Bulked seeds of the tef diversity panel grown under field conditions were obtained from the USDA-ARS National Plant Germplasm System, Plant Germplasm Introduction and Testing Research Unit (Pullman, Washington, USA). The seeds were placed on a 76mm x 25mm concavity slide and imaged using a KEYENCE BZ-X710 All-in-one Fluorescence Microscope (Keyence Corporation of America, Ithica, Il) using the setting in Supplementary Table S1 . ImageJ software v1.54g (Schneider et al., 2012) was used to determine seed size attributes including seed length, seed width, and seed area of 189 tef and 11 E. pilosa accessions. The length measurements were taken from the longest portion of the seed and the width measurements were taken from the widest portion of the seed, while the seed area corresponds to the region within the perimeter of the image ( Figure 1 ). For each of the 189 tef and 11 E. pilosa accessions, ten seeds (replicates) were measured for length, width, and area. Each seed image included a 1000 µm scale bar. Ten replicate measurements of the scale bar were taken and averaged to generate a calibration factor from pixels to µm (seed length and width; Equations 1 and 2) or pixels² to µm² (area; Equation 3); the calibration factor is assigned pixels_c in Equations 1–3. According to the user guide (https://imagej.net/imaging/spatial-calibration), ImageJ takes area measurements as a count of pixels² and straight-line measurements in pixels. We maintained the same microscope settings to acquire all the images. We found that after converting the seed measurement to a standardized unit (µm or µm²), the initial parameters were not as important, as long as the calibration was properly set.

Additionally, we also determined 1000-grain weight of representative tef accessions selected based on seed area analysis, and those accessions that are considered reference cultivars in tef research. For each accession, one thousand seeds were counted manually, and the mass was measured. This was conducted in quadruplicate and then averaged. Some common tef accessions used in this study include Dabi, a brown cultivar with medium-sized seeds and a reference cultivar; Magna, an ivory (white) variety with high consumer preference and relatively large seeds; and Dessie, another common, brown-colored cultivar widely grown in the U.S. Dabi, Magna, and Dessie have retained their common name, whereas other accessions are labeled with their Plant Introduction (PI) number (Supplementary File 1) .

Seeds of selected E. tef accessions varying in seed size (small, medium, and large) along with the reference cultivar (Dabi) and the cultivar commonly grown in U.S. (Dessie), were grown under greenhouse conditions with supplemental light at constant temperature (28°C). Twenty seeds of each accession was planted in 2 L pots containing Sun Gro professional Mix (Sun Gro Horticulture, Agawam, MA). Plants were supplied with Osmocote^® Smart-release^® controlled-release fertilizer (Scotts Company, LLC, Marysville, OH) following the manufacturer’s recommended application rates. Seedlings were thinned down to five. Plants were supplemented with 1 L of 4.74 g/L Miracle-grow fertilizer (Scotts Miracle-Gro Co., Marysville, OH) solution (24–8–16) per flat containing two pots, and 1 L of 250 mg/L M.O.S.T. soluble trace elements (JR Peters, Allentown, PA) per flat every four weeks. Selected E. pilosa accessions were grown under the same conditions. All accessions were replicated four times. Seeds of plants grown in each pot were bulked at harvest, dried and processed for mineral analysis.

A total of 500 mg of ground seed was digested using concentrated HNO₃ for 30 min in a microwave at 200°C, and mineral content was analyzed using Inductively Coupled Plasma Optical Emission Spectroscopy (ICP-OES).

The procedures for synchrotron x-ray fluorescence (SXRF) imaging were described previously with slight modifications (Chia et al., 2023). Briefly, the seeds were placed between a layer of metal-free Kapton film and Kapton tape before being mounted onto 35-mm slide frames. The spatial distribution of elements was imaged via SXRF microscopy at the Functional Materials Beamline (FMB) of the Cornell High Energy Synchrotron Source (CHESS). FMB employs an undulator source and an energy of 9.7 keV was selected using a side-bounce monochromator (Smieska et al., 2023; Stoupin et al., 2021). The beam was focused to 3 µm tall x 15 µm wide containing approximately 3e10 ph/s using a set of compound refractive lenses (RXoptics, Monschau, Germany). SXRF signal was collected in flyscan mode along the vertical axis with a pixel size of 20 µm and a dwell time of 80 ms, using a Vortex ME4 detector (Hitachi, Japan) and an Xspress3 signal processor (Quantum Detectors, UK). SXRF peak areas were fit in Praxes software (https://github.com/praxes/praxes) which is based on PyMCA (http://dx.doi.org/10.1016/j.sab.2006.12.002). Thin foil calibration standards (Micromatter, Canada) were used to obtain a flux to concentration conversion and provide area densities in units of µg/cm².

A list of seed development genes in rice was compiled in a review by Li et al. (2022) and Li and Li (2016). We used these lists as a reference to identify genes in rice that are specifically implicated in controlling grain size. Rice gene nucleotide coding sequences (CDS) were generated by searching the Gene ID number in NCBI (National Center for Biotechnology Information) database (https://www.ncbi.nlm.nih.gov/); Search criteria: All Databases). Using the rice CDS sequences as query, we searched the most similar CDS from wheat (Triticum aestivum; taxid: 4565), maize (Zea mays; taxid: 4577), barley (Hordeum vulgare L.; taxid: 4513), sorghum (Sorghum bicolor (L.) Moench; taxid: 4558) and Arabidopsis (Arabidopsis thaliana (L.) Heynh.; taxid: 3702). We also included rice (Oryza sativa (japonica cultivar group); taxid: 39947) in this search as a control comparison to ensure the correctness of the query sequence. Then, the tef CDSs were gathered using CoGeBlast, using the rice gene as the reference sequence (https://genomevolution.org/coge/CoGeBlast.pl). Rice CDS nucleotide sequences were BLAST against the tef genome (Selected Genome: Eragrostis tef (tef) id 50954 PacBio unmasked vV3; Maker, PacBio: (id 50954) vV3 unmasked 577,738,711nt). Rice sequences were used as the reference in all but two instances, where the Sorghum bicolor sequence was used to derive the RGG1 and AGO17 orthologs from tef. To simplify the phylogenetic analysis, we selected a single allele with the highest predicted similarity. However, since tef contains two sub-genomes, we have included both alleles of each putative seed size regulating gene in Supplementary File 2 . The BLAST parameters were set at an E-Value cutoff of 1e-30, with other parameters left standard. This cutoff was set to ensure the tef sequences we acquired could be classified as putative homologs or orthologs using the standard cutoff for determining homology using BLAST (E< 1e-5) (Choudhuri, 2014), or the cutoff others have indicated for added level of scrutiny for nucleotide comparisons (E< 1e-10) (Pearson, 2013).

