Dynamic properties of corticogeniculate excitatory transmission in the rat dorsal lateral geniculate nucleus in vitro

Abstract

The feedback excitation from the primary visual cortex to principal cells in the dorsal lateral geniculate nucleus (dLGN) is markedly enhanced with firing frequency. This property presumably reflects the ample short-term plasticity at the corticogeniculate synapse. The present study aims to explore corticogeniculate excitatory postsynaptic currents (EPSCs) evoked by brief trains of stimulation with whole-cell patch-clamp recordings in dLGN slices from DA-HAN rats. The EPSCs rapidly increased in amplitude with the first two or three impulses followed by a more gradual growth. A double exponential function with time constants 39 and 450 ms empirically described the growth for 5–25Hz trains. For lower train frequencies (down to 1Hz) a third component with time constant 4.8 s had to be included. The different time constants are suggested to represent fast and slow components of facilitation and augmentation. The time constant of the fast component changed with the extracellular calcium ion concentration as expected for a facilitation mechanism involving an endogenous calcium buffer that is more efficiently saturated with larger calcium influx. Concerning the function of the corticogeniculate feedback pathway, the different components of short-term plasticity interacted to increase EPSC amplitudes on a linear scale to firing frequency in the physiological range. This property makes the corticogeniculate synapse well suited to function as a neuronal amplifier that enhances the thalamic transfer of visual information to the cortex.

Thalamic neurones relay sensory information, encoded in spike trains, to the primary sensory areas of the cortex. In the visual system, the principal cells of the dorsal lateral geniculate nucleus (dLGN) forward activity from a small number of neighbouring retinal ganglion cells with circular on- or off-centre receptive fields to the primary visual cortex (Hubel & Wiesel, 1961). The sensory information is transmitted to spiny stellate neurones in cortical layer 4 and to pyramidal cells in layer 6. The latter have axons that project to layer 4 cells and back to the dLGN (Gilbert & Kelly, 1975; Gilbert & Wiesel, 1979; Ferster & Lindström, 1982). The spiny stellate cells represent the major point of entry to cortical processing, while the function of the layer 6 cells is less apparent (Ahlsén et al. 1982; Ferster & Lindström, 1985a, b; Lindström & Wróbel, 1990). It has been suggested that the corticogeniculate projection represents a feedback system that modulates the excitability of dLGN principal cells (Guillery & Sherman, 2002). More specifically it has been proposed to function as a variable neuronal amplifier that enhances the transfer of visual information to the cortex in the attentive state (Ahlsén et al. 1985;Ferster & Lindström, 1985b; Lindström & Wróbel, 1990; Granseth et al. 2002; Granseth & Lindström, 2003). This idea is supported by the finding that layer 6 cells readily fire action potentials up to 25Hz in aroused animals while their responses during sleep or anaesthesia are weak and capricious (Livingstone & Hubel, 1981).

One shared feature of the feedback originating in layer 6 in all cortical areas is a pronounced frequency-dependent enhancement of synaptic strength (Ferster & Lindström, 1985a, b; Lindström & Wróbel, 1990; McCormick & von Krosigk, 1992; Mishima & Ohta, 1992; Turner & Salt, 1998; Bartlett & Smith, 2002). Depending on the firing frequency, this synaptic enhancement can be ascribed to the actions of different phases of short-term synaptic plasticity, a characteristic of synapses in both the central and peripheral nervous system (Magleby, 1987; Zucker & Regehr, 2002). A fast and a slow component of facilitation as well as augmentation have so far been identified at the corticogeniculate synapse in the dLGN (Granseth et al. 2002; Granseth & Lindström, 2003, 2004).

The facilitation at the corticogeniculate synapse has previously been quantified using paired pulse stimulation protocols (Lindström & Wróbel, 1990; Turner & Salt, 1998; Granseth et al. 2002; Granseth & Lindström, 2003). However, such results are hard to extrapolate to the normal behaviour of the synapse since visual stimuli more often induce continuing firing in corticogeniculate neurones (Livingstone & Hubel, 1981). Their activity is maintained until the stimulus is moved out of their receptive field or, in the case of stationary stimuli, until the eyes move (Martinez-Conde et al. 2002). Train stimulation of the corticogeniculate pathway has been carried out before (Lindström & Wróbel, 1990; McCormick & von Krosigk, 1992; Turner & Salt, 1998) but without detailed quantification of the facilitation process. Such data are required to define the effective gain of the suggested neuronal amplifier.

Hence, the effects of repetitive stimulation of corticogeniculate axons were systematically investigated using patch-clamp recordings of dLGN principal cells in vitro. The present paper focuses on facilitation while augmentation is described in an accompanying paper (Granseth & Lindström, 2004). The results show that facilitation induces and maintains a manifold increase in EPSC amplitude that is Ca²⁺ dependent. The EPSC enhancement is linearly related to the firing frequency in the physiological range, a property that makes the corticogeniculate synapse particularly well suited for the suggested amplifier function.

