Seed germination, a central seed trait, is a key bottleneck in plant life history, potentially affecting the distribution and abundance of species in communities (Guo et al., 2000; Souza and Fagundes, 2014). Most life history traits are correlated (Moles and Leishman, 2008; Norden et al., 2009), and some traits may affect seed germination percentage. Understanding how life-history traits are related to germination percentage is crucial for understanding the evolution and ecology of seed germination strategies. Although relationships between life-history traits and germination have received considerable attention, most studies have focused on individual species without scaling up to a community level (but see Liu et al., 2007; Xu et al., 2014). Furthermore, variation in germination strategies among species in a community may support higher diversity by allowing temporal partitioning of environmental resources and providing a buffer against extinction (Chesson, 2000; Gremer and Venable, 2014). Therefore, understanding germination strategies of seeds within a single community has implications for trait evolution as well as population and community dynamics.

Seed mass has long been regarded as an important aspect of plant reproductive biology (Baker, 1972; Moles et al., 2005, 2007). Seed mass influences many different life-history stages in plants. For instance, small-seeded species have low rates of seedling survivorship (Moles and Westoby, 2004), and they tend to have high seed output but short lifespans (Moles and Westoby, 2004). Relationships between seed mass and germination are less well understood than relationships between seed mass and other life-history stages. Differences in seed mass among species are due in part to different levels of starch and endosperm nutrients, and may influence germination percentage (Chen et al., 2002; Soriano et al., 2011). Studies have shown that seed mass may affect germination percentage (Reich, 1994; Paz et al., 1999; Bu et al., 2007; Xu et al., 2014). Regardless, it is widely accepted that large seeds generally have a higher germination percentage than small seeds. However, there is little information on the relationship between seed mass and germination in arid or semi-arid ecosystems, especially, the degraded grasslands in arid environments.

Seed shape, which varies in three dimensions, is another important trait that affects germination (Weiher et al., 1999; Tekrony et al., 2005). Theoretical studies have predicted that species with elongate and flat seeds should germinate more frequently than those with round seeds (Grime et al., 1981). Elongate and flat seeds are more likely to experience post-dispersal predation than round seeds, and should therefore show prompt germination to avoid risks of mortality. However, strong empirical data supporting this prediction are lacking (Liu et al., 2007).

It is not yet fully known how seed germination percentage is related to various factors. Germinability has been predicted to be associated with phylogeny, seed traits, and ecological factors (Bu et al., 2007; Norden et al., 2009). These hypotheses can be tested by answering two questions. Which factors of phylogeny, seed traits, and ecology affect seed germination? Which factors are most significant in explaining seed germination percentage? Although there has been research on the relationship between seed traits and germinability, there is limited information on the proportion of germination variation among species that could be attributed to the species' phylogenetic background, seed traits, and ecological factors. The relative importance of phylogeny, seed traits such as mass, shape and color, and ecological factors such as ecotype, life form, adult longevity, dispersal type, and onset of flowering in determining seed germination percentage need to be determined for degraded sandy grassland ecosystems.

The Horqin Sandy Land located in the arid zone of Inner Mongolia, China, is a well-known grassland in northeastern China (Zhu and Chen, 1994; Wang et al., 2003). In the last few decades, it has been affected by desertification, likely the consequence of over-cultivation, intensive fire wood collection, and overgrazing. Mobile dunes, semi-stabilized dunes, and stabilized dunes are distributed over some areas of the steppe. In degraded sandy grasslands, seedlings are vulnerable to desiccation, long periods of drought, high temperatures, and overgrazing (Liu et al., 2007; Zhao et al., 2011). Seed germination strategies are, in theory, some of the most effective strategies for plants inhabiting arid environments, which are typically characterized by highly variable and unpredictable environment (Philippi, 1993). Plants inhabiting arid degraded sandy grasslands have developed different strategies for seed germination to respond to the novel environmental pressures via long-term natural selection (Li et al., 2006). However, in recent years, many studies have focused on the relationship between seed morphology and vegetation processes (Liu et al., 2007; Yan et al., 2009; Ma et al., 2010; Zhao et al., 2011), but the link between the germination and seed traits, ecological factors at community level in this highly disturbed and stressed environment has not been elucidated. Studies of the relationships between seed germination and phylogeny, seed traits, and ecological factors would provide insights into seed germination strategies in unpredictable environments, especially in degraded sandy grasslands.

In the present study, we assessed relationships between germination percentage and phylogeny, seed traits, and ecological factors for 109 species in a degraded sandy grassland at the community level. The following questions were addressed: (1) whether species with small seeds would have a higher germination percentage than species with larger seeds, (2) whether species with elongate and flat seeds would have a higher germination percentage than those with round seeds, and (3) what proportion of germination variation among species could be attributed to the species' phylogeny, seed traits (mass, shape, and color), and ecological factors (ecotype, life form, adult longevity, dispersal type, and onset of flowering).

