Shift of seed mass and fruit type spectra along longitudinal gradient: high water availability and growth allometry

Abstract. Propagule traits vary among biomes along geographical gradients such as longitude, but the mechanisms that underlie these variations remain unclear. This study aims to explore seed mass variation patterns of different biome types along a longitudinal gradient and their underlying variation mechanisms by involving an in-depth analysis on the variation of seed mass, fruit type spectra, growth forms and dispersal mode spectra in Inner Mongolia and northeastern China. Plant community characterization and seed collection were conducted in 26 sites spreading over five vegetation types and covering 622 species belonging to 66 families and 298 genera. We found there are significantly declining trends for mean seed mass, vertebrate-dispersed species richness and fleshy-fruited species richness along a longitudinal gradient from forests to desert grasslands. However, we also found the lowest average seed mass and the smallest proportion of species dispersed by vertebrates occurring at typical grasslands in the five biomes. The variations of average seed mass display high congruence with transition of growth form spectra. The selection for these propagule attributes is driven mainly by climatic factors such as precipitation, temperature, soil moisture and evaporation, as well as by internal biotic factors such as growth forms, canopy coverage and leaf area (Ackerly and Donoghue, 1998). A hypothesis was provided that environmental factors and botanical traits that favor greater water availability lead to emergence (or speciation) of species with large seeds or fleshy fruits with high water content. Due to greater water availability and increasing leaf area, much more photosynthate (photosynthesis production) and allometric growth then ultimately increase the biome average seed mass from west to east. Phylogenetic signal or diversity are not found to be significantly involved in the effect on the patterns. A novel mechanistic framework and mathematical model are provided to expound seed variation among species or biomes.

