Introduction
Seed size (here is defined by mass or geometric size metrics, such as volume, length, width, and thickness) is a trait central to many aspects of plant ecology. It is a key trait for understanding and predicting ecological processes in the early phases of a plant’s life cycle, for example, dispersal, persistence or dormancy, germination, and establishment (Leishman et al., 2000; Fenner and Thompson, 2005; Moles et al., 2005; Saatkamp et al., 2019). Seed mass is a widely applied metric of seed size in many studies of plant functional traits (Moles, 2018). However, geometric metrics, such as seed volume and seed length, are measured in most studies relevant to dispersal ecology (De Jager et al., 2019) and animal foraging behavior (Yi and Wang, 2015). Often, both seed size metrics and seed mass are not reported in the same studies; hence, we do not have enough information to accurately interchange between them (Kitamura et al., 2002; Moles et al., 2005; Pérez-Harguindeguy et al., 2013; Bello et al., 2015; Díaz, et al., 2016). We propose to develop an allometric equation that can predict seed mass from seed size and vice versa. Generally, the best-fit allometric model is the relationship between seed mass and volume, whereas the relationship with geometric measures, such as seed length or shape, vary among studies (Brewer, 2001; Wright et al., 2007; Casco and Silva Dias, 2008; Wang and Chen, 2009; Ganhão and Dias, 2019). Furthermore, most studies are biased toward Neotropical forests (Wright et al., 2007).
Ecologists have sought to explain variation of seed size within and between plant communities (Coomes and Grubb, 2003; Moles et al., 2007), because variation in seed size is crucial for understanding traits, such as competitive ability (Leishman and Murray, 2001; Turnbull et al., 2004), tolerance to environmental stressors (Baker, 1972; Westoby, 1998), seed predation (Thompson, 1987), seed dispersal (Greene and Johnson, 1993), and dormancy (Harel et al., 2011). In a tree community, seeds of pioneer species are normally smaller than those of shade-tolerant species (Ghazoul and Sheil, 2010). Seed size variation between pioneer vs. shade-tolerant species needs to be reinvestigated based on an objective classification, and systematically assessed via the probability of species occurrence in second- versus old-growth forests (e.g., Chazdon et al., 2011).
In fleshy-fruited plants, seed dispersers are regarded as important drivers of seed size, especially in tropical forests, where >70–80% of plant species are dispersed by animals (Osuri et al., 2016; Chanthorn et al., 2019). Generally, the achievement of seed dispersal is determined by whether a seed can pass into the mouth of a frugivore and be taken away from a fruiting tree. A key disperser trait is gape width, which should be wider than seed width (Fleming et al., 1993). Although many physiological and ecological factors may produce selection on seed size, the availability of effective dispersers may place an upper limit on seed size. Strong evidence that gape width of frugivores exerts selective pressure on seed size comes from a study showing a reduction of seed size in Neotropical forests with high defaunation intensity of large-bodied dispersers compared to non-defaunated forests, a phenomenon that has occurred only within the last 100 years (Galetti et al., 2013). However, since gape width data are difficult to collect, body size is often used as a proxy. The study of “anachronisms,” for example, demonstrated the consequences of large animal defaunation during the Pleistocene era that affected the dispersal of ill-suited plants with very large seed size (Janzen and Martin, 1982; Pires et al., 2017; Lim et al., 2020). As this topic has been studied mostly in Neotropical forests (Fuzessy et al., 2018; Lim et al., 2020), studies from other tropical regions are needed to gain a global perspective on this subject.
Coordination (positive correlation) and trade-off (negative correlation) of functional traits are common features in plant communities (Díaz, et al., 2004; Moles, 2018), including tropical forests (Wright et al., 2010). Coordination between tree height and seed size has been reported in a global study (Díaz et al., 2016) and in a study focusing only on trees in Neotropical forests (Bello et al., 2015). For the trade-off, species with a high leaf area normally have lower wood density, which supports fast growth in high-light environments (Chave et al., 2009). Intuitively, large seeds tend to generate large seedlings and provide an adaptive advantage to seedlings growing in shade environments, but the relationship between seed and leaf size lacks direct causation and is predicted to be weaker compared to that between seed size and plant height (Díaz et al., 2016). Alternatively, another global study showed a strong correlation between seed size with leaf life span (Moles, 2018). Interestingly, a cross-continental study found a strong relationship between seed size and leaf toughness (fracture resistance), suggesting another functional trait coordination of importance (Kraft et al., 2015).
Here, we compiled data on seed size for a majority of species in a tropical forest community, which was challenging because of the irregular fruiting or masting of most tree species (Suwanvecho et al., 2017; Kurten et al., 2018). Lianas are equally important and typically comprise about one-third of woody plant diversity in tropical forests (Schnitzer, 2018). Studies of lianas have lagged those of trees, and only a few studies have examined seed size or mass in more than a minority of liana species (e.g., Gallagher and Leishman, 2012).
