A COBRA‐like protein ACL83 acts downstream of transcription factor ROC5 to control abaxial leaf curling in rice

SUMMARY

Leaf rolling is considered an important agronomic trait in rice (Oryza sativa). Moderate leaf rolling contributes to erect leaves and a compact plant architecture, thereby improving photosynthetic efficiency and grain yield. However, the molecular mechanisms regulating leaf rolling remain incompletely understood. In this study, we isolated an acl83 mutant in indica rice, which exhibits abaxially curled leaves due to increased area of bulliform cells. Map‐based cloning and MutMap analysis revealed the candidate gene LOC_Os05g32110 (ACL83) encoding a COBRA‐like protein. Complementation and CRISPR/Cas9 knockout assays confirmed that ACL83 is responsible for the mutant phenotype of acl83. ACL83 was constitutively expressed, and its encoded protein was localized to the plasma membrane. Furthermore, the transcription factor ROC5 directly bound to the promoter of ACL83 to activate ACL83 expression. Meanwhile, the acl83/roc5 double mutant exhibited an abaxially curled leaf phenotype similar to that of the acl83 single mutant, suggesting that ACL83 is a downstream target gene of ROC5. Collectively, our findings revealed a distinct ROC5‐ACL83 genetic pathway in which the COBRA‐like protein ACL83 functions downstream of ROC5 to control abaxial leaf rolling in rice by regulating cell expansion. This work provides insight into the role of COBRA‐like proteins in the regulation of leaf morphogenesis.

Significance Statement

ACL83 encodes a COBRA‐like protein and acts as a downstream target gene of the HD‐ZIP IV transcription factor ROC5 to control abaxial leaf rolling in rice by regulating cell expansion. This work provides insight into the role of COBRA‐like proteins in the regulation of leaf morphogenesis.

INTRODUCTION

Rice (Oryza sativa) is a staple food for a large proportion of the global population (Zhang et al., 2024), and improving grain yield has long been a central objective of rice research (Liu et al., 2022; Yuan et al., 2022). Leaf functions, including photosynthesis, respiration, and transpiration, depend on leaf shape or three‐dimensional architecture (Govaerts et al., 1996). In rice, moderate leaf rolling is regarded as a desirable agronomic trait, which contributes to yield improvement by reshaping plant architecture, enhancing photosynthetic efficiency, reducing transpirational water loss, delaying leaf senescence, and optimizing canopy light interception, particularly during the later stage of plant growth (O'Toole & Cruz, 1980; Richards et al., 2002). Therefore, isolating mutants with moderately rolled leaves and identifying genes responsible for leaf rolling will facilitate the development of super‐high‐yielding rice varieties with an ideal plant architecture (Liu et al., 2016).

Bulliform cells (BCs) are widely regarded as playing a key role in the regulation of rice leaf rolling (Zou et al., 2011). These specialized large parenchymatous cells occur in groups between adjacent vascular bundles on the adaxial surface of rice leaves, and possess a large cell volume, highly vacuolated cytoplasm, and thin cell walls (Wang et al., 2020). In rice, most leaf rolling phenotypes are attributable to abnormalities in the morphology, number, size, or spatial distribution of BCs on the adaxial surface of the leaf blade (Li, Zhang, et al., 2017). Nevertheless, it was found in some studies that sclerenchyma cells, epidermal cells, mesophyll cells, and cuticular wax all contribute to the regulation of leaf rolling. For example, SLL1 (shallot‐like 1) promotes the development of abaxial sclerenchyma cells in leaves, and its mutation results in defective sclerenchyma formation on the abaxial side and consequently extremely incurved leaves (Zhang et al., 2009). CFL1 (curly flag leaf 1) negatively regulates leaf cuticle development, thereby altering leaf morphology (Wu et al., 2011). OsCHR4 regulates cuticle wax formation, and its mutation results in narrower and rolled leaves with increased cuticular wax (Guo et al., 2019). In addition, several genes involved in secondary cell wall or cellulose biosynthesis also play important roles in determining leaf rolling, such as NRL1 (narrow and rolled leaf 1)/DNL1 (dwarf and narrowed leaf 1), RL14 (rolling‐leaf 14), SRL1 (semi‐rolled leaf 1)/CLD1 (curled leaf and dwarf 1), OsMYB103L (R2R3‐MYB transcription factor), OsSND2 (NAC family transcription factor), and PSL1 (photo‐sensitive leaf rolling 1) (Ding et al., 2015; Fang et al., 2012; Hu et al., 2010; Li, Zhang, et al., 2017; Xiang et al., 2012; Yang et al., 2014; Ye et al., 2018; Zhang et al., 2021).

Based on the direction of rolling, rice leaf rolling is generally classified into two types: adaxial rolling and abaxial rolling. So far, many genes have been identified to modulate adaxially curled leaves, such as OsAGO7 (Argonaute family), REL1 (rolled and erect leaf 1), and REL2, OsLBD3‐7 (lateral organ boundaries domain), OsHox32 (HD‐ZIP III family member), SRL2 (semi‐rolled leaf2), NRL2 (narrow and rolled leaf2), LRRK1 (leaf rolling receptor‐like cytoplasmic kinase 1), OsDRB2 (double‐stranded RNA‐binding2)/SRL10 (semi‐rolled leaf 10), OsLC1 (leaf‐curling mutant) (Chen et al., 2015; Li, Shen, et al., 2016; Li, Zou, et al., 2016; Liu et al., 2016; Shi et al., 2007; Wang et al., 2023; Wu et al., 2025; Yang et al., 2016; Yuan et al., 2022; Zhao et al., 2016; Zhou et al., 2018). Meanwhile, seven genes encoding transcription factors have been found to regulate abaxially curled leaves, including ACL1 (abaxially curled leaf1) and ACL2, OsZHD1 and OsZHD2, ROC1 (Rice Outermost Cell‐specific gene1), ROC5, and ROC8/Url1 (upward rolled leaf 1). ACL1 and ACL2 encode HD‐ZIP III transcription factors, OsZHD1 and OsZHD2 encode a zinc finger homeodomain class homeobox transcription factor, and ROC1, ROC5, and ROC8/Url1 encode homeodomain‐leucine zipper (HD‐ZIP IV) family transcription factors. Overexpression of ACL1, ACL2, OsZHD1, and OsZHD2 leads to abaxially curled leaves (Li et al., 2010; Xu et al., 2014). In contrast, overexpression of ROC1, ROC5, or ROC8 causes adaxially curled leaves, whereas their knockout mutants all show abaxially curled leaves (Fang et al., 2021; Sun et al., 2020; Tao et al., 2024; Xu et al., 2021; Zou et al., 2011). Nevertheless, the involvement of non‐transcription factor genes in abaxial leaf rolling remains poorly understood.

