Found in different continents worldwide, old (ancient), climate-buffered, infertile landscapes (termed Ocbils) harbor high endemism and diversity, as exemplified by the Australian kwongkan, South African fynbos, Guayana shield tepuis (Hopper, 2009). These habitats are believed to have persisted over very long timeframes on the scale of many millions of years due to the geological stability and climatic buffering of their locations. The specialized environmental conditions of such habitats, and the long timeframe over which they have persisted, represent a useful system in which to explore the collective ecological and evolutionary response of plants that exist in them.

Brazilian rupestrian habitats, locally known as ‘campos rupestres,’ are an example of a long-standing climate-buffered infertile landscape that contains many endemic species and lineages, which are adapted to very shallow and nutrient-poor soils (Silveira et al., 2016). The highlands in which most campos rupestres occur are situated primarily in the states of Minas Gerais and Bahia, with outliers in Goiás and Tocantins. Thus, they are set within three major Brazilian biomes: to the Southeast, they are embedded in the westernmost part of the Atlantic Rainforest in the state of Minas Gerais, while to the north and west of Minas Gerais state they are found within the Cerrado savannas. In Bahia state, the campos rupestres occur within the Caatinga semi-arid biome. The campos rupestres of the highlands of the Espinhaço range, which extends through Minas Gerais and Bahia, are the better botanically explored and have long been recognized as sites of exceptional biodiversity and endemism (Stannard, 1995; Giulietti et al., 1997) and, more recently, as fulfilling several of the criteria used to characterize Ocbils (Hopper et al., 2016; Silveira et al., 2016). While campos rupestres are particularly associated with the Espinhaço range, they also occur on even more ancient rock formations in the Quadrilátero Ferrífero, to the south of the Espinhaço (Schaefer et al., 2016). Along the Espinhaço range, different vegetation types can be found (Figure 1). Forests and savannas mostly grow at lower altitudes or along the rivers (Figures 1c,d), while patches of campo rupestre grasslands develop at the more exposed sites of mountain tops (Figure 1a). Occurring on either quartzitic, arenitic or ironstone substrates, typically at altitudes >900 m above sea level, campos rupestres are estimated to occupy 66,450 km² and to be home to over five thousand known species (Silveira et al., 2016), of which over two thousand are recorded as endemic to this habitat type and occur over a relatively small area (Brazil Flora Group [BFG], 2015).

The high levels of floristic diversity and endemism for which campo rupestre vegetation is renowned have often been ascribed to the disjunct (mosaic) distribution of campo rupestre sites and resulting floristic influences from other habitats, especially savannas (Giulietti et al., 1997; Neves et al., 2018). Recent recognition of campos rupestres as an Ocbil has provided a wider theoretical framework within which the ecology and evolution of their extraordinary plant diversity is being interpreted in a global context and over extended geological timescales (Silveira et al., 2016). Two recent in-depth reviews (Hopper et al., 2016; Silveira et al., 2016) have shown several of the predictions of Ocbil theory to be clearly applicable to campos rupestres, and further studies of specific subsets of the campo rupestre flora are adding to this emerging picture. For example, in a study of 210 plant species from campo rupestre Dayrell et al. (2017) found that the majority (62.5%) had non-dormant seeds, the highest proportion of non-dormancy reported for any vegetation type to date and consistent with predictions of reduced dispersability in Ocbils. Although this study encompassed representatives of 20% of the plant families represented in campo rupestre, fewer than 5% of the plant species known from campo rupestre were treated, illustrating the scale of the challenge in understanding this exceptional flora.

