Context-dependent effects of inflammation on retina regeneration

Inflammation is required for the proliferation of Müller glia (MG) into multipotent progenitors (MGPCs) in the injured fish and avian retinas. However, its function in retina regeneration has not been fully understood. Here we investigated the role of inflammation in three different retinal regeneration paradigms in zebrafish (stab-injury, NMDA-injury and insulin treatment). We first show that different types of immune cells and levels of inflammatory cytokines were found in the retinas of these paradigms. Though zymosan injection alone was insufficient to induce MG proliferation in the uninjured retina, immune suppression significantly inhibited MGPC formation in all three paradigms. Enhancing inflammation promoted MGPC formation after stab-injury, while exhibiting a context-dependent role in the NMDA or insulin models. We further show that proper levels of inflammation promoted MG reprogramming and cell cycle re-entry after stab- or NMDA-injury, but excessive inflammation also suppressed MG proliferation in the latter model. Finally, inflammation differentially affected neuronal regeneration in various injury paradigms. Our study reveals the complex and context-dependent role of inflammation during retinal repair in fish and suggests accurate inflammation management may be crucial for successful retina regeneration in mammals.


Introduction
Degenerative retinal diseases such as macular degeneration and retinitis pigmentosa often lead to blindness, and currently, there is no therapy available to restore the lost vision in these patients. Unlike mammals, lower vertebrates such as teleost fish are able to repair their damaged retinas and restore vision [1,2]. Retina regeneration in zebrafish relies on Müller glia cells (MG) in the retina. After retinal injury, zebrafish MG undergo a reprogramming process and proliferate into multipotent progenitors (MG-derived progenitor cells, MGPCs) [3,4]. These MGPCs further proliferate, migrate to different retinal layers and eventually differentiate into major types of retinal neurons [5]. Although mammals cannot spontaneously regenerate their damaged retina, their MG share many characteristics with those of zebrafish [6], and there is increasing evidence that they have neurogenic potential and could be used as an endogenous source for retinal repair [7][8][9]. A comprehensive understanding of the mechanisms governing retinal repair in fish could therefore provide important clues for retina regeneration in mammals in the future.
After tissue injury or infection in the central nervous system (CNS), inflammation is among the earliest responses detected in both mammals and fish [10][11][12]. Tissue inflammation is characterized by the expression of proinflammatory cytokines and chemokines, which recruit resident immune cells or those from the blood stream to the site of tissue injury. These activated immune cells will modulate the local environment by further secreting cytokines and chemokines, and removal of cell debris and metabolites [13]. It has been shown that acute inflammation in the CNS can activate tissue stem cells and promote their proliferation and neurogenesis [12,14,15]. On the contrary, chronic inflammation is believed to be harmful because a prolonged inflammatory response could cause damage to the neurons and even lead to degenerative diseases [16,17].
Inflammation and microglia are known to be required for MGPC formation and photoreceptor regeneration. In the postnatal chicken retina, microglia ablation suppressed the subsequent MGPC formation after NMDA-injury [18]. In the larval zebrafish, co-ablation of photoreceptor and microglia delayed MG proliferation and slowed rod cell replacement [19]. A recent study from our laboratory showed that microglia-mediated inflammation activated the mTOR signaling in the MG, and promoted MGPC formation after a stab-injury in adult zebrafish [20]. Similarly, immune suppression or microglia ablation impaired neuronal regeneration in other retinal injury models in adult zebrafish [21,22]. However, most of these studies utilized a single injury model and lacked a comprehensive and comparative analysis of the roles of inflammation in different paradigms of retina regeneration. Furthermore, little is known about the effect of enhancing inflammation on MGPC formation and retina regeneration. In this study, we utilized three different paradigms (stab-injury, NMDA-injury, and insulin administration) and investigated the impact of immune manipulation on MG reprogramming and proliferation, MGPC formation and neuronal regeneration with or without injury in adult zebrafish. Our results show a surprisingly complex and context-dependent role of inflammation in the retinal repair in fish, and suggest that the optimization of inflammatory response according to injury paradigm, background inflammation level, and degree of tissue damage should be considered for a successful retina regeneration in mammals.

