Phosphatidylserine Decarboxylase Regulates Retinal Ganglion Cell Neurite Outgrowth with Altered Somal Membrane Fluidics and Mitochondrial Morphology

Abstract

Mitochondrial lipid metabolism is an emerging regulator of neuronal regeneration, yet its role remains poorly defined. We investigated the function of phosphatidylserine decarboxylase (PSD), a mitochondrial enzyme that converts phosphatidylserine to phosphatidylethanolamine (PE), in retinal ganglion cell (RGC) regeneration. Using human glaucomatous degenerating optic nerves, we found PE was aberrantly accumulated with an elevated PSD expression and activity. In contrast, transcriptomes of regenerating RGCs present downregulated PSD, implicating PSD as a potential negative regulator of axonal growth. Using AAV2-mediated gene modulation, we evaluated how PSD knockdown (PSDKD) and PSD overexpression (PSDOE) alter RGC neurite outgrowth in vitro while evaluating effects on mitochondrial morphology, membrane fluidity by C-Laurdan staining, and lipidomes by LC-MS analysis. PSDOE did not support RGC neurite outgrowth, fragmented mitochondria, and increased polyunsaturated triacylglycerols. PSDKD significantly enhanced RGC neurite outgrowth and increased somal membrane fluidity accompanied by decreased cholesterol and saturated triacylglycerols. Notably, Doxorubicin, which attenuates PSD activity, increased neurite growth in PSDOE RGCs, supporting PSD’s activity as a negative role for growth. Using the optic nerve crush degenerative model in C57BL/6 mice, we confirm PSDKD RGCs have higher growth competency in vivo. These findings indicate PSDKD positions RGCs in a more growth-permissive state.

1. Introduction

Lipid metabolism is integral to maintaining neuronal health and function. Disruption of lipid homeostasis is related to the development of neurodegenerative disorders, such as Parkinson’s and Alzheimer’s diseases [1,2,3]. Meanwhile, nerve regeneration involves precise lipid biosynthesis and trafficking to facilitate substantial membrane remodeling and membrane expansion [4,5,6]. The lipid enzymes involved in the axonal growth and regeneration process remain understudied as well as the lipid biophysical characteristics of growing axons.

Recent studies found glycerophospholipid synthesis as an essential metabolic pathway for regeneration [7,8,9]. Lipidomes of regenerating optic nerves have confirmed changes in glycerophospholipid classes [10,11,12,13,14]. Studies have focused on endoplasmic reticulum-based synthesis; however, glycerophospholipids, such as phosphatidylethanolamine (PE), can be synthesized via multiple organelle routes.

Glycerophospholipids are actively transported between organelles indicating widespread effects. The glycerophospholipid source may present unique axonal growth capacities as the lipid production alters organelle behavior. For example, mitochondrial glycerophospholipid metabolism can influence mitochondrial morphology and dynamics, which affect axonal growth and regeneration [15,16,17,18,19,20,21].

Membrane fluidity is an important biophysical property of growing axons, and there is evidence that high membrane fluidity supports axon regeneration [22,23,24,25]. However, these studies have been limited to the enzymatic sources studied. Lipid sources can produce lipids with unique characteristics, such as length, acyl-linkages, and unsaturation level. All these characteristics determine a lipid’s geometrical shape and intrinsic curvature that influence membrane fluidity [26,27,28,29,30,31]. Furthermore, lipid metabolism has downstream effects on how other lipids incorporate into the membrane. For example, a link between PE and cholesterol levels has been identified to maintain fluidity [32]. As such, regenerative studies that look at lipid metabolism should also consider membrane fluidic effects.

One understudied lipid metabolic enzyme is phosphatidylserine decarboxylase (PSD), which resides in the inner mitochondrial membrane and converts phosphatidylserine (PS) to PE. PSD’s expression influences mitochondrial morphology and dynamics as well [33,34]. PSD-derived PE primarily resides within mitochondria but is also exported via mitochondrial-associated membranes (MAMs) to the endoplasmic reticulum or directly interacts with autophagosomes and mitochondrial-derived vesicles [35,36]. Thus, PSD’s activity influences both mitochondrial and cellular lipid homeostasis.

In this study, we first evaluate lipid metabolism in human glaucomatous optic nerves and find PSD expression is pathologically upregulated leading to increased PE levels and decreased phosphatidylcholine and PS. Inversely, based on available transcriptomic profiles, PSD expression was downregulated in regenerating retinal ganglion cells (RGCs), whose axons make up the optic nerve. We further investigate PSD’s effects on regulating axonal growth, mitochondrial morphology, and membrane fluidity in RGCs.

Using AAV-mediated gene modulation and lipidomic profiling, we characterize the impact of PSD knockdown (PSDKD) and overexpression (PSDOE) on isolated RGC membrane structure and in vitro and in vivo growth. We employ a unique approach of using C-Laurdan dye to visualize membrane fluidity shifts of isolated RGCs. Functionally, PSDKD enhances neurite outgrowth and increases somal membrane fluidity accompanied by decreased cholesterol and altered triacylglycerols. PSDKD did not disrupt axonal mitochondrial morphology. In contrast, PSDOE induces mitochondrial fragmentation and does not support neurite extension. Using an in vivo optic nerve crush mouse model to study regeneration [37,38,39,40], we further confirm that PSDKD supports RGC regenerative capacity.

Our findings establish PSD as a regulator of lipid metabolism, mitochondrial morphology, and membrane dynamics that constrains RGC axonal growth and regenerative capacity.

2. Materials and Methods

2.1. AAV Construct and Use

All adeno-associated viruses (AAV) were designed and obtained from VectorBuilder (Chicago, IL, USA) as ultra-purified AAV2 serotype vectors, which are known to exhibit tropism for retinal ganglion cells and are widely employed in optic nerve regeneration studies.

The control (NEG) and overexpression (PSDOE) Vectors were generated using the following constructs:

• NEG: pAAV[Exp]-CMV>ORF_Stuffer:IRES:mCherry:WPRE (Vector ID:VB240217-1288tfn)
• PSDOE: pAAV[Exp]-CMV>mPisd[NM_177298.3]:IRES:mCherry:WPRE (Vector ID:VB900144-8505ksj)

For gene knockdown studies, shRNA-expressing and scrambled control vectors were purchased as follows:

• pAAV[shRNA]-mCherry-U6>mPisdshRNA or Scramble
• Scramble-VB010000-0024wah
• PSDsh1-VB240212-1188unn
• PSDsh2-VB240212-1190tpq
• PSDsh3-VB240212-1194jxb

All vectors were verified for sterility, purity, and absence of mycoplasma contamination. Physical genome copy titers were determined by quantitative polymerase chain reaction (qPCR) and are summarized in

Table 1. AAV sequence information and estimated viral titers by qPCR.

AAV	ORF/Target Sequence	Estimated Titer (×1013 GC/mL)
NEG	ORF_Stuffer, mCherry	4.92
PSDOE	mPisd[NM_177298.3], mCherry	5.48
Scramble	CCTAAGGTTAAGTCGCCCTCG, mCherry	3.36
PSDsh1	TCCTACAATGACCTGAGCTTT, mCherry	2.18
PSDsh2	CCCTGTCACTATGAATCTACT, mCherry	3.10
PSDsh3	CAGGTGTCAGAAATTTCCATA, mCherry	1.80

.

2.2. Retinal Ganglion Cell Isolation, Culture, and Transduction

Retinal ganglion cells were purified from postnatal pups using published methodology [41]. This immunopanning process results in >95% purity of RGCs due to the Thy1 marker. We verified purity by Beta-Tubulin III and Brn3a RGC marker staining compared to nuclear staining. Postnatal day 3 mouse pups were euthanized by decapitation, and retinas were dissected. Tissues were enzymatically dissociated in papain (37 °C, 30 min) followed by gentle tituration. Cell suspensions were filtered and immunopanned using a three-step dish protocol: two negative selections followed by positive selection on anti-Thy1-coated dishes. Purified RGCs were counted and plated in well dishes with Neurobasal/SATO growth media supplemented with CNTF, BDNF, Forskolin, FGF, and GDNF. Half of the volume was used during plating; the remaining half was used for viral transduction 30 min later. RGCs were maintained for 5 days in vitro (DIV) at 37 °C in 5% CO₂, with media replaced every other day (half-volume change). A multiplicity of infection of ~125,000 was used for AAV transductions. Primary neurons, especially RGCs, are difficult to transduce and require a high MOI ~100,000+ to ensure transduction and sufficient expression within 5 days. Thirty minutes after plating, the appropriate volume of concentrated AAV was diluted in prewarmed, equilibrated complete media and gently added to the wells. Transduction conditions were optimized in preliminary experiments.

2.3. RGC Immunocytochemistry

RGCs were fixed in 4% paraformaldehyde in media (20 min, 37 °C), washed with PBS, and permeabilized with 0.3% PBS-Triton X100 (PBST) (Thermo Fisher Scientific, Waltham, USA) for 60 min. RGCs were blocked with 5% serum in 0.3% PBST for 60 min and incubated overnight at 4 °C with primary antibody (1:500 β Tubulin III Abcam ab18207; 1:500 TOMM20 Abcam ab186735). Following washes, secondary antibodies (Goat Anti-Rabbit Alexa 488 Abcam ab150077), NucBlue Stain, and/or Phalloidin 647 were applied (2 h, room temperature, orbital shaker), followed by PBS washes.

