Transgenerational decline through insidious effects of drought memory

Climate change increases the occurrence of drought stress, which challenges the sustainability of agroecosystems, forests, and gardens and impairs plant productivity. Nonetheless, repeated drought episodes may train a single plant to better withstand subsequent water deficit (Ding et al., 2012) – a phenomenon referred to as priming, hardening, or acclimation (Gallusci et al., 2023; Harris et al., 2023). There is also evidence that plants may be trained across generations, which requires a transmittable form of stress 'memory'. This concept has gained biological support with the discovery of molecular mechanisms underlying stress memory, including epigenetic modifications that may be transmitted to the progeny (Crisp et al., 2016). Sexual reproduction resets most of these modifications, and the few examples of transgenerational drought memory show little to no impact on integrated phenotypes such as yield (Liu et al., 2022; Gallusci et al., 2023). Clonal reproduction – the dominant propagation mode in horticulture, arboriculture, and many ecosystems – is more likely to retain epigenetic modifications or any morpho/physiological change involved in stress memory. A pioneering study in poplar showed that clones propagated at distinct nurseries for several decades diverged in their transcriptome and physiological response to drought when grown in a common garden (Raj et al., 2011). In white clover, clonal offspring are slightly less sensitive to drought when the parent has also been exposed (Rendina González et al., 2017). To our knowledge, whether effects of drought memory may accumulate across numerous generations remains largely unexplored. Here, we tested whether training petunia plants to drought, through recurrent exposure to water deficit within and across clonal generations, may provide them with transgenerational and adaptive drought memory. We exposed petunias to recurrent water deficit for > 25 clonal generations, but trained plants did not outperform naive plants under drought and started declining after c. 15 generations, ultimately leading to extinction. Our results emphasize an underappreciated vulnerability of plants to recurrent drought through insidious effects that can be transmitted clonally. Starting from one single plant of petunia (Petunia × hybrida, generation #00), we propagated two clonal lines through cuttings for more than 25 generations. The two lines were grown together in the same growth chamber with all environmental conditions being identical, except for the irrigation regime. We generated a 'naive line' exposed to a soil water potential always above −100 hPa (sub-irrigation about once a day), corresponding to well-watered conditions, as well as a 'trained line' recurrently exposed to less than −500 hPa (sub-irrigation every 4 d), corresponding to chronic drought conditions around the wilting point (Supporting Information Fig. S1). After c. 1 month in these contrasting conditions, plants were regenerated through cuttings, subjected again to their assigned watering regime and so forth (see the Materials and Methods section). We first assessed the behavior of both lines maintained in their irrigation regime. We observed morphological and physiological effects of water restriction that were consistent across successive generations. Growth was obviously affected by chronic water deficit (Fig. 1a): shoot biomass was reduced on average by 50% in dry weight and in fresh weight across generations #02 to #10 (Fig. 1b; Fig. S2a). Shoot branching, shoot apparent area, flower area, and root development were all negatively affected by water restriction (Fig. S2b–e). Leaves of well-watered petunias showed high water content (94–95%; Fig. 1c). Chronic water deficit significantly reduced leaf water content but only by c. 1–3%, while the plants here had not been irrigated for 3–4 d (Fig. 1c). These results suggest isohydric behavior, with a tight regulation of tissue water status typically mediated by stomatal control (Tardieu & Simonneau, 1998; Hochberg et al., 2018). Consistent with this premise, analysis of gas exchange showed a consistent reduction of stomatal conductance under chronic water deficit, at the expense of photosynthesis (Fig. 1d,e). Moreover, when both lines were subjected to soil dehydration, shoot water potential decreased more rapidly in the naive line than in the trained line because naive plants were larger and thus consumed more water, but after 3 d of dry-down, water status was similar between lines due to feedback of stomatal regulation (Fig. 1f). To test for evidence of drought memory, we then performed a series of reciprocal transfer experiments. On generation #08, after a post-rooting period of 1 wk with all plants in well-watered conditions, we placed a batch of naive plants under water restriction and a batch of trained plants under well-watered conditions. We found no conspicuous effect of the clonal history on plant traits, neither in water deficit nor in well-watered conditions. There were no significant differences in shoot dry biomass, fresh biomass, branching, apparent area, or flower area between the naive and trained lines exposed to the same irrigation regimes during 3 wk (Figs 1g, S3a–d). Similar results were obtained from generations #09 and #10. We also measured the dynamics of leaf water content as water was withheld in the four conditions, corresponding to the first drought event for the well-watered plants. There was no significant difference between lines as soil desiccation progressed (Fig. 1h, three-way ANCOVA, Line × Time interaction). At incipient wilting, there was no difference in shoot water potential between the naive and the trained line (Fig. S3e). Overall, there were no strong beneficial or detrimental effects of the training regime experienced throughout 8–10 generations. Closer examination of the phenotypes revealed slight evidence of drought memory, but we did not find evidence it was adaptive. On generation #15, we compared the dynamics of response to the installation of chronic water deficit in naive and trained plants, through exposure to successive drought cycles. Biomass accumulated similarly between lines as the number of drought cycles progressed, yet the naive line performed slightly better (Fig. S4a). Total leaf area was also comparable between lines (Fig. S4b), but the trained line developed more leaves of smaller size compared to the naive line (Fig. 1i,j), suggesting plant morphology expressed a form of drought memory. Leaf water content in the trained line was slightly higher (by c. 1%), but no different behavior during soil desiccation was observed for this trait between the lines (Fig. 1k, three-way ANCOVA, Line × Time interaction). The same was true for quantum yield in darkness or light (Fig. S4e,f). Similar conclusions could be drawn from generation #17 (Fig. S5). Moreover, starch and sugar contents were not depleted in either line, even after 6 d of withholding water (Fig. 1l, generation #18). Reciprocal transfers at generations #20-24 showed virtually identical stomatal responses to light and leaf excision between the trained and the naive lines (Fig. 1m). No significant difference in net assimilation or intrinsic water use efficiency was observed either (Fig. S6). We noticed, however, that the trained line showed enhanced leaf thickness and a trend for higher mesophyll porosity (Figs 1n, S7a,b), also observed upon short-term drought hardening in tobacco (Khan et al., 2020). Compared to the naive line, leaf mass per area (LMA) was similar or higher in the trained line depending on the generation (Fig. S7c). These minor physiological and anatomical adjustments in the trained line did not translate into detectable changes in whole-plant morphology or water use efficiency when deficit irrigation was imposed in glasshouse conditions (Fig. S8). Instead, and not meeting our expectations, chronic exposure to drought across generations led insidiously to the extinction of the trained line. Line propagation started to become challenging, with more and more plants of the trained line showing impaired rooting, bud bursting, or subsequent growth: only 28% of cuttings collected from the trained line of generation #19 generated a viable plant for generation #20, most of them growing very slowly (Fig. S9). We then systematically scored the cutting success rate of subsequent generations, showing a gradual decline in the trained line up to generation #27, after which it was eventually extinct (Fig. 1o). We also established two new lines by switching their irrigation regimes. On the one hand, we generated a 'late-trained line' from cuttings taken on naive plants of generation #19 and then exposed to the same chronic drought as the trained line. This late-trained line could be easily propagated, and no decline was observed after 10 novel generations under chronic drought (Fig. 1o). Reciprocally, we built an 'escaped line' from cuttings of the trained line taken on generation #19 and then exposed to well-watered conditions as in the naive line. Strikingly, this line could not be propagated for more than 4 generations, suggesting the decline of