The links between wood traits and species demography change during tree development in a lowland tropical rainforest

Abstract One foundational assumption of trait-based ecology is that traits can predict species demography. However, the links between traits and demographic rates are, in general, not as strong as expected. These weak associations may be due to the use of traits that are distantly related to performance, and/or the lack of consideration of size-related variations in both traits and demographic rates. Here, we examined how wood traits were related to demographic rates in 19 tree species from a lowland forest in eastern Amazonia. We measured 11 wood traits (i.e. structural, anatomical and chemical traits) in sapling, juvenile and adult wood; and related them to growth and mortality rates (MR) at different ontogenetic stages. The links between wood traits and demographic rates changed during tree development. At the sapling stage, relative growth rates (RGR) were negatively related to wood specific gravity (WSG) and total parenchyma fractions, while MR decreased with radial parenchyma fractions, but increased with vessel lumen area (VA). Juvenile RGR were unrelated to wood traits, whereas juvenile MR were negatively related to WSG and axial parenchyma fractions. At the adult stage, RGR scaled with VA and wood potassium concentrations. Adult MR were not predicted by any trait. Overall, the strength of the trait-demography associations decreased at later ontogenetic stages. Our results indicate that the associations between traits and demographic rates can change as trees age. Also, wood chemical or anatomical traits may be better predictors of growth and MR than WSG. Our findings are important to expand our knowledge on tree life-history variations and community dynamics in tropical forests, by broadening our understanding on the links between wood traits and demography during tree development.


