2.1 Phenotype changes of tef seedlings exposed to low Ca²⁺

Although tef is known to accumulate high levels of Ca²⁺ in both the straw and grains, the effect of low Ca²⁺ treatment on tef growth and development are unknown. In this study, 6-day-old seedlings were transferred to control (1 mM Ca²⁺) or low Ca²⁺ (.01 mM) hydroponic solution and plants were evaluated after 4 weeks of growth. As shown in Figure 1, low Ca²⁺ treatment decreased plant growth as compared to control plants. Ca-deficient plants exhibited symptoms including leaf necrosis, leaf curling, and growth stunting, while control plants produced more biomass without marked symptoms. Roots of control plants were slightly longer than those grown in low Ca²⁺ solution, but there was no marked difference in root mass between low and optimal Ca²⁺.

2.2 Characterization of small RNAs via high-throughput sequencing

We performed here miRNA sequencing of control and Ca²⁺-deficient roots and shoots to understand the pattern of miRNA expression in tef. miRNA-seq was performed using the Illumina HiSeq 2500. After filtering raw sequencing reads, clean reads were mapped to small RNA (sRNA), transfer-RNA (tRNA), ribosomal-RNA (rRNA), and small nuclear RNA (snoRNA) (Table 1). The number of raw reads ranged from ~8.9–22 million in the four replicates of roots treated with low calcium (LCR); after filtering, ~7.5 to 18.1 million clean reads were obtained. Similarly, in shoots under low calcium conditions (LCS), the number of raw reads ranged from ~14.7–44.7 million and after filtering, the number of clean reads ranged from ~13.1–39.9 million reads. Overall, more reads were obtained from shoots as compared to roots under low Ca condition (Table 1). Furthermore, higher sRNA reads (~4.1–9.7 million in roots and over ~7.5–21.5 million in shoots) were detected as compared to tRNA, rRNA, and snoRNA. In control roots (root treated with 1 mM Ca²⁺, ConR), the number of raw reads ranged from ~22.9 million to 43.1 million; however, after filtering, the range of clean reads was ~20–37.4 million. Similarly, in control shoots, designated as ConS, the number of raw reads ranged from ~17.1–38.9 million, while after filtering, the number of clean reads ranged from ~14.9–36.1 million. Overall, higher sRNA reads were observed for samples from ConR plants as compared to low Ca²⁺ plants. Moreover, variation in the number of sRNAs was observed between samples, tissues and Ca²⁺ treatment (Table 1).

The sizes of detected sRNA in both low Ca²⁺ and control treatments were in the range of 17 to 35 nt (Figure 2 and Table S1). In most samples from low Ca²⁺ treatment, 24-nt reads were more abundant than other sRNAs. Furthermore, the read number in the LCS was higher than the LCR. The size distribution of the sRNAs observed in control samples was similar with that of low Ca²⁺ samples. Reads with 21- and 35-nt reads were more abundant in ConS (Figure 2 and Table S1) while 21- and 24-nt reads were more abundant for most ConR.

2.3 miRNA identification

Across all samples, 350 unique novel and 161 already known miRNAs were identified as shown in Tables S2 and S3, and the total novel and known miRNAs along with their read counts for all the treatments LCR, LCS, ConR, and ConS are listed in Table 4. All miRNAs were found to have predicted hairpin structure, which is conserved among regulatory miRNAs. Homologous sequences were not found in miRbase (http://www.mirbase.org/) for all novel miRNAs, which were derived from the 3′ and 5′ sequences (referring to the position within the hairpin). Furthermore, a total of 161 known miRNAs that we identified in this study are homologous to those previously reported in tef and other plant species. For example, 14 miRNAs matched those previously identified in tef including tef-miR1219b_5p, tef-miR461a_3p, and tef-miR5387a_5p tef (Martinelli et al., 2018); 45 miRNAs matched to Brachypodium distachyon miRNAs including bdi-miR160a-5p, bdi-miR168-3p, bdi-miR393a, and bdi-miR2118a (Unver & Budak, 2009); 47 miRNAs matched to miRNAs identified previously in rice including osa-miR408-5p, osa-miR399a, and osa-miR396e-5p (Li et al., 2005); 29 miRNAs matched to A. thaliana miRNAs including ath-miR164a, ath-miR167a-5p, and ath-miR156a-5p (Rajagopalan et al., 2006); six miRNAs matched to Camelina sativa (cas-miR166c-3p) (Poudel et al., 2015); and six miRNAs matched to Z. mays including zma-miR156a-3p, zma-miR162-5p, and zma-miR529-5p (Gupta et al., 2017); 10 miRNAs matched to Glycine max including gma-miR156k, gma-miR166h-3p, and gma-miR169e (Joshi et al., 2010); three miRNAs matched to Citrus trifoliata (ctr-miR166 and ctr-miR171) (Song et al., 2010); and one miRNA matched to the previously identified miRNA in Avicennia marina (ama-miR156) (Khraiwesh et al., 2013) (Table S3).

