Auxin apical dominance governed by the OsAsp1-OsTIF1 complex determines distinctive rice caryopses development on different branches

In rice (Oryza sativa), caryopses located on proximal secondary branches (CSBs) have smaller grain size and poorer grain filling than those located on apical primary branches (CPBs), greatly limiting grain yield. However, the molecular mechanism responsible for developmental differences between CPBs and CSBs remains elusive. In this transcriptome-wide expression study, we identified the gene Aspartic Protease 1 (OsAsp1), which reaches an earlier and higher transcriptional peak in CPBs than in CSBs after pollination. Disruption of OsAsp1 expression in the heterozygous T-DNA line asp1-1+/–eliminated developmental differences between CPBs and CSBs. OsAsp1 negatively regulated the transcriptional inhibitor of auxin biosynthesis, OsTAA1 transcriptional inhibition factor 1 (OsTIF1), to preserve indole-3-acetic acid (IAA) apical dominance in CPBs and CSBs. IAA also facilitated OsTIF1 translocation from the endoplasmic reticulum (ER) to the nucleus by releasing the interaction of OsTIF1 with OsAsp1 to regulate caryopses IAA levels via a feedback loop. IAA promoted transcription of OsAsp1 through MADS29 to maintain an OsAsp1 differential between CPBs and CSBs during pollination. Together, these findings provide a mechanistic explanation for the distributed auxin differential between CPBs and CSBs to regulate distinct caryopses development in different rice branches and potential targets for engineering yield improvement in crops.

Introduction Rice (Oryza sativa) is a staple food for approximately two-thirds of the global population. Due to the application of dwarf breeding and hybrid breeding in the past 50 years, rice grain yield has more than doubled in most parts of the world, and even tripled in certain countries and regions [1]. However, these newer rice cultivars have not entirely fulfilled their high-yield potential due to smaller grain size and poorer grain filling in caryopses located on proximal secondary branches (CSBs) than in those of apical primary branches (CPBs). This problem is exacerbated in recently bred 'super' rice cultivars [2].
Rice caryopsis development begins with a double-fertilisation event, following which endosperm mainly contributes to grain size and weight in two continuous developmental progresses: early-stage endosperm cell proliferation and differentiation, and late-stage grain filling. Extensive studies have shown that grain size and weight on a single spikelet are regulated by the histone acetylation and deacetylation pathway [3], the ubiquitin-proteasome pathway [4][5][6], G-protein signalling [7,8], MAPK signalling [9], and phytohormone signalling, such as that involving brassinosteroids [10][11][12] or auxins [13,14], which influence spikelet hull and endosperm growth. Grain filling is associated with carbohydrate supply, activity of the enzymes involved in starch metabolism, levels of various endogenous hormones, and environmental conditions [15][16][17][18]. Studies using omics approaches have revealed differences in the expression of various proteins [19], in the expression of genes related to starch synthesis and hormone signalling [20], and in miRNA levels between CPBs and CSBs during the middle and late stages of grain filling after fertilisation [21,22]. However, the molecular mechanisms determining the developmental differences between CPBs and CSBs remain elusive. In this transcriptome-wide expression study, we sought to identify the key genes that are differentially expressed in post-anthesis CPBs and CSBs, and further characterised the role of one of these genes, Aspartic Protease 1 (OsAsp1), in determining the distinctive development of CPBs and CSBs in rice.

Transcriptome-wide expression analysis of CPBs and CSBs to identify OsAsp1
To identify the key factors that determine the developmental differences between CPBs and CSBs in rice, we performed a transcriptome-wide expression analysis of CPB and CSB samples obtained at four different dates. Samples were chosen separately for each branch type to ensure morphological similarity of CPBs and CSBs, given that primary branches develop earlier than secondary branches. Specifically, samples were collected at 0, 5, 12, and 20 days after heading (DAH) for CPBs (CPB-0, -5, -12, and -20) and 5, 12, 25, and 35 DAH for CSBs (CSB-5, -12, -25 and -35) (S1A Fig).

