# Source paper archive - Source title: GSNOR-dependent nitric oxide homeostasis shapes methylome persistence after multi-generational exposure to future climate conditions in Arabidopsis thaliana - Author(s): Behl, R., Thanh Vo, B., Ghirardo, A., Albert, A., Mattes, E., Goessl, L., Schnitzler, J.-P., Johannes, F., Lindermayr, C. - Source landing URL: https://www.biorxiv.org/content/10.64898/2026.11.09.765246v1 - PDF URL: https://www.biorxiv.org/content/10.64898/2026.11.09.765246v1.full.pdf - PDF SHA-256: feb3ae3e4f072309616d81d472f9a3594ee1aff6aeda667b0c3befc305d85199 - Converter: pymupdf4llm 0.0.27 - Limits: Automated PDF text extraction; reading order, equations, tables and formatting may be incomplete or incorrect. Images are omitted, and no OCR was performed. Markup and links are neutralized for safe downloading. Consult the original PDF for the authoritative version. --- [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.09.765246; this version posted November 11, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.09.765246) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY-NC 4.0 International license.](http://creativecommons.org/licenses/by-nc/4.0/) 1 **GSNOR-dependent nitric oxide homeostasis shapes methylome persistence after multi-** 2 **generational exposure to future climate conditions in** _**Arabidopsis thaliana**_ 3 **Running title: NO homeostasis and multi-generational effects on methylome** 4 Behl R [1], Thanh Vo B [2], Ghirardo A [1,3], Albert A [1], Mattes E [4], Gößl L [5], Schnitzler JP [1,3,5,*], 5 Johannes F [2,*], Lindermayr C [4,7,*] 6 Affiliations: 7 1. Research Unit Environmental Simulation (EUS), Helmholtz Munich, Neuherberg, 8 Germany 9 2. Plant Epigenomics, TUM School of Life Sciences, Technical University of Munich, 10 Freising, Germany. 11 3. Institute for Translational Metabolism Research, Helmholtz Diabetes Center, 12 Helmholtz Munich / School of Medicine and Health, Technical University of Munich, 13 Munich, Germany 14 4. Institute of Lung Health and Immunity (LHI), Comprehensive Pneumology Center 15 (CPC-M), Helmholtz Munich, Neuherberg, Germany. 16 5. Plant Genome and Systems Biology, Helmholtz Munich, Neuherberg, Germany 17 6. Ecosystem Physiology, Faculty of Environment and Natural Resources, University 18 Freiburg, Germany 19 7. TUM School of Life Sciences, Technical University Munich, Weihenstephan, 20 Germany. 21 22 *: Corresponding author 23 Abbreviations: 24 AI, artificial intelligence; ANOVA, analysis of variance; CTAB, cetyltrimethylammonium 25 bromide; DMR, differentially methylated region; G1–G5, generations 1–5; _gsnor-ko_, 26 GSNOR-deficient _gsnor1-3_ mutant; GSNOR, S-nitrosoglutathione reductase; LINE, long 27 interspersed nuclear element; LTR, long terminal repeat; PCA, principal component analysis; 28 PCR, polymerase chain reaction; RC, rolling circle; RdDM, RNA-directed DNA methylation; 29 RNA-seq, RNA sequencing; SINE, short interspersed nuclear element; TAIR10, The 30 Arabidopsis Information Resource 10 reference genome; TE, transposable element; WGBS, 31 whole-genome bisulfite sequencing; WT, wild type. 32 33 **PLAIN LANGUAGE SUMMARY (1000 characters)** 34 Plants can retain molecular traces of environmental stress, but it is unclear why some traces 35 disappear while others persist in later generations. We exposed wild-type _Arabidopsis_ plants 36 and a GSNOR-deficient mutant to drought, elevated carbon dioxide, ozone, heat, or their 37 combination for three generations. GSNOR regulates nitric oxide levels and has important 38 function in stress response. After stress exposure, plants were grown under control conditions 39 for two generations. The mutant has more differentially methylated regions than wild-type 40 plants even under control conditions, developed many more changes under stress conditions, 41 and retained a much larger proportion during recovery. Many retained changes occurred near 42 genes and mobile DNA elements and coincided with altered gene activity. Our results suggest 43 that GSNOR-dependent regulation of nitric oxide helps plants restore their DNA methylation 44 landscape after repeated environmental stress. [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.09.765246; this version posted November 11, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.09.765246) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY-NC 4.0 International license.](http://creativecommons.org/licenses/by-nc/4.0/) 45 46 47 **ABSTRACT (1500 characters)** 48 Environmental stress can modify plant DNA methylation, but the roles of nitric oxide 49 homeostasis in methylome recovery and multigenerational persistence remain unclear. We 50 used whole-genome bisulfite sequencing to compare _Arabidopsis thaliana_ wild type and the 51 GSNOR-deficient mutant ( _gsnor1-3_ ) exposed to drought, elevated CO 2, ozone, heat, or their 52 combination during generation G1 to G3, followed by two generations under control 53 conditions. Methylomes from G1, G3, and G5 were analysed for global divergence, 54 differentially methylated regions, genomic annotation, and methylation trajectories and were 55 integrated with matched RNA-seq and independent datasets. Control methylomes of _gsnor1-3_ 56 were already distinct from those of the wild type, and the mutant accumulated substantially 57 more DMRs, with a strong hypermethylation bias. In _gsnor1-3_, 44.1 to 45.7% of G1 CG 58 DMRs retained their direction through G3 and G5, compared with 0.09 to 1.24% in wild 59 type; persistent CHG and CHH DMRs were also markedly more abundant. Persistent regions 60 occurred at genes, promoters, and transposable elements and were associated with broader 61 expression responses in the mutant. Independent datasets supported a subset of these regions, 62 including directional agreement for 211 of 292 published GSNOR-associated DMRs. These 63 results indicate that GSNOR-dependent nitric oxide homeostasis contributes to methylome 64 stability and resetting after repeated environmental exposure. 65 **1.** **INTRODUCTION** 66 Plants are constantly subjected to environmental fluctuations that limit their growth, 67 reproduction, and survival (Joseph et al., 2025; Zhang et al., 2023). Due to anthropogenic 68 climate change, the frequency and intensity of high-temperature events have increased, and, 69 in many areas, so have agricultural and ecological droughts. This means that plants are 70 subjected to a range of stresses, including changes in atmospheric CO 2 concentration, 71 increases in tropospheric ozone, and other environmental disturbances (IPCC, 2023a, 2023b). 72 However, these different factors rarely occur in isolation, and plants may therefore be 73 exposed to more than one stress at a time, one after another or repeatedly. The way in which 74 plants respond to a combination of stresses cannot always be deduced from their responses to 75 each stress on its own. In _Arabidopsis thaliana_, for example, simultaneous drought and heat 76 stress led to a specific physiological and transcriptional response, involving hundreds of 77 transcripts not activated by either stress alone (Behl et al., 2026; Rizhsky et al., 2004). 78 Similar results have been found in studies involving sequential stress, in which a prior stress 79 has left detectable transcriptional marks and altered either the timing or the extent of the 80 response to a subsequent stress (Coolen et al., 2016). It is therefore essential to understand 81 how plants respond to repeated and interacting environmental conditions if we are to predict 82 their performance in future climates and identify the molecular processes that contribute to 83 resilience. 84 Plants undergo changes in transcription, along with changes in hormone signalling, 85 metabolism, redox regulation, growth, and resource distribution, in response to stress 86 (Fernandes & Ghag, 2022). Even though several of the responses decrease following the 87 removal of the stress, prior exposure can affect the rate, extent or direction of the response to [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.09.765246; this version posted November 11, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.09.765246) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY-NC 4.0 International license.](http://creativecommons.org/licenses/by-nc/4.0/) 88 a subsequent stress. This is generally referred to as stress priming or stress memory (Crisp et 89 al., 2016; Sani et al., 2013). For instance, repeated dehydration in Arabidopsis leads to the 90 formation of different categories of transcriptional memory genes, some of which are induced 91 more strongly, others less strongly, and in some cases show altered response patterns upon 92 later exposure (Ding et al., 2012; Ding et al., 2013). In certain cases involving dehydration93 memory gene loci, transcription returns to baseline levels during recovery, even though 94 elevated H3K4me3 and phosphorylated RNA polymerase II remain associated with the 95 genes, indicating that a molecular record of the previous stress can persist even after 96 transcription has been restored (Ding et al., 2012). Heat-stress memory is also linked to 97 sustained H3K4 methylation and the activity of specific heat-shock transcription factors, 98 indicating that chromatin states play a role in maintaining the transcriptional competence 99 induced by stress (Liu et al., 2018). However, it is not always beneficial to maintain a stress100 related state, since effective recovery also involves switching off the protective programmes 101 and restoring growth. Plant performance thus relies on a balance between molecular memory 102 and the resetting that occurs after stress is removed (Crisp et al., 2016). 103 Epigenetic mechanisms offer a possible molecular basis for this balance by altering 104 how genes are expressed. Cytosine DNA methylation is particularly significant in plants and 105 occurs in the CG, CHG, and CHH sequence contexts, with H representing A, C, or T (Abdul 106 Aziz & Masmoudi, 2026; Johannes & Schmitz, 2019; Law & Jacobsen, 2010). The pathways 107 responsible for establishing and maintaining these three contexts differ in part (Zhang et al., 108 2018). CG methylation is mainly kept up by METHYLTRANSFERASE 1, while CHG 109 methylation is mostly maintained by CHROMOMETHYLASE 3 (Cao & Jacobsen, 2002; 110 Kankel et al., 2003). The DOMAINS REARRANGED METHYLTRANSFERASE proteins 111 are involved in de novo methylation and play a central role in RNA-directed DNA 112 methylation, and CHROMOMETHYLASE 2 makes a major contribution to CHH 113 methylation in heterochromatin that is enriched in H3K9me2 (Cao & Jacobsen, 2002; Stroud 114 et al., 2014). Together, these pathways help to limit the activity of transposable elements 115 (TE), preserve genome stability, and create both transcriptionally permissive and repressive 116 chromatin environments (Liang et al., 2022; Stroud et al., 2014). Since CG and CHG 117 methylation occur at symmetric sites in the DNA sequence, it can be copied during DNA 118 replication, whereas asymmetric CHH methylation must be re-established in each cell 119 generation. Stress-induced methylation changes in these contexts may therefore vary in terms 120 of stability, location in the genome, and ability to last during recovery or from one generation 121 to the next (Boyko et al., 2010; Johannes & Schmitz, 2019; Law & Jacobsen, 2010; Sani et 122 al., 2013; Shahryary, Symeonidi, et al., 2020). 