AMPK hyperactivation promotes dendrite retraction, synaptic loss, and neuronal dysfunction in glaucoma

Background The maintenance of complex dendritic arbors and synaptic transmission are processes that require a substantial amount of energy. Bioenergetic decline is a prominent feature of chronic neurodegenerative diseases, yet the signaling mechanisms that link energy stress with neuronal dysfunction are poorly understood. Recent work has implicated energy deficits in glaucoma, and retinal ganglion cell (RGC) dendritic pathology and synapse disassembly are key features of ocular hypertension damage. Results We show that adenosine monophosphate-activated protein kinase (AMPK), a conserved energy biosensor, is strongly activated in RGC from mice with ocular hypertension and patients with primary open angle glaucoma. Our data demonstrate that AMPK triggers RGC dendrite retraction and synapse elimination. We show that the harmful effect of AMPK is exerted through inhibition of the mammalian target of rapamycin complex 1 (mTORC1). Attenuation of AMPK activity restores mTORC1 function and rescues dendrites and synaptic contacts. Strikingly, AMPK depletion promotes recovery of light-evoked retinal responses, improves axonal transport, and extends RGC survival. Conclusions This study identifies AMPK as a critical nexus between bioenergetic decline and RGC dysfunction during pressure-induced stress, and highlights the importance of targeting energy homeostasis in glaucoma and other neurodegenerative diseases. Supplementary Information The online version contains supplementary material available at 10.1186/s13024-021-00466-z.

Adenosine monophosphate-activated protein kinase (AMPK) is a highly conserved energy sensor and an important metabolic regulator [11]. AMPK is a heterotrimeric serine/threonine kinase composed of a catalytic α subunit (PRKAA1), which is essential for AMPK function through phosphorylation of its activation loop, and regulatory β and γ subunits [12]. During energy stress, intracellular ATP levels fall and AMP rises leading to AMPK activation [13]. Active AMPK restores energy homeostasis through catabolic pathways that produce ATP while inhibiting processes that consume energy. For example, AMPK activation during developmental energetic stress inhibits neuronal polarization and axonal outgrowth [14,15] while it regulates synaptic remodeling in aging neurons [16]. Although sensing ATP is critical to restore energy, the consequence of persistent AMPK activity in neurons with compromised energy supply is unknown. Furthermore, the role of AMPK in dendritic and synaptic alterations in injured neurons is not well understood.
To address this knowledge gap, we focused on retinal ganglion cells (RGC), a population of long-projecting neurons that link the retina to the brain. The selective death of RGC is a crucial element in the pathophysiology of glaucoma, the leading cause of irreversible blindness worldwide [17]. Retraction of RGC dendrites with synapse disassembly is a key feature of ocular hypertension damage [7]. Recent work has strongly implicated energy deficits in glaucoma [18][19][20][21][22][23], and reduced AMPK correlates with anti-inflammatory responses and RGC protection [24]. However, the role of AMPK in RGC dendritic structure and connectivity is currently unknown. Our data reveal that AMPK is a critical mediator of RGC dendrite pathology and synapse elimination during ocular hypertension-induced metabolic stress. These findings highlight the importance of restoring energy homeostasis to recover neuronal function in glaucoma.

Experimental animals
All animal procedures were approved by the Centre de Recherche du Centre Hospitalier de l'Université de Montréal (CRCHUM) Animal Care Committee and the Canadian Council on Animal Care guidelines. Surgical procedures were carried out in female B6.Cg.Tg [Thy1-YFPH]2Jrs/J mice (Jackson Laboratory, Bar Harbor, ME) or wild-type littermate controls (3̶ 4 months of age) maintained under 12 h light/12 h dark cyclic light conditions with an average in-cage illumination level of 10 lx. All experiments were performed under general anesthesia using 2% isoflurane (0.8 l/min), except for electroretinogram (ERG) recordings (see below).

Magnetic microbead occlusion mouse glaucoma model
Unilateral elevation of intraocular pressure was induced by a single injection of magnetic microbeads into the anterior chamber of the eye as described [25]. Briefly, animals were anesthetized using isoflurane (2% isoflurane, 0.8 l/min) and a drop of tropicamide was applied on the cornea to induce pupil dilation (Mydriacyl, Alcon, Mississauga, ON). A custom-made sharpened microneedle attached to a microsyringe pump (World Precision Instruments, Sarasota, FL) was loaded with 1.5 μl of a homogenized magnetic microbead solution (diameter: 4.5 μm, 2.4 × 10 6 beads) (Dynabeads M-450 Epoxy, Thermo Fisher Scientific, Waltham, MA). Using a micromanipulator, the tip of the microneedle was used to gently puncture the cornea and the microbeads were injected into the anterior chamber. A hand-held magnet was used to immediately attract the magnetic microbeads to the iridocorneal angle. This procedure avoided injury to ocular structures such as the lens and iris. Sham controls received an injection of sterile phosphate buffered saline (PBS). An antibiotic eye drop was applied to the operated eye (Alcon) and the animal was allowed to recover on a heat pad. Intraocular pressure was measured before and after the procedure (bi-weekly) in awake animals using a calibrated TonoLab rebound tonometer (Icare, Vantaa, Finland). A drop of proparacaine hydrochloride (0.5%, Alcon) was applied to the cornea and, holding the tonometer perpendicular to the eye surface, ten consecutive intraocular pressure readings per eye were taken and averaged. The following preestablished exclusion criteria were used in our study: i) mice with consecutive low intraocular pressure readings (< 40-50% increase over control) over the course of 1 week after microbead injection, ii) lens opacification, and iii) eye infection. Following these criteria, less than 10% of the total number of mice subjected to this procedure were excluded.

