## Introduction

Temperatures higher than the plants’ adaptive thresholds cause heat stress (HS), which diminishes their growth, survival and productivity. Plants, by nature, have the ability to tolerate a certain degree of high temperature above their ambient temperature. This is known as ‘basal thermotolerance’. In addition to basal thermotolerance, plants also have the ability to acquire thermotolerance if they are pre-exposed to mild non-lethal higher temperatures, in a process called ‘heat priming’1,2,3,4. This heat priming induces not only an immediate response, but also leads to molecular and metabolic changes that persist for some time in the absence of the stress and that allow plants to respond more effectively to a second HS event. The second stress is known as the ‘triggering stimulus’3,4. The time period between priming and triggering stimuli is called the ‘memory phase’, during which stress memory can form and consolidate. Thermomemory refers to the maintenance of some, but not all, HS-induced changes, which ‘primes’, or prepares plants to respond more rapidly and strongly if such stress recurs before the memory fades. Experimental evidence indicates that thermomemory can range from several hours to days4,5,6 or even generations7. Thermomemory appears to be regulated, at least in part, by a set of genes different from those acting in basal thermotolerance and acquired thermotolerance1,4,5,8,9,10,11.

HS priming involves the activation of heat shock transcription factors (HSFs) that induce the expression of heat shock proteins (HSPs) which help to protect cellular proteins from denaturation and contribute to the repair or removal of misfolded proteins12,13. Several HSFs have been shown to be involved in the HS response9,14,15,16. In particular, class HSFA1s are considered as ‘master regulators’ of the HS response16, regulating the expression of several other transcription factor (TF) genes, including DEHYDRATION RESPONSIVE ELEMENT BINDING PROTEIN 2A (DREB2A), HSFA2, HSFA7a, HSFBs, and MULTIPROTEIN BRIDGING FACTOR 1C (MBF1C)17. While immediate responses to HS are relatively well studied, the molecular and physiological processes underlying thermomemory in plants are still not well understood. One of the possible mechanisms involves the accumulation of TFs in their inactive state after a priming stimulus (during the memory phase), and their activation upon experiencing the triggering stimulus18,19. For example, heat shock factor A2 (HSFA2) has been identified as a key component in the establishment of thermomemory in primed plants9. HSFA2 induces the expression of HEAT STRESS-ASSOCIATED 32 (HSA32), which was found to be required specifically for the maintenance of thermomemory and to participate in the maintenance of cellular homeostasis during high temperature stress8.

NAM/ATAF1/2/CUC2 (NAC) is a family of plant-specific TFs that have an important role in the response to different biotic and abiotic stresses, including water deficit, salinity stress and temperature stress20,21,22. Several NACs have been reported to be involved in the basal HS response in Arabidopsis thaliana and crop plants23,24,25,26,27. For example, overexpression of ANAC019 improved the thermotolerance of Arabidopsis, presumably by regulating the expression of HSFA1s and other HSFs including HSFBs25. The Arabidopsis membrane-associated NAC transcription factor gene NTL4 has also been found to be responsive to heat stress, and ntl4 mutants exhibited a higher cell viability and less H2O2 accumulation than WT, under HS27. In rice (Oryza sativa), expression of SNAC3 is induced by several stresses, including heat, and its overexpression enhances tolerance to high temperature, drought and oxidative stress26. The expression of the Triticum aestivum (wheat) NAC transcription factor TaNAC2L is also induced by high temperature, and its overexpression in Arabidopsis improves its acquired heat tolerance via regulating expression of its heat stress-related genes23. However, there is limited evidence on the involvement of NAC TFs in regulating thermomemory. Arabidopsis JUNGBRUNNEN1 (JUB1) is the only NAC TF that has so far been reported to regulate thermomemory. JUB1 expression is induced by HS and its expression pattern during the memory phase is similar to the expression pattern of other thermomemory associated genes, such as HSFA224. Overexpression of JUB1 improves the thermomemory of Arabidopsis seedlings24.

