Cytological Observation and RNA-Seq Analyses Reveal miR9564 and Its Target Associated with Pollen Sterility in Autotetraploid Rice

Understanding the regulation of autotetraploid sterility is essential for harnessing the strong advantages in genomic buffer capacity, biodiversity, and heterosis of autotetraploid rice. miRNAs play crucial roles in fertility regulation, yet information about their reproductive roles and target genes in tetraploid rice remains limited. Here, we used three tetraploid lines, H1 (fertile), HF (fertile), and LF (sterile), to investigate cytological features and identify factors associated with autotetraploid sterility. LF showed abnormal meiosis, resulting in low pollen fertility and viability, ultimately leading to scarce fertilization and a low-seed setting compared to H1 and HF. RNA-seq revealed 30 miRNA-candidate target pairs related to autotetraploid pollen sterility. These pairs showed opposite expression patterns, with differential expression between fertile lines (H1 and HF) and the sterile line (LF). qRT-PCR confirmed that miR9564, miR528, and miR27874 were highly expressed in the anthers of H1 and HF but not in LF, while opposite results were obtained in their targets (ARPS, M2T, and OsRPC53). Haplotype and expression pattern analyses revealed that ARPS was specifically expressed in lines with the same haplotype of MIR9564 (the precursor of miR9564) as LF. Furthermore, the Dual-GFP assay verified that miR9564 inhibited the fluorescence signal of ARPS-GFP. The over-expression of ARPS significantly decreased the seed setting rate (59.10%) and pollen fertility (50.44%) of neo-tetraploid rice, suggesting that ARPS plays important roles in autotetraploid pollen sterility. This study provides insights into the cytological characteristic and miRNA expression profiles of tetraploid lines with different fertility, shedding light on the role of miRNAs in polyploid rice.


Introduction
Regardless of autopolyploidy and allopolyploidy, polyploid individuals have substantial advantages in stress resistance, biosynthesis, genomic buffering capacity, biodiversity, and heterosis [1][2][3].However, the widespread complex reproductive defects in autotetraploid plants limit their application [4][5][6][7].Autotetraploid rice, resulting from doubling its diploid counterpart, represents a promising breeding germplasm, which displays high salt and heavy metal ion resistance [8][9][10][11].Partial pollen sterility stands out as a major factor contributing to a low-seed set of autotetraploid rice.Within this context of pollen sterility, abundant abnormalities had been observed in the meiotic anthers of autotetraploid rice, including abnormal chromosome behavior and tapetum development [12][13][14].Chinese scientists dedicated at least 20 years in repeated selfing and selection to overcome polyploid infertility, resulting in the successful breeding of fertile tetraploid rice, such as PMeS polyploid rice and neo-tetraploid rice (NTR, with 80% seed setting) [15][16][17][18].Notably, NTR lines have shown the ability to overcome polyploidization sterility when crossed with typical autotetraploid rice exhibiting low fertility [16].These fertile tetraploid rice germplasms have produced a valuable opportunity to identify genes associated with autotetraploid infertility.Our previous studies had reported 15 NTR lines to assess their yield traits, reproduction, and gene expression [16][17][18].Relative to diploid counterparts and NTR lines, great changes have been detected in expression levels of genes, microRNAs (miRNAs), and long non-coding RNAs during embryo sac development in autotetraploid rice, such as meiotic genes [12][13][14]19].
miRNA is a special non-coding RNA that targets specific areas of mRNA to suppress gene expression by a direct cleaving of mRNAs by miRNAs or inhibiting the translation of the target mRNA to participate in numerous crucial processes [20].For instance, miR528 regulates pollen intine formation by targeting the uclacyanin gene OsUCL23 to impact flavonoid metabolism [21], modulates flowering time by targeting OsRFI2 [22], and regulates OsSPL9 to affect the antiviral response [23].miR2118 regulates reproductive tissue development in rice [24].Additionally, miR167d, miR398b, miR7695, and miR1873 are associated with blast resistance [25][26][27][28].miR156 defines ideal plant architecture by targeting OsSPL14 and regulates seed dormancy through the gibberellin pathway [29,30].Recently, the regulatory network of OsPIL15-miR530-OsPL3 and miR1432-OsACOT was found to be related to rice grain yield [31,32].
From diploidy to autotetraploidy, the expression patterns of miRNA dynamically changed in the anther and ovary, indicating the important roles of miRNAs in autotetraploid reproduction [12,13,16].There were 172 miRNAs differentially expressed in the meiosis anther of autotetraploid rice, including 57 miRNAs specifically expressed in autotetraploid rice [12].Furthermore, 321 and 368 miRNAs showed differential expression during anther and embryo sac development in autotetraploid rice, respectively [13].In addition, 122 miRNAs were differentially expressed in the meiotic anther of the neo-tetraploid line H3 compared to autotetraploidy T452 [16].However, the functional verification about miRNAs and their negative targets in the reproductive regulation of tetraploid rice is still limited.
In this study, two neo-tetraploid lines (H1 and HF) and one low-fertility tetraploid line (LF) were utilized for cytological observation, miRNA-seq, and RNA-seq to identify common differentially expressed miRNAs and their negative targets relative to tetraploid reproduction.One key candidate target was further over-expressed in neo-tetraploid rice.As expected, the transgenic materials exhibited a low-seed set and low pollen fertility.These findings contributed to enrich our understanding of pollen development in autotetraploid rice.
Moreover, the embryogenesis and endosperm development of HF and LF were also observed using WE-CLSM at 1 h after flowering, 1 day after flowering (DAF), 3 DAF, and 5 DAF.In HF, 91.67%, 77.62%, 77.35%, and 77.01%samples were able to be fertilized at 1 h, 1 DAF, 3 DAF, and 5 DAF, respectively, while only 19.05%, 15.67%, 37.32%, and 6.18% of observed samples were fertilized in LF (Figure 2C-G).Most of the samples in LF were still unfertilized and retained the mature embryo-sac-like morphology (Figure 2H-K) or exhibited other abnormalities (Figure S3).These results indicate that LF displayed defects during pollen development, double fertilization, and embryogenesis.