Reverse transcription and quantitative real time PCR (qPCR)
Total RNA was isolated from individual retinas using the RNeasy Mini kit (Qiagen Inc., Valencia, CA). cDNAs were generated from total RNA (1 μg) using the Rever-tAid First Strand cDNA Synthesis Kit (Thermo Fisher Scientific). Real-time PCR was performed using TaqMan primers against YFP (Catalog # Mr04097229_mr) or βactin (Catalog # Mm02619580_g1) (Thermo Fisher Scientific). Amplification was performed using the 7900HT Fast Real-Time PCR System (Applied Biosystems, Waltham, MA) with the following cycle conditions: 95°C for 15 s, 60°C for 1 min, 72°C for 1 min. Reactions were run in triplicates for each sample and the 2-ΔΔCt method was used for the calculation of relative gene expression.

Dendritic arbor analysis
Dendritic arbor reconstruction and measurements were performed blind to manipulations. High-resolution images of YFP-labeled RGC dendrites were acquired using a confocal microscope (Leica Microsystems Inc.). Scans were taken at 0.5 μm intervals (1024 × 1024 pixels) with an average of 3 to 5 images per focal plane. Reconstruction of dendritic trees was carried out using the computer-aided filament tracing function of Imaris (Bitplane, South Windsor, CT). The following parameters were measured: i) total dendritic length: the sum of the length of all dendrites per neuron, ii) total dendritic field area: the area within the contour of the arbor created by a line connecting the outermost tips of the distal dendrites, iii) total number of branches: the sum of all dendritic branches per neuron, and iv) Sholl analysis: the number of dendrites that cross concentric circles at increasing distances from the soma (10 μm interval). RGC located in all retinal quadrants and eccentricities were included in our analysis.

Analysis of synaptic markers
Mice were sacrificed by decapitation under deep anesthesia (5% isoflurane), and the eyes were immediately collected. The cornea was pierced with a 30-gauge needle and the eye was submerged in ice-cold 4% carbodiimide for 30 min. Retinal cryosections (16 μm) were generated and incubated overnight with each of the following primary antibodies at 4°C: VGLUT1 (1:800, Synaptic System, Gottingen, Germany) and PSD95 (2 μg/ml, Abcam, Cambridge, UK). Sections were washed and incubated with secondary antibodies: anti-guinea pig and anti-mouse (Alexa 594 or 488, 2 μg/ml, Molecular Probes). Three retinal cross sections per eye were analyzed in 2 areas (central and peripheral) for a total of 6 output measures per mouse. Fluorescent labeling was visualized with a Leica SP5 confocal microscope (Leica Microsystems Inc.), and 7.5-μm-thick z-stacks were sequentially obtained at 0.13 μm intervals (1024 × 1024 pixels) with an average of three images per focal plane. Quantitative analysis of voxels, which measured the three-dimensional volume occupied by pre-and postsynaptic markers, was carried out using Imaris (Imaris-Coloc, Bitplane).
Human glaucoma specimens and processing , followed by anti-rabbit Alexa Fluor 594 or anti-guinea pig Alexa Fluor 488 (2 μg/ml, Molecular Probes). Glaucoma and age-matched control retinas were processed simultaneously under identical conditions. All RGC in each retinal section were measured and three retinal sections per sample case were analyzed. The quantification of fluorescence and normalization for background fluorescence was carried out as described above.

Assessment of blood-retinal-barrier integrity
Compound C (20 mg/kg, Sigma) or vehicle were administered by intraperitoneal injection using the same regimen as for dendritic and RGC density analysis (once a week for 2 weeks). Prior to imaging, mice were anesthetized and pupils were dilated as described above, and fluorescein (332 Da, 5%, Fluorescite, Novartis Pharma) was administered intraperitoneally. For fluorescein angiographic evaluation, retinal imaging was done 10 min after fluorescein administration using the Phoenix Micron IV System (Phoenix Research Labs, Pleasanton, CA). Mice subjected to transient retinal ischemia, a procedure known to compromise blood-retinal-barrier integrity [31], were used as positive controls. Briefly, under general anesthesia, the left optic nerve was exposed and the optic nerve dural sheath was opened longitudinally. A fine 10-0 nylon suture was carefully introduced between the sheath and the optic nerve and tied around the sheath to compress the ophthalmic vessels to block blood flow for 60 min [32]. Because the optic nerve sheath contains the ophthalmic artery, this procedure interrupts retinal and choroidal blood flow without damaging the optic nerve itself.

