Microglial dynamics after axotomy-induced retinal ganglion cell death

Animals undergoing surgery were anesthetized by intraperitoneal injection of a mixture of ketamine (60 mg/kg; Ketolar, Pfizer, Alcobendas, Madrid, Spain) and xylazine (10 mg/kg; Rompum, Bayer, Kiel, Germany). Analgesia was provided by subcutaneous administration of buprenorphine (0.1 mg/kg; Buprex, buprenorphine 0.3 mg/mL; Schering-Plough, Madrid, Spain). During and after surgery, the eyes were covered with an ointment (Tobrex; Alcon, S. A., Barcelona, Spain) to prevent corneal desiccation.

The animal groups are the following: (i) intact, no surgery; (ii) tracing controls, tracing from the superior colliculi (SCi), analyzed 7 days or 2 months later (9 weeks); (iii) axotomized, tracing from the SCi, and 1 week later optic nerve crush (ONC) or optic nerve transection (ONT), retinal analysis 2, 5, 9, 14 days or 2 months (9 weeks tracing) later. Injured retinas are the left ones, and contralateral retinas the right ones.

After the exposure of both SCi, a pledge of gelatin sponge soaked in 3% fluorogold (FG, Fluorochrome, LLC, USA) dissolved in 10% dimethyl sulfoxide/saline was applied onto both SCi 7 days prior optic nerve injury, or 7 days or 9 weeks before sacrifice (traced but otherwise intact retinas), following standard techniques in our laboratory [8, 23, 24, 34, 35].

Seven days after RGC tracing and according to previous studies, the left optic nerve was crushed or transected at 3 or 0.5 mm from the optic disc, respectively [2, 4, 34‐36]. With these settings, ONC elicits a slower RGC death than ONT [2, 34].

Unless otherwise stated, all reagents were from Sigma-Aldrich Quimica S.A. Madrid, Spain.

Animals were sacrificed with an intraperitoneal injection of an overdose of sodium pentobarbital (Dolethal, Vetoquinol; Especialidades Veterinarias, S.A., Alcobendas, Madrid, Spain). All animals were perfused transcardially with 0.9% saline solution followed by 4% paraformaldehyde in 0.1 M phosphate buffer.

The retinas (n = 4–12/group and time point, see tables) were dissected as flattened whole mounts as previously described [34]. Some eye cups (n = 3/group and time point) were cryoprotected in a series of sucrose gradients, embedded in optimal cutting temperature medium (OCT, Tissue-Tek, Sakura-Finetek, VWR, Barcelona, Spain), frozen, and sectioned in a cryostat at 14-μm thickness.

Double immunodetection of RGCs and MCs was carried out as previously described [2, 11, 21, 35, 37]. Healthy RGCs were detected using goat α-Brn3a (C-20; Santa-Cruz Biotechnology, Heidelberg, Germany) diluted 1:750 and microglial cells using rabbit α-Iba1 antibody (ionized calcium-binding adapter molecule 1; Wako Chemicals GmbH, Neuss, Germany) diluted 1:1000. Secondary detection was carried out with donkey α-goat and donkey α-rabbit (Alexa Fluor 594 and 488 respectively; Molecular Probes; Thermo Fisher Scientific, Madrid, Spain) diluted at 1:500. In some retinal sections, all nuclei were counterstained using antifading mounting media with DAPI (4′,6-diamidino-2-phenylindole, Vectashield mounting medium with DAPI; Vector Laboratories, Palex Medical, Barcelona, Spain).

Whole-mounted retinas where photographed focusing firstly on the GCL and secondly on the IPL. All images were acquired using an epifluorescence microscope (Axioscop 2 Plus; Zeiss Mikroskopie, Jena, Germany) equipped with a computer-driven motorized stage (ProScan H128 Series; Prior Scientific Instruments, Cambridge, UK) controlled by image analysis software (Image-Pro Plus, IPP 5.1 for Windows; Media Cybernetics, Silver Spring, MD). Retinal photomontages were reconstructed from 154 (11 × 14) individual images by zigzag tiling, as reported [2, 25, 38].

