Parenchyma fractions drive the storage capacity of nonstructural carbohydrates across a broad range of tree species

Premise: Nonstructural carbohydrates (NSCs) play a key role in tree performance and functioning and are stored in radial and axial parenchyma (RAP) cells. Whether this relationship is altered among species and climates or is linked to functional traits describing xylem structure (wood density) and tree stature is not known. Methods: In a systematic review, we collated data for NSC content and the proportion of RAP in stems for 68 tree species. To examine the relationships of NSCs and RAP with climatic factors and other functional traits, we also collected climatic data at each tree's location, as well as wood density and maximum height. A phylogenetic tree was constructed to examine the in ﬂ uence of species' evolutionary relationships on the associations among NSCs, RAP, and functional traits. Results: Across all 68 tree species, NSCs were positively correlated with RAP and mean annual temperature, but relationships were only weakly signi ﬁ cant in temperate species and

Nonstructural carbohydrates (NSCs) are essential substrates for metabolic processes such as maintenance respiration, osmoregulation, growth, and defense (Kozlowski, 1992;Morin et al., 2007;Sala et al., 2012). The dynamics of NSCs are considered as indicators of carbon source-sink balance in trees (Körner, 2003;Mei et al., 2015) and have major consequences for downstream processes such as mycorrhizal activity (Schiestl-Aalto et al., 2019). Generally, carbon fixed during photosynthesis is transported as NSCs, which mainly comprise soluble sugars and starch (Hoch et al., 2003). Soluble sugars are mostly involved in transport and various immediate functions, while starch is stored in different plant organs for future remobilization, allowing trees to maintain functionality when carbon demand is higher than supply (e.g., under severe drought stress) (Anderegg and Anderegg, 2013;Wiley et al., 2013;Klein and Hoch, 2015;Furze et al., 2019). Drought-related tree mortality has been linked to NSC starvation (McDowell et al., 2008;Galiano et al., 2011;Chuste et al., 2020), although evidence points to hydraulic failure as the main trigger of mortality (Rowland et al., 2015;Adams et al., 2017). Nonstructural carbohydrates also play an important role in drought resistance as sources of solutes for osmoregulation and turgor maintenance (O'Brien et al., 2014;Sapes et al., 2019Sapes et al., , 2021. Therefore, an understanding of potential factors that can affect the size of the NSC storage pool is essential for better evaluating the role of NSC in tree physiological processes, especially across a broad range of species and climates. Nonstructural carbohydrates are stored in vegetative tissues in the form of soluble sugars and starch and, besides in storage organs such as rhizomes and stolons, are usually most abundant in actively growing tissues such as leaves, fine roots, and sapwood (Hartmann and Trumbore, 2016;Ramírez et al., 2018;Wang et al., 2018). Within the sapwood, NSCs are mostly stored in radial and axial parenchyma (RAP) cells (Spicer, 2014;Plavcová and Jansen, 2016). Stem parenchyma cells are produced by ray and fusiform initials of the vascular cambium, which develop into radial parenchyma (RP) and axial parenchyma (AP) strands, respectively (Carlquist, 2001;Morris et al., 2016). The RAP fractions play a key role in defense against pathogenic fungi, as both RP and AP can accumulate antimicrobial compounds such as suberin (Biggs, 1987;Morris et al., 2016). The RP and AP can also have different functions. For example, the tensile strength of individual xylem rays was found to be three times higher than the radial strength of all xylem cells in Fagus sylvatica (Burgert and Eckstein, 2001), suggesting the importance of RP for tree mechanical integrity (Zheng and Martinez-Cabrera, 2013). Although total RAP proportion was not correlated with hydraulic properties Pratt et al., 2007;Poorter et al., 2010;Fortunel et al., 2014), the amount of AP was positively correlated with hydraulic capacitance in tropical species that had higher AP fractions around wide vessels (Morris et al., 2018;Aritsara et al., 2020).
