Genome-scale transcriptome augmentation during Arabidopsis thaliana photomorphogenesis

Abstract

Plant photomorphogenesis is a light-induced developmental switch that combines massive reprogramming of gene expression and a general enhancement in RNA Polymerase II activity. Yet, transcriptome analyses have failed to demonstrate any tendency toward gene upregulation. To solve this conundrum, we use a spike-in RNA-seq experimental and bioinformatic pipeline, enabling to reconcile transcriptome dynamics with epigenomic and cytogenetic observations of Arabidopsis thaliana cotyledon photomorphogenesis. During the transition, a quasi-unilateral impact of light, with 94% of the differentially expressed genes being upregulated within the first six hours, triggers a two-fold increase in cellular transcript levels. This augmentation of the transcriptome is detected at a similar strength in spike-free RNA-seq datasets re-normalized using stable endogenous transcript levels that mimic the spike-in information. Reanalyzing light-mediated gene regulatory pathways from this standpoint further reveals a quasi-exclusive positive effect of ELONGATED HYPOCOTYL 5 (HY5) and other key light-induced transcription factors on target genes. This study provides a paradigm shift for understanding global genome regulation by light and opens the way to investigate transcriptome size control during other developmentally or environmentally controlled cellular transitions in plants.

Authors present epigenomic, cytological, and transcriptomic evidence of a two-fold increase in transcriptome size during cotyledon photomorphogenesis, providing a paradigm shift in the study of genome regulation during plant cellular transitions.

Introduction

Plants display extraordinary developmental plasticity to external cues, especially when it comes to light. In addition to fueling photosynthesis, light is sensed and signaled to continuously modulate plant physiology, development, and cellular status. These dynamic adjustments to local environmental conditions are largely achieved through the regulation of gene expression patterns^1–3. The spectacular effect of light on plant gene expression has most often been studied when a germinating plant is exposed to light for the first time, thereby triggering de-etiolation —the transition from skotomorphogenesis to photomorphogenesis. In complete darkness, DE-ETIOLATED-1 (DET1), CONSTITUTIVE-PHOTOMORPHOGENIC-1 (COP1), and several other repressors of photomorphogenesis induce seedling etiolation by promoting hypocotyl (stem) elongation and inhibiting the development of embryonic leaves (cotyledons)^4–6. Once red, blue, or UV-B light is perceived by photoreceptors, complex signaling transduction events induce photomorphogenesis with the onset of chloroplast biogenesis and photosynthesis^3,7. Former studies indicated that as much as 40% of the nuclear genes are differentially expressed during the transition^8–12 in an organ- and cell-specific manner through the action of multiple light-regulated transcription factors (TF) such as PHYTOCHROME INTERACTING FACTORS (PIFs), HYPOCOTYL ELONGATION-5 (HY5) and TEOSINTE BRANCHED1/CYCLOIDEA/PCF (TCPs)^8–11,13.

In cotyledons, light-mediated transcriptome changes involve gene-specific and large-scale variations in chromatin composition and 3D organization^14,15, together with nucleus expansion and endoreduplication, a mechanism by which the nuclear DNA content increases^11,16. This general adaptation of the nuclear status may sustain a global increase in transcription, evidenced by a higher proportion of RNA Polymerase II (RNA Pol II) engaged in transcript production¹⁷ and an enrichment at most genes of monoubiquitinated histone H2B (H2Bub)^18,19, a histone modification functionally associated with RNA Pol II elongation. Induction and maintenance of a hypotranscriptional status in darkness, therefore, plausibly result from a genome-wide attenuation of transcription, which would be released upon light perception through a massive wave of gene induction. This initial evidence suggests a significant rise in cellular mRNA levels in response to light. However, transcriptomic investigations have consistently identified approximately the same number of genes being up- or down-regulated during seedling or cotyledon de-etiolation, and none have demonstrated a clear trend in gene upregulation^8–11. This conundrum most plausibly results from the capacity of classic transcriptomic analyses to report relative changes in transcript abundance while ignoring the fundamental importance of absolute quantification. Multiple conceptual and experimental studies have identified technical difficulties in quantifying absolute variations in mRNA steady-state levels using RNA-seq when cellular global transcript abundance differs between samples^20–24. Using standard bulk (averaged from a cell population) or single-cell RNA-seq methodologies, the information needed to estimate absolute transcript abundance is usually lost during sample collection or RNA extraction and further vanishes upon data normalization, as count data generated by sequencing technologies is compositional²⁵. Hence, while former studies have gained insightful information on light-mediated gene expression changes during photomorphogenesis and other developmental transitions, we ignore how transcriptional intensification impacts transcriptome dynamics in plant cells. This lack of knowledge further impedes proper integration of transcriptomic data with genomic profiling of RNA Pol II, nascent transcripts, transcription-associated factors, or epigenome changes.

In this study, combining epigenomic profiling and live imaging, we first report that light sensing triggers a genome-scale augmentation of the transcriptional regime, marked by a doubling of RNA Pol II activity in individual nuclei. We further reveal how transcription intensification functionally impacts transcriptomic patterns during Arabidopsis thaliana cotyledon photomorphogenesis. To overcome the technical bottlenecks of quantitative RNA-seq approaches, we deployed a dedicated spike-in RNA-seq approach and analysis pipeline. This showed that cotyledon photomorphogenesis leads to 94% upregulation vs 6% downregulation, resulting in a 2-fold increase in transcriptome size. This approach enabled the identification of stably expressed genes during cotyledon de-etiolation, which we then used to renormalize publicly available transcriptome datasets. This, in turn, demonstrated the possibility of recovering accurate information on light-mediated transcriptome size augmentation in independent studies lacking a spike-in approach. Our study reconciles insights into global changes in steady-state RNA levels with epigenome dynamics and RNA Pol II activity, further providing a paradigm shift for plant gene expression studies.

Results

RNA Pol II activity increases during cotyledon photomorphogenesis

To estimate the levels of RNA Pol II engaged in transcript production in single nuclei from light-grown versus dark-grown cotyledons (Fig. 1A), we employed an immunocytometry approach using two antibodies that recognize the elongating Serine 2 phosphorylated (S2P) and the inactive unphosphorylated (NP) RNA Pol II forms (Supplementary Fig. S1A). This first showed that both S2P and NP RNA Pol II levels per nucleus increased proportionally to ploidy (Fig. 1B and Supplementary Fig. S1B), in agreement with the proposed role of endoreduplication in sustaining high messenger RNA (mRNA) production rates^26–30. Corroborating previous analyses¹⁷, we also detected a 1.6- to 2-fold higher proportion of elongating RNA Pol II in cotyledon nuclei of light-grown compared to dark-grown seedlings (Fig. 1B). The effect of light was significant for each ploidy class and therefore independent of the amount of DNA serving as a template for transcription. Last, analyzing each antibody signal separately showed that the S2P form is more abundant in the light than in the dark, while the NP form is less so (Fig. 1B and Supplementary Fig. S1A). This contrast increases with ploidy, suggesting that RNA Pol II is impacted differently depending on the number of genome copies in the nucleus, the cell type, or the differentiation status.

Fig. 1. Light impacts the transcriptional status of cotyledon nuclei.