While campo rupestre can be associated with different substrates, such as quartzite, ironstone or sandstone, the diversity and specificity of iron-ore plant communities and their occurrence on iron-rich substrates of enormous mining interest means that they are amongst the most threatened vegetation types in Brazil (Jacobi et al., 2007). Unprecedented pressure on natural resources such as iron-ore make Brazil second only to Australia in the quantity of seaborne iron-ore it exports to China (Lawrence and Nehring, 2015). There are three major iron-mining districts in Brazil, namely Carajás in Pará, Urucum in Mato Grosso, and the campos rupestres associated with the Quadrilátero Ferrífero (Iron Quadrangle) in Minas Gerais, where the iron-rich substrate is known as canga. Comparative studies of the flora of campos rupestres on quartzite and canga have shown strong correlations between plant species abundance, vegetation cover, coverage of individual species and soil properties between these two habitats (Vincent and Meguro, 2008; Messias et al., 2012a, 2013). Tolerance to (and accumulation of) high metal concentration has evolved independently in different phylogenetic lineages (Broadley et al., 2001), allowing plants of different lineages coming from various ecosystem types to occupy and diversify in iron-rich canga habitats. Such tolerance may represent an environmental filter shaping canga communities in addition to the many factors they share with other campo rupestre environments such as intense solar radiation, high evapotranspiration, large daily temperature variations and soils characterized by poor water retention and low nutrient availability (Jacobi et al., 2007).

The availability of increasingly comprehensive plant phylogenies has enabled exploration of the roles of environmental filtering (selection imposed by environmental extremes that favor specific traits), and niche conservatism, in which such traits tend to be shared among species that are closely related phylogenetically (Webb et al., 2002; Cavender-Bares et al., 2009). Strong environmental drivers, such as specialized substrates, are a major factor in shaping local biodiversity (e.g., Webb, 2000). Thus, we would expect campos rupestres on quartzite and canga to differ in terms of the impacts of environmental filtering on the floristic composition and phylogenetic structure of their communities.

To date the majority of studies examining phylogenetic structure over environmental gradients in tropical plant communities have focused on climate, with only a few examining other factors such as soil fertility and even these tend to be focused on tropical trees and palms (Lehtonen et al., 2015). Until recently, very few studies have focused on open vegetation formations, or on plant habits other than trees; but this is starting to change with studies on Caatinga (Moro et al., 2015), campos gerais (Moraes et al., 2016) and campos rupestres (Miazaki et al., 2015; Pugliesi and Rapini, 2015). Both of the latter studies focus on campos rupestres on quartzite; and, although they differ markedly in geographic scale and taxonomic scope, they both evidence phylogenetic clustering within quartzitic campo rupestre assemblages. Considering distribution records for a single family, the Apocynaceae, across the northern part of the Espinhaço range (widely known as the Chapada Diamantina), Pugliesi and Rapini (2015) found significant phylogenetic clustering which they attributed to niche conservatism and limited dispersal leading to in situ diversification and high density of microendemics. On a more local spatial scale, comparing the angiosperm communities on two sites within Itacolomi State Park in Minas Gerais state, Miazaki et al. (2015) concluded that environmental severity reduces phylogenetic diversity and increases phylogenetic clustering in campo rupestre vegetation, as predicted by the stress dominance hypothesis (Swenson and Enquist, 2007).

In this study we compare the composition of campo rupestre plant assemblages having different physiognomies (open and forest formations) and occurring on contrasting substrates (quartzite and canga) in the Espinhaço range and the Quadrilátero Ferrífero. We construct the first campo rupestre supertree (believed to be the first for any Ocbil) and use it to compare the phylogenetic structure of these contrasting assemblages. Based on earlier studies in campos rupestres, and consistent with Ocbil theory, we predicted that phylogenetic clustering would be prevalent in the assemblages analyzed. Assuming the prevalence of niche conservatism, consistent with Ocbil theory, we anticipated that the high levels of metals in canga might represent an additional environmental filter potentially resulting in accentuated phylogenetic clustering in canga assemblages. We also predicted significant over-representation of clades representing relatively old lineages, consistent with the Gondwanan Heritage Hypothesis of Ocbil theory (Hopper et al., 2016). We discuss our results in the context of recent publications demonstrating or inferring environmental filtering in campos rupestres (Negreiros et al., 2014; Miazaki et al., 2015; Pugliesi and Rapini, 2015) and those highlighting the campo rupestre ecosystem as a recently recognized Ocbil and a long-neglected conservation priority (Silveira et al., 2016; Neves et al., 2018).

