Radial and vertical variation of wood nutrients in Bornean tropical forest trees

Nitrogen, phosphorus, potassium, calcium, and magnesium concentrations in woody tissue are poorly documented, but are necessary for understanding whole‐tree nutrient use and storage. Here, we report how wood macronutrient concentrations vary radially and along the length of a tree for 10 tropical tree species in Sabah, Malaysia. Bark nutrient concentrations were consistently high: 2.9–13.7 times greater than heartwood depending on the nutrient. In contrast, within the wood both the radial (sapwood vs. heartwood) and vertical (trunk bottom vs. trunk middle) variation was modest. Higher concentrations in sapwood relative to heartwood provide empirical support for wood nutrient resorption during sapwood senescence. Dipterocarp species showed resorption rates of 25.3 ± 7.1% (nitrogen), 62.7 ± 11.9% (phosphorus), and 56.2 ± 12.5% (potassium), respectively, while non‐dipterocarp species showed no evidence of nutrient resorption in wood. This suggests that while dipterocarps have lower wood nutrient concentrations, this family is able to compensate for this by using wood nutrient resorption as an efficient nutrient conservation mechanism. In contrast to other nutrients, calcium and magnesium tended to accumulate in heartwood. Wood density (WD) showed little vertical variation along the trunk. Across the species (WD range of 0.33 to 0.94 mg/cm3), WD was negatively correlated with wood P and K concentration and positively correlated with wood Ca concentration. As our study showed exceptionally high nutrient concentrations in the bark, debarking and leaving the bark of the harvested trees on site during logging operations could substantially contribute to maintaining nutrients within forest ecosystems.

While great attention has been given by ecologists to the link between interspecific variation in leaf traits and patterns of N and P cycling both within and across forest types (Chapin, 1980;Ordoñez et al., 2009;Reich et al., 1992Reich et al., , 1997, interspecific variation in wood chemical traits for nutrient dynamics has historically not been well studied (Vitousek & Sanford, 1986). Wood chemical traits have been hypothesized to covary along the "wood economics spectrum" (WES; Chave et al., 2009) with wood density, which could represent an easily measurable proxy for life-history strategy among co-occurring tree species (Wright et al., 2003). Such covariation can be expected when considering a functional tradeoff between "storage/defense" and "growth" as a selective pressure that drives interspecific variation in wood chemistry (Martin et al., 2014). However, the conceptual framework of WES acknowledges a near complete lack of information on species-specific wood chemical traits such as nutrient concentrations among trees (Chave et al., 2009;Martin et al., 2013;Martin et al., 2014). In addition to widely quantified foliar nutrient concentrations, which empirically support the conceptual framework of leaf economics spectrum (LES; Wright et al., 2004), direct measurements of wood nutrient concentrations are necessary to understand the whole-tree nutrient use of tropical tree species and its implication for ecosystem processes (Heineman et al., 2016). Furthermore, as early nutrient studies suggest that tropical forests should not be categorized as a single entity (Proctor, 1987;Vitousek & Sanford, 1986), the spatial heterogeneity of wood nutrient status is expected across the tropics.
There is likely to be broad variation in anatomical, chemical, physical and mechanical properties, as well as nutrient concentrations in different plant organs (Lachenbruch et al., 2011;Meerts, 2002).
While wood is usually considered to have the lowest nutrient concentrations among all organs (Tsutsumi et al., 1968), wood nutrient concentrations are expected to vary substantially within a tree (Hillis, 1987) and overall wood nutrient stocks may be a substantial component of ecosystem nutrient reservoirs (Bauters et al., 2022;Bond, 2010). Wood accumulates nutrient capital more slowly than foliage (Johnson et al., 2001), but serves as a long-term, slow turnover C and nutrient pool and store (Chapin et al., 1990;Poorter & Kitajima, 2007). Nutrient storage in woody biomass may be functionally important, particularly in highly weathered soils typical for tropical forests (Grau et al., 2017;Heineman et al., 2016).
It is also necessary to consider differences in nutrient contents between heartwood and sapwood to quantify nutrient budgets and fluxes for individual trees and forest stands (Augusto et al., 2000;Heineman et al., 2016;Schilling et al., 2015;Turner & Lambert, 1983). Sapwood and heartwood are anatomically different. Inner sapwood rings are eventually converted to heartwood (duramen), which no longer contains living cells and often has vessels blocked with tyloses and accumulates secondary compounds (Bamber & Fukazawa, 1985;Hillis, 1987). Nutrient resorption occurs during the transition of sapwood to heartwood, as indicated by a meta-analysis that shows substantially lower nutrient concentration in heartwood than sapwood (Meerts, 2002). A recent study in Panama also indicates P resorption as inner annulus have lower P than outer annulus (Heineman et al., 2016). Analyses on how wood nutrient concentrations vary across woody components including branches, roots, sapwood, and heartwood are especially limited in the tropics (Meerts, 2002;Nakagawa et al., 2016).
Furthermore, bark chemistry has not been considered in comparative studies of bark traits . Bark is a morphologically diverse and functionally important component of the stem (Romero, 2014), its functions including pathogen and fire protection (Paine et al., 2010;Poorter et al., 2014), corticular photosynthesis (Pfanz, 2007;Rosell et al., 2015), water and photosynthate storage, and mechanical support (Rosell & Olson, 2014). Bark plays a vital role in nutrient dynamics, maintaining higher concentrations of nutrients than other woody tissues and acting as a protein store (Jenkins et al., 2003;Wetzel & Greenwood, 1989).
Wood density (WD) is a key trait for life-history variation of tree species, because it represents C allocation to stem tissue, and tends to correlate with growth and mortality (Enquist et al., 1999;Muller-Landau, 2004;Preston et al., 2006;Swenson & Enquist, 2008;van Gelder et al., 2006). While WD positively correlated with wood N for tree species in Panama (Martin et al., 2014), no relationship was found between WD and wood N content in Uganda (Becker et al., 2012).
The aim of this study is to examine how nutrient concentrations and WD vary within trunk cross-section and along the length of the tree in a Bornean tropical forest, and to assess inter-and intra-specific patterns. Previous direct measurements of macronutrients are entirely based on extracted wood cores (Feldpausch et al., 2004;Heineman et al., 2016;Johnson et al., 2001). Ours is therefore the first study of macronutrient distribution across the woody components of tropical trees, considering both vertical and radial variation. To our knowledge, these are also the first estimates for all five macronutrients (N, P, K, Ca, Mg) in coarse roots in tropical forests.
We address the following research questions: Q1) How do wood nutrient concentrations vary within a trunk cross-section (radially) from bark to sapwood to heartwood? Q2) Is there evidence for nutrient resorption or accumulation in wood during sapwood senescence? Q3) How do wood nutrient concentrations vary along the length of the tree (vertically)? Q4) How does wood density vary vertically, and how does it correlate with wood nutrient concentrations? 2 | ME THODS

