The promise of targeted cancer therapies depends on knowing which target is truly being hit. Say you have an effective drug that you believe is inhibiting mitosis in tumors. Confirming the exact mechanism of action on the single cell level might allow finer tuning or better choices for combination therapy but in a three-dimensional (3D) tumor, it can be hard to see what’s really going on when anti-cancer drugs go to work. Recent advances in intravital microscopy have made single cell imaging in living animal tissue possible but analysis of the images is tedious. These pioneering 3D model imaging systems still depend on manual scoring of selected images while the sampling itself may bias interpretation.
To streamline this new technology with its flood of data and its human limitations comes an interdisciplinary group from Harvard Medical School’s (HMS) Cell Biology and Systems Biology departments and the Massachusetts General Hospital (MGH). The researchers have built what they say is an integrated workflow that uses advanced fluorescent markers of cell-cycle status, an implanted gold grid as a spatial reference system, high-resolution intravital microscopy, and an automated 3D image analysis framework to track cell-cycle progress in individual cells. All together, this integrated approach yielded a newly detailed look over time at drug effects in human tumor xenografts in mice, a system which the HMS researchers describe in a Nature Methods paper, just e-published ahead of print.
As a proof-of-principle demonstration, the HMS group led by ASCB member Stefan Florian in the Timothy Mitchison lab and Deepak R. Chittajallu in the Gaudenz Danuser lab used their new workflow to compare two widely used antimitotic cancer drugs with a failed drug candidate that were all specifically designed to arrest cancer cells in the midst of cell division. In collaboration with the MGH lab of Ralph Weissleder, they used intravital imaging to acquire a huge 3D dataset comprising 38,000 cells, and applied their newly developed automated image analysis framework to rate the degree of mitotic arrest in multiple regions of the tumors over time. In this preliminary study, the system found that these drugs produced a much lower rate of mitotic arrest in this 3D model than was reported in previous 2D culture experiments and that each drug had a characteristic effect on the nuclear shape of the cancer cells. All this suggests that automated single cell 3D imaging will give us a much more nuanced and complex view of real drug impacts in living tissue, and allow analysis of datasets large enough for truly unbiased, representative quantification.
The researchers began with an implanted grid, which allowed them to return to the same positions in the tumor tissue, over the eight days of the experiment. Tumors are crowded, highly heterogeneous systems, and knowing where you are in a 3D model is critical, they say. The high-resolution imaging system scanned 3D volumes of the xenografted tumors in two passes, capturing the cell cycle status of each cell first by two-photon microscopy from DNA reporter molecules and then by confocal imaging of fluorescent ubiquitination-based cell cycle indicators (FUCCI). A computer-based image analysis framework aligns the images, identifies individual cell nuclei, and scores their cell cycle status. The framework uses machine-learning techniques trained by human observers for accurate detection of tumor cell nuclei, which take on highly complex shapes, especially after drug treatment.
Although their workflow was created to measure a key indicator in cancer biology, Chittajallu and colleagues believe that their automated 3D imaging system will be useful in other fields. “We anticipate that our framework will contribute to faster, more consistent, and richer quantitative analysis of 3D image data sets from living tissues,” they say.