| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
University Laboratory of Physiology, Oxford, United Kingdom
Submitted 29 April 2004; accepted in final form 11 December 2004
 |
ABSTRACT |
|---|
|
|
|
INTRODUCTION |
|---|
|
; Sherman and Guillery 1996, 2002; Usrey et al. 1999). However, retinal afferents do not represent the large majority of inputs to the dLGN. Dominating the dLGN's afferent population are the 85–90% of synapses that provide "modulating" influences on geniculate processing (e.g., Sherman and Guillery 1996, 2002). Although these modulatory inputs cannot change fundamental receptive field (RF) properties established by retinal drivers, they can produce subtle alterations in the response properties of dLGN neurons, such increasing or decreasing the magnitude of visually evoked responses (e.g., Uhlrich et al. 1995, 2002). Thus due to modulatory inputs onto dLGN cells, the thalamic relay of visual information from retina to cortex becomes a dynamic operation.
One important way in which modulatory inputs to the dLGN can alter the retinocortical transfer of visual information is by changing the firing mode of geniculate relay cells. These neurons, like all thalamic relay cells, can fire action potentials in two different modes: "burst" and "tonic." As demonstrated by in vitro recordings (e.g., Jahnsen and Llinas 1984; Kim et al. 2001; Ramcharan et al. 2000a; Sherman 2001; Zhan et al. 1999), firing mode is controlled by a T-type calcium channel that produces IT, a voltage- and time-dependent current. IT is inactivated at relatively depolarized membrane potentials (>60 mV). In this state, depolarizing inputs stimulating a geniculate relay cell will produce conventional, tonic action potentials. If the cell's membrane potential is relatively hyperpolarized for >100 ms, however, IT is de-inactivated. Cell stimulation then produces a depolarizing waveform carried predominantly by calcium ions, known as a low-threshold spike (LTS). This LTS is usually accompanied by a burst of sodium-based action potential firing. Modulatory geniculate inputs that produce (long-lasting) depolarizations or hyperpolarizations of cell membrane potentials can therefore switch relay cell responses from burst to tonic firing, and vice versa.
One of the most important situations in which such burst-tonic switches occur is the transition between sleep and wakefulness. Thalamocortical relay neurons in naturally sleeping animals display rhythmic bouts of bursting ( McCarley et al. 1983; Weyand et al. 2001), which in vitro studies have established are due to cyclic interactions between IT and a depolarizing, hyperpolarization-activated current Ih ( McCormick and Bal 1997). With the transition from slow-wave sleep to wakefulness, modulatory inputs from the brain stem produce a progressive depolarization of thalamocortical relay cells. This inactivates IT, decreasing burst firing and promoting tonic responses ( McCarley et al. 1983; McCormick and Bal 1997).
Because rhythmic burst firing is most prominent during slow-wave sleep ( McCarley et al. 1983; Weyand et al. 2001) and because during slow-wave sleep the responsiveness of dLGN neurons to RF stimulation is decreased ( Livingstone and Hubel 1981), it might seem that burst firing in thalamocortical relay cells is unimportant as a carrier of sensory information. However, studies in the past decade have begun to uncover the possibility that both burst and tonic firing modes represent specific signals in sensory processing. Lu et al. (1992) used intracellular recordings in vivo to demonstrate that burst responses occur in the cat dLGN during periods of visual stimulation. These recordings also showed that burst spikes could be accurately identified purely on the basis of temporal patterns in spike firing ( Lu et al. 1992), allowing less techni-cally demanding extracellular recordings to confirm that bursting, like tonic firing, occurs during sensory stimulation in anesthetized (e.g., Guido et al. 1992) and awake behaving (e.g., Fanselow et al. 2001; Guido and Weyand 1995; Ramcharan et al. 2000b; Swadlow and Gusev 2001; Weyand et al. 2001) animals. Furthermore, while burst and tonic dLGN spikes carry similar amounts of visual information ( Reinagel et al. 1999), recordings in the cat dLGN have shown that the type of visual information transmitted by the two firing modes can be significantly different. Among other differences, burst spikes in the cat follow temporal changes in a sinusoidal stimulus less faithfully ( Guido et al. 1992; Lu et al. 1992), offer better stimulus detection ( Guido et al. 1995), are more tightly tuned for TF ( Mukherjee and Kaplan 1995), and have more reliable timing ( Guido and Sherman 1998) than their tonic counterparts. Changes in firing mode in thalamic relay cells might therefore affect the information relayed by those cells to cortex.
Although studies in the cat have identified several differences in the stimulus coding properties of burst and tonic spikes in the dLGN ( Guido and Sherman 1998; Guido et al. 1992, 1995; Lu et al. 1992; Mukherjee and Kaplan 1995) (see DISCUSSION), the comprehensive set of data gathered in our laboratory concerning the response properties of mouse dLGN neurons ( Grubb and Thompson 2003) offers an opportunity to compare burst and tonic firing across a large range of visual response characteristics. As well as providing a full and detailed description of the functional differences between these two firing modes, such a comparison will also offer an indication as to whether such differences are conserved across mammalian species. We know that the properties of the thalamic IT current and its accompanying burst activity are similar across different mammals (guinea pig: Jahnsen and Llinas 1984; McCormick and Feeser 1990; cat: Zhan et al. 1999; monkey: Ramcharan et al. 2000a; mouse: Kim et al. 2001), but it is unknown whether the distinct visual response characteristics of burst and tonic firing ( Guido and Sherman 1998; Guido et al. 1992, 1995; Lu et al. 1992; Mukherjee and Kaplan 1995) are a common feature of mammalian visual processing at thalamic level. The mouse dLGN data presented here represent a first step toward addressing this issue.
Perhaps more importantly, though, studying the visual coding properties of burst and tonic firing in the mouse thalamus opens up the study of firing mode function to genetic manipulation techniques. In mammals, these techniques are most advanced in mice, where they can be invaluable tools for investigating neuronal function at the molecular level. In particular, because the switch between burst and tonic thalamic firing modes can be determined by a single ion channel, investigations of firing mode function might especially benefit from approaches involving genetic manipulation. Mice that lack the T-type calcium channel underlying IT have been generated ( Kim et al. 2001), but before the consequences of this mutation on visual processing at the thalamic or cortical level can be investigated, we need to know the probable contribution of this channel in the normal visual system. We can start to address this issue by investigating the basic visual response properties of burst and tonic firing in the wild-type mouse dLGN. In addition, genetic manipulations might be extremely useful in dissecting the molecular components underlying the control of response mode in thalamic relay cells. Various modulatory systems have been proposed to switch thalamic relay cells from burst to tonic mode or vice versa (e.g., Godwin et al. 1996; Lu et al. 1993; Uhlrich et al. 2002). Manipulating the receptors or other molecular machinery involved in this modulation could be extremely informative concerning the circuitry that controls or does not control thalamocortical information transmission.
As an essential precursor to such studies, we demonstrate here that burst-like events can be identified in extracellularly recorded responses of mouse dLGN cells and that these events have some properties expected of thalamic burst firing. We go on to show that burst firing in the mouse dLGN occurs during visual stimulation. Finally, we investigate the visual response characteristics of burst and tonic spikes in the mouse dLGN across a broad range of stimulus parameters.
|
|
METHODS |
|---|
|
Animals were adult (>3 mo of age) pigmented C57Bl/6J mice of either sex weighing 20–36 g (Harlan Olac, Oxon, UK). Mice were housed under a 12:12-h cycle of light and dark. All procedures were conducted under the auspices of United Kingdom Home Office project and personal licenses held by the authors.
Electrophysiology
We carried out electrophysiological recordings as described previously ( Grubb and Thompson 2003; Grubb et al. 2003). Mice were first anesthetized with an intraperitoneal injection of 25% fentanyl and fluanisone (Hypnorm, Janssen Animal Health, Bucks, UK) and 25% midazolam, Down, UK), administered at 2.7 µl/g. Surgical anesthesia was maintained throughout the setup procedures by subsequent applications of this induction dose. To prevent the accumulation of bronchial secretions, 5 µl atropine sulfate (600 µg/ml, Animalcare, Yorks, UK) was given subcutaneously. A tracheotomy was then performed ( Schwarte et al. 2000), and a small plastic tube (1 mm OD, 0.5 mm ID, SIMS Portex, Kent, UK) inserted and secured by ligation around the trachea. Application of a spot of cyanoacrylate glue added extra stability by attaching the tube to the skin of the upper thorax.
The animal was situated in a custom-built head holder and secured using a bite bar and ear bars before the tracheal tube was connected to a respiratory pump (MiniVent 845, Hugo Sachs Elektronik, March-Hugstetten, Germany). Pump rate (usually 120–140 strokes/min) and stroke volume (usually 150–200 µl) were adjusted to maintain the expired carbon dioxide level, monitored with a Capstar-100 CO2 analyzer (CWE, Ardmore, PA), at 2.5–4%. Once on the pump, the mice breathed a 1:3 mixture of oxygen and nitrous oxide, along with 1–1.5% halothane for maintained anesthesia. The mice were not paralyzed at any stage, allowing continued assessment of anesthetic depth via the paw pinch response. Throughout the experiment, heart rate was monitored via an ECG (
5 Hz was normal). Core body temperature, measured with a rectal probe, was maintained at 37°C by combining a high ambient temperature with heat from a thermostatically coupled blanket (NP 50-7061-R, Harvard Apparatus, Kent, UK).
A circular craniotomy 0.5 mm in diameter was made unilaterally above the presumptive location of the dLGN, 2.5 mm posterior and 2 mm lateral of the Bregma suture ( Paxinos and Franklin 2001). A small durotomy was made in the center of this exposed zone and, after any whiskers occluding the eyes had been trimmed, a tungsten-in-glass recording electrode with a 5- to 10-µm exposed tip (Alan Ainsworth, Northants, UK) was lowered vertically into the brain using a microdrive. Changes in background activity while advancing through the brain were used to locate the electrode relative to the dLGN ( Grubb and Thompson 2003). Once in the dLGN, the electrode was advanced or retracted in 5-µm steps. Signals were amplified locally by a FET headstage, further amplified by a Neurolog preamp (NL104; typically 10–20 k), then filtered (Neurolog NL125, low-frequency 150 Hz, high-frequency 3,000 Hz) for display on an oscilloscope. A Schmidt trigger was used to isolate responses of single neurons by adjusting the signal threshold; the trigger was also connected to a storage oscilloscope that enabled us to monitor spike shapes and to ensure that no spikes were seen in the refractory period. To isolate the responses of single units, electrode position was adjusted until spikes were deemed to have the same shape, amplitude, and rise time. Responses of optic tract fibers—typically large and fast, with high levels of spontaneous firing and a lack of burst activity—were easily distinguished from cell soma responses. Spike events were then time-stamped with a resolution of 100 µm and were linked to the timing of visual stimulus changes using in-house software written by D. Smyth.
