Historical and Current Perspectives on the Systematics of the ‘Enigmatic’ Diatom Genus Rhoicosphenia (Bacillariophyta), with Single and Multi-Molecular Marker and Morphological Analyses and Discussion on the Monophyly of ‘Monoraphid’ Diatoms

This study seeks to determine the phylogenetic position of the diatom genus Rhoicosphenia. Currently, four hypotheses based on the morphology of the siliceous valve and its various ultrastructural components, sexual reproduction, and chloroplasts have been proposed. Two previous morphological studies have tentatively placed Rhoicosphenia near members of the Achnanthidiaceae and Gomphonemataceae, and no molecular studies have been completed. The position of Rhoicosphenia as sister to ‘monoraphid’ diatoms is problematic due to the apparent non-monophyly of that group, so hypotheses of ‘monoraphid’ monophyly are also tested. Using an analysis of morphological and cytological features, as well as sequences from three genes, SSU, LSU, and rbcL, recovered from several freshwater Rhoicosphenia populations that have similar morphology to Rhoicosphenia abbreviata (Agardh) Lange-Bertalot, we have analyzed the phylogenetic position of Rhoicosphenia in the context of raphid diatoms. Further, we have used topology testing to determine the statistical likelihoods of these relationships. The hypothesis that Rhoicosphenia is a member of the Achnanthidiaceae cannot be rejected, while the hypothesis that it is a member of the Gomphonemataceae can be rejected. In our analyses, members of the Achnanthidiaceae are basal to Rhoicosphenia, and Rhoicosphenia is basal to the Cymbellales, or a basal member of the Cymbellales, which includes the Gomphonemataceae. Hypothesis testing rejects the monophyly of ‘monoraphid’ diatoms.


Introduction
Cymbopleura, Geissleria and Encyonopsis, but excluding Rhoicosphenia, as we are testing its phylogenetic position.
Genera in the Achnanthales per Round et al. [22] include Achnanthes (Achnanthaceae), Cocconeis (Cocconeidaceae), and Achnanthidium (Achnanthidiaceae). These are often referred to as 'monoraphid' diatoms, due to the presence of a raphe system on one valve only, and over the past two decades several genera including Karayevia Round & Bukhtiyarova ex [33], Lemnicola Round & Basson [34], Planothidium Round & Bukhtiyarova [35], Platessa Lange-Bertalot in [36], Psammothidium Bukhtiyarova & Round [37], and Rossithidium Round & Bukhtiyarova ex [33] have been proposed and include many species assigned previously to Achnanthidium and other genera in this group. Molecular data have been generated for some of these taxa, and the position of Achnanthes sensu stricto has been shown [38][39][40] distinct from other 'monoraphid' genera, such as Achnanthidium, Cocconeis, and Lemnicola. Based on the distant phylogenetic position of Achnanthes sensu stricto, we will here take a narrower view of 'monoraphid' diatoms and include the genera Achnanthidium, Cocconeis, Lemnicola, Planothidium, and Psammothidium, but exclude Achnanthes. The distant phylogenetic position of Achnanthes relative to the other aforementioned monoraphid genera was proposed by Mereschkowsky [16] and has been supported by molecular phylogenies [41,42]. Mereschkowsky [16] placed Achnanthidium (then Microneis) and Cocconeis into the Heteroideae, which excluded Achnanthes, so we will test whether Rhoicosphenia is part of a monophyletic group with taxa in the Heteroideae.
