Intracellular complement factor H protects neurons during CNS inflammation

In MS, a disease that predominantly affects young adults, inflammation is followed by early and widespread neurodegeneration throughout the CNS¹. The disease is thought to be driven by autoreactive T cells infiltrating the CNS and triggering chronic myeloid cell activation that sustains low-grade inflammation and gradually drives neurodegeneration across CNS regions, including the eye^3,4. Although several immunomodulatory therapies target the peripheral immune system, none address the CNS-intrinsic inflammatory environment or reinforce stressed neurons against inflammation-induced degeneration, limiting their impact on progressive disability⁵.

In addition to motor and sensory impairments, visual disturbances are among the most common presenting symptoms in people with MS⁴. Retinal atrophy is an early feature of MS and can be detected even without a previous episode of optic neuritis⁶. Thus, acute and chronic low-grade CNS inflammation leads to substantial degeneration of RGCs, the output neurons of the retina of which the axons form the optic nerve.

RGCs are among the most diverse neuronal populations, with over 40 distinct subtypes described in mice². This diversity has driven investigations into subtype-specific vulnerability, particularly in models such as optic nerve crush, where around 80% of RGCs degenerate within the first 2 weeks². Notably, because humans evolved a fovea for high-resolution colour vision, RGCs in this region may face a trade-off of increased susceptibility to degeneration, as seen in age-related macular degeneration (AMD) and MS, emphasizing the need to study the mechanisms of retinal degeneration in humans⁷.

Recent work has begun to resolve the transcriptional heterogeneity of human RGCs^8,9,10, but cell type identification remains inconsistent and datasets are dominated by midget RGCs, limiting analysis of rarer subtypes. Although MS substantially affects the retina, single-cell studies have focused largely on the brain^11,12,13 and histopathological studies of the MS retina are sparse¹⁴, leaving a gap in our understanding of how MS-associated inflammation affects retinal neurons. Moreover, most single-cell studies describe molecular changes without translating them into clinical concepts or therapies. To address this, we performed single-nucleus sequencing analysis of RGCs from control and MS retinas to identify genes determining neuronal vulnerability to inflammation, then validated and mechanistically explored our findings to identify how neurons protect themselves against inflammation across the CNS.

To capture the heterogeneity of RGCs, which constitute only around 1% of retinal cells, we developed a fluorescence-activated nucleus sorting (FANS) protocol to extract their nuclei from human post-mortem retina⁸ (Fig. 1a). We used NeuN as a pan-neuronal marker and also labelled RGCs with an antibody against RNA-binding protein with multiple splicing (RBPMS), the principal RGC marker¹⁵, revealing the expected severe RGC loss in individuals with MS (Extended Data Fig. 1a). We sorted NeuN^highRBPMS^high nuclei (Fig. 1b) and confirmed RGC enrichment using quantitative PCR with reverse transcription (RT–qPCR), observing a strong increase in RBPMS transcripts and depletion of recoverin (RCVRN), a photoreceptor marker (Extended Data Fig. 1b).

For single-nucleus RNA-sequencing (snRNA-seq) analysis, we next focused on the macula region, where MS-associated RGC degeneration is most severe¹⁶. Macula tissue was obtained from ten donors without retinal pathology (40% female; mean age, 55.8 years) and ten donors with MS (60% female; mean age, 58.6 years; Supplementary Table 1). After quality control, including removal of doublets, low-quality nuclei, nuclei with over 5% mitochondrial gene content (Supplementary Fig. 1a) and non-RGC nuclei, we retained 105,502 high-quality RGC transcriptomes (Fig. 1c and Supplementary Fig. 1c,d), with a mean of 4,567 genes detected per nucleus (Supplementary Fig. 1d).

RGC identity was confirmed by RBPMS expression (Fig. 1d). Small or inconsistent clusters were merged with their nearest neighbour (Supplementary Fig. 1e), and the resulting clusters were annotated using gene set enrichment against three published human RGC datasets^8,9,10 (Extended Data Fig. 1c) and known RGC markers^2,8 (Extended Data Fig. 1d,e). Consistent with previous studies^8,9,10, midget RGCs were the largest population, followed by parasol RGCs (PGs), which subdivided into PG-ON and PG-OFF groups based on the transcriptional profile. The third most frequent type comprised FOXP2-expressing F-RGCs. RGC20, RGC21 and RGC25 expressed markers of ON–OFF direction-selective RGCs² (SATB2, CARTPT, NEUROD2 and MMP17), while RGC9, RGC14, RGC22, RGC24, RGC26 and the PGs expressed SPP1, a marker for alpha RGCs in mice². We also identified OPN4-expressing intrinsically photosensitive RGCs (ipRGCs) and TAFA4-expressing populations² (Extended Data Fig. 1d,e). Thus, our dataset resolved all major RGC subtypes with high concordance to existing human datasets in both control and MS samples.

To assess whether RGC subtypes differ in vulnerability to inflammation, we calculated the relative frequency of each cluster within the total RGC population in MS and control samples (Fig. 1e), revealing significant differences in survival (Fig. 1f,g). Midget RGCs and ipRGCs showed the greatest loss, whereas the PG-ON, RGC26 and RGC13 subtypes were relatively preserved (Fig. 1h,i and Extended Data Fig. 1f).

To identify transcriptional features underlying resilience while reducing the influence of subtype-specific marker genes, we generated mixed pseudobulk profiles by aggregating total gene counts from the most susceptible and most resilient RGCs for each sample and condition, followed by differential expression analysis (Extended Data Fig. 2a–f). We defined several gene modules: module A (higher in resilient than susceptible RGCs; intrinsic resilience genes) was associated with synapse organization (such as GRID2, SYNDIG1 and LRRTM3), axonogenesis (NGF, SEMA5B, EFNA5 and SLIT2), complement system (CFH) and potassium ion transmembrane transport (KCNJ16, KCNB2 and CHRNA2); module B (higher in susceptible than resilient RGCs; intrinsic susceptibility genes) was mainly involved in glutamatergic and calcium signalling (GRIA4, GRIN2B, GRIN2C and CACNG2); module C comprised genes upregulated during MS in resilient RGCs associated with stress responses (JUN, JUND, FOSB, ACSL4 and SQSTM1) and inflammatory signalling (NFKBIA, NFKBIZ and TGFB2); module D contained transcripts downregulated in resilient RGCs in MS, indicating suppression of neuronal identity (TBR1, POU4F1 and EBF2) and excitability programs (DLGAP1, SORCS1 and GRIN2C), alongside downregulation of metabolic and redox-defence pathways (GPT2 and PSAT1); module E comprised genes upregulated in susceptible RGCs during MS, reflecting extracellular-matrix remodelling (DCN, SCUBE1 and BMP5) and mitochondrial stress (MT-ATP8, MT-ND3, MT-ND4 and MT-ND4L); and module F comprised six downregulated transcripts in susceptible RGCs during MS. We validated modules A–E using AUCell analysis of pseudobulked susceptible, intermediate-susceptible, intermediate-resilient and resilient RGCs (Extended Data Fig. 3a–e), and assessed the resilience (module A) and susceptibility (module B) signatures across all 27 RGC subtypes (Extended Data Fig. 3f,g).

We next performed linear regression between resilience (change in relative frequency during MS; Fig. 1f) and pseudobulked gene expression, identifying genes that are positively (resilience candidates) and negatively (susceptibility candidates) associated with resilience (Fig. 1j,k and Extended Data Fig. 3h,i). CFH and GPR149 emerged as the strongest, most biologically plausible candidates from both analyses. CFH was enriched in resilient RGC populations (module A), particularly RGC13 and RGC26 (Extended Data Fig. 3j), correlated positively with resilience (Fig. 1j) and was upregulated in RGCs from people with MS (Extended Data Fig. 3k), suggesting an adaptive neuronal stress response. CFH encodes a secreted glycoprotein that is the principal inhibitor of the alternative complement pathway and is abundant in plasma¹⁷. Notably, CFH variants have been implicated in neurodegenerative diseases, and its expression in retinal pigment epithelium has been studied in AMD, where the Y402H CFH variant accounts for over 40% of genetic risk for AMD^18,19,20.

Conversely, GPR149 was enriched in susceptible RGCs (module B), correlated with increasing susceptibility (Fig. 1k), was de-enriched in CFH-expressing RGCs (Extended Data Fig. 3l) and has a single-nucleotide polymorphism associated with increased MS risk²¹.

Multiplex single-molecule fluorescence in situ hybridization (smFISH) validated our bioinformatic findings, showing significant preservation of CFH-expressing RGCs in MS retinas, while the relative frequency of GPR149-expressing RGCs was reduced (Fig. 1l,m). Moreover, several other module A/B genes were recovered by the regression analysis (Extended Data Fig. 3h,i). As CFH expression correlated with baseline resilience and increased further in MS, we focused on it as a candidate mediator of both intrinsic and adaptive neuroprotection.

To examine whether neuronal CFH induction during CNS inflammation extends beyond the retina and represents a general feature of CNS neurons, we performed immunohistochemistry analysis of post-mortem cortex from donors with MS and control donors without inflammatory diseases (Fig. 2a, Extended Data Fig. 4a and Supplementary Table 2). Notably, CFH expression was minimal to absent in control individuals but was significantly elevated during MS both in chronic active lesions and normal-appearing grey matter, with upregulation predominantly localized to neurons (Fig. 2a).

In experimental autoimmune encephalomyelitis (EAE), CFH protein increased significantly in spinal-cord motor neurons of C57BL/6 mice during the acute and chronic phase of EAE (Fig. 2b). This was corroborated by immunoblotting of EAE spinal cords (Extended Data Fig. 4b), and by published cell-type-specific mRNA datasets^{22,23,24,25,26} showing increased CFH primarily in neurons and CNS-associated macrophages, with no significant changes in other cell types (Extended Data Fig. 4c). Confirming that this is a general response across MS models, neuronal CFH was similarly induced during chronic EAE in SJL/JRj (hereafter, SJL) mice, a relapsing–remitting model^27,28 (Extended Data Fig. 4d). Analogous to the human retina, we observed a significant increase in CFH in RGCs of the C57BL/6 EAE mouse retina (Extended Data Fig. 4e). In CNS inflammation, CFH levels were comparable between sexes across human RGCs, cortical neurons and mouse spinal motor neurons (Extended Data Fig. 4f–h).