Analysis of environmental impacts on phylogenetic structure have largely focused on broad-scale effects, such as climate. The present study is one of the few that have considered the role of soil substrate and vegetation physiognomy, two factors of critical importance to plant distributions. The use of environmental contrasts (geologically different substrates; open versus forested vegetation) provides a useful framework for investigation of the impact of environment on patterns of biodiversity. Moreover, our study is the most comprehensive study of community phylogenetic structure in campo rupestre to date and the only one to include sites on canga, a substrate that is a target for extractive industry and thus one of the most threatened habitats in Brazil (Jacobi and Carmo, 2008). Our aim was to explore variation in floristic and phylogenetic diversity and structure across different campo rupestre assemblages on contrasting rock formations within the Espinhaço range of Eastern Brazil and to explore whether the relative ages of lineages which are significantly over-represented in these assemblages provide support for the Gondwanan Heritage Hypothesis of Ocbil theory. We found evidence of distinct floristic assemblages in each of the four habitat types studied and detected significant phylogenetic structure in three of the four habitats, the exception being forest on canga (FC). FC sites were also notable for an over-representation of clades representing relatively young lineages consistent with expectations for young, often disturbed, fertile landscapes (Yodfels sensu Hopper et al., 2016) compared to those on quartzite which had an over-representation of clades from relatively old lineages as predicted for an Ocbil.

Similarity analysis clearly showed the importance of substrate to floristic composition, with most communities on canga having a flora more similar to that of other sites on canga than to the majority of sites on quartzite. The NMDS ordination also showed the importance of substrate, but physiognomy appeared even more important, with only one site crossing an otherwise clear separation between the floras of forest and open vegetation. Considered in combination, these results based on floristic composition provide strong support for our a priori recognition of four habitat types based on substrate and physiognomy: FC, FQ, OC and OQ. This is of great importance from a biodiversity perspective. The Espinhaço has substantial environmental heterogeneity related to both substrate, altitude, topographic position and microrelief. While the windward face of a mountain can harbor forests, the top of the same mountain bears campo rupestre, with forests restricted to small depressions where deeper soils accumulate. At a landscape scale, we can see sharp transitions between vegetation types and, as we see in our analysis, each habitat type has its own flora with its own environmental processes shaping its composition, in terms not only of individual species, but also of lineages.

We found phylogenetic clustering to be common in campo rupestre communities on both canga and quartzite and in both open and forest physiognomies, but the degree of phylogenetic clustering detected differed between the habitat types and with the metric used. Phylogenetic clustering was most evident in open assemblages, with both OC and OQ having mean NRI and NTI significantly different from zero, indicating significant phylogenetic clustering both near the tips of the phylogeny (NTI) and also involving the deeper nodes in the tree (NRI). Phylogenetic clustering was particularly pronounced in open assemblages on quartzite (OQ) with mean NRI and NTI values exceeding those reported by Miazaki et al. (2015) for angiosperms in open campo rupestre vegetation on quartzite at two sites in Minas Gerais and by Pugliesi and Rapini (2015) for Apocynaceae assemblages in the Chapada Diamantina, in the northern and drier part of the Espinhaço, largely covered by open campo rupestre formations associated with quartzitic outcrops. [In forest communities, however, NTI was not significantly different from zero on either substrate while mean NRI differed significantly from zero in FQ but not in FC].