LF Showed Severe Abnormal Male Meiotic Process
Similar to diploid rice, HF underwent meiosis with fewer abnormalities.However, more abnormal chromosome behaviors were observed in meiotic pollen mother cells (PMCs) of LF, including chromosome dragging at metaphase I and metaphase II, chromosome lagging at anaphase I and anaphase II, chromosome behavior disorder, micronuclei at telophase I and telophase II, abnormal cell shape and asynchrony of the chromosome during meiosis II, and an abnormal tetrad (Figure 3).High frequencies of chromosome behavior abnormalities were observed since the early stage of meiosis in LF, which were significantly higher than HF at all observed stages (Figure 4A).Abnormal chromosome configurations were found in both LF and HF, such as univalents, trivalents, and particularly quadrivalents.The number of quadrivalents per cell was lower in LF (6.86) compared to HF at diakinesis (7.97, Figure 4B).A ring shape was the most frequent configuration of quadrivalents in both HF and LF.HF had a higher frequency of ring-shape (64.28%) and X-shape (16.91%) quadrivalents than LF (57.93%, 11.76%, respectively), while LF had a

LF Showed Severe Abnormal Male Meiotic Process
Similar to diploid rice, HF underwent meiosis with fewer abnormalities.However, more abnormal chromosome behaviors were observed in meiotic pollen mother cells (PMCs) of LF, including chromosome dragging at metaphase I and metaphase II, chromosome lagging at anaphase I and anaphase II, chromosome behavior disorder, micronuclei at telophase I and telophase II, abnormal cell shape and asynchrony of the chromosome during meiosis II, and an abnormal tetrad (Figure 3).High frequencies of chromosome behavior abnormalities were observed since the early stage of meiosis in LF, which were significantly higher than HF at all observed stages (Figure 4A).Abnormal chromosome configurations were found in both LF and HF, such as univalents, trivalents, and particularly quadrivalents.The number of quadrivalents per cell was lower in LF (6.86) compared to HF at diakinesis (7.97, Figure 4B).A ring shape was the most frequent configuration of quadrivalents in both HF and LF.HF had a higher frequency of ring-shape (64.28%) and X-shape (16.91%) quadrivalents than LF (57.93%, 11.76%, respectively), while LF had a higher frequency of chain-shaped quadrivalents (21.10%) than HF (12.54%, Figure 4C).A similar result was observed during metaphase I (Figure 4B,C).

