Rhodes grass (Chloris gayana) is a tropical C4 species native to most of Africa that is grown across the continent as hay and permanent pasture, as well as to control erosion due to its deep rooting (Skerman and Riveros, 1990). It is a member of the genus Chloris that includes 56 species and sub-species, most of which are weedy types (Anderson, 1974). Rhodes grass is most closely related to Chloris virgata, which is more commonly known as feather fingergrass or occasionally Rhodes grass or feathertop Rhodes grass (Edwards, 2012; Rojas-Sandoval, 2016). Exported in the early 1900s, C. gayana is now cultivated in predominantly tropical and sub-tropical areas across the globe (Loch et al., 2004). Demonstrated by its extensive distribution in its native Africa and its establishment in non-indigenous regions, Rhodes grass has wide environmental adaptation potential, proliferating in a wide range of soil types and with cultivars capable of growing at temperatures between 8-35°C (Loch et al., 2004). Rhodes grass is a halophytic grass, capable of tolerating higher levels of soil salinity than many other C4 grasses due to its ability to partition salt ions such that photosynthesis is unaffected (Bogdan, 1969; Oi et al., 2022). Its tolerance of salt has seen the species used to restore highly degraded soils and to provide an additional source of fodder on otherwise underutilised land (Abate et al., 2021; Singh et al., 2022). Global warming-mediated temperature changes in traditionally temperate regions has increased interest in Rhodes and other C4 tropical grasses as they are considered potentially important to the future of a range of livestock industries; despite the majority of current research focusing on C3 forages (Leegood, 2013).

As a member of the PACMAD clade that includes all C4 grasses, Rhodes grass falls under the Cynodonteae tribe within the Chloridoideae subfamily (Edwards, 2012). Only four of at least 850 recognised Cynodonteae species have whole or draft genomic sequences available (Oropetium thomaeum, 245Mb genome; Eleusine coracana, 1453Mb genome; Eleusine indica, 584Mb genome; and Cynodon transvaalensis, 454Mb genome) (VanBuren et al., 2015; Hittalmani et al., 2017; Soreng et al., 2017; Zhang et al., 2019; Cui et al., 2021). While no molecular dating of the divergence of C. gayana from other grass species is available, it belongs to the subtribe Eleusininae which is known to have originated approximately 22mya and C. gayana’s closest known relative, C. virgata, is estimated to have diverged from a different Chloris species approximately 3.6mya (Edwards, 2012; Wang et al., 2022). Of the four Cynodonteae species with available draft genome sequences, those most closely related to the Chloris genus, E. coracana and E. indica, are approximately 15 million years diverged (Wang et al., 2022). The evolutionary distance from Rhodes grass to these other species makes comparative genomics challenging. Furthermore E. coracana and E. indica are both allotetraploids, which limits their application for a diploid genome assembly, and all four Cynodonteae species with genome sequence are assembled only to a scaffold level. Setaria italica is the most closely related species to C. gayana that is both a diploid and has a chromosome-level genome assembly available (Bennetzen et al., 2012).

Rhodes grass has both diploid (2n=20) and tetraploid varieties (2n=40), with some reports of triploid (2n = 30) varieties existing (Anderson, 1974; Loch et al., 2004). The tetraploid varieties are currently proposed to be autotetraploids or segmental allotetraploids based on findings from a single study (Nakagawa et al., 1987). However, the exact genetic composition of the tetraploid varieties is unknown and to date no suggestions of speciation events have been proposed (Loch et al., 2004).