Although we did not apply contact lenses, optical quality in our setup was good and did not degrade over the course of experiments. As described previously ( Grubb and Thompson 2003), no significant correlations were found between time from experiment start (
13 h) and any measures of spatial tuning in dLGN neurons. Furthermore, within animals, there were no significant differences in these measures between cells recorded in the first versus the second half of experiments (paired t-test or Wilcoxon paired test, depending on normality, P > 0.05). Eye movements under halothane anesthesia were rare and small ( Grubb and Thompson 2003; cf. Dräger 1975; Gordon and Stryker 1996; Wagor et al. 1980). As described previously ( Grubb and Thompson 2003), remapping of RFs 0.5–1 h after their initial localization revealed a mean shift (±SE) in position of only 3.6 ± 0.5° (n = 11). In addition, over a time scale of 10 min, response features such as phase-locked raster plots (Fig. 5), sinusoidal dependence of cell responses on stimulus phase (Fig. 6), and regular shifts in response phase with increasing temporal frequency (Fig. 5), all suggest that RF position changed very little. Moreover, in a study where properties of burst and tonic firing were always compared within individual tuning curve experiments, small eye movements were unlikely to affect the present findings.
|
|
Visually responsive neurons were first identified using a hand-held ophthalmoscope. Monochrome visual stimuli were then presented on a cathode ray tube monitor under the control of a Visual Stimulus Generator 2/4 graphics card (Cambridge Research Systems, Cambridge, UK); this had a pseudo-15-bit analog output allowing precise control of the luminance of each pixel on the display and correction for expansive luminance nonlinearities. Stimuli could be presented with 256 linearly spaced gray levels. The display comprised 800 x 600 pixels and at viewing distances between 135 and 400 mm subtended 53–111 x 41–95°. With maximum and minimum luminances of 99 and 2cd/m2, respectively, the display was the main source of illumination in an otherwise dim room. The frame rate of the monitor was 160 Hz.
The display was initially placed with a given cell's presumptive RF at its center; mapping with reverse correlation ( Grubb and Thompson 2003) then confirmed this location, or directed movement of the monitor in order that the display covered the entire RF area. Only cells the RFs of which could be localized in this way were subjected to subsequent analysis of response characteristics. Stimuli were presented with both eyes open, but responses in all cells were most likely driven entirely by the contralateral eye: not only are all wild-type mouse dLGN cells monocularly driven ( Grubb et al. 2003), but in these investigations, the display was often placed in positions that could be viewed only by the contralateral eye. Furthermore, although it is still possible that interactions between the two eyes could modulate the responses of single mouse dLGN neurons, such interactions are unlikely to affect present analyses in which burst and tonic response properties were compared within cells.
For quantitative investigation of the visual response properties of burst and tonic firing, sinusoidal monochromatic vertical gratings were presented covering the full extent of the display. These gratings were drifting in all experiments except the null test, in which they were stationary and sinusoidally contrast-modulated. In all experiments, all grating parameters were kept constant except the one under study; gratings varying in this parameter were presented in a pseudorandom sequence. Grating parameters in spatial frequency (SF) and temporal frequency (TF) tuning assessments were varied systematically and logarithmically to produce 8 stimuli ranging between 0.01 and 1 c/° inclusive for SF assessments, and between 0.5 and 16 Hz inclusive for TF assessments. In contrast and null phase test assessments, grating parameters were varied systematically and linearly to produce 12 stimuli ranging between 1 and 100% inclusive for contrast assessments and between –180 and +150° inclusive for null phase tests. Every stimulus repeat also contained a blank stimulus at mean luminance from which spontaneous activity levels were calculated. Each presentation of a grating stimulus in a given repeat occurred for seven cycles; of these only responses from the last five cycles were used to discard any flash responses to the initial presentation of the stimulus. The exception was assessing TF tuning. In these experiments, stimuli 1 Hz were presented for seven cycles (with the 1st 2 discarded) as usual. However, to ensure capture of enough responses to high TF stimuli, stimuli >1 Hz were presented for seven seconds, with the first 2 s being discarded to allow for flash responses. At least two repeats of each 7-cycle (or 7 s) stimulus were presented to every cell, although in the majority of cases four repeats were utilized.
Histology
Each experiment was ended with an intraperitoneal injection of 0.3 ml pentobarbitone sodium (Sagatal, Rhône Mérieux, Essex, UK). The mouse was then transcardially perfused with phosphate-buffered saline (Sigma, Dorset, UK) followed by 4% paraformaldehyde (TAAB Laboratories, Berks, UK) in 0.1 M phosphate buffer. The brain was removed, sunk in 30% sucrose for cryoprotection, and sectioned coronally at 30–50 µm on a freezing microtome. Sections were then mounted from distilled water onto gelatinized slides. Subsequent staining for cell bodies with cresyl violet allowed the identification of small (6 µA for 6 s) electrolytic lesions made at precise locations on successful penetrations. Locating these lesions made it possible to reconstruct electrode tracks and localize recorded units. Only cells unequivocally located within the dLGN were included in subsequent physiological analyses.
Data analysis
IDENTIFYING BURST AND TONIC RESPONSE COMPONENTS.
Spikes forming part of bursts triggered by low-threshold calcium spikes (LTS bursts) were identified from temporal patterns in the firing of dLGN cells. From intracellular recordings in the cat dLGN, Lu et al. (1992) developed criteria that could reliably identify bursts driven by LTS activity; we applied the same criteria here. Specifically, the first action potential in a burst had a preceding silent period of 
100 ms and a following interspike interval (ISI) of 4 ms. Any subsequent spikes with preceding ISIs of 4 ms were designated as additional action potentials in a burst. All spikes that did not form part of LTS bursts were designated as part of tonic firing.
RHYTHMICITY.
Rhythmicity in spontaneous bursting was assessed after Weyand et al. (2001). For each 5-s presentation of a 51 cd/m2 blank screen during which at least four bursts were fired, we calculated the SD of interburst intervals, the latter being defined as the time between the first spikes of consecutive bursts. This SD is a direct measure of burst rhythmicity—perfect rhythmic bursting should produce a SD of 0. To assess the statistical significance of each SD measure, we employed a Monte Carlo approach ( Weyand et al. 2001). For each blank stimulus presentation, 999 simulations were run in which an identical number of bursts were randomly distributed within the 5-s interval. Standard deviations of interburst intervals were calculated for each simulation and were ranked alongside the original, measured value. The rank of the value of the observed SD then gave a statistical measure of burst rhythmicity: significant rhythmicity (P < 0.05) was detected if the observed value fell within the 50 lowest simulated values.
FOURIER ANALYSIS.
Fourier analysis was applied separately to a given cell's tonic and burst spikes. Burst or tonic responses to a given grating stimulus were first binned according to their poststimulus time. One hundred twenty eight bins were used per stimulus cycle, producing a poststimulus time histogram (PSTH) for each stimulus. A fast Fourier transform of these data then converted these PSTH data into amplitudes (in spikes/s) and phases (in degrees or radians) of harmonics of the stimulus' modulation frequency ( Hochstein and Shapley 1976).
RESPONSE LINEARITY.
Linearity of spatial summation was assessed using a linearity measure first applied by Hochstein and Shapley (1976). Over the many stimulus phases presented in a single modified null test, the mean second harmonic (F2) response amplitude was divided by the maximum fundamental harmonic (F1) response amplitude. Linearity values <1 are indicative of linear spatial summation, while linearity values greater than one are indicative of nonlinear spatial summation.
Another measure of response linearity, "rectification," was calculated over a given tuning curve as follows
 | (1) |
; Lu et al. 1992).
CURVE FITTING.
All curve fits were carried out using a least-squares minimalization algorithm ( Grubb and Thompson 2003).
Plots of stimulus SF versus cell F1 response amplitude were fitted with a difference of Gaussians curve ( Grubb and Thompson 2003; Rodieck 1965; So and Shapley 1981)
 | (2) |
is SF, b is baseline response, kc is the area under the RF center's Gaussian function, ks is the relative area under the RF surround's Gaussian function, and rc and rs are the radii of the center and surround Gaussian functions, respectively, at the point each mechanism has reached 1/e of its peak.
Plots of stimulus TF versus cell F1 response amplitude were fitted with an atheoretical function comprising two half-Gaussians ( Grubb and Thompson 2003)
 | (3a) |
 | (3b) |
is TF, p is peak TF, a is the response amplitude at optimum TF, s is the Gaussian spread, b1 is the baseline on the low-frequency side of the curve, and b2 is the baseline on the high-frequency side of the curve. Plots of stimulus TF versus cell F1 response phase were simply fitted with a straight line.
Plots of stimulus Michelson contrast versus cell F1 response amplitude were fitted using a hyperbolic function ( Albrecht and Hamilton 1982; Grubb and Thompson 2003) of the form
 | (4) |
SIGNAL DETECTION ANALYSIS.
Receiver operating characteristic (ROC) curves were derived from probability distributions of spike counts obtained during a sampling episode 
of visually driven (s) and spontaneous (n) activity ( Guido et al. 1995). s comprised half the temporal period of the stimulus and was centered on the phase of the F1 response component. n comprised an equal period during the presentation of a blank screen (51 cd/m2). For all spikes fired to a given stimulus, a criterion domain was set corresponding to the full range of spike counts (including 0) occurring for s and n. ROC curves were then obtained by plotting, for each criterion level, the probability P(false alarm) of obtaining a criterion response during n versus the probability P(hit) of obtaining a response during s. To produce a nonparametric measure of discrimination performance ( Green and Swets 1966; Macmillan and Creelman 1991), the area underneath each ROC curve was then calculated using simple geometry.
MEASURES OF SPIKE TIMING RELIABILITY.
Spike timing reliability was assessed using two measures. The first, "reliability," was applied to all burst or tonic spikes fired to a given stimulus, and represented the mean deviation of spike times from their F1 phase, normalized by the stimulus period
 | (5) |
First-spike reliability was applied only to the first burst or tonic spikes fired during a particular stimulus cycle and was calculated as the SD of first-spike latencies. Latencies were taken from the start of a half-cycle period centered on the F1 component of either burst or tonic firing, as appropriate
 | (6) |
10. Statistical analysis
Sample sizes in this study were large enough (n > 30) to reliably assess their Normality using a Kolmogorov-Smirnov test. Those deemed Normal were described using the means ± SE and were compared using parametric tests. Those deemed non-normal were described using the median and were compared using nonparametric tests. Paired tests were used to compare burst and tonic spikes fired by the same cell to the same stimuli. All comparison tests were two-tailed, with the level of significance set at 0.05.
|
|
RESULTS |
|---|
|
Visual response properties of burst and tonic firing were assessed in 102 mouse dLGN cells, all with classic geniculate ON- or OFF-center receptive fields (see Grubb and Thompson 2003). Responses were recorded from cells with RFs at visual eccentricities ranging from 0 to 92° (means ± SE: 44 ± 3°). Because the response properties of mouse dLGN cells change very little with increasing visual eccentricity ( Grubb and Thompson 2003), we did not limit our analysis to cells with RFs in particular regions of visual space.
Burst firing in the responses of cells to sinusoidal grating stimuli was identified on the basis of temporal patterns of spikes. The first spikes in bursts were preceded by 100 ms of silence and were followed by another spike within 4 ms. Subsequent spikes in bursts followed at ISIs of 4 ms ( Lu et al. 1992) (see METHODS). Figure 1 illustrates the way in which the responses of mouse dLGN cells to drifting sinusoidal grating stimuli were analyzed using a double interspike interval (ISI) plot, revealing distinct temporal patterns of firing associated with burst responses. For each spike, subsequent ISI is plotted against preceding ISI. The distribution of points depends on the temporal properties of the stimulus and, more importantly, the balance between burst and tonic firing. The significance of the latter can be seen by comparing Fig. 1, A and C, in which example cells fire almost all of their spikes as part of burst or tonic firing, respectively. Figure 1B illustrates the more common occurrence in which both burst and tonic spikes contribute to a cell's responses.
|
100 ms of silence and are followed by another spike within 4 ms. Spikes within bursts, with pre- and post-ISIs of 4 ms, form the majority of spikes in zone 2, while the x-axis position of the cluster in zone 3 is due to these spikes being mostly the last spikes in bursts—they follow within 4 ms of a previous spike but are themselves followed by >4 ms of nonspiking "dead" time. In the example cell in Fig. 1A, dominated by burst responses, these zones contain almost all of the spikes fired. However, the temporal structure of the stimulus also affects the positioning of clusters in double ISI plots: assuming a neuron fires spikes to every stimulus cycle, maximum pre- and post-ISIs are constrained by the grating's temporal frequency (TF). In the example cell in Fig. 1C, dominated by tonic responses, the majority of spikes fall outside zones 1–3 but are constrained by the stimulus' TF. The example cell in Fig. 1B, in contrast, displays a complex ISI pattern indicative of important contributions from both burst and tonic firing. This produces a broad distribution of post-ISIs in zone 3: while many of these last spikes in bursts are the last fired in a cycle and are thus constrained largely by stimulus TF (zone 4), many others are followed by tonic firing and thus possess much shorter post-ISIs (zone 5). Similarly, within spikes fired as part of tonic firing, spikes clustered in zone 6 are the last spikes fired in a cycle, and those clustered in zone 7 are the first fired in a cycle.
