Chromosome-scale reference genome assembly of a diploid potato clone derived from an elite variety

Abstract Potato (Solanum tuberosum L.) is one of the most important crops with a worldwide production of 370 million metric tons. The objectives of this study were (1) to create a high-quality consensus sequence across the two haplotypes of a diploid clone derived from a tetraploid elite variety and assess the sequence divergence from the available potato genome assemblies, as well as among the two haplotypes; (2) to evaluate the new assembly’s usefulness for various genomic methods; and (3) to assess the performance of phasing in diploid and tetraploid clones, using linked-read sequencing technology. We used PacBio long reads coupled with 10x Genomics reads and proximity ligation scaffolding to create the dAg1_v1.0 reference genome sequence. With a final assembly size of 812 Mb, where 750 Mb are anchored to 12 chromosomes, our assembly is larger than other available potato reference sequences and high proportions of properly paired reads were observed for clones unrelated by pedigree to dAg1. Comparisons of the new dAg1_v1.0 sequence to other potato genome sequences point out the high divergence between the different potato varieties and illustrate the potential of using dAg1_v1.0 sequence in breeding applications.


Introduction
Potato (Solanum tuberosum L.) was domesticated about 8000 years ago in the Andes from diploid wild potatoes and became a staple food of indigenous American communities (Spooner et al. 2005). Because of its high nutritional value (Jansky et al. 2019), the potato is nowadays one of the most important crops for humanity and its global production exceeds 370 million metric tons (FAO 2019).
The number of potato cultivars is in the thousands (FAO 2008), most of which are tetraploid (2n ¼ 4Â ¼ 48), with a high level of heterozygosity and strong inbreeding depression (Zhang et al. 2019). With the steady rise of the human population, and growing fears of food insecurity (Beddington 2010), it is crucial to increase potato productivity. Inter alia, considerable increases are expected to be contributed by plant breeding (Lenaerts et al. 2019). Modern breeding tools such as genome editing (Altpeter et al. 2016) and genomic selection (Stich and Van Inghelandt 2018) have the potential to enhance the gain of selection in potato. However, to utilize the full potential of these tools, high-quality reference genomes of germplasm relevant to breeding are required.
The current S. tuberosum reference genome is that of a doubled monoploid clone from the cultivar group Phureja (Xu et al. 2011;Sharma et al. 2013;Pham et al. 2020). However, group Phureja has considerable genome and phenotype differences compared to the commercially established group Tuberosum of tetraploid cultivars (Xu et al. 2011), which makes it presumably not ideal as a reference for the latter. Furthermore, preliminary comparisons between cultivars indicated substantial sequence and structural variations (SV; Xu et al. 2011;Uitdewilligen et al. 2013), which calls for cultivar-specific genome assemblies as to optimally exploit genomic tools for potato breeding.
Assembling potato genomes is challenging because of their high levels of heterozygosity. Mixed heterozygous and homozygous regions make it difficult for algorithms to find a single unique path of overlapping reads, leading to more fragmented assemblies and a requirement of higher sequencing coverage (Pryszcz and Gabaldó n 2016). If heterozygosity is very high, the alternative haplotype contigs are assumed to be separate regions of the genome, a phenomenon called undercollapsed heterozygosity (Matthews et al. 2018). This effect is more pronounced in tetraploid genomes, since they have more alternative haplotype versions of the same region. Long-read sequencing technologies such as PacBio (Shearman et al. 2020;Vollger et al. 2020) and Nanopore (Kuderna et al. 2019;Low et al. 2019;Kinkar et al. 2021) aim to overcome the problem of heterozygosity, allowing more space for overlaps during assembly (Jiao and Schneeberger 2017). Further sequencing technologies such as proximity ligation and optical mapping help resolving areas that are difficult to assemble (Field et al. 2020).
In recent years, potato genome assemblies of wild diploid potato relatives Solanum commersonii (Aversano et al. 2015) and Solanum chacoense (M6; Leisner et al. 2018) have become available. Exploiting the latest sequencing technological advances, Zhou et al. (2020) assembled the phased genome sequence of RH89-039-16 (RH89), a diploid clone derived from a cross between S. tuberosum dihaploid and a diploid clone, which in turn was generated from a cross between two S. tuberosum group Phureja hybrids (Xu et al. 2011). Finally, the first non-Phureja S. tuberosum assembly has been recently published (Solyntus_v1.1;van Lieshout et al. 2020). For the latter, however, the Phureja genome (DM_v4.03) has been used for reference-based scaffolding. Therefore, to our knowledge, no genome sequence of an elite variety is available nor any pure chromosome-level assembly of S. tuberosum group Tuberosum.
The objectives of this study were (1) to create a high-quality consensus sequence across the two haplotypes of a diploid clone derived from a tetraploid elite variety and assess the sequence divergence from the available potato genome assemblies as well as among the two haplotypes; (2) to evaluate the new assembly's usefulness for various genomic methods; and (3) to assess the performance of phasing in diploid and tetraploid clones using linked-read sequencing technology.

