The synaptic mechanisms underlying the generation of orientation and direction selectivity in layer 4 of the primary visual cortex are still largely unclear. Previous in vivo work has shown that intra-cortical inhibition plays a major role in generating the properties of orientation and direction selectivity. Excitatory and inhibitory cortical neurons differ in their receptive field properties: excitatory neurons tend to be orientation- and direction-selective, inhibitory neurons tend to be orientation-, but not direction-selective. Here we have compared the relationship between direction preference maps recorded in vivo and synaptic input maps recorded in vitro from excitatory and inhibitory stellate cells in layer 4 of ferret visual cortex. Our goal was to test whether the differences in direction tuning between these cell populations might result from different inhibitory connectivity patterns. We found that excitatory neurons, which are direction tuned in vivo, receive ∼50% of their inhibitory inputs from cortical regions of opposite direction preference whereas inhibitory cells, which are not or poorly direction tuned, receive only very few inputs from regions of opposite direction preference. This confirms that inhibitory connections arising in cortical regions of opposite direction preference may be required to create or strengthen direction tuning in their target neurons. Thus, differences in intracortical inhibitory circuit patterns may underlie the differences in receptive field properties observed between excitatory and inhibitory neurons in vivo.
Layer 4 of the primary visual cortex is the major target of afferent fibers arriving from the thalamus. Simple cells in layer 4 are also the first neurons showing orientation and direction selectivity. The synaptic mechanisms involved in creating orientation and direction tuning in layer 4 are still largely unclear. An analysis of the intrinsic circuitry of layer 4 is therefore crucial to understanding the very first stages of intracortical synaptic processing involved in generating the specific reponse properties of neurons in the primary visual cortex.
Especially the roles of the thalamic input versus intracortical synaptic processing are still controversial. The majority of synaptic inputs to layer 4 neurons originate within the cortex (Ahmed et al., 1994; Stradford et al., 1996), indicating that the intracortical circuitry might play a prominent role in setting up functional response properties. In addition, previous in vivo studies using iontophoretic application of the GABAA receptor antagonist bicuculline indicate a fundamental role for intra-cortical inhibition in creating orientation and direction tuning (Sillito, 1975, 1977; Tsumoto et al., 1979; Sillito et al., 1980; Sato et al., 1995; Crook et al., 1996, 1997, 1998; Murthy and Humphrey, 1999).
However, other studies raise the possibility that the thalamic input may be sufficient to generate higher order response properties in layer 4 of the primary visual cortex (Ferster et al., 1996; Chung and Ferster, 1998; Gillespie et al., 2001). These findings appear difficult to reconcile at present. However, receptive field properties of layer 4 neurons most likely emerge from a combined mechanism involving both the spatial alignment and temporal sequence of activation of thalamic afferents on their cortical target neurons and intracortical circuits, i.e. the thalamic input may confer some directional selectivity which is amplified by intracortical mechanisms.
Despite the strong evidence for intracortical inhibitory mechanisms playing a role in creating direction tuning, the precise mechanisms involved are not clear. Direction selectivity has been linked to the spatio-temporal structure of simple cell receptive fields (Murthy and Humphrey, 1999). It has been demonstrated for layer 4 neurons that the spatio-temporal structure of their receptive fields confers a gradient of response timing which results in a preferred direction of motion across the receptive field. Direction selectivity is then created by linear summation of responses across the receptive field followed by amplification. Based on these findings two potential roles of inhibitory inputs are feasible: (i) intracortical inhibition creates or shapes the spatio-temporal orientation of simple cell receptive fields or (ii) intracortical inhibition merely suppresses weak responses (thresholding effect), e.g. by lowering the membrane potential relative to the spike threshold. Recent work (Murthy and Humphrey, 1999) indicates that inhibition primarily shapes the spatio-temporal receptive field structure, i.e. it sets up a gradient of response timing.
