Diverse GABAergic neurons organize into subtype‐specific sublaminae in the ventral lateral geniculate nucleus

Abstract In the visual system, retinal axons convey visual information from the outside world to dozens of distinct retinorecipient brain regions and organize that information at several levels, including either at the level of retinal afferents, cytoarchitecture of intrinsic retinorecipient neurons, or a combination of the two. Two major retinorecipient nuclei which are densely innervated by retinal axons are the dorsal lateral geniculate nucleus, which is important for classical image‐forming vision, and ventral LGN (vLGN), which is associated with non‐image‐forming vision. The neurochemistry, cytoarchitecture, and retinothalamic connectivity in vLGN remain unresolved, raising fundamental questions of how it receives and processes visual information. To shed light on these important questions, used in situ hybridization, immunohistochemistry, and genetic reporter lines to identify and characterize novel neuronal cell types in mouse vLGN. Not only were a high percentage of these cells GABAergic, we discovered transcriptomically distinct GABAergic cell types reside in the two major laminae of vLGN, the retinorecipient, external vLGN (vLGNe) and the non‐retinorecipient, internal vLGN (vLGNi). Furthermore, within vLGNe, we identified transcriptionally distinct subtypes of GABAergic cells that are distributed into four adjacent sublaminae. Using trans‐synaptic viral tracing and in vitro electrophysiology, we found cells in each these vLGNe sublaminae receive monosynaptic inputs from retina. These results not only identify novel subtypes of GABAergic cells in vLGN, they suggest the subtype‐specific laminar distribution of retinorecipient cells in vLGNe may be important for receiving, processing, and transmitting light‐derived signals in parallel channels of the subcortical visual system.

While their organization is not as ordered as their retinal afferents, classes of dLGN relay cells exhibit some regional preferences in their distribution, whereas interneurons are evenly dispersed throughout the nucleus (Krahe et al., 2011). Cell type-specific circuitry and function has also been demonstrated in dLGN, where W-like relay neurons receive input from direction-selective RGCs and in turn project to the superficial layers of mouse primary visual cortex (Cruz-Martín et al., 2014).
While our understanding of subtype-specific circuits has facilitated functional studies of SC and dLGN, there remain many retinorecipient regions about which such foundational information is unknown. One such region is the ventral LGN (vLGN), a portion of ventral thalamus that neighbors dLGN and is similarly innervated by retinal axons. Although less studied, it has been shown that vLGN is remarkably distinct from its dorsal counterpart in its transcriptome, proteome, cytoarchitecture, and circuitry (Harrington, 1997;Monavarfeshani et al., 2018;Sabbagh et al., 2018;Su et al., 2011).
In fact, distinct subtypes of RGCs project to vLGN and dLGN, and the majority of dLGN-projecting RGC classes fail to send collateral axons into vLGN, despite having to pass by it (or through it) on the way to dLGN (Huberman et al., 2008(Huberman et al., , 2009Kim et al., 2008).
Here, to address these gaps, we sought to determine the celltypes populating the vLGN, and their connectivity to retinal afferents. We assessed vLGN neurochemistry and cytoarchitecture by labeling cells with canonical and novel cell type markers. We found a richly diverse and tightly organized cellular landscape in vLGN, where transcriptomically distinct cell types are distributed in laminar subdomains, which appear to receive monosynaptic inputs from the retina. Our findings not only identify a novel organization of retinorecipient cells in vLGN, they suggest this order may be important for receiving, processing, and transmitting distinct light-derived signals in parallel channels of the subcortical visual system.

| Animals
Wild-type C57BL/6 mice were obtained from Jackson Laboratory.
We obtained the following mice from Jackson Laboratory . Animals were housed in a temperature-controlled environment, in a 12 hr dark/light cycle, and with access to food and water ad libitum. Both males and females were used in these experiments. Genomic DNA was isolated from tails genotyping as previously described (Su, Gorse, Ramirez, & Fox, 2010)