environmental factors and botanical traits that favor greater water availability lead to emergence (or speciation) of 23 species with large seeds or fleshy fruits with high water content. Due to greater water availability and increasing leaf 24 area, much more photosynthate (photosynthesis production) and allometric growth then ultimately increase the biome 25 average seed mass from west to east. Phylogenetic signal or diversity are not found to be significantly involved into the 26 effect on the patterns. A novel mechanistic framework and mathematical model are provided to expound seed variation 27 among species or biomes. retaining a certain degree of plasticity being affected by the surrounding environment (Baker, 1972). Therefore, an 35 individual plant's seed size is a combined result of its taxonomic group's evolutionary history and immediate selective 36 pressures of the local environment (Westoby et al., 2002;Moles et al., 2005a). Furthermore, as an important aspect in 37 the reproductive biology of plants, seed mass is evolutionarily associated with and corresponds to other plant traits, 38 relating to growth forms (for instances, trees, shrubs and herbs), life history (for example, annual plants or perennial 39 plants) (Moles et al., 2005a), stature and canopy sizes (Venable, 1992 Torstesson, 1991), as well as to anatomical traits of flowers and fruits (Primack, 1987). 42 Numerous works show that seed mass varies along environmental gradients such as latitude, elevation and 43 longitude owing to environmental variations in temperature and precipitation both among and within communities 44 (Moles et al., 2007) and several ecological factors are proposed to explain such seed mass variation gradients or 45 patterns, for example, temperature (Moles et al., 2014), light (or solar radiation) (Murray et al., 2004;Demalach and 46 Kadmon, 2018), soil resource availability (Demalach and Kadmon, 2018), growth forms (Moles et al., 2005b), 47 dispersal modes (Moles and Westoby, 2003), soil pH (Tautenhahn et al., 2008) etc. However, a deep understanding of 48 the factors that underlie these major biogeographical variations is missing (Demalach and Kadmon, 2018), especially 49 at a continental scale along longitude. Previous work suggested that community-level average seed mass tends to 50 decrease towards higher latitudes and elevations (Moles et al., 2007). These trends can be explained by shifts in habitat 51 type, plant growth form spectra, seed disperser assemblage (Moles and Westoby, 2003), solar radiation and metabolic 52 expenditure (Murray et al., 2003; and NPP (Bu et al., 2007;Guo et al., 2010) along latitudinal and elevational 53 gradients. Additionally, species that prefer shaded habitats and late successional stages generally tend to have larger 54 seeds than those in open arid habitats or earlier successional stages (Baker, 1972;Salisbury, 1974;Foster and Janson, 55 1985;Hallett et al., 2011;Moles and Westoby, 2006), indicating a strong effect of high water availibility on seed mass 56 owing to low evaporation under close canopy coverage. Longitudinal variations of seed mass has been discussed 57 among species with a single genus (Murray et al., 2003;; however, there are few studies that focus on how 58 community-level variations of seed mass (especially across species) correspond with other plant traits along 59 longitudinal gradients, because of the difficulty to predict variations of comprehensive environmental factors arising 60 from complex topography. In this region average seed mass is expected to decrease with declining longitude due to 61 gradually less rainfall from forests to desert ecosystems (Murray et al., 2003;. Here we present a study of 62 community-level variations in seed mass in correspondence to position in the continent (relative to the sea) across 63 Inner Mongolia and northeastern China, to identify the longitudinal pattern and discuss the mechanisms that may 64 underlie them. 65 Previous works emphasize the role of high light acquisition and allometric growth (a growth pattern in which 66 different parts of an organism grow at defined rates) in shaping seed mass variation through model prediction and 67 experiment testing (Demalach and Kadmon, 2018;Demalach et al., 2019), and in this article we emphasize the 68 importance of high water availability and allometric growth for speciation and colonization of species with large seeds 69 due to environmental factors and biological traits. Allometry of biomass growth and size-asymmetry of light 70 competition became the drivers of seed mass variation owing to soil resource availability and ultimate productivity 71 heterogeneity along soil resource gradient (surely including water gradient). As we know, primary production of 72 communities increases across an increasing water gradient (Bai et al., 2008). This article presents a novel mechanistic 73 framework that integrates previous theory and hypotheses (related to climate, phylogeny, water conduction systems 74 Seed mass, longitude and precipitation were log-transformed before analysis to meet the normality and 152 homoscedasticity assumptions of linear regression models. In order to ensure that any observed seed mass variation 153 along the longitudinal gradient is independent of latitude and elevation, general linear models (GLM) were employed. 154 Seed mass and other plant traits were treated as the dependent variable in all analyses with latitude, longitude and 155 climatic variables entered into models as independent variables. 156 The proportions and species richness of plants with various seed mass and fruit types in different communities 157 were compared using analysis of variance (ANOVA). ANOVA was also used to compare average seed mass between 158 different growth forms, different community types, different fruit types and dispersal types. The GLM procedure was 159 used to examine the explanatory power of community types, dispersal types, longitude, precipitation and temperature 160 on seed mass. All analyses were performed with R-3.3.3 (R Core Team, 2018). By use of the function commonality in 161 the R package yhat (Nimon et al., 2013), we take the log-transformed seed size as dependent variable, life forms, 162 vegetation types, dispersal modes and each with latitude as independent variables, exploring predictive power of each 163 variable, respectively. 164 For the 620 species (two gymnosperm species were excluded owing to their low relatedness with most of 165 angiosperm species), a supertree was constructed using the software Phylomatic (Webb et al., 2008). The phylogenetic 166 backbone was based on the APG III tree (R20120829, http://phylodiversity.net/phylomatic/). We quantified the 167 strength of phylogenetic conservatism and tested the phylogenetic signal in seed mass using Pagel's λ (Pagel, 1999)  . Regression analyses were conducted between phylogenetic signal and longitude across the sites in five 171 community types. Using the phylogenetic tree with branch length, we calculated the phylogenetic diversity using the 172 measure PD, which was defined as the minimum total length of all the phylogenetic branches required to span a given 173 set of taxa on the phylogenetic tree (Faith, 1992). Taking mean seed mass as dependent variable and longitude 174 We considered the relations between the number of species with fleshy fruits and longitude, the number of 177 families, number of genera, and the phylogenetic diversity PD. Since there are strong correlations between the latter 178 four variables (r > 0.67, p < 0.001), they cannot be used in the same model. Therefore, we built four models. Each took 179 one of the four variables as the independent variable and the number of species with fleshy fruits as dependent variable. 180 A generalized linear regression model with Poisson family was fitted using R package stats (R Core Team, 2017). In 181 the model, we also included log (number of species) as offset.