To identify CFH-inducing stimuli, we exposed primary mouse neurons to interferon-γ (IFNγ) and glutamate. Cfh expression was induced not only by the combination of IFNγ and glutamate but also by each stimulus individually (Extended Data Fig. 4i). As glutamate causes oxidative stress among other effects, we applied the GPX4 inhibitor (1S,3R)-RSL3 (RSL3) as a more targeted oxidative stressor, which is known to induce lipid peroxidation²⁹, and the neurotoxic aldehyde malondialdehyde (MDA)³⁰. Immunocytochemistry analysis confirmed a significant increase in CFH protein under all conditions, indicating that inflammatory and oxidative stressors independently induce neuronal CFH (Extended Data Fig. 4j).

To determine its functional relevance in vivo, we used an adeno-associated virus (AAV) expressing Cre recombinase under the neuron-specific human synapsin 1 (hSYN1) promoter to achieve neuronal Cfh deficiency in Cfh^flox/flox mice before inducing EAE (Supplementary Fig. 2a–c). Neuronal Cfh-deficient mice showed a more severe disease course and greater visual impairment affecting both visual acuity and contrast sensitivity (Fig. 2c,d and Extended Data Fig. 4k–p), corroborated by increased neuronal loss in the ventral horn of the spinal cord (Fig. 2e) and accompanied by more extensive axonal damage and demyelination (Fig. 2f and Extended Data Fig. 4q). Notably, neuronal Cfh deficiency did not alter the peripheral immune cell compartment or immune cell infiltration into the spinal cord during acute EAE (Supplementary Figs. 3 and 4). Together, CFH is upregulated in neurons during MS and EAE, probably as a stress response, and its absence exacerbates disease severity and visual impairment, demonstrating a critical role in protecting neurons against inflammation-induced degeneration.

To evaluate the therapeutic potential of neuronal CFH and elucidate its mechanism of action, we examined whether expression could pre-emptively protect against inflammation-induced neurodegeneration. Given the large size of CFH, we used two published truncated versions to compare its N- and C-terminal domains. The first, mouse homodimeric minimal factor H (mHDM-FH), is an effective cofactor for human factor-I-mediated cleavage of mouse C3b and reduces glomerular C3 staining in Cfh-deficient mice^31,32. The second, CR2-FH, is a fusion of a complement receptor 2 fragment with the N-terminal short consensus repeat (SCR) domains 1–5 of CFH, reported to reduce choroidal neovascularization in an AMD model when delivered by AAV³³ (Fig. 3a). At 3 weeks before EAE induction, we administered AAVs encoding mHDM-FH, CR2-FH or a GFP control intravenously under the hSYN1 promoter for neuron-specific CNS expression (Supplementary Fig. 5a–c), without relevant expression in the enteric nervous system (Supplementary Fig. 5d). Notably, only mHDM-FH significantly reduced EAE scores in both male and female mice and neuronal loss in the chronic phase (20–29 days after immunization), whereas CR2-FH had no measurable effect (Fig. 3b,c and Extended Data Fig. 5a–k).

To determine whether protection reflected altered immune responses, we performed a comprehensive immune cell characterization using flow cytometry during acute EAE and found no significant differences in the immune cell infiltrate or activation between mHDM-FH- and GFP-expressing mice (Supplementary Fig. 6a–e). Analysis using an enzyme-linked immunosorbent assay (ELISA) validated the unchanged plasma CFH levels (Extended Data Fig. 5l). We also examined C3b and factor Bb, which together constitute the alternative-pathway C3 convertase that is inhibited by CFH. Both were significantly elevated in the spinal cord during acute EAE (Extended Data Fig. 5m), but did not differ between mHDM-FH- and GFP-expressing animals (Extended Data Fig. 5n). As plasma and spinal cord levels of C3, C3a and C5a and the C3a/C3 ratio at peak disease also did not differ between constructs (Extended Data Fig. 5o–v), there was no evidence that complement activation contributed to the phenotype. Thus, neuronal mHDM-FH, but not CR2-FH, protects against inflammatory neurodegeneration without altering the immune response or directly inhibiting the alternative complement pathway.

mHDM-FH, but not CR2-FH, contains the C-terminal SCR18–20 domains, implicating them in neuroprotection. Whereas N-terminal SCR1–4 confer complement-regulatory activity, C-terminal SCR18–20 mediate binding to cell surfaces and dying cells by recognizing danger signals including sialic-acid-containing glycosaminoglycans, heparan sulphates, surface-bound C3b and MDA³⁴. MDA is a byproduct of lipid peroxidation, an indicator of oxidative stress and a highly reactive aldehyde that modifies nearby proteins and lipids. Its levels were increased in neurons during EAE (Extended Data Fig. 6a), indicating oxidative damage, lipid peroxidation and ferroptosis³⁵, all of which are implicated in neuronal damage in MS and EAE^36,37,38. Consistently, our pseudobulk analysis of MS versus control RGCs confirmed pronounced enrichment of oxidative-stress signatures (Extended Data Fig. 6b). mHDM-FH-expressing neurons showed reduced levels of the oxidative-stress markers COX2 and inducible nitric oxide synthase as well as MDA-modified proteins (Extended Data Fig. 6c–e), whereas MDA-modified proteins were increased in neuronal Cfh-deficient mice during EAE (Extended Data Fig. 6f), leading us to hypothesize that CFH acts by mitigating oxidative stress.

For mechanistic studies, we generated neuronal Cfh-knockout cells in vitro by transducing Cfh^flox/flox neurons with AAV-hSYN1-cre (Cfh-cKO; Extended Data Fig. 6g) and used lentiviral vectors to expand our construct repertoire, generating full-length CFH and additional truncated variants (Fig. 3a). On the basis of the hypothesis that neuroprotection is mediated by the C-terminal domains, we also created constructs lacking SCR20, lacking the signal peptide for intracellular retention and a construct comprising the SCR20 domain alone³⁹.

To induce oxidative damage, we applied three stimuli: glutamate, an excitotoxic agent that increases reactive oxygen species (ROS), promotes lipid peroxidation and activates ferroptosis (Extended Data Fig. 6h); MDA, which can bind to CFH⁴⁰, reduces neuronal viability in a manner partly rescued by the ferroptosis inhibitor liproxstatin-1 (Extended Data Fig. 6i) and increases ROS and lipid peroxidation, suggesting that it sensitizes cells to ferroptosis in a feed-forward manner (Extended Data Fig. 6j,k); and the GPX4 inhibitor RSL3, which directly induces lipid peroxidation. Glutamate induced Cfh in wild-type neurons, but not in Cfh-cKO neurons (Extended Data Fig. 6l), with no baseline differences in viability, ROS or lipid peroxide content between genotypes (Extended Data Fig. 6m). After glutamate, MDA or RSL3 treatment, Cfh-cKO neurons showed reduced viability (Fig. 3d), increased ROS (Fig. 3e) and increased lipid peroxidation (Fig. 3f). These differences were further amplified by chronic IFNγ exposure (Extended Data Fig. 6n–p), which exacerbates ferroptosis susceptibility³⁷. Bulk RNA-seq also revealed an increased excitotoxicity signature in Cfh-cKO neurons treated with glutamate (Extended Data Fig. 6q–s). Conversely, CFH construct expression conferred significant protection against all stressors, but only when constructs included SCR20 (Fig. 3g–q).

The subcellular localization of each construct was validated by immunocytochemistry in primary neurons (Supplementary Fig. 7a–j and Extended Data Fig. 7a,b), Pfa1 fibroblasts⁴¹ (Supplementary Fig. 8a–i) and HEK293T cells (Supplementary Fig. 9a–i). CFH, CFH-KDEL, mHDM-FH, mHDM-FH-KDEL and the SCR20 subunit localized predominantly to the endoplasmic reticulum (ER), whereas the signal-peptide-deficient CFH(ΔSP) and mHDM-FH(ΔSP) variants distributed diffusely, partially co-localizing with ER, lysosomes, mitochondria and nucleus. To corroborate this, we performed proximity labelling with an engineered biotin ligase, TurboID⁴², targeted to the cytoplasm (TurboID-Cyto) or ER lumen (TurboID-ER), validating compartment-specific enrichment in cortical neurons (Extended Data Fig. 7c–f). Both endogenous and expressed CFH and mHDM-FH localized to the ER, whereas CFH(ΔSP) was mainly detected in the cytoplasm (Extended Data Fig. 7g). In vivo, neuron-specific ER-targeted TurboID revealed that endogenous neuronal CFH was largely undetectable at steady state but became robustly induced and enriched in the ER fraction during EAE (Extended Data Fig. 7h). Together, neuronal CFH localizes to the ER and reduces oxidative stress and lipid peroxidation in an SCR20-dependent manner.

Neurons expressing SCR20-containing constructs of CFH showed significantly lower levels of ROS and lipid peroxidation after glutamate, MDA or RSL3 exposure, whereas neurons expressing SCR20-lacking constructs exhibited markedly higher levels (Fig. 3j–q). CFH has been identified as a binding partner of MDA-modified proteins in the plasma^40,43, so we examined whether it also associates with them intracellularly under oxidative stress. Co-localization of CFH with MDA-modified proteins increased significantly after glutamate, MDA or RSL3 treatment, for both endogenous and lentivirally delivered CFH (Extended Data Fig. 8a–d). In post-mortem human retina, MDA-modified proteins increased significantly in MS (Extended Data Fig. 8e) with enhanced CFH–MDA co-localization (Extended Data Fig. 8f). CFH also overlapped strongly with the lipid-peroxidation marker BODIPY C11 after glutamate and RSL3 treatment (Extended Data Fig. 8g,h), and constructs containing SCR20 accumulated significantly fewer MDA-modified proteins (Extended Data Fig. 8i,j). To further validate this spatial association, proximity ligation assays using the CFH variants, which were reliably detected with an antibody directed against CFH (Supplementary Figs. 7–9), showed no proximity under control conditions (Supplementary Fig. 10a) but a significant increase after glutamate treatment (Extended Data Fig. 8k). A similar increase was observed in the GFP controls, consistent with the upregulation of endogenous CFH, whereas no signal was detected in Cfh-cKO cultures (Supplementary Fig. 10b).