Methods

Patch-clamp recordings

Brain slices were prepared from pigmented DA-HAN rats (BK Universal, Sollentuna, Sweden) of either sex, 23–37 days old, as previously described (Granseth et al. 2002). The animals were anaesthetized with halothane (ISC Chemicals, Avonmouth, UK) and decapitated. All procedures were approved by the Committee for Ethics in Animal Research of Linköping, in accordance with Swedish animal-welfare legislation. Slices containing the dLGN, 250–300μm thick, were cut in a plane that preserves the distal ramifications of retinogeniculate and corticogeniculate fibre tracts (Turner & Salt, 1998). After at least 1h of incubation at 37°C the slices were transferred to the recording chamber. Individual principal cells in the dLGN were identified using an Axioskop FS microscope (Zeiss, Jena, Germany) with water immersion objectives and infrared differential phase contrast optics and an infrared digital camera (C7500, Hamamatsu, Hamamatsu City, Japan). Whole-cell patch-clamp recordings were made with the EPC9 amplifier (HEKA Elektronik, Lambrecht, Germany) with slices submerged in Krebs solution (mm: NaCl, 124; NaH₂PO₄, 1.25; NaHCO₃, 26; KCl, 3.0; MgCl₂ 2.0; CaCl₂, 2.0; and glucose, 10; at 34°C equibrilated with 95% O₂ in 5% CO₂) as previously described (Granseth et al. 2002). Picrotoxin (100 μm) and dl-2-amono-5-phosphonovaleric acid (DL-APV, 100 μm) obtained from Sigma, St Louis, MO, USA, were routinely included to block GABA_A and NMDA receptor currents. Cyclothiazide (Sigma) was added to the medium to final concentrations of 50–100μm from a 100 mm stock solution in DMSO to hinder AMPA receptor desensitization in some experiments. When (S)-α-methyl-4-carboxyphenylglycine [(S)-MCPG] or (RS)-α-cyclopropyl-4-phosphonophenylglycine [(RS)-CPPG], both obtained from Tocris Cookson (Bristol, UK), were used to antagonize metabotropic glutamate receptors, they were added to final concentrations of 100 and 200μm from 10 and 100 mm stock solutions in Krebs solution and NaOH, respectively. To minimize changes in nerve fibre excitability and maintain a high-quality gigaseal, [Mg²⁺]_o was adjusted on an equimolar basis to changes in [Ca²⁺]_o as previously described (Granseth et al. 2002). Borosilicate glass microelectrodes (Clark Electromedical, Reading, UK) were filled with a caesium gluconate-based buffer (mm: caesium gluconate, 100; NaCl, 10; Hepes, 10; TEA-Cl, 20; QX-314, 5.0; EGTA, 0.10; and MgATP 1.0; pH adjusted to 7.3 and osmolality adjusted with H₂O from hyperosmolar to 300 mmol kg⁻¹). Tip resistance was 3–6MΩ. Seals were initially 1.0–2.5GΩ and after rupturing the cell membrane access resistance was less than 25MΩ. Neurones were voltage clamped at −70 mV, adjusted for a liquid junction potential of 8 mV. Corticogeniculate axons were stimulated by amplitude graded voltage pulses of 0.2 ms duration (Granseth et al. 2002). For the present study, trains of 5–40 pulses at frequencies from 1 to 50Hz were used with the trains separated by resting periods of at least 60s. Unless specified, trains at 25Hz with 10 impulses at intensities recruiting multiple corticogeniculate fibres were used (Granseth & Lindström, 2003).

Data analysis

Records were filtered at 4.0kHz and sampled at 20kHz using the Pulse software (HEKA Elektronik) and stored to the hard disk of a computer. After additional digital Bessel filtering at 1.5kHz, EPSC amplitudes were measured from baseline to peak. When EPSCs overlapped temporally, amplitudes were measured from the EPSC peak to the extrapolated decay level of the preceding EPSC. Averaging of EPSCs was made using the PulseFit or IgorPro software (Wavemetrics, Lake Oswego, OR, USA). A measurement of facilitation was obtained by EPSC_n/EPSC₁. Functions were fitted to the data using a least-sum-of-squares method with Origin (MicroCal, Northampton, MA, USA) or IgorPro software. Fits with exponential functions of the type

were weighed in favour of EPSCs early in the trains by a factor from the reciprocal of the stimulus number. Hill functions

and linear regressions

were not weighed. Values are given as mean ± standard deviation unless otherwise stated. Data were statistically evaluated by Student's paired or unpaired t test; P < 0.05 was considered significant. Multiple comparisons were evaluated by analysis of variance (ANOVA).

Results

Corticogeniculate EPSC amplitudes increase with train stimulation

Unitary corticogeniculate EPSCs are small in amplitude with low resting transmitter release probability (p_syn) (∼5 pA at −70 mV holding potential; p_syn < 0.1; Granseth & Lindström, 2003). For the present analysis compound EPSCs, typically exceeding 50 pA in amplitude were obtained by stimulation of a large group of convergent corticogeniculate axons (Fig. 1A). In this way, the response variability was minimized by averaging the activity over many synapses, a procedure justified by the rather uniform properties of corticogeniculate synapses (Granseth et al. 2002). Recordings were obtained from a total of 45 dLGN cells in slices from 36 animals.

