ABSTRACT
During tissue regeneration, proliferation, dedifferentiation and reprogramming are necessary to restore lost structures. However, it is not fully understood how metabolism intersects with these processes. Chicken embryos can regenerate their retina through retinal pigment epithelium (RPE) reprogramming when treated with fibroblast factor 2 (FGF2). Using transcriptome profiling, we uncovered extensive regulation of gene sets pertaining to proliferation, neurogenesis and glycolysis throughout RPE-to-neural retina reprogramming. By manipulating cell media composition, we determined that glucose, glutamine or pyruvate are individually sufficient to support RPE reprogramming, identifying glycolysis as a requisite. Conversely, the activation of pyruvate dehydrogenase by inhibition of pyruvate dehydrogenase kinases, induces epithelial-to-mesenchymal transition, while simultaneously blocking the activation of neural retina fate. We also identified that epithelial-to-mesenchymal transition fate is partially driven by an oxidative environment. Our findings provide evidence that metabolism controls RPE cell fate decisions and provide insights into the metabolic state of RPE cells, which are prone to fate changes in regeneration and pathologies, such as proliferative vitreoretinopathy.
INTRODUCTION
Metabolic activity is a crucial cellular property that can be altered in relation to many cellular and molecular functions, such as cell signaling, proliferation, differentiation and reprogramming (Agathocleous and Harris, 2013; Wu et al., 2016). Reprogramming somatic cells into induced pluripotent stem cells (iPSCs) using Yamanaka factors causes considerable molecular alterations, such as activation or repression of signaling pathways, epigenetic arrangements and metabolic remodeling (Polo et al., 2012; Takahashi and Yamanaka, 2006; Shi et al., 2017). This is characterized by a switch from oxidative phosphorylation (OXPHOS) to only glycolytic metabolism as cells transition to a pluripotent state (Teslaa and Teitell, 2014; Wu et al., 2016; Panopoulos et al., 2011). This metabolic switch is required for cell reprogramming, as glycolysis inhibition and OXPHOS activation decrease reprogramming efficiency (Sun et al., 2020; Panopoulos et al., 2011). These observations suggest that any cell that undergoes reprogramming or fate change may experience deterministic metabolic rearrangements. During regeneration, reprogramming of resident cells is crucial for recovering lost tissues or body parts (Min and Whited, 2023; Jopling et al., 2011; Chiba, 2014; Wan and Goldman, 2016). However, how metabolism affects cell reprogramming in the context of regeneration is a question that remains poorly studied.
One of the best examples of cell reprogramming during regeneration is blastema formation, which is necessary for many forms of epimorphic regeneration. The blastema is formed by the reprogramming of cells surrounding the injured area following the loss of body parts, such as limbs, tails and fins (Min and Whited, 2023; Jopling et al., 2011). Likewise, retina regeneration in several vertebrates also requires cell reprogramming. In zebrafish, the neural retina regenerates through reprogramming of the Müller glia, whereas in newts, retinal regeneration proceeds through retinal pigment epithelium (RPE) reprogramming (Wan and Goldman, 2016; Barbosa-Sabanero et al., 2012). Interestingly, embryonic amniotes, such as the chicken embryo, can also regenerate the neural retina through RPE reprogramming if stimulated with fibroblast growth factor 2 (FGF2) at the time of injury (Spence et al., 2007, 2004; Park and Hollenberg, 1989). In this model, RPE reprogramming takes place at day 4 of development [Hamburger–Hamilton (HH) Stage 24; Hamburger and Hamilton, 1951] but by day 5 (HH Stage 27), RPE loses this ability (Tangeman et al., 2022; Sakami et al., 2008). Cell proliferation, epigenetic remodeling, and activation of neural genes have been identified as the cellular processes required to facilitate RPE reprogramming in embryonic chickens (Tangeman et al., 2022, 2021; Luz-Madrigal et al., 2014).
Despite these findings, the metabolic foundation of embryonic RPE reprogramming remains relatively unexplored. Our recent study highlighted the activation of retinol metabolism concurrent with RPE fate restriction, suggesting that metabolic pathways associated with RPE maturation may limit reprogramming competence (Tangeman et al., 2022). However, the metabolic requirements of RPE cells as they dedifferentiate and change their identity during reprogramming are unclear. To address these questions, we dissected the role of metabolism during RPE reprogramming in embryonic chickens. Using an RPE explant system that facilitates precise modulation of the metabolic environment, we identified glycolysis as an essential pathway for cell reprogramming, and that inhibition of pyruvate dehydrogenase kinases (PDKs) provokes the activation of an epithelial-mesenchymal transition (EMT) program that redirects the cells toward a mesenchymal fate.
RESULTS
RPE explants reprogram into neural retina
Cellular reprogramming of RPE into neural retina is induced by FGF2 in chicken embryos (Spence et al., 2004). To facilitate the elucidation of the metabolic determinants of RPE reprogramming, explant cultures were used to reprogram RPE into neural retina (Sakami et al., 2008; Tangeman et al., 2022). To better understand the RPE explant model system, we performed a phenotypic characterization. The RPE was collected from embryonic day (E) 4 chickens and cultured in the presence or absence of FGF2 (Fig. 1A). By 24 h of culture, the explants with FGF2 started to show signs of neuroepithelium formation, which became more evident at 48 h and well-defined after 96 h of culture (Fig. 1B). In vivo, FGF2 is essential to induce RPE reprogramming; however, it has not been determined whether sustained FGF2 treatment is necessary for reprogramming and neuroepithelial growth. To test this, we removed FGF2 1 h or 6 h after the initiation of the culture. Surprisingly, 1 h of FGF2 treatment was sufficient to induce RPE reprogramming (Fig. S1A), suggesting that FGF2 is the trigger for cell reprogramming, but its sustained presence is not necessary for neuroepithelial growth. Similar to in vivo studies, RPE explants can reprogram when collected at E4, but lose neurocompetence at E5 (Tangeman et al., 2022; Sakami et al., 2008). However, the exact time point at which reprogramming competence is lost in culture has not yet been determined. Therefore, we added FGF2 to E4 explants at different time points in the culture. We found that after 24 h, RPE was still able to reprogram, but not after 48 h of culture, demonstrating that RPE progressively lost reprogramming competence between 24 h and 48 h of culture (Fig. S1B).
RPE explants reprogram into neural retina. (A) RPE sheets were collected from chicken embryos at E4 and cultured for up to 4 days. (B) Representative RPE explants at the indicated time points of cell culture. Red arrows indicate reprogrammed RPE. (C) Gene expression levels of the neural genes SOX2, SIX6 and PAX6, as well as the RPE genes RPE65, TYR and OTX2 determined by RT-qPCR. (D) Two-dimensional principal component analysis (PCA) displaying RNAseq samples colored by condition. (E) Row-normalized heatmap displaying expression patterns of RPE identity factors (top) and neural retina transcription factors (bottom) in response to FGF2. (F) Immunofluorescence detection of SOX2 and OTX2 in explants collected at 48 h of cell culture. Insets show higher-magnification views of the boxed areas. Data are mean±s.d. *P<0.05, ***P<0.001, ****P<0.0001. n=4.
RPE explants reprogram into neural retina. (A) RPE sheets were collected from chicken embryos at E4 and cultured for up to 4 days. (B) Representative RPE explants at the indicated time points of cell culture. Red arrows indicate reprogrammed RPE. (C) Gene expression levels of the neural genes SOX2, SIX6 and PAX6, as well as the RPE genes RPE65, TYR and OTX2 determined by RT-qPCR. (D) Two-dimensional principal component analysis (PCA) displaying RNAseq samples colored by condition. (E) Row-normalized heatmap displaying expression patterns of RPE identity factors (top) and neural retina transcription factors (bottom) in response to FGF2. (F) Immunofluorescence detection of SOX2 and OTX2 in explants collected at 48 h of cell culture. Insets show higher-magnification views of the boxed areas. Data are mean±s.d. *P<0.05, ***P<0.001, ****P<0.0001. n=4.