2.4. Neurite Outgrowth Quantification

Images were acquired at 10× magnification using a Leica Stellaris 5 confocal microscope (Leica, Durham, NC, USA) (GFP channel). Neurite lengths were traced in FIJI (ImageJ version 2.14.0/1.54f) using the NeuronJ plugin (v1.4.3), based on β-Tubulin III immunostaining. Images were gray-scaled, inverted, and scaled prior to analysis.

2.5. RGC Mitochondrial Morphology Analysis

RGCs were seeded in 8-well coverglass-bottom chambers. At 5 DIV, immunocytochemistry images of transduced RGCs were taken using a 63× oil-immersion objective on a Leica Stellaris 5. Only RGCs with axonal growth were used for analysis. TOMM20 signal was segmented using auto local thresholding (Bernsen, 20 pixel) in FIJI. Axonal regions were traced manually using the Phalloidin 647 channel. Mitochondrial features were quantified using the MitochondriaAnalyzer plugin (v2.3.1) on ImageJ, treating mitochondria as data points. Metrics included mitochondrial length (the longest shortest distance) and form factor. Statistical comparisons to the NEG control were performed using one-way ANOVA in GraphPad Prism 10.

2.6. RGC Lipid Extractions

Lipids were extracted from cultured RGCs using a butanol–methanol (BUME) with lithium chloride protocol. Culture medium was aspirated, and cells were gently washed with PBS. A 3:1 1-butanol/methanol mixture was directly added to wells, followed by mechanical scraping to release cellular content. The extract was mixed with 3:1 heptane:ethyl acetate (v/v), vortexed for 1 min, then repeated with additional heptane:ethyl acetate. Phase separation was induced by adding 50 mM Lithium Chloride and vortexing for 1 min. The samples were centrifuged at 2700× g for 10 min. The upper organic phase was collected, and the lower aqueous layer was re-extracted by phase separation twice more with 3:1 heptane/ethyl acetate. Combined organic layers were dried under vacuum using a SpeedVac concentrator (Labconco, Kansas City, MO, USA) and stored at −80 °C under argon until lipidomic analysis.

2.7. C-Laurdan Staining and Imaging

For membrane fluidity measurements, RGCs were plated in 8-well coverglass-bottom chambers. At 5 DIV, half of the culture medium was removed and replaced with C-Laurdan media (10 µM final concentration, pre-equilibrated with 5% CO₂). Cells were incubated at 37 °C for 4 h. After staining, cells were fixed in 4% paraformaldehyde and 4% sucrose in PBS for 15 min at 37 °C and washed 3× with PBS.

C-Laurdan images were acquired on a Leica Stellaris 5 confocal microscope using a 63× oil-immersion objective. Excitation was at 405 nm, and emission was collected using two HyD detectors: 415–455 nm (order phase) and 490–530 nm (disorder phase). Z-stacks were acquired at 0.5 µm intervals. Laser power was optimized to prevent saturation while maintaining consistent emission intensities across samples. Maximal intensity projections were used for general C-Laudan analysis, and single z-planes were used for 3D color mapping.

Method validation was performed in HEK293T cells treated with methyl β-cyclodextrin to disrupt ordered lipid raft domains [42].

2.8. C-Laurdan G Correction Factor

C-Laurdan calibration was performed using a 5 mM solution of C-Laurdan (Tocris BioScience Bristol, UK) in DMSO imaged under identical gain settings at three laser powers (low, medium, high) corresponding to experimental conditions (GP_m). Mean pixel intensities from Ch00 (ordered) and Ch01 (disordered) channels were used to compute GP values:

$${\mathrm{G}\mathrm{P}}_{\mathrm{m}}={\displaystyle \frac{{\overline{\mathrm{I}}}_{\mathrm{C}\mathrm{h}00}-{\mathrm{G}\overline{\mathrm{I}}}_{\mathrm{C}\mathrm{h}01}}{{\overline{\mathrm{I}}}_{\mathrm{C}\mathrm{h}00}+\mathrm{G}{\overline{\mathrm{I}}}_{\mathrm{C}\mathrm{h}01}}}$$

The correction factor (G) was computed using a target GP reference value (GP_r) of 0.25 to center the GP distribution at GP of 0. The value is influenced by the specific solvatochromic dye used (C-Laurdan) and the microscope setup. A value of 0.25 was determined for our microscope system with C-Laurdan after preliminary experiments with HEK293T cells and basal readings.

$$\mathrm{G}={\displaystyle \frac{{\mathrm{G}\mathrm{P}}_{\mathrm{r}}+{\mathrm{G}\mathrm{P}}_{\mathrm{r}}{\overline{\mathrm{G}\mathrm{P}}}_{\mathrm{m}}-{\overline{\mathrm{G}\mathrm{P}}}_{\mathrm{m}}-1}{{\overline{\mathrm{G}\mathrm{P}}}_{\mathrm{m}}+{\mathrm{G}\mathrm{P}}_{\mathrm{r}}{\overline{\mathrm{G}\mathrm{P}}}_{\mathrm{m}}-{\mathrm{G}\mathrm{P}}_{\mathrm{r}}-1}}$$

For our system, this yielded a G factor of 0.899. This C-Laurdan G correction factor corrects the GP calculation for any dye background signal within our specific confocal microscope setup.

2.9. C-Laurdan Image Analysis

Background signals were obtained from RGCs without C-Laurdan staining and applied in image processing. All ch00 and ch01 images were preprocessed using FIJI with rolling ball background subtraction (radius = 50 px). This created the most consistent image sets to account for differences during imaging. Full-cell masks were manually drawn by blind associates on the disordered (Ch01) channel, with soma and neurite masks segmented from the full-cell mask. An adapted ImageJ macro ([42], GitHub: https://github.com/MeehanS, accessed on 1 August 2023) was used for GP map generation and calculation.

GP intensity histograms (256 bins, range: −1 to 1) were normalized by total pixel count to yield distribution profiles. Average GP values were calculated by multiplying normalized bin counts by corresponding GP values and averaging across the full GP distribution. Bins with zero-pixel counts were excluded from averaging.

For visualization, hue-saturation-brightness (HSB) maps were built using the disordered channel as intensity reference. For volumetric renderings, GP images were generated per z-slice, LUTs (16-color or UnionJack) applied, and reconstructions performed in FIJI’s 3D Viewer (v5.0.1). Voxel depth was scaled to 3× the XY pixel size.

2.10. BE2-M17 Neuroblastoma Cell Culture, Transfection, and Puromycin Selection

BE2-M17 Neuroblastoma cells (CRL-2267 ATCC, Manassas, VA, USA) were maintained at 37 °C in a 5% CO₂ Incubator with media consisting of a 1:1 mixture of Eagle’s Minimum Essential Medium (MEM) and F12 Medium with 10% Fetal Bovine Serum, 1× Antibiotic-Antimycotic, 1 mM Sodium Pyruvate, and 2 mM L-glutamine. Neuroblastoma cells were transfected using Endofectin Max™ and the prescribed plasmid DNA quantities depending on the cell dish sizes used. Neuroblastoma cells transfected with pReceiverlv-CMV-ORF-IRES2-mCherry(or GFP)-IRES-puromycin (Genecopoeia, Rockville, MD, USA) were exposed to 3 ug/mL of Puromycin, which preliminary tests showed removal > 95% of Neuroblastoma cells within 3 days.

2.11. Mitochondrial Isolation

Neuroblastoma cells were grown on 15 cm culture dishes to ~80% confluency. Neuroblastoma cells were subsequently transfected and puromycin-selected. Mitochondria were isolated from BE(2)-M17 neuroblastoma cells using published methods [43,44,45]. Cells were detached with 1 mM EDTA in PBS for 5 min at 37 °C and collected using a cell scraper. After centrifugation at 600× g for 10 min at 4 °C, cells were washed once in PBS. Cells were resuspended in a hypotonic swelling buffer (10 mM NaCl, 1.5 mM MgCl₂, 10 mM Tris-HCl, pH 7.5, with cOmplete mini protease inhibitor cocktail, Roche) and subjected to one freeze-thaw cycle. Cells were homogenized with 60 strokes using a polypropylene pestle. Nuclei were pelleted by centrifugation at 1300× g for 10 min at 4 °C, and the supernatant was retained. The nuclear pellet was re-homogenized in fresh hypotonic buffer and centrifuged again. Supernatants were pooled and centrifuged three times at 1300× g for 3 min each to remove remaining debris and nuclei. Mitochondria were pelleted at 15,000× g for 10 min at 4 °C. The mitochondria pellet was washed with mitochondrial isolation buffer (10 mM HEPES, 220 mM mannitol, 70 mM sucrose, 100 mM KCl, 0.5 mM EGTA, 2 mM MgCl₂·2H₂O, 5 mM K₂HPO₄·3H₂O, pH 7.4) and re-centrifuged to remove residual hypotonic buffer. Mitochondrial protein content was determined by densitometry of GelCode™ Blue-stained protein spots, using a BSA standard curve. Mitochondrial content was used for mitochondrial phospholipidomics or PSD activity assay.

2.12. Phosphatidylserine Decarboxylase (PSD) Activity Assay

PSD activity was assayed by quantifying the conversion of phosphatidylserine (PS) to phosphatidylethanolamine (PE), as previously described [46]. Two micrograms of mitochondria were used per experimental group. NEG and PSDOE had 6 biological replicates each. Reactions were conducted in 50 mM imidazole-HCl buffer containing 5 mM brain-derived PS and 36 mM Triton X-100. Samples were incubated at 37 °C for 30 min. Control reactions were used to establish baseline PE levels, which were determined to be negligible relative to experimental conditions.

