# Source paper archive

- Source title: Recurrent evolution of an APO1 cis-regulatory module that shapes rice panicle branching
- Author(s): Makino, T., Tsuda, K., Makino, A., Saito, K., Toriba, T., Kyozuka, J.
- Source landing URL: https://www.biorxiv.org/content/10.64898/2026.11.18.767244v1
- PDF URL: https://www.biorxiv.org/content/10.64898/2026.11.18.767244v1.full.pdf
- PDF SHA-256: 4659840576454d9bbd761b6c9f051decf6dfaf7c42c948861f148cc026d8f458
- Converter: pymupdf4llm 0.0.27
- Limits: Automated PDF text extraction; reading order, equations, tables and formatting may be incomplete or incorrect. Images are omitted, and no OCR was performed. Markup and links are neutralized for safe downloading. Consult the original PDF for the authoritative version.

---

&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.18.767244; this version posted November 19, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.18.767244)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC-ND 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc-nd/4.0/)


**Recurrent evolution of an** _**APO1**_ **cis-regulatory module that shapes rice panicle**
**branching**

Takashi Makino &#91;*1&#93;, Katsutoshi Tsuda &#91;2&#93;, Asa Makino &#91;1&#93;, Kei Saito &#91;1&#93;, Taiyo Toriba &#91;3&#93;, Junko
Kyozuka &#91;1&#93;

1 Graduate School of Life Sciences, Tohoku University, Sendai, Japan
2 School of Agriculture, Meiji University, Kawasaki, Japan
3 School of Food Industrial Sciences, Miyagi University, Sendai, Japan

***** Corresponding author: Takashi Makino
Graduate School of Life Sciences, Tohoku University, 6-3,
Aramaki Aza Aoba, Aoba-ku, Sendai 980-8578, Japan
Tel: +81-22-795-5585

Fax: +81-22-795-6693

E-mail: tamakino@tohoku.ac.jp

Running head: Evolution of _APO1_ CNSs in Poaceae

Abbreviations: CNS: Conserved noncoding sequence, _APO1_ : _ABERRANT_
_PANICLE ORGANIZATION 1_, EMSA: electrophoretic
mobility shift assay, BEP: _Bambusoideae, Ehrhartoideae_
and _Pooideae_, PACMAD: _Panicoideae, Aristidoideae,_
_Chloridoideae,_ _Micrairoideae,_ _Arundinoideae_ and
_Danthonioideae_

Key words: CNS, _APO1_, inflorescence architecture, CArG-box,
MADS, grass evolution


1


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.18.767244; this version posted November 19, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.18.767244)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC-ND 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc-nd/4.0/)


1 **Abstract**
2 Quantitative variation in inflorescence branching drives grass diversification, yet the
3 cis-regulatory mechanisms tuning reproductive meristem activity remain poorly
4 understood. Here, we identified a module of two CArG boxes within a conserved
5 noncoding sequence near rice _APO1_ . Genome-edited alleles generated a quantitative
6 phenotypic series: deletions affecting both CArG boxes produced the strongest
7 increases in secondary branching, whereas deletions restricted to CArG box 1
8 produced smaller but reproducible increases. These deletions extended secondary
9 branch formation toward the panicle base at the expense of direct spikelets, while
10 spikelet number per primary branch was maintained or increased. Across eight
11 genotypes, _APO1_ transcript abundance was quantitatively associated with branching,
12 and both CArG boxes contributed to MADS15–MADS34 binding to CNS3,
13 consistent with this module limiting _APO1_ expression. Structural and CArG-box
14 changes in this region were detected across multiple phylogenetically distinct
15 lineages with branched inflorescences, indicating that _APO1_ underwent recurrent cis16 regulatory modification during grass inflorescence diversification.
17


2


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.18.767244; this version posted November 19, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.18.767244)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC-ND 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc-nd/4.0/)


18 Grass inflorescences range from compact spikes to highly branched panicles.


19 Differences in the initiation, maintenance, and fate transitions of reproductive


20 meristems generate this architectural diversity and determine the number and spatial


21 arrangement of spikelets &#91;1-3&#93; . Genetic studies have identified conserved regulators of


22 these developmental processes, but much less is known about how evolutionary


23 changes in their cis-regulatory sequences quantitatively alter inflorescence


24 architecture. Because cis-regulatory variation can modify the level, timing, or spatial


25 domain of gene expression without changing gene function, it provides a plausible


26 mechanism for generating graded evolutionary changes in branching &#91;4&#93; .


27 _ABERRANT PANICLE ORGANIZATION 1_ ( _APO1_ ), which encodes an F-box


28 protein orthologous to _Arabidopsis UNUSUAL FLORAL ORGANS_ ( _UFO_ ), is a


29 dosage-sensitive regulator of rice inflorescence development &#91;5-7&#93; . In rice, loss of


30 _APO1_ function causes premature acquisition of spikelet meristem identity and


31 reduces spikelet number, whereas increased _APO1_ expression prolongs the


32 inflorescence meristem and branch meristem phases, thereby increasing branching


33 and spikelet production &#91;5,8,9&#93; . Natural variation at the _APO1_ locus also affects panicle


34 architecture and spikelet production &#91;6,9&#93; . More recently, targeted deletion of a


35 chromatin accessibility-defined region in the _APO1_ promoter increased grain yield


36 under field conditions &#91;10&#93; . Similarly, natural and experimentally generated variation in


37 the wheat ortholog _WAPO1_ affects spikelet number per spike &#91;11&#93; . Yet the specific cis

38 regulatory motifs that control _APO1_ expression, the transcription factors that


39 recognize them, and the extent to which their divergence contributed to grass


40 inflorescence evolution have remained unresolved.


3


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.18.767244; this version posted November 19, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.18.767244)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC-ND 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc-nd/4.0/)


41 Conserved noncoding sequences (CNSs) are promising candidates for


42 developmental cis-regulatory elements because evolutionary constraint can reveal


43 functional noncoding regions that are difficult to recognize from sequence features in


44 a single species &#91;12,13&#93; . Comparative genomic studies have identified CNSs associated


45 with gene regulation and agronomic traits in maize and other grasses &#91;14,15&#93;, while a


46 recent analysis of 284 plant species revealed extensive CNS conservation across deep


47 evolutionary timescales and demonstrated the developmental importance of ancient


48 CNSs &#91;16&#93; . Integrating sequence conservation, chromatin accessibility, transcription


49 factor binding motifs, and other genomic features has also enabled targeted promoter


50 editing to generate quantitative trait variation in rice &#91;17&#93; . Editing conserved regulatory


51 regions of meristem regulators can generate allelic series with graded phenotypic


52 effects: targeted editing of a conserved promoter region of the tomato _UFO_ ortholog


53 _ANANTHA_ revealed antagonistic cis-regulatory activities that stabilize floral


54 development &#91;18&#93;, and editing a downstream CNS of rice _TAWAWA1_ ( _TAW1_ ) generated


55 graded variation in panicle branching &#91;19&#93; .


56 MADS-domain proteins are central regulators of inflorescence and floral


57 meristem identity and recognize cis-regulatory sequences known as CArG boxes &#91;20&#93; .


58 In rice, _MADS34_ ( _PAP2_ ) regulates panicle branching and spikelet development &#91;21,22&#93;,


59 while _MADS14, MADS15, MADS18,_ and _MADS34_ cooperatively regulate


60 inflorescence meristem identity, and _MADS15_ and _MADS34_ physically interact &#91;23&#93; .


61 Whether any of these factors act through defined cis-regulatory elements at the _APO1_


62 locus, however, has not been established. Here, we identify three CNSs surrounding


63 rice _APO1_ and characterize two conserved CArG-box motifs within CNS3 that


64 together constitute a cis-regulatory module.


4


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.18.767244; this version posted November 19, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.18.767244)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC-ND 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc-nd/4.0/)


65 **Results**


66 **Comparative genomics identifies three conserved noncoding sequences near**


67 _**APO1**_


68 To investigate the cis-regulatory basis of _APO1_ expression, we compared the


69 genomic region surrounding the rice _APO1_ locus across representative species of


70 Poaceae to identify conserved noncoding sequences. We identified three noncoding


71 sequences, designated CNS1, CNS2, and CNS3, that were conserved between rice


72 and other grasses (Fig. 1a–c and Supplementary Fig. 1). CNS1 showed broad


73 sequence conservation across the sampled Poaceae, whereas CNS2 was less strongly


74 conserved. CNS3 contained discrete, highly conserved segments shared across the


75 BEP and PACMAD clades. No homologous sequences meeting our detection criteria


76 were identified in the syntenic region of _Musa acuminata_, suggesting that detectable


77 conservation of these three elements is restricted to grasses at the phylogenetic


78 resolution of our analysis.


79 CNS1, CNS2 and CNS3 overlapped accessible chromatin regions in


80 developing rice panicles in the Chromatin Accessibility of Rice Tissues (CART)


81 dataset. CNS1 also showed accessibility in Shoot Apical Meristem (SAM), whereas


82 CNS2 and CNS3 showed low or undetectable accessibility in the other tissues


83 examined (Fig. 1a) &#91;24&#93; . This panicle-associated accessibility identifies the CNSs as


84 candidate cis-regulatory elements active during inflorescence development.


85 Sequence inspection of CNS3 identified two closely spaced CArG-box motifs


86 recognized by MADS-domain transcription factors, designated CArG boxes 1 and 2


87 (Fig. 1b). Sequence-logo analysis revealed that CArG box 1 was nearly invariant


88 across the sampled Poaceae, whereas CArG box 2 was more variable (Fig. 1c).


5


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.18.767244; this version posted November 19, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.18.767244)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC-ND 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc-nd/4.0/)


89


90 **Editing the** _**APO1**_ **regulatory region produces a quantitative series of branching**


91 **phenotypes**


92 To assess the functions of the three CNSs surrounding _APO1_, we generated 34


93 independent homozygous lines carrying CRISPR-induced mutations within or around


94 CNS1, CNS2, and CNS3 (Fig. 2a and Supplementary Fig. 2; Supplementary Table


95 1). We quantified primary branch number, secondary branch number, total spikelet


96 number, and lateral-organ number (secondary branches + spikelets directly borne on


97 primary branches; Fig. 2b). To account for differences in overall panicle size, we also


98 evaluated secondary branches per lateral organ as a measure of relative secondary


99 branching.


