Above- and below-ground trait coordination in tree seedlings depend on the most limiting resource: a test comparing a wet and a dry tropical forest in Mexico

Study sites

This study was conducted in two Mexican tropical forests: one located in the Montes Azules Biosphere preserve (16°04′N; 90°45′W) in Chiapas State, within the Lacandon region, and the other in the Chamela-Cuixmala preserve (19°30′N, 105°03′W) in Jalisco State, Mexico (Figs. S1, S2). Montes Azules Preserve covers an area of 331,200 ha, with an altitude range of 80 to 1,750 m a.s.l. This rainforest receives 3,000 mm of annual precipitation, distributed mostly between May and January, and a 3 month dry season with <100 mm of precipitation per month (Ibarra-Manríquez & Martínez-Ramos, 2002). The understory is highly shaded, receiving on average 1.83% of incident radiation, and soil fertility is highly spatially variable (Table S1). The seedlings obtained from this tropical moist forest were distributed in an area of low hills over humic acrisol soils, with depths of 55 to 65 cm and moderate drainage (Table S1) (Ibarra-Manríquez & Martínez-Ramos, 2002). The Chamela-Cuixmala preserve is located on the Mexican Pacific coast and is dominated by a tropical dry forest that extends over low hills with an altitude range of 300 to 800 m above sea level. The mean annual precipitation ranges between 400 and 1,100 mm and occurs mainly from July to October, resulting in a 7-month dry season (November to June); most species lose their leaves in response to drought (Lott, Bullock & Solis-Magallanes, 1987). The forest understory is relatively open, receiving 13.7% of total incident radiation during the rainy season, while soils are generally stony and highly variable in depth (between 10 and 70 cm) and fertility (Cotler, Durán & Siebe, 2002). Given that soil nutrient content in our studies sites is highly variable spatially, and available nitrogen content overlapped between the two forest types, the main physical differences between the forests are related to water availability and light in the understory, as can be seen in Table S1.

Plant material and traits

For the moist forest, during the rainy season, we collected seedlings of 43 dominant species (42 trees and one liana, see Table S2) that had no senescent cotyledons and at least one pair of leaves (average 21 seedlings per species, range 4–60). We carefully extracted the seedlings from the soil to preserve the integrity of all root tissues (following (Paz, 2003)). For the dry forest, we selected 52 woody dominant species (Table S2). It was not possible to extract the entire root system of seedlings because soils are thin and stony, so we collected seeds during the fruiting peak of each species. Seeds were stored in paper bags in the lab at room temperature until all of the seeds had been collected. We germinated the seeds in forest soil beds in a shade house, and five days after seedlings emerged, 16 seedlings per species were transplanted to 4.6 L (14 cm diameter × 30 cm tall) plastic bags with basal drainage, using forest soil as substrate. Seedlings were grown in a greenhouse for 3 months at 23.4 C (range 14–41 C) and 62% (40–83%) relative humidity, daily average photosynthetic photon flux of 805 millimoles/m²/s and maintaining high volumetric soil water content (20%). These conditions fall within the range of variation detected in the mature forest floor during the rainy season (Pineda-García, Paz & Meinzer, 2013). See details of plant collection or propagation in Supplemental information.

The seedlings were washed carefully with tap water to remove soil particles from the roots. Maximum depth of the root system (MRD) was measured by extending the radicular system over a table and measuring the maximum length from the upper to the lower part with a ruler. Seedlings were divided into roots, stem, and leaves. Leaves were wrapped in moist paper towels and stored in sealed plastic bags for 12 h at 4 °C and after that time their saturated weight was obtained. Leaves where then digitized on a flatbed scanner (EPSON Expression 10000 XL, Japan). Roots were spread in a glass tray with water, and if necessary, cut to minimize overlap, and digitized with the same scanner. Leaf area was determined using Image J software (Rasband, 2014) and total root length and average root volume and diameter were measured with WinRhizo ( Arsenault et al., 2018). A basal fragment of the stem of each seedling was cut under water and saturated in distilled water for 12 h. We obtained its saturated weight and its volume by the water displacement method. We removed the bark and obtained the wood weight and volume in the same way. Leaves, roots, wood and bark samples were then oven dried (leaves and roots for 48 h at 60 °C and wood and bark for 72 h at 70 °C) and we obtained their dry weight. Finally, we used these measurements to calculate the leaf, stem and root traits shown in Table 1. For the dry forest, we had leaf and stem data for 52 species and root data for 28 of those species, while in the moist forest all species had all traits (Table S2). We used the species mean values as data points for the statistical analyses.

