We included representatives of the main Neotropical subgroups of Habenaria, as defined by phylogenetic studies (Batista et al. 2013), species addressed in subsequent studies (Batista et al. 2014, 2016, 2017, 2021, 2022b, 2023; Pedron et al. 2014; Lau et al. 2021; Cruz-Lustre et al. 2022), as well as species sequenced for this study. For the outgroup, we used the same dataset as that used by Batista et al. (2022b). This dataset consists of a selected sample based on the studies of Jin et al. (2017) and Ngugi et al. (2020) and comprises representatives of the major clades of Asian and African Habenaria and related genera. The final dataset comprised 119 terminals from 116 taxa. Of these, 99 were from Neotropical Habenaria (96 species), while the remaining 20 were Old World Habenaria and related genera. Based on the results of Ngugi et al. (2020), we used Stenoglottis longifolia Hook.f., from subtribe Orchidinae, to root all trees. Geographical origin, voucher information, and GenBank accession numbers are provided in Supplementary material 1.

The following markers were used: rDNA ITS, matK-trnK, rps16-trnK (Batista et al. 2013; Pedron et al. 2014), and rDNA ETS (Monteiro et al. 2010; Lau et al. 2021). DNA was extracted from fresh or silica gel-dried samples, according to Doyle and Doyle (1987). The samples were amplified using the same primers and conditions as described by Monteiro et al. (2010), Batista et al. (2013), and Pedron et al. (2014). The PCR products were purified by precipitation using 125 mM EDTA and ethanol, and then sequenced in both directions using the same primers used for amplification at Macrogen Inc (Korea). Sequence reads were assembled using the Staden software package (Staden 1996), aligned using MAFFT v.7 (Katoh and Standley 2013) as implemented in the EMBL-EBI bioinformatics web tool (Madeira et al. 2019), and manually adjusted using MEGA v.7.0 (Kumar et al. 2016).

DNA sequence alignments were analysed using parsimony, maximum likelihood (ML), and Bayesian inference. To assess incongruences between datasets, searches were performed using both individual and concatenated matrices. Concatenation was performed using the same individuals sequenced for the different markers in most cases. We performed maximum parsimony (MP) phylogenetic analyses using PAUP* v.4 (Swofford 2002), applying Fitch parsimony (equal weights and unordered characters; Fitch 1971) as the optimality criterion. Each search consisted of 2,000 replicates of random taxon additions, with branch swapping performed using the Tree Bisection and Reconnection (TBR) algorithm, saving ≤ 10 trees per replicate to prevent extensive swapping on suboptimal islands. We assessed internal support using character bootstrapping (Felsenstein 1985), with 2,000 replicates, simple addition, and TBR branch swapping, and saving ≤ 15 trees per replicate. We considered bootstrap percentages (BP) of 50–70% as weak, 71–85% as moderate, and > 85% as strong (Kress et al. 2002).

We performed the ML analyses using the program RAxML v.8.2.12 (Stamatakis 2014), as implemented in the Cyberinfrastructure for Phylogenetic Research (CIPRES) Portal 2.0 (Miller et al. 2010). Following the analysis of 1,000 rapid bootstrap replicates, a search was conducted for the tree that maximised the likelihood function. This search used the default value of 25 rate categories and estimated all free model parameters for five character partitions: matK, trnK intron, ITS, ETS, and rps16-trnK intergenic spacer. The bootstrap and ML tree searches both used the GTRGAMMA nucleotide model.

We performed Bayesian analyses using MrBayes v.3.2.7a (Ronquist et al. 2012), as implemented in the Cyberinfrastructure for Phylogenetic Research (CIPRES) Portal 2.0 (Miller et al. 2010). In these analyses, we treated each DNA region as a distinct partition. We selected an evolutionary model for each DNA region using hierarchical likelihood ratio tests (hLRTs) in MrModeltest 2 (Nylander 2004). We used the “unlink” command to estimate model parameters separately for each partition. Each analysis comprised two independent runs, each with four chains for 5,000,000 generations, sampling one tree every 1,000 generations and a temperature parameter of 0.2. Convergence between the runs was assessed using the average standard deviation of split frequencies (< 0.01) and the potential scale reduction factor (= 1.0), and was achieved after 991,000 generations. Log files were also analysed using Tracer v.1.7.2 (Rambaut et al. 2018) to assess convergence between the two independent runs. The first 1,500 trees (30%) were discarded as the burn-in, after which the remaining trees were used to assess topology and posterior probabilities (PP) in a majority-rule consensus. We considered groups with PP > 0.95 as strongly supported, groups with PP ranging from 0.90–0.95 as moderately supported, and groups with PP < 0.90 as weakly supported. The tree was edited using FigTree v.1.3.1 (Rambaut 2009).

