Local-scale environmental filtering shape plant taxonomic and phylogenetic diversity in an isolated Amazonian tepui (Tepequém table mountain)

Understanding how environmental drivers induce changes in plant composition and diversity across evolutionary time can provide important insights into community assembly mechanisms. We evaluated how taxonomic and phylogenetic diversity and structure of plant communities change along a local-scale edaphic and elevational gradient in the Tepequém table mountain, Brazilian Amazon. We selected three phytophysiognomies along the altitudinal gradient: Open Rupestrian Grassland, Shrubby Rupestrian Grassland, and Forest. We compared community composition and taxonomic and phylogenetic diversity between phytophysiognomies, and analysed effects of altitude and soil properties on species richness and phylogenetic metrics using linear mixed-effects models. The highest species richness and phylogenetic diversity were found at a lower elevation for Forest. All standardised phylogenetic metrics were significantly lower in Shrubby Rupestrian Grassland. This phytophysiognomy showed phylogenetic clustering. Forest showed a cluster pattern when only terminal nodes are considered and random dispersion to deep phylogenetic nodes. Open Rupestrian Grassland also showed a random phylogenetic structure. The linear mixed-effects models showed that species richness and different phylogenetic structure metrics were explained by altitude and soil properties. Our study revealed that key plant diversity dimensions (i.e. taxonomic and phylogenetic) are shaped by a local-scale edaphic and elevational gradient on an isolated tepui of Brazilian Amazonian.


Introduction
Mountain ecosystems have been recognised as a remarkable ecological scenario to evaluate the effect of environmental drivers on plant community diversity and structure (Qian et al. 2014;Liu et al. 2019a). These ecosystems are Earth's most biodiverse region (Körner et al. 2017), due to the higher variability of abiotic-related factors (e.g. climate and soil properties) along local environmental gradients, such as elevation (Körner et al. 2017;Pashirzad et al. 2018). Thus, elevation can induce changes in temperature, precipitation, and edaphic conditions (Benites et al. 2003;Körner et al. 2017), which influence the taxonomic diversity, composition, structure (Mota et al. 2016;Campos et al. 2018;Cordeiro and Neri 2019), and evolutionary history of plant communities (Chu and Lee, 2018;Pontara et al. 2018;Liu et al. 2019a;Rezende et al. 2019;Campos et al. 2021). However, most studies on altitudinal gradients in tropical mountains have focused on the effects of abiotic variables on floristic composition and species richness (e.g. Neri et al. 2017;Campos et al. 2018;Mota et al. 2018). Thus, knowledge of their effects on communities' evolutionary history is still limited, mainly in ancient climate buffered and nutrient-poor Neotropical mountains.
The phylogenetic analysis is used to infer the mechanisms of community assembly, which is associated with stochastic processes, such as dispersal limitation and ecological drift (Hubbell 2001;Vellend 2010) and niche-related deterministic processes, such as environmental filtering and limiting similarity (Webb et al. 2002;Cadotte et al. 2011). In this context, an approach based on phylogenetic diversity of plant communities can provide insights to explain ecological processes that shape the assemblages along environmental gradients at local and regional scales (Webb 2000;Liu et al. 2019a, b). Thus, the nicherelated deterministic processes postulate that plant species interactions with their environment are mediated by different functional traits, and more related species have more similar functional traits and niche resources explored (Webb 2000;Liu et al. 2019a). Thus, if functional traits are conserved within evolutionary lineages, communities shaped by environmental filters show a more clustered phylogenetic structure (Wiens and Graham 2005;Cavender-Bares et al. 2009). Conversely, overdispersed patterns suggest that density-dependent processes (e.g. competitive exclusion) are dominant and determine the phylogenetic structure of communities (Webb et al. 2002;Cavender-Bares et al. 2009). However, biotic interactions (e.g. competition) can also lead to clustered phylogenetic structures when exclusion occurs among distantly related organisms (Mayfield and Levine 2010). Furthermore, the local ecological community can show a random phylogenetic pattern when various nichebased processes operating simultaneously in species selection or neutral factors are more important (Webb 2000;Sobral and Cianciaruso 2012).
