Loss of function at RAE2, a previously unidentified EPFL, is required for awnlessness in cultivated Asian rice

Significance

This study investigates a previously unidentified cysteine-rich peptide (CRP). CRPs have diverse roles in plant growth and development, such as control of stomata density and guidance of pollen-tube elongation. Despite numerous studies on CRPs in Arabidopsis thaliana, there are still many peptides with unknown function. We identify a previously unidentified rice CRP named Regulator of Awn Elongation 2 (RAE2) and show that it is cleaved specifically in the spikelet to promote awn elongation. We demonstrate that RAE2 was a target of selection during domestication, contributing to loss of awns in Asian but not African rice. The discovery of RAE2 simultaneously deepens our understanding of plant developmental pathways and lends insight into the complex processes underlying cereal domestication.

Abstract

Domestication of crops based on artificial selection has contributed numerous beneficial traits for agriculture. Wild characteristics such as red pericarp and seed shattering were lost in both Asian (Oryza sativa) and African (Oryza glaberrima) cultivated rice species as a result of human selection on common genes. Awnedness, in contrast, is a trait that has been lost in both cultivated species due to selection on different sets of genes. In a previous report, we revealed that at least three loci regulate awn development in rice; however, the molecular mechanism underlying awnlessness remains unknown. Here we isolate and characterize a previously unidentified EPIDERMAL PATTERNING FACTOR-LIKE (EPFL) family member named REGULATOR OF AWN ELONGATION 2 (RAE2) and identify one of its requisite processing enzymes, SUBTILISIN-LIKE PROTEASE 1 (SLP1). The RAE2 precursor is specifically cleaved by SLP1 in the rice spikelet, where the mature RAE2 peptide subsequently induces awn elongation. Analysis of RAE2 sequence diversity identified a highly variable GC-rich region harboring multiple independent mutations underlying protein-length variation that disrupt the function of the RAE2 protein and condition the awnless phenotype in Asian rice. Cultivated African rice, on the other hand, retained the functional RAE2 allele despite its awnless phenotype. Our findings illuminate the molecular function of RAE2 in awn development and shed light on the independent domestication histories of Asian and African cultivated rice.

Since the dawn of agriculture, cultivated plants have contributed to human health and prosperity. Among the three major cereals, wheat (Triticum aestivum), maize (Zea mays), and rice, only rice has been independently domesticated twice (1–3). There are two cultivated rice species, Oryza sativa derived from Oryza rufipogon in Asia and Oryza glaberrima derived from Oryza barthii in Africa (4, 5). Although the habitats of O. sativa and O. glaberrima are geographically isolated, the two species share a suite of domestication-related traits, such as decreased seed shattering, upright growth habit, and awnless seeds (6–9). This suggests that such characteristics significantly contribute to promoting agriculture. There is evidence for artificial selection at a genome-wide scale in O. sativa and O. glaberrima (10), and several domestication traits are regulated by the same genes in both species (11–14). However, convergent evolution via different genes is another scenario by which shared phenotypes may arise from geographically independent domestication processes. Recently, we reported that the genetic regulation of awnlessness differs in the two species of cultivated rice (15). Awns are sharp spine-like structures extending from the lemmas of rice florets (Fig. 1A). Their bristle-like architecture enhances seed dispersal and provides protection against seed predation (16). Thus, long awns are considered important for habitat expansion and survival of wild rice. Under domestication, the awnless phenotype has been selected to facilitate planting, harvesting, and storage of seeds (17).

Fig. 1.

Identification and functional characterization of RAE2. (A) The awn is a spine-like extension of the rice lemma. (Left) Awned rice seed anatomy. (Right) Panicles of the awned chromosome segment substitution line GLSL25. The red arrowhead points to the awn. (B–D) Seed phenotypes and graphical genotypes of Koshihikari (O. sativa ssp. japonica; yellow) (B), IRGC104038 (O. glaberrima; blue) (C), and GLSL25 (D). (E–G) Evaluation of transgenic plants with plasmid vector pGWB501 (vector control; V.C.) and RAE2 gene (pRAE2::RAE2): seed phenotype (E), frequency of awned seeds per panicle (F), and awn length (G). No visible awn was observed in pGWB501 (V.C.), indicated as “n.d.” (not detected). (H–J) Evaluation of vector control [pANDA (V.C.)] and RNAi line (RAE2-RNAi): seed phenotype (H), frequency of awned seeds per panicle (I), and awn length (J). (K) rae2/OsEPFL1 amino acid structure of Koshihikari (O. sativa ssp. japonica) and RAE2 of IRGC104038 (O. glaberrima). Yellow triangles indicate insertions. Each colored box represents a peptide region. ma, mature peptide (gray or red); pro, propeptide (green); sp, signal peptide (blue). Red bars indicate cysteine (C) residues. (Scale bars, 1 cm.) The statistical significance was at *P < 0.05 based on a two-tailed Student’s t test. Error bars represent SD of the mean.

