Genome assembly and annotation

We used nanopore sequencing in combination with Illumina sequencing of 10X Genomics Chromium-, paired-end- and HiC libraries to generate a chromosome-scale haploid assembly of the Atlantic halibut. The final scaffolded assembly (IMR_hipHip.v1) had a scaffold N50 of 27 Mb (S1A and S1B Fig) and this assembly was evaluated using BUSCO using 4584 genes in Actinopterygii odb9 data set, annotating 95.4% of analyzed genes as “Complete” in the genome assembly. The assembly is organized into 24 larger scaffolds containing 97.6% of the assembled sequence (S2 Fig), corresponding to the chromosomes inferred from karyotyping [23] and linkage analysis [20]. A number of smaller contigs and scaffolds, together comprising only 2.4% of the assembled sequence was left as unplaced. We named the scaffolds by mapping microsatellite primer sequences from a published genetic linkage map for Atlantic halibut [20] to IMR_hipHip.v1. Genome to genome alignment of IMR_hipHip.v1 to a recently finished reference genome of the Pacific halibut (NCBI accession GCF_013339905.1) revealed a largely co-linear genome organization, with few large-scale disagreements (S1C Fig) as expected due to the recent split of the two species. To generate RNA-seq data for genome annotation and to obtain transcriptomic data describing the early stages of development, we used samples from different time points throughout embryonic development from newly fertilized to 54 days post fertilization (dpf). For the first four stages (1 hour post fertilization, 5 dpf, 8 dpf and 12 dpf), four pools of eight individuals were collected and subjected to RNA extraction and Illumina HiSeq mRNA-sequencing. For the 28 dpf, 43 dpf and 54 dpf, four individual fish were sampled and sequenced for each stage. Genome annotation based on the RNA-sequencing identified 18,314 genes, of which 18,092 (98.8%) significantly matched with proteins in public databases.

Identification of the Atlantic halibut sex chromosome

In order to identify and characterize the sex chromosomes we sampled and extracted and pooled DNA from ten full siblings of each phenotypic sex and sequenced these separately using Illumina HiSeq sequencing. The sequencing data was mapped against the IMR_hipHip.v1 assembly and the sequence differentiation between male and female pools was determined for windows of 500 kb along each chromosome using delta allele frequency (ΔAF) by comparing male- and female allele frequencies (ΔAF = abs(maleAF-femaleAF)). This screen revealed chromosome 13 as the most diverged between males and females (Fig 1A), in agreement with a previously published linkage study using RAD-sequencing [19]. The divergence between phenotypic males and females on chr13 is not uniform, but gradually declines along the length of the chromosome, with the end being no more divergent than autosomes (Fig 1B). We extracted SNPs that had a male vs. female allele frequency difference of >40% while being fixed for one allele in the female pool. Genome-wide 8,566 SNPs showed this pattern, 5,715 of which were located on chr13.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Genetic differentiation between phenotypic males and females highlights chr13 as the putative Y-chromosome.

A Average male vs. female ΔAF for 500 kb windows along each assembled chromosome. The genome-average ΔAF is indicated as a horizontal line together with a line indicating Z-score = 3, i.e. the ΔAF-value corresponding to 3 SD above the genome average). B Average ΔAF for 500 kb windows along chr13 with genome average ΔAF and 3 SD above genome average ΔAF indicated by horizontal lines.

https://doi.org/10.1371/journal.pgen.1010011.g001

Identification of the Atlantic halibut sex determining gene

RNA-seq data was generated from 12 whole individuals covering three developmental stages with the assumption that both genetic sexes would be represented. As we were unable to infer phenotypic sex for these samples, an unbiased screen could involve searching for the most variably expressed genes among the 12 individuals, but after conducting a Principal Component Analysis we concluded that the samples clustered strongly by development stage (S3 Fig). An alternative strategy could involve screening for variable gene expression between known phenotypic sexes during gonadal differentiation. As known phenotypic sex was unavailable, we determined the genetic sex of the RNA-seq samples by mapping the RNA-seq samples to the genome, then intersecting observed RNA-seq alleles with the 5,715 SNPs on chr13 differentiating males from females. Hierarchical clustering of RNA-seq alleles observed at intersected SNP positions (S4 Fig), shows clearly that 7 samples were genetically male (XY) and 5 were genetically female (XX). We then produced a Volcano plot showing the global genetic male vs. female expression ratio plotted against the statistical significance of a differential expression (DE) between sexes (Fig 2A). The statistically most significantly DE gene (two-sided t-test) was gsdf, previously shown to be the male-determining MSD gene in Luzon ricefish and sablefish [24,25]. Furthermore, among genes previously identified as MSD genes in other fish species [8], only gsdf and bmpr1b [11] reside on the herein defined chrY (chr13 FRR) in Atlantic halibut (gsdf at 8.5 Mb and bmpr1b at 12.1 Mb). Bmpr1b was however not differentially expressed between males and females. Among 12 RNA-seq individuals the 5 XX fish had almost no gsdf expression, while the 7 XY fish all had high expression, consistent with Gsdf being a male determining factor. To verify sex specific expression of gsdf during development and to obtain better temporal resolution of expression levels, we performed quantitative PCR (qPCR) on PCR sex-determined samples (Fig 2B and 2C). RNA was extracted from individual embryos from 11 development stages, ranging from 1 hour post fertilization (hpf) to 13 dpf. A PCR assay based on SNPs in the 3’UTR of brx (chr13:9125004–9125007) was used to determine genetic sex of samples from 48 hpf and onwards (S5 Fig). The qPCR results show that the expression of gsdf starts in males already at 48 hpf, and from 7 dpf onwards there is a clear increase in expression compared to the earlier stages. For one individual sampled at 6 dpf determined as a genetic male, we could not detect any expression, possibly due to environmental factors having the chance to override the genotype in the determination of phenotypic sex in the species as is the case with several flounder species [17,26], including Atlantic halibut [27]. In silico searches for conserved genes located up- and downstream of gsdf in other species confirmed the gsdf identity. (S6 Fig).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Identification of gsdf as the Atlantic halibut sex determining gene.

