Natural variation in Arabidopsis uncouples leaf and flower development and reveals massive transcriptomic heterochrony

Abstract

Plant development is a sequence of precisely timed and spatially coordinated events that produce organs such as leaves and flowers. In Arabidopsis thaliana, for example, the development of leaves (called bracts in the inflorescence) halts once the first flower forms. Understanding how this transition is regulated is key for decoding how developmental programmes are coordinated during the floral transition. In this study, we investigated a natural phenotypic variation that uncouples bract repression from flower initiation. We discovered that the continued formation of bracts after the floral transition involves complex genetic interactions across at least four loci. Interestingly, none of these loci included known floral identity genes previously linked to bract repression, pointing to novel regulators in the coordination of bract and flower development. Using time-series transcriptomics and curve registration, we found that differences in gene expression levels when bracts persist are mainly driven by a massive desynchronization of gene dynamics. This affects a wide range of biological processes beyond those associated with leaf identity. These findings align with the ‘inverse hourglass’ model, which proposes that transcriptomic divergence at transitional stages contributes to morphological variation. Our results suggest that this model may also explain trait variability within species, highlighting how transcriptome dynamics shape phenotypic robustness during developmental transitions.

By studying a wild Arabidopsis plant that does not precisely synchronize leaf and flower formation, this research shows that perturbation of developmental timing in nature can arise from several unexpected genes and widespread shifts in gene expression dynamics.

Introduction

The development of living organisms generally proceeds through distinct temporal phases. As illustrated by insects developing from embryo to adult through several larval stages and metamorphosis, development is often described as stepwise ontogenetic sequences (Baressi and Gilbert, 2023). Some transitions are sharp, requiring the precise coordination of the developmental programmes of different organs to generate new structures and ultimately form the organism (Furlow and Neff, 2006; Truman, 2019; Yamaguchi, 2021). At the molecular level, these transitions involve highly dynamic changes in gene expression. Changes in the relative timing of these developmental events, or heterochrony, are recognized as an important driver of morphological evolution (Alberch et al., 1979; Keyte and Smith, 2014). Hence, understanding how developmental timing is controlled from gene expression to organ formation is key to understanding both phenotypic robustness and evolution.

Flowering is a major developmental and morphological transition, transforming plant architecture. In most species, flowers emerge along stems from previously flowerless parts, without any intermediate floral forms. Very often, the start of flower production is also phased with other phenotypic changes. The floral transition thus provides an excellent developmental system to investigate the mechanisms of phenotypic robustness, involving precise spatio-temporal coordination events.

In several species, flower production coincides with the cessation of leaf production, resulting in flower-bearing stems devoid of leaves. In botany, a leaf associated with a single flower is called a bract (Penin, 2008; Prenner et al., 2009). A bractless (i.e. leafless) inflorescence is a typical trait of the Brassicaceae family, which includes the model plant Arabidopsis thaliana (Penin, 2008). The presence of bracts in angiosperm inflorescences varies widely among species, eliciting diverse and still unresolved hypotheses about the selecting factors and adaptive values driving this evolution (Mach, 2010). From a developmental point of view, it shows that the coordination of the leaf programme with reproductive development has been independently and recurrently modified during evolution.

In this study, we use the floral transition in A. thaliana as a model system to study how two developmental programmes, of the leaf and flower, are coordinated in a short time window. For instance, several studies detailed below have explored how leaf development is halted precisely when the first flower is formed in this species.

First, morphological deformations in the inflorescence meristems (Kwiatkowska, 2006) and localized gene repression [e.g. SHOOT MERISTEMLESS (STM)] or activation [e.g. AINTEGUMENTA (ANT) or FILAMENTOUS FLOWER (FIL)] indicate that a bract domain forms before each flower but fails to grow out (Long and Barton, 2000; Dinneny et al., 2004). This domain is later incorporated in floral tissues, remaining cryptic (Heisler et al., 2005; Goldshmidt et al., 2008). Hence, the absence of a bract is due to its early repression when flower development is initiated.

Second, several genes have been implicated in bract repression through identification of mutants that restore bract formation. These genes, mostly transcriptional regulators, are all functionally associated with the floral meristem identity (FMI) pathway (Siriwardana and Lamb, 2012), including the central flower regulator LEAFY (LFY) (Schultz and Haughn, 1991; Weigel et al., 1992). The FMI pathway specifies floral rather than shoot identity (Weigel, 1995) and its impairment leads to flower-to-shoot conversions by restoring vegetative characteristics to flowers (altered floral organ identity and phyllotaxis, internode elongation, and indeterminacy), including bract outgrowth (see Supplementary Table S1). Furthermore, the genetic ablation of flowers induces bract outgrowth (Nilsson et al., 1998). This suggests that many genes within the FMI regulatory network redundantly act as bract repressors. In all these studies, bract de-repression is consistently accompanied by pleiotropic phenotypes affecting flowers, branching, or flowering time.

Conversely, only a few genes have been identified as promoting bract outgrowth, such as JAGGED (Dinneny et al., 2004; Ohno et al., 2004), ANT, and AINTEGUMENTA-LIKE6 (AIL6) (Manuela and Xu, 2024), and their misexpression also produces pleiotropic phenotypes in addition to bract development. Bracts can also be induced in wild-type A. thaliana under light treatments enriched in far red light, together with flower–shoot chimaeras (Hempel and Feldman, 1995; Hempel et al., 1998). The authors proposed a ‘conversion’ model, where a young branch meristem (already subtended by a leaf) can be converted (here photo-induction) into a flower that inherits this pre-formed leaf, interpreted then as a bract. However, the underlying genetic mechanisms of this process also remain unclear.

In summary, the current literature suggests a straightforward explanation for the precise coupling of bract repression with formation of the first flower in A. thaliana: this arises from the bract-repressing activity of the FMI pathway acting in the flower. The molecular mechanisms mediating this repressive floral effect on the bract remain unclear but they are likely to be indirect and/or non-cell autonomous, as FMI genes are mostly expressed in the developing flower and not in the bract domain. Since bracts are almost identical to leaves produced during the preceding vegetative phase (Dinneny et al., 2004), de-repressed bracts have sometimes been interpreted as the persistence of a vegetative programme (Hempel et al., 1998). Bract de-repression and improper establishment of flower identity are therefore considered as two sides of the same coin.

In this study, we provide new data showing that, in natural populations, bract repression and flower development can be uncoupled. We report the transient presence of bracts with wild-type flowers precisely at the floral transition in wild A. thaliana accessions. By combining precise phenotype characterization, quantitative genetics, genomics, and time-series transcriptomics specifically in meristem tissues, we investigated this change in the timing of flower and leaf formation at developmental, genetic, and transcriptomic levels. Altogether, our results provide two new main contributions to our understanding of the mechanisms coordinating leaf and flower development. First, bract development can be prolonged without affecting the main FMI pathway genes. Second, this prolonged bract formation at the floral transition does not correspond to a global ‘juvenile’ shift in the transcriptome and correlates with more complex transcriptomic heterochronies. In line with the ‘inverse hourglass model’, we discuss how transcriptional divergence, resulting from the ongoing adaptation of flowering timing, may lead to phenotypic variation at different evolutionary scales.

Materials and methods

Plant growth conditions

Seeds were sown on peaty-clay soil, and germination was synchronized by stratification at 4 °C for 2–3 d in the dark. No vernalization was required for the plants used in this study. Plants were watered with fertilizer (18-10-18 N-P-K) under LED lighting (sunlight spectrum NS12, 150 µmol m^–2 s^–1). Three different day/night regimes were used in the experiments: short days (SDs) with 8 h light and 16 h dark (non-inductive photo-conditions for flowering); long days (LDs) with 16 h light and 8 h dark; and continuous light (CL) with 24 h light (both photo-inductive flowering conditions). Temperature and humidity are controlled as follows: 22 °C and 60% humidity during light for LD and CL conditions, and 18 °C and 70% humidity constantly in SDs and during the night-time in LDs. For the bulk segregant analysis (BSA) and the RNA-seq time course, plants were grown for 20 d in SDs before switching to LDs. For the recombinant inbred line (RIL) phenotyping assays, plants have been directly cultured in LDs.

Plant materials

Arabidopsis thaliana natural accessions and associated RILs were obtained from the Versailles Arabidopsis Stock Centre (VASC), Col-0, 186AV; Tsu-0, 91AV; RIL set name, 3RV. F₁ and F₂ plants used for the BSA were generated by crossing the parents Tsu-0 and Col-0 in both directions. The following strains were obtained from the Nottingham Arabidopsis Stock Centre (NASC) and are in the Col-0 background if not otherwise mentioned: tfl1-13, N6237; tfl1-14, N6238. puchi-1, bop1-4×bop2-11, and puchi-1×bop1-4×bop2-11 (Karim et al., 2009), lfy-12 (Maizel and Weigel, 2004), and soc1-2 (Michaels et al., 2003) were described previously. Plants expressing pLFY::2×mCherry-N7; 35S::Lti6b–yellow fluorescent protein (YFP) (in Col-0 and Tsu-0) were generated for this study. Lobularia maritima and Allaria petiolata plants were natural specimens collected outside the laboratory, in France. The phylogeny of Brassicaceae tribes used in Supplementary Fig. S1 is extracted from the Brassibase website https://brassibase.cos.uni-heidelberg.de/ (Kiefer et al., 2014).

