The evolution of huge Y chromosomes in Coccinia grandis and its sister, Coccinia schimperi

Abstract

Microscopically dimorphic sex chromosomes in plants are rare, reducing our ability to study them. One difficulty has been the paucity of cultivatable species pairs for cytogenetic, genomic and experimental work. Here, we study the newly recognized sisters Coccinia grandis and Coccinia schimperi, both with large Y chromosomes as we here show for Co. schimperi. We built genetic maps for male and female Co. grandis using a full-sibling family, inferred gene sex-linkage, and, with Co. schimperi transcriptome data, tested whether X- and Y-alleles group by species or by sex. Most sex-linked genes for which we could include outgroups grouped the X- and Y-alleles by species, but some 10% instead grouped the two species' X-alleles. There was no relationship between XY synonymous-site divergences in these genes and gene position on the non-recombining part of the X, suggesting recombination arrest shortly before or after species divergence, here dated to about 3.6 Ma. Coccinia grandis and Co. schimperi are the species pair with the most heteromorphic sex chromosomes in vascular plants (the condition in their sister remains unknown), and future work could use them to study mechanisms of Y chromosome enlargement and parallel degeneration, or to test Haldane's rule about lower hybrid fitness in the heterogametic sex.

This article is part of the theme issue ‘Sex determination and sex chromosome evolution in land plants’.

1. Introduction

Only 47 out of thousands of dioecious seed plant species are known to have microscopically dimorphic sex chromosomes [1]. This number is surely an underestimate because almost half of the dioecious seed plants are concentrated in just a few clades, many of them tropical and poorly known cytogenetically, including Diospyros (700 species), Pandanus (600 species), Salix/Populus (500 species), Litsea (400), Calamus (370), Dioscorea (350), Baccharis (350), Piper (300 of its 1000 species), Macaranga (300), Ilex (300), Clusia (250) and even entire families, such as Menispermaceae, Myristicaceae, Picrodendraceae and Rafflesiaceae [2]. Because of this reason, and also because we lack densely sampled phylogenies for most genera with dioecious species, there are few study systems in which sister species pairs are reliably known and for which living material is available for cytogenetic studies.

Large non-recombining regions in heteromorphic plant sex chromosomes have been studied mostly in Silene latifolia [3,4], Coccinia grandis [5,6] and Cannabis sativa [7]. The number of sex-linked contigs (genes) has been estimated as greater than 1700 for S. latifolia based on segregation analysis of RNA sequencing data, sampling three males and three females from brother–sister matings [8]. Study in a larger population of S. latifolia (parents plus 20 sons and 32 daughters) revealed 931 genes with both X- and Y-linked alleles, including 201 genes with known positions in a genetic map [3]. For Ca. sativa, the number of sex-linked genes has been estimated as greater than 500 based on segregation analysis using the same approach on RNA-sequencing (RNA-seq) data from a full-sibling family of two parents, five sons and five daughters [7], and for Co. grandis, as greater than 1300, also from two parents, five sons and five daughters [6].

Coccinia is a genus of Cucurbitaceae with both homomorphic and heteromorphic sex chromosomes and a complete species-level phylogeny that comprises all its 25 species, all dioecious and all endemic in Africa except for Co. grandis [9]. The latter is a perennial climber that ranges from India to tropical China and Australia, as well as most tropical islands worldwide (figure 1a). This aggressive and often weedy species can produce fruits in the first year and can reach stem diameters of greater than 8 cm and ages of at least 20 years. The genus Coccinia belongs to the tribe Benincaseae (which includes Cucumis and Citrullus) where it is sister to Diplocyclos, a genus of four monoecious species [10]. Indian and Hawaiian plants of Co. grandis have a large Y chromosome that comprises about 200 Mb, mainly owing to the accumulation of transposable elements (TEs) and satellite repeats, resulting in a 10% difference in the size of male and female genomes [5,11–13]. No plants from other parts of the range of Co. grandis have so far been studied cytogenetically.

Figure 1.

