Taxon sampling for molecular analyses

We reused the sequences from the 58 samples published in Christenhusz & al. (2008) and additionally sequenced 243 new samples specifically for this study, thereby obtaining a dataset of 301 samples of Danaea. To increase the resolution at the species level, we also sequenced one new locus, the highly variable rpl32-trnL intergenic spacer (Shaw & al. 2007), for most of the specimens. Multiple individuals across the geographical range of each species were sampled when possible (Fig. 1). We were able to obtain genetic data for 68 of the 81 taxa we recognize as well as for several specimens that we were unable to assign to any described species. Our outgroup sampling includes 13 species, with at least one representative of each of the other five genera of Marattiaceae. Appendix 1 lists all species with DNA material and includes voucher information and GenBank numbers for each accession.

DNA isolation

DNA was extracted from silica gel dried material with the NucleoSpin Plant II kit (Macherey-Nagel, Germany). Polymerase chain reaction (PCR) amplifications were executed using PuReTaq Ready-To-Go (PCR) Beads (GE Healthcare UK Limited). PCR was used to amplify the plastid genes rbcL and atpB, and the non-coding rpl32-trnL and trnL-F regions. The PCR reactions contained approximately 32 µl of solution composed by 25 µl of ddH₂O, 1 µl of each primer and 5 µl of the extraction template. Purification and sequencing of the PCR products was done by Macrogen Inc. (Seoul, South Korea/Amsterdam, the Netherlands). PCR and sequencing primers and protocols are detailed in Table 1.

Phylogenetic analyses

Full dataset of 301 ingroup samples representing 68 named taxa of Danaea and one sample for each of 13 outgroup species from the five other genera in the family Marattiaceae were used in the phylogenetic analyses. The four loci were aligned separately with MAFFT (Katoh & Standley 2013) on the EMBL-EBI-server (Madeira & al. 2022) with default settings. In addition, the two loci containing gaps (rpl32-trnL and trnL-F) were aligned with MUSCLE (Edgar 2004) and T-COFFEE (Notredame & al. 2000) on the EMBL-EBI-server with default settings to compare the sensitivity of the results to different alignments. All sequences are from the chloroplast genome and therefore share the same phylogenetic history, so we concatenated them into a single run with SequenceMatrix (Vaidya & al. 2011). Variation in sequence lengths appeared phylogenetically informative, so we used simple gap coding (Simmons & Ochoterena 2000) to code gaps as binary characters with the software FastGaps 1.2 (Borchsenius 2009).

The sequence data was partitioned by the 1^st, 2^nd, and 3^rd positions of the genes and by non-coding regions, and this partition was run with the advanced greedy algorithm in ModelFinder (Kalyaanamoorthy & al. 2017) in IQ-TREE 1.6.12 (Nguyen & al. 2015). The rate of evolution at different loci was estimated with TIGER-rates (Frandsen & al. 2015), and the acquired rate file was fed into RatePartitions (Rota & al. 2018). We tested division factor values in RatePartitions ranging from 1.5 to 3.5, which resulted in differently partitioned alignments. These were fed into ModelFinder and run with the advanced greedy algorithm. The models with the best BIC values were the greedy algorithms partitioned with RatePartitions division factor 3.0. The models with the worst BIC values were the ones partitioned by genes, introns, and codon positions (Appendix 2). The gap-data for each alignment was also analysed with ModelFinder and used in subsequent analysis. The sequence and gap data were combined for the phylogenetic analysis with RAxML-NG 1.1.0 (Kozlov & al. 2019) and the models with the best BIC values were used. RAxML-NG was run with default settings, e.g. 20 maximum likelihood inferences of 10 parsimony and 10 random trees, scaled branch lengths, and the number of bootstrap replications decided by a bootstopping test (Pattengale & al. 2010). Branch supports were assessed with Felsenstein's bootstrap. Tree visualization was done using the packages ape (Paradis & Schliep 2019), ggplot2 (Wickham 2016), ggtree (Yu & al. 2017), and treeio (Wang & al. 2020) in the R environment (R Core Team 2022). The final data matrices and the resulting trees are available in TreeBASE (study number 30768). Since the trees made with the different alignments were almost identical and congruent in all well supported clades, we present and discuss here only the tree with the dataset aligned with MAFFT.

