Small is beautiful: models of small neuronal networks CONEUR-1039; NO. OF PAGES 6 Small is beautiful: models of small neuronal networks Damon G Lamb and Ronald L Calabrese Available online at www.sciencedirect.com Modeling has contributed a great deal to our understanding of how individual neurons and neuronal networks function. In this review, we focus on models of the small neuronal networks of invertebrates, especially rhythmically active CPG networks. Models have elucidated many aspects of these networks, from identifying key interacting membrane properties to pointing out gaps in our understanding, for example missing neurons. Even the complex CPGs of vertebrates, such as those that underlie respiration, have been reduced to small network models to great effect. Modeling of these networks spans from simplified models, which are amenable to mathematical analyses, to very complicated biophysical models. Some researchers have now adopted a population approach, where they generate and analyze many related models that differ in a few to several judiciously chosen free parameters; often these parameters show variability across animals and thus justify the approach. Models of small neuronal networks will continue to expand and refine our understanding of how neuronal networks in all animals program motor output, process sensory information and learn. Address Emory University, Department of Biology, 1510 Clifton Rd, Atlanta, GA 30322, United States Corresponding author: Calabrese, Ronald L (ronald.calabrese@emory.edu, rcalabre@biology.emory.edu) Current Opinion in Neurobiology 2012, 22:1–6 This review comes from a themed issue on Microcircuits Edited by Edward Callaway and Eve Marder 0959-4388/$ – see front matter Published by Elsevier Ltd. DOI 10.1016/j.conb.2012.01.010 Introduction Models of the small neuronal networks of invertebrates, especially rhythmically active central pattern generators (CPG), have proven to be fruitful subjects of investi- gation, revealing general principals of neuronal network function and generating hypotheses later supported by the living systems they represent. Over the past two decades, models of ‘simple’ networks, powered by effi- cient desktop computing and a wealth of physiological data, have provided guiding insights into how neuronal networks function. Over the past decade, theoretical studies, but now supported by experimental analysis in Please cite this article in press as: Lamb DG, Calabrese RL. Small is beautiful: models of smal www.sciencedirect.com several different networks and species, have shown that reliable network output can result from networks in which parameters (e.g. the intrinsic membrane proper- ties (maximal conductances) of the neurons and the strengths of the synaptic connections) show 2–5-fold animal-to-animal variability [1–3]. Consequently, to understand a neuronal network through biophysical modeling, we must construct populations of models with multiple sets of parameter values corresponding to parameters from different individuals [4–6]. A sobering consequence is that the computational effort needed to produce a state of the art biophysical model is vastly increased. The situation is clearly still fluid [4,5], but the reaction in the modeling community has ranged from a continued pursuance ‘ideal parameter sets’ or sticking to averaged values for parameters to what Prinz [6], calls ensemble modeling, where multiple functional instances are identified and examined. In this review we sample the diversity of small network modeling approaches to highlight how each continues to contribute significant new insights. A note on models and parameters Before we continue, we should distinguish between models and parameters. The models discussed in this review consist of differential equations that describe the dynamics of state variables, for example, membrane potential (Vm) and the gating variables of voltage de- pendent conductances. Embedded in these equations are a number of parameters, including maximal con- ductances as well as half activation voltages and time constants of channel gates. Some of these parameters are considered free, or variable between instances, while the remaining parameters are fixed. For example, in the pioneering work of Prinz et al. [2,3], only maximal conductances were considered free parameters. Even with powerful computing resources, it is not possible or desirable to consider