the trained line was not reversible and even worsened upon return to conditions that are normally favorable (Fig. 1o). Transgenerational legacy of chronic exposure to sub-lethal drought therefore translated into a poor capacity for plantlet establishment at the cutting stage, which ultimately led to a gradual loss of the trained line. Since the escaped line died before the trained line, it was challenging to evaluate whether transgenerational drought memory was truly maladaptive. Chronic drought rather led petunia plants to a cul-de-sac, whereby firmly established memory of a predictable stress yielded plasticity with limited benefits under drought, but at other costs (e.g. impaired capacity for vegetative propagation) that were rapidly fatal upon return to well-watered conditions. Retrospectively, we could trace back the initiation of the decline of the trained line around generation #17, when we first noticed a small yet significant reduction in biomass, even before any stress application (Fig. S5a). This decline was unlikely driven by drought-induced starvation because carbohydrates and LMA maintained high levels in the trained line (Figs 1l, S7c), consistent with growth being more sensitive to water deficit than photosynthesis (McDowell, 2011; Muller et al., 2011). This is also in line with studies on adventitious rooting in petunia cuttings that emphasized the prominent role of sink establishment over source limitation (Ahkami et al., 2009, 2013). It is also unlikely that osmotic adjustment would make sugars unavailable for metabolism and thereby limit the establishment of plantlets since the latter were regenerated in well-watered conditions – releasing the constraint on osmotic potential build-up. We also excluded that the decline was due to increased susceptibility to pests and pathogens that could arise in plants under drought stress (Szczepaniec & Finke, 2019; Dixit et al., 2024), since there was no evidence of enhanced biotic attacks in the trained line. Alternatively, irremediable shifts at some key nodes of the hormonal network may underlie plant extinction across generations of drought. However, we do not speculate that it could have arisen from spontaneous mutation because a deleterious allele would not have been fixed through our propagation method that involved independent plants. The irreversible decline of the trained line may instead be subtended by drought-induced epigenetic drift, namely heritable changes that occur independently of sequence modifications. In populations of wild strawberry from different geographic regions, a large part of the variation in DNA methylation associated with climate of origin is inherited upon clonal propagation in a common garden (Sammarco et al., 2024). In this species, epimutations induced by heat stress can persist in at least three generations of successive stolons (López et al., 2024). Similarly, in the Lombardy poplar, natural variation in DNA methylation is largely transmitted through cutting propagation (Díez Rodríguez et al., 2022a). This natural variation overlaps with methylation patterns induced by different abiotic and biotic stresses within a single generation, suggesting functional and adaptive significance (Peña-Ponton et al., 2024). Nonetheless, the epigenetic legacy of geographical origin has a weak effect on phenotypic variation compared to that of current environmental conditions (Díez Rodríguez et al., 2022b; Vanden Broeck et al., 2024). Moreover, in Arabidopsis, repeated drought exposure across five sexual generations did not trigger consistent accumulation of epialleles and did not alter plant responses to drought (Ganguly et al., 2017). Thus, it remains to be proven whether drought-induced, stable epimutations actually accumulated at target loci across clonal generations and had any causal role in the gradual loss of petunia fitness. Overall, spanning 27 generations over 5 yr, our experiment demonstrates that chronic, sub-lethal drought may cause detrimental effects on plant fitness through gradual accumulation of deleterious legacy, without major benefit for the clonal progeny. Critically, this result is not in line with the positive effects that would be expected from an environmentally induced 'soft inheritance' (Richards, 2006; Youngson & Whitelaw, 2008). Nonetheless, it does not exclude that valuable training outcomes may be achieved using other genotypes or environmental scenarios, as transgenerational effects may be negative, null, or positive depending on these factors (Herman & Sultan, 2016; Groot et al., 2017; Rendina González et al., 2018; Deng et al., 