Chromosome-scale genome assembly for the duckweed Spirodela intermedia, integrating cytogenetic maps, PacBio and Oxford Nanopore libraries

Duckweeds are small, free-floating, morphologically highly reduced organisms belonging to the monocot order Alismatales. They display the most rapid growth among flowering plants, vary ~ 14-fold in genome size and comprise five genera. Spirodela is the phylogenetically oldest genus with only two mainly asexually propagating species: S. polyrhiza (2n = 40; 160 Mbp/1C) and S. intermedia (2n = 36; 160 Mbp/1C). This study combined comparative cytogenetics and de novo genome assembly based on PacBio, Illumina and Oxford Nanopore (ON) reads to obtain the first genome reference for S. intermedia and to compare its genomic features with those of the sister species S. polyrhiza. Both species’ genomes revealed little more than 20,000 putative protein-coding genes, very low rDNA copy numbers and a low amount of repetitive sequences, mainly Ty3/gypsy retroelements. The detection of a few new small chromosome rearrangements between both Spirodela species refined the karyotype and the chromosomal sequence assignment for S. intermedia.


Results
The cytogenetic map for S. intermedia clone 7747. Previously we established a cytogenetic map for S. intermedia clone 8410 using 93 BACs anchored in the S. polyrhiza genome 29 . Now we hybridized diagnostic BACs to the chromosomes of the S. intermedia clone 7747 in order to test whether chromosomal differences occurred between the karyotypes of these two S. intermedia clones, because different chromosome numbers (2n = 36 versus 2n = 20) have been reported for S. intermedia clone 7747 30,31 . Our current chromosome counting found no chromosome number difference between clones 7747 and 8410. For both clones 2n = 36 were counted 18 (Fig. S1). In order to test whether structural chromosomal rearrangements occurred between these S. intermedia clones, we applied cross-FISH with suitable combinations of 93 S. polyrhiza BACs, as described in Hoang & Schubert, 2017 29 , on mitotic spreads of S. intermedia clone 7747. We used the same chromosome designation as in Hoang & Schubert, 2017 29 : ChrSp for chromosomes of S. polyrhiza and ChrSi for chromosomes of S. intermedia. No chromosomal rearrangements between these two S. intermedia clones were detected. Six 29 , were found also for clone 7747 (Fig. 1). The similarity between the cytogenetic maps of S. intermedia clones 7747 and 8410 enabled merging of genome assembly from PacBio reads for clone 7747 and of Illumina and ON reads for clone 8410, and yielded an apparently identical genome map for both S. intermedia clones.
Genome assembly for S. intermedia clone 7747 based on the library of PacBio sequence reads. Two rounds of PacBio-sequencing of a 20 kb library of genomic DNA of S. intermedia clone 7747 resulted in 149 Gbp of raw read data. After an initial filtering for potential bacterial contamination, reads of at least 500 nucleotides were assembled using the Canu pipeline v. 1.5 32 .
A total of 1,305,064 reads were assembled into 1172 sequence contigs of 147,613,042 nucleotides, corresponding 91.7% of the estimated genome size. All contigs of this draft assembly are covered in median 37.5-fold by raw reads. In a first round of scaffolding, the two genomes of the sister species S. polyrhiza (from clones 9505 and 7498) 23,25 were used as references to order contigs as described 33 . The resulting scaffolds (N50 = 1,7 Mbp) were super-scaffolded by SSPACE-Longread v.1-1 34 and assigned to the 18 chromosomes of S. intermedia, using 93 S. polyrhiza BACs as landmarks which were cross-hybridized to the S. intermedia chromosomes of clone 8410 29 . In addition to confirmation of the same linkage relationship in clone 7747 (Fig. 1), new cytogenetic probes using BACs from the genomic regions of interest were designed for FISH experiments to approve localization of the contigs within the pseudomolecules, to resolve mis-assemblies and/or to confirm new linkages (see below). Furthermore, contiguity of the assembly was confirmed by corrected ON reads of S. intermedia clone 8410 using minimap2 v2. 16 35 (see below).
