We first examined RGC labeling using four neurogenic tagging mouse lines (Hirata et al., 2021)¹ and found that the Neurod1^CreER (D1B) line labeled RGCs most effectively. To obtain a fine temporal profile of labeling in the retina, we crossed the Neurod1^CreER (D1B) driver line with the Tau^mGFP-nLacZ reporter mouse, which expresses dual nucleus- and membrane-localized reporters under a constitutive neuronal promoter after excision of the loxP-STOP-loxP cassette (Hippenmeyer et al., 2005) (Figure 1A). The Tau-driven reporter has been widely used to label differentiated neurons in multiple contexts, such as neurons in the interpeduncular nucleus (Ruiz-Reig et al., 2019), the jugular and nodose ganglia (Lowenstein et al., 2024), and the spinal cord (Xu et al., 2013). In our study, when a ubiquitous Cre driver line, CAG-Cre (Sakai and Miyazaki, 1997; Sato et al., 2011), was crossed with the Tau reporter line, strong reporter expression was predominantly observed in the ganglion cell layer (GCL; Supplementary Figure 1). This indicates that Tau-driven reporter expression is slightly biased toward RGCs in the retina.

In the Neurod1^CreER (D1B) driver line with the Tau^mGFP-nLacZ reporter mouse, reporter expression driven by the Tau locus was also confined to the GCL when Cre recombination was induced between E11.5 and E15.5 (hereafter referred to as TM11.5 and TM15.5, respectively; Supplementary Figure 1). In this context, we found that more than 85% of reporter-positive cells co-expressed the established RGC marker, RBPMS (Supplementary Figure 1). A single administration of TM at E13.5 most abundantly induced loxP recombination in RGCs (approximately 440 neurons per flat-mount retina; see also Supplementary Figure 1), whereas fewer recombination events were observed at earlier and later stages. This temporal profile corresponds well with the timing of RGC production previously described in the developing mouse retina (Young, 1985).

At the time of TM administration at E13.5, we characterized the cell cycle status of labeled RGCs using the thymidine analog EdU (Figure 1B), and the labeled cells were subsequently analyzed at P0. The results indicated that RGCs were most susceptible to TM-induced recombination 6 h after the final DNA synthesis. This result also suggested that a fraction of recombination may have occurred in the progenitor stage because some of the RGCs incorporated EdU, even after TM administration. This interpretation is consistent with a previous report showing that Neurod1 is expressed predominantly in postmitotic neurons but occasionally in mitotically active cells in the developing mouse retina (Liu et al., 2008). The same conclusion is also supported by RNA-seq-based transcriptomic data (Clark et al., 2019; Chen et al., 2021; Wu et al., 2021).

We next examined the spatial distribution of the neurogenic-tagged RGCs in flat-mount retinas at P21 (Figure 1C). Previous birthdating studies have shown that RGCs are generated from the center to the periphery in a wave-like pattern (Young, 1985; Lyu and Mu, 2021). Although this trend was not very clear in our analyses, the early tagged (TM11.5) neurons were more concentrated in the central retina, whereas the later-tagged (TM14.5) neurons were distributed more widely across the retina (Figure 1C). Because RGC subtypes were not distinguished in this analysis, the overlapping distributions of individual subtypes may have reduced the spatial resolution of the individual differentiation waves.

RGC axons from the eye project through the optic tract to the retinorecipient brain targets (Morin and Studholme, 2014; Martersteck et al., 2017; Kerschensteiner and Feller, 2024). We examined where the neurogenic-tagged RGCs project their axons. To label axons in an eye-specific manner, Neurod1^CreER (D1B) mice were crossed with R26-CAG-LF-mTFP1 reporter mice, in which monomeric Teal Fluorescent Protein 1 (mTFP1; Clavularia cFP484) was expressed after the removal of double STOP cassettes by two recombinases, Cre and Flippase (Figure 2A) (Imayoshi et al., 2012). We chose to analyze three stages of TM administration, TM11.5, TM13.5, and TM15.5, as a two-day interval could reveal differences, if any. After the Cre-loxP STOP cassette was removed by TM in Neurod1^CreER (D1B) mice, an AAV2 vector coding Flp recombinase (AAV2-CAG-FLPo) was intravitreally injected into one eye at the postnatal stage, and axon projections were analyzed at P28-P35. Thus, only axons from neurogenic-tagged RGCs in one eye were visualized in the brain.