100 Mutations in CNS1 and CNS2 consistently reduced secondary branches per


101 lateral organ and total spikelet number, although these reductions were statistically


102 significant in only some lines (Supplementary Fig. 2 and Supplementary Table 2). In


103 contrast, CNS3 mutations generated a clear allelic series (Fig. 2c–f). Deletions


104 encompassing both CArG boxes significantly increased secondary branches per


105 lateral organ in all seven lines (10, 11, 29, 30, 31, 32, and 56), with a 3.4-fold


106 increase on average relative to wild type (Fig. 2c; Supplementary Table 2). Primary


107 branch number and total spikelet number increased by 1.39-fold and 1.52-fold on


108 average, respectively, with significant increases in six and five of the seven lines


109 (Fig. 2d, e). Line 11 showed comparatively weak effects on these two traits, with


110 neither reaching statistical significance.


111 Deletions restricted to CArG box 1 (lines 23, 24, 26, and 27) produced


112 smaller increases in relative secondary branching than deletions affecting both CArG


6


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.18.767244; this version posted November 19, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.18.767244)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC-ND 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc-nd/4.0/)


113 boxes. These increases were statistically significant in lines 23, 24, and 26 but not in


114 line 27 (Fig. 2c; Supplementary Table 2). Primary branch number increased


115 significantly only in line 24, and total spikelet number did not increase significantly


116 in any of these lines. Thus, CArG box 1 deletions predominantly affected relative


117 secondary branching rather than overall panicle expansion.


118 The sole CArG box 2 deletion line (41) showed a distinct phenotype: primary


119 branch number and total spikelet number increased significantly, whereas relative


120 secondary branching did not (Fig. 2c–e; Supplementary Table 2).


121 Additional CNS3 alleles altered the central region while retaining both CArG


122 boxes: a 226-bp insertion in line 8, a 28-bp deletion with a 6-bp substitution in line


123 12, and a complex local rearrangement in line 25 (Fig. 2a; Supplementary Table 1).


124 Line 8 showed significant increases in primary branch number, total spikelet number,


125 and relative secondary branching, resembling the dominant _apo1-D_ phenotype (Fig.


126 2c–e) &#91;9&#93; . Line 12 significantly increased primary branch number alone, whereas line


127 25 showed no significant increase in the branching traits examined. Larger central

128 region deletions retaining both CArG boxes (lines 5, 6, and 7) did not produce


129 comparable increases: relative secondary branching and total spikelet number


130 showed no significant differences from wild type, and only line 5 showed a


131 significant increase in primary branch number. These contrasting effects indicate that


132 CNS3 regulatory output depends on local structural context as well as CArG-box


133 integrity.


134 Deletions between CNS2 and CNS3 also produced distinct phenotypic


135 effects. Smaller deletions of 143–160 bp (lines 34 and 37) did not significantly affect


136 relative secondary branching or total spikelet number, although primary branch


7


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.18.767244; this version posted November 19, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.18.767244)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC-ND 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc-nd/4.0/)


137 number increased in line 37 (Supplementary Fig. 2). In contrast, a 703-bp deletion


138 (line 36) significantly reduced relative secondary branching while increasing primary


139 branch number (Supplementary Table 2).


140 No tertiary branches were observed in any of the 821 panicles from 175


141 plants across 35 lines examined. Thus, even the strongest CNS3 mutations increased


142 branching only at the secondary-branch level.


143


144 **CNS3 editing shifts the spatial domain and fate of lateral organ formation**


145 We next examined the spatial distribution of secondary branches and the allocation


146 of lateral organs between direct spikelets and secondary branches. Primary branch


147 positions were normalized from 1 (most apical) to 0 (most basal) to compare panicles


148 with different numbers of primary branches.


149 In wild type panicles, secondary branches were concentrated in mid-apical


150 positions, with none on the most basal primary branch (Fig. 2g). Deletions


151 encompassing both CArG boxes expanded secondary branch formation toward the


152 panicle base, including the most basal primary branches. The mean onset position of


153 secondary branching shifted from 0.47 in wild type to 0.17 in both-box deletion lines,


154 with significant shifts in all seven independent lines (Supplementary Table 3). CArG


155 box 1 deletions produced an intermediate shift (mean 0.35), significant only in line


156 24.


157 Figure 2g and Supplementary Fig. 3 also show differences in the total number


158 of lateral organs per primary branch among genotypes. In wild type plants, the


159 number of lateral organs per primary branch was similar across most panicle


160 positions and lower on the most basal primary branches, a pattern generally observed


8


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.18.767244; this version posted November 19, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.18.767244)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC-ND 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc-nd/4.0/)


161 across genotypes. In lines with deletions affecting both CArG boxes, the total


162 number of lateral organs per primary branch was reduced across all positions, while


163 the secondary-branch component expanded as the direct-spikelet component


164 contracted. This pattern indicates that the increase in secondary branch number does


165 not reflect production of additional lateral organs, but rather a shift in lateral-organ


166 fate from direct spikelets toward secondary branches. Deletions restricted to CArG


167 box 1 showed an intermediate reduction in total lateral organ number and an


168 intermediate expansion of secondary branch fate, between wild type and the


169 phenotypes of lines with deletions affecting both CArG boxes.


170 We next examined the spatial distribution of spikelet production across all


171 mutation classes (Fig. 2h and Supplementary Fig. 3). Although lines with deletions


172 affecting both CArG boxes reduced the number of lateral organs per primary branch,


173 total spikelet production was maintained or increased (Fig. 2h): At the most basal


174 primary branch, total spikelet numbers were similar to wild type (5.9 vs. 5.4), but


175 approximately two spikelets were borne on secondary branches in the deletion lines


176 (Fig. 2h). At mid-panicle positions, secondary-branch-borne spikelets increased (6.5


177 vs. 3.2), compensating for fewer direct spikelets (3.2 vs. 6.0) and maintaining slightly


178 higher total spikelet numbers (9.7 vs. 9.3).


179 Central CNS3 mutants (lines 8, 12, and 25) showed a distinct phenotypic


180 profile. These lines tended to increase primary branch number and, in line 8, total


181 spikelet number (Fig. 2d, e), but showed little or no shift in secondary branch onset


182 position (Supplementary Table 3). Line 8 showed a modest but statistically


183 significant increase in secondary branches per lateral organ, whereas lines 12 and 25


184 did not (Supplementary Table 2).


9


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.18.767244; this version posted November 19, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.18.767244)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC-ND 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc-nd/4.0/)


185


186 _**APO1**_ **transcript abundance is associated with the degree of secondary**


187 **branching**


188 To examine the association between _APO1_ expression and branching phenotypes, we


189 quantified _APO1_ expression in 1–2-mm young panicles from eight genotypes


190 spanning the observed phenotypic range by RT-qPCR (Fig. 2i). Expression was


191 lowest in line 20 and highest in line 30. Genotype-level mean _APO_ 1 expression was


192 positively correlated with primary branch number (Spearman's rho = 0.905, P =


193 0.002; Fig. 2k), secondary branches per lateral organ (rho = 0.833, P = 0.010; Fig.


194 2j), and total spikelet number (rho = 0.738, P = 0.037; Fig. 2l; n = 8 genotypes, three


195 biological replicates per genotype).


196


197 **Both CArG box 1 and** **CArG box 2 contribute to MADS15–MADS34 binding to**


198 **CNS3**


199 CNS3 contains two conserved CArG-box motifs that may be recognized by MADS

200 domain factors. _APO1_ expression declines during early inflorescence development as


201 several MADS-box genes, including _MADS14_, _MADS15_, _MADS18_, and _MADS34_,


202 increase in expression &#91;23,25&#93; . Previous studies have shown that MADS34 regulates rice


203 inflorescence branching &#91;21&#93; and directly binds CArG-box-containing regulatory


204 regions of RCN4 &#91;22&#93; . Notably, MADS15 and MADS34 have been shown to physically


205 interact &#91;23&#93; . We therefore tested their binding to the CNS3 region containing CArG


206 boxes 1 and 2 by electrophoretic mobility shift assays (EMSA; Fig. 3). MADS14 and


207 MADS15 produced weak shifts individually, whereas MADS18 and MADS34


208 showed no detectable binding alone. Among the tested combinations, MADS14–


10


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.18.767244; this version posted November 19, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.18.767244)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC-ND 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc-nd/4.0/)


209 MADS34 and MADS15–MADS34 formed detectable complexes, with MADS15–


210 MADS34 showing the strongest binding (Fig. 3b).


211 Single-nucleotide deletion of CArG box 1 (mut1) or CArG box 2 (mut3)


212 reduced but did not abolish MADS15–MADS34 binding, whereas simultaneous


213 deletion of both CArG boxes (mut1&amp;3) abolished detectable binding (Fig. 3a, c). A


214 single-nucleotide deletion between the motifs (mut2) did not reduce binding.


215 We next tested CNS3 fragments amplified from the edited lines used for


216 phenotypic analysis (Fig. 3d). Wild type CNS3 and line 28, which retained both


217 intact CArG boxes, showed detectable binding. Binding was abolished in lines with


218 deletions affecting both CArG boxes (lines 10, 11, 29, 30, 31, and 32), CArG box 2


219 (line 41), or CArG box 1 (lines 23, 24, and 26). Line 27, carrying a CArG box 1


220 deletion, retained weak binding, consistent with its mild phenotype. These results


221 demonstrate that CArG-box-dependent binding occurs not only with synthetic probes


222 but also with CNS3 fragments derived from the edited lines.


223 Together, these biochemical and genetic results support a model in which


224 MADS15–MADS34 binding to CNS3 limits _APO1_ expression and restricts


225 secondary branch formation to mid-apical primary branches (Fig. 5).


226


227 **The** _**APO1**_ **regulatory module changed recurrently during grass inflorescence**


228 **evolution**


229 Grass inflorescences vary widely in architecture, ranging from unbranched spikes


230 and racemes to highly branched panicles (Fig. 4a). To determine whether the _APO1_


231 regulatory region shows phylogenetically correlated variation with this diversity, we


232 surveyed the CNS3 interval and its flanking sequences across grass species with


11


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.18.767244; this version posted November 19, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.18.767244)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC-ND 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc-nd/4.0/)


233 chromosome-scale genome assemblies and mapped regulatory structure and


234 inflorescence architecture onto the species phylogeny (Fig. 4b and Supplementary


235 Table 4).


236 CArG box 1 was nearly invariant across the sampled Poaceae, with


237 substitutions detected only in a few species. In contrast, CArG box 2 was more


238 variable, with substitutions in both spike- and panicle-bearing lineages (Fig. 1c).


