The thermal niche and phylogenetic assembly of evergreen tree metacommunities in a mid-to-upper tropical montane zone

Abstract

Frost and freezing temperatures have posed an obstacle to tropical woody evergreen plants over evolutionary time scales. Thus, along tropical elevation gradients, frost may influence woody plant community structure by filtering out lowland tropical clades and allowing extra-tropical lineages to establish at higher elevations. Here we assess the extent to which frost and freezing temperatures influence the taxonomic and phylogenetic structure of naturally patchy evergreen forests (locally known as shola) along a mid-upper montane elevation gradient in the Western Ghats, India. Specifically, we examine the role of large-scale macroclimate and factors affecting local microclimates, including shola patch size and distance from shola edge, in driving shola metacommunity structure. We find that the shola metacommunity shows phylogenetic overdispersion with elevation, with greater representation of extra-tropical lineages above 2000 m, and marked turnover in taxonomic composition of shola woody communities near the frost-affected forest edge above 2000 m, from those below 2000 m. Both minimum winter temperature and patch size were equally important in determining metacommunity structure, with plots inside very large sholas dominated by older tropical lineages, with many endemics. Phylogenetic overdispersion in the upper montane shola metacommunity thus resulted from tropical lineages persisting in the interiors of large closed frost-free sholas, where their regeneration niche has been preserved over time.

1. Introduction

Abiotic stressors such as freezing temperatures or extreme droughts are known to strongly influence the evolution of plant species and the assembly of plant communities [1–3]. Thermal gradients, such as those that occur when moving from tropical to temperate latitudes or from lower to higher elevations along mountain slopes, are characterized by predictable decreases in temperature and increases in seasonal and daily temperature variations. The ability to withstand cold temperatures and large fluctuations in temperatures are thus expected to be important drivers of plant species distributions along such gradients. Since the thermal tolerances of plants are highly conserved and adaptations to freezing temperatures require specific physiological adaptations [4,5], the evolutionary and phylogenetic histories of species become important in driving species distribution patterns across such environmental gradients [6–8].

Among the most well-recognized species distribution patterns is the latitudinal diversity gradient (LDG), where plant species richness, not only taxonomic but also phylogenetic, is highest in the warm, moist tropics and decreases towards the cold, dry poles [9,10]. While multiple hypotheses have been proposed to explain this distinct pattern, two evolutionary hypotheses have been examined at length in recent literature: the tropical niche conservatism (TNC) hypothesis [6] and the out of tropics (OTT) hypothesis [7,11]. These two hypotheses predict different patterns for the phylogenetic structure of diversity along thermal gradients. The TNC predicts phylogenetic clustering as we move from tropical to temperate climates, driven by evolutionary niche conservatism and the limited dispersal of a few related cold-adapted lineages from ancestral tropical clades. Relative to the TNC, the OTT predicts a pattern of greater phylogenetic dispersion in temperate climates, based on greater dispersal of ancient tropical lineages during warmer past climates, niche convergence rather than conservatism, and the diversification of lineages after arriving in temperate climates [6,7,10].

Similar to the LDG, but at localized geographical scales, thermal gradients occur along mountain slopes, where temperatures change rapidly with elevation. Here, studies indicate that phylogenetic structuring of plant communities along the elevation gradient differ between temperate and tropical mountains. In temperate mountains, community structure tends to be characterized by patterns of nestedness and phylogenetic clustering as elevation increases, following the TNC hypothesis [12–15]. However, in tropical mountains the pattern is one of phylogenetic overdispersion as elevation increases, driven by the co-occurrence of divergent tropical and extra-tropical lineages, which is consistent with the OTT hypothesis [11,16–19]. This pattern of phylogenetic overdispersion along the tropical montane gradient may stem from a combination of processes. First, freezing temperatures appear to have posed an obstacle to tropical species on an evolutionary time scale [4,5]. The potential physiological pathways for overcoming frost are limited, particularly for woody evergreen species [4]. At the extreme upward end of the tropical elevation gradient, where frost occurs, only evergreen species with traits necessary to survive episodes of freezing, can persist [4]. Thus, frost creates a colonization barrier for many lowland tropical clades, and an opportunity for subtropical and temperate lineages to establish in tropical montane areas affected by frost and cold temperatures. For instance, Malesian mountaintops both east and west of Wallace's Line are dominated by Southern Hemisphere conifers that are capable of long-distance dispersal, as few of the lower elevation lineages were able to evolve adaptations to colder environments [16].