We examined fresh, liquid preserved, and herbarium specimens. For the gynostemium, we used a stereoscopic microscope and measured it with a digital calliper. Data on flowering time, habitat, and geographical distribution were based on field observations and herbarium specimens. The character states and measurements of species in the H. nuda complex were supplemented with data from Batista et al. (2008, 2011a, 2011b, 2012, 2021, 2023) and examination of specimens from the BHCB and CEN herbaria (acronyms according to Thiers 2025). Morphological terminology is based on Stearn (1992) and Simpson (2006).

The survey was based on our own collections and herbarium records for the region. We consulted the three main Brazilian botanical record databases: speciesLink (2025), REFLORA - Virtual Herbarium (2025), and JABOT (2025) to find specimens collected in the Grão Mogol region; these included samples from the herbaria BHCB, CEN, HB, HUEFS, MBM, RB, SP, SPF, UB, and UEC (acronyms according to Thiers 2025). All records were examined in loco or via images and were taxonomically revised. The taxonomic status and circumscription of the species follow a synopsis of the Neotropical species of Habenaria (Batista et al. 2011a, 2011b).

The extent of occurrence (EOO) and area of occupancy (AOO) were estimated using a beta version of GeoCAT (Bachman et al. 2011) as implemented on the IUCN website (https://geocat.iucnredlist.org/), with the default setting of 2 km cell width. Preliminary conservation status was inferred using the IUCN Red List categories and criteria (IUCN 2012) and the guidelines for using the IUCN Red List categories and criteria (IUCN 2016). The distribution map was prepared using the free software QGIS v.3.10 (QGIS Development Team 2019).

The concatenated ITS, ETS, matK-trnK, and rps16-trnK DNA matrix consisted of 4,405 aligned characters, 869 (19.7%) of which were parsimony-informative. The Supplementary material 2 shows the general features of the datasets and parsimony statistics, together with a summary of the implemented models for each partition. As with previous molecular phylogenetic analyses involving similar datasets (Pedron et al. 2014; Lau et al. 2021; Batista et al. 2022b, Cruz-Lustre et al. 2022), no significant incongruence was detected between the nuclear and plastid datasets. Therefore, only the results of the combined matrix are presented and discussed. The strict consensus tree from the parsimony analysis was largely congruent with both the ML best tree and the Bayesian majority-rule consensus tree. However, since the latter was more fully resolved and had stronger overall support, it was selected for presentation and discussion (Fig. 1). For comparison, the bootstrap percentages from the MP and ML analyses are shown on the Bayesian tree.

The recovered relationships were similar to those of previous molecular phylogenetic studies (Batista et al. 2013, 2014, 2016, 2017, 2022b, 2023; Pedron et al. 2014; Lau et al. 2021; Cruz-Lustre et al. 2022). The New World Habenaria species formed a well-supported monophyletic group (1.00 PP, 100% BP-ML, 79% BP-P) and several well-supported subclades were recovered within this clade and numbered accordingly (subclades 1 to 20) (Fig. 1). As sequences for the ETS, the complete matK gene and the rps16-trnK intergenic spacer were unavailable for species in subclade 9 according to Batista et al. (2013), composed mainly of Mexican species, these were excluded from the analyses.

The three samples of the unidentified taxon from Grão Mogol were recovered within a large clade (1.00 PP, 99% BP-ML, 96% BP-P) named Cerrado clades by Batista et al. (2013). The three sequenced terminals formed a clade with high support (1.00 PP, 100% BP-ML, 100% BP-P), which was sister to the clade formed by Habenaria rolfeana Schltr. and Habenaria setacea Lindl. (1.00 PP, 100% BP-ML, 97% BP-P), but with low support (0.78 PP, 56% BP-ML, 53% BP-P). The species of the H. nuda complex are distributed across three subclades within the Cerrado clades.