123 Environmental stresses can cause changes in DNA methylation at genes, promoters, 124 TEs, and other regulatory regions, but it is still unclear to what extent these changes reflect 125 stable stress memory. Some experiments have identified methylation states that are 126 responsive to stress and linked to either changes in gene expression or phenotypic changes in 127 the offspring. For instance, in _Arabidopsis thaliana_, repeated exposure to hyperosmotic stress 128 led to methylation changes at epigenetically labile sites and resulted in a short-term, mainly 129 maternally inherited memory that was widely reset following removal of the stress (Wibowo 130 et al., 2016, 2018). In other cases, DNA methylation and small RNA-dependent responses 131 have been observed in the direct offspring of stressed plants (Boyko et al., 2010). [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.09.765246; this version posted November 11, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.09.765246) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY-NC 4.0 International license.](http://creativecommons.org/licenses/by-nc/4.0/) 132 Redox signalling could be a significant yet underexplored factor in these methylome 133 dynamics. Nitric oxide is a redox-active signalling molecule that primarily alters protein 134 function via S-nitrosylation of cysteine residues. S-nitrosoglutathione reductase regulates this 135 signalling pathway by breaking down S-nitrosoglutathione and, as a result, influencing the 136 amount of cellular S-nitrosothiols and protein S-nitrosylation (Behl et al., 2026; Feechan et 137 al., 2005; Rudolf et al., 2021). When the GSNOR1 function is lost in _Arabidopsis,_ there is a 138 buildup and improper regulation of S-nitrosothiols, along with effects on plant development, 139 fertility, thermotolerance, hormone signalling, and immunity (Feechan et al., 2005; Lee et al., 140 2008; Shi et al., 2015). The _gsnor1-3_ null mutant is thus a genetically defined system for 141 investigating how disturbed nitric oxide homeostasis affects plant responses to repeated 142 environmental stress. 143 Some of the findings provide further evidence of a mechanistic link between nitric 144 oxide homeostasis and epigenetic regulation. Nitric oxide can affect histone acetylation by 145 modulating the activity of histone deacetylases, and GSNOR-dependent changes in nitric 146 oxide levels have been linked to genome-wide alterations in histone modifications and stress147 related transcription (Ageeva-Kieferle et al., 2021; Behl et al., 2026). The chromatin 148 responses that are dependent on HDA6 are affected by both GSNOR and nitric oxide, while 149 the S-nitrosylation of HDA19 controls its association with chromatin, its histone 150 deacetylation activity, and its role in tolerating oxidative stress. These findings show that 151 protein S-nitrosylation acts as a redox-sensing mechanism which can alter chromatin activity 152 (Ageeva-Kieferle et al., 2021; Lindermayr et al., 2020; Rudolf et al., 2021; Zheng et al., 153 2023). It remains largely unknown whether a disruption in nitric oxide homeostasis also 154 affects the establishment, maintenance, or resetting of stress-associated DNA methylation 155 patterns over several generations. 156 In this study, we performed Whole-Genome Bisulfite Sequencing (WGBS) of a 157 multigenerational experiment to examine how repeated exposure to climate change-related 158 stress and subsequent recovery periods affected the DNA methylomes of both wild-type and 159 _gsnor-ko_ individuals. Methylation responses in the CG, CHG, and CHH contexts were 160 analysed across generations, and differentially methylated regions (DMRs) linked to single 161 and combined environmental treatments were identified. We then separated the areas that 162 exhibited persistent hypermethylation or hypomethylation from those that reversed their 163 methylation direction or were no longer detectable after removal of the stress, and examined 164 how these regions were distributed among genes, promoters, TEs, and intergenic regions. We 165 proposed that the majority of methylation changes associated with stress would be reversed 166 after two generations under control conditions, while a smaller group would either persist or 167 switch in direction. We also suggested that disrupting GSNOR-dependent nitric oxide 168 homeostasis would modify both the degree and the genomic location of methylation 169 persistence and resetting. By investigating these changes genome-wide, the present study 170 addresses whether redox homeostasis contributes to the stability of stress-induced epigenomic 171 states across successive generations of exposure and recovery. 172 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.09.765246; this version posted November 11, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.09.765246) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY-NC 4.0 International license.](http://creativecommons.org/licenses/by-nc/4.0/) 173 **2.** **MATERIALS AND METHODS** 174 **2.1 Plant material and multigenerational climate-stress experiment** 175 The multigenerational climate-stress experiment was conducted as described in (Behl et al., 176 2026). Wild-type (WT) _Arabidopsis thaliana_ Col-0 and _gsnor-ko_ plants were grown under 177 control, drought, elevated CO 2, elevated ozone (O 3 ), warm climate (heat), and combined 178 treatments across five successive generations. Treatments were applied during the first three 179 generations (G1-G3). Later generations (G4-G5) were maintained under ambient control 180 conditions to assess recovery and transgenerational effects. Rosette leaves were collected 181 from 4.5-week-old plants between 09:00 and 11:00. For this study, leaf material from G1, 182 G3, and G5 was used for WGBS to investigate stress-induced, inherited, reset, and genotype183 dependent DNA methylation states (Supplemental Figure S1 and Supplemental File S1). 184 **2.2 Whole-Genome Bisulfite Sequencing** 185 Genomic DNA was extracted from 65–70 mg of ground leaf tissue using a CTAB-based 186 protocol (Healey et al., 2014). DNA quantity and quality were assessed using NanoDrop 187 spectrophotometry and Qubit fluorometry. DNA was sheared to an average fragment size of 188 about 350 bp using the Covaris R230 instrument. Libraries were prepared with NEBNext 189 Ultra II DNA Library Prep Kit (E7645S, NEB, USA). Bisulfite conversion was performed 190 using the EZ-96 DNA Methylation-Gold MagPrep Kit (D5042, Zymo Research, USA). 191 Fragmented DNA was end-repaired, A-tailed, ligated to methylated adapters, size-selected, 192 bisulfite converted, and PCR-amplified using KAPA HiFi HS Uracil+ polymerase (Roche; 193 Cat. No. 07959050001). 194 WGBS libraries were sequenced by Novogene (UK) Co., Ltd. on the NovaSeq X Plus 195 platform using 150 bp paired-end sequencing. Approximately 7 Gb of sequencing data were 196 generated per sample. Three biological replicates were used for the control samples, and two 197 for each treatment in G1, G3, and G5. 198 **2.3 WGBS read processing and methylation calling** 199 WGBS reads were processed using the MethylStar pipeline, which integrates adapter and 200 quality trimming with Trimmomatic, bisulfite-aware alignment with Bismark, and cytosine201 level methylation calling with Methimpute, as described previously (Shahryary, Hazarika, et 202 al., 2020; Taudt et al., 2018). Reads were aligned to the _Arabidopsis thaliana_ TAIR10 203 reference genome. DNA methylation was analysed separately in CG, CHG, and CHH 204 sequence contexts. 205 Differentially methylated regions (DMRs) were identified using jDMRgrid (Behl et 206 al., 2025; Hazarika et al., 2022). The genome was partitioned into 200-bp sliding windows 207 with a 50-bp step size. Only windows containing at least 10 cytosines in the respective 208 methylation context were retained. Methylation region calls were generated for each sample 209 and used to construct region-call matrices for pairwise comparisons between control and 210 stress conditions. Non-epipolymorphic patterns were removed, replicate consensus filtering 211 was applied, and adjacent significant windows were merged to obtain final DMRs. 212 DMRs representing treatment responses were identified by comparing each treatment 213 with the WT control of the respective generation. Direct genotype contrasts were additionally [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.09.765246; this version posted November 11, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.09.765246) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY-NC 4.0 International license.](http://creativecommons.org/licenses/by-nc/4.0/) 214 performed by comparing _gsnor-ko_ and WT under the same treatment and generation. DMRs 215 were classified as hypermethylated or hypomethylated relative to the WT control. 216 **2.4 Global methylation divergence and AlphaBeta analysis** 217 Genome-wide methylation was analysed separately for each sequence context using 218 biological-replicate methylation estimates. Within each generation, a two-factor ANOVA 219 was fitted with genotype, treatment and their interaction as fixed effects. Post-hoc 220 comparisons were performed using the Dunnett test, with p-values adjusted using the 221 Benjamini–Hochberg procedure. 222 To evaluate whether methylation changes accumulated across generations and to 223 quantify methylation stability, spontaneous CG epimutation rates were estimated for WT and 224 _gsnor-ko_ using the AlphaBeta R package (v1.10.0) (Shahryary, Symeonidi, et al., 2020). 225 Following published approaches (Hazarika et al., 2022; van der Graaf et al., 2015), CG 226 methylation gain ( α ) and loss ( β ) rates were inferred from multigenerational WGBS data 227 using known pedigree structures. Confidence intervals were computed as implemented in 228 AlphaBeta. 229 **2.5 Annotation of DMRs** 230 To assess the fate of stress-associated methylation changes across generations, overlapping 231 DMRs from G1, G3, and G5 were merged into consensus genomic regions. For each 232 consensus region, methylation state was assigned in each generation as hypermethylated, 233 hypomethylated, mixed/both, or absent. Based on the G1–G3–G5 methylation-state pattern, 234 each region was classified into trajectory classes, including acute/generation-specific, stress235 maintained-then-reset, persistent, recurrent, G3/G5-emergent, and switcher DMRs. These 236 trajectory classes were used to distinguish methylation changes that were reset after stress 237 withdrawal from those that persisted, reappeared, or changed direction across generations. 238 Gene, TE, and genome annotation files were obtained from TAIR10. Promoters were 239 defined as 1 kb upstream of annotated gene transcription start sites. DMRs were annotated 240 according to their overlap with gene bodies, promoters, TEs, and intergenic regions. DMR241 associated genes and TEs were extracted for downstream analyses of overlap, trajectories, 242 and enrichment. Where DMRs overlapped more than one annotation class, multiple 243 annotation labels were retained for downstream feature-specific summaries. 244 **2.6 TEs and genes in their proximity** 245 To determine whether persistent methylation marks were preferentially associated with 246 specific TE classes, TE-overlapping DMRs were mapped to TAIR10 TE family and 247 superfamily annotations. TE superfamilies included, where present, Helitron, Gypsy, Copia, 248 MuDR, HAT, En-Spm, Mariner/Tc1, LINE, and SINE classes. Persistent DMR-associated 249 TEs were compared against all DMR-associated TEs and, separately, against all annotated 250 TAIR10 TEs. 251 To assess potential regulatory relationships between TE-associated methylation 252 changes and nearby gene expression, TE-DMRs were paired with genes located within 253 approximately ±1.2 kb (Behl et al., 2025). This distance threshold was selected to focus on 254 the closest quartile of TE–gene pairs. For each TE–gene pair, the TE-DMR methylation state [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.09.765246; this version posted November 11, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.09.765246) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY-NC 4.0 International license.](http://creativecommons.org/licenses/by-nc/4.0/) 255 and the corresponding gene expression state were combined to define classes: TE 256 hypermethylation with gene downregulation; TE hypermethylation with gene upregulation; 257 TE hypomethylation with gene upregulation; TE hypomethylation with gene downregulation; 258 and TE hyper/hypomethylation with unchanged gene expression. Persistent TE-DMRs and 259 switcher TE-DMRs were also compared with gene-expression trajectories across G1, G3, and 260 G5. These analyses were used to identify candidate TE-linked regulatory relationships and 261 were interpreted as associations rather than direct evidence of causality. 