Population genomics uncovers loci for trait improvement in the indigenous African cereal tef (Eragrostis tef)

Abstract

Tef (Eragrostis tef) is an indigenous African cereal that is gaining global attention as a gluten-free "superfood" with high protein, mineral, and fibre contents. However, tef yields are limited by lodging and by losses during harvest owing to its small grain size (150× lighter than wheat). Breeders must also consider a strong cultural preference for white-grained over brown-grained varieties. Tef is relatively understudied with limited "omics" resources. Here, we resequence 220 tef accessions from an Ethiopian diversity collection and also perform multi-locational phenotyping for 25 agronomic and grain traits. Grain metabolome profiling reveals differential accumulation of fatty acids and flavonoids between white and brown grains. k-mer and SNP-based genome-wide association uncover important marker-trait associations, including a significant 70 kb peak for panicle morphology containing the tef orthologue of rice qSH1-a transcription factor regulating inflorescence morphology in cereals. We also observe a previously unknown relationship between grain size, colour, and fatty acids. These traits are highly associated with retrotransposon insertions in homoeologues of TRANSPARENT TESTA 2, a known regulator of grain colour. Our study provides valuable resources for tef research and breeding, facilitating the development of improved cultivars with desirable agronomic and nutritional properties.

Fig. 1. Diversity of panicle morphology and grain colour in tef.

a Comparison of a bread wheat spike (cv. ‘Paragon’, far left) with tef accessions exemplifying four categories of panicle morphology (from left to right; very lax, lax, semi-compact, and compact). b Comparison of bread wheat grains (cv. Paragon, bottom) with grains from brown and white-grained tef varieties.

Fig. 2. Resequencing, phenotyping, and GWAS of the EIAR core tef collection.

A representative panel of Ethiopian tef accessions was resequenced and phenotyped for agronomic, grain size, and metabolomic traits. Statistical modelling was used to correct for location and within-site spatial effects. The software and numbers of accessions used for each step are indicated above each box. Vector images of the tef plant and sequencer were created by Wanda Pelin Canila/Shutterstock.com and Jaitham/Shutterstock.com, respectively.

Fig. 3. Phylogenetic analyses identify redundancy in the EIAR core collection.

a Phylogram of 220 tef accessions, arbitrarily rooted against the accession “Ada-T58”. White and brown-grained varieties are well-distributed across the phylogeny. A total of 32 redundancy groups, ranging in size from 2 to 19 accessions, were identified on the basis of small phylogenetic distances between pairs of accessions. b Phylogenetic distance plotted against the percentage of k-mer states shared for all 24,090 pairwise comparisons between accessions. There is a strong correlation between these two relatedness metrics. Accession pairs from within the previously defined redundancy groups (purple) cluster together at uniquely high shared k-mer state rates, depicted in detail in (c). The points with particularly low percentages of shared k-mer states (<78%, dashed line) represent the full set of comparisons of “DZ-01-1167” with other accessions.

Fig. 4. Best linear unbiased predictors (BLUPs) reveal correlations between key agronomic traits.

a–c Analysis of BLUPs for n = 141 accessions and redundancy groups. a Correlation tests were conducted between the BLUPs for 20 traits of interest. Significant correlations (p <  0.05) are indicated by circles whose size and colour represent the magnitude and direction of correlation. b Boxplot of plant height BLUPs for white-grained (n = 84) and brown-grained (n = 57) varieties, with individual data points overlaid. White-grained accessions tended to produce taller plants. The centre line represents the median; the lower and upper hinges correspond to the 25th and 75th percentiles, and the whisker extends to 1.5 * Interquantile range (IQR). c Scatterplot of grain area BLUPs against thousand grain weight (TGW) BLUPs. A distinctly bimodal distribution is strongly explained by grain colour, with white-grained varieties tending to produce smaller grains. Despite this, their grains are of approximately the same mass as brown-grained varieties, suggesting higher grain densities.

Fig. 5. Brown and white-grained tef accessions display differential metabolite accumulation.

a Partial least squares discriminant analysis (PLS-DA) of metabolites in grain samples of brown and white-grained accessions. Ellipses represent 95% confidence intervals around each group. b Differentially accumulated metabolites show enrichment for several metabolic pathways, notably fatty acid and flavone metabolism. Point size scales with pathway impact and colour intensity scales with significance of pathway enrichment. c Volcano plot for the 183 identifiable differentially accumulated metabolites. Fold-change (FC) was calculated as mean value in brown-grained varieties divided by that in white-grained varieties. Plotted FC thresholds are log₂(0.83) and log₂(1.2) and plotted FDR threshold is log₁₀(0.05).

Fig. 6. k-mer-based GWAS identifies multiple marker-trait associations, including regions associated with panicle morphology and grain KOROG.

a Plot summarising all trait-associated regions identified by k-mer-based GWAS. Regions positively associated with traits are plotted above their respective chromosomes, while negatively associated regions are plotted below. For grain colour, positive and negative associations indicate brown and white, respectively. b A region significantly associated with panicle morphology was detected on chromosome 3B. The arrangement of the 13 genes within this region is displayed below the plot. The candidate gene qSH1 is highlighted. c A region significantly associated with Kaempferol 3-O-rhamnoside-7-O-glucoside (KOROG) was detected on chromosome 7A. The arrangement of the 26 genes within this region is displayed below the plot. The candidate gene CYP93G1 is highlighted. In (b and c), k-mers are grouped according to their association level and genomic coordinates (10 kb bins) and coloured according to the direction of association; red for panicle laxness or low KOROG, blue for panicle compactness or high KOROG. Point size is proportional to the number of k-mers rounded upwards to the nearest 10.

Fig. 7. Co-association of grain colour, width, and EPOD concentration with multiple regions.

Plots of k-mers associated with a grain colour, b grain width, and c grain EPOD concentration. k-mers are grouped according to their association level and genomic coordinates (10 kb bins) and coloured according to the direction of association. In (a), brown denotes association with brown grain colour and white with white grain colour. In (b and c), red denotes association with lower trait values, and blue with higher trait values. Point size is proportional to the number of k-mers rounded upwards to the nearest 10. Nine regions are labelled with black and green numbers, denoting whether the region is significant or not significant for the plotted trait, respectively. Diagrams of LTR Copia insertions into TT2 homoeologs on d chromosome 4A, e chromosome 4B. Top: structure of TT2 in Dabbi (brown-grained). Centre: structure of TT2 in Tsedey (white-grained). Bottom: detail of LTR Copia insertions. Narrower exons indicate presumed protein truncations. Black DNA bases denote 5 bp target-site duplications. Green DNA bases denote the start of the retrotransposon insertions. Single-letter amino acid codes show the introduction of premature stop codons (*). The first 22 bp of the TT2 open reading frame on 4B is not assembled in the Tsedey genome (greyed out, black arrowhead). Gene annotations derive from ref. . and do not include 5’ and 3’ untranslated regions.