Mitochondrial and Plastid Genomes from Coralline Red Algae Provide Insights into the Incongruent Evolutionary Histories of Organelles

Abstract Mitochondria and plastids are generally uniparentally inherited and have a conserved gene content over hundreds of millions of years, which makes them potentially useful phylogenetic markers. Organelle single gene-based trees have long been the basis for elucidating interspecies relationships that inform taxonomy. More recently, high-throughput genome sequencing has enabled the construction of massive organelle genome databases from diverse eukaryotes, and these have been used to infer species relationships in deep evolutionary time. Here, we test the idea that despite their expected utility, conflicting phylogenetic signal may exist in mitochondrial and plastid genomes from the anciently diverged coralline red algae (Rhodophyta). We generated complete organelle genome data from five coralline red algae (Lithothamnion sp., Neogoniolithon spectabile, Renouxia sp., Rhodogorgon sp., and Synarthrophyton chejuensis) for comparative analysis with existing organelle genome data from two other species (Calliarthron tuberculosum and Sporolithon durum). We find strong evidence for incongruent phylogenetic signal from both organelle genomes that may be explained by incomplete lineage sorting that has maintained anciently derived gene copies or other molecular evolutionary processes such as hybridization or gene flow during the evolutionary history of coralline red algae.


Introduction
Mitochondria and plastids originated from prokaryotes (i.e., aproteobacteria and cyanobacteria, respectively) through independent primary endosymbioses that occurred early in eukaryote evolution, over a billion years ago (Timmis et al. 2004). These prokaryotic symbionts were integrated as intracellular compartments in the eukaryotic host cell through the process of organellogenesis, whereby their genomes were reduced due to outright gene loss or by endosymbiotic gene transfer (EGT) to the nuclear genome (Martin and Herrmann pathways involved in photosynthesis and energy cycles in eukaryotic cells (Martin and Schnarrenberger 1997;Martin and Herrmann 1998;Saraste 1999;Burger et al. 2003;Herrmann 2003). Through a series of endosymbioses involving different host cells, about ten diverse phyla (e.g., Euglenophyta, Heterokontophyta, Haptophyta, Dinozoa) became plastidbearing photosynthetic eukaryotes. The members of most of these lineages share a common evolutionary history even if they contain three genomes from two organelles and one host. Because of a conserved gene content and low mutation rates, when compared with nuclear genes, organelle genome data have been frequently applied to evolutionary studies. High-throughput DNA sequencing has empowered the construction of massive organelle genome databases from diverse eukaryotes that has resulted in many new insights into their phylogenetic relationships and the molecular evolution of the organelle genomes (Smith and Keeling 2015). For example, based on mitochondrial and plastid genome data, red algal phylogenetic relationships have been robustly reconstructed Lee, Cho, et al. 2016;Muñoz-G omez et al. 2017).
Despite the rich phylogenetic information, incongruent phylogenetic relationships between organelle (mitochondria or plastid) and nuclear genes were recently revealed. For example, it was found that the plastid (trnT-trnL, matK) and nuclear (ITS) sequence data from a flowering plant, Pilosella hawkweeds, contain phylogenetic incongruence derived from ancient intergeneric hybridization (Fehrer et al. 2007). An insect, the Hawaiian cricket, also showed incongruence between mitochondrial and nuclear DNA phylogenies (Shaw 2002). Phylogenies of the mitochondrial genome, Y chromosome, and nuclear genome fragments of bears showed incongruence caused by introgression and gene flow (Kutschera et al. 2014;Kumar et al. 2017). Such results have been frequently reported from animal groups (Toews and Brelsford 2012), including a genome-wide analysis of birds that uncovered incomplete lineage sorting (ILS) caused by a rapid radiation, particularly in early-diverged lineages (Jarvis et al 2014). In the red algae (Rhodophyta), phylogenetic incongruence was reported between two early diverged coralline species based on plastid (psbA) and nuclear (18S rRNA) markers (Broom et al. 2008). In the calcified red algal group Corallinophycidae, phylogenetic analyses using molecular markers are necessary because these species display complex morphological diversity including geniculate (branching), nongeniculate (crustose), and rhodolith (free-living) forms ( as well as convergent morphologies within these growth forms; Adey and Macintyre 1973;Bosence 1983;Foster 2001). However, complete organelle genomes have been reported for only two species within this subclass (Janou skovec et al. 2013;Lee, Cho, et al. 2016).