PLOS GENETICS
The gene expression profiles formed two separate clusters, one consisting of CPB-0, CSB-5, CPB-5, and CSB-12 and the other of CPB-12, CSB-25, CPB-20, and CSB-35, among which two pairs, CPB-5/CSB-12 and CPB-20/CSB-35, were closer than the others (Fig 1A). This pattern is consistent with the existence of two continuous phases of caryopsis development: endosperm cell proliferation, which mainly contributes to grain enlargement, followed by grain filling. We also found that CPB endosperm had a higher cell reproduction rate and faster grain filling rate than that of CSBs (S1B and S1C Fig), which is consistent with the findings of previous studies [17,23]. We propose that the difference in endosperm cell proliferation rates between CPBs and CSBs is the critical factor for grain production because it determines the sink capacity for caryopsis. Therefore, we further compared the genes that were more highly expressed in CPB-0 and CSB-12 than in CSB-5 and CPB-5 ( Fig 1A) with the results of the gene expression ratio distribution analysis, based on reads per kb per million mapped reads (RPKM) values. Then, we identified candidate genes with expression ratios higher than 4.8 in both comparisons (Fig 1B and S1 Table) and used quantitative reverse-transcription polymerase chain reaction (qRT-PCR) to confirm the relative transcript levels of the top three genes (S2 Fig): Os07g39020, encoding the SUBTILISIN-LIKE PROTEASE OsSUB53; Os06g41030,  Fig 1A) based on the reads per kb per million mapped reads (RPKM) values for CPB-0 and CSB-5 (red line) and CSB-12 and CPB-5 (green line). � Ratio distribution of a significant coincidence for genes of CPB-0 vs. CSB-5 and CSB-12 vs. CPB-5, centred on a value of 4.8 (� 4.8). C, Schematic representation of the two OsAsp1 T-DNA insertions found in the asp1-1 and asp1-2 mutant rice lines. D, Identification of the genotypes of asp1-1 and asp1-2 by polymerase chain reaction (PCR). Wild-type (WT) rice was used as a control. LP, LB, and RP primers used for PCR are listed in S4 Table. E, Fertility of the WT, asp1-1 +/-, and asp1-2 +/-. Unfertilised and fertilised seeds were counted for each plant. F, Transcription levels of OsAsp1 detected by quantitative reverse-transcription (qRT)-PCR of CPBs and CSBs in the WT, asp1-1, and asp1-2 at 0, 3, 5, and 10 DAF. Solid and dotted lines indicate the transcript levels of OsAsp1 in CPBs and CSBs, respectively. Red, blue, and green lines indicate the transcript levels of OsAsp1 in the WT, asp1-1, and asp1-2, respectively. The expression of OsAsp1 was normalised to that of OsActin1, and relative expression in CPBs in the WT at 0 DAF was set at 1.0. qRT-PCR data are means ± standard deviation (SD) of three independent experiments. G, Dry weights of developed grain from CPBs and CSBs in one whole plant of the WT, asp1-1 +/-, and asp1-2 +/-. Values are means ± SD (n = 30). Different letters indicate significant differences (P < 0.05, Student's t-test).

Disrupting OsAsp1 eliminated CPB-CSB developmental differences
To explore the role of OsAsp1 in caryopsis development, we obtained two OsAsp1 T-DNA mutation lines from the Rice Functional Genomic Express Database (RiceGE, http://signal. salk.edu): asp1-1 (PFG_3A-60042) and asp1-2 (PFG_3A-04889) (Fig 1C). Through self-crossing and back-crossing with wild-type (WT) plants, we confirmed that both asp1-1 and asp1-2 are single-insertion and homozygous-lethal mutants, because about 75% of spikelets developed into fertilised seeds and none seeds of these seeds was homozygous (S2 Table, Fig 1D and 1E). We further detected OsAsp1 transcript levels in CPBs and CSBs of the WT, asp1-1 +/-, and asp1-2 +/on four different dates, and found that OsAsp1 expression peaked in CPBs at 3 days after flowering (DAF) and in CSBs at 5 DAF in the WT, and that the peak OsAsp1 transcript level of WT CPBs at 3 DAF was higher than that of WT CSBs at 5 DAF ( Fig 1F). However, in both CPBs and CSBs of asp1-1 +/and asp1-2 +/-, OsAsp1 expression was stable and expression values at different DAF were similar to those at 0 DAF in both CPBs and CSBs of the WT ( Fig  1F). These results indicate that T-DNA insertion decreased expression of OsAsp1 in CPBs and CSBs in the independent lines asp1-1 +/and asp1-2 +/-. Then, we assessed the developmental parameters of both lines, and found that plant height was lower in both asp1-1 +/and asp1-2 +/ than in the WT, but that spikelet number per panicle in asp1-1 +/was similar to that in the WT and greater than that in asp1-2 +/-(S3 Fig). Of note, compared to the WT, the difference in the grain weight of the developed seeds between CPBs and CSBs disappeared in both asp1-1 +/ and asp1-2 +/-, whereas seed grain weights in asp1-1 +/and asp1-2 +/were the same or closer to those of CPBs in the WT (S4 Fig and Fig 1G). These results indicate that OsAsp1 is essential for rice growth and caryopsis development; therefore, we focused on asp1-1 +/in the subsequent analysis.
Next, we observed the ovary development of CPBs and CSBs in the WT and asp1-1 +/at 3 and 5 DAF, respectively, and found that at both time points, asp1-1 +/displayed faster ovary development in CSBs than in the WT, and that the ovary size of CSBs in asp1-1 +/was similar to that of CPBs in both the WT and asp1-1 +/- (Fig 2A and S5 Fig). We further used Eosin-B staining to assess the ovary development of CPBs and CSBs in the WT and asp1-1 +/at 3 and 5 DAF, and found that the CSBs in asp1-1 +/showed a higher level of endosperm cellularisation than either CSBs in the WT or CPBs in the WT and asp1-1 +/at both time points (Fig 2B). In addition, CSBs in asp1-1 +/displayed more developed embryos at 5 DAF than either CSBs in the WT or CPBs in the WT and asp1-1 +/- (Fig 2B). Moreover, DNA ploidy analysis of CPBs and CSBs in the WT and asp1-1 +/at 3 and 5 DAF by flow cytometry showed that at both time points, the relative 3C plus 6C DNA contents, which represent the numbers of endosperm cells, were significantly higher in CSBs of asp1-1 +/than in CPBs of asp1-1 +/or either CPBs or CSBs of the WT, indicating that CSBs in asp1-1 +/developed better than did CPBs of asp1-1 +/ or CPBs and CSBs of the WT (Fig 2C). Previous studies have identified Orysa;CycB1;1 (OsCycB1), Orysa;KRP1 (OsKRP1) and the rice cell cycle switch 52A gene (OsCCS52A) as three marker genes associated with cell cycle and division in rice endosperm [25][26][27]. qRT-PCR analysis showed that these three genes displayed differential expression patterns between CPBs and CSBs in the WT and asp1-1 +/at both 3 and 5 DAF (Fig 2D). Together, these data confirm that disruption of OsAsp1 transcription accelerated CSBs' developmental progress, thereby relieving the developmental differences between CPBs and CSBs in asp1-1 +/ -.

