AlphaFold DeepMind: engineering heat-tolerant Rubisco for crops

DeepMind used AlphaFold to design mutations in Rubisco, the central photosynthesis enzyme that fixes atmospheric carbon, aiming to boost heat tolerance in crop plants. Laboratory collaborators tested several predicted variants and observed improved thermal stability in biochemical assays, a step toward crops that retain photosynthetic activity under higher temperatures.

What DeepMind did

Rubisco is slow and temperature sensitive, and its inefficiency contributes to yield losses when heat stresses crops. DeepMind applied AlphaFold, its protein structure prediction model, to generate structural models of Rubisco variants and to evaluate the likely impact of specific amino acid substitutions on folding and subunit interfaces. The team then prioritized designs predicted to increase thermostability while preserving the enzyme's catalytic geometry.

Design work combined structural predictions with sequence analysis to identify candidate mutations in flexible loops and contact regions between Rubisco subunits. Selected variants were synthesized and expressed for in vitro tests. Biochemical assays measured thermal stability and retained activity after exposure to elevated temperatures, providing initial validation of the computational designs.

The effort paired computational design with standard laboratory pipelines rather than presenting field-level modifications. DeepMind framed the work as a demonstration of how high‑accuracy structure prediction can narrow the space of mutations to test, reducing experimental burden compared with unguided screening.

Results and validation

DeepMind and partners reported that a subset of designed variants showed measurable gains in thermal stability in purified enzyme assays. Structural models suggested that stabilizing substitutions clustered at intersubunit interfaces and in regions that influence local flexibility, consistent with expected mechanisms for improving thermostability.

Activity assays indicated that the more stable variants retained catalytic function after heat exposure that reduced activity for the wild-type enzyme. The teams emphasized that laboratory assays are an early step, and that translating enzyme improvements into whole-plant performance requires further work: expressing engineered Rubisco in plant cells, checking assembly and regulation in plastids, and assessing growth and yield under heat stress.

The study illustrates an iterative pipeline: predict structures with AlphaFold, propose targeted substitutions to alter stability, test candidates biochemically, then feed results back to refine designs. DeepMind described the approach as a way to accelerate protein engineering for agricultural traits, while noting that downstream engineering and regulatory steps remain substantial.

Why it matters

Using structure-prediction models like AlphaFold to guide Rubisco redesign narrows experimental search space and may speed efforts to produce crops that tolerate higher temperatures. If computationally guided variants can be deployed in plants without harming assembly or regulation, the approach could reduce heat-related yield losses. The work also highlights that translation from enzyme assay to field-ready crops will require additional genetic, physiological, and regulatory work.