This mechanism appears to require a specific spatial alignment of receptive fields in inhibitory neurons and their target cells and/or a specific temporal order of activation of thalamic or intracortical excitatory synapses and intracortical inhibitory synapses. Does it also require a particular spatial organization as for the location of inhibitory cells relative to iso-orientation and iso-direction domains? Studies performing selective inactivation of functionally connected cortical sites (Crook et al., 1996, 1997, 1998) demonstrate that cross-orientation inhibition and iso-orientation inhibition between cells with opposite direction preference sharpen orientation and direction tuning by suppressing responses to non-optimal orientations and directions. This process is supplemented by an amplification of responses to optimal stimuli by iso-orientation/ iso-direction excitatory connections and a regulation of the amplification process by iso-orientation inhibition. Thus we expect inhibitory connections between cortical regions of opposite direction preference. If these inputs arising in cortical domains of opposite direction preference are responsible for direction tuning in their target cells, we expect them predominantly in neurons with sharp tuning, i.e. excitatory rather than inhibitory neurons. The inhibitory inputs in cat cortex appear largely provided by lateral projections of large basket neurons (Kisvarday et al., 1993, 2000; Crook et al., 1998; Buzas et al., 2001). However, locally anisotropic projections of clutch cells may also play a role (Budd and Kisvarday, 2001). Especially in view of the spatial organization of orientation and direction preference maps, i.e. orientation and direction preference are mapped orthogonally and iso-orientation domains tend to be subdivided into direction domains preferring opposite directions (Shmuel and Grinvald, 1996; Weliky et al., 1995), very local interconnections between regions of opposite direction preference are feasible.
Our previous data, obtained in combined in vivo optical imaging and in vitro synaptic mapping experiments designed to physiologically characterize the functional topography of inhibitory connection that may underlie direction tuning, indicate that different inhibitory mechanisms may account for orientation-and direction-selectivity (Roerig and Katz, 1998; Roerig and Kao, 1999; Roerig and Chen, 2002). Whereas broadly tuned inhibitory inputs are sufficient for orientation tuning in upper layer pyramidal neurons (Roerig and Katz, 1998), specific lateral inhibitory inputs originating in cortical regions preferring the opposite direction of stimulus motion appear to be required for direction tuning in layer 2/3 pyramidal neurons (Crook et al., 1998; Roerig and Kao, 1999). However, layer 2/3 already represents the second synaptic processing stage in the cortex, i.e. the synaptic mechanisms operating in layer 4, where direction tuning is first observed, are still unclear.
In vivo, excitatory and inhibitory cells, including those in layer 4, differ in their receptive field properties: excitatory (regular spiking) neurons are orientation-and direction-selective whereas most inhibitory (fast spiking) neurons are orientation-, but poorly direction-tuned (Azouz et al., 1997; Gibber et al., 2001). Anatomically, excitatory neurons represent pyramidal cells and spiny stellate cells. A huge variety of anatomically and physiologically distinct types of GABAergic interneurons has been described in the visual cortex (Kisvarday, 1992; Gonchar and Burkhalter, 1997; Tamas et al., 1997, 1998).
The difference in direction tuning between excitatory and inhibitory neurons could be due to differences in intracortical inhibitory synaptic input patterns. Based on our previous findings that layer 2/3 pyramidal cells in ferret V1 receive a significant population of inhibitory inputs originating in cortical domains of opposite direction preference (Roerig and Kao, 1999), we have here tested the hypothesis that excitatory layer 4 neurons, which are direction selective, receive inhibitory synaptic inputs from domains preferring the opposite direction of stimulus motion, whereas inhibitory interneurons, which are orientation-, but not direction-selective, receive only broadly tuned inhibitory inputs. We have analyzed the relationship between orientation and direction preference maps and synaptic input maps in excitatory and inhibitory stellate cells in layer 4 of ferret visual cortex. We first optically imaged direction and orientation maps in vivo. Subsequently we recorded from individual excitatory and inhibitory layer 4 neurons in whole cell patch clamp configuration in tangential slices prepared from the imaged cortex. Synaptic inputs were evoked by local photolysis of caged glutamate — photostimulation (Katz and Dalva, 1994). Optical imaging and synaptic input maps were aligned and the orientation and direction tuning of excitatory and inhibitory inputs to individual neurons was analyzed.
Our results indicate clear differences in intracortical synaptic input patterns between excitatory and inhibitory layer 4 neurons in ferret V1: Excitatory neurons showed a bimodal distribution of inhibitory synaptic inputs with one population originating from iso-direction domains and one population from regions of opposite direction preference. In contrast, inhibitory inputs to inhibitory interneurons were broadly tuned. Generally speaking, this raises the possibility that differences in orientation and direction tuning between different cortical cell types might be based on different intracortical connectivity patterns. In addition, our results confirm previous studies suggesting that inhibitory inputs originating in regions of opposite direction tuning might be required for generating or sharpening direction tuning in cortical neurons.