(a) Range of Coccinia schimperi and Co. grandis based on 629 herbarium collections (modified from [9]); (b) Co. schimperi, coast of Kenya, photo: Q. Luke. (c) Co. grandis, Tongatapu, Tonga, Polynesia, photo: Wikicommons.

The evolution of the large Y chromosome in Co. grandis raises questions about the chromosomes of its closest relatives. Three species other than Co. grandis have been studied with molecular-cytogenetic approaches (Coccinia hirtella, Coccinia sessilifolia, Coccinia trilobata) and all have homomorphic chromosomes [13], but none of them is close to Co. grandis. The next-closest relative of Co. grandis known prior to the present investigation appeared to be Coccinia ogadensis from Ethiopia and Somalia [9,10], which is not in cultivation and whose chromosomes have not been studied.

Coccinia grandis plants from Asia have white flowers (figure 1), while plants from Africa and the Arab Peninsula have pale-yellow flowers. The white-flowered form is a noxious weed in Hawaii, many tropical islands and Western Australia, and is an important target of biological control research. The yellow-flowered form was long considered a separate species, Coccinia schimperi Naudin, but in 2015 this name was synonymized under Co. grandis [9]. This was based on the view that flower colour was insufficient to treat these entities as separate species, especially since classic experimental hybridizations of greenhouse-grown individuals resulted in viable offspring. The classic hybridization work that Holstein [9] relied on was carried out by Charles Naudin [14,15], a pioneer of research on hybridization [16]. Naudin crossed white-flowered Asian Co. grandis and pale-yellow-flowered African Co. schimperi and found that two hybrids that flowered (but one month apart), one male and one female, were intermediate in flower colour between the parents. When a male F₁ hybrid was backcrossed (in another year) with an Asian Co. grandis female, the sole offspring that flowered, a female plant, had pure white flowers. Naudin lacked sufficient plants for additional reciprocal crosses he had intended to carry out. Modern nuclear gene trees that include representatives of all species of Coccinia show that Asian Co. grandis, represented by a plant from Thailand, are sister to plants from Tanzania and Sudan [10], but with an impressive branch length difference, indicating numerous substitutions.

Because the close relationship between Co. grandis and Co. schimperi is unambiguous (based on the mentioned nuclear gene tree in [10]), we decided to obtain seeds from a yellow-flowered population (to represent Co. schimperi) for cytogenetic studies. To throw light on the evolution of sex chromosomes in these species, we use DNA and RNA sequence data to construct genetic maps for Co. grandis males and females and to test if Co. grandis X- and Y-alleles group together as would be expected if recombination arrest evolved after the reproductive isolation of the two species Co. schimperi and Co. grandis.

2. Material and methods

(a) . Plant material used in this study

Plants of Co. grandis (L.) Voigt (incl. Coccinia indica Wight & Arn., nom. illeg.) were grown from seeds in the greenhouses of the Botanical Garden Munich. Some of the original seeds were collected in 2011 on the Campus of Kakatiya University, Vidyaranyapura, northern Bangalore, state of Warangal; others were collected in 2007 in Pearl City, Oahu, Hawaii. Several herbarium vouchers of male and female plants have been deposited in the herbarium of Munich (acronym M), including Renner 2901, Renner 2902, Holstein 12 and Holstein 14. Plants of Co. schimperi Naudin were raised in June 2021 in a greenhouse at the Institute of Biophysics in Brno from seeds originally collected in the Al Baha region of Saudi Arabia (voucher S. S. Renner 2912, herbarium M).

