Highly Automated Live Cell Imaging
Modular Software Speeds Up Development of Analysis Workflows
- Highly Automated Live Cell Imaging - Modular Software Speeds Up Development of Analysis Workflows
- Fig. 1: Module chain for image analysis: a sequence of plugins is used to analyze the cell cycle. With each step more information is gained about the data. The user interface is designed that it supports the chain and lets the user intuitively follow the workflow.
- Fig. 2: Tracking of individual cells: live cell imaging allows visualizing the history of single cells.
- Fig. 3: Modular data pipeline: independent and interchangeable algorithms can be used to analyze data. The modularity is also represented on the level of data analysis and the user interface. Configuration files are used to adapt the modules to different workflows.
Live cell imaging is becoming an important tool in pharmaceutical research. Automated microscopy and image analysis are used to study, for example, the influence of compounds on cell cycle progression. Researchers want to find out which compounds arrest the cell cycle or how long they need to incubate at which concentration. Studying such effects requires complex image analysis workflows, which can be tedious to establish. Modular software helps to speed up this process.
In drug discovery processes for cancer drugs researchers study the cell cycle to find compounds that arrest it. The cell cycle is a fixed sequence of phases that a cell passes through while it is dividing. If something goes wrong during the transition from one phase to the next, the cell usually enters programmed cell death. Compounds inducing such behavior are highly interesting for companies that develop new active pharmaceutical ingredients to treat cancer. It is a long way for a compound to actually become a prescribed drug; many different requirements and tests need to be passed. One of the early steps in this process is drug discovery, where researchers screen for compounds that are able to arrest the cycle in a cellular model . They image dividing cells with an automated microscope over a period of one or two days. During that time they apply different compounds and characterize their effects on the cells. Most of the time there is no obvious reaction, but sometimes the substances have an effect. Since the researchers do not know beforehand in which phase the compound will work, they need to visualize the entire cell cycle. The arrest can happen in different phases and each compound may have a different mode of action. The better the researchers can characterize a compound at the cellular level, the easier it is to further develop it. Therefore, the images from the microscope need to provide detailed information about the cells. Plus, the images need to be analyzable with automatic procedures, thus many compounds can be screened in a short period of time.
Visualizing the Cell Cycle
To visualize the entire cell cycle, specialized cell lines are used that carry different reporters for the cell's DNA .
These reporters are based on fluorescent proteins that are fused to very small camelid antibodies. The antibodies bind to target structures inside the cell nucleus. Since this bond is very specific, the antibodies can target different structures that appear in different phases of the cell cycle. The reporters thus indicate different phases of the cycle. When the right structures are combined, the entire cell cycle can be made visible. This is done by fusing the two antibodies to different fluorescent proteins, one of them green and the other one red. This allows us to distinguish the signals from the two reporters. During the phases each of the reporters then changes its specific signal appearance. The intensity may become higher or some bright spots may appear in the cell nucleus. The reporter-antibody fusion proteins are very small and very stable. Thus the cells can produce them continually while they otherwise behave normally. This is an important prerequisite for live cell imaging studies, since the researchers want to study the effect of compounds and not the effect of the reporters on the cell cycle. The cells are put into microplates and, after they have recovered from this transfer, compounds are applied and the cells are imaged in the microscope. During this time the cell cycle is constantly visible, thanks to the reporters.
Automating Image Analysis
Since the automated microscope produces hundreds of thousands of images in a few days, powerful image analysis software is needed to discriminate the different appearances of the reporters. Vendors of automated microscopes usually also offer software for the analysis of the images that their hardware produces. The difficulty with these software packages is to find the right strategy for a specific assay. The analysis needs to be robust to cope with different conditions and at the same time fast enough to compute results for many thousands of images in a short time. Ideally, parts of the analysis strategy can be reused for other assays. To analyze data from a live cell imaging assay, some basic image analysis procedures need to be combined with a set of specialized algorithms. The basic procedures include image enhancements, e.g. shading correction or segmentation, that are used to remove typical artifacts or to recognize objects in the images. One of the adapted approaches is a special tracking routine. In live cell imaging, the images of a time series cannot be analyzed separately. It is absolutely essential to study the development inside the cells across a sequence of images. This is why a tracking routine needs to be integrated in the image analysis workflow. However, such routines are not available in most of the software packages provided by microscope vendors. Most of the time, the users cannot just pick a predefined and optimized workflow that contains all the necessary procedures. Instead, they have to develop a complete new workflow, if only some specialized algorithms are missing.
Modularizing Image Analysis Workflows
Modularity is a way of dealing with complex problems . Instead of developing a workflow with many dependencies for the analysis of the cell cycle, the problem should first be decomposed into small elements. It is a lot easier to find a solution for each element than for the complete workflow. A complex problem is decomposed into much simpler and manageable problems. After each of them has been solved, they can be stitched together to form a complex solution. This process requires, however, that the elements are self-contained sub-tasks that are mostly independent of each other. Dependencies can make the problem complex again, and, more importantly, they prevent interchangeability of the elements. This is, however, what modularity wants to achieve. Not only do we want to understand a complex problem by decomposing it, but we want to reuse the elements of the solution for future problems. If we have achieved modularity, we can understand the image analysis workflow more easily and we can reuse parts of the workflow for others.
Configuring Instead of Implementing Fixed Workflows
The Fraunhofer Institute for Applied Information Technology FIT has developed a software platform that can be adapted to different image analysis tasks. The platform provides the necessary tools and components to build a modular image analysis workflow quickly and easily. The components are self-contained and take care of image handling, parallel processing or the user interface, to name a few. But most importantly the platform provides the means to put together modular elements of the actual image analysis. In this way many elements can be reused even if the image analysis workflows as a whole are completely different. Fraunhofer FIT has achieved this by creating configurable image analysis plugins. Each plugin contains a number of analysis procedures that belong to the same logical type of task. A preprocessing plugin, for instance, comprises procedures for image enhancement and lighting corrections. These tasks are usually performed at the beginning of a workflow. Another plugin deals with problems of object recognition and a third one with the classification of the objects. This one detects, for instance, the different cell cycle phases. Each plugin is configurable so that it can be adapted to different tasks. Depending on the workflow, it may also contain different procedures, but the logical purpose is always the same. With this strategy, a new analysis workflow can be stitched together in a fairly short time. Future design goals are to give users more freedom over the configurability of plugins while maintaining the operability of the image analysis workflows.
We would like to thank Bayer Healthcare Pharmaceuticals for providing image material.
 M. Held et al.: Nat. Methods 7, 747-754 (2010)
 K. Zolghadr et al.: Methods Mol. Biol. 911, 569-575 (2012)
 C. Y. Baldwin and K. B. Clark: MIT Pr. (2000)
Dr. Andreas Pippow
Fraunhofer Institute for Applied
Information Technology FIT
High Content Analysis
Sankt Augustin, Germany