Development of retinal astroglia

Abstract

Müller cells and retinal nerve fiber layer astrocytes are the major astroglia of the mammalian retina. They have numerous important functions in adulthood for maintaining neuronal homeostasis, and also in developing retina where they facilitate key events in the assembly of the retinal tissue. Recent years have seen substantial progress in understanding how these astroglial cells develop, and how their development shapes the cells around them. Here we review the mechanisms underlying the formation, maturation, and spatial patterning of Müller glia and retinal astrocytes, with an emphasis on how they acquire their functional properties. We focus on developmental events that have a major impact on overall retinal integrity, such as formation of neuro-glial junctions at the outer limiting membrane and patterning of retinal astrocytes into a template that guides angiogenesis. Finally we discuss examples of retinal diseases that have their origins in developmental defects affecting Müller cells or retinal astrocytes. These include certain classes of inherited retinal degenerations, as well as retinopathy of prematurity.

I. Introduction: Glial cells in the developing retina

Glial cells, once a neuroscientific afterthought, are now the hottest topic in developmental neurobiology. Recent advances have been driven by studies throughout the central nervous system (CNS), and the retina is no exception. In fact, because the glial milieu in the mammalian retina differs in important ways from the brain and spinal cord, the retina offers an opportunity to gain a different perspective on how glial cells contribute to nervous system form and function. Here we review recent progress in understanding development of retinal astroglia – the Müller cells and the retinal nerve fiber layer (RNFL) astrocytes – with an emphasis on how their mature functions are established. We also touch on the emerging literature linking defects in retinal astroglial cell development to vision diseases.

I.a. Müller glia

Müller cells are the most numerous astrocyte-like cell type in the vertebrate retina. Their functions in mature retina are quite similar to protoplasmic astrocytes in the cerebral cortex: They mediate vital homeostatic functions like neurotransmitter recycling, water flux, and potassium buffering (Reichenbach & Bringmann 2019, Vecino et al. 2016). However, unlike cortical astrocytes, Müller cells are radial glia that span the entire vertical extent of the neural retina (Fig. 1), contacting all retinal layers from the inner limiting membrane (ILM) to the outer limiting membrane (OLM). While their radial morphology makes them somewhat unusual among astrocytic cells of the mammalian CNS, it is quite common for radial glia to mediate astrocytic functions within the CNS of fish and birds (Lyons & Talbot 2014). Thus, Müller cells – along with radial glia such as hypothalamic tanycytes and cerebellar Bergmann glia – have deep evolutionary roots. Radial morphology is also important to Müller cell function, endowing them with important structural roles in maintaining retinal laminar integrity (MacDonald et al. 2015, Prieto-Lopez et al. 2024, Rich et al. 1995).

Figure 1: The mammalian retina and its astroglial cell types.

A) Drawing of a mature mammalian retina with all of the principal cell types represented.

B) Individual astrocytes of adult mouse retina (adapted from Holden et al., 2023) demonstrating the variety of distinct astrocyte morphologies. Scale bars, 25 μm.

C) Individual Müller cell of bovine retina, illustrated by Santiago Ramón y Cajal (1893).

D) Drawing of embryonic mouse retina (Ramón y Cajal, 1893) illustrating the morphology of radial glial progenitor cells. The progenitors have attachments at the apical side (top), while newborn neurons are located basally.

Abbreviations: Hz, horizontal cell; RGC, retinal ganglion cell; RNFL, retinal nerve fiber layer; OLM, outer limiting membrane; ONL, outer nuclear layer; OPL, outer plexiform layer; INL, inner nuclear layer; IPL, inner plexiform layer; GCL, ganglion cell layer.

Müller cell radial morphology hints at their ontogeny: they are produced by the differentiation of radial glial progenitor cells that remain at the conclusion of retinal neurogenesis. Radial progenitors have a much simpler morphology than mature Müller cells (Fig. 1D). During their transition into Müller glia, radial progenitor cells ramify fine arbors within the retinal synaptic layers (the outer and inner plexiform layers, OPL and IPL); they further elaborate their endfeet and extend branches to contact vasculature; they form adherens junctions with photoreceptors that comprise the mature OLM; and they branch within the nuclear layers to enclose neuronal somata (Ramón y Cajal 1893) (Fig. 1C). This differentiation program endows Müller cells with remarkably distinct anatomical specializations within each retinal layer, which presumably reflect the distinct functions that Müller cells need to accomplish within each layer (Fig. 1C).

I.b. Astrocytes of the retinal nerve fiber layer

Whereas Müller cells are generalists – they are found in every vertebrate species, within every retinal layer, interacting with every other retinal cell type – retinal astrocytes are much more specialized. First of all, they are found only in the RNFL, where RGC axons course to the optic nerve head (Fig. 1A). As such, their main intercellular interactions are with axons and with the primary plexus of the retinal vasculature. Astrocyte morphology is fibrous, although the particulars of their morphology can vary substantially (Fig. 1B) (Holden et al. 2023). With their fibrous morphology, their presence in an axon-dominant region, and their high expression of glial fibrillary acidic protein (GFAP), RNFL astrocytes bear a strong resemblance to the white-matter astrocytes of the brain and spinal cord (Köhler et al. 2021). Thus, axon-associated astrocytes throughout the CNS share certain key features, despite the absence of myelin within the retina.

In adult retina, the functions of RNFL astrocytes likely overlap substantially with the RNFL endfeet of Müller glia. However, there are morphological differences in how the two glial cell types interact with axons. While Müller endfeet fill spaces between RNFL axon bundles, astrocytes insinuate their fine processes into these bundles to contact and partially wrap individual axons (Büssow 1980, Holden et al. 2023). The function of these direct axonal contacts is not known; but because brain white matter astrocytes also do this (Köhler et al. 2021), it is likely they are important for maintaining neuronal homeostasis.

A second type of RNFL astrocyte specialization is that, unlike Müller glia, these cells are found only in mammals (Schnitzer 1988). This specialization reflects the intimate relationship between retinal astrocytes and vasculature. Whereas most vertebrate species rely on choroidal and/or vitreal blood supplies to oxygenate the retina, mammals have evolved an intraretinal vascular system, comprised of a primary plexus in the RNFL and a deep plexus at the outer plexiform layer. Development of this retinal vascular system depends fundamentally upon RNFL astrocytes, in two main ways. First, at the onset of angiogenesis, astrocytes are the sole retinal source of the potent pro-angiogenic cue VEGF-A (Fig. 2C), which is required to drive the entry of endothelial cells into the retina at the optic nerve head. If the astrocytic source of VEGF-A is removed, either by genetic ablation of astrocytes or by astrocyte-specific deletion of the mouse Vegfa gene, vessels fail to enter the retina at all (O’Sullivan et al. 2017, Rattner et al. 2019, Tao & Zhang 2016). Second, prior to angiogenesis, immature astrocyte cell bodies and arbors become patterned into a honeycomb-like network within the RNFL. This network prefigures the ultimate pattern of the capillary network that will form over it (Fig. 2C). Subsequently, angiogenesis spreads outward from the optic nerve head through the RNFL, in a wave-like manner, led by specialized endothelial tip cells that crawl precisely over the astrocyte honeycomb template (Fig. 2A,D). In this way, astrocytes dictate the pattern of the developing capillaries (Dorrell et al. 2002, Fruttiger et al. 1996, Gerhardt et al. 2003). Conversely, astrocytes are absent from any avascular regions of mammalian retina, such as the primate fovea, suggesting that developing vessels cannot go where astrocytes do not go first (Ogden 1978, Provis et al. 2000, Schnitzer 1985, Schnitzer 1987, Schnitzer 1988, Stone & Dreher 1987). Together, these observations suggest that RNFL astrocytes evolved in the mammalian lineage alongside intraretinal vasculature, to promote and guide vascular development.