Electroretinography
Animals were dark adapted prior to electroretinogram (ERG) recordings and all manipulations were carried out under dim red light. Mice were anesthetized using ketamine (20 mg/ml), xylazine (2 mg/ml), and acepromazine (0.4 mg/ml). Bilateral pupil dilation was induced by applying tropicamide on the cornea (1%, Mydriacyl, Alcon). The recording system used Burian-Allen bipolar electrodes adapted for mice. The active electrode, with a corneal contact shape, was placed on the cornea following application of a drop of hydroxypropyl methylcellulose (Isopto Tears, 0.5%, Alcon, Mississauga, ON). The reference electrode was placed behind the ears, and the ground electrode in the tail. Electrical signals generated in the retina were amplified (1000x) and filtered (bandpass filter: 1-1000 Hz) using a commercial amplifier (Power Lab, ADInstruments, Oxford, UK). The recorded signals were digitized (Power Lab, ADInstruments) and displayed on a computer. Bilateral ERG recordings were performed simultaneously from both eyes. Measurements from non-injured eyes (pre-injury) served as baseline, and contralateral eyes were used for data normalization. To measure the positive scotopic threshold response (pSTR), the retina was stimulated at light intensities ranging between 10 − 6 and 10 − 4 cd.s.m. − 2 . A series of responses per flash were averaged for each light stimulus (50 recordings). The interval between light flashes was adjusted to allow for response recovery. A calibration protocol was established to ensure homogenous stimulation and recording parameters and was performed immediately prior to each experiment.

Quantification of neuronal survival
Mice were subjected to transcardial perfusion with 4% PFA, retinas were dissected out and free-floated overnight at 4°C in blocking solution (10% normal donkey serum, 2.5% bovine serum albumin, 2% Triton X-100) and incubated with Brn3a (0.3 μg/ml, Santa Cruz Biotechnologies) followed by secondary anti-goat Alexa Fluor 594 (2 μg/ml, Molecular Probes). Retinas were mounted with the nerve fiber layer side up, and visualized with a Zeiss Axio Observer (Carl Zeiss Canada). Brn3a-labeled or YFP-positive RGC were manually counted within three square areas at distances of 0.25, 0.625 and 1 mm from the optic nerve disc in each of the four retinal quadrants for a total of twelve retinal areas. For axon counts, animals received a transcardial injection of heparin in saline solution (10 U/ml) and sodium nitroprusside (10 mg/kg) followed by perfusion with 2% PFA and 2.5% glutaraldehyde.
Optic nerves were dissected out, fixed in 2% osmium tetroxide, and embedded in epon resin. Semi-thin sections (0.7 μm) were cut on a microtome (Reichert, Vienna, Austria) and stained with 1% toluidine blue. RGC axons were manually counted in three non-overlapping areas of each optic nerve section, encompassing a total area of 3500 μm 2 per nerve. The total area per optic nerve cross-section was measured using Northern Eclipse image analysis software (Empix Imaging, Toronto, ON), and this value was used to estimate the total number of axons per optic nerve. The measurements provided in this study represent a mathematical average of RGC soma and axonal density.

Axonal transport measurements
The anterograde tracer cholera toxin B 1 subunit (CTB) conjugated to Alexa Fluor 488 (Molecular Probes, Life Technologies, Eugene, OR) was injected intravitreally using a custom-made sharpened microneedle attached to a Hamilton syringe (1% diluted in sterile PBS, total volume: 2 μl). Animals were perfused transcardially with 4% PFA at 24 h after CTB administration, brains were removed, post-fixed and incubated overnight in 30% sucrose prior to embedding in optimal cutting temperature compound (Tissue-Tek, Miles Laboratories). Serial coronal cryosections of the entire superior colliculus from each animal were obtained using a cryostat (50 μm thickness). Seven sections per superior colliculus, from rostral to caudal, were selected using an unbiased systematic stereological sampling method as previously described [27]. Sections were photographed using the Zeiss Axio Observer fluorescent microscope with Apotome (Carl Zeiss Canada) and the area of the CTB signal in each section was measured using the Imaris Measurement Pro module (Bitplane, South Windsor, CT).

Data and statistical analyses
Data analysis was always carried out masked by third party concealment of treatment using uniquely coded samples. Statistical analysis was performed using Graph-Pad Instat software (GraphPad Inc., San Diego, CA) by one-way analysis of variance (ANOVA) followed by a Tukey's multiple comparison post hoc test, or by a Student's t-test as indicated in the legends. All cohorts were evaluated with normality (Kolmogorov-Smirnov test) and variance (F-test) tests. Only one eye per mouse was used for glaucoma induction (or sham) and/or treatment. Therefore, the number of mice (N) indicated is equal to the number of eyes analyzed (1 eye per mouse). The number of RGC analyzed in each experiment is indicated as (n). Both N and n values are indicated in the tables and in the figure legends.