Plasmid constructions and plant transformation

For pLFY::2×mCherry-N7, since no polymorphism was sequenced in the LFY promoter between Col-0 and Tsu-0 accessions, a sequence starting at −2277 bp upstream of the ATG was amplified by PCR from Col-0 genomic DNA using 5'GGGGACAACTTTGTATAGAAAAGTTGATCCATTTTTCGCAAAGG and 5'GGGGACTGCTTTTTTGTACAAACTTGAATCTATTTTTCTCTCTCTCTCTATC primers. PCR fragments were purified and inserted into pDONR P4-P1R with a gateway BP reaction. This plasmid was then recombined in a three-fragment gateway reaction with 2×mCherry pDONR211 and N7-tag pDONR P2R-P3 (containing the nuclear tag N7 and a stop codon) and the destination vector pK7m34GW.

For 35S::Lti6b–YFP, a pDONR P2R-P3 plasmid containing the sequence of the plasma membrane protein Lti6b (Cutler et al., 2000) was recombined in a three-fragment gateway reaction with 35S pDONR P4-P1R and YFP pDONR P2R-P3 into the destination vector pB7m34GW. Both constructs were then transformed into Agrobacterium tumefaciens C58pMP90 strain by electroporation and transformed into both Col-0 and Tsu-0 plants by floral dipping (Clough and Bent, 1998). For each construct×genotype combination, several independent transgenic lines were selected in T₂ for a single insertion event based on 3:1 resistant:sensitive Mendelian segregation of the resistance provided by the transgene. The expression patterns of pLFY were reproducible between selected lines, and matched published data for Col-0. We were unable to obtain a 35S::Lti6b–YFP line with a membrane signal as strong in Tsu-0 as in Col-0. However, despite the weak YFP signal, the morphology of the tissue could still be captured.

Microscopic meristem imaging and image analysis

Meristems were imaged using a scanning electron microscope (Hirox 3000 SEM), or with a confocal microscope Zeiss 700 LSM, according to the manufacturer’s instruction and without prior fixation. Multitrack sequential acquisitions were performed as follows: YFP, excitation wavelength (ex): 488 nm, emission wavelength (em), 300–590 nm; mCherry, ex, 555 nm, em, 300–620 nm. Detection wavelengths and laser power were identical for Col-0 and Tsu-0; the photomultiplier tube (PMT) voltage was identical for mCherry to allow comparisons of pLFY signal intensity in the two genotypes. The YFP PMT voltage was optimized on each plant. Confocal images were processed using ImageJ (Schneider et al., 2012): maximum projections of the mCherry channel and 16-bit-transformed standard deviation projections of the YFP channel were merged in a composite image. mCherry intensities were unchanged, while brightness and contrast of the YFP channel were optimized on each plant to provide the best morphological outlines of the tissues.

Macroscopic plant phenotyping

Macroscopic pictures were taken using different devices, according to the manufacturer’s instructions: Keyence VHX-900F, Canon EOS 450D camera, camera device of a Samsung Galaxy A10, and a Ulefone Armor X5 pro.

Phenotypic measurements

The basal bract score was determined by counting all bracts in the main stem and cauline branches, normalizing by the total number of branches (excluding rosette branches, Fig. 1E); the inspection was limited to the first five flowers, especially for mutants. The number of bracts was counted after internode elongation on the last upper cauline branch to ensure bracts were visible to the naked eye. Flowering time was measured with different methods (mentioned when needed in the Results): the number of days from the start of transfer to growth chambers to the day of bolting, the day when the first flower blooms (open petals), or as the cumulated number of leaves (rosette only or rosette and cauline). For plastochron measurements, several plants of both genotypes were synchronized and grown in the same condition. On each sampling day after transfer to an LD, 5–10 plants were randomly dissected under a binocular dissecting scope, removing and counting all organs (leaves or flowers) from the first leaf to the youngest organ visible on the main meristem. The youngest organs were counted as soon as they were separated by a boundary (corresponding to stage-2 flowers as defined by Smyth et al., 1990).

Fig. 1.

Desynchronization of leaf repression from flower formation in the natural Tsu-0 accession of Arabidopsis thaliana. (A) Lateral views of the main stem of Col-0 (left) and Tsu-0 (right) accessions showing the transition from branches (at the bottom) to flowers (to the top). The red arrow points to a bract (a leaf) subtending the first flower in Tsu-0. Cartoons on the side summarize the structures produced at each node. (B) Distribution (as a percentage ) of the position of bracts in Tsu-0 (plants n=50, bracts n=337) in the main stem and cauline branches, where 1 is the first flower after floral transition. (C) Number of bracts on the different branches of a Tsu-0 inflorescence (a dot is the mean, vertical bars are the standard deviation). Branch positions are numbered with increasing numbers from the base to the top of the main stem, which is noted as 0 by convention (plants n=90). A linear regression (horizontal red dotted line with the SE shaded in red) indicates a stable average value of 1 bract on each branch. (D) Col-0 floral pedicels do not show any sign of bract development (black arrowhead). (E) Variable bract development (red arrows) observable in Tsu-0, from true leaves (left) to filamentous or small rudimentary structures at the base of the floral pedicel. (F) SEM of floral pedicels at the floral transition in Col-0 (left) and Tsu-0 (right). In Col-0, the pedicel is a regular, naked cylinder at the base (black arrow) while in Tsu-0 there is a bract on the first flower (red arrow) and swollen base of pedicels on the two following flowers (red arrowhead). (G) Number of floral organs in flowers with (dark red) or without (light red) a bract in Tsu-0 plants. (H and I) SEM of the main meristem of plants captured right at the transition to flowering in Col-0 (H) and Tsu-0 (I). The first flowers after the transition are indicated by white arrows and ordered with increasing numbers in their order of production inferred from size and phyllotactic positions. Green arrowheads: cauline branch (axillary meristem and a cauline leaf), –1 indicates the last branch before floral transition. Red arrows: bracts. In Col-0, no bract is visible on the adaxial side of the developing flowers.

DNA extraction and sequencing for bulk segregant analysis

For the BSA, an F₂ mapping population of 684 plants was generated by crossing (Col-0×Tsu-0) in both directions. The BSA was split into four replicates. A 1 cm² leaf sample was sampled for each F₂ plant, kept at −20 °C, and genomic DNA was extracted individually if the plant was selected in one of the bulks. Genomic DNAs of 56 and 17 plants were selected and pooled in the bulks of low and high bract score plants, respectively. The genome of the parental lines was re-sequenced using genomic DNA from bulk Tsu-0 and Col-0 seedlings, respectively. All genomic DNAs were extracted and purified using a cetyl trimethylammonium bromide (CTAB)-based protocol, following instructions as given in Healey et al. (2014).

Each DNA bulk was prepared by pooling the genomic DNA of selected plants in equal quantities, to reach a final concentration between 13 ng µl^–1 and 25 ng µl^–1. Pooled DNA bulks and parental DNA were used to prepare libraries and sequenced on BGISEQ-500WGS according to the manufacturer’s instructions, yielding 5 Gb of data of 100 bp paired-end reads per library (target coverage of 40×).

DNA sequencing analysis and genomic variant analysis

Sequencing data consisted of two parental plus two bulks of sequencing data. A genomic variant analysis was performed on each dataset following the workflow of short variant discovery previously described in Besnard et al. (2020). This resulted in four gVCF files (one per sample) generated by the HaplotypeCaller tool of GATK (v3.8, McKenna et al., 2010). TAIR10 was used as the A. thaliana reference genome. Then, the two parental gVCFs were first joint-genotyped using the GATK tool GenotypeGVCFs to produce a common vcf file for the two samples. This file was used to select a list of specific single nucleotidee polymorphisms (SNPs) and small insertions/deletions (indels) of Tsu-0 (91AV stock) versus Col-0 (186AV stock), filtering for positions with the depth of coverage (metric DP) >10. This reference list of Tsu-0 polymorphisms was then used as the -dbsnp option in a second pass of joint genotyping using all four gVCFs as input (two parental samples plus the two bulks) to produce a common vcf file. Finally, relevant polymorphic positions from the reference list in the two bulks were selected after filtering for a depth ≥3.

Quantitative trait locus mapping from bulk segregant analysis

Quantitative trait locus (QTL) identification was carried out using the QTLseqr package (Mansfeld and Grumet, 2018) according to user instructions. After filtering data with the following parameters (refAlleleFreq=0, minTotalDepth=10, maxTotalDepth=90, minSampleDepth=10, minGQ=99, depthDifference=15), the ΔSNP index (Takagi et al., 2013) was generated and loci that reached the ∼95% confidence interval were retained.

Quantitative trait locus mapping using recombinant inbred lines

Genotyped RILs from Col-0×Tsu-0 (3RV) are publicly available in the VASC. Based on BSA results, a panel of 55 RILs was selected from their known genotypes on chromosomes 1 and 5 using GGT 2.0 software (van Berloo, 2008). The detailed genotypes of each line used in this study are available in Supplementary Table S2. The presence of basal bracts was quantified in each line using the bract score. QTL mapping was performed with R/qtl software according to user instructions (Broman et al., 2019): we used the scanone function with the mean value of the bract score for the 55 tested RILs as a trait, and significant thresholds were computed by setting a permutation number of 2000. For the bract score, the ‘non-parametric’ (np) statistical model was used, while for paraclade number, we used a normal model implemented with the hk method. To look for QTL interactions, the scantwo function was used with the normal model and the hk method since the np model is not implemented for this function. Multiple-QTL mapping (MQM) was performed with default parameters. The (broad-sense) heritability for the bract score was defined as H²=(var_T−var_E)/var_T, with var_T and var_E being the total and environmental variance, respectively. var_T was computed as the total variance of the bract score of each plant over the entire QTL mapping dataset (1830 RIL plants and 780 control parents split over 12 experiments), while var_E was computed as the pooled variance of the line variances (computed over all plants of a line across all experiments), which averages the different line variances computed as: $\left({\sum }_{i=1,N}((n_i-1)\times s_i^2)\right)/\left({\sum }_{i=1,N}(n_i-N)\right)$ where N is the number of lines; and n_i and s_i the number of plants and the variance, respectively, in line i. Confidence intervals of H² were provided by bootstrapping 70% of the strain 1000 times as in Chuffart et al. (2016).

