Genome-scale phylogenetic analyses confirm Olpidium as the closest living zoosporic fungus to the non-flagellated, terrestrial fungi

The zoosporic obligate endoparasites, Olpidium, hold a pivotal position to the reconstruction of the flagellum loss in fungi, one of the key morphological transitions associated with the colonization of land by the early fungi. We generated genome and transcriptome data from non-axenic zoospores of Olpidium bornovanus and used a metagenome approach to extract phylogenetically informative fungal markers. Our phylogenetic reconstruction strongly supported Olpidium as the closest zoosporic relative of the non-flagellated terrestrial fungi. Super-alignment analyses resolved Olpidium as sister to the non-flagellated terrestrial fungi, whereas a super-tree approach recovered different placements of Olpidium, but without strong support. Further investigations detected little conflicting signal among the sampled markers but revealed a potential polytomy in early fungal evolution associated with the branching order among Olpidium, Zoopagomycota and Mucoromycota. The branches defining the evolutionary relationships of these lineages were characterized by short branch lengths and low phylogenetic content and received equivocal support for alternative phylogenetic hypotheses from individual markers. These nodes were marked by important morphological innovations, including the transition to hyphal growth and the loss of flagellum, which enabled early fungi to explore new niches and resulted in rapid and temporally concurrent Precambrian diversifications of the ancestors of several phyla of fungi.


Results
Genome assembly and data binning analysis. The raw de novo assembly of transcriptome comprised 94,984 contigs with a total length of 41 megabases (Mb). A large proportion of contigs from this assembly (80%) were less than 1.5 kb in length and not included in the binning analysis using VIZBIN 32 . The subsampled transcriptome assembly from VIZBIN analysis, cluster 1 (Fig. 1), contained 1,946 sequences with a total of 4.4 Mb. The raw de novo assembly of Olpidium bornovanus genome comprised 160,327 contigs with a total length of 187 Mb. After two rounds of binning and conservative sampling of data ( Fig. 1), the working genome assembly contained 13,695 contigs with a total length of 29 Mb.
Phylogenetic reconstruction and dating analysis. The concatenated matrix of the 295-marker dataset comprised 83,461 aligned amino acid sites. The results based on RAxML 33 and IQ-TREE 34 analyses were generally consistent with strong support for the monophyly of fungal phyla and subphyla sampled in this study (bootstrap support 100; Fig. 2 & Table S2; referred as ConcatBestTree). The first split in Kingdom Fungi was inferred between Cryptomycota and the remaining fungal taxa. Blastocladiomycota and Chytridiomycota formed a paraphyletic grade of zoosporic fungi, subtending a clade of Olpidium and the non-flagellated terrestrial lineages. All the splits mentioned above, as well as the Olpidium + non-flagellated clade, received strong support (bootstrap support of 100) from RAxML and IQ-TREE analyses. Olpidium was resolved as the sister www.nature.com/scientificreports/ group of non-flagellated terrestrial fungi with moderate to strong bootstrap support (95 for IQ-TREE analysis and 89 for RAxML analysis on non-partitioned 295-marker dataset; Fig. 2 and Table S2). Within non-flagellated terrestrial fungi, Zoopagomycota was sister to a clade of Mucoromycota and Dikarya, again with moderate to strong bootstrap support (95 for IQ-TREE analysis and 89 for RAxML on non-partitioned 295-marker dataset). Partitioned RAxML analysis on the 295-marker dataset showed weaker support for the aforementioned nodes (66 and 68, respectively). ASTRAL 35 analyses also recovered the clade of Olpidium and the non-flagellated fungi with strong support (1.0 pp and 100 bootstrap support). For the exact placement of Olpidium, ASTRAL analyses showed decreased support for Olpidium as sister to non-flagellated group, and various placements of Olpidium were recovered in the best trees across different ASTRAL analyses (Fig. S1). We further assessed the completeness of the dataset to evaluate whether missing data may impact the phylogenetic reconstruction. All the sampled markers were present in 80% of the sampled genomes, except for one that was found in 75% of all sampled genomes (85 genomes). All the genomes possessed at least 80% of all the 295 markers, except for the genome of Encephalitozoon cuniculi (Cryptomycota) with only 137 sampled markers present. Of all the 295 markers, 283 had broad phylogenetic distributions as defined by being present in Olpidium and at least one species of each of Zoopagomycota, Mucoromycota, Ascomycota and Basidiomycota. Phylogenetic reconstruction based on these 283 markers was topologically similar to the ConcatBestTree, with comparable bootstrap support throughout the tree.