After reiterative rounds of manual curation and validation by FISH, in the final genome assembly, 18 scaffolds (N50 = 8.3 Mbp) of in total 131.4 Mbp (82.2% of the estimated genome size) could be assigned to the 18 chromosomes (Table 1). Six of them show telomeric sequences at both ends and seven at least on one end (see Fig. 2). Most of the shorter and/or repetitive sequences (16.2 Mbp, corresponding 10.1% of the estimated genome size) (N50 = 27.1 Kbp) could not yet be assigned and were considered as additional pseudomolecule "SiUn".
Genome assembly for S. intermedia clone 8410 based on Illumina and Oxford Nanopore reads. The ON/Illumina-derived scaffolds for clone 8410 were created using the MaSuRCA assembler and filtered using minimap2 to remove duplicated sequences derived from heterozygous regions. To form pseudomolecules, all 70 remaining scaffolds were corrected and ordered by manual curation using the assembly of clone 7747 as a reference. Finally, all 18 pseudomolecules ended at both sites with telomeric sequences. Merging of both assemblies revealed further small rearrangements between the karyotypes of S. polyrhiza and S. intermedia (see below and Figs. 2 and 3). The quality of both S. intermedia genome assemblies was assessed by the Gene prediction. Based on similarity to nine aquatic and non-aquatic angiosperm reference genomes, including two duckweed species, S. polyrhiza 7498 v3.1 25 and Lemna minor 5500 38 , gene model prediction via Gene Model Mapper-GeMoMa 39 suggested in total 22,245 (Pacbio, clone 7747) or 21,594 (ON/Illumina, clone 8410) protein/RNA-coding genes, some more than predicted for S. polyrhiza 25 (Table S1).
A total number of 16,162 genes of clone 7747, 16,493 of clone 8410 and 11,327 of S. polyrhiza clone 9509 are coinciding with eggNOGs (Non-supervised Orthologous Groups). Comparing the proportion of eggNOG functional categories between the genomes of the two Spirodela species, the differences were < 1%. Only the category 'Energy production and conversion' , is overrepresented in S. intermedia clone 7747 (4.2 versus 3.1% in clone 8410 and 2.6% in S. polyrhiza) and the category 'Replication, recombination and repair' in clone 8410 (5.3% versus 3.2% in clone 7747 and 2.3% in S. polyrhiza) (Fig. S2, Table S2). www.nature.com/scientificreports/  Enumeration is as in Hoang & Schubert (2017) 29 and in Table S3 (Short sequences present in S. p., but not chromosomally assigned in S. i. correspond to "SiUn" and are not considered). 13 ChrSi of the 7747 assembly show telomeric sequences at one (*) or both ends (**), while all pseudomolecules of the 8410 assembly show them at both ends. www.nature.com/scientificreports/ New linkages in S. intermedia as revealed by genome assembly and FISH. The PacBio assembly for clone 7747 indicated a new rearrangement involving ChrSp20. This chromosome did not form the entire ChrSi18 as reported 29 , instead it is split into two parts, the largest part corresponding to ChrSi18 and two rather small regions (between ChrSp20 3.72-3.78 and 3.80-3.96 Mbp) transferred to ChrSi16 (42,000-175,000 bp) (Table S3). Therefore, the previously not tested BAC 013O04 belonging ChrSp20 was selected for mcFISH experiments. The presence of this BAC sequence in S. polyrhiza ChrSp20 was confirmed by FISH (Fig. 4a). FISH results on S. intermedia chromosomes (clone 8410: Fig. 4b, and clone 7747: Fig. 4c) showed that BAC 013O04 (yellow) labeled the same chromosome (ChrSi16) as the BACs belonging to ChrSp15 (red), while the remaining part of ChrSp20 (green) labeled another chromosome pair (ChrSi18). This result confirmed that ChrSp20 became split and a very small part became translocated to ChrSp15, forming ChrSi16 (Fig. 2), as inferred by PacBio assembly. Because only 76 kbp of the BAC 013O04 sequence appeared in the ON/Illumina assembly of ChrSi16, this link is not visible in Fig. 3 where the entire chromosome ChrSp20 is represented by the S. intermedia chromosome ChrSi18, based on the assembly results for clone 8410. Furthermore, a piece of ChrSp13 (0.46-0.68 Mpbs), according to the ON assembly for ChrSp13 of S. polyrhiza clone 9509, became integrated into ChrSi02 (5. 