VEGF is an autocrine/paracrine neuroprotective factor for injured retinal ganglion neurons

Vascular endothelial growth factor-A (VEGF) is the angiogenic factor promoting the pathological neovascularization in age-related macular degeneration (AMD) or diabetic macular edema (DME). Evidences have suggested a neurotrophic and neuroprotective role of VEGF, albeit in retina, cellular mechanisms underlying the VEGF neuroprotection remain elusive. Using purified adult retinal ganglion cells (RGCs) in culture, we demonstrated here that VEGF is released by RGCs themselves to promote their own survival, while VEGF neutralization by specific antibodies or traps drastically reduced the RGC survival. These results indicate an autocrine VEGF neuroprotection on RGCs. In parallel, VEGF produced by mixed retinal cells or by mesenchymal stem cells exerted a paracrine neuroprotection on RGCs. Such neuroprotective effect was obtained using the recombinant VEGF-B, suggesting the involvement of VEGF-R1 pathway in VEGF-elicited RGC survival. Finally, glaucomatous patients injected with VEGF traps (ranibizumab or aflibercept) due to either AMD or DME comorbidity, showed a significant reduction of RGC axon fiber layer thickness, consistent with the plausible reduction of the VEGF autocrine stimulation of RGCs. Our results provide evidence of the autocrine neuroprotective function of VEGF on RGCs is crucially involved to preserve injured RGCs such as in glaucomatous patients.


Results
VeGf is a paracrine neuroprotective factor for RGcs. We previously showed that conditioned medium isolated from mixed retinal cell cultures (CM-M) can promote the survival of enriched adult rat RGCs in cultures 17 . This effect occurred either in the presence or in the absence of B27 supplement and persisted until 21 days in vitro (DIV) (Fig. S1). We identified the neuroprotective factors contained in this CM-M using the Luminex assay for rat trophic factors. All samples (n = 4) of CM-M contained a large amount of VEGF-A 164 , but no other factors known to promote RGC survival (i.e. BDNF, G-CSF, and GM-CSF; Table 1). The presence of VEGF was confirmed and quantified by ELISA technique (106.1 ± 22.6 pg/ml, mean ± SEM, n = 9). Mixed retinal cultures are composed of a monolayer of macroglial and microglial cells, upon which are found retinal neurons (mainly cone photoreceptors and bipolar cells; Fig. S2). We attributed the origin of VEGF synthesis to microglial cells because these purified cells can produce substantial levels of VEGF in most cultures after only 48 h (17.8 ± 21.7 pg/ml; n = 22; Fig. S3). P urified macroglial cells (including Müller glial cells) did not produce VEGF (Fig. S3), in contrast to results reported in previous studies on isolated Müller cells 18 or a human Müller cell line (MIO-M1) 9 .
In parallel, we found that conditioned medium from mesenchymal stem cells (CM-MSC), which contained VEGF, can also promote RGC survival (Fig. 1A-E). We assessed whether the VEGF contained in these two different conditioned media (CM-MSC and CM-M) contributed to RGC survival. First, purified adult rat RGCs were incubated with the CM-MSC, in the presence of a specific antibody against VEGF-A 164 (rabbit polyclonal IgG, 0.5 µg/ml). This anti-VEGF antibody completely suppressed the RGC survival effect produced by the CM-MSC (p < 0.01, Fig. 1C,D). Indeed, there was a linear correlation (r 2 = 0.78) between the VEGF concentration in each CM-MSC and its efficiency in promoting RGC survival (Fig. 1E). Secondly, we obtained a significant reduction of the CM-M-induced neuroprotection (p < 0.01) when the anti-VEGF antibody was added to retinal CM-M (Fig. 1F). However, the neuroprotective effect of CM-M was not completely suppressed and remained significant (p < 0.01 as compared to control) in presence of the anti-VEGF. This observation may result from the presence of other neurotrophins in CM-M (e.g. NGF or GDNF, other than BDNF which were undetectable in CM-M) since the detection of growth factors via our Luminex assay was not exhaustive. Finally, we tested a control rabbit polyclonal IgG targeting another protein (anti-NF200), and we found that the CM-M-enhanced RGC survival was not significantly affected (Fig. 1G). This result indicates that the VEGF present in the two different conditioned media is partially responsible for the trophic effects on RGCs.