A Volcano plot representation of differential expression (DE) between individuals stratified by chr13 genotyping-based sex assignment. Dot colors indicate the chromosome from which genes are expressed. The most highly differentially expressed gene, gsdf, is indicated in the upper right corner. B qPCR result showing the sex specific gsdf expression (relative to gtf3c6) across developmental stages. C Pictures of fish representing the developmental stages sampled for gene expression analyses; 1 hpf to 13 dpf.

https://doi.org/10.1371/journal.pgen.1010011.g002

Most of the Atlantic halibut genome is subject to male-only or female-only meiotic recombination

In order to validate chr13 as the Atlantic halibut sex chromosome we downloaded the pedigree RAD-sequencing data previously used to assign linkage group 13 as X/Y [19], aligned the data to our assembly IMR_hipHip.v1, called SNPs and investigated genotypes by sex phenotype in the pedigrees. The pedigree analysis verified our divergence-based assignment of IMR_hipHip.v1 chr13 as the sex chromosome and 90/94 F1 individuals were in agreement with regard to the association between phenotypic sex and chr13 genotype. A similar observation was made for the embryonal qPCR analysis where one out of 34 genetic males (brx-assay) had no gsdf expression. The reason for the non-perfect association between genetic- and phenotypic sex could be erroneous annotation of the downloaded RAD-seq dataset or a logical consequence of non-perfect penetrance of young genetic sex determination systems. In our genome assembly-guided linkage analysis, we detected suppression of meiotic recombination along a large part of chr13 in male germ cells as would be expected in an X/Y system. Surprisingly, the majority of the sex chromosome pair exhibited strict sex specific meiotic recombination. We term such regions Male-only or Female-only Regions of meiotic Recombination (MRR/FRR) (Fig 3A). The male- and female genetic maps for chr13 both sum up to 49 cM but the recombining intervals are nonoverlapping. None of the 90 F1 offspring used for analysis carried a male meiotic recombination in the first ~15 Mb, while all detected female meiotic recombination events on chr13 occurred in this region. Thus, the first ~15 Mb of chr13 behaves like a X/Y-linked region, with high male vs. female divergence (Fig 1B) and nonexistent meiotic recombination in males (FRR). The later part of chr13 (~15 Mb until chromosome end) behaves like a pseudoautosomal region (PAR), but only recombines in males (X/Y recombination), rendering sequence diversity in this part of the chromosome shared between X and Y.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Cumulative proportion of all recombination events occurring during male and female meiosis in a two-pedigree intercross.

A chr13 with gsdf location indicated. *Mapping interval of markers most highly associated with sex in a previous study [19]. B Male and female linkage maps differ strongly genome-wide. The outermost track visualizes individual chromosomes. Locations of defined Male-only and Female-only meiotic Recombination Regions (MRRs and FRRs) are shown as alternating black/grey rectangular tracks beneath each chromosome. All 24 chromosomes exhibit clear sex-restricted recombination intervals as shown in the tracks labeled MR and FR, showing the cumulative proportions (range 0–1) of all recombination events on each chromosome for males (MR) and females (FR). The innermost track, labeled ΔAF, shows the male vs. female delta allele frequency for 500kb windows along each chromosome (range 0–0.36).

https://doi.org/10.1371/journal.pgen.1010011.g003

Even more surprising than the meiotic recombination desert observed for chr13 in females was the observation that all chromosomes clustered into more or less strict MRR/FRR regions (Figs 3B and S7A, S1 Table). We limited our analysis to the placed chromosome part of the genome (chromosomes 1–24), spanning 596 Mb out of the total 611 Mb, and defined FRRs and MRRs according to our classification outlined in S2 Table. Genome-wide, this annotation defined 407 Mb (68% of the sequence) as FRRs due to preferential recombination in females and 154 Mb as MRRs due to preferential recombination in males (26% of the sequence). We investigated what distinguishes an MRR from an FRR and concluded that for all chromosomes except chr9, MRRs are much smaller than FRRs while retaining approximately the same number of recombination events (S7B Fig). Only ~6% of the sequence fell outside of this classification because both sexes recombined at similar frequencies or because meiotic recombination was not observed in either sex.