Kompetitive allele-specific PCR genotyping

We selected 19 new genotyping markers from the 140SNPvCol marker set (Lutz and Schwechheimer, 2017, Preprint) to cover the two large QTLs mapped on chromosome 1, namely 1a and 1b (Supplementary Fig. S2B), and corresponding kompetitive allele-specific PCR (KASP) oligos were ordered from Biosearch Technologies™ LGC Ltd. To further reduce the intervals, we designed two new SNP markers targeting Tsu-0 polymorphisms identified in our whole-genome re-sequencing data. Specific KASP oligos (three per SNP) were designed by LGC from the 100 bp sequence surrounding the SNP. All 21 new markers (Supplementary Table S3) were validated on parental genomic DNA (Col-0 and Tsu-0) in a KASP assay before their use with RILs. Clean genomic DNA from CTAB-based extraction was used for all samples. For each marker, 2.5 µl of DNA (diluted to a final concentration of 5 ng µl^–1), 2.5 µl of KASP-TF V4.0 2× Master Mix low ROX, and 0.07 µl of KASP assay mix were mixed in 384-well plates, and the genotyping assays were run in a QuantStudio 6 Flex (Applied Biosystems), using standard user guidelines for thermal cycling and final fluorescence analysis.

Fine mapping using heterologous inbred families

Using F₆ genotyping data for the 3RV RIL set from the VASC, we selected three RILs heterozygous in a region overlapping QTL1a, QTL1b, and the interval in between: RILs 334, 488, and 478, respectively (Supplementary Fig. S2). For RILs 334 and 488, we genotyped 160 segregating F₈ progenies by KASP and selected five and eight plants, respectively, with homozygous recombination inside the QTL region under study. For RIL 478, we genotyped 20 plants and selected only one; genotyping reactions were performed using PCR primers designed to amplify the published marker of the VASC (Simon et al., 2008) (Supplementary Table S4) and the results were read by Sanger sequencing. All selected F₈ plants were selfed to generate an inbred line of a fixed genotyped (F₉), the collection of F₉ inbred lines forming a heterologous inbred family (HIF). HIF lines (with at least 20 plants per line) were phenotyped for the bract score with parental control in the same experiment.

Biological sample preparation for the transcriptome time-series profiling

After 20 d in SDs, Tsu-0 and Col-0 meristems were dissected every day in LD conditions, to capture the precise developmental stages (especially stages T and F; see below). Three independent biological experiments were performed with 5–11 meristems per replicate. Stages are defined as follows: V, the day when plants were transferred to LDs from SDs; L, after 4–5 d of LDs (identical for both Col-0 and Tsu-0, which have the same meristem shape at this time); T, the main meristem enlarges and domes (Kwiatkowska, 2008; Kinoshita et al., 2020), and lateral meristems start to become visible at the axils of young leaf primordia; F, all young organs in the meristems were identified as flowers (note that this stage occurred several days before bolting). Stage F in mutants was defined when several rounded primordia become visible at the shoot apical meristem (SAM; which will become branch-like flowers or only branches). Dissections were performed from 09.00 h to 12.00 h, by alternating between Tsu-0 and Col-0. Micro-dissected meristems were snap-frozen in liquid nitrogen and stored at −80 °C until RNA extraction. To avoid the induction of stress-related gene expression, meristem dissection did not exceed 3 min between removal of the first organ and freezing. For each replicate, pooled meristems were ground with a RockyII tissue lyser according to the manufacturer’s instructions. RNA was extracted using the PicoPure RNA Isolation Kit (Arcturus, Catalog KIT0202) according to the standard protocol. RNA concentration ranged from 3 ng µl^–1 to 64 ng µl^–1 (average 23 ng µl^–1), with an RNA integrity number (RIN) between 5.6 and 7.6 (average 6.8).

RNA-sequencing

Library preparation was carried out using the NEBNext® Ultra™ II Directional RNA Library Prep Kit for Illumina (New England Biolabs); the NEBNext® Poly(A) mRNA Magnetic Isolation Module (New England Biolabs); and NEBNext® Multiplex Oligos for Illumina (Index Primer Sets 1, 2, and 3) from 40 ng of RNA, and sequenced using the NextSeq500 (Illumina). Sequence quality was controlled using the Sequencing Analysis Viewer; sequences that did not pass the quality filter were removed. Following quality control an average of 43 million sequences per sample was achieved.

RNA-sequencing analysis (pre-processing procedures)

Quality filtering using Trimmomatic (Bolger et al., 2014) retained 96% of the reads for all samples, which were aligned to the TAIR10 genome using STAR (Dobin et al., 2013) with the following parameters: non-canonical splice junctions were removed; multi-mapping reads were limited to 10; and only reads mapped once were considered to determine splicing junctions. Between 88% and 97% of the reads were mapped to a simple locus. Normalization of read counts was performed using the R Bioconductor package DESeq (Love et al., 2014), with the following parameters: genes for which the total number of reads was <10 were discarded, and data were transformed with the variance-stabilizing transformation (VST) function (Anders and Huber, 2010), to harmonize the variance. Consistency between biological replicates was verified using a principal component analysis (PCA) on all samples with VST-transformed data.

RNA-sequencing analysis: differential gene expression analysis

Differential analysis was performed using the R Bioconductor package edgeR (McCarthy et al., 2012). Reads were first normalized using TMM (trimmed mean of M-values) to reduce library-specific biases. Normalization factors were between 0.94 and 1.049. A generalized linear model was applied for the analysis of differentially expressed genes (DEGs). Three types of DEG analysis were considered: DEGs at each stage between the different genotypes (Supplementary Table S5), DEGs between the stage within the same genotype, and DEGs across all conditions (stages and genotypes). Multiple DEG analyses were corrected using Benjamini–Hochberg (BH) correction, and genes with a P-value <0.05 were retained.

RNA-sequencing analysis: clustering gene expression correlating with bract presence

To identify bract signature genes, we first calculated the minimum and maximum expression levels among stages with and without bracts. If the minimum expression level in the stages with bracts was greater than or equal to the maximum expression level in the stages without bracts, the genes were classified as bract signature genes whose expression levels positively correlated with the presence of bracts. If the maximum expression level in the stages with bracts was lower than or equal to the minimum expression level in the stages without bracts, these genes were classified as bract signature genes whose expression levels anti-correlated with the presence of bracts. If neither of these conditions was satisfied, genes were discarded.

Temporal registration of gene expression dynamics

To align gene expression profiles of A. thaliana Tsu-0 (query data) with Col-0 (reference data), we utilized the curve registration method in the R package greatR (Kristianingsih, 2024). This approach involved shifting the gene expression profiles of Tsu-0 across developmental stages (V, L, T, and F) using shift factors ranging from −1 to 1; a stretch factor was not applied due to the two datasets being over similar times. The optimal shift parameter for each gene pair was identified by maximizing the log-likelihood. The fit with the best registration factors for each gene was then compared with the fit with a non-registered model (without transformation) using the Bayesian information criterion (BIC) statistic. A lower BIC score for the registered model (with transformation) indicated that the gene expression dynamics of Tsu-0 and Col-0 were similar, and the expression profile differences could be resolved through registration. Conversely, a higher BIC score suggested that the profiles were best described by two individual curves (i.e. not aligned). Prior to transformation, gene expression levels in both datasets were centred and scaled using the z-score scaling method. The standard deviation for the replicates at each time point was set to 0.01. Complete results of the registration are provided as Supplementary Table S6.

Gene Ontology term analysis

Gene Ontology (GO) term enrichment analyses were performed using either the ‘Single Enrichment Analysis’ tool from AgriGOv2 (Tian et al., 2017) with default parameters (Fisher’s test with Yekutieli adjustment method) or clusterProfiler 4.0 (Wu et al., 2021) with default parameters (enrichGO and simplify functions with the BH adjustment method). The rich factor was computed by dividing the number of genes associated with a GOterm (‘Count’) by the numerator of the ‘BgRatio’ computed by the enrichoGO function.

Gene selection inside genetic mapping intervals

The R library ‘GenomicFeatures’ (Lawrence et al., 2013) was used to import gene information from the most recent annotation file of A. thaliana in the gff format (Cheng et al., 2017). Custom R scripts were used to intersect all gene loci falling within genetic mapping intervals and provide information for each gene from RNA-seq analysis and genomic variant analysis (see Supplementary Tables S2 and S4). We used snpEff v5.0d (Cingolani et al., 2012) and its putative functional categories (HIGH/MODERATE/LOW/MODIFIER) to predict the functional impacts of the genomic variants over regions covering each annotated gene of the interval, including the promoter (2 kbp upstream of the transcriptional start site) as well as the 500 bp downstream region in the case of miRNA. In Supplementary Tables S2 and S4, additional gene information was retrieved from ThaleMine (Pasha et al., 2020) and GO terms from the TAIR bulk data retrieval tool.

Data handling, data visualization, and descriptive statistics

The R software was used (R Core Team, 2018), in particular the following packages: Bioconductor packages (Huber et al., 2015) for the analysis of omics data, especially GenomicFeatures, IRange (Lawrence et al., 2013), rtracklayer (Lawrence et al., 2009), clusterProfiler (Wu et al., 2021), limma (Ritchie et al., 2015), DESseq2 (Love et al., 2014), edgeR (Chen et al., 2016, Preprint), and org.At.tair.db (Carlson, 2024); dplyr (Wickham et al., 2023), tidyr (Wickham et al., 2024), and reshape2 (Wickham, 2007) for data manipulation; and ggplot2 for almost all plots (Wickham, 2016), with viridis and plasma (Garnier et al., 2024) for some colour optimization.