| Site description
The study site is located in the state of Sabah, Malaysian Borneo (4°43′ N, 117°35′ E). The vegetation is a mixed dipterocarp lowland rainforest, which is the most extensive forest type in Malaysian Borneo (Palmiotto et al., 2004). The family Dipterocarpaceae is the most abundant with ten genera and about 270 species described (Burghouts, 1993). Soils are acrisols and luvisols (WRB classification), developed over the underlying bedrock of predominantly mudstone and sandstone (Acres, 1975). The climate is aseasonal with mean annual temperature of ca. 27.0°C and annual precipitation of 3500 mm (Imai et al., 2012). The study is part of the Stability of Altered Forest Ecosystems (SAFE) Project, one of the world's largest experiments in human-modified forest landscapes, which examines the impacts of logging and forest conversion to oil palm on biodiversity, ecosystem functioning and productivity (Ewers et al., 2011). The SAFE site has previously been logged at different intensities (Struebig et al., 2013).
The study took place in a logging site where direct felling was being conducted within the Benta Wawasan (concession holders) area. The trees sampled for the study were felled as part of concessional logging operations due to forest conversion to oil palm plantation under the Yayasan Sabah (landowners) Forest Concession.
Soil characteristics in the logging sites were not measured, but were available from a nearby plot on a similar soil type (Table S1).

