Grow slowly, persist, dominate—Explaining beech dominance in a primeval forest

Abstract Being able to persist in deep shade is an important characteristic of juvenile trees, often leading to a strong dominance of shade‐tolerant species in forests with low canopy turnover and a low disturbance rate. While leaf, growth, and storage traits are known to be key components of shade tolerance, their interplay during regeneration development and their influence on juveniles' survival time remains unclear. We assessed the ontogenetic effects of these three traits on the survival time of beech (Fagus sylvatica), and Norway and sycamore maples (Acer pseudoplatanus, Acer platanoides) in a primeval beech forest. Biomass allocation, age, and content of nonstructural carbohydrates (NSC) were measured in the stems and roots of 289 seedlings and saplings in high‐ and low‐vitality classes. Saplings experienced a trade‐off between absolute growth rate (AGR) and storage (NSC) as the leaf area ratio (LAR) decreases with biomass development. High LAR but low AGR and low NSC corresponded to beech with a marked ability to persist in deep shade while awaiting canopy release. In turn, a comparably small LAR in combination with a high AGR and higher storage (NSC), as observed in Norway maple and sycamore maple, reduced sapling survival time, thus offering an explanation for beech dominance and maple disappearance in the undergrowth of old‐growth beech forests.


| INTRODUC TI ON
In primeval monodominant Fagus sylvatica L. forests, seedlings (0-130 cm tall) and saplings (131-500 cm tall) often grow in deep shade for extended periods, due to low canopy turnover (Hobi et al., 2015;Runkle, 1985;Valverde & Silvertown, 1997). If canopy turnover is low (i.e., the mean time between recurring gap formation at any point in the forest), saplings that are able to persist for decades in shade are more likely to experience a canopy opening (i.e., a release event) enabling the subsequent promotion to canopy (Canham, 1985(Canham, , 1990. Hence, high juvenile shade tolerance is pivotal in determining the survival time of F. sylvatica and may be compromised in co-occurring species, explaining the low tree species diversity and F. sylvatica dominance frequently observed during succession (Korpel, 1995;Rey et al., 2019).
Shade tolerance can be assessed via functional traits, that is, morphological, physiological, and phenological features that reflect a species' ecological strategy (Pérez-Harguindeguy et al., 2013). One of the concepts used to explain shade tolerance, the "carbon gain" hypothesis, postulates that saplings can enhance carbon gain in the shade, either by minimizing CO 2 losses via respiration or by investing in the light-harvesting capacity (greater leaf area and crown volume, Givnish, 1988) while maintaining a higher growth rate (Popma & Bongers, 1988;Walters & Reich, 1996). Another concept, the "defence and storage" hypothesis, relates shade tolerance to the resistance to herbivory, pathogens, and mechanical damage (Kitajima, 1994) and to storage (Kobe, 1997). Accordingly, shade-tolerant species do not maximize growth in low light but invest a larger fraction of non− structural carbohydrates (NSC) in storage to buffer against stress during a prolonged period of shade (Kobe, 1997). The two concepts, "carbon gain" and "defence & storage," are not mutually exclusive, but rather present different mechanisms of the complex phenomenon of shade tolerance.
The traits of shade tolerance change during ontogeny, that is, the development from seedlings to saplings. Maintenance and construction costs increase with tree height because the proportion of non−photosynthetic support tissue increases continually (Delagrange et al., 2004). At the same time, the ratio of leaf area to total tree biomass (LAR) diminishes as young deciduous trees grow (Niinemets, 1998), and thus, the leaf area capacity may be limited in terms of providing photosynthates for both growth and storage. NSC dynamics in juvenile regeneration during ontogenetic development are not well understood . Although the NSC concentration is expected to decrease with tree height , the growing volume of support tissue suggests that allocation to storage increases in proportion to plant mass (Plavcová et al., 2016), which in turn may decrease growth under carbon limitation (Wiley & Helliker, 2012). Thus, during ontogenetic development, young trees may experience a trade-off between growth and storage, leading to a shorter survival time.
Studies on the relationships among the traits of shade tolerance in broad-leaved species have mostly been focused on seedlings in garden experiments (Gibert et al., 2016) and have not involved investigations of how leaf, growth, and storage traits develop with age, thus leading to patchy evidence (Valladares et al., 2016;Valladares & Niinemets, 2008). Moreover, we are not aware of any existing study on deciduous trees addressing how ontogenetic changes in traits of shade tolerance may affect the survival time of juvenile trees. We define "regeneration survival time" as the potential time that seedlings and saplings can survive in the unfavorable environment of deep shade, which corresponds to the time until the first canopy release. In the present study, we aimed to combine leaf, growth, and storage traits as proxies of shade tolerance to infer the survival time of juvenile beech and co-occurring species.
Due to low tree diversity in monodominant F. sylvatica forests, we studied its seedlings and saplings (0-5 m height) and the most abundant competitor species, such as Acer pseudoplatanus and Acer platanoides, in two vitality classes. Comparisons of traits between trees with high and low vitality (i.e., the capacity to grow, resist stress, and acclimate to environmental conditions; adapted from Brang, 1998;Dobbertin, 2005) made it possible to relate trait performance to survival time. In particular, we investigated the following research questions: (a) Which traits relating to leaf, growth, and storage can discriminate between low-and high-vitality regeneration? (b) Is there a trade-off between growth and storage traits among species of low and high vitality? (c) How do these traits affect regeneration survival time?