Within the current known range of grass genome sizes (the smallest being Orpetium thomaeum at 245Mb and one of the largest being Triticum aestivum L. at ~17Gb), the estimated genome size of Rhodes grass is relatively compact (1C = ~293Mb) (Bennett and Smith, 1976; VanBuren et al., 2015; Shi and Ling, 2018). As an outbreeding species, it is highly heterozygous with substantial intra- and inter-cultivar variability (Loch et al., 2004; Ubi et al., 2004). This diversity gives rise to a range of beneficial traits that can be exploited by breeders. However, modern breeding tools such as genomic selection and genome editing have not yet been applied as limited research has been conducted on the species. Hence, Rhodes grass can be considered an orphan crop as it is one of the most important perennial C4 forage grasses, yet its available genomic and genetic resources are disproportionately few (Gondo et al., 2007). Currently available resources include a genetic linkage map (Ubi et al., 2004), unassembled short Illumina sequences available on GenBank from a DNA barcoding study (Gill et al., 2019) and short sequences from unpublished research. Previous research has utilised conventional anonymous molecular marker techniques such as RAPD, AFLP, ISSR and SRAP markers to examine diploid and tetraploid Rhodes grass diversity (Pérez et al., 1999; Ubi et al., 2000; Ubi et al., 2001; Ubi et al., 2003; Ribotta et al., 2019). The main conclusions of all these studies were similar; in Rhodes grass, there is a large degree of both intra- and inter-cultivar variation. The most recent study on Rhodes grass population structure and genetic diversity assessed 104 accessions using SNP and SilicoDArT markers generated with DArTseq (Negawo et al., 2021). This study examined Rhodes grass accessions in the context of their geographical origin but found no clear relationship between country of origin and genetic similarity. Consistent with the previous molecular marker studies, Rhodes grass was found to have considerable genetic diversity between accessions (Ubi et al., 2000; Ubi et al., 2001; Ubi et al., 2003). In their study, Negawo et al. (2021) attempted to align their molecular markers to reference genomes of closely related species but only a small proportion (1-5.5% of SNP markers) could be mapped to the available Chloridoideae genomes. They noted the number of mappable markers was significantly higher when a reference genome for that species was used, highlighting the importance of whole genome sequence resources. As their study focussed on describing a geographically diverse collection of Rhodes grass, it did not sample individual accessions deeply or provide information on ploidy status.

Due to its small genome size and genome plasticity (Nakagawa et al., 1987), Chloris gayana has potential as a model C4 grass species with the development of relevant underpinning genomic resources. Here, we present the first de novo draft reference genome assembly in the Chloris genus, as well as a large SNP resource developed via whole genome resequencing of a range of current commercial diploid and tetraploid cultivars of Rhodes grass. The SNP resource enables identification of within and between cultivar relationships, while the WGR data itself can be aligned to the diploid reference genome to provide insights into tetraploid Rhodes grass origins. These genomic resources will facilitate future breeding efforts and genome editing studies, as well as providing further insight in the grass phylogeny.

An annotated draft genome sequence assembly of Rhodes grass was constructed using long-read sequencing technology and the tissue culture responsive line T108 isolated from the diploid cultivar Tolgar. At 603 Mb, the total length of the draft assembly is about the estimated diploid genome size for Rhodes grass (2C= ~586Mb) (Bennett and Smith, 1976; VanBuren et al., 2015; Shi and Ling, 2018). Due to the Flye assemblers (v-2.8.2, https://github.com/fenderglass/Flye) attempts to collapse haplotigs, the assembled genome was expected to be smaller than the actual generated assembly (603Mbp), with this larger than expected assembly potentially providing an indication of the genetic diversity of the genome. While the error rate of the ONT pore 9.4 is higher than other technologies, it is unlikely it would be sufficiently significant to cause the large assembly size of the genome. An additional hypothesis regarding the assembled data size could be seed contamination or mislabelling such that the plant sequenced to generate whole genome sequence (Tolgar 108) is tetraploid and not diploid. However, as this plant genotype was resequenced and analysed in the broader cohort of plants and subsequently shown to align more with diploid individuals ( Figure 3 ), this also may be unlikely. Comprising 5,233 contigs with a N50 of 310kb and 46,087 predicted genes (including 91.97% of expected BUSCO genes, of which 71.59% are single copy), the quality and completeness of the draft genome assembly is comparable to that of other draft orphan crop genomes; e.g. Digitaria exilis- 3329 contigs, BUSCO: 98.1% (Wang et al., 2021). As generated by EvidenceModeler from the consensus of the Augustus gene predictions and the S. italica BLASTn alignment, the predicted gene number for Rhodes grass (46,078) was in a similar range of gene number compared to closely related species E. coracana (57,180 predicted genes (Hatakeyama et al., 2018)), S. italica (24,000-29,000 predicted genes (Bennetzen et al., 2012)) and L. perenne (38,868 predicted genes (Frei et al., 2021)). The accuracy of the Rhodes grass predicted gene set could be improved by creating transcriptome resources for training prediction models, however this was outside the scope of this study.