GP in the degraded sandy grassland was affected by phylogeny, seed traits and ecological factors that interacted strongly in a factorial model. Figure 4 shows a model of the significant correlations found between mean GP and five independent variables analyzed in this study. Phylogenetic relatedness, dispersal type, and seed shape, in that order, were the three factors with the greatest contribution to the factorial model (14% of the variance in GP). Seed mass and onset of flowering combined explained only 4.6% of the variation in GP.

Several studies have advocated using phylogenetic correlates of plant species within a community to explain variance in ecological or other traits among taxa (Miles and Dunham, 1993; Figueroa and Armesto, 2001; Zhang et al., 2004; Norden et al., 2009; Xu et al., 2014). Our results clearly showed that phylogenetic groups significantly influenced GP for the species in the degraded sandy grassland. Previous research showed a phylogenetic pattern of seed germination in plant species of temperate rain forests (Figueroa, 2003), alpine meadows (Bu et al., 2007; Xu et al., 2014), and arid and semiarid zones (Wang et al., 2009). We think that variation in GP may be closely dependent on phylogeny, i.e., inherent traits of species may play a remarkable role in evolution. In Sheffield, England, germination rates of species in the same families were more similar than those of species in different families (Grime et al., 1981). Germination behavior might be more similar in closely related species than in distantly related species regardless of ecological factors. Our results suggest that phylogeny should be taken into account when investigating the role of natural selection in the regulation of germination or other life history attributes.

Our results indicated that GP was also correlated with seed traits such as shape and mass. The most significant and interesting finding of our study was the significantly negative relationship between GP and seed mass (Figure 1A, Table 1). This result was inconsistent with other studies that showed that seed mass was significantly positively correlated with GP (Vanmolken et al., 2005; Wu and Du, 2007; Galíndez et al., 2009). Traditionally, large and heavy seeds have better GP (Wu and Du, 2007), emergence, seedling survival, and growth than small seeds because of their larger reserves (Bonfil, 1998), which is consistent with the idea that heavy seed mass confers a competitive advantage, as assumed by the competition-colonization model (Rees et al., 2001). However, our results showed that small-seeded species have a higher GP than large-seeded ones in the degraded sandy grassland. There are several possible interpretations for the contradiction. Firstly, our study contained species with a range of growth forms from a wide range of families, and the species in Asteraceae, Poaceae, and Fabaceae were dominant in this study site. Asteraceae species had a lighter average seed mass (mean mass 1.59 mg; 73% were less than 1.00 mg in our study site) than the average of all species (mean mass 4.90 mg), and they had a higher GP (>80%). Legume seeds, which tend to have a hard seed coat, typically had a greater average seed mass (mean mass 11.30 mg) than species without a hard seed (4.30 mg). Hard seeds often require scarification to breach the test and do not immediately germinate (Smıkal et al., 2014). Secondly, the Horqin sandy land is characterized by Aeolian sandy soils (coarsely textured with a loose structure), and seeds seem to be more easily buried in sand ecosystems than in others. Small seeds may become more deeply buried in open, loose, sandy soil, reducing seed germination. Thirdly, certain small-seeded species in this study, such as Chloris virgata Sw., Euphorbia humifusa Willd., Datura stramonium L., and Solanum nigrum, germinate immediately after dispersal ((Liu et al., 2004a), b). Finally, it is well known that environmental factors are crucial from seed germination (Tielbörger and Petrü, 2010; Cendán et al., 2013; Li et al., 2015). Germination strategy is can be considered a coevolutionary strategy to the extreme environmental conditions in the degraded sandy grasslands. Plants in the degraded sandy grassland of Horqin Sandy Land are exposed to a highly disturbed and stressed environment due to overgrazing, wind, and drought. The small-seeded species should have greater seed density, higher germination percentage, and lower seedling survival at potential recruitment sites (Liu et al., 2004a, 2007). The large-seeded species have greater seedling establishment but a lower seed density and germination percentage (Liu et al., 2004a,b; Moles and Westoby, 2004; Li et al., 2006; Liu et al., 2007), even under favorable conditions. Thus, small-seeded species tend to produce large number of light seeds for long-distance dispersal to escape from unfavorable spatial conditions, and large-seeded species tend to be dormant to spread the risk of encountering unfavorable temporal conditions. Similarly, Rees (1996) found that seed germination was negatively related to seed size in a British grassland community with high overgrazing and stated that this was consistent with heavy seeds having better establishment success and suffering higher levels of unfavorable conditions (Rees, 1996).

We found that species with elongate and flat seeds had higher GPs than species with round seeds (Figure 1B). Our results indicated that the pattern between seed shape and GP found by Grime et al. (1981) applies to our study region. Interestingly, our study indicated that GP is more strongly correlated with seed shape than with seed mass in the Horqin sandy land (Figure 1, Table 1). We hypothesize that phylogeny may be responsible for this pattern because seeds of Poaceae and Asteraceae, which are often elongated or flattened, tend to germinate easily (Grime et al., 1981; Liu et al., 2003).