Firing in mouse dLGN cells thus conforms to particular temporal patterns, some of which correspond well to those predicted by the properties of intracellularly recorded LTS bursts in the cat dLGN ( Lu et al. 1992). Separating mouse dLGN cell responses into burst and tonic components on the basis of the preceding criteria is therefore nonarbitrary: it is effective in classifying preexisting temporal patterns of neuronal firing. However, this classification is not quite perfect. Figure 1 shows that "tonic" spikes can sometimes occupy ISI clusters (zones 2 and 3) dominated by burst firing. Indeed, Lu et al. (1992) concede that their criteria are conservative, allocating to tonic firing 5–10% of spikes that actually form part of bursts.
Characteristics of burst firing in the mouse dLGN
Bursting appears to be a well conserved thalamic phenomenon. As assessed in vitro, the properties of the IT current and its accompanying burst activity are similar across thalamic nuclei ( Jahnsen and Llinas 1984; Ramcharan et al. 2000a) and across different mammalian species (guinea pig: Jahnsen and Llinas 1984; McCormick and Feeser 1990; cat: Zhan et al. 1999; monkey: Ramcharan et al. 2000a; mouse: Kim et al. 2001). Furthermore, it was shown in the previous section that distinct patterns of firing in the mouse dLGN can be identified by criteria used to identify burst spikes in other species (cat: Lu et al. 1992; monkey: Ramcharan et al. 2000b). However, we wanted to be more confident that the bursting criteria used here were identifying the same LTS-based phenomenon studied in other species. We therefore investigated whether certain features of burst firing reported to occur in those species also existed in the mouse dLGN.
Bursts are longer if preceded by greater periods of silence
It is known from in vitro recordings in the cat ( Zhan et al. 1999) and monkey ( Ramcharan et al. 2000a) that the number of spikes in a burst is not related to the strength of the depolarizing input stimulus but rather to the hyperpolarization of the membrane potential when that depolarizing stimulus arrives. In vivo, membrane hyperpolarization should depend on the degree of inhibition on a cell. Such inhibition will vary over time, but might have most chance to hyperpolarize a cell when that cell has been silent for an extended period. One might therefore predict that, although nonspiking "dead time" and membrane hyperpolarization need not go hand in hand, the number of spikes in bursts in vivo might be greater when bursts are preceded by longer periods of silence. In fact, across all bursts recorded in all experiments in all cells here, a small but significant positive relationship was noted between burst length and the duration of the preburst silent period (Spearman r = 0.16, P < 0.0001, n = 42,400). These data were summarized by taking the median preburst silent period for different burst lengths (Fig. 2A). For bursts consisting of 5 spikes, which represented >99% of all bursts recorded, there is a clear increase in the median period of preburst silence with increasing burst length. For larger bursts, this relationship breaks down. This could be due to the relatively small sample of long bursts. Alternatively, Rowe and Fischer (2001) used paired retinal and geniculate recordings to show that the first spikes in thalamic bursts are usually stereotypical events triggered by a single retinal spike but that later spikes in long bursts can be triggered directly by subsequent retinal action potentials. In other words, long bursts in vivo may not be long because of a large initial membrane hyperpolarization ( Zhan et al. 1999), but because they are triggered by two or more retinal action potentials closely spaced in time. If the latter is true, the relationship between preburst dead time and burst length might be expected to break down for the very longest bursts recorded.
|
It has been previously reported that burst length in the cat ( McCarley et al. 1983) and human ( Radhakrishnan et al. 1999) thalamus is correlated with the duration of the first within-burst ISI. This was confirmed for the mouse dLGN here. Across all bursts there was a strong negative relationship between the first ISI within a burst and burst length (Spearman r = –0.41, P < 0.0001, n = 42,400). Taking median initial intraburst ISI values for each burst length also reveals that bursts are longer when the first intraburst ISI is shorter (Fig. 2B).
ISI duration increases within bursts
Another widely reported feature of thalamic bursting is that ISIs become progressively longer within bursts (e.g., McCarley et al. 1983; Radhakrishnan et al. 1999; Reinagel et al. 1999). This feature was also observed here. Grouping data across bursts in the mouse dLGN, there was a significant positive correlation between within-burst ISI number and ISI duration (Spearman r = 0.22, P < 0.0001, n = 77,384). Taking median ISI duration values for each ISI position also shows a gradual increase in duration from the average first through to the average fourth ISI in a burst (Fig. 2C1). Later within-burst ISIs do not continue the trend but represent <1% of the ISIs analyzed and, as noted in the preceding text, might be expected to be less stereotyped than earlier within-burst ISIs ( Rowe and Fischer 2001). A more powerful demonstration of increasing ISI duration within bursts comes, however, from analyzing differences in within-burst ISIs within individual bursts. Figure 2C2 shows that the median difference in duration between a within-burst ISI and its preceding within-burst ISI is always positive. All median difference values here are significantly greater than zero (Wilcoxon rank sum test, P < 0.05). In other words, any given intraburst ISI tends to be shorter than the intraburst ISI that follows it.
Bursts are stereotyped events
Following Reinagel et al. (1999), the coefficient of variation (CV) of burst length was calculated for each experiment. Defined as the SD divided by the mean of any given sample, the CV should be 1 if sample values vary by chance. Here, most bursts (47%) consist of only two spikes. This stereotyped length is reflected in low CV values in every tuning curve experiment (Fig. 2D). The median CV value across all tuning curve experiments was 0.25, strikingly similar to mean values reported for cat X (0.25) or Y (0.23) cells ( Reinagel et al. 1999).
Rhythmicity of spontaneous burst firing
Spontaneous bursting in the dLGN neurons of naturally sleeping animals often occurs rhythmically ( McCarley et al. 1983; Weyand et al. 2001), with rhythmicity established by cyclic interactions between two depolarizing geniculate currents IT and Ih ( McCormick and Bal 1997). We assessed whether spontaneous bursting is also rhythmic in the mouse dLGN under conditions of halothane anesthesia. Our analysis was based on 5-s periods of spontaneous activity in which the stimulus display consisted of a blank screen at mean luminance (51 cd/m2); such blank presentations occurred once for every repeat of sinusoidal grating stimuli in a given tuning experiment (see METHODS). Rhythmicity was assessed by analyzing interburst intervals during such 5-s spontaneous epochs. For each epoch that contained at least four bursts, we calculated the SD of interburst intervals, with this SD measure being directly related to bursting rhythmicity: perfect rhythmic bursting would produce a SD of zero. We then assessed the statistical significance of each SD rhythmicity measure using Monte Carlo simulations that ranked the observed rhythmicity value among 999 simulated values (see METHODS).
Examples of burst timing in the responses of a single mouse dLGN cell during several 5-s spontaneous epochs are presented in Fig. 3A. These plots show that evenly-spaced bursts produce significant rhythmicity values and also demonstrate that spontaneous bursting in the mouse dLGN could alternate between rhythmic and nonrhythmic behavior over time in the same neuron. The histogram in Fig. 3B presents P values for all 234 spontaneous epochs (involving 45 cells) that contained at least four bursts. Forty four of these epochs (19%, involving 22 cells) were accompanied by significantly rhythmic bursting (P < 0.05). Of the 22 cells showing rhythmic spontaneous bursting, most (n = 12) did so in only a single epoch, although 3 cells had five rhythmic bursting epochs each. Rhythmicity in spontaneous bursting therefore occurs 20% of the time in the dLGN cells of a halothane-anesthetized mouse and is restricted to around half of geniculate neurons under such conditions, although a minority of neurons display rhythmic bursting rather often. This level of rhythmicity is considerably lower than the 47% of rhythmic episodes observed in the naturally sleeping cat ( Weyand et al. 2001). The discrepancy could arise because our spontaneous epochs were 5 s long, whereas Weyand et al. based their analysis on "burst bouts" of 2 s. However, when we also analyzed the rhythmicity of such 2-s burst bouts in our mouse preparation, we observed a reduction in rhythmic episodes: 48 of 438 bouts (11%) displayed significantly rhythmic firing. dLGN cells in anesthetized mice therefore do burst rhythmically but compared with geniculate neurons in naturally sleeping animals do so only rather rarely.
|
Neurons in the mouse dLGN fire spikes that can be identified as part of bursts (Fig. 1) and that can be shown to possess features characteristic of burst firing in other species and preparations (Fig. 2). But do these burst spikes encode visual information? A first, crude step toward answering this question comes from a comparison of bursting prevalence during periods of spontaneous and visually driven activity. Every investigation of a dLGN cell's responses to a certain stimulus parameter (known as a "tuning experiment") involved presenting a pseudorandom series of sinusoidal grating stimuli which varied in the parameter under study (spatial frequency, contrast, etc.; see METHODS) and also included the presentation of a blank stimulus at mean luminance. For each such tuning experiment, we calculated the percentage of spikes forming part of bursts (burst %) over all presentations of all grating stimuli (visually driven activity), and over all periods of blank screen presentation (spontaneous activity). We recognize that this approach does not compare matched periods of visually driven and spontaneous spiking, but our use of a percentage measure of bursting prevalence should compensate for this.
Although some cells' visual firing was dominated by burst firing (maximum burst % = 78%; see Fig. 1A), in most experiments bursts occupied a small fraction of all spikes fired (Fig. 4A1). However, burst percentage values for periods of spontaneous activity were usually even lower with many cells not firing any burst spikes in the absence of visual stimulation (Fig. 4A2). Indeed, across all tuning experiments (n = 323 in 102 cells), burst percentage values were significantly higher for visually driven versus spontaneous activity (visual median: 19.8%; spontaneous median: 11.1%; Wilcoxon matched pairs test, P = 0.002; Fig. 4B). Because this analysis was applied across all visual stimuli in a given experiment, not just those producing the highest firing rates, this is a rather clear demonstration that bursting is more prevalent when cells are responding to visual stimuli. This begs the question: exactly what visual information does burst firing carry?
|
Investigations into the relative timing of burst and tonic spikes during presentation of sinusoidal gratings suggest that burst spikes might carry important information about stimulus onset. Figure 5A shows the burst and tonic responses of a mouse dLGN cell to a sinusoidal grating of optimum spatial and temporal frequencies. As often reported by others (e.g., Guido et al. 1992, 1995; Lu et al. 1992) burst spikes, when they occur, tend to occur at the start of a response to a given stimulus cycle. Tonic spikes tend to occur a little later. This precedence of burst firing can have two possible mechanisms: burst spikes can have shorter latencies than tonic spikes or they can be triggered by earlier phases of the stimulus. The relative contribution of each of these candidate mechanisms can be assessed using plots of stimulus temporal frequency (TF) versus cell F1 response phase. The slope of a line relating these two variables can be used as a measure of response latency ( Grubb and Thompson 2003; Hawken et al. 1996). In addition, the y intercept of the line can give an indication of relative response phase (e.g., Guido et al. 1992; Mukherjee and Kaplan 1995). In mouse dLGN cells, burst responses to low TFs are usually phase advanced relative to tonic responses to the same stimuli. At high TFs, though, response phases for the two firing modes tend to be similar (Fig. 5B). This makes the TF-F1 phase line steeper for burst firing, meaning that burst spikes possess longer latencies than tonic spikes. Across all TF experiments, burst latencies were significantly longer than the latencies of tonic spikes fired to the same stimuli (burst median: 106 ms; tonic: 90 ms; Wilcoxon matched pairs test, P = 0.001; Fig. 5C). However, burst spikes still occur earlier than tonic spikes because they are triggered by an earlier phase of the stimulus. Phase advances for burst spikes, taken by subtracting tonic from burst y-intercept values in each experiment, were almost all positive, with the sample median being significantly different from zero (more positive values reflect earlier phases; 1-sample t-test, P < 0.0001; Fig. 5D). Despite their longer latency, burst spikes usually carry the first stimulus information relayed to cortex.