Introduction
The main goal of trait-based ecology is to understand how traits are related to species demography (Shipley et al. 2016).This is of central importance for understanding life-history variations and mechanisms of community assembly (Poorter et al. 2010;Wright et al. 2010), as well as to improve dynamic global vegetation models (e.g.Worthy and Swenson 2019).Although significant advances have been made in the last decades in explaining the links between traits and demographic rates (e.g.Poorter et al. 2010;Wright et al. 2010;Hérault et al. 2011;Aubry-Kietz et al. 2015;Visser et al. 2016), the predictive power of most traits on species demography is generally low (e.g.Visser et al. 2016;Yang et al. 2018).Two main reasons may explain this pattern.First, research on this topic has focussed largely on a few relatively easy-to-measure traits, which may not fully capture some plant functions (Yang et al. 2018).This leads to the question of whether other traits more mechanistically linked to plant functions, despite being relatively more difficult to measure, can be better predictors of species demographic rates (Poorter et al. 2010;Russo et al. 2010;Fan et al. 2012).Second, most previous studies on trait-demography relationships have considered either only one stem size class or have averaged broad stem size classes (e.g.Russo et al. 2010;Wright et al. 2010), overlooking sizerelated variations in both traits and performance (e.g.Visser et al. 2016;Rungwattana and Hietz 2017).Therefore, it remains unclear to what extent the effects of traits on demography change during tree development (Iida et al. 2014).
A number of studies have shown that wood-specific gravity (WSG) can predict, to some extent, species demographic rates (Chave et al. 2009;Poorter et al. 2010;Wright et al. 2010;Hietz et al. 2016;Visser et al. 2016).In general, WSG is negatively related to mortality and diameter growth rates (Poorter et al. 2008;Wright et al. 2010; but see Russo et al. 2010; see Fig. 1).Yet, specific gravity is an emergent property of wood that is affected, in angiosperms, by the fractions and morphologies of fibres, vessels and parenchyma cells (e.g.Zieminska et al. 2013Zieminska et al. , 2015)).The main functions of these cell types are related to mechanical strength, water transport and storage, respectively (e.g.Carlquist 2001).Therefore, the relationships of WSG with growth and mortality rates (MR) may be better understood by considering wood anatomy (e.g.Poorter et al. 2010;Russo et al. 2010; see Fig. 1).
The anatomical traits underlying the relationship between WSG and diameter growth rates are relatively well studied (e.g.Russo et al. 2010;Fan et al. 2012;Hietz et al. 2016;Fig. 1).For example, low-WSG species grow fast in part because they generally have lower fibre fractions and/or thinner fibre walls (Zieminska et al. 2013), which may reduce stem construction costs (King et al. 2006).They also build wider vessels which favour high xylem hydraulic conductivity and support high leaf-photosynthetic carbon gain (Santiago et al. 2004;Poorter et al. 2010;Hietz et al. 2016).In addition, tree growth may also depend on wood nutrient concentrations (Martin et al. 2014; Fig. 1).For instance, physiological studies have shown that calcium (Ca), potassium (K) and magnesium (Mg) are involved in wood formation, particularly during cell expansion and differentiation (Ache et al. 2010;Fromm 2010).Phosphorous (P), on the other hand, plays critical role in plant metabolism as it is a structural component in ribonucleic acid (RNA), and acts as a metabolic energy unit in adenosine triphosphate (ATP, Jiang et al. 2019), among others.Furthermore, nutrient addition experiments in lowland tropical forests have shown that tree growth is commonly limited by soil P (Vitousek 1984;Turner et al. 2018) or K availability (Wright et al., 2011(Wright et al., , 2018)).Thus, there are good reasons to think wood nutrient concentrations should be related to growth rates.Yet, the extent to which wood nutrients influence growth is unclear, especially in diverse tropical forests, as studies examining the links between wood nutrient concentrations and growth rates in tropical trees have been scarce and show seemingly contradictory results (Martin et al. 2014;Heineman et al. 2016).
On the other hand, it is well known that species with higher WSG tend to have lower MR (e.g.Wright et al. 2010;Visser et al. 2016).However, the anatomical mechanisms driving this relationship are still unclear (but see Poorter et al. 2010;Osazuwa-Peters et al. 2017;Fig. 1).This is due to the fact that similar values of WSG can be the product of different wood anatomies (e.g.Zieminska et al. 2015) and also, and more importantly because tree mortality can be simultaneously influenced by different processes (Russo et al. 2010;Hietz et al. 2016;Osazuwa-Peters et al. 2017).For instance, the lower MR that characterize high-WSG species may be due to high fibre fractions and thicker fibre walls that may increase structural strength (Poorter et al. 2010) and resistance to pathogen attacks (Alvarez-Clare and Kitajima 2007), and/or due to low xylem turgor loss points that might counteract the adverse effects of water deficits on living cells (Santiago et al. 2018).
In addition, there are possible links between xylem parenchyma cells and species mortality, but they remain poorly understood (Poorter et al. 2010;Morris et al. 2016a; Fig. 1).For instance, a higher allocation to axial (AP) and radial (RP)  Figs 2, 3 and 4 for details), while dashed arrows are hypothesized, although nor realized, links between traits and demographics rates.González-Melo et al. -Associations between wood traits and demography during tree development parenchyma cells could reduce mortality (Poorter et al. 2010), because these cells store non-structural carbohydrates (NSC) and nutrients (Plavcová and Jansen 2015; Kotowska et al. 2020), which might enable faster recovery from disturbances (Kitajima and Poorter 2007;Martin et al. 2014), or can provide an active response to xylem infections and mechanical damage (Carlquist 2001;Morris et al. 2016bMorris et al. , 2019)).The effects of parenchyma cells on MR may depend, to some extent, on their stem cross-sectional fractions (Poorter et al. 2010;Morris et al. 2016b).While total parenchyma fractions (i.e.AP + RP) tend to be lower in high-WSG species (e.g.Fortunel et al. 2013;Zieminska et al. 2015), AP (Gonzalez-Melo et al., unpubl. res.) or RP (Zheng and Martinez-Cabrera 2013) fractions can scale positively with WSG.This suggests that AP and RP cells may have different patterns of associations with WSG and MR.However, very few studies have formally examined the possible links between xylem parenchyma cells and MR across tropical tree species (but see Poorter et al. 2010).
As wood traits (e.g.Martin and Thomas 2013; Rungwatanna and Hietz 2017) and demographic rates (e.g.Hérault et al. 2011;Visser et al. 2016) can vary substantially with tree size, it is likely that trait-demography links may change accordingly (Fig. 1).For instance, seed size and stature are important to predict seedling performance, while WSG seems to be the best predictor of the demography of small, but not of large trees (Lida et al. 2014;Visser et al. 2016).Similarly, in a recent study conducted in a tropical semi-deciduous forest, Osazuwa-Peters et al. (2017) found that species demography was more strongly related to wood anatomical traits, such as cell fractions, in small than in large trees (Osazuwa-Peters et al. 2017).However, it is unclear if this trend holds true for other forest types.To our knowledge, no previous studies have examined the extent to which the relationships between wood anatomical traits and demographic rates change during tree development in lowland tropical humid forests.
Here, we studied how wood traits (i.e.WSG, and wood anatomical and chemical traits) are related to growth and MR in 19 tree species from a lowland forest in eastern Amazonia.In particular, we wanted to answer the following questions: (i) How are wood traits related to growth and MR?We hypothesized that diameter growth rates will increase with wood nutrient concentrations (i.e.P, K, Mg and Ca) and xylem water transport capacity (i.e.vessel lumen area), and decrease with stem construction costs (i.e. higher WSG, fibre fraction or fibre wall thickness (F WT )).We also predicted that MR will be negatively related to traits linked to stem structural strength (i.e.WSG, fibre fraction and F WT ) and either to AP or RP fractions.(ii) Do the links between wood traits and demographic rates vary with tree size?We expected that the traits associated with growth and mortality would change during tree development and that the trait-demography links would be weaker at later ontogenetic stages.

