Electron microscopy was used in macaque monkey cortical area V1 to investigate what factors might determine the proportion of somatic membrane covered by inhibitory type 2 synapses. In a sample of 4654 excitatory neurons, synapse cover did not correlate consistently with cell variety (pyramid or spiny stellate), soma size, synaptic apposition length or thalamic input. There were significant differences in somatic synapse cover per layer, but the pattern of differences in cover among layers differed significantly between animals, suggesting that laminar environment alone is not a generally applicable determinant of amount of inhibitory synapse cover. The pattern of cover for cells in different layers was, however, similar between the two hemispheres of an individual monkey. Measures of inhibitory synapse cover on four sets of pyramidal neurons in layers 5 and 6, each with different efferent projection targets, showed that the sets differed significantly from other cells in their respective layers, and differed significantly from each other. These findings demonstrate that there is unique circuitry for different subsystems within single layers of cortex and provide a rationale for the rich variety of cortical GABAergic interneurons within single layers.
Somal inhibition is likely to be a powerful regulator of the ability of cortical spine bearing neurons to respond to excitatory inputs; the inhibitory synapses derive from particular sets of local circuit neurons, including a variety of basket neurons (Jones and Hendry, 1984; Lund and Yoshioka, 1991; Lund and Wu, 1997) and, in layer 4, a particular class of interneuron, the clewed cell or layer 4C-1 cell (Valverde, 1971; Lund, 1987). Using macaque monkey visual cortex, area V1, as a model system, this study addresses the following questions:
Is there a constant common pattern to spiny neuron somatic inhibitory synapse cover in different layers of V1?
Is the amount of synapse cover related to factors such as size of the neuron, possession of thalamic inputs or morphology (pyramidal or spiny stellate)?
Is synapse cover determined on a more individual level? If a set of cells is identified as likely to have a common function (e.g. projecting to a particular external destination), do they have a unique weight of somal inhibition?
This last possibility was suggested for the visual cortex of the cat (Farinas and DeFelipe, 1991a,b), where labelled callosally projecting neurons were found to have a significantly larger number of somal synaptic contacts than labelled cortico-thalamic projecting cells; however, these two cell populations differed in mean size, were located in different laminae, and each population was reported to have the same mean density of synapses as other, unlabeled cells in the same layers. This leaves unanswered the question of whether the difference in synapse number was really due to the neurons being of particular functional kind, or to one or more of the other factors mentioned above.
By trying to answer the apparently simple questions posed above, we hoped to provide data that would assist in modelling particular circuits in the cortex. The modeller needs to know whether the weight of inhibition can be predicted by some simple morphological scaling factor — or laminar environment — or if unique inhibitory cover can be assumed for each excitatory cell of particular function.
Of all cortical regions, most is known about intrinsic circuitry and physiological properties in primary visual cortex of the macaque monkey. It has clearly defined laminar boundaries; these layers define zones where particular efferent cell populations are located, and thalamic afferents of different kind terminate. Cells of different layers differ significantly in mean size; layers also differ in kind of excitatory cell — some contain spiny stellate neurons, others contain only pyramidal neurons (Levitt et al., 1996). These clear-cut differences within V1 provide a useful setting to try to answer the questions we posed. While both inhibitory and excitatory synaptic cover on single cells certainly change in V1 during postnatal maturation (Boothe et al., 1979; Mates and Lund, 1983), synapse cover appears to reach a stable figure by the time the animal is sexually mature, as in this study (Bourgeois and Rakic, 1993).