In this study, we set out to systematically identify NAC TFs responsive to thermomemory in Arabidopsis. We analyzed the expression patterns of nearly all (104) Arabidopsis NACs after priming HS, during the thermomemory phase, and after the triggering stimulus. One NAC identified in our screen is ARABIDOPSIS TRANSCRIPTION ACTIVATOR FACTOR 1 (ATAF1; also called ANAC002). ATAF1 was previously identified to be involved in the gene regulatory networks of senescence, drought and sugar signaling28,29,30,31. We found that overexpression of ATAF1 in transgenic Arabidopsis plants limits thermomemory capacity, while knocking out ATAF1 profoundly enhances thermomemory and survival of plants exposed to a more severe HS. To identify genes regulated by ATAF1 in thermomemory, we employed RNA-seq to compare the heat responses in wild-type plants and mutants with altered ATAF1 levels (overexpressors and knockout mutants). We found that ATAF1 is co-expressed with ANAC055. With respect to thermomemory, the ataf1/anac055 double-mutant displayed a phenotype similar to that of the single knock-out mutants ataf1 and anac055, suggesting that the two TFs cooperatively regulate thermomemory.

## Results

### Identification of priming-responsive and memory-associated NAC transcription factors

To investigate the expression pattern of NAC TFs in response to HS priming and during the memory phase, we treated Arabidopsis seedlings with a mild HS (priming) before exposing them to a severe HS (triggering stimulus). We collected samples at multiple time points after priming (up to 48 h into the thermomemory phase) and tested the expression of 104 NAC genes; seedlings kept in unprimed conditions were used as controls (Fig. 1).

Considering a 1.5-fold change as cut-off, 75 of the NACs were differentially expressed upon priming compared to control, while expression of 29 NAC genes was undetectable in the whole-seedling samples. The cut-off threshold was chosen considering that even moderate changes in TF expression levels can elicit strong downstream responses32. The expression of some NACs (including e.g., ANAC013 and ANAC029/AtNAP) in response to HS priming was similar to that reported in a previous study by Kilian et al.30. NACs were grouped in clusters based on their expression pattern using k-means clustering. Changes in the expression of many NACs was already detectable immediately after the priming treatment (Fig. 1, Supplementary Table S1). Three clusters comprising 33 genes in total (clusters 1, 2 and 3) appeared upregulated during early timepoints, and largely unchanged during later timepoints (Fig. 1), while 26 genes (in cluster 4) were initially strongly downregulated, and then strongly upregulated during later time points (Fig. 1). For the genes in cluster 5 we observed a strong downregulation at time point 0, which is immediately after completion of the 90 min HS, followed by a strong upregulation wthin the time frame of up to 2 h after the HS, and then a moderate to unchanged expression (Fig. 1). Taken together, the NAC TFs analyzed exhibited different expression profiles during HS priming and memory, suggesting different roles in the response to HS.

### The effect of heat priming and triggering on the expression of NAC transcription factors

Pre-exposure of plants to a moderate stress (priming) alters their response to upcoming severe temperature stress. We tested this, at the molecular level, by investigating the transcriptional response of NAC TFs. We tested whether or not heat-priming affects the expression of NACs after the triggering HS. To this end, we determined the expression of NAC TFs after the triggering stimulus (T plants) and compared it with the expression in primed and triggered plants (P + T), using a 1.5-fold change cut-off. To identify specific expression patterns between P + T and T plants, a k-means clustering approach was applied, using 10 clusters (Fig. 2, Supplementary Table S2). We noticed that expression of most NAC TFs was induced in primed + triggered plants, compared with triggered-only plants. Only few genes (cluster 9) showed an opposite expression pattern, being mostly repressed in primed + triggered plants and induced in triggered-only plants (Fig. 2). ATAF1 displayed a particularly interesting pattern: its expression remained almost unchanged in plants after priming + triggering (P + T), and expression was considerably induced in triggered-only conditions (without prior priming) (Fig. 2, cluster 9).

To test the hypothesis that ATAF1 is functionally involved in thermotolerance, additional physiological experiments were performed on ataf1 mutants and plants overexpressing ATAF1. Two other genes from clusters 4 (ANAC047/SHYG, AT3G04070) and 8 (ANAC013, AT1G32870) were included for comparison.

