Chromatin Accessibility and Gene Expression Vary Between a New and Evolved Autopolyploid of Arabidopsis arenosa

Abstract

Polyploids arise from whole-genome duplication (WGD) events, which have played important roles in genome evolution across eukaryotes. WGD can increase genome complexity, yield phenotypic novelty, and influence adaptation. Neo-polyploids have been reported to often show seemingly stochastic epigenetic and transcriptional changes, but this leaves open the question whether these changes persist in evolved polyploids. A powerful approach to address this is to compare diploids, neo-polyploids, and evolved polyploids of the same species. Arabidopsis arenosa is a species that allows us to do this—natural diploid and autotetraploid populations exist, while neo-tetraploids can be artificially generated. Here, we use ATAC-seq to assay local chromatin accessibility, and RNA-seq to study gene expression on matched leaf and petal samples from diploid, neo-tetraploid and evolved tetraploid A. arenosa. We found over 8,000 differentially accessible chromatin regions across all samples. These are largely tissue specific and show distinct trends across cytotypes, with roughly 70% arising upon WGD. Interestingly, only a small proportion is associated with expression changes in nearby genes. However, accessibility variation across cytotypes associates strongly with the number of nearby transposable elements. Relatively few genes were differentially expressed upon genome duplication, and ∼60% of these reverted to near-diploid levels in the evolved tetraploid, suggesting that most initial perturbations do not last. Our results provide new insights into how epigenomic and transcriptional mechanisms jointly respond to genome duplication and subsequent evolution of autopolyploids, and importantly, show that one cannot be directly predicted from the other.

Introduction

Whole-genome duplication (WGD), which results in polyploidy, has occurred throughout eukaryote evolution, increases genome complexity, and is implicated in evolutionary innovation, speciation, and adaptation (De Bodt et al. 2005; Soltis et al. 2008; Barker et al. 2009; Jiao et al. 2011). All angiosperms are estimated to have polyploid ancestry (Goldblatt 1979; Lewis 1979; Masterson 1994; Jiao et al. 2011), and roughly one-third of domesticated crops used in modern-day agriculture are polyploids (Salman-Minkov et al. 2016), while many others show evidence of ancestral genome duplication events (Blanc and Wolfe 2004). Newly formed polyploids commonly exhibit novel traits such as stress resilience, and increased organ and fruit size, which can be important in natural evolution as well as agriculture (Maherali et al. 2009; del Pozo and Ramirez-Parra 2015; Bomblies 2020; Van de Peer et al. 2021). The onset of polyploidy, however, is well known to trigger increases in cell size and nuclear volume, which can affect basic cell functions such as nucleo-cytoplasmic transport or ion homeostasis (del Pozo and Ramirez-Parra 2015; Robinson et al. 2018; Doyle and Coate 2019). Numerous studies have also reported epigenetic and transcriptomic changes in neo-polyploids (Stupar et al. 2007; Yu et al. 2010; Allario et al. 2011; Zhang et al. 2015, 2019; Xiang et al. 2019; Garcia-Lozano et al. 2021; Sun et al. 2022; Westermann et al. 2024).

Nuclear volume changes arising from WGD could directly influence chromatin organization and thus gene expression. In Arabidopsis thaliana, for example, it has been reported that biophysical constraints within the nucleus upon WGD can increase long-range interactions between chromosomes, and also affect gene expression in a subset of topological domains with altered compaction (Zhang et al. 2019). Other species, like watermelon, show more dramatic differences between diploids and neo-tetraploids, where even large chromatin regions classified into the euchromatic A compartments and heterochromatic B compartments can switch state upon genome doubling (Garcia-Lozano et al. 2021). In rice, neo-tetraploids have more accessible chromatin regions (ACRs) with the active histone modifications H3K36me2 and H3K36me3 associated with increased gene expression relative to diploid progenitors (Zhou et al. 2021). Neo-tetraploid rice has also been shown to have increased DNA methylation at transposable elements (TEs), which can reduce the expression of proximal genes (Zhang et al. 2015), induce gene expression in response to salt stress (Wang et al. 2021a), and influence three-dimensional (3D) chromatin interactions (Sun et al. 2022). Strong epigenetic regulation may also explain why neo-polyploid plants of many other species [e.g. maize (Riddle et al. 2010); potato (Stupar et al. 2007); Rangpur lime (Allario et al. 2011)] show only moderate increases in gene expression despite their doubled genome content.

ACRs can contain regulatory elements that serve as binding sites for transcription factors (TFs; Zhu et al. 2015; O’Malley et al. 2016; Maher et al. 2018; Sijacic et al. 2018; Lu et al. 2019; Reynoso et al. 2019). As such, one might expect that accessibility changes accompany gene expression changes. In plants, chromatin accessibility can dramatically change under biotic (Ding et al. 2021) and abiotic stress conditions (Reynoso et al. 2019; Zeng et al. 2019; Liang et al. 2021), during key developmental transitions (Pajoro et al. 2014; Yan et al. 2024) or when the genome is perturbed by epigenetic abnormalities (Zhong et al. 2021; Srikant et al. 2022; Zhao et al. 2022; Candela-Ferre et al. 2024). Polyploidy may also be such a perturbation. For example, chromatin accessibility differences between orthologous gene pairs in allotetraploid cotton are associated with differences in TF binding and expression divergence between subgenomes (You et al. 2022). Genes retained since WGD in the paleopolyploid soybean exhibit higher chromatin accessibility and gene expression levels than single copy genes; interestingly, chromatin accessibility was also increased at chromosomal breakpoints that occurred as the polyploid genome underwent diploidization, implicating accessibility not only in gene regulation but also genome rearrangement (Wang et al. 2021b). In allo-octoploid strawberry, on the other hand, genomic regions with high accessibility in one subgenome are less fragmented after polyploidization than their homeologs with lower accessibility in other subgenomes (Fang et al. 2024).

The above examples demonstrate that polyploidy can trigger epigenetic and accessibility changes that can affect genome structure and gene regulation. However, when comparing diploids with new polyploids, it is not generally clear whether observed changes are detrimental, neutral, or beneficial, or whether they will persist during subsequent evolution. In contrast, when comparing evolved polyploids to their diploid relatives, it is not clear which changes arose due to genome duplication per se vs. subsequent evolution. Moreover, we cannot be sure that the conserved features between diploids and evolved polyploids nevertheless may not have been perturbed by genome duplication and subsequently evolved back to the ancestral state (“Type-A trend” in Bomblies 2020). Comparing diploids, neo-polyploids, and evolved polyploids of the same species can overcome these limitations to at least some extent (Ramsey and Schemske 2002; Hegarty et al. 2013; Van Drunen and Husband 2018). Autopolyploids, which arise from within-species WGD events, are especially apt for studying these questions, as they do not have the added complexity of hybridity that allopolyploids (which arise from cross-species WGD) do (Parisod et al. 2010).

We use as our study system Arabidopsis arenosa, a relative of A. thaliana, from which it diverged around 6 Ma (Hohmann et al. 2015). This species naturally occurs both as a diploid and an autotetraploid form, which arose once, ∼30,000 generations ago (Arnold et al. 2015; Monnahan et al. 2019). Genome resequencing scans in the autopolyploid previously identified genes where novel variants in the tetraploid lineage show evidence of having undergone selective sweeps (Hollister et al. 2012; Yant et al. 2013; Bohutínská et al. 2021). These genes encode proteins involved in diverse processes, but meiosis, ion-homeostasis, cell division and growth, transcriptional regulation, and chromatin remodeling stand out as heavily represented processes. Where allele function has been explored, we have found that the derived alleles found in the evolved polyploids help stabilize polyploid meiosis (Morgan et al. 2020; Morgan et al. 2022) and polyploid pollen-tube tip growth (Westermann et al. 2024), both of which represent substantial fertility-compromising challenges for neo-polyploids. Multiple genes with evidence of having been under selection in the tetraploid lineage encode chromatin remodelers (e.g. BRM, ATRX, MOM1, and CHR11), epigenetic regulators (e.g. HDA15, HDA19, AGO1, DCL4, and RDR3) and core transcriptional regulators (e.g. NRPA1, NRPB1, and RDR4). Motivated by this finding, as well as the large body of work on epigenetic instabilities in other autopolyploid species (e.g. Mittelsten Scheid et al. 2003; Lavania et al. 2012; Zhang et al. 2019; Garcia-Lozano et al. 2021; Zhou et al. 2021; Wang et al. 2021a)), we hypothesized that chromatin architecture and together with this, gene expression, might be directly altered by WGD but subsequently stabilize, perhaps via the genes that show evidence of selection in the tetraploid lineage. That selection apparently acted on many genes encoding histone modifiers and chromatin remodelers, hints that WGD-induced changes required evolutionary adjustment, that is, they constituted a WGD-associated challenge.