Statistical analysis

Before analysis, we assessed trait data normality independently for each forest. We performed log 10 transformations for all variables except for root density (RD) and maximum root depth (MRD) data from the dry forest (Chamela), and leaf thickness (LTh) and wood density (WD) data from the moist forest (Montes Azules). We assessed the influence of total plant biomass on each trait using linear regressions. In cases where this association was significant (p < 0.05), we used the residuals in lieu of the uncorrected trait value, to avoid the effect of size on traits. Hereafter we will refer to both original trait values and these residuals as “traits” for simplicity.

Bivariate and multivariate trait relationships in the moist and the dry forest

Multivariate analysis (Fig. 1, Table S3) showed a positive association between carbon investments in leaf, stem, and roots in both forests, showing a trade-off between organ density and water content. This is evidenced by a positive association between stem and leaf water contents (SWC, LWC) and a negative relationship of these traits with WD. Interestingly, although the signal of coordination with roots was present, it was less clear (Fig. 1). Specific root length (SRL) was directly related to leaf and stem water contents in the moist forest, while maximum root depth (MRD) and root density (RD) were related to WD in the dry forest. When using phylogenetic independent contrasts (PICs), most trait relationships remained (Fig. 1, Fig. S3). In the moist forest only, we found a positive correlation between SLA and SRL, traits which are associated with the efficiency in the acquisition of resources above ground and below ground, respectively (r = 0.38, p = 0.01, Fig. 2).

In addition, we found some correlations that were significantly different between forests (Fig. 2 and Fig. S4, Table S4). The strongest differences were found in the correlation between RTh and SRL (z = 4.61, p = 0.000), and between RTh and LWC (z = 2.82, p = 0.005), both of which were negative in the moist forest but positive in the dry forest. Other differences involved the MRD, which showed a significantly stronger positive relationship with WD (z = 2.80, p = 0.005) and a stronger negative relationship with SWC (z = 2.19, p = 0.005) in the dry forest (Fig. 2 and Table S4). There was also a slight difference between forests in the correlation of SLA with RTh (negative in the moist forest, positive in the dry forest, z = 1.94, p = 0.053), and SWC (positive in the moist forest and absent in the dry forest, z = 1.9, p = 0.058) (Fig. 2 and Table S4).

Trait integration through network analysis

When considering forest networks formed by significant Pearson correlations (Table 2 and Fig. 3), the modularity values were intermediate in both forests (moist forest = 0.175; dry forest = 0.207) but were sufficiently high to identify two highly interrelated trait modules in each forest type (Fig. 3). In the moist forest, one module was formed by leaf and stem traits—LTh (leaf), MPU (leaf), LWC (leaf), and WD (stem)—while the other module was formed by leaf, stem, and root traits—SWC (stem), SLA (leaf), SRL (root), RD (root) and RTh (root). In the dry forest, one module was formed by leaf, stem, and root traits—LWC (leaf), MPU (leaf), LTh (leaf), MRD (root), SWC (stem), and WD (stem)—while the other module was formed only by root traits—RD (root), RTh (root), and SRL (root).

Principal component analysis (PCA) of the three centrality metrics (degree, betweenness centrality, and closeness centrality, Table 2) accounted for 93% of the variability in the dry forest and 85% in the moist forest. The highest PC1 scores in both forests corresponded to leaf water content (LWC, Fig. 4), indicating that this is a highly important trait, connected with many other traits by multiple direct and indirect pathways. Interestingly, the correlation between the first principal component (PC1) scores of each forest centrality PCA showed that the most important traits in one forest type were not necessarily the most important in the other forest (r ² = 0.55; p = 0.09), indicating that the most important linking traits differed between the two environments (Fig. 4). For example, while MRD showed marginal importance in the moist forest (only weakly coordinated with LWC), this same trait was highly important in the dry forest and showed multiple relationships with leaf and stem traits (Figs. 3 and 4). Likewise, while specific root length (SRL) had marginal importance in the dry forest, this trait had high importance in the moist forest (Fig. 4).