Nucleotide diversity and repeat content are correlated with sex-restricted recombination

MRRs, with the exception of chr9, exhibit higher nucleotide diversity (π) than their FRR counterparts (Fig 4A). Recombination rate is correlated with π (S8 Fig) and chr13 is an outlier with the largest MRR over FRR π ratio among all chromosomes, presumably due to a bottleneck associated with the establishment of the new Y-chromosome. Furthermore, the pool-sequenced male and female individuals came from a single cross between one female and one male, meaning that less chrX and chrY genetic diversity was transmitted to female and male offspring, respectively in comparison to autosomes. Typically, MRRs occupy one end of a chromosome and FRRs start from the other end. Atlantic halibut has subtelomeric or acrocentric centromeres [23] and for all 18 chromosomes where we could infer centromere placement by mapping centromeric markers [20] to IMR_Hiphip.v1, the FRR coincided with the centromere (S3 Table). We next sought to determine whether meiotic recombination is sex-restricted also in species closely related to Atlantic halibut but could not find any publicly available pedigree datasets for other Righteye flounders. Since Atlantic halibut π was found to be higher for MRRs than FRRs, we decided to investigate whether π is correlated with MRR/FRR classification also in other Righteye flounder species. We could not find any publicly available genome sequencing data for Pacific halibut and instead used RNA-sequencing data to estimate π for this species. For Greenland halibut (Reinhardtius hippoglossoides) and European plaice (Pleuronectes platessa) we estimated π from publicly available RAD-seq- and whole genome shotgun sequencing data, respectively. The across-species comparison of π distribution in MRRs and FRRs revealed a striking agreement among the four species analyzed, with regions corresponding to Atlantic halibut MRRs having higher π (Fig 4A–4C), indicating that heterochiasmy has been conserved among righteye flounders.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Nucleotide diversity (π) and repeat content is correlated with sex-restricted recombination throughout the genome.

A π distribution in the Atlantic halibut genome, contrasting regions of male-only vs. female-only meiotic recombination, MRRs and FRRs, respectively. B Distribution of Pacific Halibut π estimates for regions mapped back to the Atlantic halibut genome, classified as either MRRs or FRRs. Chromosomal and MRR/FRR assignments were extracted from the Atlantic halibut genome assembly coordinates. C Genome-wide distributions of estimated π for Atlantic-, Pacific- and Greenland halibut and European plaice in regions corresponding to herein defined Atlantic halibut MRRs and FRRs. Numbers above MRR distributions indicate, within species, how much higher π is in MRR regions compared to FRR regions. * The time since divergence of Greenland halibut from Hippoglossus was estimated from [28]. ** Divergence estimate for European plaice was obtained from TimeTree [29]. D Genome-wide distributions of five repeat types in MRRs and FRRs. LINE and SINE represent Long- and short interspersed nucleotide elements, respectively. LTR indicates Long Terminal Repeats. LCR and SR indicate low complexity- and simple repeats, respectively. The Y-axis shows densities of different repeat types after Z-score transformations in relation to genome averages.

https://doi.org/10.1371/journal.pgen.1010011.g004

We investigated whether MRRs and FRRs differed for repeat content and annotated repeats in IMR_Hiphip.v1 by means of RepeatMasker [30] using the Actinopterygii database. RepeatMasker raw output was parsed by repeat type and the density per type was investigated for each chromosome (S4 Table). Comparison of repeat type density in MRRs and FRRs revealed a markedly higher density of Short Interspersed Nucleotide Elements (SINEs) in FRRs than in MRRs while an opposite pattern was observed for simple- and low complexity repeats (Fig 4D). We also observed that the Long Interspersed Nucleotide Element (LINE) density increases towards the FRR ends of chromosomes, while simple repeats and low complexity repeats show the opposite pattern, being enriched towards the MRR end (S9 and S10 Figs).

The insertion on chromosome Y is of transposon origin and contains a gsdf promoter

In an effort to identify candidate Structural Variants (SVs) between Y and X that may influence gsdf expression, we aligned all the ONT reads from the male individual used for genome assembly to the IMR_hipHip.v1 assembly and used the software Sniffles [31] to call SVs along the genome. This analysis revealed a large set of putative SVs (S5 Table), two of which coincided with the gsdf locus, both detected as heterozygous deletions in relation to the reference at chr13:8503901–8505128 (1,227 bp) and chr13:8509688–8509767 (79 bp). Manual inspection of these hits revealed that both ONT reads and linked read pseudo-haplotypes from the male assembly individual validate the assignments made by Sniffles. We used alignments of the male and female pool sequencing data and the phases of ONT long reads to conclude that the insertion allele at these two indels is present on chrY. The HiC data used for scaffolding the assembly was used to screen for large-scale SVs that may differ between X and Y. This analysis indicated that no large inversions are present on chr13 and that neither Y nor X are part of any large translocations with other chromosomes, as such events would appear as off-diagonal contacts in the contact map (S11 Fig).

The 1.2 kb insertion upstream of gsdf (INS1) is not found in the corresponding Pacific halibut locus, supporting that this region is derived on the Y chromosome in Atlantic halibut. INS1 (IMR_Hiphip.v1 chr13:8503902–8505127) is highly similar to 24 other loci in the genome assembly. Sequence alignments showed high sequence similarity starting with TG and ending with CA di-nucleotides in 19/24 loci. The remaining 5 loci showed strong, further extending similarities to each other, and we therefore extracted additional sequence flanking these five loci to perform a 7–8 kb alignment, which also terminated in conserved TG/CA di-nucleotides in four out of the five sequences. The TG/CA-ends are hallmarks of retrotransposon long terminal repeat (LTR) boundaries, and we could also identify 4–6 nucleotide sequence duplications immediately flanking the insertions, indicating target site duplications (TSDs) that confirm the transposable element (TE) insertions as retrotransposons. Although these TE hallmarks were obscured at the INS1 site upstream of gsdf, the sequence similarity and alignment to the related loci in the genome support its TE origin (S1 File). One of the five longer sequences (IMR_Hiphip.v1 chr14:3028070–3045206) was used as query to search the DFAM database [32]. This search revealed a 5254 nt long hit against the 5410 nt long Danio rerio Gypsy transposable element (Gypsy6-I_DR, Bit score 2342), supporting by sequence similarity association that INS1 was derived from a Gypsy retrotransposon that inserted in antisense orientation relative to gsdf. Strong similarities between the full-length Gypsy LTR sequences and INS1 revealed that INS1 is indeed an LTR remnant of an ancient Gypsy insertion. This association is further supported by RepeatMasker analysis of the five putative full-length hits partially matching INS1, as they all contained Gypsy sequence annotations.

