A recent comparative transcriptomic study of Müller glia (MG) in vertebrate retinas revealed that fatty acid binding proteins (FABPs) are among the most highly expressed genes in chick ( Hoang et al., 2020). Here, we investigate how FABPs and fatty acid synthase (FASN) influence glial cells in the chick retina. During development, FABP7 is highly expressed by retinal progenitor cells and maturing MG, whereas FABP5 is upregulated in maturing MG. PMP2 (FABP8) is expressed by oligodendrocytes and FABP5 is expressed by non-astrocytic inner retinal glial cells, and both of these FABPs are upregulated by activated MG. In addition to suppressing the formation of Müller glia-derived progenitor cells (MGPCs), we find that FABP-inhibition suppresses the proliferation of microglia. FABP-inhibition induces distinct changes in single cell transcriptomic profiles, indicating transitions of MG from resting to reactive states and suppressed MGPC formation, with upregulation of gene modules for gliogenesis and decreases in neurogenesis. FASN-inhibition increases the proliferation of microglia and suppresses the formation of MGPCs. We conclude that fatty acid metabolism and cell signaling involving fatty acids are important in regulating the reactivity and dedifferentiation of MG, and the proliferation of microglia and MGPCs.
The process of retinal regeneration varies greatly across vertebrate species. In the fish, retinal regeneration is a robust process that restores functional cells and visual acuity following injury, whereas this process is far less robust in birds and is absent in mammals (Hitchcock and Raymond, 1992; Karl et al., 2008; Raymond, 1991). Müller glia (MG) have been identified as the cell of origin for progenitors in the retina (Bernardos et al., 2007; Fausett and Goldman, 2006; Fausett et al., 2008; Fischer and Reh, 2001; Ooto et al., 2004). In normal retinas, MG are the predominant type of support cell that provide structural, metabolic, visual cycle and synaptic support (Reichenbach and Bringmann, 2013). In response to damage, growth factors or drug treatments, MG can be stimulated to become reactive, de-differentiate, upregulate progenitor-related genes, proliferate and produce progeny that differentiate as neurons (Fischer and Bongini, 2010; Gallina et al., 2014; Wan and Goldman, 2016).
In mammalian MG, significant stimulation, such as forced expression of Ascl1, inhibition of histone deacetylases and neuronal damage, are required to reprogram MG into progenitor-like cells that produce a few neurons (Jorstad et al., 2017; Pollak et al., 2013; Ueki et al., 2015). Alternatively, deletion of Nfia, Nfib and Nfix in mature MG combined with retinal damage and treatment with insulin+FGF2 results in reprogramming of MG into cells that resemble inner retinal neurons (Hoang et al., 2020). Blockade of Hippo-signaling via forced expression of degradation-resistant YAP1 drives the proliferation of mature MG in the mouse retina (Hamon et al., 2019; Rueda et al., 2019), but it remains unknown whether any of the progeny differentiate as neurons. In addition, viral delivery of reporters β-catenin, Otx2, Crx and Nrl may reprogram MG into photoreceptors (Yao et al., 2018), but there are concerns that the viral vectors and mini-promoters used in these studies are prone to leaky expression in neurons (Blackshaw and Sanes, 2021). In the chick retina, MG readily reprogram into progenitor-like cells that proliferate, but the progeny have a limited capacity to differentiate as neurons (Fischer and Reh, 2001, 2003). Understanding the mechanisms that regulate the formation of Müller glia-derived progenitor cells (MGPCs) and neuronal differentiation of progeny is important to harnessing the regenerative potential of MG in mammals.
Fatty acid synthesis, metabolism and signaling are likely to be key components of regulating glial responses to damage and the formation of MGPCs. Fatty acid binding proteins (FABPs) are cytosolic lipid-binding proteins that mediate fatty acid metabolism and cell-signaling, and have highly conserved primary and tertiary structures across species, from Drosophila to humans (Hanhoff et al., 2002; Smathers and Petersen, 2011). FABPs are known to bind to polyunsaturated fatty acids including arachidonic acid and docosahexaenoic acid and have been shown to regulate signal transduction, neurotransmission, proliferation, differentiation and cell migration (Allen et al., 2007; Dawson and Xia, 2012; Tripathi et al., 2017; Yamashima, 2012). Very little is known about the cellular mechanisms and patterns of expression of FABPs in the retina. In mammals, FABP3, 5 and 7 have been identified in the brain, retina and radial glia, with demonstrated roles in differentiation and cell fate determination (Owada, 2008; Sellner, 1993; Sellner et al., 1995; Allen et al., 2007; Dawson and Xia, 2012; Tripathi et al., 2017; Yamashima, 2012). Further evidence indicates that FABPs in the CNS modulate peroxisome proliferator-activated receptor (PPAR), NF-kB and CREB signaling (Bogdan et al., 2018; Peng et al., 2017; Tripathi et al., 2017; Yamashima, 2012). NF-kB has been implicated as a key signaling ‘hub’ that suppresses the formation of MGPCs in chicks and mice, but not zebrafish (Hoang et al., 2020; Palazzo et al., 2020).
In a recent cross-species transcriptomic and epigenomic study, we identified FABP5 and PMP2 (also known as FABP8) among the most highly upregulated genes in MG in damaged chick retinas (Hoang et al., 2020). Further, we found that inhibition of FABPs potently suppresses the formation of proliferating MGPCs in damaged retinas (Hoang et al., 2020). However, few details are known about the mechanisms by which FABPs act to influence the formation of MGPCs, including how FABP activity influences cell-signaling pathways, reactivity and proliferation of microglia, and downstream changes in gene expression. Accordingly, this study investigates how FABPs and fatty acid synthase (FASN) influence the formation of MGPCs and analyzes single cell transcriptomic changes downstream of FABP inhibition in damaged and growth factor-treated retinas in the chick model system.
