BAX activation in mouse retinal ganglion cells occurs in two temporally and mechanistically distinct steps

Background Pro-apoptotic BAX is a central mediator of retinal ganglion cell (RGC) death after optic nerve damage. BAX activation occurs in two stages including translocation of latent BAX to the mitochondrial outer membrane (MOM) and then permeabilization of the MOM to facilitate the release of apoptotic signaling molecules. As a critical component of RGC death, BAX is an attractive target for neuroprotective therapies and an understanding of the kinetics of BAX activation and the mechanisms controlling the two stages of this process in RGCs is potentially valuable in informing the development of a neuroprotective strategy. Methods The kinetics of BAX translocation were assessed by both static and live-cell imaging of a GFP-BAX fusion protein introduced into RGCs using AAV2-mediated gene transfer in mice. Activation of BAX was achieved using an acute optic nerve crush (ONC) protocol. Live-cell imaging of GFP-BAX was achieved using explants of mouse retina harvested 7 days after ONC. Kinetics of translocation in RGCs were compared to GFP-BAX translocation in 661W tissue culture cells. Permeabilization of GFP-BAX was assessed by staining with the 6A7 monoclonal antibody, which recognizes a conformational change in this protein after MOM insertion. Assessment of individual kinases associated with both stages of activation was made using small molecule inhibitors injected into the vitreous either independently or in concert with ONC surgery. The contribution of the Dual Leucine Zipper-JUN-N-Terminal Kinase cascade was evaluated using mice with a double conditional knock-out of both Mkk4 and Mkk7. Results ONC induces the translocation of GFP-BAX in RGCs at a slower rate and with less intracellular synchronicity than 661W cells, but exhibits less variability among mitochondrial foci within a single cell. GFP-BAX was also found to translocate in all compartments of an RGC including the dendritic arbor and axon. Approximately 6% of translocating RGCs exhibited retrotranslocation of BAX immediately following translocation. Unlike tissue culture cells, which exhibit simultaneous translocation and permeabilization, RGCs exhibited a significant delay between these two stages, similar to detached cells undergoing anoikis. Translocation, with minimal permeabilization could be induced in a subset of RGCs using an inhibitor of Focal Adhesion Kinase (PF573228). Permeabilization after ONC, in a majority of RGCs, could be inhibited with a broad spectrum kinase inhibitor (sunitinib) or a selective inhibitor for p38/MAPK14 (SB203580). Intervention of DLK-JNK axis signaling abrogated GFP-BAX translocation after ONC. Conclusions A comparison between BAX activation kinetics in tissue culture cells and in cells of a complex tissue environment shows distinct differences indicating that caution should be used when translating findings from one condition to the other. RGCs exhibit both a delay between translocation and permeabilization and the ability for translocated BAX to be retrotranslocated, suggesting several stages at which intervention of the activation process could be exploited in the design of a therapeutic strategy.


Background
Proteins of the BCL2 gene family regulate mitochondrial involvement and dysfunction during the process of intrinsic apoptosis.The family is sub-classi ed into three distinct groups based on the presence of BCL2 Homology (BH) domains.These include full-length anti-apoptotic proteins (BCL2 and BCL2L1 (BCLX L )) and full-length pro-apoptotic proteins (BAX and BAK), which contain all 4 BH domains, and pro-apoptotic activator proteins which contain only the BH3 domain (BH3-only proteins such as BIM, BBC3, BID, BAD, HRK, NOXA, and BMF).Both BAX and BAK are found in an equilibrium between the surface of the mitochondrial outer membrane (MOM) and the cytosol in a process that is controlled by BCL2 and/or BCL-X L , which retrotranslocate the pro-apoptotic proteins to the cytosol (1,2,3).In this equilibrium, BAX is primarily cytosolic, while BAK is predominantly located on or near the MOM surface.
Upon stimulation of the cell death program, one or more BH3-only proteins are either post-translationally modi ed or expressed.
The process of BAX activation has been most extensively studied.BAX is translocated to the MOM (6, 7,8) where it undergoes a conformational change exposing a hydrophobic C-terminal domain allowing it to anchor in the membrane.BAX monomers then dimerize and these begin to oligomerize forming large molecular weight complexes that are often localized to the scission sites of fragmenting mitochondria (9).A principal function of BAX is to permeabilize the MOM (MOMP), which leads to the release of proapoptotic molecules, such as cytochrome c and SMAC/Diablo, that signal caspase activation.Studies in tissue culture cells show that MOMP occurs virtually instantaneously with the initiation of BAX translocation (10,11), consistent with reports that even a single activated BAX molecule can destabilize a lipid bilayer to create a proteolipid pore large enough to allow passage of cytochrome c (12).An exception to this timing of MOMP is during the process of anoikis, which is apoptosis that is stimulated by loss of cell-cell or cell-matrix adhesion (13).Loss of adhesion induces BAX translocation, but permeabilization is delayed by several hours (14), which allows for BAX retrotranslocation once the cells regain surface contacts.
Retinal ganglion cells (RGCs) are long projection neurons of the central nervous system that transmit light-stimulated electrical signals from the retina to visual centers in the brain through axons that project into the optic nerve.These neurons are highly sensitive to both acute and chronic optic nerve damage, which causes RGC death and blindness in a variety of conditions including glaucoma.Based on prevalence statistics for glaucoma alone, RGC loss represents one of the most common forms of neurodegeneration in humans, projected to affect more than 111 million individuals world-wide by 2040 (15).RGCs die by executing the intrinsic apoptotic program.Evidence for this comes from studies showing that manipulating the balance of BCL2 family proteins provides nearly complete abrogation of RGC soma death in both development and after optic nerve damage.
Overexpression of BCL2 or BCLX L in RGCs yields supernumerary cells in adult mice, indicative of failure of redundant RGCs to undergo developmental apoptosis, and dramatic preservation of RGCs after axotomy and in the DBA/2J mouse model of glaucoma (16,17,18,19,20,21).Similarly, genetic ablation of the Bax gene yields even greater inhibition of developmental apoptosis (22) and virtually complete elimination of RGC death in mouse models of acute optic nerve crush (ONC) (23,24,25) and in DBA/2J glaucoma (26,27).Importantly, most neurons, including RGCs, are exclusively reliant on BAX because Bak transcripts undergo a splice variant that introduces a premature stop codon, eliminating full length BAK as a functioning pro-apoptotic protein (28, 29,30,31).
To date, few studies have examined the process and timing of BAX activation during the progression of RGC death, even though this likely represents the most critical transition point in the commitment of cells to the apoptotic pathway (49).Immunostaining for BAX shows increased staining of RGCs beginning as little as 30 minutes after acute optic nerve damage, peaking at 3-7 days (48, 50).While these results were originally interpreted as showing an increase in BAX expression, they likely re ect the process of translocation and concentration of cytosolic BAX at the MOM of RGCs.Supporting this, more recent studies have shown that forced expression of a GFP-BAX fusion protein in RGCs failed to spontaneously induce BAX translocation and death but con rmed that optic nerve damage induced signi cant translocation in cells during the 3-7 day interval (25,32).Longitudinal imaging of GFP-BAX expressing RGCs using blue-light confocal scanning laser ophthalmoscopy also suggested that cells exhibiting translocation remained present for several days before clearance (32).
Since development of neuroprotective therapies for RGCs critically relies on BAX activation, we sought to investigate the kinetics of this process in RGCs using both static and live-cell imaging methodologies.Our results demonstrate that the translocation and permeabilization two steps of BAX activation are mechanistically and temporally distinct in RGCs.While most cells irreversibly translocate, some cells exhibit spontaneous retrotranslocation of BAX in a process that resembles the interruption of anoikis, indicating that translocation and MOMP do not occur simultaneously.Further, we show that BH3-only proteins are critical to drive the translocation step, but are not su cient to elicit permeabilization, which requires the activation of kinases distinct from the DLK-JNK axis.Together, this work highlights several points of intervention to prevent BAX activation and introduces the possibility of reversing the early stage of translocation.

