Phylogeny of marattioid ferns (Marattiaceae): inferring a root in the absence of a closely related outgroup^†

Taxonomy

Modern marattioids have most commonly been treated in a single family Marattiaceae Kaulf. (e.g., 81; 100), although some authors (e.g., 69) have further dissected this into smaller families, Danaeaceae C. Agardh, Angiopteridaceae Fée ex J. Bommer, and Christenseniaceae Ching (= Kaulfussiaceae Campb.) with varying circumscriptions (59). Marattiaceae s.l. is typically treated as having four genera: Danaea Sm., restricted to the neotropics; Christensenia Maxon, restricted to Southeast Asia; Marattia Sw., with a pantropical distribution; and Angiopteris Hoffm., native to Madagascar, Asia, and Oceania (introduced in Hawaii and the American tropics). Some authors have recognized up to four poorly defined genera segregated from Angiopteris s.l.: Archangiopteris Christ and Giesenh., Macroglossum Copel., Protangiopteris Hayata, and Protomarattia Hayata, all with highly restricted ranges in Asia. 73 subdivided Marattia into five genera (Marattia,Discostegia C. Presl, Gymnotheca C. Presl, Stibasia C. Presl, and Eupodium J. Sm.), a taxonomy also followed by 18 in their monograph of the family; however, more recent treatments (e.g., 16; 81) have treated these genera within a more broadly circumscribed Marattia (referred to in this paper as Marattia s.l.).

While marattioids are united as a group by multiple distinctive characters, they are at the same time morphologically diverse. Danaea is the only reproductively dimorphic genus in the family and is also distinguished from other genera by long synangia that are sunken into a thickened lamina of the fertile fronds, sporangia that dehisce via apical pores, and large peltate scales. Christensenia is unique among marattioids in having palmately lobed to pedately compound blades, reticulate venation, circular synangia that appear more or less scattered on the abaxial side of the frond. Marattia is the most widespread and morphologically diverse genus, the paleotropical species having sessile synangia, most neotropical species having sessile or slightly raised synangia with lower pinnae more divided, and the neotropical M. laevis Sm. complex having distinctive stalked synangia. Angiopteris is distinguished by having sporangia that are fused only at the base and by having monolete spores; all other genera have trilete spores.

While there are some taxonomic difficulties with nearly all lineages, marattioids present an especially challenging combination of problems for a taxonomist, analogous in many ways to tree ferns, and these problems have contributed to a chronically unsettled species-level taxonomy. Many marattioids grow to be extremely large with some species having fronds over 6 m long. The large size of marattioids presents problems for taxonomists due to the necessarily fragmentary nature of most collections, some of which, including type specimens, consist of only a few small pinnules. Apart from size, separating plasticity from heritable differences is perhaps the greatest challenge for systematists working on marattioid ferns. Within populations and even within individuals one can find striking differences in morphology. Light, temperature, and frond age have a strong effect on all aspects of frond morphology, and irregular frond division is common in the family (3; 1; 34; A. Murdock, personal observation); herbarium specimens commonly present only a small picture of the overall variability of any given plant. Some investigators have misinterpreted hypervariability as taxonomically informative and have named numerous poorly differentiated species, making this group nomenclaturally burdensome. Additionally, the size, slow growth rate, and the difficulty of obtaining and propagating live plants make common garden experiments and even moderately large living collections impractical.

Because of the taxonomic difficulties in marattioid ferns, species diversity estimates have varied widely among investigators. In Angiopteris, 13 recognized 62 species and an additional nine species of Archangiopteris just for China, while 80 recognized only 10 species worldwide; others have only recognized one polymorphic species in Angiopteris (e.g., 36; 3). Nearly all species of Angiopteris have at one time been synonymized with A. evecta (type from Tahiti), and nearly all Old World Marattia species have at one time been synonymized with M. fraxinea (type from Réunion). One result of this taxonomic uncertainty is that assessment of conservation status of marattioid taxa is problematic. Many species have been construed as highly restricted endemics (24; 37; 43), and some are threatened by deforestation, plant invasion, and predation by feral pigs (112; 70; 6). These factors make a strong argument in favor of using molecular data and the resulting phylogenetic inferences to make informed taxonomic and conservation decisions for marattioid ferns.