| Anterograde axon and mono-synaptic tracing
Intravitreal injection of cholera toxin subunit B (to trace retinal terminals) was performed as previously described Su et al., 2011). Briefly, mice were anesthetized with isoflurane, and 1 μl of 1 mg/ml fluorescently conjugated Alexa-647-CTB (Invitrogen, C34778) was binocularly injected with a fine glass pipette using a picospritzer. The rapid, minor, and non-invasive nature of these injections (which is not considered a surgery) precluded the need for additional analgesics. After 3 days, animals were killed and transcardially perfused with PBS followed by PFA. We opted for a binocular injection here because it is the best way to accurately delineate the border between vLGNe and vLGNi.
A similar intravitreal injection of AAV2/1-hSyn-Cre-WPRE-hGH (2.5 × 10 13 GC/mL, here referred to as AAV1-Cre) was used to monosynaptically label retinorecipient neurons in the vLGN. 1.2 μl of AAV-Cre virus was binocularly injected at an approximate 45° angle relative to the optic axis. We opted for a binocular injection here to maximize the probability of AAV-Cre infecting RGCs and the probability of trans-synaptic infection in vLGN. AAV1-Cre was a gift from James M. Wilson (Addgene viral prep #105553-AAV1; RRID:Addgene_105553). Animals were killed and perfused with PFA as described above 6-10 weeks after injection.

| Transcriptomic analyses
RNA from wild-type vLGN and dLGN was extracted and purified from mice at P3, P8, P12, and P25, then processed at the Genomics Research Laboratory at Virginia Tech's Biocomplexity Institute for RNAseq analysis. The RNA sequencing experiment was previously published and the protocol is described in detail in .

| In vitro slice preparation and wholecell recording
In vitro recordings were conducted on genetically labeled vLGN neurons using methods described previously (Hammer et al., 2014). Mice were anesthetized with isoflurane, decapitated and brains were rapidly immersed in an ice-cold, oxy-  OT stimulation consisted of a 10Hz (10 pulses) train delivered at an intensity (25-200 µA) that evoked a maximal response (Hammer et al., 2014;Jaubert-Miazza et al., 2005). In vitro data are obtained from all healthy cells which we were able to record from in six Pvalb-Cre::Thy1-Stop-YFP mice, three Sst-cre::Rosa-Stop-tdT mice, and two GAD67-GFP mice.

| Statistics
Since comparisons of electrophysiological measurements were between three groups and were not all normally distributed (D'Agostino-Pearson test; all groups were normally distributed except for the GAD67-GFP group because of sample size; GraphPad Prism), we determined statistical significance by a Kruskal-Wallis test with Dunn's correction for multiple comparisons using GraphPad Prism (version 8.0.; RRID:SCR_002798). Differences were considered significant when p < .05. We did not perform outlier tests for electrophysiological data. No exclusion criteria were predetermined before the described experiments were performed and no data (or animals) were excluded from any of the analysis.

| Quantification and imaging
No sample calculation was performed; a sample size of at least three biological replicates (mice) was determined to be appropriate (based on observed variability and previous experience (Sabbagh et al., 2018)) where quantitative comparison across replicates was performed (colocalization with Gad1 signal, spatial distribution analysis in vLGNe/vLGNi using cholera toxin subunit b (CTB) tracing, and line scan analyses). In these analyses, we obtained measurements obtained from three to four vLGN sections per mouse brain and averaged them to obtain the mean value for that biological replicate.
To quantify the GABAergic neurons as a percentage of total cells in vLGN and dLGN, we labeled and counted Gad1 + (by ISH), Boundaries of vLGN or dLGN were determined by DAPI counter-staining. For these imaging experiments, no prior sample calculation was performed; a sample size of at least 3 biological replicates (mice) was determined to be appropriate based on observed variability and previous experience to asses colocalization with Gad1 signal, spatial distribution analysis in vLGNe/vLGNi using CTB tracing, and line scan analyses (described below). For each of these analyses, we obtained measurements from three to four vLGN sections per mouse brain and averaged them to obtain the mean value for that biological replicate.
To quantify the spatial distribution of cell-type marker expression across the entire vLGN, we developed a custom line scan script (khatScan) that runs in ImageJ which overlays the vLGN with equally spaced lines. We opted for this approach over manually drawing lines to avoid user bias. A brief summary of how this script works: to determine the curvature of the vLGN in a particular image, khatScan prompts the user to draw a line along the optic tract adjacent to the vLGN, then automatically draws lines of a set length and number guided by that curve and plots the signal intensity across the x coordinates of each line. These intensities can then be averaged to determine where there is a specific enrichment for that marker in the vLGN. All imaging for quantification was performed on a confocal Zeiss LSM 700 microscope at 20x magnification and 0.5 digital zoom.