Parenchyma fractions might act as a proxy for NSC storage capacity, and recent studies showed positive correlations between RAP and NSC content in woody stems and branches Godfrey et al., 2020;Chen et al., 2020;Kawai et al., 2021;Pratt et al., 2021). Although, these studies focused on a limited number of species, data suggest that the roles of RP and AP are distinct among species and climates. For example, Chen et al. (2020) found that NSC content was related to RAP and AP, but not RP in 19 temperate broadleaf tree species, while Kawai et al. (2021) showed both AP and RP were associated with NSC content in the branches of 15 subtropical woody species. In addition, Plavcová et al. (2016) found significantly lower NSC content but higher RAP volume in juvenile stems of four tropical species compared to 12 temperate species. Nevertheless, it remains unclear to what extent relationships would hold for a broad diversity of tree species and across biomes. Field measurements provide useful data, but due to the seasonality in NSC fluxes, largescale studies are difficult to perform because repeated sampling must be performed throughout the year (Martínez-Vilalta et al., 2016). A systematic review of available data with multiple seasonal sampling points can therefore provide an alternative perspective and complementary evidence to validate the relationship between NSCs and RAP in tree stems across species and climates.
Along with vessels and fibers, RAP is a key component of wood structure and partially determines wood density (Ziemińska et al., 2013;Fortunel et al., 2014;Osazuwa-Peters et al., 2017). Martínez-Cabrera et al. (2009) showed that RP surface area decreased with increasing wood density in shrubs. Conversely, other studies on trees have shown that there is a positive correlation between wood density and xylem ray volume (Taylor, 1969;Rahman et al., 2005), although this result may be biased by the proportion of lignified cell wall present. In addition, the distribution patterns of RAP cells within xylem and the variation in ray geometry can also play an important role in wood mechanical properties (Fujiwara, 1992;Ziemińska et al., 2015). Wood density can be linked to different ecological functions (Chave et al., 2009;Fortunel et al., 2014), including growth, life span, stem respiration, and successional status (Enquist et al., 1999;Roderick, 2000;Muller-Landau, 2004;Larjavaara and Muller-Landau, 2010;Martínez-Vilalta et al., 2010), that may also be affected by NSC availability (Körner, 2003;Hartmann and Trumbore, 2016). In particular, wood density tends to be inversely related to growth rate as a greater investment in mass per unit of wood volume slows down volumetric growth (Enquist et al., 1999;Poorter et al., 2008;Martínez-Vilalta et al., 2010;Fortunel et al., 2016;Rüger et al., 2018). Additionally, the relationships among wood density, RAP, and NSCs are regulated by trade-offs that affect growth rate and ultimately tree height. Taller trees tend to have stiffer stems (Jagels et al., 2018), less-dense wood (Díaz et al., 2016), and wider vessels (Morris et al., 2018). However, we are still lacking a general consensus on the link between NSCs, RAP, wood density, and maximum tree size across a broad diversity of species.
Species from different climatic backgrounds can have specific functional strategies that reflect differences in resource allocation and use (Wright et al., 2005;Violle et al., 2007;Yu et al., 2020). For example, in North and South America, both mean wood density and tree height increased with increasing mean annual temperature (MAT), but wood density increased as the sum of annual precipitation decreased (Šímová et al., 2018). Morris et al. (2016) also found there was a strong and positive correlation between RAP and MAT, albeit nonlinear, whereas the relationship between RAP and mean annual precipitation (MAP) was weakly negative.
Evolutionary subdivisions such as angiosperms and gymnosperms may drive functional strategies (Carnicer et al., 2013;Piper et al., 2019). Herrera- Ramírez et al. (2020), who studied one angiosperm and two gymnosperm species, found that NSC age and transit time differed between groups, with older NSCs occurring in stems of angiosperm species. Generally, deciduous species have higher levels of NSCs than in evergreen species because deciduous species have greater asynchrony between supply and demand (Chapin et al., 1990;He et al., 2020;Jiang et al., 2021), so seasonal fluctuations of NSCs are assumed to be greater in deciduous than in evergreen species because of leaf-fall (Piispanen and Saranpää, 2001;Schädel et al., 2009;Martínez-Vilalta et al., 2016).
Phylogenetic relationships between species can also be leveraged to evaluate the evolutionary drivers of functional strategies (Cadotte et al., 2008;Gravel et al., 2012;Schweiger et al., 2018). Previous studies found a strong signal of evolutionary history in the relationships between wood density and anatomical traits (Chave et al., 2006;Swenson and Enquist, 2007;Zanne et al., 2010;Baraloto et al., 2012;Gleason et al., 2012;Poorter et al., 2012;Fortunel et al., 2014). However, empirical evidence is not always consistent (Hoch et al., 2003;Richardson et al., 2013), suggesting that we need more data to understand the contrast between angiosperms and gymnosperms as well as between deciduous and evergreen species. By analyzing NSCs, RAP, and wood density across species and climates, we can further clarify the contribution of climate and evolutionary factors.