Table 1.

Primers used for amplification and sequencing, PCR conditions, and appropriate references.

Taxonomic work and identification key

Our species delimitation is based on morphological and genetic discontinuities between species, and the aim was to obtain biologically meaningful species that are applicable in and useful for ecological and evolutionary studies. To identify morphological groupings and discontinuities, we have extensively compared actual herbarium specimens (made possible by loans to TUR) and digital images of specimens (available online, provided by herbaria on request or photographed by us during herbarium visits). We applied species names by comparing our material with the original type specimens.

The taxonomic work was based on a total of 3362 herbarium specimens of Danaea collected throughout the distribution of the genus (Fig. 1). Herbarium specimens were examined from A, AAU, ASU, BM, BRIT, C, CAY, CHRB, COAH, COL, CR, E, F, FLAS, G, GOET, H, HUA, HUTI, INB, INPA, K, L, LPB, M, MICH, MO, MSC, NO, NY, P, PH, PI, PMA, PRC, QCA, S, SJ, SP, STU, TUB, TUR, U, UC, US, UTCEC, VT, W, WIS, WTU, WVA, YU, and Z. In addition, many of the species have been observed, collected and photographed by us in the field.

The identification key was built in Lucid Builder 4.0 ( https://www.lucidcentral.org/, Queensland, Australia), where features are recorded on a table with a column for each taxon and a row for each feature (for quantitative traits) or state of a feature (for qualitative traits). The quantitative traits are represented by minimum and maximum values but can also have separate extreme values. We built the key using all the available material, although some specimens were considered so aberrant that they were not included in the measurements. Otherwise, the atypical specimens were labelled as extreme values in the key.

Although our species delimitations in many cases draw on field experience, we do not have characteristics of fresh material for all species, so they are not used in the key. We also chose to only use characteristics visible to the naked eye, with an emphasis on characteristics that can be seen in herbarium specimens. To keep the key manageable, we focused on the characteristics of adult plants. Juveniles would need to be treated as separate entities from the adults in the key and we do not have observations on juveniles for all species. All species are illustrated with pictures of herbarium specimens and many also with photos of live material from the field.

For species with several fertile specimens available, we used size measurements from fertile specimens only. For species with few or no fertile specimens, we used size measurements from specimens judged to be adult. Consequently, there is some uncertainty in the upper limits of the measurements, as we cannot be sure about the maximum size that the species can attain. Measurements from the earlier species descriptions were verified and supplemented with measurements of available specimens.

For the quantitative measurements, we applied a correction of measurements based on collection intensity. Measurements for species with 20 or more measured specimens were not altered, but we added and subtracted 5% of the minimum and maximum measurements for species with 10–19 measured specimens, and 10% for species with under 10 measured specimens. For elevation, we added and subtracted 100 m for species with 10–19 seen specimens, and 200 m for species with under 10 seen specimens. These were added as extreme values.

Fig. 1.

Distribution in tropical America of Danaea specimens used in this study. – A: density of all available material is shown as a count of unique collection numbers per 2° × 2° grid cell, and DNA specimens used in phylogeny are shown as dots; B: number of recognized Danaea species per 2° × 2° grid cell.

Because any given geographical area has only a limited number of Danaea species, we added geographical region as a separate character in the key. One of the species, D. excurrens, was given two entries in the key, as it contains two forms that are morphologically too different to be keyed out together.

Phylogeny and species delimitation

In our analyses, Danaea formed a clade consisting of three well supported subclades (Fig. 2). These correspond to the three subgenera as outlined in Christenhusz (2010): D. subg. Arthrodanaea C. Presl (Fig. 3), D. subg. Danaea (Fig. 4) and D. subg. Holodanaea C. Presl (Fig. 5). Each subgenus consisted of several well-supported subclades, many of which could be referred to a known species. However, many others could not, and those well-supported clades that had adequate material were described as new species (Keskiniva & Tuomisto 2024). In the end, we decided to assign the 301 sequenced samples to 68 recognized taxa and to leave 18 samples unidentified for the time being. The latter presumably represent several undescribed species, but we deemed the material insufficient to describe them at the present time.

