Augmentation of the Zebrafish Phenomic Landscape by Histological Analysis and Content-Based Image Retrieval

Open Access
Canada, Brian Ashley
Graduate Program:
Integrative Biosciences
Doctor of Philosophy
Document Type:
Date of Defense:
October 04, 2010
Committee Members:
  • James Z Wang, Dissertation Advisor
  • James Z Wang, Committee Chair
  • Keith C Cheng, Committee Member
  • Kenneth Monrad Weiss, Committee Member
  • Joan Therese Richtsmeier, Committee Member
  • Webb Colby Miller, Committee Member
  • content-based image retrieval
  • zebrafish
  • histological analysis
  • phenotyping
  • phenomics
The study of systems biology involves the computational modeling and analysis of complex biological systems. Whether the system of interest is a cell, an organ system, or a complete organism, creating a working model of any given biological process requires a comprehensive understanding of the functions of genes, interactions between gene products, and interactions between the system and its surrounding environment. The emerging field of <i>phenomics</i> addresses the problem of systematically analyzing physiological, behavioral, and morphological phenotype data from genetic manipulation experiments on a genome-wide scale. For instance, the morphological analysis of zebrafish mutants provides clues to the function of the affected genes and represents a key aspect of the proposed Zebrafish Phenome Project. While gross phenotypic analysis, generally performed by stereomicroscopy, can be used to characterize morphological phenotypes at the organism or organ level, it is necessary to use higher-resolution imaging methods, such as histology, to detect phenotypes that are only evident at the cellular level. However, in spite of the higher sensitivity relative to gross analysis, histology has been undeservedly perceived as being too costly in terms of time and effort. Consequently, most of the phenotypic annotations in the current zebrafish literature are limited by the relatively low descriptive power and resolution of the cheaper and faster phenotyping by stereomicroscopic analysis. Overcoming the perception that histology adds little value beyond gross analysis requires a clear and significant demonstration of the power of histology to augment the zebrafish phenomic landscape and also that histological analysis has the potential to become a fully-automated, high-throughput phenotyping method. In addressing these issues, this thesis makes three important contributions to the field of phenomics. First, we present the results of a pilot study that showcases the sensitivity of histology in detecting subtle multiple-organ morphological phenotypes in ten larval zebrafish mutants previously annotated with gross phenotypes that were only observed in one or two organs at the same developmental stage. While gross analysis by stereomicroscopic inspection detected an average of 1.6 affected organs per mutant, a histological characterization of these same mutants (based on 75 semi-quantitative phenotypic measurements) detected an average of 7.4 affected organs per mutant—nearly a four-fold increase in sensitivity to detect phenotypes at the organ level. In addition, a deeper evaluation of the histology of lens and seven retinal cell layers of the eyes resulted in an average of 7.2 defective eye structures per mutant. These results suggest that histology's ability to detect subtle defects in morphology can contribute significantly to insights into vertebrate gene function. The second contribution of this thesis expands upon the ten-mutant pilot study by providing a comprehensive, systematic analysis of multiple-organ histological phenotypes in 102 larval zebrafish mutants. This dataset enabled us to begin using high-dimensional phenotypes to explore the functional interrelatedness of these mutants by graph-theoretic approaches. A subset of 77 of these mutants with known homology to human genes was used to construct a series of biological networks based on phenotypic similarity. We find that, as a <i>group</i>, zebrafish mutants belonging to maximally-connected subgraphs (or “cliques”), with high phenotypic similarity to each other, appear to have shorter interaction path lengths than mutants with lower phenotypic similarity. However, we also find that the size, number, and membership of cliques varies considerably with the dimensionality and resolution of the phenotype dataset, suggesting that the reliability of functional associations between genes based on organ-level morphological phenotype similarity alone can be augmented by increasing the precision of phenotypic analysis, thus maximizing phenomic resolution. Finally, this thesis details the development and implementation of a novel content-based image retrieval (CBIR) system, called SHIRAZ (System of Histological Image Retrieval and Annotation for Zoomorphology). Using our robust set of semi-quantitative histological phenotype measurements as a training dataset, the current version of SHIRAZ is designed to be capable of automatically annotating images depicting histological abnormalities in the developing eye of the larval zebrafish, with the intention of providing a proof-of-concept demonstration of the potential for a fully automated high-throughput histological analysis laboratory workflow.