In the 1980's, there was substantial interest in the phylogenetic position of Rhoicosphenia [43][44][45][46][47][48]. Mann [43] asserted four hypotheses for the systematic position of Rhoicosphenia, which are paraphrased as follows (Fig 1); 1. a. Rhoicosphenia is an intermediate form between Achnanthes and Gomphonema, or, b. The common ancestor of 'monoraphid' and 'gomphonemoid' genera, 2. Rhoicosphenia is a 'monoraphid' diatom, 3. Rhoicosphenia is related to Gomphonema, and 4. Rhoicosphenia is unrelated to 'monoraphid' and gomphonemoid diatoms. Hypothesis 1 has two parts; (a) Rhoicosphenia is an intermediate form between Achnanthes and Gomphonema, and (b) is the common ancestor of both 'monoraphid' and gomphonemoid groups. Hypothesis 1a was proposed by Schütt [21] with Rhoicosphenia being the link between Gomphonema and Achnanthes, but we are unable to test the topology with our statistical methods and will therefore not statistically address the hypothesis in this paper. Hypothesis 1b is not testable with hypothesis testing techniques, since Rhoicosphenia would not occupy a position as a terminal taxon, but rather be placed at a node of divergence between 'monoraphid' and gomphonemoid diatoms. However, the hypothesis will be tested broadly in the context of the position of Rhoicosphenia compared to other genera. Hypothesis 2 [43] follows Grunow and Hustedt, with Rhoicosphenia being more closely related to 'monoraphid' diatoms. Hypothesis 3 [43] follows Van Heurck and Mereschkowsky and states that Rhoicosphenia is sister to Gomphonema. Finally, hypothesis 4 [43] most closely resembles Cleve's hypothesis that the phylogenetic affinity of Rhoicosphenia to 'monoraphid' diatoms is due to polyphyletic origins of the 'monoraphid' condition, but also does not lend itself to hypothesis testing because we cannot place Rhoicosphenia in an unknown position in the tree.
In studying the morphology of Rhoicosphenia valves in detail, some of Mann's [43] conclusions were that the valve symmetry of Rhoicosphenia is similar to Gomphonema and Cymbella, Rhoicosphenia valves are not similar to Achnanthes or Cocconeis, the chloroplasts of Rhoicosphenia are more similar to Achnanthidium than Achnanthes (and cites [16] chloroplast work), and Rhoicosphenia is unlike Gomphonema due to areolar occlusions differences [43]. Subsequently, Mann notes differences in sexual reproduction between the isogamous Rhoicosphenia and the physiological anisogamy of Gomphonema and Cymbella [44]. The conclusions of Mann's final paper support the 4 th hypothesis, that Rhoicosphenia 'clearly' is not allied with 'monoraphid' diatoms, but belongs in an 'isolated position' near the gompho-cymbelloid diatoms within the Naviculales and offers an emended description of the family Rhoicospheniaceae [45].
Soon after Mann's papers, a cladistic analysis of Cocconeis, Mastogloia Thwaites in [49], Achnanthes sensu lato, Gomphonema, and Rhoicosphenia was produced [48]. Using eleven morphological characters to test historical hypotheses similar to those in Mann [43], the analysis showed that Rhoicosphenia is more closely related to Gomphonema, with Achnanthes sensu lato as sister and Cocconeis more distantly related [48]. In that analysis, Rhoicosphenia did not occupy an undetermined position, but was sister to Gomphonema and only closely allied with one of the other 'monoraphid' genera, Achnanthes sensu lato. A more recent cladistic analysis using morphology that included Rhoicosphenia employed more characters (n = 35) and taxa (n = 49). This analysis placed Rhoicosphenia in an unresolved polytomy of raphid diatoms [50]. These subsequent results do not support Grunow's hypothesis of relationship, based on his decision to place his 'newly' erected genus in the Achnantheae, and also rejects the hypothesis that Rhoicosphenia is sister to Gomphonema. The results showed that some members of Cymbellales sensu Mann in [22], (Cymbella, Encyonema, Gomphonema, and Reimeria) are a natural group, but Anomoeoneis, Placoneis and Rhoicosphenia were not allied with that group [50]. Also, the 'monoraphid' diatoms in that study, Achnanthidium and Cocconeis, formed a natural group, but Rhoicosphenia was excluded from that clade [50]. In terms of the four hypotheses forwarded by Mann, the study by [50] supports hypothesis 4, that Rhoicosphenia occupies an 'enigmatic' position in the raphid diatom phylogeny [43,45]. Cox [51] discussed several morphological characters of Achnanthes sensu stricto and suggested it belongs in the Mastogloiales Mann in [22], rather than Achnanthales, again casting doubt on the monophyly of 'monoraphid' diatoms, supporting proposals made at the turn of the 20 th century [16,20]. Rhoicosphenia is also interesting because two of its potential phylogenetic positions, 'monoraphid' or Gomphonema (Cymbellales), are consistently returned as sister taxa in molecular analyses [1, 38-40, 42, 52-54], but many of these analyses are focused on other questions and have not discussed this relationship [22,31,[55][56][57][58].