(b) . Transcriptome data and assembly

For Co. grandis, we downloaded RNA-seq data from Indian plants (herbarium voucher: Tripura University Campus, Karmakar 433) from a study by Devani et al. [17] deposited in the NCBI Sequence Read Archive. Specifically, we downloaded and merged raw reads from early stage female flower buds (SRR5811941; SRR5811942) and early stage male flower buds (SRR5811937; SRR5811938). The sequences were trimmed with Trimmomatic [18] to remove adapters and low-quality sequences. Trimmed reads were assembled using two assemblers, Trinity [19] and rnaSPAdes [20]. The Trinity output consisted of 106.7 megabase pairs (Mbp) and 105 492 sequences; the rnaSPAdes output had a size of 74.1 Mbp and 137 869 sequences. Default settings were used for both assemblers. Redundancies were removed from the assemblies using EvidentialGene [21], resulting in smaller assemblies (Trinity: 40.2 Mbp, 34 984 sequences; rnaSPAdes: 35.0 Mbp, 29 998 sequences). To get the best possible reference transcriptome, we merged the outputs of Trinity and rnaSPAdes, and again used the EvidentialGene pipeline to remove redundancy, resulting in an assembly of 44.0 Mbp and 42 175 sequences. The benchmarking universal single-copy orthologues (BUSCO) approach was used to explore completeness of the transcriptome according to conserved orthologue content [22]. The BUSCO output [C:93.4%[S:92.5%,D:0.9%],F:4.7%,M:1.9%,n:1614] showed that out of 1614 BUSCOs for embryophytes, 1507 full-length BUSCOs were detected in our de novo-assembled Co. grandis transcriptome, indicating 93.4% completeness. Simultaneously, the assembly shows low level of redundancy 0.9% duplicated BUSCOs. Only 4.7% of BUSCOs were incomplete and 1.9% were missing.

For Co. schimperi and other species of Cucurbitaceae, we used RNA-seq assemblies from Guo et al. [23] deposited in figshare (links: https://figshare.com/s/9c8cc7464ae06b40e0c2 and https://figshare.com/s/111835077bd85f68d659, accessed 25 September 2020), most importantly, Diplocyclos palmatus, a representative of the sister genus to Coccinia. Cucumis melo (assembly by Guo et al. [23]) and Citrullus lanatus (our Trinity assembly based on sample SRR9317454 from the NCBI Sequence Read Archive) were included as representatives of outgroup genera [24]. Guo et al. do not cite herbarium vouchers for their material, but the senior author, Hong Ma, provided photos of Coccinia spec. 1 = XWB13686, cultivated in the greenhouses of the Wuhan Botanical Garden and indicated that the seeds were collected in Kenya. The plant, which has since died, has yellow flowers that are easily recognized as female flowers and that are typical of Co. schimperi in all aspects.

(c) . Identification of sex-linked sequences

To identify sex-linked genes, we used the model-based SEX-DETector approach [25], which uses RNA-seq or DNA-sequencing (DNA-seq) data to genotype parents and their offspring and to infer sex-linkage or autosomal segregation types, both of which have typical patterns. For example, a bi-allelic single nucleotide polymorphism (SNP) in which one allele is transmitted exclusively from father to sons, while the other is transmitted from both parents to all offspring will be identified as an X/Y SNP (with the male-specific allele being the Y allele). The information of all SNPs in a gene is then combined into a probability for the gene being sex-linked or autosomal. For this study, we had available cleaned Illumina paired-end reads (average depth of approx. 10-fold) for 102 Co. grandis individuals (mother, father and 100 full siblings of which seven were males and 93 females) obtained from the seeds in three fruits from a Hawaiian mother pollinated with pollen from an Indian father in a greenhouse in Munich (these data are part of an ongoing Co. grandis genome assembly and will be uploaded in early 2022). The DNA-seq data for 16 individuals (parents, the seven sons and seven randomly selected daughters) were used as input for SEX-DETector with the X/Y settings. For each of the 16 individuals, its DNA-seq reads were mapped to our Co. grandis RNA-seq reference genome (see Transcriptome data and assembly, above) using the Subread aligner [26]. Individuals were genotyped by reads2snp [27], using the default settings (two independent runs, -min6 and -min10) and the steps recommended in the SEX-DETector manual.

