vHOG, a multispecies vertebrate ontology of homologous organs groups.

MOTIVATION: Most anatomical ontologies are species-specific, whereas a framework for comparative studies is needed. We describe the vertebrate Homologous Organs Groups ontology, vHOG, used to compare expression patterns between species.
RESULTS: vHOG is a multispecies anatomical ontology for the vertebrate lineage. It is based on the HOGs used in the Bgee database of gene expression evolution. vHOG version 1.4 includes 1184 terms, follows OBO principles and is based on the Common Anatomy Reference Ontology (CARO). vHOG only describes structures with historical homology relations between model vertebrate species. The mapping to species-specific anatomical ontologies is provided as a separate file, so that no homology hypothesis is stated within the ontology itself. Each mapping has been manually reviewed, and we provide support codes and references when available.
AVAILABILITY AND IMPLEMENTATION: vHOG is available from the Bgee download site (http://bgee.unil.ch/), as well as from the OBO Foundry and the NCBO Bioportal websites. CONTACT: bgee@isb-sib.ch; frederic.bastian@unil.ch.

1 INTRODUCTION

One of the main approaches to understand biological objects has long been comparative studies, from comparative anatomy in the 18th century to comparative genomics in the last decade. Comparative analysis can notably help identify adaptation, as well as functional or structural constraints (Harvey and Pagel, 1991).

Many of the data which we would like to compare, such as gene expression or phenotypes, need to be mapped to anatomy and development of organisms to be of use. To facilitate the automatic manipulation of this data, there has been an important effort to build ontologies, which describe the anatomy of human and of animal model organisms (Bard, 2008). These ontologies have tended to be species-specific, resulting in an increasing number of ontologies corresponding to different projects (see the OBO Foundry and the NCBO Bioportal, Noy ; Smith ). This makes the comparison between species difficult, since differences in representational schemes and in vocabulary are added to the differences in biology. Yet automatic comparisons are increasingly necessary, with large amounts of functional data generated in diverse model organisms. An integrated view is advantageous both for a fundamental understanding of animal biology and evolution, and for the efficient transfer of information from model organisms to human or veterinary medicine.

Multispecies integration within anatomical ontologies poses a number of challenges. One is the criterion of comparison. While comparative studies can be performed in diverse frameworks, homology is the most widely recognized criterion (Hall, 1994). This raises further problems. First, homology is always a hypothesis (Haendel ), which according to the principle of reality followed by the OBO Foundry should not be included within an ontology (Smith and Ceusters, 2010) (although see Merrill, 2010). Second, there exist structures which will not be included in a homology comparison between a given pair of species, because they are specific to one or the other, and have no homolog. Third, the exclusion of analogous structures might be limiting for some studies (e.g. comparing insect and vertebrate eye development). Fourth, homologous structures can diverge in function or structure, to an extent that representing them together in an ontology might be difficult. Finally, there can be differences between species in the relationships among structures (Haendel ).

There are several ongoing efforts to create multispecies ontologies for animal groups, which have chosen different answers to the challenges outlined above.

The Teleost Anatomy Ontology (TAO) (Dahdul ) is a multispecies ontology for teleost fishes. The TAO is based on the Zebrafish Anatomy Ontology (ZFA) (from the ZFIN database, Bradford ), and uses general (higher level) terms from the Common Anatomy Reference Ontology (CARO) (Haendel ). The CARO was created to provide a common basis for all future anatomy ontologies and facilitate their interoperability. Several other efforts follow the same model as the TAO, and include the Amphibian Anatomy Ontology and the Hymenoptera Anatomy Ontology. In each of these cases, there is an effort to describe the morphological diversity of the clade. Consequently, each ontology will include terms that are found in several species of the clade. The use of a term for several species in the TAO does not imply homology (Dahdul ). A species-specific ontology (e.g. ZFA) is considered a subset of the multispecies ontology (e.g. TAO).

