Phylogenetic structure and formation mechanism of shrub communities in arid and semiarid areas of the Mongolian Plateau

Abstract The mechanisms of species coexistence within a community have always been the focus in ecological research. Community phylogenetic structure reflects the relationship of historical processes, regional environments, and interactions between species, and studying it is imperative to understand the formation and maintenance mechanisms of community composition and biodiversity. We studied the phylogenetic structure of the shrub communities in arid and semiarid areas of the Mongolian Plateau. First, the phylogenetic signals of four plant traits (height, canopy, leaf length, and leaf width) of shrubs and subshrubs were measured to determine the phylogenetic conservation of these traits. Then, the net relatedness index (NRI) of shrub communities was calculated to characterize their phylogenetic structure. Finally, the relationship between the NRI and current climate and paleoclimate (since the Last Glacial Maximum, LGM) factors was analyzed to understand the formation and maintenance mechanisms of these plant communities. We found that desert shrub communities showed a trend toward phylogenetic overdispersion; that is, limiting similarity was predominant in arid and semiarid areas of the Mongolian Plateau despite the phylogenetic structure and formation mechanisms differing across habitats. The typical desert and sandy shrub communities showed a significant phylogenetic overdispersion, while the steppified desert shrub communities showed a weak phylogenetic clustering. It was found that mean winter temperature (i.e., in the driest quarter) was the major factor limiting steppified desert shrub phylogeny distribution. Both cold and drought (despite having opposite consequences) differentiated the typical desert to steppified desert shrub communities. The increase in temperature since the LGM is conducive to the invasion of shrub plants into steppe grassland, and this process may be intensified by global warming.


