Feedforward and recurrent inhibitory receptive fields of principal cells in the cat’s dorsal lateral geniculate nucleus
Principal cells in the dorsal lateral geniculate nucleus receive both feedforward and recurrent inhibition. Despite many years of study, the receptive field structure of these inhibitory mechanisms has not been determined. Here, we have used intracellular recordings in vivo to differentiate between the two types of inhibition and map their respective receptive fields. The feedforward inhibition of a principal cell originates from the same type of retinal ganglion cells as its excitation, while the recurrent inhibition is provided by both on- and off-centre cells. Both inhibitory effects are strongest at the centre of the excitatory receptive field. The diameter of the feedforward inhibitory field is two times larger, and the recurrent two to four times larger than the excitatory field centre. The inhibitory circuitry is similar for X and Y principal cells.
KeywordsDorsal lateral geniculate nucleus Postsynaptic inhibition Receptive fields
Almost 50 years have passed since Hubel and Wiesel  described the receptive field structure of cells in the cat’s dorsal lateral geniculate nucleus (dLGN). They found geniculate cells to have concentric receptive fields of on- or off-centre type, just like retinal ganglion cells. The only obvious difference was an increased peripheral antagonism of the centre response, which they proposed to be due to local inhibition in the dLGN.
Shortly afterwards, postsynaptic inhibitory potentials were recorded in geniculate cells by several investigators . Today, it is known that geniculate principal cells are equipped with two inhibitory circuits: a feedforward inhibitory system with the responsible interneurons located within the main layers of the dLGN and a recurrent inhibitory system with the interneurons confined to the perigeniculate nucleus [18, 32, 48]. Both types of inhibitory neurons have GABA as their main transmitter. The inhibitory circuits are similar but functionally segregated for X and Y cells . Apparently, the dLGN is composed of a large number of interdigitated neuronal “processing units” formed by individual principal cells and their inhibitory circuits .
These dLGN processing units have been analysed with anatomical and electrical stimulation techniques. While such procedures are excellent to reveal the serial order of connections between cells they cannot be used to identify other functionally important aspects of the circuitry such as the spatial distribution and receptive field type of the involved neurons. Nor can they reveal whether the feedforward or the recurrent inhibitory pathway is responsible for a particular inhibitory effect in the principal cell response to visual stimulation. Many studies, primarily based on indirect techniques, have aimed to elucidate such functions [7, 48, 49, 51], and a variety of models for dLGN inhibition have been proposed, none of which has received general acceptance. The problem is clearly related to the difficulty to differentiate between true inhibition and disfacilitation, i.e. reduced excitation of the principal cells. It seems firmly established, however, that the enhanced antagonistic surround of dLGN neurons is due to GABAergic inhibition [8, 50].
Two recent studies have reiterated the described difficulties. In an elegant in vitro study in mice, Blitz and Regehr  convincingly demonstrated that many feedforward IPSPs in principal cells are time-locked to unitary excitatory postsynaptic potentials (EPSPs) from retinal ganglion cells. This coupling, observed also in the cat , implies that at least some feedforward inhibition is provided by the very same retinal ganglion cells as the excitation. The other study, focusing on principal cell responses to natural visual scenes, arrived at the contrary conclusion, namely, that the feedforward inhibition originates from the opposing type of retinal ganglion cell . There is clearly a need to resolve this issue, especially since the two inhibitory systems appear to be independently controlled from the brain stem  and the basal forebrain . Such connections open the possibility that the two types of inhibition and thus, the visual transfer through the dLGN, might be differentially regulated in various behavioral situations.
In an attempt to resolve the disagreement in this otherwise well-studied system, we have combined direct intracellular recordings of inhibitory postsynaptic potentials in dLGN principal cells with adequate visual stimulation. Our procedure with strong cell depolarisation allowed feedforward and recurrent IPSPs to be distinguished and the receptive field structure of the two inhibitory mechanisms to be analysed.
Experiments were performed on 14 young adult cats (2.1–3.0 kg). Anaesthesia was induced with an alphaxalone–alphadolone mixture (Saffan, Glaxovet; 12 mg/kg, i.m. followed by sodium pentobarbitone (Apoteksbolaget; 25–30 mg/kg, i.v.). Additional small doses of pentobarbitone (5 mg/kg) were given to maintain an appropriate depth of anaesthesia during the surgical procedure and subsequent recording session. Heart rate and arterial blood pressure were continuously monitored. Mean arterial blood pressure was maintained above 120 mm Hg by slow infusion of bicarbonate-buffered Ringer glucose. Body temperature was kept between 37.5 and 38.5°C by a feedback-controlled heating lamp. During intracellular recordings the animals were paralysed by gallamine triethiodide (Flaxedil, May and Baker) added to the infusate. Continuous ECoG recordings were used in this situation to ascertain that the animals remained in a state of slow wave sleep. They were artificially ventilated through a tracheotomy with a positive end-expiratory pressure of 1–2 cm H2O. Tidal volume was adjusted to maintain end-expiratory CO2 at 3.5–3.7%. To further reduce respiratory-linked movements of the brain, the animals were suspended by a stereotactic head holder and a clamp placed at a midthoracic vertebrae. A pneumothorax was also performed prior to the recordings. The corneas were covered with contact lenses with a curvature appropriate to focus them on a tangent screen 1.5 m in front of the cat. Pupils were fully dilated, accommodation paralysed, and nictitating membranes retracted by local application of 1% atropine and 10% neosynephrine. At the end of the experiment, animals were killed by an overdose of the anaesthetics. The experiments were approved by the Regional Animal Care and Use Committee in accordance with Swedish law.