43-5.66 Mbps) as suggested by the assemblies for both S. intermedia clones (for clone 8410 see Fig. 3). However, three BACs from this region (029A10, 028I16 and 037J09, together 649,507 bp) used as FISH probe, labeled ChrSp02 of S. polyrhiza clones 7498 and 9509 (Fig. 5) and appear on ChrSp02 also in the Bionano map (CP019095.1) of S. polyrhiza clone 9509 25 as well as in pseudomolecule 1 (corresponding ChrSp02) of S. polyrhiza clone 7498 23 (Table S4). This uncovers a hitherto overlooked error in the ON assembly for S. polyrhiza clone 9509 24 . Another new small region, for which no BAC is available, www.nature.com/scientificreports/  www.nature.com/scientificreports/ became transferred from ChrSp14 (4.9-5.4 Mbps) to ChrSi04 (9.5-9.9 Mbps) and appeared in both assemblies (for clone 8410 see Fig. 3). Although only one BAC from ChrSp16 was tested previously 29 , the assemblies from PacBio and ON reads suggests that the entire ChrSp16 is included in S. intermedia ChrSi11 together with a part of ChrSp10 (Figs. 2, 3).

Types, abundance and distribution of repetitive elements. Characterization of S. intermedia repet-
itive sequences was first performed by analyzing unassembled Illumina reads from the clone 8410 using the RepeatExplorer pipeline 40 . This analysis served as a control for repeat quantification in the assembled pseudomolecules that may be biased due to the exclusion of satellite DNA or other repeats that are hard to assemble. In addition, the RepeatExplorer output was used to compile a reference database of S. intermedia repeats that was used to annotate the genome assemblies.
The RepeatExplorer analysis of 2.4 million paired-end Illumina reads (2.25 × genome coverage) revealed relatively small proportions of highly and moderately repeated sequences in the S. intermedia genome ( Table 2). The repeats accounted for 20% of the genome, with LTR-retrotransposons representing the most abundant repeat class (14.1% of the genome, with Ty3/gypsy to Ty1/copia ratio of 2.17). Other repeats including LINEs, rDNA and satellite DNA made up only minor genome proportions.
Repeat annotations of the genome assemblies using the RepeatExplorer reference database resulted in 18.5% and 20.2% of the 8410 and 7747 sequences marked as repetitive, respectively. These proportions correspond to the estimate obtained for unassembled reads (20.0%), suggesting that repeats were not significantly depleted in the final assemblies. Inspection of individual repeat categories revealed that partial depletion resulting in smaller proportions in the assembled genomes occurred for satellite repeats in both clones and for rDNA in the clone 7747 (Table 2). To obtain an alternative estimate of repeat proportions, we also analyzed both assemblies using RepeatScout 41 that performs de-novo identification of repetitive elements based on the high frequency k-mers. This program estimated total repeat proportions of 23.1% (8410) and 25.6% (7747), most likely due to its better sensitivity for low-copy repeats ( Table 2).
An interesting observation was the relatively high abundance of simple-sequence repeats, especially the microsatellite motif (AG) n , that was revealed by Tandem Repeats Finder 42 analysis within the assembled genomes and the Illumina reads. About 37,000 loci of (AG) n with an average length of 38 bp were detected in 8410, making up 1% of the assembly (the same proportion and characteristics were found for (AG) n in the 7747 assembly). These dispersed simple repeat loci appeared in the S. intermedia genome with an average frequency of one per 3.7 kb.