(d) . Genetic map construction

For a genetic map, the DNA-seq reads of all 102 Co. grandis individuals (including the 16 analysed in SEX-DETector) were mapped to our Co. grandis RNA-seq reference genome as above and genotyped by reads2snp in nine separate batches to reduce computer grid overload. We used the fix_reference_genotype.py script on the resulting vcf file (https://gist.github.com/wdecoster/d7fa440a74afd4607bb321ae0986fccd). The fixed vcf files were then merged using bcftools. Genetic maps of all chromosomes except the sex chromosomes were constructed using the genotypes given by reads2snp. To map the sex chromosomes, SNPs of sex-linked sequences found with SEX-DETector were extracted from the reads2snp output and divided into Y-linked SNPs and X-linked SNPs. All extracted SNPs were manually inspected using the Integrative Genomics Viewer (IGV; [28]) and evaluated according to the number of reads (greater than 10), fraction of less abundant variants (greater than 20%), and patterns of SNPs in their neighbourhood (to exclude regions with homology to multiple regions in the genome). For each SEX-DETector-identified sex-linked contig, the genotype of the individual was determined based on an IGV evaluation of all SNPs in the respective contig. The contigs lacking informative SNPs were excluded from the dataset.

Separate datasets were generated for the construction of the male map and the female map (only markers heterozygous in only one parent were retained). In the final table of genotypes, the columns corresponding to the manually curated sex-linked SNPs were combined with the columns from the rest of the dataset with the autosomal and pseudo-autosomal SNPs, and the tabulated genotypes then became the input for a Lep-MAP3 analysis [29], using a custom script based on the genotypes2post.awk script from Lep-MAP2 package [30]. The mapping of the sex chromosomes was performed in two rounds. After the first round of mapping, we tested the dataset for problematic markers with the Genetic Map Comparator web application [31] in R [32], which revealed 32 problematic markers. These were excluded from the final analysis. Final maps were drawn with LinkageMapView [33] in R.

(e) . Phylogenetic analyses, computation of sequence divergence and time estimation

We built RNA nucleotide alignments using the X-linked and Y-linked alleles identified in SEX-DETector including as many individuals and species as possible for which orthologues from other cucurbit species were available using reciprocal BLASTn search [34] and RBH-v1.py script (Hong Qin, https://github.com/hongqin/Simple-reciprocal-best-blast-hit-pairs/blob/master/RBH-v1.py). This resulted in the inclusion of our Co. grandis males and females, the Co. schimperi female, Co. trilobata, Coccinia spec._GJ090 (from Africa, but without a voucher), D. palmatus, Cu. melo and Ci. lanatus. Alignments were prepared with MACSE [35] and treated with Gblocks [36] to remove poorly aligned positions. Ambiguous nucleotides and stop codons were removed using a custom script. Alignments shorter than 300 bp were excluded from further analyses. Phylogenies from each alignment were obtained with MrBayes 3.2.7a [37], setting ‘nst' to ‘mixed', which results in Markov chain sampling over the space of all possible reversible substitution models. Markov chain Monte Carlo chains (MCMC) were run for 5 million generations, sampling every 1000th generation, with the burn-in set to 20%. The runs were monitored using Tracer [38], and summary figures were prepared in Figtree [39].

Using the R package PhySortR [40], the MrBayes tree file output from each of the 360 separate alignments of X-linked and Y-linked gene copies (remaining after all the preceding steps) was sorted into two sets based on the position of Co. schimperi X relative to Co. grandis X- and Y-alleles (0.95% support demanded). A small number of genes was excluded from further analysis (20 grouped all Co. grandis X, Co. grandis Y and Co. schimperi X-alleles without resolving their relationships, and in three gene trees, Co. grandis X, Co. grandis Y, and Co. schimperi X-alleles failed to form a clade). Species trees were then inferred for each of the two sets separately using the BUCKy program, which implements Bayesian concordance analysis [41]. Independent analyses were run using different settings for the concordance parameter alpha, viz. α = 1, α = 10 and α = 1000. The first 20% generations of each input file were discarded. All analyses were run with four MCMC chains with 1.1 million of MCMC updates, with the burn-in set to 10%. A consensus tree for each of the two sets of trees was constructed using the SumTrees program from the DendroPy Python library [42]. Graphs summarizing the results of the phylogenetic and mapping analyses were produced using ggplot2 [43] in R.