239 Thus, substitutions in CArG box 2 alone were not consistently associated with


240 increased branching. Changes in CArG box 1 or loss of CArG box 2 were


241 concentrated in more highly branched lineages, whereas many spike-bearing taxa


242 retained conserved motifs (Fig. 4b, Supplementary Fig. 4a and Supplementary Table


243 5).


244 Structural changes were also observed in multiple lineages, including CNS2


245 deletions, CNS3 insertions, insertions between CNS2 and CNS3, and translocation of


246 CNS2 to a different chromosome. These changes occurred predominantly in panicle

247 bearing lineages, whereas spike-bearing taxa generally retained an intact CNS


248 configuration (Fig. 4b, c, Supplementary Table 6). To quantify these associations, we


249 scored five cis-regulatory features across 24 species: CNS2–CNS3 spacing change


250 (&gt;110 bp or translocation), CNS2 structural change (&gt;30 bp), CNS3 structural


251 change (&gt;30 bp), CArG box 1 substitution, and CArG box 2 change (nucleotide


252 substitution relative to the ancestral CTGAAATG motif, deletion, or motif absence).


253 Branched species showed higher composite divergence scores even after accounting


254 for phylogenetic non-independence (phylogenetic GLS, β = 1.42, SE = 0.46, p =


255 0.0050; Fig. 4d and Supplementary Table 7). Because minor substitutions in CArG


256 box 2 were observed across many grass species, we also tested a composite score that


12


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.18.767244; this version posted November 19, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.18.767244)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC-ND 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc-nd/4.0/)


257 excluded these minor changes from the CArG box 2 component. This association


258 remained significant when the composite divergence score was calculated using only


259 major CArG box 2 changes ( ≥ 4 nucleotide substitutions, deletion, or motif absence; β


260 = 1.64, SE = 0.50, p = 0.0036; Supplementary Table 7). Among individual features,


261 CNS2–CNS3 spacing change (phylogenetic GLS, p = 0.0060) and CNS2 structural


262 change (p = 0.018) were significantly associated with inflorescence type


263 (Supplementary Table 7).


264 CArG box 1 was conserved across spike-bearing taxa except in the _Secale_


265 _cereale_ reference assembly, which contained an insertion within the motif. Sanger


266 sequencing of an independent _S. cereale_ accession recovered an intact motif


267 (Supplementary Fig. 4b). Because the insertion was not detected by Sanger


268 sequencing, we treated CArG box 1 as intact in _S. cereale_ for the comparative


269 analysis.


270 The CNS2–CNS3 configuration in distantly related unbranched lineages,


271 including _Aegilops tauschii_ and _Zoysia japonica_, suggested ancestral conservation


272 (Fig. 4c). Under a parsimonious interpretation, these observations supported an


273 intact, closely linked CNS2–CNS3 configuration as the ancestral state of core


274 Poaceae, followed by independent changes in multiple panicle-bearing lineages (Fig.


275 4b). Although CNS2 and CNS3 were undetectable in the earliest-diverging sampled


276 grass, _Streptochaeta angustifolia_, _Pharus latifolius_, the earliest-diverging lineage


277 with detectable CNSs, retained a compact 65 bp CNS2–CNS3 interval. This supports


278 an ancestral closely linked configuration extending beyond core Poaceae. The


279 association was not absolute: some panicle-bearing taxa retained the conserved CNS


13


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.18.767244; this version posted November 19, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.18.767244)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC-ND 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc-nd/4.0/)


280 configuration. _APO1_ regulatory changes were recurrently associated with increased


281 branching but did not uniquely determine inflorescence type.


282


14


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.18.767244; this version posted November 19, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.18.767244)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC-ND 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc-nd/4.0/)


283 **Discussion**


284 The graded CNS3 allelic series links a two-CArG-box module to quantitative


285 changes in _APO1_ expression and secondary branching. Deletions affecting both


286 motifs consistently produced the strongest increase in secondary branching, whereas


287 CArG box 1 deletions produced smaller, reproducible effects. The sole CArG box 2


288 deletion allele (line 41) instead increased primary branch number without a


289 significant increase in relative secondary branching. This distinction is intriguing but


290 requires additional CArG box 2 alleles for confirmation. Importantly, _APO1_


291 promotes both inflorescence meristem growth and the maintenance of branch


292 meristem identity &#91;8,9&#93; . Thus, the contrasting single-box phenotypes suggest a possible


293 cis-regulatory basis for separating these established _APO1_ functions across


294 developmental stages, although this interpretation remains to be tested directly.


295 EMSA using probes from the edited alleles reinforced the relationship


296 between CArG-box integrity and binding. Line 28, which retained both motifs,


297 bound MADS15–MADS34 comparably to wild type; binding was abolished in the


298 tested alleles disrupting both motifs or either motif alone, except for weak residual


299 binding in line 27. In contrast to the synthetic single-nucleotide deletion probes, most


300 edited alleles removed multiple bases, plausibly accounting for the stronger loss of


301 binding. Line 27 carries a three-nucleotide deletion but retains the sequence


302 CTTTTTTTG, differing at only one position from the CC(A/T)6G consensus; this


303 may explain both residual binding and its mild phenotype. The preferential binding


304 of MADS15–MADS34 to CNS3 is consistent with evidence that the intervening


305 domain of plant MADS transcription factors contributes to dimerization partner


306 selection and DNA-binding specificity &#91;26&#93; . _MADS5_ and _MADS34_ have been shown to


15


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.18.767244; this version posted November 19, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.18.767244)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC-ND 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc-nd/4.0/)


307 limit rice panicle branching by directly repressing _RCN4_ &#91;22&#93; . Our findings suggest that


308 MADS15–MADS34 may similarly restrict secondary branching through the _APO1_


309 CNS3 module, although direct transcriptional repression of _APO1_ remains to be


310 demonstrated.


311 Earlier transgenic experiments, conducted before CNS3 was identified,


312 provide retrospective support for this model. A genomic _APO1_ construct containing


313 the region now recognized as CNS3 ( _pro5kb-APO1_ ) rescued the _apo1-3_ phenotype,


314 whereas a shorter construct lacking CNS3 ( _pro3kb-APO1_ ) increased _APO1_


315 expression and branching, albeit with abnormalities in distal panicle development &#91;8&#93; .


316 Viewed in light of our findings, these results are consistent with a role for CNS3 in


317 constraining _APO1_ expression. In addition, a deletion upstream of CNS3, in a region


318 absent from _pro3kb-APO1_, likewise increased _APO1_ expression &#91;10&#93;, suggesting


319 additional regulatory sequences outside CNS3.


320 Positional analysis shows that CNS3 regulates the basal extent of secondary

321 branch formation and lateral-organ fate without necessarily reducing spikelet output.


322 The finding that tertiary branches were absent from all _APO1_ CNS3 editing lines,


323 despite marked increases in secondary branch number, distinguishes the role of


324 _APO1_ from that of _TAWAWA1_ ( _TAW1_ ). _TAW1_ promotes inflorescence meristem


325 activity and suppresses the phase transition from branch meristems to spikelet


326 meristems; its gain-of-function alleles and CNS-deletion lines produce tertiary


327 branches, reflecting a generalized extension of meristem branching capacity &#91;19,27&#93; . By


328 contrast, the _APO1_ CNS3 phenotypes identified here highlight a different step: the


329 fate of lateral organs on established primary branches, determining whether each


330 lateral organ primordium becomes a spikelet or a secondary branch meristem. _APO1_


16


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.18.767244; this version posted November 19, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.18.767244)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC-ND 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc-nd/4.0/)


331 and _TAW1_ thus provide independent regulatory entry points into the inflorescence


332 developmental program, appearing to act at distinct steps in the sequential


333 elaboration of the grass panicle.


334 The spatial profile also provides a developmental interpretation for the


335 positional specificity of the wild type phenotype. In wild type rice, secondary


336 branches form preferentially on mid-apical primary branches, suggesting positional


337 differences in the competence to form secondary branches. The basal expansion of


338 secondary branch formation in CArG-box deletion lines is consistent with elevated


339 _APO1_ expression extending this competence to more basal primary branch meristems


340 before the transition to spikelet identity occurs. This interpretation aligns with the


341 established role of _APO1_ in delaying meristem phase transitions &#91;8,9&#93; and places the


342 CNS3 module as a cis-regulatory brake on this delay.


343 Line 8 carries a 226-bp insertion, line 12 a 28-bp deletion with a 6-bp


344 substitution, and line 25 a local replacement involving a duplicated upstream


345 sequence; their branching effects differ despite their proximity. Importantly, this


346 central region overlaps the clustered transposon insertion sites of the four


347 independently isolated dominant _apo1-D_ alleles &#91;9&#93; . The line 8 insertion coincides with


348 _apo1-D2_ and lies one nucleotide from _apo1-D1_, whereas the edits in lines 12 and 25


349 overlap the _apo1-D4/Ur1_ insertion site (Supplementary Fig. 5). Thus, two


350 complementary approaches, _apo1-D_ mutant isolation and genome editing,


351 independently identify the same relatively weakly conserved segment between


352 strongly conserved CNS3 regions as functionally sensitive.


353 Line 8 most closely resembled the dominant _apo1-D_ phenotype, increasing


354 primary and secondary branching and spikelet number, and also showed elevated


17


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.18.767244; this version posted November 19, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.18.767244)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC-ND 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc-nd/4.0/)


355 _APO1_ expression (Fig. 2i). Its insertion may alter regulatory spacing or chromatin


356 organization. Crucially, larger deletions of the central region in lines 5, 6, and 7,


357 which retained both CArG boxes and, in lines 5 and 7, removed all four _apo1-D_


358 insertion sites, did not produce comparable increases in branching. This contrast


359 argues that the line 8 phenotype cannot be explained by simple loss of central-region


360 sequence and instead may depend on the particular structural consequences of the


361 insertion. More broadly, the differing effects of these alleles suggest that CNS3 is not


362 a simple two-element switch: regulatory output depends on local structural context as


363 well as CArG-box integrity. Separating spacing from sequence effects will require


364 targeted follow-up experiments.


365 CArG box 1 was nearly invariant among spike-bearing taxa, whereas CArG


366 box 2 varied in both spike- and panicle-bearing lineages. This asymmetry in


367 conservation mirrors the asymmetry in phenotypic effects observed in the rice allelic


368 series: deletions restricted to CArG box 1 generally increased secondary branching,


369 whereas the single CArG box 2 deletion showed a distinct, trait-specific phenotype.


