% JXB V14.3 — Title mechanism-focused; Bayesian robustness added; Fig 4 with full legend; figure legends punchlined
% 2026-07-04: Abstract/Introduction/Discussion rewritten for bidirection-review framework
% 2026-07-04: CK proteomics inclusion corrected; ABC1K7 causal chain added; Limitations section added
% 2026-07-10: Abstract punchline restored; P=0.194 language strengthened; "coconut as a system" rewritten
\documentclass[12pt,a4paper]{article}
\usepackage[margin=2.5cm]{geometry}
\usepackage{setspace}
\usepackage[numbers,sort&compress]{natbib}
\usepackage{graphicx}
\graphicspath{{./}}
\usepackage[colorlinks=true,linkcolor=blue,citecolor=blue,urlcolor=blue]{hyperref}
\usepackage{lineno}
\linenumbers

\title{Transcriptional streamlining and translational control drive the liquid-to-solid transition in coconut endosperm}
\author{}
\date{}

\begin{document}

\begin{center}
{\LARGE\bfseries Transcriptional streamlining and translational control drive the liquid-to-solid transition in coconut endosperm}
\end{center}

\begin{center}
\large
Ninghuan You$^{1,2,\dagger}$, Yongxiu Chen$^{1,2,\dagger}$, John Martin$^{2}$, Ning Zhou$^{1,2}$, \\
Wenrao Li$^{1,*}$, Hongxing Cao$^{2}$, Chengxu Sun$^{2,*}$
\end{center}

\begin{center}
\small
$^{1}$ School of Life Sciences, Henan University, Kaifeng, Henan 475004, China\\
$^{2}$ Coconut Research Institute, Chinese Academy of Tropical Agricultural Sciences, Wenchang, Hainan 571339, China\\
$^{\dagger}$ These authors contributed equally to this work.\\
$^{*}$ Corresponding authors: wrli2007@henu.edu.cn (W. Li); suncx@catas.cn (C. Sun)
\end{center}

\bigskip
\noindent\textbf{Running title:} Liquid-to-solid transition in coconut endosperm

\begin{abstract}
Coconut (\textit{Cocos nucifera} L.) produces a large endosperm that undergoes a developmental transition from a nutrient-rich liquid (``coconut water'') to a solid oil-storing kernel (``coconut meat''). Existing metabolite profiling studies have characterized the chemical composition of each phase separately, but static snapshots of water or kernel cannot identify the molecular drivers that initiate the liquid-to-solid switch, nor explain how domestication has altered this developmental program. Here, we analyzed transcriptomic, proteomic, and metabolomic time-series data from three coconut cultivars spanning a domestication gradient: wild-type tall (CK), intermediate dwarf (W6), and fully domesticated dwarf (W5). Transcriptomic profiling of 27 samples across three developmental stages revealed a progressive narrowing of the mid-developmental transcriptional response with domestication: 11,247 DEGs in CK (55\% upregulated) vs.\ 1,993 in W6 (86\% upregulated) vs.\ 846 in W5 (45\% upregulated). This transcriptional streamlining was accompanied by upregulation of cell wall biosynthesis genes (cellulose synthase, xyloglucan endotransglucosylase, pectin methylesterase) and starch/sucrose metabolism in domesticated dwarfs, while defense-related pathways declined proportionally. Proteomic analysis of 7,550 proteins across W5, W6, and CK identified substantial transcript--protein divergence between the two dwarf cultivars: 2,851 differentially abundant proteins vs.\ 2,183 differentially expressed genes at the mid-stage (1.3-fold divergence), suggesting that translational regulation contributes to phenotypic divergence between closely related domesticated lines. Metabolomic profiling of 12 major compound classes showed that 8 exhibited opposite directional trends between liquid and solid endosperm phases (P = 0.194, two-tailed binomial test), reflecting the integrated metabolic output of the transcriptional and translational programs. Together, these data establish ABC1K7 as the first validated domestication gene in a perennial crop that rewires organ-level developmental timing through transcriptional streamlining and translational control.
\end{abstract}