Study site
The research was conducted in the Paracou research station, a lowland tropical rainforest located in northern French Guiana (https://paracou.cirad.fr/;5° 18ʹN, 52° 55ʹW).At Paracou, the mean annual temperature is 28.4 °C, and annual rainfall averages c. 3000 mm with a marked dry season occurring between August and November, and a distinct rainy season from March to June (Wagner et al. 2011).Dominant families in the forests of Paracou include Fabaceae, Lecythidaceae, Sapotaceae and Chrysobalanaceae (Hérault et al. 2011).The landscape is characterized by moderate hills separated by narrow streams (Ferry et al. 2011), and soils are strongly P-limited (Grau et al., 2017).

Species and sampling
We selected 19 tree species that represent a broad gradient of variation in shade tolerance, ranging from pioneers to understory or canopy shade tolerants (Table 1).These species also spanned a wide range of WSG and wood anatomical traits (Table 2).We sampled 75 large trees (i.e.>10 cm diameter at breast height (DBH)), with two to five individuals per species (Table 1).Stem discs were collected, at breast height (c.1.3 m), from previously cut-down trees.All wood samples were collected in Paracou, except for Schefflera morototoni, Cecropia obtusa and Miconia tschudyoides, which were collected in nearby secondary forests.
As wood structure is conserved as trees age, ontogenetic shifts in wood structure can be analysed by measuring radial (i.e. from pith to bark) changes in wood traits.Consequently, in each stem disc, radial segments were cut and then divided into three radial sections: (i) 1-5 cm, (ii) 5-10 cm and (iii) >10 < 50 cm [see Supporting Information-Fig.S1].Hereafter, we refer to these radial sections as sapling, juvenile and adult wood, respectively.It is important to note that here we used stem diameter to define these three ontogenetic stages (i.e.sapling, juvenile and adult wood).Yet, we acknowledge that this categorization of ontogenetic stages can differ from others found in the literature, and may not always reflect real ontogeny because species can attain maturity at different stem diameters (e.g.Wright et al. 2005).