262 **2.7 Candidate regulatory gene and Integration with WGBS and RNA-seq data from the** 263 **previous manuscripts** 264 RNA-seq results from the same experimental system were obtained from our previous study 265 (Behl et al., 2026). Because the RNA-seq and WGBS datasets were generated from the same 266 genotypes, stress treatments, and generations, RNA-seq results were integrated with DMRs 267 and gene annotations to assess methylation–expression relationships. Additionally, previously 268 published environmental stress-associated DMRs were compared with DMRs identified in 269 this study to establish a link between stress-induced DMRs and their inheritance (Behl et al., 270 2025; Rudolf et al., 2021). 271 **2.8 Statistical analysis and visualization** 272 All statistical analyses were performed in R/RStudio. Data processing and visualisation were 273 performed using R packages including data.table, dplyr, tidyr, ggplot2, GenomicRanges, 274 IRanges, rtracklayer, and ggalluvial. Final figure assembly and graphical adjustments were 275 performed in Inkscape. 276 **2.9 AI usage in Data analysis** 277 For data analysis, an AI tool (ChatGPT 5.6) was used for programming in R 278 (2024.12.1+563). After using this tool/service, the author(s) reviewed and edited the codes as 279 needed and take(s) full responsibility for the content of the published article. 280 **3.** **RESULTS** 281 We investigated genome-wide DNA methylation dynamics in WT and _gsnor-ko_ during 282 repeated environmental exposure in G1–G3 followed by two generations of recovery in G4 283 and G5. The experimental design and growth conditions are described in Methods and 284 Supplemental Figure S1. 285 **3.1 Loss of GSNOR increases DMR abundance and promotes hypermethylation across** 286 **generations** 287 We examined the methylomes of control samples to determine whether genotype-dependent 288 differences were present in the absence of environmental treatment. Principal Component 289 Analysis (PCA) separated WT and _gsnor-ko_ control samples in all three methylation contexts 290 (Supplemental Figure S2). The biological replicates were generally grouped by genotype, 291 especially in the CG context, indicating consistent methylation profiles. The separation 292 between WT and _gsnor-ko_ controls demonstrated that loss of GSNOR was associated with an 293 altered baseline methylome prior to treatment-dependent changes. [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.09.765246; this version posted November 11, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.09.765246) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY-NC 4.0 International license.](http://creativecommons.org/licenses/by-nc/4.0/) 294 Genome-wide average methylation differed among methylation contexts, generations, 295 genotypes, and treatments (Supplemental Figure S3). Absolute changes in average 296 methylation were relatively modest compared with the regional differences identified by 297 DMR analysis. Nevertheless, _gsnor-ko_ generally displayed higher methylation than WT 298 across several treatment conditions, with the clearest genotype-associated differences 299 observed in G3. Interestingly, differences remained detectable in G5 across several treatment 300 lineages, but they were not significant. These results suggested that the absence of GSNOR 301 primarily affected regional methylation patterns and resulted in measurable changes in 302 genome-wide methylation levels. 303 WT treatments exhibited ~1k- ~3k DMRs across all three sequence contexts. By 304 contrast, tens of thousands of DMRs were detected in _gsnor-ko_, with the greatest absolute 305 numbers occurring in the CG context (~30K). A large number of DMRs (from ~1k in CHG to 306 25k in CG) were already present in the comparison between _gsnor-ko_ and WT controls, 307 demonstrating substantial baseline methylome divergence between the genotypes. Most 308 environmental treatments further increased the regional methylation differences in _gsnor-ko_ . 309 The direction of methylation change also differed strongly between the genotypes. Although 310 both hyper- and hypomethylated DMRs were detected in _gsnor-ko_, hypermethylated regions 311 predominated across treatments and generations (Figure 1; Supplemental Figure S4; 312 Supplemental File S2). The largest absolute hyper-DMR burden occurred in the CG context. 313 In contrast, the strongest relative bias toward hypermethylation was observed in CHG and 314 CHH, where comparatively few hypo-DMRs (~1k) were detected. 315 Consistent with these patterns, the hyper-hypo DMR ratio was consistently greater in 316 _gsnor-ko_ than in WT (Figure 2; Supplemental File S2). In _gsnor-ko_, the ratio remained above 317 1 across the conditions examined, indicating an overall predominance of hypermethylation. 318 In WT, CG hyper:hypo ratios were commonly below or near 1 in G1 and G3, indicating 319 similar numbers of hyper- and hypo-DMRs or a modest predominance of hypomethylation. 320 The CG ratio subsequently increased in several WT treatment lineages by G5. WT CHG and 321 CHH ratios were generally lower than those of _gsnor-ko_ and showed more treatment322 dependent trajectories. For example, CHH ratios increased at G3 under heat and combination 323 treatments but declined again by G5. Therefore, WT showed context- and treatment324 dependent changes during recovery, whereas the methylation response of _gsnor-ko_ was 325 characterised by a stronger and more persistent hypermethylation bias. 326 Although _gsnor-ko_ Control already exhibited substantial baseline methylome 327 divergence from WT, the environmental-treatment lineages generally contained more DMRs 328 (~1.5-2k) than the mutant control comparison, indicating additional treatment-associated 329 remodelling superimposed on the genotype-associated background (Figure 1). Hyper-DMRs 330 shared across G1, G3 and G5 represented 39.9–42.8% of CG, 8.4–9.4% of CHG and 3.1– 331 3.4% of CHH DMR unions in _gsnor-ko_, compared with 0–0.7%, 0–0.1% and 0–0.1%, 332 respectively, in WT (Supplemental Figures S5–S7). Generation-specific DMRs were also 333 detected in both genotypes, indicating that cross-generation retention occurred alongside 334 continued methylome remodelling. Similarly, the hyper-DMR core shared among the 335 displayed treatment sets represented 34.6–47.5% of CG, 14.0–18.0% of CHG and 8.2–9.2% 336 of CHH DMR unions in _gsnor-ko_, compared with 0.5–0.9%, 0.7–1.2% and 0.9–1.5%, [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.09.765246; this version posted November 11, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.09.765246) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY-NC 4.0 International license.](http://creativecommons.org/licenses/by-nc/4.0/) 337 respectively, in WT (Supplemental Figures S8–S10). These intersections indicate a large 338 shared mutant component together with additional treatment- and generation-specific changes 339 DMR localisation remained strongly context dependent: CG DMRs were most 340 frequently associated with gene bodies, whereas CHG and CHH DMRs occurred 341 predominantly in TE-associated regions (Supplemental Figures S11–S13). Although this 342 broad context-specific pattern was conserved, the relative representation of gene-body- and 343 TE-associated DMRs differed quantitatively between genotypes. Overall, loss of GSNOR 344 was associated with baseline methylome divergence, an additional treatment-associated DMR 345 burden, a pronounced hypermethylation bias and reduced methylome resetting across 346 generations. 347 **3.2 Methylation divergence and DMR persistence differ between WT and** _**gsnor-ko**_ 348 We next used AlphaBeta analysis to determine whether methylation divergence accumulated 349 differently between genotypes and treatments. In the CG context, fitted divergence 350 trajectories were consistently higher in _gsnor-ko_ than in WT (Figure 3A). Nominally 351 significant treatment effects were detected within both genotypes, while comparisons 352 between genotypes identified differences under control, CO �, O �, and heat conditions 353 (Supplemental Table S1). CHG methylation dynamics also differed among selected 354 treatments and genotypes, with the strongest genotype difference occurring under the 355 combination treatment. By contrast, no AlphaBeta comparison was significant in the CHH 356 context (Supplemental Figure S14; Supplemental Table S1; Supplemental File S3). Thus, the 357 loss of GSNOR predominantly affected the dynamics of CG- and CHG-methylation 358 divergence. Full parameter estimates and _P_ values are provided in Supplemental Table S1. 359 We subsequently classified DMRs according to their G1–G3–G5 trajectories. A 360 persistent DMR was defined as a region that remained differentially methylated in the same 361 direction in all three generations, either hyper–hyper–hyper or hypo–hypo–hypo. A switcher 362 was present as a DMR in all three generations but changed direction at least once. To 363 distinguish treatment-associated persistence from the baseline effect of the mutation, WT 364 treatments and _gsnor-ko_ control and treatments were compared with WT Control at each 365 generation. WT contained 3–37 persistent CG DMRs across the five treatments, compared 366 with 17–74 in _gsnor-ko_ (Figure 3B; Supplemental File S4). Persistent CHG DMRs ranged 367 from 0–5 in WT and 1–12 in _gsnor-ko_, while persistent CHH DMRs ranged from 3–8 in WT 368 and 0–5 in _gsnor-ko_ . By contrast, the separate baseline comparison of _gsnor-ko_ Control with 369 WT Control identified 18,593 persistent CG, 1,478 CHG, and 693 CHH DMRs. Thus, the 370 very large persistent DMR population in _gsnor-ko_ primarily represented a stable baseline 371 genotype difference rather than treatment-specific inheritance. 372 Because the initial number of treatment-associated DMRs differed among genotypes, 373 treatments, and sequence contexts, persistence was also expressed as the percentage of G1 374 DMRs that retained the same methylation direction in G3 and G5. In the CG context, 0.9– 375 4.5% of G1 DMRs persisted in _gsnor-ko_, compared with 0.1–1.2% in WT. Persistence was 376 substantially lower in the non-CG contexts, reaching a maximum of 0.6% for CHG DMRs in 377 _gsnor-ko_ and approximately 0.1% for CHG DMRs in WT and for CHH DMRs in both 378 genotypes. Therefore, _gsnor-ko_ showed greater treatment-specific retention mainly in the CG 379 context, but most treatment-associated methylation changes were not maintained across all [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.09.765246; this version posted November 11, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.09.765246) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY-NC 4.0 International license.](http://creativecommons.org/licenses/by-nc/4.0/) 380 three generations. These results do not indicate a general loss of methylome flexibility in 381 _gsnor-ko_ ; instead, they distinguish a highly persistent baseline genotype signature from a 382 smaller increase in the retention of treatment-associated CG DMRs. 383 Switcher DMRs were fewer than persistent DMRs in _gsnor-ko_ . Depending on 384 treatment and context, only 0–15 DMRs remained present across all three generations. WT 385 displayed a few switcher DMRs, although in CHG and CHH their numbers occasionally 386 equaled or exceeded the small number of persistent regions. The alluvial plots illustrated 387 these trajectories and showed that same-direction retention represented a major component of 388 the _gsnor-ko_ DMR population, whereas flows between hypermethylated, hypomethylated, 389 and absent states accounted for the remaining dynamic regions (Supplemental Figure S16A390 C; Supplemental File S4). 