Ocular hypertension promotes RGC dendrite retraction and synapse loss
To establish whether dendrite pathology is a consequence of metabolic stress, we first characterized the response of RGC to ocular hypertension using a microbead occlusion model [25]. Magnetic microbeads were injected into the anterior chamber and attracted to the iridocorneal angle using a magnet to block aqueous humour outflow resulting in high intraocular pressure (Fig. 1A, B). Ocular hypertension developed as early as 1 week after the procedure, but significant RGC soma or axon loss was not detected until 3 weeks of glaucoma induction (Fig. 1C, D). Analysis of dendritic arbors and synapses was carried out at 2 weeks after microbead injection, a time when ocular hypertension was stable and prior to overt RGC soma loss (mean ± S.E.M.; sham = 3114 ± 58 RGC/mm 2 , N = 12 mice; ocular hypertension = 3051 ± 73 RGC/mm 2 , N = 15 mice, Student's t- test, P = 0.5242, n.s.: not significant). Microbead occlusion was induced in mice expressing yellow fluorescent protein (YFP) under control of the Thy1 promoter (Thy1-YFPH) [33], a transgenic line that allows visualization of individual RGC dendritic trees [34]. The number of YFP-positive RGC, and the level of retinal YFP mRNA did not differ from those in non-injured retinas at 2 weeks after microbead injection (Fig. S1A, B) confirming stable reporter gene expression [35]. The antibody SMI-32, which recognizes nonphosphorylated neurofilament heavy chain, was used to identify alpha RGC characterized by strongly labeled somata and large dendritic arbors [36]. YFP-positive RGC that co-labeled with SMI-32 and had a clearly identifiable axon were selected for dendritic arbor imaging and three-dimensional reconstruction (see Methods, Fig. S1C-E). RGC located in all retinal quadrants and locations were included in our analysis and measurements were performed masked to manipulations. At 2 weeks after glaucoma induction, dendrites had visibly retracted relative to non-injured neurons from sham-injected eyes ( Fig. 1E-G). Analysis of total dendritic area, length and number of branches demonstrated a reduction of 22, 28, and 24%, respectively, compared to SMI-32-positive RGC in sham controls ( Fig.  1H-J, Table 1). Sholl analysis confirmed a global decrease in the number of branch intersections indicative of reduced arbor complexity (Fig. 1K).
Next, we asked whether ocular hypertension-induced dendrite retraction correlated with loss of excitatory synapses. We examined changes in endogenous vesicular glutamate transporter 1 (VGLUT-1), a presynaptic protein expressed at bipolar ribbon synapses [37], and postsynaptic density protein 95 (PSD95) in the inner plexiform layer, where RGC dendrites are located. A pronounced decrease in the expression of both VGLUT1 and PSD95 was observed at 2 weeks after glaucoma induction relative to control retinas ( Fig. 1L-O). Quantitative analysis of pre-and postsynaptic voxels, which represent the three-dimensional volume occupied by (See figure on previous page.) Fig. 1 Ocular hypertension promotes early RGC dendrite retraction and synapse loss. A Schematic of magnetic microbeads injection into the mouse anterior chamber to induce ocular hypertension (OHT). B This procedure results in gradual increase of intraocular pressure (IOP) leading to RGC soma and axon loss (ANOVA with Tukey's multiple comparison post-hoc test, ***p < 0.001, N = 10-14 mice/group). C, D Quantitative analysis demonstrates a statistically significant loss of RGC soma and axons at 3 weeks after microbeads injection (ANOVA with Tukey's multiple comparison post-hoc test, n.s. = not significant, ***p < 0.001, N = 5-14 mice per group). (E) YFP-positive RGC with a clearly identifiable axon (arrowhead) were selected for dendritic arbor imaging and 3D reconstruction. F, G Representative examples of RGC dendritic arbors from non-injured (sham) and glaucomatous retinas at 2 weeks after induction of OHT. H-K Quantitative analysis of dendritic parameters reveals a substantial reduction in dendritic area, length, number of branches, and complexity (Sholl analysis) in OHT retinas (blue) relative to sham controls (white) (Student's t-test, **p < 0.01, ***p < 0.001, N = 6 mice/group, n = 40-51 RGC/group Table 1). L-N Glutamatergic synapses are visualized in the inner plexiform layer (IPL) on retinal cross-sections using immunolabeling against PSD95 (green) and VGLUT1 (red), post-and pre-synaptic markers, respectively. OHT injury (2 weeks) induces a pronounced loss of both VGLUT1 and PSD95 expression in the IPL. O High magnification of inset in (N) shows VGLUT1 and PSD95 puncta in the IPL at the level of the OFF sublamina in OHT and sham retinas. P Quantitative analysis of pre-and postsynaptic colocalized voxels, which measured the three-dimensional volume occupied by both VGLUT1 and PSD95 in the IPL, confirms that OHT promotes a striking loss of synapses (Student's t-test, ***p < 0.001, N = 4 mice/group). Values are expressed as the mean ± S.E.M VGLUT1 and PSD95 in the inner plexiform layer, showed a 65% decrease in glaucomatous retinas compared to non-injured controls (Fig. 1P). These data confirm that RGC undergo dendritic retraction and synapse elimination in response to ocular hypertensive stress.

