in silico Plants

The assimilation of carbon dioxide by plants can be predicted by the Farquhar, von Caemmerer and Berry model of photosynthesis. This largely mechanistic model is central to understanding how plants influence Earths climate. However, it represents the use of light by photosynthesis using an empirical formulation. Johnson and Berry proposed an alternative mechanistic formulation based on the functioning of the cytochrome b6f complex that includes key steps in light harvesting and electron transport. We compared both formulations using photosynthetic light response measurements from 146 C3 species spanning arctic to tropical biomes and implemented them in the terrestrial biosphere model ELM-FATES to simulate global photosynthesis. The Johnson and Berry formulation better fitted the measured response of leaf-level photosynthesis to light, and predicted lower photosynthetic rates at intermediate light levels, which decreased global estimations of terrestrial photosynthesis by 8%. Our findings support adopting the Johnson and Berry formulation to improve model representation of global carbon cycle modeling.

Intercropping is a promising strategy to improve productivity and sustainability in agricultural systems, but designing effective genotype combinations remains a major challenge owing to the rapid increase in possible pairings as the number of candidate genotypes increases. This creates a practical bottleneck because field evaluation of all combinations is infeasible under realistic resource constraints. Here, we propose a framework that integrates genomic prediction and Bayesian optimization to support efficient decision-making for intercropping system design. Using genome-wide marker data from sorghum and soybean, we simulated intercropping performance across 5,214 genotype pairs under certain genetic architectures, including variation in heritability, correlations between direct and indirect genetic effects, and the contribution of pair-specific interactions. Genomic prediction models incorporating direct and indirect genetic effects substantially improved prediction accuracy compared with models based on direct genetic effects alone, and inclusion of specific mixing ability further enhanced the performance under high-heritability conditions. When coupled with Bayesian optimization, the models rapidly identified superior genotype pairs, requiring fewer evaluation cycles than random or prediction-only search strategies. Acquisition functions that account for predicted uncertainty were most effective in complex scenarios involving interaction effects or negative correlations between direct and indirect effects. These results demonstrate that combining genomic prediction with Bayesian optimization can substantially reduce the experimental burden associated with intercropping design, while improving the efficiency of identifying high-performing genotype pairs. The proposed framework provides a practical approach for prioritizing candidate mixtures in breeding and field evaluation, and contributes to the development of data-driven strategies for sustainable agricultural systems. HighlightsO_LIA data-driven framework was developed to optimize genotype pairs in intercropping. C_LIO_LIModeling indirect effects improved prediction accuracy across genotype pairs. C_LIO_LIPair-specific interactions enhanced prediction under high-heritability conditions. C_LIO_LIBayesian optimization identified superior pairs under limited evaluation capacity. C_LIO_LIThe framework reduces field-testing requirements for intercropping system design. C_LI

Growth kinematics and soil mechanics are key to explain how roots overcome the mechanical resistance of soil, yet few studies are linking these two factors. Formulas for cone penetration tests are typically used to infer the friction experienced by roots, but these fail to consider how growth affects the external forces applied on the root. This study formalised how expansive growth in the root apical meristem can reduce soil friction, and applied the framework to analyse the growth strategy of 6 plant species. The results of the analysis revealed trade-offs between reducing frictions, maintaining a desired growth trajectory and elongation rate. A shorter elongation zone can reduce the fraction of the mechanical energy lost to friction, but this is done at the expense of the elongation rate. A sharper tip or increased radius can help roots maintain the elongation rate at no energetic cost, but these strategies come with the cost of growth instability (tortuous roots) and decrease in specific root length respectively. During establishment, root strategies may therefore occupy a 2-dimensional trait space in which the mechanical efficiency of growth is balanced against the explorative-exploitative trade-off. HighlightsGrowth and form of root tips explain how plants overcome mechanical resistance from the soil Trade-offs link the energy lost by friction, growth stability and elongation rate of roots Larger roots allow faster growth independently of these trade-offs New framework formalises plants strategies to acquire soil resources

