Reproductive sorus tissue from a wild population of S. latissima sporophytes was sampled from Black Ledge, Groton, Connecticut, US (41°31′N, 72°07′W, 26 June 2014). Following induced sporulation, individual gametophytes were isolated to establish monoclonal gametophyte cultures in a laboratory setting, as outlined by Redmond et al. (2014) and Alsuwaiyan et al. (2019). One female S. latissima gametophyte (var. SL-CT1-FG3) was selected for long-read sequencing for reference genome assembly and cultured in Erlenmeyer flasks under red light with a 12:12 h light:dark photoperiod at 10°C to inhibit reproduction and promote growth and mitotic division (Redmond et al., 2014; Augyte et al., 2018) for genomic DNA extraction. Augyte et al. (2018) at the Marine Biotechnology Laboratory at the University of Connecticut extracted DNA from a 24 mg (fresh) female gametophyte culture of S. latissima (var. SL-CT1-FG3) using a modified protocol of the NucleoSpin Plant II Maxi Kit (Macherey-Nagel, Düren, Germany; cat # 740609). Gametophyte biomass was spun down in 1.5-mL tubes in an Eppendorf 5424 microcentrifuge (21,000 rcf, 2 min). Sealed tubes of gametophyte material were frozen in liquid nitrogen for 20 s before being ground with a plastic pestle for 30 s. Ground samples were extracted with CTAB buffer using repeated wash steps (Doyle and Doyle, 1990; Augyte et al., 2018).

We collected samples from Saccharina latissima sporophytes cultivated and harvested by University of Connecticut and GreenWave from a farm site in Branford, Connecticut, USA (41°15′13″N, 72°46′5″W, May 2020). Sporophytes were replicates of a single cross from S. latissima gametophyte collections (Augyte et al., 2018; Mao et al., 2020): a female gametophyte (var. LIS-F1-3) originating from Southern New England, and a male gametophyte (var. SL-CT1-MG2) cultured from the spore release that also generated SL-CT1-FG3. We flash-froze the samples in liquid nitrogen immediately after collection and subsequently stored at −80°C for less than 1 month before being sent on dry ice to Cornell University for RNA extraction. Sporophyte samples were ground in liquid nitrogen and RNA was extracted using the Quick-RNA Microprep Kit (Zymo Research, Irvine, CA, USA; cat #R1052), yielding 8.31 µg total RNA (average 1.66 µg/sample). An additional RNA extraction was performed at the HudsonAlpha Institute (Huntsville, Alabama, USA) from SL-CT1-FG3 gametophyte culture using the RNeasy Micro Kit (Qiagen Inc., Valencia, CA, USA; cat # 74004).

Extracted DNA from the S. latissima female gametophyte (var. SL-CT1-FG3) was sent to the HudsonAlpha Institute for whole genome sequencing using a whole genome shotgun sequencing strategy and standard sequencing protocols. Sequencing reads were collected using two platforms: Illumina reads were sequenced using the Illumina NovaSeq 6000 platform, and the PacBio reads were sequenced using the Sequel II platform. One 400bp insert 2 × 250 Illumina fragment library (66.03x) was sequenced along with one 2 × 150 Dovetail Hi-C library (145.33x) (Supplementary Table S1). Prior to use, the Illumina fragment reads were screened for phix contamination. Reads composed of >95% simple sequence were removed. Illumina reads <50bp after trimming for adapter and quality (q < 20) were removed. The final Illumina read set consisted of 282,564,006 reads for a total of 66.03x of high-quality bases. PacBio sequencing yielded 111.14 Gb of total raw sequence, representing 180.56x genomic coverage (Supplementary Table S2).