On the other hand, tropical elevation gradients are much older than either latitudinal gradients or temperate elevation gradients [11]. Facilitated by this long time scale of their existence, many tropical lineages from lower elevations have been able to colonize and establish in higher elevations, such that tropical lineages also constitute a substantial and even major portion of tropical upper montane forests [16,20–23]. Here the underlying processes may be ecological rather than evolutionary, where otherwise frost-intolerant species may establish beneath canopies where frost and freezing temperatures do not penetrate. In these warm microclimates where freezing does not occur despite the high elevation, multiple tropical lineages, as long as they are able to disperse up the elevation gradient, may establish because their regeneration niche is preserved [24–26]. Such facilitative interactions that overcome environmental filtering by frost can also result in a pattern of phylogenetic overdispersion in tropical montane communities, by allowing deeply divergent tropical as well as temperate lineages to co-exist.

Here, we examine patterns of phylogenetic and taxonomic structure in non-climbing woody plant communities in evergreen forests (locally known as sholas), along a tropical elevation gradient in the Western Ghats of India, a global biodiversity hotspot. Shola forests exist as distinctive patches, characterized by stunted evergreen trees, within a grassland mosaic in the mountain tops of the southern Western Ghats, and are home to a biodiverse flora and fauna, many of which are endemic to this region [17,27]. Recent work establishes that frost inhibits the survival of native shola tree seedlings in open, high-elevation areas [25], thus likely acting as a colonization barrier. We hypothesized that: (a) shola forest communities will abruptly turnover in taxonomic composition between middle and upper montane zones (approx. 1700–2000 m and greater than 2000 m above sea level), where the latter are regularly exposed to freezing temperatures over larger areas but the former are not [28,29], (b) niche convergence rather than conservatism will drive community assembly, leading to a pattern of increasing phylogenetic dispersion from middle to upper montane zones, driven by tolerance of cold temperature. Specifically, we expect more representation of extra-tropical lineages in upper montane sholas and (c) since frost and cold temperatures limit shola tree seedling establishment in open high-elevation areas, but adult trees exist within forest patches at the same elevations, phylogenetic dispersion in upper montane sholas will also result from habitat-driven microclimatic variation that allows tropical lineages to establish and persist within closed shola forests.

2. Methods

(a) . Study area

The study was conducted over 60 km² in the southern and western parts of the upper Nilgiris plateau in south India (11.17° N, 76.77° E and 11.50° N, 76.43° E), between 1750 and 2400 m above sea level (electronic supplementary material, figure S1). Tropical montane evergreen forest (locally known as shola) forests occur as naturally discontinuous patches in a mosaic with native grasslands. In some areas, native grasslands have been converted to exotic tree plantations [30]. See [31] for a detailed description of the vegetation. Studies of soil carbon isotopes and fossil pollen show that sholas have expanded into grasslands during warmer, wetter periods and contracted during cooler, drier ones [32–34]. The western side of the plateau receives the highest rainfall (2500–5000 mm), mainly during the southwest monsoon, the southern and eastern portions of the plateau receive 1500–2000 mm annually from both the southwest and northeast monsoon and the central Nilgiris receive 900–1200 mm on average [34]. The dry season extends from December to March. Temperature ranges from a mean maximum of 24°C in April to a mean minimum of 5°C in December (see electronic supplementary material, table S1 for mean annual temperature and precipitation ranges by elevation zone). Absolute minimum temperature recorded at various observatories ranges from −6.7°C to 0°C [29]. Frost usually occurs between November and March, mainly in the depressions and valleys [29]. The average number of frost days per year shows high spatial variation (10–46), generally increasing with elevation. There is also high temporal variation; in more extreme years, frost can occur on one-third to half of the days during winter, with enough severity to cause significant economic damage, while in other years there have been less than 10 frost days [29]. Soils are acidic, rich in organic matter and described as non-allophanic andisols [34]. In general, human population density and therefore direct anthropogenic disturbance is relatively low in this region [35].

(b) . Data collection

Shola woody communities were sampled using 87 20 × 20 m (0.04 ha) plots, that were located using a stratified random sampling design based on topography and surrounding land cover. Shola patch size is an important determinant of community composition [31,35] and we sampled a total of 52 shola patches that varied widely in size (0.56 to more than 300 ha, median 7.9 ha). In very large patches (greater than 60 ha), three transects of plots were placed at a minimum of 250 m apart, with a distance of at least 50 m between each plot along the transect. In other patches, plots were spaced at least 100 m apart. The farthest distance between any pair of plots in the study was 31 km (electronic supplementary material, figure S1). All individuals greater than 1 cm dbh were censused within the plots, with species identity, diameter at breast height and height recorded. In addition, the distance to the nearest forest edge was noted, as well as the GPS location of the plot corner, elevation, slope and aspect. We confirmed species identities using field guides, flora [36,37] and the help of an experienced taxonomist.

