A potential role for somatostatin signaling in regulating retinal neurogenesis

Neuropeptides have been reported to regulate progenitor proliferation and neurogenesis in the central nervous system. However, these studies have typically been conducted using pharmacological agents in ex vivo preparations, and in vivo evidence for their developmental function is generally lacking. Recent scRNA-Seq studies have identified multiple neuropeptides and their receptors as being selectively expressed in neurogenic progenitors of the embryonic mouse and human retina. This includes Sstr2, whose ligand somatostatin is transiently expressed by immature retinal ganglion cells. By analyzing retinal explants treated with selective ligands that target these receptors, we found that Sstr2-dependent somatostatin signaling induces a modest, dose-dependent inhibition of photoreceptor generation, while correspondingly increasing the relative fraction of primary progenitor cells. These effects were confirmed by scRNA-Seq analysis of retinal explants but abolished in Sstr2-deficient retinas. Although no changes in the relative fraction of primary progenitors or photoreceptor precursors were observed in Sstr2-deficient retinas in vivo, scRNA-Seq analysis demonstrated accelerated differentiation of neurogenic progenitors. We conclude that, while Sstr2 signaling may act to negatively regulate retinal neurogenesis in combination with other retinal ganglion cell-derived secreted factors such as Shh, it is dispensable for normal retinal development.

Neurotransmitters play important roles in regulating neurogenesis and development. Although neurotransmitters are generally considered in terms of neuronal signaling, neurotransmitter expression precedes the nervous system on both evolutionary and developmental time scales 1 . In fact, neurotransmitter genes are found in species without a central nervous system like sponges and the social amoeba Dictysotelium 2 . The receptors for several neurotransmitters such as GABA, glutamate, and serotonin are expressed on progenitor cells of the central nervous system prior to the formation of functional synapses [3][4][5] . The neurotransmitters themselves likely act as extrinsic signaling factors to guide neural development 1,[3][4][5] .
It may be that developmental signaling represents the ancestral role for neurotransmitters before their use in signaling by mature neurons 2 . Neurotransmitters and neurotransmitter receptors are expressed in neural progenitors and/or immature neurons in many different regions of the vertebrate CNS [6][7][8] . Over the past three decades, there have also been a large number of studies in which pharmacological manipulation of a number of neurotransmitter systems have been reported to modulate multiple different aspects of neuronal development in different brain regions, including progenitor proliferation, neurogenesis, axonal guidance, synaptogenesis and neuronal survival 3,9,10 . The hippocampus is the brain region with the greatest number of reported neurotransmitter developmental signals: ATP, serotonin, VGF, vasoactive intestinal peptide (VIP), histamine, glycine, acetylcholine, dopamine, cholecystokinin, oxytocin, corticotropin-releasing hormone (CRH), ghrelin, nitric oxide (NO), GABA, and glutamate have all been found to play different roles in developmental and/or adult rodent hippocampal neurogenesis 3,6,[11][12][13][14][15][16][17][18][19][20][21][22] . Other regions have fewer reported neurotransmitter developmental signals, like GABA, glutamate, histamine, VIP, and taurine 3,9,13,15 in neocortical development, and pituitary adenylate cyclase-activating peptide (PACAP) in cerebellar development. Serotonin regulates the development of multiple brain regions. However, the specificity of these ligands-particularly at the high concentrations often used in these studies-is unclear. Although mutants for many genes directly related to neurotransmission are available, www.nature.com/scientificreports/ developmental phenotypes are seldom characterized. The precise role of neurotransmitters in regulating brain development thus remains largely unclear. The retina is an accessible and relatively less complex CNS region that serves as a useful model for understanding molecular mechanisms controlling the development