Four of the markers contained identical sequences from two or more different species in the individual alignments after trimming and were excluded in the RAxML inference of individual gene trees and the ASTRAL analyses 36 . The topology recovered in the ASTRAL analysis based on all remaining 291 individual gene trees was mostly consistent with the ConcatBestTree (Fig. 2). One relevant discrepancy, however, was the placement Figure 1. Metagenomic data binning of the transcriptome and genome assemblies of Olpidium bornovanus based on tetramer nucleotide composition using Vizbin. (A) Vizbin analysis on the transcriptome. Contigs shorter than 1.5 kb were removed from the analysis. The ovals delimit the seven data clusters recovered in the Vizbin analysis. (B) Vizbin analysis on the Olpidium-raw-genome assembly. Contigs longer than 1 kb in the genome assembly (blue data points) and contigs from transcriptome cluster 1 (red data points) were included in the analysis. The oval delimits Olpidium-subsample1, the subset of contigs used in the third Vizbin analysis. (C) Vizbin analysis on Olpidium-subsample1 (blue data points). Genome sequences from another seven fungal species were included in the same Vizbin analysis to serve as spike sequences. Contigs from the area that the Olpidium-subsample1 contigs show little overlap with the spike sequences were sampled (Olpidium-subsample1′) and further binned using methods based on sequence similarity (mmseq2 and DIAMOND blastx) to produce the working assembly.  (Fig. 2).
Investigation of the uncertain local placement of Olpidium in fungal tree of life and the branching order between Blastocladiomycota and Chytridiomycota. RAxML analyses with fast-evolving sites removed from the concatenated 295-marker matrix recovered the same topology of [Olpidium, (Zoopagomycota, (Mucoromycota, Dikarya))], with a minor to moderate decrease of branch support compared with analyses included all sites (Table S3). We performed two sets of the Approximately Unbiased (AU) 38 tests, one set on the concatenated matrices of 295 markers and the other set on individual markers (Fig. S1). For the AU tests on individual markers, a total of 283 markers were included from the analysis. Four of the 12 excluded markers contained identical sequences from two or more different species in the trimmed individual alignments. For the other eight excluded markers, one or more constraints used in RAxML analysis could not be applied to due to missing taxa and thus the AU test could not be performed on these markers.
The AU test on the concatenated 295-marker matrix was consistent with the ConcatBestTree being significantly better than the alternative topologies. For the AU tests on individual markers, the ConcatBestTree was not rejected as a significantly worse tree than the best individual gene trees for 236 markers (the ' AU_similar' set of markers), while the best gene trees for 47 markers were significantly better than the ConcatBestTree (the ' AU_better' set of markers) (Fig. S1). The remaining 12 markers were not eligible for AU test due to the presence of missing data. Out of the 236 markers in the AU_similar set, 188 of them showed equivalent support for at least one other alternative topology in addition to the best gene tree and the ConcatBest tree. (Fig. S1). In the quartet mapping analysis on the concatenated 295-marker matrix, 48.2% of the sampled quartet favored the topology of [(Olpidium, Mucoromycota), (Zoopagomycota, Dikarya)], 39.6% favored [(Olpidium, Zoopagomycota), (Mucoromycota, Dikarya)], and the remaining 11.4% favored [(Olpidium, Dikarya), (Mucoromycota, Zoopagomycota)]. In the quartet mapping analysis on individual markers, only one marker had more than 75% of the sampled quartet favored the topology of [(Olpidium, Zoopagomycota), (Mucoromycota, Dikarya)], while all the remaining markers show no or little preferences over any of the three competing topologies (Fig. 3).
