- Top of page
This paper describes a new ontology of human developmental anatomy covering the first 49 days [Carnegie stages (CS)1–20], primarily structured around the parts of organ systems and their development. The ontology includes more than 2000 anatomical entities (AEs) that range from the whole embryo, through organ systems and organ parts down to simple or leaf tissues (groups of cells with the same morphological phenotype), as well as features such as cavities. Each AE has assigned to it a set of facts of the form <AE><relationship><parent>, with the relationships including starts_at and ends_at (CSs), part_of (there can be several parents) and is_a (this gives the type of tissue, from an organ system down to one of ~ 80 simple tissues predominantly composed of a single cell kind, which is also specified). Leaf tissues also have a develops_from link to its parent tissue. The ontology includes ~14 000 such facts, which are mainly from the literature and an earlier ontology of human developmental anatomy (EHDAA, now withdrawn). The relationships enable these facts to be integrated into a single, complex hierarchy (or mathematical graph) that was made and can be viewed in the OBO-Edit browser (oboedit.org). Each AE has an EHDAA2 ID that may be useful in an informatics context, while the ontology as a whole can be used for organizing databases of human development. It is also a knowledge resource: a user can trace the lineage of any tissue back to the egg, study the changes in cell phenotype that occur as a tissue develops, and use the structure to add further (e.g. molecular) information. The ontology may be downloaded from www.obofoundry.org. Queries and corrections should be sent to firstname.lastname@example.org.
- Top of page
Bioinformatics needs formal anatomies (anatomy ontologies) primarily to structure databases that handle tissue-associated data (Burger et al. 2008). Such formalisms not only provide numerical identifiers (IDs) that can be used for searching and for providing normal data for mutant material, but can also make available to the user information that goes well beyond what might be found in anatomy textbooks (e.g. a hierarchy of all the tissues present at a given stage of development). There is now a wide range of such ontologies for adult and developing model organisms, and they are available at www.obofoundry.org (the location for all ontologies mentioned in this paper, other than the FMA). Standard ontology editors such as OBO-Edit (oboedit.org/) integrate and display the ontology as a parts or classification hierarchy.
A very detailed ontology of adult human anatomy, the FMA1 (Rosse & Mejino, 2008) is now available. The current ontology for human developmental anatomy (ID: EHDA; Hunter et al. 2003) is less complete: it comprises a set of ontologies, one for each Carnegie stage (CS) (1–20) that only includes basic part_of data. Its structure was derived from the original ontology of mouse developmental anatomy (Bard et al. 1988), and its content was based partly on this and partly on a limited study of sectioned human material. This ontology has not been maintained to meet current standards, and contains naming and structural inconsistencies. This is mainly because, for reasons of terseness and simplicity, it was decided to give each tissue a single part_of relationship (the femur could be part_of the skeleton or part_of the lower limb, but not both) and to make the full pathway the unique name for the tissue. In practice, this slightly odd approach can be used computationally, but it is not ideal. There is also an abstract version of the ontology (ID: EHDAA) that integrates the ontology set into a single hierarchy. This was made computationally and is unfortunately incoherent because the stage-dependent hierarchical structures were not integrated.
This paper reports the production of a new ontology for human developmental anatomy covering CS1–20 (the first 7 weeks), referred to here as EHDAA2. It represents a complete rebuilding of the EHDA intended to meet current standards. The intention has been to include as much information about human developmental anatomy as is practical and as is available in the literature. The ontology is structured using anatomical entities (AEs), and these include all solid structures from the embryo down to individual leaf tissues, which are groups of cells with the same morphological phenotype, together with immaterial volume features such as cavities and surface elements such as pits. The ontology contains a rich partonomy (an AE can be part_of many higher-level structures), a class link (this gives the tissue type) and complete timing details (each AE has assigned to it a starts_at and ends_at CS). Leaf tissues in the hierarchy are also assigned develops_from lineage information and, via the class link, a cell-type classification from the cell-type ontology. These relationships allow a user, in principle at least, to retrace the lineage of a tissue, stage by stage, back to the original fertilized egg and also to follow cell differentiation. It is hoped that the domains of knowledge included in the ontology are reasonably complete.
This paper first summarizes the sources of information used in making EHDAA2, and there is then a short section on the computational methodology and the links that exist to other bioinformatics resources. The Results section describes how the various link relationships have been used to assign information to AEs and reports briefly on how various organ systems have been compiled and integrated; the Discussion considers the strengths, weaknesses and uses of the new ontology. The ontology itself replaces the earlier ones (EHDA and EHDAA), which have now been withdrawn. Like all other ontologies mentioned in this paper, it can be downloaded from www.obofoundry.org and viewed in the OBO-Edit browser (oboedit.org).