To better understand organelle genome evolution and inspect the phylogenetic signal encoded by mitochondrial and plastid genes, we generated a total of 10 complete organelle genomes (five plastids and five mitochondria) from five coralline species (Lithothamnion sp., Neogoniolithon spectabile, Renouxia sp., Rhodogorgon sp., and Synarthrophyton chejuensis). These data were then compared with existing organelle genomes from two coralline red algae (Calliarthron tuberculosum and Sporolithon durum). These seven genomes encompass the four major orders (Sporolithales, Rhodogorgonales, Hapalidiales, and Corallinales) of the subclass Corallinophycidae (class Florideophyceae; Le Gall and Saunders 2007;Nelson et al. 2015). From these analyses, we report incongruent phylogenetic histories between mitochondrial and plastid genome data between the order Sporolithales and Rhodogorgonales. Because phylogenetic incongruence between mitochondrial and plastid trees and between 18S and 28S rRNA trees was unexpected, we describe here these major trends in the genealogical history of organelle genomes and discuss how to interpret these incongruent phylogenetic signals in the coralline algae.  (Sep. 20, 2015, Song-do beach, Busan, Korea) were collected from the subtidal zone. Genomic DNA was extracted using the DNeasy Plant Mini Kit (Qiagen, Hilden, Germany). These coralline species were identified based on morphological features as well as phylogenetic analyses using the psbA and cox1 genes (supplementary fig. S1, Supplementary Material online). However, three species (Lithothamnion, Renouxia, and Rhodogorgon species) were not resolved at the species-level due to a lack of authentic sequences from type material. The following species were chosen for analysis: a rhodolith-forming taxon (i.e., Lithothamnion sp.), two nongeniculate species (N. spectabile, Syn. chejuensis), and two fleshy species (Renouxia sp. and Rhodogorgon sp.). The coralline taxa were subsampled with care to avoid contamination; therefore, it is unlikely (though not impossible) that the selected samples were contaminated by different species in their natural habitats as may be the case with crustose species where one species may grow on top of another crust. The HiSeq2000 sequencing platform (Illumina, San Diego) was applied to generate genome sequencing data of Syn. chejuensis using 100 bp paired-end sequencing library. Other coralline genome sequencing data were generated using the Ion Torrent PGM platform (Thermo Fisher Scientific, San Francisco, California) with 400 bp-sized sequencing libraries. To check for cross contamination, or mixed samples, several molecular markers (i.e., 18S rRNA, rbcL, and psbA) were identified in the assembled genome data, and their phylogenetic analysis confirmed that there were no mixed coralline species in each data set (i.e., the single, expected marker was found). Organelle genome assemblies and annotations followed Song et al. (2016). The sequenced raw reads were assembled using the CLC Genomics Workbench 5.5.1 (CLC bio, Aarhus, Denmark) and MIRA assemblers. Contigs of organelle genome were sorted by customized Python scripts with a local BLAST program and reassembled to construct consensus genome sequences. Draft organelle genomes were confirmed using the read-mapping method in CLC Genomics Workbench 5.5.1. Gap sequences were verified by PCR-based Sanger sequencing.

Materials and Methods
Organelle gene prediction and annotation were done using Geneious 8.1.2 (Kearse et al. 2012) based on BlastX search results (e-value 1.0e À05 ) with codon table 4 (The Mold, Protozoan, and Coelenterate Mitochondrial Code and the Mycoplasma/Spiroplasma Code) and codon table 11 (Bacterial, Archaeal and Plant Plastid Code). Ribosomal RNAs (rRNAs) and transfer RNAs (tRNAs) were predicted by the web-based programs RNAmmer 1.2 Server and ARAGORN (Laslett and Canback 2004;Lagesen et al. 2007). RNAweasel (http://megasun.bch.umontreal.ca/cgi-bin/ RNAweasel/RNAweaselInterface.pl) was used to predict group II introns. The 18S and 28S rRNA regions were sorted from the genome assembly of each sequencing data using BlastN search (e-value ! 1.e À20 ) and then the full length was confirmed using the web-based RNAmmer 1.2 program (Lagesen et al. 2007).