Previous work

A morphological analysis by 33) was the first study to examine the evolutionary relationships of marattioid ferns in a cladistic framework and includes one of the most detailed discussions of morphological characters in the group to date. In addition to extant taxa, 33 included several well-characterized fossil taxa; 47 used a similar approach, but they were more concerned with interrelationships of fossil taxa. 33, using Ophioglossales as an outgroup, resolved Christensenia as sister to a clade containing all other marattioids and resolved Angiopteris as monophyletic only if defined in a broad sense to include Archangiopteris,Macroglossum, and Protomarattia. Both Danaea and Marattia were resolved as monophyletic; however, relationships among Danaea,Marattia, and Angiopteris s.l. remained unresolved. However, 33, p. 276) stated that Marattia,Macroglossum,Angiopteris, and Archangiopteris were all found to be paraphyletic in their analysis, despite the fact that their figures show only Angiopteris to be paraphyletic, and then only if narrowly circumscribed. Additionally, they implied that Marattia laevis should be segregated as a more narrowly defined genus based on cladistic principles and then discussed multiple alternatives to recognizing a paraphyletic Marattia; however, these conclusions and discussions are not consistent with their trees, where Marattia s.l. is shown as monophyletic.

Apart from recent phylogenetic work on Danaea (14) and the complete plastid genome sequence of Angiopteris evecta (Roper et al., 2007), relatively few marattioid ferns have been sequenced. Species that have been sequenced have mostly been part of large-scale phylogenetic studies (e.g., 32; 50, 51; 74, 8; 115; 76) with minimal sampling of marattioids. There have also been several recent papers focused specifically on Chinese and Taiwanese species of Angiopteris or Archangiopteris (37; 12; 46; S. Lin, Shenzhen Fairy Lake Botanical Gardens, unpublished manuscript), which have uncovered very little genetic differentiation between the many named species in the region. No molecular analysis has included the genus Christensenia, apart from 14), where it was used as an outgroup to Danaea. Inclusion of Christensenia in nucleotide sequence data sets could potentially improve the resolution of the placement of marattioid ferns in relation to other early-diverging fern lineages.

For this study, a phylogeny of marattioid ferns was inferred using nucleotide sequence data and morphology. This phylogeny helps to clarify the poorly understood taxonomy of marattioids and to interpret the morphological evolution and biogeographic history of the group. The understanding of polarity of morphological evolution in marattioid ferns is crucial for understanding relationships of extant taxa to putative fossil relatives, dating divergence times, and assessing morphological homologies with related lineages. Because confidence in rooting is essential for determining polarity and rooting can be difficult in groups that are phylogenetically isolated from other lineages, rooting hypotheses and outgroup relationships were explored using a multifaceted approach designed to isolate potential sources of bias and produce a consensus root position.

Sampling

Sampling for this study represented the broad range of morphology and geographic distribution in marattioids, including multiple individuals from all currently recognized genera (Christensenia, Danaea, Angiopteris, Archangiopteris, Macroglossum, and Marattia) as well as individuals from all previously recognized genera and subgenera within Marattia s.l. (i.e., Discostegia C. Presl, Eupodium J. Sm., Gymnotheca C. Presl, Stibasia C. Presl, and subgenus Mesosorus Rosenst.). Generic names are used in the sense of 16. Species names used are based on regional floristic treatments, taxonomic revisions, and the taxonomic opinions of the author based on examination of types and numerous other collections. Names shown in quotes are those that are used in regional floras but are likely to change in upcoming revisions (e.g., M. Christenhusz, University of Turku, unpublished manuscript, for Danaea).

Morphology

The previously published morphological analysis by 33 provided the starting point for building a new morphological matrix for marattioid ferns. Many characters used by Hill and Camus were at least partially nonindependent, and character states were questionably discrete in some cases, so the data set was critically reviewed and extensively recoded. Many characters useful for taxonomic purposes were excluded due to a lack of independence or a nondiscrete nature. While there are several cladistic analyses of fossils that included some marattioid ferns (e.g., 84; 85), these contained few additional characters that were useful for resolution within marattioid ferns. Additional characters have been added from a variety of literature sources and from observation of specimens and living material.

