The genome of S. mediterranea and the evolution of cellular core mechanisms

Summary The planarian Schmidtea mediterranea is an important model for stem cell research and regeneration. We report the first highly contiguous genome assembly of Schmidtea mediterranea, using long-read sequencing and a de novo assembler (MARVEL) enhanced for low complexity reads. The S. mediterranea genome is highly polymorphic and repetitive genome, and harbors a novel class of giant Gypsy retroelements. Further, the genome assembly lacks a number of highly conserved genes, including critical components of the mitotic spindle assembly checkpoint, yet planarians maintain checkpoint function. Our genome assembly provides a key model system resource that will be useful for studying regeneration and the evolutionary plasticity of cell biological core mechanisms.


Introduction
Rapid regeneration from tiny tissue pieces makes planarians a prime model system for regeneration. Abundant adult pluripotent stem cells termed neoblasts power regeneration and the continuous turn-over of all cell types1-3 and transplantation of a single neoblast can rescue a lethally irradiated animal4. Planarians therefore constitute also a prime model system for stem cell pluripotency and its evolutionary underpinnings5. The taxonomic clade Platyhelminthes ("flatworms") also harbors parasitic lineages with a massive impact on human health, such as blood flukes (Trematoda) and tape worms (Cestoda)6. Here, the phylogenetic position of planarians as free-living flatworms7 provides a reference point towards an understanding of the evolution of parasitism8.
Despite modest genome sizes mostly in the range of 1-2 Gbp, planarian genome resources are so far limited. Although the model species Schmidtea mediterranea (Smed) was sequenced by Sanger sequencing, even 11.6x coverage of ~600 bp Sanger reads yielded only a highly fragmented assembly (N50 19 Kbp)9. Recent high coverage short-read approaches yielded similarly fragmented assemblies10, 11. The high A/T content of ~70% represents one known assembly challenge. Further, standard DNA isolation procedures perform poorly on planarians, which so far precluded the application of long-read sequencing approaches or BAC-clone scaffolding.
We here report a first highly contiguous PacBio SMRT long-read sequencing12 assembly of the Smed genome. Giant Gypsy/Ty3 retroelements, abundant AT-rich microsatellites and inbreeding-resistant heterozygosity collectively provide an explaination for why previous short-read approaches were unsuccessful. We find a loss of gene synteny in the genome of S. mediterranea and other flatworms. In analysis of highly conserved genes, we find a loss of MAD1 and MAD2, suggesting a MAD1-MAD2 independent spindle assembly check point. Our Smed genome assembly provides a resource for probing the evolutionary plasticity of cell biological core mechanisms, as well as the genomic underpinnings of regeneration and the many other fascinating phenomena that planarians so uniquely expose to experimental scrutiny.

De novo long read assembly of the planarian genome
In preparation for genome sequencing, we inbred the sexual strain of S. mediterranea (Smed) (Fig. 1a) for > 17 successive sib-mating generations in the hope of decreasing heterozygosity. Further, we developed a new DNA isolation protocol that meets the purity and high molecular weight requirements of PacBio long-read sequencing12 (Extended Data Fig. 1a-d, Supplementary Information S1-2). We used MARVEL, a new long-read genome assembler developed for low complexity read data (Supplementary Information S3, Nowoshilow et al., The axolotl genome and the evolution of key tissue formation regulators. Nature https://doi.org/10.1038/nature25458 (2018). An initial de novo MARVEL assembly of reads > 4 kbp with approximately 60x genome coverage show improvement over Canu, the PacBio assembly tool (Canu15) and showed substantial improvements over existing Smed assemblies based on short read sequencing (Extended Data To assess the quality of this genome assembly, we back-mapped a transcriptome of the sequenced strain (Supplementary Information S6) and found mapping of >99 % of transcripts, thus confirming both near-completeness and accuracy of the assembly (Supplementary Information S7, Extended Data Fig. 1f,g). To assess global assembly contiguity, we analyzed structural conflicts between the MARVEL assembly and Chicago/ HiRise scaffolding. Out of a total of 51 such events across the 782.1 Mbp of assembled genome sequence, only two represented unambiguous MARVEL assembly mistakes (Fig.  1b, Supplementary Information S4.3). Further, high-stringency back-mapping of high confidence cDNA sequences (Supplementary information S7.3) confirmed assembly contiguity below the ~ 1 kbp resolution limit of the Chicago/HiRise method, with smallscale sequence duplications near assembly gaps as only minor inconsistencies (Extended Data Fig. 2).
Our Smed genome assembly represents a major improvement over existing Smed assemblies10 (Fig. 1c) and more generally, the first long-range contiguous assembly of a non-parasitic flatworm species. A UCSC genome browser instance with supplemental quality control, annotation and experimental data tracks (Supplementary Information S8) is available at PlanMine17 (http://planmine.mpi-cbg.de). All analyses in this manuscript refer to the assembly release version dd_Smed_g4. The current source code of the MARVEL assembler is available at https://github.com/schloi/MARVEL. The execution scripts used for Smed can be found in the respective subfolder of the examples folder.

