Background

Image-based profiling is an ‘omics’ technique with increasing popularity, that uses images from high-throughput microscopy to analyze biologically relevant effects on cells. Recently, high-content cell profiling has shown use for clinical application (e.g. diagnosis, prognosis). However, readout is traditionally limited to only one or few time points possibly missing out important morphological dynamics. Therefore the aim of this study was to improve our established morphological profiling platform C-MORE by adding the temporal dimension.

Methods

Primary neonatal rat cardiomyocytes were treated with phenylephrine (PE) or insulin (INS) for 60h. At 0, 3, 6, 9, 12, 24, 30, 36, 48 and 60h treated cells were fixed with paraformaldehyde. Cells were stained (desmin, DAPI) and imaged (INCell Analyzer 2200). Image segmentation, feature extraction and selection were performed as a modified version of the previously described C-MORE assay¹. Treatment-specific profiles were created by median-aggregation of selected features. Unsupervised Leiden clustering was performed on a uniform manifold approximation and projection (UMAP) representation. The most frequent treatment condition within a cluster was chosen as the predictive label for the entire cluster (majority vote). Accuracy was evaluated as concordance of inferred clustering labels and true treatments.

Results

After stimulation with PE and INS, cellular subpopulations were identified whose morphological trajectories deviated from the control. In general, overall accuracy increased with time and ranged between 0.95 and 1. Interestingly, PE and INS showed multiple trajectories of subpopulations with differing time points of maximum deviation from control. To evaluate whether information from time resolved trajectories compared to classical single timepoint experiments can improve the accuracy for treatment prediction, we compared a single timepoint (60h) to the morphodynamics based identified timepoints of maximum deviation to control. This increased accuracy to 1.

Conclusion

Morphodynamics enables the analysis of time-dependent patterns and identification of cellular subpopulation with characteristic temporal development (trajectories), which can be used for improving the accuracy of treatment prediction. Herein we present a live cell imaging assay that combines advantages of our established high-content image-based cardiomyocyte profiling (C-MORE) with temporal resolution. Our findings hint towards interim differentiation and specialization of subpopulations of cardiomyocytes which are only detectable in temporal resolution. In a next step this methodology could contribute to deepening our understanding of cardiac disease pathophysiology  and pave the way to personalized medicine.

1: J. Furkel, M. Knoll, S. Din, N.V. Bogert, T. Seeger, N. Frey, A. Abdollahi, H.A. Katus, M.H. KonstandinC-MORE: A high-content single-cell morphology recognition methodology for liquid biopsies toward personalized cardiovascular medicine