(c) . Data analyses

(i) . Patterns in shola metacommunity taxonomic and phylogenetic structure along the elevation gradient

To analyse the phylogenetic structure of shola metacommunities, we derived a phylogeny for our study species from an angiosperm megaphylogeny [4,38,39]. Species names were checked and standardized to match accepted nomenclature from The Plant List (TPL, v. 1.1; www.theplantlist.org). We created the phylogeny in R using package ‘V.Phylomaker’ [39], and the function ‘build.nodes.1’. All the genera present in our study plots are represented in the megaphylogeny. However, only 34% of the study species matched species present in the megaphylogeny. Missing species were added within their respective genus nodes, using scenario ‘S3’, based on the analysis by Qian & Jin [40], which showed that S1 and S3 performed better than S2 for derivation of metrics from a megaphylogeny. We felt that S3 was evolutionarily more realistic compared to S1. Divergence in phylogenetic structure at the plot level was quantified using species abundance data to calculate two commonly used metrics—mean pair-wise phylogenetic distance (MPD) and mean nearest taxon distance (MNTD) [41–43]. The former measures mean phylogenetic relatedness among co-occurring individuals, resulting from both ancient as well as recent evolutionary events, while the latter measures the mean distance separating each individual in the community from its closest relative (recent evolutionary history). We tested the hypothesis that, on average, individuals were more (or less) phylogenetically related than random, by calculating the standardized effect sizes (ses) for each metric, using a null model (replicated 9999 times), that held both species richness and occurrence frequency constant (independentswap) [44]. The indices were therefore standardized for species richness [45] and positive sesMPD or sesMNTD values, with high quantiles (greater than 0.95), indicate phylogenetic overdispersion, while negative values, with low quantiles (less than 0.05), indicate phylogenetic clustering [42]. The analysis was conducted in R using the package ‘picante’ [46].

To assess whether phylogenetic metacommunity structure was related to elevation, we obtained Pearson's R correlations between plot-level sesMPD and sesMNTD values and elevation. Since there were only seven plots less than 2000 m, we tested whether the observed relationship was affected by uneven sampling, by randomly sampling seven plots from different shola patches in the 2000–2199 m and 2200–2400 m zones, respectively, calculating sesMNTD and sesMPD, and then regressing these against elevation. We repeated this process 100 times to check whether the regression slopes were consistent. While the slope coefficients of the regression of sesMPD with elevation were not sensitive to smaller sample size, those of sesMNTD were (electronic supplementary material, figure S1). Therefore, the results of sesMNTD correlation with elevation are not presented.

We assessed the pattern of taxonomic metacommunity structure along the elevation gradient using the ‘elements of metacommunity structure’ (EMS) framework [47,48]. This approach is based on three metrics (coherence, turnover, and boundary clumping), calculated using an ordinated species presence–absence matrix. The statistical significance of these metrics was tested against randomized null matrices in which row and column totals were held constant (‘fixed-fixed null’ model) [49], using a z-test. The analysis was conducted in R using the package ‘metacom’ [50].

We then examined whether there was a large change in taxonomic metacommunity structure above and below 2000 m, by dividing the elevation gradient into three zones (zone 1: 1750–1999 m; zone 2: 2000–2199 m; zone 3: 2200–2400 m), and conducting a pair-wise comparison of taxonomic similarity between plots in each of these successive zones, using Sorenson's index. This analysis was implemented using the ‘betapart’ package in R [51].

Finally, since plots at shola edges are more likely to be affected by low and freezing temperatures [25,52], we used Ward's minimum variance clustering [53], which minimizes within cluster sums of squares, to determine whether the shola woody communities occurring below 2000 m and close (less than or equal to 50 m) to the forest edge, differed in terms of their species composition from similar communities, in similar-sized shola patches, above 2000 m elevation. For this comparison, we used a total of 11 plots, as the remaining plots were either greater than 50 m from the edge or were in forest patches that were much larger or smaller than the less than 2000 m sholas.

(ii) . The influence of macroclimate and microclimate on taxonomic and phylogenetic metacommunity structure

To test whether taxonomic variation in metacommunity structure was influenced by climate and habitat-based surrogates for microclimate (i.e. forest patch size and distance to edge) we ran a partial redundancy analysis (RDA) on the Hellinger-transformed species abundance data (all stems greater than 1 cm dbh). This allowed us to control for the effects of spatial autocorrelation in the data by partitioning out the effect of space (x and y location of plot corners). The predictors included in the main RDA model were: mean diurnal temperature range, temperature seasonality, maximum temperature of the warmest month, minimum temperature of the coldest month, seasonality of precipitation, log of the shola patch area and distance of the plot to the nearest forest edge. All predictors were standardized and checked for collinearity and none were found to have Pearson's r > 0.7. We used a permutation test to assess model significance [53]. Data for the climatic predictors, at 1 km resolution, was obtained from the BIOCLIM database [54]. Shola patch size was calculated in QGIS [55], after digitizing each sampled patch using 2.5 m resolution QuickBird imagery in Google Earth (earth.google.com; Digital Globe Imagery from 2008 to 2011, accessed August 2011). The analysis was conducted in R using the ‘vegan’ package [56].