Glaucoma-induced energy stress triggers AMPK activation
Energy stress is characterized by low ATP and high AMP intracellular levels. When AMP binds to AMPK, it promotes allosteric changes that enhance phosphorylation of threonine 172 (Thr172) on the activation loop of the α subunit by the upstream Liver Kinase B1 (LKB1) [38,39]. To investigate whether AMPK serves as a sensor of energy stress in glaucoma, we first performed western blot analysis of retinal homogenates using an antibody that recognizes phosphorylated AMPK on Thr172 (pAMPK Thr172 ). We observed a significant increase in retinal pAMPK Thr172 at 1 week after microbead injection compared to sham-injected controls ( Fig. 2A, B) concomitant with enhanced LKB1 activity (Fig. 2C,  D). Retinal immunohistochemistry showed pAMPK Thr172 increase in ocular hypertension in several cell types, consistent with the highly conserved nature of AMPK as an energy sensor and metabolic regulator [40], including RGC visualized with the cell-specific marker Brn3a (brain-specific homeobox/POU domain protein 3A) [41] ( Fig. 2E-H). Quantification of active AMPK in RGC was performed by co-localization of anti-pAMPK Thr172 with the RGC-specific markers Brn3a or RBPMS [41,42]. The number of pAMPK Thr172 -positive RGC and the intensity of pAMPK Thr172 epifluorescence per cell confirmed enhanced AMPK activity in these neurons (Fig.  2I, J). pAMPK Thr172 labeling, indicative of AMPK activation, was also detected in intraretinal RGC axons and increased with glaucoma (Fig. S2). Next, we sought to determine whether AMPK is activated in human glaucoma hence serving as an indicator of metabolic stress. For this purpose, retinal sections from 27 surgical eye specimens of patients diagnosed with primary open angle glaucoma and 15 age-matched controls were labeled for pAMPK Thr172 (Table 2). A substantial increase of pAMPK Thr172 in RGC was found in individuals with glaucoma relative to controls (Fig. 2K, L). Co-labeling of pAMPK Thr172 with RBPMS revealed robust AMPK activation in these neurons (Fig. 2M, N). Quantification of pAMPK Thr172 -positive epifluorescence confirmed increased AMPK activity in RGC from glaucoma patients (Fig. 2O, n = 100 RGC/human retinal specimen, Student's t-test, p < 0.001, see Methods).
To test whether AMPK activation mediates dendritic pathology, we used compound C, a small molecule cellpermeable AMPK blocker [43]. Compound C or vehicle were administered by intraperitoneal injection, once a week for 2 weeks after glaucoma induction, followed by analysis of AMPK activity, RGC dendritic morphology, and RGC density. Compound C administration effectively reduced AMPK activation (Fig. 2P) and promoted robust protection of RGC dendrites preserving arbor area, process length, and complexity relative to shamoperated controls (Fig. 2Q-V). In addition, compound C effectively protected RGC soma from ocular hypertension damage (Fig. 2W, X). To rule out adverse effects of compound C including vascular leakage, we examined blood-retinal-barrier (BRB) integrity in the presence of fluorescein, a small molecular weight tracer. Retinal imaging showed no fluorescein extravasation in eyes injected with compound C or vehicle (Fig. S3A, B). In (See figure on previous page.) Fig. 2 Glaucoma-induced energy stress triggers AMPK activation. A, B Western blot and densitometry analysis of retinal homogenates demonstrate a substantial increase in active AMPK (pAMPK Thr172 ), a readout of metabolic stress, as early as 1 week after induction of OHT (Student's t-test, * = p<0.05, N = 5 mice/group). The lower panel is the same blot probed with an antibody against total AMPK and β-actin for normalization. C, D Western blot and densitometry analysis of retinal homogenates show increased LKB1 activity (pLKB1) in glaucomatous retinas (Student's t-test, ** = p<0.01, N = 7-9 mice/group). The lower panel is the same blot probed with an antibody against total LKB1 and β-actin for normalization. E, F Immunohistochemical analysis of mouse retinal cross sections with an antibody against pAMPK Thr172 , reveals robust AMPK activity in retinal cells subjected to OHT. G, H Co-labeling with antibodies against pAMPK Thr172 and the RGC-specific marker Brn3a demonstrates AMPK hyperactivity in RGC. I, J Quantification of the number of pAMPK Thr172 -positive RGC as well as epifluorescence intensity per neuron confirms a significant increase in AMPK activity (Student's t-test, ** = p<0.01, N = 5 mice/group, n = 50 RGC/group). K, L pAMPK Thr172 retinal immunostaining of primary open angle glaucoma patients and age-matched controls ( Table 2) Table 1). O, P Rapamycin also blocks siAMPK-mediated rescue of synapses, visualized with the post-and pre-synaptic markers PSD95 and VGLUT1, respectively. Q Quantification of synaptic voxels confirms that siAMPK-induced synaptic protection is abolished by rapamycin, confirming that this process is mTORC1 dependent (Student's t-test, ***p < 0,001, N = 5 mice/group). Values are expressed as the mean ± S.E.M. IPL: Inner Plexiform Layer, GCL: Ganglion Cell Layer contrast, marked fluorescein extravasation was found in ischemic retinas subjected to central retinal artery ligature, a procedure known to compromise the BRB [31] and used as positive control (Fig. S3C). Consistent with this, we did not detect T cell infiltration in retinas treated with compound C (Fig. S3D), confirming that this agent does not promote vascular leakage. Collectively, our findings indicate that AMPK is markedly activated in glaucoma, a sign of metabolic stress, and that AMPK plays a role in RGC dendritic pathology.