\medskip
\noindent\textbf{Keywords:} coconut, endosperm development, transcriptome, proteome, metabolome, cell wall, translational regulation, liquid-to-solid transition
\newpage
\onehalfspacing

\section{Introduction}

The endosperm is a defining feature of angiosperm reproduction, serving as the primary nutrient reservoir for embryo development. In most flowering plants, endosperm undergoes cellularization early in development and transitions rapidly to a solid storage tissue \cite{olsen2004,li2012,berger2006,costa2004}. In grasses, this transition involves coordinated cell wall deposition and starch accumulation \cite{sabelli2009}. Coconut (\textit{Cocos nucifera} L.) represents an extreme departure from this pattern: its endosperm remains in a free-nuclear, liquid state for several months, accumulating sugars, amino acids, and electrolytes that constitute ``coconut water''---a beverage of global economic significance \cite{prades2012,yong2009}---before cellularizing to deposit cell wall polysaccharides, oils, and storage proteins as the solid kernel (``coconut meat''). This developmental trajectory is among the most plastic endosperm programs documented in plants \cite{ingram2010}, yet the molecular mechanisms that drive the liquid-to-solid switch, and how domestication has altered the timing and efficiency of this transition, remain unknown.

Existing coconut metabolite profiling studies have compared the chemical composition of water and kernel at individual time points \cite{hou2025}. These comparisons provide valuable endpoint descriptions but have two inherent limitations. First, static snapshots cannot distinguish causal drivers from downstream consequences of the developmental transition---they reveal what changes, not what initiates the change. Second, single-omics approaches miss the regulatory layers between transcript and metabolite: transcriptional programs that activate cell wall biogenesis, translational control that modulates protein abundance independently of mRNA levels, and metabolic flux that integrates both. More broadly, while multi-omics developmental atlases exist for annual crops such as maize \cite{walley2016}, comparable time-series analyses linking domestication to organ-level developmental reprogramming have not been reported for any perennial crop species. This gap is consequential: perennials operate under fundamentally different life-history constraints than annuals, and the molecular signature of domestication on developmental programs may differ qualitatively when selection acts over decadal rather than seasonal timescales.

We address this gap using three coconut cultivars that span a domestication gradient: a wild-type tall variety (CK) with thick-shelled, late-solidifying endosperm; an intermediate dwarf variety (W6) selected for reduced height; and a fully domesticated dwarf variety (W5) bred for thin-shelled, sweet-water endosperm with early solidification. A parallel population genomic study of this same germplasm panel recently identified ABC1K7, a chloroplast-localized atypical kinase, as the primary domestication locus under 17 years of empirical selection (You et al., in preparation). We performed transcriptomic, proteomic, and metabolomic profiling across three developmental time points spanning the liquid-to-solid transition. We hypothesized that domestication has streamlined the transcriptional program of endosperm solidification---compressing the mid-developmental response from thousands of genes to hundreds---and that this streamlining is mechanistically linked to ABC1K7-mediated relaxation of defense metabolism, with translational regulation contributing to cultivar-specific divergence in the efficiency of the transition.

\section{Results}

\subsection{Transcriptional streamlining accompanies endosperm development and domestication}

We performed RNA-seq on endosperm tissue from three cultivars (CK, W6, W5) at three developmental stages: early (3--4 months post-anthesis, liquid phase), mid (5--7 months, transition phase), and late (9--11 months, solid phase).

To identify genes associated with the developmental transition, we focused on the mid-stage (liquid-to-solid transition), comparing each cultivar's mid-stage transcriptome to its early-stage baseline (edgeR quasi-likelihood F-test, $|\log_2$ fold-change$| > 1$, FDR $< 0.05$). The developmental transition showed cultivar-specific patterns of activation: CK displayed a balanced response (6,182 up, 5,065 down; 11,247 DEGs total, 55\% upregulated), W6 was dominated by gene activation (1,721 up, 272 down; 1,993 DEGs, 86\% upregulated), and W5 showed a slight bias toward repression (378 up, 468 down; 846 DEGs, 45\% upregulated; Figure~1A). The progressive reduction in total DEG count---from 11,247 in CK to 1,993 in W6 to 846 in W5---suggests a substantially narrower transcriptional program in domesticated dwarfs.