Comparative miRNA Expression Profiles in Meiotic Anthers of H1, HF, and LF
To investigate the differences in miRNA expression during anther development among tetraploid lines with opposite fertility, three small RNA libraries from meiotic anthers of H1, HF, and LF were sequenced by using Illumina Solexa high-throughput sequencing technology.A total of 38,569,001, 39,942,383, and 45,913,426 raw reads were generated, yielding 28,129,531 (72.93%), 27,818,269 (69.65%), and 27,648,526 (60.22%) valid reads from H1, HF, and LF, respectively (Table S1).A significant correlation was observed among three biological replicates of each library with a correlation coefficient exceeding 0.95 (Figure S4).
A total of 929 miRNAs were detected across the three libraries (Table S2).A total of 692 miRNAs were detected in H1, comprising 268 known and 424 novel miRNAs.Among

Comparative miRNA Expression Profiles in Meiotic Anthers of H1, HF, and LF
To investigate the differences in miRNA expression during anther development among tetraploid lines with opposite fertility, three small RNA libraries from meiotic anthers of H1, HF, and LF were sequenced by using Illumina Solexa high-throughput sequencing technology.A total of 38,569,001, 39,942,383, and 45,913,426 raw reads were generated, yielding 28,129,531 (72.93%), 27,818,269 (69.65%), and 27,648,526 (60.22%) valid reads from H1, HF, and LF, respectively (Table S1).A significant correlation was observed among three biological replicates of each library with a correlation coefficient exceeding 0.95 (Figure S4).
To gain further insight into the functions of coDEMs, we conducted an analysis of their potential targets.In total, 5944 mRNAs transcribed by 4866 genes were annotated as potential targets of 60 coDEMs.Gene Ontology (GO) analyses of 3476 predicted target genes of 44 cdDEMs revealed significant enrichment in various processes including the protein modification process (GO: 0006464), regulation of gene expression and epigenetics (GO: 0040029), signal transduction (GO: 0007165), pollen-pistil interaction (GO: 0009875), cell differentiation (GO: 0030154), post-embryonic development (GO: 0009791), anatomical structure morphogenesis (GO: 0009653), plasma membrane (GO: 0005886), kinase activity (GO: 0016301), and nucleotide binding (GO: 0000166) (Figure S5).Most of the enriched GO items were associated with reproductive development, suggesting that these coDEMs might play important roles during reproductive development in tetraploid rice.
qRT-PCR was performed to verify the expression levels of the four aforementioned coDEM-DEG pairs during anther development of H1, HF, and LF.miR9564 showed upregulated expression in developmental stage 7 to 10 (S7-S10) anthers of HF and H1 compared to LF, while its target ARPS showed down-regulated expression during the same stage.Similarly, miR528 and miR27874 displayed up-regulated expression in S7-S10 anthers of HF and H1, while their targets M2T and M3T showed down-regulated expression during the same stage.Another cuDEM, miR818d, was confirmed with up-regulated expression in S8a-S8b anthers of HF and H1, but its potential target M4T was also up-regulated in S8a anthers.M4T1 encoded an expressed protein without known function and was mainly highly expressed in vegetative organs.Opposite expression patterns were found in miR9564-ARPS, miR528-M2T, and miR27874-M3T, but not in miR818d-M4T (Figure 6B).Because of its most substantial expression differences, the miR9564-ARPS couple was selected as autotetraploid rice pollen sterility candidates for further verification.qRT-PCR was performed to verify the expression levels of the four aforementioned coDEM-DEG pairs during anther development of H1, HF, and LF.miR9564 showed upregulated expression in developmental stage 7 to 10 (S7-S10) anthers of HF and H1 compared to LF, while its target ARPS showed down-regulated expression during the same stage.Similarly, miR528 and miR27874 displayed up-regulated expression in S7-S10 anthers of HF and H1, while their targets M2T and M3T showed down-regulated expression during the same stage.Another cuDEM, miR818d, was confirmed with up-regulated expression in S8a-S8b anthers of HF and H1, but its potential target M4T was also up-regulated in S8a anthers.M4T1 encoded an expressed protein without known function and was mainly highly expressed in vegetative organs.Opposite expression patterns were found in miR9564-ARPS, miR528-M2T, and miR27874-M3T, but not in miR818d-M4T (Figure 6B).Because of its most substantial expression differences, the miR9564-ARPS couple was selected as autotetraploid rice pollen sterility candidates for further verification.