Two additional hypotheses are added that are not strictly related to Rhoicosphenia, but more broadly to 'monoraphid' diatoms. The first, H 5 , addresses the issue of whether or not all 'monoraphid' diatoms are monophyletic. Several molecular and one morphological [50] have suggested that this is not the case, as Achnanthes sensu stricto is not part of a monophyletic group with the other 'monoraphid' diatoms, such as Achnanthidium and Cocconeis, and in fact is quite distantly related to them. The second, H 6 , tests the hypothesis, forwarded by Cox [51], that Achnanthes sensu stricto is closely related to the genus Mastogloia.
The major goal of this project is to use single and multi-marker molecular analyses, as well as analysis of morphological data to determine the systematic position of Rhoicosphenia in the diatom tree of life within the context of previous taxonomic hypotheses.

Molecular Analyses
Taxon collections. Three Rhoicosphenia populations were isolated from freshwater streams into monoculture via micro-pipette serial dilution from collections made in California, Colorado and Oregon, USA, and were grown in freshwater WC medium [59]. After isolation, the cultures were maintained at a temperature of approximately 25C, with a 12:12 light dark cycle at an irradiance of 50 μmol cm -2 s -1 . The other 4 sets of sequences were obtained via a Chelex extraction from colonies found in live samples. Colonies were chosen to ensure that DNA was obtain from one genetic clonal line. Table 1 contains information on sampling locations of sequenced specimens. Samples in California were collected with a Scientific Collecting Permit from the California Department of Fish and Wildlife, issued to Evan W. Thomas. The Oregon Department of Fish and Wildlife and Colorado Department of Natural Resources did not require permits for microalgal collections. All collections were made from state, county, and city parks, or from waterways accessible from public roads and no field sites had endangered or protected species. Prepared diatom slides, referenced by Collection number in Table 1, containing sequenced population are housed in the Kociolek Diatom Collection, University of Colorado, Museum of Natural History, Boulder, Colorado, USA.
Seven Rhoicosphenia populations were sequenced for this analysis with 7 isolates yielding partial 18S small subunit rDNA (SSU) sequences, 6 sequences from the D1-D2 region of the 28S large subunit rDNA (LSU), and 4 sequences from the chloroplast encoded large subunit of RUBISCO (rbcL). Only three populations yielded sequences for all 3 markers. The list of populations studied, including taxon name, ID, and GenBank accession numbers is presented in Table 2. Additionally, GenBank was used to obtain an additional 140 sequences for SSU, 80 sequences for LSU, and 100 sequences for rbcL and a list of these taxa are included as supplemental document S1 Table. The concatenated three marker tree includes 3 Rhoicosphenia sequences and 78 GenBank sequences (S2 Table).  [60] was used to extract DNA from monocultures and was modified to a volume of 20 μL Chelex for colonies of Rhoicosphenia. The molecular markers chosen, include the conserved (SSU) and variable (LSU, rbcL), which have been shown to provide order [1,38,42,55] and species [61][62][63] level resolution. Further, due to the widespread use of these markers in diatom phylogenetics [1, 31, 38-40, 42, 53, 55, 61, 63], it allowed for the broadest taxon sampling of non-Rhoicosphenia Gen-Bank sequences from the raphid diatoms. Primers used in amplification and sequencing of these markers are listed in Table 3.
Using GE Healthcare illustra Ready-To-Go™ PCR beads (GE Healthcare Biosciences, Pittsburgh, Pennsylvania) following the manufacturer's instructions, all markers were amplified by polymerase chain reaction (PCR). PCR was performed in an Eppendorf Mastercycler 1 using the program: 94 C for 3:30, 36 cycles of 94 C for 50 seconds, 52 C for 50 seconds, 72 C for 80 seconds, with a final extension at 72 C for 15 minutes. After amplification, the PCR products were purified with ExoSAP-IT (Affymetrix, Santa Clara, California) using the manufacturers protocol. Purified PCR products were sequenced at Functional Biosciences, Inc. (Madison, Wisconsin) and Geneious ver. 5.6 [68] was used to assemble and edit sequences. Sequences for the seven Rhoicosphenia taxa included in this analysis are deposited in GenBank and accession numbers for SSU, LSU, and rbcL sequences are listed in Table 1.
