# Source paper archive - Source title: Leaf ionomic stability shapes drought adaptation in pearl millet - Author(s): Nakombo Gbassault, P., Arenas, S., Affortit, P., Faye, A., Flis, P., Sine, B., Moukouanga, D., Gantet, P., KOSH-KOMBA, E., Kane, N., Bennett, M. J., Wells, D. M., Cubry, P. M., Bailey, E., Segura, V., Vigouroux, Y., Laplaze, L., Grondin, A. - Source landing URL: https://www.biorxiv.org/content/10.64898/2026.11.12.766145v1 - PDF URL: https://www.biorxiv.org/content/10.64898/2026.11.12.766145v1.full.pdf - PDF SHA-256: 4c1f681c7fe4a5da4890ff7ddb484d3f3753fc02dbd84f48c24472a31ca1e170 - Converter: pymupdf4llm 0.0.27 - Limits: Automated PDF text extraction; reading order, equations, tables and formatting may be incomplete or incorrect. Images are omitted, and no OCR was performed. Markup and links are neutralized for safe downloading. Consult the original PDF for the authoritative version. --- [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.12.766145; this version posted November 13, 2026. The copyright holder for this](https://doi.org/10.64898/2026.11.12.766145) preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in [perpetuity. It is made available under a CC-BY-NC-ND 4.0 International license.](http://creativecommons.org/licenses/by-nc-nd/4.0/) 1 **Leaf ionomic stability shapes drought adaptation in pearl millet** 2 3 Princia Nakombo-Gbassault [1,2], Sebastian Arenas [1,$], Pablo Affortit [1], Awa Faye [3], Paulina Flis [4], 4 Bassirou Sine [3], Daniel Moukouanga [1], Pascal Gantet [1], Ephrem Kosh-Komba [2], Ndjido Kane [3], 5 Malcolm Bennett [4], Darren Wells [4], Philippe Cubry [1], Elizabeth Bailey [4], Vincent Segura [5], Yves 6 Vigouroux [1], Laurent Laplaze [1], Alexandre Grondin [1] 7 8 1 DIADE, Université de Montpellier, IRD, CIRAD, Montpellier France 9 2 JEAI AgrobiodiveRCA, Université de Bangui, Bangui, Central African Republic 10 3 CERAAS, Institut Sénégalais des Recherches Agricoles (ISRA), Thiès, Senegal 11 4 School of Biosciences, University of Nottingham, Sutton Bonington, United Kingdom 12 5 UMR AGAP Institut, Université de Montpellier, CIRAD, INRAE, Institut Agro Montpellier, F 13 34398 Montpellier, France 14 $ Present address: Department of Plant Protection Biology, Swedish University of 15 Agricultural Sciences, Alnarp, Sweden. 16 1 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.12.766145; this version posted November 13, 2026. The copyright holder for this](https://doi.org/10.64898/2026.11.12.766145) preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in [perpetuity. It is made available under a CC-BY-NC-ND 4.0 International license.](http://creativecommons.org/licenses/by-nc-nd/4.0/) 17 **Abstract** 18 Pearl millet is a staple cereal of semi-arid West Africa, where early-season water deficit 19 constrains vegetative establishment and yield. While the genetic architecture of the pearl millet 20 foliar ionome has been characterised under well-watered conditions, how this ionome 21 reorganises under drought, and whether this reorganisation has a distinct genetic basis, remains 22 unknown. We profiled the drought-response of 17 foliar elements in 175 diverse pearl millet 23 inbred lines over two seasons in Senegal. The foliar concentration of two co-regulated groups 24 of elements changed reproducibly under early water stress across both years. Leaf ion stability, 25 rather than plasticity, was associated with the maintenance of flowering time and plant biomass, 26 suggesting that drought adaptation in pearl millet relies more on ionomic stability than on 27 adaptive shifts in elemental composition. Among the 17 elements, only potassium showed a 28 significant genotype-by-treatment interaction suggesting that genetic variation contributes to 29 differences in potassium response to drought. Genome-wide association mapping on potassium 30 plasticity in response to drought identified a locus, designated qKP-6, that limits potassium 31 depletion under stress without penalising constitutive potassium levels. This locus represents a 32 candidate region for the genetic control of potassium plasticity under drought. This work 33 provides new insights into the genetic control of mineral nutrition under water deficit in pearl 34 millet, and highlights leaf potassium concentration as a genetically controlled component of 35 drought-responsive ionomic variation with potential implications for drought tolerance. 36 37 **Key-words:** Pearl millet, drought, ionome, plasticity, GWAS 38 2 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.12.766145; this version posted November 13, 2026. The copyright holder for this](https://doi.org/10.64898/2026.11.12.766145) preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in [perpetuity. It is made available under a CC-BY-NC-ND 4.0 International license.](http://creativecommons.org/licenses/by-nc-nd/4.0/) 39 **Introduction** 40 Water scarcity, exacerbated by climate change, poses a major threat to the sustainability of 41 agricultural production in arid and semi-arid regions where rainfall is highly variable and 42 drought events are projected to increase in both frequency and severity (Intergovernmental 43 Panel On Climate Change (IPCC), 2023; Rezaei _et al._, 2023). In these predominantly rainfed 44 systems, food security critically depends on drought-tolerant crops capable of sustaining at 45 least minimal but reliable yields under water-limited conditions (Satyavathi _et al._, 2021). 46 Among drought-tolerant crops, pearl millet ( _Pennisetum glaucum (L.) R. Br._ ), a C4 47 cereal domesticated in the Sahel, stands out for its exceptional resilience to both water and heat 48 deficits (Debieu _et al._, 2017). It is widely cultivated in regions receiving as little as 300 to 500 49 mm of annual rainfall, where only few other cereal species can survive (Bidinger _et al._, 1987; 50 Pucher _et al._, 2015). This resilience relies on a diversity of adaptive strategies, including 51 tolerance, avoidance and escape mechanisms, operating at different developmental stages from 52 vegetative growth to flowering and grain filling (Shrestha _et al._, 2023). Recent crop-modelling 53 evidence indicates that early drought stress, occurring before or during panicle initiation, is the 54 most damaging water-stress pattern for pearl millet in the Sahel, causing grain yield losses of 55 up to 44% and biomass reductions of approximately 40% (De la Fuente _et al._, 2024). Climate 56 based simulations further show that early drought stress occurs more frequently than terminal 57 stress for pearl millet in this region, a trend expected to be reinforced in the context of changing 58 climate, underscoring its critical impact on crop establishment in rainfed systems (De la Fuente 59 _et al._, 2024). Because this stress coincides with vegetative development, a period of rapid 60 mineral uptake and partitioning (Marschner and Marschner, 2012), it may have particularly 61 pronounced effects on the leaf ionome, defined as the composition of plant tissue in mineral 62 nutrients and trace elements (Salt _et al._, 2008; Srivastava _et al._, 2022), with potential 63 implications for photosynthesis, osmotic adjustment, and stress signalling (Farooq _et al._, 2009). 64 Water availability is a major determinant of plant growth and mineral nutrition. Beyond 65 its direct effects on cell expansion and photosynthesis, reduced water availability profoundly 66 reshapes plant mineral composition and regulation (D’Oria _et al._, 2022). Nutrient responses to 67 drought are no longer viewed as a passive consequence of reduced transpiration or growth 68 alone, but as actively regulated processes involving specific membrane transporters, signalling 69 pathways, and metabolic adjustments (Bisht _et al._, 2026). Among essential macronutrients, 70 potassium (K), calcium (Ca), and magnesium (Mg) play central but contrasting roles under 71 water limitation: K contributes to osmotic adjustment and stomatal regulation, Ca functions as 72 a key secondary messenger in stress signalling, and Mg deficiency reduces whole-plant water 3 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.12.766145; this version posted November 13, 2026. The copyright holder for this](https://doi.org/10.64898/2026.11.12.766145) preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in [perpetuity. It is made available under a CC-BY-NC-ND 4.0 International license.](http://creativecommons.org/licenses/by-nc-nd/4.0/) 73 use efficiency (Tränkner _et al._, 2016; Johnson _et al._, 2022; Liu _et al._, 2022). Under water 74 deficit, studies have revealed reproducible shifts in elemental profiles, including changes in the 75 concentrations of Na, K, Mg, Mn or Cu, whose direction and magnitude depend on stress 76 intensity, duration, and species (D’Oria _et al._, 2022; Quagliata _et al._, 2025). Importantly, these 77 changes can occur prior to the appearance of visible morphological symptoms, indicating that 78 the ionome is an early and sensitive indicator of drought responses (Danku _et al._, 2013; D’Oria 79 _et al._, 2022). Thus, drought-induced changes in elemental composition can offer an integrative 80 view of physiological adjustment to water limitation, while genotypic variation in the 81 magnitude of these changes may reflect ionomic plasticity. The differential regulation of these 82 elements under drought therefore reflects an integrated reprogramming of the plant ionome, 83 which may influence stress tolerance and ultimately yield. 84 Ionomics, the high-throughput profiling of elemental composition across genotypes and 85 environments, provides an integrative framework to investigate the genetic determinants of 86 plant responses to environmental constraints (Salt _et al._, 2008; White and Broadley, 2009; 87 Bouain _et al._, 2019). Yet, the genetic bases of ionomic variation and plasticity under water 88 limited conditions remain incompletely characterized, particularly in crops adapted to marginal 89 environments. Genome-wide association studies of the ionome have been conducted primarily 90 under near-optimal conditions, limiting our understanding of ionomic plasticity under drought 91 and the detection of genotype × environment interactions in response to this stress. In pearl 92 millet specifically, the genetic architecture of foliar mineral variation and responsiveness under 93 water deficit remains unexplored, leaving unresolved to what extent water stress-induced 94 plasticity is under genetic control and how it contributes to plant performance under water 95 limitation. 96 Here, we combined field-based ionomic phenotyping with genome-wide association 97 analyses in a diverse pearl millet panel evaluated under two contrasting water regimes over two 98 years to identify the genomic regions associated with constitutive foliar elemental variation and 99 its plastic response to water deficit. We hypothesised that 1) the leaf ionomic profiles under 100 drought would exhibit substantial genotypic variation in pearl millet and be associated with 101 plant performance, and 2) ion content are under genetic control, with loci underlying 102 constitutive ionomic content being largely distinct from those controlling ionomic plasticity 103 under drought. 4 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.12.766145; this version posted November 13, 2026. The copyright holder for this](https://doi.org/10.64898/2026.11.12.766145) preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in [perpetuity. It is made available under a CC-BY-NC-ND 4.0 International license.](http://creativecommons.org/licenses/by-nc-nd/4.0/) 104 105 **Materials and Methods** 106 **Plant material** 107 A total of 175 pearl millet ( _Pennisetum glaucum_ (L.) R. Br.) inbred lines were selected from 108 the Pearl Millet Genetic Association Panel (PMiGAP; Sehgal _et al._, 2015; Varshney _et al._, 109 2017), which captures the genetic diversity of the species from Africa and India. Detailed 110 composition of the lines used here is provided in Nakombo-Gbassault _et al._ (2025). 