RNA extracted from five S. latissima sporophyte samples was sent to the HudsonAlpha Institute for library preparation and sequencing using standard protocols. All RNA libraries were prepared using TruSeq Stranded mRNA Library Prep (96 samples) (Illumina, San Diego, CA, USA; cat # 20020595) and indexed with IDT for Illumina TruSeq RNA UD Indexes v2 (96 indexes) (Illumina, San Diego, CA, USA; cat # 20040871) according to manufacturer instructions. cDNA was sequenced on the Illumina NovaSeq 6000 platform and generated a total of 287 million reads, with an average of 96 million reads per sample.

The initial assembly version 0 was generated by assembling 11,430,834 PacBio CCS reads of the female S. latissima gametophyte SL-CT1-FG3 using hifiasm v0.7 (Cheng et al., 2021) and subsequently polished using RACON v1.4 (Vaser et al., 2017). This produced an initial assembly (SL-CT1-FG3 v0) consisting of 4,854 contigs, with a contig N50 of 863.2 Kb, and a total assembly size of 925.8 Mb (Supplementary Table S3).

Hi-C sequencing of SL-CT1-FG3 yielded 626,664,456 2 × 150 Hi-C Illumina reads, an estimated 145.33x coverage. The reads were aligned to the SL-CT1-FG3 v0 assembly using BWA-MEM v0.7.17 (Li, 2013). Paired-end reads were mapped independently (as single-ends) due to the nature of the Hi-C pair, which captures conformation via proximity-ligated fragments. A small fraction of single-end mapped reads will contain a ligation junction, an indicator that they are chimeric because they do not originate from a contiguous piece of DNA. In these cases, only the 5′-side was retained, as the 3′-end generally originates from the same contiguous DNA as the 5′-side of the mated read. The resulting single end alignments were combined into a BAM file containing the paired, chimera-filtered Hi-C read alignments. The 3D-DNA (Dudchenko et al., 2017) suite of internal tools was used to generate a contact map using the resultant BAM file, and the contact map was visualized using Juicebox (Durand et al., 2016). Chromosome-scale scaffolding was attempted using 3D-DNA, but high levels of contamination in the v0 genome initially prevented meaningful scaffolding.

The assembled contigs were screened against bacterial proteins, organelle sequences, and the NCBI non-redundant protein sequence database (NR) (Sayers et al., 2024) and removed if found to be a contaminant according to standard practice (e.g., (Bennetzen et al., 2012; Motamayor et al., 2013; Bartholomé et al., 2015). Contigs were classified into bins depending on sequence content. Contamination was identified using BLASTn (Camacho et al., 2009) against the NCBI non-redundant nucleotide database (NT) (Sayers et al., 2024) and BLASTx using a set of known microbial proteins. Additional contigs were classified in the version 1 release as contaminants (2,863 contigs, 283.9 Mb), chloroplast (181 contigs, 11.0 Mb), prokaryote (14 contigs, 10.6 Mb), redundant (>95% masked with 24mers that occur more than 2 times in all contigs) (233 contigs, 6.2 Mb), repetitive (>95% masked with 24mers that occur more than 4 times in contigs greater than the contig N50) (24 contigs, 1.8 Mb), and unanchored rDNA (Ding et al., 2022) (3 contigs, 165.1 Kb).

With contaminant contigs removed, we attempted to scaffold contigs together using the contact information from 3D-DNA (Dudchenko et al., 2017) and scaffold graph construction, as described by Ghurye et al. (2019). In brief, a graph was formed between all contigs in which graph edge weights, w(u, v), between any two contigs u and v were computed by dividing the number of counts, N(u, v), by the total number of cut sites, C, in both u and v: (Ghurye et al., 2019) w u , v = N u , v C u + C v

We then computed a normalized Best Buddy Weight, BBW(u, v), as the weight, w(u, v), divided by the maximal weight of any edge incident upon contigs u or v, excluding the (u, v) edge itself (Ghurye et al., 2019). All BBW(u, v) values >1 were retained, and a reverse Dijkstra’s algorithm (highest weight graph) (Dijkstra, 1959) was then utilized to generate the contig order and orientation for the join file. Using this algorithm, a total of 333 joins were made to form an additional 218 scaffolded contig sets. Each join was padded with an unsized gap of 10,000 Ns.

