letters to nature ... Feedback inhibition controls spike transfer in hybrid thalamic circuits

Transcription

1 16. Furst, J. et al. ICln ion channel splice variants in Caenorhabditis elegans. Voltage dependence and interaction with an operon partner protein. J. Biol. Chem. 277, (2002). 17. Culetto, E. & Sattelle, D. B. A role for Caenorhabditis elegans in understanding the function and interactions of human disease genes. Hum. Mol. Genet. 9, (2000). 18. Reinke, V. et al. A global profile of germline gene expression in C. elegans. Mol. Cell 6, (2000). 19. Jiang, M. et al. Genome-wide analysis of developmental and sex-regulated gene expression profiles in Caenorhabditis elegans. Proc. Natl Acad. Sci. USA 98, (2001). Supplementary Information accompanies the paper on Nature s website ( Acknowledgements We thank J. Spieth, J. Kent, A. Zahler and L. Stein for help with navigation of the C. elegans databases, Y. Kohara for cdna data, M. Huang for discussions, I. Shah for statistical advice, D. Guiliano and M. Blaxter for communication of unpublished results, and P. MacMorris for advice on the manuscript. This work was supported by the NIH (T.B., C.D.L. and S.K.K.). Competing interests statement The authors declare that they have no competing financial interests. Correspondence and requests for materials should be addressed to T.B. ( tom.blumenthal@uchsc.edu). arousal to sleep 1 5,8. Although much is known about the anatomy and the synaptic and cellular properties of the thalamic networks, the nature of the sensory information processing throughout selective arousal and sleep wake stages is not yet understood. The goal of this work was to investigate the mechanisms responsible for changes in the efficiency of sensory spike transfer in the retinothalamic network, during different states of arousal. We used hybrid biological neuromimetic networks that allow direct control of cellular and synaptic components 11. We measured the variations of spike-to-spike correlation between identified input and output neurons, reflecting the efficiency and reliability of signal transfer in different activity states. In our hybrid networks (Fig. 1a), synaptic-like interactions between realistic conductance-based model neurons and an intracellularly recorded biological neuron run in real time, following the natural dynamics of the biological cell or network. Individual membrane currents of the simulated and biological neurons and the properties of their synaptic connections can be selectively and quantitatively controlled throughout their dynamic range, in vitro or in vivo. The required speed of real-time computation is achieved by using both programmable digital signal processors (DSPs) and newly designed analog integrated circuits 11,12. A dynamic clamp procedure was used to simulate synaptic conductances by current injection through the intracellular recording pipette Feedback inhibition controls spike transfer in hybrid thalamic circuits Gwendal Le Masson*, Sylvie Renaud-Le Masson, Damien Debay & Thierry Bal * Laboratoire de Physiopathologie des Réseaux Neuronaux Médullaires, EPI INSERM 9914, Institut François Magendie, Université Victor Segalen Bordeaux 2, 1 Rue Camille Saint Saëns, Bordeaux Cedex, France Laboratoire IXL, CNRS UMR 5818, ENSEIRB, Université de Bordeaux 1, 351 Cours de la Libération, Talence Cedex, France Unité de Neurosciences Intégratives et Computationnelles, CNRS UPR 2191, Institut de Neurobiologie Alfred Fessard, 1 Avenue de la Terrasse, Gif-sur- Yvette Cedex, France... Sensory information reaches the cerebral cortex through the thalamus, which differentially relays this input depending on the state of arousal 1 5. Such gating involves inhibition of the thalamocortical relay neurons by the reticular nucleus of the thalamus 6 8, but the underlying mechanisms are poorly understood. We reconstructed the thalamocortical circuit as an artificial and biological hybrid network in vitro. With visual input simulated as retinal cell activity, we show here that when the gain in the thalamic inhibitory feedback loop is greater than a critical value, the circuit tends towards oscillations and thus imposes a temporal decorrelation of retinal cell input and thalamic relay output. This results in the functional disconnection of the cortex from the sensory drive, a feature typical of sleep states. Conversely, low gain in the feedback