The Mammalian Phenotype (MP) Ontology was initially created for annotating gene alleles at the Mouse Genome Informatics (MGI) database (Smith et al., 2005). For the MP ontology, “phenotype refers to the observable morphological, physiological, and behavioral characteristics of an individual in the context of the environment” (Smith and Eppig, 2009). Because MP was designed to be used with mouse knockouts, mutations, and other types of alleles, there is an underlying assumption of a comparison to the trait exhibited by a mouse with the genetic background from which the allele has been constructed or a comparison to a normal or “wild type” trait. Thus the terms often contain words such as “abnormal,” “increased,” or “decreased” with the implication of “relative to” an assumed observation. The actual measured values for observed traits are not connected to these annotations. MP follows the open-source Open Biological and Biomedical Ontology (OBO) file format and is organized on the Directed Acyclic Graph (DAG) structure with the highest nodes related to physiological systems such as cardiovascular, immune system as well as behavioral, life span, and cellular phenotypes. Each physiological system node is followed by a basic division into physiological and morphological phenotype branches. MGI currently has over 41,000 genotypes annotated with MP terms for a total of more than 193,000 annotations. MP is considered a pre-coordinated term ontology since both the entity (i.e., anatomical site or physiological process) and the quality of it (i.e., abnormal, increased, decreased) are included in the term. MP is also being used for the EuroPhenome project to annotate mutant mouse phenotype data generated using standard phenotyping platforms (Morgan et al., 2010). The advantages in terms of annotation are significant since curators only have to search a single ontology and has terms that more closely mimic those seen in literature and commonly used in laboratory settings.

Another approach to phenotype ontologies has been the Phentoype and Trait Ontology (PATO) project³ (Gkoutos et al., 2004). Unlike MP which is considered a pre-coordinated term phenotype ontology, PATO uses the EQ approach (entity + quality; Gkoutos et al., 2004; Smith and Eppig, 2009). Thus PATO presents terms related to qualities and attributes that are then linked to terms from other ontologies such as anatomy ontologies to describe phenotypic characteristics. Thus, “big ears” would be represented by the term “increased size” from PATO and the word “ear” from an anatomy ontology (Mungall et al., 2010). One of the advantages of this approach is the re-use of existing ontologies such as the anatomy ontology. Representing morphological traits through the use of anatomy ontologies and the qualities described in PATO is relatively straight forward. This is one reason that resources housing data for some organisms such as drosophila and zebrafish have found this approach useful (Mungall et al., 2010); the majority of their reported traits and phenotypes are morphological in nature. However, the representation of physiological traits and specific clinical measurement types is more problematic (Smith and Eppig, 2009). First, there is not necessarily a single ontology that adequately represents the physiological trait corresponding with the quality expressed by PATO; and second, a single EQ term may not adequately express the phenotype observed. For example while the morphological trait of “big ears” is relatively easy to represent by EQ, an MP term of “abnormal cochlear outer hair cell electromotility” provides a greater challenge. The disadvantages of the PATO approach for annotation are also significant. Curators would have to browse multiple ontologies to create a term on the fly and this approach creates terms and phrases that sometimes are stilted or not commonly used in the literature or laboratory settings.

The Human Phenotype Ontology (HPO) was developed in part to address the shortcomings of information presented in the Online Mendelian Inheritance in Man (OMIM) database⁴ (Amberger et al., 2009). OMIM has traditionally been the most commonly used resource for information on genetic diseases. Unfortunately, the information housed there is in free text format, making it difficult to mine computationally because of the non-standard way in which traits and abnormalities are described. For instance, OMIM uses the synonymous descriptions “generalized amyotrophy,” “generalized muscular atrophy,” and “muscular atrophy, generalized” so even simple searches may not return the results a user desires. While a human reader going through the free text of multiple entries will recognize similar meanings, computers will not. The initial version of HPO was created using the information at OMIM in an effort to merge synonyms and create links and relationships among the terms and concepts. This initial structure has been expanded and refined through manual curation of information from a variety of sources and consistent development of definitions and relationships (Robinson and Mundlos, 2010). As with MP, the emphasis has been on phenotypes which diverge from the normal or expected and disease states and terms are pre-constructed as with the MP.

Ontologies such as MP, PATO, and HPO were originally designed for use in simple annotations to a single data (e.g., gene product or allele) or an individual and the term was expected to appear alone in the annotation with the minimal accompanying information of an evidence code indicating level of experimental evidence to support the annotation and the reference from which the annotation was made. The existing phenotype ontologies were not developed to be attached to actual measurement values but to indicate a state or characteristic observed relative to that which has been determined to be “normal” or “wild type,” or relative to that exhibited by an individual with a known genotype. Information on experimental conditions and measurement assays used are vital parts of the phenotype record and the use of multiple ontologies to represent these has been advocated as a way to accomplish this (Shimoyama et al., 2005; Hancock et al., 2007). Clearly, developing separate ontologies for the elements of phenotype measurement, method of measurement, and conditions under which the measurement was made along with provisions for additional information on actual values, duration of conditions, and so on, will allow these aspects of the phenotype record to be linked. Database structures which allow re-use of information and multiple associations will facilitate data integration, data mining, and data presentation.

In this paper, we present three ontologies created to standardize phenotype measurement records for use in human studies and those using laboratory animals: Clinical Measurement Ontology (CMO), Measurement Method Ontology (MMO), Experimental Condition Ontology.