- Top of page
The first step in making the ontology was to revise the original time-dependent ontology of human development (EHDA) in which each CS (1–20) had its own hierarchy so that, as far as possible, the internal hierarchies for each organ system at each stage were consistent. The 25 hierarchies (some CSs have sub-stages) were then integrated computationally (by Mike Wicks of the MRC Human Genetics Unit) into a single abstract ontology. In these, all AEs that are present in a sequence of stages are collapsed into a single term with a start and end stage. In this so-called abstract ontology, each AE was assigned a new ID of the form EHDAA2:000abcd (the changes were so extensive that it was not practical to maintain the old EHDAA IDs). Each AE was associated with: a parent of which it was a part; and a start and end stage (the way in which time is handled in abstract formats). The next step was to regularize naming: previously, each tissue's unique name had been based on its path and there were, for example, > 100 associated mesenchymes. In practice, > 1500 renamings were needed to provide every AE with a unique name (ontologies require this so, for example, the name of the primordial germ cells (PGCs) has to change as they migrate through the various tissues along their path – see below).
Each organ system in the ontology was then examined in turn to ensure that its organization and partonomy were coherent, consistent and complete. For this, each material AE was examined in turn to ensure that: its relationship to other organ systems was included (e.g. the femur is now part_of both the lower limb and the skeleton); and the starts_at and ends_at stages were correct. It was also given an appropriate anatomical type term [an is_a or classification link to the Anatomical Entity Ontology (AEO)]. Leaf tissues were also linked via a develops_from relationship to an earlier parent tissue. Immaterial AEs, such as cavities and surface features, were linked within the partonomy via a located_in relationship, and their lineage was given via a develops_in relationship.
The starting place for the information in EHDAA2 was EHDA; the developmental anatomy for this (parts and timings) was primarily compiled by Matthew Kaufman (Hunter et al. 2003) on the basis of limited access to sections (see http://www.ana.ed.ac.uk/database/humat/notes/ for Kaufman's comments and references). Most of the EHDA terms had to be renamed to ensure uniqueness, several hundred had to be removed on grounds of redundancy, several hundred more had to be added for completeness, and many timings needed to be corrected (see below). The main sources for these changes were the key texts (Hamilton et al. 1972; Kaufman, 1992; Larsen, 2001; O'Rahilly & Müller, 2001, 2006; Standring, 2008). A few minor AEs were added on logical grounds; these were mainly leaf tissues, such as associated mesenchymes that were too small to justify mentioning in books. No automated annotation tools were used; all links were made manually. It is worth noting that EHDAA2, like all other anatomical ontologies, defines a simple tissue on the basis of standard morphology rather than by using molecular markers, which often gives an earlier start stage, but a boundary that is less precise because it may be marker-specific.
The ontology was assembled and viewed using the OBO-Edit program (Day-Richter et al. 2007); this automatically assigns an ID to all terms and allows relationships to be made through grab-and-drop procedures, with the user deciding on the nature of the relationship link. Files are stored in the OBO format, which is essentially a set of terse paragraphs or stanzas (Table 1) associated with each AE, that include lines for the ID and all the relationship links, together with ones for the occasional alternate ID from the Uberon (Mungall et al. 2012), flybase (Drosophila) and zfin (Danio rerio) ontologies and for synonyms. The OBO format is designed to be visually comprehensible and can be manipulated directly in text-editing programs (not WORD as the file is too long for the program to handle); this facility was used for manual checking and to save time when adding class links to standard sets of AEs (e.g. in the nervous system, each of the 39 ventricular layers is_a proliferating neuroepithelium).
Table 1. An example of an entry in an OBO file
|name: 6th arch mesenchyme from neural crest|
|is_a: AEO:0000146 ! dense mesenchymal tissue|
|relationship: develops_from EHDAA2:0001560 ! pro-rhombomere c neural crest|
|relationship: starts_at CS12 ! CS12|
|relationship: ends_at CS16 ! CS16|
|relationship: part_of EHDAA2:0004076 ! 6th arch mesenchyme|
|relationship: part_of EHDAA2:0004423 ! mesenchyme from rhombencephalic neural crest|
The current version of the ontology has about 2500 terms and about 14 000 links, and was not easy to check for errors. However, as each additional relationship was added, the earlier ones could be inspected on the screen and corrected if necessary. The final draft was checked by eye to pick up any obvious mistakes and then computationally for is_a classifications and timing inconsistencies (carried out by Mike Wicks); additional checks were performed by aligning EHDAA2 to the Uberon metazoan anatomy ontology (carried out by Chris Mungall). All errors so identified were corrected manually.