Phylogenetic markers

Multiple regions of the chloroplast genome, including rps4 plus the rps4–trnS^GGA intergenic spacer (rps4–trnS) (∼900 bp), and the region from trnS^GCU to trnG^UUC (including the trnG intron) (trnSGG) (∼1600–2100 bp) were sequenced to provide resolution within the ingroup. Eighty-eight accessions were sequenced for rps4–trnS and trnSGG for maximal resolution within marattioid ferns. The chloroplast region rps4–trnS was first used in fern phylogenetics by 99 and has been widely used in other groups of green plants. Apart from 97, a study that assessed the potential utility of numerous chloroplast noncoding regions for fern phylogenetics, trnSGG has not been widely used in fern phylogenetic studies. It appears, however, to be a highly promising phylogenetic marker for lower-level phylogeny reconstruction in many groups of plants (31; 92, 46; 107). In most nonflowering plants, trnSGG includes multiple short coding regions (trnS^GCU, psaM, ycf12, and trnG^UUC), which serve as positions for primer design and assist with sequence alignment. Three other chloroplast regions with relatively slow sequence evolution, atpB (∼1270 bp), rbcL (∼1370 bp), and rps4 (∼600 bp) were sequenced to provide resolution at deeper levels and assist with outgroup rooting. Because rbcL, atpB, and rps4 had limited variability in the ingroup, a smaller data set of 36 taxa (with 22 ingroup taxa selected to represent all major clades, and 14 outgroup taxa) was assembled for outgroup rooting.

Primers

Primers used in this study are shown in Table 1, with corresponding primer locations for trnSGG shown in Fig. 1. Novel primers were designed when necessary after initial sequencing efforts using published primer sequences and comparison with sequences in GenBank. The forward PCR primer for trnSGG from 92) is not optimal for pteridophytes because of base mismatches, but the reverse primer is generally applicable. Nor are the updated primers in 93, designed for angiosperms, not for pteridophytes because of base mismatches in both the forward and reverse primers and the location of the forward primer over an insertion found in the trnS gene in Equisetum. For trnSGG, forward and reverse PCR primers trnS-GCU-F1 and 3′-trnG-UUC, and internal sequencing primers 5′-trnG2G and 5′-trnG2S can be used for most nonflowering plants; additional internal primers were designed specifically for use in Marattiaceae and may work in related fern families; with minor modifications, these could be used in more distantly related ferns and other nonflowering plant lineages. Locations of primer sites in trnSGG are shown in Fig. 1. Internal primers for rbcL were designed specifically for this study and will likely have limited applicability in other lineages; other primers used for rbcL and atpB are variations on those used in other fern studies (e.g., 32; 116; 74) but have been redesigned for use in Marattiaceae and other early-diverging fern lineages.

Sequencing

Extraction of genomic DNA was performed using the Qiagen DNeasy Plant Mini Kit (Valencia, California, USA) from fresh, silica-dried, or recent herbarium specimen leaf material. In general, fresh or silica-dried material was needed; amplification of PCR products from herbarium specimens was often problematic. PCR was performed on 1:1 to 1:50 dilutions of total genomic extraction, using Bioneer PCR Premix Tubes (Bioneer, Alameda, California, USA), Amersham PCR Beads, or AmpliTaq Gold (Applied Biosystems, Foster City, California, USA) (specific reaction mixes available on request from the author). PCR was performed on an MJ Research (Waltham, Massachusetts, USA) PTC-200 thermal cycler with protocols following standard PCR protocols for atpB and rbcL, the protocol of 99 for rps4–trnS, and the protocol of 92 for trnSGG. Touchdown PCR (20) was used to amplify several samples that were problematic under standard protocols. PCR product was visualized and quantified on agarose gels stained with ethidium bromide and subsequently purified using USB ExoSAP-IT or USB PCR Pre-Sequencing Kit (USB Corp., Cleveland, Ohio, USA) prior to sequencing. Nucleotide sequencing was done on Applied Biosystems 96 capillary 3730xl DNA Analyzers or an Applied Biosystems 377 sequencer.

Sequence lignment and phylogenetic analysis

Contigs of sequences were constructed and proofread using the program Sequencher (Gene Codes, Ann Arbor, Michigan, USA). The rooting data sets were supplemented with sequences from GenBank (see Appendix 1); outgroup taxa were mostly single species, but in some cases were a composite of sequences from two related species when necessary due to availability of sequences and plant material. Coding regions were confidently alignable by eye. Where indels occurred in coding regions, genes were translated into amino acids using the program MacClade version 4.08 (49) to accurately identify the triplets involved in the indels. Noncoding regions were aligned first by the program ClustalW (http://www.ebi.ac.uk/clustalw/) and then fixed by eye where obvious alignment errors occurred. Areas of uncertain alignment and complex nested indels were excluded in final analyses. Microstructural characters were coded as present/absent where discrete, following 29). Identical taxa (i.e., with a pairwise difference of 0 in informative characters), of which there were several, particularly in Angiopteris, were merged prior to analysis when necessary for longer analyses.