A global nuclear war could inject soot into the stratosphere, blocking sunlight and causing rapid cooling. Assessments of the resulting agricultural collapse rely on crop models never validated under such conditions. We grew wheat, canola, and potato in growth chambers simulating the light and temperature of an extreme nuclear winter at tropical and temperate sites. In the tropical chamber (18-20 {degrees}C, 200 mol m-2 s-1 PAR), all three crops produced viable yields. Wheat yielded 2.1-2.3 t/ha (n=3 well-watered, n=3 water-stressed pots), 60% of the global average, and single-pot observations of canola and potato suggested biological yields comparable to global averages. In the temperate chamber simulating nuclear winter irradiance (60-360 mol m-2 s-1), wheat stems collapsed under their own weight. Although hand-harvesting recovered 0.6-2.8 t/ha of grain, mechanical field harvest of a flat canopy would recover substantially less. This failure mode was not observed in a higher-light control chamber and is not captured by existing crop models, which may therefore overestimate temperate cereal production under nuclear winter. Canola produced comparable yields under both temperate light regimes without lodging. Empirical screening of additional staples is needed to identify which remain viable under nuclear winter.

Genomic prediction models that account genotype-by-environment (GxE) have the potential to accelerate the rate of genetic gain for yield and agronomic performance, yet relatively few studies have applied GxE prediction in public soft red winter wheat (Triticum aestivum) breeding programs. In this study, we extended a reaction norm-based genomic prediction framework by integrating weather-based environmental covariates to more effectively capture genotype- environment interactions. Key agronomic traits, including seed yield, plant height, test weight, and heading date, were evaluated across 33 environments (location-year) using over 3,200 breeding lines from the North Carolina State University small grains breeding program. Multiple genomic prediction models were compared using several cross-validation (CV) schemes representing common breeding scenarios. Across traits, the reaction norm M5 model, which incorporates both GxE and genotype-by-environmental covariate interactions (GxO), achieved the highest prediction accuracy (PA) in CV2 (predicting incomplete field trials) and CV1 for yield and test weight (predicting new lines). The highest PA was observed for test weight under CV2 (0.54) and for yield under CV1 (0.41). Under CV0 (predicting new environments), the M3 model incorporating GxE produced highest PA across traits, with the greatest accuracy for plant height (0.45), although differences among M2, M3, and M4 were small. Prediction under CV00 (predicting new lines in new environments) remained more challenging, with PA values 0.10 - 0.20 across traits. Overall, our results demonstrate that integrating environmental covariates into genomic prediction models can improve predictive performance across diverse wheat-growing environments in North Carolina, supporting their utility for applied breeding efforts. CORE IDEASO_LIIntegrating genotype-by-environment (GxE) interactions with environmental covariates improves prediction accuracy across environments. C_LIO_LIModel performance varies by prediction scenario, with different approaches performing best for new lines, incomplete trials, or new environments. C_LIO_LIPrediction of new lines in new environments remains challenging. C_LI PLAIN LANGUAGE SUMMARYThis study explores how adding environmental information to genomic prediction models can improve prediction accuracy in a public winter wheat breeding program. Using data from multi-environment trials conducted across diverse conditions in North Carolina, we evaluated statistical models that capture how different wheat lines respond to changing environments. By incorporating weather data, we improved the ability to predict performance across locations and years. These findings provide practical insights for refining selection strategies and accelerating genetic gain in wheat breeding.