Gene phylogenies were grouped by their common regulatory pathway or functional category. Sequences were aligned using MUSCLE and the gene phylogenies constructed using a Maximum Likelihood tree building algorithm in MEGA v11.0.13 (Tamura et al., 2021). The trees were converted to Newick file format and uploaded to Interactive Tree of Life (ITOL) for final formatting and annotation (https://itol.embl.de/) (Letunic and Bork, 2024). The Newick files are included as Supplementary Data .

Functional analysis of rice genes was conducted using ShinyGO v0.80 (http://bioinformatics.sdstate.edu/go/), which was used to identify key biological processes involved in seed size regulation in rice (Ge et al., 2020). Our list consisted of reported seed size regulating genes that have been cloned and characterized in rice, compared against the rice genome background (Oryza sativa Japonica Group genes IRGSP-1.0; TaxID: 39947). Genes were clustered by GO Biological Processes ( Supplementary File 3 ), and extracted using FDR (false discovery rate; FDR< 0.05) and enrichment score. ShinyGO defines Fold Enrichment as the percentage of genes in the list belonging to a pathway, divided by the corresponding percentage in the background (Ge et al., 2020; http://bioinformatics.sdstate.edu/go74/#:~:text=Fold%20Enrichment%20is%20defined%20as,a%20certain%20pathway%20is%20overrepresented). Quoted from the ShinyGO v.80 website, “FDR refers to how likely the enrichment is by chance; Fold Enrichment indicates how drastically genes of a certain pathway is overrepresented.” The purpose of this analysis was to highlight the pathways which are important to the regulation of seed size in rice. For those focused on tef breeding, genes of those pathways may be a key starting place for gene manipulation in tef, bearing the mechanisms regulating seed size in both cereals are similar.

Analysis of seed size attributes and ICP-OES results was performed in SAS Enterprise Guide v8.3.8.206 (SAS Institute Inc., Cary, NC USA). We conducted one-way analysis of variance (ANOVA) to determine if there was significant difference in mean seed width, seed length, and seed area among the 189 tef and 11 E. pilosa accessions. For the tef accessions, we binned accessions from each seed size attribute as small, medium, or large, and selected 10 accessions from each group for simplicity. These represent the 10 largest, medium, and smallest accessions for seed area, length, or width. For seed length, seed width, and seed area of the 30 representative accessions, we performed another one-way ANOVA, followed by a Student-Newman-Keuls (SNK) post-hoc test for means comparison. Lastly, ICP-OES results were averaged and analyzed by one-way ANOVA, followed by a SNK post-hoc test. Figures were generated using GraphPad Prism v10.0.3 (GraphPad software, Boston, Massachusetts, USA).

In this study, we analyzed the natural variation in seed size of 189 tef and 11 E. pilosa genotypes. Seed images were acquired by microscopy and seeds size attributes, including seed length, width, and area was measured. Additionally, 1000-grain weight was generated for selected tef and E. pilosa accessions.

The distribution of seed areas of sampled tef accessions is normal (x̄ = 525,565 µm²; sd = 49,102 µm²) but slightly left skewed ( Supplementary Table S2 ; Figure 2A ). The center half (interquartile range, IQR) of the tef accessions fall within a range of 498,457 µm² and 555,9739 µm², with the largest and smallest accessions being nearly 25% larger and 27% smaller than the mean, respectively. One-way ANOVA revealed a statistically significant difference (p< 0.0001) between the 189 tef varieties in mean seed area. Seed areas of accessions PI494370 and PI494453 were identified as the largest and smallest, respectively ( Figure 3A ). Seeds of PI494370 are about 51% larger than the smallest accession PI494453. To better represent our data, we selected ten accessions with the largest, medium, and smallest areas (30 in total). One-way ANOVA revealed that the difference in seed area among the 30 selected accessions is statistically significant ( Figure 3A ). A post-hoc test indicated that the ten largest accessions were significantly larger than the ten smallest. However, no significant difference was found between the medium ten accessions and six of the larger and one of the smaller tef accessions.

The distribution of tef accessions by seed length is normal (x̄ = 1142 µm; sd = 65.75 µm), and slightly left skewed, with the IQR of the accessions falling between 1095µm and 1185µm ( Figure 2B ). The 189 tef accessions were significantly different (p< 0.0001) in seed length. The longest accession was identified as PI494370, which also has the largest seed area. The seed length of PI494370 is 31% larger than PI193511, the shortest accession ( Figure 3B ), and were found to be significantly different in the 30-accession comparison (p< 0.05; Figure 3B ).

Like seed area and length, the distribution of accessions by seed width is normal (x̄ = 582.4 µm; sd = 33.98 µm) and slightly left skewed. The IQR falls within 559.1 µm and 605.7 µm. One-way ANOVA showed a statistically significant difference in seed width for the 189 tef accessions (p< 0.0001; Figure 2C ). The seed width of accessions PI524446 and PI494307 were the widest and most narrow, respectively, among the 189 tef accessions ( Figure 3C ). Seed of PI524446 are 30% wider than that of PI494307. One-way ANOVA of the 30-accession comparison of seed width revealed significant differences among the accessions. A post-hoc test indicated significant differences between the ten widest and ten most narrow accessions ( Figure 3C ). No significant difference was found between the ten medium width accessions and seven widest and six most narrow accessions. A summary of these findings are in Supplementary Table S2 .

To determine the 1000-grain weight we counted 1000 seeds of seven tef accessions representing large, medium, and small seeds, and took the weights of the seeds. PI524446, which also had the widest seeds, had the highest 1000-grain weight, while PI494307 was found to have the lowest 1000-grain weight. The latter also has the smallest seed width. The 1000-grain weight of PI524446 was 1.54 times higher than that of PI494307, whose difference was statistically significant ( Figure 2D ).