To take the uncertainty of individual gene trees into account, we collapsed the weakly supported nodes (bootstrap support < 30) in the ASTRAL polytomy tests. In the polytomy test based on all the 291 individual gene trees, the length of the internode between the divergence of Olpidium and Zoopagomycota was reduced to zero (Fig. S1). The polytomy test based on the 236 markers from the AU_similar set identified a polytomy regarding the relationships among Olpidium, Zoopagomycota, Mucoromycota and Dikarya. For the 47 markers from AU_better set, there were insufficient quartets remaining to perform the polytomy test when the weak nodes in the gene trees (bootstrap support < 30) were collapsed. When the original gene trees were used, the polytomy among Olpidium, Zoopagomycota, Mucoromycota and Dikarya was again recovered.
We also performed AU tests on the individual gene trees on the relationships among Blastocladiomycota, Chytridiomycota and the clade of Olpidium and non-flagellated terrestrial fungi, which were also represented by a polytomy in tests by ASTRAL. We performed AU tests on the individual gene trees regarding the diverging order between Blastocladiomycota and Chytridiomycota. Among the 262 markers that were eligible for AU tests, 212 markers showed equivocal support for the individual best gene tree and for the branching pattern found in the best tree, i.e., [Blastocladiomycota, (Chytridiomycota, (Olpidium, non-flagellated terrestrial fungi))] (referred as Blasto_sister arrangement). The best gene trees of the remaining 50 markers differed from and were significantly more likely than the Blasto_sister arrangement.

Discussion
Generating DNA data for phylogenetic reconstruction for fungi present in complex biological systems. The obligate endoparasitic nature of Olpidium makes it challenging to collect DNA of the quantity and quality necessary for whole-genome sequencing. The fungus produced zoosporangia in the roots of inoculated cucumbers grown in sand. Transferring the roots with zoosporangia to distilled water triggered differentiation and release of zoospores into the water, but it also allowed dispersal of other organisms living within or on the surface of the host plant roots. Hence the genomic data generated from the Olpidium zoospores is a mixture of DNA sequences with different origins. We were able to retrieve Olpidium sequences with strong confidence by applying binning methods based on oligonucleotide compositions by VIZBIN and supplemented by BLAST and mmseq2 searches. While our conservative sampling approach in VIZBIN analysis might lead to the exclusion of some Olpidium sequences from the working assembly, it assures that the sequences included in the working assembly were highly likely from Olpidium. The quality of the working assembly was further boosted by the additional filtering of data with sequence-similarity-based methods of BLAST and mmseq2. After completion of the conservative binning process, the size of the working assembly is only 16% of the raw assembly. However, we managed to retrieve high proportion of the coding regions of the genome. The BUSCO3.0.2 39 score of the working assembly is 75%, based on comparison to fungi_obd9. Out of the 434-marker set that proved useful www.nature.com/scientificreports/ to reconstruct high-level fungal relationships 6,40 , we were able to retrieve 295 of them for use in the subsequent phylogenetic analysis. Generating adequate genomic data from non-culturable fungi represents a challenge in the incorporation of these organisms in phylogenetic analyses, especially for fungi that are microscopic for their entire life histories and/or reside in a complex biological system (e.g., endophytes, endoparasites, soil fungi etc.). Our approach provides a solution to effectively retrieve markers for phylogenomic inference from these problematic taxa without demanding extraordinary measures in acquisition of sequence data. It can also serve as an approach to survey the fungal diversity of complex systems (such as the microbiome of soil samples) and potentially perform functional analyses of these systems, especially with the help of the rapid advancement of binning methods targeting eukaryotic sequences 30 . Olpidium is the sister group to non-flagellated, terrestrial fungi.. Loss of the flagellated stage is considered to be one of the most important morphological transitions in the evolution of fungi and has been interpreted to be associated with the colonization of the terrestrial landscape and rise of the non-flagellated fungi 41,42 . Being a zoospore-producing fungus, Olpidium was long considered a member of Chytridiomycota, but multi-gene studies by James et al. 2 and Sekimoto et al. 18 questioned this hypothesis. These analyses grouped Olpidium within the non-flagellated terrestrial fungi, but taxon sampling was relatively sparse and the exact phylogenetic placement of Olpidium remained unresolved. In this study we densely sampled within 'zygomycete' fungi, i.e., Mucoromycota and Zoopagomycota 2 , aiming to include the closest extant relatives of Olpidium. We also sampled multiple species from each of the individual classes within Dikarya and within Chytridiomycota to capture the deepest nodes, i.e., the MRCAs of all extant phyla and subphyla, within these two groups, allowing us to break potential long branches between these major fungal lineages.
