Integrated multi-omics analysis reveals molecular changes associated with chronic lipid accumulation following contusive spinal cord injury

Functional and pathological recovery after spinal cord injury (SCI) is often incomplete due to the limited regenerative capacity of the central nervous system (CNS), which is further impaired by several mechanisms that sustain tissue damage. Among these, the chronic activation of immune cells can cause a persistent state of local CNS inflammation and damage. However, the mechanisms that sustain this persistent maladaptive immune response in SCI have not been fully clarified yet. In this study, we integrated histological analyses with proteomic, lipidomic, transcriptomic, and epitran-scriptomic approaches to study the pathological and molecular alterations that develop in a mouse model of cervical spinal cord hemicontusion. We found significant pathological alterations of the lesion rim with myelin damage and axonal loss that persisted throughout the late chronic phase of SCI. This was coupled by a progressive lipid accumulation in myeloid cells, including resident microglia and infiltrating monocyte-derived macrophages. At tissue level, we found significant changes of proteins indicative of glycolytic, tricarboxylic acid cycle (TCA), and fatty acid metabolic pathways with an accumulation of triacylglycerides with C16:0 fatty acyl chains in chronic SCI. Following transcriptomic, proteomic, and epitranscriptomic studies identified an increase of cholesterol and m 6 A methylation in lipid-droplet-accumulating myeloid cells as a core feature of chronic SCI. By characterizing the multiple metabolic pathways altered in SCI, our work highlights a key role of lipid metabolism in the chronic response of the immune and central nervous system to damage.


Background
Spinal cord injury (SCI) is increasingly recognized as a global health issue due to its poor neurological outcome requiring specialized care and leading to a significant burden on patients, their families, and society (GBD 2016 Traumatic Brain Injury andSpinal Cord Injury Collaborators, 2019;Zipser et al., 2022).
Growing evidence suggests that insufficient neurological recovery after SCI is linked with a state of persistent inflammation of the injured central nervous system (CNS) causing neurodegeneration and halting regenerative responses (Hamel et al., 2023;Zrzavy et al., 2021).In chronic SCI, both infiltrating leukocytes and resident microglia acquire pathogenic roles resulting in axonal damage, as well as loss of neuronal and oligodendrocytic populations (Yong et al., 2019).In the acute/ subacute phase after SCI, inflammatory responses are instead beneficial to the injured CNS, as they guide compensatory or reparative processes, promoting acute neuroprotection and regeneration (Tran et al., 2018;Yong et al., 2019).Therefore, understanding how inflammatory effector responses are temporally regulated in SCI may provide valuable insights on new mechanisms that could be modulated to promote the recovery of the injured CNS.
Integrative multi-omics is emerging as a promising approach to comprehensively map molecular changes driving pathology (Shafik et al., 2021;Yoon et al., 2022).We previously employed an untargeted proteomic profiling of the acute and subacute phases of a rodent model of SCI to study different myeloid activation states characterized by altered lipid metabolism and lysosomal function (Yao et al., 2021).We found significant apolipoprotein E (Apoe) upregulation in macrophages and microglia, which acted as a hub gene regulating lipid metabolism both in the subacute and chronic phases of SCI (Yao et al., 2022).Interestingly, Apoe − /− mice showed worse neurological dysfunction and neuroinflammation, thus suggesting a putative neuroprotective role for Apoe + myeloid subpopulations (Yao et al., 2022).Despite this evidence, the molecular mechanisms underlying the association between altered lipid metabolism and persistent neuroinflammation after SCI still need to be fully elucidated.
In this study, we used histopathology coupled with transmission electron microscopy (TEM) to characterize the CNS and myeloid responses in a mouse model of spinal cord hemicontusion injury (Huang et al., 2022).Next, we mapped pathological changes to fluctuations in the spinal cord proteome via weighted gene co-expression network analysis (WGCNA) and temporal clustering analysis.Lastly, we performed integrated lipidomics, transcriptomics, targeted proteomics, and m 6 A epitranscriptomics analyses to profile the molecular mechanisms of the chronic SCI phase.

Mice
N = 107 male C57BL/6 mice (8 weeks old) were acquired from the Laboratory Animal Center of Southern Medical University and housed in a temperature-controlled environment under a 12-h light-dark cycle with ad libitum access to food and water at the Laboratory Animal Center of Nanfang Hospital.Procedures were regulated under the authority of the Laboratory Animal Care and Use Committee of Nanfang Hospital, Southern Medical University.

C5 spinal cord hemicontusion injury
Mice were anesthetized using isoflurane (3% for induction; 1.5-2% for maintenance) and a C5 spinal cord hemicontusion injury was conducted as previously described (Huang et al., 2022).Briefly, following the removal of the C5 lamina, contusion SCI was induced using an electromagnetic servo material testing machine (Instron E1000; Instron, United States).The diameter of the impact tip was 1.0 mm, and the applied displacement and speed were set to 1.2 mm and 300 mm/s respectively.Biomechanical parameters were recorded in the process of contusion.Sham-operated control mice underwent the same surgical procedure without SCI contusion.

