# Source paper archive

- Source title: Ambivalent roles of nonsense-mediated mRNA decay in heat stress tolerance
- Author(s): Syed Hussam Abbas, Henrik Mihaly Szaker, Eva Darko, Tunde Eros, Daniel Silhavy, Tibor Csorba
- Source landing URL: https://www.biorxiv.org/content/10.64898/2026.10.30.763111v1
- PDF URL: https://www.biorxiv.org/content/10.64898/2026.10.30.763111v1.full.pdf
- PDF SHA-256: b9e49d85cdc6ea847a45e048535928767fc418805c01c01829a96bc263b7a7e0
- 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/10.64898/2026.10.30.763111; this version posted November 2, 2026. The copyright holder for this](https://doi.org/10.64898/2026.10.30.763111)

preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.


**Ambivalent roles of nonsense-mediated mRNA decay in heat stress tolerance**

Syed Hussam Abbas [1,2], Henrik Mihály Szaker [1], Éva Darkó [3], Tünde Erős [1], Dániel Silhavy [4],
Tibor Csorba [1]  

1 Genetics and Biotechnology Institute, Department of Plant Biotechnology, Hungarian
University of Agriculture and Life Sciences, 2100, Gödöllő, Hungary; [2] Plant and Horticulture
Program, Doctoral School of Agricultural and Food Sciences, Hungarian University of
Agriculture and Life Sciences (MATE), 2100, Gödöllő, Hungary; [3] HUN-REN, Agricultural
Institute, Center for Agricultural Research, 2462, Martonvásár, Hungary; [4] HUN-REN, Plant
Photo and Chronobiology group, Biological Research Centre, Szeged.
*Csorba.Tibor.Levente@uni-mate.hu

**Short title** : _NMD Regulates Early Sensitivity and Late Tolerance to Heat Stress_
**Keywords** : _Nonsense-mediated decay, heat stress memory, HsfA2, HsfA3, Arabidopsis_
**Highlight statement** : _Arabidopsis_ _NMD Mutants Exhibit a Biphasic Heat Stress Response,_
_with Early HsfA2/A3-Independent Tolerance and Late HsfA2-Dependent Sensitivity._

**Abstract**


RNA homeostasis is essential for gene regulation and cellular survival; however, how RNA
degradation pathways contribute to stress adaptation remains incompletely understood. Here,
we investigated the role of the nonsense-mediated mRNA decay (NMD) in regulating heat
stress (HS) memory in a long-term acquired thermotolerance assay. NMD-deficient
_Arabidopsis thaliana_ mutants _upf1-5_ and _upf3-1_ exhibited a biphasic phenotype: they displayed
transiently elevated heat tolerance during early recovery but became more sensitive to heat
during late recovery. Genetic analysis of _upf3;hsfa2_ and _upf3;hsfa3_ double mutants
demonstrated that the early heat tolerance occurs independently of both HsfA2 and HsfA3 HSmemory transcription factors. Under non-treated conditions, the _upf3-1_ transcriptome exhibited
baseline priming of heat-shock responsive transcripts, likely accounting for the initial heat
tolerance. Further transcriptome profiling revealed that _upf3-1_ mutants exhibited reduced
levels of the functional _HsfA2_ isoform ( _HsfA2-I_ ) and attenuated expression of the HsfA2dependent thermomemory regulon following the triggering stress. NMD deficiency also
upregulated immune-response genes while repressing developmental and cell-wall-associated
pathways. Together, impaired HsfA2-dependent thermomemory and a compromised growth–
defense balance underlies the late heat sensitivity of NMD-deficient plants. Overall, NMD acts
as a dual-action post-transcriptional regulator that coordinates the transition from stress
response to growth resumption.


**Introduction**

High temperature is a major abiotic stress factor that impacts the viability and reproductive
fitness of all organisms. Exposure to heat stress (HS) inflicts cellular damage that must be
prevented or, if already present, reversed. The collection of pathways dedicated to this


1


[bioRxiv preprint doi: https://doi.org/10.64898/2026.10.30.763111; this version posted November 2, 2026. The copyright holder for this](https://doi.org/10.64898/2026.10.30.763111)

preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.


mitigation is termed the Heat Shock Response (HSR). HSR operates on many layers, a central
level of which is transcriptional regulation. Upon HS a large part of developmental genes is
shutdown, while the stress-responsive genes are hyper-transcribed (Akerfelt, Morimoto, &
Sistonen, 2010; Charng, Mitra, & Yu, 2023; H. C. Liu & Charng, 2013; Obermeyer et al., 2023;
Ohama, Sato, Shinozaki, & Yamaguchi-Shinozaki, 2017; Szadeczky-Kardoss et al., 2022;
Yoshida et al., 2011).
The induction of the HSR is a multi-layered process that begins with the activation of Heat
Stress Factor A1 family transcription factors (in _Arabidopsis_ HsfA1a, A1b, A1d and A1e).
Once triggered, HsfA1s launch a transcriptional cascade of secondary HSFs, and other HStranscripts including heat stress proteins (HSPs), antioxidants, etc. (Akerfelt et al., 2010;
Charng et al., 2023; H. C. Liu & Charng, 2013; Ohama et al., 2017). Besides the immediate
HSR, plants are capable to remember heat exposures and become primed (acclimated) for a
subsequent thermal stress; the process is called thermomemory (or long acquired
thermotolerance, LAT). For thermomemory activation the plant utilizes dedicated targets of
HsfA1s, the HsfA2 and HsfA3 factors (Charng et al., 2007; Friedrich et al., 2021; Kappel et
al., 2023). The HsfA2 and A3 proteins establishes heat memory through promoting the
deposition of epigenetic marks that persist on target loci for days and ensure that
thermomemory genes remain active (type I memory genes) or are reactivated more robustly
(type II memory genes) upon a second heat exposure (Friedrich et al., 2021; Weng et al., 2014).
NMD (Nonsense-Mediated mRNA Decay) is a translation-coupled RNA quality control
system conserved in eukaryotes, which is essential for development and stress responses; NMD
contributes broadly to transcriptome homeostasis, it eliminates aberrant and wild-type
transcripts having NMD inducing features (F. Kerenyi, Wawer, Sikorski, Kufel, & Silhavy,
2013; Z. Kerenyi et al., 2008; Kertesz et al., 2006; Kurihara et al., 2009; J. P. Lloyd & Davies,
2013; J. P. B. Lloyd, 2018; Luha, Rana, Vainstein, & Kumar, 2024; Nasim, Karim, Blilou, &
Ahn, 2025; Ohtani & Wachter, 2019). The process is activated against mRNAs in which the
translational machinery identifies a premature termination codon (PTC). PTC recognition is
based on the presence of exon-junction complexes (EJC) located in the 3’ untranslated region
(3’UTR), and an unusually long 3’UTR. Besides, if a short open reading frame is present in
the 5’UTR (upstream ORF, uORF) before the main protein-coding frame, it also triggers NMDmediated mRNA decay (Li, Srivastava, Howell, Chudalayandi, & Severin, 2026; Luha et al.,
2024). The alternatively spliced (AS) mRNA isoforms are a major source of PTC (Kurihara et
al., 2009). The core NMD _trans_ -factors are the UP-Frameshift 1 (UPF1), UPF2, UPF3, and the
late-acting Suppressor with Morphogenetic effects on Genitalia 1 (SMG1) and SMG7 proteins
(F. Kerenyi et al., 2013; Z. Kerenyi et al., 2008; Kertesz et al., 2006; Kurihara et al., 2009; J.
P. B. Lloyd, 2018; Ohtani & Wachter, 2019). Plant NMD is autoregulated, the SMG7 mRNA
is a direct NMD target (Z. Kerenyi et al., 2008). In _A. thaliana_, a complete loss of function of
UPF genes leads to seedling lethality, while hypomorphic _upf_ mutants exhibit various
developmental abnormalities (Arciga-Reyes, Wootton, Kieffer, & Davies, 2006).
It was shown that NMD regulates several Hsfs, including HSFA2, through the destabilization
of AS isoforms. The _HsfA2-I_, the functional form, encoding the HsfA2 protein, is not the target
of NMD. The _HSFA2-II_, a non-functional mRNA isoform (Sugio, Dreos, Aparicio, & Maule,
2009), and the _HSFA2-III_, which encodes a short peptide named S-HsfA2, are both targets of
NMD (J. Liu et al., 2013). S-HsfA2 feedback regulates positively the HSFA2 locus


2


[bioRxiv preprint doi: https://doi.org/10.64898/2026.10.30.763111; this version posted November 2, 2026. The copyright holder for this](https://doi.org/10.64898/2026.10.30.763111)

preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.


transcriptional activity (Chen et al., 2025; J. Liu et al., 2013). Based on these, it was implied
that NMD fine-tunes plant thermomemory through the HsfA2 pathway, but this hypothesis was
not tested experimentally. Neither is known whether NMD regulates other aspects of heat stress
memory through HsfA3, or if the impact of AS-NMD on other proposed Hsf targets, including
HsfA4c, A7b, B1, B2a (J. Liu et al., 2013)) is biologically relevant.
In the present work, we demonstrate that NMD has an ambivalent role during thermal
resilience. On one hand (i) NMD activity establishes a non-stress status at ambient condition
through repression of unneeded HS-inducible transcripts, while on the other hand it promotes
thermomemory through (ii) increasing _HsfA2-I_ and consequently downstream HsfA2- (but not
HsfA3-) regulon expression, (iii) by suppressing aberrant induction of the immunity pathway
during HS, and (iv) by promoting growth through cell wall and auxin routes during recovery
period.
In sum, our work untangles the complex roles of NMD during thermal adaptation and shows
that RNA quality control -coupled turnover is a crucial determinant of efficient stress defense
and recovery.

**Materials and methods**


**Plant materials**


_Arabidopsis thaliana_ seeds were surface sterilized by bleach treatment, stratified two days at
4°C in the dark and germinated on MS (Duchefa M0222) medium agar plate (1x MS salts, 1%
glucose, 1% agar, pH 5.8). Plants were routinely grown in a Sanyo MLR-350 growth chamber
under cool white light at 21°C long-day condition photoperiod.


**Mutant** **lines used in the study**


T-DNA insertion lines _upf1-5_ (SALK_081178), _upf3-1_ (SALK_025175), _hsfa2-1_
(SALK_008978) _and_ _hsfa3-1_ (SALK_011107) were obtained from NASC
(https://arabidopsis.info). Double mutants were generated by crossing the respective single
mutants, and homozygous lines were confirmed by genotyping.


