Axon topography of layer 6 spiny cells to orientation map in the primary visual cortex of the cat (area 18)
To uncover the functional topography of layer 6 neurons, optical imaging was combined with three-dimensional neuronal reconstruction. Apical dendrite morphology of 23 neurons revealed three distinct types. Type Aa possessed a short apical dendrite with many oblique branches, Type Ab was characterized by a short and less branched apical dendrite, whereas Type B had a long apical dendrite with tufts in layer 2. Each type had a similar number of boutons, yet their spatial distribution differed from each other in both radial and horizontal extent. Boutons of Type Aa and Ab were almost restricted to the column of the parent soma with a laminar preference to layer 4 and 5/6, respectively. Only Type B contributed to long horizontal connections (up to 1.5 mm) mostly in deep layers. For all types, bouton distribution on orientation map showed an almost equal occurrence at iso- (52.6 ± 18.8 %) and non-iso-orientation (oblique, 27.7 ± 14.9 % and cross-orientation 19.7 ± 10.9 %) sites. Spatial convergence of axons of nearby layer 6 spiny neurons depended on soma separation of the parent cells, but only weakly on orientation preference, contrary to orientation dependence of converging axons of layer 4 spiny cells. The results show that layer 6 connections have only a weak dependence on orientation preference compared with those of layers 2/3 (Buzás et al., J Comp Neurol 499:861–881, 2006) and 4 (Karube and Kisvárday, Cereb Cortex 21:1443–1458, 2011).
KeywordsApical dendrite Bouton overlap Optical imaging Pyramidal cell 3D reconstruction
Unraveling the connectivity of individual cortical neurons is an important step towards understanding their placement and role in the cerebral cortex. One of the approaches that has been used successfully is to compare connectivity features of single cortical neurons with functional architectures and determine their interrelationship, whereby single cell connectivity can be directly related to the spatial distribution of functional properties. Since the neocortex is a six-layered structure and neurons show marked structural and functional differences depending on laminar location, the above-mentioned comparison must be made at best for each layer separately. For the primary visual cortex, it is known that in the upper layers long-range horizontal connections link regions which represent similar functional attributes such as orientation preference (in cat: Gilbert and Wiesel 1989; Kisvárday et al. 1997; but see also Martin et al. 2014, in monkey: Malach et al. 1993; Yoshioka et al. 1996; in tree shrew: Bosking et al. 1997). When, however, deeper cortical layers are under scrutiny, a marked change in the morphological and functional interrelationship is observed compared to their upper layer counterparts (Hubel and Wiesel 1962; Gilbert 1977; Tsumoto et al. 1978; Henry et al. 1979; Harvey 1980a, b; Martin and Whitteridge 1984). Indeed, it was shown that the orientation distribution of layer 4 (L4) cortico-cortical connections differs from that of the superficial layers, i.e., the connectivity distribution of single L4 spiny neurons is not only less iso-orientation specific (Yousef et al. 1999), but also functional selectivity varies from axon branch to axon branch for the same L4 spiny neuron (Karube and Kisvárday 2011).
Layer 6 (L6) represents one of the chief output layers of the visual cortex to other cortical areas and subcortical structures (Briggs 2010; Thomson 2010). On the other hand, it receives inputs from the dorsal lateral geniculate nucleus (LGN; LeVay and Gilbert 1976; da Costa and Martin 2009; Bannister et al. 2002) and the visual claustrum (LeVay and Sherk 1981; da Costa et al. 2010) and thus establishes cortico-subcortical loops (Gilbert and Kelly 1975; LeVay and Sherk 1981; McCourt et al. 1986; Katz 1987). As a consequence, the connections of L6 spiny cells must have an impact on the entire visual signaling pathway, including other cortical areas as well as subcortical visual centers (for review, see Cudeiro and Sillito 2006; Sillito et al. 2006; Briggs 2010; Thomson 2010). In line with this, there are studies which indicate that L6 neurons are functionally heterogeneous in cat (Hirsch et al. 1998) and other species (Usrey and Fitzpatrick 1996; Wiser and Callaway 1996). For example, in rodent, L6 neurons are assumed to be involved in gain control of visually evoked responses both in cortex and thalamus (Olsen et al. 2012; Bortone et al. 2014) and results of a more recent study indicates that the functional tuning properties of cortically projecting and those of thalamic projecting L6 neurons differ markedly from each other (Velez-Fort et al. 2014). Although a newly developed method allows direct imaging of deep layers in rodents (Andermann et al. 2013; Masamizu et al. 2014), preserving the integrity of anatomical connection of the imaged regions in not resolved.
L6 is also named as multiform layer composed of a heterogeneous neuron population in cat (Gilbert 1977; Harvey 1980b; Katz 1987; Grieve and Sillito 1991a, b, 1995; Hirsch et al. 1998), tree-shrew (Usrey and Fitzpatrick 1996), monkey (Lund and Boothe 1975; Wiser and Callaway 1996) and rodents cortices (Zhang and Deschenes 1997; Zarrinpar and Callaway 2006; Kumar and Ohana 2008; Watakabe et al. 2012). In the cat primary visual cortex, L6 pyramidal neurons are commonly distinguished on the basis of apical dendrite morphology (Lund and Boothe 1975; Katz 1987), which correlates with their subcortical projection patterns. Accordingly, L6 pyramidal cells have been considered to form three main populations: (1) ~50 % of L6 cells project into the LGN (Gilbert and Kelly 1975; McCourt et al. 1986; Katz 1987; Budd 2004) and have a short apical dendrite typically terminating in L4 (type 1 and 2 cells in Katz 1987); (2) <10 % of cells project to the visual claustrum (LeVay and Sherk 1981; Katz 1987) and characterized by a long, slender apical dendrite towards superficial layers; and (3) the remaining cell population provides cortico-cortical connections (McCourt et al. 1986; Katz 1987) either locally, i.e., intrinsic cortico-cortical connections, or to other cortical regions, i.e., extrinsic cortico-cortical connections. It should be noted that a marked difference between L6 neuron types concerns not only morphology, but also their biochemical content. For example, in the non-human primate, L6 cortico-thalamic neurons lack the activity-related neuromodulator synaptic zinc, whereas a subpopulation of cortico-cortical neurons contains synaptic zinc (Ichinohe et al. 2010, for review, see Rockland 2014).