Data from each gene region were analyzed separately and combined into three data sets: {1} rps4–trnS+trnSGG; {2} rps4–trnS+trnSGG with morphology and coded indels; and {3} atpB+rbcL+rps4 (for rooting analysis)—these data sets will be referred to later by their respective numbers inside the braces.

Maximum parsimony (MP), maximum likelihood (ML), and Bayesian inference using a Markov chain Monte Carlo approach (BI) were used to analyze phylogenetic data in this study using PAUP* version 4.0b10 (104) and MrBayes version 3.1 (38; 82). The programs ModelTest version 3.7 (72) and MrModelTest version 2.2 (67) were used to select models for ML and BI analyses, respectively, with model selection based on the Akaike information criterion (AIC). For MP and ML analyses, all data sets were analyzed using a heuristic search with 10 random sequence addition replicates and tree-bisection-reconnection (TBR) branch swapping. Starting trees for full ingroup MP analyses were obtained using 100000 random addition sequence replicates with no swapping and MulTrees unselected; the best trees found were then used as starting trees for complete TBR branch swapping with MulTrees selected. MP bootstrap analyses were run using heuristic searches with 10 random addition sequence replicates or were limited to 10 million rearrangements per replicate when necessary to complete an analysis. Decay analyses used commands generated in MacClade, which were then executed in PAUP*. BI analyses were continued until apparent stationarity, diagnosed using the program Tracer version 1.3 (77). The BI analysis of all data sets was based on 3000000 generations, four parallel chains, sampled every 1000 generations, with a burn-in of 500000 generations excluded from the final analysis. Tests for the presence of a molecular clock in the ingroup were performed using a likelihood ratio test (23) for all genes separately and combined in PAUP*.

Rooting

One hundred random outgroup sequences used in rooting trials were generated in MacClade using the average base composition of the ingroup (which did not significantly differ from the outgroups). Each of the 100 random sequences was used individually as an outgroup to the rooting data set {3} and analyzed under MP, and the first 20 sequences were analyzed under ML, to assess tendency of random outgroups to attach to particular ingroup branches. In both MP and ML random rooting trials, ingroup topology was constrained to match that of Fig. 2 using a backbone constraint tree, and random sequence addition was used. Shimodaira–Hasegawa tests (94; 26) comparing the likelihood scores, and Wilcoxon signed-rank tests (110) comparing the parsimony tree length scores of 41 constraint trees corresponding to all possible root attachment points within the atpB+rbcL+rps4 data set were performed with three possible outgroups (Equisetaceae, Ophioglossaceae, Osmundaceae) following the general procedures of 28) and 7 in PAUP* using RELL estimation. Constraint trees for all root positions analyzed were generated using MacClade. Lundberg root vectors were created by reconstructing the nodal states for a polytomy containing all outgroup taxa in MacClade with equivocal reconstructions coded as unknown. Midpoint and Lundberg rootings were calculated using PAUP*.

Phylogenetic relationships

Sequences of trnSGG proved to be the most informative within the 88 taxon ingroup data set, with 483 variable characters (367 parsimony informative) of an aligned length of 1992 characters; rps4–trnS had 246 variable characters (213 parsimony informative) of an aligned length of 992 characters. Both regions contained microstructural characters (indels and repeat motifs): 56 in trnSGG and 14 in rsp4–trnS (with only one of these in the coding region of rps4). In the 36-taxon rooting data set {3} (22 ingroup taxa and 14 outgroup taxa), rbcL had 566 variable characters (432 parsimony informative) of an aligned length of 1371 characters; atpB had 585 variable characters (426 parsimony informative) of an aligned length of 1269 characters; rps4 had 331 variable characters (266 parsimony informative) of an aligned length of 523 characters.

Support for the monophyly of Marattiaceae was consistently high in all analyses that included outgroups (100% Bayesian posterior probability and MP bootstrap), regardless of root placement within the family (described later). Support for internal branches was variable: in general, clades of closely related (and morphologically similar taxa) were highly supported on long branches with 100% Bayesian posterior probability, 100% MP bootstrap, and very high decay index values ≥43; however, support for relationships between and within these clades was often much lower (Fig. 2). Major clades that received high support included Danaea,Christensenia, and Angiopteris s.l. (i.e., including Archangiopteris and Macroglossum). Marattia s.l. was resolved as paraphyletic in all analyses, regardless of optimality criterion or inclusion of morphology. Marattia s.s., i.e., the clade that includes the type Marattia alata Sw., was resolved as sister to Angiopteris s.l. (91% Bayesian posterior probability, 81% MP bootstrap, decay index = 6). Two groups of taxa generally treated in Marattia s.l., the neotropical Marattia laevis complex, which corresponds to Eupodium J. Sm., and a clade of the paleotropical taxa, were both strongly supported as monophyletic; these two clades (referred to as “Marattia A” and “Marattia B,” respectively) were resolved as sisters in all molecular analyses; however, support for this relationship was lower (97% Bayesian posterior probability, 68% MP bootstrap, decay index = 6).

