|
|
||||||||
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
1Center for Theoretical Biological Physics; 2Department of Physics; 3Graduate Program in Neurosciences, University of California San Diego, La Jolla, California; 4Department of Physiology and Zlotowski Center for Neuroscience, Ben Gurion University of the Negev, Beer-Sheva; and 5Department of Neurobiology, The Weizmann Institute of Science, Rehovot, Israel
Submitted 12 July 2005; accepted in final form 26 October 2005
| ABSTRACT |
|---|
| INTRODUCTION |
|---|
|
We consider the hypothesis that the essential determinants of the latency coding observed in POm thalamus depend on both features of the internal circuitry, i.e., the strong facilitation as well as delayed and prolonged response of GABAB-mediated synaptic transmission (Kim et al. 1997
), and features of the external stimulus, i.e., the gradually increasing spike rate of the brain stem input to thalamic nuclei. Our challenge is to address this hypothesis in terms of a neural-based model and account for the differences in the response of neurons in POm versus VPm thalamus to periodic stimulation. We ask: why does the latency in the response of POm neurons to prolonged stimuli increase substantially with frequency during steady state conditions? Why is there a distinct difference in latency coding for POm versus VPm thalamus? And why is there no latency coding for brief stimuli? A distinctive feature of our approach is the use of reduced models that are amenable to analytical treatment as a means to gain intuition into the role of synaptic versus external inputs as well to gain insight into the stability and sensitivity of solutions of the model.
| MODEL |
|---|
The architecture of the model includes the three thalamic nuclei: POm, VPm, and Rt. The Rt nucleus receives fast excitatory input from both POm and VPm nuclei. In turn, the POm and VPm nuclei receive inhibitory feedback from Rt thalamus as a fast GABAA- and delayed and prolonged GABAB-mediated current (Fig. 1A). As a means to obtain insight into the neuronal dynamics and to develop a set of robust predictions, we examine a succession of models of increasing complexity. We begin with a reduced but analytically tractable circuit comprising only the POm and the Rt nuclei (Fig. 1B). Then, we investigate the effects of feedforward input from the VPm to the Rt (Fig. 1C). Finally, we study the full thalamic architecture that further includes feedback inhibition from the Rt to the VPm nucleus (Fig. 1D).
We note that the difference in latency coding between the VPm and POm thalamus is also reflected by neurons in cortical layers that receive axonal projection from these nuclei, i.e., layers IV and Va, respectively (Ahissar et al. 2000
, 2001
; Ahrens et al. 2002
). We focus on feedback dynamics at the level of the thalamus. We test if GABAB-mediated currents and the brain stem stimulus shape are sufficient to explain the observed differences in latency.
Measured spike rates
The measured peristimulus time histograms (PSTHs) in brain stem nuclei Pr5 and SpVI and in the thalamic nuclei VPm and POm, based on population-average data from all the recorded multiunits (Ahissar et al. 2000
, 2001
; Sosnik et al. 2001
), show different steady-state responses to periodic air-puff stimuli, each 50 ms in duration, delivered at frequencies of fstim = 2, 5, 8, and 11 Hz (Fig. 1A). The population-averaged temporal responses for both the Pr5 and the Sp5I nuclei of the brain stem may be described as a dual-sloped ramp in which a brief, rapidly rising phase is followed by a prolonged, slowly rising phase. The neuronal activity then gradually decays after the activity reaches its maximum value. In contrast to the prompt response in brain stem, the activity of units in POm thalamus are delayed and then rise to their maximal value. Critically, the response rises more gradually to its peak value as the stimulus frequency, fstim, is increased (cf. 2- and 11-Hz data in Fig. 1A), so that the rise time is, very roughly, inversely proportional to fstim, whereas the maximal activity decreases with fstim. In contrast to the case for POm thalamus, the activity of units in VPm thalamus begins with shorter delay and then rises rapidly to a peak. The rise time is almost independent of fstim, but the maximal activity decreases with fstim.