Sequence alignment and phylogenetic analysis. A muscle alignment algorithm [69] in Geneious was used for all alignments. The three molecular markers were aligned separately prior to concatenation in the two and three-molecular marker alignments. The ends were trimmed from each of the alignments to minimize missing characters. A variable 63 base pair region of SSU, corresponding to region 579-641 in the initial alignment, was removed due to the ambiguity in the alignment, creating a final trimmed length of 1566 sites. The final trimmed length of LSU was 604 base pairs and rbcL had a final trimmed length of 799 base pairs. The three-marker concatenated alignment for 81 taxa was 2969 sites. The SSU alignment included 140 non-Rhoicosphenia taxa with representatives from all available raphid diatom orders sensu [22]. The LSU and rbcL alignments included less taxa, but attempted to maintain coverage of raphid diatom groups based on available sequences. The number of taxa included in alignments are as follows: SSU-147; LSU-86; rbcL-104; SSU + LSU-85; SSU + rbcL-97; LSU + rbcL-81; and SSU + LSU + rbcL-81. To understand the position of Rhoicosphenia All seven alignments were analyzed using the general time reversible (GTR) model with a gamma distribution (Γ) and a proportion of invariable sites (I) [1,40]. SeaView version 4.3.4 [70] was used to perform maximum likelihood (ML) analysis with PhyML version 3.0 [71] using the GTR+ Γ+I model with four rates classes and 500 bootstrap replicates to estimated branch support. MrBayes version 3.2.1 [72] was used to perform Bayesian analyses. Analyses were run using the default settings and a GTR+Γ+I model with four rate classes. The single and two-molecular marker alignments were run for 10 million generations with a burn-in of 2 million generations, and the three-molecular marker alignment was run for 30 million generations with a burn-in of 6 million generations; all alignments were analyzed using two runs of four MCMC chains sampled every 1000 generations. Maximum likelihood phylograms are presented in this paper and nodes are labelled with maximum likelihood bootstrap values (BS)/Bayesian posterior probabilities (BPP) reported as percentages. In situations where the ML and Bayesian trees are incongruent, the Bayesian node support is denoted as (-). Hypothesis testing. Hypotheses concerning the monophyly of Rhoicosphenia were tested using tree likelihoods and the Approximately Unbiased (AU) test [73]. For the test using the two and three-molecular marker alignments, an unconstrained tree (H 0 ) was tested against four constrained alternative topologies:  [74] and the graphical user interface raxmlGUI ver. 1.3.1 [75] were used to generate maximum likelihood trees from the unconstrained and constrained alignments for hypotheses 2 and 3 (A & B), using GTR+ Γ+I model. The probability that the alternative topologies were as likely as the null topology (unconstrained tree) was tested by calculating per site log likelihood values using RAxML and implementing the AU in the program CONSEL using default settings [76]. In CONSEL the AU test compares a hypothesized tree topology to a set of trees generated through a multi-scale bootstrap technique of per site log likelihoods. A statistically significant result, p-value less than or equal to 0.05, means that the hypothesized tree topology can be rejected, while a p-value greater than 0.05 does not allow the rejection of the hypothesized constrained tree.

Morphological analyses
The taxa, character matrix, and character states used in this analysis were published in [50]. Our analysis used 33 of the 49 taxa published in [50] to maximize taxa shared between our morphological and molecular analyses. The characters used, as well as their coding, has been left unchanged from the original dataset [50], but we ran all data, protoplast and frustule, together in our analysis. The explanation and coding of characters can be found in S3 Table  and the taxon and character matrix is presented in S4 Table. Phylogenetic analysis was performed in PAUP Ã 4.0b10 [77], and all 35 characters were unordered and equally weighted. Trees were generated using the branch-and-bound search option to determine the 200 most parsimonious trees that were then used to compute a strict consensus tree, which can be found as S1 Fig.