Materials and Methods
Animals, Injection Procedures and Histology for Light Microscopy (LM) and Electron Microscopy (EM)
Two normal adult macaque monkeys (no. 1205 — a female Macaca fascicularis, wt 4.5 kg; and no. 146 — a female M. mulatta, wt 8.4 kg) were used for these studies. All experimental procedures were approved by the British Home Office and were carried out in accordance with the guidelines published in the NIH ‘Guide for Care and Use of Laboratory Animals’ (NIH publication no. 86-23, revised 1987) with adequate measures taken to minimize pain or discomfort. Monkey 146 was placed under deep anaesthesia and large pressure injections of wheat germ agglutinin–horseradish peroxidase (WGA–HRP) (15% in distilled water; 0.3 μl total in each hemisphere) were made via micropipette into the lateral geniculate nucleus of the right hemisphere and into the pulvinar nucleus of the left hemisphere (each region was located by stereotaxic coordinates, followed by microelectrode recordings). Animal 146 was allowed to recover from anaesthesia, and survived 2 days before terminal anaesthesia and perfusion with saline and sodium nitrite, followed by 2% paraformaldehyde plus 2% gluteraldehyde in phosphate buffer, pH 7.2. The fixative was allowed to remain in the brain for 1 h, then rinsed out by perfusion with 0.1 M phosphate buffer (pH 7.4). Monkey 1205, without tracer injections, was deeply anaesthetized and perfused with saline followed by 4% paraformadehyde, 0.5% gluteraldehyde in phosphate buffer. The brains were removed, and small blocks from the visual cortex, area V1, of each hemisphere of the animal (1205) without tracer injections were osmicated and embedded for EM.
The visual cortex and thalamus of each hemisphere from the animal with tracer injections (animal 146) were removed. The thalamic tissue blocks were allowed to sink in 30% sucrose fixative and then frozen sectioned at 50 μm and a 1/5 series reacted for WGA–HRP label using cobalt chloride intensification (Adams, 1977). The positions of the labelled injection sites were verified by LM in the series through the thalamus of each hemisphere. The visual cortices of the injected animal were serially vibrotome sectioned at 50 μm and a 1/10 section series reacted for WGA–HRP. These sections were mounted for LM and scanned to locate regions of retrogradely labelled corticogeniculate projecting cells in layer 6 in the right hemisphere, and retrogradely labelled corticopulvinar cells in layer 5 of the left hemisphere. Additional sections were then reacted for WGA–HRP from positions in the section series between the scanned sections containing labelled cells; these additional sections were osmicated in 1% Os in phosphate buffer for 20 min, washed in phosphate buffer and dehydrated in a graded (25–100%) series of alcohols. The sections were infiltrated with acetone/Epon (Agar Scientific) 1:1 for 3 h followed by overnight in Epon. The sections were then mounted on Teflon (PTFE from RS Components) coated slides using freshly prepared Epon as a mounting medium. Sections were covered with silicon (Sigma, Sigmacote) treated coverslips and pressed flat. Polymerization was carried out at 60°C for 48 h.
Sectioning Labelled cells and a Sample of Unlabelled Layer 6 Meynert Cells for EM
HRP–WGA labelled cells in layer 5B and at the upper and lower borders of layer 6, and a set of unlabelled giant Meynert cell pyramids in layer 6, were located using LM in the Epon-embedded sections; the darkly staining WGA–HRP inclusions in their cytoplasm served to identify the retrogradely labelled cells. The large unlabelled layer 6 Meynert cells were clearly identifiable by their large size compared to other pyramids in the sections surveyed by LM. Neurons lying well within the depth of the section were selected to allow for sufficient sampling in later resectioning. Each cell was photographed, then the coverslip was removed and each cell, together with a small amount of surrounding tissue, was excised from the slide and re-embedded in a flat-bottomed BEEM capsule using Epon. Sequential semithin (0.75–1 μm) sections were cut from the resulting block using an ultramicrotome. Once the HRP–WGA-labelled cell was identified in a semithin section, the subsequent sections were serially collected and each mounted on plastic (Thermanox, AGAR Scientific) coverslips, mounted on slides. These sections were stained with toluidine blue in 1% borax, temporarily mounted in glycerol and photographed under a light microscope. The slide-mounted semithin sections were then re-embedded using an inverted capsule on the slide and polymerized at 60°C for 24 h. The resulting block was broken away from the slide, mounted on an ultramicrotome and serial ultrathin sections cut. The ultrathin series were collected on formvar coated 2 × 1 mm slotted copper grids, stained with uranyl acetate and lead citrate (Reynolds). Grids were examined by EM and the labelled cell located using the photograph of the semithin section as a guide.
