2.1 Taxonomic Sampling

We included a sampling of 154 species of Pteridaceae occurring in Brazil (approximately 74%), in addition to 648 Pteridaceae species found in other regions and/or continents. All rbcL and atpA sequences were obtained through searches on GenBank (see Supporting Information S1). As an outgroup, we considered 17 species belonging to the families Aspleniaceae, Blechnaceae, Didymochlaenaceae, Dryopteridaceae, Lindsaeaceae, Nephrolepidaceae, Polypodiaceae, Saccolomataceae, Thelypteridaceae and Woodsiaceae, whose sequences were also obtained from GenBank (see Supporting Information S1).

2.2 Sequence Alignment

For each marker, sequences were aligned using MAFFT v. 7.505 (Katoh and Toh 2010) and subsequently reviewed manually. Regions with a large amount of missing data, primarily those located at the ends of each alignment, were excluded. Additionally, for each marker, a preliminary phylogenetic tree was generated (see parameters in the next section) using maximum likelihood (ML) in IQTree software (Nguyen et al. 2015; Trifinopoulos et al. 2016) to identify anomalous sequences, which were subsequently removed.

2.3 Phylogenetic Analyses

From the concatenated dataset, we generated an ML phylogeny in IQTree software. ModelFinder (Kalyaanamoorthy et al. 2017), implemented in IQTree, was used to select the most suitable substitution model for the data. For all phylogenetic analyses performed, the GTR+F+I+G4 model presented the best fit and was automatically selected by IQTree according to the Bayesian Information Criterion and Corrected Akaike Information Criterion. The other parameters used in the analysis were: bootstrap analysis ultrafast with 1000 alignments, maximum iterations = 1000, minimum correlation coefficient = 0.99 and single branch test with 1000 replicates; for IQTree search parameters, perturbation strength = 0.5 and stopping rule = 100.

2.4 Divergence Time Estimation

The calibration of the ML tree obtained in IQTree was performed using penalised likelihood in treePL (Smith and O'Meara 2012). Three fossil constraints extracted from the Fern Tree of Life (Nitta et al. 2022) were used in this analysis: Heinrichsia cheilanthoides Regalado, A.R. Schmidt, M. Krings & H. Schneid., with an estimated age of 100.5–113 Mya (Regalado et al. 2019), positioned at the crown Pteridaceae; Acrostichum intertrappeum Bonde & Kumaran, with 66–72.1 Mya (Bonde and Kumaran 2002), in the stem Acrostichum+Ceratopteris; and Acrostichum palaeoaureum Beauchamp, Lemoigne & Petrescu, with 23.03–27.82 Mya (García-Massinni et al. 2006), in the Acrostichum stem. The position of the fossils in the phylogeny is in agreement with Nitta et al. (2022). In addition, we incorporated a secondary calibration point at the root node of Pteridaceae. Age estimates for the family obtained from previous studies were used to fix the age of this node. We considered 110.8 Mya obtained by Schuettpelz and Pryer (2009), 184.5 Mya from Testo and Sundue (2016) and 227.9 Mya from Nitta et al. (2022). We chose to test these three estimates as they represent the minimum and maximum ages previously recorded for Pteridaceae. Finally, we applied a smoothing parameter of 0.000001 based on the smallest chi-squared value obtained from treePL.

2.5 Biogeographical Analyses

We explored the biogeographical history of Pteridaceae and estimated the variation of ancestral distribution across the three chronograms obtained in treePL. Due to computational limitations and the maximum number of taxa allowed in each analysis, subfamilies were pruned from each calibrated tree. Thus, a total of 15 analyses were performed for the 5 subfamilies from each of the three chronograms.