COBRA encodes a glycosylphosphatidylinositol (GPI)‐anchored protein and belongs to a multigene family (COBRA‐like family) in plants (Roudier et al., 2002). In rice, COBRA‐like proteins primarily function in the mechanical strength of tissues/organs and/or cell wall biosynthesis (Dai et al., 2011; Li et al., 2003), and have also been implicated in the regulation of β‐glucan levels in the endosperm cell wall and in stomatal development (Li et al., 2021; Midorikawa et al., 2019). However, their roles in the regulation of leaf rolling remain unclear.

In this study, we isolated a leaf‐rolling mutant, acl83, from indica rice, which exhibited an abaxially curled leaf phenotype due to increased area of BCs. Map‐based cloning and MutMap analysis suggested that the candidate gene ACL83 encodes a COBRA‐like protein. Complementation and CRISPR/Cas9 knockout assays confirmed that ACL83 is responsible for the mutant phenotype of acl83. This gene was constitutively expressed, and protein ACL83 was localized to the plasma membrane. Furthermore, our data demonstrated that ACL83 is a downstream target gene of the HD‐ZIP IV transcription factor ROC5. Therefore, our findings revealed a distinct ROC5‐ACL83 genetic pathway in which the COBRA‐like protein ACL83 functions downstream of ROC5 to control abaxial leaf rolling in rice by regulating cell expansion.

RESULTS

Phenotypic characterization of the acl83 mutant

We isolated the acl83 mutant from an ethyl methane sulfonate (EMS) mutagenesis library of indica rice restorer line 676R. The mutant did not exhibit a leaf‐rolling phenotype during early growth. However, from the late tillering stage onward, acl83 developed an abaxially curled leaf phenotype that persisted until maturity, with the flag leaf showing a milder rolling phenotype than the lower leaves (Figure 1A–C). At the grain‐filling stage, the leaf rolling indices (LRIs) of the flag leaf, second, and third leaves in acl83 plants were 35.4, 70.0, and 64.6%, respectively, whereas the corresponding wild‐type leaves showed slight adaxial rolling with LRI values of 11.9, 9.0, and 8.6%, respectively (Figure 1D). At maturity, the investigation of agronomic traits showed that plant height, number of productive panicles per plant, and number of spikelets per panicle of acl83 were significantly reduced by 19.7, 25.6, and 18.7%, respectively, compared with those of the wild type (WT) (Figure 1E–G). In contrast, seed setting rate and 1000‐grain weight showed no significant differences between acl83 and WT (Figure 1H,I).

Figure 1.

Phenotypes characterization of the acl83 mutant.

(A) Plant phenotypes of acl83 and its wild type (WT) at grain‐filling stage. Scale bar, 20 cm.

(B, C) Blade phenotypes and transverse sections of the second leaves from the top (the top 2nd leaves) at the grain‐filling stage, respectively. Ab, abaxial; Ad, adaxial. Scale bars, 10 cm in (B) and 0.5 cm in (C).

(D) Leaf rolling indices of the top three leaves (the flag, 2nd, and 3rd leaves from the top of the main stem of a single plant) at the grain‐filling stage. Ad and Ab indicate adaxial and abaxial rolling, respectively.

(E–I) Plant height, number of productive panicles per plant, number of spikelets per panicle, seed setting rate, and 1000‐grain weight at the maturity stage. Values are means ± SD of three biological replicates. Asterisks indicate significant differences between acl83 and WT by Student's t‐test (*P < 0.05, **P < 0.01).

To investigate the formation of leaf rolling, we performed paraffin cross‐sectioning on the middle region of the second leaf from the top (the top 2nd leaf) at the heading stage. As a result, the number of bulliform cells (BCs) in the midrib, large, and small veins did not differ significantly between acl83 and WT (Figure 2A,B). However, the total area of BCs adjacent to the midrib, large, and small veins in acl83 was significantly increased by 56.3, 91.7, and 110.0%, respectively, compared with that in WT (Figure 2A,C). This result suggested that the abaxial leaf rolling could be due to the increased size of BCs in the mutant.

Figure 2.

Comparison of bulliform cells between acl83 and WT.

(A) Paraffin sections of the middle region of the top 2nd leaves at the heading stage. Ab, abaxial; Ad, adaxial; LV, large vein; MV, midrib vein; SV, small vein. Red arrows indicate bulliform cells (BCs). Scale bars, 250 μm.

(B, C) Number and total area of bulliform cells in the top 2nd leaves at the heading stage. Values are means ± SD of three biological replicates. Asterisks indicate significant differences between acl83 and WT by Student's t‐test (**P < 0.01).

Gene cloning of ACL83

For genetic analysis, we crossed acl83 with 676R (WT) plants. All individuals in the F₁ population exhibited a normal phenotype as WT. While the F₂ population displayed a segregation of the two phenotypes, the segregation ratio of WT‐like to acl83‐like individuals demonstrated a ratio of 3:1 (χ ² = 2.43 < ${\chi }_{0.05}^2=3.84$) (Table S1). The result suggested that the acl83 mutant phenotype was controlled by a single recessive nuclear gene.

To identify the causal gene for acl83, SSR, and InDel markers, along with an F₂ population derived from the acl83 × ZH11 cross, were used to map the mutated locus. The initial mapping using 350 F₂ individuals with abaxially curled leaves showed that the acl83 locus was linked to the SSR marker RM163 on the long arm of chromosome 5 (Figure 3A). After that, 895 F₂ plants with the mutant phenotype and eight InDel markers with polymorphism between the two parents were used to fine map the acl83 locus (Figure 3B). The target locus was ultimately narrowed down to a 196‐kb region between InDel markers P5 and P7, at a genetic distance of 0.11 cM and 0.28 cM, respectively (Figure 3C). Within the 196‐kb region, 27 genes are annotated in the Rice Genome Annotation Project (https://rice.uga.edu) (Table S2).

Figure 3.

Mapping of the acl83 locus and identification of the acl83 mutation.