In this study, our aim is to analyze the functional topography of different types of L6 neurons and determine their axonal bouton distribution patterns in relation to local orientation map obtained with intrinsic signal optical imaging in area 18 of the cat primary visual cortex (Payne and Peters 2002). To investigate the potential difference between different types of neuron, we applied a morphological classification of L6 spiny cells based on a quantitative comparison of apical dendrite morphology (see also Katz 1987). In addition to this, in a few cases the functional topography of nearby L6 pyramidal cells was compared with that of L4 spiny neurons. We will discuss how the L6 circuitry is organized with regard to functional architectures and in the context of other cortical layers.
Materials and methods
Nine adult cats (1.8–3.4 kg) were used in this study. All experiments were performed according to institutional (Local Committee at the University of Debrecen) and governmental requirements [Hungarian Enactment of the Protection of Animals (1998.XXVIII.) Section 4] and the guidelines of the European Convention for the Protection of Vertebrate Animals Used for Experimental and Other Scientific Purposes (Strasbourg, 18.III, 1986). The detailed method used here was previously reported (Karube and Kisvárday 2011).
Anesthesia was induced by intra-muscular injection of a mixture of ketamine (7 mg/kg of Ketavit, Pharmacia and Upjohn GmbH, Erlangen, Germany) and xylazine (1 mg/kg of Primazin, Alfasan, Woerden, The Netherlands). Eye drops of Isopto-Max (Alcon Pharma GmbH, Freiburg, Germany) containing antibiotics for protecting the corneal surface, 5 % phenylephrine-hydrochloride (Neosynephrin-POS, Ursapharm, Saarbrücken, Germany) for relaxing the nictitating membrane and 1 % atropine sulfate (Atropine-POS, Ursapharm) for dilating the pupils were applied. All surgical wounds and pressure points were treated with the topical anesthetic, Lidocaine (Xylocaine gel, Astra Chemicals, Wedel/Holstein, Germany). A catheter was inserted in the femoral artery and a Y-shaped cannula in the trachea. For muscle relaxation and nutrition, a mixture of alcuronium chloride (0.15 mg/kg/h of Alloferin, ICN Pharmaceuticals Germany GmbH, Frankfurt, Germany) and glucose (24 mg/kg/h) diluted in Ringer solution was infused through the arterial catheter. In some animals, gallamine tetrathiodide was used instead of Alloferin. A few seconds after the initial dose of muscle relaxant (4 ml/kg), ventilation was commenced with a mixture of N2O/O2 (70 %/30 %) gas containing 0.3–0.6 % halothane using an Ugo-Basile cat/rabbit ventilator (10–13 strokes/min, 50–70 ml per stroke). Blood pressure (100–135 mmHg), end-tidal CO2 (3.8 ± 0.3 %), body temperature (~38.5 °C) were continuously monitored throughout the experiments. Craniotomy was made in the posterior part of both hemispheres to expose area 18 of the visual cortex [Horsely–Clark coordinates antero-posterior (AP): (−5)-(+15), latero-medial (LM): (+0.5)–(+3.5)] (Tusa et al. 1979). A metal chamber (diameter, 30 mm) was fixed to the skull using dental cement (Paladur, Heraeus Kulzer, Wehrheim, Germany). The eyes were fitted with appropriate correction lenses to focus on the monitor screen (GDM-F520, Sony Electronics Inc., Park Ridge, NJ, USA). Viewing distance was 28.5 or 57 cm. During the experiments, eye positions were repeatedly checked on the basis of tapetal back projection of retinal blood vessels.
Optical imaging of intrinsic signals
Optical imaging was carried out to map orientation columns in the central visual representation of area 18 (Horsley–Clarke coordinates AP, 0–12 and LM, 0.5–3.5). Dura mater was removed and the chamber filled with silicon oil (50 cSt viscosity; Aldrich, Milwaukee, WI, USA). The vascular pattern of the cortical surface was imaged using green light (λ = 545 ± 10 nm). During imaging of intrinsic signals, the cortex was illuminated with orange light (λ = 609 ± 5 nm) using fiber-optic cables fitted with either an adjustable fiber-optic head-lens or a fiber-optic ring fixed to the objective. Intrinsic signals were recorded with a CCD camera (CS8310C, Tokyo Electric Industry Inc., Tokyo, Japan) in slow-scan mode through two objectives arranged in a ‘tandem configuration’ (SMC Pentax A 1:1.25, 50 mm). During data acquisition, the camera was focused 700 μm below the cortical surface. Images were acquired by a computer using the Imager 2001 and VDAQ software (Optical Imaging Inc., New York, NY, USA). Visual stimuli were generated using the VSG Series Three (Cambridge Research System, Oxford, UK) system. Stimuli were presented on a B/W monitor (SONY, GDM-F520, 100 HZ, non-interlaced mode) placed at 28.5 cm in front of the cat’s eye. They contained full field square wave gratings with high contrast at eight equally spaced orientations that moved forth and back (bi-directional drift) along the orthogonal axis of the orientation. The mean luminescence of the stimuli including the blank was 13 lx. Spatial and temporal frequency was optimized for area 18 (Movshon et al. 1978; Orban 1984) and each stimulus was presented 25–56 times in a pseudo-random manner. Either the blank screen or a stationary image of the next stimulus was presented during interstimulus interval lasting for 10 s. Image acquisition commenced 1 s after stimulus onset and lasted for 4500 ms during which ten data frames were collected. The camera frames underwent spatial summation (two times binning) resulting in a spatial resolution of 21.28 μm per image pixel. Typically, gratings <0.2 cycle per degree and >1 Hz temporal frequency yielded strong activity change and elongated orientation domains characteristic for area 18 (Swindale et al. 1987; Bonhoeffer and Grinvald 1993; Bonhoeffer et al. 1995; Ohki et al. 2000; Shmuel and Grinvald 2000).