Results
GFP-BAX translocation in RGCs after optic nerve crush RGC susceptibility to ONC is in uenced by genetic variation in mouse strains (51,52).Therefore, we rst assessed the time line of GFP-BAX translocation in response to ONC in BALB/cByJ mice, which we planned to use in subsequent live-cell imaging experiments.Adult BALB/cByJ mice were transduced with AAV2/2-Pgk-GFP-BAX and one month later were subjected to ONC.Localization of GFP-BAX in RGCs in retinal whole mounts harvested at different times after ONC (Fig. 1A-F) showed an increasing percentage of transduced cells exhibiting punctate localization, indicative of BAX translocation to the MOM (7,8,9,11,25,53).A signi cantly elevated percentage of cells with punctate GFP-BAX, relative to contralateral retinas, was observed by 3 days and persisted through 7 days after ONC before declining (P < 0.0001, t-tests, Fig. 1G), similar to other reports in rodents with acute optic nerve damage (32,48,50).
Examination of punctate GFP-BAX localization also revealed that many cells exhibited this feature not only in the soma, but in different RGC compartments, including the dendritic arbor and the intraretinal axons proximal to the crush lesion (Fig. 1B, Fig. 2A-C).
To determine if BAX could also be translocated in axons distal to the glial lamina of RGCs undergoing apoptosis, we initiated apoptosis at the level of the soma by intravitreal injection of staurosporine (STS).Sections of the optic nerve of GFP-BAX expressing mouse eyes showed axons with patterns of diffuse BAX and axons with BAX puncta, the latter of which co-localized with the accumulation of phosphorylated neuro lament commonly exhibited by degenerating RGC axons (54) (Fig. 2D).These results indicate that multiple RGC compartments can exhibit BAX translocation in response to an apoptotic stimulus.

Live ex vivo imaging of RGCs reveals reduced GFP-BAX translocation kinetics
To measure the kinetics of GFP-BAX translocation in RGCs, we transduced RGCs in BALB/cByJ mouse retinas using AAV2/2-Pgk-GFP-BAX and performed ONC one month later.After 7 days, retinas were harvested and immediately prepared for ex vivo live-cell imaging.GFP-BAX translocation occurred in the RGCs of the crushed retina as visualized by the transition from a diffuse to a punctate localization (Fig. 3A, Supplemental Video S1).Contralateral retinas showed no signi cant increase in the number of imaged cells exhibiting punctate GFP-BAX localization during an imaging session of 3 hours (Fig. 3B).In contrast, a signi cant proportion of cells in the ONC retinas converted to punctate GFP-BAX localization (P < 0.0001, t-test, Fig. 3B), indicating that translocation ex vivo was a consequence of ONC in vivo and not the process of preparing the retinas for imaging.To provide comparison for GFP-BAX translocation kinetics in ONC-induced mouse RGCs, we also conducted kinetic analyses in mouse RGCs from retinas exposed to the apoptosis-inducer STS, or in differentiated 661W cells overexpressing histone deacetylase 3 (HDAC3).
HDAC3 is selectively toxic to differentiated 661W cells (31) and is known to play an important role in RGC pathology after ONC in mice (55,56,57,58).
Table 1 Comparison of BAX translocation kinetics between tissue culture cells and retinal ganglion cells (RGCs) in ex vivo retinal imaging experiments.Apoptotic stimuli included staurosporine (STS), overexpression of histone deacetylase 3 (HDAC3), or optic nerve crush (ONC)."Zero times" for the initiation of translocation are based on either the addition of STS (D407, HeLa, HCT116 cells, and the RGC STS cohort), time from transfection of cells (661W cells), or time of animal euthanasia (RGC ONC and Retro cells).Retinal explants were harvested from eyes at 7 days after ONC (naïve retinas were used for STS experiments).The initiation times and translocation rates were calculated from the translocation data after tting with a sigmoid function as previously described [11].Means and standard deviations are derived from cell populations, where multiple foci within a cell were rst averaged, and cells with fewer than 3 foci were omitted.Values for Coe cient of Variation within the cell population were calculated as described in the Methods.Y-Maximum refers to the number of relative uorescence units (RFU) of uorescently tagged-BAX that had accumulated at individual foci during an experiment.Values of RFU intensity were set to 1.0 at baseline.All data represents experiments using GFP-BAX.a Data for D407, HeLa, and HCT116 cells were described previously in reference [11].