Oil Red O staining
Sections were pre-warmed and rinsed thrice using PBS, followed by 15 min staining with Oil Red O solution (G1263, Solarbio).Sections were then re-stained with hematoxylin, mounted with glycerol, and imaged using a Leica DM4000 microscope.
For analysis of demyelinated and damaged axons, at least 100 axons were analyzed in each mouse (n = 4 per time point) using the ImageJ software.Under TEM, we characterized typical microglia based on their relative small body, thin processes, bean-shaped or round nuclei and distinct heterochromatin, as described previously (Savage et al., 2018), whereas infiltrating monocyte-derived macrophages were mainly located in the lesion core and identified on the basis of abundant inclusions, large body, bi-lobulated or irregular nuclei (Yamasaki et al., 2014;Zrzavy et al., 2021).For quantification of LDs and lysosomes in microglia/macrophages over time, images of 8 microglia/macrophages from each mouse (n = 4 per time point) were analyzed.For morphological and ultrastructural comparison of resident microglia and infiltrating macrophages in the chronic phase of SCI, a total of 32 microglia in the lesion rim and 32 macrophages in the lesion core from 4 animals at 6-wpi were imaged and the area, circularity, aspect ratio, LDs, lysosomes, and cholesterol crystals were measured using the ImageJ software (St-Pierre et al., 2023).

Label-free proteomic analysis
At 3-dpi, 7-dpi and 6-wpi, mice were anesthetized and perfused with only PBS to harvest spinal cord tissues (5 mm rostral to 5 mm caudal to the epicenter).Proteins were extracted from the tissue samples (n = 6/ time point) using 8 M urea (U4883, Sigma-Aldrich).The concentration of each sample was determined using the bicinchoninic acid assay.Proteins were reduced using 5 mM dithiothreitol (A39255, Thermo Fisher Scientific) at 56 • C for 30 min, followed by alkylation with 10 mM iodoacetamide (A39271, Thermo Fisher Scientific) at room temperature for 30 min in the dark.Proteins were then digested overnight in trypsin (V5280, Promega) at 37 • C via filter-aided sample preparation using a 10 KD filter (VN01H02, Sartorious), an enzyme-to-protein ratio of 1:50.Data-dependent acquisition (DDA) was performed using an Orbitrap Fusion™ Tribrid™ High-Resolution Mass Spectrometer equipped with an Easy-nLC ™ liquid chromatography system (Thermo Fisher Scientific).Mobile phase A was 0.1% formic acid (FA) and mobile phase B was a mixture of acetonitrile (ACN) and 0.1% FA (4:1).Peptides were separated using a 15 cm × 75 μm NanoViper™ C18 column (particle size 2 μm, Thermo Fisher Scientific), flow rate of 300 nL/min with the following 120 min gradient: (1) 5-10% solvent B for the first 28 min, (2) 10-22% for 55 min, (3) 22-30% for 27 min and (4) 100% solvent B for the last 10 min.The precursors and fragments were analyzed using a detector operated in positive ion mode with a scan range of 350-1500 m/z.
Raw mass spectroscopy data was analyzed using a Linux-compatible MaxQuant version (Sinitcyn et al., 2018).Data filtration, normalization, imputation, and differential protein expression analysis was performed using the R package DEP (Zhang et al., 2018).Proteins were identified as differentially expressed proteins (DEPs) if meeting the following criteria: P < 0.05 and fold change>1.5 or < 0.67.Clusters of highly correlated proteins were identified using WGCNA (Langfelder and Horvath, 2008).Protein modules with consistent expression patterns following SCI were identified through fuzzy c-mean clustering using Mfuzz (Kumar and Futschik, 2007).

Untargeted lipidomics
Lipids were extracted from spinal cord tissues at 6-wpi (sham group: n = 6; SCI group: n = 6) using isopropanol precipitation, an established sample preparation procedure for untargeted lipid profiling using ultraperformance liquid chromatography− mass spectrometry (UPLC− MS) (Sarafian et al., 2014).Briefly, tissue homogenates (50 μL) were precipitated by adding three volumes of − 20 • C precooled isopropanol, vortexed for 1 min, incubated at 24 • C for 10 min, then stored overnight at − 80 • C. Homogenates were centrifuged at 16000 ×g for 20 min the following day.The quality control sample was constituted by pooling 20 μL of supernatant from each sample.Samples were analyzed using an Orbitrap Fusion™ Tribrid™ High-Resolution Mass Spectrometer equipped with a Vanquish™ Flex ultra-high-performance liquid chromatograph (Thermo Fisher Scientific).The extracts were separated using a 150 × 2.1 mm Acclaim™ C30 chromatographic column (particle size 3 μm, Thermo Fisher Scientific) at a flow rate of 0.3 mL/min at 40 • C. mobile phase A included 10 mM HCOONH 4 , 0.1% FA, and ACN: H 2 O (60:40), while mobile phase B included 10 mM HCOONH 4 , 0.1% FA, and isopropanol:ACN (90:10).Samples were analyzed in both the positive and negative ion mode.A full scan was obtained at 60,000 resolution and m/z values ranging from 200 to 2000, followed by an MS/ MS scan with DDA (dd-MS2, Top Speed) at a resolution of 15,000 in higher-energy collision dissociation (HCD) mode.
Lipid species were identified using the Lipid Search 4.1 software (Thermo Fisher Scientific).Intensities within individual samples were normalized to the protein concentrations.Lipidomic analysis was performed using MetaboAnalyst (www.metaboanalyst.ca)(Pang et al., 2022).Principal component analysis (PCA) and orthogonal projections to latent structures discriminant analysis (OPLS-DA) were used to separate the sample groups and identify lipids contributing to class separation relevant to SCI.The contribution of lipid classes to the OPLS-DA model was evaluated using the variable importance in projection (VIP) score.Statistical significance of a given lipid among experimental groups was assessed via student's t-test.Lipids of potential biological relevance were identified based on the cut-off values P < 0.05 and VIP > 1.