AMPK hyperactivation promotes dendritic retraction and synapse loss in glaucoma
To rule out potential off target effects of pharmacological AMPK inhibition [44], we selectively inhibited AMPK function using a short interfering RNA (siRNA) that targets the α catalytic subunit, which is essential for AMPK activation [39]. First, we examined whether siRNA delivered intraocularly was taken up by murine RGC. A single intravitreal injection of non-targeting Cy3-tagged control siRNA resulted in RGC labeling as early as 3 h after administration (Fig.  S4A-D). The mechanism for this preferential uptake is unknown, but likely reflects the proximity of the ganglion cell layer to the vitreous chamber allowing siRNA diffusion. Next, we assessed the efficacy of a targeted siRNA against AMPK (siAMPK) to knockdown retinal AMPK protein levels and activity. Western blot analysis showed that intravitreal administration of siAMPK significantly reduced total and phosphorylated (active) AMPK, while a non-targeting control siRNA (siCTL) had no effect (Fig. 3A-C).
To assess the role of siAMPK on dendrite morphology, each siRNA (siAMPK or siCTL) was independently administered by intravitreal injection once a week for a total of 2 weeks after glaucoma induction followed by characterization of RGC morphology (Fig. 3D). Retinal immunohistochemistry confirmed a reduction in the number of pAMPK Thr172 -positive RGC at 2 weeks of ocular hypertension in eyes treated with siAMPK relative to siCTL (Fig. 3E-G). Our data show that siAMPKtreated RGC had longer dendrites and more elaborate arbors than control neurons treated with siCTL (Fig.  3H, I). Quantitative analysis revealed that siAMPK protected dendritic field area (90%), length (86%), and number of branches (93%) while preserving arbor complexity during glaucoma-induced damage relative to siCTL (length: 66%, area: 62%, branches: 80%) (Fig. 3J-M, Table  1). siAMPK also rescued VGLUT1 and PSD95 expression in the inner plexiform layer (Fig. 3N-P), and quantitative analysis of pre-and postsynaptic voxels confirmed robust synaptic protection compared to controls (Fig.  3Q). Based on these findings, we conclude that early AMPK activation promotes RGC dendritic damage and synapse loss in glaucoma.

AMPK knockdown rescues RGC dendrites and synapses through mTORC1 activation
AMPK is a potent inhibitor of mTORC1 [45], which we previously showed is a key regulator of RGC dendritic morphology [34,46]. To test the hypothesis that AMPK mediates dendritic pathology through inhibition of mTORC1, we first examined whether arbor retraction correlated with changes in endogenous mTORC1 function. mTORC1 activates the p70-S6 kinase leading to phosphorylation of the ribosomal protein S6 at Ser240/244 residues (pS6 Ser240/244 ) thus stimulating protein translation [47]. Immunolabeling of noninjured sham retinas with pS6 Ser240/244 revealed two cell populations endowed with robust constitutively active mTORC1: one located in the ganglion cell layer and another in the inner nuclear layer (Fig. 4A). Robust mTORC1 activity, visualized through co-labeling of pS6 Ser240/244 and RBPMS, was found in non-injured RGC (Fig. 4B), and decreased markedly at 2 weeks of ocular hypertension (Fig. 4C-E). In the inner nuclear layer, pS6 Ser240/244 co-localized with the calciumbinding protein calbindin, a marker of horizontal cells (Fig.  S5A). In contrast to RGC, pS6 Ser240/244 labeling in horizontal cells remained unchanged after microbead injection indicating that glaucoma-induced mTORC1 downregulation was RGC-specific (Fig. S5B). Glaucomatous retinas treated with siAMPK had substantially more pS6 Ser240/244 -positive RGC (45%) than controls (Fig. 4F-H). Co-administration of siAMPK and rapamycin, an inhibitor of mTORC1 [30], blocked the effect of siAMPK on dendritic and synaptic rescue ( Fig. 4I-Q) confirming that these are mTORC1dependent processes. Administration of siAMPK or rapamycin alone in sham-operated retinas did not elicit significant changes (Fig. S5C, D, Table 1). None of the pharmacological or siRNA reagents used in this study altered intraocular pressure (Table 3). These data identify the AMPK-mTORC1 axis as a critical regulator of RGC dendritic arbor morphology in energetically stressed neurons.