To examine whether phylogenetic metacommunity structure was influenced by climate, we regressed abundance-weighted sesMPD and sesMNTD against the macro and microclimatic predictors described above.

Finally, we analysed the relationship between elevation and log patch area and the number of endemic species within a plot, using a Poisson regression.

(iii) . Spatial pattern of phylogenetic structure and its relationship with macro and microclimate

Earlier work has established that shola communities are compositionally heterogeneous in space [31]. In order to understand the spatial pattern of phylogenetic variation across the metacommunity, and how it relates to micro- and macroclimate, we analysed the phylogenetic tree using specific-overrepresentation scores (SOS) and associated geographic node divergence scores (GND) [57]. SOS compares the species richness of sister clades, for each node in the phylogeny, at every site, to the expectation from a null model. We obtained the null distribution by running the ‘quasiswap’ algorithm 100 times. The latter randomizes species occurrences, while holding site richness and species range size constant [57]. GND is calculated from the mean of logit P across all sites, where P is the proportion of the simulated distribution of richness values that is more extreme than the observed value. GND ranges between 0 and 1, with high values for a given node indicating that there is high divergence in spatial distribution between its descendant clades. This analysis was run using the ‘nodiv’ package in R [57]. For each of the nodes with high GND scores, we modelled the relationship between the predictors listed above and plot-level SOS scores (for that node) using multiple regression.

3. Results

A total of 75 tree and shrub species, belonging to 31 families and 50 genera (electronic supplementary material, table S2), were recorded from the study plots. The dominant families in terms of species number and basal area are Lauraceae, Rubiaceae and Myrtaceae, followed by Symplocaceae, Acanthaceae, Celastraceae and Sapotaceae. These are also among the families with the highest levels of endemism (electronic supplementary material, figure S3). Litsea and Syzygium were the most species-rich genera, followed by Cinnamomum and Symplocos. The most abundant tree species were, Symplocos foliosa Wight and Litsea wightiana (Nees) Hook. f., while some of the rarer species were: Syzygium densiflorum Wall. ex Wight & Arn., Elaeocarpus recurvatus Corner, Isonandra montana (Thwaites) Gamble and Olea paniculata R.Br. There was wide variation in the relative abundances of endemic species, with some endemics showing greater abundance and dominance than cosmopolitan species such as, Ilex denticulate Wall. ex Wight, Eurya nitida Korth. and Ternstroemia gymnanthera (Wight & Arn.) Sprague. Approximately 54% of the species belong to families with a tropical–subtropical distribution, while 25% from tropical families, another 16% from extra-tropical families and 6% from cosmopolitan families (electronic supplementary material, table S2). A detailed description of the vegetation structure and taxonomic composition of the shola woody community is provided in [31].

(a) . Taxonomic metacommunity structure along the elevation gradient

The results of the EMS framework analysis indicate that the metacommunity contained fewer embedded absences than expected under null model simulations (z = 2.49, p = 0.013), implying positive coherence. There were also more species' replacements than expected (z = −4.85, p < 0.001), implying significantly high turnover among sites.

Finally, species’ range boundaries were highly clumped and significantly different from the null expectation (Morisita's index = 3.52, p = 0). The site scores of the ordinated species presence-absence matrix were significantly correlated with plot elevation (R_s = −0.5, p < 0.001). Therefore, at the scale of the study area, the metacommunity exhibits a Clementsian structure, in which co-occurring species occupy similar ranges and range boundaries coincide along the elevation gradient.

(b) . Phylogenetic metacommunity structure along the elevation gradient

The abundance-weighted community phylogenetic index, sesMPD was significantly positively correlated with elevation (Pearson's r = 0.39; electronic supplementary material, figure S4) indicating phylogenetic overdispersion along the elevation gradient. sesMNTD values were not significantly different from random at the plot level and did not show a significant relationship with elevation.