Lipid extraction was performed using a butanol–methanol and lithium chloride protocol [47]. A 3:1 (v/v) n-butanol/methanol mixture was added to the activity reaction mixture, vortexed, followed by the addition of 3:1 heptane/ethyl acetate (v/v) and additional vortexing. Phase separation was induced by adding 50 mM lithium chloride, and samples were centrifuged at 2700× g for 10 min. The upper organic layer was collected, and the aqueous phase was re-extracted twice using the same procedure. Combined organic phases were dried under vacuum in a SpeedVac concentrator (Labconco, Kansas City, MO, USA) and stored under argon gas at −80 °C until mass spectrometry analysis.

2.13. Neuroblastoma Mitochondrial Morphology Staining and Analysis

Neuroblastoma cells were grown on coverglass-bottom 8-well plates. Mitochondria in BE2-M17 Neuroblastoma cells were stained with 300 nM MitoTracker™ DeepRed (ThermoFisher, Eugene, OR, USA) for 30 min at 37 °C in vitro. Cells were then washed with pre-warmed culture medium and fixed in 4% PFA in culture medium for 15 min. After fixation, cells were washed thoroughly with PBS.

Confocal imaging was performed using a Leica Stellaris 5 microscope with a 63× oil-immersion objective at 0.5 µm z-step intervals. Only flat, adherent neuroblastoma cells isolated from large cell clusters displaying complete mitochondrial networks were selected for analysis (n ~ 50 cells per treatment). Transfected cells were identified by GFP expression.

Image preprocessing included background subtraction, unsharp masking (radius = 2, 0.6 mask), and auto-local thresholding (Bernsen, 10 pixel). Background signal was evaluated using secondary antibody-only controls and background signal within the nucleus. Only neuroblastoma cells with a substantial mitotracker signal above the background signal that could be accurately segmented were included. Mitochondrial morphology was quantified using the MitochondriaAnalyzer plugin in FIJI using complete cell analysis. Parameters measured included mean branch length, mean form factor, and the number of branches per mitochondrion. Comparisons between control (NEG) and PSD overexpression (PSDOE) cells were performed using unpaired t-tests in GraphPad Prism 10.

2.14. Neuro2a Cell Culture and Transfection

Neuro2a cells (CCL-131 ATCC, Manassas, VA, USA) were maintained at 37 °C in a humidified 5% CO₂ Incubator. Cells were cultured in Eagle’s Minimum Essential Medium (Thermo Fisher Scientific, Grand Island, NY, USA), supplemented with 1 mM Sodium pyruvate, 2 mM L-glutamine, 10% Fetal Bovine Serum, and 1× Antibiotic-Antimycotic. Transfections were performed using Endofectin Max™ (Genecopoeia, Rockville, MD, USA) according to the manufacturer’s instructions, and DNA quantities were scaled based on culture vessel size.

2.15. Neuroblastoma and Neuro2a Cell Lysis, Protein Estimation, and Western Blot

BE(2)-M17 Neuroblastoma and Neuro2a cells were lysed in buffer containing 20 mM TrisHCl, 50 mM NaCl, 1% Sodium Dodecyl Sulfate, 1× cOmplete Protease Inhibitors (Roche). Cells were homogenized using the Precellys Homogenizer (Bertin Technologies, Montigny-le-Bretonneux, France). Cell lysate was centrifuged at 14,000× g for 10 min at 4 °C.

Protein estimation was performed by densitometry. GelCode™ Blue-stained protein spots were quantified using ImageJ’s gel analysis tools and compared to a BSA protein ladder. Thirty-five micrograms of total protein resolved on a 4–20% Tris-glycine Gel at 80 V for 75 min with ice packs. Proteins were transferred to the PVDF membrane using the Biorad Turbo Transfer system.

Membranes were stained with Ponceau S solution to verify protein transfer, then destained with TBS washes and blocked for 1 h in LI-COR Blocking Buffer. Blots were incubated overnight at 4 °C with shaking using the following primary antibodies diluted in blocking buffer:

• Anti-PISD (1:500, Proteintech #16401-1-AP)
• Anti-PISD (1:500, Santa Cruz #sc-390070, H-2)
• Anti-GAPDH (1:1000, Abcam #ab8245)

After washing in TBS-T (0.1% Tween-20), membranes were probed with IRDye 680 or 800 secondary antibodies (1:5000, LI-COR) for 1 h at room temperature. Final washes included four rinses with TBS-T and one final rinse with TBS before imaging.

2.16. RNA Isolation and Quantitative Polymerase Chain Reaction (qPCR)

Total RNA was isolated from 200 k Neuro2a cells using the RNeasy Mini Kit (Qiagen, Düsseldorf, Germany) following the manufacturer’s instructions. Genomic DNA contamination was removed using the TURBO DNA-free kit (Invitrogen, Vilnius, Lithuania). Complementary DNA (cDNA) synthesis was performed using the High-Capacity RNA-to-cDNA Kit (Applied Biosystems, Vilnius, Lithuania), with incubation at 37 °C for 60 min followed by enzyme inactivation.

Quantitative PCR was carried out using PowerUp™ SYBR Green Master Mix (Thermo Fisher, Vilnius, Lithuania) and primers obtained from Bio-Rad (Pisd (qMmuCED0026090) and Actb (qMmuCED0027505)).

Amplification was performed on a QuantStudio™ 5 Real-Time PCR System under the following cycling conditions: 50 °C for 2 min, 95 °C for 2 min, followed by 40 cycles of 95 °C for 5 s and 60 °C for 30 s. Melt curve analysis was performed from 65 °C to 95 °C in 0.5 °C increments, with 5 s per step.

Actb, MAPK1, and GAPDH were all evaluated as housekeeping genes, but GAPDH and MAPK1 had poor melting curves and high variability. Actb primer was validated by amplification and melt curve analysis using no reverse transcriptase, no template control, and control runs. No Actb amplification appeared in the no template control, nor did any reverse transcriptase runs. PSD had no amplification in the no reverse transcriptase run, but had a Cq value of 33.683 in the no template control that was well above the measured Sample PSD Cq’s. Melt curve analysis of the control Actb run had a single peak at 82.7 °C. Melt curve analysis for the PSD primer showed a single peak at 84.8 °C. These results are consistent with Bio-Rad validation data. PSD primer efficiency was 99% efficiency with a R² value of 0.9986 and a Cq y-intercept of 35.44. Actb primer efficiency was 101% with a R² of 0.9996 and Cq y-intercept of 35.67.

Knockdown Efficiency was calculated using 2^−ΔΔCT where the housekeeping gene was Actb, PSD was the target gene, and the average Scramble ΔCT was used as the control. Sample qPCR runs were conducted as three technical replicates and three biological replicates. The qPCR experiment and analysis were repeated with another set of transfected Neuro2a cells to verify knockdown consistency. Three PSD-targeting shRNA constructs (PSDsh1-3) were tested. PSDsh3 achieved ~50% reduction while pSdhs1 and PSDsh2 showed variable knockdown (40–60%). Due to limited material, Western blot validation in primary RGCs was not performed. Potential off-target effects of shRNAs cannot be fully excluded, though PSD-related experimental results support specificity.

2.17. Liquid Chromatography–Mass Spectrometry Analysis–Retinal Ganglion Cells and Mitochondria

Dried Mitochondrial lipid extracts were reconstituted in 19 µL of isopropanol:acetonitrile (1:1, v/v) with 1 µL of EquiSPLASH™ LIPIDOMIX^® Quantitative Internal Standard (Avanti Polar Lipids, Cat# 330731, Alabaster, AL, USA). A 3 µL injection volume was used per run and three technical replicates per biological replicate.

Dried RGC lipid extracts were reconstituted in 9.5 µL of isopropanol:acetonitrile (1:1, v/v) with 0.5 µL of EquiSPLASH™ LIPIDOMIX^® Quantitative Internal Standard. Samples were sonicated in a water bath for 15 min to ensure complete solubilization. A 1 µL injection volume was used with one technical replicate per biological replicate. A smaller reconsitution volume was used to maximize the lipid signal.

Chromatographic separation was performed on a Vanquish Horizon UHPLC system (Thermo Scientific, Asheville, NC, USA) equipped with an Accucore Vanquish C18+ UHPLC column (150 mm × 2.1 mm, 1.5 µm particle size; Thermo Scientific).

Samples were run at a flow rate of 0.260 mL/min and a column temperature of 55 °C. with Mobile phase A: 50% acetonitrile, 50% water, 5 mM ammonium formate, 0.1% formic acid, and Mobile phase B: 88% isopropanol, 10% acetonitrile, 2% water, 5 mM ammonium formate, 0.1% formic acid. Ionization and detection were performed using a heated electrospray ionization (HESI) source coupled to a Q Exactive Orbitrap mass spectrometer (Thermo Scientific). Data was acquired in both positive and negative ionization modes using optimized parameters based on the manufacturer’s protocols. A quality control injection using EquiSPLASH was used to verify liquid chromatography–mass spectrometry performance.

2.18. Mass Spectrometry Lipid Identification and Quantification–Retinal Ganglion Cells and Mitochondria

Raw data files (*.RAW) were processed using LipidSearch 5.0 software (Thermo Scientific) with the following parameters:

Search settings: Mammalian lipid database

MS1 precursor tolerance: 5.0 ppm

MS2 product ion tolerance: 10.0 ppm

Product threshold: 1.0

MS1 absolute intensity threshold: 30,000

Top Rank Filter

M-Score Threshold: 5.0

Alignment settings:

Calculation method: Max

RT tolerance: 0.05 min

RT correction tolerance: 0.5 min

Signal-to-noise threshold: 3.0

Intensity ratio threshold: 1.5

Valid peak rate threshold: 0.5

Internal standard recovery was monitored across samples and remained between 70–130% of the batch mean. Lipid intensities were normalized to spiked-in EquiSPLASH™ deuterated standards. For mitochondrial lipid analyses, lipid quantities were also normalized to estimated mitochondrial protein content.