Molecular markers have been used to characterize RGCs and some RGC subtypes, such as alpha RGCs, ipRGCs, ON–OFF direction-selective RGCs, and F-RGCs (Hattar et al., 2002; Kwong et al., 2010; Kay et al., 2011; Rodriguez et al., 2014; Rousso et al., 2016; Krieger et al., 2017). We analyzed marker expression in the neurogenic-tagged RGCs at P21 (Figure 5). For this purpose, the neurogenic tagging driver, Neurod1^CreER (D1B), was crossed with the Tau^mGFP-nLacZ reporter line. Nuclear-localized β-galactosidase (nls-β-gal) was then detected by immunohistochemistry together with RGC markers (Figure 5A). Although the number of neurogenic-tagged RGCs (nls-β-gal-positive) varied among the TM injection stages (Supplementary Figure 1), the total number of single marker-positive RGCs remained constant. These results suggest that these molecular markers reliably identified the RGC subtypes in this analysis.

Figure 5B shows a bubble plot summarizing the results of this analysis, including the percentage and number of marker-positive cells among the neuron-tagged cells in the ganglion cell layer. Specifically, the pan-RGC marker, RBPMS, was expressed in the majority of neurogenic-tagged RGCs at all TM induction stages. The analyzed RGC subtypes accounted for less than 25% of the TM-labeled RGCs at any of the tagged time points. Among them, ON–OFF direction-selective RGCs marked by CART were slightly more abundant in TM13.5- and TM15.5-labeled RGC populations compared with those labeled at TM11.5; however, this difference did not reach statistical significance based on Fisher’s exact test (significance threshold: p < 0.05). The proportions of other RGC subtypes did not differ among the TM stages (Fisher’s exact test).

We next analyzed the soma size and dendritic morphology of the neurogenic-tagged RGCs at P21 (Figure 6) because soma size and dendritic field size differ among RGC subtypes (Rockhill et al., 2002; Goetz et al., 2022). For this purpose, the neurogenic tagging driver Neurod1^CreER (D1B) was crossed with the Thy1-STOP-EYFP reporter line, which has been widely used to label RGCs and their dendrites in previous studies (Buffelli et al., 2003; Kim et al., 2008) (Figure 6A). The morphology of labeled RGCs was examined in flat-mount retinal preparations. We found that the soma size of TM11.5 RGCs was slightly, but significantly, larger than that of TM15.5 RGCs (Figures 6B,E). The dendritic field size showed a trend toward group differences (Figure 6C). The relationship between dendritic field diameter and soma diameter showed a positive correlation (Figure 6D; Spearman’s r: TM11.5, r = 0.6587, p < 0.0001; TM13.5, r = 0.5920, p = 0.0004; TM15.5, r = 0.5463, p = 0.0008).

The position of dendrites within retinal sublaminas is another criterion for categorizing RGC subtypes (Figures 6F–L). We examined the dendritic positioning of the neurogenic-tagged RGCs by co-immunostaining with an antibody against the vesicular acetylcholine transporter (VAChT). VAChT immunostaining labels two layers of processes from the starburst amacrine cells in the inner plexiform layer (IPL). These layers are closely associated with the ON- and OFF-type RGCs in the retina (Huberman et al., 2009; Martersteck et al., 2017). Based on the stratification of their dendrites within the IPL, we categorized RGCs into four types: (1) ON type, in which the dendritic arbor is confined to the ON sublamina (Figures 6G,G’); (2) OFF type, in which the dendritic arbor is ramified within the OFF sublamina (Figures 6I,I’); (3) bistratified type, in which dendrite branches are distinctly confined to both ON and OFF sublaminas (Figures 6J,K); and (4) diffuse types, in which the dendritic arbor is not confined to a single sublamina but spans the IPL (Figures 6H,L). The proportions of these four types were compared among TM11.5, TM13.5, and TM15.5, using a chi-square test, which showed significant differences (χ² = 44.01, df = 6, p < 0.0001). Post hoc comparisons with Bonferroni correction also revealed significant differences between TM11.5 and TM13.5 (χ² = 12.69, p = 0.0053), TM11.5 and TM15.5 (χ² = 36.35, p < 0.0001), and TM13.5 and TM15.5 (χ² = 26.18, p < 0.0001) (adjusted significance threshold: α = 0.05/3 = 0.0167). Standardized residual analysis indicated that ON-type cells were overrepresented among TM11.5 RGCs, whereas this cell type was rare among TM15.5 RGCs. In contrast, diffuse-type cells were underrepresented among the TM13.5 RGCs.