Seed mass variations along the longitudinal gradient 185
Although the majority of species had medium-sized seeds (Figure 1), variations among all species were great. There 186 were considerable differences in average seed mass and seed spectra among the five biome types (Figure 1). Forests 187 have the largest average seed mass (23.45 ± 18.34 mg) and both typical grasslands (4.75 ± 3.93 mg) and sparse forests 188 There are declining trends for herbaceous species richness and canopy coverages from forests to desert along 197 decreasing longitudinal gradient in this region (Table 2). Typical steppe was found to have the lowest woody species 198 richness and highest herbaceous abundance in five community types (Table 2). 199

Seed mass relations to environmental variables 216
Average seed mass was minimum at approximately 114 degrees longitude where typical grasslands occur ( Figure 3). 217 However, phylogenetic diversity (PD) was not a significantly explanatory variable (p > 0.8) (Figure 3). Linear 218 regression model shows that there is no significant decreasing trend from forests to deserts along declining longitude 219 (F = 2.289, p = 0.143) in this region. If the westernmost sample site (Ejinaqi) is excluded, seed mass significantly 220 decrease inland (R 2 = 0.2434, F = 7.398, p = 0.012). 221 Significant negative relationships were found between seed mass and MAT (R 2 = 0.1752, p = 0.01915) and 222 elevation (R 2 = 0.1221, p = 0.0449) across all sample sites, but no significant relationships were found between seed 223 mass and latitude (R 2 = -0.028, p = 0.576) and MAP (R 2 = -0.008, p = 0.380). Across 23 sample sites from desert 224 through desert grassland to typical grassland, average seed mass had significantly negative relationship with longitude 225 (R 2 = 0.232, p = 0.012) and MAP (R 2 = 0.48, p = 0.00015), while across 20 sample sites from typical steppe to forests 226 average seed mass had significantly positive relationship with longitude (R 2 = 0.232, p = 0.012) and MAP (R 2 = 0.48, p 227 = 0.00015). Average seed mass was found to just be weakly positive relationship with MAT both from desert through 228 desert grassland to typical grassland and from typical grassland to the forests (R 2 = 0.09207, p = 0.08665). According 229 to above analysis, MAP should be crucial environmental drive factor for seed mass variation. 230 In addition, average seed mass is significantly related with soil moisture (R 2 = 0.8259, p = 0.0017) and soil 231 moisture significantly decrease with declining longitude from typical to desert grasslands (R 2 = 0.6019, p = 0.0018). 232 233

Species richness and proportion of fleshy fruited species 234
Among the five community types, forests have the highest number (7.44 ± 1.26) and proportion (28.05 ± 6.16) of

Variation of seed mass spectra and environmental factors 269
There is strong and consistent effect of community type (along a longitudinal gradient) on seed mass (Figure 1, Figure  270 3). The average seed mass displays a significantly declining trend along decreasing longitude from forests to typical 271 grasslands and then to some sites in desert grasslands in this region (Figure 3). In these sites, average seed mass was It is possible that larger seeds are more common in drought-prone habitats most likely because they allow 293 seedlings to establish large root systems early, with a better chance of surviving drought (Baker, 1972;Salisbury, 1974). 294 In this study, desert grassland and desert ecosystems are found to be dominated by shrubs that often possess larger 295 seeds (Figure 2). In Inner Mongolian Plateau these species are seldom exposed to strong interspecific competition or 296 shading that make the plants invest more in propagules than in vegetative apparatus for competitive strength (Bai et al., 297 2008). In addition, relatively high species richness and the highest number of species occurred in this typical steppe 298 grassland (Table 2), and in contrast, desert steppe had very low species richness and number of individuals (abundance) 299 dispersed species in typical grasslands was the lowest in comparison to other communities (S4). The patterns of seed 302 dispersal syndromes observed in this study are congruent with previous findings in Australia's subtropics (Butler et al., 303 2007). Biotic dispersal agents exert a strong selective pressure on angiosperm species with various seed size in Inner 304 Mongolian plateau, as evidenced by the evolution of a wide range of adaptations for animals (such as ants, birds, 305 squirrels ) dispersal. 306 307