Molecular Phylogenies
In the analysis of the three-molecular marker concatenated alignment (Fig 2), both the ML and Bayesian analyses support a clade consisting of 'monoraphid' diatoms, members of the Cymbellales sensu lato, and Rhoicosphenia, to the exclusion of all other diatoms. In the ML threemolecular marker concatenated tree, Rhoicosphenia is not sister to Cocconeis, but is sister to the Cymbellales clade, with Achnanthidium and Cocconeis forming a grade basal to Rhoicosphenia. In the Bayesian three-molecular marker concatenated tree, Achnanthidium and Rhoicosphenia + Cocconeis are a 'monoraphid' grade basal to the Cymbellales. When concatenated, the two nuclear markers, SSU and LSU, show consistent topologies in both ML and Bayesian analyses (S2 Fig). Rhoicosphenia strains are monophyletic, and sister to Anomoeoneis, that clade is sister to a large portion of the Cymbellales, including the genera Encyonema, Cymbella, Cymbopleura, Didymosphenia, Geissleria, Placoneis, Gomphonema, and Gomphoneis. Basal to the clade containing Rhoicosphenia and the aforementioned genera is Adlafia Moser, Lange-Bertalot & Metzeltin [78], and sister to Adlafia + Rhoicosphenia + Cymbellales is a basal grade of the 'monoraphid' genera Achnanthidium and Cocconeis. ML and Bayesian analyses recover congruent topologies for SSU and rbcL when concatenated (S3 Fig). Rhoicosphenia strains are sister to Cocconeis placentula and C. pediculus, and the other 'monoraphid' taxa (C. stauroneiformis, Lemnicola hungarica, and Achnanthidium minutissimum) + Rhoicosphenia and the two Cocconeis are represented as a grade of taxa basal to the Cymbellales. These analyses show Adlafia as basal to the Cymbellales. The other 'monoraphid' taxa in these analyses, Achnanthes sensu stricto (four sequences), are not closely related to the previously mentioned 'monoraphid' diatoms and Rhoicosphenia.
LSU and rbcL results (S4 Fig) recover a monophyletic clade consisting of Rhoicosphenia + Cocconeis placentula and C. pediculus + Achnanthidium minutissimum, however, C. stauroneiformis is not part of that group. The clade of Rhoicosphenia + C. placentula and C. pediculus + A. minutissimum is not sister to the Cymbellales, however there is very low bootstrap support (44) for the node separating them from the intermediate clade made of biraphid naviculoid diatoms.
Both ML and Bayesian SSU analyses (S5 Fig) provide congruent results with the concatenated alignment that the genus Rhoicosphenia is basal to the Cymbellales. The SSU topology shows a well-supported (95 ML BS) lineage consisting of 'monoraphid' genera and the Cymbellales. Cocconeis and Achnanthidium, two 'monoraphid' genera, are non-monophyletic and are basal to a clade consisting of Rhoicosphenia + Cymbellales. The node where Rhoicosphenia splits from the Cymbellales has a bootstrap value of 45.
LSU results (S6 Fig) recover a topology where Rhoicosphenia is sister to two Cocconeis species, with another Cocconeis species sister to Achnanthidium and those two are not sister to Rhoicosphenia + Cocconeis. However Rhoicosphenia + Cocconeis are not sister to the Cymbellales, and are in a weakly supported (3 ML BS) clade with naviculoid diatoms. The Cymbellales clade recovered is similar to the clade in the three molecular marker and SSU analysis.
rbcL sequences result (S7 Fig) in a topology similar to the LSU analysis in that Rhoicosphenia is sister to Cocconeis. Unlike SSU, the rbcL phylogeny has more 'monoraphid' taxa (excluding Achnanthes sensu stricto) that form a weakly supported clade (10 ML BS) sister to the Cymbellales. Unlike LSU, rbcL does not result in a polytomy, but assigns branching order with Rhoicosphenia sister to Cocconeis, which together are sister to the Cymbellales.

Hypothesis Testing on Molecular Phylogenies
Full results of hypothesis testing for all seven alignments; SSU, LSU, rbcL, SSU + LSU, SSU + rbcL, LSU + rbcL, and SSU + LSU + rbcL; can be found in Table 4. In testing alternate constrained topologies against the unconstrained phylogeny, examining the three molecular marker concatenated tree, we cannot reject H 2a : that Rhoicosphenia is a Heteroideae diatom, H 2b : that Rhoicosphenia is sister to Achnanthidium, and H 2c : that Rhoicosphenia is sister to Cocconeis. The hypothesis that Rhoicosphenia is sister to Gomphonema (H 3 ), could be rejected (p = 0.029). In the SSU + LSU analysis, we can only reject hypothesis 3, that Rhoicosphenia is sister to Gomphonema (p = 0.042).
For LSU, we cannot reject any of the alternative hypotheses, H 2a,b,c or H 3a,b . For rbcL, we can reject H 2c , that Rhoicosphenia is sister to Cocconeis (p = 0.019), and H 5 , that all 'monoraphid' diatoms are monophyletic (p < 0.001).