miR9564 Could Negatively Regulate the Expression Level of ARPS
miR9564 was 24 nt long and its precursor MIR9564 was 92 nt long on chromosome 11.A re-sequencing analysis revealed different haplotypes of MIR9564 between HF-H1 and LF.The haplotype in HF and H1 (labeled as Ha1) contained two SNPs, including the 10th G:C and the 27th C:T, while the haplotype in LF (labeled as Ha2) was the same as the reference genome (Figure 7A,B).The mRNAs of Ha2-MIR9564 were predicted to form three RNA secondary structures different from that of Ha1-MIR9564 (Figure 7C).Interestingly, previously finished RNA-seq and re-sequencing datasets revealed that 33 samples with Ha2-MIR9564 and 22 samples with heterozygous MIR9564 displayed a high expression level of ARPS, but not in 29 samples with Ha1-MIR9564 (Figure 7D).

miR9564 Could Negatively Regulate the Expression Level of ARPS
miR9564 was 24 nt long and its precursor MIR9564 was 92 nt long on chromosome 11.A re-sequencing analysis revealed different haplotypes of MIR9564 between HF-H1 and LF.The haplotype in HF and H1 (labeled as Ha1) contained two SNPs, including the 10th G:C and the 27th C:T, while the haplotype in LF (labeled as Ha2) was the same as the reference genome (Figure 7A,B).The mRNAs of Ha2-MIR9564 were predicted to form three RNA secondary structures different from that of Ha1-MIR9564 (Figure 7C).Interestingly, previously finished RNA-seq and re-sequencing datasets revealed that 33 samples with Ha2-MIR9564 and 22 samples with heterozygous MIR9564 displayed a high expression level of ARPS, but not in 29 samples with Ha1-MIR9564 (Figure 7D).
To verify if the expression level of ARPS is repressed by miR9564, a Dual-GFP assay was further performed, which used a nuclear-localized GFP signal as the internal control (Figure S6).In the control, strong ARPS-GFP fluorescence was distributed in both the nucleus and cytoplasm of protoplasts.While co-expressing the Ubi::MIR9564 effecter (pOX9564) and APRS-DGFP reporter, APRS-GFP signals were repressed in the cytoplasm, and only nuclear GFP signals were observed.Once ARPS was mutated in the target site, the pOX9564 effecter lost its ability to repress the fluorescence signal of ARPSm-GFP (Figure 8).Taken together, these results suggested that miR9564 might negatively affect the expression level of ARPS.To verify if the expression level of ARPS is repressed by miR9564, a Dual-GFP assay was further performed, which used a nuclear-localized GFP signal as the internal control (Figure S6).In the control, strong ARPS-GFP fluorescence was distributed in both the nucleus and cytoplasm of protoplasts.While co-expressing the Ubi::MIR9564 effecter (pOX9564) and APRS-DGFP reporter, APRS-GFP signals were repressed in the cytoplasm, and only nuclear GFP signals were observed.Once ARPS was mutated in the target site, the pOX9564 effecter lost its ability to repress the fluorescence signal of ARPSm-GFP (Figure 8).Taken together, these results suggested that miR9564 might negatively affect the expression level of ARPS.