all instances of a model. Making a model then involves deciding on a neuronal structure (single or multiple compartments), network connec- tivity, descriptive equations (often derivatives of the Hodgkin–Huxley formalism), which parameters are free and the range over which each may vary. These decisions will all be driven by the data available and by the investigators’ intuition for which parameters are likely to be significant in controlling neuronal activity. In short, the ability to consider multiple instances of a model does not free one from making a good model, and making a good model requires detailed knowledge of the system and judgment about what details can be ignored and which parameters fixed. l neuronal networks, Curr Opin Neurobiol (2012), doi:10.1016/j.conb.2012.01.010 Current Opinion in Neurobiology 2012, 22:1–6 http://dx.doi.org/10.1016/j.conb.2012.01.010 mailto:ronald.calabrese@emory.edu mailto:ronald.calabrese@emory.edu http://dx.doi.org/10.1016/j.conb.2012.01.010 http://www.sciencedirect.com/science/journal/09594388 2 Microcircuits CONEUR-1039; NO. OF PAGES 6 Swimming in Tritonia The swimming CPG of Tritonia has long been the object of experimental analyses and modeling [7,8]. Calin-Jage- man et al. [9��], updated the Getting model [10] to reflect new data on synaptic connectivity, intrinsic properties and intrinsic neuromodulation with a careful refitting of intrinsic and synaptic parameters, only to find that the network model did not produce the swim motor pattern in either the unmodulated or modulated state. Unde- terred, they first compared their model’s parameters to those of the original model and adjusted those that significantly differed to the values of the original model. This process restored the ability of the model to produce the swim pattern. They also performed a brute force parameter space investigation of the differing parameters and identified all models that produced the swim pattern (2%). They showed that all parameters sets leading to proper network function were contiguous in parameter space. Then, using discriminant analysis and dimensional stacking, they identified key parameters contributing to proper network function. Somewhat at a loss to explain how their carefully fitted model failed while the Getting model was successful, they suggest that, ‘even if it reflects a nonphysiological configuration, it has still been a useful sign-post toward understanding the conditions that could enable the swim-motor program.’ Feeding in Lymnea The feeding CPG of Lymnea is well characterized [11] and has been analyzed for mechanisms of associative learning in the form of single-trial, food-reward classical condition- ing [12,13]. Vavoulis et al. [14], developed a model of the core CPG with fixed parameters selected to match excit- ability criteria and other physiological data. This model captured feeding activity and predicted physiological phenomena that were later verified experimentally. Cor- responding work in the Aplysia feeding CPG [15] failed to capture feeding activity without an additional putative neuron which has yet to be identified physiologically. An interesting aspect of the Aplysia study is the extensive sensitivity analysis, including perturbations during simu- lations, which demonstrated the model activity’s robust- ness to parameter variation. Vavoulis et al. [16�], have continued the Lymnea work by focusing on the mechanisms of persistent depolarization of the cerebral giant cells (CGCs) that underlies single- trial, food-reward classical conditioning. To construct the model CGC neuron they used a combination of fitting to available voltage-clamp data and parameter optimization techniques for determining maximal conductances and dynamic parameters. Remarkably, optimization led to tight ranges for most parameters, though some time constant parameters were not tightly constrained. A model based on median parameter values captured CGC depolarization without altered excitability, Please cite this article in press as: Lamb DG, Calabrese RL. Small is beautiful: models of smal Current Opinion in Neurobiology 2012, 22:1–6 observed in the living system, and identified two critical maximal conductances in this process; simultaneous increases in maxgNaP, a persistent sodium current, and maxgD, a delayed rectifier. Heartbeat in Hirudo The heartbeat CPG of Hirudo has been analyzed [17] and modeled, for example [18], extensively. Weaver et al. [19], presented a model of the entire core CPG that shows the utility of systematic parameter variation of a small subset of parameters in network analysis. Two bilateral pairs of premotor interneurons, phased differently with respect to the timing kernel of the CPG, show a phase progression on one side and near synchrony on the other. These phase differences are achieved by blending inhibitory synaptic input and electrical coupling. Making simplifying assumptions based on symmetry in the network, these investigators probed a range of maximal conductances for the inhibitory synapse and the electrical coupling to each premotor interneuron. The analysis established parameter values that produced model activity within the range observed in a large number of preparations. The relative parameter values predicted by this model were confirmed physiologically in voltage-clamp exper- iments [19] (see Figure 1). The control of motor neurons by this CPG has also been extensively analyzed and modeled [20,21,22 � ]. Garcia et al. [23] used averaged data to define input phasing and synaptic strength profiles and a simplified single compartment model for the motor neurons. That model failed to achieve quantitatively accurate average output phasing [23,24]. An analysis of 12 preparations for the timing of the activity of the four premotor interneurons of the CPG (input phasing) and of their strength of inhibi- tory output (synaptic strength profiles) onto motor neurons, and the timing (output phasing) of the motor neurons revealed wide animal-to-animal variability in synaptic strengths and input and output phasing [20]. Wright and Calabrese [22 � ], used input phasing, synaptic strength profiles and phasing targets from individual animals as inputs to the Garcia model [23], but were still not able to achieve greater accuracy. Those results, as well as dynamic clamp experiments also using individualized synaptic input patterns, indicated the involvement of motor neuron intrinsic properties not encompassed by the motor neuron model, thus leading to the conclusion that a multi-compartmental motor neuron model with more sophisticated intrinsic properties was required [21,22 � ]. Respiration in vertebrates The respiratory CPG of vertebrates has been studied in detail at many levels. It is a complex system consisting of millions of neurons with a great diversity of intrinsic and synaptic properties [25]. Moreover, the CPG is influ- enced by myriad extrinsic inputs, primarily related to l neuronal networks, Curr Opin Neurobiol (2012), doi:10.1016/j.conb.2012.01.010 www.sciencedirect.com http://dx.doi.org/10.1016/j.conb.2012.01.010 Small is beautiful Lamb and Calabrese 3 CONEUR-1039; NO. OF PAGES 6 Please cite this article in press as: Lamb DG, Calabrese RL. Small is beautiful: models of small neuronal networks, Curr Opin Neurobiol (2012), doi:10.1016/j.conb.2012.01.010 Figure 1 HN 1,2 3,4 6,7 6,7 5 5 3,4 1,2 HN silent 2 s HN(P,4) HN(P,3) HN(P,6) HN(P,7) HN(S,4) HN(S,3) HN(S,6) HN(S,7) HN(S,7) HN(S,6) HN(S,5) HN(R,4) HN(S,3) HN(P,7) HN(P,6) HN(P,4) HN(P,3) Peristaltic Synchronous 2 s50 mV 80 70 60 50 40 30 20 10 0 1 80 70 60 50 40 30 20 10 0 1 10 80 70 60 50 40 30 20 10 0 1 10 80 70 60 50 40 30 20 10 0 1 10 HN(R,7) SynchronousHN(L,7) Peristaltic HN(R,6) SynchronousHN(L,6) Peristaltic Ipsilateral Electrical Coupling Ip si la te ra l I n h ib ito ry S yn a p tic S tr e n g th g S yn S (n S ) g S yn S (n S ) gCoup (nS)gCoup (nS) 10 Expt. Range Only Expt. Avg. ± 1.5*SD Both Constraints Met (a) (b) (c) (d) HN(S,5) Current Opinion in Neurobiology (a) Bilateral activity (recorded extracellularly) in the premotor heart interneurons (HN(3), HN(4), HN(6) and HN(7) interneurons) of the core heartbeat CPG showing these neurons in peristaltic (P) and synchronous (S) coordination modes. The middle spike of the peristaltic HN(4) interneuron is used as a reference to compute phase: vertical dashed lines ease comparison of relative (unilateral) phase in the two coordination modes. The bilateral record is artificially reconstructed from a unilateral recording that switched between coordination modes and aligned so that peristaltic and synchronous HN(4) interneurons fire out of phase (0.5). (b) Circuit diagram showing synaptic connections among interneurons of the core heartbeat CPG. Small colored/ black circles indicate inhibitory chemical synapses, and diodes indicate