| Study area and plot selection
The Uholka-Shyrokyi Luh reserve in Ukraine belongs to one of the most investigated F. sylvatica-dominated primeval forests of Europe and is listed as a UNESCO World Heritage site (Stillhard et al., 2019;Trotsiuk et al., 2012;Zenner et al., 2020). In this study, we focused on the Uholka part of the forest (coordinates: 48°16′N, 23°40′E), which was selected because it has a greater share of Acer spp. than in the Shyrokyi Luh part of the reserve. The Uholka part covers 4,729 ha, ranging from 400 to 1,300 m a.s.l., with a mean annual temperature of about 8°C (−3°C in January and 18°C in July at 430 m latitude) and a mean annual precipitation of 1,134 mm (Commarmot et al., 2013).
The share of A. pseudoplatanus shrinks from 15% in the 10-39.9 cm height class to 3% in the 3-3.9 cm DBH class, while the share of A. platanoides is almost zero for trees with a DBH >2 cm (inventory 2010, analysis not shown). The forest is dominated by a small-scale disturbance regime with a mosaic of mainly small canopy gaps (98% are <200 m 2 ); only a few large, stand-replacing events were detected in a study using high-resolution satellite imagery (Hobi et al., 2015).
We randomly selected six plots (total area 2.53 ha) varying from 0.2 to 0.7 ha in size, in which mixed regeneration of the three species was present in subplots. Within the six plots, nine subplots from 140 to 520 m 2 (total 0.26 ha) were delineated to contain as many seedlings/saplings of the target species/sizes/vitality classes as possible.
Among this regeneration, 289 target seedlings and saplings were randomly selected and marked according to the following criteria: three species (F. sylvatica, A. pseudoplatanus, and A. platanoides), two vitality classes (low and high), and eight height classes: 0-10, 11-20, 21-35, 36-60, 61-90, and 91-130 cm as seedlings, 131-200 and 201-500 cm as saplings, one seedling/sapling per plot in each height and vitality class (see Table S1). Browsing was apparent on all plots, with many recovered Acer spp. trees having scars, while F. sylvatica regeneration was almost untouched.

| Classification into vitality classes
We developed criteria for juvenile trees based on the vitality assessment used for adult trees, in which tree crowns are assessed visually (Eichhorn et al., 2016;Roloff, 1991) and growth is measured in the field (Dobbertin, 2005). Crown transparency has been shown to correlate well with relative growth rate (Lorenz et al., 2004;Solberg, 1999) and also with subsequent tree mortality and survival (Dobbertin & Brang, 2001;Schmid-Haas, 1993). Hence, we classified seedlings and saplings (Table 1), taking into account crown transparency (leaf loss and/or dieback) and the increment of the apical shoot for several years, but we used the branching pattern and stem condition as additional discriminators to differentiate between high-and low-vitality trees (Collet et al., 2011;Roloff et al., 2016). To avoid an inconsistent crown transparency assessment (Dobbertin, 2005), only one evaluator assessed all seedlings and saplings, using site-specific reference trees. A reference tree is a tree with full foliage (defoliation 0%) that grows at a particular site, considering altitude/latitude, site conditions, and social status (Eichhorn et al., 2016). Trees browsed during the current season were not considered.