Accurate modeling of genotype-by-environment (GxE) interaction is critical for genomic prediction in plant breeding but remains challenging due to complex interaction structures. Conventional models often use the Hadamard product of genotype and environment covariance matrices to capture joint similarity, which may not fully represent GxE complexity. Here we propose a novel framework that derives covariance structures from the matrix multiplication of genotype and environment kernels, decomposing these into symmetric components incorporated as random effects in mixed models. Evaluated for 11 wheat and rice multi-environment datasets and across, this approach consistently outperformed the traditional Hadamard-based model, improving prediction accuracy by up to 13.2% in Pearsons correlation and enhancing top-selection accuracy. Combining both methods yielded the highest performance, indicating complementary information capture. This framework offers a flexible, interpretable, and computationally feasible extension for modeling GxE interaction, potentially enhancing genomic selection effectiveness under diverse environmental conditions.

Crop improvement can enhance food security, but side effects, such as trade-offs between valuable agronomic traits, are common. Likewise, fertilisation helps ensure high yields, but can devalue mutualisms with soil microbes that would otherwise be essential for nutrient acquisition. If the need for nutritional mutualisms is reduced in crops, mutualisms could be disrupted by selection relaxation or allocation trade-offs. Thus, crops could achieve high yields in spite of, or because of, disruption of nutritional mutualisms. Alternatively, the highest-yielding plants might flourish because they maximise nutrient acquisition from both symbionts and the soil. Here, enhanced mutualism could evolve over the course of agricultural crop improvement. To investigate whether high yields in cultivars and wild accessions are negatively or positively genetically correlated with outcomes in the legume-rhizobia mutualism, we measured whether yield and symbiosis traits trade-off or are positively genetically correlated among cultivars and wild accessions. We also tested whether this relationship differs between accessions released before or after 1950. We measured genetic correlations between yield and mutualism traits in 87 domesticated pea (Pisum sativum) accessions in a common garden agricultural field across three years. Seed yield and N2 fixation (%Ndfa) were positively genetically correlated. While N fixation was more strongly predictive of yield in the pre-1950 accessions than the post-1950 accessions, the underlying positive genetic correlation between the traits did not differ between the groups. The positive genetic correlation between yield and N2 fixation indicates that selection to increase yields has maintained or increased the benefits of the rhizobial mutualism in pea. Our findings predict that breeding to increase yield may continue to produce pea cultivars that get a greater proportion of their N from rhizobia, enhancing symbiotic mutualism and reducing the proportion of N supplied by fertilisation.

PurposeGenotype-by-environment (G x E) interactions represent a major obstacle to increasing genetic gain in crop breeding, with the underlying physiological drivers often remaining obscured within conventional statistical models. This case study presents a novel framework that transforms the latent factors from Factor Analytic (FA) multi-environment trial (MET) models into heritable quantitative traits, enabling the genetic dissection of adaptive response patterns. MethodsA Factor Analytical Linear Mixed Model (FA-LMM) was fit to plot-level yield data for 1,036 barley genotypes across eight Australian trials. ResultsCorrelation of the factor loadings with APSIM-simulated environmental covariates demonstrated that the second latent factor FA2 was strongly correlated with the Water Stress Index (r = -0.83) during the critical flowering period, establishing water availability as the main biological axis of crossover Gx E. Genotypic scores for the derived traits, Overall Performance (OP) and Water Stress Response (WSR), were subjected to high-resolution haplotype-based mapping using local Genomic Estimated Breeding Values (GEBV). ConclusionThis analysis successfully identified major genomic regions that accounted for a substantial proportion of the additive genetic variance. Gene Ontology enrichment of candidate genes within the top haploblocks implicated fundamental pathways related to energy homeostasis, root development, and stress response, with notable candidates including FTsH11, BPS1, and TDP1. The distribution of favourable Haplotypes of Interest (HOI) in elite cultivars suggested a historical signature of inadvertent selection for these adaptive mechanisms. This framework provides an explicit bridge between statistical modelling and functional genomics, offering breeders actionable genetic targets for accelerated development of climate-resilient cereals.