(c) . Metacommunity turnover above and below 2000 m

The mean value of the pair-wise Sorenson's Index between plots in elevation zone 1 (1750–1999 m) and zone 2 (2000–2199 m) was 0.51, while the mean value of the pair-wise Sorenson's Index between plots in zones 2 and 3 (2200–2400 m) was 0.45, indicating that woody communities showed greater taxonomic differences above and below 2000 m compared to those above 2000 m. We also found increased total basal area of extra-tropical and cosmopolitan lineages in shola woody communities above 2000 m (figure 1).

Figure 1.

Bar chart indicating that most extra-tropical (EXT) and cosmopolitan (COS) angiosperm families show an increase in proportion of total basal area sampled with elevation in the upper Nilgiris plateau. (Online version in colour.)

(d) . The influence of macroclimate and microclimate on taxonomic metacommunity structure

The results of the partial RDA showed that after accounting for the variation in species' abundances explained by space, the selected combination of macro and micro climatic predictors was able to explain 21% of the remaining variation ($R_{\mathrm{adj}}^2=0.213$, p < 0.001). RDA 1 and 2 together accounted for 15.5% of the explained variation. Minimum temperature of the coldest month, log of patch area, maximum temperature of the warmest month and mean diurnal temperature range were the main factors behind the dispersion of sites on RDA 1, while for RDA 2 the main constraints were distance from edge, mean diurnal temperature range and log patch area (figure 2). Litsea wightiana (LIWI), Berberis leschenaultii (BELE) and Syzygium grande (SYGR) abundances were strongly negatively correlated with minimum temperature of the coldest month, while Litsea stocksii (LIST), Neolitsea zeylanica (NEZE) and Actinodaphne lanata (ACLA) abundances were highly positively correlated with the same predictor. The dominant understorey shrubs, Psychotria nilgirensis (PSNI), Lasianthus venulosus (LASI) were strongly positively correlated with mean diurnal temperature range, as was Symplocos foliosa (SYFO). Finally, abundances of Cryptocarya lawsonii (CRLA), Litsea floribunda (LIFL) and Saprosma foetens (SAFO) were strongly positively correlated with distance from forest edge and log patch area (figure 2).

Figure 2.

Partial RDA triplot showing RDA 1 and 2, with fitted scores and correlations between species (cross symbols) and predictors related to macro and micro climate (blue arrows). Species showing high correlations with predictors are labelled as follows: ACLA, Actinodaphne lawsonii; CRLA, Cryptocarya lawsonii; BELE, Berberis leschenaultii; LASI, Lasianthus venulosus; LIFL, Litsea floribunda; LIST, L. stocksii; LIWI, L. wightiana; MESI, Meliosma simplicifolia; NESE, Neolitsea zeylanica; PSNI, Psychotria nilgirensis; SAFO, Saprosma foetens; SYFO, Symplocos foliosa; SYGR, Syzygium grande. (Online version in colour.)

Woody communities at or near the shola edge below 2000 m elevation were taxonomically distinct from those above 2000 m (figure 3).

Figure 3.

Dendrogram showing clustering of plots in edge and near-edge shola woody communities, using Ward's minimum variance clustering algorithm on Hellinger-transformed species abundance data. Number shown at tips of dendrogram correspond to plot numbers.

(e) . The influence of macroclimate and microclimate on phylogenetic metacommunity structure

Results of the multiple regression of abundance-weighted sesMPD with macro and microclimatic predictors ($R_{\mathrm{adj}}^2=0.14$, p < 0.01), indicated that it was significantly negatively correlated with temperature seasonality, minimum temp of the coldest month and log of patch area (table 1). These results suggest there was greater phylogenetic overdispersion in plots with lower temperature seasonality and lower minimum temperatures, which occurred above 2000 m. However, plots in the interiors of a few very large shola patches, which occurred above 2000 m, showed greater phylogenetic clustering at both basal (table 1) and terminal branches of the phylogeny; abundance-weighted sesMNTD was significantly negatively related to log of patch area (β = −0.31, p < 0.05). sesMNTD was not significantly related to any of the macroclimatic predictors. Further, endemic species richness was more strongly influenced by log patch area (β = 0.11, p < 0.001) compared to elevation (β = 0.0008, p < 0.05; electronic supplementary material, table S3).

Table 1.