Electron Microscopic Analysis of Large Populations of Unlabelled Neurons
Estimates of Type 2 Synapse Coverage on Excitatory Neurons in Different Laminae of V1
A representative semithin section, stained with toluidine blue, was cut from the pia to white matter extent of each of five blocks of tissue, or five tissue sections, embedded for EM, from the outer operculum of area V1 of each hemisphere of animals 146 and 1205. The boundaries of layers within depth of the cortex were mapped, based on the fibre and cell density patterns [compared with cytochrome oxidase staining patterns in many of our previous studies (Fitzpatrick et al., 1985), and see Fig. 1, and known to be accurate indicators of laminar position (Lund, 1987)] by LM onto drawings of these sections, together with the detailed pattern of blood vessels. A light microscope photomontage was made of relevant laminae in the semithin sections and the block trimmed to include those laminae.
EM thin sections were then taken, with at least 50 μm between sections from the faces of these blocks or from different embedded histological sections to avoid sampling the same cells twice. These blocks or retrieved histological sections were trimmed to include particular sets of laminae. The laminar boundaries were recognized under the electron microscope by the patterns of blood vessels and cells compared to their positions in the photomontage. The tissue within each lamina was surveyed and each neuron somal cross section encountered was examined; neuron somata were accepted for the sample if the nucleus was present within the cell profile. The perimeter of each cell soma was then surveyed at ×50 000 magnification to identify synaptic appositions [see Fig. 2 for representative synapses — Colonnier's classification of synapses was used (Colonnier, 1968)].
Any cell perimeter with both type 1 (asymmetric postsynaptic density, round vesicles) and type 2 synapses (symmetric synaptic density, pleiomorphic vesicles) was rejected from the sample on the basis that the presence of both types of synapses, rather than purely type 2 synapses, is generally characteristic of a GABAergic inhibitory inter- neuron (Colonnier, 1968). Profiles of the neurons with only type 2 contacts on their surface, or ring profiles with no synaptic contacts at all, were accepted for the sample. A total of 2955 cells were used in the sample for animal 146 and 1699 cells for animal 1205. The length of each type 2 apposition found on the somata ring profiles was measured using the on-screen measuring system of the Jeol 1010 electron microscope, and the number of type 2 contacts on each ring profile noted. Twenty-five ring profiles of somata fitting these criteria were measured for each layer from each of five blocks or five sections, giving a total sample size of 125 somatic rings per layer per animal. The density of synaptic appositions was determined for the total length of somatic membrane surveyed for each of the five samples of 25 rings for each layer, as described above.
In order to ensure we used sample sizes that were large enough to avoid random variation, we took the samples of 125 somatic ring profiles surveyed for two single layers with least and most variance (layers 5A and 2/3, in the left hemisphere of animal 146) and randomly divided the rings into 100 sets of five samples, each comprised of the same number of somatic ring profiles. This was repeated using different numbers of individual rings (10, 15, 20, 25) per sample. We then determined what number of rings had to be included in the samples to ensure that mean coverage for each sample lay within the confidence limits of mean ± 2 SD between different samples of rings collected in the same way from the same lamina. This number of rings per sample was then used throughout the study. Statistical comparisons (methods outlined below) were then made among all layers to determine significance of any differences among them in mean coverage of neurons by type 2 synapses.
There has been considerable debate in the literature on the best methods for evaluating synapse numbers in EM material (DeFelipe et al., 1999). In this study, where measurement was made of relative synaptic coverage amoung the cell populations, rather than absolute numbers of synapses on each cell, we believe that the methods used were sufficient to answer the questions posed above, and had the advantage that the measures could be accomplished within a reasonable time, using a much larger sample of cells than would be possible if absolute numbers were required.
Layers Chosen for Analysis
The layers chosen for separate analysis — layers 2/3, 3B, 4B, 4C (with three equal divisions, 1, 2, 3 in depth), 5A, 5B and layer 6 (6a: upper third: 6b: middle third; 6c: lower third) — are indicated in Figure 1.