Figure 1. Increase in amplitude of corticogeniculate EPSCs with repetitive stimulation.

A, averaged EPSCs evoked by stimulation of corticogeniculate axons in the optic radiation by a train of 10 pulses (3 V) at 25Hz. Stimulation artefacts are truncated. B, development of facilitation (EPSC_n/EPSC₁) at 25Hz; mean ±s.e.m. for 30 cells. The data points could be empirically described by the sum of two exponential functions (f(t) = F_ss−K₁× exp(−t/τ₁) −K₂× exp(−t/τ₂)). The continuous line is a least-sum-of-squares fit providing time constants τ₁= 39 ms and τ₂= 450 ms. The dotted line is a linear regression for EPSC₃− EPSC₁₀ with r²= 0.98. C, EPSC₁ (thin black line), EPSC₃ (thick grey line) and EPSC₁₀ (dotted grey line) from record in A, scaled in amplitude and overlaid at the same time scale. Note that the waveforms are similar with a slightly increased duration of EPSC₁₀.

Brief trains of stimulation (10 pulses at 25Hz) evoked EPSCs that grew rapidly in amplitude for the first two or three stimuli with a more moderate growth rate for later EPSCs (Fig. 1B). The facilitation (EPSC_n/EPSC₁) of EPSC₂ was 3.7 ± 1.2, similar to the paired pulse facilitation previously described (Granseth et al. 2002), and increased to 5.1 ± 1.9 for EPSC₃ (30 cells). In all cells, such enhanced EPSC amplitudes were maintained for subsequent EPSCs without obvious signs of synaptic depression. Instead, the EPSCs continued to grow, reaching 6.3 ± 2.7 at EPSC₁₀. This multifold increase in synaptic strength occurred at the standard [Ca²⁺]_o of 2.0 mm (see also below). A comparably dramatic facilitation that can be supported for prolonged periods of firing has, to my knowledge, not been seen for any other mammalian synapse.

To further investigate for EPSC depression during repetitive synaptic activity, trains were prolonged to contain 40 impulses (Fig. 2). In all five cells the EPSC amplitudes were maintained without obvious signs of depression. Thus, at physiological firing frequency, facilitated transmitter release can be upheld well beyond normal train durations (Livingstone & Hubel, 1981; Martinez-Conde et al. 2002).

Figure 2. Prolonged train stimulation with different degrees of recruitment.

A, compound EPSCs evoked at 2 different intensities (2.5 V upper trace; 3.5 V lower trace) to recruit different numbers of corticogeniculate synapses at the same dLGN cell. Averages of 10 (upper trace) and 5 records (lower trace). Stimulation artefacts are truncated for clarity. B, facilitation of EPSCs over time for the trains in A; 2.5 V (□) and 3.5 V (▪); means ±s.e.m. Note the lack of depression and that the facilitation is similar despite a fourfold increase in EPSC amplitudes.

The EPSC facilitation could be empirically described by a double exponential function, f(t) = F_ss−K₁× exp(−t/τ₁) −K₂× exp(−t/τ₂), where F_ss is the theoretical steady-state EPSC ratio approached with time constants τ₁ and τ₂. A least-sum-of-squares fit to the pooled data of 30 cells gave F_ss= 7.0 with the time constants τ₁= 39 ms and τ₂= 450 ms for trains at 25Hz (Fig. 1B, continuous line). The moderate increase in amplitude of late EPSCs could also be approximated by a linear function (Fig. 1B, dotted line). The likely reason for this is that the fast component already reaches 95% of steady state within 120 ms, with growth thereafter defined primarily by the slow component. A linear regression would satisfactorily describe the latter when train duration is short relative to the slow time constant.

The sample traces in Fig. 1A and C are from one cell where EPSC₃ was 4.6 times the size of EPSC₁. In spite of this multifold increase in size there was no change in EPSC waveform between the first and third stimuli (Fig. 1C). For the last stimulus in the train, when the amplitude ratio had increased to 5.0, the EPSC duration was slightly longer (Fig. 1C, dotted line). However, such a change was not consistently encountered in all 30 cells investigated. The average EPSC₃ and EPSC₁₀ ratios were 5.1 ± 1.9 and 6.3 ± 2.7, while the EPSC duration at half-peak amplitude was not significantly different (EPSC₁= 8.5 ± 2.9 ms; EPSC₃= 9.0 ± 2.8 ms; EPSC₁₀= 9.3 ± 2.4 ms; P= 0.49, single-factor ANOVA). Thus, the facilitation was not paralleled by a corresponding change in EPSC duration.

The magnitude and time course of facilitation was the same for compound EPSCs of different amplitudes obtained by varying the stimulation intensity (4 cells; Fig. 2). This uniformity with different degrees of axon recruitment is in agreement with earlier findings for paired pulse facilitation and confirms that corticogeniculate synapses constitute a quite homogeneous population of synapses with regard to short-term synaptic plasticity (Granseth et al. 2002).