### Functional analysis of selected NACs for thermomemory

To investigate the impact of the selected NAC genes on thermomemory, we assessed their functional involvement in the process by analyzing the phenotype of overexpressor and knockout (or knockdown) mutants in response to a triggering HS given 3 days after heat priming (3 days of thermomemory). As shown in Supplementary Fig. S1, ANAC013 and SHYG overexpressors, shyg knockout and anac013 knockdown lines were not significantly different from wild type when compared in the thermomemory assay. However, ATAF1 appeared to be involved in thermomemory. A phenotypic analysis of ATAF1 transgenic plants for thermomemory showed a strong phenotype, where the ataf1–2 and ataf1–4 mutants had a significantly higher survival rate and fresh mass compared with WT plants, while plants overexpressing ATAF1 (called ATAF1-OE in the following) exhibited a severely reduced thermomemory (Fig. 3a–c). When the ataf1–4 mutant was transformed with an ATAF1 allele (pATAF1::ATAF1-GFP) which expresses an ATAF1-GFP fusion from the native ATAF1 promoter, thermomemory was restored to wild-type response (Supplementary Fig. S2). Taken together, these results suggest a negative regulatory role of ATAF1 in thermomemory.

### Potential target genes of ATAF1

As our data suggested that ATAF1 is a negative regulator of thermomemory we sought to understand the role of ATAF1 in response to a heat priming stimulus by identifying its regulated target genes. To this end, WT, ATAF1-OE and ataf1–4 mutant seedlings were treated with the priming stimulus (37 °C for 90 min), and samples were then collected at three time points (0 h, 1 h and 4 h; Fig. 4a) after the heat priming for gene expression profiling by RNA-seq. To eliminate the potential influence from other environmental factors, control samples (unprimed control) were collected at the same time points as the heat-treated samples.

To understand the relationship between samples in the RNA-seq, a hierarchical clustering analysis was performed and a heat map of the Pearson’s correlation coefficients of the expression profiles between all possible pairs of the samples was established (Fig. 4b). We observed a clustering between biological replicates, suggesting a high reproducibility of the experiment. The heat map also showed a strong separation between samples based on HS conditions, and by genotype (Fig. 4b). The overall similarities and differences between samples were confirmed by multidimensional scaling (MDS) (Fig. 4c), supporting a stronger separation by condition (heat and control) than by genotype. We also noticed a similarity of samples under control temperature and those after 4 h of heat stress, suggesting a rapid recovery of most changes in gene expression. Of note, the expression of ATAF1 in WT seedlings during the memory phase, as revealed by RNA-seq, followed the pattern determined by qRT-PCR (Supplementary Fig. S3).

For all three genotypes, we found the highest number of differentially expressed genes (DEGs) at time point 0, right at the end of the 90-min heat priming treatment (Fig. 5a). To identify priming HS-associated genes, we used the same criteria that were previously used by Sedaghatmehr et al.4. We investigated genes whose expression was induced after priming and remained high at all examined time points in the memory phase, as well as genes whose expression was down-regulated and remained low during the memory phase (Fig. 5a). Among the DEGs are thermomemory-associated HSPs4 whose expression in response to heat priming was similarly induced in ATAF1-OE, ataf1–4 mutant and WT, such as HSP22, HSP21, HSP17.4 and HSP18.2 (Supplementary Table S3). This finding corroborates the conclusion that these HSPs contribute to a general heat-response, which is common to all three genotypes.