111 112 **Field trials and experimental design** 113 Field trials were conducted at the National Center for Agronomic Research (CNRA, ISRA, 114 Bambey, Senegal; 14°42'N, 16°28'W) during the hot dry off-seasons of 2021 and 2022 on a 115 loamy sand soil (84.8% sand, 7.9% silt, 7.2% clay; Diongue _et al._, 2022). An alpha-lattice 116 design was used with two water regimes: well-watered (WW) and vegetative-stage water stress 117 (WS), with four replicates per treatment. In both years, prior to planting, soil samples were 118 collected from thirteen locations distributed across the four replicate blocks, at four depth 119 intervals (0-20, 20-60, 60-100, and 100-140 cm). 120 Since no rainfall occurs during this period of the year, water stress was imposed simply 121 by withholding irrigation from 21 to 49 days after sowing (DAS) in 2021 and from 21 to 42 122 DAS in 2022, corresponding to stress durations of 28 and 21 days respectively. These 123 conditions defined four environments used in subsequent analyses: WW21 (well-watered) and 124 WS21 (water-stress) for the 2021 trial, and WW22 and WS22 for the 2022 trial. Relative soil 125 water content was monitored from 0 to 160 cm using DIVINER probes spread across both 126 treatments. 127 At the end of the stress period, three plants per plot were sampled for leaf ionomic 128 profiling, root anatomical phenotyping, and aboveground biomass measurement, as described 129 in Nakombo-Gbassault _et al._ (2025). For ionomic analysis, the last fully ligulated leaf from the 130 main tiller of each harvested plant was collected, washed, dried, and pooled per plot before 131 digestion and quantification by inductively coupled plasma mass spectrometry (ICP-MS) as 132 described (Nakombo-Gbassault _et al._, 2025). Foliar concentrations of 17 elements were 133 quantified, including macro- and micronutrients: potassium (K), sodium (Na), magnesium 134 (Mg), calcium (Ca), phosphorus (P), sulfur (S), iron (Fe), zinc (Zn), copper (Cu), cadmium 135 (Cd), nickel (Ni), manganese (Mn), molybdenum (Mo), cobalt (Co), lithium (Li), rubidium 136 (Rb), and strontium (Sr). Plant available soil elements were also quantified using 1M NH 4 CO 3 137 DTPA extraction. For root anatomy, crown root segments were sampled and analysed by laser 5 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.12.766145; this version posted November 13, 2026. The copyright holder for this](https://doi.org/10.64898/2026.11.12.766145) preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in [perpetuity. It is made available under a CC-BY-NC-ND 4.0 International license.](http://creativecommons.org/licenses/by-nc-nd/4.0/) 138 ablation tomography to quantify stele area, total metaxylem vessel area and number, 139 sclerenchyma proportion, and overall root cross-sectional area, as described in Affortit _et al._ 140 (2024). 141 After the stress period, plants were re-irrigated until maturity, and phenology as well as 142 yield and yield-component traits were recorded, including plant height, shoot biomass at 143 maturity, spike number, grain weight per plant and harvest index, following procedures 144 described in Affortit _et al._, (2024). Trait list and acronyms are described in Supplementary 145 Table S1. 146 147 **Stress response analyses** 148 For each phenotypic trait including element concentrations, root anatomical and 149 agromorphological traits, best linear unbiased estimates (BLUEs) and broad-sense heritability 150 were estimated for each treatment-year combination using the SpATS spatial model 151 (Rodríguez-Álvarez _et al._, 2018), implemented in the statgenSTA R package (Van Rossum, 152 2020). The model was fitted to data from a resolvable incomplete block design. Genotype was 153 fitted as a fixed effect to obtain BLUEs and as a random effect to estimate broad-sense 154 heritability using the generalised heritability approach described by Oakey _et al._, (2006). 155 Two complementary stress indices were computed per trait and genotype, restricted to 156 genotypes present in both water regimes within the same year: 157 (i) the plasticity index, expressed as a log2 ratio (LR): 158 LR = log2(BLUEs_WS/BLUEs_WW) (1) 159 quantifies relative plasticity, with LR = 0 indicating no change, negative values indicating 160 stress-induced depletion, and positive values indicating accumulation under stress; 161 (ii) the Stress Tolerance Index (STI; Fernandez, 1992): 162 STI = (BLUEs_WW × BLUEs_WS) / mean (BLUEs_WW)² (2) 163 captures absolute performance across both water regimes. Both indices were computed for 164 ionomic and agromorphological traits. 165 To identify elements whose response to early water deficit is repeatable across seasons, 166 the direction and magnitude of each element’s response were characterised per year by the 167 percentage change in mean concentration and the mean log2 ratio (LR) across genotypes. WW 168 and WS BLUEs of the same genotypes were compared with a paired t-test when residuals were 169 normally distributed (Shapiro-Wilk _p_ - 0.05), and with a Wilcoxon signed-rank test otherwise. 170 An element response was considered robust when significant in both years with the same 171 direction of change. 6 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.12.766145; this version posted November 13, 2026. The copyright holder for this](https://doi.org/10.64898/2026.11.12.766145) preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in [perpetuity. It is made available under a CC-BY-NC-ND 4.0 International license.](http://creativecommons.org/licenses/by-nc-nd/4.0/) 172 Because soil sampling preceded treatment imposition, it reflects baseline spatial 173 heterogeneity rather than treatment-induced variation in soil ion availability. Multivariate 174 correspondence between soil (0-20 cm layer, ICP-MS values) and foliar ion concentrations was 175 tested separately for each of the four environments using a Mantel test on the 15 elements 176 common to both datasets (standardised Euclidean distances, 999 permutations; Mantel, 1967). 177 178 **Multivariate analysis of the foliar ionome** 179 Principal component analysis (PCA) was performed on foliar ion concentrations (BLUEs) to 180 characterise the variation of the 17 elements. A global PCA was first conducted using data from 181 all four environments (WW21, WS21, WW22, and WS22) to characterise the overall structure 182 of ionomic variation. To isolate the treatment-driven component of variation from inter-annual 183 differences, separate PCAs were then performed for each year using the corresponding WW 184 and WS samples. For these year-specific analyses only, the sign of the principal components 185 was reversed when necessary, so that water-stressed genotypes consistently projected towards 186 negative scores on the primary stress-response axis, facilitating direct comparison of this axis 187 between years. For each configuration, the contribution of each element to each component 188 (percentage of the component's total inertia) and the element loading (coordinate on the 189 component axis) were reported to identify the principal drivers of the stress axis. For the stress 190 response axis, scores under each treatment (PC_WW, PC_WS) and their difference (ΔPC = 191 PC_WS - PC_WW) were computed per genotype and year. 192 193 **Correlation analyses** 194 Pairwise Pearson correlations among the 17 foliar elements were computed within each 195 environment on BLUEs and their significance assessed with a 5% false discovery rate (FDR) 196 correction (Benjamini and Hochberg, 1995). Pairs were classified as constitutive (significant 197 with consistent sign in all four environments), never significant, or context dependent. 198 Conservation of the correlation structure between water regimes was tested per year using 199 Mantel tests (Mantel, 1967) and its significance tested using 999 permutations. Hierarchical 200 clustering (average linkage) was applied to the mean correlation matrix across the four 201 environments to identify co-regulation modules of ions. Ions within each module were then 202 combined into a composite score, which was used in subsequent analyses in place of the 203 individual ions. 204 Correlations between the ionome and agromorphological traits were assessed at three 205 levels: individual elements (mono-ion), modules, and the combined-PCA stress-response axis 7 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.12.766145; this version posted November 13, 2026. The copyright holder for this](https://doi.org/10.64898/2026.11.12.766145) preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in [perpetuity. It is made available under a CC-BY-NC-ND 4.0 International license.](http://creativecommons.org/licenses/by-nc-nd/4.0/) 206 (multi-ion). Two complementary approaches were applied at each level: (i) within-environment 207 Pearson correlations, retaining associations significant in both years with a consistent sign; and 208 (ii) linear regression pooling both years, with year as a fixed covariate and cluster-robust 209 standard errors at the genotype level (vcovCL, Zeileis _et al._, 2020). FDR correction at 5% was 210 applied within each trait category, with STI and LR indices corrected independently. For all 211 pooled models, a predictor x Year interaction term additionally tested whether association 212 strength differed between years. 213 214 **Predicting genotype-level drought tolerance from the constitutive ionome** 215 To predict genotype-level drought tolerance, the STI of shoot biomass and of grain weight were 216 used. Three predictors were compared. The first approach used a simple linear regression to 217 assess the well-watered score of the PC2 axis as a predictor of the STI. The second approach 218 used a linear regression to identify the ion most strongly correlated with the STI as a predictor. 219 The third approach used a penalized regression (LASSO; Tibshirani, 1996; glmnet package, 220 Friedman et al., 2010) on all measured elements to identify a multi-ion signature predictor, 221 with λ chosen at the one-standard-error rule and with an ion selection frequency over 20 refits. 222 In the three approaches, the predictive accuracy, namely the correlation between observed and 223 out-of-fold predicted values, was estimated by repeated genotype-grouped 5-fold cross 224 validation (20 repetitions) with identical partitions for all predictors, and predictors were 225 compared using bootstrap intervals of the differences (10,000 resamples of genotypes). All 226 predictors were evaluated on the same held-out genotypes, so that accuracy reflects prediction 227 of new genotypes rather than model fit. 228 The predicted score (centred on the year mean) for each STI, calculated as the mean 229 out-of-fold for the best predictor (highest mean predictive accuracy) of the three approaches, 230 was used to classify genotypes as tolerant, intermediate or sensitive using tercile thresholds, as 231 was the observed STI. To verify that predicted scores separated genotypes that were tolerant 232 from those that were sensitive for the observed STI, the scores of these two groups were 233 compared with a Wilcoxon test. Fisher’s exact and Spearman tests were used to test whether 234 genotypic classification between STI for shoot biomass and STI for grain weight was 235 conserved, on predicted and observed values. To test if population structure was a potential 236 confounder, a PCA using the genotypic data from the panel was performed and the first two 237 components were selected. The first two components were correlated with the predicted scores 238 and included as covariates in the regression of the observed STI on the predicted score. 