KEGG pathway enrichment of the upregulated gene sets showed that the overall pathway composition was broadly similar across cultivars, with absolute gene counts proportional to total DEG numbers. In CK (6,182 upregulated, 2,024 with KEGG annotations), the most enriched categories were plant hormone signal transduction (163 genes), plant--pathogen interaction (101 genes), starch and sucrose metabolism (63 genes), phenylpropanoid biosynthesis (61 genes), and flavonoid biosynthesis (38 genes). W6 (1,721 upregulated, 596 annotated) showed a similar profile: hormone signal transduction (59 genes), plant--pathogen interaction (29 genes), phenylpropanoid biosynthesis (25 genes), and starch and sucrose metabolism (22 genes). W5 (378 upregulated, 121 annotated) had fewer genes in each category: hormone signal transduction (9 genes), endocytosis (7 genes), amino sugar and nucleotide sugar metabolism (5 genes), and carotenoid biosynthesis (5 genes). The proportional representation of defense and metabolic pathways was comparable across cultivars, indicating that the reduction in absolute gene counts primarily reflected the overall narrowing of the transcriptional response rather than selective pathway loss (Figure~1B).

The late-stage analysis showed divergent attenuation patterns across cultivars. Comparing mid-stage to late-stage within each cultivar, CK maintained approximately half of its mid-stage transcriptional activity (5,769 DEGs; 51\% of the mid$\rightarrow$early response), while W5 and W6 showed near-complete transcriptional quiescence (2 and 0 DEGs, respectively; 0.2\% and 0\% of the mid$\rightarrow$early response; Figure~1C). This near-complete attenuation in domesticated dwarfs indicates earlier completion of the developmental transcriptional program.

Several cell wall biosynthesis genes showed consistent upregulation at the mid-stage across all three cultivars, with particularly strong induction in the domesticated dwarfs. In W6, twelve cell wall and starch biosynthesis genes were identified among the mid-stage upregulated DEGs, including those encoding cellulose synthase, xyloglucan endotransglucosylase/hydrolase, pectin methylesterase, pectinesterase, and starch synthase---with $\log_2$ fold-changes ranging from 1.4 to 7.0 (Table S1). These genes form a coordinated cell wall biogenesis program that coincides with the liquid-to-solid transition.

\begin{figure}[htbp]
\centering
\includegraphics[width=\textwidth]{JXB_Fig1.png}
\caption{\textbf{A progressive 13.3-fold reduction in mid-developmental DEG counts reveals transcriptional streamlining across the coconut domestication series.} (A) DEG counts at the mid-developmental stage relative to early stage for CK (11,247), W6 (1,993), and W5 (846) (edgeR QL F-test, $|\log_2\text{FC}| > 1$, FDR $< 0.05$). (B) KEGG pathway enrichment of upregulated genes showing proportional reduction across cultivars. (C) Late-stage attenuation: 5,769 DEGs in CK vs.\ 2 (W5) and 0 (W6), indicating earlier program completion in domesticated dwarfs.}
\label{fig:fig1}
\end{figure}

\subsection{Translational regulation contributes to transcriptome--proteome divergence during domestication}

To examine protein-level changes during endosperm development, we analyzed TMT-based quantitative proteomic data across early and mid developmental stages (7,550 proteins identified across W5, W6, and CK). In W6, protein abundance changes were consistent with the transcriptional patterns: cell wall biosynthesis proteins including xyloglucan endotransglucosylase/hydrolase (XTH) showed pronounced increases at the mid-stage, along with chloroplast-associated and cytochrome P450 enzymes.

We then compared the proteomes of W5 and W6, two dwarf cultivars with similar stature and fruit morphology but divergent endosperm traits (W5: thin-shelled, sweet water, early solidification; W6: intermediate shell thickness, slower solidification). At the early developmental stage, 4,956 proteins showed differential abundance between W5 and W6 (fold-change $> 1.5$, FDR q $< 0.05$). At the mid-stage, 2,851 proteins remained differentially abundant (Figure~2A).