Morphological Phylogeny
The strict consensus tree of the 200 trees returned from the branch-and-bound parsimony analysis was similar to the consensus tree using all data from [50]. Our tree (S7 Fig) returned Rhoicosphenia in an unresolved polytomy of 20 taxa, however within that polytomy members of the same genus did group together. Although our tree was unable to resolve relationships with any more detail than [50], we are still including the tree in this paper. The consistency (CI) and retention indices (RI) from our analysis, CI = 0.4727 & RI = 0.7434, are similar to those of [50], CI = 0.39 & RI = 0.77.

Discussion
The results of the molecular analyses from this study provide insights into the evolution of the 'monoraphid' condition, and also lend support to the Cymbellales sensu Mann in [22], with both of these results having implications for the systematic position of Rhoicosphenia. First, SSU + rbcL (S3 Fig), SSU (S5 Fig), and rbcL (S7 Fig), do not support a monophyletic lineage of 'monoraphid' diatoms of the genera Achnanthes, Achnanthidium, Cocconeis, Lemnicola, Planothidium, and Psammothidium (Table 4). Past molecular results have indicated that Achnanthes is more closely related to the Bacillariales than the other genera previously listed [38-40, 42, 52-55], however Cox [51] hypothesized that Achnanthes sensu stricto and Mastogloia are sister taxa. Hypothesis testing for monophyly of these genera in the analyses of SSU + rbcL, SSU, and rbcL yields mixed results with SSU rejecting that relationship, while rbcL and SSU + rbcL failed to reject that relationship (Table 4). In light of these results, instead of testing the position of Rhoicosphenia against the non-monophyletic 'monoraphid' diatoms, we tested its position against the Heteroideae [16] consisting of the families Achnanthidiaceae (Achnanthidium, Lemnicola, Planothidium, and Psammothidium) and Cocconeidaceae (Cocconeis).
Our three-molecular marker analysis yields a well-supported relationship with Rhoicosphenia as sister to a monophyletic clade of the Cymbellales, and a grade of 'monoraphid' taxa including Achnanthidium and Cocconeis is sister to Rhoicosphenia + the Cymbellales (Fig 2). Hypothesis testing on the three-molecular marker topology rejects the hypothesis that Rhoicosphenia is sister to Gomphonema, but does not reject the hypothesis that Rhoicosphenia is a member of the Heteroideae. The three-gene, SSU, and rbcL phylogenies also support the sister relationship of the Heteroideae and the Cymbellales + Adlafia. This is not a novel topology, as it has been evident in other molecular analyses [1,[38][39][40]42], but has only been discussed in [42]. The only topology rejected by hypothesis testing on the three-molecular marker analysis was the sister relationship between Rhoicosphenia and Gomphonema. The Heteroideae were monophyletic in the three-molecular marker tree, so hypotheses H 2b,c were not tested and H 2a was not rejected ( Table 4).
Analyses of concatenated alignments of two molecular markers generated three different topologies. The phylogeny based on SSU + LSU shows Rhoicosphenia as sister to Anomoeoneis, within the Cymbellales. This combination of molecular markers is the only one out of the seven molecular analyses to return this topology. It is interesting for two reasons. First, it is the only tree in which Rhoicosphenia is within, as opposed to outside the Cymbellales sensu Mann in [22]. Second, neither SSU nor LSU, when analyzed alone, return this result (S5 and S6 Figs). Although parts of the tree have low support, the node that places Rhoicosphenia within the Cymbellales has moderate support (83 BS, 97 BPP). Hypothesis testing only rejects the sister relationship between Rhoicosphenia and Gomphonema, and fails to reject the three different hypothesis in regards to the position of Rhoicosphenia relative to the Heteroideae. SSU + rbcL, show a sister relationship between Rhoicosphenia and the two freshwater Cocconeis species. The clade including these taxa, along with the 'monoraphid' genera Lemnicola and Achnanthidium is sister to a clade of Adlafia + Cymbellales with moderate support (71 BS, 100 BPP). Cocconeis stauroneiformis is not sister to the 'monoraphid' genera, but is basal to the other Heteroideae + Cymbellales. Hypothesis H 2c was rejected, meaning that even though the most likely tree places Rhoicosphenia and the two freshwater Cocconeis species as sister taxa, this relationship has very low support. This alignment allowed the testing of all 'monoraphid' genera, including Achnanthes sensu stricto, and the monophyly of these genera was rejected, while the hypothesis of Achnanthes sensu stricto as sister to Mastogloia was not rejected.