239 8 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.12.766145; this version posted November 13, 2026. The copyright holder for this](https://doi.org/10.64898/2026.11.12.766145) preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in [perpetuity. It is made available under a CC-BY-NC-ND 4.0 International license.](http://creativecommons.org/licenses/by-nc-nd/4.0/) 240 **Estimation of phenotypic variance components** 241 To partition the variation of each principal component and each element into constitutive and 242 condition-specific genetic components, a linear mixed model was fitted to genotype BLUEs by 243 restricted maximum likelihood (REML, lme4; Bates _et al._, 2015): 244 P gty = µ + T t + Yy + (TxY) ty + Gg + (GxT) gt + (GxY) gy + ε gty (3) 245 where T and Y are fixed effects of treatment and year, (T×Y) their fixed interaction, G the 246 random effect of genotype, and G×T and G×Y the random genotype-by-treatment and 247 genotype-by-year interactions. Year was treated as fixed because it has only two levels. 248 Random effects were tested by likelihood-ratio tests comparing the full model to the model 249 without the effect (with statistical threshold _p_ < 0.05) and, for principal components, fixed 250 effects by Type II F-tests (Satterthwaite approximation, lmerTest, Kuznetsova _et al._, 2017). 251 For elements, each random component is expressed as a percentage of the total random 252 variance. For principal components, the variance of each fixed term (sum-to-zero coding) was 253 added to the random components so that the treatment effect could be compared with the 254 genetic components (Nakagawa and Schielzeth, 2013). The principal component with the 255 highest treatment percentage was retained as the stress-response axis. In the year-specific 256 PCAs, the treatment share was estimated as η² = SS_T/SS_total from the linear model: 257 PC i = μ + G g + T t + ε i (4) 258 with Genotype (G) and Treatment (T) as fixed effect. 259 260 **Genome-wide association analysis (GWAS)** 261 A total of 269,848 SNPs were obtained from whole-genome sequencing of 147 inbred lines 262 from the PMiGAP panel, after filtering for minor allele frequency (MAF > 0.05) and missing 263 data rate (< 20% per SNP and per genotype). Procedures for filtering, imputation, and 264 population structure inference are described in Nakombo-Gbassault _et al._, (2025). 265 GWAS was performed on ion BLUEs for each of the four environments separately 266 (WW21, WW22, WS21, WS22), using four complementary models correcting for population 267 structure and/or genetic relatedness through different approaches: LFMM2 (Frichot _et al._, 268 2014), BLINK (Huang _et al._, 2019), EMMA (Kang _et al._, 2008) and CMLM (Zhang _et al._, 269 2010). For ions showing a significant genotype × treatment interaction, GWAS was performed 270 on the corresponding plasticity index (LR) to identify loci associated with variation in the 271 response to water deficit. 272 273 **Multi-environment meta-analysis of GWAS** 9 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.12.766145; this version posted November 13, 2026. The copyright holder for this](https://doi.org/10.64898/2026.11.12.766145) preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in [perpetuity. It is made available under a CC-BY-NC-ND 4.0 International license.](http://creativecommons.org/licenses/by-nc-nd/4.0/) 274 For each of the four GWAS models, results across environments were integrated independently 275 using the fixed-effect ( _Fe_ ) procedure of the metaGE framework in R (De Walsche _et al._, 2025). 276 Because the same genotypes were evaluated across environments, the correlation between 277 environment-specific GWAS was first estimated to account for this non-independence in the 278 model. 279 In metaGE, the estimated effect (β) and _p_ -value of each marker from each environment 280 specific GWAS is converted into a signed Z-score: 281 Z mk = sign(β mk ) × Φ⁻¹(1 - p mk /2) (5) 282 where Z mk is the Z-score of marker m in environment k, β mk and p mk its estimated effect and _p_ 283 value, and Φ⁻¹ the quantile function of the standard normal distribution. 284 Two complementary tests were applied: (i) a model without treatment contrast, 285 identifying markers with effects consistent across all environments, and (ii) a model with an 286 explicit contrast opposing irrigation and water deficit, testing for marker effects within each 287 water regime and differences between regimes. The no-contrast model was applied (i) across 288 all four environments, (ii) within each water regime separately (WW21 and WW22 combined 289 or WS21 and WS22 combined), and (iii) on the plasticity index (LR, both years combined). 290 The contrast model was applied across all four environments. Significance was set at _p_ < 10 [⁻5] . 291 To delimit a region of interest around each significant SNP, pairwise linkage disequilibrium 292 (r²) was calculated using PLINK v1.9 (Chang _et al._, 2015) within a ± 50 kb window centred on 293 the significant marker. SNPs with r² > 0.5 were considered genetically linked and grouped with 294 the significant SNP to define the boundaries of the associated region. Candidate genes within 295 this region were identified using the functional annotation of the pearl millet reference genome 296 (ASM217483v2; (Varshney _et al._, 2017). 297 298 **Allelic class effect analysis** 299 For each significant marker, trait associations were tested across the three allelic classes 300 (homozygous reference, heterozygous, homozygous alternative) determined from VCF 301 genotype calls, using linear mixed models fitted by maximum likelihood (REML = FALSE), 302 which allows likelihood ratio tests between models differing in fixed effects: 303 trait = μ + SNP_class + Year + Genotype + ε (6) 304 where SNP_class and Year are fixed effects, Genotype is a random effect of line identity 305 accounting for repeated measurements across years, and ε is the residual. Significance was 306 assessed by likelihood ratio test against a null model without the SNP_class term, and effect 307 size was estimated by the increase in marginal R² relative to this null model (ΔR²m; Nakagawa 10 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.12.766145; this version posted November 13, 2026. The copyright holder for this](https://doi.org/10.64898/2026.11.12.766145) preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in [perpetuity. It is made available under a CC-BY-NC-ND 4.0 International license.](http://creativecommons.org/licenses/by-nc-nd/4.0/) 308 and Schielzeth, 2013). Estimated marginal means of allelic classes were compared using 309 Tukey-adjusted pairwise contrasts. Dependence of the allelic effect on water regime was tested 310 using a SNP class × treatment interaction: 311 Trait = μ + SNP_class + Treatment + (SNP_class×Treatment) + Year + (Year×Treatment) + 312 Genotype + ε (7) 313 To assess whether the SNP effect was independent of the phenological variation, the 314 association between each trait and flowering time under well-watered conditions (DTF_WW) 315 was first tested by replacing SNP_class with DTF_WW in model (6); DTF_WW was then 316 included as a covariate in both the full and null models: 317 trait = μ + SNP_class + DTF_WW + Year + Genotype + ε (8) 318 and the change in the difference between homozygous classes was expressed as a percentage 319 reduction. 320 321 **Result** 322 **Genotypic and interannual variation in the foliar ionomic response to early water deficit** 323 To characterise the foliar ionomic response to early water stress, we quantified 17 mineral 324 elements in leaves of 175 pearl millet inbred lines evaluated under well-watered (WW) and 325 water-stressed (WS) conditions over two seasons (2021 and 2022) at Bambey (Senegal). Under 326 water stress, the 17 measured elements varied widely among genotypes, with coefficients of 327 variation ranging from 12.4% (S) to 44.8% (Na) in 2021 and from 12.4% (S) to 51.9% (Li) in 328 2022 (Table 1). Broad-sense heritability also showed substantial variation, from low for lithium 329 (H² = 0.01 in 2022) and nickel (H² = 0.04 in 2022) to high for cadmium (H² = 0.77-0.84) and 330 cobalt (H² = 0.80-0.82) indicating strong differences in the genetic control of ion accumulation 331 in pearl millet (Table 1). Soil ion concentrations were sampled from thirteen locations spread 332 across the four replicate blocks, at four depths per location. These soil values did not match 333 mean foliar ion concentrations in any of the four environments (Mantel tests, _r_ = -0.37 to 0.09, 334 all _p_ - 0.45), suggesting that differences between genotypes were independent of differences 335 in soil ion contents (Supplementary Table S2). 336 11 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.12.766145; this version posted November 13, 2026. The copyright holder for this](https://doi.org/10.64898/2026.11.12.766145) preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in [perpetuity. It is made available under a CC-BY-NC-ND 4.0 International license.](http://creativecommons.org/licenses/by-nc-nd/4.0/) 337 **Table 1: Descriptive statistics and broad-sense heritability of foliar ion concentrations** 338 **under well-watered and water-stressed conditions.** WW WS Element Year Mean concentration ( mg kg [-1] ) SD CV Min Max H2 n Mean concentration ( mg kg [-1 ] ) SD n Ca Cd Co Cu Fe K Li Mg Mn Mo Na Ni 2021 119 10920.98 4314.2 96 9516.1 2495.26 26.2 5123.92 17380.9 0.81 2022 121 7187.06 1573.15 121 7688.06 1514.75 19.7 4088.81 11779.92 0.74 2021 113 0.41 0.16 95 0.42 0.18 41.8 0.09 1.13 0.77 2022 120 0.45 0.14 115 0.58 0.21 35.9 0.25 1.41 0.84 2021 117 0.08 0.02 93 0.1 0.03 31.3 0.04 0.2 0.82 2022 121 0.06 0.02 117 0.07 0.03 41 0.02 0.19 0.80 2021 118 8.81 1.81 98 10.66 02.09 19.6 6.61 15.99 0.82 2022 124 6.29 1.17 124 6.83 1.17 17.2 3.99 10.59 0.76 2021 118 98.61 25.55 97 103.44 34.55 33.4 47.05 218.99 0.80 2022 121 62.86 14.57 118 59.12 9.7 16.4 41.58 100.07 0.64 2021 120 30486.69 7383.72 98 15481.56 4579.16 29.6 5554.01 29228.41 0.29 2022 125 41030.1 9373.68 125 36132.62 7086.5 19.6 20888.37 58864.14 0.33 2021 119 0.06 0.02 96 0.04 0.01 32.3 0.01 0.08 0.44 2022 121 0.03 0.01 117 0.01 0.01 51.9 0 0.03 0.01 2021 120 7425.09 1730.51 98 10883.43 3394.66 31.2 4329.01 20689.75 0.73 2022 124 7670.91 1431.56 123 10000.37 2164.93 21.6 4939.59 15572.19 0.69 2021 119 55.41 19.83 97 79.26 22.8 28.8 30.3 160.07 0.66 2022 121 39.32 9.3 120 48.76 10.35 21.2 26.37 79.55 0.54 2021 116 1.32 0.68 92 1.33 0.54 40.3 0.46 2.96 0.80 2022 118 0.96 0.32 120 0.78 0.25 32.5 0.32 1.71 0.72 2021 114 1234.74 497.24 95 216.52 97 44.8 85.07 518.1 0.78 2022 119 1423.95 393.22 114 498.35 228.49 45.9 127.28 1338.37 0.83 2021 101 1.74 0.41 82 1.77 0.52 29.1 0.74 3.25 0.42 2022 104 0.78 0.24 104 0.88 0.23 26.4 0.45 1.68 0.04 12 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.12.766145; this version posted November 13, 2026. The copyright holder for this](https://doi.org/10.64898/2026.11.12.766145) preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in [perpetuity. It is made available under a CC-BY-NC-ND 4.0 International license.](http://creativecommons.org/licenses/by-nc-nd/4.0/) 2021 119 4412.61 781.59 98 4052.16 782.95 19.3 2349.25 6065.8 0.71 P 2022 125 2117.45 389.3 121 1918.73 375.47 19.6 921.17 3551.05 0.69 2021 120 38.79 7.76 97 29.18 8.62 29.5 12.72 51.34 0.64 Rb 2022 124 46.41 8.44 124 46.07 8.24 17.9 25.75 75.12 0.53 2021 119 2848.12 384.98 98 2916.58 361.73 12.4 1991.22 3729.47 0.74 S 2022 124 2043.58 287.28 121 2032.71 252.53 12.4 1424.01 2814.37 0.65 2021 118 298.22 111.4 97 243.13 73.78 30.3 86.41 466.86 0.69 Sr 2022 124 255.52 58.14 123 276.39 57.42 20.8 134.3 482.69 0.61 2021 108 65.77 17 98 45.64 9.82 21.5 20.17 70.92 0.64 Zn 2022 121 37.55 8.42 119 32.26 7.91 24.5 15.03 53.81 0.63 339 For each of the 17 foliar elements and each year (2021, 2022), the number of genotypes with 340 available BLUEs (n), mean, and standard deviation are reported for well-watered (WW) and 341 water-stressed (WS) conditions; coefficient of variation (CV), minimum and maximum values 342 are additionally reported for the water-stressed condition. Broad-sense heritability (H²) was 343 estimated jointly with BLUEs under water stress. 