The mouse is a significant model organism in the study of human genetic disorders. Mice and humans are physiologically and genetically very similar. Mice are experimentally tractable: they are small and easy to maintain mammals; mice breed quickly and have a short lifespan; and, unlike humans, all life stages, including embryogenesis, are accessible to study. The environment in which mice live can be controlled to carefully introduce exogenous factors, such as diet, drugs, social interactions, or stimuli that may influence phenotype. Importantly, in addition to many spontaneous and induced mouse models available, the genome of mice can be directly and specifically manipulated to introduce exogenous genetic material, reporters, and specific genetic mutations found in human disease.

Information on mouse strains and mutants, and their associated genetic alleles or QTL (quantitative trait loci) has been collected for many years. Tens of thousands of mouse mutations already exist and their phenotypes have been systematically curated in the Mouse Genome Informatics (http://www.informatics.jax.org) database. Hundreds of new mutations continue to be reported each month in the scientific literature. Most mouse phenotype data appear in publications as a combination of textual description, figures, and tables with no standardization, other than adherence to the publishing format. In addition to these published reports, a wealth of data have been or are being generated internationally from large-scale ENU mutagenesis centers [c.f., 10, 14, 17, 22] and the International Mouse Knockout Consortium projects, KOMP (US Knockout Mouse Project), NORCOMM (North American Conditional Mouse Mutagenesis), EUCOMM (European Conditional Mouse Mutagenesis), and TIGM (Texas A&M Institute for Genomic Medicine) [1, 2, 9, 18], much of which will only be available in electronic form. Given the richness and complexity of written language, that by its nature is inherently inconsistent in usage and content, and the avalanche of new data being generated by large-scale projects, there needs to be a standardized, robust way of representing phenotype information in a manner that allows for computational methodologies to fully exploit and realize the potential of these data.

Ontologies are being developed in many domains of biology to facilitate information extraction from literature, aid in scientific discovery and experimental analysis, and to support database interoperability (e.g., Gene Ontology (GO) [6]; Mouse Adult Anatomy (MA) [15]; and Cell Line Ontology (CL) [5]). Ontologies describe the concepts that exist in an area of knowledge and the relationships that exist between these concepts [4]. We have built an ontology describing mammalian phenotype to classify and organize these descriptions for mouse and other mammalian species and for facilitating comparisons across species.

The Mammalian Phenotype Ontology (MP) is a standardized structured vocabulary. MP terms are organized into a directed acyclic graph (DAG) with terms placed hierarchically from the very general at the top of the graph to more specific terms as one moves down the structure. The highest level terms describe physiological systems, survival, and behavior. The physiological systems branch into morphological and physiological phenotype terms at the next node level. A browser is available to view the MP at (http://www.informatics.jax.org/searches/MP_form.shtml) and a sample page is shown in Figure 1. Each vocabulary term includes a unique accession ID, term name, synonyms and definition. The relationship to child nodes and parent nodes is also represented. Any term with more than one parent is shown in multiple sequential graphs.

The organization of the MP allows annotation of data to the level of resolution available, whether general or highly specific. For example, a published phenotype referred to as “moderate behavioral impairment” may be annotated to the high level term “abnormal behavior” [MP:0004924] while a more detailed description of “opisthotonus” [MP:0002880] could be used if more specific observations or studies had been done. Each term shown in the browser is followed by a hypertext link listing the number of genotype annotations in MGI to each term or to a child of that term (Figure 1, “opisthotonus” has 12 annotations to 12 different genotypes, as of this writing). This hypertext link leads to a list of mouse genotypes annotated in MGI and the annotated term and reference(s) cited for that annotation. This feature allows a narrow or broad search that can return genotypes annotated to detailed phenotypic descriptions or to genotypes that have less well-characterized data.