SNP and INDEL errors in the consensus were corrected with 282,564,006 Illumina fragment 2 × 250 reads (66.03x coverage) by aligning the reads using BWA-MEM (Li, 2013) and identifying SNPs and INDELs with GATK3 UnifiedGenotyper (Van der Auwera and O’Connor, 2020). A total of 502 SNPs and 20,723 INDELs were corrected in the release. The final version 1 release contained 612.21 Mb of sequence, consisting of 218 scaffolds and 1,513 contigs, with a contig N50 of 971.7 Kb (Supplementary Table S4).

The JGI Annotation Pipeline (Grigoriev et al., 2006) was used to annotate the S. latissima nuclear genome assembly. The pipeline automatedly predicts, filters, and functionally annotates gene models as described by Grigoriev et al. (2006). Briefly: repeats in the genome assembly were masked with RepeatMasker (Smit et al., 2004), RepBase (Jurka et al., 2005), and RepeatScout (Price et al., 2005). The masked assembly was then used to predict protein-coding gene models with ab initio, homology, and transcriptomic modeling methods. Ab initio gene predictions were performed with Fgenesh (Salamov and Solovyev, 2000) and GeneMark (Ter-Hovhannisyan et al., 2008). Homology was assessed by BLASTx (Camacho et al., 2009) of the assembly against NCBI NR (Sayers et al., 2024). Resulting alignments were used to seed Fgenesh+ (Salamov and Solovyev, 2000) and Genewise (Birney et al., 2004) for homology-based gene prediction. A transcriptome assembly was generated for S. latissima with Illumina RNAseq reads with Trinity v2.11.0 (Grabherr et al., 2011), and RNA reads were mapped back to the genome assembly with HISAT2 (Kim et al., 2019b). With these inputs, the transcriptome-based programs Fgenesh (Salamov and Solovyev, 2000) and combest (Zhou et al., 2015) were used to generate gene models. A total of 41,561 gene models predicted across all methods (all models) were filtered based on protein homology and transcriptome evidence to a single representative model at each genomic locus (filtered models), generating a set of 24,790 gene models; for a detailed description of JGI model filtration, refer to Grigoriev et al. (2006). Protein sequences predicted from the set of filtered gene models were functionally annotated. Proteins were classified using SignalP v3 (Nielsen et al., 1997) for signal sequences, TMHMM (Melén et al., 2003) for transmembrane domains, and InterproScan (Quevillon et al., 2005) for functional domains. BLASTp (Camacho et al., 2009) alignments of proteins against NCBI NR (Sayers et al., 2024), Swiss-Prot (UniProt Consortium, 2012), KEGG (Kanehisa, 2004), and KOG (Koonin et al., 2004) databases to further informed functional interpretation of predicted proteins. The version 1 S. latissima genome assembly, associated annotations, and metadata are hosted on the JGI PhycoCosm (Grigoriev et al., 2021) comparative algal genome portal (https://phycocosm.jgi.doe.gov/SlaSLCT1FG3_1).

Gene content was scored for the analyzed brown macroalgal genomes using BUSCO v5.7.1 (Manni et al., 2021) in genome mode with gene predictor set to Augustus v3.5.0 (Stanke et al., 2008) against the ortholog databases for Eukaryota (eukaryota_odb10) and Stramenopiles (stramenopiles_odb10). QUAST-LG v5.2.0 (Mikheenko et al., 2018) was used to compute relevant assembly quality metrics for each. Synteny of our v1 S. latissima assembly to related species was investigated by aligning to published genomes of S. japonica, Macrocystis pyrifera, U. pinnatifida, and Ectocarpus sp. using the Progressive Cactus v2.6.7 pipeline (Armstrong et al., 2020). The phylogenetic tree used to seed the Cactus alignment was pruned using tidytree v0.4.6 (Yu, 2022) and treeio v1.30.0 (Wang et al., 2020) from the Starko et al. (2019) kelp phylogeny constructed from plastid, mitochondrial, and ribosomal genes.

Blocks of synteny were extracted from the five-species hierarchical whole genome alignment (HAL) format (Hickey et al., 2013) into the BLAT-defined PSL format (Kent, 2002) using halSynteny v2.2 (Krasheninnikova et al., 2020). We applied a method put forward by Nosil et al. (2023) to establish one-to-one homology between chromosomes of related species using synteny blocks derived from pairwise Cactus alignments. To appropriately map our S. latissima scaffolds onto longer reference chromosomes, we modified the method to a many-to-one approach, allowing multiple S. latissima scaffolds to align to a single chromosome. For each alignment of a S. latissima scaffold to a reference species chromosome, lengths of syntenic blocks (“matches” in PSL format (Kent, 2002)) were summed. Best scaffold-chromosome pairs were identified with respect to each S. latissima scaffold, hereafter referred to as a “maximal syntenic match”, by calculating the maximum summed exact match per scaffold amongst the aligned chromosomes. FASTA genome assembly files for each of the five species, and a whole genome alignment file converted from HAL to MAF using hal2maf v2.2 (Hickey et al., 2013), were given as input to Ragout v2.3 (Kolmogorov et al., 2018), a reference-assisted scaffolding tool used to improve assembly contiguity.

Heatmaps representing synteny between reference chromosomes of each species versus S. latissima scaffolds were generated using ggplot2 v3.5.1 (Wickham, 2016) in R v4.4.0 (R Core Team, 2024). Ordering of our genome v1 assembly scaffolds onto synteny-constructed pseudochromosomes was rendered into genetic map representation using ggplot2 v3.5.1 (Wickham, 2016) in R v4.4.0 (R Core Team, 2024). Scripts used to perform these analyses and generate figures are hosted in a public GitHub repository (https://github.com/kdews/s-latissima-genome).

The v1 assembly was screened for vector using the NCBI UniVec database (https://www.ncbi.nlm.nih.gov/tools/vecscreen/univec) using the standard command blastall -p blastn -d Saccharina_latissima.mainGenome.fasta -i UniVec -q −5 -G 3 -E 3 -F ″m D” -e 700 -Y 1.75e12 -m 8 (Camacho et al., 2009). A total of 43 vector hits were identified with≥95% identity and≥31bp in length. Contaminants were identified using the contaminant screener FCS-GX (Astashyn et al., 2024). A total of 90 contaminated regions were identified (63 TRIM, 21 FIX, 6 EXCLUDE). All identified vector and contaminant regions were removed from the sequence. If a vector/contaminant region fell within a scaffold, then the region was replaced with the same number of N’s. If the vector/contaminant region fell on the front/end of a scaffold, the bases were eliminated, and the annotation GFF file was translated appropriately. These changes resulted in a net loss of 53 scaffolds, 4,498,393 bp, and a total of 209 annotated genes.

Organelle genomes from the same female S. latissima gametophyte (SL-CT1-FG3) were also assembled using the PacBio reads generated for the nuclear genome assembly. Reference organelle genomes for S. latissima have been published (Wang et al., 2016; Fan et al., 2020b), but these genomes do not have corresponding nuclear genomes from the same individual. Gene transfers between organelle and nuclear genomes are more easily identified when using genomes sourced from the same individual (Cui et al., 2021). To identify organelle reads, raw PacBio reads were aligned to the S. latissima mitochondrial (Wang et al., 2016) and chloroplast (Fan et al., 2020b) genomes using minimap2 standard settings (Li, 2018). Read IDs that aligned to the organelle genomes were extracted using samtools (Li et al., 2009), and then corresponding PacBio reads that aligned to the respective organelle genomes were subset using seqtk subseq (Li, 2012). Organelle genomes were then assembled using the assembler flye v2.9.2-b1786 (Kolmogorov et al., 2019), with the--pacbio-raw flag and genome size estimates based on previously published sugar kelp chloroplast (Fan et al., 2020b) and mitochondrial (Wang et al., 2016) genomes. We used GeSeq2 (Tillich et al., 2017) to annotate and compare our sugar kelp organelle genomes versus available published reference S. latissima organelle genomes (Wang et al., 2016; Fan et al., 2020b), using genome annotations of U. pinnatifida mitochondria (Li et al., 2015) and chloroplast (Zhang Y. et al., 2016) as outgroups. A collection of scripts used in this analysis have been placed in a public GitHub repository (https://github.com/kdews/s-latissima-organelles).