FG⁺RGCs in intact and contralateral to the lesion retinas, and Brn3a⁺RGCs in all retinas, were automatically quantified using our previously described methods [25, 34]. After injury, microglial cells became transcellularly labeled (PMCs: FG⁺Iba1⁺), being for the automated routine impossible to tell them apart from traced RGCs Thus, FG⁺RGCs were manually quantified (FG⁺Iba1⁻) and their total number estimated using previously reported methods [11]. Briefly, the retinas were divided into three concentric rings (central, equatorial, and peripheral) and four samples within each were selected (0.25 mm²/sample). Each sample corresponded to a retinal quadrant, therefore, three samples per quadrant and four samples per ring. The area of each ring was measured, and the total number of FG⁺RGCs in a given ring were estimated from the mean number quantified in its four samples, thus giving back the number of FG⁺RGCs in the central, medial, and peripheral retina. The total number of FG⁺RGCs per retina was the sum of the partial values. This method was validated in intact retinas, where FG⁺RGCs were quantified automatically and manually (for more details, see [11]).

All non-phagocytic MCs (Iba1⁺FG⁻) and PMCs (Iba1⁺FG⁺) were manually dotted on the retinal photomontages using Adobe Photoshop CS 8.0.1; (Adobe Systems, Inc., San Jose, CA), and the dots were automatically quantified (IPP software) [2, 8, 24, 39]. Quantitative data from all populations were exported to a spreadsheet application (Microsoft Office Excel 2003; Microsoft Corporation, Redmond, WA) for further analyses.

The spatial distribution of RGCs and MCs was assessed by the near neighbor algorithm as reported [2, 21, 24, 39, 40]. The position of each Brn3a⁺RGC or dot (microglial cell) in the photomontage was translated to an X,Y location using as reference the optic nerve (0,0 point). Maps were plotted using SigmaPlot (SigmaPlot 9.0 for Windows; Systat Software, Inc., Richmond, CA, USA). In addition, using a color code, these maps depict the number of neighbors around a given cell in a given radius (see Fig. 1 legend).

Data were analyzed with GraphPad Prism v.7 (GraphPad San Diego, USA). Each cell type was analyzed independently along time within each lesion and between lesions (one-way ANOVA, post hoc Tukey’s test). Differences were considered significant when p < 0.05. Data are presented as mean ± standard deviation. Correlation analyses (microglial cells vs. RGCs, or cells vs. time) were modeled using the same program and were considered significant when R ² (goodness of the fit) was ≥ 0.9.

As observed in the retinal section in Fig. 1a, SMCs are found in the GCL, IPL, and OPL. In the IPL, some MCs lie on the inner limit of the INL. Using flat mounts, we quantified their total number in the GCL or IPL (Fig. 1b, b’), which contain, respectively, RGC axons and somata, or their dendrites and dRGCs.

The total number of SMCs in the GCL and IPL is very similar (Table 1), although slightly higher in the GCL. The ratio RGC/MC is 6:1 if we take into account only MCs in the GCL, but if we add the MCs in the IPL, that is MCs in the GC complex, this ratio becomes 3:1. In both layers, SMCs are homogenously distributed (Fig. 1c, c’, Table 1).

In traced retinas, FG is observed filling the RGC cytoplasm, while Brn3a expression is nuclear (Fig. 1e). When the tracer is left for 7 days (standard procedure), there are usually no visible PMCs (Table 1). By 2 months, few PMCs are observed, both in the GCL and IPL. Note that at 2 months, the number of FG⁺RGCs decreases because FG is not a persistent label [39, 41, 42] but the number of Brn3a⁺RGCs remains unaltered (Fig. 1f–i, Table 1). In the right retinas, contralateral to the lesion, there is no RGC loss and the presence of PMCs is testimonial (Fig. 1i, j’, Table 1).

After ONC or ONT, traced RGCs disappear and PMCs appear (Fig. 2a–j). The morphology of MCs after the lesion changes from ramified to fusiform or ameboid. Fusiform or rod-like MCs are preferentially found in chains, along the RGC axons. Ameboid MCs are found more often engulfing a traced RGC which, normally, does not express Brn3a anymore (Figs. 2 and 3a).

As previously reported [2, 34], RGC loss after ONC is slower than after ONT, and this is observed with both identification methods (Table 2 and Fig. 3b, c). However, there are differences between the course of RGC death (Brn3a) and RGC clearance (FG).