Our analyses based on the super-alignment approach (RAxML and IQ-TREE) inferred Olpidium as more closely related to the non-flagellated terrestrial fungi, consistent with the findings by James et al. 17 and Sekimoto et al. 18 . In contrast to these two studies where Olpidium was found nested within non-flagellated fungi, however, Olpidium was recovered as the sister group to the clade of non-flagellated terrestrial fungi. This topology supports the most parsimonious model of morphological evolution with the production of zoospores through www.nature.com/scientificreports/ zoosporangia being a symplesiomorphic character shared by Olpidium and the 'core chytrids' , and a single loss of the flagellum leading to the MRCA of (Zoopagomycota, (Mucoromycota, Dikyara)), rather than multiple losses among the fungi now classified in Zoopagomycota.
Lack of strong support for recalcitrant nodes along the fungal backbone is due to lack of phylogenetic information. The analyses performed here do not support missing data as being a major contributing factor to the uncertain placement of Olpidium relative to the non-flagellated terrestrial fungi. The removal of fast-evolving sites consistently supported the branching order of [Olpidium, (Zoopagomycota, (Mucoromycota, Dikarya))] with minor to moderate decrease of branch support (Table S3), suggesting that the noise potentially brought in by fast-evolving sites was limited in the inference of the placement of Olpidium. Quartet analysis, AU tests, and direct examination of individual gene trees revealed little conflict among the sampled markers. Only eight markers showed strong support for an alternative placement of Olpidium. The removal of these eight markers from the 295-marker dataset did not improve the support for the placement of Olpidium in the subsequent phylogenetic reconstruction, suggesting that these conflicting signals were not the major contributing factor to the uncertain local placement of Olpidium (Table 1). When individual markers were ranked based on their phylogenetic informativeness (PI) and added to a concatenated dataset in a stepwise fashion from highest to lowest PI, the node supporting Olpidium with the non-flagellated fungi was strongly supported and stably resolved with twenty markers ( Table 1). Resolution of the relationships between Zoopagomycota, Mucoromycota and Dikarya required more than 200 genes, however, and never reached maximum values of support, a finding consistent with limited phylogenetic signal. The equivocal support in AU tests and quartet analyses for three or more competing phylogenetic positions of Olpidium suggests that limited phylogenetic information contained in the markers is the main cause of the lack of strong support for the nodes defining the relationships between Olpidium, Zoopagomycota and Mucoromycota. Similarly, AU tests on the placement of Blastocladiomycota and Chytridiomycota also suggested lack of phylogenetic signals in individual markers. Consistent with the AU tests, the ASTRAL polytomy tests on the super-tree datasets recovered a polytomy for the branching order among Olpidium, Zoopagomycota, Mucoromycota, and Dikarya. These findings are consistent with the branches involved in these ASTRAL polytomies being one-third to one-half the lengths of the flanking backbone branches (Fig. 2). These short branches would be best captured by more quickly evolving genes/sites, however, the ancient age of these evolutionary events further compounds this problem and creates a situation that the genes/sites with the most appropriate substitution rate are also more susceptible to signal decay over time. While these nodes will be difficult to resolve using sequence-based methods, changes in genome structure and content among different lineages may provide information necessary to infer the order of divergence. In addition, increasing taxon sampling, especially those understudied parasitic members from the early-diverging lineages, may shed light on how fungi made the transition to terrestrial ecosystems.