Trait measurements
Wood specific gravity.In each sapling, juvenile and adult wood section [see Supporting Information-Fig.S1], samples of c. 2 × 2 × 1 cm were cut every 1 cm.In these wood samples, we measured fresh volume and dry mass.Fresh volume was measured with the water displacement method, and dry mass was obtained after drying the segments at 103 °C to a constant mass for 72 h.WSG per sample was defined as dry mass over fresh volume (Kollman and Coté 1968).Then, mean WSG values for sapling, juvenile and adult wood were calculated by averaging values from the 1-cm samples.As the deposition of wood extractives (WE) in heartwood or sapwood can increase wood dry mass and, thus, alter WSG values (e.g.Lehnbach et al. 2019), it is possible that the links between WSG and demographic rates may be affected by the concentration of WE.To examine the effects of WE on the WSG-demography associations, we recalculated sapling, juvenile and adult WSG (i.e.WSG REC ) of five species (Bocoa prouacensis, Virola michelii, Sextonia rubra, Bagassa guianensis and Dicorynia guianensis) by removing the effect of WE concentrations on wood dry mass, and run analyses (see below) using WSG REC values.We obtained WE concentrations from both published (Amusant et al. 2014) and unpublished studies (Rodrigues 2010).
Wood anatomy.In sapling and juvenile wood, anatomical analyses were conducted on one wood sample (c. 2 × 2 × 1 cm); whereas in adult wood, anatomy was analysed every 1.5 cm until reaching the bark, and an averaged value was calculated [see Supporting Information-Fig.S1].To characterize wood anatomy, cross-sectional surfaces were sanded using a polishing machine with 1200-grit diamond discs, and then samples were cut with a GLS-1 sledge microtome to get a plane surface.Photographs were taken at 5-10 × objective lenses using a reflected light (episcopic) microscope (BFMX, Olympus, Tokyo, Japan), equipped with a digital camera (Canon EOS T6i; Canon Inc., Tokyo, Japan).For each wood sample, between 10 and 20 partially focussed images were taken and were then combined using Helicon Focus (Helicon Focus Ltd., Kharkov, Ukraine).From these anatomical images, vessel lumen area (V A , mm 2 ) and cross-sectional area fractions of fibres (F F ), vessels (F V , lumen + wall), axial (F AP ), radial (F RP ) and total parenchyma cells (F TP = F AP + F RP ), were measured.
To calculate cell fractions and V A , wood cell types were manually coloured using Photoshop (Adobe Systems Incorporated, USA), and traits were calculated automatically using the batch function in the software ImageJ (https:// imagej.nih.gov/ij/).F WT (μm) was measured by taking photographs at 50-100× objective lenses using a laser microscope (VK 8850, Keyence).To do so, each anatomical image was Table 1.Study species, family, number of trees sampled (n), mean diameter at breast height (DBH) of sampled individuals and ecological guilds for 19 Amazonian tree species from a lowland forest in eastern Amazonia.Ecological guilds were assigned based on Favrichon (1994), Uriarte et al. (2005), Scotti et al. (2010) and Bossu et al., (2016).

Species
Family  divided into 4 equal sections, and 8 pairs of fibres were randomly selected in each section, for a total of 32 pairs of fibres per image.To obtain F WT , double wall thickness was measured and then divided by two using ImageJ.In total, we measured eight wood anatomical traits (Table 2).
Wood nutrient concentrations.We measured wood nutrient concentrations in both sapling and adult wood [see Supporting Information-Fig.S1].Yet, as wood nutrients tend to be reabsorbed from inner to outer wood, during heartwood formation, in several of our study species (A. Gonzalez-Melo, et al., unpub. res.), nutrient concentrations from sapling wood were not considered reliable for this study.Therefore, here we only considered values of wood nutrient concentrations from adult wood.Wood nutrient concentrations were estimated on the sample closest to the bark.Whenever possible, nutrient analyses were conducted on the same wood samples that were used to calculate WSG and anatomical traits.To calculate wood nutrient concentrations, wood samples were ground and dry-ashed at 550 °C for 1 h, and then the ash was dissolved in HNO 3 (1 M).Wood base cations and P were calculated using inductively coupled plasma-optimal emission spectrometry (ICP-OES) on an Optima 7300 DV (Perkin-Elmer Ltd., Shelton, CT, USA), with apple leaves (NIST 1515) as reference samples.
Demographic rates.To calculate adult (i.e.>10 cm DHB; see Supporting Information-Fig.S1) demographic rates, we used two datasets belonging to the Guyafor network of forest permanent plots (https://paracou.cirad.fr/).The first dataset is based on eight permanent plots: one 25-ha and seven 6.25-ha plots (plots 9-15) set up between 1991 and 1992 to monitor the functioning and dynamics of both disturbed and undisturbed forests in Paracou (Gourlet-Fleury et al., 2004).Between 1992 and 2015, censuses of all stems with a DBH > 10 cm were conducted every 5 years in the 25-ha plot, and every 2 years in the seven 6.25-ha permanent plots.
The second dataset is from five 1-ha permanent plots established in undisturbed forests in Nouragues, a research station located c. 120 km from Paracou.In these plots, all stems with a DBH > 10 cm have been monitored at 2-4-year intervals since 1993.
To calculate sapling (1-5 cm DBH; see Supporting Information-Fig.S1) and juvenile (5-10 cm DBH; see Supporting Information-Fig.S1) demographic rates, we used 2 datasets: the first dataset belongs to the Guyafor network and is based on c. 750 8-m radii circular subplots established in 12 6.25-ha permanent plots to monitor forest regeneration dynamics in Paracou.Diameter increments and MR of saplings and juveniles have been monitored in these circular subplots at 3-or 5-year intervals between 1992 and 2016.The second dataset is from ten 20 × 250 m transects established in Paracou in 1994 by Molino and Sabatier (2001).In these transects, inventories of diameter growth and MR of saplings and juveniles were conducted for a 7-year period (1995)(1996)(1997)(1998)(1999)(2000)(2001)(2002).
Diameter relative growth rates (RGR, mm•mm −1 y −1 ) were calculated as ln(DBH f /DBH i )/Δt, where DBH f and DBH i refer to final and initial diameters, respectively and Δt is the time in years between the latest and earlier censuses.MR (% y −1 ) were calculated as 100 × (1−(N f /N i )) × 1/t, where N i is the initial number of individuals, N f is the number of survivors and t is the time in years between measurements (Sheil and May 1996).RGR and MR were calculated separately for sapling (1-5 cm DBH), juvenile (5-10 cm DBH) and adult wood (10-50 cm DBH).The number of individuals and census intervals used to calculate species demographic rates at each ontogenetic stage are shown in Supporting Information-Table S1.