OsAsp1 activated TAA1 transcription to upregulate IAA levels
A prior study showed that auxin biosynthesis couples central cell division for endosperm development to fertilisation in Arabidopsis [28]. Therefore, we monitored the levels of indole-3-acetic acid (IAA) in CPBs and CSBs of the WT and asp1-1 +/using high-performance liquid chromatography and found that in the WT, free IAA content was significantly higher in CPBs than in CSBs at 0, 3, and 5 DAF (Fig 3A), which is consistent with a previous report [16]. However, in asp1-1 +/-, there was no difference in IAA levels between CPBs and CSBs; IAA content was very low and almost identical in CPBs and CSBs at 0, 3, and 5 DAF (Fig 3A). In rice, IAA is mainly synthesised via the indole-3-pyruvic acid (IPA) pathway, catalysed by tryptophan aminotransferase (OsTAA) and flavin monooxygenase-like enzymes (OsYUCCAs) [29,30]. As expected, in the WT, expression of OsTAA1 gradually increased after fertilisation, reaching a peak at 5 DAF in both CPBs and CSBs, with a higher peak in CPBs. However, in asp1-1 +/-, the expression peaks of OsTAA1 in both CPBs and CSBs disappeared at 5 DAF, but occurred at 3 DAF. Of note, OsTAA1 expression was lower in CPBs than in CSBs of asp1-1 +/- (Fig 3B). By contrast, the expression patterns of OsYUCCA1, which was more highly expressed in CPBs than in CSBs at 5 and 10 DAF, respectively, were similar in CPBs and CSBs in the WT and asp1-1 +/-, except that expression levels of OsYUCCA1 were lower in CPBs of asp1-1 +/than in the WT at 10 DAF ( Fig 3C). In addition, we heterologously expressed OsAsp1 in DR5::GUS transgenic Arabidopsis thaliana plants, and found upregulated transcription of AtTAA1 and AtYUCCA8, but not AtYUCCA9 ( Fig 3D); GUS activities were also strongly increased in root apical meristem of the main and lateral roots of 3-day-old seedlings ( Fig 3E and 3F). These results imply that OsTAA1 is upregulated by OsAsp1 to preserve IAA levels and ensure normal development of CPBs and CSBs. Values are means ± SD of three independent experiments. B, Transcript levels of OsTAA1 detected by qRT-PCR in CPBs and CSBs of the WT and asp1-1 +/at 0, 3, 5, and 10 DAF. OsTAA1 expression was normalised to that of OsActin1, and relative expression in CPBs of the WT at 0 DAF was set at 1.0. qRT-PCR data are means ± SD of three independent experiments. C, Transcription levels of OsYUCCA1 in CPBs and CSBs of the WT and asp1-1 +/on 5 and 10 DAF. OsYUCCA1 expression was normalised to that of OsActin1, and relative expression of OsYUCCA1 in CPBs of the WT at 5 DAF was set at 1.0. Values are means ± SD of three independent experiments. NS, no significant difference; � 0.01 < P < 0.5; �� 0.001 < P < 0.01 (Student's t-test). D, Expression of OsAsp1, AtTAA1, AtYUCCA8, and AtYUCCA9 in Arabidopsis WT and OsAsp1-overexpressed lines OE-5, OE-15, and OE-19. Target gene expression was normalised to that of AtUBQ10, and relative expression in the WT was set at 1.

