Morphological and Electrophysiological Characterization of a Novel Displaced Astrocyte in the Mouse Retina

Abstract

Astrocytes throughout the central nervous system are heterogeneous in both structure and function. This diversity leads to tissue-specific specialization where morphology is adapted to the surrounding neuronal circuitry, as seen in Bergman glia of the cerebellum and Müller glia of the retina. Because morphology can be a differentiating factor for cellular classification, we recently developed a mouse where glial-fibrillary acidic protein (GFAP)-expressing cells stochastically label for full membranous morphology. Here we utilize this tool to investigate whether morphological and electrophysiological features separate types of mouse retinal astrocytes. In this work, we report on a novel glial population found in the inner plexiform layer and ganglion cell layer which expresses the canonical astrocyte markers GFAP, S100β, connexin-43, Sox2 and Sox9. Apart from their retinal layer localization, these cells are unique in their radial distribution. They are notably absent from the mid-retina but are heavily concentrated near the optic nerve head, and to a lesser degree the peripheral retina. Additionally, their morphology is distinct from both nerve fiber layer astrocytes and Müller glia, appearing more similar to amacrine cells. Despite this structural similarity, these cells lack protein expression of common neuronal markers. Additionally, they do not exhibit action potentials, but rather resemble astrocytes and Müller glia in their small amplitude, graded depolarization to both light onset and offset. Their structure, protein expression, physiology, and intercellular connections suggest that these cells are astrocytic, displaced from their counterparts in the nerve fiber layer. As such, we refer to these cells as displaced retinal astrocytes.

Graphical Abstract

Introduction

Astrocytes in the brain show regional heterogeneity in both structure and function, playing crucial roles in energy production, blood-brain barrier maintenance as well as synaptic maintenance and plasticity^1,2. Recently, we used the G-MORF transgenic mouse strain to explore retinal astrocyte morphology, finding complexity could be reduced through recurring structural motifs which were suggestive of function^3,4. In the retina, astrocytes help maintain proper function of retinal ganglion cell (RGC) neurons^5–8, whose axons comprise the optic nerve. These glia form a dense plexus with cell bodies in the nerve fiber layer (NFL), extending processes that make intricate connections with not only RGC axons, but also vascular and other glial elements of the NFL and ganglion cell layer (GCL). Like astrocytes in the brain, these connections mediate metabolic support to surrounding neurons, blood flow regulation, resource distribution, and maintenance of extracellular ion concentrations^5–9. Because astrocytes in the retina show extensive morphological diversity, it’s possible that regional heterogeneity drives differential physiology in this area of the central nervous system (CNS) as well³.

Here we use once again G-MORF mice in conjunction with a strain producing tdTomato in glial fibrillary acidic protein (GFAP)-expressing cells to investigate whether morphological and physiological features separate types of retinal astrocytes. Through this work, we found a novel GFAP-expressing glial cell in the inner plexiform layer (IPL) and GCL. While distinct from Müller glia, these cells share similarities in both protein expression and light responses with NFL astrocytes despite differences in morphology, retinal distribution, and localization. These putative displaced retinal astrocytes likely monitor electrical and synaptic activity of the inner plexiform layer in order to regulate blood flow, making connections with neuronal processes and blood vessels found there.

Methods

Animals:

All transgenic mice used in this study were either a product of a cross between GFAP Cre 77.6 (Jackson #024098) and floxed tdTomato (Jackson #007914, Ai14) or between GFAP Cre 77.6 and MORF3 (Jackson # 035403). Progeny from the second cross are referred to as G-MORF mice³. We additionally used non-transgenic 129S6/SvEvTac (129SVE, Taconic Biosciences) mice. All animals used in this study were adult mice (2–5 months old) with equal numbers of males and females. Animals were housed at the Vanderbilt University Division of Animal Care facility and subjected to a 12-hr. light/dark cycle. Animals were provided with water and rodent chow ad libitum. Animals whose retinae were used in wholemount preparations were euthanized and retinas dissected as previously described³. After fresh dissection, retinas were transferred to a solution of 4% paraformaldehyde to shake at room temperature for 1 hr., then stored in 1X PBS with azide at 4°C. Animals whose retinae were sectioned were euthanized by intraperitoneal pentobarbital injection followed by transcardial perfusion of 1X PBS and 4% paraformaldehyde. Eyes were stored in a 30% sucrose solution in PBS prior to sectioning.