The current geographical distribution for each species (see Supporting Information S1) considered in the analyses was obtained from herbaria reviews (for species occurring in Brazil), online databases (Tropicos [https://www.tropicos.org/home], GBIF [https://www.gbif.org] and speciesLink [https://specieslink.net/search/]), and the literature. We considered seven areas: A: North and Central America, Mexico and the Antilles; B: South America (excluding G, H and I); C: Africa and Madagascar; D: Eurasia and Australasia; E: Amazon; F: Atlantic Forest; G: Dry Diagonal (Huiet et al. 2018; Pereira et al. 2021). Only a single discrete area was specified for Eurasia and Australasia, and another one for North and Central America, Mexico and the Antilles due to computational limitations.

We employed the BioGeoBEARS package (Biogeography with Bayesian Evolutionary Analysis in R Scripts; Matzke 2013) in RASP 4.0 (Yu et al. 2020). This package implements three commonly employed approaches: dispersal-extinction-cladogenesis (DEC; Ree and Smith 2008), dispersal-vicariance analysis (DIVA; Ronquist 1997) and bayesian inference for discrete areas (BayArea; Landis et al. 2013). Since BioGeoBEARS uses probability-based versions of DIVA and BAYAREA, they are referred to as DIVALIKE and BAYAREALIKE, respectively.

Analyses were conducted using these three methods, also considering the additional free parameter j (thus, DEC+j, DIVALIKE+j, BAYAREALIKE+j). The parameter j represents the additional speciation process from the founder effect, which occurs when a daughter lineage disperses to a new area (where the ancestor did not occur) (Matzke 2014). This is particularly relevant for ferns, which generally exhibit a high dispersal capacity due to their small and light spores (Barrington 1993; Wolf et al. 2001). From the six models employed, the most suitable for each subfamily tree was selected based on the Akaike Information Criterion (AIC). Using the selected model, distribution probabilities for ancestral nodes were inferred. The maximum number of coded areas was set to seven, as there are species in the analyses that occur in all seven areas.

3.1 Phylogenetic Relationships

The atpA alignment included 583 species and 1515 bp, while the rbcL alignment presented 813 taxa and 1309 bp. ML analyses for these gene regions resulted in congruent topologies (see Supporting Information S1), with good bootstrap support. These markers were subsequently concatenated, resulting in a matrix with 819 taxa and 2824 bp, comprising 1918 distinct patterns, 1293 parsimony-informative sites, 191 singleton sites and 1340 constant sites. The phylogeny based on the concatenated data, constructed using ML, is available in the Supporting Information S1.

Pteridaceae is monophyletic, as are the subfamilies Parkerioideae, Cryptogrammoideae, Pteridoideae and Vittarioideae. Subfamily Cheilanthoideae was not resolved as monophyletic, due to the phylogenetic positioning of Calciphilopteris ludens (Wall. ex Hook.) Yesilyurt & H. Schneid., which in our analysis appeared as the sister group to Vittarioideae, instead of being the sister group to the remainder of Cheilanthoideae. The relationships among the subfamilies and between most genera are well-resolved and well-supported. The majority of the genera are monophyletic, with a few exceptions (mainly Cheilanthoideae).

3.2 Divergence Time Estimates

Three chronograms (Figures 1-3) were obtained using fossil information and a secondary calibration point. The differences between these trees are due to the fixed age of the Pteridaceae root. In analysis 1, we fixed this node at 110.8 Mya; in the second analysis at 184.5 Mya; and in the third at 227.9 Mya. The secondary calibration of this node generates variations in the ages of the family, as well as the subfamilies and genera; see Table 1.

Overall, the period of greatest diversification for the Pteridaceae (with the highest number of cladogenetic events) encompasses the last 10 Mya. Regarding the species from the areas of endemism defined by Della and Prado (2024), it is observed that those delimiting southeastern Brazil and southeastern Bahia are quite recent compared to the taxa from southern Brazil and the Guiana Shield (see Table 1). The species from southeastern Brazil emerged in the last 8 Mya, and the only species sampled from southeastern Bahia, Adiantopsis alata Prantl, is at most 2 Mya. In contrast, the taxa from southern Brazil range between 47 and 10 Mya (except for Doryopteris rivalis Sehnem, which in the most recent estimate has 3.5 Mya and Adiantopsis dichotoma (Sw.) T. Moore with 6.5–13 Mya). For the Guiana Shield, it can be noted that Pterozonium Fée dates to approximately 55–27 Mya.