From the ONT data used for genome assembly we could assess DNA methylation at the gsdf locus. Interestingly, we observed low DNA methylation within INS1, in a region coinciding with a predicted CpG-island (Figs 5 and S12). Further inspection of the lowly methylated CpG-island revealed predicted Sp1 binding sites indicating that this locus may act as a promoter region for gsdf. It is clear from both ONT read alignment and the linked read pseudo-haplotype alignments that the chrY haplotype carries INS1. We developed a PCR assay to amplify into INS1 and observed that this assay gave clear bands of the expected size for 10 adult phenotypic males while no bands were observed for 10 phenotypic females (S13 Fig). Expression of embryonal gsdf (28–54 dpf) from the Y-allele (gsdf-Y) extends further upstream of the transcription start site than gene models both in adult Atlantic halibut testis and in male- and female reproductive tissues in adult Pacific halibut. The longer gsdf-Y isoforms contain 5’UTRs extending into the Gypsy LTR INS1 (Figs 5 and S12), which supports promoter activity of INS1.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Chr. Y carries a transposon-derived LTR upstream of gsdf.

A Agreement between reference individual Chromium pseudo-haplotype assemblies and chrY-markers from Pool-seq of males and females. Black lines show markers along chr13 where pseudo-haplotype 1 (outer track) and pseudo-haplotype 2 (inner track) are in agreement with the Y-allele defined from pool-seq. In the interval 0-10Mb pseudo-haplotype 1 shows a striking agreement with Y-markers defined from pool-seq data. B Schematic representation of the Atlantic Halibut sex chromosomes as inferred from linkage analysis alone and from comparison of Pool-seq defined Y-markers with Chromium pseudo-haplotypes of the assembly male. X and Y haplotypes differ for a 1.2 kb segment derived from a transposable element 2 kb upstream of the gsdf transcription start site. Males carry the insertion allele which contains a lowly methylated CpG-island inferred to act as a derived promoter of gsdf uniquely in males. The shorter gsdf isoforms were observed in Atlantic halibut adult testis in as well as adult ovary and testis of Pacific halibut. The two gsdf gene models labeled chr13Y-specific indicate isoforms detected only in males from three separate developmental stages. An IGV-plot of this region is presented in S12 Fig.

https://doi.org/10.1371/journal.pgen.1010011.g005

Analyses of Y decay and X/Y allelic imbalance

In addition to the ONT genome assembly, we also sequenced the same male using a Chromium Genome linked read library and used the resulting data to generate two separate pseudo-haplotype (PH1 and PH2) assemblies which successfully phased the first 9–10 Mb of chr13 into separate scaffolds for the X and the Y chromosome (Fig 5A). Loss of X/Y recombination is expected to lead to chrY decay, and we investigated whether we could find evidence of this by contrasting coding changes against Pacific halibut gene models for chrX (PH2) and chrY (PH1) haplotypes along the first 9 Mb of chr13. Numbers of non-synonymous substitutions for the interval corresponding to Atlantic halibut X/Y (chr13: 0–9 Mb) were 68 and 70 for PH1 and PH2, respectively, suggesting little if any decay of the Atlantic halibut chrY (S14 Fig). That no large-scale loss of genetic material has occurred on chrY is supported by the observation that PH1 and PH2 are both 20.4 Mb in size. It is clear that the overall divergence, as well as non-synonymous and synonymous divergence is higher between species than between Atlantic halibut X and. Y (S14 Fig), showing that the Atlantic halibut sex-chromosome evolved after the split from Pacific halibut, i.e. < 4.5 MYA. Altogether, at 7611 single nucleotide alleles Atlantic halibut chrY differs from another allele shared among Atlantic halibut chrX and Pacific halibut. The majority of these (excluding the 68 nonsynonymous sites, giving 7543 sites) are expected to be neutral. Using the 7543 putatively neutral sites together with mutation rates determined in Atlantic herring (2x10^-9 per site per generation [33] and Cichlid fish (3.5x10^-9 per site per generation [34] and a mean generation time of seven years, we estimate that the age of the ~10Mb Atlantic halibut chrY-locus is at most 1.5–2.6 million years. This calculation should be interpreted as the absolute upper possible age and is likely a gross overestimation since Atlantic halibut π dictates that most of the 7543 observed differences could represent standing variation present at the time of the X/Y split, rather than mutations which occurred after the split. From the intersected RNA-seq alleles on chr13 (S4 Fig), we assessed allelic imbalance for 130 SNPs in the 7 male RNA-seq samples and six of these SNPs showed consistent allelic imbalance, three expressed in higher levels from X and three in higher levels from Y, including gsdf (S15 Fig).

Ethics statement

For this study the halibut were reared under conditions similar to standard commercial fish farming conditions. Such conditions are listed as an exception in The Norwegian Regulation on Animal Experimentation, thus approval of the experimental protocol of this experiment by NARA (the governmental Norwegian Animal Research Authority) was not required.