(A) Initial mapping of the acl83 locus on the long arm of chromosome 5 using 350 individuals exhibiting abaxially curled leaves from the acl83 × ZH11 F₂ population.

(B) Fine mapping of the acl83 locus using 895 mutant individuals, which was narrowed down to a 196‐kb region between InDel markers P5 and P7.

(C) Only one gene (LOC_Os05g32110 indicated with the red box) in the 196‐kb region was found to carry a mutation by MutMap analysis.

(D) Gene structure of the candidate gene ACL83 (LOC_Os05g32110). Black boxes represent exons and black lines represent introns. A red arrow denotes the single nucleotide G‐to‐A substitution occurring at position 2067, which is located at the first base of the fourth intron of ACL83 in the acl83 mutant.

(E) Genomic DNA sequencing confirming the G‐to‐A substitution in acl83 compared with the wild type (WT).

(F) cDNA sequencing revealing an additional “ATGA” insertion in acl83 compared with WT.

Subsequently, we performed whole‐genome resequencing and MutMap analysis using 30 F₂ individuals with abaxially curled leaves from the acl83 × 676R cross and found a single nonsynonymous mutation in LOC_Os05g32110, the only variant within the 196‐kb candidate region. We sequenced the LOC_Os05g32110 gene in genomic DNA from the acl83 mutant and the wild‐type parent 676R. Consistent with the MutMap analysis, a single nucleotide G to A substitution occurred at position 2067 of this gene sequence (corresponding to the first nucleotide of the fourth intron) in the acl83 mutant (Figure 3D,E). We next sequenced cDNA of LOC_Os05g32110 from acl83 mutant and the wild type using reverse transcription (RT)‐PCR. The result showed that the acl83 cDNA harbored a 4‐bp insertion (ATGA) (Figure 3F), which introduced a premature termination codon and resulted in a truncated protein. Therefore, LOC_Os05g32110 was identified as the candidate gene responsible for the acl83 mutant phenotype and designated ACL83.

Sequencing analysis revealed that ACL83 (LOC_Os05g32110) comprises six exons and five introns (Figure 3D), with a genomic span of 2851 bp and a coding sequence (CDS) of 1374 bp. The gene encodes a putative GPI‐anchored COBRA‐like protein (Figure S1A) (Dai et al., 2009). Phylogenetic analysis showed that ACL83 is more closely related to its maize (Zea mays) homolog than to the homologs from other species (Figure S1B).

Previous studies have revealed that LOC_Os05g32110 (OsBC1L4) participates in cell wall biosynthesis (Dai et al., 2011). Therefore, we analyzed the cell wall components of leaves from the acl83 mutant at the heading stage, when the abaxial leaf rolling phenotype was clearly evident. As a result, hemicellulose and lignin contents in the acl83 leaves were reduced by 26.3 and 22.3%, respectively, compared to the wild type (Figure S2B,C), whereas cellulose and pectin contents did not differ significantly (Figure S2A,D). These results indicate that the cell wall composition is altered in the acl83 mutant.

Complementation and CRISPR/Cas9 knockout of ACL83

To confirm that ACL83 is the causal gene for acl83, we conducted a complementation experiment. The full‐length cDNA of the wild‐type ACL83 gene was inserted into the pCAMBIA2300‐Actin1 vector to generate a complementation construct. Because the indica rice variety 676R is recalcitrant to Agrobacterium‐mediated transformation, we generated an acl83 near‐isogenic line (NIL) in the genetic background of the transformable japonica variety Zhonghua 11 (ZH11) through repeated backcrossing and selfing (Figure S3). Then, the complementation construct was introduced into the acl83‐NIL (NIL^ZH11), yielding 10 positive transgenic lines that were confirmed by PCR (Figure S4A). These transgenic lines all exhibited a normal leaf phenotype indistinguishable from the wild type (ZH11), with restored leaf rolling indices and BC morphology and size (Figure 4A,B,G,I,J). In addition, the transgenic lines expressing ACL83 fully rescued major agronomic traits, including plant height, number of productive panicles per plant, panicle length, number of spikelets per panicle, seed setting rate, and 1000‐grain weight (Figure S4B–G).

Figure 4.

Functional validation of ACL83 by complementation and CRISPR/Cas9‐mediated knockout.

(A, B) Plant architectures of ZH11, NIL^ZH11 and transgenic plants (TP1 and TP2) at the grain‐filling stage, and their transverse sections of the top 2nd leaves, respectively. Scale bars, 20 cm in (A) and 1 cm in (B).

(C) Gene structure of ACL83 (LOC_Os05g32110) and its sgRNA target sites used for CRISPR/Cas9‐mediated gene knockout.

(D) Alignment of the CRISPR/Cas9 target sites in ACL83 between the japonica cultivar ZH11 and the knockout lines (acl83 ^ZH11 #1, #2 and #3). Blue letters indicate target sequences; green letters indicate the protospacer adjacent motif (PAM) sites; red dashes and letters indicate nucleotide deletions and insertions, respectively.

(E, F) Plant architectures of ZH11 and the knockout lines at the heading stage, and their transverse sections of the top 2nd leaves, respectively. Scale bars, 20 cm in (E) and 0.5 cm in (F).

(G, H) Leaf rolling indices of the top 2nd leaves in ZH11, NIL^ZH11, TP and knockout lines at the grain‐filling stage. Ad and Ab indicate adaxial and abaxial rolling respectively.

(I, J) Paraffin sections and area quantification of bulliform cells, respectively, in the top 2nd leaves in ZH11, NIL^ZH11, TP and acl83 ^ZH11 at the grain‐filling stage. LV, large vein; SV, small vein. Red arrows indicate bulliform cells (BCs). Scale bar, 200 μm. Values are means ± SD of three biological replicates. In panels (G) and (J), statistical significance was determined by one‐way ANOVA with Tukey's multiple‐comparison test (P < 0.05), and different lowercase letters indicate significant differences. In panel (H), asterisks indicate significant differences from ZH11 as determined by Student's t‐test (**P < 0.01).