Analysis of Labelled cells, Meynert Cells and Neighbouring Unlabelled Cells
The on-screen measuring system of the Jeol 1010 was used to measure the mean diameter (longest axis + axis orthogonal to the longest axis/2) of each labelled neuronal somatic profile (‘ring’) at ×10 000 magnification where the cytoplasm contained a significant portion of the cell's nucleus (to avoid regions where the section might be very oblique to the somatic membrane and appositions sites therefore less clear, and also to avoid sections of large proximal dendrites which might be mistaken for somata). The circumference of the cell somata was derived using the formula 2πr, where r = mean diameter/2. Synapses on the surface of the soma were counted, and the length of their synaptic apposition sites measured. This process was repeated for two further ultrathin sections from each cell, separated in depth through the block by at least 1 μm to avoid counting the same synapse twice. This procedure was used to provide three somatic rings for each of 10 labelled cells for two samples of lateral geniculate nucleus (LGN) projecting neurons from the right hemisphere of animal 146 (10 cells in the deepest third of layer 6, and 10 cells in the upper third of layer 6); a similar procedure was used to sample 10 labelled pulvinar projecting neurons from layer 5B of the left hemisphere of animal 146; and for 10 large, unlabelled layer 6 Meynert cell pyramids taken from both right and left hemipheres of animal 146. The Meynert cells were readily identified by their very large size compared to other neurons in layer 6 (Lund, 1973; Winfield et al., 1981). At the same time as making measures of the WGA–HRP labelled cells and their synapses, an unlabelled ‘near neighbour’ control cell, close to each labelled cell, was also measured, together with its synapses, in the same fashion as the labelled cell. This gave a sample of three rings from each of 10 labelled control cells (30 rings) for comparison with similar sized data sets from each set of labelled neurons. Thus the sample size for labelled cells and their near neighbour control cells was different from that used in the analysis of random somatic ring data collected for each layer (five samples of 25 rings per layer).The 30 somatic ‘ring’ profiles from each of these labelled cell samples were used to estimate synaptic coverage for each set of neurons as outlined below.
Statistical Methods Employed
Optimal Sample Size
The workload in processing the tissue for EM and counting the synapses was considerable. However, it was desirable to have as large a database as possible to allow robust statistical predictions about the characteristics of the sample under examination. The task was to find a good compromise between these two factors. To do this, we drew random sets of five samples with somatic ring numbers per sample of n = 5, 10, 15 or 25 from the pooled data (collected as five samples per layer with 25 rings per sample). In this way we generated 100 bootstrapped data sets for each of the four sample sizes, and tested mean and variance of those bootstrapped samples, to check whether they lay within the range of the original sample. This procedure was performed for two representative layers in animal 146 (layers 3 and 5A where values were at the extremes of the distribution).
Comparison of Differences in Cell Size and Length of Synaptic Appositions between Animals, Hemispheres and Layers
The frequency distribution of these variables followed a Gaussian distribution and sample size was always n > 50, thus t-tests were used to test for differences in mean and variance.
Comparison of Differences in Synaptic Coverage between Animals, Hemispheres and Layers
As the data did not follow a uniform or normal distribution, and were not symmetrical, a non-parametric procedure, the sign–rank test (Rees, 1987), was used. Variables were compared pairwise with two related samples to test the hypothesis that two variables have the same distribution. The actual form of the sample frequency distribution is of no importance here, with the disadvantage that the test statistic is less powerful, i.e. declaring two distributions to be not significantly different when they actually are (Rees, 1987) (see also Prophet StatGuide: www.prophet.abtech.com/statguide/sghome.html).
Synaptic Cover on Neurons Projecting to Different Subcortical sites in Animal 146
In total 10 cells were selected for each projection target. For each cell, three somatic rings were measured. For control cells, we selected and measured, in the same fashion as the labelled cells, 10 unlabelled neurons in the vicinity of the prelabelled ones. In order to compare these sets of 30 rings with the larger sample of unlabelled cells in the same layer (n = 125) we had to randomly sample the data (i.e. bootstrap) from the labelled and nearest neighbour unlabelled cells. A non-parametric bootstrap algorithm was used (Zoubir and Boashash, 1998; Zoubir, 1999). Resampling was done 32 times for each data set, each with n = 120.