Halibut material

All samples were taken during routine production of Atlantic halibut at the Austevoll Research Station, Institute of Marine Research, Storebø, Norway. Halibut breeding individuals were stripped, eggs fertilized and incubated, and juvenile halibut produced according to standard aquaculture protocols developed at the Austevoll Research Station [54–56]. Briefly, females were stripped according to their individual ovulatory rhythms to ensure high (>90%) fertilization. Fertilized eggs were incubated in 250 l conical incubators with upwelling water for 60 day degrees (i.e. 10 days at 6°C) and transferred to larger incubation units, silos (vol. 50,000 l) before hatching. Yolk sac larvae were held in the silos until ca. 265 day degrees posthatch when they were moved to first feeding tanks and fed Artemia for 70 days until metamorphosis was complete. At this stage, the larvae were weaned onto a formulated diet. Juvenile halibut were moved to ongrowing tanks when they had reached an average size of ca. 10 cm and held there until they had reached a harvest size of 2–5 kg. One adult, three-year old male was euthanized with metacain (100 mg/l) and muscle, spleen and blood samples were taken for reference genome sequencing. For sequencing of pooled family samples, tissue samples were taken from ten full siblings of mixed sex (ten males and ten females). All samples were snap frozen in liquid nitrogen (N₂) and stored at -80°C before DNA extraction. RNA-sequencing was made on a series of samples taken through embryonic development. Before hatching, embryo samples were snap frozen in liquid N₂ and stored at -80°C. Larvae were euthanized in metacain (100 mg/l), transferred into RNA later, kept at 4°C for 24 h and stored at -20°C until RNA isolation. The samples for qPCR were snap frozen in liquid N₂ and stored at -80°C. Digital images of all embryo-larval stages were captured using video cameras Moticam 1080 (Motic) mounted on Olympus SZX10 stereomicroscope using Motic Live Imaging Module. Living embryos and larvae were individually immobilized in a Petri dish containing 2% methylcellylose/98% seawater. Images were then scaled and processed by ImageJ software [57].

DNA extraction and sequencing of male and female pools

DNA was extracted from muscle samples using Qiagen DNeasy Blood and Tissue Kit. Paired-end libraries of equimolar male (10 individuals) and female (10 individuals) DNA pools and from one single individual of each sex were constructed using the Genomic DNA Sample Preparation Kit (Illumina) according to manufacturer’s instructions and sequenced on the Illumina HiSeq2000 platform (Illumina) at the Norwegian Sequencing center (https://www.sequencing.uio.no, Oslo, Norway).

High Molecular Weight DNA extraction and sequencing

Several tissues were collected from a single male Atlantic Halibut. Muscle and spleen were snap frozen in liquid nitrogen and stored at -80°C until further use. Fresh blood was used to isolate erythrocyte nuclei and then prepare High Molecular Weight (HMW) DNA from these nuclei according to a Phenol/Chloroform protocol (Quick). Extracted DNA was either left unfragmented or sheared using either MegaRuptor (Diagenode) or by passing DNA through a 100μl pipette tip 10–20 times. From these DNA samples we produced MinION sequencing libraries with the standard adapter ligation protocol kit (LSK-108), 1D^2 sequencing kit (LSK-308) and the RAD002 and RAD003 kits (Oxford Nanopore Technologies). For libraries prepared with the RAD002 and RAD003 kits we used a protocol developed to produce ultra–long reads (Quick, protocols.io). In total we used 9 flow cells and sequenced 18.8 Gbp in 2,963,108 reads with a read N50 of 17.8 kb. Sequenced fast5 raw signal data files were basecalled using albacore v.2. The size distribution of sequenced reads is presented in S1A Fig. We also used the collected muscle samples to produce HiC libraries (Dovetail), sequenced on one lane of HiSeq X (Illumina). One of the DNA preps inferred to have HMW DNA was used to prepare a linked reads Chromium library (10x genomics) and was sequenced on one lane of HiSeqX (Illumina).

Genome assembly, polishing and scaffolding

Nanopore reads passing filters were used in assembly by first constructing all-vs-all alignments between all reads using minimap2 [58]. These all-vs-all alignments were then used in the software miniasm to produce a haploid contig assembly. The contig assembly was polished using racon [59] with the nanopore data followed by pilon [60] using the Illumina data from the male pool. Scaffolding of the polished miniasm assembly was performed using the HiRise pipeline (Dovetail Genomics). The quality of the generated assembly was assessed using BUSCO [61]. Initial nanopore contigs assembled with miniasm [58] had a contig N50 of 3.2Mb and the resulting final assembly had a scaffold N50 of 27.2 Mb.

Assembly of linked read data

The software Supernova2 [62] (10X Genomics) was used to produce a genome assembly of the sequencing data resulting from the Illumina sequencing of the Chromium library. For this assembly, reads were subsampled to get close to the recommended 56x genome coverage recommended in the Supernova2 user manual. The two pseudo-haplotype linked read assemblies featured one scaffold each which covered the whole of IMR_Hiphip.v1 chr13. We used SNP markers identified as divergent between the male and female short read data to investigate the pseudo-haplotypes in relation to sex divergent markers, i.e. if one pseudo-haplotype could represent the Y chromosome and the other the X-chromosome. This analysis revealed that for the first 10 Mb of chr13, the allele being heterozygous in males and not present in females (the perceived Y-allele) was observed on pseudo-haplotype number 1 (1,954 SNPs). The perceived Y-allele was observed only for 2 SNPs in this region for pseudo-haplotype number 2. The corresponding counts for the rest of chr13 was 125 and 213 alleles on pseudo-haplotype numbers 1 and 2, respectively.