| Field measurements
The following measurements were taken before tree excavation: diameter at root collar (DRC), tree height, and height of the crown base (height of the lowest foliage, excluding epicormic shoots).
We measured crown area projection by two perpendicular crown diameters using a pendulum suspended from the outermost branches to the ground. Stem height increment was measured for the most recent 5-10 years (until the last visible bud scale scar) to the nearest millimeter. Leaf area index and indirect site factor (ISF), that is, the proportion of diffuse solar radiation at a given location relative to that in the open, were assessed with hemispherical photographs (Coolpix 4500, Nikon, Japan) with a 183° fish-eye lens (Nikon FC-E8) mounted on a tripod (Thimonier et al., 2010). Photographs were taken just above the uppermost leaves of every tree, bending saplings taller than 1.5 m to allow photograph shooting of canopy. We then excavated trees manually and cleaned roots with water to avoid damage to the fine roots.

| Postharvest processing (mid-May to mid-July)
The sampled trees were separated at the root collar into aboveground biomass (foliage, stem, branches) and belowground biomass (roots). Pieces of 5-cm length from the stem at the level of the root collar and from the coarse roots (diameter >2 mm) were cut for NSC analysis and placed in a microwave at 900 W twice for 15 s immediately after the harvest (Popp et al., 1996). In the case of seedlings without coarse roots, we used the taproot. All fresh leaves per tree were scanned with a smartphone (Petiole, version 2.0.1, Petiole Ltd. 2019) after calibration of the camera. Foliage, stems, branches, and roots were dried at 65°C for 3 days until a constant weight was reached and then weighed to the nearest 0.01 g.

| Calculations
Crown area projection was calculated based on the quadratic mean radius (Pretzsch et al., 2015). We multiplied crown area projection by the difference between tree height and the height of the crown base to obtain crown volume (assuming it is a cylinder). Hemispherical photographs were analyzed with the program Hemisfer (version 2.2, ©Patrick Schleppi, WSL). ISF was estimated using the method introduced by Thimonier et al. (2010). We calculated the trait variables according to the formulae in Table 2 and present the final results in Table S1.

| NSC analysis
Non−structural carbohydrates (NSC) represent the storage trait in the "defence and storage" concept of our study; they are sugars of low molecular weight (glucose, fructose, and sucrose) and starch. NSCs were analyzed according to the Wong (1990) protocol modified by Hoch et al. (2003). The harvested (mid-May to mid-July) coarse root sections of larger saplings (2-4 cm DRC) were limited to 5-10 mm diameter and the harvested stem sections to 10 mm of wood directly under the bark. The milled stem sections (without bark) and the root (without bark if possible) of each sapling (10-12 mg) were boiled in 2 ml of distilled water for 30 min. After centrifugation, we added TA B L E 1 Criteria used to classify juvenile trees into high-and low-vitality classes based on (a) crown transparency, (b) apical shoot increment, (c) branching pattern, and (d) stem damage We took 500 μl of the extract (including sugars and starch) and incubated it with a fungal amyloglucosidase from Aspergillus niger (Sigma-Aldrich) for 15 hr at 49°C to break starch into glucose. Total glucose (corresponding to NSC) was determined photometrically as described above. The concentration of starch was calculated as NSC concentration minus the free sugar concentration determined in the first step. Standards of pure starch and glucose, fructose, and sucrose solutions were used as controls, and standard plant powder (orchard leaves; Leco) was included to test the reproducibility of the extraction. NSC concentrations were expressed on a gram per dry matter basis and scaled to the whole stem and root dry biomass to obtain the absolute value of total NSCs pool per tree. We assumed no large vertical or horizontal NSC gradient within the wood (which is all sapwood); therefore, upscaling to the whole stem and dry root mass should give realistic absolute values for the NSC content per tree. We ran the analysis in the same laboratory with no change in protocol (Quentin et al., 2015).

| Dendrochronological analysis
From each harvested tree, a stem disk was cut at the level of the root collar using a microtome to determine age and radial growth. The stained disks were photographed (Canon EOS 700D) and analyzed with WinDENDRO™ (Regent Instruments Inc.) under a microscope.
The number and width of the rings were measured in 2-4 perpendicular directions because of the eccentric tree piths and then arithmetically averaged.