Expression of FABPs during embryonic retinal development
Single cell RNA-sequencing (scRNA-seq) libraries were established for retinal cells at embryonic day (E) 5, E8, E12 and E15. These libraries yielded 22,698 cells after filtering to exclude doublets, cells with low unique molecular identifier/genes per cell, and high mitochondrial gene content. Uniform manifold approximation and projection (UMAP) plots of aggregate libraries revealed clusters of cells that correlated to developmental stage and cell type (Fig. 1A). Embryonic retinal progenitor cells (eRPCs) from E5 and E8 retinas were identified by expression of ASCL1, CDK1 and TOP2A (Fig. 1B). Maturing MG were identified by expression of GLUL, RLBP1 and SLC1A3 (Fig. 1C). FABP5 and FABP7 were expressed by different types of developing cells at different stages of development, whereas PMP2 was not expressed (Fig. 1D). FABP5 was expressed by maturing bipolar and amacrine cells, whereas FAPB7 was expressed by eRPCs from E5 and E8 and at elevated levels in immature MG at E8 (Fig. 1D,E). Levels of FABP7 decreased in maturing MG at E12 and E15 (Fig. 1D,E). FABP7 was also expressed by maturing bipolar and amacrine cells from E8 retinas (Fig. 1D,F). Immunolabeling for PMP2 confirmed findings from scRNA-seq. PMP2 was not expressed in the developing retina until E16, when PMP2-immunofluorescence was detected in cells that resembled oligodendrocytes near the vitreal surface of the retina (Fig. 1G). These findings indicate that FABP5 and FABP7 are dynamically and highly expressed in eRPCs and MG through the course of embryonic development.
Expression of FABPs in damaged chick retina
We sought to provide a detailed description of expression patterns of FABPs in the retinas of normal and damaged hatched chicks. scRNA-seq libraries were aggregated for cells obtained from control and N-methyl-D-aspartate (NMDA)-damaged retinas at different time points (24, 48 and 72 h) after treatment for a total of 57,230 cells (Fig. 2A). We have previously used these chick scRNA-seq databases to compare MG and MGPCs across fish, chick and mouse (Hoang et al., 2020) and characterize expression patterns of genes related to NFkB, midkine, matrix metalloproteases and endocannabinoids (Campbell et al., 2019, 2021a, 2021b). UMAP-clustered cells were identified based on well-established patterns of expression (Fig. 2A,B). Resting MG formed a discrete cluster of cells and expressed high levels of GLUL, RLBP1 and SLC1A3 (Fig. 2C,D). After damage, MG downregulate markers of mature glia as they transition into reactive glial cells and into progenitor-like cells that upregulate TOP2A, CDK1, SPC25 and PCNA (Fig. 2C,D). FABP5 and PMP2 were expressed at low levels in relatively few resting MG from undamaged retinas, whereas FABP7 was widely expressed by the majority of resting MG (Fig. 2E,F). FABP7 and PMP2 were detected in oligodendrocytes and non-astrocytic inner retinal glia (NIRGs). NIRG cells are a distinct type of glial cell that has been described in the retinas of birds (Rompani and Cepko, 2010; Fischer et al., 2010) and some types of reptiles (Todd et al., 2016b). Following NMDA-induced retinal damage, levels of FABP5, FABP7 and PMP2 were significantly increased in activated MG at 24 h (Fig. 2E,F). Levels of FABP5 and PMP2 were significantly reduced in activated MG at 48 h and 72 h, but remained elevated in proliferating MGPCs (Fig. 2E,F).
Immunolabeling for PMP2
To validate some of the scRNA-seq data, we characterized PMP2 immunolabeling in normal and NMDA-damaged retinas. In undamaged retinas, PMP2 was observed in NIRG cells, but only in peripheral regions of retina (Fig. 2G). Although the NIRG cells were often observed in the proximal inner nuclear layer (INL) and had morphologies reminiscent of amacrine cells; these cells were negative for amacrine cell markers including AP2α, islet1 and tyrosine hydroxylase (Fig. 2G). PMP2-immunolabeling was prevalent in oligodendrocytes that extended processes into the nerve fiber layer (NFL). The PMP2-positive oligodendrocytes were negative for glutamine synthase (GS; Fig. 2H), but positive for Olig2 (Fig. 2I) and Sox10 (Fig. 2M). A few PMP2-positive oligodendrocytes were labeled for EdU in central and peripheral regions of the damaged retinas (Fig. 2J-L). We did not observe PMP2 in NIRG cells [Sox2+/Nkx2.2+ cells in the inner plexiform layer (IPL)] in central regions of control or NMDA-damaged retinas (Fig. 2N,O), but the labeling may have been obscured by labeling in MG processes. Collectively, these patterns of immunolabeling are consistent with scRNA-seq data for oligodendrocytes and NIRG cells and patterns of expression for PMP2, SOX10 and OLIG2 (Fig. 2P). Further, NIRG cells expressed FABP7, and upregulated FABP5, FABP7 and PMP2 after damage (Fig. S1A). Oligodendrocytes expressed PMP2 alone (Fig. 2P), but upregulated FABP5 and FABP7 after damage (Fig. S1B).
FABPs in retinas treated with insulin and FGF2
We next examined FABP expression in MGPCs in the absence of neuronal damage. In the chick retina the formation of MGPCs can be induced by consecutive daily injections of fibroblast growth factor 2 (FGF2) and insulin in the absence of neuronal damage (Fischer et al., 2002, 2009b, 2014; Ritchey et al., 2012). Eyes were treated with two or three consecutive daily doses of FGF2 and insulin, and scRNA-seq retinal libraries were generated. UMAP ordering of cells revealed distinct clusters that were segregated based on cell type (Fig. 3A). MG glia were identified by expression of VIM, GLUL and SLC1A3, and MGPCs were identified by expression of TOP2A, nestin, CCNB2 and CDK1 (Fig. 3B). Resting MG from saline-treated retinas formed a cluster distinct from MG treated with either two or three doses of FGF2 and insulin (Fig. 3B). Similar to patterns of expression seen in NMDA-damaged retinas, FABP5, FABP7 and PMP2 were significantly increased in MG treated with insulin and FGF2 (Fig. 3C,D).
