Label-Free Estimation of Sarcomere Orientation from Brightfield Microscopy Images of Induced Pluripotent Stem Cell Derived Cardiomyocyte Nuclei

Antti Ahola1, Birhanu Belay1, Carolina W√§hlby2, Jari Hyttinen1
1Faculty of Medicine and Health Technology, Tampere University, 2Department of Information Technology, Uppsala University


Abstract

Human induced pluripotent stem cell derived cardiomyocytes (hiPSC-CMs) provide a platform for studying disease models and physiological conditions. For quantitative assessment of structure and function of cell cultures, novel methods are needed. Brightfield microscopy methods are a way to measure these characteristics in vitro without labeling the cells. The organization of sarcomeres is a key characteristic of an hiPSC-CM culture maturity. Continuous evaluation of this structure is challenging, and typically requires labelling by genetic modification (GFP tagging). In this work, we propose the label-free evaluation of sarcomere organization from the morphology and orientation of nuclei in brightfield images.

We used a publicly available hiPSC-CM image dataset consisting of brightfield, Hoechst nuclear stain and endogenously GFP-tagged alpha-actinin-2 channels. The dataset consists of >5500 imaged nuclei and related sarcomere structure. We extracted the orientation and aspect ratio (major/minor axes) of the stained nuclei and determined the sarcomere orientations from the alpha-actinin-2 channel from the same cells. Based on the quantified parameters, we estimated the relation between sarcomere orientation, and the orientation and morphology of nuclei. For label-free estimation, we trained a U-Net-based network for segmenting nuclei from brightfield microscopy images.

The analysis of Hoechst-stained nuclei orientation and sarcomere structure revealed a significant correlation between the orientation of nuclei and orientation of sarcomeric structure. The correlation was higher with larger axes aspect ratios of nuclei. For segmenting brightfield nuclei images, the trained U-Net-based network reached over 0.7 intersection-over-union score when comparing to the Hoechst-stained nuclei.

Together, these results indicate that brightfield data can be used to provide estimates of cellular structures without the need of staining sarcomeres. Advances in deriving data from brightfield images provide the means to assess the structure and maturity of cell cultures in repeated measurements, enabling higher throughput and new measures of the in-vitro cardiomyocyte mechanobiology.