To characterize RPE reprogramming ex vivo at the molecular level, explants were collected at different time points during cell culture (Fig. S1C) for gene expression analysis of neural retina markers (SOX2, PAX6 and SIX6) and RPE markers (RPE65, TYR and OTX2). We found that FGF2 treatment induced upregulation of SOX2 and SIX6 at 6 h, whereas PAX6 was upregulated at 24 h (Fig. 1C). In contrast, the RPE markers were acutely downregulated in the presence of FGF2 (Fig. 1C). The downregulation of TYR and OTX2 was observed after 12 h of culture, whereas RPE65 was significantly downregulated after 24 h (Fig. 1C). In the PBS-treated controls, RPE65 and OTX2 abundance increased after 48 h, indicating RPE maturation (Fig. 1C). Based on these results, we collected another set of explants at 24 h and 48 h for RNA sequencing (RNAseq) (Fig. S2). Principal component analysis (PCA) showed that the sequenced samples clustered by condition (Fig. 1D). Normalized count plots for retinal and RPE markers were plotted from the RNAseq data, showing the same regulatory patterns for genes previously assessed by RT-qPCR (Fig. 1C; Fig. S3A). As expected, FGF2 treatment led to an increased abundance of genes encoding eye field transcription factors (RAX, SIX3 and LHX2), a neural retina specification marker (VSX2), and retinal ganglion cell determinants (POU4F2 and ATOH7) (Fig. 1E). Additionally, genes associated with the term ‘Generation of neurons’ (GO:0022008) were upregulated in reprogramed explants (Fig. S3B). In contrast, RPE genes were downregulated in the presence of FGF2 after 48 h (Fig. 1E). To confirm the gene expression results, immunohistochemistry was performed for SOX2 and OTX2 in the explants collected at 48 h of culture. Explants treated with FGF2 showed a positive signal for SOX2, which was absent in untreated explants (Fig. 1F). Instead, explants that were not exposed to FGF2 were OTX2 positive, whereas in FGF2-treated explants, OTX2 abundance was low and mostly restricted to pigmented areas (Fig. 1F). Altogether, these data show that the molecular reprogramming of RPE into neural retina in vitro is observable at 24 h and robustly detectable after 48 h.
Cell proliferation is tightly regulated during RPE reprogramming
From the RNAseq experiment, we detected 19,194 genes, of which 3682 were differentially expressed genes (DEGs), fitting the criteria log fold change ≥1 and adjusted P-value≤0.05 (Fig. 2A). The top 10 DEGs that were repressed or activated by FGF2 at 24 h and 48 h are shown in Fig. 2B. We found that all the FGF2-repressed genes at 24 and 48 h were different, whereas, amongst the top FGF2-activated genes, FEZF2, KK34 and NOS1 were observed at both time points. Therefore, to identify the possible pathways and cellular processes that act as determinants for RPE reprogramming, we performed pathway enrichment analysis using the DEGs classified as FGF2-activated or FGF2-repressed genes at both 24 h and 48 h (Fig. 2B). The regulated processes included ‘MAPK signaling’, ‘mesenchyme differentiation’, ‘reactive oxygen species (ROS) metabolic process’, ‘DNA metabolic process’ and ‘cell cycle’ (Fig. 2B). It was previously reported that during RPE reprogramming, cells proliferate to form a neuroepithelium (Luz-Madrigal et al., 2020; Tangeman et al., 2022). Accordingly, in RPE explants, genes associated with the cell cycle (KEGG: map04110) were robustly upregulated by FGF2 at 24 h and 48 h (Fig. 2C), which was confirmed by 5-ethynyl-2′-deoxyuridine (EdU) incorporation and phospho-histone H3 (pHH3) staining (Fig. 2D,E). Taken together, these data confirm that cell proliferation is one of the major cellular processes associated with RPE reprogramming ex vivo.
Cell proliferation is highly activated during RPE reprogramming. (A) Heatmap displays row-normalized expression patterns of 3682 identified DEGs, defined by the criteria |log2(fold change)|≥1 and adjusted P-value≤0.05. (B) Top: The top 10 up- and downregulated DEGs, ranked by log-fold change. Log-fold change values represent expression change in response to FGF2 relative to PBS, after 24 h or 48 h in culture. Bottom: Pathway enrichment analysis was used to assign biological functions to up- and downregulated gene sets in response to FGF2. The adjusted P-values for select terms are displayed as a bar chart. (C) The top 30 upregulated genes associated with the term ‘Cell cycle’ (KEGG:04110). (D) EdU and pHH3 staining of explants cultured for 24 h and 48 h. Insets show higher-magnification views of the boxed areas. (E) Quantification of EdU- or pHH3-positive cells from D. Data are mean±s.d. *P<0.05, ***P<0.001, ****P<0.0001. n=5.
Cell proliferation is highly activated during RPE reprogramming. (A) Heatmap displays row-normalized expression patterns of 3682 identified DEGs, defined by the criteria |log2(fold change)|≥1 and adjusted P-value≤0.05. (B) Top: The top 10 up- and downregulated DEGs, ranked by log-fold change. Log-fold change values represent expression change in response to FGF2 relative to PBS, after 24 h or 48 h in culture. Bottom: Pathway enrichment analysis was used to assign biological functions to up- and downregulated gene sets in response to FGF2. The adjusted P-values for select terms are displayed as a bar chart. (C) The top 30 upregulated genes associated with the term ‘Cell cycle’ (KEGG:04110). (D) EdU and pHH3 staining of explants cultured for 24 h and 48 h. Insets show higher-magnification views of the boxed areas. (E) Quantification of EdU- or pHH3-positive cells from D. Data are mean±s.d. *P<0.05, ***P<0.001, ****P<0.0001. n=5.
Glycolysis is necessary for RPE reprogramming
Cells with high proliferative activity, such as stem cells, progenitors (including neural progenitors), and cancer cells, require large amounts of energy provided by metabolic activity, especially glycolysis (Abdel-Haleem et al., 2017; Lunt and Vander Heiden, 2011; DeBerardinis et al., 2008). Thus, we hypothesized that RPE reprogramming requires glycolysis owing to the high cell proliferation observed. Glycolysis is a pathway of sequential biochemical reactions that produces ATP, pyruvate, and metabolites for amino acid synthesis (Lunt and Vander Heiden, 2011). In contrast, gluconeogenesis results in glucose production (Hers and Hue, 1983). Glycolysis and gluconeogenesis share enzymes that catalyze reversible reactions, whereas rate-limiting enzymes are specific to each pathway. In glucose metabolism, hexokinases 1 and 2 (HK1, 2), phosphofructokinase (PFKP, PFKL) and pyruvate kinase (PKLR) are glycolytic enzymes, whereas pyruvate carboxylase (PC), phosphoenolpyruvate carboxylase 1 (PCK1), fructose 1,6-bisphosphatase 1 (FBP1) and glucose 6-phosphatase (G6P) are specific gluconeogenic enzymes (Fig. 3A). RNAseq analysis revealed that the genes encoding enzymes essential for glycolysis and gluconeogenesis were differentially expressed in cultured RPE explants (Fig. 3A,B). Interestingly, genes encoding most glycolytic enzymes, including rate-limiting enzymes, except ALDOB, were upregulated in the presence of FGF2 (Fig. 3B). LDHA and LDHB, which encode lactate dehydrogenases A and B, respectively, were also expressed and regulated during cell reprogramming. LDHA was upregulated at 48 h, whereas LDHB was transiently upregulated at 24 h (Fig. 3B). To confirm these observations, the expression levels of selected metabolic genes were determined by RT-qPCR (Fig. S4) at different time points, including 24 and 48 h. HK1 was upregulated by FGF2 at 24 h, whereas HK2 was significantly upregulated at 12 h (Fig. S4). GADPH was also upregulated in the presence of FGF2 and LDHA was positively regulated by FGF2, whereas LDHB was unaffected (Fig. S4). LDHA and LDHB are mainly differentiated by their affinity for pyruvate and lactate; LDHA has a higher affinity for pyruvate and LDHB for lactate; therefore, their differential expression might affect the glycolytic rate (Mishra and Banerjee, 2019; Read et al., 2001). In contrast, the levels of FBP1, a gluconeogenic gene, were repressed by FGF2 at 48 h (Fig. S4). These data suggest that glycolysis may be favored and is required for RPE reprogramming.
The glycolysis pathway is activated during RPE reprogramming and its inhibition affects cell reprogramming. (A) The glycolysis and gluconeogenesis pathway, including a pathway that glutamine can follow to enter glycolysis/gluconeogenesis. Green indicates glycolytic enzymes, red gluconeogenesis-specific enzymes and blue indicates lactic enzymes. (B) Heatmap displaying row-normalized expression values of glycolysis (green), gluconeogenesis (red) and lactic enzyme (blue) genes. (C) Representative RPE explants cultured for 96 h in 3C-media at the indicated conditions. (D) Quantification of the explant phenotype shown in C. Class I: no reprogramming; Class II: signs RPE reprogramming; Class III: reprogrammed RPE. (E,F) Glucose (E) and lactate (F) determination in RPE cultured in 3C-media after 48 h under the indicated conditions. (G-I) RT-qPCR gene expression analysis of retina and RPE genes (G), metabolic genes (H) and cell proliferation genes (I) of RPE explants cultured in 3C-media at the indicated conditions and time points. Data are mean±s.d. *P<0.05, **P<0.01, ***P<0.001, ****P<0.0001. n=5 (E,F); n=4 (G-I). For E, $P<0.05 and $$P<0.01 compared with PBS+2DG condition.