2021). Perhaps the overpredictability of our controlled drought scenario prevented creating epigenetic noise that could have improved long-term fitness at the population level, following a 'bet hedging' strategy (Herman et al., 2014). Thus, the remaining gaps of our study – identification of molecular mechanisms and assessment of broader genotype × environment combinations – represent important opportunities for further investigation of how physiological and/or epigenetic memory may be leveraged to safeguard plants against recurrent drought. One single mother-plant was raised from a seed of Petunia hybrida cv 'Carnival Red Star' (kindly provided by Prof. Masayoshi Nakayama, Institute of Floricultural Science, NARO, Tsukuba, Japan). From this mother-plant (generation #00), single-node cuttings bearing one leaf were produced. Cutting bases were dipped into a β-indole butyric acid solution (Rhizopon AA 20%, Hazerswoude-Rijndijk, Netherlands), prior to being inserted into well-watered microplugs (4-cm diameter, Fertil, Boulogne-Billancourt, France), transferred into a mini-glasshouse maintaining high humidity and placed for 4 wk in a growth chamber at 22°C and 16-h photoperiod with white light (280 μmol m−2 s−1) provided by LEDs (Ledpower, Saint-Calais, France). Then, well-rooted cuttings were potted into 0.3-l pots (SOPARCO, Condé-sur-Huisne, France) containing growing medium composed of Irish peat, perlite, and coconut fibers (50 : 40 : 10; v/v/v) amended with Pg Mix (1 kg m−3) and fertilized with nutrient solution (15 N : 10 P2O5 : 30 K2O, Mg, Fe and trace elements, Yara, Nanterre, France, stabilized at pH 6.2). Plantlets were then acclimated for 1 wk in a glasshouse at 18–22°C and brought back in the growth chamber for a second acclimation week. Plants were then split into two batches forming the naive and trained clonal lines (generation #01). Part of these plants were used for assays, while another part was kept permanently either under well-watered conditions for the naive line or under chronic drought for the trained line (see 'Irrigation regimes' in the Materials and Methods section) to raise the next clonal generation (#02) from cuttings as described above. To secure plantlet establishment, cuttings of both lines were kept in well-watered conditions during the propagation process (c. 1 month). While this necessary step could have mitigated potential training effects, it also allowed us to minimize the morphological differences between lines at the onset of chronic drought at each generation. The same scheme was followed to produce generation #03 from generation #02, and so forth. About 20–30 independent plants of each line were used as mother-plants for cuttings at each generation. Similarly, we generated an 'escaped' and a 'late-trained' line from generation #19. The escaped line was built from an initial batch of nine plants obtained from successful cuttings of the trained line at generation #19, and then exposed to well-watered conditions as in the naive line. Conversely, the late-trained line was built from an initial batch of 37 plants obtained from successful cuttings of the naive line at generation #19, and then exposed to the same chronic drought as the trained line. In the growth chamber, potted plants were grouped in trays and sub-irrigated with a mineral solution (Peters Excel CalMag Finisher 13 N : 5 P2O5 : 20 K2O, 7 CaO, 2 MgO plus Kanieltra 15 Fe DTPA, Yara) delivered in the tray until the growing medium saturated and then drained. Water potential of the growing medium was continuously monitored between 0 and −500 hPa using tensiometers (ES-A500, Bambach GbR, Geisenheim, Germany). Under well-watered conditions, irrigation was launched automatically as soon as soil water potential dropped below −100 hPa (corresponding to one irrigation per day during the experiment periods). Chronic drought cycles consisted of one irrigation every 4 d, so that soil water potential oscillated from 0 hPa (just after irrigation) to less than −500 hPa (gradual soil water depletion due to transpiration), depending on the plant stage (Fig. S1a). Under this drought scenario, adult petunia plants showed incipient leaf wilting after 4 d without irrigation (Fig. S1b), which rapidly recovered upon rehydration and could be observed visually (Fig. S1c). Morphological and physiological analyses were carried out through ad hoc destructive harvests. Depending on the generation, different traits were measured at various growth stages and times after irrigation (as specified in the figure legends). Unless otherwise specified, plants exposed to well-watered conditions