Coordination is similar between forests, but trait relationships are not

We hypothesized that water shortage in the dry forest selects for stronger coordination among leaf, stem, and root traits, while shade in the understory of the wet forests favors strong coordination only between leaf and stem traits. However, we found no support for such hypothesis; in both forests we detected modules that included traits from all three organs. This is similar to findings by Flores-Moreno et al. (2019), who also found high connectivity across trait networks within and between tissue types in both tropical and temperate forests. A novel contribution from our study is the use of network analysis to study inter-trait coordination not only between stem and leaf traits, but also with root traits. Interestingly, the finding that the pattern of coordination of key traits was different between the dry and the moist forest supports the hypothesis that the selective value of the same traits differs under different conditions (Dwyer & Laughlin, 2017; Flores-Moreno et al., 2019). For example, in the moist forest we found weak coordination between the morphological efficiency to capture resources above ground and below ground (i.e., direct correlation between SLA-SRL, Fig. 2), but these traits were not correlated in the dry forest. The existence of a growth-defense trade-off has been well described among moist forest species (Poorter & Bongers, 2006); fast growth in gaps is promoted by cheap, short-lived, and physiologically active leaves (indicated by high SLA), while high survival in the forest understory is enhanced by the formation of long-lived, tough leaves that reduce herbivory, mechanical damage, or leaf turnover (Poorter & Bongers, 2006). The coordination between SRL and SLA detected in our study (Figs. 2 and 3), suggests that the growth-defense trade-off involves both above- and below-ground morphological traits. Species with high SLA that are adapted to rapidly acquire light, seem to also have roots that are capable of absorbing soil resources efficiently. Meanwhile, species with high investment in structural defense of leaves (low SLA), tend to also have structurally well-defended roots (low SRL). However, the weak SLA-SRL correlation (r = 0.38, p = 0.01) and the inconsistency of the relationship between these traits in published literature (Weigelt et al., 2021) may be related to the fact that SRL depends on root density (RD) and root thickness (RTh), and plants can construct roots with many combinations of these values. In addition, RTh is also related to symbiosis with mycorrhizal fungi. Consequently, roots vary not only along a conservation-acquisition trade-off, but also, and orthogonally, along a collaboration-do it yourself trade-off (Bergmann et al., 2020). The role played by symbiosis in root traits (SRL and RTh) needs further investigation in this moist forest.

The lack of SLA-SRL coordination in the dry forest may also reflect the fact that the development of dense tissues with high carbon investment is not necessarily the predominant strategy to deal with drought. Commonly, drought-deciduous species have short-lived, low-cost leaves with high SLA, coupled with low-density stems and thick root tissues containing high water and carbohydrate reserves (Powers & Tiffin, 2010; Pineda-García, Paz & Tinoco-Ojanguren, 2011; Palomo-Kumul et al., 2021). Conversely, maximum root depth (MRD) was coordinated with leaf and stem traits in the dry forest, but not in the moist forest (Fig. 3). The relevance of MRD in this and other dry forests is not surprising, given the importance of this trait for seasonal water uptake in seedlings in arid environments (Ackerly, 2004; Maeght, Rewald & Pierret, 2013), especially during drought periods (León et al., 2011; Padilla & Pugnaire, 2007). Furthermore, the coordination of dense stems with deep roots in the dry forest (Figs. 2 and 3) highlights the importance of a more permanent source of water when stem storage capacity is low (Schwinning & Ehleringer, 2001; Hasselquist, Allen & Santiago, 2010; Paz, Pineda-García & Pinzón-Pérez, 2015). Likewise, thick roots (RTh) were directly related with high SRL and leaf water content (LWC) in the dry forest (and inversely in the moist forest), an indication that RTh may be related to water economy in the dry forest. Commonly in dry forests, roots store important amounts of water in their tissues (Paz, Pineda-García & Pinzón-Pérez, 2015), while in moist forests thick roots are commonly fibrous.

Although our moist forest seedlings were obtained directly from the field while dry forest seedlings were grown in pots, we are confident that this did not affect root trait measurements, particularly MRD. First, in our study of dry forest plants, we rarely observed roots reaching the bottom or lateral edges of the pots, indicating the soil volume did not impede root growth in a specific direction. Second, in a previous study where MRD was measured in four neotropical forests with the same methodology, extracting seedlings from the ground, roots were deeper in the site with the longest dry season (Paz, 2003), similar to our study.