Here we compiled a database of NSC content, RAP fractions, wood density, and maximum tree height for 68 tree species that span a wide spectrum of characteristics, to provide a new outlook and further insight on their relationships across climates and evolutionary subdivisions. In line with previous findings, we hypothesized that (1) tree stems with large RAP fractions also have high NSC content and that RP fractions play a greater role in NSC storage than AP fractions; (2) RAP fractions and NSC content increase with increasing wood density but are smaller in taller trees; (3) RAP fractions, NSC content, and wood density all increase with warmer MAT, but decrease with increasing MAP.

Description of the data set
We combined available NSC and RAP data from specialized databases and a literature survey. We focused on data from tree stems only. Firstly, data on radial parenchyma (RP), axial parenchyma (AP) and RAP for a total of 1439 tree species were taken from the global wood parenchyma database  and Zheng and Martinez-Cabrera (2013) via the TRY Plant Trait Database (https://www.try-db.org/TryWeb/ Home.php; Kattge et al., 2011). Data on NSCs for 177 tree species were taken from a global NSC database (https://doi.org/ 10.5061/dryad.j6r5k; Martínez-Vilalta et al., 2016). When a tree species in the original database had multiple values, we calculated the mean. A comparison of the global wood parenchyma and NSC databases showed that 39 tree species had both NSC and RAP data. Secondly, according to the selection criteria of Martínez-Vilalta et al. (2016), we used the terms "NSC" or "nonstructural carbohydrates" in Google Scholar to search for peer-reviewed scientific papers that were published after the year 2015 (i.e., were not included in the database developed by Martínez-Vilalta et al., 2016), and that reported NSC content (Appendix S1). We found NSC data for another 18 tree species that were also included in the global wood parenchyma database. NSC data were obtained from tables or extracted from figures using Engauge Digitizer 4.1 (Mitchell et al., 2022). In addition, we also searched the xylem collection at CIRAD (Montpellier, France) that comprises microscope slides (tangential, transversal and radial sections of stem wood). In the CIRAD xylem collection, we found 11 tree species for which NSC data also existed in the global NSC database. Three microphotographs of transversal sections for each species were taken with an APO x5 lens using a digital camera (EOS 500D; Canon, Tokyo, Japan) mounted on a light microscope (BX 60 F; Olympus, Tokyo, Japan). Parenchyma areas were manually colored in Photoshop (Adobe, San Jose, CA, USA) and the proportions of RAP (the total surface area of parenchyma divided by xylem surface area, %) for each species were quantified using ImageJ 1.52 software (National Institutes of Health, Bethesda, MD, USA). An example of an image is shown in Appendix S2.
In most studies, NSC content was expressed as milligrams per gram dry mass; any NSC values in other units were converted to milligrams per gram for comparison. All RAP units were expressed as percentages. The database of NSC and RAP contained data for 68 tree species (Appendix S3), and only species' averages were used. Because NSC data were collected in different months, we collated the mean and maximum of NSC values and recorded them as NSC mean and NSC max , respectively. NSC mean represents the average state of the NSC during the year and thus balances the influence of seasonal variation. NSC max is the maximum content of NSC throughout the year. In addition to NSC (n = 65) and RAP (n = 60) values, data for mean values of soluble sugars (n = 58), mean values of starch (n = 62), AP (n = 39), and RP (n = 64) were also included in this database. All species' names were checked against standard taxonomic nomenclature. Based on tree evolutionary subdivision, tree species were divided into angiosperms (n = 49) and gymnosperms (n = 19). We also divided tree species into deciduous species (n = 27) and evergreen species (n = 41) according to leaf habit.
We recovered wood density data (n = 65) for the above tree species from the global wood density database (Chave et al., 2009; https://doi.org/10.5061/dryad.234, Zanne et al., 2009) and additional literature sources (Appendix S1), where wood density is defined as the oven-dry mass divided by green volume. We also obtained maximum tree height (H max ) data (n = 63) via the TRY Plant Trait Database (https:// www.try-db.org/TryWeb/Home.php; Kattge et al., 2011) and additional literature sources (Appendix S1).