Transcriptomic analysis
Total RNA was extracted from spinal cord tissues at 6wpi (sham group: n = 5; SCI group: n = 5); mRNA was then purified using poly T oligo-attached magnetic beads.cDNA library fragments were purified using the AMPure XP system (Beckman Coulter, USA), followed by PCR amplification and purification using AMPure XP beads.RNA and library quality were assessed using an Agilent 2100 Bioanalyzer.Sequencing was performed using an Illumina NovaSeq 6000 with 150-bp paired-end reads.
Quality control check of the RNA sequencing data was performed using FastQC (https://www.bioinformatics.babraham.ac.uk/projects/fa stqc/).Raw reads were subjected to adapter trimming and low-quality base removal using Trim Galore (https://www.bioinformatics.babraham.ac.uk/projects/trim_galore), then mapped to the reference genome using HISAT2 (Kim et al., 2019).Transcript expression was quantified using Salmon (Patro et al., 2017).Differentially expressed genes (DEGs) across experimental groups were determined using the DESeq2 software based on cut-off values P < 0.05 and fold change >1.5 or < 0.67.Spearman's rank correlation analysis was used to evaluate correlations between transcriptomic and proteomic profiling.

Targeted proteomic analysis
Parallel reaction monitoring (PRM)-targeted proteomics was used to validate the DEPs identified in the unbiased label-free analysis at 6-wpi (sham group: n = 5; SCI group: n = 5).Peptides generated from protein enzymatic digestion were desalted using Pierce™ C18 Tips (87,782, Thermo Fisher Scientific) according to the manufacturer's instructions.
Quality control samples were prepared by mixing 2 μL of the supernatant from each sample.DDA of the pooled samples provided key information required to identify peptides (1-3 unique peptides of each protein), including amino acid sequences, retention time, charge state, and m/z values (Rauniyar, 2015).Mobile phase A was 0.1% FA and mobile phase B was a mixture of ACN and 0.1% FA (4:1).Identified peptides were monitored using an Orbitrap Fusion™ Tribrid™ High-Resolution Mass Spectrometer coupled to an Easy-nLC™ liquid chromatography system (Thermo Fisher Scientific) with a 15 cm × 75 μm column, flow rate of 300 nL/min with the same 120 min gradient described in label-free analysis.A full mass spectrum was acquired with a resolution of 60,000, target automatic gain control (AGC) values of 1.0 × 10 6 , maximum injection time of 50 ms, and m/z values ranging from 350 to 2000.This was followed by an MS/MS scan with a resolution of 30,000, target AGC values of 1.0 × 10 5 , maximum injection time of 100 ms, HCD collision energy of 30%, m/z values ranging from 120 to 2000, and an isolation window of 1 m/z.Ions peak area was used for peptide X.-Q.Yao et al. quantification using the Skyline software (MacLean et al., 2010).

Methylated RNA immunoprecipitation and sequencing
Total RNA was extracted from the spinal cord samples at 6-wpi (sham group: n = 3; SCI group: n = 3), then underwent integrity and concentration assessment using an Agilent 2100 Bioanalyzer (Agilent) and a SimpliNano spectrophotometer (GE Healthcare) respectively.Fragmented RNA (100 nt) was incubated with anti-m 6 A polyclonal antibody (Merck Millipore) at 4 • C for 2 h.Immunoprecipitated and input RNA was used for library generation using the Ovation SoLo RNA-Seq System Core Kit (NuGEN).Library preparations underwent 150 bp paired-end sequencing on the Illumina Novaseq platform as per standard protocols.

Functional enrichment analysis
Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analyses of DEGs, DEPs, and genes with differential m 6 A modification were performed using the package clus-terProfiler (version 4.2.2) (Wu et al., 2021).Pathway terms with P < 0.05 were considered statistically significant.