Materials and methods
Genetic material, DNA, and RNA extraction Three gynogenic dihaploid S. tuberosum clones (dAg1, dAg2, and dAg3) were created from S. tuberosum group Tuberosum tetraploid cv. Agria (tAg). The haploid inducer was S. tuberosum group Phureja IVP06-153. Besides tAg, its parental clones tPa1 and tPa2 as well as five tetraploid elite potato clones (tV1-tV5) were included in this study. DNA was extracted from the leaves of all clones according to Mayjonade et al. (2016). For RNA sequencing, 10 tubers of tAg were grown in a cultivation chamber set to 25 C during day (6-22 h) and 20 C during night. The light intensity was about 300 lmol/m 2 s in the leaf canopy. Samples of leaves, stolons, and flowers were harvested at 15 (leaves and stolons) and 45 (flowers) days after planting. Total RNA was extracted using RNeasy Plant MiniKit (Qiagen, Hilden, Germany) following the manufacturer's instructions. RNA was pooled to equal concentration for the following library preparation.

Preparation of libraries and sequencing
For all clones, 10x Genomics (10xG; Pleasanton, CA, USA) libraries were prepared (Supplementary Table S1) following the manufacturer's recommendations, using 1 ng of DNA input, where size selection was performed before library preparation on BluePippin (SAGE Sciences, Beverly, MA, USA) with a high-pass protocol allowing a size selection start at 40 kb. The quantity and quality control of size-selected DNA were performed with Qubit (Thermo) and with a Genomic tape (Agilent TapeStation). Sequencing of 10xG libraries was performed on an Illumina (San Diego, CA, USA) HiSeq3000 in paired-end read mode.
For dAg1, SMRTbell libraries were prepared as recommended by Pacific Biosciences (Menlo Park, CA, USA, SMRTbell Template Prep Kit 1.0-SPv3), including a final size selection on Blue Pippin to remove fragments lower than 10 kb. Sequencing was performed on a PacBio Sequel I with Binding Kit 2.0 and Sequencing chemistry 2.0 for 10 h or Binding Kit 3.0 and Sequencing chemistry 3.0 for 20 h, as recommended by Pacific Biosciences.
Pooled RNA from leaves, stolons, and flowers was used to prepare an Iso-Seq library following manufacturer instructions. Sequencing was performed on the PacBio Sequel II using the Sequel II Sequencing Kit 2.0 chemistry. Iso-seq v3 pipeline (https://github.com/PacificBiosciences/IsoSeq) was used to generate final RNA sequencing data.