Comparison of Organelle Genome Structure and Phylogenetic Analysis
Structures of organelle genomes were compared using MAUVE 2.3.1 (Darling et al. 2004) with "default options". To construct maximum likelihood (ML) trees, organelle coding genes were aligned using MAFFT 7.110 under default settings (Katoh and Toh 2008) and ML trees were constructed using IQ tree (Minh et al. 2013;Flouri et al. 2015;Nguyen et al. 2015) with the predicted amino acid sequences. The phylogenetic model was chosen through the model test option (-m TEST), followed by the ML tree search, and ultrafast bootstrapping with 1,000 replications (-bb 1,000). Concatenated alignments of 22 mitochondrial and 195 plastid genes were constructed using a customized Python script and then analyzed with the gene partition information (-q). The approximately unbiased test (AU test) was done by IQ tree with the possible tree topologies (-z topology.treefile -zb 1000 -au). To sort subtopologies from constructed ML trees, the PyCogent python module was used and the results were manually confirmed (https://github.com/pycogent/pycogent). Several representative ML trees were merged by the method of intertwining phylogenetic networks (Schliep et al. 2017) and visualized with the phangorn package in R (https://github.com/ KlausVigo/phangorn).

Results and Discussion
General Features of Coralline Mitochondrial Genomes Mitochondrial genomes (mtDNAs) of five coralline species were assembled using high-throughput sequencing data from Renouxia sp.  Yang et al. 2015). The mtDNAs of most coralline species contain two rRNAs and around 25 protein-coding sequences, except the mtDNA of N. spectabile that showed pseudogenization of several conserved CDSs (e.g., atp8, rpl20, and sdh4). Nineteen to twentyfive tRNAs were commonly found between the secY and atp6 genes as in other red algal mtDNAs (Lee et al. 2015;Yang et al. 2015). All seven coralline species encoded a group II intron-containing tRNA (trnI) between the nad5 and nad4 genes in mtDNA (supplementary table S2, Supplementary Material online).
It is notable that two Rhodogorgonales species (Renouxia sp. and Rhodogorgon sp.) contained additional introns in the cox1 (two introns with intronic orf780 and orf790) and rrl (one intron) regions. The total sequence lengths of these introns were 4,875 bp in Renouxia sp. (cox1: 463 þ 3,794 bp and rrl: 618 bp) and 4,774 bp in Rhodogorgon sp. (cox1: 491 þ 3,658 bp and rrl: 625 bp). These introns and intronic ORFs were one of the major contributors to size variation, together with noncoding regions (Spo. durum: 4,626 bp, Renouxia sp.: 6,139 bp, Rhodogorgon sp.: 6,188 bp, Lithothamnion sp.: 3,260 bp, Syn. chejuensis: 6,050 bp, C. tuberculosum: 4,220 bp, and N. spectabile: 5,552 bp). The homologs of Rhodogorgonales orf780 gene were also found in various eukaryotes (i.e., rhodophytes, Viridiplantae, stramenopiles, fungi, and cryptophytes), but prokaryotic homologs were not identified from the public database (blastp e-value ! 1.e À05 to local RefSeq database; supplementary fig. S3, Supplementary Material online). On the basis of the Conserved Domain search (Marchler-Bauer et al. 2017), most of these homologous genes encode group II intronderived reverse transcriptase domain superfamily members (Intron_maturas2 domain superfamily, cl03174). Homologs of the Rhodogorgonales orf790 gene were also distributed in the mtDNAs of various eukaryotes; however, this gene showed a close relationship to diverse prokaryotic lineages (e.g., Proteobacteria, Bacteroidetes, Cyanobacteria, Chloroflexi, Firmicutes) as well as with plastid-encoded genes (e.g., Viridiplantae, cryptophytes, and euglenophytes; supplementary fig. S4, Supplementary Material online). On the basis of this result, we postulate that orf790 originated from endosymbiotic prokaryotes and then spread into eukaryotic organelle genomes. Only Rhodogorgonales contains the orf790 gene among seven coralline mtDNAs, but this gene is found in other red algae including Ahnfeltia plicata (class Florideophyceae; subclass Ahnfeltiophycidae) and Bangiophyceae species (genus Bangia, Porphyra, and Pyropia). Although most eukaryote copies contain the reverse transcriptase domain superfamilies (RVT_1, cl26764 and RVT_N, cl16337; Marchler-Bauer et al. 2017), these other red algal orf790 homologs did not form a monophyletic group (supplementary fig. S4, Supplementary Material online). The means and timing of spread of orf790 homologs in eukaryotes are still unclear, however, one possible scenario might be due to opportunistic gene transfer from prokaryotes into organelle genomes through a genetic vector (e.g., plasmid). This is because plasmid-mediated horizontal gene transfers have been frequently observed in red algal organelle genomes (Lee, Kim, et al. 2016).