Rate model
Our technical approach makes use of rate equations (Hopfield 1982
; Wilson and Cowan 1973
), where the instantaneous spike rate of a population of neurons is expressed in terms of the presynaptic activation for each synaptic current (Ermentrout 1994
; Kang et al. 2003
; Rinzel and Frankel 1992
; Shriki et al. 2003
). The activation parameters are denoted by subscripted ui(t), where the subscript i is used to denote the presynaptic neuronal population. For inhibitory conductances, a second subscript, A or B, denotes whether the synapse is mediated by GABAA or by GABAB, respectively. Thus as examples, uPOm(t) describes activation of the excitatory current that originated from POm neurons and uRt,B(t) describes activation of the slow inhibitory current that originates from the Rt nucleus. We further note the aggregate measure of the strength of the synaptic connection of the projection from nucleus j to nucleus i by the synaptic conductance gi;j. For example, gPOm;Rt,B denotes the GABAB-mediated input from Rt to POm thalamus. The activation parameters ui(t), multiplied by the synaptic conductances gi;j, determine the total synaptic current to the postsynaptic cell, e.g., gPOm;Rt,BuRt,B(t) for GABAB-mediated input to POm neurons from Rt thalamus.
NONLINEAR, DELAY DIFFERENTIAL EQUATIONS THAT DEFINE THE DYNAMICS.
The observable quantities in a network are the instantaneous spike rates for each population of neurons. This rate is denoted by Mi for nucleus i, e.g., MPOm for the POm nucleus. Without intrinsic adaptation, the presynaptic spiking rate Mi is approximated by the instantaneous input-output relation, or f-I curve, Mi(t) = [I(t)]+, and is taken to be the rectification (linear-threshold) function, that is
![]() | (1) |
The total current I(t) is the sum of the synaptic currents and the brain stem input, i.e., IPOm(t) and IVPm(t) for the POm and VPm nuclei, respectively, relative to the threshold for spiking, e.g.,
POm for neurons in POm thalamus. Thus the instantaneous spiking rates of the three thalamic nuclei are
![]() | (2) |
![]() | (3) |
![]() | (4) |
The dynamics for glutamatergic fast excitatory synapses and GABAA-mediated inhibitory synapses, but not GABAB-mediated inhibitory synapses, are defined by pairs of delay differential equations
![]() | (5) |
![]() | (6) |
![]() | (7) |
![]() | (8) |
G,
G, and tG are the synaptic rise, decay, and delay times, respectively, of the excitatory synapses,
A,
A, and tA are the synaptic rise, decay, and delay times, respectively, of the GABAA-mediated synapses, the variable xi is an auxiliary variable, and the subscript i is POm thalamus or VPm thalamus. The synaptic delays depend only on the synaptic phenotype in our architecture.
Inhibitory synapses mediated by GABAB are nonlinear and facilitating. In practice, they have a delay of
3040 ms and respond much more strongly to a prolonged burst of spikes than to a brief burst (Golomb et al. 1996
; Kim et al. 1997
). We use a version of a nonlinear model for GABAB-mediated synapses (Golomb et al. 1996
)
![]() | (9) |
![]() | (10) |
B,
B, and tB are the synaptic rise, decay, and delay times, respectively, of the GABAB-mediated synapses. The quadratic nonlinearity in the first term in the right-hand side of Eq. 10 is responsible for the facilitating nature of the synapse.
ADAPTATION.
We incorporate an idealization of cellular adaptation in the rate equations for the VPm and POm thalamic nuclei with an adaptation time scale that is much slower than the spiking time scale. A thalamic neuron fires a small number of spikes in response to a single stimulus in the relay mode (Minnery and Simons 2003
; Sosnik et al. 2001
). In particular, cells in POm thalamus rarely fire more than one spike in response to a single stimulus (Sosnik et al. 2001
). We therefore model adaptation as a process of inactivation and slow recovery of the "neuronal pool" that is capable of spiking (Eggert and van Hemmen 2000
, 2001
). Such a process can be described as a multiplicative dynamical process with two time scales, one of activation as a result of neuronal activity and one of inactivation when neurons are silent. The equations for the adaptation variables ai (Eqs. 2 and 3) and the auxiliary variables bi, identified with the activation dynamics, are
![]() | (11) |
![]() | (12) |
We assume that POm thalamus has stronger adaptation than VPm thalamus; this allows us to model the fast decay of POm activity at low frequencies in comparison to the more sustained activity of VPM neurons (Fig. 1A). The adaptation in the VPm neurons rises faster than that in POm neurons as a result of the higher spiking rate in the former nucleus. The activity in the Rt nucleus is more prolonged than the activity in thalamic relay nuclei (Hartings and Simons 2000
), and therefore we do not introduce adaptation for the Rt nucleus.