Over-Expression of ARPS Reduced Pollen Fertility and Seed Set in Neo-Tetraploid Rice
To further understand the reproductive roles of ARPS, three over-expression lines of ARPS driven by the Ubique promoter (ARPS-OE1, ARPS-OE2, and ARPS-OE3) were constructed in the H1 background (Figure 9A).ARPS showed 6856.03~37238.58-foldup-regulation in obtained ARPS-OE lines (Figure 9B).The plant height (14.05-22.85%reduction), panicle length (2.47-11.95%reduction), and grain number per panicle (0.00-39.56% reduction) slightly reduced relative to WT plants (Figure 9C-E), which is consistent with the difference between HF and LF (Figure S1C-E).As expected, the seed setting rates of ARPS-OE1 (45.68%) and ARPS-OE3 (64.12%) were significantly lower than that of WT plants (77.27%) (Figure 9F,G).Moreover, abundant abortive pollen grains were found in ARPS-OE plants.The pollen fertility of ARPS-OE lines ranged from 48.00% to 53.03%, which was significantly lower than that of WT plants (97.10%) (Figure 9H,I).These results indicate that the up-regulation of ARPS expression in anthers would be detrimental to normal pollen development.

Over-Expression of ARPS Reduced Pollen Fertility and Seed Set in Neo-Tetraploid Rice
To further understand the reproductive roles of ARPS, three over-expression lines of ARPS driven by the Ubique promoter (ARPS-OE1, ARPS-OE2, and ARPS-OE3) were constructed in the H1 background (Figure 9A).ARPS showed 6856.03~37238.58-foldupregulation in obtained ARPS-OE lines (Figure 9B).The plant height (14.05-22.85%reduction), panicle length (2.47-11.95%reduction), and grain number per panicle (0.00-39.56% reduction) slightly reduced relative to WT plants (Figure 9C-E), which is consistent with the difference between HF and LF (Figure S1C-E).As expected, the seed setting rates of ARPS-OE1 (45.68%) and ARPS-OE3 (64.12%) were significantly lower than that of WT plants (77.27%) (Figure 9F,G).Moreover, abundant abortive pollen grains were found in ARPS-OE plants.The pollen fertility of ARPS-OE lines ranged from 48.00% to 53.03%, which was significantly lower than that of WT plants (97.10%) (Figure 9H,I).These results indicate that the up-regulation of ARPS expression in anthers would be detrimental to normal pollen development.
In this study, 62 coDEMs were identified from meiotic anthers of two neo-tetraploid lines to further enrich the understanding about miRNAs related to autotetraploid rice pollen sterility.Moreover, 30 miRNA-target pairs were highlighted because of their opposite expression patterns in both H1/LF and HF/LF comparative RNA-seq analyses.One of these coDEMs, miR528, has been previously reported to regulate pollen intine formation by targeting OsUCL23 to influence flavonoid metabolism [21].The STTM (Short Tandem Target Mimic) lines of miR528 in neo-tetraploid rice exhibited a significant reduction in seed setting and pollen fertility [34].Here, we identified the OsINH2 gene as a novel candidate target of miR528, which functions in regulating pollen viability and grain numbers [35].In both H1/LF and HF/LF, miR528 was up-regulated during pollen development, while OsINH2 showed down-regulation.The altered expression pattern of miR528-OsINH2 from sterile autotetraploid rice to fertile neo-tetraploid rice suggests its potential importance in enhancing pollen fertility.
In this study, 62 coDEMs were identified from meiotic anthers of two neo-tetraploid lines to further enrich the understanding about miRNAs related to autotetraploid rice pollen sterility.Moreover, 30 miRNA-target pairs were highlighted because of their opposite expression patterns in both H1/LF and HF/LF comparative RNA-seq analyses.One of these coDEMs, miR528, has been previously reported to regulate pollen intine formation by targeting OsUCL23 to influence flavonoid metabolism [21].The STTM (Short Tandem Target Mimic) lines of miR528 in neo-tetraploid rice exhibited a significant reduction in seed setting and pollen fertility [34].Here, we identified the OsINH2 gene as a novel candidate target of miR528, which functions in regulating pollen viability and grain numbers [35].In both H1/LF and HF/LF, miR528 was up-regulated during pollen development, while OsINH2 showed down-regulation.The altered expression pattern of miR528-OsINH2 from sterile autotetraploid rice to fertile neo-tetraploid rice suggests its potential importance in enhancing pollen fertility.
OsRPC53 encodes a subunit of RNA polymerase C (III), which is required for the hybrid pollen sterility, grain length, and grain number in rice [36,37].The mutation of OsRPC53 caused pollen sterility in interspecific hybrid progeny of cultivated rice and Asian annual wild rice (Oryza nivara) [36].In this study, a negative relationship in expression between miR27874 and OsRPC53 was identified and verified by qRT-PCR.miR27874 showed high-level expression in developing anthers of H1 and HF, but not in LF, while its target, OsRPC53, maintained a high expression level in LF anthers.These results suggest that miR27874-OsRPC53 might play important roles in regulating pollen fertility of neo-tetraploid rice.
In rice, STS1 encodes an endoplasmic reticulum-localized protein with lipase activity, essential for tapetal degeneration and pollen wall formation [38].In this study, the expression of STS1 was up-regulated in both H1 and HF compared to LF, while its regulator miRNA, osa-MIR5522-p3, was down-regulated.These results suggested that the expression level of STS1 may be insufficient for post-remodeled reproduction in autotetraploid rice, leading to partial pollen sterility.However, the down-regulated expression of osa-MIR5522-p3 might facilitate the rescue of a higher expression level of STS1 during pollen development in neo-tetraploid rice, contributing to its high pollen fertility.
miR9564 was initially identified in the flowers of Brassica rapa, but its function remained unknown [39].Here, we identified the homologous miR9564 from the meiosis anther of H1 and HF, which displayed a high expression level in developing anthers of H1 and HF, but not in LF.Conversely, its predicted target (ARPS) maintained a high expression level only in LF anthers.Moreover, the Dual-GFP assay revealed that the expression of MIR9564 could inhibit the fluorescence signals of ARPS-GFP, but not in mutant ARPS-GFP.These results indicate that ARPS is a target of miR9564.ARPS is a homologous gene of AtULP1a from Arabidopsis, which encodes an ESD4-like SUMO (small ubiquitin-related modifier) protein kinase 1 [40].SUMO is a micromolecule with a similar structure to ubiquitin molecules and participates in protein modification after translation.It has been found that the SUMO genes regulate the nitrogen homeostasis, grain size, anther dehiscence, pollen fertility, and seed set in rice [41,42].In this study, two haplotypes of MIR9564 were identified, in which one haplotype related to H1 and HF exhibited a low expression level of ARPS.Furthermore, the over-expression of ARPS was detrimental to pollen fertility and seed setting in neo-tetraploid rice.These results suggested that the high expression of miR9564 may suppress ARPS expression, preventing the partial autotetraploid pollen sterility caused by polyploidization in neo-tetraploid rice.