| Sampling of tree components
Tree components were sampled in August 2014. Thirty trees ≥10 cm diameter at breast height (DBH), which consisted of 10 tree species with three replicates were selected in the study site (Table 1)

| Selection of tree species
To characterize the species composition in the study area, tree censuses were conducted during 2011 and 2012, and 1580 stems (≥10 cm DBH) of 405 species were identified in five 1-ha selectively logged plots at the SAFE Project area (Both et al., 2019;Riutta et al., 2018). The species sampled for this study represented 25% of total number of individual stems, including the most abundant species Macaranga pearsonii, and some of the dominant species: Parashorea malaanonan and Neolamarckia cadamba.
Dipterocarpaceae is the most dominant family in Borneo (Palmiotto et al., 2004), accounting for 19% of the stems in the study area, and a half of our sampled trees belongs to this family. This family is also favored for timber. Macaranga pearsonii, Neolamarckia cadamba, and Trema orientalis are considered as pioneer species (Table 1) (Davies & Semui, 2006;Saner et al., 2012). Hence, we sampled a broad taxonomic range from light-demanding pioneer to shade-tolerant late-successional species. Although in most cases the three replicates of each species were of similar size (Table 1), standardizing the DBH size-class within replicates was not feasible due to high species diversity in the study site.
Sampled trees had no major visible damage (e.g., no insect damage, chlorotic leaves, and bent stems). All measurements (the number of branches, diameters, and length of all sampled locations) were conducted using a measuring tape within hours after the sampled trees were felled. Directly following felling, sampled woody components were cut using a chainsaw. The samples were brought to TA B L E 1 The list of sampled tree species grouped by successional types (n = 30; 10 species with three replicates).  Figure 1b).
Bark is defined as all tissues external to the vascular cambium.
The analyzed bark layer comprised outer bark which was composed of dead cells known as periderm, and inner living bark including secondary phloem, cortex, and phelloderm (Evert et al., 2006;Rosell et al., 2015). The vascular cambium was often only one or two cells thick, and required a microscope to see it well.
The outermost wood layers containing living cells (xylem) are referred to as sapwood, whereas the inner wood layers containing no living cells are referred to as heartwood (Bamber & Fukazawa, 1985;Hillis, 1987). The collected trunk disk was divided into sapwood and heartwood based on a distinguishable color difference after mechanical sanding. If there was no clear color difference, the section of the rim of the disk was used as a sapwood sample and the wood around the pith as a heartwood sample ( Figure 1b). Due to the absence of color boundary between sapwood and heartwood in most branch and coarse root samples, they were disaggregated into wood and bark only. The twig tips of sampled branches (less than 2 cm diameter) were not debarked.

| Chemical analyses
Woody samples for the chemical analysis totaled 420 samples (10 species × 3 replicates × 14 sampled locations within a tree). Most of the laboratory analysis protocols are described in more detail in Majalap and Chu (1992), and a summary of the methods used is as follows. The procedures included both references samples and procedural blanks. Each sample was digested following the sulfuric acid-hydrogen peroxide-lithium sulphate digest procedure for vegetation described in Allen (1989). Phosphorus in the digest was determined using the molybdenum-blue method described in Anderson and Ingram (1993) and read at 880 nm on a spectrophotometer (HITACHI UV-VIS), while K, Ca, and Mg contents were measured on an atomic absorption spectrophotometer (GBC Scientific Equipment). Total C and N contents (as total element contents)

Concentrations
were determined by a dry combustion method at 900°C using an Elementar Vario Max CN analyzer (Elementar Analysensysteme).

| Estimation of wood density (WD)
Wood density was measured by following the procedure ISO 3131:1975 (E) Wood -Determination of Density for Physical and Mechanical Tests. Wood density was defined as the oven-dry mass per unit volume of green wood cube (tangent 20 mm × axial 20 mm × radial 20 mm length), given in g cm −3 . Dry mass was measured after oven drying at 103 ± 2°C until the dry mass was stabilized.

| Estimation of wood nutrient resorption rate
Wood nutrient (x) resorption rate (%) was computed for each tree as where x is an element of either N, P, K, Ca, Mg, or C, xR is the x resorption rate, x sapwood is the x concentration in sapwood, x heartwood is the x concentration in heartwood.