The glycolysis pathway is activated during RPE reprogramming and its inhibition affects cell reprogramming. (A) The glycolysis and gluconeogenesis pathway, including a pathway that glutamine can follow to enter glycolysis/gluconeogenesis. Green indicates glycolytic enzymes, red gluconeogenesis-specific enzymes and blue indicates lactic enzymes. (B) Heatmap displaying row-normalized expression values of glycolysis (green), gluconeogenesis (red) and lactic enzyme (blue) genes. (C) Representative RPE explants cultured for 96 h in 3C-media at the indicated conditions. (D) Quantification of the explant phenotype shown in C. Class I: no reprogramming; Class II: signs RPE reprogramming; Class III: reprogrammed RPE. (E,F) Glucose (E) and lactate (F) determination in RPE cultured in 3C-media after 48 h under the indicated conditions. (G-I) RT-qPCR gene expression analysis of retina and RPE genes (G), metabolic genes (H) and cell proliferation genes (I) of RPE explants cultured in 3C-media at the indicated conditions and time points. Data are mean±s.d. *P<0.05, **P<0.01, ***P<0.001, ****P<0.0001. n=5 (E,F); n=4 (G-I). For E, $P<0.05 and $$P<0.01 compared with PBS+2DG condition.
To test whether glycolysis is indispensable for RPE reprogramming, we used 2-deoxy-D-glucose (2DG), a glucose analog that inhibits glycolysis by inhibiting hexokinase and glucose-6-phosphate isomerase in a competitive and non-competitive manner (Pajak et al., 2019; Aft et al., 2002; Shiraishi et al., 2014). 2DG up to 20 mM concentration did not inhibit RPE reprogramming, a phenotype confirmed by morphological examination and gene expression determination of neural retina and RPE markers (Fig. S5A,B). However, because 2DG glycolysis inhibition can be competitive, we calculated the glucose concentration based on the media composition as previously published (Sakami et al., 2008) and found it to be 60 mM, suggesting that 20 mM 2DG might be insufficient to inhibit glycolysis. Therefore, we cultured RPE explants in DMEM without glucose or pyruvate, in the presence or absence of FGF2. The RPE with FGF2 was able to reprogram (Fig. S5C), suggesting that the cells used a different metabolite, such as glutamine, as an energy source. Thus, RPE explants were cultured in media depleted of three major carbon sources: glucose, pyruvate and glutamine (3C-media). In this case, RPE reprogramming was limited, but if glucose, pyruvate or glutamine was added at a concentration of 2 mM, the phenotype was rescued. Interestingly, RPE reprogramming in the presence of any of the three major carbon sources was inhibited by 2DG at equimolar concentration (Fig. 3C). To quantify the effect of 2DG, we classified RPE reprogramming, based on phenotypic observations on day 4 of culture, into three classes: class I, not reprogrammed RPE; class II, rudimentary signs of RPE reprogramming; and class III, extensively reprogrammed RPE. In the absence of FGF2, the control explants did not reprogram (16/16). In the presence of FGF2, class II (5/16) and class III (3/16) explants were detected in 3C-media alone, suggesting that the amino acids present might be used as energy and carbon sources. In the presence of 2DG only, none of the explants (16/16) reprogrammed (Fig. 3D). In the presence of FGF2, glucose and 2DG, most of the explants were class I (12/16), one was class II, and three were class III. Interestingly, most explants cultured in the presence of FGF2, pyruvate and 2DG showed signs of reprogramming (class II, 7/8), one did not reprogram (class I), and none was classified as class III. In contrast, in the presence of FGF2, glutamine and 2DG, most explants were class I (6/8) and the remainder class II (2/8) (Fig. 3D). At glucose concentration of 4 mM and 2DG 2 mM, no class I explants were detected; instead, most of the explants were class III (6/8) and a few were class II (2/8), reflecting the competitive nature of 2DG (Fig. S5D,E). Similarly, in 4 mM glutamine or pyruvate and 2DG 2 mM, only class II explants were detected (Fig. S5D,E). Thus, the 2DG effect on RPE reprogramming when metabolites such as pyruvate, glutamine or amino acids are the main carbon sources, may indicate that these metabolites are converted into glucose prior to being metabolized by glycolysis.
To validate the inhibition of glycolysis and the possible gluconeogenic activity during RPE reprogramming, we determined the amount of glucose and lactate present in the explants cultured for 48 h and treated with 2 mM 2DG in combination with any of the major carbon sources at 2 mM concentration, similar to what was demonstrated in Fig. 3C. We found that 2DG caused glucose accumulation when any of the carbon sources and 2DG were present, an effect independent of the presence of FGF2 (Fig. 3E). Lactate was not detected in the explants cultured in PBS alone, but was detected when FGF2 or any of the carbon sources were present (Fig. 3F). As expected, 2DG depleted lactate production, even when the carbon source was pyruvate or glutamine (Fig. 3F). Lactate depletion in the presence of pyruvate and 2DG was unexpected, as pyruvate can be directly used for lactate production (Fig. 3A). Therefore, the fact that 2DG causes glucose accumulation and lactate depletion confirmed the inhibition of glycolysis. This also strongly suggests that pyruvate and glutamine, along with other carbon sources, such as amino acids, were used for glucose production, which was subsequently used by glycolysis to produce lactate as a by-product. Interestingly, the significant accumulation of lactate induced by FGF2 in the absence of any carbon source confirmed the activation of glycolysis during RPE reprogramming.
To corroborate the effects of 2DG on RPE reprogramming, we determined the gene expression levels of neural retina and RPE markers at 24 h and 48 h of cell culture. The results confirmed that 2DG stymied the upregulation of retina markers induced by FGF2 (Fig. 3G). However, the expression of RPE markers was downregulated in explants treated with either FGF2 or 2DG alone, suggesting that glycolysis inhibition might also disrupt RPE identity (Fig. 3G). Gene expression analysis of metabolic genes also showed that 2DG prevented the upregulation of glycolytic genes (Fig. 3H; Fig. S5F). In addition, treatment with 2DG and FGF2 led to the downregulation of the cell cycle genes E2F1 and PCNA relative to FGF2-only explants, suggesting that 2DG may limit cell proliferation during reprogramming (Fig. 3I). Altogether, these data show that glycolysis is necessary for RPE reprogramming and points toward glucose as a required metabolite.
Inhibition of pyruvate dehydrogenase kinases inhibits RPE reprogramming
Glycolysis inhibition demonstrates the importance of proper metabolism in RPE reprogramming. Therefore, we next investigated the effect of promoting pyruvate dehydrogenase (PDH) activity by inhibiting PDKs using dichloroacetate (DCA), a condition that promotes the catabolism of pyruvate through the tricarboxylic acid (TCA) cycle and, consequently, OXPHOS (Wang et al., 2021; Stacpoole and Greene, 1992; Michelakis et al., 2008; Kluza et al., 2012) (Fig. 4A). We hypothesized that PDKs inhibition would activate OXPHOS and inhibit RPE reprogramming, similar to what is observed during iPSC formation (Sun et al., 2020). DCA alone did not affect the phenotype of RPE explants, except for an apparent increase in the size of the explants treated with 20 mM DCA (Fig. 4B). Interestingly, DCA inhibited RPE reprogramming in a concentration-dependent manner, with 100% inhibition at 20 mM (Fig. 4B). The phenotypic characteristics of the explants differed from those treated with 2DG (Figs 3C and 4B). Inhibition of RPE reprogramming by DCA was also observed in RPE explants cultured in DMEM containing only glutamine (Fig. S6A), suggesting that the DCA effect only depends on PDK inhibition and it is independent of the carbon source. Hematoxylin and Eosin staining revealed apparent disorganization of the RPE and lack of neuroepithelium formation in FGF2+DCA-treated explants (Fig. S6B). At the molecular level, we observed overexpression of SOX2 and SIX6 induced by FGF2 in the presence of DCA, but their expression levels were significantly lower than those observed with FGF2 alone, except for SOX2 at 24 h. However, PAX6 levels were mostly unaffected, and DCA alone upregulated its gene expression at 24 h and 48 h (Fig. 4C). DCA treatment alone and in combination with FGF2 led to the repression of RPE65 and TYR, whereas OTX2 downregulation by FGF2 was not observed in the presence of DCA (Fig. 4C). These observations indicate that PDK inhibition not only affects RPE reprogramming but also perturbs RPE identity.
Inhibition of PDKs blocks RPE reprogramming, inhibits cell proliferation, and activates ERK. (A) Effect of inhibition of PDKs via DCA on OXPHOS metabolism. (B) Representative RPE explants after 96 h of cell culture treated with FGF2 and DCA at varying concentrations. (C) RT-qPCR gene expression analysis of neural (top) and RPE (bottom) genes. (D) EdU and pHH3 staining performed in explants after 48 h of cell culture. Insets show higher-magnification views of the boxed areas. (E) Quantification of EdU- and pHH3-positive cells from D. (F) Lactate determination in RPE explants after 48 h of cell culture at indicated conditions. (G) Representative western blot of RPE explants at 48 h of cell culture under the indicated conditions. (H) Protein band quantification by densitometry of the western blot shown in F. Data are mean±s.d. *P<0.05, **P<0.01, ***P<0.001, ****P<0.0001. n=4 (C); n=6 (E); n=5 (F); n=3 (H). ND, not determined.
