RGCs are a heterogeneous population of neurons that play a crucial role by transmitting visual information from early processing in the retina to the brain (Levin, 2005; Kerschensteiner and Feller, 2024) (Figure 1). They are responsible for encoding various aspects of visual stimuli, such as color, shape, and motion (Yan et al., 2020). Certain subtypes of RGCs also have indirect visual functions, such as regulating the size of the pupil in response to changes in light levels. Unlike photoreceptors, which are comprised of only four major human subtypes, with often only a single synaptic connection, RGC diversity is much more complex.

Primates and non-primates alike possess many RGC subtypes that differ from each other in multiple ways, including: molecularly and biochemically (e.g., having a diverse gene expression profile) (Kang et al., 2017; Mead et al., 2014; Polyak and Willmer, 1949; Yan et al., 2020), morphologically (i.e., cell soma size and/or stratification) (Dacey, 2004; Poljak, 1935), and physiologically (e.g., some respond to movement while others respond to light, or to regulate circadian rhythms) (Gouras, 1968; Graham et al., 1978; Johnson and Samuels, 1997; Kuffler, 1953; Pokorny, 2011; Stiles, 1959). These functionally and topographically unique subtypes of RGCs receive afferent input from complex inner retinal circuits that can include dozens of presynaptic bipolar and amacrine cells. All RGCs then extend lengthy axons through the optic nerve and into one of several visual centres in the brain. By combining these features, RGCs have been classified into multiple distinct subtypes that can respond differently to injury or disease (Figure 2). Recent single-cell RNA sequencing (scRNA-seq) breakthroughs have revolutionized subtype mapping, revealing gene expression profiles that delineate functional classes. For instance, in mice, comprehensive scRNA-seq atlases have identified over 40 subtypes, characterized by markers like SPP1/KCNG4 for alpha RGCs (motion-sensitive) and OPN4 for ipRGCs (regulating circadian rhythms), enabling spatial and developmental insights through integration with spatial transcriptomics (Budoff and Poleg-Polsky, 2025; Li et al., 2024). In primates (including humans), scRNA-seq has uncovered 18–25 subtypes. Cross-species comparisons have highlighted evolutionary divergences, such as enriched midget RGCs for foveal high-acuity color vision in primates versus broader motion-detection subtypes in nocturnal mice (Yi et al., 2021; Lu et al., 2020). These transcriptomic atlases, including multi-species datasets and aging-focused profiles, underscore species-specific adaptations. Mouse studies leverage advanced genetic tools for finer classification, while human/primate studies face tissue access limitations but benefit from emerging databases like scRetinaDB, aggregating over 2.79 million cells across species for subtype vulnerability analysis in diseases (Tang et al., 2025).

Adding additional complexity, this diversity of subtypes is species-specific. Distinct RGC subtypes have been identified in humans in comparison to mice. Curiously, the mouse has many more identified subtypes when compared to humans (more than 40 subtypes in the mouse eye, compared to about 18 in the human eye) (Corral-Domenge et al., 2022; Rheaume et al., 2018; Sanes and Masland, 2015; Tran et al., 2019; Yamada et al., 2005). The higher number of RGC subtypes in mice compared to primates likely stems from evolutionary adaptations and differences in research methodologies. From an evolutionary perspective, mice are nocturnal animals that lack a specialized fovea. They may require a broader array of specialized cell subtypes to perform complex computational visual processing, such as motion and looming detection, across a uniform retina. These adaptations facilitate rapid, innate survival behaviors. In contrast, primates prioritize fewer specialized types for high-acuity trichromatic vision, with a retina dominated by midget cells that act as “pixel encoders.” These cells transmit high-resolution, relatively raw data to the brain, shifting the burden of complex visual processing from the retina to the visual cortex. Furthermore, the current disparity between species is reinforced by methodological asymmetries. The availability of sophisticated transgenic mouse lines and high-throughput transcriptomic tools allows for the identification of extremely rare murine RGC subtypes that comprise less than 1% of the population. In the primate retinas, the overwhelming numerical dominance of midget and parasol cells, combined with more limited access to similar genetic tools, makes the characterization of rare subtypes significantly more challenging, potentially undercounting the true diversity (Kim et al., 2021; Hahn et al., 2023; Huang et al., 2024).