Plant Material
Anthers during meiosis of two fertile neo-tetraploid lines, Huaduo1 (H1) and High-Fertility Tetraploidy (HF), and one low-fertility tetraploid line (LF), were used for miR-NAs' analysis.HF and LF were sister lines derived from the progenies of Linglun-4x (infertile) × H1.H1 was the receptor for over-expression lines of ARPS.

Cytological Observation
Three florets from each of the three plants for each line were collected for evaluation on pollen fertility or viability.Pollen grains fixed in a Carnoy solution (ethanol/acetic acid = 3:1 v/v) for a minimum of 24 h were stained by a 1% iodine-potassium iodide solution (w/v) to evaluate pollen fertility.Pollen viability was evaluated with live pollen grains subjected to a 1% 2,3,5-Triphenyl-2H-tetrazolium chloride (TTC, w/v) solution at 31 • C. Subsequently, viable pollen grains turned red, while devitalized pollen grains retained their original color (Figure S7).Each sample was then photographed, obtaining five pictures from different perspectives under a Motic BA200 microscope, to count the pollen grains.
Florets were collected and fixed in the Carnoy solution for at least 24 h.Pollen mother cells were isolated and placed into a 1% acetocarmine solution for 3-5 min.Slides were briefly exposed to a flame for 5-7 s to induce thermal expansion and squashed to obtain separated meiotic chromosomes.Meiotic chromosomes were observed and photographed under a Motic BA200 microscope.The quantification of chromosome behaviors and chromosome configurations was performed as described in our previous study [18].