Inhibition of PDKs blocks RPE reprogramming, inhibits cell proliferation, and activates ERK. (A) Effect of inhibition of PDKs via DCA on OXPHOS metabolism. (B) Representative RPE explants after 96 h of cell culture treated with FGF2 and DCA at varying concentrations. (C) RT-qPCR gene expression analysis of neural (top) and RPE (bottom) genes. (D) EdU and pHH3 staining performed in explants after 48 h of cell culture. Insets show higher-magnification views of the boxed areas. (E) Quantification of EdU- and pHH3-positive cells from D. (F) Lactate determination in RPE explants after 48 h of cell culture at indicated conditions. (G) Representative western blot of RPE explants at 48 h of cell culture under the indicated conditions. (H) Protein band quantification by densitometry of the western blot shown in F. Data are mean±s.d. *P<0.05, **P<0.01, ***P<0.001, ****P<0.0001. n=4 (C); n=6 (E); n=5 (F); n=3 (H). ND, not determined.
Because cell proliferation is a key cellular process for RPE reprogramming, we investigated whether cell proliferation was also affected by PDK inhibition. A comparable number of EdU- and pHH3-positive cells in PBS- and DCA-treated explants at 24 h and 48 h were observed; however, cell proliferation induced by FGF2 was inhibited by DCA (Fig. 4D,E; Fig. S6C,D). Furthermore, PCNA and E2F1 expression was downregulated in explants treated with FGF2+DCA (Fig. S6E). To ensure that DCA exerted the intended effect, we measured lactate in 48 h cultures in explant medium (glucose 60 mM) and detected the phosphorylated form of PDH (pPDH). Lactate determination revealed that lactate levels were unaffected in DCA-treated explants. However, significantly less lactate was observed in DCA+FGF2-treated explants in comparison with FGF2 alone (Fig. 4F). This indicates that PDK inhibition promoted pyruvate metabolism via the TCA cycle. It is important to highlight that, in the absence of FGF2, lactate was detected, but it was not detected in 3C-media (Fig. 3F). We attributed this difference to the presence of glucose (0 versus 60 mM) and other metabolites. In concordance with lactate determination, we observed a decrease in pPDH induced by DCA at 24 or 48 h of cell culture, an effect more pronounced at 48 h (Fig. 4G,H; Fig. S6F,G). This result confirmed that DCA inhibited PDKs and, consequently, PDH was activated (not phosphorylated), thereby affecting lactate production in the presence of FGF2 (Fig. 4F). Given that FGF2 signals through the MAPK kinase pathway, which is required for RPE reprogramming (Spence et al., 2007, 2004), we investigated whether DCA might affect ERK phosphorylation. Surprisingly, DCA alone promoted ERK phosphorylation, which became more prominent in the presence of FGF2 (Fig. 4G,H; Fig. S6F,G). Interestingly, α-tubulin, a protein that is highly present in neural tissues (Gloster et al., 1994; Lewis et al., 1985), accumulated in FGF2-treated explants, but was almost undetectable in PBS-, DCA- and FGF2+DCA-treated explants (Fig. 4G,H), confirming that DCA inhibited the neural fate induced by FGF2. DCA effects on lactate and pPDH suggest that OXPHOS was activated. Using a fluorogenic probe that accumulates in mitochondria via the mitochondrial membrane gradient (Yang et al., 2022), we observed an increase of fluorescent signal in DCA explants compared with PBS control explants (Fig. S6H,I). An increase of fluorescent intensity was also observed in FGF2- and FGF2+DCA-treated explants; however, this effect was neutralized when adjusted for the size and morphological changes of the explants. Altogether, these data denote that PDK inhibition promotes PDH activity and OXPHOS, which inhibits RPE reprogramming.
Inhibition of PDKs redirects RPE reprogramming toward an EMT program
The phenotypic and molecular characteristics observed following PDK inhibition suggested that cell reprogramming was initiated without resulting in retina formation as a final cellular fate. To elucidate this, we collected explants treated with DCA for RNAseq at 24 h and 48 h and compared them with the samples treated with FGF2 alone (Fig. S2). PCA revealed clear clustering of the samples per treatment (Fig. 5A; Fig. S7A-D), indicating a robust effect of PDK inhibition on gene expression. Although we observed a clear effect of DCA at both 24 h and 48 h, the measured gene expression responses were generally much stronger at 48 h; therefore, we decided to focus our study on the samples collected at 48 h. We noted that some genes normally affected by FGF2, such as FGFR1, were not affected by DCA (Fig. 5B). However, other classes of genes upregulated by FGF2, such as the neural retina factor ASCL1, were downregulated in the presence of DCA (Fig. 5B). Genes with the opposite behavior were also found; this was the case for MAP1LC3B, a gene important for autophagy, a process that can be regulated by FGF2/FGFR1 (Yuan et al., 2017) (Fig. 5B). In another set of genes, DCA potentiated the effect of FGF2; for example, LDHA was upregulated by FGF2, which was exacerbated by FGF2+DCA. Moreover, TYRP1 was downregulated by FGF2 and further downregulated in the presence of both FGF2 and DCA (Fig. 5B).
Inhibition of PDKs during RPE reprogramming turns on an EMT program. (A) Two-dimensional principal component analysis (PCA) displaying RNAseq samples colored by condition. (B) Representative RNAseq count plots displaying normalized expression values for genes of interest from explants after 48 h of cell culture. (C) An interaction analysis was performed to identify genes that exhibit altered responses to FGF2 in the presence of DCA at 48 h. Genes with adjusted P-value≤0.05 and |log2(fold change)|≥1 are shown. (D) Pathway enrichment analysis performed on gene sets with positive or negative log2(fold change) values, as identified in C. (E) Genes of interest that display a significant DCA:FGF2 interaction are displayed in a row-normalized heatmap. (F) Gene expression of EMT- and mesenchyme-associated genes at the indicated conditions and time points. Data are mean±s.d. *P<0.05, **P<0.01, ***P<0.001; ****P<0.0001; n=3 for (B); n=4 for (F).
Inhibition of PDKs during RPE reprogramming turns on an EMT program. (A) Two-dimensional principal component analysis (PCA) displaying RNAseq samples colored by condition. (B) Representative RNAseq count plots displaying normalized expression values for genes of interest from explants after 48 h of cell culture. (C) An interaction analysis was performed to identify genes that exhibit altered responses to FGF2 in the presence of DCA at 48 h. Genes with adjusted P-value≤0.05 and |log2(fold change)|≥1 are shown. (D) Pathway enrichment analysis performed on gene sets with positive or negative log2(fold change) values, as identified in C. (E) Genes of interest that display a significant DCA:FGF2 interaction are displayed in a row-normalized heatmap. (F) Gene expression of EMT- and mesenchyme-associated genes at the indicated conditions and time points. Data are mean±s.d. *P<0.05, **P<0.01, ***P<0.001; ****P<0.0001; n=3 for (B); n=4 for (F).
These observations indicated that the expression of many genes was altered by the interaction between FGF2 and DCA. Therefore, we performed an interaction analysis to answer how the role of FGF2 on RPE reprogramming is modified in the presence of DCA. From this analysis, we identified 1463 genes with significant interactions (Fig. 5C). Pathway enrichment analysis of the identified genes set with positive log2(fold change) values revealed that the addition of DCA enhanced the treatment effects of FGF2 on genes associated with the ‘collagen-containing ECM’, ‘MAPK cascade’, ‘mesenchyme development’, ‘ROS metabolic process’, ‘production of TGFβ’ and ‘regulation of EMT’ (Fig. 5D). By contrast, we observed a negative interaction for genes associated with ‘cell cycle’, ‘DNA metabolic process’, ‘eye development’, ‘the generation of neurons’ and ‘neural retina development’ (Fig. 5D). Thus, these negatively regulated genes and their associated processes represent a class of biological effects elicited by FGF2 that are muted by the addition of DCA. Accordingly, these findings can be reconciled with the observed inhibition of cell proliferation observed with FGF2+DCA treatment (Fig. 4D,E; Fig. S6C-E), as well as the accumulation of pERK in the presence of DCA (Fig. 4G,H; Fig. S6F,G). Selected genes were plotted in a heatmap, revealing the nature of the interactions between genes associated with FGF2 response, EMT, BMP signaling and ECM. These gene sets were downregulated by FGF2 alone, but, upon the addition of both FGF2 and DCA, they failed to be downregulated or even increased in expression (Fig. 5E). In contrast, genes that are essential for neural development and proliferation were robustly activated by FGF2 alone, an effect that was largely abrogated by combined treatment with FGF2 and DCA, demonstrating that DCA impairs the activation of essential neural reprogramming factors.