| Identification of distinct subtypes of GABAergic cells in vLGN
We first determined precisely what proportion of vLGN cells were GABAergic. In the brain, GABAergic neurons express Gad1 and/ or Gad2, genes which encode glutamate decarboxylases (GAD67 and GAD65, respectively), the biosynthetic enzymes necessary for the production of the neurotransmitter GABA. We therefore took two complementary approaches to label GABAergic interneurons: we performed in situ hybridization (ISH) with riboprobes gener- we generated >40 riboprobes to perform ISH (Table S1). We also took advantage of available cell type-specific reporter mice and antibodies for this screen.
From this unbiased riboprobe screen, we identified two genes, Nxph1 and Arx, whose expression in vLGN was restricted to cells in largely non-overlapping domains. Nxph1, which encodes the α-Nrxn we found that > 90% of Pvalb + cells generated Gad1 mRNA and were therefore GABAergic (Figure 2o-p). When we labeled retinal projections in vLGNe with CTB and Pvalb + GABAergic neurons by immunolabeling, we determined that Pvalb + cells were exclusively present in vLGNe (Figure 2q-r). It was noteworthy that Sst + , Calb1 + , and Pvalb + cells were largely absent from the neighboring dLGN, suggesting that these GABAergic cell types were unique from previously studied dLGN interneurons.
Since neither the Pvalb + neurons (which preferred the vLGNe) nor Calb + neurons (which preferred the vLGNi) labeled all of the cells in their respective subdomains, we hypothesized that an even richer Taken together, these experiments reveal novel markers of transcriptomically and regionally distinct subsets of GABAergic neurons in vLGN. Importantly, not only did these cells have regional preferences, but Pvalb + , Penk + , Spp1 + , Ecel1 + , and Lypd1 + neurons each appeared to be organized into segregated strata that span the dorso-ventral axis of the vLGN.

| Transcriptomically distinct GABAergic neurons organize into adjacent sublaminae of vLGNe
We next asked whether these were in fact mutually exclusive groups of cells and whether they corresponded to distinct vLGNe sublaminae. We started by assessing whether Spp1 and Pvalb mRNAs were generated by the same neurons or occupied the same dorsoventral zone. Performing ISH on Pvalb-Cre::Thy1-Stop-YFP tissue revealed that Spp1 mRNA was generated by some, but not all, Pvalb + cells (~50% Spp1 + neurons were Pvalb -) and vice versa (~25% Pvalb + neurons were Spp1 -) (Figure 4a-d). Spp1 + Pvalb + , Spp + Pvalb -, and Pvalb -Spp1 + cells all appeared to reside within the same sublamina of vLGNe. To quantitatively assess cellular distribution in vLGN, we developed an automated script in ImageJ to measure fluorescent intensity along the medial-lateral axis of vLGN (Figure 4c inset).
Fluorescent signals at each medial-lateral coordinate were averaged along the entire dorsoventral extent of vLGN (and between biological replicates) and the quantified data identified the same spatial region of vLGN as populated both Spp1 + and Pvalb + cells (Figure 4e).
Next, we asked whether Penk was generated by either Spp1 + or Pvalb + cells. For this, we used double ISH or genetic reporter lines. In both cases, we were unable to identify a single occurrence in which

| GABAergic neurons in the four sublaminae of vLGNe are retinorecipient
The  (Figure 6a and b). This trans-synaptic viral transfection strategy has previously been shown to accurately map monosynaptic connections and, in our hands, resulted in sparse tdT + labeling of cells in retinorecipient brain regions (Figure 6b) (Zingg et al., 2017).
Using this approach, we trans-synaptically labeled 53 retinorecipient We next sought to functionally confirm that some of these transcriptomically distinct cell types receive direct retinal input and to characterize their electrophysiological response properties.
We utilized available transgenic reporter lines to record from and characterize several GABAergic subtypes in vLGN. Figure Guillery, 1966;Seabrook et al., 2013). The voltage responses to current injection also revealed differences in their intrinsic membrane and firing properties. For example, the resting membrane potential (Figure 7j