The DNA sequences had 7 % missing data, with most data missing for the least informative locus (atpB: 12 %) and the least missing data for the two most informative loci (rpl32-trnL: 7 %, trnL-F: 1 %). Especially the newly sequenced locus rpl32-trnL was more variable than the other loci and helped to improve the resolution of the phylogeny, although a few problematic groups still remain.

Below we describe the phylogeny in more detail and discuss its contribution to our taxonomic decisions especially in the most difficult species complexes. We also provide a full list of accepted species and their synonyms by subgenus.

Subgeneric classification of Danaea

I. Danaea subg. Arthrodanaea C. Presl (Presl 1845) Species:

We were able to include 15 species of Danaea subg. Arthrodanaea in the phylogeny, only lacking DNA for two species (D. alansmithii and D. dilatata). Out of the three subgenera, Arthrodanaea (Fig. 3) has the smallest number of species and generally also the least amount of genetic variation. There are some well supported clades, however: D. simplicifolia + D. ushana Christenh. (BP = 100); D. antillensis + D. trinitatensis + D. trifoliata (BP = 100); and D. bipinnata + D. ulei Christ (BP = 90). Six of the species in the phylogeny form clades with a bootstrap support > 80: D. antillensis (BP = 100), D. danaëpinna (BP = 82), D. leprieurii (BP = 86), D. opaca (BP = 87), D. trifoliata (BP = 87), and D. trinitatensis (BP = 84). Two species are represented by only one specimen each, so more material is needed to test their genetic coherence (D. lingua-cervina and D. zamiopsis).

Fig. 2.

Simplified cladogram of Marattiaceae. Taxa of uncertain identity and duplicate specimens of each species pruned using drop.tip-function in ape package (Paradis & Schliep 2019) in R environment (R Core Team 2022).

The biggest problem that remains to be solved is centred around Danaea arbuscula. We have opted for a rather broad circumscription of this species, which has led to it becoming paraphyletic: D. geniculata, D. danaëpinna, D. polymorpha and D. zamiopsis are embedded in different parts of the same clade. In the case of D. danaëpinna, it is both monophyletic and morphologically distinct, while D. polymorpha and D. zamiopsis are morphologically identifiable, but the small number of DNA samples does not provide a proper test of their genetic distinctness. As to D. geniculata, the material from the Brazilian Atlantic coast (where the type is from) is both genetically and morphologically distinct from the type of D. arbuscula from northern Peru, but the relationships are complicated especially among the samples from the Amazonian side of the Andes (from Bolivia to Ecuador) and the Caribbean (including Central America). It is quite possible that both D. arbuscula and D. geniculata are species complexes, but we were unable to sort them out with the material at hand. In practice, we solved the problem by ignoring the phylogeny and assigning to D. arbuscula all specimens whose sterile leaves have relatively thick lamina texture and (dark) brownish colour and the fertile pinnae are elliptic with (long-)attenuate bases and apices. To D. geniculata we assigned those specimens whose sterile leaves have relatively thin lamina texture and greyish to bluish green colour and the fertile pinnae are lanceolate with truncate bases and acute to obtuse apices.

The placement of Danaea ushana in the phylogeny was a complete surprise, because until now it has been thought to belong to D. subg. Danaea: it has a creeping dorsiventral rhizome (vs erect and radially arranged in all species of D. subg. Arthrodanaea), no nodes on the petiole (vs present in almost all specimens of D. subg. Arthrodanaea) and also its pinnae look more like D. subg. Danaea than D. subg. Arthrodanaea. Nevertheless, both DNA accessions of D. ushana were firmly embedded within the D. simplicifolia clade, which suggests that they might be hybrids. The only species of D. subg. Danaea growing in French Guiana alongside D. simplicifolia and D. ushana is D. nigrescens Jenman. We hereby propose that D. ×ushana is a hybrid between D. simplicifolia and D. nigrescens.

Although Danaea ulei and D. bipinnata did not separate cleanly into two clades, their morphological differences are so big that we have no doubt they represent different species.

The Caribbean material that had previously been identified as D. geniculata formed a well-supported clade as sister to D. trifoliata. After comparing the voucher specimens to the material from Trinidad and Tobago, we concluded that the material from Mexico, Puerto Rico and Jamaica should be identified as D. trinitatensis. This resulted in a large range extension for D. trinitatensis from a single island (Trinidad) to much of the Caribbean, and a corresponding range contraction for D. geniculata.