A Analyses in this panel employed cotyledons dissected from dark- or light-grown seedlings. B RNA Pol II activity in individual cotyledon nuclei was determined by flow cytometry as the ratio of fluorescent signals after immunolabeling of its Serine 2 phosphorylated (S2P) and non-phosphorylated (NP) forms for each ploidy class (Supplementary Fig. S1A). The intensity of individual signals was measured consistently with the same settings and is plotted as arbitrary units. For the intensity ratio, the mean of the medians of the 2 C Dark samples was arbitrarily set to 1. Each dot represents the measure in a single nucleus. Three independent biological replicates were used, each comprising cotyledons from several seedlings. Values from at least 437 nuclei were aggregated (i.e., the number of measured nuclei in a single ploidy class of a single biological replicate). The p-value from a two-sided t test (see “Methods”) is given between dark and light for each ploidy class (Source data are provided as a Source Data file). C Live imaging of cotyledon mesophyll nuclei expressing RNA Pol II S2P mintbody-GFP and H2B-mRuby reporting local signals of the elongating RNA Pol II and chromatin. 3D projections of deconvolved images showing the levels of RNA Pol II S2P (green), H2B-mRuby (magenta), and their ratio (“Fire”)³¹. See Supplementary Fig. S1C for further details. Source data are provided as a Source Data file. D Quantification of RNA Pol II-S2P and H2B total nuclear signals calculated as the sum of pixel intensity in each analyzed nucleus, and RNA Pol II-S2P/H2B signal ratio calculated as the mean of the intensity ratio across the pixels of each nucleus (n nuclei = 14, n plants = 3 and n nuclei = 11, n plants = 5 for dark- and light-grown seedlings, respectively). The p-values are calculated from a two-sided Welch’s t test comparing the means of biological replicates (individual plants) between Dark and Light. E RT-qPCR analysis of AtRS31 pre-mRNA accumulation. The bars indicate the ratio of the signal obtained with primer pairs recognizing a proximal and a distal position on the AtRS31 pre-mRNA. Error bars represent the standard deviation between two independent biological replicates. Source data are provided as a Source Data file. F ChIP-seq meta-profiles of RNA Pol II S2P and H2Bub over the ensemble of genes marked in Dark or Light samples.

To track RNA Pol II activity in live cotyledons, we imaged a ratiometric Modification-specific Intracellular antiBODY (mintbody) S2P RNA Pol II reporter line³¹ (Fig. 1C, D). Following two-photon 3D microscopy, we quantified S2P mintbody-GFP vs H2B-mRuby live fluorescence levels per pixel in cotyledon nuclei of dark and light-grown seedlings (Supplementary Fig. S1C). On average, total S2P RNA Pol II in whole nuclei and local S2P signals normalized to chromatin signal (H2B) were higher in the light (Fig. 1C, D), showing, in vivo, that RNA Pol II activity is induced ~ 2-fold during photomorphogenesis independently of DNA content.

We then asked whether light-triggered enrichment of active RNA Pol II is associated with differences in RNA Pol II kinetics. We exploited At-RS31 pre-mRNA accumulation in proximal (P) vs distal (D) regions, with the P/D ratio serving as a proxy for RNA Pol II processivity^32,33. RT-qPCR analysis of At-RS31 pre-mRNA detected a higher P/D ratio in dark-grown than in light-grown cotyledons (Fig. 1E), indicating a lower processivity of RNA Pol II in the dark at this locus.

Finally, to gain a genome-wide view of RNA Pol II activity during cotyledon photomorphogenesis, we profiled H2Bub and S2P RNA Pol II in dissected cotyledons by ChIP-seq. In line with a general hypotranscriptional status of skotomorphogenic cotyledon nuclei, both S2P RNA Pol II and H2Bub chromatin levels at gene bodies were lower in dark than in light-grown cotyledons (Fig. 1F and Supplementary Fig. S2A). Collectively, our cytogenetic and epigenomic analyses showed that cotyledon photomorphogenesis involves a general increase of RNA Pol II content and activity over the genome.

Light-mediated transcription augmentation is undetected using standard RNA-seq

Having established that RNA Pol II activity increases during cotyledon photomorphogenesis, we questioned to what extent such genome-scale control would impact transcriptomic patterns. RNA-seq analysis of 200 cotyledons dissected from dark- or light-grown seedlings showed no tendency for a global increase in mean transcript levels or gene expression fold changes (Log₂ FC distributed around 0, i.e., no difference), and only a minor bias in the number of down- or up-regulated genes (43 vs 57%, respectively, with an FDR < 0.01 and a |Log₂ FC | > 0.5) (Fig. 2A and Supplementary Data 1, 2). Hence, as in previous studies^10,11, a standard RNA-seq methodology detected no global difference that would reflect a 2-fold intensification of transcription. To assess this discrepancy at the gene level, we compared S2P-RNA Pol II and H2Bub enrichments at differentially expressed genes (FDR < 0.01 for a |Log₂ FC | > 0.5) with those at genes with similar transcript levels between dark and light conditions (FDR < 0.01 for a |Log₂ FC | ≤ 0.5). While light-induced genes displayed higher S2P-RNA Pol II and H2Bub mean enrichment in light than in darkness, down-regulated genes showed very poor or no change in these two transcription hallmarks (Fig. 2B and Supplementary Fig. S2B). This suggested that considering this gene set downregulated may be artifactual. Conversely, the set of genes with similar transcript levels between dark and light was enriched for S2P-RNA Pol II and H2Bub in the light compared to darkness (Fig. 2C), suggesting that changes in transcript levels were underestimated. Hence, discrepancies between S2P-RNA Pol II and H2Bub profiles and standard RNA-seq analysis suggested that effects of transcription intensification were largely missed, possibly underestimating the impact of light on transcript levels during cotyledon photomorphogenesis.

Fig. 2. Transcription intensification is not detected using standard RNA-seq.

A Distribution of transcript levels, expression fold changes (Log₂), and numbers of differentially expressed (DE) genes (q-value < 0.01 for a |Log2 FC | > 0.5) using a standard RNA-seq analysis of 5 biological replicates comparing cotyledons dissected from dark- or light-grown seedlings. The p-value was calculated from a two-sided Welch’s t test. B RNA Pol II S2P and H2Bub chromatin enrichment at differentially expressed genes analyzed in A, genes displaying similar transcript levels in dark and light (Similar: q-value < 0.01 for |Log2 FC | ≤ 0.5), and genes neither up nor downregulated (non-DEGs) in standard RNA-seq analysis of dissected cotyledons. C RNA Pol II S2P and H2Bub meta-profiles over all the genes displaying similar transcript levels in dark and light as defined in (B). The p-values are calculated from a two-sided Wilcoxon Rank Sum test. Three independent biological replicates were analyzed for RNA Pol II S2P, and two for H2Bub. Source data are provided as a Source Data file. D Experimental setup for introducing Drosophila S2 cells as an exogenous spike-in reference (RNA-Rx). As a proof-of-concept experiment, 5000 S2 cells were mixed with either 100 or 200 dissected cotyledons prior to RNA extraction. Created in BioRender. Richet-Bourbousse, C. (2025) https://BioRender.com/rrgd2r4E RT-qPCR analysis of mRNAs extracted from 100 or 200 dissected cotyledons, both including the same number of Drosophila cells as a spike-in. The ratio of RNA levels between the two sample types was determined either by normalizing to Arabidopsis reference genes (AT3G02065, PP2A, and AT2G41020) or to Drosophila genes (RpL21, cg25c, vkg, B4, and pum). Error bars indicate the standard deviation between two technical replicates. F Distribution of expression changes (Log₂ FC). RNA sequencing reads were analyzed by normalizing solely to library size (Standard) or using the reads mapping to Drosophila genes (Spike-in normalization). The analysis uses four independent biological replicates. Source data are provided as a Source Data file.