Characterization of rDNA loci. Chromosome analysis of S. intermedia by FISH using 5S and (18S + 26S) rDNA probes, detected one major locus of 5S rDNA on ChrSi15 (corresponding to ChrSp13) (Fig. S3a) and a major locus of 45S rDNA on ChrSi01 (corresponding to ChrSp01) (Fig. S3b). The chromosome-scale assembly and ON reads for Si8410 confirmed the presence of a 5S rDNA locus on ChrSi15, and revealed an additional www.nature.com/scientificreports/ locus on ChrSi14. The extra-long ON reads showed that the locus on ChrSi15 contains a cluster of thirty-one 5S rDNA repeats, whereas the locus on ChrSi14 is composed of seven and 13 5S rDNA repeats interrupted by two 6 kb long repeats of another sequence (Fig. 6b). The chromosome assembly, based on PacBio sequences, showed similar arrangement of 5S rDNA in Si7747 with slightly different copy numbers in each of the loci. Quantitative PCR estimation of 5S rDNA copy number in Si7747 and Si8410 genomes supports this data with 57 ± 10 copies of 5S rRNA genes for Si8410 and, 70 ± 20 copies for Si7747. Sequence alignments suggested that the 5SrDNA units of the two loci contain slightly different non-transcribed spacers (NTS). This was confirmed by sequencing of the 5S rDNA repeats, amplified from genomic DNA of clones 8410 and 7747. Analysis of individual 5S rDNA clones of Si8410 and Si7747 identified the 119 bp long 5S rRNA genes with 100% identity to the previously sequenced 5S rRNA gene of S. polyrhiza 43 , and two variants of the NTS for each ecotype (Fig. S4), with variant 8410-1 corresponding to 5S rDNA locus on chromosome ChrSi15 and the variant 8410-2 corresponding to the locus on chromosome ChrSi14 (Fig. 6a).
In agreement with the FISH data, the genome assembly revealed a 45S rDNA locus in a distal position on ChrSi01 for both clones, Si7747 and Si8410. Extra-long ON reads confirmed this locus for Si8410, about 83 kb upstream of one ChrSi01 chromosome telomere. Moreover, ON reads revealed an additional cluster of 45S rDNA repeats about 8 kb upstream the telomere on ChrSi06. Unlike the situation with 5S rDNA loci, there are no readthrough sequences containing the whole 45S rDNA loci. Therefore, it is not possible to determine exactly how many of the 93 ± 25 45S rDNA copies estimated by qPCR for Si8410, are located in the minor locus.

Discussion
Chromosome rearrangements between the two Spirodela species. Previously, chromosome homeology and rearrangements between S. polyrhiza and S. intermedia were uncovered by comparative serial multicolor cross-hybridization to the S. intermedia clone 8410 29 . Here, we applied the same approach, using informative BACs out of the 96 ones anchored in the S. polyrhiza genome, and found the same chromosomal rearrangements in S. intermedia clone 7747 as reported for clone 8410 29 . Reiterative rounds of genome assembly and validation by FISH revealed a new linkage compared to the S. polyrhiza genome, in addition to the eight translocations detected previously 29 . This new linkage involves ChrSi16, which received small parts from one end of ChrSp20 (Fig. 2). A small piece, apparently transferred from one end of ChrSp13 to ChrSi02, turned out to be an error within the previous ON assembly for ChrSp13 of clone 9509 24 . Another small region, transferred from ChrSp14 to ChrSi04 was not studied by FISH, but it appeared in both assemblies. The S. intermedia chromosomes ChrSi03, 04, 06, 07, 09 and 11, previously found to be involved in evolutionary rearrangements, correspond in total to 10 S. polyrhiza chromosomes (Figs. 1, 2, 3 and 29 ). Now, at least two more rearrangements (in addition to the previously postulated one inversion and eight translocations) have to be assumed if the S. polyrhiza karyotype is more similar to that of the ancestor. Alternatively, eight (instead of six) translocations (involving ChrSi03, 04, 06, 07, 08, 11,12,16,18), and one fission (ChrSi09) 29 were required if the S. intermedia karyotype is the more ancestral one. Additionally, we found a positional change of the smaller 5S rDNA locus to ChrSi14 (instead of ChrSi07 which harbors sequences adjacent to the minor 5S rDNA locus on ChrSp06).