STIMULUS SHAPE.
The input from the brain stem nuclei is considered as an external variable that monotonically tracks the stimulus. We do not discriminate between inputs from the Pr5 and SP5I nuclei as the average responses of neurons to air-puff stimuli are very similar in the two areas (Ahissar et al. 2000
). In both dorsal thalamic nuclei, the input from the brain stem begins after a short delay that represents the onset of the stimulus and the time that the stimulus drives the thalamic nuclei. We model the "double ramp" shape of the brain stem input to VPm thalamus, IVPm, as a piecewise linear function. The brain stem input to POm thalamus is proportional to the brain stem input to the VPm, and it is delayed to allow larger latency in POm even for low stimulus frequencies
![]() | (13) |
The parameters of the brain stem stimuli to the two thalamic nuclei are given in APPENDIX A.
PARAMETERS.
We use the following parameters throughout the analysis unless stated otherwise: thresholds:
VPm =
POm =
Rt = 0; excitatory synapses:
G = 1 ms,
G = 2 ms, tG = 0 (Golomb and Amitai 1997
); GABAA synapses:
A = 1 ms,
A = 10 ms, tA = 3 ms; GABAB synapses:
B = 40 ms,
B = 150 ms, tB = 35 ms (Golomb et al. 1996
; Kim et al. 1997
); synaptic conductances: gVPm;Rt,A = 0.5, gVPm;Rt,B = 2, gPOm;Rt,A = 0.16, gPOm;Rt,B = 3.5, gRt;VPm = 1.2, gRt;POm = 1; and adaptation: ka,POm = 0.33 ms1, ka,VPm = 0.1 ms1,
b,VPm = 10 ms;
b,POm = 30 ms, kb,POm = kb,VPm = 0.05 ms1, and
a,VPm =
a,POm = 100 ms.
Derived quantities
SPIKE NUMBER.
The function MPOm(t) (Eq. 3) is proportional to the instantaneous spiking rate of POm neurons. A measure that is of further utility in our analysis is the total number of spikes that are fired during a time interval by these neurons. This number is proportional to NPOmspikes, where
![]() | (14) |
![]() | (15) |
LATENCY.
The onset latency, t0, is computed in our analytical treatment. We further define the time between the start of the stimulus and the time that the instantaneous spiking rate, Mi(t), reaches half of its maximal value at the midpoint latency, as t0.5. This second definition is consistent with that used to quantify the latency in the experimental observations (Ahissar et al. 2000
, 2001
; Sosnik et al. 2001
).
Computation
The delay differential equations and the auxiliary equations that define the model (Eqs. 112) are solved numerically by the Euler method with a time step of
t = 0.02 ms.
| RESULTS |
|---|
GABAB-mediated inhibition has slow kinetics and facilitates with increased presynaptic spiking. Under steady-state conditions, the slow kinetics cause a rhythmic input to lead to a delayed inhibition of thalamic cells. For a rhythmic stimulus, the extent of the delay will depend on the rate of rise of the brain stem input and the frequency of the stimulus. If the rise is gradual, facilitation will transform an increasing rate of stimulus-induced spiking into an increased latency in the response of thalamic cells because higher spike-rates lead to stronger inhibition. The model enables us to assess the consequences and stability of this mechanism and the specific role of the three thalamic nuclei.