miRNA Analysis
RNA samples were extended from anthers during meiosis, which were used for RNAseq [33].Small RNA libraries were constructed by Illumina's TruSeq small RNA sample preparation Kits (San Diego, CA, USA).The miRNA sequencing was performed by LC-BIO (Hangzhou, China).The ACGT101-miR program was used to eliminate junk, adapter dimers, common RNA families (snoRNA, snRNA, tRNA, rRNA), low complexity, and repeats from raw reads.Subsequently, unique sequences (18-25 nt in length) were aligned against in miRBase 21.0 (ftp://mirbase.org/pub/mirbase/(accessed on 21 November 2018)) to identify known or derived miRNAs.Unmapped sequences were used to predict novel miRNAs using BLAST tools and RNAfold software (http://rna.tbi.univie.ac.at/ cgi-bin/RNAWebSuite/RNAfold.cgi(accessed on 9 November 2018)).Normalized deepsequencing counts represented the expression level of miRNA.MiRNAs with log 2 (fold change ratio) > 1 and p-value < 0.05 were considered as differentially expressed miRNAs (DEMs).miRNAs with a maximum expression level among all samples less than 10.00 were categorized as low-expression miRNAs, those with a maximum expression level higher than the average expression level of all miRNAs were considered as high-expression miRNAs, and the remaining miRNAs were categorized as middle-expression miRNAs.

Bioinformatics Analysis Tools
The RNA structure of the precursor of miRNA was predicted by Mfold [43].psR-NATarget was employed to identify targeted mRNA of DEMs [44].The targeted genes were annotated on the National Rice Data Center website (http://www.ricedata.cn/gene(accessed on 8 April 2019)).Gene expression levels of ARPS were analyzed using our 84 RNA-seq datasets, as described in our previous study [33].The Gene Ontology (GO) enriched terms of miRNA targets were analyzed by using AgriGO (https://systemsbiology. cau.edu.cn/agriGOv2/(accessed on 14 October 2018)).Venn diagram analyses and a heatmap diagram were created by TBtools [45,46].

Quantitative Real-Time Polymerase Chain Reaction (qRT-PCR) Analysis
The total RNA samples were extracted as templates.The Transcroptor First Strand cDNA Synthesis Kit (Roche, Mannheim, Germany) was used for reverse transcription.The cDNA of targeted genes was typically reverse-transcribed following the manufacturer's instructions of the kit.Ubiquitin was used as the internal control gene for qRT-PCR of candidate genes.For the reverse transcription of DEMs, miRNA-specific stem-loop RT

Figure 4 .
Figure 4. Frequency of abnormal chromosome behavior and quadrivalent distribution in meiotic pollen mother cells.(A) Frequency of abnormalities in LF and HF during different stages of meiosis.(B) Average of quadrivalents in LF and HF at diakinesis and metaphase Ⅰ. (C) Frequency of different types of quadrivalents at diakinesis and metaphase Ⅰ. "Ring", "Chain", "Frying pan", "OK", and "X" indicate different-type quadrivalents shown in Figure 3D-F.n, number of pollen mother cells observed.

Figure 4 .
Figure 4. Frequency of abnormal chromosome behavior and quadrivalent distribution in meiotic pollen mother cells.(A) Frequency of abnormalities in LF and HF during different stages of meiosis.(B) Average of quadrivalents in LF and HF at diakinesis and metaphase I. (C) Frequency of different types of quadrivalents at diakinesis and metaphase I. "Ring", "Chain", "Frying pan", "OK", and "X" indicate different-type quadrivalents shown in Figure 3D-F.n, number of pollen mother cells observed.
Plants 2024, 13, x FOR PEER REVIEW 8 of 17 378 were considered as moderately expressed.The 422 low-expressed miRNAs were subsequently excluded from the further analysis (Figure5D).