| INTRODUC TI ON
The formation and maintenance mechanisms of community biodiversity have been an interesting theme in ecological research for several decades (Roughgarden, 1983;Tilman, 2004;Yang et al., 2011). With the rapid development of genetic information, gene sequencing, and computational power, evolutionary history combined with regional processes is used to understand the mechanism of species coexistence in biological communities (Cavender-Bares, Kozak, Fine, & Kembel, 2009).
At present, two major mechanisms are recognized for species coexistence in communities: a deterministic process based on niche theory, and a stochastic process based on neutral theory. Niche theory emphasizes the role of niche differentiation and habitat filtering (Webb, 2000) and generally holds that species, which are phylogenetically closer share similar niches (Burns & Strauss, 2011). By calculating the phylogenetic distance of species within a community and comparing with the regional species pool null model, the phylogenetic community structure is determined (Webb, 2000). If the phylogenetic relationship between species within a community is closer than that predicted by the null model (i.e., phylogenetic clustering), habitat filtering is more important, since it has served to cluster species with similar niches. On the contrary, if more species with distant phylogenetic relationships appear than expected (i.e., phylogenetic overdispersion), limiting similarity (such as competitive exclusion and negative density dependent) is thought to be predominant (Webb, Ackerly, Mcpeek, & Donoghue, 2002). Neutral theory suggests that equal niches exist among different species and species composition depends on the constraints of stochastic processes and spatial distances (Bell, 2000).
Since Webb (2000) first applied the phylogenetic community structure to explain community construction, this concept has been widely used in the study of tropical rainforest and temperate forests (e.g., Döbert, Webber, Sugau, Dickinson, & Didham, 2017;Kraft, Valencia, & Ackerly, 2008;Letcher, 2010). However, the phylogenetic community structure of arid and semiarid shrub communities, especially temperate desert communities, is poorly understood to date. Environmental aridity could either serve as an environmental filter to cluster closely related species, or aggravate niche differentiation in order to use different resources (Cavender-Bares et al., 2009). For example, Qian, Jin, and Ricklefs (2017) found that native communities in California were more phylogenetically clustered in dry areas, while Valiente-Banuet and Verdú (2007) found that significant phylogenetic overdispersion of the communities in semiarid regions of Mexico was chiefly due to facilitation between distant species. Soliveres, Torices, and Maestre (2012) investigated 11 Stipa grassland communities in semiarid regions of Spain and suggested that most were phylogenetically random under the combination of habitat filtering and competitive exclusion, but that interaction between species (i.e., competitive exclusion) had a greater impact on community construction. Nevertheless, either facilitation or competition may result in phylogenetic overdispersion (Cavender-Bares et al., 2009). Therefore, the role of habitat filtering in community construction (i.e., phylogenetic clustering) is not necessarily related to environmental stress. Under the premise of niche conservation, phylogenetic overdispersion could result from either competitive exclusion between close relatives or facilitation between distant species (Valiente-Banuet & Verdú, 2007;Webb et al., 2002). Other researches indicated that low temperature was a strong habitat filter, especially for wood species (Qian, 2018;Qian, Hao, & Zhang, 2014). Yan, Yang, and Tang (2013) demonstrated an overall phylogenetic clustering on the Qinghai-Tibetan Plateau and proved that both cold and dry climates were responsible for the formation of this lineage pattern. Thus, phylogenetic community structure in arid and semiarid regions may have different performance and formation mechanisms.
The western Mongolian Plateau is located in the interior of Eurasia beyond the reach of the warm and humid air currents of the Pacific and Indian oceans, and covered largely by sand and desert (Zhang, Hu, Zhuang, Qi, & Ma, 2009). With its ancient geological and floristic history, the region is rich in endemic and ancient remnant species, particularly in the Alashan-Ordos region, which is regarded as one of the eight biodiversity centers of China (Wang, 2005;Zhu, Ma, Liu, & Zhao, 1999). Given that the study of phylogenetic structure combines regional environment, interspecific relationships, and geological history processes (Fine & Kembel, 2011), the western Mongolian Plateau constitutes an ideal location for phylogenetic study because of its ancient flora and harsh modern environment.
In the present study, we took plant communities in the western Mongolian plateau as a model, for exploring species coexistence mechanisms and how phylogenetic structure responds to climate change in temperate arid and semiarid areas. We hypothesized that low temperature was more restrictive than drought for construction of shrub communities in temperate arid and semiarid areas, and that shrub communities would become phylogenetically more overdispersed under climate warming. This research will allow better understanding of the underlying mechanisms of community assembly and provide insight on biodiversity and ecosystem stability in arid and semiarid areas under global warming.

| Study area
The Mongolian Plateau, located in the hinterland of Asia, includes the whole territory of Mongolia, parts of southern Russia and most of China's Inner Mongolia Autonomous Region (Figure 1). The terrain is characterized by broad high plains and marginal mountains.
Most of the plateau is at an altitude of between 1,000 and 1,500 m above sea level (John et al., 2018). It has a typical temperate continental climate, with precipitation falling mainly in summer, and is cold and dry in winter (Fang, Bai, & Wu, 2015). The eastern part of the Mongolian Plateau is covered dominantly by temperate steppes, the central and western parts by vast deserts (Liu et al., 2013). We focused our study on xeromorphic shrub and sandy shrub communities in arid and semiarid regions of the Mongolian Plateau. The average annual temperature and average annual precipitation  in this region range from −0.8 to 10.3°C and 37 to 426 mm, respectively. The plant communities in the region include desert steppe, steppified desert, and typical desert, successively distributed along the climatic gradient of increasing aridity (Wu, Zhang, Li, & Liang, 2015 (Zhu et al., 1999).

| Field survey
This study centered on Erdos-Alashan and extended eastward to the desert steppe of the Wulanchabu Plateau and the western part of Xilin Gol in China's Inner Mongolia Autonomous Region, and East Gobi, South Gobi, and Bayankhongor in southern Mongolia.
We surveyed 116 vegetation plots in total (Figure 1). Typical plant communities of local habitats with little human interference were selected. At each plot, we recorded all shrub and subshrub species within a 10 m × 10 m square; the height and canopy (length and width) of each shrub and subshrub were also measured, and latitude and longitude were recorded by GPS.