During the course of the seed size measurement and analysis, we identified a unique accession PI442115, which was substantially smaller than the tef accessions in our sample. PI442115 was a total outlier in seed size and color compared to the other tef accessions. PI442115 more closely resembled E. pilosa in color and in size than it did to tef ( Figure 4 ). It was significantly smaller and much darker in color ( Figure 5 ). This accession was mischaracterized in the USDA-ARS Germplasm Resource Information Network (GRIN) database as tef. This information prompted us to expand the study to include all the E. pilosa accessions available at the U.S. National Plant Germplasm Center.

We also measured seed length, width, and area for 11 E. pilosa accessions. Our analysis revealed significant variation among the seed size attributes of E. pilosa. Seed area of accession PI442487 (350,174 µm²) was 1.79 times that of PI442115 (151,744µm²). PI378082 has the longest seeds, and PI442115 has the shortest seeds among the E. pilosa accessions. The widest E. pilosa accession, PI263510, was 42% wider than PI442115. The 1000-grain weight of PI378082 is twice that of PI442115. Seed size attributes (area, length, width, and 1000-grain weight) of PI442115 are the smallest of all the tef and E. pilosa accessions we analyzed. A summary of these findings is in Supplementary Table S3 . The analysis of E. pilosa grain length, width, area, and 1000-grain weight revealed significant variation in seed size among all of the available genotypes. We suspect there is a genetic basis behind this natural variation, but this remains to be validated. Furthermore, there is the potential for identifying alleles from tef’s progenitor species that can shed light onto the evolution of tef domestication or can be utilized as a source of genes for introgression of agronomically desirable traits into tef. In rice, a handful of alleles regulating seed size have been identified in a similar way, using wild rice species such as Oryza rufipogon, the O. sativa progenitor, and O. barthii, the progenitor of African rice O. glaberrima, to identify alleles that were lost or selected for during domestication (Jiang et al., 2019; Luo et al., 2013; Wu et al., 2017).

Additionally, we identified morphological differences in both the tef and E. pilosa seeds. Most E. pilosa grains appear distinctly darker than tef grains. In comparison, tef grains have a lighter colored seed coat, with a smooth outer surface. Some of the E. pilosa grains showed clear darkened reticulate patterned seed coat (For an example: Figure 5C ; PI222988). This was also reported by Kreichstitz et al. (2009). None of the tef accessions we sampled expressed this type of patterning. Interestingly, not all of the E. pilosa accessions showed reticulate seed coat patterning but retained the darker seed coat coloration. Specifically, PI211030, PI221925, and PI223259 have fingerprint-like patterning and more closely resemble tef grains in this aspect (For an example: Figure 5C ; PI211030). Additionally, the non-reticulate E. pilosa accessions are more ovular ( Figure 5C ; PI211030), whereas the reticulate patterned seeds are oblong, with more pronounced emarginated apices ( Figure 5C ; PI222988). The seeds from PI263510 (E. pilosa) appeared to be more spherical, with a highly diminished length-to- width ratio ( Figure 5C ) (data not shown).

To understand the association between seed size and mineral content, we analyzed the mineral concentration of tef and E. pilosa accessions contrasting in seed size, along with a few reference varieties of tef using ICP-OES. As shown in Figure 6 , the differences in all mineral concentrations were statistically significant (p< 0.0001). The Ca concentrations in most E. pilosa accessions were higher than the tef accessions. The highest Ca concentration (3 g/kg) was detected for PI222988 and PI219588 ( Figure 6A ). The lowest Ca level (1.5 g/kg) was detected in the reference tef cultivar Dabi, along with the tef accessions with small seeds (PI494293), and large seeds (PI524446). The K concentration varied from 6.5-7.5g/kg for all accessions, and the highest concentration was detected for tef accession PI494293 ( Figure 6B ). The Fe concentration of two E. pilosa accessions PI219588 (73.47 mg/kg) and PI222988 (69.03 mg/kg) was significantly higher than tef and other E. pilosa accessions ( Figure 6C ). Seed Mn concentration was highest for PI442115 (93.37 mg/kg) followed by Dabi (68.91 mg/kg) ( Figure 6E ) while the concentration of Zn and Cu was higher for PI442115, P524446, and Dabi compared to the other accessions ( Figures 6E, F ). Our findings show that accession PI442115 tends to accumulate more micronutrients (Mn, Zn and Cu) than the other E. pilosa accession, but only some of the tef accessions. On the other hand, there was no marked difference in P and Mg concentration among the accessions ( Supplementary Figure 1 ) while the concentration of S and B was higher in PI442115 ( Supplementary Figure 1 ).

Next, using the ICP-OES data discussed above, we analyzed the relationship between seed size and mineral concentration. The seed area was plotted against mineral concentration, and a simple linear regression analysis was conducted to determine if there was a correlation between seed size and mineral content in tef or E. pilosa. We found no significant relationship between tef seed area and P (p = 0.131; R = 0.87), K (p = 0.57; R = 0.43); Ca (p = 0.84; R = 0.17), and Mg (p = 0.64; R = 0.36) concentration ( Supplementary File 4 ; Supplementary Figure 2 ). We detected a significant positive correlation between tef seed area and S concentration (p< 0.01; R = 0.99) ( Supplementary Figure 2 ). Among the micronutrients, we detected no significant relationship between tef seed area and Fe (p = 0.59; R = 0.41), Mn (p = 0.57; R = 0.43), Zn (p = 0.12; R = 0.88), Cu (p = 0.10; R = 0.89), or B (p = 0.32; R = 0.68) concentration. We found no significant association between E. pilosa seed area and P (p = 0.25; R = 0.56), K (p = 0.71; R = 0.20), Ca (p = 0.14; R = 0.67), and S (p = 0.052; R = 0.89) concentration. We detected a significant but weak negative correlation between E. pilosa seed area and Mg concentration (p< 0.05; R = 0.89). Lastly, we found no significant relationship between E. pilosa seed area and Fe (p = 0.65; R = R = 0.24), Mn (p = 0.126; R = 0.69), Zn (p = 0.40; R = 0.42), and Cu (p = 0.42; R = 0.41) concentration.

