Using automated image analysis of live neurons to screen for natural compounds effective against Alzheimer's disease

Abstract

Sustained technological advances in microscopy are enabling biologists to observe living organisms with unprecedented resolution and to reveal new information about their cellular structure, dynamics and function. Accurate and systematic methods for analysing, organising and visualising the information in the corresponding images are, however, still lacking. This thesis addresses several aspects related to the quantification of biological structures relevant to Alzheimer's disease - a disease associated with the death of neurons, causing a slow deterioration of a person's memory, speech, behaviour, motor skills and thinking abilities. The fully automated tools generated as an outcome of this thesis have been designed to aid biologists in their search for a better understanding of the neuronal architecture and of the interactions that neurons have with neighbouring cells. The development of these tools will contribute to identifying compounds, either natural or synthetic, that will reduce, prevent or even reverse the symptoms of this fatal disease - even small delays in the onset of the disease will significantly reduce its costly impact on society. Of particular interest is the automated analysis of the neuronal trees. The reliable identification of neurite traces is very important, but is also challenging. Developing rules that associate neurites with their true mother cells is an unsolved problem. Obtaining precise estimates of neurite properties uniformly along their length without bias induced by the neurite's width, brightness or location in the image, also remains a challenging task. One of the methods presented concerns the detection of weakly contrasted neurites in noisy images and the preservation of continuous neurite traces over their full length. These scenarios are often prevalent when acquiring images of neurons under low light conditions, when photobleaching and photo-toxicity are to be avoided. Other circumstances require the reconstruction of the complete neuronal tree, with neurites connected to their true mother cell. This is particularly important for measuring the influence that different growth conditions have on the total neurite length as well as the subtle changes they induce in the neurite morphology. Another novel method introduced in this thesis involves the fitting of smoothing splines to digital traces of neurites. This enables precise estimation of the local neurite properties, such as local tangents, uniformly along the neurite length. Another hallmark of Alzheimer's disease is the formation of senile amyloid beta plaques in the Alzheimer's brain. The statistical association between the familial form of the disease with genetic mutations, has led to the development of many strains of transgenic animal models that mimic pathological symptoms of Alzheimer's disease. In particular, amyloid beta peptides have been the centre of attention of much research over previous decades. Consequently, there is a need to accurately quantify the plaque load in the Alzheimer's brain. A framework has, therefore, been developed for segmenting amyloid plaques made visible through staining of histological sections of transgenic mice brains. The quantification of plaque load, such as plaque number, individual size, and coverage of critical brain regions, was shown to be able to measure the neuroprotective potential afforded by natural compounds. For example, Vitamin D from enriched mushrooms is shown to reduce the plaque load. Furthermore, correlations between amyloid beta plaques and other biomarkers could also be characterised, thus contributing towards an improved understanding of the Alzheimer's pathology. The final contributions relate to the growing interest in multiple particle tracking. This involved the development of a user-friendly graphical user interface, which generates synthetic data sets corresponding to fields of moving, point-like targets. Thus, the ground truth is available and algorithms may be compared on relevant dynamics. For the biologist, this tool can help optimise the quantitative tracking results in a vast range of applications. In studies of dynamic changes in the neuronal architecture, this tool may be used to optimise the tracking of critical locations of the neuronal tree, for example bifurcation points and neurite extremities.

Date of Award	2014
Original language	English