VeGf is an autocrine factor produced by RGcs to promote their own survival. Surprisingly, the antibody directed against VEGF-A 164 has also reduced per se the survival of purified RGCs by 34.8% when they were cultured in a commercial VEGF-free culture medium (p < 0.05, see Fig. 1F). This result suggests that RGCs themselves can release VEGF into culture medium. Indeed, when RGCs were initially seeded at 8000 cells per well, measurable concentrations of VEGF-A 164 were detected in the culture medium at 6 and 12 DIV by ELISA, whereas VEGF was non-detectable at 1 DIV (not shown). From 70 to 75% of harvested supernatants contained detectable VEGF-A 164 after 6 to 12 DIV, with concentrations reaching 81.3 ± 12.7 pg/ml at 6 DIV.
At 6 DIV, the increase of seeding density from 8 × 10 3 to 50 × 10 3 cells/well improved the RGC survival ( Fig. 2A,B) and increased the percentage of VEGF-positive supernatants from 12 to 100%, respectively (Fig. 2C). VEGF concentrations also substantially and significantly increased with higher seeding density (p < 0.001, Kruskal-Wallis test; Fig. 2C). More importantly, the VEGF-A 164 concentrations highly correlated with the actual number of surviving RGCs (p < 0.001, Spearman test), the curve being best fitted with an exponential regression (r 2 = 0.65) (Fig. 2D). These measures are therefore consistent with a correlation between the level of free VEGF (0-600 pg/ml) and RGC survival (0-6000 RGCs per well). The expression of VEGF in healthy RGCs as in cultured RGCs was confirmed by anti-VEGF immunolabelling of retinal sections and purified RGC cultures (Fig. S4) in agreement with previous reports 9,19 . These measures are also consistent with the detection of VEGF-A mRNA in embryonic RGCs 20 and our adult rat RGCs (data not shown). Altogether, these results indicate that VEGF is an autocrine factor for RGCs.
The importance of VEGF for RGC survival was finally verified using various VEGF trapping molecules: (i) a mouse monoclonal anti-human VEGF (anti-pan, non-selective for VEGF-A isoforms), (ii) a rabbit polyclonal antibody against murine VEGF-A 164 (i.e. selective for isoform 164), and (iii) ranibizumab (the Fab fragment from a humanized mouse monoclonal anti-VEGF-A, non-selective for isoforms). These anti-VEGF proteins all induced a statistically significant dose-dependent reduction of RGC survival after 6 DIV (Fig. 2E). The efficacy of ranibizumab became significant at relatively high concentrations (above 250 µg/ml; Fig. 2E). Both the specific Table 1. Growth factors in conditioned medium harvested from mixed retinal cultures. Growth factors were measured in control and conditioned medium harvested from mixed retinal cultures (CM-M). VEGF concentration, expressed as pg per ml, is the mean ± SEM from independent preparations (n = 4), all showing detectable VEGF levels (4/4). ND, no-detectable; VEGF, vascular growth factor; BDNF, brain derived growth factor; G-CSF, granulocyte colony-stimulating factor; MG-CSF, granulocyte-macrophage colonystimulating factor. www.nature.com/scientificreports/ rabbit polyclonal antibody against murine VEGF-A 164 (2 µg/ml), as well as ranibizumab (at least above 250 µg/ ml), prevented the detection of VEGF produced by RGCs since 100% of samples displayed undetectable VEFG levels ( Fig. 2F). In contrast, the pan antibody (non-selective for VEGF-A isoforms) reduced the proportion of samples with a detectable level of VEGF but did not significantly affect the mean VEGF concentrations in the medium, although it affected its neuroprotective effectiveness, probably due to the reduction of free VEGF (not bound to an antibody). The mean values of VEGF concentrations were calculated, including undetectable values defined as zero. We found that if anti-pan did not significantly decrease the VEGF concentrations whatever the doses used (0.5, 1. or 2.0 µg/ml), the selective anti-VEGF-A 164 induced a significant decrease in VEGF concentrations in the culture medium at the two doses tested (0.5 and 2.0 µg/ml, p < 0.05 as compared to control condition). Similarly, ranibizumab induced a dose-dependent decrease of VEGF concentrations in culture medium, which appeared statistically significant at the dose of 250, 500 and 1,000 µg/ml, respectively (p < 0.05 as compared to control condition).

