The scent released by a flower does not necessarily reach a bee unchanged. Because compounds differ in how readily they react in the air, reactions with ozone and other substances alter the scent blend. How important are those changes to a bee’s recognition of a floral odor it has learned?
A study published by Sprayberry and colleagues in 2026 examined a model that quantifies molecular features of scents and relates them to western honeybee behavior. It analyzed behavioral data from three earlier studies and predicted what would happen as the floral scents of four crops changed. It was not a new field study measuring bees or crop yields.
Looking at shared features rather than molecular names
Floral scent is a mixture of multiple volatile compounds. The effect of losing a compound is not determined simply by the number of compounds lost. Whether it shares features with other compounds or contributes distinctive features to the mixture may influence how the odor is perceived.
The method used in the study, called CWB, sums compounds’ relative amounts for structural features such as carbon-chain length, the number of carbons in rings, and groups such as alcohols and esters. It thus represents a mixture as a combination of features rather than a list of molecular names.
That combination is represented by a mathematical arrow, and the difference between two odors is calculated as the angle between their arrows. A larger angle means a greater difference as captured by this method. There is no literal angle in a bee’s brain, nor is this value a direct reading of neural activity.
Will a bee extend its mouthparts to a learned odor?
In the three earlier studies used for the analysis, bees learned to associate an odor before pollution-induced alteration with a sugar reward. They were then exposed to mixtures recreating pollution-induced changes, and the frequency of proboscis extension was measured. This response served as an indicator of how far recognition of the learned odor was retained.
With the original CWB calculation, responses declined as the angle increased within each study. Across studies, however, the angles at which responses declined did not agree. Experiments removing compounds entirely differed from experiments changing their proportions.
The authors therefore modified the method to give a structural feature ten times the weight when it disappeared completely or was newly added. This reflects the idea that losing a feature may affect neural processing more strongly than simply reducing its amount. The tenfold weighting is not a measured amplification factor in an actual neural circuit.
After the modification, pooling the three studies showed response rates falling to around 50% in the 10–15-degree range and below 50% for data above 15 degrees. This is a guide derived from these data and this calculation method, not an established universal perceptual boundary for all bees and all odors.
Four crops do not change in the same way
Next, the researchers compared white mustard, strawberry, apple, and oilseed rape using previously reported floral scent compositions. The strawberry data were for the cultivar Malwina and the apple data for Golden Delicious.
To represent daytime reactions, they held the concentration of highly reactive hydroxyl radicals constant and varied ozone from 10 to 200 ppb. Reaction rates for individual compounds came from the literature, with missing values estimated from molecular structure.
In the calculations, oilseed rape scent changed most strongly and rapidly. At ozone concentrations above 140 ppb, for example, its modeled angle reached at least 15 degrees within about three minutes. By contrast, the apple scent composition used here was predicted to change very little in angle over 60 minutes.
These are calculations for particular scent compositions and reaction conditions. They do not establish that bees lose track of flowers after three minutes in a field or that apples are unaffected by air pollution. The 60-minute calculation explored an upper bound under sustained exposure; the authors particularly emphasized the first 15 minutes as a period relevant to odor-guided behavior outdoors.
Losing more compounds does not necessarily make recognition harder
An important result was the comparison between strawberry and white mustard. Looking only at molecular identities and proportions, strawberry changed more. In the feature-based model, however, white mustard was predicted to reach a greater impact on recognition sooner.
The reason is the overlap in features among compounds. During the initial modeled changes, losing three strawberry compounds removed two features, whereas losing two white mustard compounds removed six. How the entire mixture changes depends on whether other compounds continue to provide the same features.
The authors argue that assessment should consider not only chemical changes but also the sensory system of the organism receiving the signal. However, they did not directly test this ranking of the four crops in bee behavioral experiments.
Assumptions also shape the predictions
The model makes an important simplification. As a compound declines through reaction, it is treated as absent after its atmospheric lifetime has elapsed. In reality, about 37% remains at that point. Because the model places substantial weight on complete feature loss, this assumption affects its predictions.
Conversely, dilution of odors by wind and turbulence is not included. The authors explain that compounds remaining after chemical reactions would also be diluted outdoors, but how the two effects balance in practice needs verification under specific conditions.
Further measurements are needed to predict outcomes that include scent variation among cultivars and environments, bees’ use of multiple foraging cues, and real airflow. This model provides a starting point for deciding which plants and conditions deserve priority in further study.
Summary
When floral scent changes in the air, counting the compounds lost may not fully capture the effects on bees. This study sought to predict differences that chemical analysis alone may miss by considering overlap in molecular features and the receiver’s senses.
The authors’ aim is to use models to identify vulnerable plants and conditions, test those predictions experimentally, and refine the models. Repeating that process could provide a basis for anticipating effects on pollination across landscapes and considering ways to protect flower–insect relationships from air pollution.
References
Sprayberry, J. D. H., Girling, R. D., Ryalls, J. M. W., Blande, J. D., & Langford, B. (2026). A perceptual model indicates air pollution-induced shifts in honeybee floral-scent recognition. Communications Earth & Environment, 7, 357. DOI: 10.1038/s43247-026-03351-z / Full text. Original article: © The Author(s) 2026, CC BY-NC-ND 4.0.