Animals
Wild type (AB) and Tg(mpx:GFP) transgenic zebrafish [23] were obtained from China Zebrafish Resource Center (CZRC). The Tg(mpeg1:EGFP) fish [24] was obtained from the Institute of Neuroscience, Chinese Academy of Sciences. All fish were treated in accordance with the Guidelines for Animal Use and Care at Nantong University. Fish was maintained at 28 °C for a light/dark cycle of 14 h/10 h. Adult zebrafish of both sexes (4-6 months of age) were used in this study.

Retinal injury paradigms and insulin administration
The method of mechanical injury has been described previously [25]. Briefly, fish were anesthetized in 0.02% Tricaine methane sulfonate (Sigma-Aldrich, E10521) in system water. Under a dissecting microscope, the right eye was gently rotated in the socket and stabbed four times (once in each quadrant) through the sclera with a sterile 30 gauge needle to the length of the bevel. The uninjured left eye served as a negative control.
For chemical lesions, 1 μl of NMDA (N-methyl-D-aspartate, Sigma-Aldrich, M3262) was injected intravitreally through the front of the eye at the first day of experiment. The default concentration of NMDA is 50 mM unless otherwise indicated. Care was taken not to damage the retina or other tissues.
For insulin-induced retina regeneration [26], 2 μg of bovine insulin (Solarbio Life Science, China, I8040) in PBS was injected intravitreously through the front of the eye. Care was taken not to damage the retina or other tissues. Fish received daily insulin injection for 4 days before the sacrifice. PBSinjected eyes were used as a negative control.

Quantitative PCR (qPCR)
The methods for qPCR have been described previously [25]. At least three fish were used for each group and qPCR was performed in triplicate using the Taq Pro Universal SYBR qPCR Master Mix (Vazyme, China, Q712-02). The relative mRNA expression was deciphered using the ΔΔCt method and normalized to rp113 mRNA levels. Primers used in the study are listed in Table 1.

Immune suppression and activation
In order to suppress the retinal immune response, zebrafish were immersed in dexamethasone (Dex, Sigma-Aldrich, D4902) in system water or solvent control (0.1% DMSO) for a week prior to injury or insulin administration. The default concentration of Dex is 15 mg/L unless otherwise indicated. Dex treatment continued after retinal injury or insulin injection for 4 days, and then, fish were returned to the system water for recovery. Dex or DMSO was changed daily, and zebrafish were fed two hours before the water change. To enhance the immune response in the retina, fish received an intravitreal injection of indicated amount of zymosan A (Zym, Sigma-Aldrich, Z4250) at the time of retinal injury or on the first day of the experiment.

TUNEL assay
The TUNEL experiment was performed using an In Situ Cell Death Detection Kit, Fluorescein (Roche Applied Science, 11,684,795,910) according to manufacturer's instruction. The DNAse I-treated sections were used as a positive control.

BrdU incorporation and lineage tracing
To label proliferating cells in the retina, 20 μl of 20 mM BrdU (Sigma-Aldrich, B5002) was injected intraperitoneally into the zebrafish 4 h before sacrifice. In order to trace the lineage of proliferating MGPCs, a pulse of BrdU (20 μl of 20 mM) was intraperitoneally injected at 4 days post-injury (dpi) before sacrificing at 30 dpi or at indicated time points to examine their distribution and differentiation.

Microscopy and cell counting
A Zeiss Imager M2 upright microscope (Carl Zeiss AG) equipped with an Axiocam 506 monochrome camera was used to capture the fluorescence images of retinal sections or flat mounts. Fluorescence images were captured with a 10 × or 20 × objective, and cells were counted using the ImageJ software (Cell Counter plugin) as previously described [27]. For cell counting, at least 3 fish were used for each group. In the stab-injury paradigm, target cells at the injury site were counted on each cryosection, and their total numbers were added. In the NMDA or insulin paradigms, target cells in the entire retina were counted and summed from each section. For the quantification of neuronal regeneration in the NMDA paradigm, target cells were counted in a segment of retina that spans 300 μm in length.

Quantification and Statistics
All of the experiments were performed at least in triplicate and repeated at least two times. The number of samples used in each experiment was indicated in the figures or figure legends. Student's t test (2 tails, unpaired) was used for single comparison, and a one-way analysis of variance (ANOVA) was used, followed by a Tukey test for multiple comparisons. Error bars represent standard error (SEM). Statistical significance was defined as p < 0.05.