Histology:

Retinal sections were acquired through cryostat sectioning at a thickness of 30μm. Sections were washed 3X in 1X PBS to remove Optimal Cutting Temperature compound (Fisher 23–730-571). Tissue was blocked in a solution of 5% normal donkey serum (NDS) in 0.1% Triton-X PBS (PBST) for 3 hr. at room temperature. Primary antibody, in a solution of 3% NDS in PBST, was incubated overnight at 4°C. Tissue was washed 3X and a solution of secondary antibody in 1% NDS in PBST added to incubate at room temperature for 2 hours. Slides were washed 3X in PBS and then mounted for imaging. Wholemount tissue was treated similarly except the incubation time for primary antibody was 3 days instead of overnight. All primary antibodies, unless otherwise stated, were used at a concentration of 1:500. Primary antibodies used in this study include the following: goat GFAP (Abcam, ab53554), rabbit S100β (abcam, ab52642), mouse Sox2 (Santa Cruz, sc-365823), rabbit Sox9 (abcam, AB5535), goat choline acetyl transferase (1:100; Millipore, AB144P), rabbit glutamate decarboxylase (GAD65/67, Abcam, AB11070), rabbit V5 (Bethyl Laboratories, A190–120A), rabbit lucifer yellow (Invitrogen, A-5750), rabbit Pax6 (1:300, Millipore, AB2237), mouse syntaxin 1a (Millipore, S0664), rabbit glycine transporter 1 (Glyt1, Antibodies Online, ABIN1841935), chicken tyrosine hydroxylase (Abcam, ab76442), rabbit connexin 43 (Cell Signaling Technology, 3512S), goat IBA-1 (Novus, NB100–1028), rabbit RBPMS (GeneTex, 118619), and goat KLF4 (R&D Systems, AF3158-SP).

Imaging:

Confocal images were taken on a Nikon Ti-E spinning disk confocal microscope at 60X and 100x magnification. Z-stacks were taken at a step size of 0.3μm. When stacked, individual pages in each image file were combined using the standard deviation Z-stack method in Fiji ImageJ version 2.3.0. Images of amacrine cell markers were subjected to 25 iterations of Richardson-Lucy deconvolution using Nikon Nis-Elements AR 5.21.03. 3D cellular reconstruction was accomplished using Imaris 10.0.0. Cell density plots were created by first obtaining cell coordinates using Fiji. These coordinates were transferred to the open-source, R package RETISTRUCT to get geographic coordinates of cells mapped to native retinal space¹⁰. Cells were binned according to distance from the optic nerve head in increments of 100μm. Cell density was taken as the number of cells in each bin divided by the area of the annulus corresponding to each bin.

Sequencing Data:

Data for single-cell RNA sequencing was taken from a study which generated an atlas of mouse amacrine cell types¹¹. Data was downloaded from the Gene Expression Omnibus with accession number GSE149715 and parsed in Python.

Whole-Cell Patch Clamp Electrophysiology:

Electrophysiology was performed as previously described^3,12,13. Animals were euthanized by cervical dislocation and eyes dissected under ambient red light. Retinas were immediately transferred to a carbogen-saturated Ames’ medium (US Biologic) supplemented with 20 mM D-glucose and 22.6 mM NaHCO3 (pH 7.4, 285–295 Osm). Retinas were flat mounted onto a perfusion chamber receiving Ames’ media at a flow rate of 2 mL/min at 35°C (Warner Instruments). Individual astrocytes and Müller glia endfeet were identified by tdTomato fluorescence using a 40X water immersion objective on an Olympus BX50 microscope. Amacrine cells were blind-patched using IR imaging. Cells were patched onto in a whole-cell configuration using a borosilicate pipette (I.D. 0.86 mm, O.D. 1.5 mm; Sutter Instruments) filled with (in mM): 125 K-gluconate, 10 KCl, 10 HEPES, 10 EGTA, 4 Mg-ATP, 1 Na-GTP, 0.1 Alexa 488 (Invitrogen) or 0.8 Lucifer Yellow dye (Thermo-Fischer). The intracellular solution pH was 7.35 and osmotic concentration was 285 Osm. Pipettes containing intracellular solution had a resistance between 8 and 14MΩ. A holding potential was set at −80mV upon approach to help achieve a successful patch. After achieving a whole-cell configuration, we switched from voltage-clamp to current-clamp mode to record voltages responses to light stimulation. Light stimulation protocol was recorded in a current clamp configuration exposing cells to 525 nm light (full-field, 300 μW/cm2, 3-s duration, CoolLED, pE-4000). Raw data was parsed using the Python library pyABF and Clampfit 10.7. Current clamp light stimulation data was smoothed using the Savitsky-Golay algorithm in the Python library SciPy.

Results

Fibrous Cells with Astrocyte Markers are Found Sparsely in the Inner Plexiform and Ganglion Cell Layers of Mice.