In this study, we aimed to study the immediate epigenomic and transcriptional responses to WGD and compare this to the “evolved” state in natural established autotetraploids of A. arenosa. The latter are ∼30,000 generations old and well adapted to life as polyploids (Arnold et al. 2015). We describe genome-wide chromatin accessibility changes using ATAC-seq (Buenrostro et al. 2015), and gene expression changes using RNA-seq, in a three-way genotype/cytotype comparison, including diploids, neo-tetraploids, and evolved tetraploids. We find that some of the initial responses to genome doubling are maintained or further elaborated during subsequent evolution, while others appear to revert to the ancestral state. Chromatin accessibility across cytotypes is most dynamic in regions where the reference genome contains TEs yet shows only limited correlation with differential gene expression. The majority of accessibility and gene expression changes we observed in neo-polyploids relative to diploids reverted back in established tetraploids to nearly diploid levels. Incidentally, these tend to correlate with regions where chromatin accessibility variation is low. Together our results demonstrate how dynamic the effects of WGD can be at the nuclear level, while also suggesting that many initial changes may be temporary, perhaps because they pose challenges that are thus reverted. On the other hand, a subset does produce apparently persistent novel states that are retained or further enhanced in evolved polyploids.

Materials and Methods

Plant Growth and Tissue Collection

Seeds of natural populations (SNO, TBG) were sterilized with ethanol and a 6% Bleach solution, and stratified at 4 °C for 1 week. Stratified seeds were then sown on plates with ½ MS-Agar medium and kept in plant growth chambers for 15 d. Seedlings were then transplanted to soil pots (½ sand + ½ soil) and grown under greenhouse conditions at the ETH Research Station for Plant Sciences at Lindau, Zurich.

At ∼4 weeks after transplantation, rosette leaves were collected from three biological replicates per population. For each biological replicate, two leaves each were collected for whole-genome sequencing and RNA-seq, respectively, and frozen immediately in liquid nitrogen (and subsequently at −80 °C). Two additional leaves were collected for ATAC-seq and subjected to an additional fixation step before freezing in liquid nitrogen. In short, the leaves were fixed with 0.1% formaldehyde in phosphate buffer saline using syringe infiltration, followed by treatment with 0.125 M glycine in phosphate buffer saline, washed with autoclaved water thrice and dried before freezing in liquid nitrogen (and subsequently at −80 °C).

Approximately, 2 weeks after the onset of flowering, petals were collected from all samples. For each biological replicate, ten open flowers were collected in three sets, and petals were carefully dissected from each set. The first set was used for formaldehyde fixation and freezing (similar to rosette leaves) for ATAC-seq, and the second and third set were directly frozen in liquid nitrogen for whole-genome sequencing and RNA-seq, respectively. Due to insufficient RNA derived from the third set (using the Qiagen RNeasy Plant Mini kit), petals were once again collected (from ten healthy open flowers per biological replicate) at 9 weeks after the onset of flowering and subjected to RNA extraction using the Spectrum Plant Total RNA Kit (Sigma), which resulted in higher RNA yield.

Generation of Neo-tetraploid Lines

The apical meristem of 14-d-old seedlings was treated with 0.05% colchicine (Sigma C9754) diluted in sterile water with 0.05% Silwet-77 (Anawa 30630216). Neo-tetraploid branches were identified by flow cytometry performed on young flowers. Neo-tetraploid branches from different plants were used for crosses, to generate neo-4X F1s. The seeds of neo-4X (F1) lines used in this study were derived from two such independent crosses (Fig. 1b).

Fig. 1.

dACRs identified in rosette leaf and petal samples. a) Geographical origins of the diploid and autotetraploid populations used in this study. b) Cartoon demonstrating the generation of neo-tetraploids (“neo-4X” lines) from the F1 generation (the progeny of two colchicine-treated plants). c) Libraries prepared and tissues used in this study. d) Heatmap of z-scaled accessibility scores (measure of standard deviations above or below the mean) across dACRs for rosette samples of SNO (2X), neo-4X, and TBG (Est-4X). The dACRs are split into groups based on k-means clustering of accessibility trends across genotypes. The adjacent heatmap on the left shows the mean accessibility value for each dACR. e) Trends in mean accessibility for each k-group indicated in d). Error bars represent the standard error between three biological replicates for every genotype. f, g) Metaplots showing mean accessibility in proximity to genes f) and TEs and repeat regions g), for each k-group in d) and e). For f) TSS and TES represent transcription start site and transcription end site and for g) transposable element/repeat region start site and transposable element/repeat region end site, respectively. Panels h), i), j), k) represent Petal dACRs and are similar to d), e), f), g) respectively. CPM, counts per million.

ATAC-seq and gDNA Library Prep, Sequencing, and Mapping

Nuclei Isolation

For ATAC-seq analyses, each of the biological replicates was processed individually. Fixed tissue was chopped finely with 500 µL of General-Purpose buffer (0.5 mM spermine · 4 HCl, 30 mM sodium citrate, 20 mM MOPS, 80 mM KCl, 20 mM NaCl, pH 7.0, and sterile filtered with 0.2 µm filter, followed by the addition of 0.5% of Triton-X-100 before usage). The slurry was filtered through one-layered Miracloth (pore size: 22 to 25 µm), followed by filtration twice through a cell-strainer (pore size: 40 µm) to collect nuclei.

Fluorescence-Activated Cell Sorting

Nuclei were subjected to fluorescence-activated cell sorting (FACS) using a BD FACS Aria III—BSL1 sorter at the Flow Cytometry Core Facility, ETH Zürich. In brief, extracted nuclei were stained with DAPI (4′,6-diamidino-2-phenylindole), and the 2C population of nuclei (which were not endoreplicated) were gated and sorted into an empty microfuge tube using a 70 µm nozzle. For each biological replicate, 2 technical replicates of 20,000 nuclei from the gated population were sorted.

ATAC-seq Library Prep

Sorted nuclei were heated at 60 °C for 5 min, followed by centrifugation at 4 °C (1,000 × g, 5 min). The supernatant was removed, and nuclei were resuspended with a transposition mix [1 µL Tagment TDE1 enzyme (Illumina 20034197), 10 µL of 2X Tagment TD buffer (Illumina 20034197) and 9 µL autoclaved water] followed by a 37 °C treatment in a heat block for 30 min. About 200 µL sodium dodecyl sulfate buffer (50 mM Tris HCl pH 8.0, 1% sodium dodecyl sulfate, 10 mM Ethylenediaminetetraacetic acid pH 8.0) and 8 µL 5 M NaCl were added to the reaction mixture, followed by 65 °C treatment overnight. Nuclear fragments were then cleaned up using the Zymo DNA Clean and Concentrator kit (D4013). Three microliters of eluted DNA were subjected to 12 PCR cycles, incorporating Illumina indices, followed by a 1.8:1 ratio cleanup using AMPure XP beads (Beckman Coulter).

gDNA Library Prep

gDNA was extracted from each sample using the Macherey Nagel NucleoSpin PlantII kit. gDNA libraries were generated for each individual sample, using a maximum of 50 ng in a 9 µL volume as input for the library preparation. The transposition reaction was carried out by adding 1 µL Tagment TDE1 enzyme (Illumina) and 10 µL of 2X Tagment TD buffer (Illumina) to the gDNA, followed by a 37 °C treatment in a heat block for 30 min. Nuclear fragments were then immediately cleaned up using Zymo column-purification (DNA Clean and Concentrator). Three microliters of eluted DNA were subjected to 12 PCR cycles for incorporating Illumina indices, followed by a 1.8:1 ratio cleanup using AMPure XP beads.