Materials and Methods
Optical Imaging of Direction Preference Maps In Vivo
Ferrets [postnatal days 37 (P37)–P48; Marshall Farms, New Rose, NY] were anesthetized and prepared for in vivo optical imaging as previously described (Weliky et al., 1995, 1996). The cortex was illuminated with red light (707 nm). A 50 × 50 tandem lens combination and a Peltier-cooled slow-scan CCD camera (Optical Imaging Europe, Martinsried-Munich, Germany) was used for optical imaging. Visual stimulation was provided monocularly through the contralateral eye. Visual stimuli were presented at a distance of 30 cm. For mapping of direction preference, animals were presented with a dot pattern (1.5 × 1.5° dot size) moving in different directions at a velocity of 15°/s interleaved with blank screen presentations. The average dot density was 2.2 dots per 10 × 10° region. For imaging orientation domains, a high contrast bar grating pattern was used (1.2° bar width, 6° bar spacing, 18°/s drift). Sixteen directions and eight orientations were imaged per animal; stimuli were presented in a randomly interleaved manner. Single condition responses (averages of 120–180 trials) were divided by images acquired during blank screen presentations. The single condition images were vector summed to produce an angle map of direction or orientation preference (Bonhoeffer and Grinvald, 1993). Although a more appropriate method for creating direction preference maps based on the vector maximum instead of the vector sum has been published recently (Kisvarday et al., 2001), we feel confident in using the vector sum method for the following reasons: (i) we quite frequently found an orthogonal relationship between orientation and direction preference in our maps; (ii) the orientation and direction preference recorded electrically from single units corresponds well with the imaged orientation and direction preference maps (Shmuel and Grinvald, 1996; Roerig and Chen, 2002); and (iii) we found statistically significant differences between tuning of synaptic inputs in different cell populations repeatedly and in maps recorded from a number of different animals, which renders the possibility of a pure imaging artifact unlikely.
Intracortical injections of rhodamine conjugated latex microspheres (70–150 nl) were made to guide alignment of in vivo and in vitro maps. The imaged cortex was then removed and tangential slices (400 μm thickness) were prepared. The block of tissue removed from the animal was placed in chilled artificial cerebrospinal fluid (sucrose–ACSF, composition: 125 mM NaCl; 5 mM KCl; 5.3 mM KH2PO4; 1.3 mM MgSO4; 3.2 mM CaCl2; 10 mM dextrose; 25 mM NaHCO3) oxygenated with a mixture of 95% O2 and 5% CO2, pH 7.4). Usually the second slice of each imaged hemisphere was used for layer 4 recordings. Histological sections were stained with cresyl violet to verify that cell somata were small as an additional indicator for layer 4.
Whole cell patch clamp recordings were made at 33°C in a temperature controlled recording chamber mounted on the stage of an upright microscope (BX50WI; Olympus Optical, Tokyo, Japan). Fluorescent bead marks were viewed using epifluorescence and either a rhodamine or fluorescein filter set (exciter G 546 nm, beam splitter FT 580, barrier LP 590 for rhodamine; 450–490 nm excitation, FT 510 dichroic mirror, LP 520 barrier filter for fluorescein; Zeiss). Bead injections were visible in the living slices and could directly be used to guide positioning of patch pipettes. Slice overview images were taken using a SONY XC-75 CCD camera and a SNAPPY (Play) video frame acquisition module in conjunction with SNAPPY software. Electrophysiological recordings from single neurons were performed using standard whole cell patch clamp methods. The intracellular solution consists of 110 mM d-gluconic acid, 110 mM CsOH, 11 mM EGTA, 10 mM CsCl, 1 mM MgCl2, 1 mM CaCl2, 10 mM HEPES, 1.8 mM GTP, 3 mM ATP, pH 7.2 and contained 0.5% N-(2-aminoethyl)biotinamide (Neurobiotin; Molecular Probes, Eugene, OR).
To estimate the spatial resolution of our photostimulation approach neurons were recorded in whole-cell, current clamp mode and were photostimulated directly and up to 500 mm away from their cell bodies (see Results section).
Synaptic Mapping Experiments
Voltage clamp recordings are conducted using an Axopatch 1D amplifier (Axon Instruments, USA). The holding potential was either –60 or –20 mV to distinguish between excitatory and inhibitory synaptic inputs (Katz and Dalva, 1994). Recordings were filtered at 1 kHz and digitized at 8 kHz. Series resistances ranged from 11 to 17 MΩ, a 30–50% compensation was usually achieved using the amplifyer adjustments.