We identified the locations (latitude and longitude) from the original publications that contained the NSC data for the 68 tree species. When mean annual temperature (MAT) and mean annual precipitation (MAP) were not presented in the original articles, we used their locations to extract them from Bioclim layers based on the WorldClim Global Climate Database (Fick and Hijmans, 2017) for the period 1970-2000 using ArcMap software (version 10.5, ESRI, San Jose, California, USA). According to this geographical information, the climates where species were growing were divided into temperate (n = 41), subtropical (n = 6), and tropical (n = 21) climate.
As in any systematic review of data, there are several caveats. First, studies may differ in trait measuring protocols. Second, some species are widespread and, accordingly, NSC, RAP, and wood density data from independent studies may be sampled in different countries, continents or climates, reducing the quality of the data matching (due to intraspecific variability in traits). To limit these effects, the following criteria for data selection were applied. First, for a given trait, we considered that species included in the same published database followed the same or similar protocols, and thus are comparable by default. For data from supplementary individual papers, we checked the measuring protocol and assessed whether it was compatible with those described in the global wood parenchyma database and global NSC database cited above. Finally, for each species, we checked the sampling location and year for each of the traits. Eighteen species were found to have trait values from different continents or climates. We conducted the main statistical analyses with and without those species, and the results were not significantly different. Therefore, we present the results for all 68 species. We also analyzed the variability of RAP, NSCs, and wood density within a single species across the different locations. We calculated the coefficient of variation for %RAP, NSC mean , NSC max , and wood density of the same tree species from different locations (when more than three values were available). The maximum coefficient of variation for %RAP was 0.41 (for Swietenia macrophylla King) and the minimum was 0.06 (for Pterocarpus angolensis DC.). Most of the coefficients of variation were less than 0.2. The maximum coefficient of variation for NSC mean was 0.58 (for Pinus sylvestris L.) and the minimum was 0.08 [for Picea abies (L.) H. Karst]. The maximum coefficient of variation for NSC max was 0.58 (for Fagus sylvatica L.), and the minimum was 0.02 (for Carpinus betulus L.). The maximum coefficient of variation for wood density was 0.30 [for Leucaena leucocephala (Lam.) de Wit], and the minimum was 0.01 (for Baikiaea plurijuga Harms).

Statistical analyses
We used one-way analysis of variance (ANOVA) followed by Tukey's honestly significant difference (HSD) post hoc test to test for differences in NSC mean and NSC max and a Kruskal-Wallis test to investigate differences in %RAP (1) between gymnosperms and angiosperms and (2) between climates. We ran linear regressions to examine the effects of MAT and MAP on NSC mean , NSC max , %RAP, wood density, and maximum tree height. To check whether the strength of relationships among NSC mean , NSC max , %RAP, wood density, and maximum tree height held in all species pooled and in each evolutionary subdivision and climate, we used linear models (1) with all species pooled, (2) separately for angiosperms and gymnosperms, and (3) by climatic region. We also investigated the relationships between NSC mean , NSC max , wood density, and %RAP using linear models for deciduous and evergreen species, respectively.
To test for multivariate correlations and bivariate relationships between NSC mean , NSC max , %RAP, wood density, and H max , we performed (1) a principal component analysis (PCA) between wood traits (including climate as supplementary variables) and (2) Pearson correlation tests across wood traits and climate variables (MAT, MAP, and latitude). The PCA and correlation analyses were repeated using phylogenetically independent contrasts (PICs; Felsenstein, 1985). In addition, phylogenetic generalized least squares (PGLS; Freckleton et al., 2002) were used to examine the role of species' evolutionary relationships on the associations between NSC mean , NSC max , %RAP, wood density, and H max . To this aim, we generated a phylogenetic tree for our 68 species with Phylomatic version 3 (Webb and Donoghue, 2005) using Phylomatic tree version R20120829 as the backbone, based on the Slik et al. (2018) phylogenetic hypothesis. We also performed a Mantel test between correlation matrices with and without PICs to determine whether correlation patterns differ when including PICs. As correlation tests are sensitive to missing data, we removed the species for which wood trait data were missing (n = 15) when running the PCA. Some tree species were located in the southern hemisphere, therefore latitude values were negative, so we took the absolute value of the latitude value when performing PCA and Pearson analyses.

RESULTS
Variation in mean and maximum NSC content and parenchyma fractions across evolutionary subdivisions and climates NSC mean and NSC max of tree stems for all species were 64.4 ± 5.1 mg g −1 and 87.9 ± 7.2 mg g −1 , respectively, and varied between evolutionary subdivisions and climates (Table 1). NSC mean , NSC max , soluble sugar, and starch content were greater in angiosperms than gymnosperms. NSC mean , NSC max , soluble sugar, and starch content were also greater in subtropical and tropical species, compared to temperate species (Table 1).