Inductively coupled plasma optical emission spectrometry (ICP-OES) is a useful strategy for quantifying minerals in different plant tissues including seeds. However, this method does not provide information on the spatial distribution of mineral elements in those tissues. Therefore, we used synchrotron-based X-ray fluorescence (SXRF) microscopy to visualize minerals in the mature seeds. This method provides semiquantitative elemental distribution at high sensitivity and high spatial resolution (Donner et al., 2012, 2013). Here we report the results of SXRF imaging of the reference tef variety Dabi, small (PI494293) and large (PI524446) seed tef accession, and the E. pilosa accession with the smallest seed (PI442115).

SXRF imaging detected calcium (Ca) apparently localized to the seed coat and was detected in the embryo ( Figure 7A ). It is barely detected in the endosperm. The Ca signal is lower in the reference cultivar Dabi, while there is no marked difference between PI494293, PI524446, and PI442115. When Ca concentration was quantified, the median concentration of Ca was lower in Dabi and slightly elevated in PI442115, with no clear difference between PI494293 and PI524446 ( Supplementary File 5 ). Potassium (K) was detected in the outer embryo and the seed coat while it appears absent from the endosperm ( Figure 7B ). For K, we also observed a substantial difference among the accessions. The K signal was stronger in accession PI494293 and PI524446 while it was weakest in the small-seeded E. pilosa accession PI442115. Quantification of K concentration showed that PI494493 and PI524446 exhibit elevated median levels of K, in comparison to Dabi and PI442115 ( Supplementary File 5 ). Furthermore, there is a greater proportion of signals that exhibit high levels of K in PI494293 and PI524446 in comparison to the other accessions. It should be noted that since the K and Ca SXRF lines have low energy, these signals are measured mainly from the surface of the seed, on the order of 10s of µm. For heavier elements, SXRF signals are measured from greater depths (100s to 1000 µm).

Iron (Fe) is detected in the embryo and the seed coat ( Figure 7C ). There is no marked difference in the signal intensity of Fe among the accessions, although it is not visible in the seed coat of PI442115, which could be due to its small seed size. Quantification of Fe shows that there is no difference in median levels of Fe among the accessions in our sample ( Supplementary File 5 ). Like Fe, zinc (Zn) is also localized in the embryo, and the seed coat ( Figure 7D ). The signal of Zn looks slightly higher in PI524446, followed by Dabi, compared to accessions PI494293 and PI442115. Quantification of Zn shows that median levels of Zn are slightly higher in PI524446, but otherwise there is no difference detected in other accessions ( Supplementary File 5 ). The manganese (Mn) signal was higher in PI442115, followed by Dabi, where it is detected in the embryo and the seed coat ( Figure 7E ). The Mn signal was lower in the other accessions. Quantification of Mn showed that the median levels of Mn are elevated in Dabi and PI442115, but that a larger proportion of detected signals exhibit high levels of Mn in PI442115 ( Supplementary File 5 ). The signal of copper (Cu) was higher in PI524446 as compared to Dabi and accessions PI494293 and PI442115 ( Figure 7F ). Quantification of Cu indicated elevated levels of Cu in PI524446, and a greater proportion of signals from PI524446 show higher levels of Cu than the other accessions ( Supplementary File 5 ). Cu is detected in the embryo and the seed coat. Overall, seed size does not appear to affect the general patterns of mineral localization in the seed.

This study established that there is natural intraspecific variation in seed size, which indicates that there is likely a genetic basis for this trait. It is known that there are a number of factors that influence final grain size, including numerous genetic factors that are under scrutiny (Li et al., 2022). We hypothesized that the mechanisms regulating seed size in both tef and rice are potentially similar, with other studies indicating high conservation of seed size regulating mechanisms among the cereals (Tao et al., 2020a; Long et al., 2024; Zhang et al., 2015). We gathered the sequences of genes reported to influence grain size in rice and identified putative orthologs in tef using the rice genes as a reference. Although C₄ grasses such as the millets and sorghum are more closely related to tef, the genetic mechanisms regulating seed size are less established in these species. The mechanisms regulating these traits are better characterized in rice. As a result, we chose to use rice as the reference.

72 seed size regulating genes in rice were collected, representing numerous regulatory mechanisms and gene families, and identified putative orthologs in tef. Then, we conducted a phylogenetic analysis to determine the similarity of the tef sequences to highly similar sequences from other common cereals. The phylogenetic analysis was limited to four gene families, including G-protein pathway genes, mitogen-activated protein kinase (MAPK) signaling, the ubiquitin-proteasome pathway, and brassinosteroid (BR) signaling and biosynthesis, all of which have been shown to influence grain size in rice. Lastly, we conducted a functional analysis of reported seed size regulating genes in rice to establish which mechanisms were most represented and highly influential in rice seed size regulation. For the analysis we used a list of seed size regulating genes from rice, analyzed against the background rice genome. The genes were clustered by the gene ontology (GO) term Biological Processes. Only genes that have been functionally cloned and characterized from rice were included in the list. For those focused on tef breeding, genes of highly enriched pathways may be a key starting place for gene manipulation in tef, if the mechanisms regulating seed size in both cereals are similar.

The functional analysis of seed size regulating genes suggested that G-protein-coupled receptor signaling (FDR = 1.9E-10) was the most significantly enriched biological process, followed by brassinosteroid-mediated signaling (FDR = 2.4E-18) ( Figures 8 , 9 ). The other highly enriched biological processes involve steroid hormone signaling and response. This indicates that BR signaling, BR biosynthetic regulation, and G-protein mediated signaling are highly important for the regulation of seed size in rice.