Library Pooling and Sequencing

For both ATAC-seq and pooled gDNA libraries, diploid samples and tetraploid samples (including neo-tetraploids) were pooled separately. The library pools were then subjected to a right-sided size selection to retain fragments <800 bp. Library QC and sequencing were performed at the Functional Genomics Centre (FGCZ, University of Zurich) in an Illumina NovaSeq 6000 instrument (Full SP Flow cell, three batches) with 2 × 150 bp paired-end reads.

Mapping of ATAC-seq and gDNA Libraries to the Strečno Reference Genome

For ATAC-seq libraries, reads were aligned as two single-end files to the Strečno (A. arenosa) reference genome (Barragan et al. 2024) using bwa-mem (default options), filtered for the SAM flags 0 and 16 (only reads mapped uniquely to the forward and reverse strands), and separately converted into bam files. The bam files of the forward and reverse reads were then merged, sorted, and PCR duplicates were removed using picardtools. This resulted in an average of 12.8 million mapped reads for each diploid technical replicate and 19.5 million mapped reads for each tetraploid technical replicate. The sorted bam files from technical replicates of each sample were subsequently merged (samtools merge—default options) to obtain a final average of 25.6 million mapped reads for each diploid biological replicate and 36.9 million mapped reads for each tetraploid biological replicate. gDNA libraries were similarly aligned to the Strečno reference, with an average of 12.8 million mapped reads per diploid library and 18 million mapped reads per tetraploid library.

Peak Calling and Identification of dACRs

Peak calling was carried out for each biological replicate using MACS2 (https://github.com/macs3-project/MACS; Zhang et al. 2008) with the following parameters:

macs2 callpeak -t [ATACseqlibrary].bam -c [Control gDNA library].bam -f BAM –nomodel –extsize 147 –keep-dup = all -g 1.5e8 -n [Output_Peaks] -B -q 0.05.

After peak calling, every peak set was further filtered based on their respective q-values in the MACS2 peaks.xls files, retaining peaks with q ≤ 0.01. The R package DiffBind [3.4.11, (Stark and Brown 2021)] was used to identify consensus peaks from different sample groups. In brief, filtered peak files and .bam alignment files were processed to identify consensus peaks that overlapped in at least two out of three biological replicates per group, and represented peaks unique to at least one group (FDR-adjusted P-value <0.01). These ACRs identified by DiffBind were then used to generate accessibility scores in counts per million (CPM) after trimmed mean of M-value normalization. Next, the CV of mean accessibility levels between cytotypes (measured in log₂ CPM) was calculated for each ACR, and those with the top 25% of CV values were retained as dACRs.

For downstream analyses, the dACRs (rosette dACRs and petal dACRs separately) were split into four k-groups based on k-means clustering (an unsupervised machine learning algorithm) of their accessibility levels across all samples, using the kmeans() function in R with centers = 4. The optimal number of k-groups was chosen as four based on an “elbow” plot comparing the number of k-groups with the within-cluster sum of squares of transformed accessibility counts. Accessibility levels at each dACR were measured in CPM, or in transformed counts (by applying log₂{CPM accessibility value + 1} which has been denoted as log₂ CPM in figures for simplicity). Accessibility dynamics were measured from the CV across cytotypes using a custom R function: CoV < −function(x){CV < − (SD(x)/mean(x)) × 100; return(CV)}, which was applied to a data frame containing mean transformed accessibility counts for diploids (SNO), all neo-tetraploids (neoL2, neoL3), and evolved tetraploids (TBG).

Generation of Randomly Sampled Controls for dACRs

For both rosette leaf and petal dACRs, the shuffle function of bedtools (v2.30.0) was used to generate randomly shuffled control regions with the same size distribution as dACRs (including the distribution across k-groups). These controls for dACRs were also used to estimate overlaps with DEGs, non-DEGs, and TEs (see below).

Metaplots of dACRs Near Genes and TEs/Repeats

Accessibility levels near the start and end sites of annotated TEs, repeats, and genes were computed using the computeMatrix function of deepTools (https://deeptools.readthedocs.io/en/develop/index.html). The output files were used to generate metaplots using a custom R script (https://github.com/groverj3/genomics_visualizations/blob/master/metaplotteR.r).

Genome Browser Visualization of dACRs

The MACS2 function bdgcmp was used to generate bedgraph files (.bdg) of fold enrichment by comparing the treatment (treat_pileup.bdg) and control (control_lambda.bdg) files generated during peak calling, and individual.bdg files of various samples were subsequently loaded on IGV (Integrative Genomics Viewer) for visualization.

RNA-seq Library Prep, Sequencing, and Mapping

RNA Extraction

For rosette leaf samples, RNA was isolated using the Qiagen RNeasy Plant Mini Kit (74904), and checked for quality using Nanodrop (A260/280 and A260/230 ratios >2) followed by concentration measurement using Qubit BR RNA reagents. About 2,000 ng of RNA (in a maximum volume of 17 µL, suspended in RNase-free water) was used as input for a 20 µL DNase I reaction. In brief, 1 µL of DNase I enzyme (Thermo Scientific) and 2 µL of 10X DNase I buffer (Thermo Scientific) was added to 17 µL of normalized RNA, and the reaction mixture was incubated at 37 °C for 30 min. The reaction mixture was then directly subjected to a cleanup with the Zymo RNA Clean and Concentrator kit (R1015) and eluted in 30µL of RNase-free water. Purified RNA was once again checked for quality and concentration.

For petal samples, RNA isolation with the Qiagen RNeasy Plant Mini Kit (74904) resulted in low yield and quality for many of our samples. Therefore, petals were once again collected from ten flowers per biological replicate at 9 weeks after the onset of flowering and subjected to RNA extraction using the Spectrum Plant Total RNA Kit (Sigma, STRN250), which resulted in higher RNA yield and quality. DNaseI treatment was performed similarly to rosette leaf samples, followed by cleanup using the Zymo RNA Clean and Concentrator Kit (R1015).

RNA-seq Library Prep and Sequencing

RNA-seq libraries were prepared by Novogene UK Ltd (Directional mRNA enrichment libraries) and subjected to NovaSeq paired-end sequencing (2 × 150 bp) at an average coverage of 17 million reads per library.

Mapping of RNA-seq Libraries

Fastq files of samples were aligned using bowtie2 to the Strečno reference genome (Barragan et al. 2024), prepared using the rsem-prepare-reference function of the RSEM software. Aligned bam files were sorted and indexed using samtools V1.9. Gene transcript counts for each sample were estimated using rsem-calculate-expression. For each sample, transcript counts and metadata were imported into the R software using the packages “tximport” and “tximportData” for creating a DESeq object [for processing with the “DESeq2” package (Love et al. 2014)].

DEG-calling

A DESeq object containing all 24 samples was first generated. After filtering genes with low read counts, a common set of 18,226 genes were retained (out of the 22,630 annotated genes in the Strečno genome reference), and the DESeq function was applied under a two-factor interaction model (∼design = ∼Ploidy + Tissue + Ploidy:Tissue) and default parameters (nbinomWald test). DEGs were subsequently identified by generating various pairwise contrasts (across different sample groups) and identifying genes with a P-value <0.01 and |log₂ FoldChange| > 1. The pairwise contrasts were made as follows: a first contrast (“contrast 1”) was performed to compare diploids with neo-tetraploids (neo-4X vs. 2X), “contrast 2” compared neo-tetraploids with evolved tetraploids (Est-4X vs. neo-4X), and “contrast 3” compared diploids with evolved tetraploids (Est-4X vs. 2X; supplementary table S1 and dataset, Supplementary Material online). We discovered a total of 4,692 DEGs (from both rosette and petal samples put together; supplementary table S1, Supplementary Material online) in our experiment, which together represented approximately a quarter of the 18,226 genes analyzed.