Figure 2: Colonization of the retina by astrocytes and vasculature.

A) Astrocytes and vasculature colonize the mouse retina via sequential waves of migration outward from the optic nerve (asterisk). Illustrations show progression of astrocytes (top) and vasculature (bottom) across the retina at three different stages of mouse development.

B) Cross-sectional view of an E12 mouse retina highlighting the fate-committed progenitor region that generates RNFL astrocytes. Astrocytes migrate out from this region to colonize the RNFL (A)

C) Top: Morphology of the astrocyte honeycomb-like template that guides angiogenesis. Green, astrocyte cell surface marker (anti-PDGFRα); magenta, vascular cell surface marker (IB4 lectin). Arrows: Endothelial tip cells at the advancing vessel wavefront; arrowheads: tip cell filopodia. Bottom: Phenotypic differences between astrocyte precursor cells (APCs) located ahead of the vascular wavefront and mature astrocytes behind the wavefront. VEGF-A (green) is highly expressed by APCs. Red, pan-astrocyte marker (Sox9). Blue, vasculature (IB4 lectin).

D) High-magnification view of endothelial tip cells (arrows) and tip cell filopodia (arrowheads). Scale bars: 50 μm (C); 25 μm (D).

II. Development of Müller glia

II.a. Generation of Müller glia: From radial progenitor to Müller cell and back

During embryonic development, the retina is comprised of proliferative neuroepithelial progenitor cells (Fig. 1D) that will ultimately give rise to all retinal neurons and Müller glia. Progeny of dividing progenitors may remain in the progenitor pool, or they may exit the cell cycle to become a neuron. Later, near the end of neurogenesis, remaining progenitors begin to lose their competence to generate neurons. At this point, they exit the cell cycle and initiate a differentiation program that converts them into Müller glia. Differentiation begins around postnatal day (P)4-5 in mice and lasts until about P10, when most anatomical features of mature Müller cells are established (Wang et al. 2017).

Mature Müller glia have a lot in common with the progenitor cells that give rise to them: they share a radial morphology, some functional aspects including formation of inner and outer retinal borders, and a remarkably similar overall transcriptional profile (Nelson et al. 2011). Thus, the differentiation process is far more reversible than for a neuron. Indeed, Müller cells of certain fish and amphibians can de-differentiate in response to injury, returning them to a neurogenic progenitor state to replace lost neurons. While such reversals do not normally occur in mammals, there is great interest in figuring out how to induce de-differentiation and neurogenesis for the purpose of regenerative therapies. We will not cover this topic in detail, but readers are referred to several excellent recent reviews (Lahne et al. 2020, Pavlou & Reh 2023, Taylor et al. 2024).

In this section we highlight key molecular pathways controlling each step of the Müller cell fate and differentiation program. Any such discussion must begin with Notch signaling, because this pathway influences each step in Müller gliogenesis. Notch receptors are widely used throughout the animal kingdom to control cell fate determination, usually in a manner that opposes neurogenesis. Activation of Notch signaling within neural progenitor cells, including retinal progenitors, directly induces expression of basic helix-loop-helix (bHLH) transcription factors of the Hairy and Enhancer of Split (HES) family. Key HES factors in the retina are Hes1 and Hes5 (Bao & Cepko 1997). HES factors block neurogenesis by repressing expression of another set of “proneural” bHLH factors. Conversely, the proneural factors can repress HES genes. This cross-repression creates a winner-take-all scenario: In cells with high Notch signaling, HES genes gain the upper hand leading to repression of proneural genes; but in cells with low Notch signaling, lowered HES expression allows proneural genes to “win” the cross-repression battle (Beatus & Lendahl 1998).

This Notch-driven tug of war between HES and proneural factors is used repeatedly throughout Müller cell genesis – it controls glial fate selection, differentiation, and even the maintenance of the adult Müller phenotype. At each stage, Müller development is promoted by high Notch signaling activity and inhibited when proneural genes gain the upper hand. First, during neurogenesis, the frequency with which the proneural pathway overcomes the Notch-HES pathway determines how many progenitor cells are left at the end of the neurogenic competence window to become Müller glia. Loss of Notch, Hes1, or other downstream Notch effectors leads to precocious neural differentiation, depletion of the progenitor pool, and decreased Müller cell production (Bosze et al. 2023, Jadhav et al. 2006, Riesenberg et al. 2009, Yaron et al. 2006, Zheng et al. 2009). Conversely, genetic enhancement of Notch signaling preserves a larger progenitor pool, overproducing Müller glia at the expense of neurons (Furukawa et al. 2000, Scheer et al. 2001).

Second, near the end of neurogenesis when progenitors are starting to give rise to Müller glia, Notch acts at the time of cell cycle exit to control a binary neuron-glial fate choice. Progenitors in which proneural genes gain the upper hand during the final cell cycle become late-born neurons, such as rod photoreceptors or bipolar cells; whereas those that maintain high levels of Notch signaling become Müller cells (Clark et al. 2019, MacDonald et al. 2015). Accordingly, when Notch1 or Hes5 is removed from late progenitors, they fail to adopt glial fates and instead become neurons (Hojo et al. 2000, Mizeracka et al. 2013). Conversely, overexpression of Hes1 or Hes5 in late progenitors promotes glial fate at the expense of neurons (de Melo et al. 2016a, Furukawa et al. 2000, Matsuda & Cepko 2007).

Finally, after cell cycle exit, Notch signaling promotes Müller glia differentiation – that is, the transformation of radial progenitors into mature Müller cells. This was shown by pharmacological inhibition of Notch signaling in larval zebrafish or mouse retinal explants. The Notch inhibitor DAPT was applied at a time when most Müller cells had exited the cell cycle but had yet to express genetic or anatomical features of mature Müller cells (MacDonald et al. 2015, Nelson et al. 2011). In both species, DAPT blocked Müller cell differentiation, including elaboration of specialized lamina-specific arbors in the zebrafish case. Remarkably, the requirement for Notch signaling continues even after differentiation is complete. A tamoxifen-CreER system was used to delete the Notch transcriptional effector Rbpj selectively from adult mouse Müller glia. This caused a rapid dedifferentiation of Müller cells, as judged by single-cell sequencing, returning them to a progenitor-like state (Le et al. 2024). Thus, Notch signaling is required not only to establish but also to maintain the adult Müller glial phenotype.