| Variable selection and discrimination between vitality classes
Principal component analysis (PCA) was used to select variables among leaf and growth traits with the highest contribution to principal components ( Figure S1, Tables S2). The comparison of means among vitality classes (Table S3) was made with Yuen's trimmed t test (Yuen, 1974) with the Benjamini-Hochberg p-value adjustment, with the significance level set to 0.05 (Benjamini & Hochberg, 1995); and among species with a heteroscedastic two-way factorial ANOVA   (1) where LA is the leaf area per tree and m total is the total dry mass per tree where LA is the leaf area per tree and m leaf is the total leaf dry mass per tree where m leaf is the total leaf dry mass per tree and m total is the total dry mass per tree and r 1 …r 4 are radii in four cardinal directions m 2

Growth
Absolute growth rate, AGR AGR = m total t (5) where m total is the total dry mass per tree and t is tree age g/year Shoot mass growth rate (leaves + branches) where m shoot is the total dry shoot mass per tree and t is tree age where is the error term that follows a standard normal distribution ( Figure S3). All continuous variables were log-transformed, except for tree age, and centered. The assumption of homogeneity of covariance (Box's M test at α < 0.001)) was not violated (p = 0.03).
Correlation between the covariates was moderate: r = 0.50 between shoot growth rate and tree age, low correlation r = −0.11 between shoot growth rate and LMF, and r = −0.36 between LMF and age.
Multicollinearity was tested with variance inflation factor VIF (R package olsrr) and did not exceed 3, indicating low-to-moderate multicollinearity. The analyses and visualization were run in R, version 3.6.1 (R Core Team, 2019).

| Biomass allocation
In deep shade (mean ISF 1.95%-3.34% on the six plots, Table S1), F. sylvatica invested more heavily in leaf area development and crown volume starting from the height class >36 cm in both highand low-vitality trees than its competitor species, while its mean leaf area and crown volume were smaller than those of the competitors for smaller seedlings <36 cm height (Figure 1). Mean biomass allocation to leaves increased with tree height for all species, with F. sylvatica investing more in leaves than Acer spp. above a tree height of 60 cm (Figure 1). Patterns in branch biomass allocation were similar to those for leaf biomass, but maple seedlings <60 cm tall generally did not grow branches if not browsed. Allocation to the stem was more similar in the three species than allocation to leaves and branches. Mean biomass partitioning to roots was higher in Acer spp.
up to a height of 90 cm. In short, Acer spp. focused on conservative investments in the stem and roots, while F. sylvatica pursued lightharvesting and space occupation strategies by allocating biomass to leaves and branches.
As expected, species identity significantly affected leaf traits such as leaf area (p = 0.018) and crown volume (p = 0.025) in the heteroscedastic two-way factorial ANOVA where the two vitality classes and two height classes were pooled together ( Figure S2). The species effect was also significant for the growth trait shoot biomass (leaf and branch mass, p = 0.004), but not for total dry mass (p = 0.335).
The total mean biomass of Acer spp. trees (39 ± 6.7 g for A. pseudoplatanus and 44 ± 8.4 g for A. platanoides) was smaller, although not significantly so, than that of F. sylvatica (86 ± 17 g, mean ± SE).

| Discrimination between vitality classes
Leaf traits (LAR, leaf area, crown projection area, crown volume), together with growth traits (AGR, root collar diameter, mean height increment; growth rates of leaf, stem, roots, and shoots biomass per year) and storage (NSC), contributed 56.6% to the first axis in the PCA, while LAR, LMF, SLA, and tree ring width contributed 13.6% to the second axis, summing to 70.1% of the total variance in the tree traits. The major contributors to the two principal components were AGR (11.0%), NSC pool (7.6%), LAR (32.6%), and LMF (33.0%) ( Figure S1, Table S2). High-vitality trees differed from low-vitality individuals in that they had a significantly higher LAR  Table S3). Unlike leaf and storage traits, growth (AGR) did not differ between high-and low-vitality trees (p = 0.181). Also, the mean annual height increment was similar for both vitality classes and varied between 5.2 and 6.5 cm for high-vitality trees and between 4.4 and 5.7 cm for low-vitality individuals ( Figure S4).