Rosette leaf union DEGs and petal union DEGs were generated by combining DEGs from all three contrasts. These “union DEGs” were subsequently each divided into four groups in the following manner: first, mean expression levels across samples (vsd counts/transformed read counts) were calculated for each cytotype (i.e. three samples of SNO, six samples of neo-4X, three samples of TBG). Based on the mean expression levels, differences between cytotypes were then calculated, such that:

• Transition 1 Expression difference = mean expression (neo-4X) − mean expression (SNO)

• Transition 2 Expression difference = mean expression (TBG) − mean expression (neo-4X)

DEGs were then classified into one of four groups based on the Transition 1 and 2 Expression differences:

• Group 1: Upregulated in Transition 1, Upregulated in Transition 2

• Group 2: Downregulated in Transition 1, Upregulated in Transition 2

• Group 3: Upregulated in Transition 1, Downregulated in Transition 2

• Group 4: Downregulated in Transition 1, Downregulated in Transition 2

These groups were used to generate the heatmaps in Fig. 2.

Fig. 2.

Transcriptome differences across cytotypes. a) Schematic of comparisons between SNO (2X), neo-4X, and TBG (Est-4X). b) Venn diagrams denoting the number of DEGs identified in pairwise comparisons of cytotypes in rosette leaf and petal tissues. c) Heatmap representing union DEGs of rosette leaf samples, split by four groups. Trendline plots for every group indicate mean expression changes across cytotypes. Error bars represent the standard deviation between three biological replicates for every genotype. d) Heatmap similar to c) for petal union DEGs.

Strečno Genome and Annotations

The diploid Strečno reference genome (Barragan et al. 2024) was used as the reference genome for this study. Genome sequence and gene annotations were downloaded from NCBI-SRA. Gene orthogs to the A. thaliana TAIR10.1 assembly and the A. lyrata v2.0 (Rawat et al. 2015) were identified using OrthoFinder (Emms and Kelly 2019). TEs and repeats were annotated using EDTA (https://github.com/oushujun/EDTA; Ou et al. 2019). For LTRs annotated based on structural homology, only the parent feature was retained. TE superfamilies were inferred from the TE sequence ontologies provided in the output files of the EDTA pipeline.

Feature	#Instances
Annotated genes	22,630 (16,867 with 1-1 TAIR10.1 orthologs, 16,554 with 1-1 A. lyrata orthologs)
EDTA annotated TEs/repeats	56,545 total sequences
	TE sequence ontology	Count
	LINE	1,055
	LTR
	• Copia	6,985
	• Gypsy	3,334
	• Unknown (other)	2,472
	TIR
	• CACTA	2,131
	• Mutator	3,748
	• PIF Harbinger	1,918
	• Tc1 Mariner	1,581
	• hAT	2,311
	Non-TIR
	• Helitron	17,452
	Repeat region	13,558

Gene Ontology Enrichment

For all Strečno genes, TAIR orthologs identified using OrthoFinder were filtered to retain only those that had single copy orthologs, or those that had multiple transcript variants orthologous to the same gene. The TAIR orthologs of different gene sets were then used as input for gene ontology enrichment using STRING (www.string-db.org).

Motif Enrichment Analysis

Identification and enrichment of TF-binding motifs in dACR regions was performed using the MEME suite (Bailey et al. 2015). In brief, FASTA sequences of dACRs (all dACRs, k-group specific dACRs, and dACRs with DEGs for each tissue) were used as input for STREME, and the resulting motifs were compared with the A. thaliana CISBP and DAP-seq databases using TOMTOM. The motifs enriched from all the above dACR groups were then combined together, and filtered to retain motifs with the following features: E-value ≤0.001, q-value ≤0.0001, Offset to query ≤4 and length difference to query ≤5 bp. This resulted in 51 motifs, which were further filtered to 25 motifs based on the redundancy between CISBP and DAP-seq motifs (where DAP-seq motifs, including amp-DAP-seq motifs, were preferentially retained), and whether the genes encoding the TFs had sufficiently detectable read counts in our rosette leaf and petal RNA-seq datasets. The .meme files of the 25 motifs were subsequently used as queries for FIMO, to identify and count motif occurrences near all dACR groups, as well as randomly shuffled control regions with the same size distribution as dACRs in each of the groups [generated using the shuffle function of bedtools (Quinlan and Hall 2010) 2.30.0]. For each motif, the fold enrichment value was calculated by dividing the number of dACRs containing the motifs by the number of control regions containing the same motif. The percentage occupancy of the motif was calculated by dividing the number of dACRs containing the motif by the total number of dACRs in a given group. The candidate gene targets for each motif were determined by identifying the closest gene (within 1.5 kb) to the dACR in which the motif occurred.

Associations Between dACRs, Genes, and TEs

Genes (DEGs and non-DEGs) and TEs in cis to dACRs were identified using bedtools (2.30.0) closest, and filtered to retain the closest hits that occurred within 1.5 kb upstream or downstream the dACR region, including those features that overlapped with the dACR region. The list of TEs used for intersection with dACRs also included repeat regions annotated by EDTA. Chi-square tests were carried out using the chisq.test function in R to measure the significance of enrichment of these cis features near dACRs, relative to their occurrence near randomly sampled control regions with the same length distribution as dACRs.

Intersection of dACRs and DEGs With Genes Under Selection

We examined two lists of genes under selection in tetraploid A. arenosa—the first list combined A. thaliana orthologs of all genes under selection from Hollister et al. (2012) and Yant et al. (2013), and was compared with A. thaliana orthologs of rosette leaf DEGs, petal DEGs, and all genes (DEGs and non-DEGs) in cis to rosette leaf and petal dACRs. The same DEGs and genes (Strečno annotation) were also intersected with a second list of the top 0.1% outliers from Bohutínská et al. (2021) (using Strečno orthologs of the A. lyrata genes).

Results

Chromatin Accessibility Changes Across Cytotypes Cluster into Distinct Trends

We generated and sequenced ATAC-seq, RNA-seq, and genomic DNA (gDNA) libraries from rosette leaves and petals of A. arenosa plants of three different cytotypes—diploids (“2X”, population SNO from Strečno, Slovakia), established autotetraploid plants (“Est-4X,” population TBG from Triberg, Germany) and neo-tetraploid lines generated in the lab using colchicine treatment (Fig. 1a to c, Materials and Methods). Two neo-tetraploid lines (“neoL2” and “neoL3,” which together comprise the cytotype referred to as “neo-4X” hereafter) used in this study were derived from independent crosses between two colchicine-treated neo-tetraploid SNO plants (Fig. 1b). This is important because it means the neo-tetraploids analyzed were not directly exposed to colchicine (which is toxic) and are not chimeric for diploid and tetraploid sectors, which we commonly observe in colchicine-treated plants. For each genotype, three individual plants were chosen as biological replicates, and sequencing libraries were generated from leaf and petal tissues collected from the same individuals. We mapped all of our sequenced libraries to a high-quality reference genome of the diploid Strečno population (Barragan et al. 2024), from which we also derived gene and TE annotations (Materials and Methods). We do caution that this approach could introduce some reference bias in calling accessible regions or differentially expressed genes (DEGs; below), but it is unlikely to affect our overall conclusions.

We first compared the chromatin accessibility landscape of 2X, neo-4X, and Est-4X using ATAC-seq (normalized to sample-specific gDNA libraries as a control). We identified a consensus set of 20,150 and 22,693 ACRs across all rosette leaf and petal samples, respectively. These lists were then filtered to retain only the top 25% most variable in accessibility across cytotypes—we refer to these hereafter as “differentially accessible chromatin regions” (dACRs; Materials and Methods). We obtained a final set of 5,046 rosette leaf dACRs and 5,674 petal dACRs (supplementary dataset, Supplementary Material online). We used a k-means clustering approach (Materials and Methods) to identify four major clusters of dACRs (which we call as “k-groups”) based on their accessibility trends across cytotypes. (Figure 1d, e, supplementary table S1, Supplementary Material online, Materials and Methods). Each k-group exhibited a distinct change in accessibility levels when comparing diploids and neo-tetraploids (WGD response, referred as “Transition 1” hereafter), and between neo-tetraploids and evolved tetraploids (subsequent evolution and/or genotypic differences, referred as “Transition 2” hereafter; Fig. 1h, i). Although the dACRs and k-groups were generated independently for rosette leaf and petal samples, we could identify strong similarities for three out of four k-groups (rosette leaf k2, k3, k4 being similar to petal k4, k3, k1, respectively). This was, however, not due to a high overlap between rosette leaf and petal dACRs (only 1,841 dACRs were common to both tissues), suggesting that different chromatin regions across tissues may exhibit similar trends of accessibility changes during polyploidy evolution. Conversely, we found that the k-groups with the highest positional overlap between tissues (rosette leaf k-group 1 and petal k-group 2 having 437 dACRs in common, supplementary fig. S1, Supplementary Material online) showed opposite trends of accessibility changes in Transition 1, despite both increasing in accessibility levels in Transition 2.