In order for Notch signaling to mediate so many different aspects of Müller cell development, Notch acts in concert with other stage-specific transcriptional mechanisms to implement each stage of gliogenesis. During embryonic development, transcription factors Rax, Lhx2, and Sox2 all perform similar functions in support of Notch signaling, acting to preserve the progenitor pool (de Melo et al. 2016b, Rodgers et al. 2018, Taranova et al. 2006). Lhx2 accomplishes this by directly upregulating Notch signaling through binding to cis-regulatory elements of Hes5. Progenitor pool maintenance is also regulated by the polycomb repressive complex 2 (PRC2), which represses gene expression by catalyzing trimethylation of histone H3 lysine 27 (H3K27). The Hes1 locus is normally H3K27 trimethylated during neurogenesis, but this repressive mark is lost in the absence of PRC2, leading to Hes1 upregulation and precocious Müller glia differentiation (Iida et al. 2015, Ueno et al. 2017, Zhang et al. 2015).

At the late progenitor stage, when progenitors are choosing between late-born neuron or Müller glial fates, a distinct set of transcription factors impinges on Notch signaling to bias the choice towards glia. Ikzf4, a zinc finger transcription factor, promotes gliogenesis by directly binding to cis-regulatory elements that upregulate Notch pathway genes, including Hes1 (Javed et al. 2023). Nuclear Factor I (NFI)-family transcription factors, NFIA, NFIB, and NFIX, can also upregulate Notch pathway components in late progenitors. Triple deletion of all three NFI genes prevents retinal progenitors from adopting any late-born fates, including bipolar cells and Müller glia, illustrating the importance of these factors (Clark et al. 2019). A third example is the Rax homeodomain factor: When Rax is selectively deleted from late progenitors, Müller cells are overproduced at the expense of rods (Yoshimoto et al. 2023). Direct interactions between Rax and Notch pathways have not been explored, but seem likely given their similar effects on both embryonic and late progenitors.

Finally, at the differentiation stage, a partially overlapping set of transcription factors cooperates with Notch to promote acquisition of mature Müller cell properties. Late progenitor deletion of Lhx2, or of individual NFI factors, mainly affects Müller cell differentiation without major effects on cell fate. In each case, the glial phenotype involves loss of radial morphology and OLM junctions, loss of early Müller cell markers, and enhanced Müller cell death (Clark et al. 2019, de Melo et al. 2016b). The HMG-box transcription factors Sox2, Sox8, and Sox9 influence astrogliogenesis throughout the CNS, including the retina. While it was initially suggested that they might influence the rod-versus-Müller cell fate choice (Muto et al. 2009), evidence from mouse gain- and loss-of-function studies is more consistent with a role in differentiation (de Melo et al. 2016a, Lin et al. 2009, Muto et al. 2009, Poche et al. 2008). There is likely to be cross-talk between the NFI and HMG-family transcription factors during Müller differentiation because, in the spinal cord, NFI-mediated expression of glial differentiation genes involves heterodimerization with Sox9 (Kang et al. 2012). In addition to these transcriptional mechanisms, cell-extrinsic signaling through bone morphogenetic proteins also promotes molecular and morphological differentiation of Müller cells (Ueki et al. 2015). One important open question is why so many different factors are needed for Müller differentiation. Future studies will need to determine whether the individual factors can be arranged into a linear pathway, or if instead there are multiple independent pathways that control different aspects of the mature Müller phenotype.

II.b. Functional maturation of Müller glia

As part of their differentiation program, Müller glia acquire specialized anatomical and molecular features that support their mature functions. In general, far less is known about these developmental events than the earlier processes of specification and terminal differentiation. In this section, we highlight several particularly important phenotypes expressed by mature Müller glia about which open developmental questions remain.

II.b.1. Layer-specific and regional specialization.

One of the most notable features of Müller glial anatomy is how individual cells adopt strikingly different morphologies within each retinal layer (Fig. 1C). This observation suggests that local cues from the neuronal elements present within each layer sculpt the growing arbors of Müller cells. For example, several lines of evidence suggest that development of fine arbors in the IPL likely involve interactions with synapses. First, arbor morphology varies across IPL sublayers containing different types of synapses (Fig. 1; Wang et al., 2017). Second, the timing of arbor elaboration is well matched to the timing of synaptogenesis (Wang et al. 2017, Williams et al. 2010). During the synaptogenic period, live imaging in zebrafish and mice shows that Müller IPL arbors are highly dynamic, whereas arbors cease moving at a time corresponding to the end of synaptogenesis (Fisher 1979, Tworig et al. 2021). Neural activity or retinal waves do not seem to control the dynamic morphogenesis of Müller arbors. However, a recent CRISPR screen in zebrafish identified a number of genes required for elaboration of Müller cell arbors in particular layers, paving the way for mechanistic studies to address how arbors form (Charlton-Perkins et al. 2019).

In addition to adapting their anatomy in a layer-specific manner, Müller cells also adapt to regional specializations such as the primate fovea (Reichenbach & Bringmann 2019). At the foveola – the center of the foveal pit – a small handful of Müller cells adopt a distinctive shape that differs from peripheral glia or even those elsewhere within the fovea. Because these foveolar glia do not interact with the retinal layers containing neuronal somata or synapses, their morphology is much simpler than a textbook Müller cell. Additionally, the basal processes of foveolar Müller cells fan out to form a cone-shaped glial structure that lines the inner border of the foveal pit (Bringmann et al. 2018, Gass 1999, Syrbe et al. 2018). This “glial cone” is thought to provide important structural support to the thinnest part of the retina within the foveal pit (Syrbe et al. 2018). The glial cone also contains high levels of macular pigment, absorbing blue light, which is thought to enhance visual acuity by diminishing the impact of chromatic aberration on the red and green cones that populate the fovea.

In accordance with their distinct anatomy and function, Müller glia of the fovea are also molecularly distinct, as shown by single-cell sequencing and immunohistochemical studies (Bringmann et al. 2018, Masri et al. 2025, Zhang et al. 2024). Indeed, in marmoset, fovea-to-periphery gene expression differences were greater for Müller glia than for any neuronal cell type. How these regional distinctions arise during development remains an important open question. They could reflect glial adaptation to the needs of the local foveal environment; or alternatively, foveal glia could be a distinct Müller cell type specified during gliogenesis. To distinguish between these models, it will be important to determine whether foveal Müller cells are already molecularly distinct at embryonic stages when the fovea begins to form.

II.b.2. Tiling and spatial patterning.

Throughout the CNS, astroglial arbors insinuate themselves into any and all empty spaces larger than a synaptic cleft. Electron micrographs show that Müller cell processes fill the spaces between neuronal cell bodies in the nuclear layers, and between synapses in the plexiform layers (Bringmann et al. 2006, Burris et al. 2002, Grimes et al. 2024). This anatomy reflects the fact that, to perform their structural and physiological functions, astroglial arbors need to be present throughout the retinal neuropil. In addition to ensuring that essential functions such as neurotransmitter recycling and potassium buffering are universally available, space-filling itself likely serves a structural role.

To achieve their space-filling goal, developing Müller cell arbors grow according to a developmental rule known as tiling. The rule specifies that arbors should grow until they touch the arbors belonging to a neighboring Müller cell. When arbors grow by such a rule, the result is a mature arrangement whereby each Müller cell occupies a unique neuropil territory, minimally overlapping adjacent arbors (Fig. 3). The notion that Müller cell arbors develop according to such a rule is supported by live imaging in zebrafish as well as anatomical studies in mice, both of which show that tiling arises upon first contact between neighboring arbors during their initial outgrowth (Wang et al. 2017, Williams et al. 2010). Müller arbors tile within all retinal layers despite their substantial morphological differences (Fig. 3). Even the fine arbors contacting vasculature exhibit tiling, such that neighboring glia collaborate to completely surround capillaries (Grimes et al. 2024).

Figure 3: Müller cell tiling.

Adult mouse retinas in which Müller cells express a combination of membrane-targeted fluorescent proteins through the Brainbow virus system. Left, cross-section view. The entire Müller cell is labelled from endfeet at RNFL to microvilli at OLM. Note lack of overlap between adjacent cells at all retinal layers. Right, en face view of a retinal wholemount, imaged at the IPL level to show synapse-associated Müller arbors. The lack of arbor overlap is best appreciated in high-magnification view (inset). Abbreviations as in Fig. 1. Scale bars, 10 μm.

Nonoverlapping territories do not arise by chance but instead likely arise through local cell-cell interactions (Wang et al. 2017). Accordingly, ablation of an individual Müller cell in zebrafish induces neighboring cells to grow laterally and take over its territory (Williams et al. 2010). Astrocyte tiling has also been documented in the CNS of many species, including Drosophila, zebrafish, mice, and ferrets (Bushong et al. 2002, Chen et al. 2020, López-Hidalgo et al. 2016, Stork et al. 2014). Thus, the functional impetus for tiling, as well as the developmental mechanisms that produce it are likely to be broadly shared.

The molecular basis for cell-cell repulsion during establishment of tiling remains unclear. A zebrafish CRISPR screen identified multiple genes affecting Müller cell tiling, which may provide a molecular handle for seeking these repulsive recognition molecules; however, the screen did not identify such molecules directly (Charlton-Perkins et al. 2019). A recent study in cortical astrocytes focused on HepaCAM, a member of the immunoglobulin superfamily that is highly expressed on the astrocyte cell surface. Sparse deletion of HepaCAM from individual astrocytes in mouse cortex leads to a dramatic reduction in territory size, whereas broad deletion in all astrocytes has no effect (Baldwin et al. 2021). This finding suggests that mutant astrocytes compete unsuccessfully for territory with surrounding wild-type astrocytes. The competition appears to involve encroachment of wild-type cells into mutant astrocyte territories, as neighboring wild-type and mutant astrocytes overlap more with each other than do astrocytes of the same genotype. Together these results establish a role for cell-surface HepaCAM levels in mediating competitive homotypic interactions underlying both astrocyte tiling and establishment of arbor size. HepaCAM is also expressed by Müller glia, but the role of HepaCAM in the retina has yet to be studied.

To ensure space-filling, there is also a need for developmental control over Müller cell number and spacing – if the cells are too sparse or too unevenly spaced for their arbors to touch, tiling cannot occur. Accordingly, Müller somata are tightly packed against one another within the zone of the inner nuclear layer (INL) where they reside. This cell-to-cell packing imposes order and regularity onto the Müller cell array (Wang et al. 2017). In using a passive mechanism like cell packing to control soma spacing, Müller cells stand in contrast to retinal neurons, which can achieve regular “mosaic” spacing of their somata via active homotypic repulsion mechanisms (Keeley et al. 2020). However, the spatial order achieved by soma packing is sufficient to ensure that cells are close enough together to engage the tiling mechanism. Packing depends on a developmental mechanism that confines Müller cell bodies within a single substratum of the INL; if somata were free to reside throughout the INL, their density would not be high enough to ensure neighbor contact. This laminar restriction mechanism is poorly understood, but it seems to involve genes expressed during Müller cell differentiation, because mutations that disrupt differentiation cause a loss of soma laminar confinement (Charlton-Perkins et al. 2019, Clark et al. 2019, Surzenko et al. 2013).

II.b.3. Formation of the outer limiting membrane.

The OLM is one of the most important functional specializations of Müller glia, with key structural and barrier functions that are necessary for photoreceptor survival. It is not really a “membrane” – rather, it is a network of adherens junctions that connects the apical-most processes of Müller glia with photoreceptor inner segments and with each other (Fig. 4). This junctional network is quite similar, in molecular structure and function, to the network of adherens junctions connecting the apical side of most epithelia, including the neuroepithelium of the developing CNS. Apical epithelial junctions provide structural integrity to tissues and regulate the passage of solutes. In the embryonic CNS, including the retina, progenitor apical junctions are a key part of the barrier between the neural parenchyma and the ventricles (Aaku-Saraste et al. 1996, Chenn et al. 1998). By adulthood, when neuroepithelial progenitors are gone, most regions of the CNS have transferred this barrier function to ependymal cells and subependymal astrocytes. By contrast, in the retina, Müller glia inherit this barrier function from their closely related radial glial precursors. The OLM is far more anatomically conspicuous than equivalent structures elsewhere in the CNS for one key reason: the photoreceptor inner and outer segments project across the Müller apical barrier into the subretinal space. Thus, instead of lurking at the tissue margin, OLM junctions are situated in a prominent location at the outer edge of the ONL (Fig. 4).

Figure 4: Anatomy of the outer limiting membrane.

A) Representation of the mouse outer retina showing OLM (red arrow) in cross-section view. Schematic drawing (right) is aligned to the photomicrograph (left). OLM separates nuclear layer from inner segments.

B) En face views of the OLM. Top: Brainbow-labeled Müller glia in a retinal wholemount (as in Fig. 3), imaged by confocal microscopy at the OLM level. Small unlabelled circles within Müller arbor territories correspond to photoreceptor inner segments. Middle: En face electron micrograph of adult mouse OLM. Bottom: schematic illustrating the cellular compartments within middle image. Müller cells tile the OLM, meeting at borders containing electron dense homotypic adherens junctions. Heterotypic adherens junctions surround inner segments. Numbers indicate territories belonging to individual Müller cells in the real image (middle) and schematic (bottom). Scale bars, 10 μm (top); 1 μm (middle).

OLM junctions have two crucial functions. First, they serve as a diffusion barrier that creates a distinct extracellular milieu within the subretinal space (Bunt-Milam et al. 1985). Second, the OLM serves essential structural roles. The belt of adherens junctions laced through the center of the retina provides tensile strength to the curved tissue that opposes any tendency to buckling (MacDonald et al. 2015). When OLM junctions are disrupted, the outer retina buckles inwards to form characteristic U-shaped dysplasias known as hemi-rosettes (Clark et al. 2019, de Melo et al. 2016b, Rich et al. 1995, Stuck et al. 2012, Surzenko et al. 2013, West et al. 2008). Individual photoreceptors also receive important structural support, as each cell is completely surrounded by junctions at the base of its inner segment (Fig. 4). The importance of these barrier and structural roles is illustrated by the fact that photoreceptors die when OLM integrity is lost (Cisneros et al. 2020, Mehalow et al. 2003, Ray et al. 2020, West et al. 2008).

Given the importance of the OLM, it is surprising how little is known about its development. During progenitor differentiation, neuroepithelial apical junctions need to be replaced by Müller OLM junctions in a coordinated fashion that preserves the apical retinal boundary. How this coordination might be achieved has rarely been considered. One plausible scenario is that Müller cells inherit their homotypic junctions from progenitors, while making new junctions with photoreceptors as they protrude their nascent inner segments across the apical boundary. This model is consistent with the observation that mouse OLM first becomes visible around the end of the first postnatal week, when photoreceptors are beginning to elaborate their inner/outer segments (Stuck et al. 2012). Formation of the Müller-photoreceptor junctions appears to require a concerted effort from both cell types, because if differentiation of either cell type is perturbed, OLM junction formation is disrupted leading to hemi-rosette dysplasias. The cellular events leading to formation of these heterotypic junctions remain an important area for future study.

II.c. Müller glia and the OLM: Roles in inherited retinal degeneration

Inherited retinal degenerations (IRDs) are a diverse collection of disorders characterized by progressive vision loss due to the death of photoreceptors. A large fraction of IRDs are caused by mutations that affect phototransduction or other basic functions of photoreceptors. However, a subset of IRDs involve genes that impact Müller cell development, particularly at the OLM. Here we focus on IRDs caused by mutations in the CRB1 gene, as this is the best studied example of how OLM development influences photoreceptor survival.

Crumbs-homolog 1 (CRB1) is part of a protein complex that is conserved across animal development, and is critical for forming and/or maintaining apical epithelial junctions in a wide variety of tissues (Thompson et al. 2013). In the retina, the Crumbs complex localizes to apical junctions among neuroepithelial progenitor cells, as well as at OLM junctions. Mouse and zebrafish mutants lacking various Crumbs complex components exhibit severe retinal laminar disorganization phenotypes, reflecting the early loss of progenitor apical junctions (Alves et al. 2013, Cho et al. 2012, Cho et al. 2019, Horne-Badovinac et al. 2001, Koike et al. 2005, Omori & Malicki 2006, Park et al. 2011, Wei & Malicki 2002). In such mutants, the OLM is often unrecognizable. However, milder mutations that allow assessment of the OLM reveal a role for this complex in integrity of OLM junctions as well (Mehalow et al. 2003, Ray et al. 2020, Weatherly et al. 2022).

Mutations in the human CRB1 gene are among the top ten causes of autosomal recessive IRDs, which manifest with an unusually wide spectrum of disease phenotypes (Ehrenberg et al. 2013, Hanany et al. 2020). Most affected individuals fall into two broad phenotypic classes: those with Leber congenital amaurosis (LCA)-type disease, and those with retinitis pigmentosa (RP)-type disease (Daich Varela et al. 2023, Talib et al. 2022, Talib et al. 2017). In CRB1-LCA patients, retinal lamination defects and early disease onset suggest that pathology originates during fetal development. By contrast, CRB1-RP patients have normal retinal layers and a later disease onset, suggesting that pathologies arise postnatally. An appealing model for this phenotypic difference, which is supported by the phenotypes of mice or fish lacking Crumbs complex proteins, is that defects in fetal neuroepithelial junctions could cause the early laminar disorganization that characterizes CRB1-LCA. Defects at OLM junctions, meanwhile, could cause photoreceptor stress leading to progressive degeneration as seen in both versions of the disease.

It has been surprisingly tricky to obtain experimental evidence in animal models for the premise that junction defects cause CRB1 disease. The first issue is a different expression pattern for CRB2 – a close homolog of CRB1 that can compensate for its function – in animal models as compared to humans. Single-cell sequencing data show that human retinal progenitor cells express lower CRB2 levels than mouse progenitors (Clark et al. 2019, Lu et al. 2020), which could account for the fact that Crb1 mutant mice do not exhibit early retinal disorganizations resembling LCA (Mehalow et al. 2003, Ray et al. 2020). By contrast, disorganization of the neonatal retina and other LCA-like phenotypes are observed in mice carrying Crb1 and Crb2 mutations (Cho et al. 2024, Cho et al. 2019, Pellissier et al. 2013). These findings support the idea that LCA-type CRB1 disease originates from defects in apical progenitor junctions.

A second challenge in proving a link between junction defects and retinal disease arises from the lack of consistent OLM or degenerative phenotypes in Crb1 mutant mice. OLM junction phenotypes are strikingly different in the two best-studied Crb1 mutant mouse strains: Animals homozygous for a point mutation called Crb1^Rd8 do have notable OLM junction defects, but a targeted mutation deleting Crb1 exon 1 did not show this effect (Mehalow et al. 2003, van de Pavert et al. 2004). Moreover, neither mutant line yields strong or consistent photoreceptor degeneration, raising questions as to whether OLM defects are sufficient to cause degeneration. In a recent study, our group found that neither the exon1 nor Rd8 alleles completely abrogates Crb1 gene function, because neither allele disrupts all Crb1 mRNA isoforms. This was determined through a long-read sequencing study that comprehensively identified all CRB1 mRNA isoforms expressed in mouse and human retina (Ray et al. 2020). Leveraging the isoform map, we used CRISPR to generate a true null allele that eliminates all Crb1 isoforms. In contrast to the other two alleles, these Crb1^null mice reliably exhibit an RP-like degenerative phenotype by ~3 months of age (Ray et al. 2020). Furthermore, degeneration is preceded by OLM junctional defects that are more extensive than those in Rd8 mutants, suggesting that the severity of degeneration is correlated with the severity of earlier OLM damage. These data support the idea that OLM defects do indeed contribute to the pathobiology of the human disease, but further work in the Crb1^null animal model, as well as in human patients, will be required to rigorously test this hypothesis.

Our studies of CRB1 isoforms revealed additional complexity that may shed light on how the two OLM cell types regulate junction formation. In addition to the canonical CRB1 isoform – denoted CRB1-A – we also detected an additional highly expressed isoform, which we named CRB1-B. Within both mouse and human retina, CRB1-B is by far the most abundant of all CRB1 isoforms, with CRB1-A a distant second (Ray et al. 2020). Remarkably, these two most prominent isoforms are expressed in different cell types: CRB1-A is expressed by embryonic progenitor cells and Müller glia, while CRB1-B is expressed by rods and cones. Despite their expression in different cell types, CRB1-A and CRB1-B must both be deleted to generate OLM defects and photoreceptor degeneration (Ray et al. 2020, van de Pavert et al. 2004). This finding implies that the presence of CRB1 protein on one side of the OLM junction is sufficient to induce junctions on the opposite side, even if the opposite cell type is lacking CRB1.

Altogether, the recent work on CRB1 emphasizes that interactions between developing photoreceptors and Müller glia at the OLM can have a long-lasting impact on retinal health. Other IRDs may also involve similar pathobiological mechanisms. For example, two proteins encoded by Usher syndrome disease genes, Harmonin and Clarin-1, localize to the OLM and/or Müller apical microvilli, where they may mediate glial-photoreceptor interactions (Cowan et al. 2020, Li et al. 2023, Nagel-Wolfrum et al. 2022, Nonarath et al. 2024, Xu et al. 2020). Additionally, OLM defects have been documented in rodent models of X-linked retinoschisis (Ye et al. 2022, Zeng et al. 2004). Gaining a better understanding of the mechanisms underlying OLM development will be important for devising new therapeutic strategies targeting these IRDs.

III. Development of RNFL astrocytes

While retinal neurons and Müller glia are derived from progenitor cells located throughout the neuroepithelium, astrocytes are derived from a specialized population of retinal progenitor cells located at the junction of the neural retina and the optic stalk (Fig. 2B) (Dakubo et al. 2003, Morcillo et al. 2006, Watanabe & Raff 1988). These progenitors give rise exclusively to astrocyte precursor cells (APCs) – a population of immature glial cells that retain proliferative capacity but are fully committed to the astrocyte fate (Chan-Ling et al. 2009, Chu et al. 2001). From the site of their birth, APCs disperse peripherally to colonize the RNFL (Fig. 2A,B) (Ling et al. 1989, Ling & Stone 1988, Stone & Dreher 1987, Watanabe & Raff 1988). This migration establishes the astrocytic honeycomb-like network that will serve as the template for subsequent angiogenesis (Fig. 2C). Arrival of vessels triggers the differentiation of APCs into mature astrocytes, as well as a remodeling of the astrocyte network involving morphological changes and naturally-occurring cell death. In this section we will consider the mechanisms underlying each of these developmental events that ultimately establish the mature RNFL astrocyte population.

III.a. Astrocyte birth and migration into the retinal nerve fiber layer

When the impressive astrocytic migration into the retina (Fig. 2A) was first discovered in the 1980s, it was thought that astrocytes were immigrants from the optic nerve (Ling & Stone 1988, Stone & Dreher 1987, Watanabe & Raff 1988). While it may still be possible that some RNFL astrocytes derive from the nerve itself, we now know that there is a specialized progenitor zone within the neural retina, surrounding the optic disc, that gives rise solely to astrocytes (Chu et al. 2001, Dakubo et al. 2003, Morcillo et al. 2006). Such progenitors are probably the main or even the exclusive astrocyte source. Patterning and specification of this progenitor zone has been reviewed elsewhere (Paisley & Kay 2021). This intra-retinal progenitor population expresses key astrocyte/APC markers such as Pax2, Pdgfra and Megf10, suggesting that they are lineage-committed to the astrocyte fate (Chan-Ling et al. 2009, Chu et al. 2001, Dakubo et al. 2003, Paisley & Kay 2021, Soukkarieh et al. 2007).

Once they have exited the optic disc progenitor zone, APCs undertake a long-range migration to colonize the RNFL. Migration occurs during the late embryonic and early postnatal period in mice (Fig. 2), and during the second trimester in humans, with APCs reaching the retinal boundary by 28 gestational weeks (Chu et al. 2001). Migrating APCs adopt a highly polarized morphology, with their somata and nuclei aligned along the center-to-peripheral axis (Chu et al. 2001, Fruttiger 2002, O’Sullivan et al. 2017). Transit to the periphery requires intact ILM extracellular matrix as a permissive migratory substrate (for review see Paisley and Kay, 2021) and is guided directionally by RGC axons: In both mice and humans, migrating APCs are closely associated with axons and are polarized in alignment with them (O’Sullivan et al. 2017, Provis et al. 1997). When axons are genetically removed, APCs polarize randomly, as if they lacked directional information, and they fail to reach the retinal periphery (Derbyshire et al. 2023, O’Sullivan et al. 2017). Chains of migrating astrocytes are often observed along axons, suggesting that there could be some mechanistic similarities to other systems that use collective migration (Helmbacher 2022, Mayor & Etienne-Manneville 2016). Guidance by axons likely explains the absence of astrocytes from the primate fovea, as fetal RGC fibers course around the region where the fovea will ultimately form without entering it (Bringmann et al. 2018).

III.b. Astrocyte guidance of retinal angiogenesis

Endothelial cells first enter the retina at P0 in mice and around gestational week 14-15 in humans; at this point, APC migration is well underway but not yet completed (Fig. 2A) (Provis 2001). The wave of sprouting angiogenesis then sweeps through the RNFL, guided by the astrocyte honeycomb template (Fig. 2A,C). Primary RNFL angiogenesis is complete by P8 in mouse; but in humans, wavefront progression slows markedly as it reaches the far periphery, such that vessels do not fill the RNFL until shortly before birth (Provis 2001). As the primary plexus is nearing completion, angiogenic sprouts dive to form the deep vascular plexus. Here we focus on establishment of the primary plexus because this is the aspect of retinal angiogenesis controlled by RNFL astrocytes. However, it is important to note that Müller glia critically influence deep plexus formation, through the expression of growth factors such as VEGF-A and Norrin (Rattner et al. 2019, Selvam et al. 2018).

III.b.1. Astrocyte molecular cues regulating angiogenesis

Astrocytes guide angiogenesis in two ways: they provide both pro-angiogenic molecular cues and a physical template to guide vascular growth. Among pro-angiogenic growth factors, the most important is VEGF-A. As noted in Section I above, APCs are the key source of VEGF-A during primary angiogenesis, without which vessels cannot enter the retina. APCs express VEGF-A under control of the hypoxia-induced factor (HIF) signaling pathway – the major mechanism by which cells initiate transcriptional responses to alleviate hypoxia (Kaelin & Ratcliffe 2008). During primary angiogenesis, retinal tissue ahead of the advancing vascular wavefront is hypoxic due to the lack of nearby vessels. Hypoxia induces HIF signaling within this avascular zone, which leads APCs to express VEGF-A via the HIF2α transcriptional effector (Duan et al. 2014, Perelli et al. 2021, Stone et al. 1995, West et al. 2005). Other retinal cells within the avascular zone are also hypoxic, but they do not express VEGF-A – most likely because they express a different HIF effector, HIF1α (Nakamura-Ishizu et al. 2012, Perelli et al. 2021). Accordingly, deletion of HIF2α from astrocytes abolishes VEGF-A expression and prevents vasculature from entering the retina (Duan et al. 2014, Perelli et al. 2021).

As the vascular wavefront advances, APCs ahead of the wavefront maintain high levels of VEGF-A expression, whereas maturing astrocytes behind the wavefront gain access to circulation and downregulate VEGF-A (Fig. 2C). This creates a gradient of VEGF-A expression which is detected by the receptor VEGFR-2 on tip cells. Binding to VEGFR-2 has two main effects: it mediates tip cell guidance to steer them in the proper direction; and it stimulates vessel branching by inducing formation of new tip cells and new filopodia (Chappell et al. 2019, Gerhardt et al. 2003). Angiogenesis can be acutely blocked by intravitreal injection of soluble VEGF-A binding proteins, which sequester the protein and prevent it from accessing VEGFR-2. This is the basis for a key class of anti-angiogenic ophthalmic drugs in wide use today in the clinic.

Despite the potent effects of VEGF-A on growing RNFL tip cells, and its key role recruiting vessels into the retina in the first place, it does not appear to be absolutely required for advance of the vascular wavefront once angiogenesis is underway. A variety of approaches have been used to selectively diminish VEGF-A function during wavefront progression; while these can delay RNFL angiogenesis, none prevents its completion (Gerhardt et al. 2003, Perelli et al. 2021, Stalmans et al. 2002). Thus, other APC-derived cues besides VEGF-A may also be important for wavefront progression. Angiopoietin (Angpt)-1 has been suggested as such a cue, and indeed genetic studies suggest it is pro-angiogeneic in the developing retina (Lee et al. 2013). However, recent single-cell sequencing data (Li et al. 2023) suggest Angpt1 is provided by pericytes at or behind the angiogenic wavefront, not by astrocytes ahead of it. Angpt4, by contrast is expressed by APCs ahead of the wavefront in a HIF-dependent manner (Elamaa et al. 2018). Another potential candidate is adrenomedullin, a known pro-angiogenic cue that is expressed by astrocytes and tip cells (Li et al. 2023, Strasser et al. 2010). There is some genetic evidence to suggest it has a role in RNFL angiogenesis (Iesato et al. 2013, Mackie et al. 2019). Overall, however, more work is needed to understand the role of non-VEGF factors in driving the angiogenic wavefront.

III.b.2. Patterning of astrocytes into a physical template for angiogenesis

The second way in which astrocytes guide primary angiogenesis is through the honeycomb-like pattern of the APC cellular network (Fig. 2C). Genetic studies in mice have firmly established that the honeycomb network dictates patterning of developing vasculature. When the local density of APCs is experimentally perturbed, the spatial pattern of the honeycomb network is also altered, becoming either more or less dense depending on the change in APC numbers. In these studies, the density and patterning of capillaries changed to match the alterations in the honeycomb network, leading to severe vascular defects (Fruttiger et al. 1996, Morita et al. 2017, O’Sullivan et al. 2017, Perelli et al. 2021). Chemogenetic ablation of astrocytes during wavefront progression also causes striking vascular changes: While advance of the wavefront continues in the absence of the honeycomb network, capillary patterning is lost entirely as vessels grow straight towards the periphery (O’Sullivan et al. 2017).

The effects of astrocyte ablation on vessel patterning (O’Sullivan et al. 2017) is far stronger than what is observed following equivalently timed VEGF-A manipulations, indicating that there are non-VEGF molecular cues situated within the honeycomb template that guide tip cells and pattern capillaries. Several studies have reported potential astrocytic guidance molecules, such as Fibronectin-1 (Fn1), an ECM protein expressed selectively by APCs ahead of the wavefront; and R-cadherin (also known as Cdh4), a homophilic adhesion molecule (Dorrell et al. 2002, Stenzel et al. 2011). However, astrocyte-specific Fn1 knockout mice yielded a vascular phenotype that was far milder than astrocyte-ablated mice; and a role for R-cadherin has never been established using Cdh4 knockout animals. Thus, the identity of the tip cell guidance molecules expressed by the APC template remains a key open question meriting future study.

Because the pattern of the APC template is transferred with such high fidelity to the vasculature, the developmental mechanisms that establish APC patterning are clearly of vital importance for vessel development. Several of the key APC patterning mechanisms involve developmental control over APC cell numbers. As noted earlier in this section, the arrangement and density of the honeycomb network is highly sensitive to both increases and decreases in APC density. Thus, APC density must be under tight developmental control to ensure proper vessel development. Here we consider three developmental mechanisms – migration, proliferation, and naturally-occurring cell death – that impact astrocyte patterning by regulating cell numbers.

Migration.

As we have already seen, a long-range migration is required to deliver APCs to peripheral retina (Fig. 2A). How individual APCs decide where to stop migrating remains unknown, but this process is undoubtedly under regulatory control to ensure appropriate local APC density. The influence of these migratory control mechanisms over APC patterning is illustrated by mouse mutants in which migration is impaired. Examples include mutants affecting the RGC axons (Derbyshire et al. 2023, O’Sullivan et al. 2017), as well as several mutants affecting ILM integrity (Clements et al. 2017, Gnanaguru et al. 2013, Tao & Zhang 2016). In all of these mutants, migratory defects cause astrocyte density to become abnormally high in central retina and abnormally low in the periphery, leading to striking alterations in the patterning of the honeycomb template with severe consequences for vessel patterning.

Cell proliferation.

Competence to divide is one of the defining features of APCs (Chan-Ling et al. 2009, Provis 2001), and proliferation has a major impact on APC patterning. Proliferation is influenced by at least three cell-extrinsic factors that each contribute to ensuring that the number of APCs is appropriate to build the honeycomb template and support angiogenesis. First, there appears to be a homotypic recognition mechanism that induces proliferation when APC density becomes too low. Our laboratory has observed that if astrocyte density is experimentally diminished, remaining astrocytes can proliferate to replace them (Perelli et al. 2021). Little is known about how this mechanism works.

A second cell-extrinsic signal promoting APC proliferation is PDGF-A, derived from RGCs (Fruttiger et al. 1996). PDGF-A is a potent APC mitogen that acts by binding to PDGFRα receptors on astrocytes. Deletion of PDGFRα using an astrocyte-specific Cre driver causes a complete absence of APCs, suggesting that PDGFRα-mediated proliferation and/or survival is essential to the establishment of the APC population (Tao & Zhang 2016).

Third, APC proliferation is strongly regulated by tissue hypoxia. This is evident from studies in which neonatal mice are reared in high or low oxygen environments: excess oxygen suppresses proliferation, while a hypoxic environment causes APC overproduction and vascular defects (Morita et al. 2016, Perelli et al. 2021, Stone et al. 1995, West et al. 2005). Hypoxia drives astrocyte proliferation via the HIF pathway, in two different ways. First, there is a cell-autonomous mechanism involving astrocytic hypoxia sensing via HIF2α (Duan & Fong 2019, Duan et al. 2014, Perelli et al. 2021). Second, there is a more circuitous mechanism in which hypoxia activates HIF1α within retinal neurons, thereby stimulating production of the APC mitogen PDGF-A (Nakamura-Ishizu et al. 2012). The need for hypoxia-induced proliferation likely relates to the key role of APCs as the cellular source of VEGF-A. Given this fact, alleviating hypoxia requires not only transcriptional regulation of VEGF-A, but also a cellular mechanism to ensure that the VEGF-A expressing cells – i.e. APCs – are present in sufficient numbers to effectively promote angiogenesis.

Developmental cell death.

Naturally-occurring cell death is a normal part of nervous system development (Arya & White 2015). Mouse RNFL astrocytes are subject to a particularly striking developmental decline in cell numbers: between the peak at P5 and the end of the second postnatal week, when adult numbers are reached, astrocyte numbers decrease by over three-fold (Puñal et al. 2019). As a patterning mechanism, death may be important as a homeostatic counterbalance to the effects of migration and proliferation: Downward adjustments to APC density would serve to ensure proper cell numbers/patterning, and also to prevent excessive VEGF-A levels (Gerhardt et al. 2003, Tata et al. 2015).