Metabotropic glutamate receptors

In addition to ionotropic glutamate receptors that mediate the fast EPSC, metabotropic glutamate receptors have been identified at the corticogeniculate synapse. The Group I metabotropic glutamate receptor mGlu1 causes a slow depolarizing response associated with an increase in membrane resistance of dLGN principal cells (McCormick & von Krosigk, 1992; Turner & Salt, 1998, 2000). However, GTP wash-out with whole cell recordings and the fact that EPSC peak currents were measured from the extrapolated decay of the preceding EPSC (see Methods) assure that metabotropic currents would not have much bearing on the above estimates of facilitation. As expected, EPSC growth was unaffected with 200μm (S)-MCPG to antagonize Group I metabotropic receptors (P= 0.43, two-factor ANOVA with replication, 3 cells).

Activation of Group III glutamate receptors has been shown to reduce the strength of corticogeniculate synapses, possibly acting directly on the presynaptic terminal (Turner & Salt, 1999). To explore if this receptor influences EPSC growth during trains, the Group III antagonist (RS)-CPPG was added to a final concentration of 100μm (4 cells). The facilitation remained unchanged (P= 0.59, two factor ANOVA with replication), arguing against any major role for metabotropic glutamate receptor activation in the present experiments.

EPSC growth with different train frequencies

The EPSC facilitation varied with stimulation frequency. Double exponential functions with the same time constants as earlier (τ₁= 39 ms and τ₂= 450 ms) could well describe the EPSC build-up in the frequency range 5–25Hz (pooled data from 10 cells; Fig. 3A). The steady-state facilitation (F_ss) inferred by the fitted functions increased linearly with train frequency (Fig. 3B, dotted line). The steady-state level would essentially be attained after 1–2s but corticogeniculate neurones in vivo are unlikely to maintain elevated firing this long. Animals make eye and head movements several times per second in order to prevent fading of stationary stimuli (Ditchburn & Ginsborg, 1952; Fuller, 1985; Chelazzi et al. 1989; Martinez-Conde et al. 2002). Hence, the EPSC amplitude attained at train durations as short as 0.2 and 0.4 s would be physiologically relevant. When the EPSC size ratios were plotted for EPSCs at or immediately preceding t= 0.2 s and 0.4 s they also varied linearly with train frequency. Slopes were similar (Δy_{t = 0.2s}= 0.23; Δy_{t = 0.4s}= 0.25; Δy_{t =∞}= 0.26) but amplitudes were generally lower than for F_ss (y_{0t = 0.2s}= 0.86; y_{0t = 0.4s}= 0.99; y_{0t =∞}= 1.7) (Fig. 3B). This simple relationship between facilitation, firing frequency and train duration might be useful in the construction of neuronal network models aimed at probing the function of the corticogeniculate feedback system.

Figure 3. Corticogeniculate EPSC facilitation with 5–25Hz trains.

A, facilitation of EPSCs over time in trains at 25 (▪), 16.7 (□), 10 (⊠) and 5Hz (). Lines are least-sum-of-squares fits of a double exponential function as in Fig. 1 with the same time constants τ₁ = 39 ms and τ₂ = 450 ms. B, diagram comparing EPSC amplitudes at different points in time (t) for 5–25Hz trains. Data points are EPSC amplitude at, or immediately preceding t = 0.2 or 0.4 s. Continuous lines are linear regressions with Δy_{t = 0.2s}= 0.23 and Δy_{t = 0.4s}= 0.25. The dotted line is the theoretical F_ss (×) regression line with Δy_{t =∞}= 0.26 at the theoretical point in time t=∞. Other markers are the same as in A. Data points in A and B are average values ±s.e.m. from the same 10 cells.

There was no further growth in EPSC amplitude when train frequency increased from 25 to 50Hz (P= 0.34, two-factor ANOVA with replication;Fig. 4B). As a result of this response saturation, the time constants used above were not apt for the EPSC growth at 50Hz. To test if AMPA receptor desensitization limits the EPSC growth at such high frequencies, the effect from either the desensitization blocker cyclothiazide (50–100μm; 5 cells) or partial receptor blockade with the fast antagonist kyunurenic acid (1.0mm; 4 cells) was investigated (Wong et al. 2003). Neither drug enhanced the facilitation during 50Hz trains (P= 0.55 and 0.22, respectively, two-factor ANOVA with replication), arguing against AMPA receptor desensitization being responsible for the EPSC saturation.

Figure 4. EPSCs growth with 1–50Hz trains.

