Osmunda pulchella sp. nov. from the Jurassic of Sweden—reconciling molecular and fossil evidence in the phylogeny of modern royal ferns (Osmundaceae)

Background The classification of royal ferns (Osmundaceae) has long remained controversial. Recent molecular phylogenies indicate that Osmunda is paraphyletic and needs to be separated into Osmundastrum and Osmunda s.str. Here, however, we describe an exquisitely preserved Jurassic Osmunda rhizome (O. pulchella sp. nov.) that combines diagnostic features of both Osmundastrum and Osmunda, calling molecular evidence for paraphyly into question. We assembled a new morphological matrix based on rhizome anatomy, and used network analyses to establish phylogenetic relationships between fossil and extant members of modern Osmundaceae. We re-analysed the original molecular data to evaluate root-placement support. Finally, we integrated morphological and molecular data-sets using the evolutionary placement algorithm. Results Osmunda pulchella and five additional Jurassic rhizome species show anatomical character suites intermediate between Osmundastrum and Osmunda. Molecular evidence for paraphyly is ambiguous: a previously unrecognized signal from spacer sequences favours an alternative root placement that would resolve Osmunda s.l. as monophyletic. Our evolutionary placement analysis identifies fossil species as probable ancestral members of modern genera and subgenera, which accords with recent evidence from Bayesian dating. Conclusions Osmunda pulchella is likely a precursor of the Osmundastrum lineage. The recently proposed root placement in Osmundaceae—based solely on molecular data—stems from possibly misinformative outgroup signals in rbcL and atpA genes. We conclude that the seemingly conflicting evidence from morphological, anatomical, molecular, and palaeontological data can instead be elegantly reconciled under the assumption that Osmunda is indeed monophyletic. Electronic supplementary material The online version of this article (doi:10.1186/s12862-015-0400-7) contains supplementary material, which is available to authorized users.

TAB. S1.1. The range of values and the results of the k-clustering for XSpPSt.

Category
"0" "1" "2" Minimum difference between categories k = 2 0.37-1.78 1.87-2.86 n/a 0.09 k = 3 0.37-1.27 1. 40 Miller (1971) put strong emphasis on this character, which he classified into four character states ["in outer cortex to petiole base", "near boundary of inner and outer cortex", "in inner cortex", and "in stele" (the latter being a putative synapomorphy of subgenus Plenasium)] weighted with distance values up to 10 in his analysis of all species (matrix I) and into three categories (weighted as 0, 0.5, and 1) in that of the Osmundacaulis kolbei line (matrix II). We argue that the character cannot be accurately determined enough to warrant the classification into that many categories; moreover, the scoring of Plenasium is problematic, since leaf traces in this subgenus arise from two protoxylem strands that initiate independently in two adjacent stem-xylem segments. Hence, we treat this character as being not applicable in Plenasium species (scored as "?").  (1) heterogeneous.
This character is based on observations of Hewitson (1962) and Miller (1967Miller ( , 1971, and is Following the results of the cluster analysis, the binary scoring provided the most distinct NLTOC cut-off values (Table S1.3). See character 18 for details.
TAB. S1.3. The range of values and the results of the k-clustering for NLTOC.
Category "0" "1" "2" Minimum difference between categories k = 2 0.02-0.30 0.38-0.98 n/a 0.08 k = 3 0.02-0.14 0.18-0. The development of a heterogeneous sclerenchyma ring in the petiole is considered a critical diagnostic character for the identification of modern Osmundaceae in the fossil record (see Miller, 1971;Rothwell, 2002;Vera, 2008); fossil rhizomes that are superficially similar to those of modern Osmunda but that lack the heterogeneous sclerenchyma ring have been initially assigned to Osmundacaulis (Miller, 1971) and are currently accommodated in Millerocaulis (Tidwell, 1986) or Ashicaulis (Tidwell, 1994; see Vera, 2008, for a critical discussion of the genera).
The configuration of patches of particularly thick-walled fibres in the petiole sclerenchyma ring is highly diagnostic (Hewitson, 1962;Miller, 1967Miller, , 1971, and is scored in the four following characters. BS analysis under LS relied on 10,000 replicate trees inferred via the BioNJ algorithm (Gascuel, 1997) based on a matrix of mean (Hamming) pairwise morphological distances computed from the matrix using PAUP* (Swofford, 2002).
For MP-BS analysis, 10,000 replicate trees were inferred using PAUP* with the following settings (Müller, 2005). A single MP tree was inferred on each replicate matrix (option "MulTrees" deactivated) using heuristic search algorithm with option "AddSeq" set to "Furthest". All other settings left to PAUP* defaults.

ML-BS support was estimated via the fast bootstrapping (option -x) implementation in
RAxML v. 7.4.2 (Stamatakis, 2006b, Stamatakis et al., 2008 and 10,000 replicates. Both available transition models for categorical (multistate) data, the general time-reversible model (Rodriguez et al., 1990; -K GTR) and Lewis' (2001) model (-K MK), were applied. The GTR model will allow for different transitions rates between character states, whereas the Lewis' model estimate a single parameter, a general probability of character state change. For both models, we allowed for site-specific rate variation modelled via a Gamma (+) distribution with 25 distinct rate categories (default in RAxML; -m MULTIGAMMA).
1,000,000 generations were computed in ten parallel runs with one (cold) Monte-Carlo Markov chain each and allowing for parameter and topology swapping between runs following the recommendations in the manual for analysis with few taxa and characters. The topology of every 1000 th generation was sampled. Posterior probabilities take into account all sampled topologies from all ten runs saved for the first one, in total 10,000 sampled topologies. In contrast to molecular data sets, morphological data sets like the one used here converge directly to a plateau, why additional heated chains or high burn-in fractions are not

Set-up for evolutionary placement algorithm
Molecular data. -The evolutionary placement algorithm (EPA) implemented in RAxML Stamatakis, 2010, Berger et al., 2011) was used to (1) investigate the position of the outgroup-inferred root based on the concatenated and single-gene molecular data sets and (2) to place the fossil taxa individually within the molecular framework of modern taxa using a probabilistic approach.
Root test. -The ingroup-only topology was used as reference tree and then the EPA was invoked to find the optimal placement of the outgroup sequences within the ingroup topology (-f v). Results are shown in main text, Figure 9.
Independent optimization of the placement of fossils within the molecular framework. -EPA was invoked using three different weighting schemes. ML-based weights for each character were estimated by RAxML using 1000 replicates (-f u -# 1000) under the GTR+ and MK+ transition models using a molecular-based species-consensus ML tree as reference (files