The loss of the flagellum in fungi and the radiation of the early terrestrial fungi. The MRCAs of the Chytridiomycota, Zoopagomycota, Mucoromycota and Dikarya are all estimated to be of similar and overlapping ages (Fig. 2), lending support to the hypothesis of rapid radiation among the early terrestrial fungal groups 18,43 . These findings are consistent with a Precambrian origin of the non-flagellated fungi 6,44,45 , and that they experienced their first radiation in Neoproterozoic (Fig. 2), concurrent with Ediacaran fauna (625-539 Ma) 46 and predating the earliest evidence of land plants in the late Ordovician and Silurian (450-420 Mya) 47 . While animal phyla likely diversified in early marine environments, the loss of flagellum and concurrent ages of the MRCA of Zoopagomycota, Mucoromycota and Dikarya suggest that these lineages likely diversified in early terrestrial environments.
The Neoproterozoic was an era characterized by multiple glaciation events, commonly referred to as the 'snowball' earth 48 . Under this model, the MRCA of the non-flagellated terrestrial lineages and its ancestors must have resided in the microbial biocrust, which has been proposed as the earliest terrestrial ecosystem [49][50][51] . These early terrestrial fungi would have possessed a zoosporic stage after they first moved to the microbial biocrust, and the glaciations, especially the Sturtian glaciation, might have led to the extinction of much of the diversity. The survivors of the glaciation rapidly radiated during the inter-glaciation period or post-glaciation early Ediacaran, which was characterized by increased temperatures, newly created terrestrial niches created by the retreating glaciers, and new food sources by the diversification of other eukaryotes (e.g., protists and algae) [52][53][54] .
The loss of the flagellum might have either occurred during the glaciations, or before fungi began to diversify on land after the glaciation. In either scenario, Olpidium and other zoosporic species retained their flagella and remained dependent on liquid for dispersal. Whereas the remaining terrestrial fungal lineage lost the flagellated zoospore stage and invented means to disperse spores aerially, which freed them from dependence on free water for dispersal. In addition, hyphal growth form of these fungi allowed them to efficiently explore the terrestrial landscape. As a result, these non-flagellated fungi radiated and gave rise to 1.5 million species and became the dominant fungi in terrestrial, plant-based ecosystems.

Methods
Genome and transcriptome sequencing and assembly. Olpidium bornovanus S191 was co-cultured with cucumber roots in sand. At maturity, the zoospores of O. bornovanus were collected by washing the cucumber root with sterile water. The collected zoospores were pelleted at 5000×g for five minutes. Total DNA and RNA were extracted from the pelleted zoospores using Qiagen DNeasy Plant Mini kit and Qiagen RNeasy Plant Mini kit, respectively, following the procedures outlined by the manufacturer.
The Olpidium genome was sequenced using both mate-pair sequencing and standard Illumina pair-end sequencing. For mate-pair sequencing, the amplified paired-end library was size selected to obtain the desired www.nature.com/scientificreports/ fragments (300-600 bp) and sequenced on the Illumina HiSeq2000 sequencer. The paired-end library was sequenced on an Illumina HiSeq2000, following a 2 × 100 indexed run recipes. The Olpidium-raw-genome assembly was generated using MEGAHIT 55 based on raw reads from both the paired end and mate pair sequencing runs.
To prepare the library for transcriptome sequencing, messenger RNA was purified from total RNA using Dyanbeade mRNA isolation kit (Invitrogen) and was fragmented using RNA Fragmentation Reagents (Ambion) with targeting fragments range of 300 bp. Complementary DNA was generated by reverse transcription of the fragmented DNA using SuperScript II Reverse Transcription (Invitrogen). The cDNA library was prepared was sequenced using Illumina HiSeq2000 sequencing, following a 2 × 150 indexed run recipe. The raw RNA-Seq reads were assembled using TRINITY v.2.5.1 56 .
Binning analysis on Olpidium transcriptome and genome assemblies. Olpidium transcriptome contigs longer than 1.5 kb were binned based on tetramer composition of individual contigs using VIZBIN. We identified a total of seven clusters of contigs (Fig. 1A). Contigs in each data cluster were searched against Refseq database 57 and an in-house database composed of 350 fungal proteomes. Cluster 1 (Fig. 1A) was dominated by sequences showing fungal affinities and was used in the subsequent binning analysis on the raw genome assembly of Olpidium (Olpidium-raw-genome).