A, growth in EPSC size compared to position in the train (n= sequential pulse number) for 25 (▪), 16.7 (□), 10 (⊠), 5 (), 2 () and 1Hz (). Lines are exponential functions with time constants (τ₁= 39 ms, τ₂= 0.45 s and τ₃= 4.8 s) transformed from the temporal domain to stimulus number (n). B, EPSC_n/EPSC₁ at 50 (□) and 25Hz (▪) over time. C, contribution of each exponential component in curve fits in A plotted against frequency (K₁, ⋄; K₂, ♦; K₃, ). Continuous lines are curve fits; for K₁ a linear regression between 2 and 25Hz (Δy_K1= 0.20 versus frequency), for K₂ a linear regression for data between 5 and 25Hz (Δy_K2= 0.07) and for K₃ an exponential decay. Dotted line is a regression line for K₂+K₃ (×; Δy_K2+K3= 0.06). Further details in the text. Average values ±s.e.m. from the same 10 cells as in Fig. 2.

For frequencies < 5Hz a component with a very slow time constant (τ₃= 4.8s) had to be included in the curve fits (Fig. 4A). This component, presumably being augmentation, could solely account for the data at 1Hz. When this time constant was included in triple exponential curve fits f(t) =F_ss−K₁× exp(−t/τ₁) −K₂× exp(−t/τ₂) −K₃× exp(−t/τ₃), it did not contribute to the response at train frequencies >10Hz.

The values of K for the different exponentials in the above curve fits are plotted against frequency in Fig. 4C. K₁ (fast facilitation) increased linearly with frequencies >2Hz (Δy_K1= 0.20), while the increase of K₂ (slow facilitation) was roughly linear at frequencies > 5Hz (Δy_K2= 0.07). K₃ (augmentation) contributed at the lowest frequencies but given short gaze holding times, it may be of little relevance in the present context. This component is further investigated and discussed in an accompanying paper (Granseth & Lindström, 2004).

The rate of EPSC growth was most pronounced for the first two or three stimuli at all train frequencies. This was particularly obvious when EPSC ratios were plotted against sequential stimulus number (Fig. 4A). The EPSC₃/EPSC₁ ratio varied from 78 to 83% (mean 80%) of that for EPSC₁₀/EPSC₁ when going from 1 to 25Hz. Most of the EPSC enhancement at the higher frequency came from the fast component of facilitation that did not contribute at all at the lowest frequency. It follows that the pronounced frequency modulation of corticogeniculate EPSCs is primarily due to fast facilitation. Its relative contribution compared to the sum of the two slower components was 3.3:1, as estimated from the Δy_K1: Δy_K2+K3 ratio (0.2/0.06; 77%; Fig. 4C).

EPSC growth at different extracellular calcium ion concentrations

Fast transmitter release is Ca²⁺-dependent and the same appears to be valid for the mechanism of facilitation (Katz & Miledi, 1968; Zucker & Regehr, 2002; Felmy et al. 2003; Blatow et al. 2003; Trommershäuser et al. 2003). When the Ca²⁺ inflow into the corticogeniculate terminal was altered by changing [Ca²⁺]_o, there was a dramatic effect on the EPSCs evoked by a 25Hz train. Sample records, at 1.0 and 4.0mm, well below and above the standard [Ca²⁺]_o of 2.0mm, are shown for a representative cell in Fig. 5A. Note that there was no indication of EPSC depression during the trains, not even at the highest Ca²⁺ level.

Figure 5. Effect of changes in [Ca²⁺]_o on amplitude and facilitation of corticogeniculate EPSCs.

A, averaged traces for 25Hz trains (2.5 V) at different [Ca²⁺]_o; 10 records in 1 cell. Stimulation artefacts are truncated. B, average EPSC amplitudes (±s.e.m.) plotted against [Ca²⁺]_o on a logarithmic scale. ○, EPSC₁; •, EPSC₂; ⊗, EPSC₇. Lines are least-sum-of-squares fits of Hill functions, f([Ca²⁺]_o) = EPSC_max×[Ca²⁺]_o^N/(K_½^N+[Ca²⁺]_o^N). Exponential (N) was 1.6 for EPSC₁, 2.6 for EPSC₂ and 2.7 for EPSC₇. Binding constants (K_½) were 8.3, 2.5 and 1.5, respectively. C, facilitation relative to EPSC₁ for EPSC₂ (▪) and EPSC₇ (squares with cross) versus[Ca²⁺]_o.

The [Ca²⁺]_o dependency of EPSC₁, EPSC₂ and EPSC₇, determined at a wider [Ca²⁺]_o range for the same cell (Fig. 5B), could be described by Hill functions, f([Ca²⁺]_o) = EPSC_max×[Ca²⁺]_o^N/(K_½^N+[Ca²⁺]_o^N). In the functions, amplitudes increased from 0 to EPSC_max, with [Ca²⁺]_o defined by the cooperativity constant (N) and binding constant (K_½). EPSC₁ was less affected by changes in [Ca²⁺]_o than subsequent EPSCs which resulted in a lower value for N (1.6) in the best fit Hill function (EPSC_max was constrained to the same value as for EPSC₇). The [Ca²⁺]_o relationship was steeper for EPSC₂ and EPSC₇, both with similar values for N (2.6 and 2.7, respectively). Furthermore, the apparent K_½ decreased with sequential stimulus number, shifting the curves leftwards in the diagram. Qualitatively similar results were seen in two more cells. These changes in the Hill functions are compatible with a mechanism where substantially more Ca²⁺ becomes available at the presynaptic Ca²⁺ sensor with repetitive activity.