inhibition and the action of noradrenaline, a known modulator of arousal 4,9,10, converge to increase input output correlation in relay neurons. Combining gain control of feedback inhibition and modulation of membrane excitability thus enables thalamic circuits to finely tune the gating of spike transmission from sensory organs to the cortex. The thalamus is the major gateway for the flow of sensory information to the cerebral cortex. Far from being a passive relay, this structure actively processes information before cortical integration. It is the first stage at which sensory signals can be gated during selective attention or during the transition from general Figure 1 Design of hybrid thalamic circuits. a, Artificial synaptic connections between a biological TC cell recorded intracellularly in an LGNd slice and DSP-based and analog integrated circuit (IC) neurons. Wiring diagram in a ferret LGNd slice: þ, excitatory; 2, inhibitory. b, One-to-one coupling in ferret networks: a burst of spikes evoked in a single TC neuron can trigger burst firing of a target PGN neuron (not shown), which generates feedback inhibition (arrow) and rebound burst (asterisk). Middle, tetrodotoxin (TTX) block of PGN activity prevents feedback inhibition. Bottom, synaptic interaction between PGN and TC neurons leads to repetitive TC bursts. c, Hybrid circuit reconstruction using nrt/pgn model cell, in guinea-pig LGNd slices where TC cells are initially synaptically isolated: effect of incrementing nrt/pgn-mediated GABA conductance. Calibration bars, 0.1 s, 20 mv, 0.35 na Nature Publishing Group

2 GABA (g-aminobutyric acid)-releasing reticular interneurons (nucleus reticularis (nrt)/perigeniculate (PGN)) provided a strategically located inhibitory feedback to the corticopetal afferent sensory pathway (Fig. 1a), activated by recurrent collaterals of the thalamocortical (TC) cells and by corticothalamic projections 1,7 and modulated by brainstem afferents 1,9. We hypothesized that changes in the gain of reticular-mediated inhibition 14 affect the temporal correlation between input and output spikes, thus providing a mechanism for spike transfer control. We tested this hypothesis using canonical hybrid thalamic circuits of the lateral geniculate nucleus (LGN dorsal, LGNd). These consist of a biological TC neuron reciprocally connected to a model reticular interneuron. The marked similarity of the behaviour of the hybrid circuits compared with the known reciprocal interactions of the wholly biological TC nrt/pgn circuit validates the hybrid equivalent reconstruction (Fig. 1b, c). In the wholly biological circuit, synaptic interactions between nrt/pgn and TC cells can generate sustained network oscillations, known as spindle waves 5, a landmark of early stages of sleep. The basic properties of this intrathalamic loop (Fig. 1b) can be reproduced in the hybrid circuit with full control of the ionic conductances modulating synaptic strength (Fig. 1c). Progressive increase of the maximal conductance (G max ) of the nrt/pgn-to-tc GABA-mediated synapse (mixed GABA A / GABA B conductances) from 0 to 73 ns (n ¼ 14) increased the probability of rebound burst generation in the TC neuron (Fig. 1c). Oscillations resembling spindle waves (Fig. 1b, c) were obtained for a threshold value of GABA G max that showed little variation between TC cells (29 ^ 4.2 ns, mean ^ s.e.m.; n ¼ 9) (see Methods). We explored the signal transfer capability of thalamic loops by adding a simulated retinal input to the hybrid circuit. In vivo, individual TC neurons receive the bulk of their excitation from a single retinal ganglion cell 17. In darkness or under constant illumination, different types of ganglion cells discharge irregularly with a mean rate ranging from 5 to 60 Hz and interspike interval (ISI) distributions adequately modelled 18 as renewal processes with g- distributed intervals (see Methods). In hybrid circuits, analog model neurons were programmed to simulate the temporal structure of retinal cell discharges and the kinetics of real AMPA (aamino-3-hydroxy-5-methyl-4-isoxazole propionic acid) receptormediated excitatory postsynaptic potentials (EPSPs). Hybrid thalamic circuits receiving this realistically emulated synaptic bombardment spontaneously generated short epochs of oscillation (frequency 9.26 ^ 0.87 Hz; duration 1.74 ^ 0.36 s; n ¼ 27), recurring periodically