**Genotyping**


Genomic DNA was extracted using a nucleic acid extraction buffer (100 mM glycine, 10 mM
EDTA, 100 mM NaCl, 2% SDS), purified with phenol-chloroform-isoamyl alcohol (25:24:1,
pH 8.0), precipitated in 96% ethanol and resuspended in sterile water. Genotyping PCR was
performed using Taq DNA polymerase (NEB, M0273S) according to the manufacturer’s
instructions. Genotyping primers were described previously (Friedrich et al., 2021; SzadeczkyKardoss et al., 2022; Szaker et al., 2019; Szaker et al., 2025).


**Heat stress treatments**


For the long acquired thermotolerance (LAT) heat stress treatment, the 4 days-old seedlings
were primed with two-step priming (P) protocol in a water bath in the dark: 37°C/60 min (ZT6ZT7), 21°C/90 min, and at 44°C/45 min (ZT8,5-ZT9,5). Three days after acclimation, primed
seedlings were exposed to a triggering (P/T) heat treatment at 44°C/50-80 min in the water
bath. Seedlings were returned to 21°C for post-stress recovery (P/T+recovery). Experiments


3


[bioRxiv preprint doi: https://doi.org/10.64898/2026.10.30.763111; this version posted November 2, 2026. The copyright holder for this](https://doi.org/10.64898/2026.10.30.763111)

preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.


were performed in at least three independent replicates. For molecular analyses, seedlings were
harvested immediately after the priming (P) step and following one day of recovery (P+1d) at
21°C (RT-qPCR assay), or after priming and treatment (P/T) and P/T followed one day
(P/T+1d) recovery (RNA-seq). Samples were flash-frozen in liquid nitrogen and stored at 80°C. For physiological characterization, and for phenotyping and statistical analysis, plants
on agar plates were analyzed 4 days (P/T+4d) or 14 days (P/T+14d) after the priming and
triggering treatment (P/T), and in parallel non-treated controls (NT, plant age was 10, 12 or 20
days old, respectively) as shown in figure 1.


**Phenotypic scoring**


Seedling phenotypes were scored visually at early (+4d) and late (+14d) recovery after P/T,
alongside non-treated controls grown on the same plate. Each seedling was assigned one of the
four categories: not affected (green, expanded cotyledons and true leaves), mildly affected
(partial wilting, reduced leaf expansion or brownish-green pigmentation), severely affected
(extensive wilting and chlorosis, with some green tissue remaining) or dying (withered, yellowbrown cotyledons and true leaves without green tissue). For each genotype and treatment, an
average of 30 seedlings per plate were scored, and experiments were repeated at least three
times independently. The proportion of seedlings in each category was calculated per plate,
and differences between genotypes were tested using an unpaired two-tailed Student’s t-test (P
< 0.05).


**RNA extraction and RT-qPCR**


Seedlings (about 30 mg) were flash-frozen in liquid nitrogen and stored at −80°C until the next
step of the process. Total RNA was extracted from the frozen seedling tissue ground to a fine
powder in liquid nitrogen using a Quick-RNA Miniprep kit with on-column DNase I digestion
(ZYMO Research, Irvine, CA). First-strand cDNA was synthesized from 1 μg total RNA using
random primers and the ProtoScript II First Strand cDNA Synthesis Kit (NEB, E6300S,
https://www.neb.com) according to the manufacturer’s protocol. Quantitative PCR (qPCR)
was performed using qPCR Master Mix (NEB, M3003S, https://www.neb.com) and qPCR
reactions were run in LightCycler96 real-time PCR instrument (Roche). At least three
independent biological replicates were generated per genotype and treatment condition; error
bars represent the standard error of the mean. _P_ -values were calculated using unpaired twotailed Student’s _t_ -test to assess the significance of differences. Differences were considered
statistically significant at _P_ < 0.05. All comparisons were made against Col-0 within the same
treatment condition unless otherwise stated. Gene expression levels were normalized to the
reference gene _PP2AA3_ _(AT1G13320)_, previously extensively proved as a very stable
transcript under both control ambient or heat stress conditions (Karanyi et al., 2022; SzadeczkyKardoss et al., 2022; Szaker et al., 2025). Genotyping primers were described previously
(Friedrich et al., 2021; J. Liu et al., 2013; Sugio et al., 2009; Szadeczky-Kardoss et al., 2022;
Szaker et al., 2019; Szaker et al., 2025).


**Physiological measurements**


Chlorophyll _a_ fluorescence measurements were performed on intact plants using IMAGING
PAM Maxi (Walz, Effeltrich, Germany), as described previously (Darko _et al_ ., 2011), except


4


[bioRxiv preprint doi: https://doi.org/10.64898/2026.10.30.763111; this version posted November 2, 2026. The copyright holder for this](https://doi.org/10.64898/2026.10.30.763111)

preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.


that 56 umol m [-2] s [-1] actinic light intensity was used for 15 min to induce photosynthesis
(Lichtenthaler, 1987) after 20 min dark adaptation period. The fluorescence parameters were
determined according to (Lichtenthaler, 1987). Chlorophyll (a+b) content was determined
using a Cary-100 UV-VIS spectrophotometer (Varian, Middelburg, Netherlands) after
extraction of 50 mg shoot plantlets in 80% acetone, using the method of (Lichtethaler, 1987).
Electrolyte leakage (EL%) was determined as described previously by (Asghar _et al_ ., 2021)
with the use of 0.05 shoot plants for each sample. Six biological repetitions were used for each
genotype and treatment.


**Transcriptome analysis**


Col-0 (wild-type) and _upf3-1_ mutant plants were profiled by RNA-seq under three conditions:
non-treated (NT), heat priming+triggering (P/T), and one day of recovery after triggering (P/T
+1d). Raw RNA-seq reads were trimmed using Cutadapt (v4.7.) (Martin, 2011). The trimmed
reads were aligned to the _Arabidopsis thaliana_ TAIR10 (v62.) reference using HISAT2 and
were processed with SAMtools (v1.22.1.) to generate BAM files (Danecek et al., 2021). Genelevel raw read counts were assembled into a count matrix with featureCounts (v2.0.8.) (Liao,
Smyth, & Shi, 2014) for downstream differential expression analysis with DESeq2 (v1.50.2.)
(Love, Huber, & Anders, 2014). Differential expression between _upf3-1_ and wild-type Col-0
was assessed within each condition (log 2 FC > 1, BH adjusted p < 0.05).


Gene Ontology (GO) term enrichment was performed with clusterProfiler (v4.18.4.) (T. Wu et
al., 2021). GO terms with a BH adjusted p-value below 0.05 were considered significant. To
reduce redundancy among related GO terms, semantic similarity reduction was performed
using the Wang similarity measure with a cutoff of 0.7.


For each Araport11 transcript, we computed 3′UTR length, whether an intron overlaps the
3′UTR, and whether an upstream open reading frame (uORF) is present. Each feature was
reduced to one value per gene: the longest 3′UTR across isoforms, and an intron or uORF flag
if any isoform had one. Experimental uORF evidence came from the RiboTaper (primary
analysis) and CiPS (sensitivity analyses) datasets (H. L. Wu et al., 2024). A gene with too little
ribosome-profiling signal can’t show that it lacks a uORF. We therefore called a gene Riboseq-detectable if its TPM was at least 1 in GSE183264. Genes with a positive uORF call were
always kept, and genes without one counted as negatives only if they were detectable. 3′UTR
length was compared between expression-status groups with Wilcoxon rank-sum tests. The
prevalence of 3′UTR introns and uORFs was compared with Fisher’s exact tests. P-values were
adjusted with the Benjamini–Hochberg procedure within each feature family.


Expression trajectories were constructed using DESeq2-normalized counts. Counts were
transformed as log 2 (normalized count + 1), and mean expression was calculated for each gene
within each of the six genotype–treatment combinations. The 2,000 most variable genes were
selected for the primary analysis and each selected gene’s six-condition profile was
standardized to a z-score. K-means clustering was performed on these standardized profiles
with six clusters, 50 random starts and a maximum of 100 iterations, using a fixed random
seed. Cluster trajectories were summarized by genotype and treatment.


5


[bioRxiv preprint doi: https://doi.org/10.64898/2026.10.30.763111; this version posted November 2, 2026. The copyright holder for this](https://doi.org/10.64898/2026.10.30.763111)

preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.


For _cis_ -regulatory element enrichment, candidate transcription-factor binding motifs were
searched in the 1-kb region upstream of gene's transcription start site; motif presence was
compared against the genome-wide promoter background by Fisher's exact test, reported as
odds ratio and significance.


Heat shock response (HSR) gene sets were defined using the treatment-associated DESeq2

contrasts in Col-0. Genes induced under P/T but not P/T+1d were classified as acute or transient

responders; genes induced under both P/T and P/T+1d were classified as sustained responders;
and genes induced under P/T+1d but not P/T were classified as delayed responders. The
broader HSR-responsive set comprised genes induced under either treatment condition, while
genes expressed in at least one Col-0 sample but not meeting the induction criteria were
classified as other expressed genes. Paired Wilcoxon signed-rank tests were applied to pergene expression means for descriptive Col-0 versus _upf3_ comparisons.


For each HSR gene, Pearson’s correlation coefficient was calculated between its Col-0 wild
type expression profile across NT, P/T, and P/T+1d and the corresponding wild type expression
profile of HSFA2 or HSFA3. Genes with a Pearson correlation coefficient ( _r_ ) > 0.7 with
HSFA2 were assigned to the HSFA2-correlated group, while those with _r_  - 0.7 with HSFA3
were assigned to the HSFA3-correlated group. Genes exceeding the threshold for both
transcription factors were included in both groups and genes with correlations ≤ 0.7 for both
factors were classified as ‘neither-correlated’. HSFA2, the adjacent gene AT2G26140 and
HSFA3 were excluded from the target-gene correlation analysis to avoid self-correlation or
inclusion of the specified neighbouring locus. The corresponding standardized wild type
HSFA2 or HSFA3 expression profile was used as the reference trajectory in the relevant
transcription-factor-correlated groups.


A priming index was calculated for HSR-classified genes using the genotype-associated
expression effect under NT relative to the largest Col-0 treatment induction estimate. The index
was defined as the log 2 FC for _upf3_ versus Col-0 under NT divided by the maximum of the Col0 log 2 FC fold changes for P/T versus NT and P/T+1d versus NT. Genes were retained for this
analysis only when the denominator was finite and positive and the resulting index was finite.


Analyses and figures were performed and generated in R using DESeq2 and the associated
Bioconductor and CRAN packages (including clusterProfiler, org.At.tair.db, rtracklayer,
readxl, rrvgo, ggplot2, dplyr, tidyr, tibble and patchwork). Unless otherwise specified,
differential-expression significance was defined using a Benjamini–Hochberg-adjusted pvalue threshold of 0.05. All R scripts used for analysis can be found on
https://github.com/TCsorbaGroup/Heat_Stress_Recovery_UPF3_RNAseq_analysis. Raw
RNA-seq data have been made available in the SRA repository under the BioProject ID:

PRJNA1552212.