Statistical analyses
Trait-demography relationships, for each ontogenetic stage, were examined using pairwise Pearson's correlations (stast library (version 4.4.0),cor function), with species as data points.We also used Pearson's correlations to evaluate traittrait associations, as well as the effects of WE on WSGdemography associations (see above).Growth-mortality, for each ontogenetic stage, was examined using linear regressions (lme4 library (version 1.1-34), lm function).All statistical analyses were conducted in R version 4.3.0(R Development Core Team, 2019).

Results
Overall, there was substantial variation in mean trait values both among species and ontogenetic stages.Most species mean trait values were higher in adults than in juveniles or sapling wood, except for fibre fraction (F F ) and vessel number (V N ), which were higher in sapling wood.Overall, the ranges of variation in wood traits were higher for saplings than for adult or juvenile wood.Species mean RGR was higher for juveniles, while MR was higher for saplings (Table 2).

Effect of WE on WSG-demography associations
The concentration (i.e. % wood dry mass) of WE, in both heartwood and sapwood, is known for five of our study species (i.e.Bocoa prouacensis, Virola michelii, Sextonia rubra, Bagassa guianensis and Dicorynia guianensis).In these species, WE concentrations are low (<7%), and the differences between heartwood and sapwood WE concentrations are small (<3%, Rodrigues 2010; Amusant et al. 2014).The Pearson correlations that included WSG REC (see 'Materials and Methods' section) did not differ, in terms of the correlation coefficients or level of significance, from the Pearson correlation that included WSG (data not shown).