These results suggest that one of the main processes induced by PDK inhibition during RPE reprogramming is an EMT program and, consequently, the redirection of RPE reprogramming toward mesenchymal fate. Determination of the expression of selected gene markers of EMT and mesenchymal fate revealed that TGFB2 and vimentin (VIM) were robustly upregulated in the FGF2+DCA condition by 48 h, whereas SNAI1 and PITX1, which were downregulated by FGF2 alone, were unchanged upon the addition of FGF2+DCA (Fig. 5F). Notably, the addition of DCA downregulated ACTA2 expression relative to PBS condition and did not modify the expression of ACTA2 relative to FGF2 alone; similar gene expression patterns were observed in the RNAseq data (Fig. S7E), indicating that the RPE was reprogrammed into mesenchyme. However, during explant collection, a small amount of periocular mesenchyme tissue remained associated with the RPE, leaving the possibility that the observed gene expression alterations could be attributed to this remaining tissue. Thus, immunofluorescence against S100A4, a marker of EMT that is also present when the RPE acquires mesenchymal characteristics (Chen et al., 2012; Lo et al., 2011), was performed on explants collected at 48 h. S100A4 was not detected in the RPE or the reprogramed RPE, but was detected at low levels in the RPE of DCA-treated explants (Fig. 6). Interestingly, S100A4 was robustly present throughout the corresponding RPE regions of explants treated with FGF2+DCA, which was discernible by the presence of pigment (Fig. 6). Altogether, these observations show that PDK inhibition during FGF2 treatment activates an EMT program that redirects RPE away from neural lineages and toward a mesenchymal fate.
Inhibition of PDKs induces the presence of EMT protein S100A4 in the RPE. Representative immunofluorescence of EMT protein S100A4 in explants collected after 48 h of cell culture. Insets show higher-magnification views of the boxed areas.
EMT activation by PDK inhibition is partially driven by ROS
Inhibition of PDKs by DCA has been shown to increase mitochondrial respiration and ROS production (Kluza et al., 2012). In addition, it has been shown that ROS can induce EMT (Radisky et al., 2005; Jiang et al., 2017); therefore, the formation of ROS following inhibition of PDKs might have an active role in the induction of EMT. To test this hypothesis, we supplemented the medium with the antioxidant N-acetylcysteine (NAC). NAC at 1 mM or 10 mM did not show any obvious phenotypic effects on RPE explants (Fig. S8A) and did not affect RPE reprogramming induced by FGF2 at 10 mM (Fig. 7A). Surprisingly, NAC at 10 mM, but not 1 mM, partially rescued the inhibition of RPE reprogramming by DCA (Fig. 7A; Fig. S8B). Classification of the phenotype of the explants on day 4 of culture in class I (no reprogramming), class II (partially reprogrammed) and class III (reprogrammed) showed that DCA inhibited 100% of the reprogramming of the RPE; however, the addition of both DCA and NAC partially rescued reprogramming competence, with 30% of the explants showing signs of reprogramming, and 70% being fully reprogrammed, although the total amount of neuroepithelium produced was lower than that of the explants treated with FGF2 alone or FGF2+NAC (Fig. 7A,B). Thus, we evaluated the effect of NAC at 10 mM on RPE reprogramming in explants collected at 48 h and observed that the expression of neural retina markers was unaffected by NAC or DCA in the absence of FGF2 (Fig. 7C). In FGF2- and FGF2+DCA-treated explants, NAC did not affect SIX6 or PAX6 gene expression (Fig. 7C). However, DCA induced the downregulation of SOX2 in FGF2-treated explants, consistent with our previous results (Fig. 4C), but this downregulation was blocked by the addition of NAC (Fig. 7C). In contrast, NAC did not affect the expression of RPE markers previously observed to be regulated by DCA treatment (Fig. 4C), in both FGF2-treated and untreated explants (Fig. 7C). Also, NAC did not affect EMT marker expression in the absence of FGF2 (Fig. 7C). Furthermore, in FGF2-treated explants, TGFB2 and VIM gene expression induction by PDK inhibition was reduced by NAC, but no effect was detected on SNAI1 or ACTA2 gene expression (Fig. 7D). Regardless of the effect of NAC on EMT genes, no rescue effect was observed on the cell cycle genes E2F1 and PCNA (Fig. 7E). Altogether, these data suggest that limiting ROS production by the addition of NAC may partially inhibit the activation of the EMT machinery caused by the DCA:FGF2 interaction and also indicates that ROS production is only one component of EMT activation, as ROS alone cannot explain the totality of the DCA:FGF2 interaction, such as effects related to proliferation and the full scope of the changes in the expression of EMT machinery.
EMT activation by PDK inhibition is partially reverted by the antioxidant NAC. (A) Representative RPE explants after 96 h of culture treated with the indicated molecules. (B) Quantification of the occurrence of explant phenotypes shown in A. Class I: no reprogramming; Class II: signs of reprogramming; Class III: reprogrammed. (C,D) Gene expression of retina and RPE genes (C), and EMT and mesenchyme genes (D). (E) Gene expression of cell proliferation genes. Data are mean±s.d. *P<0.05, **P<0.01, ***P<0.001, ****P<0.0001. n=4.
EMT activation by PDK inhibition is partially reverted by the antioxidant NAC. (A) Representative RPE explants after 96 h of culture treated with the indicated molecules. (B) Quantification of the occurrence of explant phenotypes shown in A. Class I: no reprogramming; Class II: signs of reprogramming; Class III: reprogrammed. (C,D) Gene expression of retina and RPE genes (C), and EMT and mesenchyme genes (D). (E) Gene expression of cell proliferation genes. Data are mean±s.d. *P<0.05, **P<0.01, ***P<0.001, ****P<0.0001. n=4.
DISCUSSION
In this study, we comprehensively characterized RPE chicken embryo explants as a model for studying RPE reprogramming into neural retina ex vivo. Our studies confirmed that RPE reprogramming ex vivo is induced by FGF2, which causes the loss of RPE identity, and upregulation of neural and proliferation genes, similar to what is observed in vivo (Tangeman et al., 2022, 2021; Luz-Madrigal et al., 2014). Interestingly, our results show that FGF2 is a necessary trigger for reprogramming, but is not required for reprogramed RPE growth, a phenomenon that has not been possible to determine before.
The RPE explant model revealed remodeling of metabolic activity during RPE reprogramming, favoring glycolytic metabolism. Interestingly, we found that RPE cells can use glucose, pyruvate or glutamine as a carbon source for reprogramming. Inhibition of glycolysis by 2DG blocked RPE reprogramming, even when pyruvate or glutamine served as the sole carbon source. Pyruvate and glutamine metabolism do not require glycolysis to produce energy as both can be directly metabolized by the mitochondria via the TCA cycle (Fig. 3A) (Stumvoll et al., 1999). However, we found that glucose accumulated, and lactate dissipated, when only pyruvate or glutamine was present and glycolysis was inhibited by 2DG. This indicated that RPE synthesized glucose by gluconeogenesis, a pathway that was fed by pyruvate, glutamine, and probably by other metabolites, such as amino acids. Gluconeogenesis has been described as primarily limited to the liver, muscle, kidney and intestine (Scrutton and Utter, 1968; Mithieux and Gautier-Stein, 2014). Our RNAseq analysis revealed the expression of gluconeogenic genes in RPE, according to the gluconeogenic potential of the tissue. To date, it has not been shown that the RPE can perform gluconeogenesis, although RPE and retina have been described as metabolically active tissues (MacGregor et al., 1986; Hurley et al., 2015; Xu et al., 2020) and can produce pyruvate, malate and citrate from both glucose and lactate (Kanow et al., 2017). However, gluconeogenesis has been observed in the retina of both amphibians and mammals, although this is limited to Müller glia, cells that also store glycogen (Kanow et al., 2017; Viegas and Neuhauss, 2021; Scrutton and Utter, 1968), and it is unclear whether this activity is also present in photoreceptors (Hurley et al., 2015; Goldman, 1990; Mamczur et al., 2010). Our studies show that in the presence of either pyruvate or glutamine, glucose and lactate synthesis were unaffected by FGF2 presence, which may indicate that both non-reprogrammed and reprogrammed RPE have the same gluconeogenesis capacity.
