Brain‐derived neurotrophic factor prevents dendritic retraction of adult mouse retinal ganglion cells

Abstract We used cultured adult mouse retinae as a model system to follow and quantify the retraction of dendrites using diolistic labelling of retinal ganglion cells (RGCs) following explantation. Cell death was monitored in parallel by nuclear staining as ‘labelling’ with RGC and apoptotic markers was inconsistent and exceedingly difficult to quantify reliably. Nuclear staining allowed us to delineate a lengthy time window during which dendrite retraction can be monitored in the absence of RGC death. The addition of brain‐derived neurotrophic factor (BDNF) produced a marked reduction in dendritic degeneration, even when application was delayed for 3 days after retinal explantation. These results suggest that the delayed addition of trophic factors may be functionally beneficial before the loss of cell bodies in the course of conditions such as glaucoma.


Introduction
The adult murine retina provides an attractive model to study detailed structural alterations of neurons following axotomy (Johnson & Martin, 2008;Bull et al., 2011;Guerin et al., 2011;Wood et al., 2011;Denk et al., 2015;White et al., 2015). While the loss of retinal ganglion cells (RGCs) has typically been used as the primary read-out in this preparation (Bull et al., 2011), fewer studies address the changes that might occur in the dendrites of axotomized RGCs prior to cell loss (see Discussion for a very recent example, Johnson et al., 2016). Yet dendritic atrophy following axotomy may be of considerable functional relevance while also offering a therapeutic opportunity in conditions such as glaucoma (Weber et al., 1998;Morgan, 2012;El-Danaf & Huberman, 2015). In principle, dendritic arbourization can be readily quantified by analytical tools, such as Sholl analysis (Sholl, 1953), in which proximal dendrite loss is reflected by a lowered peak amplitude and reduced terminal branching density is indicated by a leftward shift in the profile. Sholl analysis has been further developed with the introduction of the notion of a 'branching index' (Garcia-Segura & Perez-Marquez, 2014). While Sholl analysis is as such well-established and used extensively to quantify dendritic arbourization in numerous central nervous system (CNS) structures, the procedures allowing unbiased and sparse labelling of individual cells are less straightforward. In particular, the RGC morphology varies greatly between cell types in the rodent eye, reflecting the large number of individual cell types (Sun et al., 2002;Coombs et al., 2006). In view of this, random labelling of RGCs would be desirable to minimize bias. Sparse labelling using transgenic animals is frequently used to analyse dendritic arbours in the CNS including those of RGCs, typically using the Thy1 promoter coupled to genetically encoded fluorescent reporters. While elegant and convenient, this approach is not without problems given observations by us and others that the Thy1 promoter may be up or downregulated following axotomy (Lee et al., 1998;Schlamp et al., 2001;Huang et al., 2006;Astafurov et al., 2014). As the mechanisms underlying sparse expression of the Thy1 promoter in transgenic animals are still unknown, it is also unclear whether RGC labelling is truly random or not. In this study, we use an acute labelling technique designated diolistics (Gan et al., 2000) to determine the time course of neuronal dendritic atrophy following RGC axotomy. The use of this technique revealed that dendritic alterations occur within hours following axotomy and precede by days the loss of RGCs, thus allowing us to explore the effect of brain-derived neurotrophic factor (BDNF) on dendrites days before any confounding effects caused by the rescue of RGC cell bodies. This is potentially a significant confounding factor as BDNF has been known for a long time to prevent the death of RGCs in a number of species, both in vitro and in vivo after axotomy (Rodriguez-Tebar et al., 1989;Meyer-Franke et al., 1995;Di Polo et al., 1998).