344 345 The ionomic profile was more strongly disturbed in response to vegetative drought 346 stress in 2021 than in 2022, with a mean plasticity index (LR) across the 17 elements of -0.23 347 in 2021 versus -0.12 in 2022. Comparing the two water regimes revealed three clear and 348 repeatable patterns of ionomic response to vegetative drought (Fig. 1). A first group of four 349 elements increased significantly under water deficit in both years. This group included 350 magnesium (Mg), cobalt (Co), manganese (Mn), and copper (Cu). A second group of five 351 elements decreased significantly under water deficit in both years. This group comprised 352 sodium (Na), potassium (K), zinc (Zn), lithium (Li), and phosphorus (P). A third group, made 353 up of calcium and strontium, reversed direction between years: both elements decreased under 354 water deficit in 2021 but increased in 2022. The treatment-by-year interaction was significant 355 for 14 of the 17 elements (Supplementary Table S3). Lithium, sodium and nickel were the only 356 elements with no significant interaction. Together with the moderate-to-high heritability of 357 several elements, these results indicate that the early ionomic response to water deficit in pearl 358 millet is reproducible and structured. 359 360 **Structure of the foliar ionome network across water regimes** 361 To investigate the functional interdependencies underlying foliar ionomic variation, pairwise 362 correlation matrices among the 17 elements were computed separately for each water regime 13 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.12.766145; this version posted November 13, 2026. The copyright holder for this](https://doi.org/10.64898/2026.11.12.766145) preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in [perpetuity. It is made available under a CC-BY-NC-ND 4.0 International license.](http://creativecommons.org/licenses/by-nc-nd/4.0/) 363 and year (Supplementary Fig. S1). Overall, correlation patterns were largely conserved 364 between well-watered and water-deficit conditions, as indicated by significant Mantel 365 correlations between the two matrices in both years ( _r_ = 0.88 in 2021; _r_ = 0.81 in 2022; _p_ = 366 0.001). However, the proportion of significant pairwise correlations decreased under water 367 deficit, from 66.9% to 59.6% in 2021 and from 44.1% to 25.7% in 2022, indicating an overall 368 weakening of ion-ion associations under stress. Hierarchical clustering on the correlation 369 matrix averaged across all four environments identified two anti-correlated modules: Module 370 1, comprising thirteen elements (Ca, Cd, Co, Cu, Fe, Li, Mn, Mo, Ni, P, S, Sr, and Zn), and 371 Module 2, comprising four elements (K, Mg, Na, and Rb). This two-module partition was 372 robust to FDR correction at 5% of the correlation matrix. Within Module 1, correlations were 373 observed within the four environments for Ca-Sr (r = 0.86-0.98), Cu-Fe ( _r_ = 0.66-0.84), Cu-S 374 ( _r_ = 0.55-0.73), and Cu-P ( _r_ = 0.56-0.72). Within Module 2, K-Rb was the only pair significant 375 in all four environments ( _r_ = 0.50-0.66), indicating that Module 2 cohesion is primarily carried 376 by these two elements. Between the two modules, negative correlations were observed as 377 significant in all four environments for Fe-K ( _r_ = -0.43 to -0.26) and Cu-K ( _r_ = -0.37 to -0.28), 378 while it was significant in three of the four environments for Rb-Sr, Mo-Na, Cu-Rb and Mg-P. 379 These results show that while the foliar ion correlation network remains structured under early 380 drought, its connectivity and strength are partially reconfigured, reflecting a degree of plasticity 381 in the ionomic covariance structure. 382 383 **Relationships between foliar ionome and agro-morphological traits across water regimes** 384 We next examined relationships between individual foliar ion concentrations and agro 385 morphological traits separately under well-watered and water-deficit conditions, retaining only 386 correlations that were significant in both 2021 and 2022 and with a consistent sign 387 (Supplementary Fig. S2, S3). These correlations were structured according to the two ionomic 388 modules previously identified. Under well-watered conditions, Module 1 elements, particularly 389 Fe, Cu and Mo, were predominantly negatively correlated with phenological traits and plant 390 height, whereas Module 2 elements, particularly K and Na, showed positive correlations with 391 time to flowering and maturity. Under water deficit, the overall contrast between the elements 392 from the two modules was maintained. Cu and Mo were negatively associated with flowering 393 and maturity time, while Fe, Cu, and Mo were negatively associated with plant height. In 394 contrast, K and Mg showed positive associations with flowering and maturity time, and Rb was 395 positively associated with plant height. Relationships with biomass and yield were element 396 and water-regime-specific, including a positive correlation between Mn and grain weight under 14 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.12.766145; this version posted November 13, 2026. The copyright holder for this](https://doi.org/10.64898/2026.11.12.766145) preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in [perpetuity. It is made available under a CC-BY-NC-ND 4.0 International license.](http://creativecommons.org/licenses/by-nc-nd/4.0/) 397 well-watered conditions, and negative correlations of Cu and P with shoot dry weight at 398 maturity under water deficit. 399 We then asked whether, in each of the four environments, correlations observed for 400 individual elements aligned with those observed at the module-level. Composite module scores 401 largely reproduced the contrast already described and extended it to additional traits 402 (Supplementary Table S4). Under well-watered conditions, Module 2 was positively associated 403 with days to flowering, days to maturity, plant height and negatively correlated with harvest 404 index, whereas Module 1 associated negatively with flowering, maturity and plant height, and 405 positively with spike number. Under water deficit, the same contrast persisted for flowering, 406 maturity and plant height, and Module 2 was additionally associated with shoot dry weight at 407 maturity (Supplementary Table S4). The same analysis performed on the combined four 408 environments largely confirmed the main associations identified in the year-specific analyses 409 and revealed additional relationships with grain weight and root anatomical traits under water 410 deficit (Supplementary Tables S5 and S6). The low frequency of significant year interactions 411 indicated that most ion-trait associations were consistent across years. 412 We next examined whether ion-trait relationships were associated with ionic stability, 413 measured as the stress tolerance index (STI), or with the magnitude of the ionic response to 414 water deficit, measured as the log-ratio index (LR; Supplementary Fig. S4, S5, S6 and S7). The 415 STI-STI correlations recapitulated the major patterns observed for raw ion concentrations, with 416 STI of Fe, Cu and Mo negatively correlated with STI of flowering and maturity time, whereas 417 STI of K showed positive correlations with the STI of the same traits. This phenological pattern 418 was also observed at the module level, with Module 1 negatively and Module 2 positively 419 associated with STI of flowering and maturity time respectively (Supplementary Table S7). LR 420 of Ca and Sr was positively correlated with STI of flowering time and maturity time. No LR 421 based association was detected at the module level (Supplementary Table S7). Thus, the 422 reproducible ion-trait network was dominated by correlations between ionic stability and 423 phenological stability, while ionic plasticity contributed a smaller, element-specific, set of 424 correlations involving Ca and Sr. A complementary analysis pooling both years, at both the 425 individual-ion and module levels, recovered these correlations and additionally identified 426 relationships involving trait plasticity, not detected under the stricter two-year criterion 427 (Supplementary Table S8). Year interactions were rare among these correlations. 428 Overall, foliar ion concentrations were consistently associated with flowering and 429 maturity time and plant height, with opposite patterns between the two ionomic modules across 430 years and water regimes. The stability of ion concentrations across water regimes, more than 15 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.12.766145; this version posted November 13, 2026. The copyright holder for this](https://doi.org/10.64898/2026.11.12.766145) preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in [perpetuity. It is made available under a CC-BY-NC-ND 4.0 International license.](http://creativecommons.org/licenses/by-nc-nd/4.0/) 431 their plasticity, was associated with the stability of phenological traits, plant height and, to a 432 lesser extent, biomass and yield components. 433 434 **Multivariate analysis of foliar ionomic variation** 435 Principal component analysis (PCA) of foliar ion concentrations across the four environments 436 resolved two major axes of ionomic variation, together explaining 60.8% of the total variance 437 (PC1, 46.2%; PC2, 14.6%; Fig. 2a). A joint model partitioning each axis by genotype, year, 438 treatment and their interaction showed that PC1 was predominantly associated with year, which 439 accounted for 70.5% of the variance in PC1 scores ( _p_ < 0.001; Fig. 2b). In contrast, PC2 was 440 predominantly associated with water treatment, which accounted for 65.9 % of the variance in 441 PC2 scores ( _p_ < 0.001). The main effect of year was not significant for PC2 ( _p_ = 0.3), 442 suggesting that this axis is predominantly associated with water treatment across years. PC2 443 was therefore retained as the stress-response axis. The loading structure of PC2 further revealed 444 a coordinated contrast among multiple elements rather than the contribution of a single ion (Na 445 showed the largest positive loading while Mg showed the largest negative loading; Fig. 2c). 446 PC2 also retained a significant genotype component, with genotype accounting for 447 9.3% of its variance ( _p_ < 0.001), indicating that genotypes differed significantly in their 448 position along the multivariate ionomic axis associated with water treatment. A significant year 449 × treatment interaction was also detected for PC2 (5.4%, _p_ < 0.001), indicating that the 450 magnitude of the ionomic response to water deficit varied between years. No principal 451 component showed a significant G×T variance ( _p_ ≥ 0.30), whereas G×Y was significant for 452 PC1, PC3 and PC5 (p < 0.01). When PCA variance was partitioned separately by year PC2 453 remained the component most strongly associated with water treatment in both years, with 454 treatment accounting for 77% of the variance in 2021 and 67% in 2022 (Supplementary Fig. 455 S8). 