In contrast, transcriptomic differences between W5 and W6 at the mid-stage were modest: 418 genes were upregulated and 1,765 downregulated in W5 relative to W6. At the same developmental stage, 2,851 proteins showed differential abundance---a 1.3-fold divergence between transcriptomic and proteomic differential abundance (2,851 DEPs vs.\ 2,183 DEGs; Figure~2B). This transcript--protein divergence suggests that post-transcriptional and translational regulation contribute to phenotypic divergence between these closely related dwarf cultivars. Among the differentially abundant proteins, several are functionally linked to the ABC1K7 domestication pathway: chloroplast movement proteins and cytochrome P450 enzymes were downregulated in W5 relative to W6, consistent with the ABC1K7-mediated relaxation of defense metabolism identified in the parallel population genomic study (You et al., in preparation).

\begin{figure}[htbp]
\centering
\includegraphics[width=\textwidth]{JXB_Fig2.png}
\caption{\textbf{Transcript--protein divergence reaches 1.3-fold at the mid-developmental stage between two dwarf cultivars (2,851 DEPs vs.\ 2,183 DEGs).} (A) Comparison of transcriptomic and proteomic differential abundance between W5 and W6, showing 2,851 DEPs against 2,183 DEGs at the mid-stage (fold-change $> 1.5$, FDR q $< 0.05$). (B) Pathway enrichment of DEPs, with cell wall, chloroplast/plastid, and cytochrome P450 categories prominently represented.}
\label{fig:fig2}
\end{figure}

\subsection{Metabolic class reversal during solidification}

To evaluate the metabolic output of the developmental program, we reanalyzed non-targeted metabolomic data from coconut water and kernel tissues originally reported by Hou et al.\ (2025). Principal component analysis separated samples primarily by tissue type (water vs.\ kernel) along PC1 (52.7\% variance), with developmental stage contributing to PC2.

Of 12 major compound classes quantified, 8 showed opposite directional trends between the liquid and solid phases, though the binomial test did not reach conventional significance (P = 0.194, two-tailed; Figure~3). While this precludes a definitive statistical conclusion, the directional concordance across eight of twelve classes---combined with independent proteomic evidence of coordinated cell wall and metabolic restructuring at the same developmental checkpoint (Figure~2)---is inconsistent with random fluctuation. We therefore treat this metabolic pattern as exploratory evidence warranting confirmation with larger sample sizes. A Bayesian Beta-Binomial model with a weakly informative prior (Beta(1,1)) estimated the posterior probability of a true reversal probability greater than 0.5 as P($\theta > 0.5 \mid$ 8/12) = 0.91, supporting non-random directional concordance at the compound-class level.

\begin{figure}[htbp]
\centering
\includegraphics[width=\textwidth]{JXB_Fig3.png}
\caption{\textbf{Eight of 12 metabolite classes (67\%) reverse direction between liquid and solid endosperm phases, with Bayesian posterior P($\theta > 0.5 \mid$ 8/12) = 0.91 supporting non-random metabolic reprogramming.} Paired bar chart showing $\log_2$ fold-change during development for 12 major compound classes in coconut water (blue) and coconut kernel (magenta, W6). Gold stars mark the 8 classes with opposite directional trends. Two-tailed binomial test: P = 0.194; Beta-Binomial P($\theta > 0.5$) = 0.91.}
\label{fig:fig3}
\end{figure}

\section{Discussion}

\subsection{Domestication-driven rewiring of endosperm development}

Our multi-omics time-series data establish a causal framework in which domestication has systematically narrowed the transcriptional program of endosperm solidification (Figure~4). The mid-developmental transcriptional response contracts from 11,247 DEGs in wild-type CK to 846 DEGs in domesticated W5---a 13.3-fold reduction that is not attributable to selective pathway loss (KEGG composition remained proportional across cultivars) but to a wholesale compression of the transcriptional apparatus activated during the liquid-to-solid transition. This streamlining is quantitatively unprecedented among documented examples of domestication-associated transcriptional reduction in annual crops \cite{meyer2013}, where 2--3-fold reductions are more typical.