Previous studies in the mountains documented the environmental filtering effect by harsh climatic and edaphic conditions at higher elevations and competitive exclusion at lower elevations driving the increased and decreased phylogenetic relatedness, respectively (Li et al. 2014;Qian et al. 2014;Manish and Pandit 2018). Environmental filters select species from the overall species pool via environmental constraints (Gastauer and Meira-Neto 2014;Kraft et al. 2015;Ramos et al. 2015), whereby only those species with certain adaptive traits are fit to survive and reproduce in local communities (Webb et al. 2002;Cavender-Bares et al. 2009). Another niche-related process that often influences community assembly at small spatial scales is competitive exclusion (de Oliveira et al. 2014). This hypothesis assumes that the co-occurrence of species is possible only if they have distinct traits (i.e. low niche overlap; MacArthur and Levins 1967;Cavender-Bares et al. 2009).
Among Neotropical mountain ecosystems, the isolated table mountains (called tepuis), Guayana Shield region in the northern South America (Huber 1987;Safont et al. 2014), are considered a natural laboratory to evaluate key ecological and evolutionary processes of the biota worldwide . The tepuis are modelled on Precambrian quartzites/sandstones (Huber 1987;Rull 2007;Safont et al. 2014), with elevation ranging from approximately 1000 to 3000 m (Huber 1995). They are considered virtually pristine environments, with high host levels of biodiversity and endemism (Rull 2007;Safont et al. 2016). Similarly, to the rupestrian grassland complex, in eastern Brazilian mountains , tepuis are characterised by distinct physiognomies, including forests, shrublands, high-mountain meadows, grasslands, and pioneer vegetation (Huber 1995). In addition, both rupestrian ecosystems share similar environmental conditions, such as high mountain climate, extensive areas of rocky outcrops, and soil that is generally poor, acidic, shallow, and water-deficient .
In South American highland rupestrian ecosystems, small differences in altimetric and soil conditions are important drivers of plant community assembly, influencing the taxonomic diversity patterns (Neri et al. 2017;Campos et al. 2018;Mota et al. 2018). Studies have revealed that different abiotic factors (e.g. climate and soil) at fine-scale altitudinal gradients play significant roles in plant phylogenetic diversity (Li et al. 2014;Qian et al. 2014;Chun and Lee 2018). However, no detailed study has evaluated these patterns along the altitudinal gradient in tepuis. Attempts to address this knowledge gap could be important for understanding the mechanisms and processes that shape tepuis plant communities and establishing strategies and actions for biodiversity conservation.
In this context, we aimed to evaluate how taxonomic and phylogenetic patterns of plant communities change along a local-scale edaphic and elevational gradientin the Tepequém table mountain, Brazilian Amazon. Thus, we established the following research questions: (i) How do plant community composition, structure, and diversity (taxonomic and phylogenetic) change along a local-scale edaphic and elevational gradient? (ii) What are the effect of different environmental predictors (altitude and edaphic-related factors) on taxonomic and phylogenetic community structure and diversity? We hypothesis that the localscale environmental filtering (altitude and edaphic-related factors) is the main process that shapes plant community assemblages in Tepequém table mountain. Thus, we predict that harsh edaphic conditions and altitude impose a strong environmental filter to species richness, composition, and phylogenetic diversity, but lead a similar pattern of community phylogenetic structure along phytophysiognomies. Therefore, we expected that these communities in edaphic conditions would show more phylogenetic clustering because of the strong effect of environmental filtering.

Study area
The study was conducted on Tepequém table mountain in Roraima state, northwestern Brazil (03°45′54,6″ N and 61°41′17,5″ W; Fig. 1). This tepui reaches approximately 1100 m of altitude (Nóbrega et al. 2016) and comprises 70 km 2 in total area (Reis and Carvalho 1996). According to Köppen's classification, the region has a humid tropical climate (Am), with a mean annual temperature and precipitation are 28 °C and 1600 mm, respectively (Barbosa and Miranda 2004).