in a manner very similar to biological spindle waves occurring in sleep-related states (Fig. 2). Each spindling epoch is terminated by an after-depolarization that is similar in the biological and the hybrid network. This after-depolarization, resulting from the hyperpolarization-activated current I h (ref. 19), GABA synapse Figure 2 Spontaneous spindle activity in sleeping hybrid retinothalamic circuit. a, Patterned firing (mean rate 42 Hz; g ¼ 1.5) of the analog retinal neuron generates an artificial synaptic bombardment in the biological TC cell. b, Detail of a. c, Total synaptic current injected into the biological TC cell and simultaneous voltage records. Outward synaptic current is downward. Horizontal dashed line indicates null net current. A positive AMPA current triggered by a retinal cell spike induced an EPSP in the biological TC cell (arrows). TC cell spikes triggered AMPA EPSPs and bursts of spikes in the nrt/pgn cell (circle). The latter triggered mixed GABA A /GABA B negative currents (asterisk) that summed in the TC cell. Calibration bars: 1 s, 20 mv (a); 0.3 s, 20 mv (b); 0.1 s, 20 mv, 1nA(c). Figure 3 The strength of inhibition regulates temporal correlation between input retinal and output TC cell spikes. a, High-conductance, mixed GABA A /GABA B (GABA G max ), nrt/ PGN-to-TC inhibitory synapse. Right panels, contribution index (2-ms bins). Asterisk, TC spikes triggered within a delay of,10 ms after retinal spike (5 Hz; g ¼ 3). b, c, Decreasing the inhibition increases the probability of input output correlation. d, e, Averaged contribution (CI, d) and correlation indices (CC, e) versus synaptic strength. Six screening experiments, 2-ms bins, input rate 21.6 ^ 6.9 Hz, g ¼ 3. Mann Whitney U-test, P, Calibration bars, 0.5 s, 20 mv Nature Publishing Group 855

3 sums to the synaptic bombardment to form a depolarization that causes inactivation of the low-threshold Ca 2þ current and subsequent block of input signal transfer. During this emulated sleep-like state, the firing pattern of TC cells, which is very different from the retinal cell input, suggests that the transfer of retinal cell spikes to the cortex is low (Fig. 2a). The question of whether this low transfer could nonetheless be reliable was examined by calculating two different indices. First, the contribution index (CI) examines the TC cell discharge and quantifies the percentage of spikes that are precisely correlated with retinal input spikes. It estimates the reliability with which a TC spike can be considered as being triggered by the input rather than being spontaneous. It is computed as the peak of the cross-correlation normalized by the number of TC spikes (output). Second, the correlation index (CC) measures the global efficiency of the input output spike transfer and indicates the ratio of input spikes being actually transmitted as output spikes in the TC neuron. It is computed as the peak of the cross-correlation between retinal and Figure 4 Noradrenaline (NA) enhances input output correlation in TC neurons. a, Application of NA 500 mm (horizontal bar) on biological TC. Retinal input, 30 Hz; g ¼ 3. b, Detail from a: left, sleep-like state, GABA G max ¼ 36 ns; middle, GABA G max ¼ 36 ns; right, under NA, GABA G max ¼ 0. Calibration bars: 10 s (a) or 200 ms (b), 20 mv. c, Normalized cross-correlation histograms (1-ms bins) for conditions in b. Left, low correlation; middle, increased correlation under NA; right, NA and no nrt/pgn inhibition. d, e, Transfer efficiency (CC, d) and contribution indices (CI, e) versus experimental conditions (averaged peaks of cross-correlation histograms, 1-ms bins, six cells). Asterisk, P, 0.05, nonparametric Wilcoxon matched-pairs test. g ¼ 3; 20 Hz for five retinal cells, 30 Hz for one cell; GABA G max ¼ 38 ^ 1 ns; AMPA G max ¼ 21 ^ 2 ns. f, Converging influences of the strength of feedback inhibition and membrane property changes of TC cells. TC neuron spikes, normalized by the number of retinal cell spikes (input). In the presence of strong inhibitory feedback (GABA G max ¼ 47 ns), the CI was low (Fig. 3a), indicating that most of the TC spikes were not correlated with the retinal input and thus that the thalamus was not transferring spikes in a one-to-one manner. We then systematically screened the strength of inhibition to test whether the degree of inhibition produced by nrt/pgn interneurons