We did not have genetic material of the type of Danaea atlantica, but we did have DNA of two paratypes (M. J. M. Christenhusz 4937 and 4911). These grouped with D. geniculata in the phylogeny. We were unable to find reliable morphological characteristics to separate between these two species in the herbarium and, therefore, treat D. atlantica as a synonym of D. geniculata. We also failed to find any systematic difference between D. polymorpha and D. draco Christenh. and treat D. draco as a synonym of D. polymorpha.

II. Danaea Sm. subg. Danaea Species:

Danaea subg. Danaea (Fig. 4) splits into two genetically distinct clades; an Amazonian clade (BP = 99), and a clade occurring everywhere else (BP = 99). Inside the Amazonian clade, D. elongata + D. cartilaginea form a well-supported clade (BP = 100), as do D. kessleri + D. nigrescens (BP = 87). In the extra-Amazonian clade, D. erecta forms a sister to a well-supported clade (BP = 100) containing the rest of the subgenus in two well supported clades, one containing D. latipinna + D. panamensis (BP = 91), another containing D. nodosa + the remaining nine species of D. subg. Danaea (BP = 96).

Fig. 3.

A: maximum likelihood tree of Danaea subg. Arthrodanaea; B: placement of D. subg. Arthrodanaea within Marattiaceae.

We were able to include 17 species of Danaea subg. Danaea in the phylogeny, but for three species we did not have DNA (D. epilithica, D. epiphytica, and D. megaphylla). Eight of the species in the phylogeny form clades with a bootstrap support > 70 (D. alba (BP = 89), D. ampla (BP = 81), D. elongata (BP = 78), D. erecta (BP = 100), D. kalevala (BP = 71), D. latipinna (BP = 82), D. leussinkiana (BP = 96), and D. panamensis (BP = 91)), two are represented by only one specimen in the phylogeny (D. antioquiana and D. longicaudata), and two form clades with a bootstrap support < 70 (D. cartilaginea and D. sellowiana). Three species pairs are somewhat ambiguous in the phylogeny, and these include the remaining five species (D. nodosa and D. pterorachis; D. grandifolia and D. sellowiana; D. nigrescens and D. kessleri).

We do not have DNA of Danaea nodosa from Hispaniola, from where the species was described, but the material from Jamaica and Puerto Rico is morphologically very similar, so we consider it justified to assume that they represent true D. nodosa. These samples form a clade with some of the Costa Rican and Mexican material that would traditionally have been identified as D. nodosa but has recently been assigned to D. media. Some of the Costa Rican specimens have nodes on the petiole and rhizomes with leaf bases in 3–5 rows, both characteristics that conflict with typical D. nodosa (no nodes on the petiole, rhizomes with leaf bases in exactly two rows). We assign these Costa Rican specimens to D. pterorachis, as scrutiny of the type material suggested that they do not conform with D. media and D. elata after all (Keskiniva & Tuomisto 2024). Unfortunately, D. nodosa and D. pterorachis do not separate to different clades, so further studies are needed to figure out their relationship.

All material of Danaea subg. Danaea from the Atlantic coastal forests of Brazil groups into one clade, which morphologically conforms with D. sellowiana, described from the same area. At the base of this clade is a polytomy consisting of all the specimens of D. grandifolia. Even though the latter does not form a clade, the two are morphologically clearly distinct (rhizome with leaf bases in two rows vs several rows, elliptic often slightly falcate pinnae vs parallel-sided straight pinnae, acuminate vs cuspidate pinna apices).

Within the Amazonian clade, the oldest available name is Danaea nigrescens, which we apply to the large clade containing material from the Guianas, as the species was described from that region. This species is remarkably widespread in the Amazonian lowlands and is genetically rather uniform, although three samples are separated into a polytomy with the closely related D. kessleri. Danaea nigrescens and D. kessleri were monophyletic with one of the alignments, albeit with poor support. Given that the two are not too difficult to tell apart based on their general appearance, we considered the evidence sufficient to recognize them as separate species.

We decided to treat Danaea moralesiana (from Costa Rica) as a synonym of D. erecta (widespread in the northern Andes) although we have no genetic material and have not seen the type of D. moralesiana. However, we have seen a picture of a paratype (A. Rojas 899, CR) and another specimen collected near the type locality (J. T. Mickel 3049, NY). The protologue of D. moralesiana provides no comparison with D. erecta and we were not able to find any separating morphological characters between these two.