Among commercial cultivations of Rhodes grass, cultivars descended from Katambora are common. Hierarchical clustering of 104 Rhodes grass accessions by Negawo et al. (2021) showed the presence of two main subpopulations, with Katambora clustering within the larger group. As Tolgar is a descendent of Katambora, the draft genome sequence generated is expected to be useful as a representative for diploid Rhodes grass and broadly applicable for a majority of Rhodes grass cultivars.

The reference genome for Setaria italica, that is assembled to chromosome level, was used to order the assembled draft Rhodes grass genome contigs, based on synteny and visualised using Strudel (Bayer et al., 2011; Bennetzen et al., 2012). Setaria italica is the most closely related species to Rhodes grass with an available whole genome sequence and has a similar number of chromosomes to diploid Rhodes grass (S. italica 2n = 18; diploid C. gayana 2n = 20). The S. italica subfamily Panicoideae is estimated to have diverged from the other PACMAD subfamilies about 20mya (Cotton et al., 2015). The comparative alignment and syntenic comparison provides an indication of genome assembly completeness, and a preliminary indication of how the Rhodes grass contigs may be ordered at a chromosome level. The contig ordering and alignment from this approach has been provided for reference only and has not been used for any further analysis. The availability of a Rhodes grass high-density genetic linkage or optical map anchored to chromosomes would allow for the contigs to be more accurately assembled into pseudochromosomes (You et al., 2018). However, to date molecular genetic studies for Rhodes grass have generally focussed only on assessing cultivar diversity (Pérez et al., 1999; Ubi et al., 2000; Ubi et al., 2001; Ubi et al., 2003; Ribotta et al., 2019). To our knowledge, only one genetic linkage map has been constructed for Rhodes grass based on 25 restriction fragment length polymorphisms (RFLPs) and augmented with amplified fragment length polymorphisms (AFLP) markers (Ubi et al., 2004). The density of this linkage map is insufficient for anchoring and ordering the draft Rhodes grass genome assembly.

It is known from available breeding information ( Supplementary Table 1 ) that the diploid cultivars Finecut, Tolgar and Endura are all descendants of Katambora, a landrace derived from seed collected along the Northern Rhodesian bank of the Zambesi River in Zimbabwe (CSIRO, 1972b). Hence, the overlapped grouping of the Rhodes grass diploid cultivars was expected due to their shared ancestry and may also reflect the limited selection and outcrossing nature of its breeding system ( Figure 3 ). In general, the diploid samples clustered according to the cultivar from which they were derived, although there was some overlap and mixing most likely due to their shared ancestry from cv. Katambora. The tetraploid cultivars Toro and Mariner clustered into their own group, separate from the diploid samples, with some mixing between cultivars. Despite this, these cultivars have not been reported to share close genetic or geographic origin. Mariner and Toro are descended from cultivars Samford and Callide, respectively (Plant Breeders Rights, 2022; Barenbrug, 2022a; Barenbrug, 2022b). Callide originates from what is now Tanzania, while Samford was derived from ecotypes grown in Sierra Leone that arose from a prior introduction from Kenya (CSIRO, 1972a; CSIRO, 1972c). The tetraploids cluster away from the diploids in our phylogeny, as they are likely derived from a speciation event and therefore have more closely linked genetic ancestry. These three parental cultivars, Katambora, Callide, and Samford, were shown by Negawo et al. (2021) to cluster together when evaluated against other Rhodes grass cultivars, indicating a degree of similarity despite differences in geographic origin and ploidy. The pedigree of the cultivars likely contributes to the overlapping and integrated clades within the Rhodes grass phylogenetic tree. It is also possible that some overlap of individuals among cultivars results from inadvertent seed mixing or interbreeding in the field. The SNP lists used to construct the phylogeny were filtered with strict parameters, which likely resulted in the exclusion of some genuine SNPs and rare alleles. However, as these data are the first of their kind for Rhodes grass we have chosen to provide the most conservative highest quality SNP set as a foundation for future research, also considering the modest average sequence coverage and samples per cultivar that the study generated.