While MP analyses of data set {1} (with or without inclusion of outgroup taxa) and trnSGG alone were congruent with the topology produced by all ML and BI analyses, other MP analyses produced an alternate topology. In MP analyses of data set {2}, and rps4–trnS alone, Christensenia was resolved as sister to a clade of Marattia s.l. + Angiopteris s.l. Relaxation of parsimony by only a single step is all that is required to produce the ML topology, indicating the low level of support for either placement of Christensenia under MP. Apart from the change in position of Christensenia, all other relationships were congruent with the topology shown in Fig. 2, with bootstrap support values varying slightly but without showing any trend for increased or decreased support.

Relative to related lineages, even the longest branches within Marattiaceae are short (Fig. 3), suggesting either recent radiation or reduced rates of molecular evolution in the family relative to other lineages, as proposed by 101. While trnSGG proved to be the most variable region used in this study, this region is far more conserved in Marattiaceae than in other groups, including the filmy ferns (Hymenophyllaceae) (J. Nitta, University of California, Berkeley, unpublished manuscript); all regions used in this study proved to be much less informative than would be expected given the variation found in numerous studies of much younger lineages using the same markers. Tests for the presence of a molecular clock in the ingroup were rejected by a likelihood ratio test for all genes separately and combined. The relatively short length of branches, in combination with the distance from the nearest outgroups, likely contribute to the difficulty in rooting the phylogeny (see section Rooting).

Morphology

The MP analysis of the morphological data set produced a topology that is incongruent with molecular trees. While many elements agree with the molecular results (e.g., Marattia paraphyletic; Danaea, Christensenia,Eupodium,Ptisana, and Angiopteris monophyletic), the interrelationships of these groups differed, albeit with very few characters supporting any branch. Character descriptions, morphological matrix, and cladogram are included in Appendix S1 (see Supplemental Data with online version of this article).

Rooting

Rooting of the phylogeny was inconsistent between data sets, model choice, and optimality criteria. Additionally, both rooting and ingroup topology were affected by outgroup choice and analytical method. Figure 4 summarizes the rooting positions resolved by various analyses on the combined atpB+rbcL+rps4 data set. As a convention, root positions in this paper are referred to by the clade subtended by the branch to which the root attaches, e.g., “the branch subtending Danaea” and will be followed by the root position number in Fig. 5 in brackets, e.g., [1].

Parsimony rooting

With the exception of atpB analyzed separately, MP consistently resolved the root on the branch subtending Danaea [1], the longest internal branch, in analyses of data set {1}, data set {3}, rbcl alone, and rps4 alone. To test for the possibility of long-branch attraction (LBA), the ingroup portion of data set {3} was rooted with 100 random outgroups with base composition equal to the average composition of the ingroup taxa. In all 100 trials, MP attached the root to the branch subtending Danaea [1]. In 76% of the random outgroup rooting trials, Christensenia was resolved on the branch between Danaea and the rest of the taxa (i.e., sister to all other taxa), but in 24% of the trials Christensenia was separated from Danaea and placed in the same position recovered by ingroup analyses (i.e., in a clade with Marattia s.s. and Angiopteris). When Danaea was excluded from the analysis, MP resolved the root on the branch subtending Christensenia [24] in 97% of the trials. Lundberg rooting (48) agrees with all other MP rooting results, attaching the root to the branch subtending Danaea [1], as do results of analyses in which third position nucleotides were downweighted or excluded.

Wilcoxon signed-rank tests (110), comparing the parsimony tree length scores of 41 constraint trees corresponding to all possible root attachment points within data set {3}, were performed with three possible outgroups (Equisetaceae, Ophioglossaceae, Osmundaceae) (Figs. 5, 6). In the three analyses, only two branches were found to be not significantly less parsimonious (p≤ 0.05) than the MP root placement (the branch subtending Christensenia [24] and the branch subtending “Marattia A” [7]).