Variation of fruit type spectrum and associations of seed mass with fruit types 308
Fleshy fruited species richness significantly corresponded to gradual changes of climate, especially for MAP (Table 1). 309 The smallest proportion of fleshy fruited species occurred in typical grasslands and desert grasslands (Figure 3 closed vegetation (Mazer, 1989;Hammond and Brown, 1995) and with late successional stages (Hammond and Brown, 330 1995). All those phenomena indicate that seed mass may be related to low evaporation and high water availability in 331 plants ( Figure 6). We suggest that, as an ecological strategy, the derivation and evolution of species with large seeds 332 may be due to improved water accommodation in plants by strong resource acquisition ability (such as having strong 333 water absorbing root system and advanced water conductive ability) or water retention ability (such as habituating 334 shaded environment or developing small, thick leaves and hair or waxiness on leaf to prevent water loss) (Baker, 1972;335 Fonseca et al., 2000). Plant species have evolved various ecological strategies to match their environments (Laughlin, 336 2014). These strategies are manifested in many plant organs and traits. In the present study, seed mass is strongly 337 connected with other biological characteristics such as plant dispersal ability (SP 4, Table 3 Anatomical structures of lots of species indicated that the species with large seeds or fleshy fruits often have wide and 342 long vessel elements that can provide much more water (Carlquist, 1975;Zimmermann, 1983). As suggested before, 343 seed mass also is likely to be a result of co-evolution among various organs that determine plant responses to changing 344 abiotic factors (Dí az and Cabido, 1997; Sandel et al., 2010). 345 In light of growth allometry theory, average seeds mass variation should converge with community total biomass 346  (Table 1). 354 In previous studies, soil moisture was found to not correlate with the relative abundance of fleshy-fruited species 355 due to low temperature on water availability constraints (Yu et al., 2017). As we know, seed plants employ two main 356 strategies to increase water use efficiency: one is to take up more water through strong root systems and the other is 357 reducing water loss through low evapotranspiration. In our study, canopy coverage decreases from forests to sparse 358 forests and then to grasslands and desert grasslands (Table 2), leads to gradual reduction in fleshy-fruited or 359 large-seeded species richness ( Figure 5, Table 2). However, since fleshy fruits have high water content and thus inquire 360 higher plant internal water content (Yu et al., 2017), we suggest the correspondence of seed size and fruit water content 361 imply that some species evolved to contain more water or photosynthate in multiple body parts. Furthermore, CO2 362 concentration is generally the same everywhere although there is some small variation during growth seasons (Wang et 363 al., 2002), its impact on seed mass variation patterns should be expelled. Solar radiation variation is not very large 364 along longitude (see site description) especially among typical grasslands, desert grasslands and deserts with similar 365 elevation, therefore, its effect on seed mass variation is very small, moreover, since light is not a limited factor for 366 growth in northern China according to our observation. Variation trend of sunshine hours or light intensity are contrary 367 to that of rainfall amount along longitude. Only when water remain sufficient, strong light may favor plant growth and 368 increase seed mass. For example, combination of much more belowground water with more sunshine hours or higher 369 light intensity in Erjina may increase its average seed mass, and this may be responsible for larger seed mass in desert 370 than in some sites of desert grasslands. Therefore, combined with previous results of other studies, we deduce that 371 drivers of seed mass spatial distribution patterns include temperature, rainfall, solar radiation, soil moisture and 372 nutrients, leaf area, canopy coverage and their interactions, however, high water availability in plant body may be the 373 most vital driving factor in shaping seed mass spatial distribution patterns. According to growth allometry, a fraction of 374 coverage, is considered to be allocated to seeds. In addition, biological structures (such as fair or waxiness on leaf or 376 fruits to prevent water loss), that favor water retention in plant body would also be useful in increasing seed mass or 377 fruit water content. 378 In order to understand variation mechanism of seed mass better, a simple mechanistic model is provided to trying 379 explain quantitatively average (or total) seed mass variation between communities for one species as following: 380 St is the total seed mass of all species in a community, Sa is the average seed mass per species taken from the total 383 community (St/n), n is number of species in a community, Ci1 is the allometric growth coefficient (or allocation portion 384 to seeds) that differ among species. Bt is total biomass from photosynthate per species. Bid value is the biomass of 385 photosynthate related to water from conducting issues for one species, Bi0 is the biomass of photosynthate related to 386 water from other approach (for instances, lessening evaporation), Bl is the biomass of photosynthate related to leaf area 387 ( Figure 6). As we know, ecological factors affecting St are numerous. St will be developed according to other sufficient 388 data basis. For instances, seed developing time, sunshine duration and intensity and belowground water may affect Bt, 389 however, how affect and what extent will be done further in the future to improve and perfect Bt,. 390 Generally, seed mass is quite phylogenetically conservative (Lord et al., 1995). However, in this study, 391 phylogenetic signal is weak across the 26 sites (Table 1) and the phylogenetic signal are found to be little involved in 392 the relationships between seed mass and longitude, MAP and MAT in the five biomes. This proves that the 393 environmental factors affect seed mass variation in the community context and phylogenetic constraints are not 394 significant (Figure 3, 4). The five communities are in middle or late successional stages in which the main construction 395 process is environmental filtering (effect) rather than competitive exclusion (Norden et al., 2012). 396 In additions, in this study we just measure the soil moisture of top 10cm which mainly influence growth of herbs, 397 but for the growth of shrubs and trees, rich soil water below the depth of 10 cm in some area of Ejinaqi also is useful. 398 intensity in Erjina may increase seed mass and shape the present seed mass variation patterns in this region. Moreover, 400 ecological scale and micro-environmental heterogeneity often affects results of functional traits along biogeographical 401 gradients, so further study may be necessary in larger scale (or large investigation area) to identify the results of this 402 article. 403