370 The ancestral inflorescence form of Poaceae remains debated. Classical


371 morphological models proposed a panicle-like branching system as the ancestral


372 condition, from which simpler inflorescence architectures arose largely through


373 reductive processes &#91;28&#93;, a view also adopted in subsequent studies of grass


374 inflorescence evolution &#91;29&#93; . In contrast, phylogenetic and developmental analyses


375 have emphasized that superficially similar branched inflorescences can arise from


376 distinct developmental and evolutionary trajectories &#91;30&#93; . Notably, spike-bearing


377 species in both the BEP and PACMAD clades retain an intact, closely linked CNS2–


378 CNS3 configuration (Fig. 4b, c). Under a parsimonious interpretation, these


18


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.18.767244; this version posted November 19, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.18.767244)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC-ND 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc-nd/4.0/)


379 observations support an intact, closely linked CNS2–CNS3 configuration in the


380 ancestor of core Poaceae, consistent with the possibility that its inflorescence was


381 spike-like. Recurrent modification of this regulatory configuration may have


382 provided a route to increased inflorescence branching in descendant lineages (Fig.


383 4b). The uncertain status of CNS2 and CNS3 in _Streptochaeta_, where the elements


384 may have been lost or diverged beyond recognition, limits inference concerning the


385 deeper ancestor of all Poaceae but does not affect the reconstructed state for the core


386 Poaceae ancestor.


387 Independent insertions within CNS3 or between CNS2 and CNS3 occur in


388 panicle-bearing _Oryza sativa_, _Leersia perrieri_, and _Saccharum spontaneum_ . Their


389 locations recall the dominant _apo1-D_ transposon insertions, which increased _APO1_


390 expression and branching &#91;9&#93;, suggesting that structural disruption may represent one


391 recurrent route to altered regulation. However, some branched Pooideae species


392 retain an intact CNS2–CNS3 configuration, indicating that structural modification of


393 this module is not required for the evolution of branched inflorescences. The effect of


394 a deletion in the intergenic region between CNS2 and CNS3 in rice further suggests


395 that structural changes in this region can alter panicle branching. Together, these


396 findings refine the model: _APO1_ regulatory output may be modified through distinct


397 changes in motif sequence, module organization, or spacing, rather than through a


398 single structural route. The comparative associations do not establish causation in


399 other grasses.


400 Several questions remain unresolved: direct repression and in vivo MADS15–


401 MADS34 occupancy at CNS3 have not been demonstrated; central-region edits do


19


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.18.767244; this version posted November 19, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.18.767244)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC-ND 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc-nd/4.0/)


402 not isolate spacing effects from sequence effects; and natural CNS configurations


403 have not been functionally tested in their native genetic backgrounds.


404 The allelic series also has potential value for crop improvement. Unlike


405 constitutive _APO1_ overexpression, regulatory editing may allow moderate increases


406 in branch or spikelet number while retaining otherwise normal development.


407 However, this study does not directly evaluate grain yield or possible trade-offs


408 involving grain filling, fertility, and panicle number. The appropriate agricultural


409 conclusion is therefore that _APO1_ CNS3 provides candidate alleles for evaluating


410 yield-related trade-offs, not that the present mutants have already been demonstrated


411 to improve grain yield.


20


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.18.767244; this version posted November 19, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.18.767244)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC-ND 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc-nd/4.0/)


412 **Materials and methods**


413 **Identification of CNSs in rice chromosome 6**


414 To identify CNSs in the _APO1_ regulatory region, we obtained nucleotide sequences


415 of chromosome 6 with the _APO1_ ortholog for _Leersia perrieri_, _Brachypodium_


416 _distachyon_, _Aegilops tauschii_, _Hordeum vulgare_, and _Oryza sativa_ Japonica from


417 EnsemblPlants release 60 (http://plants.ensembl.org). Pairwise whole-chromosome


418 alignments were generated between rice chromosome 6 (IRGSP-1.0) and orthologous


419 chromosomes from the four BEP clade species using LASTZ v1.04.41 with the


420 parameters --inner=2000 and --chain. Multiple sequence alignments of these five


421 BEP clade species were constructed with MULTIZ/TBA v11.2; single coverage was


422 enforced with single_cov2, and roast produced the combined alignment under the


423 guide topology (( _O. sativa_, _L. perrieri_ ) ( _B. distachyon_, ( _A. tauschii_, _H. vulgare_ ))).


424 Neutral substitution patterns were modeled by extracting fourfold degenerate (4D)


425 sites from the rice reference annotation (IRGSP-1.0) with msa_view in PHAST v1.5


426 &#91;31&#93; . The 4D sites were used to build a neutral phylogenetic model with phyloFit and


427 phyloBoot, yielding a nonconserved model. For the conserved model, the


428 chromosome 6 alignments were partitioned into 1 Mb windows using msa_split, and


429 local conserved models were estimated with phastCons. The window-based


430 conserved models were then averaged with phyloBoot to produce the final conserved


431 phylogenetic model. We computed basewise conservation scores with phastCons


432 using the neutral and conserved models learned as described above, with target


433 coverage = 0.25 and expected conserved length = 12 bp. Conserved elements were


434 then called as contiguous runs of positions with phastCons posterior probability ≥ 0.7


435 and a minimum length of 20 bp. Elements overlapping annotated coding sequences


21


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.18.767244; this version posted November 19, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.18.767244)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC-ND 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc-nd/4.0/)


436 were classified as coding and excluded from downstream analyses; the remaining


437 elements were considered CNSs.


438 To assess conservation at a broader phylogenetic scale, nucleotide sequences


439 of the chromosome containing the APO1 ortholog for _Sorghum bicolor_, _Zea mays_,


440 _Setaria italica_, and _Panicum hallii FIL2_ and _Musa acuminata_ were obtained from


441 EnsemblPlants. Whole-chromosome alignments were generated for these four


442 additional PACMAD species and _Musa acuminata_ using the same LASTZ


443 parameters, and a combined alignment of all nine Poaceae species plus Musa was


444 constructed under the guide topology ((((( _O. sativa_, _L. perrieri_ ) ( _B. distachyon_, ( _A._


445 _tauschii_, _H. vulgare_ ))) (( _S. bicolor_, _Z. mays_ ) ( _S. italica_, _P. hallii_ ))) _Musa_


446 _acuminata_ )).


447 We surveyed the presence of CNSs in upstream and downstream of _APO1_


448 (chr6:27480082-27481453) in _O. sativa,_ and found three CNSs (CNS1:


449 chr6:27478213-27478295; CNS2: chr6:27483827-27483993; CNS3: chr6:27484894

450 27485103) neighboring _APO1_ . Orthologous CNSs in additional Poaceae species,


451 including _Avena sativa OT3098, Lolium perenne, Secale cereale, Saccharum_


452 _spontaneum, Echinochloa crus-galli, Digitaria exilis_ and _Eragrostis tef_ in


453 EnsemblPlants, _Eremochloa ophiuroides_ &#91;32&#93; and _Zoysia japonica_ &#91;33&#93;, _Pharus latifolius_


454 34 and _Streptochaeta angustifolia_ 35, as well as _Poa annua_ (GCA_026929855.1),


455 _Dactylis glomerata_ (GCA_007115705.1), _Agrostis stolonifera_ (GCA_052724435.1),


456 _Holcus mollis_ (PRJEB94103), _Deschampsia cespitosa_ (PRJEB122840), and _Bromus_


457 _sterilis_ (GCA_950022295.1) from NCBI were detected by BLASTN. For _Z. mays, A._


458 _sativa, S. spontaneum, E. crus-galli, D. exilis,_ and _E. tef,_ _P. annua,_ _A. stolonifera_,


459 and _H. mollis_, which have undergone whole-genome duplication or polyploidization


22


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.18.767244; this version posted November 19, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.18.767244)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC-ND 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc-nd/4.0/)


460 and therefore carry more than one _APO1_ locus, one representative locus with the


461 most diverged CNS2 and CNS3 was included in Figure 4b.


462


463 **Identification of orthologous CNSs and CArG box motifs**


464 The genomic positions of CNS2 and CNS3 were examined in each species, and their


465 location in the vicinity of the _APO1_ ortholog was confirmed based on the


466 corresponding genome annotation. CNS sequences were aligned with MAFFT


467 v7.505. CArG boxes 1 and 2 were identified by inspection of the aligned CNS3


468 interval. The canonical CArG box 1 sequence (CCATTTTTG) and CArG box 2


469 sequence (CTGAAATG) are highly conserved across Poaceae and are consistently


470 located in the 3' region of CNS3. Substitutions relative to these ancestral sequences


471 are reported in Supplementary Table 5. Cases where the motif was absent or


472 undetectable are indicated in Supplementary Table 5. Sequence logos for CArG box


473 1 and CArG box 2 were generated using logomaker v1.2 &#91;36&#93; in Python. Sequences


474 were collected from species with an identifiable motif (CArG box 1, n = 25; CArG


475 box 2, n = 23), using one representative sequence per species. Species without an


476 identifiable motif were excluded: _Streptochaeta angustifolia_ for CArG box 1, and _S._


477 _angustifolia,_ _Pharus latifolius_, and _Echinochloa crus-galli_ for CArG box 2. For


478 polyploid species, a single representative sequence was used. Information content at


479 each position was calculated from the observed nucleotide frequencies relative to a


480 uniform background.


481


482 **Vector construction for CRISPR/Cas9 genome editing**


23


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.18.767244; this version posted November 19, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.18.767244)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC-ND 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc-nd/4.0/)


483 To generate genome-edited lines in the _APO1_ regulatory regions, pairs of guide


484 RNAs (gRNAs) were incorporated into each CRISPR/Cas9 vector and expressed


485 simultaneously. Fragments containing each gRNA pair were amplified by PCR using


486 the pUCAMdE-U3U6 vector as a template and inserted into the pZNH2GTR-U6


487 vector by multiple steps of enzyme digestion and ligation. The pZNH2GTR-U6


488 vector, derived from pZ2028 37, contains a functional Cas9 cassette and


489 spectinomycin- and hygromycin-resistance genes. Escherichia coli strain DH5 α was


490 grown in 2×YT medium containing spectinomycin at 37°C for vector cloning.


491 Agrobacterium tumefaciens strain EHA105 was transformed with the constructed


492 vector and grown in 2×YT medium containing spectinomycin at 28°C. The positions


493 of the gRNA target sites are shown in Figure 2 and Supplementary Figure 2. The


494 gRNA sequences are listed in Supplementary Table 8, and the gRNA pairs used for


495 genome editing are listed in Supplementary Table 9.