In summary, our analysis revealed that RGC axon projections in the LGN and MTN were segregated depending on the stage of tamoxifen injection (Figure 7). In addition, morphological analyses of labeled RGCs showed that TM11.5 RGCs had slightly larger somata and included a higher proportion of ON-type RGCs compared with TM15.5 RGCs.

Axon projections to the MTN are a unique feature of ON-type direction-selective RGCs (ON DSGCs) belonging to the accessory optic system (AOS) (Yonehara et al., 2009; Dhande et al., 2013). Therefore, TM11.5 RGCs are expected to include this subtype. The AOS stabilizes retinal images by driving compensatory eye movements against head movements via projections to the MTN. Although the early generation of this subtype has not been previously described, it has been investigated as SPIG1-RGCs (Yonehara et al., 2009) (SPIG1 is also known as FSTL4), Hoxd10-RGCs (Dhande et al., 2013), and Pcdh9-Cre-labeled RGCs (Lilley et al., 2019). Our observations show that approximately 50% of TM11.5 RGCs have monostratified dendrites confined to the ON sublamina (Figure 6F). In addition, TM11.5 RGCs exhibited relatively large soma sizes. These features are consistent with the known characteristics of ON-DSGCs (Yonehara et al., 2009; Dhande et al., 2013; Lilley et al., 2019).

Another possible subtype of TM11.5 RGCs is ipRGCs. Although the ON DSGCs described above do not project to the LGN, the LGN projections in our study were clearly labeled by TM11.5 RGC axons. In particular, axon targeting of the vLGN was prominent, and these characteristics were similar to those reported for ipRGCs (Hattar et al., 2006). The occupation of axons in the deep (furthest from the pia) and dorsal regions of the contralateral optic tract also resembles the reported pattern of ipRGCs (Hattar et al., 2006). The observation that TM11.5-labeled RGCs are preferentially located in the central retina and exhibit relatively large soma sizes is consistent with the characteristics of ipRGCs. However, this possibility is not supported by our findings at the TM stage, which showed a low percentage of melanopsin-expressing RGCs and no detectable projections to the SCN.

Because ipsilateral projections to the LGN and SC were observed from TM13.5 and TM15.5 axons, TM13.5 RGCs (as well as TM15.5 RGCs) clearly include the ipsilateral RGC population. In both the LGN and SC, axon terminals were confined to ipsilateral target regions, consistent with the findings of previous studies (Morin and Studholme, 2014). The peak of production of ipsilateral RGCs is estimated to occur around E13-E14 in mice (Marcucci et al., 2019). This time course of ipsilateral RGC production is consistent with our observations of ipsilateral projections revealed by neurogenic tagging.

There are several possible subtypes that could account for the contralateral projections of TM13.5 RGCs. Projections to the core region of the dLGN and superficial layer of the SC are consistent with alpha RGCs (Martersteck et al., 2017). Another candidate subtype is ON–OFF DSGCs, such as BD-RGCs, which are CART-positive and have moderately sized dendritic fields (Kim et al., 2010; Kay et al., 2011). As conclusive evidence is still lacking, we do not further speculate on these possibilities.

It is highly likely that TM15.5 RGCs include ON–OFF DSGCs, such as DRD4-RGCs or W9-RGCs (Huberman et al., 2009; Kay et al., 2011). Axons of these subtypes have been shown to project to the shell region of the dLGN and superficial SC, but not to the vLGN or MTN. These RGC subtypes are CART-positive and have relatively small dendritic fields. All these characteristics are consistent with the observed properties of TM15.5 RGCs. Moreover, many TM15.5 RGCs exhibited bistratified dendrites in both the ON and OFF sublaminas of the retina, indicating that they are indeed ON–OFF RGCs. One potential discrepancy is that the birth timing of DRD4-RGCs was previously determined to be E12.5 (Osterhout et al., 2014). This timing appears slightly earlier than the stage estimated using our neurogenic tagging method.