Assembly challenges in the Smed genome
To understand why the Smed genome was recalcitrant to prior short-read assembly, we first analyzed its repeat content (Supplementary Information S9). A repetitive fraction of 61.7 % (Fig. 2a) significantly exceeds the 38 % or 46 % repeat content of the mouse or human genomes18. We detected > 7,000 insertions of 11 distinct families of Long Terminal Repeat (LTR) retroelements ( Fig. 2b; Extended Data Fig. 3a, Supplementary Information S10). These do not cluster with known Metaviridae (Fig. 2b), suggesting that they represent either extremely divergent or so far undescribed retroelement families. Three families reach an exceptional size of > 30 kbp, which is more than 3-times longer than the 5-10 kbp typically observed in vertebrates (Fig. 2c, Extended Data Fig 3b). The only known similar-sized LTRs are the plant-specific Ogre-elements19, which is why we refer to the giant Smed repeat families Burro (Big, Unknown Repeat Rivaling Ogre; Supplementary Information S10.3). Burro elements are pervasively transcribed (Extended Data Fig. 3c,d. Supplementary Information S10.4), yet their high degree of intra-family sequence divergence suggests a relatively ancient invasion (Supplementary Table 1, Supplementary Information S10.5, Extended Data Fig. 3e). Burro-1, with 130 fully assembled copies the most abundant giant retroelement, is highly overrepresented at contig ends and 50 % of all current scaffolds terminate in a Burro-1 element (Fig. 2d, Supplementary Information S10.6). Therefore, these abundant > 30 kbp repeat elements still limit the size of the current assembly. Additionally, abundant AT-rich microsatellite regions disrupt the alignment of spanning reads and thus also reduce contig contiguity (Extended data Fig. 4, Supplementary Information S11). Finally, the Smed assembly graphs showed substantial structural heterogeneity (Supplementary Information S12) in form of bubbles (transient divergences in sequencing read alignments) and spurs (divergences without re-connection), which were largely absent from a comparable genome assembly, of Drosophila melanogaster using PacBio sequencing and MARVEL aseembly ( Overall, the combination of giant repeat elements, low-complexity regions and inbreedingresistant heterozygosity provides an explanation for why prior short-read sequencing assemblies of Smed have proven so challenging. The long-range contiguity that we achieved in the Smed genome assembly and similarly substantial improvements of the recently published PacBio genome assembly of the flatworm species Macrostomum lignano21 (Supplementary Table 2), further emphasizes the improvements that the combination of long-read sequencing with our MARVEL assembler offers in the assembly of challenging genomes.

Comparative analysis of the planarian gene complement
We next annotated the Smed gene complement, relying on our planarian transcriptome resources17 (Supplementary Information S13). Our analysis showed a high divergence of Smed gene sequences (Supplementary Information S14) en par with Caenorhabditis elegans (Fig. 3a). In contrast, the low degree of sequence substitutions between the sexual and asexual Smed strains (Fig. 3a) and nearly identical mapping statistics of the two transcriptomes to the genome (Supplementary Information S7.1, Extended Data Fig. 1f) establish the utility of our assembly for both strains. To evaluate the Smed genome structure, we performed whole genome alignments (Supplementary Information S15) with the available parasitic flatworm genomes6 and a draft genome of the platyhelminth M. lignano21 (Fig. 3b). The highest alignment similarity was found between Smed and the parasitic flatworm Schistosoma mansoni, which is consistent with the platyhelminth phylogeny7. However, alignments were mostly limited to individual exons of specific genes irrespective of the quality of the various assemblies (Extended Data Fig. 5a,b). In general, flatworm genome comparisons resulted in alignment chains much shorter and lower-scoring than those obtained from comparisons across the tetrapod (human-frog) or vertebrate (human-zebrafish) clade (Fig. 3b). Together with likely > 1,000 planarian-specific protein coding genes (Supplementary Information S16; Supplementary Table 5, Extended Data Figure 6a-g), our data show a high degree of genome divergence in Smed and other flatworms.
We therefore next investigated gene loss in planarians. Our analysis deliberately focused on highly conserved genes, such that the absence of sequence similarity alone provides a strong indication of loss (Supplementary Information S17). We identified 452 highly conserved gene losses shared between Smed and other planarians ( Fig. 3c), which compares to 284 and 757 such losses in D. melanogaster and C. elegans (Extended Data Fig. 5c). Gene loss in planarians is therefore broadly in range with established invertebrate model organisms. However, the lost genes included 124 homologues of essential genes in humans or mice (Supplementary Table 6) and generally key components of multiple cell biological core mechanisms (Fig. 3c). Specifically, planarians lack multiple highly conserved components of DNA double strand break (DSB) repair, including Rad52, XRCC4, XLF, SMC5/6 and the entire condensin II complex22. A possibly consequent reliance on mutagenic DSB repair pathways (e.g., micro-homology mediated end joining)23 could account for both the abundance of microsatellite repeats and the structural divergence of the Smed genome ( Fig.   3b), but raises questions regarding the extraordinary resistance of planarians to DSB inducing γ-irradiation4.
Further, planarians are missing recognizable homologues of key metabolic genes. Loss of Fatty Acid Synthase (FASN) is striking in face of its essential role in eukaryotic de novo fatty acid synthesis and may indicate a particular dependence of planarians on dietary lipids. The loss of the heme break-down enzymes HMOX1 and BRVB despite maintained heme biosynthesis capacity24 is similarly unusual for a free living eukaryote (C. elegans lost both25). Remarkably, the above and multiple other genes were missing not only in planarians, but also in the parasite genomes6 and the transcriptome of the macrostomid M. lignano26 (Fig. 3c). Given their broad conservation in the lophotrochozoan sister clade, their broad absence in flatworms represents a likely ancestral loss. This complicates for example the interpretation of FASN loss in the parasitic lineages as specific adaptation to parasitism6. Conversely, the absence of key metabolic genes as phylogenetic signal underscores the utility of free-living flatworms as model systems for the parasitic lineages and the development of anti-helminthic reagents8. A Mad1/Mad2-independent spindle check-point?
The apparent absence of Mad1 and Mad2 in planarians (Fig. 3c) raises the question of whether planarians have a functional SAC, and how essential cellular functions can be maintained in absence of supposed core components. Both are near-universally conserved due to essential roles in the spindle assembly checkpoint (SAC), which guards against aneuploidy27 by inhibiting cell cycle progression as long as even a single chromosome remains unattached to the mitotic spindle14. Though Mad1 and Mad2 homologues are easily identifiable in all other flatworms examined (Extended Data Figure 7, 8), not even flatworm queries could identify significant homologues in Smed or the transcriptomes of 5 other planarian species. Therefore, planarians have very likely lost Mad1, Mad2 and multiple other SAC components (Fig. 4a). The known M-phase arrest of planarian cells upon pharmacological interference with spindle function28 (Fig. 4b) is therefore remarkable, as it indicates the maintenance of a SAC-like response despite a lack of supposed SAC core components.
In order to explore the underlying mechanisms, we targeted remaining components of the SAC network (  Figure 9b). Therefore, planarian Rod-Zwilch-Zw10 either control APC/C-Cdc20 independently of Mad1/Mad2 or in concert with homologues that have lost defining sequence features (Extended Data Figure 6, 7). Our results motivate the examination of putative Mad1/2 independent roles of the Rod-Zwilch-Zw10 complex also in other model systems and, together with the striking evolutionary plasticity of the SAC network in eukaryotes13, generally challenge our understanding of a cell biological core mechanism.