Intracellular recordings were obtained from dLGN principal cells with glass micropipettes filled with 3 M sodium acetate. Such electrodes gave better intracellular recordings than conventional electrodes filled with potassium salts . The microelectrodes had their tips broken back to a diameter of about 0.5 μm, giving a DC resistance of about 15–30 MΩ.
To obtain maximal stability, only a small craniotomy (2 × 3 mm) was made above dLGN. The microelectrode was then advanced through the overlying cortex down to the dLGN with a stepping motor micromanipulator. After the beginning of each track, the exposed cortex was covered by body-warm agar to further reduce pulsation. Principal cells were recognized in extracellular recordings by their position in the dLGN, by their receptive field properties and by antidromic activation from the visual cortex. When penetrated, the cells were rapidly hyperpolarised by a small negative current through the microelectrode in order to stabilise the recordings and to stop the injury discharge.
Only cells with reasonably stable recordings lasting for more than half an hour could be used for this study. In order to analyse inhibitory and excitatory potentials in isolation, it was necessary to polarise the cells by large current injections through the recording electrode (see “Results”). This was a brutal procedure and most cells eventually deteriorated. Since the current injections usually caused electrode polarisation and drifts in recorded DC potentials, it was, for most cells, impossible to deduce the actual membrane potentials during different phases of the analysis. Only a handful of cells remained stable, as far as we could judge, during the entire recording session (more than 2 h). When stepping out of these cells, DC potential changes in the 60–75 mV range were recorded. Since we intentionally tried to inactivate the spike mechanism by strong depolarisation, spike amplitudes could neither be used to monitor cell quality. As an alternative, we used the amplitude of unitary EPSPs evoked by optic nerve stimulation to monitor the stability of the recordings.
Receptive fields were plotted against a mesopic background. Their location and approximate size were first determined with extracellular recordings. To classify cells as X or Y types, we primarily relied on spatial resolution tests: receptive field centre size in relation to eccentricity and/or spatial resolution as tested with moving gratings of different spatial frequencies [13, 57]. There was sufficiently good correlation between these measures and cell classification based on thresholds and latencies with optic nerve stimulation to render more sophisticated visual classification schemes obsolete . To establish the eccentricity, the area centralis and optic disk position were mapped for each cell. The expected progression of receptive field positions with electrode advancement through the dLGN and the changes of ocular preference at laminar borders made histological confirmations unnecessary.
Once the cells were penetrated, their excitatory receptive fields were remapped by the recording of visually evoked unitary EPSPs. The cells were then depolarised to reveal unitary feedforward IPSPs and recurrent IPSPs. The size and contrast of the small light spots were adjusted individually for each cell to give a reproducible but weak response with resolvable unitary PSPs. Typical test spots were 0.3–0.6 times the diameter of the excitatory receptive field centre. These adjustments were necessary to allow the differentiation of feedforward and recurrent IPSPs. Thus, most stimuli were suboptimal with respect to response latency and amplitude, implying that data from different cells could only be compared qualitatively. The shutter of the projector was controlled by the stimulator, and consecutive points within the receptive fields were adjusted manually and marked on the screen.
The visual stimuli were delivered at a repetition rate of 0.5 Hz and lasted 0.5 or 1.0 s. The evoked responses were appropriately amplified and stored for offline analysis. Due to background synaptic noise, it was impossible to use automatic detection procedures, each response had to be inspected by eye at large amplification and expanded time resolution in order to identify evoked unitary PSPs. To compensate for the variability in visually evoked responses, each stimulus configuration was repeated five to ten times so that consistent components could be identified. To quantify feedforward inhibitory effects, we counted unitary EPSPs and IPSPs, scaled with an appropriate size factor (comp. the middle column of Fig. 8), during a predetermined time interval (typically including the time of phasic part of responses, approximately 100 ms). Quantitative data from many stimulus trials were than normalised with respect to the largest response and plotted as weighted means ±95% confidence intervals (CI). For some cells, the responses were averaged offline, once individual feedforward IPSPs had been identified. Recurrent inhibitory effects were also estimated by offline averaging, after subtracting identifiable feedforward IPSPs (comp. supplementary Fig. S3). The recurrent IPSP were quantified by integrating the IPSP area in averages of 5–10 repetitions using a straight horizontal baseline and own software (e.g. Fig. 9, position 6).
The present report is based on intracellular recordings from 51 functionally identified principal cells (25 X and 26 Y types). Half the sample consisted of on-centre cells (14 X, 12 Y), the rest were off-centre cells (11 X, 14 Y). Six X cells had antidromic latencies (for cortical stimulation) above 2.2 ms and slow responses to visual stimulation, indicating that they belonged to the lagged X cell category . Such cells are known to be among the smallest X cells . They did not differ in any other respect from the rest of the X cell sample, but including them made us to believe that we have studied a representative sample of X and Y principal cells. A summary of various cell parameters is provided in the supplement (Fig. S1).
Identification of feedforward and recurrent IPSPs
Our goal for this study required means to differentiate between feedforward and recurrent IPSPs during visual stimulation. These IPSPs differ considerably in time course, when evoked by electrical stimulation [32, 34]. Feedforward IPSPs have much shorter rise and decay times than recurrent IPSPs, both in X and Y cells (Fig. 1), and it seemed possible to utilise these characteristics also when analysing visually evoked responses.