of more complex brain regions 23 . Extrinsic signaling also plays a critical role in regulating retinal development. In addition to classical growth and differentiation factors that are present in the developing retina-such as bFGF and TGF alpha 24 , GDF11 25 , VEGF 26 and SHH 27 -a number of neurotransmitters are expressed by retinal progenitors and/or immature retinal neurons 28 . Pharmacological manipulation of serotonin, glutamate, or GABA signaling impacts retinal development in Xenopus laevis 1 , and endogenous opiate-like peptides have been reported to regulate neurogenesis in rodent retina 27,29 . However, genetic data to support these findings is lacking. Moreover, until recently, comprehensive and quantitative analysis of the cellular expression patterns of neurotransmitter biosynthetic enzymes and receptors during retinal development has not been available. This has changed with the generation of several large-scale single-cell RNA-sequencing (scRNA-Seq) datasets from developing mouse and human retina 30,31 . This work has identified four different neuropeptides and/or their receptors as being strongly and selectively expressed during early stages of retinal development-somatostatin, neuropeptide Y, galanin, and proenkephalin. Each of these neuropeptides has been reported to regulate brain development. Somatostatin acts as a generalized growth factor inhibitor, and somatostatin expression in the hypothalamus decreases with age along with hypothalamic neurogenesis 13,32 . Neuropeptide Y promotes proliferation in the subventricular zone and hippocampal precursor cells 6,13,15,33 . It also promotes proliferation of dissociated rat retinal progenitor cells, apparently through NO 34 . Galanin, acting through its receptor GalR3, promotes neurogenesis in the adult hippocampus 6,13,17,35 . Finally, proenkephalin impacts cerebellar development and adult hippocampal neurogenesis 36,37 .
To define any of these neuropeptides as an endogenous regulator of retinal development, we need to identify (1) a source of the ligand, (2) impacts on neurogenesis from activation and knockout of the appropriate receptors, and (3) an in vivo phenotype 38 . To accomplish this, we use publicly-available single-cell RNA-seq data of mouse and human retinal development 30,31 , a pharmacologic screen on murine embryonic retinal explants, and single-cell sequencing of retinal explants and mutant retinas. We find that though they play roles in development elsewhere in the CNS or in other in vitro situations, neuropeptide Y, galanin, and proenkephalin do not impact neurogenesis in embryonic retinal explants. Activation of the somatostatin receptor Sstr2 reduces the production of photoreceptors in this model. However, somatostatin likely plays a redundant role in retinal development as its knockout does not produce the same effect in vivo.

ScRNA-seq identifies neuropeptides dynamically expressed during retinal neurogenesis. A
previously published single-cell RNA sequencing (scRNA-seq) study profiled mouse retinal development across ten-time points. We reasoned that this gene expression dataset could be used to identify candidate neurotransmitters impacting mouse retinal development. We downloaded the dataset and used the stored cell type and age data to investigate the expression of neuropeptides in mouse development.
Cellular expression patterns of neuropeptides are readily identified using scRNA-seq. The neuropeptides somatostatin (Sst), galanin (Gal), neuropeptide Y (Npy), and proenkephalin (Penk) were selected for further study as they showed the greatest correlation of expression with a single cell type (Fig. 1A). Sst and Gal are expressed by mouse retinal ganglion cells most strongly between E14 and P0 ( Fig. 1A,C, S1A), and by human retinal ganglion cells between gestational day 42 and gestational week 13 (Fig. S1D,F). Gal is also strongly expressed in early-stage neurogenic progenitors 39 . Npy is expressed primarily by late-stage primary retinal progenitor cells between E18 and P2, and Penk is expressed by neurogenic progenitor cells mostly between E14 and P2 ( Fig. 1A, S1B,C). The expression of neuropeptides by specific cell types over delimited windows during retinal development suggests that they may be acting as signaling molecules impacting retinogenesis.