could directly control the precision of the input output temporal correlation. As the inhibitory synaptic strength (G max ) decreased, the spike-to-spike correlation gradually increased (Fig. 3a d). Thus, the recurrent nrt/pgn feedback inhibition has a direct de-correlating effect capable of reducing the reliability of spike transfer. Despite these highly significant inhibition-dependent variations of CI (Fig. 3d), the non-selective global efficiency of spike transfer, measured by the CC, remained low and was not significantly different in the presence or absence of feedback inhibition (Fig. 3e). This low transfer efficiency was predicted by computational models 20 and can be explained by the low-threshold calcium current that, at hyperpolarized membrane potentials, endows TC cells with low-pass filtering properties typical of sleep states 4,5. These results show that high gain of feedback synaptic inhibition strengthens the filtering properties of TC cells and ensures a nearly complete decorrelation of the output activity from the sensory input. The resulting combination of low reliability and low efficiency of spike transfer explains how sensory input may be disconnected within the thalamic circuit during sleep. In contrast, during arousal, reliable synaptic transmission and short-latency responses to afferent stimuli are required 4,5. It is well known that membrane properties of thalamic cells are modulated by neurotransmitters released by cortical, brainstem and hypothalamic afferents in the thalamus 9,10, switching their activity from bursting mode during sleep-related states to an activity dominated by tonic firing during circadian or phasic episodes of arousal 4,5,9. The transition from sleep to waking can be mimicked in vitro by local application of noradrenaline 21, known to increase the excitability of relay TC neurons and to decrease the gain of intrathalamic inhibition 22 through activation of postsynaptic a1- and b-adrenoreceptors 4,5,9,22. Here, we used hybrid circuits to investigate how the modulation of membrane properties alters the balance of cell cell interactions, and thus also affects input output spike correlation (Fig. 4). Noradrenaline increased both spike transfer efficiency and reliability, measured respectively by CC and CI (Fig. 4b e; compare left with middle and right cross-correlation histograms in Fig. 4b, and bars in Fig. 4d and e), even when associated with a high value of feedback inhibition from nrt cells. Furthermore, TC spike latency became shorter and less variable (control, 5.3 ^ 1.3 ms, versus noradrenaline, 2.8 ^ 0.3 ms, P, 0.05). These effects were often associated with a tonic depolarization of the TC cell (Fig. 4a). In this neurotransmitter-dependent activated state, spike transfer efficiency (CC) could be further enhanced by removing the nrt/ PGN-mediated inhibition (Fig. 4b d; compare middle and right cross-correlation histograms in Fig. 4c and bars in Fig. 4d) and in four cells this increased the reliability estimate (CI; Fig. 4e). In the presence of noradrenaline, the remaining two cells tended to generate supplementary uncorrelated spikes. When synaptic inhibition was absent, this resulted in an input/output spike ratio of greater than one, which decreased CI. This suggests that the presence of feedback inhibitory postsynaptic potentials (IPSPs) in these cells can make transfer more reliable by filtering out spikes not correlated with the simulated retinal input. Our results show that hybrid thalamic circuits sample and filter input spikes with a temporal precision that is regulated by both the strength of intrathalamic inhibitory synapses and the state of membrane properties of TC neurons (Fig. 4f). It is probable that in vivo these combined mechanisms constitute a versatile control Nature Publishing Group