Axonal projection analyses also suggest that TM15.5 RGCs include ipsilateral RGCs as additional subtypes. This finding is consistent with previous observations that the production of ipsilateral RGCs decreases but continues to be observed at E15 in mice (Marcucci et al., 2019).

Annotation methods for RGC subtyping using immunohistochemical markers can identify only a limited subset of RGC subtypes, collectively accounting for approximately 20% of the total RGC population (Sanes and Masland, 2015; Nguyen-ba-charvet and Rebsam, 2020). In contrast, our analyses targeted many more RGC subtypes, including unidentified subtypes. Although our classification was based on a limited number of molecular markers, this approach could provide an initial framework for distinguishing these populations. The use of additional molecular and morphological markers is necessary for a more exhaustive analysis of RGC subtypes. Recent transcriptomic studies may help in further classification because they provide new combinations of numerous subtype-specific markers in RGC populations (Baden et al., 2016; Rheaume et al., 2018; Tran et al., 2019; Goetz et al., 2022). In the future, to further refine subtype identity and establish functional relevance, the neurogenic tagging mouse could be combined with electrophysiological imaging or recording, as well as with activating or silencing manipulations (Yonehara et al., 2009; Kay et al., 2011; Kofuji et al., 2016). For example, calcium imaging would enable selective labeling and functional analysis of these neurons in vivo. In addition, crossing the neurogenic tagging mouse line with Rosa26-loxP-STOP-loxP-DTA or DTR mice would allow selective ablation of the targeted cells. Using these mice, visual assays, such as optokinetic response tests, could reveal the functional roles of RGC subtypes in vivo.

Temporal precision may represent another limitation in distinguishing closely related RGC subtypes. Higher-resolution approaches, such as the direct injection of 4-hydroxytamoxifen (4-OHT), the active metabolite of tamoxifen, may allow separation within a narrower developmental window (Guenthner et al., 2013; Hatanaka et al., 2024). Unlike nucleotide-based neuronal birthdating methods, the neurogenic tagging approach used in this study does not directly provide information regarding the timing of the cell cycle exit (Llorens-Martín and Trejo, 2011; Govindan et al., 2018). This limitation may restrict certain applications but also represents an advantage in other contexts, as the two methods are based on distinct biological mechanisms.

Finally, it is possible that the Neurod1 promoter does not label all progenitors contributing to the full diversity of RGC subtypes, thereby creating bias in the current analyses of RGC subtypes. As Neurod1 has been shown to partially compensate for the loss of Math5 (Atoh7) (Mao et al., 2008), RGC differentiation is likely to involve other transcription factors.

The Neurod1^CreER (D1B) mouse line (official name: C57BL/6-Tg (Neurod1-cre/ERT2) D1BTahi) was generated as previously described (Hirata et al., 2021). CAG-Cre mice, constitutively expressing Cre recombinase, have been described in previous studies (Sakai and Miyazaki, 1997; Sato et al., 2011). Tau-STOP-mGFP-IRES-NLS-lacZ reporter mice (RRID: IMSR_JAX: 021162) (Hippenmeyer et al., 2005), maintained on a mixed CD-1 background, were provided by Dr. Silvia Arber (Friedrich Miescher Institute for Biomedical Research) and backcrossed with C57BL/6 wild-type mice (RRID: MGI: 5295404) for at least four generations. Thy1-STOP-YFP reporter mice (B6.Cg-Tg (Thy1-EYFP) 15Jrs/J; RRID: IMSR_JAX:005630) (Buffelli et al., 2003) were obtained from The Jackson Laboratory. R26-CAG-LF-mTFP1 reporter mice (B6;129S6-Gt(ROSA)26Sor^{<tm1(CAG-mTFP1)Imayo>}/ImayoRbrc; RRID: IMSR_RBRC 05146) (Imayoshi et al., 2012) were obtained from the RIKEN BRC through the National BioResource Project of the MEXT/AMED, Japan. Heterozygous Neurod1^CreER (D1B) mice were crossed with homozygous reporter mice. In some experiments, CAG-Cre mice were crossed with Tau-STOP-mGFP-IRES-NLS-lacZ reporter mice. Embryonic day (E) 0.5 was defined as the day a vaginal plug was detected, and the day of birth was designated as postnatal day 0 (P0). Animals of both sexes were used for the analysis. All animal care and procedures were approved by the Institutional Animal Committees of the National Institute of Genetics and conducted in accordance with the institutional guidelines.