To identify differential responses of ATAF1-OE and ataf1–4 mutant plants to heat priming, we analysed the RNA-seq data for genes upregulated in ATAF1-OE, compared to WT, but downregulated in ataf1–4 mutant plants, and vice versa (Fig. 5b). Next, to identify potential direct target genes of ATAF1, we analyzed the promoters of differentially expressed genes for either the presence of the ATAF1 binding site (reported by Garapati et al.31) or for binding by ATAF1 as determined by DNA affinity purification sequencing (DAP-seq) experiments33. The analysis was performed for each time point after the heat priming (0 h, 1 h and 4 h). In total, we identified 64 genes as likely direct ATAF1 targets (Supplementary Table S4). We then refined our analysis by searching for those potential target genes that are commonly regulated by ATAF1 in all three, or two, time points after heat priming. Five genes, i.e., AT2G31260 (ATG9), AT2G41640 (GT61), AT3G44990 (XTH31), AT4G27720 and AT3G23540 were commonly regulated at two of the three timepoints, suggesting they might be priming-associated direct targets of ATAF1 (Fig. 5c); no gene was regulated at all three time points. ATG9 is an autophagy gene34 and experimental evidence indicates that autophagy plays a role in the heat stress response35,36. Its involvement in thermomemory is yet to be confirmed. The two genes GT61 and XTH31 are related to cell wall biosynthesis and expansion. Glycosyltransferase 61 (GT61) belongs to the glycosyltransferase (GT) family. GT proteins have diverse functions in plants, but most of them are likely involved in the biosynthesis of polysaccharides and glycoproteins in the cell wall. In grasses, including rice (Oryza sativa) and wheat (Triticum aestivum), GT61 family enzymes are involved in the synthesis of xylans, one of the main components of the cell wall37,38,39. GT61 has not yet been shown to be involved in heat stress priming or tolerance. XTH31 belongs to the xyloglucan endotransglucosylase/hydrolase (XTH) family. Generally, members of the XTH family are involved in cell wall remodeling, expansion and morphogenesis suggesting a potential role in stress responses40. In Arabidopsis, XTH31 is involved in regulating cell wall xyloglucan content41. The xth31 loss-of-function mutant has a reduced sensitivity to ABA, and seeds germinate faster than those of the WT41,42. Transgenic soybean (Glycine max) plants overexpressing XTH31 from Arabidopsis display enhanced tolerance to flooding along with more adventitious roots and longer primary roots43. The two genes AT4G27720 and AT3G23540 are not well characterized. AT4G27720 is annotated to encode a major facilitator superfamily protein with a molybdate ion transporter function, while AT3G23540 is annotated to encode a protein of the alpha/beta-hydrolase superfamily. The α/β-hydrolase enzymes are involved in various processes, including biosynthesis, metabolism, signal transduction, gene regulation44, and in the plants’ response and tolerance to salinity stress45.

### Co-regulatory network of ATAF1 reveals potential co-regulation with ANAC055

Genes with similar expression patterns often share similar functions and are potentially regulated by the same transcription factors. In order to identify genes co-regulated with ATAF1 in response to heat priming, a co-expression analysis was performed using the transcriptomic data of ATAF1-OE, ataf1–4 mutant and WT plants, under control condition and upon exposure to heat. A weighted correlation network analysis was used to find modules of highly correlated genes. This method is commonly used to investigate relationships between genes and to identify gene candidates for further analysis46. The co-expression analysis revealed 40 modules in our data set (Fig. 6a). These modules contained groups of genes expressed in a similar way. It is likely that genes regulated directly by ATAF1, in response to heat, display similar expression patterns; hence, we took publicly available data on genes that have already been identified to be potentially targeted by ATAF1 (in general, not just by heat stress) and queried if these genes were present in our co-expression modules. The publicly available data of the potential direct targets were generated using a DAP-seq assay33. Putative target genes were designated as having a DAP-seq peak within the first 1000 bp upstream of the transcription start site. Thus, we compared the list of genes that are proposed direct targets of ATAF1 (from O’Malley et al.33) with the genes that are clustered in the co-expression analysis; we noticed that the potential targets of ATAF1 were overrepresented in co-expression clusters 10 and 13 (Fig. 6b). These clusters needed further investigation because they likely contained direct target genes of ATAF1, in response to heat stress. The clusters were further interrogated with the proposed TF targets identified using the DAP-seq assay developed by O’Malley et al.33. We found that three of the clusters (10, 13, 21) showed significant enrichment for genes targeted by both ATAF1 and ANAC055, suggesting that the genes in these clusters might be co-regulated by the two NACs (Fig. 6b).

### ATAF1 and ANAC055 appear to have a common role in thermomemory

Our co-expression analysis showed that ATAF1 and ANAC055 had a large number of shared co-expressed targets in response to HS (Supplementary Table S5). We therefore tested if ANAC055 is also functionally involved in regulating thermomemory. To this end, we assessed the thermomemory phenotype of transgenic lines with altered expression of ANAC055 (Fig. 7a,b). Two anac055 mutants (anac055-1 and anac055-2) were tested, and both showed a substantial increase in fresh mass ratio, survival, and chlorophyll content compared with WT, while plants overexpressing ANAC055 showed a decrease in those parameters (Fig. 7b, Supplementary Fig. S4). These results indicate a negative role for ANAC055 in the regulation of thermomemory, similar to that of ATAF1. To test whether or not ATAF1 and ANAC055 worked in a fully or partially redundant manner, we generated an ataf1/anac055 double mutant and exposed seedlings to HS and compared their phenotypes with the phenotypes of the corresponding single-gene mutants. The fresh mass ratio of the ataf1/anac055 double-mutant was higher than that of the WT, but not significantly different from that of the single-gene knockout mutants ataf1–4 and anac055-1 (Fig. 7c). This result indicates that ATAF1 and ANAC055 require each others function to control thermomemory.