To further validate the function of ACL83, we used the CRISPR/Cas9 system to generate knockout lines (acl83 ^ZH11 ) in the ZH11 background. Sequence analysis confirmed that three knockout lines harbored insertions or deletions at both sgRNA target sites SG1 and SG2 (Figure 4C,D; Figure S5A), resulting in predicted premature termination of the ACL83 protein (Figure S5B). Importantly, the three homozygous knockout lines all exhibited an abaxially curled leaf phenotype similar to that of the acl83‐NIL (NIL^ZH11) (Figure 4E,F). The leaf rolling indices of the top 2nd leaves were 37.9%, 38.1%, and 38.8% in the three lines, respectively, whereas the wild‐type ZH11 leaf showed only slight adaxial rolling (LRI = 6.9%) (Figure 4H). Meanwhile, the total area of BCs adjacent to large veins and small veins was significantly increased in the knockout lines by 47.8 and 64.0%, respectively, compared to that in the wild‐type ZH11 (Figure 4I,J). In addition, the major agronomic traits were significantly impaired in the knockout lines (Figure S6A–F).

Finally, we generated ACL83‐overexpressing lines in the ZH11 background to assess the effect of elevated ACL83 expression. qRT‐PCR analysis confirmed that ACL83 transcript levels were markedly upregulated in the overexpression lines (Figure S7A). Phenotypic observations revealed no significant differences in leaf morphology or leaf rolling index between the overexpression lines and the wild‐type ZH11 plants (Figure S7B–D).

Collectively, both complementation and knockout assays demonstrated that ACL83 is essential for leaf morphology development in rice, and that a single‐nucleotide mutation in this gene causes the abaxially curled leaf phenotype of the acl83 mutant.

Expression pattern and subcellular localization

To explore the expression pattern of ACL83, we quantified its transcript levels in various tissues of wild‐type 676R plants at both the seedling and booting stages by qRT‐PCR. ACL83 was expressed in all tissues examined, including roots, stems, leaf blades, leaf sheaths, and young panicles, although the expression levels varied among various organs. Notably, the highest expression at the booting stage was observed in stems and leaf blades (Figure 5A). We further analyzed ACL83 expression in leaf blades during different developmental stages and found that transcript levels gradually increased from the 10th to the 15th leaf, with peak expression detected in the 15th leaf (the flag leaf) (Figure 5B).

Figure 5.

Expression pattern of the ACL83 gene and subcellular localization of the ACL83 protein.

(A) The spatiotemporal expression of ACL83 in root (R), leaf blade (L), stem (S), leaf sheath (LS), and young panicle (P) of the wild‐type plants at seedling stage and booting stage.

(B) Expression analysis of ACL83 in the wild‐type leaf blades at different developmental stages. L10, L11, L12, L13, L14, and L15 indicate the full‐expanded leaves from the stem base, respectively (L15 was the flag leaf). Values are means ± SD of three biological replicates.

(C) Subcellular localization of ACL83 protein. The fusion protein ACL83‐GFP and the empty GFP were co‐expressed with the plasma membrane marker RAC3‐mCherry, respectively. Green fluorescence indicates the signals from GFP protein, red fluorescence indicates the signals from mCherry protein, and yellow fluorescence shows images with two types of fluorescence merged. Scale bars, 10 μm.

To investigate subcellular localization of the ACL83 protein, an ACL83‐GFP recombinant vector was constructed and co‐expressed with the plasma membrane marker RAC3‐mCherry in rice protoplasts. As a result, the GFP signals from the ACL83‐GFP fusion protein precisely overlapped with the red fluorescence of the plasma membrane marker RAC3‐mCherry, whereas the green fluorescence from the GFP control showed no overlap with the red fluorescence of RAC3‐mCherry (Figure 5C). These results indicated that ACL83 is localized to the plasma membrane.

ROC5 activates the promoter of ACL83 via direct binding to the L1‐box motif

The leaf rolling phenotype of the acl83 mutant was similar to that of the roc5 mutant (Zou et al., 2011). Given that ROC5 is a HD‐ZIP IV transcription factor, we hypothesized that ROC5 might act upstream of ACL83 in a common genetic pathway. To test this, we performed a yeast one‐hybrid (Y1H) assay, which showed that ROC5 could bind to the ACL83 promoter (the 2071‐bp region upstream of the ATG) and activate the LacZ reporter gene (Figure 6A). Subsequently, we conducted a dual‐luciferase (LUC) assay. The firefly luciferase reporter (ACL83pro::LUC) was driven by the ACL83 promoter, with 35S‐driven Renilla luciferase (REN) serving as an internal control. The reporter construct was co‐transformed into tobacco leaf epidermal cells with either pC2300‐CaMV35S‐ROC5 or pC2300‐CaMV35S‐GFP (negative control) (Figure 6B). The results showed that the LUC fluorescence signal was markedly higher in the leaf cells co‐expressing ROC5 and ACL83pro::LUC than in those co‐expressing GFP and ACL83pro::LUC (control) (Figure 6C). Quantification of LUC and REN activities revealed that the LUC/REN ratio in the ROC5 effector group was 4.6‐fold higher than that in the GFP control (Figure 6D). A similar trend was observed in rice protoplasts (Figure S8).

Figure 6.

ROC5 directly regulates the target gene ACL83.

(A) Yeast one‐hybrid assay showing that AD‐ROC5 activates the ACL83 promoter‐driven LacZ reporter.

(B) Schematic representation of the effector and reporter constructs used for the dual‐luciferase assay. LUC, firefly luciferase; REN, Renilla luciferase.

(C) Chemiluminescence imaging of Nicotiana benthamiana leaves co‐infiltrated with ACL83pro::LUC together with GFP (left) or ROC5 (right).

(D) Activation of the ACL83 promoter by ROC5 in N. benthamiana leaves. The LUC/REN ratio of the GFP control was set to 1.

(E) Schematic diagram of the ACL83 promoter divided into 12 fragments (A–L) for DAP‐qPCR analysis. DNA sequences containing the L1‐box motif are indicated by white boxes.

(F) DAP‐qPCR analysis of ROC5 binding to ACL83 promoter fragments. Input DNA fragments without DAP incubation were used as a negative control and normalized to 1.

(G) Electrophoretic mobility shift assay (EMSA) showing binding of His‐ROC5 to the L1‐box‐containing probe in the ACL83 promoter. Increasing amounts (5×, 10× and 100×) of unlabeled wild‐type competitor probe were added as indicated.

(H) Gene structure of ROC5 and design of CRISPR/Cas9 target sites.

(I) Sequence alignment of the CRISPR/Cas9 target sites in ROC5 between the japonica cultivar ZH11 and the ROC5‐knockout lines (roc5 ^ZH11 #1, #2 and #3). Blue letters indicate target sequences; green letters indicate PAM sites; red dashes and letters indicate nucleotide deletions and insertions, respectively.