Mean stem %RAP for all species was 22.0 ± 1.8%, with differences in %RAP between evolutionary subdivisions and climates (Table 1). Stem %RAP and %RP were greater in angiosperms compared to gymnosperms. However, %AP was not significantly different between evolutionary subdivisions. In addition, stem %RAP and %AP were also greater in subtropical and tropical species compared to temperate species, respectively (Table 1).
NSC mean , NSC max , and %RAP were positively correlated with MAT ( Figure 1A, C, E). MAP was weakly and positively correlated with NSC mean and NSC max but not with %RAP ( Figure 1B, D, F). Neither wood density or maximum tree height were correlated with either MAT or MAP (Appendix S4).
Correlations between mean and maximum NSC content, parenchyma fractions, wood density, and maximum tree height across evolutionary subdivisions and climates As expected, NSC mean was positively correlated with %RAP ( Figure 2A). However, this correlation was found in angiosperms but not gymnosperms ( Figure 2B) and only in temperate species when focusing on climatic regions ( Figure 2C). In both deciduous and evergreen species, NSC mean was positively correlated with %RAP (Appendix S5). Across all species, NSC mean was positively correlated with %RP but not with %AP ( Figure 2D, G). When analyzing separately angiosperms and gymnosperms, no relationships were found between %AP or %RP and NSC mean ( Figure 2E, H). When dividing by climate zones, NSC mean was positively correlated with RP in temperate species ( Figure 2F, I). Similar patterns were found between NSC max and %RAP, %RP and %AP (Appendices S5, S6).
We found no relationship between NSC mean , NSC max , and wood density (Appendix S7), but we found a positive correlation between %RAP and wood density ( Figure 3A). However, when evolutionary subdivisions and climates were taken into account, significant correlations between %RAP and wood density occurred in gymnosperm species ( Figure 3B) and temperate species ( Figure 3C) only. Wood density was positively correlated with %RAP in both deciduous and evergreen species (Appendix S5). Across all species, wood density was positively correlated with %RP, but not with %AP ( Figure 3D, G). We found the same pattern when separately analyzing angiosperm and gymnosperm species (Figure 3E, H). Within climate zones, %AP and %RP were correlated with wood density only in temperate species ( Figure 3F, I).
We found that H max was strongly and negatively correlated with wood density and %RAP, but not with NSC mean and NSC max when all species were pooled together (Appendix S8). When analyzing separately angiosperms and gymnosperms, only wood density was significantly correlated with H max (Appendix S8). When data were divided by climate zones, NSC mean was weakly and negatively correlated with H max in temperate species (Appendix S8), and this relationship was largely driven by the presence of tall, coniferous and temperate species in the data set. Both NSC mean and NSC max were positively correlated with H max in tropical species (Appendix S8). Wood density and %RAP were negatively correlated with H max in temperate species only (Appendix S8). Similar patterns were found between % RP and H max , while no significant correlations were found between %AP and H max (Appendix S9).
Among the 68 species, only 53 species had complete data for NSC mean , NSC max , %RAP, wood density, H max , MAT, MAP, and latitude. The first two PCA axes for the five plant traits explained 56.9% and 24.0% of overall trait variation, respectively ( Figure 4A). NSC mean , NSC max , and %RAP loaded strongly on the first PCA axis, while wood density and H max mainly contributed to the second PCA axis (with opposite signs). Pearson correlation showed that %RAP was significantly and positively correlated with NSC mean , NSC max , and wood density, while H max was T A B L E 1 Mean values of soluble sugars, starch, NSCs, RP, AP, and RAP in tree stems in different climates and for angiosperms and gymnosperms (mean ± SE). Different lowercase letters indicate significant differences among climates; different uppercase letters indicate significant differences between evolutionary subdivisions (P < 0.05). NSC mean , mean value of nonstructural carbohydrates over one year; NSC max , maximum value of nonstructural carbohydrates in one year; AP, axial parenchyma; RP, radial parenchyma; RAP, radial and axial parenchyma; n, sample size. See Appendix S11 for statistical results of the difference analyses significantly and negatively correlated with %RAP and wood density (Table 2). When PICs were used in trait PCA analyses, we found that there were no changes in the patterns of correlations among wood traits. The first PCA axis explained 40.6% of overall variation, mainly loading with NSC mean , NSC max , and %RAP and the second PCA axis explained 27.9%, through coordinated tree traits such as H max and wood density ( Figure 4B). In a Pearson correlation, when PICs were included, NSC max and %RAP, wood density and H max were significantly correlated, but there were no significant correlations among other traits ( Table 2). The matrices of pairwise correlations among wood traits were similar between Pearson correlation matrices with and without PICs (R Mantel = 0.58, P < 0.01). The PGLS models showed that both NSC mean and NSC max content and wood density were significantly correlated with %RAP, while NSC mean and NSC max content were not correlated with wood density (Appendix S10).