496


497 **Identification and genotyping of edited lines**


498 Genomic DNA was extracted from leaves of wild type and edited rice lines using the


499 Maxwell RSC Plant DNA Kit (Promega) according to the manufacturer's


500 instructions. Edited _APO1_ regulatory regions were amplified from genomic DNA


501 using the primers listed in Supplementary Table 9. PCR products were separated by


502 agarose gel electrophoresis; bands of expected size were excised and subjected to


503 Sanger sequencing. For lines in which small indels prevented reliable band


504 separation, genotyping was performed in the T1 generation. Homozygous alleles


505 were confirmed by Sanger sequencing of T1 plants. Deletion or insertion junctions


506 were determined relative to the Nipponbare reference sequence. The category, target


24


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.18.767244; this version posted November 19, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.18.767244)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC-ND 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc-nd/4.0/)


507 region, and allele of each edited line are reported in Supplementary Table 1. All lines


508 were confirmed homozygous by genotyping. Independently derived lines were


509 obtained from independent transformation events or from T1 siblings originating


510 from the same T0 plant.


511


512 **Rice transformation**


513 Rice ( _Oryza sativa_ L. ssp. _japonica_ ) cv. Nipponbare served as the wild type in this


514 study. Nipponbare seeds were de-husked, surface-sterilized with 70% ethanol and


515 then with sodium hypochlorite 1% for 30 min, followed by 3 washes in sterile


516 distilled water. Surface-sterilized seeds on 2N6 medium were kept at 32 °C in the


517 dark for 4 weeks for inducing calli from seeds. The seeds were replanted in new 2N6


518 medium every two weeks. Agrobacterium strain EHA105 harboring the constructed


519 vector were cultured in AB medium with spectinomycin at 23 °C in the dark for 3


520 days. Before the infection of the Agrobacterium, calli were cultured in new 2N6


521 medium at 32 °C in the dark for 3 days. Agrobacterium strain EHA105 was


522 suspended by liquid MSL medium with acetosyringone, and added to calli. The calli


523 were grown in N6CO medium at 23 °C in the dark. The calli were washed by sterile


524 distilled water and meropenem trihydrate solution, and were grown in N6SE medium


525 with meropenem trihydrate and hygromycin at 32 °C in the dark for 4 weeks. Calli


526 were grown in ReIII medium with meropenem trihydrate and hygromycin at 28 °C


527 under continuous light for about 6 weeks. Plants approximately 10 mm in height


528 were grown in MSF medium at 28 °C under continuous light. Plants approximately


529 150 mm in height were transplanted in soil and grown in a biohazard room at 28 °C


530 under the regime of a 14 h light/10 h dark photoperiod.


25


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.18.767244; this version posted November 19, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.18.767244)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC-ND 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc-nd/4.0/)


531


532 **Plant growth and panicle phenotyping**


533 Plants used for phenotyping were grown as described above. Line 1 was a


534 non-transgenic Nipponbare plant grown alongside the edited lines. Five plants were


535 analyzed per line, and up to five mature panicles were collected per plant in order of


536 emergence; panicles beyond the fifth were excluded.


537 For each panicle, we counted primary branches, secondary branches, total


538 spikelets, and lateral organs. Primary branches were defined as branches arising


539 directly from the main rachis, and secondary branches as branches arising from


540 primary branches. Lateral-organ number was defined as the number of secondary


541 branches plus the number of spikelets borne directly on primary branches, excluding


542 the terminal spikelet of each primary branch. Relative secondary branching was


543 calculated for each primary branch as the ratio of secondary branches to lateral


544 organs, and the resulting ratios were averaged across primary branches within each


545 panicle. Tertiary branches were recorded but were absent from all panicles examined.


546 For each trait, measurements from multiple panicles were averaged within


547 each plant, and the resulting plant means were used as independent biological


548 replicates for statistical analysis.


549 For the spatial analysis of lateral-organ formation, primary branches were


550 numbered sequentially from the basal to the apical end of each panicle. Because the


551 number of primary branches differed among panicles, the position of each primary


552 branch was normalized as ( _r −_ 1)/( _N_ − 1), where _r_ is its basal-to-apical rank and _N_ is


553 the total number of primary branches on that panicle, as determined from the panicle


554 summary record. The resulting coordinate ranged from 1 for the most apical primary


26


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.18.767244; this version posted November 19, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.18.767244)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC-ND 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc-nd/4.0/)


555 branch to 0 for the most basal primary branch. Normalized positions were used to


556 determine the onset position of secondary branching. For visualization and


557 descriptive summaries, the normalized coordinate was divided into five equal-width


558 intervals: 0–&lt;0.2, 0.2–&lt;0.4, 0.4–&lt;0.6, 0.6–&lt;0.8, and 0.8–1.0, corresponding to


559 progressively more apical regions of the panicle. The onset position of secondary


560 branching was defined for each panicle as the normalized position of the most basal


561 primary branch bearing at least one secondary branch; it was treated as missing for


562 panicles without secondary branches.


563 For each primary branch, we recorded the number of secondary branches, the


564 number of spikelets borne directly on the primary branch, and the number of


565 spikelets borne on its secondary branches. The terminal spikelet of each primary


566 branch was included in the direct-spikelet count but was not classified as a lateral


567 organ. Accordingly, the number of lateral organs on a primary branch was calculated


568 as the number of secondary branches plus the number of directly borne spikelets


569 excluding the terminal spikelet. Lateral organs were classified according to whether


570 they developed as direct spikelets or secondary branches. Total spikelet number per


571 primary branch was calculated as the sum of directly borne spikelets, including the


572 terminal spikelet, and spikelets borne on secondary branches. Position-specific


573 measurements were first calculated for each panicle and then averaged across


574 panicles from the same plant. Plant means were treated as independent biological


575 replicates for comparisons among lines. For summaries of mutation classes, plant


576 means were averaged within each line, and class-level values were calculated as


577 unweighted means of line means so that each independently derived line contributed


578 equally. Error bars in class-level positional plots represent the standard error of the


27


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.18.767244; this version posted November 19, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.18.767244)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC-ND 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc-nd/4.0/)


579 mean among line means and were not calculated for mutation classes represented by


580 a single line.


581


582 **Statistical analysis of panicle phenotypes**


583 Trait values were first calculated for individual panicles and then averaged within


584 each plant. Plant means were treated as independent biological replicates in all


585 statistical analyses; individual panicle measurements were retained for visualization


586 but were not treated as independent replicates. When fewer than five panicles were


587 available, the plant mean was calculated from the available measurements, and each


588 plant contributed one value to the line-level analysis. Each line was compared with


589 line 1 (wild type) using a two-sided Mann–Whitney U test applied separately to each


590 of the three traits (primary branch number, total spikelet number, and secondary


591 branches per lateral organ). P values from all pairwise tests (34 lines × 3 traits = 102


592 comparisons) were adjusted together using the Benjamini–Hochberg false discovery


593 rate (FDR) procedure; adjusted p values below 0.05 were considered statistically


594 significant. Statistical analyses were performed in Python (version 3.12) using the


595 SciPy (version 1.13) and statsmodels (version 0.14) libraries. Exact numbers of


596 plants and panicles contributing to each line, together with U statistics, unadjusted p


597 values, FDR-adjusted p values, and line means, are provided in Supplementary Table


598 2.


599 For the spatial analysis, the onset position of secondary branching was


600 compared between each line and wild type using a one-sided Mann–Whitney U test


601 (alternative: line &lt; wild type). Plant means of onset position of secondary branching


602 were used as the unit of comparison. P values from all pairwise tests (34


28


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.18.767244; this version posted November 19, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.18.767244)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC-ND 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc-nd/4.0/)


603 comparisons) were adjusted together using the Benjamini–Hochberg false discovery


604 rate (FDR) procedure. Detailed results are provided in Supplementary Table 3.


605


606 **RNA extraction and expression analysis**


607 Total RNA was extracted from 1–2 mm young panicles from wild type and edited


608 lines using the Maxwell RSC Plant RNA Kit (Promega) according to the


609 manufacturer's instructions. First-strand cDNA was synthesized from 1 μ g of total


610 RNA using the SuperScript VILO cDNA Synthesis Kit (Invitrogen) with oligo(dT)


611 primers. qRT-PCR was performed using KOD SYBR qPCR Mix (TOYOBO) on a


612 StepOnePlus Real-Time PCR System (Applied Biosystems). Each 20- μ L reaction


613 contained 2 μ L of 20-fold-diluted cDNA, 10 μ L of KOD SYBR qPCR Mix, and 0.32


614 μ M of each primer. The thermal cycling conditions comprised an initial denaturation


615 at 98°C for 2 min, followed by 40 cycles of 98°C for 10 s, 60°C for 10 s, and 68°C


616 for 30 s. Amplification specificity was verified by melting-curve analysis. Primers


617 used were as follows: APO1, 5 ′ -TTCGCCGTCAAGAACATCT-3 ′ and 5 ′   

618 GGGATCGAGGGAGGAGAG-3 ′ ; OsUBQ5, 5 ′ -AACCAGCTGAGGCCCAAGA-3 ′


′ ′
619 and 5 -ACGATTGATTTAACCAGTCCATGA-3 . _APO1_ transcript abundance was


620 normalized to _OsUBQ5,_ and relative expression was calculated using the 2 &#91;-&#93; &#91;ΔΔ&#93; &#91;Ct &#93;


621 method.


622 Associations between _APO1_ transcript abundance measured by qRT-PCR and


623 panicle traits were evaluated using genotype-level means for the eight selected


624 genotypes. Individual plants and panicles were not treated as independent


625 expression–phenotype pairs. Associations were tested using Spearman’s rank


626 correlation, with Spearman’s ρ and P values reported for each trait.