(J) Plant architectures of ZH11 and the ROC5‐knockout lines at the grain‐filling stage. Scale bar, 20 cm.

(K) Leaf rolling indices of the top 2nd leaves in ZH11 and the ROC5‐knockout lines at the grain‐filling stage. Ad and Ab indicate adaxial and abaxial rolling, respectively.

(L, M) Relative expression levels of ACL83 in ROC5‐knockout lines and ROC5 in ACL83‐knockout lines, respectively. Total RNA was extracted from leaves at the booting stage, and relative expression levels were normalized to Actin1. In (L), the expression level of ACL83 in ZH11 was set to 1.0, and values in the ROC5‐knockout lines were calculated accordingly. In (M), the expression level of ROC5 in ZH11 was set to 1.0, and values in the ACL83‐knockout lines were calculated accordingly. Values are means ± SD of three biological replicates. Asterisks indicate significant differences from ZH11 as determined by Student's t‐test (**P < 0.01).

To further confirm the direct binding of ROC5 to the ACL83 promoter, we expressed and purified recombinant ROC5 protein (Figure S9), and then performed DNA affinity purification followed by qPCR (DAP‐qPCR). The result suggested that promoter fragments J and K were enriched 6.0‐fold and 1.0‐fold, respectively, relative to the control (Figure 6E,F), indicating that ROC5 directly binds to fragment J of the ACL83 promoter in vitro. HD‐ZIP IV transcription factor regulates downstream target genes by binding to the L1‐box motif (5′‐TAAATGYA‐3′) (Abe et al., 2003; Fang et al., 2021; Wu et al., 2011). Notably, this motif is present in fragment J of the ACL83 promoter. Therefore, we performed an electrophoretic mobility shift assay (EMSA) to test whether ROC5 directly binds to a probe containing the L1‐box sequence. The results showed that His‐ROC5, but not the His tag alone, specifically bound to the L1‐box motif within fragment J of the ACL83 promoter (Figure 6G).

Taken together, our in vivo and in vitro results demonstrated that ROC5 positively regulates ACL83 by directly binding to its promoter, suggesting that ACL83 could act downstream of ROC5 to regulate leaf development.

ACL83 acts downstream of ROC5 in a genetic pathway regulating abaxial leaf rolling

To confirm the genetic relationship between ACL83 and ROC5, we used the CRISPR/Cas9 system to generate a roc5 single‐knockout mutant by targeting ROC5 alone (Figure 6H), and an acl83/roc5 double‐knockout mutant by simultaneously targeting both ACL83 and ROC5 (Figure 7A), all in the ZH11 genetic background.

Figure 7.

Genetic evidence showing that ACL83 acts downstream of ROC5 to regulate bulliform cell development and leaf rolling.

(A) Schematic representation of the CRISPR/Cas9 construct containing four sgRNA expression cassettes (SG1 and SG2 targeting ACL83, and SG3 and SG4 targeting ROC5).

(B) Gene structures of ACL83 and ROC5, and their sequence alignment of the CRISPR/Cas9 target sites between the ZH11 and the acl83/roc5 double‐mutant lines (#1, #2, and #3). Blue letters indicate target sequences; green letters indicate PAM sites; red dashes and letters indicate nucleotide deletions and insertions, respectively.

(C, D) Plant architectures of ZH11, roc5 ^ZH11 ‐2, acl8 ^3ZH11 ‐1, and the double mutant acl83/roc5, and their transverse sections of the top 2nd leaves at the grain‐filling stage. Scale bars, 20 cm in (C) and 0.5 cm in (D).

(E) Leaf rolling indices of the top 2nd leaves in ZH11, roc5 ^ZH11 , acl83 ^ZH11 , and acl83/roc5 at the grain‐filling stage. Ad and Ab indicate adaxial and abaxial rolling respectively.

(F) Paraffin sections of the leaves in ZH11, roc5 ^ZH11 ‐2, acl83 ^ZH11 ‐1, and acl83/roc5 at the heading stage. LV, large vein; SV, small vein. Red arrows indicate bulliform cells (BCs). Scale bars, 200 μm.

(G, H) Number and total area of bulliform cells in the top 2nd leaves from ZH11, roc5 ^ZH11 , acl83 ^ZH11 , and acl83/roc5. LV, large vein; SV, small vein. Values are means ± SD of three biological replicates. Statistical significance was determined by one‐way ANOVA with Tukey's multiple‐comparison test (P < 0.05), and different lowercase letters indicate significant differences.

In the roc5 single‐knockout mutant, three independent lines harbored insertions or deletions at two CRISPR/Cas9 target sites SG3 and SG4 (Figure 6I; Figure S10A), resulting in premature translational termination and a truncated ROC5 protein (Figure S10B). As expected, all three lines exhibited a markedly abaxially curled leaf phenotype (Figure 6J), with leaf rolling indices of 32.4, 35.5, and 48.5%, respectively (Figure 6K). qRT‐PCR analysis revealed that ACL83 expression was significantly reduced in these roc5 knockout lines (Figure 6L). In contrast, ROC5 expression was unchanged in the three independent acl83 knockout lines (Figure 6M).

In the acl83/roc5 double‐knockout mutant, three homozygous lines harbored deletions or insertions at all four CRISPR/Cas9 target sites (SG1 to SG4) in both ACL83 and ROC5 (Figure 7B; Figure S11A,B), resulting in truncated ACL83 and ROC5 proteins (Figure S11C,D). Compared with wild‐type ZH11, all three acl83/roc5 double‐knockout lines exhibited a markedly abaxially curled leaf phenotype similar to those of the acl83 and roc5 single‐knockout mutants (Figure 7C,D). Leaf rolling indices of the roc5, acl83, and acl83/roc5 mutants were 36.2, 39.6, and 41.6%, respectively (Figure 7E), with no significant difference among them. Histological analysis showed that the number of BCs in the top 2nd leaves from all three mutants (roc5, acl83, and acl83/roc5) was comparable to that in ZH11 (Figure 7F,G), whereas the total area of BCs was markedly enlarged (Figure 7F,H). Specifically, in large veins, BC area increased by 54.5, 45.5, and 50.0% in the roc5, acl83, and acl83 roc5 mutants, respectively, with no significant difference among them. A similar trend was observed in small veins, with increases of 75.0, 62.5, and 68.8%, respectively. Additionally, the acl83/roc5 double mutant displayed agronomic traits similar to those of the acl83 single‐knockout mutant, including plant height, number of productive panicles per plant, panicle length, number of spikelets per panicle, and 1000‐grain weight (Figure S12A–F).