In sharp contrast, with one exception, the receptors for Sst, Gal, Npy, and Penk were either not detected or only barely detectable in both developing mouse and human retina. The one notable exception was somatostatin receptor 2 (Sstr2) which is strongly and selectively expressed by neurogenic RPCs from E12 to P2 in mice (Fig. 1B,D). This matches the period when Sst is prominently expressed in immature RGCs. SSTR2 is also expressed by neurogenic RPCs, and undifferentiated cones, bipolar cells, and RGCs, in the human retina, while SST is also expressed in immature RGCs ( Figure S1D-G). To confirm the expression of Sstr2 in the neurogenic progenitors of developing mouse retina, we used RNAscope single molecule fluorescent in situ hybridization to probe for Sstr2 and Neurog2 mRNA expression in E14 and P0 retinas (Fig. 1E,F). We found a high level of overlap in expression with a mean of 70% (+/− 3%) of Sstr2 + cells expressing Neurog2 and 80% (+/− 4%) of Neurog2 + cells expressing Sstr2 at E14, while at P0, 78% (+/− 7%) of Sstr2 + cells expressed Neurog2 and 83% (+/− 8%) of Neurog2 + cells expressed Sstr2 (Fig. 1G). This led us to hypothesize that Sst could be acting as a signal released by retinal ganglion cells to influence neurogenic progenitors through Sstr2.
Sstr2 signaling inhibits expression of photoreceptor marker genes in retinal explants. We decided to use a pharmacologic screen, treating mouse embryonic retinal explants with small molecule agonists and antagonists to the receptors of the four neuropeptides of interest, as a cost-effective test for an effect on retinal development. Because Sstr2 was highly and selectively expressed in neurogenic RPCs, while other Sst receptors were not expressed or barely detectable, we used an Sstr2-specific agonist and antagonist. For the other candidates, we ordered the neuropeptides themselves and antagonists general to all of their receptors; except Npy, for which no general antagonist was available. The explants were cultured for a period roughly equivalent to the expression of the relevant neuropeptide in vivo with either agonist, antagonist, or solvent control. We used www.nature.com/scientificreports/ qRT-PCR to measure the relative levels of multiple cell type marker genes between treated retinas and control as a readout of the effects of the treatments on the expression of cell type-specific markers. Treatment with Npy, galanin, and the galanin receptor antagonist M40 did not produce significant changes in expression for the tested marker genes ( Fig. 2A,B). Treatment with Met5-enkephalin, a bioactive processed product of Penk, decreased expression of the neurogenic progenitor markers Atoh7, Neurog2, and Sstr2 while the broad-spectrum opiate receptor antagonist naltrexone increased expression for Atoh7 and Sstr2 as measured by ANOVA (P-value < 0.0045, Fig. 2C). These results suggest that Penk signaling may reduce the proportion of neurogenic progenitor cells in retinal explants. Meanwhile, treatment with the Sstr2-specific agonist (1R,1'S,3'R/1R,1'R,3'S)-L-054,264 consistently nominally significantly decreased expression of the photoreceptor markers Crx and Otx2 (P-value < 0.05, Fig. 2D, S2B). Treatment with the Sstr2-specific antagonist CYN 154,806 did not impact marker gene expression, possibly because retinal explants lack RGCs, the endogenous source of Sst (Fig. S2A). This similarly suggests that Sst signaling through Sstr2 negatively regulates photoreceptor differentiation.
We next conducted further studies of the potential role of Penk and Sst/Sstr2 in regulating retinal development. To increase our ability to detect an impact on retinal development, the initial assay had been performed using concentrations orders of magnitude in excess of the ligand's reported EC50 or IC50 values. It is therefore possible that the observed effects were due to nonspecific or off-target effects of these ligands, rather than selective modulation of neuropeptide receptors. In order to test this, we cultured explants in a range of treatment concentrations below that tested previously. We then tested for expression of Atoh7, Neurog2, and Sstr2 following Met5-enkephalin and naltrexone treatment, and for Crx and Otx2 following treatment with the Sstr2 agonist. Treating with reduced concentrations of Met5-enkephalin and naltrexone failed to reproduce the effect seen at saturating concentrations ( Fig. S2C-H). However, treatment with lower concentrations of the Sstr2 agonist reduced photoreceptor marker gene expression in a dose-dependent manner, suggesting that it was indeed acting through a specific receptor (P < 0.05, Fig. 2E,F). Moreover, treatment with 500 nM Sstr2 agonist had no significant effect (P > 0.1) on Crx and Otx2 expression in explants obtained from Sstr2 −/− mice, as measured by t-test (Fig. 2G,H), demonstrating that these effects are selectively mediated by activation of Sstr2 signaling.