This seeming contradiction most likely resulted from the well-described inability of standard RNA-seq analyses to inform on absolute variations of transcript levels^20,22,23,34. Indeed, because count data generated by sequencing technologies are compositional, a strong limitation of RNA-seq standard normalization procedures is that they estimate changes in transcript levels relative to one another but not absolute changes between samples²⁵. To circumvent this issue and obtain absolute quantifications, we used an exogenous reference. Instead of spiking exogenous RNA before library preparation, which cannot inform on absolute levels in the starting material because of RNA extraction and processing biases (Supplementary Fig. S3A), we spiked in a fixed number of D. melanogaster S2 cells directly in cotyledon lysates and extracted RNA from A. thaliana and D. melanogaster together. As a proof of concept, we compared transcript levels in pools of 100 and 200 cotyledons dissected from light-grown seedlings, meant to display a doubling of all transcript levels when normalizing to the spike-in reference (Fig. 2D). While standard RT-qPCR normalization using housekeeping genes expectedly gave similar relative transcript levels in both sample types, using Drosophila transcripts as reference allowed the detection of twice more A. thaliana transcripts in the “200 cot” than in the “100 cot” samples for all tested genes (Fig. 2E). Extending this principle to RNA-seq expectedly enabled the detection of a two-fold shift in the relative proportion of Arabidopsis vs Drosophila reads in all four biological replicates comparing 100 and 200 cotyledons (Supplementary Fig. S3B). To robustly normalize RNA-seq data to the spike-in, we implemented an ad hoc analysis pipeline, hereafter called RNA-Rx (RNA-seq with reference exogenous transcripts), that mitigates the technical variability induced by the spike-in (See the methods section and https://github.com/vidal-adrien/RNA-Rx-Pipeline-Pub³⁵) while reducing data dispersion to a level similar to standard RNA-seq procedures (Supplementary Fig. S3C). This confirmed a 2-fold change using the RNA-Rx pipeline, whereas no global change was detected using a standard approach (Fig. 2F and Supplementary Fig. S3D). These observations demonstrate that RNA-Rx enables absolute comparisons of transcript levels between samples.

Transcriptome size increases during cotyledon photomorphogenesis

A prerequisite for inter-sample comparisons with RNA-Rx is the use of samples with similar cell or nuclear counts. This is roughly the case of cotyledons from dark- and light-grown seedlings³⁶. Yet, in light-grown seedlings, about 20% of the cotyledon cells are subjected to one more endoreduplication cycle, producing extra genome copies compared to dark-grown cotyledons (average ploidy level of 3.5 vs 4.3; Supplementary Fig. S4B). Conversely, cotyledons display neither cell division nor ploidy changes during the first 24 h of de-etiolation and therefore are ideally suited for an RNA-Rx approach (Supplementary Fig. S4A, B, López-Juez et al. 2008¹¹). We ensured that RNA Pol II processivity increases after 1, 6, and 24 h of light by determining pre-mRNA proximal vs distal sites at At-RS31. Light exposure triggered a fast and strong decrease in the P/D ratio, reaching 2-fold after 1 h (Supplementary Fig. S4C). Consequently, in subsequent analyses, dark-grown seedlings were exposed to continuous light for 1 to 24 h before cotyledon dissection (Fig. 3A).

Fig. 3. Light induces an increase in transcriptome size in cotyledon cells during de-etiolation.

A The analyses employed cotyledons dissected from 5-day-old seedlings grown in darkness (Dark) and exposed to light for the indicated duration (0, 1, 6, and 24 h). Five independent biological replicates were analyzed for each time point. B Transcriptome size of the five biological replicates was estimated from the size factor of DESeq2 differential analyses of Standard RNA-seq and Spike-in RNA-Rx, divided by library size. For comparison, the median values of the Dark samples were arbitrarily set to 1. One dot represents the value for one biological replicate at each time point. C Distribution of mean transcript levels and Log₂ FC of expression. The Dark time point was used as a reference for each Standard RNA-seq and Spike-in RNA-Rx analysis. Source data are provided as a Source Data file. D Number of differentially expressed genes (DEGs, q-value < 0.01 for a |Log2 FC | > 0.5) in standard RNA-seq and Spike-in RNA-Rx analyses. The Venn diagrams show the number of DEGs commonly called by both methods (see Supplementary Data 3). E Log₂ FC of expression at the three time points of light exposure compared to the Dark time point for all genes analyzed in D. Genes are ranked according to their Log₂ FC at the 24 h time point. F Heatmap representing the Log₂ FC of expression at the indicated time points compared to the Dark time point for genes ordinarily used as RT-qPCR references or considered as housekeeping. Genes are ranked by their Log₂ FC at the 24 h time point. The sidebar indicates which gene sets were called DEGs in Standard RNA-seq or Spike-in RNA-Rx analyses, using the color code shown in D.

We used this setup for an RNA-Rx time series analysis of cotyledon photomorphogenesis. An Arabidopsis cotyledon being made of ca. 4000 cells³⁷, the number of cells in each sample was estimated at ~ 800,000 irrespective of the light condition, to which we added 5000 Drosophila S2 cells (0.625%) per sample in five independent biological replicates (Supplementary Fig. S5A). Dual mapping of the sequencing reads to the Drosophila and Arabidopsis reference genomes first showed that the percentage of Drosophila vs Arabidopsis reads decreased with light exposure from an average of 2.26% at time 0 (T0 Dark) to 0.81 after 24 h (Supplementary Fig. S5B), indicating a progressive augmentation of Arabidopsis transcriptome sizes during de-etiolation (Fig. 3B). The distribution of transcript levels and gene expression fold changes were expectedly stable in response to light using standard normalization, but increased over time with spike-in normalization, reaching a 2-fold rise after 6 h (Fig. 3C, Supplementary Fig. S5C, F and Supplementary Data 1-2). Both observations support an average doubling of transcriptome size in cotyledon cells during photomorphogenesis.

As anticipated, transcript level variations were more dispersed for cotyledon photomorphogenesis than when comparing 100 vs 200 cotyledons, reflecting the extensive gene reprogramming events in addition to transcriptome size doubling (Fig. 3C and Supplementary Fig. S5C). Compared to standard RNA-seq, RNA-Rx identified almost twice as many differentially expressed genes at all time points, affecting 46% of the total number of Arabidopsis genes (11,104 among the 23,898 genes analyzed) (Fig. 3D). While standard RNA-seq detected comparable numbers of down- and upregulated genes as in former studies^9–11,13, RNA-Rx identified a considerable bias toward gene upregulation with 8500 genes upregulated and only 648 genes downregulated after 24 h (Fig. 3D and Supplementary Fig. S5C and Supplementary Data 3). Accordingly, about 6000 genes were found to be upregulated exclusively upon spike-in normalization. Conversely, a large majority of the genes identified as downregulated by standard RNA-seq were not differentially expressed in the RNA-Rx analysis (Fig. 3D, E, Supplementary Fig. S5E, and Supplementary Data 3). Consistent with our observation of greater RNA Pol II levels in individual nuclei (Fig. 1), RNA-Rx identified genes encoding RNA Pol II subunits as upregulated (Supplementary Fig. S5D). RNA-Rx also identified many genes commonly used as references in expression studies, such as ACT2, UBQ10, PP2A, and GAPDH, as upregulated (Fig. 3F), indicating that housekeeping genes are part of the transcriptome augmentation process. Overall, RNA-seq with spike-in followed by ad hoc normalization revealed that cotyledon de-etiolation involves a wide increase in transcript levels that was formerly masked when using standard methodologies, and that hundreds of genes previously identified as downregulated are actually stable.