To date, two genes have been identified that regulate awn development in rice and show evidence of selection during domestication: a basic helix-loop-helix transcription factor called AWN1 (An-1) and a cytokinin-activating enzyme named LONG AND BARBED AWN1 (LABA1), both residing on chromosome 4 (18, 19). These reports focus on allelic variation solely within Asian rice, and it remains unclear whether mutations in these same genes might be responsible for the awnless phenotype in African cultivated rice. Using a library of chromosome segment substitution lines (CSSLs) derived from a cross between two awnless parents, O. sativa ssp. japonica and O. glaberrima (20) (Fig. 1 B and C), two of the lines were observed to have long awns: Line GLSL13 carried an O. glaberrima introgression on chromosome 4 that included Regulator of Awn Elongation 1 (RAE1), which corresponds to An-1, and line GLSL25 carried an O. glaberrima introgression on chromosome 8 containing RAE2 (Fig. 1D) (15). RAE1 and RAE2 have been shown to have redundant roles based on the fact that a functional allele at one or both loci confers awns in O. sativa. Although quantitative trait loci associated with awn elongation around the RAE2 region have been reported, the responsible gene has not been identified (5, 21, 22).

Here, we show that RAE2 is an EPIDERMAL PATTERNING FACTOR-LIKE 1 (EPFL1) gene and demonstrate that the RAE2 allele found in O. glaberrima induces awn development in O. sativa. We also identify SUBTILISIN-LIKE PROTEASE 1 (SLP1) as a protease required for RAE2 processing in the rice spikelet. Finally, through RAE2 sequence analysis in a diverse collection of Asian and African rice, we reveal a highly variable GC-rich region that harbors multiple independent mutations underlying protein-length variation. We show that the array of differences found in this GC-rich region gives rise to a suite of dysfunctional RAE2 variants that were selected during domestication and condition the awnless phenotype in Asian rice.

Results

O. sativa Has a Dysfunctional RAE2 Protein, Whereas O. glaberrima Carries a Functional RAE2.

To identify the functional RAE2 allele from O. glaberrima that induces awn elongation in the genetic background of O. sativa ssp. japonica, cv. Koshihikari, we undertook positional cloning in a mapping population derived from GLSL25 × Koshihikari (SI Appendix, Fig. S1A). Genetic linkage analysis using ∼8,000 individuals from filial generation 2 (F₂) delimited the candidate region on chromosome 8 to about 80 kb, which encompassed 12 predicted genes (SI Appendix, Fig. S1 B and C). We screened an O. glaberrima bacterial artificial chromosome (BAC) library and identified a clone, Ogla0006B21, that encompassed this candidate region (SI Appendix, Fig. S1C). Five subclones of Ogla0006B21 were systematically introduced into Taichung 65 (T65), an awnless O. sativa ssp. japonica cultivar (SI Appendix, Fig. S1D). Two of the subclones (33 and 89) recovered awns in the transgenic lines (SI Appendix, Fig. S1 E–G and Table S1). A single open reading frame, Os08g0485500, was shared between subclones 33 and 89 (SI Appendix, Fig. S1 C and D). When a construct containing only Os08g0485500 (O. glaberrima allele) was introduced into Nipponbare (awnless O. sativa ssp. japonica), the resultant transgenic plants produced awns of comparable frequency and length to the awns of GLSL25 (Fig. 1 E–G and SI Appendix, Table S1). RNA interference (RNAi) experiments using a construct harboring the Os08g0485500 3′ UTR transformed into GIL116, an awned introgression line carrying the chromosome 8 segment derived from O. glaberrima in the T65 genetic background (SI Appendix, Fig. S2), generated lines with awn lengths significantly shorter than those of the vector controls, whereas awn frequencies were not different (Fig. 1 H–J and SI Appendix, Table S1). Together, these results indicate that Os08g0485500 is RAE2, and that this gene acts to regulate awn elongation.