Tamoxifen (Sigma-Aldrich, T5648) was dissolved in corn oil (Sigma-Aldrich, C8267) at a final concentration of 9 mM together with 5 mM progesterone (Fujifilm Wako, 160-24511), and 250 μL of this solution was administered by intraperitoneal injection to staged pregnant mice. Injections were performed once on each indicated day from E11.5 to E15.5 (hereafter referred to as TM11.5 to TM15.5, respectively). Because tamoxifen treatment often delayed parturition, pups that were not delivered by E19.5 were collected by cesarean section and transferred to ICR foster mothers (RRID: MGI:5462094, Japan SLC Inc.). For dendrite morphology analyses, the injection volume of the tamoxifen solution was reduced to 100 μL at TM11.5, 50 μL at TM13.5, and 50 µL at TM15.5 per mouse to decrease the number of labeled neurons and minimize overlap.

The retina and brain for cryosections were prepared as previously described (Hirata et al., 2021). Briefly, the animals were anesthetized with isoflurane and transcardially perfused with phosphate-buffered saline (PBS) followed by 4% paraformaldehyde (PFA). The brains and eyes were dissected and further fixed with 4% PFA for 12–24 h at 4 °C. After fixation, the tissue was cryoprotected overnight in 30% sucrose in PBS and frozen in a mixture of OCT compound (Sakura Finetek) and 30% sucrose in PBS at a 2:1 ratio. Coronal sections (14 μm for retinal sections and 20 μm for brain sections) were cut using a cryostat and placed on MAS-coated glass slides (Matsunami Glass). The sections were treated with 0.5% blocking reagent (FP1020, PerkinElmer) containing 0.2–0.4% Triton X-100 (Nacalai Tesque Inc., 35501-15) for 30 min at room temperature and incubated overnight at 4 °C with primary antibodies: chicken anti- β-galactosidase (1:1000, Abcam Cat#ab9361, RRID: AB_307210), rabbit anti-GFP (1:1000, MBL International Cat#598S, RRID: AB_591816), rat anti-mTFP1 (1:500, ximbio Cat#155264), rabbit anti-RBPMS (1:100, Abcam Cat# ab194213, RRID: AB_2920590), goat anti-VAChT (1:3000, Millipore Cat#ABN100, RRID: AB_2630394), rabbit anti-Melanopsin (1:5000, Advanced Targeting Systems, Cat#AB-N38, RRID: AB_1608077), mouse anti-SMI-32 (1:3000, Covance, Cat#SMI-32P, RRID: AB_2314912), rabbit anti-FOXP2 (1:20000, Abcam, Cat#ab16046, RRID: AB_2107107), rabbit anti-CART (1:2000, Phoenix Pharmaceuticals, Cat#H-003-62, RRID: AB_2313614). The sections were then incubated for 2 h at room temperature with secondary antibodies: donkey Alexa Fluor 488-conjugated anti-rabbit IgG (1: 500–1000, Molecular Probes Cat#A-21206, RRID: AB_2535792), donkey Cy3-conjugated anti-rabbit IgG (1:1000, Jackson ImmunoResearch Labs Cat#711–165-152, RRID: AB_2307443), donkey Cy3-conjugated anti-mouse IgG (1:1000, Jackson ImmunoResearch Labs Cat#715–165-150, RRID: AB_2340813), donkey Alexa Fluor 488-conjugated anti-chicken IgY (1:1000, Jackson ImmunoResearch Labs Cat#703–545-155, RRID: AB_2340375), and donkey Cy3-conjugated anti-goat IgG (1:1000, Jackson ImmunoResearch Labs, Cat#705–165-147, RRID: AB_2307351). The sections were counterstained with DAPI (4′, 6-diamidino-2-phenylindole, Fujifilm Wako, Cat#045–30361). For staining of nucleus-localized-β-galactosidase, the sections were pre-treated for antigen retrieval by incubating in 10 mM sodium citrate with 0.05% Tween-20 (pH 6.0) at 95 °C for 20 min before blocking. For staining of flat-mounted retina, animals were fixed as described above, and a small radial slit was made at the dorsal pole of the retina for future orientation. The cornea was cut, the lens was removed, and the neural retina was detached from the eye. Dissected retinae were processed for staining, as described above.