4 system that allows precise up- and downregulation of signal transfer, over a continuum from sleep to arousal. These results were obtained in a reduced retinothalamic circuit, but generate principles that are applicable to the larger-scale thalamic network. Filtering of the global retinal input can be seen as a result of the interactive operation of similar canonical circuits reiterated across the retinotopic space. Functional and anatomical studies describe significant overlap between parallel sensory channels through convergent and divergent synaptic connections both between the thalamic cell types 14, and within corticothalamic feedback projections. Crosstalk between channels is a functional variable, permitting the synchronization of the activity of large populations of thalamic cells during slow-wave-sleep 4,5,23 or epileptic absence seizures 4,24,25. We therefore propose that during these states of low perceptual awareness, input output spiking decorrelation caused by nrt/pgn-mediated inhibition could be generalized across the entire thalamic network, thus contributing to reduced sensory perception. In the transition from sleep to arousal, activation of TC and nrt/ PGN cells by extrinsic neuromodulatory influences decreases their ability to fire in burst mode 5,9,22. This, in turn, weakens feedback inhibition of TC cells 14,26 and ultimately blocks slow recurrent circuit oscillations 21. We postulate that, in arousal, decreased inhibition across the network results in a generalized increase in input output spike correlation in many parallel sensory channels. This would tend to shift the time constant of integration of sensory input to the millisecond range. Such a change in network behaviour might also act in a more restricted local manner, contributing to focal attention 8 processes. The gain of nrt/pgn-mediated inhibition could also be tuned down in restricted parts of the thalamic network by corticothalamic or brainstem afferents. This would provide a mechanism for selective sensory attention by creating islands of highly efficient sensory spike transfer in a network otherwise decoupled from its inputs. A Methods Biological preparations Biological neurons were recorded at C from slices 25 of guinea-pig (1 4 months old) or ferret (2 15 months old) LGNd. Intracellular records were made with glass micropipettes filled with 1.2 M potassium acetate ( MQ after bevelling 25 ). Drugs were applied locally by the pressure-pulse technique 25. Hybrid networks were built in guinea-pig preparations in which TC neurons were synaptically isolated from the extrinsic afferent inputs (including nrt) in the process of making the LGNd slice. This was translated by the complete absence of spindle waves and conferred the important advantage that it was not necessary to use pharmacological blockers, which might also have affected the response properties of the cells. Mean input resistance of guinea-pig TC cells was 68.1 ^ 3.1 MQ (n ¼ 19). Drifts in membrane potential were corrected, if necessary, to maintain the threshold for action potentials measured during rebound bursts at ^ 0.6 mv (n ¼ 11). Spindle generation in biological and hybrid networks Spindle waves, generated in thalamic circuits, appeared in the electroencephalogram as 7 14-Hz oscillations that waxed and waned in amplitude over a 2 4-s period and reappeared once every 3 10 s 4,5. Knowledge of their mechanisms, acquired in the ferret LGNd slice 14 16, was used to design and validate our hybrid circuits. GABA-mediated PGN neurons generated direct inhibitory feedback to TC neurons, which resulted in intermittent activation of low-threshold calcium spikes and associated bursts of action potentials. In turn, as in the hybrid network, rebound bursts of TC cells reactivated the TC PGN TC feedback loop, thus entraining the circuit in spindle wave oscillation. In the hybrid circuit, rhythmic IPSPs activated a rebound burst in the biological cell at each cycle of spindle oscillation. Similar rebound bursting can be observed in the purely biological circuit (Fig. 1b, bottom trace). Implementation of model neurons and synapses Hybrid interaction between biological neurons and models 11 was based on a dynamic clamp 13 procedure. The principal advantage of the hybrid network was the ability to selectively and quantitatively control membrane properties and synaptic gain (Fig. 1c), throughout their dynamic range, in a systematic manner. Real-time connections require fast computation, which was achieved with DSP boards, coupled with digital-to-analog and analog-to-digital (8 bits, 128 khz) conversion stages (Innovative Integration) and controlled by dedicated software. Numerical models of thalamic neurons were singlecompartment conductance-based models Analog silicon integrated circuits, performing real-time processing independently of the model s complexity, were developed using a full custom current-mode design with 1.2 mm BiCMOS technology 