Within each lesion, RGC loss is detected earlier with Brn3a than with FG. Indeed, half of the RGCs have died 5.6 days after ONC or 4.2 days after ONT, while half of the RGC clearance occurs 3 days later for both lesions (at 7.2 or 8.4 days respectively). Furthermore, RGC death adjusts to a segmental lineal regression with two phases of loss, one quick that lasts up to 7 days for ONC (daily loss of 7326 RGCs, m ₁) or to 5.8 days for ONT (m ₁, 9762 RGCs/day) and a slower one thereafter with a comparable rate of loss for both lesions (m2 values). For its part, RGC clearance is linear and the daily disappearance rate is similar between lesions (there are no significant differences in the slopes between ONC and ONT p = 0.59), albeit delayed for 1 day (see above) and less pronounced after ONC (5249 RGCs/day) than after ONT (5795 RGCs/day).

PMC appearance highly correlates with FG clearance (Fig. 3c), but not as much with RGC death (Brn3a⁺RGCs vs. PMCs, regression coefficient 0.84, not shown). Incidentally, the FG⁺RGCs vs. PMCs correlation tells us that for each 3288 or 3044 cleared RGCs, there is an increase of 1000 PMCs, meaning that one PMC is able to engulf ~ 3 RGCs. This ratio can also be estimated from the slope of FG⁺RGCs vs. time by the slope of PMCs vs. time (i.e., 5249/1789 = 2.9).

Quantitative analyses of cell topography show that more RGCs die in the central retina (Fig. 3d, first graph). Proportionately, though, their death is synchronous along the retina because the percent of loss is similar at any given distance from the optic nerve (Fig. 3d, second graph). In agreement, PMCs appear across the GCL, increasing in number with increasing post-lesion intervals, following the topographic pattern of RGC loss (Fig. 3d, compare the third graph with the first one). However, proportionally, the central retina bears a higher percent of PMCs than the periphery (Fig. 3d, fourth graph). This can be explained because there are more PMCs where more RGCs are dying, and this is clearly observed when the data are graphed as neighbor maps (Fig. 4); while RGC loss is diffuse, the emergence of PMCs seems centripetal being their density higher in the retinal areas where RGCs are more abundant (RGC distribution in an intact retina, Fig. 1g).

PMCs peak at 14 days (Table 2). At this time point, ~ 90% of MCs in the GCL are PMCs which, as abovementioned, are more abundant in the areas of higher RGC density and death (Fig. 5a, e). The remaining 10% are non-PMCs and are distributed fairly homogenously, except for a dense accumulation around the optic nerve and along the extreme periphery (Fig. 5a’, e’).

At its highest, the number of PMCs + non-PMCs in the GCL surpasses the number of SMCs found in intact retinas by ~ 10,000. In parallel, there is a uniform decrease of MCs in the IPL (around 50% of their initial population) that justifies half of the excess observed in the GCL, suggesting that there is microglial migration between these layers (Table 2, Fig. 5b–b”, f–f”). Indeed, PMCs are observed orthogonally placed in the IPL and INL from 9 days onwards (Fig. 6).

Two months after the lesions, when RGC death has slowed down, the number of MCs in the GCL has decreased and is numerically (Table 2) and topographically (Fig. 5c–c”, g–g”) normal, albeit most of these MCs are still transcellularly labeled. For its part, the IPL has partially recovered by the addition of PMCs which, most probably, come from the GCL (Table 2, Fig. 5d–d”, h–h”). However, the IPL, and therefore the GC complex, are deficient on 3000–5000 MCs compared to that on intact retinas.

In agreement with previous reports, MCs are homogenously distributed in the GCL and IPL of healthy retinas [46, 49]. This topography reflects their surveying function, explained by the continuous mobility of MCs and their multidirectional projections used to maintain the distance between surrounding cells [50].

In traced but otherwise intact rat retinas, the number of PMCs depends on the time of tracing, as it has been shown in mice [21]. In these retinas, we did not observe a significant loss of Brn3a⁺RGCs (at least up to 2 months). Thus, the presence of PMCs can be explained by phagocytosis of the FG excreted by the traced RGCs, by pruning of RGC neurites, and/or by direct toxicity of the tracer on a few RGCs. Our study does not allow to discriminate among these reasons, but given the very low number of PMCs quantified in these control retinas, tracing itself does not influence PMC quantification after injury.

In mice, after unilateral ON axotomy, there is a strong PMC reaction in the contralateral (uninjured) retina that is constant, scattered, and independent of the rate of RGC survival in the axotomized retina [21, 24]. In rats, a contralateral response involving activation of MCs, astrocytes, and Müller cells has been observed after axotomy [47, 51‐54], or ocular hypertension [55], but to our knowledge, this is the first report describing that in this species, there is not a PMC reaction. At present, we do not know why, but it is possible that the scarce retino-retinal projection is implicated because in adult rats, this projection is missing in some animals and, when present, is smaller than in mice [56‐58]. Thus, retino-retinal RGCs, which are directly affected by the axotomy, could transmit a danger signal that would spread along the retina, recruiting and activating MCs. These signals would not be sufficient to elicit a PMC response in those rats with very few retino-retinal RGCs, and inexistent in those animals without them. In addition, these signals might combine with a systemic inflammatory response that would account for the inflammatory reaction observed in both species.

