Opendata, web and dolomites


Teaser, summary, work performed and final results

Periodic Reporting for period 2 - SIMBIONT (A data-driven multiscale simulation of organogenesis)


The goal of SIMBIONT is to build the first full computer simulation of a paradigmatic example of organogenesis – mammalian limb development. Understanding the genes, signalling pathways and molecular networks that underlie organogenesis has enormous potential impact, both...


The goal of SIMBIONT is to build the first full computer simulation of a paradigmatic example of organogenesis – mammalian limb development. Understanding the genes, signalling pathways and molecular networks that underlie organogenesis has enormous potential impact, both scientifically and medically.

Scientifically, it is a prime biological example of complex multi-scale control – in which macroscopic and microscopic phenomena feedback to control each other. In particular, the macroscopic state of the system (at the organ scale) feeds back to control at least two types of microscopic cellular decisions: Firstly, during development cells constantly make choices about which cell type to adopt (for example bone cells versus muscle cells). Secondly, cells also continually make choices about which physical movements to make – to migrate, contract, divide or die.

At the heart of these decisions are many hundreds of genes and proteins wired together into cellular networks (or “control circuits”). The dynamically changing states of these cellular networks reflect the decisions being made, but this process cannot be understood by molecular analysis alone. As a cell’s position changes during morphogenesis, the range of signals it receives from neighbours also changes, simply as a consequence of these geometric rearrangements. Thus, genes control cells, but cell movements equally control genes, creating a multi-scale feedback loop. Such complex feedback systems can display very non-intuitive behaviour, and we are therefore still far from understanding how organs are reliably constructed. To do so will require integrating many types of data and sophisticated computer modelling, and thus represents a scientific grand challenge.

Medically, developmental genes and networks are central to: (a) human congenital abnormalities – such as heart defects or polydactyly (which affects 1 in 500 births), (b) cancer - many pathways of interest for tumor control were discovered as examples of de-regulated development, (c) stem cells - the process of organogenesis is essentially the control of large populations of mulitpotent progenitor cells, and most excitingly (d) regenerative medicine - these pathways clearly underlie the promise of tissue and organ regeneration.

We chose limb development for the SIMBIONT project because it is the most tractable example of mammalian organogenesis. It was already studied intensively by embryologists in the 1940s and 50s, long before molecular biology entered the field, which encouraged the development of strong conceptual frameworks regarding “organisers” – regions of tissue which secrete diffusible signals to coordinate growth and patterning. Studies of limb development have thus contributed some of the key principles to the field, which remain invaluable today. Since the 1980’s this conceptual framework has been complemented by a wealth of molecular data, and experimental genetics from the mouse. Over 1,800 mutant mouse strains have been documented with defects in limb develoment, and the ability to generate sophisticated genotypes now means that double or even triple limb-specific conditional knock-outs have become increasingly common within the field. Finally, it is clear that both the principles of multi-scale coordination and the specific genes involved will be very relevant to many other organ systems.

The key objectives of SIMBIONT can be organized into the following points:

1) The primary objective is to close the multiscale loop – to create a computer model in which Gene Regulatory Networks (GRNs) and global tissue movements dynamically control molecular patterns, which in turn feedback to control tissue growth and thus organ shape change. Chronologically, this primary objective will be achieved as the last big step of the project, as it depends on many prior steps.

2) SIMBIONT is rigorously data-driven, (especially from quantitative image-based data), so the project involves 3

Work performed

In addition to the primary goal of a computer model, a number of different data-sets must be obtained for the SIMBIONT project. The first of these is a complete numerical description of 3D tissue movements during limb development. This has been pursued in 2 different ways. The first is to improve in vitro limb bud culture techniques, so that we can simply image the live dynamic growth of the tissue under a suitable type of microscope. We have performed a lot of work in trying to improve our in vitro techniques, but so far they have not been very successful. We are continuing with this route, but the second alternative approach is to digitally reconstruct the dynamic behavior of limb development from 3D static data-sets of various different read-outs. On this front the project has been much more successful. We have developed new software for creating a standard 4D trajectory of the limb bud morphology over time. The key is not to simply “morph” 3D data-sets together, but to find a method which genuinely takes advantage of as much of the data as possible. To this end we have developed a computational method which uses spherical harmonics as a way of mapping the 3D shapes of different aged limbs into a standard reference coordinate system. In addition to the overall shape, our standard trajectory must also capture the internal movements of the growing mesenchymal tissue.

A second critical data-set for the SIMBIONT project is a 4D database, or spatio-temporal atlas (STA) of the dynamics of all relevant molecular states involved in limb development. This includes data on where different genes are expressed, and also information on the activated states of signalling pathways (which can be obtained by immunohistochemistry with antibodies specific for the activated forms of various signal transduction molecules. The original proposal – Tomo-Seq – does not work as well as hoped, and as a result we have focused on an atlas-based approach. We have worked to map multiple gene expression patterns (scanned by Optical Projection Tomography) into standard 3D atlases at fixed time-points, and also performing multiple trials of RNA-Seq with gradually smaller sample sizes, to work our way towards single-cell trasncriptomics (which can then each be mapped back into the 3D atlas). We have also spent the last 12 months of the project working on a new alternative method for 3D spatial transcriptomics. Multiple rounds of testing have been performed, with multiple rounds of related RNA-Seq. The proposed approached is totally novel, and we hope to have concrete proof of concept within one year from now.

Beyond the empirical data-sets, the core goal of SIMBIONT is to create the first full 3D + time computer simulation of limb development, which includes the dynamics of the gene regulatory networks which control the process, all relevant cellular activities, and makes good “macroscopic” predictions of the morphologies – both wildtype and mutant phenotypes. It was clear that linking all of these features together into a single simulation framework would require a new software platform. Our most exciting results of the project so far have been in the creation of just such a new computer program. Named yalla, this software was written in the lab from scratch and designed from day one to employ GPUs (graphics processing units), rather than CPUs. In other words, the code of the simulator was written in the GPU-specific language CUDA, such that it is completely dedicated to this task, and runs extremely fast (on either a standard desktop GPU, or a GPU cluster).

We have already demonstrated the power of yalla in simulating proof-of-concept morphogenetic processes. In particular, we programed it to emulate both mesenchymal tissue and also epithelial sheets. In the latter case it employs a cell-based polarity to represent the apical-basal axis. This allows a computationally-efficient method to endow the epithelial sheet with a controlled degree o

Final results

Some of the experimental results of SIMBIONT are still under development (generation of data-sets), however the new software yalla represents a core goal for the SIMBIONT project, and a big step forwards in the field of multicellular simulation software. It is a general-purpose simulator for morphogenesis (including tissue homeostasis and regeneration) which we believe will be a valuable tool for multiple labs around the world. It displays a number key features which when combined in this way make it a novel system taking it beyond the state-of-the-art:
• Simulation algorithm designed explicitly for maximum benefit of the GPU architecture.
• The use of GPUs allows extremely parallel computation.
• The simplest model was sought for implementing the cell vector force calculations.
• No dedicated graphical user face was integrated into the code, again to maximize speed.
The resulting software is able to simulate complex morphogenetic processes, involving the active movements of hundreds of thousands of virtual cells, controlled by local molecular reactions and efficient diffusion of signaling molecules (morphogens). Our technical paper illustrates all the major capabilities of the program, including two examples that integrate patterning, signaling, directional control of cellular intercalation and large-scale tissue shape changes.