VEGF-elicited RGC survival involves VEGF-R1 pathway.
Using the recombinant VEGF-A 164 protein (from rat), we consolidated that VEGF per se promotes the RGC survival. At the 10 ng/ml concentration the recombinant protein significantly increased the RGC survival (+ 41.8%, p < 0.01), whereas incubation at 1 ng/ml produced a slight increase (+ 14.3%, Fig. 3A), but not statistically significant. To further assess the pharmacological pathway of the VEGF-elicited RGC survival, we investigated the involvement of VEGF receptors in VEGFelicited RGC survival. Because VEGF-A binds at two distinct receptors named VEGF-R1 and VEGF-R2 21 , we used pharmacological agents in order to discriminate which of these receptors are activated by VEGF to promote the RGC survival. The VEGF-R2 receptor antagonist, ZM 323,881 (ZM) allows discerning the activity of VEGF-R1 versus -R2 receptors. Co application of ZM (5 nM) in the culture medium with the recombinant VEGF-A (10 ng/ml) did not affect the VEGF-mediated increase in RGC survival (ZM + VEGF-A: + 31.4%, VEGF-A: + 38.5%; Fig. 3B), while incubation of ZM alone did not show any significant changes on RGC survival (+ 8.0%). These data suggested that the RGC survival is rather mediated through the VEGF-R1 activation. To verify this hypothesis, we applied a recombinant VEFG-B protein, which stimulates selectively the VEGF-R1. Addition of RGCs were seeded at initial density of 8000 cells/well. ***p < 0.001, **p < 0.01, and *p < 0.05 as compared to the control group, and ## p < 0.01 compared between indicated groups (Kruskal-Wallis ANOVA followed by a Dunns post-hoc test).
In order to check the absence of VEGF-R3 implication in this neuroprotective efficacy, we finally applied the recombinant VEGF-D protein, which activates VEGF-R3 and in some proportion VEGF-R2 (4). However, VEGF-D (0.5 ng/ml) failed to increase the RGC survival (+ 1.7%; Fig. 3C). These results are consistent with the involvement of VEGF-R1 activation by VEGF, to promote RGC survival by VEGF-A. We could show expression of both VEGF-R1 and VEGF-R2-at the mRNA level-in freshly purified RGCs (n = 3), as shown by RT-PCR experiments (Fig. 3D,E). We concluded that RGC survival relies on the VEGF-R1 activation.
Anti-VEGF treatments alter the retinal nerve fiber layer (RNFL) thickness in glaucomatous patients. One report has already suggested that anti-VEGF therapy could negatively affect RGC survival 13 . www.nature.com/scientificreports/ Such negative effects could be more important in patients with already compromised RGCs. Thus, we focused our investigations by measuring the RNFL thickness of glaucomatous patients, with either AMD or DME comorbidity, as compared to non-glaucomatous patients, all receiving anti-VEGF therapy. There was a statistically significant decrease of RNFL thickness in the treated eyes (injected eyes) from the glaucomatous group, which appeared during the third month of treatment (Fig. 4D). We can observe a quick decrease of the RNFL thickness during the first 3 months corresponding to the induction phase (one intravitreal injection [IVI] each month) whereas during the next phase the decrease is continuing, but less intensely (but remaining significant), probably due to a less intensive treatment (3.4 and 3.2 IVI respectively for non-glaucomatous and glaucomatous patients during the following 9 months). In contrast, the RNFL thickness of the fellow control eye was stable (Fig. 4D). This decrease was not associated with significant changes in intra-ocular pressure (IOP) levels in either the injected or control eyes (Fig. 4B).  Scientific RepoRtS | (2020) 10:12409 | https://doi.org/10.1038/s41598-020-68488-z www.nature.com/scientificreports/ There were no differences in RNFL thickness or IOP in either eye of the control group (non-glaucomatous patients; Fig. 4C,E). These observations suggest that anti-VEGF therapies could affect RGC survival in patients with already injured RGCs.