The CMO provides the standardized vocabulary necessary to indicate the type of measurement made to assess a trait. For the purposes of this project and these ontologies, trait and clinical measurement are defined as follows:

A critical element in the description of a phenotype is the measurement method used. Several types of methods are commonly used to measure such things as blood pressure resulting in differing clinical measurement values so the inclusion of method as part of the measurement reading is necessary for the integration of data from multiple studies. The MMO is designed to provide this information. As described above, this ontology was developed in parallel with the CMO as trait areas are targeted. The MMO is organized by the underlying principle or mechanism of the method (Figure 4) with two major branches, “ex vivo method” and “in vivo method.” Methods were identified from protocol descriptions and data labels from the targeted data sources. As with the CMO, for completeness in the ontology, additional sources of method information such as vendors’ catalogs, laboratory manuals, and published literature were reviewed for associated methods. Thus, if one of the protocols for one of the originating data sources indicated that a balloon tipped catheter was used to measure blood pressure, a quick review of a variety of publications revealed that the basic category is vascular indwelling catheter with a variety of types including fluid filled catheter, intravascular electromagnetic flow sensor, and transducer tipped catheter. There are currently 195 terms in the MMO.

While many phenotype measurements are made under baseline conditions, changes to diet, atmosphere, activity level, and other conditions are common aspects of phenotype experiments. Often this information is added as part of the phenotype label in the individual laboratory’s database or only included as part of a lengthy text protocol. Creation of standardized terminology and format for presenting this information with phenotype measurements is crucial to the integration of phenotype data from multiple datasets. An XCO was created to provide standardization and structure for this important information (Figure 5).

In addition, to the “is_a” relationship, in certain areas a “part_of” relationship is utilized so that parts of a whole can be described as in “air oxygen content” is “part_of” “atmosphere composition.” The ontology was designed so that conditions related to existing ontologies such as those involving chemicals or drugs represented in ChEBI (Degtyarenko et al., 2009), follow the structure and terminology of these ontologies and provide appropriate linkages through identifiers (Figure 5). Initial emphasis was on the conditions used in the targeted data sets and expanded to those conditions most commonly used in experiments involving the targeted trait domains. Structural provisions in the database structures of projects using the ontology provide ordinality information to indicate whether multiple conditions were simultaneous or sequential. Use of this ontology in annotating phenotype data allows users to retrieve multiple, disparate phenotype information in which similar experimental conditions were imposed. There are currently 110 terms in the XCO.

The three ontologies have been used to integrate multiple data sets for two major projects, one involving human data and the other involving rat data. The Cardiovascular Ontologies and Vocabularies in Epidemiological Research project was designed to integrate demographic and phenotype measurement data from three family blood pressure studies, Hypertension Genetic Epidemiology Network (HyperGEN; Williams et al., 2000); Genetic Epidemiology Network of Salt Sensitivity (GenSalt; Gu et al., 2007); and HEalth, RIsk factors exercise Training And GEnetics (HERITAGE; Bouchard et al., 1995). While all three studies focused on cardiovascular disease and associated risk factors, they were disparate in the types of interventions used, variety of measurements taken and the methods used to make the measurements. Measurements related to blood pressure, blood chemistry and lipid levels, body weight and body fat were included as well as interventions ranging from sodium controlled diets to exercise. Invasive, non-invasive and imaging techniques were also used. The HyperGEN study was designed to characterize the genes influencing hypertension by recruiting hypertensive sibships (i.e., each participant with two or more hypertensive sibs) from across multiple field centers and ethnic groups. The GenSalt study is an intervention study of the genetic and environmental factors related to dietary sodium and potassium effects on blood pressure in rural Chinese families. The HERITAGE study is an intervention study designed to assess genetic and environmental factors underlying the effects of endurance exercise training on several cardiorespiratory and cardiovascular disease risk factors. The CMO, MMO, and Experimental Condition Ontology were used to map data elements from each of the studies to a common format for integration into a single resource⁷. To date, 16 phenotype classes with records for 8,778 subjects have been integrated for the three studies and made available at the website. Additionally, all variables across the studies are being mapped and modeled with their associated ontology terms to facilitate querying and access to raw data fields of interest; to date 11 classes have been created representing over 100 phenotype measurements.

The rat PhenoMiner project⁸ (Figure 6) also has used the three ontologies to define data formats and standards for integrating rat phenotype measurement data from a variety of sources including two large scale phenotyping projects and published literature. PhysGen Program for Genomic Applications⁹ (Kwitek et al., 2006), one of the large scale phenotyping projects was designed to conduct high throughput phenotype screening for a targeted set of inbred strains, as well as consomic and mutant strains. The screens involved hundreds of different types of phenotype measurements for heart, lung, renal, vascular, and blood function under baseline conditions as well as varying diet, atmosphere, and activity conditions. Data was organized, stored, and presented by protocol so even though some similar measurements such as weight or blood pressure were measured in multiple protocols, the data was not integrated across protocols. The National BioResource Project for the Rat in Japan (Serikawa et al., 2009) was the second large scale rat phenotyping project. Phenotype screens for body weight, activity, behavior, blood pressure, blood chemistry, urine analysis, and organ weights have been conducted under baseline conditions for inbred and mutant strains.

Because the rat is an ideal model organism for pharmacology, biochemistry, and physiology research, the published literature is a rich resource of data on phenotype measurements for particular strains. Papers reporting cardiovascular, respiratory, renal, morphological, and blood chemistry measurement data as well as those with measurements related to cancer were targeted for the initial phase of the PhenoMiner resource. Over 13,000 measurement records from these three sources have been mapped to the three ontologies and integrated into PhenoMiner (Figure 7).