Metabolomic changes in the mouse retina after optic nerve injury

In glaucoma, although axonal injury drives retinal ganglion cell (RGC) death, little is known about the underlying pathomechanisms. To provide new mechanistic insights and identify new biomarkers, we combined latest non-targeting metabolomics analyses to profile altered metabolites in the mouse whole retina 2, 4, and 7 days after optic nerve crush (NC). Ultra-high-performance liquid chromatography quadrupole time-of-flight mass spectrometry and liquid chromatography Fourier transform mass spectrometry covering wide spectrum of metabolites in combination highlighted 30 metabolites that changed its concentration after NC. The analysis displayed similar changes for purine nucleotide and glutathione as reported previously in another animal model of axonal injury and detected multiple metabolites that increased after the injury. After studying the specificity of the identified metabolites to RGCs in histological sections using imaging mass spectrometry, two metabolites, i.e., L-acetylcarnitine and phosphatidylcholine were increased not only preceding the peak of RGC death in the whole retina but also at the RGC layer (2.3-fold and 1.2-fold, respectively). These phospholipids propose novel mechanisms of RGC death and may serve as early biomarkers of axonal injury. The combinatory metabolomics analyses promise to illuminate pathomechanisms, reveal biomarkers, and allow the discovery of new therapeutic targets of glaucoma.


Principal component analysis and discriminant analysis. After t R alignment and deconvolution
with Progenesis QI, the HILIC column in positive ion (HILICpos) mode revealed 5,902 features, the negative ion (HILICneg) mode revealed 1,833 features, the C18pos mode revealed 7,675 features, and the C18neg mode revealed 5,398 features. We then applied a normalization procedure for the abundance of retinal metabolites. After normalization, there were 1,578 HILICpos assay-detected features, 272 HILICneg-detected features, 1,194 C18pos-detected features, and 1,645 C18neg-detected features. All these features passed our selection criteria for a multivariate analysis of the retinal samples, which we obtained from control (Ctrl) animals and on days 2, 4, and 7 after NC. Figure 2a,b shows a typical principal component analysis (PCA) score plot and an orthogonal partial least square-discriminant analysis (OPLS-DA) score plot from the C18pos analysis for all four groups of mice. These results clearly show large variations between the retinas from the Ctrl and NC groups. Next, based on our OPLS-DA findings, we selected features that contributed to the differences between the Ctrl and NC day 2 groups. We defined large changes in abundance based on a correlation value (p(corr) [1]P) greater than 0.60 (up to 1.0) and less than −0.55 (as low as −1.0), as derived from S-plot analyses in the four assays (Fig. 2c,d). All selected features in the four assays are listed in Supplementary Table 1.
Metabolomic Analysis. Eighty-three metabolites in the mouse retinas had significantly different concentrations after NC when compared with the non-treated group, (P < 0.01) and 30 of these 83 metabolites were identifiable by UHPLC-QTOF/MS and LC-FTMS (Table 1). To analyze these metabolites in further detail, we sorted them with hierarchical clustering and classified them into 4 subgroups (Fig. 3).
Group A contained 9 metabolites that decreased on days 2 and 4 after NC, but increased again on day 7. Group B contained 6 metabolites that maintained a consistently reduced level after NC. Group C contained 8 metabolites that transiently increased on day 2 after NC and then decreased slightly on days 4 and 7. Group D contained 7 metabolites that increased gradually and peaked on day 4 after NC. The metabolites that were identified are listed in Table 1.
Imaging MS Analysis. Next Fig. 4a-c). These changes were consistent with the results of our MS spectrometric analysis (Fig. 3). IMS was also able to detect other metabolites, including adenine, glycerophosphocholine, adenosine, lysophosphatidylcholine (LPC) (18:2) and phosphatidylglycerol (32:0), but could not detect the precise location of changes in these metabolites in the GCL (Supplementary Fig. 1).