Reinterpreting the gene regulatory paths of cotyledon photomorphogenesis

Given that switching from standard to spike-in normalization may affect current views of light-driven gene regulation, we examined the expression of genes targeted by the major transcription factors mediating light signaling (Supplementary Data 4). Compared to standard RNA-seq, a much higher fraction of the total gene sets bound in the light by HY5, GLK1/2, FHY1, BZR1, TCP4, or GI were induced during cotyledon de-etiolation (Fig. 4A). Focusing the analysis of target genes on those that are differentially expressed, RNA-Rx identified fewer downregulated genes in response to light compared to standard RNA-seq. For example, RNA-Rx identified that 93% (185 of 198 DE genes) of the HY5 target genes are upregulated, rather than 66% (90 of 135 DE genes). The most notable example is GLK1/2, whose differentially expressed target genes were 98.4% upregulated, rather than 93% after 24 h (Fig. 4B). Hence, supporting previous functional analyses^13,38,39, RNA-Rx identified a quasi-exclusive propensity of GLK1/2 and HY5 for gene activation. For dark-activated TFs such as PIF3, FHY3, ARF6, and BZR1, RNA-Rx also detected more light-induced genes and fewer light-repressed genes than standard RNA-seq. In the case of PIF3 target genes, RNA-Rx detected 240 instead of 83 upregulated genes and 145 instead of 265 downregulated genes after 24 h of light (Fig. 4A, B), in agreement with PIF3 acting both as a transcriptional activator and repressor in etiolated cotyledons^40–42.

Fig. 4. RNA-Rx enables refining gene regulatory pathways during cotyledon de-etiolation.

A Percentage of genes up- or downregulated after 24 h of light (q-value < 0.01 for a |Log2 FC | > 0.5) at target genes of major transcription factors involved in light signaling. All analyzed TF target genes were previously identified, as detailed in Supplementary Data 4. Source data are provided as a Source Data file. B Expression Log2 FC of genes bound by GLK1/2, HY5, FHY1, and PIF3 transcription factors that are also differentially expressed upon 24 h of light exposure using Standard or Spike-in RNA-Rx analyses, respectively. Genes are ranked by their Log2 FC at 24 h. C DREM analysis of coregulated gene paths using expression Log₂ FC from Standard RNA-seq and Spike-in RNA-Rx. Seven paths were found in both cases, called A to G (Standard) or A’ to G’ (Spike-in), that largely overlap. The inlays display the percentage of DEGs in paths A to G at the 24 h time point and the percentage of genes shared between paths A to G and A’ to G’. D Metaplot profiles of RNA Pol II S2P and H2Bub enrichment in dark and light-grown cotyledons at the genes corresponding to the DREM paths A’ to G’. E Distribution of mRNA half-lives in the DREM paths A’ to G’ according to Sorenson et al. (2018), Chantarachot et al. (2020), and Szabo et al. (2020)^44–46. The distributions over all TAIR10 genes are shown as reference in gray and compared to each DREM path with a Welch’s t test (* < 0.1, ** < 0.01 and ***< 0.001). Source data are provided as a Source Data file. F Over-represented gene ontology (GO) categories within the DREM path A’ to G’. Numbers correspond to the genes falling in each GO category. G Variations of expression Log₂ FC upon Standard RNA-seq or Spike-in RNA-Rx analysis of genes encoding proteins of the photosynthetic chain, ribosomes, and replication machinery. Source data are provided as a Source Data file.

We then compared the regulatory networks inferred from RNA-Rx vs. standard RNA-seq using Dynamic Regulatory Events Miner (DREM)⁴³. Employing identical gene sets and parameters, the two analyses identified seven modules of expression with similar gene sets but different patterns (Fig. 4C, Supplementary Fig. S6A and Supplementary Data 5). Genes that were highly induced in response to light displayed an upregulation twice as strong using RNA-Rx (paths A’-C’) than standard RNA-seq (paths A-C) data (Supplementary Fig. S6A). Path D and D’ mainly consist of genes upregulated in RNA-Rx but not in standard RNA-seq data (75%). Path E and E’ are mainly composed of genes downregulated in standard RNA-seq (68%) but not in RNA-Rx. Finally, paths F-G and F’-G’ correspond to genes downregulated in both analyses but with a milder intensity in RNA-Rx datasets. Accordingly, genes in the induced paths A’ to C’ are significantly enriched in target genes of GLKs and HY5, while genes in repressed paths F’ and G’ are significantly enriched in target genes of PIF3, MYC2, EIN3, ARF6, and BES1. The target genes of FHY1, TCP4, GI, and BZR1 are found enriched in both induced and repressed paths. A small proportion of genes in the stable path E’ were found to be targeted by PIF3 (150/2021; 7.4%) and MYC2 (104/2021; 5.1%), and no enrichment in the major TFs mediating light signaling was found in the large induced gene set of path D’ (Supplementary Fig. S6B, C).

To test the validity of the gene expression paths defined using RNA-Rx, we probed S2P-RNA Pol II and H2Bub chromatin enrichment at genes composing the seven DREM modules. Paths A’ to C’ displayed stronger enrichments of these transcription hallmarks in light than in dark. This was also true when comparing path D’, in line with the idea that a majority of genes are upregulated instead of being stably expressed (Fig. 4D and Supplementary Fig. S7A). Interestingly, genes in path E’ displayed a mild increase in H2Bub and S2P-RNA Pol II mean levels in the light, indicating that epigenomic data imperfectly match both RNA-seq and RNA-Rx analyses for this intermediate path. In contrast, genes in paths F’ and G’ displayed low markings in the light, especially S2P, in line with their downregulation upon the transition.

One of the striking findings using RNA-Rx was a massive correction for less downregulation than previously assumed. Given that transcriptional shutdown and mRNA decay can both contribute to transcript depletion, we investigated whether transcripts from downregulated genes were particularly unstable. Such a property could explain a loss of transcripts per cell in the absence of division. The transcript half-life estimates from recent studies^44–46 were used to determine the intrinsic mRNA stability of genes that compose the seven DREM paths. Transcripts of the downregulated paths F’-G’ genes displayed a tendency for short half-lives (Fig. 4E). Based on studies employing cordycepin (3’-deoxyadenosine) to block mRNA synthesis (Sorenson et al., 2018⁴³ and Chantarachot et al., 2020⁴⁵), the paths A’ and B’ corresponding to highly upregulated genes also tend to encode short-lived transcripts. In contrast, with thousands of moderately upregulated genes (~ 8700), paths C’ and D’ are enriched in genes that encode transcripts with a long half-life (Fig. 4E).