Our first goal is to gain qualitative insight into the role of GABAB-mediated inhibition and the stimulus shape in the production of a frequency-dependent latency. To achieve this goal, we simplify the rate model to obtained a reduced, analytically solvable model. The analysis is extended to circuits with increasing level of complexity. Finally, the outcome of the reduced model is compared with simulations of the full rate model to demonstrate that the approximations used for the reduction do not disrupt the basic dynamical mechanisms. The simulations further allow us to make a detailed comparison between experimental and model results as well as to make predictions for the outcome of future experimental investigations.
Reduced model of the thalamic circuit
A simplified model is obtained through the inclusion of four assumptions. First, the GABAB-mediated decay-time
B is larger than all the other synaptic time constants in the system. Therefore with the exception of the GABAB-mediated delay (which plays a special role in the dynamics) and decay, all of the synaptic processes are considered to be instantaneous. This allows us to focus on the special role of GABAB-mediated currents in shaping the thalamic dynamics. In particular, we set tPOm,delay = 0 in Eq. 13 because this time constant is much smaller than the GABAB decay time so that IPOm(t) =
IVPm(t). Further, since the time constants of the fast excitatory synapses (
G,
G, and tG) are set to zero (Eqs. 5 and 6), we replace uVPm by MVPm and replace uPOm by MPOm in the expressions for the firing rate (Eqs. 24).
The second assumption concerns adaptation. Because adaptation does not affect the thalamic response at the onset of activity, we simply ignore it and take aPOm = aVPm = 0.
The third set of assumptions concern the fast conductances. GABAA-mediated inhibition is ignored because this inhibition has fast decay rate, which is now taken to be instantaneous. Therefore inhibition generated in response to a given stimulus decays before the next stimulus arrives and does not participate in generating the latency. The final assumption is that, for simplicity, the thresholds
i are set to 0 for all nuclei i.
Based on the first assumption, the GABAB-mediated conductance in the Rt nucleus is simplified by taking the onset to be instantaneous, i.e.,
B = 0 (Eq. 9). This allows us to substitute xRt(t) by MRt,B(t tB) (Eqs. 8 and 9). This yields
![]() | (16) |
To realize analytical treatment, we treat only cases where Tstim = 1/fstim >2tB. Furthermore, because the rise time of the brain stem stimulus-evoked response is
50 ms (Fig. 1A), which is only slightly greater than the delay of GABAB-mediated synapses, tB, we take tB to equal the stimulus duration. Further, to demonstrate the importance of this stimulus shape for the latency in a manner that avoids the algebraic complexity associated with a "double-ramp" treatment, we consider first a triangular stimulus to mimic the slowly rising phase and next a square stimulus to mimics the fast rising phase. The shapes of stimuli are, for 0
t < Tstim
![]() | (17) |
![]() | (18) |
The value NPOmspikes, proportional to the number of spikes fired by POm neurons, is (Eq. 14)
![]() | (19) |
Analytical results for reduced POm-Rt circuits
POm-Rt CIRCUIT FOR TRIANGULAR STIMULUS: ESSENTIAL FEATURES FOR CODING STIMULUS FREQUENCY BY RESPONSE LATENCY.