Visualization of glial fibrillary acidic protein (GFAP)-expressing cells in the mouse retina reveals those which express the glial markers GFAP, S100β, Sox2, Sox9, and connexin-43 in the inner plexiform and ganglion cell layers (IPL and GCL, Fig. 1). These cells are sparse and primarily localized to the central retina (Fig. 1B), first appearing just beyond the annulus of intense GFAP labeling which surrounds the optic nerve head. They localize at a median distance of 431μm from the ONH center with an interquartile range of 355 μm (Fig. 1B). The highest concentration exists out to a radius of 800μm (80% of cells). Between 800 and 1600μm, the number drops sharply and accounts for 13% of cells. Between 1600 and 2200μm, few cells are observed (1%) but then reappear at a low frequency out to the retinal border (7%). We have observed at least one of these cells in 100% of the GFAP-Cre-tdTomato^fl/fl retinas we imaged (30 retinae from 15 animals) as well as retinae from non-transgenic strains 129S6 (2 retinas from 2 animals, Fig. S1A) and Cre-negative Ai14 mice on a C57BL/6 background (2 retinas from 2 animals, Fig. S1B). Additionally, the co-labeling of S100β, GFAP, and tdTomato demonstrates that our observation of these cells is not an artifact of developmentally leaky tdTomato expression (Fig. 1D). Due to their expression of astrocyte markers and atypical localization, we refer to these cells as displaced retinal astrocytes (DRA).

Fig. 1: Fibrous Cells with Astrocyte Markers are Found Sparsely in the Inner Plexiform and Ganglion Cell Layers of Mice.

(A) Cells with glial markers are found directly beneath and amongst cell bodies of the GCL and extend fibrous processes. 120 tdTomato-positive cell bodies were identified in the GCL and IPL of histological slices from 8 retinas of 4 mice. (B) The majority of cells are found within 1,600μm of the ONH center but are also found in the extreme periphery. Blue plot shows cell density as a function of the distance from the ONH center. Bars indicate SEM. The second distribution indicates individual cell locations that were quantified. Bars indicate median and interquartile range. 252 cells were analyzed from 9 wholemount retinae of 5 mice. These cells label positively for (C) Sox2, Sox9, (D) GFAP, S100β, and (E) connexin-43 (primarily localized to cell soma). White arrows indicate cell soma. Scale is 25μm for A, D, and full-size image in E while 15μm for C and zoomed E.

Displaced Astrocytes Differ in Morphology and Spatial Distribution to NFL Astrocytes

Immunolabeling for glutamate decarboxylase (GAD) and choline acetyltransferase (ChAT) can reveal distinct substrata of the inner plexiform layer. ChAT prominently labels sublamina 2 and 4, whereas signal for GAD is strongest in sublamina 1, 3, and 5^14,15. Labeling for these as reference points (not astrocyte markers) in conjunction with tdTomato visualization shows that cell bodies of displaced astrocytes localize to all substrata of the IPL as well as the GCL (Fig. 2). Soma shape is not regular, ranging from ellipsoid to spherical (Fig. 1–2, 4). Soma orientation can also be horizontal within a single retinal sublamina, or vertical, spanning multiple (Fig. 2). Processes similarly localize to all strata of the IPL, but do not continue into the outer plexiform layer (OPL).

Fig. 2: Displaced Astrocytes are Found in All Sublaminae of the Inner Plexiform Layer and Ganglion Cell Layer.

In the first column of each row, displaced astrocytes are indicated by white arrows. tdTomato-labeled cells that are not indicated by an arrow are nerve fiber layer astrocytes or Müller glia. * indicates Müller glia processes. Immunolabeling for choline acetyltransferase (ChAT) and glutamate decarboxylase (GAD) is used to delineate the IPL sublaminae. These borders indicate that displaced astrocytes in (A) have soma localized to S5 and S3 of the IPL. Those in (B) localize to the GCL, S5, and S2. Cells in (C) have soma in S5/S4, and S3. Some cells send vertical processes throughout the entire IPL. Panel (D) visualizes a region near the optic nerve head where multiple displaced astrocytes are observed in all IPL layers except S3. Dashed Boxes indicate the zoomed ChAT, GAD, and Merge column regions. 120 tdTomato-positive cell bodies were identified in the GCL and IPL of 8 retinas from 4 mice. Scale bar indicates 25μm.

Fig. 4: Displaced astrocytes have a morphology which differs from NFL astrocytes.

Displaced astrocytes are fibrous in appearance and exhibit many bead and bristle motifs along their processes (bristles at yellow arrows and beads at magenta arrows, zoom column). They lack large regions of flat membrane like their counterparts in the NFL exhibit. Each image shows a V5-labeled displaced astrocyte. Cell bodies are indicated with a cyan arrow. Blood vessels are indicated with purple lines. * indicates amacrine processes. 20 retinas from 10 G-MORF mice were analyzed. Scale bar indicates 25μm.