Viewing the ontology
To view EHDAA2 on a desktop computer, first download the EHDAA2 file in its OBO format. To do this, go to the obofoundry site http://www.obofoundry.org and right-click on the ‘human abstract version developmental anatomy, v2’. Choose the ‘save as’ option from the menu and choose a file location for saving it (e.g. as EHDAA2.txt on the desktop, in the first instance). The ontology can be viewed in either OBO-Edit or Protégé 4, but the former is much simpler to use, and the appropriate version can be downloaded from oboedit.org.
To display the required viewing windows in OBO-Edit, double click on the file name and first click on the ‘editors’ tab, second on the ‘Ontology Tree Editor’ (this shows the hierarchy) and then on the ‘Parent Editor’ (this shows all the relationships associated with a term); under the ‘search’ tab, click on the ‘Search Panel’. Finally, load EHDAA2 by clicking on ‘load ontologies’ (under the file tab of OBO-Edit); this opens the filing system and allows the user to click on EHDAA2.txt to open the file in the ‘Ontology Tree Editor’. This displays the top levels of the ontology hierarchy. Clicking on + and − buttons opens and closes sub-hierarchies.
Options for browsing on the web include OntoBee2 and BioPortal3.
- Top of page
The EHDAA2 ontology, like EHDA, covers the stages from CS1 (the fertilized egg or zygote) to CS20 (day 49). EHDA did not extend further than this as it is at about CS21 that a great deal of detail is filled in as, for example, the skeletal system starts to differentiate towards its final structure. A more difficult problem would have been reorganizing the hierarchies of maturing structures such as the brain. While it is not possible to put an exact figure on the extension of the ontology that would have been needed to include CS21, a rough estimate is about 25% in terms of new AEs. In short, it was not practical to include further CSs.
All anatomical ontologies are designed on the basis of the answers to four substantive questions: first, what is the granularity of the ontology (how fine is the detail); second, what tissues are to be excluded; third, is a tissue to be identified by an arrow pointing to a central point or by a boundary; fourth, how is one to identify migrating tissues where there is no central point and no boundary.
EHDAA2 attempts to include all AEs that are named and all of their obvious parts (e.g. their associated mesenchymes) down to the resolution of simple tissues composed of a single cell type, whether or not it has a well-defined boundary. The problem of how to handle tissues with boundaries that contain mixtures of cells (e.g. brain nuclei) is handled by treating the tissue as a leaf, and using the is_a classification as a means of handling more than one cell type (e.g. neuronal nuclei include both neurons and support cells). The obvious exclusions are minor blood vessels and nerve fibres, while the many minor mesenchymal condensation within looser mesenchyme that are seen in the later-stage embryos have also not been included, mainly because, as their fate is unclear, it is not possible to name them. The difficulty in handling mesenchymal condensations extends further: they have no obvious boundaries and the default convention for this ontology is that a tissue is included if it can be pointed to – many tissues (e.g. the many regions of the heart) are named (e.g. the septa) where their boundaries cannot be recognized (these boundaries probably do not exist in any biological sense). Migrating cell populations are placed as part_of their location at a particular stage (migrating PGC populations are placed in different tissues at different stages – see below). The fact that they are migrating is also handled via their classification (e.g. is_a migrating mesenchymal population).
Visualization of the ontology
OBO-Edit displays the EHDAA2 ontology as a set of major hierarchies (for the anatomy, for CSs, for the AEO classification and for a subset of the cell-type ontology that has been imported; Fig. 1), together with two essentially irrelevant minor hierarchies (for the relationships used and obsolete terms). The only child of the anatomy hierarchy is the conceptus, and this is split into the zygote (CS1), the morula (CS2), and embryonic (CS3–20) and extra-embryonic parts (Fig. 1). Both of these latter hierarchies open to give a host of AEs that are structured around the part_of and develops_from relationships. The timing relationships are not shown directly: each AE has its own start and end stage (available in the ‘Parent Editor’ and which is narrower or the same as the tissue of which it is part). Visualization of the tissues present in individual CSs (currently being planned, see Discussion) will simplify things: for early stages, the organ system and limb hierarchies will be excluded; for later stages, all the early tissues will have been lost to leave the organ systems, the limbs, the early coelomic cavity and (up to CS14) the neural crest.