A different approach is taken by the Uberon project (Mungall et al. 2012), which maps terms from several animal anatomy ontologies, but also anatomy-related terms from the Gene Ontology or medical ontologies, and other multispecies ontologies such as the TAO or our organs groups (Bastian ). Uberon aims to provide anatomical information in relation to the Gene Ontology and the Cell Ontology, and is neutral relative to the criterion of homology (C.Mungall, personal communication). Uberon thus groups structures based on a criterion of similarity (Dahdul ; C.Mungall, personal communication) that is the parent concept of homology, and also of homoplasy or of functional similarity (Roux and Robinson-Rechavi, 2010).

Despite the issues raised by the use of the criterion of homology, we feel that it also presents important advantages. First, restricting to one criterion allows a clear interpretation of the ontology when used in a database; it especially allows a clear use of automatic reasoning. Second, homology is transitive, which allows us to form an ontology of ‘organs groups’, rather than encode all pairwise relations between terms. Finally, it is the one criterion that permits correct formulation of hypotheses of adaptation or constraints in evolution (Harvey and Pagel, 1991).

Our software Homolonto to align ontologies (Parmentier ) generates a multispecies ontology of Homologous Organs Groups (HOGs). These HOGs are used in our database of gene expression evolution, Bgee (Bastian ), as well as in Uberon. Mappings in the HOGs are restricted to manually curated relations of homology. We use a strict definition of historical homology: ‘Homology that is defined by common descent’ (HOM:0000007, Roux and Robinson-Rechavi, 2010).

The homology in Bgee has allowed the comparison of expression patterns between species in several applications, such as the characterization of gene interactions in development (Comte ), the study of orthologs and paralogs (Huerta-Cepas ) (Bastian,F.B. et al., unpublished data) or the study of microRNA evolution (Roux et al., unpublished data).

The HOGs used in Bgee are part of the database schema, are constrained according to the database optimization, and are not formatted for easy external use. Yet, it is also desirable to provide an ontology which is optimized for inter-operability and reuse by the community. Thus, we present here a CARO-compliant version of the HOG ontology, with all terms and relations carefully curated. The present version of this ontology is limited to vertebrates. Based on the alignment of two mammals, one frog and one fish, it covers many of the morphological terms needed to describe most vertebrates. Thus, we present both the first large and high-quality ontology of vertebrate anatomy, and the first ontology strictly limited to homologous structures: vHOG.

2 RESULTS

All terms of the vHOG ontology were manually curated, to verify that they correspond to groups of homologous organs between vertebrate species, linked by relations of strict historical homology, as defined in the HOM ontology (Roux and Robinson-Rechavi, 2010). When the structures are homologous but divergent, we use a combination of terms, such as ‘limb - fin bud’ (VHOG:000125).

Relations between the terms were also manually curated (see algorithm in Parmentier ), and set either to is_a or to part_of. We aim to provide exactly one is_a relation for each term, according to OBO Foundry guidelines. As noted in the TAO publication (Dahdul ), it is at present difficult to implement this guideline in practice for anatomical ontologies. In future versions of the vHOG, we will continue working toward this aim.

The terms and relations thus generated were incorporated into the framework of the CARO (Haendel ). We were careful to implement this in a manner consistent with other anatomical ontologies using CARO: TAO, XAO and ZFA. This especially concerns the relations between vHOG terms derived from species-specific ontology alignments and higher level CARO terms, e.g. ‘digit’ is_a ‘anatomical cluster’.

The vHOG ontology version 1.4 (December 2011) has 1184 terms, 506 is_a relations and 1181 part_of relations. There are 664 synonyms; 547 terms have definitions; 67% of terms have only one parent, 25% have two parents, and the others have three or more. There are 493 cross-references (xref) to other multispecies OBO ontologies. These do not include the mappings to species-specific ontologies, as we consider that mappings from multispecies ontologies to species-specific ontologies should not be treated as xrefs, but should be encoded separately in an association file, similar to the Gene Ontology Annotation mapping (Barrell ). This differs from the approach of Uberon, which includes cross-references to the source ontologies in the multispecies ontology itself (Dahdul ).

