Biophysics and Deep Learning Approach to Investigate Tissue Dynamics During Ovulation of Young and Aged Follicles.

J. Turley¹, A. Shivankar¹, L. Wang², R. Prevedel^2,3,4,5, C. J. Chan^1,6

1. Mechanobiology Institute, National University of Singapore, Singapore


2. Cell Biology and Biophysics Unit, European Molecular Biology Laboratory, Heidelberg, Germany


3. Developmental Biology Unit, European Molecular Biology Laboratory, Heidelberg, Germany


4. Epigenetics and Neurobiology Unit, European Molecular Biology Laboratory, Monterotondo, Italy


5. Molecular Medicine Partnership Unit (MMPU), European Molecular Biology Laboratory, Heidelberg, Germany


6. Department of Biological Sciences, National University of Singapore, Singapore

The release of oocytes from inside an ovarian follicle is a critical biological process required for reproduction. While the hormonal signals regulating this process are well studied, less is known about the role of tissue hydraulics during ovulation. Ovulation is fundamentally a mechanical process, where factors such as increased follicle volume, degradation of the follicle wall and a build-up of internal pressure may all contribute to the rupturing of follicles. Anovulation, failure of eggs to be released from the ovary, is a hallmark of infertility in ageing and diseases such as polycystic ovary syndrome (PCOS). Aged ovaries are known to be both stiffer and have dysregulated hyaluronic acid synthesis. We hypothesise that these and likely other mechanisms would lead to altered follicle mechanics with less effective rupture and release of oocytes.

In this project, we aim to use a combination of mouse follicle ex vivo cultures, biophysical measurements, advanced microscopy and deep learning image analysis to determine the biomechanical basis of follicle rupture during ovulation and how this is altered in aged ovaries. Until recently it has been challenging to study this process, due to difficulties in live imaging such a large tissue ( >400 µm) undergoing highly dynamic morphological changes. To overcome these challenges, we use optical coherence microscopy, a label free imaging technique, to characterise the ovulation dynamics. This microscope can image the entire 3D volume of follicles within 30 seconds, while incurring no phototoxicity. Such a high temporal resolution allows us to study the highly dynamic nature of follicle rupture and oocyte ejection.

One challenge with using label free imaging techniques is the lack of unique fluorescent markers to label distinct cell types. Instead, follicles components have subtle differences in the intensity and texture which are distinguishable by the human eye, but not by simple image segment techniques. Therefore, we developed a deep learning model to segment components such as the oocyte, somatic cells and antrum, and achieved a model accuracy of over 97%. This allows us to successfully quantify the spatiotemporal dynamics of tissues across the whole follicles and over the entire process of ovulation. Specifically, we found that following ovulation induction, the follicle volume increases, and its shape becomes more spherical. This likely indicates a buildup of internal pressure inside the tissue, which we aim to validate using micropressure probe. We also found a smaller volume increase in the aged follicles compared to that of the young ones. During cumulus oocyte complex (COC) expansion, follicles exhibit pulsatile activity with a periodicity of ~10mins. Prior to rupture, the follicle contracted along with partial degradation of the follicle wall at the rupture site. In aged follicles, we observed many of the oocytes showed impaired ovulation characterised by failed or lagged ejection, as compared to the typically rapid expulsion of oocytes during ovulation (~ mins) in the young follicles. Our deep learning analysis revealed that the follicles with lagged ejection have a lower amount of antrum fluid, indicating a possible link between antrum formation and effective ovulation. Altogether, our results highlight the benefits of combining label free imaging and machine learning to analyse complex 3D tissues dynamics in large tissues, which can be applied to other developmental and disease processes, hence providing a useful resource for the scientific community.