3
The regional geomorphology is characterised mainly by erosive scarps, steep slopes, valleys and residual mountains (Nascimento et al. 2012). Geologically this tepui represents residuals of ancient erosional landsurfaces, supported by thick tabular sandstones of the Proterozoic Roraima Supergroup (Almeida-Filho and Shimabukuro 2002). The top stratigraphic unit corresponds to the Tepequém Formation composed mainly of sandstone and conglomerates, lying over acid to intermediate Early Proterozoic volcanites of the Surumu Group (Almeida-Filho and Shimabukuro 2002).
The Tepequém table mountain vegetation is characterised predominantly by Ombrophilous Dense Forest extending through the slopes and scarps associated with Haplic Cambisol and Yellow Red Latosol (Nascimento et al. 2012). Moreover, this table mountain has a typical Rupestrian Grassland Complex distributed mainly between 650 and 1100 m a.s.l., influenced by savannic formation, with sclerophyllous vegetation ranging from grassy to the shrubby formation, usually on Quartzarenic and Litholic Neosols (Silva 1997).

Data collection
The sampling was performed at three main Rupestrian Grassland Complex phytophysiognomies along the altitudinal gradient: (i) Open Rupestrian Grassland (ORG) in plots ranging from 923 to 1100 m altitude; (ii) Shrubby Rupestrian Grassland (SRG), from 690 to 855 m; and (iii) Forest (FOR), from 450 to 600 m. In each of the three different phytophysiognomies adequate plot sizes were used for each predominant life form. In the herbaceous vegetation from ORG, we randomly distributed 80 plots (1 × 1 m). Plant community structures were evaluated by the cover-abundance scale proposed by Braun-Blanquet (1979). Thus, 60 plots (20 × 20 m) were randomly established to woody vegetation, i.e. 30 plots along the SRG and 30 plots along the FOR. Each plot in the same phytophysiognomies was at least 40 m apart from each other. In the SRG, all individuals presenting circumference at soil height (CSH) ≥ 3 cm were sampled, while in the FOR sampling was performed of all individuals presenting circumference at 1.30 m from the soil (CBH) ≥ 15 cm (Moro and Martins 2011). The number of individuals for each species was registered in each plot. Finally, the botanical specimen was deposited in the Herbarium VIC at the Universidade Federal de Viçosa (Minas Gerais, Brazil). Identifications were made by consulting specialists and the literature. Taxonomic classification followed APG IV (Angiosperm Phylogeny Group 2016).
The physicochemical properties of the soil were sampled as abiotic variables along an altitudinal gradient of Tepequém table mountain. A composite sample of three topsoil subsamples (0-10 cm depth) was collected for each plot. Samples were air-dried and sifted through a 2 mm mesh sieve. Analyses were conducted at the Laboratory of Soil Analysis, Universidade Federal de Viçosa, following international standards (Embrapa 2017). The soil pH was determined in water. Acidic components (H + + Al 3+ ) were extracted with Ca(OAc) 2 0.5 mol L −1 buffered to pH 7.0 and quantified via titration with NaOH 0.0606 mol L −1 . Exchangeable cátions were extracted in KCl 1 mol.L −1 , and determined via atomic absorption spectroscopy (Ca 2+ and Mg 2+ ) and titration with NaOH (Al 3+ ).The available phosphorus (P),Na + , K + , Fe, Cu, Mn and Zn were extracted with Mehlich-1, and quantified using inductively coupled plasma optical emission spectrometry (ICP-OES). The remaining P (P-rem) was obtained using a fine air-dried soil sample containing 60 mg L −1 of P (KH 2 PO 4 ) and determined by photo colorimetry. Effective cation exchange capacity (ECEC) was calculated by determining the sum of cations (Ca 2+ , Mg 2+ , Na + , K + , and Al 3+ ). In contrast, the total cation exchange capacity (CEC) was estimated using the bases of sum (BS) and potential acidity (H + + Al 3+ ). We determined the bases saturation index (V) and Al saturation index (m). Organic C was determined by the Walkley-Black method without heating. The organic matter (OM) content was estimated by multiplying organic C by 1.724 (OM = Walkley-Black C × 1.724). Sodium Saturation Index (ISNa) indicates the proportion of soluble sodium concerning total cation exchange capacity. Granulometric analysis (clay, silt, coarse and fine sand contents) was performed using the pipette method.