Given these observations, we revisited the RNA-Rx-defined co-expression modules to gain an updated understanding of light-regulated biological functions. Upon Gene Ontology (GO) analysis, path D’ with discordant trends between standard RNA-seq and RNA-Rx was significantly enriched in genes involved in replication and translation, including most ribosomal protein genes as well as RNA Pol I and RNA Pol III subunits (Fig. 4F, G and Supplementary Fig. S7B). Hence, contrary to standard RNA-seq, RNA-Rx identified upregulation of replication and translation functions, supporting the known positive effects of light on translation^11,47–51 and the onset of endoreduplication 24–48 h after exposure to light¹¹. Taken together, these results demonstrate that regulatory networks defined using RNA-Rx better support epigenomic profiling data and are more coherent in terms of biology than standard RNA-seq.

Identification of stable genes to quantify transcriptome size in standard RNA-seq data

Considering that defining an internal reference would open the way to the absolute quantification of transcriptomic variation in RNA-seq datasets lacking a spike-in and may further reduce technical variability, we sought to look for genes whose expression patterns mimic the spike-in information. We identified genes with moderate to high transcript levels and a high signal-to-noise ratio, and ranked them based on likelihood ratios, estimating their ability to recapitulate the spike-in information (Fig. 5A; see “Methods”). As detailed in Supplementary Fig. S8A, the top 32 genes were selected as our candidate set of stable genes (Fig. 5B, C). These genes encode proteins with basic molecular and cellular functions, such as organelle and vesicle trafficking, transcription and RNA metabolism, splicing, ribosome biogenesis, and vacuolar H⁺ ATPase subunits (Fig. 5B). Interestingly, 28 of them are poorly translated in response to darkness or hypoxia^48,50,52 (Supplementary Table 1) indicating that these genes, stably expressed during cotyledon de-etiolation, may predominantly be regulated by light at the translational level.

Fig. 5. Genes with stable absolute expression recapitulate spike-in information in independent studies of cotyledon de-etiolation.

A The pipeline used to identify reference genes with stable absolute expression levels during cotyledon de-etiolation led to the selection of 32 genes (see “Methods” for details). TPM values from the five biological replicates at the four time points (Dark, 1 h, 6 h and 24 h) were used. Created in BioRender. Richet-Bourbousse, C. (2025) https://BioRender.com/rj05ueh. B Identity of the selected stable genes. C MA-plot analyses after Standard RNA-seq or Spike-in RNA-Rx, highlighting the 32 stable genes (orange). D Differential gene expression in our study and in 3 published datasets^8,9,13 using the set of 32 stable genes as an internal reference for normalization. In each analysis, the dark time point was used as a reference. E Number of differentially expressed genes (q-value < 0.01 for a |Log2 FC | > 0.5) using the 32 reference genes as an internal control for normalization of our RNA-seq data and of the same published datasets as in (D).

We tested the capacity of the stable gene set to provide information equivalent to that of the exogenous spike-in. Renormalization of the de-etiolation RNA-seq time series on this gene set as reference resulted in differential expression changes highly correlated to those obtained with the Drosophila spike-in as reference (R² ≥ 0.89 at all time points; Supplementary Fig. S8B) and a comparable doubling of transcriptome size at 24 hours (Fig. 5D, E and Supplementary Fig. S8C). Consequently, we tested whether this approach could be used to faithfully detect transcriptome size changes in RNA-seq data generated without a spike-in. We remapped and normalized three publicly available transcriptomic datasets of cotyledon de-etiolation (Supplementary Fig. S9A; Han et al. 2023⁸; Sun et al. 2016⁹; Burko et al. 2020¹³) using the set of 32 genes as an internal reference. Remarkably, the distribution of transcript levels and differential expression changes consistently detected a similar augmentation of transcriptome size in the four independent laboratory setups (Fig. 5D and Supplementary Fig. S9B–D). In our RNA-seq time series, transcriptome size variations were nonetheless consistently lower at the 1-hour and 6-hour time points using the stable gene set than the Drosophila spike-in (Supplementary Fig. S8B, C), indicating that normalization with stable genes may slightly underestimate global variations. Accordingly, the number of upregulated genes was lower in the four renormalized datasets than in those using our spike-in normalization (Fig. 5E and Supplementary Fig. S9D), especially in the study by Burko et al., which had fewer replicates and smaller library sizes. Yet, with an average trend of 84% upregulated versus 16% downregulated genes at 6 hours, normalization using the set of 32 stable genes reproducibly detected clear trends toward gene upregulation in all four studies (Fig. 5E and Supplementary Data 6).

Lastly, we asked whether the stable genes defined by spike-in could serve as an endogenous reference for absolute quantification of gene expression using RT-qPCR. Accordingly, RT-qPCR normalization using transcripts of three stable genes resulted in increased transcript levels for all tested genes (CAB2, EXPA1, ACT2, ACT7, RS31, AGL31, PIL1, and XTR7) at both the 6 h and 24 h time points compared to using housekeeping genes (Supplementary Fig. S10). This further revealed a clear but moderate induction by light of the classically used housekeeping genes ACTIN2 and ACTIN7, consistent with their association with path D’. This also shows a stable expression of RS31 and AGL31, two genes of path E’ that were called as being downregulated in standard RNA-seq but stable in RNA-Rx. Our data renormalization method thus enabled the reliable detection of a transcriptome size increase that had previously been missed in independent studies on cotyledon de-etiolation.

Discussion

Arabidopsis photomorphogenesis is a developmental transition that profoundly reshapes the transcriptional program^1–3,15,53. In previous studies, we proposed that the nuclei of skotomorphogenic cotyledons exhibit a quiescent transcriptional state^17,19. Here, we revealed that cotyledon photomorphogenesis involves a rapid increase in the transcriptional regime, characterized by a quasi-exclusive trend of gene upregulation and transcriptome expansion. The existence of hyper- and hypotranscriptional states in plants has long been overlooked, and the capacity of eukaryotic cells to adjust their transcriptional regime to developmental or external signals is just starting to emerge^24,54. The biological significance and underlying mechanisms of hypertranscription are best understood in Embryonic Stem Cells (ESCs), tumors, and multiple progenitor cell lineages, which all overexpress c-Myc^{20,23,55–57}. Vice versa, hypotranscription has been linked to specific metabolic states, such as during yeast cell quiescence^58–60 and mouse blastocyst diapause^54,61. In plants, hypotranscription has been suggested in spore mother cells⁶², the egg cell and the resulting embryo⁶³, and in dry seeds^64,65. Global variations in RNA Pol II activity were also found to follow changes in genome content during endoreduplication or polyploidization, as well as during the cell cycle, with tissue-specific effects^26–30. Our study further unveiled that environmentally induced plant cellular transitions can involve changes in the transcriptional regime at the genome scale and additionally decrypted their impact on transcriptome dynamics.