Results

Bract repression is not strictly synchronized with the floral transition in the natural accession Tsu-0

Like in A. thaliana and its reference accession (Col-0) (Fig. 1A), the typical ebracteate flower (i.e. without bracts) of Brassicaceae generally appears as soon as the first flower is produced, indicating synchronization of bract repression with the onset of the flower programme. However, in a number of Brassicaceae species, bract repression is delayed for several rounds of flower formation, generating bracteate flowers (i.e. with bracts) at the base of their flowering branch (a raceme) (Endress, 2006; German et al., 2023): some species display bracts up to the first half of the raceme (e.g. Supplementary Fig. S1A, C), while others only have one or two bracts at the base (Supplementary Fig. S1B, D; or, for example, in Al-Shehbaz, 2015).

Likewise, we discovered that the natural Tsu-0 accession in A. thaliana displays bracts on the first 1–5 flowers of the raceme (Fig. 1A, B), in the main stem and cauline branches at the same frequency (Fig. 1C). A gradient of bract outgrowth can be observed in Tsu-0, from fully developed bracts resembling the cauline leaves associated with branches, up to filamentous, short rudimentary structures (Fig. 1D, E). SEM showed that a mild swelling of the peduncle bases could sometimes occur in Tsu-0, unlike in Col-0 (Fig. 1F). Both bracteate and ebracteate Tsu-0 flowers were normal, containing the invariant organ number typical of each whorl. Thus, developmental variation is restricted to the bract (Fig. 1G). To exclude that these leaves could be secondary outgrowths from mature floral peduncles, as is the case for floral prophylls, we imaged the emergence of the first flower by SEM. In Col-0, the last lateral meristem subtended by a cauline leaf was immediately followed by ebracteate flowers (Fig. 1H). In Tsu-0, the first flowers produced after the last branch displayed bracts or rudimentary bracts (Fig. 1I), recapitulating the gradient of bract forms observed in mature stems. Their abaxial position and their precocious development also suggest that these structures developed from the cryptic bract, that developed despite the onset of flower formation.

Since Tsu-0 is a wild accession, we wondered whether a ‘conversion’ model as proposed by Hempel and Feldman (Hempel and Feldman, 1995; Hempel et al., 1998) for other wild accessions could explain this phenotype. This model proposes that floral bracts arise from a lateral meristem that is first initiated with the identity of an axillary meristem, allowing a cauline leaf to develop, but is then converted to a floral meristem. The authors notably observe shoot–flower chimaeras supporting this identity conversion. However, phenotyping hundreds of plants did not identify any shoot–flower chimaera (Fig. 1D, F, H). Also, Tsu-0 plants did not exhibit the negative correlation between the numbers of bracts and cauline branches (Fig. 2A) that would be expected under the ‘conversion’ model. This suggests that Tsu-0 bracts are not produced by shoot-to-flower conversions.

Fig. 2.

Desynchronization of leaf repression from flower formation in Tsu-0 and other natural accessions exhibits complex relationships with flowering time and photo-induction conditions. (A) Correlation study between the number of lateral (cauline) branches and the number of bracts on the main stem in Tsu-0 (Pearson test, P-value>0.1), which do not show a negative correlation expected by the ‘conversion’ model (see text). (B) Definition of a quantitative bract score for a single plant. (C) Quantification of basal bracts in different accessions using the bract score. (D) Correlation study between the bract score and the flowering time (assayed when the first flower opens) in different accessions, labelled with different colours. Each dot is a plant, and a linear regression with standard deviation (colour shade) is computed for each accession. (E) Effect of different light regimes on bract scores in Col-0 and Tsu-0 plants (green and red, respectively). LD and SD stand for long- and short-day conditions, respectively, and CL stands for continuous light (see the Materials and methods). Numbers (e.g. 20SD>LD) indicate the number of days in the first condition before transfer to the second. The number of plants scored is indicated below each bar.

To summarize, in the natural accession Tsu-0, bract repression is no longer synchronized with the onset of flower development. Instead, it is transiently altered to varying and generally decreasing degrees in the first flowers of a branch, resulting in what we have termed ‘basal bracts’.

Desynchronization of bract repression with the floral transition is common among Arabidopsis natural accessions

To measure the ability of different genotypes to generate basal bracts, we defined a bract score (Fig. 2B) that takes into account several branches (hence several floral transitions) per plant, normalizing by the number of scored branches. Similar to Tsu-0, other genetically diverse natural accessions produce basal bracts, albeit at highly variable rates (Fig. 2C). We could not identify any correlation between basal bracts and geographic or genetic origins (Supplementary Fig. S3). The bract score shows a high level of intra-genotype variability. Despite this variability, there are clear differences between the basal bracts of different genotypes (Fig. 2C). The production of basal bracts is thus a common phenotypic trait among A. thaliana natural accessions from different geographical origins and varying genetic backgrounds.

Basal bract frequency shows complex correlations with flowering time but is not affected by variation in photo-induction

We then explored whether variation in the frequency of basal bracts in natural accessions could be linked to variation in flowering time. Indeed, flowering time is known to be highly variable among natural accessions (Salomé et al., 2011) and depends on both genetic and environmental factors (Maple et al., 2024). We measured both bract score and flowering time in a panel of five accessions and observed a global positive correlation across genotypes (Fig. 2D). For instance, Tsu-0 flowers later than Col-0 in either absolute time (i.e. the number of days from germination) or developmental time (measured by the number of leaves produced), consistently across culture conditions (Supplementary Fig. S4). This suggests that a longer flowering time could promote the alteration of synchrony between bract repression and the onset of flower development. However, within an accession, this correlation can be negative (e.g. in Ler and Tsu-0, Fig. 2D), suggesting that the relationships between basal bracts and flowering time could be more complex depending on the genetic backgrounds.

To explore further the effects of varying flowering time in a genetic background, Col-0 and Tsu-0 plants were cultured in different photo-induction regimes that vary the flowering time (see the Materials and methods; see also Supplementary Fig. S4). The bract score remains remarkably constant across all tested conditions for both accessions (Fig. 2E). This indicates that basal bract formation does not depend either on photoperiod or on the absolute flowering time, and rather suggests the importance of intrinsic genetic determinism specific to each accession.

Delayed bract repression is controlled by multiple quantitative trait loci that do not contain genes previously associated with bract development

To analyse the genetics of basal bract formation, we conducted quantitative genetics using crosses between Tsu-0 and Col-0 since these two accessions ranked as high and low bract producers, respectively (Fig. 2C). F₁ progeny showed intermediate bract scores, while F₂ bract scores were spread within the parental range, with a distribution that does not indicate a simple monogenic trait (Fig. 3A; Supplementary Fig. S5A). BSA with mapping-by-sequencing identified four QTLs: two on chromosome 1 and two on chromosome 5 (Supplementary Fig. S5B, C). The large spread of F₁ phenotypic values constrained F₂ plant selection, with few plants in the high-score bulk and possibly spurious homozygous in the low-score bulk. These two limits may explain the large intervals obtained.

Fig. 3.

Basal bract formation in Tsu-0 is controlled by four major QTLs that do not contain the FMI genes associated with bract repression. (A) Genetic transmission of the bract score in F₁ and F₂ hybrids from a cross between Col-0 and Tsu-0 parent plants. (B) QTL mapping for the bract score in a set of RILs identifies four putative QTLs, two in chromosome 1 (1a and 1b) and two in chromosome 5 (5a and 5b). Dotted horizontal red lines indicate three different significance thresholds computed from 2000 permutations. At the top of the graph, H² indicates the (broad-sense) heritability of the bract score computed among the RIL set. (C) Genotype of the F₆ generation of the line RIL334 in chromosome 1, showing a heterozygotic region overlapping with QTL1a. (D) Chromosome 1 genotypes of selected lines forming a heterologous inbred family (HIF) obtained from RIL334 after two more generations of selfing (F₈). (E) Box plots of bract scores of the different lines from the ‘334 family’ (blue boxes) with parental controls (Col-0, green; Tsu-0, red; n>20 plants per line). Segregating bract scores within the family map a narrower region, QTL1a*, spanning 541 annotated genes. (F) Genotype of the F₆ generation of the line RIL488 in chromosome 1, showing a heterozygotic region overlapping with QTL1b. (G) Chromosome 1 genotypes of selected HIF lines obtained from RIL488 at F₈ and (H) a box plot of their bract scores (blue boxes), with parental controls (Col-0, green; Tsu-0, red; n>21 plants per line). This maps a narrower region, QTL1b*, spanning 332 annotated genes. In (E) and (H), lines not sharing the same letter(s) are statistically different (post-hoc Tukey analysis with 0.05 significance level from a glm of bract scores fitted with a quasi-poisson distribution). (I) Final genetic mapping of the ‘basal bract’ trait in Tsu-0 and the location of genes reported to impact bract development in previous mutant studies.