Response linearity
When all spikes are taken into consideration, almost all mouse dLGN cells sum inputs in a linear manner across their RFs ( Grubb and Thompson 2003). It is unlikely that such linear behavior could have a substantially nonlinear component, so it was reassuring to note that in almost all mouse dLGN cells both burst and tonic spikes show linear spatial summation. The plots in Fig. 6A display typical results of modified null tests ( Hochstein and Shapley 1976) applied separately to the burst and tonic components of cells' responses. These tests involve displaying stationary sinusoidal gratings the contrast of which is sinusoidally modulated in time, at various spatial phases across a cell's RF. Null tests were always initially applied with stimuli at a cell's optimal spatial frequency (SF) but were often repeated at higher SFs because nonlinearities of spatial summation can become more prevalent as SF increases (e.g., Hochstein and Shapley 1976). Twenty-eight cells were tested at optimal SF, 12 at twice optimal SF, and 3 at three times optimal SF. Temporal frequency of these stimuli was 1 Hz (11 cells) or optimal (17 cells), maximum stimulus contrast was 70%. By convention, fundamental (F1) response amplitudes are represented as negative in null test plots when their phase differs by 90–270° from that of the maximum response; this explains why burst and tonic plots from the same cell sometimes look like mirror images of each other.
In cells that sum inputs linearly across their RFs, responses to counterphased stimuli should be largest at the fundamental (F1) stimulus harmonic and the amplitude of this harmonic should vary sinusoidally with stimulus phase ( Hochstein and Shapley 1976). In addition, a linearly summating cell should possess two null phases where excitatory and inhibitory inputs are perfectly balanced across the RF ( Enroth-Cugell and Robson 1966). These null phases are represented by zero crossings in a null test plot ( Hochstein and Shapley 1976). In almost all cases in our mouse dLGN sample (27/28 cells, 42/43 null tests, Fig. 6A, 1 and 2), the F1 components of both burst and tonic firing were largest, varied sinusoidally in amplitude with stimulus phase, and crossed the zero line at 2 null locations. Such behavior was accompanied by linear "linearity" values (see METHODS) ( Hochstein and Shapley 1976) of <1.
One cell showed evidence of SF-dependent nonlinear spatial summation but did so in both response modes (Fig. 6A3). In both burst and tonic firing, although appearing to possess two null phases, the F2 response component of this neuron to stimuli at twice optimal SF was generally larger than the F1 component and varied sinusoidally with stimulus phase. Such a response pattern is characteristic of a RF with overlapping, nonantagonistic On and Off subregions ( Lennie and Perry 1981). While this cell displayed evidence of such ON-OFF nonlinearities in both its burst and tonic firing (Fig. 6A3), a high maximum F1 response in its burst firing meant that only its tonic component possessed a linearity value greater than unity, at 1.64.
Across all null test experiments, linearity values of burst spikes were not significantly different from those of tonic spikes fired to the same stimuli (burst means ± SE: 0.42 ± 0.02; tonic: 0.37 ± 0.04; paired t-test, P = 0.17; Fig. 6B). Both burst firing and tonic firing in the responses of mouse dLGN cells therefore sum spatial inputs in an equally linear manner.
Although burst and tonic firing do not differ in terms of their linearity of spatial summation, they do differ on a very different measure of response linearity, rectification. The response of a linear neuron to a sinusoidal grating stimulus should be sinusoidal, with a dominant F1 component (e.g., Shapley and Lennie 1985). However, even when a neuron displays wholly linear spatial summation its response to a sinusoid may not be sinusoidal because of response rectification: when spontaneous activity is low and visual responses strong and transient, cells cannot decrease their responses to the nonpreferred half cycle of a given grating, and may not precisely follow the temporal form of the grating's preferred half cycle. Such response rectification acts to increase all nonfundamental harmonics in a cell's responses. Because burst spikes tend to be high-frequency and transient, one might expect their rectification to be greater. Indeed, over all tuning experiments (n = 332 in 102 cells), the "rectification" of burst spikes (Eq. 1) was significantly greater than that of tonic spikes fired to the same stimuli (burst median: 0.46; tonic: 0.36; Wilcoxon matched pairs test, P < 0.0001; Fig. 6C). Thus although both burst and tonic spikes display linear spatial summation, burst spikes are significantly more nonlinear in terms of response rectification.
Spatial frequency tuning
As well as being identical with regard to spatial summation, burst and tonic spikes in the mouse dLGN are no different with regard to their SF tuning properties. SF tuning was assessed by fitting separate Difference of Gaussians (DoG) curves (Eq. 2) to the F1 amplitudes of burst and tonic responses to sinusoidal gratings of various SFs. Temporal frequency of these stimuli was 1 Hz, contrast was 70%. The DoG equation ( Rodieck 1965, So and Shapley 1981) models the center and surround mechanisms of a cell as symmetrical, antagonistic Gaussian functions, and can provide measures of the strengths and sizes of a given cell's center and surround RF regions. Because a DoG curve also provides a smooth fit to raw data, it can also be used to precisely calculate a cell's SF optimum (or peak) and cutoff (taken as the high SF at which response amplitude decayed to 1% of its maximum). However, due to the large size of RFs in the mouse dLGN, coupled with the limited size of our stimulus display at appropriate viewing distances, we were unable to accurately describe a cell's full low-SF roll-off (see Grubb and Thompson 2003).
As shown in Fig. 7A, DoG curves fitted to burst and tonic responses sometimes differed in their relative amplitudes (though not consistently in favor of either burst or tonic firing), but were usually similar in shape. We included in our analysis all cells (n = 61) in which the DoG function accounted for 80% of the variance of both burst and tonic raw data. In this sample, there were no significant differences in spatial tuning between the two response modes. Burst and tonic spikes fired to the same stimuli did not differ in terms of kc, the strength of the RF center (burst median: 7.59; tonic: 13.42; Wilcoxon matched pairs test, P = 0.43), ks, the relative strength of the RF surround (burst median: 0.98; tonic: 0.981; Wilcoxon matched pairs test, P = 0.06), rc, the radius of the RF center (burst median: 5.0°; tonic: 4.5°, Wilcoxon matched pairs test, P = 0.38), rs, the radius of the RF surround (burst median: 13.6°; tonic: 17.1°, Wilcoxon matched pairs test, P = 0.96), peak SF (burst means ± SE: 0.033 ± 0.003c/°; tonic: 0.032 ± 0.002c/°; paired t-test, P = 0.64; Fig. 7B), or SF cutoff (burst means ± SE: 0.17 ± 0.01; tonic: 0.16 ± 0.01; paired t-test, P = 0.56; Fig. 7C).
|
Burst and tonic spikes differ with respect to their temporal tuning. Temporal frequency (TF) tuning was assessed by presenting drifting sinusoidal gratings of various TFs at optimal SF and 70% contrast. F1 response amplitudes of burst and tonic firing to these stimuli were then separately fitted with two-half-Gaussian equations (Eq. 3). These functions tended to be narrower, peak at lower TFs, and fall off at lower high TFs for burst versus tonic firing (Fig. 8A). We included in our analysis all cells (n = 42) in which the two-half-Gaussians equation accounted for 80% of the variance of both burst and tonic raw data. In this sample, burst peak TFs were significantly lower than those of tonic spikes fired to the same stimuli (burst means ± SE: 3.9 ± 0.2 Hz; tonic: 4.5 ± 0.3 Hz; paired t-test, P = 0.007; Fig. 8B). Burst high50 values, taken as the high TF at which response amplitude dropped to half of its maximum, were also significantly lower than their tonic counterparts (burst means ± SE: 6.0 ± 0.3 Hz; tonic: 8.0 ± 0.5 Hz; paired t-test, P < 0.0001; Fig. 8C). To assess the tightness of TF tuning, a bandwidth measure was introduced. This measured, in octaves, the width of a TF tuning curve at half-maximum height. The bandwidth of burst spikes was significantly lower than that for tonic firing (burst means ± SE: 1.6 ± 0.1 octaves; tonic: 2.3 ± 0.2 octaves; paired t-test, P = 0.002; Fig. 8D). However, the low-frequency roll-off of burst and tonic TF tuning curves was not different. The measure low50, the low TF at which a curve reached half its maximum height, did not differ significantly between burst and tonic activity (burst means ± SE: 2.2 ± 0.2 Hz; tonic: 1.8 ± 0.3 Hz; paired t-test, P = 0.25).
|
100 ms (see METHODS). For this reason, spikes triggered by LTS activity might go undetected as bursts when, as is the case at high TFs, the period of sinusoidal grating stimuli approaches or is <100 ms. This is unlikely to affect any other data presented in this paper because stimuli in all non-TF tuning experiments were presented at 1 Hz or peak TF—always with a stimulus period well above 100 ms. However, it could affect our TF tuning results. We therefore investigated the effects of relaxing the requisite preburst silent period to 50 ms for stimuli with TF 8 Hz, following a precedent set by Sherman and colleagues ( Guido et al. 1992; Lu et al. 1992). Surprisingly, this had remarkably little effect on burst-tonic differences in temporal tuning. With bursts preceded by 50-ms silence when stimulus TFs were 8 Hz, we still saw significant differences between burst and tonic spikes in high50 (burst means ± SE: 5.9 ± 0.3 Hz; tonic: 7.6 ± 0.4 Hz; paired t-test, P = 0.0003) and bandwidth (burst means ± SE: 1.6 ± 0.1 octaves; tonic: 2.3 ± 0.2 octaves; paired t-test, P = 0.002), and there remained no burst-tonic difference in low50 (burst means ± SE: 2.0 ± 0.2 Hz; tonic: 1.6 ± 0.2 Hz; paired t-test, P = 0.16). In addition, response timing results, obtained using data collected during TF tuning experiments (see Fig. 5) were also unchanged: bursts were still phase-advanced (mean advance ± SE: 1.3 ± 0.2 rad; 1-sample t-test vs. 0, P < 0.0001) with a longer latency (burst median: 107 ms; tonic: 90 ms; Wilcoxon matched pairs test, P = 0.0008). The only different result noted when the preburst silent period was relaxed at high TFs was that peak TF was no longer significantly different between burst and tonic firing (burst means ± SE: 4.0 ± 0.3 Hz; tonic: 4.4 ± 0.3 Hz; paired t-test, P = 0.11). Note, however, that the trend for a higher TF peak in tonic firing remained. It therefore appears that the large majority, perhaps all, of burst-tonic differences in temporal tuning stem from fundamental differences in stimulus response, rather than from the 100-ms silence criterion used to identify burst firing in the present study. Contrast response characteristics
The contrast response functions of burst and tonic spikes were assessed by presenting drifting sinusoidal gratings of various contrasts at optimal SF. TF of these stimuli was either 1 Hz (18 cells) or optimal (14 cells). F1 response amplitudes of burst and tonic firing were then separately fitted with a hyperbolic function (Eq. 4) ( Albrecht and Hamilton 1982) and measures were extracted from the portion of this curve between 0 and 100% contrast ( Grubb and Thompson 2003). In most cases, both burst and tonic response amplitude grew rather gradually with increasing stimulus contrast (Fig. 9A1) or at least changed in a similar manner as contrast grew stronger (Fig. 9A3). In a few cases, though, a step-like function was found only in a cell's burst responses (Fig. 9A2). Contrast gain was calculated as the slope of a tangent to the curve where response amplitude was 20% of its value at 100% contrast. We included in our analysis all cells (n = 32) in which the hyperbolic function accounted for 80% of the variance of both burst and tonic raw data. In this sample, a trend was observed toward higher gain values for tonic firing. This trend, however, was not significant (burst median: 0.26 spikes · s–1 · %–1; tonic: 0.28 spikes · s–1 · %–1; Wilcoxon matched pairs test, P = 0.06; Fig. 9B). Neither was a significant difference found between burst and tonic spikes in terms of c50, the contrast at which response amplitude reached 50% of its value at 100% contrast (burst means ± SE: 33.5 ± 3.2%; tonic: 36.8 ± 3.2%; paired t-test, P = 0.33; Fig. 9C). However, because in vitro experiments have suggested that the contrast response function for burst spikes might be more step-like than that of tonic spikes ( McCormick and Feeser 1990; Ramcharan et al. 2000a; Zhan et al. 1999), the two firing modes were also compared with respect to the fitted curve parameter n (see Eq. 4). The larger this rate-of-change variable, the more step-like a curve is. n was significantly higher for burst spikes than for tonic spikes fired to the same stimuli (burst median: 3.34; tonic: 1.86; Wilcoxon matched pairs test, P = 0.0005; Fig. 9D).