O_LIIn fluctuating environments, the kinetics of stomatal opening and closing influence the balance between carbon gain and water loss. Smaller guard cells may respond faster to fluctuating environmental conditions because of their greater surface area for osmolyte flux relative to cell volume. A related hypothesis is that operational stomatal conductance (gop) is often well below its theoretical maximum (gmax) because at this stomatal aperture, guard cell volume is poised to change rapidly with small changes in turgor pressure. C_LIO_LIWe analyzed 2,124 estimates of stomatal closure kinetics in response to an abrupt increase in vapor pressure deficit (VPD) among 29 diverse wild tomato populations in the genus Solanum. C_LIO_LILeaves with small guard cells and a lower gop to gmax ratio (fgmax) closed faster, but explained variation in kinetic parameters at different levels of biological organization. Guard cell size had high phylogenetic heritability and varied relatively little within populations, whereas fgmax varied mostly among individuals and between light intensity treatments. C_LIO_LISmaller stomata can be speedier, but only if stomata are held at an aperture where they are responsive to changing turgor pressure. Selection on stomatal speed may influence not only anatomical traits like guard cell size, but also physiological controls on gop. C_LI

The performance of a single cross is determined by the average additive effects of the parents, as well as the interactions between them. These quantities can be estimated using an appropriate genetic design, allowing for the estimation of general (GCA) and specific (SCA) combining abilities. The prediction of GCA for new parents and the total genetic value of unrealized crosses can be made when genome-wide marker information is available. Several studies in crops such as maize and rice have demonstrated the potential of genomic-assisted prediction of single-cross performance in economically important crops. However, no study to date has explored its relevance in perennial ryegrass, an obligate allogamous species that is bred in genetically heterogeneous families. In this study, we aimed to estimate genetic parameters and assess the ability of genomic models to predict the performance of F2 families in terms of dry matter yield and nutritive quality traits. We used data from a large partial diallel involving 104 parents from two distinct subpopulations, as inferred by admixture analysis. F2 families were evaluated in multiple environments and under two nitrogen availability conditions. Genotyping-by-sequencing of the parent plants produced 42,145 variants after quality control, which were used to estimate genomic relationships based on identity-by-state. Variance component estimation revealed limited GCA and SCA interactions with the environment, and particularly with nitrogen management. The predictive abilities of two parental models exceeded 0.60 and often surpassed 0.70 for most traits. However, incorporating non-additive effects into the model did not improve predictive ability. We leveraged the genetic diversity among parents to map genomic regions associated with all recorded traits. Genome-wide association studies (GWAS) by genomic best linear unbiased prediction (GBLUP) identified six quantitative trait loci (QTL) regions, with 45 candidate genes within the linkage disequilibrium range, estimated at approximately 92 kb. Our results demonstrate that genomic prediction of single crosses can be performed with high accuracy, especially when both parents are also progenitors of families in the training set.

Parameter estimation in nonlinear biological dynamical systems is a difficult inverse problem because the governing equations are often stiff or oscillatory, the data are sparse and noisy, and the objective landscape is non-convex. Physics-informed neural networks (PINNs) offer an alternative to purely simulation-based calibration by representing state trajectories with neural networks while penalizing violations of the governing equations. This paper studies the empirical reliability of PINNs for recovering the parameters of the repressilator, a synthetic genetic oscillator formed by three cyclically repressive genes. We use synthetic time-series generated from the standard ordinary differential equation model and train inverse PINNs to estimate the production parameter {beta} and the Hill coefficient n. The study varies observation noise, partial observation of repressors, sampling density, sensitivity to initial parameter guesses, and the difference between stable and oscillatory regimes. The results show that PINNs can reconstruct trajectories accurately when the model structure is correct and the three repressors are observed, but parameter recovery is more fragile than trajectory fitting. Noise, sparse sampling, unobserved variables, and unfavorable initial guesses increase the risk of biased estimates. The stable regime is easier to reconstruct, whereas the oscillatory regime provides richer information but also exposes optimization sensitivity. These findings support PINNs as a useful reverse-engineering tool for small gene-regulatory ODE models, while highlighting the need for repeated runs, uncertainty reporting, and experimental designs that improve identifiability.