OsAsp1 dismissed transcriptional inhibition of OsTIF1 to OsTAA1 to sustain IAA levels
OsAsp1 encodes an aspartic protease [24], implying that it may regulate the transcription of OsTAA1 via an uncharacterised transcriptional regulator. As shown in our flowchart of OsTAA1 transcription regulator prediction in rice (Fig 4A), we first used the Arabidopsis coexpression networks of AtTAA1 (http://atted.jp/) [31] to identify a C 2 H 2 -type zinc-finger protein (At1G75710) as the candidate (S6A Fig  Values are means ± SD of three independent experiments. � 0.01 < P < 0.5; ��� P < 0.001 (Student's t-test). C, Promoter activity of P OsTAA1 under the regulation of OsAsp1 and/or OsTIFs in the protoplast of rice leaves. The P OsTAA1 :GUS construct was transfected along with P UBI :OsAsp1, P UBI :OsTIFs, or P UBI :OsAsp1 and P UBI :OsTIFs, or with empty vector as control. We included 35S:LUC as an internal control. Relative GUS activity normalised to luciferase activity was adopted as the promoter activity of P OsTAA1 . Values are means ± SD of three independent experiments. D, Chromatin immunoprecipitation (ChIP)-qPCR was used to detect the enrichment of OsTIF1-binding motifs at positions -2349, -1748, or -377 nt in the promoter region of OsTAA1. OsTIF1-HA was overexpressed in rice calli and purified with HA antibody. NC, negative control. Values are means ± SD of three independent experiments. E, EMSA showing recombinant OsTIF1-His directly binding the promoter region of OsTAA1 at its potential motifs. The predicted OsTIF1 binding motifs at positions -2349, -1748, or -377 nt were labelled with biotin as the probe, and the unlabelled DNA fragments were used as the competitor. Red and black arrows indicate the shifted band and free probe, respectively. F, and G, Transcript levels of OsAsp1, OsTIF1, and OsTAA1 in rice calli overexpressing OsAsp1-GFP (F) or OsTIF1-HA (G). The expression of each target gene was normalised to that of OsActin1, and calli transformed with empty vector were used as a control. qRT-PCR data are means ± SD of three independent experiments.
https://doi.org/10.1371/journal.pgen.1009157.g004 we identified two candidate genes, Os04g59380 and Os04g02510, which we named OsTAA1 Transcription Inhibition Factor 1 (OsTIF1) and OsTIF2, respectively. We further detected the transcription levels of OsTIF1 and OsTIF2 in CPBs and CSBs of the WT and asp1-1 +/-, and found that expression of both genes was higher in both CPBs and CSBs of asp1-1 +/than in the WT (Fig 4B). We transiently co-transformed P OsTAA1 :GUS with either OsAsp1 and OsTIF1 or OsTIF2 (S7A Fig) in rice mesophyll protoplasts, and found that both OsTIF1 and OsTIF2 inhibited the transcription activity of P OsTAA1 , and OsTIF1 more strongly (Fig 4C). In addition, OsAsp1 relieved the inhibitory effects of OsTIF1 or OsTIF2 on the activity of P OsTAA1 in rice mesophyll protoplasts ( Fig 4C).
Next, according to the target sequences of C 2 H 2 -type transcription factors reported previously [32], three potential OsTIF1 binding sites at -2349, -1748, and -377 nt were predicted in the 5' region upstream from the ATG site of OsTAA1 (S7B Fig). Chromatin immunoprecipitation (ChIP)-qPCR showed that sites at positions -2349 and -1748 nt, but not at -377 nt, were enriched by OsTIF1 in transgenic rice calli (Fig 4D). An electrophoretic mobility shift assay (EMSA) showed that recombination OsTIF1-His could separately bind these three sites to produce the band shift visible in Fig 4E. We also found that overexpression of OsAsp1 effectively upregulated transcription of OsTAA1 (Fig 4F), whereas overexpression of OsTIF1 strongly downregulated expression of OsTAA1 in rice calli ( Fig 4G). However, overexpression of OsAsp1 and OsTIF1 in rice calli slightly activated transcription of OsTIF1 and OsAsp1, respectively (Fig 4F and 4G). These results demonstrate that OsAsp1 activates OsTAA1 transcription by relieving the inhibition of OsTIF1 to ensure IAA biosynthesis.

IAA induced OsTIF1 ER-nucleus transfer by reducing its interaction with OsAsp1 to form a feedback loop
We further investigated the possibility of an interaction between OsAsp1 and OsTIF1 to characterise the regulation mechanism of OsTAA1 transcription. First, after separately co-transforming OsAsp1-GFP with an ER marker, OsTIF1-GFP with an ER marker, and OsAsp1-GFP with OsTIF1-mCherry, we showed that OsAsp1 and OsTIF1 were co-localised in the ER of rice mesophyll protoplasts (Fig 5A), which raised the question of how OsTIF1 enters the nucleus to inhibit OsTAA1 expression. Due to the involvement of OsTAA1 in IAA formation in rice caryopses, we treated rice mesophyll protoplasts transformed with OsAsp1 and/or OsTIF1 with 50 μm IAA, and found that OsTIF1 was transferred into the nucleus, but that OsAsp1 remained localised in the ER after IAA treatment ( Fig 5B). Then, using a yeast twohybrid assay, we showed that OsAsp1 and OsTIF1 separately formed homodimers and interacted with each other (Fig 5C). Furthermore, IAA inhibited the interaction between OsAsp1 and OsTIF1 in the yeast two-hybrid system (Fig 5C). A co-immunoprecipitation (Co-IP) assay also demonstrated that OsAsp1-GFP interacted with OsTIF1-HA in rice mesophyll protoplasts and that their interaction was significantly depressed upon the addition of IAA (Fig 5D). These results indicate that IAA mediates a regulatory feedback loop involving the OsAsp1-OsTIF1 complex to fine-tune IAA levels between CPBs and CSBs during caryopsis development.