Inflammatory responses in three different retinal regeneration paradigms
To assess the inflammatory responses within each paradigm, we first characterized the dynamics of inflammatory cells in the retina. For this purpose, retinal microglia/macrophage was labeled by an intravitreous injection of the isolectin IB4 [20,28]. In the retina of Tg(mpeg1:EGFP) transgenic zebrafish in which GFP was specifically expressed by microglia/macrophage [24], the majority of IB4 signals co-localized with GFP ( Fig. 1A), further confirming the specificity of this staining. In the uninjured retina, a small number of IB4 + cells could be seen scattered in the retina and vitreous (Fig. 1B, left panel, Fig. 1E). Following stab injury, a large number of IB4 + cells accumulated at the injury site at 2 days post-injury (dpi) (Fig. 1B,C; p < 0.001), indicating a robust microglia/macrophage response in the retina as shown previously [20]. Intravitreal PBS injection into the uninjured eye did not significantly increase the total number of IB4 + cells in the retina and vitreous (Fig. 1B,D,E; p = 0.356), though a small increase in the vitreous was observed at 2 d (Fig. 1E, p < 0.001), indicating that some of the IB4 + cells in the retina migrated to the vitreous after PBS injection. Two days after the NMDA-injury or insulin-treatment, significantly increased numbers of IB4 + cells compared with the PBS control could be seen in the retina and vitreous (Fig. 1D 1D, lower panels). We have previously shown in the stab-injury model that the accumulated IB4 + cells could be seen as early as 6 h post injury (hpi), and they peaked at 2 dpi before declining at 4 dpi [20]. Interestingly, similar time courses were observed following NMDA-injury or insulintreatment, though the initial rise in the number of IB4 + cells occurred slightly later than following stab-injury (12 hpi vs 6 hpi, Fig. 1F). These results indicate that an acute microglia/ macrophage response with similar time courses occurred in the retina and vitreous in the three paradigms. It has been previously shown that neutrophil infiltrates into the retina after stab injury, but not after sterile retinal cell loss in zebrafish larvae [19]. To characterize the potential neutrophil response in the three paradigms, we took advantage of the Tg(mpx:GFP) transgenic zebrafish in which GFP was selectively expressed in neutrophils but not microglia/macrophages [23]. 3 h after stab injury, a large number of GFP + cells were observed infiltrating the injured retina of the Tg(mpx:GFP) zebrafish ( Fig. 2A,B). The number of neutrophils at the injury site rapidly decreased at 6 hpi ( Fig. 2A-C), and by 12 and 24 hpi, only a few neutrophils could be seen in this region (Fig. 2B,C and data not shown). In contrast to the early neutrophil response, an increased presence of microglia/macrophage within the injured retina was not observed until 6 hpi (Fig. 2B). Interestingly, numerous neutrophils were also observed accumulating in the space between the peripheral retina and iris ( Fig. 2A, white boxes) after stab injury at 3 hpi, and some appeared to migrate along the inner retinal surface toward the injury region ( Fig. 2A, upper left panel). After 6 hpi, neutrophil accumulation in the peripheral region decreased rapidly in a manner similar to that of the injury site ( Fig. 2A and data not shown). In contrast to the stab-injury paradigm, no obvious neutrophil infiltration into the retina or vitreous was found after NMDA injury or insulin injection at 3 and 6 hpi, though some neutrophils were seen in the iris ( Fig. 2A, arrowheads). These findings demonstrate a different neutrophil response between the stab-injury and other two models.
To investigate if the inflammation levels are comparable in the three models, RNA samples were extracted from the whole retina at different time points (0 h, 6 h, 12 h, 1 d, 2 d and 4 d), and qPCR was performed to determine the expression of typical inflammatory cytokines including il1b, il6, il11a and il11b. A typical inflammatory cytokine TNFα was not chosen due to its very low expression in the stab injury and insulin models. In the control retina, there was a relatively low induction of il1b and il6 but not il11a/11b after PBS injection ( Fig. 2D-G), suggesting that the injection assay itself may cause a weak inflammatory response. qPCR showed that inflammatory cytokines were strongly induced in the retinas of all three models, but their highest levels were observed following NMDA-injury ( Fig. 2D-G). In insulin-treated retinas, the expression levels of il1b and il11a were higher than those of the stab injury, while their levels of il6 and il11b were largely comparable ( Fig. 2D-G). Interestingly, qPCR also revealed that the expression of inflammatory cytokines peaked significantly earlier than the number of IB4 + cells (Fig. 1F,  Fig. 2). Specifically, the expression of inflammatory cytokines peaked at 6 h (NMDA and insulin groups) or 12 h (stab-injury group). In contrast, the number of IB4 + cells peaked at 2 days in all three paradigms. This inconsistency indicates that cell subtypes or expression patterns of microglia/macrophage may vary between early and later time-points following injury or growth factor treatment.