Collectively, our data confirmed that ACL83 acts downstream of ROC5 in a genetic pathway to control abaxial leaf rolling in rice by regulating cell expansion.

DISCUSSION

Leaf rolling is an important agronomic trait in rice. To date, many genes involved in the regulation of adaxially curled leaves have been cloned and functionally characterized, such as SLL1, RL14, CLD1/SRL1, OsMYB103L, SRL2, NRL2, LRRK1, PSL1, DRB2/SRL10, and OsLC1 (Fang et al., 2012; Li, Zhang, et al., 2017; Liu et al., 2016; Wang et al., 2023; Wu et al., 2025; Xiang et al., 2012; Yang et al., 2014; Yuan et al., 2022; Zhang et al., 2009, 2021; Zhao et al., 2016; Zhou et al., 2018). However, only a few genes associated with abaxially curled leaves have been identified, including ACL1, ACL2, OsZHD1, OsZHD2, ROC1, ROC5, and ROC8/Url1 (Fang et al., 2021; Li et al., 2010; Sun et al., 2020; Tao et al., 2024; Xu et al., 2014, 2021; Zou et al., 2011). Notably, all these genes encode transcription factors. Overexpression of ACL1, ACL2, OsZHD1, or OsZHD2 leads to abaxially curled leaves. In contrast, overexpression of ROC1, ROC5, or ROC8 causes an adaxially curled leaf phenotype, whereas their loss‐of‐function results in abaxially curled leaves. In this study, we isolated mutant acl83, which showed an abaxially curled leaf phenotype due to increased area of bulliform cells (BCs). The candidate gene for acl83 is LOC_Os05g32110 (ACL83) encoding a putative COBRA family protein. In the acl83 mutant, the mutation causes aberrant splicing, leading to the premature translation termination. Complementation and knockout assays confirmed that ACL83 is the causal gene for the acl83 mutant phenotype. Therefore, our data indicated that ACL83 is a distinct gene controlling abaxial leaf rolling in rice.

COBRA belongs to a plant‐specific multigene family (Roudier et al., 2002). This family comprises 12 members in Arabidopsis (COBRA and AtCOBL1‐11), 9 in maize (ZmBK2, ZmBK2L1, and ZmBK2L3‐9), and 11 in rice (BC1, OsBC1Lp1, and OsBC1L1‐9) (Dai et al., 2009). In rice, BC1 regulates the mechanical strength of tissues/organs (Li et al., 2003); OsBC1L1 and OsBC1L8 are involved in regulating stomatal development (Li et al., 2021); OsBC1L6 modulates β‐glucan levels in the endosperm cell wall (Midorikawa et al., 2019); and OsBC1L4 affects cellulose synthesis in cell walls, with decreased cellulose content and increased pectin content in the mature internodes of the osbc1l4 mutant (Dai et al., 2011). Nevertheless, the functional role of COBRA‐like proteins in the regulation of leaf rolling remains unclear. In our study, we demonstrated that acl83 is allelic to osbc1l4, and that ACL83 localizes to the plasma membrane in rice protoplasts (Figure 5C). In the acl83 leaves at the heading stage, hemicellulose and lignin contents were reduced, whereas cellulose and pectin contents did not differ significantly (Figure S2). We speculate that the observed differences in cell wall components between acl83 and osbc1l4 may result from variations in the developmental stages and tissue types sampled. Importantly, the number of BCs remained unchanged in acl83, whereas their area was markedly enlarged (Figure 2), supporting a role for COBRA‐like proteins in regulating cell expansion. Thus, ACL83 is a COBRA‐like protein that controls abaxial leaf rolling in rice via the regulation of cell expansion. Nevertheless, whether there is a direct causal relationship between cell wall composition changes and bulliform cell expansion remains to be determined and requires further investigation.

ROC5 encodes a HD‐ZIP IV transcription factor in rice. ROC5‐knockout leads to abaxial leaf rolling by increasing the number and size of BCs on the adaxial surface, whereas ROC5 overexpression causes adaxial leaf rolling by reducing the number and size of BCs (Xu et al., 2021; Zou et al., 2011). In addition, ROC5 participates in a transcriptional regulatory network that coordinates multiple downstream genes to modulate BC development and leaf rolling. For example, ROC5 regulates the epidermis‐specific gene PFL (protodermal factor like), and silencing PFL (via PFL‐RNAi) causes abaxial leaf rolling by increasing the number and size of BCs (Zou et al., 2011). ROC5 directly activates CFL2, a cytochrome P450 gene whose mutation leads to abnormal epidermal development, enlarged BCs, and curling at the base of the flag leaf (Zhang et al., 2020). Additionally, ROC5 can form a transcriptional repressor complex with URL1 (ROC8) and TPL2 to negatively regulate ACL1, thereby modulating BC development and leaf rolling (Fang et al., 2021). In this study, we confirmed that ROC5 binds to the ACL83 promoter and activates ACL83 expression using yeast one‐hybrid and dual‐luciferase (LUC) assays (Figure 6A–D; Figure S8A). DAP‐qPCR demonstrated significant enrichment of a specific ACL83 promoter fragment by ROC5 (Figure 6E,F), and EMSA revealed that ROC5 directly binds to an L1‐box motif within this fragment (Figure 6G). Furthermore, we generated roc5 single‐knockout mutants in the ZH11 background, which exhibited an abaxially curled leaf phenotype (Figure 6J,K) and an enlarged BC area (Figure 7F–H)—phenotypes similar to those of the acl83 single‐knockout mutant. ACL83 transcript levels were significantly reduced in ROC5‐knockout lines (Figure 6L), whereas ROC5 expression was unchanged in the ACL83‐knockout lines (Figure 6M). Importantly, the leaf rolling index and BC area in the acl83/roc5 double mutant are not significantly greater than those in either single mutant (Figure 7E,F,H), and major agronomic traits of the double mutant largely resemble those of the acl83 single‐knockout mutant (Figure S12A–F). Taken together, the molecular and genetic evidence support a model in which the absence of additive or synergistic effects in the acl83/roc5 double mutant suggests that ACL83 and ROC5 act in the same genetic pathway, and the direct activation of ACL83 by ROC5 and the reduced expression of ACL83 in the roc5 single‐knockout mutant reveals that ACL83 functions downstream of ROC5. In wild‐type plants, ROC5 activates ACL83, forming a ROC5‐ACL83 regulatory module that restricts BC expansion and helps maintain the balance between the adaxial and abaxial sides of the leaf blade, thereby limiting abaxial leaf rolling. When ROC5 or ACL83 is disrupted, reduced ACL83 activity permits excessive BC expansion, leading to abaxial leaf rolling. In a word, our findings reveal a distinct ROC5‐ACL83 genetic pathway in which the COBRA‐like protein ACL83 functions downstream of ROC5 to control abaxial leaf rolling in rice by regulating cell expansion.