Bursty transcription in single cells typically produces over-dispersed, skewed, and sometimes heavy-tailed expression distributions that are explained by two-state Markov models of the promoters. While the gold standard for simulation is exact stochastic sampling with Gillespies algorithm, obtaining thousands of timed traces is computationally costly. Surrogate models based on stochastic differential equations (SDEs) are widely used to speed up this simulation process. An example is the Chemical Langevin Equation based on Gaussian noise, which, however, does not capture heavy-tailed noise. In this work, we present a unified SDE framework that combines deterministic drift, Gaussian fluctuations, and additive sporadic jumps of arbitrary distributions, and provide an open-source Python implementation, bcrnnoise. The framework subsumes standard surrogate models and allows for vectorized generation of batches of transcription traces. We assess computational speed and accuracy of common surrogate models along with new models, showing that high accuracy can be obtained while reducing computational cost up to two orders of magnitude.

Plant disease detection using deep learning is essential for precision agriculture, enabling early and automated crop health monitoring. This study proposes an end-to-end transfer learning pipeline, LeafyVGG-16, for multi-class classification of plant diseases and nutrient deficiencies using a tomato leaf dataset. The framework integrates data preprocessing, augmentation, and a VGG-16 backbone with a two-stage fine-tuning strategy. The proposed model is evaluated against CNN, DenseNet-121, Inception-V3, EfficientNetB0, and ResNet-50, achieving an accuracy of 0.93 with precision, recall, and F1-scores of 0.93, 0.90, and 0.92, respectively. These results demonstrate the effectiveness of transfer learning for fine-grained plant disease recognition. We further evaluate model robustness under adversarial cyber attacks to assess deployment reliability in agricultural systems. Under Fast Gradient Sign Method (FGSM) attacks ({epsilon} = 0.01- 0.05), the model shows an accuracy drop of 1%-7.5%, while Projected Gradient Descent (PGD) attacks ({epsilon} = 0.05, step size = 0.005, 10 iterations) produce similar degradation, highlighting the models vulnerability to adversarial perturbations. These findings highlight potential security and reliability risks in AI-based agricultural decision-making systems. Future work will focus on improving robustness and cyber-resilience and extending this framework to other crops for secure and context-aware deployment in resource-constrained environments.

Ensuring global food security under rapid climate change demands accelerated genetic gain and breeding strategies that address complex Genotype-by-Environment (GxE) interactions. Traditional genomic selection models often fail to account for novel or extreme climates.Furthermore, integrating mechanistic crop growth models (CGMs) using traditional Bayesian frameworks to solve this issue presents severe computational bottlenecks. Here, we introduce DeepBioGS, a novel hybrid framework that integrates genomic selection with biophysical growth modelling via a fully differentiable deep learning architecture. DeepBioGS utilises a parameter-prediction multi-layer perceptron to map high-dimensional genomic markers to latent, highly heritable physiological traits (Genotype-Specific Parameters; GSP). These parameters mechanistically predict crop phenology across diverse environments. Using two multi-environment wheat datasets comprising over 6,000 genotypes, DeepBioGS extracted latent traits with near-perfect SNP-based heritability values (0.95-1.00). Crucially, the framework demonstrated superior or comparable predictive accuracy (up to r2 = 0.77) against standard genomic best linear unbiased prediction (GBLUP) and traditional Bayesian CGM-WGP models. Its architecture drastically improved computational scalability by enabling standard backpropagation, effectively bypassing the stochastic sampling limitations of approximate Bayesian methods. Most importantly for climate adaptation, DeepBioGS allowed accurate forecasting of genotype performance in entirely unobserved environmental conditions. By merging the representational power of deep learning with the structural constraints of biophysics, DeepBioGS provides a highly scalable, interpretable tool to navigate GxE interactions, enabling the assessment of cultivars under future climate scenarios, thus optimising crop breeding for a changing global environment.