Discussion
This study provides evidence that VEGF is a neuroprotective factor for RGCs through an autocrine mechanism. We demonstrate in vitro that enriched adult rat RGCs cultured under serum-deprivation-a situation that mimics ischemic conditions-can release VEGF to promote their own survival. To our knowledge, such ability of RGCs to produce themselves VEGF has never been described yet for these retinal neurons, except in one study showing detectable VEGF amounts in supernatant harvested from RGC-5 cell line 22 ; the nature of this cell line remaining controversial. Hence, this autocrine mechanism of neuroprotection may underlie the VEGF-elicited neuroprotection of RGCs, as observed in vivo following optic nerve transection 15 and in glaucomatous animal models 16 .
In retina, such neuroprotective action of VEGF involving an autocrine mechanism was already described for Müller cell line (MIO-M1) and retinal pigment epithelium (ARPE-19) which can secrete VEGF to promote their own survival 8,9 . An autocrine mechanism was also suggested to explain the neuroprotective action of VEGF on embryonic cortical neurons 23 , and may be also involved in the VEGF neuroprotection of motoneurons 24 .
In addition, we found that VEGF is released by microglial cells suggesting that it also exerts neuroprotection through paracrine pathway. Other studies attributed the paracrine action of VEGF through a secretion from retinal pigment epithelium (RPE) or from Müller cells. Indeed, VEGF is particularly expressed by RPE 25 and was also found in inner retina (see Fig. 4S). In human retina, VEGF immunoreactivity was specifically expressed in Müller cell bodies 26 . Moreover, we demonstrate here that VEGF is among the essential factors of MSC secretome conferring a direct RGC protection. A paracrine action of VEGF may also explain part of the therapeutic effect provided by bone marrow-derived MSCs injected into eyes following optic nerve injury or glaucoma 27,28 .
Our pharmacological investigations reveal that VEGF-elicited RGC survival is mediated through the VEGF-R1 stimulation. Indeed, VEGF neuroprotection was not blocked by the addition of VEGF-R2 antagonist, while it was reproduced by incubation of VEGF-B which only stimulates VEGF-R1 (see 4 ). VEGF-B was shown to stimulate the survival of RGC-5 subjected to oxidative-stress, and it inhibits the apoptosis in the retina induced by axotomy, NMDA intra-vitreous injection or ischemia 29 , suggesting the neuroprotective effect of the VEGF-R1 signaling. On the other hand, it has been shown that VEGF can protect hippocampal neurons against glutamate excitotoxicity through the activation of VEGF-R2 signaling 5,6 . Furthermore, VEGF-R2 blockade aggravates neuronal degeneration in brain 22,30 , supporting the involvement of VEGF-R2 in VEGF neuroprotection. VEGF-R1 and VEGF-R2 belong to the tyrosine kinase receptor family and both trigger the PI3K/AKT and the MEK/ERK pathways involved in the VEGF neuroprotection 5 . Accordingly, VEGF-R1 and VEGF-R2 can be differentially implied in the neuroprotective effects of VEGF according to the cell types, but in RGCs, VEGF-R1 seems to be preferentially implied.
VEGF production and release by adult rat RGCs demonstrate its role as an autocrine factor for these spiking retinal neurons. In vivo, VEGF is highly expressed by healthy RGCs as observed here by immunolabeling on retinal sections, while a lower expression is found in other cell types (Fig. S4). VEGF release could play a major physiological role because the extracellular space is small and RGC density is very high in the RGC layer. During development, the release of VEGF by RGCs may contribute to the development of the inner retinal vasculature 20 .
We previously showed that cultured RGCs in low nutritive conditions (without B27 serum) have features highly reminiscent of ischemic conditions 17 . Therefore, such culture conditions can be considered as representative of glaucomatous ischemic conditions. As a consequence, these results suggest that the VEGF synthesis is also occurring in RGCs from glaucomatous patients. However, VEGF immunolabeling also demonstrated VEGF synthesis in healthy RGCs 14,15 . Although VEGF concentrations in cell cultures and vitreous are difficult to compare, it is surprising to note that VEGF concentrations released by our cultured RGCs were in a similar range or higher (< 600 pg/ml) than those measured in vitreous of control patients (< 150 pg/ml) [31][32][33] . It remains unclear if retinal pathologies affecting RGC survival like glaucoma alter VEGF release, thus participating in the dysregulation of ocular blood flow 34 . Apart from its effect on blood flow, reduced VEGF release due to RGC loss could affect the survival of the remaining RGCs in a vicious cycle.
In fact, VEGF contents in the vitreous of glaucomatous patients were only measured in a recent study on patients affected by neovascular glaucoma, showing vitreous VEGF concentrations raising to 220 pg/ml 35 . VEGF contents were higher in vitreous from patients affected by retinal neovascularization, like proliferative diabetic retinopathy, going up to 500 pg/ml, and more (> 1000 pg/ml) according to various studies [31][32][33]36 . Interestingly, anti-VEGF treatment allowed to drastically reduce the VEGF concentrations in vitreous of these patients 33 .
Regarding the healthy subjects, the vitreous VEGF concentrations do not exceed 150 pg/ml (from 1.7 to 146 pg/ Table 2. Patient characteristics. *Glaucomatous patients received anti-glaucomatous treatment through eye-drop consisting in either monotherapy, bi-therapy, tri-therapy, or quadri-therapy with drugs from the 4 distinct classes (Beta blockers, Alpha-adrenergic agonists, Prostaglandins or Carbonic anhydrase inhibitors). Non glaucomatous patients did not receive any eye treatment.
Age (mean ± S.D.) Gender (% female) Glaucoma treatments* (numbers of drugs; mean ± S.D) Glaucoma group (n = 10) 73.8 ± 7.9 60 1.8 ± 0.6 Non glaucoma group (n = 10) 73.0 ± 9.0 70 -Scientific RepoRtS | (2020) 10:12409 | https://doi.org/10.1038/s41598-020-68488-z www.nature.com/scientificreports/ ml). Although, these VEGF concentrations are in the same range or below those measured in RGC culture media (up to 600 pg/ml), it is unlikely that RGCs provide the only source of VEGF. Indeed, we showed that microglial cells can also synthesize VEGF (Fig. S3) and we previously confirmed other studies showing VEGF production by RPE cells 37 . In addition, others reported that a Müller cell line can also secrete VEGF amounts reaching more than 1000 pg/ml 9 . Therefore, vitreous VEGF is likely resulting from the production by all these retinal cells in vivo.
The chronic administration of anti-VEGF therapies may enhance VEGF loss, leading to deleterious consequences for vulnerable RGCs, such as in glaucoma. This safety concern for anti-VEGF therapies is consistent with a previous study showing a significant loss of RNFL thickness after 12 months of intravitreal anti-VEGF injections in AMD patients 13 . This conclusion is further supported by a recent retrospective study showing that patients with continuous unilateral anti-VEGF treatment showed a significant decrease in both RNFL and RGC layer thickness during at least 24 months of follow-up 38 . However, other studies in patients with only AMD have not found any correlation between RNFL changes and anti-VEGF therapy [39][40][41] . In a recent study 42 which looked at patients with both glaucoma and AMD, the authors did not find a correlation between decrease nerve fiber layer and treatment. However, the main concern with this study is that they used a macular analysis of ganglion cells or RNFL, and the macular edema due to AMD may induce artifacts on the measures of thickness for each macular layer (ganglion cells, RNFL) with artificial increase, thus probably masking a thinning of these layers. Given these overall contradictory data, future clinical trials are needed to assess further if an anti-VEGF therapies sustainably affect RGC survival and thus the RNFL thickness in patients.