Phylogenetic relationships

While it was hoped that sequence data from Christensenia might break up the long branch leading to marattioid ferns and help resolve relationships among the early-diverging fern lineages, this was not the case. Christensenia could be sister to all other marattioid ferns, depending on root placement (see Discussion, section Rooting) however, nucleotide sequences from Christensenia are highly similar to those from other marattioids, and Christensenia is in no way a genetic “missing link” between marattioids and other early-diverging fern lineages.

Long-branch attraction, but not long-branch repulsion (LBR), appears to be affecting the analyses: the ML ingroup topology was consistent with or without outgroups, and the MP topology was either congruent with the ML topology or produced an alternate (but equally poorly supported) topology when outgroups were included. If LBR were affecting the ML analysis, MP analysis would not be expected to produce the same topology because parsimony has the opposite bias. Additionally, removing Danaea and/or Christensenia from the analysis had no effect on the ML topology. Under MP, real outgroups performed no differently than random outgroups. While this result does not indicate that the root chosen by MP is incorrect, it does provide evidence that LBA may be occluding the true signal. In contrast, when random outgroups were used to root the ingroup under ML, each of the 41 possible root positions was selected at least once, with a weak tendency to attach the root to longer internal branches. While LBA may be biasing MP analyses, the root position selected by MP corresponds to that chosen by midpoint rooting, enforcement of molecular clock assumptions under ML, MP Lundberg rooting, and morphology (85; 60); it is this consensus rooting that is shown in Fig. 2.

The phylogeny of marattioid ferns presented here has several important impacts on the understanding of the taxonomy of marattioids. Marattiaceae s.l. is resolved as monophyletic with strong support, in agreement with earlier phylogenies (e.g., Pryer et al., 2001; 19; Pryer et al., 2004; 115). Marattia s.l. is paraphyletic based on the evidence presented in this study, and is resolved in all analyses (excluding morphology alone) as three well-supported clades: Marattia s.s. (neotropical taxa and M. douglasii of Hawaii), “Marattia A,” which corresponds to Eupodium (the neotropical Marattia laevis complex with stalked synangia), and “Marattia B,” a clade comprising the paleotropical taxa. A taxonomic revision of Marattiaceae based on the phylogenetic results presented here, including new combinations and a more detailed discussion of nomenclature can be found in a forthcoming paper (Murdock, 2008).

The resolution of Danaea as sister to the other marattioid ferns (i.e., the consensus root position shown in Fig. 2) agrees with all previous molecular phylogenies (Pryer et al., 2001, 2004; 19; 57; 115; 85; 76; 89), but disagrees with the morphological analysis of 33). The sampling in this study is insufficient to make any clear conclusions about relationships of species groups within Danaea; 14, contains more comprehensive sampling within Danaea than is presented here.

Christensenia has been previously described as the most primitive of the marattioid ferns (e.g., 9; 33); the results of this study do not support this hypothesis. Christensenia is certainly distinctive, and the unique morphology of this genus is mirrored in the nucleotide sequences: while similar to Angiopteris and Marattia s.l., the sequences of Christensenia contain multiple autapomorphies, mostly in the form of unique indels that provide no grouping information within the family. The origin of Christensenia from within a clade of pinnately compound, free-veined taxa may seem surprising, but is the inevitable conclusion unless Christensenia was resolved as sister to all other marattioid taxa. 1 and 33 hypothesized that the leaf morphology and venation of Christensenia resulted from the compression of a rachis of a complex pinnate frond; the position in the marattioid phylogeny and the disposition of sori in irregular rows on both sides of main vein branches support this hypothesis. Christensenia has likely undergone significant morphological change as adaptation to extreme low-light environments: low, creeping habit; large, raised stomata; and high levels of chlorophyll concentrated in cells on the adaxial leaf surface (with abaxial surface cells possibly providing internal light reflection) (62; 79).