were sampled shortly after the last irrigation and drainage, while plants subjected to drought conditions were sampled 3 d after the last irrigation. For whole-plant projected area and shape parameters, a side-view photograph of each plant was taken (Fig. S2d). For leaf and axis count, all leaves and axes longer than 1 cm were counted on each plant. For total leaf area, all leaves of each plant were detached, flattened, and scanned. Mean leaf area was obtained by dividing total leaf area by leaf count. For mean flower area, fully opened flowers were detached, placed vertically on a custom-made support, and photographed from above (Fig. S2c). For root coverage at the bottom of the pot, the bottom surface of the root ball was photographed once the pot was removed (Fig. S2e). All images were processed using ImageJ for segmentation and parameter quantification. Total aerial fresh biomass was determined from the fresh weight of all leaves, flowers, and stems of a given plant (all aerial organs together or separately depending on the generation). Roots were washed to eliminate the growing medium, wiped with paper towel, and weighed for fresh root biomass determination. The weighed organs were then oven dried (60°C) for 72 h, and their dry biomass determined. For LMA, dry mass was determined on leaf discs (13-mm diameter). For cutting success rate, plantlets were observed 3 wk after cutting, before any drought cycle. Rooting was scored based on the number of visible roots emerging from the microplug, and bud bursting was scored based on the development of axillary shoots. For propagation, only the plantlets with at least 10 visible roots and 1 axillary shoot were eventually potted. Unless otherwise specified, fully expanded leaves of the median part of shoots were used for physiological analyses. For leaf water content (LWC) and relative water content (RWC), a piece of lamina (without rib) was collected using a scalpel and immediately weighed for fresh weight (FW) determination. Leaf pieces were then floated onto distilled water in a closed Petri dish at 4°C for 24 h prior to being wiped and weighed for the determination of turgid weight (TW). Then, leaf pieces were oven-dried (60°C) for at least 48 h and weighed for dry weight (DW) determination, allowing LWC to be computed as (FW − DW)/FW, and RWC as (FW − DW)/(TW − DW). We noticed, however, that water soaking could occur on some leaves, especially those subjected to low water potential, and this was exacerbated if the tissues were immersed – making TW overestimated and RWC underestimated in these samples (Fig. S10). We therefore present only LWC in this manuscript, but source data for RWC remain available (Data availability). For gas exchange analyses, a Li-Cor 6400XT equipped with a 6 cm2 clamp-on leaf cuvette topped by a 6400-02B LED light source (Li-Cor Inc., Lincoln, NE, USA) was used on intact leaves in the growth chamber. Reference CO2 was tuned at 400 ppm. For snapshot measurements (generation #07), block temperature was set at 22°C and PPFD at 250 μmol m−2 s−1, while RH was c. 60%. For time-course measurements of gas exchange in response to a step increase in irradiance and leaf excision (generations #20, #21 and #24), an Li610 dew-point generator (Li-Cor Inc.) with reference dew-point temperature set to 14°C was connected to the system inlet while block temperature was set at 25°C, so that air VPD was c. 1.5 kPa. PPFD was initially set to 100 μmol m−2 s−1 until obtaining a steady gs (15–90 min depending on the leaf) and then switched to 1000 μmol m−2 s−1. One hour after this switch, the petiole was excised and the dynamics of stomatal movements were monitored for at least 4 h, showing a large and rapid closure after the initial wrong-way response (Powles et al., 2006). Snapshot measurements of abaxial gs were further carried out using an SC1 leaf porometer (Decagon Devices, Pullman, WA, USA). Quantum yield of photosystem II was measured in dark- and light-adapted leaves (generations #15 and #17), using a FluorPen FP 110/P Chl fluorometer (Photon Systems Instruments, Drásov, Czech Republic). Dark adaptation was achieved in the growth chamber by applying a leaf clip (30-mm diameter) for 15 min before attaching the fluorometer (without exposing the measured leaf portion to light). Shoot water potential was measured using a Scholander pressure bomb (PMS Instruments, Albany, OR, USA). The distal part of the plant shoot bearing a mature terminal flower was inserted downwards with the cut end visible out of the pressure chamber, so that the balancing pressure could be determined through gradual