Considering this cellular and morphological complexity, the preservation or replacement of damaged RGCs due to injury or degenerative disease remains a major challenge. Yet, irreversible RGC degeneration drives vision loss in glaucoma; a leading cause of blindness worldwide (Alqawlaq et al., 2019; Weinreb et al., 2016; Almasieh et al., 2012). Addressing this important issue requires understanding how specific subtypes function and respond to stress and injury, followed by methods for the therapeutic manipulation of cellular pathways involving neuronal survival, migration, dendritogenesis and axogenesis, pathfinding, synaptogenesis, and myelination. These data will also facilitate the development of protocols to enrich and/or differentiate for specific subtypes, particularly those of the human retina. Recent research has been pioneering use of in vitro models to grow human retinal neurons, including RGCs, from human embryonic stem cells (ESCs) and induced pluripotent stem cells (iPSCs). However, the isolation of and study of RGC subtypes from these models is still unclear. To date, labs have tended to use either pan-RGC markers (expressed by all subtypes) or mouse-specific markers that may not have the same specificity in human cells. Yet, there is increasing published information concerning human RGC subtype identification and roles, and their responses to injury and disease. To comprehensively address this point, this review will: 1) explore the different subtypes of human RGCs, 2) profile relevant human molecular markers available for identifying and isolating each subtype, 3) evaluate evidence for subtype vulnerability in injury and disease, and 4) assess current strategies for subtype differentiation in human PSC (hPSC) cultures, leveraging transcriptomic insights for therapeutic translation.

Unlike in other species, the identification of RGC subtypes in humans is faced with many hurdles. Acquiring eyes donated from different age groups and healthy or sick individuals is challenging, as most samples are collected postmortem. Using these tissues, researchers have tried to understand the development of human RGCs and their specification in the retina as well as their connections to the brain’s visual cortex. Sectioning and staining these human samples to identify subtype-specific molecular markers is still a technical challenge. In addition, testing the physiological response in living or postmortem eyes is extremely difficult and often unreliable. One alternative approach has been to use non-human primates as an RGC model to study physiological responses of different subtypes. Clinically, optical coherence tomography (OCT) imaging has provided additional tools to define changes in RGC and nerve fiber architecture, and the pathological progression of inherited and acquired optic neuropathies (Kim et al., 2021).

There are about 18 peripheral and 16 central RGC subtypes in primates (Dacey, 2004; Masri et al., 2019; Peng et al., 2019; Rheaume et al., 2018; Yan et al., 2020). Of these, the three major RGC subtypes are: ON and OFF Midget RGCs, ON and OFF parasol RGCs and small bistratified RGCs (Grünert et al., 2021; Lee et al., 2010; Merigan and Maunsell, 1993). The remaining subtypes include recursive bistratified, recursive monostratified, broad, and narrow thorny, smooth monostratified, large sparse, and giant sparse melanopsin ganglion cells (Ghosh et al., 1996; Grünert et al., 2021; Hughes et al., 2016; Kim et al., 2021; Mao et al., 2014). Interestingly, in primates the distribution of RGC subtypes appears to be spatially organized. For example, of the total RGCs in the human fovea, midget cells constitute about 90%, parasol cells about 5%, and small bistratified cells about 1%. In the periphery, midget cells make up about 40%–45%, parasol cells about 15% and small bistratified cells about 10% of the total. (Dacey, 1994; Field et al., 2007; Masri et al., 2019; Peng et al., 2019; Vogelsang et al., 2025; Zhang et al., 2024). Each of these subtypes has received varying degrees of research and so range from well-studied to virtually unknown. Known classification and roles for various primate RGC subtypes are described in Table 1 and in the section below.

According to many published studies, RGCs are particularly vulnerable neurons in the retina, and their degeneration is the immediate cause of vision loss in a variety of retinal diseases, such as glaucoma and ischemic optic neuropathies (Alqawlaq et al., 2019; Soucy et al., 2023). Yet, there is controversy in the field as to whether certain RGC subtypes are more or less vulnerable to injury (Santina and Ou, 2017) (Figure 3). Questions have been raised as to whether soma size (small vs. large), stratification (ON vs. OFF), firing (sustained vs. transient) and location (central vs. peripheral) affect the susceptibility of RGCs to injury. This applies to glaucoma and a various optic nerve injury model, including exposure to elevated intraocular pressure (IOP), ONC or transection, and ischemic or excitotoxic stresses. Answering these questions will help researchers overcome current challenges in engineering cells with increased resilience and protecting the more vulnerable subtypes. This strategy can be applied to future efforts to preserve vision and prevent blindness.