|
The more step-like contrast response functions of burst spikes (Fig. 9D), as well as their strong rectification (Fig. 6C), are consistent with the hypothesis that burst firing in an individual dLGN cell provides more information about whether a given stimulus is present than about that stimulus' fine visual details (e.g., Sherman and Guillery 1996). In this hypothesis, burst spikes are primarily important for detecting visual stimuli. The relationship between burst firing and stimulus detection can be directly tested using receiver operating characteristic (ROC) analysis ( Green and Swets 1966; Guido et al. 1995; MacMillan and Creelman 1991). This approach assesses how well a cell's responses discriminate visual stimulation from periods of spontaneous activity, without ever employing a fixed detection criterion. This is achieved through the construction of ROC curves (Fig. 10A), which for a given stimulus plot the probability of observing different response sizes during spontaneous activity, P (false alarm), against the probability of observing those response sizes during visual stimulation, P (hit) (see METHODS). The curve always runs from (0,0) to (1,1), and if real signal detection is occurring all points lie above the unity line. Area underneath the ROC curve (ROC area) can then be used as a nonparametric measure of stimulus detection ( Green and Swets 1966; Guido et al. 1995; MacMillan and Creelman 1991).
|
) but cannot be carried out with confidence when data are obtained from extracellular recordings. This is because the construction of ROC curves requires a direct comparison of spontaneously and visually driven activity, a comparison that is considerably complicated by recording periods in which no spikes are fired. During intracellular recordings, it is a simple matter to identify the underlying response mode of a neuron based on its membrane potential. An intracellularly recorded lack of spontaneous firing, for example, can then be easily categorized as a lack of either burst or tonic firing and can be compared with a visually driven response (or lack of response) that occurred while the cell was in the same firing mode. When responses are recorded extracellularly, though, a response failure during, for example, visual stimulation, cannot similarly be compared with the appropriate period of spontaneously driven activity. With extracellular recording techniques therefore, we were not able to effectively compare spontaneous and visually driven activity in burst and tonic firing. Furthermore, we felt we could not simply ignore all stimulus cycles that produced a lack of burst or tonic activity because such cycles constituted a rather large proportion of all stimulus presentations: 75% of all stimulus cycles did not evoke a burst response, whereas 22% of all stimulus cycles did not evoke a single tonic spike.
However, we were able to assess the contribution of burst firing to stimulus detection more indirectly by asking whether the latter improved as the former became more prominent (cf. Guido et al. 1995). For each sinusoidal grating stimulus presented to each cell, we therefore calculated ROC area based on the total complement (burst + tonic) of the cell's responses. Assigning response failures to a particular firing mode was therefore not an issue. We also calculated a "burst index" for each stimulus defined as the proportion of all stimulus cycles generating at least one burst response ( Guido et al. 1995). The example ROC curves in Fig. 10A show that ROC area, and thus signal detection ability, tended to be larger when the burst index was higher. Indeed, over all stimuli presented to all cells, we observed a positive and significant correlation between ROC area and burst index (Spearman r = 0.59; P < 0.0001; n = 3,040). This relationship was similarly positive and significant when assessed only for the optimal stimulus in each tuning experiment (Spearman r = 0.54; P < 0.0001; n = 339), only for the nonoptimal stimuli in each tuning experiment (Spearman r = 0.58; P < 0.0001; n = 2701), or only for those stimuli associated with ROC areas >0.5, representing significant stimulus detection (Spearman r = 0.54; P < 0.0001; n = 2347). The correlation between ROC area and burst index was also positive and significant if, instead of using one ROC area and one burst index per stimulus, we averaged ROC area and burst index values across all stimuli presented to each cell (Spearman r = 0.34; P =0.0005; n =102; Fig. 10B). Indirect evidence therefore suggests that burst firing is related to greater signal detection in the responses of mouse dLGN cells.
Reliability in spike timing
If burst firing tends to occur first in a dLGN cell's responses to visual stimuli (Fig 5), does it provide reliable information concerning the time of stimulus onset? Although reliability in the absolute timing of spikes from stimulus onset cannot be assessed using the cyclic sinusoidal grating stimuli presented here, this question can be answered by taking a fixed point in the stimulus cycle and assessing reliability in the relative timing of spikes from that point. Reliability in burst and tonic spike timing was first assessed by calculating the mean deviation of spike times from the appropriate F1 response phase, normalized by the stimulus period (Eq. 6, Fig. 11A ). Note that higher timing reliability here is reflected in lower values of the reliability measure. Over all grating stimuli, burst spikes occupy less of the stimulus period than tonic spikes (burst reliability median: 0.07; tonic: 0.13; Wilcoxon matched pairs test, P < 0.0001; Fig. 11B). An individual burst spike thus informs the brain more about stimulus timing than an individual tonic spike fired to the same grating. In addition, this advantage in timing reliability for burst spikes increases with increasing response size. While both burst and tonic spike times become less variable as overall F1 response size moves closer to optimal (burst Spearman r = –0.33, P < 0.0001, n = 2,178; tonic Spearman r = –0.53, P < 0.0001, n = 2,178), the ratio of burst to tonic reliability also becomes lower (i.e., burst spikes are relatively less spread than tonic spikes) with increasing response optimality (Spearman r = –0.15, P < 0.0001, n = 2,178).
|
This relationship between response strength and reliability in spike timing was investigated in more detail for contrast response characteristics because of a specific prediction arising from an in vitro electrophysiological study. Zhan et al. (1999) found that the latency of the first burst action potential caused by a low-threshold calcium spike was longer after weaker depolarizing current injections. Presumably assuming that longer latencies are associated with greater variability in response timing, these authors predicted that increasing stimulus contrast in vivo would be met with increased reliability in burst spike timing. Indeed, the present data show that this is the case. Across all stimuli in all contrast experiments, stimulus contrast is negatively and significantly correlated with burst reliability (Spearman r = –0.22, P < 0.0001, n = 652) and burst first-spike reliability (Spearman r = –0.29, P < 0.0001, n = 405; see Fig. 11, D and E). And within contrast experiments, burst reliability and first-spike reliability values were both higher for the highest contrast for which they could be measured versus the lowest contrast for which they could be measured (reliability: low contrast means ± SE: 0.11 ± 0.01; high contrast: 0.07 ± 0.007; paired t-test, P < 0.0001; 1st-spike reliability: low contrast means ± SE: 126 ± 15 ms; high contrast: 100 ± 15 ms; paired t-test, P = 0.01). However, increasing reliability with increasing stimulus contrast is not a feature restricted only to burst firing. Across all contrast stimuli, the variability of tonic spike timing also decreased with increasing contrast (reliability: Pearson r = –0.43, P < 0.0001, n = 691; 1st-spike reliability: Pearson r = –0.34, P < 0.0001, n = 792; see Fig. 11, D and E). Within contrast experiments, too, the highest contrasts for which reliability values could be measured produced significantly more reliable tonic spike timing than the lowest contrasts for which these values could be obtained (reliability: low contrast median: 0.20; high contrast: 0.11; Wilcoxon matched pairs test, P < 0.0001; first-spike reliability: low contrast means ± SE: 154 ± 11 ms; high contrast: 70 ± 7 ms; paired t-test, P < 0.0001). Thus although the reliability of burst spike timing increases as stimulus contrast increases, as predicted by Zhan et al. (1999), tonic firing reliability becomes better with increasing contrast as well.
Analyses treating bursts as unitary events
In the preceding analyses, visual response properties are compared between spikes that form part of bursts and spikes that do not form part of bursts. We extended these analyses by asking what happens if instead of treating bursts as events made up of component spikes, bursts are analyzed as unitary events. In other words, are the same burst-tonic functional differences (and similarities) observed if we analyze the response properties, not of burst spikes, but of bursts per se? To address this issue, bursts were treated as unitary events, occurring at the poststimulus time of their first spike. Using this approach, we found an almost identical pattern of burst-tonic functional relationships to that outlined in the spike-based analyses described in the preceding text. As was the case when burst and tonic spikes were compared, burst events, in terms of bursts per second, were not significantly different (P > 0.05) from tonic spikes with respect to linearity of spatial summation, spatial frequency tuning, and contrast c50. Also in line with the preceding analyses, bursts as unitary events had higher visual than spontaneous firing rates, had advanced response phases but longer latencies than tonic firing, were more rectified than tonic spikes, had smaller ROC areas than tonic spikes, had more step-like contrast response curves than tonic firing, and, in their TF responses, had lower peak, high50, and bandwidth values (all P < 0.05). The only difference between analyses based on burst spikes versus those based on bursts per se occurred for the TF tuning measure low50—the low TF at which responses fell to half their maximal height. While burst spikes show a nonsignificant trend toward higher low50 values than tonic spikes (see preceding text), bursts as unitary events do possess significantly higher low50 values compared with tonic firing (burst median: 1.8 Hz; tonic: 1.4 Hz; Wilcoxon test, P = 0.03). This very small difference between two separate ways of assessing burst response characteristics suggests that the burst-tonic functional differences outlined in the preceding text are rather robust phenomena.
Burst-tonic differences in ON-center and OFF-center cells
In our previous description of visual response properties in the mouse dLGN ( Grubb and Thompson 2003), we looked for evidence of parallel processing of visual information by different cell types and found it only in functional differences between cells that respond primarily to increases in stimulus luminance (ON center) and cells that respond primarily to decreases in stimulus luminance (OFF center). It was therefore worth asking whether the burst-tonic differences in visual processing described in the preceding text occurred predominantly and/or consistently in either of the two main RF types in the dLGN.