Cell
b Not applicable (n/a).Retrotranslocation was found to occur immediately after completion of maximum translocation in these cells.
c Not calculated due to the low number of data points available.
Previous analyses of BAX translocation kinetics in 3 different tissue culture cell lines showed average rapid rates of BAX translocation (between − 1.0 and − 1.8 Log 2 relative uorescence units (RFU)/min), that was initiated nearly simultaneously at different mitochondrial loci within a single cell (Initiation coe cient of variation (CoV) values between 1% and 5%).Within any given cell, however, the rates of translocation at individual loci exhibited a high degree of variation (Rate CoV values between 47% and 150%) (Table 1).Differentiated 661W cells exhibited a comparable mean rate of translocation (-2.2 Log 2 RFU/min) compared to other in vitro cell-experiments (Table 1), along with a low CoV (1%) for initiation times within a given cell and a high CoV (53%) for intracellular rates (Table 1).In contrast, analysis of GFP-BAX translocation in ex vivo RGCs yielded several differences compared to in vitro tissue culture cells.Regardless of the damaging stimulus (ONC or STS injection), RGCs yielded less overall translocation of GFP-BAX to the MOM than 661W cells (RFU Y-Maximum, Table 1, Fig. 3C, Supplemental Figure S1), which may re ect differences between imaging the RGCs in a thick segment of retina versus tissue culture cells adherent to a coverslip.RGCs also exhibited signi cantly slower rates of translocation (-5.1 and − 6.0 Log 2 RFU/min for ONC and STS respectively, Table  S2).These data demonstrate that the kinetics of intracellular initiation and rate of translocation, both as a global average and as independent events within a single cell, do not directly translate between cells in vitro and cells contained in a complex tissue.
A subset of ONC-induced RGCs exhibit retrotranslocation of translocated GFP-BAX Live imaging of GFP-BAX translocation in AAV2/2-Pgk-GFP-BAX transduced retinas showed that some RGCs (5/33 cells) in ONCinduced retinas exhibited an unexpected phenomenon.GFP-BAX translocation was followed almost immediately by its retrotranslocation back into the cytosol (Fig. 4A, Supplemental Video S2).Kinetic analysis of the GFP-BAX translocation (ON) rate of these cells was statistically similar to RGCs that translocated GFP-BAX and stably maintained a punctate phenotype.However, the rate of retrotranslocation (OFF rate) from the same cells was signi cantly slower than the ON rates for both stably converted or retrotranslocating RGCs (P < 0.0001, ANOVA, Figs.4B and C, Supplemental Figure S3, Table 1).Measurements of the difference in total amount of GFP-BAX uorescence indicated that retrotranslocating cells translocated a similar amount of GFP-BAX as stably converted cells, and that the process of retrotranslocation removed translocated GFP-BAX to a level indistinguishable from cytosolic localization (Fig. 4D, Table 1).
RGCs with translocated BAX exhibit a delay before activation of MOMP Studies of BAX activation in tissue culture cells indicate that BAX-mediated MOMP events such as cytochrome c release (10,11,59,60,61) occur rapidly at the onset of BAX translocation to the MOM.The conformational change in BAX that facilitates MOM insertion also exposes its N-terminal domain, which can be detected with the 6A7 monoclonal antibody (8, 62, 63, 64) (Fig. 5A,B).
When we assessed 6A7 immunostaining in GFP-BAX expressing RGCs after ONC, approximately 50% of cells with punctate GFP-BAX were 6A7-negative at 3 days post ONC, which progressively decreased over 5 and 7 days (Fig. 5C).Importantly, while we did observe some 6A7-positive cells that were negative for the GFP-BAX fusion protein, we did not observe any 6A7-positive cells that exhibited diffuse localization of the fusion protein.These results suggest that some RGCs exhibit a delay in the permeabilization step of BAX activation, which would result in a delay of MOMP and the release of cytochrome c.To con rm this, we injected mouse eyes with AAV2/2-Pgk-mCherry-BAX and AAV2/2-Pgk-cytochrome c-GFP, which resulted in co-transduction in 80.9 ± 5.1% of cells (n = 9 retinas, data not shown).Retinal whole mounts harvested at 3, 5, and 7 days after ONC were assessed for cytochrome c-GFP release in cells with punctate mCherry-BAX localization.As with the scoring of cells for 6A7 immunostaining, we observed the same inverse correlation between time and the proportion of cells with punctate BAX that exhibited mitochondrial localization of cytochrome c (Fig. 5D, Supplemental Figure S4).

BAX translocation and permeabilization in RGCs is in uenced by signaling pathways associated with anoikis
The delay between BAX translocation and permeabilization, and the phenomenon of retrotranslocation, are features associated with anoikis (13).In tissue culture cells, disruption of cell contacts results in a loss of signaling by Focal Adhesion Kinase (FAK), stimulating BAX translocation through the activation of latent BMF, a BH3-only protein (65, 66).To assess whether BAX activation in RGCs can be driven by anoikis, we rst determined if FAK inhibition could induce GFP-BAX translocation and permeabilization in naïve retinas.Figure 6A shows the induction of GFP-BAX translocation in a small percentage of RGCs in response to increasing concentrations of the FAK inhibitor PF573228.Maximum translocation was achieved 24 hrs after injection of the lowest dose of inhibitor (1 µL of a 10 µM solution injected).This percentage was comparable to the percentage of translocated cells in retinas 24 hrs after ONC, but less than the level observed by 5 days after ONC.No signi cant change in the percentage of translocating cells was observed up to 5 days from the single injection (data not shown) suggesting either clearance of the inhibitor, or that PF573228 was affecting a subset of RGCs.Examination of translocated cells in 10 µM PF573228-injected eyes showed that an average of 82.79% of RGCs with translocated GFP-BAX were 6A7 negative, while only 50% of ONC-induced RGCs were 6A7-negative (P < 0.0001 for 10 µM injection, t-test, Fig. 6B).These results suggest that FAK inhibition may primarily induce translocation but not MOM permeabilization.
Importantly, this axis contributes to the up-regulation of several BH3-only proteins that participate in RGC death through the activation of both JUN-and p53-mediated transcription (41,42,43,69,70,71).Analysis of transcript abundance for pJUN-and p53regulated genes showed no signi cant accumulation of mRNAs from these target genes after PF573228 injection, compared to signi cant induction in retinas one day after ONC (Supplemental Figure S5).Immunolabeling for accumulating pJUN, however, did show that PF573228 induced the accumulation of this activated transcription factor within 24 hrs after injection (Fig. 6C-F) in a fraction of cells (6.27% of total neurons) albeit in signi cantly fewer cells than ONC after the same time period (Fig. 6G) (P < 0.0001, t-test).Aligning with both reduced pJUN immunostaining and pJUN-mediated gene expression observations, PF573228 induced translocation of GFP-BAX in signi cantly more cells that were lacking pJUN accumulation than ONC (Fig. 6H) (49.5% v 19%, P < 0.0001, χ 2 test).Our interpretation of these data is that ONC can induce BAX translocation through both DLK-JNK-dependent and independent pathways, and that the independent pathway is enriched by direct inhibition of FAK.
We also noted that injection of PF573228 was associated with a propensity to affect larger cells and was statistically similar to the population of cells affected within 24 hrs of ONC (Fig. 6I).By 4-5 days pONC, GFP-BAX translocation was present in cells that were statistically smaller (P < 0.0003 relative to 1 day post-ONC and 1 day PF573228 treatment, ANOVA, data not shown).Interestingly, a small proportion of RGCs in both PF573228-treated eyes, and after 24 hrs post-ONC, exhibited BAX translocation that was restricted to regions of the dendritic arbor (Fig. 6J) including different secondary branches off the same primary branch.
Since FAK inhibition-induced GFP-BAX translocation could be attributed to only a subset of RGCs, we explored further if the DLK-JNK axis affected either translocation and/or permeabilization after ONC.We tested whether two inhibitors could affect BAX activation if immediately injected after ONC surgery.First, SB203580, a selective p38/MAPK14 inhibitor which is also implicated in anoikis (72) and second, sunitinib, an inhibitor with broad selectivity which has been used to disrupt the DLK-JNK axis (36).Importantly, sunitinib does not affect p38/MAPK14 (73).Assessment of each inhibitor to suppress the DLK-JNK axis was made by staining for pJUN nuclear accumulation as a surrogate for activation of the axis.At 24 hrs after ONC neither inhibitor prevented the accumulation of pJUN (Fig. 7A), nor prevented the expression of either pJUN-or p53-mediated gene expression at 5 days post-ONC (Figs.7B, C).
While no effect on the DLK-JNK axis was expected for p38/MAPK14 inhibition, this result was unexpected for sunitinib.GFP-BAX translocation after 5 days was unimpeded by both inhibitors relative to ONC surgery alone (Fig. 7D), but both signi cantly attenuated BAX permeabilization as a function of 6A7-staining (P < 0.0001, ANOVA, Fig. 7E).Sunitinib yielded a greater effect at preventing permeabilization than the p38/MAPK14-selective inhibitor SB203580 (P = 0.0053, t-test).
Since sunitinib had no effect on blocking the DLK-JNK axis, we instead interrogated the relative contribution of this pathway in mice with conditional ablation of both Mkk4 and Mkk7 genes, which encode two kinases central to the axis (38).To avoid the developmental defects associated with combined Mkk4/7 deletion (38), these genes were targeted in adult mice using AAV2mediated delivery of Cre recombinase at the time cells were also transduced with AAV2/2-Pgk-GFP-BAX.At 5 days post-ONC, Mkk4/7 de cient cells exhibited signi cantly fewer cells exhibiting GFP-BAX translocation compared to wild type mice (P < 0.0001, ttest) and no increase in punctate cells relative to contralateral uncrushed retinas (Fig. 7F).Together, these data indicate that the DLK-JNK axis is a critical component in facilitating the translocation of GFP-BAX in a majority of RGCs but is insu cient to stimulate permeabilization in the absence of active kinases such as p38/MAPK14.