OsAsp1 transcription was regulated by auxin-inducible MADS29 after pollination
Our analyses showed that IAA levels were always higher in CPBs than in CSBs from the day of pollination (Fig 3A), which is consistent with the phenomenon of auxin apical dominance in plants. Therefore, we hypothesised that IAA regulates OsAsp1 expression to maintain its differential level between CPBs and CSBs in rice. To test this hypothesis, we first cultured unpollinated spikelets (-1 DAF) in vitro with IAA, and found that transcription of OsAsp1 was induced in a concentration-dependent manner (Fig 6A). In addition, in vitro culture of unpollinated spikelets with 2,4-dichlorophenoxyacetic acid (2,4-D) also induced OsAsp1 expression (S8 Fig). Moreover, OsAsp1 transcription in spikelets of CSBs strongly increased after treatment with IAA at 3 days before fertilisation (-3 DAF) compared with OsAsp1 transcription in spikelets given a mock treatment (methanol) (Fig 6B). The grain weight of mature CSB seeds after IAA treatment was also greater than that of CSB seeds treated with mock solution and similar to that of CPB seeds (Fig 6C and S9 Fig). We further treated CPBs of asp1-1 +/with IAA to restore the gradient of IAA levels between CPBs and CSBs, and found that the grain weight of mature CPB seeds of asp1-1 +/after IAA treatment was greater than that of CSB seeds (S10A Fig), and that this difference was similar to the grain weight difference between CPB and CSB seeds in the WT. In addition, the difference in OsAsp1 expression between CPBs and CSBs in asp1-1 +/was also complemented by IAA treatment of CPBs in asp1-1 +/-(S10B Fig). OsTIF1 transfers from the endoplasmic reticulum (ER) into the nucleus after IAA releases its interaction with OsAsp1. A, OsAsp1 co-localised with OsTIF1 in the ER of rice mesophyll protoplast. The ER marker was ER-rk CD3-959, purchased from the Arabidopsis Stock Center (http://www.arabidopsis.org/). Scale bars, 5 μm. B, OsTIF1, but not OsAsp1, was transferred from the ER into the nucleus in rice mesophyll protoplasts in response to IAA treatment (50 μM). The nuclear marker was AtBZR2(1-158 aa)-mCherry. Scale bars, 5 μm. C, Interaction between OsAsp1 and OsTIF1 detected by yeast two-hybrid assay.-LH indicates SD medium without LEU and TRP, and-LTH indicates SD medium without LEU, TRP, and HIS. D, Interaction between OsAsp1 and OsTIF1 detected by a Co-IP assay. P UBI :OsTIF1-HA was co-transformed with P UBI :OsAsp1-GFP or P UBI :GFP into rice mesophyll protoplast, and then IAA (50 μM) was added, along with OsTIF1-HA as bait to pull down OsAsp1-GFP. Anti-GFP (a-GFP) was used to detect OsAsp1-GFP or green fluorescent protein (GFP), and anti-hemagglutinin (a-HA) was used to detect OsTIF1-HA.
https://doi.org/10.1371/journal.pgen.1009157.g005 To further investigate the molecular mechanism regulating OsAsp1 expression, we identified genes that were co-expressed with OsAsp1 using the Rice Oligo Array Database [33], and found five MADS family members with enriched expression (S3 Table). A prior study showed that MADS29 is induced by IAA and is involved in rice seed development [34]. We found that the expression pattern of MADS29 was similar to that of OsAsp1 during the development of CPBs and CSBs after fertilisation (Fig 6D), and that IAA treatment of CSB spikelets at -3 DAF increased MADS29 transcription compared to that in mock-treated spikelets (Fig 6E). Furthermore, OsAsp1 expression was almost the same and downregulated in the CPBs and CSBs of Antisense-MADS29 transgenic rice lines A2 and A14, respectively, but not in A33. (Fig 6F and  S11 Fig). The difference in mature grain dry weight between CPBs and CSBs was absent in Antisense-MADS29 lines A2 and A14, but not in A33, which was similar to that of asp1-1 +/ - (Fig 6G and S12 Fig). Three MADS protein binding sites, CArG boxes at positions -1412, -1217, and -770 nt were predicted in the 5' region upstream from the ATG site of OsAsp1 ( Fig  6H). We found that MADS29 effectively promoted the activity of P OsAsp1 :GUS in Arabidopsis mesophyll protoplasts (Fig 6I). ChIP-qPCR analysis showed that the CArG box at position -770 nt, but not those at positions -1412 or -1217 nt, was enriched by MADS29-His in transgenic rice calli (Fig 6J). An EMSA further showed that recombinant MADS29-His bound all three of these CArG boxes to produce a specific shifted band (Fig 6K). These results indicate that IAA promotes the expression of OsAsp1 via MADS29 to sustain higher OsAsp1 levels in CPBs than in CSBs during pollination.
Based on these results, we propose a model to explain how OsAsp1 maintains auxin apical dominance to regulate the developmental differential between CPBs and CSBs (Fig 7). In this model, high IAA levels upregulate OsAsp1 transcription via MADS29 at 0 DAF to produce higher OsAsp1 levels in CPBs than in CSBs, which explains why OsAsp1 was identified as differentially expressed in our transcriptome-wide expression analysis. Then, OsAsp1 interacts with OsTIF1 in the ER to dispel OsTIF1 transcriptional inhibition of OsTAA1, causing IAA content to peak at 5 DAF. At the same time, IAA induces transfer of OsTIF1 into the nucleus by reducing its interaction with OsAsp1, causing IAA biosynthesis to decrease, and thereby forming a feedback regulation loop controlling IAA levels. From this model, we conclude that differential IAA levels between CPBs and CSBs are determined by the OsAsp1-OsTIF1 complex, which is essential for the distinct development of CPBs and CSBs.