| Statistical analyses
For each measured element, we used linear mixed-effects models to test whether wood nutrient concentration varied radially or vertically. We created separate models for each nutrient, and for radial and vertical variation. In each model, the response variable was the nutrient concentration. In the radial models, the fixed effect was the radial location (bark, sapwood, or heartwood), with separate models run for trunk bottom and trunk middle. In the vertical models, the fixed effect was the vertical location, with separate models for bark, sapwood, and heartwood. For bark and sapwood, the samples spanned the whole length of the tree from coarse root, trunk bottom, trunk middle, branch bottom to branch middle, while for heartwood there was only the comparison between trunk bottom and trunk middle (the other components did not have distinct heartwood). In all models, the tree species and the individuals within each species were included as nested random effects. Nutrient concentrations were log-transformed to meet the assumption of normality of errors. We also used linear mixed-effects models to test whether wood nutrient resorption varied between tree species groups (dipterocarps, other non-pioneers and pioneers, Table 1). We created a separate model for each nutrient. In these models, the response variable was the resorption rate, the fixed effect was the species group and random effect was tree species. Tukey HSD post hoc test was used to determine which pairs differed significantly, and the Bonferroni correction was performed to account for multiple comparisons.
We calculated Pearson correlation coefficients for the six elements × five organs combinations (bark, sapwood, and heartwood at trunk bottom, branch wood at branch middle, and coarse root wood), using the species' means (n = 10). We also examined the covariation among nutrients in sapwood at trunk bottom and WD with principal component analysis (PCA; vegan package, Oksanen et al., 2020). We used a major axis regression to examine the relationship between WD at the trunk bottom and trunk middle. We also examined the correlation between WD and nutrient concentrations in the sapwood and heartwood using Pearson correlation coefficients, both with untransformed and log-log transformed data. Statistical analyses were performed in the R programming language version 4.0.4 (R Core Team, 2021) and the packages "ggplot2" (Wickham, 2016), "lme4" (Bates et al., 2015), "smatr" (Warton et al., 2012), and "vegan" (Oksanen et al., 2020).

| Variation in nutrient concentrations between bark and wood
Bark had substantially higher nutrient concentrations than sapwood or heartwood (Figure 2, Figure S3,  (Figure 3, Table S4).
The dipterocarp family had particularly high nutrient concentrations in bark ( Figure 2). Shorea micans had among the highest N ratio between bark and heartwood at the bottom of the trunk (25.47 ± 4.56). Shorea johorensis showed among the highest P ratio between bark and heartwood (33.57 ± 14.53). Shorea johorensis also had exceptionally high K ratio between bark and heartwood (56.43 ± 30.13). While Trema orientalis, a pioneer species, had higher Ca ratio between bark and heartwood (11.68 ± 0.53), Ca and Mg ratios were generally more constrained across species. Across elements, intra-specific variation was considerable.

| Wood nutrient resorption from heartwood to sapwood
The mean nutrient concentrations across species indicated no significant difference between sapwood and heartwood ( Table 2). The ratio of sapwood to heartwood nutrient concentration varied from below to above one, depending on the nutrient and tree species ( Figure 3 for results in trunk bottom; Figure S1 for both trunk bottom and trunk middle). The former pattern suggests accumulation in heartwood, whereas the latter indicates wood nutrient resorption during sapwood senescence (Bamber & Fukazawa, 1985;Hillis, 1987;Saka & Mimori, 1994;Smith & Shortle, 1996). The difference in the nutrient concentrations between sapwood and heartwood was larger for the dipterocarp species than for the other species for N, P, and K (Figure 3, Figure S1). While dipterocarps had consistently higher nutrient concentrations in sapwood than in heartwood, the other species showed either smaller differences or in several cases lower concentrations in sapwood.
A difference between sapwood and heartwood by dipterocarp species suggested wood nutrient resorption at the rates of 25.3 ± 7.1% (N), 62.7 ± 11.9% (P), and 56.2 ± 12.5% (K), respectively, at the bottom of trunk (Table S4), which was significantly higher than the resorption rate of the non-dipterocarps species for N (p = .011) and P (p = .005), and close to significant for K (p = 0.057). At the middle of trunk, the resorption rates in dipterocarps declined compared to the bottom of trunk at 10.7 ± 13.6% (N), 41.9 ± 13.0% (P), and 20.4 ± 15.2% (K), respectively. In contrast, Ca and Mg tended to accumulate in heartwood in 9 out of 10 species (Ca), and all species (Mg), with no difference between dipterocarps and non-dipterocarps (p = .265 and p = .080 for Ca and Mg, respectively).  Table 1). The triangle indicates total mean across species (n = 10). Note the different scales and log transformation of the y-axes.