| Determination and analysis of traits
Plant height, canopy area, leaf width, and leaf length were used to test trait conservation. The plant height and crown were measured with a tape measure. The canopy was approximated as an ellipse, and canopy area was thus calculated as canopy length × width × π/4. Height and canopy for each species in our study were taken as the average height and canopy area of the individuals measured (no fewer than 10 for most species). Leaf length and width data were derived from the median provided by the Flora Reipublicae Popularis Sinicae (FRPS, http://frps.iplant.cn/). For example, if the FRPS described the leaf width of N. sphaerocarpa as 2-4 mm, then the leaf width was set as 3 mm. Phylogenetic signals were measured using Pagel's λ model (Freckleton, Harvey, & Pagel, 2002;Pagel, 1999). The coefficient λ defines the weighting of phylogeny on traits. Under a completely Brownian motion model of evolution, phylogeny alone defines the expected covariance matrix of traits. However, as the F I G U R E 1 Distribution of surveyed communities, showing phylogenetically clustered (solid symbols) and overdispersed (outline symbols) shrub communities in arid and semiarid regions on the Mongolian Plateau (squares, triangles, and circles represent sand shrub, steppified shrub, and desert shrub, respectively) influence of factors unrelated to phylogenetic history increases, the weighting of phylogeny will be reduced. λ is generally between 0 and 1, with values closer to 1 indicating that a trait is conserved (Münkemüller et al., 2012).

| Phylogenetic analysis of plant communities
A list of all species recorded in our survey, and their general and family information based on the APG III (Angiosperm Phylogeny Group, 2009), was entered into the Phylomatic online plant database (https ://phylo diver sity.net/phylo matic/ , storedtree = "Zanne et al. (2014)" and clean = "true") to obtain a phylogenetic tree with differentiation time (Webb & Donoghue, 2005). It is worth noting that while gymnosperms such as Ephedra kaschgarica occupy an important position in some communities in the study area, the proportion of gymnosperms is quite small on a regional scale.
However, the presence of gymnosperms strongly biases results (Feng et al., 2015;Letcher, 2010), so gymnosperms were removed in our final calculation to avoid distortion of the results at the broader scale.
The net relatedness index (NRI) was used to define the phylogenetic community structure (Webb et al., 2002). This index assumes that all species in the study area constitute a local species pool. First, the mean phylogenetic distance (MPD observed ) of all species pairs in the sample is calculated. Then, without changing the number of species and the number of individuals in each species, the species in the sample are repeated randomly from the species pool 999 times to obtain the distribution of MPD of the random null model (i.e., MPD rand ). Finally, the MPD rand is normalized by stand diversion (SD).
The NRI is calculated as follows: In general, if NRI > 0, it means that the average phylogenetic distance between species in the sample is closer than the phylogenetic distance in the random null model; that is, the community is phylogenetically clustered. By contrast, if NRI < 0, the average phylogenetic distance between species in the sample is greater than in the random null model, and the community is phylogenetically overdispersed. If NRI ≈ 0, it indicates that the community is phylogenetically random (Swenson, Enquist, Jill, & Zimmerman, 2007).

| Climate data
Climate data were taken from CHELSA (Climatologies at high resolution for the earth's land surface areas, http://chelsa-clima te.org/), a global climate data set based on the time period 1979-2013. This contains 19 core bioclimatic data sets including annual mean temperature (AMT), mean annual precipitation (MAP), seasonal range of temperature variation, and precipitation in the driest quarter (defined as three consecutive months with least precipitation, generally between November and the following March; Table 1). More details and calculation methods of 19 climate factors were described by Karger et al. (2017). Potential evapotranspiration (PET) and aridity index (AI = MAP/PET) were downloaded from CGIAR-CSI (http://www.cgiar-csi.org). PET and AI are averages calculated from meteorological data from 1970 to 2000 (Zomer, Trabucco, Bossio, & Verchot, 2008). A low AI value indicates a drier climate, with an AI of 0.3 usually defined as the boundary between arid and semiarid regions (Cardy, 1997