**Results**

**NMD restricts early heat tolerance but supports long-term recovery**


6


[bioRxiv preprint doi: https://doi.org/10.64898/2026.10.30.763111; this version posted November 2, 2026. The copyright holder for this](https://doi.org/10.64898/2026.10.30.763111)

preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.


LAT assays were used to evaluate the role of NMD in regulating thermomemory. We subjected
the NMD mutants _upf3-1_ and _upf1-5_ (referred to as _upf3_ and _upf1_ ) to a triggering heat stress
assay (T) following a priming (P) treatment (primed and triggered, P/T), and then we monitored
plant recovery over a two-week post-stress period (Fig 1A-E and Fig S1). Non-treated (NT)
plants were used as controls.
During early recovery (P/T+4d), both NMD mutants displayed a mild heat-tolerant phenotype
compared to the wild-type Arabidopsis (Columbia-0, Col-0): their leaves were larger and
exhibited less shrinking, though they developed a distinct brownish-green pigmentation (Fig
1B,D and S1A). In contrast, Col-0 plants were more affected at this stage, displaying withered,
yellow-brown cotyledons and true leaves.
Notably, the phenotypic difference gradually reversed during the two weeks recovery period,
the mutant plants lost their pigmentation and developed a HS sensitivity phenotype (Fig 1C,E).
At the end of the two-week recovery, the Col-0 plants recovered much more successfully than
the NMD mutants (Fig 1C,E and S1C,D).
To analyze the impact of NMD absence at the physiological level, we performed a suite of
measurements at early (P/T+4d) and late stage of recovery (P/T+14d), in accordance with the
visual inspection timepoints on Col-0 and _upf3_ (being a stronger allele (Arciga-Reyes et al.,
2006))(Fig 2). Under ambient conditions, the total chlorophyll content of the _upf3_ was slightly
lower than that of Col-0 (Fig 2A), but following 4-day recovery the chlorophyll levels
decreased more strongly in Col-0 than in _upf3_ (Fig. 2A, _left_ ). During the late recovery
(P/T+14d) however, Col-0 plants recovered significantly better (Fig. 2A, _right_ ).
To assess membrane integrity, we measured electrolyte leakage. Under NT conditions _upf3_
displayed slightly higher leakage rates (Fig 2B). During recovery, ion leakage increased in both
genotypes, but _upf3_ exhibited significantly lower leakage at the early recovery stage and higher
leakage rates at the late recovery stage (Fig. 2B).
To evaluate photosynthetic performance, we determined key chlorophyll fluorescence
parameters. While the effective quantum yield of photosystem II (Y(II)) expectedly declined
in response to heat during the early (P/T+4d) recovery phase, the _upf3_ mutant retained a
significantly higher Y(II) (Fig 2C, _top_ ). During the later phase of recovery (P/T+14d), the Col0 plants recovered more efficiently, as shown by higher Y(II) (Fig.2C, _bottom_ ). The lower level
of Y(II) in _upf3_ was accompanied by higher levels of regulated Y(NPQ) and non-regulated
Y(NO) non-photochemical quenching compounds (Fig. 2C). The elevated Y(NO) indicates an
inability to safely dissipate excess excitation energy, leading to photoinhibition. The maximum
photochemical efficiency of PSII (Fv/Fm) was mildly and equally reduced in both genotypes
at early recovery and slightly more reduced in the mutant at late recovery stage (Fig. 2D).
Taken together, the visual observations and physiological measurements consistently show that
although the NMD mutants are more thermotolerant in the early recovery phase, over a twoweek period, the wild-type plants recover much better. Thus, NMD plays a dual role in
thermotolerance; it inhibits early but promotes late HS recovery. These data also suggest that
at least two contrasting layers of recovery pathways act during the stress-to-regrowth
physiological switch.

**NMD remains active during and after heat stress**


7


[bioRxiv preprint doi: https://doi.org/10.64898/2026.10.30.763111; this version posted November 2, 2026. The copyright holder for this](https://doi.org/10.64898/2026.10.30.763111)

preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.


To compare the molecular changes in wild-type and _upf3_ NMD mutant plants during the
execution of heat stress memory and very early recovery, we profiled the transcriptomes of
non-treated (NT), primed-and-triggered (P/T) and recovering (P/T+1d) plants (Fig. 1A).
First, we studied whether NMD functions during and after the HS on selected NMD _trans_
factors. As expected, the UPF3 mRNA showed strongly reduced expression in the _upf3_ mutant.
The SMG7 mRNA, which is a strong direct target of NMD, was overexpressed in _upf3_ under
all conditions. By contrast, the UPF1 mRNA was elevated only after HS, and UPF2 was
unchanged (Fig 3A). To assess genome wide functions of NMD pathway, we compared the
expression of a previously defined core NMD-target set (300 genes, the common targets of
_upf1_ and _upf3_ mutants (Szaker et al., 2025)) Fig. 3B). The _upf3_ /Col-0 ratio for NMD targets
remained significantly elevated in all states (1.93-, 1.87- and 1.81-fold at NT, P/T and P/T+1d;
p < 0.001) indicating that NMD is active throughout the HS-recovery experiment.
Next, we analyzed the genes that showed differential expression (DEGs, defined as log 2 FC >
1, adjusted p < 0.05;) in _upf3_ plants relative to the wild-type. In all three states, more
upregulated than downregulated DEGs were found in _upf3_ (n = 606, 547 and 608 upregulated
vs. n = 248, 317 and 349 downregulated in NT, P/T and P/T+1d, respectively; Fig. S2A). This
is consistent with the role of NMD in destabilizing target transcripts (Gloggnitzer et al., 2014;
Riehs-Kearnan, Gloggnitzer, Dekrout, Jonak, & Riha, 2012; Szaker et al., 2025). A significant
part of these upregulated transcripts were genotype-dependent (n = 163 up in all conditions),
in contrast to the downregulated ones (n = 12) (Fig S2B,C).
The nature of the NMD mechanism suggests that upregulated gene set contains both direct and
indirect NMD targets, whereas downregulated genes are always indirect targets. To assess the
ratio of direct/indirect NMD targets in the upregulated DEGs (Up), we analyzed the canonical
NMD _cis_ -features (3′UTR length, intron in the 3′UTR (Z. Kerenyi et al., 2008) or uORF in the
5’UTR (H. L. Wu et al., 2024)) in the Up and Down genes and compared them with the
unchanged (Unch.) control genes (Fig. 3C–E). The NMD features were not overrepresented in
the Down set (relative to the control), which is consistent with these being indirect NMD
targets. By contrast, Up genes had an increasing trend in 3’UTR lengths (NT and P/T+1d, Fig
S3C), and were significantly enriched for 3′UTR introns (2.5-, 1.8- and 2-fold at NT, P/T and
P/T+1d, vs unchanged set; Fig. 3D) and uORF (2.6-, 1.9- and 2.4-fold at NT, P/T and P/T+1d,
vs unchanged set; Fig. 3E) NMD features. The enrichment of NMD features in _upf3_
upregulated genes reflects the stabilization of direct targets in the absence of NMD. However,
these features were present in only a fraction of upregulated genes (<10.3% 3’UTR intron- and
<25.5% uORF-containing genes), indicating that most upregulated genes are not direct NMD

targets.
Together, these data show that NMD remains active throughout and after the heat stress regime,
and that although its loss stabilizes several direct targets, many of the upregulated transcripts
are probably secondarily changed.

**NMD deficiency causes heat-dependent de-repression of immunity pathways**
To unravel the molecular basis of the altered heat tolerance, the _upf3_ DEGs were analyzed for
enriched GO terms (Fig. S2D). Notably, GO studies revealed state-specific responses in _upf3_ .
In NT plants, the only term enriched among upregulated genes was “RNA modification”, which
remained overrepresented at P/T and P/T+1d, suggesting a basal alteration of RNA


8


[bioRxiv preprint doi: https://doi.org/10.64898/2026.10.30.763111; this version posted November 2, 2026. The copyright holder for this](https://doi.org/10.64898/2026.10.30.763111)

preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.


homeostasis in the absence of NMD. Downregulated genes at NT were moderately enriched
for GO terms related to cell wall metabolism and oxidative-stress. After heat (P/T) and during
recovery (P/T+1d), upregulated genes in _upf3_ were strongly enriched for immunity-related
terms ( _see detailed list_, Fig S2D). Surprisingly, although _upf3_ showed early heat tolerance
(Fig1-2), typical heat-response GO terms were not enriched, suggesting that immunity, rather
than a canonical HSR, is activated during HS in the NMD-deficient mutant. The downregulated
genes at P/T and P/T+1d were enriched for growth- and cell wall-associated terms, predicting
a decreased recovery capacity in _upf3_ following HS.
Together, these changes are consistent with a shift in the growth–immunity balance toward
immunity in the absence of NMD during HS and early recovery.

**Heat-dependent immune activation in** _**upf3**_ **is consistent with transcriptional induction**
Since we did not find HSR differences when _upf3_ DEGs were analyzed, we employed a
different approach. We partitioned the 2,000 most variable genes of the transcriptome into six
clusters (see _Materials and Methods_, Fig 4A) and then these clusters were analyzed. Cluster 1
was enriched for growth- and metabolism-related GO terms, cluster 2 for heat stress and
proteostasis terms, cluster 3 for “response to wounding”, and cluster 4 for “RNA modification”
and “macromolecule modification” terms. Cluster 5 showed strong enrichment for immunityrelated GO terms, whereas cluster 6 did not show any GO enrichment (Fig S3). Importantly,
apart from cluster 5, none of the clusters showed strong genotype-specific differences between
the wild-type and _upf3_ (Fig 4A), indicating that most of the variance of these genes is driven
by heat rather than by genotype. Cluster 5, the only group in which gene expression was
strongly NMD-dependent, was further analyzed.
Cluster 5 (n = 279) displayed both genotype- and condition-specific trajectories. To separate
these effects, cluster 5 was subdivided into three subclusters (subclusters 5/1–5/3) (Fig. 4B).
Subcluster 5/1 and 5/3 genes, but not subcluster 5/2 genes, were overexpressed in the _upf3_
mutant under all three conditions (NT, P/T and P/T+1d, respectively). Thus, NMD direct
targets are mainly present in these subclusters, whereas subcluster 5/2 genes are predominantly
indirect targets. Subcluster 5/1 was enriched for “RNA modification”, while subcluster 5/3 was
weakly enriched for “defense response to symbiont” term (Fig. 4B,C). Subcluster 5/2 genes
were expressed similarly at NT in the wild-type and _upf3_, remained unchanged in treated Col0, but were induced in _upf3_ during heat (P/T) and further elevated upon recovery (P/T+1d).
This subcluster was strongly enriched for immune-related terms, mirroring the genome-wide
trends (Fig. 4C vs Fig S2D). These data suggest that the HS-inducible immune-related genes,
which are suppressed by NMD, are mainly indirect NMD targets.
To distinguish direct transcriptional induction from NMD-dependent RNA destabilization, we
analyzed promoter _cis_ -elements alongside NMD _cis_ -features (Fig 4D,E). The promoters of
subcluster 5/2 were enriched for the W-box (OR = 1.89, adjusted p = 0.08) and heat-responsive
HSE motifs (OR = 2.02, adjusted p < 0.05), whereas subcluster 5/1 was mildly enriched for the
DRE/CRT motif (OR = 1.7)(Fig 4D). While subclusters 5/1 and 5/3 were enriched in genes
having uORF NMD feature (28% and 21.3% vs 10% in background set; Fig. 4E, _right_ ), and
subcluster 5/3 showed enrichment for 3′UTRs with introns (8,9% vs 4,3%, Fig 4E, _middle_ ),
NMD features were underrepresented in subcluster 5/2 genes (Fig 4E, _left_ ). These results
suggest that the immune genes present in subcluster 5/2 are upregulated mainly through