456 We next tested whether the ionomic signature associated to response to water stress 457 (PC2) was correlated to the changes in agromorphological traits (LR and STI). The well 458 watered score (PC2_WW) was negatively correlated with STI for grain weight, plant height 459 and spike weight, while the water-stressed score (PC2_WS) was negatively correlated with STI 460 for spike weight (Supplementary Fig. S9). A model with both years, with year as a fixed 461 covariate and standard errors adjusted for repeated measurement of genotypes, confirmed these 462 five associations and identified new ones (Supplementary Table S9). PC2_WW was 463 additionally negatively correlated with STI for shoot dry weight both at 42/49 DAS and at 464 maturity, and days to flowering and maturity, while PC2_WS was negatively associated with 16 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.12.766145; this version posted November 13, 2026. The copyright holder for this](https://doi.org/10.64898/2026.11.12.766145) preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in [perpetuity. It is made available under a CC-BY-NC-ND 4.0 International license.](http://creativecommons.org/licenses/by-nc-nd/4.0/) 465 STI for grain weight and tiller number. Associations between PC2 or the stress-induced 466 displacement of the axis (ΔPC2), and trait stability or plasticity were rare. 467 Altogether, water availability shaped foliar ionomic variation along a single 468 coordinated gradient (PC2). The position of genotypes along this gradient under well-watered 469 conditions, more than the shift induced by water deficit, was associated with STI for shoot 470 biomass, grain weight, spike weight and plant height, suggesting that the leaf mineral profile 471 of unstressed plants could help identify tolerant genotypes. 472 473 **Ionomic prediction of genotype-level drought tolerance** 474 Having established that the PC2 axis under well-watered conditions was associated with STI 475 for biomass and grain weight, we tested whether the ionome measured under well-watered 476 conditions could predict drought tolerance in pearl millet. Three predictors were compared: (i) 477 the well-watered score of PC2 (PC2_WW), (ii) the single ion most strongly correlated with the 478 trait, and (iii) a multi-ion signature obtained by penalized regression. 479 For STI grain weight, PC2_WW showed the highest accuracy, ahead of the multi-ion 480 signature and the single ion selected within training sets (mean cross-validated r = 0.300, 0.197 481 and 0.195, respectively), suggesting a better prediction of grain weight maintenance 482 (Supplementary Table S10). For STI shoot dry weight at maturity, the multi-ion signature, 483 which retained Mg, Fe and Mo in all 20 refits, significantly outperformed the single ion and 484 PC2_WW (r = 0.422, 0356 and 0.317, respectively), indicating improved prediction from a 485 specific multi-ion signature (Supplementary Table S10). 486 The tercile analysis identified 42 observed tolerant and 42 observed sensitive genotypes 487 whose predicted scores differed significantly for both traits (Wilcoxon, p = 0.015 and p = 488 1.7×10⁻¹¹, respectively; Supplementary Table S10). Classification in each group for shoot 489 biomass and grain weight was non-random (Fisher’s exact test, OR = 8.2, p = 3×10⁻⁷), genotype 490 ranking was conserved (Spearman ρ = 0.54, _p_ < 2.2×10⁻¹⁶), and similar for observed STI (OR 491 = 4.2, p = 5×10⁻⁴; ρ = 0.38, _p_ = 1×10⁻⁵), with 27 genotypes jointly classified as tolerant for 492 shoot biomass and grain weight (Supplementary Table S11). Predicted scores were not 493 substantially confounded by population structure, which was limited in this panel (genomic 494 PC1 explained 4.6% of the variance). 495 Overall, foliar ionomic profiles measured on well-watered plants predicted the 496 maintenance of grain weight and biomass under water deficit. The most informative predictor 497 differed between traits, the global PC2 axis for grain weight and a multi-ion model consistently 17 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.12.766145; this version posted November 13, 2026. The copyright holder for this](https://doi.org/10.64898/2026.11.12.766145) preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in [perpetuity. It is made available under a CC-BY-NC-ND 4.0 International license.](http://creativecommons.org/licenses/by-nc-nd/4.0/) 498 retaining Mg, Fe and Mo across refits for biomass. Based on predicted scores, 27 genotypes 499 were classified as tolerant for both traits, identifying drought-tolerant germplasm and 500 illustrating the use of pre-stress ionomic profiling for early, non-destructive screening. 501 502 **Genetic variance in foliar ion accumulation and its interaction with water treatment** 503 To quantify the respective contributions of genetic and environmental factors to foliar ion 504 accumulation, linear mixed-effects models were fitted for each element, with water regime, 505 year, and their interaction as fixed effects, and genotype, genotype by treatment (G×T), and 506 genotype by year (G×Y) as random effects. The random genotype effect was significant for 507 multiple ions, reflecting substantial genetic variation in mean foliar concentrations 508 independently of water regime. The proportion of total phenotypic variance attributable to 509 genetic effects ranged widely across elements (from 21% for Zn to 40.6% for Cu; Fig. 3; 510 Supplementary Table S12). Among all elements analysed, only potassium showed a significant 511 G×T interaction ( _p_ = 5.48×10⁻³), accounting for 15.9% of its total variance (Supplementary 512 Table S12), indicating that pearl millet genotypes differ specifically in their foliar potassium 513 content in response to early water deficit. Drought-induced changes in the other elements 514 analysed were largely independent of the genotypes. 515 516 **Constitutive genetic architecture of the pearl millet foliar ionome** 517 To identify markers with effects consistent across all four environments, genome-wide 518 association analysis was performed using the metaGE meta-analysis framework without 519 contrast for treatments. Significant loci were identified for 12 of the 17 quantified elements, 520 yielding 48 significant SNPs that resolved into 42 independent QTLs after grouping markers 521 within linkage disequilibrium blocks (Table 2; Supplementary Table S13). Four instances of 522 co-localisation between markers associated with different elements were observed. The SNP 523 CM007983_27393436 on chromosome 2 was associated with both K and Cu accumulation, the 524 SNP CM007986_6220019 on chromosome 5 was associated with both Fe and Ca 525 accumulation, and the SNP LKME02014639.1_317356 on an unanchored contig and 526 CM007983_157814421 on chromosome 2 were associated with both K and Rb accumulation 527 (Table 2; Supplementary Table S13). Among all elements, K showed the broadest 528 chromosomal distribution of associated loci. Genes present within the linkage disequilibrium 529 (LD) block surrounding each QTL and the Z-scores of all significant SNPs in each environment 530 are summarised in Supplementary Table S13. These 42 QTLs thus represent loci with 531 consistent effects on the foliar ionome across all four year × water regime combinations (well 18 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.12.766145; this version posted November 13, 2026. The copyright holder for this](https://doi.org/10.64898/2026.11.12.766145) preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in [perpetuity. It is made available under a CC-BY-NC-ND 4.0 International license.](http://creativecommons.org/licenses/by-nc-nd/4.0/) 532 watered and drought-stressed). 533 534 **Table 2. Constitutive QTLs for leaf element concentrations detected by multi-** 535 **environment GWAS meta-analysis across four environments in the PMiGAP pearl millet** 536 **panel.** **Element** **Markers** **QTLs** **Chromosomes** **Co-localised QTLs** Ca 4 4 1, 3, 5, contig 1 (Fe) Cd 2 1 1 – Cu 5 3 2, 3, 5 1 (K) Fe 5 4 2, 5, 7 1 (Ca) K 9 8 1, 2, 6, 7, contig 3 (Cu, Rb) Li 5 4 2, 4, contig – Mg 2 2 3, contig – Mn 2 2 2, contig – Mo 1 1 6 – Na 2 2 3, 4 – Rb 10 10 1, 2, 6, contig 2 (K) Zn 1 1 2 – 537 Markers: total number of significant SNPs ( _p_ < 10⁻⁵) from the metaGE fixed-effect meta538 analysis without contrast, performed for four GWAS methods (LFMM2, EMMA, BLINK and 539 CMLM). QTLs: number of QTL associated; markers of the same element located within the 540 same LD region (r² > 0.5) were grouped into one QTL. Chromosomes: chromosomal positions 541 of the identified QTLs; contig refers to unanchored genomic contigs that could not be assigned 542 to a specific chromosome. Co-localised QTLs: number of QTLs sharing a marker with a QTL 543 for another element (element in brackets). Details for each marker are given in Supplementary 544 Table S13. 545 546 **Identification of QTLs controlling leaf potassium plasticity** 547 We next looked for QTL controlling ionomics plasticity in response to vegetative water stress. 548 Among the 17 elements, K was the only one with a significant genotype × treatment variance 549 component ( _p_ = 5.48×10⁻³). To dissect the genetic architecture of potassium plasticity in 550 response to water deficit, three complementary analyses were applied: (i) condition-specific 551 GWAS within each water regime, (ii) a genotype by treatment contrast model in MetaGE 19 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.12.766145; this version posted November 13, 2026. The copyright holder for this](https://doi.org/10.64898/2026.11.12.766145) preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in [perpetuity. It is made available under a CC-BY-NC-ND 4.0 International license.](http://creativecommons.org/licenses/by-nc-nd/4.0/) 552 within all four environments, and (iii) GWAS on the potassium plasticity index LR_K. These 553 analyses identified a set of loci whose effects are either condition specific or relate to the 554 potassium response to water stress. 555 Under well-watered conditions, five significant SNPs were detected on chromosomes 556 2, 3, 6 and 7, resolving into five independent QTLs (Supplementary Table S14). Among these, 557 CM007983_27393436 on chromosome 2 (qKC-2a) and CM007987_104928234 on 558 chromosome 6 (qKC-6b) were also identified in the four-environment analysis without contrast 559 for treatments. Under water stress, two significant SNPs were detected: CM007987_17176446 560 on chromosome 6 (qKC-6a) and LKME02014639.1_317356 on an unanchored contig (qKC 561 uc1). Both SNPs were also identified in the four-environment analysis without contrast, with 562 Z-scores of greater magnitudes under water stress than under well-watered conditions. Three 563 SNPs detected under well-watered conditions were absent from both the four-environment and 564 water stress analyses and defined well-watered-specific QTLs: CM007984_207071465 on 565 chromosome 3 (qKE-3), CM007988_37716247 on chromosome 7 (qKE-7a) and 566 CM007988_136644636 on chromosome 7 (qKE-7b). 567 In the genotype by treatment contrast model applied across four environments, a single 568 SNP reached significance: CM007983_27393436 on chromosome 2 (qKC-2a), already 569 identified in the four-environment without contrast analysis, and well-watered analyses 570 (Supplementary Table S14). This SNP had Z-scores of the same sign in all four environments, 571 but larger under well-watered than under water stress conditions (Z = 5.34 and 4.19 under well 572 watered conditions; Z = 3.29 and 2.76 under water stress, for 2021 and 2022 respectively). The 573 C allele was associated with higher K concentration under both water regimes, with a larger 574 difference under well-watered conditions (C/C vs T/T: 39,689 vs 32,294 mg kg⁻¹) than under 575 water stress (28,114 vs 24,322 mg kg⁻¹). The SNP class × treatment interaction on K 576 concentration was significant ( _p_ = 0.002), i.e. the effect of qKC-2a differed in size between 577 water regimes but not in direction. 