Community Assembly as the Basis for Species Selection for Tropical Forest Restoration in Global Change Scenario

Background Tropical forests hold high biodiversity, which challenges the selection of species for forest restoration. For a site-specic restoration is required the understanding of the main inuences ruling the assemblages. We aimed to answer three questions. 1) how do environmental variables inuence taxonomic, phylogenetic diversities, and the phylogenetic structure in the of Rio Doce Basin (TFRD)? 2) How do environmental variables, phylogenetic structure and the main types of seed dispersal relate to each other? 3) Which information of the TFRD assemblages can be used for ecological restoration and conservation in global change scenario? We used 78 sites with their checklists to calculate taxonomic, and phylogenetic diversities, phylogenetic structures, and dispersal proportions. Then, we related the diversities of the sites to their bioclimatic variables and built GLM models. Results Species richness was inuenced negatively by water excess duration, by water decit duration, and positively by maximum temperature, and temperature seasonality. Water regime drives diversity and phylogenetic community structure in the TFRD more than other variables. Annual precipitation and maximum temperature presented the clearer inuences on diversity and phylogenetic structure. Zoochory was positively, and anemochory, autochory were negatively related to sesMPD. Conclusions By choosing the lineages with high tness for each site, the functioning and the stability of ecosystems should increase. The addition of species with anemochory and autochory functional and phylogenetic diversity in areas extreme water excess or water decit, important in a global change scenario.


Background
Environmental degradation is one of the largest threats that are being looked at in the current global change scenario. Large-scale land degradation by agriculture, urbanization and mining are increasing in the tropics ( For deeper understanding of which factors play in the assemblage of those Tropical Forests of the Rio Doce Basin (TFRD), rules of community assembly can be assessed by means of tree species composition of forest remnants responding to environmental variables, in uencing the taxonomic diversity as well as phylogenetic diversity and structure. This is because (Niño et al. 2014) ecological niches are major drivers of tropical communities largely determined by environmental variables that limit in different ways the species distributions ( (Fig 1). The TFRD are essentially secondary forests with only few of them classi ed as old growth forests. The TFRD area may be in uenced by different oras (i.e., Cerrado and Campos Rupestres) on western watersheds, and mountain tops, but Atlantic Tropical Forest ora de nes the basin within the domain of Atlantic Forest. We analyzed woody species from subtypes of Atlantic Tropical Forests: seasonal tropical forest and tropical rainforest.
The seasonal tropical forest is de ned by a seasonality characterized by a rainy season characteristic of tropical forests followed by a dry season during the winter. As a response to that seasonality, around 20-50% of the trees

Phylogenetic community structure
We produced a phylogenetic tree with all woody angiosperms recorded in the basin using Phylomatic function in

Phylogenetic signal of dispersal types
Phylogenetic signal of dispersal types means the trend of species with zoochory, anemochory, and autochory to be more related than by chance or to be less related than by chance. If the signal shows the pool of species with a dispersal type more related between each other than null models, the phylogenetic signal is clustered and means that the dispersal type has been predominantly conserved within the phylogenetic lineages under niche conservatism. The opposite, when a signal shows the species less related than calculated by null models, the phylogenetic signal is overdispersed and means that the dispersal type is predominantly convergent (homoplasy).
In order to test if dispersal types presented phylogenetic signal, the net relatedness index -NRI among all species of the same dispersal type (DispersalNRI) was calculated. The same was calculated for nearest taxon distance -NTI (DispersalNTI). This calculation was done reordering species labels 999 times through phylogeny.
DispersalNRI values signi cantly greater than zero (more than 1.96 sd, and more than -1.96 sd) indicate phylogenetic signal. Phylogenetic signal of dispersal type was considered signi cant when the observed MPD or MNTD occurred in the lowest 2.5% of the null models (signi cance of 0.05) (Gastauer et al. 2017).

Modelling
To outline the environmental predictor variables that best describe species richness and phylogenetic structure of sampled tree communities, we build for each response variable a global general linearized model containing all 15 predictor variables.
We scaled the explaining variables. Then, we used the dredge function of the MuMIn package (Bartón 2018) to model-select all possible combinations of ve or less uncorrelated predictive variables (r <0.6, https://github.com/rojaff/dredge_mc) that most parsimoniously explained our data. We considered models with ve or less predictive variables to avoid over tting. We used the Akaike Information Criterion (AIC) for selection of the best models (Symonds and Moussalli 2011). All models with ΔAIC less than 2 were considered equally parsimonious. When more than one model was selected, we calculated model-averaged parameters and unconditioned standard errors weighted by the likelihood ratio using the 'model.avg' function of the MuMIn package. The function performs signi cance tests for each predictor estimated in this average model. The resulting global models were con rmed by plotting residuals versus adjusted values as well as residuals versus predicts.
For modelling percentages of zoochorous, anemochorous and autochorous species within surveys, we build global models containing the same 15 environmental predictor variables plus species richness, sesMPD, sesMNTD and sesPD. For model selection, we proceed as described above.