The increased number of rearrangements refined our knowledge about karyotype evolution between S. polyrhiza and S. intermedia and made the corresponding genome assembly for both S. intermedia clones robust by www.nature.com/scientificreports/ independent confirmation. These results demonstrate the benefit of our novel approach, combining comparative cytogenomics and hybrid sequencing technologies, especially for species for which genetic maps are not available.
Genome assembly for both S. intermedia clones. The assembly of PacBio reads for clone 7747 is more fragmented than the assembly of ON/Illumina reads for clone 8410. It displays a higher percentage of ambiguous bases (0.003%) and a lower coverage (Table 1). Several draft contigs representing contamination from plastids, mitochondria, and a virus (Human alpha herpesvirus 1 strain) were detected which show largely increased coverages of 2925-, 17,126-and 73-fold respectively. These overrepresented sequences caused a diminished overall coverage in both assemblies. Canu performs best with more than 50X coverage, while lower values might decrease the contiguity of an assembly 32 .
Comparing read length N50 of the two S. intermedia clones, the ON library of clone 8410 (16,322 kb) yielded a larger amount of long reads than clone 7747 with an N50 of 9879 kb. As a consequence, longer repeats may not be sequenced completely, leading to a more fragmented assembly 44 , as in case of clone 7747.
In summary, the PacBio read assembly of S.intermedia 7747 provides continuous pseudomolecules but these might comprise ambiguous and low coverage nucleotide stretches, and possibly missing sequences, as described for other plant species 45 . Despite the lower library quality for clone 7747, the non-hybrid PacBio assembly shows (as expected 46 ) good results for the scaffold N50 (8.3 Mbp similar to 9.25 of the ON/Illumina assembly, Table 1), depicting the synergy of cytogenetics and long read sequencing technologies.
The distribution of functional eggNOG groups within the S. intermedia clones and S. polyrhiza is largely balanced, indicating a high similarity in overall functional gene content between both species which amounts with ~ 20,000 genes, a rather low number compared to most land plants. The BUSCO results reflect the increased fragmentation of the 7747 assembly by an increased number of fragmented and missing genes (~ 20% in total). With 3.0% of fragmented and 3.9% of missing candidate genes, the ON based assembly of clone 8410 exhibits a good representation of the expected gene content. It has to be noted, that the embryophyta set of orthologue sequences (mostly from land plants) might not be optimal for aquatic plants, because unnecessary genes might have been lost during evolutionary organ reduction and aquatic life style of duckweeds 21 .  25 reported a higher proportion of mobile elements (25%) for that species. Possible explanations might be: (1) There is a difference in repeat content between S. polyrhiza and S. intermedia; this seems less likely, because of the close relationship and same genome size of both species. (2) The assembly by Michael et al. 25 is more complete; this is probably not the case, because we found a lower content of mobile elements also in unassembled reads. (3) Their methods including structure-based detection revealed low copy elements (one or a few copies/C) which are not captured by our clustering and similarity-based searches. (4) Michael et al. 25 described a ratio of soloLTRs to complete elements of 8 to 1. Because our tools did not determine soloLTRs, the repeat content for S. intermedia might be underestimated. A high proportion of soloLTRs suggests that the genome size of Spirodela decreased by deletion-biased DNA double-strand repair 49 via the single-strand annealing pathway since its separation from the other duckweed lineages. Remarkable is the high abundance of SSR repeats, especially that of dispersed (AG) n microsatellite arrays, which in rice contribute to regulate gene expression by binding transcription factors 50 . Such arrays were reported also for S. polyrhiza, where they "severely impeded elongation during sequence assembly" 21 .