Previous cytologic work has suggested that tetraploid Rhodes grass is likely to be an autotetraploid or segmental allotetraploid due to the observed presence of multiple multivalent formations during meiosis (Nakagawa et al., 1987). Autotetraploids have two duplicated copies of the genome originating from a single ancestor, while segmental allopolyploids originate from doubled diploid hybrids of closely related species with chromosome pairing generally favoured between common ancestors (Sybenga, 1996). Given there was little difference (0.39%) in the proportion of diploid and tetraploid sample reads that were mapped to the diploid reference genome, our results suggest the second sub-genome of the tetraploid cultivars is sufficiently similar to the diploid reference to correctly align. The consistency of alignment suggests that the tetraploids originated from a self-genome replication event consistent of an autotetraploid. If the tetraploids were segmental allotetraploids, as has been previously suggested, a considerable difference in the alignment statistics might be expected (Nakagawa et al., 1987; Loch et al., 2004). A much higher frequency of heterozygous SNP genotype calls would also be expected for the tetraploid samples when their sequence reads were mapped to the diploid genome assembly due to sequence divergence between some chromosomes of one genome and those of the other genome in the tetraploid samples. Consistent with previous cytological studies which showed that tetraploid Rhodes grass forms multivalents during meiosis, our results allow us to rule out the possibility that the tetraploids are allotetraploids (two distinct genomes brought together through hybridisation between two species followed by doubling of the chromosomes of the resulting hybrid). If allotetraploids, sequence reads from the second unrepresented sub-genome would be expected to misalign with the diploid Rhodes grass genome assembly, resulting in a high frequency of heterozygous SNP genotype calls which was not observed. Further, a proportion of the SNP from the tetraploid samples would be expected to possess a nucleotide variant not observed among SNP discovered in the diploid samples. Hence, our results suggest that tetraploid Rhodes grass is an autopolyploid.

This understanding of Rhodes grass as an autopolyploid can facilitate breeding and genome editing for improved forage qualities. As the chromosome sets in tetraploid Rhodes are homologous, polysomic inheritance can occur with recombination of alleles possible between sub genomes, allowing for inheritance of desirable traits (Klie et al., 2014). Genome editing polyploids can be complicated due to circumstances including high heterozygosity, presence of homoeologous alleles and repetitive DNA. Nevertheless, genome editing of several polyploid species has been achieved with various site-directed nuclease (SDN) technologies, such as Talens and CRISPR-Cas9 (Schaart et al., 2021). Previous work on editing autotetraploid crops has been facilitated by the presence of the whole genome sequence of the autotetraploid, the comparison of autotetraploid short read sequence to a diploid genome, or by characterising specific target regions then looking for genetic variation between targets (Johansen et al., 2019; Chen et al., 2020; Sevestre et al., 2020). By generating short read sequence and SNP data for Rhodes grass tetraploids, as well as a diploid whole genome sequence, our work could facilitate future breeding and genome editing of tetraploid Rhodes grass.

The draft genome sequence generated here enables transcriptome studies on the variation of important traits in Rhodes grass, like salt tolerance and resilience to varying growth temperatures (Bogdan, 1969; Loch et al., 2004; Oi et al., 2022). Additionally, the SNP annotation data gives an insight into variations in the sequence that may impact gene function between individuals and can better explain varying phenotypes between individuals. These data will be of significant value for the research community that is gaining increased interest in this species as it will enable a wide range of molecular breeding tools as well as an array of biotechnology approaches to now be possible.

The draft diploid Rhodes grass genome assembly presented here represents the second draft genome assembly among Cynodonteae forages and first within the Chloris genus. It provides an important resource to underpin the genetic improvement of important breeding traits such as forage digestibility. Our results provide supporting evidence that is consistent with tetraploid Rhodes grass having evolved through an autopolyploid process. The genomic resources generated can be used to facilitate genomic breeding approaches and genome editing, and provide the opportunity to use Rhodes grass as a model C4 forage species for functional genomic studies. Rhodes grass is a resilient warm season forage crop, which would benefit from genetic improvement of its forage quality since it has been identified as a potential future forage species for dairy and livestock industries in regions where global warming is expected to reduce temperate pasture quality and productivity.

The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found below: Bioproject accession number: PRJNA974075.

The authors contributed to the paper as follows, study conception; NC, MH, KM-R, and DI: data collection; KM-R: data analysis and interpretation; KM-R: paper drafting; KM-R: paper revision and editing; KM-R, NC, DI, and MH. All authors contributed to the article and approved the submitted version.