Presynaptic inputs were stimulated using the scanning laser photo-stimulation approach detailed in (Chung and Ferster, 1998). Slices were bathed in a 1 mM solution of Nmoc-caged glutamate (Rossi et al., 1997). An argon/krypton ion laser (Spectraphysics Stabilite 2017) was used as a UV light source. The laser beam was coupled into a 50 μm diameter fiber optic attached to a motorized X/Y-stage. The fiber was moved within an oil droplet below the recording chamber. The laser beam was manually focused into the middle of the slice preparation (Fig. 1). Opening of the external shutter, scanning of the laser beam and data acquisition were controlled by a National Instruments AD board (AT-MIO/AI E-10) and custom written software. (Labview, National Instruments). The flash duration was 5 ms and the interstimulus interval 3–5 s. The acquisition period was 1 s, with the shutter opening occurring after 500 ms. The prestimulus acquisition period served to monitor spontaneous synaptic activity. Photostimulation evoked responses were identified by their constant latency and shape. We have stimulated each site four times and only events that occurred at least three times and had a fairly constant latency and shape were included in the analysis. In addition, an average of the baseline window (i.e. the 500 ms pre-stimulation period) recorded in each trace is created by the analysis program and subtracted from the stimulation traces. Photostimulation evoked responses were analyzed within the first 100 ms following uncaging. The spacing of stimulation sites was 50 μm. Typical maps consist of 500–1000 stimulation sites, corresponding to an area of 1.5–2.5 mm2. Only one or two cells were recorded per slice to facilitate alignment and unambiguous assigning of input maps to postsynaptic cells. Maps were recorded at –20 mV to distinguish between inhibitory and excitatory inputs, at –60 mV to allow for a larger driving force for small excitatory events and in the presence of tetrodotoxin (TTX, 2 μM) to distinguish between direct activation and local synaptic inputs. Following the photostimulation mapping of synaptic inputs, brief current clamp recordings were done to characterize the neurons’ spike pattern. All cells subsequently identified as spiny stellate cells showed regular spiking characteristics (McCormick et al., 1985) and seven out of eight recorded aspiny stellate cells showed fast spiking characteristics.
Micropipettes were loaded with 1% biocytin. Following recording, slices were fixed in 4% paraformaldehyde in phosphate buffered saline (pH 7.4) for subsequent histological processing of neurobiotin filled cells. Slices were resectioned at 70 mm on a freezing microtome and processed for immunohistochemical visualization using a standard avidin–biotin (Elite) kit and DAB as the chromagen. Spiny (putative excitatory) and aspiny (putative inhibitory) stellate cells were distinguished by their morphology.
Alignment and Analysis
Alignment of orientation and direction maps and photostimulation maps was guided by the fluorescent bead injections. The in vivo images, the video image of the living slices and the histological sections were overlaid with the photostimulation (synaptic input) maps using the layer menu of Adobe Photoshop (Moutain View, CA). The position of the stained postsynaptic cell in the histological section and the direct activation area in the photostimulation maps served as additional markers. Linear scaling and rotation was applied to the images until the bead marks were at least 50% overlapping. The bead marks are 100–200 μm in diameter, which results in a maximum alignment error of 75 mm (Weliky et al., 1995). Synaptic input maps were then superimposed on the orientation/ direction maps. For each site giving rise to a synaptic input as well as for the location of the postsynaptic cell, the orientation/direction value was calculated as the mean of four pixels in the color coded maps (pixel sixe, 10 μm). The orientation/direction tuning difference between the location of the postsynaptic cell and the sites of origin of synaptic inputs was calculated for each EPSC and IPSC and the difference values were used to generate input tuning histograms.
The analysis of photostimulation evoked synaptic currents was done in the following way. We analyzed EPSCs and IPSCs that occurred during the first 100 ms following the laser flash. Our current clamp data show that all photostimulation evoked action potential in presynaptic neurons occur within this time frame. We omitted direct activation currents evoked that occurred mainly within 50 mm around the postsynaptic cell body. Direct activation currents were identified by their short latency and their persistence in the presence of TTX (1 mM). To assess spontaneous activity a 100 ms time window preceding the laser flash was analyzed in the same way as the stimulation window for each individual recorded trace. Spontaneous and evoked currents were detected and their number and amplitudes stored for each individual stimulation site using custom written LabView software. The detection of events was not automated. To control for variability of spontaneous activity and serious resistance between individual neurons we adopted a previously published method by (Dantzker and Callaway, 2000). A normalized synaptic input value was generated by subtracting the number (fs) × mean amplitude (as) of spontaneous events from the number (fp) × mean amplitude (ap) of photostimulation evoked events. To correct for access resistance variability among cells this value was divided by the mean amplitude of spontaneous synaptic events (as) for each trace. This analysis was done twice for each trace recorded at –20 mV, once for analysis of IPSCs and once for analysis of EPSCs. The results of this analysis provided the synaptic input values used for the construction of histograms in the lower panels of Figure 9.
To generate tuning histograms (Figs 6–8) we used the percentage of sites generating an evoked inhibitory or excitatory synaptic response in the postsynaptic neurons that fell into each tuning difference category. Here, each stimulation site was counted only once, independent of whether the evoked synaptic response consisted of a single event or a burst of events.