The difference in IPSP time course is explained by a much stronger synaptic coupling in the feedforward than in the recurrent inhibitory pathway. Intrageniculate interneurons that mediate the feedforward inhibition receive large unitary EPSPs from a small number of retinal ganglion cells and fire impulses time-locked to such EPSPs [1, 33, 41]. The subsequent IPSPs evoked in principal cells are also large  and time-locked to spikes in the responsible optic nerve fiber, whether evoked by electrical or visual stimuli.
The recurrent system differs in all these aspects. Axon collaterals from many principal cells converge onto the same perigeniculate cell with small unitary EPSPs [3, Lindström and Wróbel, unpublished observations]. Their unitary IPSPs in principal cells are also small, compared to the feedforward IPSPs, and difficult to resolve individually . Compound IPSPs, as in Fig. 1, are formed by temporal summation of a large number of asynchronous unitary events, which explains their slow rise and decay time course. We expected the IPSPs to be even slower with asynchronous excitation of the perigeniculate cells from many principal cells during visual stimulation.
Note that the recurrent IPSPs in Fig. 1 were evoked at a stimulation intensity subthreshold for activation of corticogeniculate axons as revealed by the lack of fast IPSPs from intrageniculate interneurones superimposed on the slow recurrent IPSPs and by the lack of corticogeniculate EPSPs in the principal cells (not shown [6, 33, 35]).
With depolarising current injections (upper pairs of traces), the optic nerve stimulation evoked a combined EPSP-IPSP response (A). The IPSP had a disynaptic latency indicating that it was mediated by the feedforward pathway . Similar PSP combinations were seen in the records with spontaneous activity (B) with the EPSP indicated by a small upwards deflection followed by an IPSP. Visual stimulation of the receptive field centre (C) evoked a burst of IPSPs that were also similar in size and time course as the feedforward IPSPs evoked by electrical stimulation.
Note that practically all PSPs were of the combined type with the IPSP time-locked to a small positive deflection, representing a preceding EPSP. There was a fixed latency difference of 0.8 ms, between the two PSP components, whether evoked by electrical stimulation, visual stimulation or during spontaneous activity. The interval was also the same after reversal of the IPSP by the injection of a strong hyperpolarising current (not illustrated, see Ref. ). This latency difference proves that the biphasic responses were combined EPSPs-IPSPs and not partially reversed EPSPs. The tight temporal coupling of the two responses demonstrates that they were triggered by single spikes from the same excitatory retinal ganglion cell [10, 34].
Combined EPSPs-IPSPs were found in the large majority of the recorded cells (46/51), especially during visual stimulation. For the receptive field analysis, we usually depolarised the cells, so that biphasic PSP could be resolved, rather than to the EPSP reversal level. In the following, feedforward IPSPs time-locked to EPSPs will be referred to as “locked IPSPs,” while those without preceding EPSPs will be called “nonlocked IPSPs” (following the suggestion by Blitz and Regehr ). The lower response to visual stimulation starts by a large nonlocked feedforward IPSP (asterisk, Fig. 2c, second response). A smaller nonlocked IPSP occurred after an EPSP failure in the response to OT stimulation (Fig. 2a). Note that both types of IPSPs could vary considerably in amplitude. Measurements from a long sequence of recordings from the illustrated cell showed feedforward IPSPs of at least four amplitudes and occasional EPSPs without an IPSP. The stepwise change in IPSP amplitude is explained by the convergence of several feedforward inhibitory interneurons onto the same principal cell , with the interneurons being influenced by the same retinal ganglion cell(s). A variation in the firing probability of these interneurons would give IPSPs of different amplitudes coupled to the EPSPs. Interestingly, the maximal IPSP that occurred spontaneously or after visual stimulation had in this and several other cells the same amplitude as the maximal feedforward IPSP evoked by electrical stimulation of the optic nerve.
In the illustrated case (Fig. 3a), the slow stimulus evoked hyperpolarisation increased in amplitude with depolarisation and reversed polarity with the hyperpolarising current, as expected for a true IPSP. A similar late slow IPSP component was hidden by the fast feedforward IPSPs in the light off mode of stimulation (Fig. 3b, d). Since perigeniculate neurons receive convergent excitation, via axon collaterals of both on- and off-centre principal cells, it is not a farfetched conclusion that both these slow IPSPs were of the recurrent type (although we cannot exclude some contribution from calcium spikes in intrageniculate interneurons in the off mode; ). Note in the diagrams (Fig. 3d, e) that the recurrent IPSP at light on had about the same reversal level as the feedforward IPSP evoked in the same cell by optic tract stimulation (displayed at higher time resolution in Fig. 3c). This correspondence would be expected if both IPSPs involved GABA-A regulated Cl- channels. We did not observed slow IPSPs with more negative reversal levels that could qualify as GABA-B mediated effects, connected to K+ channels.
Basic pattern of feedforward inhibition
An enlarged spot covering most of the excitatory receptive field centre produced stronger inhibition with shorter latency at light off (spot 2). A close inspection at larger amplification revealed that also this inhibitory response was evoked by temporally summating fast IPSPs of feedforward type (not illustrated). Only about a fifth of the fast IPSPs were nonlocked as judged by many similar stimulation sequences. At light on, there was a late, slow hyperpolarisation but no fast IPSPs. This late hyperpolarisation was a recurrent IPSP, which will be described later.
A small spot in the receptive field periphery (Spot 3) was completely ineffective both at light off and on. With an annulus, only a slow hyperpolarisation was evoked at light off. The single fast IPSP occurring after about 60 ms was due to ongoing spontaneous activity and was not seen in other records. At annulus on there was a late burst of EPSPs representing the surround response of the excitatory retinal ganglion cell. As with the centre response, there were feedforward IPSPs locked to these EPSPs.