Discussion
We here report the first highly contiguous genome sequence of the planarian model species Schmidtea mediterranea, which enables the genomic analysis of whole body regeneration, stem cell pluripotency, lack of organismal ageing and other fascinating features of this model system. The resulting bird's eye view of a "difficult" genome using long-read sequencing and de novo assembly also highlights significant challenges remaining to be overcome. In the case of Smed, these include an abundance of low complexity microsatellite indicates that similar challenges may be wide-spread. We therefore expect that the specific improvements of the MARVEL assembler towards heterozygous and/or compositionally biased sequencing data (Novojilow et al., coordinated in press at Nature) will be useful for enhancing assembly contiguity in de novo genome sequencing projects.
Our genome assembly also shows a high extent of structural rearrangements and the absence of a number of conserved genes in the Smed genome. However, also D. melanogaster, C. elegans or other animals show loss of "essential" genes13,25,31, which raises a general conundrum: How can animals survive and compete while lacking core components of essential mechanisms? In cell biological terminology, "core mechanism" signifies a chain of molecular interactions that explain a given process in multiple species, while "essentiality" indicates importance for organismal survival. The emergence of viable yeast strains upon deletion of essential genes32 or the competitiveness of hundreds of extant planarian species in a diversity of habitats worldwide33 both relativize "essentiality". Our demonstration of SAC function in likely absence of Mad1 and Mad2 suggests that our genetic and mechanistic understanding of SAC function is incomplete. Further studies on planarians and other "non-traditional" model organisms are needed to understand the basis and mechanism of these cellular functions. Such function-oriented, rather than gene-centric view of biological mechanism abstracts general function from individual molecules and is therefore likely to ultimately also facilitate the reverse engineering of biology.

Europe PMC Funders Author Manuscripts
Europe PMC Funders Author Manuscripts results of Extended Data Fig. 1g. These 2 were designated as "false positive", since both mapped to the Smed genome with > 90 % query coverage and sequence identity using BLAT. c) UCSC genome browser screenshot (75 kbp window) of the genomic mapping location of one of the two "unknown" HC-cDNAs as single example of a mapping failure due to an actual assembly error. The example documents inversion of the 5'-end of the cDNA within a low confidence stretch at a contig end (lack of coverage in the Quiver track). Europe PMC Funders Author Manuscripts assembly contiguity due to an increased probability of read alignment loss. d) Genome-wide coverage ratios of insertion/deletion sequences > 99 bp and excluding AT-repeats. e) Read length variation analysis across AT-rich repeat regions (AAT) in regular PacBio sequencing data compared to Circular Consensus Sequencing (CCS) coverage of the same region. CCS reads sample the same genomic region multiple times. The lack of a clear difference in the length variation of specific AT-repeats (AAT) between repetitive sequencing of the same DNA molecule (CCS data set) versus sequencing reads representing different DNA molecules (regular PacBio data) indicates that repeat length variations are mainly technical in nature. Rather than repeat length polymorphisms, the most likely cause of the detrimental effect of the repeats is the increased ambiguity in low complexity sequence alignments (Supplementary Information S11. Extended Data Figure 5. comparative genomics. a) Table listing contig and scaffold N50 statistics of the genomes used for the comparative genome alignments in Fig. 3b. The table reveals that the basal vertebrate lamprey genome assembly is more fragmented (similar or lower N50 values) than most other platyhelminth genomes. Nevertheless, the human to lamprey genome alignment has equivalent or even higher alignment chain scores and spans, indicating that the true extent of sequence divergence and loss of conserved gene order in platyhelminths is likely an underestimate. contains predicted domains. d) Identity of detected domains (Pfam and SUPERFAMILY). "unintegrated signatures" designates recurring sequence motifs that are not grouped into InterPro entries. These might represent so far un-curated or weakly supported motifs that do not pass InterPro's integration standards. e) Differential expression of 626 planarian-specific genes in published Smed RNAseq data sets of different regeneration phases (left), stem cells or progeny populations (middle) or specific developmental stages (right). Red lines indicate differential expression relative to the control of each series (white = no change). Genes were ordered using rank by sum. The high proportion of differential expression indicates the widespread contribution of lineage-specific genes to planarian biology.