## Discussion

HS is one of the major abiotic stresses negatively affecting plant growth and reproduction globally. The impact of HS on crop production increases with climate change. Prior exposure to mild HS (heat priming) enables plants to acquire and maintain tolerance to HS by establishing a cellular memory of a prior HS (thermomemory)1,2,3,4. Here, we studied the role of NAC TFs in thermomemory. We found that several NACs substantially responded at their expression level after heat stress and that the changes in expression persisted during the memory phase. These findings are similar to those reported for the expression pattern of other thermomemory-associated genes during the memory phase (such as HSFA2, HSA32, and JUB1)8,9,24.

We then analyzed the expression of NAC TFs after the triggering stimulus (second heat stress) to investigate if their expression was affected by a priming treatment. Our data show that NAC expression was altered after the triggering stimulus, indicating the presence of transcriptional memory that controls their expression. A similar expression pattern has previously been reported for drought memory, where the expression of a subset of DEGs is altered after a second drought stress. Ding et al.47 classified the memory genes into four distinct groups, each showing different expression patterns: genes with increased (or decreased) expression levels after a first stress, and a further increased (or decreased) level after a second stress, which they, respectively, denoted as (+/+, or −/−); ‘revised memory genes’ that are induced after a first stress and then decreased after the second stress (+/−), or the opposite (−/+). Many NACs showed an expression pattern similar to those reported by Ding et al.47.

We found that ATAF1 (ANAC002) expression increases rapidly right after the priming HS, and then decreases to below control levels during the memory phase (Fig. 1; cluster 2). We also found that ataf1 knockout mutants were more tolerant to heat stress than the wild type, revealing a negative regulatory role of ATAF1 in HS memory. ATAF1 was previously identified as an important regulator of senescence and the response to drought stress29,31,48. It has been well established that ATAF1 promotes senescence by directly regulating the expression of two TFs involved in senescence and chloroplast maintenance: ANAC092 (ORE1) and GOLDEN2-LIKE1 (GLK1)31. ATAF1 also regulates drought response by regulating the expression of 9-CIS-EPOXYCAROTENOID DIOXYGENASE 3 (NCED3), which encodes a key enzyme of ABA biosynthesis31,49.

We used RNA-seq analysis to identifying early transcriptional changes controlled by ATAF1. Several mechanisms (genes) have been previously identified to be involved in basal heat tolerance: they include ROS detoxification, lipid signaling, and activation of heat shock proteins and heat shock factors8,9,10,27,50. Our data found similar responses in ATAF1-OE, ataf1–4 mutant and WT plants, where these mechanisms were similarly activated.

Co-expression analysis has been established as a powerful tool to identify TF target genes51,52. For instance, Jensen et al.49 performed a co-expression analysis across publicly available microarray datasets and identified 25 genes co-expressed with ATAF1. The promoter regions of these ATAF1 co-expressors were significantly enriched for ATAF1 binding sites. We performed a co-expression analysis and combined this information with publicly available data from a DAP-seq approach33. We found that four co-expression clusters were enriched with predicted ATAF1 target genes; and those were also enriched for target genes of another NAC TF, ANAC055. We found many target genes commonly regulated by ATAF1 and ANAC055, suggesting at least partial overlap between the two TFs with respect to their regulatory capacity. Often, TFs act in a redundant manner. For example, the transcription factors WRKY11 and WRKY17 operate as negative regulators of basic resistance to Pseudomonas syringae pv. tomato (Pst)50. The loss of function of WRKY11 increases the resistance toward Pst. An expression analysis of selected genes in single and double mutants revealed that both TFs modulate transcriptional changes in response to pathogen infection; however, each one of them acted either specifically or in a partly redundant manner, depending on the specific target gene50. In our work, we identified ATAF1 and ANAC055 as negative regulators of thermotolerance. Phenotypic and gene expression data of ataf1 and anac055 single and double mutants showed a partial regulatory redundancy between ATAF1 and ANAC055.