Discussion
To our knowledge this study represents the rst kinetic analysis of BAX translocation and permeabilization in cells present in a complex tissue environment.The process of translocation can be initiated several days after the initial damage to the RGC axon, but once initiated proceeds rapidly in RGCs albeit with statistically slower kinetics than we have observed in tissue culture cells.Slower translocation of BAX to the MOM may be a consistent feature of RGCs regardless of the apoptotic stimulus, including exposure to STS, a kinase inhibitor that has frequently been used to assess the kinetics of BAX activation in tissue culture cells.This may be a phenomenon that is intrinsic to RGCs, but at this time we cannot rule out that this process is in uenced by surrounding cells, such as supporting macro-and microglia that are responding to the damaged environment.Such responses could be the generation of trophic factors, clearance mechanisms of damaged RGC compartments, and the efforts to maintain ion homeostasis and synaptic function (74,75,76).
We observed translocation of GFP-BAX in several RGC compartments, including the dendritic arbor and both the intraretinal and optic nerve axon.This observation suggests that BAX-mediated catabolism may play a role in degeneration of these compartments and could explain why axonal degeneration in glaucomatous DBA/2J mice is delayed in the absence of a functional Bax gene (26).BAX and/or caspase-dependent axonal degeneration has been described in several neuronal cell types (77, 78, 79) and may precede SARM-1-mediated calpain-dependent catabolism, observed in other cell types (80).Similarly, the pruning of dendritic branches has been linked to caspase activation (81, 82) and recent studies suggest that arbor remodeling in some RGCs is mediated by a BAXdependent mechanism (83).
We also observed a small percentage of cells with punctate GFP-BAX restricted to the dendritic arbor in both ONC retinas and in retinas treated with the FAK inhibitor PF573228.The arbor may be a critically susceptible region of the RGC where the process of BAX translocation is initiated and may account for why larger RGCs, which have the largest and most complex dendritic arbor branching (84), were the earliest to exhibit BAX translocation in both ONC and FAK inhibition experiments.We speculate that once initiated, BAX translocation may then propagate to other compartments of the RGC.The propagation of apoptosis, including BAX translocation, MOMP, mitochondrial membrane depolarization, and caspase activation has been described in tissue culture cells and some neurons (85, 86, 87, 88, 89, 90,91,92,93).With respect to BAX translocation in tissue culture cells, propagation is initiated preferentially in distal tips of processes and then travels to the soma and even into additional processes at a velocity of ~ 7 µm/sec (85).Within the soma, however, BAX translocation initiates simultaneously at mitochondrial foci.If RGCs also propagate BAX translocation at a similar rate, this would explain why we observed so few cells with punctate BAX restricted to the arbor, suggesting we were capturing a rare early stage of propagation.The concept that BAX translocation is initiated rst in the dendritic arbor aligns with observations made by others that remodeling of this compartment represents one of the earliest morphological changes associated with RGC pathology after optic nerve damage (94,95,96,97).
Our results also showed that the processes of BAX translocation and permeabilization in RGCs were separated by a signi cant delay.This contrasts dramatically from studies in tissue culture cells undergoing BAX-dependent apoptosis, where BAX permeabilization leading to MOMP occurs virtually simultaneously with the translocation of the rst BAX molecules (11,98,99).The precise length of the delay interval in RGCs was not clear from our experiments, but we posit that it may range from several hours to days.Longitudinal imaging of GFP-BAX translocation in eyes of mice subjected to ONC suggests that cells with translocated BAX can remain present for up to 5 days before being cleared (32).
A small proportion of RGCs exhibited rst translocation of BAX followed immediately by its retrotranslocation back into the cytosol, a phenomenon that has also been documented in detached tissue culture cells that have initiated anoikis, but are then allowed to reattach with the surrounding matrix (14,72).Studies on anoikis suggest that the process is initiated by the loss of function of FAK.
In this pathway, FAK inhibition leads to the activation of the latent BH3-only protein BMF (66, 100).Anoikis also exhibits a delay between translocation and permeabilization that has been estimated to be up to 3-4 hrs.We hypothesized that a similar mechanism may be a critical feature of some damaged RGCs.Consistent with this, we observed a signi cant increase in BAX translocation in naïve retinas exposed to the FAK inhibitor PF573228.Importantly, PF573228 induced a similar level of translocation after 24 hrs as ONC at this time point but did not induce high levels of BAX permeabilization.We interpret the effect of PF573228 as priming BAX in RGCs without necessarily activating it.This would explain previous observations that FAK inhibition exacerbates RGC loss but only after optic nerve damage (101,102).
The mechanism of BAX translocation is still not elucidated but is likely tied to the activity of BH3-only proteins.A predicted primary function of BH3-only proteins is that they sequester anti-apoptotic proteins like BCLX L allowing for a shift in the cytosolic-MOM equilibrium of pro-apoptotic counterparts (103).This concept is reenforced by data showing the BH3-only proteins have a stronger binding a nity to anti-apoptotic members, and only exhibit transient interactions with pro-apoptotic proteins (103,104).BH3-only proteins, however, are also thought to play a direct role in BAX translocation.Evidence suggests that some BH3-only members bind to the surface of inactive BAX dimers in the cytosol to facilitate their separation into monomers and translocation to the MOM (105,106).This form of BAX exposes the C-terminal membrane anchor domain allowing it to associate with Voltage-dependent anion channel type 2 (VDAC2) on the MOM.This is also the likely state of BAX and BAK in healthy cells when they are in the BCLX L /BCL2mediated equilibrium with the cytosol, and we predict, the state of translocated BAX in affected RGCs that have not undergone permeabilization.
Protein structural studies also indicate that BH3 mimetics play a role in the BAX permeabilization step.Mimetics can bind to a region of BAX distinct from its BH3-domain causing it to undergo a conformational change leading to unfolding of the "latch" domain that exposes hydrophobic alpha-helices that further strengthen the MOM-BAX interaction (4, 107) and the initiation of proteolipid pores.This model is di cult to reconcile with our results using kinase inhibitors.Both SB203580 and sunitinib failed to block the expression of BH3-only genes that are linked to the DLK-JNK axis (Hrk and Noxa) suggesting that these BH3-only proteins (and presumably Bbc3 and Bim, which are also tied to p53 and pJUN activation, respectively) were active in these experiments, yet we still observed a signi cant attenuation of BAX permeabilization.Additionally, combined deletion of Mkk4/7, which inhibits the expression of these BH3-only genes, effectively blocked translocation of BAX in a signi cant proportion of cells.Therefore, our results are more consistent with BH3-only proteins playing a critical role in the translocation of BAX but are, by themselves, not su cient for permeabilization.
The mechanism of retrotranslocation in damaged RGCs is also not clear.In healthy cells, BAX is retrotranslocated by BCLX L (1), but it is unknown if this same mechanism is active in reattached cells during anoikis or in RGCs that exhibit spontaneous retrotranslocation.Overexpression of BCLX L in both tissue culture cells and RGCs prevents BAX translocation in the rst place (21), making study of its role in retrotranslocation di cult.We predict a model, however, in which a subset of RGCs undergoing anoikis are able to recover cell-cell contacts and have su ciently high enough levels of endogenous BCLX L to initiate retrotranslocation.
Further studies to explore this model are ongoing.
The permeabilization of the MOM by translocated BAX in anoikis is linked to the activity of p38/MAPK14 (72), although it is not clear if this kinase modi es BAX directly, or if the mechanism is indirectly linked to p38/MAPK14 modi cation of an intermediate substrate.A recent report implicates DRP1 as directly interacting with BAX and mediating its activation (108) and p38/MAPK14dependent phosphorylation of DRP1 has been linked to other neurodegenerative conditions (109,110).p38/MAPK14 has previously been shown to have a role in RGC pathology after both acute and chronic optic nerve damage (111,112), but the actual mechanism of neuroprotection afforded by blocking this kinase has not been studied in detail.Blocking p38/MAPK14 with SB203580 had no impact on preventing BAX translocation, but it did suppress the permeabilization of translocated BAX up to 5 days after ONC.The inability to block translocation contradicts previous reports that p38/MAPK14 is required for this process in cultured cells (113,114,115), although these studies relied on cell fractionation experiments or used antibodies that only recognized permeabilized BAX.
Alternatively, the difference between these previous observations and ours could be associated with a difference in activation mechanisms between isolated cultured cells and RGCs in a complex tissue.
Interestingly, we also observed that a low dose of sunitinib also blocked BAX permeabilization, but not translocation.Sunitinib is a broad-spectrum kinase inhibitor, but does not directly affect p38/MAPK14 activity (73).Additionally, at the concentration we used, we did not affect activation of the DLK-JNK axis, thus modulation of either MKK4/7 or JNK2/3, both of which have been linked to the activation of p38/MAPK14 (116), would not be expected to be affected in our experiments.The implication of this is that low levels of sunitinib were affecting one or more kinases that have yet to be identi ed in the BAX activation process.Based on known selectivity measures (73), the concentration of sunitinib we used could target as many as 99 other known kinases before reaching e cacy to inhibit DLK.Identi cation of other kinases in the BAX activation process in RGCs is ongoing.
Acute damage to the axons of the optic nerve leads to nearly complete loss of the entire RGC population in the mouse retina (117), but elegant studies using single cell RNA-Seq clearly show that RGC populations exhibit varying levels of susceptibility to this damage paradigm (118).One interpretation of these differences in susceptibility is that damage induced by ONC is not monolithic and is instead multifactorial.For example, some RGCs may be susceptible to loss of FAK signaling, while others are more susceptible to ER stress or accumulation of reactive oxygen species, etc, as part of a continuum of damaging stimuli.Since Bax deletion blocks essentially all RGC soma death after optic nerve injury (23, 24, 26) it is likely the common denominator of all the proapoptotic signaling pathways in these cells after optic nerve damage.Consequently, we predict that an analysis of the pattern of BH3-only protein activation/expression would act as a signature of the critical damaging stimulus of a given RGC subtype, consistent with studies that show that the pattern of BH3-only proteins activated in apoptosis varies with the type of stimulus (5,119).Ultimately, nal susceptibility may be a function of the cooperation of different members of the BH3-only protein family.For example, RGCs susceptible to FAK inhibition may be more sensitive to anoikis-driven BMF activation, which would account for why a higher proportion of cells exposed to PF573228 exhibited punctate GFP-BAX without DLK-JNK driven accumulation of pJUN and its associated BH3-only proteins.Variable RGC susceptibility as a function of BH3-only protein levels also explains observations that selective deletion of these genes only provides partial protection to RGCs after ONC (41,44,45,46,47,48).The implication of a model in which RGC death is a function of multiple different, and possibly sequential stimuli, is that there is no single early molecular event that can be targeted therapeutically, complicating the development of a global neuroprotective strategy.Conversely, targeting later BAX activation, particularly the permeabilization step, could offer promise as a strategy to facilitate recovery of RGCs after damage.