The accessibility trends for each k-group were consistent over gene bodies, promoters, and downstream regions (Fig. 1f, j). For both rosette leaf and petal dACRs, three of the four k-groups (comprising 68.7% and 73.3% of dACRs, respectively) showed marked accessibility changes already in Transition 1, indicating substantial immediate changes triggered by WGD. Furthermore, accessibility levels for certain regions (e.g. in petal k-groups 1 and 2), exhibited large variability across neo-tetraploid lines, but were overall quite similar between both neo-tetraploid lines (neoL2 and neoL3) used in our study.

With respect to their genome-wide distribution, dACRs from each k-group did not show any obvious overrepresentation on particular chromosomes, nor any overall differences in the fraction of dACRs that have protein-coding genes and/or TEs (including repeats) in their proximity (supplementary fig. S1, Supplementary Material online, Materials and Methods). We next asked whether the unique patterns of each k-group could be related to their having functionally similar genes in cis—reflecting some kind of coordinated response. Gene Ontology enrichment analysis identified a statistically significant enrichment of chloroplast genes in rosette leaf k-groups 1 and 3, both of which become more accessible upon WGD, but differ in the “later evolution” (Transition 2) with about half decreasing and half increasing further in established tetraploids. Rosette leaf k-group 2 was enriched (1.9-fold) for DNA-binding TF activity (supplementary fig. S2, Supplementary Material online). For petal dACRs, genes in cis to k-group 1, which are regions in which accessibility declines upon WGD, were enriched for “anatomical structure development.” Genes lying near petal k-group 2 regions, which decrease in accessibility upon WGD and increase again in established tetraploids, were enriched for response to oxidative stress, metabolic processes and reproductive structure development (supplementary fig. S3, Supplementary Material online).

We also noted that chromatin accessibility could be highly variable over TEs and repeat regions (annotated in the diploid reference genome) for different k-groups (Fig. 1g, k), which could result from both genetic and epigenetic instabilities in these regions during WGD. The changes in accessibility across cytotypes were also higher over TEs than over genes. Finally, we observed that all petal dACRs grouped together had a lower mean accessibility level (both over genes and TEs) than rosette leaf dACRs (supplementary fig. S1, Supplementary Material online), although the magnitude of their accessibility changes across cytotypes were similar (supplementary fig. S1, Supplementary Material online). Overall, these results suggest that chromatin accessibility levels show tissue-specific changes both at the onset of polyploidy (Transition 1) and in subsequent stabilization (Transition 2).

Gene Expression Changes Upon WGD are Often Reversed in Evolved Tetraploids

To explore whether dACRs are associated with altered transcriptional regulation during polyploidy, we first analyzed RNA-seq data from the same individuals used for ATAC-seq. We generated pairwise comparisons of the three cytotypes to identify DEGs among 18,226 genes with sufficient read counts out of 22,630 total annotated genes (Materials and Methods, Fig. 2a, supplementary table S1 and dataset, Supplementary Material online). We defined DEGs as genes that exhibited at least a 2-fold change in expression in a given pairwise comparison, and with an adjusted P-value ≤0.01. We discovered a total of 4,692 DEGs (supplementary table S1, Supplementary Material online) in our experiment, which together represented approximately a quarter of the 18,226 genes analyzed. Thus, while the expression of the majority of genes appeared unchanged in response to WGD, a substantial fraction did change.

For both rosette leaf and petal samples, the highest number of DEGs was observed in a comparison between neo-tetraploids and evolved tetraploids (1,177 DEGs and 2,483 DEGs respectively; Fig. 2b, supplementary table S1 and dataset, Supplementary Material online), which is unsurprising as this represents both adaptation to WGD as well as neutral or adaptive genotypic divergence. However, the total number of DEGs identified when comparing neo-4X and 2X or Est-4X and neo-4X was higher than the number of DEGs between Est-4X and 2X (1,588 vs. 1,071 for rosette leaves, and 3,167 vs. 1,840 for petals; Fig. 2b, supplementary table S1 and dataset, Supplementary Material online). This pattern could be explained if a substantial portion of genes that change expression after WGD (neo-4X vs. 2X) or during subsequent evolution (Est-4X vs. neo-4X) return to more diploid-like levels, such that the diploid SNO and established tetraploid TBG have more similarly expressed genes than either does relative to neo-4X. To further explore this, we combined DEGs from all pairwise comparisons, to obtain 1,967 rosette leaf “union” DEGs and 3,665 petal “union” DEGs (Fig. 2b, supplementary dataset, Supplementary Material online). We then clustered these union DEGs into groups (Fig. 2c, d) based on expression changes between 2X and neo-4X (“Transition 1” or WGD response, Fig. 2a) and between neo-4X and Est-4X (“Transition 2” or subsequent evolution, Fig. 2a; Materials and Methods). Both for rosette leaf and petal samples, Group 3 DEGs (which showed increased expression in Transition 1 and reduced expression in Transition 2) were the most abundant (32.1% and 35.9% for rosette leaf and petal DEGs, respectively). Both Group 2 and 3 DEGs (comprising 57.1% to 62.9% of the total union DEGs) represent genes whose expression changes during Transition 1 but were reversed in Transition 2, resembling a “Type A” evolutionary trend as proposed in (Bomblies 2020). As described above for dACRs, this pattern is expected when a WGD-induced change constitutes a “challenge” that is then overcome during subsequent evolution (see Discussion).

Groups 1 and 4 (which comprise genes whose expression changes upon WGD but then become more extreme in evolved polyploids) contain 37.1% to 42.9% of the union DEGs (“Type B” trends; Bomblies 2020). These trends may represent favorable or neutral novelties that persist or are further elaborated in Est-4X. All rosette leaf union DEGs were functionally enriched for genes involved in response to hypoxia, ethylene signaling, and water deprivation (supplementary fig. S4, Supplementary Material online), which may relate to the observation that many polyploids are immediately, and often lastingly, more drought resistant than their diploid progenitors (see for review, Bomblies 2020). Petal DEGs of Group 2 were particularly enriched for functions in metabolic processes and the biosynthesis of amino acids, which suggests these processes may be perturbed by WGD but return to diploid-like levels over time (Fig. 2d, supplementary fig. S4, Supplementary Material online).

We identified 940 DEGs common to both tissues. These represented 48% of rosette leaf union DEGs and 25% of petal union DEGs, which is consistent with many more DEGs being identified in petal samples than in rosette leaf samples (Fig. 2b). The gene expression changes at these DEGs were similarly clustered for both tissues (supplementary fig. S4, Supplementary Material online), although their proportions moderately varied from the overall set of DEGs (Fig. 2). This list of common DEGs was not significantly enriched for any functional category.

The Relationship Between Accessibility Changes and Expression Changes During WGD is Complex

Accessible regions are often known to be hotspots for TF binding (O’Malley et al. 2016) and are also involved in regulating cell-type or tissue-specific gene networks [e.g. (Maher et al. 2018; Sijacic et al. 2018; Marand et al. 2021; Zhang et al. 2024)]. We thus investigated the extent to which chromatin accessibility changes correlate with expression changes in the polyploidy context by integrating our ATAC-seq and RNA-seq datasets. Hence, we examined the union DEGs which occurred in cis (those overlapping or within 1.5 kb upstream or downstream) to dACRs, thereby identifying positionally co-occurring dACR-DEG pairs. Therefore, we sorted dACR-DEG pairs based on the consistency of their trend across both Transitions 1 and 2. We first searched for dACR-DEG pairs where, in both Transitions 1 and 2, increased accessibility accompanied increased expression levels (or decreased accessibility accompanied decreased expression levels). We called these “positively associated dACR-DEG pairs.” Similarly, “negatively associated dACR-DEG pairs” showed consistent negative associations for both Transitions 1 and 2, where increased accessibility is associated with reduced expression or vice versa. All other dACR-DEG pairs were called “inconsistent association pairs,” since the relationship between accessibility and expression changes was not consistent between Transitions 1 and 2. When the dACR-DEG pairs were split based on the dACR k-groups (Fig. 1), we could also observe that expression changes across cytotypes followed the patterns of accessibility changes for the “positive association” pairs and mirrored the patterns of accessibility changes for the “negative association” pairs (Fig. 3a, b).