Our results suggest that ATAF1 and ANAC055 work both as negative regulators of thermomemory because (i) the single ataf1 and anac055 mutants as well as the double-mutant plants have a similar phenotype in thermomemory, (ii) they appear to be co-expressed, and (iii) their potential target genes overlap. Further protein-interaction studies are needed to test if ATAF1 and ANAC055 also interact physically, for instance as heterodimers or if they are only connected via their gene-regulatory networks. This interaction needs to be investigated in a temporal and spatial context to evaluate its relevance for thermomemory.

It has recently been suggested that the heat-inducible transcription factor HSFA2 is involved in thermomemory, particularly regulating HS memory genes in the shoot apical meristem (SAM), suggesting that the establishment of thermomemory involves organ-specific mechanisms53,54. Our results show that in ataf1–4 lines, a higher proportion of plants displayed full recovery, i.e. seedlings and cotyledons remained green compared to WT plants. In plants that only partially recovered, cotyledons remained bleached and the developing true leaves emerged green (Supplementary Fig. S4). The green emerging leaves suggest that mechanisms that protect the SAM, where new leaves develop, play a role in thermomemory.

Most of the current understanding of thermotolerance and thermopriming is based on Arabidopsis seedlings. However, there have been some limited studies performed to assess the thermopriming in crop plants. For instance, heat priming of winter wheat during the stem-elongation, bolting and anthesis stages significantly enhanced grain-yield under heat stress during the grain filling phase55. Another mechanism for improving thermotolerance may be through root endophytes which appear to enhance thermotolerance of crops by impacting the expression of thermomemory associated gene HSFA256. Our study shows the importance of two genes, ATAF1 and ANAC055, as negative regulators of thermomemory in seedlings; it remains to be tested if later growth or seed yield are also modified under recursive HS. At least for seedlings tested here, a biomass penalty under non-stress conditions was not observed. Once a better understanding of thermomemory is established in model organisms and crops, genome editing technology could be employed to enhance crop performance in the field in anticipated future climatic conditions.

## Materials and methods

### Plant materials

Arabidopsis thaliana ecotype Col-0 was used as wild type and obtained from the European Arabidopsis Stock Center (NASC) (http://arabidopsis.info/). The ATAF1-OE, ataf1-2 (SALK-057618) and ataf1–4 (GABI-Kat GK565H08) seeds were described previously in Garapati et al.31. Seeds of the T-DNA insertion lines of ANAC055 (anac055-1; SALK_014331 and anac055-2; SALK_011069) were obtained from The European Arabidopsis Stock Centre (NASC) seed collection (http://arabidopsis.info/). The ANAC047 transgenic lines, i.e., ANAC047-OE (35S::ANAC047-GFP) and the two knockout lines (T-DNA insertion lines shyg-1/SALK-066615 and shyg-2/GABI-Kat GK-343D11) were previously described by Rauf et al.57. Seeds of ANAC013-OE and knockdown (anac013-kd) lines were provided by Prof. Frank van Breusegem (Ghent University, Ghent, Belgium) and described in De Clercq et al.58.

### Plasmid construction and plant transformation

For 35S::ANAC055 (ANAC055-OE), the ANAC055 open reading frame was amplified by PCR from Arabidopsis Col-0 leaf cDNA and inserted into pUni/V5-His-TOPO (Invitrogen). The cDNA was then cloned via added PmeI-PacI sites into a modified pGreen0229-35S plant transformation vector59. The resulting pGreen0229-35S::ANAC055 vector was electroporated into Agrobacterium tumefaciens strain GV3101 and transformed into Arabidopsis wild type via floral dip. To construct pATAF1::ATAF1-GFP, a genomic fragment encompassing the 1.5-kb ATAF1 promoter and the ATAF1 coding region (without the translation stop codon) was inserted upstream of the green fluorescence protein coding sequence (in-frame fusion) in plasmid pK7FWG2,060. The cloning replaced the CaMV 35S promoter present in the plasmid with the ATAF1 promoter. The ataf1 knockout complementation line was generated by transforming the ataf1–4 mutant31 with the pATAF1::ATAF1-GFP construct.

### Double mutant

The ataf1/anac055 double-knockout mutant was generated by crossing the homozygous ataf1–4 mutant with the homozygous anac055-1 mutant. Plants from the F2 generation were PCR-screened to select homozygous double-mutant plants. The list of primers used for genotyping is given in Supplementary Table S6.