Conclusions
Acute optic nerve damage induces BAX activation which can be temporally separated into a stage of BAX translocation to the MOM and a stage where it permeabilizes the MOM leading to the release of cytochrome c.Translocation is associated with the ability of the cells to express and/or activate BH3-only proteins, while permeabilization is associated with one or more kinases, including p38/MAPK14.At least of subset of RGCs exhibit sensitivity to loss of signaling pathways linked to cell-adhesion and FAK activity similar to the process of anoikis.

Animals
Adult

Optic nerve crush
ONC was performed as previously described (52,120).Brie y, mice were anaesthetized via intraperitoneal injection of ketamine (120 mg/kg) and xylazine (11.3 mg/kg).One drop of 0.5% proparacaine hydrochloride (Akorn, Lake Forest, IL) was used to numb the experimental eye.A lateral canthotomy exposed the optic nerve and was followed by an incision in the conjunctiva at the limbal junction.The optic nerve was clamped for 3 seconds using self-closing N7 forceps (Roboz Surgical Instrument Co, Gaithersburgs, MD).After surgery, triple antibiotic ointment was used to cover the eye, and a subcutaneous injection of buprenorphine (0.2 mg/kg) was delivered to mitigate pain.The right eye was not surgerized and evaluated as the contralateral condition.

Intravitreal injections
After anesthesia as described above, a sterile 30G needle was used to puncture a small hole through the conjunctiva and scleral tissue.A beveled 35G Nano l needle a xed to a Nano l syringe (World Precision Instruments, Inc, Sarasota, FL) was inserted in the pre-poked hole to deliver desired virus, drug or vehicle.The contents were slowly delivered over 30 seconds, and the needle remained within the globe for an additional 30 seconds with caution, such that the lens was not damaged.Antibiotic ointment was placed over the injection site and buprenorphine (0.2 mg/kg) was administered to relieve pain.AAV2/2-Pgk-GFP-BAX, AAV2/2-Pgk-mCherry-BAX, and AAV2/2-Pgk-Cytochrome c-GFP were packaged by the University of North Carolina (UNC) Vector Core (Chapel Hill, NC) (1-5 x 10 12 genome copies/ml).AAV2/2-CMV-Cre virus (> 10 12 gc/mL) was purchased from the UNC Vector Core.All injections of virus were conducted at least 1 month prior to conducting optic nerve crush surgery.

Clones and plasmids
The GFP-BAX construct used for tissue culture cell experiments was previously described (121).The HDAC3-Flag construct was a gift from Eric Verdin (Addgene plasmid no.13819: Addgene, Cambridge, MA).The mitoBFP construct was described previously (11,53).Transgenes in these plasmids were all driven by the CMV promoter.Plasmids used for generating AAV2 viral particles were derived from the parent plasmid AAV-Pgk-Cre (Addgene plasmid no.24593), a gift from Dr. Patrick Aebischer.The coding sequence for Cre was removed by digestion with EcoRI and SalI and was replaced with a sequence bridge of single cutting restriction sites.
Tissue culture cell maintenance 661W cells (mouse retinal progenitor cells) were cultured in DMEM (Cellgro, Mediatech, Inc., Manassas, VA) supplemented with 10% FBS (Atlanta Biologicals, Norcross, GA) and 1% Antibiotic/Antimycotic (Cellgro, Mediatech, Inc.) at 37°C and 5% CO2.Treatment with 316nM staurosporine for 24 hours induced 661W differentiation (122).For transfection, prewarmed media containing 2 µg each of the GFP-BAX, mitoBFP, and HDAC3-Flag plasmids was mixed with 2 volumes of Transfast reagent (E2431, Promega, Madison, WI), vortexed and incubated at room temperature for 15 min.After removal of regular media, the transfection mixture was added to cells and they were incubated at 37˚C for 1 hr, after which fresh media was added to the cells and they were incubated for a further 15-24hr.Overexpression of HDAC3-Flag is selectively toxic to differentiated neurons (123), including differentiated 661W cells (31) and plays a critical role in the apoptotic program executed by RGCs (56, 58).661W cells, originally described as RGC-5 cells from Rattus norvegicus, were con rmed in our laboratory as Mus musculus origin by PCR of the mitochondria d-loop (124) and the upregulation of TUBB3 after differentiation (31).