12. These multipurpose analog neurons were conductance-based realistic models implemented on programmable, application-specific integrated circuits (ASICs). Synaptic input from a model cell to a biological neuron is achieved by generating a voltage signal proportional to a synaptic current according to the equation I syn ¼ G syn (t,v m )(V r 2 E s ). This voltage signal was converted to current in an Axoclamp 2B (Axon Instruments) amplifier used in discontinuous current-clamp mode and fed to the recorded neuron via the intracellular microelectrode. G syn is the synaptic conductance expressed as a function of time and presynaptic model membrane potential (V m ); unless stated in the text, in control conditions G syn values were (in ns): GABA A G max, 35; GABA B G max, 1.4; TC to nrt AMPA G max, 20; and retino-tc AMPA G max, 28. The formalism and parameters describing the dynamics of G syn for AMPA and GABA synapses were based on previous models 28,29 that include a kinetic scheme of neurotransmitteur release. V r is the biological postsynaptic neuron membrane potential. E s is the reversal potential for the ion species involved and was set at (in mv): Na þ, þ55; Cl 2, 290; and K þ, To reproduce natural intraspindle frequency and duration, the parameter space of GABA receptor subtypes (GABA A G max versus GABA B G max ) was screened systematically and given an optimal conductance ratio of 96% GABA A and 4% GABA B (n ¼ 18), consistent with previous findings 4,5,14 16,25,26. The AMPA-type excitatory synaptic inputs from retinal ganglion cells to TC, as well as from TC to nrt, were simulated using the model of Destexhe et al. 28. Because inputs from retinal ganglion cells are located near the soma 1, on the proximal dendrites of TC cells, artificial somatic injection using dynamic clamp is a realistic procedure for mimicking retinal inputs. Simulated recording at a proximal dendritic site 35 mm from the soma of a 200-compartment TC model cell, shows that AMPA receptor-mediated EPSPs generated in the soma or the dendrite give almost identical responses (A. Destexhe, unpublished simulation). Thus, artificial AMPA EPSPs efficiently trigger low-threshold calcium spikes because the distribution of T (transient) Ca 2þ channels is non-uniform and highest in proximal dendrites 30. The ISI distribution of retinal ganglion cells can be accurately described by an Erlang renewal process, where the width of the distribution is set by the g order 18.Ahighg-value leads to highly regular ISI, whereas lower g-values generate irregular ISIs. Renewal processes with g orders within the physiological range (g ¼ 0.7 to 12) 18 were used to generate ISIs for our analog retinal cells. Received 3 December 2001; accepted 8 April 2002; doi: /nature Sherman, S. M. & Guillery, R. W. Functional organization of thalamocortical relays. J. Neurophysiol. 76, (1996). 2. Coenen, A. M. L. & Vendrick, A. J. H. Determination of the transfer ratio of cat s geniculate neurons through quasi-intracellular recordings and the relation with the level of alertness. Exp. Brain Res. 14, (1972). 3. Livingstone, M. S. & Hubel, D. H. Effects of sleep and arousal on the processing of visual information in the cat. Nature 291, (1981). 4. Steriade, M., Jones, E. G. & McCormick, D. A. Thalamus, Organization and Function (Elsevier Science, Oxford, 1997). 5. Steriade, M., McCormick, D. A. & Sejnowski, T. J. Thalamocortical oscillations in the sleeping and aroused brain. Science 262, (1993). 6. Yingling, C. D. & Skinner, J. E. Gating of thalamic input to cerebral cortex by nucleus reticularis thalami. Attention, voluntary contraction and event-related cerebral potentials. Prog. Clin. Neurophysiol. 1, (1977). 7. Ahlsen, G., Lindström, S. & Lo, F. S. Interaction between inhibitory pathways to principal cells in the lateral geniculate nucleus of the cat. Exp. Brain Res. 58, (1985). 8. Montero, V. M. Amblyopia decreases activation of the corticogeniculate pathway and visual thalamic reticularis in attentive rats: a focal attention hypothesis. Neuroscience 91, (1999). 9. McCormick, D. A. Neurotransmitter actions in the thalamus and cerebral cortex and their role in neuromodulation of thalamocortical activity. Prog. Neurobiol. 39, (1992). 10. Aston-Jones, G., Chiang, C. & Alexinsky, T. Discharge of noradrenergic locus coeruleus neurons in behaving rats and monkeys suggests a role in vigilance. Prog. Brain. Res. 88, (1991). 