| Data analysis
All data analysis was performed in R version 3.5.1. The package "picante" was used to calculate abundance weighted (expressed as important value) NRI and Faith's phylogenetic diversity (Kembel et al., 2010).
Principal component analysis (PCA) was performed on 21 climate data variables using the "princomp" function in R. The principal components with explanatory variables above 85% were selected, and the climatic distance between the samples was represented by the Euclidean distance on the selected principal component axis; the unweighted pair group method using arithmetic average (UPGMA) clustering (using the "hclust" function in R and set "method" = "average") was then performed on the climate distance to divide climate habitats into groups.
Spearman correlation coefficients between 21 climate factors and the NRI of each climate habitat group were calculated in order to identify the climatic factors with the greatest impact on each habitat group.
The Wilcoxon signed-rank test was used to test whether the NRIs were significantly different from zero. The Wilcoxon rank-sum test was used to assess differences in NRI, PD, and species richness (SR) between habitat groups. Partial least squares regression (PLS) was used to analyze the power of climate factors in explaining NRI.

| Phylogenetic community structure
A total of 47 species of angiospermous shrubs and semishrubs were recorded in 116 samples in this study, belonging to 14 families and 29 genera according to the APG III system. The three families of Fabaceae, Amaranthaceae, and Asteraceae comprised 57.14% of the total species recorded (Table 3 and Table S1).
The 21 climatic factors were decomposed by principal components, with the first three components representing 88.15% of the variation in all variables. The surveyed plots were clustered into three groups using UPGMA based on environmental distance: shrub communities in relatively warm and humid sandy habitats ("sandy shrub," 12 plots), shrub communities in relatively dry and cold habitats in vegetation transition zones between desert and steppe, that is, desert steppe or steppified desert communities ("steppified shrub," 17 plots), and typical desert shrub communities in extremely arid habitats ("desert shrub," 87 plots; Figures 1,   2a, and Table S2).
The NRI of the studied shrub communities was significantly less than zero (p < .05), with phylogenetically clustered and overdispersed communities accounting for 29.66% and 70.34%, respectively.
Phylogenetic diversity (PD) and species richness (SR) in desert shrub communities were significantly greater than in sandy shrub communities (p < .05), and somewhat higher than in steppified shrub communities (p = .557; Figure 3).

| Relationship between current climate factors and phylogenetic community structure
The NRI of desert shrub and steppified shrub communities increased with increasing AI (p = .01 and 0.14 when analyzed, respectively  was positively correlated with AMT anomaly from the LGM and that of steppified shrub communities was negatively related to MTD anomaly and positively related to AP anomaly since the LGM.

| Significant phylogenetic conservation of shrub leaf traits
Niche conservation is a necessary assumption for applying phylogenetic analysis to community assembly (