Taxonomic and phylogenetic composition
We performed a non-metric multidimensional scaling (NMDS) multivariate analysis based on Bray-Curtis dissimilarity (based in abundance data) to compare species composition between phytophysiognomies (Bray and Curtis 1957). After, a permutational multivariate analysis of variance (PERMANOVA, 9999 permutations) was used to determine differences in species composition between phytophysiognomies using the adonis routine available within the "vegan" package (Oksanen et al. 2018).
We generated a phylogenetic tree including pooling of all species using a mega-tree approach in the V. PhyloMaker package (Jin and Qian 2019). This mega-tree contains 74,533 vascular plant species based on the APG IV phylogenetic system of angiosperm classification and includes all plant families of the world (Jin and Qian 2019). We used the phylo.maker function makes phylogenetic hypotheses under scenario 3 which consist in the tips of a new genus or species not included in the mega-tree were bound to the half-point of the family or genus branch, representing the branch between the family and genus root node and the basal node (Jin and Qian 2019).
We performed an evolutionary principal component analysis based on Hellinger distance (evoPCAHellinger) to analyze differences in phylogenetic composition across phytophysiognomies (Pavoine 2018). This approach balances the influence of deep and shallow nodes on the ordination analysis represents one of the more powerful methods to study phylogenetic patterns across underlying environmental variation (Pavoine 2016). We implemented this approach using the evoPCAHellinger function in the "adiv" package (Pavoine 2018).

Taxonomic and phylogenetic diversity and structure
We determined the species richness as the sum of all species found in all plots in each phytophysiognomy. We used individual-based rarefaction and extrapolation curves with the first Hill number (species richness, q = 0) to analyse differences in species richness among phytophysiognomies (Colwell et al. 2012;Chao et al. 2014). Individual-based rarefaction/ extrapolations were computed using the 'iNEXT' package (Hsieh et al. 2016; Supplementary Material, Fig. S1).
We calculated metrics that evaluate the evolutionary history for each phytophysiognomies; such as phylogenetic diversity (PD, in millions of years, myrs), correspond the sum of the branch lengths of a phylogenetic tree connecting all species in a community (Faith 1992); to remove the influence of species richness, we calculated their equivalents standardized based on a null model (ses.PD, ses.MPD, ses.MNTD) (Swenson 2014;Xu et al. 2016). Negative ses.MPD and ses.MNTD values indicate phylogenetic clustering (species are distributed within clades with relatively recent common ancestors, or are more closely related than expected by chance). In contrast, positive values indicate phylogenetic overdispersion (species more distantly related to each other than expected by chance) (Webb et al. 2002). Our tree was compared with 10,000 null model randomizations using the algorithm 'phylogeny pool' for the standardised effect size calculations. The phylogenetic analyses were based on plant abundance. We calculated these metrics using "picante" package (Kembel et al. 2015).

Statistical analyses
The Shapiro-Wilk test for normality and Q-Q graphs, were used to verified data distribution (Crawley 2013). Differences in the species richness and phylogenetic metrics across the phytophysiognomies were analyzed by using one-way analysis of variance (ANOVA; for normally distributed data) followed by a post hoc Tukey's test (p < 0.05). Post-hoc Tukey HSD tests were applied to outline significance levels of phylogenetic structure differences among single phytophysiognomy.
We used a Kruskal-Wallis test to compare soil attributes (non-normally distributed data) (Supplementary Material, Table A1) followed by a posteriori Dunn´s test (Dinno 2017). All analyses were calculated using the "stats" and "dunn.test" packages (Dinno 2017). Soil variables were summarized using principal component analysis (PCA), proceeded by standardisation by logarithmic transformation, to equalise their contributions on the axis (Supplementary Material, Fig. S2). Soil pH (pH_H 2 O) values were not transformed because they are already expressed on a logarithmic scale. We also calculated Pearson correlations among soil properties (Supplementary Material, Fig.  S3). The PCA was performed using the "FactoMineR" package (Husson et al. 2017).
To reduce any strong correlations among soil attributes the first and second axis was considered to be a proxy for soil fertility (PCA1f) and variability of physical properties related to soil texture (PCA1t) across analyses Schmitz et al. 2020). Thus, we defined the first PCA axis for soil fertility and texture variables across all the tested models.