9


[bioRxiv preprint doi: https://doi.org/10.64898/2026.10.30.763111; this version posted November 2, 2026. The copyright holder for this](https://doi.org/10.64898/2026.10.30.763111)

preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.


transcriptional activation, possibly via both WRKY and HSF transcription factor activity,
whereas transcripts in 5/1 and 5/3 may be stabilized in the absence of NMD activity, in
agreement with their trajectories. Importantly, the activation of the immune subcluster 5/2
depended on heat: the _upf3_ /Col-0 ratio was 1.1- at NT, 3.0- at P/T and 3.8-fold at P/T+1d, while
the genes did not respond to heat in Col-0.
Collectively, clustering analysis revealed that heat induces a transcriptional immune response,
which is suppressed by NMD in wild-type plants but occurs in the NMD-deficient mutant.

**Reduced** _**HSFA2**_ **levels is associated with a dampened HSR in** _**upf3**_
It was previously suggested that NMD regulates _HsfA2_ isoforms’ stability (J. Liu et al., 2013;
Sugio et al., 2009) and through this also thermomemory. For this reason, we examined selected
hallmark transcripts of heat stress-memory (Fig. 5A). _HSFA2_ dropped to ∼78% in _upf3_ vs Col0 at P/T (BH-adjusted p = 0.06, n = 3), whereas the expression of memory factor _HsfA3_
remained low at P/T in both genotypes. The memory HSP targets _sHSP18.2_ and _sHSP22.0_

∼
were mildly induced, both were higher in Col-0 plants vs _upf3_ ( 0.77- and 0.65-fold decrease,
respectively). The control non-memory _HSP101_ was induced at P/T, non-significantly lower
in _upf3_ (∼88% of Col-0). During recovery the level of _HsfA3_ and memory chaperons sHSP18.2
and sHSP22 was strongly elevated, with no significant difference between genotypes (Fig 5A).
To examine the HSR genome-widely, we defined heat-responsive genes as those induced in
Col-0 relative to NT (HSR genes, n = 2416, BH-adjusted p < 0.05, log 2 FC > 1), and analyzed
the difference in HSR expression between Col-0 and _upf3_, using the non-HSR set (n = 17,711)
as control. Importantly, the mean expression of the HSR set was significantly higher in _upf3_

∼
relative to Col-0 at NT ( 1.27-fold), but significantly lower at P/T (0.86-fold) (Fig 5B). The
difference disappeared at P/T+1d. These results suggest that NMD-deficiency (i) already
creates a weak heat-stress-primed state under NT conditions, and (ii) dampens the response
during the acute phase of heat stress, an effect that is mostly lost during recovery. The findings

that both HSFA2 induction and the HSR activation are weaker in the NMD-deficient mutant at

P/T suggest that NMD is required for optimal HSFA2 mRNA level and HS responses.
Although both HSFA2 and HSFA3 are essential for thermomemory, our findings that HSFA2
is upregulated at P/T and decreases at P/T+1d, while HSFA3 is induced at P/T+1d indicate that
they could regulate the HSR at different stages. Based on these, we grouped the heat-responsive
genes into transient (P/T-specific, n = 664), sustained (increased level during both P/T and
P/T+1d, n = 623) and delayed (P/T+1d-specific, n = 1,129) sets (Fig. 5B). We found the
transient genes are enriched in cluster 6, the sustained genes in cluster 2 and mildly cluster 6,
while the delayed genes in cluster 4 (Fig. S4). Expression of most transient and sustained genes
correlated with _HsfA2_ (n = 583, 93.6%, and n = 589, 94.5%, respectively); while the delayed
genes correlated mainly with _HsfA3_ (n = 1045) (Fig 5C, S5A). Relevantly, the largest genotype
differences were seen in the HSFA2-correlated transient and sustained sets at P/T (∼0.74- and
0.77-fold decrease in _upf3_ ), and the smallest in the HsfA3-correlated delayed set at both P/T
and P/T+1d (∼0.99-fold decrease in _upf3_ ) (Fig 5C), indicating that specifically HsfA2 pathway
is affected by NMD loss.
The direct target loci of HsfA2/HsfA3 were defined by ChIP-seq (Kappel et al., 2023): the two
factors bind to nearly identical loci, in total 380 genes. We analyzed the overlap of our three


10


[bioRxiv preprint doi: https://doi.org/10.64898/2026.10.30.763111; this version posted November 2, 2026. The copyright holder for this](https://doi.org/10.64898/2026.10.30.763111)

preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.


HSR groups with these ChIP loci: 11 of 664 transient genes (1.7%), 100 of 623 sustained genes
(16%) and 54 of 1,129 delayed genes (4.8%) overlapped with HsfA2/A3-ChIP loci (Fig. 5D).
Of the 215 ‘ChIP-only’ genes left, 94 showed no significant change, 65 were weakly induced,
30 were not or only lowly expressed and 26 were repressed by heat in our dataset (Fig. S4B);
overall, ChIP-only genes had a lower expression level in _upf3_ (Fig. S4C) and were enriched
for GO terms "nucleotide catabolic processes” and weakly for “response to heat” (Fig S4D).
Notably, the transient and sustained HSR genes were found to be significantly lower expressed
at P/T in _upf3_, but no difference was found in the delayed genes between genotypes (Fig 5C).
The strongest induction occurred in the ChIP-bound (much less in the unbound) sustained
genes, at both P/T and P/T+1d, with lower median in _upf3_ (Fig. S4E).
To test whether the lower transcript levels in _upf3_ plants reflect transcription or decay, we
analyzed promoter and NMD _cis_ -features (Fig S5B,C). For the transient set we found promoter
enrichment of HSE (OR = 1.26, adjp < 0.05), in case of the sustained set for HSE (OR = 1.25,
adjp < 0.05) and ABRE (OR = 1.79, adjp < 0.001), and for the delayed set of ABRE (OR =
1.31, adjp < 0.01) and DRE (OR = 1.21, adjp < 0.05) _cis_ elements (Fig S5B). In addition, no
significant enrichment of genes with NMD features could be found in the three HSR sets (Fig.
S5C). These findings are consistent with HSR genes being transcriptionally regulated through
HSE (transient set), HSE and ABRE (sustained set), and ABRE and DRE (delayed set) _cis_
elements within their promoters and being rarely direct NMD targets.
Based on these results, we propose that the transient and sustained sets of HSR genes are
regulated primarily by HsfA2, while the delayed HSR set is controlled by HsfA3, and that, the
decreased level of _HsfA2_ mRNA in _upf3_ leads to the drop of transient and sustained genes at

P/T.

**NMD deficiency leads to HsfA2- and HSFA3-independent early heat tolerance**
The transcriptome analysis showed a reduced level of HSFA2 mRNA in _upf3_ and a dampened
induction of the HsfA2-correlated, but not the HsfA3-correlated branch (Fig. 5). This may
explain the ineffective late recovery of _upf3_, but not the early heat tolerance of the NMD
mutants (Fig. 1, S1). We therefore hypothesized that NMD can also act independently of the
HsfA2/HsfA3 pathways and tested this using the double mutants _upf3-1;hsfa2-1_ ( _upf3;hsfa2_ )
and _upf3-1;hsfa3-1_ ( _upf3;hsfa3_ ). Under NT conditions, _upf3_ rosettes were slightly smaller than
those of Col-0 (Arciga-Reyes et al., 2006), whereas _hsfa2_ and _hsfa3_ rosettes appeared slightly
larger. _hsfa2_ plants had started to bolt earlier than _Col-0, upf3_ and _hsfa3_ . The _upf3;hsfa3_
rosettes resembled _upf3_ in size, whereas _upf3;hsfa2_ rosettes were larger and were also bolting
(Fig. S6).


First, we analyzed NMD activity in the _upf3_ and _hsfa2_ single and the _upf3,hsfa2_ double mutants
using the _SMG7_ (Z. Kerenyi et al., 2008) and _At5g45430_ (Degtiar et al., 2015) NMD reporters.
Both NMD-sensitive transcripts were elevated in the _upf3_ and _upf3,hsfa2_ genotypes under all
conditions and were unchanged in the _hsfa2_ mutant, indicating that NMD is not affected by the
_hsfa2_ mutation (Fig. S7).


Further, we compared the heat sensitivity to P/T treatments in a time series treatment set (50,
60 and 80 min; Fig. 6, S8). The _hsfa2_ and _hsfa3_ single mutants were markedly heat-sensitive
(Fig. 6A,B; S8A,B), confirming the roles of HsfA2 and HsfA3 in thermomemory. During early


11


[bioRxiv preprint doi: https://doi.org/10.64898/2026.10.30.763111; this version posted November 2, 2026. The copyright holder for this](https://doi.org/10.64898/2026.10.30.763111)

preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.


recovery (P/T+4d), _upf3_ was less affected than the Col-0. The double mutants were less
sensitive than the respective _hsfa_ single mutants but remained more sensitive than the _upf3_
(Fig. 6B, S8B). This suggests that the loss of UPF3 improves early heat tolerance likely
independently of HsfA2 and HsfA3. At late recovery (P/T+14d), the _upf3;hsfa2_ and _upf3;hsfa3_
plants were not significantly different from _hsfa2_ and _hsfa3_ (Fig. 6C,D, S8C,D), thus, the loss
of NMD could not compensate for the high mortality caused by the _hsfa2_ and _hsfa3_ mutations.