Genome assembly
Our objective was to create one contiguous consensus assembly across the two haplotypes of dAg1 and phase the existing intragenomic variants for diploid and tetraploid clones in a second step. We have evaluated two different assembly strategies to obtain the dAg1_v1.0 genome sequence (Supplementary Figure S1), but in this manuscript, only the final assembly strategy and results are presented.
Final assembly strategy: PacBio assembly as backbone All PacBio reads that had 200Â coverage, an error rate <15% after error correction, and a length !1000 bp were assembled with Canu v1.8 . Parallelly, the same reads were also assembled using Falcon and Falcon-unzip (Chin et al. 2016). To deal with the higher error rate of PacBio reads, both assemblies were polished using Pilon (v1.22; Walker et al. 2014) with the less error-prone 10xG linked reads, where mapping was performed with longranger align (v2.2.2) (Zheng et al. 2016). Furthermore, the polished Canu assembly was filtered with Purge Haplotigs (Roach et al. 2018) to avoid undercollapsed heterozygosity (Matthews et al. 2018) by discarding alternative haplotigs.
A hybrid assembly was created using quickmerge (v0.3; Chakraborty et al. 2016), where the polished Falcon assembly was used as reference and the polished and deduplicated Canu assembly as query. This was followed by a second round of Pilon polishing with mapped 10xG linked reads. These mapped reads were additionally used to correct misassemblies using Tigmint (Jackman et al. 2018) and the assembly was filtered with Purge Haplotigs. Arcs (v1.0.6; Yeo et al. 2018) and Links (v1.8.7;Warren et al. 2015) were used to scaffold contigs of the polished, corrected, deduplicated quickmerge assembly with the 10xG library 1, lowering the minimum aligned reads to 3 instead of 5 (-c 3) and using k-mers of size 20 (-k 20). Thereafter, the step was iterated with 10xG library 2. Finally, a last round of polishing with Pilon and filtering with Purge Haplotigs was performed.

Hi-C scaffolding
The reads of the Hi-C library were mapped against the scaffolded hybrid assembly in two steps with different software. In the first step, we used BWA-MEM (v0.7.15; Li and Durbin 2010) for mapping and Salsa (Ghurye et al. 2017) for scaffolding, with misassembly correction activated (-m yes). In the second step, Juicer was used for mapping (Durand et al. 2016) and 3D-DNA (Dudchenko et al. 2017 for scaffolding. Contigs smaller than 12.5 kb were ignored during scaffolding and the repeat coverage misjoin threshold (-editor-repeat-coverage) was set to 3. The resulting contact maps were visualized using Juicebox (Durand et al. 2016;Dudchenko et al. 2017Dudchenko et al. , 2018) and a final manual curation and scaffolding were performed.

Evaluation of assemblies
A custom python script was used at all steps of the assembly to obtain several statistics, namely the N50, N90, L50, L90, number of Ns per 100 kb, as well as scaffold number, and total sequence length. Benchmarking Universal Single-Copy Orthologs (BUSCO; Simão et al. 2015) were used to assess gene completeness compared to the Solanaceae gene set (odb10; Kriventseva et al. 2019).
In order to evaluate our final assembly, mapping of 10xG linked reads from various diploid and tetraploid clones against our and the existing potato reference assemblies (dAg1_v1.0, DM_v4.04, DM_v6.1, RH89, M6, and Solyntus_v1.1) was performed using longranger align. Illumina sequencing data of the diploid wild potato species (Solanum bukasovii, dW; Kyriakidou et al. 2020) were downloaded from the SRA database and mapped against the six genomes using BWA-MEM. Samtools (v1.10; Li et al. 2009) was used to calculate the proportion of mapped reads and properly paired reads (Thankaswamy-Kosalai et al. 2017).

Gene annotation
The MAKER pipeline (Campbell et al. 2014) was used to annotate genes. A custom repeat library was created with Repeatmodeler (Smit et al. 2013) and Mite Hunter (Han and Wessler 2010) according to Campbell et al. (2014). Repeatmasker (Smit et al. 2013) was then used to mask these repeat regions in the genome. The Isoseq RNA data generated for tAg in this project as well as published mRNA reads (SRX4882701; Caruana et al. 2019) from tAg, assembled into a transcriptome with Trinity (v2.11.0) (Haas et al. 2013), were used as EST evidences. Protein evidences were UniProt proteins of Solanum (The UniProt Consortium 2019). Snap (Korf 2004) and Augustus (Stanke et al. 2008) were used as gene predictors. Orthologous analysis with UniProt proteins of Solanum was done with Orthofinder (Emms and Kelly 2019).