Figure 1A illustrates, in a section stained for Nissl substance, the laminar numbering and organization of area V1 in the macaque monkey and the layers recognized in this study. The same layers could be identified by cell size and packing density, together with fibre density and orientation, in the semithin sections studied by LM (Fig. 1C) and in the osmicated frozen section series by changes in fibre staining and background density (Fig. 1B). Both animals' cortices had similar laminar organization. Figure 2 shows representative electron micrographs of type 1 and type 2 synapses from area V1 of each animal. While the tissue from animal 146 had undergone processing to reveal HRP–WGA retrograde label, it was still adequately preserved for study of synaptic morphology though not as good as the tissue from animal 1205, which had been directly prepared for EM. Figures 3 and 4 each show the sequence of identifying a retrogradely labelled cell in the osmicated sections of animal 146 at low and high magnification LM (Figs 3A,B and 4A,B), resectioning the tissue to find the same cells in semithin sections (Figs 3C and 4C), then finding the same cell for electron microscopic examination of the somal membrane and its synaptic appositions (Figs 3D and 4D). Two examples of this sequence of locating the labelled cells for EM are shown since it is very important to demonstrate that the label was clearly identifiable and tissue well preserved, despite undergoing several different procedures. Meynert cells were readily identified in the semithin sections by their very large size compared to the rest of the neurons within layer 6 (Fig. 5).
From our analysis of variance, creating 100 sets of five samples for each of four different sample sizes (10, 15, 20 or 25) of somatic rings from a layer with most variance and one with least (layer 5A and layers 2/3 of animal 146), it was found that ring samples should include at least 15 rings, and ideally no less than 20 rings, to have mean synaptic coverage for the samples fall within a 95% confidence limit. On this basis, the five sets with sample size of 25 rings each, actually collected from each layer or cell group throughout this study to determine the mean percent of membrane occupied by synapse apposition sites, was deemed statistically valid in terms of its likelihood of producing a consistent figure for synapse cover from the particular environment sampled.
Differences between Animals
Figure 6 shows the results of our EM measures of mean somatic ring profile circumferences (calculated from mean diameters) as a rough indicator of relative sizes of the neuron somatic profiles examined, mean type 2 synaptic apposition length, and proportion of the total somatic membrane occupied by type 2 synapses (percent cover) in the two animals in this study. EM tissue surveyed from animal 146 was derived from single sections processed first for WGA–HRP, while in the case of animal 1205 the cortical tissue was derived from blocks cut directly from the fixed brain and osmicated in blocks; since overall size of all components between the two animals may differ due to different preparation, the actual differences in soma diameters and contact lengths between the two animals were deemed irrelevant. Instead, it was asked if the relative similarities or differences between parameters measured for neurons of particular layers in one animal also existed in the other animal, despite it being both another macaque species and another individual.
Figure 6 shows that in both animals the smallest cells were found in layers 4C and 5A, but that the pattern of differences in mean synaptic length between layers differed between the two animals. Furthermore, while the highest percent synaptic cover was found in layer 4C in both animals, it was the deep layers in animal 146 and the superficial layers in animal 1205 where cover was least. We conclude that there was no clear commonality in the overall pattern of synaptic cover between layers in the two animals.
Differences between Hemispheres of Single Animal
The samples taken from right and left hemispheres of animal 146 are shown in Figure 7. While not apparent from the histograms in Figure 7, somatic ring circumference was significantly different between the hemispheres in layers 2/3 and 4B (P = 0.01) due to differences in the distributions of the sample ring sizes; however, interhemispheric differences in synaptic length and synaptic cover were not significant in any layer in animal 146.
Relationship between Cell Size, Synaptic Apposition Size and Synaptic Cover
Synaptic cover was not generally related to overall size of the neuron [see Fig. 8 where mean ring circumference for cells in single layers in animal 146 (both hemispheres) is plotted against synaptic cover]. Differences in synaptic cover were also not related to difference in synaptic apposition length. This is apparent from Figures 6 and 7. While the largest pyramidal cells (Meynert cells of layer 6) did have much higher synapse cover than other neurons (see Fig. 9), the mean synaptic length of synaptic appositions on the Meynert cell surface was not significantly different from synapses on the general cell population in layer 6.
Synapse Cover on Spiny Stellate versus Pyramidal Neurons
Comparison of synaptic cover between neurons of layer 4C, where excitatory cells are almost exclusively spiny stellate neurons, and the other layers, where excitatory cells are largely or exclusively pyramidal neurons (see Figs 6 and 7), showed layer 4C to have somewhat higher synaptic coverage than other layers, both deep and superficial, but the significance of that difference from deep and superficial layers varied between hemispheres of 146 and between the two animals. In animal 1205 layer 4C neurons had significantly (P = 0.02–0.001) higher cover than all other layers but layer 6; in animal 146 neurons of layer 4C had significantly higher coverage (P = 0.02–0.001) than the deeper layers and layer 4B but were not different from neurons of layers 2/3. These findings provide only slight support for the suggestion that spiny stellate neurons might differ from pyramidal neurons in inhibitory synapse cover, either because of intrinsic nature or perhaps because the thalamic input to these cells might require higher inhibitory cover on the postsynaptic neurons [the GABA boutons certainly appear to be larger (Fitzpatrick et al., 1987)].