Discussion
This study used G-Met to identify new phospholipids in the retina after NC, in addition to sugar, amino acids and purine nucleobases. We categorized these metabolites into four groups with a hierarchical analysis: among the metabolites in these groups, purine metabolites, D-glucose and glutathione metabolites decreased in the early stages of RGC damage (before significant RGC loss), and then increased in the late stages. Glycerophosphatidylcholine and LacCer decreased continuously after NC. Most elevated metabolites in the early stages of RGC damage were phosphatidylglycerol (PG) metabolites, and carnitine metabolites in the late stages. Moreover, some phospholipids with altered levels in the retina after NC were observed to be specifically elevated in the GCL. These findings demonstrate that our new method, which combines LC-MS and IMS, can identify new, promising biomarkers and should help in the understanding of the pathophysiological mechanisms associated with RGC-specific neuronal degeneration. Although dysfunctional RGCs can be observed in the retina in the early stages of glaucoma 14 , these cells represent only a small proportion of the total cells in the retina. Therefore, analyses based on the entire mass of the retina are less likely to accurately detect the distribution of metabolites specific to the RGCs, or to provide sufficient sensitivity and spatial resolution to adequately reveal pathophysiological events in the RGCs. A promising alternative is matrix assisted laser desorption/ionization imaging MS (MALDI IMS) technology, which has high enough spatial resolution to allow the distribution analysis of metabolites 15 . Methods for distribution analysis focused on the central metabolites and on lipids in the retina have been established based on MALDI IMS [16][17][18][19] . Furthermore, the sensitivity and spatial resolution of focused lipid localization analysis have been found to be high 20 . Therefore, the combination of LC-MS and MALDI IMS technology for G-Met promises to reveal new prognostic biomarkers of glaucoma, and to enable new ways of identifying the disease. Additionally, this method may clarify the underlying pathological mechanisms, and should help to find new ways to prevent disease progression at an early stage.
In this study, metabolites in group A decreased 2 days and 4 days after NC, but increased on day 7, in comparison with the non-treated retinas. Group A included guanosine and adenosine, which are classified as purine nucleosides and are well known to modulate the homeostatic function of Muller cells, astrocytes and microglia [21][22][23] . In addition, purinergic signaling protects against cell death in injured neurons. For example, guanosine has a neuroprotective effect in the hippocampus, reducing oxidative stress and inhibiting the PI3K/Akt/GSK3b pathway-mediated inflammatory response and iNOS [24][25][26][27] . Adenosine inhibits P2X7 receptor-induced calcium influx and protects the RGCs 28,29 . Nicotinamide adenine dinucleotide (NAD), one of the products of adenine, prevents axonal loss and protects against cell body degeneration in the RGCs 30,31 , suggesting that adenine might contribute to the maintenance of the RGCs.       The metabolic cycle of glutathione, which is regulated by glutathione reductase (GSR) and glutathione peroxidase (Gpx1), is well known to include a redox function and to defend cells from oxidative-stress injury 32 . Previous work has shown that the glutathione level in blood samples taken from glaucoma patients is lower than in age-matched controls 33 . Our data also showed that the glutathione level in the retina on day 2 after optic nerve crush was less than in the uninjured retinas. These findings suggest that lower levels of glutathione in glaucoma patients might be associated with optic nerve damage, and that they may cause progressive RGC loss via insufficient redox. The reduction of glutathione after NC may also be due to defective enzymatic regulation.
Interestingly, the metabolites in group A that had decreased in the earlier stages increased on day 7 after NC, reaching levels that were higher than the non-treated baseline. It is known that nucleotides and nucleosides released from injured neurons activate the microglia. Furthermore, adenosine-and ATP-induced microglial activation promotes the recruitment of microglia around the site of neuronal damage, acting via the A3 and P2Y12 receptors 34 . Previous research has also revealed that on day 7 after NC, most RGCs disappear from the retina and macrophages and microglia are recruited 35 . This suggests that in group A, the recruitment of microglia was regulated by adenosine and ATP released from injured RGCs.
Metabolites in group B maintained a consistently reduced level after NC and were mostly classified as chorines. Phosphatidylcholines are one of the components of the plasma membrane and are catalyzed by phospholipase D (PLD) 36 . In damaged RGCs in mice after NC, the plasma membrane may be disrupted by PLD activation. Chorine has crucial roles in maintaining plasma membranes, in the synthesis of neurotransmitters (in which it acts as a substrate) and in supplying methyl groups with material. LacCer (d18:1/22:0), categorized as a sphingolipid, is also a component of plasma membranes. Citicoline, an exogenous CDP-choline, has been shown to improve retinal function in glaucoma patients and to prevent apoptotic cell death in the RGCs [37][38][39][40] . Previous studies also suggest that the supply of choline has a crucial role in maintaining the RGCs and visual function, and that low levels of cholines after optic nerve injury may contribute to glaucoma progression.