391 The genomic distribution of persistent DMRs was context-dependent (Supplemental 392 Figures S17 and S18). Persistent CG DMRs most frequently overlapped gene bodies, whereas 393 persistent CHG and CHH DMRs were primarily associated with TEs. Promoter-associated 394 and intergenic DMRs constituted smaller fractions of the persistent sets. Switcher DMRs 395 showed more variable annotation profiles, consistent with their low abundance, but CHG and 396 CHH switchers were also frequently associated with TEs (Supplemental Figures S17 and 397 S19). These annotation patterns were broadly similar between genotypes, indicating that the 398 major difference between WT and _gsnor-ko_ was the extent of DMR retention rather than a 399 complete redistribution among genomic features. 400 **3.3 Persistent and switching DMRs highlight genes associated with stress and epigenetic** 401 **regulation** 402 Candidate genes associated with persistent DMRs were organised into functional groups 403 related to epigenetic and small-RNA regulation, DNA methylation and chromatin 404 organisation, redox and stress signalling, hormone, calcium and nutrient signalling, and 405 reproduction or transgenerational regulation (Figure 4A; Supplemental File S5). Persistent 406 promoter- and gene-body DMRs occurred across all these functional classes and included 407 both hypermethylated and hypomethylated regions. These associations were detected 408 predominantly in _gsnor-ko_, frequently across multiple treatments, whereas only a limited 409 number of WT-specific persistent DMRs were identified. Their retention in the same 410 methylation direction across G1, G3, and G5 indicates that methylation changes affecting 411 genes involved in stress responses and epigenetic regulation remained detectable after 412 treatment withdrawal. 413 Switcher DMRs were associated with genes belonging to functional classes related to 414 epigenetic and small-RNA regulation, redox and stress signalling, hormone, calcium and 415 nutrient signalling, and reproduction or transgenerational regulation. These patterns occurred 416 in both genotypes, although switcher-associated candidates were more widely distributed 417 across treatments in _gsnor-ko_ (Figure 4B; Supplemental Figure S20; Supplemental File S5). 418 **3.4 DMRs correlating with altered gene expression** 419 We next performed transcriptome-guided cross-omics validation using the same epigenetic420 regulator genes previously identified in our RNA-seq study, which showed transgenerational 421 expression differences between WT and _gsnor-ko_ (Behl et al., 2026). Gene-associated DMRs [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.09.765246; this version posted November 11, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.09.765246) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY-NC 4.0 International license.](http://creativecommons.org/licenses/by-nc/4.0/) 422 were detected for many of these independently defined RNA-seq candidates. WT generally 423 displayed fewer and more treatment- or generation-restricted methylation changes. By 424 contrast, _gsnor-ko_ showed a denser distribution of promoter- and gene-body DMRs across 425 treatments and generations, including hypermethylated, hypomethylated, and mixed 426 methylation states. Particularly extensive methylation differences were observed among 427 genes associated with DNA methylation and silencing, Polycomb regulation, histone and 428 chromatin organisation, RNA methylation, and methyl-donor metabolism (Supplemental 429 Figure S21). This demonstrates that genes differing transgenerationally between WT and 430 _gsnor-ko_ at the transcript level are frequently associated with local methylation differences in 431 the corresponding multigenerational WGBS dataset. This convergence provides cross-omics 432 support for an association between GSNOR-dependent transcriptomic reconfiguration and 433 altered methylation at epigenetic regulatory genes. However, because the analysis is 434 correlative, it does not establish that the observed DMRs directly caused the corresponding 435 expression differences. 436 To further determine whether methylation changes at the selected candidate genes 437 coincided with transcriptional responses, the WGBS results were integrated with RNA-seq 438 data. Methylation and expression measurements were matched by genotype, treatment, and 439 generation. Integration of persistent DMRs with RNA-seq responses revealed marked 440 differences between WT and _gsnor-ko_ (Figure 5A; Supplemental File S6). In WT, candidates 441 generally displayed fewer significant expression responses, consistent with the comparatively 442 limited occurrence of persistent candidate-associated DMRs. By contrast, a larger proportion 443 of the persistent candidates in _gsnor-ko_ showed altered expression in one or more treatment– 444 generation combinations. These expression responses were observed at genes carrying 445 persistent promoter DMRs, gene-body DMRs, or DMRs overlapping both genomic features. 446 Several candidates showed expression responses across multiple generations or treatments, 447 whereas others exhibited expression changes restricted to a particular environmental 448 condition or generation. Switcher candidates also displayed genotype- and condition449 dependent transcriptional responses (Figure 5B; Supplemental File S6). Significant 450 expression changes were detected for a subset of switcher genes, particularly in _gsnor-ko_, 451 although the responses were generally more heterogeneous than those associated with 452 persistent DMRs. Changes in methylation direction across G1, G3, and G5 were not 453 invariably accompanied by corresponding reversals in gene-expression direction. Instead, 454 some switcher candidates exhibited generation-specific expression responses, while others 455 retained similar expression directions despite transitions between hypermethylated and 456 hypomethylated states. 457 **3.5 Persistent differential TE methylation is associated with altered expression of** 458 **nearby genes** 459 Persistent DMRs within TEs were examined separately from gene-associated DMRs to 460 determine whether stable TE methylation changes coincided with transcriptional responses in 461 neighbouring genes. TEs were classified according to their annotated TE class or 462 superfamily, including DNA/MuDR, RC/Helitron, LTR/Copia, and LTR/Gypsy elements 463 (Figure 6; Supplemental File S7). Persistent TEs (DMRs) were detected predominantly in 464 _gsnor-ko_, whereas corresponding persistent TE methylation was limited or absent in WT for [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.09.765246; this version posted November 11, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.09.765246) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY-NC 4.0 International license.](http://creativecommons.org/licenses/by-nc/4.0/) 465 most candidates. In _gsnor-ko_, several TE loci retained the same hypermethylated or 466 hypomethylated state across G1, G3, and G5 under one or more treatment conditions. The 467 distribution of these persistent states varied across TE classes and treatments, indicating that 468 TE methylation retention was locus- and condition-dependent rather than uniform. Several 469 genes located near persistently methylated TEs showed altered expression, with the majority 470 of these responses occurring in _gsnor-ko_ . However, significant expression differences were 471 not detected for every persistently methylated TE–gene pair. The direction of nearby-gene 472 expression did not show a uniform relationship with the methylation state of the associated 473 TE. Persistent TE hypermethylation coincided with reduced expression for some nearby 474 genes but with increased, unchanged, or statistically nonsignificant expression for others. 475 Expression responses also varied among treatments and generations, even when the 476 corresponding TE retained the same methylation direction. These patterns suggest that the 477 relationship between TE methylation and nearby-gene expression depends on the individual 478 TE–gene pair and its genomic and environmental context. 479 **4.** **DISCUSSION** 480 This study shows that GSNOR-dependent nitric oxide homeostasis is associated with both the 481 magnitude and the persistence of DNA methylation variation during repeated environmental 482 exposure and subsequent recovery. The main findings are that loss of GSNOR produced a 483 strongly divergent baseline methylome, greatly increased the number of DMRs, favoured 484 hypermethylation, and reduced the resetting of regional methylation states after treatment 485 withdrawal. 486 487 **4.1 Loss of GSNOR shifts the methylome towards regional hypermethylation** 488 The separation of WT and _gsnor-ko_ control methylomes, along with the large number of 489 DMRs already present under control conditions, demonstrates that GSNOR has a substantial 490 genotype-dependent effect on the methylome. This finding is consistent with earlier evidence 491 that GSNOR1 contributes to chromatin accessibility, DNA demethylation, and the expression 492 of TEs and stress-responsive genes (Rudolf et al., 2021). _Rudolf et al._ linked GSNOR1 493 activity to the methylation index and to H3K9me2 abundance, providing a mechanistic 494 precedent for altered methylation in _gsnor-ko_ . Other studies have shown that nitric oxide 495 regulates histone acetylation through GSNOR- and HDA6-dependent processes and that S496 nitrosylation can directly modify the activity and activities of HDACs (Ageeva-Kieferle et 497 al., 2021; Mengel et al., 2017; Zheng et al., 2023). These studies established a connection 498 between nitric oxide homeostasis and chromatin regulation, but they did not determine how 499 loss of GSNOR affects DNA methylation during repeated stress and recovery across 500 generations. 501 Our results extend this literature in three respects. First, the effect of GSNOR loss was 502 not restricted to a particular stress: a high DMR burden was detected in the mutant control 503 and across drought, elevated CO 2, O 3, heat, and combined-treatment lineages. Second, 504 differences in average methylation were comparatively modest, whereas tens of thousands of 505 DMRs were detected in _gsnor-ko_ . This distinction is important because opposing or spatially 506 restricted methylation changes can produce limited changes in the genome-wide mean while 507 extensively remodelling regulatory regions. Third, _gsnor-ko_ showed a persistent bias towards [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.09.765246; this version posted November 11, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.09.765246) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY-NC 4.0 International license.](http://creativecommons.org/licenses/by-nc/4.0/) 508 hypermethylation, especially as a high absolute number of CG hyper-DMRs and a marked 509 hyper: hypo imbalance in CHG and CHH. These results demonstrate that loss of GSNOR 510 alters the balance of methylation states but do not establish whether this phenotype results 511 from reduced demethylation, increased methylation establishment or maintenance, or another 512 indirect mechanism, as also reported by (Rudolf et al., 2021). 513 The genomic distribution of the DMRs followed the established architecture of the 514 Arabidopsis methylome. CG DMRs were frequently associated with gene bodies, whereas 515 CHG and CHH DMRs were concentrated at TEs. CG, CHG, and CHH methylation are 516 maintained by partly distinct pathways, and non-CG methylation is particularly important for 517 heterochromatin and TE silencing (Johannes & Schmitz, 2019; Law & Jacobsen, 2010; 518 Stroud et al., 2014; Zhang et al., 2018). Although the proportions of gene- and TE-associated 519 DMRs differed between WT and _gsnor-ko_, both genotypes showed the same general context520 dependent pattern. Thus, loss of GSNOR mainly affected the number and direction of DMRs 521 rather than producing a completely different distribution among genes, TEs, promoters, and 522 intergenic regions. 