Discussion
Auxin apical dominance has evolutionary advantages for plant growth and reproduction because it plays an essential role in regulating plant shoot architecture and ensuring the production of at least some high-quality seed within a short period [35,36]. In this study, we demonstrated that differential distribution of auxin determines developmental differences between CPBs and CSBs in rice, causing the grain-filling problem observed in CSBs and thereby reducing total grain production and quality. We found that auxin-dependent transcription of OsAsp1 and interaction of OsAsp1 with OsTIF1 regulate the transcription of OsTAA1 to maintain auxin apical dominance between CPBs and CSBs after fertilisation, providing a potential strategy for enhancing grain size and grain filling and thereby increasing rice yield.
Asps comprise a large endopeptidase family that is widely distributed in all three domains of life [37,38]. The rice genome contains 96 putative OsAsp genes [39], among which atypical and nucellin-like Asps are characterised by the absence of a plant-specific insert and an unusually high number of cysteine residues, which play regulatory roles in biotic and abiotic stress responses, chloroplast metabolism, and reproductive development [40]. A prior study showed that EAT1 (ETERNAL TAPETUM 1), a basic helix-loop-helix transcription factor, promotes tapetal programmed cell death (PCD) by directly regulating the expression of OsAsp25 and OsAsp37 in rice anthers [41]. However, caspase activity of OsAsp25 or OsAsp37 was not detected, indicating that OsAsp25 and OsAsp37 regulate downstream caspase-like protease(s) involved in PCD rather than directly acting on caspase-specific substrates [41]. Rice nucellin gene OsAsp1 is strongly expressed in zygotic embryos at 1-2 DAF [24]. However, the biological function of OsAsp1 remains unclear. Auxin has been found to induce OsMADS29 transcription to promote PCD of the nucellus and nucellar projection during rice seed development [34]. The present study was the first to determine that OsAsp1 is essential for rice caryopse development because both the asp1-1 and asp1-2 T-DNA insertion mutant lines were homozygous-lethal and, compared to the WT, the difference in mature seed grain weight between CPBs and CSBs disappeared in both lines. Furthermore, we showed that auxin regulates expression of OsAsp1 through OsMADS29 to maintain high levels of OsAsp1 in CPBs, but not in CSBs, after fertilisation, and that OsAsp1 then interacts with OsTIF1 and dismisses its transcriptional inhibition for OsTAA1 to promote IAA biosynthesis. In addition, IAA facilitates OsTIF1 translocation from the ER to the nucleus by releasing the interaction of OsTIF1 with OsAsp1 to form a feedback regulation of IAA levels in caryopses. The disruption of OsAsp1 levels in asp1-1 +/and overexpression of OsAsp1 in rice calli activates transcription of OsTIF1, indicating that OsTIF1 expression is regulated by an uncharacterised and sophisticated mechanism. Therefore, our results demonstrate that OsAsp1 functions as a transcriptional regulator to form a complex with OsTIF1 to regulate the differential development of CPBs and CSBs after fertilisation.
As a critical phytohormone, auxin regulates nearly all aspects of plant development including morphogenesis and adaptive responses [42]. Auxin distribution depends on both auxin metabolism (biosynthesis, conjugation, and degradation) and cellular auxin transport [43,44]. Auxin is mainly synthesised in shoot apical meristem to maintain increased shoot apical dominance, which transfers to inflorescence meristem when plants flower. In higher plants such as Arabidopsis, maize, and rice, the IPA pathway, catalysed by TAA and YUCCA, has been shown to be the major auxin biosynthetic pathway [29,30,[45][46][47]. The rice genome contains two OsTAA (synonym of OsTAR) and 14 OsYUCCA genes; increased IAA content during development of rice grains was strongly correlated with the expression of OsTAR1, OsYUCCA9, and OsYUCCA11 [48]. IAA content has been shown to be higher in CPBs than in CSBs at the early grain-filling stage [16], which is consistent with our results. However, few studies have focused on the differential expression of OsTAAs and OsYUCCAs in CPBs and CSBs. In this study, we showed that OsTAA1/OsTAR2 expression gradually increased after fertilisation, reaching a peak at 5 DAF in both CPBs and CSBs in the WT, and with a higher peak in CPBs, which was correlated with the IAA content of CPBs and CSBs in the WT. However, in asp1-1 +/-, expression peaks of OsTAA1 disappeared at 5 DAF, but occurred at 3 DAF in both CPBs and CSBs, indicating that OsTAA1 transcription is regulated by OsAsp1 and that OsTAA1 is a key enzyme for IAA biosynthesis, maintaining differential auxin distribution between CPBs and CSBs. By contrast, the expression pattern of OsYUCCA1, which was more highly expressed in CPBs than in CSBs at 5 and 10 DAF, respectively, was similar in the WT and asp1-1 +/-, except that OsYUCCA1 expression was lower in CPBs in asp1-1 +/than in the WT at 10 DAF. These results imply that OsYUCCA1 is not a main enzyme for IAA biosynthesis in CPBs and CSBs. Therefore, we will aim to detect the expression levels of 13 other OsYUCCAs of CPBs and CSBs in asp1-1 +/in a future study to determine which OsYUCCA is regulated by OsAsp1 and associated with IAA biosynthesis in CPBs and CSBs. The heterologous expression of OsAsp1 in Arabidopsis improved IAA levels in root apical meristem in 3-day-old seedlings, as well as transcription levels of AtTAA1 and AtYUCCA8, but not AtYUCCA9, providing additional support for the detection of expression of other OsYUCCAs in asp1-1 +/-.
In this study, we characterised the effect of differential auxin distribution mediation by OsAsp1 on the developmental difference between CPBs and CSBs. First, we found that two independent T-DNA insertion mutant lines, asp1-1 and asp1-2, present a homozygous-lethal phenotype and that differences in ovary development and size, endosperm cellularisation and development, and mature seed grain weight between CPBs and CSBs disappeared, and were closer to CPBs in asp1-1 +/-. These results indicate that OsAsp1 is critical for grain development and maintenance of the developmental difference between CPBs and CSBs. We further showed that IAA content was very low and almost identical in CPBs and CSBs in asp1-1 +/during the early stages of caryopse development, indicating that low auxin levels are sufficient to trigger embryo and endosperm development in CPBs and CSBs, and promote similar development of CSBs and CPBs in asp1-1 +/-. In vitro treatment of CPBs in asp1-1 +/with IAA phenocopied the expression difference of OsAsp1 and the development difference between CPBs and CSBs in the WT. These results imply that the higher auxin levels in CPBs than in CSBs, but not absolute IAA concentration, are critical for maintaining differential development of CPBs and CSBs. Treatment of CSBs in the WT with extra IAA before pollination promoted the development of CSBs, allowing the grain weight of CSBs to increase to that of CPBs, which provides strong evidence for differential IAA distribution, but not concentration, determining the developmental difference between CPBs and CSBs in rice. Auxin is produced post-fertilisation in seeds, which drives central cell division and is necessary for MADS-box transcription factor AGL62 (AGAMOUS-LIKE 62)-dependent endosperm development in Arabidopsis [28]. In the present study, we showed that auxin upregulates OsAsp1 transcription via MADS29 after fertilisation to produce higher levels of OsAsp1 in CPBs than in CSBs. The interaction of OsAsp1-OsTIF1 regulated by auxin formed a feedback regulation loop to maintain auxin apical dominance during rice caryopsis development. Therefore, our results present a mechanistic explanation for differential distribution of auxin between CPBs and CSBs to regulate distinct caryopse development patterns in rice.