We consider first a minimal circuit with feedback between the POm and Rt nuclei, in addition to input from the brain stem to POm thalamus (Fig. 1B). The dynamics of the GABAB activation variable, uRt,B(t), are determined by a single delay-differential equation with five parameters (Eq. 16) and the expression for the triangular stimulus (Eq. 17). Three of these parameters are time constants, i.e., the time period of the stimulus Tstim, the stimulus duration tB, and the GABAB decay rate
B. The other two parameters are the POm-to-Rt excitatory conductance gRt;POm and the Rt-to-POm GABAB-mediated inhibitory conductance gPOm;Rt,B. The dynamics of activity in POm thalamus are determined by the difference between the excitatory stimulus IPOm(t) and the GABAB-mediated inhibition gPOm;Rt,B uRt,B(t) (Fig. 2A). Thus the latency for activation of POm neurons corresponds to the time when the difference between the stimulus and the inhibition becomes positive, namely (APPENDIX B, Eq. B2)
![]() | (20) |
![]() | (21) |
|
![]() | (22) |
![]() | (23) |
3.6 for the parameters of Fig. 2C. Last, although the reduced model has multiple parameters, the value and stability of the fixed point depends only on the products gRt,POm2gPom,Rt,B, gPOm;Rt,BuRt,B(0), and gRt,POm2/uRt,B(0). Thus an increase in gRt;POm can be offset by a decrease in gPOm;Rt,B and a rescaling of the value of the activity at the fixed point. The latency t0 is a monotonic function of gPOm;Rt,B up to the point of period doubling (Fig. 3A). The number of spikes per cycle NPOmspikes thus decreases with increasing stimulus frequency (Eq. 21; Fig. 2B). For values of gPOm;Rt,B that permit period doubling, i.e., 3.6 < gPOm;Rt,B < 7.1 for the parameters of Figs. 2C and 3A, the latency alternates with a period of two stimulus cycles. More complex behavior emerges for still larger values of gPOm;Rt,B. In general, we can evaluate the stability of the solution with a single stimulus cycle for arbitrary values of stimulus frequency, tB, and gPOm;Rt,B (Fig. 3B). This solution is stable for all frequencies for small gPOm;Rt,B and is unstable for stimulus frequencies above, roughly, fstim = 2 Hz for large values of gPOm;Rt,B. The dependence of the latency t0 on the stimulus frequency fstim and gPOm;Rt,B is shown in Fig. 3C. For values of gPOm;Rt,B that do not lead to period doubling, the latency increases with fstim and can be even somewhat larger than tB/2 but not close to tB. Thus the minimal model captures the essential scaling of increased latency with increased stimulation frequency, albeit for a constrained set of parameter values.
|
|
![]() | (24) |
To gain insight into the role of VPm input to Rt thalamus, and to examine latency as a function of the coupling between these nuclei gRt;VPm, we first consider the circuit with POm-to-Rt feedback excitation turned off, i.e., gRt;POm = 0. For this case, the steady state with the period of the stimulus is always stable, but the output of POm thalamus is silent above the value gRt;VPm*. For both triangular (Eq. 17) and rectangular stimuli (Eq. 18), the spiking activity decays gradually from a maximal value to zero; this is shown for fstim = 8 Hz in the examples of Fig. 5, A and B. For the triangular stimulus, the latency increases gradually from 0 to tB (Fig. 5A). In contrast, for the rectangular stimulus, the latency remains zero for an extended interval of values of gRt;VPm that satisfy gPOm;Rt,B u(0)
1; note that u(0) increases gradually with gRt;VPm. The latency then jumps steeply and reaches t0 = 1 at gRt;VPm*, for which gPOm;Rt,B u(0) = exp(tB/
B) (Fig. 5B). Therefore for a rectangular stimulus, the latency increases with frequency only for a restricted range of values of gRt;VPm for which POm thalamic activity is very small. This implies that a gradual increase of the stimulus activity with time is necessary to code stimulus frequency by latency for this purely feedforward architecture.
|
|
-1gPOm;Rt,B for VPm versus POm thalamus, respectively. The former value should be the smaller of the two to ensure that the latency in VPm thalamus is essentially constant. Numerical results for the full POm-Rt-VPm circuit
The analytical theory suggests an explanation for the latency coding. Still, several questions still remain open. First, are the analytical results valid despite the approximations that were made? Second, what is the effect of the dual-sloped rising phase of the brain stem stimulus, as observed experimentally (Fig. 1A), as opposed to just a triangular shape? Third, can we explain the detailed shape of the PSTHs recorded in the POm and the VPm nuclei? Specifically, why does the POm activity decay rapidly with time after the initial onset at low stimulus frequency, whereas such decay is seen neither in the POm in response to high-frequency stimulation nor in the VPm?