An independent phylogenetic analysis was carried out using coding sequences mapped to autosomal linkage groups identified in Co. grandis by Lep-MAP3 [29]. We again searched for orthologous sequences in related species and included Co. grandis males and females, our Co. schimperi female, Co. trilobata, Coccinia spec._GJ090, D. palmatus, Cu. melo and Ci. lanatus. The orthologous sequences showing a single copy in each of the studied species were isolated using Orthofinder [44]. We were able to include 1740 single-copy autosomal orthogroups with alignments longer than 300 bp after treatment with Gblocks.

Synonymous sequence divergence (dS) values were computed using codeml from the PAML package [45]. To translate dS values into absolute time, we used the divergence time of Diplocyclos and Coccinia (15 ± 3 Ma) estimated in a fossil-calibrated chronogram for the Cucurbitaceae as a calibration point [46]. Separate estimations of divergence times were calculated for the Co. grandis X and the Co. grandis Y based on mean synonymous divergences between Diplocyclos and Co. grandis X, Diplocyclos and Co. grandis Y, and Diplocyclos and Co. schimperi X. We used only the 27 genes that yielded trees with the grouping (Co. grandis X, Co. schimperi X) and that had alignments longer than 500 bp. For the autosomal genes, we used distances only from the 1076 genes with alignments longer than 500 bp, s.e. for the total means were calculated using the sample decomposition procedure in the utilities package in R.

(f) . Chromosome preparation and cytogenetic analysis

Seeds of Co. schimperi were washed three times 1 h at room temperature (RT) in distilled sterile water and kept in the dark for 3–4 days. Young seedlings were moved to greenhouse conditions under a 16 L : 8 D cycle at 24°C. After 10 days, plants with young leaves (0.5–2 cm long) were used for chromosome preparation. Young leaves were collected in 1.5 ml tubes containing 2 mM 8-hydroxyquinoline (MERCK; 252 565) and incubated for 4 h in the dark (2 h at RT, 2 h at 4°C). After 4 h, leaves were fixed in Clarke's fixative (ethanol : acetic acid, 3 : 1, v/v) for 24 h and stored in 70% EtOH at −20°C. The chromosome preparation and image capture were as described by Bacovsky et al. [47] with minor modification. Young leaves were macerated in an enzyme mix for 25 min and before squashing placed for an additional 5 min in 45% acetic acid. Each karyotype was scored in three males and females in at least 15 metaphases.

3. Results and discussion

(a) . Genetic maps for the X and Y chromosomes of Coccinia grandis

The male and female maps inferred here for Co. grandis share 436 loci (mapped genes), 273 in the X-Y non-recombining part and 163 in the pseudo-autosomal region, revealing a single large pseudo-autosomal region (figure 2). The length of the X-Y non-recombining region on the female meiotic map is 77.6 cM, and the length of the corresponding pseudo-autosomal region is 40.4 cM (or 41.5 cM if one marker with a non-mapped Y-homologue is included). The length of the pseudo-autosomal region estimated in male meiosis is 42 cM, which is almost identical to the value obtained in female meiosis. This suggests that in male Co. grandis, there is no compensation for the missing crossing-overs in the non-recombining region. The SEX-DETector approach here applied to a family of 16 individuals identified 926 potentially sex-linked sequences of which 742 remained after manual inspection of all SNPs and evaluation of the number and pattern of reads (Material and methods). Of these 742 genes, 118 had only an X-linked copy and are probably hemizygous in males (with a probability of hemizygosity of over 80% as inferred by SEX-DETector). A female map of the 118 hemizygous X-linked genes is shown as the electronic supplementary material, figure S1.

Figure 2.

Genetic map of the sex chromosomes of Coccinia grandis. The long black lines connect the names of markers in the female map with the corresponding mapping positions on the male map. The triangles connect markers with their map positions if there are two or more markers mapping to the same position. The ruler shows genetic distances in cM. The short names of the genes are based on the numbering of markers having X and Y copies. Full gene names can be found in the electronic supplementary material, file S1. (Online version in colour.)

