2. Outline
Heart Morphogenesis – what happens?
Information Modelling – what do we mean?
Multiscale Modelling - how do we do it?
Conclusion and Future Work
9. Outline
Heart Morphogenesis – what happens?
Information Modelling – what do we mean?
Multiscale Modelling - how do we do it?
Conclusion and Future Work
10. Information Modelling – what do we mean?
The concepts, relationships and constraints in
a domain - these are ontologies
Use the same information models across
multiple data sources
Thus we know how one piece of information
relates to another
17. Outline
Heart Morphogenesis – what happens?
Information Modelling – what do we mean?
Multiscale Modelling - how do we do it?
Conclusion and Future Work
18. Multiscale Modelling – how do we do it?
Different types of computational model are
suitable for different levels of biological scale
E.g. Biochemical reactions can be
represented as networks or ODEs. Cellular
behaviour can be modelled with agent based
models.
Use models at one level of scale, to pass
information to models at another level of
scale
19.
20.
21. Conclusion and Future Work
Gene to phenotype annotation tends to use a
surgical or anatomical perspective – but we
know very little about mechanism or causes
Post-compositional annotation allows a link
between multiscale measurement and
multiscale modelling
Future plans include scale-linking between
SBML models and agent based models, for
simulation of Endocardial Cushion Growth
For more information, visit: http://www-staff.lboro.ac.uk/~elta2/index.htm
1) At an early stage the heart is a tube. It has no valves but is somehow pumping blood in one direction, perhaps using impedance pumping. In the third and fourth week of human development, heart looping takes place, and we end up with the four chambered double pump. Let’s look at this in more detail.2) One of the key things that happens is heart looping and wedging. There’s a lot of jargon here, so let’s break it down a little bit. The long thing, which is looping behind is known as the Outflow Tract (OFT) or conotruncus. You can divide it into two parts the conus and the truncus. Now while it is doing this looping, the Atrioventricular canal is septating (this means a wall is growing to divide it). Also, a septum is growing in the conotruncus. Because the conus is looping, the shape of this septum is spiral. When the OFT wedges into the AV canal, in normal development the conal septum lines up with the AV septum. This allows blood to flow from right ventricle to pulmonary artery (to the lungs) and from left ventricle to the aorta and around the body.3) In this bottom image we see two Scanning Electron Micrograph (SEM) images of mouse hearts during looping. The first one has looped in the correct direction – clockwise, while the second one has looped in the direction – anticlockwise. In this case, we have genetically induced situsinversus – where the organs develop on the opposite side of the body. Images: [1] http://www.helmholtz-muenchen.de/en/ieg/group-functional-genetics/deltanotch-pathway/index.html[2] Kirby et al. Cardiac Development[3] J. Schleich, C. Almange, J. Dillenseger, and J. Coatrieux, "Understanding normal cardiac development using animated models," IEEE Computer Graphics and Applications, vol. 22, 2002, pp. 14-19.
So... This re-modelling of the Outflow Tract is a crucial part of heart development... Among CHD this is the most common thing that can go wrong. But there are several mechanisms that can disrupt this remodelling – either individually or in combination. Contribution from the Second heart field can lead to a shortened OFT, which causes shorter rotation. Neural crest cells migrate from the neural tube to the heart tube, and abnormal migration of these lead to septation defects. The cardiac muscle (myocardium) itself may have some defect, which causes incorrect rotation. Finally the endocardium may undertake abnormal Epithelial-Mesenchymal Transisition. This will affect the growth of the endocardial cushions in the OFT and the AVC, causing septation and valve defects.Image:F. Bajolle, S. Zaffran, and D. Bonnet, "Genetics and embryological mechanisms of congenital heart diseases.," Archives of cardiovascular diseases, vol. 102, 2009, pp. 59-63.
We have noted that one genetic disruption may lead to multiple types of CHD, and one type of CHD might be caused by many different genetic mutations. With this image we can gain further insight as to how CHD sit on a spectrum. We have different disease classifications corresponding to different degrees of rotation.In the normal situation the OFT should rotate about 150 degrees, clockwise.So... Diseases are usually classified by their symptoms, but in reality diseases can have common causes, and even be overlapping. The classification of Double Outlet Right Ventricle (DORV) overlaps with that of Tetralogy of Fallot (TOF) with rotation varying from about 90 to 140 degrees.Image: after L.F. Donnelly, "Adapting disease concepts to changes in imaging modalities in complex congenital heart disease Imaging and staging of Wilms ’ tumour," Pediatric Radiology, vol. 27, 1997, pp. 284-285.
Because tissue ends up in quite different places, a defect in one place, at one stage of development leads to defects in different places later on. The endocardial cushions in the AVC end up as the atrioventicular valves (blue). Those in the conotruncus end up as the semilunar valves and the membranous septum (yellow). This is the most common location for a ventricular septal defect. Some of these names things have real boundaries, and others have only fiat boundaries. The names and locations are quite disputed, and this is one case in which information modelling – and more specifically ontologies – will come into play.Image:D. Srivastava and E.N. Olson, "A genetic blueprint for cardiac development.," Nature, vol. 407, 2000, pp. 221-6.
This is essentially what happens during Epithelial to Mesenchymnal Transition (EMT) in the heart.To begin with you have a field myocardium, underlying cushion forming regions, which has lower expression of VEGF.
Here we are looking at the same thing, but closer up. We can see the intracellular signalling that takes place within a single cell.In the myocardium VEGF signalling induces the expression of Calcium ions In the Endocardium... Notch signalling has pre-sepecified cells that may undergo EMT. In combination with BMP and TGF-beta signalling from the myocardium, this increases the expression of Snail protein, which in turn reduces the expression of VE-Cadherin. This causes endocardial cells to lose their adhesiveness and undergo EMT, which leads to growth of the cushions.We will return to this later, when I talk about multiscale modelling.
All OBO ontologies map to the Basic Formal Ontology (BFO). Thus this constitutes a basic upper ontology. The idea is that everything can then exist under a common framework, and all these different, specialised, ontologies are compatible.‘Entity’ is the root term of anything. Then this is the most basic distinction we can make.Independent continuants are entities that exist through time. Dependent continuants are qualities that inhere in either independent continuants or in occurents. They are called ‘dependent’ because, for example, you can not have a “Red” only a “red eye” or a “red cell”, or a “Discontinuous Growth”. Occurents are processes.You may be wondering. What is the difference between Morphogenesis, growth and Development? Well growth simply means something increasing in size, could be a cell or an embryo, or a child. Developmental Growth is a type of growth, where the increase in size is a result of progression over time. Morphogenesis is a type of Developmental Growth, that takes place when anatomical structures are taking their shape.I am going to run through some different examples of ontologies now, as this is a good way to introduce how all this works.
Source:AmiGO. What we are looking at, at the top here is a part of the Gene Ontology. This is one of the most sucessful biomedical ontologies. It is split into three branches: Biological Process, Cellular Component, and Molecular Function. We are showing Biological Process – which is at the top there. Now as we go down the tree, and get more specific, fewer gene products are associated with each term. Until eventually, there are only 5 gene products that have been associated – somewhere in GO annotated biomedical literature - with “endocardial cushion formation”, GO term 3272And these are those 5, essentially they are all Bone Morphogenetic Protein, but in different species.
So... For example, here is a part of the Mammalian Phenotype Ontology (MP). Everything in the MP is a phenotype that mammals can have – hance the root node is “Mammalian Phenotype”. Everything under this is a subtype of “Mammalian Phenotype”. The arrows here are “is_a” relations. This means we go from broader to narrower terms. Shown here are some phenotypes that are types of “abnormal heart development”. Some child nodes are hidden to keep it tidier. We have phenotypes like “abnormal looping morphogenesis”, which could be delayed heart looping, abnormal direction, or failure of looping.Transposition of great arteries – by itself ok, but there are lots of other chd phenotypes associated with abnoral degrees of OFT rotation.We have “abnormal outflow tract development” and in this case, the only child nodes are abnormal septation or ‘transposition of great arteries’ – as we have seen already this can mean many different things. And then there are things like dextrocardia – where the heart develops to the right, or abnormal EMT. And under abnormal endocardial cushion morphology, well they can be absent, decreased in size increased in size thin – or there can be a failure to close.Now this may be called a pre-composed phenotype ontology. The curators attempt to populate it with all the phenotype terms that might ever be applicable, and in some circumstances this can be helpful. But there are always going to be situations where you just don’t have terms to the required specificity. Getting new terms added to an ontology might be a somewhat cumbersome process. For example... When this ontology defines ‘decreased size of endocardial cushion’ are we talking about the endocardial cushions in the AVC or in the OFT? This is a real problem, sometimes an induced genetic mutation leads to an increase in one set of endocardial cushions, but not in the other.A better approach might be post-composition. We can see that terms in the MP consists of a relation between an anatomical entity – say ‘endocardial cushion’ and a quality “decreased size”Ontology view from: http://bioportal.bioontology.org/
So here is a view of part of PATO – PATO deals only with qualities – things like size and all the many subtypes of size.This can only really be used in a post-composition approach – in which we combine these terms with terms from other ontologies such as anatomical ontologies. “decreased size” doesn’t make sense unless we are talking about the decreased size of something.Note that we have more options here. The MP ontology stopped at “decreased size of endocardial cushion” – but if we were to use this ontology, in a post-composition approach we could specify whether it was hypoplastic (fewer cells) or hyopotrophic (smaller cells). The same is true of increased size with hyperplastic and hypertrophic).Ontology view from: http://bioportal.bioontology.org/
Here is a view of a species-specific developmental anatomy ontology – the EHDA. Note that this is primarily orgainsied along part_of relationships, as this is generally easier from an anatomical perspective. We also have starts_at and ends_at relationships, which link to entities representing the Carnegie Stages (CS) of human development. The heart starts developing at CS06 and ends at CS20. The AV canal and OFT start at CS10.
Now here we demon an example of post-composition . We define a new term as an intersection entity that has_quality of being “mislocalisedradially” and inheres_in the outflow tract.Because the “aortic component” is part of the “outflow tract”, an automatic reasoner could infer that “outflow tract”+”mislocalisedradially” has a part “aortic component”+”mislocalisedradially”.Now even from this simple example, we can see that we are starting to work across multiple scales,
We’ll need a moment for this diagram. We are showing the levels of scale that can be represented. Different levels of spatial scale correspond to different levels of Temporal Scale.We are demonstrating that here for EMT. Individual signal transduction in a cell influences cell behaviour, which leads to a tissue transformation, which affects overall heart tube morphogenesis.There are particular modelling approaches applicable to each level of scale. From pathway models at the subcellular level, to reaction-diffusion equations at the cellular level, agent based models at the multicell/tissue level, and finite element models at the developing organ level.The biological modelling community have been developing interchange formats which allow models to be shared between different tools. These are XML specifications. For example, SBML allows the sharing of biochem reaction networks, CellML of biophysical cellular models. FieldML of finite element models. Cell Behaviour Markup Language is still being defined, but this would allow sharing models of cellular behaviours.Because the interchange formats are XML, they are compatible with ontologies. Which brings us to the bottom of the diagram. The “ontology-space” that is needed here is divided between Independent continuants and dependent continuants, which concern the spatial scale, and Occurents which concern temporal events. Different ontologies are used for annotation at different levels of scale. E.g. FMA for organ level, Cell Type (CL) ontology for cell level, Protein Ontology (PRO) for the protein level.Now, one particular challenge is how to link between terms in these different ontologies, which represent different levels of scale. Doing this brings us part of the way to understanding where different models and model components lie in relation to one another.
Now... So far we have talked about models being stored in XML, which allows for easy organisation with ontologies. But results and data, can also be stored as XML, in web accessible databases. These include both results of biomedical measurements, and the results of simulations of biomedical models.Along the left we have data sources at different levels of scale. These can be annotated, for example with segmentation of MRI images using the FMA. Doing this allows automatic generation of realistic 3d models of organs. Histochemical data can be annoted with the Protein Ontology and the CellularComponent (GO-CC) in which they are located. Gel electrophoresis data can be annoted with GO-Molecular Function and the Protein Ontology.We can build composite annotations. These same composite annotations can then be used for annotating Biosimulation variables, parameters and modules. Finally there are disease classifictions, such as “Ventricular Septal Defect” which might be inferred from a decreased volume of the membranous portion of the cardiac septum.
…we can tell what diseases are related to particular genetic mutations but know very little about the causes or mechanisms.