We next examined whether CFH-mediated protection extended to other cell types and cell death modalities. In Pfa1 fibroblasts, MDA or RSL3 reduced viability and increased lipid peroxidation, both rescued by the radical-trapping agent liproxstatin-1 (Supplementary Fig. 11a,b). SCR20-containing CFH constructs conferred robust protection, preserving viability and markedly reducing lipid peroxidation after RSL3 treatment (Supplementary Fig. 11c,d). Protection was ferroptosis specific, as MDA-modified proteins increased only after RSL3-induced ferroptosis, not after apoptosis, necroptosis or proteotoxic stress (Supplementary Fig. 11e), and neither CFH nor mHDM-FH protected against these non-ferroptotic pathways (Supplementary Fig. 11f). Consistently, while ferrostatin-1 and liproxstatin-1 alone conferred some protection against glutamate-induced stress, their combination with CFH or mHDM-FH expression added no further benefit, probably because CFH already exerted substantial protection (Supplementary Fig. 11g,h). To test SCR20 dependence in vivo, AAVs encoding mHDM-FH, the truncated mHDM-FH(Δ20) (lacking SCR20) or GFP under the hSYN1 promoter were delivered 3 weeks before EAE (Fig. 4a and Extended Data Fig. 9a–d). mHDM-FH, but not mHDM-FH(Δ20), conferred significant neuroprotection, preserving ventral horn neuronal counts, reducing MDA levels and significantly protecting transduced RGCs (Fig. 4b–d). Thus, the C-terminal SCR20 domain is essential for neuroprotection against oxidative stress in neuronal and non-neuronal cells.

Despite the established extracellular functions of CFH, including binding MDA modifications on cell surfaces⁴⁰ and regulating complement^39,43, our data support an intracellular mechanism in neurons. We first confirmed that CFH was secreted into the supernatant after lentiviral transduction (Extended Data Fig. 9e). To test whether extracellular CFH contributes to protection, we performed a medium-exchange experiment; primary cultures expressing CFH or an mScarlet control construct were incubated with swapped medium before stimulation with glutamate, MDA or RSL3. Applying medium from CFH-expressing cultures to mScarlet cultures did not improve viability, ROS or lipid peroxidation, whereas CFH-expressing cultures retained protection when given mScarlet medium (Extended Data Fig. 9f–h). Similarly, a C-terminal CFH-blocking antibody⁴⁴ did not diminish protection of viability or ROS production after stimulation with glutamate, MDA or RSL3 (Extended Data Fig. 9i,j), further supporting an intracellular mechanism independent of extracellular C-terminal interactions.

To confirm that intracellular CFH is sufficient for neuroprotection, we generated signal-peptide-deficient variants (Fig. 3a and Supplementary Figs. 7e,i, 8c,f and 9c,f) that were undetectable in the supernatant by ELISA (Extended Data Fig. 9k). Intracellular mHDM-FH(ΔSP) protected against cell death, ROS production, lipid peroxidation and accumulation of MDA-modified proteins, whereas mHDM-FH(Δ20ΔSP) did not (Fig. 4e–h). Likewise, CFH(ΔSP) was as neuroprotective as CFH (Extended Data Fig. 9l–o). We next engineered KDEL-tagged constructs to retain CFH and mHDM-FH in the ER (Fig. 3a and Supplementary Figs. 7c,g, 8b,e and 9b,e); ELISA confirmed significantly reduced secretion (Extended Data Fig. 9p) and these ER-retained constructs also robustly reduced cell death, ROS and lipid peroxidation after glutamate, MDA and RSL3 (Extended Data Fig. 9q–s). To understand how differently localized variants confer protection, we mapped the subcellular distribution of lipid peroxidation. Consistent with reports implicating the ER and lysosomes as primary ferroptotic sites^45,46,47, lipid peroxidation localized early to lysosomes and the ER in Pfa1 cells after RSL3 (Supplementary Fig. 12a–f) and predominantly to the ER, followed by lysosomes, mitochondria, other membranes and cytosol, in neurons (Supplementary Fig. 13a–f). These findings support ER involvement in neuronal ferroptosis and are consistent with protection by ER-retained CFH and mHDM-FH variants.

Using the retina as a discovery platform, we leveraged RGC heterogeneity to identify neuron-intrinsic determinants of resilience to inflammation (a graphical summary is shown in Supplementary Fig. 17). In the retina, light exposure generates free radicals, subjecting neurons to constant oxidative stress and making it an ideal environment for studying intrinsic neuronal defence mechanisms against oxidative damage, a hallmark of neurodegeneration in MS and other neurological diseases¹. Midget RGCs, the retinal neurons of the parvocellular colour-vision pathway, were most vulnerable in MS, consistent with dyschromatopsia in optic neuritis and reports of preferential parvocellular degeneration after optic neuritis^51,52. Melanopsin-expressing ipRGCs were also susceptible, echoing clinical evidence of early ipRGC loss and their vulnerability in Alzheimer’s disease^53,54.

Among candidate genes, CFH correlated with RGC survival and was induced in MS. Notably, CFH was absent at the baseline in cortical and spinal motor neurons but was induced in MS and EAE, suggesting that constantly oxidatively stressed RGCs constitutively deploy a defence that other CNS neurons activate only under inflammation, representing a broadly conserved neuronal strategy. The identified intracellular function of CFH is distinct from its canonical secreted role and may reflect a distinct splice variant or conformation⁵⁵. Consistent with this, the SCR7 region, which contains the AMD-associated Y402H variant, was dispensable for neuroprotection against oxidative stress^40,56, whereas SCR20 was essential. Mechanistically, inflammatory (IFNγ) and oxidative-stress-induced neuronal CFH interrupted a self-perpetuating cycle of ROS accumulation, lipid peroxidation and ferroptosis in an SCR20-dependent, C3-independent manner. Protection was maintained when CFH was retained in the ER or cytoplasm, probably by sequestration of locally generated MDA-modified proteins, consistent with the spatial association between protective CFH constructs and MDA-modified proteins. Protection extended from RGCs to cortical and spinal motor neurons, guarding against both visual and motor deficits. While the lipid-peroxidation-dependent cell death process ferroptosis is a key driver of neuronal loss in MS^36,37,38, our data position the ER as a prominent site of neuronal lipid peroxidation, as recently suggested in other cell types^{35,45,46,47,57,58}. We identify intracellular CFH as a key regulator of the protective neuronal inflammatory stress response and a potential counterbalance to STING-driven GPX4 degradation^37,59, positioning enhancement of intracellular CFH as a promising therapeutic strategy for inflammation-induced neurodegeneration.

Human tissue

Post-mortem human tissue samples

Eyeballs from individuals diagnosed with MS were obtained from the Netherlands Brain Bank, and those from control donors without recognizable neuropathological changes were obtained from Johns Hopkins University. Detailed information about donors used for snRNA-seq, immunohistochemistry, qPCR and smFiSH are provided in Supplementary Table 1. The macula region of these donors was used for snRNA-seq while smFISH was performed on tissue adjacent to the macula. qPCR and immunohistochemistry were conducted with the peripheral retina. Paraffin-embedded cortex tissue for histopathology was obtained from the UK Multiple Sclerosis Tissue Bank at Imperial College London. Samples were classified as chronic active lesions or normal appearing grey matter according to the histopathological assessment provided by the UK Multiple Sclerosis Tissue Bank’s histopathology reports. Although no statistical methods were used to predetermine sample size, our sample size is comparable to those reported in previous studies^12,13,60.

Nucleus isolation and library preparation

Eyeballs were enucleated from deceased healthy controls and people with MS. Only tissue from patients with a post-mortem interval ≤24 h was used. To isolate the macula region, fresh-frozen eyeballs were positioned in a CM3050 S cryostat (Leica Microsystems) at –20 °C, with the lens facing downward and the optic disc oriented toward the examiner. The macula was identified as a distinct yellow spot on the retina, and the tissue was stored at −80 °C until use. All of the subsequent steps were performed on ice according to a previously published protocol with minor adaptations. In brief, on each experimental day, the frozen tissue was transferred to ice-cold NP40 lysis buffer (0.1% NP-40 (Thermo Fisher Scientific, 85124), 10 mM Tris pH 8.0, 1 mM CaCl₂, 8 mM MgCl₂, 15 mM NaCl, 0.02 U µl^–1 DNase I (Merck Millipore, D4527)). The retina was transferred to a Dounce homogenizer in 1 ml lysis buffer supplemented with 0.2 U µl⁻¹ Ribolock RNase Inhibitor (Thermo Fisher Scientific, EO0382) and homogenized 20 times with both loose and tight pestles. The homogenate was passed through a 100 µm cell strainer and centrifuged at 500g for 5 min. All buffers, except for the washing buffer, were supplemented with 0.16 U µl⁻¹ Ribolock RNase inhibitor. Pelleted nuclei were resuspended in staining buffer (Tris base buffer: 10 mM Tris pH 8.0, 1 mM CaCl₂, 8 mM MgCl₂, 15 mM NaCl, 1 U ml⁻¹ DNase I) containing 0.02% Tween-20 and 2% BSA, with primary antibodies against NeuN (1:250, Merck Millipore, ABN91) and RBPMS (1:250, Abcam, ab194213) for 15 min at 4 °C. Nuclei were washed, centrifuged at 500g for 5 min and stained with secondary antibodies (anti-chicken 647, 1:500; Jackson ImmunoResearch, 703-605-155; and anti-rabbit PE, 1:200, BioLegend, Poly4064) for 15 min at 4 °C. After another wash step, nuclei were filtered through a 70 µm strainer, resuspended in sorting buffer (Tris base buffer with 2% BSA), and Hoechst (1:2,000) was added to visualize nuclei. NeuN⁺RBPMS⁺ nuclei were sorted using the BD FACS Aria III device running BD FACSDiva v.9.0.1 into Ames medium (Sigma-Aldrich, A1420) with 1.5% BSA. Sorted NeuN⁺RBPMS⁺ nuclei were pelleted at 500g for 5 min at 4 °C, resuspended in approximately 20 µl of 1% BSA in PBS, visually inspected, counted in a Neubauer chamber and adjusted to a concentration of around 1,000 nuclei per µl. Nuclei were then loaded onto the 10x Chromium Single Cell Chip G (10x Genomics) with a targeted recovery of about 12,000 nuclei per channel. Libraries were generated according to the manufacturer’s protocol using the Chromium Single Cell 3′ Reagent Kit version 3.1 (dual index), measured using the Agilent Bioanalyzer (TapeStation 4150) and sequenced on the Illumina NovaSeq 6000 platform (paired-end), aiming for a sequencing depth of around 30,000 reads per nucleus.