rectifying electrical junctions. For simplicity, in the CPG diagram, cells with similar input and output connections and function are combined. Only one HN(5) interneuron is rhythmically active at a time, and it determines synchronous coordination ipsilaterally and peristaltic coordination contralaterally. (c) Covarying maximal conductances for the inhibitory synapse and the electrical coupling to identify appropriate values (colored asterisks and arrows). For every combination of maximal conductances, phase and duty cycle for that middle premotor interneuron were calculated and compared with those of individual neuron’s experimentally recorded values (see graph legend). (d) Core CPG model activity (bilateral) with parameters selected in (c). After Weaver et al. [19]. www.sciencedirect.com Current Opinion in Neurobiology 2012, 22:1–6 http://dx.doi.org/10.1016/j.conb.2012.01.010 4 Microcircuits CONEUR-1039; NO. OF PAGES 6 relevant respiratory signals such as pCO2. A large-scale computational model of the respiratory CPG [26] suc- cessfully reproduced data from arterially perfused brainstem-spinal cord rat preparations in which tran- sections sequentially removed rostral components of the respiratory network. That model was based on populations of Hodgkin–Huxley style neurons in the Bötzinger complex (BötC), pre-Bötzinger complex (pre-BötC), and ventral respiratory group, with excit- atory drive from the pons and retrotrapezoid nucleus (RTN). Building on this work, Rubin et al. [27��,28], reduced the complexity and employed the small network approach used for invertebrate models. The authors used activity- based, non-spiking, single neuron models to represent populations of spiking neurons, allowing for the appli- cation of bifurcation analysis. The reduced model repro- duced the three major dynamic regimes observed in previous experimental studies and large-scale models [26,29], supporting the idea that the proposed architec- ture and drive structure are reasonable. Similar to exper- imental studies using brain stem transections, the removal of pontine drive converted the initial three-phase oscillations characteristic of in vivo respiration to two- phase oscillations, which lack the post-inspiratory phase, characteristic of en bloc preparations that retain rostral nuclei. Subsequent removal of inhibition from BötC expiratory neurons associated with removal of tonic drive from RTN converted the two-phase oscillations to one- phase inspiratory oscillations characteristic of slice prep- arations of the pre-BötC. Two oscillatory mechanisms appear to underlie these oscillatory patterns at the cellular level, in particular in the pre-BötC complex: a persistent sodium-driven oscil- lation and a calcium-driven oscillation. Two recent papers [30,31 � ] present models which include both Nap and Ca driven oscillatory behavior, although the specifics of the models presented differ. An important aspect of the Toporikova and Butera model [31 � ] is that several import- ant responses to neuromodulator, pharmacological, and environmental influences are replicated, despite its rela- tively simple construction. Food processing in Crustaceans The food processing CPGs in the stomatogastric nervous system (STN) of decapod crustaceans continue to be a focus of modeling studies with wide implications for how neuronal networks achieve functional output. Early last decade, researchers began to use the STG system to address questions of model degeneracy, where the same functional output resulted from models with widely dif- fering underlying parameters. This raised the question of how reliable network activity as well as sensitivity to neuromodulation and perturbation could be achieved in these small neural networks. Please cite this article in press as: Lamb DG, Calabrese RL. Small is beautiful: models of smal Current Opinion in Neurobiology 2012, 22:1–6 Recent investigation by Grashow et al. [32], used a sim- plified neuronal model to approach the broad question of how underlying neural parameters contribute to overall network performance. The authors coupled a Morris Lecar model neuron with a pharmacologically isolated living STG neuron via dynamic clamp, and then varied the maximal conductance of the artificial synapses and a model Ih current injected into the STG neuron. Echoing Prinz [3], the results showed that diverse parameter values can lead to similar network output. In other