The mosaic oat genome gives insights into a uniquely healthy cereal crop

1Oat is a member of Poaceae, an economically important grass family that includes wheat, rice, barley, common millet, maize, sorghum and sugarcane. Avena species exist in nature as diploids, tetraploids and hexaploids and exhibit the greatest genetic diversity around the Mediterranean, Middle East, Canary Islands and Himalayas. Currently, oat is a global crop with production ranking seventh among cereals (http://www.fao.org/faostat/en/, accessed May 2021). Compared with that of other cereals, oat cultivation requires fewer treatments with insecticides, fungicides or fertilizers. Whole-grain oats are a healthy source of antioxidants, polyunsaturated fatty acids, proteins and dietary fibre such as β-glucan, which is important in post-meal glycaemic responses and for preventing cardiovascular disease–. Cereals such as wheat, barley and rye store high amounts of gluten proteins in their grain; by contrast, oat and rice store globular proteins in their grain.

2We produced a chromosome-scale reference sequence of oat cv. ‘Sang’ comprising 21 pseudochromosomes (Fig. 1, Extended Data Fig. 1a and Supplementary Table 1), with a BUSCO (v5.1.2; ref. ) score of 98.7% (Extended Data Fig. 2a), following the short-read strategy used for wheat, barley and rye. Inspection of Hi-C contact matrices (Supplementary Fig. 1) and the consensus genetic map (Supplementary Fig. 2a) and their comparison with the independent assembly (long-read) of hexaploid oat OT3098 (ref.; version 2; Supplementary Table 2) verified the integrity of the assembly (Extended Data Fig. 2b and Supplementary Fig. 2). We also assembled pseudochromosomes of the diploid Avena longiglumis and tetraploid Avena insularis, which are presumed A and CD subgenome progenitors of Avena sativa (Extended Data Figs. 1a and 2a and Supplementary Figs. 3 and 4). Phylogenomic analyses (Supplementary Fig. 5) used to assign A. sativa chromosomes to subgenomes showed that gene order is conserved in the proximal chromosomal regions. The 21 A. sativa chromosomes, named 1A–7D following the subgenome assignments of ref., were oriented to preserve the orientations of core regions across homoeologues and possibly between Avena and Triticeae. Alignments to barley (Extended Data Fig. 1b), Avena eriantha (Supplementary Fig. 6), A. longiglumis (Supplementary Fig. 7a) and A. insularis (Supplementary Fig. 7b) confirmed the validity of this revised nomenclature, which is accepted by the International Oat Nomenclature Committee.

3We predicted gene models in the oat genome using an automated annotation pipeline, assisted by RNA-sequencing (RNA-seq) and Iso-seq transcriptome data, protein homology and ab initio prediction. This yielded 80,608 high-confidence protein-coding loci (98.5% BUSCO; Extended Data Fig. 2c and Supplementary Table 3), 83.5% of which showed evidence of transcription in at least one condition. Another 71,727 low-confidence protein-coding loci primarily represent gene fragments, pseudogenes and gene models with weak support. The overall amount and composition of transposable elements is very similar between the Sang and OT3098 assemblies (Supplementary Tables 4 and 5 and Supplementary Fig. 8). Transposable elements accounted for 64% of the oat genome sequence. The size difference of about 1 Gb observed between the C and A or D subgenome probably reflects higher transposon activity in the diploid ancestor of the C subgenome, as evidenced by a 1.3-fold increase in the number of full-length long terminal repeat retrotransposons, an enrichment in specific transposable element-related Pfam domains and Csubgenome specific transposon families, higher repetitivity, more tandem repeats and higher numbers of transposable element and low-confidence genes (Extended Data Fig. 1c). Several tandem repeat subfamilies were unequally distributed across the subgenomes, highlighting potentially rearranged genomic regions (Extended Data Fig. 1d). However, limitations of the short-read assembly arising from lower contiguity (Supplementary Table 1) were apparent in the overall reduced representation of tandem repeats and ribosomal DNA loci (Supplementary Tables 4 and 6) as well as in regions of reduced gene density mainly in centromeric and pericentromeric regions and unplaced genes (Supplementary Fig. 9, Supplementary Table 7 and Supplementary Methods).