11. Le Masson, G., Le Masson, S. & Moulins, M. From conductances to neural network properties: analysis of simple circuits using the hybrid network method. Prog. Biophys. Mol. Biol. 64, (1995). 12. Le Masson, S., Laflaquière, A., Bal, T. & Le Masson, G. Analog circuits for modeling biological neural networks: design and applications. IEEE Trans. Biomed. Eng. 64, (1999). 13. Sharp, A. A., O Neil, M. B., Abbott, L. F. & Marder, E. Dynamic clamp: computer generated conductances in real neurons. J. Neurophysiol. 69, (1993). 14. Kim, U. & McCormick, D. A. The functional influence of burst and tonic firing mode on synaptic interactions in the thalamus. J. Neurosci. 18, (1998). 15. McCormick, D. A. & Bal, T. Sleep and arousal: thalamocortical mechanisms. Annu. Rev. Neurosci. 20, (1997). 16. von Krosigk, M., Bal, T. & McCormick, D. A. Cellular mechanisms of a synchronized oscillation in the thalamus. Science 261, (1993). 17. Usrey, W. M. & Reid, R. C. Synchronous activity in the visual system. Annu. Rev. Physiol. 61, (1999). 18. Troy, J. B. & Robson, J. G. Steady discharges of X and Y retinal ganglion cells of cat under photopic illuminance. Vis. Neurosci. 9, (1992). 19. Bal, T. & McCormick, D. A. What stops synchronized thalamocortical oscillations? Neuron 17, (1996). 20. Mukherjee, P. & Kaplan, E. Dynamics of neurons in the cat lateral geniculate nucleus: in vivo electrophysiology and computational modeling. J. Neurophysiol. 74, (1995). 21. Lee, K. H. & McCormick, D. A. Abolition of spindle oscillations by serotonin and norepinephrine in the ferret lateral geniculate and perigeniculate nuclei in vitro. Neuron 17, (1996). 22. Funke, K., Pape, H. C. & Eysel, U. T. Noradrenergic modulation of retinogeniculate transmission in the cat. J. Physiol. (Lond.) 463, (1993). 23. Contreras, D., Destexhe, A., Sejnowski, T. J. & Steriade, M. Control of spatiotemporal coherence of a thalamic oscillation by corticothalamic feedback. Science 274, (1996). 24. Snead, O. C. III Basic mechanisms of generalized absence seizures. Ann. Neurol. 37, (1995) Nature Publishing Group 857

5 25. Bal, T., Debay, D. & Destexhe, A. Cortical feedback controls the frequency and synchrony of oscillations in the visual thalamus. J. Neurosci. 20, (2000). 26. Kim, U., Sanchez-Vives, M. V. & McCormick, D. A. Functional dynamics of GABAergic inhibition in the thalamus. Science 278, (1997). 27. McCormick, D. A. & Huguenard, J. R. A model of the electrophysiological properties of thalamocortical relay neurons. J. Neurophysiol. 68, (1992). 28. Destexhe, A., Mainen, Z. F. & Sejnowski, T. J. Synthesis of models for excitable membranes, synaptic transmission and neuromodulation using a common kinetic formalism. J. Comput. Neurosci. 3, (1994). 29. Destexhe, A., Bal, T., McCormick, D. A. & Sejnowski, T. J. Ionic mechanisms underlying synchronized oscillations and propagating waves in a model of ferret thalamic slices. J. Neurophysiol. 76, (1996). 30. Williams, S. R. & Stuart, G. J. Action potential backpropagation and somato-dendritic distribution of ion channels in thalamocortical neurons. J. Neurosci. 15, (2000). Acknowledgements We are grateful to K. Grant, Y. Fregnac, S. Oliet, F. Nagy, A. Destexhe, M. Rudolph and B. Gutkin for in-depth discussion and comments on the manuscript; G. Sadoc, N. Gazère and E. Barbe for their technical input; and A. Destexhe for theoretical simulations. This research was supported by the Groupement d Intérêt Scientifique Sciences de la Cognition, Centre National de la Recherche Scientifique, Institut National de la Santé et de la Recherche Médicale, Plan Pluriformation du Ministère de la Recherche, Fondation pour la Recherche sur l Epilepsie, and the Institut Electricité Santé de France. Competing interests statement The authors declare that they have no competing financial interests. Correspondence and requests for materials should be addressed to T.B. ( thierry.bal@iaf.cnrs-gif.fr).... Na 1 /H 1 exchanger regulatory factor 2 directs parathyroid hormone 1 receptor signalling In most cells, the activated PTH1R increases cyclic AMP robustly through stimulation of adenylyl cyclase, and increases hydrolysis of inositol phosphates only modestly through stimulation of phospholipase C (PLC) 1,6. The PTH1R couples to stimulatory G proteins (G s ) and protein of the G q subfamily, but also couples to G i/o protein(s), as