Immunohistochemistry
At the end of the culture period 16 retinas (n = 14 mice of both sex, aged 2-7 months) were fixed and cryosectioned (v.s.). Sections were labelled with the RGC markers NeuN, TUJ1 and Thy1.2, and the apoptotic marker active caspase-3. Antibody working concentrations and sources are given in Table 1. Incubations were at room temperature, unless stated otherwise. Sections were blocked with 5% chicken serum (Invitrogen Ltd., 30 min), then incubated with primary antibody (4 h or overnight at 4°C). Sections were incubated with secondary antibody (1.5 h), nuclear stained with TO-PRO-3 (10 min), and coverslipped with Prolong Gold anti-fade reagent. Positive controls were anterior sections of sagittally cut mouse brain (NeuN, TUJ1 and Thy1.2) or mouse spleen sections (active caspase-3). Negative controls were as follows: primary antibody only, secondary antibody only, no antibody. Sections were imaged at 209 by confocal microscopy using separate channels for the fluorophore (488 nm argon laser with 500-530 nm bandpass filter) and nuclear stain (633 nm laser). Z-stacked 1024 9 1024 pixel images were obtained in 1 lm steps. The central 1 lm slice was analysed. NeuNand TUJ1-positive cells in the GCL were double stained for nuclear stain and antibody. Mean green channel intensity was calculated with the image processing software FIJI (Schindelin et al., 2012) as the [mean green channel count for the layer] À [mean green channel count for background outside the region of interest].

Viability assay
The terminal deoxynucleotidyl transferase dUTP nick-end labelling (TUNEL) assay (Millipore, UK Ltd.) was used according to the manufacturer's protocol to quantify apoptosis in retinas cultured for up to 14 days. Briefly, frozen sections of 14 explants (n = 13 mice of both sex, aged 2-7 months) were digested with proteinase-K before labelling with TdT end-labelling cocktail, and then Avidin-FITC. Sections were counterstained with TO-PRO-3, and coverslipped with Prolong Gold Anti-fade reagent. Positive controls were retinal sections pre-incubated in proteinase-K for 1 h. Negative controls were prepared by omission of Avidin-FITC. Sections were imaged at 209 by confocal microscopy using separate channels for Avidin-FITC (488 nm argon laser with 500-530 nm bandpass filter) and TO-PRO-3 (633 nm laser). Z-stacked 1024 9 1024 pixel images were obtained and analysed as previously described. Mean green channel intensity was calculated as before.

Diolistics and Sholl analysis
Fifty explants (n = 38 mice of both sex, aged 1.5-5 months) were cultured for up to 3 days. Tungsten particles (200 mg) were coated with 3 mg 1-1-dioctadecyl-3,3,3,3-tetramethylindocarbocyanine perchlorate (DiI) and 6 mg 3,3-dioctadecyloxacarbocyanine perchlorate (DiO), and distributed along Tefzel tubing (Bio-rad, Hertfordshire, UK) using a Tubing Prep Station (Bio-rad). The tubing was cut into 1.2 cm 'bullets' (Gan et al., 2000). At the end of the culture period, DiI/DiO-coated tungsten particles were fired using a Helios gene gun (Bio-rad) 5 cm from the retinal surface at 120 psi (helium) through a 3 lm polyethylene terephthalate membrane filter (Scientific Laboratory Supplies Ltd., Yorkshire, UK) to prevent clumping of dye particles. Explants were then incubated at 37°C, 5% CO 2 for 30 min, followed by fixation with 4% paraformaldehyde (10 min), and nuclear stained with TO-PRO-3 (10 min). Finally, Diolistically labelled RGCs were identified as having a soma in the GCL and an axon projecting to the optic nerve. RGCs were imaged at 209 by confocal microscopy using separate channels for each dye (DiI 543 nm laser with 565-615 nm bandpass filter; DiO 488 nm argon laser with 500-530 nm bandpass filter). Z-stacked 512 9 512 or 1024 9 1024 8-bit pixel images of the entire cell were obtained in 1 lm steps. Cell eccentricities relative to the optic nerve head were noted using the Stage Controller Stepper (Zeiss). RGC images were imported into FIJI and dendritic arbours traced in 3D. Dendritic integrity was quantified by 3D Sholl analysis using the Simple Neurite Tracer plugin (Longair et al., 2011) centred on the soma centre with a sphere interval of 10 lm. The area under the Sholl profile was calculated using the trapezoidal model. The branching index was calculated as described by Garcia-Segura and Perez-Marquez using the number of dendrite intersections, I, at each radial distance from the soma centre, x (Garcia-Segura & Perez-Marquez, 2014). Briefly, at each radial distance > 0 lm the number of new dendrites was multiplied by the radial distance. The sum equated to the branching index for that cell. Note that this index accounts for branching events only.