29


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.18.767244; this version posted November 19, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.18.767244)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC-ND 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc-nd/4.0/)


627


628 **Gel shift assays**


629 Gel shift assays were performed as previously described &#91;38&#93; with minor modifications.


630 The coding sequences of _OsMADS14, OsMADS15, OsMADS18_, and _OsMADS34_


631 were amplified by PCR and cloned into XhoI- and XbaI-digested pTnT vector


632 (Promega) using the NEBuilder HiFi DNA Assembly kit (NEB). In vitro


633 transcription and translation were carried out using 5 � µg of each plasmid with the


634 TnT SP6 High-Yield Wheat Germ Protein Expression System (Promega) in 25 � µL


635 reactions. The wild type CNS3 fragment was amplified from Nipponbare genomic


636 DNA and cloned into the pCR-Blunt vector using the Zero Blunt TOPO cloning kit


637 (Thermo Fisher). Mutant probes were generated by inverse PCR and self-ligation,


638 using the plasmid containing the wild-type CNS3 fragment described above as a


639 template. Similarly, the mut1&amp;3 probe was generated by inverse PCR using a


640 plasmid containing mut1 as a template and primers for mut3. Finally, using each


641 plasmid as a template, biotin-labeled probes were synthesized by PCR using 5'

642 biotinylated primers annealing to the vector backbone. Primers were listed in


643 Supplementary Table 10.


644 Binding reactions (20 μ L total volume) were carried out in binding buffer (10


645 mM Tris-HCl, pH 7.5, 50 mM NaCl, 1% CHAPS, 0.5 mM EDTA, 1 mM MgCl �,


646 0.5 mM DTT, 60 μ g/mL salmon sperm DNA, and 8% glycerol), with 20 fmol of


647 probe and 2 μ L of in vitro translated protein. Samples were electrophoresed on 5%


648 non-denaturing polyacrylamide gels pre-run at 100 V for 30 min at 4°C in 0.5× TBE


649 buffer, followed by electrophoresis for 2 h under the same conditions. Biotin-labeled


30


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.18.767244; this version posted November 19, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.18.767244)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC-ND 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc-nd/4.0/)


650 probes were detected using the LightShift Chemiluminescent EMSA Kit (Thermo


651 Fisher).


652 Three sets of EMSAs were performed. First, binding of OsMADS14,


653 OsMADS15, OsMADS18, and OsMADS34, individually and in combination, to the


654 wild type CNS3 probe was examined. Reactions without an expression plasmid


655 during in vitro translation were used as negative controls.


656 Second, the contribution of the two CArG boxes to MADS15–MADS34


657 binding was examined using five CNS3 probes: wild type; mut1, containing a 1-bp


658 deletion in CArG box 1; mut2, containing a 1-bp deletion in the intervening sequence


659 between CArG boxes 1 and 2; mut3, containing a 1-bp deletion in CArG box 2; and


660 mut1&amp;3, containing both the mut1 and mut3 deletions. These probes were assayed


661 with MADS15 and MADS34.


662 Third, CNS3 fragments corresponding to the actual genome-edited alleles


663 were PCR-amplified from genomic DNA using primers KT2641 and KT2642 and


664 cloned into the pCR-Blunt vector. Biotin-labelled fragments were generated as


665 described above, and used as EMSA probes. The assayed fragments were derived


666 from the wild type and edited lines 28, 27, 31, 10, 30, 29, 24, 23, 26, 11, 32, and 41.


667 Each probe was assayed in the presence or absence of MADS15 and MADS34 under


668 the same binding conditions described above. Primer sequences used for probe


669 preparation are listed in Supplementary Table 10.


670


671 **Validation of the** _**Secale cereale**_ **CNS3 sequence**


672 Seeds of _Secale cereale_ commercially sold as ‘Raitaro’ were obtained from Takii &amp;


673 Co., Ltd. (Kyoto, Japan) and grown for genomic DNA extraction. Genomic DNA


31


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.18.767244; this version posted November 19, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.18.767244)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC-ND 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc-nd/4.0/)


674 was extracted from leaves using the Maxwell RSC Plant DNA Kit (Promega)


675 according to the manufacturer's instructions. The CNS3 region containing the CArG

676 box was amplified by PCR using the primers listed in Supplementary Table 9 and


677 subjected to Sanger sequencing. The obtained sequence was compared with the


678 corresponding sequence in the Rye_Lo7_2018_v1p1p1 reference genome. The


679 experimentally determined sequence and its discrepancy from the reference genome


680 are shown in Supplementary Figure 4b.


681


682 **Inflorescence type classification**


683 Inflorescence types were classified based on two structural features: spikelet


684 attachment and branching pattern &#91;30&#93; . Spikelets were categorized as sessile (directly


685 attached), pedicellate (borne on stalks), or mixed (both). Branching was defined as


686 either unbranched (no lateral branches) or branched (with lateral branches). Branched


687 architectures included both hierarchically subdivided lateral axes and digitate


688 arrangements. Three canonical inflorescence types were defined using these criteria:


689 spike (sessile, unbranched), raceme (pedicellate, unbranched), and panicle


690 (pedicellate, hierarchically branched). In addition, spike-like racemes and digitate


691 inflorescences were recognized as structurally distinct forms. All classification


692 assignments are listed in Supplementary Table 4.


693


694 **Phylogenetic comparative analysis**


695 The species phylogeny and branch lengths were obtained from TimeTree &#91;39&#93; .


696 Inflorescence architecture was classified as branched or unbranched as described


697 above. CNS architecture and CArG-box state were coded using the definitions in


32


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.18.767244; this version posted November 19, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.18.767244)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC-ND 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc-nd/4.0/)


698 Supplementary Tables 6 and 5, respectively. The ancestral CNS configuration of core


699 Poaceae was interpreted under parsimony: because phylogenetically distant


700 unbranched lineages, including Pooideae wheat relatives and the PACMAD


701 chloridoid _Zoysia japonica_, share an identical CNS2–CNS3 configuration, the


702 conserved configuration was treated as the ancestral state for core Poaceae.


703 For the comparative analysis of cis-regulatory features across species, five


704 binary features were scored for each of 24 Poaceae species and summed to generate a


705 composite divergence score (0–5): CNS2–CNS3 spacing change (&gt;110 bp,


706 corresponding to the maximum CNS2–CNS3 spacing observed among unbranched


707 species &#91;110 bp in _Eremochloa ophiuroides_ &#93;, or translocation to a different


708 chromosome), CNS2 structural change (&gt;30 bp), CNS3 structural change (&gt;30 bp),


709 CArG box 1 nucleotide substitution from the canonical motif (CCATTTTTG), and


710 CArG box 2 change (nucleotide substitution relative to the ancestral motif


711 CTGAAATG, deletion, or motif absence). CArG box 2 changes were further


712 classified as minor (1–3 nucleotide substitutions relative to the ancestral motif) or


713 major ( ≥ 4 nucleotide substitutions, deletion, or absence of an identifiable motif).


714 Two composite divergence scores were calculated, one including both minor and


715 major CArG box 2 changes and the other including only major changes. For


716 polyploid species with multiple ohnologs (WGD-derived paralog), a feature was


717 scored as changed if it was altered in at least one ohnolog.


718 The association between composite divergence score and inflorescence type


719 (branched or unbranched) was tested using phylogenetic generalized least squares


720 (PGLS) under a Brownian motion model, implemented in Python using a


721 phylogenetic covariance matrix derived from root-to-tip and patristic distances


33


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.18.767244; this version posted November 19, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.18.767244)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC-ND 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc-nd/4.0/)


722 computed from the TimeTree phylogeny. Regression coefficients were estimated by


723 generalized least squares using SciPy (version 1.13). Individual features were tested


724 using the same framework. Full results are provided in Supplementary Table 7.


725


726 **Data availability**


727 Raw panicle measurements, species coding, phylogenetic trees, expression data,


728 analysis scripts, CNS sequences across analyzed grass species, and genome-edited


729 line sequences are available on Zenodo &#91;40&#93; .


730


731 **Code availability**


732 Analysis scripts are available on Zenodo &#91;40&#93; .


733


734 **Acknowledgements**


735 We thank Dr. Kuroda for providing vectors, pZNH2GTR-U6, pZNH2GTR(NG)-U6


736 and pUC198AMdE-U61, U62, and U63.


737


738 **Funding statement**


739 This research was supported by the research program on development of innovative


740 technology grant JPJ011937 from the Project of the Bio-oriented Technology


741 Research Advancement Institution (BRAIN).


742


743 **Author Contributions**


34


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.18.767244; this version posted November 19, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.18.767244)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC-ND 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc-nd/4.0/)


744 T.M. and J.K. conceived and designed the project. T.M., K.T., A.M., K.S., T.T., and


745 J.K. conducted experiments. T.M. analyzed the data. T.M. wrote the first draft of the


746 manuscript. K.T., T.T., and J.K. contributed to manuscript revision and finalization.


747


748 **Competing interests**


749 T.M. and J.K. are inventors on patent applications related to this work (JP 2023

750 118524; PCT/JP2024/024726, published as WO2025/018218). The other authors


751 declare that there are no competing interests.


752


753 **References**


754 1 Bommert, P. &amp; Whipple, C. Grass inflorescence architecture and meristem
755 determinacy. _Semin Cell Dev Biol_ **79**, 37-47 (2018).
756 2 Kyozuka, J., Tokunaga, H. &amp; Yoshida, A. Control of grass inflorescence form
757 by the fine-tuning of meristem phase change. _Curr Opin Plant Biol_ **17**, 110-115
758 (2014).
759 3 Tanaka, W., Yamauchi, T. &amp; Tsuda, K. Genetic basis controlling rice plant
760 architecture and its modification for breeding. _Breed Sci_ **73**, 3-45 (2023).
761 4 Wray, G. A. The evolutionary significance of cis-regulatory mutations. _Nat Rev_
762 _Genet_ **8**, 206-216 (2007).
763 5 Ikeda, K., Nagasawa, N. &amp; Nagato, Y. ABERRANT PANICLE
764 ORGANIZATION 1 temporally regulates meristem identity in rice. _Dev Biol_
765 **282**, 349-360 (2005).
766 6 Ookawa, T. _et al._ New approach for rice improvement using a pleiotropic QTL
767 gene for lodging resistance and yield. _Nat Commun_ **1**, 132 (2010).
768 7 Samach, A. _et al._ The UNUSUAL FLORAL ORGANS gene of Arabidopsis
769 thaliana is an F-box protein required for normal patterning and growth in the
770 floral meristem. _Plant J_ **20**, 433-445 (1999).
771 8 Ikeda, K., Ito, M., Nagasawa, N., Kyozuka, J. &amp; Nagato, Y. Rice ABERRANT
772 PANICLE ORGANIZATION 1, encoding an F-box protein, regulates meristem
773 fate. _Plant J_ **51**, 1030-1040 (2007).
774 9 Ikeda-Kawakatsu, K. _et al._ Expression level of ABERRANT PANICLE
775 ORGANIZATION1 determines rice inflorescence form through control of cell
776 proliferation in the meristem. _Plant Physiol_ **150**, 736-747 (2009).
777 10 Zhang, S. _et al._ Genome Editing of the qPL6 Promoter Creates Novel Alleles for
778 High-Yielding Rice. _Rice (N Y)_ **18**, 46 (2025).
779 11 Kuzay, S. _et al._ WAPO-A1 is the causal gene of the 7AL QTL for spikelet
780 number per spike in wheat. _Plos Genet_ **18**, e1009747 (2022).