Data preprocessing and quality control

Count matrices for each sample were generated by aligning each library to the human reference mRNA transcriptome GRCh38-2020-A using Cell Ranger (v.7.0.1), including both exonic and intronic reads⁶¹. The Cell Ranger output was processed using CellBender v.0.3.0 with the default settings (epochs = 150, fpr = 0.01, learning rate = 10⁻⁴) to remove ambient RNA and other background noise⁶². Downstream analysis was performed with Seurat (v.5)⁶³ in R Studio (R v.4.4.1). For each RNA count matrix, the following steps were carried out: cell counts were normalized to a total library size of 10,000 and log transformed (Seurat, NormalizeData). The top 2,000 variable features were identified (Seurat, FindVariableFeatures), followed by data scaling (Seurat, ScaleData) and dimensionality reduction (Seurat, RunPCA, npcs = 50). RPCA integration was performed to integrate the principal components (Seurat, IntegrateLayers) and used as input for k nearest-neighbour graph construction (Seurat, FindNeighbors, dims = 50) and Leiden clustering with a resolution of 2.5 (Seurat, FindClusters), initially deliberately overclustering the dataset. An initial dataset of 351,737 nuclei was subjected to quality control. Potential doublets were identified using the scDblFinder⁶⁴ package. Given the defined cluster structure in our dataset, we used a cluster-based approach for doublet identification, estimating the standard 10x doublet rate of 0.8% per 1,000 nuclei. To remove non-RGC nuclei, we filtered the dataset for clusters with high expression of RBPMS, the main RGC marker gene. After subsampling, nuclei with abnormally high (greater than mean + 3 s.d.) gene counts, fewer than 2,300 gene counts, a doublet score of >0.5, as well as mitochondrial counts of >5% were removed.

Clustering and cell type annotation

After removing low-quality and non-RGC nuclei from the dataset, we repeated the normalization and clustering pipeline at a resolution of 0.5 (RunPCA, npcs = 30; FindNeighbors, dims = 30). For each cluster, we calculated differentially expressed marker genes compared with every other cluster (Seurat, FindMarkers). Clusters were merged if ≤5 differentially expressed genes were found between them, with an average log₂-transformed fold change of >2 or <–2 and a P value < 0.05. Moreover, if a cluster contained fewer than 200 nuclei in the control condition or was absent in two or more samples, it was fused with its nearest neighbour based on a Euclidean distance matrix constructed in gene expression space (Seurat, BuildClusterTree; Supplementary Fig. 1e). The remaining 27 clusters were annotated manually based on their respective marker gene expression, using three published healthy human RGC datasets as reference^8,9,10. To avoid disease-related transcriptional changes influencing cell type annotation, marker genes were identified exclusively from healthy control samples using Seurat’s FindAllMarkers function. Of the top 30 markers of each cluster, 2–3 are shown in Extended Data Fig. 1e. For each reference dataset, the top 50 marker genes per cell type were selected and used for gene set enrichment analysis (GSEA) through the ClusterProfiler⁶⁵ package, applying the ranked cluster markers from the healthy control samples of our dataset. For the enrichment analysis in Extended Data Fig. 1c and the cell type marker identification in Extended Data Fig. 1e, the three midget OFF RGC subtypes (MG-OFF1, MG-OFF2, MG-OFF3) and the four midget ON RGC subtypes (MG-ON1, MG-ON2, MG-ON3, MG-ON4) were combined into single midget OFF and midget ON types. This consolidation streamlined marker gene identification and cell type annotation. The merged midget ON and midget OFF RGC clusters were also used for the UMAP representation in Fig. 1c.

Differential gene expression analysis

To analyse the transcriptional response of all RGCs during MS, we performed sample- and condition-wise pseudobulk aggregation of unnormalized counts (Seurat, aggregateExpression). The DESeq2 package⁶⁶ (v.1.44) was used for normalization and differential gene expression analysis, considering genes with false discovery rate adjusted P value < 0.05 as differentially expressed. Enrichment analysis of biological themes was performed with the ClusterProfiler⁶⁵ (v4.12.6) package.

Analysis of resilient and susceptible RGC subtypes

To assess differences in RGC vulnerability among cell types in MS, we calculated the relative frequency of each cell type in both control and MS RGC populations. Outliers were identified and excluded using Grubbs’ test. Significant differences in the relative frequencies between control and MS samples were assessed using Mann–Whitney U-tests. Clusters with significantly lower relative frequencies in MS compared with controls (P < 0.05, log₂-transformed fold change < 0) were defined as susceptible. Clusters with a significantly higher relative frequency in MS compared with controls (P < 0.05, log₂-transformed fold change > 0) were defined as resilient. Clusters falling into either term but not reaching significance (P > 0.05) were classified as intermediate susceptible (log₂-transformed fold change < 0) or intermediate-resilient (log₂-transformed fold change > 0). We performed sample- and condition-wise pseudobulk aggregation of significantly de-enriched (MG-ON4, MG-OFF2, ipRGC) and significantly enriched clusters (PG-ON, RGC26, RGC13) to compare transcriptional profiles between susceptible and resilient clusters. DESeq2 analysis was conducted to determine differentially expressed genes between resilient MS, resilient control, susceptible MS and susceptible control pseudobulks. Gene regulation modules were defined as follows: module A, genes upregulated in resilient RGCs compared to susceptible RGCs; module B, genes upregulated in susceptible RGCs compared with resilient RGCs; module C, genes upregulated in resilient RGCs during MS; module D, genes downregulated in resilient RGCs during MS; module E, genes upregulated in susceptible RGCs during MS; module F, genes downregulated in susceptible RGCs during MS. Gene Ontology analysis with the genes in each module was performed using the clusterProfiler⁶⁵ package. To calculate resilience and vulnerability scores for the susceptible, intermediate susceptible, intermediate resilient and resilient pseudobulks of each donor and condition, we applied the AUCell algorithm⁶⁷, using the top 200 module-defining genes from modules A–E as input, after pseudobulk aggregation and normalization of gene expression. To calculate intrinsic resilience signatures for each pseudobulked RGC subtype and donor, we used the top 200 genes of module A (intrinsic resilience signature) and module B (intrinsic susceptibility signature) as the input for AUCell analysis, after aggregation and normalization of gene expression by donor and cell type.

Linear regression analysis of intrinsic susceptibility and resilience genes

To identify genes of which the expression correlated with resilience, we extended the analysis to include all RGC types rather than focusing solely on the most resilient or susceptible clusters. Clusters were ranked from most susceptible to most resilient based on the signed negative logarithmic P value of their relative frequency in MS compared with controls. We focused on the ten samples from control individuals. To minimize biases arising from differences in cluster size across RGC types, we standardized cluster sizes by bootstrapping the dataset, randomly selecting 100 nuclei from each cluster within each sample. Clusters containing fewer than 10 nuclei were excluded. For clusters with fewer than 100 nuclei, sampling was performed with replacement; for larger clusters, it was performed without replacement. This process was repeated five times, and pseudobulk count vectors were summed across iterations. The DESeq2 package (v.1.44) was used to normalize gene counts of the size-adjusted pseudobulks, accounting for differences in sequencing depth and library size across samples. Variance stabilization was performed using the rlog function within DEseq2. For each gene, a linear regression was performed using the lm function in R (formula: normalized expression ~ resilience), modelling the normalized gene expression in a given cluster and sample as a function of the cluster resilience score, as defined previously by the signed negative logarithmic P value of a cluster’s relative frequency in MS compared with controls. P values for significance were adjusted for multiple comparisons using the FDR method.

Single-molecule fluorescence multiplex in situ RNA hybridization

smFISH was performed on representative post-mortem, unfixed MS and control cryosections (16 µm thickness) using the RNAscope Multiplex Fluorescent Detection Kit v2 (ACD Biotechne, 323100) according to the manufacturer’s protocol. The following human RNAscope assay probes were used (ACD Biotechne): POU4F1 (438441), TBR1 (425571-C2), FOXP2 (407261-C2), GPR149 (1040901-C3) and CFH (428731-C3). Probes were labelled with TSA Vivid Fluorophores (fluorescein, cyanine 3, cyanine 5, Akoya Biosciences). Each run included quality-control slides stained with the human 3-plex RNAscope positive-control probe (320861) and 3-plex negative-control probe (320871) from ACD Biotechne.

Image acquisition and quantification of fluorescence multiplex in situ RNA hybridization

Multiplex fluorescence images were acquired using the Leica DM6 B microscope equipped with a Leica K5 camera and the THUNDER imaging system for preprocessing. Images were captured at ×20 magnification as z stacks to ensure thorough signal capture within tissue sections. Acquisition was performed with LAS-X software, and images were exported as LIF files for subsequent analysis with QuPath⁶⁸ (v.0.4.3). In QuPath, analysis was performed using the subcellular spot detection tool. The RGC layer was first selected, followed by cell detection based on the DAPI channel. Subcellular spot detection was then conducted for POU4F1, FOXP2, TBR1, CFH and GPR149. Double-positive cells (such as POU4F1-positive cells co-expressing specific subpopulation markers) were identified using an initial single-measurement classifier, followed by a composite classifier.

RT–qPCR

We reverse-transcribed RNA to complementary DNA using the RevertAid H Minus First Strand cDNA Synthesis Kit (Thermo Fisher Scientific, K1632) according to the manufacturer’s instructions. Samples were measured on the QuantStudio Flex Real-Time PCR System using TaqMan Gene Expression Assays (Thermo Fisher Scientific) for Cfh (mouse, Mm01299248_m1), RBPMS (human, Hs01060992_m1) and RCVRN (human, Hs00610056_m1). All analyses were performed in triplicate. We calculated gene expression as \({2}^{-\varDelta {C}_{{\rm{t}}}}\) relative to Tbp (mouse, Mm01277042_m1) or GAPDH (human, Hs99999905_m1) as the endogenous control.

Human histopathology of post-mortem tissues

Tissue paraffin sections were cut at 3 μm and stained with haematoxylin and eosin (H&E) according to standard procedures. For immunohistochemical staining, brain tissue sections were processed as follows. After dewaxing and inactivation of endogenous peroxidases (3% hydrogen peroxide), antibody-specific antigen retrieval was performed using the Ventana BenchMark XT autostainer. The sections were blocked with rabbit serum and then incubated with the primary antibody against human CFH (1:40; R&D, AF4779). For detection of specific binding, the anti-goat Histofine Simple Stain MAX PO immune-enzyme polymer (Nichirei, 414161F) was used as the secondary antibody. For DAB staining, the UltraView Universal DAB Detection Kit (Roche, 760-500) was used. Counterstaining and bluing were performed using haematoxylin (Ventana Roche, 760-2021) and Bluing Reagent (Ventana Roche, 760-2037) for 4 min. Subsequently, stained sections were mounted in mounting medium. For quantification, DAB-positive neurons were counted manually per area of cortical tissue within ImageJ (Fiji, v.2.14.0). Neurons were identified on the basis of their characteristic morphology in H&E-stained sections. Information about donors used for histopathology is provided in Supplementary Table 2.