For simplicity, each stimulus type is only illustrated by single representative records in Fig. 4. Due to the spontaneous activity of retinal ganglion cells such traces contain much synaptic noise. To identify stimulus dependent activity it was always necessary to study many individual responses at high resolution. The reproducibility of the responses in this cell can be judged from more single traces displayed in Fig. 8.
For X cells, the situation was straightforward and simple: with the opposite phase of centre stimuli, the IPSP activity ceased together with that of the EPSPs (Fig. 5a, b). With extracellular recordings, there is a pause in the spike activity of principal cells in this phase. Clearly, this pause is primarily due to lack of excitation (disfacilitation), caused by arrested activity in the input retinal ganglion cell(s) and not by active inhibition. The same is true for the pause in principal cell firing, caused by an opposing annulus in the periphery. Feedforward IPSPs could neither be seen in this phase. From the surround of X principal cells, feedforward IPSPs only occurred together with long-latency EPSPs resulting from surround activation of the input ganglion cell(s). It should be stressed that our visual stimuli were strong enough to activate neighbouring principal cells of on and off-centre types in each preparation. So our failure to detect opposing feedforward IPSPs cannot be due to inadequate visual stimulation. In many X cells, the largest unitary IPSPs were similar in amplitude to the maximal feedforward IPSPs evoked by electrical stimulation. so at least for these cells, there was no further feedforward inhibition to be accounted for.
For Y principal cells, the analysis was more complicated. They were “noisier” than X cells, due to excitatory convergence from several retinal ganglion cells. Their input ganglion cells also gave phasic on–off discharges to spot stimuli placed in an overlap zone between the centre and the surround. These overlap zones were spatially displaced for the different converging ganglion cells, resulting in on–off EPSPs from a large portion of the principal cell receptive field. Usually, the Y principal cells also had large recurrent IPSPs of short latency. Such difficulties made it complicated to illustrate the selectivity of connections to such cells. However, after viewing many stimulus response sequences for each cell we became convinced that no Y cell had a significant feedforward inhibitory input from ganglion cells of the opposite type to those providing centre excitation. No unlocked IPSPs of feedforward type were evoked by opposing visual stimuli in their receptive field centre—only occasional spontaneous IPSPs at a low rate (cf. traces with spontaneous activity, Fig. 5c, d). From the surround region, IPSPs of feedforward type only occurred in association with excitatory surround responses from driving retinal ganglion cells. As expected, these surround responses invariably had a longer latency than the centre response. Thus, it is concluded that also Y cells lack feedforward inhibition from the opposite type of retinal ganglion cells.
Feedforward inhibitory receptive fields
The spatial structure of the feedforward inhibitory receptive field was determined quantitatively in 16 principal cells (ten X and six Y cells, half of each category were on-centre and half off-centre cells, four X cells were lagged). Two measurements were obtained, a receptive field profile determined by small test light spots placed in different positions along one or two axis through the excitatory receptive field centre and an area-response function with centred spots of different diameters.
Feedforward receptive field profiles
As an independent measure we averaged the feedforward IPSPs and integrated the area of the inhibitory response (Fig. 6, right column). Note that the slow rise of the IPSPs in these traces is due to smoothing by the averaging procedure. The measurements were restricted to the first 70 ms after stimulus on, since recurrent IPSPs contributed at longer intervals. With this method the size of the inhibitory response is somewhat underestimated for the receptive field centre due to interaction with EPSP currents. This interaction explains why the resulting inhibitory profile (Fig. 6, dotted line) is shallower than that based on IPSP counts. Even so, the width of the inhibitory field at 1/e was about the same. Similar profiles were found along the orthogonal axis of the receptive field with 1/e diameters of 0.5° for the EPSP and 0.8° for the feedforward IPSP field (not illustrated).
The organisation of the feedforward inhibitory field was similar in all the other tested principal cells with the peak of inhibition centred at the excitatory field centre. Further, the estimated inhibitory field diameter was not more than twice that of the EPSP field. For the total sample, the ratio of IPSP/EPSP field diameters ranged from 1.0 to 2.1 with no obvious difference between X and Y principal cells of either on- or off-centre type.
Feedforward area-response functions
We were intrigued by the findings that the feedforward inhibitory effect was strongest from the receptive field centre and that the effect decayed so rapidly at positions outside the excitatory centre. It was intuitively difficult to understand that an inhibitory system organised in this way could provide the enhanced antagonistic surround, as described by Hubel and Wiesel . One possible explanation could be that the inhibitory input from peripheral regions required more spatial summation and was underestimated by our small test spots. For this reason we investigated area-response functions for feedforward IPSPs in 12 X (six on-centre and six off-centre) and 11 Y (eight on- and three off-centre) neurons. Three X and three Y cells were recorded sufficiently long to determine area-response curves with full range of spot diameters with representation of both on- and off-centre types.
From the diagram, it can be seen that the highest EPSP count was obtained with a spot size of 0.6°. This stimulus covered the entire receptive field centre, as determined by the initial recording of extracellular spike activity. The EPSP count decreased with larger spot diameters covering more and more of the receptive field surround of the input retinal ganglion cell. The feedforward inhibition was strongest with a spot diameter twice as large (1.2°) as that for optimal excitation. With larger stimuli also the inhibition decreased with a similar slope as for the excitation.