| Phylogenetic structure and formation mechanisms of shrub communities
Our results show that the phylogenetic structures of shrub communities in the western Mongolian Plateau demonstrate a general trend toward phylogenetic overdispersion, though differences exist among the three groups corresponding to sandy, steppified desert, and typical desert habitats.
Desert shrub communities account for more than 3/4 of the sample communities in this study. The desert habitat is harsh with an extremely dry climate (annual precipitation generally < 200 mm); however, desert shrub communities show a significant phylogenetic overdispersion trend (Figure 3a), indicating the coexistence of distantly related species in these communities. Species competitive exclusion, facilitation between distantly related species, and other density-dependent species interactions may contribute to the coexistence of distantly related species (Webb, Gilbert, & Donoghue, 2006), although it is difficult to isolate which process is responsible (Cavender- Bares et al., 2009). Competition for below-ground resources (groundwater especially) is key for plant survival when resources are limited (Wilson, 1988). Therefore, fierce competition for limited resources will inevitably make it difficult for species with similar ecological niches to coexist, and only species with differentiated niches can be retained in the community (Webb et al., 2002).
As a result, communities were expected to be phylogenetically that compared to trees, shrubs and herbs showed less sensitivity to climate, and attributed this to their smaller size allowing them to take advantage of microhabitat protection.
As an example, Reaumuria songarica + N. sphaerocarpa is a widely distributed typical desert community. Reaumuria songarica is a shallow-rooted plant with more lateral roots and most roots distributed in the range of 20-60 cm underground (Wu, Zhou, Zheng, Li, & Tang, 2014). In contrast, N. sphaerocarpa has deeper root distribution, enabling it to use groundwater below 80cm underground (Cui, Ma, Sun, Sun, & Duan, 2015). Coexisting species with differential water niches are expected, under the premise of a conservative niche, to be phylogenetically overdispersed (Webb et al., 2002). Our results were in contrast to those of Yan et al. (2013) on the Qinghai-Tibetan Plateau, which showed an overall phylogenetic clustering. The possible reason for the discrepancy in the phylogenetic structures of shrub communities between these two plateaus may be difference in temperature. The elevation is much higher and the mean average temperature lower on the Qinghai-Tibetan Plateau (over 4,000 m and AMT < 0°C and above 4,000 m) than on the Mongolian Plateau (below 1,740 m and MAT 8.7°C; Zheng, 1996), and low temperature rather than dryness is a strong filter for survival of woody plants (Qian et al., 2014). We will discuss this further below.
Steppified shrub communities show a weak phylogenetic clustering (p = .08), indicating an effect of habitat filtering on community construction. The weak (nonsignificant) phylogenetic clustering is F I G U R E 4 Changes in net relatedness index (NRI) of shrub communities in the three habitats along climate aridity (AI) gradient (*p < .01).
(a) Linear regressions of sandy, steppified, and desert to AI; (b) smooth fitting of steppified-desert and sandy to AI F I G U R E 5 Spearman correlations between the NRI of shrub communities in different habitats and current climate as well as paleoclimate anomaly (AMT, Annual mean temperature; AP, Annual precipitation; MTD: Mean temperature of driest quarter). Significant values are in bold most likely related to the fact that climate-based habitat clustering cannot fully reflect shrub community structure. For example, some communities in this steppified shrub community group (phylogenetically overdispersed) show obvious desert shrub characteristics.