pressurization. At generation #18, a batch of plants from the naive and the trained lines was grown in controlled deficit irrigation during 2 months in a glasshouse. Each pot containing 75 g of growing medium and one plantlet was weighed at potting, and the initial soil water content was recorded. The water retention curve of the growing medium was determined (Fig. S11) using a tension table draining at pressure potentials from −5 (pF0.7) to −1000 hPa (pF3) (Caron et al., 2007). Based on this retention curve, we computed the target weight to reach a soil water potential of −400 hPa upon re-irrigation. Using a weighing scale, plants were re-irrigated three times a week so that their soil water potential was not above −400 hPa. The amount of water used for irrigation was recorded for each pot. At 4, 6, and 8 wk after potting, five plants were destructively harvested for fresh and dry weight determination. Cumulative water use efficiency at a given time was computed as the dry mass at this given time divided by the total amount of supplied water from the beginning of the experiment. At generation #18, leaves of the naive and trained lines were harvested in liquid nitrogen throughout a 6-d dry-down experiment. Leaves were ground and lyophilized, and then DW was determined. One mililiter 80% methanol containing 40 μl of 4 g l−1 melizitose as an internal sugar standard was added to 15–20 mg powder and the mix was vortexed. After heating at 76°C for 15 min, the liquid was evaporated under vacuum in a Speedvac AES1010 (Savant Instruments, Hyderabad, India). The residue was dissolved in 1 ml milliQ water, vortexed, and spun for 5 min at 20 000 g at 4°C. Supernatant was used for soluble sugars analysis while the pellet was used for starch content. After appropriate dilution of the supernatant, soluble sugars were analyzed by high-performance liquid chromatography (HPLC) on a Carbopac PA-1 column (Dionex Corp., Sunnyvale, CA, USA) as previously described (Vandecasteele et al., 2011). Starch content was determined from the pellet using a Thermo Scientific Gallery photometric analyzer (Thermo Fisher Scientific, Waltham, MA, USA) after enzymatic digestion of the pellet (Crespel et al., 2022) and expressed as mg equivalent glucose g-1 dry weight. Wheat starch (Sigma S5127) was used as an internal standard. At generation #21, one leaf was collected on six different plants per line and irrigation regime. All pots were rehydrated 2 h before harvest. Lamina sections of c. 1 cm2 were cut and placed in 2 ml of freshly prepared glutaraldehyde 3.5% (v/v) in phosphate buffer 0.2 M (pH 7.2) in a glass vial on ice, and then vacuum-infiltrated for several hours until deposition of tissues at the bottom of the vial. After replacement with fresh glutaraldehyde solution, tissues were left overnight at 4°C for fixation. The next day, tissues were rinsed twice with phosphate buffer 0.2 M (pH 7.2) at 4°C, then rinsed three times in distilled water and dehydrated in a series of ethanol solutions (50°, 70°, 80°, 95°, and 3 times at 100°) before pre-infiltration in ethanol 100°/Technovit 7100 resin 50 : 50 (v/v) under vacuum for 2 h on ice followed by 12 h storage at 4°C. The pre-infiltration solution was next removed and replaced by Technovit 7100 resin solution 100% (v/v) followed by vacuum infiltration for 30 min on ice and 12 h storage at 4°C. Leaf tissues were then placed in moulds with 700 μl resin on a heating plate at 35°C for 2 h and transferred to 37°C in a heating cabinet for at least 12 h until total polymerization and inclusion. 3-μm thick transverse sections were produced using an automated microtome and stained in toluidine blue (0.5%) for 10 s. Leaf sections were observed under a macroscope (Zeiss Axio Zoom.V16). Mosaic images were obtained in the bright field and in the GFP domain (long path filter) to capture residual autofluorescence from chloroplasts, enhancing our capacity to distinguish cells from intercellular spaces. Both channels were merged for subsequent image analysis. The thickness of each tissue (upper and lower epidermis, palisade and spongy mesophyll) was measured manually in ImageJ (each biological replicate being the mean of 5–10 measurements at different regions of the same leaf sample). Leaf porosity was determined as the fraction of airspace in the mesophyll using a semi-automatic procedure. The Zeiss Zen 3.5 Image Analysis module was used to segment mesophyll cells and intercellular spaces, before each detected object was manually assigned in either class. Data management, statistical analyses, and figures were handled in R 4.2.0 (R Development