| Influence of LAR, AGR, and NSC pool on species' juvenile survival time
Effects of tree age, shoot growth (leaf + branch mass per year), species identity, LMF, and the interaction between tree age and shoot growth were significant for the response variables LAR, AGR, and NSC in the MANCOVA model, based on a Pillai test ( Figure S4). For every 1% increase in tree age, LAR decreased by 1.33% and AGR by 0.74%, while NSC increased by 9.37% ( = 0 + 1 species + 2 ln(LMF) + 3 ln(shoot growth) * age + 4 ln(shoot growth) + 5 age + (prediction errors or RMSE) except for NSC, and a high goodness-offit (R 2 = 0.83-0.98).
The predictions of the MANCOVA model for LAR, AGR, and NSC over tree age suggest that juvenile trees of the three species are facing similar trade-offs between investment to leaves (LAR), growth (AGR), and storage (NSC) (Figure 2). The decline in LAR over time is due to an increasing tree biomass, while the increase in AGR means that tree biomass is increasing at ever higher rates, in particular given the log scale of the vertical axis ( Figure 2). In line with the increasing AGR, the absolute value of the NSC pool is also increasing, due to the growing parenchyma tissue. The relevant pattern in Figure 2 is, therefore, the difference in slope between LAR on the one hand and AGR and NSC on the other hand. Unfortunately, we lack data for trees in the height class >5 m (and thus of older age) and can only hypothesize that the trajectory of AGR and NSC would have continued to develop in a linear (on a log scale) manner or would have taken another trajectory. Another trajectory assumes a slowing of growth and/or storage, when both may reach a plateau.
Still, the pattern emerging from Figure 2 is that, in high-vitality trees of the same age, the two Acer spp. have a smaller LAR at their disposal (smaller LAR intercept values) than F. sylvatica, due to faster growth (higher AGR intercept value) than in F. sylvatica, and also need more storage than F. sylvatica (visible from the higher NSC intercept value in Figure 2).
This relationship is more evident in low-vitality Acer spp. trees where LAR is lower or decreasing faster (steeper slope) than in high-vitality regeneration while AGR and NSC increase faster. Lowvitality F. sylvatica trees differ from high-vitality trees in that they have a lower LAR and exhibit faster growth of AGR and NSC values, as indicated by the comparably steeper lines. The different rates of change in LAR, AGR, and NSC lead to a trade-off point between AGR and NSC due to carbon limitation, which is reached earlier in lowvitality trees than in high-vitality trees (point 1, Figure 2). Carbon limitation occurs as a result of the ever-shrinking capacity of LAR to support both growth (point 2) and storage (point 3), assuming no improvement of light availability during this development.
The three species have different allocation strategies, as confirmed by the predicted marginal mean of the MANCOVA model (assuming tree age = 14.1, shoot growth = 0.4 g, and LMF = 0.1).

| Traits of shade tolerance and their impact on survival time
The been shown so far (Palacio et al., 2014). In our study, the trade-off between growth (AGR) and storage (NSC) at decreasing capacity of LAR can lead to a reduction in survival time under carbon limitation (point 1, Figure 2). Thus, our model can partly explain the long survival time of F. sylvatica in old-growth deciduous forests.
The "carbon gain" concept is based on the assumption that shadetolerant species have a higher LAR, and therefore improved light interception, than shade-intolerant species and thus a larger fraction of biomass is allocated to foliage (Bazzaz, 1979;Givnish, 1988). Indeed, in our study the TA B L E 3 MANCOVA summary: effect of explanatory variables tree age, shoot growth rate, species identity, leaf mass fraction (LMF), and interaction between tree age and shoot growth rate on leaf area ratio (LAR), absolute growth rate (AGR), and content of non−structural carbohydrates (NSC) Abbreviations: CI-0.95 confidence intervals, SE-standard error, RMSE-root-mean-square error or prediction error, R2-the variance of the response variable explained by the explanatory variables.
marginal mean LAR of F. sylvatica was higher and declined more slowly during ontogeny compared with its competitors (Figure 2), confirming findings by other authors (Annighöfer et al., 2017;Niinemets, 1998).
Such phenotypic plasticity can be explained by two factors: (a) low annual growth allows F. sylvatica to balance leaf area per unit mass and (b) the energy (glucose) required for building a unit of leaf area is lower in juvenile F. sylvatica than in A. pseudoplatanus (Petriţan et al., 2010). A failure to balance leaf area with growth leads to consistently higher mortality in seedlings of species with a higher relative growth rate than in seedlings with a lower relative growth rate in deep shade (Walters & Reich, 1996).
Hence, F. sylvatica maintains a higher LAR for more efficient capture of diffuse light at a relatively "cheap" leaf construction cost.