In the second VIZBIN analysis, we included contigs from the Olpidium-raw-genome assembly and contigs from cluster 1 of the Olpidium transcriptome assembly. We subsampled the contig cluster of Olpidium-rawgenome contigs that overlapped with the majority of transcriptome cluster 1 contigs (dataset Olpidium-sub1; Fig. 1B).
Olpidium-raw-genome contigs longer than 1 kb were searched against the in-house database of fungal proteomes using DIAMOND blastx 58 . The genome sequences of the two fungi with the highest number of hits among Dikarya taxa (Coprinopsis cinereal and Ustilago hordei), together with genome sequences from two Mucoromycota fungi (Mortierella elongata and Umbelopsis ramanniana,), two Zoopagomycota fungi (Basidiobolus meristosporus and Syncephalis plumigaleata), and one Chytridiomycota fungus (Spizellomyces punctatus), together with contigs from Olpidium-sub1, were included in a third VIZBIN binning analysis. Prior to the third binning analysis, the aforementioned seven fungal genome sequences were divided into 3-kb fragments using a custom script. We removed the contigs from Olpidium-subsample1 that clustered with the non-Olpidium reference genome sequences. The remaining Olpidium-subsample1 (dataset Olpidium-subsample1′; Fig. 1C) was used to query the NCBI nr and Mycocosm database using mmseq2 59 . Each contig with the top hit from a non-fungal organism was extracted and compared against an in-house database composed of 350 fungal proteomes using DIAMOND with the blastx option. The contigs without significant fungal hits (1e-30) in the DIAMOND search were removed from Olpidium-subsample1′. The remaining contigs from Olpidium-subsample1′ were used as the working assembly of Olpidium.
Genome annotation. Prediction of protein-coding genes in the working assembly of Olpidium was performed using MAKER pipeline v2.31.8 60 5  10  20  30  40  50  60  70  80  90  110  130  150  170  190  210  230  250  270  290   (O, Z, M, D)  ---81  92  100  100  100  100  100  100  100  100  100  100  100  100  100  100  100   (Z, M, D)  --99  -63  85  96  87  76  52  63  61  67  48  73  67  77  92  93  96   (M, D)  -----84  96  86  76  52  63  61  66  48  72  67  77  93  93  www.nature.com/scientificreports/ Phylogenetic reconstruction and molecular dating analysis. We sampled proteome sequences from a total of 113 species (106 fungi and seven non-fungus outgroups; Table S1) in our phylogenetic reconstruction. We employed 434 protein markers that proved useful for the inference of higher-level phylogenetic relationships among fungi 31,40 (https ://githu b.com/1KFG/Phylo genom ics_HMMs). We used the pipeline PHYling (Stajich JE; http://githu b.com/staji chlab /PHYli ng_unifi ed) to search for the target markers in the sampled proteomes with an e-value of − 20 and performed alignment for each marker. A total of 295 targeted markers were found present in the working assembly of Olpidium and these markers were the main focus in the subsequent analyses. We took three approaches to infer the phylogenetic relationships of the sampled taxa: the super-alignment approaches by using RAxML v.8.0.26 and IQ-TREE on the concatenated alignment, and the super-tree approach by using ASTRAL-III on the individual markers. The RAxML analysis on the concatenated matrix was performed on the CIPRES web portal 66 , using the "-f a" option with 100 bootstrap replicates. We performed non-partitioned analysis with PROTGAMMALG model and partitioned analysis using a partition scheme generated by PartitionFinder2 67 . For IQ-TREE analyses, the best ML tree was based on the non-partitioned 295-marker dataset, using a FreeRate heterogeneity model (LG + F + R10) and using a heterotachy model (LG + F + R10*H4) 68,69 . Branch supports for IQ-TREE analyses were estimated using the ultrafast bootstrap approximation approach with 1000 bootstrap replicates (Hoang et al., 2018). For the ASTRAL analysis, we initially scored the topology obtained in the RAxML analysis of the concatenated alignment (ConcatBestTree) with individual gene trees. We then inferred the best ASTRAL topology based on individual gene trees. To account for the uncertainty in the best gene tree, we collapsed the weakly supported nodes (bootstrap support < 30) in the individual gene trees. We also performed a bootstrap analysis (100 replicates) and generated a greedy consensus of the 100 bootstrap replicates.