AMPK attenuation promotes RGC functional recovery and survival
To assess the impact of siAMPK on RGC function, we measured the positive scotopic threshold response (pSTR), a component of the electroretinogram (ERG) that derives predominantly from RGC activity [48,49]. We followed light responses in the same mice before and after microbead injection and administration of siAMPK or siCTL using the regimen outlined in Fig. 3D. Figure 5 shows representative pSTR recordings before (Fig. 5A, yellow trace) and after microbead injection in mice treated with siCTL (Fig. 5B, grey trace) or siAMPK (Fig. 5C, red trace) relative to contralateral eyes (black trace). The pSTR amplitude diminished in siCTL-treated glaucomatous mice while siAMPK preserved retinal responses ( Fig. 5D) indicative of RGC functional recovery. Defective anterograde transport along RGC axons is an early pathological feature of glaucoma and a sign of unhealthy neurons [50]. To investigate whether AMPK attenuation improved axonal transport, the anterograde tracer cholera toxin B subunit (CTB) was injected intraocularly and its accumulation in the contralateral superior colliculus was quantified at 3 weeks of glaucoma induction (Fig. 5E). Unbiased stereological sampling demonstrated a substantial reduction of tracer accumulation in the superior colliculus of glaucomatous mice (Fig. 5F, G), while siAMPK substantially improved CTB transport (93%) relative to controls (62%) (Fig. 5H, I).
To determine whether attenuation of AMPK activity impacted the ability of RGC to survive after injury, we used the same regimen as above and quantified RGC soma and axonal density at 3 weeks of glaucoma induction. Retinas from eyes treated with siAMPK consistently displayed higher RGC soma densities than those treated with control siRNA (Fig. 5J-L). Quantitative analysis confirmed that siAMPK promoted RGC soma survival relative to control eyes (siAMPK: 93%, siCTL: 75%) (Fig.  5M, Table 4). To rule out the possibility that changes in retinal area caused by ocular hypertension could affect RGC density, we measured the total retinal area in sham-operated and microbead-injected eyes and found no significant differences (Fig. S5E). Analysis of optic nerve cross sections showed marked protection of RGC axons in siAMPK-treated eyes compared to control nerves (siAMPK: 98%, siCTL: 78%) (Fig. 5N-Q, Table 4). This level of axonal protection together with increased pAMPK Thr172 in RGC axons (Fig. S2) suggest that siAMPK might act on the axonal compartment. Early changes at the optic nerve head (ONH), including infiltrating cells and reactive astrocytes, have been propose to contribute to RGC damage in glaucoma [51]. Therefore, we examined the effect of siAMPK on microglia/ macrophages (Iba1) and reactive astrocytes (GFAP) at 2 weeks after glaucoma induction. Our data show that although Iba1-positive cells and GFAP immunoreactivity in the ONH increased with high intraocular pressure, siAMPK treatment did not alter this response (Fig. S6) suggesting that these events do not contribute to siAMPK-mediated RGC soma or axon protection. Collectively, these results demonstrate that attenuation of AMPK activity restores light-evoked retinal responses, improves anterograde axonal transport, and promotes robust RGC survival in glaucoma.
(See figure on previous page.) Fig. 5 AMPK attenuation promotes RGC functional recovery and survival. A-C Representative positive scotopic threshold response (pSTR) recordings prior to induction of ocular hypertension (pre-OHT, yellow trace) or after injury and treatment with siCTL (grey trace) or siAMPK (red trace). Pre-and post-OHT recordings are normalized relative to the contralateral, non-injured eye (black traces). D Quantitative analysis of the pSTR amplitude demonstrates RGC function rescue in siAMPK-treated eyes relative to controls at 3 weeks after OHT induction (ANOVA with Tukey's multiple comparison post-hoc test, * = p<0.05, N = 8-18 mice/group). E The tracer cholera toxin B subunit (CTB) conjugated to Alexa Fluor 488 is injected intravitreally and its accumulation in the contralateral superior colliculus is quantified as a readout of active anterograde axonal transport. F-H Unbiased stereological rostral-to-caudal sampling of the superior colliculi demonstrates a substantial reduction of the CTB-labeled target area in siCTL-treated mice relative to sham controls. In contrast, siAMPK-treated animals display an increase in brain CTB accumulation. I Quantification of the total CTB-positive area in the superior colliculus confirms a marked increase in anterograde axonal transport in siAMPKtreated mice compared to siCTL (ANOVA with Tukey's multiple comparison post-hoc test, * = p<0.05, N = 5-9 mice/group). J-M Brn3a-labeled flatmounted retinas show higher RGC soma density at 3 weeks of OHT following siAMPK treatment relative to controls (ANOVA with Tukey's multiple comparison post-hoc test, * = p<0.05, N = 5-12 mice/group). N-Q siAMPK also promotes survival of RGC axons, quantified in optic nerve cross sections, compared to retinas treated with siCTL (ANOVA with Tukey's multiple comparison post-hoc test, * = p<0.05, N = 5-6 mice/group). Values are expressed as the mean ± S.E.M