To better estimate the QTLs, we also used RILs, which allowed averaging of bract scores across isogenic plants. Using an existing RIL set (Simon et al., 2008), we measured the average bract scores of several lines (Supplementary Fig. S6A) for which we checked the absence of recombination bias (Supplementary Fig. S6B). The heritability of the bract score was high among RILs (65.45±1.7%) and a single QTL scanning mapped four peaks: two most significant on chromosome 1 [logarithm of odds (LOD) score >4 with a P-value <0.1%, named 1a and 1b] and two less significant on chromosome 5 (LOD score >3, named 5a and 5b; Fig. 3B). Consistent with the BSA result, this analysis provided slightly different QTL positions and shorter intervals. A two-QTL scan suggested additive effects between QTLs 1a and 1b (Supplementary Fig. S6C). To confirm and reduce the intervals of these two major QTLs, we used two RILs identified as heterozygous in a region overlapping with these QTLs in the F₆ generation (Fig. 3C–H). F₈ plants from these RILs were re-genotyped at a higher density with new genetic markers (Supplementary Fig. S2A, B) to select recombining intervals homozygous for either allele of the two parents (Fig. 3D, G; Supplementary Fig. S2F). Selected plants were used to generate new lines forming a HIF with identical genotypes except in a small interval, which allows testing of the effect of alleles in that region only (see Fig. 3D). Scoring bracts among HIFs confirmed both QTLs and mapped them to smaller intervals named 1a* and 1b*. Additive effects of QTLs 1a and 1b appeared when the Tsu-0 allele in 1b* was combined with the Col-0 allele in 1a*, the bract score reaching about half of the Tsu-0 parent bract score (see in 488 families, Fig. 3G, H). However, the Tsu-0 parent score is equalle only when both QTLs bear Tsu-0 alleles (see in 334 families, Fig. 3D, E). As a control, we generated a HIF to test the region between 1a and 1b and demonstrated that it did not influence the bract score (Supplementary Fig. S2C–E).

The final mapped intervals remained large (4.1–11.8 cM), containing many genes (332–931, with 2324 annotated genes for the four QTLs; Fig. 3I;  Supplementary Table S7). The parental accessions that we re-sequenced differ by >770 000 small genomic variations (SNPs and indels), with more than one variation every 175 bp on average, leaving few invariant genes to exclude (Supplementary Table S7). However, most known bract-related genes were absent from the QTLs (Fig. 3I), whether those genes were reported as promoting bract development (e.g. JAGGED) or as being involved in the FMI pathway and acting as bract repressor genes (Supplementary Table S1). LIGHT-DEPENDENT SHORT HYPOCOTYLS (LSH1) was found to lie at the border outside QTL5a. We noted that SQUAMOSA PROMOTER BINDING PROTEIN-LIKE 2 (SPL2) falls within QTL5b, although this specific SPL gene has not been implicated in bract development (Shikata et al., 2009; Xu et al., 2016). Similarly, MIR156F, one of the 12 miRNAs targeting SPL genes, is located in QTL5a, although there is no evidence implicating it in bract development (He et al., 2018). This absence of major FMI genes in our QTLs is consistent with the wild-type phenotype of flowers associated with bracts in Tsu-0 (Fig. 1G).

To decipher the causal variation responsible for basal bracts, we took advantage of pleiotropic phenotypes co-segregating with basal bracts in the RILs. The number of cauline branches positively correlated with the bract score, and QTL1b overlaps with a QTL controlling cauline branch number (Supplementary Fig. S7). This suggests that the gene controlling basal bracts in QTL1b* might also promote more cauline branches. Second, ‘branched siliques’ could sometimes be observed among RILs and HIFs at low penetrance (Supplementary Fig. S8A, B). We interpreted this transgressive phenotype as resulting from an incomplete floral determinism with reiteration of floral bud formation in whorls, as reported for mutants such as ap1 (Irish and Sussex, 1990). In HIF 488, this phenotype occurred only when Tsu-0 QTL1b* was associated with Col-0 haplotypes in other mapped QTLs (Supplementary Fig. S8C). In HIF 334, however, this phenotype was not correlated with a consistent genetic combination of mapped QTLs (Supplementary Fig. S8C). This transgression suggests that the maintenance of a wild-type phenotype in Tsu-0 bracteate flowers relies on genetic interactions and may interfere with the FMI pathway.

To conclude, our genetic analysis demonstrates that the genetic control of basal bracts in Tsu-0 is complex, involving several loci with additive effects on bract development and some epistasis in the control of other pleiotropic traits. It also suggests that the causal genes differ from those previously identified in mutants with bracts.

Time-series transcriptomics reveal a peak of gene expression divergence at the floral transition, while FMI genes remain synchronized

Developmental programmes are governed by gene regulatory networks, often leading to substantial transcriptional changes. We therefore sought to identify genes whose expression changes with bract development. We profiled meristem transcriptomes before, during, and after floral transition in both accessions. Microdissected meristems were matched to the same four developmental stages: vegetative, late vegetative, transition, and floral (V, L, T, and F, respectively, Fig. 4A). We performed RNA-seq and used PCA to assess the consistency between replicates (Supplementary Fig. S9B).

Fig. 4.

Transcriptomic divergence between Tsu-0 and Col-0 meristems peaks at the floral transition despite synchronization of the FMI pathway genes. (A) SEM showing the evolution of the main meristem in both Col-0 and Tsu-0 at the four different stages (V, L, T, and F) used for RNA-seq. Plants were first synchronized by 21 d of a non-inductive SD light regime before a transfer to inductive LDs. The date after the transfer is indicated in the top-right corner of each image. Green arrowheads point to branches (with a leaf) while magenta arrowheads point to the first flowers produced after floral transition. In Tsu-0, red arrows show bracts (the bottom-right image—the last time point for Tsu-0—is identical to Fig. 1I). Scale bars: 50 µm. (B) Dynamics of the expression levels of FT, LFY, and AP1 (three important regulators of floral transition and identity) in both Col-0 (green) and Tsu-0 (red) over the four developmental stages. Differential expression analysis reveals no difference at any stage between the two accessions. (C and D) Confocal live imaging of the main meristems of Col-0 (C) and Tsu-0 (D) plants at floral transition, expressing a pLFY transcriptional reporter (magenta) and a membrane marker (green) for morphology (top-view projections of stack acquisitions). Green arrowheads, branches, ordered with decreasing numbers from the floral transition; white arrows, flowers ordered with increasing numbers from the floral transition. In both genotypes, pLFY expression is absent from branch meristems and suddenly appears in the first flower onwards. Representative images of at least six plants per genotype captured at floral transition. (E) Side-view of the first flower from (D; Tsu-0): a bract (red arrow) is visible on the abaxial side of this young flower. Scale bars: 50 µm. (F) Bar plots of the number of genes differentially expressed between two consecutive stages in Col-0 (green) and Tsu-0 (red). The number of changes peaks at the V-to-T stage transition, especially in Tsu-0. (G) Bar plots of the number of DEGs between Col-0 and Tsu-0 at each stage of the time course. T stage is when the highest number of DEGs is measured (on top of the bar, different letters indicate the statistical difference with a χ² test of homogeneity after post-hoc analysis). (H) Dynamics of the expression levels of five genes reported to be involved in bract development, showing a significant difference at least at the T stage between Col-0 (green) and Tsu-0 (red). *, Genes are differentially expressed; **, the absolute fold change is >1.

To monitor the progression from vegetative growth to flowering at the transcript level, we analysed the expression dynamics of several key regulators of floral transition, such as FLOWERING LOCUS T (FT) (Wigge et al., 2005), and of flower identity (e.g. LFY or AP1). The synchronized expression of these genes in the two accessions validated the precision of our morphological staging (Fig. 4B): despite different absolute flowering times (Fig. 4A; Supplementary Figs S3, S9A), our experimental design realigned the developmental progression of the two accessions using the floral transition as a common timer. Hence, gene expression that is desynchronized from flower formation can easily be tracked. To ensure that our bulk RNA-seq data do not obscure changes in gene expression patterns related to flower identity, we introduced an LFY transcriptional reporter in both accessions to precisely monitor the activation of this master floral identity gene in space and time during the floral transition. Live confocal imaging revealed similar dynamics in both genotypes: while LFY expression is not detected in branch meristems, it progressively increases in each floral primordium from the first flower onwards, and reaches high levels in the floral meristem proper, but remains absent or low on the external side of the primodium where the bract grows out in Tsu-0 (Fig. 4C–E; Supplementary Fig. S10). These results, in line with our genetic experiments, confirm that alteration of LFY dynamics during floral transition is not likely to be involved in the delayed repression of bracts. As previously reported (Chahtane et al., 2018), bracts in lfy mutants are not precisely located at the base of branches (Supplementary Fig. S10), in contrast to Tsu-0 basal bracts (Fig. 1B). This indicates that the developmental contexts leading to the formation of bracts in mutants versus basal bracts in natural accessions may be really distinct, in line with the involvement of distinct genetic pathways shown above (Fig. 3).

In the time-course, stage T should be the most relevant to study bract formation even if no bract is visible at this stage. Indeed, bract and flower initiations start before any morphological event (Heisler et al., 2005). The delay between stages T and F (a median of 1–2 d; Supplementary Fig. S9A) is compatible with the initiation of the first flowers and their bracts at stage T. Hence, transcriptional changes associated with the delay of bract repression should transiently appear at stage T and progressively fade by stage F (since only the first 1–5 flowers present a bract, Fig. 1). At the global transcriptome scale, stage T stood out as a critical transcriptomic transition in both accessions, with the highest number of gene expression changes upon reaching this stage (Fig. 4F). Interestingly, the divergence between the transcriptomes of the two accessions (as measured by the number of DEGs) also peaks at stage T, with up to 4759 DEGs (Fig. 4G; Supplementary Table S5), including genes potentially responsible for bract formation. In conclusion, the relevant time window during which bract repression is uncoupled from flower development corresponds to a period of highly dynamic gene expression, both throughout development of a plant and among plants with different genetic backgrounds.