Molecular dating analyses were performed on the best RAxML tree based on the 295-marker matrix using MCMCTree of the PAML4.9 package, with only the fungal ingroups included. We used six fossils as the lower minimum bound for the MRCAs of Blastocladiomycota (407 Ma), Chytridiomycota (407 Ma), Endogonaceae (247 Ma), Mucorales (315 Ma), Ascomycota (407 Ma) and Basidiomycota (330 Ma) (Table S4). A truncated Cauchy distribution was applied for each of the calibration points. In addition, we set a loose maximum age constraint on the node of the MRCA of Chytridiomycota, Olpidium and the non-flagellate terrestrial fungi (MRCA-COT) based on expansion of fungal pectinases and the age of pectin-containing streptophytes as discussed in Chang et al. 6 . The recent estimates on the origin of streptophytes range from 750 Ma to about 1100 Ma and the origin of the pectin-containing lineages were around 850 Ma or younger 45,70,71 . We applied 850 Ma as the upper bound of MRCA-COT. We used a relaxed molecular clock with the independent rate model (clock = 2) 72 , and divergence times were estimated using the approximation method 73 . We performed two independent MCMC runs with the following parameters: burn in = 250,000; sampling frequency = 100; number of samples = 20,000. With this setting, the first 250,000 iterations were thus discarded as burn-in, and then the MCMC was run for another 2,000,000 iterations, sampling every 100 iterations. The 20,000 samples were then summarized to estimate mean divergence date and associated 95% credibility intervals.
Investigation of the uncertainty in the local placement of Olpidium in fungal tree of life. To explore the impact of fast-evolving sites on the phylogenetic reconstruction, we used TIGER 1.02 74 to classify all the aligned amino acid sites into ten rate categories. We sequentially removed the sites with faster substitution rates from the matrix and performed tree inference using RAxML with 100 bootstrap replicates.
To investigate whether the lack of strong support for the local placement Olpidium in Kingdom Fungi was due to conflicting signals from individual genes or due to lack of informative sites, we performed the Approximately Unbiased (AU) tests and the polytomy tests implemented in ASTRAL (Fig. 2). We performed AU tests on the concatenated matrix based on the 295-marker dataset and on the individual markers. We identified the alternative placements of Olpidium present in the individual RAxML bootstrap trees based on the 295-dataset and in the ASTRAL analysis. We applied each of the Olpidium placements as topology constraints in RAxML analyses to generate the alternative trees used in AU tests. For AU tests on individual markers, we applied an additional ConcatBestTree constraint with Olpidium being sister to the non-flagellated terrestrial fungi. For each AU test, the site likelihood was calculated using RAxML and the final AU test was performed using CONSEL 75 . The ASTRAL polytomy test was applied to the concatenated matrices and to subsets of the matrices which were generated based on the results of AU tests (Fig. S1). For each dataset we performed one test on the original gene trees and one test on the modified gene trees with their low-supported nodes collapsed (bootstrap value < 30) to account for the uncertainty of individual gene trees.
To further explore the support for the relationships among Olpidium, Zoopagomycota, Mucoromycota and Dikarya, quartet mapping analyses were performed in IQ-TREE using the options -lmclust to define the aforementioned four groups. Additionally, phylogenetic informativeness of the 295 loci was measured using PhyDesign web server 76,77 (http://phyde sign.towns end.yale.edu/). Phylogenetic informativeness (PI) was estimated from the concatenated and partitioned 295-marker dataset and the chronogram generated from MCMCTREE. We ranked the 295 markers according to their phylogenetic informativeness and sequentially added the marker to a series of concatenated matrices according to their PI rank from highest to lowest. We inferred phylogenetic trees based on each of the resulting matrices using IQ-TREE.

Data availability
The whole genome data for Olpidium bornovanus S191 has been deposited at DDBJ/ENA/GenBank under the accession JAEFCI000000000. The version described in this paper is version JAEFCI010000000.