Based on the phylogenetic analysis, most genes reported to control grain size in rice show high similarity to the loci we identified in the tef genome. For all trees, only the most similar tef sub-genome feature was displayed for each gene. In the phylogenetic analysis of genes grouped into ubiquitin-mediated pathways, we clustered a total of 57 sequences from seven plant species, including rice, wheat, maize, sorghum, barley, tef, and Arabidopsis. These genes clustered into eight groups (LG1, GRX8, TUD1, GW6a, bZIP47, GW2, CLG1, and WTG1) ( Figure 10 ). For genes grouped into the MAPK pathway for the phylogenetic analysis, we gathered a total of 56 sequences from seven plant species. These genes clustered into seven groups (MKKK10, MKKK70, MKP1, MPK6, MAPKK4, MKKK62, MKK62+MKKK70). For the phylogenetic analysis we display one ortholog identified from tef from each of the rice reference sequences except in the MKKK70 cluster, where BLAST for MKKK62 and MKKK70 returned identical sequences in tef and sorghum ( Figure 11 ). From the G-protein mediated signaling genes, we gathered 47 sequences among the seven species and clustered them into nine groups (RGG2, RGG1, DEP1, RGA1, RGB2, GS3+DEP1+GGC2, GS3, and DEP1+GGC2). Two additional groups formed from BLAST returning identical sequences from GS3, DEP1, and GGC2 in Arabidopsis, and BLAST returning identical sequences from DEP1 and GGC2 in barley. In the G-protein phylogenetic analysis, we report one tef ortholog in seven of the clusters RGG2, RGG1, DEP1, RGA1, RGB2, and GS3 ( Figure 12 ). In BR signaling or biosynthetic-family genes, we collected 154 sequences from seven species, which clustered into 22 groups. An additional group resulted from BLAST of GSK2 and GSK3 returning a single Arabidopsis sequence. For the purposes of the phylogenetic analysis, we report 21 tef putative orthologs. We identified other putative orthologs from tef which showed significant similarity (E< 1e-30) with rice genes from other biological processes, but they are not included in our phylogenetic analysis. We have also included the corresponding homologous sub-genome feature from tef. The complete list of putative orthologous sequences that were identified in tef is available in Supplementary File 2 .

For simplicity, the phylogenetic analysis of ubiquitin-proteasome family genes includes a single tef sub-genome feature with the highest degree of similarity. From genes involved in the ubiquitin-proteasome pathway, we identified putative orthologous sequences from tef for the genes encoding a ubiquitin-specific protease (UBP15; LG1), a RING ubiquitin E3 ligase (HRZ2), an otubain-like protease (OTUB1; WTG1), a CC-type glutaredoxin protein (GRX8;WIDE-GRAIN 1; WG1), a histone acetyltransferase (OslIHAT1; GW6a), a U-box E3 ubiquitin ligase (TUD1), an E3 ubiquitin ligase (GW2), and a bZIP transcription factor (bZIP47) ( Figure 10 ). The tef ortholog of OsUBP15 forms a single cluster with close similarity to the maize and sorghum UBP15 orthologs, both of which encode UBP15. Our analysis showed a single cluster of genes referenced from OsGRX8, with high confidence of similarity between the rice and tef sequences (94.1% bootstrap value). Rice and tef also share some similarities for TUD1 but to a lesser extent than UBP15. Our analysis indicates that both TUD1 and lHAT1 found in rice are conserved among our sampled Poaceae species, with little similarity to the predicted Arabidopsis sequences. Tef sequences share the same level of similarity to OslHAT1 and the predicted barley and wheat orthologs. Similarly, tef shows equal similarity to the rice bZIP47 and other grass-family bZIP47 orthologs. GW2 referenced genes form a single cluster, with the tef ortholog sharing the closest similarity to the maize and sorghum sequences. CLG1/HRZ2 from rice is highly similar to the wheat sequence and shares a sub-cluster with tef. Oddly, the predicted CLG1/HRZ1 ortholog in barley is most similar to the sorghum OTUB1 sequence. The tef ortholog of OTUB1 shares a high bootstrap value with OsOTUB1 and other grass-family OTUB1 predicted orthologs. Our analysis suggests the tef genome contains two copies of each of the aforementioned genes, i.e. two homologous sub-genome features with high degrees of similarity. In tef, we identified an additional, highly similar copy of GW6a on chromosome 1A, which was the second closest genomic feature returned from the BLAST results.

We report five orthologs in tef that resemble genes involved in MAPK-mediated regulation of seed size in rice. Tef has at least two homologous sub-genome features for each MAPK-family gene we analyzed. Notably, we found an additional copy of MKK4 with significant similarity on chromosomes 1A and 1B. The putative tef ortholog of OsMAPK6 shows low similarity to the rice gene, but closer similarity to the T. aestivum (wheat) MAPK6 ortholog (XM_044568146.1), predicted to encode a MAP Kinase ( Figure 11 ; Supplementary File 2 ). Both the tef and wheat orthologs of MAPK6 form a cluster that shows higher similarity to the clusters for MAPKKK70 and MAPKKK62. The tef ortholog of MKK4 forms a cluster of MKK’s in wheat, sorghum, rice, corn, and Arabidopsis. Our phylogenetic analysis indicates a high confidence of similarity (100% bootstrap value) between the predicted tef MKKK10 sequence and the rice MKKK10 sequence and form an obvious cluster with a high bootstrap value with the other five species we analyzed. Interestingly, the predicted tef ortholog of MKP1 shows close similarity with the rice sequence for MKKK70, but next to the cluster of rice MAPK6 and four other MAPK6 orthologs. OsMKKK62 shows high similarity to the sorghum and maize predicted orthologs of MKKK70, but little resemblance to the redundant kinase OsMKKK70. Both rice sequences MKKK62 and MKKK70 returned identical sequences in tef and sorghum. This indicates that tef and sorghum potentially lack the redundant version of MKKK70.

Using the sequences for the standard seed size regulating G-proteins in rice ( Figure 12 ), we were able to identify all but one of tef putative orthologs using rice sequences. Each G-protein family gene we analyzed returned two homologous sub-genome features in tef. For RGG1, we used the sequence from S. bicolor (sorghum) (XM_002464159.2) to generate the CDS in tef. From the phylogenetic analysis, the RGG1 tef ortholog shows no significant similarity to the rice RGG1, but high similarity to the Z. mays (maize) and sorghum sequences, which are annotated as Gγ proteins. We also predict coding sequences in tef that show high similarity to the rice genes RGA1 and RGB2. In addition to RGG1, DEP1 and GGC2 orthologous tef sequences show closer similarity to the maize and sorghum sequences, than to the rice sequences. The tef ortholog of GS3 shows lower similarity to the rice sequence but shares a high bootstrap value with a cluster that includes the GS3 gene in rice, in addition to the barley, wheat, sorghum, and maize GS3 putative orthologs.