Results
Modelling species richness, only one global model containing ve environmental variables was selected, with four signi cant trends (Table S1), indicating that richness is in uenced negatively by water excess duration and by water de cit duration meanwhile is positively in uenced by the Maximum temperature of the warmest month as well as is positively in uenced by seasonal temperature (Figure 1).
For further response variables, more than one model was selected (Table S1). 14 models containing the all variables except Precipitation during Wet Period, and Minimum Temperature explain sesPD. However, only two trends were signi cant: annual precipitation reduces sesPD, and water excess duration increases sesPD (Table S2, Figure 2).
For sesMPD, six models were selected containing 10 out of the 15 predictor variables (Tables S1, S2). In the averaged model, only two trends were signi cant: annual precipitation reduces and water excess duration increases sesMPD (Figure 2).
For sesMNTD, the two signi cant trends were: increasing maximum temperature of the warmest month, and increasing isothermality increase sesMNTD (Table S2, Figure 2).
The percentage of zoochory species increases with increasing sesMPD, with water excess duration, with water de cit severity, and decreases with water excess severity ( Figure 3).
The percentage of species with anemochory decreases with increasing sesMPD and increases with increasing water excess severity (Table S3, Figure 4).
The percentage of autochory increases as sesMPD decreases, as water de cit severity decreases, and is positively related to the water excess severity (Table S3, Figure 4).
Anemochory, and autochory presented DispersalNRI clustered phylogenetically showing that these dispersal types are conserved within phylogenetic lineages throughout the phylogenetic tree (Table 1). Zoochory, and anemochory preented clustered DispersalNTI, showing that hese dispersal types are conserved within phylogenetic lineages towards the tips of the phylogenetic branches (Table 1) Water regime drives species richness and phylogenetic community structure in the TFRD more than other types of environmental variables. At least one among annual precipitation, water excess duration, water excess severity, and water de cit severity were signi cant in all averaged models, except for the model for sesMNTD. Species richness, sesPD, sesMPD and percentage of zoochory were affected by water excess duration. All dispersal types were explained partially by water excess severity. Temperature regime was the second mst important important regime. Maximum temperature of the warmest month and isothermality explained sesMNTD. Maximum temperature of the warmest month, and temperature seasonality explained partially species richness. Among the variables, water excess duration, annual precipitation, maximum temperature and isothermality in uenced the phylogenetic diversity and phylogenetic structure. The effects of temperature on richness, and on sesMNTD suggest environmental ltering effects causing phylogenetic clustering driven by decreasing maximum temperature of the warmest month and decreasing isothermality (i.e., increasing temperature variation). Therefore, the lower temperature and the less temperature variation in different seasons, the more environmental ltering.
The percentage of species with zoochory, anemochory or autochory were related to phylogenetic structures (i.e., sesMPD) in different ways. Increasing zoochory increases phylogenetic distances between species (i.e., increasing sesMPD) meanwhile increasing anemochory, and autochory decrease phylogenetic distances between species (i.e., decreasing sesMPD). The opposite relation of zoochory with sesMPD was not because zoochory is convergent, but because the less zoochory, the more abiotic dispersion (i.e., anemochory + autochory) that is clustered for MPD (i.e., DispersalNRI more related than by chance), suggesting anemochory, and autochory as homologies in the TFRD metacommunity. Only zoochory decreased with increasing water excess severity, and appears to be the only dispersal type bene ted with increasing water surplus. Water excess severity was positively related to anemochory, and autochory, and might be interpreted as stresses in uencing positively these dispersal types. Water de cit severity in uenced negatively autochory.
The answer of the rst question shows how environmental variables in uence taxonomic, and phylogenetic diversities as well as phylogenetic structure in the tropical forests of that large river basin (TFRD). Seven environmental variables were found to in uence the taxonomic diversity, phylogenetic diversity and phylogenetic structure in averaged models: water excess duration, water excess severity, water de cit duration, water excess severity, maximum temperature, temperature seasonality, and isothermality. As water excesses and water de cits increase in number of days, the species richness decreases. As the maximum temperature and the temperature seasonality decrease, the species richness decrease. Thus, more days of water excesses or more days of water de cits in TFRD are associated with loss of species richness as well as lower maximum temperatures and lower temperature variation between seasons. Therefore, the found global model shows that increasing the number of days with excess or de cit of water, decreasing temperatures, and decreasing temperature variation, decrease species richness.
Still answering the rst question, as annual precipitation decreases and there are more days of water excess, sesPD and sesMPD increase; as maximum temperature and isothermality increase, sesMNTD increases. Thus, the lower annual precipitation, the more water excess days, the higher temperature variation, and the higher maximum temperatures in TFRD, the greater the phylogenetic diversity and phylogenetic distances between species. However, as the more days with water excess (i.e., days with higher precipitation than evapotranspiration) cause higher sesPD and sesMPD, it is possibly because bene ts the zoochory that is positively-related to mean phylogenetic distances (see zoochory discussion below). Therefore, higher precipitation promotes decreasing sesPDs and sesMPDs (i.e., phylogenetic clustering) in the TFRD similarly to effects of environmental lters found The percentage of species with zoochory, anemochory and autochory was signi cantly related to sesMPD and not to sesMNTD or sesPD, which means that the main association of dispersal types with the communities' phylogenies is throughout the entire phylogenetic tree (i.e., sesMPD). As the DispersalNRI is clustered for abiotic dispersal types, the clustered sesMPD reinforce that anemochory and autochory are conserved in phylogenetic branches originated in old nodes congruently with the decreasing sesMPD as anemochory and autochory. As a consequence, the increasing proportion of zoochory is not caused by its preeminence as homoplasy in the TFRD metacommunity, but because abiotic dispersal decreases. Thus, increasing anemochory and autochory reduce the sesMPD distances between species congruently with their clustered DispersalNRI in the TFRD metacommunity.
Zoochory is phylogenetic clustered towards the tip of phylogenetic tree according to its DispersalMNTD. That means the species with zoochory are predominantly clustered in lineages with conserved zoochory diversi ed from recent nodes. Despite zoochory is clustered towards the tip of phylogenetic tree, zoochory is not clustered throughout the entire phylogenetic tree. Anemochory is also clustered clustered according to its DispersalMNTD.
Autochory is not clustered towards the tip of phylogenetic tree in TFRD, suggesting lack of recent diversi cation of lineages with conserved autochory.
The dispersal types responded to water regime. The dominant dispersal type in TFRD is zoochory that varies from 50% up to more than 95% of the tree species in evergreen and seasonal Atlantic Tropical Forests (Tabarelli and Peres 2002). In the TFRD, the minimum zoochory percentage found in a forest was 53% and the maximum was 81% (data not shown). As dominant dispersal type, zoochory responded positively to water excess duration, and water de cit severity meanwhile anemochory and autochory were not in uenced by those explaining variables.
However, zoochory responded in an opposite way from the two abiotic dispersal types to water excess severity. Zoochory also responded in an opposite way from autochory to water de cit severity. Thus, zoochory seems to be impaired meanwhile anemochory and autochory seem to be bene tted in sites with extreme severity of water excess or water de cit.
As anemochory and autochory are clustered in some phylogenetic branches in the TFRD metacommunity, the presence of these dispersal types in restored TFRD will allow those species to improve the functional and phylogenetic diversity of restored forests by representing entire phylogenetic branches, especially in those TFRD with extreme water regimes (i.e., high severity of water excess, and high severity of water de cit). Therefore, the relation between severity of water excess, and severity of water de cit with the two abiotic dispersal types deserves attention since extreme water regimes are increasing according to forecasted climate scenarios in the most of the Atlantic Forest distribution (IPCC 2013). As a consequence, the proportion of anemochory and autochory may increase in the TFRD during the next decades in a global change scenario.