We recently described morphology of over a thousand well-separated, mouse retinal astrocytes in the NFL³. We found that NFL astrocytes are largely non-fibrous and contain extensive regions of flat membrane which make contact with neuronal and vascular structures in the inner retina. Additionally, they exhibit recurring microstructures, or motifs, which have characteristic cellular contact partners. In Fig. 3, we highlight some of this morphology to draw comparisons to displaced astrocytes. Beyond differences in retinal localization, displaced astrocytes contrast nerve fiber layer (NFL) astrocytes in their fibrous appearance, lacking large, flat regions of plasma membrane (Fig. 4). A minority of these cells sends flat protrusions into the NFL which resemble Müller glia endfeet and contact NFL astrocytes (Fig. 5). We additionally do not observe any displaced astrocytes engaging with the vasculature through complete enveloping connections as described previously (example NFL astrocyte with an enveloping connection in Fig. 3 iv)³. However, displaced astrocytes possess the bead and bristle motifs along their fibrous projections which are characteristic of thin processes of NFL astrocytes as well (Fig. 3–4)³.

Fig. 3: Nerve fiber layer astrocytes are uniformly distributed with largely non-fibrous morphology and extensive regions of flat membrane.

(A) Genetic architecture of the MORF3 and Ai14 mouse lines. When crossed with a Cre-expressing line, MORF3 animals have sparse (2%) labeling of cells with a membrane-directed, highly antigenic, V5-tagged fluorescent protein³⁹ and Ai14 animals have cytosolic expression of tdTomato. (B) NFL astrocyte density is relatively homogenous across retinal eccentricities. Example distribution from a Cre 77.6 × Ai14 retina representing all of its 4,489 NFL astrocytes. (C) NFL astrocytes are largely non-fibrous in their morphology with extensive regions of flat membrane to interact with neuronal and vascular structures of the NFL and GCL. Yellow boxes indicate zoomed regions in the far-right column. Dotted purple lines show where blood vessels are located. Scale for full-sized images is 25μm and for zoomed regions is 5μm. Magenta arrows indicate the neuronal-associated bristle (iii) and bead motifs in (vi). For a detailed description of NFL astrocyte heterogeneity from over 1,357 individual cells as well as motif nomenclature, see Holden et al., 2023.

Fig. 5: Some displaced astrocytes extend flat protrusions into the nerve fiber layer.

Some displaced astrocytes send flat protrusions, reminiscent of Müller glia endfeet, into the NFL. (A) Confocal images of the same cell at two different retinal depths (NFL and GCL). * In the NFL, the endfoot structure contacts astrocytes as indicated by GFAP labeling. (B) 3D reconstruction of the cell in A.

Displaced Astrocytes Are Not Amacrine Cells

Because their morphology more closely resembles amacrine cells than astrocytes, we investigated whether protein expression of displaced astrocytes was consistent with known amacrine cell subtypes. Using an open-source single-cell RNA sequencing dataset of amacrine cells, we looked for cells with a signature characteristic of displaced astrocytes (Fig. 6)¹¹. Of the 32,523 cells in the dataset, 173 express GFAP and 874 express S100β (Fig. 6B). Only 5 cells express both genes, with each of these also co-expressing Pax6 (Fig. 6B, D). Immunolabeling displaced astrocytes shows they are Pax6-negative (Fig. 6E). We also found evidence for the GFAP-positive, Pax6-positive amacrine cells in the dataset. They were unsurprisingly localized to the inner nuclear layer. Similarly, the dataset did not contain any cells which co-express both Sox2 and Sox9 while also expressing either GFAP or S100β (data not shown).

Fig. 6: Protein Expression of Displaced Astrocytes Does Not Match scRNA Sequencing Data for Amacrine Cells.

(A) Single-cell RNA expression profile of amacrine cells adapted from Yan et al., 2020. Numbers above each column indicate the percent of cells which express the indicated gene (have normalized gene expression levels above 0.1). Bars indicate mean expression level. Most amacrine cells do not express canonical glial markers. (B) Of the 32,523 amacrine cells in the data set, only 5 express both S100β and GFAP. (C) Most amacrine cells express Pax6. (D) Of the 5 amacrine cells indicated in B which express GFAP and S100β, all express Pax6. (E) Displaced astrocytes do not express Pax6. Pax6-positive, tdTomato-positive cells are found in the inner nuclear layer. These are likely the GFAP-positive, Pax6-positive amacrine cells in B. Arrows indicate cell bodies and dashed path indicates their outline.