To test whether PMP2 was increased when MG were treated with FGF2 alone we immunolabeled retinal sections. Four consecutive daily intraocular injections of FGF2 are known to be sufficient to induce the formation of proliferating MGPCs in the absence of damage (Fischer et al., 2014). We found that PMP2-immunoreactivity was increased in FGF2-treated MG (Fig. 3E). This effect was not evident in central regions of the retina, but was apparent in peripheral regions (Fig. 3E). Some of the PMP2+/Sox2+ MG were labeled for EdU (Fig. 3F), indicating that these cells were proliferating MGPCs. We next investigated whether FABP-inhibition influenced the formation of MGPCs in the absence of neuronal damage. Accordingly, we applied BMS309403 (hereafter referred to as BMS) with FGF2 alone and probed for the formation of proliferating MGPCs. BMS is a pan-FABP inhibitor with demonstrated inhibitory activity for FABP3, FABP4 and FABP5 (Sulsky et al., 2007). BMS interacts with the fatty-acid-binding pocket within the interior of the protein and competitively inhibits the binding of endogenous fatty acids, and is expected to inhibit all FABPs, including FABP7 and PMP2 (reviewed by Furuhashi and Hotamisligil, 2008). Inhibition of FABPs significantly reduced numbers of Sox2/EdU-labeled MGPCs in the INL of FGF2-treated retinas (Fig. 3G,H). There were no dying cells detected in retinas treated with FGF2±BMS (not shown). Application of BMS as three consecutive daily intraocular injections had no effect upon cell death (TUNEL+ cells), proliferation of microglia or proliferation of MGPCs (not shown).
FABPs in eRPC and MGPCs
We next compared levels of FABPs between different types of retinal progenitor cells. We aggregated and normalized scRNA-seq libraries for progenitors from E5 and E8 retinas, and MGPCs from retinas treated with NMDA and/or insulin and FGF2 (Fig. 4A). UMAP-ordering of cells revealed two distinct clusters of cells for eRPCs and MGPCs (Fig. 4B), with both clusters expressing high levels of proliferation-associated genes; PCNA, SPC25, TOP2A and CDK1 (Fig. 4B,C). MGPCs expressed FABP5 and PMP2 with significantly higher levels in MGPCs from retinas treated with insulin and FGF2, whereas eRPCs did not express FABP5 or PMP2 (Fig. 4D-G). By contrast, levels of FABP7 were higher in proliferating eRPCs than in proliferating MGPCs, with significantly lower levels in MGPCs from retinas treated with insulin and FGF2 (Fig. 4G).
Genes regulated by FABP-inhibition
To identify transcriptional changes downstream of FABPs we generated scRNA-seq libraries for control retinas and retinas treated with inhibitor±NMDA. Aggregation of the scRNA-seq libraries revealed distinct UMAP clusters of neurons and glia (Fig. 5A,B). UMAP-ordering did not result in distinct clustering of neurons based on treatment, whereas MG were differentially clustered according to damage and BMS treatment (Fig. 5A,B). Activated MG upregulated HBEGF, TUBB6 and TGFB2 (Fig. 5C,D), MGPCs upregulated CDK1, TOP2A and ESPL1 (Fig. 5E), and resting MG expressed high levels of GLUL, RLBP1 and SLC13A, which were downregulated in activated MG (Fig. 5F). We identified differentially expressed genes (DEGs) in MG from retinas treated with BMS, NMDA and BMS+NMDA (Fig. 5C; Tables S1, S2, S3). BMS treatment of undamaged retinas resulted in downregulation of 393 genes, including markers for resting mature glia such as GLUL, CRABP-I, RLBP1, CA2, SFRP1 and ID4 (Fig. 5G; Table S1). By comparison, BMS treatment resulted in upregulation of 495 genes including FABPs, secreted factors associated with glial activation and receptors for TNF-related ligands (Fig. 5G; Table S1). Gene ontology (GO) enrichment analysis of DEGs from saline-BMS-treated MG revealed significant enrichment for upregulated gene modules associated with gliogenesis, wound healing and developmental processes (Fig. 5I), whereas downregulated gene modules were associated with neurogenesis and cell proliferation (Fig. 5I). BMS treatment of NMDA-damaged retinas resulted in significant downregulation of 192 genes including markers for proliferation, pro-glial genes NFIX and ID4, and High Mobility Group (HMG) genes (Fig. 5H; Table S2). By comparison, BMS treatment resulted in upregulation of 114 genes including FABP5 and FABP7, secreted factors BMP4, WNT4 and WNT6, and glial reactivity genes such as CD44 (Fig. 5H; Table S2). GO enrichment analysis for NMDA/BMS-treated MG revealed significant enrichment for gene modules associated with proliferation and organelle fission (Fig. 5J). By comparison, GO enrichment analysis for NMDA/BMS-treated MG revealed significant enrichment for gene modules associated with growth factor signaling, and cell motility (Fig. 5J).
We isolated MG from four treatment groups and re-ordered these cells in a UMAP plot. The MG formed five distinct clusters; two clusters of resting MG (occupied predominantly by cells from retinas treated with saline and BMS-saline), two clusters of activated glia (occupied predominantly by cells from retinas treated with NMDA, BMS-NMDA and BMS-saline) and MGPCs (∼73% occupied by cells from retinas treated with NMDA alone) (Fig. S2A-C). This finding suggest that BMS treatment induces MG to acquire a reactive phenotype. Genes for mature resting MG were significantly downregulated in activated MG and MGPCs, activated MG significantly upregulated genes associated with reactivity, and MGPCs had elevated levels of proliferation genes (Fig. S2D-G). Consistent with findings that BMS treatment resulted in upregulation of TNF receptors in MG in saline and NMDA-damaged retinas (Fig. 5G), we found increased levels and percentage of cells expressing genes associated with NFkB-signaling including NFKBIA, NFKBIB (NKIRAS2), NFKB1, CHUK and IKBKB in BMS-treated MG (Fig. S2H). Consistent with the notion that BMS is not toxic to MG, GO analysis indicated that there was no upregulation of gene modules associated with cell death in MG treated with BMS.
