, with several implications from the economic point of view , identifying traits that can be measured by non-destructive methods renders benefits for both genetic programs and wood industry, optimizing tree selection and industrial processes. In this regard, significant relations between growth stresses and basic wood density variation from pith to bark has been described for several species, including E. dunni (Hernández et al., 2014), E. grandis and interspecific hybrids . In our study, we described the form of that variation in the studied two species, as well as how it may change due to site and genetic effects (which must be explored more in detail). The implication of these changes on the technological characteristics of the wood needs still to be explored. Zobel & Jett (1995) argued that a possible genetic gain in growth could have an indirect effect on wood properties due to a possible unfavourable genetic correlation between growth and wood density. Even when in this study we only explored phenotypic relationships, they were highly complex in E. grandis. The negative relationship between MAG and wood density (measured by x-ray densitometry) was described by Aparicio et al. (2011). However, the study of this relationship at the site level showed that there exists a negative correlation between MAG and density in CAAC, the site with higher mean growth, but a non-significant correlation in CONC, where growth was more limited. It is important to note that although CAAC was the site with the highest growth of both studied, it is not the best site for the species in the studied region.

Phenotypic correlations between wood density and mean stem growth
In the case of E. viminalis, the correlations MAG-Med r and MAG-Max r although close to zero, were positive in both sites evaluated. This type of positive relationship between growth and wood density has also been described for E. urophylla (Kien et al., 2008) and E. globulus (Igartua & Monteoliva, 2010). However, for this last species, a negative relationship has also been described for Phenotypic variation of wood density in Eucalyptus Chilean sites (Lizana, 2006), which demonstrates that there are not unique relationships between growth and wood density even at a species level.

Conclusions
In both studied species, the results show that there is a significant variation in wood density along the radial profile (within-tree), between trees (in E. viminalis), between sites and between genetic entities (provenances, clones or families) for the wood density variables assessed. Despite considering only two sites for each species and a limited number of provenances, clones and families, the existence of a significant difference in relation to the common gene pool, indicates the possibility of identifying genotypes with desirable technological properties. This is particularly relevant in the frame of the genetic improvement program of E. viminalis considering the potentially reduced genetic basis (i.e. a limited representation of the large natural distribution area of the species). Depending on the requirements of the industry, whether low wood density variation is desired or high or low mean wood density values, these different wood characteristics can be explored and potentially selected in the genetic program.
Due to differences in mean growth rates, E. grandis supplies similar log diameters in half of the time compared to E. viminalis. This raw material, composed of mostly juvenile wood in the former species, has a higher wood density in the pith than closed to the bark. In contrast, probably due to differences in wood type (juvenile-mature) in E. viminalis, an opposite pattern was observed, with higher wood density towards the outer wood.
Although in E. grandis there is a negative association between stem growth and wood density, described here and in previous studies, this relation is complex and probably driven by sites or clones that allow high growth rates. This compromise seems to disappear in sites of relatively low growth potential, but that can still contribute an important proportion of the area planted with this species in Argentina. More research is needed considering growth-wood density relationships in a broad range of site qualities and plant origins (clonal vs seed-origin materials). In E. viminalis, even when the mean wood density was similar in both sites and no trade-off was observed with growth, the least productive site presented lower minimum but higher maximum wood density. This variability associated with site conditions could have implications in determining the quality of E. viminalis wood for sawmill and deserves been explored in a broader range of environmental conditions. The results also suggest that even if no differences between sites are found for mean wood density, its variation along the radial profile may differ between sites and should be taken into account to better explain wood properties related to the quality for solid uses.
Site conditions not studied here, such as the effective soil depth, soil water retention capacity, seasonal and interannual patterns of rainfall, would play a major role on wood characteristics, with a higher effect than the mean temperature. This highlights the complexity of the studied processes and the risks of extrapolating relationships between wood density and productivity from other species of the same genus, or even from the same species but of different genetic source and/or growing in other environmental conditions. Further research is needed to elucidate the effect of these environmental factors on the wood properties, in interaction with growth rates, in woody species.