The larger summation area for the feedforward inhibition seemed sufficient to account for the enhanced surround inhibition of this principal cell. This fact is best appreciated if we compare the responses to a small and large spot giving the same amount of excitation of the principal cell. In fact, Hubel and Wiesel  used this procedure in combination with extracellular recordings to propose that the increased peripheral suppression at the dLGN level results from a local inhibitory process. For the illustrated cell, this same amount of excitation was obtained with stimulus spots of 0.2° and 1.8°, which both produced EPSP counts slightly above 60% of the peak activity. The larger of these spots evoked considerably more feedforward inhibition, however, which is easily seen from the sample records to the right.
A close scrutiny of many records at high magnification revealed that this difference was partly due to an increased number of nonlocked feedforward IPSPs, i.e. IPSPs without a preceding unitary EPSP. Indeed, twofold increase of nonlocked inhibition was observed with stimulus spot change from 0.2° to 4.8° (Fig. 8, diagram to the left; n = 10; t-test, p > 0.01). The inhibitory responses in two of the traces to the right started with such nonlocked IPSPs (small arrows).These feedforward IPSPs were obviously evoked by activity in other off-centre ganglion cells than that providing excitation to the recorded principal cell. The responsible cells apparently had their receptive fields slightly displaced with respect to the excitatory field centre. Note that the curve for inhibition is below that for excitation up to a diameter of 0.9° and above at larger stimulus diameters. Although the weight of the excitation and inhibition are not quite comparable, the shift of the curves illustrates that excitation dominates for smaller stimuli and inhibition for larger. In addition, for the other studied cells, including the Y cells, the diameter of the summation area for the feedforward inhibition was about twice as large as that of the excitatory centre.
Basic pattern of recurrent inhibition
As already described, slow presumed recurrent IPSPs were regularly evoked in the principal cells by centre stimuli of the opposite type to their excitatory input (Figs. 3, 4 and 5). Such IPSPs were also elicited by on and off stimulation in the surround of the principal cell receptive field. For many cells, there was also a late IPSP component in the excitatory phase of centre stimulation which was not accounted for by summation of feedforward IPSPs (Fig. 3). The underlying IPSP component had a slow rising phase with an overall latency similar to that of recurrent IPSP in the opposite phase of stimulation. These patterns are consistent with the known convergence of excitation of individual perigeniculate neurons as discussed above.
Threshold recurrent IPSPs required a higher stimulus intensity or summation from a somewhat larger stimulus area than feedforward IPSPs. For suboptimal stimuli, the latency of the recurrent IPSPs was also much longer than for the feedforward centre response, especially in X cells. We did not try to quantify these differences since they are mainly anaesthesia-dependent. Perigeniculate cells, especially those of X type, have a sluggish response in barbiturate anaesthetised animals . Thus, the long latency of the recurrent IPSPs resulted from our experimental procedure and does not reflect the normal operation of the recurrent pathway during active vision. The effect was convenient for the present analysis, however, since it simplified the differentiation between visually evoked feedforward and recurrent IPSPs.
For more optimal visual stimuli, the latency difference between recurrent and feedforward IPSPs became negligible. The recurrent IPSPs had also similar latencies in on- and off-centre cells (not illustrated), whether evoked from the receptive field centre or its surround or by light on or off stimuli (supplement; Fig. S4). This similarity is interesting since it implies that the recurrent IPSPs originated from centre stimulation of both on- and off-centre principal cells. Direct proof for the involvement of centre responses was obtained by stimulation with annuli. In such cases, the recurrent IPSPs always preceded the EPSPs elicited by surround activation of the ganglion cells (supplement; Figs. S2 and S3). These findings are in keeping with previous observations from perigeniculate cells [4, 70].
Recurrent inhibitory receptive fields
Recurrent receptive field profiles
Centre stimulation also evoked the strongest recurrent inhibition in the other cells. The diameters of the recurrent field profiles were typically larger than those of feedforward IPSPs. Compared to the EPSP fields, the recurrent field diameters were 1.5 to three times larger, with no obvious difference between X and Y cells. These figures should be taken with some caution since the recurrent responses were very variable from trial to trial in our anaesthetised preparation, especially for stimuli in the marginal zone. The variability was particularly pronounced during periods of spindles in the ECoG when spontaneous recurrent IPSPs interfered with the visually evoked IPSPs. In several cells, the recurrent inhibitory fields were asymmetrical and different for on and off stimuli (Fig. 9, positions 2 and 8). In some parts of the receptive field, light on stimuli gave stronger effects than light off, in other parts the opposite could be true. We found no rules for this asymmetry. It presumably reflects an uneven, limited convergence of excitation from on- and off-centre principal cells onto individual perigeniculate neurons.
Recurrent area-response functions
Recurrent IPSPs of about the same amplitudes also occurred in the off phase of stimulation, but here, they were preceded by feedforward IPSPs forming the early part of the response. (Note that the fast rise time of unitary feedforward IPSPs is smoothed by the averaging procedure). The recurrent contribution in this phase appeared as a second inhibitory component (arrows, Spot 3) with about the same latency as the recurrent IPSP in the on phase of stimulation.