Statistical analyses
GraphPad Prism (version 8.2.1) was used to conduct remaining statistical analyses.One-way ANOVA was used to compare data from different time points.Kruskal-Wallis test followed by Dunn's test post hoc was used to analyze the immunofluorescence staining and the TEM results.The Mann-Whitney test was used to analyze BODIPY-cholesterol esters, m 6 A and BODIPY-neutral lipid staining, as well as PRM results.Unpaired t-test was used to analyze morphological changes between resident microglia and infiltrating macrophages.P < 0.05 was used as cut-off for statistically significance.

The injured spinal cord shows persistent myelin damage and axonal loss in the chronic phase of SCI
To characterize the CNS tissue response after SCI, we performed histological analyses during the acute (3-dpi), sub-acute (7-dpi) and chronic (6-wpi) phases of a mouse model of C5 spinal cord hemicontusion (Pukos et al., 2019) (Fig. 1A).This model is clinically relevant to the majority of traumatic SCI in humans (Steward and Willenberg, 2017) and employs displacement control to guarantee its reproducibility (Fig. S1A-D).
Immunofluorescence stainings showed consistent white matter damage after SCI (Fig. 1B) coupled with a reduction of the expression of myelin basic protein (MBP) and the pan-axonal marker SMI312 in the lesion rim, which persisted up to 6wpi (Fig. 1C).This effect was preceded by an increase of SMI32 + dephosphorylated neurofilaments, which peaked at 3-dpi, consistent with acute axonal damage (Schäffner et al., 2023) (Fig. 1D).
Temporally spaced TEM analysis of the lesion rim confirmed these findings, showing an increase of demyelinated and damaged axons at 3dpi that persisted up to 6-wpi, in line with a chronic state of demyelination and axonal loss after SCI (Fig. 1E).

Microglia and macrophages show progressive lipid accumulation after SCI
We next characterized the glial responses to SCI by performing immunostainings for GFAP and Iba1 in the lesion rim.We found that both Iba1 + microglia/macrophages and GFAP + astrocytes accumulated after SCI, persisting at 6 wpi (Fig. 2A).
Following TEM analysis of the lesion core and the surrounding rim showed major, temporally defined, intracellular changes in myeloid cells (Fig. 2B).At 3-dpi, resident microglia were activated and engaging in phagocytosis, while infiltrating macrophages were migrating to the lesion site and performed cellular debris clearance.At 7-dpi, both microglia and macrophages acquired a lipid-laden cellular state, characterized by the presence of lipid droplets (LDs), which were engulfed by "wrapping lysosomes".In the chronic phase of SCI (at 6-wpi), both microglia and macrophages contained significantly more LDs and lysosomes, suggesting their accumulation in myeloid cells over time.These lipid-droplet-accumulating myeloid cells were associated with significant lipid accumulation in the lesion core and rim, which also peaked at 6-wpi, as detected by the Oil Red O staining (Fig. 2C) and persisted even at longer time points after SCI (16wpi) (Fig. S2A-C).
To gain further insights into the unique features of lipid-dropletaccumulating myeloid cells at 6-wpi, TEM images were then mapped to semi-thin sections stained with toluidine blue (Fig. S3).We found that macrophages were aggregating in the lesion core, while microglia within the rim showed different morphological changes consistent with the known diverse myeloid responses in the lesion core and the surrounding rim (Milich et al., 2019;Zrzavy et al., 2021).Specifically, morphological analysis showed that microglia in the lesion rim had a smaller area being more elongated than macrophages in the lesion core (Fig. 2D).We also found that microglia in the lesion rim had less LDs and cholesterol crystals but similar number of lysosomes than macrophages in the lesion core (Fig. 2D).
We next performed confirmatory immunofluorescence stainings, which validated this chronic lipid-droplet-accumulating myeloid phenotype and unveiled a significant upregulation of the autophagy marker LC3 co-localizing with BODIPY-labeled LDs in F4/80 + myeloid cells (Fig. 2E).
Altogether, these data suggest that SCI is characterized by persistent glial reactivity, significant tissue lipid accumulation peaking at 6 wpi, and the presence of lipid-droplet-accumulating microglia and macrophages engaging in lysosomal-based autophagy (Van Broeckhoven et al., 2021).

