Research in Computational Neuroscience aims at developing detailed biophysical and abstract theoretical models of single cells, microcircuits and large neuronal networks in order to understand information processing and memory formation in the brain. Specifically, we built computational models and use them to investigate the biophysical and morphological mechanisms underlying learning and memory processes in various brain regions (hippocampus, prefrontal cortex, amygdala), paying particular attention to dendritic computations.
Computational modeling of place cell formation
Studying the formation of place cells is an important step in understanding how representation of the external environment is coded in neural networks that constitute spatial maps. It is not currently known how place fields emerge in CA1 neurons. An influential model of place cell formation predicts the convergence of various grid field inputs which combine linearly to create the place field output of CA1 cells. In this study, we constructed a model of CA1 place cell formation through the convergence of grid field inputs to the distal dendrites of our model neuron. Firstly, a model of grid cell activity is created and represents the firing of grid cells. Then, different grid fields are used as synaptic inputs to stimulate the distal dendrites of the CA1 neuron, which is a biophysically constrained, detailed and compartmental neuron model. Apart from the excitatory, inhibitory synapses were placed in both the distal and proximal dendrites.
Structured connectivity and microcircuits function in the Prefrontal Cortex
PA is the spiking activity that persists beyond the stimulus presentation and is considered to be the cellular correlate of working memory. In the prefrontal cortex (PFC) in particular, pyramidal neurons were shown to form hyper-clusters, compared to other sensory regions. Yet, very little is known about the functional properties of these microcircuits and their role in Persistent Activity (PA). Motivated by the above this work probes the role of realistic connectivity constraints in shaping the functional output of PFC, through simulations of biophysically and morphologically detailed PFC circuits.
The role of dendritic functions in interneurons behavior.
The role of inhibition in regulation and proper function of neural circuits is one of the most challenging and open questions. Research on this topic focuses mainly on the effect of inhibitory neurons on the activity of principal neuronal populations, but not the opposite. Very little is known about the intrinsic and dendritic integration properties of inhibitory cells and how excitatory inputs shape their firing. In this work, we use computational modelling to characterize the dendritic integration properties of a fast spiking PV interneuron model, one of the main subtypes in layer V of the mammalian prefrontal cortex. A detailed biophysical model of a fast-spiking interneuron was created and extensively validated against experimental data. The model exhibits sodium spikes as well as supralinear calcium dynamics in its dendrites. These findings are in line with experimental data showing calcium-dependent nonlinear summation in PV interneurons and point to the ionic conductances that generate these non-linearities. Activation of uniformly distributed synapses along the dendrites generates two major modes of dendritic operations, the supralinear and the sublinear (Tzilivaki et al 2015).
It has been suggested that, due to their widespread connectivity, fast-spiking interneurons act as a uniform inhibitory blanket across the cortex and exhibit limited specificity. However their integrative properties at the single cell level remain unknown. As dendritic integration is key for neural computations, ongoing work investigates how different spatiotemporal patterns of synaptic activation may exploit interneurons nonlinear integration profile to shape the output of the PV interneuron and to maximize its dynamic range.
Pattern separation in Dentate Gyrus.
The aim of the project is to investigate the role of different mechanisms involved in pattern separation. Pattern separation is a computational task accomplished by hippocampal Dentate Gyrus (DG), and specifically by its principal neurons, the Granule Cells (GC). Towards this direction, we use a simplified, yet biologically relevant, computational model of DG. Using this model, we investigate the role of the indirect inhibitory circuitry formed by Mossy (MC) and Basket Cells and we examine its effect under MC loss condition. In addition, we examine the role of GC dendrites in the aforementioned task by alternating their morphological and biophysical characteristics.
Coding visual information in apical and basal dendrites.
In collaboration with the laboratory of Prof. Smiranakis, we are studying how visual information is encoded in the apical and basal dendritic trees. Towards this goal we simulate different input properties / synaptic location / active mechanisms and untangle the conditions under which layer 2/3 pyramidal neurons encode visual information.
Single-neuron model of persistent activity
In this work, we investigated how NMDA spikes in the basal dendrites of single PFC model neurons participate in the emergence of the persistent spiking acivity. We have pinpointed that input structure to the distal versus the proximal basal dendrites participates in this phenotype.