Figure 5 .
Figure 5. Identification of miRNAs from meiotic anthers of H1, HF, and LF.(A) Number of miRNAs detected in H1, HF, and LF.Yes, miRNAs detected in miRBase.Diff, miRNAs detected in miRBase with differing sequence.New, miRNAs not present in miRBase.(B) Venn diagram analysis of miR-NAs among H1, HF, and LF.(C) Length distribution of unique miRNAs.(D) Number of miRNAs with different expression levels, categorized as High, Middle, and Low.(E) Number of differentially expressed miRNAs (DEMs) in H1 and HF; (F) identification of coDEMs between H1 and HF; (G) hierarchical cluster analysis among coDEMs; (H) upset plot analysis of coDEM targets and coDEGs to identify negative regulatory groups.DEGs, differentially expressed genes.

Figure 5 .
Figure 5. Identification of miRNAs from meiotic anthers of H1, HF, and LF.(A) Number of miRNAs detected in H1, HF, and LF.Yes, miRNAs detected in miRBase.Diff, miRNAs detected in miRBase with differing sequence.New, miRNAs not present in miRBase.(B) Venn diagram analysis of miR-NAs among H1, HF, and LF.(C) Length distribution of unique miRNAs.(D) Number of miRNAs with different expression levels, categorized as High, Middle, and Low.(E) Number of differentially expressed miRNAs (DEMs) in H1 and HF; (F) identification of coDEMs between H1 and HF; (G) hierarchical cluster analysis among coDEMs; (H) upset plot analysis of coDEM targets and coDEGs to identify negative regulatory groups.DEGs, differentially expressed genes.

Figure 6 .
Figure 6.Expression analyses of four key coDEMs and coDEGs in developing anthers.Expression levels of miR9564-ARPS, miR528-M1T, miR27874-OsRPC53, and miR818d-M4T were analyzed using RNA-seq data (A) and qRT-PCR (B) in H1, HF, and LF anthers.Fc, fold change relative to sample with lowest expression.Ubiquitin was used as internal control gene.S7 to S10 represent anther stages.Error bars indicate standard error (SE) with n = 3.

Figure 7 .
Figure 7.The relationship between MIR9564 haplotypes and ARPS expression levels.(A,B) Information about the precursor of miR9564 (MIR9564).Ha1 and Ha2 indicate two haplotypes of MIR9564.(C) Predicted RNA secondary structures of Ha1-MIR9564 and Ha2-MIR9564; (D) expression levels of ARPS in plants with different MIR9564 haplotypes.Het indicates the heterozygote of Ha1 and Ha2.N, indicated numbers of RNA-seq samples for expression analyses of ARPS.

Figure 7 .
Figure 7.The relationship between MIR9564 haplotypes and ARPS expression levels.(A,B) Information about the precursor of miR9564 (MIR9564).Ha1 and Ha2 indicate two haplotypes of MIR9564.(C) Predicted RNA secondary structures of Ha1-MIR9564 and Ha2-MIR9564; (D) expression levels of ARPS in plants with different MIR9564 haplotypes.Het indicates the heterozygote of Ha1 and Ha2.N, indicated numbers of RNA-seq samples for expression analyses of ARPS.

Plants 2024 , 17 Figure 8 .
Figure 8. Dual-GFP assay for negative regulation between MIR9564 and ARPS.(A) Schematic diagrams illustrating information about effectors and reporters.NLS, nuclear localization signal; DG, Dual-GFP.(B) Subcellular localization of Dual-GFP signal for testing MIR9564 regulation on ARPS-GFP activity.(C) Frequency of protoplasts with different types of GFP signals.N, number of observed protoplasts.

Figure 8 .
Figure 8. Dual-GFP assay for negative regulation between MIR9564 and ARPS.(A) Schematic diagrams illustrating information about effectors and reporters.NLS, nuclear localization signal; DG, Dual-GFP.(B) Subcellular localization of Dual-GFP signal for testing MIR9564 regulation on ARPS-GFP activity.(C) Frequency of protoplasts with different types of GFP signals.N, number of observed protoplasts.
Ubiquitin was used as internal control gene.S7 to S10 represent anther stages.Error bars indicate standard error (SE) with n = 3.