shown by pertussis-toxin blockade of PTH activation of PLC and increased activation of adenylyl cyclase In COS7 (African green monkey kidney) cells, pertussis-toxin sensitivity was observed only with intact PTH1R, but not with PTH1R, which had a carboxy-terminal deletion of 111 amino acids 11. In addition, PTH increased camp accumulation in cells expressing the truncated PTH1R by 4 6-fold that of cells expressing full-length PTH1R, but PTH-stimulated increases of inositol phosphates were unaffected. These findings suggest that the C-terminal tail contains determinants that modulate signalling by the PTH1R. Using the C-terminal, cytoplasmic tail of the PTH1R as bait in a yeast two-hybrid screen of a human kidney complementary DNA library, we found a strong interaction with a NHERF2 fragment, termed 12A. 12A includes PDZ2 and the C-terminal ERM-binding region, but lacks amino-terminal sequences, including PDZ1. It contains a consensus splice sequence and 105 extra-exonic bases, which are spliced in-frame and encode 35 amino acids N-terminal to amino acid 139 of NHERF2. 12A is most probably an incompletely processed messenger RNA, because the 5 0 region could not be identified by 5 0 rapid amplification of cdna ends (RACE) using a3 0 primer that included a portion of 12A. The high homology between 12A and NHERF2, however, led us to study NHERF2 PTH1R interactions. NHERF2 contains two PDZ (PSD95, discs large protein, ZO1) domains that assemble target proteins by binding to a C-terminal motif having a consensus sequence (D/E)-(S/T)-X-(L/V) 12 and a C-terminal ERM-binding domain that mediates interactions with the actin-binding proteins ezrin, radixin and moesin 13. NHERF2 Matthew J. Mahon*, Mark Donowitz, C. Chris Yun & Gino V. Segre* * Endocrine Unit, Massachusetts General Hospital and Department of Medicine, Harvard Medical School, Boston, Massachusetts 02114, USA Department of Medicine, Gastroenterology Division, The Johns Hopkins University School of Medicine, Boston, Massachusetts 21205, USA... The parathyroid hormone 1 receptor (PTH1R) is a class II G- protein-coupled receptor 1. PTH1R agonists include both PTH, a hormone that regulates blood calcium and phosphate, and PTHrelated protein (PTHrP), a paracrine/autocrine factor that is essential for development, particularly of the skeleton. Adenylyl cyclase activation is thought to be responsible for most cellular responses to PTH and PTHrP, although many actions appear to be independent of adenylyl cyclase 1 5. Here we show that the PTH1R binds to Na 1 /H 1 exchanger regulatory factors (NHERF) 1 and 2 through a PDZ-domain interaction in vitro and in PTH target cells. NHERF2 simultaneously binds phospholipase Cb1 and an atypical, carboxyl-terminal PDZ consensus motif, ETVM, of the PTH1R through PDZ1 and PDZ2, respectively. PTH treatment of cells that express the NHERF2 PTH1R complex markedly activates phospholipase Cb and inhibits adenylyl cyclase through stimulation of inhibitory G proteins (G i/o proteins). NHERF-mediated assembly of PTH1R and phospholipase Cb is a unique mechanism to regulate PTH signalling in cells and membranes of polarized cells that express NHERF, which may account for many tissue- and cell-specific actions of PTH/PTHrP and may also be relevant to signalling by many G-proteincoupled receptors. Figure 1 In vitro interactions of PTH1R and NHERF1 and 2 are PDZ-domain specific. a, NHERF2 interacts with the C-terminus of PTH1R. Full-length NHERF2 was overlaid on a membrane containing GST, GST C-terminal tail of PTH1R and mutants lacking 20 C-terminal amino acids (CD20), and 20 (ND20), 40 (ND40), and 60 (ND60) N-terminal amino acids (top); Coomassie blue (bottom). b, NHERF2 binds to a C-terminal, PTH1R PDZ interaction motif. Full-length NHERF2 was overlaid on a membrane containing PTH1R GST C-tail (ETVM), and the C-tail with individual alanine substitutions (top, in bold); Coomassie blue (bottom). c, PTH1R interacts with PDZ2 of NHERF2. GST C-tail of PTH1R WT (top) or PTH1R-M591L (bottom) were overlaid on duplicate membranes containing full-length NHERF2, NHERF2 PDZ1, NHERF2 PDZ2 and clone 12A.d, NHERF1 binds PTH1R through a PDZ-domain-specific interactions. Full-length NHERF1 was overlaid on a membrane containing PTH1R GST C-tail (ETVM), and the C-tail with individual alanine substitutions (top); Coomassie blue (bottom) Nature Publishing Group