We next isolated the MG, microglia and NIRG cells, re-embedded these cells in UMAP plots and probed for cell signaling networks and putative ligand-receptor (LR) interactions using SingleCellSignalR (Cabello-Aguilar et al., 2020). We focused our analyses on MG, microglia and NIRG cells because evidence supports autocrine and paracrine signaling among these glia during activation and the formation of proliferating MGPCs (Fischer et al., 2014; Wan et al., 2012; White et al., 2017; Zelinka et al., 2012). Resting MG included cells from saline- and BMS-treatment groups, activated MG included cells mostly from BMS-saline-, NMDA- and BMS-NMDA-treatment groups, and MGPCs were predominantly derived from the NMDA-treatment group (Fig. 6A-C). Numbers of LR interactions (significant upregulation of ligand and receptor) between cell types in the different treatment groups varied between 70 and 315 (Fig. 6D-G). We performed analyses on glial cells from each treatment group and compared changes across the most significant LR interactions. For example, LR interactions included IGF1R and FGFR1 in activated MG to MGPCs, but not when treated with BMS (Fig. 6F,G). By comparison, BMS-treated MG included LR interactions with TGFBR2 and interactions involving TNFRSF11B, MDK, PTN and PTPRZ1 (Fig. 6F,G). We compared significant changes in LR interactions among glial cells and interactions unique to treatment groups for undamaged and damaged retinas. We identified 33 LR interactions specific to saline-treated glia and 148 LR interactions specific to BMS-treated glia in undamaged retinas (Fig. 6H,I,L,M,P). Glia in undamaged saline-treated retinas included unique LR interactions for FGF9-FGFR3/4, BMP2-ACVR1 and BMP6-BMPR2 (Fig. 6H,I,L,M,P). By comparison, glia in undamaged BMS-treated retinas included unique LR interactions associated with activated glial phenotypes such as IL1B-IL1RAP, TGFB1/2-TGFBR2, HBEGF-CD9/CD82/ERBB2/EGFR and JAG1/JAG2/PSEN1-NOTCH2 (Fig. 6H,I,L,M,P). We identified 40 LR interactions specific to saline/NMDA-treated glia and 86 LR interactions specific to BMS/NMDA-treated glia in damaged retinas (Fig. 6J,K,N,O,Q). LR interactions unique to glia in NMDA-damaged retinas included BMP, MDK, FGF and DLL1-NOTCH1 signaling (Fig. 6J,K,N,O,Q). By comparison, LR interactions unique to glia in BMS/NMDA-damaged retinas included JAG1/2/PSEN1-NOTCH2, INHBA-ACVR2, TGFB1-ITGB3 and WNT5A-LRP2/FZD4 (Fig. 6J,K,N,O,Q).
Inhibition of FABPs in resting MG
We next investigated and validated changes in cell signaling in damaged retinas treated with FABP-inhibitor. One day after treatment with NMDA±BMS there was a significant increase in the number of dying TUNEL+ cells (Fig. 7A,B). NMDA treatment of the chick retina is known to result in the death of only amacrine and bipolar neurons (Fischer et al., 1998). By contrast, we observed a significant decrease in pS6 in MG in damaged retinas treated with BMS (Fig. 7A,B), consistent with findings from SingleCellSignalR where LR interactions for FGF1-FGFR1 and MDK-ITGA4 are missing with BMS treatment (Fig. 6Q). MDK and FGF are known to active mTor-signaling and upregulated pS6 in MG in the chick retina (Campbell et al., 2021a; Zelinka et al., 2016). We also observed reduced levels of pStat3 in MG nuclei in damaged retinas treated with BMS (Fig. 7A,B). Stat phosphorylation is known to be downstream of PDGF-signaling (Li et al., 2012; Popielarczyk et al., 2019) and PDGFA-PDGFRA LR interactions are missing from damaged retinas treated with BMS (Fig. 6Q). Similarly, we found reduced levels of pSMAD1/5/8 in MG nuclei in damaged retinas treated with BMS (Fig. 7A,B). This may result from increased signaling through ACVR2A and TGFB1/ITGB3 which may antagonize BMP/SMAD-signaling (Todd et al., 2017). Signaling through mTor (pS6), Jak/Stat (pStat3) and BMP/SMAD (pSmad1/5/8) is known to be required for the formation of proliferating MGPCs (Todd et al., 2016a, 2017; Zelinka et al., 2016). At 24 h after treatment, numbers of Sox2+ MG nuclei in retinas treated with NMDA+BMS were not significantly different from numbers of MG in retinas treated with NMDA alone (NMDA – 23.6±3.4; NMDA+BMS – 21.4±2.4; P=0.27, n=5). This suggests that decreases in immunofluorescence for different cell signaling pathways did not result from BMS-induced death of MG.
scRNA-seq data indicated that BMS treatment of undamaged retina had a significant impact on the transcriptional profile of MG with profiles resembling de-differentiation and reactive glia. We verified some of the scRNA-seq data by labeling BMS-treated retinas with antibodies to vimentin or PMP2. BMS treatment significantly increased immunofluorescence for vimentin and PMP2 in MG (Fig. 7C,D). As BMS treatment appeared to stimulate MG to become reactive and de-differentiate, processes that occur during a transition to a progenitor-like phenotype (Hoang et al., 2020), we tested whether BMS treatment primes MG to become proliferating progenitors in damaged retinas. BMS treatment should have inhibited FABP7 in resting MG and this inhibition was predicted to enhance the ability of MG to become proliferating MGPCs. Contrary to expectations, we found that BMS treatment before NMDA-induced damage suppressed the formation of proliferating MGPCs (Fig. 7E,F).
Effects of FABP-inhibition in microglia
Retinal microglia and infiltrating macrophage are known to promote the formation of MGPCs in chick and zebrafish retinas (Fischer et al., 2014; Palazzo et al., 2020; White et al., 2017) and suppress the neuronal differentiation of reprogrammed MG in mouse retinas (Todd et al., 2020). Accordingly, we investigated whether microglia were influenced by treatment with FABP inhibitor. The application of BMS to NMDA-damaged retinas suppressed the accumulation of microglia and significantly reduced numbers of EdU+/CD45+ cells (Fig. 8A,B). The microglia in BMS-NMDA-treated retinas appeared to retain a reactive morphology (Fig. 8A). Thus, it is possible that reduced numbers of proliferating MGPCs resulted, in part, from reduced accumulation of reactive monocytes in BMS-treated retinas.