Additionally, displaced astrocytes are negative for all amacrine cell markers probed (Fig. 7). In amacrine cells, syntaxin 1a, glycine transporter 1 (GlyT1), and glutamate decarboxylase (GAD) label the soma membrane in subtypes which express these genes. This labeling is absent in displaced astrocytes (Fig. 7A–C). Moreover, choline acetyltransferase (ChAT) and tyrosine hydroxylase (TH) label the entirety of select amacrine soma in their respective layers of the INL and GCL. Displaced astrocyte soma are negative for both markers as well (Fig. 7D, E).

Fig. 7: Protein Expression of Displaced Astrocytes is Not Consistent With Amacrine Cells.

Soma for displaced astrocytes are negative for a variety of amacrine cell markers. They lack positive immunolabeling for (A) syntaxin 1a, (B) glycine transporter 1, (C) glutamate decarboxylase, (D) choline acetyltransferase, and (E) tyrosine hydroxylase. The number of cells deconvolved and examined for colocalization are as follows and in the same order as above: 7, 6, 9, 9, and 7. Scale of full-size images is 15μm and for zoomed images is 3μm. * indicates blood vessel.

Displaced Astrocytes Contact Neurons, Glia, and Vasculature.

While labeling for amacrine markers, we noticed that displaced astrocytes often had depressions in tdTomato labeling (Fig. 8A). These depressions colocalize with amacrine marker signal which extends beyond the border of tdTomato, indicative of contacts with neuronal processes. To investigate this possibility further, amacrine cells superior to displaced astrocytes were blind-patched and filled with dye (Fig. 8Bi). Amacrine cell identity was determined through light response and morphology. We find that their dendritic arbor contacts displaced astrocytes along both their soma and processes (Fig. 8Bi–iii).

Fig. 8: Displaced astrocytes contact retinal neurons, glia, and the vasculature.

Displaced astrocytes contact amacrine cell processes. (A) Displaced astrocyte with depressions in tdTomato signal which is filled with glutamate decarboxylase signal (arrows) that continues beyond the tdTomato border. Scale full 15μm, zoom 3μm. (B) Patched amacrine cell (AC) superior to a displaced retinal astrocyte (DRA). (i) immunolabeling for tdTomato and Lucifer Yellow. Yellow arrow indicates patched amacrine cell soma. * Connections between displaced astrocyte and amacrine cell. Scale 25μm. (ii-iii) 3D rendering of confocal scene in i. Scale 5μm. (iv) Light responses for patched cell to verify amacrine identity. Two cells were patched in this way from different retinas. Scale 25μm. (C) Displaced astrocyte with soma in GCL/IPL (arrow) showing dye-coupling to GCL neurons. Scale 25μm. (D) Wholemount retina showing displaced astrocytes contacting other displaced astrocytes*, blood vessels, and Müller glia in the IPL. Full scale 50μm, zoom 25μm. (E) Retinal section labeled with tdTomato showing displaced astrocyte making connections with a NFL astrocyte, a blood vessel, and a Müller glia*. Scale 15μm. (F) A patched Müller glia endfoot with processes enveloping a displaced astrocyte soma. Scale (IHC) 10μm, (3D) 5μm.

Additionally, displaced astrocytes in the GCL and IPL were patched, and positive dye labeling was evident in both surrounding RBPMS-negative and, to a lesser degree, RBPMS-positive cells in the GCL (6/7 displaced astrocytes show dye spread, with the average distance to the furthest dye-positive cell being 261 ± 42μm², Fig. 8C). However, it should be noted that in order to patch onto these cells deep in the IPL without sectioning the tissue, positive pressure was maintained upon approach to avoid clogging the pipette tip. Some dye may be contributed to leakage from the pipette upon approach as both lucifer yellow and Alexa-488 are weakly capable of entering cells exogenously. Considering the scarcity of these cells, patching from sectioned retinas was not feasible.

Displaced astrocytes also contact the retinal macroglia and vasculature, making connections with blood vessels, NFL astrocytes, Müller glia, and one another (Fig. 8D–F). Whole-cell dye filling of individual Müller glia endfeet superior to displaced astrocyte soma reveals extensive contact between the two cell types (Fig. 8F).

Displaced and NFL Astrocytes Have Similar Responses to Light

Finally, we characterized cellular electrophysiology as a means for cell-type discrimination. We obtained whole-cell recordings of both displaced and NFL astrocytes as well as Müller glia endfeet (Fig. 9). Resting membrane potential was most similar between displaced astrocytes and NFL astrocytes (Displaced astrocytes: RMP = −75.7 ± 1.4 mV, n=7. NFL astrocytes: RMP = −74.5 ± 0.8 mV, n=7. Müller Glia: RMP=−81.71 ± 2.02mV, n=7). RMP was not significantly different between displaced astrocytes and NFL astrocytes (p=0.789, one way ANOVA, Tukey’s post-hoc) but did differ with Müller glia (p=0.014, one way ANOVA, Tukey’s post-hoc).