TA B L E 2
Mean ratio (±1 SE) between nutrient and carbon concentrations in bark, sapwood, and heartwood in the trunk bottom (TB) and trunk middle (TM); the first component divided by the second component in each case. The values are averages of the ten studied tree species (see Table 1). The values in bold indicated significant differences (p < .05), that is, they are significantly different from the expectation of a 1:1 ratio for each nutrient, tested using linear mixed-effects models with the Tukey HSD post hoc test. The p-values were Bonferroni corrected to account for multiple comparisons.

| Vertical variation in nutrient concentrations
There was no significant variation in nutrient concentrations between trunk bottom and trunk middle in bark, sapwood, or heartwood for most of the nutrients (Tables S2 and S3). The exception was bark P concentration, which was lower in the trunk bottom (p = .005).
Across the whole length of the tree, from coarse roots through trunk sapwood to branch bottom and branch middle, there was a trend in wood nutrients, with the lowest concentrations in the trunk, increasing both towards coarse roots and branches, with significant differences found for N, P, and C concentrations (Figure 3, Table S2).
Branch bottom had exceptionally high N, P, and K relative to other components. However, this vertical pattern varied in magnitude among species: While there was extraordinarily large vertical variation in the dipterocarp family for N and P (green symbols in Figure 3), with significantly higher concentration at branch bottom compared to other components, other species showed nearly no vertical variation. Vertical variation was modest (not significant) for K, Ca, and Mg across species (Table S2). Note that this comparison was done using trunk sapwood, as we consider coarse roots and branches to consist of mostly sapwood.
In contrast to the vertical variation in wood nutrient concentration, there were significant differences between trunk bark, branch bark, and coarse root bark in P, K, and Ca ( Figure S3, Table S3).
Phosphorus and K concentrations were highest in branches, intermediate in the trunk, and lowest in coarse roots, while Ca concentration peaked at branch bottom, decreasing towards branch middle and towards trunk and coarse roots.

| Wood density (WD)
There was high variability in WD among the studied species: Shorea micans had exceptionally high WD: 0.94 ± 0.01 g/cm 3 in trunk bottom and 0.83 ± 0.02 g/cm 3 in trunk middle, respectively (Figure 4).
Macaranga pearsonii had the lowest density of 0.33 ± 0.01 g/cm 3 in trunk bottom and 0.30 ± 0.01 g/cm 3 in trunk middle, respectively.
There were no significant differences in the mean WD between species groups of dipterocarps, other non-pioneers or pioneers  Table 1). The triangle indicates total mean across species (n = 10). Note the different scales and log transformation of the y-axes.
(p = .437). Major Axis Regression indicated little or no difference between WD in trunk bottom and trunk middle, as the 95% confidence intervals (0.80 and 1.20) of the slope (0.90) included zero (Figure 4).
Across the species, WD was negatively correlated with wood P and K concentration and positively correlated with wood Ca concentration (Table 3). These correlations were found in sapwood at both trunk bottom and trunk middle (P, K, and Ca) and in heartwood at the trunk middle (P and Ca), but not in heartwood at trunk bottom.

| Covariation across nutrients and organs
There were many significant correlations (Pearson correlation, p < .05) across organs for the same nutrient (e.g., sapwood and heartwood nutrient concentrations had a significant positive correlation for each nutrient, except Ca; Figures S4 and S5), and across different nutrients for the same organs (e.g., positive correlation between branch N and branch P, and between bark P and bark K; Figures S4 and S6). While the strongest correlations tended to be either for the same nutrient or the same organ, there were also several significant correlations for different organ × nutrient combinations, such as a positive correlation between branch Mg and bark K, and between branch K and sapwood N. All significant correlations (not corrected for multiple testing) are reported in Figure S4. In this dataset, branch K and sapwood K had the highest number of significant correlations with other organ × nutrient combinations.
Major gradients in the data were visualized with the principal component analysis ( Figure 5). Together, the first two PCA axes explained 62% of the variation in the data. The strongest gradients F I G U R E 4 Linear relationship of wood density between trunk bottom and trunk middle. We sampled ten species (grouped into dipterocarps, other non-pioneers and pioneers) with three replicate trees. Species codes are listed in Table 1. Each data point is the mean (±SE) of an individual tree with two to five samples taken per tree. The continuous line shows the fit of the major axis regression, and the dashed line is the 1:1 relationship.

