The Edited Field

AEvery few decades an agricultural technology arrives with the claim that it will end an old compromise. The one now making that claim is gene editing, and in particular the CRISPR systems that let a breeder cut a chosen sequence and leave the cell's own repair machinery to disable it. The promise is usually explained by contrast with the transgenic crops of the 1990s, which worked by inserting genes taken from other organisms and which needed, on average, twelve years and a dossier running to some thirty volumes before a variety could be sown outside a glasshouse. Editing leaves no foreign DNA behind, and the earliest edited lines reached field trial about three years after the first cut was designed. Whether that saving of time and paperwork has produced crops that actually do more in a field, or merely crops that arrive sooner, is the question the past decade of trials was meant to answer.

BLena Vosberg, a plant geneticist at the Halstrom Institute for Crop Science, has spent that decade counting. Between 2016 and 2022 her group grew edited bread wheat in 118 replicated plots across four sites, each plant carrying a disabled copy of a single susceptibility gene that mildew is known to exploit. Where the fungus was allowed to establish, the edited plots out-yielded their unedited controls by 22 per cent, a figure large enough to be reported on its own in the farming press. What that figure omits is the other half of the design. In the plots kept clear of mildew by fungicide, the same edited lines yielded about 5 per cent less than the controls, a penalty her group traces to the everyday work the disabled gene does when no fungus is present. The gain, in short, was confined to the seasons in which the crop came under attack.

CThe account behind those numbers makes a prediction, and Vosberg has tested it. Editing of this kind removes; it is at its most powerful when a desirable trait is simply what remains once one gene has been switched off, and the same reasoning should hold in reverse wherever a trait is governed by many genes of small effect rather than by one. Drought tolerance is the standard example. In trials run at the same four sites, edited lines aimed at drought tolerance improved measurably in 9 per cent of attempts, against roughly 70 per cent for the single-gene disease traits. The asymmetry matters, because it is the many-gene traits that conventional crossing and selection have always handled best. Vosberg is nonetheless careful to present the figure as a boundary for one method rather than a verdict on the trait.

DPia Ovchinnikov, an agronomist at the Rioseco Field Station, rejects the inference, on the ground that a line which fails on a research station is telling you about the station at least as much as about the gene. His own edited sorghum lines, all of them aimed at heat tolerance, a trait governed, as he accepts, by many genes at once, were grown by farmers on their own land, with their own spacing, weeding and manure, across two seasons; 57 per cent of them out-performed the local variety the farmer had planted the year before. A trial which strips a plant of the management its growers actually practise, Ovchinnikov argues, is measuring the station rather than the seed. He concedes two things, however. His work covered eleven lines of a single crop, and adoption was estimated indirectly, from the rate at which households reordered seed, because no dependable record of on-farm planting exists in the district.

EThe dispute has not stayed inside the field station. A court ruling in 2018 placed edited crops under the same authorisation regime as transgenic ones in several jurisdictions, a decision that turned on the wording of a directive drafted long before the technique existed, and a partial relaxation in 2023 has left breeders licensing one tool under two sets of rules depending on the border. Vosberg points out, uncomfortably for her own position, that nothing in her yield analysis bears on whether the public will accept the crops, since precision at the cut site and public trust are not alternatives. Ovchinnikov replies that the two cannot be held apart in practice, because legislators read yield figures as evidence about safety, and read a poor harvest as evidence against it. Both agree that the habit of publishing single striking successes, rather than counts of what lines do across ordinary seasons, has served the subject badly.

FA settlement of sorts is emerging in which the two accounts are given different work to do rather than ranked against each other. Where a trait is the visible consequence of one gene being switched off, editing delivers it in a few seasons instead of a decade, which explains both the speed of the wheat result and the modesty of its scope; where a trait is assembled from many small contributions, the older business of crossing, selecting and waiting still does the assembling, and editing at best adjusts what that business hands over. The unresolved question is whether the traits that fail on research stations are genuinely beyond reach on farms, and no trial design now in use can settle it, since the two settings differ in far more than management. What both researchers reject is the assumption that produced the original promise, namely that a new tool must either replace breeding or be a faster copy of it.