Trends in Neurosciences
Microcircuits Special FeatureBiophysically detailed modelling of microcircuits and beyond
Introduction
Quantitative computational modelling is becoming an important tool in neuroscience research. Models are developed and studied at all levels, from the molecular processes underlying cellular and synaptic properties to brain-scale neuronal networks. Two complementary modelling strategies are used. ‘Bottom-up’ simulations start from biophysically realistic models that mimic many details of the system under study and enable open-ended investigation of its properties. ‘Top-down’ approaches use abstract models or pure mathematics to cast general principles of the system under study into a minimal model, fully describing its essential properties with as few parameters as possible.
The central role of modelling is to promote synthesis of experimental data from different sources into a coherent picture of the system under study. The resulting model can then, for instance, demonstrate how seemingly unexplained phenomena are in fact a consequence of what is already known. Exploration of the model can lead to truly unexpected findings, which then provide important input for the planning of new experiments. In this manner, modelling enables us to extract maximal knowledge from existing data and to find the most promising way ahead.
Bottom-up models have been successful in simulating microcircuits* 1, 2 (Grillner et al. in this issue) and we first review recent progress made using this approach to simulate the behaviour of two central pattern generators (CPGs), in the leech and the lamprey. Such models traditionally consist of small networks of synaptically connected ‘point neurons’ (i.e. models without morphology). The active conductances responsible for the excitable properties of the neurons and the synaptic conductances are simulated in a physiologically realistic manner [3]. Mathematical approaches used in top-down modelling have also contributed to better understanding of the two CPGs, but results from these methods are not always congruent with those from the bottom-up approaches. Such differences can be resolved only by intensive interactions between the modellers and experimenters involved. Some authors have advocated combining bottom-up and top-down models for the same system [4], but these techniques are not yet widely used.
In the second part of this review, we consider how one can make the simple network models more elaborate, to increase their realism. We describe simulating neuromechanical feedback, adding biochemical networks involved in synaptic learning, the incorporation of realistic neuron morphology and increasing the size of the network to full-scale (with all neurons modelled). This part uses examples from simulations of the other microcircuits reviewed by Grillner et al. in this TINS special feature.
Section snippets
Modelling the leech CPG
Because one can identify individual invertebrate neurons and because their cellular properties and connectivity are stereotypical it is, in principle, possible to create complete models of invertebrate CPGs, replicating all important properties. A system where much progress has been made is the circuit responsible for generating the heartbeat rhythm in the medicinal leech. The sub-circuit of heart interneurons (HN neurons) in the first four segmental ganglia of the nerve cord (Figure 1b)
Modelling the lamprey spinal pattern generator
The lamprey spinal locomotor system 1, 15 generates a wave of neural activity along the body during swimming, normally travelling in the head-to-tail direction but reversed during occasions of backward swimming. It comprises two main types of premotor interneurons, in addition to the motoneurons driving the swimming muscles: ipsilaterally projecting excitatory glutamatergic interneurons, and contralaterally projecting glycinergic inhibitory interneurons (see Grillner et al. in this issue for a
Incorporating neuromechanical models
The lamprey CPG is influenced by sensory feedback from the generated body undulations [15]. Can such interactions also be incorporated into the neuronal circuit models? More generally, can a model of a spinal pattern generator be connected to models of muscles and limbs to capture the entire movement-generation process?
Because all neural circuits have evolved to function in specific natural environments, any study, whether in vitro or a simulation, in which the mechanical context is ignored
Interaction with subcellular modelling
Intracellular signalling cascades can change microcircuit computations by modifying the intrinsic properties of neurons or the strengths of synaptic connections. They are important in the homeostatic control of these properties [34] but, because the signalling processes involved are at present poorly understood, no detailed biophysical models exist. In future such models will have to bridge the gap in timescales between electrical neuronal activity (milliseconds to minutes) and homeostasis
Interaction with cellular modelling
Most microcircuit models consist of networks of point neurons 1, 2 (see Grillner et al. in this issue). But because dendritic integration of synaptic input [48] can have a strong effect on neuronal input–output function, there is growing interest in incorporating morphologically realistic models of neurons into microcircuit models. The standard technology for modelling dendrites is compartmental modelling with active conductances [3] but using this in microcircuit models introduces several
Interaction with large-scale network modelling
A microcircuit is, by definition, a component of a much larger network. It can be simulated to some degree in isolation, especially when it can be related to a reduced in vitro preparation such as a piece of lamprey spinal cord. But often it is in reality embedded in a mosaic of similar modules that requires large-scale network modelling. A major benefit of a large-scale model is that, by providing a realistic number of presynaptic inputs, it removes the need to increase synaptic connection
Concluding remarks
Quantitative computational modelling has come to neuroscience to stay. As these techniques become increasingly integrated with experimental research, there will be more knowledge extracted from existing experimental data and model predictions will enter more routinely into the planning of new experiments. But to succeed, better integration of bottom-up modelling, which will rapidly increase in detail because of affordable massively parallel computers, with top-down approaches is needed.
Acknowledgements
We thank R. Calabrese, R. Maex and T. Szilagyi for comments on earlier versions of this manuscript and A. Kozlov for supplying material for Figure 2. This work was funded by FWO and IUAP (Belgium), by the Swedish Research Council and by the European Commission.
References (69)
- et al.
Interneuron Diversity series: inhibitory interneurons and network oscillations in vitro
Trends Neurosci.
(2003) Modelling of intersegmental coordination in the lamprey central pattern generator for locomotion
Trends Neurosci.
(1992)Dynamic simulation of insect walking
Arthropod Struct. Dev.
(2004)Strong activation of parallel fibers produces localized calcium transients and a form of LTD that spreads to distant synapses
Neuron
(1996)Cerebellar long-term depression might normalize excitation of Purkinje cells: a hypothesis
Trends Neurosci.
(1995)What the cerebellum computes
Trends Neurosci.
(2003)- et al.
Dendrites: bug or feature?
Curr. Opin. Neurobiol.
(2003) Synchronization of Purkinje cell pairs along the parallel fiber axis: a model
Neurocomput.
(2003)Neosim: portable plug and play neuronal modelling
Neurocomput.
(2001)- et al.
Principles of rhythmic motor pattern generation
Physiol. Rev.
(1996)