Inflammation is essential for MGPC formation in all three paradigms
To investigate whether inflammation is required for the formation of MGPCs in each paradigm, glucocorticoid dexamethasone (Dex, 15 mg/L) was used to suppress retinal inflammation and previous studies have validated its efficacy in adult zebrafish retina [20,22]. As it has been shown that 98% of the BrdU + cells in the inner nuclear layer ( Figure S1) [20]. These results indicate that inflammation is essential for MGPC formation in all three models.
We next ask if inflammation is required for the cell cycle re-entry of MG. Since almost all of the INL BrdU + cells are proliferating MG at 2 dpi in the stab-injury paradigm [3], we labeled these cells by BrdU and PCNA immunostaining at this time point. Immunofluorescence microscopy showed that Dex-treatment significantly reduced the number of INL BrdU + or PCNA + cells at 2 dpi ( , p = 0.032), suggesting that inflammation is necessary for MG proliferation. This finding is consistent with the result from our previous study, in which we showed that Dex-treatment significantly reduced the number of proliferating MG, as indicated by number of GFP + /BrdU + columns in the Tg(1016tuba1a:GFP) transgenic zebrafish [20]. Furthermore, qPCR was performed to examine the expression of regeneration-associated genes (RAGs) and cell cycle-related genes in control and Dex-treated retina. Our results showed that Dex-treatment significantly reduced the expression of RAGs such as ascl1a, lin28a, hbegfa, igf1, socs3a and socs3b, as well as several cell cycle-related genes including ccna2, ccnb1, ccnd1, ccne1 and cdk2 (Fig. 4D). Together these results suggest that inflammation regulates the expression of RAGs and cell cycle-related genes and is necessary for MG proliferation and MGPC formation.
Nevertheless, since there was no definitive markers to separate MG from MGPCs, Dex-treatment may also decreased progenitor proliferation.

Zymosan injection alone is insufficient to elicit a regenerative response in the intact retina without causing cell death
The above data show an essential role of inflammation in the regenerative response in zebrafish retina in different paradigms. To investigate whether inflammation alone is sufficient to induce the reprogramming and proliferation of MG, fish received intravitreal injection of zymosan A (Zym) to activate inflammation in the intact retina. The efficacy of Zym to induce inflammation in the zebrafish CNS has been reported [12], and our previous study also validated its capability to enhance inflammation in stab-injured zebrafish retinas [20]. To better understand the effect of Zym-treatment on retinal cell survival and inflammation, different doses of Zym were tested, and TUNEL experiment showed that 20 and 40 μg/eye of Zym caused significant cell death in the intact retina (  . Therefore, Zym doses at 10 μg/eye or lower (0.1 or 1 μg/eye) were used in the following experiments of this study. We first characterized the inflammatory time course after Zym injection. qPCR of typical inflammatory cytokines showed that intravitreal injection of 10 μg of Zym stimulated an acute inflammatory response which peaked at 12 h in the uninjured retina (Fig. 5B). Importantly, Zym injection significantly increased the number of IB4 + cells in the vitreous ( Fig. 5C; [0.1, 1, 10 μg], p = 0.005, < 0.001, and < 0.001, respectively), as well as the expression of inflammatory cytokines in the retina in a largely dose-dependent manner (Fig. 5D). These findings suggest that Zym injection could dose-dependently induce the inflammatory response in the zebrafish retina.
We next investigated the effect of Zym injection on MG proliferation and MGPC formation in the intact retina. BrdU immunofluorescence showed that all three doses of Zym failed to induce MG proliferation or MGPC formation at 2 days and 4 days after injection (Fig. 5E,F 1 and 10 μg], p = 0.961, 0.908, and 0.716, respectively). Importantly, qPCR showed that Zym injection failed to induce the expression of key RAGs such as ascl1a, lin28a and hbegfa (Fig. 5G), many of the cell cycle-related genes ( Fig. 5H), and critical cytokines required for MG reprogramming (Fig. 5I). Zym injection did cause a significant induction of the Jak-Stat reporter socs3a (Fig. 5G), suggesting that inflammation is sufficient to activate this signaling in the retina. Together these results indicate that inflammation alone may be insufficient to drive MG into reprogramming and proliferation in the intact retina.