Sstr2 signaling inhibits photoreceptor specification in embryonic retinal explants in a dose-dependent manner.
It is unclear whether Sstr2 signaling actually inhibits photoreceptor generation or reduces expression levels of Otx2 and Crx. While qPCR cannot distinguish between these possibilities, scRNA-seq can readily do so. We applied MULTI-seq 40 to retinal explants grown in a range of Sstr2 agonist concentrations. MULTI-seq allows us to generate a single expression library with cells labeled for each treatment www.nature.com/scientificreports/ condition ( Fig. 3A). After clustering and cluster marker identification, we identified the cell type corresponding to each cluster ( Fig. 3B, S3A). With the sample and cell type identity known for each cell, we quantified and compared the proportions of cell types across the different treatment conditions (Fig. 3C). Increasing the concentration of Sstr2 agonists decreased the proportion of photoreceptors and increased the proportion of primary progenitors as a fraction of cells in the explants. Cell type proportions across different treatment conditions were significantly different by chi-squared test (P < 2.2 E-16). In addition, the level of photoreceptor marker gene expression in photoreceptor clusters was not significantly changed in response to different agonist concentrations by Wilcoxon rank-sum test (Fig. 3D, S3B). We confirmed the differences in cell type proportions by immunostaining for the primary progenitor marker gene Lhx2, the cell cycle marker Ki67, and the photoreceptor marker Recoverin. Treatment with saturating levels of Sstr2 agonist increased the relative fraction of Lhx2 + cells and the thickness of the explants, while decreasing the relative ratio of Recoverin + cells to Lhx2 + cells ( Fig. 3E-I). Though suggestive, these differences did not reach significance by Welch's t-test. This demonstrates that somatostatin signaling exerts, at best, modest changes in retinal cell type composition in retinal explants. The treatment did not impact the area-normalized number of cells expressing Ki67, though the observed increase in explant thickness may have affected this measurement (Fig. S3C,D). While not statistically significant, the immunostaining results are in agreement with both the qPCR and MULTI-seq data. The qPCR, MULTI-seq, and immunostaining results taken together demonstrate that, in retinal explants, Sstr2 signaling modestly inhibits photoreceptor specification while increasing the relative fraction of primary RPCs.

No significant change in cell composition is detected in Sstr2 −/− retinas. To test whether Sstr2
signaling regulates photoreceptor differentiation in vivo, we applied MULTI-seq to the retinas of littermates that were wild type, heterozygous, and homozygous for a targeted deletion of Sstr2. We hypothesized that KO mice would have greater proportions of photoreceptors than their littermates. Mice were collected at P0 to match the age at which the explants were tested and P14 to determine whether terminal differentiation and/or survival of specific retinal cell types were affected. Following the same procedure for the sample and cell-type identification as we used for the explants (Fig. 4A,B, S4A,E-G), we found no significant difference in cell-type proportions between littermates of different genotypes at P0 by chi-squared test (Fig. 4C). The genotypes at P14 had significantly different cell type proportions, but this was likely driven by the over enrichment of rods in the Het cells compared to both WT and KO genotypes (Fig. S4H). While the significance of this is unclear, the smaller total number of cells profiled in each sample in the P14 dataset may have led to greater variability in the observed cell type proportions, leading to the observed discrepancy. We confirmed the similarity in cell-type proportions across genotypes by immunostaining for Lhx2, Ki67, and Recoverin, which showed no difference in retinal thickness, the density of Lhx2 + or Ki67 + cells, or in the ratio of Recoverin + to Lhx2 + cells by Welch's t-test at P0 (Fig. 4E-H, S4B-D). There was also no difference in Lhx2 + cell density or the combined thickness of the outer segment and outer nuclear layer at P14, or in the total retinal area controlled for body length in the adult mouse as measured by ANOVA (Fig S5A-E). All results found in the Sstr2 KO mouse by immunostaining were recapitulated in the Sst KO mouse (Fig. S6A-K). However, Sstr2 mRNA levels were increased in neurogenic RPCs in P0 Sstr2 mutants relative to controls as measured by Wilcoxon rank-sum test (P-value = 3.35E-5) (Fig. 4D), implying that Sstr2 signaling may negatively regulate Sstr2 transcription.