The Venn diagram in Figure 3 shows miRNAs uniquely or commonly expressed in the four groups (LCR, LCS, ConR, and ConS). In low Ca²⁺ treatment, the number of miRNAs detected in roots was 619, 563 (68%) miRNAs were commonly expressed across all the samples, 11 were commonly expressed in LCR and ConR, two were expressed in LCR and ConS, 10 were expressed in LCR, LCS, and ConR, 19 were common to LCR, ConR and ConS, and 13 were common to LCR, LCS, and ConS. Only one miRNA was uniquely detected in the LCR. The total number of miRNAs in the shoots of low Ca²⁺ treatment (LCS) was 761 including the 563 miRNAs that are commonly detected under all treatments. Four were commonly expressed in LCS and LCR, 10 were common to LCS, LCR, and ConR, 13 were common to LCS, LCR, and ConS, 65 were common between LCS and ConS, and 97 were commonly expressed among the LCS, ConR, and ConS while nine were uniquely detected in LCS. Similarly, a total of 707 miRNA were detected in the ConR. Pairwise comparison of miRNAs and those commonly detected in the four groups (ConS, ConR, LCS, and LCR) is shown in Figure 3. A total of 788 miRNAs were detected in ConS, of which only 27 miRNAs were uniquely expressed while the remaining were commonly expressed also in roots and under low Ca²⁺ condition.

2.4 Differentially expressed miRNAs

The expression levels of miRNAs were analyzed in shoots and roots of control and low Ca²⁺ treatment. For both Ca²⁺ treatments, the number of DEMs was higher in roots than shoots (LCR vs. LCS and ConR vs. ConS) (Figure 4). Pairwise comparison of miRNAs detected in different treatments is shown in Figure 4a. A total of 19 DEMs were detected in LCS compared to ConS, whereas 38 DEMs were detected in LCR compared to ConR. Moreover, 166 DEMs were detected in LCR compared to LCS, while 184 were expressed in ConR compared to ConS. Interestingly, there were no DEMs common to the different comparisons (LCS vs. ConS, LCR vs. ConR, LCR vs. LCS and ConR vs. ConS). Similarly, there was no DEMs common to LCS versus ConS, LCR versus ConR, and LCR versus LCS. In the LCS, 19 miRNAs were upregulated compared to ConS, while there were no downregulated miRNAs detected in LCS compared to ConS (Figure 4b and Table S5). Expression of 26 miRNAs was higher in LCR compared to ConR including tef-novel-201_5p, known112_5p and known046_3p, while 12 miRNAs were downregulated in LCR compared to ConR including tef-novel-314_3p and known157_5p (Figure 4b and Table S5). The number of DEMs between root and shoot tissues is greater than those differentially expressed between the Ca²⁺ treatments. A total of 96 miRNAs were upregulated including known155_3p, tef-novel-043_5p, tef-novel-043_3p, known117_5p, and known118_5p in LCR as compared to LCS, while 70 miRNAs were downregulated in LCR including tef-novel-067_5p, tef-novel-067_3p, known084_5p, known085_5p, known084_3p, and known085_3p compared to LCS (Figure 4b and Table S5). Similarly, 117 miRNAs were upregulated, while 67 miRNAs were downregulated in ConR compared to ConS (Figure 4b and Table S5). Comparing both roots and shoots of plants grown under low calcium condition, 96 miRNAs were upregulated while 70 miRNAs were downregulated in LCR compared to LCS. Taken together, miRNA expression is tef appears to be more influenced by the tissue type than the Ca²⁺ levels.

2.5 Identifying miRNA targets

As described in methods section, an online database “psRNA Target Server” (http://biocomp5.noble.org/psRNATarget/) (Dai & Zhao, 2011) was used to understand the possible function of the miRNAs., miRNA targets with the expectation scores of 0 to 3.5 were selected as target genes (Table 6). Target genes which are involved in certain functions were identified for each DEMs and listed in Table S7. A total of 11,458 putative target genes were identified for the miRNAs. Among these, 5606 genes were annotated. Furthermore, among the annotated set, 817 genes belong to an uncharacterized gene family (Table S7) while the remaining target genes belong to gene families that are known to participate in various biological processes including ion transport, signaling, and transcriptional regulation.