Glycolysis and gluconeogenesis are intimately related pathways (Hers and Hue, 1983; Scrutton and Utter, 1968). They share several enzymes (Fig. 3A), but glycolysis is catabolic and gluconeogenesis is anabolic. Glycolysis produces energy: two ATP molecules, two NADH and two pyruvates, whereas gluconeogenesis requires two pyruvates, two ATP, two GTP and two NADH to synthesize a molecule of glucose. Lactate production is used to maintain the release of NAD+, which is necessary for maintaining glycolysis flux (Xie et al., 2020). In our study, we observed that, in the presence of pyruvate, lactate accumulated in the RPE either in the presence or absence of FGF2. This can be observed as a direct synthesis of lactate using available pyruvate (Fig. 3A). However, upon inhibition of glycolysis, glucose accumulated, and lactate levels were depleted; a similar effect was observed in the presence of glutamine as the only carbon source (Fig. 3E,F). Glucose accumulation can be explained by gluconeogenesis, but lactate depletion suggests that it was not directly synthesized by the available pyruvate, probably because of the limited amount of NADH. However, the fact that glucose was produced indicates that NADH was available. Canonical and non-canonical TCA cycle have been reported. Although the canonical TCA cycle is necessary to maintain pluripotency and occurs entirely in the mitochondria, the exit of pluripotency requires a switch to non-canonical TCA, which is characterized by oxalate formation in the cytoplasm by ATP citrate lyase (ACLY; Arnold et al., 2022). Therefore, we speculate that glucose is a key metabolite for RPE reprogramming, and its metabolism may occur through alternative pathways and organelles, a hypothesis that requires further investigation.
Interestingly, our study also showed that DCA, an inhibitor of PDKs, blocks RPE reprogramming into neural retina by activating an EMT program and acquisition of mesenchymal fate. Inhibition of PDKs promotes mitochondrial metabolism, or OXPHOS, causing an increase in TCA cycle intermediates and a reduction of lactate levels (Kluza et al., 2012). Similarly, we observed an increase in mitochondrial activity and a decrease in lactate production in RPE explants treated with DCA, suggesting an increase of OXPHOS. OXPHOS activation can also cause ROS formation via the TCA cycle and the electron transport chain (Hernansanz-Agustín and Enríquez, 2021; Hamanaka and Chandel, 2010), a phenomenon also observed upon PDK inhibition by DCA (Ruggieri et al., 2014). Interestingly, we observed that the ROS scavenger NAC partially inhibited EMT progression in the RPE, and overexpression of TGFB2 was observed after DCA treatment. Mitochondrial ROS can affect and regulate cellular processes such as proliferation, differentiation and apoptosis through the regulation of signaling pathways, including PI3K and MAPK, or by activating transcription factors, such as HIF1a and FOXO (Hamanaka and Chandel, 2010; Schieber and Chandel, 2014). Furthermore, ROS can induce EMT (Jiang et al., 2017; Radisky et al., 2005). Therefore, inhibition of PDKs may cause also ROS production, contributing to the EMT program observed during RPE reprogramming, which could represent a linear sequence of events supported by our data.
The interaction analysis performed for RNAseq data revealed that inhibition of PDKs increased the expression of genes normally downregulated by FGF2, such as those involved in ECM, BMP signaling, FGF response, and EMT. Amongst the upregulated genes are INHBA (activin), BMP2/6 and TGFB1/2, which belong to the TGFβ superfamily of structurally related cytokines. BMP factors are crucial effectors of RPE and neural retina fate (Steinfeld et al., 2017; Müller et al., 2007), and activin signaling has been shown to restrict RPE neurocompetence (Sakami et al., 2008; Fuhrmann et al., 2000). In addition, several members of the TGFβ superfamily, including TGFβ1 and activin, are positive regulators of EMT (Kahata et al., 2018) as well as S100A4, a calcium-binding protein present in EMT progression across several cellular contexts, including RPE (Lo et al., 2011; Chen et al., 2012). Recently, Zhu et al., 2023, showed in human umbilical artery endothelial cells that acetate controls endothelial-to-mesenchymal transition, a process that requires TFGβ signaling (Zhu et al., 2023). Zhu et al. found that TFGβ suppresses the expression of PDK4, promoting the synthesis of acetate, which is used by ACSS2 to produce acetyl-CoA. Acetyl-CoA overproduction causes acetylation of the TGFβ receptor ALK5 (TGFBR1) as well as the mediators SMAD2 and SMAD4, allowing long-term TGFβ signaling and endothelial-to-mesenchymal transition (Zhu et al., 2023). Surprisingly, we found that PDKs inhibition by DCA resulted in overexpression of TGFB1 and TGFB2. Thus, a potential molecular mechanism of the EMT program observed might be the overproduction of acetate because of the inhibition of PDKs.
Mitochondrial defects have been identified as major alterations associated with RPE dysfunction and disease (Brown et al., 2019; Tong et al., 2022). In RPE obtained from human donors with age-related macular degeneration (AMD), treatment with NAC or rapamycin improved mitochondrial function, which was not observed in RPE from individuals without AMD (Ebeling et al., 2020). Similarly, in postnatal mice, the ablation of mitochondrial functions specifically in the RPE by knocking out Tfam, which is essential for mtDNA transcription and replication (Larsson et al., 1998), provoked gradual RPE dedifferentiation, hypertrophy, and activation of glycolysis, causing a decrease in the retinal response to light and photoreceptor degeneration (Zhao et al., 2011). RPE dedifferentiation and hypertrophy are associated with mTOR activation. However, rapamycin treatment partially counteracts RPE dedifferentiation and photoreceptor degeneration (Zhao et al., 2011). Likewise, in primary RPE fetal human cells, inhibition of glycolysis promotes RPE differentiation, whereas OXPHOS inhibition causes downregulation of RPE-specific genes, indicating that OXPHOS metabolism is necessary to keep RPE identity. In the human cell line ARPE-19, DCA prevented EMT induced by TGFβ (Shukal et al., 2020). The authors found that TGFβ induced the expression of EMT markers, wound healing response, and ERK phosphorylation, which were reduced by DCA (Shukal et al., 2020). Altogether, these observations appear contradictory to those of our study, which may be because of differences in RPE origin. RPE explants were collected from chicken embryos and were studied in the context of cell reprogramming. This RPE is not completely differentiated and retains neural competence, a necessary characteristic to be able to respond to FGF2 and reprogram into the neural retina (Tangeman et al., 2022). However, overall, our study, together with others, highlights the importance of mitochondrial metabolism in RPE function, physiology, identity, and regenerative potential.
Eye disorders, such as diabetic retinopathy, proliferative vitreous retinopathy and AMD, are multifactorial diseases characterized by RPE dysfunction and degeneration (Yang et al., 2021). Accumulation of connective tissue, a process known as fibrosis, is commonly observed in vitreoretinal diseases and is caused by EMT of RPE cells (Hiscott et al., 1999). Several studies have shown that ROS, TGFβ and TNFα are drivers of EMT in RPE cells (Boles et al., 2020; Yang et al., 2020). Furthermore, glucose imbalance is strongly correlated with the development of diabetic retinopathy and EMT of RPE cells (Cheung and Wong, 2007; Che et al., 2016), indicating that metabolic state might play an active role in EMT and vitreoretinal diseases. Therefore, our data provide new insights into the importance of metabolism in RPE cell decisions and suggest a potential role for metabolic dysfunction in vitreoretinal disease progression. We show that RPE reprogramming into neural retina requires glycolytic metabolism (Fig. 8A), and that inhibition of PDKs during this process affects RPE cellular fate by activation of an EMT program, a mechanism that we speculate could be mediated by overproduction of acetate, ROS as well as regulation of EMT-related proteins, such as ALKs and SMADs by acetylation (Fig. 8B).
Proposed mechanisms for the role of glycolysis and PDKs in directing cell fate decisions during RPE reprogramming. (A) Plastic RPE requires glycolytic metabolism to reprogram into neural retina. (B) PDK inhibition may induce the synthesis of acetate by PDH, a metabolite that may be utilized in the cytoplasm to regulate the activity of EMT key factors such as SMAD and ALK by acetylation, a process turning on an EMT program that requires FGF2 and redirect RPE fate into mesenchyme. Ac, acetyl group.
Proposed mechanisms for the role of glycolysis and PDKs in directing cell fate decisions during RPE reprogramming. (A) Plastic RPE requires glycolytic metabolism to reprogram into neural retina. (B) PDK inhibition may induce the synthesis of acetate by PDH, a metabolite that may be utilized in the cytoplasm to regulate the activity of EMT key factors such as SMAD and ALK by acetylation, a process turning on an EMT program that requires FGF2 and redirect RPE fate into mesenchyme. Ac, acetyl group.