SCI induces dynamic tissue metabolic changes
To investigate the molecular mechanisms sustaining the lipid accumulation seen after SCI, we performed label-free proteomic profiling of spinal cord tissues from the acute, subacute, and chronic phases of SCI.
After data quality control, we used WGCNA on 1755 proteins to identify eight distinct modules representing highly correlated protein networks (Fig. 3A; Table S1).Among them, the modules most significantly correlated with the acute (3-dpi) and the subacute (7-dpi) SCI samples were the green eigen module (149 proteins, r = 0.84, P = 4 × 10 − 7 ) and the turquoise eigen module (341 proteins, r = 0.62, P = 0.001), respectively (Fig. 3B).KEGG pathway analysis of the green module revealed enrichment in complement and coagulation cascades (for example, A2M, C3, CLU), proteasome (for example, PSMA3, PSMC2), and glutathione metabolism (for example, LANCL1, GCLC, MGST3) (Fig. S4A).Instead, we found that the proteins in the turquoise module were significantly enriched in the pathways associated with leukocyte trans-endothelial migration, tight junctions, oxidative phosphorylation, endocytosis, lysosome, phagosome, neurotrophin signaling pathway, proteasome and axon guidance (Fig. S4B).We also found that  the turquoise module was significantly enriched in GO-terms related to regulation of neuronal death, axonogenesis, positive regulation of neurogenesis and neuron development (Fig. S4C), which may be indicative of initial regenerative responses occurring within the inflammatory lesional environment in the subacute phase of SCI.When focusing on the chronic 6-wpi SCI samples, we found that the module most significantly correlated (and therefore likely associated with the lipid accumulation seen in chronic SCI) was the blue eigen module (305 proteins, r = 0.96, P = 2 × 10 − 13 ) (Fig. 3B).Further pathway analysis of this module identified an enrichment of pathways related to neurodegenerative diseases, along with proteins involved in metabolic pathways of oxidative phosphorylation (such as NDUFB4 and SDHC), tricarboxylic acid (TCA) cycle (such as IDH2 and SDHC), pyruvate metabolism (such as ALDH3A2 and ME2), glycolysis/gluconeogenesis (such as PFKL and PFKM), and fatty acid (FA) degradation (such as ALDH3A2 and ADH5) (Fig. 3C-D).
As these results were potentially indicative of a metabolic rewiring in the chronic phase of SCI, we then applied fuzzy c-means algorithm to further define protein clusters whose temporal expression changed across SCI phases (Fig. 3E).We found that cluster 4, which was upregulated at 3-dpi and remained high at 7-dpi and 6-wpi, exhibited an enrichment of lysosomal, phagosome, autophagy, and FA biosynthesis pathways (Fig. 3F).Analysis of proteins assigned to cluster 4 revealed an up-regulation of the rate-limiting enzyme in FA biosynthesis (ACACA) and lysosomal formation and structural proteins (LAMP1/2, CTSB, CTSD, HEXB) across all SCI phases (Fig. 3F).
Immunofluorescence stainings showed a significant upregulation of the glycolytic enzyme PFKL in Iba1 + microglia/macrophages in the lesion rim during the subacute and chronic phases of SCI (Figs.3H).
This temporal protein profiling is indicative of a form of metabolic rewiring in the SCI microenvironment, which persists in the chronic phases and involves the upregulation of proteins regulating different metabolic pathways, particularly those involved in lipid metabolism.

Triacylglycerides (TG) accumulate in the chronic SCI phase
We next focused on the 6-wpi tissues to identify which individual lipid species characterized the chronic SCI phase as compared to shamoperated controls.
A total of 1038 individual lipid species were identified (Fig. 4A; Table S2), of which half were phosphatidylcholine (PC) and phosphatidylethanolamine (PE) species, in line with the expected CNS lipid composition (Yoon et al., 2022).
Overall, these data suggest that several lipid classes characterize the chronic SCI, among which TG with C16:0 fatty acyl chains involved in processes such as de novo lipogenesis, lipid overload, and LD formation, play a key role.

Integrative untargeted analyses reveal lysosome-based autophagy and altered cholesterol metabolism in chronic SCI
To further investigate the mechanisms associated with lipid accumulation and LD formation in chronic SCI, we next performed a transcriptomic and proteomic profiling of spinal cord tissues collected at 6wpi vs sham-operated controls.
We next performed PRM-targeted proteomics to validate these labelfree proteomic hits (Fig. 5G).Using Skyline's peptide transition identification and peak integration (Figs.5H-I), we reliably quantified the expression trends of protein hits involved in cholesterol metabolism and lysosomes.Via fragment ions with the highest intensities for quantification, we detected an increase in HEXB and LAMP1 [markers of microglia (Masuda et al., 2020) and lysosomes (Cheng et al., 2018), respectively], in 6-wpi vs sham-operated controls (Fig. 5J).
We also found a downregulation of the cholesterol biosynthesis enzyme HMGCS1 (Itoh et al., 2018) and an upregulation of both lipid-Fig.2. Chronic SCI is characterized by tissue lipid accumulation and LD formation in microglia and macrophages.(A) Immunofluorescence staining showing significantly accumulation of GFAP + astrocytes and Iba1 + microglia/macrophages in the lesion rim of different phases of SCI.The yellow box area is shown below at higher magnification.n = 5 replicates per group.ns P > 0.05, * P < 0.05, ** P < 0.01, *** P < 0.001, using Kruskal-Wallis test followed by Dunn's test post hoc.Scale bars = 20 μm at low magnification and 10 μm at higher magnification.(B) Representative TEM images and quantification of LDs and lysosomes in microglia in the lesion rim (blue-green pseudo-coloring) and macrophages in the lesion core (orange pseudo-coloring) sampled at 3-dpi, 7-dpi, and 6-wpi.n = 4 replicates per group.ns P > 0.05, * P < 0.05, ** P < 0.01, *** P < 0.001, using Kruskal  binding proteins APOE and APOD, which have been involved in cholesterol efflux and ROS scavenging, respectively (Pascua-Maestro et al., 2017).Differences in proteins involved in PPAR signaling and ECM-receptor interaction such as DHRS1 and COL1A1 respectively failed instead to reach statistical significance.
Altogether these data indicated a regulated transcriptomic and proteomic response in chronic SCI, which involves microglia, lysosomes, as well as pathways regulating cholesterol synthesis and efflux.