Because of the more or less free sporangia and trilete spores, it might be presumed that Angiopteris represents a plesiomorphic form among extant marattioids. 5 hypothesized the opposite case, arguing that the semifree state seen in Angiopteris sporangia is secondarily derived from a fully synangiate state: “The sporangia of Angiopteris, although discrete at maturity, have no morphological integrity when initiated.… In the Marattiales, the difference between free sporangia and synangia is directly attributable to differential growth of the intersporangial areas.” The phylogeny presented here supports Bierhorst's hypothesis: partially free sporangia and trilete spores are a derived state in extant marattioids. Fossil marattioids had both monolete and trilete spores (some fossils showing a mixture of the two in the same sporangium), and fused to free sporangia. Because monolete, ellipsoid spores are generally held to be a derived state in fern lineages, with the tetrahedral spores of the lycophyte and bryophyte lineages (and the trilete marks found in the majority of these lineages) assumed to be the plesiomorphic state, the trilete, tetrahedral spores in Angiopteris may represent a reversal to the ancestral state (111; 42). However, given that basal fern lineages have both monolete and trilete spores, and fossil marattioids also produce both types of spores, the polarity of the character state change in spore morphology is difficult to interpret. The finding that Angiopteris is monophyletic only if it is circumscribed to include Macroglossum and Archangiopteris (which also encompasses the rarely recognized Protomarattia and Protangiopteris) agrees with the results of 33, and the morphological arguments presented in 33 and 11.

Marattia douglasii

35 originally identified Marattia douglasii as M. alata, known only from the American tropics. Later authors either considered this species to represent an entirely new genus, Stibasia (e.g., 73; 18), or to be more closely related to the M. melanesica complex in Southeast Asia (45; 16). Nucleotide sequence data support Hooker and Arnott's original interpretation that M. douglasii is related to M. alata. The most likely biogeographic scenario that led to this distribution pattern is that the progenitor of M. douglasii dispersed to Hawaii from the American tropics, possibly carried by a tropical storm track (78). However, given the current sampling and topology, it remains a remote possibility that Marattia s.s. originated in Hawaii and dispersed to the neotropics. Further sampling within Marattia s.s. and clarification of fossil affinities may help answer this question.

Marattia purpurascens

One of the rarest species in the family, M. purpurascens, is endemic to Ascension Island and is known from only a few remaining populations (17; 30). While morphologically similar to the African M. fraxinea, the sequences from this species are clearly distinct from the South African M. fraxinea sequenced in this study. It is likely, given the wide range of morphological variation in M. fraxinea, that several species are subsumed under this name; while this has been previously proposed, species delimitation has proven difficult due to intermediate and overlapping morphologies (68). Because Ascension Island is quite young, ∼1 million years old, and the habitat for M. purpurascens on Ascension Island is probably much younger than that, given the relatively recent volcanic activity (64; 2), it would be remarkable (albeit not impossible) for the observed sequence divergence to have arisen in that time. Additional sampling from within the M. fraxinea complex, particularly from West African populations, is needed to clarify the origin of M. purpurascens and other lineages in the M. fraxinea complex.

Marattia howeana

Marattia howeana, another rare, endangered species, is endemic to Lord Howe Island. Threatened due to habitat destruction and grazing by introduced pigs, the few remaining known populations of M. howeana are being protected and monitored. While treated in the past as a variety of M. fraxinea, M. howeana is both morphologically and genetically distinct from M. fraxinea as well as nearby Australasian species.

Angiopteris evecta

This species, the type of Angiopteris, is extremely widespread and encompasses a wide range of morphological variation. The morphological plasticity of this genus is such that nearly all geographic and taxonomic distinctions break down upon close scrutiny. While there are indications of phylogenetic and biogeographic structure within the A. evecta complex, the markers used for this study are insufficiently informative to clearly elucidate relationships in this clade. Angiopteris longifolia, treated in most floras as a synonym of A. evecta, is distinguished mainly by having sori extending to the apex of the pinnule (as opposed to stopping well below the apex in A. evecta). In the Society Islands, these two forms can be seen in mixed populations with no other apparent differences (A. Murdock, personal observation). The form introduced to the Hawaiian Islands, purportedly from Samoa, is comparable to A. longifolia. While there is little basis for maintaining these as separate species, particularly given the range of morphology that A. evecta encompasses (as currently understood), it is interesting to note that MP analysis groups A. “longifolia” from Tahaa (Society Islands) with the three accessions from the Hawaiian islands, albeit with poor support (Fig. 2).

Morphology

Morphology suffers from the same general problem as the molecular data: the major clades are highly supported by multiple characters; however, there are very few characters that can inform interrelationships of these clades, and there is often homoplasy in these few characters. While inclusion of morphology in phylogenetic analyses has increased phylogenetic resolution in other plant groups (e.g., 21; 56), in Marattiaceae the clades that received strong support from the morphology and microstructural data are clades that were already strongly supported by sequence data.