The most interesting observation, with respect to the recurrent inhibition, occurred with wide field stimuli. Such stimuli completely failed to evoke any recurrent inhibition in X cells. This failure was apparent both in the off- and on-phase of stimulation with the largest spot (Fig. 10, Spot 6). At the expected latency for the recurrent IPSPs, there were no inhibitory responses, only small depolarisations representing disinhibition of the principal cell. Both the lack of recurrent inhibition and the disinhibition can be ascribed to the fact that X principal cells stop to fire spikes in response to wide field visual stimulation. Note that the feedforward inhibitory pathway was still operative as revealed by the early IPSP in the off response and the late hyperpolarisation in the on response. The latter was due to feedforward IPSPs evoked by surround activation of the off-centre retinal ganglion cell. Similar effects of wide field stimulation were found for all 11 X cells tested with wide field stimuli. It follows from this kind of observation that the recurrent inhibitory pathway cannot be the main source of enhanced lateral inhibition of X principal cells, at least not for large stimuli. This finding makes it also unlikely that the slow IPSPs, inferred by us to represent recurrent IPSPs, included a substantial component of slow IPSPs generated locally at synaptic triads by intrageniculate interneurones, whether induced by dendritic calcium spikes  or activation of dendritic metabotropic glutamate receptors . Both effects would be expected to be enhanced rather than suppressed by massive visual stimulation. Note that X-type intrageniculate interneurones maintain a substantial response to diffuse illumination (just like retinal ganglion cells and in contrast to X-type principal cells; ).
The situation might be different for Y cells. All tested Y principal cells (including nine additional cells with incomplete area response plots) displayed clear recurrent IPSPs in responses to wide field stimuli. Most Y principal cells also respond to such stimuli with a short burst of spikes both at light on and off. For the three Y cells, it was only feasible to determine the width of the recurrent inhibitory field using stimuli of the opposite phase to the centre excitation. In the latter phase of stimulation, it was impossible to quantify the recurrent IPSPs due to temporal overlap with feedforward IPSPs (supplement; Fig. S2). The recurrent summation area was considerably larger than that of feedforward inhibition and exceeded the diameter of the EPSP summation area at least three to four times. The inhibition decreased in amplitude with larger stimuli, but about half its peak amplitude remained with wide field stimuli.
The present results demonstrate that the feedforward and recurrent inhibitory pathways of the dLGN carry different information and thus may subserve different functions. While the recurrent pathway combines information from the on- and off-centre systems, the feedforward pathway is type-specific, with the inhibition exclusively originating from the same type of retinal ganglion cells as the excitation. The receptive fields of both types of inhibition are bell-shaped and aligned with the excitatory receptive field. The diameter of the feedforward inhibitory field is about twice that for excitation while the recurrent inhibitory field is about two to four times larger. The structural organisation of the inhibitory fields is the same for X and Y principal cells, extending our previous observation that the inhibitory circuits of X and Y principal cells are similar although functionally segregated .
The strength of our observations is that visually evoked IPSPs were observed directly by intracellular recordings from dLGN principal cells. By polarising the cells, it was easy to differentiate between true postsynaptic inhibition and decreased excitation (disfacilitation), a distinction rarely made in the visual literature. From the shape and latency of the inhibitory potentials, it was also possible to differentiate between effects mediated by the feedforward and recurrent inhibitory pathways. The rationale behind this distinction is elaborated upon in the Results and will not be considered further.
There is a small possibility that some of the visually evoked IPSPs observed in our experiment would reflect the interneuronal activity evoked by the excitatory feedback loop via layer six pyramidal cells in the primary visual cortex . Such activity would remain separated according to our criteria and contribute separately to the described feedforward and recurrent IPSP fields. We think, however, that such additional inputs are negligible in shaping the inhibitory fields as described here. First, layer 6 cells are virtually unresponsive to small spot stimuli as used here, especially in pentobarbital anaesthesia . Secondly, the perigeniculate cells would inhibit other perigeniculate cells and feedforward interneurons shortly after their initial response [6, 71]. Thirdly, even in the nonanaesthetised preparation, the cortical input was shown to be only modulatory to the spontaneous activity: increasing it for principal cells and decreasing for perigeniculate cells .
Only postsynaptic inhibitory effects were revealed by our recordings. Any presynaptic inhibition would have been undetected. At present, there is little evidence for specific presynaptic inhibition in the dLGN , although several transmitter substances have been found to modulate EPSP amplitudes in vitro . Slow, long-latency IPSPs of GABA-B type [16, 55] were not observed in our responses whether evoked by electrical or visual stimulation. This negative result was probably due to a pronounced frequency depression of GABA-B IPSPs  caused by the high spontaneous activity of geniculate interneurons in vivo. It emphasises that known inhibitory effects on visual responses of dLGN principal cells is removed by GABA-A antagonists [8, 50, 61].
The feedforward inhibitory circuit of principal cells
The tight coupling to the excitatory input explains why the feedforward inhibition is strongest in the receptive field centre. Since the inhibitory field is about twice as large as the excitatory centre, also the nearest neighbours in the retinal mosaic would contribute to the inhibition. Only a small number of cells would be engaged as judged from the ganglion cell packing in the retina. Here, X cells of the same receptive field type are arranged in a somewhat noisy hexagonal pattern with the dendritic fields displaced by about a centre radius [41, 46]. Therefore, the feedforward inhibitory field could be formed by as little as seven retinal ganglion cells. Most likely, these neighbouring retinal ganglion cells would converge onto common intrageniculate interneurons, which in turn would distribute their inhibition to several principal cells with spatially overlapping but separate excitatory inputs. Again, most of these connections would occur at triads, thereby forming the anatomical substrate for nonlocked feedforward IPSPs. Unlocked IPSPs could also be mediated by locally ramifying axons from the interneurons [59, 71].