Plant materials and growth conditions
The WT rice (Oryza sativa L. subspecies japonica) varieties used in this study were Zhong Hua 15, for RNA sequencing (RNA-seq) and auxin treatment; Zhong Hua 11 for A-MADS29 lines; and Dong Jin for asp1-1 and asp1-2. Two T-DNA mutant lines, PFG_3A-60042 and PFG_3A-04889, were obtained from http://signal.salk.edu, named asp1-1 and asp1-2, respectively, and confirmed by genomic PCR and RT-PCR. Both rice lines were cultivated in a field at Beijing Normal University during the natural growing season for observation of their growth phenotypes during the vegetative and reproductive periods.

RNA-seq and data analysis
Caryopses were collected from growing Zhong Hua 15 rice at different stages after heading for total RNA extraction. Libraries for each sample were constructed using a Next Multiplex RNA Library Prep Set (NEB) and sequenced on an Illumina HiSeq 2000 Platform. Clean reads were mapped to the rice genome (version 7.0) [49] using Cufflink software (http://cufflinks.cbcb. umd.edu/). For gene expression analysis, the numbers of reads per kilobase per million mapped reads (RPKM) were calculated. Differentially expressed genes were defined as having a fold change of � 2 or � 0.5 and a false discovery rate (FDR) � 0.05.

qRT-PCR and semi-quantitative RT-PCR
Total RNA was separately extracted from caryopses at different developmental stages, transgenic rice calli, and 5-day-old Arabidopsis seedlings using TRIzol reagent (Invitrogen, USA) and purified using a PureLink RNA Mini Kit (Invitrogen) and a PureLink DNase Kit (Invitrogen), according to the manufacturer's protocol. Approximately 2 μg RNA was reverse transcribed using a Reverse-Aid First Strand cDNA Synthesis Kit (Thermo Fisher Scientific, USA) and used as template for RT-PCR and RT-qPCR. PCR reactions were carried out in a MyCycler thermal cycler (Bio-Rad, USA) using extension cycling conditions of 94˚C for 30 s, 55˚C for 30 s, and 72˚C for 30 s. The amplifications were performed for 28-33 cycles. qRT-PCR was performed using a 7500 Fast Real-Time PCR System (Applied Biosystems, USA) with Power SYBR Green PCR Master Mix (Applied Biosystems). The thermal program was 2 min at 50˚C, 10 min at 95˚C, followed by 40 cycles of 15 s at 95˚C and 60 s at 60˚C. The dissociation curve program was used to confirm the specificity of the target amplification product. Three independent biological replicates were performed for RT-qPCR analyses. OsActin1 or AtUBQ10 was used as an internal control in both qRT-PCR and RT-PCR analyses. The primers for the target genes are listed in S4 Table.

Eosin staining
Whole-mount stain-clearing laser-scanning confocal microscopy (WCLSM) was used to evaluate the structure of mature embryo sacs of WT and asp1 +/mutant rice. The spikelets of WT and asp1 +/mutants were collected, fixed, hydrated, stained in Eosin Y water solution, dehydrated, and cleared as previously described [50]. The developmental phenotype of the embryo sacs was examined using confocal laser-scanning microscopy (LSM 700, Zeiss, Germany).

Flow cytometry analysis
Ovaries of WT and asp1-1 +/rice sampled at 3 or 5 DAF were chopped with a razor blade (six ovaries per group). We added 500 μL GS buffer (45 mM MgCl 2 , 30 mM sodium citrate, 20 mM MOPS, 0.1 v/v Triton X-100) to each group and the mixture was filtered by passing it through a sieve (48 μm 2 mesh). DNA was stained by adding 10 mM propidium iodide. The DNA content of the nuclei was measured using a flow cytometer (BD FACSAria II; BD Biosciences) as previously reported [51].

Liquid chromatography-mass spectrometry (LC-MS) analysis
IAA content was quantified using LC-MS as previously reported [52]. Fresh caryopses were collected and ground into powder using liquid nitrogen. Approximately 100 ± 5 mg powder was extracted using 1.5 mL methanol. The supernatant was lyophilised in a freeze-drier (Thermo Fisher Scientific) and then resuspended in 85% (v/v) methanol in water for LC-MS analysis. IAA (Sigma-Aldrich) was used as a standard to prepare the standard curve.

EMSA analysis
The coding sequences (CDSs) of MADS29 and OsTIF1 were separately amplified and cloned into the vector pET32a. MADS29-His and OsTIF1-His were expressed in Escherichia coli (BL21) and affinity-purified using Ni-NTA Resin. Synthesised fragments of the DNA probes (S4 Table) were labelled using a Biotin 3'-End DNA Labeling Kit according to the manufacturer's instructions (Thermo Fisher Scientific). The protein-DNA binding reactions were performed using a LightShift Chemiluminescent EMSA Kit according to the manufacturer's instructions (Thermo Fisher Scientific).