To answer the preceding questions and to draw a close comparison between responses calculated for the model and those observed in experiments, we have carried out numerical simulations of the full rate model (APPENDIX A) with the architecture of Fig. 1D. We incorporate inhibitory feedback that satisfies gPOm;Rt,B > gVPm;Rt,B; adaptation in the activity of POm thalamus, and weaker adaptation in VPm thalamus, to account for the reduction in the population of neurons that are available to spike after each stimulus (Eggert and van Hemmen 2001
2000
); and a dual-sloped rising phase for the stimulus (Fig. 7, A and B).
|
GABAB-mediated inhibition by Rt thalamus becomes strongly activated with increased values of the stimulation frequency above fstim = 2 Hz (Fig. 7C). The excitatory stimulus minus the GABAB-mediated inhibition is positive only near the peak of the stimulus, where the stimulus rises gradually, as in the case of a triangular stimulus. As a result, the activity in POm thalamus is strongly suppressed just after the onset of the stimulus (Fig. 7E). With increasing time, the brain stem input overcomes the GABAB-mediated inhibition and the activity in POm rises until the stimulus ends. For times
5090 ms after the stimulus onset, the GABAB synaptic activation is balanced by the decreasing adaptation and the activity in POm thalamus is essentially independent of fstim (Fig. 7E). These dynamics result in a calculated instantaneous spike rate whose latency to peak and overall amplitude compares favorably with those in the experimental data (cf. Figs. 7E with 1A).
With regard to VPm thalamus, the slow GABAB-mediated inhibition reduces the overall activity with increasing stimulus frequency (Fig. 7F). Yet, as a result of a larger input from the brain stem (
= 0.6 in Eq. 13) and the weaker GABAB-mediated inhibition compared with the case for POm thalamus, the excitatory stimulus minus the GABAB-mediated inhibition is positive everywhere except just after the onset of the stimulus. The reduction of activity by GABAB-mediated inhibition decreases the firing activity just following the stimulus onset but is insufficiently strong to affect the latency of the response. Thus there is negligible change in the latency to peak, with increasing stimulus frequency. As in the case of POm thalamus, the time dependence of the instantaneous spike rates in our simulations compares favorably with those in the data (cf. Figs. 7F with 1A).
COMPARISON OF THEORY WITH EXPERIMENT: LATENCY AND SPIKE COUNT. The calculated and observed latencies were expressed in terms of t1/2, which is the time for the instantaneous rates of spiking, MPOm(t) and MVPm(t), to reach one-half of their maximum value. The values of t1/2 computed from the model well approximate the observations for both the POm and VPm nuclei (Fig. 7, G and H). Further, the calculated increase in the value of t1/2 with increasing stimulus frequency for POm thalamus is a robust, monotonic function of gPOm;Rt,B up to the frequency where POm neurons are silenced by inhibition (Fig. 7I). In contrast to the close match of theory and observation for the latency effect, the calculated values of the number of spikes per cycle in POm thalamus decreases more sharply with frequency than in the observations (Figs. 7G). This deviation results from adaptation at low frequencies of stimulation that is relatively weak in the model. Nonetheless, the overall match between calculated and observed values (Fig. 7, EH) lends support to the idea that the heightened frequency-dependent latency and reduced spike-count in POm versus VPm thalamus results from a stronger GABAB-mediated inhibitory feedback, a weaker brain stem input for neurons in POm thalamus (Eq. 13), and the double-ramp, monotonically increasing brain stem stimulus.
EFFECT OF STIMULUS DURATION AND SHAPE. We seek to account for the effect of stimulus width on the latency effect in POm thalamus and further predict the nature of the spiking response in the POm and VPm nuclei for arbitrary stimuli. Intuitively, the strongly facilitating nature of the GABAB-mediated currents should make the rate of spiking sensitive to the duration of the stimulus.