Figure 1. An OBO-Edit screen shot showing the ontology tree editor (the EHDAA2 hierarchy), the search panel (choice = epiblast) and the Parent Editor (showing all the parents of epiblast and their relationships). The ontology shows the top-level partonomy to display how the anatomy is laid out; the organ system group and the embryonic membranes have been expanded to show the next level of parts.
Download figure to PowerPoint
Material AEs (this term excludes entities, such as surface features and spaces that are known here as immaterial AEs) are viewed as being part_of larger structures. Unlike the FMA ontology of adult human anatomy, which has a range of part_of relationships (Rosse & Mejino, 2008), EHDAA2, like most other anatomical ontologies, has a single part_of relationship that covers two rather different meanings: first, an included part with a functional relationship (the femur is part_of the skeleton); and second, a natural member of the class of objects that is distributed (the thymus is part_of the glandular system). A third meaning is included_in (e.g. the bone marrow is part_of the femur); EHDAA2 mainly uses this latter relationship for immaterial AEs (an exception is the partonomy of the eye).
The part_of relationship was used in EHDA to locate a tissue to the organ system in which it had its major functional relationship, and this meant that each term had a single parent in a hierarchy appropriate to that CS. The part_of relationship in EHDAA2 is used widely, and many anatomical tissues have two or more part_of relationships. Some of these are because a tissue can be viewed as part of several different structures, others are for user convenience: the main hierarchy of the gut, for example, is from the oral region through the foregut to the anus and each region is handled separately. For convenience, there is a separate heading that groups all the main gut tissues (epithelium, mesentery and mesothelia); the heart and brain similarly have a dual organization that allows the user to choose whichever is more convenient.
Very occasionally, the attached_to relationship was used where the part_of relationship seemed inappropriate. Two examples are the brain arteries that locally come together in the circle of Willis, and the extra-embryonic and embryonic vascular systems. Unlike all the other relationships used, attached_to is reciprocal (if A is attached to B, then B is attached to A), and this dual link implies the possibilities of a loop in the ontology. As loops can cause complications, the reciprocity possibility was explicitly avoided: thus, while the smaller extra-embryonic vascular system is detailed as being attached_to the larger embryonic vascular system, the former was not linked to the latter; similarly, the cerebral arteries are attached_to the circle of Willis, but the reciprocal relationship is not given.
Making the partonomies for most organ systems turned out to be difficult because they had to allow low-level terms to differentiate from stage to stage within a hierarchy where the upper-level terms made anatomical sense, and this sometimes required the use of semi-artificial AEs (e.g. the natural development of the hindbrain rhombomeres requires the presence of a future myelencephalon; Fig. 4 below). One thing that was interesting here was that, while early drafts of the partonomies for individual organ systems were incoherent, the hierarchies became simpler and clearer as detail and timings were added and upper-level AEs from separate stages were rationalized; achieving this clarity gave an indication that a reasonable structure was in place. The final results are sometimes still a little cumbersome, but they should be clear.
Immaterial AEs: cavities and surface features are embedded within the partonomy through located_in relationships, with lineage relationships being represented as develops_in links. The located_in relationship is also occasionally used to place tissues present within cavities (e.g. the reticulum in the chorionic cavity, the various tissues within the coelomic cavity, and the blood vessels in the vitreous cavity of the eye), and this is one of the few places in the ontology where an attempt has been made to include a geographic location. This formalism does lead to the occasional oddity such as the diaphragm being located_in the early coelomic cavity even though it is a solid tissue with several sources; here, the partonomy of the diaphragm is really best understood via the lineage links.
Choosing the start and end CSs for each tissue was one of the hardest tasks in making the ontology. The original timings came from EHDA and were based on a single embryo at each stage; many of these have been superseded by those of O'Rahilly & Müller (2001, 2006) who looked at many embryos. There are, however, tissues in EHDAA2 whose staging was not mentioned in either of these resources. As it was not practical to explore the histology in detail, some timings had to be chosen on a logical basis. It was, for example, assumed that all somites were present for a single CS before they broke up into sclerotomes and dermomyotomes, which each took some four CSs to differentiate (recognizing a particular somite or dermomyotome in a section is not usually possible so any ambiguity in the ontology is unlikely to lead to serious errors). Where there were no direct clues from the literature about the start and end stages of a particular tissue, these were worked out on the basis that a transitional tissue could not form until its parent was lost and would itself be lost when its descendents were in place.