Fig. 9: Displaced Astrocytes Have Similar Light Responses to NFL Astrocytes and Müller Glia Endfeet.

(A) Displaced astrocytes in the GCL and IPL were patched in a whole-cell configuration and filled with Lucifer Yellow (i). At light onset and light offset, these cells respond with small graded depolarization potentials (n=7, RMP = −75.7 ± 1.4 mV, i-ii). (B) Astrocytes in the nerve fiber layer also respond with small depolarizations to light onset and offset (n=7, RMP = −74.5 ± 0.8 mV, i-iii). (C) Müller glia endfeet patched in the nerve fiber layer have similar responses but more negative resting potentials (n=7, RMP=−81.71 ± 2.02mV, i-iii). For all cells, the decay rate of the potential waveform during stimulus presentation varied from cell to cell. Scale indicates 25μm. Each displaced astrocyte recording represents a single cell from a single retina of a different mouse (n=7 mice). The NFL astrocyte recordings were from 4 retinas from 4 mice. The Müller glia endfoot recordings were from 5 retinas from 5 mice.

In response to light stimulation, current-clamp traces for all groups show small depolarizations to both light onset and offset (Fig. 9). Action potentials were not observed in response to either stimulus. For all cell types, the decay rate of the potential waveform during stimulus presentation varied from cell to cell (Fig. 9).

Discussion

Recently, we described morphological diversity of retinal astrocytes in the NFL³. Here, we describe a novel population of retinal glia which localizes to the GCL and IPL and expresses canonical astrocyte markers such as GFAP, S100β, Sox2, Sox9 and connexin-43 (Fig. 1–2)^3,16–18. We refer to these cells as displaced astrocytes. Of note in Fig. 1E, the connexin-43 (Cx43) labeling does not completely localize to astrocytes. Based on the amacrine cell scRNA sequencing from Yan et al. and a recently generated Cx43-reporter mouse, non-zero Cx43 expression is observed in a minority of amacrine cells (~0.3%)^11,19. Additional studies have also shown that Cx43 is expressed in rodent Müller glia both through immunolabeling and transgenic protein expression from Cx43-reporter mice^19,20. This likely accounts for our observation here as Cx43 is not a completely selective marker for astrocytes. However, morphology, gene expression, and electrophysiology rule out displaced astrocytes being either amacrine cells or Müller glia. In Fig. S2 we image through a single column of mid-retinal space, and the non-astrocytic IPL labeling of Cx43 is largely regional near the optic nerve head. It is possible that Cx43-positive amacrine cells found there are uniquely positioned to interact with astrocytes which also express it. In the NFL, the Cx43 labeling is selective for astrocytes (Manders’ coefficient of 0.968 ± 0.004, n=6 images). We have published using this antibody in the past with a Cx43 knockdown strain, and it is selective (reduces fluorescence labeling by 98% in retina and brain)⁹. This was consistent with the efficacy of the knockout, which was never 100% with oral gavage of tamoxifen.

Unlike NFL glia, displaced astrocytes are fibrous and stellate in appearance (Fig. 1, 3–4, 9). However, like NFL astrocytes, these fibrous processes harbor the bead and bristle motifs we described previously³. In that work, we found that these motifs were highly predictive of underlying neuronal architecture, signaling potential structural connections between displaced astrocytes and neuronal structures in the deeper retinal layers. In fact, we find that these cells contact not only neuronal architecture but also other displaced astrocytes, NFL astrocytes, Müller glia, and the vasculature- as do their counterparts in the NFL (Fig. 5, 8)^3,21. In addition to their thin processes throughout the IPL, some displaced astrocytes also extend protrusions into the NFL which form structures akin to Müller glia endfeet and make connections with NFL astrocytes (Fig. 5)²².

Because their retinal localization and morphology have neuronal characteristics, we investigated whether their expression profile matched gene and protein expression of known amacrine cell subtypes. We used the Yan et al. amacrine cell scRNA sequencing dataset to find that most amacrine cells do not express canonical glial cell markers (Fig. 6A)¹¹. Of the few cells which co-expressed GFAP and S100β, all expressed Pax6 (Fig. 6B, D–E). Because of this, none of the cells in this sequencing dataset correspond to displaced astrocytes as they do not express Pax6 (Fig. 6E). Moreover, we immunolabeled displaced astrocytes for common amacrine markers and did not find evidence for their expression (Fig. 7). It is also exceedingly rare for any species to harbor amacrine cells with soma in layers other than the ganglion cell or inner nuclear layers, providing further confirmation that this is an undescribed glial cell^23–28. Interestingly, immunolabeling for amacrine markers and dye filling displaced amacrine cells shows that their processes contact displaced astrocyte processes and soma (Fig. 8A–C). This opens the possibility that direct communication occurs between displaced astrocytes and retinal neurons, potentially influencing electrical activity and corresponding neuronal computations. In line with this, we found limited evidence for gap junctional coupling between these two populations and that displaced astrocytes express the gap junction protein connexin-43 (Fig. 1, 8C, S2).