35


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.18.767244; this version posted November 19, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.18.767244)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC-ND 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc-nd/4.0/)


781 12 Freeling, M. &amp; Subramaniam, S. Conserved noncoding sequences (CNSs) in
782 higher plants. _Curr Opin Plant Biol_ **12**, 126-132 (2009).
783 13 Van de Velde, J., Van Bel, M., Vaneechoutte, D. &amp; Vandepoele, K. A
784 Collection of Conserved Noncoding Sequences to Study Gene Regulation in
785 Flowering Plants. _Plant Physiol_ **171**, 2586-2598 (2016).
786 14 Salvi, S. _et al._ Conserved noncoding genomic sequences associated with a
787 flowering-time quantitative trait locus in maize. _Proc Natl Acad Sci U S A_ **104**,
788 11376-11381 (2007).
789 15 Song, B. _et al._ Conserved noncoding sequences provide insights into regulatory
790 sequence and loss of gene expression in maize. _Genome Res_ **31**, 1245-1257
791 (2021).
792 16 Amundson, K. R. _et al._ A deep-time landscape of plant cis-regulatory sequence
793 evolution. _Science_ **392**, eadt8983 (2026).
794 17 Zhou, J. _et al._ An efficient CRISPR-Cas12a promoter editing system for crop
795 improvement. _Nat Plants_ **9**, 588-604 (2023).
796 18 Lanctot, A., Hendelman, A., Udilovich, P., Robitaille, G. M. &amp; Lippman, Z. B.
797 Antagonizing cis- regulatory elements of a conserved flowering gene mediate
798 developmental robustness. _P Natl Acad Sci USA_ **122**, e2421990122 (2025).
799 19 Kuroha, T. _et al._ Modification of TAWAWA1-mediated panicle architecture by
800 genome editing of a downstream conserved noncoding sequence in rice. _Plant_
801 _Biotechnol J_ **23**, 2667-2669 (2025).
802 20 Smaczniak, C., Immink, R. G. H., Angenent, G. C. &amp; Kaufmann, K.
803 Developmental and evolutionary diversity of plant MADS-domain factors:
804 insights from recent studies. _Development_ **139**, 3081-3098 (2012).
805 21 Gao, X. _et al._ The SEPALLATA-like gene OsMADS34 is required for rice
806 inflorescence and spikelet development. _Plant Physiol_ **153**, 728-740 (2010).
807 22 Zhu, W. _et al._ Rice SEPALLATA genes OsMADS5 and OsMADS34 cooperate
808 to limit inflorescence branching by repressing the TERMINAL FLOWER1-like
809 gene RCN4. _New Phytol_ **233**, 1682-1700 (2022).
810 23 Kobayashi, K. _et al._ Inflorescence meristem identity in rice is specified by
811 overlapping functions of three AP1/FUL-like MADS box genes and PAP2, a
812 SEPALLATA MADS box gene. _Plant Cell_ **24**, 1848-1859 (2012).
813 24 Zhu, T. _et al._ Comprehensive mapping and modelling of the rice regulome
814 landscape unveils the regulatory architecture underlying complex traits. _Nat_
815 _Commun_ **15**, 6562 (2024).
816 25 Miao, Y. _et al._ ABERRANT PANICLE ORGANIZATION2 controls multiple
817 steps in panicle formation through common direct-target genes. _Plant Physiol_
818 **189**, 2210-2226 (2022).
819 26 Lai, X. L. _et al._ The intervening domain is required for DNA-binding and
820 functional identity of plant MADS transcription factors. _Nat Commun_ **12**, 4760
821 (2021).
822 27 Yoshida, A. _et al._ TAWAWA1, a regulator of rice inflorescence architecture,
823 functions through the suppression of meristem phase transition. _Proc Natl Acad_
824 _Sci U S A_ **110**, 767-772 (2013).
825 28 Vegetti, A. &amp; Anton, A. M. Some Evolution Trends in the Inflorescence of
826 Poaceae. _Flora_ **190**, 225-228 (1995).


36


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.18.767244; this version posted November 19, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.18.767244)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC-ND 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc-nd/4.0/)


827 29 Poursarebani, N. _et al._ COMPOSITUM 1 contributes to the architectural
828 simplification of barley inflorescence via meristem identity signals. _Nat_
829 _Commun_ **11**, 5138 (2020).
830 30 Kellogg, E. A. _et al._ Early inflorescence development in the grasses (Poaceae).
831 _Front Plant Sci_ **4**, 250 (2013).
832 31 Siepel, A. _et al._ Evolutionarily conserved elements in vertebrate, insect, worm,
833 and yeast genomes. _Genome Res_ **15**, 1034-1050 (2005).
834 32 Wang, J. _et al._ A high-quality chromosome-scale assembly of the centipedegrass
835 &#91;Eremochloa ophiuroides (Munro) Hack.&#93; genome provides insights into
836 chromosomal structural evolution and prostrate growth habit. _Hortic Res_ **8**, 201
837 (2021).
838 33 Tanaka, H. _et al._ Sequencing and comparative analyses of the genomes of
839 zoysiagrasses. _DNA Res_ **23**, 171-180 (2016).
840 34 Ma, P. F. _et al._ The Pharus latifolius genome bridges the gap of early grass
841 evolution. _Plant Cell_ **33**, 846-864 (2021).
842 35 Seetharam, A. S. _et al._ The Streptochaeta Genome and the Evolution of the
843 Grasses. _Front Plant Sci_ **12**, 710383 (2021).
844 36 Tareen, A. &amp; Kinney, J. B. Logomaker: beautiful sequence logos in Python.
845 _Bioinformatics_ **36**, 2272-2274 (2020).
846 37 Kuroda, M., Kimizu, M. &amp; Mikami, C. A simple set of plasmids for the
847 production of transgenic plants. _Biosci Biotechnol Biochem_ **74**, 2348-2351
848 (2010).
849 38 Tsuda, K. _et al._ YABBY and diverged KNOX1 genes shape nodes and
850 internodes in the stem. _Science_ **384**, 1241-1247 (2024).
851 39 Kumar, S. _et al._ TimeTree 5: An Expanded Resource for Species Divergence
852 Times. _Mol Biol Evol_ **39**, msac174 (2022).
853 40 Makino, T. _et al._ _APO1_ CNS Cis-regulatory Module - Data and Analysis Scripts.
854 Zenodo (2026). https://doi.org/10.5281/zenodo.23161144


37


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.18.767244; this version posted November 19, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.18.767244)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC-ND 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc-nd/4.0/)


855 **Figure legends**
856
857 **Fig. 1|A cis-regulatory module with paired CArG boxes flanks rice** _**APO1**_ **.** **a,**
858 Genomic organization of the rice _APO1_ locus and chromatin accessibility across the
859 regulatory region. Upper: the _APO1_ gene model and transcriptional orientation, CNS1,
860 CNS2, and CNS3 with their distances from _APO1_, and Poaceae conservation tracks
861 (Monocot and BEP clade). Lower: chromatin-accessibility profiles from the CART
862 dataset &#91;24&#93; in panicle, SAM, leaf, and root tissues. **b,** Multiple-sequence alignment of the
863 CArG box 1 and CArG box 2 intervals within CNS3 across representative Poaceae
864 species. Conserved positions, substitutions, indels, and motif boundaries are indicated.
865 Full-length CNS alignments for representative species are shown in Supplementary Fig.
866 1. **c,** Sequence logos of CArG box 1 (n = 25) and CArG box 2 (n = 23) across the
867 analyzed Poaceae species. Height at each position represents information content in bits.
868 Species for which a motif was absent or undetectable were excluded.
869
870 **Fig. 2|CNS3 editing alters panicle branching, lateral organ fate, and APO1**
871 **expression.** **a,** CNS3 edited alleles showing phastCons conservation scores, CArG
872 boxes 1 and 2, and guide RNA target sites (inverted triangles). Each bar represents a
873 homozygous line; deletions (dark blue), substitutions (sequence colors), insertions
874 (pink), and inversions (purple) are indicated. **b,** Panicle architecture showing primary
875 branches (PBs), secondary branches (SBs), and terminal (black), lateral (gray), and SB876 borne (red) spikelets. **c,** secondary branches per lateral organ, **d,** primary branch
877 number, and **e,** total spikelet number **.** Small open circles indicate individual panicles
878 and large filled circles plant means (up to five panicles from each of five plants per
879 line). Plant means were used as independent biological replicates. For **c,** ratios were
880 calculated for each primary branch and averaged within panicles. Plants within each line
881 are arranged in ascending trait order; lines are ordered by mean secondary branches per
882 lateral organ and colored by mutation class. Asterisks indicate significant differences
883 from wild type (Mann–Whitney U test, Benjamini–Hochberg-adjusted p &lt; 0.05; exact
884 values in Supplementary Table 2). Phenotypes of CNS1, CNS2, and CNS2–CNS3
885 intergenic mutants are shown in Supplementary Fig. 2. **f,** Representative panicles from
886 lines 1 (wild type), 21 (other CNS3 mutant), 8 (central CNS3 mutant), 41 (CArG box 2
887 deletion), 24 (CArG box 1 deletion), and 32 (CArG box 1+2 deletion). Scale bar, 5 cm.
888 g, h, Spatial distributions of lateral organs (g) and spikelets (h) across five normalized
889 primary-branch positional bins (0, basal; 1, apical). Bars distinguish secondary branches
890 (colored) from direct-spikelet lateral organs (gray) in **g,** and SB-borne (colored) from
891 direct (gray) spikelets in **h.** Terminal spikelets were excluded from g but included in h.
892 Mutation-class values are unweighted means of line means, with error bars indicating
893 SEM among line means for components and totals. Colors indicate mutation classes,
894 and numbers of independent lines are shown. i, _APO1_ expression measured by qRT895 PCR and normalized to OsUBQ5 (n = 3 biological replicates per genotype). Points
896 indicate replicates; bars show mean ± SD. **j,** Genotype-mean _APO1_ expression versus
897 secondary branches per lateral organ; **k,** primary branch number; and **l,** total spikelet
898 number. Points represent genotypes (n = 8), labeled by line number and colored by
899 mutation class. Dashed lines guide the eye; Spearman's ρ and P values are indicated.
900