Animal models

Mice

All mice, including C57BL/6J WT (The Jackson Laboratory), SJL/JRj WT (Janvier Labs), C3 knockout (B6.129S4-C3^tm1Crr/J, The Jackson Laboratory) and Cfh^flox/flox mice were maintained under specific pathogen-free conditions in the central animal facility of the University Medical Center Hamburg-Eppendorf (UKE). Adult mice aged 8–20 weeks were used for experiments. For EAE experiments involving C57BL/6J WT, SJL/JRj WT and C3-KO mice, only female animals were included. To assess potential sex dependent effects, an additional cohort of male C57BL/6J mice was used to compare neuronal mHDM-FH expression with a GFP control. For experiments with Cfh^flox/flox mice both sexes were used. The mice were kept under a 12 h light–12 h dark cycle at 22 ± 2 °C and 40–60% relative humidity with ad libitum access to standard chow (Altromin, 1328P) and water. EAE mice additionally received DietGel Recovery (Ssniff, H007-72065). For EAE experiments, if both sexes were included, equal numbers of male and female mice were allocated to each experimental group. Furthermore, animals from each litter were evenly distributed across groups to minimize potential litter effects. Within these constraints, mice were randomly assigned to receive either effector-AAVs or control AAVs. Sample sizes were selected based on the principles of the 3Rs (replacement, reduction and refinement), the use of genetically homogeneous inbred mouse strains and sample sizes commonly used in previous studies employing the EAE model that detected comparable biological effects^25,37,38. Clinical EAE scoring was conducted in a blinded manner, with researchers unaware of both the genotype and the injected AAV to prevent observer bias.

Generation of Cfh neuronal knockout mice

Cfh^{tm1a(EUCOMM)Wtsi} sperm (The Jackson Laboratory) were revitalized on C57BL/6 wild-type female mice (The Jackson Laboratory), resulting in litters positive for the Cfh^{tm1a(EUCOMM)Wtsi} allele. To obtain the conditional (tm1c) allele (Cfh-flox), these animals were bred with ACTB:FLPe B6J mice expressing the Flp recombinase (The Jackson laboratory, 005703)⁶⁹. To generate neuronal Cfh-KO mice, we performed retrobulbar intravenous injections of an AAV (serotype PhP.eB) carrying cre recombinase under the control of the hSYN1 promoter in Cfh^flox/flox mice.

EAE model

For experiments involving AAV-mediated expression, mice received retrobulbar injections of 100 μl of the respective AAV (serotype PhP.eB) at a titre of 3 × 10¹¹ viral genomes in PBS 3 weeks before EAE induction. Mice were monitored daily and weighed starting 1 week after the injection. Mice injected with effector constructs were compared to littermates receiving GFP-only control AAVs. To induce EAE in C57BL/6 mice, they were immunized subcutaneously with 200 μg MOG_35–55 peptide (Schafer-N) emulsified in Complete Freund’s adjuvant (CFA) (Difco, DF0639-60-6) containing 2 mg ml⁻¹ Mycobacterium tuberculosis (Difco, DF3114-33-8). Moreover, 300 ng of pertussis toxin (Calbiochem, CAS70323-44-3, 516562) was administered intraperitoneally on the day of immunization and again 48 h later. For experiments in SJL mice, EAE was induced by subcutaneous immunization with 100 µl of PLP_139–151 peptide (1.5 mg ml⁻¹) emulsified 1:1 with 100 µl of CFA (Difco, DF0639-60-6) containing M. tuberculosis H37RA (1 mg ml⁻¹, 100 µg per mouse). Clinical signs were assessed daily using the following scoring system: 0, no clinical deficits; 1, tail weakness; 2, hind limb paresis; 3, partial hind limb paralysis; 3.5, full hind limb paralysis; 4, full hind limb paralysis with forelimb paresis; and 5, premorbid state or death. Animals reaching a clinical score of ≥4 were euthanized in accordance with the regulations of the German Animal Welfare Act. All EAE experiments were conducted with investigators blinded to the genotype and treatment conditions.

Visual assessment in mice

To evaluate visual acuity and contrast sensitivity in EAE mice, automated optomotor testing was performed using the OptoDrum system (Striatech)^70,71. Mice were placed individually onto an elevated central platform surrounded by computer monitors and monitored from above by a camera. The optomotor reflex, a head-tracking movement elicited by rotating visual stimuli, was induced by a moving vertical black-and-white grating displayed on the surrounding screens. Visual acuity was determined by progressively increasing the spatial frequency (that is, the number of stripes per degree of visual angle) until the optomotor reflex could no longer be elicited. Contrast sensitivity was assessed by gradually reducing the contrast of the grating until the reflex was no longer observed. For each stimulus condition (spatial frequency or contrast level), two successful responses were required to advance to the next step, whereas three consecutive failures were defined as an unsuccessful trial.

Analysis of published EAE sequencing datasets

Raw count matrices and metadata from publicly available RNA-seq datasets were retrieved from the Gene Expression Omnibus (GEO) repository. The GSE118948 dataset includes single-cell sequencing data of CD45⁺ cells isolated from the spinal cords of EAE mice 15 days after immunization. Analysis of the single-cell data was performed using Seurat, with cell type annotations from the original publication²². Differential expression analysis was conducted with the FindMarkers function within Seurat using a Wilcoxon rank sum test to compare gene expression in healthy mice and those with acute EAE across T cells, neutrophils, dendritic cells, macrophages, monocytes, CNS-associated monocytes and microglia. Bulk RNA-seq data from GEO GSE194071 (ref. ²³) were used to examine spinal cord microglia in acute and chronic recovery EAE mice, while GSE100329 (ref. ²⁴) provided data for analysing spinal cord astrocytes in acute and chronic progressive EAE mice. Moreover, GSE104899 (ref. ²⁵) and GSE279707 (ref. ²⁶) were used to study spinal cord motor neurons in acute EAE mice. All bulk sequencing data were analysed using DESeq2 (v.1.44), and signed negative logarithmic FDR-adjusted P values were calculated for visualization.

Immunohistochemistry of spinal cord tissue

Mouse spinal cord tissue was obtained and processed as described previously⁷². Images were acquired using a Zeiss LSM 900 Airyscan 2 laser-scanning confocal microscope equipped with ZEN blue software v.3.9. Antibodies and their dilutions used for immunohistochemistry are specified in Supplementary Table 3.

Immunohistochemistry analysis of mouse retinal whole mounts

After transcardial perfusion of mice with ice-cold 4% PFA, the orientation of the eye was marked in situ, and eyes were enucleated and post-fixed in 4% PFA for 1 h. The eyeballs were then transferred to ice-cold PBS. Whole-mount dissection of retinas was performed under a dissection microscope. For immunohistochemistry, retinas were washed in PBS 0.5% Triton X-100 and permeabilized by freezing them at −70 °C for 15 min. After thawing, the retinas were rinsed with PBS 0.5% Triton X-100 and incubated overnight with primary antibodies at 4 °C in blocking buffer (PBS with 2% Triton X-100 and 2% normal donkey serum). The retinas were subsequently washed in PBS 0.5% Triton X-100 and incubated with secondary antibodies in PBS with 2% Triton X-100 for 2 h at room temperature. The retinas were washed with PBS and mounted vitreal side up. All antibodies used are listed in Supplementary Table 3. Retinal whole mounts were imaged using the Zeiss Axiocam 705 mono microscope. Semi-automated quantification of GFP-positive RGCs was performed within Fiji (ImageJ, v.2.14.0). Background subtraction was applied to correct for uneven illumination and contrast enhancement was performed to optimize illumination. Thresholding used the default autothreshold method and the image was then converted into a binary mask. To reduce noise, the Despeckle function was applied. Segmentation was refined with the Watershed algorithm to separate cells. Particle analysis was conducted on segmented objects with inclusion of particles with a size of 30–300 pixels and circularity range of 0.4–1.0.

Luxol fast blue staining

Myelination in EAE mice was assessed by Luxol Fast Blue staining in transverse sections of the dorsal columns in the cervical, thoracic and lumbar spinal cord as previously described⁷³. The slides were imaged on the NanoZoomer 2.0-RS digital slide scanner with NDP.view2 software (Hamamatsu). A customized counting mask within Fiji (ImageJ, v.2.14.0) was created to quantify the area of Luxol Fast Blue-positive axons in the white matter of the spinal cord. The Luxol Fast Blue-positive area was normalized to the total analysed area within the dorsal column.

Immunoblot

Spinal cords of healthy and EAE mice were homogenized in 2 ml radioimmunoprecipitation buffer (50 mM Tris, 150 mM NaCl, 0.5 mM EDTA, 10% SDS, 1% NP-40, 10% sodium deoxycholate, with protease and phosphate inhibitor cocktails (cOmplete, Sigma-Aldrich, 11836170001)) using a tissue grinder and incubated at 4 °C for 30 min on a rotating wheel. The lysates were centrifuged for 5 min to remove the cell debris. Protein concentrations were determined by BCA assay (Pierce BCA Protein Assay Kit, Thermo Fisher Scientific, 23228 and 23224) according to the manufacturer’s protocol, and 25 µg of protein were loaded onto SDS–PAGE (NuPAGE, Thermo Fisher Scientific, NW04125BOX) followed by wet transfer to polyvinylidene fluoride membranes. Blocking was performed using 5% BSA for 1 h at room temperature. Membranes were incubated with primary antibodies overnight at 4 °C. Horseradish-peroxidase-coupled secondary antibodies were applied for 1 h at room temperature, and chemiluminescence was visualized using WesternSure PREMIUM Chemiluminescent Substrate (LI-COR, 926-95000) according to the manufacturer’s protocol. A list of all of the antibodies used is provided in Supplementary Table 3. All uncropped western blot membranes are provided in Supplementary Fig. 18.