The physiological [Ca²⁺]_o is at the lower part of the Hill functions, where EPSC amplitudes increase just about exponentially with [Ca²⁺]_o. Facilitation was as a result more prominent in this range of [Ca²⁺]_o (Fig. 5C). This effect was more expressed for EPSC₇ than for EPSC₂. It was also visible when facilitation for sequential EPSCs in 25Hz trains was compared at three different [Ca²⁺]_o (1.0, 2.0 and 3.0mm; pooled data from 6 cells; Fig. 6). These plots clearly demonstrate that the temporal pattern of EPSC growth changed with [Ca²⁺]_o, in parallel to the absolute amplitudes of evoked EPSCs (Fig. 6A). EPSCs grew more gradually at low [Ca²⁺]_o compatible with a lengthened time constant for the fast facilitation (Fig. 6A, inset). The opposite was true for high [Ca²⁺]_o. This change in the fast time constant corresponds to the increase in N between Hill functions for EPSC₁ and EPSC₂ (Fig. 5B). Moreover, in the double exponential functions in Fig. 6B, the relative contribution by each component of facilitation (K₁ and K₂) changes in favour of the fast facilitation with increasing [Ca²⁺]_o.

Figure 6. Train stimulation at 1.0, 2.0 and 3.0mm[Ca²⁺]_o.

A, EPSC amplitudes at [Ca²⁺]_o 1.0 (⊗), 2.0 (•) and 3.0mm (○). Normalized to EPSC₁ at 2.0mm. Lines are double exponential curve fits with τ₂= 450 at all [Ca²⁺]_o, while τ₁ was more rapid with increasing [Ca²⁺]_o as shown in inset. B, facilitation of EPSCs at the different [Ca²⁺]_o in A. Lines are double exponential functions as in A. Train stimulation at 25Hz. All points are averages ±s.e.m. from 6 cells.

Reducing [Ca²⁺]_o reveals facilitation at many synapses that normally display depression for repetitive stimulation (Katz & Miledi, 1968; Paulsen & Heggelund, 1994; von Gersdorff & Borst, 2001). One scenario is that the facilitation, present at these synapses at normal [Ca²⁺]_o, is masked by depression from depletion of releasable synaptic vesicles. If transmitter release is reduced by lowering [Ca²⁺]_o, this facilitation would be unveiled. The corticogeniculate synapse displays pronounced facilitation at normal [Ca²⁺]_o. Nevertheless the EPSC enhancement was increased with a reduction in [Ca²⁺]_o (Fig. 6B). At 1.0mm EPSC₁₀/EPSC₁ was 220% of the corresponding ratio at 2.0mm. If depletion does limit facilitation at the corticogeniculate synapse during normal [Ca²⁺]_o, a critical EPSC size should exist, after which EPSC growth reverses to depression (Dittman et al. 2000). As can be seen in Fig. 4, this was not found for the frequencies tested in the present study. Moreover, there was no depression when transmitter release was increased by raising [Ca²⁺]_o (Fig. 5). In the absence of depression the conclusion must be that the Ca²⁺ dependency is inherent in the mechanism of facilitation itself.

Discussion

Brief trains of stimulation of corticogeniculate synapses evoked EPSCs that increased severalfold in amplitude with the first two or three impulses followed by a more gradual growth for later EPSCs. The facilitation never reversed to depression, not even with long train duration, high frequency or at high [Ca²⁺]_o. The EPSC build-up for trains at 5–25Hz could be described with an exponential function with two time constants of ∼40 and 450ms. The growth of compound EPSCs in the present study was similar to the facilitation previously seen for unitary EPSCs of corticogeniculate fibres (Granseth & Lindström, 2003). At lower frequencies, a third component with time constant of ∼4.8 s was apparent, presumably representing augmentation (Granseth & Lindström, 2004).

Short-term plasticity during trains

It is appealing to ascribe the two time constants that describe EPSC growth during trains to the two components that can be seen in the decay of paired pulse facilitation (Magleby, 1987; Zucker & Regehr, 2002; Granseth et al. 2002). For the simplest model containing two elements of facilitation with immediate onset and separate monoexponential decays, a linear superposition during trains would yield the same time constants for the build-up as for the decay (Helmchen et al. 1996). Such an assumption would be appropriate for facilitation that has been shown to enhance transmitter release independently of the history of preceding release (Katz & Miledi, 1968; Granseth & Lindström, 2003). It would not hold for EPSC depression, since it is a release-dependent process, whether it arises from depletion of transmitter vesicles, receptor desensitization or both (Dittman et al. 2000; Wong et al. 2003).

