MICRO AND MACRO VIEWS OF THE MAIZE-SETOSPHAERIA TURCICA PATHOSYSTEM
No Access Until
Permanent Link(s)
Collections
Other Titles
Author(s)
Abstract
Interactions between host and pathogen can be understood at many different spatial scales, from nanometers to kilometers. In this dissertation, I explored the economically important disease Northern Leaf Blight and the components of the pathosystem- the host, maize, and the fungal pathogen, Setosphaeria turcica- at two very different spatial scales. At the “micro” scale, I used RNA-seq to explore the transcriptomic aspects of infection, with a focus on the pathogen’s transition from biotrophy to necrotrophy and the impacts of pathogen virulence/avirulence in the presence of the host Ht2 R gene. Gene expression in both host and pathogen shifted dramatically between biotrophy and necrotrophy, with specific trends demonstrating the different molecular mechanisms of infection and host defense during each phase. Pathogen avirulence, due to R-gene mediated resistance, led to an apparent arrest of the pathogen in the biotrophic phase. The importance of gene-sparse regions of the S. turcica genome for pathogenesis was shown for the first time. Expression of NLB-induced genes in diverse non-inoculated maize lines could be used to predict their NLB phenotypes, suggesting a potential role for baseline expression of these genes in resistance. At the “macro” scale, I combined crowdsourcing and machine learning to develop a new method for aerial detection of disease symptoms in the field. The task of annotating thousands of disease lesions in order to train a machine learning model was split in two, with experts annotating lesions in low resolution and numerous non-experts performing the more time-consuming task of outlining lesions, using the expert annotations as a base. This method allowed us to generate a large amount of reliable training data very quickly and at low cost. These data were used to train a convolutional neural network (CNN) to high accuracy, and a fully-connected conditional random field (CRF) was used to segment images into lesion and non-lesion areas using the CNN output. The final model was able to delineate lesions in aerial images down to the millimeter level, a finer spatial scale than any previously reported method. It also outperformed human experts by identifying lesions that they had missed. Though the techniques, findings, and impacts involved in work at these two very different scales are accordingly varied, they all contribute to a holistic understanding of the pathosystem and our ability to make practical change.
Journal / Series
Volume & Issue
Description
Sponsorship
Date Issued
Publisher
Keywords
Location
Effective Date
Expiration Date
Sector
Employer
Union
Union Local
NAICS
Number of Workers
Committee Chair
Committee Co-Chair
Committee Member
Gore, Michael