Cholesterol adaptive responses in SCI are linked with m 6 A modifications
So far, our multi-omics data have shown that after SCI proteins and lipids involved in FA and TG metabolism are upregulated, in line with increased lipid deposition and LD formation in chronic SCI.This effect is coupled by a chronic transcriptional and translational response that involves both microglia and cholesterol metabolism.Therefore, we next investigated if any epigenetic modifications, specifically those of the m 6 A profile, which have been implicated in various neurological  diseases (Shafik et al., 2021), could play a role in regulating these pathways in chronic SCI.
We found that m 6 A peaks were characterized by canonical GGACU motifs (Fig. 6A) and were distributed near the coding sequence (CDS) and the 3′ untranslated region (3´UTR) (Fig. 6B).
At 6 wpi, we identified 815 genes with differential m 6 A peaks.These genes exhibited an enrichment of pathways involved in axonogenesis and guidance, TGF-β signaling, myelination, autophagy, fat cell differentiation and lipid metabolism (Figs.6C-D).
We next correlated differentially methylated genes with their mRNA  and protein levels to study their influence on gene/protein expression.This approach allowed us to consistently overlap methylated profiles to 6 differentially expressed proteins (DEPs) and 30 DEGs, which converged on the cholesterol biosynthesis enzyme HMGCS1 (Figs. 6E-F; Table S5).Chronic SCI was also characterized by significantly upregulated m 6 A methylation in transcription factor Mafb, TGF-β signaling factor (Tgif1), and the 3´UTR region of collagen domain-binding receptor Lair1, which have all been previously associated with myeloid cell function (Keerthivasan et al., 2021;Matcovitch-Natan et al., 2016).
Given the significant downregulation of HMGCS1 previously seen at 6wpi (Fig. 5J), we next correlated the mRNA and protein of Hmgcs1 to hypomethylation at multiple m 6 A sites (Fig. 6G).We found that these changes occurred on a background of elevated m 6 A levels at Mylip, Daam2 and Pdk4 sites.While Pdk4 is a key enzyme regulating glycolysis, Mylip and Daam2 are involved in cholesterol homeostasis, suggesting reduced low-density lipoprotein (LDL) cholesterol uptake and synthesis, but increased efflux (Cristobal et al., 2022;Li et al., 2020).
Therefore, to understand how cholesterol changes may be linked to myeloid cells, we finally assessed the levels of BODIPY-cholesterol esters in F4/80 + cells.We found that in chronic SCI, myeloid cells showed a significant increase of BODIPY-cholesterol esters LDs (Fig. 6H), which was coupled by a significant increase in m 6 A + punctae co-localizing with LDs (Fig. 6I).
Collectively these results show cholesterol accumulation and LD formation in myeloid cells during the chronic phase of SCI is linked with increased m 6 A methylation.