2.19. Statistical Analysis of Lipidomics Data

Lipid abundance data were log₁₀-transformed and Pareto-scaled using MetaboAnalyst 5.0 to approximate a Gaussian distribution. Total lipid class levels were calculated as the sum of all normalized lipid species within each class.

• Statistical comparisons:

  ◦

  Two-way ANOVA with multiple comparison correction was used to compare PSDOE, PSDsh1, PSDsh2, and PSDsh3 groups against NEG controls.

  ◦

  Mitochondrial lipidomes of NEG and PSDOE groups were compared using an unpaired Student’s t-test.

2.20. Mouse Husbandry and Experimental Design

All animal procedures were conducted in accordance with protocols approved by the University of Miami Institutional Animal Care & Use Committee (IACUC protocol number: IPROTO202300002266). Mice were maintained in the McKnight vivarium under AAALAC-accredited conditions, housed on a 12 h light/dark cycle with ad libitum access to standard chow and water. All mice used in this study were of the C57BL/6 background and sourced from the Jackson Laboratory. At study endpoints, animals were euthanized via CO₂ asphyxiation followed by exsanguination, in accordance with AAALAC guidelines.

Sample size determination was in line with relevant literature where sample sizes range from 3 to 6 for optic nerve regenerative research. All experimental mouse groups originated from the same 2 litters and equal male/female distribution (2M/2F). Each mouse was a biological replicate. Each experimental group was present in each mouse cage to control for differences in cage/location. Mice were randomly selected within the cages and provided a nonidentifiable numerical code during initial group allocation. All injections and optic nerve crush surgeries were performed blind to the mouse treatments. The order of the cages was randomly chosen for surgical treatments.

2.21. Mouse Intravitreal Injection

Intravitreal injections were performed using a 32-gauge needle to create a small entry site into the eye. Subsequently, 1 µL of Alexa Fluor™ 488-conjugated Cholera Toxin subunit b (CTB488; 2 µg/µL; Thermo Fisher Scientific) or 1 µL of concentrated AAV was injected into the vitreous chamber using a Hamilton syringe (Hamilton Company, Reno, NV, USA). The syringe bevel was oriented outward, angled toward the optic nerve head to avoid contact with the lens. The needle was held in place for approximately 10 s post-injection to allow for dye diffusion and to minimize reflux upon withdrawal.

2.22. Optic Nerve Crush

Optic nerve crush (ONC) was performed following previously established protocols. Mice were anesthetized with an intraperitoneal injection of ketamine (100 mg/kg) and xylazine (10 mg/kg). Under a dissecting microscope, the conjunctiva was incised using angled Vannas scissors to expose the posterior eye. The orbital tissues were gently retracted to access the optic nerve. The nerve was crushed approximately 2 mm posterior to the globe using calibrated forceps (Dumont #5) for 5 s, taking care to avoid damaging the ophthalmic artery. Mice exhibiting excessive bleeding or structural damage were excluded from the study. Following the procedure, ophthalmic antibiotic ointment (erythromycin) was applied to the eye, and mice received a single subcutaneous dose of sustained-release buprenorphine (0.1 mg/kg) for postoperative analgesia.

2.23. Pattern Electroretinography

PERG recordings were obtained using the Jörvec PERG system (Jörvec Visual Stimulation Box, M014760L, Miami, FL, USA). Mice were anesthetized as described. Electrodes were inserted subcutaneously: a recording electrode between the eyes near the nasal tip, a reference electrode at the scalp, and a ground electrode near the base of the tail. A balanced salt solution was applied to the corneal surface to maintain hydration. Mice were maintained at 37 °C using a temperature-controlled heating pad with rectal thermistor feedback (Physitemp TCAT-2LV). PERG responses were recorded as the average of three consecutive sweeps of 600 contrast reversals. Amplifier settings were: 10.0 K gain, 1.0 Hz high-pass filter, 100.0 Hz low-pass filter, and 360.0 µV artifact rejection. PERG amplitude was defined as the peak-to-trough voltage in microvolts (µV) of the major waveform components.

2.24. Intraocular Pressure Measurement by Tonometer

Intraocular pressure was measured using a rebound tonometer (TonoLab, Icare Finland, Helsinki, Finland) set to mouse mode. Mice were anesthetized as described. The head was gently stabilized in a neutral position, and measurements were performed with a clean probe placed 2–3 mm from the central cornea. For each eye, six consecutive technical readings were averaged into one measurement; this process was repeated four times per eye. Only measurements with acceptable variability (low standard deviation) were retained for analysis. Final IOP values represent the average of these four replicate measurements per eye per time point.

2.25. Ocular Tissue Dissection, Embedding, and Sectioning

At designated experimental endpoints, mice were euthanized by CO₂ inhalation followed by cervical dislocation in accordance with institutional animal care guidelines. Eyes (globes) and optic nerves were rapidly dissected, rinsed in phosphate-buffered saline (PBS), and fixed in 4% paraformaldehyde (PFA) for 2 h at room temperature. Tissues were then cryoprotected in a 30% sucrose solution at 4 °C until equilibration. Samples were embedded in optimal cutting temperature (OCT) compound and cryosectioned using a Leica cryostat (CM1950). Optic nerves were sectioned at 10 µm and globes at 14 µm thickness. Five sections were placed per slide and stored at −20 °C until further processing.

2.26. Immunohistochemistry of Ocular Tissues

Slides containing optic nerve and retinal tissue sections were air-dried and pre-warmed on a 55 °C hot plate for 10 min. Slides were post-fixed in 4% PFA for 15 min, followed by two washes in PBS (10 min each).

For retina sections, permeabilization was performed with 0.1% PBS-Tween (PBS-T) for 10 min, followed by blocking with 5% goat serum in PBS-T for 1 h at room temperature under gentle agitation. Sections were incubated overnight at 4 °C with the following primary antibodies diluted in 1% goat serum in PBS-T: anti-β-Tubulin III (1:200, Abcam ab18207) and anti-Brn3a (1:200, Abcam ab245230). The next day, slides were washed three times with PBS-T (10 min each) and incubated with secondary antibody (Goat anti-Rabbit Alexa Fluor 647, 1:1000 Abcam ab150079) in 1% goat serum in PBS-T for 2 h at room temperature with gentle shaking. After four final washes in PBS-T and one in PBS, sections were coverslipped using Vectashield Vibrance Mounting Medium containing DAPI (Vector Laboratories, Newark, CA, USA).

Optic nerve slides underwent the same PFA fixation and PBS wash steps and were also mounted with DAPI-containing medium for imaging.

2.27. Quantification of Regenerating Optic Nerve Axons

Axon regeneration was assessed using confocal imaging of optic nerve sections using a Leica Stellaris 5 confocal microscope with a 40× oil-immersion objective. Three to four z-stacks per optic nerve were acquired at comparable mid-section depths (z-step = 2 µm), followed by maximal intensity projection. Regenerating axons were manually quantified in ImageJ based on colocalization of mCherry and CTB-488 fluorescence signals at 100, 200, and 300 μm distances from the crush site. Only axons exhibiting both signals above background levels were included. For each optic nerve, axon densities were an average determined by three separate 10 μm optic nerve cryosections. Axon densities (axon count/mm) were calculated by the axon counts normalized by the optic nerve width. Associates were blinded during data collection and axon quantification.

2.28. Human Optic Nerve Procurement and Donor Criteria

Human donor tissues were obtained under Institutional Review Board approval (University of Miami) and in accordance with the Declaration of Helsinki. Cadaveric donor eyes were procured from the Midwest Eye Bank (Cincinnati, OH, USA) and Lions Eye Bank (Miami, FL, USA). Donors (both male and female, all Caucasian) included control and primary open-angle glaucoma (POAG) cases, with mean ages of 72.3 ± 5.9 and 70.3 ± 10.5 years, respectively.

Eligible donor eyes were limited to those enucleated within 21 h postmortem. Eyes were transported in phosphate-buffered saline (PBS) at 4 °C and shipped immediately. Dissections were performed within 36 h postmortem. Optic nerves (ONs) were transected at the nerve head, cleared of surrounding fat, washed with PBS, and stored at –80 °C. POAG diagnosis was confirmed via medical records including ≥ 2 visual field tests within a 1–2-year interval showing progressive loss. Control eyes were confirmed to lack any optic neuropathy history. Additional ON samples were collected for Western blot and enzyme activity assays.

2.29. Lipidomics Profiling of Human Optic Nerve Tissue

Lipid extraction was followed by reverse-phase chromatography using an Acclaim C30 column (150 × 2.1 mm, 3.0 µm; Thermo Scientific) on an HPLC Accela system coupled to a Q Exactive mass spectrometer (Thermo Scientific Asheville, NC, USA).

• Column temperature: 30 °C; injection volume: 5 µL; flow rate: 260 µL/min
• Solvent A: Methanol:H₂O (60:40, v/v) + 0.2% formic acid + 10 mM ammonium acetate
• Solvent B: Methanol:chloroform (60:40, v/v) + 0.2% formic acid + 10 mM ammonium acetate

A 20-min linear gradient from 15% to 65% solvent B was applied at a constant flow rate.