A problem that has no good solution is how to handle the timings of a tissue whose start and end stages depend on location (e.g. gastrulation). An exact solution would require that the tissue be artificially segmented by stage boundaries, but this would be hard to handle as there are no physical boundaries to mesh with the temporal ones. The solution adopted was to assign a start stage when the event first happens somewhere and an end stage when the tissue change has been completed everywhere.
Timings cannot of course be given a unique CS with confidence for two reasons. In early development, change is rapid and many events can happen during the course of a single CS. EHDAA2 handles this by including develops_from links within a single stage so that at least the order of development is explicit. For later development, O'Rahilly & Müller (2001) report that the first stage at which a tissue can be noted in human embryos often ranges over three CSs, and occasionally by as much as four stages (e.g. the epiglottis). EHDAA2 represents a single idealized but typical developing human embryo, and it is hoped that the start and end CSs are accurate to within a stage either way. Readers who notice a timing that falls outside these bounds should send details to email@example.com for correction.
For technical reasons, formal ontologies require a classification for each entity (given through an is_a relationship), and for anatomical ontologies this has, for the past few years, been handled using the Common Anatomy Reference Ontology (CARO; Haendel et al. 2007); this essentially assigns a class on the basis of the complexity of that AE (from portion of tissue up to organism). However, the relatively few terms that CARO includes are of limited use if one wishes to capture something of the richness of anatomy. The solution adopted here has been to use the Anatomical Entity Ontology or AEO (Bard, 2012a), a substantial expansion of CARO that includes, for example, ~ 80 simple tissues (those containing a single cell type) and a wide variety of immaterial AEs (e.g. pits, rows, lumens, cavities) rather than the four (based on dimension: point, line, etc.) in the CARO.
One important new hierarchy in the AEO classifies developing AEs, albeit that its use is slightly problematical in the sense that all tissues in an embryo are developing. The solution adopted here has been to restrict the use of these terms to tissues that were transitory in the sense that they were migrating or had a short period of existence (e.g. mesenchymal condensations).
The other problem in classifying tissues is that one tissue can easily have several classes assigned to it: an epithelium can be both columnar and a tube; a gland could be an organ or a multi-tissue structure (to use CARO terminology). The general solution adopted has been to choose what appears to be the single most appropriate class for a tissue, allowing a second class if it seemed that this would be intuitively useful. In practice, a class seems least useful where there are many members (e.g. multi-tissue structure) and most helpful where there are few (e.g. neuronal nucleus).
There is one other feature of the AEO worth mentioning: each of its ~ 80 simple tissues has a link (has_part) to its main constituent cell type as categorized in the cell-type ontology (e.g. cartilage condensation has_part chondroblast). These links allow a user to assign histological knowledge to the anatomy through upwards links: the cell types in a multi-tissue structure are those in its constituent simple tissues. This feature allows a user to follow cell-type transitions over time, albeit that this is easier to do computationally than manually.
The aim here was that every leaf tissue could be traced back to the zygote, and the information on which this part of the ontology is based comes from the literature on mammalian development. This is mainly on the mouse (e.g. Kaufman & Bard, 1999), but there is no reason to suppose that, once embryogenesis is under way, i.e. CS5 when the inner cell mass is partitioning, there are any lineage differences between mouse and human embryos. Because every higher-level tissue is the sum of its leaf tissues, the develops_from links make it possible, in principle, to follow the development of every AE back in time by tracing the development of its parts. In terms of visualization, however, this currently has to be done by hand as the ontology shows children rather than parents [the descendents of the neural crest are shown in Fig. 2; another clear example are the descendents of the ectoderm (not shown)]. A means of providing the lineage of an AE directly is currently being planned (see below).
Figure 2. The neural crest lineage in EHDAA2. OBO-Edit shows the develops_from relationship as a white D in a dark blue diamond. See Table 1 for an example of the full details for a neural crest derivative.
Download figure to PowerPoint
Lineage is one area of the ontology that will certainly need curation, as there are areas such as haematopoiesis where the full details are still not in place (Medvinsky et al. 2011).
The major organ systems
It is not practical to go through the whole ontology describing every decision that has been made. The summary below is, however, intended to indicate some of the major features, while Kaufman's referenced notes describing how decisions were made for the original EHDA ontology, together with details of the early literature, are available at http://www.ana.ed.ac.uk/anatomy/humat/notes/.