Together, these results suggest that UPF3 loss confers a HSFA2- and HSFA3-independent,
transient heat tolerance during early recovery, whereas HsfA2 and HsfA3 remain the main
drivers of long-term memory.


**NMD deficiency confers early heat tolerance independently of the HsfA2 regulon**


Several HSFs (HsfA4c, HsfA7b, HsfB1 and HsfB2a) were previously proposed to be NMD
targets (J. Liu et al., 2013). To test whether the altered heat tolerance of the NMD mutants is
based on the altered priming response of these NMD targets, we measured the amount of these
transcripts by RT-qPCR at NT, P and P+1d (Fig S9). Surprisingly, none of them was a genuine
NMD target, they were not upregulated in the _upf3_ mutant at NT. Moreover, their expression
changed similarly in the Col-0 and _upf3_ plants during the treatments, indicating that these
HSFs, which were previously proposed to be NMD-regulated, were not responsible for the
early heat tolerance of NMD mutant (Fig S9).


Previously we observed that HSR transcripts are upregulated in _upf3_ already at NT condition
(Fig 5B,C). These could contribute to the early heat tolerance of _upf3_ plants. To evaluate the
extent to which these mRNAs confer an overall stress-primed status to NMD deficient plants,
we calculated a priming index (PI, 1680 out of the total 2358 HS-responsive transcripts were
significantly induced in _upf3_ ): the _upf3_ /Col-0 log 2 FC values at NT (genotype difference) was
divided by the maximum WT induction at P/T (condition difference). In this ratio the PI=1
means full heat stress level of the given transcript. Across all 1,680 HSR genes the median PI
was 22% in _upf3_ plants (Fig S10). When the PI was calculated on ChIP-bound vs -unbound
sets to assess the HsfA2/A3 impact: the PI for the ChIP-bound group was almost zero. This
shows that the effect was entirely confined to genes not bound by HsfA2/A3 (35%, 30% and
18%; ChIP-bound groups −21%, −2% and −7%). These results suggest that the mild enhanced
HSR expression in the NMD mutant at NT is not caused by HsfA2/A3 binding and induction
(Fig S10A). Further, removing 49 known NMD targets (Fig S10A, _middle panel_ ), or 277 genes
with known or predicted NMD structural features (Fig S10. _bottom panel_ ) changed the PI
median only modestly (all HSR classes 22%, 22%, 20%; transient 34%, 33%, 33%; sustained
22%, 22%, 19%; delayed 16%, 15%, 14%), suggesting that the primed state is mostly an

indirect effect of NMD loss.


To try to define the nature of this priming, we next analyzed the GO enrichment of the HSR
genes already elevated in _upf3_ at NT (n = 1680). No typical heat-related terms were enriched.
Instead, “RNA modification” was the only term enriched, mirroring the genome-wide GO
analysis (Fig. S10B). This suggests that the stress priming status of _upf3_ under ambient


12


[bioRxiv preprint doi: https://doi.org/10.64898/2026.10.30.763111; this version posted November 2, 2026. The copyright holder for this](https://doi.org/10.64898/2026.10.30.763111)

preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.


∼
conditions (PI 22%) is achieved through non-canonical pathways, namely RNA modification.
This may transiently improve the heat tolerance of NMD mutants during early recovery,
although this protection does not seem to be lasting on longer time periods (Fig. 6D).


**NMD-controlled** _**HsfA2-II**_ **and** _**HsfA2-III**_ **isoforms may indirectly lower** _**HsfA2-I**_

**abundance**


NMD has previously been proposed to influence heat stress memory through the _HSFA2_ splice
isoforms _HsfA2-II_ and _HsfA2-III_, of which _HsfA2-III_ acts as a positive regulator of HSFA2
transcription (J. Liu et al., 2013; Sugio et al., 2009). Our RNA-seq data showed that the overall
_HSFA2_ transcript level is lower in _upf3_ at P/T (Fig. 5A). To see how the signal of the HSFA2
locus is distributed between isoforms, we measured the functional, non-NMD-target isoform
_HsfA2-I_ and the NMD-target isoforms _HsfA2-II_ and _HsfA2-III_ separately by RT-qPCR after
priming (P) in Col-0, _upf3_, _hsfa2_ and the double mutant _upf3;hsfa2_ (Fig. 7A–C). All three
isoforms were induced at P in Col-0 and _upf3_ and returned to basal levels at P+1d. None of
them were detected in _hsfa2_ or _upf3;hsfa2_, confirming their mutant status. Compared with Col0, _upf3_ had ∼35% less _HsfA2-I_ but about 6-fold more _HsfA2-II_ . _HsfA2-III_ showed the same
trend (∼2-2.5-fold increase). The accumulation of _HsfA2-II_ and _HsfA2-III_ in _upf3_ is consistent
with their degradation by NMD (J. Liu et al., 2013; Sugio et al., 2009).


_HsfA2-I_ is not an NMD target, so its decrease in _upf3_ must be indirect. One possible explanation
is feedback on the transcription of the locus. To test this, we measured the unspliced _HSFA2_
transcript ( _uHSFA2_ ) as a proxy for transcriptional activity. The _uHSFA2_ levels did not differ
between Col-0 and _upf3_ (Fig. 7D; note, the signal in _hsfa2_ and _upf3;hsfa2_ most likely arises
from a chimeric T-DNA transcript), arguing against reduced transcription. We therefore
propose that, with unchanged transcription, a shift in splicing from _HsfA2-I_ toward the NMDcontrolled isoforms _HsfA2-II/III_ may indirectly reduce _HsfA2-I_ abundance in _upf3_ (Fig. 7E).


**Discussion**


The heat shock response (HSR) is a multilayered process in which several regulatory
mechanisms act in parallel, including changes in RNA and protein homeostasis (Chung et al.,
2025; Li, Tang, Srivastava, Bassham, & Howell, 2020; Rahaman et al., 2025; Staacke, MuellerRoeber, & Balazadeh, 2025). RNA decay is one of these layers. The exoribonuclease XRN4
degrades transcripts that are no longer needed (Merret et al., 2013; Merret et al., 2015; Nguyen
et al., 2015) whereas NMD is a surveillance pathway that safeguards transcript fidelity by
eliminating erroneous transcripts (Szaker et al., 2025). Moreover, NMD also directly targets
normal mRNAs possessing NMD inducing features and indirectly controls the genes that are
regulated by these direct targets. Timely RNA decay facilitates efficient transcriptome
reprogramming and releases cellular resources for growth and recovery.
Building on our previous work (Szaker et al., 2025), we examined how NMD contributes to
the heat stress recovery in _A. thaliana_, and tested whether it regulates the HsfA2 pathway, a
link previously proposed between NMD and thermomemory (J. Liu et al., 2013; Sugio et al.,
2009). By comparing the thermotolerance of wild-type plants with the _nmd_ single and


13


[bioRxiv preprint doi: https://doi.org/10.64898/2026.10.30.763111; this version posted November 2, 2026. The copyright holder for this](https://doi.org/10.64898/2026.10.30.763111)

preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.


_nmd;hsfa2_ and _nmd;hsfa3_ double mutants in LAT assays, we showed that NMD plays a dual
role in thermotolerance regulation; it inhibits short-term heat tolerance in a HSFA2/3independent manner, whereas it stimulates long-term HS tolerance. To reveal the molecular
basis of this dual role, transcriptomes of wild-type and NMD mutants were compared at the
NT, P/T and P/T+1d timepoints. We found that NMD was active throughout the LAT assay
(Fig 3). Transcriptome analysis of NMD-deficient mutants revealed four main features. First,
more genes were upregulated than downregulated in the NMD mutant, and many of the
upregulated transcripts carry direct NMD features (Fig 3, S2). Second, NMD deficiency creates
a stress-primed state in NT plants through the upregulation of HsfA2-independent transcripts
involved in RNA modification and HSR (Fig 5B,C, S10). We propose that this stress-primed
state explains the early, transient thermotolerance; although the mechanism remains to be
determined. Third, HSFA2 activity is reduced in the _upf3_ plant, _HSFA2-I_ mRNA accumulates

to a lower level and the balance between functional and non-functional HsfA2 isoforms is

shifted at P/T (Fig 7). We propose that this low _HSFA2-I_ level leads to the diminished induction
of HSFA2-dependent memory genes; they accumulate to about 80% of wild-type levels (Fig.
5). Fourth, NMD deficiency shifted the transcriptome toward immune signaling at the expense
of auxin-related growth and cell-wall biogenesis during HS and recovery (Fig 4, S2). Because
maintaining a stress state is metabolically costly (Cadena-Ramos, Rodriguez-Pina, & De-laPena, 2026; Staacke et al., 2025; Zhou et al., 2013), this altered growth–defense balance and
the weakened memory regulon may explain the physiological decline (reduced chlorophyll,
lower photosynthetic efficiency, higher ion leakage, Fig 2) and late heat sensitivity of NMD
mutants (Fig 1, 6, S1, S8).
The elevated expression of immune genes in _upf3_, which appears mainly when plants are
exposed to heat, suggests that NMD restrains defense responses during heat stress, which may
allow plants to shift resources from defense to regrowth. This trade-off could matter in nature
when heat waves are followed by pathogen infections. Infection, in contrast, inhibits NMD, in
part by reducing the levels of UPF proteins (Gloggnitzer et al., 2014), whereas heat does not
reduce NMD efficiency (Fig. 5). Whether this crosstalk between heat and biotic stress is
biologically relevant remains to be tested, e.g. by generating NMD and immune double
mutants, and then infecting them upon/after heat treatment.
Our data indicate that the NMD-dependent regulation of HsfA2 expression and isoform
imbalance affects the HsfA2-regulon, but we detected no effect on the HsfA3-regulon. Since
HSFA2/HSFA3 heteromers promote HS memory (Friedrich et al., 2021), NMD-mediated
control of HsfA2 plays a modulatory role in thermomemory. HsfA2 and HsfA3 share their
targets (Kappel et al., 2023) but differ in kinetics (HsfA2 is strongly induced at P/T; HsfA3 is
mildly elevated at P/T and strongly at P/T+1d). The NMD-dependent modulation of HsfA2
may finetune the temporal composition of HSF complexes, and optimize the HS responses.
The S-HsfA2 peptide translated from _HsfA2-III_ binds HREs, thereby dampening the induction
of HS-responsive genes, and attenuating full-length HsfA2 activity (Chen et al., 2025). The
stabilized _HsfA2-III_ in _upf3_ may therefore contribute to the downregulation of the HS regulon.
However, this cannot explain the isoform changes in _upf3_ ( _HsfA2-I_ ~35% down, _HsfA2-II_ ~6fold up, _HsfA2-III_ ~2.2-fold up, _uHsfA2_ unchanged). The unchanged expression of the _uHsfA2_
transcripts excludes altered transcription, so the effect must be post-transcriptional. NMD acts
downstream of splicing and should destabilize _HsfA2-II/III_ without affecting _HsfA2-I_ . Non

14


[bioRxiv preprint doi: https://doi.org/10.64898/2026.10.30.763111; this version posted November 2, 2026. The copyright holder for this](https://doi.org/10.64898/2026.10.30.763111)

preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.


mutually exclusive feedback regulatory routes could include competition for splicing factors,
altered nuclear export and positive regulation of unproductive splicing. Regulation of a
functional isoform by an NMD-target isoform may be a widespread phenomenon and deserves
further study.
In summary, NMD functions as a key regulator of thermal acclimation. By destabilizing stressrelated transcripts under ambient conditions, constraining immune gene overexpression during
both ambient and heat stress, and fine-tuning the functional isoforms of HsfA2 (and
consequently the HsfA2 thermomemory pathway), NMD operates as an RNA quality control
pathway to orchestrate the transition from a stress response to physiological resetting/regrowth.
Together with recent evidence, these findings support a model in which NMD is a fundamental
mechanism for maintaining transcriptome integrity in fluctuating environments.