Sstr2 signaling promotes retinal neurogenesis and inhibits expression of progenitor-specific genes in retinal explants. We next examined our scRNA-Seq data in more detail to determine which genes showed altered expression in Sstr2 −/− retinas. We hypothesized that though we did not see a change in cell-  www.nature.com/scientificreports/ type proportions in vivo that were complementary to those seen in agonist-treated ex vivo retinal explants, we might nonetheless observe complementary changes in gene expression. We compared differentially expressed genes of wild type vs. Sstr2 −/− neurogenic RPCs in vivo and 500 nM versus 0 nM Sstr2 agonist-treated neurogenic cells ex vivo. We identified only one differentially expressed gene in common between these two comparisons. This was Xist, and likely simply reflected differences in the sexes of the animals among the different samples. In total, only eleven genes were identified in vivo and two identified ex vivo by Wilcoxon rank-sum test (P-value < 5E-5) (Suppl. Tables 1, 2). Neither of these lists of genes suggested a clear transcriptional impact from Sstr2 perturbation. We felt that traditional gene expression tests may be underpowered when looking for modest changes in gene expression in single cell types. We then looked for a mechanism of action for Sstr2 signaling by testing whether genes that differ between genotypes in vivo could distinguish cells from different treatments ex vivo, as a more nuanced way to evaluate differences in gene expression. To do this, we integrated our ex vivo and in vivo datasets (Fig. 5A,B) and, reasoning that the neurogenic cells that express Sstr2 are the most likely to be affected by differences in Sstr2 signaling, selected neurogenic cells that express Sstr2 from cluster 11 (Fig. 5C,D). We then used the 58 genes that are differentially expressed between Sstr2-positive neurogenic RPCs from wildtype, Sstr2 +/− , and Sstr2 −/− mice to guide the pseudotime program Monocle 2 41 in aligning all of the Sstr2-positive neurogenic cells along a pseudotime trajectory (Fig. 5E, Suppl. Table 3). This analysis identified an unbranched trajectory. While high Sstr2activated (wild type and heterozygous retinal, and 500 nM-treated explant) cells were distributed equally along this trajectory, low Sstr2-activated (Sstr2 −/− retinal, and 0, 5, and 50 nM-treated explant) cells were significantly comparatively enriched at lower pseudotime values (one-sided KS test, p value = 0.0054). None of the samples combined in the low or high groups had significantly different pseudotime distributions from each other as measured by a two-sided KS test. These results led us to conclude that Sstr2-related perturbations created some shared differences across the two systems.
We next tested for genes that changed in expression across our pseudotime trajectory to see if we could resolve which genes are impacted most by Sstr2 signaling. We identified 331 genes, 170 of which increased with pseudotime and Sstr2 signaling, and 161 of which decreased by likelihood ratio test (q-value < 0.05, Fig. 5F, Suppl. Tables 4, 5). The online gene set enrichment tool DAVID 42 showed a strong enrichment for cell-cycle genes among the genes that increased with pseudotime, including Cdc20, Cenpf, and Mki67. Genes that regulate neuronal development and differentiation showed decreased expression with pseudotime, including Neurod4, Nnat, and Sox9. It appears that the pseudotime trajectory captures the transition from proliferative neurogenic RPCs (high pseudotime values) to differentiating post-mitotic neural precursors (low pseudotime values) and www.nature.com/scientificreports/ that genotypes and conditions with no or low Sstr2 signaling lead to increased cell cycle exit and neurogenesis.