3.3 Ancestral Areas Estimation

From the chronograms of each subfamily (considering the three calibrations), we identified the models that best fit the data according to the AIC (Table 2). For Cheilanthoideae and Vittarioideae, in all three analyses, the best model was BAYAREALIKE+J; for Cryptogrammoideae, it was DEC; for Parkerioideae, it was BAYAREALIKE and DEC; and for Pteridoideae, it was DEC and BAYAREALIKE+J. The reconstructions of Cryptogrammoideae are not discussed here, since they do not include species that occur in South America.

Figures 4 and 5 present the biogeographic history of only a few clades of Pteridaceae, focusing primarily on the largest genera from each subfamily and the groups most relevant to the discussion of the hypotheses tested here. These figures were created using the oldest estimated age for the Pteridaceae root node, ensuring that the most recent estimates are also covered. Complete reconstructions for each subfamily are provided in the Supporting Information S1.

The biogeographical histories obtained through the reconstructions show us that species from the Dry Diagonal, Brazilian Atlantic Forest and Amazon Forest tend to originate in three periods: before 20 Mya, around 10 Mya and in the last 2 Mya. Vicariance events explain the origin of many of these species. However, dispersal events play an important role in the evolutionary history of all subfamilies.

The Dry Diagonal does not constitute a barrier to species dispersal, as it is common to find shared taxa with the Brazilian Atlantic Forest and (or) Amazon Forest. Species occurring in the Cerrado, Caatinga and Chaco are mostly recent, dating from the Late Miocene to the Pleistocene. Additionally, some taxa with disjunct distributions in the Amazon Forest and the Brazilian Atlantic Forest, primarily within the subfamily Vittarioideae, are ancient, originating between 28 and 62 Mya.

The Amazon region does not appear to be the primary source area for species of the Pteridaceae subfamilies. Table 3 shows the estimates of dispersal frequency and speciation in each area. This table does not show data on Cryptogrammoideae, as they do not occur in South America, and on Parkerioideae, as they have few species, which are widely distributed. The primary source areas are North and Central America, and Africa for Cheilanthoideae; South America (excluding the Amazon Forest, Brazilian Atlantic Forest and Dry Diagonal) and Eurasia and Australasia for Pteridoideae and Vittarioideae. The sink areas are South American vegetation (excluding the Amazon Forest and Brazilian Atlantic Forest) and the Dry Diagonal for Cheilanthoideae; North, Central and South America (excluding the Amazon Forest, Brazilian Atlantic Forest and Dry Diagonal) for Pteridoideae; and North and Central America, and the Brazilian Atlantic Forest for Vittarioideae.

4.1 Phylogenetic Relationships

The phylogeny we obtained is generally consistent with previously published studies of the Pteridaceae (Prado et al. 2007; Schuettpelz et al. 2007), its subfamilies (Zhang et al. 2005, 2017), and genera therein (Sánchez-Baracaldo 2004; Link-Pérez et al. 2011; Lu et al. 2011; Sigel et al. 2011; Chao et al. 2014; Grusz and Windham 2014; Cochran et al. 2014; Yesilyurt et al. 2015; Zhang et al. 2015; Schuettpelz et al. 2016; Regalado et al. 2017; Huiet et al. 2018). The position of Calciphilopteris ludens was perhaps most discordant relative to what are now understood to be its affinities. Although unusual, this Asian species has typically been considered to be part of the Cheilanthoideae. While most molecular phylogenetic analyses have revealed it to be sister to the remainder of that subfamily, support is often poor (e.g., Schuettpelz et al. 2007) and, in some cases, the species is instead resolved as sister to Vittarioideae (e.g., Pryer et al. 2016); as we find it here. Ultimately, this inconsistency reflects the distinct nature of Calciphilopteris and most likely stems from a deep and rapid divergence separating it from the (other) Cheilanthoideae and Vittarioideae. It is likely that additional sequencing (of both chloroplast and nuclear loci) will be necessary to find strong support for its phylogenetic placement.