The Relationship between Synapse Cover and Type of Thalamic Axon Input
Comparison was made of the significance of differences in synaptic cover between layer 4C1 neurons in the upper third of the 4C layer (receiving exclusively the terminations of M LGN axons) and neurons in the 4C3 division (receiving exclusively P LGN axons) for the two hemispheres of animal 146 and for animal 1205. Synaptic cover did not differ significantly between these divisions in animal 1205 or in animal 146 (P = 0.4 for 1205, and P = 0.35 for 146). We conclude that there was no evidence that thalamic axons of different type required different postsynaptic neuron inhibitory cover.
Comparison between Synapse Cover on Neurons in Layers 4B and 6, where some Proportion of Neurons Share the Property of Direction Selectivity, and Layers 2/3 and 5, where Neurons Lack Direction Specificity
The data shown in Figures 6 and 7 provide no evidence for similarities in somatic inhibitory cover between layers having direction specificity, or evidence for difference between neurons in these layers and neurons in layers 2/3 and 5 that lack directionality Synaptic cover differed between the two layers sharing direction specificity in both animals (animal 146: layer 4b versus layer 6: P = 0.001; animal 1205: layer 4b versus layer 6: P = 0.001), and there was no similarity in cover between layers that lack direction specificity (animal 146: layer 2/3 versus layer 5: P = 0.001; 1205 layer 2/3 versus layer 5: P = 0.05).
Synaptic Cover on Neurons Projecting to Different Subcortical Sites in Animal 146
Figure 9 shows, for animal 146, the mean ring membrane length, mean synapse apposition length and mean synaptic cover for the 30 ring samples from three groups of 10 retrogradely labelled cells (left hemisphere labelled layer 5 cells; right hemisphere labelled layer 6a cells and 6c cells), and a group of 10 large unlabelled upper layer 6 Meynert cells taken from both hemispheres. Similar data are shown for 10 unlabelled near neighbour cells (in the case of each of the three labelled groups) and for same layer, same hemisphere, unlabelled somatic ring populations for all four groups.
Average ring sizes for the three populations of labelled cells were found to be significantly larger than the control sets of unlabelled nearest neighbour cells or the randomly selected large populations of cell somatic ring profiles. It is probable that this was due to choice of the largest labelled cells because they were more prominent in the light histology and it was easier to obtain three sections for EM, spaced at least 1 μm apart, from the larger cells. It should be noted that the 30 ring samples taken from 10 nearest neighbour cells around each set of labelled neurons did not differ significantly in synaptic cover from the large ring sample taken from the same layer. This provided evidence that the method of taking 30 ring samples from the labelled cells did not in itself lead to differences in synaptic cover, compared to the methods used to obtain the large ring sample from the same layer.
The sample of labelled layer 5B cells had significantly (P ≤ 0.001) lower synaptic cover than the random ring population taken from the same layer and from the near neighbour sample. Upper layer 6a labelled cells had significantly (P ≤ 0.001) higher synaptic cover than either their near neighbours or the same sublayer random ring sample. Lower layer 6 (division 6c) labelled cells did not differ significantly from either nearest neighbours or the random ring sample from the same division. Meynert cells had a markedly higher synaptic cover than the random ring sample from layer 6 as a whole (P ≤ 0.001). The three labelled cell sets and Meynert cells all differed significantly (P ≤ 0.001 in all cases) in synapse apposition cover from each other.