Live imaging and analysis
Live-cell imaging was performed using an Andor Revolution XD spinning disc confocal microscopy system (Andor, Belfast, Northern Ireland) comprised of the Nikon Eclipse Ti inverted microscope, Nikon objectives, an iXon x3 897 EM-CCD camera, a Yokogawa CSU-X1 confocal spinning disk head, the Andor Laser Combiner with four solid state lasers, an ASI motorized stage with Piezo-Z, and an Okolab CO 2 cage incubator for temperature and CO 2 control at 37°C and 5% CO 2 .
661W cells were plated at a density of 100,000-300,000 cells per well on 4-well chamber slides containing #1.5 optical grade plastic (Ibidi, Madison, WI).Cell treatments were staggered for an appropriate time for the cell type to ensure the ability to image each well on a slide.Prior to imaging, phenol-containing media was removed from the well, and replaced with a recording media prepared inhouse (HBSS with 1.26 mM calcium chloride, 0.49 mM magnesium chloride, 0.4 mM magnesium sulfate, no phenol red, 4.5 g/L glucose and 10 mM HEPES) supplemented with appropriate serum concentration.Under temperature and CO 2 control, time-lapse imaging was performed by collecting z-stacks consisting of 20-25 optical sections at 0.22µm (Nyquist) taken every minute for one to two hours.
Live ex vivo imaging of mouse retina was performed with the same microscopy conditions as discussed above.To prepare the retina, animals were euthanized with a lethal overdose of pentobarbital sodium, stained with live-cell nuclear dye DRAQ5 (ThermoFisher Scienti c, Waltham, MA), then mounted on a coverslip bottom (#1.5) chamber slide (Ibidi).For DRAQ5 staining, once euthanasia was con rmed, an intravitreal injection of DRAQ5 was administered (200 µm nal concentration), then the enucleated eye globe was incubated in nutrient rich medium at 37°C for 10 minutes.DRAQ5 labeling was stable for ~ 3 hours.To mount the retina, an eye cup was created from the eye globe, then a single relaxing cut through the retina and sclera was made in the anterior region.Without removing the dissecting scissors from the relaxed cut, in one quick motion, the lens was removed with the dissecting scissors as the retina was ipped such that the ganglion cell layer was facing down onto the coverslip portion of the chamber slide.
A second relaxing cut was made, allowing the retina to rest as at as possible.A drop of 2% low melt agarose (ThermoFisher Scienti c) medium was placed on the retina, followed by a small piece of coverglass to aid in attening the retina for imaging purposes.Once positioned, the retina was overlayed in the agarose medium, allowed to polymerize brie y, and covered with a layer of sterile water to prevent evaporation and image drifting during acquisition.The low melt agar was made in 50% 'recording media' (above) and 50% Neurobasal medium (ThermoFisher Scienti c) supplemented with 1X Neurocult SM1 (Stemcell Technologies, Cambridge, MA) and 10% FBS (Atlanta Biologicals).The FBS and Neurocult supplements were added after the melted agarose had been cooled to 60˚C.Both experimental and control retinas were mounted in the same way, and this process was completed in approximately 45 minutes from the time of euthanasia.Z-stacked images were collected every 5 minutes for up to three hours after the start of time-lapse imaging.DRAQ5 staining provided a control for tissue drift during imaging.Only RGCs that maintained a xed position, based on nuclear visibility, were used for subsequent kinetic analyses.
For STS treatment, naïve animals were euthanized and the eyes injected with DRAQ5 and STS to a nal concentration of 1 µM and incubated at 37˚C for 10 min.The retinas were then harvested and mounted for imaging as described above except that the media contained 1 µM STS during the imaging procedure.
All images were subject to the same background subtraction and Gaussian ltering within the IMARIS 7.7 software (Bitplane, Concord, MA), where each lter had its own constant algorithm and threshold setting.The spots function was used to identify and track GFP-BAX foci within a cell and extract the maximum uorescence of the desired uorophore from each surface.A subset of BAX foci was selected for computational analysis based on the longevity of the track, whether the track encompassed the translocation event and remained within the de ned 3-dimensional space of the cell, and whether the BAX foci were spatially isolated.For ONC experiments, a total of 33 cells, from 7 retinas, exhibiting translocation during an imaging session were recorded.
Of these cells, we were able to extract kinetic data for a minimum of 3 mitochondrial foci per cell for 27 cells (5 of which also exhibited retrotranslocation).For STS treated retinas, we were able to record 10 cells from 4 retinas, all of which were suitable for kinetic analysis.

Computational Analysis
BAX translocation analysis was performed as described previously (11).Brie y, raw maximum uorescence data extracted from the IMARIS program was normalized (to its initial value) for each foci within a cell to allow for comparison among many GFP-BAX foci.These data were then t to a sigmoid curve using a linear regression model.The code using the SciPy library in the Python programming language is available in (11).Quanti cation for retrotranslocating BAX cells was performed by applying sigmoid functions to each phase separately (translocation and retro-translocation).The retrotranslocation curve was rst t using a gaussian function to determine the maximum point of the curve, which would de ne the two phases.

Retinal tissue preparation and immunostaining
Enucleated eyes were punctured on the anterior portion of the limbus with a 30 gauge needle and then immersed in 4% paraformaldehyde in PBS for 1 hour.Eye cups were made by removing the anterior chamber of the eye by cutting along the limbus.
In preparation for immuno uorescent staining, retinal eye cups were rst wiped with the tip of a twisted Kimwipe to remove the residual vitreous and then equilibrated in 30% sucrose in PBS overnight at 4˚C.They were then subjected to 3 rounds of freezethawing on dry ice and blocked in PBS containing 2% goat serum and 0.1% Triton X-100 overnight at 4˚C.Primary antibodies were added to the same buffer and incubated with the eyecups for a minimum of 3 days at 4˚C.After incubation, the eyecups were washed 3 times in PBS and then incubated in PBS containing 2% goat serum and 0.1% Triton X-100 overnight at 4˚C.
For immunostaining longitudinal sections of the optic nerve, xed eyecups containing ~ 1 mm of the optic nerve were equilibrated in 30% sucrose in phosphate buffered saline (PBS) (50 mM phosphate, pH 7.4, 150 mM NaCl) overnight at 4˚C and then embedded in Optimal Cutting Temperature media (Scigen Scienti c, Gardena, CA) on dry ice.Cryosections (10 µm) were selected that included the optic nerve head and an extended portion of the optic nerve.These were rst incubated in PBS at room temperature for 10 min and then blocked in PBS containing 10% horse serum (Lonza, Basel, Switzerland) and 0.1% Triton X-100 for 1 hr at room temperature in humidi ed chambers.Sections were then covered in PBS containing 2% BSA and 0.1% Triton X-100 with primary antibody (SMI-31 see below) overnight at 4˚C.After incubation, the sections were washed 3 times in PBS and incubated in the same buffer (PBS, 2% BSA, 0.1% Triton X-100) with the secondary antibody for 1 hr at room temperature.