We also recorded whole-cell patch clamp electrophysiology from displaced astrocytes using in-tact retinae (Fig. 9). These cells show small amplitude, graded depolarization to both light onset and offset. These responses are consistent with our recordings of NFL astrocytes and Müller glia (Fig. 9)²⁹. Interestingly, in Müller glia, these depolarizations are thought to arise in part due to calcium and potassium currents in the plexiform layers³⁰. Because displaced astrocytes are intimately connected to Müller glia and reside predominantly in the IPL, it is likely that their light response is driven by similar currents, either independently or directly through Müller glia which they contact (Fig. 8D–F). It is also possible that these currents are tied to the coupling observed between displaced astrocytes and amacrine cells which produce graded potentials to light onset and offset (Fig. 8B, C). Action potentials were not observed in the current clamp recordings which means they cannot be any neuronal population which produces spikes instead of graded potentials. Furthermore, it is noteworthy that displaced astrocytes are found throughout all substrata of the IPL while also having both ON and OFF responses to light (Fig. 2, 9)^22,31,32.

In our previous work, we found that retinal eccentricity and local neurovascular architecture dictates NFL astrocyte morphology³. These cells adopt morphology which is capable of interacting effectively with cellular substrate in their environment. In the NFL, this morphology often consists of large membranous regions which can effectively cover ganglion cell axon bundles or neural soma in the GCL³. In the inner plexiform layer, this morphology is not possible due to the vast dendritic arborization of amacrine and retinal ganglion cells (Fig. 7–8A–B). Immunolabeling for amacrine markers, cellular dye filling, and their location in the IPL suggests that displaced amacrine cells directly communicate with neuronal cells in the GCL and IPL (Fig. 1–2, 7, 8C). Instead of large, membranous sheets, the IPL environment requires slender, fibrous morphology to efficiently weave between the dendritic architecture. It is reasonable then to assume that like NFL astrocytes, displaced astrocytes adopt their morphology in response to their local environment. This feature is likely important in disease where local neurovascular architecture changes, likely promoting compensatory structural changes in astrocytes^18,33.

There is growing appreciation that the optic nerve head lamina region is a stem cell niche, capable of producing all macroglial cell types of the anterior optic nerve, including astrocytes³⁴. Because of their high concentration at the optic nerve head and central to peripheral gradient, it is possible that displaced astrocytes arise from this population of progenitor cells and migrate into the retina. While they are positive for the Yamanaka factor Sox2 (Fig. 1), they are negative for KLF4 (data not shown). Full lineage tracing and studies which follow this population throughout the course of aging and diseases like glaucoma which affect the optic nerve head would be valuable additions to the literature.

Displaced astrocytes have gone unrecognized in both their presence in the murine retina and their morphological distinction from NFL astrocytes. However, other species feature glial cells in their IPL which resemble displaced astrocytes in morphology and, in part, their gene expression^35,36. In the chick retina, non-astrocytic inner-retinal glia-like (NIRG) cells inhabit the IPL³⁵. While they express Sox2 and Sox9, they do not express GFAP so cannot be the same population³⁵. However, because of the similarity in structure and localization, both cell types may perform similar functions or have a common response to tissue damage^37,38. Single-cell RNA sequencing of displaced astrocytes may reveal unexpected parallels between the mammalian and avian retinae. Because our data shows the utility in morphological classification of retinal astrocytes, future studies seeking to examine heterogeneity in response to injury and disease would likely benefit from incorporating morphological information into their existing transcriptomic approaches.

A natural question that arises from our study is, why have displaced astrocytes gone unrecognized thus far in the mouse retina? We believe there are a few reasons for this. The first is likely due to their scarcity. In a given retina, displaced astrocytes number in the tens of cells whereas there are multiple thousands of NFL astrocytes (Fig. 1, 3B) Although the GFAP-Cre-tdTomato^fl/fl mouse is a commonly used cross to visualize astrocytes, tdTomato is also expressed in non-astrocytic cell types such as Müller glia and occasionally amacrine cells (Fig. 1A, 6E). Because of this, it is likely that these cells have been observed before but were assumed (as we have for years using these mice) to be amacrine cells and were thus ignored. We verify in this paper that the observance of displaced astrocytes is not due to leaky expression of fluorescent markers. These cells are also observed using S100β labeling, revealing displaced astrocytes in not only transgenic C57s but also non-transgenic 129S6 mice and Cre-negative Ai14 mice (Fig. S1). Our observation that displaced astrocytes represented a distinct astrocyte population arose in our previous work using a MORF3 mouse line that sparsely labels GFAP-expressing cells³. Because of their genetics (Fig. 3A), only ~2% of total cells expressing GFAP label with the V5-fluorescent protein. We imaged thousands of V5-labeled astrocytes, and in the process observed a small number of displaced astrocytes in these MORF3-derived retinas (0–5 per retina). Their morphology drew our attention, and because we co-labeled with GFAP at the time, we could easily verify that these cells were GFAP-positive and not expressing V5-fluorescent protein as an artifact. The frequency of V5-positive labeling for displaced astrocytes also points to greater numbers than we observe in GFAP-Cre-tdTomato^fl/fl mice (5 labeled cells at 2% labeling suggests 250 total cells per retina). This leads to the hypothesis that during retinal development, a smaller number of these cells may serve as progenitors which directly pass on expression of the V5-fluorescent protein to daughter cells (frameshift mutation in the Rosa26 locus). This idea is supported by occasional observation of these cells in clusters (Fig. S1A). An additional possibility is that astrocytes which form a ring about the optic nerve head may migrate out of this region through the IPL, becoming displaced astrocytes.

Displaced astrocytes are also difficult to observe in histological sections due to sampling density. Sectioning a single mouse eye generates well over a hundred sections, even when the sections are thick (30μm). Unless each section were imaged, these cells would be easily missed, especially if optic nerve head sections were not of interest. An additional reason why they may have gone unnoticed is that they are found in the GCL and IPL. When imaging with non-confocal microscopy, as is commonly done, the distinction between layers in wholemount retinas is difficult to distinguish as you gather light from a thicker plane. Finally, until our paper detailed full membranous morphology for individual astrocytes, we lacked a sufficient understanding of baseline astrocytic morphology³. Without this, it would have been difficult to recognize a morphological subtype of astrocyte simply due to the large population-level heterogeneity. A combination of all these factors helps explain why these cells have not been described before.

Supplementary Material

Fig S1

Fig. S1: Displaced astrocytes are also found in non-transgenic 129S6 mice and Cre-negative Ai14 mice. In addition to the transgenic G-MORF and GFAP-Cre-tdTomato^fl/fl animals used in this study, we observe displaced astrocytes in the non-transgenic 129S6 mouse line as well as Cre-negative Ai14 animals. (A) To determine whether observation of displaced astrocytes is strain-specific, we labeled for S100β in 129S6 mice. The cells are present in these mice as well. The first row shows a cluster of displaced astrocytes (i-ii). Scale is 50μm in i and 25μm in ii. The second row shows two additional example cells in a separate retina from another mouse. (2 retina from 2 mice, 100% have displaced astrocytes). Scale is 25μm in iii-iv. (B) Cre-negative Ai14 mice (C57BL/6 background) also have displaced astrocytes as indicated by S100β labeling. 2 examples are shown from two different retina. (2 retina from 2 mice, 100% have displaced astrocytes). Scale is 25μm.

Fig S2

Fig. S2: Specificity of Cx43 antibody labeling. (A) Testing Cx43 antibody specificity and normal appearance in nerve fiber layer (NFL) astrocytes. Cx43 colocalizes with tdTomato produced in NFL astrocytes. Manders’ colocalization coefficient (Cx43 signal within NFL tdTomato signal) for manually thresholded images is 0.968 ± 0.004 (n=6 images, 226×322 μm²). (B) Shows NFL astrocytes from G-MORF mice. The full membranous morphology helps build the point in A that the Cx43 signal is localized to astrocytes. (C) Cx43 expression is largely localized to the NFL in the mid retina. Each quadrant corresponds to a single slice in a single column of retinal space (NFL, GCL, IPL, INL) where no displaced astrocytes are present. (D) Additional example of Cx43-positive displaced astrocyte near the nerve head. Labeling is most intense at blood vessel contacts (yellow dotted line), soma, and process terminals (white arrow, zoom 1). tdTomato-negative, Cx43-positive labeling in some regions is linear, suggesting neuronal process expression (white arrow, zoom 2). This is strengthened by the last row image showing a stack of the entire IPL. Distinct tracks of Cx43 labeling are not random. Scale is 25μm for all images except zoom insets which are 15μm.

Main Points:

1. Fibrous cells with glial markers are found in the mouse inner plexiform and ganglion cell layers.

2. These cells are distinct from nerve fiber layer astrocytes and Müller glia. While they resemble amacrine cells, they are not neuronal.