4.2 Divergence Time Estimates

The selection and positioning of fossil constraints is a critical step in dating phylogenies, which can lead to markedly different age estimates (Sauquet et al. 2012). This can be seen in the different age estimates already published for Pteridaceae in the works of Schuettpelz and Pryer (2009), Testo and Sundue (2016) and Nitta et al. (2022). Additionally, the statistical methods used to obtain the chronogram must be considered, as discussed by Bauret et al. (2017). In the aforementioned studies, two used penalised likelihood (the first and third studies) and one employed an uncorrelated lognormal relaxed clock. Another relevant factor is the use of secondary calibrations, which can negatively influence results and lead to the propagation of errors if the original primary calibration includes fossils that are phylogenetically misplaced or assigned incorrect ages (Parham et al. 2012; Schenk 2016). Furthermore, using secondary calibration points tends to produce younger age estimates (Sauquet et al. 2012).

In a study with Pteris, Chao et al. (2014) estimated that the genus diverged from other Pteridoideae lineages around 47 million years ago. The methodology used was an uncorrelated lognormal relaxed model, with two secondary calibration points obtained from Schuettpelz and Pryer (2009). Therefore, our estimate, which used a fixed age for the root node of Pteridaceae at 110.8 Mya, obtained the closest age, 54.5 Mya. However, under the other analyses, we obtained ages of 89 and 116 Mya. This demonstrates that the secondary calibration point directly impacts the estimated age of the genus.

Two other examples, using Adiantum as a study model, can also illustrate this. Regalado et al. (2017) suggested that the crown group of Adiantum originated in the Palaeocene (~70 Mya), using three local clocks and secondary calibrations from Rothfels and Schuettpelz (2014). Our three analyses showed that the crown Adiantum ages were 68, 111 and 143 Mya. Lu et al. (2011) estimated the age of the A. pedatum complex as 4.27 Mya (2.2–6.5) using a lognormal relaxed clock model, fossils and secondary calibrations (according to Schuettpelz and Pryer (2009)). The result of our analysis 1, with secondary calibration from Schuettpelz and Pryer (2009), obtained similar results for the complex, 6 Mya, while analyses 2 and 3 obtained 10.5 and 13.5 Mya, respectively. Therefore, we prefer to consider the three chronograms to perform the reconstructions of the ancestral areas, even if this implies large intervals for maximum and minimum ages.

4.3 Origin of Pteridaceae Species Occurring in the Amazon Forest, Brazilian Atlantic Forest and Dry Diagonal

4.4 Was the Dry Diagonal a Barrier to Species Dispersal?

Based on our ancestral reconstructions within subfamilies, it can be concluded that the Dry Diagonal of South America did not constitute a barrier to the dispersal of Pteridaceae species. On the contrary, many taxa that occur in the Brazilian Atlantic Forest or Amazon Forest are also found in the Dry Diagonal. Examples include Adiantum subcordatum Sw., Lytoneuron ornithopus and Adiantum mathewsianum Hook., which are present in the Brazilian Atlantic Forest and Dry Diagonal; and Adiantopsis senae, Cheilanthes pohliana Mett. and Adiantum cinnamomeum Lellinger & J. Prado, which are found in the Amazon Forest and Dry Diagonal. Some of these species are relatively old, originating in the last 48 Mya. However, the vast majority appeared within the past 10 Mya or even within the last 1 Mya. These younger ages agree with previous studies that found the Cerrado flora was assembled from numerous plant lineages from different phytogeographic domains, mainly in the last 4–5 Mya (Simon et al. 2009; Rando et al. 2016; Antonelli et al. 2018). Furthermore, earlier work estimates that the biotic exchange from the Brazilian Atlantic Forest to the Dry Diagonal may have been a geologically recent event, occurring at least 5 Mya (Batalha-Filho et al. 2013; Thomé et al. 2016; Antonelli et al. 2018; Matos-Maraví et al. 2021).