Determining the impact of higher-order transcriptional control on individual gene expression and transcriptome size requires absolute normalization approaches, a bottleneck we addressed using an exogenous reference. Previous studies assessing the effect of plant polyploidization on transcriptome size have used spike-in strategies for normalization per biomass, cell, or genome content^24,28–30. Cell number-controlled RT-qPCR enabled absolute quantification of diurnal variations in the transcript levels of Arabidopsis central clock genes⁶⁶. A ‘per-leaf normalization’ 3’-RNA-seq analysis further unveiled that many Sorghum genes display higher average transcript levels in the evening than in the morning⁶⁷. Interestingly, these loci include genes involved in transcription and translation. This suggests global regulation of transcript production during the photoperiod, which aligns with our observations of an increase in transcriptome size during photomorphogenesis. Given that cotyledon de-etiolation initiates independently of mitosis and endoreduplication, we were able to reach an absolute normalization averaged per cell, which was essential for detecting and quantifying a global increase in mRNA cellular content during cotyledon photomorphogenesis. The doubling in transcriptome size identified during the first six hours cannot be explained by endoreduplication, which is negligible over this short time period. It may primarily result from the 2-fold increase in nuclear amounts of active RNA Pol II. This likely initiates before the onset of endocycles to sustain mRNA and protein production upon full adaptation to light. Light sensing indeed strongly enhances global translation in plant cells⁵³, a function evidenced in our RNA-Rx data by the upregulation of genes involved in protein synthesis. Studies employing exogenous spike-in RNA as a normalization reference to quantify polysome-bound mRNAs reported an overall increase in translation four hours after induction of Arabidopsis photomorphogenesis^47,48,50.

Translation and transcription are both energetically costly⁶⁸. Hence, attenuation of these two basic cellular activities plausibly constitutes an adaptation to the low metabolic status upon prolonged growth in darkness. Etiolated development relies on limited seed resources that, in the absence of photosynthesis, can only last a few days⁴. Yet, the control of translation is surprisingly independent of photosynthesis but relies on an auxin-TOR signaling pathway acting downstream of COP1⁵⁰. Translation inhibition partially relies on mRNA sequestration in cytosolic p-bodies that are particularly abundant in darkness⁴⁸. Interestingly, most of the stably expressed genes identified here using an RNA-Rx approach correspond to genes regulated at the translational rather than transcriptional level. Considering the hypotranscriptional state of etiolated cotyledons, we hypothesize that long-lived transcripts stored within p-bodies are immediately available for translation to enable the seedling to overcome the dark-to-light transition before reaching full transcription competency. We identified that transcripts of the strongly induced gene paths (DREM paths A’ and B’) tend to be short-lived, while transcripts of the thousands of moderately upregulated genes (~ 6545) from path D’ tend to display a long half-life. This observation aligns with studies in mammals showing that transcripts mediating constitutive cellular processes, such as housekeeping genes, tend to be more stable than those of transcription factors and signaling components⁶⁹. A slight increase in transcription rates of transcripts with a long life may therefore also contribute substantially to transcriptome size doubling during cotyledon de-etiolation.

Translation regulation may also impact the transcriptional regime, for example, by sustaining the production of unstable chromatin modifiers, as achieved by the TOR pathway in mouse embryonic stem cells⁷⁰. In etiolated cotyledons, COP1/auxin/TOR-mediated translation control may similarly attenuate the transcriptional regime through epigenome adaptations. Considering that DET1 regulates the stability of a histone H2Bub deubiquitination module, targeted degradation of chromatin modifiers likely contributes to adjusting the epigenome landscape to a low transcriptional status in darkness¹⁹. Chromatin regulatory processes are at the nexus of many light-driven signaling pathways, and their possible implication in energy-saving mechanisms remains to be tested.

In this study, the absolute quantification of transcript levels led to a drastic revision of former views on the activity of transcription factors at thousands of light-regulated genes. This further sheds light on higher-order regulatory mechanisms at the whole transcriptome level. Disentangling the contribution of genome-scale transcription intensification from sequence-specific control at individual genes will therefore be essential for decrypting the molecular mechanisms that modulate transcriptome size augmentation. A second layer of complexity is the existence of cell-to-cell variability in gene transcriptional responses to environmental changes, as identified through in situ imaging of transcription reporter transgenes^71–73 and RNA FISH⁷⁴. These case studies identified that organ-level transcriptional regulation combines two patterns: an ‘analogic’ mode, involving gradual adjustments of gene transcription across all cells, and a ‘digital’ mode in which genes are switched on and off in distinct cells. Accordingly, single-cell and single-nucleus RNA-seq^8,75,76, combined with ad hoc absolute normalization methods, may allow for disentangling the contributions of analog and digital regulation modes to transcriptome size changes.

Methods

Plant material

All seed lots were Arabidopsis thaliana Col-0 wild-type plants, except for mintbody microscopy, which used a pRPS5a:Ser2P-mintbodyIntF2A-H2B-mRuby line³¹ and a CYCB1;1::DB-GUS reporter line⁷⁷.

Growth conditions

Seeds were surface-sterilized and sown on half-strength Murashige and Skoog media (Duchefa Biochemie, The Netherlands) with 1% (w/v) Phytoagar, pH 5.7. After 3 days of stratification at 4 °C in darkness, seeds were exposed to light for 4–6 h to synchronize germination. Plants were either grown under light conditions for 5 days (with a 16 h-light 8 h-dark photoperiod) or under three layers of aluminum foil for 5 days before being harvested (Light and Dark samples, respectively) or exposed to 1 h, 6 h, or 24 h of continuous light (80 µmol.m⁻².s⁻¹ irradiation from cool white fluorescent tubes (Philips F25T8/TL841 (Hg)) in a Percival CU36L5 growth chamber. Dark samples were collected under a dim green light.

RNA extraction and library preparation

The light-exposed or dark-grown samples were collected in 80% acetone at − 20 °C before being vacuum-infiltrated twice for 5 min under white or green light, respectively. Unless stated otherwise, 200 cotyledons were dissected in 100% ethanol under a dissection scope and frozen in liquid nitrogen before grinding with a Tissue Lyser (Qiagen). RNA was extracted with the microRNeasy kit (Qiagen). Five million Drosophila melanogaster S2 cells were resuspended in 7 mL of RLT buffer containing ß-mercaptoethanol. For this, cotyledons were ground in liquid nitrogen and resuspended in 700 µL of the RLT buffer containing the S2 cell lysate by vortexing and repeated resuspension through a 20-gauge (0.9 mm) needle attached to a sterile plastic syringe. Combined purification of A. thaliana and D. melanogaster RNA was then conducted according to the manufacturer’s instructions. Sequencing libraries were prepared using the Illumina TruSeq Stranded mRNA kit with polyA selection. Paired-end sequencing was performed on a DNBSEQ-G400 platform by BGI (Hong Kong) with a read length of 100 base pairs. Transcriptomic analyses comparing 100 vs 200 cotyledons in Fig. 2 employed four independent biological replicates, while transcriptomic analyses of cotyledon photomorphogenesis employed five independent biological replicates for each time point.

Read mapping

After trimming, reads were mapped using Salmon⁷⁸ (version 1.9.0) to a merged reference transcriptome composed of the AtRTD3⁷⁹ Arabidopsis and the dmel_r6.48 Drosophila transcriptomes, using the Arabidopsis (TAIR10) and Drosophila (dmel_r6.48) genomes as a decoy sequence. Salmon options to correct for sequence and GC bias were used, and 100 Gibbs inferential replicates of the count assignment were sampled with a thinning factor of 100. The read counts were obtained from the median of those inferential replicates via the tximport⁸⁰ R package (version 1.30.0, R version 4.3.2) with the ‘scaledTPM’ option. For the remaining processing steps, only transcripts annotated as mRNAs in AtRTD3⁷⁹ and having at least 1 read count in at least 1 sample were considered.