We UMAP-embedded microglia from retinas treated with saline, BMS and NMDA. Plots revealed distinct clusters of resting microglia, activated and proliferating cells (Fig. 8C). The resting microglia UMAP clusters were predominantly occupied by microglia from saline-treated retinas, whereas microglia from BMS-saline treated retinas were clustered among activated cells (Fig. 8C,D). BMS treatment of microglia in undamaged retinas resulted in upregulation of 545 genes, whereas no genes were significantly downregulated (Fig. 8E,G; Table S4). BMS treatment of undamaged retinas stimulated microglia to upregulate genes associated with proliferation, cell signaling, complement, integrins and glial transcription factors (Fig. 8E; Table S4). NMDA treatment stimulated microglia to significantly upregulate nearly 1400 genes including FABP5, FABP7, PMP2 and FASN, whereas only eight genes were downregulated (Fig. 8F,G; Table S5). BMS treatment of NMDA-damaged retinas resulted in changes in expression of only eight genes, and PMP2 and proliferation-associated genes were among the few DEGs in microglia in BMS/NMDA-damaged retinas (Fig. 8E,F). GO enrichment analysis of microglia from retinas treated with saline±BMS indicated gene modules associated with cellular biogenesis, regulation of cell death and hydrolytic/catabolic processes (Fig. 8H). Thus, it is not surprising that the BMS-treated microglia were embedded among microglia from NMDA-damaged retinas in UMAP plots (Fig. 8C,D). GO enrichment analysis of microglia from retinas treated with saline±NMDA indicated groups of genes associated with lytic enzyme activity, lysosomal activity and cellular biogenesis (Fig. 8I). There were only six DEGs in microglia from retinas treated with NMDA±BMS, consistent with the co-clustering of microglia from these treatments in UMAP plots (Fig. 8D). These microglia were harvested at 48 h after NMDA treatment and it is therefore likely that significant differences in gene expression that led to decreased accumulation of reactive microglia in BMS/NMDA-treated retinas occurred shortly after damage and BMS treatment was not included. This is consistent with scRNA-seq findings in NMDA-damaged mouse retinas wherein microglia significantly upregulated genes for pro-inflammatory cytokines between 3 and 12 h after damage (Todd et al., 2019).
FASN influences MGPCs, neuronal survival and the accumulation of reactive microglia
FASN-dependent fatty acid synthesis is necessary for FABP activity. scRNA-seq data indicated that FASN is widely expressed by most retinal cell types (Fig. 9A). In MG, levels of FASN were elevated in resting MG and downregulated in MG at 24, 48 and 72 h after NMDA treatment, and remained low and prevalent in MGPCs (Fig. 9A,B). We further assessed patterns of expression of FASN in aggregate libraries from time points soon after NMDA, at 3 and 12 h after treatment. Levels of FASN were significantly upregulated in MG at 3 h after NMDA and back down at 12 and 48 h after NMDA (Fig. 9C,D), suggesting a rapid and transient need for elevated fatty acid synthesis in MG shortly after NMDA treatment.
We targeted FASN activity with small molecule inhibitors G28UCM and C75. G28UCM is a polyphenol with demonstrated specificity as an FASN inhibitor (Puig et al., 2009). C75 is a stabilized synthetic derivative of cerulenin, an endogenous FASN inhibitor (Kuhajda et al., 2000). C75 inhibits FASN activity and may weakly activate carnitine palmitoyltransferase-1, the enzyme responsible for the regulation of mitochondrial fatty acid oxidation (Thupari et al., 2002). Treatment of NMDA-damaged retinas with G28UCM or C75 resulted in significantly fewer proliferating MGPCs (Fig. 9E-G). These large decreases in numbers of proliferating MGPCs occurred despite small, but significant, decreases in cell death (Fig. 9H-J); levels of retinal damage positively correlate with the number of proliferating MGPCs (Fischer et al., 2004). By comparison, FASN inhibitors significantly increased numbers of proliferating microglia in damaged retinas (Fig. 9K-P). Increases in numbers of proliferating microglia occurred without increasing total numbers of CD45+ cells (Fig. 9K,P). NIRG cells were not affected by FASN inhibitors (not shown). Three consecutive daily applications of G28UCM to normal retinas did not induce cell death (TUNEL+ cells; not shown).
We next sought to investigate whether cell signaling in damaged retinas was influenced by FASN-inhibitor. We observed a significant decrease in levels of pS6 in the cytoplasm of MG treated with FASN-inhibitor in damaged retinas (Fig. 9Q,R). Similarly, levels of pStat3 and pSmad1/5/8 were undetectable in MG treated with FASN-inhibitor in damaged retinas (n≥5) (Fig. 9S,T).
In this study we investigated the function of FASN and FABPs in glial cells in the chick retina. We observed that FABP7 is highly expressed by eRPCs and maturing MG during embryonic development. FABP7, FABP5 and PMP2 are upregulated during the activation of MG after damage or treatment with FGF2 and insulin. This pattern of FABP expression was also observed for NIRG cells and microglia in damaged retinas. Inhibition of FABPs or FASN influenced the proliferation of different types of cells including microglia and MGPCs. Inhibition of FABPs in undamaged retinas induced reactivity in MG, but also both decreased levels of genes associated with resting mature MG and neurogenesis and increased genes associated with gliogenesis and inflammation. Inhibition of FABPs and FASN in damaged retinas selectively suppressed cell signaling pathways in MG that are known to promote the formation of MGPCs. These findings indicate the importance of FASN and FABPs in mediating the transition of MG into a proliferating MGPCs (Fig. 10).
Different FABP isoforms are known to be expressed by different cell types in the maturing mammalian brain (Owada, 2008). Similarly, FABP isoforms have been identified in the chick retina (Sellner, 1993). FABP7 is often used as a marker for radial glia (also known as brain lipid binding protein, Blbp) in the developing mouse brain and is presumed to facilitate cortical development (Anthony et al., 2005; Feng et al., 1994). FABPs are expressed in different types of tumors, particularly during metastasis (Ohmachi et al., 2006; Senga et al., 2018). We found that FABP7 was upregulated in developing and maturing MG in embryonic chick retina. In addition, FABP7 was detected in developing amacrine and bipolar interneurons, but was downregulated in mature neurons. By comparison, FABP5 was expressed at high levels in mature interneurons. These cell-type specific patterns of expression for FABPs may indicate isoform-specific rolls despite overlap in ligand-binding among FABPs.
During embryonic development glial precursor cells migrate into the retina from the optic nerve (Rompani and Cepko, 2010). These precursor cells undergo a cell division to generate oligodendrocytes and diacytes (Rompani and Cepko, 2010), also known as NIRG cells, which reside predominantly in the IPL (Fischer et al., 2010). This unique type of glial cell has been identified in the retinas of birds and some types of reptiles (Todd et al., 2016b). The function of the NIRG cells is unknown, but these cells are known to proliferate in response to IGF1 (Fischer et al., 2010) and their survival is tethered to retinal microglia (Zelinka et al., 2012). We found that NIRG cells express PMP2, but only in peripheral regions of the retina or at later times (>7 days) after NMDA-induced damage. By comparison, inhibition of FASN had no effect upon the proliferation of NIRG cells.