Immune cell isolation for flow cytometry

Spinal cord tissue was collected after transcardial PBS perfusion and dissociated into single-cell suspensions using 1 mg ml⁻¹ collagenase A (Roche, 11088793001) and 0.1 mg ml⁻¹ DNase I from bovine pancreas (Merck Millipore, 260913) with the gentleMACS Octo Dissociator (Miltenyi Biotec, program: Multi_F). The dissociated tissue was passed through a 70 µm cell strainer, and immune and glial cells were enriched using a discontinuous density gradient. After centrifugation at 2,500 rpm, 4 °C for 30 min, cells were collected from the interphase between the 30% Percoll and 70% Percoll layers. Non-specific Fc-receptor-mediated antibody binding was blocked by pre-incubation with TruStain FcX anti-mouse CD16/32 antibody (BioLegend, 101320) for 10 min at 4 °C, followed by surface antibody staining in Brilliant Stain Buffer (BD Biosciences, 563794) for 30 min at 4 °C. Dead cells were excluded from analysis using Zombie Green Fixable Viability Stain (BioLegend, 423112). For intracellular staining of CD206 and CD68, cell suspensions were fixed for 20 min at room temperature using fixation buffer (BioLegend, 420801), followed by 20 min incubation with anti-CD68 antibody in intracellular staining permeabilization wash buffer (BioLegend, 421002). Absolute cell numbers for CD45^high leukocytes and CD45^med microglia were determined using Precision Count Beads (BioLegend, 424902). We obtained data using the BD Symphony A3 flow cytometer (BD Biosciences) and analysed data using FlowJo v.10.9 (BD Biosciences).

Viral vectors

Vector construction

All primers and oligonucleotides used in this study are listed in Supplementary Table 4. For expression of synthetic minimal CFH-derived constructs, clonal genes were synthesized by Twist Bioscience (Supplementary Table 4 (gene synthesis tab)). In mHDM-FH, the N-terminal SCR1–5 are linked to C-terminal SCR18–20 through an SGSG linker. Amino acids 24–152 of mouse CFHR-1 were used as a dimerization unit to increase protein half-life. The amino acid sequence of mHDM-FH was derived from a previous study³¹. CR2-FH is a fusion protein consisting of a complement receptor 2 fragment linked to the N-terminal SCR domains 1–5 of CFH. The amino acid sequence of CR2-FH contains the signal peptide of mouse IgG κ light chain for optimal secretion, amino acid residues 1–257 of mature mouse CR2, a (G₄S)₂ linker, followed by the 5 N-terminal SCR domains of mouse factor H³³. Both sequences were codon-optimized for expression in mouse cells. A Pfl23II restriction-enzyme-recognition site was attached to the 5′ end, and BamHI and SalI sites were added to the 3′ end. Clonal genes were digested with Pfl23II and SalI for ligation into a customized pAAV backbone with a hSYN1 promoter derived from pAAV-hSYN1-mTurquoise2 (gift from V. Gradinaru, Addgene 99125), an eGFP:KASH cassette (AAV:ITR-U6-sgRNA(backbone)-hSyn-cre-2A-EGFP-KASH-WPRE-shortPA-ITR, gift from Feng Zhang, Addgene 60231) and a P2A-T2A cleavage peptide. A control plasmid containing a stop codon was cloned downstream of the P2A-T2A cleavage peptide. For lentiviral expression in primary neuronal cultures, clonal genes were digested with Pfl23II and BamHI for ligation into a custom lentiviral backbone with an hSYN1 promoter, a mScarlet cassette and a P2A cleavage peptide. For lentiviral expression of full-length mouse CFH, two separate PCR fragments were generated and linked with a triple-ligation strategy. The full-length Cfh sequence was amplified from mouse kidney cDNA with primers containing Pfl23II and XbaI, or XbaI and SgsI restriction sites, respectively. The inserts were ligated into the same lentiviral backbone described above. These constructs were subsequently used as templates to generate different mHDM-FH and CFH variants using the primers in Supplementary Table 4. To create constructs lacking the signal peptide, forward primers excluding the first 54 base pairs (P15–P18) were used. To exclude SCR20, we used reverse primers complementary to SCR19 (P10–P12) and inserted the construct upstream of the dimerization domain. The 2×KDEL sequences for retention in the secretory pathway were included in the reverse primers (P19–P20). For expressing the SCR20 alone (P13–P14), SCR20 was added downstream of an Lck-transmembrane domain (P8–P9) for intracellular retention at the cell membrane. The Lck domain was amplified from PZac2.1 hSYN1-Lck-13×Linker-BioID2-BioID2-HA (gift from B. Khakh, Addgene 176854). The lentiviral IFNγ overexpression construct was generated as previously described³⁷. We used an AAV carrying cre recombinase under the hSYN1 promoter to generate conditional knockouts in neuronal cultures (pAAV-hSYN1-cre-P2A-dTomato, gift from R. Larsen (Addgene 107738). To generate conditional knockouts in mice, the plasmid pENN.AAV.hSYN.HI.eGFP-cre.WPRE.SV40, a gift from J. M. Wilson (Addgene 105540) was used. For expression of TurboID constructs, the sequence for the biotin ligase was amplified from pAAV-TBG-Cyto-TurboID (gift from J. Long, Addgene 149414) using primers for a C-terminal HA protein tag and inserted into a lentiviral backbone with a hSYN1 promoter using NheI and BamHI (Turbo-Cyto; P21-P22) or Nhel and EcoRI (Turbo-ER; P23-P24). For ER retention of the TurboID-ER construct, we included an N-terminal signal peptide and a C-terminal KDEL sequence, separated by a GGGGS linker. For in vivo delivery, the TurboID-ER cassette was transferred into a pAAV backbone with a hSYN1 promoter using NheI and SgsI (P25–P26). The following constructs were generated and used for in vitro or in vivo experiments in this study: Lenti-hS-mmCfh-P2a-mScarlet, Lenti-hS-mmCfhΔ20-P2a-mScarlet, Lenti-hS-mScarlet-P2a-mmCfh-2×KDEL, Lenti-hS-mmCfhΔSP-P2a-mScarlet, Lenti-hS-mmCfhΔ20ΔSP-P2a-mScarlet, Lenti-hS-mScarlet-P2a-CR2-FH, Lenti-hS-mHDM-FH-P2a-mScarlet, Lenti-hS-mHDM-FHΔ20-P2a-mScarlet, Lenti-hS-mScarlet-P2a-mHDM-FH-2xKDEL, Lenti-hS-mScarlet-P2a-mHDM-FHΔ20-2×KDEL, Lenti-hS-mHDMΔSP-FH-P2a-mScarlet, Lenti-hS-mHDM-FHΔ20ΔSP-P2a-mScarlet, Lenti-hS-Lck-scr20-P2a-mScarlet, Lenti-CMV-mmIfng-P2a-mScarlet, Lenti-hS-mScarlet, Lenti-CMV-mScarlet, Lenti-hS-TurboID-ER, Lenti-hS-TurboID-Cyto, pAAV-hS-eGFP(-KASH)-P2a(-T2a)-mHDM-FH, pAAV-hS-eGFP-P2a-mHDM-FHΔ20, pAAV-hS-eGFP-P2a-mHDM-FHΔSP, pAAV-hS-eGFP-P2a-Lck-scr20, pAAV-hS-eGFP(-KASH), pAAV-hS-eGFP-KASH-P2a-T2a-CR2-FH, pAAV-hS-TurboID-ER. All final products were confirmed by Sanger sequencing.

Lentiviral production

To produce VSV-G-pseudotyped lentiviruses, HEK293T cells were seeded out at 6 × 10⁴ per cm² 1 day before transfection in DMEM containing glutamine and high glucose (Thermo Fisher Scientific), together with helper plasmids pMDLg/pRRE, pRSV-Rev and pMD2.G. pMDLg/pRRE (gift from D. Trono, Addgene 12251, 12253, 12259). For a 10 cm cell culture plate, we used 15 µg transfer plasmid, 10 µg pMDLg/pRRE, 5 µg pRSV-Rev and 2 µg pMD2.G. Plasmids were mixed in 1× HEPES-buffered saline (HBS) and 125 mM CaCl₂ solution and applied to the HEK293T cells in the presence of 25 µM chloroquine diphosphate. After 12 h, the medium was changed and the supernatant was collected at 36 h after transfection, filtered through a 0.45 µm PES filter and immediately snap-frozen and stored at −80 °C. For CFH expression lentivirus production, lentiviral particles were concentrated using the Lenti-X Concentrator (Takara Bio, 631478) according to the manufacturer’s instructions. Primary neurons were transduced at 5–8 days in vitro (d.i.v.) with 70–80% efficiency, visually confirmed by fluorescent protein expression.

AAV production

For vector production, we selected the PhP.eB serotype, which has shown high efficiency in transducing the CNS⁷⁴. AAV particles were produced according to the standard procedures of the UKE vector facility^38,75.

Cell culture

Primary neuronal cultures

For primary cortical cultures, pregnant C57BL/6J or Cfh^flox/flox mice were euthanized, and cortices were isolated and dissociated. Cells were plated at a density of 6 × 10⁴ cells per cm² on poly-d-lysine-coated wells (5 µM, Sigma-Aldrich) and maintained in PNGM medium (Lonza, CC4461) at 37 °C, 5% CO₂ and 98% relative humidity. Experiments were performed using cultures between 14 and 23 d.i.v. To generate Cfh-cKO neurons, neuronal cultures from Cfh^flox/flox mice were transduced at 4 d.i.v. with an AAV PhP.eB carrying cre recombinase (Addgene 107738) under the control of an hSYN1 promoter at a multiplicity of transfection (MOI) of 30,000. The transduction efficiency, estimated visually based on eGFP expression, was 70–80%. Chronic IFNγ exposure was achieved by delivering a lentivirus containing an Ifng expression construct under the control of a CMV promoter at 7 d.i.v.³⁷. Either mScarlet or GFP-only controls were used for all AAV and lentiviral expression experiments.