III. Danaea subg. Holodanaea C. Presl (Presl 1845) (= Heterodanaea C. Presl; Presl 1845) Species:

Fig. 4.

A: maximum likelihood tree of Danaea subg. Danaea; B: placement of D. subg. Danaea within Marattiaceae.

Danaea subg. Holodanaea (Fig. 5) has the highest number of species and generally the largest genetic differences between species and clades. We were able to include 36 species of D. subg. Holodanaea in the phylogeny but did not have DNA for 7 described species and one proposed hybrid (D. betancurii, D. chococola, D. erosa, D. nasua, D. oblanceolata, D. peruviana, D. tuomistoana, and D. ×plicata).

Of the 36 species in the phylogeny, ten form clades with a bootstrap support > 70: Danaea alata (BP = 100), D. andina (BP = 100), D. bicolor (BP = 97), D. crispa (BP = 98), D. cuspidata (BP = 100), D. excurrens (BP = 100), D. lanceolata (BP = 75), D. pumila (BP = 79), D. vanderwerffii (BP = 90), and D. wendlandii (BP = 100). Three formed clades but with bootstrap support < 70 (D. imbricata, D. mazeana, and D. tenuicaulis), nine were resolved as not monophyletic, and 14 are represented by only a single specimen. The main observations for each of the six main clades of the phylogeny are described below.

Danaea acuminata clade

This is a well-supported clade (BP = 100) of five species, all of which were described in 2001–2010: Danaea acuminata, D. falcata, D. lucens, D. vivax, and D. xenium. The species are relatively similar in general appearance: creeping rhizomes, concolorous laminae that dry dark brown, and serrate pinna apices. The other clades in Holodanaea are more heterogeneous in appearance. Geographically this clade is concentrated to the Amazonian lowlands adjacent to the Andes in Peru and Ecuador, with one species (D. lucens) between the Cordilleras in Colombia.

Danaea alata clade

This is a moderately supported clade (BP = 72) that contains species from all corners of the wet American tropics: the Atlantic coastal forests of Brazil (Danaea ubatubensis), the Lesser Antilles and the Venezuelan coast (D. alata), Central America (D. wendlandii), the Pacific coast of Colombia and Ecuador (D. gracilis), and western Amazonia (D. lanceolata).

Danaea wendlandii, D. gracilis, and D. lanceolata form a well-supported clade (BP = 100) and are rather uniform in appearance, being rather small and having lanceolate laminae with narrow pinnae and terminal buds. Danaea alata and D. ubatubensis are clearly larger species, although D. ubatubensis shares the propensity to produce terminal buds. Three of the species were strongly supported as monophyletic, but D. gracilis was resolved as paraphyletic.

Danaea excurrens clade

Danaea excurrens was described from the Atlantic rainforests of Brazil, and our DNA samples from that area form a strongly supported clade (BP = 100) with similar-looking material from Bolivia. The DNA samples from Brazil represent both individuals with spathulate pinnae that conform with the type of the species and individuals that have a more classic Holodanaea appearance. Therefore, we have concluded that both forms belong to the same species, even though we have not discovered any obvious reason for such heterophylly.

Danaea crispa clade

This is a well-supported (BP = 100) clade containing three small-statured species from Central America that we would never have grouped together based on their morphology. One of them is unique in having simple, bicolorous laminae (Danaea carillensis), one resembles D. humilis of the D. cuspidata clade in having many small, bicolorous pinnae (D. pumila), and the third one is similar to D. gracilis of the D. alata clade in having dark green laminae that are so thin as to be translucent (D. crispa). The D. crispa clade forms a well-supported clade (BP = 93) with the D. moritziana clade.

Danaea moritziana clade

This clade consists of six species from the Caribbean region that are similar in having bicolorous laminae, large number of pinnae and clearly serrate pinna apices: Danaea moritziana (Coastal Venezuela and adjacent Colombia), D. mazeana (Lesser Antilles), D. urbanii (Greater Antilles), D. cuspidopsis (Costa Rica to northern Andes), D. jamaicensis (Greater Antilles), and D. jenmanii (Greater Antilles). They form a well-supported clade (BP = 98) with relatively small interspecific genetic differences.