(b) . Phylogenetic analyses, recombination arrest and time estimation

Phylogenies require relevant outgroups, which reduced the number of potentially sex-linked genes that we could include in our analyses to 360 (of 926 or 742 potentially sex-linked genes; above). Three-hundred (83%) of the 360 genes yielded topologies in which the X- and Y-alleles of Co. grandis grouped together (figure 3a), 37 (10%) grouped the Co. grandis X allele with the Co. schimperi X allele (figure 3b), 6% grouped all Co. grandis X, Co. grandis Y and Co. schimperi X-alleles without resolving their relationships, and in 1% of trees, Co. grandis X, Co. grandis Y and Co. schimperi X-alleles failed to form a clade. The Bayesian concordance analyses showed significant support for the two main topologies (with a posterior probability greater than 0.95) except for the Co. grandis/Co. schimperi X-alleles (figure 3a versus b). A consensus tree of individual autosomal gene trees and the results of Bayesian concordance analysis (electronic supplementary material, figure S2) showed the same topology as obtained from the sex-linked genes (figure 3).

Figure 3.

Majority rule consensus trees of the individual gene trees obtained from X-linked and Y-linked alleles detected in RNA-seq data from a Co. grandis male and female and a Co. schimperi female (no male was available). Numbers at nodes represent posterior probabilities, followed by concordance factors for alpha 1, 10 and 1000 (Material and methods). The dashed lines highlight the sole difference between the two topologies. (a) The consensus topology obtained from the 300 genes that grouped Co. grandis X- and Y-alleles together. (b) The consensus topology obtained from the 37 genes that grouped Co. grandis X- with Co. schimperi X-alleles. (Online version in colour.)

To test for gradual or stepwise arrest of recombination along the non-recombining part of the X chromosome, we plotted the dS_XY values seen in the two main types of tree topologies—(Co. grandis X/Y) versus (Co. grandis X, Co. schimperi X)—against the position of the respective gene on the genetic map of the Co. grandis X (figure 4). The sex-linked genes that yielded the (Co. grandis X, Co. schimperi X) grouping had significantly higher mean dS_XY values (0.019) than the genes that yielded the (Co. grandis X/Y) grouping (0.011; p-value in Welch t-test = 8.097 × 10⁻⁵). However, there was no significant relationship between dS values and gene positions on the female map nor a clustering of the 118 hemizygous X-linked genes (electronic supplementary material, figure S1), suggesting that the entire region stopped recombining at the same time. Alternatively, chromosomal strata or a gradient of recombination suppression could have been present early during the evolution of these sex chromosomes and have been lost owing to translocations or inversions on the X chromosome. Such inversions have been invoked in Silene dioica [48]. Another explanation might be that our dS values may be unreliable owing to substitution hotspots in some genes (although it is hard to image how this might be the full explanation) and because of probably large confidence intervals (CIs), given our single-individual sampling for D. palmatus and Co. schimperi. Divergence estimates and gene tree topologies can also be influenced by the preferential mapping of reads carrying the variant most similar to the reference, as found in a study of the sex chromosomes of Humulus lupulus and Ca. sativa [49].

Figure 4.

The synonymous divergence (dS) between X-linked and Y-linked gene copies in Coccinia grandis plotted against the position of the corresponding gene on the non-recombining part of the Co. grandis X chromosome (shown in figure 2; starting at ca 40 cM). Black circles correspond to genes for which the Co. grandis X and Y group together. Open circles correspond to genes for which the Co. grandis X-copy groups with the Co. schimperi X-copy.