523 524 **4.2 GSNOR is associated with methylation divergence and resetting after stress** 525 **withdrawal** 526 The AlphaBeta analysis and the region-level trajectory analysis provide complementary 527 views of methylome dynamics. The consistently higher CG divergence fitted for _gsnor-ko_, 528 together with the significant difference between the WT and mutant controls, indicates that 529 altered methylation dynamics pre-existed the treatment-dependent responses. Significant 530 effects of CO 2, O 3, and heat relative to the _gsnor-ko_ control further show that environmental 531 conditions modified this altered baseline. The absence of a significant difference between 532 _gsnor-ko_ Combination and _gsnor-ko_ Control does not mean that the combined treatment had 533 no effect on the methylome; probably, already divergent mutant control reduced the 534 additional contrast attributable to the combined treatment. By comparison, WT Combination 535 differed from WT Control for the methylation-gain parameter, indicating that the same 536 environmental regime was superimposed on a less divergent baseline in WT. 537 Previous studies have reported conflicting evidence regarding the stability of stress538 associated methylation. Hyperosmotic priming can generate locus-specific chromatin and 539 DNA methylation states that influence later responses, but much of this memory is transient 540 or extensively reset when stress is removed (Sani et al., 2013; Wibowo et al., 2016, 2018). 541 Repeated drought experiments have found few conserved stress-induced DMRs and have 542 attributed much of the apparent transgenerational variation to stochastic or pre-existing 543 differences, particularly in repetitive regions (Ganguly et al., 2017). Other studies have 544 likewise concluded that transgenerational stress memory is not a universal response in 545 Arabidopsis (Pecinka et al., 2009). These findings emphasise that persistence depends on 546 genotype, stress regime, tissue, developmental stage, and the number of untreated generations 547 used to test recovery. 548 In the present study, the marked contrast between WT and _gsnor-ko_ is an important 549 finding. Only a very small fraction of WT G1 DMRs retained the same direction through G3 550 and G5, consistent with efficient resetting and with earlier reports of a generally stable WT 551 methylome. In _gsnor-ko_, however, approximately 44–46% of G1 CG DMRs, 17–21% of [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.09.765246; this version posted November 11, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.09.765246) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY-NC 4.0 International license.](http://creativecommons.org/licenses/by-nc/4.0/) 552 CHG DMRs, and 8–10% of CHH DMRs remained in the same direction across G1, G3 and 553 G5. Persistence, therefore, remained much greater in the mutant even after normalisation to 554 its larger G1 DMR population. Because G4 and G5 were grown under control conditions, the 555 G5 states cannot be explained by continued direct exposure. They instead represent 556 methylation differences retained within the descendant lineage after two generations of 557 recovery. On the other hand, switcher DMRs were rare, indicating that complete reversals 558 between hypermethylated and hypomethylated states were not the dominant outcome. The 559 combination of extensive same-direction retention and a smaller set of directional switches 560 suggests that loss of GSNOR changes both the capacity to erase methylation differences and 561 the route by which some loci are remodelled during recovery. This provides a genome-wide 562 extension of the known role of nitric oxide as a regulator of chromatin state and supports a 563 model in which balanced S-nitrosothiol signalling contributes to epigenomic homeostasis 564 (Ageeva-Kieferle et al., 2021; Lindermayr et al., 2020). 565 566 **4.3 Persistent and switching DMRs implicate interconnected regulatory pathways** 567 We identified genes that overlapped with persistent or switching DMRs and grouped them by 568 their known biological functions. These groups included small-RNA pathways, DNA 569 methylation and chromatin organisation, redox and stress signalling, hormone responses, and 570 plant development. This literature-based classification was used to identify biologically 571 relevant candidates and should not be interpreted as a statistical enrichment analysis. 572 Persistent DMRs associated with _CLSY3, FDM3, and IDN2_ are notable because these 573 genes function in or are connected to RNA-directed DNA methylation, while _CMT3, IBM1,_ 574 _ROS1_, and _SUVH6_ participate in the establishment, interpretation, or removal of repressive 575 methylation and chromatin states. The presence of persistent promoter- or gene-body DMRs 576 at these loci suggests feedback between altered methylation and the machinery that regulates 577 it. However, a DMR associated with a methylation regulator does not by itself demonstrate 578 altered protein activity or pathway output, and this interpretation must remain a hypothesis 579 until tested experimentally. The persistent candidates also connected methylation dynamics 580 with pathways expected to be sensitive to altered nitric oxide homeostasis. Genes related to 581 redox balance, autophagy, photosynthetic electron transfer, ABA and auxin signalling, 582 calcium signalling, and reproductive development carried persistent DMRs predominantly in 583 _gsnor-ko_ . GSNOR is already known to influence development, fertility, thermotolerance, 584 auxin signalling, and stress responses (Feechan et al., 2005; Lee et al., 2008; Shi et al., 2015). 585 Our results do not show that methylation causes these phenotypes, but they identify stable 586 local epigenomic states at genes positioned within the same biological processes. 587 Switcher DMRs provided a complementary set of candidates whose methylation 588 direction changed across generations. These candidates were associated mainly with 589 epigenetic regulation, stress responses, and plant development, suggesting that some loci 590 continued to be remodelled during the transition from repeated treatment to recovery. As with 591 the persistent DMRs, these functional associations are descriptive and require experimental 592 validation. 593 594 **4.4 Methylation–expression relationships are locus- and context-dependent** [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.09.765246; this version posted November 11, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.09.765246) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY-NC 4.0 International license.](http://creativecommons.org/licenses/by-nc/4.0/) 595 Integration of the WGBS and RNA-seq datasets showed that more persistent and switcher 596 candidates exhibited altered expression in _gsnor-ko_ compared with WT. Because methylation 597 and expression were matched by genotype, treatment, and generation, these analyses resolve 598 associations that would be obscured by comparisons across unmatched conditions. Persistent 599 DMRs were often accompanied by expression changes in one or more treatment–generation 600 combinations, whereas switcher candidates tended to show more heterogeneous responses. 601 The broader transcriptional response in _gsnor-ko_ is consistent with our previous finding that 602 loss of GSNOR produced stronger persistence or reconfiguration of stress-associated 603 expression states during recovery (Behl et al., 2026). This convergence supports an 604 association between GSNOR-dependent transcriptome reconfiguration and local methylation 605 differences at epigenetic regulators. It does not establish whether methylation preceded 606 expression change, resulted from altered transcription or chromatin state, or participated in a 607 reciprocal feedback loop. There was, however, no universal inverse relationship between 608 methylation and expression. Promoter hypermethylation can restrict transcription in some 609 settings, but gene-body methylation has more variable associations with transcription, and the 610 effect of a DMR depends on its position relative to regulatory elements, its sequence context, 611 the surrounding chromatin, and the cell types contributing to the bulk tissue (Law & 612 Jacobsen, 2010; Zhang et al., 2018). Consistent with this complexity, hypermethylation in our 613 dataset was associated with decreased, increased, unchanged, or non-significant expression, 614 depending on the locus and condition; hypomethylation was similarly not synonymous with 615 activation. Ganguly et al. (2017) likewise found that drought-associated methylation variants 616 did not show a general relationship with drought-responsive expression. Conversely, Wibowo 617 et al _._ (2016) demonstrated at the CNI1 locus that a stress-induced DMR can regulate a long 618 non-coding RNA and influence gene expression, illustrating that causal methylation– 619 expression coupling may exist at selected loci. 620 Accordingly, the value of the integrated heatmaps is not evidence for a single 621 direction of regulation but the identification of candidates with coordinated, condition622 matched molecular changes. Loci that combine a persistent promoter DMR with reproducible 623 expression differences across generations are especially suitable for targeted validation. 624 Bisulfite amplicon sequencing, locus-specific expression assays, chromatin profiling, and 625 targeted methylation or demethylation would be required to distinguish causal regulation 626 from correlated responses to a common upstream change in nitric oxide or chromatin state. 627 628 **4.5 Persistent TE methylation extends the connection between GSNOR and transposon** 629 **regulation** 630 The number of persistent CHG and CHH DMRs at TEs is consistent with the central role of 631 non-CG methylation in heterochromatic TE silencing (Liang et al., 2022; Stroud et al., 2014). 632 It also agrees with the independent observation that GSNOR1 contributes to TE 633 demethylation and expression (Rudolf et al., 2021). In the present study, persistent TE634 associated DMRs occurred mainly in _gsnor-ko_ and included DNA/MuDR, RC/Helitron, 635 LTR/Copia, and LTR/Gypsy elements. Their persistence across G1, G3, and G5 extends the 636 earlier single-generation connection between GSNOR and TEs by showing that altered TE 637 methylation can remain detectable through repeated exposure and subsequent recovery. Some 638 genes located near persistently methylated TEs also showed altered expression, but the [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.09.765246; this version posted November 11, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.09.765246) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY-NC 4.0 International license.](http://creativecommons.org/licenses/by-nc/4.0/) 639 direction of expression could not be predicted from TE methylation alone. We tested this 640 because TEs can influence nearby genes through local chromatin spreading, regulatory 641 sequences, or TE-derived small RNAs, whereas many methylated TEs remain 642 transcriptionally silent and have no measurable effects on adjacent genes (Ahmed et al., 643 2011; Chuong et al., 2017; Dubin et al., 2018; Ito, 2022; Ramakrishnan et al., 2021). The 644 paired TE–gene analysis therefore identifies potential regulatory neighbourhoods rather than 645 proving direct cis regulation. Moreover, this study did not measure TE mobilisation, and 646 altered TE methylation or expression should not be equated with transposition. 