578 Direct GWAS on potassium plasticity index LR_K identified two significant SNPs on 579 chromosome 6: CM007987_110250858 and CM007987_110250945, located 87 bp apart 580 within the same linkage disequilibrium block (Supplementary Table S14). Together they define 581 a single QTL designated qKP-6. Neither SNP was detected in previous GWAS on absolute 582 potassium concentrations. Pairwise linkage disequilibrium between CM007987_110250858 583 and all other significant potassium SNPs was r² < 0.09 regardless of chromosomal position, 584 indicating that qKP-6 is distinct from all previously identified potassium loci. G/G genotypes 585 at the CM007987_110250858 locus accumulated more potassium than A/A genotypes under 20 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.12.766145; this version posted November 13, 2026. The copyright holder for this](https://doi.org/10.64898/2026.11.12.766145) preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in [perpetuity. It is made available under a CC-BY-NC-ND 4.0 International license.](http://creativecommons.org/licenses/by-nc-nd/4.0/) 586 well-watered conditions, whereas potassium concentrations did not differ significantly between 587 genotypes under water stress. A/A genotypes showed a mean potassium loss of 27.4% under 588 water stress relative to well-watered conditions, compared to 40.1% for G/G genotypes. A 589 significant SNP class × treatment interaction ( _p_ = 2.4×10⁻⁴) indicated that allelic effects at this 590 locus depend on the water regime. Because LR_K was associated with flowering time ( _p_ = 591 1.6×10⁻⁴) and A/A genotypes flowered 4.4 days earlier than G/G genotypes ( _p_ = 0.003), models 592 were additionally adjusted for flowering time. The qKP-6 effect on LR_K remained significant 593 ( _p_ < 0.001; ΔR²m = 0.037), with a 14.7% reduction between homozygous classes, whereas the 594 effect on potassium concentration under well-watered conditions was no longer significant ( _p_ 595 = 0.10), suggesting that the association with K plasticity was not fully accounted for by 596 flowering time. The effect on LR_K was significant in each year (2021: _p_ = 0.017; 2022: _p_ = 597 6.5×10⁻⁵). Altogether, our analyses identified a new QTL, qKP-6, associated with the plasticity 598 of leaf K concentration under vegetative drought stress and distinct from the loci for K 599 concentration. 600 601 **Identification of a candidate gene in qKP-6** 602 At qKP-6, linkage desequilibrium analysis around SNP CM007987_110250858 identified a 603 14.3-kb LD block spanning 110,236,620 to 110,250,945 bp and containing 604 CM007987_110250945, a second SNP significantly associated with the potassium plasticity 605 index (Figure 4; Supplementary Table S14). The block also encompassed several additional 606 variants, including four SNPs located within _Pgl_GLEAN_10012096_ (110,235,838 607 110,237,176 bp) which partially overlaps the LD block. These four SNPs showed strong LD 608 with the index SNP ( _r²_ = 0.77-0.86), supporting _Pgl_GLEAN_10012096_ as the strongest 609 positional candidate at this locus. BLASTp analysis of the predicted 348-amino-acid protein 610 revealed sequence similarity to uncharacterised proteins from related grass species, including 611 _Urochloa humidicola_ (59% sequence identity), _Setaria italica_ (52%) and _Paspalum vaginatum_ 612 (46%). InterProScan identified no recognisable functional domains in the predicted protein. 613 Thus, _Pgl_GLEAN_10012096_ is prioritised as a positional candidate for potassium plasticity 614 at qKP-6, although its molecular function remains unknown. 615 616 **Discussion** 617 In this study, we investigated the drought response of the leaf ionome in pearl millet, aiming 618 to identify ions associated with early drought tolerance and to evaluate whether plasticity in 619 ion content contributes to this tolerance. In addition, we searched for genetic determinants 21 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.12.766145; this version posted November 13, 2026. The copyright holder for this](https://doi.org/10.64898/2026.11.12.766145) preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in [perpetuity. It is made available under a CC-BY-NC-ND 4.0 International license.](http://creativecommons.org/licenses/by-nc-nd/4.0/) 620 associated with ion content stability across water regimes and with drought-induced plasticity, 621 with the aim of supporting breeding towards improving pearl millet resilience to drought in 622 terms of nutrient use efficiency and forage quality. Our findings suggest that the stability of 623 several leaf ions is associated with drought tolerance in pearl millet. We further identified the 624 genetic architecture underlying both ion stability across water regimes and drought-induced 625 ion plasticity highlighting potassium as a particularly variable component of the drought 626 response. 627 Leaf ionomic composition has been shown to be remodelled by drought stress in several 628 crops (Fischer _et al._, 2019; D’Oria _et al._, 2022). In young wheat leaf blades, moderate to severe 629 drought stress induced a significant reduction in the concentration of most ions (D’Oria _et al._, 630 2022), whereas in maize cobs, mild drought increased ion concentrations, while severe drought 631 caused a strong reduction (Fischer _et al._, 2019). Substantial variation among genotypes has also 632 been observed in ionomic responses to drought (D’Oria _et al._, 2021, 2022; Quagliata _et al._, 633 2025). In our study, the use of a large panel of pearl millet genotypes confirmed that the leaf 634 ion content under drought varied largely among genotypes and that this variation was partly 635 under genetic control. The strongest source of environmental variation in ionomic composition 636 was associated with the year, as observed in the principal component analysis. The effect of 637 the drought treatment was secondary and was mainly captured by the second principal 638 component axis. Despite the strong year effect, correlations among ions observed under 639 irrigated conditions were generally conserved under drought stress, although they tended to 640 weaken. Indeed, two covariation modules were consistently identified across conditions: one 641 comprising Ca, Cd, Co, Cu, Fe, Li, Mn, Mo, Ni, P, S, Sr and Zn, and another comprising Na, 642 K, Rb and Mg. The concentrations of some of these ions were consistently affected by drought 643 across both years: Na, K, Zn and Li concentrations decreased, whereas those of Mg, Co, Mn 644 and Cu increased. Together, these results suggest that water deficit does not disrupt the overall 645 structure of the leaf ionome but rather alters elemental concentrations while largely preserving 646 their covariation structure. This interpretation is consistent with the concept of ionomic 647 homeostasis, whereby coordinated relationships among elements can be maintained despite 648 changes in their absolute concentrations under water deficit (Salt _et al._, 2008). 649 The concentrations of several ions have been associated with improved drought 650 tolerance in crops. In alfalfa, changes in the concentrations of ions such as Na, K and Ca were 651 associated with the maintenance of cellular homeostasis and water balance under drought 652 (Chen _et al._, 2026). Furthermore, several studies have illustrated the beneficial role of K in 653 mitigating the effects of drought stress in wheat (Alzahrani _et al._, 2026), barley (Feng _et al._, 22 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.12.766145; this version posted November 13, 2026. The copyright holder for this](https://doi.org/10.64898/2026.11.12.766145) preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in [perpetuity. It is made available under a CC-BY-NC-ND 4.0 International license.](http://creativecommons.org/licenses/by-nc-nd/4.0/) 654 2020) or poplar (Zhang _et al._, 2023). In line with these findings, our results showed that Na 655 and K concentrations were positively correlated with time to flowering, while parameters 656 describing the ion module containing these two elements were also positively correlated with 657 shoot dry weight at maturity. However, Na and K concentrations were significantly reduced 658 under drought. Such reductions may result from decreased K uptake and translocation to the 659 shoots due to drought-induced limitations in transpiration and may illustrate a potential trade 660 off between water conservation and nutrient uptake required for osmoregulation (Nieves 661 Cordones _et al._, 2014, 2019). Conversely, Cu and Fe concentrations were negatively associated 662 with plant height and shoot dry weight at maturity under drought. Similarly, although Mg 663 concentration was not directly correlated with agro-morphological traits, the computation of an 664 ionomic signature based on a principal component analysis revealed that the water stress score 665 (PC2_WS), to which Mg was a major contributor, was negatively associated with the stress 666 tolerance index (STI) for grain weight, tiller number and shoot dry weight at maturity. Leaf Mg 667 and Cu concentrations increased under drought and, although Fe concentration did not increase 668 significantly in these conditions, it was closely associated with Cu concentration. This pattern 669 is consistent with the possibility that a variation in Cu and Fe levels contributes to oxidative 670 stress under drought conditions (Ravet and Pilon, 2013). Therefore, both ions deficiency (likely 671 Na and K) and excessive accumulation of specific ions (Mg, Cu and Fe) may have contributed 672 to the reduction in shoot biomass production and yield under drought. However, these 673 associations should be interpreted as relationships rather than evidence of direct causal effects. 674 Beyond individual ion associations, the constitutive foliar ionome under well-watered 675 conditions also distinguished tolerant and sensitive genotypes. Grain yield maintenance under 676 drought tended to be best captured by the well-watered score (PC2_WW). Biomass tolerance, 677 in contrast, was better predicted by a sparse Mg-Fe-Mo signature than by PC2_WW, indicating 678 that a combination of elements carried more information than any single elemental 679 concentration. The significant, though partial, overlap between genotypes tolerant for both 680 traits is consistent with a shared tolerance component, and the resulting dual-tolerant genotypes 681 provide promising candidates for subsequent field validation. 682 Phenotypic plasticity can play important roles in the ability of plants to cope with 683 drought stress (Schneider, 2022). However, adaptive responses to drought require appropriate 684 coordination, as maladaptive plasticity may negatively affect plant growth and fitness (Brooker 685 _et al._, 2022). In our study, no correlations were observed between the plasticity of individual 686 ions, or of the leaf ionome, and the plasticity of agro-morphological traits. In contrast, the 687 stability of leaf ion concentration across water regimes was associated with the stability of 23 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.12.766145; this version posted November 13, 2026. The copyright holder for this](https://doi.org/10.64898/2026.11.12.766145) preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in [perpetuity. It is made available under a CC-BY-NC-ND 4.0 International license.](http://creativecommons.org/licenses/by-nc-nd/4.0/) 688 agronomic traits. These results suggest that, in pearl millet, adaptation to drought may rely 689 more on the maintenance of ion homeostasis than on the adaptive changes in leaf ion content. 690 Correlation analyses further suggest that this strategy may be beneficial under drought for some 691 ions, such as Na and K to avoid a reduction in leaf concentration, but also Mg and Cu to avoid 692 an increase in leaf concentration. Interestingly, among all the ions studied, only leaf K 693 concentration showed a significant genotype by treatment interaction, which explained 15.9 % 694 of its variance. This suggests that the ionomic reorganisation induced by drought may represent 695 a conserved physiological response in pearl millet. In contrast, the genotype-specific response 696 of K indicates that this otherwise conserved response contains a heritable component that may 697 be exploitable for breeding. Genotype-dependent K response may reflect its multifunctional 698 role in plant water relations and growth, including osmotic adjustment, stomatal regulation or 699 transport (Mishra _et al._, 2026). Genetic variation in these processes may therefore contribute 700 to the differences observed among genotypes in K homeostasis under drought stress. 