The phylogenetic analysis of BR pathway associated genes were limited to 21 reference genes ( Figure 13 ; Supplementary File 2 ). Fourteen of the reference genes returned two highly similar homologous sub-genome values in tef. We found a significantly similar third copy of BRD2 on chromosome 6A. GW5 and GS5 returned two sub-genome values, but in both cases, other loci on different chromosomes were ahead of the second homolog in similarity. We also found several highly similar copies of D11/CYP724B1 along chromosomes 7A and 7B. GW10 returned a third sub-genome feature on chromosome 5B (E-value = ~ 0.0). From the gene phylogeny, we identified a locus in tef that shows high similarity to the rice GSK2 sequence, along with the sorghum and maize GSK2 orthologs. The predicted tef GSK3 ortholog shows high similarity to the maize and sorghum GSK3 sequences, which both encode SHAGGY-related kinases. BRD1, BRD2, GL3.1, GS5, BAK1 and BZR1 show a high degree of conservation among the seven species we analyzed, with clear clustering of sequences with high bootstrap values. The cluster of genes generated from the OsD11/CYP724B1 sequence shows high similarity among the monocots, with Arabidopsis forming an outgroup of the cluster. The tef orthologous sequence of OsD11/CYP724B1 was most closely related to the maize gene, which also encodes a CYP724B1. The cluster forming LAC-related sequences formed a single cluster with a high bootstrap value. However, within the LAC cluster, the separation between the homologs is less clear. Regardless, the tef and sorghum OsLAC orthologs show the closest similarity. The BRI, GW5, GS2/GRF4, OFP1, AGO17 and OFP8 clusters indicate high conservation among the monocots, with low similarity to the Arabidopsis sequences. Interestingly, the tef orthologous sequences to WRKY53 and POW1 show low similarity to the rice and other sequences we analyzed. The tef ortholog of GS9 sequences appeared to be a partial sequence and shows low similarity to the main cluster with OsGS9 and GS9 orthologs in the other monocots. The tef GS9 and DLT sequences form a two-leafed group in our analysis, both of which are not included in the main clusters. The cluster for GW10 forms a clear group, including Arabidopsis, with a 99.2% bootstrap value. Oddly, the maize GW10 ortholog is much more distant. The sorghum GW10 sequence is predicted to encode CYP450-89A2, the same as OsGW10 and exhibits the highest similarity to the putative tef ortholog.

Studying the spatial distribution of minerals in seeds is important for understanding the fundamentals of seed development, nutrient bioavailability, bio-accessibility, and strategies for biofortification. Cereals provide a substantial portion of the mineral nutrients acquired through diet (Arafsha et al., 2023) and mineral malnutrition remains a global health concern and is widespread in both the developed and developing world (Zimmermann and Hurrell, 2007). To study the mineral content of biological samples such as grains, inductively coupled plasma mass spectrometry (ICP-MS) or ICP-OES has been widely used (Dame, 2020; Woldetsadik et al., 2024). However, it gives no insight into localization patterns and potential bioavailability.

To study the spatial distribution of minerals, researchers have turned to synchrotron-based x-ray fluorescence (SXRF) imaging to visually study and quantify the minerals within biological tissues. XRF imaging is a non-destructive imaging technique that allows in situ 2D quantification of elements (Donner et al., 2012, 2013). Some studies have utilized XRF spectroscopy to study hyperaccumulators (Goudard et al., 2024). Others have utilized XRF imaging for studying micronutrient mobilization during seed germination in vivo (Takahashi et al., 2009). SXRF microscopy has proven to be useful in studying alterations in mineral accumulation in loss of function studies (Chia et al., 2023; Sheng et al., 2021). In this report, we utilized SXRF imaging to explore the distribution of minerals within the tef grains for the first time. Our findings show that Ca is associated with the seed coat and the embryo while K is detected in the outer embryo and seed coat, and the micronutrients are predominantly associated with the embryo and seed coat. The concentration of minerals was also variable among the genotypes we analyzed ( Figure 5 ). A similar distribution pattern has been reported in rice, where minerals were shown to localize in the embryo and remain absent from the endosperm (Lu et al., 2013).

For processing and consumption, mineral localization is nutritionally important. In many cereals, the milling process separates the embryo and endosperm (Shewry et al., 2020). Because mineral localization is variable, this may inadvertently remove micronutrients from the final product. The whole seed is milled in some cereals, such as tef, leaving the endosperm and embryo components together. Milling the entire seed potentially increases the overall amount of nutrients in the flour. However, the localization of minerals in different seed tissues may alter bio-accessibility or dictate how easily minerals are accessed from seed tissues upon digestion (Arafsha et al., 2023). For example, the plant cell wall is resistant to digestion in the human digestive tract. The cells of the aleurone layer are composed of water-insoluble indigestible fibers (Arafsha et al., 2023; Brouns et al., 2012). The starchy endosperm is the most digestible part of the seed, but lacks minerals, which was evident in the tef accessions we analyzed (Gidley, 2024; Khalid et al., 2023) ( Figure 5 ). Furthermore, the accessibility of minerals can be affected by how the mineral is stored in the seed chemically. For example, prominent levels of phytate can bind minerals in the seeds, making them less accessible during digestion (Mandha and Raes, 2023). Phytate tends to localize at the highest concentration in the aleurone layer and the embryo (Bohn et al., 2008). Baye et al. (2015) addressed this issue though the additions of phytate degrading enzymes in tef flour to improve iron bioavailability. Furthermore, tef is usually consumed after fermentation, which was reported to decrease the concentration of antinutrients such as phytate and increase mineral bioavailability (Fischer et al., 2014; Shumoy and Raes, 2017).