Repetitive elements. S. intermedia
The unusually low copy number of both 45S and 5S rDNA repeats revealed in the genome of S. polyrhiza 24,25 , inspired curiosity about the number and arrangement of those genes in related S. intermedia. The qPCR-based estimation showed ~ 93 copies of the 25S rRNA genes for both Si8410 and Si7747 genomes, a number very close to the estimate for S. polyrhiza 25 . Unlike S. polyrhiza, where the 45S rDNA was shown to locate in a single chromosome locus, our deep coverage ON sequencing of Si8410 revealed at least two 45S rDNA loci, located at ~ 83 kb upstream of the telomere at SiChr01, and at ~ 8 kb upstream of the telomere at SiChr06. The fact that FISH revealed only one signal on SiChr01, suggests that this location is the major 45S rDNA locus with the majority of the ~ 93 gene repeats. However, the exact 45S rDNA copy number distribution between the two loci remains unclear, because no read-through sequences are available. For the 5S rDNA, a range of generated ON read-through sequences, revealed two distinct loci, one containing a cluster of 31 repeats on SiChr15 which has been also visualized by FISH, and a second split locus on SiChr14 containing two clusters with 7 and 13 repeats, separated by ~ 13 kb long region of non-rDNA sequence (Fig. 6b). While the general arrangement of the 5S rRNA genes in S. intermedia resembles that of the S. polyrhiza, the total 5S rDNA copy number in S. intermedia is even lower, 51 vs 73 in S. polyrhiza 24 . This is the smallest number reported so far for any plant species. Usually, the copy number of 5S RNA genes in land plants varies from 2000 to 75,000 51 . Therefore, our findings for S. polyrhiza and S. intermedia proclaim that the small copy number of rDNA is a unique phenomenon of the genus Spirodela. www.nature.com/scientificreports/ (clone 8410) was isolated for Oxford Nanopore and Illumina sequencing as follows: the plants were kept three days in the darkness. After harvesting, 10 g of the fronds were used for DNA isolation from purified nuclei according to the protocol of Vondrak et al. 53 , with two minor modifications: the centrifugation of nuclei was performed at 650 × g instead of 200 × g due to their small size, and 2 × CTAB isolation buffer was supplemented with 2% PVP-360 (polyvinylpyrrolidone, avg. molecular weight 360,000).

Material and methods
Genome sequencing. PacBio. After shearing of genomic DNA (S. intermedia clone 7747), a size-selected 20 kb library was sequenced on the Pacific Biosciences RS II platform (GATC Biotech, Konstanz, Germany) combining the P6-C4 polymerase-chemistry and 240 min of movie duration. Two rounds of sequencing resulted in 149 Gb of raw read data.
Oxford nanopore. The sequencing libraries were prepared from 3 μg of purified HMW DNA using a Ligation Sequencing Kit SQK-LSK109 (Oxford Nanopore Technologies) as described by Vondrak et al. 53 . Briefly, the DNA was treated with 2 μl of NEBNext FFPE DNA Repair Mix and 3 μl of NEBNext Ultra II End-prep enzyme mix in a 60 μl volume that also included 3.5 μl of FFPE and 3.5 μl of End-prep reaction buffers (New England Biolabs). The reaction was performed at 20 °C for 5 min and 65 °C for 5 min. Then, the DNA was purified using a 0.4 × volume of AMPure XP beads (Beckman Coulter). Because long DNA fragments caused clumping of the beads and were difficult to detach, the elution was performed with 3 mM TRIS-HCl (pH 8.5) and was extended up to 40 min. Subsequent steps including adapter ligation using NEBNext Quick T4 DNA Ligase and library preparation for sequencing were performed as recommended. The whole library was loaded onto FLO-MIN106 R9.4 flow cell and sequenced on MinION instrument until the number of active pores dropped below 40 (21-24 h). Basecalling of the raw reads was done using Albacore 2.3.3 (Oxford Nanopore Technologies).