We have optically recorded orientation and direction maps from ferret primary visual cortex (n = 11; age, postnatal days 37–48). Subsequently the imaged brain region was removed and tangential slices (400 μm thickness) prepared. Whole cell patch clamp recordings from individual excitatory and inhibitory layer 4 neurons were done and synaptic inputs were scanned by local photolysis of Nmoc-caged glutamate (1 mM). Postsynaptic cells were filled with biocytin and histological sections were aligned with the synaptic input maps and the optical images obtained in vivo to determine the spatial distribution of presynaptic inputs.
Control for Spatial Resolution
To allow for a reliable interpretation of our mapping experiments, especially regarding the fact that synaptic input maps are aligned with the optical imaging maps obtained in vivo, we have to assure that stimulation of presynaptic sites is spatially confined to one or a few neurons. To estimate the spatial resolution of our photostimulation approach, neurons were recorded in whole-cell, current-clamp mode and were photostimulated directly and up to 500 μm away from their cell bodies. Photostimulation evoked multiple action potentials in both pyramidal and stellate neurons (n = 9; mean number of action potentials per stimulation site, 2.3 ± 0.68). All neurons tested fired 90% of their action potentials when stimulated at locations in a radius of 80 μm of their somata. The average was 67 ± 26 mm from the soma. These results indicate that the location of neurons providing presynaptic input to a recorded cell can be determined with an accuracy of <50 mm. All tested cells fired action potentials only when stimulated directly, i.e. no polysynaptic activation of more distant sites occurred.
We recorded from n = 12 spiny (putative excitatory) and n = 8 aspiny (putative inhibitory) stellate cells. The total number of excitatory inputs analyzed was 356 for aspiny and 321 for spiny stellate cells. The total number of inhibitory inputs was 234 for aspiny stellate cells and 378 for spiny stellate cells. These numbers refer to the number of sites generating an evoked excitatory or inhibitory synaptic response.
Evoked EPSC amplitudes ranged from 10 to 127 pA for aspiny stellate cells and from 13 to 78 pA in spiny stellates. Evoked IPSC amplitudes ranged from 8 to 85 pA in aspiny stellates and from 12 to 106 pA in spiny stellate cells.
The synaptic input values used for tuning analysis were corrected for spontaneous synaptic activity (see Materials and Methods section). The mean frequencies of spontaneous EPSCs were 4.5 ± 3.1 Hz (mean ± SD) in spiny stellate cells (n = 12) and 5.6 ± 2.8 Hz in aspiny stellate cells (n = 8). The mean spontaneous EPSC amplitudes were 28.7 ± 14.3 pA in spiny stellates and 36.1 ± 26.4 pA in aspiny stellate cells. The mean frequencies of spontaneous IPSCs were 7.8 ± 3.8 HZ (mean ± SD) in spiny stellate cells (n = 12) and 9.1 ± 4.8 Hz in aspiny stellate cells (n = 8). The mean spontaneous IPSC amplitudes were 35.9 ± 22.1 pA in spiny stellates and 42.7 ± 18.3 pA in aspiny stellate cells. Differences between frequencies and amplitudes of spontaneous synaptic currents were not statistically significant between excitatory and inhibitory stellate cells (P < 0.8, Students’s t-test).
Latency Analysis of Direct Activation and Synaptically Evoked Currents
To distinguish between direct activation and synaptically evoked currents we employed two different methods: (i) we recorded maps under control conditions and in the presence of tetrodotoxin (TTX) and (ii) we compared the latency distributions of direct activation and synaptic currents (Figs 2–4).
Mapping in the presence of TTX was done in n = 5 spiny and n = 3 aspiny neurons (Fig. 2). Excitatory and inhibitory synaptic currents were completely blocked in the presence of TTX and the area of direct activation was confined to the soma and proximal dendrites of the recorded neurons. The correspondence of the area of direct activation with the anatomical extent of the recorded cell was verified by overlaying the direct activation maps with a histological sections containing the biocytin filled neurons (Fig. 2).
The direct activation current had a considerably shorter latency (of the order of 2–6 ms in the majority of tested neurons) compared to synaptically evoked currents (10–60 ms latency range). To estimate the variability in the latencies of both direct and excitatory and inhibitory synaptic currents the same site was stimulated up to 30 times and the latency plotted as a function of the flash number. Figures 3 and 4 show example recordings and plots for a spiny stellate cell (Fig. 3) and an aspiny stellate cell (Fig. 4). Although the latencies of both direct and synaptic currents showed variability, the latencies recorded for individual events were fairly constant and events evoked from different sites in the same neuron could clearly be distinguished based on their latency and time course. The latency distributions of synaptically evoked currents matched the latency distribution of presynaptic action potentials.