This scheme is consistent with the finding that only 20–25 % of the neurons in the cat´s dLGN are GABAergic interneurons . It also agrees with the variation in amplitude of locked and nonlocked feedforward IPSPs, whether evoked by visual stimulation or by spontaneous ganglion cell activity. We have assumed that this variability is caused by a variation in the number of activated intrageniculate interneurons, each discrete step representing the inhibitory input of one spike in one interneuron. The number of discrete steps may thus give an estimate of the number of converging interneurons (4–6). This interpretation is supported by the finding of a similar stepwise gradation of feedforward IPSPs evoked by electrical stimulation of the optic nerve . In this case, there was a threshold difference between steps proving that axons of different retinal ganglion cells contributed excitation to the pool of interneurons.
The alternative interpretation that the stepwise change in feedforward IPSPs represent independent local release from single triads [14, 49] seems less likely. Intrageniculate interneurons of both X and Y types fire some spikes in response to excitation from ganglion cells and corticogeniculate neurons [33, 41, 43, 45, 71]. The delay and time course of the feedforward IPSPs in principal cells and the occurrence of spatial facilitation at interneuronal level indicates that the IPSPs are spike evoked . Further, principal cells receive excitation from a single retinal ganglion cell at a large number of synaptic sites . Why should the feedforward inhibitory input at these sites operate independently? A somewhat disappointing implication is that the synaptic triads of the dLGN may be trivial from a functional point of view. The structure may simply have emerged as the most economical way of constructing a type specific feedforward inhibitory network with a limited number of interneurons.
The recurrent inhibitory circuit of principal cells
Recurrent IPSPs were evoked by light on and off stimuli in all parts of the receptive fields, similarly for X and Y cells. The latency of these inhibitory responses was the same and consistently shorter than for EPSPs evoked by surround stimulation of ganglion cells. It follows that the recurrent inhibitory effects were mainly due to centre stimulation of both on- and off-centre ganglion cells. This finding agrees with the observations that the excitation of individual perigeniculate neurons originates from both on- and off-centre principal cells with overlapping receptive fields [4, 70]. For clarity, the diagram in Fig. 11 is oversimplified with respect to the recurrent circuit. The mutual inhibitory connections between perigeniculate neurons or their projection to intrageniculate interneurons are not shown [6, 71], nor is the full spatial extent of their connections.
The convergence number for principal cells onto single perigeniculate cells is not known but might be as high as 20–40. This estimate comes from their binocular input from both on- and off-centre principal cells and response fields about twice the field centre of principal cells at the same eccentricity [4, 67]. The number of perigeniculate neurons converging onto a single principal cell is neither known, but it may be about the same as for the feedforward pathway. Such a convergence would account for the bell-shaped recurrent inhibitory receptive fields about three to four times larger than the EPSP fields of the principal cells, as found here.
These considerations agree with the small number of perigeniculate neurons that supply recurrent inhibition to the entire dLGN. In fact, the connectivity seem to be at the margin for a complete coverage of the receptive fields. Thus, we frequently observed asymmetrical on and off inhibitory responses in parts of the recurrent inhibitory field. Such asymmetries may explain some of the orientation tuning observed for geniculate cells . Whatever, the recurrent inhibitory fields of principal cells are much smaller than generally assumed for this pathway [49, 51] suggesting that it is inappropriate to refer to the recurrent inhibitory system in terms of “diffuse inhibition.” Even if our measure is somewhat underestimated due to the low excitability of perigeniculate cells in barbiturate anaesthesia, the size of the inhibitory field agrees rather well with the termination arbor of perigeniculate axons in the dLGN . Lateral spread of recurrent inhibition may also be limited by the mutual inhibitory connections of perigeniculate cells . Such a mechanism could explain why “long-range inhibition” appears in the dLGN after acute local destruction of the retina .
Comparison with other studies
Several procedures have been used in the past to study the organisation and function of the dLGN inhibition: (1) comparison of the firing responses of input retinal ganglion cell and target dLGN cell to different visual stimuli [27, 31, 37, 58, etc.]; (2) cross-correlation between impulse activity of two geniculate neurons , retinal and geniculate neurons [39, 41] or mutual recordings from dLGN and PGN cells ; (3) comparison of geniculate cell responses before and after iontophoretic blockade of GABA receptors by specific antagonist [8, 19, 50, 61]; (4) quasi-intracellular or intracellular recordings of visually evoked responses [44, 52, 53, 63].
All but the latter studies have relied on indirect methods to identify inhibitory effects in the dLGN. This fact may explain why almost every conceivable model has been proposed to account for the increased peripheral antagonism of principal cells . All models contain some components in common with the inhibitory circuitry as revealed here. For instance, Levick et al.  were correct with respect to the organisation of recurrent inhibitory pathway, and Singer et al.  in identifying “synergistic” inhibition from the receptive field centre. Likewise our scheme for feedforward inhibition is identical to that proposed by Mastronarde [39, 41] for lagged X cells, the difference being that we find a similar circuitry also for non-lagged X cells and for Y cells. The push–pull hypothesis proposed by Wang et al.  was based on an indirect procedure to estimate inhibition rather than direct observations of IPSP events. Our results suggest that their “pull” effect from the receptive centre was mainly due to disfacilitation, possibly primed by some recurrent inhibition. So the main merit of the present study is that the correct circuits could be sorted out among a list of proposed alternatives. In addition, we were able to show that the inhibitory fields were bell-shaped with the strongest effect in the excitatory field centre, as originally proposed for the surround mechanism of retinal ganglion cells by Rodieck and Stone .