We found, however, an unusual 7 kbp insertion in the Renouxia sp. ptDNA located between the rpl19 and ilvB genes that included several plasmid-mediated gene transfers (six orfs and one pseudogenized orf; supplementary table S3, Supplementary Material online). A similar region was found in the Gracilariales species (subclass Rhodymeniophycidae), which is a distantly related red algal order (Lee, Cho, et al. 2016). These plasmid-related sequences were frequently found in red algal organelle genomes and are likely derived from the plasmid itself or from foreign genetic materials encoded on the plasmid (Lee, Kim, et al. 2016). Conspicuously, there were two novel transferred genes located in the ptDNA of Renouxia sp. One of these hypothetical proteins is clustered with the bacterial cupin domain containing proteins (blastp results; e-value ! 1.e À05 to NCBI nr database) in the ML tree ( is griffithsin (synthetic protein)-like protein that is clustered with the bacterial jacalin-related lectin protein and its homologs (domain code: cl03205) from diverse bacteria including several cyanobacterial species (supplementary fig. S6B, Supplementary Material online). The jacalin-like lectins are sugar-binding protein domains that are mostly found in land plants (Peumans et al. 2001), however, there were no land plant homologs in the blastp search in this study of the griffithsin-like protein (e-value ! 1.e À05 to NCBI nr database). We postulate that these two hypothetical proteins in ptDNA of Renouxia sp. were independently transferred from bacteria and were mediated by red algal plasmids. It is likely that this mobile element (i.e., plasmid) plays a key role in the acquisition of foreign genes (i.e., novel genetic resources) and thereby, organelle genome evolution.

Phylogenetic Analyses of Coralline Species
It has been generally accepted that the order Sporolithales is the earliest branching group of coralline red algae, based on the application of different molecular markers (Le Gall and Saunders 2007;Nelson et al. 2015), as well as the fossil record, although fossils of Rhodogorgonales species are unknown (Aguirre et al. 2000;2010). To study the genealogical history of coralline algae, we compared four ML trees using the complete sequences of concatenated rRNAs (18S þ 28S rRNA), mitochondrial (cMT; 22 mitochondrial genes) and plastid genes (cPT; 195 plastid genes) ( fig. 1). The ML tree of concatenated rRNAs showed two monophyletic clusters ( fig. 1A), one comprised of Sporolithales (Spo. durum) and Rhodogorgonales (Renouxia sp. and Rhodogorogon sp.) with moderate support (BS: 81%), and the other of Hapalidiales (Lithothamnion sp. and S. chejuensis) and Corallinales (C. tuberculosum and N. spectabile) with strong support (BS: 100%). To compare the genealogical histories of these coralline nuclear rRNAs, we constructed ML trees of each rRNA data set, and recovered two different tree topologies (supplementary fig. S7, Supplementary Material online). The ML tree of 18S rRNA showed that the early divergence of Sporolithales was followed by the Rhodogorgonales, with a monophyletic cluster formed by Hapalidiales and Corallinales. It is worth noting that the bootstrap supporting (BS) value for the divergence point of the Rhodogorgonales was relatively low (BS: 51%; supplementary fig. S7A, Supplementary Material online). In contrast, the ML tree of 28S rRNA (i.e., a member of the same operon) showed an identical topology to the concatenated rRNA tree (i.e., the monophyly of the Sporolithales and Rhodogorgonales with moderate support, BS: 83%; supplementary fig. S7B, Supplementary Material online). Interestingly, these two classes of conflicting topologies were recovered when we used mitochondrial ( fig. 1B) and plastid ( fig. 1C) genome data (all branches BS ! 90%).
To address this conflict, we analyzed individual gene data sets from each organelle genome (mtDNA 22 and ptDNA 195 genes, respectively). We determined which individual gene supports (or rejects) these two alternative topologies being addressed using the AU test. One mitochondrial and 18 plastid gene data significantly rejected the mitochondrial tree topology (P-value 0.05; supplementary table S4, Supplementary Material online), suggesting that 8.7% of organelle genes have strong conflicts with regard to the basal position of Spo. durum (i.e., mitochondrial tree topology).