Consistent with previous reports (Kohsaka et al., 1980, 1983), the chicken retina contains axons that are thinly myelinated by oligodendrocytes that express Sox10, Olig2 and PMP2. There were rare PMP2+ cells in the inner INL of peripheral retinal regions that did not express neuronal markers, but expressed a set of markers associated with NIRG cells. One week after damage there was an increase in the number of EdU-labeled oligodendrocytes, which suggests de novo myelination of axons in the NFL from newly generated oligodendrocytes. Myelination of RGC axons in the chick retina is known to be relatively sparse and thin (Kohsaka et al., 1980; Nakazawa et al., 1993). The origin of newly generated oligodendrocytes in NMDA-damaged retinas remains uncertain, but could include quiescent oligodendrocyte precursors within the retina, or migration and recruitment of newly generated oligodendrocytes from precursors that reside in the optic nerve. However, we cannot exclude the possibility that the EdU-labeling of oligodendrocytes resulted from DNA replication without mitosis. Further studies are required to determine the origin of the EdU-labeled oligodendrocytes and whether the newly generated oligodendrocytes provide additional axon myelination. Notably, NMDA damage is not expected to result in demyelination, which raises questions about the signals that promote the proliferation of oligodendrocytes. The presence of oligodendrocyte precursor cells in mature chick retinas remains uncertain.
scRNA-seq data indicate that FABP7 is expressed by resting MG in the postnatal chick retina. When the retina is damaged or treated with FGF2 and insulin the MG robustly upregulate FABP5, FABP7 and PMP2 (Table 1). When FABPs are inhibited in damaged retinas, significantly fewer proliferating MGPCs are generated (Hoang et al., 2020). Similarly, MGPC proliferation was suppressed when inhibitor was applied before damage when FABP7 is highly expressed in MG; although FABP5 and PMP2 become elevated in response to inhibition, perhaps to compensate for diminished overall FABP activity. scRNA-seq data suggest that inhibition of FABPs pushes MG into an activated/reactive state, away from resting and progenitor-like states. It is thought that it is necessary for MG to become reactive before transitioning to progenitor-like state in fish and chick retinas (Hoang et al., 2020; Thomas et al., 2015). Our findings suggest that the rapid activation of FABPs and FASN for fatty acid metabolism is a first step in responding to stress and later transitioning to a progenitor-like state. The failure to proceed to a progenitor-like state in response to FABP/FASN inhibitors is supported by findings in which key cell signaling pathways required for the formation of MGPCs are suppressed. By comparison, our scRNA-seq databases (Hoang et al., 2020) indicate that the MG in NMDA-damaged mouse retinas upregulate Fabp5 and Pmp22, and downregulate Fasn, but do not express Fabp7 or Pmp2 (Fig. S3). The MG in NMDA-damaged zebrafish retinas upregulate fabp3 and fabp7a, and downregulate fasn, whereas fabp5 and pmp2 were not annotated (Fig. S3). Collectively, damage-induced changes in expression of FABPs and FASN suggest that fatty acid metabolism plays important roles in the responses of MG to damage and reprogramming into MGPCs in the retinas of chicks, and possibly also in the retinas of fish and mammals.
FABP and FASN activity may be upstream of important cell signaling pathways that drive the transition of MG to MGPCs. We observed suppressed cell signaling through BMP/SMAD, Jak/Stat and mTor in MG treated with FABP or FASN inhibitors. These findings are consistent with the loss of specific LR interactions in glia treated with FABP inhibitor, as revealed by SingleCellSignalR analyses. These cell signaling pathways are required for the formation of proliferating MGPCs in the chick retina (Todd et al., 2016a, 2017). Further, we found loss of LR interactions involving FGF, midkine and Notch1 in retinal glia treated with FABP inhibitor; pathways that are necessary for the formation of proliferating MGPCs (Fischer et al., 2009a,b; Ghai et al., 2010; Hayes et al., 2007). Consistent with our observations, Notch-signaling is known to be downstream of FABPs (Anthony et al., 2005).
FABP isoforms serve many different functions including cellular metabolism and cellular trafficking of lipid metabolites (Storch and Corsico, 2008). In addition to facilitating lipid metabolism, FABPs can also facilitate nuclear transport of hydrophobic ligands for cell signaling such as PPAR (Tripathi et al., 2017), retinoic acid (Dawson and Xia, 2012) and endocannabinoids (Haj-Dahmane et al., 2018). Given the well-established involvement of FABPs in lipid metabolism, we found associations with the expression of FASN, which is important for producing long-chain fatty acids (Kuhajda, 2006). When we antagonized FASN with different small molecule inhibitors there were significant decreases in numbers of proliferating MGPCs. These findings provide further evidence that lipid metabolism is required for the transition of resting MG to activated states, and subsequent proliferation as progenitor-like cells. scRNA-seq data indicate that MG acquire reactive transcriptomic profile with FABP inhibition in the absence of neuronal damage. This broad shift in the gene-expression modules to induce reactive phenotypes was supported by evidence for LR interactions associated with reactivity including responses to ligands such as IL1β, TGFβ and HB-EGF.
Although inhibition of FABPs in undamaged retinas stimulated MG to adopt a reactive phenotype and acquire a transcriptomic profile characteristic of activation and de-differentiation, FABP-inhibition before NMDA treatment did not prime MG to become MGPCs. This likely resulted from FABP-inhibition upregulating gene modules associated with gliogenesis and downregulation of gene modules associated with neurogenesis and proliferation. FABP inhibition in undamaged retinas should have inhibited FABP7 in resting MGs, as levels of FABP5 and PMP2 are very low in these glia, and this inhibition likely resulted in the altered transcriptomic profiles seen in MG and microglia.