The mean synonymous divergence from D. palmatus seen in the (Co. grandis X, Co. schimperi X) group of genes was 0.094 (Material and methods), and given our calibration, this translates to 6.17 × 10⁻⁹ substitutions per synonymous site per year (CI 5.30 × 10⁻⁹–7.04 × 10⁻⁹ subs. syn.⁻¹ site yr⁻¹), which corresponds to a divergence time of 3.1 (CI 2.1–4.3) million years (Myr) for the Co. grandis X- and Y-alleles. The mean synonymous divergence of the two species Co. grandis and Co. schimperi of 0.025 (s.e. = 0.001) subs. syn.⁻¹ site yr⁻¹ seen in the phylogenies from autosomal single-copy orthogroups with alignments longer than 500 bp (n = 1076) translates into a divergence time of 3.64 (CI 3.37–3.9) Ma for these species. The divergence time of Co. grandis from Co. ogadensis, the next-closest relative, has earlier been estimated with a fossil-calibrated chronogram as about 3.1 Myr [10] and that of the Co. grandis sex chromosomes from each other, using different generation time assumptions, as 9–51 Myr [6], a range that illustrates the uncertainty of all these estimates.

Based on our estimates, the sex chromosomes of Co. grandis and Co. schimperi appear to have evolved separately for the last 3–4 Myr, and today, there is little opportunity for natural hybridization because of the species' geographical separation (figure 1). Unfortunately, their sister species, Co. ogadensis from Ethiopia and Somalia [9,10], is not in cultivation and its chromosomes are unknown. One can therefore not be sure if the large Y chromosomes with their accumulated repetitive sequences originated before Co. grandis and Co. schimperi diverged from each other. The Y chromosomes of Co. schimperi are metacentric (electronic supplementary material, figure S3) as are those of Co. grandis [5], which would fit a common origin and similar further evolution. However, since we lack transcriptome and DNA data for a male Co. schimperi, we cannot so far infer Y chromosome homology. For the X chromosomes in the two species, the 37 sex-linked genes that yield the topology (Co. grandis X, Co. schimperi X) point to chromosome homology, but future chromosome-level genome assemblies and synteny analysis are needed.

With the data now in hand, and living material of both species, future work could involve the re-sequencing of a Co. schimperi family (mother, father, offspring) and inference of Co. schimperi male and female maps. Another question is whether the Y elongation in Co. schimperi is owing to the same types of repetitive DNA as previously found in Co. grandis [5,12,13]. If degeneration of their Y chromosomes is occurring independently, genes that are hemizygous in Co. grandis might not be in Co. schimperi, and vice versa—an easy thing to check in future work. In addition, the Co. grandis/Co. schimperi species pair could be used to test Haldane's [50] rule, which predicts that in recently diverged species with heteromorphic sex chromosomes, the heterogametic sex should have had lower fitness than the homogametic sex and therefore be absent, rare, or sterile. The viability (up to the point of flowering) of the two experimental F₁ hybrids between Co. grandis and Co. schimperi obtained by Naudin [15], one of them male, the other female, is insufficient to test Haldane's rule, but since both species are now in cultivation, future work could include experimental crosses. In plants, Haldane's rule has only ever been tested in reciprocal crosses between dioecious and heterogametic Silene dioica, Silene diclinis and S. latifolia in which the rule holds but the genetic basis is insufficiently understood [51,52].

Data accessibility

For this study, we had available cleaned Illumina paired-end reads (average depth of approx. 10-fold) for 102 Co. grandis individuals (mother, father and 100 full siblings of which seven were males and 93 females) obtained from the seeds in three fruits from a Hawaiian mother pollinated with pollen from an Indian father in a greenhouse in Munich (these data are part of an ongoing Co. grandis genome assembly and will be uploaded in early 2022). The gene names and positions are shown in the electronic supplementary material, file S1 [53].

Authors' contributions

B.J.: formal analysis, investigation, supervision, validation, visualization, writing—review and editing; R.G.: data curation, formal analysis, investigation; V.B.: formal analysis, investigation, writing—review and editing; S.S.R.: conceptualization, project administration, resources, writing—original draft.

All authors gave final approval for publication and agreed to be held accountable for the work performed therein.

Competing interests

We declare we have no competing interests.

Funding

V.B. was supported by the Czech Science Foundation (Grantová agentura České republiky) grant no. 19-02476Y, and computational resources came from the large infrastructure projects e-INFRA LM2018140 and ELIXIR-CZ LM2018131, supported by the Czech government.