Spatial Distribution and Tuning of Synaptic Events
The number of stimulation sites recorded per cell ranged from 870 to 1200, i.e. the size of the mapped area was similar for most neurons. Figure 5 shows examples of input patterns recorded from a spiny (A) and an aspiny (B) neuron superimposed on the direction/orientation preference maps. The size of the mapped area is indicated by the white line. The number of sites generating an input ranged from 48 to 87 in spiny cells and from 56 to 116 in aspiny neurons. Since the differences between individual cells were not very large within each cell group, data from each group were pooled without weighting to generate the histograms in Figures 6–9.
The majority (83%) of both excitatory and inhibitory inputs to layer 4 stellate cells originated from within 500 mm distance around the postsynaptic cell body. Long-range projections were rare in layer 4; however, occasionally both EPSCs and IPSCs could be evoked from distances >800 mm. There was no significant difference in the distance range from which the two cell classes received excitatory and inhibitory inputs.
Most (68%) of excitatory inputs to both spiny (putative excitatory) and aspiny (putative inhibitory) stellate cells originated from cortical regions preferring the same orientation and direction as the postsynaptic cell (Figs 6 and 7). However, the inhibitory input patterns were significantly different for the two cell populations: excitatory layer 4 cells received two populations of inhibitory inputs, ∼56% originated in iso-direction domains whereas the remaining inputs originated in cortical regions preferring the opposite direction of stimulus motion (Figs 7 and 8). Inhibitory layer 4 neurons only received a small number of inhibitory synaptic inputs from cortical regions of opposite direction preference (Figs 7 and 8).
The distributions of EPSCs and IPSCs have been compared for n = 12 spiny neurons and n = 8 aspiny neurons. The percentages of EPSCs and IPSCs falling into each direction tuning difference category (0–20 to 160–180°) were calculated for each individual neuron. The percentages of events in each category were statistically compared (Mann–Whitney rank sum test). In spiny stellate cells the percentages of EPSCs and IPSCs were significantly different in each bin (P < 0.001; Fig. 8). In aspiny stellates, however, EPSC and IPSC distribution were not significantly different (P > 0.05, Fig. 8), indicating that a subpopulation of synaptic inputs originating in cortical domains tuned to the opposite direction are confined to excitatory neurons. We have also compared the average IPSC tuning histograms for the 12 spiny stellates and the eight aspiny stellates (Fig. 8C), the distributions were significantly different (P < 0.001) for the direction tuning difference range of 60–180°. We also performed a statistical analysis based on the NEI values. This also revealed a significant difference among IPSC distribution in spiny and aspiny stellate cells (Fig. 10F).
The efficacy of a synaptic input related to its impact on receptive field properties is determined not only by its specificity but also by its amplitude. To determine whether iso-orientation/direction tuned inputs are also the strongest ones we have plotted the average amplitude of EPSCs and IPSCs recorded from each neuron against the orientation/direction tuning difference between pre-and postsynaptic cells. We statistically compared the event amplitudes in the 0–20° bin, i.e. the events showing the strongest iso-direction tuning, with all the other categories using one-way ANOVA. Summary histograms are shown in Figure 9. The median values for the 160–180, 120–140 and 100–120 bins were smaller than the median of the 0–20° bin. In most neurons the iso-orientation/direction tuned inputs were also the strongest ones; however, the average amplitudes of IPSCs originating in cortical sites tuned to the opposite direction were of equal strength as compared to the IPSCs originating in iso-direction domains.
To compensate for differences in spontaneous activity rates and series resistance between individual neurons we have also created a normalized input (see Materials and Methods), which is plotted in Figure 10. Although in most cases the normalized input showed a similar variation over the orientation-/direction-tuning difference range between stimulation and recording sites as the mean amplitude, it showed a stronger tendency for iso-direction tuned IPSCs being most efficient in aspiny stellate cells and it indicates a bimodal distribution (iso-and opposite direction inputs being strongest) in spiny stellate cells.
A number of in vivo studies have shown that blocking intra-cortical GABAA receptor mediated inhibition reduces direction selectivity (Sillito, 1975, 1977, 1980; Tsumoto et al., 1979; Sato et al., 1995; Murthy and Humphrey, 1999), thus indicating that intracortical inhibition plays a major role in generating this receptive field property. The synaptic mechanisms involved, however, are still controversial and may be different in layer 4 — where direction tuning is first created — and other cortical layers (Murthy and Humphrey, 1999). Inhibition also appears to play a larger role in creating simple cell direction tuning than creating complex cell direction selectivity (Goodwin and Henry, 1975; Eysel et al., 1988; Marlin et al., 1988; Sato et al., 1995; Murthy and Humphrey, 1999).
Even among simple cells the effects of blocking inhibition can range from small reductions to a complete loss of direction tuning (Tsumoto et al., 1979; Sato et al., 1995; Murthy and Humphrey, 1999). Since GABAB receptor mediated inhibition does not seem to play a significant role in generating direction tuning (Baumfalk and Albus, 1988), the residual direction selectivity observed following blockade of GABAA receptor mediated inhibition most likely reflects excitatory mechanisms, either at the level of the thalamocortical synapse or mediated by intracortical circuits.
However, the experimental evidence for a role of intracortical inhibition in generating direction tuning is strong and a number of different mechanisms have been proposed. How does the intrinsic circuitry of layer 4 and the functional topography of synaptic connections support a specific role for intracortical inhibitory circuits in establishing direction selectivity? A fraction of inhibitory interneurons, mostly basket cells, receive direct thalamic input (Freund et al., 1985; Ahmed et al., 1997). Intracortically, these neurons receive excitatory synaptic input from spiny stellate and pyramidal neurons (Tarczy-Hornoch et al., 1997). Basket cells, the major population of GABAergic neurons involved in lateral inhibition (Crook et al., 1996, 1997), synapse onto both spiny (excitatory) and smooth (inhibitory) stellate cells in layer 4. In addition, they form inhibitory connections with other basket cells (Ahmed et al., 1997).
How specific are inhibitory interactions as far as the functional topography of the visual cortex is concerned, i.e. do inhibitory inputs to a given target cell preferentially originate in columns preferring the same or opposite directions or are they nonspecific? Answering this question may provide important clues as for the mechanisms underlying the creation of direction selective receptive fields in layer 4 of V1. Selective inhibition originating at cortical sites of opposite direction preference suppressing responses to the non-preferred direction have been reported in cat (Creutzfeld et al., 1974; Eysel et al., 1988; Crook et al., 1996, 1997) and monkey (Livingstone, 1998) primary visual cortex. Our previous study on layer 2/3 pyramidal neurons (Roerig and Kao, 1999), as well as our recordings from excitatory layer 4 stellate cells reported in this study, are in line with these observations. We find a significant proportion of inhibitory inputs originating in regions tuned to the non-preferred direction in both cell populations. Other studies indicate that that both excitatory and inhibitory visually evoked potentials are tuned to the preferred direction (Douglas et al., 1991; Jagadeesh et al., 1997). To some extent our results also support those findings since the majority of intracortical EPSCs and a considerable fraction of IPSCs were iso-orientation tuned in all cell types studied. However, in excitatory neurons, i.e. pyramidal and spiny stellate cells, we find a bimodal distribution of IPSCs, one population of inhibitory inputs originates in iso-direction domains, whereas the other population preferentially originates in regions preferring the opposite direction. This may indicate that the mechanism underlying direction tuning in these neurons involves iso-direction tuned inhibitory inputs which threshold out excitatory inputs from the non-preferred direction and, in addition, inhibitory inputs from domains of opposite direction preference may provide further refinement.
Cortical neurons differ in their receptive field properties. These different functional properties are likely to be due to differences in synaptic inputs, both from intrinsic and extrinsic sources. In vitro studies show that different populations of cortical neurons receive synaptic inputs of different strength and different spatial organization (Dantzker and Callaway, 2000). However, so far it has not been investigated whether differences in receptive field properties correlate with different intracortical connectivity patterns. Although some layer 4 basket cells show a direction bias in vivo (Ahmed et al., 1997) in both ferret (Gibber et al., 2001) and cat (Azouz et al., 1997), excitatory neurons tend to be both orientation- and direction-tuned, whereas the majority of inhibitory neurons tend to be orientation- but not direction-tuned.
Can we correlate these differences in response properties with differences in the spatial pattern of intrinsic inputs as related to the functional organization of direction preference? In our previous study (Roerig and Kao, 1999) we showed that layer 2/3 pyramidal neurons in ferret primary visual cortex receive inhibitory synaptic inputs which originate in cortical regions of opposite direction preference. This is in line with in vivo studies showing specific inhibitory inputs from cortical sites tuned to the opposite direction suppressing responses to the non-preferred direction (Creuzfeld et al., 1974; Eysel et al., 1988; Crook et al., 1996, 1997, 1998; Livingstone, 1998).