Variant calling, phasing, and annotation
The dAg1_v1.0 assembly was used as reference to call single nucleotide variants (SNV), and small insertions and deletions (indels, <50 bp) for all potato clones. The corresponding 10xG linked reads of the diploid clones dAg1, dAg2, and dAg3 were aligned with longranger wgs (v2.2.2), and phased SNV and indels were called using freebayes (v1.3.2-40; Garrison and Marth 2012). 10xG linked reads of the three tetraploid clones tAg, tPa1, and tPa2 were mapped against the dAg1_v1.0 assembly using longranger align (v2.2.2) and variants were called by freebayes. Variants of the samples dAg1, dAg2, dAg3, tAg, tPa1, and tPa2 were filtered for a minimum depth of 10. The variants of the clones were phased with whatshap polyphase (Schrinner et al. 2020). The allele profiles of regions for which phase information was available for the offspring (dAg1, dAg2, dAg3, and tAg) were compared with that of the respective parents (tAg, tPa1, and tPa2). The proportion of regions with correctly phased allele profiles in the offspring compared to the allele profiles of the parental clones was calculated.
Sorting Intolerant From Tolerant 4G (SIFT4G, v2.4) was used to annotate tolerant (score >0.05) and deleterious (score 0.05) variants based on the conversion of amino acid sequences (Vaser et al. 2016). The SIFT4G database was built using SIFT4_Create_Genomic_DB with the uniref90 database, the dAg1_v1.0 sequence, and its corresponding predicted genes and proteins. The number of genes with at least one putative deleterious variant was estimated.
Pericentromeric regions of the potato chromosomes of DM_v4.03 were determined based on the recombination rates reported for the DRH population (Manrique-Carpintero et al. 2016). Thereafter, we determined the pericentromeric regions in the dAg1_v1.0 sequence based on the coordinates of the wholegenome alignment between dAg1_v1.0 and DM_v4.03 using show-coords from the MUMMER package. We then used a t-test to examine the difference of the proportion of genes with at least one deleterious variant between 1dAg1-3 and 1Pa1-2 in pericentromeric to subtelomeric regions for its statistical significance. Additionally, a t-test was used to test for a mean difference of the proportion of genes with at least one deleterious variant, calculated in 1-Mb windows across the genome, between diploid (1dAg1-3) and tetraploid clones (1tPa1-2).
SV between the two haplotypes of dAg1 were identified from PacBio reads, using the CuteSV algorithm (v1.0.8; Jiang et al. 2020) after mapping the reads with Minimap2. Sequence divergence between the assembly sequences was estimated as the proportion of the number of bp affected by SV, where the latter was extracted from the whole-genome alignments (-l 1000 -c 1000 -d 10), including all final primary scaffolds of the four potato genomes using show-diff from the MUMMER package.

Genome assembly
Two PacBio assemblies were created for the final assembly strategy: the first with Canu comprising 14,037 contigs and the second with Falcon comprising 2,109 contigs. The Canu assembly had a larger than expected assembly size and the BUSCO analysis indicated a high proportion of duplications, both signs of undercollapsed heterozygosity. In addition, the N50 value of the Falcon assembly (0.618 Mb) was higher than that of the Canu assembly (0.203 Mb). Consequently, the Falcon assembly was used as reference and the Canu assembly as query in creating the hybrid assembly. The resulting hybrid assembly had a reduced number of contigs (1,592) and the N50 increased to 0.865 Mb. After two rounds of 10xG scaffolding, the number of scaffolds decreased to 704 and the N50 increased to 1.656 Mb, where, for the Solanaceae gene set, a BUSCO statistic of 95% was observed.
This assembly strategy based on PacBio long reads as backbone resulted in a more contiguous assembly with a closer-toexpected assembly size and a lower number of Ns (Table 1) compared to another strategy with 10xG linked reads as backbone (Supplementary Text and Table S2). Therefore, this former assembly was used for further scaffolding using Hi-C data. The two Hi-C scaffolding approaches led to drastically increased N50 values up to 57.4 Mb and a slight decrease in BUSCO statistics. The latter phenomenon was already observed by Kadota et al. (2020) and might be due to misassembly over-correction and/or imperfect manual curation at the last stage.
The unscaffolded minor contigs were concatenated as ChrUn (65.88 Mb), which represents 8.3% of the genome and hosts 2,124 (4.7%) of the annotated genes. When ignoring Ns and including ChrUn, the final assembly size of the dAg1_v1.0 genome was 812 Mb, which is in a similar range compared to 731, 807, and 1,674 Mb (diploid genome size) for DM_v6.1, M6, and RH89, respectively (Supplementary Table S3). Considering only the 12 final scaffolds as chromosomes, the genome size of 744 Mb is larger than in M6, Solyntus_v1.1, and DM_v6.1 (499, 716, and 731 Mb).