Discussion
In this study, we employed a temporal ultrastructural approach to provide a comprehensive characterization of the molecular changes that occur in the spinal cord after injury.By doing so, we were able to uncover significant transcriptomic, proteomic and metabolic changes that support extensive metabolic rewiring after SCI.
Metabolic rewiring occurs when cells undergo a shift in their utilization of metabolic pathways upon changes in substrate availability in the microenvironment (Forteza et al., 2023;Peruzzotti-Jametti et al., 2021).In the acute phases after SCI, we found that proteins involved in oxidative phosphorylation, endocytosis and neurogenesis, were all upregulated.We interpret these changes as supportive of an initial proregenerative response within the lesional microenvironment (Forteza et al., 2023;Peruzzotti-Jametti et al., 2021).In the chronic phase of SCI, other pathways became more prominent instead.These included a shift towards glycolysis, accompanied by de novo lipogenesis and LD formation.
De novo lipogenesis has been described as both a pro-regenerative response during remyelination (Batchuluun et al., 2022;Dimas et al., 2019), and a driver of oxidative stress and inflammation (Haney et al., 2024;Marschallinger et al., 2020;Yoon et al., 2021).Specifically, this latter function has been described in cases of de novo lipogenesis associated with intracellular accumulation of lipids in LD-accumulating glia found in Alzheimer's disease (AD) and aged human brain (Haney et al., 2024;Marschallinger et al., 2020).Herein we demonstrate that similar changes occur also in chronic SCI on a background of epi-transcriptomic regulation of cholesterol metabolism.
Cholesterol is produced by glia, but CNS cells cannot catabolize it and therefore its efflux and/or storage in LDs is promoted during lipid overload (Berghoff et al., 2021;Luo et al., 2020;Vitali et al., 2014).
In chronic SCI, we show that these mechanisms are tightly regulated by a m 6 A-mediated downregulation of cholesterol synthesis and uptake (HMGCS1 and FDPS) and upregulation of its efflux (APOE) and storage (LDs), with the latter being specifically observed in myeloid cells.However, these changes may also represent an adaptive response to restore cholesterol homeostasis and promote repair in chronic SCI, in line with previous evidence showing worsened SCI recovery upon Apoe knockout and improved remyelination upon boosting cholesterol efflux during demyelination (Blanchard et al., 2022;Yao et al., 2022).Future functional studies using myeloid cell specific knockouts of genes involved in these pathways will be needed to clarify their putative detrimental/protective role in chronic SCI.
We also acknowledge further limitations of our current study.The use of whole-tissue, rather than single-cell, multi-omics analysis limits our ability to identify cell-type-specific changes.Therefore, we cannot infer whether these changes are specific for any microglial activation state or correspond to a general cellular response to lipid overload also found in microglia.Further work is needed to characterize myeloidspecific metabolic processes and their role in SCI.In addition, some experiments may have been underpowered to fully cover changes in rare metabolic pathways which, coupled with unequal sample sizes, may explain discordances between our transcriptome and proteome data.Finally, few cholesterol species passed quality control using the isopropanol extraction method (Sarafian et al., 2014), which will need to be further optimized in follow up experiments.
Despite these limitations, our study represents the first and most extensive characterization of the gene, protein, metabolic and epigenetic changes specific to the chronic SCI.In addition, our multi-omics integration with temporal ultra-structural profiling has revealed how activated myeloid cells adapt to a hostile lipid overload environment, showcasing the importance of epi-transcriptomic regulation of metabolic rewiring in chronic inflammation.Finally, we reveal that reactive myeloid cells accumulate LDs enriched in m 6 A modifications in an environment chronically characterized by a m 6 A-mediated, adaptive glycolytic and cholesterol responses.A deeper understanding of these molecular changes might have important therapeutic implications for SCI.Indeed, microglia cholesterol synthesis blockade promotes axonal regeneration upon CNS injury (Shabanzadeh et al., 2021), whereas preserving an epigenetic acetylation programme induces an antiinflammatory phenotype in myeloid cells from patients with neuroinflammatory disorders (Zierfuss et al., 2020).
In conclusion, we use a clinically relevant SCI model to show that chronic lipid accumulation, adaptive metabolic responses and epitranscriptomic modifications are core features of chronic CNS damage after SCI.The molecular changes described in this work will provide the backbone for future studies aiming to develop new immunomodulatory therapeutic strategies for chronic disorders of the CNS.

Fig. 1 .
Fig. 1.SCI leads to persistent myelin and axonal damage.(A) Schematic illustration of the experimental design of the study, including C5 spinal cord hemicontusion injury and integrated analysis of pathological characteristics and multi-omics.(B) Representative confocal images and quantification of the area of spared white matter in different phases of SCI.Purple, MBP; green, SMI312; blue, DAPI.Scale bars = 200 μm.(C) Representative immunofluorescence staining and quantification of the MBP-positive myelin and SMI312-positive axons in different phases of SCI.n = 5 replicates per group.Statistical analysis was performed using Kruskal-Wallis test followed by Dunn's test post hoc.Scale bars = 20 μm.(D) Representative immunofluorescence staining and quantification of the SMI32-positive damaged axons in the lesion rim of different phases of SCI.The yellow box area is shown below at higher magnification.n = 5 replicates per group.Statistical analysis was performed using Kruskal-Wallis test followed by Dunn's test post hoc.Scale bars = 20 μm at low magnification and 10 μm at higher magnification.(E) Representative TEM images and quantification of the ratio of demyelinated and damaged axons to total axons in the spinal cord of sham, 3-dpi and 7-dpi and 6-wpi.* indicates myelinated axons, + indicates demyelinated axons, × indicates damaged axons.n = 4 animals per group.Statistical analysis was performed using Kruskal-Wallis test followed by Dunn's test post hoc.Scale bars = 1 μm.
X.-Q.Yao et al.   (caption on next page) X.-Q.Yao et al.
-Wallis test followed by Dunn's test post hoc.* indicates LDs, × indicates damaged myelin components, arrow indicates "wrapping lysosomes".Scale bars = 1 μm.(C) Representative images and quantification of Oil Red O staining showing chronic lipid accumulation in the lesion after SCI.White boxes correspond to high-magnification images.n = 4 animals per group; ns P > 0.05, * P < 0.05, *** P < 0.001, using Kruskal-Wallis test followed by Dunn's test post hoc.Scale bars = 20 μm (high magnification images) and 50 μm (low magnification images).(D) Representative TEM images and comparison of morphological and ultrastructural changes between microglia in the lesion rim (blue-green pseudo-coloring) and macrophages in the lesion core (orange pseudo-coloring) at 6-wpi. 1 indicates primary lysosomes, 2 indicates secondary lysosomes, 3 indicates tertiary lysosomes, × indicates myelin components.n = 32 cells per group from 4 animals.ns P > 0.05, *** P < 0.001, using unpaired t-test.Scale bars = 1 μm.(E) Representative immunofluorescence staining and quantification in the lesion rim of LC3 + BODIPY-neutral lipid + F4/80 + microglia/macrophages in chronic SCI.n = 4 replicates per group.* P < 0.05, using the Mann-Whitney test.Scale bars = 10 μm.(For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)X.-Q.Yao et al.