In the human eye, lost or damaged RGCs cannot be replaced endogenously, as the stem cells that build the retina during development enter a state of quiescence after birth (Tropepe et al., 2000). Several methods have been proposed to generate new RGCs, including the use of autologous adult retinal stem cells (Tropepe et al., 2000; Coles et al., 2004; Cho et al., 2012). In addition, multiple studies have attempted to differentiate RGCs from other extraretinal stem cell populations, with varying degrees of success. These efforts include adult mesenchymal stem cells (MSCs) (Eraslan et al., 2023; Dodina et al., 2024), autologous adipose-derived stem cells (Rezanejad et al., 2014), autologous bone marrow derived stem cells and Müller glia (MG) cells (Cen and Ng, 2018; Silva-Junior et al., 2021; Soucy et al., 2023). However, these studies frequently utilize rodent models and highlight paracrine-mediated neuroprotection rather than the robust generation of functional human RGC subtypes. Although MSCs and MG (Silva-Junior et al., 2021; Soucy et al., 2023) offer potential therapeutic avenues, their capacity to recapitulate the complex diversity of the human RGC landscape remains limited. Consequently, human PSCs including ESCs and iPSCs, have emerged as the gold standard for modeling human RGC subtype specification and functional maturation (Kosior-Jarecka and Grzybowski, 2024; Zhang et al., 2021) (Figure 4). Pluripotent stem cells give rise to a broader range of cell types compared to adult stem cells, making them preferred source for many scientists.

Many previous studies focused on growing RGC subtypes in vitro have relied upon animal models, which have enabled the identification of these cells and the study their functional characteristics (Dhande et al., 2015; Marcucci et al., 2016; Sanes and Masland, 2015). However, comparably few studies have been performed in humans due to the limited availability of adult tissue and the inaccessibility of the human retina at early developmental stages.

hPSCs provide a powerful tool for developmental studies because they can self-renew and give rise to all cell types of the body (Takahashi et al., 2007; Trimarchi et al., 2007; 2008; Yu et al., 2007). Many previous efforts have examined the ability of hPSCs to give rise to RGCs (Fligor et al., 2018; Gill et al., 2016; Lei et al., 2024; Luo and Chang, 2024; Ohlemacher et al., 2016; Riazifar et al., 2014; Sluch et al., 2015; Tanaka et al., 2016; Teotia et al., 2017). However, this differentiation has primarily focused on the generation of general RGCs without addressing the numerous subtypes that exist. RGC differentiation strategies from hPSCs generally fall into three primary categories: small molecule-directed 2D differentiation, 3D retinal organoid differentiation, and direct transcription factor-driven differentiation.

Small molecule-directed 2D RGC differentiation protocols commonly work by directing hPSCs toward a neuroectodermal fate through the dual inhibition of BMP and TGF-β signaling (e.g., using drugs like dorsomorphin or SB431542). This is followed by retinal progenitor specification using FGF2 and IGF1 to promote the formation of optic vesicle-like structures. Subsequent RGC commitment is achieved through Notch inhibition and the expression of proneural factors such as ATOH7 (the human ortholog of Math5), which drives cell cycle exit. Maturation is further guided by POU-domain transcription factors, particularly BRN3B and ISL1, which are essential for axonal outgrowth and terminal differentiation (Gill et al., 2014; Ji and Tang, 2019; Sluch et al., 2015; Lee et al., 2018; Luo et al., 2022).

3D retinal organoid differentiation leverages the intrinsic self-organizing capacity of hPSCs in suspension culture to form laminated, optic-vesicle-like structures. Within this 3D architecture, RGCs emerge spontaneously, offering superior biological complexity and mimicry of in vivo retinogenesis and subtype diversity. Modern protocols can achieve a stratified retinal structure over several months, providing a more physiologically relevant model than 2D system (Fligor et al., 2018).

Finally, the direct transcription factor-driven differentiation approach involves the exogenous, forced expression of key proneural and POU-domain transcription factors (e.g., BRN3B, or ISL1) to bypass traditional developmental stages. This method rapidly convert hPSCs or somatic cells directly into RGC-like neurons, often within one to 2 weeks, which is significantly faster than the organoid-based methods (Agarwal et al., 2023; Costea et al., 2025; Wang J. et al., 2020).