BDNF treatment
To test a possible role of BDNF in delaying RGC dendropathy, the medium was supplemented with 100 ng/mL E. coli recombinant BDNF (Regeneron/Amgen) diluted in 19 PBS containing 0.1% bovine serum albumin (BSA). Ten explants (n = 10 mice of both sex, aged 2-5.5 months) were cultured in BDNF-supplemented medium for 3 days initiated at 0 day. Controls had medium with the same volume aliquot (2 lL/1 mL medium) of vehicle. As before, the medium was replaced daily, along with fresh BDNF (100 ng/mL) or vehicle. At the end of the culture period RGCs were diolistically labelled and analysed by Sholl analysis. Dendritic field area was calculated as the area enclosed by a linear polygon connecting the terminal dendrites. All cells were masked prior to analysis.

Pan-caspase inhibition
To assess the role of caspase activation in RGC dendritic retraction, the pan-caspase inhibitor, Q-VD (R&D Systems, Abingdon, UK), was applied to nine explants (n = 6 mice, female, aged 3-4 months). Q-VD has been shown to be one of the most effective broad-spectrum caspase inhibitors and at 100 lM Q-VD is non-toxic and highly effective at preventing apoptosis in mouse lymphoblast cells (Caserta et al., 2003) and 52.6 lM Q-VD protects against RGC death in the rat retina (Patil & Sharma, 2004). A topical aliquot (100 lL) of 100 lM Q-VD, or vehicle (DMSO) as control, was applied daily. After 2 days retinas were labelled (diolistics) and nuclear stained (TO-PRO-3). Sholl profiles were derived for labelled cells as described. Cell analysis was undertaken in a masked fashion.

Delayed BDNF treatment
To test the neuroprotective role of BDNF in a more clinically relevant setting, six explants (n = 6 mice, male, aged 8.5 months) were cultured in BDNF-free medium for 3 days, followed by culture in BDNF-supplemented medium (100 ng/mL BDNF) for a further 3 days (6 days total). Fresh BDNF was added with each daily medium change. Controls had vehicle only. RGCs were labelled diolistically and analysed in a masked fashion as before.

Cell sub-type analysis
To account for cell-subtype bias, cells were classified as ON, OFF and bistratified according to dendritic termination within the inner plexiform layer (IPL); cells with dendrites terminating within the first 3/5 (~9 lm) of the IPL were categorized as ON, those with dendrites terminating in the end 2/5 (~6 lm) of the IPL were categorized as OFF. Bistratified cells had dendrites terminating in both sublaminae.

