The concealed geometry behind breeding constraints

Breeding crops for thousands of years has narrowed genetic diversity into tight corridors, delivering steady gains—yet today progress is slowing. Heritability-based expectations often suggest more hidden variance should exist, but breeding programmes struggle to uncover it. In a new Perspective, Ortiz-Barrientos, Jordan and Cooper argue that the gap is not simply about insufficient data or weak heritability signals. Instead, they propose that practical breeding constraints force genetic exploration to occur on curved, lower-dimensional surfaces inside the much larger space of possible genotypes.

The core idea is that crops are not evolved in isolation: they are continually filtered through management realities. Fertilizer regimes, sowing density, weed and pest control, irrigation schedules and harvest logistics jointly define the environments where particular genotypes are routinely viable. These environments act like “feasibility boundaries” that restrict which combinations can be tested at scale.

Within this restricted space, gene interactions may be hidden. Even if underlying biology is fundamentally epistatic—dependent on combinations of alleles—constraint-driven filtering can make fitness effects look additive. Breeding then appears to work through incremental, single-step improvements, because selection responses measured within the constrained environment align with additive models. The authors emphasize that this appearance can mask a different reality: the true adaptive trajectory is shaped by the geometry of constraints.

Selection thus proceeds as additive steps on a curved surface. Short-term movement can seem linear and fast, but the long-term path follows the contour dictated by what the agricultural system allows. The result is a slowing “plateau” effect: the programme runs out of accessible directions that remain productive under those constraints.

This pattern is expected to be strongest in major crops subjected to intensive, directional selection for stable agricultural performance across many generations. In settings where selection is newer or constraints are weaker, epistatic structure may be less masked, and apparent plateaus may be less pronounced.

The Perspective frames constraint-based filtering as a potential mechanism for why breeding has not recovered the amount of genetic variance implied by classic estimates. If epistatic combinations drive fitness, then boundaries imposed by breeding and agronomic practice could be excluding exactly the genetic interactions that would unlock further gains.

Finally, the authors outline potential “escape routes” from these stagnating dynamics: strategic wide crosses to reintroduce distant variation, combining transgenes that were previously tested separately, and targeted genome edits designed to access variants currently unreachable within the constrained genetic manifold.