After retinal damage in the chick, microglia and macrophage normally proliferate and acquire a reactive phenotype (Zelinka et al., 2012). The presence of reactive microglia/macrophage is required for the formation of MGPCs (Fischer et al., 2014). Further, signals from reactive microglia mediate inflammatory signaling in MG through pathways such as NFkB (Palazzo et al., 2020). Given that microglia rapidly respond to retinal damage with upregulation of pro-inflammatory signals (Todd et al., 2019), it is possible that reduced numbers of microglia in damaged retinas treated with FABP inhibitor influenced the formation of proliferating MGPCs. In damaged retinas treated with FABP inhibitor, microglia appeared to be highly reactive, but the numbers of these cells were significantly reduced. Further, we detected few microglial genes that were differentially expressed in damaged retinas treated with FABP inhibitor. We observed increased expression of proliferation-related genes in microglia in undamaged retinas treated with FABP inhibitor, whereas we observed decreased expression of proliferation-related genes in microglia in NMDA-damaged retinas treated with FABP-inhibitor. These findings suggest context-specific effects of FABP inhibition on the proliferation of microglia. It remains uncertain whether the effects of FABP inhibitors on microglia are direct, indirect via MG or a combination of these possibilities (Fig. 10). By comparison, in damaged retinas treated with FASN-inhibitors we found increased numbers of proliferating microglia without increases in total numbers of CD45+ microglia (Fig. 10). This finding suggests that there may have been less recruitment of peripheral monocytes coincident with increased proliferation of resident microglia to result in no net change in total numbers of CD45+ cells in the retina. The decrease in proliferating microglia may have resulted from decreased cell death in retinas treated with FASN inhibitors. Given the differences in patterns of expression and different functions of FABPs and FASN, it is not surprising that inhibition of FABPs and FASN had different effects on microglia. We cannot exclude the possibility that ‘suppressed’ microglia influenced the diminished formation of MGPCs in retinas treated with FASN and FABP inhibitors.
FASN and FABPs are novel targets of investigation with respect to retinal glia and reprogramming of MG into MGPCs. We found that FABPs are highly expressed by MG during reprogramming into proliferating MGPCs. Inhibition of FABPs results in the upregulation of genes associated with gliogenesis and inflammation while concurrently reducing the expression of genes associated with proliferation and neurogenesis. The anti-proliferative effects of FABP inhibition were not specific to MG, as microglia also showed reduced proliferation in inhibitor-treated retinas. Our findings suggest that FABPs mediate glial reactivity and de-differentiation through lipid-associated cell signaling, whereas proliferation requires lipid metabolism. Consistent with this hypothesis, inhibition of FASN potently inhibited the formation of proliferating MGPCs, while decreasing cell death and increasing microglial proliferation. Microglia express FABPs, and inhibition of FABPs alters cytokine production and reactivity, which is expected to secondarily impact signaling with MG. Collectively, our data suggest the activity of FASN and FABPs in MG is required for the cells to become activated before forming proliferating MGPCs in the chick retina.
MATERIALS AND METHODS
The use of animals in these experiments was in accordance with the guidelines established by the National Institutes of Health and the Institutional Animal Care and Use Committee at The Ohio State University, USA. Newly hatched P0 wild-type leghorn chicks (Gallus gallus domesticus) were obtained from Meyer Hatchery. Post-hatch chicks were maintained in a regular diurnal cycle of 12 h light, 12 h dark (08.00 h-20.00 h). Chicks were housed in stainless-steel brooders at 25°C and received water and Purina chick starter ad libitum. All experiments were completed prior to P24. Fertilized eggs were obtained from the Michigan State University, Department of Animal Science. Eggs were incubated at a constant 37.5°C, with a 1 h period at room temperature with a cool-down period every 24 h. The eggs were rocked every 45 min and held at a constant relative humidity of 45%. Embryos were harvested at various time points after incubation and staged according to guidelines established by Hamburger and Hamilton (1992).
Chicks were anesthetized with 2.5% isoflurane mixed with oxygen from a non-rebreathing vaporizer. The technical procedures for intraocular injections were performed as previously described (Fischer et al., 1998). With all injection paradigms, both pharmacological and vehicle treatments were administered to the right and left eye, respectively. Compounds were injected in 20 µl sterile saline or 30% DMSO in saline. Compounds included: NMDA (38.5 nmol or 154 µg/dose; Millipore Sigma), FGF2 (250 ng/dose; R&D Systems), BMS309403 (2.5 µg/dose; Tocris Biotechne), C75 (2.5 µg/dose; Tocris Biotechne), G28UCM (2.5 µg/dose; Tocris Biotechne). 5-Ethynyl-2′-deoxyuridine (1.3 µg/dose; EdU, Thermo Fisher Scientific) was injected into the vitreous chamber to label proliferating cells. Injection paradigms are included in each figure.
scRNA-seq of retinas
Retinas were obtained from embryonic and hatched chicks. Retinas were dissociated in a 0.25% papain solution in Hank's balanced salt solution (HBSS) (pH 7.4) for 30 min, and suspensions were frequently triturated. The dissociated cells were passed through a sterile 70 µm filter to remove large particulate debris. Dissociated cells were assessed for viability (Countess II; Invitrogen) and cell density diluted to 700 cell/µl. Each single cell cDNA library was prepared for a target of 10,000 cells per sample. The cell suspension and Chromium Single Cell 3′ V2 or V3 reagents (10x Genomics) were loaded onto chips to capture individual cells with individual gel beads in emulsion (GEMs) using the 10x Chromium Cell Controller. cDNA and library amplification for an optimal signal was 12 and 10 cycles, respectively. Sequencing was conducted on Illumina HiSeq2500 (Genomics Resource Core Facility, Johns Hopkins University, USA), or Novaseq6000 (Novogene) using 150 paired-end reads. Fasta sequence files were de-multiplexed, aligned and annotated using the chick ENSMBL database (GRCg6a, Ensembl release 94) using 10x Cell Ranger software. Gene expression was counted using UMI bar codes and gene-cell matrices were constructed. Using Seurat toolkits UMAP plots were generated from aggregates of multiple scRNA-seq libraries (Butler et al., 2018; Satija et al., 2015). Seurat was used to construct gene lists for DEGs, violin/scatter plots and dot plots. Significance of difference in violin/scatter plots was determined using a Wilcoxon Rank Sum test with Bonferroni correction. SingleCellSignalR was used to assess potential LR interactions between cells within scRNA-seq datasets (Cabello-Aguilar et al., 2020). Genes that were used to identify different types of retinal cells included the following: (1) Müller glia: GLUL, VIM, SCL1A3, RLBP1; (2) MGPCs: PCNA, CDK1, TOP2A, ASCL1; (3) microglia: C1QA, C1QB, CCL4, CSF1R, TMEM22 (SLC35G2); (4) ganglion cells: THY1, POU4F2, RBPMS2, NEFL, NEFM; (5) amacrine cells: GAD67, CALB2, TFAP2A; (6) horizontal cells: PROX1, CALB2, NTRK1; (7) bipolar cells: VSX1, OTX2, GRIK1, GABRA1; (8) cone photoreceptors: CALB1, GNAT2, OPN1LW; and (9) rod photoreceptors: RHO, NR2E3, ARR1. GO enrichment analysis was performed using ShinyGO V0.65 (http://bioinformatics.sdstate.edu/go/). scRNA-seq libraries can be queried at https://proteinpaint.stjude.org/F/2019.retina.scRNA.html or gene-cell matrix files can be downloaded from GitHub at https://github.com/fischerlab3140/ and https://github.com/jiewwwang/Single-cell-retinal-regeneration.