Enhancing inflammation influences MGPC formation in a context-dependent manner
As immune suppression inhibited MGPC formation, we predicted that enhancing inflammation would increase MGPC formation. Indeed, intravitreal injection of Zym (10 μg/eye) significantly increased the number of MGPCs in the stab-injury model at 4 dpi (Fig. 3A,B, p< 0.001), consistent with the result from immune suppression (Fig. 3A,B). Surprisingly, enhancing inflammation by intravitreous injection of the same dose of Zym following NMDA-injury or insulin-treatment significantly inhibited MGPC formation (Fig. 3C-F, p< 0.001 for both groups). These results were unlikely caused by a significant change in retinal cell death, as Zym injection did not affect the number of TUNEL + cells in the NMDA-treated retina ( Figure S1), and very few TUNEL + cells were seen in  Figure S1). The unexpected results prompted us to further examine the influence of inflammation on MGPC formation. Since our previous data showed that the levels of typical inflammatory cytokines following NMDA-injury or insulin-treatment were higher than that of stab-injury in the retina (Fig. 2), we hypothesize that Zym's inhibitory effect on MGPC formation in these two models was caused by an excessive level of inflammation. To answer this question, various doses of Zym were used and their effects on MGPC formation in the three paradigms were examined. In the stab-injury model, Zym injection significantly increased the number of MGPCs at 4 dpi in a dose-dependent manner (Fig. 6A,D; [0.1, 1, 10 μg], p = 0.012, p < 0.001 and p < 0.001, respectively). In the NMDA-injury paradigm, low doses of Zym (0.01, 0.05, 0.1 and 1 μg/eye) dosedependently increased MGPC formation (Fig. 6B,E; [0.01, 0.05, 0.1 and 1 μg], p = 0.007, 0.003, 0.006, and 0.002, respectively). However, this promoting effect vanished at the dose of 2.5 μg (Fig. 6E; p = 0.415) and was reversed at the dose of 10 μg (Fig. 6B,E, p < 0.001). A similar dual effect of Zym on MGPC formation was also observed in the insulin model (Fig. 6C,F). These results support our hypothesis and indicate a dual role of inflammation in MGPC formation in the NMDA and insulin models.
To investigate if the effect of enhancing inflammation also depends on the degree of retinal damage, retinas were treated with different concentrations of NMDA (0.5 mM, 5 mM, or  (Fig. 7C,D; [50, 5 and 0.5 mM], p = 0.003, p < 0.001, and p = 0.012, respectively). 10 μg of Zym significantly inhibited MGPC formation in retinas treated with 50 mM of NMDA (Fig. 7C,D, p < 0.001), consistent with the data above (Fig. 6E). Surprisingly, this inhibitory effect of 10 μg of Zym was completely reversed in retinas treated with 5 mM or 0.5 mM of NMDA (Fig. 7C,D; [5 mM and 0.5 mM], p < 0.001 and p = 0.008, respectively). The reversal of 10 μg Zym's effect was unlikely caused by a marked decrease in the overall background level of inflammation, as qPCR showed no significant difference in the levels of typical inflammatory cytokines between the 5 mM and 50 mM NMDA groups, and 4 out of 6 of these cytokines were also comparable between the 0.5 mM and 50 mM NMDA groups (Fig. 7E). These results indicate that the influence of inflammation on MGPC formation also depends on the degree of retinal damage. Taken together, these data demonstrate a complex function of inflammation following retinal injury, and indicate the influence of inflammation on MGPC formation is dependent on the injury paradigm, the level of inflammation, as well as the degree of retinal damage.