RAE2 Encodes an EPIDERMAL PATTERNING FACTOR-LIKE 1 PROTEIN in Rice.

Based on amino acid sequence analysis, RAE2 is predicted to encode an EPIDERMAL PATTERNING FACTOR-LIKE 1 (EPFL1) protein (23). This protein is a member of the EPF/EPFL family, a group of plant-specific secreted peptides that regulates a range of developmental processes (24–28). In Arabidopsis thaliana, the most extensively studied EPFs/EPFLs are Stomagen (also known as AtEPFL9) and its competitive factors, EPF1 and EPF2. In contrast, there are no reports about this family in rice. Members of this peptide family share a conserved cysteine-rich region that mediates formation of disulfide bonds essential for functional conformation as a ligand (29).

Using cysteine-rich region sequences, we evaluated the phylogenetic relationship of RAE2 with other members of the EPF/EPFL family from A. thaliana and several awned grass species (Brachypodium distachyon, Triticum urartu, and Hordeum vulgare) (SI Appendix, Fig. S3A). Phylogenetic analysis revealed that RAE2 is classified into the AtEPFL1-3 clade, a group of genes with unknown function. Comparison of RAE2 and rae2 amino acid sequences with other EPFL relatives showed that all sequences except rae2 contained the six cysteine residues typical of EPFL peptides (SI Appendix, Fig. S3B).

Three-dimensional structure modeling revealed clear structural similarity between RAE2 and Stomagen (SI Appendix, Fig. S4). The loss of cysteine residues has been reported to cause dysfunction of Stomagen due to the nonformation of a critical scaffold mediated by disulfide bonds (30). Comparative sequence analysis of RAE2 and rae2 from IRGC104038 (O. glaberrima) and Koshihikari (O. sativa ssp. japonica), respectively, revealed several SNPs and insertions in the promoter and coding regions (SI Appendix, Fig. S5A). A 2-bp insertion in the second exon of rae2 gave rise to a truncated protein through a frameshift mutation (Fig. 1K and SI Appendix, Fig. S5 B and C), and was hypothesized to be the causal mutation leading to dysfunctional conformation.

Coordination of RAE2 Expression with Awn Development.

We next analyzed the expression of RAE2 throughout the rice plant to determine its correlation with awn development. Plant-wide, RAE2 expression was about 10-fold higher in young panicles than in other organs (Fig. 2A). Within the young panicle, expression of RAE2 was significantly greater than that of rae2. We additionally observed awn development in Koshihikari and GLSL25 by scanning electron microscopy (SEM). SEM observations showed that lemma and palea morphology did not differ between the two lines until the spikelet development stage 7 (Sp7) stage (Fig. 2 B and F). The awn primordium protruded at the distal end of the lemma only in GLSL25 at the beginning of Sp8 (Fig. 2 C–E and G–I). In situ hybridization showed similar expression patterns for both RAE2 and rae2 from Sp4 through Sp7 (Fig. 2 J–L and O–Q). RAE2 transcripts, however, exhibited prolonged expression compared with rae2 transcripts in subsequent stages (Fig. 2 M and R). In GLSL25, especially high expression was observed in the vascular bundles of the awn primordium compared with Koshihikari (SI Appendix, Fig. S6). Together, these observations provide evidence of the importance of spatiotemporal regulation of RAE2 expression for awn elongation.

Fig. 2.

RAE2 expression pattern and correlation with awn development. (A) Quantitative RT-PCR showing RAE2 mRNA levels in different organs of Koshihikari and GLSL25 (IN, internode; LB, leaf blade; LS, leaf sheath; RO, root; PA, young panicle). OsUBI was used as an internal control. Error bars represent SD of the mean. (B–I) SEM images of spikelets at different developmental stages in Koshihikari (B–E) and GLSL25 (F–I). Developmental stages are classified into Sp7 (B and F), Sp8 (C and G), and post Sp8 (D, E, H, and I) according to Oryzabase classification (www.shigen.nig.ac.jp/rice/oryzabase/devstageineachorgan/list). (Scale bars, 50 µm.) Red arrowheads point to the awn. (J–S) In situ hybridization using antisense probes of rae2 (J–M), RAE2 (O–R), and sense probes (N and S) during spikelet development in Koshihikari and GLSL25. Blue arrowheads indicate the tips of paleae; red arrowheads indicate the tips of lemmas; green asterisks show anthers (an, anther; le, lemma; pa, palea). [Scale bars, 50 µm (J–L and O–Q) and 100 µm (M, N, R, and S).]

Protein and Allelic Diversity of RAE2 and Geographical Distribution.

Characterizing the diversity and frequency of nucleotide polymorphisms in domestication genes across divergent populations gives insight into the evolutionary history of rice. To understand RAE2 variation across diverse accessions, we sequenced RAE2 across a panel of 130 accessions made up of Asian (cultivated, n = 42; wild, n = 65) and African rice (cultivated, n = 12; wild, n = 11) (listed in SI Appendix, Table S2). The 2-bp insertion in the second exon of rae2 corresponding to the functional mutation in Koshihikari (Fig. 1K and SI Appendix, Fig. S5) occurred in a highly variable GC-rich repeat region in our diversity panel (SI Appendix, Fig. S7A). We observed seven different length polymorphisms in this region (SI Appendix, Table S3). Among the seven variants, four translated into functional RAE2 proteins (i.e., six cysteine residues), whereas the other three gave rise to putatively dysfunctional RAE2 proteins: either a truncated, short protein in cv. Koshihikari (four cysteines) or a long protein (seven cysteines). We named these translated products according to the number of cysteine residues harbored in the cysteine-rich region and grouped them into three protein-length classes: 4C/short, 6C/medium, and 7C/long (SI Appendix, Fig. S7B and Table S3). These results demonstrated that nucleotide variations in the GC-rich repeat region are responsible for RAE2 protein length and affect protein function. In addition, three singleton variants of RAE2 were identified that independently gave rise to translated products with five cysteine residues, resulting from polymorphisms that occurred outside the GC-rich repeat region (listed in SI Appendix, Table S4). All of these were predicted to cause medium-length proteins but are compositionally divergent from the 6C/medium peptide class (SI Appendix, Fig. S7B).

RAE2 Loss-of-Function Alleles Have Been Selected in Asian but Not African Rice.

To understand the functionality of the RAE2 protein variants in relation to awn development, we evaluated overexpression lines of each variant. Only the RAE2-6C type exhibited the awned phenotype, whereas the other three lines did not show awns, regardless of expression level (Fig. 3A and SI Appendix, Fig. S8). This indicates that the RAE2-6C type is functional for awn elongation but the other three types of protein (RAE2-4C, RAE2-5C, and RAE2-7C) are not functional. We also evaluated RAE2 variants in two species of Asian wild rice, Oryza nivara (accession W0054), which is an annual type of O. rufipogon carrying the RAE2-6C type, and O. rufipogon (accession W0106), carrying the RAE2-7C type. These wild accessions were used as donors for two previously developed CSSL populations (WBSL and RSL, respectively) in the background of cv. Koshihikari (SI Appendix, Fig. S9A) (15). Phenotyping the two lines from these CSSL populations that harbored the RAE2 locus on chromosome 8 derived from O. nivara (WBSL18) or O. rufipogon (RSL23) demonstrated that line WBSL18 formed awns but RSL23 did not (SI Appendix, Fig. S9B). Consistent with our results from overexpression analysis of RAE2 alleles, O. nivara harbored the functional 6C/medium RAE2, whereas O. rufipogon possessed the dysfunctional 7C/long RAE2 type (SI Appendix, Fig. S9C).

Fig. 3.

Distribution of RAE2 protein variants across diverse rice accessions. (A) Awned phenotypes of overexpression lines carrying different RAE2 types (4C, 5C, 6C, 7C). pCAMBIA1380 was used as the vector control. (Scale bar, 1 cm.) (B) Distribution of the four RAE2 protein variants within O. barthii (i; n = 11), O. glaberrima (ii; n = 12), O. rufipogon/O. nivara (iii; n = 65), and O. sativa (iv; n = 42). (C) Geographical distribution of RAE2 protein variants found across these same 130 accession [107 Asian rice accessions (circles) and 23 African rice accession (triangles)]. Symbol sizes are proportional to the number of accessions and are indicated by the numbers in the rectangle. Colors represent each protein variant: yellow, 7C/long; green, 6C/medium; blue, 5C/medium; and red, 4C/short. (D) Awned phenotypes across the four RAE2 protein variants. Numbers of awned (gray bars) and awnless (white bars) accessions for O. barthii (i), O. glaberrima (ii), O. rufipogon/O. nivara (iii), and O. sativa (iv). (E, Upper) Nucleotide diversity of O. sativa individuals (n = 67) relative to nucleotide diversity of O. rufipogon/O. nivara individuals (n = 65) across chromosome 8. The gray box represents reducing relative diversity surrounding RAE2 (green line), consistent with a selective sweep. (E, Lower) Genetic distance between O. sativa and O. rufipogon/O. nivara for all RAE2 types (blue) and dysfunctional ones (red; 4C and 7C). Decreased distance in the dysfunctional class relative to the distance in the “all” class in a 1.5-Mb region surrounding RAE2 (gray box).

The geographic distribution of the 130 diverse rice accessions reflects the fact that Asian rice varieties are widely planted around the world, whereas African rice is confined to Africa (Fig. 3B). Interestingly, all cultivated African rice varieties (O. glaberrima) and the wild ancestor, O. barthii, were found to carry the RAE2-6C type, despite the fact that O. glaberrima does not have awns and O. barthii possesses awns (Fig. 3 C, i and ii and D, i and ii). These results support our previous report that a different gene(s), other than RAE2, is responsible for the awnless phenotype in O. glaberrima (15). The gene reported to cause awnlessness in African domesticated rice is RAE3.

In Asia, dysfunctional RAE2 protein types were present in 32% of O. rufipogon and O. nivara wild accessions (Fig. 3 B, iii), despite the fact that almost all individuals possessed awns (Fig. 3 D, iii). This distinguished the RAE2 gene from previously reported awn domestication genes, such as An-1/RAE1 or LABA1, which were documented to persist as functional alleles in the vast majority of wild Asian rice populations (18, 19). Although O. sativa was nearly fixed (93%) for dysfunctional alleles at the RAE2 locus and most accessions were awnless (Fig. 3 C, iv and D, iv), the five subpopulations of O. sativa harbored different frequencies of RAE2 protein classes (SI Appendix, Table S5). Upon examination, it was found that Asian rice cultivars that had the awned phenotype always carried a functional allele at An-1/RAE1 and/or RAE2.

To test for evidence of a selective sweep in the region of RAE2, the nucleotide diversity (π) of O. sativa (n = 67) relative to the π of O. rufipogon (n = 65) was estimated using data from 100-SNP sliding windows across chromosome 8. A drastic decrease was observed in the ratio π O. sativa/π O. rufipogon across a 1.5-Mb region flanking RAE2 (Fig. 3E, Upper), consistent with a selective sweep in O. sativa. A second decrease in the ratio was observed 0.5 Mb downstream of RAE2, suggesting the possibility of another target of selection nearby. Using the Kolmogorov–Smirnov test, we tested whether the distribution of this diversity ratio from the RAE2 region differed from the rest of the chromosome. It was indeed highly unusual (P < 2.2 × 10⁻¹⁶). Analyzing the genetic distance (d) between O. sativa and O. rufipogon using the same 100-SNP windows across chromosome 8 revealed that the dysfunctional protein types observed in wild accessions were likely the result of recent back-introgression from O. sativa to O. rufipogon or O. nivara. This was supported by a decrease in d across the 1.5-Mb region surrounding RAE2 in the dysfunctional classes (4C and 7C) relative to d in all functional (6C) RAE2 types (Fig. 3E, Lower). This result is consistent with recent gene flow back from cultivated to wild Asian rice (31).

RAE2 Is Cleaved Specifically in the Spikelet.

The EPFL family of peptides typically requires posttranslational cleavage of a propeptide to become a mature peptide (SI Appendix, Fig. S10A) (32, 33). To test whether RAE2 is cleaved, we generated transgenic plants carrying pACT::RAE2-3×FLAG. Immunoblot analysis demonstrated that RAE2 was cleaved into an ∼11-kDa peptide only in the spikelet but not in the other organs (Fig. 4 A and B and SI Appendix, Fig. S10B). To confirm this spikelet-specific cleavage of RAE2, we conducted an in vitro processing assay by mixing the recombinant RAE2 propeptide-fused 3×FLAG tag (RAE2-pro) with each of the protein extracts from various organs (callus, stem, leaf, and spikelet). The result of immunoblot analysis using the FLAG antibody showed that only spikelet extract could cleave RAE2-pro (Fig. 4C and SI Appendix, Fig. S10C). In addition, we proved that the ∼11-kDa band contained the C-terminal region of RAE2 by using an anti-RAE2 antibody made at the end of the RAE2 C terminus (149-RDRLFDP-125) (SI Appendix, Fig. S10D). Furthermore, the cleavage of RAE2-pro was inhibited by a protease inhibitor mixture, Complete (SI Appendix, Fig. S10E). Together, these data suggest that the RAE2-targeted protease(s) specifically functions in rice spikelets.