In MGI, mouse phenotype data are annotated to genotype objects. Each genotype consists of one or more mutant allele pairs combined with associated genetic background strain information. Each MP term annotated to a genotype is supported by one or more references and may also include additional unstructured notes that describe specific comparative details such as incidence, experimental conditions, e.g., diet or stressor, age of onset or quantitative values. Figure 2 shows an example of MP annotations to genotypes involving the Eng^tm1Mle mutant allele (the first targeted mutation of the endoglin gene made in the Michelle Letarte laboratory). Users can view information about the Eng^tm1Mle mutant allele from two perspectives on its allele detail page: 1) The phenotype summary section presents a matrix view showing phenotypes (MP Ontology terms) on one axis and genotypes involving Eng^tm1Mle on the other axis. Using the expansion toggles to view more or fewer specific MP terms, it is easy to compare across genotypes to observe similarities or differences. 2) The phenotype data by genotype section lists each genotype that involves Eng^tm1Mle that has been phenotypically characterized. Links in the genotype column lead to the greatest detail available for descriptions of that genotype, including MP annotations and additional annotated detail. If an author states that a particular mouse genotype is a model of human disease or syndrome, an association with an Online Mendelian Inheritance in Man (OMIM) human disease term is made, and links to both OMIM and MGI’s Human Disease and Mouse Model Detail pages are provided. In addition, published images highlighting phenotype data are frequently associated with both allele and genotype records. Each of these phenotype sections is organized by the biological system affected and is based on the structure of the MP Ontology. As of 12/20/08, MGI contained 139,751 MP annotations to 27,778 genotypes, representing 20,987 independent alleles, and 8,286 genes and 4,023 QTL (Table 1).

In addition to mouse data, the Mammalian Phenotype Ontology has been used to classify phenotype data from rat and other mammalian species (RGD, OMIA). Species-specific terms are included in the MP. For example, species such as rodents, felines and others exhibit vibrissae while primates do not. In such cases, annotations can be made to vibrissae-related terms when appropriate to a species and ignored when the terms do not apply. The MP also contains related species-specific terms such as the primate-specific term abnormal menstrual cycle [MP:0003375 and the term abnormal estrous cycle [MP:0001927] that is applicable to many other non-primate mammals. By using the parent term abnormal ovulation cycle [MP:0009344] data annotated to this parent term and both of its children could be retrieved from multiple sources.

The MP Ontology, as well as other commonly used biomedical ontologies (e.g., GO, MA, and CL), is maintained using the OBO-Edit software developed by Lewis and Richter [11]. Daily edits include the addition of new term and synonyms, new relationships based on comparisons to orthogonal ontologies and revisions to existing terms. We adhere to Open Biomedical Ontologies (OBO) Foundry principles (http://www.obofoundry.org/about.shtml) in the construction and revision of the ontology, such as orthogonality, public availability and collaborative development [26]. In addition to the MP browser mentioned above, the MP ontology is available in OBO format from the MGI ftp site (ftp://ftp.informatics.jax.org/pub/reports/index.html#pheno) and in OBO and OWL (Web Ontology Language) format from the OBO Foundry site (http://www.obofoundry.org/cgi-bin/detail.cgi?id=mammalian_phenotype).

The advantage of the EQ approach is that existing ontologies are used and standardized in coordination with other vocabularies that provide supporting data for phenotypic information. It is a useful method in high-throughput screens when phenotypic data are collected for many individuals, where phenotyping protocols are tightly controlled, and phenotyping is generally done at a very broad (vs. deep) level. However, data described in publications generally uses precise clinical terminology that describes summary results for cohorts of animals that have been phenotyped at a very deep and precise level. Further, clinical terms are not always easily represented by EQ statements and may require multiple statements to describe them, the sum of which may not accurately reflect the complexity and precision of clinical term’s meaning. Table 2 shows one straightforward example and one complex example of MP terms and how they could be represented in EQ format. The term cataracts [MP:0001304], defined as “complete or partial opacity of the lens,” can be related to the anatomy term lens [MA:0000275, or FMA:58241] and the qualifier opaque [PATO: 0000963]. However, the term jaundice [MP:0000611], defined as “clinical manifestation of hyperbilirubinemia, with deposition of bile pigments in the skin, resulting in yellowish staining of the skin and mucous membranes” needs several EQ statements to describe the phenotype that together do not accurately reflect the biological meaning that “jaundice” imparts to a clinical researcher.