Transcriptomic data identify candidate genes responsible for the desynchronization of bract repression

To identify putative genes impacting bract formation among DEGs at stage T, we first cross-referenced our transcriptomic (4759 DEGs) and QTL mapping data (2324 genes in intervals). We further filtered out genes that had no genetic polymorphisms (including variations with no putative functional impact; see the Materials and methods) between Col-0 and Tsu-0 (Fig. 5A, B). This selection yielded a list of 362 candidate genes spread over the four QTLs (Supplementary Table S8), including 24 transcription factors, and none of them had previously been directly associated with bract formation. SPL2 was found to be more expressed at stage T in Tsu-0 than in Col-0 (+29%, Fig. 4H). Three other SPL genes not located within mapped QTLs (SPL3, SPL5, and SPL10) also showed increased expression at stage T (Supplementary Fig. S11A). However, the only two reports of a bract phenotype following SPL perturbations are contradictory: one was observed after SPL3 overexpression (Cardon et al., 1997) and the other after global SPL down-regulation using a pFD::MIR156 construct (Wang et al., 2009). To our knowledge, despite a wide variety of genetic perturbations (single or combined spl mutants, strong overexpression of distinct SPL genes, milder overexpression of rSPL constructs resistant to miR156 targeting, and use of 35S::MIM156 for target mimicry), this bract phenotype has not been reproduced elsewhere in the Arabidopsis SPL literature (Cardon et al., 1999; Schwab et al., 2005; Wu and Poethig, 2006; Gandikota et al., 2007; Schwarz et al., 2008; Wang et al., 2008; Shikata et al., 2009; Usami et al., 2009; Wang et al., 2009; Yamaguchi et al., 2009; Hyun et al., 2016; Xu et al., 2016; Werner et al., 2021; Hu et al., 2023). Hence, further work is needed to test the role of these candidate genes, including SPL2, as causal genetic factors driving bract formation in Tsu-0. However, our analysis suggests that basal bract formation in a natural accession can be genetically determined by a novel set of genes, probably present in this list of candidates.

Fig. 5.

Identifying a transcriptomic signature of leaf development in shoot meristems at the floral transition stage using time-series RNA-seq. (A) Intersecting genetic mapping and transcriptomic data: if a genetic variant causing Tsu-0 bracts also induces the differential expression of the gene in cis, this gene should verify the conditions defined by conditions A, B, C, and D. (B) Evolution of the number of candidate genes mapped in the four QTLs (condition A) by the successive addition of the other conditions B–D. (C) Strategy to cluster genes based on the presence (ellipse with the solid line) or the absence (dotted ellipse) of a leaf and/or bract in the different combinations of stages and genotypes (see text for details). (D) Volcano plot of gene expression at stage T between Tsu-0 and Col-0. All genes expressed in the SAM are plotted (n=21 652 grey dots) but only the genes fulfilling the clustering condition defined in (C) (‘bract signature genes’) are highlighted in orange and blue for up- and down-regulation, respectively (n=124). Vertical dashed lines, absolute fold change >1; horizontal dashed line, significance threshold at 5×10⁻² (adjusted P-value with the FDR method). (E) Seven ‘biological process’ GO terms are statistically enriched among the up-regulated genes identified in (D) (bar plot providing the number of related genes). They are all related to the anthocyanin biosynthesis pathway, which is cited in the two most significant terms (asterisks). (F) Only three ‘biological process’ GO terms are statistically enriched among the down-regulated genes identified in (D) (bar plot providing the number of related genes); they collectively point to the defence response and, more specifically, the SA pathway (asterisk). In (E and F), colour gradients indicate the statistical significance of the GO term enrichment (Fisher’s test with Yekutieli adjustment method). (G) Representative images of micro-dissected meristems in Col-0 (upper row) and Tsu-0 (lower row) just at, before (left), or after (middle) stage T (n>15 for each genotype) and a close-up of the base of the bolted main stem (right). At stage T, Tsu-0 meristems display a typical anthocyanin red coloration just below the meristem, which is not observed in Col-0. After bolting, both genotypes show anthocyanin coloration at the base of the stem. Scale bars: 100 µm (left and middle), 1 cm (right). (H) Intersection of the gene set in (A) with the ‘bract signature genes’. (I) Examples of the expression profile of two candidate genes, showing an up- (left) and a down- (right) regulation at the T stage. DFR is an enzyme involved in anthocyanin biosynthesis while FMOGS-OX7 is annotated as an SA-responding enzyme. Two asterisks indicate that the genes are differentially expressed and the absolute fold change is >1. The bract and bract-less clusters are outlined with a solid and dashed ellipse, respectively, while the horizontal dotted line highlights their separation.

To determine whether these causal genetic variations influence the expression of other genes known for their role in bract development (Supplementary Table S1) or act through an independent mechanism, we systematically analysed the expression dynamics of these genes. Only PUCHI, SUPPRESSOR OF CONSTANS 1 (SOC1), TERMINAL FLOWER1 (TFL1), and LSH1 significantly differed at stage T between Tsu-0 and Col-0 (Fig. 4H; Supplementary Fig. S9D). Single puchi, soc1, or lsh1 null mutations alone do not induce basal bracts (Rieu et al., 2024; Supplementary Fig. S12C, D), so the isolated effect of their reduced expression in Tsu-0 at stage T (54, 52, and 41%, respectively) probably does not impact bract development. By transforming lateral branches into flowers, tfl1 mutants display basal bracts (Supplementary Fig. S12C, D). However, tfl1 phenotypes are recessive: in heterozygous TFL1/tfl1 plants, termination (the apical meristem is transformed into a terminal flower) does not occur and we observed no basal bract formation. If TFL1 expression at stage T in Tsu-0 is reduced (−71%), the absence of termination indicates that this expression level in the apical meristem is sufficient to prevent its transformation to a terminal flower. Therefore, why this reduced TFL1 level would affect bract development remains unclear. Furthermore, reducing TFL1 activity reduces the number of branches (Ratcliffe et al., 1999). However, Tsu-0 plants form more branches than Col-0 (Supplementary Fig. S7E), indicating another TFL1-controlled trait that is insensitive to the reduction in its expression. Altogether, this supports the idea that basal bract formation in Tsu-0 does not primarily rely on variation in TFL1 expression levels. To our knowledge, however, combinations of these mutations have not been reported. Moreover, these gene expression differences persisted in stage F while bracts disappeared with older flowers: hence, the reduced expression of PUCHI, SOC1, LSH1, and TFL1 is not correlated with bract formation at all developmental stages. Altogether, this suggests that both the natural genetic determinism of delayed bract repression and the underlying transcriptional programme acting in the meristem are largely independent of the genetic pathway identified in bract mutants.

Using time-series to identify a transcriptomic signature of leaf/bract development in the shoot meristem

To identify biological functions specific to the critical stage T in Tsu-0 without prior assumptions, we performed a GO term analysis on all DEGs compared with Col-0. However, the results were not very informative, notably lacking terms related to leaf development (Supplementary Fig. S13A-C). Since Tsu-0 and Col-0 are distant backgrounds, it is not surprising that many gene expression differences at stage T are not related to bract formation or even to development (see the second axis of the PCA, Supplementary Fig. S9B). To specifically isolate the gene regulatory network responsible for delayed bract repression in Tsu-0, and thanks to the information provided by the time-series, we designed a clustering method to isolate a transcriptomic signature common to all leaf-producing stages versus non-leaf producing stages. Briefly, expression data were gathered in two groups corresponding to either bract-less (Col-0 T, Col-0 F, and Tsu-0 F) or leaf/bract-producing meristems (the other samples, including Tsu-0 T) and we selected genes whose expression profile allows definition of a unique expression level that separtes these two groups (Fig. 5C). These genes can be considered ‘bract signature genes’, with a positive or inverse correlation with the presence of a bract if their expression is up- or down-regulated (respectively) at stage T in Tsu-0.

Only 124 genes met these criteria (Fig. 5D; Supplementary Table S9), including SOC1, whose expression anti-correlates with the presence of a bract. Interestingly, two enriched GO terms emerged from this list of bract signature genes: anthocyanin biosynthesis and salicylic acid (SA), associated with up- and down-regulated genes, respectively (Fig. 5E). Backtracking these ontology terms for all the 4759 DEGs at stage T (Supplementary Tables S10, S11) retrieved even more genes associated with these pathways, supporting that their activity levels differ between Tsu-0 and Col-0 at the critical stage T (Supplementary Fig. S13D). High anthocyanin production in Tsu-0 at stages L and T was evident from the frequent purple coloration just below the meristems, contrasting with the green tissues in Col-0 (Fig. 5G). Sometimes, this purple colour extended to young organs where bracts were initiated (Supplementary Fig. S13E). As the stem grew, the pigment receded to the rosette junction in both accessions (Fig. 5G), vanishing from the apical meristem. This trait specific to Tsu-0 validates our clustering strategy: it is able to capture genes whose expression differs between Tsu-0 and Col-0 at stage T and with overall contrasted levels between vegetative and flower stages, as expected for a gene regulating bract repression in synchronization with flower development.

Interestingly, among these 124 bract signature genes, 12 also lie in the mapped QTLs (Fig. 5H), including one anthocyanin biosynthetic enzyme (dihydroflavonol reductase, DFR) and five SA-responsive genes (see Supplementary Table S9). For example, Fig. 5I shows the expression profiles of two such genes. However, further work is required to test whether these candidates are responsible for basal bract formation in Tsu-0 and whether the anthocyanin and/or SA pathways are involved in this natural variation.

Curve registration revealed massive gene desynchronization and challenged the hypothesis of a ‘prolonged vegetative state’ during floral transition in Tsu-0

In the previous clustering approach, both up- and down-regulated genes display similar temporal dynamics: they tend to be delayed in Tsu-0 compared with Col-0 (Fig. 5I). This result suggests that a transcriptomic leaf signature is maintained at stage T in Tsu-0 compared with Col-0. This desynchronization of the expression of some genes with the developmental clock imposed in our experiment (from stage V to F) identifies transcriptomic heterochronies (i.e. changes in the timing of gene expression).