647 648 **4.6 Independent comparisons support a reproducible subset of GSNOR-associated** 649 **DMRs** 650 Comparison with published _Arabidopsis_ stress-methylome datasets showed that some of the 651 persistent and switching DMRs, candidate genes, and TEs identified in the present study were 652 also detected in independent experiments (Supplemental Figures S22–S24; Supplemental File 653 S8) (Behl et al., 2025). CHH showed the largest absolute number of overlapping DMRs, but 654 CHH DMRs were also the most abundant overall. Therefore, this result does not mean that 655 CHH changes were proportionally more reproducible than CG or CHG changes. Variation 656 among studies was expected because they differed in stress conditions, plant age, tissue, 657 genotype, sequencing depth, and methods used to identify DMRs. 658 The published _gsnor1-3_ -versus-WT dataset provided a more direct comparison of the 659 baseline mutant phenotype (Supplemental Figure S25; Supplemental File S9). Of the 292 660 published DMRs, 215 (73.6%) overlapped at least one DMR detected in the current G1 661 _gsnor-ko_ Control comparison, and 211 of these regions showed the same methylation 662 direction. Viewed from the opposite direction, 1,378 of the 55,942 current G1 DMRs (2.5%) 663 overlapped a published region. These percentages differ because the published DMR set was 664 small and focused, whereas the present analysis detected many more context-specific and 665 overlapping 200-bp windows. One published region could therefore overlap several current 666 DMR windows. The high agreement in methylation direction among overlapping regions 667 supports the presence of a reproducible subset of GSNOR-associated DMRs. However, the 668 published dataset did not classify individual DMRs as CG, CHG, or CHH, so concordance 669 within the different sequence contexts could not be assessed. 670 The independently supported G1 regions were subsequently followed across G3 and 671 G5 to determine whether their methylation states persisted or changed over generations 672 (Supplemental Figure S25; Supplemental File S9). We also distinguished DMRs shared with 673 _gsnor-ko_ Control from those detected only in the treatment comparisons. This separated 674 stable baseline genotype effects from additional treatment-associated responses. The analysis 675 therefore extends the Rudolf et al. (2021) dataset by showing not only which GSNOR676 associated DMRs were reproduced at G1, but also how these regions behaved during 677 multigenerational treatment and recovery. 678 Future studies should validate the highest-priority regions using targeted bisulfite 679 sequencing and determine whether their methylation states affect gene expression. Analyses 680 of chromatin accessibility, histone modifications, small RNAs, and S-nitrosylated proteins 681 could help explain how GSNOR-dependent redox signalling is connected to DNA [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.09.765246; this version posted November 11, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.09.765246) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY-NC 4.0 International license.](http://creativecommons.org/licenses/by-nc/4.0/) 682 methylation. For TE-associated DMRs, direct measurements of TE expression and 683 mobilisation will be necessary, as methylation changes alone do not indicate altered TE 684 activity. 685 686 **ACKNOWLEDGEMENTS** 687 This work (for RB and CL) was supported by the Initiative and Networking Fund of the 688 Helmholtz Association under the call and by EpiCrossBorders: International Helmholtz689 Edinburgh Research School for Epigenetics. 690 We thank Peiyuan Zhu, Moritz Popp, Ina Zimmer, Petra Seibel, Franz Buegger, Mustafa 691 Oezden, Alexandros Sigalas, Georg Gerl, Ulrich Junghans, Baris Weber, and Armin Richter 692 for their technical support during experimental work, and Zhilin Zhang and Rohan Bhardwaj 693 for help with WGBS data preprocessing. We would also like to thank Prof. Dr. Aloys 694 Schepers for his fruitful input during thesis meetings. 695 During the preparation of this work, the author(s) used AI tools (such as ChatGPT 5.64 and 696 Grammarly 1.162.2.0) to polish the writing framework and to search for relevant research 697 articles (literature overview). After using this tool/service, the author(s) reviewed and edited 698 the content as needed and take(s) full responsibility for the content of the published article. 699 **CONFLICT OF INTEREST** 700 Not declared 701 **ORCID** 702 rachna.behl@helmholtz-munich.de (ORCID ID- https://orcid.org/0009-0004-5318-9481) 703 andrea.ghirardo@helmholtz-munich.de (ORCID ID- https://orcid.org/0000-0003-1973-4007) 704 binh.vo@tum.de 705 andreas.albert@bayern-mail.de 706 elkemattes@gmx.de 707 lucia.goessl@helmholtz-munich.de 708 joergpeter.schnitzler@helmholtz-munich.de (ORCID ID- https://orcid.org/0000-0002-9825709 867X) 710 frank@johanneslab.org (ORCID ID- https://orcid.org/0000-0002-7962-2907) 711 christian.lindermayr@helmholtz-munich.de (ORCID ID-https://orcid.org/0000-0002-9343712 4996) 713 **AUTHOR CONTRIBUTIONS** 714 Experiments were designed by CL & FJ; experimental work was conducted by RB, AG, EM, 715 LG, AA, JPS & CL. WGBS library preparation was performed by BV. Analysis was 716 performed by RB; statistical analysis and WGBS analysis support were provided by JPS, FJ 717 & CL; data interpretation was performed by RB, JPS, FJ & CL; and the paper was written by 718 RB with input from JPS, AG, FJ & CL. The written paper was reviewed by all the co-authors. 719 720 **SUPPLEMENTAL MATERIAL** 721 **Supplemental File S1. WGBS sample names and experimental classifications.** 722 List of the WGBS samples used in this study, including sample name, genotype, experimental [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.09.765246; this version posted November 11, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.09.765246) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY-NC 4.0 International license.](http://creativecommons.org/licenses/by-nc/4.0/) 723 lineage, generation, and experimental phase. Treatment labels in G5 reflect the preceding 724 treatment lineage, as plants were grown under control conditions during the recovery 725 generations. 726 727 **Supplemental File S2. Complete hypermethylated and hypomethylated DMR dataset.** 728 Coordinate-level catalogue of hypermethylated and hypomethylated regions identified in G1, 729 G3, and G5. Records are classified by genotype, treatment lineage, methylation context (CG, 730 CHG, or CHH), generation, and methylation direction. Reads were aligned to the TAIR10 731 reference genome, whereas DMR direction was defined relative to the corresponding WT 732 Control comparison baseline. 733 734 **Supplemental Table S3. AlphaBeta fitted estimates and pairwise statistical comparisons.** 735 Fitted intercept, methylation-gain ( α ), and methylation-loss ( β ) estimates are provided 736 separately for the CG, CHG, and CHH contexts. The workbook also contains pairwise 737 comparisons between genotypes and treatments, including estimated differences, pooled 738 standard errors, and nominal _P_ -values for α and β . Nominal _P_ < 0.05 is highlighted. 739 740 **Supplemental File S4. Summary of persistent and switching DMRs across G1, G3, and** 741 **G5.** Numbers of DMR regions classified as persistent or switching, stratified by genotype, 742 treatment, and methylation context. Persistent DMRs retained the same hypermethylated or 743 hypomethylated direction in G1, G3, and G5. Switcher DMRs were detected across all three 744 sampled generations, but changed methylation direction in at least one generation. 745 746 **Supplemental File S5. Literature-supported persistent and switching DMR candidates.** 747 Literature-guided prioritisation of genes and transposable elements associated with persistent 748 or switching DMRs. The persistent section contains the curated candidate-gene set, direct 749 promoter or gene-body DMR associations, candidate-selection summaries, supporting 750 references, and the candidate universe. The switcher section contains directly associated 751 genes and transposable elements, recurrent genomic loci, gene–TE co-overlapping loci, 752 prioritisation tiers, and supporting references. 753 754 **Supplemental File S6. Integrated DMR and RNA-seq data for candidate genes.** 755 Matched WGBS and RNA-seq data underlying the persistent- and switcher-candidate 756 heatmaps. Separate worksheets provide DMR methylation state, promoter or gene-body 757 association, RNA-seq log 2 fold change, DESeq2-adjusted _P_ -value, and expression758 significance classification for each matched genotype, treatment, and generation. Significance 759 was defined as adjusted _P_ < 0.05. 760 761 **Supplemental File S7. Persistent TE-associated DMRs and expression of nearby genes.** 762 Condition-matched integration of persistent TE-associated DMRs with significant RNA-seq 763 responses of nearby genes. The detailed worksheet reports TE identity, family and 764 superfamily, genomic coordinates, nearby gene, TE–gene distance and positional 765 relationship, genotype, treatment, generation, TE methylation state, gene-expression log � [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.09.765246; this version posted November 11, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.09.765246) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY-NC 4.0 International license.](http://creativecommons.org/licenses/by-nc/4.0/) 766 fold change, and adjusted _P_ -value. Additional worksheets summarise unique TEs, genes, TE– 767 gene pairs, and matched events by TE superfamily. 768 769 **Supplemental File S8. Validation of current DMRs against previous Arabidopsis stress-** 770 **methylome studies.** Coordinate-level comparison of the current G1–G3–G5 DMR trajectory 771 atlas with a compiled atlas of previously reported Arabidopsis stress-associated DMRs. 772 Overlaps were required to occur in the same methylation context. The workbook contains the 773 current and published DMR atlases; individual current–published overlap relationships; 774 summaries by genotype, treatment, context, trajectory class, and published study; enrichment 775 tests with raw and Benjamini–Hochberg-adjusted _P_ -values; and nonexclusive promoter, 776 gene-body, TE, and intergenic annotations. Each row in the overlap table represents an 777 overlap relationship rather than a unique DMR; therefore, one current DMR may be linked to 778 multiple published DMRs or studies. 779 780 **Supplemental File S9. Independent validation against the Rudolf et al. (2021) WT–** 781 **gsnor1-3 methylome dataset.** 782 Coordinate- and direction-based comparison of current G1 _gsnor-ko_ Control DMRs with 783 published _gsnor1-3_ -versus-WT DMRs from Rudolf et al. (2021). The workbook contains all 784 pairwise coordinate overlaps; classifications of current DMRs according to overlap and 785 directional concordance; the concordantly validated G1 anchor set; reciprocal summaries 786 using either current or published DMRs as the denominator; and audits of one-to-many and 787 many-to-many overlap relationships. Directional agreement was defined by concordance 788 between the current and published hypermethylated or hypomethylated state. The published 789 DMR table did not report individual CG, CHG, or CHH contexts; consequently, methylation790 context agreement could not be assessed. 791 792 **DATA AVAILABILITY** 793 The raw WGBS data supporting this study have been deposited in ArrayExpress under 794 accession **E-MTAB-17204** . Processed data and associated supporting materials, including 795 supplemental material and FastQC quality-control reports, have been deposited in Zenodo – 796 **https://doi.org/10.5281/zenodo.22960793** . 797 798 **REFERENCES** 799 Abdul Aziz, M., & Masmoudi, K. (2026). Epigenetic stress memory for predictive and 800 heritable crop epibreeding. _Plant_ _Stress_, _21_, 101363. 801 https://doi.org/https://doi.org/10.1016/j.stress.2026.101363 802 Ageeva-Kieferle, A., Georgii, E., Winkler, B., Ghirardo, A., Albert, A., Hüther, P., Mengel, 803 A., Becker, C., Schnitzler, J. P., Durner, J., & Lindermayr, C. (2021). Nitric oxide 804 coordinates growth, development, and stress response via histone modification and 805 gene expression. _Plant_ _Physiol_, _187_ (1), 336-360. 