Date of Award

2025-12-01

Degree Name

Doctor of Philosophy

Department

Ecology and Evolutionary Biology

Advisor(s)

Philip Lavretsky

Abstract

Understanding the genetic underpinning and distribution of phenotypic variation within and between divergent groups is core towards shedding light into how populations diverge and adapt, as well as how hybridization breaks or builds on these scenarios; and thus, central to evolutionary biology. In wild organisms, however, quantifying and linking phenotypic traits to underlying genetic processes, like mutation, gene expression, epigenetics and allele interactions, remains challenging. This difficulty arises from the complex interplay between morphology, environment, and gene regulation, as well as the logistical barriers of collecting and standardizing large-scale data across individuals and populations. As a result, researchers are increasingly turning toward interdisciplinary frameworks that integrate traditional morphometric analyses, artificial intelligence (AI)-based phenotyping, and genomic approaches to illuminate the biological mechanisms that generate and maintain variation in natural systems.

Species comprising the Mallard Complex provide an exceptional model for integrative work linking artificial intelligence and genetics, as they share much of their genomes due to incomplete lineage sorting with the exception of those genic regions that are responsible for each species' morphological and ecological differences. Importantly, the same group of organisms share ancestry due to frequent gene flow when in contact. For example, the New Zealand Grey Duck (Anas superciliosa superciliosa) and the introduced Mallard (Anas platyrhynchos) are genetically and morphologically distinct, but their contact has resulted in hybrids that span the genomic and morphological spectrum of both parental taxa. Moreover, the interplay between these species leads to both conservation and evolutionary questions surrounding the persistence of native lineages and dynamics of morphological introgression. Similarly, species of Mallard-like ducks found in North America diverged from the Mallard only in the last half-million years, resulting in high rates of retained genetic and phenotypic ancestry. Among these, the Mexican Duck (Anas diazi) hybridizes with the Mallard in regions of sympatry, producing a gradient of phenotypic intermediates that blur species boundaries. This overlap has practical implications as both species are currently managed as a single group due to the difficulty of field identification without expert input. Methods that standardize and automate trait recognition are thus required to enable wildlife agencies to distinguish between these species more reliably, as to support finer-scale management and conservation policies that better reflect their biological and ecological differences. Together, these parallel systems underscore the need for integrative tools that bridge field observation, phenotypic analysis, and genetic insight to clarify patterns of divergence and hybridization in wild waterfowl. The overarching goal of this dissertation is to develop and apply integrated morphological, computational and genomic approaches to better understand phenotypic variation and its genetic underpinnings in wild waterfowl. In Chapter 1, I test whether morphological traits can reliably distinguish hybrid and parental individuals in the New Zealand duck system, while also bringing forth a case of hybrid vigor within their system. Specifically, I examine morphological differentiation between New Zealand Grey Ducks, Mallards and their hybrids using traditional morphometric approaches. Doing so, I establish a phenotypic baseline of variation between these closely related, recently diverged species, while evaluating which traits most effectively discriminate between groups. I then tie in morphological data with individual fecundity and demonstrate that New Zealand Mallards and hybrids have a higher nesting success compared to that of wild Mallards of North America, thereby showing greater robustness in viability rates compared to one of their parental species. Thus, I conclude that introduced Mallards and their hybrids may be a case of hybrid vigor.

Next, I move towards developing methods to facilitate reliable species identification through a novel AI-based computer vision model based on images of genetically-vetted individuals in Chapter 2. In short, I introduce a deep learning approach for automating the identification of plumage traits used for species identification. I deploy methods on the Mexican Duck and Mallard system for which previous field keys were developed. In doing so, I assess the potential of AI to standardize phenotypic data collection and reduce observer bias, while importantly stressing interpretability of model results, and thereby addressing a major limitation in large-scale ecological and evolutionary studies. Although, methods were not as accurate as earlier field keys, this study provides a foundational backbone towards better integration of wild systems that often suffer from image quality and quantity.

Finally, I deploy the AI methods developed in Chapter 2 that provide binary codes for phenotypic traits to establish whether AI-assisted genome-wide association studies are possible in Chapter 3. Coupling binary codes across traits with genomic data associated with images used for model development, I attempt to link phenotypic variation with specific genomic regions and provide insight into the molecular mechanisms shaping complex trait diversity. Specifically, I apply genome-wide association studies (GWAS) to both expert- and AI-derived phenotypic datasets to investigate the genetic architecture underlying plumage coloration and patterning. This analysis explores how genomic variation contributes to visible trait difference and tests the feasibility of integrating AI-based trait data into population genomic analyses. These efforts identified several key genetic regions associated with plumage trait expressions when analyzing expert-derived trait datasets. Despite the AI-derived trait datasets not converging with the expert-led dataset due to poor model performance, I demonstrate the potential of such efforts towards building my capacity to study complex trait evolution in wild systems.

Together, my dissertation attempts to understand how morphological and phenotypic traits evolved in closely related species, while assessing what happens to these traits during hybridization. Importantly, my efforts demonstrated the utility of coupling phenotypic and genomic data within a Machine Learning environment and novel AI models when attempting to maximizes data collected, including promoting reproducibility, scalability, and insights of wild systems, and those mechanisms driving biodiversity.

Language

en

Provenance

Received from ProQuest

File Size

176 p.

File Format

application/pdf

Rights Holder

Sara Gonzalez

Download

Share

COinS