RNA-sequencing and gene annotation

Total RNA was extracted from whole larvae from seven different stages of development using miRNeasy Mini Kit (Qiagen) and treated with TURBO DNase-Free kit (Ambion). TruSeq cDNA libraries were sequenced on the HiSeq 4000 sequencing platform (Illumina). The average number of reads was 50,320,639 per sample. Twenty-two samples (four samples from 54 dpf and three samples from each other stage) were assembled individually with Trinity using the Agalma pipeline [63]. The transcriptome assemblies were annotated with BLAST against Swissprot (download date 20191227) and the Ensembl gene models of zebrafish (GRCz11) and stickleback (BROAD S1). The hits with the best scores were kept. Transcripts from all assemblies were mapped to the halibut genome with BLAT. The thresholds for mapping were set to 98% match for sequences with length > = 401 nt and 95% match for sequences with length < = 400 nt. Long sequences were mapped before short sequences. The mapped sequences were clustered on genome location, selecting the sequence with the highest score for each non-overlapping cluster. Only clusters with an annotation (>300 in BLAST score threshold) were saved. Neighboring clusters annotated with the same gene from either Swissprot, zebrafish or stickleback were merged in cases where the annotation indicated that the gene was split in several clusters (the neighboring clusters were only merged in cases where the annotation was from different parts of the same gene). For the merged clusters, the sequence with the highest BLAT score from all merged clusters were chosen. Additionally, the transcriptome assembly sequence with the highest scoring BLAT match to the merged cluster was added. We found 15,273 annotated genes based on the Trinity assemblies. The zebrafish and stickleback ENSEMBL gene models were mapped to the halibut genome using exonerate in order to add genes without RNA-seq evidence. The exonerate hits were clustered by location, using the exonerate hit with the highest score for each cluster. Sequences for each exonerate based gene were extracted from the halibut genome sequence. 2,269 genes were added based on the exonerate analysis. We added 222 genes that had no annotation (BLAST threshold score <300) but had RNA-seq expression. After running BUSCO 3.1 with the Actinopterygii odb9 dataset on the non-filtered Trinity assemblies that mapped to the genome [61], we added 550 genes from the BUSCO analysis that were missing from our filtered list. This resulted in a total number of 18,314 genes and 20,579 transcripts giving a 80.9% complete, 8.7% fragmented and 10.4% missing BUSCO coverage.

Nanopore sequencing of testis RNA

Testis tissue was homogenized at 4°C in 1ml QIAzol Lysis Reagent (Qiagen, Hilden, Germany) together with RNase-Free Zirconium Oxide Beads (NextAdvance, Inc., Troy, NY, USA) during 1 min, maximum effect, in a BulletBlender (NextAdvance) and total RNA was prepared according to the QIAzol (Qiagen) protocol. Polyadenylated RNA was isolated with the Dynabeads mRNA direct purification kit (Thermo Fisher Scientific), following the kit protocol. For full-length cDNA Oxford Nanopore sequencing, the cDNA-PCR Sequencing SQK-PCS108 kit (Oxford Nanopore Technologies) was used according to the kit protocol with minor modifications. The finished cDNA library was sequenced on a MinION device using a R9.4 flow cell (Oxford Nanopore Technologies).

Screening for male vs. female divergence

Illumina sequences from the male- and female DNA pools were filtered by trimmomatic [64], then aligned to the assembly IMR_hipHip.v1 using BWA-MEM [65]. Aligned reads were subjected to duplicate flagging (“Picard Toolkit.” 2019. Broad Institute, GitHub Repository. http://broadinstitute.github.io/picard/; Broad Institute) and were then used as input for calling SNPs as well as small Insertions/Deletions using Genome Analysis Toolkit (GATK) [66] UnifiedGenotyper. Raw variant vcf files were subjected to variant filtration using GATK Best Practices procedure with SNP-clusters being added as an additional filter (more than 5 SNPs in a window of 20 bases were filtered out). SNPs passing filters (n = 2,500,656) were isolated from the vcf file and allele frequencies from the male- and female pools were extracted and used for calculating male vs. female ΔAF, first calculated for each individual SNP having read depths > = 10, and average ΔAF was then calculated for 500kb windows along each chromosome.

Sex differentiating PCR assay

Genetic sex of each embryo was determined by PCR using sex-specific primers (Table 1). The thermal cycling conditions were as follows: 94°C for 5 min, 32 cycles with 94°C for 30 sec, 65°C for 1 min and 72°C for 1 min, and finally 72°C for 5 min. Each 12 μl PCR reaction contained 5X GOTaq Flexi Buffer, MgCl, dNTPs, GoTaq Polymerase, PCR primers and template cDNA.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Brx PCR primers.

https://doi.org/10.1371/journal.pgen.1010011.t001

Differential gene expression between sexes

We generated RNA-sequencing data from 12 Atlantic halibut whole embryos representing stages 28 (n = 4), 43 (n = 4) and 54 (n = 4) dpf. Since one cannot determine the phenotypic sex at this stage we aligned the RNA-seq data using Bowtie 2 [67] to IMR_hipHip.v1 and extracted expressed allele counts at positions identified as divergent between males and females in the pool DNA-seq. Extracted RNA-seq allele ratios were used to cluster the individuals, revealing two well-separated clusters of XY and XX individuals and this assignment was used to identify genetic male and female individuals for the differential expression screen. The differential expression screen involved contrasting across-sample normalized gene expression levels of the seven inferred males to the five inferred females in a two-tailed t-test. Principal Component Analysis (PCA) was performed, using the R function “prcomp”, on the normalized gene expression levels from the 12 embryos.