Fig. 4.

RAE2 maturation caused by cleavage with SLP1 protease at the spikelet. (A) Immunoblot analysis of RAE2-3×FLAG in transgenic plants with an anti-FLAG antibody. The gray line indicates erased space between the callus and stem lane, although all samples were applied on the same membrane. (B) The expected size of the RAE2-3×FLAG peptide after cleavage in transgenic plants of overexpression constructs: signal peptide (blue), propeptide (green), and mature peptide (pink). (C) In vitro processing assay of recombinant RAE2 peptide incubated with plant extracts of Koshihikari or buffer (mock). The ∼30-kD band is a tag-fused recombinant RAE2 propeptide (indicated by −tag+pro). Asterisks indicate nonspecific bands; the red arrowhead indicates the expected mature RAE2-3×FLAG peptide (∼11 kDa; indicated by −ma). (D) MS ion spectrum for the synthetic peptide (52-AGEEEKVRLGSSPPSCYSK-70), which was cleaved at the position between P65 and S66, indicated by the red line. The full length of the synthetic peptide was cleaved on the other site (+, 54-EEEKVRLGSSPPSCYSK-70; ++, 54-EEEKVRLGSSPP-65). (E) In vitro processing assay of a series of alanine-substituted recombinant RAE2 peptides (mu-RAE2) using the spikelet extract of Koshihikari or buffer (mock). The table (Lower) shows the amino acid sequence around the predicted cleavage point. Other descriptions are the same as in C. (F) Predicted sequence of RAE2, which encodes a 125-amino acid peptide composed of a signal peptide (sp; blue), a propeptide (pro; green), and a mature peptide (ma; pink). The dotted and solid arrows indicate the cleavage sites of Stomagen and EPF2, respectively. See also SI Appendix, Figs. S10 and S12.

Identification of RAE2-Targeted Protease.

To find the candidate protease(s), we compared expression patterns of 63 members of the rice subtilisin-like protease (SLP) family (listed in SI Appendix, Table S6). This protease family has been shown to cleave EPF/EPFL peptides in A. thaliana [SDD1 (34); CRSP (35)]. Among the 63 proteases examined, we identified one promising candidate, Os01g0702300, which was specifically expressed around the young inflorescence according to RiceXPro (ricexpro.dna.affrc.go.jp) (SI Appendix, Fig. S11A). We confirmed this expression pattern experimentally (SI Appendix, Fig. S11 B and C) and named this gene SLP1. An analysis of phylogenetic relationships classified this gene in the same clade as CRSP (36), supporting the hypothesis that SLP1 is the most promising candidate gene targeting RAE2. To determine whether the RAE2 protein could be cleaved by SLP1, we performed an in vitro processing assay using a synthetic RAE2 short peptide (synRAE2) spanning the predicted cleavage site (52-AGEEEKVRLGSSPPSCYSK-70) based on Stomagen (27) and an EPF2 (35) with an in vitro SLP1 protein. As a result, three cleaved peptides were detected (Fig. 4D and SI Appendix, Fig. S12A). Cleavage of synRAE2 was documented between amino acid positions G53 and E54, P65 and S66, or both sites by SLP1. This suggested that SLP1 cleaves two positions of this peptide; however, the shorter version would work as a ligand and, according to the Stomagen and EPF2 reports, the site between P65 and S66 is most appropriate as the cleavage site. In addition, we tested whether SLP1 digested a series of mutated synRAE2 peptides with a single amino acid substitution near the predicted cleavage site (mu-synRAE2). Cleavage of mu-synRAE2 (P64G, P65G P65A, and S66D) by in vitro-synthesized SLP1 was inhibited (SI Appendix, Fig. S12B). Furthermore, the alanine substitution closer to the cleavage point repressed RAE2 cleavage using mutated RAE2-pro (mu-RAE2 1–4) (Fig. 4E). These results suggested that SLP1 cleaves RAE2 between amino acid positions P65 and S66, and that cleavage occurs specifically in the spikelet (Fig. 4F).