Heated electrospray ionization (HESI) was used with the following parameters:

• Spray voltage: 4.4 kV; capillary temperature: 350 °C; heater temperature: 275 °C
• S-lens RF level: 70; sheath gas: 45; auxiliary gas: 15

Full scan resolution was set to 70,000 at m/z 200 (AGC target: 1 × 10⁶, max injection time: 100 ms). For data-dependent acquisition, resolution was 17,500; isolation window: 1.3 m/z; dynamic exclusion: 3 s; collision energies: 30 and 19 eV (positive mode).

2.30. Human Optic Nerve Lipid Identification and Data Processing

Raw mass spectrometry files were processed in LipidSearch 4.1 with the following settings:

• Search tolerances: Parent and product ion, 5 ppm
• Filters: Top rank, main isomer peak, FA priority
• Quantification parameters: m/z tolerance: 5 ppm; RT tolerance: 0.5 min
• Adducts (positive mode): [M+H]+, [M+NH₄]+, [M+H–H₂O]+, [M+H–2H₂O]+, [M+2H]²+
• All lipid classes were selected for identification. Each sample was analyzed in technical triplicate.

For alignment, retention time tolerance was set to 0.1 min, with top rank and main isomer peak filters applied. Annotated peaks were merged by lipid species. Alignment scores were set to M-score ≥ 5.0, C-score ≥ 2.0, and FA chain identification enabled (types A and B).

Lipid data were grouped by lipid class, fatty acid, and molecular species. Glycerophospholipid class totals were analyzed using two-way ANOVA with multiple comparison correction. PE subclasses (acyl and plasmalogen) were compared by unpaired t-tests.

2.31. Western Blot Analysis of Human Optic Nerve Proteins

ON proteins were extracted and subjected to SDS-PAGE (4–15% gradient gels; Invitrogen) and transferred to PVDF membranes (Ready Gel, Bio-Rad, Hercules, CA, USA). Electrophoresis was performed at 70 V for 90 min. Membranes were blocked in 5% nonfat milk in TBS (pH 8.0).

Primary antibodies against PSD were used for detection. HRP-conjugated secondary antibodies (1:2000; goat anti-rabbit and anti-mouse, Abcam; donkey anti-sheep, Thermo Fisher) were applied. Blots were developed using chemiluminescent detection (GE ImageQuant LAS4000).

2.32. Enzymatic Activity Assays for Glycerophospholipid Biosynthesis

Enzymatic assays were performed in triplicate using human ON protein extracts from control and glaucomatous donors. Assays were evaluated by LC-MS-based lipidomics as described above.

• PSD assay:

2 µg of mitochondrial protein per replicate. Assay buffer: 50 mM imidazole-HCl, 5 mM phosphatidylserine (brain-derived), 36 mM Triton X-100. Reactions were terminated with chloroform:methanol (2:1, v/v), and lipids were extracted, dried under SpeedVac, and stored at −80 °C.

• PSS1 assay:

20 µg of ON protein was reconstituted in 50 mM Tris-HCl (pH 8.0), 0.6 mM MnCl₂, 0.5 mM L-serine, 0.2 mM CDP-diacylglycerol, 4 mM Triton X-100CaCl₂. Incubation: 20 min at 37 °C. Reaction was stopped and lipids extracted as above.

• PSS2 assay:

20 µg of ON protein in 100 mM CaCl₂, 40 mM hydroxylamine, 20 mM HEPES (pH 7.4), 20 µL each of L-serine and ethanolamine. Incubation: 20 min at 37 °C. Reaction was stopped and lipids extracted.

• PEMT assay:

20 µg of ON protein in 125 mM Tris-HCl (pH 9.2), 5 mM DTT, 1 mM Triton X-100, 2 mM PE, and 0.4 mM PMME. Incubation: 20 min at 25 °C. Reaction was terminated with chloroform:methanol (2:1, v/v) and 0.9% KCl. Lipids were extracted and stored.

Lipid product species were quantified and compared between control and glaucomatous ONs to determine enzyme activity rates. All comparisons were performed using two-way ANOVA.

2.33. Statistical Analysis

Statistical analyses were performed on GraphPad Prism 10 using t-test, One-Way ANOVA, or two-way ANOVA where appropriate. All error bars presented as +/− SEM unless stated otherwise. All ANOVA analyses were compared to NEG or Scramble controls with a multiple comparison correction applied using Dunnett’s test. Correlation matrices were generated using Pearson’s R coefficient on GraphPad Prism. RGC, Tissue, and Mitochondrial Lipidomic species data processing, PCA plots, and normalization plots were generated using Metaboanalyst 5.0.

3. Results

3.1. Upregulated Phosphatidylethanolamine and Phosphatidylserine Decarboxylase in Human Glaucomatous Optic Nerves

Neurodegenerative disorders, including ocular neuropathies such as glaucoma, can induce observable tissue-level lipidome changes [48,49,50]. Glaucoma is a progressive loss of retinal ganglion cell (RGC) axons resulting in loss of peripheral vision followed by central vision. Currently, treatments only stop or slow down disease progression but fail to restore the lost vision. We utilized human glaucomatous optic nerves as our neurodegenerative model to identify aberrant lipidomes and potential lipid metabolic enzymes associated with degenerative states.

We collected human optic nerve tissues from age- and sex-matched glaucoma and control donors. To reduce potential confounding variables, we maintained a postmortem interval of 36 h and an age range of 72.3 ± 5.9 and 70.3 ± 10.5 years for control and glaucoma patients, respectively. As the pathology and etiology are still being elucidated for lipid metabolism and glaucoma, we aimed to ensure that we had pathological tissues for analysis. To reduce variation, we need a confirmed primary open-angle glaucoma diagnosis confirmed by ≥2 visual field tests within a 1–2-year interval. This indicates progressive vision loss, and their enzymatic activity changes reflect pathological differences. Control eyes had no optic neuropathy history. Glycerophospholipid profiling, performed using Bligh and Dyer extraction followed by high-performance liquid chromatography-tandem mass spectrometry (LC-MS), revealed increased phosphatidylethanolamine (PE) and decreased phosphatidylserine (PS) and phosphatidylcholine (PC) in glaucomatous optic nerves (Figure 1A). We further investigated the increased PE species and found significantly increased acyl-linked chains in glaucomatous optic nerves (Figure 1B).

To investigate the imbalance of PC, PE, and PS groups, we evaluated the enzyme activity of proteins involved in the interconversion of acyl-linked lipids within these lipid classes. PS can be synthesized from PC and PE using phosphatidylserine synthase 1 (PSS1) and phosphatidylserine synthase 2 (PSS2), respectively. PE can be synthesized from PS using phosphatidylserine decarboxylase (PSD). PC is synthesized from PE by phosphatidylethanolamine N-methyltransferase (PEMT). Total protein content was determined by densitometry and used to calculate the appropriate tissue prep aliquot for each assay. We used established enzyme activity assays [46,51,52] and presented the enzymatic rate as pmol lipid product per minute normalized by the tissue protein content. In glaucomatous optic nerves, PSD activity was significantly upregulated while PEMT, PSS1, and PSS2 activities had no significant difference (Figure 1C). Compared to age- and sex-matched controls, Western blot analysis validated increased PSD expression in glaucomatous optic nerves (Figure 1D). This increased PSD expression and activity indicate PSD is producing a high PE flux leading to reduced PS and PC levels.

Phosphatidylserine decarboxylase (PSD), an inner mitochondrial membrane enzyme, converts PS to PE, preferentially using polyunsaturated sn-2 acyl chains. PSD is translated as an ⍺-β proenzyme that undergoes autocleavage after mitochondrial import.

Partial loss-of-function and missense variants of PSD are associated with Liberfarb’s syndrome, optic nerve atrophy, retinal atrophy, reduced electroretinogram, and decreased myelination [53,54]. Additional studies have found PSD activity regulates autophagy associated with neurodegenerative disorders, such as tauopathies [55,56]. Here, we discovered a unique finding of PSD overexpression associated with glaucomatous ocular neurodegeneration.

3.2. PSD Overexpression Fragments Mitochondrial Networks and Alters Their Glycerophospholipidomes

Within glaucomatous mouse models, RGC axonal mitochondria display abnormal, spherical morphologies lacking cristae curvature [57,58]. Changes in PSD expression influence mitochondrial shape and dynamics [59]. PSD’s cristae localization and upregulated activity are a likely culprit for glaucomatous mitochondrial abnormalities.

To determine PSD’s elevated activity effects on mitochondrial morphology and mitochondrial lipidomes, we first evaluated PSD overexpression (PSDOE) effects in a human BE(2)-M17 neuroblastoma cell line. Neuroblastoma cells were transfected with pReceiver vectors and isolated by puromycin selection. Western blot analysis of puromycin-selected PSDOE neuroblastoma cells revealed increases in PSD β subunit, while the ⍺-β proenzyme levels remained constant (Figure 2A). This increased β subunit is consistent with published overexpressed PSD models as PSD undergoes cleavage after mitochondrial import [54]. PSDOE’s increased activity was validated by using two micrograms of isolated mitochondria from puromycin-selected transfected neuroblastoma cells. Using published methods, exogenous PS was administered to mitochondria, and PS conversion by PSD was permitted to proceed for 30 min at 37 °C [46]. Enzymatic activity was halted, and phospholipids were extracted by butanol–methanol extraction with lithium chloride. After mass spectrometry analysis, normalized PE levels were compared between NEG and PSDOE groups (6 biological replicates), and PSDOE displayed a log₁₀-fold increase in normalized PE concentration (p = 0.0147; Figure 2B). Using organelle markers and immunocytochemistry analysis, we verified appropriate PSD colocalization with mitochondria compared to other organelles (Figure S1A). After these validation experiments, our PSDOE model could be applied to neuroblastoma cells for mitochondrial analysis.