4The overall structure of the oat genome is similar to that of Triticeae genomes, although frequent genomic rearrangements in oat have resulted in a mosaic-like genome architecture. In many oat chromosomes, gene and recombination density is not a monotonic function of distance from the centromere (Extended Data Fig. 3), as is mostly observed in the Triticeae. Examination of whole-genome alignments, subgenome-specific k-mers and orthologous and homoeologous genes clustering as syntenic blocks in genomic neighbourhoods in four Avena species (Extended Data Figs. 1d and 4) revealed numerous large-scale genomic rearrangements affecting the order of these blocks within and between subgenomes (Fig. 2a). We detected seven large-scale genomic rearrangements in A. sativa and traced them back to eight translocation events between the A, C and D subgenomes (Fig. 2b, c, Extended Data Figs. 4a and 5a, Supplementary Fig. 10 and Supplementary Table 8), spanning 4.3% of the genome and approximately 7.9% of the high-confidence genes. Two of the translocation events were specific to A. sativa. Unlike those in wheat, the oat subgenomes exhibit unbalanced gene counts; specifically, the C subgenome appears to have 12% fewer genes than the A or D subgenome (Extended Data Fig. 2d and Supplementary Table 9). Analysis of orthologous gene groups (Supplementary Table 10 and Supplementary Figs. 11 and 12) showed that unbalanced gene families were associated with significant spatial clustering (Supplementary Fig. 13) in genomic rearrangements. Ancestral state reconstruction of the oat chromosomes revealed a loss of at least 226 Mb of gene-rich regions from the C subgenome to the A and D subgenomes (Supplementary Table 9). This implies that the translocations fully account for the lower gene count in the C subgenome and not gene loss or subfractionation after formation of the hexaploid.

5Previous molecular marker studies using oat mapping and breeding populations have provided independent evidence for frequent translocations among oat subgenomes,. Using the oat genome to reanalyse the data (Extended Data Fig. 6), we observed inter-chromosomal pseudo-linkage in a population that segregates for the 1C translocation on 1A. Such pseudo-linkage has been implicated in the propensity for cold hardiness to remain associated with non-carriers of this translocation. An accompanying study details similar associated opportunities and barriers in genomic breeding strategies. The mosaic nature of the oat genome may be associated with the apparent lack of an orthologue of TaZIP4-B2 (located within the Ph1 locus), which in bread wheat stabilizes the genome structure during meiosis and suppresses crossovers between homoeologues– (Extended Data Fig. 5b and Supplementary Figs. 14 and 15). In contrast to wheat, interploidy crosses and alien introgressions have been extremely challenging in Avena, suggesting that incompatible genome architecture is an additional barrier preventing genetic gains in oat.

6After polyploidization, sub- and neofunctionalization and gene loss modify the gene content in the new species,. Systematic differences in subgenome/homoeologue gene expression (homoeologue expression bias) may also be prevalent. In fact, quantitative variation for many agronomic traits may reflect genetic interactions between homoeologues such as functional redundancy (buffering) or dominant phenotypes attributed to one homoeologue. To investigate homoeologue expression bias in hexaploid oat, we defined 7,726 homoeologous gene triads with a 1:1:1 correspondence across the three oat subgenomes (Supplementary Table 11), referred to as ancestral triads. Average expression values across transcriptome samples from six tissues showed that C-subgenome genes were slightly less expressed (32.32%) than those in the D (33.53%) and A (33.76%) subgenomes (Kruskal–Wallis, P = 0.054). We considered six homoeologous expression categories and found that most ancestral triads (84.1%) showed balanced expression, 3.4% showed single-homoeologue dominance and 12.6% showed single-homoeologue suppression. The relative contributions of the different categories (Extended Data Fig. 7a) indicated no major overall bias for one of the subgenomes (Fig. 2d). A co-expression network approach revealed that genes from the C subgenome were found in divergent expression modules more frequently than their A- and D-subgenome homoeologues (χ2 test, P = 2.085 × 10−6; Extended Data Fig. 7b and Supplementary Table 12).

7In another 1,508 triad gene clusters containing at least one member positioned in a translocated region (relocated triads; Supplementary Table 13), the overall expression patterns were similar to those of the ancestral triads (Extended Data Fig. 7c). The C-suppressed category was slightly larger (5.1%) in the ancestral triads compared with the A-suppressed (3.5%) or D-suppressed (4.1%) triads, but the subgenome suppression patterns were reversed (4.5% A, 4.2% C and 5.2% D) in the relocated triads (χ2 test, P = 0.019; Extended Data Fig. 7c). Our results indicate that translocations and rearrangements in the oat genome may affect global and homoeologous gene expression patterns. Understanding how homoeologues interact to influence gene expression and identifying functional single-copy genes showing non-balanced expression will inform crop improvement in oats.