LSU + rbcL recover a moderately-supported sister relationship between Rhoicosphenia and Cocconeis (76 BS, 98 BPP), and a less well-supported sister relationship between Rhoicosphenia + Cocconeis and Achnanthidium (45 BS, 98 BPP), the other 'monoraphid' taxon in the analysis. However, the sister relationship between the 'monoraphid' genera and Cymbellales is not supported in this analysis and Cocconeis stauroneiformis does not fall with the 'monoraphid' genera. Hypothesis testing rejected the hypothesis that Rhoicosphenia and Gomphonema are sister taxa.
The single molecular marker trees generated in this study supported different hypotheses of relationships for Rhoicosphenia. Other studies of diatoms analyzing multiple single molecular marker and concatenated alignments [38,42,55] demonstrate similar results, that is, not all single molecular marker trees recover the same tree topologies as each other or the concatenated alignment. Our single molecular marker analyses of SSU (8 BS) and rbcL (39 BS) suggest a weakly supported relationship between 'monoraphid' diatoms and Rhoicosphenia, together being sister to a moderately to poorly supported (SSU 63 BS, rbcL 26 BS) Cymbellales clade (S5 and S7 Figs). In the SSU analysis, Rhoicosphenia is sister to the Cymbellales clade with a branch support of 64 (ML bootstrap). Hypothesis testing could not reject Rhoicosphenia as either part of the Heteroideae, or as sister to Gomphonema. However, the hypothesis that all 'monoraphid' diatoms are monophyletic was rejected, while the hypothesis (H 6 ) that Achnanthes sensu stricto is sister to Mastogloia was not rejected.
rbcL has weak support, 26 (ML BS), for a sister relationship between the Heteroideae and the Cymbellales, with Rhoicosphenia being sister to Cocconeis 39 (ML BS) deep within the Heteroideae. Hypothesis H 2c was rejected, meaning that even though the most likely trees places Rhoicosphenia and the two freshwater Cocconeis species as sister taxa, this relationship has very low support. Both the SSU and rbcL results support Mereschkowsky's Pyrenophoreae [16], based on chloroplast number and structure but including diverse valve morphologies. Hypothesis testing of all 'monoraphid' diatoms, H 5 , was rejected with rbcL, however the hypothesis (H 6 ) that Achnanthes sensu stricto is sister to Mastogloia was not rejected. Unlike SSU and rbcL, LSU places Rhoicosphenia sister to Cocconeis with weak support 34 (ML BS), with taxa not sister to the Cymbellales. However, deeper nodes in the LSU phylogram are very weakly supported <10 (ML BS), which could be reflective of LSU being a faster evolving marker in diatoms [79]. Our results with LSU and LSU + rbcL are similar to the LSU trees generated in [42,55], in that their LSU returned the most unique topology of the three single molecular marker analyses. After analyzing all trees based on single, two-, and three-molecular markers we, similar to previous investigators [38,40,42,55], have decided to base our conclusions on the three molecular marker concatenated alignment.
With regards to morphological analysis the strict consensus tree generated from 200 most parsimonious trees produced a large polytomy of taxa, with only congeneric species within the analysis being resolved together (S1 Fig). This result only differs from [50] (their Fig 5 and 6) in that their analysis groups some genera together, within a larger unresolved polytomy. This result, when compared to [50], indicates that our documentation and understanding of morphological characters that can inform a broad phylogeny of the raphid diatoms is currently insufficient.
In addition to the systematic position of Rhoicosphenia, our SSU analysis shows that the 'monoraphid' condition evolved multiple times, once in Achnanthes sensu stricto, and at least once in the other 'monoraphid' genera near the Cymbellales (S5 Fig), supporting hypotheses of Cleve [20] and Mereschkowsky [16]. Phylogenies showing this result have been returned in all analyses that include Achnanthes sensu stricto and other 'monoraphid' taxa [39,40,42,[52][53][54][55]. When considering morphology, the systematic position of Achnanthes sensu stricto is also quite interesting. Cox [51] suggested Achnanthes is closely related to Mastogloia, based on similarities in chloroplast, pore (cribrate), and raphe structure and cite their position in a cladistic analysis of morphology [50]. Our single molecular marker SSU, LSU and rbcL and multimolecular marker analyses do not support a relationship between Achnanthes and Mastogloia, but instead place Achnanthes within the Bacillariales, similar to other molecular studies [42,54,55]. Mereschkowsky [16] showed the chloroplast of Achnanthes sensu stricto to be similar to Hantzschia Grunow [80], a genus within the Bacillariales. Placement of Achnanthes within the Bacillariales is problematic based on morphology, and more extensive taxon sampling in this region of the raphid diatom tree of life may help to resolve the phylogenetic position of this 'monoraphid' genus. Our molecular results, however, support the relationship between Achnanthes and the Bacillariales, but results of hypothesis testing do not rule out the possibility that Achnanthes is related to genera in the Mastogloiales. This appears to be another case, in addition to the relationships of 'monoraphid' diatoms and Rhoicosphenia with the Cymbellales, where molecular data support Mereschkowsky's [16] suggestion of a close relationship between taxa with diverse valve morphologies, based on chloroplast similarities.