Results of multiple regression of abundance-weighted sesMPD against temperature and habitat-based surrogates for microclimate, showing predictors and their effect sizes.

predictor	β	2.5% CI	97.5% CI	p
temperature seasonality	−0.333	−0.57	−0.09	<0.01
log of patch area	−0.327	−0.64	−0.02	<0.05
min. temp coldest month	−0.317	−0.63	−0.0004	<0.05
distance from forest edge	0.252	−0.03	0.53	0.08
mean diurnal temp range	−0.077	−0.36	0.21	0.59

(f) . Spatial variation in phylogenetic patterns and its relationship with macro and microclimate

The spatial patterns in phylogenetic metacommunity structure across the study area appear to be driven by ancient evolutionary history, as nodes showing high GND scores tended to occur along the deeper branches of the phylogeny (figure 4), with the root node, separating the magnoliids from eudicots, showing the greatest GND in the study area. Similarly, the nodes separating the Asterids and the Rosids, the Fabidae from the Malvidae and the Ericales from the rest of the Asterid lineages also showed higher GND (0.39; figure 4). However, more recent evolutionary events, within the most dominant lineage in the study area, Lauraceae, have also contributed to the observed variation in metacommunity phylogenetic structure. High GND was observed in the node separating Cryptocarya and Beilschmeida from younger genera and terminal nodes separating the genus Cinnamomum from Litsea, Neolitsea and Actinodaphne, as well as the node separating Litsea from Neolitsea and Actinodaphne (figure 4).

Figure 4.

GND scores for each node in the phylogeny of shola woody species. Nodes are coloured and scaled according to GND. Nodes with GND scores greater than 0.3 are labelled with respective node numbers. Nodes with descendant clades having a single species were excluded from the analysis. Unresolved species are shown as polytomies. Node 1 separates Magnolids from all other angiosperms; node 5, Rosids from Asterids; node 6, Ericales from other Asterids; node 7, Gentianales, Icacinales, Lamiales from Aquifoliales and Apiales; node 21, Primulaceae, Theaceae and Sapotaceae from Symplocaceae and Ericaceae; node 33, Malvidae from Fabidae; node 55, Cryptocarya spp. and Beilschmeida spp. from other Lauraceae; node 57, Cinnamomum spp. from Litsea, Neolitsea and Actinodaphne spp., node 59, Litsea spp. from Neolitsea and Actinodaphne spp. (Online version in colour.)

Patch area and minimum temperature were the main drivers of spatial pattern in phylogenetic metacommunity structure and were significant predictors of SOS scores for several of the nodes with high GND—both at deeper branches of the tree as well as terminal ones (table 2 and figure 4). Magnoliid lineages had higher SOS scores in larger patches, compared to eudicots. Within the magnoliid clade, differences in spatial distribution between Cryptocarya and Beilschmeida and younger Lauraceae genera were also significantly related to patch area and minimum temperature (node 55 in figure 4 and table 2). Mean diurnal temperature range was significantly related to spatial variation in the distribution of genera within Lauraceae (node 55 in figure 4 and table 2), but to a lesser degree compared with minimum temperature. Symplocaceae and Ericaceae showed significantly greater SOS in sites with lower minimum temperatures and greater precipitation seasonality, compared to Primulaceae, Theaceae and Sapotaceae (table 2).

Table 2.

Results of multiple regression of specific-overrepresentation score (SOS) scores for nodes in phylogeny with high geographic node divergence (GND) scores (greater than 0.3), with macro and microclimatic predictors. Results only shown for nodes related to the dominant lineages in the study area. Node number corresponds to the node of the phylogenetic tree in figure 5.

node	separation of daughter lineages	significant predictors	β	2.5% CI	97.5% CI	p
1	Magnoliids from Eudicots	ln shola patch area	−0.75	−1.09	−0.41	<0.001
5	Asterids from Rosids	ln shola patch area	0.44	0.14	0.75	<0.01
		precipitation seasonality	−0.32	−0.65	−0.0006	<0.05
21	Symplocaceae, Ericaceae from Primulaceae, Theaceae, Sapotaceae	min. temp. coldest month	−0.63	−0.93	−0.32	<0.001
		precipitation seasonality	0.39	0.06	0.72	0.02
55	Cryptocarya and Beilschmeida spp. from other Lauraceae	min. temp. coldest month	0.40	0.16	0.65	<0.01
		ln shola patch area	−0.39	−0.63	−0.15	<0.01
		mean diurnal temp range	0.29	0.08	0.49	<0.01
57	Cinnamomum spp. from Litsea, Neolitsea & Actinodaphne spp.	min. temp. coldest month	−0.43	−0.80	−0.07	<0.05
		temp. seasonality	−0.41	−0.81	−0.02	<0.05
59	Litsea spp. from Neolitsea & Actinodaphne spp.	ln shola patch area	0.57	0.23	0.90	<0.01
		min. temp. coldest month	−0.54	−0.88	−0.20	<0.01

4. Discussion

In this study, we show that, along a tropical elevation gradient in southern India, the taxonomic and phylogenetic structure of the shola (montane forest) metacommunity is strongly influenced by the temperature gradient, specifically, minimum winter temperature and temperature variation, both diurnal and seasonal. Above and below 2000 m, there are marked shifts in metacommunity composition, driven by species turnover. Further, woody communities near forest edges, that are more likely to be exposed to greater diurnal and seasonal temperature fluctuations, as well as to the effects of frost [25,52], show distinct differences in species composition above and below 2000 m.

