Computational Approaches and Models for Ovarian Ageing: From 2D to 4D

Abstract

The theme of the work presented in this multi-disciplinary PhD is the development of new computational tools and techniques to study and understand spatio-temporal follicle growth in neonatal mouse ovaries. The female ovary is endowed at birth with a finite, non-renewable supply of oocytes, each enclosed in a layer of supporting somatic (granulosa) cells to form a quiescent follicle. From birth, a steady trickle of follicles initiate growth to maintain a supply of mature oocytes for regular ovulation. Disruption in the regulation of initiation of follicle growth can result in various pathologies, such as premature ovarian failure and polycystic ovary syndrome. The mechanism of regulation of the initiation of follicle growth remains unclear, but may involve inter-follicle signaling via paracrine growth factors. To investigate this hypothesis, a new technique for quantifying and analyzing spatial distributions of quiescent and growing follicles in the adult human has been developed, as an extension of a novel technique previously developed in neonatal mice in our laboratory. As in the mouse study, we have found evidence that in the human ovary neighbouring quiescent follicles inhibit follicle growth, at a small range. This approach has been further extended to cultured neonatal mouse ovaries, which in vitro lack a systemic blood supply, to investigate the relative contributions of inter-follicle paracrine signaling and endocrine growth factor/nutrient signaling to the regulation of initiation of follicle growth. Accurate counts of the numbers of follicles in ovaries are important for a wide variety of studies of ovarian physiology, including investigating the effects of age, toxins, chemotherapeutics, endocrine disruptors and specific genes (knock out/transgenic studies) on follicle formation, endowment and development. Many published studies use frequent sampling of a small number of ovaries (often as few as three) to obtain estimates of the number of follicles. We have tested the validity of this approach by generating 3D spherical simulated ovaries which contain realistic numbers of follicles at different stages and which are realistically positioned within these ovaries. The number and position of follicles is based on real biological data. This model enables us to rapidly ‘virtually’ section the ovary in silico and obtain computer-generated counts of the numbers of follicles in sections at different frequencies, such as one every fifth section (1/5), 1/20 or 1/50. As we know precisely how many follicles each simulated ovary contains, we can compare the accuracy using different sampling frequencies of varying numbers of ovaries. This has enabled us to demonstrate that the error is smaller when infrequent sampling of a large number of ovaries (≥8) is carried out, and that this actually involves analyzing fewer sections overall. We have gone on to generate simulated ovaries from knockout mice, with more or fewer follicles, and can predict how many ovaries are required to make robust comparisons between knockout and control animals. This has shown that biological variability contributes more to counting error than the method of sampling. These simulated ovaries provide a unique resource to model large studies. Currently follicle counts are obtained by fixing and serially sectioning ovaries, and manually counting the follicles in sections. This is laborious and time-consuming. Faster methods of obtaining follicle estimates are required. With the use of confocal microscopy and immunohistochemistry for an oocyte-specific protein, we were able to establish a protocol that allows us to image and computationally reconstruct a whole neonatal mouse ovary in 3D. Follicle number can be estimated rapidly using a stereologic method. The stereologic technique error was estimated using the simulated ovary model, leading to the conclusion that the method can be safely used to obtain rapid estimates of follicle number. The time required can be further reduced by using image processing to detect the stained follicles on the sections. We have developed an algorithmic technique that can instantaneously identify stained oocytes, count them, and calculate their spatial distribution. A fundamental unanswered question is whether follicles move in the ovary, particularly as they grow. This question has arisen from the observation that small follicles tend to be situated close to the ovarian surface, while large ones are closer to the medulla. This question has implications for interfollicle signaling. We have developed a protocol to image the ovary while in culture using timelapse confocal and live lipid stains to visualize the follicles. Results show that small follicles are not moving significantly over a period of 12h. This project can be extended in the future with the use of transgenic mice for GFP tagging, to accurately monitor changes in structures of interest within cultured ovaries.