For those response property parameters, which were normally distributed or which could be rendered normal by simple logarithmic transforms, we employed repeated-measures ANOVA tests, in which firing mode (burst/tonic) was the within-subject variable and center type (ON-OFF) was the between-subject variable. Any significance in the firing mode x center type interaction would then imply that a given burst-tonic difference was significantly stronger in cells of one or other center type. This ANOVA analysis was possible for 16 response parameters: latency, phase, linearity, rectification, kc, rc, rs, SF peak, SF cutoff, TF peak, high50, bandwidth, low50, contrast gain, c50, and contrast n. In most cases, the firing mode x center type interaction was not significant (P > 0.05). In four cases, however, the interaction was significant (P < 0.05), suggesting that burst-tonic differences in these parameters were stronger in either ON- or OFF-center neurons. Burst-tonic differences in response phase and rectification were significantly larger in ON-center cells, while differences between the two firing modes in latency and high50 were significantly greater in OFF-center cells. Still, the fact that burst-tonic comparisons were not significantly different between ON- and OFF-center cells for the majority of visual response measures, coupled with the observation that the four significant effects described in the preceding text were split evenly between ON- and OFF-center neurons, show that the burst-tonic differences described in the present paper are not due primarily to firing mode distinctions within cells of one single center type.
This conclusion is backed by consideration of those parameters that could not be rendered normally distributed by simple transforms. These response measures were compared by simply noting the results of paired burst-tonic comparisons applied to ON- and OFF-center data separately. We found positive and significant correlations between ROC area and burst index in both ON- and OFF-center neurons, observed that spike timing reliability is significantly better in the burst spikes fired by both ON- and OFF-center cells and noted that both center types possess no significant difference between burst and tonic spikes with respect to the RF surround parameter ks. Only in the remaining 2 response measures did burst-tonic differences appear to be dominated by cells of one particular center type. Like our dLGN sample as a whole, ON-center cells showed a significantly greater percentage of burst spikes in visually driven versus spontaneous firing. OFF-center cells, however, did not. Conversely, while OFF-center cells mirrored our overall results in that their burst first-spike reliability was significantly worse than that of tonic spikes, this effect was not significant in ON-center cells. Still, one burst-tonic difference due mainly to the responses of ON-center cells, and one due mainly to the responses of OFF-center cells, along with the ON-OFF similarities described in the preceding text, suggests that, overall, burst-tonic differences in the visual response properties of mouse dLGN cells are not due predominantly to the responses of cells of a single center type.
Why are burst spikes functionally different from tonic spikes?
Bursts possess two properties that distinguish them from the large majority of tonic spikes: high-frequency firing and a long preceding silent period. The possibility exists that the differences between burst and tonic spikes outlined in the preceding text might be due primarily to one or other of these distinguishing features—for example, differences in temporal tuning could be caused purely by the long silent period required for burst spikes to occur rather than being caused by LTS activity per se. To investigate this possibility, we compared the visual response properties of burst spikes with three types of tonic responses, all identified, like burst spikes themselves (see METHODS), on the basis of temporal patterns in spike firing. High-frequency tonic (HFT) spikes were separated by short interspike intervals of 4 ms, but, unlike bursts, groups of HFT spikes were not preceded by 100 ms of nonspiking dead time. In contrast, gap spikes were preceded by 100 ms of silence but unlike bursts had following ISIs of >4 ms. The final group, TX spikes, comprised the remainder of nonburst firing, and represented normal tonic activity (Fig. 12A). By comparing the visual response properties of these different types of activity, we hoped to glean some understanding of the underlying reasons for burst-tonic functional differences. For instance, the 4-way analysis might find that in the majority of cases where burst spikes differ from tonic spikes, HFT spikes differ from TX spikes in the same direction as the burst-tonic difference, but gap spikes do not. This might then indicate that the functional distinctions between burst and tonic responses outlined above are due mainly to the high-frequency firing component of bursts, rather than the requisite preceding silent period.
|
These results show that it was not possible to attribute burst-tonic functional differences, as a whole, to a single feature of burst firing. Perhaps assessing the firing properties of different subgroups of tonic spikes is not the best approach to this problem, but it was a sensible first step. Maybe intracellular recordings, by allowing a more accurate classification of burst, gap, and HFT spikes based on membrane hyperpolarizations and IT activity, might shed more light on the issue. Until such methods are employed, however, it is perhaps most straightforward to conclude that the collection of response properties that differentiate burst spikes from their tonic counterparts is produced by a complex combination of both high-frequency firing and a requisite preceding silent period.
|
|
DISCUSSION |
|---|
|
Comparison with previous studies
INCIDENCE OF BURSTING.
Criteria previously used to identify burst firing from extracellular recordings in the cat dLGN (e.g., Guido et al. 1992; Lu et al. 1992) were used successfully here in mouse dLGN neurons to classify firing events that already showed distinct temporal patterns (Fig. 1) and that possessed properties similar to those reported for burst firing in other species and preparations (Fig. 2). Coupled with in vitro thalamic recordings showing that IT and burst spikes are very similar across different mammalian species (guinea pig: Jahnsen and Llinas 1984; McCormick and Feeser 1990; cat: Zhan et al. 1999; monkey: Ramcharan et al. 2000a; mouse: Kim et al. 2001), these data suggest that burst firing is well conserved across the mammalian order. It will require intracellular recordings from mouse dLGN cells in vivo, however, for us to know for sure that the action potentials classified as burst firing in this study really were riding on a low-threshold calcium spike.
In terms of burst prevalence, 20% of spikes form part of bursts in the anesthetized mouse dLGN during presentation of sinusoidal grating stimuli (Fig. 4). This is comparable to the value of 20–25% reported for X cells during visual stimulation in the anesthetized cat ( Lesica and Stanley 2004; Reinagel et al. 1999), but much lower than the Y cell value under the same conditions (60%) ( Reinagel et al. 1999), and much higher than the 1–2% reported during visual stimulation in the awake monkey dLGN ( Ramcharan et al. 2000b). Bursting was likely facilitated by the anesthetized preparation employed here (cf. Guido and Weyand 1995; Reinagel et al. 1999). If this is the case, experiments in anesthetized animals may not produce results that can translate directly to behavioral situations but do present the best opportunity to compare the response properties of many burst and many tonic spikes fired by the same cells to the same sensory stimuli.
RELATIVE SPIKE TIMING.
Burst spikes in the mouse dLGN have longer latencies, but because they are triggered by earlier stimulus phases tend to occur before tonic spikes in responses to sinusoidal gratings (Fig. 5). The situation in the cat dLGN is almost identical. Cells that fire more burst spikes have longer latencies and larger phase advances ( Mukherjee and Kaplan 1995), while stimulating the histaminergic tuberomammillary nucleus increases tonic firing and produces a phase lag in responses to grating stimuli ( Uhlrich et al. 2002). Furthermore, a Fourier analysis of in vitro responses to sinusoidal inputs shows that the line relating TF to F1 response phase is steeper (indicative of longer latency), with a higher y intercept (indicative of a phase advance), for burst versus tonic spikes ( Smith et al. 2000). And although Guido et al. (1992) report that burst and tonic spikes in the cat dLGN have similar latencies, with burst spikes phase advanced, close inspection of their raw data reveals that their plots of TF versus F1 response phase are strikingly similar to those shown in Fig. 5 here: burst spikes are phase advanced at low but not high TFs. Had line fitting procedures not differed between the two studies, with Guido et al. (1992) employing response-weighted fits that decrease the influence of weak responses to high TFs, both might have reported longer latencies for burst spikes.
The phase advance and the longer latency of burst spikes could be consequences of the LTS that produces burst firing ( Jahnsen and Llinas 1984). By definition, the LTS is low threshold, and there is in vitro evidence that burst spikes can be elicited by smaller stimulation amplitudes than tonic spikes (e.g., Zhan et al. 1999). In vivo, smaller stimulation strengths, in the form of low rates of firing in retinal afferents, should occur at lower stimulus contrasts and therefore at earlier phases of the stimulating half cycle (dark or bright) of a sinusoidal grating. Burst spikes would thus be the earliest spikes triggered by a given cycle of such a stimulus. Long latencies, on the other hand, could be a consequence of the LTS's slow dynamics. In vitro, burst spikes require longer durations of stimulation than tonic spikes ( McCormick and Feeser 1990), and in vivo, retinal spikes preceding burst firing occur much longer before geniculate activity than retinal spikes preceding tonic firing ( Rowe and Fischer 2001).
Linearity
Burst and tonic spikes in the mouse dLGN do not differ with respect to linearity of spatial summation, but burst firing is more rectified than tonic activity (Fig. 6). This is the first study to use a null test ( Enroth-Cugell and Robson 1966; Hochstein and Shapley 1976) to assess linearity of spatial summation in the two thalamic firing modes. Because both burst and tonic spikes display linear summation in linear mouse dLGN cells, it would be extremely interesting to know whether the ON-OFF doubling nonlinearity of cat Y cells (e.g., So and Shapley 1979) is carried by burst spikes, tonic spikes, or both.
In contrast, rectification nonlinearities have been well studied previously. As in the mouse dLGN, burst spikes in the cat dLGN in vivo ( Guido et al. 1992; Lu et al. 1992; Mukherjee and Kaplan 1995) and in vitro ( Smith et al. 2000) follow the temporal form of a sinusoidal stimulus less closely than tonic spikes.
Spatial and temporal tuning
Burst and tonic firing in the mouse dLGN do not differ in terms of SF tuning (Fig. 7). No other studies have directly compared SF tuning in the two firing modes, although Guido et al. (1992) did compare the percentage of burst spikes fired to different stimuli across cells and found no variation with SF. In addition, stimulation of the tuberomammillary nucleus, which increases tonic firing in the cat dLGN, produced very little change in the SF tuning properties of geniculate neurons ( Uhlrich et al. 2002). Those small but significant changes that were produced—increased RF center strength (kc) and decreased RF center radius (rc)—could well have been produced by nonburst related effects of the stimulation but were nevertheless mirrored by nonsignificant trends in the mouse dLGN data. In reverse correlation studies of the spatial RF structure of burst and tonic spikes, initial results are inconsistent: whereas Rivadulla et al. (2003) find RF center size to be smaller for burst responses, Allitto et al. (2003) do not but do observe increased surround strength during burst firing. There are therefore no large or consistent differences in the spatial response properties of burst and tonic spikes.
Burst spikes in the mouse dLGN prefer lower TFs and show sharper TF tuning than tonic spikes (Fig. 8). Because burst firing can only occur after 100 ms of hyperpolarization, one might expect it to have difficulty in following higher TFs. Indeed, in vitro studies in the guinea pig ( McCormick and Feeser 1990) and cat ( Smith et al. 2000) also report that burst spikes stop responding at lower cutoff TFs than tonic spikes. The present finding of sharper TF tuning (i.e., lower bandwidth) in burst firing is also well supported by previous data. Tonic TF tuning in slices taken from the cat dLGN is more broadband than its burst counterpart ( Smith et al. 2000), while a higher incidence of bursting was correlated with more band-pass TF tuning in the intact cat dLGN ( Mukherjee and Kaplan 1995). There is some slight disagreement between the present data and the latter in vivo cat study, however. While burst spikes here displayed sharper TF tuning mainly by dint of reduced responses at high TFs, cat dLGN cells showed sharpened tuning associated with bursting because of reduced responses at low TFs ( Mukherjee and Kaplan 1995). This may reflect species or methodological differences between the two studies but could also be caused by random fluctuations between different samples. There was certainly a trend toward higher low50 values for burst firing in the mouse dLGN.