Notably, ACL83 expression was not completely abolished in roc5 knockout mutants (Figure 6L), suggesting that ACL83 is not exclusively regulated by ROC5. Instead, functional redundancy among other members of the HD‐ZIP IV transcription factor family, or regulation by uncharacterized transcription factors, likely sustains basal ACL83 expression.

EXPERIMENTAL PROCEDURES

Plant materials and growth conditions

The acl83 mutant was obtained through screening an EMS mutagenesis library constructed in the indica rice restorer line 676R. For genetic and MutMap analyses, the mutant was crossed with the wild‐type 676R. To perform gene mapping, acl83 was crossed with the japonica rice variety Zhonghua 11 (ZH11). Rice materials were grown during the normal rice growing season in the experimental fields of Sichuan Agricultural University in Chengdu, China. All the primers used in this study were listed in Table S3.

Measurement of Leaf rolling index (LRI)

The adaxial (Ad) and abaxial (Ab) leaf rolling indices (LRIs) were measured as previously described (Shi et al., 2007). LRI value was calculated using the formula: LRI (%) = ([Lw − Ln] ÷ Lw) × 100, where Ln denotes the leaf width under the natural (folded) condition, and Lw denotes the maximum leaf width at the same position when fully unfolded.

Histology analysis

For paraffin section analysis, leaf samples were collected from the middle region of the second leaf from the top of each plant at the heading stage, unless otherwise stated. The leaf samples were immediately submerged in 70% FAA solution (70% ethanol, 5% formaldehyde, 5% glacial acetic acid) (G1103, Servicebio, China) and fixed for 24 h, and then processed for paraffin embedding and sectioning by Servicebio (Wuhan, China). The sections were imaged and analyzed using SlideViewer software (3DHIESTECH, Hungary). Bulliform cells adjacent to the midrib, large, and small veins in the leaf samples were counted, and their total areas were measured using ImageJ software (http://rsb.info.nih.gov/ij/).

Cell wall composition analysis

At the heading stage, flag leaves were collected from wild‐type 676R and acl83 mutant plants, immediately frozen in liquid nitrogen, and stored at −80°C until further analysis. For cell wall composition analysis, the samples were dried at 105°C for 1 h and then at 65°C for 24 h and subsequently ground into fine powder. The contents of cellulose, hemicellulose, lignin, and pectin were determined as previously described (Dai et al., 2011).

Gene mapping

To identify the candidate gene of acl83, map‐based cloning was performed using molecular markers and an F₂ population derived from the acl83 × ZH11 cross. A total of 895 F₂ individuals exhibiting an abaxially curled leaf phenotype were selected for linkage mapping analysis. On the other hand, leaf blades from 30 F₂ individuals with the mutant phenotype in the acl83 × 676R population were used to prepare DNA pools, which were subjected to bulk sequencing (Novogene Co., Ltd., Beijing, China). MutMap analysis was then performed as previously described (Abe et al., 2012), using the wild‐type 676R genome as the reference.

Vector construction and plant transformation

For complementation and overexpression assay, the 1374‐bp full‐length coding sequence of ACL83 (LOC_Os05g32110) was amplified from cDNA of the wild‐type 676R and inserted into the binary vectors pCAMBIA2300‐Actin1 and pCAMBIA2300‐35S, respectively. The complementation construct was transformed into the near‐isogenic line of acl83 in the ZH11 genetic background (NIL^ZH11), while the overexpression construct was introduced into wild‐type ZH11 using the Agrobacterium‐mediated method. For genome editing, CRISPR/Cas9 vectors were constructed as described previously (Cong et al., 2013). The targeting sequences for ACL83 (SG1 and SG2), ROC5 (SG3 and SG4), and both genes (SG1–SG4) were synthesized and annealed to form the oligo adapters. These constructs were introduced into the ZH11 genetic background via Agrobacterium‐mediated transformation as described previously (Hiei et al., 1994). ACL83 and ROC5 were individually knocked out to generate single mutants. Meanwhile, ACL83 and ROC5 were simultaneously knocked out to create an acl83/roc5 double mutant.

RNA isolation and qRT‐PCR analysis

Total RNA was extracted using Total RNA Extraction Reagent (R401‐01, Vazyme, China) from rice leaf blades, roots, leaf sheaths, stems, and young panicles collected at various developmental stages. First‐strand cDNA was synthesized from 2 μg total RNA with HiScript II Q RT SuperMix (R223‐01, Vazyme, China). qRT‐PCR was performed using SYBRPrime (BG0014, Bioground, China) on a CFX96 real‐time PCR system (qTOWER3 G, Jena, Germany). The PCR reaction mixture and program were as described previously (Guo et al., 2025). The Actin1 gene was used as the internal control. Relative expression levels were calculated using the 2^−ΔΔCT method from at least three independent biological replicates.

Subcellular localization

The 1371‐bp full‐length coding sequence of ACL83 was amplified from the wild‐type cDNA of 676R, and inserted into the pCAMBIA2300‐GFP vector to generate the ACL83‐GFP fusion construct. The ACL83‐GFP construct and the empty vector (control) were each co‐transformed with the plasma membrane marker (RAC3‐mCherry) (Chen et al., 2010) into rice protoplasts. The transformation procedure was performed as described previously (Zhang et al., 2011). After overnight incubation in the dark, fluorescent signals were observed using a laser scanning confocal microscope (STELLARIS STED, Leica, Germany).