The Mammalian Phenotype Ontology uses precoordinated terms that include both the entity and quality concepts. A primary consideration is the use of terms by clinicians and scientists to describe phenotypes and use of terms commonly seen in scientific literature. However, there are two other practical reasons in using the precoordinated approach for mouse phenotype data, both focused on efficiency in the curation process, which is the most costly part of maintaining a database [16]. First, the precoordinated approach provides specific defined terms for curators to use, and thus increases curation consistency. And, second, the EQ approach requires look-up of terms from multiple ontologies for the correct “entity” (and sometimes requesting new terms to be added by ontology providers), thus slowing the curation process.

The EQ term mappings of the Mammalian Phenotype Ontology to other foundational ontology terms (i.e. the logical definitions) may be stored, allowing ontological reasoning and interoperability with other ontological resources such as the GO, FMA, MA, PATO, CL and others. Phenotypic comparisons to data annotated with other precoordinated phenotype ontologies, such as the recently described Human Phenotype Ontology (HPO) [25] can be made if such logical definitions are provided, or by storing term mappings to the HPO directly. Creating logical definitions will also allow comparisons to phenotypic data annotated using the postcoordinated approach. We are assisting in collaboration with others in the preparation of a logical definitions file for the MP (http://www.obofoundry.org/cgi-bin/detail.cgi?id=mammalian_phenotype_xp, unpublished data).

A summary of mouse phenotype data for any gene at MGI may be viewed on a gene detail page. A gene detail page for Pax6 (paired box gene 6) is shown in Figure 3. In addition to summary information about sequence, genetic location, expression, domain structure, Gene Ontology classifications and references, the phenotype data section contains a hyperlink to a listing of the thirty-two phenotypic alleles of Pax6. A brief description of the phenotypic effects of the mutations in Pax6 is presented. Phenotype images are available for mutations in this gene and the hyperlink leads to a summary page for all images.

High-level phenotype data can be viewed on the mouse Genome Browser (Figure 4). Specific genome tracks can be viewed representing top-level phenotype terms in the MP Ontology. Annotations to any term in that class will be represented on a separate track in the genome browser. For each icon represented on the track, a link is present that will return a phenotype detail page representing that allele or QTL. Large regions of the genome can be scanned for phenotypes of interest to predict candidate genes for QTL or spontaneous or induced mutations.

The MP Ontology and the phenotype annotations in MGI have served the scientific community as an analysis tool and information resource. For example, Martinez-Morales, et al. (2007) used MGI’s mouse phenotype annotations describing neural crest cells and neural crest derivatives to select genes that control differentiation of these structures. Phylogenetic comparison of these genes showed that 9% are unique to vertebrates, much higher than for other tissues. Over half of these unique genes encode soluble ligands that control specification of neural crest cells into derivative lineages [21]. Ramanathan et al. (2008) have used comparative expression profiling from inbred strains with high and low lactation capacity to identify signaling pathways involved in enhanced lactation performance. Many differentially expressed genes identified had phenotypic alleles annotated to Mammalian Phenotype Ontology terms related to maternal performance or mammary-gland development [23]. Reed et al. (2008) analyzed mouse body weight annotations in MGI and showed that body weight is frequently affected in mice carrying knockout alleles. Based on a sample set of 1,977 viable mouse knockout strains, 34% of these were associated with annotations to body weight changes and suggests that thousands of genes may influence this complex trait [24].

Mammalian Phenotype ontology annotations in MGI have also been used for candidate and comparative gene analysis in mouse and human. Bao L, et al. (2007) identified a set of mouse genes with alleles that are known to affect behavioral and neurological phenotypes. Analysis of expression of these genes in recombinant inbred strains identified expression quantitative trait loci (eQTL) that affected expression of members of the gene set. Mapping of these eQTL led to the identification of novel genetic loci and causal genes for a number of behavioral and neurological phenotypes [3]. Chen, et al. describe a method using comparisons to mouse Mammalian Phenotype annotations to prioritize candidates for novel human disease gene identification [8].