| D ISCUSS I ON
In this study, we identified novel vLGN cell-type markers which label GABAergic cells throughout the nucleus and distinguish it from its dorsal counterpart-the dLGN. In some cases, these novel GABAergic cell types also distinguished vLGN from IGL, although some transcriptionally defined cell types also appear to be shared between these two regions. By performing a series of multiplex labeling experiments using these newly identified markers, we revealed a remarkably organized laminar architecture of vLGNe, composed of at least four adjacent, transcriptionally distinct sublaminae. Interestingly, such organization was absent from IGL, even for cell types present in both regions.
Using anterograde trans-synaptic viral tracing and patch-clamp electrophysiology, we determined that many of these regionally and tran-

| Different types of hidden laminae exist in vLGN and dLGN
In contrast to the clear lamination of the primate and carnivore lateral geniculate nucleus, the rodent dLGN and vLGN have no obvious cytoarchitectonic lamination-save for the division of the vLGN into retinorecipient vLGNe and non-retinorecipient vLGNi (Gabbott & Bacon, 1994;Harrington, 1997;Hickey & Spear, 1976;Niimi et al., 1963;Sabbagh et al., 2018). The neuronal cytoarchitecture of the rodent dLGN is composed of three well-defined classes of glutamatergic thalamocortical relay cells and just one or two classes of GABAergic inhibitory interneurons (roughly 10% of its neuronal population) (Arcelli et al., 1997;Evangelio et al., 2018;Jaubert-Miazza et al., 2005). While interneurons are present throughout the nucleus, relay cells of each class appear to exhibit regional preferences in their distribution within the dLGN (Krahe et al., 2011).
These regional preferences in the rodent dLGN, however, do not alone capture the level of organization seen in primate and carnivore LGN. Instead, it appears that retinal afferents impart order in the rodent dLGN by segregating their arbors into "hidden laminae" both in terms of eye-specific domains and subtype-specific lamina (Hong & Chen, 2011;Martin, 1986;Reese, 1988 (Hattar et al., 2006;Monavarfeshani et al., 2017;Osterhout et al., 2011). The diffuse terminal arborization of these non-image forming subtypes of RGCs raises the question of whether visual information is, in fact, parsed into parallel channels in vLGN (Hattar et al., 2006;Osterhout et al., 2011). The stratification of transcriptomically distinct neurons presented in this study into adjacent sublaminae in vLGNe may contribute to the parallel processing of sensory information in this brain region. Just as the organization of retinal inputs imposes order on the otherwise less organized cytoarchitecture of dLGN, the diversity and organization of intrinsic cells in vLGN may impose order on the unorganized input it receives from the retina.

| Do laminated retinorecipient circuits organize visual pathways through the vLGN?
Is the quantifiable segregation of distinct GABAergic subtypes into sublaminae a potential means of organizing visual information arriving from the retina? To address this, we sought to determine whether these subtypes were directly innervated by RGCs.
Using anterograde trans-synaptic tracing, we identified Ecel1 + , Pvalb + , Spp1 + , and Penk + cells as retinorecipient, together representing at least three sublaminae of vLGNe. While we failed to observe any retinorecipient cells in vLGNi using this method, it certainly remains possible that the dendrites of cells in vLGNi extend into vLGNe and receive monosynaptic input from the retina.
Nevertheless, our viral tracing results suggest that the cell-typespecific organization of the vLGN is relevant to organizing visual input. Such organization of retinorecipient cells hints at a potentially novel role for vLGN in visual processing, by which incoming visual input is sampled by specific GABAergic cell types and parsed into parallel channels of sensory information to be trans-  (Cadusseau & Roger, 1991;Swanson, Cowan, & Jones, 1974;Trejo & Cicerone, 1984), the suprachiasmatic nucleus (at least in hamsters) (Harrington, 1997), the habenula (Huang et al., 2019;Oh et al., 2014), and zona incerta (Ribak & Peters, 1975), all contributing to visuomotor, ocular, vestibular, circadian, and other innate behaviors (Monavarfeshani et al., 2017). It is known that some GABAergic cells in vLGN receive retinal input and project to the lateral habenula, although it remains unclear which GABAergic subtypes this includes (Huang et al., 2019).
Might the different transcriptionally distinct GABAergic cell types in vLGN each project to different downstream nuclei and contribute to unique functions and behaviours? First, we note here that it is conceivable that vLGN not only organizes and transmits visual information in separate channels, but also samples specific features of retinal input in sublamina-specific manner-consistent with labeled-line theory. Unfortunately, we do not yet have subtype specific resolution of vLGN projection neurons but hope that the data presented here will help to create a molecular toolkit for such circuit tracing in future studies. Such experiments will be crucial in a) determining whether parsing visual information into these hidden laminae is important for parallel processing and b) whether there is a functional and/or projection-specific logic to the lamination of vLGN cell types.