Statistical analysis
Data processing was carried out using MICROSOFT EXCEL (Office 2013). Statistical analysis was carried out using IBM SPSS (version 20). Normality of data was tested using the Shapiro-Wilk test. To compare two groups independent samples t-test, pairwise one-way ANOVA or pairwise Kruskal-Wallis one-way ANOVA was used. Comparison between multiple groups was made by one-way ANOVA with TUKEY post hoc, or Mann-Whitney with Bonferroni correction. P < 0.05 was the level of significance. Mean values are given AE SEM. The regression model for factors influencing the Sholl profile included culture period/culture conditions, age and sex (1 = male, 2 = female) as explanatory variables.
Having established that no significant signs of cell death is observed in the GCL during the first week of culture, RGCs were labelled diolistically for morphometric analysis (Fig. 4A-C). Surprisingly, dendritic shrinkage, measured as a decrease in Sholl area under the curve (AUC), could be observed as early as after 6 h post-mortem (P = 4.91 9 10 À2 , Mann-Whitney tests with Bonferroni correction, Fig. 4A-C), which further progressed over the next 3 days (P = 2.42 9 10 À9 , Fig. 4D and E). The branching index reduced by 41.9 AE 6.0% by 1 day (P = 0.007, ANOVA with Tukey post hoc), after which it remained relatively unchanged (Fig. 4F). Having defined a window of several days during which alterations in the dendritic profile could be analysed in the absence of cell death, the Sholl profiles were also determined separately for ON, OFF, bistratified RGC sub-types. No bias could be detected at any of the time points examined. Thus, compared with the same cell type at time 0, the Sholl AUC of ON cells decreased by 49.0 AE 7.1% (P = 4.02 9 10 À4 , Mann-Whitney tests with Bonferroni correction) after 2 days and by 61.0 AE 5.6% (P = 2.50 9 10 À5 ) after 3 days. The corresponding values for the OFF cells decreased by 37.9 AE 6.5% (P = 1.91 9 10 À3 ) after 1 days and by 30.1 AE 17.0% (P = 4.96 9 10 À2 ) after 2 days and for the bistratified cells 48.0 AE 8.1% (P = 4.83 9 10 À2 ) after 2 days (see also Fig. 4G). To examine any potential role of confounding variables, a multiple regression analysis was run with Sholl AUC as the dependent variable and culture period, age and sex as predictors. The model provided a good fit for the data (P = 7.27 9 10 À12 , ANOVA) with the time of culture period as the only significant predictor (B = À 228, P = 7.08 9 10 À10 ) of Sholl AUC.
Having established that our culture conditions and labelling technique allow us to analyse the role of BDNF in modulating dendritic atrophy following axotomy without the confounding factor of RGC loss or of biases towards particular RGC populations, we next supplemented the culture medium for a period of 3 days with BDNF added at 0 or 3 days (see Fig. 5A for a summary of the procedure, with representative cells illustrated in Fig. 5B and C). Following incubation with BDNF, the Sholl profile increased at 20-80 lm from the soma centre (P < 0.05, Kruskal-Wallis), relative to controls (Fig. 5D). The Sholl AUC increased by 75.8 AE 14.8% (P = 0.002, ANOVA with Tukey post hoc) relative to controls and was not significantly different to that measured at 0 day (Fig. 5E). The addition of BDNF increased branching by 81.0 AE 15.7% (P = 1.94 9 10 À4 , ANOVA with Tukey post hoc), relative to controls and was not significantly different compared to 0 day cells (Fig. 5F). The dendritic field area of BDNF-treated cells was 61.2 AE 13.9% (P = 0.011, ANOVA with Tukey post hoc) larger compared with controls and 28.8 AE 6.1% (P = 0.004) smaller than that of 0 day cells (Fig. 5G and H). We next examined possible subtype-specific effects of BDNF and analysed ON, OFF and ON-OFF RGCs based on IPL stratification depth. The Sholl AUC of BDNFtreated ON cells increased by 108.9 AE 22.4% (P = 0.003, ANOVA with Tukey post hoc) relative to controls, not