Following Cavender-Bares et al. (2009), our results could be interpreted as likely due to the effects of environmental filtering, in an ecosystem where niche conservatism is prevalent. The congruence of our results for OQ with those of earlier authors could be argued to support the impact of relatively strong environmental filters inferred by Miazaki et al. (2015) and Pugliesi and Rapini (2015) and earlier documented by Negreiros et al. (2014) for non-woody species in open campos rupestres where a clear pre-dominance of stress-tolerant strategies was reported [and the few species with a greater level of competitiveness had relatively low abundance]. However, recent reviews of the application of phylostructure metrics to community ecology have cautioned against the use of phylogenetic patterns as proxies for community assembly mechanisms (Gerhold et al., 2015) and warned that the metaphor of the environmental filter has likely resulted in overstating the role of abiotic tolerance in shaping community structure (Kraft et al., 2015). These latter authors argue for a stricter application of the term environmental filtering, restricting it to situations where the abiotic environment can be demonstrated to prevent establishment or persistence in the absence of biotic interactions. Since experimental data of this kind are lacking for the vast majority of species in our Espinhaco megatree analysis, we cannot aspire to apply this stricter definition in the discussion that follows. Instead we exercise caution in discussing our results, indicating where evidence suggests there may be a role for environmental filtering but accepting that other potential equally valid explanations have not been explored.

Overall, phylogenetic structure in campo rupestre is influenced more by physiognomy than by substrate. Unpacking the relative importance of substrate and physiognomy and their interaction on our two measures of phylogenetic structure, analysis of variance showed that vegetation physiognomy was highly significant in explaining variation in NTI, while substrate and physiognomy were of equal importance (each only marginally significant) in explaining variation in NRI. Surprisingly, substrate x physiognomy interactions did not explain significant variation in either metric. The significance of physiognomy in explaining variation in NTI in our study is consistent with the observations that across campo rupestre the development of forest tends to occur in areas that are more favorable in terms of water and/or nutrient availability (Ferrari et al., 2016) and that in these fragmented campo rupestre landscapes the co-existence of forests and grasslands at the same altitude indicates that the distribution of tree species might be attributable to additional factors such as soil physical and chemical parameters (Coelho et al., 2018). In this context the lack of phylogenetic clustering of terminals in phylotrees representing assemblages from forest environments may be attributable to a relatively smaller role for environmental filtering in forests than in more environmentally stressful areas where open vegetation develops. Within a series of forest islands embedded in open campos rupestres on quartzite, Coelho et al. (2018) showed that high soil fertility and canopy cover were associated with greater height and basal area of trees, suggesting a relaxation of abiotic filters in the less harsh environment which supported forest, which could be consistent with the lack of signal in NTI observed in our forest samples.

Phylogenetic clustering was most evident in OQ, and the clade-by-clade results generated by nodesig revealed over-representation in at least half of the OQ sites in our study of two major clades of monocots: the larger being the commelinid monocot clade, which includes the Poales and three other orders of monocots, a result consistent with the wide diversity of different types of open vegetation encompassed within our OQ study sites. The smaller over-represented clade in OQ is nested within Poales, and comprises the cyperid clade (Cyperaceae, Juncaceae and Mayaca) together with the xyrid clade (Eriocaulaceae + Xyridaceae). These latter families include ‘everlasting’ flowers from the iconic megadiverse genera Paepalanthus and Xyris, long considered emblematic of open campo rupestre communities and an important source of income for rural, low-income families (Giulietti et al., 1996). The over-representation of these two monocot clades in OQ may be primarily attributable to niche conservatism with respect to life form, since monocots are mostly herbaceous species more prevalent in open habitats. However, our analysis shows that the xyrid clade is significantly under-represented in most study sites for OC, suggesting a difference between OQ and OC habitats attributable to factors other than life-form. One possible explanation lies in the nutrient-acquisition strategies of Eriocaulaceae and Xyridaceae. In their pioneering study of mineral nutrition strategies of campo rupestre plants, Oliveira et al. (2015) studied 50 species on quartzite- and arenite-derived substrates, and reported a range of root specializations associated with campo rupestre soils which are comparable to the most phosphorus-impoverished soils in the world. Roots covered in very fine root hairs over their entire length were reported for 32 species, most of them in the Eriocaulaceae and Xyridaceae, but also in Asteraceae. The authors invoke convergent evolution and Ocbil theory in interpreting the striking similarities in patterns of nutrient-acquisition strategies between campos rupestres, kwongan and fynbos, concluding that campos rupestres are home to several very old Gondwanan lineages which have evolved (in situ) a wide range of root specializations to survive P limitation. The ages of the two monocot clades which are over-represented in OQ sites are 67 and 96 Mya, respectively (Table 4), lending further support to the interpretation of open campo rupestre systems on quartzite as Ocbils. In contrast to the monocot over-representation in OQ sites, the only clade that our study showed to be over-represented in most OC sites is the genus Eremanthus (estimated age c. 20 Mya). These tree-like, woody Asteraceae are characteristic of the transition between forest and open habitats on canga, often forming homogenous stands known as ‘candeias.’