### Plant growth conditions

Arabidopsis seeds were surface-sterilized by washing them with 70% ethanol for 2 min; the ethanol solution was then replaced with 20% sodium hypochlorite and seeds were incubated in the solution for 20 min, with continuous shaking. Subsequently, seeds were washed four times with sterile water and left to dry on sterilized filter paper, in a laminar clean bench. Approximately equal numbers of seeds (around 500 seeds) were germinated on small petri dishes containing half-strength Murashige and Skoog (MS) medium supplemented with 1% (w/v) sucrose. For stratification, plates were kept in a dark cold room (8 °C) for 2 days before placing them in a growth chamber (photoperiod 16 h/8 h: light/dark at 22.5 °C, under 120 μmol m−2 s−1 photon irradiance). Five-day-old seedlings were used for HS assays.

### Heat stress treatment

For NAC expression profiling, 5-day-old Arabidopsis seedlings were exposed to a HS regime. Plants were primed with mild heat stress at 37 °C for 90 min in a growth chamber, and subsequently returned back to normal growth condition to recover for 2 days to assess their thermomemory. Plants were then challenged with a triggering HS of 44 °C for 45 min. Arabidopsis seedlings were collected during the memory phase, and after the triggering stimulus, at the time points indicated in Fig. 1. Whole seedlings collected from one single plate were considered as one biological replicate. Control samples (without treatment) were collected at the same time to reduce circadian effects. Three biological replicates per treatment, per time point, and per genotype were analysed. Samples were directly frozen in liquid nitrogen, and stored at − 80 °C until further analysis.

### Chlorophyll content

The Chlorophyll content was measure using the protocol of Hiscox and Israelstam61. Frozen seedling of ten were immersed in 0.5 mL preheated dimethylsulfoxid (DMSO, 65 °C) and incubated at 65 °C for 20 min. For the measurement the extract was diluted 1:10 or 1:100 with DMSO. Absorbance was recorded at 663 nm and 645 nm, respectively, for chlorophyll a and b. The following formula62 was used to calculate the concentration of chlorophyll a, b and total chlorophyll as mg g−1 fresh mass (FM):

$${Chl}_{a }\left(\text{g L} ^{-1}\right)=0.0127\left({OD}_{663}\right)-0.00269\left({OD}_{645}\right),$$
$${Chl}_{b}\left(\text{g L}^{-1}\right)=0.229\left({OD}_{645}\right)-0.00488\left({OD}_{663}\right),$$
$${Total}_{Chl}\left(\text{g L}^{-1}\right)= {0.202Chl}_{a}+ {0.00802Chl}_{b},$$
$${Total}_{Chl} \left(\text{mg g}^{-1}\right)=\frac{20.2{Chl}_{a}+8.02{Chl}_{b}}{FM}.$$

Data was normalized to the respective control of each genotype.

### RNA isolation, cDNA synthesis and qRT-PCR for NAC expression profiling

For expression profiling of NAC genes, total RNA was extracted using Trizol (Ambion, Life Technologies), followed by purification of RNA using columns of the PureLink RNA Mini kit (Ambion, Life Technologies) according to the manufacturer’s instructions. RNA concentration was measured using a NanoDrop spectrophotometer (Thermo Scientific); genomic DNA was digested using the TURBO DNA-free Kit (Ambion) according to the manufacturer’s instructions. The absence of genomic DNA contamination was confirmed by qRT-PCR, using intron-specific primers of control gene AT5G65080 (for primer sequences, see Supplementary Table S6). cDNA was synthesized from 4 µg high-quality RNA, using the Revert Aid H Minus First Strand cDNA Synthesis Kit (Thermo Scientific), following the instructions of the manufacturer.

The qRT-PCR reaction was performed using SYBR Green master mix (Applied Biosystems): cDNA 0.5 μL, SYBR Green master mix (2×) 2.5 μL, forward and reverse primer mix (each 0.5 μM) 2 μL. The reactions were incubated in an ABI PRISM 7900 HT sequence-detection system (Applied Biosystems) using the following protocol: temperature was first set at 95 °C and kept constant for 10 min, and then underwent a series of 40 cycles of 95 °C for 15 s and 60 °C for 1 min. ACTIN2 (AT3G18780) was used as the reference gene for data analysis. The primer platform of the 104 Arabidopsis NAC TF genes was established using the QuantPrime tool63. Primer sequences are given in Supplementary Table S6. After 40 reaction cycles, a melting curve was generated by increasing the temperature from 60 to 95 °C, with a ramp speed of 1.9 °C min−1, to verify the amplification of the desired products. The efficiency of the reaction was estimated using LinRegPCR software, as described by Caldana et al.64.