cases, however, the same parameter values can also result in wildly different network output, showing that differences between the intrinsic properties of the biological neurons can drastically alter the resulting pattern in the hybrid network. Thus, network activity is more resilient to variations in some regions of parameter space than others, strongly supporting earlier modeling work [1]. Nadim et al. [33�], also used a simplified model to inves- tigate how a specific synapse influences network activity. Their stripped down model of a primary STN pacemaker network, composed of the anterior burster (AB) and pyloric dilator (PD) neurons, allowed them to apply phase plane analysis to investigate the role of the only known chemical synapse onto this network, the lateral pyloric (LP) to PD synapse. In the living system, the removal of this synapse has no effect under control conditions, although it was proposed that the LP-PD synapse would stabilize the AB/PD cycle period [34,35]. Nadim et al. [33 � ], verified this experimentally, and then used their model to help explain how. Essentially, the synapse reinforces the stability of the pacemaker by overriding the influence of perturbations — either slowing down incipient advances or speeding up incipient delays. Moving from highly reduced to more complicated models, Taylor et al. [36��], uses what is variously referred to as ensemble [6], family [4], or population modeling. In this approach, many similar models are considered and each instance, or individual, is a different combination of free parameter values. They constructed a biophysical baseline model, tuned to a subset of experimental data, and then explored the parameter space around this model. By randomly sampling the parameter space and then simulating and perturbing each instance, they could identify models which were acceptable across all of several metrics. Those acceptable models had widely differing parameters, but did not have the strong corre- lations between these parameters that were expected based on experimental correlations between channel mRNA [37–39]. Conclusions Each of these models has advantages and limitations, but all contribute to our understanding of neuronal net- works — from circuit-specific findings, such as putative new members, to broad conclusions about the likely l neuronal networks, Curr Opin Neurobiol (2012), doi:10.1016/j.conb.2012.01.010 www.sciencedirect.com http://dx.doi.org/10.1016/j.conb.2012.01.010 Small is beautiful Lamb and Calabrese 5 CONEUR-1039; NO. OF PAGES 6 structure and regulation of network topology and cellular properties. We were only able to present a selection of small network models, leaving out many other fascinating systems which are actively modeled, especially neuronal circuits underlying locomotion [40–45,46 � ,47 � ]. The abil- ity of small network models to facilitate a mechanistic understanding of such a broad array of complex systems speaks to their heuristic power. As we move forward, models of small networks will no doubt clarify many interesting issues. For example, although some studies find that variability of intrinsic properties at the cellular level becomes less important at the network level [32], others suggested that network topology and neural dynamics strongly interact and both are of critical importance [48]. Furthermore, we know that neuromodulation plays a key role in pattern generation in many systems, yet much of the research on the small neural networks which underlie behavior appears to touch only superficially on the 2nd messenger systems involved. The levels of regulation available to neurons are myriad: epigenetics, transcription and translation regulatory mechanisms, splice variation, interrupted or delayed translation, and post-translational modification. The next logical step is to extend our investigations into the 2nd messenger pathways which drive observable change in electrophysiological properties. Not only will a more complete understanding of the cellular pathways which influence electrophysiological activity help us understand the unperturbed state of small neural networks, but also the effects of and interactions between neuromodulators and extrinsic perturbations. In addition to expanding the complexity of our models to include more cellular com- plexity, the ensemble modeling approach is likely to expand as we move forward, especially given the ever- increasing