We found 29 specimens encompassing 12 species of Habenaria from Grão Mogol (Fig. 2; Table 1). One specimen had poorly preserved flowers and could not be identified, while another belonging to the H. nuda complex could not be identified with confidence. Three of the four species reported for Grão Mogol by Barros and Pinheiro (2004) were misidentified: Habenaria aff. ayangannensis Renz is referable to Habenaria espinhacensis J.A.N.Bat. & A.A.Vale (Kameyama et al. CFCR 9015 – SP, SPF; Pansarin & Simões 785 – UEC), Habenaria inconspicua Cogn. is referable to Habenaria parviflora (Silva et al. CFCR 12525 – SPF), and Habenaria repens var. gracilis Lüderw. & Hoehne is Habenaria subrepens J.A.N.Bat., B.L.Lau & Machado-Costa (Hensold et al. CFCR 3457 – SP, SPF). With eight records, H. espinhacensis seems to be the most common species in the region. Although the type of the species originates from Grão Mogol, its distribution extends both north and south of the EMR (Batista et al. 2016). Nine species (75%) are known from only one or two records. Most of these species are found in rupestrian grasslands on dry or seasonally wet soils. Only Habenaria gourlieana Gillies ex Lindl. and H. subrepens are found in wet or flooded areas for most of the year.

Our phylogenetic analysis revealed patterns within the Neotropical Habenaria species that were comparable to those identified by Batista et al. (2013) and Vale (2014), but with better resolution and support. As mentioned by Batista et al. (2021), the H. nuda complex is not monophyletic; however, all its species belong to the clade named Cerrado clades. This clade unites species that were originally classified into five sections: sect. Macroceratitae Kraenzl., sect. Microdactylae Kraenzl., sect. Nudae, sect. Pentadactylae Kraenzl., and sect. Spathaceae Kraenzl. (Kränzlin 1892, 1901; Cogniaux 1893). Of these sections, H. adamantina fits better in sect. Nudae, due to its morphology and phylogenetic relationship with other species of the section. However, both a molecular phylogenetic analysis of Neotropical Habenaria (Batista et al. 2013) and the morphological and phylogenetic analyses of sect. Pentadactylae (Pedron et al. 2014), sect. Microdactylae (Batista et al. 2018), and sect. Micranthae Kraenzl. (Cruz-Lustre et al. 2022) have shown that the sections defined by Kränzlin (1892, 1901) and Cogniaux (1893) comprise non-monophyletic and morphologically heterogeneous groups, and require extensive revision.

The position of H. adamantina within the Cerrado clades was expected, given its similar morphology, habitat, and geographical distribution to that of most species in the clade. With a few exceptions, the morphology of these species is characterised by slender plants with linear leaves that are commonly adpressed to the stem. The Cerrado clades includes species that are mainly distributed in rupestrian grasslands and the Cerrado domain in central and south-eastern Brazil (Batista et al. 2013).

Habenaria adamantina was recovered as a sister species to the clade comprising H. rolfeana and H. setacea, although with low statistical support. These two species are morphologically very similar to each other (Batista et al. 2011b) and clearly distinct from the new species due to their strongly arched ovaries (vs straight ovary in H. adamantina), flowers facing downwards and perpendicular to the rachis (vs facing forwards and appressed to the rachis), and longer lateral segments of petals and lip (vs shorter segments). While all three species occur in the EMR in Minas Gerais, H. rolfeana has the widest distribution, occurring in other parts of the state, as well as in Rio de Janeiro, São Paulo, and Paraná (Batista et al. 2011b). Both H. setacea and H. adamantina are endemic to the EMR in Minas Gerais, with the former restricted to the southern part of the EMR, while the latter is restricted to Grão Mogol, in the north. There are no records of either H. rolfeana or H. setacea for Grão Mogol so far.

Among the species in the H. nuda complex, H. adamantina is more morphologically similar to H. australis. However, in H. adamantina, the anterior petal segment is inserted at the base of the posterior segment, whereas in H. australis it is inserted 3.0–3.5 mm above the base. The two species also differ in their distribution: H. adamantina is restricted to campo rupestre in the northern EMR of Minas Gerais, while H. australis occurs in the Serra Gaúcha in the Pampa domain (Batista et al. 2012). Habenaria adamantina can be distinguished from other species in the H. nuda complex by the following combination of characteristics: patent lateral sepals; petal anterior segment and lip lateral segments shorter than the petal posterior segment and lip median segment, respectively; and small auricles located below the rostellum arms and anther canals.

In general, species tend to be morphologically similar within the clades identified in phylogenetic analyses of Neotropical Habenaria (Batista et al. 2013; Pedron et al. 2014; Batista et al. 2018; Cruz-Lustre et al. 2022); however, there are several notable exceptions. In some cases, such as sect. Microstylinae, floral morphology is quite variable within the clade (Batista et al. 2013). In other cases, one species has a different floral morphology to the rest of the clade. Examples include H. montevidensis Spreng. in sect. Pentadactylae (Pedron et al. 2014) and H. karstica J.A.N.Bat. in sect. Spathaceae (Batista et al. 2022b). A further notable example is H. caldensis and H. magniscutata Catling, which are sister species with significantly different morphologies (Batista et al. 2013). In the case of H. montevidensis, Pedron et al. (2014) suggested that the morphological differences represent a shift in pollinators from nocturnal moths to diurnal butterflies. Pollinator shifts are also likely responsible for the aforementioned differences, given that species within each clade generally exhibit similar vegetative morphologies and are biogeographically related.