ChIP-qPCR
Ubi1:OsTIF1-HA and Ubi1:MADS29-GFP were transformed into rice callus by particle bombardment using a PDS-1000/He biolistic particle delivery system (Bio-Rad) as described previously [53]. ChIP of OsTIF1-HA and MADS29-GFP was performed as previously described, with minor modifications [54]. Briefly, 1 g transgenic callus was ground to a fine powder with liquid nitrogen and combined with 10 mL ChIP extraction buffer (0.4 M sucrose, 10 mM Tris-HCl, 10 mM MgCl 2 , 1 mM DTT, 0.1 mM PMSF and protease inhibitor cocktail, pH 8.0). Then, 270 μL 37% formaldehyde solution (to a final concentration of 1%) was added and the mixture was incubated at 4˚C for 10 min to cross-link DNA to protein. The cross-linking reaction was quenched by adding 0.63 mL 2 M glycine, and the mixture was incubated at 4˚C for 5 min. Finally, the nuclear pellet was isolated by centrifugation and resuspended with ChIP lysis buffer (50 mM Tris-HCl, 10 mM EDTA, and 1% sodium dodecyl sulfate [SDS], pH 8.0) and kept on ice for 30 min. One volume of ChIP dilution buffer (16.7 mM Tris-HCl, 167 mM NaCl and 1.1% Triton X-100, pH 8.0) was added to the samples before sonication for 12 min (30 s on, 30 s off, high level) in a Bioruptor (Diagenode) to yield DNA fragments of 0.2-1.0 kb in length. The lysates were diluted fivefold in ChIP dilution buffer to decrease the concentration of SDS to 0.1% and cleared by centrifugation (16,000 × g for 5 min at 4˚C). After keeping 5% of the sample as an input control, the remaining supernatant was incubated with antibodybound Dynabeads Protein A or G (Invitrogen) overnight at 4˚C. Washing, elution, reverse cross-linking, and DNA purification were performed according to the methods of a previous study [55]. The antibodies used for ChIP were hemagglutinin (HA, Sigma) and green fluorescent protein (GFP, Sigma). DNA isolated by ChIP was used for qPCR analysis. The 100-bp 5'region of OsTAA1 was used as a negative control for OsTIF1, and the 100-bp N-terminal region of GUS was used as a negative control for MADS29. The primers used for qPCR are listed in S4 Table.

Transient transactivation assay
The promoters of OsTAA1 or OsAsp1 were amplified by PCR from rice genomic DNA and cloned into the pGPTV vector to generate the P OsTAA1 :GUS and P OsAsp1 :GUS constructs, respectively. P OsTAA1 :GUS was co-transformed with either 35S:Luc and 35S:OsAsp1-GFP or 35S:OsTIF1-HA into rice mesophyll protoplasts [56,57]. P OsAsp1 :GUS was co-transformed with 35S:Luc and 35S:MADS29-GFP into Arabidopsis mesophyll protoplasts [58]. 35S:Luc was used as an internal control. GUS and luciferase (Luc) activities were measured as previously described [59], and relative GUS activity (GUS/Luc) was calculated to determine the activity of P OsTAA1 or P OsAsp1 .

Yeast two-hybrid assay
The CDS of OsAsp1 or OsTIF1 was subcloned into pGAD7 or pGBK7 (Clontech) as prey and bait constructs, respectively. Then, the pairs of bait and prey constructs were co-transformed into yeast strain AH109 and screened by growth on selective medium (SD-Leu-Trp-His). The same strains were grown on a control medium (SD-Leu-Trp) for use as controls.

Co-IP assays
Ubi1:OsTIF1-HA was co-transformed with Ubi1:OsAsp1-GFP or Ubi1:GFP into rice mesophyll protoplasts, which then were incubated with or without 50 μM IAA for 16 h. Total protein was extracted and immunoprecipitated with GPF or HA antibody incubated with antibody-bound Dynabeads Protein A or G. The immunoprecipitates were separated by 10% SDS-polyacrylamide gel electrophoresis (PAGE) and detected with GFP antibody. Ubi1:HA and Ubi1: OsAsp1-GFP transgenic protoplasts were used as a negative control.

IAA and 2,4-D treatments
Spikelets were collected from panicles at -1 DAF and cultured in Murashige and Skoog (MS) medium plus IAA at different concentrations (100, 1.0, or 0.1 μM) or 2,4-dichlorophenoxyacetic acid (2,4-D, Sigma) at different concentrations (10, 1.0, or 0.1 μM) in 0.1 v/v Triton X-100. Some of the collected spikelets were cultured in the MS medium with methanol in 0.1 v/v Triton X-100 as a control, as previously described [34,60]. Spikelets located on CSBs in the WT and on CPBs in asp1-1 +/were separately treated with 100 μM IAA (Sigma) in 0.1 v/v Triton X-100 using a fine brush once per day from -3 to 0 DAF. Methanol treatment used as a negative control (mock), following a method modified from a previous study [13].
Supporting information S1 Table. Candidate genes were more highly expressed in apical primary branches (CPBs) at 0 days after heading (DAH) than in 5-DAH proximal secondary branches (CSBs), and expression was higher in 12-DAH CSBs than in 5-DAH CPBs.