Here we collected a seed size dataset of >5,000 seeds measured in five metrics: seed mass, seed volume, seed length, seed width, and seed thickness, covering >98 and > 70% of tree and liana stems, including many liana species not previously reported in TRY global databases (see Supplementary Table 3). We have, in addition to the Mo Singto plot, which is covered with relatively old-growth forest, a chronosequence of smaller plots covering all successional stages in the same landscape, allowing us to classify all tree species into pioneer, shade-tolerant, and generalist types (Chazdon et al., 2011). The life-history traits available for analysis fall into four groups: leaf traits (leaf area, leaf toughness, specific leaf area, and leaf greenness), structural traits (maximum plant height and wood specific gravity), successional niche, and seed dispersal agents or mean body size of frugivores.
The objectives of our study were to create allometric equations among seed size metrics, to examine phylogenetic influence on seed size variation, and to evaluate the relationship (coordination or trade-off) between seed size and plant traits by testing four hypotheses: (1) Large seed size is coordinated with other plant traits that are associated with shade tolerance and longer leaf life span, as opposed to faster growth; (2) Large seed size is associated with old-growth forest as opposed to early successional forest; (3) seed size is correlated with disperser size, as effective dispersal is the initial major hurdle to overcome in a plant’s life history; and (4) trees and lianas are expected to show different seed size–life-history trait relationships, because although reproductive lianas occupy similar light environments in the forest crown, they have different growth trajectories and means of support. We compared models with and without phylogenetic correction to examine the independence of variables and perhaps gain insight related to evolution.
Discussion
Allometry and Phylogeny
Seed size is a major trait in two large research topics: functional traits and seed dispersal (Jansen et al., 2004; Wang and Chen, 2009; Wang and Yang, 2014). However, data are generally created from different seed size metrics, for example, seed mass and geometric size (volume, length, width, thickness). Measuring two metrics of seed size together can be a laborious task, but our allometric equation helps to reduce this problem. We found high interchangeability of our models because of well-fitting log-linear models using seed mass and volume. Studies in Neotropical forests show well-fitted models similar to ours (Wright et al., 2007). Contrary to our expectation, wind dispersal did not create systematic error. However, based on our visualization, there was a cluster of highly negative residuals from the single species, Toona ciliata. Moreover, our best selected model was the one without any phylogenetic effects, suggesting that our allometric equations could apply in other seasonal tropical forests of Southeast Asia where our forest is well represented and shares similar phylogenetic structure (Ashton, 2014; Brockelman et al., 2017).
Species with similar seed size were not clustered in certain lineages on the phylogeny tree. They were distributed among several lineages and groups of lineages across the phylogenetic trees. In our visualization (Figure 4), we did not see any clear pattern between Gymnosperms and Angiosperms. This contrasts with the discovery of large differences between both groups in the global study of Moles et al. (2005). However, note that our Gymnosperms comprise only four species belonging to two genera, a significant limitation of our study.
Trait–Trait Relationships: Driver and Trait Coordination
We interpret a significant positive correlation as either a driver or trait coordination, but interpret a negative correlation as a trade-off. Our results for tree species found only significant positive correlations. The strongest positive relationship was between mean body size of frugivore dispersers and seed size, without any phylogenetic effect. We interpret this trait as a major driver of seed size variation, acting as a constraint. This is a consequence of natural selection in which frugivores tend to feed on fruits with the largest seeds that they can swallow, for single-seeded fruits, such as drupes. Selection on seed size driven by frugivore gape size may occur within a few centuries (Galetti et al., 2013; Dirzo et al., 2014). In Southeast Asia, the seed size of Choerospondias axillaris, a primarily deer-dispersed species (Brodie et al., 2013), is smaller in the north of Thailand where deer have been extirpated for ca. 80 years compared to our study site, where deer are still abundant. A possible consequence is secondary extinction (Brodie et al., 2014), which may also effect ecosystem services, such as carbon sequestration (Osuri et al., 2016; Chanthorn et al., 2019). Evidence from Neotropical forests, that is, “anachronisms,” suggests that the extinction of large-sized frugivores during the Pleistocene era may have led to reduction of opportunities for long-distance dispersal in many Neotropical trees with very large-sized fruit (Janzen and Martin, 1982).