Endemic taxa of the Dry Diagonal predominantly emerged in the last 10 Mya, with exceptions such as Adiantum patens, which is very recent, estimated to have originated in the last 400,000 years, and Adiantopsis flexuosa, which is quite ancient, with an age ranging from 18 to 38 Mya. This also aligns with previous studies that dated the formation of the Cerrado to around 10 Mya (Simon et al. 2009; Rando et al. 2016).

For some relatively ancient groups, the disjunct distributions between the Atlantic Forest and Amazon could be explained by the formation of the Dry Diagonal. It is estimated that periods of global cooling and increased aridity in the late Eocene and part of the early Oligocene led to the formation of the Dry Diagonal (Prado and Gibbs 1993; Simon et al. 2009; Pinheiro and Monteiro 2010; Sobral-Souza et al. 2015; Thode et al. 2019; Bacci et al. 2022). Therefore, species such as Hecistopteris pumila, with an origin estimated between 29 and 62 Mya, and Polytaenium citrifolium, with an age of 28–59 Mya (see Figure 5), probably arose before the separation of the Amazon and Atlantic Forest. The establishment of the Caatinga, Cerrado and Chaco therefore led to the formation of these disjunct populations.

Species that occur in the Brazilian Atlantic Forest, as well as in the Dry Diagonal and Amazon Forest, are mostly very young, having appeared in the last 2 Mya (such as Adiantum latifolium L. and Adiantum terminatum). However, some taxa arose around 10 Mya (e.g., Vittaria lineata and Adiantum delicatulum), and a few that are older than 20 Mya (e.g., Ananthacorus angustifolius and Adiantum deflectens). Some of these species are epiphytic, but the majority are terrestrial and found in moist areas. In the Dry Diagonal, they occur mainly in riparian forests (Xavier 2007; Xavier et al. 2012; Prado et al. 2020).

4.5 Was the Amazon a Source Area for Pteridaceae Species?

The Amazon was not recovered as the primary source for Pteridaceae species when considering the biogeographic history of their subfamilies in a global context. North, Central and South America (excluding the Amazon Forest, Brazilian Atlantic Forest and Dry Diagonal), as well as Eurasia and Australasia, are the main source areas for these subfamilies. The sink areas are the vegetation of South America (excluding the Brazilian Atlantic Forest and Amazon Forest), as well as North and Central America (see Table 3 and Supporting Information S1).

However, in the context of South America, the Brazilian Atlantic Forest (for Cheilanthoideae) and the Amazon Forest (for Vittarioideae) are areas where many species originate and from which many lineages disperse. On the other hand, the Dry Diagonal acts as a sink area for Cheilanthoideae, Pteridoideae and Vittarioideae. Antonelli et al. (2018) identified the Amazon as the most important primary source of Neotropical for several groups of organisms, including ferns in general. The second main source area for ferns was the Brazilian Atlantic Forest (Antonelli et al. 2018). Gentry (1982) had already suggested that the Brazilian Atlantic Forest was a source area of species for various Gondwanan vegetation. For ferns, the Amazon Forest is considered an area poor in species, mainly due to the homogeneous relief (Tryon and Conant 1975; Tryon 1986). In contrast, the Brazilian Atlantic Forest hosts a large number of species, many of which are endemic (Prado et al. 2020; Della and Prado 2024). Antonelli et al. (2018) also found that dispersals from forested/wet environments to open/dry habitats are relatively common. This would explain the recent origin of many endemic taxa in the Dry Diagonal and the high number of species that co-occur in this region as well as in the Brazilian Atlantic Forest and (or) the Amazon Forest.