The cyperid + xyrid clade, so characteristic of OQ, is significantly under-represented in most of the study sites for FC and for FQ. FQ, was revealed by nodesig to have significant over-representation of the early branching clades in our Espinhaço megatree: specifically the Eudicot clade, the clade comprising Sabiales to Asterales and the pentapetalids. These three clades represent the oldest lineages which are significantly over-represented in any habitat in our study (Table 4), a result consistent with the inclusion of forest assemblages on quartzite within the campo rupestre Ocbil. While potentially attributable to niche conservatism of lifeform, this result is also consistent with our tree-wide phylostructure metrics for which phyloclustering was detected by mean NRI (reflecting relatively deep nodes) but not at the level of NTI and may reflect the relative isolation of most FQ sites from the humid forest biome. In a study of forest islands on quartzite in the Serra do Cipó, Coelho et al. (2018) described dispersal of tree species to mountain top forest patches via gallery forests on their eastern slopes but our FQ sites were, for the most part, embedded in the savannas of the Cerrado biome and or, in the case of the more northern sites, in the semi-arid Caatinga biome.

Our tree-wide metrics detected no significant phylogenetic structure across study sites for FC, with neither mean NRI nor mean NTI differing significantly from zero, but clade-by-clade analysis indicated significant over-representation of several dicot clades in FC. These included: the family Solanaceae and a subclade within it encompassing Solanum and closely related genera; a suborder of Sapindales soon to be recognized as ‘core Sapindales’ (J. R. Pirani personal communication); the clade formed by Myrtaceae together with Vochysiaceae, and nested within, the family Myrtaceae and the genus Myrcia. The prominence of Solanum and Solanaceae in FC is consistent with qualitative and quantitative descriptions of canga vegetation by earlier authors (Jacobi et al., 2007; Mourão and Stehmann, 2007). The importance of Myrtaceae and, specifically Myrcia, has also been highlighted previously, though not to the same extent, and may reflect the geographic location of the FC sites which are all close to the megadiverse Atlantic Forest biome (Stehmann et al., 2009; Brazil Flora Group [BFG], 2015) in which Myrcia is exceptionally diverse (Murray-Smith et al., 2009; Lucas et al., 2011; Nic Lughadha et al., 2012; Lucas and Bünger, 2015). Further evidence that proximity to the megadiverse Atlantic Forest biome may be a source of species from genera not commonly associated with campos rupestres is provided by the fact that three of the five genera which are most diverse in the Atlantic Forest in Brazil (Eugenia, Croton and Solanum) are represented by multiple species in FC, but not elsewhere in our study. Notably, the six clades reported above as significantly over-represented in the majority of FC sites are among the seven youngest over-represented lineages in our study (see Table 4).

