Randal A. Koene

Current Research
Theoretical models and computer simulations created in my research strive to take into account characteristic features of the neuroanatomy and neurophysiology of specific brain regions, as well as the neural pathways that connect those regions. Simulations replicate the macroscopic behavior of humans or animals, neural spiking activity and whole circuit dynamics. Data that is used to tune and evaluate model simulations comes from relevant major publications, as well as from in-vivo recordings with implanted electrodes and in-situ recordings done in the laboratories of Dr.Michael Hasselmo and Dr.Howard Eichenbaum within the Center for Memory and Brain (http://www.bu.edu/cmb/) at Boston University. Animal experiments and simulations of theoretical models are explicitly complementary in Dr.Michael Hasselmo's laboratory of Computational Neurophysiology. The remaining labs in the Center are those of Dr.Nancy Kopel, Dr.Chantal E. Stern and Dr.John White. The Hasselmo, Eichenbaum and Stern labs also coordinate research efforts with eleven other labs at Boston University, including Dr.Stephen Grossberg's department of Cognitive and Neural Systems, as well as labs at MIT (Dr.Earl Miller), at the University of Pennsylvania (Dr.Michael Kahana), and at Brandeis University (Dr.Robert Sekuler), to constitute a Center of Excellence for Learning in Education, Science, and Technology (CELEST) funded by the NSF (http://www.cns.bu.edu/CELEST/).

Many of my simulation studies were done using the Catacomb modeling environment, developed by Dr.Robert Cannon. My integrate-and-fire neural network simulations explicitly take into account: (1) spike timing, (2) dynamic post-synaptic response functions specific to receptor types, (3) dynamic response functions of the neuronal membrane potential, (4) spike timing dependent modification of synaptic strength, which can incorporate models of long-term potentiation and long-term depression, and (5) rhythmic modulation of membrane potential, of synaptic transmission and of synaptic plasticity. Recent evidence for specific biophysical mechanisms in neocortical and medial temporal lobe regions of the brain is used to devise realistic models at the resolution expressed in this list of explicit considerations. The possible role of modulating oscillations, especially those elicited by the theta brain rhythm (3-12 Hz) and by the gamma brain rhythm (25-50 Hz), is explored in the context of encoding and retrieval in short-term and long-term memory.

At this modeling resolution, my research focuses on neural mechanisms in prefrontal cortex and in the medial temporal lobes: (1) Processes of executive decision making and rule learning in prefrontal cortex. (2) Processes of temporal context dependent episodic memory in entorhinal cortex and hippocampal regions. I have emphasized specific models for short-term buffers of sequences of spike patterns, upon which processing in both of the brain areas may depend. Specific spatial and operant tasks that require conditional and delayed responses (e.g. non-match to sample, delayed alternation) depend on the combined capabilities of prefrontal and medial temporal lobe regions.

Spiking neuron models that I have developed and published address neural mechanisms for specific functions, including: (1) Short-term memory of ordered sequences of spike patterns based on persistent spiking in layer II of entorhinal cortex and in prefrontal cortex (Koene et al., 2003; Koene and Hasselmo, 2005). (2) First-in-first-out queued replacement in short-term buffers based on persistent spiking neurons (Koene and Hasselmo, 2006). (3) Alternating modes of encoding and retrieval within each cycle of oscillation during theta rhythm, during which theta modulated synaptic transmission reduces the potential effects of interference between retrieval and encoding activity (Koene et al., 2003). (4) Encoding and retrieval of temporal context dependent episodic memory in dentate gyrus and hippocampal region CA3, which combines with a feature specific stream of retrieval in layer III of entorhinal cortex (Koene and Hasselmo, 2006). (5) A reduction of interference during one-shot encoding in the hippocampus achieved by recruiting spiking neurons into non-overlapping representations (Koene and Hasselmo, 2006). (6) Autoassociative pattern correction in the reciprocal loop formed by dentate gyrus and hippocampal region CA3 to avoid cumulative errors during sequence retrieval (Koene and Hasselmo, 2006). (7) A biologically plausible integrate-and-fire model of reinforcement learning in prefrontal cortex that depends on short-term buffers and spike-timing dependent plasticity (Koene and Hasselmo, 2004, 2005). (8) Decision making during prefrontal cortex dependent goal-directed tasks, through converging streams of spiking activity elicited by current state and goal representations, and propagated between model prefrontal minicolumns (Koene and Hasselmo, 2005).

Behavioral tasks simulated with the models listed above include (http://askja.bu.edu/): goal-directed navigation using sensory input about spatial environments, drug-seeking operant behavior (this also involves modeling functions of amygdala and nucleus accumbens), reward seeking operant behavior in response to visual cues, and delayed non-match to sample operant tasks (different neural mechanisms may be responsible for behavior when novel and when familiar stimuli are encountered). Phenomena observed during normal function were demonstrated, such as phase precession of hippocampal pyramidal cell spiking during simulated linear motion and event-selective spiking of prefrontal neurons. In many of these tasks, simulated lesions were applied to specific brain regions or to communication pathways between regions, and deficits in rhythmic modulatory effects were simulated to predict behavioral differences from normal function.

The results of my research at the Laboratory of Computational Neurophysiology have contributed to our understanding of the possible functional roles of oscillatory brain rhythms: Theta rhythm at specific phase offsets may enable synchronization of functions performed in multiple brain regions, by alternately depolarizing and hyperpolarizing neurons. It may enable the regular and ordered repetition of spike patterns in a buffer of persistent spiking neurons. Theta rhythm may enable time-multiplexing of different operations (encoding and retrieval) by using theta modulation of transmission and plasticity to prevent interference. In general, the theoretical models of my research described above combine a top-down approach to the integration of functions provided by multiple brain regions (systems neuroscience) with a selective focus on underlying biophysical details (neurophysiology). This general approach has made it possible to propose neural mechanisms and make specific predictions for observable behavior in experiments at the Center for Memory and Brain. A rationale for this approach, as an effective way to deal with the nature of biological investigation, and a description of the Catacomb simulation environment (http://askja.bu.edu/catacomb/index.html) have been published (Cannon, Hasselmo and Koene, 2003).