O_LISymbiosis between legumes and rhizobia is beneficial on nutrient-poor soils, as it enables the fixation of atmospheric N2. To establish this symbiosis, gene expression in both the host plant and the symbiont has to be regulated. To understand the underlying RNA-mediated regulation of host gene expression, we designed experiments to identify competing endogenous networks involving circular RNA, microRNA, and linear transcripts during symbiosis, using wt and symbiosis-deficient Lotus japonicus mutants with the rhizobium Mesorhizobium loti (M. loti). C_LIO_LICircRNA, miRNA, and linear transcripts were identified from Lotus japonicus wildtype and CCamK mutant (ccamk-13; snf-1) seedlings without inoculation or with M. loti inoculation using deep short-read sequencing with rRNA-depletion and random primers. C_LIO_LIDifferentially expressed miRNAs showed negative correlations to predicted target genes and may regulate symbiotic processes. The symbiosis essential iron-sensor LjnsRING/BRUTUS expresses a circRNA which was upregulated in symbiotic treatments. This circRNA may act as a target mimic and contribute to nodule longevity. CircRNAs are predicted to act predominantly as trans-regulatory molecules with similar frequencies in Arabidopsis thaliania, Oryza sativa, and Lotus japonicus. C_LIO_LIWe identified novel miRNAs, long noncoding RNAs, and circRNAs, and nominated several as potential new regulatory non-coding RNAs that may act as target mimics to stabilize genes and support symbiosis. C_LI SummarySymbiosis between Lotus japonicus and Mesorhizobium loti involves treatment-specific regulation of competing endogenous RNA networks involving circular RNA, miRNA, and linear transcripts.

Time-resolved measurements are central to calibrating mechanistic dynamical models, but current inference frameworks typically assume that reported measurement times are exact. In practice, actual sampling times may deviate from reported times because of sample-handling delays, imper-fect synchronization, or reporting errors. Here, we present a Bayesian framework for parameter inference in ordinary differential equation models that explicitly accounts for uncertainty in measurement times. We formulate latent measurement times as random variables and derive a joint and marginalized posterior. To compute the marginal likelihood efficiently, we augment the original dynamical system with additional state variables that evaluate the required integrals during numerical simulation. This reduces the dimensionality of the estimation problems and allows for efficient and reliable Markov chain Monte Carlo sampling. Across synthetic examples and a published model of carotenoid cleavage in Arabidopsis thaliana, neglecting time uncertainty led to biased estimates and overconfident uncertainty quantification, whereas the proposed marginalized formulation recovered reliable parameter estimates while substantially improving sampling efficiency and scalability. These results identify measurement time uncertainty as an important source of variability in dynamic modeling and establish posterior marginalization as a practical strategy for robust mechanistic inference.

Flowering time and monocarpic senescence are tightly environmentally and genetically controlled. Typically, early flowering and staygreen traits are associated with opposing life-history strategies; stress avoidance versus adaptation; with flowering time an overarching regulator of crop cycle length. We developed RIL populations segregating for Ppd-1 and NAM-1 variation, which are otherwise isogenic. Multi-year field experiments enabled exploration and uncoupling of the relationship between heading and staygreen traits. Heading date manipulation enabled introduction of staygreen traits to their target breeding environments, characterised by a hot-finish. Under moderate stress, we report a 2.9% and 1.9% increase in grain width (P<0.0001), and 5.8% and 3.7% increase in TGW (P<0.0001), plus significantly greater yield (P<0.1) for late heading staygreen RILs homozygous for NAM-A1, and NAM-D1 missense variants, respectively. Grain yield increases were proportionate to the delay in senescence, being greater for the NAM-A1 than the NAM-D1 variant. For RIL populations segregating for both traits, senescence variation was observed relative to heading-date. Regarding grain yield, the staygreen trait-associated increase in source size could not compensate for the Ppd-1a associated pleiotropic reduction in sink size, even under hypothesised continental target breeding environments, with trait competition identified. Therefore, to maximise the benefits associated with staygreen traits, especially in early-heading favouring environments required targeted manipulation of source-sink dynamics, and we propose multiple strategies. HighlightStaygreen traits were associated with extending grain fill duration, increasing grain width, TGW and grain yield. There appears an antagonist relationship between earlier heading and staygreen traits.