Metabolites categorized in group C increased gradually and peaked on day 2 after NC, suggesting that metabolites in group C have potential as predictive biomarkers of RGC damage during the early stages. Interestingly, retinal metabolites in group C included elevated lysophosphatidylcholine (LPC) levels. Previous work has demonstrated that exposure to LPC induces demyelination and inflammation of the optic nerve in a model of optic neuritis 41 . LPC also induces the recruitment of macrophages and modulates neutrophil oxidant production 42 . Here, we also found that phosphatidylcholine (PC) (38:7) was elevated in the GCL after NC in mice. The pathophysiological effects of elevated LPC might contribute to RGC degeneration after NC via demyelination, inflammation and oxidant production. In addition, LPC contributes to age-related macular degeneration (AMD), suggesting that LPC might be involved not only in pathological neurodegeneration, such as occurs during glaucoma progression, but also in neovascularization in other retinal diseases.
Another interesting finding of this study was that PG (16:0/16:0) and PG (18:0/22:6), both included in group C, increased after NC. PG is converted from lysophosphatidylglycerol by lysophospholipid acyltransferase (Lpt1p) 43 . Elevated PG in the mouse retina after NC may depend on the activation of Lpt1p and promote the production of cardiolipin, which is a catalysis product of PG. Cardiolipin, a subspecies of PG, is a component of the mitochondrial inner membrane. Cardiolipin is needed for the translocation of caspase-8 to the mitochondria in apoptotic cells 44 . Cardiolipin is also required for the functioning of inflammasome, and directly binds with Nlrp3 inflammasome 45 . In injured cells, PG is released from the mitochondria and induces the activation of apoptosis via cytochrome C, caspase 8 and inflammasome signaling. Autoantibodies against PG have been found in glaucoma patients 46 . This suggests that axonal damage induces excess levels of PG in the RGCs, and that it might be involved in the loss of RGCs in glaucoma via apoptosis signaling, inflammation and autoantibody antigens.
S-adenosylmethionine (SAM), a metabolite of polyamine and an important methyl donor in cells, also increased after NC in the mice in this study, and was included in group C. SAM is synthesized in a reaction catalyzed by methionine adenosyltransferases (MATs) 47 and catabolized with glycine N-methyltransferase (GNMT) 48 . Previous studies have reported that SAM has a neuroprotective effect against L-dopa toxicity in vitro and ischemic damage in vivo [49][50][51] . SAM also functions to reduce inflammation via the suppression of oxidative stress in a mouse model of chronic asthma 52 , and is well known as a substrate of spermidine and spermine. Spermidine treatment promotes autophagy, suppressing oxidative stress and necrosis 53 . In the retina, SAM restores photoreceptor function 54 and protects the RGCs in a mouse model of glaucoma 35,55 . SAM not only has a neuroprotective effect, but can also promote optic nerve regeneration 35 . These previous findings suggest that the elevation in SAM we observed after optic nerve crush in mice might act to promote RGC survival and regenerate the optic nerve. In the current study, glutathione decreased 2 days and 4 days after NC. SAM plays an important role in the synthesis of glutathione by providing homosysteine. Glutathione reduction and SAM elevation after NC may be due to the inhibition of some enzymatic activity related to SAM and the glutathione pathway, including the activity of enzymes including GNMT, S-adenosylhomocysteine hydrolase (SAHH) and cystathionine beta-synthase (CBS). Additionally, SAM is a methyl donor and converts phosphatidylethanolamine (PE) to phosphatidylcholine (PC) 56 . In the current study, PE (42:8) increased and PC decreased slightly after NC, suggesting that the conversion of SAM to S-adenosylhomocysteine (SAH) by PE N-methyltransferase (PEMT) and GNMT may be suppressed after NC. PE (42:8), was also included in group C in this study. It is associated with Raf-1 kinase inhibitory protein (RKIP), which functions to promote RGC survival and axonal regeneration after NC in mice 57 . It is possible that elevated PE (42:8) is also part of a protective response in injured RGCs after NC.