We tested how altitude, variability of chemical properties related to soil fertility (PCA1f) and variability of physical properties related to soil texture (PCA1t) on richness and phylogenetic metrics (ses.PD, ses.MPD and ses.MNTD) using linear mixedeffects models (LMMs, with random and fixed effects). The phytophysiognomies and plots were used as a random effect (1│plots: phytophysiognomies) in all models tested. We tested the distributions of residuals to select the most suitable distribution and link function, i.e. Gaussian error distribution by the Q-Q graph and Shapiro-Wilk test (Crawley 2013). We selected predictor variables using Pearson correlation to avoid collinearity (r ≥ 0.7), then we included them in univariate models. Finally, we applied a multi-model inference approach with the Dredge function of the 'MuMIn' package (Barton 2017), to evaluate the best models (LMMs) tested using the information theoretical approach based on the Akaike Information Criterion (AIC), considering all models with AIC < 2.0 as equally plausible (Burnham et al. 2011;Barton 2017). We also used the estimates of the predictors' coefficients to interpret parameter estimates on a comparable scale. All models were calculated using the package 'lme4' (Bates et al. 2019). The graphics were performed with the 'ggplot2' package (Hadley 2015). All analyses were carried out using R software version 3.6.2 (R Core Team 2020).

Taxonomic and phylogenetic composition
Overall, 7325 individuals were sampled, among 108 angiosperm species, 88 genera, and 37 families ( Fig. 2; Table A2). The NMDS showed that species composition varied considerably among phytophysiognomies along the altitudinal gradient, mainly among FOR with SRG and ORG (Fig. 3a). In the evolutionary composition ordination, a contrasting phylogenetic composition pattern was observed between phytophysiognomies (Fig. 3b). Three groups were clearly separated according to the first two evoPCA axes, which together explained 48.6% of the total data variation. The first axis (PC 1 , 27%) separated plots from FOR with positive values contrasting with ORG and SRG with negative values. This axis is strongly and negatively correlated with lineages related to the Poales clade, especially the families Poaceae, Cyperaceae, and Xyridaceae, with the highest number of species in ORG (Fig. 3c). The second axis (PC 2 , 21.6%) separated most of the plots from FOR and ORG, with negative values, of the SRG plots with positive values. The second axis is strongly and negatively correlated with the Pandanales (family Velloziaceae), Asparagales (family Orchidaceae) and Poales clades, with a high number of species in ORG; and strongly and positively correlated with Nitrogen fixers to COM, mainly Fabales (Fabaceae) and Malpighiales clades, with a higher number of species in SRG and FOR (Fig. 3c). The second axis also correlated with Magnoliales (family Annonaceae) and Laurales (Lauraceae), both with a higher number of species in FOR.

Taxonomic and phylogenetic diversity and structure
Species richness showed marked differences between FOR (55 species) compared to SRG (31) and ORG (29) Table A3). Phylogenetic diversity (PD, in myrs) showed similar patterns as we found for species richness, once that metrics both are highly correlated (Supplementary Material, Fig. S4). The highest value was found for FOR (1716.465 Ma), while the lowest value was found for ORG (366.047 Ma) (Fig. 4a). The SRG showed intermediate values (848.602 Ma). Standard-effect transformations of phylogenetic diversity (ses.PD) and mean pairwise distance (ses.MPD) values were highest and close to zero in the ORG and FOR. Only, ORG showed the highest value for mean nearest neighbour distance (ses.MNTD; Fig. 4d). For ses.MNTD, SRG was not significantly different from FOR. Both phytophysiognomies showed clustering pattern. SRG showed strong negative values for all standardized metrics; ses.MPD and ses.MNTD indicated phylogenetic clustering for deep and shallow nodes, respectively.