806 https://doi.org/10.1093/plphys/kiab222 807 Ahmed, I., Sarazin, A., Bowler, C., Colot, V., & Quesneville, H. (2011). Genome-wide 808 evidence for local DNA methylation spreading from small RNA-targeted sequences in 809 Arabidopsis. _Nucleic_ _Acids_ _Res_, _39_ (16), 6919-6931. 810 https://doi.org/10.1093/nar/gkr324 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.09.765246; this version posted November 11, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.09.765246) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY-NC 4.0 International license.](http://creativecommons.org/licenses/by-nc/4.0/) 811 Behl, R., Gallo-Franco, J., Hazarika, R., Zhang, Z., Wilming, F., Schnitzler, J., Lindermayr, 812 C., & Johannes, F. (2025). DNA methylome responses to biotic and abiotic stress in 813 <em>Arabidopsis thaliana</em>: A multi-study analysis. _bioRxiv_ . 814 Behl, R., Ghirardo, A., Winkler, J., Jafarian, S., Frungillo, L., Spoel, S., Fischback, A., 815 Huber, G., Koller, R., Albert, A., Mattes, E., Gößl, L., Mayer K, F. X., Johannes, F., 816 Lindermayr, C., & Schnitzler, J. (2026). GSNOR-dependent nitric oxide homeostasis 817 promotes recovery from repeated climate stress across generations in 818 <em>Arabidopsis thaliana</em>. _bioRxiv_, 2026.2008.2005.742973. 819 https://doi.org/10.64898/2026.08.05.742973 820 Boyko, A., Blevins, T., Yao, Y., Golubov, A., Bilichak, A., Ilnytskyy, Y., Hollunder, J., 821 Meins, F., Jr., & Kovalchuk, I. (2010). Transgenerational adaptation of Arabidopsis to 822 stress requires DNA methylation and the function of Dicer-like proteins. _PLoS One_, 823 _5_ (3), e9514. https://doi.org/10.1371/journal.pone.0009514 824 Cao, X., & Jacobsen, S. E. (2002). Role of the Arabidopsis DRM Methyltransferases in De 825 Novo DNA Methylation and Gene Silencing. _Current Biology_, _12_ (13), 1138-1144. 826 https://doi.org/https://doi.org/10.1016/S0960-9822(02)00925-9 827 Chuong, E. B., Elde, N. C., & Feschotte, C. (2017). Regulatory activities of transposable 828 elements: from conflicts to benefits. _Nat_ _Rev_ _Genet_, _18_ (2), 71-86. 829 https://doi.org/10.1038/nrg.2016.139 830 Coolen, S., Proietti, S., Hickman, R., Davila Olivas, N. H., Huang, P. P., Van Verk, M. C., 831 Van Pelt, J. A., Wittenberg, A. H., De Vos, M., Prins, M., Van Loon, J. J., Aarts, M. 832 G., Dicke, M., Pieterse, C. M., & Van Wees, S. C. (2016). Transcriptome dynamics of 833 Arabidopsis during sequential biotic and abiotic stresses. _Plant J_, _86_ (3), 249-267. 834 https://doi.org/10.1111/tpj.13167 835 Crisp, P. A., Ganguly, D., Eichten, S. R., Borevitz, J. O., & Pogson, B. J. (2016). 836 Reconsidering plant memory: Intersections between stress recovery, RNA turnover, 837 and epigenetics. _Sci Adv_, _2_ (2), e1501340. https://doi.org/10.1126/sciadv.1501340 838 Ding, Y., Fromm, M., & Avramova, Z. (2012). Multiple exposures to drought 'train' 839 transcriptional responses in Arabidopsis. _Nat_ _Commun_, _3_, 740. 840 https://doi.org/10.1038/ncomms1732 841 Ding, Y., Liu, N., Virlouvet, L., Riethoven, J. J., Fromm, M., & Avramova, Z. (2013). Four 842 distinct types of dehydration stress memory genes in Arabidopsis thaliana. _BMC Plant_ 843 _Biol_, _13_, 229. https://doi.org/10.1186/1471-2229-13-229 844 Dubin, M. J., Mittelsten Scheid, O., & Becker, C. (2018). Transposons: a blessing curse. _Curr_ 845 _Opin Plant Biol_, _42_, 23-29. https://doi.org/10.1016/j.pbi.2018.01.003 846 Feechan, A., Kwon, E., Yun, B. W., Wang, Y., Pallas, J. A., & Loake, G. J. (2005). A central 847 role for S-nitrosothiols in plant disease resistance. _Proc Natl Acad Sci U S A_, _102_ (22), 848 8054-8059. https://doi.org/10.1073/pnas.0501456102 849 Fernandes, L. B., & Ghag, S. B. (2022). Molecular insights into the jasmonate signaling and 850 associated defense responses against wilt caused by Fusarium oxysporum. _Plant_ 851 _Physiol Biochem_, _174_, 22-34. https://doi.org/10.1016/j.plaphy.2022.01.032 852 Ganguly, D. R., Crisp, P. A., Eichten, S. R., & Pogson, B. J. (2017). The Arabidopsis DNA 853 Methylome Is Stable under Transgenerational Drought Stress. _Plant Physiol_, _175_ (4), 854 1893-1912. https://doi.org/10.1104/pp.17.00744 855 Hazarika, R. R., Serra, M., Zhang, Z., Zhang, Y., Schmitz, R. J., & Johannes, F. (2022). 856 Molecular properties of epimutation hotspots. _Nat Plants_, _8_ (2), 146-156. 857 https://doi.org/10.1038/s41477-021-01086-7 858 Healey, A., Furtado, A., Cooper, T., & Henry, R. J. (2014). Protocol: a simple method for 859 extracting next-generation sequencing quality genomic DNA from recalcitrant plant 860 species. _Plant Methods_, _10_, 21. https://doi.org/10.1186/1746-4811-10-21 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.09.765246; this version posted November 11, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.09.765246) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY-NC 4.0 International license.](http://creativecommons.org/licenses/by-nc/4.0/) 861 IPCC. (2023a). Climate Change Information for Regional Impact and for Risk Assessment. 862 In C. Intergovernmental Panel on Climate (Ed.), _Climate Change 2021 – The Physical_ 863 _Science Basis: Working Group I Contribution to the Sixth Assessment Report of the_ 864 _Intergovernmental Panel on Climate Change_ (pp. 1767-1926). Cambridge University 865 Press. https://doi.org/DOI: 10.1017/9781009157896.014 866 IPCC. (2023b). Weather and Climate Extreme Events in a Changing Climate. In C. 867 Intergovernmental Panel on Climate (Ed.), _Climate Change 2021 – The Physical_ 868 _Science Basis: Working Group I Contribution to the Sixth Assessment Report of the_ 869 _Intergovernmental Panel on Climate Change_ (pp. 1513-1766). Cambridge University 870 Press. https://doi.org/DOI: 10.1017/9781009157896.013 871 Ito, H. (2022). Environmental stress and transposons in plants. _Genes Genet Syst_, _97_ (4), 169872 175. https://doi.org/10.1266/ggs.22-00045 873 Johannes, F., & Schmitz, R. J. (2019). Spontaneous epimutations in plants. _New Phytol_, 874 _221_ (3), 1253-1259. https://doi.org/10.1111/nph.15434 875 Joseph, R., Apollon, W., & De Oliveira, A. C. (2025). The stressing point: how plants 876 respond to environmental stimuli. _Biotechnol_ _Adv_, _84_, 108691. 877 https://doi.org/10.1016/j.biotechadv.2025.108691 878 Kankel, M. W., Ramsey, D. E., Stokes, T. L., Flowers, S. K., Haag, J. R., Jeddeloh, J. A., 879 Riddle, N. C., Verbsky, M. L., & Richards, E. J. (2003). Arabidopsis MET1 Cytosine 880 Methyltransferase Mutants. _Genetics_, _163_ (3), 1109-1122. 881 https://doi.org/10.1093/genetics/163.3.1109 882 Law, J. A., & Jacobsen, S. E. (2010). Establishing, maintaining and modifying DNA 883 methylation patterns in plants and animals. _Nat Rev Genet_, _11_ (3), 204-220. 884 https://doi.org/10.1038/nrg2719 885 Lee, U., Wie, C., Fernandez, B. O., Feelisch, M., & Vierling, E. (2008). Modulation of 886 nitrosative stress by S-nitrosoglutathione reductase is critical for thermotolerance and 887 plant growth in Arabidopsis. _Plant_ _Cell_, _20_ (3), 786-802. 888 https://doi.org/10.1105/tpc.107.052647 889 Liang, W., Li, J., Sun, L., Liu, Y., Lan, Z., & Qian, W. (2022). Deciphering the synergistic 890 and redundant roles of CG and non-CG DNA methylation in plant development and 891 transposable element silencing. _New_ _Phytologist_, _233_ (2), 722-737. 892 https://doi.org/https://doi.org/10.1111/nph.17804 893 Lindermayr, C., Rudolf, E. E., Durner, J., & Groth, M. (2020). Interactions between 894 metabolism and chromatin in plant models. _Mol Metab_, _38_, 100951. 895 https://doi.org/10.1016/j.molmet.2020.01.015 896 Liu, H. C., Lämke, J., Lin, S. Y., Hung, M. J., Liu, K. M., Charng, Y. Y., & Bäurle, I. (2018). 897 Distinct heat shock factors and chromatin modifications mediate the organ898 autonomous transcriptional memory of heat stress. _Plant J_, _95_ (3), 401-413. 899 https://doi.org/10.1111/tpj.13958 900 Mengel, A., Ageeva, A., Georgii, E., Bernhardt, J., Wu, K., Durner, J., & Lindermayr, C. 901 (2017). Nitric Oxide Modulates Histone Acetylation at Stress Genes by Inhibition of 902 Histone Deacetylases. _Plant_ _Physiology_, _173_ (2), 1434-1452. 903 https://doi.org/10.1104/pp.16.01734 904 Pecinka, A., Rosa, M., Schikora, A., Berlinger, M., Hirt, H., Luschnig, C., & Mittelsten 905 Scheid, O. (2009). Transgenerational stress memory is not a general response in 906 Arabidopsis. _PLoS One_, _4_ (4), e5202. https://doi.org/10.1371/journal.pone.0005202 907 Ramakrishnan, M., Satish, L., Kalendar, R., Narayanan, M., Kandasamy, S., Sharma, A., 908 Emamverdian, A., Wei, Q., & Zhou, M. (2021). The Dynamism of Transposon 909 Methylation for Plant Development and Stress Adaptation. _Int J Mol Sci_, _22_ (21). 910 https://doi.org/10.3390/ijms222111387 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.09.765246; this version posted November 11, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.09.765246) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY-NC 4.0 International license.](http://creativecommons.org/licenses/by-nc/4.0/) 911 Rizhsky, L., Liang, H., Shuman, J., Shulaev, V., Davletova, S., & Mittler, R. (2004). When 912 defense pathways collide. The response of Arabidopsis to a combination of drought 913 and heat stress. _Plant_ _Physiol_, _134_ (4), 1683-1696. 914 https://doi.org/10.1104/pp.103.033431 915 Rudolf, E. E., Hüther, P., Forné, I., Georgii, E., Han, Y., Hell, R., Wirtz, M., Imhof, A., 916 Becker, C., Durner, J., & Lindermayr, C. (2021). GSNOR Contributes to 917 Demethylation and Expression of Transposable Elements and Stress-Responsive 918 Genes. _Antioxidants (Basel)_, _10_ (7). https://doi.org/10.3390/antiox10071128 919 Sani, E., Herzyk, P., Perrella, G., Colot, V., & Amtmann, A. (2013). Hyperosmotic priming 920 of Arabidopsis seedlings establishes a long-term somatic memory accompanied by 921 specific changes of the epigenome. _Genome_ _Biol_, _14_ (6), R59. 922 https://doi.org/10.1186/gb-2013-14-6-r59 923 Shahryary, Y., Hazarika, R. R., & Johannes, F. (2020). MethylStar: A fast and robust pre924 processing pipeline for bulk or single-cell whole-genome bisulfite sequencing data. 925 _BMC Genomics_, _21_ (1), 479. https://doi.org/10.1186/s12864-020-06886-3 926 Shahryary, Y., Symeonidi, A., Hazarika, R. R., Denkena, J., Mubeen, T., Hofmeister, B., van 927 Gurp, T., Colomé-Tatché, M., Verhoeven, K. J. F., Tuskan, G., Schmitz, R. J., & 928 Johannes, F. (2020). AlphaBeta: computational inference of epimutation rates and 929 spectra from high-throughput DNA methylation data in plants. _Genome Biol_, _21_ (1), 930 260. https://doi.org/10.1186/s13059-020-02161-6 931 Shi, Y. F., Wang, D. L., Wang, C., Culler, A. H., Kreiser, M. A., Suresh, J., Cohen, J. D., 932 Pan, J., Baker, B., & Liu, J. Z. (2015). Loss of GSNOR1 Function Leads to 933 Compromised Auxin Signaling and Polar Auxin Transport. _Mol Plant_, _8_ (9), 1350934 1365. https://doi.org/10.1016/j.molp.2015.04.008 935 Stroud, H., Do, T., Du, J., Zhong, X., Feng, S., Johnson, L., Patel, D. J., & Jacobsen, S. E. 936 (2014). Non-CG methylation patterns shape the epigenetic landscape in Arabidopsis. 937 _Nat Struct Mol Biol_, _21_ (1), 64-72. https://doi.org/10.1038/nsmb.2735 938 Taudt, A., Roquis, D., Vidalis, A., Wardenaar, R., Johannes, F., & Colomé-Tatché, M. 939 (2018). METHimpute: imputation-guided construction of complete methylomes from 940 WGBS data. _BMC Genomics_, _19_ (1), 444. https://doi.org/10.1186/s12864-018-4641-x 941 van der Graaf, A., Wardenaar, R., Neumann, D. A., Taudt, A., Shaw, R. G., Jansen, R. C., 942 Schmitz, R. J., Colomé-Tatché, M., & Johannes, F. (2015). Rate, spectrum, and 943 evolutionary dynamics of spontaneous epimutations. _Proc Natl Acad Sci U S A_, 944 _112_ (21), 6676-6681. https://doi.org/10.1073/pnas.1424254112 945 Wibowo, A., Becker, C., Marconi, G., Durr, J., Price, J., Hagmann, J., Papareddy, R., Putra, 946 H., Kageyama, J., Becker, J., Weigel, D., & Gutierrez-Marcos, J. (2016). 