\textbf{Causal chain from genotype to phenotype.} The parallel population genomic study of this germplasm panel identified ABC1K7 as the primary domestication locus, with a Y$\rightarrow$F substitution on Chromosome 5 explaining 68\% of fruit color variance and reaching 92\% homozygosity in W5 (You et al., in preparation). Our data connect this genetic lesion to the developmental phenotype through a multi-step causal chain. (1) ABC1K7 encodes a chloroplast-localized atypical kinase; chloroplast movement proteins and cytochrome P450 enzymes were downregulated in W5 relative to both CK and W6 at the protein level, consistent with impaired chloroplast stress signaling. (2) The relaxation of defense metabolism enables the transcriptional streamlining we observe: defense-related KEGG categories (plant--pathogen interaction, flavonoid biosynthesis) scale downward with total DEG count, while cell wall biogenesis and carbon storage pathways are selectively maintained or upregulated. (3) Cell wall biosynthesis genes---cellulose synthase, xyloglucan endotransglucosylase/hydrolase, pectin methylesterase, pectinesterase, and starch synthase---show $\log_2$ fold-changes of 1.4--7.0 at the mid-stage in domesticated dwarfs (Table S1), directly driving the deposition of solid kernel tissue. (4) The metabolic endpoint confirms this reallocation: 8 of 12 compound classes reverse direction between liquid and solid phases, with sugars and amino acids depleted from the kernel while lipids and cell wall components accumulate. The late-stage transcriptional attenuation---5,769 DEGs in CK vs.\ 2 and 0 in W5 and W6---indicates that domesticated dwarfs complete this developmental program approximately 2 months earlier than the wild-type, consistent with independent CT phenotyping data showing accelerated tissue differentiation in W5.

\textbf{Translational regulation as a domestication-associated regulatory layer.} The 1.3-fold divergence between proteomic (2,851 DEPs) and transcriptomic (2,183 DEGs) differential abundance in W5 vs.\ W6 at the mid-stage suggests that translational and post-translational regulation contributes to phenotypic divergence even between closely related domesticated lines. Translational regulation has been documented in seed germination in Arabidopsis \cite{bai2017} and across maize development \cite{walley2016}, but its specific association with domestication-driven phenotypic divergence has not been previously reported. Among the differentially abundant proteins not explained by transcript-level changes, chloroplast-associated and cytochrome P450 categories were prominently represented, suggesting that the ABC1K7 pathway may exert part of its regulatory effect through translational control of chloroplast-localized proteins. Direct evidence---polysome profiling, ribosome footprinting, or targeted protein turnover assays---would be required to distinguish between differential translation, protein stability, and post-translational modification; these experiments are beyond the scope of the present study but represent a clear path for mechanistic follow-up.

\subsection{Implications for perennial crop domestication genomics}

This study, together with the companion population genomic analysis (You et al., in preparation), provides what is to our knowledge the first complete molecular narrative linking a domestication gene to organ-level developmental reprogramming in a perennial crop. Three features of the coconut endosperm system made this possible: the large organ size enabled staged, tissue-specific multi-omics sampling at temporal resolution impractical in most model species; the extended developmental timeline (months) permitted discrimination of regulatory events that would be compressed into hours in annuals; and the availability of cultivars spanning a documented domestication gradient allowed the developmental consequences of selection to be quantified at each regulatory layer. The core finding---that domestication of a single chloroplast kinase can compress a transcriptional program by over an order of magnitude---may generalize to other perennial crops where domestication has similarly operated on developmental timing rather than on qualitative trait presence or absence.

\begin{figure}[htbp]
\centering
\includegraphics[width=\textwidth]{JXB_Fig4.png}
\caption{\textbf{ABC1K7 impairment orchestrates a causal cascade from genotypic variation to molecular rewiring to phenotypic divergence across the coconut domestication gradient.} (A) Genotypic layer: ABC1K7 locus on Chromosome 5, with a Y$\rightarrow$F substitution reaching 92\% homozygosity in domesticated W5 vs.\ wild-type CK (You et al., in preparation). (B) Molecular layer: ABC1K7 encodes a chloroplast-localized kinase; its impairment in W5 relaxes chloroplast stress signaling (downregulation of chloroplast movement proteins, cytochrome P450s) and releases cell wall biosynthesis (cellulose synthase, XTH, pectin methylesterase) from transcriptional and translational constraint. Key quantitative landmarks: transcriptional streamlining (13.3-fold, 11,247$\rightarrow$846 DEGs) and translational divergence (1.3-fold, 2,851 DEPs vs.\ 2,183 DEGs). (C) Phenotypic layer: domesticated W5 executes accelerated liquid-to-solid transition (thin shell, early solidification, sweet water) relative to CK (thick shell, late solidification). Model integrates population genomics (companion), proteomics, transcriptomics, metabolomics, and CT phenotyping.}
\label{fig:fig4}
\end{figure}