computational capabilities available. Models of small networks are beautiful windows into how neuronal networks in all animals function, and we look forward to exciting new models that will continue to expand and challenge our understanding of them. Acknowledgements Funding: NS024072 to RLC. Angela Wenning and Eleanore D. Sternberg for proofreading. References and recommended reading Papers of particular interest, published within the period of review, have been highlighted as: � of special interest �� of outstanding interest 1. Goldman MS, Golowasch J, Marder E, Abbott LF: Global structure, robustness, and modulation of neuronal models. J Neurosci 2001, 21:5229-5238. 2. Prinz AA, Billimoria CP, Marder E: Alternative to hand-tuning conductance-based models: construction and analysis of databases of model neurons. J Neurophysiol 2003, 90:3998-4015. Please cite this article in press as: Lamb DG, Calabrese RL. Small is beautiful: models of smal www.sciencedirect.com 3. Prinz AA, Bucher D, Marder E: Similar network activity from disparate circuit parameters. Nat Neurosci 2004, 7:1345-1352. 4. Marder E: Variability, compensation, and modulation in neurons and circuits. Proc Natl Acad Sci U S A 2011, 108(Suppl. 3):15542-15548. 5. Marder E, Taylor AL: Multiple models to capture the variability in biological neurons and networks. Nat Neurosci 2011, 14:133-138. 6. Prinz AA: Computational approaches to neuronal network analysis. Philos Trans R Soc Lond B: Biol Sci 2010, 365:2397-2405. 7. Getting PA, Dekin MS: Mechanisms of pattern generation underlying swimming in Tritonia. IV. Gating of central pattern generator. J Neurophysiol 1985, 53:466-480. 8. Getting PA: Emerging principles governing the operation of neural networks. Annu Rev Neurosci 1989, 12:185-204. 9. �� Calin-Jageman RJ, Tunstall MJ, Mensh BD, Katz PS, Frost WN: Parameter space analysis suggests multi-site plasticity contributes to motor pattern initiation in Tritonia. J Neurophysiol 2007, 98:2382-2398. This paper builds on the small network model of Getting (1989) that galvanized the field of small network modeling. It applies modern approaches to enhance this venerable model and comes to the realization that, although Getting model may lack physiological verisimilitude, it was definitely heuristic. 10. Getting PA: Reconstruction of small neural networks. Methods in Neuronal Modeling. MIT Press; 1989:. pp. 171–194. 11. Kemenes G, Staras K, Benjamin PR: Multiple types of control by identified interneurons in a sensory-activated rhythmic motor pattern. J Neurosci 2001, 21:2903-2911. 12. Benjamin PR, Staras K, Kemenes G: A systems approach to the cellular analysis of associative learning in the pond snail Lymnaea. Learn Mem 2000, 7:124-131. 13. Benjamin PR, Kemenes G, Kemenes I: Non-synaptic neuronal mechanisms of learning and memory in gastropod molluscs. Front Biosci 2008, 13:4051-4057. 14. Vavoulis DV, Straub VA, Kemenes I, Kemenes G, Feng J, Benjamin PR: Dynamic control of a central pattern generator circuit: a computational model of the snail feeding network. Eur J Neurosci 2007, 25:2805-2818. 15. Cataldo E, Byrne JH, Baxter DA: Computational model of a central pattern generator. Comput Methods Syst Biol Proc 2006, 4210:242-256. 16. � Vavoulis DV, Nikitin ES, Kemenes I, Marra V, Feng J, Benjamin PR, Kemenes G: Balanced plasticity and stability of the electrical properties of a molluscan modulatory interneuron after classical conditioning: a computational study. Front Behav Neurosci 2010, 4:19. This paper provides a very interesting example of using parameter optimization techniques to generate a family of model instances. Para- meters of the difference instances are then compared to determine constrained and unconstrained parameters and a baseline model based on median values generated and mechanistically analyzed. 17. Kristan WB Jr, Calabrese RL, Friesen WO: Neuronal control of leech behavior. Prog Neurobiol 2005, 76:279-327. 18. Doloc-Mihu A, Calabrese R: A database of computational models of a half-center oscillator for analyzing how neuronal parameters influence network activity. J Biol Phys 2011, 37:263-283. 19. Weaver AL, Roffman RC, Norris BJ, Calabrese RL: A role for compromise: synaptic inhibition and electrical coupling interact to control phasing in the leech heartbeat CpG. Front Behav Neurosci 2010:4. 20. Norris BJ, Wenning A, Wright TM, Calabrese RL: Constancy and variability in the output of a central pattern generator. J Neurosci 2011, 31:4663-4674. 