Fixed tissue imaging and analysis
For all images used for quanti cation, samples were imaged on a Zeiss Axioimager Z2 upright microscope (Zeiss, Oberkochen, Germany).All images were analyzed using Zen Blue software (Zeiss).Separate imaging modalities were used to quantify the percentage of cells with punctate GFP-BAX and the percentage of 6A7 positive cells.To quantify the percentage of GFP-BAX labeled cells with punctate GFP-BAX, 4 images per retinal quadrant were taken using a 20X air objective.The retinal eccentricity was held constant in the mid-periphery.For those 16 images, every GFP-BAX positive cell was counted.A minimum of 700 cells were counted for each retina.To quantify the percentage of 6A7 positive cells, a second set of images was taken of each sample.Beginning at one side of the retina, the camera was panned across the retina, and every eld that contained a cell with punctate GFP-BAX was imaged.This was necessary to maximize the number of cells with punctate GFP-BAX counted per retina.An average of 94 punctate cells were counted for each retina with a range of 33-177 cells counted.The images of GFP-BAX puncta in the dendritic arbors of RGCs were imaged on a Leica SP8 (Leica Camera, Wetzlar, Germany) confocal microscope using a 1.4 NA 63X oil objective.

Realtime quantitative polymerase chain reaction
Immediately after euthanasia, eyes were enucleated and the retinas were extracted and ash frozen in microcentrifuge tubes on dry ice.A total of 12 retinas were used for each experimental group.These were split into groups of 3 and their OS and OD retinas were frozen in separate tubes.Therefore, each experimental group in the experiment consists of four replicate values obtained from three mice each.RNA was extracted from the pooled retinas using a DNA/RNA/Protein extraction kit (IBI Scienti c, Dubuque, IA).RNA quality was veri ed using a Nanodrop One spectrophotometer (ThermoFisher Scienti c).All RNA samples had a A260/280 value between 2.10 and 2.13 and an A260/230 value between 2.00 and 2.30.cDNA was created by annealing 2 µg of total RNA from each sample to Oligo(dT) 15 Primers (Promega) and then incubating with MMLV-RTase (Promega) in a 40˚C water bath for 1 hr.The 25 µL total volume of each reaction was diluted by addition of 200 µL of diethyl pyrocarbonate treated, nuclease free water (ThermoFisher Scienti c).Quantitative PCR was performed in a Quant Studio 7 Flex Real Time PCR machine (ThermoFisher Scienti c) using a customized TaqMan Array Card (ThermoFisher Scienti c) that was designed to assess the abundance of the following mouse RNA transcripts: 18S, Hrk, Ecel1, Atf3, Noxa, and Cdkn1a (p21).Changes in transcript abundance were calculated using the △△C t method (125) using the 18S rRNA value from each sample as a reference to calculate △C t and then the change caused by the experimental paradigm was calculated as the difference in △C t values between the experimental eyes and the contralateral eyes.

Statistical analysis
Calculations of sample size for statistical power were made using previous data on outcome metric variances for ONC experiments.Cohort sizes of n = 5 retinas provide a power set to 0.8 to detect a difference of 12% between groups and n = 4 retinas to detect a 20% difference.Translocation rates from live-cell imaging experiments did not t a normal distribution and were Log 2 transformed for analysis.One-way analysis of variance (ANOVA) was used to evaluate differences when comparing multiple groups.A two-sided Student's t-test was used to compare the means of two isolated cohorts.A χ 2 test was used to compare the proportions of punctate GFP-BAX cells showing accumulation of pJUN after ONC or injection of PF537228.To assess the intracellular synchronicity of both the time of initiation and the rate of translocation, we used data collected from multiple points within a single cell to determine the coe cient of variation (CoV) for each variable per cell.These statistics were normalized to a percentage (absolute value) of the mean for each cell within a single group to facilitate comparison between groups.Data was generally represented as scatter plots  Punctate labeled intraretinal axons could be identi ed in retinas at all time points after ONC examined.Scale bar = 25 µm.(D) Punctate labeling of GFP-BAX in axons of the optic nerve distal to the optic nerve head.Cell death in this experiment was induced by intravitreal injection of staurosporine 24 hrs previously in order to preserve axon integrity with the RGC soma.An axon with diffuse GFP-BAX (arrow) is compared to an axon with punctate GFP-BAX (arrowhead).Note that the axon with punctate GFP-BAX also counterstains with an antibody against phosphorylated neuro lament (pNF) consistent with reports that degenerating neurons, including RGCs, accumulate this antigen (54,126).Scale bar = 4 µm.

Figure 3 Ex
Figure 3 the time after transfection of differentiated cells with a plasmid expressing human HDAC3.The time scale for graphs (F and G) was set to "0" when imaging of the explant began.Foci from 2 individual cells are plotted as red and yellow circles, respectively.The more stacked the alignment of points relative to the X-axis is indicative of simultaneous initiation of GFP-BAX translocation at all sites within a cell.The broad range of points relative to the Y-axis is indicative of variable rates of translocation at different mitochondrial foci within a single cell.Analysis of the coe cient of variation for both initiation and rate indicate that RGCs exhibit greater variability in initiation times and less variability of the intracellular rates, than 661W cells (see Supplemental FigureS2).

Figure 4 Ex
Figure 4

Figure 7 The
Figure 7 (51)/cByJ or C57BL/6J mice (Jackson Laboratory, Bar Harbor, ME) were housed in microisolator cages with a 4% fat diet (8604 M/R; Harland Teklad, Madison, WI) and maintained in a 12 hour light/dark cycle.C57BL/6J mice carrying dual oxed alleles of Mkk4 and Mkk7 were developed as previously reported (38).Biological variability involving sex-related differences in response to optic nerve damage was considered in the design of experiments.While no sex differences have been detected using the acute ONC protocol(51), experimental groups were each assigned an equal mixture of male and female mice.Animals were handled in accordance with the Association for Research in Vision and Ophthalmology statement on the use of animals in research.The Institutional Animal Care and Use Committees of the University of Wisconsin and the University of Rochester approved the experimental protocols and ethical care for mice used in these experiments.Any animals that exhibited distress were removed from the study.Euthanasia was performed by intraperitoneal injection of 0.05 mL of Euthasol (Virbac AH, Fort Worth, TX).Following cessation of a detectable heartbeat, mice were cervically dislocated.