947 Hyperosmotic stress memory in Arabidopsis is mediated by distinct epigenetically 948 labile sites in the genome and is restricted in the male germline by DNA glycosylase 949 activity. _Elife_, _5_ . https://doi.org/10.7554/eLife.13546 950 Wibowo, A., Becker, C., Marconi, G., Durr, J., Price, J., Hagmann, J., Papareddy, R., Putra, 951 H., Kageyama, J., Becker, J., Weigel, D., & Gutierrez-Marcos, J. (2018). Correction: 952 Hyperosmotic stress memory in Arabidopsis is mediated by distinct epigenetically 953 labile sites in the genome and is restricted in the male germline by DNA glycosylase 954 activity. _Elife_, _7_ . https://doi.org/10.7554/eLife.44302 955 Zhang, H., Lang, Z., & Zhu, J. K. (2018). Dynamics and function of DNA methylation in 956 plants. _Nat Rev Mol Cell Biol_, _19_ (8), 489-506. https://doi.org/10.1038/s41580-018957 0016-z 958 Zhang, Y., Xu, J., Li, R., Ge, Y., Li, Y., & Li, R. (2023). Plants' Response to Abiotic Stress: 959 Mechanisms and Strategies. _Int_ _J_ _Mol_ _Sci_, _24_ (13). 960 https://doi.org/10.3390/ijms241310915 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.09.765246; this version posted November 11, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.09.765246) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY-NC 4.0 International license.](http://creativecommons.org/licenses/by-nc/4.0/) 961 Zheng, Y., Li, Z., Cui, X., Yang, Z., Bao, C., Pan, L., Liu, X., Chatel-Innocenti, G., 962 Vanacker, H., Noctor, G., Dard, A., Reichheld, J. P., Issakidis-Bourguet, E., & Zhou, 963 D. X. (2023). S-Nitrosylation of the histone deacetylase HDA19 stimulates its activity 964 to enhance plant stress tolerance in Arabidopsis. _Plant J_, _114_ (4), 836-854. 965 https://doi.org/10.1111/tpj.16174 966 967 968 969 970 971 972 973 974 975 976 977 978 979 **FIGURES AND TABLES** 980 981 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.09.765246; this version posted November 11, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.09.765246) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY-NC 4.0 International license.](http://creativecommons.org/licenses/by-nc/4.0/) 982 983 **Figure 1. Differential methylation across generations in wild type and GSNOR-deficient** 984 _**A. thaliana**_ **.** Numbers of hypermethylated (orange) and hypomethylated (blue) differentially 985 methylated regions (DMRs) in wild type (WT) and _gsnor-ko_ plants in G1, G3 and G5, 986 separated by CG, CHG and CHH sequence context. All comparisons used WT Control from 987 the corresponding generation as the reference. Treatments were applied during G1–G3, 988 followed by two generations under control conditions; G5 labels therefore indicate ancestral 989 treatment lineages. DMRs were identified using jDMRgrid with 200-bp windows, a 50-bp 990 step, a minimum of 10 cytosines per context, exclusion of intermediate methylation calls and 991 replicate.consensus = 1. Hyper- and hypomethylated regions were extracted from the filtered 992 state-call matrices. WGBS included three biological replicates per control group and two per 993 treatment group in each sampled generation. Bars represent region counts rather than 994 replicate means; leftward bars indicate hypomethylated DMR counts. 995 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.09.765246; this version posted November 11, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.09.765246) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY-NC 4.0 International license.](http://creativecommons.org/licenses/by-nc/4.0/) 996 997 **Figure 2. Changes in the balance between hypermethylated and hypomethylated regions** 998 **across generations.** Hyper DMR ratios in WT and _gsnor-ko_ plants across G1, G3 and G5, 999 separated by treatment lineage and methylation context. Ratios were calculated by dividing 1000 hypermethylated by hypomethylated DMR counts, using WT Control from the corresponding 1001 generation as the reference. Circles connected by solid lines represent WT, and triangles 1002 connected by dashed lines represent _gsnor-ko_ . The dotted magenta line marks a ratio of 1, 1003 indicating equal numbers of hypermethylated and hypomethylated DMRs. Values above and 1004 below 1 indicate a predominance of hypermethylation and hypomethylation, respectively. 1005 The vertical axis is logarithmic. G5 labels indicate ancestral treatments followed by two [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.09.765246; this version posted November 11, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.09.765246) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY-NC 4.0 International license.](http://creativecommons.org/licenses/by-nc/4.0/) 1006 generations under control conditions. Ratios are descriptive summaries of the DMR counts 1007 shown in Figure 1. 1008 1009 **Figure 3. CG methylation divergence and persistence of regional methylation** 1010 **differences across generations.** **(A)** Observed CG methylation divergence and fitted 1011 AlphaBeta trajectories for WT and _gsnor-ko_ treatment lineages. Circles and solid lines 1012 represent WT observations and fitted trajectories, respectively; triangles and dashed lines 1013 represent _gsnor-ko_ . Colours distinguish treatments. Δ t denotes the summed generational 1014 distance separating two sampled individuals through their common ancestor. Methylation [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.09.765246; this version posted November 11, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.09.765246) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY-NC 4.0 International license.](http://creativecommons.org/licenses/by-nc/4.0/) 1015 gain ( α ) and methylation-loss ( β ) parameters were estimated using the AlphaBeta ABneutral 1016 model, with a posteriorMaxFilter setting of 0.99 and 50 optimisation starts. Parameter 1017 uncertainty was assessed using 100 bootstrap iterations. Pale-yellow shading highlights the 1018 area above a boundary placed just below the lowest fitted _gsnor-ko_ trajectory at each Δ t; it is 1019 a visual aid and does not represent a confidence interval or significance threshold. **(B)** 1020 Numbers of persistent (salmon) and switcher (turquoise) DMRs by genotype, treatment 1021 lineage and methylation context. All differential methylation states were defined relative to 1022 WT Control within the corresponding generation. Persistent DMRs retained the same 1023 direction at G1, G3 and G5; switcher DMRs were detected at all three sampling points but 1024 changed direction at least once. Vertical-axis breaks accommodate differences in abundance. 1025 Bars show region counts rather than replicate means. [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.09.765246; this version posted November 11, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.09.765246) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY-NC 4.0 International license.](http://creativecommons.org/licenses/by-nc/4.0/) 1026 1027 **Figure 4. Persistent and switching methylation patterns at selected candidate genes.** 1028 Heatmaps showing candidate genes associated with **(A)** persistent and **(B)** switcher DMRs in 1029 WT and _gsnor-ko_ plants. Methylation states were defined relative to WT Control within the 1030 corresponding generation. Columns within each treatment lineage represent G1, G3 and G5. 1031 Orange and blue indicate hypermethylated and hypomethylated DMR associations, 1032 respectively. White cells indicate no displayed DMR association, except where × denotes 1033 both hypermethylated and hypomethylated DMRs associated with the same gene. P, B and 1034 PB indicate promoter, gene-body, and combined promoter/gene-body associations, 1035 respectively. Promoters were defined as 1 kb upstream of the annotated transcription start 1036 site. Persistent DMRs retained the same direction across all three sampled generations; [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.09.765246; this version posted November 11, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.09.765246) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY-NC 4.0 International license.](http://creativecommons.org/licenses/by-nc/4.0/) 1037 switcher DMRs changed direction while remaining detectable at all three sampling points. 1038 Genes were selected and grouped according to literature-supported functions; these categories 1039 do not represent statistical enrichment results. G5 labels indicate ancestral treatment lineages 1040 following two generations under control conditions. 1041 1042 **Figure 5. Integration of candidate-gene methylation patterns with transcriptional** 1043 **responses.** Paired methylation and expression heatmaps for genes associated with **(A)** 1044 persistent and **(B)** switcher DMRs in WT and gsnor-ko plants. Columns within each 1045 treatment lineage represent G1, G3 and G5. For each gene, the upper track displays DMR 1046 associations and the lower track displays the corresponding RNA-seq log � fold change from [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.09.765246; this version posted November 11, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.09.765246) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY-NC 4.0 International license.](http://creativecommons.org/licenses/by-nc/4.0/) 1047 Behl et al. (2026). Methylation states were defined relative to WT Control within the 1048 corresponding generation. Orange, blue and magenta indicate hypermethylated, 1049 hypomethylated and combined hyper-/hypomethylated associations, respectively; white 1050 indicates no displayed DMR association. P, B and PB denote promoter, gene-body, and 1051 combined promoter/gene-body associations; × denotes both methylation directions. RNA 1052 colours represent unshrunken log � fold changes, including nonsignificant results, with blue 1053 indicating negative and red indicating positive values. Asterisks indicate DESeq2-adjusted P 1054 < 0.05, without an additional fold-change cutoff. Grey RNA cells indicate unavailable 1055 matched expression values; near-white expression colours indicate values close to zero. 1056 WGBS and RNA-seq results were matched by genotype, treatment and generation. G5 labels 1057 indicate ancestral treatment lineages following two generations under control conditions. 1058 1059 **Figure 6. Persistent TE-associated hypermethylation and expression of neighbouring** 1060 **genes.** Paired heatmaps showing eight selected transposable-element (TE)–gene candidate 1061 groups in WT and _gsnor-ko_ plants, organised by TE superfamily: DNA/MuDR, RC/Helitron, 1062 LTR/Copia and LTR/Gypsy. The upper track displays persistent TE-associated methylation 1063 and the lower track displays the neighbouring gene’s RNA-seq log � fold change. Columns 1064 within each treatment lineage represent G1, G3 and G5. Orange indicates persistent 1065 hypermethylation relative to WT Control within the corresponding generation; white 1066 indicates no displayed persistent TE-associated DMR. Persistent DMRs retained the same 1067 differential methylation direction across all three sampled generations. Candidate TEs were 1068 linked to nearby genes using a 1.2-kb proximity threshold. RNA results from Behl et al. 1069 (2026) were retained only for significant, condition-matched TE–gene associations with 1070 DESeq2-adjusted P < 0.05, without an additional fold-change cutoff. Blue and red indicate 1071 negative and positive log � fold changes, respectively; asterisks mark significant expression 1072 results. Grey RNA cells indicate absence from the filtered, condition-matched significant 1073 TE–gene table and do not establish unchanged expression. G5 labels indicate ancestral 1074 treatment lineages following two generations under control conditions. [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.09.765246; this version posted November 11, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.09.765246) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY-NC 4.0 International license.](http://creativecommons.org/licenses/by-nc/4.0/) 1075 1076 **Graphical abstract** 1077 _Created in BioRender.com_