Discussion
This study revealed significant differences in plant community composition, structure, and diversity (taxonomic and phylogenetic) along a local-scale edaphic and altitudinal gradient. Our results showed that significant effects of altitude and edaphic-related factors on species richness and phylogenetic structure of plant communities highlight the importance of fine-scale environmental heterogeneity to mantain high biodiversity levels in tepuis (Vegas-Vilarrúbia et al. 2012;Safont et al. 2014Safont et al. , 2016. We suggest that harsh edaphic conditions and altitude would impose a strong environmental filter on vegetation, resulting in phylogenetic clustering. However, we highlight that other environmental filters, stochastic factors, and biotic interactions may also be responsible for Tepequém table mountain community assembly, a tepui with high ecological importance for the Amazon region. The community composition changes considerably among phytophysiognomies along the altitudinal gradient. The physical environment influences the species distributions and community composition (Nunes et al. 2015;. ORG was strongly associated with members of the following families: Velloziaceae (Pandales); Orchidaceae (Asparagales); and Poaceae, Cyperaceae, and Xyridaceae (Poales). In open physiognomies of rupestrian ecosystems, these families are the most important by dominating the herbaceous monocot stratum (Porembski 2007;Le Stradic et al. 2015;Silveira et al. 2016;Silva et al. 2019). Its high representativeness is due to the ability of the species to colonise harsh environments under severe stress conditions, with a broad array of adaptations (Porembski 2007). Similarly, Fabaceae, Annonaceae, and Lauraceae, with high representativeness of lineages in SRG and FOR, are essential in the woody strata structuration of rupestrian ecosystems. For example, the taxa of the Fabaceae family are important for the dynamics of poor-nutrient ecosystems due to morphological adaptations, such as nodules in the roots (Oliveira et al. 2012).
Variation in environmental conditions across phytophysiognomies along the altitudinal gradient, such as lower temperatures and shallow soils with low water-holding capacity seems to be a relevant factor leading to variation of species richness, ses.PD, ses.MPD, and ses.MNTD in plant communities in the Tepequém table mountain. Harsh climatic and edaphic conditions in high altitudes represent a filter to plant species' establishment and growth Cordeiro and Neri 2019), consequently influencing the richness and evolutionary history of communities. Between ORG and FOR, there is an altitudinal difference of 650 m, and a corresponding decrease of ca. 3.90 °C in temperature. Such a correlation has been observed in other tepuis, which show an adiabatic temperature decrease of 0.60 °C/100 m elevation (Huber 1995;Vegas-Vilarrúbia et al. 2012), and in several tropical mountain ecosystems, which show values ranging between 0.50 and 0.70 °C at each 100 m (Safford 1999;McCain and Grytnes 2010). Along the altitudinal gradient, other harsh climate conditions related to high altitudes, such as high solar radiation, high evapotranspiration, and wind exposure may prevent the establishment of many species Zhang et al. 2014) and can account for the results obtained.
The ses.MPD of the ORG and FOR was close to zero, suggesting a random phylogenetic structure when deep (old) phylogenetic nodes are considered. ORG showed a similar pattern when only shallow (recent; ses.MNTD) nodes were considered. This pattern may suggest that neutral processes, such as dispersal limitation and ecological drift (Hubbel 2001;Vellend 2010), can shape local communities (e.g. Chun and Lee 2018;Liu et al. 2019a) or that environmental filtering and competitive exclusion act simultaneously in different phytophysiognomies in the Tepequém table mountain (Mayfield and Levine 2010; Liu et al. 2019a). Conversely, ses.MNTD in FOR resulted in a clustered pattern. Our findings indicate that although the Forest phytophysiognomy is characterised by a phylogenetically random tree community for deep phylogenetic nodes (i.e. order and family level), this same community has species more closely related than expected by chance when only terminal nodes are analysed (i.e. intra-familial and/or intra-generic level). We suggest that the greater species richness and clustering pattern found in FOR might be related to specific edaphic conditions, such as lower nutrient availability (i.e. Ca 2+ , Mg 2+ , remaining phosphorus, Mn, Zn, organic matter). In rupestrian ecosystems, the soil is an important driver at the local scale on plants' taxonomic (Nunes et al. 2015;Mota et al. 2018) and phylogenetic diversity (Miazaki et al. 2015;Zappi et al. 2017;Pontara et al. 2018;Campos et al. 2021).