Minimum winter temperature was an influential predictor of the taxonomic and phylogenetic structure of shola woody metacommunities along the elevation gradient, driving differences in the distributions of tropical and subtropical lineages, of older and younger genera in the dominant family of Lauraceae, and also among younger genera within Lauraceae. Meher-Homji [52] speculated that diurnal temperature variation, leading to excessive transpiration, was more likely to cause mortality of evergreen shola saplings, rather than frost per se. We did find that diurnal temperature range was a significant predictor of variation in taxonomic structure, lending some support to this hypothesis. However, minimum winter temperature, which is correlated with frost formation [29], was a more influential predictor of variation in both taxonomic and phylogenetic structure. Since very few tropical evergreen lineages have been able to adapt to tolerate freezing [4], frost and freezing temperatures are likely to be more of a limiting factor for tropical evergreen woody communities compared to high diurnal temperature fluctuations, which are found to varying degrees in many different kinds of environments.

(a) . Phylogenetic overdispersion with elevation: niche convergence or facilitation

Our finding of a negative relationship between sesMPD and elevation indicates increasing phylogenetic dispersion at higher elevations, similar to those reported from other tropical elevation gradients [11,17,58,59]. A caveat here is that, despite standardization, these indices (sesMPD and sesMNTD) are known to be influenced by variation in species richness in the presence of environmental or biotic filtering, making their interpretation problematic [60]. In this study however, species richness (rarefied to account for variation in stem density across plots) does not vary systematically along the elevation gradient (β = −2.86 × 10⁻⁰⁵, p > 0.5), suggesting that the positive relationship between sesMPD and elevation is unlikely to be an artefact of underlying changes in species richness.

Previous studies have interpreted phylogenetic overdispersion to be an indication of the importance of interspecific interactions (i.e. competitive exclusion [41,61], or facilitation [24]), while phylogenetic clustering has often been interpreted as an indication of environmental filtering by abiotic factors, selecting for species with similar traits and therefore shared evolutionary histories [12,15,17,62]. However, phylogenetic overdispersion could also result from environmental filtering on distantly related taxa based on ecologically important convergent traits [41,63]. Therefore, phylogenetic overdispersion may result either from processes linked to environmental filtering or from those linked to biotic interactions.

The importance of minimum winter temperatures in driving community structure in our study suggest that, as reported from other tropical upper montane habitats [26], environmental filtering by frost and freezing temperatures may be important in structuring evergreen woody assemblages. Zanne et al. [4] show that the evolution of traits necessary to allow woody evergreen species to survive in freezing environments preceded their colonization of these habitats. In these shola communities, phylogenetic overdispersion at higher elevations is driven by greater evenness in the abundances of magnoliids and eudicots at higher elevations (electronic supplementary material, figure S5; see also [45,64]). Plots above 2000 m show a reduction in the relative dominance of magnoliid lineages and an increase in the abundance of eudicots. The former have largely tropical distributions [52,65] and are likely to have boreo-tropical or Gondwanan ancestry [65,66], whereas all of the lineages with extra-tropical distributions (and adaptation to cold temperature) fall within eudicot clades. Therefore, increasing phylogenetic overdispersion with elevation, driven mainly by the deepest branches of the phylogenetic tree, could indeed be the result of environmental filtering for cold and frost tolerance acting at the longer time scale of biogeographic processes that have shaped the current mix of tropical, subtropical and temperate lineages in the study area [16,17,52,59,67]. Hence niche convergence, rather than niche conservatism (as in the case of the latitudinal and temperate montane thermal gradient), could be driving metacommunity assembly in shola forests along the elevation gradient. Further studies on plant functional traits linked to cold and frost tolerance, and their relationship with phylogenetic history are required to confirm this hypothesis, both within tropical and subtropical lineages, as well as between the tropical and extra-tropical lineages that constitute these upper montane communities.

A more proximate explanation for the presence of phylogenetic overdispersion in the study area is the facilitation of frost and cold intolerant evergreen woody shola species by frost and cold tolerant species with temperate or sub-tropical affinities [32,52]. Such facilitative interactions are known to generate phylogenetic overdispersion in arid and semi-arid ecosystems [24,68]. As the regeneration niche tends to be strongly phylogenetically conserved [24], tropical evergreen woody species may not be able to establish in freezing climates without facilitation by frost-tolerant, extra-tropical ‘nurse’ species that could ameliorate microclimatic conditions [69], gradually creating suitable conditions for the establishment and survival of cold and frost-intolerant tropical species. Recent experiments demonstrate that dominant shola species fail to recruit in open grasslands, where they suffer high mortality from cold temperature and frost, whereas they establish within sholas where temperatures do not drop below freezing [25]. Further, studies based on fossil pollen indicate that the establishment of shola vegetation in open grasslands, about 35 000 years ago, followed the establishment of shrubby genera with temperate and sub-tropical affinities [32].