**Author contribution**

SH Abbas: genotyping, crosses, heat stress treatments, phenotyping, molecular measurements,
data analysis, writing; HM Szaker performed bioinformatic and data analysis; E Darkó:
physiological assays and data analysis, T Erős: genotyping, HS-phenotyping; D Silhavy: data
analysis, funding, writing, T Csorba: conceptualization, funding, writing. All authors

contributed to the final version.

**Conflict of Interest**

The authors have no conflicts to declare.

**Funding**
This work was supported by the National Research, Development and Innovation Fund [K139349,K-137722, K146300]; by Flagship Research Group Program of the MATE University;
by the EKÖP-MATE/2026/27/K University Research Scholarship Programme of the Ministry
of Science and Technology from the source of the National Research, Development and
Innovation Fund [H.M.SZ.]; by Tempus Public Foundation under Stipendium Hungaricum
Scholarship [S.H.A.]; open access charge by [K-146300]; [T.Cs.] supported by the Research
Excellence Programme of Hungarian University of Agriculture and Life Sciences.

**Data availability**
Data and materials are available upon request.


15


[bioRxiv preprint doi: https://doi.org/10.64898/2026.10.30.763111; this version posted November 2, 2026. The copyright holder for this](https://doi.org/10.64898/2026.10.30.763111)

preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.


1

2 **References:**

3

4 Akerfelt, M., Morimoto, R. I., & Sistonen, L. (2010). Heat shock factors: integrators of cell
5 stress, development and lifespan. _Nat Rev Mol Cell Biol, 11_ (8), 545-555.
6 doi:10.1038/nrm2938
7 Asghar, M.A., Balogh, E., Szalai, G., Galiba, G., Kocsy, G. (2022). Differences in the light8 dependent changes of the glutathione metabolism during cold acclimation in wheat
9 varieties with different freezing tolerance. _Journal of Agronomy and Crop Science_
10 _208_ (1). 65-75. doi: 10.1111/jac.12566
11 Arciga-Reyes, L., Wootton, L., Kieffer, M., & Davies, B. (2006). UPF1 is required for
12 nonsense-mediated mRNA decay (NMD) and RNAi in Arabidopsis. _Plant J, 47_ (3),
13 480-489. doi:10.1111/j.1365-313X.2006.02802.x
14 Cadena-Ramos, A. I., Rodriguez-Pina, A. L., & De-la-Pena, C. (2026). ROS and autophagy in
15 plant stress responses: adaptive partners in survival. _Plant Physiol Biochem, 231_,
16 110991. doi:10.1016/j.plaphy.2025.110991
17 Charng, Y. Y., Liu, H. C., Liu, N. Y., Chi, W. T., Wang, C. N., Chang, S. H., & Wang, T. T.
18 (2007). A heat-inducible transcription factor, HsfA2, is required for extension of
19 acquired thermotolerance in Arabidopsis. _Plant Physiol, 143_ (1), 251-262.
20 doi:10.1104/pp.106.091322
21 Charng, Y. Y., Mitra, S., & Yu, S. J. (2023). Maintenance of abiotic stress memory in plants:
22 Lessons learned from heat acclimation. _Plant_ _Cell,_ _35_ (1), 187-200.
23 doi:10.1093/plcell/koac313
24 Chen, W., Zhao, J., Tao, Z., Zhang, S., Bei, X., Lu, W., & Qi, X. (2025). Short heat shock
25 factor A2 regulates heat resistance and growth balance in Arabidopsis. _Elife, 13_ .
26 doi:10.7554/eLife.99937
27 Chung, K. K., Law, K. C., Zhao, Z., Ma, J., Zhan, X. T., Chiang, C. H., . . . Zhuang, X. (2025).
28 Biomolecular condensation of ERC1 recruits ATG8 and NBR1 to drive autophagosome
29 formation for plant heat tolerance. _Proc Natl Acad Sci U S A, 122_ (46), e2425689122.
30 doi:10.1073/pnas.2425689122
31 Danecek, P., Bonfield, J. K., Liddle, J., Marshall, J., Ohan, V., Pollard, M. O., . . . Li, H. (2021).
32 Twelve years of SAMtools and BCFtools. _Gigascience,_ _10_ (2).
33 doi:10.1093/gigascience/giab008
34 Darkó, É., Fodor, J., Dulai, S., Ambrus, H., Szenzenstein, A., Király, Z., Barnabás, B. (2011).
35 Improved Cold and Drought Tolerance of Doubled Haploid Maize Plants Selected for
36 Resistance to Prooxidant tert‐Butyl Hydroperoxide. _Journal of agronomy and crop_
37 _science 197_ (6), 454-465. doi: 10.1111/j.1439-037X.2011.00479.x
38 Degtiar, E., Fridman, A., Gottlieb, D., Vexler, K., Berezin, I., Farhi, R., . . . Shaul, O. (2015).
39 The feedback control of UPF3 is crucial for RNA surveillance in plants. _Nucleic Acids_
40 _Res, 43_ (8), 4219-4235. doi:10.1093/nar/gkv237
41 Friedrich, T., Oberkofler, V., Trindade, I., Altmann, S., Brzezinka, K., Lamke, J., . . . Baurle,
42 I. (2021). Heteromeric HSFA2/HSFA3 complexes drive transcriptional memory after
43 heat stress in Arabidopsis. _Nat Commun, 12_ (1), 3426. doi:10.1038/s41467-021-2378644 6
45 Gloggnitzer, J., Akimcheva, S., Srinivasan, A., Kusenda, B., Riehs, N., Stampfl, H., . . . Riha,
46 K. (2014). Nonsense-mediated mRNA decay modulates immune receptor levels to
47 regulate plant antibacterial defense. _Cell Host Microbe, 16_ (3), 376-390.
48 doi:10.1016/j.chom.2014.08.010


1


[bioRxiv preprint doi: https://doi.org/10.64898/2026.10.30.763111; this version posted November 2, 2026. The copyright holder for this](https://doi.org/10.64898/2026.10.30.763111)

preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.


49 Kappel, C., Friedrich, T., Oberkofler, V., Jiang, L., Crawford, T., Lenhard, M., & Baurle, I.
50 (2023). Genomic and epigenomic determinants of heat stress-induced transcriptional
51 memory in Arabidopsis. _Genome Biol, 24_ (1), 129. doi:10.1186/s13059-023-02970-5
52 Karanyi, Z., Mosolygo, L. A., Fero, O., Horvath, A., Boros-Olah, B., Nagy, E., . . . Szekvolgyi,
53 L. (2022). NODULIN HOMEOBOX is required for heterochromatin homeostasis in
54 Arabidopsis. _Nat Commun, 13_ (1), 5058. doi:10.1038/s41467-022-32709-y
55 Kerenyi, F., Wawer, I., Sikorski, P. J., Kufel, J., & Silhavy, D. (2013). Phosphorylation of the
56 N- and C-terminal UPF1 domains plays a critical role in plant nonsense-mediated
57 mRNA decay. _Plant J, 76_ (5), 836-848. doi:10.1111/tpj.12346
58 Kerenyi, Z., Merai, Z., Hiripi, L., Benkovics, A., Gyula, P., Lacomme, C., . . . Silhavy, D.
59 (2008). Inter-kingdom conservation of mechanism of nonsense-mediated mRNA
60 decay. _EMBO J, 27_ (11), 1585-1595. doi:10.1038/emboj.2008.88
61 Kertesz, S., Kerenyi, Z., Merai, Z., Bartos, I., Palfy, T., Barta, E., & Silhavy, D. (2006). Both
62 introns and long 3'-UTRs operate as cis-acting elements to trigger nonsense-mediated
63 decay in plants. _Nucleic Acids Res, 34_ (21), 6147-6157. doi:10.1093/nar/gkl737
64 Kurihara, Y., Matsui, A., Hanada, K., Kawashima, M., Ishida, J., Morosawa, T., . . . Seki, M.
65 (2009). Genome-wide suppression of aberrant mRNA-like noncoding RNAs by NMD
66 in Arabidopsis. _Proc_ _Natl_ _Acad_ _Sci_ _U_ _S_ _A,_ _106_ (7), 2453-2458.
67 doi:10.1073/pnas.0808902106
68 Li, Z., Srivastava, R., Howell, S. H., Chudalayandi, S., & Severin, A. (2026). The Role of
69 Nonsense-Mediated Decay in Alleviating Nonproductive RNAs Produced by Heat
70 Stress-Induced Alternative Splicing in Arabidopsis. _Plant Cell Environ, 49_ (3), 156171 1565. doi:10.1111/pce.70321
72 Li, Z., Tang, J., Srivastava, R., Bassham, D. C., & Howell, S. H. (2020). The Transcription
73 Factor bZIP60 Links the Unfolded Protein Response to the Heat Stress Response in
74 Maize. _Plant Cell, 32_ (11), 3559-3575. doi:10.1105/tpc.20.00260
75 Liao, Y., Smyth, G. K., & Shi, W. (2014). featureCounts: an efficient general purpose program
76 for assigning sequence reads to genomic features. _Bioinformatics, 30_ (7), 923-930.
77 doi:10.1093/bioinformatics/btt656
78 Lichtenthaler, H. K. (1987). Chlorophylls and carotenoids: Pigments of photosynthetic
79 biomembranes. _Methods in Enzymology 148_, 350-382. doi: 10.1016/007680 6879(87)48036-1
81 Liu, H. C., & Charng, Y. Y. (2013). Common and distinct functions of Arabidopsis class A1
82 and A2 heat shock factors in diverse abiotic stress responses and development. _Plant_
83 _Physiol, 163_ (1), 276-290. doi:10.1104/pp.113.221168
84 Liu, J., Sun, N., Liu, M., Liu, J., Du, B., Wang, X., & Qi, X. (2013). An autoregulatory loop
85 controlling Arabidopsis HsfA2 expression: role of heat shock-induced alternative
86 splicing. _Plant Physiol, 162_ (1), 512-521. doi:10.1104/pp.112.205864
87 Lloyd, J. P., & Davies, B. (2013). SMG1 is an ancient nonsense-mediated mRNA decay
88 effector. _Plant J, 76_ (5), 800-810. doi:10.1111/tpj.12329
89 Lloyd, J. P. B. (2018). The evolution and diversity of the nonsense-mediated mRNA decay
90 pathway. _F1000Res, 7_, 1299. doi:10.12688/f1000research.15872.2
91 Love, M. I., Huber, W., & Anders, S. (2014). Moderated estimation of fold change and
92 dispersion for RNA-seq data with DESeq2. _Genome Biol, 15_ (12), 550.
93 doi:10.1186/s13059-014-0550-8
94 Luha, R., Rana, V., Vainstein, A., & Kumar, V. (2024). Nonsense-mediated mRNA decay
95 pathway in plants under stress: general gene regulatory mechanism and advances.
96 _Planta, 259_ (3), 51. doi:10.1007/s00425-023-04317-7
97 Martin, M. (2011). Cutadapt removes adapter sequences from high-throughput sequencing
98 reads. _EMBnet J. 17_ (1), doi: https://doi.org/10.14806/ej.17.1.200