significantly different from 0 day cells. BDNF-treated OFF cells increased by 73.6 AE 30.2% (P = 0.047) relative to controls, with no significant difference from 0 day cells, as was also the case for bistratified cells (Fig. 5I). The possible confounding effects of age and sex were tested by multiple regression analysis. Area under the Sholl profile was the dependent variable and culture condition (1 = control and 2 = BDNF-treated), age and sex were predictors. The model provided a good fit of the data (P = 0.022, ANOVA) and in this case, the culture condition, plus or minus BDNF, was found to be the only significant predictor (B = 665, P = 0.008) of Sholl AUC.  We next determined the extent to which dendritic degeneration could be mitigated by caspase inhibition within 2 days of culture ( Fig. 6A-C), a time interval selected as this presented the peak for the change in dendrite configuration. Indeed, recent results suggest that the degeneration of neuronal processes may involve mechanisms requiring caspase activity (Williams et al., 2006;Nikolaev et al., 2009;D'Amelio et al., 2010;Hyman & Yuan, 2012;Erturk et al., 2014). While the Sholl profile was greater at 10-30 lm compared with untreated controls (P < 0.05, Kruskal-Wallis, Fig. 6D), the Sholl AUC, branching index and dendritic field area while showing a similar trend, did not reach significance ( Fig. 6E-G). Given this trend and the possibility of cell selectivity in the effect of Q-VD, RGC sub-type-specificity was also examined with ON, OFF and ON-OFF separately examined. However, this analysis did not reveal any evidence of a differential protective effect of pan-caspase inhibitors. Age was also checked as a possible confounder by multiple regression analysis but again the only variable found to be a predictor of Sholl AUC was the presence or absence of Q-VD in the culture medium (B = 464, P = 0.005).
Given the potential clinical relevance of the finding that BDNF retards the progression of dendritic atrophy, we next examined whether its delayed addition would also be effective, given the clear progression of dendritic retraction following axotomy in the absence of RGC death (see Fig. 7A-C). The corresponding Sholl profile was greater at 20-100 lm (P < 0.05, Kruskal-Wallis) compared with untreated controls (Fig. 7D). Sholl AUC increased by 135.6 AE 29.6% (P = 1.08 9 10 À4 , ANOVA with Tukey post hoc) relative to controls and 56.6 AE 19.7% (P = 0.003) relative to 3 days, but decreased by 32.2 AE 8.5% (P = 0.009) relative to 0 day (Fig. 7E). The branching index for delayed BDNF-treated cells was increased by 106.8 AE 29.2% (P = 0.009, ANOVA with Tukey post  hoc), compared to controls, but decreased by 45.1 AE 7.7% relative to 0 day (P = 0.002), relative to 0 day (Fig. 7F). The dendritic field area was 133.9 AE 32.8% increased (P = 1.93 9 10 À5 , ANOVA with Tukey post hoc), compared with untreated controls, but was 43.1 AE 8.0% decreased (P = 4.59 9 10 À5 ), compared to 0 day (Fig. 7G). To check for sub-type-specific effects of delayed BDNF treatment, ON, OFF and ON-OFF cells were also examined separately. The Sholl AUC for ON cells was increased by 95.8 AE 30.4% (P = 0.020, ANOVA with Tukey post hoc) relative to untreated cells and increased by 55.2 AE 24.1% (P = 0.025) relative to 3 days cells, and 39.5 AE 9.4% decreased (P = 0.033), compared with 0 day cells. The Sholl AUC for OFF cells was increased by 253.3 AE 80.2% (P = 0.001) compared with untreated cells, though this was not significantly different compared with 0 and 3 days cells. The Sholl AUC of bistratified cells was not significantly different between groups (Fig. 7H). There was no significant difference for BDNF-treated cells between sub-types, indicating an absence of sub-type-specific effects of BDNF. Regression analysis confirmed culture condition (1 = control, 2 = delayed BDNF-treated) as the only factor influencing dendritic integrity (B = 671, P = 0.001).