To evaluate mitochondrial networks, transfected neuroblastoma cells were stained with Mitotracker DeepRed while seeded on coverglass-bottom dishes. For analysis, ~50 neuroblastoma cells were imaged per group at 63× magnification, and the processed mitochondrial signal was analyzed using the FIJI plugin, MitochondriaAnalyzer. PSDOE neuroblastoma cells displayed increased mitochondrial fragmentation, evident as punctate mitochondrial staining (Figure 2C). Quantitative analysis showed PSDOE had significantly reduced mitochondrial branch length (p < 0.0001), form factor (p = 0.0003), and branch number per mitochondrion (p = 0.0009; Figure 2D–F), consistent with previous reports [59,60].

Accompanying this fragmentated mitochondrial population, we expected a significant shift in mitochondrial phospholipidomes. Mitochondrial phospholipidomic profiles were generated using isolated mitochondria from transfected neuroblastoma cells. Phospholipid quantities were normalized to mitochondrial protein estimates and spiked-in deuterated lipid standards. These profiles show distinct clustering in principal component analysis and the highest correlation within groups (Figure 3A,B).

As expected, polyunsaturated PE comprised most of mitochondrial PE, but total PE and polyunsaturated PE content were only trending higher in PSDOE but not significantly (Figure 3C). No significant PE lipid species were identified either (Figure 3D). This is consistent with other studies that did not find significant changes in mitochondrial PE levels with PSDOE. These results further validate the idea that mitochondrial PE content is tightly controlled [61]. The increased PE flux from PSDOE is likely beyond what mitochondria can handle leading to a significant PE efflux to other membranes. From there, PE can be incorporated or further metabolized. Future studies should investigate these contextual differences between mitochondrial and plasma membrane lipid pools using fractionated studies.

However, we observed significant changes in other phospholipid groups. PSDOE mitochondria showed increased phosphatidylinositol (PI) and decreased phosphatidylglycerol (PG) (Figure 3E). PG is a cardiolipin precursor, suggesting reduced PG may indicate impaired cardiolipin synthesis. Both cardiolipin and polyunsaturated PE maintain the cristae lipid curvature due to their intrinsic curvature and conical shapes. Multiple cardiolipin species were reduced indicating potential disruption of cristae integrity (Figure 3F).

Overall, PSD overexpression induces mitochondrial fragmentation, alters lipid composition, and disrupts cristae-forming lipids. All are features observed in glaucomatous degeneration indicating increased PSD expression and activity may lead to neurodegenerative states.

3.3. Reduced PE and PSD Expression in Optic Nerve Regeneration

PE is one major glycerophospholipid comprising the axonal membrane. As mentioned, glycerophospholipid synthesis supports axon regeneration, but can have multiple sources. Phosphatidylethanolamine (PE) is synthesized via two primary routes: the CDP-ethanolamine Kennedy pathway in the endoplasmic reticulum and mitochondrial conversion of PS by PSD. Previous studies have confirmed that Kennedy-based PE supports regeneration in regenerative models and in overexpression models, but PSD’s involvement in axonal growth or regeneration has not been studied.

Lipidomic profiles of regenerating optic nerve tissues have been collected and evaluated for glycerophospholipid trends [10,11,12,13,14]. Within the early stages of optic nerve regeneration, there is a consensus amongst regenerative models indicating reduced negative intrinsic curvature lipids. This was attributed to reduced PE and polyunsaturated PE species. The functional reason for this is unclear, but it suggests that PSD and PSD-derived PE are not essential for axonal growth. We further evaluated phospholipid metabolic genes (45 genes) in transcriptomic data sets of regenerating PTENKO RGCs (GSE206626, GSE202155) and found a significantly reduced PSD expression, while no significant change in PSD expression was observed in RGCs after optic nerve injury (Figure S2). This finding suggests that PSD downregulation may be a characteristic of successful axonal regeneration

3.4. PSD Knockdown Increases RGC Neurite Outgrowth In Vitro

To investigate the impact of PSD on axonal growth, we isolated RGCs from postnatal day 3 (p3) C57BL/6 mice and transduced them with AAV vectors expressing ORFStuffer-mCherry (NEG control), PSD overexpression (PSDOE), or PSD-targeting shRNAs (sh1, sh2, sh3). PSD overexpression and knockdown were validated in Neuro2a cells (Figure S3). Knockdown efficiency was determined to be ~50%; however, Western blot confirmation in primary RGCs was not performed, and off-target effects cannot be entirely ruled out. The concordance between lipidomic, mitochondrial, and growth phenotypes supports that these effects are likely PSD-dependent. AAVs were administered 30 min post-seeding, and RGCs were cultured for 5 days in vitro (DIV).

Neurite extension was assessed using β-tubulin III (TUBB3) immunolabeling. Only RGCs bearing neurites were included in the analysis. Representative images and tracings are shown in Figure 4A, highlighting longer neurites in the PSDsh3 group. Sample sizes ranged from 20 to 80 RGCs per group. One-way ANOVA analysis revealed a significant increase in neurite length in PSDsh3-treated RGCs (1470 µm) compared to the NEG control (725 µm, p = 0.0194). PSDsh2 also shows increased outgrowth (1210 µm), while PSDsh1 was modestly reduced (582 µm). The observation that the shRNA producing the strongest neurite outgrowth did not correspond to maximal knockdown suggests that PSD activity may exhibit a non-linear relationship with neurite extension. PSDOE resulted in the shortest neurite extension (435 µm).

Notably, PSDsh3 shifted the entire outgrowth distribution, suggesting a broader effect across RGC subtypes. We further confirmed PSDsh3 increased neurite outgrowth in a repeated, independent RGC experiment. These findings indicate that PSD knockdown (PSDKD) enhances RGC neurite extension, while PSD overexpression may inhibit it.

3.5. Doxorubicin Rescues Neurite Outgrowth Deficits in PSDOE RGCs

Due to neurite outgrowth impairment observed in PSDOE RGCs, we investigated whether pharmacological inhibition of PSD could mitigate this phenotype. Doxorubicin, an anthracycline previously reported to suppress PSD activity [62], was tested for its ability to modulate RGC neurite growth. Dose-response analysis revealed a significant, concentration-dependent inhibition of neurite extension in wild-type RGCs following Doxorubicin treatment at 3 DIV (Figure 5A). Despite the growth-suppressive effects, treatment of PSDOE RGCs with Doxorubicin beginning at 1 DIV led to a significant restoration of neurite outgrowth at 5 DIV compared to untreated PSDOE controls (878 µm, p = 0.0206; Figure 5B). These findings indicate that suppression of excessive PSD activity can rescue neurite extension, supporting a role for PSD as a negative regulator of axon growth.

We next assessed whether supplementation with lysophosphatidylethanolamine (LPE), a PE precursor in the Lands cycle capable of restoring deficient mitochondrial PE [59], could alter neurite effects in PSDKD RGCs. Natively, LPE treatment (10 nM–1 µM) had no significant effect on neurite extension in wild-type RGCs (Figure S4A). In PSDKD conditions, 10 nM LPE was selected for further testing to minimize off-target receptor activation. At 5 DIV, LPE treatment had no significant effect on PSDKD neurite outgrowth, although a trend toward increased neurite length was observed in PSDsh1 (945 µm, p-value 0.0854) (Figure S4B). These results suggest that exogenous LPE has a limited effect on PSDKD RGCs displaying increased neurite outgrowth.

3.6. PSDOE Disrupts Mitochondrial Morphology in RGC Axons

Although PSD overexpression was shown above to fragment mitochondrial networks in neuroblastoma cells, its impact on mitochondrial morphology in primary neurons remains unexplored. Given that mitochondrial dynamics are critical for axonal extension and growth cone motility [21], and that increased mitochondrial fusion supports axonal growth and regeneration [15], we investigated how PSD modulation affects mitochondrial morphology in RGCs.

We used AAV-mediated PSD overexpression and knockdown models and TOMM20 staining to determine the RGC mitochondrial signal. For analysis, we avoided using somatic regions due to their rounded shapes, which limit network visualization and analysis. TOMM20-labeled mitochondria were isolated within manually traced axonal masks for analysis (Figure 6A). Mitochondrial morphology varied across groups. PSDOE RGCs exhibited small, rounded mitochondria, whereas other groups showed more elongated or mixed morphologies. Quantification of total TOMM20 signal area normalized to axonal area revealed no significant differences between groups (Figure 6B). However, further analysis using the ImageJ Mitochondria Analyzer plugin [63], treating each mitochondrion as a discrete data point (n = 100–950 mitochondria from 10–30 RGCs per group), revealed PSDOE significantly reduces mitochondrial length and form factor (Figure 6C,D). In contrast, PSDKD did not significantly alter mitochondrial morphology or form factor. These effects mirrored those observed in neuroblastoma cells. Together, this data indicates PSD overexpression disrupts mitochondrial integrity in RGC axons associated with a reduced neurite outgrowth phenotype. Conversely, PSD knockdown preserves mitochondrial structure and supports axonal elongation.

3.7. RGC Membrane Fluidity Profiling Using C-Laurdan Imaging

During axonal regeneration, RGCs engage lipid synthesis pathways that support membrane expansion. Phospholipidomic analyses have shown a consistent reduction in lipids with negative intrinsic curvature, suggesting biophysical membrane adaptations favoring growth [10]. An indirect evaluation of membrane fluidics by removal of cholesterol or inhibiting cholesterol synthesis indicates increased membrane fluidity supports axon regeneration and branching [23,24,25]. Notably, embryonic dorsal root ganglia (DRG) neurons, during peak neuritogenesis, display higher membrane fluidity than adult DRG, as revealed by fluorescence recovery after photobleaching (FRAP) measurements [64].