Here, we investigated whether the difference in direction tuning between excitatory and inhibitory layer 4 neurons in the primary visual cortex of the ferret corresponds to differences in the specificity and spatial organization of inhibitory inputs to the two neuron populations. Using a combination of in vivo optical imaging and in vitro synaptic mapping, we compared the orientation- and direction-tuning of excitatory and inhibitory inputs to spiny (putative excitatory) and aspiny (putative inhibitory) layer 4 stellate cells.
The orientation tuning of both excitatory and inhibitory inputs to both cell classes was similar: EPSCs and IPSCs were largely iso-orientation tuned. The tuning of inputs to aspiny (putative inhibitory) stellates tended to be broader compared to the tuning of inputs to spiny (putative excitatory) stellates. By contrast, the two cell classes differed considerably in the direction tuning of their inputs. Whereas excitatory inputs were iso-direction tuned in both cell types, the pattern of inhibitory inputs was significantly different. Excitatory stellates received a significant number of inhibitory inputs originating from cortical regions tuned to the opposite direction. This is in line with previous studies demonstrating a role for iso-orientation, but opposite direction tuned inhibitory projections in sharpening direction selectivity (Crook et al., 1996, 1997, 1998). By contrast, inhibitory layer 4 neurons only received a small number of inhibitory inputs of opposite direction preference. Inhibitory inputs to aspiny stellates tended to be iso- or broadly direction-tuned. This confirms that specific inhibitory connections originating in regions tuned to the opposite direction are important for direction tuning of cortical neurons and that differences in response properties in different populations of cortical neurons might be explained by their different intra-cortical connectivity patterns.
So far our analysis is confined to smooth stellate cells, which represent only one population of inhibitory neurons. Future studies will include different population, especially basket neurons which have been strongly implicated in direction tuning of their target neurons in the upper layers of cat visual cortex (Crook et al., 1996, 1997, 1998).
Previous and our own studies seem to establish: (i) that GABAA receptor mediated inhibition contributes to direction tuning and (ii) that the majority of GABAergic neurons do not show much direction selectivity themselves. An obvious question, of course, is how can a population of cells which is not or only weekly direction tuned sharpen or even generate this property in excitatory neurons.
Murthy and Humphrey (Murthy and Humphrey, 1999) have shown that in layer 4 simple cells block of intracortical inhibition affects the spatio-temporal structure of receptive fields and direction selectivity, suggesting that direction tuning is created by gradients of response timing. There is no obvious need for inhibitory cells to be direction tuned if their role is to establish a spatio-temporal pattern in the excitatory response. However, the spatial distribution and time of activation of inhibitory inputs seems critical. Inhibitory cells are to some extend directly activated by thalamocortical afferents; their main excitatory input, however, is provided by intracortical excitatory synapses (Ahmed et al., 1997; Tarczy-Hornoch et al., 1998). The majority of inhibitory connections themselves are very local; a fraction of inhibitory axons, however, projects over a range of several millimeters (Martin et al., 1983). It is therefore feasible that inhibitory neurons are locally activated by excitatory neurons or thalamic afferents and then exert their inhibitory action via projections to sites of opposite direction preference. This effect can be mediated by locally projecting neurons (e.g. stellate cells) if the domain of opposite direction preference is located in close vicinity, e.g. within same iso-orientation domain, or by long-range projecting inhibitory neurons (e.g. large basket cells) if the target neuron is located in a remote domain. Alternatively, locally projecting interneurons could provide the iso-direction inhibition whereas the inputs originating in sites of opposite direction preference may be supplied by long-range projecting neurons. Our data on both layer 4 and layer 2/3 neurons (Roerig and Kao, 1999) support the first hypothesis since we find inhibitory inputs tuned to both iso-directions and opposite directions over the entire distance range we have investigated. However, we cannot at present exclude the possibility that the two populations of inputs are generated by two or more different types of GABAergic interneurons since we do not know the anatomical nature of the stimulated presynaptic neurons.
Our results and those of Crook et al. (Crook et al., 1996, 1997, 1998) suggest a specific role for inhibitory connections originating in cortical domains of opposite direction preference in establishing tuning sharpness in excitatory target cells. Our results reported here suggest that the comparatively poor direction tuning of inhibitory cells may be explained by a lack of these inputs. So far our analysis is confined to one subpopulation of GABAergic interneurons: the spiny stellate cell of layer 4. It will be important to investigate whether a similar connectivity pattern is found in other interneuron types, e.g. large basket cells, to analyze whether there are differences in the degree of direction tuning or direction bias found in different interneuron types and to correlate these findings with an analysis of projection patterns of different interneuron classes to establish their different roles in shaping visual response properties.
This work was supported by EY 12702 (B.R.) and GM56481 (J.P.Y.K.).