Discussion
The key novel finding here is that delayed application of BDNF significantly retards dendritic atrophy of RGCs following retinal explantation. While this manuscript was being revised, a related study was published using mouse retinal explants and time-series imaging of Thy1-YFP-labelled RGCs (Johnson et al., 2016). This study showed that the joint addition of ciliary neurotrophic factor and BDNF at the time of explantation attenuates dendritic loss over the effects of forskolin alone, thus broadly supporting our conclusions (see below for further discussion).
To ensure that under our experimental conditions, the loss of dendrites can be studied in the absence of RGC death, we first established the time course of cell loss in the GCL. We found that the pattern of cell loss identified by the RGC markers NeuN, Tuj1 and Thy1.2 was inconsistent, likely reflecting the lack of truly specific markers (Perry et al., 1984;Gan et al., 1996). In addition, the expression of proteins, including Thy1 (Lee et al., 1998;Schlamp et al., 2001;Huang et al., 2006;Astafurov et al., 2014), may alter following injury, as we report here (Fig. 2). Therefore, we concluded that the use of nuclear staining to quantify RGC loss is more reliable, given that amacrine cells are resistant to degeneration (Kunzevitzky et al., 2010) and that the numbers of apoptotic cells in the GCL detected by TUNEL and active caspase-3 were negligible over 14 days. Using nuclear staining, we found that RGC counts do not decrease significantly during the course of the first week (Fig. 1B). As the loss of RGC dendrites can already be observed 6 h after explantation (Fig. 4D), a substantial time window exists during which the protective effects of BDNF can be studied, even under conditions of delayed addition (Figs 5 and 7). A detailed analysis of the process of dendritic retraction (Fig. 4D) revealed an initial leftward shift of the (Sholl) profile, possibly resulting from selective loss of distal dendrites, as reported in axotomized mouse RGCs (Morquette et al., 2015). In contrast, we can discount sub-type bias after finding no significant differences in Sholl AUC between ON, OFF or bistratified cells at any time point, and the absence of sampling bias as a function of eccentricity (Fig. 4B). A key difference between our study and the results published by Johnson et al., 2016 lies in the method used to assess dendritic arbourization. While we randomly labelled cells by diolistics at different time points, Johnson et al. use a Thy-1 YFP line that also sparsely labelled RGCs, with the advantage that individual cells can be followed over time. However, in addition to the persistent unresolved question of the mechanisms leading to single-cell marking that may or may not select for subclasses of neurons in the CNS, we previously noted that genetically encoded reporters such as YFP can underreport dendritic pathology (Williams et al., 2013), possibly because probes are more likely to be expressed in cells less affected by the lesion, with a biosynthetic machinery that is still intact. In contrast, diolistics labelling relies on temperature-dependent diffusion within the plasma membrane, regardless of the health of the cell. This may explain why in the very recent study by Johnson and colleagues, dendritic pathology was not reported before 7 days of culture, a time point at which the death of RGCs was already significant (Johnson et al., 2016). This temporal coincidence makes it difficult to separate cell death from dendritic atrophy and while we have no explanation as to why under our experimental conditions, no significant RGC death (determined by nuclear staining) can be observed during the first week following explantation, we think that our much earlier detection of dendritic atrophy may be explained by a combination of the use of an unbiased labelling method (Sherazee & Alvarez, 2013) coupled with the underreporting dendritic pruning when using Thy1-YFP animals (Williams et al., 2013).
As such, the effects of BDNF on dendrites may not be too surprising given previous reports by others in various areas of the CNS, typically during development (McAllister et al., 1995(McAllister et al., , 1996Bosco & Linden, 1999;Horch et al., 1999;Lom & Cohen-Cory, 1999;Jin et al., 2003;Baquet et al., 2004;Rauskolb et al., 2010). Of special relevance are previous experiments indicating that BDNF treatment reduces dendritic retraction of RGCs in vivo in a cat optic nerve crush model (Weber & Harman, 2008) and an optic nerve transection model (Rodger et al., 2012).
In conclusion, we show that application of BDNF alone attenuates dendritic retraction of randomly labelled RGCs in the adult retinal explant. Crucially, this BDNF protective effect can also be observed following delayed application of BDNF, a significant new observation that may have important implications for the treatment of neurodegenerative disorders, such as glaucoma.