The name Danaea moritziana C. Presl has traditionally been applied in a very wide sense. Our phylogenetic results suggest that the species should be circumscribed more narrowly. Our interpretation of the phylogenetic position of D. moritziana is based on material from the Caribbean coast of Colombia that morphologically matches the type of D. moritziana from Venezuela. Unfortunately, we do not have DNA from the type locality, but this interpretation leads to a morphologically coherent species whose distribution is limited to the northernmost parts of Venezuela and Colombia. Notably, none of the Mexican Holodanaea samples were resolved to the D. moritziana clade, strongly supporting the recognition of D. cuspidata as a distinct species. The situation was complicated by the fact that some Costa Rican and Colombian specimens were deeply embedded in the D. moritziana clade even though they were morphologically very similar to D. cuspidata. These have now been described as D. cuspidopsis (Keskiniva & Tuomisto 2024).

Christenhusz (2010) synonymized Danaea jamaicensis under D. mazeana, but the phylogeny shows that D. mazeana forms a clade of its own at the base of the D. moritziana clade, whereas D. jamaicensis groups with D. jenmanii. The latter two species are intermingled in the phylogeny, but we decided to keep them as separate species because of their morphological differences: D. jenmanii is generally smaller and has fewer pairs of shorter pinnae, shorter pinna apices, more scaly petioles and rachises, and the terminal pinna (or a part thereof) is replaced by a bud (vs terminal pinna usually present in D. jamaicensis). In our interpretation, D. mazeana is restricted to the Lesser Antilles, whereas D. jamaicensis and D. jenmanii occur in the Greater Antilles. The material from Lesser Antilles that we identified to D. mazeana has generally narrower pinnae and broader pinna apices, and they dry to a darker colour than D. jamaicensis.

Fig. 5.

A: maximum likelihood tree of Danaea subg. Holodanaea with major clades named; B: placement of D. subg. Holodanaea within Marattiaceae.

Danaea cuspidata clade

This moderately supported (BP = 77) clade is a mixture of large and small plants, translucent and thick laminae and distributions from Mexico to Bolivia and the Guianas (including Danaea ypori, the only species in D. subg. Holodanaea with a Guianan distribution) but no species from the Caribbean islands.

Both Danaea cuspidata (from Mexico) and the newly described D. andina (from Ecuador) formed strongly supported clades within the D. cuspidata clade, confirming their distinctness from D. moritziana. However, their positions within the D. cuspidata clade varied between analyses done with different alignments and was not well resolved in any of them. In general, the species in this clade were supported by the phylogenetic results, although most of them had too few samples for their coherence to be properly tested. The morphologically relatively similar D. imbricata and D. trichomanoides were resolved to different subclades, supporting their recognition as distinct species.

Species incertae sedis

Excluded species

Identification key

An open-access online identification key to the 81 taxa of Danaea is available at  https://keys.lucidcentral.org/keys/v4/neo_fern_genus_danaea/.

The list of features contains a total of 48 morphological characters. Twenty of these have negative dependencies (they disappear if an incongruent feature has been selected) and four have positive dependencies (they appear if a specific other feature has been selected). Both quantitative measurements and qualitative descriptions often overlap among species, even when they are not closely related, but a combination of traits should narrow the search down to one species when all necessary traits are present. However, often all traits are not present. For example, fertile leaves or rhizomes may be crucial to separate between two otherwise similar species, but information of them is often missing in herbarium specimens.

In general, the most useful morphological features for identifying Danaea species are leaf and pinna size, number of pinna pairs, number of petiole nodes, rhizome habit, fertile pinna shape, adaxial vs abaxial side colour, and the margins of the pinna apices.

Even though most species can (tentatively) be keyed out with morphological features alone, in practice the most effective way to start the identification process is to choose the geographical region where the specimen comes from. This is because the number of species that is known (or expected) to occur in any one region is almost always fewer than 20 and very often fewer than 10 (out of the total of 79 species and 2 hybrids). The main exceptions are Colombia (33 species), Ecuador (26 species), Panama (21 species), and Peru (21 species), all of which have both lowland and montane species. In the case of Colombia and Ecuador, the key also allows making a choice between the Pacific and the Amazonian side of the Andes, which reduces the number of relevant species to at most 18 per region.