701 K was identified as being associated with improved drought tolerance, although its 702 concentration tended to decrease under drought. Therefore, QTL contributing to the 703 maintenance of K concentration across water regimes may be particularly relevant for breeding. 704 GWAS identified eight constitutive loci (qKC) for K concentration that were not affected by 705 water treatment. Interestingly, two of these loci (qKC-6a and qKC-uc1) were also identified 706 specifically for K concentration under drought stress, suggesting that they may be of particular 707 interest for increasing K concentration constitutively across water regimes. Because the 708 drought response of leaf K concentration was genetically controlled, we used plasticity indices 709 to quantify this response and applied genetic models based on meta-analysis of contrasts 710 between water treatments to identify its genetic basis. Among the different plasticity indices, 711 we selected the ratio, which has been shown to be suitable for uncovering the genetic 712 architecture of plasticity (Arenas _et al._, 2025). One locus on chromosome 6 (qKP-6) was 713 identified as associated with this index. The alternate allele limits K depletion under drought 714 without affecting constitutive K levels. Although no agro-morphological trait showed a 715 significant association with this SNP that could reach genome-wide significance in GWAS 716 analyses, the favourable allele was significantly associated with greater stability of flowering 717 time under drought. This association raises the possibility that qKP-6 contributes to coordinated 718 regulation of K homeostasis and developmental stability under drought, although the 719 underlying mechanism remains to be established. Within the LD block surrounding qKP-6, the 720 only annotated gene ( _Pgl_GLEAN_10012096_ ) encodes a protein of unknown function. Its 721 strong LD with the lead SNP makes it a positional candidate, but functional validation will be 24 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.12.766145; this version posted November 13, 2026. The copyright holder for this](https://doi.org/10.64898/2026.11.12.766145) preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in [perpetuity. It is made available under a CC-BY-NC-ND 4.0 International license.](http://creativecommons.org/licenses/by-nc-nd/4.0/) 722 required to determine whether it directly contributes to K plasticity. The identification of these 723 QTL provides promising opportunities for improving drought tolerance through the 724 manipulation of leaf K concentration. 725 726 **Conclusion** 727 Pearl millet is a cereal crop of major importance for food security, particularly in the driest 728 regions of the world. Accelerating its genetic improvement is crucial to maintain food security 729 under current and future climate conditions. Our results suggest that the response of pearl millet 730 ionome to drought involves the maintenance of ion homeostasis, while highlighting the central 731 role of K both in plant growth and in drought tolerance. The identification of distinct QTL 732 associated with K accumulation and its plasticity in response to drought provides new insights 733 into the genetic basis of K nutrition and plasticity in pearl millet, opening new prospects for 734 breeding. Ultimately, these findings contribute to a better understanding of the mechanisms 735 underlying plant adaptation to drought, a critical challenge in the context of climate change. 736 737 **Declaration of competing interest** 738 The authors report no competing interest. 739 740 **Acknowledgements** 741 This work was supported by the European Plant Phenotyping Network (EPPN2020 project 742 Ionomil n°386), the Royal Society (Anatomics grant ICA-R1-180356 to MB and NK) and the 743 Agence Nationale de la Recherche (ANR-20-CE20-0016 to AG). P.N.G. was supported by a 744 joint PhD grant from the Institut de Recherche pour le Développement (IRD, France) and the 745 French Embassy in the Central African Republic. 746 747 **Data availability** 748 Data are available in the DataSuds database (https://doi.org/10.23708/DWGEAJ). 749 750 25 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.12.766145; this version posted November 13, 2026. The copyright holder for this](https://doi.org/10.64898/2026.11.12.766145) preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in [perpetuity. It is made available under a CC-BY-NC-ND 4.0 International license.](http://creativecommons.org/licenses/by-nc-nd/4.0/) 751 **References** 752 **Affortit P, Faye A, Jones DH,** _**et al.**_ 2024 _a_ . Root metaxylem area influences drought 753 tolerance and transpiration in pearl millet in a soil texture dependent manner. Physiology doi: 754 10.1101/2024.11.09.622826. [Preprint]. 755 **Affortit P, Faye A, Jones DH,** _**et al.**_ 2024 _b_ . Root metaxylem area influences drought 756 tolerance and transpiration in pearl millet in a soil texture dependent manner. bioRxiv. 757 **Alzahrani Y, Alsamadany H, Shah ZH** . 2026. The foliar potassium silicate enhances wheat 758 terminal-drought tolerance and yield via physiological, biochemical and genetic regulation. 759 Frontiers in Plant Science **17**, 1766913. 760 **Arenas S, Djabali Y, Rincent R, Cubry P, Martin M-L, Blein-Nicolas M, Laplaze L,** 761 **Schneider H, Grondin A** . 2025. Modeling plant phenotypic plasticity and its underlying 762 genetic architecture: a comparative study. (M Byrne, Ed.). Journal of Experimental Botany 763 **76**, 2239–2251. 764 **Bates D, Mächler M, Bolker B, Walker S** . 2015. Fitting Linear Mixed-Effects Models 765 Using lme4. Journal of Statistical Software **67** . 766 **Benjamini Y, Hochberg Y** . 1995. Controlling the False Discovery Rate: A Practical and 767 Powerful Approach to Multiple Testing. Journal of the Royal Statistical Society Series B: 768 Statistical Methodology **57**, 289–300. 769 **Bidinger F, Mahalakshmi V, Rao G** . 1987. Assessment of drought resistance in pearl millet 770 (Pennisetum americanum (L.) Leeke). II. Estimation of genotype response to stress. 771 Australian Journal of Agricultural Research **38**, 49. 772 **Bisht N, Singh T, Ansari MohdM, Chauhan PS** . 2026. Mapping the Molecular Circuitry of 773 Plant Nutrient Acquisition and Homeostasis in Stressful Environments: Insights into Sensing, 774 Signaling, and Transport Mechanisms. Journal of Soil Science and Plant Nutrition **26**, 34–52. 775 **Bouain N, Korte A, Satbhai SB, Nam H-I, Rhee SY, Busch W, Rouached H** . 2019. 776 Systems genomics approaches provide new insights into Arabidopsis thaliana root growth 777 regulation under combinatorial mineral nutrient limitation. PLOS Genetics **15**, e1008392. 778 **Brooker R, Brown LK, George TS, Pakeman RJ, Palmer S, Ramsay L, Schöb C,** 779 **Schurch N, Wilkinson MJ** . 2022. Active and adaptive plasticity in a changing climate. 780 Trends in Plant Science **27**, 717–728. 781 **Chang CC, Chow CC, Tellier LC, Vattikuti S, Purcell SM, Lee JJ** . 2015. Second782 generation PLINK: rising to the challenge of larger and richer datasets. Gigascience **4**, 783 s13742-015-0047–8. 784 **Chen F, Ha X, Gao R, Ma H** . 2026. Integrated ion, hormone, and proteomic analyses reveal 785 the mechanisms of drought resistance in alfalfa (Medicago sativa). BMC Plant Biology **26**, 786 490. 787 **Danku JMC, Lahner B, Yakubova E, Salt DE** . 2013. Large-Scale Plant Ionomics. In: 788 Maathuis FJM, ed. Plant Mineral Nutrients: Methods and Protocols. Totowa, NJ: Humana 789 Press, 255–276. 26 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.12.766145; this version posted November 13, 2026. The copyright holder for this](https://doi.org/10.64898/2026.11.12.766145) preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in [perpetuity. It is made available under a CC-BY-NC-ND 4.0 International license.](http://creativecommons.org/licenses/by-nc-nd/4.0/) 790 **De la Fuente C, Grondin A, Sine B,** _**et al.**_ 2024. Glutaredoxin regulation of primary root 791 growth is associated with early drought stress tolerance in pearl millet. (M Nordborg and J 792 Kleine-Vehn, Eds). eLife **12**, RP86169. 793 **De Walsche A, Vergne A, Rincent R, Roux F, Nicolas S, Welcker C, Mezmouk S,** 794 **Charcosset A, Mary-Huard T** . 2025. metaGE: Investigating genotype x environment 795 interactions through GWAS meta-analysis. (A Hancock, Ed.). PLOS Genetics **21**, e1011553. 796 **Debieu M, Kanfany G, Laplaze L** . 2017. Pearl Millet Genome: Lessons from a Tough 797 Crop. Trends in Plant Science **22**, 911–913. 798 **Diongue DML, Roupsard O, Do FC, Stumpp C, Orange D, Sow S, Jourdan C, Faye S** . 799 2022. Evaluation of parameterisation approaches for estimating soil hydraulic parameters 800 with HYDRUS-1D in the groundnut basin of Senegal. Hydrological Sciences Journal **67**, 801 2327–2343. 802 **D’Oria A, Courbet G, Billiot B,** _**et al.**_ 2022. Drought specifically downregulates mineral 803 nutrition: Plant ionomic content and associated gene expression. Plant Direct **6**, e402. 804 **D’Oria A, Courbet G, Lornac A, Pluchon S, Arkoun M, Maillard A, Etienne P,** 805 **Diquélou S, Ourry A** . 2021. Specificity and Plasticity of the Functional Ionome of _Brassica_ 806 _napus_ and _Triticum aestivum_ Exposed to Micronutrient or Beneficial Nutrient Deprivation 807 and Predictive Sensitivity of the Ionomic Signatures. Frontiers in plant science **12**, 641678. 808 **Farooq M, Wahid A, Kobayashi N, Fujita D, Basra SMA** . 2009. Plant drought stress: 809 effects, mechanisms and management. Agronomy for Sustainable Development **29**, 185–212. 810 **Feng X, Liu W, Cao F, Wang Y, Zhang G, Chen Z-H, Wu F** . 2020. Overexpression of 811 HvAKT1 improves drought tolerance in barley by regulating root ion homeostasis and ROS 812 and NO signaling. (R Sunkar, Ed.). Journal of Experimental Botany **71**, 6587–6600. 813 **Fernandez GCJ** . 1992. Effective selection criteria for assessing plant stress tolerance. In: 814 Kuo CG, ed. Proceedings of the International Symposium on Adaptation of Vegetables and 815 Other Food Crops in Temperature and Water Stress. Shanhua, Taiwan: AVRDC, 257–270. 816 **Fischer S, Hilger T, Piepho H-P, Jordan I, Cadisch G** . 2019. Do we need more drought for 817 better nutrition? The effect of precipitation on nutrient concentration in East African food 818 crops. Science of The Total Environment **658**, 405–415. 819 **Frichot E, Mathieu F, Trouillon T, Bouchard G, François O** . 2014. Fast and efficient 820 estimation of individual ancestry coefficients. Genetics **196**, 973–983. 821 **Friedman J, Hastie T, Tibshirani R** . 2010. Regularization Paths for Generalized Linear 822 Models via Coordinate Descent. Journal of Statistical Software **33** . 823 **Huang M, Liu X, Zhou Y, Summers RM, Zhang Z** . 2019. BLINK: a package for the next 824 level of genome-wide association studies with both individuals and markers in the millions. 825 GigaScience **8** . 826 **Intergovernmental Panel On Climate Change (Ipcc)** . 2023. _Climate Change 2022 –_ 827 _Impacts, Adaptation and Vulnerability: Working Group II Contribution to the Sixth_ 27 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.12.766145; this version posted November 13, 2026. The copyright holder for this](https://doi.org/10.64898/2026.11.12.766145) preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in [perpetuity. It is made available under a CC-BY-NC-ND 4.0 International license.](http://creativecommons.org/licenses/by-nc-nd/4.0/) 828 _Assessment Report of the Intergovernmental Panel on Climate Change_ . Cambridge 829 University Press. 830 **Johnson R, Vishwakarma K, Hossen MS, Kumar V, Shackira AM, Puthur JT, Abdi G,** 831 **Sarraf M, Hasanuzzaman M** . 