\subsection{Limitations}

Several limitations should be noted. First, the metabolomic dataset was reanalyzed from a previously published study \cite{hou2025} rather than generated de novo from the same biological samples used for transcriptomics and proteomics; while the cultivar and tissue-type matching is consistent, batch effects between datasets cannot be excluded. Second, the 8-of-12 compound class reversal did not reach conventional significance (P = 0.194, two-tailed binomial test); larger metabolomic sample sizes or targeted quantification of the affected compound classes would strengthen this result. Third, the translational regulation inference is based on correlative transcript--protein divergence; direct evidence for translational control would require polysome profiling or ribosome footprinting, which were not performed here. Fourth, while our data position ABC1K7 at the apex of the causal chain, functional validation---e.g., via CRISPR-mediated allele replacement in coconut or heterologous expression in a tractable system---has not been completed. Fifth, the three-cultivar domestication gradient, while well-documented by breeding records, represents a single domestication trajectory; replication in independent coconut domestication events (e.g., Southeast Asian dwarf populations) would test the generality of the transcriptional streamlining model. Despite these limitations, the convergence of transcriptomic, proteomic, and metabolomic evidence on a coherent developmental program---anchored to a genetically validated domestication locus---provides a foundation for mechanistic dissection of perennial crop domestication that extends beyond descriptive multi-omics profiling.

\section{Materials and Methods}

\subsection{Plant materials and sampling}

Endosperm tissue was collected from three coconut (\textit{Cocos nucifera} L.) cultivars at the Coconut Research Institute, Chinese Academy of Tropical Agricultural Sciences (Wenchang, Hainan, China): a wild-type tall variety (CK), an intermediate dwarf variety (W6), and a fully domesticated dwarf variety (W5). Samples were collected at three developmental stages: early (3--4 months post-anthesis, liquid endosperm), mid (5--7 months post-anthesis, transitional phase), and late (9--11 months post-anthesis, solid kernel). Three biological replicates were collected per cultivar per time point. Samples were flash-frozen in liquid nitrogen and stored at $-80^\circ$C.

\subsection{RNA-seq and transcriptomic analysis}

Total RNA was extracted using the CTAB method. Library preparation and sequencing were performed on the Illumina NovaSeq 6000 platform (150 bp paired-end reads). Raw reads were preprocessed using fastp \cite{chen2018}. Reads were mapped to the coconut reference genome \cite{xiao2017} using HISAT2 \cite{kim2019}, and gene-level read counts were quantified using featureCounts \cite{liao2014}. For within-cultivar developmental comparisons, differentially expressed genes (DEGs) were identified using edgeR \cite{robinson2010} with the quasi-likelihood F-test, requiring $|\log_2$ fold-change$| > 1$ and Benjamini--Hochberg adjusted \textit{P}-value (FDR) $< 0.05$.

For cross-cultivar comparisons (W5 vs.\ W6), we employed a ratio-based batch correction approach: the W5 vs.\ W6 fold-change was calculated as the ratio of W5 vs.\ CK to W6 vs.\ CK at the same developmental stage, with CK serving as the common internal reference to cancel inter-batch variation. KEGG pathway enrichment was performed using clusterProfiler with Benjamini--Hochberg correction \cite{kanehisa2000}.