Contrast response characteristics
Comparison of contrast response characteristics showed that burst and tonic firing modes do not differ in terms of contrast gain or c50. However, the contrast-response curves of burst firing are more step-like (reflected in higher values of the parameter n) than those of tonic spikes (Fig. 9). This latter difference might have been predicted on the basis of in vitro investigations. Due to the all-or-nothing nature of the LTS, the amplitude of which depends not on the strength of depolarizing inputs but rather on the relative hyperpolarization of the membrane potential when those inputs arrive, the burst response of a dLGN cell to increasing current injection at a constant membrane potential resembles a step function. Tonic firing, in contrast, increases almost linearly with increasing stimulation strength ( McCormick and Feeser 1990; Ramcharan et al. 2000a; Zhan et al. 1999). Burst contrast-response functions observed in vivo may not be entirely step-like, and in vivo tonic functions are certainly not always completely linear (see Fig. 9A). Nevertheless, the relatively more step-like relationship between burst responses and stimulus contrast might be expected based on the properties of the LTS and represents a significant functional difference between burst and tonic firing.
Stimulus detection
The preceding discussion shows that differences in burst and tonic visual coding in the mouse dLGN usually reflect similar differences reported in the cat dLGN. The functions of the two firing modes therefore seem rather well conserved across these two mammalian species. This conclusion is bolstered by our analysis of stimulus detection: in the mouse, as in the cat ( Guido et al. 1995), better signal detection is associated with more prominent burst firing in a dLGN cell's responses (Fig. 10). It appears, however, that bursting may not be associated with better stimulus detection in all thalamic nuclei (or maybe in all species), given that a greater incidence of burst firing is not related to better stimulus detection in the guinea pig auditory thalamus ( Massaux et al. 2004).
Reliability of spike timing
In the mouse dLGN, the relative timing of all burst spikes is more reliable than that of all tonic spikes, but tonic firing is more reliable in terms of the first spikes fired to a stimulus (Fig. 11). A single burst spike therefore carries more information concerning the timing of a stimulus event than a single tonic spike. Combined with the fact that burst spikes occur earlier than tonic spikes in responses to a given stimulus, this may make bursting ideal for signaling to cortex when a stimulus appeared, even if the first spikes in bursts are not quite so reliable in their timing as the first spikes in tonic responses. This latter result contradicts a finding in the cat dLGN, where the reliability of first-spike timing was better for burst than for tonic responses ( Guido and Sherman 1998). This contradiction may reflect true species differences, possible differences in halothane anesthetic depth (see preceding text), or differences in the stimuli employed in the two studies. Unlike the flashed spots employed by Guido and Sherman (1998), the sinusoidal stimuli used here do not possess a single onset time and therefore might not be entirely appropriate for assessing spike timing properties. Still, the increased variability in the timing of burst first spikes in the mouse dLGN does appear to fit well with a different cat study, in which paired retinal and dLGN recordings showed burst soikes to be far more loosely and variably coupled to retinal inputs than tonic spikes (Rowe and Fischer 2001).
Although the main focus of this paper was simply to describe the different types of visual information that burst and tonic spikes in the mouse dLGN might send to visual cortex, it is worth briefly considering some implications of the mechanism by which burst and tonic stimulus selectivity arises in the first place. Tonic firing in thalamocortical relay cells is largely determined by excitatory retinal inputs with the correspondence rather tight between presynaptic spikes in retinal ganglion cells and postsynaptic spikes in geniculate neurons ( Rowe and Fischer 2001). The properties of the IT current underlying burst responses, however, mean that although burst spikes are normally triggered by excitatory retinal inputs, their stimulus selectivity should be largely determined by local inhibitory circuitry. Not only is membrane hyperpolarization needed to de-inactivate IT and allow burst firing to occur (e.g., Jahnsen and Llinas 1984), the degree of membrane hyperpolarization also determines the amplitude of burst responses: the more hyperpolarized the membrane, the more burst spikes result from a constant excitatory trigger ( Zhan et al. 1999). Measures of F1 response amplitude, such as those widely employed to assess stimulus selectivity in this paper and elsewhere, depend on both the probability of a response occurring to a particular stimulus cycle and the size of that response when it does occur. In burst firing, both of these factors will be largely determined by the degree of hyperpolarization present in dLGN cells before they receive an excitatory triggering input from the retina. It follows that burst spike response selectivity during the presentation of drifting grating stimuli is determined in large part by the stimulus half-cycle that inhibits the RF center rather than the stimulus half-cycle that excites the RF center and actually causes a spike response to occur. Because the excitatory "push" and inhibitory "pull" mechanisms operating on dLGN neurons are extremely well matched, at least spatially ( Hirsch 2003), this might not render burst and tonic stimulus selectivity too different. Indeed, we show that differences in burst and tonic visual response properties in the mouse dLGN, although significant, are rather small. However, this similarity might arise because of the cyclic nature of the stimuli used. Perhaps nonperiodic stimuli, such as natural scenes (cf. Lesica and Stanley 2004), that do not provide balanced and rhythmic durations of inhibitory pull and excitatory push will produce large differences in the visual response properties of burst and tonic firing and will allow us to better understand the functions of the two thalamic response modes.
What do bursts do?
The analyses presented here extend our knowledge of the visual encoding properties of burst and tonic firing in the dLGN and will be important when future studies investigate the molecular basis of thalamic processing using mutant mice (see INTRODUCTION), but do they shed any light on the fundamental functions of the two firing modes? They certainly confirm that tonic spikes are used to extract detailed information about visual stimulus attributes. Where tuning properties differ between burst and tonic firing, tonic spikes, with less rectification, broader TF tuning, better responsivity to high TFs, and more linear responses to increasing contrast, are better placed to signal information about a wider range of visual stimulus attributes. But it is not at all clear from the present analysis what the role of burst firing might be.
The prevailing hypothesis of burst function revolves around previous reports of greater rectification and signal detection in burst firing (e.g., Guido et al. 1992; Sherman 2001; Sherman and Guillery 1996). Burst firing is seen as a "wake-up call" for cortex, detecting stimuli but then leaving any detailed analysis to tonic firing. Indeed, the increase in stimulus detection that occurs with increased burst prominence in mouse dLGN cells (Fig. 10), along with the phase advance of burst firing (Fig. 5) ( Guido et al. 1992; Mukherjee and Kaplan 1995) and the reliable timing of burst spiking as a whole (Fig. 11) suggest that burst spikes may be ideal for signaling to cortex ( Swadlow and Gusev 2001; Swadlow et al. 2002) that a stimulus has occurred and that it started a certain time ago. Furthermore, given that the visual encoding differences between tonic and burst spikes are not huge (Figs. 5–8), this bursting detection signal could also contain a great deal of information concerning stimulus attributes, something for cortex to start to process before more detailed tonic spike-carried information arrives. In awake cats, burst spikes are fired at the very start of stimulus presentations and visual fixation periods ( Guido and Weyand 1995; Weyand et al. 2001), suggesting that bursting may indeed signal the presence of new stimuli to cortex. In addition, burst spikes fired by X cells in the anesthetized cat dLGN tend to occur when excitatory features follow the prolonged presence of inhibitory features in natural movie stimuli ( Lesica and Stanley 2004). In natural scenes, therefore bursts might encode the appearance of new objects. However, burst firing is also more prevalent during "passive" viewing in awake cats ( Weyand et al. 2001) and, when occurring before stimulus onset, increases the probability of a subsequent stimulus-induced response in rat somatosensory thalamus ( Fanselow et al. 2001). This suggests that burst spikes may not be detecting stimuli per se but are instead priming the brain to do so. It would be extremely useful to know whether bursting during stimulus presentation correlates with the animal's detection of that stimulus.
Another view of firing mode function, based on the different temporal tuning properties of burst and tonic spikes ( Mukherjee and Kaplan 1995) (Fig. 8), argues that the thalamus is a "tuneable filter" for visual information. Various modulatory inputs to the dLGN, by altering the probability of burst firing in relay cells (e.g., Godwin et al. 1996; Uhlrich et al. 2002), could alter the tuning of information sent to visual cortex. Given the present data this could certainly work in the contrast (Fig. 9) and TF (Fig. 8) domains, although it is unlikely to apply to spatial information (Fig. 7). However, it is unclear to what extent bursting tuning properties are functional adaptations and to what extent they are simply artifacts of burst properties that might have evolved for different reasons. It certainly seems strange that so much circuitry would be invested into changing temporal tuning in the dLGN when such a large amount of temporal information is discarded at the geniculocortical synapse (e.g., Hawken et al. 1996). And while increasing bursting might tighten temporal tuning in the dLGN, this would necessarily occur at the expense of detailed information about stimulus contrast (Fig. 9) and stimulus temporal form (increased rectification; Fig. 6). Still, until a definitive test of this hypothesis is formulated, it remains a valid suggestion.
Finally, there is a view that burst firing might be vital not for signaling specific stimulus attributes but for synchronizing activity across thalamic neurons. Sillito et al. (1994) showed that synchronous activity in dLGN cells is dependent on cortical feedback. In a model of these data, Kirkland and Gerstein (1998) show that this cortically induced synchronization is critically dependent on burst firing. The reliability of burst timing (Fig. 11) might be useful in this regard, although direct tests of the hypothesis, and a better understanding of whether bursts are linked more strongly to changes in sensory stimuli or changes in internal state, await multi-unit thalamic recordings.
|
|
ACKNOWLEDGMENTS |
|---|
|
|
|
FOOTNOTES |
|---|
Address for reprint requests and other correspondence: I. D. Thompson University Laboratory of Physiology, Parks Road, Oxford, OX1 3PT UK (E-mail: ian.thompson{at}physiol.ox.ac.uk)
|
|
REFERENCES |
|---|
|
Allito HJ, Weyand TG, and Usrey WM. Corticothalamic interactions and the role of thalamic bursts in visual processing. Soc Neurosci Abstr 29: 485.16, 2003.
Dräger UC. Receptive fields of single cells and topography in mouse visual cortex. J Comp Neurol 160: 269–290, 1975.[CrossRef][Web of Science][Medline]
Enroth-Cugell C and Robson JG. The contrast sensitivity of retinal ganglion cells of the cat. J Physiol 187: 517–552, 1966.
Fanselow EE, Sameshima K, Baccala LA, and Nicolelis MA. Thalamic bursting in rats during different awake behavioral states. Proc Natl Acad Sci USA 98: 15330–15335, 2001.
Godwin DW, Vaughan JW, and Sherman SM. Metabotropic glutamate receptors switch visual response mode of lateral geniculate nucleus cells from burst to tonic. J Neurophysiol 76: 1800–1816, 1996.
Gordon JA and Stryker MP. Experience-dependent plasticity of binocular responses in the primary visual cortex of the mouse. J Neurosci 16: 3274–3286, 1996.
Green DM and Swets JA. Signal Detection Theory and Psychophysics. New York: Wiley, 1966.
Grubb MS, Rossi FM, Changeux JP, and Thompson ID. Abnormal functional organization in the dorsal lateral geniculate nucleus of mice lacking the 
2 subunit of the nicotinic acetylcholine receptor. Neuron 40: 1161–1172, 2003.[CrossRef][Web of Science][Medline]
Grubb MS and Thompson ID. Quantitative characterization of visual response properties in the mouse dorsal lateral geniculate nucleus. J Neurophysiol 90: 3594–3607, 2003.
Guido W, Lu SM, and Sherman SM. Relative contributions of burst and tonic responses to the receptive field properties of lateral geniculate neurons in the cat. J Neurophysiol 68: 2199–2211, 1992.