Plant domestication has followed convergent, but highly similar processes globally among many cultivated crops (Fuller et al., 2014). In cereals, the major traits selected were shattering and lodging resistance, along with greater seed size to improve yield and quality. Many domestication traits were selected for purposely, whereas others may have been inadvertently acquired through genetic linkage, pleiotropy, or otherwise unconsciously (Purugganan, 2019). The increase in seed size among the cereals is contested, with some arguing that enlarged seed size was consciously selected for, which would bring higher yield and ease of handling. Others have argued that seed size was linked to characteristics such as height, suggesting that the enlargement of seed size was unconscious (Purugganan, 2019; Meyer and Purugganan, 2013). In our study, we show that tef seeds can be as much as 4.44 times larger by area and 2.35 times longer than its progenitor E. pilosa. The evolution of grain size in tef remains to be fully understood.

In some cases, individual genes that produce highly desirable traits and share commonality among closely related cultivars or between species, show signs of selection at the genetic level (Meyer and Purugganan, 2013). The capability to sequence genomes reliably and efficiently for genome wide association studies (GWAS) and quantitative trait loci (QTL) mapping has provided novel insights to the genomic evolution of domesticated cereals. In rice, this has revealed the genetic basis of seed size, a key determinant of yield in cereals (Tao et al., 2020b). We anticipate that future research will reveal such insights in tef.

Seed size is an extraordinarily complex trait that is influenced by both genetic and environmental conditions. Furthermore, there seems to be a tradeoff between offspring number and size, as reported by those studying natural variations and single gene mutants (Gnan et al., 2014; Zhang et al., 2017a, 2017b; Xiao et al., 2019). In natural populations, there is strong environmental pressure that determines seed size, and variation exists across species (Venable, 1992). Razzaque et al. (2023) showed that local adaptation to climate was a major predictor of seed size of the perennial grass Panicum hallii. They showed that annual temperature showcased the strongest predictability in determining seed size among P. hallii ecotypes. During the course of domestication, humans artificially selected against the natural plant dispersal mechanisms (shattering) and selected for increased seed size (Harlan et al., 1973). Harlan et al. (1973) revealed that the transition from wild grasses to their domesticated counterparts follow highly similar trajectories among species. Fuller et al. (2014) substantiated these ideas, highlighting extensive parallelism (convergence) in the evolution of non- shattering species and the enlargement of seeds in cereals. In tef, one estimate indicates a 72% difference in grain thickness in tef and E. pilosa, which is consistent with our estimates (D’Andrea, 2008). D’Andrea (2008) speculated that early tef cultivators encountered several obstacles during domestication and that tef’s domestication history was highly unconventional in comparison to other cereals. In speculation about why tef grains have remained so small, one hypothesis suggests that natural introgression with E. pilosa may have helped maintain small seed sizes. Early cultivators more likely recognized tef’s susceptibility to shattering and lodging and may have selected against large panicles and large, heavy grains (D’Andrea, 2008; Ketema, 1997). Lastly, water-conservation practices and attempts to reduce lodging resulted in minimal tilling practices in Ethiopia’s semi-arid highlands, which is hypothesized to reduce selection pressure for large seeds (Harlan et al., 1973). Furthermore, D’Andrea (2008) notes that modern Ethiopian farmers do not select for large seeds, but rather for color, appearance, increased panicle branching, and more numerous grains, suggesting that any improvement to grain size in tef has been non-deliberate or unconscious (Purugganan, 2019; Meyer and Purugganan, 2013). For those interested in breeding large seed cultivars of tef, these limitations remain a major point of consideration.

The relationship between seed size and mineral content remains elusive in tef. Previously, there has been some speculation regarding the relationship between seed size and mineral content. It is suggested that higher mineral contents of the seed may confer a major evolutionary advantage. Additionally, smaller seeds are more easily dispersed and evolutionarily beneficial but have become agronomically undesirable in many seed-bearing crops. However, some have suggested that any benefit conferred from a larger seed diminishes if the external nutrient availability is low, when smaller, nutrient-dense seeds may be much more advantageous (Grubb and Burslem, 1998). Curiously, Grubb and Burslem (1998) found an inverse relationship between mineral content and seed size. They showed that larger seed size diminished N, P, K, Mg, and Ca concentrations in several different crop species. It is interesting to speculate that the ancestral species of tef, E. pilosa, may have benefited evolutionarily from this kind of selection. However, this remains to be validated. The accession with the smallest and lightest grains from either the tef or E. pilosa accessions was PI442115 (E. pilosa), which showed the highest concentration of Cu, Zn, and Mn from the ICP-OES, and had clearly higher levels of Mn in the SXRF ( Figures 6 , 7 ). However, in PI442115, the mineral concentration of other elements was relatively lower, potentially indicating some kind of trade off. The preliminary results of whether there is any relationship between seed size and mineral concentration in tef or E. pilosa seeds are inconclusive. We will continue to pursue this by analyzing the mineral content of more tef genotypes, where we can begin to unravel the relationship between seed size and mineral content further.

The genetic mechanisms regulating seed development and seed size are complex and poorly understood. Even in a model monocot such as rice, the underlying mechanisms regulating seed size remain fragmented. Regardless, developing novel rice cultivars with enhanced seed size is a major area of research, as larger seed size is often associated with improved yield and quality. In tef cultivation, its small seed size presents numerous challenges to its cultivation. Much of the general biology, let alone the genetic mechanisms regulating seed development, remains to be elucidated in tef. Additionally, the tef genome remains to be fully annotated.

In rice there are ~80 genes that have been implicated in seed size regulation (Li et al., 2022). Some of these have been selected for over the course of domestication, whereas many others represent rare alleles identified from mutagenic studies or evaluation of rice progenitor species. Most of these genes regulate seed size via changes to cell proliferation, elongation, and enlargement in the seed hull (and related tissues), or in maternally derived tissues such as the embryo or endosperm (Li et al., 2022). Many of the genes encode receptors, proteins involved in various signaling mechanisms, hormone biosynthetic enzymes, transcriptional regulators, and transcription factors. For simplicity, they can be grouped by a shared pathway or function, but many have been shown to be multi-pathway regulating or multifunctional.