Metabolites categorized in group D increased gradually and peaked on day 4 after NC, suggesting that metabolites in group D have potential as predictive biomarkers of RGC damage during the advanced stages. All identified metabolites in group D were categorized as carnitines. Additionally, L-acetylcarnitine increased in the GCL after NC. L-acetiylcarnitine is synthesized by carnitine acetyltransferase (CAT) from acetyl-CoA and carnitine 58,59 . The elevation of L-acetylcarnitine after NC may be due to the increase or activity of CAT in the mouse retina. Acetylcarnitine can improve mitochondrial function via attenuation of mitochondrial protein SCIENTIfIC REPORtS | (2018) 8:11930 | DOI:10.1038/s41598-018-30464-z acetylation. 60 Carnitine treatment has been shown to prevent the loss of RGCs in the retinas of mice with induced high IOP, acting as an antioxidant. It is possible that this is because L-acetylcarnitine is the main component of the inner membrane of mitochondria 61 . Thus, elevated carnitine may be a biomarker of progressive damage in the RGCs, and elevated carnitine may play a neuroprotective role in maintaining mitochondrial function during RGC degeneration after NC in mice. Previously, we found that SYTOX orange-positive dead RGCs were mainly detectible on day 4 after NC 62 . The current study confirms that L-acetylcarnitine was also highly elevated 4 days after NC in the GCL. This suggests that elevated L-acetylcarnitine is associated with RGC death, and that it might be a candidate biomarker of RGC death. However, the GCL contains not only RGCs, but also other cells, such as displaced amacrine cells. Therefore, amacrine cells may also have increased L-acetylcarnitine and PC (38:7) in the GCL after optic nerve crush. However, it is hard to determine which cells upregulate these metabolites in the GCL. This was therefore a limitation of the present study that will require further consideration in a future study.
In conclusion, our MS-based approach identified novel metabolites that changed in the mouse retina after NC. Categorizing the identified metabolites into subgroups provided insights that may lead to the discovery of previously unknown pathomechanisms and offer a better understanding of already identified mechanisms of RGC degeneration after NC. In particular, IMS technology allowed the identification of new potential biomarkers of RGC damage: PC (38:7) in the early stages and L-acetylcarnitine in the advanced stages. These metabolomic alterations, including alterations in phospholipids, may help the clinical diagnosis of glaucoma and lead to the discovery of new candidate molecules for therapies targeting RGC loss caused by axonal injury.

Material and Methods
Animals. Eight-to twelve-week-old male C57BL/6 J mice were obtained from Clea (Tokyo, Japan) and main-

Induction of axonal injury in mice.
For anesthesia, a mixture of ketamine (180 mg/kg) and xylazine (90 mg/kg) was used intramuscularly and NC was performed to induce damage to the RGCs, as previously described 4,63 . In brief, the optic nerve was exposed and crushed approximately 2 mm posterior to the eyeball with forceps for 5 s. After surgery, an ointment containing levofloxacin (Santen Pharmaceutical Co., Ltd., Osaka, Japan) was applied and the animals were kept on a heat pad. In all experiments, only the right eye was used.
Immunohistochemistry. Staining with anti-RBPMS antibodies was performed after the mouse retinas were fixed with 4% paraformaldehyde. Cryosections were then stained as previously described 64 . Briefly, the cryosections were incubated with rabbit anti-RBPMS (Abcam, #194213; dilution 1:200) and then incubated with goat anti-rabbit IgG Alexa Fluo 488. Nuclear staining was performed with Vectashield, including DAPI (Vector). Immunofluorescence images were captured through a microscope (Axiovert 200; Carl Zeiss, Berkochen, Germany) and immunofluorescence images of the entire retina were obtained and quantified with a fluorescence microscope (BZ-9000; Keyence, Osaka, Japan).
Western blotting. The mouse retinas were homogenized, the extracted protein concentration was calculated, the retinal proteins were separated with SDS-PAGE, and the retinal proteins were transferred to a PVDF membrane, according to a method previously described 65 . The membrane was blocked with 1% skim milk in Tw-PBS for 1 h at room temperature and then incubated with rabbit anti-RBPMS antibody (Abcam; dilution 1:1000) overnight at 4 °C. After washing the membranes with Tw-PBS, they were incubated with HRP-conjugated donkey anti-rabbit IgG (Sigma; dilution 1:5000) at room temperature for 1 h. The immunoreactive signal was developed with ECL prime reagent (GE Healthcare, Piscataway, NJ) and was captured with ChemiDoc (Bio-Rad). The membranes were then reblotted with Restore Western Blot Stripping Buffer (Thermo Scientific) and incubated with mouse anti-beta-actin antibody (Sigma; dilution 1:5000) as an internal control.

Sample preparation for LC-MS analysis.
For the G-met analysis, 8 mouse retinas were used per condition for the controls and for NC day 2, and 4 mouse retinas were used per condition for NC day 4 and day 7. The obtained retinal samples were placed into sample tubes (2.0 mL). Two hundred µL of methanol containing 0.1% formic acid was added to the frozen sample. The mixture was homogenized using a lysis and homogenization system (Precellys) (5,000 × rpm, 15 s, 2 Zr bead). After homogenization in an ultrasonic bath for 10 min, the samples were then centrifuged at 16,400 g for 20 min at 4 °C, and the supernatant was passed through a 96-well plate for deproteinization (Sirocco, Waters Corp.) and then washed 3 times with 100 µL of methanol containing 0.1% formic acid. Thirty µL of each sample was collected from the 96-well plate and mixed in a 15-mL tube. Then, the mixture was transferred into a well as a study quality control (SQC). From SQC, a series of dilution quality controls (dQC) were prepared by dilution with 50% methanol containing 0.1% formic acid as follows: dilution at 2-fold (d2QC), 4-fold (d4QC), 8-fold (d8QC), and 16-fold (d16QC). Finally, the dilution plate was replicated; the SCIENTIfIC REPORtS | (2018) 8:11930 | DOI:10.1038/s41598-018-30464-z original plate was used for UHPLC-QTOF/MS analysis, and the replicate plate was used for LC-FTMS analysis, respectively, with 4 µL and 3 µL sample sizes.
Quality Control Sequences. The required frequency of SQC injections was determined with reference to previous reports 12 . The study-samples were injected in randomized run orders, and the SQC, which was mixed in all study-samples, was injected after every eight study-samples (2 h). In addition, 10 consecutive injections of SQC were made at the start of the chromatographic run to initialize the column. Finally, diluted (x times) QCs (dxQC) were injected three times at each concentration in the following order: d16QC, d8QC, d4QC, d2QC and SQC at the end of the sequence.

UHPLC-QTOF/MS and LC-FTMS Methods. The UHPLC-QTOF/MS analysis was performed on an
Acquity Ultra Performance LC I-class system, equipped with a binary solvent manager, a sample manager, and a column heater (Waters Corp.). This system interfaced with a Waters Synapt G2-Si QTOF MS with electrospray ionization (ESI) system, operated in both positive and negative ion modes. LC separation was performed using a C18 column (Acquity HSS T3; 150 mm × 2.1 mm i.d., 1.8 µm particle size; Waters) with a gradient elution of solvent A (water containing 0.01% formic acid) and solvent B (acetonitrile containing 0.01% formic acid) at 400 µL min −1 . The data were collected using MassLynx, v4.1 software (Waters Corp., Manchester, UK).
The LC-FTMS system consisted of a NANOSPACE SI-II HPLC, equipped with a dual pump system, an auto sampler, and a column oven (Shiseido, Tokyo, Japan), and a Q Exactive Orbitrap MS (Thermo Fisher Scientific, San Jose, CA) equipped with a heated-ESI-II (HESI-II) source for positive and negative ion mode analysis. LC separation was performed using a HILIC column (ZIC R ○ -pHILIC; 100 mm × 2.1 mm i.d., 5 µm particle size; Sequant, Darmstadt, Germany) with a gradient elution of solvent A (10 mmol L −1 ammonium bicarbonate in water, pH 9.2) and solvent B (acetonitrile) at 300 µL min−1. The data were collected using Xcalibur v4.1 software (Thermo Fisher Scientific, San Jose, CA). Details of the UHPLC-QTOF/MS and LC-FTMS operating conditions can also be found in previous reports 12 .
Data Processing. All data obtained from the four assays in the two systems, in both the positive and negative ion modes, were processed with Progenesis QI data analysis software (Nonlinear Dynamics, Newcastle, UK) for peak picking, alignment, and normalization, to produce peak intensities for t R and m/z data pairs. The range of automatic peak picking for the C18 and HILIC assays was between 0.5 and 13.0 min and between 0.5 and 9.0 min, respectively; the 'more 5' mode was selected in setting the threshold for the sensitivity of picking. Then, the adduct ions of each "feature" (m/z, t R ) were deconvoluted, and these features were identified from the human metabolome database (HMDB) and Lipidmaps. Features were selected based on their coefficient of variation (CV) with the SQC samples, which were injected after every 8 study samples; features with CV over 30% were eliminated. Features were also positively selected according to the inverse correlation of the dilution fold and the peak intensity to the dQC samples, as well as their CV with 3 injections of the same dQC samples. Then, the values of the compounds were imported to the Quantbolome (software) for log-median-regression. The normalization process has been described previously 12 . Finally, the values were normalized to the volume (in mg) of retinal material.