The positive correlation between nearly all traits with seed size suggests that traits may be physiologically co-adapted either with or without any mechanistic links. The positive significant relation between leaf area and chlorophyll content (indicated by leaf greenness) may be explained as co-evolved traits which support the energy demand of large seed size. Normally, a large leaf has a high photosynthetic rate, indicating a fast-growing strategy, especially in pioneer species (Poorter and Bongers, 2006; Reich, 2014). In our old-growth forest, species with large-sized leaves may have been selected to produce more energy to sustain large seeds. Although this trend is similar to observations in the global study (Díaz et al., 2016), there has been little explanation for it. Species with large leaves are typically pioneer species (Reich, 2014), which may have been selected to produce large numbers of small seeds more rapidly. In Southeast Asia, Macaranga spp. are examples of pioneer trees that support this hypothesis. However, leaves of dominant or common species in our secondary forests are not as large as those of Macaranga or Hibiscus (Chanthorn et al., 2017). The leaves of our pioneer species seem to be smaller and tend to have lower specific leaf area (see Supplementary Figure 2). Moreover, there was a strong negative correlation between seed size and successional niche in the secondary forest, suggesting that many pioneer species produce small seeds coordinated with small leaf size. Also, in meta-analyses of functional traits, plant size and seed size pattern correlate positively along the same principal component axis (Díaz et al., 2004, 2016). Additionally, this relates to vertical niche position in the forest canopy, or to the expectation that a canopy tree may have larger leaves to take advantage of greater exposure to the sun. We also found a positive correlation between maximum tree height and seed size. This is because the old-growth forest has taller trees, and trees that require larger seeds that have a better chance to survive in the shady understory. Although the energy demand of large seeds is supported by high leaf chlorophyll content, the exclusion of specific leaf area (SLA) among all leaf traits from the regression model suggests that larger leaf size is required by a higher photosynthetic rate rather than needed for storing substances or strengthening leaf structure. Furthermore, SLA is largely independent of leaf area (Díaz et al., 2016). Studies at regional and community scales have found that SLA is weakly related to plant height and seed size (Díaz et al., 2004) and that it often shows no relationship with wood density and leaf size across many species and community types (Wright et al., 2010). Consistent with our result that large seeds coordinate with large leaf area and high chlorophyll content, the relationship with SLA is very weak.
Interestingly, our strongest leaf economic trait was toughness (justified by the p-value). Our result confirms the trend found in other studies which compared forests in three continents including a tropical rainforest of Southeast Asia (Kraft et al., 2015). This trait coordination reflects the syndrome associated with a slow life-history/reproductive strategy characteristic of large-seeded species adapted to resource-limited environments, with prolonged leaf lifespans and well-defended leaves (Moles et al., 2005). These findings provide supporting evidence for correlated components of the life-history strategy, between seed size and mechanical leaf defense (Dalling et al., 2020).
However, all these relationships were absent in lianas. Leaf trait variability was generally lower in lianas than in tree species in the plot. This is likely because most mature liana species grow to the top of forest canopy, where their leaves are exposed to the same high-light conditions as tall trees (Schnitzer, 2018). Lianas are structural parasites, which germinate on the ground, stay rooted until finding other plants to use as structural support (Visser et al., 2018). Due to this growth form, lianas can acquire resources faster than trees, because they invest fewer resources in structure, and more into productive leaves and reproductive biomass (Collins et al., 2016; Schnitzer, 2018; Werden et al., 2018). For structural traits, such as height, the strong positive correlation was also similar to that found in the global study (Díaz et al., 2016). The explanation is the fecundity trade-off throughout the plants’ lifetime, in which large trees compensate low fecundity with larger seed size and high survival throughout their longer life spans (Moles, 2018). In addition, the seed volume of second-growth species was smallest, which may also be explained by the fecundity trade-off. Smaller seeded species have the fecundity advantage which allows them to disperse widely and recruit in light gaps and forest edge environments (Rose, 2005).
Wood density has been proposed as one of the most important functional traits organizing tropical forests at the population level, affecting species survival and competitive ability (Chave et al., 2009). We expected that species with large seed size would tend to have a high wood density (stem specific gravity) There is a well-supported tendency for large hardwood species to have larger fruits and seeds (Wright et al., 2007; Bello et al., 2015). However, this relationship was not significant in our study, suggesting that more studies from Southeast Asia are required to support a pattern different from that in Neotropical forests. For instance, the study by Krishnan et al. (2019) showed that tropical dry-forest trees with large seeds and low wood density had greater survival rates under drought conditions in the Indian subcontinent, which should allow them to cope with global change more efficiently. Several studies have demonstrated the impact of defaunation on carbon storage, which is based on an underlying coordination between wood density and seeds size in many tropical forests, including Asia (Osuri et al., 2016). In our forest, the wood density of trees dispersed by large-bodied frugivores tends to be higher than that of species dispersed by small birds (Chanthorn et al., 2019). However, some species with large seeds are not dispersed by large-bodied frugivores, for example, species belonging to the family Fagaceae and those with unknown dispersal mechanisms, Excoecaria oppositifolia (Euphorbiaceae) and Gonocaryum lobbianum (Cardiopteridaceae; Chanthorn et al., 2019).