In combination with our results obtained from retinal explants, where increasing Sstr2 activation reduced photoreceptor generation, we conclude that Sstr2 signaling modestly inhibits retinal neurogenesis in retinal explants. Because of the relatively modest impact of Sst/Sstr2 signaling on retinal neurogenesis, we hypothesize that additional signals may be released by RGCs to modulate retinal neurogenesis. To investigate this possibility further, we again analyzed the publicly available scRNA-Seq mouse retinal development dataset. Mouse RGCs express the neuropeptide Pacap and signaling molecule Sonic Hedgehog (Shh) during the same interval that Sst and Sstr2 are expressed (Fig. S7A,C,E). When examining the full course of mouse retinal development, neither the Shh receptor Ptch1 nor the Pacap receptor Vipr2 are specifically expressed in neurogenic RPCs, though they are specifically expressed in late-stage primary RPCs and photoreceptor precursors, respectively (Fig. S7A). Additionally, both genes are coexpressed with Sstr2 in neurogenic RPCs (Fig. S7D,F). When examining the full course of human retinal development, both Pacap, and Shh are more strongly expressed in Müller glia than RGCs (Fig. S7B), although both are co-expressed with Sst in RGCs (Fig. S7C,E). Both Ptch1 and Vipr2 are co-expressed in neurogenic RPCs and late-stage primary RPCs in developing human retina, while Vipr2 is also expressed in early-stage RPCs and RGCs (Fig. S7B). As in mouse, both Ptch1 and Vipr2 are coexpressed with Sstr2 in neurogenic RPCs (Fig. S7D,F). Pacap and Shh may therefore act in combination with Sst as potentially redundant signals that modulate retinal neurogenesis.

Discussion
We identified four neuropeptides that are transiently, strongly, and specifically expressed during mouse retinal development by analyzing scRNA-seq data obtained from developing mouse retina. We tested whether they regulate retinal development using a combination of small-molecule ligands and qRT-PCR analysis of major retinal cell type markers in embryonic retinal explants. Agonists and antagonists of these neuropeptides-Npy, Gal, and Penk-did not produce significant dose-dependent changes in any retinal cell type markers tested. These results agree with Brodie-Kommit et al., who found no effect on RGC density in Gal-deficient mice 39 . However, agonist-dependent activation of the Sst receptor Sstr2 inhibited photoreceptor generation and led to a compensatory increase in the proportion of primary progenitors. The specificity of these effects was confirmed by their loss in Sstr2-deficient retinal explants. However, while Sstr2 signaling exerts a modest inhibitory effect on photoreceptor specification in retinal explants, this was not seen in vivo, with relative numbers of major retinal cell types unchanged in the Sstr2 mutant relative to wild type. This could reflect differences in the signaling mechanisms downstream of Sstr2 in explants compared to intact retinas that result from stresses associated with explant culture. ScRNA-Seq analysis revealed that retinal explants treated with Sstr2 agonists showed changes  www.nature.com/scientificreports/ phenotype seen in our embryonic retinal explants where treatment with an Sstr2-specific agonist resulted in reduced photoreceptor proportion and an increase in the number of primary progenitors. Wang et al. also found that ablation of Shh signaling resulted in a decrease of Ccnd1 and Hes1 gene expression, while RPC-enriched genes such as Hes2 and Ccnb1 and other cell cycle genes were also decreased upon disruption of Sstr2 signaling. Though we did not observe changes in cell composition in Sstr2 mutant retinas in vivo, pseudotime analysis revealed that neurogenic progenitors in these mutants were relatively more mature than controls, while neurogenic progenitors were relatively less mature in explants treated with Sstr2 agonists. Sstr2 + neurogenic cells in both intact Sstr2 −/− mutant retina and in explants treated with no or low levels of Sstr2 agonist showed increased expression of genes associated with terminal differentiation relative to wildtype, Sstr2 +/− , or Sstr2-expressing neurogenic cells in retinal explants that were treated with high levels of Sstr2 agonist, which in turn showed higher expression of genes enriched in retinal progenitors. Sstr2 signaling may therefore work in tandem with Shh signaling to prevent maturation and differentiation of neurogenic progenitor cells. A similar dependence on context for the impact of signaling on retinal cell type proportions was reported by Pearson and colleagues 44 . They found that an agonist for the ATP receptor P2RY2 only impacted chick retinal explant development in the absence of the RPE, the endogenous source for ATP in chick retinogenesis. The retinal explants used in the current study lack viable RGCs -the endogenous source of secreted Sst and Shh at this age. We suggest that in the developing retina, RGCs release multiple redundant signals to inhibit cell cycle exit and differentiation in neurogenic progenitors. When retinas are grown as explants, all of the signals are lost, allowing a single exogenous agonist to influence the fate of retinal progenitors. This would explain why Sstr2-mediated signaling regulates cell type proportions ex vivo but not in vivo.