21. Wright TM Jr, Calabrese RL: Patterns of presynaptic activity and synaptic strength interact to produce motor output. J Neurosci 2011, 31:17555-17571. l neuronal networks, Curr Opin Neurobiol (2012), doi:10.1016/j.conb.2012.01.010 Current Opinion in Neurobiology 2012, 22:1–6 http://dx.doi.org/10.1016/j.conb.2012.01.010 6 Microcircuits CONEUR-1039; NO. OF PAGES 6 22. � Wright TM Jr, Calabrese RL: Contribution of motoneuron intrinsic properties to fictive motor pattern generation. J Neurophysiol 2011, 106:538-553. This paper investigates the often overlooked contribution of motor neu- rons in shaping motor output using both modeling as well as hybrid dynamic clamp experiments. The authors found that the influence of intrinsic properties is mostly dominated by the synaptic strength profile and electrical coupling, but that key attributes of motor neurons in the living system contribute meaningful phase shifts to the motor output. 23. Garcia PS, Wright TM, Cunningham IR, Calabrese RL: Using a model to assess the role of the spatiotemporal pattern of inhibitory input and intrasegmental electrical coupling in the intersegmental and side-to-side coordination of motor neurons by the leech heartbeat central pattern generator. J Neurophysiol 2008, 100:1354-1371. 24. Wenning A, Norris BJ, Doloc-Mihu A, Calabrese RL: Bringing up the rear: new premotor interneurons add regional complexity to a segmentally distributed motor pattern. J Neurophysiol 2011, 106:2201-2215. 25. Smith JC, Abdala AP, Rybak IA, Paton JF: Structural and functional architecture of respiratory networks in the mammalian brainstem. Philos Trans R Soc Lond B: Biol Sci 2009, 364:2577-2587. 26. Rybak IA, Abdala AP, Markin SN, Paton JF, Smith JC: Spatial organization and state-dependent mechanisms for respiratory rhythm and pattern generation. Prog Brain Res 2007, 165:201-220. 27. �� Rubin JE, Shevtsova NA, Ermentrout GB, Smith JC, Rybak IA: Multiple rhythmic states in a model of the respiratory central pattern generator. J Neurophysiol 2009, 101:2146-2165. This simplified network model of a mammalian respiratory CPG captures the gross structure of the living system’s CPG and can replicate many pharmacological and physiological results, in particular changes to the rhythmic states when different nuclei are disconnected by transection. The authors then thoughtfully examine this model in detail, and explore how various components of the model contribute to important phenom- ena. This paper demonstrates the efficacy and utility of the small network approach in studying complex vertebrate neuronal networks. 28. Rubin JE, Bacak BJ, Molkov YI, Shevtsova NA, Smith JC, Rybak IA: Interacting oscillations in neural control of breathing: modeling and qualitative analysis. J Comput Neurosci 2011, 30:607-632. 29. Smith JC, Abdala AP, Koizumi H, Rybak IA, Paton JF: Spatial and functional architecture of the mammalian brain stem respiratory network: a hierarchy of three oscillatory mechanisms. J Neurophysiol 2007, 98:3370-3387. 30. Dunmyre JR, Del Negro CA, Rubin JE: Interactions of persistent sodium and calcium-activated nonspecific cationic currents yield dynamically distinct bursting regimes in a model of respiratory neurons. J Comput Neurosci 2011, 31:305-328. 31. � Toporikova N, Butera RJ: Two types of independent bursting mechanisms in inspiratory neurons: an integrative model. J Comput Neurosci 2011, 30:515-528. The authors present a model of inspiratory pre-Bötzinger complex neu- rons which extends an earlier single-compartmental model to include two compartments and two oscillatory interactions: a calcium and a NaP driven oscillator. The resulting model can account for previously contra- dictory experimental pharmacology while incorporating neuromodulatory influences from norepinephrine, hypoxia, and Ca modulation. 32. Grashow R, Brookings T, Marder E: Compensation for variable intrinsic neuronal excitability by circuit–synaptic interactions. J Neurosci 2010, 30:9145-9156. 33. � Nadim F, Zhao S, Zhou L, Bose A: Inhibitory feedback promotes stability in an oscillatory network. J Neural Eng 2011, 8:065001. The greatly simplified model used in this paper allows for a beautiful example of the utility of phase plane analysis in developing mechanistic explanations for neuronal network function. 