Core Team, 2022) using the packages rstatix and ggpubr (Kassambara, 2023a,b), ARTool (Kay et al., 2025) and emmeans (Lenth et al., 2025). For each experiment, the number of replicates can be visualized as the number of points in each corresponding jitter plot and is specified in the figure or its legend otherwise. Morphological and physiological traits were analysed using 1-, 2-, or 3-way ANOVA or non-parametric ANOVA through Aligned Rank Transform (ART-ANOVA; Wobbrock et al., 2011) when heteroscedasticity or non-normality of residuals was detected. Pairwise comparisons were performed using estimated marginal means (emmeans; also developed for ART, Elkin et al., 2021), with the significance threshold adjusted by the Bonferroni method unless otherwise stated. ANOVA effect size of individual factors (generation, line, irrigation and/or cycle) and their interactions were computed using the partial eta squared (η2p). When traits were analysed during time since last irrigation, ANCOVA was performed if the response proved linear, and effect sizes and significance levels were obtained similarly. When trait response to desiccation was not linear, a loess curve was fitted and emmeans comparisons were performed at each time point. Note that the actual number of replicates for each analysis can be visualized as the number of points in each corresponding jitter- or scatter-plot. We thank Camille Lebras and Bénédicte Dubuc for technical assistance during morpho-physiological measurements, Pascale Satour for the analysis of sugar contents, and Yvette Barraud-Roussel for the determination of the soil-water retention curve. We thank all interns (Myriam Tisserand, Félix Rétif, Etienne Brémand, Juliette Vergnaud, Youssra Daccouch) who contributed to this project at different stages. We gratefully acknowledge the PHENOTIC platform (doi: 10.17180/YKBZ-2V85), especially Christine Boursier for plant care in glasshouse conditions and the IMAC facility (Aurélia Rolland and Fabienne Simonneau) for assistance during microscopy analyses. This project benefited from financial support of Région Pays de la Loire (RFI Objectif Végétal, project Epiclones 2017-2019) and of Régions Bretagne and Pays de la Loire and ADEME (project ClimatVeg 2021-2025 driven by Vegepolys Valley). FP acknowledges support from INRAE – AgroEcoSystem Division and Institut Agro for the secondment period at IRHS. None declared. NL conceived and supervised the project. LH-T, PM, NB and NL performed morpho-physiological measurements. LH-T, PM and NL planned the experiments and line propagation, and analyzed data. FP conducted data analysis, contributed to experiments (gas exchange, microscopy, RWC) and prepared the figures and the manuscript. LH-T and NL revised the manuscript. FP and LH-T contributed equally to this work. Phenotypic data and scripts for their analyses are publicly available at doi: 10.57745/CNJOZ8. Fig. S1 Typical time-course of irrigation regimes. Fig. S2 Chronic drought consistently affects morphological traits of petunia across multiple clonal generations. Fig. S3 Irrigation regime of the past petunia clonal generations has weak legacy on petunia phenotypes at generations #08–10. Fig. S4 Irrigation regime of the past petunia clonal generations has only subtle effects on petunia responses to chronic drought installation at generation #15. Fig. S5 Irrigation regime of the past petunia clonal generations has only subtle effects on petunia responses to chronic drought installation at generation #17. Fig. S6 Irrigation regime of the past petunia clonal generations has no significant effects on petunia gas exchange at generations #20–24. Fig. S7 Chronic drought applied on numerous petunia generations tends to enhance mesophyll porosity and leaf mass per area. Fig. S8 Chronic drought applied on numerous petunia generations does not enhance water use efficiency or morphological traits under deficit irrigation in glasshouse conditions. Fig. S9 Challenging vegetative propagation in the petunia trained line at generation #20. Fig. S10 Petunia leaf water soaking occasionally occurs upon water saturation and biases the estimation of relative water content. Fig. S11 Soil retention curve. Please note: Wiley is not responsible for the content or functionality of any Supporting Information supplied by the authors. Any queries (other than missing material) should be directed to the New Phytologist Central Office. Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article. The New Phytologist Foundation remains neutral with regard to jurisdictional claims in maps and in any institutional affiliations.