Normalization and differential analysis for Standard RNA-seq

Differential analysis was performed using DESeq2³⁴ (version 1.42.0, R version 4.3.2) with read counts for Arabidopsis genes as input. The DESeq2 design matrix included an additive ‘batch’ effect, thus taking into consideration the fact that five biological replicates were obtained in two series. Genes were considered differentially expressed at a q-value threshold of 0.01 for an absolute Log₂ fold change greater (or lesser for stably expressed genes) than 0.5 by setting the ‘lfcThreshold’ option in DESeq2 ‘results’ function to 0.5. Normalized transcript levels were obtained from the DESeq2 normalized read counts divided by gene length and scaled to arbitrarily set the dark sample’s median to 1 for each analysis. For scatter plot comparisons of datasets, the Log₂ fold change determined using DESeq2 was shrunk with the ashr⁸¹ package (version 2.2.63, R version 4.3.2) using a mixture of Normal prior, no point mass at 0 and the ‘estimate’ option for the prior mode.

Normalization and differential analysis for Spike-in RNA-Rx

To mitigate read count dispersion due to spike-in variability (Supplementary Fig. 3C), we used the CATE⁸² package (version 1.1.1, R version 4.3.2). First, each sample was normalized using the DESeq2 median of ratios computed from Drosophila gene read counts. The Log2 of the obtained Arabidopsis genes’ normalized read counts using a 0.5 offset and the same design matrix used for Standard RNA-seq (including a ‘batch’ effect) was used as input for CATE to adjust for the principal latent confounder (‘r’ set to 1 in ‘cate’ function). The ‘cate’ function was run in ‘negative controls’ mode using Drosophila genes with at least five reads in all samples. The obtained matrix of corrected counts was used for DESeq2 differential analysis, with the size factor calculated from the median of ratios of corrected counts for Drosophila genes with the previously described design matrix that included an additive ‘batch’ effect. Differentially expressed genes, normalized transcript levels, and scatter plots were retrieved in the same way as for Standard RNA-seq.

DREM analysis

Log₂ fold change of expression from Standard RNA-seq and Spike-in RNA-Rx differential analyses were used as inputs, using the same DREM parameters as in Bourbousse et al. 2018⁸³.

Selection of stably expressed genes

Plastid, mitochondrial, AtRTD3-newly identified transcripts, and AtRTD3 annotations overlapping multiple genes were excluded. Only genes whose TPM median and signal-to-noise ratio minimum across samples were both greater than 10 were considered. TPM values were retrieved directly from Salmon⁷⁸ (version 1.9.0) using tximport⁸⁰. The signal-to-noise ratios were calculated as the median divided by the square root of the gene-count variance across the inferential replicates for each sample.

For each of the 1763 remaining genes, size factors were calculated for each sample (similar to DESeq2’s median of ratios method, i.e., as the ratio of the sample’s gene counts over the geometric mean of gene counts across all samples) using CATE corrected counts. We compared the size factors obtained for each of these 1763 genes with those obtained with the Drosophila spike-in genes using the coefficients and their standard error derived from the linear regression of log-transformed size factors on the sample’s design matrix. Specifically, for each candidate gene, we calculated (1) the ‘absolute deviation’ as the maximum of the absolute difference between the candidate gene’s coefficients and the Drosophila spike-in coefficients, (2) the ‘variability’ as the maximum of the candidate gene’s coefficient’s standard deviations and (3) the ‘likelihood ratio’ as the minimum of the ratio of the spike-in model coefficients’ likelihoods in the gene model over the spike-in model. For those three measures, the coefficient corresponding to the intercept was ignored. Genes were ranked using the likelihood ratio as a ranking score, employing a method similar to the stability ranking procedure described in Siebourg et al. (2012)⁸⁴ to increase robustness. 100,000 rankings of the candidate genes were computed from count matrices built by sampling gene counts independently from Salmon inferential replicates of each sample, and the final ranking was obtained with a probability threshold⁸⁴ π = 0.95. The counts of each inferential replicate were also corrected separately using the CATE package as described above.

According to this final ranking, the ‘absolute deviation’, ‘variability’, and ‘likelihood ratio’ were calculated this time using an increasing number of candidate genes. The ‘likelihood ratio’ is stable across the first 100 genes, yet we defined the final set of stable genes as the first 32 ranked genes, given the presence of a local maximum at this position (Supplementary Fig. S7A).

Normalization and differential analysis using stable genes as endogenous references

Normalization, differential analysis, and absolute transcript-level computation with an endogenous reference were conducted similarly to the Drosophila spike-in method described above, by replacing the Drosophila genes with the 32 stable genes and without implementing a ‘batch’ effect correction for the publicly available dataset.

ChIP-seq

Dark and light-grown seedlings were fixed in 1% formaldehyde under dim green or white ambient light, respectively. A fixed number of 400 cotyledons was dissected, frozen, and ground. For S2P RNA Pol II ChIP, a second crosslinking step was performed, and chromatin was extracted as in Bourbousse et al. 2018⁸³, sheared using a Covaris sonicator, and immunoprecipitated using the Abcam Ab5095 antibody. For H2Bub ChIP, a chromatin spike-in was included, and immunoprecipitation was conducted as in Nassrallah et al. 2018¹⁹ using the Medimabs MM-0029-P antibody. Libraries were prepared using NEBNext® Ultra™ II DNA Library Prep Kit for Illumina® (NEB, E7645). Paired-end sequencing was performed on a DNBSEQ-G400 platform by BGI (Hong Kong). ChIP-seq bioinformatic analyses were conducted following the pipeline available on our GitHub resource: https://github.com/vidal-adrien/ChIP-Rx-Pipeline-Pub⁸⁵.

RT-qPCR

Reverse transcription was performed using the High-Capacity cDNA Reverse Transcription Kit (Applied Biosystem) using random hexamers, followed by quantitative PCR using the SYBR Green I Master Mix (Roche) with the oligonucleotides listed in Supplementary Table 2. PCR analyses of AtRS31 pre-mRNA were conducted as described in Herz et al. 2019³³. Absence of DNA contamination was checked for all RNA samples by amplification without the RT enzyme.