Discussion
RGC undergo substantial metabolic deficits in glaucoma, but how these neurons sense and respond to energetic stress is not well understood. Here, we show that RGC subjected to ocular hypertension display early signs of bioenergetic decline characterized by hyperactivation of the energy sensor AMPK. Our data show that persistent AMPK activity not only reflects metabolic strain but also exerts a marked detrimental effect on RGC structure and function. Indeed, AMPK triggered swift RGC dendritic retraction and synapse disassembly accompanied by loss of light-evoked responses and impaired axonal transport. These findings reveal that AMPK is a critical homeostatic hub which, if chronically active, contributes to RGC damage by shutting down energetically expensive processes, notably synaptic transmission and axonal transport. Mitochondrial deficits and vascular dysregulation have been proposed to compromise energy availability in glaucoma [18][19][20][21][22]. Age-dependent decline in retinal nicotinamide adenine dinucleotide (NAD) was reported in an inherited glaucoma model, and administration of the NAD precursor vitamin B3 promoted neuroprotection in mice and improved retinal function in patients [23,52]. A spectrum of mutations in mitochondrial DNA and nuclear genes encoding mitochondrial proteins have been reported in primary open angle and normal tension glaucoma patients [53][54][55][56][57]. We show a striking increase in AMPK activity (pAMPK Thr172 ) in RGC from both microbead-injected mice and patients affected by the disease, suggesting that AMPK is rapidly engaged in metabolically stressed RGC neurons. Of interest, a comprehensive molecular atlas of 46 RGC subtypes in the adult retina generated by single-cell RNA-seq [58] shows that Prkaa1 expression is found in many RGC subtypes including αON sustained and transient, αOFF sustained and transient, melanopsin-positive (M1, M2 and M5), and ON-OFF direction selective RGCs (oo-DS). The widespread Prkaa1 gene expression in most RGC classes suggests that AMPK plays an important role in the regulation of energy homeostasis beyond the αRGCs analyzed in this study.
Given the importance of AMPK to sense energy stress and mount an adaptive response, its complete inhibition can be detrimental. For example, irreversible AMPK blockade in hypoxic-ischemic brain damage was shown to exacerbate neuronal death [59]. The siRNA-based approach used here, which conferred only partial inhibition of AMPK, was surprisingly effective to protect RGC dendrites/synapses and rescue light-evoked responses. This approach is likely beneficial because it reduces the damaging consequences of AMPK overactivation while preserving its homeostatic functions. In line with this, strategies that attenuate AMPK activity rescued cognitive deficits in Alzheimer's disease models [60,61]. These findings suggest that the timing and duration of AMPK modulation is critical for a successful outcome in neurodegenerative diseases with a metabolic stress component.
Although AMPK has been implicated in models of brain injury and neurodegeneration [62], it is not clear how sustained AMPK activation is deleterious to neurons. We previously showed that mTORC1 is essential for RGC dendritic maintenance and regeneration [34,46]. Here, we demonstrate that ocular hypertensioninduced AMPK hyperactivity strongly inhibits mTORC1 triggering early dendrite and synapse loss in vulnerable neurons. Indeed, AMPK knockdown restores mTORC1 activity and protects dendrites in injured RGC. These data identify AMPK as a critical regulator of mTORC1mediated maintenance of synaptic connections in energetically stressed neurons. Our results contrast with previous findings showing that the mTOR inhibitor rapamycin promoted RGC survival [63]. Rapamycin has been shown to have both beneficial and detrimental effects on metabolism, a disparity that has been largely attributed to non-specific effects depending on the timecourse of administration [64,65]. Specifically, compensatory pathways are activated in response to chronic rapamycin treatment [66,67]. Here, we used a narrow therapeutic window to minimize off target effects (2 weeks), however sustained daily rapamycin treatment can promote neuroprotection through activation of off target pathways rather than mTORC1 inhibition [68].

Conclusions
Our study identifies AMPK as a critical molecular link between metabolic stress and RGC dysfunction, and highlights the importance of intervention therapies that capitalize on the restoration of energy homeostasis for glaucoma.