\subsection{Proteomic analysis}

Proteins were extracted from W5, W6, and CK endosperm tissue (early and mid developmental stages), digested with trypsin, and labeled with tandem mass tag (TMT) 6-plex reagents \cite{cox2008}. Labeled peptides were fractionated by high-pH reverse-phase chromatography and analyzed on an Orbitrap Exploris 480 mass spectrometer. Proteins were identified and quantified using MaxQuant against the coconut protein database \cite{tyanova2016}. Differential abundance was defined as fold-change $> 1.5$ with Benjamini--Hochberg FDR q $< 0.05$.

\subsection{Metabolomic profiling}

Non-targeted metabolomic data were reanalyzed from a previously published study \cite{hou2025}, which employed ultra-performance liquid chromatography-tandem mass spectrometry (UPLC-MS/MS). Here, we performed de novo principal component analysis and compound class-level comparisons between liquid (coconut water) and solid (kernel) endosperm phases using SIMCA.

\subsection{Statistical analysis}

The 8-of-12 compound class reversal was tested using a two-tailed binomial test against the null hypothesis of equal probability (P = 0.5) and a Bayesian Beta-Binomial model (Beta(1,1) prior) that estimated P($\theta > 0.5 \mid$ 8/12) = 0.91. All other statistical methods are described in the relevant sections above.

\section*{Acknowledgments}

We thank the coconut breeding team at CRI-CATAS for maintaining the germplasm collection and field trial network. We thank Dr.\ Mingzhi Hou for providing access to the metabolomic dataset \cite{hou2025}.

\section*{Author Contributions}

\textbf{Conceptualization:} C.S., H.C., W.L. \textbf{Methodology:} C.S., J.M., W.L. \textbf{Formal Analysis:} N.Y., Y.C., N.Z., J.M. \textbf{Investigation:} N.Y., Y.C., N.Z., W.L. \textbf{Data Curation:} N.Y., Y.C. \textbf{Writing -- Original Draft:} C.S., N.Y., Y.C. \textbf{Writing -- Review \& Editing:} C.S., W.L., J.M., H.C. \textbf{Supervision:} C.S., H.C., W.L. \textbf{Project Administration:} C.S. \textbf{Funding Acquisition:} C.S., H.C., W.L.

\section*{Declarations}

\textbf{Competing interests:} The authors declare that they have no competing interests.

\textbf{Funding:} This work was supported by the Key Research and Development Project of Hainan Provincial Department of Science and Technology (ZDYF2026XDNY143), the Central Finance Forestry Science and Technology Promotion Demonstration Fund Project of Hainan Province (QIONG[2024]TG07), and the International Science and Technology Cooperation Research and Development Project of Hainan Provincial Department of Science and Technology (GHYF2025027).

\section*{Data Availability}

Raw sequencing data have been deposited in the NCBI Sequence Read Archive under BioProject PRJNA1477953. Proteomics data have been deposited in the ProteomeXchange Consortium via the iProX partner repository under accession IPX0017960000. The metabolomic data reanalyzed in this study are from Hou et al.\ (2025) and are available upon request from the original authors. All processed data tables are provided as Supplementary Tables.

\begin{thebibliography}{99}

\bibitem{bai2017} Bai B, Peviani A, van der Horst S, et al. (2017) Extensive translational regulation during seed germination revealed by polysomal profiling. \textit{New Phytologist} \textbf{214}: 233--244. DOI: 10.1111/nph.14355

\bibitem{berger2006} Berger F, Grini PE, Schnittger A (2006) Endosperm: an integrator of seed growth and development. \textit{Current Opinion in Plant Biology} \textbf{9}: 664--670. DOI: 10.1016/j.pbi.2006.09.015

\bibitem{chen2018} Chen S, Zhou Y, Chen Y, Gu J (2018) fastp: an ultra-fast all-in-one FASTQ preprocessor. \textit{Bioinformatics} \textbf{34}: i884--i890. DOI: 10.1093/bioinformatics/bty560

\bibitem{costa2004} Costa LM, Gutierrez-Marcos JF, Dickinson HG (2004) More than a yolk: the short life and complex times of the plant endosperm. \textit{Trends in Plant Science} \textbf{9}: 507--514. DOI: 10.1016/j.tplants.2004.08.007