What of the answers to the previous questions can be used in the ecological restoration of TFRD in a global
change scenario, as well as in the conservation of biodiversity and ecosystem services?
Species selection is not the restoration project itself. Most selected species may not be able to thrive under the harsh initial conditions of the restoration sites. This means that the selected species must be used in TFRD restorations to add taxonomic and phylogenetic diversity in order to improve the suitability and long-term stability of the restored ecosystems. For that, our databank delivered as supplementary material will offer the basis for the species selection. The results sequence shows how the TFRD community assembly rules are central to the species selection for the planned large-scale restoration in the basin, which includes mitigating the Mariana disaster that compromised the service of water puri cation in the basin. 1) The environmental ltering promoted by low temperatures (i.e., low maximum temperatures) drops the taxonomic diversity shortening the phylogenetic distance between species, predominantly ltering in species of some recently diversi ed lineages. Temperatures in TFRD are also related with altitude (data not shown). Thus, the use of some lineages better tted for TFRD restoration in sites with low maximum temperatures (i.e., high altitudes inside the basin) should enhance the performance of restored forests. 2) Possibly the water puri cation is the most affected ecosystem service by the main players of global change, such as forests degradation, agriculture (Cabel et al. 1982 All sampling and analysis were carried out in accordance with Brazilian law. The data were shared between the authors.

Consent for publication
All authors agree with the publication of this manuscript.
Availability of data and material Data will be shared as demanded.

Competing interest
None.
Authors contributions JAAMN conceived and designed the analyses, wrote and edited the paper, performed the analyses; NDOJ collected data and performed analyses, wrote the paper; NS collected data and performed analyses; ATOF collected the data and developed part of the analyses tools; MLB performed the analyses, wrote the paper; VP performed the analyses and wrote the paper; MG performed the analyses, wrote the paper, developed part of the analyses tools.