Each system offers distinct advantages for differentiating hPSCs into RGCs. Cultures in 2D provide simplicity, scalability, and precise control over signaling cues, facilitating higher RGC yields of RGC progenitors and easier downstream analyses like immunopurification. However, they often lack the physiological architecture and mature synaptic connectivity found in vivo, resulting in reduced cell-cell interactions and limited mimicry of retinal lamination (Oswald and Petr, 2018). In contrast, 3D retinal organoids better recapitulate in vivo retinogenesis, promoting self-organization into laminated structures with improved RGC maturation, synaptic formation, and subtype diversity but faces challenges such as high variability, reproducibility, long differentiation timelines, axon growth, and nutrient diffusion limits in the absence of vascularization (Wang et al., 2025; Harkin et al., 2024). Markers expression also varies widely between the two systems. In 2D cultures, small molecule-directed differentiation enables rapid induction of markers like BRN3B within 4–6 weeks (Gudiseva et al., 2021) and high THY1 expression for efficient isolation, though RBPMS is often lower or delayed (Chavali et al., 2020). In contrast, 3D organoids more closely mimic in vivo retinogenesis with sequential marker emergence ATOH7 in progenitors, followed by BRN3B/ISL1 and later RBPMS/neurofilaments. Yet THY1 remains low unless enriched via dissociation and 2D replating (Aparicio et al., 2017; Rabesandratana et al., 2020). Hybrid approaches, such as transitioning from 3D to 2D for RGC enrichment, may optimize outcomes for research and therapeutic applications (Li et al., 2021; Luo et al., 2022).

However, while pluripotent stem cell-derived RGCs offer promising systems for disease modeling and regenerative therapies, several important limitations persist. These limitations include incomplete maturation, where derived RGCs often exhibit immature phenotypes with limited axonal outgrowth. Also, incomplete synaptic connectivity, and poor electrophysiological responses have been reported compared to native cells (Singh and Nasonkin, 2020; Kurzawa-Akanbi et al., 2024). Heterogeneity in differentiation efficiency leads to mixed populations, complicating subtype-specific studies, and the absence of vascular and immune components in vitro restricts modeling of complex pathologies (Kurzawa-Akanbi et al., 2024). Post-transplantation challenges, such as poor integration, immune rejection, and limited long-term survival (e.g., up to 4 months in some models), further hinder clinical translation (Lei et al., 2024; Yang et al., 2021).

To move beyond heterogeneous pan-RGC populations, recent strategies have focused on manipulating the transcriptional landscape. Beyond the core ATOH7/BRN3B/ISL1 axis that drives general RGC commitment, researchers are now targeting specific factors such as SoxC (Sox4, 11, 12) and GDF signaling to direct subtype-specific fates or to bias cells toward specific projection identities (Kuwajima et al., 2017; Norsworthy et al., 2017; Enriquez et al., 2025). Furthermore, the functional maturation of these in vitro models is limited by the absence of the native microenvironment. To overcome this ‘maturation bottleneck,’ approaches have started to be integrated into differentiation workflows. These methods include co-culture with astrocytes or microglia to enhance synaptic pruning and ionic current maturation (VanderWall et al., 2019), the development of retinal-thalamic assembloids to provide RGCs with their physiological axonal targets (Fligor et al., 2021), and the application of extracellular matrix (ECM) scaffolds or mechanical patterning to sustain long-term culture (Gomes et al., 2025). Additionally, chronic electrical stimulation has been shown to improve the electrophysiological excitability of RGCs, bringing their functional profiles closer to their adult in vivo counterparts (Huberman et al., 2008). By shifting from pan-RGC generation to these subtype-targeted and maturation-enhanced protocols, in vitro models are becoming increasingly relevant for studying the subtype-specific vulnerability observed in diseases like glaucoma.

The ability to more accurately generate these cells from hPSCs, allows for the study of the cellular mosaicism that exists among RGCs of the human retina, with important implications for how these subtypes differ in their functionality as well as how they may be affected in disease states.

Retinal organoids are three-dimensional structure derived from stem cells that recapitulate the temporal development and spatial lamination of the in vivo retina. They display characteristic stratification, with RGCs located in inner layers and photoreceptors in the outer peripheral layers. While many studies have utilized organoids to study normal histology, physiology and diseases of the outer retina including photoreceptors and RPE (Phillips et al., 2012; Völkner et al., 2016; Wahlin et al., 2017; Zhong and Gutierrez, 2014), fewer have focused specifically on RGC development (Ohlemacher et al., 2019). Because RGCs are among the first cell types to develop, they provide a more feasible research timeline than that of photoreceptors, which can take over 200 days to mature. Several protocols have been developed to enrich organoids for RGCs using various strategies (Freude et al., 2020; Gill et al., 2014; Gill et al., 2016; Rabesandratana et al., 2020; Reichman et al., 2017; Soucy et al., 2023; Tanaka et al., 2016; Zhu et al., 2018). However, a recurring limitation in these studies is the focus on pan-RGC markers rather than specific subtype identification. Examples of most used protocols are briefly summarized in the following paragraphs.