Two very recent, synthetic studies of campo rupestre soils (Ferrari et al., 2016; Schaefer et al., 2016) have been of particular interest in the interpretation of our results because they include detailed comparisons of campo rupestre on quartzite and on canga. Using a conceptual framework very similar to the present study, Ferrari et al. (2016) monitored thermic and hydric dynamics of soils in four sites, contrasting open campo rupestre vegetation on canga (OC) and FC with open campo rupestre vegetation on quartzite (OQ) and FQ. Their results confirm the observations of many earlier authors (Meguro et al., 1996a,b), that soil depth is key, in both quartzite and canga landscapes, with forest occurring on much deeper soils than grassland. Furthermore, detailed monitoring of diurnal and seasonal changes in soil moisture showed that, in general, there was greater seasonal variation in soil moisture on canga, especially at the surface. Campo rupestre grasslands on canga (OC) and on quartzite (OQ) showed greater variation in soil moisture than did forest sites, with recorded moisture levels frequently indicating severe water deficits in grassland on canga (OC) and even more frequently in grassland on quartzite (OQ), a difference which Ferrari et al. (2016) attribute to the greater clay and silt content of the canga soils resulting in greater water retention and availability. No water deficits were recorded for FQ while interpretation of results for FC was complicated by technical/methodological issues. These results suggest that soil moisture factors may represent particularly strong filters in OQ.

Parallel monitoring of air and soil temperatures showed different patterns across the four sites compared by Ferrari et al. (2016). In general, air temperatures of grasslands were similar to those of forest, but soil temperatures in forest more closely reflected air temperature while grassland soil temperatures showed more seasonal and annual variation. The range of this variation is much greater in grassland on canga (OC) and soil temperature usually exceeds air temperature to a greater extent than in grassland on quartzite (OQ), a difference which the authors attribute to substrate albedo: dark, red, dense canga absorbing more energy than light, pale quartzite. Furthermore, the frequency and maximum duration of events in which soil temperature reached >35°C was highest in grassland on canga (OC). Surprisingly, events where the soil temperature exceeded 35°C were also recorded in FQ, albeit with lower frequency and duration than in grassland on canga (OC). Thus soil temperature factors may represent particularly strong filters in OC but also, secondarily in FQ.

A cluster analysis of soil profiles from high altitude rocky complexes across Brazil (Schaefer et al., 2016), distinguished two main types of campo rupestre soils differing more in physical than in chemical attributes: sandy soils with greater levels of exchangeable Al³⁺, associated with granite/gneiss and quartzite substrates as compared to soils with greater accumulation of organic matter and clayey/silty textures found on canga and itabirite. These authors conclude that the low biomass of campo rupestre vegetation in general is attributable to low nutrient levels, especially to low phosphorus (P) rather than to metal toxicity. They report P amounts to be particularly critical for vegetation on canga from mining areas including those in the Iron Quadrangle of Minas Gerais though, puzzlingly, the lowest P levels they report for any site within our study area are actually for a quartzite site within the Quadrilátero Ferrífero.

In summary, low nutrient levels are prevalent across campo rupestre, particularly acute in quartzitic soils, relative to those on canga, and a more significant constraint to growth than metal toxicity. Other factors such as soil moisture, water deficits and very high temperatures vary in frequency and intensity across the four habitat types studied and are variously reported to be more severe in grassland (on OQ and/or OC in the case of water deficits and soil temperatures) or FQ (soil temperatures). In these studies no single potential environmental filter was reported to be most severe for FC, a result congruent with the lack of phylogenetic clustering detected in this habitat. Together these considerations may suggest that this assemblage is less likely to be a result of relatively severe environmental filtering than others in our study, though of course in practice environmental factors need to be considered in combination rather than individually in order for their filtering effects to be understood.

Considered as a whole and in the context of the several recent studies on campo rupestre flora and soils on quartzite and canga substrates, our results on phylogenetic structure, ages of over-represented clades in each habitat and the identities of these clades, lend support to the recognition of both open and forest assemblages on quartzite as Ocbils, with their significant over-representation of clades from several ancient lineages consistent with Gondwanan Heritage Hypothesis predictions (Hopper et al., 2016; Silveira et al., 2016). In contrast, all the evidence suggests that forest assemblages on canga represent Yodfels.