2


[bioRxiv preprint doi: https://doi.org/10.64898/2026.10.30.763111; this version posted November 2, 2026. The copyright holder for this](https://doi.org/10.64898/2026.10.30.763111)

preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.


99 Merret, R., Descombin, J., Juan, Y. T., Favory, J. J., Carpentier, M. C., Chaparro, C., . . .
100 Bousquet-Antonelli, C. (2013). XRN4 and LARP1 are required for a heat-triggered
101 mRNA decay pathway involved in plant acclimation and survival during thermal stress.
102 _Cell Rep, 5_ (5), 1279-1293. doi:10.1016/j.celrep.2013.11.019
103 Merret, R., Nagarajan, V. K., Carpentier, M. C., Park, S., Favory, J. J., Descombin, J., . . .
104 Bousquet-Antonelli, C. (2015). Heat-induced ribosome pausing triggers mRNA co105 translational decay in Arabidopsis thaliana. _Nucleic Acids Res, 43_ (8), 4121-4132.
106 doi:10.1093/nar/gkv234
107 Nasim, Z., Karim, N., Blilou, I., & Ahn, J. H. (2025). NMD-mediated posttranscriptional
108 regulation fine-tunes the NLR-WRKY regulatory module to modulate bacterial defense
109 response. _Plant Sci, 356_, 112528. doi:10.1016/j.plantsci.2025.112528
110 Nguyen, A. H., Matsui, A., Tanaka, M., Mizunashi, K., Nakaminami, K., Hayashi, M., . . .
111 Seki, M. (2015). Loss of Arabidopsis 5'-3' Exoribonuclease AtXRN4 Function
112 Enhances Heat Stress Tolerance of Plants Subjected to Severe Heat Stress. _Plant Cell_
113 _Physiol, 56_ (9), 1762-1772. doi:10.1093/pcp/pcv096
114 Obermeyer, S., Stockl, R., Schnekenburger, T., Kapoor, H., Stempfl, T., Schwartz, U., &
115 Grasser, K. D. (2023). TFIIS Is Crucial During Early Transcript Elongation for
116 Transcriptional Reprogramming in Response to Heat Stress. _J Mol Biol, 435_ (2),
117 167917. doi:10.1016/j.jmb.2022.167917
118 Ohama, N., Sato, H., Shinozaki, K., & Yamaguchi-Shinozaki, K. (2017). Transcriptional
119 Regulatory Network of Plant Heat Stress Response. _Trends Plant Sci, 22_ (1), 53-65.
120 doi:10.1016/j.tplants.2016.08.015
121 Ohtani, M., & Wachter, A. (2019). NMD-Based Gene Regulation-A Strategy for Fitness
122 Enhancement in Plants? _Plant Cell Physiol, 60_ (9), 1953-1960. doi:10.1093/pcp/pcz090
123 Rahaman, S., Schiffelholz, N., Mittal, N., Frohlich, K. E., Zavolan, M., & Becskei, A. (2025).
124 Heat shock induces silent ribosomes and reorganizes mRNA turnover. _Cell Rep, 44_ (10),
125 116447. doi:10.1016/j.celrep.2025.116447
126 Riehs-Kearnan, N., Gloggnitzer, J., Dekrout, B., Jonak, C., & Riha, K. (2012). Aberrant growth
127 and lethality of Arabidopsis deficient in nonsense-mediated RNA decay factors is
128 caused by autoimmune-like response. _Nucleic Acids Res, 40_ (12), 5615-5624.
129 doi:10.1093/nar/gks195
130 Staacke, T., Mueller-Roeber, B., & Balazadeh, S. (2025). Stress resilience in plants: the
131 complex interplay between heat stress memory and resetting. _New Phytol, 245_ (6),
132 2402-2421. doi:10.1111/nph.20377
133 Sugio, A., Dreos, R., Aparicio, F., & Maule, A. J. (2009). The cytosolic protein response as a
134 subcomponent of the wider heat shock response in Arabidopsis. _Plant Cell, 21_ (2), 642135 654. doi:10.1105/tpc.108.062596
136 Szadeczky-Kardoss, I., Szaker, H. M., Verma, R., Darko, E., Pettko-Szandtner, A., Silhavy,
137 D., & Csorba, T. (2022). Elongation factor TFIIS is essential for heat stress adaptation
138 in plants. _Nucleic Acids Res, 50_ (4), 1927-1950. doi:10.1093/nar/gkac020
139 Szaker, H. M., Darko, E., Medzihradszky, A., Janda, T., Liu, H. C., Charng, Y. Y., & Csorba,
140 T. (2019). miR824/AGAMOUS-LIKE16 Module Integrates Recurring Environmental
141 Heat Stress Changes to Fine-Tune Poststress Development. _Front Plant Sci, 10_, 1454.
142 doi:10.3389/fpls.2019.01454
143 Szaker, H. M., Verma, R., Szadeczky-Kardoss, I., Gal, N., Abbas, S. H., Darko, E., . . . Csorba,
144 T. (2025). Transcription fidelity and control of alternative splicing contribute to heat
145 stress survival in Arabidopsis. _Plant Cell, 37_ (11). doi:10.1093/plcell/koaf256
146 Weng, M., Yang, Y., Feng, H., Pan, Z., Shen, W. H., Zhu, Y., & Dong, A. (2014). Histone
147 chaperone ASF1 is involved in gene transcription activation in response to heat stress
148 in Arabidopsis thaliana. _Plant Cell Environ, 37_ (9), 2128-2138. doi:10.1111/pce.12299


3


[bioRxiv preprint doi: https://doi.org/10.64898/2026.10.30.763111; this version posted November 2, 2026. The copyright holder for this](https://doi.org/10.64898/2026.10.30.763111)

preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.


149 Wu, H. L., Ai, Q., Teixeira, R. T., Nguyen, P. H. T., Song, G., Montes, C., . . . Hsu, P. Y.
150 (2024). Improved super-resolution ribosome profiling reveals prevalent translation of
151 upstream ORFs and small ORFs in Arabidopsis. _Plant Cell, 36_ (3), 510-539.
152 doi:10.1093/plcell/koad290
153 Wu, T., Hu, E., Xu, S., Chen, M., Guo, P., Dai, Z., . . . Yu, G. (2021). clusterProfiler 4.0: A
154 universal enrichment tool for interpreting omics data. _Innovation (Camb), 2_ (3), 100141.
155 doi:10.1016/j.xinn.2021.100141
156 Yoshida, T., Ohama, N., Nakajima, J., Kidokoro, S., Mizoi, J., Nakashima, K., . . . Yamaguchi157 Shinozaki, K. (2011). Arabidopsis HsfA1 transcription factors function as the main
158 positive regulators in heat shock-responsive gene expression. _Mol Genet Genomics,_
159 _286_ (5-6), 321-332. doi:10.1007/s00438-011-0647-7
160 Zhou, J., Wang, J., Cheng, Y., Chi, Y. J., Fan, B., Yu, J. Q., & Chen, Z. (2013). NBR1-mediated
161 selective autophagy targets insoluble ubiquitinated protein aggregates in plant stress
162 responses. _PLoS Genet, 9_ (1), e1003196. doi:10.1371/journal.pgen.1003196
163

164


4


[bioRxiv preprint doi: https://doi.org/10.64898/2026.10.30.763111; this version posted November 2, 2026. The copyright holder for this](https://doi.org/10.64898/2026.10.30.763111)

preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.


# A



















**NT** **P/T 50’** **P/T 60’** **P/T 80’**


**Col** **-0** _**upf3**_ **Col** **-0** _**upf3**_ **Col** **-0** _**upf3**_ **Col** **-0** _**upf3**_


**Col** **-0** _**upf3**_ **Col-0** _**upf3**_ **Col** **-0** _**upf3**_ **Col** **-0** _**upf3**_





0,0 .



0,0 .



dying






|Col1|***<br>*** **<br>*** *** **<br>--00 upf3 --00 upf3 --00 upf3 --00 upf3<br>CCooll ooll CCooll CCooll<br>CC<br>NT 50' 60' 80'|
|---|---|
|||
|||
|||


|Col1|*** *** *<br>*** ***<br>*<br>--00 upf3 --00 upf3 --00 upf3 --00 upf3<br>CCooll CCooll CCooll CCooll<br>NT 50' 60' 80'|
|---|---|
|||
|||
|||


### Figure 1: NMD has ambivalent roles during heat stress response. (A) Treatment scheme for long- acquired thermotolerance (LAT) regime to test HS tolerance; priming (P: 37°C/60′, 21°C/90′, 44°C/45′) was followed by 2 days interval at 21°C, than triggering (T) heat treatment for 50′, 60′, or 80′ (P/T); red cicles show sample collection times for transcriptome analysis ( see latter ); (B) Phenotypes of wild-type (Col-0) and mutant upf3 taken on day 4 post-treatment (P/T+4d, early recovery) or (C) 14 days post-treatment (P/T+14d, late recovery). (D) Quantification of early phenotypes (+4d) shown in panel B ; (E) quantification of late phenotypes (+14d) shown in panel C ; seedlings were scored as not affected, mildly affected, severely affected, or dying ( see legend on right for each ). Data are mean ± SD of three independent experiments, and asterisks indicate significant differences to Col-0 (*p < 0.05; **p < 0.01; ***p < 0.001, unpaired two-sided t-test).