gsdf quantitative PCR

Genomic DNA and total RNA was extracted from single embryos using the Allprep DNA/RNA/miRNA Universal Kit (Qiagen), according to the manufacturer’s instructions. The RNA was DNase treated as part of the extraction process, according to the manufacturer’s instructions. For each sample, cDNA was synthesized from 125 ng RNA using the Superscript VILO cDNA synthesis Kit (ThermoFisher Scientific). The primers used for qPCR were designed using BatchPrimer3 (https://wheat.pw.usda.gov/demos/BatchPrimer3/) (Table 2). The gene gtf3c6 was used as an internal reference gene for normalization of the qPCR data due to its stable expression in the developmental stages analysed in this study (S17 Fig). Using different RNA concentrations, the PCR efficiencies (E = -1+10(-1/slope)) for the primers were calculated to be 108.5% (gsdf) and 92.4% (gtf3c6). A melt curve analysis was performed to confirm that the primers did not amplify more than one PCR product. qPCR was performed in duplicates in 384-well optical plates in a QuantStudio 5 Real-Time PCR system (ThermoFisher Scientific) using default settings. 2 μl cDNA (diluted 1:20) was used in a 6 μl Fast SYBR Green qPCR reaction (ThermoFisher Scientific). No-template controls for each gene were run in all qPCR plates. The relative gene expression level was calculated using the comparative Ct (or 2^−ΔΔCt) method. All values were normalized to gtf3c6 and calibrated to the sample with the lowest ΔCt.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Primers used for qPCR.

https://doi.org/10.1371/journal.pgen.1010011.t002

Detection of structural variants, small insertion/deletions and annotation of repeat sequences

The long-read SV identification tool Sniffles [31] was used to identify SVs from the ONT data. The ONT data from the single male individual used for genome assembly was aligned to IMR_hipHip.v1 using minimap2 [58] with the “-x map-ont” option and the resulting bam file was used in sniffles, requiring > = 5 observations of a variant allele to be retained as a potential SV.

Analysis of the evolutionary origin of the 1.2 kb insertion upstream of gsdf

The 1.2 kb insertion (IMR_Hiphip.v1 chr13:8503902–8505127) was used as a query sequence in BLAT [68] searches against IMR_HipHip.v1. Full length hits were retained (n = 30) and were extended to include 2kb up- and downstream sequences and these approximately 5.3 kb sequences were used as input for multiple sequence alignment using MUSCLE [69]. The resulting alignments were visualized in AliView [70] and abrupt drops in sequence similarities were observed for 24 of the sequences on each flank. The extent of highly similar sequences started and ended at perfectly conserved TG and CA di-nucleotides, the hallmark motifs of LTR boundaries, and the insertion was immediately flanked by 4–6 nucleotide duplications of loci TE target sites (TSD) confirming the LTR-retrotransposon insertion.

DNA methylation calling from ONT data and CpG island prediction

Basecalled ONT reads previously used for the genome assembly were aligned back to the IMR_Hiphip.v1 assembly using minimap2. The software nanopolish [71] was used to index the raw ONT fast5-files to enable extraction of likelihoods of CpG site methylation along individual mapped reads in genome assembly coordinates. Next, the nanopolish script “calculate_methylation_frequency.py” was used to extract the methylation frequency at each adequately covered CpG cluster in the assembly. CpG islands were predicted for the IMR_Hiphip.v1 assembly using gcluster [72] employing default settings.

Estimation of nucleotide diversity

For Atlantic Halibut we estimated π by creating chromosome-specific samtools mpileup files from the aligned data previously used for variant calling, using only positions with a mapping quality > = 20 and read depths ranging between 10 and 50. The script “WinNuclDivmPileup.py” (https://github.com/cjrubinlab/python_scripts2) was used to estimate π in 500kb windows along the genome for the male- and female pools. For Pacific halibut no DNA-sequencing data was available in public sequence repositories, so we devised a way to estimate π from RNA-sequencing data. Sequencing data from 12 RNA-seq experiments of various tissues and individuals was downloaded from the SRA (SRR11826178-SRR11826189) and these data were aligned to the Oxford nanopore derived Pacific Halibut genome assembly (GCF_013339905.1) using the STAR aligner [73]. The resulting alignments were randomly subsampled 20 times and then subjected to duplicate-pair removal using picard-tools MarkDuplicates, retaining 70% of the reads after filtering. From the bam file containing filtered reads an mpileup file was generated using samtools (mapping quality exclusion filter < 20). For the Greenland Halibut (Reinhandtius Hippoglossoides) we downloaded the genome assembly from NCBI (GCA_006182925.2_MU_Rhippo_v1) as well as RAD-sequencing data from 20 individuals (SRA accessions: SRR11351808-SRR11351827)[74], aligned the short-read data to the assembly using minimap2 with the “-x sr” option. From mpileup files derived from samtools mpileup for Pacific Halibut and Greenland halibut the estimated heterozygosity (He) was calculated for positions having coverage> = 10 and mapping quality > = 20 using the WinNuclDivmPileup.py” script. For European plaice (Pleuronectes platessa), a species lacking a genome assembly, we downloaded paired genome resequencing Illumina reads (SRA accession: SRX8240416), aligned these reads to the IMR_Hiphip.v1 assembly using BWA-MEM and proceeded with the same pipeline as was used for estimation of Atlantic halibut π. In order to lift over genomic intervals annotated as MRRs and FRRs in the Atlantic halibut to the Pacific- and Greenland halibut assemblies we used the satsuma [75] commands “Chromosemble” to generate a detailed list of lift-over intervals and “BlockDisplaySatsuma” to summarize the detailed list into coherent blocks. We then converted these coordinates to the bed-format and used the bedtools [76] “intersectbed” command to link coordinates in the other assemblies to the corresponding coordinates in the Atlantic halibut assembly and MRRs and FRR classification.