Yeast one‐hybrid assay

As described previously (Liang et al., 2014), the full‐length coding sequence (CDS) of ROC5 was amplified from japonica rice variety ZH11 using the primers containing homologous arms and an EcoRI restriction site, and then inserted into the pJG4‐5 vector to generate AD‐ROC5. The ACL83p:LacZ reporter was constructed by PCR amplification of the ACL83 promoter (2071 bp upstream of ATG) using primers containing homologous arms and an XhoI restriction site, followed by cloning into the pLacZi2μ vector. The ACL83p:LacZ reporter plasmid was co‐transformed with the AD‐ROC5 into the yeast strain EGY48. Transformants were selected on SD/‐Trp/‐Ura medium and incubated at 28°C for 2–4 days. Single colonies were streaked onto SD/‐Trp/‐Ura plates supplemented with X‐gal and incubated at 28°C in the dark for 1–2 days. Similar results were obtained in three independent biological replicates; a representative result is shown.

Dual‐LUC assay

Effector constructs were generated by cloning the full‐length coding sequence of ROC5 into pCAMBIA2300 under the control of the CaMV 35S promoter. For the reporter, a 2071‐bp fragment of the ACL83 promoter was inserted into pGreenII 0800‐LUC double reporter vector to generate the ACL83p::LUC construct. In this system, Renilla luciferase (REN), driven by the CaMV 35S promoter, served as an internal control. Effector and reporter plasmids were co‐delivered into Nicotiana benthamiana leaf epidermal cells via Agrobacterium‐mediated infiltration or co‐introduced into rice protoplasts by PEG‐mediated transformation, as described previously (Tao et al., 2021). For N. benthamiana assay, 1 mM D‐luciferin potassium salt (40902ES01, Yeasen, China) was smeared onto the leaf surface, and luminescence was captured using a FUSION‐FX7 imaging system (VILBER LOURMAT, Frence). Firefly luciferase (LUC) and REN activities were quantified with the Dual Luciferase Reporter Assay Kit (DL101‐01, Vazyme, China) on a GloMax® 96 Microplate Luminometer (Promega, America) according to the manufacturer's instruction. At least three independent replicates were performed.

DAP‐qPCR assay

A prokaryotic expression system was employed to produce recombinant ROC5 protein for DNA affinity purification (DAP) assay. The full‐length coding sequence of ROC5 was amplified from the ZH11 and inserted into the pCold‐TF vector to generate a TF‐His‐ROC5 fusion construct. The fusion protein was expressed in Escherichia coli BL21 and purified by Ni²⁺‐affinity chromatography using Ni‐TED 6FF resin (Sangon, China). The corresponding His‐TF protein (lacking ROC5) was prepared and purified in parallel as a negative control. Genomic DNA was extracted from 3‐week‐old seedlings of ZH11 and fragmented to approximately 100–500 bp by ultrasonication. To interrogate ROC5 binding across the ACL83 promoter, 12 primer pairs generating 100–300 bp amplicons were designed to tile the promoter region. The DAP procedure was performed as described previously (Li, Zhu, et al., 2017; Tao et al., 2021). DNA eluted from the DAP assay was quantified by qPCR using SYBRPrime (BG0014, Bioground, China) on a CFX96 real‐time PCR system (qTOWER3 G, Jena, Germany), with cycling conditions as described previously (Li, Zhu, et al., 2017). The DNA sequence of the UBQ5 promoter was used as an internal control. For enrichment analysis, input DNA fragments (without DAP incubation) were used as a negative control and normalized to 1.

EMSA

Electrophoretic mobility shift assay (EMSA) was performed as described previously (Sun et al., 2021). Purified TF‐His‐ROC5 protein was used for the binding reactions. A biotin‐labelled 59‐bp DNA probe encompassing the L1‐box cis‐element from the ACL83 promoter was synthesized by Tsingke (Beijing, China) and used in the assay. The corresponding unlabeled probe was used as a competitor probe, and purified TF‐His alone protein was used as a negative control. EMSA reactions and chemiluminescent detection were carried out using the Chemiluminescent EMSA Kit (GS009, Beyotime, China) according to the manufacturer's instructions.

Statistical analysis

Statistical analyses were performed as indicated in the figure legends.

ACCESSION NUMBERS

Sequence data from this article can be found in the MSU Rice Genome Annotation Project Database under the following accession numbers: ACL83 (LOC_Os05g32110), ROC5 (LOC_Os02g45250).

AUTHOR CONTRIBUTIONS

PW and XD designed the research and supervised the study. CC performed most of the research and drafted the manuscript. RL conducted gene mapping and genetic transformation for functional complementation. JD, RL, HZ, DL, YX, JL, HS, YL, XY, JG, YZ, and BY performed the experiments and conducted fieldwork. CC, PW, XD, and CS analyzed data. XD, PW, and CC revised the manuscript. All authors reviewed the results and approved the final manuscript.

CONFLICT OF INTEREST

The authors declare that they have no conflicts of interest.

Supporting information

Figure S1. Sequence alignment and phylogenetic tree of ACL83 and its homologs.

Figure S2. Analysis of cell wall components in the leaves of acl83.

Figure S3. Construction of the acl83 near‐isogenic line (NIL^ZH11) in the ZH11 genetic background.

Figure S4. Identification of transgenic plants (TP) for genetic complementation and comparison of agronomic traits among ZH11, NIL^ZH11, and TP.

Figure S5. Sequencing chromatograms and protein sequence alignment between ZH11 and ACL83 knockout lines.

Figure S6. Comparison of agronomic traits between ZH11 and ACL83‐knockout lines.

Figure S7. Phenotypic analysis of ACL83‐overexpression lines.

Figure S8. Activation of the ACL83 promoter by ROC5 in protoplasts.

Figure S9. Expression and purification of ROC5 recombinant protein in Escherichia coli for DAP‐qPCR and EMSA.

Figure S10. Sequencing chromatograms and protein sequence alignment between ZH11 and ROC5‐knockout lines.

Figure S11. Sequencing chromatograms and protein sequence alignment of ACL83 and ROC5 between ZH11 and acl83/roc5 double‐mutant lines.

Figure S12. Comparison of agronomic traits among ZH11, roc5 ^ZH11, acl83 ^ZH11, and acl83/roc5 double mutant.

Table S1. Genetic segregation of the abaxially curled leaf trait in an F₂ population derived from the cross between the acl83 mutant and the wild‐type 676R.

Table S2. Putative genes within the 196‐kb mapping region.

Table S3. Primers used in the study.

DATA AVAILABILITY STATEMENT

All relevant data can be found within the manuscript and its Supplementary Information materials.