Guido W, Lu SM, Vaughan JW, Godwin DW, and Sherman SM. Receiver operating characteristic (ROC) analysis of neurons in the cat's lateral geniculate nucleus during tonic and burst response mode. Vis Neurosci 12: 723–741, 1995.[Web of Science][Medline]
Guido W and Sherman SM. Response latencies of cells in the cat's lateral geniculate nucleus are less variable during burst than tonic firing. Vis Neurosci 15: 231–237, 1998.[CrossRef][Web of Science][Medline]
Guido W and Weyand T. Burst responses in thalamic relay cells of the awake behaving cat. J Neurophysiol 74: 1782–1786, 1995.
Hawken MJ, Shapley RM, and Grosof DH. Temporal-frequency selectivity in monkey visual cortex. Vis Neurosci 13: 477–492, 1996.[Web of Science][Medline]
Hirsch JA. Synaptic physiology and receptive field structure in the early visual pathway of the cat. Cereb Cortex 13: 63–69, 2003.
Hochstein S and Shapley RM. Quantitative analysis of retinal ganglion cell classifications. J Physiol 262: 237–264, 1976.
Hubel DH and Wiesel TN. Integrative action in the cat's lateral geniculate body. J Physiol 155: 385–398, 1961.
Jahnsen H and Llinas R. Electrophysiological properties of guinea-pig thalamic neurons: an in vitro study. J Physiol 349: 205–226, 1984.
Kim D, Song I, Keum S, Lee T, Jeong MJ, Kim SS, McEnery MW, and Shin HS. Lack of the burst firing of thalamocortical relay neurons and resistance to absence seizures in mice lacking alpha(1G) T-type Ca(2+) channels. Neuron 31: 35–45, 2001.[CrossRef][Web of Science][Medline]
Kirkland KL and Gerstein GL. A model of cortically induced synchronization in the lateral geniculate nucleus of the cat: a role for low-threshold calcium channels. Vision Res 38: 2007–2022, 1998.[CrossRef][Web of Science][Medline]
Lennie P and Perry VH. Spatial contrast sensitivity of cells in the lateral geniculate nucleus of the rat. J Physiol: 31: 569–579, 1981.
Lesica NA and Stanley GB. Encoding of natural scene movies by tonic and burst spikes in the lateral geniculate nucleus. J Neurosci 24: 10731–10740, 2004.
Livingstone MS and Hubel DH. Effects of sleep and arousal on the processing of visual information in the cat. Nature 291: 554–561, 1981.[CrossRef][Medline]
Lu SM, Guido W, and Sherman SM. Effects of membrane voltage on receptive field properties of lateral geniculate neurons in the cat: contributions of the low-threshold Ca2+ conductance. J Neurophysiol 68: 2185–2198, 1992.
Lu SM, Guido W, and Sherman SM. The brain-stem parabrachial region controls mode of response to visual stimulation of neurons in the cat's lateral geniculate nucleus. Vis Neurosci 10: 631–642, 1993.[Web of Science][Medline]
Macmillan NA and Creelman CD. Detection Theory: A User's Guide. New York: Cambridge, 1991.
Massaux A, Dutrieux G, Cotillon-Williams N, Manunta Y, and Edeline JM. Auditory thalamus bursts in anesthetized and non-anesthetized states: contribution to functional properties. J Neurophysiol 91: 2117–2134, 2004.
McCarley RW, Benoit O, and Barrionuevo G. Lateral geniculate nucleus unitary discharge in sleep and waking: state- and rate-specific aspects. J Neurophysiol 50: 798–818, 1983.
McCormick DA and Bal T. Sleep and arousal: thalamocortical mechanisms. Annu Rev Neurosci 20: 185–215, 1997.[CrossRef][Web of Science][Medline]
McCormick DA and Feeser HR. Functional implications of burst firing and single spike activity in lateral geniculate relay neurons. Neuroscience 39: 103–113, 1990.[CrossRef][Web of Science][Medline]
Mukherjee P and Kaplan E. Dynamics of neurons in the cat lateral geniculate nucleus: in vivo electrophysiology and computational modeling. J Neurophysiol 74: 1222–1243, 1995.
Paxinos G and Franklin K. The Mouse Brain in Stereotaxic Coordinates: New York: Academic, 2001.
Radhakrishnan V, Tsoukatos J, Davis KD, Tasker RR, Lozano AM, and Dostrovsky JO. A comparison of the burst activity of lateral thalamic neurons in chronic pain and non-pain patients. Pain 80: 567–575, 1999.[CrossRef][Web of Science][Medline]
Ramcharan EJ, Cox CL, Zhan XJ, Sherman SM, and Gnadt JW. Cellular mechanisms underlying activity patterns in the monkey thalamus during visual behavior. J Neurophysiol 84: 1982–1987, 2000a.
Ramcharan EJ, Gnadt JW, and Sherman SM. Burst and tonic firing in thalamic cells of unanesthetized, behaving monkeys. Vis Neurosci 17: 55–62, 2000b.[CrossRef][Web of Science][Medline]
Reinagel P, Godwin D, Sherman SM, and Koch C. Encoding of visual information by LGN bursts. J Neurophysiol 81: 2558–2569, 1999.
Rivadulla C, Martinez L, Grieve KL, and Cudeiro J. Receptive field structure of burst and tonic firing in feline lateral geniculate nucleus. J Physiol 553: 601–610, 2003.
Rodieck RW. Quantitative analysis of cat retinal ganglion cell response to visual stimuli. Vision Res 5: 583–601, 1965.[CrossRef][Medline]
Rowe MH and Fischer Q. Dynamic properties of retino-geniculate synapses in the cat. Vis Neurosci 18: 219–231, 2001.[CrossRef][Web of Science][Medline]
Schwarte LA, Zuurbier CJ, and Ince C. Mechanical ventilation of mice. Basic Res Cardiol 95: 510–520, 2000.[CrossRef][Web of Science][Medline]
Shapley R and Lennie P. Spatial frequency analysis in the visual system. Annu Rev Neurosci 8: 547–583, 1985.[CrossRef][Web of Science][Medline]
Sherman SM. Tonic and burst firing: dual modes of thalamocortical relay. Trends Neurosci 24: 122–126, 2001.[CrossRef][Web of Science][Medline]
Sherman SM and Guillery RW. Functional organization of thalamocortical relays. J Neurophysiol 76: 1367–1395, 1996.
Sherman SM and Guillery RW. The role of the thalamus in the flow of information to the cortex. Philos Trans R Soc Lond B Biol Sci 357: 1695–1708, 2002.
Sillito AM, Jones HE, Gerstein GL, and West DC. Feature-linked synchronisation of thalamic relay cell firing induced by feedback from the visual cortex. Nature 369: 479–482, 1994.[CrossRef][Medline]
Smith GD, Cox CL, Sherman SM, and Rinzel J. Fourier analysis of sinusoidally driven thalamocortical relay neurons and a minimal integrate-and-fire-or-burst model. J Neurophysiol 83: 588–610, 2000.
So YT and Shapley R. Spatial properties of X and Y cells in the lateral geniculate nucleus of the cat and conduction velocities of their inputs. Exp Brain Res 36: 533–550, 1979.[Web of Science][Medline]
So YT and Shapley R. Spatial tuning of cells in and around lateral geniculate nucleus of the cat: X and Y relay cells and perigeniculate interneurons. J Neurophysiol 45: 107–120, 1981.
Swadlow HA and Gusev AG. The impact of "bursting" thalamic impulses at a neocortical synapse. Nat Neurosci 4: 402–408, 2001.[CrossRef][Web of Science][Medline]
Swadlow HA, Gusev AG, and Bezdudnaya T. Activation of a cortical column by a thalamocortical impulse. J Neurosci 22: 7766–7773, 2002.
Uhlrich DJ, Manning KA, and Xue JT. Effects of activation of the histaminergic tuberomammillary nucleus on visual responses of neurons in the dorsal lateral geniculate nucleus. J Neurosci 22: 1098–1107, 2002.
Uhlrich DJ, Tamamaki N, Murphy PC, and Sherman SM. Effects of brain stem parabrachial activation on receptive field properties of cells in the cat's lateral geniculate nucleus. J Neurophysiol 73: 2428–2447, 1995.
Usrey WM, Reppas JB, and Reid RC. Specificity and strength of retinogeniculate connections. J Neurophysiol 82: 3527–3540, 1999.
Wagor E, Mangini NJ, and Pearlman AL. Retinotopic organization of striate and extrastriate visual cortex in the mouse. J Comp Neurol 193: 187–202, 1980.[CrossRef][Web of Science][Medline]
Weyand TG, Boudreaux M, and Guido W. Burst and tonic response modes in thalamic neurons during sleep and wakefulness. J Neurophysiol 85: 1107–1118, 2001.
Yu Y and Bradley A. Engineering chromosomal rearrangements in mice. Nature Rev Genet 2: 780–790, 2001.[CrossRef][Web of Science][Medline]
Zhan XJ, Cox CL, Rinzel J, and Sherman SM. Current clamp and modeling studies of low-threshold calcium spikes in cells of the cat's lateral geniculate nucleus. J Neurophysiol 81: 2360–2373, 1999.
This article has been cited by other articles:
|
|
 |
|
  R. R. Llinas and M. Steriade Bursting of Thalamic Neurons and States of Vigilance J Neurophysiol, June 1, 2006; 95(6): 3297 - 3308. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| Visit Other APS Journals Online |
TOP
ABSTRACT
INTRODUCTION
, 1 spike. In poststimulus time histograms (PSTHs; right) spikes are binned across all stimulus cycles according to their poststimulus time (pst). The "all spikes" PSTH possesses 2 peaks. Separating responses into burst and tonic components shows that these 2 peaks correspond to these 2 different firing modes. Burst spikes (middle) are concentrated at the start of the cell's response to the stimulus, whereas tonic spikes (bottom) tend to occur later in the stimulus cycle. B: dissecting possible factors underlying the precedence of burst spikes. In these plots of temporal frequency (TF) vs. F1 response phase, calculated separately for a given cell's burst and tonic spikes, the slope of the line relating the 2 variables can be taken as a measure of response latency and the y intercept of this line can give an indication of relative response phase (e.g., Guido et al. 1992
).
, show F1 response amplitudes to drifting sinusoidal gratings of various SFs; —, the best fit to these data of a Difference of Gaussians model (see METHODS). Note the different y-axis scales for the 3 examples, and the logarithmic x axis. Across these examples, and the mouse dLGN sample as a whole, the shape of SF tuning curves tended to be very similar for burst and tonic components of a given cell's responses. In some cells, burst responses were stronger than tonic responses (A1). More commonly, tonic firing was stronger than burst firing (A2). In other cells, burst and tonic tuning curves were similar in amplitude as well as shape (A3). Spatial tuning parameters for all 3 cells: cell 1: burst firing: kc 18.4, ks 0.67, rc 2.8°, rs 4.0°, peak 0.063 c/°, cutoff 0.27c/°; tonic firing: kc 6.7, ks 0.70, rc 2.9°, rs 6.0°, peak 0.064 c/°, cutoff 0.25 c/°; cell 2: burst firing: kc 7.4, ks 0.69, rc 3.8°, rs 17.0°, peak 0.031 c/°, cutoff 0.18 c/°; tonic firing: kc 12.4, ks 1.0, rc 2.6°, rs 22.9°, peak 0.029 c/°, cutoff 0.26 c/°; cell 3: burst firing: kc 5.2, ks 0.96, rc 3.8°, rs 12.4°, peak 0.041 c/°, cutoff 0.19 c/°; tonic firing: kc 4.3, ks 1.20, rc 3.5°, rs 14.0°, peak 0.04 c/°, cutoff 0.20 c/°. B: burst and tonic firing modes do not have different peak SFs. 