Our analysis suggests that G-protein mediated signaling is the most significantly (FDR< 0.05) enriched biological process regulating grain size in rice ( Figures 8 , 9 ). Several G-proteins have been implicated in seed size regulation in rice, many of which were identified as major quantitative loci. The rice genome encodes one Gα subunit (RGA1), one Gβ subunit (RGB1), and five Gγ subunits (RGG1, RGG2, GS3, DEP1, and GGC2), which compose the major G-proteins in rice (Mao et al., 2010; Fan et al., 2006; Sun et al., 2018; Tao et al., 2020b; Miao et al., 2019). Plants also exhibit a number of non-canonical extra-large GTP-binding proteins (XLGs), which are large atypical Gα subunits (Maruta et al., 2021). XLGs have been shown to regulate several complex traits including panicle architecture, stress tolerance, and some agronomic traits such as 1000-grain weight (Biswal et al., 2022; Zhao et al., 2022; Cantos et al., 2023). However, compared to the other G-proteins we used in our analysis, these XLGs are not as well characterized and their effects on seed size regulation are still under active investigation. As a result, we chose to focus our study on the better characterized G-protein subunits and omit XLGs from our analysis. Unlike animals, plants lack G-protein coupled receptors (GPCR’s) and are self-activated (Urano and Jones, 2014). To transduce the signal, Gβ and Gγ subunits form a Gβγ heterodimer, which go on to interact with Gα (Maruta et al., 2021; Urano and Jones, 2014). Evidence suggests that various combinations of G-protein subunits exhibit unique alterations to key developmental processes and specific alterations to various grain size attributes, either positively or negatively (Sun et al., 2018; Pandey, 2020). We were able to identify putative orthologs in tef for all but one of the G-protein family genes using the rice CDS (Criterion from BLAST: E< 1e-30). We found no significantly similar sequences in tef using the OsRGG1 sequence but found that the maize and sorghum RGG1 orthologs produced the same tef sequence, which was validated by our phylogenetic analysis ( Figure 12 ). Otherwise, the other tef sequences we identified pass the standard cutoff for the determination of homology using BLAST (Choudhuri, 2014; Pearson, 2013).

The ubiquitin-proteasome pathway regulates protein stability, activity, and degradation in eukaryotes. In plants, ubiquitin-mediated processes contribute to many biological processes, such as embryogenesis, hormone signaling, and senescence (Moon et al., 2004). Eight genes involved in ubiquitin-mediated regulation of seed size have been reported in rice. Many of these genes have been identified in other grasses and in Arabidopsis, where they have exhibited highly conserved functions (Du et al., 2014; Tang et al., 2023; Shi et al., 2019; Keren et al., 2020; Li and Li, 2014; Wang et al., 2020; Zombori et al., 2020; Brinton et al., 2018). Most of the genes grouped in ubiquitin-mediated pathways act as regulators of other systems, and exhibit crosstalk with the G-protein, BR, GA, and MAPK pathways, in addition to regulating the activity of some transcription factors (Bai et al., 2024; Yang et al., 2021; Hu et al., 2013; Hao et al., 2021; Tian et al., 2021). From tef, we report eight putative orthologs involved in ubiquitin-mediated regulation of grain size, along with their homologous sub-genome feature ( Figure 10 ; Supplementary File 2 ) (Choudhuri, 2014; Pearson, 2013).

The mitogen activated protein kinase (MAPK) cascade is a central signaling mechanism in eukaryotes (Taj et al., 2010). MAPK cascades are comprised of closely related kinases that are sequentially phosphorylated, eventually leading to a final activated MAPK that phosphorylates transcription factors and other transcriptional regulators (Hardie, 1999). In rice, a three-gene MAPK module serves as a positive regulator of grain size (Liu et al., 2021; Xu et al., 2018; Liu et al., 2015; Duan et al., 2014), with a single phosphatase acting as the negative regulator of the module (Guo et al., 2018; Xu et al., 2018). Furthermore, two redundant MKKK’s have been shown to regulate the activity of the transcription factor OsWRKY53 and influence grain size via the MAPK module mentioned above (Liu et al., 2021; Tian et al., 2021). Of the six MAPK-related seed size regulating genes reported in rice, only five putative orthologs could be described in tef, along with their sub-genome homolog ( Figure 11 ; Supplementary File 2 ). BLAST of the tef genome with OsMKKK62 and OsMKKK70 returned the same tef sequence, indicating that the tef genome likely lacks the paralog of this kinase.

Several phytohormones and their associated regulatory mechanisms have been linked to seed development, including brassinosteroids (BRs), auxin (IAA), gibberellins (GA), and cytokinin’s (Li et al., 2022). The functional analysis concluded that brassinosteroid signaling and homeostasis was a significantly enriched biological process involved in seed size regulation in rice ( Figures 8 , 9 ; Supplementary File 3 ). We report 21 putative orthologs in tef from rice genes involved in BR biosynthesis and regulation, along with their sub-genome homolog ( Figure 13 ). Furthermore, several transcription factors, functional proteins, microRNAs (miRNA’s), and proteins involved in endosperm development are reported to determine final grain size and weight in rice. Of such genes, we report 29 putative orthologs in tef, along with their corresponding sub-genome homolog ( Supplementary File 2 ).

This study characterized the seed size phenotype of 189 tef and 11 E. pilosa genotypes. Unfortunately, this is reflective of most, but not all of the tef population available at the USDA-ARS germplasm center. We feel our analysis likely reflects the true nature of the tef population but remains incomplete without more research. In our analysis we only utilized tef and E. pilosa accessions whose seeds exhibited relatively homogenous coloration and treated samples with different colored seeds as impure. To ensure that our seed size analysis was as accurate as possible, we excluded samples with multi-colored seeds. Additionally, the seeds used in this study were from stock collections from the U.S. National Plant Germplasm Center to ensure they were sourced from a single location, grown under the same field conditions. Notably, replication of many of these accessions under greenhouse conditions indicated that seed size is true to type (data not shown).

Furthermore, there is no literature on the anatomy of tef seeds, so the conclusions we make for the SXRF imaging are highly general. To better understand seed development in tef, the impact of genetic manipulation, and the localization of minerals, more research into the developmental anatomy of tef is required. Moreover, the tef genome annotation remains to be completed, so validation of the findings from our comparative genomic analysis is impossible without further research. Future research into tef should aim to address these limitations. Regardless, we hope the findings from this research can provide a starting point for those interested in breeding high quality, nutritious tef, and for those interested in the evolutionary genetics of tef domestication.