Fluorescent images were captured with a confocal microscope (Olympus IX81 FV1000) using FluoView software (FV-ASW version 4.02; Evident Scientific, Japan) and processed as previously described (Hirata et al., 2019). Some images (Figures 5A, 6G–L) were captured with a CCD camera (Olympus DP71) mounted on a conventional fluorescence microscope (Zeiss Axioplan 2) using the Cell Sens Standard software. Brightness and contrast were adjusted equally across the entire image using the Adobe Photoshop software.

Neurod1^CreER (D1B); Tau^mGFP-nLacZ mice that had received tamoxifen at E13.5 were used for this analysis. A thymidine analog, 5-Ethynyl-2′-deoxyuridine (EdU, Tokyo Chemical Industry Cat#E1057; 50 mg/kg body weight), was injected intraperitoneally into pregnant mice at selected time points before or after tamoxifen injection. Eyes were dissected at P0, and coronal sections (16 μm thickness) were prepared as described above. The sections were subjected to antigen retrieval and immunostaining, as previously described (Hirata et al., 2021). Primary antibodies used for immunostaining were chicken anti-β-galactosidase (Abcam Cat#ab9361, RRID: AB_307210), followed by donkey Alexa Fluor 488-conjugated anti-chicken IgY (Jackson ImmunoResearch Labs Cat#703–545-155, RRID: AB_2340375). Incorporated EdU was detected by incubation in a solution containing 0.1 M Tris (pH 7.6), 2 mM CuSO₄, 3 μM Alexa Fluor 555 azide, triethylammonium salt (Thermo Fisher Scientific Cat# A20012), and 10 mM ascorbic acid for 40 min at room temperature.

Neurod1^CreER (D1B); R26-CAG-LF-mTFP1 mice were administered tamoxifen at a selected embryonic stage (TM11.5, TM13.5, or TM15.5). Juvenile mice (P8-P21) were used for the virus injection. Under anesthesia with 1.0–1.5% isoflurane, a small hole was made in the eye just below the limbus, using an insect pin to release intraocular pressure. One to two microliters of AAV2-CAG-FLPo solution (4.5 × 10¹² genome copies/ml; Vector Biolabs, Cat# VB1313) was injected through the hole using a pulled and beveled glass pipette (Drummond Cat#5–000-1001-X10). A glass pipette was attached to a Hamilton syringe and micromanipulator to control the injection volume. Two to three weeks after virus injection, the mice were transcardially perfused with 4% PFA in PBS, and the eyes and brains were processed for immunohistochemistry as described above.

The number of tagged neurons containing EdU was counted using a conventional fluorescence microscope (Zeiss Axioplan 2). The number of tagged neurons expressing markers of RGC subtypes was counted from the photographed images. The proportions of cells were calculated using Microsoft Excel.

Soma and dendritic field sizes were analyzed as previously described (Berson et al., 2010; Estevez et al., 2012). Briefly, the outlines of convex polygons that minimally enclosed the somatic or dendritic profiles of individual labeled cells were traced. These outlines were exported to ImageJ software² to calculate the diameter of a circle with the same area as the convex polygon. Dendritic stratification was analyzed with reference to cholinergic bands visualized using VAChT staining. For the ON and OFF subtypes, the boundary between ON and OFF was assigned at the midpoint between the two VAChT bands (approximately 50% IPL depth), as previously described (Landi et al., 2007).

Scatter and bubble plots were generated using GraphPad Prism, version 10. Statistical analyses were performed using GraphPad Prism 10 and Microsoft Excel after confirming that the data met the assumptions of the respective tests (Figures 5B, 6B–D,F). In brief, for soma diameter and dendritic field diameter measurements, Shapiro–Wilk and Levene’s tests revealed non-normal distributions and unequal variance across groups, respectively. Based on these results, the Kruskal–Wallis test with Dunn’s multiple comparison test was conducted in this analysis.