For confirming optical imaging results, single- and multi-unit responses to visual stimulation were recorded extracellularly to define the minimum spatial response field (RF), preferred orientation and direction of motion and ocularity using hand-held stimulation at random locations within the mapped area. To this end, glass electrodes (8–10 μm of tip diameter) were filled with 0.5–1 M NaCl (~5 MΩ). The pipettes were advanced perpendicularly to the cortical surface using a manual micro-drive fitted with a digital depth meter (10 μm accuracy, Sylvac SA, Crissier, Switzerland). Along each electrode track single and multi-units were recorded at several depths (300–2000 µm) from the cortical surface. To stabilize recordings, after advancing the electrode into the cortex, the chamber was filled with 3 % agar and covered with low-melting point paraffin (~43 °C, Merck, Darmstadt, Germany). Signals were amplified by AxoClamp-2A (Axon Instruments, Foster City, Canada) and filtered (0.3–10 kHz). Spike activity was monitored using a window discriminator (World Precision Instruments Inc., Sarasota, FL, USA) coupled to an audio monitor. The recorded units had large minimum discharge field (average: 9.5° ± 6.8° of diagonal length, N = 71 from six animals). In some cases RF had large size (maximum, 51°), especially at depths corresponding to layers 5/6.
Tracer injection, histology, and neuron reconstruction
At selected locations in the optically imaged regions, small extracellular tracer injections were carried out in layer 6 (>1200 μm below the cortical surface). Care was taken in choosing locations which represented flat parts of the lateral gyrus so that an interpretation of the labeled structures with respect to the columnar organization of orientation preference could be readily addressed. For example, cortical locations 1 mm from the inter-hemispheric midline, near to the lateral sulcus and major blood vessels were avoided (see Supplementary Figure 1). In each animal several such injections were deposited (6, 10, 11, 10, 8 and 12 injections, respectively) with a lateral separation of 0.5–1 mm. For neuronal labeling, 1–2 % biotinylated dextran amine (BDA, 10000 MW, Thermo Fisher Scientific, Waltham, MA) was diluted in 0.5–1 M NaCl and filled in recording glass micropipettes (8–10 μm of tip diameter). After unit recordings (see above), BDA was injected thorough the same glass micropipette. In this way, we could visualize the location and approximate depth of recordings. For each injection, positive 0.8 µA square wave current was applied for 20 min (250 ms ON/250 ms OFF). In order to help aligning the histological sections with optical images, 5–8 reference penetrations were placed in the region of interest using empty glass pipettes (~20 µm tip diameter) at stereotaxically determined locations (Supplementary Fig. 1; for details, see also Buzás et al. 1998; Karube and Kisvárday 2011). Finally, the animals received a lethal dose of anesthetics and were perfused transcardially with Tyrode’s solution (~3 min) followed by 2 L of fixative containing 2 % paraformaldehyde (PF, Merck) and 0.1 % gultaraldehyde (GA, Merck) in 0.1 M phosphate buffer (PB, pH 7.4). Tissue blocks containing the region of interest were dissected and 80-μm-thick sections cut on a vibratome (Leica S-1000, Wetzlar, Germany). Importantly, the sectioning plane was set as parallel as possible with the imaged cortical surface. Washing of sections was carried out at room temperature (RT), whereas all other steps of the histological procedure at 4 °C. Briefly, sections were washed in PB (10 min) and 0.05 M Tris buffered saline (TBS, 2× 10 min) followed by incubation in avidin–biotin-complexed HRP [ABC-Elite (Vector, Burlingame, CA, USA) at 1:200 dilution in 0.05 M TBS containing 0.1 % Triton X] overnight. After washing in TBS (10 min) and tris-buffer (TB, pH 7.6), the sections were incubated in 0.05 % of 3,3-diaminobenzidine (DAB, Sigma-Aldrich, St. Louis, MO, USA) in TB containing 0.0025–0.005 % CoCl2. DAB was visualized in the presence of H2O2 (0.01 % at final concentration, for 1–3 min). The reaction was stopped by adding an excess of TB in succession. The sections were post-fixed in OsO4 (1 % in 0.1 M PB, 45 min, Sigma-Aldrich) and dehydrated in ascending series of ethanol and two times propylene oxide (Merck). Then the sections were transferred in resin (Durcupan, Sigma-Aldrich) for 24 h, mounted on slides, coverslipped and cured at 56° for 24 h (Somogyi and Freund 1989).
Optical imaging data were analyzed using the Mix software (Optical Imaging Inc.) and a custom-made program written in IDL (Research Systems Inc., Colorado, CO, USA), and MATLAB (The MathWorks Inc., Natick, MA, USA). Single condition maps (SCMs) were obtained by images recorded to a particular orientation divided by the “cocktail blank” (Bonhoeffer and Grinvald 1996). To reduce low-frequency noise due to uneven illumination and surface blood vessels, SCMs were first high-pass filtered (boxcar filter, 50-pixel kernel size), then low-pass filtered (boxcar filter, 5-pixel kernel size). Orientation maps were calculated from filtered SCMs using the vector summation method followed by interpolation (Blasdel and Salama 1986; Bonhoeffer and Grinvald 1991). Large blood vessels, typically at the edge of the maps, were masked out and not included in data analysis. Quantitative evaluation of the imaged orientation preference was made on the basis of 10 degrees binning of orientation map values. In this way, the results could be directly compared with previously published data.
Basic morphological parameters of layer 6 pyramidal cells
Type Aa (N = 8)
Type Ab (N = 10)
Type B (N = 5)
4.31 ± 0.95
2.43 ± 0.76
2.82 ± 1.62
Aa > Ab**
Number of nodes
15 ± 4
9 ± 4
11 ± 7
Aa > Ab*
Nodes on the trunk
>600 μm from the soma
0 ± 1
0 ± 0
4 ± 4
Aa > Ab*
>100 μm from the soma
7 ± 4
2 ± 2
5 ± 5
Number of trunks
7 ± 4
5 ± 1
7 ± 3
Total length (mm)
4.16 ± 1.10
5.22 ± 0.93
6.69 ± 3.14
Total number of nodes
15 ± 5
21 ± 7
21 ± 4
13.96 ± 6.81
16.00 ± 5.28
20.16 ± 7.18
Number of nodes
41 ± 22
47 ± 18
44 ± 20
Number of boutons
1482 ± 765
1308 ± 485
1448 ± 535
Proportion of superficial layer boutons (%)
4.8 ± 8.9
21.2 ± 19.2
28.7 ± 26.8
Mean bouton density/100 μm
11.3 ± 3.6
8.6 ± 2.0
8.1 ± 2.4
Cell body area (μm2)
247.78 ± 31.75
264.91 ± 67.7
302.78 ± 230.59
Cortical layers were determined on the basis of a host of qualitative features such as relative neuron and nerve fiber density, soma size, and the presence of layer-specific cell types, for example, large pyramidal cells at the layer 3/4 border and in layer 5 (O’Leary 1941; Otsuka and Hassler 1962; Harvey 1980a). It should be noted that due to horizontal sectioning and the lack of counterstaining laminar borders could not always be unequivocally determined across all axon ranges. Hence, for quantitative purposes we subdivided the cortical thickness into superficial (L1–3) and deep layers (L4–6).