The predicted miRNA target transporter genes include cation transporters (calcium-transporting-ATPase-10,-plasma-membrane-type, calcium-transporting-ATPase-4,-endoplasmic-reticulum-type, copper-transporting ATPase HMA5, and zinc_transporter_6), phosphate transporters (inorganic-phosphate-transporter-2-1,-chloroplastic), various sugar transporters (sugar-transport-protein-7, sugar-transport-protein-MST3, MST4, and MST6), a sucrose transporter (sucrose_transport_protein_SUT1), several polyamine transporters, chloride channels (Table S9), as well as plasma membrane-type and endoplasmic reticulum-type ATPase 4 (Table S7). Similarly, several genes involved in signaling, including calcium-dependent protein kinases (CDPK 1, CDPK 8, CDPK 9, CDPK 12, CDPK 13, CDPK 20, and CDPK 4 isoform X2) were identified as target of the miRNAs (Table S7). Other identified miRNA target genes were auxin response factors (ARF8, ARF10, ARF12, ARF14, ARF17, ARF18, ARF22, and ARF25). Moreover, transcription factors such as WRKY (WRKY24, WRKY27, and WRKY48), leucin zipper protein targets (HOX9, HOX10, HOX11, HOX14, HOX20, HOX32, and HOX33), heat stress factor gene (heat-stress-transcription-factor-A-3) (Table S7), and NAC-domain containing proteins (NAC7, NAC21/22, NAC43, NAC79, and NAC92) were detected (Table S8). However, none of these proteins have been characterized in tef, and their role in plant growth and development remains unknown.

2.6 Gene ontology (GO) analysis

To predict the involvement of miRNAs targets in various processes, we performed GO analysis using the AgriGo website (http://bioinfo.cau.edu.cn/agriGO/). A total of 15,498 identified target genes were classified into three major categories; 3742 genes were grouped into the biological process, 886 genes were grouped into the cellular component, and 10,870 genes were grouped into the molecular functions categories (Figure 5). Of the genes grouped into the biological process category, the majority of them may be involved in metabolic process including macromolecule modification, phosphorous metabolism, metabolism of phosphate containing compounds, and protein modification processes including protein phosphorylation. From the cellular components category, the majority of target genes were associated with organelles, membrane, and the nucleus. In the molecular functions category, most of the miRNA target genes were associated with catalytic activity and substrate binding functions including heterocyclic compound binding, organic cyclic compound binding, small molecule binding, nucleotide binding, and nucleotide phosphate binding (Figure 5).

Furthermore, GO analysis of target genes of DEMs was performed for the pairwise comparisons (LCR vs. ConR, LCS vs. ConS, LCR vs. LCS, and ConR vs. ConS; Figure 6). For all comparisons, most of the miRNA target genes are involved in nucleic acid binding and/or DNA bining from the molecular functions category. Whereas in comparisons LCS versus ConS, LCR versus LCS, and ConR versus ConS, most target genes are also involved in trasport activitites. From the cellular compontent category, most target genes were associated with membranes for these comparisons (Figure 6). In the biological process category, there appears to be no similarity in the fuctons of the miRNA target genes between the four comparisons (LCR vs. ConR, LCS vs. ConS, LCR vs. LCS, and ConR vs. ConS).

4.1 Plant materials and growth conditions

E. tefaccession (PI-494307), previously selected for high seed Ca²⁺ content (Ligaba-Osena et al., 2021), was used in this study. A sample of 25 seeds was surface sterilized using 70% ethanol followed by 1% NaOCl solution containing .1% Tween-20 for 20 min. The seeds were then washed with sterile ultrapure or Milli-Q® (18.2 milliohms) water. Sterilized seeds were transferred onto moist filter papers and grown for 6 days. The seedlings were transferred to modified Hoagland solution containing [in mM]; KNO3[1.5], NH₄CO₃ [.5], NH₄H₂PO₄ [.5], MgSO₄.7H₂O [.25], and [in mM], KCl [12.5], Fe (III)-EDTA-2Na [.125], H₃BO₃ [6.25], MnSO_4.H₂O [.5], ZnSO4.7H₂O [.5], CuSO_4.5H₂O [.125], Na₆Mo₇O₂₄ [.025]. NH₄CO₃ is a substitute of Ca (NO₃)₂·4H₂O for the low Ca²⁺ treatment (10 μM) while for control plants, Ca (NO₃)₂·4H₂O [1 mM] was applied. Four biological replicates were used for each root and shoots of control (control 1 to 4) and low calcium treated (low calcium 1 to 4) plants. The pH of the hydroponic solution was adjusted to 5.8 using 1 N KOH solution. The seedlings in hydroponics were transferred to growth chamber (28°C day and 25°C night temperatures, and 12-h day and night cycles). The nutrient solution was renewed every 4 days and plants were grown for 4 weeks until root and shoot tissues were collected for RNA isolation.