TA B L E 3
Pearson correlation coefficients for the correlations between wood density and nutrient and carbon concentrations in sapwood and heartwood in trunk bottom (TB) and trunk middle (each species is a data point, n = 10). Note: Coefficients with p < .05 are denoted with ** and coefficients with p < .10 are denoted with *. Non-significant (n.s.) coefficients (p ≥ .10) are labeled as '(n.s)'. Correlations based on log-log transformed data are shown in italics, we report the strongest correlation in each case, also for the non-significant correlations.
were related to Ca and WD (highest loadings on the first axis) and to C, N, and P (highest loadings on the second axis).

| Bark
Bark is known to host higher concentrations of C and N compared to other woody tissues (Jenkins et al., 2003;Wetzel & Greenwood, 1989), and our results present the same pattern in P, K, Ca, and Mg. The high nutrient concentrations in bark may be attributed to inner bark acting as nutrient storage (Romero, 2014;Scholz et al., 2007). For example, temperate tree species store N, derived from the breakdown of leaf proteins, in the bark during winter to support new growth in spring (Millard, 1996). Tropical tree species have particularly large proportion of inner bark (Rosell, 2016), indicating a large nutrient storage capacity (Rosell & Olson, 2014). Our results showed that bark nutrient concentrations were particularly high in branches; this might be due to the fact that branches have a higher proportion of this nutrient rich inner bark compared to the trunk.
We found that Ca concentrations in bark, in particular, were consistently high relative to sapwood and heartwood for all species. High Ca concentration in bark has also been reported in tropical tree species (Jones et al., 2019;Wang et al., 1991) and in temperate tree species (Österås & Greger, 2006;Trockenbrodt, 1995). This may be associated with cambial activity and xylem development (Fromm, 2010). Excess Ca ions can be deposited as Ca oxalate crystals in the bark or phloem, the cell walls of which serve as a Ca reservoir (Fromm, 2010).
Calcium cycling in forests may be susceptible to biomass removal and disturbance, as Ca accumulates in woody biomass (McGrath et al., 2001). Repeated timber harvesting causes substantial losses of ecosystem Ca due to removed Ca-rich woody biomass from the site (Brouwer & Riezebos, 1998;Huntington et al., 2000;Spangenberg et al., 1996;Zech & Drechsel, 1998). In Sabah, the loss of Ca via leaching exceeds the annual input of Ca via wet deposition (Grip et al., 1994), which is below 2 kg ha −1 year −1 (Nykvist, 2000). The harvesting of stem wood and bark results in the loss of 19% of the total Ca in the ecosystem and soil to −50 cm depth in Sabah (Nykvist, 1997). Sustainable forestry is not possible unless liming, fertilization or wood ash is applied (Nykvist, 2000).
As several studies report Ca deficiency after logging in Sabah and elsewhere (Brouwer & Riezebos, 1998;Grip et al., 1994;Huntington et al., 2000;Nykvist, 1997Nykvist, , 1998Nykvist, , 2000Spangenberg et al., 1996;Zech & Drechsel, 1998), we suggest that the debarking of the extracted logs on site, leaving the bark behind, could be one of the nutrient conservation practices for foresters and conservationists.

| Sapwood, heartwood, and wood nutrient resorption
Dipterocarps showed substantial decreases in N, P, and K concentrations in heartwood relative to sapwood, while most non-dipterocarps F I G U R E 5 Principal component analysis of the variation in wood density (WD) and in nutrient concentrations in sapwood at the bottom of the trunk. The embedded table shows the loadings of the variables on the first three axes. The tree species are displayed based on their mean scores, with dipterocarps in green, other non-pioneers in gray and pioneers in blue (see Table 1 for species abbreviations). See Table S5 for the pair-wise correlations of the six elements × five organs.
did not (Figure 3). These dipterocarp results are consistent with a global meta-analysis of 71 angiosperm species by Meerts (2002) showing that N, P, and K concentrations are generally lower in heartwood. These higher concentrations in sapwood, however, may originate either from the structural wood tissue or the nutrients in the liquid component of the xylem, which might have been retained in the samples during the drying process or both.
For dipterocarps, our results provide empirical support for a hypothesis that nutrient resorption from woody stems occurs during the conversion of sapwood into heartwood. This may indicate an efficient nutrient conservation mechanism of N, P, and K in dipterocarps, which overall have low concentrations in wood. Although there are no direct studies on wood nutrient resorption in tropical forests, resorption of P from sapwood appears to be predominant in Panama as wood P was lower in the inner annulus than the outer annulus (Heineman et al., 2016). In the analysis of 40 subarctic vascular species, resorption by fine stems (<3 mm diameter) is also substantial for N and P (Freschet et al., 2010).
In contrast, we found that Ca and Mg showed either no resorption or accumulation in heartwood ( Figure 3). Such results may reflect Ca accumulation in the form of oxalate crystals (Hillis, 1987). Calcium is immobile in the phloem and is mainly accumulated in the cell walls, becoming unavailable for metabolic processes in the symplast (Fromm, 2010). Frangi et al. (2005) also report that Ca and Mg in sapwood tend to be low in boreal montane forests, whereas the global meta-analysis shows the complex pattern of these cations, which could be explained by phylogenetic constraints (Meerts, 2002).