Zhu et al. (2018): This protocol treated hPSCs with BMP, Wnt, blockers, supplemented with IGF1 for 5–6 days during maintenance passages, to facilitate retinal induction (Zhu et al., 2018). While the resulting 6- and 12-week organoids expressed BRN3 and ISL1, no absolute cell counts were performed to assess the fold increase in RGCs, and no subtype-specific markers were utilized.

Ohlemacher et al. (2016) & Langer et al. (2018): These researchers found that RGCs expressed RGC-associated markers that reached high levels by day 50 in cultures (about 8–12 times compared to the expression level at day 25), including: ATOH7, PAX6, BRN3B, ISL1, RBPMS, SNCG, and OPSIN4. By day 70, these RGCs expressed MAP2 in somatodendritic regions and TAU in axonal extensions (Ohlemacher et al., 2016; Langer et al., 2018). While successful in enriching the general RGC population, this protocol did not differentiate between specific subtypes.

Tanaka et al. (2015): Using a modified protocol adapted from Nakano et al. (2012) without BMP, Wnt, Notch, TGFB blockers (Tanaka et al., 2015). This group observed a 30-fold increase in the expression of BRN3B, ATOH7, ISL1, SNCG, and TUJ1 by day 34 (Tanaka et al., 2015). Again, the analysis remained at the pan-RGC level.

Freude et al. (2020): This study explored the theory that enriching for MG could support RGC growth. The authors identified MG via GFAP, RLBP1, and CD44 expression, while RGCs were identified using BRN3A. Although they used magnetic-activated cell sorting (MAC) to isolate RGC progenitors (expressing CHX10 and Nestin), the study did not employ markers to distinguish between RGC subtypes.

Reichman et al. (2017) & Rabesandratana et al. (2020): These studies utilized a chemically defined E6 medium followed by ProB27 medium supplemented with FGF2 (Reichman et al., 2017). They observed that while BRN3A, RBPMS, and HuC/D were present between days 56 and 84, RGC density decreased by day 98, a common “loss of RGCs” phenomenon in aging organoids. To counter this, they dissociated day-56 organoids and replated them in 2D culture, which significantly enhanced RGC survival and maturation. Flow cytometry confirmed that 60% of the resulting cells were THY1-positive RGCs (Rabesandratana et al., 2020).

Kim et al. (2023): This group utilized a pro-neural induction medium and transitioned manually isolated optic vesicles to 3D suspension at day 25, adding retinoic acid at day 42. RNA-seq confirmed THY1 expression (Kim et al., 2023). Similarly, organoids grown from glaucoma-patient iPSCs (e.g., OPTN E50K mutants) demonstrated increased apoptosis and impaired axonal transport (VanderWall et al., 2020). These models are vital for drug screening but currently generate heterogeneous RGC populations without specific subtype enrichment (Daniszewski et al., 2022; Hameed and Sharma, 2025).

In summary, all the above-mentioned retinal organoids grown from PSCs derived from either healthy or diseased individuals never elaborated on the importance of enriching specific subtypes nor were they developed with this idea in mind. There remains a significant gap in the field regarding the directed differentiation of high-acuity midget cells, motion-sensitive parasol cells, or specialized types like recursive RGCs. Future studies should address this gap by developing protocols that utilize subtype-specific molecular markers and transcription factor biasing. This will enable more precise investigations into the pathogenesis of glaucoma and facilitate the development of targeted therapeutic interventions for the most vulnerable RGC subtypes.

In contrast to other retinal neurons, RGCs are a diverse population, with many distinct subtypes identified in varied species, including humans. Each subtype is associated with characteristic topography, functionality, and responses to disease. To replace RGCs lost during disease or injury, fully characterized human-derived stem cell culture and differentiation protocols need to be established for each subtype. However, the identification, characterization, and culture of RGC subtypes in humans still faces many challenges. The development of this field will depend on an understanding of how each subtype is generated during development, and the discovery of new selective markers, particularly for vulnerable subtypes. This knowledge will facilitate efforts to preserve subtype function and survival, and to enrich and manipulate them in cell rescue and replacement therapies.