The role of MCs in neurodegenerative diseases is controversial, but it is clear that they are not simple bystanders [27]. For one part, MCs restore tissue homeostasis, keeping the neuronal interactions as healthy as possible [9, 10], and their suppression improves axonal regeneration and RGC survival after axotomy [59]. On the other hand, MC over-activation may be damaging [31], as they act as mediators of the inflammatory response observed in several retinal degenerative diseases (reviewed in [60]), although in this respect there is some debate [61].

Upon neuronal death, MCs change their morphology from a surveying ramified shape to an ameboid-like one, M1 phenotype, which is considered harmful [28, 62‐65]. Here, we show that ameboid PMCs are often found engulfing RGC somas. We also found phagocytic rod-shaped MCs, especially along the intraretinal axons, in agreement with other works describing that rod-shaped MCs are induced by the interruption of axonal transport after insults such as axotomy or ocular hypertension [46, 66], but also during aging [67]. Rod-shaped MCs bear the M2 phenotype, and although their function is unknown, it is hypothesized that they protect the degenerating neurons [29, 67, 68]. In apparent contradiction with that hypothesis, we show here that rod-shaped MCs are or have been involved in the phagocytic response. In other words, either they engulf FG in their rod-shape state, or they could change from an ameboid phagocytic state to a protective rod-shape during the course of RGC degeneration. Because FG is accumulated in the phagolysosomes of MCs for a long time after cleansing the tissue, at least 2 months as shown here, we cannot rule out either possibility.

One of the most interesting results of this work is the 3 days delay between RGC death and RGC clearance that was observed after both lesions. This means that independently of the course of RGC loss (slower after ONC than after ONT), 3 days are needed for the MCs to become phagocytic. Therefore, to assess RGC death, tracing is not the best approach, particularly when manipulating MCs as neuroprotective tool, because MC inhibition may further delay the removal of dead RGCs from the tissue.

At 14 days, the number of PMCs in the GCL exceeds the population of SMCs in intact retinas by > 10,000. There are several theories to explain this increase: (I) microglial proliferation [69‐71], (II) extravasation of monocytes [72], and (III) migration from other retinal layers [59]. With our data, we cannot rule out (I) or (II). We found at 14 days, non-PMCs spread along the GCL, except around the optic nerve and the retinal periphery where they accumulated. They may reflect either local division, or retinal infiltration of monocytes from the ciliary bodies [73], and the peripapillary region. It is worth noting that these non-PMCs may or may not be activated, and although they have not engulfed RGC somas, they may have phagocytosed RGC axons, which do not accumulate FG. However, we do demonstrate that there is MC migration between the IPL and the GCL that would account for half of the MC surplus. Interestingly, 2 months after either lesion, the number of MCs in the GCL decreases, increasing in the IPL with cells from the GCL as the returning ones are PMCs. Finally, while the number of MCs in the GCL is back to normal, the IPL is short of ~ 4000 MCs. Have the missing MCs died? Or have they left the retina? We think that the second possibility is more plausible, because MC death would probably cause a second wave of MC activation and phagocytosis.

Our state-of-the-art topographical analyses show that more PMCs accumulate in the retinal regions where there is, proportionally, more RGC death. It is important to have in mind that RGC death is diffuse along the retina, meaning that RGCs die synchronously, but more of them die where there are more. Thus, there must be signals from the degenerating neurons and sensors in the MCs that inform of the amount of cell bodies to be removed, and upon them, more or less MCs migrate to the injured zone, having into account that one MC is able to engulf three RGC somatas. All in all, these results indicate that phagocytosis is activated by the death of RGCs rather than by the inflicted axonal damage.

Finally, our present data together with a previous work in mice [21] suggest that after optic nerve axotomy, MCs perform a cleansing role, but there is no indication of them being involved in the death of RGCs, at least during the quick phase of loss. In fact, Hilla et al. [74] have recently demonstrated, using an in vivo approach to deplete retinal MCs, that these cells do not play an active role on the death of RGCs after optic nerve axotomy.