All mappings of terms from the species-specific ontologies to the vHOG terms were manually curated. There are 5129 terms from species-specific ontologies mapped to 1169 vHOG terms, which represent 2259 hypotheses of homology between vertebrates (i.e. 2259 pair-wise relations between structures in different species). There are 15 vHOG terms with no mappings, which correspond to higher level CARO terms (e.g. ‘material anatomical entity’). The semantics of the mapping is ‘treat-xrefs-as-equivalent’ for CARO, and equivalent to ‘treat-xref-as-genus-differentia’ for the mapping to species-specific ontologies.

The mappings were annotated with ‘support codes’ to provide confidence information (Table 1). Means to provide confidence metadata information in support of annotations are currently being discussed in the framework of the Evidence Code Ontology (ECO) (Gene Ontology Consortium, 2010) and of the International Society for Biocuration. Our objective is to rely on the use of the ECO, once a standard to address the issue of confidence has emerged in the biocuration and ontology communities.

Table 1.

Support for mapping of species-specific anatomical ontologies to vHOG

Support code	Meaning	Referencesa	Terms mapped	vHOGs with mappingb
Obvious	General knowledge, no need for reference	No	44	11
Well established	No debate in the literature	Yes	3754	815
Debated	Debate in the literaturec	Yes	27	5
Uncertain	Not clearly established	Variabled	378	103
Inferred	Deduced from references which do not discuss this mapping explicitly; or personal communication from experts	No	926	236

Yes if at least one bibliographic reference is provided for each mapping.

The total is more than the number of vHOG terms, because different mappings to a same vHOG term can have different support.

A consensus is chosen and presented, but the debate is documented.

Either there is a reference in which the homology is discussed as uncertain, and it is provided; or this code is used when well-established or obvious relations between closely related species (e.g. human and mouse) are extended to other species (e.g. zebrafish).

In practice, the support codes ‘obvious’ and ‘inferred’ are essentially used for homology between the two mammals considered in our ontology: 97% of all mappings with the code ‘inferred’ are for human and mouse. All mappings with the code ‘obvious’ are for human and mouse, except ‘whole organism’ which has the code ‘obvious’ for all ontologies.

Among the vHOG groups noted ‘debated’, several concern bones and the skeletal system, due to citations such as the following: “Whether this ‘biomineralization toolkit’ of genes reflects a parallel co-option of a common suite of genes or the inheritance of a skeletogenic gene regulatory network from a biomineralizing common ancestor remains an open debate” (Murdock and Donoghue, 2011) (more references in the association file). Another case of debate is the ‘vomeronasal organ’ (VHOG:0000665). None of the three bibliographical references cited in the association file is conclusive concerning the existence and homology of this organ in different tetrapodes (Doving and Trotier, 1998; Kardong, 2006 p. 669; Smith ), and Doving and Trotier (1998) specify: ‘The opinions concerning the presence and functioning of the vomeronasal organ in humans are controversial’.

Conversely, 87% of all mappings for zebrafish, the most divergent species included in vHOG, are ‘well established’ in the literature. Still, 7% are ‘uncertain’, as for example the ovary (ZFA:0000403 mapped to VHOG:0000251): ‘ (…) while it is likely that Urbilateria lacked a complex somatic reproductive system, it is at present impossible to speculate on whether or not it possessed a true gonad’ (Extavour, 2007) (full citation in the association file) (see also DeFalco and Capel, 2009).

Finally, it should be noted that terms which represent different developmental stages of the same organ are mapped together. For example, ‘presumptive midbrain’ (ZFA:0000148) and ‘midbrain’ (ZFA:0000128) are both mapped to ‘midbrain’ (VHOG:0000069). This leads to some loss of information concerning developmental relations, and notably the relation develops_from is not implemented in the vHOG. But it provides a simpler description of vertebrate anatomy, which has proven relevant and useful to studying gene expression patterns (e.g. it is used in the Prosite database) (Sigrist ). This has proven especially useful for human and mouse, for which different ontologies describe anatomy during development, and in the adult; vHOG (and the HOGs in Bgee) provide a high-quality mapping between these ontologies. Of note, for queries on gene expression in Bgee, a developmental stage can be specified, which then recovers only data from the correct structure.

3 DISCUSSION

With increasingly abundant in vivo functional data from different model organisms, it is necessary to be able to relate and compare information between species. Different criteria for comparison can be relevant in different contexts: similarity, functional equivalence, evolutionary relationships or the implication in similar phenotypes (Mabee ; McGary ; Roux and Robinson-Rechavi, 2010). The most widely recognized criterion for comparisons of anatomical structures is homology. For large scale and reproducible studies, it must be implemented computationally. To answer these needs, we propose the first ontology of vertebrate homologous organs, the vHOG. We use a strict definition of historical homology. The vHOG aims at following OBO Foundry rules, and makes use of the CARO framework.

We believe that the vHOG ontology (and the HOG ontology used in Bgee) provides answers to the main challenges of implementing homology in an ontology. Since homology is always a hypothesis, the mappings of species-specific structures to vHOG terms are kept in a separate association file (see also Dahdul ; Haendel ). Structures that have no homolog are not included in the vHOG ontology, but can still be found in each of the ontologies which are mapped to it. Thus, inclusion in the multispecies ontology carries a clear biological meaning, without hampering the fine description of each species. Finally, divergent homologous structures can be mapped to the same vHOG term, while keeping their individual definitions.

At present, the vHOG is limited to homologies between those model species for which anatomical ontologies are publicly available. We plan to extend it to more diverse species, while maintaining the restriction to terms describing organs or tissues with evidence of homology.

The advantages and drawbacks of our strict homology approach are clear when comparing the vHOG with the Uberon. The Uberon contains many more terms (6806 as of October 2011). It includes homology mappings from our project, since the Bgee HOGs are one of Uberon's source ontologies. Since it is not limited to homology, Uberon includes for example a term ‘eye’ (UBERON:0000970), which has the children ‘compound eye’ (UBERON:0000018) and ‘simple eye’ (UBERON:0000047). An automatic reasoner cannot distinguish the case of compound eyes, which are all homologous, from the case of the parent ‘eye’, which includes homologs and analogs. And in less obvious cases, it can be difficult to recover such information even for non-automated reasoning, i.e. by a biologist user. For example, auditory ossicles are all mapped to UBERON:0001686, whereas the amphibian ‘auditory ossicle’ (XAO:0000214) is not homologous to the mammalian ‘auditory ossicles’; in vHOG it is mapped to ‘hyomandibula – stapes’ (VHOG:0000688). On the other hand, Uberon provides information that is not included in vHOG, such as develops_from relations.

Thus, Uberon and vHOG are complementary projects, the one focused on function and on integrating as many terms as possible, the other focused on a more restrictive set of terms, with strict homology definitions.

As an example of application of an ontology based on homology, we have queried gene expression available in Bgee for human, mouse and zebrafish, for HoxA5 orthologs (http://tinyurl.com/bgee10-hoxa5). In human, there are 63 organs or tissues with expression, in zebrafish 12 and in mouse 201. Unsurprisingly, given that the data are from targeted in situ hybridizations, most of the expression detected in zebrafish is shared with mammals. But there is also evidence from three high-quality in situ hybridization experiments of zebrafish-specific expression in the pharyngeal arch. Importantly, the homolog of this structure is defined, and has been studied, in mouse and human (the branchial arch), confirming that the expression pattern of HoxA5 is probably zebrafish-specific. Conversely, the abundance of large-scale reports of expression in many organs leads to an uninformative mouse expression pattern, i.e. HoxA5 is detected to some degree in many structures where no biological role has been reported. Here homology information allows us to filter the data to recover the signal. Restricting to structures with homologous expression in human, for example, highlights expression in structures in which HoxA5 has been shown to play a functional role (Boucherat ; Chen ), such as the reproductive system (ovary, testis, uterus), forelimb, gut, bone and components of the respiratory system. Thus, the homology information in the ontology allowed both the identification of a species-specific patterns and of functionally important conserved expression.

4 CONCLUSION

Fine-grained yet large-scale comparisons between model organisms, especially vertebrates such as mouse or zebrafish and humans, is increasingly important. In addition to providing a framework for evolutionary studies, the vHOG provides a unique tool for relating humans and model organisms. Additionally, the association files of vHOG and HOG are unique resources in providing detailed judgments of homology between anatomical structures, with supporting evidence from the literature.