[bioRxiv preprint doi: https://doi.org/10.64898/2026.10.30.763111; this version posted November 2, 2026. The copyright holder for this](https://doi.org/10.64898/2026.10.30.763111)

preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.


## A


## B



80


60


40


20


0



Col-0



1600


1200


800


400


0



Col-0



80


60


40


20


0













1600


1200


800


400


0


|Col1|a<br>a<br>b<br>c|
|---|---|
|||
|||


|Col1|upf3 a<br>b<br>c<br>d|
|---|---|
|||
|||



LATP/T NT LAP/ **T**

**+4d** +14d+14d



NT LATP/T



NT LAP/ **T**



LAP/



**+** 4d4d



**T** NT LAP/


4d4d



**+4d**



NT LAP/ **T**
+14d+14d



NT LAP/ **T**

**+14d**


## C D



0,8





Col-0

_upf3_



0,6


0,4


0,2


0


0,8


0,6


0,4


0,2


0












|Col1|Col2|
|---|---|
|||


|Col1|Col2|Col3|
|---|---|---|
||||


|Col1|Col2|
|---|---|
|||





NT LATP/T

+4d+4d


Col-0
_upf3_



Y(II) Y(NPQ) Y(NO)













1


0,8


0,6


0,4


0,2


0


1


0,8


0,6


0,4


0,2


0




|Col1|Col2|
|---|---|
|||


|Col1|Col2|
|---|---|
|||





NT LATP/T

**+14d**


|Col1|a<br>b<br>b<br>c<br>ab<br>ab a<br>b<br>b b ab a<br>CC ooll --00 upf3 CCooll --00 upf3<br>NNTT LP A/ TT+ +1 14 4d d|
|---|---|
|||
|||
|||
|||


### **Figure 2: Physiological parameter changes of wild-type (Col-0) and NMD mutant upf3 plants in** response to heat. Treatment was done as shown in Figure 1A. (A) Total chlorophyll content (Chl a+b, µg/g FW); (B) Electrolyte leakage (%); (C) Effective quantum yield of PSII (Y(II)), regulated non- photochemical (Y(NPQ)), and non-regulated non-photochemical (Y(NO)) chlorophyll a fluorescence parameters; (D) Optimal quantum yield of PSII (Fv/Fm). All measurements were taken at ambient (non- treated, NT) and following 4 or 14 days of P/T recovery (P/T+4d or P/T+14d, respectively); for details see main text. Data are mean ± SD of six independent experiments ( A, B panels ) or 24 measurements ( C, D panels ), and letters indicate significantly different groups (p<0.05, Tukey’s post-hoc test).


## A B D



[bioRxiv preprint doi: https://doi.org/10.64898/2026.10.30.763111; this version posted November 2, 2026. The copyright holder for this](https://doi.org/10.64898/2026.10.30.763111)

preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

## C

## E



**Figure 3: NMD regulation of target genes at ambient (NT), heat stress (P+T) and recovery (P/T+1d).** (A) Expression
of core NMD _trans_ factors (UPF3, SMG7, UPF1, UPF2) in Col-0 (grey) vs _upf3_ (red); asterisks indicate genotype
significance at each condition, see below. (B) NMD-target derepression: estimated mean expression of previously defined
NMD-targets (Szaker et al., 2025) versus non-target genes in Col-0 (grey) and _upf3_ (red); asterisks indicate genotype
significant difference (Col-0 vs. _upf3_ ) at each condition (mixed-effects model). (C-E) NMD _cis_ -feature analysis of genes
differentially down-regulated (Dw; NT n=137, P+T n=313, REC n=265), all expressed genes (All; NT n=23,279, P/T
n=22,980, REC n=22,981), and up-regulated (Up; NT n=352, P/T n=475, REC n=552) in _upf3_ vs. Col-0 wild-type : (A)
3'UTR length, (B) percentage of genes carrying with intron in their 3'UTR, and (C) percentage of genes carrying an
upstream open reading frame (uORF). significance - p<0.05, ** p<0.01, *** p<0.001, ns = not significant, comparing
within each condition; Wilcoxon rank-sum test for 3'UTR length, Fisher's exact test for intron/uORF relative to background.


**Figure** **4:** **Immunity** **transcripts** **accumulate** **during** **heat** **stress** **in** _**upf3**_ **.** (A) K-means clustering (k=6) of expression
trajectories for top 2000 genes-by variance in _upf3_ vs. Col-0 plants, based on z-score normalized log 2 (FPKM+1) values under
non-treated (NT), heat-stress (P/T), and recovery (P/T+1d); thin lines are individual genes, bold lines show mean values. (B)
Sub-clustering of Cluster 5 (k=3); (C) GO enrichment analysis of Cluster 5 subclusters, computed with clusterProfiler
(hypergeometric test, BH-adjp<0.05); non-redundant BP terms were obtained by semantic-similarity reduction (rrvgo, threshold
0.7); (D) _Cis_ -regulatory element enrichment of 1 kb promoters (relative to TSS) for Cluster 5 subclusters vs. genome
background, tested by Fisher's exact test. Candidate elements: W-box (WRKY-binding), as-1 (TGA-binding, SA-responsive),
HSE (heat-shock element), GCC-box (ERF-binding, ethylene-responsive), ABRE (bZIP-binding, ABA-responsive), DRE/CRT
(drought/cold-responsive), TCA-element (SA-responsive); core elements used as controls: TATA-box, CAAT-box, GC-box.
significance * p<0.05, ** p<0.01, *** p<0.001; ns = not significant. (E) NMD _cis_ -feature comparison of Cluster 5 subclusters
vs. background (Bg): 3'UTR length, percentage of genes with an intron in the 3'UTR, and with a uORF; significance done by
Wilcoxon rank-sum test for 3'UTR length, Fisher's exact test for intron/uORF relative to background.


[bioRxiv preprint doi: https://doi.org/10.64898/2026.10.30.763111; this version posted November 2, 2026. The copyright holder for this](https://doi.org/10.64898/2026.10.30.763111)

preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.







hypergeometric test, BH-adjp ***p<0.001). (B) Estimated mean from a genome-wide mixed-effects model for the all HS
responsive, transient, sustained and delayed HSR genes vs. the non-HSR control, hypergeometric test, BH-adjp ***p<0.001, nonsignificant, ns; (C) Mean scaled expression trajectories of the three HSR groups with HsfA2 and HsfA3 trajectories (blue dashed
lines) at NT, P/T and REC; WT transcripts (grey), _upf3_ (red), thin lines represent single gene trajectories; (D) Overlap between
the HSR gene groups and the combined HsfA2/HsfA3 ChIP-seq target gene set (Kappel et al., 2023), gene numbers are shown.


[bioRxiv preprint doi: https://doi.org/10.64898/2026.10.30.763111; this version posted November 2, 2026. The copyright holder for this](https://doi.org/10.64898/2026.10.30.763111)

preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.



0,2.


0,0.



**NT** _**upf3**_ **P/T 50’** _**upf3**_ **P/T 60’** _**upf3**_ **P/T 80’** _**upf3**_

















_**hsfa2**_ _**u3;a2**_ _**hsfa2**_ _**u3;a2**_ _**hsfa2**_ _**u3;a2**_ _**hsfa2**_ _**u3;a2**_







dying











.

.

























_**hsfa2**_ _**u3;a2**_ _**hsfa2**_ _**u3;a2**_ _**hsfa2**_ _**u3;a2**_ _**hsfa2**_ _**u3;a2**_











dying







**Figure** **6:** **NMD's** **impact** **on** **heat-stress** **recovery** **is** **independent** **of** **the** **HsfA2** **pathway.** (A) Heat-stress phenotypes of
wild-type (Col-0), _upf3_, _hsfa2_, and _upf3;hsfa2_ ( _u3;a2_ ) mutants on day 4 post-treatment (P/T+4d, early recovery); (B)
Quantification of early-recovery phenotypes shown in _panel A_ ; (C) phenotypes of the same genotypes on day 14 post-treatment
(P/T+14d, late recovery). (D) quantification of late-recovery phenotypes shown in _panel_ _B_ ; data are mean ± SD of three
independent experiments, and asterisks indicate significant differences of single mutants vs. Col-0 or of the double mutant vs.
_hsfa2_ (*p < 0.05; **p < 0.01; ***p < 0.001, unpaired two-sided t-test).


## A C E


## B


|Col1|*<br>*** ***<br>* *** ***<br>--00 upf3 hsfa2 u3;a2 -00 upf3 hsfa2 u3;a2 --00 upf3 hsfa2 u3;a2<br>CCo ol l Co ol l- ooll<br>C CC<br>NT P P+1d|
|---|---|
|||
|||
|||
|||


|Col1|* *<br>--00 upf3 hsfa2 u3;a2 --00 upf3 hsfa2 u3;a2 --00 upf3 hsfa2 u3;a2<br>CCooll CCooll CCooll<br>NT P P+1d|
|---|---|
|||
|||
|||
|||



6,E-01 .


5,E-01 .


3,E-01 .


2,E-01 .



10,0

7,5

5,0

2,5

0,0



1,00.

0,75.

0,50.

0,25.

0,00.


## D



_HSFA2-II_

### ***


_uHSFA2_



.



0,8.

0,6.

0,4.

0,2.

0



_HSFA2-I_


_HSFA2-III_



0,E+00 .



.



.



.






### Figure 7: NMD modulates HSR through HsfA2-independent and -dependent routes. RT-qPCR of (A) HsfA2-I, (B) HsfA2-II, (C) HsfA2-III and (D) unspliced HsfA2 ( uHsfA2 ) transcripts in Col-0, upf3, hsfa2 and upf3;hsfa2 ( u3;a2 ) plants that were non-treated (NT), sampled after priming (P), or one day later (P+1d); expression is relative to PP2AA3 . Data are mean ± SD of at least three biological replicates; P-values are from two-tailed Student’s t-tests for differences relative to Col-0 within each condition (*P < 0.05, **P < 0.01, ***P < 0.001; ns, not significant). In (D), dashed bars ( hsfa2 and u3;a2 ) indicate signal most likely arising from a chimeric, non-functional hsfa2 T-DNA transcript, since the primers lie upstream of the T-DNA insertion. (E) Working model: black arrows debote activation, red bars inhibition, dotted blue arrows/bars indirect effects, and “?” untested links; for details see main text.