Nicole L. Washington; Melissa A. Haendel; Christopher J. Mungall; Michael Ashburner; Monte Westerfield; Suzanna E. Lewis

“ We describe here a novel method using ontologies to record and quantify the similarity between phenotypes. We tested our method by using the annotated mutant phenotype of one member of the Hedgehog signaling pathway in zebrafish to identify other pathway members with similar recorded phenotypes. We also compared human disease phenotypes to those produced by mutation in model organisms, and show that orthologous and biologically relevant genes can be identified by this method. ”

“ Our final hypothesis was that sequence orthologs would exhibit similar mutant phenotypes and therefore phenotype descriptions alone would be sufficient to identify orthologs and pathway members. To test this, we queried the complete set of zebrafish and mouse phenotypes, using the phenotypic profiles of the 11 human disease genes annotated from OMIM and our four scoring metrics. Table 6 shows the score and rank of the mouse and zebrafish orthologs when compared to the human disease gene for all four metrics. ”

“ Model organism communities benefit from centralized collections of curated research, where a scientist can search for extensively cross-referenced gene expression, phenotype, and genomic data, referred to as “model organism databases” (MODs) . Research in the field of human biology suffers because there is no equivalent resource for the human research community, and linking these diverse datasets requires searching many detached resources. ”

“ In the future, we aim to create a “gold standard” set of genotype-phenotype annotations that would minimize literature or experimental bias and is independently annotated by different curators to eliminate errors of commission and omission. This would allow statistical testing of sensitivity and specificity with regard to these similarity metrics. Nevertheless, our results demonstrate conclusively that one can compare phenotypes across organisms using ontology-based metrics to find biologically meaningful results. ”

“ Some key players in the zebrafish shha search were not included in our analysis because they were based on morpholino knockdowns rather than traditional mutants. Similarly, morpholino phenotype data from five of the 11 orthologs of the human disease genes examined were also not included in the ortholog analysis (shha, sox9a, sox10, tnnt2, and ttna) . Future enhancements to our database structure will accommodate various mechanisms for diminishing gene function such as gene-specific morpholinos, siRNAs, or chemicals, and this will greatly expand the available dataset for comparison. ”

“ We expect that, as more data are accumulated using the EQ methodology, annotation to generic nodes will comprise an increasingly smaller percentage of the total annotations.
The genes identified by this similarity algorithm are good candidates for further investigation of biological function, pathway elucidation, and identification as animal models of disease. Although some models of disease may already be in existence, the importance of having a variety of animal models for the same disease should not be underestimated. ”

“ For example, Figure 1 shows phenotypic profiles for eye phenotypes of PAX6 ortholog mutations in mouse, human, zebrafish, and fruitfly (also see Table 2) . An important thing to note is that any given individual organism presenting a phenotype may manifest only a subset of the EQ descriptions of a complete phenotypic profile for a particular genotype. The PAX6 and ortholog EQ descriptions are based on gross observations of individual eyes, at a particular developmental stage. ”

“ This goal presented two challenges; first, we needed to include more annotations from additional sources, specifically mouse annotations from MGI [64] , [65] , and disease associations from the human Gene Association Database (GAD) [66] . However, these annotations were described using neither PATO nor an anatomical ontology. The MGI annotations use the Mammalian Phenotype (MP) ontology, and the GAD uses textual descriptors. To integrate these valuable data, we first created an equivalence mapping of MP terms to EQ descriptions [20] . ”

“ Approximately 53% of the gene records have detailed allelic variant descriptions and/or general clinical synopses, while 43% of phenotype/disease records have a known molecular basis. OMIM is a text-based resource, and retrieval of information suffers from this fact, as the Entrez searches in Table 1 show. For an individual researcher wanting to know which human mutations may result in an increase in bone size, or a computer script mining OMIM data, free text annotations do not provide the rigor necessary for querying. ”

“ Using this EQ method, we annotated the phenotypes of 11 gene-linked human diseases described in Online Mendelian Inheritance in Man (OMIM) . These human annotations were loaded into our Ontology-Based Database (OBD) along with other ontology-based phenotype descriptions of mutants from various model organism databases. ”