DISCUSSION
We found that trees with large RAP fractions also had more NSCs but only in angiosperms. In addition, we did not find more NSCs in denser wood, although the radial parenchyma (but not axial parenchyma) fraction was positively related to wood density. Taller trees had less dense wood and less RAP, but we found no significant relationships between NSCs and tree height. NSCs and RAP were more abundant in tropical species, as a result of strong positive relationships with MAT. Our analysis of evolutionary relationships demonstrated that RAP fractions and NSCs were always closely related, suggesting that RAP can act as a reliable proxy for maximum NSC storage capacity. Unless specified otherwise, as patterns for NSC mean and NSC max were similar, we use the generic term NSCs from here onward.

Relationships between NSCs and RAP fractions
In agreement with our first hypothesis, that tree stems with large RAP fractions have more NSCs, we found a positive relationship between NSCs and %RAP across 68 species and three climate regions (Figure 2A). Nevertheless, variability in data was high, partially due to the combination of data originating from diverse sources and seasons (Martínez-Vilalta et al., 2016;Furze et al., 2019). Also, we found a positive relationship between NSCs and %RAP in both deciduous and evergreen species, which emphasized our first hypothesis and highlighted the storage role of parenchyma. However, we found no significant relationships between NSCs and %RAP in gymnosperms, but % RAP did diverge between angiosperms and gymnosperms, as also found by Morris et al. (2016). Angiosperms have a wide range of ray types and dimensions, whereas gymnosperms usually only have uni-/biseriate rays. The amount of NSCs was also greater in angiosperms than in gymnosperms, which translated into different associations between NSCs and RAP. In conifers, the proportion of axial parenchyma that constitutes secondary xylem is usually much lower than that of angiosperms (Spicer, 2014;Aritsara et al., 2020), although we found no significant differences. This configuration may emphasize the role of axial parenchyma in maintaining water transport in the angiosperm vessels (Spicer, 2014). Also, distinct differences in wood structure and function between angiosperms and gymnosperms could mask effects, e.g., live wood fibers in angiosperms can accumulate starch and further contribute to NSC storage (Yamada et al., 2011;Carlquist, 2014;Plavcová et al., 2016). Additionally, von Arx et al. (2017) explained the lack of relationships between NSCs and RAP in Pinus sylvestris by suggesting that radial parenchyma fractions usually indicate the maximum potential NSC storage capacity in tree stems, and not the actual NSC content at any one time.
When we examined the relationships between NSCs and radial and axial parenchyma separately, we found a positive relationship between NSCs and radial parenchyma across F I G U R E 4 Loadings plot of wood traits on the first two principal components axes (A) using raw data and (B) using phylogenetically independent contrasts (PICs). Mean annual temperature (MAT), mean annual precipitation (MAP), and latitude (Lat.) were added as supplementary variables without any effect on the analyses. Density, wood density; H, potential maximum tree height; NSC mean , mean value of nonstructural carbohydrates over 1 year; NSC max , maximum value of nonstructural carbohydrates in 1 year; RAP, radial and axial parenchyma. The light blue area represents temperate tree species; the yellow area represents subtropical tree species; the light red area represents tropical tree species. Blue and red arrows represent abiotic variables and wood traits, respectively T A B L E 2 Pearson correlations (ρ values) among NSC content, RAP fractions, wood density, and climatic factors in 68 species. Phylogenetic independent contrasts are not considered below the central diagonal line of the matrix, but are considered above the diagonal line. Significant effects are in bold (P < 0.05). Density, wood density; H max , maximum tree potential height; MAP, mean annual precipitation; MAT, mean annual temperature; NSC mean , mean value of nonstructural carbohydrates over 1 year; NSC max , maximum value of nonstructural carbohydrates in 1 year; RAP, radial and axial parenchyma all species, but no relationships with axial parenchyma ( Figure 2D, G). These results indicate that radial parenchyma plays a greater role in NSC storage than axial parenchyma. It has been suggested that axial parenchyma can influence hydraulic capacitance in angiosperms by releasing water into vessels in periods of drought (Morris et al., 2018). Additionally, vessel to axial parenchyma