Evaluation of dAg1_v1.0 genome sequence
The visual inspection of the Hi-C contact maps of the dAg1_v1.0 sequence suggested the presence of clear contact areas between the ends of all chromosomes ( Figure 1A). This has been observed earlier for plant genome (Liu and Weigel 2015;Liu et al. 2017) and supports the quality of our assembly. As additional quality control, especially to evaluate the successful purging of the second haplotype from the consensus assembly, we have evaluated the genome-wide distribution of the read depth. Only a few coverage spikes were observed for the final assembly (Supplementary Figure S2). A first analysis of these regions with particularly high coverage suggests that they are related to repetitive sequences. These attributes indicate that our assembly has a high quality.
Visual inspections of the dot plots of whole-genome alignments between dAg1_v1.0 vs DM_v6.1 and between dAg1_v1.0 vs RH89 (Figure 1, B and C) suggested a high level of correspondence between dAg1_v1.0 and the other two potato genomes. A reduction of the minimum length of a single exact alignment match from 1,000 to 100 bp was necessary to visualize the whole-genome alignment of dAg1_v1.0 and Solyntus_v1.1 ( Figure 1D) which suggests a lower level of correspondence, which might be explained by misassemblies in the Solyntus_v1.1 genome scaffolded by DM_v4.03 (DM_v4.04 without unscaffolded contigs). This previous Phureja genome presumably included misassemblies and sequencing errors due to the limited sequencing resources available in 2011 when the genome was assembled (Pham et al. 2020). This explanation is supported by the observation of similar differences in the whole-genome alignment between dAg1_v1.0 and DM_v4.04 (Supplementary Figure S3) but not between dAg1_v1.0 and DM_v6.1.
The reason for larger gaps in the abovementioned whole-genome alignments might be a lower assembly quality in these regions, especially in the pericentromeric regions. The latter regions are characterized by high repeat frequencies which are difficult to assemble.
Visual inspections of the dot plots of whole-genome alignments of RH89 vs dAg1_v1.0 and of RH89 vs DM_v6.1 (Supplementary Figure S4) suggested that the RH89 and DM_v6.1 genomes are more similar than the RH89 and dAg1_v1.0 genomes. This visual impression is supported by the observation of a lower sequence divergence of $8.5% between RH89 and DM_v6.1 compared to $10.8% between RH89 and dAg_v1.0, calculated based on the sequence differences due to SV. This finding is in agreement with the RH89 pedigree, which implies a higher relatedness between RH89 and DM than between RH89 and dAg. The sequence divergence between dAg_v1.0 and DM_v6.1 and Solyntus_v1.1 was 8.2% and 8.6%, respectively.
To assess the completeness and correctness of the dAg1_v1.0 sequence relative to other potato sequences, we mapped 10xG linked reads of various potato varieties (tV1-tV5) and one wild species (dW) to dAg1_v1.0, DM_v4.04, DM_v6.1, RH89, Solyntus_v1.1, and M6 genome sequences. The percentage of mapped reads of all examined clones was higher against S. tuberosum reference sequences than against M6 (Figure 2A). The proportion of mapped reads against dAg1_v1.0 was high and similar to the other S. tuberosum genomes. However, the proportion of properly paired reads, considered to be a more accurate quality measure (Thankaswamy-Kosalai et al. 2017), was on average across all examined clones the highest for the dAg1_v1.0 genome ( Figure 2B). PacBio and Iso-seq reads from diploid and tetraploid Agria were also mapped against the reference assemblies and high percentages of reads were mapped in all cases (Figure 3). These observations together indicated the high completeness and especially the correctness of the dAg1_v1.0 genome assembly, which will therefore be highly useful for genome-assisted breeding applications in potato and the basis for many future research projects on diploid and tetraploid potato.