Early development (CS1–9) and extra-embryonic tissues
The partonomy and lineage for these early stages of embryonic and extra-embryonic development follow those given in O'Rahilly & Müller (2001). In EHDA, the development of most extra-embryonic tissues was only given up to CS9: EHDAA2 now includes tissues and timings up to CS20, with the extra-embryonic membranes in particular having been completely reorganized (Fig. 1). It is worth noting that the cavities of both the extra-embryonic and embryonic parts of the conceptus are linked using the located_in and develops_in relationships. One area where timings have had to be estimated is the development of the haematopoietic system, as the stage at which the yolk sac ceases to be a source of stem cells is not known. Following a comment from O'Rahilly & Müller (2001) on the activity of the yolk sac (or umbilical vesicle), this has been set at CS16, but better evidence is needed.
The most complicated system that develops in these early stages is the neural crest, which forms from all parts of the early neural tube except the telencephalon. The various neural crest domains are primarily located under the region from which they originate, but, for convenience, they have also been given their own heading in the hierarchy. This allows a user to follow its future development as a whole (Fig. 2). This heading illustrates the strengths and weaknesses of the ontology: a strength is that a lot of material is grouped together in a terse and readable way; the weaknesses include the lack of spatial resolution in the listing of, for example, the spinal ganglia and the lack of detail in the enteric and peripheral nervous systems, a reflection of the difficulty of following their early stages. In contrast, the development of the somites is straightforward and detailed, with the development of each somite being specified, albeit on the assumption that all behave in a regular way.
The development of the heart in the pericardial cavity is rapid: the early tissues are in place and the primitive heart tube start to form at CS9, while the first steps in the development of the multi-chambered heart are apparent at CS11. The basic partonomy shows the initial tissues together with these two terms (Fig. 3), with the latter being particularly complicated. As mentioned earlier, the cardiac system has a dual organization so that a user can either look at the geographical parts or examine the cardiac tissues as a list (Fig. 3). Unfortunately, OBO-Edit displays entities alphabetically, and it is not possible to list tissues in an anatomically sensible order.
Figure 3. An OBO-Edit screen shot showing features of the heart partonomy. In the lower part of the picture, the primitive heart tube gives the early tissues; the upper part includes the hierarchies for the endocardium and the myocardium.
Download figure to PowerPoint
It should be mentioned that there is a parallel ontology of heart development that is part of the Gene Ontology or GO4 (Khodiyar et al. 2011). That ontology is very different from the one described here as it views heart development as a series of processes: it carries no develops_from, lineage or timing relationships, and relatively limited anatomy. Another difference is that EHDAA2 is focused specifically on humans, but GO covers all organisms. Thus, some of the broader heart development terms in GO are applicable to development of the Drosophila dorsal organ, and the more specific terms are applicable across vertebrates. Because there is little relationship between the two descriptions and it is hard to identify many EHDAAA2 heart tissues with those in the GO, EHDAA2 includes no GO IDs for the heart. It is, however, worth pointing out that the generic process-oriented description in the GO is unified with the human-specific anatomical description in EHDAA2 in Uberon anatomy ontology using ontology cross-products (Mungall et al. 2011).
EHDAA2 partitions this into the oral and gut regions with an intervening buccopharyngeal membrane (up to CS11). The former includes the jaws and tongue, the latter covers the various regions of the gut epithelium and associated tissues. Here, the most complicated domain is the anal region of the hindgut, partly because it has an external region that includes the area of surface ectoderm in the vicinity of the anus, and partly because its internal region has to handle the development of the cloaca with its septation into the bladder (which is also part of the urinary system) and the rectum.
The alimentary system has many parts, and its standard division into fore-, mid- and hindgut makes it hard to keep a sense of the fact that these regions are continuous. EHDAA2 tries to compensate for this by having a dual hierarchy: one for the gut regions and another for the gut tissues: in this latter hierarchy, there are sub-hierarchies for the epithelium, mesenteries and mesothelia. The user can choose whichever partonomy is more convenient.
Urinary and reproductive systems
Because the reproductive tissues are initially an integral part of the early urinary system, they are included in this hierarchy up to CS14. The reproductive system is initiated as a separate hierarchy at CS15 when the indifferent gonad is first apparent; it separates into the male and female reproductive systems at CS18 when morphological differences are first apparent. The external genitalia (the fold and tubercle) remain at the indifferent stage until CS20, and thus show no sexual dimorphism over the period of the ontology.