Influence of enhancing inflammation on MG reprogramming and proliferation
To further understand how enhancing inflammation influences MGPC formation, the cell cycle re-entry of MG was Since it has been shown that most of the proliferating cells in the INL at 2 dpi are MG [3,20], this suggests enhancing inflammation promoted the entry of more MG into the cell cycle as shown by the PCNA staining at 2 dpi, but they had not reached the S phase yet as indicated by the BrdU incorporation. Indeed, BrdU staining at 3 dpi showed a significant increase of INL BrdU + cells in Zym-treated retinas (Fig. 8C,D, p < 0.001), consistent with our notion. This finding is also supported by the result from our previous study, in which we showed that Zym-treatment significantly increased the number of GFP + column that was generated from a proliferating MG in the Tg(1016tuba1a:GFP) transgenic line [20]. Importantly, qPCR analysis showed that Zym injection significantly increased the expression of key RAGs such as ascl1a, lin28a, hbegfa, igf1 and socs3a, as well as multiple cell cycle-related genes in the retina (Fig. 4D). These results indicate that in the stab-injury paradigm, enhancing inflammation promotes MG reprograming and drives more MG from quiescence into the cell cycle, which in turn generates more MGPCs at 4 dpi.
We then investigated the effect of enhancing inflammation on MG proliferation by a low or high dose of Zym (0.1 μg or 10 μg) following NMDA injury (50 mM). In line with the aforementioned data, intravitreous injection of 10 μg Zym significantly reduced the number of INL BrdU + or PCNA + MG at 2 dpi (Fig. 8E,F; [BrdU], p = 0.009; [PCNA], p < 0.001), suggesting a high level of inflammation prevents MG from entering the cell cycle in the NMDA model. Interestingly, no difference in MG proliferation measured by PCNA or BrdU immunofluorescence was observed between the 0.1 μg Zym group and the control at 2 dpi (Fig. 8E-F, [BrdU], p = 0.097; [PCNA], p = 0.167; Fig. 8G). We speculate there was a delay in MG response to Zym in the NMDA model compared with the stab injury, therefore the cell proliferation was further examined at 3 dpi. Indeed, 0.1 μg of Zym significantly increased the number of INL proliferating cells at 3 dpi (Fig. 8H,I; [BrdU], p = 0.027; [PCNA], p = 0.013). To further explore if this phenomenon was caused by the cell cycle entrance from a new population of MG, proliferating MG were labeled by a pulse of BrdU at 2 dpi, and their progeny cells were  (Fig. 8J). In the control retina, most of the INL PCNA + cells were also BrdU + at 3 dpi (94.6 ± 0.9%, Fig. 8K), suggesting that the BrdU pulse labeled the majority of proliferating MG at 2 dpi. In contrast, a significantly lower proportion of INL PCNA + cells at 3 dpi were BrdU + in retinas received 0.1 μg of Zym (Fig. 8K, 61.9 ± 2.2%, p < 0.001). The presence of many BrdU-/PCNA + cells in this group suggests the cell cycle entry of a new MG population between 2-3 dpi that were not labeled by the 2 dpi BrdU pulse. Nevertheless, we cannot exclude the possibility of an increased MGPC proliferation induced by Zym. Taken together, our findings indicate that enhancing inflammation following stab or NMDA injury may drive more MG into the cell cycle, but if excessive inflammation is elicited, it may also exhibit an opposite effect in the NMDA model.

Inflammation differentially regulates retinal neuron regeneration in various injury paradigms
To investigate the influence of inflammation on the regeneration of retinal neurons in the stab-injury paradigm, MGPCs were labeled with a pulse of BrdU at 4 dpi, and their distribution and differentiation were examined at 30 dpi (Fig. 9A). BrdU immunofluorescence showed that Dex-treatment significantly reduced the total number of BrdU + cells in the retina at this time point, while Zym injection had an opposite effect (Fig. 9A  Similar to the experiments above, we also labeled the proliferating MGPCs by a pulse of BrdU at 4 dpi in the NMDA-injury paradigm, and examined the influence of inflammation on their distribution and differentiation at 30 dpi. Immunofluorescence showed that the majority of BrdU + cells in the NMDA control group were in the ONL at 30 dpi (Fig. 10A,C). As expected, enhancing inflammation by 0.1 μg of Zym resulted in significantly more BrdU + cells (Fig. 10A,B, p < 0.001), with the most obvious increase in the INL (Fig. 10A,C; [ONL], p = 0.019; [INL], p < 0.001; [GCL], p = 0.655). Surprisingly, early immune suppression by Dex generated the largest number of BrdU + cells in the retina at 30 dpi (Fig. 10A,B, p < 0.001). Analysis of their distribution showed that Dex treatment significantly increased the number of BrdU + cells in the ONL and INL, with the most dramatic change observed in the latter layer (Fig. 10A,C;  As Dex treatment reduced the number of MGPCs in the NMDA model at 4 dpi (Fig. 3), we asked how immune suppression could produce the above result. To address this question, MGPCs from the three groups were labeled with BrdU at 4 dpi, and their number and distribution were examined at 6, 14 and 30 dpi (Fig. 10F, Figure S2). In the NMDA control and Zym-treated retinas, an initial increase of ONL BrdU + cells was observed from 6-14 dpi  and Zym-treated retinas decreased dramatically from 6-14 dpi, with partial recovery in the latter group from 14-30 dpi (Fig. 10F, middle panel, p < 0.001 for both groups; Figure S2). In contrast, there was an initial increase of INL BrdU + cells from 6-14 dpi, followed by a slight decrease from 14-30 dpi in Dex-treated retinas (Fig. 10F, middle panel; [14 d], p = 0.044; [30 d], p = 0.128; Figure S2). In the GCL, a gradual but significant decrease of BrdU + cells was found in all three groups during this time period (Fig. 10F, Figure S2). Taken together, these results indicate a dramatic decrease in the number of BrdUlabeled MGPCs from 6-30 dpi in the NMDA control and Zym-treated retinas, presumably due to cell death. Importantly, immune suppression prevented MGPC depletion in the ONL and INL, which may contribute to the best neuronal regeneration in the NMDA paradigm.