MATERIALS AND METHODS
Chicken embryos, RPE explant culture, and small molecule treatments
Fertilized white leghorn chicken eggs were obtained from Michigan State University. Eggs were incubated in a humidified rotating incubator at 37°C. At day 4 of development (E4, HH Stage 24; Hamburger and Hamilton, 1951), RPE sheets were collected and cultured as previously described, with slight modifications (Sakami et al., 2008). Briefly, RPE tissue was dissected from chicken embryos at E4, with a small amount of associated mesenchyme, in modified HBSS buffer (without calcium chloride, magnesium sulfate, and sodium bicarbonate, supplemented with 5 mM HEPES and 0.6% D-glucose). RPE sheets were washed in HBSS solution (140 mM NaCl, 5 mM KCl, 1 mM CaCl2, 0.4 mM MgSO4·7H2O, 0.5 mM MgCl2·6H2O, 0.3 mM Na2HPO4, 0.4 mM KH2PO4, 4 mM NaHCO3 and 6 mM glucose) and cultured in 500 µl of explant medium in 24-well plates. The plates were incubated in an orbital shaker at 50 rpm (3-D, Fixed Tilt Platform Rotator, Grant Instruments) at 37°C with 5% CO2. An explant medium was prepared as described previously (Sakami et al., 2008). The explant medium consisted of DMEM/F12 (HyClone, SH30026), supplemented with 0.9% D-glucose (Sigma-Aldrich, G-7021), 0.1125% NaHCO3 (Sigma-Aldrich, S5761), 20 mM HEPES (Boston BioProducts Inc., BB-2076-K), 5% fetal bovine serum (FBS) (Fisherbrand, FB12999102) and antibiotics (Gibco, 15-240-062). The glucose- and pyruvate-depleted medium was prepared by supplementation of DMEM with no glucose (Thermo Fisher Scientific, 11966025), 20 mM HEPES, 5% FBS and antibiotics, and used in the indicated experiments. In addition, media depleted of three major carbon sources (glucose, pyruvate and glutamine; 3C-media), were prepared using DMEM, with no glucose, no Phenol Red and no glutamine (Thermo Fisher Scientific, A14430), supplemented with 20 mM HEPES, 5% FBS and antibiotics, and used in the indicated experiments. FGF2 (R&D Systems, 3718-FB-025) was added at a concentration of 100 ng/ml as specified. All the explants used for experiments were maintained in glucose- and glutamine-depleted medium as well as 3C-media, and were collected and rinsed three times in HBSS modified buffer (140 mM NaCl, 5 mM KCl, 1 mM CaCl2, 0.4 mM MgSO4·7H2O, 0.5 mM MgCl2·6H2O, 0.3 mM Na2HPO4, 0.4 mM KH2PO4 and 4 mM NaHCO3) to ensure that glucose was not replenished by the tissue previous cell culture conditions. Sodium dichloroacetate (Sigma-Aldrich, 347795-10G) 1 M, 2DG (Sigma-Aldrich, D-8375-1G) 1 M, and N-acetyl cysteine (Sigma-Aldrich, A9165-5G) 0.5 M stock solutions were prepared in water and added directly to the cell media, as specified. The pH of the NAC stock was adjusted to 7 using NaOH. A stock of D-glucose (Sigma-Aldrich, G-7021) was prepared in water at a final concentration of 2 M and used for the indicated experiments. Pyruvate (HyClone, SH30239.01) and GlutaMAX (Gibco, 35050079), a substitute for L-glutamine, were used in the indicated experiments.
Immunoblotting
Three RPE explants were used per biological sample. RPE explants were collected in 60 µl of lysis buffer (Tris-HCl 62.5 mM, SDS 2%, pH 6.8) with protease (Roche, 1183617001) and phosphatase (Thermo Fisher Scientific, A32957) inhibitors. The samples were sonicated, and protein quantification was performed using the Bradford assay (Thermo Fisher Scientific, 1863028). Ten micrograms of protein was separated by electrophoresis on 4-20% Mini-PROTEAN TGX Precast Protein (Bio-Rad, 4561096) and transferred to PVDF membranes (Bio-Rad, 1704274). The membranes were blocked with EveryBlot Blocking Buffer (Bio-Rad, 12010020) for 10 min and incubated overnight at room temperature with primary antibodies diluted in EveryBlot Blocking Buffer at the indicated concentrations (Table S1). After overnight incubation, the membranes were washed three times, 5 min each, with 1× Tris-Buffered Saline, 0.1% Tween 20 Detergent (TBST) buffer and incubated with secondary antibodies diluted in TBST at the indicated concentrations (Table S1). When the secondary antibody was HRP-linked, SuperSignal chemiluminescent substrate was used (Thermo Fisher Scientific, 43580) following the manufacturer's instructions. The membranes were scanned in a ChemiDoc MP imaging system (Bio-Rad). Images obtained from the ChemiDoc MP were analyzed and quantified using Image Lab version 6.1.0 (Bio-Rad). Actin, selected as a housekeeping protein, was detected using a monoclonal primary antibody and a fluorescence-conjugated secondary antibody (Table S1). It was detected together with other target proteins only when the other primary antibody corresponded to a different host species (Table S1). Because the primary antibodies to detect the total and phosphorylated isoforms of ERK and PDH were developed in the same host species, it was not possible to detect both isoform proteins precisely from the same blot. Therefore, two western blots were run at the same time using the same loading mix of each sample. One of the blots was used to detect the total protein and the other for the phosphorylated isoform. Following this procedure, we could accurately determine the phosphorylated status of ERK and PDH.
Glucose and lactate determination
A pool of three RPE explants was used per biological sample. RPE explants cultured for 48 h, either in explant medium or 3C-media, were collected in a 1.7 ml tube on ice. The leftover media were removed by pipetting, and three washes of 2 min each with 150 µl of cold PBS were performed. Then, 60 µl of cold inactivation solution (0.6 N HCl) was added and the explants were sonicated for 1 min at intervals of 10 s. Immediately after, 60 µl of cold Neutralization Solution (1 M Tris Base) was added and the sample was vortexed for 20 s and sonicated for 30 s. The explant lysates were stored at −80°C for posterior glucose or lactate determination. Glucose was measured using the Glucose-Glo Assay (Promega, J6021), following the manufacturer's instructions. Briefly, 5 ml of Glucose Detection Reagent was prepared according to the manufacturer's protocol and 50 µl of this was pipetted into each well of a 96-well assay plate (Costar, 3603) and a glucose standard curve was prepared using 1.8 µg/µl glucose solution at the following concentrations: 0, 5, 50, 250, 1000, 3000 and 5000 ng/µl. Then, 50 µl of explant lysates were added, including the standards, and the plate was incubated for 10 min. The luminescence was recorded using a SpectraMax iD plate reader (Molecular Devices). The total glucose level of the sample was calculated using the linear regression equation of the standard curve. All the samples were measured in duplicate. A pool of three explants was enough to obtain only one measurement in duplicate. Five biological samples were analyzed by condition. For lactate determination, the same protocol used for glucose was used, except that the kit was Lactate-Glo Assay (Promega, J5022).
Histology, immunostaining and EdU detection
Explants used for EdU detection were exposed to EdU (Invitrogen, A10044) for 1 h prior to collection by adding 5 µl of EdU 2 mM to 500 µl of cell media. Explants just used for immunodetection were not exposed to EdU. Then, RPE explants were washed thrice with PBS for 5 min each wash and fixed with 4% paraformaldehyde for 15 min. After fixation, explants were washed three times with PBS and embedded in O.C.T. Compound (Fisher Scientific, 4585) and cryosections of 10 µm thickness were obtained using a cryotome (Cryostar NX50, Thermo Fisher Scientific) and collected on Superfrost Plus microscope slides (Fisher Scientific, 22-037-246). For Hematoxylin and Eosin staining, the slides were rinsed with PBS and histology staining was performed following a previously described protocol (Luz-Madrigal et al., 2014). For immunofluorescence and EdU detection, the slides with tissue sections were rinsed three times with PBS and incubated for 5 min in 5% saponin (Sigma-Aldrich, S7900) solutions following three washes of PBS. For EdU, Click-iT solutions were prepared according to the kit protocol (Invitrogen, C10337), and each slide was covered with 150 µl of Click-iT solution for 30 min. Then, the Click-iT solution was removed, and the explants were washed three times with PBS. Slides were then blocked with 200 µl donkey serum (Sigma-Aldrich, D9663) prepared at 10% in PBS with 0.1% Tween 20 Detergent (PBST) for 1 h. Then, the blocking solution was removed and 150 µl primary antibody (Table S1) was prepared in blocking solution and added to incubate overnight at 4°C. The primary antibody was removed, and three washes of PBS were performed, followed by three washes of PBST. Then, the slides were incubated in 100 µl of secondary antibody prepared in a blocking solution for 2 h. Next, three washes with PBS, for 5 min each, were performed, and the slides were incubated for 5 min in 100 µl of DAPI solution 1 µg/µl (Sigma-Aldrich, 10236276001) before finishing with three more washes of PBS. If the samples were just processed for immunofluorescence, EdU detection steps were omitted. The slides were mounted with Fluoromount (Sigma-Aldrich, F4680) and coverslipped. For imaging, confocal images were obtained using a Zeiss LSM 710 Laser Scanning Confocal System (Jena) using a 20×/0.80 NA=0.55 WD objective lens or EC Plan-Neofluar.
Mitochondrial activity determination
RPE explants were cultured for 48 h in an explant medium and treated with the indicated conditions. Afterwards, 250 µl of the media was removed and replaced with 250 µl of working solution and incubated at 37°C with 5% CO2 for 1 h. The working solution was prepared diluting 20 µl of 500X MitoLite dyes (AAT Bioquest, 22676) in 10 ml of HBSS buffer. After incubation, the stained explants were washed with PBS three times, 5 min each, and transferred to 48-well cell culture plates for imaging. Fluorescence microscopy images were obtained using a Leica Stellaris 8 confocal microscope set with 10×/0.40 PL APO objective. Total relative intensity fluorescence was measured using ImageJ software (Version 2.14.0/1.54f) and normalized to the area of the explant. Four explants were analyzed per condition.