Fixation, sectioning and immunocytochemistry
Retinal tissue samples were formaldehyde fixed, sectioned and labeled via immunohistochemistry as described previously (Fischer et al., 2006; Ghai et al., 2009). Antibody dilutions and commercial sources used in this study are described in Table S6. Observed labeling was not due to off-target labeling of secondary antibodies or tissue auto-fluorescence because sections incubated exclusively with secondary antibodies were devoid of fluorescence. In addition, patterns of immunolabeling for all antibodies precisely match patterns of expression in scRNA-libraries of control retinas, similar to previous descriptions (Campbell et al., 2019, 2021a,b; El-Hodiri et al., 2021). Secondary antibodies used include donkey-anti-goat-Alexa488/568, goat-anti-rabbit-Alexa488/568/647, goat-anti-mouse-Alexa488/568/647 and goat-anti-rat-Alexa488 (Life Technologies) diluted to 1:1000 in PBS and 0.2% Triton X-100.
Labeling for EdU
For the detection of nuclei that incorporated EdU, immunolabeled sections were fixed in 4% formaldehyde in 0.1 M PBS (pH 7.4) for 5 min at room temperature. Samples were washed for 5 min with PBS, permeabilized with 0.5% Triton X-100 in PBS for 1 min at room temperature and washed twice for 5 min in PBS. Sections were incubated for 30 min at room temperature in a buffer consisting of 100 mM Tris, 8 mM CuSO4 and 100 mM ascorbic acid in dH2O. The Alexa Fluor 568 Azide (Thermo Fisher Scientific) was added to the buffer at a 1:100 dilution.
Terminal deoxynucleotidyl transferase dUTP nick end labeling (TUNEL)
The TUNEL assay was implemented to identify dying cells by imaging fluorescent labeling of double-stranded DNA breaks in nuclei. The In Situ Cell Death Kit (TMR red; Roche Applied Science) was applied to fixed retinal sections as per the manufacturer's instructions.
Photography, measurements, cell counts and statistics
Microscopy images of retinal sections were captured using the Leica DM5000B microscope with epifluorescence and the Leica DC500 digital camera. High resolution confocal images were obtained using a Leica SP8 available in The Department of Neuroscience Imaging Facility at The Ohio State University, USA. Representative images are modified to have enhanced color, brightness and contrast for improved clarity using Adobe Photoshop. In EdU proliferation assays, a fixed region of retina was analyzed and average numbers of Sox2 and EdU co-labeled cells counted. The retinal region selected for investigation was standardized between treatment and control groups to reduce variability and improve reproducibility.
Similar to previous reports (Ghai et al., 2009), immunofluorescence intensity was quantified using ImageJ. Identical illumination, microscope and camera settings were used to obtain images for quantification. Retinal areas were sampled from digital images. Areas were randomly sampled over the INL, outer plexiform layer (OPL) and outer nuclear layer (ONL). Measurements for immunofluorescence in MG/MGPCs were made from single optical confocal sections. Measurements of pS6, pStat3 and pSmad1/5/8 immunofluorescence were made for a fixed, cropped area (25,000 µm2) of INL, OPL and ONL. Activation of cell signaling through mTor (pS6), Jak/Stat (pStat3) and BMP/SMAD (pSmad1/5/8) in outer layers of NMDA-damaged retinas is known to manifest exclusively in MG (Todd et al., 2016a, 2017; Zelinka et al., 2016). Measurements were made for regions containing pixels with intensity values of 70 or greater (0=black and 255=saturated). The cropped areas contain between 80 and 140 MG or MGPCs; numbers of cells vary depending on treatments that influence the proliferation of MGPCs. The intensity sum was calculated as the total of pixel values for all pixels within threshold regions.
We performed a Levene's test to determine whether data from control and treatment groups had equal variance. For treatment groups where the Levene's test indicated unequal variance, suggesting non-parametric data, we performed a Mann–Whitney U-test (Wilcoxon Rank Sum Test). For statistical evaluation of parametric data we performed a two-tailed paired t-test to account for intra-individual variability where each biological sample served as its own control (left eye – control; right eye – treated). For multivariate analysis across >2 treatment groups we performed an ANOVA with the associated Tukey Test to evaluate significant differences between multiple groups.
Conceptualization: W.A.C., A.J.F.; Methodology: W.A.C.; Validation: A.T.; Formal analysis: W.A.C., H.M.E.-H., E.C.H., A.J.F.; Investigation: W.A.C., A.T., H.M.E.-H., E.C.H., M.H., S. Blum; Resources: T.H., S. Blackshaw; Writing - original draft: W.A.C., A.J.F.; Writing - review & editing: W.A.C., H.M.E.-H., S. Blackshaw, A.J.F.; Supervision: H.M.E.-H., A.J.F.; Project administration: A.J.F.; Funding acquisition: S. Blackshaw, A.J.F.
This work was supported by National Institutes of Health grants R01 EY032141-01 (to A.J.F.) and U01 EY027267-03 (to A.J.F. and S. Blackshaw). Deposited in PMC for release after 12 months.
scRNA-seq libraries can be queried at https://proteinpaint.stjude.org/F/2019.retina.scRNA.html. Gene-cell matrix files can be downloaded from GitHub at https://github.com/fischerlab3140/ and https://github.com/jiewwwang/Single-cell-retinal-regeneration.
Peer review history
The peer review history is available online at https://journals.biologists.com/dev/article-lookup/doi/10.1242/dev.200127.
S. Blackshaw co-founded and is a shareholder of CDI Labs, LLC. S. Blackshaw is also a consultant for Third Rock Ventures, LLC.