Immunocytometry of RNA Pol II

Seedlings were fixed under safe green light in Galbraith buffer (20 mM MOPS, 45 mM MgCl2, 30 mM sodium citrate, 0.3% Triton X-100) containing 1% formaldehyde, under vacuum for 20 min, then released from vacuum and left in the fixative without vacuum for 20 min. The fixed seedlings were rinsed three times with water and transferred to Galbraith buffer containing 5 mM sodium metabisulfite (chopping buffer) on ice. A fixed number of 160 etiolated or 80 de-etiolated cotyledons were dissected in a small drop of chopping buffer and then chopped with a razor blade on glass Petri dishes. Upon sample collection in 1.8 mL of chopping buffer and filtration through a pre-wetted 30 μm nylon mesh filter, the nuclei were pelleted by centrifugation at 1500 × g for 5 min at 4 °C and resuspended in 100 μL of antibody buffer (30 mM MOPS, 22 mM MgCl2, 15 mM sodium citrate, 300 mM sorbitol, 0.5% BSA, 0.3% Triton X-100) containing the Abcam ab5095 and Abcam ab817 primary antibodies at a 1:500 dilution. Upon incubation overnight at 4 °C the nuclei were washed with 1.9 mL of antibody buffer (devoid of antibody), pelleted at 1500 × g for 5 min at 4 °C, and resuspended in 150 μL of antibody buffer containing Alexa488 coupled anti-mouse (Thermo Fisher A11001) and Alexa568 coupled anti-rabbit (Thermo Fisher A11011) secondary antibodies at 1:400 dilution. Upon incubation for 2 h at ambient temperature in the dark, the nuclei were washed with 1.85 mL of antibody buffer, pelleted at 1500 × g for 5 min at 4 °C, and resuspended in 200 μL of antibody buffer containing 5 μg/mL DAPI.

The immunolabeled nuclei were acquired with a Cytoflex S bench-top cytometer (Beckman-Colter) in a 96-well plate, driven by cytExpert 2.5. Data analysis was made with the Kaluza 2.1 software (Beckman-Colter). Nuclei were identified and classified by ploidy level using DAPI staining for gDNA. Doublet nuclei were excluded using the cytogram of DAPI-Area versus DAPI-Heights signals. The negative population for immunostaining was defined on the Alexa Fluor 488 and 568 background signals from a control sample lacking the primary antibodies. From this, a NOT gate was applied to all fully-labeled samples. Finally, a customized parameter was used to measure the fluorescence ratio of Alexa488 and Alexa568 for each nucleus from each ploidy level. For each ploidy level, gated data were extracted from Kaluza in CSV file format for further R analysis.

For statistical analysis, we employed a linear model on the log-transformed response variable (S2P/NP ratio, S2P signal, or NP signal) to account for heteroscedasticity and stabilize variance across treatments. The model was fitted using the following formula: ~ Condition * Ploidy + Experiment, where “Condition” refers to the light condition and “Experiment” to the experimental batch (one for replicates 1 and 2 and another one for replicate 3). Following model fitting, we used specific contrasts to test hypotheses using a t test with the “HC3” heteroscedasticity-consistent covariance matrix estimator⁸⁶ implemented manually using the variance-covariance matrix derived from the GLM.jl package (version 1.9.0). The p-value associated with the t-statistic of a given contrast was obtained from the cumulative distribution function of a Student's T distribution with the appropriate degrees of freedom using the Distributions.jl package (version 0.25.107).

Ploidy measurements

Cotyledons were chopped in Galbraith buffer with 0.3% Triton X-100. DAPI was added to a final concentration of 20 µg.mL⁻¹. Nuclei were counted in each sample according to their ploidy level, determined by fluorescence intensity using a Cytoflex bench-top cytometer (Beckman-Colter). For statistical analysis, the ploidy count data for each sample were first manually transformed using the additive log-ratio transform (ALR) with a Bayesian-multiplicative approach for count zero imputation, using a Bayes-Laplace prior⁸⁷. The p-values were obtained using an unequal variance Hotelling T2 test using the HypothesisTests.jl package (version 0.11.0).

Plot representation

All violin and box plots were drawn with ggplot2. The central line of the boxplots represents the median. The lower and upper hinges correspond to the first and third quartiles, and the whiskers extend to the largest and smallest values no further than 1.5 x the distance between the first and third quartiles. Outliers were not plotted.

GUS staining

CYCB1;1::DB-GUS seedlings were fixed in 80% cold acetone before incubation with 2 mM X-Gluc staining solution (50 mM NaPO4 at pH 7.2, 0.2% Triton X-100) at 37 °C. After several washes in 70% EtOH, seedlings were further discolored with chloral hydrate (2.5 g.mL⁻¹ in 30% glycerol).

Mintbody sample preparation, image acquisition, and 3D segmentation

Live seedlings were mounted in Immersol W 2010 (Carl Zeiss Microscopy, LLC, USA). Living nuclei from mesophyll cotyledon cells were imaged with a Leica TCS SP8-MP (Leica, Germany) equipped with a resonant scanner (8 kHz) with a 63x water objective (HC PL APO 63x/1.20 W motCORR CS2 1.2 NA). EGFP and mRuby were excited with 960 nm and 1040 nm lasers (Insight DS + Dual (680–1300 nm & 1041 nm) ultrafast NIR laser for multiphoton excitation), respectively. Fluorescence was detected using a non-descanned super-sensitive photon-counting hybrid detector (HyD), operated in photon-counting mode with 8x frame accumulation. Z-stacks were composed of 8-bit images acquired with a resolution of 568 × 568 pixels, and the voxel size was 0.0903 × 0.0903 × 0.0901 μm. All images were deconvolved with Huygens Professional version [23.04] (Scientific Volume Imaging, The Netherlands) and segmented with Imaris (bitplane, Switzerland) using the Surface creation tool in manual contouring mode for the nucleolus and ML Pixel Classifier mode for the nucleus. For each nucleus, channel signals (EGFP, mRuby, and their ratio) were exported and used to compute sums and means. The ratio between Ser2P-Mintbody-GFP and H2B-mRuby was calculated using the channel arithmetic function in Imaris.

Statistics & reproducibility

All biological replicates were independent of each other, i.e., corresponding to different seed lots growing in different petri dishes. Petri dishes were randomly positioned inside the growth chamber to minimize biases. No blinding was used as etiolated and green cotyledons are easily discriminated by the experimenter. No statistical method was used to predetermine sample size. The adequacy of the data with the statistical test’s assumptions was assessed, but the choice of the test was not conditioned on testing the assumptions. The statistical tests were done in R (version 4.3.2) or in Julia (version 1.10.0) using the GLM.jl (version 1.9.0), Distributions.jl (version 0.25.107), and HypothesisTests.jl (version 0.11.0) packages.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

C. Bo. and F.B. conceived and coordinated the study. RNA-seq experiments were performed by C. Bo. and L.W.; ChIP-seq experiments by C. Bo., E.A., and D.D.C.; RNA Pol II immunocytometry by M.L.H.D. with the contribution of M.B.; RNA Pol II imaging of nuclei and analysis by F.M.M. and C. Ba. G. S. developed the RNA-Rx bioinformatics pipeline and analyzed the RNA-seq data. C. Bo. analyzed the ChIP-seq data. A.V. developed the GitHub repository. C. Bo. prepared the figures. F.B. and C. Bo. wrote the manuscript. All authors had the opportunity to edit the manuscript and have approved it.

Peer review

Peer review information

Nature Communications thanks the anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Data availability

Raw sequencing data and processed files for RNA-seq and ChIP-seq have been uploaded to the NCBI GEO repository under the accessions GSE191042 and GSE275426. Published datasets used in this study (Han et al. 2023⁸; Sun et al. 2016⁹; Burko et al. 2020¹³) were retrieved from accessions: PRJCA016521, GSE79576, and GSE132861. Source data are provided in this paper.

Code availability

The code is available as R pipelines presented in a public GitHub repository (https://github.com/vidal-adrien/RNA-Rx-Pipeline-Pub³⁵).

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-025-66359-7.