| Vertical variation
The magnitude of wood mean nutrient concentrations decreased following the order, N > Ca > K > Mg > P, which was consistent with the pattern in a recent study in Panama (Heineman et al., 2016). For most species, there was negligible variation along the length of the main trunk, suggesting that wood sample collection near the bottom of the tree is usually representative of entire main trunk, where the bulk of woody biomass is found.
Branches, however, tended to show higher nutrient concentrations than the main trunk ( Figure S2). Sampling of outer branches may be in some cases easier for wood trait analyses than sampling of main trunk (e.g., when leaf traits are being collected). Given the high variability in nutrient concentrations along the length of the tree, particularly among dipterocarp species (green symbols in Figure S2), our findings suggest that it is not straightforward to accurately estimate whole-tree nutrient concentrations in wood based on branch samples.

| Nutrient covariation and wood density
We found a negative correlation between WD and P, and also WD and K. In a previous study in Panama, there was a also a negative correlation with P, but no relationship with K or Ca (Heineman et al., 2016). We did not find a relationship between WD and N; previous results from tropical forests are mixed, with a positive correlation found in Panama (Martin et al., 2014) and no relationship in Uganda (Becker et al., 2012). These inconsistent findings may indicate that these relationships vary biogeographically and maybe also at regional and local scales.
Interspecific variation in wood chemical traits for nutrient dynamics has historically not been well studied (e.g., Martin et al., 2014;Vitousek & Sanford, 1986), although such data would be essential for further developing the WES framework. In addition, WD needs to be quantified in combination with wood nutrients to better understand plant functional traits in tropical forests (Heineman et al., 2016). To our best knowledge, this dataset is the most detailed description of nutrient and WD covariation in SE Asian tropical forests to date. In this dataset, the strongest gradients form a "structural stability axis" (highest loadings of the first PCA axis being WD and wood Ca concentration; Figure 5) and a "nutrient richness axis" (highest loadings of the second PCA axis being wood C in one direction and N and P concentration in the opposite direction). Across a soil fertility gradient, which our study did not have, it has been suggested that tree species with high WD and low wood P concentration are associated with low fertility soils (Heineman et al., 2016).
Our dataset did not have a sufficient DBH range within species (Table 1) Feldpausch et al. (2004) report that wood concentrations of N and P decrease with secondary forest age in Amazonia, with an average reduction of 50% (N) and 60% (P) from the youngest to the oldest forests. However, wood core measurements in that study included bark, which may lead to overestimation in young stands (Feldpausch et al., 2004), because saplings have a higher proportion of bark than mature trees.

| CON CLUS ION
To the best of our knowledge, this study is the first to examine in detail the distribution of macronutrients across woody components within whole tropical trees. Since the only available published meta-analysis for wood nutrient concentrations to date is deficient in tropical species data (Meerts, 2002), our study provides essential information, considering the extensive number of samples: 420 woody components of 30 trees for the observation of six elements.
Our results provide empirical support for wood nutrient resorption during sapwood senescence, but only for the dipterocarp family.
This suggests an efficient conservation mechanism of N, P, and K, and that generally lower wood nutrient concentrations in dipterocarps are largely compensated by wood nutrient resorption. The present study focused on Sabah, Malaysia; however, further sites are needed across the tropics, for example, within the Global Ecosystem Monitoring (GEM) network using a standardized protocol to cover the biogeographical variation in wood nutrients. As a practical implication of the study, due to the high calcium concentration in bark, and calcium deficiency being typical in logged forests, we recommend debarking the logs and leaving the bark on site, to conserve this nutrient.

ACK N OWLED G M ENTS
This paper is a product of the Global Ecosystems Monitoring (

CO N FLI C T O F I NTER E S T S TATEM ENT
The corresponding author confirms on behalf of all authors that there have been no involvements that might raise the question of bias in the work reported or in the conclusions, implications, or opinions stated.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are openly available in the SAFE Project repository, which is part of Zenodo (https://zenodo.org/commu nitie s/safe) at https://zenodo.org/recor d/8158811