Discussion

Our study identifies RAE2 as Os08g0485500, a previously unidentified EPFL gene that is preferentially expressed during early panicle development to promote awn elongation. RAE2 harbors a conserved six-cysteine peptide region characteristic of the EPF/EPFL family, whereas rae2 alleles, which are found in most awnless rice accessions, have alternative numbers of cysteine residues due to frameshift mutations. Ohki et al. reported that the consensus scaffold stabilized by three pairs of disulfide bonds is structurally required for EPF/EPFL activity in Stomagen (30). They showed that substituting cysteine residues with serine residues in Stomagen led to incorrect disulfide bonds, which resulted in improper conformation. It is likely that the lack of disulfide bonds impeded active conformation. According to this report, rae2 variants without two cysteine residues, or with extra cysteine residues in improper locations, also failed to achieve proper conformation of the RAE2 protein and were not able to promote awn elongation.

We also showed that SLP1 cleaves RAE2 from a propeptide to a mature peptide. It is known that EPF/EPFL peptides are processed into mature peptides, which then bind to receptors such as the ERECTA family of receptor kinases (37, 38). We hypothesize that RAE2 may mediate awn elongation through binding to an unidentified rice ERECTA kinase(s). Improper scaffolding of the mature RAE2 peptide resulting from mutations may hinder binding by this receptor(s) and lead to an awnless phenotype. At this moment, we do not know whether the cleavage of RAE2 by SLP1 is essential for awn elongation. Further evaluation of transgenic plants transduced with RAE2 constructs with and without mutations at the cleavage site is necessary to fully elucidate the role of RAE2 cleavage in the spikelet. The observation that transgenic Nipponbare plants transduced with O. glaberrima RAE2 alleles were able to produce awns demonstrates that cv. Nipponbare retains a functional RAE2 receptor(s). As for the cleavage of RAE2, an in vitro processing assay showed that O. sativa also possesses a functional SLP1, and sequence comparisons of SLP1 alleles in O. sativa and O. glaberrima identified no deleterious frameshift mutations in the coding region of either (SI Appendix, Fig. S13). Taken together, the conservation of these downstream components suggests that they are necessary to mediate the interactions with other EPF/EPFL peptides (other than RAE2) and may interact with important, but as yet unidentified, functions in the rice plant.

Because vascular bundles penetrate the center of awns, the RAE2 gene may function to promote the proliferation of vasculature cells necessary for awn elongation. A similar phenomenon has been reported in Arabidopsis; AtEPFL4 and AtEPFL6 coordinate development of inflorescence architecture (28). In a previous mutant study, DL and OsETT2 were reported to be involved in rice awn development (39). DL is a YABBY gene, essential for forming leaf midribs by promoting cell proliferation, and OsETT2 is an auxin-responsive factor. Future identification of the receptor(s) of RAE2 and downstream factors affecting awn development will reveal the relationship among these components.

Sequence comparison of Asian and African rice suggested that the critical mutations occurred in the GC-rich repeat region, and that speciation led to RAE2 diversification. We infer that there is selective pressure to conserve RAE2 functionality in wild Asian rice, as its reading frame is preserved in most individuals despite some nucleotide-level variation. Because other awn genes (e.g., RAE1/An-1 or LABA1) can mask RAE2’s loss of function, this raises the possibility that RAE2 is conserved in the wild due to pleiotropic effects on other fitness phenotypes. On the other hand, although African rice maintained a functional RAE2 protein, we hypothesize that a mutation(s) in a different locus, designated RAE3, represses the awned phenotype in O. glaberrima (15). The story of RAE2 is part of a larger narrative about human selection. As a trait that has been targeted for selection multiple times via multiple genes, the awn serves as a unique lens through which to study the divergent domestication history of Asian and African rice.

Both molecular and genetic evidence demonstrate that RAE2 positively regulates awn elongation. Natural variation in RAE2 plays crucial roles in the domestication and evolution of rice morphology. Identifying the receptor for RAE2 and investigating the relationship among An-1/RAE1, LABA1, RAE2, and RAE3 will be important to further understand the molecular basis of the regulation of awn development in rice.