Fig. 3.

dACR and DEG associations. a, b) Trendline plots showing chromatin accessibility patterns across cytotypes (dark lines, y axis on the left) and expression patterns across cytotypes (lighter lines, y axis on the right) of dACRs and cis DEGs, split by k-groups for rosette leaf a) and petal b) samples, respectively. Numbers in the triangles represent the percentage of dACR-DEG pairs with positive/negative/inconsistent associations compared with all dACR-DEG pairs. c, d) Venn diagrams representing overlaps between all dACRs and all DEGs in rosette leaf c) and petal d) samples, respectively. e) Heatmap of fold enrichment values for 25 TF-binding motifs at dACRs. Fold enrichment values represent the ratio of TF-binding motifs in dACRs compared with randomly sampled control regions of the same length. TFs which are also identified as DEGs, are marked in green (rosette leaf DEGs) and pink (petal DEGs). Columns representing fold enrichment of TFs at all rosette leaf dACRs and all petal dACRs are highlighted with a black dotted line. f) Proportion of target genes for each motif in e) which are classified as rosette leaf DEGs or non-DEGs (heatmap on the left) and petal DEGs or non-DEGs (heatmap on the right). CPM, counts per million.

Remarkably, both rosette leaf and petal samples showed a highly similar proportion of dACR-DEG pairs that were positively associated (39.8% and 43.4%, respectively), negatively associated (14.1% and 15%) and pairs which showed inconsistent associations between accessibility and expression changes (46% and 41.5%; Fig. 3a, b, supplementary table S1, Supplementary Material online). The large number of dACR-DEG pairs with inconsistent associations suggests that accessibility and expression changes may be independent of each other in most cases. Consistent with this, only one-fourth of all rosette leaf and petal DEGs were in cis to a dACR, while an even smaller fraction of all dACRs (12% of the total for rosette leaves and 20% for petals) had cis DEGs (supplementary table S1, Supplementary Material online, Fig. 3c, d). Nevertheless, there was a significant enrichment in the overlap between dACRs and DEGs, compared with the overlap between randomly sampled regions (with similar length distribution as dACRs) and DEGs (Materials and Methods, Fig. 3c, d, supplementary table S1, Supplementary Material online). We also found that regions containing more non-DEGs show low dynamism in chromatin accessibility across cytotypes (supplementary fig. S5, Supplementary Material online). This suggests some dACRs do affect local gene expression.

We next investigated whether dACRs were enriched for any TF-binding motifs, which might be expected if there is a “ploidy response” mediated by particular pioneer TFs. We performed motif enrichment analysis separately for all dACRs, k-group dACRs and dACRs with cis DEGs, across both rosette leaf and petal samples, and identified known binding motifs for 24 TFs (Fig. 3e, Materials and Methods, supplementary fig. S6, Supplementary Material online) which were also found to be sufficiently expressed based in our RNA-seq datasets (supplementary fig. S6, Supplementary Material online). These TFs included eight involved in response to light stimulus (PIF3, PIF4, PIF7, PIF8, BBX31, ATHB23, CIB4, and GBF1), two involved in brassinosteroid signaling (BIM1, BIM2) and three involved in regulation of flowering time (REM19, CIB4, and BBX31; Fig. 3e, supplementary table S2, Supplementary Material online). CIB4 motifs were particularly highly enriched in petal dACRs and dACRs with cis DEGs. This TF regulates cell differentiation in the shoot apex during the onset of flowering (Hou et al. 2019). Apart from chromatin accessibility changes facilitating TF-binding and target gene regulation, the differential expression of the enriched TFs themselves (supplementary fig. S6 and table S2, Supplementary Material online) may also play a role. We observed that 3 of the 25 TFs in our lists were identified as DEGs. Two TFs (GBF1 and PIF3) are implicated in regulation of photosynthesis and response to blue light (Norén Lindbäck et al. 2023) and both are DEGs in rosette leaf samples (GBF1 differentially expressed between neo-4X vs. 2X and Est-4X vs. 2X; PIF3 differentially expressed between Est-4X vs. neo-4X and Est-4X vs. 2X). One (DROP2) which is associated with pollen and root hair development (Karas et al. 2009; Zhang et al. 2017) is a petal DEG (differentially expressed between Est-4X vs. neo-4X and Est-4X vs. 2X; Fig. 3e, supplementary fig. S6 and table S2, Supplementary Material online). It is unclear why DROP2 motifs and PIF3 (which are known to be more expressed in leaves than petals in A. thaliana; supplementary  table S2, Supplementary Material online) are enriched in petal dACRs. Further experiments would be needed to explore the relevance of this (if any). Finally, we examined all genes in cis to all dACRs with TF-binding motifs and observed that a majority of these “TF-target genes” were non-DEGs (Fig. 3f). This speaks against the idea that there is a strong, coordinated response to WGD mediated by TFs in accessible regions.

TEs in cis to dACRs are Quantitatively Associated with Accessibility Dynamics Across Cytotypes

Since we found that the number of dACRs with cis DEGs (and vice versa) was low, we next counted all cis features (DEGs, non-DEGs, TEs and repeats) within 1.5 kb upstream or downstream of dACRs (Materials and Methods). We found that a large majority of dACRs co-localized with cis TEs (>65% including repeats), cis non-DEGs (>45%) or a combination of both (>25%; Fig. 4a, supplementary table S1, Supplementary Material online). Incidentally, only a very small proportion (<7%) of all annotated TEs (60,005 in total, including 14,018 repeat regions) and <17% of all non-DEGs (16,259 for rosette leaves and 14,561 for petals) were in cis to dACRs, suggesting that their abundance near dACRs was not expected from their genome-wide prevalence alone (Fig. 4a, supplementary table S1, Supplementary Material online). Based on the TEs annotated in the diploid reference genome, we had initially found that accessibility could increase over TEs for certain dACR k-groups (Fig. 1g, k) which prompted us to examine whether these could be due to enrichment of certain TE superfamilies in cis to dACRs. When comparing cis TEs of dACRs to those in cis to randomly sampled control regions (Materials and Methods), we found that three superfamilies (Mutator, PIF Harbinger, and hAT), which belong to the Class II terminal inverted repeat (TIR) DNA transposons, were consistently enriched in both rosette leaf and petal dACRs (Fig. 4b). Incidentally, TIR transposons have previously been shown to be enriched also in ACRs identified among various maize inbred lines (Noshay et al. 2021), in subgenome-specific ACRs of allo-octoploid strawberry (Fang et al. 2024), and in gene-distal ACRs of 13 diverse plant species (Lu et al. 2019). This group of TEs is also known to be differentially expressed under varying temperature and irradiation conditions in certain A. arenosa populations (Wos et al. 2021).

Fig. 4.

Cis features of dACRs are associated with accessibility variation across cytotypes. a) Number of rosette leaf and petal dACRs with various combinations (as indicated in the heatmap on the left) of cis features—TEs/repeats, DEGs, non-DEGs. Heatmaps to the right of the bar plots represent the fraction of cis features occurring in proximity to dACRs. b) Stacked barplots showing distribution of various TE sequence ontologies (annotated in the 2X reference genome) across dACRs, randomly sampled control regions, and all TEs/repeats in the 2X reference genome. Solid outlines demarcate three TIR TE sequences that are enriched in dACRs, while dotted lines demarcate TE sequences that show high accessibility changes across cytotypes. c, d) Accessibility dynamics across a varying number of TE/repeat regions, for various k-groups of rosette leaf c) and petal d) dACRs. CV refers to the coefficient of variation in mean accessibility across cytotypes, as a measure of accessibility dynamics. e, f) Accessibility dynamics of rosette leaf e) and petal f) dACRs with a single cis TE, for various TE sequence ontologies. For e, f), boxplots on the right of the dotted line indicate the accessibility CV distribution for dACRs without cis TEs. dACRs with cis TE sequences not denoted by “ns” show significantly different (P < 0.05) mean accessibility CV than dACRs without any cis TEs (far right). For c, d), asterisks indicate statistical significance for pairwise Wilcoxon tests.