Research conducted through my affiliation with the Netherlands Institute for Neuroscience, which incorporates the former Netherlands Institute for Brain Research (http://www.nih.knaw.nl/), involves the development of computer modeling approaches for large-scale neuronal networks (thousands of neurons) that exhibit biologically realistic patterns of activity. I focus on phenomenological and mechanistic models of neural morphological development (morphogenesis) for the stochastic generation of networks that establish realistic structural and functional connectivity. With these models, I simulate the development of dendrite and axon arbor, and the formation of synapses between morphologically detailed neurons. Resulting two and three dimensional neural networks resemble those generated in the laboratory of Giorgio Ascoli, but by using neural growth models pioneered by Jaap van Pelt, I can simulate a sequence of realistic states of development in a network. Such a sequence enables direct investigation of the link between changing morphology and changes in emergent patterns of activity. Simulation results in terms of connectivity and patterns of activity are compared with microscope images and recorded activity in neuronal networks that are cultured on electrode arrays (Netherlands Institute for Neuroscience), and with data recorded in cortical slices (Vrije Universiteit Amsterdam).

I created a software framework for the generation of networks with realistic neuronal morphology, NETMORPH, that is based on a set of phenomenological models for neurite growth and synapse formation. Simulated network development with this framework will be used in conjunction with statistical and biophysical modeling approaches that are developed by collaborators in the Computational Analysis of Spatiotemporal Patterns of Activity in Neuronal Networks (CASPAN) project (http://www.neurodynamics.nl/). Emerging patterns of activity will be investigated in developing neuronal networks. Results of this work have been presented at meetings in 2004 and 2006. A paper describing the framework is in preparation.

___

I actively seek out research projects that combine elements of systems neuroscience with the study of biophysical processes in developmental neuroscience and neurophysiology, by applying my neuroscience and engineering experience to the problem domains described above. Integrative projects that involve modeling large-scale neural networks and detailed neuron morphology address my goal to identify significant functions encoded in neural ensembles, and their dependence on the biophysics of specific components. Understanding this dependency, and the ability to extract encoded functions, can advance basic neuroscience and applied neuroscience, such as medical neural prostheses and neural interfaces.

My scientific efforts are supported by an effective collaborative network that includes Michael Hasselmo (BU, computational neurophysiology and systems neuroscience), Jaap van Pelt (NIN, neural morphogenesis), Arjen van Ooyen (VU, neuroinformatics), Arjen Brussaard (VU, neurogenomics), Bruce McCormick (TAM, visualization, brain mapping, and in particular knife-edge-scanning of mouse brain), Jerry Schneider (MIT, neural regeneration, nanotechnology), Henry Markram (Blue Brain Project, reconstruction from morphological scan data) and Ben Goertzel (A.I.).

Ph.D. Research
My Ph.D. research focused on the short-, intermediate- and long-term memory (STM, ITM and LTM) processes of the cortico-hippocampal loop. A plausible model of STM has been proposed by Lisman and Idiart (Science, 1995), and a model for its integration with ITM has been proposed by Jensen et al. (Learning and Memory, 1996). First, I addressed the postulate that attentional control kindles consolidation. This postulate may be applicable as a selectivity function during several transitions from one stage of human memory to another. I then enhanced the representation of the interneuronal network in the Jensen et al. model, the synaptic update function and the local distribution of connectivity. I characterized the timing requirements for the synchronous cooperation of multiple network stages that exhibit oscillations of coherent activity. Simulated acquisition and ITM storage of episodes consisting of arbitrarily large patterns of activation was achieved. I included the effects of modulation through presynaptic GABA_B receptors, as a way to modulate transmission during encoding and retrieval functions in models of dentate gyrus and hippocampus. Simulated encoding and retrieval with interacting spike patterns demonstrated that, while STM could maintain interacting patterns in consecutive gamma cycles when combined with autoassociative pattern completion, cross-talk was a problem for synaptic storage. Subsequently, I realized that associating elicited spike patterns with coincidentally coactivating neurons can multiply instantiate initially interacting patterns into sparse representations, as are supposed in the hippocampal system. Memory established with multiply instantiated representations is ideally suited to rapid acquisition, but limited in resources and possibly non-permanent. Therefore, I hypothesized that the attentional focus implied when spike patterns are rapidly acquired in ITM may also serve to elevate the likelihood of later activity that can consolidate representations in robust LTM. My model proposed that increasing the likelihood of activation in target areas is achieved by ``attentional highlighting'' through long-term potentiation. I created a biophysical model for the synaptic sensitivity to frequent activity and the presence of CREB, and simulated the establishment of LTM traces at highlighted synapses.

Curriculum Vitae:
Letter paper PDF
A4 paper PDF

and Research Addendum:
Letter paper PDF
A4 paper PDF

Publications and Ph.D.

BU-CMB Computational Neurophysiology

VU-CNCR Computational Neurophysiology

Computational Neurophysiology at the Center for Memory and Brain in the Department of Psychology and Program in Neuroscience of Boston University

Psychology at McGill

Interests:

Neuroprosthetics
MindUploading.org
Neural Prosthesis Program (NPP)

Nature News
Science Now
Salon
NRC Handelsblad
Slashdot

Weather, Webster, PubMed