34. Mamiya A, Nadim F: Dynamic interaction of oscillatory neurons coupled with reciprocally inhibitory synapses acts to stabilize the rhythm period. J Neurosci 2004, 24:5140-5150. Please cite this article in press as: Lamb DG, Calabrese RL. Small is beautiful: models of smal Current Opinion in Neurobiology 2012, 22:1–6 35. Thirumalai V, Prinz AA, Johnson CD, Marder E: Red pigment concentrating hormone strongly enhances the strength of the feedback to the pyloric rhythm oscillator but has little effect on pyloric rhythm period. J Neurophysiol 2006, 95:1762-1770. 36. �� Taylor AL, Goaillard JM, Marder E: How multiple conductances determine electrophysiological properties in a multicompartment model. J Neurosci 2009, 29:5573-5586. The authors take a population approach, generating a database of closely related models. Not only were the models produced better than previous models, but the population approach also allowed for a more detailed understanding of the interaction between membrane conductances in the generation of functional activity. 37. Goaillard JM, Taylor AL, Schulz DJ, Marder E: Functional consequences of animal-to-animal variation in circuit parameters. Nat Neurosci 2009, 12:1424-1430. 38. Schulz DJ, Goaillard JM, Marder EE: Quantitative expression profiling of identified neurons reveals cell-specific constraints on highly variable levels of gene expression. Proc Natl Acad Sci U S A 2007, 104:13187-13191. 39. Tobin AE, Cruz-Bermudez ND, Marder E, Schulz DJ: Correlations in ion channel mRNA in rhythmically active neurons. PLoS One 2009, 4:e6742. 40. Ritzmann RE, Buschges A: Adaptive motor behavior in insects. Curr Opin Neurobiol 2007, 17:629-636. 41. Buschges A, Scholz H, El Manira A: New moves in motor control. Curr Biol 2011, 21:R513-R524. 42. Lamb D, Calabrese R: Neural circuits controlling behavior and autonomic functions in medicinal leeches. Neural Syst Circuits 2011, 1:13. 43. Mullins OJ, Hackett JT, Buchanan JT, Friesen WO: Neuronal control of swimming behavior: comparison of vertebrate and invertebrate model systems. Prog Neurobiol 2011, 93:244-269. 44. Grillner S, Jessell TM: Measured motion: searching for simplicity in spinal locomotor networks. Curr Opin Neurobiol 2009, 19:572-586. 45. Lockery SR: The computational worm: spatial orientation and its neuronal basis in C. elegans. Curr Opin Neurobiol 2011, 21:782-790. 46. � Borisyuk R, Al Azad AK, Conte D, Roberts A, Soffe SR: Modeling the connectome of a simple spinal cord. Front Neuroinformatics 2011, 5:20. Using a wealth of physiological and anatomical data and probabilistic developmental rules, the authors make a model, based on axonal growth and connection, of the connectome of spinal cord neurons involved in swimming in hatchling Xenopus tadpoles. The end result is about 1500 neurons of six different cell-types with a total of about 120 000 connec- tions. Each repetition of the modeling process results in a different connectome instantiation, each conforming to experimentally determined distributions of cell bodies, dendrites, and axon lengths. The model, because it can generate multiple instantiations, provides an ideal match to the ensemble modeling approach and shows a way forward for modeling the complex neuronal networks of vertebrates. 47. � Roberts A, Li WC, Soffe SR: How neurons generate behavior in a hatchling amphibian tadpole: an outline. Front Behav Neurosci 2010, 4:16. The authors summarize an extensive body of behavioral, physiological, and anatomical data on the behavioral repertoire and its neuronal bases of hatchling Xenopus tadpoles. Models (consisting of only 7 cell types) of the neuronal networks underlying swimming and struggling behaviors are presented and mechanistically compared. Combining such models with the connectome model of Borisyuk et al. (2011) in the future promises significant new insights into complex spinal neuronal net- works. 48. Gaiteri C, Rubin JE: The interaction of intrinsic dynamics and network topology in determining network burst synchrony. Front Comput Neurosci 2011, 5:10. l neuronal networks, Curr Opin Neurobiol (2012), doi:10.1016/j.conb.2012.01.010 www.sciencedirect.com http://dx.doi.org/10.1016/j.conb.2012.01.010 Small is beautiful: models of small neuronal networks Introduction A note on models and parameters Swimming in Tritonia Feeding in Lymnea Heartbeat in Hirudo Respiration in vertebrates Food processing in Crustaceans Conclusions Acknowledgements References and recommended reading