8Mixed-linkage β-glucans are soluble fibres present at high levels in oat endosperm cell walls (3.8–6.1 g per 100 g dry weight) that reduce blood cholesterol and post-meal glycaemic responses,. The cellulose synthase-like gene CslF6 is central for β-glucan biosynthesis in cereals,. We catalogued the cellulose synthase (GT2) and callose synthase (GT48) families of glycosyltransferases to identify the genetic foundation underlying oat β-glucan biosynthesis. The hexaploid oat genome encodes 134 members of the cellulose synthase gene superfamily (Fig. 3a), representing the cellulose synthase (CesA) subfamily (Supplementary Fig. 16) and seven cellulose synthase-like subfamilies, including CslA, CslC, CslD, CslE, CslF, CslH and CslJ. The GT48 family comprised 28 members (Supplementary Fig. 17). Genes within the CesA and CslF subfamilies were most highly expressed over multiple stages of seed development (Fig. 3a and Supplementary Fig. 18). Investigation of differentially expressed genes between stages indicated specific roles for particular subfamilies such as CslE and CslF (including the C-subgenome copy of CslF6), which were upregulated in late stages of seed development (Supplementary Fig. 19), as shown in barley. Compared with other grasses, the oat cellulose synthase superfamily showed no significant expansions apart from duplication events in the CesA, CslC, CslE and CslJ subfamilies (Supplementary Fig. 20 and Extended Data Fig. 8). These findings suggest that the high content and quality of β-glucan in oat are not driven by major differences in the copy number of cellulose synthase superfamily genes relative to other grasses but rather by allelic variation and transcription factors, as previously reported.

9Oat globulins constitute 75–80% of grain protein content, with prolamins (avenins) accounting for approximately 10–15%. Prolamin superfamily members trigger coeliac disease, food allergies and baker’s asthma. We identified genes encoding 25 avenins, 6 high-molecular-weight glutenins (HMW-GS) and 61 genes representing α-amylase/trypsin inhibitors (ATIs) and other prolamin superfamily members related to protein accumulation and immunogenicity (Supplementary Table 14). Hexaploid oat has avenin loci on chromosomes 1D, 3D and 7A; seed storage globulin loci (135 genes) on chromosomes 1A, 1D, 3D, 7A, 4A and 4D; and no storage protein loci mapping to the C subgenome (Fig. 3b and Supplementary Table 14).

10Unlike that of wheat, the oat genome harboured no α- or ω-gliadin genes, and the identified avenins co-clustered with γ-gliadins, low-molecular-weight glutenins and B-hordeins (Fig. 3b). We detected four complete, highly conserved oat HMW-GS gene models as two distinct loci on 1A and one locus pair on 1D, with no HMW-GS genes mapping to 1C. We identified a prolamin type, the 19-kDa globulin-like proteins, with an unknown function that is distinct from the avenins yet shares sequence similarity with HMW-GS and 19-kDa globulins (Fig. 3b). The predicted oat HMW-GS and avenins were highly conserved in their Pfam domains (Fig. 3b) and cysteine patterns (Extended Data Fig. 9). Glutamine- and proline-rich repetitive peptides were fewer in these oat proteins, making them shorter than those in wheat or barley (Extended Data Fig. 9).

11We detected transcripts for most of the avenin genes, which showed gene expression patterns that aligned with their wheat orthologues, with increased transcript levels from the middle phase of seed development (Fig. 3c), and protein levels by using liquid chromatography with tandem mass spectrometry (Fig. 3b). We identified inactive genes and pseudogenes among avenin-encoding genes (Fig. 3b, c and Supplementary Table 14) in a similar proportion as in wheat γ-gliadins,. This indicates a lower level of gene expansion and pseudogenization compared with the highly immunogenic wheat ɑ-gliadin genes. Moreover, the expression of 11S globulin genes initiated early in seed development and was higher than that of the avenin genes (Fig. 3c). Discovery proteomics detected thirty-six distinct 11S globulins, five globulin-1 proteins and two 7S globulins, with an average of 83% protein sequence coverage at a 1% false discovery rate.

12The oat avenins and globulins showed opposite trends compared with their wheat orthologues in gene copy number, protein length and enrichment in glutamine and asparagine residues that serve as a nitrogen storage sink (Extended Data Fig. 10a). Together with pronounced differences in transcription factor-binding sites specific to the nitrate response (Extended Data Fig. 10b and Supplementary Table 15), this may contribute to the primary role of oat globulins in nitrogen storage. These results confirm that the genomic organization, sequence characteristics and expression patterns of oat storage proteins share more similarities with rice and dicotyledonous plants than with wheat and other gluten-rich cereals,.