Illumina. Illumina sequencing was performed by Admera Health, LLC (South Plainfield, NJ, USA) using a KAPA DNA Library kit (Roche) and resulted in 130 million paired-end reads (2 × 150 nt).

Sequence assembly and pseudomolecule construction.
After an initial filtering for potential bacterial contamination (blastn against microbial NCBI refseq database from Aug 2017) and minimum read length (500 nucleotides), PacBio reads were assembled using the Canu pipeline v. 1.5 32 (options: 'genomeSize = 160 m, correctedErrorRate = 0.105′) consisting of the following steps: (1) Trimming, error correction and contig construction: Reads were corrected and trimmed by comparing overlaps. A minimum length of 500 nucleotides and a maximum error rate of 10.5% was chosen for extending a contig. Only reads consisting of more than 1000 nucleotides in length were considered in this step. Afterwards, the corrected reads were trimmed to improve overall read quality by using overlap information to detect high confidence regions. Contigs of insufficient read coverage and/or containing 'noisy' sequence were categorized as 'unsupported regions' and divided at weak sequence positions into subcontigs with higher support. After further contig construction on the basis of overlaps, a consensus sequence was built by removing the remaining sequencing errors to raise the overall assembly quality.
(2) Scaffolding and gap filling In a first round of scaffolding, the two genomes of the sister species S. polyrhiza (from clones 9505 and 7498) 23,25 were used as references for Mauve Genome Aligner v20150522 33 to order contigs. Scaffolding was performed by SSPACE-Longread v.1-1 34 (default options). The resulting scaffold assembly was used for the super-scaffolding approach. For this aim, contigs were assigned to 18 putative pseudomolecules (corresponding to the 18 S. intermedia chromosomes) using the information of cross-FISH of 93 S. polyrhiza BACs on the chromosomes of S. intermedia clone 8410 29 . New cytogenetic probes using BACs from the genomic regions of interest were designed for FISH experiments to approve localization of the contigs within the pseudomolecules and to resolve mis-assemblies. Additionally, bacterial contamination was filtered as described previously.
The quality of both S. intermedia genome assemblies was assessed by the BUSCO program 36,37 with the Embryophyta odb10 dataset comprising 1375 conserved genes.
The ONT/Illumina assembly for clone 8410 was performed using MaSuRCA 54 . In the first step, super-reads were assembled from 68 million Illumina 2 × 151 nt paired-end reads (128-fold coverage). Subsequently, 307,111 nanopore reads 10,000-425,377 nt in length (total length 9,963,752,003) and representing a 62-fold genome coverage, were used in scaffolding step. The resulting MaSuRCA assembly consisted of 386 scaffolds (N50 408,333 bp, total length 191,862,084 bp). Scaffolds were assigned/ordered into pseudomolecules using the 7747 as reference for a Mauve Genome alignment 33 . Additionally, all the formation of super-scaffolds was accompanied by manual curation steps based on FISH results.
Comparative genomics and visualization. Best bi-directional hits (BBHs) were identified using BLASTN 2.2.31 + ('-evalue 1e-5 -out outfmt 6′). Additional filtering for length and percent identity was applied afterwards (minimum length: 100 nt; minimum identity: 80%). Annotation of gene features and repetitive elements in S. polyrhiza 9505 was taken from Michael et al., 2017 25 . Both, length of selected repeat features and gene density were computed across a 0.5 Mbp window using the bedops suite 60 .