connectivity would facilitate xylem hydraulic optimization, suggesting that plants may need axial parenchyma to surround and service vessels (Aritsara et al., 2020;. Therefore, the physiological function of both types of parenchyma could be very different, calling for a careful consideration of xylem structure for a better understanding of tree ecological strategies. We did not find significant relationships between NSCs and %RAP in subtropical and tropical tree species, therefore partially refuting our first hypothesis. Because our results are contradictory to the results of Kawai et al. (2021), who found that RAP fractions were significantly associated with NSC content in the branches of 15 subtropical woody species, our low sample size (21 species) might not fully represent the variability in NSCs and RAP encountered in these climates. Additionally, RAP performs several major functions in tropical species, such as water storage and defense against pathogens, that may be less important in temperate species . Also, photosynthesis and growth can continue throughout the year in the tropics (in the absence of marked dry seasons), minimizing seasonal mismatches between carbon source and sinks. Future studies adding more data from tropical and subtropical species are needed to further clarify the relationship between parenchyma fractions and NSC content.

Relationships among NSCs, RAP, wood density, and maximum height
We did not find any relationships between NSC content and wood density across species (Appendix S7), but we found that %RAP increased in more dense wood, partially corroborating our second hypothesis. Although wood density was positively correlated with %RAP and radial parenchyma across all species, it was not related to axial parenchyma. Because parenchyma cells are relatively thin-walled, parenchyma-rich wood may need relatively more tracheids or fibers to contribute to mechanical strength (Martínez-Cabrera et al., 2009). Wood rays usually have large amounts of lignin packed among parenchyma cells, that will also increase overall wood density and stem bending strength (Burgert and Eckstein, 2001;Rana et al., 2009). However, in 61 shrubs from North and South America, Martínez-Cabrera et al. (2009) reported that radial parenchyma was negatively correlated with wood density, while axial parenchyma showed the opposite trend, suggesting that radial parenchyma may possess different roles in shrubs and trees, that have different needs in terms of mechanical strength. In addition, fibers and fiber tracheids contribute to mechanical strength in angiosperms (Ziemińska et al., 2013;Fortunel et al., 2014), while the mechanical strength in gymnosperm xylem is largely provided by tracheid cells (Patten et al., 2010). Apart from providing a storage function, axial parenchyma also has a role related to hydraulics. For example, Zheng and Martínez-Cabrera (2013) found axial parenchyma was directly related to conducting efficiency, and could provide extra water storage and embolism repair capacity in low density wood. Additionally, Janssen et al. (2020) found that axial parenchyma was not related to wood density but increased with sapwood specific hydraulic conductivity in neotropical tree species, which may be due to the role of parenchyma cells in the refilling of embolized vessels (Secchi et al., 2017). When dividing species by climatic zones, we found significant correlations between wood density and parenchyma fractions in temperate species only, potentially because of lower variation in wood density in temperate than tropical tree species (Swenson and Enquist, 2007).
In our study, wood density was negatively correlated with potential maximum tree height, as also found by Thomas (1996). Taller trees have wider conduits in their trunks because they have longer path lengths and therefore need wider vessels (thus reducing wood density) to maintain hydraulic conductivity and minimize the increase in hydraulic resistance (Coomes et al., 2008;Olson et al., 2021). In tropical species, NSCs were more abundant in taller trees, although this pattern was not explained by increases in axial or ray parenchyma. Taller angiosperms usually have more axial parenchyma cells packed around vessels, reducing overall wood density, and protecting against xylem embolism through the controlled water release into vessels (Preston et al., 2006;Morris et al., 2018). However, our angiosperm data set was limited to trees with a maximum potential height of 10-60 m, compared to 20-110 m for temperate conifers. Therefore, the smaller height range, combined with the increased diversity of cell types in angiosperms, may mask patterns of xylem variability linked to tree height.