2022. Potassium in plants: Growth regulation, signaling, and 832 environmental stress tolerance. Plant physiology and biochemistry: PPB **172**, 56–69. 833 **Kang HM, Zaitlen NA, Wade CM, Kirby A, Heckerman D, Daly MJ, Eskin E** . 2008. 834 Efficient Control of Population Structure in Model Organism Association Mapping. Genetics 835 **178**, 1709–1723. 836 **Kuznetsova A, Brockhoff PB, Christensen RHB** . 2017. **lmerTest** Package: Tests in Linear 837 Mixed Effects Models. Journal of Statistical Software **82** . 838 **Liu H, Song S, Zhang H, Li Y, Niu L, Zhang J, Wang W** . 2022. Signaling Transduction of 839 ABA, ROS, and Ca2+ in Plant Stomatal Closure in Response to Drought. International 840 Journal of Molecular Sciences **23**, 14824. 841 **Mantel N** . 1967. The detection of disease clustering and a generalized regression approach. 842 Cancer Research **27**, 209–220. 843 **Marschner H, Marschner P (Eds)** . 2012. _Marschner’s mineral nutrition of higher plants_ . 844 London ; Waltham, MA: Elsevier/Academic Press. 845 **Mishra S, Bisht D, Amtmann A, Srivastava AK, Pandey GK** . 2026. Potassium Deficiency 846 and Hormone Signalling in Plants. Plant, Cell & Environment **49**, 2935–2951. 847 **Nakagawa S, Schielzeth H** . 2013. A general and simple method for obtaining _R_ [2] from 848 generalized linear mixed‐effects models. (RB O’Hara, Ed.). Methods in Ecology and 849 Evolution **4**, 133–142. 850 **Nakombo-Gbassault P, Arenas S, Affortit P,** _**et al.**_ 2025. Genetic control of the leaf 851 ionome in pearl millet and correlation with root and agromorphological traits. (PK Subudhi, 852 Ed.). PLOS One **20**, e0319140. 853 **Nieves-Cordones M, Alemán F, Martínez V, Rubio F** . 2014. K+ uptake in plant roots. The 854 systems involved, their regulation and parallels in other organisms. Journal of Plant 855 Physiology **171**, 688–695. 856 **Nieves-Cordones M, García-Sánchez F, Pérez-Pérez JG, Colmenero-Flores JM, Rubio** 857 **F, Rosales MA** . 2019. Coping With Water Shortage: An Update on the Role of K+, Cl-, and 858 Water Membrane Transport Mechanisms on Drought Resistance. Frontiers in Plant Science 859 **10**, 1619. 860 **Oakey H, Verbyla A, Pitchford W, Cullis B, Kuchel H** . 2006. Joint modeling of additive 861 and non-additive genetic line effects in single field trials. Theoretical and Applied Genetics 862 **113**, 809–819. 863 **Pucher A, Sy O, Angarawai II, Gondah J, Zangre R, Ouedraogo M, Sanogo MD,** 864 **Boureima S, Hash CT, Haussmann BIG** . 2015. Agro‐morphological Characterization of 865 West and Central African Pearl Millet Accessions. Crop Science **55**, 737–748. 28 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.12.766145; this version posted November 13, 2026. The copyright holder for this](https://doi.org/10.64898/2026.11.12.766145) preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in [perpetuity. It is made available under a CC-BY-NC-ND 4.0 International license.](http://creativecommons.org/licenses/by-nc-nd/4.0/) 866 **Quagliata G, Ferrucci A, Marín-Sanz M, Barro F, Vigani G, Astolfi S** . 2025. Ionome 867 profiling discriminate genotype-dependent responses to drought in durum wheat. Journal of 868 Plant Physiology **308**, 154487. 869 **Ravet K, Pilon M** . 2013. Copper and Iron Homeostasis in Plants: The Challenges of 870 Oxidative Stress. Antioxidants & Redox Signaling **19**, 919–932. 871 **Rezaei EE, Webber H, Asseng S, Boote K, Durand JL, Ewert F, Martre P, MacCarthy** 872 **DS** . 2023. Climate change impacts on crop yields. Nature Reviews Earth & Environment **4**, 873 831–846. 874 **Rodríguez-Álvarez MX, Boer MP, van Eeuwijk FA, Eilers PHC** . 2018. Correcting for 875 spatial heterogeneity in plant breeding experiments with P-splines. Spatial Statistics **23**, 52– 876 71. 877 **Salt DE, Baxter I, Lahner B** . 2008. Ionomics and the Study of the Plant Ionome. Annual 878 Review of Plant Biology **59**, 709–733. 879 **Satyavathi CT, Ambawat S, Khandelwal V, Srivastava RK** . 2021. Pearl Millet: A 880 Climate-Resilient Nutricereal for Mitigating Hidden Hunger and Provide Nutritional 881 Security. Frontiers in Plant Science **12**, 659938. 882 **Schneider HM** . 2022. Characterization, costs, cues and future perspectives of phenotypic 883 plasticity. Annals of Botany **130**, 131–148. 884 **Sehgal D, Skot L, Singh R,** _**et al.**_ 2015. Exploring Potential of Pearl Millet Germplasm 885 Association Panel for Association Mapping of Drought Tolerance Traits. PLoS ONE **10**, 886 e0122165. 887 **Shrestha N, Hu H, Shrestha K, Doust AN** . 2023. Pearl millet response to drought: A 888 review. Frontiers in Plant Science **14** . 889 **Srivastava RK, Yadav OP, Kaliamoorthy S,** _**et al.**_ 2022. Breeding Drought-Tolerant Pearl 890 Millet Using Conventional and Genomic Approaches: Achievements and Prospects. Frontiers 891 in Plant Science **13**, 781524. 892 **Tibshirani R** . 1996. Regression Shrinkage and Selection Via the Lasso. Journal of the Royal 893 Statistical Society Series B: Statistical Methodology **58**, 267–288. 894 **Tränkner M, Jákli B, Tavakol E, Geilfus C-M, Cakmak I, Dittert K, Senbayram M** . 895 2016. Magnesium deficiency decreases biomass water-use efficiency and increases leaf 896 water-use efficiency and oxidative stress in barley plants. Plant and Soil **406**, 409–423. 897 **Van Rossum B-J** . 2020. statgenSTA: Single Trial Analysis (STA) of Field Trials. 1.0.15. 898 **Varshney RK, Shi C, Thudi M,** _**et al.**_ 2017. Pearl millet genome sequence provides a 899 resource to improve agronomic traits in arid environments. Nature Biotechnology **35**, 969– 900 976. 901 **White PJ, Broadley MR** . 2009. Biofortification of crops with seven mineral elements often 902 lacking in human diets – iron, zinc, copper, calcium, magnesium, selenium and iodine. New 903 Phytologist **182**, 49–84. 29 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.12.766145; this version posted November 13, 2026. The copyright holder for this](https://doi.org/10.64898/2026.11.12.766145) preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in [perpetuity. It is made available under a CC-BY-NC-ND 4.0 International license.](http://creativecommons.org/licenses/by-nc-nd/4.0/) 904 **Zeileis A, Köll S, Graham N** . 2020. Various Versatile Variances: An Object-Oriented 905 Implementation of Clustered Covariances in _R_ . Journal of Statistical Software **95** . 906 **Zhang Z, Ersoz E, Lai C-Q,** _**et al.**_ 2010. Mixed linear model approach adapted for genome907 wide association studies. Nature Genetics **42**, 355–360. 908 **Zhang Y, Liu B, Kong F, Chen L** . 2023. Nutrient-mediated modulation of flowering time. 909 Frontiers in Plant Science **14** . 910 911 30 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.12.766145; this version posted November 13, 2026. The copyright holder for this](https://doi.org/10.64898/2026.11.12.766145) preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in [perpetuity. It is made available under a CC-BY-NC-ND 4.0 International license.](http://creativecommons.org/licenses/by-nc-nd/4.0/) 912 **Figures** **legend** 913 914 915 **Fig. 1. Foliar concentrations of the 17 mineral elements under well-watered and water-** 916 **stressed conditions, in 2021 and 2022.** Boxplots show genotype BLUEs. Asterisks indicate a 917 significant difference between water regimes within a year (paired t-test or Wilcoxon signed 918 rank test, depending on residual normality; Benjamini-Hochberg FDR-adjusted p-values, 919 corrected within each year: * p<0.05, ** p<0.01, *** p<0.001; ns: not significant). 920 31 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.12.766145; this version posted November 13, 2026. The copyright holder for this](https://doi.org/10.64898/2026.11.12.766145) preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in [perpetuity. It is made available under a CC-BY-NC-ND 4.0 International license.](http://creativecommons.org/licenses/by-nc-nd/4.0/) 921 922 **Fig. 2. Multivariate structure of the pearl millet leaf ionome under water deficit.** 923 (A) Biplot of the first two principal components (PC1, 46.2%; PC2, 14.6%) of the global PCA 924 computed on foliar concentrations (BLUEs) of 17 elements across four environments (well 925 watered, WW, and water-stressed, WS, in 2021 and 2022). Points are genotype × environment 926 observations, dashed ellipses are 95% ellipses for each environment, and arrows show the 927 element loadings (scaled for display). (B) Percentage of the variance of PC1 to PC5 explained 928 by year, treatment, year × treatment, genotype, genotype × year, genotype × treatment and 929 residual, from sequential sums of squares of a linear model. Hatched segments are not 930 significant (p ≥ 0.05) and percentages are shown for significant terms of at least 4.5%. The 931 percentage of total variance explained by each component is given in brackets. (C) Loadings 932 (eigenvector coefficients) of the 17 elements on PC2, ranked from highest to lowest. Positive 933 loadings (purple) indicate elements higher under well-watered conditions and negative 934 loadings (red) elements higher under water deficit 935 32 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.12.766145; this version posted November 13, 2026. The copyright holder for this](https://doi.org/10.64898/2026.11.12.766145) preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in [perpetuity. It is made available under a CC-BY-NC-ND 4.0 International license.](http://creativecommons.org/licenses/by-nc-nd/4.0/) 936 937 **Fig. 3. Contribution of variance components to total variance of leaf ion concentrations.** 938 Variance is partitioned among random effects (genotype, treatment, genotype × treatment 939 interaction), expressed as a percentage of total trait variance. Asterisks indicate significant 940 genotype × treatment interactions ( _p_ < 0.05). 941 942 33 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.12.766145; this version posted November 13, 2026. The copyright holder for this](https://doi.org/10.64898/2026.11.12.766145) preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in [perpetuity. It is made available under a CC-BY-NC-ND 4.0 International license.](http://creativecommons.org/licenses/by-nc-nd/4.0/) 943 944 **Fig. 4. Genome-wide association analysis of foliar potassium plasticity and** 945 **characterization of the qKP-6 locus on chromosome 6.** 946 (A) Manhattan plot of -log₁₀(p) values for all SNPs from the GWAS of the potassium plasticity 947 index (LR_K). Colours distinguish chromosomes 1 to 7 and unanchored contigs (grey). The 948 blue line indicates the genome-wide significance threshold ( _-log₁₀ p = 5_ ). 949 (B) Quantile-quantile plot of observed against expected -log₁₀(p) values. The red line is the 950 expectation under the null hypothesis. 951 (C) Regional Manhattan plot of the significant locus on chromosome 6 (110,236 to 110,261 952 kb), with the two significant SNPs in red and the local linkage disequilibrium (LD) structure. 953 Pairwise LD (r²) values among SNPs are displayed as a colour gradient from 0 to 1. The yellow 954 segment marks the transcribed region of _Pgl_GLEAN_10012096._ 955 (D) LR_K by allelic class at CM007987_110250858 (G/G, G/A, A/A; n = number of lines). 956 Points are line × year observations, boxes show the median and interquartile range, and white 34 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.12.766145; this version posted November 13, 2026. The copyright holder for this](https://doi.org/10.64898/2026.11.12.766145) preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in [perpetuity. It is made available under a CC-BY-NC-ND 4.0 International license.](http://creativecommons.org/licenses/by-nc-nd/4.0/) 957 diamonds the means. Different letters indicate significant differences between classes ( _p <_ 958 0.05; Tukey-adjusted pairwise contrasts of estimated marginal means from the linear mixed 959 model with year as fixed effect and genotype as random effect). 35