To check whether dACRs with cis TEs showed different accessibility patterns than those distal to TEs, we first examined the distribution of accessibility levels of dACRs from diploids, neo-tetraploids, and established tetraploids, classified by the number of TEs in cis to every dACR (supplementary fig. S7a, b, Supplementary Material online). We observed a pattern where the differences in accessibility between cytotypes were more prominent when multiple TEs were found in cis to dACRs, i.e. the degree of change in accessibility among genotypes/cytotypes was strongly associated with an increasing number of nearby TEs. We quantitatively verified these “accessibility dynamics” by measuring the coefficient of variation (CV) of accessibility levels between cytotypes, for dACRs with a varying number of cis TEs (Fig. 4c, d). It should be noted that these TEs are those annotated in diploids, and it is likely that some of these are either presence/absence polymorphisms in A. arenosa or can mobilize upon WGD and may not remain in the same genomic position in the neo-tetraploid and evolved tetraploid. Nevertheless, it remains that these TE-containing regions in the diploid are particularly prone to high accessibility changes during polyploid formation and/or subsequent evolution, though cause vs. consequence is difficult to define.

When we examined accessibility dynamics across dACRs with different cis TE sequences (single copies), we noticed that regions where the reference contains Gypsy and other LTR TEs (but not Copia LTRs) and CACTA (TIR) TEs showed the largest accessibility variation among all TEs, both for rosette and petal dACRs (Fig. 4e, f). Among the TIRs that were enriched in dACRs, the hAT TIRs showed slightly higher accessibility changes than the PIF Harbinger and Mutator sequences. Given that LTRs are Class I TEs which can mobilize through a copy–paste mechanism, it is likely that the accessibility changes across cytotypes that we observed may reflect chromatin instabilities in or around the “mother” copies of these elements. On the other hand, CACTA and hAT elements which mobilize by excising themselves through a “cut-and-paste” mechanism (Class II TEs) and are known move at high frequency in autotetraploid A. arenosa (Baduel et al. 2019). However, the previous study was carried out using the Arabidopsis lyrata genome reference, while we have used a diploid A. arenosa reference for our study. Since we do not know whether and where these elements have mobilized in neo-tetraploids and evolved tetraploids, it is difficult to interpret the mechanisms by which their possible excision could affect accessibility changes across cytotypes or potentially cause artifacts in our dACR calls in some cases.

Interestingly, when comparing all singular TE occurrences near dACRs, we found that the fewest accessibility changes occur near Tc1 Mariner elements (also TIRs; Fig. 4e, f). These elements also show evidence of considerable mobilization in A. arenosa, but minimally differ in their overall density when comparing diploid and autotetraploid cytotypes (Baduel et al. 2019). For dACRs with two or more TEs in cis, the presence of two Helitron elements was most frequently observed, but the low numbers of TE sequences in various other combinations made it difficult to derive any general conclusions regarding their putative effects on accessibility dynamics (supplementary fig. S7c, d, Supplementary Material online).

When integrating all combinations of cis features near all dACRs, we found that the diploid-annotated TEs indeed appear to have the strongest positive effect on accessibility dynamics across cytotypes, and this positive correlation remains whether they co-occur with DEGs or non-DEGs (Fig. 5a). Standalone dACRs with no cis features, and dACRs with only cis non-DEGs, exhibited the lowest accessibility dynamics (Fig. 5a). These patterns were strikingly similar for both rosette leaf and petal dACRs, despite the large differences in their genomic positions as well as the tissue-specific DEGs and non-DEGs in cis to these dACRs.

Fig. 5.

TEs exhibit the largest changes in accessibility across cytotypes. a) Accessibility dynamics of rosette leaf and petal dACRs with different combinations of cis features. Accessibility dynamics are represented by CV in mean accessibility across cytotypes. P-values of significance are indicated for pairwise comparisons between dACRs with no cis features (“-”) and dACRs with various combinations of cis features, with significance levels provided to the right of a). “ns” refers to non-significant. b) Genome Browser screenshot showing large accessibility changes at two dACRs surrounded by Gypsy (LTR) TEs in a locus on Chromosome 4 for rosette leaf samples. The dACR region is indicated with gray shading across the tracks and marked in red at the locus. Each track represents a distinct sample with accessibility represented in fold enrichment values from [0 to 3] calculated from the respective ATAC-seq and control libraries (using the MACS2 bdgcmp function).

Discussion

Levin hypothesized more than 40 years ago that phenotypic novelty in polyploids could arise from nuclear enlargement, which may affect surface-to-volume ratios and in turn influence how transcripts may pass through the nuclear pores into the cytoplasm (Levin 1983). We now have evidence that polyploidy can indeed induce changes in chromatin architecture, epigenetic regulation, and transcription (Zhou et al. 2021; Wang et al. 2021b; Han et al. 2022; Sun et al. 2022; Fang et al. 2024; Long et al. 2024), but most of these results come from studies on allopolyploid plants, which are further complicated by their hybrid origin (del Pozo and Ramirez-Parra 2015; Long et al. 2024). In contrast, autopolyploids arise within species and thus should reflect primarily the effects of genome duplication per se, without the added complication of hybridity (Parisod et al. 2010). Like allopolyploidy, autopolyploidy (within-species WGD) is also expected to be accompanied by major changes within the cell, as it also triggers a cell size and nuclear volume increase (del Pozo and Ramirez-Parra 2015; Robinson et al. 2018; Doyle and Coate 2019).

In addition to understanding the immediate response to genome duplication itself, it is also important to investigate what happens in the longer term evolution of a polyploid lineage to understand how it evolves to a “new normal.” This can help us address to what extent the changes that accompany genome duplication are challenges (in which case, we expect them to revert to at least some extent to the ancestral state) vs. opportunities (in which case, we expect them to be maintained or further elaborated). To do this, a comparison that includes diploids, neo-tetraploids, and evolved tetraploids within the same species is particularly informative (Ramsey and Schemske 2002; Maherali et al. 2009; Hegarty et al. 2013; Bomblies 2020). Here, we used the plant A. arenosa as a model system to investigate chromatin accessibility and transcription changes in a “three-way” comparison of diploids, neo-tetraploids, and evolved tetraploids. The focus on chromatin was motivated by prior evidence that chromatin remodeling and core transcription genes were under selection in evolved autotetraploid populations of A. arenosa (Hollister et al. 2012; Yant et al. 2013).

Using ATAC-seq and RNA-seq on rosette leaf and petal samples, we evaluated changes that occurred upon genome duplication by comparing diploids to neo-tetraploids derived from them (Transition 1), as well as the later evolution of a polyploid lineage by comparing the neo-tetraploids to established tetraploids (Transition 2). We found >8,000 dACRs across all our samples (including both rosette leaf and petal tissues), which clustered into groups exhibiting distinct patterns of accessibility variation across cytotypes. These changes differed among tissues. Interestingly, for >38% of rosette leaf dACRs and >70% of petal dACRs, accessibility changes that occurred in Transition 1 (the WGD response) were reversed in Transition 2 (subsequent evolution). The same trend was also observed in roughly 60% of rosette leaf and petal DEGs. That the genetically more similar diploid and neo-tetraploid differ more than the evolved tetraploid and diploid, indicates that many of the WGD-associated changes were perhaps detrimental, such that the subsequent evolutionary “adjustment” to the new polyploid state may often require a return toward the ancestral diploid state; we previously called this a Type A trend (Bomblies 2020).