Our assembled S. latissima chloroplast genome (130,613 bp) is almost the same length as the published chloroplast genome (130,619 bp) (Fan et al., 2020b) (Table 2; Supplementary Figure S4; Supplementary Figure S6B). Our assembled S. latissima mitochondrial genome is 37,510 bp and contains 39 genes. It is slightly smaller than the previously published mitochondrial genome (37,659 bp) (Wang et al., 2016), but contains an additional tRNA annotation (Table 2; Supplementary Figure S5; Supplementary Figure S6A). For each of the new organelle genomes reported here, a consensus sequence was generated through multiple separate assemblies with flye (Kolmogorov et al., 2019), a long-read assembler especially robust to sequencing errors and specialized to resolve repetitive regions. Long reads allowed for resolution of inverted repeat (IR) regions in the chloroplast genome that typically span ∼5.5 kb in brown macroalgae (Rana et al., 2019). The average sequencing read length (8,389 bp) also aided with assembly, as each read covers ∼22% and ∼6% of mitochondria and chloroplast genome lengths, respectively.

The decrease in sequencing costs has led to an increase in genome assemblies for non-model species including brown algae, which until recently have lacked genomic resources. Nuclear genomes are now becoming available for some Phaeophyta species, e.g., Ectocarpus sp. (Cock et al., 2010; Cormier et al., 2017), S. japonica (Ye et al., 2015; Liu et al., 2019; Fan et al., 2020a), U. pinnatifida (Shan et al., 2020; Graf et al., 2021), and S. latissima (Denoeud et al., 2024; this publication). This annotated and scaffolded genome of North American sugar kelp (S. latissima) represents a major step forward in kelp research and development in the United States. Our assembly provides a reference for future research into population genetics and gene expression in sugar kelp populations along the coast of the northeastern US and supports both marine conservation efforts and the nascent US kelp industry.

Many tools used in breeding rely on the availability of high-quality genomes and rigorous gene annotation. Demands on genome assembly coverage, accuracy, and annotation have increased as selective breeding methods have progressed (Varshney et al., 2014; International Wheat Genome Sequencing Consortium et al., 2018; Bayer et al., 2020). With sparse maps of genetic markers, marker-assisted selection (MAS) achieved modest improvements over phenotypic selection in the prediction of breeding values for crops and livestock (Dekkers and Hospital, 2002). Modern genomic selection (GS) models depend on high-density genotypic data to generate genomic estimated breeding values (GEBVs) (Meuwissen et al., 2001) that take genome-wide variation into account (Calus and Veerkamp, 2007; Heffner et al., 2009; Christensen and Lund, 2010). To accurately predict the performance of genetic crosses with these methods, chromosome rearrangements between them must be known (Nuzhdin et al., 2012). Additionally, genome engineering (e.g., CRISPR/Cas9 gene editing) relies on the construction of accurate genome assemblies for cultivated species to mitigate potential off-target genome modification (Guo et al., 2023). Here, we have assembled a genome for S. latissima of sufficient quality to assess genetic variation through a variety of markers (e.g., SNPs, INDELs, structural variants) useful for the progression of intraspecific breeding programs (Li et al., 2022), as well as possible crossbreeding with closely related species such as S. japonica (Zhang J. et al., 2016; 2018).

