Unlocking plant metabolic resilience: how enzyme-constrained metabolic models illuminate thermal responses

While glasshouse-grown plants benefit from controlled environments, the majority of plants in the world are directly exposed to continuous changes in ambient temperatures. In 2024, the global surface temperature was 1.28°C above the 1951–1980 average (Bardan, 2025), eclipsing the previous record set in 2023 (Esper et al., 2024). The Intergovernmental Panel on Climate Change reports the likelihood of a consistent trajectory of rising temperature, although projections vary in magnitude under different scenarios (Masson-Delmotte et al., 2021). Rising temperatures along with extreme weather events will significantly challenge the survival of wild plant populations and global agricultural stability. Understanding plant metabolic responses to temperature changes at a metabolic level is critical for engineering climate-resilient plants. Recently published in New Phytologist, Wendering et al. (2025; doi: 10.1111/nph.20420) present the first enzyme-constrained, genome-scale metabolic model of Arabidopsis thaliana. By integrating temperature-dependent constraints on enzyme kinetics, protein content, and photosynthetic capacity, this model not only advances our understanding of how plant metabolism responds to thermal stress at a systemic level but also provides a valuable framework for identifying metabolic and genetic targets to enhance temperature resistance, which could apply to crops. Furthermore, such insights may also help to preserve wild plant species facing climate-driven extinction risks (Nievola et al., 2017).

By integrating thermal proteomics experimental data with a machine learning algorithm, this hybrid parameter estimation strategy resolves a critical limitation in the development of large-scale models…

Plants have a limited capacity to regulate their canopy temperature (Guo et al., 2023), which means that all internal metabolic reactions are influenced by external temperature fluctuations. Research on signaling transduction, epigenetic regulation, transcriptional networks, and post-translational regulation of heat and cold stress has gained significant attention (Ohama et al., 2017; Ding & Yang, 2022). However, responses to temperature changes in plants are initially observed at the metabolic level, with subsequent changes in gene expression to restore homeostasis (Casal & Balasubramanian, 2019).

Predicting how temperature fluctuations affect overall plant metabolism remains a challenge because the direct temperature effects on individual metabolic enzymes are often not well defined quantitatively. To overcome this challenge, the authors have developed the ecAraCore model, which is an enzyme-constrained extension of the AraCore model (Arnold & Nikoloski, 2014). The AraCore model is a widely used tool for predicting primary metabolism in Arabidopsis that uses flux balance analysis (FBA) to predict metabolic fluxes under steady-state conditions. However, the traditional FBA model ignores the limitations imposed by enzyme availability and catalytic efficiency.

The new model integrated enzyme concentrations and turnover numbers (k_cat) into the metabolic network. This integration ensures that reaction fluxes are limited by both stoichiometric and enzymatic constraints. Furthermore, by incorporating this with temperature-dependent adjustments into k_cat and the total protein content, the model enables accurate predictions of metabolic efficiency, growth, and resource allocation across various temperatures.

The model uses thermal proteome profiling (TPP) data to infer the protein thermostability parameter (optimal temperature, T_opt). Given the limited coverage of TPP data, the authors trained a Random Forest model on amino acid sequence features to predict T_opt for Arabidopsis proteins. By integrating thermal proteomics experimental data with a machine learning algorithm, this hybrid parameter estimation strategy resolves a critical limitation in the development of large-scale models and sets a precedent for future studies in other plant species.

The model predicts net CO₂ assimilation rates (A) and relative growth rates (RGR) across a biologically relevant temperature range (10–40°C) by coupling the Farquhar–von Caemmerer–Berry photosynthesis model (Farquhar et al., 1980) with genome-scale metabolism. Strong correlations between predicted and observed RGR and A data confirm the model's accuracy. Through constraint-based simulations, the authors identify metabolites and proteins that limit growth at specific temperatures. For example, at high temperatures (35–40°C), amino acids such as l-arginine and l-glutamine are critical, while Rubisco activase (RCA) and cytochrome b₆f subunits limit growth at both low (10°C) and high (40°C) extremes. These findings align with previous studies on the thermolability of RCA in crops such as tobacco, spinach, and sweet potato, validating the model's biological relevance.

A significant strength of this study is the experimental validation of model predictions. The authors tested Arabidopsis T-DNA insertion lines for genes predicted to affect growth at 17°C. Two out of three knockouts (e.g. genes linked to pyruvate metabolism) exhibited significant reductions in dry weight, while 10 out of 11 negative controls showed no phenotype. This validation highlights the model's potential for prioritizing candidate genes for breeding or gene editing.

The methodology developed in this study has the potential to be applied to agricultural crops by incorporating crop-specific k_cat values and TPP data. This approach could uncover both conserved and species-specific thermal stability patterns within the metabolic systems of crops. Over the past decade, many genome-scale metabolic models similar to the AraCore model have been established for major crops, such as rice, soybean, maize, and tomato. Additionally, tools for reconstructing, simulating, and analyzing enzyme-constrained metabolic models have advanced significantly. For example, the GECKO Toolbox 3.0 (Chen et al., 2024) that integrates deep learning for k_cat estimation and enhances integration with proteomics and metabolomics, significantly improves the efficiency of model development. However, accurate prediction of enzyme activity responses to temperature fluctuations critically depends on species-specific TPP data combined with the predictive methods described in this study. This accuracy is crucial for the precision of the final model predictions. Another key challenge lies in the accuracy of estimations for k_cat, which remains a major source of uncertainty, largely due to the limited availability of experimental data. While recent advancements in deep learning have made progress in addressing these gaps (Li et al., 2022; Kroll et al., 2023), there is still considerable potential for improving their precision.

The model relies on literature-derived total protein content and operates under the assumption of a single objective function. However, both protein content and metabolic objectives, such as stress-induced shifts from growth to defense, may vary with genotype and environmental conditions. Incorporating tissue-specific proteomics and dynamic biomass composition data would refine the model predictions.

While the model effectively captures the direct response of metabolic enzymes to temperature, it does not consider the role of temperature signal transduction or the activation of temperature-responsive genes, which are also critical for plant tolerance to temperature stress (Ding & Yang, 2022). Specialized proteins such as heat-responsive transcription factors, like HsfA1, help stabilize proteins and cellular structures under thermal stress. Additionally, hormones such as abscisic acid and signaling molecules such as reactive oxygen species regulate key temperature responses, including stomatal closure during heat stress. Temperature-induced gene expression also involves epigenetic modifications, such as histone methylation, which modulate chromatin structure and influence the transcription of stress-responsive genes (He et al., 2021). Integrating these multilevel regulatory networks with metabolic models to simulate plant responses to temperature, akin to the concept of digital twins, represents an exciting direction for future research.

This study exemplifies the power of systems biology in addressing the challenges agriculture faces due to a changing climate by linking enzyme thermodynamics to whole plant phenotypes. The authors also provide a valuable framework for studying other abiotic stresses, such as drought. The integration of machine learning with metabolic modeling highlights the potential of artificial intelligence-driven tools to accelerate crop improvement. When coupled with quantitative metabolomics to characterize temperature-dependent changes in key biomass components, this model could be expanded to optimize resource allocation across diverse environmental conditions.