In support of this, we found that patch size appears to be an equally influential predictor of plot-level community taxonomic and phylogenetic structure relative to minimum winter temperature. Plots in the interiors of very large shola patches (greater than 100 ha) were significantly more phylogenetically clustered than plots in smaller and medium sholas. While the known correlations between species richness and sesMPD and sesMNTD [60] advise caution in interpreting these data, we did find that Lauraceae, a dominant magnoliid family with the greatest number of endemics in the study area (electronic supplementary material, figure S3), tended be the most abundant in the interiors of the very large shola patches. This was especially true for older, pantropical genera such as Cryptocarya, which are likely to have Gondwanan ancestry [65]. Our phylogenetic analysis therefore supports findings of previous taxonomic and phytogeographical studies, which show that larger patches appear to represent older successional stages—with a greater proportion of ancient tropical lineages, while sub-tropical and temperate species are better represented within smaller shola patches and edge habitats [35,52].

Larger forest patches have larger core areas, that are buffered from the effects of cold and freezing temperatures and diurnal temperature fluctuations [70,71]. By contrast, smaller patches can be affected by a pervasive edge microclimate—the result of having more area exposed to multiple edges [72], and therefore to colder temperatures and higher diurnal temperature fluctuations. The ability of patch area to explain variation in phylogenetic metacommunity structure highlights the critical role of local microclimates and facilitative interactions in large patches, in allowing the persistence of tropical evergreen woody species in these tropical upper montane environments.

(b) . Conservation implications for tropical upper montane forest-grassland mosaics

In this tropical montane forest ecosystem, we find evidence of a distinct upper montane woody assemblage above 2000 m. Several species that are community dominants below 2000 m (e.g. Litsea stocksii Hook. f., Canthium dicoccum (Gaertn.) Merr., Scolopia crenata Clos, Syzygium lanceolatum (Lam.) Wight & Arn.), did not occur above 2000 m. Conversely, 10 species (including five endemics) that were common in plots above 2000 m, were not found below this elevation. Further, eight species were only found above 2100 m, constituting the highest elevation group: Microtropis ramiflora Wight, Vaccinium leschenaultia Wight, Elaeocarpus recurvatus Corner, Sarcococca saligna Müll. Arg., Ternstroemia gymnanthera (Wight & Arn.) Bedd., Hedyotis articularis R. Br. ex G. Don, Rhodomyrtus tomentosa (Aiton) Hask. and Rhododendron arboreum ssp. Nilagiricum (Zenk.). The shola woody assemblage occurring above 2000 m is therefore a distinct upper montane community and should be explicitly considered as such, which is not the case at this time. Future conservation, management and restoration of this tropical montane forest system must thus separately target the upper and lower elevation sholas to conserve the entirety of its biodiversity.

The high spatial variation in phylogenetic diversity found in this study corroborates the ancient nature of these forest-grassland mosaics [73] and supports the crucial role of microclimatic refuges in upper montane habitats [74]. Area under tropical evergreen forest decreases significantly above 2000 m elevation in the Western Ghats [75], making large (greater than 50 ha) shola patches rare above 2000 m. This steep reduction in availability of core forest niche space could increase competitive pressure between more closely related tropical and subtropical lineages, leading to increased diversification within genera [17,63], and higher levels of endemism, as shown here and in previous studies [17,27]. Large shola patches above 2000 m could therefore be critical, not just to the persistence of tropical lineages in upper montane areas, but also in driving the further diversification of these lineages by providing temporally stable climatic refuges [17].

Data accessibility

Data supporting this paper are available from the Dryad Digital Repository: https://doi.org/10.5061/dryad.70rxwdc14 [76].

The data are provided in electronic supplementary material [77].

Authors' contributions

A.A.D.: conceptualization, data curation, formal analysis, methodology, visualization, writing—original draft, writing—review and editing; J.R.: conceptualization, methodology, visualization, writing—original draft, writing—review and editing.

All authors gave final approval for publication and agreed to be held accountable for the work performed therein.

Conflict of interest declaration

We declare we have no competing interests.

Funding

A.A.D. was funded by the DBT-RA programme in Biotechnology and Life Sciences, Government of India, the support of which is gratefully acknowledged.