Hierarchical Cluster Analysis. A heat map was generated for the list of z scores for each metabolite, which were selected with the Kruskal-Wallis test (p < 0.04) when the CV was under 30% in each group (Ctrl, NC day 2, NC day 4 and NC day 7), using gplots package in the R program (v. 3.2.0). A dendrogram of the metabolites was made with the complete linkage clustering method and correlation distance measuring with the amap package. The colored scale bar, running from blue to white and red, represents low, medium, and high intensity metabolites, respectively.

Sample preparation for IMS analysis.
Eight-µm sections of the eye were obtained from the mice with a cryostat (CM 3050 S; Leica Microsystems, Wetzlar, Germany) and set on indium-tin oxide slides (100 ohm/sq; Matsunami, Osaka, Japan). The slides were then put into 50-mL plastic tubes with silica gel and the matrix was applied. Regions of the tissue samples exposed to laser irradiation were identified by light microscopic observation. Then, 660 mg of CHCA or 9-AA was deposited on the slides, at a thickness of 1.4 μm in an iMLayer (Shimadzu, Kyoto, Japan), to analyze the positive ion mode and negative ion mode. The slide glass was set on a box (cm x cm x cm) with a filter paper, which was impregnated with 350 µL of water/methanol = 95/5 (v/v). Then, the sample was incubated at 85 °C for CHCA, or 40 °C for 9-AA for 3 min. The sample was then dried in a desiccator for 30 min. The samples were immediately analyzed with MALDI-IMS (iMScope, Shimadzu).
MALDI-IMS analysis. MALDI-IMS analysis was performed with iMScope (Shimadzu). The mass spectra of the designated areas on a specimen photographed before matrix application were acquired in the positive and negative ion modes. Mass spectra were acquired under the following conditions: laser frequency and scanning mass ranged from m/z 130 to 280, m/z 500 to 750 and m/z 750 to 1,000 for positive, or m/z 750 to 1,000 for negative. Regions of the tissue samples exposed to laser irradiation were identified by light microscopic observation. The laser irradiation time, laser power, laser irradiation diameter, laser frequency, detection voltage, sample voltage and accumulated number of MALDI-IMS were 100 shots, 22 (pos) and 21 (neg), 10 μm, 1000 Hz, 2.1 kV, 3.5 (pos) and 3.0 (neg) and 1/pixel, respectively. A raster scan on the tissue surface was performed automatically. The number of pixels per scan was 57 × 53 (Ctrl) and 61 × 53 (NC day 2) for m/z 130 to 280, 68 × 53 (Ctrl) and 70 × 53 (NC day 2) for m/z 500 to 750, 55 × 52 (Ctrl) and 38 × 53 (NC day 2) for m/z 750 to 1,000 with the positive ion mode, and 69 × 48 (Ctrl) and 43 × 53 (NC day 2) with the negative ion mode. The spatial interval of data points was 10 μm, giving 3,121, 3,233, 3,604, 3,710, 2,860, 2,014, 3,312 and 2,279 data points in total for each section. The metabolites were identified with the MS/MS spectrum using the following chemical standards: SCIENTIfIC REPORtS | (2018) 8:11930 | DOI:10.1038/s41598-018-30464-z spermine and PCs. The data collected through the microscopic system were digitally processed with imaging MS solution analysis software (Shimadzu). The mass spectrum signal intensity at each point in the IMS analysis in the selected region of the GCL was detected automatically after a region of interest (ROI) was set with MS solution imaging software. The signal intensity in the ROI was measured with a slight modification of a method reported in previous work 66 . In total, an 18-point area in the controls and a 26-point area for NC day 2 were used to measure L-acetylcarnitine, a 143-point area was used for the controls, and a 120-point area was used for NC day 2 to measure PC (38:7) in two mouse retinas. Statistical analysis. The statistical significance of the number of RBPMS-positive cells was determined with Dunnett's multiple comparison test. P values < 0.05, compared to uninjured controls, were considered to be statistically significant. The intensities of the identified features were imported to the SIMCA 13.0 software (Umetrcxs, Umea, Sweden) for the multivariate analysis, and their relative quantities were evaluated with a PCA and an OPLS-DA. P-values were calculated with the Student's t-test or Wilcoxon rank sum test. For IMS analysis, P values were calculated with the Student's t-test.