Bouton clusters of single cells were identified by partitioning the three-dimensional bouton distributions using a custom-made MATLAB script based on the mean-shift algorithm (Binzegger et al. 2007. For details, see Karube and Kisvárday 2011). In brief, appropriate kernel size for the algorithm was determined by the similarity index (SI; Gusfield 2002; Giurcăneanu and Tăbuş 2004; Binzegger et al. 2007). In the first calculation the kernel size was set to 10 µm, and then increased in 5-µm steps. Then, SI between each consecutive calculation was obtained as 1 − m/(u − 1), where m is the minimum number of points to be eliminated for making two partitions equal, and u is the number of total boutons. Then the kernel size (k) was specified which provides stable SI (>0.975 for longer than 15 µm range). During calculation, we simply used the distance between the centers of bouton clusters, k/2, as threshold to be merged the clusters. The value of 10−3 k was used for the stop threshold of the mean-shift algorithm. Although our analysis resulted in a relatively larger number of clusters per cell than in the study of Binzegger et al. (2007), our algorithms and the applied calculations were much simpler to use, and basically providing similar results (see Fig. 6, Supplementary Fig. 4 in Karube and Kisvárday 2011).
Angle difference (∆ori) of preferred orientation between parent soma and the boutons was calculated on the basis of their location within the orientation map from which frequency distribution of ∆ori was generated using 10° binning. For comparing the results with previously published findings, ∆ori was additionally binned into three categories representing, respectively, iso (0°–30°), oblique (30°–60°) and cross-orientation (60°–90°) (Kisvárday et al. 1994).
To determine the exact laminar extent and sublaminar organization of L6 histochemical detection of thalamocortical terminals, vesicular glutamate transporter 2 (VGluT2) was employed (Kaneko et al. 2002; Nahmani and Erisir 2005) on tissue samples taken from the primary visual cortex of three normal adult cats. For this purpose, the fixative contained a mixture of 2 % PF and 0.05 % GA in 0.1 M PB. Tissue block containing areas 17, 18, and 19 was dissected and 20-µm-thick coronal sections cut on vibratome. Sections were immersed in 15 % then 30 % sucrose in PB and freeze-thawed twice in liquid N2. In order to reduce non-specific staining (endogenous peroxidase activity), sections were washed with 0.05 M TBS and treated with 1 % H2O2 in TBS for an hour. Primary antibody to vesicular glutamate transporter 2 (VGluT2, Chemicon International, Temecula, CA) was used (1:5000 in TBS containing 5 % normal goat serum, 2 % bovine serum albumin and 0.5 % Triton X) for 24 h at 4 °C or overnight at RT. After 3 × 10 min rinses in TBS, the sections were incubated with secondary antibody (biotinylated anti-guinea pig antibody (Vector), 1:200) for overnight at 4 °C or 4 h at RT. VGluT2 immunoreactivity was visualized using standard ABC/DAB protocol (see above). The sections were washed in PB, and dry-mounted on a gelatin-coated glass slide, dehydrated and defatted in ascending series of ethanol and xylene and coverslipped in Depex (SERVA Electrophoresis GmbH, Germany).
For double immunohistochemistry, a mixture of anti-VGluT2 and mouse anti-parvalbumin (1:2000, PV235, Swant, Bellinzona Switzerland) was used as primary antibodies. VGluT2 labeling was revealed with biotinylated anti-guinea pig antibody (Vector) using standard ABC/DAB reaction in the presence of CoCl2 resulting in dark blue end product. Then, the sections were treated with 2 % NaN3 for 10 min followed by 2 % H2O2 and 0.07 % p-cresol in TBS for 15 min to inactivate peroxidase and complete DAB oxidization. For revealing PV labeling, biotinylated anti-mouse IgG antibody (Vector, 1:200) was used as secondary antibody. Here the labeling was developed with ABC/DAB in the absence of CoCl2, providing a brown color of the chromogen.
For quantification of VGluT2-positive terminals, light micrographs were taken with an Olympus BX50 (Olympus, Tokyo, Japan) or Leica DMR microscopes equipped with a digital camera (Optronics, Cremona Drive, CA, USA) at 400× magnification. The number of VGluT2-positive terminals in L6 was counted and evaluated using Neurolucida and StereoInvestigator (MBF Bioscience), and Image J (National Institute of Health, USA) software. To better identify laminar and area borders Nissl- and cytochrome oxidase staining (Boyd and Matsubara 1996) of adjacent sections were also used.
Statistical significance among three groups was examined using pairwise Wilcoxon test with Holm’s correction or post hoc Tukey test following ANOVA. For comparison between two groups, Mann–Whitney U test was applied. Kolmogorov–Smirnov test was used for comparison of cumulative frequency distribution curves. For categorical data, Fisher’s exact test was used. All data are given as mean ± SD.
Morphological types of L6 pyramidal cells
All but one spiny cell in our sample had a distinct apical dendrite and they were identified as pyramidal cells. For the one spiny cell, only short dendrites emerged of which the longest issuing from the pial side of the cell body was considered as apical dendrite (Fig. 2. Nr. 12042). As shown in Figs. 1 and 2, the morphology and spatial extent of apical dendrites varied considerably from cell to cell not only in length, but the distribution of daughter branches (oblique branches and tufts). To quantify these differences three morphological parameters were measured: path length, vertical distance to dendritic endings, and vertical distance to nodes, i.e., dendritic branch points (Fig. 1b2–d; see also “Materials and methods”). Examples of path length distribution are shown in Fig. 1b2. The histograms well reflect the actual dendrite morphology that depends on the spatial distribution of dendritic branches. Cells with oblique dendritic branches close to the soma had a relatively large proportion of short path values (middle row in Fig. 1b), whereas cells possessing apical dendritic tufts had a relatively large proportion of long path values (bottom in Fig. 1b). Because path length represents three-dimensional distances, it reflects not only laminar distribution of dendritic branches but also informs about their horizontal extent. The two other parameters which were used for quantitative grouping reflected purely the radial extent of apical dendrites. Vertical distance to nodes was defined as the distance parallel to the pia–white matter axis from the origin of the apical dendrite to each dendritic branch point (node) and vertical distance to endings as the vertical distance from the origin of the apical dendrite to each apical dendritic ending (terminal). Scatter plots in Fig. 1c shows the proportion of path length longer than 600 μm against the median of vertical distance to nodes and Fig. 1d the median of vertical distance to nodes against the 75 % percentile of the vertical distance to tips. In both scatter plots different shapes of apical dendrites illustrated in Fig. 1b1 were segregated.
