Pheromones, nectar, and computer vision: investigating large-scale patterns of chemodiversity.
Loading...
Authors
Grele, Ari
Issue Date
2025
Type
Dissertation
Language
en_US
Keywords
Biodiversity , Chemical ecology , Computer vision , Nectar chemistry , Pollinator ecology , Semiochemistry
Alternative Title
Abstract
At its core, chemical ecology attempts to describe the complex forces that generate the vast chemical diversity observed in nature and which allow natural products to mediate interactions across biological scales. Chemodiversity is a major component of functional biodiversity in all ecosystems, and in plant – insect interactions especially mediates mutualisms, antagonisms, defense, reproduction, and many more fundamentally important ecosystem processes. Working with large datasets across multiple insect systems, this dissertation provides insight into major drivers of chemical diversity across taxa, ecological functions of specialized chemistry in plant – pollinator interactions, and methods of automating data collection for monitoring insect diversity. In particular, I investigate the role of specialized nectar chemistry under several existing hypotheses, generate and test novel hypotheses regarding global patterns of insect semiochemical diversity, and discuss methods for rapidly and scalably monitoring insect biodiversity using machine learning. Chapter 1 builds a framework for understanding large scale patterns of chemodiversity. Here I apply ideas from information theory to semiochemical communication, especially concepts surrounding communication in noisy channels, to predict variation in semiochemical diversity at global scales. While an understanding of chemical noise allows the interpretation of existing hypotheses in the chemical ecology of plant and animal systems, it also suggests several patterns which have yet to be observed. Testing two predictions made under this framework, that insect volatile semiochemical blends should be less rich in biodiverse regions and contact semiochemicals more rich, using a large scale meta-analysis approach, I show that insect semiochemistry broadly follows the patterns expected under information theory. These results suggest that large scale patterns of chemodiversity may arise due to selection imposed by chemical noise, and that information theory can allow interpretation of these patterns across systems.
Chapter 2 investigates a specific aspect of chemical communication: the role of specialized nectar chemistry in mediating plant – pollinator interactions. The Nectar Pleiotropy hypothesis suggests that the presence of specialized metabolites in nectar is non-adaptive, and that these compounds are maintained in the nectar solely through selective pressures for antiherbivore chemistry in leaves and floral tissues. Using common garden experiments with milkweed flowers in Idaho and Arizona, I found that nectar chemistry is consistently distinct from leaf chemistry within and across species, with large numbers of nectar compounds not found in leaves, and the concentrations of the majority of shared compounds uncorrelated between leaves and nectar. These findings demonstrate that the Nectar Pleiotropy hypothesis does not apply to most compounds in the milkweed nectar metabolome, suggesting that nectar chemistry is maintained separately from leaf chemistry to mediate interactions with nectar feeding organisms.
Chapter 3 builds on the work of chapter two by directly linking milkweed nectar chemistry to pollinator visitation, behavior, and plant fitness. Using gardens of milkweeds in Nevada and California, I show that compounds which strongly associate with insect feeding are also strongly associated with average pollinator efficiency, and that the concentration of these compounds in nectar indirectly alters plant fitness through the manipulation of high efficiency pollinators, but not nectar larcenists. These results indicate that nectar chemistry in these species acts primarily to manipulate high quality pollinating insects, with little impact on non-pollinating insects.
Finally, chapter 4 moves from interrogations of chemical diversity in plants and insects to discuss methods of monitoring insect biodiversity as a whole. Here I developed a computer vision system and data pipeline for rapidly gathering training data and deploying automated systems for insect monitoring. By incorporating hierarchical data into computer vision models, I demonstrate that model performance can be greatly improved when working in systems where the majority of taxa are unknown to science. These systems, while only capable of measuring a specific dimension of insect diversity, pave the way to more generalized and scalable models for automated insect monitoring.