Trait centrality

When considering the traits’ centrality values (the relative importance of each trait as a connecting node), we found that LWC was the most important trait in both forests (Fig. 4), reflecting the key role of leaf hydration in plant growth and many other physiological processes, regardless of forest type. Previous studies have proposed that in addition to leaf turgor, maintenance of leaf water content may be critical for growth processes such as cell elongation and division, as well as for leaf gas exchange (Bartlett, Scoffoni & Sack, 2012). In addition, leaf water content has been found to be associated with hydraulic traits at the stem level, such as stem water content and stem conductivity (Pineda-García et al., 2015) and is a good predictor of growth and survival under dry conditions in other tropical forests (Cifuentes et al., 2020). The frequent correlations of wood density with hydraulic and leaf traits in adult trees (Santiago et al., 2004; McCulloh et al., 2011; Greenwood et al., 2017, among others) have led to the idea that wood density could represent a central trait affecting the stem and leaf economy (Chave et al., 2009; McCulloh et al., 2011; Méndez-Alonzo et al., 2013). However, strikingly, in our study we detected strong evidence suggesting that among tropical seedlings, LWC may be another key functional trait; this is an idea worth testing in other tropical forests.

Interestingly, SWC was the second most important trait in the dry forest, but not in the moist forest. This may be explained by the close association of SWC with the water storage capacity versus soil vertical foraging and water exploitation versus drought tolerance trade-offs described in tropical seasonally dry forest (Paz, Pineda-García & Pinzón-Pérez, 2015; Pineda-García et al., 2015). Drought avoiders, which have high photosynthetic rates, xylem hydraulic conductivity, and growth rate when water is available (Pineda-García et al., 2015), maintain a narrow safety margin between plant water potential and P₅₀ (the potential that would induce 50% of hydraulic conductivity loss by the formation of emboli) (Pineda-García, Paz & Meinzer, 2013; Markesteijn et al., 2011b). Given that these species tend to have shallow roots, they have evolved a great capacity to store water in the stem and leaves, which allows them to survive as the soil desiccates during the dry season (Paz, Pineda-García & Pinzón-Pérez, 2015). On the contrary, drought-tolerant species, which have lower photosynthetic rates, xylem hydraulic conductivity, and growth rate, have a denser stem with limited water storage capacity (low SWC) (Pineda-García et al., 2015; McCulloh et al., 2019). Due to their lower capacity to decouple hydraulically from the soil, these dense-tissue species are associated with deep roots that penetrate deeper into the soil and rock interstices and rely on a more constant, although unsaturated, soil water content (Schwinning & Ehleringer, 2001; Padilla & Pugnaire, 2007; Zhou et al., 2020). Thus, differential responses to drought explain the importance of SWC, a trait strongly involved in water economy in the dry forest.

As expected, SLA had very low centrality in the dry forest (Fig. 4), where competition for light is likely not as strong as competition for water, and where SLA is strongly related to non-morphological traits such as leaf phenology. However, the low centrality of SLA in the moist forest is intriguing. This lack of finely tuned connection between a central trait in the leaf economy spectrum and other leaf and stem traits suggests that different combinations of leaf traits and plant architectures can yield similar capacities for growth in shaded forest (Valladares, Skillman & Pearcy, 2002). This hypothesis needs further investigation in our study system. Contrary to expectation, SRL, a trait typically claimed to be a key morphological determinant of the efficiency of carbon investment in water absorption, was poorly connected in the dry forest, but highly connected in the wet forest (Fig. 4). Although it is possible that this was due to differences in soil nutrients between forests, this seems unlikely since nutrient levels overlap strongly between the study sites. Our field observations suggest that in the dry forest, high values of SRL may be indicative of different functions depending on the species: fine roots deploying large absorptive surfaces, or thick roots with high water and low carbon contents acting as water storage more than for water absorption. Together, the lack of SRL centrality and the clear correlations of MRD with leaf and stem traits in the dry forest suggest that in habitats with a high risk of drought at the seedling stage, developing deep roots has a higher selective value than developing thin, efficiently absorptive roots (Padilla & Pugnaire, 2007; León et al., 2011; Paz, Pineda-García & Pinzón-Pérez, 2015). Conversely, in the moist forest, in the context of strong competition for light, SRL and WD are important under a growth-survival trade-off, where species that acquire soil resources efficiently grow fast (high SRL) and have low-density stems (low WD), and vice versa (Poorter, 2009; Pineda-García et al., 2015).