Partial least squares (PLS) regression of NRI against climatic factors
shows that community lineage in the semiarid area is strongly influenced by the mean temperature of the driest quarter (MTD, i.e., temperature in winter). That is, under the dual stress of low temperature (mainly) and drought, phylogenetic community structure shifts from overdispersion to clustering. The average annual temperature and the lowest winter temperature in the region where steppified shrub communities are predominate are the lowest among the three habitats: The temperature in winter often reaches −20°C, which is a challenge for shrub overwintering. Qian et al. (2014) showed that the woody phylogenetic community structure in a mountain ecosystem in China was limited by low temperature, tending toward phylogenetic clustering as temperature decreased. Qian (2018) also found that temperature (especially cold temperature) was more strongly correlated with angiosperm woody species in tropical South America.
In addition, herbaceous plants play an important role in steppified communities, so that competitive pressure from these (biological filtration) may be the reason for community phylogenetic clustering.
In recent years, shrub encroachment into grassland has attracted the attention of researchers (e.g., Chen et al., 2015;Eldridge et al., 2011). Our study highlights the effects of low temperature on the phylogenetic structure of steppified shrub communities and indicates that, as global temperature rises, that is, as low temperature stress reduces, the process of regional shrub encroachment may be aggravated (Chen et al., 2015;Moleele, Ringrose, Matheson, & Vanderpost, 2002).
Sandy shrub communities are distributed in markedly different habitats from desert or steppified shrub communities, in regions with higher precipitation (Figure 2a) and sandy substrates conducive to preserving precipitation and groundwater. The significant phylogenetic overdispersion of this group of shrub communities indicates that limiting similarity is the main driver of community construction (Kraft, Cornwell, Webb, & Ackerly, 2007). Species composition in sandy shrub communities is relatively simple, consisting of A. ordosica + leguminous shrub (e.g., Caragana sp. or Corethrodendron lignosum), with A. ordosica, the predominant species (Zhang, 1994).
Phylogenetic overdispersion of plant communities may be related to negative density dependence, which limits distribution of the dominant species (Webb et al., 2006).
We combined desert and steppified shrub community types to examine the changes in shrub lineage along the aridity index (AI) gradient, and found a significant positive correlation between NRI and AI (Figure 4b). This result indicates that communities with phylogenetic clustering are more common in relatively wetter habitats, a phenomenon clearly shown in vegetation transition zones with a climate AI around 0.3 (see Figure 1). While only shrub plants were studied, herbaceous plants (synusia) in the steppified desert communities are generally developed, even dominant in some areas (Eldridge et al., 2011;Liu et al., 2015). The phylogenetic clustering of shrub communities may result from strong competition with herbaceous plants in this transition zone (Clary, Save, Biel, & De Herralde, 2004). A study in shrub-encroached grasslands in Inner Mongolia showed that shrub patch density is negatively associated with precipitation and temperature and native vegetation (especially tall native herbs) is more likely to resist shrub encroachment (Chen et al., 2015). While herbs can occupy a large number of suitable niches, alternative niches for shrubs are limited and only a few species can survive in steppified shrub communities (Fowler, 1986). This is also consistent with shrub species richness and phylogenetic diversity being higher in desert shrub communities than in steppified shrub communities (Figure 3b,c). That is, biotic filtering plays an important role in forming shrub communities (Goberna, García, & Verdú, 2015).
As climate becomes relatively more humid (i.e., AI continues to rise), the desert plant community is replaced by steppe community, and the dominant role of shrubs is also shared with herbs, except in sandy conditions .

| Effect of paleoclimate and geological history on phylogenetic structure
Regional historical processes strongly affect species assembly and community phylogeny (Cavender-Bares et al., 2009). That is, current community phylogeny is affected not only by the modern environment, but also by paleoclimate and geological process (Kissling et al., 2012;Valiente-Banuet, Rumebe, Verdú, & Callaway, 2006). Sandel et al. (2011) showed that high climatechange velocity since the Last Glacial Maximum is responsible for the global absence of endemism (e.g., amphibians, mammals, and birds). Variation between the Last Glacial Maximum (LGM) and the current climate has also had a high impact on the phylogenetic diversity structure of Chinese woody community lineage (Feng et al., 2014). Kissling et al. (2012)  The typical desert area in this study belongs to the ancient Tethys region (Zhu et al., 1999) and has experienced a relatively stable arid climate since the late Tertiary (Liu, 1995). A large number of ancient Tethys flora and endemic Tertiary plants have been preserved in the area (Zhu et al., 1999). The long and stable evolutionary history and the existence of large numbers of endemic species provide abundant alternative regional species pools, rich in phylogeny with different niches (Anacker & Harrison, 2012) for community assembly (Cavender-Bares et al., 2009;Pärtel, 2002).
Therefore, this region has high pedigree and species diversity ( Figure 3b,c), which may also be a cause of phylogenetic overdispersion of the desert shrub community (Gerhold, Pärtel, Liira, Zobel, & Prinzing, 2008).

| CON CLUS IONS
The

CO N FLI C T O F I NTE R E S T
None declared.

DATA AVA I L A B I L I T Y S TAT E M E N T
Data are in the Dryad Digital Repository https ://doi.org/10.5061/ dryad.6m905 qfvv.