4.2 RNA isolation, RNA library generation, and sequencing

Root and shoot samples were ground into powder under liquid nitrogen using a mortar and pestle. Total RNA was isolated using the GeneJET RNA purification kit following manufacturer's procedure (Fisher Scientific). Small RNA libraries for miRNA-seq were prepared using the NEBNext® Multiplex small RNA library preparation set according to user instructions for the Illumina (E7300 and E7580, NEB). Sequencing was performed using Illumina Hiseq2500 platform using 50-nt read length with single end sequencing protocol (Saus et al., 2018).

4.3 Sequence analysis and identification of novel and conserved miRNAs

The raw reads were filtered for adapters, ambiguous residues, and low-quality reads prior to sRNA analysis using a Perl script Cutadapt v2.10 (Martin, 2011); the parameters were cutadapt -a AGATCGG -q 30 --discard-untrimmed –o. small RNAs of 17–35-nt reads were counted. For novel miRNA prediction, we selected sRNA reads with a minimum raw read count of 10 per library and then combined these into one sRNA library for miRNA prediction (Jin & Wu, 2015). These reads were mapped to the Tef genomic sequence (Pacbio Eragrostis_tef_tef-ft-CDS-gid-50954, https://genomevolution.org/coge/GenomeInfo.pl?gid=50954) using Bowtie 2 software (Langmead & Salzberg, 2012) with two mismatches at maximum. With one end attached 20 nt away from the mapped sRNA site, sequences in the range of 120 to 360 nt each with the extension of 20 nt were collected that covered the region of sRNA. Under similar conditions used by Meyers et al. (2008) and Thakur et al. (2011), at the sRNA location the stem loop structure having three or less gaps with ≤8 bases, and the miRNA-miRNA duplexes mapped to the precursor locus with more than 75% of reads were considered the candidate miRNA precursors. The miRNAs identified with no mismatch to any known miRNA in the miRBase dataset (miRBase, 21.0) were classified as known miRNAs while the remaining miRNAs were classified as novel miRNAs.

4.4 miRNA target identification

For miRNA target prediction, the psRNA Target software (http://plantgrn.noble.org/psRNATarget/) was used with its default parameters and published tef transcriptome (VanBuren et al., 2020). During the result filtration, only those with expectation scores from 0 to 3.5 were included. Genes targeted by differently regulated miRNAs were determined using psRNATarget (a plant-based miRNA target analysis server) (Dai & Zhao, 2011). The psRNATarget site was determined using default parameters to scan the tef transcriptome assembled by VanBuren et al. (2020) for differentially regulated miRNAs in tef. These targets were then utilized in PageMan (Usadel et al., 2006) to uncover functional ontologies that were over- and under-represented. Visualization was done using MapMan (Thimm et al., 2004). The SeqTar method (Zheng et al., 2012) was used to predict miRNA targets. Targets with less than or equal to four mismatches were considered for further investigation in the case of conserved miRNAs. Only targets with at least one valid read and fewer than four mismatches were used for novel miRNAs.

4.5 miRNA target annotation and GO analysis

miRNA target genes were predicted using the online database psRNA Target Server (http://biocomp5.noble.org/psRNATarget/) (Dai & Zhao, 2011) by using the default parameters. Function annotations and analysis were further performed by the AgriGO (agriGO: GO Analysis Toolkit and Database for Agricultural Community (cau.edu.cn)). After target predictions, all the targets of novel and conserved miRNAs were processed via the SEA tool of agriGO (an online toolkit version 1.2 for the GO analysis) (Tian et al., 2017). The AgriGO toolkit was used to assess the enriched GO terms in our dataset in relation to total annotated genes using Fisher's exact test at a significant P value of .05. The result of the software defines three GO categorization categories: biological processes, cellular components and molecular functions.

4.6 Analysis of miRNA expression patterns

Differential accumulation of miRNA was determined using the DeSeq2 package using the shrinkage estimation of fold change and dispersion for improving estimation interpretability as well as stability (Love et al., 2014). R statistical software packages ggplot2 (Wickham, 2011) and gplots (Warnes et al., 2009) were used for all the plot presentations.