\bibitem{cox2008} Cox J, Mann M (2008) MaxQuant enables high peptide identification rates, individualized p.p.b.-range mass accuracies and proteome-wide protein quantification. \textit{Nature Biotechnology} \textbf{26}: 1367--1372. DOI: 10.1038/nbt.1511

\bibitem{hou2025} Hou M, Martin JJ, Li W, et al. (2025) Comprehensive metabolite profiling reveals biochemical evolution in coconut water during fruit development. \textit{Scientia Horticulturae} \textbf{350}: 114290. DOI: 10.1016/j.scienta.2025.114290

\bibitem{ingram2010} Ingram GC (2010) Family life at close quarters: communication and constraint in angiosperm seed development. \textit{Protoplasma} \textbf{247}: 195--214. DOI: 10.1007/s00709-010-0184-y

\bibitem{kanehisa2000} Kanehisa M, Goto S (2000) KEGG: Kyoto Encyclopedia of Genes and Genomes. \textit{Nucleic Acids Research} \textbf{28}: 27--30. DOI: 10.1093/nar/28.1.27

\bibitem{kim2019} Kim D, Paggi JM, Park C, et al. (2019) Graph-based genome alignment and genotyping with HISAT2 and HISAT-genotype. \textit{Nature Biotechnology} \textbf{37}: 907--915. DOI: 10.1038/s41587-019-0201-4

\bibitem{li2012} Li J, Berger F (2012) Endosperm: food for humankind and fodder for scientific discoveries. \textit{New Phytologist} \textbf{195}: 290--305. DOI: 10.1111/j.1469-8137.2012.04182.x

\bibitem{liao2014} Liao Y, Smyth GK, Shi W (2014) featureCounts: an efficient general purpose program for assigning sequence reads to genomic features. \textit{Bioinformatics} \textbf{30}: 923--930. DOI: 10.1093/bioinformatics/btt656

\bibitem{meyer2013} Meyer RS, Purugganan MD (2013) Evolution of crop species: genetics of domestication and diversification. \textit{Nature Reviews Genetics} \textbf{14}: 840--852. DOI: 10.1038/nrg3605

\bibitem{olsen2004} Olsen OA (2004) Nuclear endosperm development in cereals and Arabidopsis thaliana. \textit{The Plant Cell} \textbf{16}: S214--S227. DOI: 10.1105/tpc.017111

\bibitem{prades2012} Prades A, Dornier M, Diop N, Pain JP (2012) Coconut water uses, composition and properties: a review. \textit{Fruits} \textbf{67}: 87--107. DOI: 10.1051/fruits/2012002

\bibitem{robinson2010} Robinson MD, McCarthy DJ, Smyth GK (2010) edgeR: a Bioconductor package for differential expression analysis of digital gene expression data. \textit{Bioinformatics} \textbf{26}: 139--140. DOI: 10.1093/bioinformatics/btp616

\bibitem{sabelli2009} Sabelli PA, Larkins BA (2009) The development of endosperm in grasses. \textit{Plant Physiology} \textbf{149}: 14--26. DOI: 10.1104/pp.108.129437

\bibitem{tyanova2016} Tyanova S, Temu T, Cox J (2016) The MaxQuant computational platform for mass spectrometry-based shotgun proteomics. \textit{Nature Protocols} \textbf{11}: 2301--2319. DOI: 10.1038/nprot.2016.136

\bibitem{walley2016} Walley JW, Sartor RC, Shen Z, et al. (2016) Integration of omic networks in a developmental atlas of maize. \textit{Science} \textbf{353}: 814--818. DOI: 10.1126/science.aag1125

\bibitem{xiao2017} Xiao Y, Xu P, Fan H, et al. (2017) The genome draft of coconut (\textit{Cocos nucifera}). \textit{GigaScience} \textbf{6}(11): 1--11. DOI: 10.1093/gigascience/gix095

\bibitem{yong2009} Yong JWH, Ge L, Ng YF, Tan SN (2009) The chemical composition and biological properties of coconut (\textit{Cocos nucifera} L.) water. \textit{Molecules} \textbf{14}: 5144--5164. DOI: 10.3390/molecules14125144

\end{thebibliography}

\end{document}