Discussion
Inflammation plays important roles in MG-dependent retina regeneration [18][19][20]22]. However, most of the previous studies employed a single injury paradigm, lacking comprehensive analysis and comparison of different models. In this study, we employed three different paradigms of retina regeneration in adult zebrafish, and investigated the influence of immune manipulation on MGPC formation and neuronal regeneration in these models. Our results show a context-dependent effect of inflammation on MG proliferation and retina regeneration, and may have important implications in designing future therapeutic strategies for mammalian retinal repair.
By IB4 staining, we found an acute microglia/macrophage response in all three models. The time course of microglia/macrophage accumulation was similar in these models, implying involvement of common mechanisms. Following NMDA injury, microglia/macrophages accumulated preferentially in the inner retina, correlating with a damage of inner retinal neurons. Similarly, ouabain treatment also resulted in microglia/macrophage accumulation in the damaged inner retina of adult zebrafish [29]. In our study, a marked difference in neutrophil response was found among the three models. Significant early neutrophil infiltration into the retina/vitreous was observed after stab injury, but not in the other two paradigms. This finding is consistent with a previous study in which neutrophils entered the retina following puncture injury but not after sterile retinal cell loss in zebrafish larvae [19]. Since the fish retina does not have a direct blood supply and neutrophils do not enter the retina in physiological conditions, it is likely that the stab injury provides a physical access to the retina. Excessive neutrophils recruitment to the site of injury has been traditionally thought to cause collateral tissue damage and chronic inflammation [30]. However, novel evidences are also emerging in recent years demonstrating their contribution to inflammation resolution and tissue repair [31,32]. Neutrophils may thus play a role in the stab injury response, and further functional investigation is required to address this question.
By immune suppression, we found inflammation was essential for MG proliferation and MGPC formation in all three paradigms in adult zebrafish. These results are consistent with previous studies carried out in chicken and fish [18][19][20][21][22], and highlight the beneficial effects of acute inflammation on retinal repair. Our further study showed that inflammation was required for the expression of key reprogramming factors ascl1a and lin28a [7,33], as well as important growth factors hbegfa and igf1 [26,34]. In both the stab-injury and NMDA-injury paradigms, enhancing inflammation with a proper dose of Zym drove more MG into the cell cycle, which in turn generated more MGPCs. These results indicate an essential role of inflammation in MG reprogramming and proliferation. Nevertheless, as no definitive markers to separate MG from MGPCs were currently available, we cannot rule out the possibility that inflammation also promotes progenitor proliferation. Immune manipulation at later time points, such as from 2 to 4 dpi when MGPCs are rapidly proliferating, will help clarify the influence of inflammation on progenitor proliferation.
Investigating the effect of inflammation demonstrated that Zym injection alone is insufficient to trigger a regenerative response in the intact retina without causing cell death. It is likely that inflammation-activated signaling pathways, such as mTOR [20], Jak-Stat, and MAPK [21] cooperate with other injury-induced signals to initiate the regeneration process. Though Zym injection caused a significant inflammatory response in the intact retina, it is possible that some aspects of this inflammation were different to that from injury. In adult zebrafish brain, however, Zym-induced sterile inflammation was sufficient to induce radial glia proliferation and subsequent neurogenesis [12]. The different response to Zym in the brain and retina could be attributed to the inherent differences between MG and radial glia, and/or distinct local environments. While Zym injection increased the number of L-plastin + cells in the zebrafish brain, it was unclear what types of leukocytes they were, and whether their secretome was distinct to that of microglia/macrophage in the retina. Continuous intravitreal injections of recombinant human IL-6 were shown to be sufficient to promote MG proliferation in the uninjured zebrafish retina [35]. In our study, Zym injection resulted in a significant increase of il6 expression in the uninjured zebrafish retina while no MG proliferation was observed. This discrepancy could be explained by a difference in the actual concentration and duration of IL-6 in the retina, and/or a different response to endogenous and recombinant IL-6 proteins. Additional experiments, such as daily Zym injections or treatment with endogenous zebrafish IL-6 proteins, are required to better understand the role of IL-6 in MG proliferation.