“ In some cases, the rankings of the orthologous gene were very similar by the different metrics. For example, comparison of human and mouse EPB41 ranked the mouse ortholog first in the case of ICCS and maxIC, sixth for the simJ metric, and third for the simIC metric. In other cases, the rankings were more variable for the different metrics. For example, mouse Pax2 was ranked as only 45th by the simJ metric, but in the top 10 most similar genes by the simIC and ICCS metrics.
Table 6 "Ortholog rankings of phenotype similarity search using human disease genes. ”

“ Phenotypic profile comparison and phenotype promotion.Multiple EQ descriptions annotated to a genotype comprise a phenotypic profile, and these profiles can be compared using subsumption logic. Phenotypes annotated to genotypes are propagated to their allele (s) , and in turn to the gene, indicated with upward arrows. Similarity is analyzed between any two nodes of the same type, such as between gene A-vs-B, allele A3-vs-B1, genotypes A1/A1-vs-A3/A3, or A3/A3-vs-B1/B1. Genotypes are shown as rounded boxes, alleles as circles, and genes as squares. ”

“ Although many MODs curate phenotype data using the EQ method, no such annotations existed for human disease genes. Because we required annotations of human diseases in the EQ style for comparison, we proceeded to annotate a small set of gene records from OMIM: ATP2A1, EPB41, EXT2, EYA1, FECH, PAX2, SHH, SOX9, SOX10, TTN, and TNNT2. These 11 genes were selected because they were known to be causal for a variety of human diseases and had known mutant orthologs in flies, mice, and/or fish with corresponding EQ descriptions available for comparative analysis. ”

“ Similarity scores between a pair of alleles were sorted into intra-gene (same gene) and inter-gene (different genes) sets, and the mean scores for each gene compared. The significance of the difference between the mean scores for each gene was calculated using a two-tailed Student's t-test.
For the zebrafish shha query, we also compared this gene against all other zebrafish genes (2,908 genes in the total set) . For the inter-species queries, we exhaustively compared each gene against all other genes using simJ and then computed all metrics on the top 250. ”

“ Where an OMIM gene record referred to a disease record, the annotators would capture as much general phenotype information about that disease as possible.
Annotation Software and Storage
We write ontology terms prefixed with the name of the ontology; abbreviations are provided at the beginning of this paper. We use ZFA:gut in place of ZFA:0000112 for legibility purposes. The actual computationally parseable form would use the numeric IDs.
All OMIM annotations were created with Phenote [25] software, using the “human” configuration. ”

“ For example, a computer program would not be able to recognize the fact that there is a significant similarity between the PAX6 mutations that result in “small eyed” mice, “opaque cornea” in humans, a “malformed retina” in zebrafish, and “eyeless” Drosophila (Figure 1) .
Representation of phenotypes.Phenotypes of wild-type (top) and PAX6 ortholog mutations (bottom) in human, mouse, zebrafish, and fly can be described with the EQ method. EQ annotations of the abnormal phenotypes are listed below each set of images per organism. ”

“ Table S7 Comparison between human PAX2 and zebrafish genes. The top 250 genes (by simJ) are reported, together with the rank and score for each metric (simIC, simJ, avgICCS, and maxIC) . Additionally, the maxIC phenotype is reported using ontology identifiers. See Materials and Methods for a list of ontology prefixes. Known sequence orthologs are highlighted in yellow and listed in Table 6, if present. Genes at rank 1 are listed in Table 7. ”

“ All known shha pathway members are shown; those with recorded mutant EQ annotations are yellow. Pathway members retrieved in the top 23 most similar genes are indicated by red boxes. Known pathway members in ZFIN are shown with their current nomenclature, with the exception of those with uninformative nomenclature, which are listed with their KEGG reference gene family nomenclature and are capitalized. KEGG reference pathway members not yet identified in zebrafish (Fu) are grayed out. Table 5 "Zebrafish genes with similar phenotypes to zebrafish shha. ”

“ Note that the anatomical entities are from ssAOs and qualities are from the PATO ontology. These PAX6 phenotypes have been described textually as follows. Human mutations may result in aniridia (absence of iris) , corneal opacity (aniridia-related keratopathy) , cataract (lens clouding) , glaucoma, and long-term retinal degeneration. For mouse, the mutants exhibit extreme microphthalmia with lens/corneal opacity and iris abnormality, and there is a large plug of persistent epithelial cells that remains attached between the cornea and the lens. ”