Interestingly, the proportion of neurogenic progenitors remained consistent across explants at each concentration of Sstr2 agonist, even though fewer cells underwent neurogenesis in explants treated with higher concentrations of agonist. The transition of primary progenitor cells to neurogenic progenitors may therefore be inhibited by the presence of a high concentration of neurogenic cells. If fewer neurogenic cells are exiting mitosis to generate neurons, fewer primary progenitors are entering a neurogenic state. Cell-cell signaling via diffusible factors would be a possible mechanism of action for control of the number of neurogenic progenitors.
We propose that the discovery of an ex vivo phenotype that is not maintained in vivo and the expression of multiple neuropeptides in RGCs and their receptors in neurogenic progenitor cells may point to the existence of a complex, redundant system of cell-cell signaling that modulates retinal neurogenesis. Future research could interrogate this redundant system to find a role for cell-cell signaling in development in vivo through knockout of multiple neuropeptides or receptors. Loss of function of Vipr2 in combination with Sstr2 might phenocopy changes in cell-type proportions similar to what is seen with perturbation of Sstr2 signaling ex vivo and is seen in vivo with loss of Shh. We would expect that specifically the cell types produced during the interval in which Sstr2 and Vipr2 are expressed would be increased in proportion relative to later-born cell types. This study demonstrates that though in vivo experiments are the gold standard for biological relevance, the lack of replication of ex vivo results in a more native context does not necessarily equate to a null result.

Methods
Retinal development scRNA-Seq analysis. Sequences and metadata for the mouse (accession number GSE118614) and human (GSE138002) datasets were downloaded from the Gene Expression Omnibus and used to create new Seurat objects following the Seurat SCT workflow 45,46 . Cell type classifications and ages were taken from the downloaded metadata, and used to query the expression of various neuropeptides and receptors. Correlation of neuropeptide expression and each cell type was calculated using the base R function 'cor()' . Heatmaps were generated with R package pheatmap 47 . Mice. The use of animals for these studies was conducted using protocols approved by the Johns Hopkins Animal Care and Use Committee, in compliance with ARRIVE guidelines, and were performed in accordance with relevant guidelines and regulations. Timed pregnant mice were ordered from Charles River Laboratories for the smFISH experiments and embryonic retinal explant pharmacologic screen. A breeding pair of Sstr2 +/− (C57BL/6NCrl-Sstr2 em1(IMPC)Mbp /Mmucd) mice were ordered from the Mutant Mouse Resource and Research Centers (Davis, CA) and outbred to CD1 mice. The Sstr2 +/− F1 progeny were bred to produce the Sstr2 +/+ , Sstr2 +/− , and Sstr2 −/− mice used in this study. Similarly, two Sst +/− male mice 48 were received from Malcolm Low at the University of Michigan and outbred to CD1 mice. The Sst +/− F1 progeny were then bred to produce the Sst +/+ , Sst +/− , and Sst −/− mice used in this study.
