When Rudolf Virchow started studying medicine in Berlin in October 1839, the theory of the four humors of antiquity was still the mainstay of medical belief. Barely two decades later, the textbooks had to be rewritten: Virchow, who had since received his doctorate in medicine, succeeded in showing that the entire human body consists of cells – tiny units that can undergo morphological changes. Virchow’s revolutionary “cellular pathology” offered an entirely new understanding of causes of disease. His teachings, valid to this day, laid the foundation for modern, science-based medicine.
And yet, by today’s standards, the prerequisites for such a momentous discovery were modest. Virchow examined tissue samples under a simple microscope with the aid of a mirror and sunlight. To make cell structures visible, he stained tissue with dyes mixed together for him by chemists. Using this method, the “father of cellular pathology” managed to diagnose 20 diseases, including leukemia and thrombosis.
21st century physicians and scientists still use staining techniques to identify morphological patterns of cells, the detailed diagnosis of cancer being a prominent example. Only today, the dyes are fluorescent and the equipment differs somewhat from that of the 19th century.
A glance at the Screening Unit at the FMP in Berlin-Buch shows just how far things have progressed: The state-of-the-art technology features a fully automated confocal microscope equipped with two cameras that automatically records 1,000 morphological characteristics for each individual single cell. Given that almost 384 experiments can be performed simultaneously on one test plate (the FMP compound library has 200 of them), no fewer than 400 million data sets per each single plate are generated in one run alone. Not even a mastermind like Rudolf Virchow would have been able to analyze such huge volumes of data. The high-resolution microscope is therefore connected to a fleet of supercomputers that use artificial intelligence to detect tiny changes in cells and assign them to specific classes or diseases.
The Unit works in the tradition of Virchow, but using computer power that was unimaginable back then. “Virchow 2.0” is the term given to the concept of computer-aided pattern recognition, which is ideal for both drug discovery and disease diagnosis.
The Screening Unit of FMP is currently using the new technology for cell toxicity profiling of the EU-OPENSCREEN compound library. The researchers want to find out which of the provided 150.000 drug-like chemical substances in the EU-compound library are potentially and generally cell-toxic. This classification should make drug screening assisted by robotic arms even more efficient in the future.
“Virchow 2.0” is soon to be used for personalized medicine. When cancer patients have developed resistance to drugs, for example, researchers can use tissue samples to search for alternative medications. Requests to this effect have already been received from Virchow’s long-time workplace – Charité. There are plans for further fields of application, and machine pattern recognition still has further potential. “Virchow pushed the boundaries and we are trying to do the same using current technological means.”
Text: Beatrice Hamberger, Translation: Teresa Gehrs
CELLULAR PATHOLOGY & COMPUTER AIDED PATTERN RECOGNITION: About 160 years ago Rudolf Virchow realized that human diseases are often reflected in an aberrant cellular morphology and that different disease stages are also reflected in different grades of aberrant morphology (Cellular Pathology). Recent developments in computer aided pattern recognition, including Artificial Intelligence and machine learning software tools, plus automated high-content screening with confocal microscopes for 3D analysis, encouraged us to apply these novel techniques for cellular morphology profiling. We are actually testing drug application and genome-wide gene-function-interference techniques for this. In principle, we test for, if specific pattern recognition of cellular responses may replace cellular reporter genes for HTS in future projects.
We decided to start with and select for toxic reference compounds linked with known cellular targets and molecular mechanisms to induce cellular dysfunction in different cell types (Painting Assay, Anne Carpenter, Broad). The aim is to generate specific morphological fingerprints for each kind of perturbation of cell function and correlated mechanisms of compound action. For this purpose, we used eight fluorescent dyes to stain and visualize cellular components like nuclei, mitochondria or actin filaments and membranes. About 1.500 parameters per individual cell are analyzed and compared in the process of computer aided image and cell structure recognition plus automated analysis and conversion to multiparameter tables. These included fluorescence intensity, object areas and contact zones between objects, orientation and symmetries, plus patterns of objects. For the identification of disease relevant patterns, we will also employ machine learning tools already integrated into the analysis software. Later on we plan to add gene editing by CRISPR-Cas to introduce disease driving mutations or to use patient derived cells to compare individual morphological fingerprints with reference patterns from our database for diagnostic and therapheutic predictions.
Our platform is open to data exchange and very interested in this, with research groups interested in data to feed predictive models for disease to pattern correlation on a mathematical/statistical basis.