Rooting

Rooting can be problematic in cases where the clade in question is distantly related to the closest available outgroup and when internal branches of the clade are relatively short (28; 88); Marattiaceae is an excellent example of just such a clade. Because accurate root placement is essential for making conclusions about character evolution, estimating divergence dates, and for testing most other tree-based hypotheses, it is prudent to check for possible rooting biases even when relatively close outgroups are available.

Multiple methods for dealing with problematic rooting and long-branch attraction (LBA) have been proposed by previous authors (see e.g., 113; 6564; 53; 87; 86; 28; 39). 48 suggested a method in which the ingroup is rooted using a vector statement representing a hypothetical ancestor. Lundberg rooting is of particular utility when outgroups are readily identifiable but are so distant that they are affecting the topology of the ingroup taxa; however, this is not really the case in the current study, as the ingroup topology is consistent with different outgroups, only the root position changes. Moreover, the composition of the Lundberg vector has a strong effect on rooting, and the best method of determining this vector is not always clear when there are multiple outgroups to choose from and their character composition is highly heterogeneous.

If a data set is clock-like, midpoint rooting or enforcing a molecular clock in a ML analysis are robust methods for selecting a root position. 39 found that even if data violates clock assumptions, enforcing a molecular clock may still identify the correct root. Midpoint rooting may choose the correct root under even wider ranges of rate heterogeneity (105). However, with both midpoint rooting and molecular clock enforcement, if clock assumptions are violated by the data, confidence in these rootings is impossible to assess, and alternative criteria are always desirable. While nonreversible character change models and asymmetric step-matrices have been used for rooting trees, these methods do not appear to provide consistently accurate results and often require unjustifiable assumptions (118; 39).

Neither MP nor ML is immune from the effects of LBA; MP tends to be more strongly affected, although ML can suffer similarly if models are poorly suited to the data due to violated assumptions or mismatch in the selection process (103; 106; 102). Additionally, ML might in some cases produce positively misleading results in the form of long-branch repulsion (LBR) (95; 96). While simulation studies have disagreed on the question of whether MP or ML are better suited to dealing with the challenges of LBA, the general consensus is that MP will outperform ML when long-branch taxa are sister in the true phylogeny (44; 4); as 71 stated, it is not clear “whether parsimony outperforms likelihood for reasons other than succumbing to its own vices at the right time.” The most effective measures for preventing LBA and LBR are to (1) use a model-based method with a good-fitting model, (2) increase taxon sampling as much as possible to break up the long branches, and (3) choose appropriately varying character sets in which the branches are not particularly long (55). While model-based methods (e.g., ML and BI) are readily available and widely applied, we currently have very few tools to accurately assess a priori the fit of models to data and to determine how model mismatch might affect the results (but see 27). Additionally, increasingly parameter-rich and better-fitting models are not necessarily better for phylogenetic inference (108; 102; 41). Increased sampling will be maximally effective for rooting if additional taxa fall along long branches; additional sampling within crown groups will likely have little effect. Unfortunately, in many cases, including Marattiaceae, extinction has removed the possibility of sampling below the crown group for molecular analyses.

The rooting problem encountered in Marattiaceae is similar to that found in several previous studies (e.g., 108; 114; 28; 88); rooting techniques used in this paper may prove useful in similar situations regardless of the organisms in question. In Marattiaceae, the rate heterogeneity and the absence of any near outgroup are the most severe issues; nucleotide compositional differences between the ingroup and the outgroup do not appear to be a factor as they were in Drosophila (108). Because model selection and outgroup composition have a strong effect on root placement, methods such as midpoint rooting, enforcement of a molecular clock, or Lundberg rooting may be the best options for rooting the family. All three of these methods resolve the same root position, the branch subtending Danaea [1], which is the same position found by MP and ML under simpler models. In addition, when the more parameter-rich models used in ML select alternative root positions, these positions are never significantly more optimal than root position [1] on the branch subtending Danaea.

Other possible data sources for determining the root of any clade include fossil morphology and external evidence such as stratigraphy and historical biogeography. While not always available, these are valuable sources of information on rooting and character polarity that should not be overlooked. In the case of Marattiaceae, fossil morphology, as currently understood, does not provide clear rooting information because of large amounts of missing data, seemingly random distribution of character states in fossil taxa, and the absence of Tertiary fossils that could be more directly compared with extant taxa. If the relationships of fossil marattioids and polarity of character change were more clearly understood, it is likely that more information on the rooting of the extant marattioid ferns could be obtained. Additionally, genomic structural data or gene duplications may provide clearer rooting information for the family in the future.