Relationships between wood traits and MR
We hypothesized that MR will decrease with stem traits related to wood strength (i.e.WSG, fibre fractions or FWT) and the fractions of either axial (AP) or radial (RP) parenchyma.In line with this expectation, we found that juvenile MR were negatively related to WSG (Fig. 3a; r = −0.71,P < 0.05).This agrees with a number of studies showing a negative relationship between WSG and juvenile MR in closed-canopy tropical forests (e.g.Wright et al. 2010;Phillipson et al. 2014).Different and non-exclusive reasons may explain this finding.First, in the shaded understory, where carbon gain is limited by low light availability, selection should favour denser tissues that increase survival due to reduced mechanical damages from falling debris or lower risk of pathogen attacks (Alvarez-Clare and Kitaima 2007; Kitajima et al. 2012;Reich 2014).For instance, stem tissue density was strongly correlated with the modulus of elasticity and toughness of stems, and was the best predictor of species MR in saplings of eight lowland tropical tree species (Alvarez-Clare and Kitajima 2007).Second, high-WSG species may have lower MR because they can be more tolerant to drought events.For example, at Paracou, high-WSG species tend to have lower xylem turgor loss points than low-WSG species, which may reduce the detrimental effects of water deficits on living cells (Santiago et al. 2018).Third, as both F WT and T PF scale with WSG at the juvenile stage [Supporting Information-Fig.S3b], it is possible that these traits might mediate the negative association between WSG and juvenile MR, as thicker fibres increase stem strength and living parenchyma cells can provide an active response against pathogens (Morris et al. 2016b).Further studies could examine the mediating effects of wood anatomical traits on the links between WSG and plant performance.We also found a significant and positive relationship between vessel lumen area (V A ) and sapling MR (Fig. 2d; r = 0.53, P < 0.05).This finding agrees with that of Osazuwa-Peters et al. (2017), who showed that sapling MR increased with vessel lumen size in a semi-deciduous forest in Panama.According to the 'rare pith hypothesis', wider conduits can have more  interconduit pits and pit membranes and, therefore, should be potentially more vulnerable than narrow conduits to airseeding through pits (Wheeler et al. 2005).Therefore, one possible explanation for this result should be that species with wider conduits have a higher risk of drought-induced xylem embolisms, which may eventually cause mortality.Yet, we ruled out this possibility, as at our study site vessel lumen size was found to be independent from pit traits, and is weakly associated with xylem cavitation (Levionnois et al., 2020).Instead, we suggest that the observed association between V A and MR, at the sapling stage, can reflect allocation and demographic differences between species.Fast-growing species typically have wider conduits because they favour water transport and ultimately growth (e.g.Santiago et al. 2004;Russo et al. 2010;Hietz et al. 2016).Yet, at the same time, species that grow faster tend to have higher MR (e.g.Poorter et al. 2008;Russo et al., 2020).For instance, among our study species, growth rates were traded-off against MR at the sapling stage [Supporting Information-Fig.S2a].Evidence suggests that saplings of fast-growing species have higher MR because they tend to have poorly defended tissues (Alvarez-Clare and Kitajima 2007; Kitajima et al. 2012).It is likely, then, that saplings with wider conduits (i.e.fast-growing saplings) die more often because they are more susceptible to pathogens or herbivores (e.g.Alvarez-Clare and Kitajima 2007; Zhu et al. 2018), rather than as a consequence of hydraulic failure.
In line with our expectation, sapling and juvenile MR were negatively related to radial (F RP; Fig. 2c; r = -0.59,P < 0.05) and axial (F AP ; Fig. 3a; r = −0.59,P < 0.05) parenchyma fractions, respectively.Storage of nutrients and NSC (i.e.starch, soluble sugars and lipids) is thought to be one of the main functions of xylem parenchyma cells (e.g.Morris et al. 2016a).For example, both axial and radial parenchyma fractions have been positively related to the concentrations of NSC (e.g.Plavcová et al. 2015;Godfrey et al., 2019).Therefore, a higher allocation of wood volume to F RP or F AP might enable higher storage of NSC, which, in turn, may increase survival because NSC favour a faster recovery from periods of drought, defoliation or shading (Poorter and Kitajima 2007;Herrera-Ramírez et al., 2021).An alternative, but not mutually exclusive, explanation is that radial and axial parenchyma cells increase survival, as they play a critical role in plant defense by providing an active response against xylem infections and mechanical damages (Morris et al. 2016b).Although F RP and F AP were significantly related to sapling and juvenile MR, parenchyma fractions were unrelated to MR at the adult stage.This may be due to the possibility that other parenchyma traits, such as spatial arrangement and morphology, can have effects on tree functioning, in addition to the cross-sectional fractions measured here.For instance, axial parenchyma associated with vessels (i.e.paratracheal AP) is suggested to be of central importance in hydraulic balance (e.g.Morris et al. 2017;Aritsara et al., 2020), while banded axial parenchyma (i.e.AP arranged in bands) may play a role in limiting the spreading of decay (Morris et al. 2016(Morris et al. , 2019)).Moreover, the morphology of radial parenchyma cells (i.e.uniseriate or multiseriate rays) may affect stem hydraulics and, to some extent, mechanics (e.g.Zheng and Martinez-Cabrera 2013).Thus, future studies should consider measuring the spatial arrangements and classify parenchyma cells into more specific categories.