Transcript analysis
To illustrate further the usefulness of the dAg1_v1.0 genome for research on tetraploid potatoes, a transcript analysis was performed. High-quality Iso-seq reads of tAg were mapped against dAg1_v1.0, DM_v6.1, and RH89 genomes. After collapsing the mapped reads, unique transcripts were compared with the dAg1_v1.0 annotation set as well as those published for DM_v6.1 and RH89 (Pham et al. 2020;Zhou et al. 2020). SQANTI3 analyses showed a higher ratio of transcripts associated with annotated genes compared to novel genes in dAg1_v1.0 and a higher number of annotated genes in dAg1_v1.0 compared to the other genomes (Supplementary Figure S5A). Though this finding may be biased by the fact that Iso-seq reads were used as evidences in obtaining gene models in dAg1_v1.0, the results of SQANTI3 suggest a good quality of annotation for dAg1_v1.0.
Iso-seq reads were obtained from an RNA bulk of leaves, stolons, and flowers. So a broad representation of genes should be expected. Nevertheless, some tissues, like tubers, are not present in Iso-seq reads. To assess whether dAg_v1.0 annotation presents some bias, an orthologous analysis was done between dAg_v1.0 genes and Solanum proteins from the UniProt database. Up to 90% of the tuber proteins have an orthologous in dAg_v1.0 genes, the same proportion found in RH89 and DM_v6.1. So, despite the absence of tuber Iso-seq reads, no clear bias was found in dAg_v1.0 annotated genes.
The proportion of genes with more than one isoform was greater than the number of genes with only one isoform  (Supplementary Figure S6). This illustrates the importance of alternative splicing and polyadenylation (Supplementary Figure  S5B). The detailed analyses of these aspects using SQANTI3 did not reveal any systematic differences (Supplementary Figure S7

Intragenomic diversity
We detected across the dAg1 genome 7,829,534 heterozygous SNV and indels which resulted in a sequence diversity between haplotypes of $1% (Table 2). A similar amount of heterozygous variants was identified for the two other diploid Agria clones dAg2 and dAg3. These values are in the range of what was reported previously for diploid wild species S. commersonii (1.49%; Aversano et al. 2015) and M6 (0.68%; Leisner et al. 2018). In addition, 32,028 SV were detected between the two dAg1 haplotypes which are in the similar range of what was reported for RH89 (Zhou et al. 2020). For the three tetraploid clones tAg, tPa1, and tPa2, the frequency of heterozygous variants was with 3.1%, 3.6%, and 3.3%, respectively, about thrice higher than for the diploid ones. This is due to the fact that in tetraploid clones, more variants between haplotypes can occur compared to diploid clones, as more haplotypes are present. These results are in accordance with those of Hardigan et al. (2017), who found a similar relation between the variant frequencies of diploid and tetraploid clones which were 1.05% for diploid landraces and 2.73% for tetraploid cultivars. The number of genes with at least one deleterious variant was assessed in our study. This number was for the diploid clones with values between 13,287 and 16,766 considerably higher than the deleterious mutations in 10,642 annotated genes described by Zhou et al. (2020) for the RH89 genome. This finding might be due to the usage of different approaches to detect deleterious mutations. In our study, short reads were mapped against the dAg1_v1.0 sequence, whereas Zhou et al. (2020) aligned the assembled chromosomes of RH89 to the DM_v4.03 sequence.
The number of genes with at least one deleterious variant observed for the tetraploid clones (tPa1 and tPa2) was with values of 27,927 and 28,357 about twice as high as for the diploid clones. Hence, the proportion of genes with at least one deleterious variant in 1-Mb windows across the genome is higher for tetraploids than for diploids (Figure 4). This might be due to that in tetraploid clones deleterious alleles can be more easily masked by non-deleterious alleles due to the higher number of alleles per gene. This explanation is supported by the higher number of genes with at  least one homozygous deleterious variant for diploids (dAg2: 1,838; dAg3: 1,436) than for tetraploids (tPa1: 669; tPa2: 1,060). A similar number of genes with homozygous deleterious variants (1,753) was observed by Zhou et al. (2020). These findings indicate the higher masking potential of deleterious alleles in tetraploids, compared to diploids. Furthermore, our observation illustrates the high efforts that will be required to breed potato as a diploid hybrid crop. This is especially true as the proportion of the number of genes with at least one deleterious variant between 1tPa1-2 and 1dAg1-3 was significantly (P < 0.001, t-test, sample size: The next inner circles report the distributions of genes (black), repeats (green) measured as percentage of masked bp, and structural variations (blue). The four most inner circles illustrate the proportion of genes with at least one deleterious variant in 1dAg1-3 (black) and 1tPa1-2 (orange), and heterozygous variants in dAg1, dAg2, and dAg3 in 1-Mb windows, respectively. The gray bars mark the pericentromeric regions, whereas the yellow bars mark the regions where the highest difference between the proportion of genes with at least one deleterious variant of 1dAg1-3 and 1tPa1-2 was identified. 863) higher in pericentromeric regions, compared to subtelomeric regions. In the former, a purging of alleles is considerably more difficult due to the reduced recombination.
In addition to the frequency of sequence variants, the phasing of alleles is relevant to evaluate the possibilities of combining or separating alleles at neighboring loci by recombination. Recently, methods have been proposed for phasing that rely on long-read sequencing (e.g., Schrinner et al. 2020). We have evaluated the use of linked-read sequencing for phasing the heterozygous variants for diploid and tetraploid clones. The resulting blocks of phased regions across the genomes had a median length of 116 bp for tPa2 and 6,824 bp for dAg1 ( Table 3). The figures are discouraging with respect to the use of phasing information e.g., in the context of genomic selection approaches (Stich and Van Inghelandt 2018). Nevertheless, these lengths are in accordance to the results of Yang et al. (2017) who phased the hexaploid sweet potato genome and obtained 542,361 phased regions, which covered about 30% of the genome.
Despite the short block length, the phased regions of parental and offspring clones were compared to each other with respect to the present alleles. In only 2.0-2.3% of the cases, the haplotypes (i.e., phased variants) observed in the three diploid clones were not observed in tAg (Table 4). For the grandparents tPa1&2, these figures were with 2.1-3.3% slightly higher but still indicating a good phasing accuracy. More than 50% of all haplotypes of tAg were also observed in the four haplotypes of the parental clones tPa1&2. It was expected that two haplotypes of tAg would occur in tPa1 and the other two in tPa2. However, an in-depth evaluation of the phasing accuracy of tetraploids using 10xG linked reads was not possible due to the short phased regions.