Analysis of Individual Gene Phylogenies in Coralline Organelle Genomes
To identify the phylogenetic signal that underlies the conflict in tree topologies, we compared the terminal branching patterns from all ML trees of individual mitochondrial and plastid genes to the concatenated mtDNA and ptDNA (cMT and cPT) topologies. We counted all observed gene numbers that show the monophyly (BS ! 50%) of target species to other species or to the group of species at each divergence point (supplementary figs. S8 and S9, Supplementary Material online). For example, the five genes nad5 (BS 99%), rps3 (BS 99%), nad4 (BS 98%), rpl16 (BS 91%), and nad1 (BS 72%) all showed a monophyletic relationship between Spo. durum and the other six coralline species ( fig. 2A-i). Additionally, the atp8 gene phylogeny supported this topology (BS 86%), but the atp8 gene of N. spectabile was absent due to sequence degradation (low similarity of 428 bp of intergenic sequences between two conserved flanking genes) in the mtDNA (supplementary fig. S8A, Supplementary Material online). A total of six mitochondrial gene phylogenies (atp8, nad5, rps3, nad4, rpl16, and nad1) supported the early diver- identical topology patterns within the cMT or cPT phylogeny, we postulate that the mitochondrial genes of these seven corallines contained both classes of genetic information that support the conflicting topologies: that is, the early divergence of Sporolithales (e.g., cMT topology), and the early divergence of the clades Sporolithales þ Rhodogorgonales and Corallinales þ Hapalidiales (e.g., cPT topology).
The One was the early divergence of the Sporolithales as shown in the cMT topology that was supported by 44 plastid genes; however, if different subtopology patterns were excluded, only five plastid gene phylogenies were identical to the cMT phylogeny (BS in all branches ! 50%; supplementary table S5, Supplementary Material online). The other was the monophyly of Spo. durum and the Hapalidiales-Corallinales clade that was supported by 27 plastid genes. Among these genes, one additional tree pattern encompassing 10 plastid gene phylogenies supported the early divergence of Rhodogorgonales, followed by the divergence of Spo. durum, and then by the monophyly of Hapalidiales and of Corallinales (BS in all branches ! 50%; supplementary table S5, Supplementary Material online).
The monophyletic orders of the Rhodogorgonales (170 genes; fig. 2B-ii), Hapalidiales (103 genes; fig. 2B-iii) and Corallinales (88 genes; fig. 2B-iv) were well-supported, among 182 plastid genes ( fig. 2B and supplementary  fig. S9H, Supplementary Material online), although there were frequent incongruent topology patterns in the Corallinales (Calliarthron and Neogoniolithon) that was found using mitochondrial genes (supplementary fig.  S9B-H, Supplementary Material online). These incongruences were already reported from previous studies including a larger coralline taxon sample using one nuclear and three plastid markers (18S rRNA, psaA, psbA and rbcL; Nelson et al. 2015). In summary, we found three major phylogenetic tree topology patterns among coralline species using individual gene analysis. In addition, we analyzed subsampled data set-based phylogenies of mitochondria and plastid genes using the TIGER program, that progressively excludes highly variable sites from the alignments (Cummins and McInerney 2011). We constructed subsampled data sets from concatenated mitochondria and plastid alignments with high-level gradient (option: -b 50), and then generated the phylogeny using IQ-tree. The proportions of maximum subsampled data set were 53% (plastid) and 60% (mitochondria) of alignments, which indicates $40% of highly variable amino acid sites (supplementary fig. S10, Supplementary Material online). When we excluded these highly variable sites, the tree topologies were unchanged both in mitochondrial and plastid gene analyses, however support values gradually decreased only in node "b" (i.e., monophyly of all coralline species except Spo. durum; supplementary fig. S10A, Supplementary Material online). Similar trends were found in the plastid data sets. Tree topologies were not changed until 38% of the original data set was used. Bootstrap support of node "b" (i.e., monophyly of Spo. durum and two Rhodogorgonales species) gradually decreased (see supplementary fig. S10B, Supplementary Material online). When we used this subsampling strategy on individual genes, most nodes collapsed with very short branches, likely caused by insufficient phylogenetic signal (results not shown). Thus, we found that mitochondrial and plastid tree topologies are well conserved across most nodes when using conserved or variable sequences. However, two particular nodes (see above) are supported only by highly variable sequences.

Three Evolutionary Scenarios of Coralline Organelle Genomes
On the basis of these tree topologies, we propose three alternate evolutionary scenarios to summarize coralline organelle genome evolution: 1) the Sporolithales diverged first (Sporofirst), 2) the cluster Sporolithales-Rhodogorgonales diverged first (Sporo-Rhodo-first), and 3) the Rhodogorgonales diverged first (Rhodo-first; fig. 3A). These scenarios are reflected in the fossil record in which the Sporolithales diverged earlier than the Hapalidiales and Corallinales (Aguirre et al. 2000(Aguirre et al. , 2010, although fossils of Rhodogorgonales have not yet been found because thalli in this lineage are less extensively calcified. Although these three evolutionary scenarios were supported by only circa 10% of genes from two organelle genome data (mtDNA: three genes and ptDNA: 32 genes; fig. 3B), we suggest that these complex evolutionary histories could be explained by ILS, in particular among the ancestors of Sporolithales and Rhodogorgonales. By this we mean that individual gene trees conflict with the overall genome tree because some alleles, surprisingly, failed to coalesce during the several hundred million years of coralline algal evolution. To identify potential ILS-impacted organelle genes, we focused on the early diverged taxa when compared with the outgroup (BS value in first branch ! 50%; fig. 3C and supplementary table S6, Supplementary Material online) regardless of other internal relationships (i.e., ignoring potential independent gene mutations in ingroup taxa). Almost one-half of each organelle genome (mtDNA: 10 genes and ptDNA: 120 genes) is involved in the conflicting phylogenetic signal in the early diverging branches, particularly in the plastid genome data ( fig. 3C). To determine whether natural selection played a (dominant) role in the spectrum of genes putatively impacted by ILS, we compared the functional categories of these genes that support different topologies. This analysis provided no obvious evidence of correlation between gene function and the three evolutionary scenarios (Sporo-Rhodo-first/ Sporo-first/Rhodo-first; supplementary table S6, Supplementary Material online) including ribosomal proteins (5/9/11 genes), photosystem related proteins (0/7/3 genes) and cytochrome related proteins (1/4/2 genes). In addition, each gene category was not highly clustered or concentrated in a specific region of organelle genomes (supplementary fig. S11, Supplementary Material online). On the basis of these comparisons, it is likely that there was no selection acting on the retention of ancestral polymorphisms with regard to gene function or genome structure in the different evolutionary histories of coralline organelle genomes (see test of diversifying selection below).
To compare phylogenetic relationships of coralline species under the three different evolutionary scenarios, we constructed concatenated ML trees using each differently categorized organelle genes: six mitochondrial genes for Sporo-first, three mitochondrial genes for Sporo-Rhodo-first ( fig. 4A), 46 plastid genes for Sporo-first, 28 plastid genes for Sporo-Rhodofirst, and 46 plastid genes for Rhodo-first scenarios ( fig. 4B)  spectabile. The monophyletic relationship of Hapalidiales and Corallinales was recovered in our study consistent with the fossil record (Aguirre et al. 2000(Aguirre et al. , 2010. However, Sporolithales and Rhodogorgonales contained at least 2-3 different evolutionary histories in their organelle genomes. To test whether these genes had undergone different selective pressures with respect to sequence evolution under the different evolutionary scenarios, we analyzed using the PAML package (alignment-based pairwise comparisons; runmode ¼ -2; Yang 2007) nonsynonymous substitution rates (dN) of these genes between coralline species and the outgroup Palmaria palmata (supplementary table S7  In addition, there were no significant differences between Sporolithales and Rhodogorgonales species (P-value > 0.1; Wilcoxon rank sum test; supplementary table S8, Supplementary Material online). On the basis of these results, it is hard to determine if these conflicts are derived from atypical mutation patterns in some organellar genes or in particular species. Therefore, we postulate that the incongruent topologies between the Sporolithales and Rhodogorgonales (e.g., fig. 1B and C) could be explained by ILS in the stem lineages of corallines resulting from a rapid radiation (i.e., fig. 3C) of taxa that contained many ancestral polymorphisms (supplementary fig. S10, Supplementary Material online). The rapid radiation of Florideophyceae, including the Corallinophycidae has already been reported (Lee, Cho, et al. 2016). In angiosperms, chloroplast phylogenomic analysis of 53 grape species shows incongruent phylogenetic relationships that were explained by both hybridizations and a rapid radiation (Wen et al. 2018). Phylogenomic analysis of the sunflower Espeletia using a large taxon sampling (41 species) also showed explosive adaptive radiation-derived ILS (Pouchon et al. 2018). A similar case revealed that the nuclear and mitochondrial phylogenies showed conflicts due to the ILS and introgression within the bear lineage (Kutschera et al. 2014;Kumar et al. 2017). A genome-wide analysis uncovered ILS in modern birds caused by a rapid radiation, particularly among early-diverged species (Jarvis et al. 2014). Another possible explanation is that the complex evolutionary history of coralline organelle genomes reflects gene flow by natural hybridization between ancestral coralline species.