Data were analyzed using the comparative Ct method, as described in Kamranfar et al.65. Briefly, the delta Ct value (ΔCt) was calculated by normalizing each Ct value with the Ct value of the reference gene ACTIN2. Then, the level of gene expression was expressed as the difference between an arbitrary value of 40 and the ΔCt value (a high 40-ΔCt value indicates a high gene expression level). A threshold of 1.5-fold change was used to select differentially expressed genes.

### RNA isolation and RNA-seq

For the RNA-seq analysis, 5-day-old Arabidopsis seedlings of the three genotypes were used: WT, ataf1–4 mutant, and ATAF1-OE. After treatment with the heat priming temperature (37 °C for 90 min), seedlings were harvested at time points 0 h, 1 h and 4 h during the memory phase.

The transcriptomes of the three genotypes WT, ataf1–4 and ATAF1-OE was analyzed using RNA-seq technology. Total RNA was isolated from 5-day-old Arabidopsis seedlings of the three genotypes. Whole seedlings collected from each plate (pool of at least 10 seedlings) were considered as one biological replicate. Three biological replicates per treatment, per time point, and per genotype, were used, which made 72 samples in total. RNA was isolated using Direct-zol RNA MiniPrep Plus (Zymo Research, CA, USA) according to the manufacturer’s instructions. Quality was assessed using a Bioanalyzer (Agilent Technologies, Waldbrann, Germany), and quantity was determined using Qubit (Invitrogen, Life Technologies). High-quality RNA (3 µg) was used to prepare libraries for sequencing. RNA-sequencing 150-bp paired-end (PE) libraries were prepared from mRNA-enriched samples, using the NEB Next Ultra DNA Library Prep Kit for Illumina sequencing (New England Biolabs), according to the manufacturers’ instructions. A total of 12 samples were pooled in one lane of the flow cell. RNA was sequenced, using an Illumina HiSeq4000 machine, at KAUST Core Labs.

### RNA-seq data analysis

HISAT was used to align the RNA-seq reads onto the genome of Arabidopsis thaliana (TAIR10)66. Only sequencing reads that mapped uniquely on the genome were retained. Htseq-count was used to count the number of sequencing reads of the transcripts67. VOOM, a transformation method that allows the use of RNA-seq data for the linear model framework, was used to calculate the transcript abundance, and quantile normalization was used to normalize the abundance distribution between samples68. EdgeR and LIMMA69,70 were used as a framework for calculating the differentially expressed genes. R, a general statistical environment, was used for data management, analyses, and generating figures71.

The customized linear contrast was set up to calculate the differential expression between genotypes, or temperatures. More specifically, second-order interaction effects were calculated using the contrast framework70. For example, a calculation could be made to quantitatively assess the changes in gene expression under heat, in comparison with the control for ataf1 and WT: (heatataf1 − controlataf1) − (heatWT − controlWT). All statistical significance cut-offs were adjusted for multiple testing by controlling the False Discovery Rate (Benjamini-Hochberg) at 0.0572 using “decideTests” (…, method = “global”, adjust.method = “BH”, p.value = “0.05”, lfc = 0.

For all genes and samples, a weighted gene correlation network analysis was performed to identify groups of co-expressed genes46, using default parameters. The soft thresholding power parameter was set at 6, which is the value, based on graphical examination, that the optimal network conditions are met (maximum scale-free fit index and minimum mean connectivity). For multi-dimensional scaling, the cmscale function in R was used.

Transcription factor binding sites (TFBS) acquired via DAP-seq as reported by O’Malley et al.33 were extracted from the publicly available plant cistrome database. DAP-seq binding motif peaks were extracted in narrowPeak format and mapped to the Arabidopsis genome. Distance from peak location to closest transcription start site (TSS) was then computed via the bedtools ‘closest’ command73, where TSS locations were taken from the TAIR 10 annotation file74. Putative ATAF1 and ANAC055 target genes were defined as having at least one DAP-seq peak within the first 1000 bp upstream of the respective TSSs. Statistical analysis for enrichment of target genes within the co-expression clusters was performed in R, using a hypergeometric test for overrepresentation71.

### Legislation statement

The experimental research and field studies on plant materials comply with relevant institutional, national, and international guidelines and legislation.