Because ontologies require each distinct term to have a unique name, handling the migration and development of the PGCs was particularly complex as these cells are part of several different AEs: there are thus terms for the yolk sac PGCs (CS6b–10), the allantoic PGCs (CS11), the hindgut PGCs (CS12–13), the mesonephros PGCs (CS13), the gonadal ridge PGCs (CS14), the indifferent gonad PGCs (CS15–17), and the male and females PGCs (CS18–20). Each of these is classified as a reproductive cell population, while the migrating PGCs (allantoic and hindgut) carry a secondary is_a migrating developing tissue link.
The development of the spinal cord and peripheral nervous systems is relatively straightforward, unlike that of the brain, which is extremely complicated and still not fully understood (Bystron et al. 2008): the main technical problem in making the ontology was to keep track of the neuromeres and their child tissues in the prosencephalon, mesencephalon and rhombencephalon, and their descendents, the forebrain, midbrain and hindbrain. The way that the ontology handles this complexity in, for example, the hindbrain is shown in Fig. 4: for coherence, it turns out to be necessary to include early pro-rhombomeres (CS9–11) and early rhombomeres (CS12–13) in the rhombencephalon, rhombomeres 1–3 in the future metencephalon and rhombomeres 4–8 in the future myelencephalon (CS14–15). The rhombomeres then break down in both the metencephalon and myelencephalon, with the former giving rise to the pons, isthmus and cerebellum (CS16), and the latter forming the medulla oblongata at CS17 (Fig. 4). Similar events take place as the fore- and midbrain develop.
Figure 4. The components of the rhombencephalon and early hindbrain show, in particular, the development of pro-rhombomeres (CS9–11) to early rhombomeres (CS12–13) in the rhombencephalon, to mature rhombomeres (CS14–15) in the future myelencephalon and metencephalon. The different names are needed because ontologies require time-dependent tissues to be distinguishable.
Download figure to PowerPoint
One novelty in EHDAA2 as compared with EHDA is the partitioning of the cerebral cortex into the archi-, neo- and paleo-pallium; here, the spatial and temporal information in O'Rahilly & Müller (2006) has been used. Information on their differentiation and that of the other parts of the brain neuroepithelium is taken from EHDA and reflects a traditional view of early neuronal development: early on, each region forms a roof plate, basal plate and lateral wall, with the latter soon becoming the ventricular layer (is_a proliferating neuroepithelium) whose cells give rise to the mantle layer (is_a differentiating neuroepithelium) and marginal layer (is_a neuronal white matter).
- Top of page
This paper reports the availability of a formal description of human development covering the first 7 weeks of embryogenesis (CS1–20). This description includes a complex partonomy, lineage data, temporal information and a formal classification of anatomical tissue types that incorporates cell-type detail. The intention has been to capture much of what is known about the morphological details of early human development, and the ontology contains some 14 000 facts of the general form <tissue><link><parent>.
The granularity of the ontology is intended to be at the level of the simple tissue, broadly defined as a group of cells whose great majority has the same morphological phenotype and usually with defined borders. This requirement could not be strictly adhered to: the early embryo includes populations of migrating cells that move through other tissues and mixtures of cells differentiating along different trajectories; it also includes tissues where the boundaries are artificial in the sense that the tissues on either side of the border have the same cell phenotype (e.g. the septa and tissues of the heart). Nevertheless, this definition of a simple tissue generally works, except where the obvious boundaries include more than one cell type (e.g. neuronal nuclei include both neurons and support cells). In short, the use of the term ‘simple tissue’ is a bit loose, but unlikely to cause any misunderstanding.
The most contentious part of the ontology lies in the timings, as these represent an idealization of what might be found in practice. O'Rahilly & Müller (2001, 2006), have looked at many embryos at each CS, and find that the stage at which a tissue is first apparent may, as mentioned earlier, vary over as many as four CSs. For EHDAA2, where precise choices have had to be made for the start and end stages of many tissues (ontologies do not handle ambiguity!), all that can be said is that one basic rule has been strictly followed: where tissue B forms from tissue A, tissue A forms before tissue B. In early development when change is particularly rapid, both tissues may be present at the same stage, but the timing is made clear via the relationship tissue B develops_from tissue A. While it is hoped that EHDAA2 matches the typical or average embryo, users may find that when they compare theirs with EHDAA2, stagings may differ by a stage (occasionally, even two stages).