Because such phylogenetic incongruence was also found in two nuclear rRNA markers (i.e., 18S and 28S rRNAs; fig. 1A and B), we presume that the coralline nuclear genomes may also contain a complex evolutionary history. Complex hybridization or ILS, as well as a high level of divergence (homoplasy, i.e., highly variable sequences) could explain these incongruences. For instance, potential hybridization has been suggested for the coralline genus Chiharaea (Corallinales) based on phylogenetic analyses using the nuclear, mitochondrial, and plastid markers (ITS, COI, rbcL, and psbA; Hind and Saunders 2013). However, to test the hypothesis of rapid radiation-derived ILS in the coralline algae, additional taxa need to be studied with their nuclear genomes.
To further study ILS in corallines (i.e., beyond incongruent phylogenies) we merged phylogenetic trees into consensus networks using the intertwining phylogenetic tree method ( fig. 5A; the cutoff value of proportion from present topology patterns ¼ 0.2; Schliep et al. 2017). For this approach, we only used reliable individual gene trees (all branches BS ! 50% with all taxa; supplementary table S9, Supplementary Material online) to minimize error from unresolved phylogenetic nodes, missing taxa, and a high divergence level. The intertwining phylogenetic trees show all possible wellsupported phylogenetic variations among coralline organelle genomes (grey color in fig. 5A). It would be useful to document such complex phylogenetic relationships including some cryptic relationships. Although we cannot clearly establish "what is the first diverged coralline order within the subclass Corallinophycidae?" due to the cryptic relationship between these early diverged coralline orders, we nevertheless postulate that, from the ancestral divergence point of view regarding coralline species, ILS likely contributed to the different phylogenetic patterns ( fig. 5B) and generated the conflicts between mtDNA and ptDNA trees and perhaps the intergenic features within each organelle.

Conclusion
This study investigated the major trends in organelle gene phylogeny among coralline species. At least 600 million years-old (Doushantuo formation; Xiao et al. 1998;Xiao et al. 2004;Condon et al. 2005), the Corallinophycidae is a florideophycean lineage without any report of prominent organelle horizontal gene transfers, including replacement of conserved genes. Therefore, organelle genealogical histories are expected to be consistent. However, a consensus evolutionary history between Sporolithales and Rhodogorgonales using mitochondrial and plastid genome data was difficult to find. There were at least two or three different evolutionary histories apparent in the two organelle genomes, likely caused by ILS. On the basis of the analyses of individual gene trees in coralline species, we found several incongruences in terminal nodes (i.e., specieslevel), and this might be the case when the ancestors of stem groups (i.e., ordinal-level) diverge into different lineages. Because ILS-derived phylogenetic incongruence is generally understood to be derived from recent rapid radiations, natural hybridization, and introgression (Shaw 2002;Fehrer et al. 2007;Toews and Brelsford 2012;Jarvis et al. 2014;Kumar et al. 2017), we postulate that the individual genes of organelle genomes in coralline red algae underwent a complex evolutionary history independently, but in deep time. Regardless of how it happened, our work demonstrates that care must be taken when analyzing phylogenies based on mitochondrial and plastid markers. If complete nuclear genome data from coralline species with a broad taxon sampling were to become available, these may provide a more detailed understanding of the evolutionary patterns revealed in our study. It is also important to determine if ILS (or other sources of phylogenetic conflict) are widespread among other red algae and therefore poses a potentially significant hurdle to the use of organelle genomebased phylogenies in this phylum.

Supplementary Material
Supplementary data are available at Genome Biology and Evolution online.