Although this study relied on visual stimulation, its main goal was to analyse the inhibitory circuits of the dLGN rather than their function in vision. Our intracellular recordings were essentially qualitative and quantitative differences may exist between the inhibition of X and Y cells or lagged and non-lagged cells. Such differences may affect their response characteristics even if their basic circuits are the same. The anaesthesia may also influence the relative importance of the feedforward and recurrent inhibitory systems in the control of principal cell activity, especially since the excitability of perigeniculate cells is much depressed in barbiturate anaesthesia [4, 42]. Finally, we used step-modulated light stimuli in order to obtain optimal time separation between feedforward and recurrent IPSPs. Such stimuli introduce artificial transients in the excitatory and inhibitory inputs to the principal cells .
Some functional implications can be derived from our results, however. Both the feedforward and recurrent inhibitory fields were larger than the excitatory field of the principal cells so both mechanisms could conceivably contribute to the increased peripheral antagonism of the centre response . For X cells, however, it is evident that only the feedforward inhibitory pathway was operative with wide field stimulation. Such stimuli produced enough feedforward inhibition to keep the principal cells below their firing thresholds and, accordingly, the perigeniculate neurons without excitation. So the recurrent inhibitory system could not account for the enhanced surround inhibition in this situation.
The feedforward system may have the same role for Y principal cells although they typically respond with a brief burst of spikes to wide field stimuli and receive short latency recurrent IPSP to such stimuli. However, the surround inhibition of their centre response is less effective than in X cells , which is contrary to what would be expected if additional inhibition was provided by the recurrent pathway. The role of the feedforward inhibitory system may thus be the same in X and Y principal cells, namely, to improve the detection of contrast borders  or, in other terms, to sharpen their spatial filtering in the low frequency range [8, 37, 38].
Why is such filtering performed in two stages, first at the retinal level and then in the dLGN? The explanation may be that biological restrains preclude the construction of filtering circuits without superimposition of excitatory and inhibitory fields. Such a superimposition inevitable involves a loss in absolute sensitivity of the excitatory centre mechanism. This would be true even if the majority of the feedforward IPSPs are time-locked to EPSPs with their peak amplitude well after the EPSP peak, as found here. By influencing the firing probability for subsequent EPSPs locked IPSPs may still have a substantial effect on the overall signal transfer through the dLGN . This sensitivity loss may be partly compensated for by an external inhibitory control of the intrageniculate interneurons. Some brain stem neurons with selective projection to the main dLGN layers may have such a role [2, 5]. A sensitivity increase at the expense of spatial resolution might be useful during vision at very low levels of illumination [11, 37, 65] or at low contrast as may happen in deep fog.
What is then the role of the recurrent inhibitory system? We have already proposed that it is a component in a variable gain regulator for the transmission of visual signals through the dLGN , the other component being the recurrent excitatory pathway through corticogeniculate cells in layer 6 of the primary visual cortex [21, 23, 35, 62]. These two complimentary recurrent pathways are suggested to be the executive links in an attention system composed of a neuronal attenuator (inhibition) and an amplifier (excitation). Central to this hypothesis is the idea that both recurrent systems can be independently controlled by extravisual neuronal systems [5, 9, 17, 36, 54]. The hypothesis is a concretisation of earlier ideas about the function of the dLGN [26, 51]. It is basically similar to the searchlight hypothesis of Crick , with the important difference that the corticogeniculate pathway, rather than the perigeniculate, is given the role of a positive amplifier. Our recent studies on behaving animals [66, 67, 68, 69] are in line with the attentional hypothesis and give an experimental support to the proposed function of the recurrent system.
The present study was supported by the Swedish Medical Research Council (Project No. 04767) and the Polish Ministry of Science and Higher Education (Project No. COST/127/2007).
Conflicts of interest
This article is distributed under the terms of the Creative Commons Attribution Noncommercial License which permits any noncommercial use, distribution, and reproduction in any medium, provided the original author(s) and source are credited.
- 19.Eysel UT, Pape H-C, van Schayck R (1986) Excitatory and differential disinhibitory actions of acetylcholine in the lateral geniculate nucleus of the cat. J Physiol (London) 370:233–254Google Scholar
- 21.Ferster D, Lindström S (1985) Synaptic excitation of neurones in area 17 of the cat by intracortical axon collaterals of cortico-geniculate cells. J Physiol 376:233–252Google Scholar
- 26.Hernandez-Peon R (1966) Physiological mechanisms in attention. In: Russel RW (ed) Frontiers in physiological psychology. Academic, New YorkGoogle Scholar
- 31.Levick WA, Cleland BG, Dubin MW (1972) Lateral geniculate neurons of the cat: retinal inputs and physiology. Invest Ophtalmol 11:302–311Google Scholar
- 44.McIlwain JT, Creutzfeldt O (1967) Microelectrode study of synaptic excitation and inhibition in the lateral geniculate nucleus of the cat. J Neurophysiol 30:1–21Google Scholar
- 47.Rodieck RW, Stone J (1965) Analysis of receptive fields of cat retinal ganglion cells. J Neurophysiol 28:833–849Google Scholar
- 58.Uglesich R, Casti A, Hayot F, Kaplan E (2009) Stimulus size dependence of information transfer from retina to thalamus. Front Systems Neurosci 3, Article 10Google Scholar
- 65.Wróbel A (1981) Light level induced reorganization of cat’s lateral geniculate nucleus receptive fields. Spatiotemporal study. Acta Neurobiol Exp (Wars) 41:447–466Google Scholar