A key finding of this study is the complex and contextdependent effect of inflammation on retina regeneration. A previous study in zebrafish larvae also reported that the effect of immune suppression on rod replacement depends on the timing of treatment [19]. By manipulating inflammation in three different models, we show that the influence of inflammation on MGPC formation depends on many factors, including the injury paradigm, the level of inflammation, and the degree of retinal damage. The exact molecules responsible for the adverse effect of inflammation in the NMDA and insulin paradigms remain unclear at this point, but studies in other systems have shown that some inflammatory cytokines such as interferons, TNFα and IL-1β could inhibit the proliferation of neural stem/progenitor cells [36][37][38]. In a recent study in mice, microglia ablation increased the neurogenic ability of MG in Ascl1-induced retina regeneration after NMDA-injury, and the authors suggest that the role of inflammation in retina regeneration is mechanistically different in fish and mice [39]. However, results from our study likely explain the difference. We show that by reducing the Zym dose or the degree of tissue damage in the NMDA-injured zebrafish retina, we could actually increase MGPC formation. It's possible that high levels of inflammation suppressed Ascl1-induced regeneration in NMDAinjured mice retinas, whereas microglia ablation reduced the inflammation to a more beneficial level and thus increased neurogenesis.
The most unexpected result of this study is the opposite outcome of immune manipulation on the regeneration of retinal neurons from the stab-injury and NMDAinjury paradigms. In the stab-injury model, Zym injection increased the number of BrdU + regenerated retinal neurons. In contrast, the highest number of BrdU-labeled newborn neurons was found in Dex-treated retinas in the NMDA model. Lineage tracing in the NMDA control and Zym-treated retinas revealed a dramatic decrease of BrdU-labeled MGPCs from 6-30 dpi. The NMDA model possessed the highest level of inflammatory cytokine expression compared to the other two and, in particular, a sustained high level of IL-6 after injury. Since overproduced inflammatory cytokines are well known for their inhibitory effect on neurogenesis and neuronal survival [40,41], this highly inflammatory environment may cause the cell death of MPGCs and/or newborn neurons in the control and Zym-treated retinas. Consistent with this idea, an enhanced neuronal regeneration was found in Dex-treated retinas possibly due to improved survival of MGPCs and/or newborn neurons. Similarly, a recent study also found that late Dex treatment had regenerationenhancing effects in the retina of mmp9 mutant fish with elevated TNFα levels [22]. Given the fact that in our study fish were pre-treated with Dex for 7 days before NMDA injury, and Dex significantly reduced MGPC formation at 4 dpi, its final result in retina regeneration is indeed striking. It will be interesting to further investigate whether the combination of early Zym injection to expand the MGPC population and late Dex-treatment to improve cell survival will result in an even better regeneration in this model.
In summary, this study revealed a complex and contextdependent function of inflammation in retina regeneration. Inflammation clearly played an essential role in MG proliferation and MGPC formation, but in order to achieve the best regeneration, our results indicate that it needs to be optimized according to the context, including the injury paradigm, the inflammation level, and the degree of retinal damage. In some scenarios such as the NMDA-injury paradigm, immune suppression or early inflammation enhancement followed by late immune suppression may result in the best outcome. The findings of this study also suggest that in future retinal repairs in mammals including humans, accurate inflammation management may be key to successful regeneration.