RNA isolation and RT-qPCR gene expression
RPE explants were collected in 200 µl DNA/RNA Shield buffer (Zymo Research, 1220-25) and stored at −20°C. Each biological sample was a pool consisting of three explants. Total RNA was isolated using the Quick-RNA Microprep Plus Kit Microprep (Zymo Research, R1051) following the manufacturer's instructions. Total RNA samples were analyzed for quantity and quality using NanoDrop ND-2000 Spectrophotometer (Thermo Fisher Scientific) and Agilent 2100 Bioanalyzer (Agilent Technologies), respectively. Two-hundred nanograms of RNA was used as a template to synthesize cDNA using QuantiTect Reverse Transcription kit (QIAGEN, 205313) according to the manufacturer's instructions. The synthesized cDNA was diluted at 1:10 ratio with pure water and 2 µl of the cDNA dilution were used for the quantitative PCR (qPCR) reaction. The final qPCR reaction contained 2 µl of diluted cDNA, 10 µl of TB Green® Advantage® qPCR Premix (Takara Bio, 639676), and 500 nM of each primer adjusted to 20 µl with water. qPCR reactions were set up in duplicate in the Rotor-Gene Q thermocycler 5 plex (QIAGEN). Primers reported here were designed using Primer BLAST (https://www.ncbi.nlm.nih.gov/tools/primer-blast/; Table S2) and obtained from IDT Technologies. The comparative ΔΔCt method was used to determine the relative gene expression levels compared with the housekeeping gene (RPLP0). Four biological samples were used for each condition.
RNAseq library preparation and sequencing
Purified RNA was collected in triplicate, as described above, and quantified using the Qubit 4 and Qubit RNA HS Kit (Thermo Fisher Scientific, Q32852). RNA integrity was validated using the Agilent 6000 Pico Kit (Agilent, 5067-1513). For each sample, 100 ng of total intact RNA was used with the NEBNext® Poly(A) mRNA Magnetic Isolation Module (NEB, E7490S), and the enriched mRNA was prepped for sequencing with the NEBNext® Ultra™ II Directional RNA Library Prep with Sample Purification Beads (NEB, E7765S) according to the manufacturer's instructions. Thirteen PCR amplification cycles were used, and indexing was performed using single index oligos (NEB, 7710/7730). The final amplified libraries were validated using the Agilent High Sensitivity DNA Kit (Agilent, 5067-4626) and quantified using the Qubit dsDNA HS Kit (Thermo Fisher Scientific, Q32851). Samples were pooled in equimolar ratios before sequencing across two lanes of Illumina HiSeq X Ten at the Novogene sequencing core using 150 base pair paired end reads.
RNA-seq data analysis
Raw reads were analyzed using FastQC v0.11.9 (https://www.bioinformatics.babraham.ac.uk/projects/fastqc/) and MultiQC v1.9 (Ewels et al., 2016) for quality assessments. Adapters and low-quality bases were removed with Trim Galore v 0.6.4_dev (https://zenodo.org/doi/10.5281/zenodo.5127898) with the parameters –stringency 3 –paired –length 36. Trimmed reads were aligned to the chicken genome GRCg6a using the STAR alignment tool v2.7.5b (Dobin et al., 2013) and parameters –quantMode GeneCounts TranscriptomeSAM –runThreadN 30 –limitBAMsortRAM 60000000000 –readFilesCommand zcat –genomeLoad LoadAndKeep –outSAMstrandField intronMotif –outSAMtype BAM SortedByCoordinate. The genome index for star was generated with the parameters –runMode genomeGenerate –sjdbOverhang 149 –genomeSAindexNbases 13, and splice sites were incorporated from Ensembl release 106 (Cunningham et al., 2022). Gene counts were generated with Stringtie v2.1.4 (Pertea et al., 2015) using the parameters –rf -e, and genes with fewer than ten raw counts were filtered from the analysis. Differential expression testing was performed using DESeq2 v1.34.0, and genes with P-adjusted value ≤0.05 and |log2(fold change)|≥1 were considered differentially expressed (Love et al., 2014). For the interaction analysis, the formula design used employed a 7-term model: ∼FGF2+DCA+time+FGF2:time+DCA:time+FGF2:DCA:time+DCA:FGF2. To test whether the effect of FGF2 was different across DCA levels at 48 h, the DESeq2 results () function was used to add the parameter estimates FGF2Yes.DCAYes and FGF2Yes.DCAYes.time48h. Pathway enrichment analysis was performed with g:Profiler with chicken genes converted to human orthologs (Kolberg et al., 2020; Raudvere et al., 2019). Multiple hypothesis testing for enriched pathways was performed using the g:SCS algorithm within g:Profiler.
Statistical analysis
Experiments measuring relative gene expression by RT-qPCR or RNAseq (Figs 1C, 3G-I, 4C, 5B,F, 7C-E; Figs S3A, S4, S5B, S5F, S6E) were log(2)-transformed and then analyzed using ANOVA. Specific comparisons were made, and these P-values were adjusted to control for multiple comparisons by the false discovery rate (FDR) procedure (Benjamini and Hochberg, 1995), with the exception of analyses comparing treatments with control (Fig. 1C; Fig. S4), in which case we used Dunnett's multiple comparisons procedure. EdU and pHH3 quantification (count responses; Figs 2E and 4E; Fig. S6D) were analyzed using negative binomial regression (Lindén and Mäntyniemi, 2011; White and Bennetts, 1996; Hoef and Boveng, 2007), and multiple comparisons were also corrected using FDR. Immunoblotting experiments, which involved ratios related to PDH/pPDH, pERK/ERK and tubulin/actin (Fig. 4H; Fig. S6G), were also log(2)-transformed to stabilize unit-to-unit variance of these ratios, before analyzing the responses via ANOVA and FDR-adjusted P-values. Glucose and lactate determinations (Figs 3E,F and 4F) and the experiment characterizing mitochondria activity (Fig. S6I) were analyzed similarly. One of the key assumptions for the use of the linear model is that each observation has the same variance. In this study, there are a large number of statistical tests performed based on ANOVA models, and, unsurprisingly, there are some instances of what seems to be visual evidence of nonconstant variance across experimental groups, even after log(2) transformation. We acknowledge that because of this some of the resulting P-values may only be approximately valid. However, our conclusions are not based upon any particular P-value, but on the weight of the results as a whole. Thus, even if some of the experimental data do not perfectly meet the model assumptions, the analysis given here still provides evidence in support of our basic contentions. See supplementary Materials and Methods for detailed statistical analyses and Table S3 for raw data. All statistics were performed in the R environment. For data plotting, R environment or Prism 10.0.1 for Mac were used.
Acknowledgements
We acknowledge and thank the staff (Dr Andor Kiss and Ms. Xiaoyun Deng) of the Center for Bioinformatics and Functional Genomics (CBFG) at Miami University for instrumentation and computational support. We also thank Zachery Oestreicher of the Center for Advanced Microscopy and Imaging (CAMI) at Miami University for providing microscope instrumentation support. We wish to thank Chet Closson from the University of Cincinnati for his support and technical assistance in the use of a Leica Stellaris 8 confocal microscope. We also thank Erika Grajales-Esquivel for her managerial assistance and critical reading of the manuscript.
Footnotes
Author contributions
Conceptualization: J.R.P.-E., K.D.R.-T.; Methodology: J.R.P.-E.; Formal analysis: B.S.; Investigation: J.R.P.-E., J.A.T., M.P.-N., H.S.; Data curation: J.A.T.; Writing - original draft: J.R.P.-E.; Writing - review & editing: J.R.P.-E., K.D.R.-T.; Visualization: J.R.P.-E.; Supervision: K.D.R.-T.; Funding acquisition: K.D.R.-T.
Funding
This work was supported by a National Eye Institute grant (R01 EY026816), a Miami University Rapid Grant, and the John W. Steube endowed Professorship to K.D.R.-T. Further support was provided by a National Institute of Neurological Disorders and Stroke grant (F99 NS129167 to J.A.T.). Support was also derived from the Undergraduate Research Award Miami University program to M.P.-N. and H.S., as well as the Miami University Honors Program and Undergraduate Summer Scholars Miami University program to M.P.-N. and H.S., respectively. Deposited in PMC for release after 12 months.
Data availability
RNAseq data discussed in this publication have been deposited in the NCBI Gene Expression Omnibus (Edgar et al., 2002) under accession number GSE210526.
Peer review history
The peer review history is available online at https://journals.biologists.com/dev/lookup/doi/10.1242/dev.202462.reviewer-comments.pdf
References
Competing interests
The authors declare no competing or financial interests.