4.6 Areas of Endemism of Pteridaceae in Brazil

The areas of endemism of Pteridaceae in Brazil have relatively distinct biogeographical histories. Regarding the Guiana Shield, although none of the taxa listed by Della and Prado (2024) were sampled in our phylogeny, there are two species belonging to the genus Pterozonium (Pterozonium brevifrons (A.C.Sm.) Lellinger and Pterozonium reniforme (Mart.) Fée) that are endemic to the Pantepuis. These two species diverged between 12 and 25.5 Mya; however, the origin of the genus is dated to 27–55 Mya. Thus, although the Pantepuis are formed by very ancient rocks from the Precambrian, the genus Pterozonium is seemingly more recent (see Table 1). The age estimates obtained here for Pterozonium are older than those found for Stegolepis (Rapateaceae), Brocchinia and Lindmannia (both Bromeliaceae) (Givnish et al. 2000, 2004, 2007; Rull and Vegas-Vilarrúbia 2020). Stegolepis would have occupied the Pantepuis around 12 Mya through dispersal and began local diversification around 6 Mya. Brocchinia species would have emerged between 12 and 17 Mya, while Lindmannia diverged in the last 2.5 Mya (early Pleistocene). Based on our results from ancestral area reconstructions, dispersals from the Andes to the Pantepuis (between 27 and 55 Mya) likely occurred, followed by habitat specialisation.

Southern Brazil encompasses part of the Serra Geral (Della and Prado 2024), which dates back approximately 133 Mya (Cretaceous) (Reis 2013). The endemic taxa in this area are also relatively old, with estimated ages ranging from 47 to 10 Mya (except Doryopteris rivalis, which in the most recent estimate has 3.5 Mya, and Adiantopsis dichotoma with 6.5–13 Mya—see Table 1). Based on reconstructions, it is estimated that all sampled species from this area originated from vicariance events. Since paleoclimatic models predicted the existence of forest refugia with relative climatic stability along the Serra Geral (Carnaval et al. 2014), these lineages may have diversified and specialised due to the microclimatic conditions present in the mountain's altitude gradients. More recent taxa, such as Doryopteris rivalis, may have originated from glacial cycles, which led to the extinction of subtropical forests and the expansion of vegetation from arid and semi-arid areas (Arana et al. 2013).

The only sampled species from southeastern Bahia, Adiantopsis alata, is about 1–2 Mya and likely originated from vicariance. This region has been interpreted as a forest refuge (‘Bahia refuge’) because it showed climatic stability during the last glacial maximum (Mori et al. 1983; Carnaval and Moritz 2008; Carnaval et al. 2009). Thus, this area likely retained lineages that diversified (in situ speciation) later due to habitat specialisation (Staggemeier et al. 2015; Machado et al. 2018).

The taxa from southeastern Brazil are also very recent, with most having originated in the last 2 Mya (except for Lytoneuron itatiaiense (Fée) Yesilyurt and Pteris congesta J. Prado, which, according to older estimates, may be 8 and 6 Mya, respectively—see Table 1). This area includes Brazil's coastal mountains, the Serra do Mar and Serra da Mantiqueira, which emerged about 65 Mya during the transition from the Cretaceous to the Paleogene (Almeida and Carneiro 1998; Vieira and Gramani 2015). Most species in this area arose through vicariance events, which may have resulted from specialisation to the heterogeneous microclimatic and soil conditions along the altitude gradients of these mountains (Kluge and Kessler 2006; Rahbek et al. 2019; Vasconcelos et al. 2020). An alternative explanation would be the diversification of these lineages due to Pleistocene glacial cycles, which may have led to repeated vicariance events (Haffer 1969). Jamesonia brasiliensis and J. cheilanthoides likely originated through long-distance dispersal (in the last 1 Mya) from the Andes and (or) Central America.