Ecological factors, such as seed dispersal type and onset of flowering, significantly influenced GP and could independently explain 5.9% and 2.4% of the total variance in GP in our study area, respectively (Tables 1, 3). Seeds of well-dispersed species germinated more frequently (Figure 2A), with anemochorous, zoochorous, and ombrohydrochorous species having higher GPs than either barochorous or autochorous species (Figure 2A). These results are consistent with those reported by Bu et al. (2007). One possible interpretation for this relationship between GP and dispersal type is that seed dispersal may be adapted to avoid or minimize sibling competition and place seeds at “safe sites,” where they can successfully germinate and establish (Gutterman, 1994; Lu et al., 2013; Bolmgern and Eriksson, 2015). In most germinated seeds, restricted (barochorous) dispersal can lead to competition among the siblings produced by a fecund maternal plant (Cheplick, 1996; Fragoso et al., 2003). The dispersal potential of seeds with different dispersal types varies greatly, with both wind- and vertebrate-dispersed seeds potentially having a greater dispersal distance from the parent than seeds of barochorous and autochorous species. In some vertebrate-dispersed seeds, germination may be promoted by passage through the animal's digestive system (Navarro et al., 2009). Thus, seeds with long dispersal distances (anemochorous, zoochorous, and ombrohydrochorous species) may germinate more frequently than seeds with short-range unassisted dispersal (which are more likely to experience sibling competition). Early onset of flowering may be correlated with early dispersal and germination, and thus a longer period of juvenile growth. A long juvenile period that includes the unfavorable season increases the risk of mortality before reproduction. These selection pressures involve tradeoffs that may hinder simple, direct effects of onset of flowering on germination. Our results revealed that most of the late-flowering species belong to Poaceae and Asteraceae. These species had a higher GP in the degraded sandy grassland community, which explains why onset of flowering species had a significant effect.

In this study, we found that ecological factors such as ecotype, life form, and adult longevity had insignificant effects on GP. In other words, the difference in GP between woody, graminoid, and forb species, and between annuals and perennials, was not significant. Previous studies have presented conflicting evidence of the relationships between life form, adult longevity, ecotype, and seed germination strategies. For example, Xu et al. (2014) indicated that there were no significant effects of life form (between annual and perennial) on GP in eastern Tibetan plateau species (Xu et al., 2014), while Rees suggested that perennials tended to have higher germination percentage than annuals based on a large grass data set (Rees, 1994). Bu et al. (2008) found life form (woody, graminoid, and forb) and adult longevity (annual and perennial) had a significant effect on seed germination time in eastern Qinghai-Tibet plateau species (Bu et al., 2008). The effects of life form and adult longevity on seed germination might vary by habitat or flora.

In this study, we provided insights into correlations between seed germination and phylogeny, seed traits, and ecological factors in a degraded sandy grassland of Horqin Sandy Land. Our most important finding is that seed germination strategies in this habitat were affected by the interactions of phylogeny, seed traits, and ecological factors. Phylogeny could explain the biggest proportion in variance of GP, which suggests that there are strong phylogenetic constraints on germination. Smaller seeds, elongate and flat seeds, seeds with long dispersal distances (anemochorous, zoochorous, and ombrohydrochorous species), and seeds from late-flowering species had higher GPs. Moreover, we found that phylogeny had a close association and interaction with seed size, shape, and dispersal strategy (Tables 1–3). Thus, we suggest that phylogeny may constrain seed germination, and because seed mass, shape, dispersal, and flowering time may be related to phylogeny, they also constrain germination. Therefore, phylogeny and life-history attributes could not be considered separately. Seed germination, like any other trait, is shaped by short-term selective pressures and long-term phylogenetic constraints.

In this study of the Horqin Sandy Land, a degraded sandy grassland in northeastern Inner Mongolia, China, we found that (1) GP was negatively and significantly correlated with seed mass, (2) species with elongate and flat seeds had higher GPs than species with round seeds, and (3) phylogenetic relatedness, dispersal type, and seed shape, in that order, were the three factors with the greatest contribution to the complete model (14% of the variance in GP). Furthermore, (1) GP, like any other trait, is shaped both by short-term selective pressures and long-term phylogenetic constraints, and (2) comprehensive studies at the community level that use multivariate approaches are needed to enhance our understanding of the evolutionary and ecological forces shaping germination strategies.

BL designed the research and collected and analyzed the data. ZW analyzed the data and drafted the manuscript. LW and ZL revised English language and contributed to paper writing. QL and YL assisted with data collection. All authors took part in writing the manuscript.

Funding was provided by the Key Project of Chinese National Programs for Fundamental Research and Development (No. 2013CB429903) and National Natural Science Foundation of China (No. 31570448) and the New Century Excellent Talents Program of Fujian Province (K8015053A).