In addition to nuclear genomes of brown algae, publications of plastid and mitochondria genome assemblies are also mounting (Oudot-Le Secq et al., 2006; Chen et al., 2019; Rana et al., 2021). While genomics has greatly increased breeding efficiency in plants, animals, and kelp, most of these breeding programs rely strictly on nuclear genomes for establishing markers for selective breeding. However, integrating organelle genomes into breeding models may increase breeding efficiency (Kersten et al., 2016). By optimizing cytonuclear interactions, breeding efficiency for sugar kelp in the future can be accelerated (Colombo, 2019). As nuclear and organelle genomes can differ between individuals in the same species, pairing nuclear and cytoplasmic genomes from a single genotype can provide a reference for future breeding experiments. For S. latissima, short-read organelle genome assemblies have become available over the past decade (Wang et al., 2016; Fan et al., 2020b; Rana et al., 2021). Our use of long-read sequencing increases our confidence in the nuclear and organelle S. latissima genome assemblies described here to serve as a foundation for future sugar kelp scientific research and breeding.

The whole-genome hierarchal alignments conducted in this study produced syntenic data applicable to both our genome improvement and comparative genomic analyses, and future research into brown algae evolution, phylogenetics, and genomic breeding. High molecular weight DNA extraction is particularly challenging for macroalgal species (Snirc et al., 2010; Greco et al., 2014), which can pose a significant barrier to generating chromosome-level assemblies. In general, genome quality and completeness are assessed through gene annotation and evaluation of sequence contiguity. From a structural perspective, genome contiguity (i.e., N50, gaps/N content) informs the degree to which assembly and scaffolding methods have succeeded in reconstructing chromosomes from whole-genome shotgun sequencing.

In brown macroalgae, Ectocarpus sp. Ec32 v2 (Cormier et al., 2017), U. pinnatifida (Shan et al., 2020), and M. pyrifera (Diesel et al., 2023) represent the most highly scaffolded genome assemblies, with greater than 90% of sequence contained in chromosomal scaffolds. In contrast, although the S. japonica assembly (Fan et al., 2020a) has been mapped into chromosomes, almost 35% of the genome is not scaffolded. A similar pattern exists in the European S. latissima v2 assembly (Denoeud et al., 2024), in which >20% of the genome was not incorporated into pseudochromosomes. Our North American S. latissima v1 genome assembly contains 218 scaffolds that incorporate ∼45% of the genome by length. These results are despite high (>100x) genomic coverage with long reads and Hi-C-assisted assembly methods for all genomes compared in this study (Table 1). Though not conclusive, the relative difficulty of de novo chromosomal reconstruction in Saccharina genome assemblies could indicate physical traits or genome architecture specific to this genus that impede efforts to sequence and scaffold DNA, for example, elevated concentrations of polysaccharides and/or phenols or higher genomic repeat content in the genus Saccharina as compared to other kelps. To overcome obstacles to chromosome-level de novo assembly, we demonstrate the utility of syntenic information to estimate chromosome number and assist in scaffold placement along chromosomes in S. latissima.

Importantly, our comparative genomics analysis has identified syntenic regions and homologous chromosomes between five species of brown algae. Construction of the brown algal phylogeny has historically been complicated and contested, both in morphological and genetic studies, owing to a sparse brown algal fossil record, rapid diversification, and convergent evolution (Bolton, 2010; Silberfeld et al., 2010; Starko et al., 2019; Bringloe et al., 2020; Choi et al., 2024; Denoeud et al., 2024). Building on syntenic blocks we identified among brown algal genomes, phylogenomic analyses that evaluate whole genomes and investigate chromosome homology and evolution have the potential to resolve outstanding problems of placement within this clade radiation (Steenwyk and King, 2024). Comparative genomic analyses like those presented here help correlate chromosomes between species for more robust evolutionary analysis and genomic breeding prediction that leverage the increasing amount of high-quality, functionally annotated brown macroalgal genome assemblies.