F I G U R E 2
Linear regression predicting the development of LAR, AGR, and NSC with tree age: Dots represent observed values, bands represent 0.95 confidence intervals (CI), point 1 indicates carbon limitation as a result of a trade-off between AGR (growth) and NSC pool (storage), point 2 indicates a limiting LAR capacity to support growth, and point 3 indicates a limiting LAR capacity to support storage F I G U R E 3 Pairwise comparison among species with the post hoc Tukey HSD test: Different letters indicate significant differences, whiskers represent 0.95 CI, and marginal mean assumes tree age = 14.3, shoot growth = 0.4 g/year, and LMF = 0.1 According to several studies (Popma & Bongers, 1988;Walters & Reich, 1996), the growth rate of young trees in low light should be higher for shade-tolerant species and lower for shade-intolerant species. Our findings imply the opposite, however, as F. sylvatica grew slowly at a young age and focused biomass allocation to branches and leaves (horizontal growth), supporting the findings of Petriţan et al. (2010) and Collet et al. (2011). The negative relationship between growth and survival time in F. sylvatica for the first 70 years of its life was observed in studies by Di Filippo et al. (2012), Di Filippo et al. (2015. Similar relationships have been observed between the growth rate of conifers and lifespan in the first 50 years of life; that is, fast early growth was associated with decreased lifespan (Bigler, 2016).
Following the "defence and storage" concept, shade-tolerant species partition a major fraction of photosynthates to internal stores at the expense of rapid growth (Kitajima, 1994;Kobe, 1997).
In our study, F. sylvatica trees had a lower NSC content during slow growth compared with the less shade-tolerant Acer spp., thus contradicting the "defence and storage" concept. Similar results as in our study were found for Acer saccharum (Kobe, 1997) and for evergreen shade-tolerant species (Lusk & Piper, 2007;Piper et al., 2009); however, no difference has been reported for other deciduous species-Castanea crenata and Quercus mongolica (Imaji & Seiwa, 2010). On the one hand, early spring leaf-out of A. pseudoplatanus compared with juvenile F. sylvatica produces more photosynthates before canopy closure (Vitasse, 2013).

A high concentration of NSCs in
Acer spp. may also reflect high levels of browsing and defoliation, as starch and sugars are used to survive periods of a negative net carbon balance after defoliation (Myers & Kitajima, 2007) or stem loss (Latt et al., 2000). On the other hand, faster-growing juvenile trees increase total storage (Canham et al., 1999;Niinemets, 1998).
In our study, both Acer spp. had higher AGR values than F. sylvatica and also maintained higher NSC content, while F. sylvatica showed the opposite (Figure 2).
Assuming that light had not improved and LAR reached its carrying capacity in carbon supply to growth (point 2) and storage (point 3, Figure 2), a tree can have two strategies: (a) slow down growth to save storage while keeping LAR high; or (b) deplete storage to maintain growth while decreasing LAR. Many empirical studies have shown that storage is prioritized over growth under carbon limitation (Weber et al., 2018;Wiley et al., 2013). A replenishment of a certain level of NSCs before growth leads to allocation of carbon first to storage and then to growth (Imaji & Seiwa, 2010;Weber et al., 2018). However, larger saplings require more storage to support the increasing operational costs of tissue maintenance and defense (Wiley & Helliker, 2012). This is why allocation to storage may increase disproportionately compared with allocation to growth (point 1, Figure 2), leading to a trade-off between growth and storage.
Under environmental conditions leading to carbon limitation (long-term deep shade or sustained severe defoliation), a trade-off between storage and growth is possible (Palacio et al., 2014), causing carbon starvation and tree death (to the right of point 1, Figure 2) (Weber et al., 2018). Presently, it is unclear whether saplings deplete their NSC reserves before growth reduction or death in the shade.
In an experiment by Weber et al. (2019), mortality of shaded F. sylvatica seedlings (0-60 cm tall) occurred after NSC concentrations in the stem dropped to approx. 3% (dry mass basis) after insect herbivore attack. In our study, the mean NSC concentration in the stem of low-vitality F. sylvatica seedlings was 7.8% of dry mass for the same height class (comparison is valid if the method is the same). The mortality of A. pseudoplatanus happened at less than 1% (dry mass basis) NSC concentration in the study by Weber et al. (2018), while our low-vitality A. pseudoplatanus seedlings maintained mean NSC levels of 9.7% (dry mass basis) for the same height classes. In our study, we did not observe growth reduction, as LAR was still able to support both growth and storage, hence lethal carbon starvation could not be assumed. However, unlike trees in the canopy, young trees do not have the opportunity to replenish NSC reserves in autumn (Hoch et al., 2003) because leaf senescence of trees and regeneration occurs at same time (Varsamis et al., 2019;Vitasse et al., 2009