To assess the importance of heterochrony in the divergence of Tsu-0 and Col-0 transcriptomes, we first tested whether its effects could be detected in a global unsupervised analysis of gene expression levels over the time-course. We therefore performed a PCA and plotted the two main axes of variance after grouping samples by genotype and time points (Fig. 6A). While the second axis of variance (∼24.4%) differentiates genotypes, the main axis (47.9% of variance) orders the sampling time points chronologically in both genotypes, representing a transcriptomic age. Surprisingly, Tsu-0 stage T does not align with Col-0 stage T on this axis: it is closer to stage F, explaining the limited expression changes when progressing to the next stage F (see Fig. 4F).

Fig. 6.

Massive and complex transcriptional desynchronization patterns co-exist in Tsu-0 during the floral transition. (A) PCA of RNA-seq data from microdissected meristems of Col-0 and Tsu-0 during the four sampled stages (biological replicates averaged per time point). The two main axes of the PCA can be interpreted as developmental time and genotype, respectively. (B) Examples of temporal registration of gene dynamics (right panels) between Col-0 (green) and Tsu-0 (red) from scaled expression levels (left panel). Dots represent the expression levels of independent biological replicates, while lines indicate the mean expression level at each time point. In the right panel, the green (or red) dotted curves represent the fitted models for Col-0 (or Tsu-0) independently, while the grey dotted curve represents the joint model for both Col-0 and Tsu-0. If the green and red dotted curves are used, it means the two time series are best explained by two independent models, indicating that they are not similar. Conversely, if the grey dotted curve is used, it suggests that a single model best explains both time series, indicating that they are similar. The name of the gene plotted is indicated on the left of each row. These four genes exemplify the four possible categories (from top to bottom, respectively): genes that cannot be registered (n=43, e.g. CYP705A9); genes with identical temporal dynamics (shift=0, e.g. AP1, see also Fig. 4B); and genes whose expression dynamics in Tsu-0 must be shifted negatively (e.g. DFR; see also Fig. 5I) or positively (e.g. AG) to align with Col-0. The last column provides a biological interpretation of the computed shift (see text): a null shift indicates that the expression dynamics in Tsu-0 stay ‘in phase’ with the floral transition, while negative or positive shifts indicate that the expression dynamics in Tsu-0 are desynchronized later or earlier, respectively, than the phenotypic progression of floral transition. (C) Distribution of heterochronic shifts between Tsu-0 and Col-0 on the entire meristematic transcriptome, computed by the registration method from (B). The shift value is colour-coded in a red-to-blue gradient from −1 to 1. (D) Correlation between the heterochronic shifts in gene expression and the fraction of genes presenting differential expression between Tsu-0 and Col-0 for each stage, colour-coded from green to orange. The blue dotted lines are two piecewise linear regressions modelling the effect of the shift with a binomial model for positive and negative values (P-value <2×10⁻¹⁶) surrounded by a 5% confidence interval (grey shading). Asterisks highlight the significantly higher proportion of DEGs at the T stage (P-value: ***<0.001, *<0.05 ; see also Supplementary Fig. S14 for the complete statistical analysis). (E) GO term enrichment analysis associated with the three categories of heterochronic shifts. The list of all significant ‘biological process’ GO terms (BH-adjusted P-value <0.05) was simplified using semantic similarity (cut-off=0.7), and the rich factor was computed for the remaining terms, revealing the proportion of genes involved among all the genes associated with this GO term. Dot size indicates the count of genes, and the colour scale is the statistical significance (BH-adjusted P-value) of the enrichment in the shift category. Asterisks indicate GO terms referring to developmental processes.

As suggested by Calderwood et al. (2021), the transcriptomes of two genotypes during flowering cannot be aligned to a single developmental time; each gene may desynchronize differently, sometimes in opposite directions. To quantify desynchronization of each gene between Col-0 and Tsu-0, we used curve registration (Calderwood et al., 2021) that can predict subtle temporal shifts of gene expression (Kristianingsih, 2024). We analysed the shift relative to the floral transition, used as a common reference clock between the two accessions. Thus, genes with no shift, such as AP1 (Figs 4B, 6B), can nevertheless be shifted in absolute time, because Tsu-0 plants flower later with an older absolute age (Supplementary Figs S3, S9A). In our approach, positive and negative shifts reflect desynchronization relative to the floral transition, and gene expression dynamics may shift earlier or later. Unlike our previous clustering approach (Fig. 5C), this registration method uses a common scaled expression level and computes only temporal shifts, regardless of absolute expression levels (Fig. 6B).

Only a few genes could not be registered (Fig. 6B, n=43), suggesting that most genes follow similar temporal dynamics in both accessions (see examples in Fig. 4B and H, Supplementary Fig. S9C and D, or Supplementary Fig. S11A). Most genes could be classified into three categories: null, negative, and positive shifts (Fig. 6B). A null shift indicates that the gene expression in Tsu-0 stays ‘in phase’ with flowering, like AP1 (compared with Fig. 4B). A positive or negative shift indicates that gene dynamics in Tsu-0 occur earlier (e.g. AG) or later (e.g. DFR) relative to the floral transition, revealing heterochronies between transcriptomic and phenotypic processes. Transcriptome-wide, the shifts were broadly distributed, showing massive and complex desynchronization of transcriptome dynamics (Fig. 6C). Hence, the floral transition in Tsu-0 did not impose a synchronization on the entire meristematic transcriptome: only ∼18% of genes (n=3879) stayed in phase with this process, while the majority showed delayed or advanced expression dynamics. Even more genes shifted earlier relative to the floral transition, supporting the PCA interpretation (Fig. 6A). Since curve registration computes temporal shifts in gene expression regardless of absolute expression levels, we investigated whether these shifts could affect differences in gene expression levels. Indeed, there is a strong correlation (P-value <2×10⁻¹⁶) between the absolute shift value and the proportion of DEGs between Tsu-0 and Col-0 at all developmental stages (Fig. 6D). This suggests that the desynchronization of gene dynamics between the two accessions contributes to differences in gene expression levels. Furthermore, this effect is significantly higher at the T stage for positive shift values ranging from 0.25 to 0.75 (Fig. 6D; Supplementary Fig. S14): this suggests that genes with ‘earlier’ dynamics in Tsu-0 explain at least partially the peak of transcriptome divergence observed at the T stage (Fig. 4G).

Altogether, the differences in the transcriptomes of Tsu-0 and Col-0 are impacted by a significant level of heterochrony, with complex combinations of temporal shifts. To assess this idea, we focused on the MIR156/157-regulated SPL genes, that are key promoters of maturation in vegetative phase change, shoot maturation, flowering, and branching. Their activity builds up progressively as the levels of MIR156/157, their antagonistic juvenile factor, decrease with time (Wang and Wang, 2015). Strikingly, this small functional gene group recapitulates the findings established for the entire genome: (i) the expression dynamics of this group are not driven by a single heterochrony and the different SPLs show a spread of shifts relative to the meristematic floral transition; (ii) there are more SPL genes with an ‘early’ shift than a late or zero shift; and (iii) there are more genes differentially expressed at stage T among the genes with an early shift (Supplementary Fig. S11C). In conclusion, although the bract is often considered a ‘juvenile’ vegetative structure, its formation at the floral transition does not indicate a global juvenilization at the transcriptomic level.

Investigating which biological processes are affected by desynchronization of gene expression in meristems during flowering

As expected, genes in phase were primarily linked to flowering and developmental phase change, as shown by key regulators of these pathways (Supplementary Fig. S15), which also emerges from GO term enrichment analysis (Fig. 6E). In contrast, genes previously associated with bract mutants exhibit a wide range of shifts, from very early to very late. This shows that they are variably affected by these desynchronizations (Supplementary Fig. S15A, C), reinforcing the idea that they are not involved in basal bract formation. GO term analysis for each shift category highlighted the particular processes desynchronized from flowering in Tsu-0 compared with Col-0 (Fig. 6D). For instance, vascular differentiation (two GO terms mentioning tracheary element and secondary cell wall) emerges as a process shifted earlier in Tsu-0 relative to the floral transition (Fig. 6D). Vascular differentiation dynamics may be driven by other progression factors such as absolute age, suggesting a loose connection with the flowering pathway. Conversely, processes shifted later in Tsu-0 revealed terms related to cell division (spindle, cell cycle, and mitosis) and ribosomal biogenesis (six GO terms mentioning ribosome, rRNA, ribonucleoprotein, and protein–RNA complexes; Fig. 6D), which suggests that core meristematic functions are prolonged in Tsu-0. Further work is needed to explore whether and how the desynchronization of these processes relative to floral transition impacts bract de-repression.

Discussion

In this study, we investigated the transient formation of ‘basal bracts’ at the base of inflorescences in natural accessions of the model plant in A. thaliana. We characterized basal bract formation at developmental, genetic, and transcriptomic levels. We found that this phenotype is associated with a desynchronization of bract repression from flower formation during the floral transition.

The mechanisms desynchronizing bract repression and flower development

Evidence from the literature suggests that flower development, through redundant actions of the genes from the FMI pathway, represses bract development, hence naturally coupling flower establishment and bract repression. However, natural basal bracts demonstrate that these two processes can be uncoupled in a short time window during the floral transition when both flowers and bracts develop together.

The wild-type phenotype of basal bract-associated flowers (Fig. 1) and the proper dynamics of many FMI genes (e.g. LFY and AP1; Fig. 4) suggest that the transient bract de-repression at floral transition is not a consequence of an impaired FMI establishment, in contrast to previously reported mutants. In line with this, bracts are rarely produced with the first flowers in mutants of the FMI pathway (Supplementary Fig. S12). Our quantitative genetic (Fig. 3) and transcriptomic data (Fig. 5) further show that the genes causing this transient bract de-repression are different from the handful of FMI genes whose mutants display bracts.