Feature annotation of each annotated repeats and genes as well as chromosome sizes have been converted to bed format using 'gtf2bed' . From these files the 0.5 Mbp windows across each chromosome have been calculated with bedops -chop. Finally, the counts of gene and repeat features within each chromosome has been determined by bedmap -count.
Synteny was determined using minimap2 ('minimap2 -x') 35 between S. polyrhiza and S. intermedia. 8410. Overlapping intervals have been merged by bedtools 61 merge (default options). Links shorter than 10.000 have been excluded from the analysis.
To compare the two Spirodela genomes, synteny between pseudomolecules, gene and repeat distribution have been plotted using Circos tool v0.67-1 62 .
Repeat annotation and analysis. Repeat analysis in unassembled Illumina reads from the clone 8410 was performed using similarity-based clustering implemented in the RepeatExplorer pipeline 40 . The pipeline was run with modified settings in order to increase its sensitivity towards divergent repeats. The similarity search step was done with BLASTN instead of mgblast and the similarity threshold was lowered from 90 to 80% identity. A total of 2.4 million of 150 nt paired-end reads were used as an input, and repeats were annotated in clusters representing at least 0.005% of the input reads.
Repeat annotation in the assembled pseudomolecules was performed using several alternative approaches. PROFREP and DANTE modules available at the public RepeatExplorer server (https ://repea texpl orer-elixi r.cerit -sc.cz/) were used to annotate repeats based on similarities to the reference database compiled from the repeat clustering analysis described above, and to the REXdb database of conserved protein domains of mobile elements 63 , respectively. Additionally, repetitive sequences in the assemblies were annotated using RepeatScout v. 1.0.5 41 . First, a library of repeats based on frequent k-mers was created using default parameters. Repeats with a frequency below 10 copies were then removed from the library. The resulting library was then used for annotating genome assemblies using RepeatMasker (https ://www.repea tmask er.org).
Identification and quantification of microsatellites (simple sequence repeats) was done by means of Tandem Repeat Finder 42 using the settings "trf input_file 2 5 7 80 10 25 25 -f " and parsing the output using TRAP 64 . Molecular characterization of rDNA. The estimation of 25S and 5S rRNA gene copies was carried out by qPCR, relating the rates of sample DNA amplification to the standard curve. The standard curve was constructed based on the amplification reads of dilution series of a reference plasmid, containing part of the actin gene (single copy in Spirodela polyrhiza, 9509), a whole PCR amplified 5S rDNA unit of S. polyrhiza and a part of the 25S gene amplified from genomic DNA of S. polyrhiza, 9509, using primers with internal restriction sites for XbaI and EcoRI. The rDNA copy number was determined in qPCR reactions prepared with the UltraSybr Mixture (CWBio, Taizhou, China), run on the CFX Connect Real-Time detection system (Bio-Rad, Hercules, USA). For quantification of the 25S rDNA we used the 5′-TCC CAC TGT CCC TGT CTA CT and 5′-CCC ACT TAT CCT ACA CCT CT primers, and for the 5S rDNA the set of primers was: 5′-GGG TGC GAT CAT ACC AGC AC and 5′-GGG TGC AAC ACG AGG ACT TC. The samples and tenfold dilution series of the reference plasmid were assayed in the same run. The quality of products was checked by thermal denaturation cycle. Only the experiments providing a single peak were considered. Three technical replicates were performed for each sample. The obtained data were analyzed using the program BIO-RAD CFX Manager 3.1 (Hercules, USA) and Microsoft Excel 2016 software.
For sequencing, the 5S rDNA genes were amplified from genomic DNA by PCR using 5S rRNA gene-specific primers DW-5S-F: CTT GGG CGA GAG TAG TAC TAGG and DW-5S-R: CAC GCT TAA CTT CGG AGT TCTG, purified by gel electrophoresis and cloned into the vector pMD19 (Takara, Dalian, China). The obtained sequences were analyzed using the "Online Analysis Tools" package (https ://molbi ol-tools .ca).