RNA-seq analysis of neuronal Cfh-cKO cultures

RNA from neuronal Cfh-cKO and control cultures with or without glutamate stimulation were isolated using the RNeasy Mini Kit (Qiagen). RNA-seq libraries were prepared using the TruSeq stranded mRNA Library Prep Kit (Illumina) according to the manufacturer’s instructions. After pooling, libraries were sequenced on the NovaSeq 6000 sequencer (Illumina) generating 150 bp paired-end reads. Reads were aligned to the mouse reference genome (mm10, 2020A) using STAR v.2.5.2b with the default parameters. Overlaps with annotated gene loci were counted using featureCounts v.1.5.1. Differential expression analysis was performed using DESeq2 (v.1.44), with genes showing a FDR-adjusted P < 0.05 considered differentially expressed. Gene lists were annotated using biomaRt (v.2.60.1). We then performed an AUCell (v.1.26) analysis to directly compare a transcriptional excitotoxicity signature between WT control, Cfh-cKO control, and glutamate-stimulated WT and Cfh-cKO neuronal cultures. The transcriptional excitotoxicity signature was defined as all differentially upregulated genes in our glutamate-stimulated WT neuronal cultures in comparison to the WT control condition. Moreover, we performed GSEA with the top 200 glutamate-induced genes in the Cfh-cKO glutamate versus WT glutamate condition.

Pfa1 and HEK293T cell lines

Immortalized mouse fibroblasts (Pfa1 cells) were received from the originating laboratory (Conrad laboratory, Helmholtz Institute)^41,76 and were not further authenticated. HEK293T cells were purchased from an authorized vendor (ACC 635, DMSZ). Pfa1 cells and HEK293T cells were maintained in DMEM high glucose (4.5 g l⁻¹ glucose, 21969-035, Gibco) supplemented with 10% FBS, 2 mM l-glutamine and 1% penicillin–streptomycin at 37 °C with 5% CO₂. Cells were passaged after reaching approximately 80% confluency and were routinely tested for mycoplasma contamination. For experiments, they were seeded at 30,000 cells per cm² 1 day before stimulation. The respective compounds and stimulation times are specified in the corresponding figure legends. All cell lines were regularly checked for mycoplasma contamination using the VenorGeM Advance kit (Minerva biolabs, 11-7024) according to the manufacturer’s instructions. Cell lines were free of mycoplasma contamination.

Cell viability assay

The RealTime-Glo MT Cell Viability Substrate and NanoLuc Enzyme (Promega, G9711) were combined, added to neuronal cultures and incubated for 5 h to allow equilibration of the luminescence signal before treatment application. Toxicity was assessed after exposure to 50 μM glutamate, 100 nM MDA (Sigma-Aldrich, 63287) or 6 µM RSL3 (Sigma-Aldrich, SML2234). Luminescence was recorded immediately before the stressors were applied and 15 h later using the Spark 10 M multimode microplate reader (Tecan) at 37 °C and 5% CO₂. At least four technical replicates were included for each condition; each well’s datapoints were normalized to its final luminescence value before stressor application. Subsequently, the data were normalized to the mean of the control wells at each timepoint to account for well-to-well variability in cell seeding and to the maximal neuronal death (2 mM glutamate) condition. Statistical comparisons were performed using end-point measurements.

For the medium change, CFH antibody and IFNγ experiments, we measured cell viability by quantifying condensed nuclear signal. Neuronal cultures were exposed to 50 µM glutamate, 100 nM MDA or 6 µM RSL3 for 10 h. Where indicated, additional treatments were applied 30 min before the stressors. Hoechst 33342, a cell-permeant nuclear counterstain, was added to the culture wells at a final concentration of 20 µM and incubated for 1 h. After incubation, cells were washed twice with preconditioned medium before imaging. Image acquisition was performed using the Zeiss LSM 900 Airyscan 2 confocal microscope with a ×40 magnification objective. Using Fiji, we generated a mask to capture all nuclei and measured the Hoechst 33342 fluorescence intensity. We then defined condensed nuclei using a threshold corresponding to the mean signal of the control conditions plus 2.5 s.d. This was used to estimate the number of dying and surviving neurons after the respective stimuli were applied.

ROS imaging

ROS levels in response to glutamate, MDA and RSL3-induced stress were measured using CellROX Green Reagent (Thermo Fisher Scientific, C10444). Neuronal cultures were stimulated with 50 μM glutamate for 2 h, 100 nM MDA for 10 h or 6 µM RSL3 for 10 h. CellROX Green was added to each well at a final concentration of 5 μM after 30 min of stressor exposure, alongside Hoechst 33342, as a nuclear counterstain for visualization. After the incubation, cells were washed twice with preconditioned medium to remove excess reagents and imaged using the Zeiss LSM 900 Airyscan 2 confocal microscope at ×20 magnification. Nuclear fluorescence of the CellROX signal was quantified using Fiji (ImageJ, v.2.14.0) to assess ROS levels.

BODIPY C11 imaging

To assess lipid peroxidation, the BODIPY C11 probe (Image-iT, Thermo Fisher Scientific, D3861) was used. Neuronal cultures were exposed to 50 µM glutamate, 100 nM MDA or 6 µM RSL3 for 10 h. Where indicated, additional treatments were applied 30 min before the stressors. BODIPY C11 was added to the culture wells at a final concentration of 20 µM and incubated for 1 h. Simultaneously, Hoechst 33342, a cell-permeant nuclear counterstain, was included. After incubation, cells were washed twice with pre-conditioned medium before imaging. Image acquisition was performed using the Zeiss LSM 900 Airyscan 2 confocal microscope with a ×40 magnification objective. Lipid peroxidation was quantified using Fiji (ImageJ, v.2.14.0) software. In lentiviral expression experiments, the ratio of oxidized to non-oxidized lipids was calculated due to the negligible or undetectable mScarlet signal from lentiviral transduction under the microscopy settings used. Owing to the stronger tdTomato signal emitted after AAV transduction of our pAAV-hSYN1-cre-P2A-dTomato constructs, only the MFI of the oxidized lipids was quantified in these experiments. For time-resolved and subcellularly resolved lipid peroxidation analyses, we used the following live cell imaging dyes: 1 µM MitoTracker Deep Red FM (Thermo Fisher Scientific), 1 µM ER-Tracker Blue-White DPX (Thermo Fisher Scientific), 500 nM SiR-lysosome kit (Cytoskeleton), 10 µg ml⁻¹ Hoechst 33342 (Thermo Fisher Scientific) and 500 nM SiR700 actine kit (Cytoskeleton). Dyes were applied in the indicated concentrations and washed out together with the BODIPY C11 probe. All imaging experiments, except for the ER-Tracker, included Hoechst 33342 staining. The oxidized BODIPY C11 signal in the overlap with the respective cell organelle marker was quantified.

Quantification of complement component levels by ELISA

Peripheral blood was collected from the vena cava at the time of euthanasia. The samples were centrifuged at 1,000g for 10 min at 4 °C, and plasma was stored at −80 °C until use. Spinal cord tissue was collected after perfusion with ice-cold PBS and snap-frozen in liquid nitrogen. Tissues were lysed in protein extraction buffer (100 mM Tris-HCl, pH 7.4, 150 mM NaCl, 1 mM EGTA, 1 mM EDTA, 1% Triton X-100, 0.5% sodium deoxycholate) supplemented with phosphatase and protease inhibitor cocktails and PMSF (final concentration 1 mM) added immediately before use. ELISAs were performed according to the manufacturers’ instructions. CFH levels were determined using the Mouse Complement Factor H ELISA Kit (Hycultec, RD-CFH-Mu); plasma samples were diluted 1:100,000 in PBS, and cell culture supernatants were used undiluted. Total C3 concentrations were measured using the Mouse C3 ELISA Kit (Abcam, ab263884); plasma was diluted 1:100,000 in sample diluent NS, and tissue lysates were diluted 1:1,000 in 1× cell extraction buffer PTR. C3a levels were quantified using the Mouse C3a ELISA Kit (MyBioSource, MBS2701721); plasma was diluted 1:500 in PBS, and tissue lysates were diluted 1:20 in reagent diluent. C5a concentrations were measured using the Mouse C5a DuoSet ELISA Kit (R&D Systems, DY2150); plasma was diluted 1:100 in PBS, and tissue lysates were diluted 1:10 in reagent diluent. To assess complement activation, C3 consumption was calculated by dividing total C3a concentrations by total C3 concentrations for each individual sample (C3a/C3 ratio). Standard curves for all ELISAs were generated using the provided recombinant standards and fitted with a four-parameter logistic regression model.

Immunocytochemistry

Neuronal cultures were grown on 12-mm-diameter coverslips and stimulated with 50 μM glutamate, 100 nM MDA or 6 µM RSL3 for 6 h. After stimulation, cells were fixed in 4% PFA for 15 min and blocked in 10% normal donkey serum containing 0.1% Triton X-100. Immunolabelling was performed using the specified antibodies (details are provided in Supplementary Table 3). Images were captured using the Zeiss LSM900 Airyscan 2 laser-scanning confocal microscope. Co-localization was analysed using Fiji (ImageJ, v.2.14.0). The percentage of overlap with CFH (from total CFH within the cell) was reported.

PLA between CFH and MDA-modified proteins

To investigate the interaction between CFH and MDA-modified proteins, we performed a Duolink in situ proximity ligation assay (PLA) according to the manufacturer’s instructions (Merck). In brief, PLA technology uses pairs of primary antibodies conjugated to complementary oligonucleotides that, when bound within around 40 nm of each other, enable rolling-circle amplification and generate a discrete fluorescence signal after hybridization with labelled probes. For detection of CFH–MDA interactions, a Cfh-KO-validated goat anti-CFH antibody (Quidel, A313, 1:200) was combined with the PLA Goat MINUS probe (Merck, DUO92006), and a rabbit anti-MDA-modified protein antibody (Abcam, ab27642, 1:200) was paired with the PLA Rabbit PLUS probe (Merck, DUO92002). Signal amplification was achieved using the Duolink in situ detection reagent, far-red channel (Merck, DUO92013). To identify neurons, actin co-staining was performed in parallel. Images were acquired using the Zeiss LSM900 Airyscan 2 laser-scanning confocal microscope. Puncta per neuron were counted and quantified.