Here, we provide further clinical evidence for this deleterious effect of anti-VEGF therapy, but in a glaucomatous population with already compromised RGCs. Repeated injection of anti-VEGF for AMD or DME induced a significant reduction of RNFL thickness over a period of 12 months in these glaucomatous patients. These results strongly suggest that VEGF is a major autocrine factor for adult human RGCs. Further studies are needed to demonstrate the potential therapeutic interest of VEGF receptor agonists in retinal diseases. Mixed retinal cultures were maintained until confluence (10-12 DIV) and then used to prepare conditioned medium from retinal mixed cultures (CM-M). Primary cultures of RGCs were isolated using an immunopanning technique, according to the protocol adapted from that used on adult rats 43,44 . Purified RGCs were seeded at various densities and cultured in Neurobasal-A medium (Invitrogen, Carlsbad, CA, USA) containing glutamine and supplemented with B27 (Invitrogen).
Automated counting of viable RGcs. RGC counting was realized after 6 DIV on alive cells labeled with calcein-AM (producing a green fluorescence; Invitrogen). Briefly, RGGs cultured in 96-well plate, were incubated in a calcein-AM solution (1.3 µg/ml, diluted in Neurobasal-A medium) for 1 h into the incubator (humidified chamber, 37 °C, 5%CO 2 ). Calcein-positive RCGs were then automatically counted after a frame per frame scanning of the full well surface (objective X10), by using an inverted epifluorescence microscope (TiE, Nikon, Champigny-sur-Marne, France) equipped with a motorized stage and a CCD camera (Roper Scientific, Evry, France).

Measurement of growth factors.
A microparticle bead-based multiplex assay was used to measure the growth factors VEGF, BDNF, G-CSF, and GM-CSF in supernatants from mixed retinal cells, using the Luminex Scientific RepoRtS | (2020) 10:12409 | https://doi.org/10.1038/s41598-020-68488-z www.nature.com/scientificreports/ 200 technology. VEGF-A 164 was specifically measured in supernatants from cell cultures using ELISA technique. Supernatants were harvested from mixed, glial or microglial or RGC cultures and frozen at − 80 °C until use. VEGF-A164 was then measured in supernatants using the "Rat VEGF Quantikine ELISA Kit" (R&D system) according the instructions of the kit.
patients and clinical examinations. We conducted a prospective single-center real-life study on 20 patients (10 non-glaucomatous and 10 stabilized glaucomatous patients) treated by intravitreal injections (IVI) of VEGF traps (ranibizumab or aflibercept; at least six intravitreal injections-6.4 vs 6.2-during 12 months). Patient characteristics are summarized in Table 2. The eyes receiving IVI were included in the "injected" group, whereas the follow eyes (without IVI) were included in the "non-injected" group. Informed consent was obtained from all patients in agreement with the Declaration of Helsinki for research involving human subjects. This study did not include a placebo group because it consists in a real-life study, and not a randomized, placebo-controlled and double-blind trial.
The study was approved by the Ethics Committee of the French Society of Ophthalmology (IRB 00008855 Société Française d'Ophtalmologie IRB#1).
Statistics. All data are expressed as means ± SEM. In ELISA experiments for detection of VEGF in culture medium, undetectable values (below the detection threshold) were defined as zero and included in the mean calculations.
The Mann Whitney test was used to compare means of 2 groups. When more than 2 groups were compared, a non-parametric Kruskal-Wallis analysis of variance was performed and followed, in case of significance by a Dunns post-hoc test to compared means between each group. For clinical observations, the Friedman test was used for analysis of variance for repeated measures and followed by a Dunns post-hoc test to compared means between each group. For all statistical analysis, differences were considered significant at *p ≤ 0.05, **p ≤ 0.01 and ***p ≤ 0.001.