“ Listed are the ontologies used in the annotations from each data source, along with the number of genes annotated, and the number of unique phenotype descriptions (EQs) . Annotations from MGI and GAD were made using the pre-coordinated ontologies MP and DO.1this study.2http://mgi.org.3http://www.zfin.org.4http://www.gad.org. OBD assigns an IC score to every term or EQ description used to annotate a gene, allele, or genotype. The IC score is a measure of how informative a term or a description is, based on the frequency of annotations with the term and depth in the ontology. ”

“ Cdon encodes an Ig/fibronectin repeat-containing protein that has been shown to bind to Shh at the cell surface and positively regulate Shh signaling in Shh expression domains in mouse [74] . Cdon and SHH mutations result in similar phenotypes such as premaxilla morphology, lip morphology, and lateral ventricle quality. Cdon has not yet been added to KEGG, and zebrafish cdon has no phenotypes annotated at this time. Based on these results, mouse Cdon and zebrafish cdon mutants may be helpful in the further analysis of the hedgehog pathway and may provide additional models of disease. ”

“ A fair degree of effort is required to build and maintain ontologies and the relationships between them and this effort must be informed and guided by collaborative interactions with the curators. Some domains, such as behavior, which is minimally represented in the GO, remain poorly represented by ontologies. ”

“ For zebrafish, the mutants express a variable and modifiable phenotype that consists of decreased eye size, reduced lens size, and malformation of the retina. Drosophila ey (a PAX6 ortholog) mutations cause loss of eye development. The genotypes shown are E15 mouse Pax614Neu/14Neu [2] , 5 day zebrafish pax6btq253a/tq253a [3] , human PAX6+/− [4] , and Drosophila ey−/− [5] . Current methodologies traditionally identify animal models on the basis of sequence orthology between the mutant animal model and a human gene. ”

“ File:Pbio.1000247.t007Shown are the highest scoring genes when comparing a human gene versus either mouse or zebrafish, using the four different similarity metrics (as in Table 6) . Sequence orthologs that are the top hit are in bold. For maxIC, there were often ties for the top rank, which are all listed with the exception of Tnnt2 versus mouse, where there were 20 genes ranked as the most similar. ”

“ Conlin, T & Dunn, N, “The Zebrafish Information Network: the zebrafish model organism database provides expanded support for genotypes and phenotypes.”, Nucleic Acids Res 36: D768–D772, pmid:17991680 ↑ Bard, J. B, “Anatomics: the intersection of anatomy and bioinformatics.”, J Anat 206: 1–16, pmid:15679867 ↑ Beck, T; Morgan, H; Blake, A; Wells, S & Hancock, J. M, “Practical application of ontologies to annotate and analyse large scale raw mouse phenotype data.”, BMC Bioinformatics 10 (Suppl 5) : S2, pmid:19426459 1 2 3 4 Haendel, M ”

“ Evans, T. M & Keenan, D. G, “Scube2 mediates Hedgehog signalling in the zebrafish embryo.”, Dev Biol 294: 104–118, pmid:16626681 ↑ Roy, S; Wolff, C & Ingham, P. W, “The u-boot mutation identifies a Hedgehog-regulated myogenic switch for fiber-type diversification in the zebrafish embryo.”, Genes Dev 15: 1563–1576, pmid:11410536 ↑ Ryu, S. L; Fujii, R; Yamanaka, Y; Shimizu, T & Yabe, T, “Regulation of dharma/bozozok by the Wnt pathway.”, Dev Biol 231: 397–409, pmid:11237468 ↑ Karlstrom, R. O; Tyurina, O. V; Kawakami, A ”

“ Dahm, R & Schonthaler, H. B, “Subfunctionalization of duplicated zebrafish pax6 genes by cis-regulatory divergence.”, PLoS Genet 4: e29, [ [pmid:18282108 doi:10.1371/journal.pgen.0040029] ] ↑ ↑ Singh, A, Drosophila ey−/− ↑ Schuhmacher, A. J; Guerra, C; Sauzeau, V; Canamero, M & Bustelo, X. R, “A mouse model for Costello syndrome reveals an Ang II-mediated hypertensive condition.”, J Clin Invest 118: 2169–2179, pmid:18483625 ↑ Collin, G. B ”