Proximity labelling with TurboID

Proximity-dependent biotinylation was performed using the biotin ligase TurboID targeted either to the ER lumen or to the cytoplasm. Cytosolic TurboID (TurboID-Cyto) expresses an HA-tagged TurboID biotin ligase under the hSYN1 promoter for cytosolic localization. ER-TurboID contains an HA-tagged TurboID (TurboID-ER) with an N-terminal signal peptide (SP) and a C-terminal KDEL ER-retention motif under the hSYN1 promoter. Cells were transduced with lentiviruses expressing either cytoplasmatic or ER-targeted TurboID at d.i.v. 7. Proximity labelling was induced by supplementing the cell culture medium with 50 µM biotin (IBA Lifescience, 6-6325-001) for 30 min at 37 °C. Cells were then rapidly cooled on ice, washed with ice-cold PBS and lysed in denaturing lysis buffer (8 M urea, 10 mM Tris buffer, 100 mM NaH₂PO₄). The lysates were incubated with streptavidin-coated magnetic beads (Thermo Fisher Scientific, 88817) overnight, washed twice with RIPA buffer, once with 1 M KCl, once with 0.1 M Na₂CO₃, once with 2 M urea in 10 mM Tris-HCl (pH 8.0) and twice again with RIPA buffer. Bound proteins were eluted by boiling in SDS sample buffer at 95 °C for 5 min for immunoblot analysis. For in vivo proximity labelling, mice received retrobulbar injections of 100 μl of an AAV carrying ER-retained TurboID (serotype PhP.eB) at a titre of 3 × 10¹¹ viral genomes in PBS 3 weeks before EAE induction. For biotin pulsing, chow was soaked with 2.5 mg biotin (IBA Lifescience, 6-6325-001) per mouse and day dissolved in water starting 3 days before termination. At the day of perfusion, mice received an i.p. injection of 2.5 mg biotin dissolved in 100 µl PBS in addition. Spinal cord tissue was collected after transcardial perfusion with 10 ml ice-cold PBS. Tissues were homogenized in lysis buffer using a tissue disruptor (twice for 1 min at 4 °C with intermittent cooling on ice) and lysates were cleared by centrifugation (20,000g, 30 min, 4 °C). Supernatants were incubated overnight at 4 °C with equilibrated streptavidin magnetic beads (100 µl per sample) under rotation. Beads were subsequently washed twice with RIPA buffer, once with 1 M KCl, once with 0.1 M Na₂CO₃, once with 2 M urea in 10 mM Tris-HCl (pH 8.0), and twice again with RIPA buffer. Bound proteins were eluted in 2× reducing sample buffer by heating at 95 °C for 5 min before downstream analysis.

LC–MS/MS based proteomics

Samples were boiled at 95 °C for 5 min and sonicated with a probe sonicator. The digestion was performed in a 96-well LoBind plate (Eppendorf) semi-automated on the Andrew+ Pipetting Robot (Waters). Disulfide bonds were reduced in 10 mM dithiothreitol for 30 min at 56 °C while shaking at 800 rpm and alkylated in presence of 20 mM iodoacetamide for 30 min at 37 °C. Carboxylate-modified magnetic E3 and E7 speed beads (Cytvia Sera-Mag) at a 1:1 ratio in liquid chromatography–mass spectrometry (LC–MS)-grade water were added in a 10:1 (beads/protein) ratio adapted from the SP3-protocol workflow⁷⁷. Protein binding was performed at 50% acetonitrile during shaking at 600 rpm for 18 min. Magnetic beads were washed twice with 80% ethanol and 100% acetonitrile. Digestion with trypsin in 100 mM AmBiCa was performed (sequencing grade, Promega) at a 1:100 (enzyme:protein) ratio at 37 °C overnight while shaking at 500 rpm. Trifluoroacetic acid was added to a final concentration of 1% and shaken at 500 rpm for 5 min. The supernatant containing tryptic peptides was transferred into a new 96-well LoBind plate, ready for subsequent LC–MS/MS analysis. Chromatography separation of tryptic peptides was achieved with a two-buffer system (buffer A: 0.1% formic acid in H₂O, buffer B: 0.1% formic acid in 80% acetonitrile) on a UHPLC (Vanquish neo UHPLC system, Thermo Fisher Scientific) at a flow rate of 0.4 µl min⁻¹. Attached to the UHPLC was a peptide trap (300 µm × 5 mm, C18, PepMap Neo Trap Cartridge, Thermo Fisher Scientific) for online desalting and purification, followed by a 25 cm C18 reversed-phase column (120 Å, 1.7 µm, 75 µm × 250 mm, Aurora Ultimate, IonOptics). Peptides were separated using a 70 min method with linearly increasing acetonitrile concentration from 3% to 34% buffer B over 60 min. MS/MS measurements were performed on a quadrupole-orbitrap hybrid mass spectrometer (Exploris 480, Thermo Fisher Scientific). Eluting peptides were ionized using a nano-electrospray ionization source (nano-ESI) with a spray voltage of 1,700 and analysed in data-independent acquisition (DIA) mode. For each MS1 scan, the ion-accumulation time was set to automatic, with an AGC target of 300%. The scan was set to m/z 400–800 with a resolution of 120,000 at m/z 200. Within a precursor mass range of m/z 400–800, fragmentation in DIA mode with m/z 12 isolation windows and m/z 1 window overlaps was performed (total of 33 scan events). Fragmentation was performed at a normalized collision energy of 30% using higher-energy collisional dissociation. The Orbitrap resolution was set to 60,000. LC–MS/MS data were searched using the CHIMERYS algorithm integrated into the Proteome Discoverer software (v.3.1.0.638, Thermo Fisher Scientific) with the Inferys 3.0.0 as the prediction model against a reviewed Mus musculus Swissprot database. Carbamidomethylation was set as a fixed modification for cysteine residues. The oxidation of methionine was allowed as variable modification. A maximum number of one missing tryptic cleavage was set. Peptides between 7 and 30 amino acids were considered. A strict cut-off (FDR < 0.01) was set for peptide and protein identification.

Ethics

All experimental procedures involving animals complied with institutional guidelines and adhered to the German Animal Welfare Act. Ethical approval for animal experiments was granted by the State Authority of Hamburg, Germany (approval numbers 41/22, 53/24, 40/25). For human eye tissue obtained from the Johns Hopkins University, all procedures were approved within the IRB00374036 “Characterizing mechanisms involved in pathological processes of retina neurodegeneration and optic neuritis present in MS in post-mortem samples”. For the human tissue analyses, as the samples could no longer be linked to identifiable individuals, the work did not qualify as a “research project on humans” under § 9 para. 2 of the Hamburg Chamber of Commerce Act for the Health Professions. Consequently, consultation required under § 15 para. 1 of the Professional Code of Conduct for Physicians in Hamburg was not necessary. For cortex tissue obtained from the UK Multiple Sclerosis Tissue Bank at Imperial College London, all procedures used by the Tissue Bank in the procurement, storage and distribution of tissue have been approved by the relevant Multicentre Research Ethics Committee (08/MRE09/31+5).

Statistical analysis

All statistical analysis was performed in R (v.4.4.1). The statistical analyses used for snRNA-seq data are described in the respective sections of the Article. A detailed description of statistical methods for experimental data is provided in the figure legends. Unless stated otherwise, data are presented as mean values. Differences between two experimental groups were determined using unpaired two-tailed Student’s t-tests or Mann–Whitney U-tests. The reported n represents biologically independent samples and not technical replicates. Statistical analysis of the clinical scores in EAE experiments was performed using the Mann–Whitney U-test on the areas under the curve for each animal.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Raw human retina snRNA-seq data have been deposited in the European Nucleotide Archive under accession number PRJEB102624. Processed snRNA-seq data, including the raw count matrix after quality control and corresponding cell- and sample-level metadata, are available at Figshare⁷⁸ (https://doi.org/10.6084/m9.figshare.33075182). Sequencing reads were aligned using the 10x Genomics Cell Ranger GRCh38 reference (2020-A; https://www.10xgenomics.com/support/software/cell-ranger/downloads/cr-ref-build-steps#human-ref-2020-a). Raw bulk mRNA-seq data are available from the GEO under accession number GSE332759. Mouse sequencing reads were aligned using the 10x Genomics Cell Ranger mm10 reference (2020-A; https://www.10xgenomics.com/support/software/cell-ranger/downloads/cr-ref-build-steps#mouse-ref-2020-a). The MS proteomics data have been deposited to the ProteomeXchange Consortium through the PRIDE⁷⁹ partner repository under dataset identifier PXD080433. Source data are provided with this paper.

1. Woo, M. S., Engler, J. B. & Friese, M. A. The neuropathobiology of multiple sclerosis. Nat. Rev. Neurosci. 25, 493–513 (2024).

   Article  CAS  PubMed  Google Scholar 

2. Tran, N. M. et al. Single-cell profiles of retinal ganglion cells differing in resilience to injury reveal neuroprotective genes. Neuron 104, 1039–1055 (2019).

   Article  CAS  PubMed  PubMed Central  Google Scholar 

3. Attfield, K. E., Jensen, L. T., Kaufmann, M., Friese, M. A. & Fugger, L. The immunology of multiple sclerosis. Nat. Rev. Immunol. 22, 734–750 (2022).

   Article  CAS  PubMed  Google Scholar 

4. Reich, D. S., Lucchinetti, C. F. & Calabresi, P. A. Multiple sclerosis. New Engl. J. Med. 378, 169–180 (2018).

   Article  CAS  PubMed  PubMed Central  Google Scholar 

5. Kuhlmann, T. et al. Multiple sclerosis progression: time for a new mechanism-driven framework. Lancet Neurol. 22, 78–88 (2023).

   Article  PubMed  Google Scholar 

6. Zimmermann, H. G. et al. Association of retinal ganglion cell layer thickness with future disease activity in patients with clinically isolated syndrome. JAMA Neurol. 75, 1071 (2018).

   Article  PubMed  PubMed Central  Google Scholar 

7. Baden, T., Euler, T. & Berens, P. Understanding the retinal basis of vision across species. Nat. Rev. Neurosci. 21, 5–20 (2020).

   Article  CAS  PubMed  Google Scholar 

8. Hahn, J. et al. Evolution of neuronal cell classes and types in the vertebrate retina. Nature 624, 415–424 (2023).

   Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

9. Yan, W. et al. Cell atlas of the human fovea and peripheral retina. Sci. Rep. 10, 9802 (2020).

   Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

10. Li, J. et al. Single-cell atlas of the transcriptome and chromatin accessibility in the human retina. Nat. Genet. 58, 418–433 https://doi.org/10.1038/s41588-025-02454-1 (2026).

11. Macnair, W. et al. snRNA-seq stratifies multiple sclerosis patients into distinct white matter glial responses. Neuron https://doi.org/10.1016/j.neuron.2024.11.016 (2024).

    Article  PubMed  Google Scholar 

12. Lerma-Martin, C. et al. Cell type mapping reveals tissue niches and interactions in subcortical multiple sclerosis lesions. