Visual Overview
Visual Abstract
Abstract
In the mammalian visual system, three functionally distinct parallel processing streams extend from the retina to the visual thalamus and then to the visual cortex: magnocellular (M), parvocellular (P), and koniocellular (K). Tree shrews (Tupaia belangeri), a preprimate species, provide an advantageous model to study the K pathway in isolation because, while M and P pathways remain mixed in Lamina 1 (L1), L2, L4, and L5 of the lateral geniculate nucleus (LGN), L3 and L6 receive strictly K-input from the contralateral eye. Additionally, K-input laminae selectively receive glutamatergic axons from the superior colliculus. To reveal how cellular and synaptic properties of K geniculate laminae may differ from M/P laminae and how tectal input may shape the K relay to the cortex, we studied the morphology and connectivity of retinal and tectal terminals in pathway-specific laminae. While confirming that K laminae relay cells contain calbindin, we also found its expression in GABAergic cells across all laminae. No cell-type or lamina specificity was observed for parvalbumin. Ultrastructurally, retinal terminals are morphologically distinct in M/P versus K laminae. Tectogeniculate axons in L3 and L6 resemble retinal terminals in their morphology and synaptic targets, while corticogeniculate terminals are sparse in L6. VGluT2, the molecular marker for large-sized driver terminals, is expressed prominently in one of the three tectal cell types that project to LGN. Morphological differences in synaptic circuitry between L3 and L6 provide further evidence that two geniculate K laminae are differentially innervated to relay distinct sets of information to the cortex.
Significance Statement
The current study provides new insights into the unique organization and functional roles of koniocellular (K) pathways in visual processing. Using the tree shrew's unique lamina organization in the lateral geniculate nucleus, where magnocellular (M)/parvocellular (P) and K pathways remain segregated, we reveal structural and neurochemical differences between these processing streams. Our findings are consistent with the idea that two geniculate K laminae are distinct and their circuitry may support adaptive roles, such as motion detection and visual processing and potentially maintaining visual function in conditions like blindsight. This research underscores the K pathway's heterogeneity and distinctions from M/P pathways and its importance in visual processing and adaptability.
Introduction
Anatomical and physiological studies have identified three main types of relay cells that extend from the retina to the cortex in most mammalian species examined: magnocellular (M), parvocellular (P), and koniocellular (K) in primates (Kaas et al., 1978; Irvin et al., 1986; Kaplan, 2003, 2014) and X, Y, and W cells in other mammals (Enroth-Cugell and Robson, 1966; Friedlander et al., 1981; Holdefer and Norton, 1995; Van Hooser et al., 2003). In most species, the lateral geniculate nucleus (LGN) is organized into laminae which are devoted to one or more parallel processing streams (Campbell et al., 1967; Kaas et al., 1978; Holdefer and Norton, 1995; Van Horn et al., 2000; Kaplan, 2003, 2014; Kerschensteiner and Guido, 2017). The geniculate neurons are located in pathway-specific laminae and may show distinct morphological and response property characteristics that reflect the RGC origin of the primary, driving input, indicating functionally distinct parallel pathways.
In primates, tree shrews, and some carnivores, parallel processing streams originating from the retina remain segregated in the LGN, although segregation patterns of these parallel pathways vary across species. In the preprimate species, tree shrew (Tupaia belangeri), Lamina (L)1, L2, L4, and L5 all receive mixed projections comprised of both M and P RGC axons, while L3 and L6 are solely comprised of K axons (Conway and Schiller, 1983; Conley et al., 1984; Holdefer and Norton, 1995). Thus, M and P pathways overlap in geniculate laminae, while the K pathway is segregated, which highlights the tree shrew species as a suitable model to study the K pathway in isolation.
Tree shrews also serve as an advantageous animal model because of their LGN laminae organization and parallel pathway segregation. The organization of the tree shrew dLGN parallel pathways contains three pairs of layers (Holdefer and Norton, 1995): L1 and L2, L4 and L5, and L3 and L6. L1 and L2 form a pair that receives ipsilateral and contralateral projections, respectively, from medium- to large-sized RGCs and display ON center receptive field properties. L4 and L5 form a pair that receives contralateral and ipsilateral projections, respectively, from medium- to large-sized RGCs and display OFF center receptive field properties. The only layers that receive input from the superior colliculus (SC), L3 and L6 (Diamond et al., 1991), form a pair that receives contralateral projections from small-sized RGCs and display W-like (K) and ON–OFF center receptive field organization (Conway and Schiller, 1983; Holdefer and Norton, 1995). These patterns reveal that the K pathway is a distinct part of the afferent visual stream in tree shrews with apparent differences from M/P parallel pathways.
In this study, we asked if the morphological, molecular, and synaptic properties of neurons and axonal inputs in K-specific lamina of the tree shrew LGN are different from those in non-K lamina. While three major classes of RGCs have been identified morphologically based on their axon diameter, soma area, and dendritic field (DeBruyn, 1983; Drenhaus et al., 1997; Lu and Petry, 2003), there is no direct demonstration showing that these phenotypic features are maintained in the LGN, although clustering analyses of retinal terminals in mouse LGN suggests that different RGC types may have unique bouton morphology and size (Maher et al., 2023). We utilized immunohistochemical and tract-tracing methods to study the cellular and morphological properties of the parallel processing streams in geniculate lamina. Overall, our findings suggest that there are chemical, synaptic, and morphological distinctions of parallel pathways in the tree shrew dLGN, with evidence that suggests the existence of two subpathways of the K processing stream.
Materials and Methods
Animals
Data for this study were collected from the brains of 10 adult tree shrews (Tupaia belangeri) of both sexes. All procedures were approved by the University of Virginia Institutional Animal Care and Use Committee and were conducted in accordance with the National Institutes of Health guidelines.
Tract-tracer injections
In order to label tectogeniculate (TG) projections via anterograde transport, adult tree shrews were anesthetized with 2–5% isoflurane and placed on a stereotaxic apparatus. An incision was made along the scalp, and a craniotomy was performed above the SC. The stereotaxic coordinates used for SC injections were as follows: AP, −6.0 to 7.5; ML, 2.0–2.5; and DV, 3.5–4.0 mm (Zhou and Ni, 2016). A NanoFil 10 μl syringe [World Precision Instruments (WPI)] containing 300 nl of a 5% solution of biotinylated dextran amine (BDA-10,000; Invitrogen) in saline or 400 nl of undiluted AAV-CAG-tdTomato (Table 1) was lowered into the SC and infused at a rate of 1 nl/s using Legato 130 Nanosystem (KD Scientific). The needle was left at the target for an additional 5 min at the end of the injection cycle to minimize backflow before it was retracted. The craniotomy was sealed with bone wax, the scalp was sutured, and the animals were placed on a heating pad until mobile. After the surgery, animals were monitored for 3 d to ensure proper wound healing and were observed for any behaviors indicative of pain or discomfort.
Antibodies, reagents, and tracers used in this study
In order to label SC cells that project to LGN via retrograde tracers, adult tree shrews were anesthetized with 2–5% isoflurane and placed on a stereotaxic apparatus. An incision was made along the scalp, and a craniotomy was performed above the LGN. The stereotaxic coordinates used for LGN injections were as follows: AP, −3.8 to 4.8; ML, 4.6–4.8; and DV, 5.5–6.5 mm (Zhou and Ni, 2016). A Nanofil 10 μl syringe (WPI) containing 300 nl of undiluted AAVrg-EGFP (Table 1) was lowered into the LGN and infused at a rate of 1 nl/s using Legato 130 Nanosystem (KD Scientific).
For the eye injections, the animals were anesthetized with isoflurane (SomnoFlo; Kent Scientific). A Hamilton 25 μl syringe (Hamilton Company, Reno, NV; #80408) was filled with cholera toxin B (CTB; 5 mg/ml; Thermo Fisher Scientific) conjugated to Alexa Fluor 555 or BDA (5%). The needle was inserted at the nictitating membrane close to the lateral canthus to access the posterior chamber, and 5–10 μl of the tracer was injected.
Tissue preparation
After a minimum of 21 d required for tracer transport, the animals were deeply anesthetized with an overdose of Euthasol (excess of 0.25 ml/kg, i.p.) and transcardially perfused with Tyrode's solution (in mm: 137 NaCl, 2 KCl, 0.9 CaCl2, 1.2 MgCl2, 11.9 NaHCO3, 0.4 NaH2PO4, 5.5 glucose; 281 mOsm), pH 7.4, for 1–2 min, followed by 300 ml of a fixative solution containing 4% paraformaldehyde (PFA) and 0.0 (for fluorescent labeling experiments) or 0.5% (for EM experiments) glutaraldehyde in 0.1 M phosphate buffer (PB), pH 7.4, at 30–35°C. Brains were extracted and postfixed overnight in 4% PFA at 4°C. Subsequently, brains were blocked and sectioned coronally at 50–60 μm on a Leica VT 1000 S vibratome (Leica Biosystems). Sections were then rinsed in 1% sodium borohydride and stored in 0.05% sodium azide (NaN3) in 0.01 M PBS at 4°C prior to immunohistochemistry.
Immunohistochemistry
Table 1 lists the primary and secondary antibodies and the dilutions used in the current study. Sections for confocal imaging were placed in the primary antibody diluted in PBS solution containing 1% bovine serum albumin (BSA), 0.3% Triton X-100, and 0.05% NaN3 for 24–48 h at room temperature (RT) on a shaker. To terminate the incubation, we rinsed the sections in 0.01 M PBS and incubated in a secondary antibody conjugated to Alexa fluorophores for 2 h. Sections were then rinsed, mounted onto subbed slides, and coverslipped with VECTASHIELD Mounting Media.
For electron microscopy, sections were preincubated in 0.01 M PBS containing 1% BSA and 0.1% Triton X-100 for 30 min. Sections were then transferred into a primary antibody diluted in 0.01 M PBS containing 1% BSA and 0.05% NaN3 for 72 h on a shaker. To terminate the incubation, we rinsed the sections in 0.01 M PBS and incubated them in a secondary antibody conjugated to biotin for 2 h. This was followed by an incubation in the avidin–biotin complex (Vector Laboratories) solution for 2 h. Sections were then rinsed in 0.01 M PBS, and the labeling was visualized in a solution of 0.02% hydrogen peroxide (H2O2) and 0.05% diaminobenzidine (DAB) for 2–7 min.
RNA fluorescence in situ hybridization (FISH)
Free floating sections containing the SC were dissected out and incubated in hydrogen peroxide for 10 min in a well plate. After incubation, slices were washed three times for 10 min in PBS to remove bubbles, mounted onto a SuperFrost Plus slide (Thermo Fisher Scientific; catalog #12-550-15) in a square configuration, and left to air-dry. The slide was then baked in an oven at 40°C for 30 min, followed by incubation in 4% PFA at 4°C for 15 min and dehydration in 50, 70, and 100% ethanol at RT for 5 min each. The slide was then air-dried for 5 min at RT, boiled in distilled water at 99°C for 10 s, incubated in target retrieval buffer at 99°C for 5 min, rinsed in distilled water for 15 s at RT, and lastly dehydrated in 100% ethanol for 3 min at RT. The slide was air-dried and left overnight in a closed container at RT.
The following day, RNAscope Multiplex v2 (Advanced Cell Diagnostics, catalog #323270) was used to label specific mRNA transcripts. The slide-mounted sections of the SC were outlined with a hydrophobic barrier, rinsed with distilled water, digested with Protease IV at 40°C for 30 min, and incubated in a Slc17a6 probe for VGluT2 (vesicular glutamate transporter type 2; catalog #1269961-C1) at 40°C for 2 h. The slide was then treated with amplification reagents (AMP 1–3) at 40°C for 15–30 min each. Following this, HRP-C1 and TSA Vivid Dye 650 were applied for 15–30 min at 40°C to bind to AMP 3 and activate fluorescence. The slide was then washed in a series of 0.1 M PBS and distilled water rinses and left to air-dry before being cover slipped with Aqua-Poly/Mount mounting medium (catalog #18606).
Embedding for electron microscopy
Using routine protocols, sections were treated with 1% osmium tetroxide (OsO4) in 0.1 M PB for 1 h. Sections were then treated with filtered 4% uranyl acetate in 70% alcohol for 1 h, followed by dehydration in acetone and treatment with a 1:1 acetone/resin mixture overnight. The following day, sections were transferred to full resin [EMBED 812; Electron Microscopy Sciences (EMS)] and left overnight. Sections were then flat embedded between two Aclar sheets (EMS) and cured in a 60°C oven overnight. Sections of LGN to be used for EM were identified from flat embed sections and photographed with a light microscope. The sections containing the tree shrew LGN were excised and placed in BEEM capsules (EMS). The capsules were filled with resin and cured at 60°C for 24–48 h or until polymerized. The region of interest including laminar borders and the other landmarks such as capillaries and myelinated axon bundles was traced with a camera lucida, and the capsule embedded tissue was trimmed down to a 1 × 2 mm trapezoid containing the entirety of LGN laminae and the optic tract. Ultrathin sections of ∼50–80 nm thickness were collected on 400 mesh copper grids (Ted Pella) using an ultramicrotome (Ultracut UCT7; Leica Microsystems). The geniculate laminae in ultrathin sections were identified with the help of trapezoid orientation over capsule-embed and the tissue landmarks including optic tract border.
Imaging, statistical modeling, and analyses
For confocal microscopy, 60-μm-thick coronal sections were mounted on slides, dried, and coverslipped with Aqua-Poly/Mount mounting medium (catalog #18606). Images were subsequently collected using Leica Stellaris 5 laser scanning confocal microscope using HC PL APO 40×/1.3 NA and HC PL APO 63×/1.4 NA with a maximum voxel resolution of 0.06 × 0.06 × 0.3 µm for highest-resolution images.
For EM images, ultrathin sections on copper grids were examined on a JEOL1010 electron microscope equipped with a 16 megapixel CCD camera (SIA). For quantitative morphology and immunolabeled terminal analysis, overlapping images were captured within laminar borders at the lipid membrane bilayer resolution. Images for quantitative and immunolabeled terminal analysis were taken at 8,000–15,000× magnification, yielding a pixel size resolution of 0.8–1.7 nm. ImageJ (NIH) and the Microscope Measurement Tools plugin were used to quantify the terminal area and synapse length for both unlabeled and labeled terminals. Other morphological characteristics, such as the contrast of mitochondria, the presence of protrusions, the number of synapses made, and the postsynaptic targets were noted as categorical data when possible.
For model-based classification of soma sizes in geniculate laminae, the Gaussian normal mixture modeling feature of the R package MClust (Scrucca et al., 2016), which assumes that distinct subpopulations that make up a heterogeneous population are normally distributed, was used to determine the parameters of subpopulation clusters within the dataset of cell soma sizes across lamina. The selection of a model is determined by the highest Bayesian information criteria (BIC) between models of different cluster numbers. The MClust package then assigns a classification for every object into the determined clusters, along with each cluster's sample size, mean, and standard deviation, which are used to construct distribution curves of model subpopulations.
For three-dimensional object–based confocal colocalization analysis, we designed a new quantitative approach that accounts for object-size variabilities of boutons visualized by tracer-filling of axons versus the immunolabeling of multiple vesicle clusters within a single bouton: First, 3D renderings of terminal boutons were made using the “Surfaces” creation tool in the Imaris 9.5 software (Oxford Instruments). The surface detail was confined to 0.10 µm, and the intensity threshold was adjusted to define object boundaries. To minimize the inclusion of false objects created from background fluorescence, any rendering that was smaller than the volume of 15 voxels (0.0185 μm3) was excluded from analyses. For every object, volume and the centroid X/Y/Z coordinates were collected, and object radius was calculated from its volume. Object information was then funneled into the nearest neighbor search algorithm in the R package FNN (Arya et al., 1998) to obtain the ID of the nearest neighbor of each object and the Euclidean distance between the two objects. Interbouton distance was calculated by subtracting the radii of each object from the distance between them. Any two objects were deemed to be colocalized if their interbouton distance was <0. Percentages of VGluT2+ boutons that are not colocalized with a retinal bouton are calculated from five images from each lamina; each image has at least 200 boutons. To account for false-positive identification of objects, a colocalization index (CI) was calculated for each image by normalizing the percentage of VGluT2 objects containing CTB to the percentage of CTB objects containing VGluT2, as VGluT2 is assumed to be present in every retinal (i.e., CTB) terminal bouton. A CI value of 1.0 indicates that every VGluT2+ object also contained CTB, whereas a CI value of <1.0 indicates the presence of singly labeled VGluT2 objects.
Adobe Creative Cloud Photoshop was used for compiling figures, including pseudocoloring and annotation of EM images. For graphs and statistical analysis with nonparametric testing, including unpaired Mann–Whitney U (MW-U) tests, Kolmogorov–Smirnov (KS) and Kruskal–Wallis (KW) comparisons, and descriptive statistics, the GraphPad Prism software (GraphPad.com) was used. Table 2 lists all statistical parameters.
Statistical parameters
Results
Characterization of neurons in K versus M/P laminae
Relay cell differences across geniculate laminae
In order to reveal if the cytoarchitectural differences across LGN laminae lend evidence for the cell-type specificity in M/P versus K pathways, we have quantified the sizes of cell somas that are outlined by neuronal markers. Retinogeniculate (RG) axons, labeled by an injection of CTB into the eye, as well as axon boutons stained for VGluT2, were used to delineate laminar borders (Fig. 1a). For measuring soma sizes, we weighed the advantages and limitations of a histochemical stain NeuroTrace (NT; Thermo Fisher Scientific) and the antibody based NeuN staining to visualize geniculate cells. While selective for neuronal nuclear protein, NeuN stain inherits limitations including dependence on antibody penetration and potential exclusion of some neuron types in the brain (Kumar and Buckmaster, 2007). On the other hand, NT reliably labels all cells with Nissl bodies, including glial cells (García-Cabezas et al., 2016). For our purposes, NT proved preferable because glial cell somas can be excluded based on size criteria. To that aim, we measured the sizes of geniculate GABA+ somata (n = 878; >100 cells per laminae; 125.2 ± 1.3 μm2; Fig. 1h) and determined that 99% of GABAergic cells were larger than 40 μm2, which is larger than glia measured in various species (Stolzenburg et al., 1989; Rajkowska et al., 1998; Verdonk et al., 2016; Liu et al., 2022). Any NT cells smaller than this cutoff value were classified as GABA+ glial cells and excluded from soma size analyses.
Cellular characteristics of M/P versus K dLGN laminae. a, Schematic for delineation of geniculate laminae by anterograde tracers from eye injections (CTB, blue), immunostaining for VGluT2 (red), GABA (magenta), and NT (cyan). b, c, Sections dually stained for NT and GABA were used to quantify soma sizes of relay cells (NT + GABA−, cyan) and interneurons (NT + GABA+, magenta). Small glial cell bodies (yellow) were excluded from analyses. Scale bar, 20 μm. d, e, Frequency distribution histogram of soma sizes for all NT+ cells collected in L1–L6 that display two distinct peaks (d). When separated out by M/P (blue) versus K (maroon), the first peak, presumably interneurons, lines up, while larger-sized, relay cell populations reveal differences in their distributions (e). f, Frequency distribution of soma sizes in M/P laminae with overlaid normal distributions of four subpopulations modeled via BIC. Inset shows the BIC for models with different numbers of components. The dashed vertical line marks the model with the highest BIC number. g, Frequency distribution of soma sizes in K laminae with overlaid normal distributions of three subpopulations modeled via BIC. Inset shows the BIC for models with different numbers of components. The dashed vertical line marks the model with the highest BIC number. h, Collapsed across lamina, all relay cells were significantly larger than all interneurons (p < 0.0001). i, Distributions of soma sizes for relay cells and interneurons, sorted by lamina. In each lamina, relay cells are significantly larger than interneurons (p < 0.0001). j, Relay cell soma size distributions, sorted by pathway. Within each lamina pair, there are no significant differences in the soma size. Relay cells in M/P L1/2 and M/P L4/5 are significantly larger from those in K L3/6. ***p < 0.0001.
At exploratory size frequency distribution histograms (Fig. 1d,e), the NT+ neuron somata (n = 2,414) range between 40.2 and 523 μm2 and display two major peaks representing the size differences between interneurons and relay cells. When neurons in M/P and K laminae are plotted separately (Fig. 1e), the first, presumed interneuron, peaks align, whereas larger-sized, relay cell populations reveal nonoverlapping distributions with discrete peaks and thus potential differences.
To elucidate for the multiple cell subpopulations that contribute to distributions in M/P and K laminae, we used a statistical modeling approach, Mclust, which iteratively applies BIC to reveal the best fit number of normally distributed subpopulations that make up a multimodal distribution (Scrucca et al., 2016). MClust analysis revealed four distinct populations in M/P laminae and three distinct populations in the K laminae (Fig. 1f,g). When modeled distributions of MClust subpopulations were superimposed on the M/P and K laminae measured distributions, several findings were revealed: First, the smallest subpopulation in each laminae category completely accounts for the interneuron peaks, and these are not statistically different than each other. Thus, interneuron soma sizes in M/P and K laminae are similar. Second, M/P laminae are composed of three distinct cell types, potentially corresponding to M and P cell types in ON and OFF geniculate laminae of the tree shrew LGN with distinct soma morphology (Friedlander et al., 1981). Third, K laminae are also composed of two distinct cell types. Cell size distributions between K L3 and K L6 show similar trends and do not significantly differ, so this finding may support the existence of multiple K relay cell subtypes.
Comparison of relay cells and interneurons across geniculate laminae
In order to identify cell body differences across LGN laminae, we used histochemical and immunocytochemical markers to distinguish relay cells (NT+ and GABA−) from interneuronal (NT+ and GABA+) cell bodies (Fig. 1b,c). As relay cells are encountered more frequently than interneurons in the LGN, we employed a sampling strategy that aimed to collect at least 100 cells of each type from each lamina, thereby oversampling interneurons. Thus, the dataset is optimized to compare size differences across cell types and laminae, but it is not suitable for revealing the ratio of relay versus interneuron density in any given region. In a dataset of 1,592 cell somas (714 relay cells and 878 interneurons; >100 cells per lamina), the geniculate relay cells displayed significantly larger soma sizes than interneurons (251.7 ± 2.7 μm2 vs 125.2 ± 1.3 μm2; Fig. 1f; MW-U, p < 0.0001). The relay–interneuron soma size differences are also evident in each individual lamina (Fig. 1i; KW; p < 0.0001 for all laminar pairwise comparisons). Because the tree shrew LGN laminae are uniquely segregated in distinct pathways (L1 and L2, M/P ON; L4 and L5, M/P OFF; and L3 and L6, K ON/OFF), we asked whether pathway-specific differences in relay soma sizes exist within and across lamina pairs. We found no differences within M/P ipsi/contra pairs or within the contralateral K pair. However, relay cells in both ON and OFF M/P laminae are significantly larger from those in K laminae (Fig. 1j; KW; p < 0.0001; pairwise comparisons). Complimented by Mclust results above, this analysis confirms that M/P and K laminae contain distinct relay cell populations.
Calbindin and parvalbumin specificity across geniculate laminae
In order to further characterize the cell-type–specific markers in tree shrew dLGN, we quantified the distribution of the calcium-binding protein calbindin (CALB) and parvalbumin (PARV) across geniculate laminae in four tree shrews (Fig. 2a–c). Previously, CALB has been qualitatively described as a marker for K (or W)-type relay cells in tree shrews and primates (Jones and Hendry, 1989; Diamond et al., 1993; Johnson and Casagrande, 1995; Goodchild and Martin, 1998); however, the exclusivity of this staining neither to relay cells nor to K laminae was confirmed. Our quantitative laminar analysis revealed that CALB-stained cells are present in every lamina (Fig. 2b,g–l); thus, CALB is not exclusive to the K pathway. To elucidate differences in CALB+ relay cells and interneurons in these distinct pathways, the tissue was dually stained, CALB+ cells were sampled across all laminae in seven LGN tissues sections (n = 1,539), and the colocalization of GABA was assessed in those cells (Fig. 2a,b,g–l). K laminae displayed the majority of CALB+ cells (L3, 794; L6, 232) and ∼20% of those contained GABA, suggesting that CALB is utilized not only in relay cells but also in interneurons. In contrast, while CALB+ cells in M/P laminae were not as numerous (50–190 cells), 65–95% of CALB+ somata contained GABA (Fig. 2g–i,m), suggesting that CALB is indeed expressed in the M/P laminae and preferentially in interneurons (Fig. 2m). It should be noted that as these experiments did not include imaging of a third channel for all geniculate cells, they do not reveal whether all K relay cells, or all M/P interneurons, are CALB-positive. Regardless, they provide definitive confirmation that interneurons across all laminae contain CALB; however only relay neurons in K laminae predominantly express this calcium-binding protein.
CALB, PARV, and GABA specificity of M/P versus K dLGN laminae. a–c, GABA (magenta), CALB (green), and PARV (orange) staining across L1–L6. Scale bar, 500 μm. d–f, Triple immunofluorescence staining reveals that PARV is present in all neuronal cell bodies, including non-GABAergic relay cells (white asterisks) and GABAergic interneurons (magenta asterisks), but not in small, glial cell types (white arrow) revealed by the Nissl stain (NT). Scale bar, 50 μm. g–l. High (63×)-magnification views from each lamina, showing colocalization (white arrowheads) of GABA and CALB in M/P pairs (g–j) and K pair (k, l). Scale bar, 50 μm. m, The proportion of CALB+ cells that are relay cells versus interneurons in each lamina. Interneurons accounted for the majority of CALB+ cells in M/P lamina pairs, whereas relay cells accounted for the majority in the K lamina pair. *p < 0.01; **p < 0.001; ***p < 0.0001.
PARV, a calcium-binding protein that is more prevalently specific for GABAergic neurons in the brain, has previously been described as a selective marker for M/P layers in both tree shrew and primates (Jones and Hendry, 1989; Diamond et al., 1993; Goodchild and Martin, 1998). In the current study, as expected, immunostaining with anti-PARV antibodies revealed denser staining in geniculate M/P laminae (Fig. 2c). However, this apparent selectivity at low magnifications was merely due to denser labeling of the neuropil in those laminae. Upon closer inspection we found labeling of cells in every geniculate lamina. Furthermore, PARV labeling was not confined solely to GABAergic cells (Fig. 2d–f); both GABA+/NT+ and GABA−/NT+ cells in all geniculate laminae were immunopositive for PARV. Thus, PARV is expressed in every neuron type in all geniculate laminae.
Terminal bouton morphology of RG axons
Ultrastructural morphology of RG projections
In order to characterize the morphological differences of RGC terminals in M/P and K laminae, we examined LGNs after BDA injections into the eye using electron microscopy. Eye injections led to visualization of fibers in all ipsilateral or contralateral geniculate laminae, spanning the entirety of the topographic representation, suggesting that the spread of the tracer across the retina and the transport of tracer in RGC axons were near 100%. A total of 1,203 terminals that contained at least one synaptic zone at the cross section were examined; 596 of these were BDA-labeled RG terminals, and 607 were unlabeled terminals within the same regions. Retinal terminals labeled by the BDA injection exhibited a range of morphologies across all geniculate laminae. Most exhibited the typical morphology previously described (Guillery, 1969; Erişir et al., 1997; Bickford et al., 2010), exhibiting large cross-section areas and thick postsynaptic densities, making multiple synaptic contacts, containing many mitochondria and having protrusions (Fig. 3a,b).
Retinal terminals in K L3 are distinct from all other geniculate laminae. a, b, Electron micrographs of RG terminals labeled anterogradely from the eye (t-BDA) in K L3 (a) and M/P L5 (b). RLP, unlabeled terminal with LM; d-rel, relay cell dendrite; F2-int, vesicle-containing interneuron dendrite; d, dendrite. Scale bar, 1 μm. c, Quantitative comparison of terminal areas of labeled RG terminals (BDA+ RG) and unlabeled terminals (BDA−) in L1–L6. In every lamina, RG terminal cross-section areas are significantly larger than non-RG (p < 0.0001). The inset shows comparison of unlabeled terminals with DM across the three pairs of layers. BDA− DM terminals in the K pair show a greater range in sizes and are significantly larger than BDA− DM terminals in M/P L1–L2 (p < 0.01). d–f, Cumulative frequency distributions of RG terminal areas in each pair of laminae. Within pairs, M/P L1–L2 (d) and M/P L4–L5 (e) have similar distributions, whereas K L3/L6 (f) significantly differ from one another (p < 0.001). g, Comparison of distributions across lamina. K laminae significantly differ from M/P lamina, with L3 differing more so than L6. *p < 0.01; **p < 0.001; ***p < 0.0001.
Quantitative comparison of retinal terminals across geniculate laminae
In order to compare retinal terminals across geniculate laminae quantitatively, we measured the areas of RG terminals labeled after eye injections, as well as the areas of all unlabeled terminals in the same tissue. In all laminae, retinal terminals were the largest in size and were significantly larger than the unlabeled terminals within the same lamina (Fig. 3c; MW-U; p < 0.0001). However, interestingly, unlabeled terminals classified as having dark mitochondria (DM; nonretinal) in the contralateral K pair were the largest on average, ranging from 0.21 to 2.88 μm2 in size (Fig. 3c, inset) and significantly differed from one of the M/P ipsi/contra pairs (L3–L6, 0.88 ± 0.06 μm2 vs L1–L2, 0.65 ± 0.04 μm2; KW; p = 0.003; vs L4–L5, 0.75 ± 0.04 μm2, ns), consistent with the idea that K laminae receive an additional input that bear large terminal boutons that do not display light mitochondrion, the typical morphology of retinal axons.
Across geniculate laminae, the smallest-sized labeled retinal terminals resided in K-recipient L3 (1.30 ± 0.08 μm2), and the largest-sized labeled retinal terminals resided in M/P-recipient L5 (2.75 ± 0.15 μm2). The pair of M/P-recipient ON layers did not significantly differ from one another (Fig. 3d; L1 RG terminals, 2.35 ± 0.11 μm2; L2 RG terminals, 2.52 ± 0.11 μm2). The other pair of M/P-recipient OFF layers also did not significantly differ from one another (Fig. 3e; L4 RG terminals, 2.74 ± 0.17 μm2; L5 RG terminals, 2.75 ± 0.15 μm2). However, the pair of K-recipient ON/OFF layers were the only pair that significantly differed from one another (Fig. 3f; L3 RG terminals, 1.30 ± 0.08 μm2; L6 RG terminals, 1.93 ± 0.10 μm2; p < 0.0001), suggesting that the two K laminae may be involved in distinct functions. Additionally, retinal terminals in K-recipient L3 significantly differed from every other lamina (Fig. 3g; p < 0.001–0.0001). Similarly, the other K-recipient lamina L6 also significantly differed from every other layer, except M/P L1, but to a lesser extent (Fig. 3g; p < 0.01–0.001). Together, these results reveal that retinal terminals in M/P lamina pairs are more similar to each other than they are to the K pair and that retinal terminals in K-input laminae are distinct from all other geniculate laminae.
Ultrastructural morphology of geniculate terminals
To capture the laminar differences in geniculate synaptic circuitry contributed by all projections, we used unlabeled tissue sections collected from one animal and examined the synaptic connections of a total of 781 terminals. We measured 242 synaptic terminals in M/P-recipient L1 (Fig. 4a; 0.84 ± 0.07 μm2), 281 synaptic terminals in K-recipient L3 (0.61 ± 0.04 μm2), and 258 synaptic terminals in K-recipient L6 (0.77 ± 0.05 μm2). Frequency distributions of cross-section areas in these laminae suggest potential differences in contribution of input subpopulations with unique sizes: M/P-recipient L1 contains a greater percentage of large-sized terminals, especially compared with K-recipient L3, while K-recipient L6 contains fewer of the smallest-sized terminals, typically originating from cortex (Fig. 4b).
Differences in input contribution in M/P versus K laminae. a, Electron micrograph of a large terminal, identified by its LM (RLP), making multiple synapses, including onto a relay cell dendrite (d-rel). Scale bar, 1 μm. b, Frequency distributions of terminal cross-section areas in M/P L1 (top, green), K L3 (middle, maroon), and K L6 (bottom, pink). Dashed lines and percentages indicate the proportion of terminals smaller than 0.20 μm2 and the proportion of terminals larger than 1.50 μm2. M/P L1 contains a greater percentage of large-sized terminals, typically of retinal origin, while K L6 contains fewer of the smallest-sized terminals, typically originating from the cortex. c, Cumulative frequency distributions of terminal areas of presumed retinal terminals, identified by their LM, in M/P L1, K L3, and K L6. LM terminals were the smallest in L3. d, Comparisons of terminal area distributions between tracer-identified retinal terminals (RG) and presumed retinal terminals identified by LM. LM terminals only differed significantly from RG terminals in K-recipient L3 (p < 0.001). **p < 0.001.
On nonimmunolabeled EM preparations, putative retinal and nonretinal terminals are traditionally identified based on whether they contain light mitochondria or DM (Erişir et al., 1998; Bickford et al., 2015). We applied the same approach to examine laminar differences among terminals that meet this morphological criterion. While unlabeled terminals with light mitochondria (LM) displayed the smallest mean in K-recipient L3, no statistical differences in their size distribution were evident when compared with L1 (Fig. 4c; 1.83 ± 0.15 μm2 vs 2.55 ± 0.20 μm2; KS, p = 0.10) or L6 (2.08 ± 0.14 μm2; p = 0.37). However, terminals identified as retinal based on their pale mitochondrial appearance differed significantly from tracer-identified RG terminals in K-recipient L3 (Fig. 4d; L3 LM terminals 1.83 ± 0.15 μm2 vs L3 RG terminals 1.30 ± 0.08 μm2; p < 0.001). This suggests that either retinal terminals do not account for all terminals with LM or that mitochondrial appearance is not a reliable criterion for retinal terminal identification, at least in K L3.
Terminal bouton morphology of TG axons
Ultrastructure of TG projections
BDA tracer injections into the SC or brachium of SC yielded robust labeling of TG axons with beaded appearance in both K L3 and L6 (Fig. 5a,b). Unlike previous studies (Fitzpatrick et al., 1980; Harting et al., 1991), our SC injections did not reveal many fibers in the interlaminar regions other than a few isolated axons on route from the optic tract to L3; the interlaminar tectal axons are not regarded further. Also to note, we observed no labeled axons in M/P L1–L2 or L4–L5. This finding argues against the possibility that our SC injections retrogradely filled M/P retinal ganglion cell axon collaterals, as occurs with Y-retinal axons in the cat (Tamamaki et al., 1995; Datskovskaia et al., 2001). Consequently, we ruled out the possibility that retrogradely filled RGC axon collaterals intermingle with TG axons in the K laminae of the tree shrew LGN.
TG terminals in K L3 and L6 have retinal-like morphology. a, Anterogradely labeled TG axons traveling in the brachium of SC (bSC) and optic tract (OT) to terminate in the LGN at 5× magnification. Scale bar, 100 μm. b, TG axon bouton terminals in K L3 shown at 40× magnification. Scale bar, 50 μm. c, d, Electron micrographs of large, labeled TG terminals (t-BDA) in K L3 (c) and K L6 (d). RLP, unlabeled retinal terminal with LM; ax, myelinated axon; F2-int, vesicle-containing interneuron dendrite; d, dendrite; t, unlabeled terminal. Scale bar, 1 μm. e, Quantitative comparison of terminal areas of labeled TG terminals (BDA+ TG) and unlabeled terminals (BDA−) in L3 and L6. In both K lamina, TG terminal cross-section areas are significantly larger than BDA− terminals (p < 0.0001). L3 TG terminal cross-section areas do not differ significantly from L6 TG terminals. f, Cumulative frequency distributions of labeled TG and RG terminal areas in K L3 and K L6. In each lamina, TG terminals do not differ significantly from RG terminals. ***p < 0.0001.
We examined morphological properties of a total of 526 synaptic terminals (209 labeled TG terminals and 317 unlabeled terminals in the same sections as the labeled TG terminals). Most TG terminals in both K-recipient L3 and L6 were medium to large in size and significantly differed from the unlabeled (BDA−) terminals collected from the same lamina (L3 TG terminals 1.41 ± 0.09 μm2 vs L3 unlabeled terminals 0.49 ± 0.03 μm2; p < 0.0001; L6 TG terminals 1.69 ± 0.10 μm2 vs L6 unlabeled terminals 0.60 ± 0.07 μm2; p < 0.0001). The morphological characteristics of TG terminals resembled those originating from the retina (Fig. 5c,d), as they contained many mitochondria, made multiple synaptic contacts, and had protrusions, suggesting that TG axons may act as functional drivers in K laminae of LGN. Whether or not TG terminals contain light mitochondria or DM could not be assessed unambiguously because DAB chromogen in terminals could confound the mitochondria contrast.
Quantitative comparison of RG and TG projections
In both K-recipient laminae, TG terminals were significantly larger than the unlabeled terminals within the same lamina (Fig. 5e; MW-U; p < 0.0001). TG terminals in K-recipient laminae did not differ significantly from one another (Fig. 5e), unlike their RG counterparts (Fig. 3f,g). In L3, anterogradely labeled TG terminals were found to not differ significantly from RG terminals collected from the same laminae (Fig. 5f; L3 TG terminals 1.41 ± 0.09 μm2 vs L3 RG terminals 1.30 ± 0.08 μm2). Our analysis also revealed no significant difference in the size distributions of TG terminals from RG terminals collected in L6 (Fig. 5f; L6 TG terminals 1.69 ± 0.10 μm2 vs L6 RG terminals 1.93 ± 0.10 μm2).
In order to further ascertain if tectal inputs mimic driver-like properties of retinal inputs, we characterized the postsynaptic target of every recorded synapse across both RG and TG boutons. Postsynaptic targets were either classified as a dendrite, which included thin appendages and protrusions, or a vesicle-filled profile (presumed F2). This analysis revealed that, similar to the RG terminals, TG terminals engaged in triadic arrangements (Fig. 6a,b), although less frequently compared with RG boutons in the same K laminae (Fig. 6c). However, considering both RG and TG axons innervate L3 and L6, this suggests that K laminae may contribute to local, inhibitory circuitry more than their M/P counterparts.
Tectal inputs mimic driver-like properties of retinal terminals. a, Electron micrograph of a labeled RG terminal and an unlabeled, DM terminal (non-RG), presumed TG, converging on the same dendrite and forming triadic arrangements with a vesicle-containing interneuron dendrite (F2). Scale bar: 1 μm. b, Electron micrograph of a labeled TG terminal forming a triadic arrangement with a vesicle-containing interneuron dendrite (F2). Scale bar, 1 μm. c, Proportion of postsynaptic targets for RG and TG terminals. TG terminals contacted F2s, but to a lesser degree compared with RG terminals in the same lamina pair (L3, L6).
To determine whether TG terminals utilize the same glutamate transporter as retinal driver terminals, we injected CTB in one eye to label RG axons in the LGN and then costained the sections with VGluT2, which strongly labeled and delineated all six laminae (Fig. 7a). The Z-scans of dually labeled laminae revealed CTB-filled axons with their swellings, as well as VGluT2+ boutons (Fig. 7b,c; note that VGlutT2 rarely fills axonal fibers, which are devoid of vesicles). Using the Imaris software to analyze the Z-scans, we identified all potential VGluT2+ regions, assessed the colocalization of CTB+ pixels within these regions, and observed areas with only VGluT2 labeling (Fig. 7b,c). Particularly in L6, clusters of VGluT2+ regions were apparent.
The K pathway receives an extra VGluT2-positive input from the SC. a, LGN immunostained for VGluT2 (green) and contralateral RG projections labeled by CTB (CTB-RG, magenta). b, c, Dually labeled LGN sections in L3 (b) and L6 (c) showing colocalization between VGluT2+ boutons and anterogradely filled retinal terminals (white arrowheads) and singly labeled VGluT2 boutons nonretinal in origin (green arrowheads). Scale bar, 10 μm. d, Example of dynamic volume rendering of three colocalized boutons in the Imaris software. Scale bar, 2 μm. e, Example field of 3D-rendered boutons in L6 revealing VGluT2 boutons surrounded by CTB (white arrowheads), retinal boutons not containing VGluT2 (magenta arrowheads), and singly labeled VGluT2 boutons (green arrowheads). Scale bar, 2 μm. f, CI of CTB+ and VGluT2+ objects in L1–L6. M/P laminae have CI values closer to 1.0 than K laminae, indicating that K laminae have a greater prevalence of singly labeled VGluT2+ objects that are not retinal in origin. g, Schematic of BDA injection into SC (left) and labeled needle track at injection site (right). h, i, Labeled projections in K L3 (h) and K L6 (i) in dLGN of the thalamus. j–l, The LGN tissue stained for VGluT2 (green) and anterogradely filled fibers and boutons (BDA-TG, magenta). There was overlap between VGluT2+ boutons and TG terminals in both K L3 (j) and K L6 (k), and BDA-TG boutons were filled substantially with VGluT2 (white, l). Scale bar, 10 μm. ***p < 0.0001.
Because confocal images do not provide detailed information on discrete bouton morphologies—especially with VGluT2 immunolabeling—and to assess the prevalence of VGluT2 labeling outside of retinal axons, we devised an object-based quantification approach using dynamic volume rendering of fluorescently labeled boutons (Fig. 7d,e). Assuming that the absence of VGluT2 in CTB+ terminals represents background noise or false-negative colocalization, we calculated a CI for each image by normalizing the percentage of CTB localization within VGluT2-labeled regions to the percentage of VGluT2 localization within CTB-labeled regions. An index value of 1.0 indicates complete colocalization, whereas values below 1.0 suggest the presence of VGluT2 boutons that are not colabeled with CTB. The M/P laminae exhibited indices close to 1 (1.06 ± 0.03), while the K laminae showed lower values (0.65 ± 0.04; Fig. 7f), and these two groups were significantly different from one another (Welch's t test; p < 0.0001). While there were more singly labeled VGluT2 boutons in L6 (L3 0.70 ± 0.06 vs L6 0.60 ± 0.05), indicating differences between two K laminae, a statistical difference was not evident (Welch's, p = 0.25). These results indicate that L3 and L6 display a higher prevalence of VGluT2 boutons that are not of retinal origin.
To test whether TG axons may account for the nonretinal VGluT2+ boutons in L3 and L6, we injected the anterograde tracer BDA into the superficial SC and visualized the TG axons using a streptavidin fluorescence procedure (Fig. 7e–h) on sections dually labeled for VGluT2, VGluT1, and GABA. We found no colocalization of VGluT1 or GABA with anterogradely transported BDA in TG boutons (data not shown) in either L3 or L6. On the other hand, qualitative evidence was present for VGluT2-ir colocalization with BDA in both K L3 and L6 (Fig. 7i–o). Quantitative assessment for the rate of VGluT2/SC axon colocalization was not feasible, because, unlike for the assessment of VGluT2/retinal axon described above (Fig. 7f), a background normalization condition, where 100% colocalization could be expected, was not present in SC injection experiments given that axons only project to K laminae. We should also note that in similar experiments where we used AAV-CAG-tdTomato as the anterograde tracer of TG axons, we failed to demonstrate any colocalization, underscoring that viral transfection may sometimes occlude any subsequent immunolabeling (Watakabe et al., 2015).
To confirm the presence of VGluT2 transcript in SC neurons that project to the LGN, we stereotaxically injected the retrograde tracer AAV-CAG-GFP into the LGN (Fig. 8a) and then detected VGluT2 transcript using RNA-FISH. Injections confined to the geniculate laminae produced retrograde labeling in cells located in the upper superficial gray (SGS) and the zonal layer (Fig. 8b), while no labeled cells were observed in the lower SGS, where pulvinar-projecting wide–field vertical cells reside. The geniculate-projecting cells exhibited three distinct morphologies. Stellate cells—with round somata and radially emerging primary dendrites (Fig. 8f,g)—were the most common, accounting for ∼40% of labeled cells encountered across three SC sections (n = 131). Cells with fusiform somata oriented perpendicular to the surface, which gave off bipolar dendrites and are suggestive of narrow-field vertical (NFV) cell morphology (Fig. 8d,e), comprised 27.5% of retrogradely labeled cells. Finally, marginal cells with small somata found immediately ventral to the SC pial surface and gave off fine, ramifying dendrites (Fig. 8c), made up 22.9% of the geniculate-projecting neurons. These findings are consistent with previous descriptions of geniculate-projecting tectal cells in the tree shrew (Albano et al., 1979; Graham and Casagrande, 1980; Diamond et al., 1991; Harting et al., 1991).
Three tectal cell types project to dLGN, but only one may account for the VGluT2-positive input. a, Schematic of AAVrg injection into dLGN (top) and labeled injection site that is confined to dLGN and spans across L3 and L6 (bottom). b, Retrogradely filled cells (magenta) shown in the upper superficial gray (SGS) of the SC, with most cell bodies residing in the ventral upper SGS (USGSv) and none in the lower superficial gray (LSGS), where pulvinar-projecting cells reside. SZ, zonal layer; USGSd, dorsal upper superficial gray. c, Retrogradely filled marginal cell (MAR, magenta arrowhead) with a small soma at the dorsal surface of the SC that does not contain VGluT2 mRNA transcript (green). Scale bar, 1 μm. d, e, A field of retrogradely cells with fusiform somata and bipolar dendrites, suggestive of NFV cell morphology (d). Two of these NFV-like cells do not contain VGluT2 transcript (magenta arrowheads), while another is positive for VGluT2 FISH signal (white arrowhead, e). f, g, Retrogradely filled stellate cell with a round soma and radially emerging dendrites (STL, f) that contains VGluT2 FISH signal (white arrowhead, g). h, Quantitative comparison of prevalence of VGluT2 transcript across identified cell types. VGluT2 transcript (green striped bar) was present most prominently in stellate cells (STL), whereas the majority of NFV-like and marginal (MAR) cells did not contain VGluT2 mRNA (magenta bars).
RNA-FISH labeling revealed that VGluT2 transcript was most prominent in stellate cells, with 88.7% of these cells expressing VGluT2 (Fig. 8h). In contrast, only 16.7% of NFV-like cells and none of the marginal cells expressed VGluT2 mRNA (n = 36 and 30, respectively). Thus, TG projections in the tree shrew are morphologically and molecularly heterogeneous: three distinct cell types contribute to the TG terminals in the K geniculate laminae, but only one appears to provide most of the VGluT2-positive input. However, because our retrograde tracer injections spanned all geniculate laminae, our experiments cannot determine whether any of these three tectal cell types preferentially project to L3 or L6.
Discussion
Our analysis of anatomical properties of M/P and K laminae of the tree shrew LGN revealed: (1) Relay cells in K-recipient laminae are morphologically and molecularly distinct from those in the Magno/Parvo-recipient laminae: they have smaller somata and contain CALB, suggesting that morphological distinctions in RGCs are preserved through parallel pathway-specific channels in LGN; (2) calcium-binding proteins display interesting patterns of specificity; CALB is expressed in GABAergic cells across all laminae, in addition to being a marker for K laminae relay cells. In contrast, PARV is not specific to any neuron type or laminae. This suggests that tree shrew relay cells differ neurochemically in their CALB, but not PARV, expression in the LGN; (3) RGC axon terminals projecting to the Magno/Parvo versus K-recipient laminae are morphologically distinct: K retinal driver terminals, especially in L3, are significantly smaller, which is consistent with innervation by distinct phenotypes of RGC axons in these pathways; (4) TG axons, which originate from three distinct cell types in the SC, closely resemble retinal terminals and many are VGluT2-positive, suggesting that TG axons may act as functional drivers in the K laminae of the tree shrew LGN.
Morphological distinction of the K pathway
Parallel pathways of the vision originate from structurally and functionally distinct ganglion cells of the retina (Rodieck and Watanabe, 1993; Dacey, 2000; Dacey and Packer, 2003; Callaway, 2005). Similar to their putative retinal ganglion cell counterparts, relay cells in X/Y/W or M/P/K geniculate laminae display differences in soma sizes (LeVay and Ferster, 1977; Friedlander et al., 1981), and these differences were also evident in soma sizes of the tree shrew K lamina relay cells in our study. In addition, our study has brought clarification to whether or not M/P/K RGC axon terminals can be morphologically distinct. While the prototypical geniculate retinal terminal, as described by Guillery (1969), displays strikingly large cross sections in all mammalian species studied to date and in fact constitutes the largest-sized terminal boutons (Erişir et al., 1997, 1998; Van Horn et al., 2000), the studies of their size distribution have revealed that a wide range existed (Erişir et al., 1997, 1998). In fact, a more recent study revealed that multiple subpopulations make up the range of retinal bouton volume distributions (Maher et al., 2023), suggesting that pathway-specific RGC axons may provide morphologically distinct retinal boutons in the geniculate circuitry. With the unique advantage of segregated pathways of the tree shrew LGN, the current study now provides evidence that retinal terminals synapsing in the K laminae constitute the smaller-sized subpopulation.
Neurochemical distinction of the K pathway
A key argument for classifying K cells as a separate group arose from evidence demonstrating that the K pathway is neurochemically distinct from the M and P pathways, particularly in the expression patterns of calcium-binding proteins. In nonhuman primates and tree shrew, researchers observed that many cells in K-recipient LGN, but only a few in M/P-recipient laminae, displayed immunoreactivity for CaMKII (Benson et al., 1991; Hendry and Jones, 1991; Hendry and Yoshioka, 1994) and CALB (Jones and Hendry, 1989; Diamond et al., 1993; Hendry and Yoshioka, 1994; Johnson and Casagrande, 1995; Goodchild and Martin, 1998), suggesting that CALB expression can be a selective marker of K lamina cells (for review: Hendry and Reid, 2000). An omission in these studies has been the characterization of CALB cells that sparsely yet persistently appear in M/P laminae. We have advanced this understanding in the tree shrew demonstrating that, while K relay cells are neurochemically distinct from M/P relay cells through their selective expression of CALB, GABAergic cells across all geniculate laminae, including the K laminae, also express CALB. Furthermore, our results revealed that, unlike in the 1993 study in the tree shrew (Diamond et al., 1993) or in primates (Jones and Hendry, 1989; Diamond et al., 1993; Goodchild and Martin, 1998), immunostaining with PARV antibodies visualized cells indiscriminately across all laminae, suggesting that PARV staining was not a reliable marker for M/P relay cells in the tree shrew.
Why may it be of importance for relay cells in the K pathway to selectively express CALB? Given the role of calcium-binding proteins as a second messenger activated by increases in intracellular calcium, their role in synaptic transmission and plasticity has been well established (Blatow et al., 2003; Catterall and Few, 2008; Westerink et al., 2012). The LGN has long been regarded as a rigid, unmodifiable relay center with the assumption that plasticity typically occurs in the cortex. However, studies have shown plasticity at the RG synapse in the mature LGN, partially due to postsynaptic processes of differential regulation of AMPA and NMDA receptors (Chen et al., 2002) and Ca2+-dependent postsynaptic processes. Similarly, metabotropic glutamate receptors, in particular, mGluR1 and mGluR5 which are coupled to phosphatidylinositide hydrolysis and intracellular calcium mobilization, are expressed differentially in triadic relay dendrites and presynaptic appendages of interneurons (Godwin et al., 1996). While the presence of CALB and PARV in geniculate interneurons is consistent with mGluR5-mediated intracellular calcium-activated processes, the specificity of CALB in K-type relay cells indicate a selective capacity of plasticity in geniculate K-type relay cells.
Are there multiple K pathways?
Functional evidence supports an heterogeneity of K-type relay cells (Norton and Casagrande, 1982; Irvin et al., 1986; Norton et al., 1988; Holdefer and Norton, 1995; White et al., 1998, 2001; Eiber et al., 2018). K cells exhibit diverse response properties: some show unique functional signatures, such as blue- on or suppressed-by-contrast responses (Tailby et al., 2007; Roy et al., 2009; Jayakumar et al., 2013), some display direction selectivity or visual properties absent in M/P cells (Solomon et al., 2002), and other K cells respond minimally or not at all to traditional visual stimuli (White et al., 2001; Xu et al., 2001). Additionally, two distinct groups of W cells in the cat retina were distinguished based on the soma size and firing patterns (Stanford et al., 1981; Stanford, 1987), and retinal terminals in W-cell recipient zones of the cat LGN display morphological heterogeneity (Chen et al., 1996). In the macaque, there's also evidence for morphological distinctions of corticothalamic feedback neurons within different parallel processing streams (Hasse et al., 2019). While our results in the tree shrew do not reveal heterogeneity in neurochemical markers and relay cell soma sizes within K laminae, morphological properties of synaptic inputs in two K laminae suggest key differences. First, our terminal size analyses in K-recipient LGN laminae revealed two morphologically distinct populations, with K L3 having smaller retinal terminals than K L6, suggesting at least two distinct RGC types innervate these two laminae selectively. Second, the K laminae pair displayed evidence of distinct circuitries: analysis of postsynaptic targets of RG inputs showed that K L6 had a higher incidence of engagement in triadic arrangements compared with L3. Additionally, K L6, but not K L3, was impoverished of small-sized glutamatergic terminals (presumed corticothalamic) that constitute up to one-third of synaptic inputs to geniculate laminae, suggesting major differences in feedback excitation that impinges on cells in two K laminae. Furthermore, in the tree shrew, the K pathway contains the only pair of laminae that has differential outputs to the striate cortex, suggestive of distinct circuitries: K L3 and K L6 primarily terminate in different subdivisions of Layer 3, with L3 axons targeting 3b and L6 projecting to 3c (Usrey et al., 1992). Finally, previous work has shown that different collicular cell types selectively target K laminae of the tree shrew dLGN, with stellate cells selectively terminating in K L3 while NFV-like cells preferentially target K L6 (Diamond et al., 1991). While our study does not provide definitive clarification for where each tectal cell type projects to, it provides evidence for potential differences in neurotransmitter release from these distinct cell types, thus suggesting differences in excitation and modulation impinging on the two K laminae.
In thalamic circuitry, inputs have been canonically classified as either “drivers” or “modulators” (Sherman and Guillery, 1998, 2002), with retinal input constituting the primary driver across all visual pathways. However, in the mouse, Bickford and colleagues have found that optogenetic stimulation of TG terminals led to excitatory postsynaptic potentials similar to those elicited by optic nerve stimulations, leading them to propose that TG inputs had the signature characteristics of an additional driver input on geniculate W-type relay cells (Bickford et al., 2015). Drivers and modulators are also distinguished by their transporters in the LGN: the driver, typically retinal, input utilizes VGluT2 while cortical input utilizes VGluT1 (Land et al., 2004; Balaram et al., 2011, 2013, 2015; Rovó et al., 2012). The LGN's third glutamatergic input, the one from SC, was speculated to be VGluT2-positive based on morphological and physiological properties of the SC terminals (Bickford et al., 2015) and the presence of VGluT2 mRNA transcript in superficial SC (Balaram et al., 2015), where LGN-projecting stellate cells reside. The current study adds to this evidence by confirming the presence of VGluT2 in SC terminals and VGluT2 mRNA in LGN-projecting stellate cells, which is consistent with the idea that at least some of the TG axons may be exerting driver-like influences on relay cells in K laminae. However, we also demonstrated that not all LGN-projecting cells utilize VGluT2: a substantial portion of geniculate-projecting tectal neurons does not express VGluT2 mRNA, including a morphological subtype of NFV cells, raising the question of which neurotransmitters may be released from those axons. The presence of a nonglutamatergic NFV cell type is not novel: unlike the glutamatergic NFV cells projecting to deep SC (Gale and Murphy, 2014), a subpopulation of NFV cells are GABAergic and project to PBG (Whyland et al., 2020). It will be interesting to know whether those same cells also project to geniculate K laminae via axon collaterals. Similarly, while marginal cells have been shown to be direction-selective (Mooney et al., 1985) and project to the LGN (Mooney et al., 1988), the neurotransmitters expressed in this cell type have not been identified, except for our RNA-FISH experiments, which reveal that they do not express VGluT2.
Our study admittedly has a technical limitation in quantifying the ratio of axons that originate from three different TG cell types that terminate in two K laminae: the tracer injections in the SC are bound to be incomplete in filling all TG cells located across the entire depth and topography of the SC. Thus, the proportion of filled terminal boutons after any injection may represent a greater percentage of one cell type over another. We expect this question could be best addressed with multiple injections throughout the mediolateral extent of the sSC using cell type-specific AAVs, which could allow for the assessment of the prevalence of such axons separately in LGN laminae. Regardless, our findings suggest that not only are the K laminae circuitry heterogeneous but also that two K laminae may be influenced differentially by distinct tectal pathways with unique functions.
Implications for blindsight
Blindsight, the phenomenon where humans and nonhuman primates can detect visual stimuli following damage to V1 (Sanders et al., 1974), is a condition that must rely on an existing processing stream that relays visual information beyond the striate cortex. There are several features of the K pathway that make it suitable for this purpose. First, tectal projections are a distinct, specific property of the K pathway in all mammalian species examined (Harting et al., 1991; Lachica and Casagrande, 1993; Bickford et al., 2015; Baldwin and Bourne, 2020), and they may exert strong excitation on relay cells, akin to retinal input. Second, SC is the origin of two main parallel projections that bypass V1: one reaches multiple areas of the association cortex through the visual pulvinar, and another to the extrastriate visual cortex via K layers of LGN (Benevento and Yoshida, 1981; Yoshida and Benevento, 1981; Yukie and Iwai, 1981; Bullier and Kennedy, 1983; Rodman et al., 2001). Third, studies in macaques with V1 ablation have shown that while a widespread relay cell degeneration predominate, many cells that express CALB and project to visual association areas survive (Cowey and Stoerig, 1989; Rodman et al., 2001; Atapour et al., 2021, 2022) and maintain intact visual processing (Yu et al., 2018). Similarly, in a monkey model of blindsight, pharmacological inactivation of LGN neurons suppressed functional activation of the visual association cortex and eliminated visual detection abilities (Schmid et al., 2010), suggesting that the geniculate K pathway is crucial for blindsight. Finally, the prominent tectal input we demonstrate in the tree shrew dLGN may be indicative of a specialized visual channel that relays information about stimulus motion and eye movement to higher visual areas. While direct evidence for tree shrew K laminae relay cells projecting to MT analog extrastriate cortical areas in a similar fashion demonstrated in the macaque and marmosets (Sincich et al., 2004; Atapour et al., 2022) is elusive, whether or not distinct K pathway information processed in geniculate L3 and L6 have selective roles in blindsight will be an interesting question to address.
Conclusion
In a seminal 1994 review (Casagrande, 1994), Vivian Casagrande provided extensive anatomical, neurochemical, and functional evidence for a third parallel visual pathway, the K pathway, suggesting it may play both a modulatory role in vision and a perceptual role during eye movements. By examining tree shrew LGN where K pathways are uniquely segregated, we revealed structural distinctions of parallel pathways and revealed further evidence for two K subpathways, each with their own inputs and distinct synaptic circuitries, and this is consistent with multiple roles K relay cells may play on visual processing as suggested by Casagrande.
Footnotes
The authors declare no competing financial interests.
We thank and acknowledge Rebecca Roberts for her technical assistance and colony care and Laura Jamison and Stephanie Zhang for their assistance with the R code for three-dimensional confocal analyses. This work was supported by National Institutes of Health–National Institute of Neurological Disorders and Stroke (U01NS122040).
This is an open-access article distributed under the terms of the Creative Commons Attribution 4.0 International license, which permits unrestricted use, distribution and reproduction in any medium provided that the original work is properly attributed.
References
In this issue
