Extracellular signatures of pathological activity in the subthalamic nucleus

The electric nature of information processing in the brain, makes electrophysiology one of the most common approaches of measuring neuronal activity in neuroscience. Extracellular potentials depend on a multitude of factors such as the local neuronal properties, the tissue conductivity and the precise connectivity structures to different brain areas. Understanding the nature of these field potentials is important for our understanding of brain activity and information processing as well as clinical applications such as deep brain stimulation (DBS) in Parkinson's disease. The focus of chapters 2, 3 and 4 of this thesis is on modelling measured extracellular potentials, whereas the last chapter focuses on measures ensuring the reproducibility of computational models in neuroscience. A main target for DBS is the subthalamic nucleus (STN). During surgeries extracellular potentials are often used as guidance signal. Mainly specific local pathological rhythmic activity as well as patterns of action potentials. These guidance signals are used for more precise localisation which is crucial for clinical efficacy. Modelling the locally generated extracellular potentials therefore is important for our understanding of the measured signals. In the second chapter we therefore study a model of the STN local field potential. We studied a variety of common assumptions in relation to pathological activity in other modelling studies. Thereby we were able to show critical dependence of the orientation of individual neurons to the spread of coherent activity in the LFP and due to exponential amplitude decay of coherent activity a decreased reach compared to older studies. We then used this model in chapter three to thoroughly test a method aimed at localising pathological activity. First this method disentangles the population activity of different populations by using a Fourier based Independent component analysis and then using kernel-current-source-density deconvolves the spread out LFP to localise the signal generating populations of neurons. This method makes use of simultaneous recorded LFPs and could in future be used in clinical applications to improve the usage of guidance signals as well as to monitor the spatio-temporal activity in experimental studies. In the high-frequency filtered component of the extracellular potential one can observe action potentials of single neurons. Due to high density of neurons in the brain, one usually measures several neurons at once. To study the properties of single neurons and their relation to the surrounding or remote population activity one first has to assign the observed action potentials to so called single units by means of spike-sorting. In chapter four we propose a recipe of how to test ones spike-sorting method by generating a ground-truth data-set based on data from intra-operative measurements during DBS implantation surgery. With this we were able to show that automated algorithms can achieve human hand sorted performance while taking only a fraction of the time. In the last chapter five we revisited a popular model for explaining information processing in the brain and uncovered several technical inaccuracies. Moreover we found that these technical inaccuracies largely invalidate the original findings. We therefore decided to write a guideline to make it easier for authors and reviewer to uncover such inaccuracies and thereby increasing the scientific credibility of a scientific study