While the methods used for this study did not allow an accurate measure of the real size of the neurons studied, the mean ring diameter provided a means of comparing the relative sizes of the different populations of neurons studied. Similarly, mean synapse length provided a means of comparing relative synapse sizes between samples. When the measures of the percent of somatic membrane covered by synapses were applied to the surface areas of cells of the mean diameters measured, and the percent of that area occupied by synapses is divided by the average area of single synaptic apposition sites (derived from the mean length of appositions on that cell type), the relative density of synapses on cells per layer, or on labelled sets of cells, can be calculated. This assessment of relative density of synapses per cell is shown for all layers studied in animals 146 and 1205, together with the labelled efferent cell sets, in Figure 10. However, the strength of somatic inhibition appears to be best correlated with the cumulative area of the synapses (if it is assumed that area of the apposition site is proportional to the number of receptors of similar type). In which case percent synapse cover per cell (which takes into account differences in cell and synapse size) would appear to be the best measure of the functional impact of somatic inhibition rather than numerical density of synapses.
In this work we were not trying to arrive at absolute numbers of synapses per cell soma; the questions posed above can be satisfactorily answered by knowing the relative differences in somatic synapse cover amoung the groups of neuron identified (percent of total somatic membrane occupied by synaptic sites). We have ignored here the fact that GABAergic axon terminals can differ in kind and that their receptor characteristics can also differ; we have, instead, concentrated on sampling as many different neurons as possible and determining if and how somatic areal cover by type 2 contact sites (Colonnier, 1968), as a general class, may differ, given the factors outlined above.
Since spine bearing neurons comprise ~80% and GABAergic neurons ~20% of neurons in area V1 (Hendry et al., 1987; Beaulieu et al., 1992), the cells sampled for this study are likely to largely represent pyramidal or spiny stellate neurons (Lund, 1984); there will, however, be some small percent contamination of the sample by interneurons where no type 1 contacts were identified. If GABAergic cells were to have no higher synaptic coverage than the total cell population, we estimate that contamination to be no greater than 17% of the somatic rings (those GABAergic cells with no visible synaptic contacts and those with just type 2 contacts visible).
The rationale for choosing to analyse separately particular layers and divisions of layers, as shown in Figure 1, was as follows. Layer 4C excitatory neurons are almost entirely spiny stellate cells in morphology and they can therefore be compared to cells in layers that have only pyramidal neurons (e.g. layers 5–6 or layers 2–3). Layer 4C receives the bulk of terminations of thalamocortical axons from the LGN. The upper half of layer 4C is innervated by axons from the magnocellular division of the LGN, the lower half is innervated by axons from the parvocellular LGN (Blasdel and Lund, 1983). We could, therefore, in layer 4C compare neuron populations that receive different types of thalamic axons by comparing the upper third (4C1) with the lower third of 4C (4C3), omitting the zone where the dendrites of postsynaptic spiny stellate cells might receive both types of input (division 4C2). In the primary visual cortex there are two layers where neurons can be strongly directionally selective — layer 4B and layer 6 (Hawken et al., 1988). These layers were compared to ones with cells in layers 5 and 2–3 that lack this property. We compared synaptic cover on pyramidal neurons of different sizes at various cortex depths; synaptic coverage was compared among cell groups of different mean ring sizes. This comparison allowed us to determine if cell size alone determines inhibitory synapse cover. Cells in layers that project to different destinations were also compared; cells of layer 2/3A, a layer that projects to other cortical areas, were compared to cells of layer 5, that projects subcortically. Our samples of cells retrogradely labelled from the LGN in the upper third of 6 versus the lower third of layer 6 also allowed us to examine this question more specifically since cells in these two depths are likely to have different projections; the lower layer 6 cells project to both the magnocellular and parvocellar LGN and have collaterals to upper layer 4C, while the upper layer 6 labelled cells project to parvocellular LGN alone and have collaterals to lower layer 4C (Fitzpatrick et al., 1994; Callaway, 1998). Our layer 5b labelled cells projecting to the pulvinar nucleus were also compared to the labelled LGN projecting neurons of upper and lower layer 6. Unlabelled layer 6 Meynert cells (excluded from our ring samples of layer 6) were also used as a population for comparison since they are known to project to both superior colliculus and cortical area MT (Fries et al., 1985).