Linkage analysis

RAD-sequencing data from an Atlantic halibut half-sib pedigree altogether comprising three parents and 90 offspring were downloaded from the SRA (accession = SRX193134) [19]. The downloaded individual files were split by barcodes and were then connected with annotation information from the original publication regarding phenotypic sex, parent/offspring status and family B/C [19]. The paired-end data from each individual in the pedigree was then aligned to IMR_hipHip.v1 using BWA-MEM. Resulting bam-files were subjected to duplicate removal using picard-tools “MarkDuplicates”, then raw variant (SNP and indel) calling using GATK UnifiedGenotyper. Raw SNPs were filtered using GATK “VariantFiltration” with the following arguments, also with a list of raw indel positions given to the “InDel” mask; “—maskName InDel”, “—maskExtension = 5”, “—clusterSize = 5”, “—clusterWindowSize = 15”, "BaseQRankSum<-7.0", "BaseQRankSum > 7.0", "MQRankSum < -7.0", "MQRankSum > 7.0", "QD < 5.0","FS > 50.0","ReadPosRankSum > 7.0", "ReadPosRankSum < -7.0" and "DP > 40". SNPs remaining after filtration were saved in a new vcf file which was given as input to Lep-MAP3 [77] to generate recombination maps for the two families. The commands used in Lep-MAP were: “ParentCall2”, with the options (removeNonInformative = 1 XLimit = 2 halfSibs = 1), then “Filtering2”, with the option data = - (dataTolerance = 0.01), then “SeparateChromosomes2” with the options (lodLimit = 10 distortionLod = 1) to identify linkage groups in the pedigrees. SeparateChromosomes2 with lodLimit = 10 generated a map of 31 linkage groups, out of which 20 had 964–3063 markers assigned to them (24 were expected based on karyotype). We continued with the output from SeparateChromosomes2 (lodLimit = 10) to run “OrderMarkers2” to, for each identified linkage group, order the markers without having given the software access to genome assembly coordinates. The output files from the first iteration of “OrderMarkers2” were then subjected to a second round of “OrderMarkers2”, this time giving the Lep-MAP3 access to genome coordinates of markers on each linkage group using the option “evaluateOrder”. The resulting linkage group specific files containing male- and female centiMorgan (cM) positions of each marker were filtered to prune obviously inflated cM positions in the very beginning and the very end of assembled chromosomes and by splitting four linkage groups where two assembled chromosomes had been merged into one linkage group, thereby increasing the observed linkage group count from 20 to the expected 24. Resulting maps from this stage were retained for downstream analyses.

Analysis of chrY decay

The two linked-read pseudo-haplotype assemblies (PH1 and PH2) of the male Atlantic halibut used for the genome assembly were aligned against the nanopore assembly and it was concluded that scaffold 45 from the two pseudo-haplotypes corresponded to chr13. Scaffold 45 from PH1 and PH2 were independently converted to overlapping short reads which were separately aligned using BWA-MEM to the Pacific halibut reference genome. SNPs and small indels were called using bcftools and these were annotated using SNPEff, with gene models derived from the Pacific halibut IPHC_HiSten_1.0 gtf-file “GCF_013339905.1_IPHC_HiSten_1.0_genomic.gtf”. SNPEff “missense” and “synonymous” annotations were counted for SNPs that were uniquely observed in one pseudo-haplotype Annotation frequencies were compared between the two pseudohaplotypes for the interval where they represent the X and Y (Atlantic halibut: chr13:0–9 Mb, Pacific halibut: CM023513.1:11–21 Mb).

Pacific halibut RNA-seq contrasts

Sequencing data from 12 RNA-seq experiments of various tissues and individuals (SRR11826178-SRR11826189) was downloaded from SRA. Four of these tissues (testis, ovary, white muscle and brain) were used for characterization of gsdf gene expression levels and validation of gene models from the gff-file accompanying the Pacific halibut assembly on NCBI. Alignments were performed using STAR as described above for π estimation.

RepeatMasker analysis

RepeatMasker v4.1.0 [30] was used to annotate both simple and interspersed repeats. We deployed the RMBLASTN (v2.6.0.+) search engine to scan for known repeats among ray-finned fishes (“-species actinopterygii”) using a database that combined Dfam_3.1 and RepBase-20181026 repeat sequences. A custom Perl script was used to parse the main output table from RepeatMasker and compute the densities of major repeat classes (i.e. “Simple_repeat”, “LINE”, “LTR”, “Low_complexity”, “DNA”, “RC”, “SINE”, “Satellite”, “Unknown”, “Retroposon”) in windows of 50 kbp and 500 kbp, respectively.