The highest values observed in species richness may be expected in environments with lower nutrient availability (Huston 1993;Nadeau and Sullivan 2015) since soils with lower fertility normally have reduced interspecific competition (Gastauer and Meira-Neto 2014); this is because efficient competitors may lack resources to outcompete inferior competitors, thereby causing higher tree species number (Tilman 1985). In addition, this supports the idea that harsh environments can select for species with similar adaptive mechanisms in response to environmental stress (Qian and Sandel 2017).According to Muscarella et al. (2018), association between phylogenetic clustering and soil fertility may reflect selection on phylogenetically conserved traits in low-fertility areas in the Amazonian rain forests.
Given the tendency of predominance towards niche conservatism in the rupestrian grasslands (Zappi et al. 2017), we suggest that the phylogenetic clustering pattern found in the shrubby community assembly might also be related to a specific combination of abiotic constraints (e.g. variability of soil texture, fertility properties, or altitude; Cavender- Bares et al. 2009). Multiple drivers may generate similar phylogenetic patterns (Cadotte and Tucker 2017), with environmental filters most common in rupestrian habitats (Miazaki et al. 2015;Zappi et al. 2017Zappi et al. , 2019.Compared to other phytophysiognomies, the SRG showed the smallest finer soil particle contents (i.e. clay, silt, and fine sand); favoring the lowest moisture retention and nutrient availability . The physical properties of the soil, such as texture, help to explain the different niches by determining which limiting resources are distinctly available to species ). However, we suggest that other abiotic filters not evaluated in this study area could shape the phytophysiognomies in the phylogenetic clustering pattern, such as the temperature and soil water status (Ferrari et al. 2016) and rockiness (Pontara et al. 2018).
In addition, the phylogenetic clustering pattern found in SRG may be related to the higher competitive abilities of some species in this phytophysiognomy. Clustered phylogenetic patterns can also result from density-dependent biotic interactions (i.e. competition), once closely related species have similar competitive abilities (Mayfield and Levine 2010). This occurs when competitive exclusion occur among distantly related taxa (Mayfield and Levine 2010; Goberna et al. 2014). For example, SRG showed a high density of Byrsonima crassifolia (L.) Kunth and Byrsonima crispa A. Juss. individuals. Both species concentrate approximately 53% of the total individual numbers for this phytophysiognomy. The genus is known for the high production of secondary metabolites, mainly in leaves, with high phytotoxic and cytotoxic activity (Amâncio et al. 2019). However, we highlight that phylogenetic patterns discussed here should be improved by exact measurements of ecological traits for each species in the communities, followed by specific testing for the presence of a strong phylogenetic signal concerning these traits (Cavender-Bares et al. 2009). (Tepequém table mountain) of Brazilian Amazonian, plant community composition and key plant diversity dimensions (i.e. taxonomic and phylogenetic) varied considerably along the local-scale edaphic and elevational gradient. In this tepui, the floristic and phylogenetic compositions differed among the three main phytophysiognomies (Open Rupestrian Grassland, Shrubby Rupestrian Grassland and Forest). 2. Species richness and phylogenetic diversity decrease with increasing altitude. 3. The shrubby vegetation showed negative values for all standardised phylogenetic metrics (ses.PD, ses.MPD, ses.MNTD), indicating phylogenetic clustering at middle altitude, probably due to biotic interactions. At lower altitudes, the Forest showed a cluster pattern when only terminal nodes (recent; ses.MNTD) are considered, suggesting the disappearance of some of the rare species from the community. 4. The Forest (lowest altitude) and Open Rupestrian Grassland (highest altitude) showed random dispersion to deep (old; ses.MPD) phylogenetic nodes. Thus, both the neutral and deterministic processes may simultaneously shape the phylogenetic diversity and structure in these phytophysiognomies.

In an isolated tepui
5. The variation of species richness is explained mainly by altitude, meanwhile more variation in different phylogenetic metrics is explained mainly by variability of physical properties related to soil texture. 6. Comprehensive studies, including the role of environmental drivers in plant evolutionary history along the altitudinal gradient, are essential for understanding the mechanism involved in mountaintop plant community assemblies, and to provide additional information for conservation planning of this highly threatened ecosystem worldwide.