A Type A trend could also arise, however, if the changes that arise in response to genome duplication are variable, such that the original neo-tetraploid that gave rise to the evolved tetraploids may not have had the same changes. Indeed, that there is not a true “ploidy response” in plants, and that there is lots of stochasticity in the response itself is supported by the different results observed for various studies in the same species, e.g. between intra-specific strains in A. thaliana (Yu et al. 2009, 2010; Pignatta et al. 2010; Pacey et al. 2022). In our study, we observed that variation among biological replicates of neo-tetraploids was minimal (supplementary fig. S8, Supplementary Material online). However, we believe that selection for reversion to the diploid-like state may in many cases be the cause of the Type A pattern, because the evolved tetraploids are (both in terms of dACRs and DEGs) more similar to the diploids despite their genetic divergence, than either is to the neo-tetraploid. This is also consistent with the aforementioned observation that chromatin remodelers show evidence of selection in the established tetraploid lineage hinting they may be responsible for some kind of evolutionary “re-tuning” of genome organization. On a practical level, these findings have important implications for interpreting studies that compare only diploids with either neo-tetraploids or evolved tetraploids but not both.

Because one might naively expect accessibility changes to correlate with expression changes, we were surprised to find that the genomic positions of dACRs and DEGs were largely non-overlapping. However, this study is not the first to observe such a pattern; a minimal overlap between differentially accessible regions and DEGs has also been observed previously in comparisons between A. thaliana accessions (Alexandre et al. 2018; Srikant et al. 2022). In a polyploidy context, one interpretation for this could be that differential chromatin accessibility may help maintain the same relative expression levels during genome doubling. Alternatively, it could also be that accessibility is not a major driver of WGD-associated expression changes at this evolutionary scale (since the A. arenosa autotetraploids arose recently, ∼30,000 years ago). In any case, one cannot be assumed to predict the other.

We observed that the most frequently occurring features in cis to dACRs were TEs (including repeats) and non-DEGs. Not only TEs are well known to be targeted by DNA methylation and other chromatin marks for silencing (Lippman et al. 2004; Underwood et al. 2017), which can in turn affect 3D chromatin interactions (He et al. 2024) but they are also strongly enriched in species-specific dACRs in plants (Lu et al. 2019). With respect to chromatin accessibility, we observed that dACRs with one or more TEs in cis in the 2X reference genome show high accessibility variation across cytotypes and significantly higher variation than dACRs that lack nearby TEs. Accessibility changes across cytotypes are also higher for DEGs and non-DEGs when they co-occur with TEs (Fig. 5a). While dACRs were enriched for certain TIR TE superfamilies (hAT, Mutator, PIF-Harbinger), it was the Gypsy and CACTA TIRs that were associated with particularly large accessibility changes. That dACRs near TEs are most variable in accessibility could be due to genetic or epigenetic changes at these TEs during polyploidization. While autopolyploid plants including A. arenosa are not known to exhibit transposition bursts in response to WGD, the Gypsy, CACTA, and hAT sequences show evidence of high-frequency insertions outside genic regions in autotetraploid A. arenosa (Baduel et al. 2019). The mobilization of these TEs (although previously identified using the A. lyrata reference genome) may indeed be associated with the large accessibility changes we observe, although this may be easier to explain for Class I TEs like the Gypsy LTRs (Fig. 5b). The systematic identification of TE sequence insertions and deletions across individuals [such as in Baduel et al. (2019)], using A. arenosa diploid and autotetraploid long-read genome references would greatly improve our understanding of the interplay between genetic and epigenetic mechanisms during WGD. Nevertheless, accessibility changes at Class II TEs like CACTA and hAT may also be due to residual epigenetic perturbations (such as DNA methylation) following their excision during polyploidization. Such a scenario is supported by evidence from rice neo-tetraploids, where DNA methylation levels at Class II TEs increase compared with diploid progenitors, and this may suppress the expression of proximal genes (Zhang et al. 2015).

It is notable that the vast majority of genes (74%; 13,534 out of the 18,226 with detectable expression in our experiment) were in fact non-DEGs for both rosette leaf and petal samples (<2-fold change in expression in any comparison across genotypes/cytotypes). Similar results have been reported in neo-tetraploid Rangpur lime (Allario et al. 2011), maize (Riddle et al. 2010) and potato (Stupar et al. 2007). These studies, and ours, suggest that “pure” ploidy changes may often have relatively subtle effects on gene expression. In comparison to these other studies, however, we detected more genes (1,967 rosette leaf DEGs and 3,665 petal DEGs) with strong expression changes across the three genotypes/cytotypes.

A few caveats bear mentioning: first, there may be some biases that arise because our reads were aligned to a reference genome derived from the diploid SNO population. This could affect calls of some dACRs and DEGs, especially in regions with structural or presence/absence variation (including regions near TE). In addition, our transcriptome analyses compare the relative expression levels of genes to the total transcriptome across cytotypes. However, we have not estimated the true transcriptome sizes of the neo-tetraploid and autotetraploid individuals (Coate and Doyle 2010, 2015; Doyle and Coate 2019). We therefore cannot rule out a scenario where there is an overall change in the total number of transcripts following WGD that we cannot detect since we have not normalized transcripts for cell number and biomass content (Doyle and Coate 2019; Coate 2023). Future RNA-seq experiments that include exogenous spike-in controls, or reverse transcription-quantitative polymerase chain reaction (RT-qPCR) of candidate genes after controlling for cell numbers or biomass, would enable accurate identification of differences in transcriptome sizes across cytotypes and scaling of RNA-seq derived read counts accordingly (Coate 2023). Nevertheless, even if transcriptome sizes were different, a large number of genes with balanced expression levels across cytotypes further supports genetic evidence (Baduel et al. 2019) that autopolyploid A. arenosa does not experience much of a “genomic shock” (McClintock 1984; Doyle and Coate 2019).

We noted that <4% of all DEGs and <3% of all genes in cis to dACRs overlapped with candidates from previous selection scans in autotetraploid A. arenosa [gene lists from Hollister et al. (2012), Yant et al. (2013), and Bohutínská et al. (2021), supplementary table S1, Supplementary Material online]. That most genes under selection are not themselves differentially regulated or differently accessible was not unexpected since most of them have signatures of selection primarily in coding rather than regulatory regions. This suggested that most of the observed chromatin architecture and transcriptional changes observed between neo-4X and Est-4X were independent of genetically fixed variants associated with cytotype. Aside from ploidy-differentiated chromatin remodelers and core transcription machinery genes that motivated this study, we noted that several DEGs and non-DEGs in cis to dACRs were also involved in chromatin remodeling (e.g. ATXR6, HDA2, SWI-3D, CENPC, H2B, HDA13, and CHR27) and transcription (e.g. POL2A, POL2B), indicating that chromatin accessibility changes may be responsible for influencing regulatory genes. Future experiments could be aimed at bridging the gap between these molecular phenotypes and the mechanisms driven by the differentiated genes.

Conclusion

This study shows evidence of the immediate and gradual cellular changes that accompany WGD, revealing new insights on the relationship between chromatin architecture and gene expression in polyploid genome stability, both in response to WGD and longer-term evolution. Strikingly, the genetically distinct 2X and Est-4X were more similar than the genetically identical, but cytologically distinct 2X and neo-4X, which supports the idea that many changes that arise after WGD represent challenges that are returned to the ancestral state over time. We show a link between TEs and accessibility variation, and that accessibility variation is only weakly associated with gene expression changes. Importantly, we show that several genes whose expression changes upon WGD return to more diploid-like levels in the evolved tetraploids. We hypothesize that the majority of these reflect that most WGD-associated changes may be detrimental and necessitate an evolutionary return to ancestral levels.

Supplementary Material

Supplementary material is available at Molecular Biology and Evolution online.

Author Contributions

T.S. and K.B. designed the study and wrote the manuscript. T.S. generated the various sequencing libraries, and A.G. generated the neo-tetraploid lines. T.S. analyzed the data. T.S., A.G., and K.B. reviewed and edited the final manuscript.

Funding

This work was supported by core funds from ETH Zürich and a postdoctoral fellowship grant from the Swiss National Science Foundation (217182; awarded to T.S.). This project has also received funding from the European Research Council Marie Skłodowska-Curie Actions (MSC) grant no. 101029732 (awarded to A.G.).

Data Availability

All sequencing datasets generated for this study are available through the European Nucleotide Archive (https://www.ebi.ac.uk/ena/browser/home) under accession numbers PRJEB77194/ERP161671 (ATAC-seq and gDNA-seq) and PRJEB77201/ERP161678 (RNA-seq). Seeds available on request: kirsten.bomblies@biol.ethz.ch.