We first consider variations in the total duration of the stimulus (Fig. 8A). The period of the initial rise was kept fixed but the duration of the second rising phase and the falling phase were co-varied. The duration was parameterized by a multiplicative factor, denoted
, where
= 1 is the value for a stimulus with a total rising phase of 50 ms (Fig. 8A1). The change in time course of MPOm(t) and MVPm(t) as a function of frequency is relatively weak for the case of a brief stimulus, i.e.,
= 0.4 for a rise-time of 23 ms (Fig. 8A, 2 and 3), compared with a long stimulus (Fig. 7, E and F). The calculated time course compares favorably with those recorded with a 20 ms stimulus (Fig. 7 in Sosnik et al. 2001
). In general, the range of the frequency-dependent latency for spiking in POm thalamus is predicted to increase as the duration of the stimulus increases toward its full width for
= 1 (Fig. 8A4).
|
| DISCUSSION |
|---|
Comparison with experimental results
Analysis of our model reveals that the latency increases with frequency in POm thalamus but not in the VPm thalamus if the brain stem input to the VPm is larger than the brain stem input to the POm and the GABABmediated inhibition is stronger in the POm than in the VPm. The first condition is clearly seen in in vivo experiments because the firing rate of VPm neurons in low stimulus frequencies is considerably higher in the VPm than in the POm (Diamond 1995
; Diamond et al. 1992
; Sosnik et al. 2001
). Furthermore, in recent in vitro experiments, C. E. Landisman and B. W. Connors (unpublished data) found that VPm neurons exhibit higher maximal firing rates from POm neurons when stimulated with injected current. These researchers also found that VPm cells were 2.5 times less likely than POm cells to have a GABAB receptor-like component as part of their inhibitory responses. Hence these recent experimental results support our view of the role of GABABmediated inhibition in coding frequency by latency in the POm but not in the VPm.
We find that strong feedback connections between POm and Rt thalamus destabilize the steady-state activity with the period of the stimulus via a period doubling bifurcation. Analysis of the experimental data (Sosnik et al. 2001
) (APPENDIX C) did not reveal any indication for activity with double period. Recent observations (Webber and Stanley 2004
) have revealed period doubling in cortical activity in response to stimuli with duty cycles of 0.5 and frequencies between 4 and 8 Hz. This behavior may reflect similar behavior in the thalamic input to cortex. The stimuli used by those authors are more prolonged than the stimuli analyzed here and may generate a strong GABAB-mediated inhibitory response.
The GABAB kinetics used in the model (Eqs. 9 and 10) are based on the results from dual recordings (Kim et al. 1997
). These observations show that a single presynaptic neuron should fire a high-frequency burst of action potentials to evoke a substantial GABAB-mediated response from the postsynaptic neuron. Indeed, Rt cells fire at a high rate in comparison with thalamic relay nuclei (Hartings et al. 2000
). Furthermore because GABABmediated synapses are metabotropic and involve the cooperative activation of G proteins (Destexhe and Sejnowski 1995
; Golomb et al. 1996
), near-synchronous spiking of several presynaptic cells may act cooperatively to facilitate the stronger GABAB-mediated response.
Consequences of approximations in our model
Single-cell properties may well affect the latency and adaptation effects in the POm nucleus. Although there is little data on the cellular properties of neurons in POm thalamus, the lateral posterior (LP) nucleus is an analogous nucleus in the visual stream (Peters 1985
) and the nonlemniscal nuclei are analogous nuclei in the auditory stream (Yu et al. 2004
). In vitro intracellular recordings from neurons in the LP nucleus revealed significant potassium A-type and calcium T-type currents (Li et al. 2003
). The A current can further delay the occurrence of the first spike in a stimulus cycle, whereas the combination of a T current and inhibitory input may contribute to the production of a burst of spikes followed by an afterhyperpolarization that terminates neuronal activity (Sherman 2001
; Steriade et al. 1993
). These cellular properties are likely to affect the details of our analysis of the full model (Fig. 7), but not our conclusions regarding the latency effect (Figs. 26).
The difference in latency coding between the VPm and POm thalamus is also reflected by neurons in cortical layers that receive axonal projection from these nuclei, i.e., layers IV and Va, respectively (Ahissar et al. 2000
, 2001
; Ahrens et al. 2002
). Cortical feedback will act to supplement the brain stem input to thalamic relay cells (Deschenes et al. 1998
; Diamond et al. 1992
), and especially to excite Rt thalamus (Golshani et al. 2001
). Recalling that thalamocortical (Gil and Amitai 1996
) and even corticothalamic (Swadlow 1994
) synaptic connections are much faster than the decay time of GABAB