We investigated sublaminar distribution of our sample of L6 cells. The mean distance from the top of L6 was 353 ± 181 μm for Type Aa, 230 ± 174 μm for Type Ab and 240 ± 190 μm for Type B. Since the distance values did not differ significantly among cell types (p = 0.55 by one-way ANOVA) their distribution can be regarded as unbiased.
Axon distribution of L6 cell types
Each apical dendritic type could be associated with a characteristic axon distribution as shown in Fig. 3 (for drawings see Fig. 2). With regard to horizontal extent, Type Aa and Ab had a large proportion of axons close to the parent somata, i.e., 82.5 % of axon length for Type Aa and 69.9 % for Type Ab were located within 500 μm around their somata, which, however, decreased rapidly at larger distances (Fig. 3a). Contrary to this, the lateral distribution of axons of Type B contained only less than half (43.1 %) of axon length within 500 μm from the somata. In fact Type B axons had the widest lateral extent beyond 1 mm. Here the differences of horizontal distribution of axons were statistically significant among all three types (pairwise Wilcoxon test in Fig. 3a1 and Kolmogorov–Smirnov test in Fig. 3a2). Considering the wide lateral extent of the axon and the long apical dendrite of Type B cells, they resembled the so-called “horizontal” L6 cells reviewed by Briggs (2010). On the other hand, the amount of radially distributed axons also differed among types as shown in Fig. 3b. Type Aa had a larger proportion of axons above the cell body level which reached the peak around 500 μm above the soma or corresponding L4 and L5. Below the cell body their proportion decreased rapidly. For the other two types, the overall proportion of axons was the largest in L6, i.e., in the cell body layer and the overlying L5. It should be mentioned that Type B had another smaller peak more superficial depth, around 700–800 μm above the cell body (Fig. 3b1). Radial distribution of axons differed between Type Ab and B, and Aa and Ab in cumulative proportion (Kolmogorov–Smirnov test in Fig. 3b2). To sum up the above findings, Type Aa axons which had a rather columnar termination field (to some extent Type Ab also) preferred L4 and L5, whereas Type Ab preferred L6. On the other hand, Type B had a considerably wider axonal field extending over 1 mm and was dense in deep layers mainly, but was also present in superficial layers. Important to note that the above quantitative results were consistent with the results of a similar analysis carried out for boutons instead of axon segments (Supplementary Fig. 4).
It is tempting to compare the above morphological properties of both dendrites and axons with earlier reports of L6 spiny cell morphology (Katz 1987; Hirsch et al. 1998). In this regard, our Type Aa neurons can correspond to those projecting into the LGN (cortico-thalamic cells), Type Ab to other cortical areas (cortico-cortical cells), and Type B neurons those projecting into the claustrum (cortico-claustral cells). Although, the possibility remains that as Katz’s Type 2 cells (Katz 1987), all types described here may contain non-projection cells. Nonetheless, multiple projection types of cells cannot be excluded either. Next, the axonal distribution of each type to functional architectures was investigated.
Bouton clustering of L6 cells
In the mammalian visual cortex, boutons of many superficial layer (L2/3) pyramidal cells are known to terminate in distinct clusters forming patchy projection of the axons (Rockland and Lund 1982; Gilbert and Wiesel 1983; Rockland and Lund 1983; Martin and Whitteridge 1984; Kisvárday and Eysel 1992; Binzegger et al. 2007). L4 neurons also provide clustered projections although with different spatial constraints compared to their superficial layer counterparts (Karube and Kisvárday 2011). However, there is no comparable data for deep layer neurons despite the fact that in a few studies clustered axons provided by L5 and L6 cells have been described (Gabbott et al. 1987; Binzegger et al. 2007).
Orientation distribution of boutons
One of the main goals of this study is to determine how boutons of L6 axons are distributed in orientation maps. Because the intrinsic signals to the orientation map derived mainly from superficial layers of the cortex (estimated layer 4 and above; see Tian et al. 2011) and that L6 cells examined here were located at least 1 mm below this zone, a set of strict criteria had to be imposed as to which of the labeled cells are selected for this study. Along this line we sought evidence to clarify that RF properties were preserved through the entire cortical thickness including L6. Unit recordings were examined in the area where optical imaging was carried out and found that only a subtle shift of the RF center position along the same recording track perpendicular to the cortical surface (average: 1.56° ± 2.38°, N = 71 penetrations) was present. Orientation preference derived from imaging showed a good match with that of unit recordings (average difference is −2.6° ± 32.4°, N = 65 penetrations; Supplementary Fig. 6).
Next, we compared the distribution of iso- and cross-orientation preferred boutons with regard to laminar location and horizontal extent from soma origin. To investigate this point, the proportion of iso- and cross-orientation preferred boutons was plotted for the two main cortical tiers, upper (L1–L3) and lower (L4–L6) tiers, as a function of lateral separation from soma location (Fig. 6). For lower tier boutons (N = 21 cells), iso-orientation preference was dominant near to the parent soma (<200 μm lateral), although a substantial proportion of boutons showed cross-orientation preference for Type Aa and Ab cells. For the same cell types, at larger lateral distances, (>250 μm, i.e., beyond lateral extent of their basal dendrites), the proportion of iso-orientation preferred boutons decreased steeply until the proportion of iso- and cross-orientation preferred boutons were near equal, implying an orientation-independent distal projection. On the contrary, for Type B cells, the proportion of iso-orientation preferred boutons remained higher than cross-orientation preferred boutons across long horizontal distances (>1 mm from soma; right column). In all cells, upper tier boutons were less numerous than lower tier boutons. For the 15 cells, which had axon collaterals in upper tier, an average of 391 ± 323 boutons per cell was found, while remaining 8 cells did not have upper tier boutons. The proportion of upper tier boutons to all boutons of the respective cell type was 4.8 % for Type Aa, 21.2 % for Type Ab, and 28.7 % for Type B (average 17.1 % of all boutons, range 0–57 %). It should be also mentioned that only those cells were included in this analysis which had >70 % of their upper tier boutons in the reliable map region (2 of Type Aa, 3 of Type Ab, and 2 of Type B). For these upper tier boutons no clear orientation bias could be found probably due the small sample size of the available boutons. Another point that needs to be mentioned here is that oblique-orientations (omitted for clarity of graphs in Fig. 6) had consistently represented mean values between those of iso- and cross-orientation.
Convergence of boutons from nearby cells
L6 is the deepest and often the thickest layer where columnar organization—both functional and structural—can be disturbed by two main factors: closeness of the underlying white matter and curvature of the cortex. In a flat cortical region fiber bundles of the white matter enter the grey matter typically perpendicular to the cortical surface. However, curvature of the cortex can distort columnar organization in deep layers rendering any functional and structural interpretation difficult here. This is particularly true for interpreting intrinsic signals, which derive largely from upper cortical layers therefore extrapolating such signals to L6 has to be treated with caution. Primary visual cortex of the cat is situated chiefly along the lateral gyrus, which comprises a centrally located rostro-caudally elongated flat zone on the top of the gyrus. Here, the columnar organization of the cortex is rather regular as confirmed by unit recordings across all layers (see “Materials and methods”; Murphy and Sillito 1986; Karube and Kisvárday 2011). We used only this flat zone for labeling L6 spiny cells to investigate correlation of bouton distribution with orientation map.
The reconstructed cells studied here were labeled using extracellular (see “Materials and methods”) tracer injections that raises the question as to completeness of the labeling. We used only those cells which showed strong labeling in the soma and the dendrites as well as the axons to their ends. However, a common drawback of extracellular labeling is that some of the labeled axon collaterals cannot be traced due to the high density of other labeled structures at the center of the injection site. We assume that this feature can explain why the number of boutons in our L6 sample (mean 1399 ± 584; range 640–2878) is smaller than expected on the basis of intracellularly labeled L6 cells for which all axons can be readily traced (Binzegger et al. 2004). Another explanation could be that our cell sample simply lacked certain types of L6 cell possessing large number of boutons. However, the latter explanation is unlikely because the mean dendritic length of both apical and basal dendrites (8.34 ± 2.49 mm, range 5.54–17.86 mm) of our L6 sample was almost the same as that reported by Binzegger et al. (2004) indicating that both studies analyzed comparable cell populations.
Types of L6 spiny cells
The morphological features of the L6 cell types analyzed here are consistent with previously published data obtained in the cat (Gilbert and Wiesel 1979; Lund et al. 1979; Martin and Whitteridge 1984; Katz 1987; Lorente de No 1992; for review see Briggs 2010; Thomson 2010). With regard to the target regions L6 excitatory cells are heterogeneous of which about 50 % project into the LGN, >10 % into the claustrum and about 20 % are cortico-cortical (Gilbert and Kelly 1975; Harvey 1980a, b; LeVay and Sherk 1981; McCourt et al. 1986; Katz 1987; Hirsch et al. 1998). The above L6 types differ not only in their projection target areas, but also can be distinguished on the basis of morphological features such as dendritic branching pattern and axon distribution. Katz (1987) used the Sholl analysis for comparing dendritic morphology of pyramidal cells projecting into different target areas. Although no statistical values were reported, the results revealed minor differences between projection types (see Fig. 17 in Katz 1987). Here we also employed the Sholl analysis and found subtle differences for basal dendrites but only at the proximal parts (p < 0.05 between Type Aa and Type B, and between Type Aa and Type Ab at the range of 10–30 μm form the soma, paired Wilcoxon test with compensation for multiple comparisons). Apical dendrites differed significantly (p < 0.05) among types at the ranges of 70–190, 310–410 and 720–780 μm from the soma (Supplementary Fig. 3). A possible interpretation of the morphological similarity of basal dendrites among types is that they sample a rather similar input, whereas apical dendrites are exposed to a wide range of input of which each type is engaged with a particular set of inputs.
In addition to morphological attributes L6 cells show rather distinct physiological characteristics: most LGN projecting L6 cells have simple RF characteristics (Gilbert 1977; Henry et al. 1979; Harvey 1980a, b; Martin and Whitteridge 1984; Grieve and Sillito 1995; Hirsch et al. 1998), although a subpopulation of them display complex RF features and additionally project into the extrageniculate and perigeniculate nuclei (Tsumoto and Suda 1980; Murphy and Sillito 1996; Murphy et al. 2000; Jones 2007). Claustrum projecting cells are typically monocular and have simple-like RFs with extreme length (10° or longer, Grieve and Sillito 1995). On the basis of distinct morphological signatures, here we make an assumption that our Type Aa corresponds to LGN projecting simple cells and Type B to claustrum projecting cells. On the other hand, Type Ab shares features with those cortico-cortical projecting cells which typically show complex RFs structure and restrict their lateral connections chiefly to infragranular layers (Hirsch et al. 1998). Hence, it is deemed that our cell sample represents the major neuron types of the otherwise highly heterogeneous L6 cell population (for reviews, Briggs 2010; Thomson 2010).
For depth distribution of our L6 cell samples, there was no significant difference among the three types. In the primate and the tree-shrew visual cortex, L6 can be subdivided into zones which parallel the termination of subcortical input fibers (Hubel and Wiesel 1972; Lund and Boothe 1975; LeVay and Gilbert 1976; Blasdel and Lund 1983; Fitzpatrick et al. 1983; Braak 1984; Conley et al. 1984; Usrey and Fitzpatrick 1996). In the cat, such a stratification of the L6 circuitry is still elusive. The distribution of VGluT2-positive terminals within L6 was only slightly different across the full depth of L6 (Supplementary Fig. 8) and does not support a clear stratification within L6. Altogether, the distribution of our reconstructed cells corroborate results of previous studies in finding no marked difference between the cell population of upper and lower subzones of L6 (Lund et al. 1979; McCourt et al. 1986). Contrary to anatomical data the few functional studies that are available pointed out the existence of subzones within L6. Hirsch et al. (1998) reported that deep zone of L6 contains mainly complex cells, while the study of Bullier and Henry (1979) showed that the proportion of complex cells is moderately larger in the upper half of L6 than in the lower half. Obviously, to disclose whether neurons of the same or similar phenotype may possess distinct functional properties and whether they are organized in distinct subzones of L6 further studies are necessary.
Comparison of bouton clustering between L6 and L4 cells
Basic properties of bouton clusters in layer 4 and 6 pyramidal cells
Number of total clusters
5.9 ± 3.5
5.1 ± 2.4
442.5 ± 328.2
394.2 ± 283.7
0.169 ± 0.199
0.196 ± 0.246
Distance from the soma to the cluster center
850.0 ± 574.6
711.2 ± 439.0
L4 > L6*
1st rank cluster
0.512 ± 0.195
0.629 ± 0.208
316.4 ± 337.3
520.9 ± 423.4
900.8 ± 366.8
771.7 ± 226.3
2nd rank cluster
0.253 ± 0.085
0.191 ± 0.105
L4 > L6*
777.9 ± 378.1
672.9 ± 319.1
577.0 ± 203.5
491.6 ± 265.8
0.59 ± 0.30
0.38 ± 0.27
L4 > L6*
Bouton overlap between neighboring cells in L4 and L6
Bouton overlap is an emerging result of our experimental approach that can be used for estimating to what extent cortico-cortical inputs converge at similar or dissimilar orientations. Here we took advantage of a previous study that had focused on the orientation distribution of L4 spiny cells, some of which also had spatially overlapping axons (Karube and Kisvárday 2011). By employing identical analysis, it is seen that bouton overlap between L4 spiny cells depends weakly on orientation preference (see Fig. 9b, c), whereas bouton overlap between L6 cells does not (see Fig. 8b, c). Since on average, there is a dominance of iso-orientation preferred boutons of L6 neurons almost at all lateral distances from soma location (see right column in Fig. 6), it is surprising that their bouton overlap between L6 cells lacks orientation dependence. However, when the data are inspected in more detail it turns out that bouton overlap is more extensive between cells which are located closer to each other or between those whose parent soma and boutons occupy iso-orientation (Supplementary Fig. 7a). Interestingly this is not the case for L4 cells, which in turn show a bias to overlapping boutons which are far from their parent soma and represent cross-orientation (Supplementary Fig. 7b). The complexity of bouton overlap observed for L6 cells is probably due to different spatial organization of connections established by the different cell types. It is conceivable that Type Aa and Ab axons whose cross-orientation preference increases with distance account for most of the non-iso-orientation biased bouton overlap. On the other hand, this is compensated by an iso-orientation biased bouton overlap typically by Type B cells. Consequently, the L6 population data for bouton overlap show no correlation with orientation preference. Inevitably, it would have been useful to carry out a cell type related quantitative analysis of bouton overlap. However, there was no sufficient number of labeled Type B cells in our sample, hence a cell type-dependent analysis of bouton overlap could not be carried out. Finally, it should be mentioned, that due to geometrical constraints and the distance of L6 from the imaged cortical surface, we cannot exclude the limited accuracy of alignment between the orientation map and the anatomical reconstruction that is inherent to the method used here.
Functional implications of L6 cell types
As a population, L6 cells showed an iso-orientation bias, similar to other layers (Fig. 5). From the circuitry point of view, L6 cells are implicated at least in two functional streams: the major intracortical input sources to L4 on the one hand and as one of the output sources to subcortical regions, mainly LGN and claustrum, on the other hand. Electron microscopy of L6 cells estimated that they account for 20–40 % of all synapses on L4 spiny stellate cells (Ahmed et al. 1994, 1997; Binzegger et al. 2004; Tanaka et al. 2011; Pichon et al. 2012). In addition to this, L6 outputs were shown to activate inhibitory neurons too (West et al. 2006; Olsen et al. 2012; Bortone et al. 2014), although the relative proportion of the targeted excitatory and inhibitory cells is still a matter of debate (McGuire et al. 1984; Somogyi 1989; Ahmed et al. 1994, 1997; Staiger et al. 1996). Importantly, the functional consequence of these findings is that facilitatory and suppressive inputs must go hand in hand and mutually modify the activity level of the entire L4 circuitry through which they may contribute to emerging novel receptive field properties. Here we showed that Type Aa (putative LGN-projection) cells provide columnarly arranged, chiefly iso-orientation input to L4, and to a lesser extent, Type Ab (putative cortico-cortical) cells also do so (Figs. 3, 5). On the other hand, the same L6 cells can contact a relatively large proportion of cross-orientation sites in L4 whereby sharpening of orientation preference possibly via lateral inhibition can be envisaged (Crook et al., 1991; Crook and Eysel 1992). Another functional interpretation of the relatively rich input from L6 to L4 is not necessarily to affect spatial tuning characteristics such as orientation preference. For example, recent data obtained in rodent visual cortex indicate that L6 cells can improve signal to noise ratio and control synaptic gain in L4 (Olsen et al. 2012; Bortone et al. 2014). It is also conceivable that L6 connections are engaged with temporal characteristics of RFs without changing spatial tuning characteristics as was found in the monkey visual cortex (Ringach et al. 1997). Clearly, the many different modes of operation of how L6 may act on L4 by modifying or even generating novel RF properties (see below) indicate innate sophisticated mechanisms in elaborating orientation-specific excitatory responses of first order L4 cells. Indeed signatures of the above plastic changes, either facilitatory or suppressive were detected in slice preparation showing small amplitude and high variance in the synaptic transmission involved in L6 and L4 cells which, nonetheless, differed between the two projection types, i.e., cortico-cortical vs thalamocortical projecting L6 cells (Stratford et al. 1996; Tarczy-Hornoch et al. 1999; Mercer et al. 2005; West et al. 2006).
Regarding the functional underpinnings of different types of L6 cell the most specific interaction studied so far concerns the generation of length tuning. Inspired by the very long RFs (8°) of a substantial proportion of L6 cells (Gilbert 1977) it was suggested that they are implicated in the generation of length tuning property, i.e., end-inhibition, of “hypercomplex” cells (Bolz and Gilbert, 1986) via L4 inhibitory neurons (see McGuire et al. 1984). However, data from other laboratories casted doubt on the significance of L6 neurons in generating end-inhibition because silencing of L6 by local iontophoretic application of either GABA or muscimol (a potent GABAa receptor agonist) revealed a decreased responsiveness to the optimal, short stimulus in L4 cells rather than an increased responsiveness to non-optimal, long stimulus (Grieve and Sillito 1991a, b). Further support to the latter observations was provided by postembedding GABA-electron microscopy which showed a smaller proportion of L6 boutons terminating on L4 GABAergic dendrites (14 % in cat: in Somogyi 1989; 32 % in rat: in Staiger et al. 1996) than had been reported solely on the basis of ultrastructural features (93 % in cat: in McGuire et al. 1984). The present anatomical findings and those of former intracellular studies (Gilbert and Wiesel, 1979; Martin and Whitteridge 1984; Katz 1987; Hirsch et al. 1998) are, in fact, more consistent with the conclusion reached by Grieve and Sillito (1991a, b). Firstly, Type Aa cells have a spatially restricted projection into L4 and it is tempting to speculate that they represent the population with an RF size too small for exerting length tuning (Grieve and Sillito 1995). Secondly, Type Ab and B cells which do have relatively long horizontal axons and presumably possess long RFs do not typically send a rich input to L4 (see Figs. 2, 3). It is noteworthy that despite the aforementioned contradictions the layer 6–4 circuitry does contain elements which are in line with the notion of a putative role in length tuning. Consequently, for a thorough understanding of the above process future studies shall employ, for example, in vivo pair-recordings of L6 and L4.
As mentioned above an estimated 50 % of L6 cells have unusually long RFs summing to 8 degrees or more (Gilbert 1977; Bolz and Gilbert 1986, 1989; Grieve and Sillito 1995). In this regard a reasonable assumption is that Type Ab and B population providing the longest axons may have the longest RFs. From this it follows that according to the orientation distribution of Type Ab and B axons quite a mixed effect can be elicited on the recipient cells in L5, i.e., local- and distal (>1 mm) axon collaterals could activate iso-orientation sites whereas in the 400–800 μm range also non-iso-orientation sites (see Fig. 5b2, 6). Because in the cat visual cortex the targeted L5 cells can project to other cortical layers as well as several subcortical structures (Hübener et al. 1990; Kasper et al. 1994), including superior colliculus (Garey 1965), thalamic nuclei (LeVay and Sherk 1981) and visual striatum (Khibnik et al. 2014), the most obvious scenario is that the above orientationally mixed effect can (through the L5 network) modify the activity of several brain regions.
Of the L6 cell population an estimated 4–10 % is known to project into the visual claustrum where RFs can be as long as 40° (Creutzfeldt et al. 1980; Sherk and LeVay 1981). The present data provide support for a two-step mechanism how these large and orientation-specific receptive fields are established in the claustrum. First, a spatial convergence of long-range axons deriving from L5 and L6 spiny cells onto Type B would generate an RF length that often exceeds 10° or longer. Second, Type B axons, which establish iso-orientation connections more frequently than any other L6 cell types (Fig. 5) forward an iso-orientation based input on claustral cells whereby a several fold increase in RF elongation is achieved in the targeted cells. It remains to be seen by future studies for what type of visual processing the very long RFs are utilized by the claustrum (for review, see Crick and Koch 2005; Smythies 2015).
Taken together the present data revealed a diverse relationship between the different morphological types of L6 cells and orientation selectivity. The integrative nature of L6, being one of the main output layers of the visual cortex and its widespread relation in receiving input from other layers render L6 a versatile stage in processing visual information.
Authors thank the three anonymous reviewers and Dr. Julian ML Budd for helpful comments on the manuscript. This work was supported by MTA-Debreceni Egyetem, Neuroscience Research Group and the European Union (Grant number FP7-2010-IST-FETPI 269921) to ZK.
- Bonhoeffer T, Grinvald A (1996) Optical imaging based on intrinsic signals. The methodology. In: Toga AW, Mazziotta JC (eds) Brain mapping: the methods. Academic Press, San Diego, pp 55–97Google Scholar
- Boyd JD, Matsubara JA (1996) Laminar and columnar patterns of geniculocortical projections in the cat: relationship to cytochrome oxidase. J Comp Neurol 365:659–682. doi:10.1002/(SICI)1096-9861(19960219)365:4<659:AID-CNE11>3.0.CO;2-C PubMedCrossRefGoogle Scholar
- Braak H (1984) Architecture as seen by Lipofuscin stains. In: Peters A, Jones EG (eds) Cerebral cortex, vol cellular components of the cerebral corte. Plenum Press, New York, pp 59–104Google Scholar
- Creutzfeldt O, Mucke L, Bang-Olsen R (1980) Responses of claustral neurons to visual stimulation. Exp Brain Res 41:A10Google Scholar
- Jones EG (2007) The thalamus, 2nd edn. Cambridge University Press, CambridgeGoogle Scholar
- Somogyi P (1989) synaptic organization of GABAergic neurons and GABA receptors in the lateral geniculate nucleus and visual cortex. In: Lam DKT, Gilbert CD (eds) neural mechanisms of visual perception. Portfolio Publishing Co., Woodlands, pp 35–62Google Scholar
- Tanaka YR, Tanaka YH, Konno M, Fujiyama F, Sonomura T, Okamoto-Furuta K, Kameda H, Hioki H, Furuta T, Nakamura KC, Kaneko T (2011) Local connections of excitatory neurons to corticothalamic neurons in the rat barrel cortex. J Neurosci 31:18223–18236. doi:10.1523/JNEUROSCI.3139-11.2011 PubMedCrossRefGoogle Scholar
- Velez-Fort M, Rousseau CV, Niedworok CJ, Wickersham IR, Rancz EA, Brown AP, Strom M, Margrie TW (2014) The stimulus selectivity and connectivity of layer six principal cells reveals cortical microcircuits underlying visual processing. Neuron 83:1431–1443. doi:10.1016/j.neuron.2014.08.001 PubMedPubMedCentralCrossRefGoogle Scholar
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.