13We mapped previously reported coeliac disease-associated T cell epitopes to the predicted oat avenin proteins and compared them with the T cell epitope patterns of wheat and barley prolamins. The results showed that only a subset of encoded avenin proteins contain coeliac disease-associated immune-reactive regions compared with the high prevalence found in wheat or barley (Fig. 3b). Taken together, the low copy number of genes encoding coeliac disease epitopes, low frequency of detected T cell epitopes in the protein sequence, low occurrence of other highly immunogenic proteins, proportion of avenins within total oat protein and relative immunogenicity of avenin epitopes all support the inclusion of oats in gluten-free diets.

14To demonstrate how an annotated reference genome enables greater use of resources such as TILLING populations,, we mapped the causal mutation in the epicuticular wax mutant glossy.1 (Fig. 4a, b). Epicuticular waxes have a role in biotic and abiotic stress resistance, and provide an important target for oat breeding. We identified homozygous polymorphisms unique to the mutant, which mapped to chromosome 1C (Fig. 4c and Supplementary Fig. 21), and identified a single gene annotated as an α/β-hydrolase (AVESA.00010b.r2.UnG1403470) as a likely candidate that is orthologous to barley Cer-q (HORVU.MOREX.r3.2HG0097460) (Supplementary Fig. 22). An independent mutant line (glossy.2) exhibited the same glossy phenotype (Fig. 4d, Supplementary Fig. 23 and Supplementary Table 16). Barley Cer-q mutants are deficient in the same β-diketone (hentriacontane-14,16-dione) and wax tubules that are absent in the glossy mutants (Fig. 4e–g and Supplementary Figs. 24–26). The scaffold containing the candidate gene was localized to the region of chromosome 1C (Fig. 4c and Supplementary Table 17). The presumed glossy.1 mutation introduced a P243S substitution in the encoded protein adjacent to a deleterious F219L substitution known to inactivate barley CER-Q (Fig. 4d and Supplementary Fig. 27). We identified gene clusters on oat chromosomes 1C, 2C and 3A and in wild Avena species (Supplementary Figs. 22 and 28–33) that are homologous to the barley Cer-cqu cluster,. We also noted genes encoding proteins with similarity to Arabidopsis wax ester synthase/diacylglycerol acyltransferase 1 (WSD1), a Myb-domain transcription factor and a short-chain dehydrogenase/reductase (SDR) protein near the Cer-cqu homologues in the Avena genomes. All genes from the 1C cluster except SDR were expressed at levels 3–6 times higher than those of the 3A cluster, with very low expression from 2C cluster genes and with no differential expression between the glossy and glaucous glume tissue (two-sided Wald test, null hypothesis logarithmic fold change = 0, adjusted P < 0.01; Fig. 4h and Supplementary Fig. 34). Together, these results suggest that AVESA.00010b.r2.UnG1403470 is the oat Cer-q gene. The reference genome thus facilitated a major advance in understanding β-diketone biosynthesis in oat and can help breeders manipulate tissue-specific epicuticular wax composition in future oat cultivars adapted for hotter climates.

15In summary, this fully annotated hexaploid oat reference genome lays the foundation for advances in oat breeding and basic oat biology and for the ongoing pan-genome project. With the chromosome rearrangements in a typical spring oat cultivar now delineated, breeders and researchers will have access to a resource equal in calibre to Triticeae genomes, which may help them to overcome the breeding barriers and segregation anomalies described in numerous mapping studies. Using the reference genome to map genes associated with agronomic and human nutrition-related traits is a viable approach for precisely adapting oat varieties. Known quantitative trait loci can be anchored to the Sang reference, and the transcriptome atlas co-expression networks can be leveraged to identify candidate genes in the vicinity of specific quantitative trait loci. Modern breeding strategies such as genome editing and gene pyramiding can now more easily be applied in oat to develop varieties that meet the increasing global demand for oat-derived products. Our proteogenomic investigation of oat storage proteins confirms qualitative and quantitative differences in the expression of proteins compared with the more abundant and immunogenic sequences in wheat, barley and rye, which supports the safety of oats in gluten-free diets. The detailed genome annotation and case studies presented here provide examples of the myriad possibilities for the discovery and exploitation of functional genetic mechanisms in oat.

16Further information on research design is available in the Nature Research Reporting Summary linked to this paper.

17Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at 10.1038/s41586-022-04732-y.