Conclusions 404
Mean seed mass, seed dispersal spectra, fruit type spectra and plant growth form spectra of five biome types vary 405 significantly along a longitudinal gradient, with the lowest average seed mass and the smallest proportion of species 406 dispersed by vertebrates occurring at the middle longitude (typical grasslands). The selection for these propagule 407 attributes is most likely to be driven by external and internal drivers ( Figure 6), however, water availability potentials 408 and growth-allometry may be key drivers of seed-mass variation along climatic gradients or resource gradients. Larger 409 seeded species or species with fleshy fruits may have evolved due to much photosynthate or high water availability in 410 plants. Our findings can provide help in understanding origin and evolution of species with large seeds or fleshy fruits.    Dear Anja Rammig, Thank you very much for careful works! According to your opinions, we had revised those mistakes as following.

Shunli Yu
----1) Table 1: Information about geographic…. of the 26 sampling sites… K-value: phylogenetic signal values with smaller values indicating a weaker signal. In the caption of Table 1, you do not need to give a definition for evaporation. Please remove the sentence: Evaporation: the change process of evaporating from a liquid to a vapor.
We had removed the sentence: Evaporation: the change process of evaporating from a liquid to a vapor. Table 3 last column, please revise and change to "Proportional occurrence", is that correct?

2)
We had changed "Proportion in the whole" into "Occurrence proportion" in Table 3 last column.
3) The text for the caption of Figure 1 is not clear, please revise.
We had revised the caption of Figure 1 and it had become "Seed mass spectra vary among five community types in Inner Mongolia (A) and proportions of larger seeds (A) and average seed mass (B) decline from forests to desert grasslands but increase in deserts (Average seed mass bearing the same letter are insignificantly different at p > 0.05)". Figure 1A, y-axis label, please change to "Proportion (%)".
5) The resolution of Fig. 1A and B is not very good, please improve.
We had improved resolution of Fig. 1A and B.
6) Caption of Figure 3, please revise: "… Average seed mass has a U-shape and declines with increasing longitude with lowest level at around 114 degrees. Average seed mass do not have significant relationships with phylogenetic diversity (p>0.05)….". Please also correct x-axis labels in Fig. 3D, longitude values are different from A and B.
We had changed the caption into "Relationships between average seed mass of communities and longitude (A, B) and phylogenetic diversity (C, D). Average seed mass declines as longitude rises and it reaches its bottom at around 114 degrees, and after that it increases. But average seed mass do not have significant relationships with phylogenetic diversity (p>0.05)".
We also corrected x-axis labels in Fig. 3D as "Phylogenetic diversity".

7)
The text for the caption of Figure 6 is not clear, please revise.
We had revised the caption of Figure 6 and turned it into "Mechanistic frameworks of large seeded species formation and corresponding increment process of community average seed mass".

8)
In addition, we had revised Fig.2A (for instances, scal to scale in y-axis, seed masses to seed mass) and Fig.2B (for instances, fleshy fruits, f to F, dry fruits, d to D).