| Species biomass allocation and tree vitality
Biomass allocation of trees with low and high vitality differed significantly in our study, reflecting different performance in shade.
Biomass allocation patterns to leaves (LMF, leaf area, LAR) and storage pool (NSC) define vitality, whereas AGR and height increment do not and do not differ between low-and high-vitality regeneration (Table S3). In addition, similar height increments for all species do not indicate shade avoidance (Henry & Aarssen, 2001). In previous studies, F. sylvatica saplings in the height class 201-500 cm featured greater leaf areas compared with Acer spp., while greater biomass allocation to branches in F. sylvatica led to larger crowns compared with the more slender form of Acer spp. (Annighöfer et al., 2017;Petriţan et al., 2009). Both Acer spp. studied here, even in the larger height classes, tend to grow long petioles over branches (Beaudet & Messier, 1998). Such kind of leaf display may be an adaptation to browsing (Modrý et al., 2004), enabling minimization of biomass loss by allocating less biomass to branches and leaves and concentrating it on the top to avoid self-shading. In our study, F. sylvatica and Acer spp. of low vitality had a smaller leaf area and crown volume than high-vitality trees. As a result of crown decline and reduced leaf area, the level of NSCs was reduced in trees with low vitality (

| Limitations
Although our model offers a plausible explanation for the regeneration growth processes, it is, nevertheless, linear and can thus be applied only to the period of intensive growth when response and explanatory variables are developing mostly linearly. Furthermore, it assumes equal annual AGR and shoot growth, which may be theoretically possible but in reality varies with tree age and size (Gibert et al., 2016). The model does not take into account NSC pool of branches and the seasonal variation, an aspect that might be needed for a better understanding of the annual storage balance. Moreover, NSC concentrations were scaled to the whole stem and root. This is justified for relatively small saplings with wood consisting almost entirely of sapwood but seems problematic for saplings with larger vertical and horizontal NSC gradients within the stem.

| CON CLUS ION
The proposed integrated model of shade tolerance explains the longer survival time of F. sylvatica juveniles compared with Acer spp.
species in deep shade. It combines traits from the "carbon gain" (LAR and AGR) and the "defence and storage" (NSC) hypotheses. Despite mechanistic approaches for the explanation of trait development, the model leads to inferences about the survival time of young trees without its direct measurement. Due to shifts in the above three traits with increasing tree height and age, juvenile trees may increasingly face a trade-off because a diminishing LAR becomes insufficient to produce the photosynthates needed to support both growth and storage. In this case, a tree can either reduce growth to retain storage, or deplete storage to achieve growth. The ability of a species to balance LAR, AGR, and NSC to postpone or avoid this trade-off defines its shade tolerance and thus its regeneration survival time.
Fagus sylvatica is able to minimize both AGR and NSC, maintaining a high LAR, while the two Acer spp. cannot reduce storage and/or growth. The increased storage in Acer spp. may be explained by early leaf-out in spring and a focus on defense, sacrificing investment into leaf and branch biomass.
Our findings only partly confirm the "carbon gain" concept: beech optimizes carbon gain with an extensive leaf display, large crown volume, and slowly decreasing LAR, but it grows slowly in the shade. However, with only storage in focus, our results also cannot fully support the "defence and storage" concept, as shade-tolerant F. sylvatica stores less NSCs than its competitors.

ACK N OWLED G M ENTS
We are grateful to Natalia and Vita Mayor, as well as Vasyl