Conclusions
In this study, we have created a chromosome-scale consensus sequence across the two haplotypes of a diploid clone derived from a tetraploid elite potato variety. This de novo assembly was performed with an optimal combination of today's sequencing technologies, comprising 10xG linked reads, PacBio long reads, and Hi-C reads. Comparisons of the new dAg1_v1.0 sequence to other potato genome assemblies pointed out the high divergence between the different potato clones and illustrated the potential of using dAg1_v1.0 sequence in breeding applications. The high amount of heterozygous SNV and indels, SV, and genes with at least one deleterious variant highlights the intragenomic diversity of the dAg1_v1.0 genome. Finally, in this study, we have shown that sequence variants of diploid potato clones could be phased using cost-efficient 10xG linked reads and the dAg1_v1.0 sequence. However, further improvements are needed to enlarge the phased regions to enable this approach in a breeding-related context.

Data availability
Supplementary File_S1 contains Supplementary Tables S1-S3 and Figures S1-S7. Supplementary Table S1 contains statistics of the sequencing technology data used in this study, including potato clones, data type, number of reads, median, Q 5 , Q 95 , and raw coverage. Supplementary Table S2 contains assembly statistics, namely number of contigs, assembly size, largest contig, N50, N90, L50, L90, Ns per 100 kb, and percentage of BUSCO genes, for an alternative assembly strategy. Supplementary Table S3 contains statistics of the different published potato assemblies with number of scaffolds, assembly size, assembly size considering only 12 chromosomes, scaffold N50, and Ns per 100 kb for 12 chromosomes. Supplementary Figure S1 shows the pipeline of the two different assembly strategies evaluated in this study. Supplementary Figure  S2 shows the read depth and the percentage of masked bp in windows across all potato chromosomes. Supplementary Figure S3 depicts the alignment dot plot of dAg1_v1.0 and DM_v4.04 genomes. Supplementary Figure S4 represents the alignment dot plot of RH89 and DM_v6.1 genomes. Supplementary Figure S5 depicts the number of transcripts assigned to annotated and novel genes and alternative splicing events. Supplementary Figure S6 shows the number of isoforms per gene. Supplementary Figure    Ethics approval and consent to participate