| Transcriptomic heterogeneity underlying cellular diversity in vLGN
There is a current push to use unbiased approaches to identify all of the cell types in the brain-essentially to create a "parts" list.
These studies have typically employed single cell RNA sequencing (scRNAseq) to understand the development, structure, and evolution of the brain (Krienen, Goldman, & Zhang, 2019;Peng et al., 2019;Saunders et al., 2018). In a neuroscience community that consists of "lumpers" and "splitters," it is clear we are currently in an era of "splitting"-as our technology to detect transcriptomic heterogeneity of cell types evolves, so too does the granularity of their discretization.
Here, we have begun to "split" the rodent vLGN into many distinct GABAergic cell types.
Rather than performing scRNAseq, we tackled the heterogeneity question using bulk RNAseq and generating riboprobes with no a priori knowledge to perform a battery of in situ hybridizations for transcriptional heterogeneity in vLGN. By generating riboprobes against single molecules, we delineated distinct populations of transcriptomically and spatially distinct cells, an advantage of this approach over scRNAseq. Conservatively, our results demonstrate the existence of at least a half dozen discrete and separable subtypes of GABAergic neurons in vLGN, (although the number is likely much higher than this) and four distinct, adjacent sublaminae of GABAergic neurons in vLGNe.
Whether there are six GABAergic cell types in vLGN or many more has made us ponder the old question (Cajal, 1893) of how does one define a cell type? Traditionally, classes and types of neurons have been characterized on the basis of morphological, electrophysiological, neurochemical, connectomic, or genetic information (Sanes & Masland, 2015). Unfortunately, rarely are all these aspects of neuronal identity accounted for in a comprehensive way to glean a more accurate understanding about the structure of the nervous system.
This has led to discrepant subtype classification across technical methodologies and challenges to comparing results between research groups. Here, we used spatial distribution and transcriptional profiles to classify neurons into distinct subtypes. Molecular markers remain at present a leading characteristic for such classifications, though they are not without their limitations. It may well be that vLGN neurons can be further subdivided if classified by 2-3 molecular markers rather than one (as we observed in Spp1 + and Pvalb + neurons). However, the very fact that differential expression of one molecule was sufficient to differentiate two vLGN populations is a strong indicator that the nucleus as a whole is more diversely populated than previously appreciated. Nevertheless, we acknowledge here the efforts to create a more comprehensive framework to classifying cell types in the field. A set of recent studies represent a major step towards this goal by utilizing Patch-seq to simultaneously characterize cortical GABAergic neurons electrophysiologically, morphologically, and transcriptomically (Gouwens, Sorensen, & Baftizadeh, 2020;Scala et al., 2020). Our approach here did not take into account these additional functional aspects of neuronal identity for all of the GABAergic cell types identified.
Our data, when taken together, suggest the possibility that functional organization of non-image-forming information from retina to vLGN is extracted from the segregation of transcriptionally distinct retinorecipient cells. We view these results as a framework for further dissecting the structure, circuitry, and functions of the vLGN at a cell type-specific level. How this heterogeneity and organization contributes to the yet-to-be determined functions of the vLGN remains to be defined.

ACK N OWLED G M ENTS
This work was supported in part by the following grants from the We thank Dr A.S. LaMantia (Fralin Biomedical Research Institute, Virginia Tech) for thoughtful comments on the manuscript. A preprint of this manuscript was previously posted on bioRxiv .
All experiments were conducted in compliance with the ARRIVE guidelines.

CO N FLI C T S O F I NTE R E S T D I S CLOS U R E
The authors have no conflicts of interest to declare.

O PE N R E S E A RCH BA D G E S
This article has received a badge for *Open Materials* because it provided all relevant information to reproduce the study in the manuscript. More information about the Open Science badges can be found at https://cos.io/our-servi ces/open-scien ce-badge s/.