Conversely, there are notable cases of convergence. For example, the H. repens Nutt. complex comprises species with very similar floral morphology, despite belonging to phylogenetically distant clades (Lau et al. 2021). For the H. nuda complex, the currently available results indicate that the complex is polyphyletic (Batista et al. 2013, 2021) and based on superficial morphological characteristics that have little or no phylogenetic significance. For instance, several species along the Neotropical clade have elongated lateral segments of petals and labellum (Batista et al. 2021), whereas linear leaves adpressed to the stem are found in most species and clades of the Cerrado clades (Batista et al. 2013). Further inferences for the H. nuda complex are beyond the scope of this study and are currently limited by the fact that a significant proportion of the group’s diversity, in the form of various morphotypes, has not yet been adequately analysed or incorporated into phylogenetic and comparative morphological studies.

These discrepancies and species complexes among Neotropical Habenaria pose a taxonomic challenge but also represent a unique opportunity to study the evolution and diversification of the group. These studies should invariably take an integrative approach and the use of phylogenomic data combined with other tools is a promising alternative.

Campo rupestre constitutes a highly diverse environment in Brazil, with around 15% of the country’s flora concentrated in an area covering less than 1 per cent of its territory (Silveira et al. 2016). However, this diversity is not evenly distributed throughout this vegetation. Analyses of the endemism of the vascular flora of the rupestrian grasslands of the EMR (e.g. Echternacht et al. 2011; Colli-Silva et al. 2019) have identified areas characterised by endemic species, supporting similar proposals for regionalisation. Similarly, comparing orchid floras from different areas of rupestrian grasslands (Azevedo and van den Berg 2007; Vieira et al. 2014) revealed low similarity indices, suggesting floristic heterogeneity and a high frequency of micro-endemic species.

The Grão Mogol region, located in the northern part of the EMR in the state of Minas Gerais, has a unique flora (Pirani et al. 2003; Colli-Silva et al. 2019). It has a different species composition to other regions of this mountain system; for example, it shares few species with areas in the central portion, such as the Serra do Cipó, and is very different from the southern portion (Echternacht et al. 2011; Vieira et al. 2014). Differentiation of lineages in the region has also been observed at infraspecific levels for other groups (e.g. Vriesea Lindl.; Dantas-Queiroz et al. 2021). The location of Grão Mogol in an area of transition between the Cerrado and the Caatinga domains may be a key factor in the region’s uniqueness (Pirani et al. 2003; Dantas-Queiroz et al. 2021), given the environmental discontinuity (Echternacht et al. 2011).

The diversity of Habenaria in Grão Mogol is comparable to that in Pico das Almas in the state of Bahia (12 species), but evidently lower than in Chapada dos Veadeiros in the state of Goiás (59 species) and Serra do Cipó (52 species; see Batista et al. [2023], for a comparison). Of the 12 species recorded so far in Grão Mogol, only H. adamantina is apparently endemic. This is consistent with the findings of Batista et al. (2023) regarding the reduced endemism of the genus at the regional level in Brazil. This differs from the situation with Eriocaulaceae, for example, which has high endemism in the central portion of the EMR in Minas Gerais (Echternacht et al. 2011), and the genus Mimosa L., where 48% of the 189 species recorded for the Cerrado domain are narrow endemics (Simon and Proença 2000).

The current known distribution of H. adamantina, only found in Grão Mogol, suggests at least two possibilities: (1) the species is truly endemic to this region due to environmental conditions or evolutionary processes, or (2) the distribution is the result of sampling bias. In any case, the new species is rare and scarce, as it has not been recorded elsewhere.

The new records from the northern region of Minas Gerais demonstrate the increasing diversity of one of the EMR’s most taxonomically underestimated areas. Historically, this region has attracted less attention from collectors than other areas of the EMR, particularly with regard to certain plant species (Rapini et al. 2002; Pirani et al. 2003; Batista et al. 2025). However, momentum has built over the last two decades in documenting the flora of Grão Mogol, revealing new species as exploration has expanded. The existence of a state park in Grão Mogol enables the conservation of its unique biodiversity. Further exploration and taxonomic work in the campo rupestre is necessary, particularly in seasonal and morphologically inconspicuous groups. A more complete knowledge of the flora will refine analyses of diversity, distribution and species evolution patterns in these ecosystems.