In quantitative genetics, candidate SNPs are identified through genotype-phenotype associations inferred with genome-wide association studies (GWAS). In this study, we explore an alternative approach to detect genetic variants with non-neutral effects by tracking temporal trends in allele frequency in a winter wheat (Triticum aestivum L.) breeding population over an eight-year period, from which signals of selection may be inferred. Selection signatures were inferred with a generalized linear model, where we modeled trends in allele frequency as a function of time (crossing year). These signatures of selection were used to prioritize variants. Associations between phenotypic performance and individual load of prioritized variants were then investigated. Furthermore, we assessed whether incorporating selection information into a genomic best linear unbiased prediction (GBLUP) model improves model performance in terms of quality of fit and prediction ability. Our findings indicate that the inferred signals of selection are effective in identifying non-neutral variants. Variants under strong negative selection were associated with a decrease in protein content adjusted for grain yield (p-value < 0.01), while genetic variants that had been under moderate to high levels of positive selection were associated with increased grain yield (p-value < 0.01). However, incorporating selection information did not improve prediction accuracy. In conclusion, temporal trends in allele frequency can be used to detect non-neutral variants. The proposed approach may hence complement traditional quantitative genetic methods for detecting non-neutral genetic variation. This approach may allow breeders to detect non-neutral variants earlier in the breeding cycle, without resorting to phenotypic data.

Belowground plant trait research has predominantly focused on trade-offs in fine root traits via the root economics space. Yet, this fine root framework captures only a fraction of the functional strategies plants employ beneath the soil surface. Here, we broaden the perspective on belowground plant functioning by integrating traits related to root system extent, clonality and bud banks, using data from the new UNDERPLOT database. This integration links measurable traits to key belowground functions: resource acquisition, spatial exploration, and persistence. Our analysis shows that the fine root economics space explains less than 5% of the variation in traits related to root system extent, clonality, and bud banks. Instead, an expanded trait analysis reveals three significant dimensions, explaining 62% of total trait variation. The third dimension, represents an independent, persistence-related gradient, not captured by existing root economics frameworks. We propose that understanding belowground plant strategies requires embracing additional functional gradients. The strategy of persistence, in particular, varies significantly across growth forms and is a critical dimension of plant response to resource limitation and stress, becoming increasingly important as global change shifts disturbance regimes.

Grain legumes such as pea (Pisum sativum L.) accumulate large amounts of seed storage proteins without nitrogen fertilization due to their symbiosis with nitrogen-fixing bacteria, making them a key source of plant-based proteins. Seed growth and the accumulation of seed storage proteins are tightly regulated by complex gene networks; however, the mechanisms governing these processes in pea remain poorly understood. In this study, we generated a comprehensive seed expression atlas covering six developmental stages in pea (cv Cameor), including the key transition stage from embryogenesis to early seed filling, providing a detailed temporal resolution of transcriptional dynamics throughout seed development in this species. Co-expression network analysis highlighted several candidate transcription factors potentially involved in the transition towards seed filling. Among them, we characterized the seed-specific NF-YB transcription factor PsLEC1-like (PsL1L), the major LEC1-type factor expressed during early pea seed development. Functional analyses using TILLING mutants demonstrated that loss of PsL1L function reduces seed size and seed nitrogen content and impairs early embryo growth from the end of embryogenesis. Finally, we show that the expression of the B3-domain transcription factor PsFUS3, but not that of PsLEC2 or PsABI3, is reduced in the loss-of-function l1l mutant, suggesting that PsL1L acts upstream of PsFUS3 to control seed size.