Relationships between wood traits and RGR
As expected, adult RGR scaled positively with wood K (Fig. 4a, r = 0.47, P < 0.05).This result is in line with fertilization experiments in lowland tropical forests showing that tree growth is strongly limited by soil K availability, although differed from those of Heineman et al. (2016), which reported no association between wood K and RGR.Potassium is suggested to play an important role in wood formation, mainly during cell expansion and osmoregulation (Ache et al., 2009;Fromm 2010).Specifically, K is thought to be a driver of vessel formation (Fromm 2010), and studies using secondary ion mass spectrometry have reported higher concentrations of K in vessels than in other xylem cells (e.g.Langer et al. 2002).As fast-growing species need large vessels to increase xylem hydraulic conductivity and sustain high growth rates (e.g.Hietz et al. 2016), it is likely that they allocate proportionally more K to grow vessels than slow-growing species.
We found that V A was positively associated with adult RGR (Fig. 4b; r = 0.46, P < 0.05).This result is in agreement with our hypothesis and with the results of several other studies (Russo et al. 2010;Fan et al. 2012;Hietz et al. 2016) which indicate that species with wider vessels grow faster.As xylem-specific hydraulic conductivity (Ks) increases exponentially with vessel lumen area, but only linearly with V N (Tyree and Zimmermann 2002), Ks is expected to be significantly higher in species that build large conduits.In turn, Ks tends to be positively related to stomatal conductance and leaf maximum photosynthetic carbon gain (Santiago et al. 2004), which ultimately favours diameter growth rates (Poorter et al. 2010;Russo et al. 2010).We also predicted that RGR would decrease with traits related to stem structural strength (i.e.WSG, F F or F WT ).Consistent with this prediction, we found that sapling RGR was negatively related to WSG (Fig. 2a; r = −0.63,P < 0.05), which agrees with other studies in tropical forests (Poorter et al. 2008;Wright et al. 2010).One main reason to explain this result is the fact that, in sapling wood, species with high WSG have thicker fibre walls (Supporting Information-Fig.S3a), representing higher stem construction costs.
Total parenchyma fractions (F TP ) were negatively related to sapling RGR (Fig. 2b; r = −0.56,P < 0.05).We are aware of only one study that has shown a significant, although positive, link between F TP and RGR (Poorter et al. 2010).Parenchyma cells represent the majority of living cells in wood (Morris et al. 2016), besides living-fibres (Carlquist 2015).Living parenchyma cells, and, in particular, contact cells (i.e. cells having functional connections with vessels), are metabolically highly active (Spicer and Holbrook 2007).Therefore, although living parenchyma have been linked to several simultaneous functions such as storage, defense and transport (e.g.Pfautsch et al. 2015, Morris et al. 2016, 2019), they may also represent a significant maintenance cost for trees.Thus, stem maintenance costs could be mediating the negative relationship between F TP with growth RGR in adult wood (Fig. 4, Table 3).

Changes in trait-demography relationships during tree development
A number of studies have shown that both traits (Hietz et al. 2016;Rungwattana and Hietz 2017) and demographic rates (Wright et al. 2010;Hérault et al. 2011) can vary considerably during tree development.Thus, we expected that the González-Melo et al. -Associations between wood traits and demography during tree development trait-demography relationships may change accordingly.In line with this expectation, we found that WSG was negatively associated with MR only at the juvenile stage (Fig. 2d; r = −0.71,P < 0.05).High-WSG may represent a competitive advantage for individuals growing in the shaded understory (see discussion above), but not for adult trees that have reached the sunlit canopy.At Paracou, adult tree mortality tends to increase during the rainy season when storms are common (Aubry-Kientz et al. 2015; but see Pillet et al. 2018), and is also strongly associated with soil topography and drainage, with higher treefall rates in bottomlands with waterlogged soils where root anchorage is limited (Ferry et al. 2011).It is reasonable to expect, then, that traits determining tree biomechanical stability, such as rooting depth, crown architecture or the presence of buttressed roots may be better predictors of adult tree mortality in Paracou than WSG.
The Pearson correlation coefficients of the growth-trait relationships tended to be higher at the sapling compared to the adult stage (Table 3; Figs 2 and 3).This, together with the fact that adult MR were unrelated to wood traits, suggests that wood traits tended to be weakly related, or unrelated, to demographic rates at later ontogenetic stages.This agrees with previous findings at local (Visser et al. 2016;Osazuwa-Peters et al. 2017) and regional scales (Poorter et al. 2008;Wright et al. 2010).One possible factor explaining this result is the strong gradient in light availability that characterizes our study site (e.g.Baraloto et al. 2005;Laurans et al. 2012).In closed-canopy forests, light availability is an important factor mediating morphological and demographic variations across species, because it limits carbon gain (Wright et al. 2010).In general, saplings growing in the understory experience lower and more variable light availability than adult trees that grow in the canopy or subcanopy (Wright et al. 2010).As a consequence, interspecific differences in traits and demographic rates are expected to be higher in saplings than in adults (e.g.Poorter et al. 2010).Among our study species, the range of variation of most traits, as well as of MR, is higher for saplings than for juveniles or adults (Table 2).This may have favoured the detection of stronger relationships between traits and demographic rates at early ontogenetic stages.Another factor that may contribute to the observed changes in the strength of the trait-growth relationships during tree development may be related to size-related shifts in growth rates.For instance, for many tree species in Paracou, growth rates are lower at larger than at intermediate stem diameters (Hérault et al. 2011), possibly due to a shift in resource allocation from diameter growth to reproduction (Thomas 1996).
A large body of literature has shown that WSG is in general a good proxy of growth and MR (e.g.Poorter et al. 2008;Wright et al. 2010;Hérault et al. 2011).Yet, our results indicate that the associations between WSG and demographic rates are not always supported for different ontogenetic stages.For instance, at the sapling stage, WSG was negatively related to RGR, but unrelated to sapling MR; while the opposite trend was found at the juvenile stage.These findings add to recent studies showing that demographic rates may be better explained in some cases by wood anatomical (Poorter et al. 2010;Russo et al. 2010;Fan et al. 2012;Osazuwa-Peters et al. 2017) or chemical traits (Martin et al. 2014) than by WSG.This may be due to the fact that these traits are more closely related to performance than WSG.For example, vessel lumen area is more directly related to water transport capacity, and consequently to growth rates, than WSG (e.g.Fan et al. 2012).Furthermore, species with similar values of WSG may be functionally different, because similar values of WSG may be product of different wood anatomies (Zieminska et al. 2015).
Finally, it is important to note that while our study species reflect a wide spectrum of ecological strategies and wood structures, the number of species we sampled is relatively low, which might limit the generalizability of our findings.Thus, we suggest that further studies should examine whether the trait-demography associations reported in this study are consistent for larger sets of species and for different forest types as well.