Despite these findings, the impact of phospholipid metabolism on RGC membrane fluidity remains uncharacterized. While PSD generates PE, which can integrate into other cellular membranes and influence other lipid metabolism, PSD’s role in modulating membrane fluidity in mammalian neurons has not been studied.

The development of solvatochromic dyes such as C-Laurdan enables measurement of membrane fluidity in biological systems [65]. C-Laurdan is compatible with conventional confocal microscopy, excited at 405 nm and emits in two spectral ranges (415–455 nm: order and 490–530 nm: disorder) depending on lipid packing order [66]. Ratiometric analysis of these emissions generates generalized polarization (GP) values, which allow visualization of ordered (low fluidity, high GP) and disordered (high fluidity, low GP) membrane regions [42,67]. This method has been used to examine the effects of cholesterol depletion [42], exogenous phospholipids [10], and polyunsaturated fatty acids on membrane properties [28].

To our knowledge, C-Laurdan has not been applied to RGCs for membrane fluidity analysis. To establish a baseline, we isolated p3 RGCs and stained them with C-Laurdan at 5 DIV. Confocal imaging revealed that the soma exhibited higher GP values (i.e., lower fluidity), whereas neurites showed a broad distribution of GP values averaging at lower GP values, suggesting heterogeneous membrane dynamics (Figure 7A). No spatial pattern of fluidity along the neurites was observed; however, some branches exhibited uniformly higher fluidity than others.

GP values from manually segmented full-cell masks were plotted and fit to a Gaussian distribution to visualize variability (Figure 7B). Soma compartments had a narrow distribution centered at higher GP values, while neurites displayed a broader range (GP −0.5 to 0.6) and a lower mean GP, indicating higher fluidity. Additional z-step ratiometric mapping confirmed these trends (Figure S5). These results demonstrate that native RGCs possess a soma of lower membrane fluidity compared to neurites, exhibiting dynamic and spatially varied fluidic profiles.

3.8. PSD Knockdown Increases RGC Somal Membrane Fluidity

To investigate how PSD expression affects membrane fluidity, we analyzed AAV-transduced RGCs expressing mCherry and stained with C-Laurdan. Only RGCs with confirmed neurite outgrowth and sufficient C-Laurdan signal were included, resulting in a sample size of 7–28 RGCs per group. Total cell masks were created using the higher-intensity disorder channel, and averaged GP values were compared to the NEG control using one-way ANOVA analysis.

Both PSDsh1 and PSDsh3 groups exhibited significantly lower total cell GP values compared to the NEG control, indicating increased membrane fluidity (Figure S6A). GP distribution plots supported this finding, showing a clear leftward shift for these groups (Figure S6B). To determine which cellular compartment contributed to these changes, total masks were subdivided into soma and neurite regions. No significant differences in average GP values were observed within the neurite compartments across experimental groups (Figure S6C,D). Neurite GP distributions were broad, consistent with native RGCs, and only showed minor shifts.

Figure 7C highlights the somal compartment GP values. Visually, PSDsh1 and PSDsh3 cells displayed lower GP values localized along the soma periphery. Quantitative analysis confirmed this observation, where GP distributions showed a pronounced leftward shift (Figure 7D), and mean somal GP values were significantly reduced in both PSDsh1 and PSDsh3 groups compared to controls (Figure 7E). As expected, somal GP values remained higher on average than neurite GP values. These results demonstrate that PSDKD significantly increases somal membrane fluidity.

Further, 3D color mapping of GP values revealed that the fluidity increase in PSDsh3 is present along the somal membrane (Figure 7F). Collectively, these findings indicate that reduced PSD expression selectively increases membrane fluidity in the soma, potentially altering lipid dynamics relevant to axonal growth.

3.9. PSD Knockdown Alters Lipidomic Profiles Linked to Membrane Fluidity

To investigate whether the observed changes in membrane fluidity were associated with underlying lipidomic shifts, we conducted untargeted lipidomic profiling on 15 k RGCs per group using LC-MS. Lipids were extracted using a butanol–methanol and lithium chloride protocol, and quantification was performed by normalization using internal spike-in deuterated standards [47,68,69].

PCA analysis of RGC lipidomes depicted overlapping distributions among groups, but NEG and PSDsh3 were the most distinctly separated (Figure 8A). Across the dataset, nine lipid classes were identified. Total class abundances were compared to the NEG control using one-way ANOVA (Figure 8B). In the PSDsh3 group, three lipid classes exhibited significant changes: cholesterol and PC levels were reduced, while sphingomyelin was increased. These lipid classes are notable components of lipid raft microdomains, which are critical to membrane organization and dynamics.

Interestingly, while all PSD-modified groups (PSDOE, PSDsh1, PSDsh2, PSDsh3) exhibited reduced cholesterol, the most pronounced reductions were observed in PSDsh1 and PSDsh3, which parallels the increased membrane fluidity detected in C-Laurdan GP measurements. At the lipid species level, PSDsh3 significantly altered triacylglycerol (TAG) profiles, particularly decreased species enriched in saturated acyl chains (Figure 8C,D). In contrast, PSDOE significantly increased polyunsaturated TAGs. These changes suggest a broader impact of PSDOE and PSDKD on lipid remodeling beyond membrane components.

Together, these results show that PSD knockdown leads to reduced cholesterol and phosphatidylcholine levels and modifies TAG composition. The combination of lipidomic shifts likely contribute to the increased membrane fluidity observed in PSDsh1 and PSDsh3 RGCs and may underlie the enhanced neurite outgrowth phenotype.

3.10. PSD Knockdown Enhances In Vivo RGC Regeneration Following Optic Nerve Injury

To evaluate the regenerative potential of PSD knockdown in vivo, we utilized AAV2-based delivery of shRNAs targeting PSD in 2-month-old C57BL/6 mice (n = 4 per group, equal sexes), followed by a standardized optic nerve crush (ONC) surgery to induce degeneration. Each PSDshRNA vector used a CMV-mCherry-U6-shRNA backbone, while a CMV-mCherry-U6-scramble vector served as the control (Scramble). We present here the scramble as the control as the primary focus is on PSDKD supporting optic nerve regeneration, but ultimately both NEG and Scramble had the same outcomes. The experimental timeline is outlined in Figure S7. AAVs were administered via intravitreal injection 14 days prior to ONC to ensure robust transgene expression. RGC transduction was verified using mCherry colocalization with β-Tubulin III (TUBB3) and Brn3a (Figure S8A–C).

In the two weeks preceding ONC, intraocular pressure (IOP) and pattern electroretinography (PERG) were monitored to assess baseline ocular health and RGC function. IOP values remained below 20 mmHg across all groups, and no significant differences were detected between groups by two-way ANOVA (Figure S9A). PERG amplitudes were also consistent across all time points and treatment groups, confirming stable RGC function prior to injury. After ONC surgery, loss of PERG amplitudes and waveform P100 peak was similar across experimental groups indicating successful ONC surgery (Figure S9B–F).

Mice were maintained for 14 days after ONC to allow for regenerative growth. Forty-eight hours before euthanasia, Alexa 488-conjugated cholera toxin subunit B (CTB488) was intravitreally injected to anterogradely label regenerating RGC axons. Regenerating axons were quantified by colocalized CTB488 and mCherry axon signals extending beyond the crush site at 100 µm intervals and normalized to the optic nerve width. Notably, all CTB⁺mCherry⁺ axons extended within 300 µm past the lesion (Figure 9A). Axon counts were averaged from three longitudinal 10 µm sections per optic nerve.

The Scramble control exhibited a baseline average of 5.4 CTB⁺mCherry⁺ axons/mm per section. Among the experimental groups, PSDsh3 demonstrated a significant increase in growing axons at 100 μm from the injury site with 19.4 CTB⁺mCherry⁺ axons/mm (p = <0.0001, one-way ANOVA; Figure 9B). PSDsh3 shows an average of 4.087 CTB⁺mCherry⁺ axons/mm at 200 μm (p = 0.22). The progressive decline in axon density by distance and the nonlinear trajectory of the axons are signs of growth rather than survival or spared axons [70]. No significant differences were observed for PSDsh1, PSDsh2, or PSDOE, and a full panel is shown in Figure S10. Although axonal growth was limited to the proximal 100 μm of the distal nerves, this region exhibited a significant, injury-dependent increase in regenerated axons, indicating PSDsh3 enhanced regenerative initiation in vivo.

4. Discussion

Lipid metabolism has a balancing role in neurobiology that, if not maintained, can lead to neurodegeneration, but is also a suitable target for improving regenerative growth. In this study, we identify phosphatidylserine decarboxylase (PSD), a mitochondrial lipid metabolic enzyme, as a negative regulator of RGC axonal growth and regenerative competence. Using in vitro and in vivo models, our findings support a model in which PSD-dependent lipid remodeling alters mitochondrial morphology and membrane characteristics that collectively influence neuronal growth capacity.

PSD is the second largest contributor of PE, producing PE from PS within the inner mitochondrial membrane, whereas the largest PE contributor is from the endoplasmic reticulum Kennedy de novo synthesis pathway. The precursor for the PSD pathway, PS, is synthesized through base exchange reactions by PSS1/2 at MAM sites, where it is transported to mitochondria [71]. Importantly, PSD preferentially uses polyunsaturated PS species, producing PE species that have stronger effects on lipid membrane packing and curvature. Mitochondrial PE is not confined to mitochondria and can be redistributed to other cellular membranes. For that reason, PSD activity has the potential to influence mitochondrial lipid composition and the broader cellular lipid membranes.