However, the precise mechanisms involved remain unresolved. Finer mapping or an alternative quantitative genetic approach such as a genome-wide association study (GWAS) could help pinpoint causal genes for their definite validation. However, the four identified QTLs, including two with additive effects (Fig. 4; Supplementary Fig. S6), indicate that several complex genetic mechanisms may be at work. Our transcriptomic analysis established correlations of DEGs and pathways in the meristem at the time when bract de-repression occurs: up-regulation of the anthocyanin pathway, down-regulation of the SA pathway (Fig. 5), or prolonged activity of ribosome biogenesis or the cell cycle (Fig. 6). This. however, does not demonstrate their role in promoting bract development: a transcriptomic signature of leaf development in stage T meristems could simply be a consequence of the developing bracts rather than its cause. Also, despite the specificity of our dataset for micro-dissected meristems, bulk RNA-seq has technical limitations. First, subtle gene expression changes, possibly restricted to specific meristematic subregions, are likely to fall below the detection threshold. Second, variations at the post-transcriptional level are simply not accessible with this technique.

Interestingly, several plant species such as grasses (Poaceae) evolved leaf-less inflorescences, through apparently convergent and unrelated mechanisms (Mach, 2010). Indeed, Arabidopsis and Poaceae mutant bracts are not strictly homologous and, so far, different genetic mechanisms have been reported (Whipple et al., 2010; Whipple, 2017; Xiao et al., 2022). However, some of these mechanisms are still partially unresolved (Kawakatsu et al., 2009; Wang et al., 2021). Since in both groups, bract exclusion from wild-type inflorescences involves a leaf repression mechanism synchronized with a developmental phase change, further research on these two groups may reveal whether they share more similarities than those currently described.

Linking floral transition, transcriptome divergence, and developmental robustness

We found that gene expression differs the most between the two accessions at the floral transition, at the time when bracts are de-repressed in Tsu-0 (Fig. 4). Furthermore, curve registration identified an important level of gene desynchronization, which probably contributes to the divergence in gene expression levels, especially at stage T (Fig. 6). Specifically, these desynchronizations alter the expression timing of many genes relative to a minor gene pool driving the floral transition, which we used as a reference in our experiments.

As flowering is a key adaptive trait that involves transcriptional reprogramming, the floral transition may be a developmental stage that leads to large transcriptome differences among natural populations. Indeed, massive and complex gene desynchronizations during floral transition were reported at both the intra- and inter-species level in Brassica rapa cultivars and A. thaliana (Calderwood et al., 2021). This phenomenon may arise from the strong selection pressure adapting flowering time to the environment (Bloomer and Dean, 2017). As a result, the flowering-related genes could shift and desynchronize from numerous others not functionally related to flowering, creating a peak of transcriptome divergence in the short time window when flowering-related genes vary rapidly (Supplementary Fig. S16). In line with this, a peak of gene expression divergence has also been reported in meristems transitioning from vegetative to floral stages between different species of Solanaceae (Lemmon et al., 2016).

In different unrelated phyla such as animals, plants, brown algae, or fungi, the ‘hourglass’ (Kalinka et al., 2010; Quint et al., 2012; Cheng et al., 2015; Lotharukpong et al., 2024) and ‘inverse hourglass’ (Lemmon et al., 2016; Levin et al., 2016) models propose that reduced or increased, respectively, transcriptome divergence at specific stages of development drive lower (respectively higher) phenotypic variations in the final body plans. For instance, the high transcriptome divergence between Solanaceae species during floral transition correlates with the diversification of their inflorescence architectures (Lemmon et al., 2016). Here, we report a similar correlation within natural populations of A. thaliana: the transcriptome divergence peaks at floral transition, which correlates with a developmental variation, the formation of basal bracts. Developmental robustness measures the invariance of development to stochastic, environmental, and genetic perturbations (Félix and Wagner, 2008; Hallgrimsson et al., 2019). Here, we report that natural genetic variation alters bract suppression in A. thaliana during the short time window of floral transition. Interestingly, several other observations suggest that developmental robustness is reduced during floral transition in Brassicaceae, at both inter- and intra-species levels. In addition to basal bracts (Supplementary Fig. S1), botanists have also documented examples of frequent flower-to-shoot transformations or flower dimorphisms (Arber, 1931a, b). A possible explanation would be that, as a consequence of the constant adaptation of flowering time, the many changes in gene expression lead to desynchronization of genetic programmes occurring during the short floral transition, creating new, transient gene expression states that fuel developmental variation (Supplementary Fig. S16). More generally, this questions whether rapid developmental transitions of adaptive traits that change the timings of events may result in more gene desynchronization and developmental variation. To test this hypothesis for the case of bract development, further investigation into the identification of the genetic and developmental mechanisms that uncouple bract repression from flower formation in natural A. thaliana accessions will be important. Such follow-up studies may also shed light on how the presence of bracts has frequently evolved within and between species, both at the base of branches linked with floral transition and also in entire inflorescences.

Supplementary data

The following supplementary data are available at JXB online.

Fig. S1. Example of natural transient bract formation at the base of branches among Brassicaceæ tribes.

Fig. S2. Fine mapping of the QTLs controlling basal bract formation in Tsu-0.

Fig. S3. The presence of basal bracts in A. thaliana accessions does not correlate with their geographic or genetic origin.

Fig. S4. Tsu-0 plants flower later than Col-0 plants in different culture conditions.

Fig. S5. Bulk segregant analysis of basal bract formation (Tsu-0×Col-0) identifies four putative QTLs.

Fig. S6. Quantitative genetics of formation of basal bracts in Tsu-0 using a set of RILs with the reference accession Col-0.

Fig. S7. QTL1b* identified for basal bracts correlates with a higher cauline branch number.

Fig. S8. Transgressive phenotypes in HIFs reveal epistasis controlling floral determinacy between alleles of Col-0 and Tsu-0.

Fig. S9. Gene expression dynamics during the floral transition in micro-dissected meristems of Col-0 and Tsu-0 plants.

Fig. S10. Time-series of confocal images showing the sudden up-regulation of pLFY at the floral transition in Col-0 and Tsu-0 apical meristems.

Fig. S11. Transcriptional dynamics of SPL genes in Tsu-0 and Col-0 during the floral transition.

Fig. S12. Phenotypic differences between the bracts of the mutants affected in the FMI pathway and the natural basal bracts observed in Tsu-0.

Fig. S13. Functional analysis of the genes differentially expressed in Tsu-0 meristems at stage T when bract repression is delayed.

Fig. S14. Evolution of the proportions of DEGs between Tsu-0 and Col-0 depending on the heterochronic shift and the developmental stage

Fig. S15. Temporal registration of expression dynamics in Tsu-0 during the floral transition for genes related to bract development, floral identity, and floral transition.

Fig. S16. A working model for natural basal bract formation in Arabidopsis thaliana.

Table S1. Literature presenting mutants in genes functionally associated with the FMI pathway in Arabidopsis thaliana and showing de-repressed bract formation.

Table S2. Genotypes of all RILs and HIF lines used in this study.

Table S3. SNP information relative to new KASP genotyping markers used in this study.

Table S4. SNP information relative to new Sanger genotyping markers used in this study and associated primers.

Table S5. Differential gene expression analysis and log2CPM for the genes expressed in microdissected meristems of Tsu-0 and Col-0 during the floral transition.

Table S6. Results of curve registration.

Table S7. List of annotated genes lying in the four mapped QTLs controlling bracts in Tsu-0, with additional information from RNA-seq and genomic variant analysis.

Table S8. Intersection of the genetic mapping, genomic variant analysis, and RNA-seq data.

Table S9. Bract signature genes in Tsu-0 identified by clustering apart expression from bract- and non-bract-forming stages in Tsu-0 and Col-0 time-series.

Table S10. Details of all genes associated with ‘response to salicylic acid’, their expression at T stage in Tsu-0, and their identification as bract signature genes.

Table S11. Details of all genes associated with anthocyanin metabolism, their expression at T stage in Tsu-0, and their identification as bract signature genes.

Artificial intelligence-based language correction software was used throughout the text to ensure accurate spelling and grammar, as well as to achieve the most comprehensible style possible, without influencing the content.

Author contributions

FB and SD: conceptualization; FB: project administration and writing—original draft; FB, SD, and SH: investigation; FB, RK, and SD: software—implementation, methodology, and visualization; FB and RK: software—programming; RK: software—design and testing; BJ, FB, RK, and SD: formal analysis; FB, RK, RM, RW, and SD: writing—review and editing; FB, RM, RW, and VV: supervision; FB, RM, and RW: funding acquisition.

Funding

We thank INRAE for a grant from the BAP department (IB2019) and the French Embassy in the UK for a short-term travel fellowship awarded to RK. RM and RW gratefully acknowledge support from the Biotechnology and Biological Science Research Council Institute Strategic Programme ‘Building Resilience in Crops’ (BB/X01102X/1) and BBSRC Strategic Longer and Larger Grant ‘BRAVO’ (BB/P003095/1). SD was supported by a PhD contract of the French Ministry of Research and Higher Education.

Data availability

Whole-genome DNA-seq data of the parental lines (Col-0 and Tsu-0) and of the two F₂ pools used for BSA have been deposited in the Recherche Data Gouv database; https://doi.org/10.57745/Z80SIM (Dieudonné, 2024a). Time-course RNA-seq data of Col-0 and Tsu-0 micro-dissected meristems during flowering are also available via the Recherche Data Gouv database; https://doi.org/10.57745/DKMQ06 (Dieudonné, 2024b) and https://doi.org/10.57745/7JI3E7 (Dieudonné, 2024c), respectively.