Since the description of Rhoicosphenia [2], multiple hypotheses of its phylogenetic position have been made based on valve [2] and chloroplast [16] morphology. Detailed investigations into the valve morphology [43], sexual reproduction [44], relation to other diatom genera [46], and initial cells and size reduction [45,47] were unable to support or reject any of the hypotheses from the past century as summarized in [43], but did support Mann's hypothesis (H 4 ) that Rhoicosphenia belongs in an 'enigmatic' position [45]. Mann presented multiple lines of morphological evidence, without any formal analysis, that support the similarities of Rhoicosphenia to 'monoraphid' diatoms and Gomphonema, but explains their similarities as convergent evolution [43][44][45]. However, he did not question that the specific morphological traits he considers-pore occlusions, shape, heteropolarity, mucilage pads, pseudosepta, copulae, raphe structure and number, etc.-may look similar in different groups due to convergence (they are not homologous) and therefore would not be helpful in building phylogenies [43][44][45].
Based on the concatenated three molecular marker analysis, we suggest that Rhoicosphenia occupies a position basal to the Cymbellales. In terms of diatom classification, with the addition of the genera Geissleria [30,31] and Adlafia, the Order Cymbellales sensu Round are a natural group-interestingly it is noted that Adlafia has a single chloroplast (as Navicula brockmanii Hustedt [81] in [42,55]), similar to the chloroplast structure Mereschkowsky [16] used to unite the Monoplacatae, the group in which he placed members of the Cymbellales and Rhoicosphenia. While our data support Mereschkowsky's Monoplacatae consisting of Heteroideae and Cymbellales, hypothesis testing rejects one specific proposal of Mereschkowsky, that is, the placement of Rhoicosphenia as sister to Gomphonema (Table 4). Our analysis supports the classification of [22] that places Rhoicosphenia in the Cymbellales, but we add phylogenetic structure to this grouping, with Rhoicosphenia in a basal position to the rest of the genera in the order. The order Cymbellales would now include the genera Adlafia, Anomoeoneis, Cymbella, Cymbopleura, Didymosphenia, Encyonema, Encyonopsis, Geissleria, Gomphoneis, Gomphonema, Placoneis, and Reimeria. The relationship between diatoms in the Heteroideae and the Cymbellales (including Rhoicosphenia) could be assigned a Linnaean taxonomic rank of superorder named Cymbellidae that would include Achnanthidiaceae + Cocconeidaceae + Rhoicosphenia + Cymbellales, within the subclass Bacillariophycidae. This superorder would be very similar to Mereschkowsky's Monoplacatae, with the addition of genera that were not yet recognized in the early 20 th century, and would also represent a monophyletic clade in the context of PhyloCode [82]. The Cymbellales would remain an order in our classification, but two unnamed clades between the Order and Superorder ranks would also be recognized, one consisting of Cocconeidaceae + Rhoicosphenia + Cymbellales, the other would consist of Rhoicosphenia + Cymbellales. Additionally, our results support Mereschkowsky [16] and Cox [51] that Achnanthes sensu stricto should not be considered part of a monophyletic clade of 'monoraphid' diatoms, however cannot fully support or reject their specific placements of the genus. Finally, our analyses support Cleve's [20] hypothesis that 'monoraphid' diatoms are polyphyletic. A classification scheme based on our results is presented below.
S2 Table. Sequences included in three-marker concatenated alignment. Taxa and GenBank Accession numbers that were included in three-marker (SSU, LSU, rbcL) alignment. (DOCX) S3 Table. Characters and character states. Characters and character states used in morphological phylogenetic analysis. (PDF) S4 Table. Taxon and character matrix. Taxa and corresponding characters used in morphological phylogenetic analysis. (PDF)