The build-up time constants for EPSC facilitation during trains were indeed similar, although consistently ∼3 times slower than the decay time constants for paired pulses previously identified at the corticogeniculate synapse (Granseth et al. 2002). It is unlikely that this disparity originates from a release-dependent activation of metabotropic glutamate receptors since facilitation of corticogeniculate EPSCs was the same in the presence of receptor antagonists. A more likely explanation would be that one or both components of facilitation do not scale linearly during trains. Since facilitation is a Ca²⁺-dependent process, intracellular Ca²⁺ buffer binding and diffusion rates could produce such non-linearity. It is, however, noteworthy that the relation between fast and slow time constants was consistent, τ₂ was ∼15 times longer than τ₁ in both growth and decay. Considering the limitations of exponential curve fitting, the moderate disparity between build-up and decay should not be over-emphasized and the two time constants most likely reflect fast and slow components of facilitation. The third time constant (τ₃), identified at trains with low frequency, is likely to result from augmentation since it develops and decays in the range of seconds (Magleby, 1987; Zucker & Regehr, 2002; Granseth & Lindström, 2004).

Facilitation increases the synaptic release probability (Katz & Miledi, 1968; Hanse & Gustafsson, 2001; Granseth & Lindström, 2003; Felmy et al. 2003). At the corticogeniculate synapse, transmitter release is raised >6 times during trains but the store of synaptic vesicles seems not to be depleted with ensuing EPSC depression. Corticogeniculate terminals are small (cross-sectional area ∼0.5μm²) but they contain large numbers of synaptic vesicles (Erisir et al. 1997). If vesicles located at some distance from the active zone belong to a reserve pool, the competence for maintaining transmitter release during prolonged periods in time should be substantial. However, the rate of mobilization of such reserve vesicles also has to be quite rapid to compensate for the increased transmitter release at high frequencies. The resistance to depletion combined with a low basal synaptic release probability (p_syn < 0.1; Granseth & Lindström, 2003) implies that transmitter release at the corticogeniculate synapse is primarily mediated by synaptic vesicles with properties similar to the slowly releasable pool of vesicles identified at the calyx of Held giant synapse (Sakaba & Neher, 2001a, b).

In a recent model it has been suggested that slowly (or reluctantly) releasable vesicles are positioned at some distance from the release-triggering Ca²⁺ channels at the active zone (Trommershäuser et al. 2003). The unfavourable position with respect to the Ca²⁺ influx leads to a low vesicular release probability that is susceptible to modulation by intraterminal Ca²⁺ buffers (Meinrenken et al. 2003). The low cooperativity constant (N) in the Hill function for EPSC₁ at the corticogeniculate synapse supports this notion. The trigger for transmitter release is usually considered to bind four or five Ca²⁺ with positive cooperativity producing an N with the same value (Bollmann et al. 2000; Schneggenburger & Neher, 2000). If the conditions for high cooperativity cannot be fulfilled, such as when [Ca²⁺] is minute at the calcium sensor and the binding of Ca²⁺ significantly reduces the concentration of free Ca²⁺, the constant will be lower (Weiss, 1997). Hence, the particularly low N for EPSC₁ could be due to a very small amount of Ca²⁺ reaching the sensor at the first impulse, consistent with the modest basal release probability.

The fast component of facilitation accounts for most of the modulation of EPSC amplitude in trains of physiological duration. Steady state is essentially reached within 0.2 s of neuronal firing and the contribution by fast facilitation increases linearly with firing frequency (Fig. 4C). This component resembles the ‘pseudofacilitation’ that can be observed after loading synapses with a low amount of a high affinity calcium buffer (Rozov et al. 2001; Blatow et al. 2003). The efficient binding of free Ca²⁺ by the buffer inhibits transmitter release for the first presynaptic spike but as the free buffer is consumed the buffer capacity starts to saturate. The Ca²⁺ influx for the next spike would then encounter a partially saturated buffer with more Ca²⁺ reaching the release trigger. Recent evidence suggests that this kind of facilitation is actually present under normal conditions in neocortical and hippocampal terminals containing the fast Ca²⁺ buffer calbindin-D28k (Blatow et al. 2003). Since the time constant of fast facilitation at the corticogeniculate synapse was Ca²⁺ dependent, such an endogenous buffer mechanism might also operate at the present synapse (although the molecular nature of the buffer remains unidentified). A larger amount of Ca²⁺ entering the presynaptic terminal with an increased [Ca²⁺]_o would saturate an intra-terminal buffer with fewer spikes, resulting in a more rapid time constant describing the build-up of facilitation. At sufficiently low [Ca²⁺]_o the endogenous buffer would not saturate, and the fast time constant would merge with the slow component of facilitation. The reduced paired pulse facilitation at 0.5mm[Ca²⁺]_o encountered in a previous study at the corticogeniculate synapse is in line with this prediction (Granseth et al. 2002; see also Fig. 5C). Further support for an increase in the Ca²⁺ reaching the release trigger comes from the increase in the cooperativity constant (N) in Hill functions for EPSCs subsequent to EPSC₁. However, a facilitation mechanism where Ca²⁺ remains bound to a high affinity binding site at the secretory trigger could also be surmised (Bertram et al. 1996).