Significance of Results
This study set out to answer a number of questions that addressed what factors may determine the percent of somatic membrane covered by inhibitory synapses on excitatory neurons that are significant to macaque monkey primary visual cortex. Significant differences in somatic inhibitory coverage are not determined by synaptic apposition length or simply related to overall size of the somata. There were significant differences in the patterns of laminar cover between the two animals studied, suggesting that either species or individual differences can be marked in this regard. Comparison between the two animals showed that highest cover was present in layer 4C of both animals; however, the difference in cover was not significantly different between layer 4C and superficial layers in animal 146. Within layer 4C there was no significant difference in cover between upper and lower divisions (which are served by different populations of thalamic axons). This suggests that cell type (spiny stellate in 4C, pyramidal in other layers), presence of thalamic input (predominantly to layer 4C), and type of thalamic input, are not reliable determinants of higher or lower synaptic cover. The comparison between the two hemispheres of animal 146 showed no significant difference in synaptic cover, suggesting that interhemispheric differences in macaque V1 are slight and unlikely to determine synaptic cover differences. We have shown that inhibitory synapse cover on four sets of pyramidal neurons in layers 5 and 6, each with different efferent projection targets, did differ significantly from that of other cells in those same layers and from each other. These consistent differences argue against somatic cover by inhibitory synapse appositions simply being stochastic in nature, even though coverage seemed unrelated to the other factors described above.
We were reassured from our results — where random large samples of cell ring profiles had similar cover to the sample of nearest neighbour cells — that our method of sampling labelled cells did not lead to significant differences in synapse cover. Significant differences in synaptic cover between cell populations projecting to different destinations, even within single layers, and the very different synapse cover on two of the three labelled sets of neurons and on the Meynert cells compared to unlabelled nearest neighbours and random ring counts in the same layers, suggests that inhibitory control on at least the soma region is uniquely determined for each different functional set of neurons in area V1. This finding is not surprising since our own previous studies of inhibitory neuron populations in area V1 (Lund, 1987; Lund and Wu, 1997; Lund and Yoshioka, 1991) showed each layer to have its own unique set of inhibitory interneurons and to receive unique interlaminar projections from interneurons located in other layers. However, we cannot distinguish whether the specific patterns of inhibitory cover on different cell types are due to specific types of inhibitory neurons or due to the particular circuitry they engage in, since we do not know the origins of the type 2 contacts. The unlabelled neurons we studied in the same layers as the labelled neurons presumably comprise diverse functional subsets, each with unique inhibitory control and synaptic cover. While we see in this study that the mean synaptic cover of random samples of cells in all cases, apart from deep layer 6, differed significantly from the labelled sets of neurons from the same layer, even the finding that the deep layer 6 labelled cells did not differ significantly in synaptic cover from the mean cover of cells in the same environment cannot be taken to indicate a lack of difference between different functional groups at that level in layer 6.
It is difficult to conceive what the functional implications of these differences in inhibitory cover may be. We do not know the sources of these inputs, or their relative efficacy. While it has been suggested that inhibition on the soma and proximal dendrites should be more effective than when placed on the distal dendrites, this may not be strictly true. Pyramidal neuron dendrites contain active conductances (Stuart and Sakmann, 1994) and it has been reported that pyramidal cells may be more electronically compact than previously believed from in vitro data (Bernander et al., 1991). The findings of this study indicate that pyramidal neurons can vary greatly in how their synaptic load is apportioned on the soma. We have no information as to whether the synaptic load on the dendrites may also vary greatly between neuron groups or if there is any easily understood relationship between the excitatory and inhibitory synaptic cover on different regions of single neurons. However, taking the observed variability in synaptic load into account in models of pyramidal neuron and cortical circuitry may allow for more realistic simulation of cortical function.
The results of this study confirm the conclusions of the studies by Farinas and DeFelipe in the cat visual cortex (Farinas and DeFelipe, 1991a,b), namely that neurons making projections to different targets outside area V1 have significantly different amounts of inhibitory synapse apposition cover on their somata. Moreover, our study shows that neither cell size nor laminar location are likely to be significant factors underlying differences in synaptic cover in the monkey. In terms of understanding and modelling visual cortex functions, it is clear that inhibitory control of functionally distinct excitatory neurons, even within single layers, is likely to be uniquely determined for each functional set.
This work was supported by MRC grants G9203679N and G9408137, and Wellcome Trust grant 03927/Z93/Z1.2.
Address correspondence to Professor J.S. Lund, Department of Ophthalmology, John A. Moran Eye Center, 50 University Health Drive, University of Utah, Salt Lake City, UT 84132, USA. Email email@example.com.