Fig. 3 .
Fig. 3. WGCNA and temporal clustering analysis of untargeted proteomic profiling in spinal cord samples at the three phases of SCI.(A) Clustering dendrogram and (B) module-trait relationships of 24 samples showing distinct co-expression modules identified by WGCNA (n = 6 3-dpi SCI, n = 6 7-dpi SCI, n = 6 6-wpi SCI and n = 6 sham group).(C) Bubble plot and (D) chord plot showing the KEGG pathway enrichment analysis of the proteins in the blue module.(E) Temporal clustering analysis identifying six distinct clusters using fuzzy c-means algorithm.Chord plot showing the KEGG pathway enrichment analysis, as well as individual proteins most significantly assigned to each term, of the proteins in (F) cluster 4 and (G) cluster 6. (H) Immunofluorescence staining showing a significant accumulation of Iba1 + PFKL + microglia/macrophages in the lesion rim during the subacute and chronic phases of SCI.The yellow box area is shown below at higher magnification.n = 5 replicates per group.Scale bars = 50 μm at low magnification and 20 μm at higher magnification.ns P > 0.05, ** P < 0.01, *** P < 0.001, using Kruskal-Wallis test followed by Dunn's test post hoc.(For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 4 .
Fig. 4. Untargeted lipidomic analysis of the spinal cord samples at the chronic SCI phase.(A) Pie chart showing the distribution of the 1038 profiled individual lipid species according to their lipid classes.(B) PCA of the lipidomics dataset from 12 differentiated samples (n = 6 6-wpi SCI, n = 6 sham group).(C) Top 20 critical lipids identified by OPLS-DA.(D) VIP scores of lipid classes obtained from OPLS-DA.(E) Bubble plot showing the distribution and corresponding concentration of 235 critical lipid species (P < 0.05, VIP > 1).Each dot is representing individual lipid species of a specific lipid class.(F) Heatmap showing the 148 critical lipid species with differential changes (P < 0.05, VIP > 1, fold change >1.5 or < 0.67).

Fig. 5 .
Fig. 5. Integrated analysis of the transcriptome and proteome profiling at the chronic SCI phase.PCA showing good intra-consistency and inter-group differences at both the (A) RNA and (B) protein levels between chronic SCI and controls.(C) Correlation between 1625 pairs of mRNA levels and protein abundance.(D) Upset plot showing dysregulated genes at both RNA and protein levels.(E) Heatmap showing the 52 mutual upregulated and downregulated genes in the 12 proteome samples.(F) Chord plot showing KEGG pathway enrichment analysis of the mutually upregulated and downregulated genes.(G) KEGG pathway enrichment analysis of the DEPs at 6-wpi (n = 6 6-wpi SCI) as compared to controls (n = 6 sham group) in the label-free proteomics dataset.Representative images of (H) MS/MS spectra for the identification and extraction of fragment ion chromatograms in the peptide TANLGAGAAQPLR and (I) total integrated fragment ion signal for peptide at different samples (contribution from each individual fragment ion is represented as a different colour in the bars).(J) Expression levels of Hexb (top left), Lamp1 (top middle), Hmgcs1 (top right), Apoe (bottom left), Apod, Dhrs1 (bottom middle) and Col1a1 proteins (bottom right), all identified via PRM-targeted proteomics.

Fig. 6 .
Fig. 6. m 6 A profiling of the spinal cord samples at the chronic SCI phase.(A) Canonical GGACU motifs of the m 6 A peaks.(B) Distribution of m 6 A sites in the transcript regions.Abbreviations: CDS = coding sequence, UTR = untranslated region.Biological processes enriched in (C) KEGG and (D) GO pathway enrichment analysis of the 815 genes with differential m 6 A peaks in chronic SCI (n = 3 6-wpi SCI) as compared to controls (n = 3 sham group).(E) Venn diagram showing overlap of the differentially methylated genes, DEPs, and DEGs.(F) Heatmap showing the 30 DEGs with differential m 6 A peaks.(G) IGV visualization of differential m 6 A peaks: DEGs/DEPs overlap such as Hmgcs1 (top left), DEGs only Mylip (top middle), DEPs only Daam2 (top right), DEGs only Pdk4 (top right), DEGs only Mafb and Tgif1 (bottom left) and DEGs only Lair1 and Colec12 (bottom right).Representative immunofluorescence staining and quantification in the lesion rim of BODIPYcholesterol esters + F4/80 + microglia/macrophages (H) and m 6 A + BODIPY-neutral lipid + F4/80 + microglia/macrophages (I) in chronic SCI.n = 5 replicates per group.** P < 0.01, using the Mann-Whitney test.Scale bars = 5 μm.