Microcircuit Models of Sustained Activity in the Prefrontal Cortex
Neurons in the prefrontal cortex display sustained activity in response to environmental or internal stimuli, that is continue to fire until the behavioral outcome or a reward signal. Mostly large-scale modeling studies have proposed intensive recurrence and slow excitation mediated by NMDA receptors as crucial mechanisms able to support the sustained excitation in these neurons. In addition, electrophysiological studies suggest that single-cell intrinsic currents also underlie the delayed excitation of prefrontal neurons. This project is focused on the interplay of both the computational and electrophysiological approaches in characterizing the activity observed at layer 5 prefrontal pyramidal neurons when connected in small microcircuits. Towards this goal we used morphologically simplified compartmental models of layer 5 neurons (both pyramidal and interneurons) implemented in the NEURON simulation environment. These neurons were fully interconnected in a small network, the properties of which are extensively based on anatomical and electrophysiological data. This microcircuit was used to characterize: a) the interplay of single cell ionic with synaptic currents for the emergence of sustained excitability (Papoutsi 2013) b) the role of dendritic non-linearities in the emergence of persistent firing (Papoutsi 2014) and c) the role of different types of interneurons in sustained activity porperties (Konstantoudaki 2014). Understanding the properties that make these neurons special in carrying temporal distinct information by using a bottom-up approach is a key issue in unraveling the complicated dynamics and flexibility of prefrontal neurons during behavioral tasks.
Dendritic Computations in Connection with Learning and Memory.
We are mostly interested in understanding the type of neural computations performed by various classes of neurons in the brain, and in particular the ones that are involved in learning and memory. Towards this target, our work focuses on the development and analysis of neurally inspired theoretical and machine-learning algorithms used to model memory capacity in the brain. This work deals with the effects of dendritic morphology and the biophysics underlying learning induced by synaptic plasticity, on the computational (memory) capacity of pyramidal neurons and their possible role in the formation/retrieval of memories. Resulting findings indicate that such model neurons with stellate-like dendritic morphology and multiple side branches that contain voltage-dependent membrane mechanisms are capable of performing nonlinear computations. Furthermore, the memory capacity of such neurons can outstrip that of linear cells by more than an order of magnitude (see Poirazi and Mel, Neuron, 2001). This difference in the memory capacity, as measured on a classification task, is even more prominent in populations of such model neurons. Towards a more thorough analysis of the types of synaptic integration performed in these cells, we have developed of a very detailed biophysical model of a CA1 pyramidal neuron, which can be downloaded from SenseLab and our Software site. The CA1 simulator is used to address questions regarding the type of neural computations performed in these cells and their possible role in short versus long-term memory storage, as well as the underlying morphological and biophysical mechanisms involved. Results of this work show that integration of synaptic inputs impinging on the dendrites or the apical trunk of the model cell can be linear or supralinear depending on the type (single pulse or high frequency), strength and location of the synaptic stimuli (see Poirazi et al, Neuron, 2003a). Furthermore, the cell’s nonlinear arithmetic can be described by a very simple paper-and-pencil calculation: the cell’s mean firing rate can be predicted by a simple formula that happens to also describe a conventional 2-layer “neural network”. In the first layer, synaptic inputs drive several dozen separately thresholded sigmoidal subunits—physically corresponding to the long, thin terminal dendrites that make up the bulk of the cell’s receptive surface. In the second layer, subunit outputs are summed within the main trunk and cell body prior to final thresholding (see Poirazi et al, Neuron, 2003b). This work provides significant insight information about the type of computations performed by CA1 neurons under various stimulus conditions.
Poirazi, P. and Mel, B.W. Impact of Active Dendritic Processing and Structural Plasticity on Learning and Memory.Neuron, vol 29, pg. 779-796, March 2001.
Poirazi, P. Brannon, T. and Mel, B.W. Arithmetic of Subthreshold Synaptic Summation in a Model CA1 Pyramidal Cell. Neuron, vol 37, pg. 977-987, March 2003.
Poirazi, P. Brannon, T. and Mel, B.W. Pyramidal Neuron as 2-Layer Neural Network. Neuron, vol 37, pg. 989-999, March 2003.
Poirazi, P. Brannon, T. and Mel, B.W. About the Model (online supplement) Neuron, vol 37, pg. 988, March 2003.