38


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.18.767244; this version posted November 19, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.18.767244)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC-ND 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc-nd/4.0/)


901 **Fig. 3|MADS15–MADS34 binding to** _**APO1**_ **CNS3 is sensitive to disruption of the**
902 **two CArG boxes. a,** CNS3 probe design. Wild type and mutant probes used to examine
903 the contribution of the two CArG boxes to MADS15–MADS34 binding are shown. The
904 mutant probes contain a single-nucleotide deletion in CArG box 1 (mut1), a single905 nucleotide deletion in the sequence between the two CArG boxes (mut2), a single906 nucleotide deletion in CArG box 2 (mut3), or the combined mut1 and mut3 deletions
907 (mut1&amp;3). CArG-box sequences are shown in red, deleted nucleotides are indicated in
908 gray, and gray circles above the sequences mark the positions of the introduced
909 deletions. **b,** EMSA of MADS14, MADS15, MADS18, and MADS34 with the wild
910 type CNS3 probe. The proteins were assayed individually and in the indicated
911 combinations. MADS15–MADS34 produced the strongest shifted signal among the
912 combinations tested. **c,** EMSA of the wild type, mut1, mut2, mut3, and mut1&amp;3 CNS3
913 probes with MADS15 and MADS34. Single-nucleotide deletion of either CArG box
914 reduced but did not abolish detectable binding, whereas mut2 retained binding.
915 Simultaneous deletion of both CArG boxes (mut1&amp;3) abolished detectable binding
916 under the assay conditions. **d,** EMSA using CNS3 probes corresponding to the actual
917 genome-edited alleles. Probes derived from the wild type and edited lines 28, 27, 31, 10,
918 30, 29, 24, 23, 26, 11, 32, and 41 were assayed in the presence or absence of MADS15
919 and MADS34. The wild type and line 28 probes showed clear shifted signals, whereas
920 line 27 showed weak residual binding. Detectable binding was abolished in lines 23, 24,
921 and 26 carrying CArG-box-1 deletions, line 41 carrying a CArG-box-2 deletion, and
922 lines 10, 11, 29, 30, 31, and 32 carrying deletions affecting both CArG boxes.
923
924 **Fig. 4|Recurrent modification of the** _**APO1**_ **cis-regulatory module during grass**
925 **inflorescence evolution.** **a,** Diagrams and photographs of unbranched (spike or spike926 like raceme) and branched (panicle or branched digitate) inflorescences from the BEP
927 and PACMAD clades. Photographs were cropped to highlight inflorescence
928 architecture. The _Oryza sativa_ photograph was taken by the authors; others were
929 obtained from Wikimedia Commons: _Aegilops tauschii_, Vengolis (CC BY-SA 3.0);
930 _Sorghum bicolor_, Stefan.lefnaer (CC BY-SA 4.0); and _Zoysia japonica_, bastus917 (CC
931 BY-SA 2.0). **b,** Species phylogeny showing inflorescence architecture and _APO1_ cis932 regulatory structures. Internal branch symbols and CNS structures represent putative
933 ancestral states based on parsimony. Orange and green indicate unbranched and
934 branched architectures, respectively. Extant species are labeled U (unbranched) or B
935 (branched). CNS2 (blue) and CNS3 (red) are positioned according to their intergenic
936 distances. Black segments indicate internal CNS deletions or terminal regions lacking
937 detectable similarity (potential deletion or sequence divergence). Purple and pink
938 triangles indicate changes in CArG boxes 1 and 2, respectively. Structural changes,
939 translocations, and intergenic distances are annotated. **c,** Alignment of the _APO1_ cis940 regulatory region between the unbranched BEP species _Aegilops tauschii_ and
941 PACMAD species _Zoysia japonica_ . CNS2 (blue), CNS3 (red), and both CArG boxes
942 are highlighted; asterisks indicate identical nucleotides. **d,** Changes in five regulatory
943 features relative to the inferred ancestral state: CNS2–CNS3 spacing, CNS2 structure,
944 CNS3 structure, CArG box 1, and CArG box 2. Red squares indicate changes; absence
945 of a square indicates conservation. For CArG box 2, filled squares represent major
946 changes ( ≥ 4 substitutions, deletion, or motif absence) and open squares minor changes


39


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.18.767244; this version posted November 19, 2026. The copyright holder for this preprint&#93;(https://doi.org/10.64898/2026.11.18.767244)

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
&#91;available under a CC-BY-NC-ND 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc-nd/4.0/)


947 (1–3 substitutions). Gray bars show the composite divergence score (number of changed
948 features, 0–5). U and B indicate unbranched and branched species.
949
950 **Fig. 5|Model linking the CNS3 CArG-box module to the spatial domain of**
951 **secondary-branch formation.** MADS15–MADS34 binding to intact CArG boxes is
952 associated with lower _APO1_ expression and restriction of secondary-branch fate to mid953 apical primary branches. CArG-box disruption reduces MADS15–MADS34 binding
954 and is associated with higher _APO1_ expression, shifting lateral meristem fate toward
955 secondary branches and expanding the secondary-branching zone toward basal primary
956 branches.


40


### **a** **b**



Conservation


Chromatin
accessibility


_O. sativa_


_L. perrieri_


_H. vulgare_


_A. tauschii_


_S. bicolor_


_S. italica_


_P. hallii_







CArG box 1 CArG box 2



_E. ophiuroides_


_Z. japonica_

### **c**


## **Figure 1**


### **a**

### **g** **h**



1 = Apical


0 = Basal


1 = Apical


0 = Basal









**Line1 (WT)** **Line21** **Line8** **Line41** **Line24** **Line32**


### **i**


## **Figure 2**


### **a**


# ・ ・・



WT GCATTTTCGCGTTTGCTCTGCTACTGCTACTCCACTTCCATTTTTGCACAGTGCTATCCCTTCCTGAAATGCTTCTTCGTACATTCGCTCATCAATCGTTTAGAGAG


mut1 GCATTTTCGCGTTTGCTCTGCTACTGCTACTCCACTT_CATTTTTGCACAGTGCTATCCCTTCCTGAAATGCTTCTTCGTACATTCGCTCATCAATCGTTTAGAGAG


mut2 GCATTTTCGCGTTTGCTCTGCTACTGCTACTCCACTTCCATTTTTGCACAGTGCTAT_CCTTCCTGAAATGCTTCTTCGTACATTCGCTCATCAATCGTTTAGAGAG


mut3 GCATTTTCGCGTTTGCTCTGCTACTGCTACTCCACTTCCATTTTTGCACAGTGCTATCCCTTC_TGAAATGCTTCTTCGTACATTCGCTCATCAATCGTTTAGAGAG


mut1&amp;3 GCATTTTCGCGTTTGCTCTGCTACTGCTACTCCACTT_CATTTTTGCACAGTGCTATCCCTTC_TGAAATGCTTCTTCGTACATTCGCTCATCAATCGTTTAGAGAG


CArG box 1 CArG box 2


### **b c**



**WT** **mut1** **mut2** **mut3** **mut1&amp;3**


### **d WT L28 L27 L31 L10 L30 L29 WT L24 L23 L26 L11 L32 L41**

MADS15+34 −  + −  + −  + −  + −  + −  + −  + −  + −  + −  + −  + −  + −  + −  +


## **Figure 3**


### **a**

### **c**



**U** nbranched


Spike or
Spike-like raceme


**CNS2**



BEP PACMAD **B** ranched BEP PACMAD


Panicle or

Branched digitate

### **d**



BEP **U**


## **Figure 4**



PACMAD



**U**



A. tauschii TTTGTTGCTTCCGTGGAGGACAACCACACTA---GTACTTTGGA--CAGTA-CAGTGCAGTGCAGGAGTAGTGGCTACGGCTGTGGCTAAACCTAGCTAGCCAGGGCACTTCCTAGCTGCTGAACTCTAGCTCCAACGAATTGACATGCATGAAGTAGCCAACTCGAAGCTTATCACAGGGCCATCAGCCTTTCTCTTCCAAAAGGCTGTCACTGAGATACTGAGATG


Z. japonica TTTGTTGCGCCTGCGGTGGACAACGAGACTACTCGTACTGTAGATCTAGTAGCAGTGCAGGAGCTGAGTAGTGGCTA-GGCCATGGCATGCAGGACCTAGCTAGGGCACTTGGTAGCTG--------------------GACGATTGACATGGAGTAGCCAACACAGAACGAAC----------GCAGGCTTTCCCTTCCAAAAAGCTGAGAGCGAGGGAGAGACGAG


********..*.*.**.*******.*.****  *****.*.** .**** ********.....************ ***..****.......*.*****.*********..******          ...**....****.**********.*..*.*..*.     .***.*****.*********.****..*..***..*..**...*


GAGAGAGAGAGAGAGAAAGAGAGAG---------------------AGAGACCTGCATGCATGCCTGCCCATACATCGTCACCCCCATCGCCGAGCACCAAC---CGGTTTGCCTGTTACCACTTGCGCTCTATTGTCGCTGTCCCTCCCGTCTTCTCCCGGCATTTTCGTGTTACTCCATTTCCATTTTTGCACAGTGATTCTCCCTTTCTAAAATACCCCCAAGTC


GCCAAAGATTGAGTTGACGCGAACGCAGCATGAGTATGACCATGCCAGCAGCTAGCAAGCTTACCGGCCAATCGATCAC--------TCGCCAAGCATCGATATGCAGTTTGTCCGTTGCCATTCCCACTCTCGTGTCGCTGTCGC-GGCGTTGCGTCCCGCTGCTGC-------CCCCGTTTCCATTTTTGAACAGTGCTGCTCCCTTTCTGAAATGCCGCTTGCAC


*..*.***..***...*.*.**..*           **...*..***.**.*.**.***.**..***..    *****.****.*.*.  *.*****.*.***.***.*..*.****..**********.* ..***....*****....*..    *.**.************.******.*.**********.****.**.*.....*



**CNS3**



CArG box 1 CArG box 2


**Intact CArG box** **CArG box disruption**





_**APO1**_ CArG box _**APO1**_ CArG box



**Lateral meristem fate**

**toward spikelets**


SB

zone



**Lateral meristem fate shifted**
**toward secondary branches (SBs)**





SB zone

expanded



Spikelet fate


Secondary
branch fate


## **Figure 5**