As has been mentioned earlier, the lineage data for simple tissues is mainly from the mouse, but there is no reason to suppose that once the inner cell mass has formed, there is any difference in the lineages of the two embryos. The early developments of the two are of course quite different, and the lineages used for the extra-embryonic tissues of EHDAA2 have been taken from O'Rahilly & Müller (2001).
In formal ontologies, the key relationship is the classification, but, although most anatomical ontologies structure themselves around the CARO, this has little practical use: its 16 or so useful terms (e.g. multi-tissue structure) include so many examples that it cannot be used in a helpful way. EHDAA2 uses CARO IDs where appropriate, and AEO IDs to classify AEs at a more detailed level. The groupings under these classification terms allow a user to identify tissues with a similar phenotype across a range of systems and organs that would not usually be grouped (e.g. neuronal nucleus, ganglion; Fig. 5a,b). This facility may be helpful in looking for common patterns of gene expression (Bard et al. 2008). Should the CARO community decide to expand their ontology to include all the tissue types used in EHDAA2, the intention would be to add these additional IDs to EHDAA2.
Figure 5. OBO-Edit screen shots of the classification hierarchy. (a) the ganglia; (b) the neuronal nuclei. The Parent Editor shows the timing and other relationships for each of these tissues in another window.
Download figure to PowerPoint
Ontologies are not intended as academic exercises, but are meant to be used. For anatomical ontologies, the most common role is to structure databases for holding tissue-associated data; examples are the gene-expression patterns in zfin for the zebrafish (http://zfin.org/) and GXD for the mouse (http://www.informatics.jax.org/expression.shtml). In the latter case, the Mouse Atlas image resource (http://www.emouseatlas.org/emage/) makes access to GXD simple: clicking on the image of a mouse tissue at a particular stage activates a program that sends the ID to GXD, which returns the gene-expression profile of the tissue at that age. These IDs have a further use in that they provide baseline keys for mutants: an abnormal tissue can be annotated using PATO (Phenotype and Traits Ontology; Gkoutos et al. 2009), but additionally annotating the file with its normal anatomy ontology ID allows for automatic linkage to any associated data. Anatomical ontologies also have roles outside the bioinformatics community: they include a great deal of information that can be of use to developmental biologists, often far more than can be found in the literature and in textbooks. The problem here is that such scientists are often unaccustomed to working with the informatics tools needed to display the information that the ontologies contain.
This will certainly be the case with EHDAA2, although it is hoped that the figures in this article give some indication that using a browser such as OBO-Edit is not particularly tricky. However, while the various OBO-Edit windows show lineage, the essential partonomy and the links associated with a particular AE, the browser does not allow a user to follow cell-type changes, analyse the tissues present at a particular stage or look at stage-specific development. Some of this can be done by taking detailed notes as one browses, or by opening the EHDAA2 file in a text editor (not WORD – see above), choosing a ID and seeing where it is used. These procedures are, however, unsatisfactory, and the file really needs a website with a functionality sufficiently sophisticated that it would allow users to see the tissues present at each CS, to follow their development backwards to the zygote and to analyse cellular differentiation. Plans are currently being discussed with the MRC Human Genetics Unit for doing this, and the intended website may be subsumed within the Human Developmental Studies network (http://www.hudsen.org/).
One further use of the ontology is that specific aspects of tissue development can be taken from the ontology and used as the basis for a more detailed graph to which can be added associated molecular data with appropriate links (e.g. signals_to). In this way, a rich picture of a developmental event can be produced (e.g. Bard, 2011) that integrates anatomical development with the underlying molecular events that drive change, and particularly the signalling pathways and limited set of downstream processes (e.g. differentiation, growth, apoptosis, migration) that they initiate (Bard, 2012b).
Finally, it needs to be emphasized that ontologies are not, as originally thought, stable representations of truth, but need to be updated to correct the inevitable errors that they contain and to incorporate new information, and for EHDAA2 this particularly applies to the developing brain (e.g. Bystron et al. 2008). Indeed, as ontologies are community resources, they need to represent what the community needs, and this may include further relationships such as connected_to. Users are invited to send corrections and suggestions to firstname.lastname@example.org, and this address can be used for the next 2–3 years. The intention is that curation will then be handled via the new human development database mentioned above. Improved versions of EHDAA2 will be posted on the obofoundry.org site, with details of the associated changes being made available on the EHDAA2 site5 associated with the obofoundry.