Recent studies have emphasized the importance of including information on phylogenetic relationships between species to better understand the role of evolutionary history on trait variation (Chave et al., 2006;Cadotte et al., 2012;Fortunel et al., 2014;Schweiger et al., 2018). In our study, the only significant relationships found after accounting for species phylogenetic relationships were %RAP with NSC max , and wood density with maximum potential tree height ( Table 2). The underlying cause for these patterns may be that traits are labile, for instance, when distant relatives show convergent evolution or when close relatives show trait differentiation (Cavender-Bares et al., 2004;Gravel et al., 2012). The relationship between %RAP and NSC max indicates that RAP is a reliable proxy for maximum potential NSC storage capacity, supported by phylogenetically corrected regression analyses (PGLS), showing that NSCs and wood density were always related with %RAP. Also, the matrices of pairwise correlations with and without PICs among wood traits were similar, indicating that phylogenetic relatedness did not affect the relationships among wood traits. Therefore, our results show for the first time, the close evolutionary relationship between stem parenchyma cells and NSC storage.

Effect of climatic factors on wood traits
In agreement with our third hypothesis, we found that NSCs and %RAP increased with warmer temperatures. The strong, positive relationships we found among latitude, NSC content, and RAP fractions reinforce the relationships that we found with MAT because there were more NSCs and RAP in tropical tree species. Our results suggest that as temperature increases, photosynthesis increases more rapidly than the combination of all carbon sinks (growth, respiration, defense), and thus, more photosynthates are stored in NSC pools, providing an important mechanism to offset rising metabolic costs (Hoch and Körner, 2009;Gough et al., 2010). We found that NSCs and %RAP were positively correlated with MAT, but only NSCs were positively correlated with MAP. This result is similar to that found by Moles et al. (2014), who demonstrated that most plant traits correlate more strongly with temperature than precipitation. Precipitation is usually interpreted as an indicator of water availability to plants, which could be affected by a suite of factors including hydrology, soil type, soil depth, and access to groundwater (Gardner, 1965). Global variation in any or all of these factors would weaken the relationship between MAP and plant traits (Moles et al., 2014). Although RAP and NSCs were influenced by climate, we found no significant relationships between MAT, MAP, and wood density, which is inconsistent with results from a study by Šímová et al. (2018). Our result indicates that other factors not considered here, such as soil fertility, stem diameter, tree age, and stand density, are more important determinants of wood density variation (Larjavaara and Muller-Landau, 2010;Nabais et al., 2018).

CONCLUSIONS
We investigated the relationships among NSCs, RAP, wood density, and potential maximum tree height, together with climatic factors, for 68 tree species. We found that NSC content was strongly related to RAP when all species were pooled together, demonstrating that parenchyma cells are a major repository for NSCs. Wood density was positively correlated with the proportion of ray parenchyma in tree stems, but there were no significant relationships between wood density and NSCs. Tree height and wood density were negatively related overall, but NSCs were more abundant in taller, tropical species. Both NSCs and RAP increased in tropical species, reflecting the strong, positive relationship with mean annual temperature. Phylogenetic relatedness did not affect the correlation patterns among wood functional traits showing that RAP is a suitable proxy for potential NSC storage capacity in stems. Our study helps to better understand the trade-offs among xylem ecological strategies, but more data, especially from the tropics and subtropics, are required on a large scale to build a robust understanding of how different sapwood functions (including NSC storage, water transport, respiration, mechanical integrity, and resistance to pathogens) are related across species, growth forms, and climate.
AUTHOR CONTRIBUTIONS G.Z. and A.S. designed the study and collected data. G.Z. and Z.M. analyzed data. G.Z. wrote the first draft of the manuscript that was further improved by inputs from all co-authors. All the authors contributed critically to the drafts and gave final approval for publication.

ACKNOWLEDGMENTS
This work was supported by China Scholarship Council (Ph.D. bursary for G.Z.). This work has benefited from support of a grant from Investissement d'Avenir grants of the Agence Nationale de la Recherche (CEBA: ANR-10-LABX-25-01). This study was co-financed by a French-Chinese program PHC XU GUANGQI 2019 (ref.: 43379XF). We thank Dr. P. Langbour (CIRAD, France) for access to the CIRAD Xylotech, Montpellier. We thank Prof. Y. Y. Zhang (Xiamen University, China.) for her valuable suggestions on data analysis. We are very thankful for the constructive comments by the Associated Editor and reviewers, which greatly improved this manuscript.

DATA AVAILABILITY STATEMENT
Data used in the systematic review can be accessed at Data INRAE at https://doi.org/10.15454/H5RLTP.