Conclusions
In this study, we examined trait-demography associations of sapling, juvenile and adult wood of 19 tree species from eastern Amazonia.We found that, in general, demographic rates of saplings, juveniles and adults were associated with different wood traits.We also showed that the strength of the trait-demography associations tended to decrease at later ontogenetic stages.Overall, our results support the growing evidence that the effects of traits on species demographic rates can change during tree development.Moreover, our findings indicate that WSG may not always be a good proxy of growth and mortality, particularly when traits such as wood chemical or anatomical traits are considered.These findings also support the general expectation that traits can be strongly related to species demographic rates and, hence, that traits are important to increase our understanding of life-history variations and community dynamics in lowland tropical forests.

Figure 1 .
Figure 1.Conceptual framework showing theoretical and realized associations among wood traits and demographic rates.Black and white arrows represent positive and negative associations, respectively.Continuous arrows are theoretical associations supported by our results (see Figs 2, 3 and 4 for details), while dashed arrows are hypothesized, although nor realized, links between traits and demographics rates.

Figure 2 .
Figure 2. Significant pairwise relationships between demographic rates and wood traits at the sapling stage.A) WSG, B) total parenchyma fractions, C) radial parenchyma fractions and D) vessel lumen area.Points represent species means.Black and solid lines represent significant (P < 0.05) relationships.P values and Pearson correlation coefficients (r) are shown for each pairwise relationship.

Figure 3 .
Figure 3. Significant pairwise relationships between MR and wood traits at the juvenile stage.A) WSG, and B) A PF .Black and solid lines represent significant (P < 0.05) relationships.Points represent species means.P values and Pearson correlation coefficients (r) are shown for each pairwise relationship.

Figure 4 .
Figure 4. Significant pairwise relationships between growth rates and wood traits at the adult stage.A) wood potassium concentrations and B) vessel lumen area.Black and solid lines represent significant (P < 0.05) relationships.Points represent species means.P values and Pearson correlation coefficients (r) are shown for each pairwise relationship.

Table 2 .
Summary characteristics of wood traits measured on sapling, juvenile and adult wood for 19 tree species from a lowland tropical forest in eastern Amazonia.Mean, standard deviation (SD) and range of variation are shown.González-Melo et al.-Associations between wood traits and demography during tree development

Table 3 .
Pairwise Pearson correlations between wood traits and demographic rates for sapling, juvenile and adult wood.See Table2for trait abbreviations.Significant correlations (P < 0.05) are shown in bold.
Note: Multiple comparisons corrections were not accounted for.