Retinal sectioning. The explants and retinas used for immunostaining or smFISH were fixed in 4% paraformaldehyde in PBS rocking at room temperature for 2 h before being transferred to 30% sucrose in PBS at 4 degrees Celsius overnight. The samples were then transferred to VWR Clear Frozen Section Compound, frozen on dry ice, and stored at − 80 °C. 16 μm sections were then cut using a Leica 3050S cryostat.
smFISH. Retinas were dissected from embryonic day (E)14 (age verified by crown-rump length) or postnatal day (P)0 mice from timed pregnant mothers. Sections were hybridized with RNAscope probes for Sstr2 and Neu-rog2 and DAPI following the RNAscope protocol. The sections were imaged on an LSM 700 confocal microscope at 20 × and 63 × resolution.
Retinal explant culture. Retinas  www.nature.com/scientificreports/ cin in 12-well plates. The explants were incubated at 37 degrees Celsius, 5% CO2 for 6 days for E14 explants, and 4 days for E18 explants. On even-numbered days, 90% of the explant culture media was replaced with new media. Treatment compounds were included in the explant culture media at the concentrations listed in the text and ordered from Tocris Bioscience.
Explant RNA extraction and cDNA preparation. Embryonic retinal explants were removed from the nuclepore membrane and media and quick-frozen at − 80 °C or on dry ice. Total RNA was extracted from explants using the Qiagen miRNeasy Mini Kit with optional DNase digest and drying steps. RNA was extracted from 2 replicates from each condition in each of 3 rounds of experimental replicates, with a total of 6 retinas examined for each condition. The concentration of the resultant RNA was determined using a ThermoScientific NanoDrop 1000. 400 ng RNA was reversed transcribed to cDNA using the Invitrogen SuperScript IV First-Strand Synthesis System. The cDNA was diluted to 1 ng/ul in nuclease-free water.
Explant qPCR analysis. We used the bimake.com 2 × SYBR Green qPCR Master Mix. Four ng cDNA was used for each 20 ul reaction. Three technical replicates were assayed for each sample. Samples were run on a standard 2-step amplification program on Applied Biosystems MicroAmp Fast Optical 96-Well or Micro-Amp Optical 384-Well reaction plates on an Applied Biosystems StepOne Plus or ViiA 7, respectively. Primer sequences used either had been previously validated in the Blackshaw lab or were taken from the Harvard Primer Bank or generated using Primer3 40,49 . Those generated on Primer3 were required to span an exon-exon junction. All primers were verified to produce a single reaction product with traditional PCR before use for qPCR. All plots generated with ggplot2 R package 50 .

MULTI-seq.
Samples were dissociated using the Worthington papain cell dissociation system. MULTI-seq was performed to pool samples prior to 10 × Chromium Single Cell 3' v3.0 GEM generation and library prep following the Gartner lab's protocol using lipid-tagged oligos provided by the Gartner lab. Quality control and sequencing were performed by the Johns Hopkins Transcriptomics and Deep Sequencing Core. scRNAseq analysis. Sequencing results were processed through the 10 × CellRanger pipeline and the output loaded into Seurat 45 . Sample of origin for each cell was determined using the deMULTIplex R package out of the Gartner lab 40 . Seurat objects were analyzed following the Seurat SCT workflow. Cell type was determined cluster by cluster based on marker gene expression derived from the Seurat function 'FindAllMarkers' using default parameters 45 . This function was also used to identify differential gene expression across conditions/genotypes. Alternatively, and where noted in Results, the Seurat function 'FindMarkers' was used to find differential expression between two groups of cells. Seurat objects were integrated following the Seurat Integration and Label Transfer workflow. The pseudotime trajectory was constructed and differential genes identified using the semi-supervised Monocle 2 workflow 41 . Density plots generated using ggplot2. Differential genes were clustered using the base R function 'k-means' , and heatmap was generated using pheatmap.
Whole retinas were dissected from adult mice, flattened using a series of radial cuts, and imaged by camera next to a ruler to determine total retinal area. Area of retinas averaged for each animal.
Cell counting was performed blinded, with images randomized, using the Cell Counter plugin on ImageJ. Cells were counted in 2 200 μm long samplings and averaged for each technical replicate. Image metrics such as retinal thickness or total area were also quantified on ImageJ. www.nature.com/scientificreports/