An unusual feature of RNFL astrocyte death is that, unlike the vast majority of developmental cell death, apoptosis is not the death mechanism. Instead, death is mediated by microglia, most likely through phagocytic removal of live astrocytes (Puñal et al. 2019). While this phenomenon has so far only been demonstrated in mice, there are hints that it could also occur in primates: RNFL astrocytes withdraw from the perifoveal area during macaque retinal development, in a manner that does not involve apoptosis (Distler et al. 2000). The molecular mechanism by which microglia kill astrocytes remains an important topic for future study. Many candidate phagocytic receptors have been knocked out from microglia, but effects on astrocyte survival have been mild, suggesting there is substantial molecular redundancy in this system (Gnanaguru et al. 2023, Puñal et al. 2019). There is also cellular redundancy, as astrocytes can take on the task of killing and engulfing each other when microglia are absent (Puñal et al. 2019). The extent of redundancy in this system emphasizes the importance of the astrocyte death mechanism – but also makes determination of the molecular mechanism more challenging.

III.c. Maturation of APCs into astrocytes

The role of astrocytes in vascular development changes instantaneously as the vessel wavefront passes over them. Before vessel arrival, the APC template needs to express growth and guidance cues that recruit angiogenic sprouts; but after arrival, the presence of such cues would disturb tip cell guidance and would prevent differentiation of newly formed vessels (Tata et al. 2015). Thus, arrival of vessels induces APCs to undergo a rapid and dramatic phenotypic change, erasing pro-angiogenic aspects of the APC phenotype and promoting expression of mature astrocyte genes. Arrival of vessels also induces striking anatomical changes to the astrocyte network, as the honeycomb meshwork is disassembled and reorganized into an adult pattern. Whereas APC somata are arranged in circles around the edge of the honeycomb loops, mature astrocytes have a very different arrangement: A subset of astrocytes remains closely associated with major vessels (Stone & Dreher 1987) while the remainder adopt a uniform mosaic pattern in which individual cells exhibit local cell-cell avoidance (Chan-Ling & Stone 1991, O’Sullivan et al. 2017). This mature pattern is thought to ensure complete coverage of the RNFL axon population by their astrocytic supporting cells.

Experiments in which vascular development was blocked show that vessels themselves are the source of molecular cues that drive the APC-to-astrocyte transition: Absence of vessels causes sustained high expression of VEGF-A, a marker of the APC state (Fig. 2C), while also inhibiting cell cycle exit and GFAP expression – two markers of mature astrocytes (Duan et al. 2017, Morita et al. 2017, West et al. 2005). Single-cell sequencing ongoing in our laboratory suggests that these three markers are just the tip of the iceberg: Vessel arrival induces massive gene expression changes involving hundreds of genes (A.M.M. and J.N.K., unpublished observations). How do vessels rapidly induce such a dramatic phenotypic switch? One important emerging principle is that there may be distinct mechanisms driving different aspects of the APC-to-astrocyte transition. To become a mature astrocyte, APCs must execute three distinct genetic programs: 1) suppression of APC genes; 2) upregulation of mature astrocyte genes; and 3) cell cycle exit. While each of these programs are triggered by vasculature-derived signals, there are important differences. One such difference is evident from the timecourse of program activation. In mice, suppression of the APC state and activation of the mature state seem to occur almost instantaneously upon vessel arrival, given that the vessel wavefront delineates a sharp boundary between astrocytes highly expressing Vegfa and those highly expressing Gfap (Fig. 2C) (West et al. 2005). By contrast, cell cycle exit occurs in a far more delayed manner, such that dividing cells can still be detected behind the wavefront for several days after it passes (Perelli et al. 2021). These distinct kinetics indicate that the mechanism driving cell cycle exit is quite different, although it remains unknown.

While suppression of APC genes and activation of mature astrocyte genes share fast kinetics, there are also important mechanistic differences in how vessels drive these two programs. Suppression of the APC state is controlled by vessel-derived molecular oxygen, acting through the HIF pathway. Ahead of the wavefront, hypoxia maintains APC gene expression via HIF2α, which induces APC marker genes such as Vegfa, Angptl4, and Nr2e1 (Elamaa et al. 2018, Uemura et al. 2006). In turn, the transcription factor encoded by Nr2e1 – known as Tlx – forms a heteromeric complex with HIF2α to regulate many aspects of APC function and identity (Duan et al. 2023, Duan et al. 2014, Uemura et al. 2006). Upon arrival of vessels and relief of hypoxia, HIF signaling is terminated leading to suppression of this APC transcriptional program.

Mature astrocyte gene activation, by contrast, is unlikely to use this same oxygen-dependent molecular mechanism. Because HIF is a transcriptional activator, termination of HIF signaling could only lead to upregulation of mature genes by acting indirectly through additional “sign-reversing” transcription factors, which have yet to be identified. Such factors may exist, as GFAP expression becomes upregulated when HIF2α-Tlx signaling is perturbed (Duan et al. 2023). Thus, it is still possible that vessel-derived oxygen contributes to induction of the mature astrocyte state. However, evidence to date suggests a more important role for vessel-derived growth factors such as Apelin and leukemia inhibitory factor (LIF). Both Apelin and LIF are sufficient to induce astrocytic GFAP expression; and in knockout mice that eliminate LIF or Apelin signaling, GFAP expression is blocked (Kubota et al. 2008, Mi et al. 2001, Sakimoto et al. 2012). Therefore, the Apelin-LIF system is ideally placed to trigger the mature astrocyte transcriptional program at the vascular wavefront.

A key priority for future work will be to understand the mechanisms underlying the anatomical rearrangement of the astrocyte network. Disassembly of the honeycomb template coincides in time with the massive decline in astrocyte cell numbers, suggesting that microglia-mediated pruning and death may be involved (Puñal et al. 2019), but this aspect of astrocyte maturation merits further study.

III.d. Retinal astrocytes and retinopathy of prematurity

Because of their central role in the onset and patterning of angiogenesis, the basic mechanisms of retinal astrocyte development are critical to establishing and maintaining the retina in a healthy state. As such, defects or disruptions to astrocyte development have consequences that manifest as retinal vascular diseases. Retinopathy of Prematurity (ROP) is a prime example. This disorder is characterized by irregular retinal vascular development in preterm infants. The primary risk factor for ROP is the administration of supplemental oxygen, a critical intervention for infant survival but one that can disrupt normal retinal vascular development. In the presence of elevated environmental oxygen, vessel growth ceases, and does not re-initiate properly upon return to ambient oxygen levels. Haphazard vessel growth leads to pathologies such as neovascularization, inflammation, and potential retinal detachment (Hellström et al. 2013).

Considering retinal astrocytes directly sense oxygen levels and play a critical role in guiding angiogenesis, they are undoubtedly involved in ROP. To better define their role, we developed a neonatal oxygen-induced retinopathy (NOIR) mouse model, in which mice are exposed to high oxygen during astrocyte development – from P0 to P4. Upon returning to room air, these NOIR mice develop astrocyte template patterning defects; these defects were found to result from excessive astrocyte proliferation induced by the HIF pathway, which becomes activated by the relative decrease in oxygen levels (Perelli et al. 2021). In this NOIR model, astrocyte template defects lead to several human ROP pathological hallmarks that are absent from traditional OIR models (Smith et al. 1994), such as delayed angiogenesis; persistent hyaloid vasculature; and altered vessel patterning with hemorrhage. These findings suggest that astrocytes do indeed have a role in ROP pathobiology. Going forward, the NOIR model may be a useful complement to traditional OIR models in efforts to develop new ROP therapeutic strategies.