In human glaucomatous optic neuropathy, PSD expression and activity are elevated leading to altered optic nerve lipidomes, specifically accumulated PE and reduced PC and PS. In our neuroblastoma and RGC models, increased PSD activity leads to mitochondrial fragmentation and spherical morphologies typically indicative of mitochondrial stress. These morphologies are analogous to the mitochondria seen in glaucoma models. PSDOE’s abnormal mitochondrial morphologies were associated with shifted phospholipidomes, particularly with increased PI and reduced PG and Cardiolipin species, suggesting disrupted inter-organelle membrane contacts and cristae disorganization, respectively [72]. These observations indicate that mitochondrial structure is sensitive to PSD-driven lipid flux from excessive PSD activity.

Conversely, downregulated PSD expression promoted neurite outgrowth and regenerative competency. In regenerative transcriptomic profiles, PSD expression was downregulated in the robust PTENKO regenerative mouse model, while injury alone does not alter PSD expression. PTENKO and other regenerative models stimulate the mTOR pathway, increasing growth-related pathways. Conversely, mTOR inhibitor efficacy was found to require PSD activity and mitochondrial PE [73]. We further confirm the transcriptomic findings by demonstrating that partial PSD knockdown enhanced RGC neurite extension in vitro and in vivo. In our in vitro study, we evaluate three PSDshRNA constructs ranging in PSD knockdown from 40–60% based on Neuro2a results, where PSDsh3 had the highest performance with a ~50% knockdown. The variability of shRNA neurite outgrowth has two potential indications. First, there is an ideal PSD knockdown window for increased growth competency. Second, the PSDsh3 binding region for the PSD transcript is optimal for performance in retinal ganglion cells. In contrast, PSD overexpression had the lowest neurite outgrowth, and this was rescued by Doxorubicin inhibition of PSDOE activity.

In vivo, PSDKD by PSDsh3 increased the proportion of regenerating axons within the proximal region of the injured optic nerve, indicating enhanced initiation of regeneration rather than sustained long-distance growth. This finding is consistent with the notion that modulating lipid metabolic pathways is insufficient to activate intrinsic growth pathways necessary for extensive regeneration. PSD expression modulation can create higher regenerative competent RGCs and potentially serve as a combinatorial strategy with other regenerative models. Based on PSDshRNA knockdown efficiencies, there is an ideal PSD knockdown that supports a growth-permissive window for RGCs in vivo as well. We note that some of our experiments were conducted with limited sample sizes due to the scarcity of human donor tissue and the constraints of low-density RGC cultures. Despite these limitations, the PSDsh3 knockdown phenotype was consistently observed across repeated in vitro studies and in vivo regeneration, supporting the robustness of these findings. Future studies with larger sample sizes will be important to confirm and validate these observations.

While partial PSD knockdown enhances regenerative potential without obvious mitochondrial stress in our models, PSD is an essential mitochondrial enzyme, and complete loss is embryonic lethal. Careful titration of PSD expression will be necessary for the therapeutic strategy. Potential safety considerations include off-target effects, long-term mitochondrial dysfunction, and immune responses associated with PSD modulation and AAV-mediated gene delivery [74]. Preclinical studies would need to address these concerns to properly consider PSDKD’s translatability.

As discussed, PSD’s PE production first directly feeds into the mitochondrial lipid membrane. Mitochondrial morphology is highly sensitive to these lipid metabolic changes and indicates mitochondrial health. Both overexpression and severe defect of PSD expression can induce mitochondrial fragmentation and dysfunction [33,75]. Due to the knockdown range of our PSDshRNA constructs, we did not see any significant alterations to RGC axonal mitochondrial content or morphology as they presented as normal, elongated structures. The mitochondrial morphological differences we observe between our PSDKD and PSDOE models reflect the differences in the growth-permissive states created from PSD manipulation.

As with many lipid metabolic enzymes, overexpression will likely introduce non-native effects, such as altered reactant stoichiometry, excessive metabolic flux, or membrane stress. Accordingly, pronounced mitochondrial fragmentation and growth suppression observed with PSD expression may reflect a supraphysiological state. In contrast, partial PSD knockdown more closely resembles endogenous modulation and provides more information on PSD’s role in creating a growth-permissive state.

It is well established that lipids directly affect membrane fluidics due to intrinsic characteristics and their influence on other lipid classes. The membrane fluidity state is the combined effect of multiple properties: lipid class composition, acyl-chain saturation, and membrane organization. Higher membrane fluidity is seen in postnatal neuronal states when neuritogenesis is more active. Our native RGC C-Laurdan analysis demonstrates compartmental differences in membrane fluidity between the soma and neurites. RGC neurites possess higher and more dynamic membrane fluidity measurements compared to somas. This is consistent with the biophysical demands of neurite and axon growth with flexibility and fast adaptation to extracellular cues.

From C-Laurdan analysis, PSD knockdown displayed a selective increase in somal membrane fluidity, which aligns with previous observations with PSD deletion in Candida albicans [76]. Further lipidomic analysis indicated PSDsh3 had reduced cholesterol and PC levels, which is a mechanistic explanation for the measured fluidic shifts. Previously, the relationship between polyunsaturated PE and cholesterol quantities and the formation of cholesterol rafts has been described [77,78]. This is a potential downstream effect of PSD knockdown, where polyunsaturated PE levels are decreased leading to reduced formation of cholesterol lipid domains and thus higher somal fluidity. It bears noting that cholesterol lipid raft disruption increased axonal growth in previous studies [23]. The observation that cholesterol is reduced in both PSDOE and PSDKD, yet increased somal membrane fluidity occurs only in PSDKD. It is likely that a broader lipid remodeling is responsible for fluidic shifts. A broader lipid profile should be investigated with a larger scale RGC study as lipid signal here was limited by 15k RGCs per replicate. Our findings highlight that a combination of these effects and cellular context are necessary to drive axonal growth.

To determine whether PE availability alone can alter neurite outgrowth, we supplemented PSDKD RGCs with exogenous LPE. However, we found no significant effect. A possible explanation is that PSDKD does not require the PE replenishment seen in severe PSD deficits to properly observe outgrowth behavior changes. Additionally, LPE metabolism does not follow the same route as PSD-derived PE metabolism, nor does it represent polyunsaturated PE. In other words, LPE supplementation is not a complete mimic of PSD-derived PE, which is a possible reason for the minimal effect when combined with these investigated PSDshRNAs.

Our study further links PSD expression to TAG metabolism in RGCs. TAG metabolism and remodeling can represent differences in lipid buffering and metabolic stress. In PSD knockdown, reduced saturated TAG species may indicate increased utilization of stored fatty acids to support membrane expansion during neurite outgrowth. This indicates saturated fatty acids may selectively be mobilized during phospholipid synthesis for neurite extension within the PSDKD growth model. Alternatively, PSDOE resulted in increased polyunsaturated TAG species. Polyunsaturated fatty acids are highly dynamic but are oxidation-prone lipids that require tight regulation to avoid death [79]. Their sequestration into TAGs can be a protective mechanism to buffer excess polyunsaturated acyl chains and limit lipotoxicity [80]. Given PSD’s preference for polyunsaturated PS substrates, increased PSD activity likely elevates polyunsaturated PE production beyond normal membrane demand. This shifts excess polyunsaturated fatty acids into TAG pools. The lipid redistribution response coincides with mitochondrial fragmentation and reduced neurite outgrowth observed in PSDOE conditions, suggesting that excessive PSD-derived PE flux disrupts lipid homeostasis and does not support membrane expansion [81,82]. Together, these findings indicate that PSD activity balances fatty acid utilization and changes in expression can lead to compensatory TAG remodeling.

Our data supports a model that PSD acts as a regulator of RGC growth competency by coordinating mitochondrial morphology, membrane fluidity, and TAG metabolism. Reduced PSD activity creates a permissive state characterized by intact mitochondria, increased somal fluidity, reduced cholesterol, and diminished saturated TAG accumulation. Characteristics that collectively support neurite outgrowth and regenerative capacity. These findings do not place any single downstream effect as the driver for neurite outgrowth, but PSD can alter regenerative growth potential. Further investigative studies where PSDKD is combined with other growth models to determine long-distance and functional growth would be necessary to determine its full therapeutic potential.

5. Conclusions

Lipid metabolism continues to be an essential facet of neurological health and is now a promising target for improving axonal growth for regenerative therapeutics. Lipids are sourced from different organelles, each containing enzymes that have the potential to synthesize unique lipid species. Here, we investigated a mitochondrial lipid metabolic enzyme, phosphatidylserine decarboxylase (PSD), that converts phosphatidylserine (PS) to phosphatidylethanolamine (PE), with a preference for polyunsaturated species. Increased PSD expression associates with glaucomatous neurodegeneration and supports a nongrowth-permissive state with fragmented mitochondria and increased polyunsaturated TAG species. PSD Knockdown enhances RGC growth competency both in vitro and in vivo while preserving normal mitochondrial morphology.

Beyond mitochondrial effects, reducing PSD expression broadly affects RGC lipid organization, including changes to somal membrane fluidity, cholesterol content, and TAG metabolism. While the direct mechanistic relationships linking PSD activity to these downstream features remain to be elucidated as well as the ultimate drivers of growth, our findings support a model where PSD activity restricts neuronal growth capacity, while reduced PSD creates a growth-permissive environment. These results suggest that targeting PSD may enhance axonal growth competency in neurodegeneration, such as glaucoma, and highlight the potential to combine PSD modulation with regenerative models to further improve outgrowth.