Analysis of phylogenetic structure of communities has proved a valuable tool in exploring contemporary ecological interactions and in linking community ecology with biogeography and evolution (Vamosi et al., 2009) and with conservation and macroecology (Tucker et al., 2017). Any attempt at synthesis on the scale of the study reported here presents significant challenges in terms of data standardization and quality control, choice of appropriate analytical approaches and interpretation of results. Vamosi et al. (2009) provide a useful checklist for evaluating phylogenetic community structure analyses. We consider the broad taxonomic scope (all angiosperms), broad geographical coverage, attention to taxonomic consistency and data integrity to be particular strengths of our study. However, we acknowledge and explain limitations in certain other aspects which may be addressed in future studies.

Although beyond the scope of the present study, a useful next step would be to collate information on regional species pools, an approach that has proven effective in understanding the historical relationships among ecologically similar sites across a continental scale (e.g., DRYFLOR et al., 2016; Dexter et al., 2017). Our species pool (metacommunity) was the combined taxon list for all the sites included in our study. An alternative approach would be to use a published list of species reported from campo rupestre s.l. (Brazil Flora Group [BFG], 2015). This would result in a species pool >60% larger than that we analyzed, or up to five times larger if we allowed for the possibility of Atlantic Forest species forming part of campo rupestre assemblages. However, this larger species pool would present greater challenges for data management and analysis and would also greatly reduce the proportion of the species pool represented in any single site, with potential detrimental effects on the statistical power to detect phylogenetic structure. Kraft et al. (2007) concluded on the basis of simulation studies that local communities comprising 30–60% of the regional pool would offer greatest statistical power. Vamosi et al. (2009) showed that, in practice, few published studies fall within this range, with most having mean local richness less than 30% of the total regional species pool, and our study is no exception.

Our study provides insights for those concerned with the conservation and sustainable management of areas of campo rupestre, including those on canga which are currently the subject of exceptional pressure from mining interests. The clear floristic distinctions between the four habitat types considered are a pointer to the importance of adequate representation of each type (and sub-types thereof where they are recognized) in protected area networks and sustainable management plans. Our conclusion that the forest communities on canga are quite different in phylogenetic structure from other campo rupestre assemblages and should in fact be interpreted as Yodfels requires further exploration before it can be considered sufficiently robust to inform appropriate conservation measures, which differ greatly between Ocbils and Yodfels (Hopper et al., 2016). However, the possibility that these forests on canga may harbor diverse subsets of Atlantic Forest diversity could lead to the identification of win–win approaches, whereby the vegetation restoration projects which are a legal requirement following mine closure (Skirycz et al., 2014) could also represent opportunities to safeguard and even increase populations of some of the most threatened species of the Atlantic Forest.

To our knowledge, this study is one of the most inclusive to date to focus primarily on non-forest ecosystems. However, increasing adoption of a very broad definition of ‘Rupestrian Grasslands’ to include all high altitude rocky complexes in Brazil highlights the opportunity for an even more inclusive analysis including other rupestrian environments associated with highlands, such as the tepuis of Roraima and the cangas of Carajás in the Brazilian Amazon. Increased investment in floristic inventory in these areas over the past decade is resulting in the publication of site-based lists suitable for inclusion in such studies (Viana and Lombardi, 2007; Nadruz-Coelho et al., 2015; Barbosa-Silva et al., 2016).

As more comprehensive data on species distributions becomes better understood and accessible, covering other rupestrian environments on highlands in different parts of the continent, studies such as the present one will elucidate processes underlying biodiversity dynamics at ever-increasing spatial scales and ecological heterogeneity.

DZ co-designed and initiated this study, collated the plot data used in the analysis, contributed to integration and interpretation of the data, and contributed to the final manuscript. MM contributed to collation of the plot data used in the analysis, conducted and documented the bulk of the similarity and phylogenetic analyses, contributed to integration and interpretation of the results, and contributed to the final manuscript. TM conducted statistical analysis, contributed to integration, visualization and interpretation of the results, also contributing toward the final manuscript. EN co-designed this study, contributed to statistical analysis and to integration, and interpretation of the results, and played a lead role in drafting the final manuscript. All authors have approved the final version to be published and agree to be accountable for all aspects of the work, ensuring questions related to its accuracy and integrity are appropriately investigated and resolved.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.