The slow component of facilitation dominates the growth in amplitude after the fast component has reached steady state. Given the short duration of fixation of the retinal image, the slow facilitation has little time to develop and would therefore add only marginally to the attained EPSC amplitude during viewing. It is well established that residual Ca²⁺ accumulating in the presynaptic terminal with repetitive spikes contributes to synaptic facilitation (Katz & Miledi, 1968; Atluri & Regehr, 1996; Zucker & Regehr, 2002; Felmy et al. 2003). At the corticogeniculate synapse the slow component of facilitation becomes less prominent with higher [Ca²⁺]_o (Fig. 6), in agreement with a direct summation of residual [Ca²⁺]_i at the secretory trigger. If the Ca²⁺ that remains in the presynaptic terminal adds to the Ca²⁺ influx of subsequent spikes, the effect on transmitter release will be greatest at the exponentially growing part of the Hill functions, that is at low [Ca²⁺]_o (Rozov et al. 2001; Felmy et al. 2003). The subsistence of [Ca²⁺]_i at the corticogeniculate synapse has, however, not been confirmed with fluorescent Ca²⁺ dyes or probed with intracellular Ca²⁺ chelators, as is the case at other synapses (Atluri & Regehr, 1996; Rozov et al. 2001; Felmy et al. 2003; Jackson & Redman, 2003). Such experiments, although elaborate because of the small size of this synapse, are desirable for elucidating the mechanism of corticogeniculate facilitation, not only since it seems unrivalled in magnitude in the central nervous system, but also because it could define the function of the corticothalamic feedback pathway.

Facilitation defines the gain of the neuronal amplifier

The observed EPSC facilitation at different firing frequencies was comparable to findings in the cat in vivo (Lindström & Wróbel, 1990). Since EPSPs evoked in the latter study superimposed on a steady depolarizing shift, the steady-state enhancement was underestimated for high stimulation frequencies. Therefore, trains > 20Hz were excluded when comparing the steady-state facilitation. The linear firing frequency relationship Δy= 0.23in vivo is then in good agreement with the value obtained in the present study (∼0.25). This confirms that the marked short-term plasticity is also present under physiological conditions and almost certainly affects the function of the visual system in living animals.

Corticogeniculate feedback has been suggested to function as a neuronal amplifier for the thalamic relay of visual input to the cortex (Ahlsén et al. 1985; Ferster & Lindström, 1985b; Lindström & Wróbel, 1990; Granseth et al. 2002; Granseth & Lindström, 2003). The visually driven feedback excitation would add to the visual input from the retina at the dLGN principal cells and increase the transfer to the primary visual cortex (Fig.7A). As illustrated by the black line in Fig.7B, such mechanism would increase the slope of the theoretical firing rate–input current relationship, in other words, increase the dynamic range of neuronal firing. The increase mediated by the neuronal amplifier is different from a general thalamic depolarization resulting from modulatory input from the brainstem (Ahlsén et al. 1984; McCormick & Bal, 1997; Fjeld et al. 2002), as the latter would have no effect on the dynamic range. A general increase in excitability of the principal cells would rather shift the cell firing to higher frequencies, as illustrated by the grey line in Fig. 7B.

Figure 7. The corticogeniculate neuronal amplifier increases the dynamic range of dLGN cell firing in response to visual stimuli.

A, schematic illustration of excitatory input to principal cells in the dorsal lateral geniculate nucleus (dLGN). Glutamatergic synapses carry information from the retina with axons in the optic tract (OT) and feedback from cortex with axons in the optic radiation (OR). Modulatory input (M, shown in grey) may be from the brainstem. B, model input–output relationships for a principal cell for different mechanisms that would increase the relay to cortex. Dashed line illustrates basal conditions in the spike-firing mode where low threshold calcium currents are inactive. A further depolarization mediated by a modulatory input would increase the output frequency without a change in the dynamic range (grey continuous line). The corticogeniculate neuronal amplifier would, when activated, increase the frequency span of the output by adding progressively more feedback excitation per impulse at higher frequencies (black continuous line).

The specificity of the geniculo-corticogeniculate system, in combination with a low resting release probability that increases with facilitation, ensures that only visual stimuli that effectively recruit the corticogeniculate neurones in the first place are amplified (Granseth & Lindström, 2003). Such input specificity limits the gain and prevents the amplifier from escaping equilibrium and diverging into saturation (Douglas et al. 1995; Wróbel et al. 1998). The level of feedback excitation reached with short trains of spikes from a visual stimulus would define the amplifier gain. Several forms of short-term plasticity might affect the feedback, including the two components of facilitation and augmentation. Interestingly, the different components of short-term plasticity interact to increase the steady-state excitatory current on a linear scale for physiologically relevant firing frequencies and gaze fixation intervals. This relationship would increase the dynamic range of firing versus input current for principal cells, without distorting the basic characteristics of the thalamic relay (Fig. 7B) and thus make the corticogeniculate system ideally suited for its proposed function as a neuronal amplifier.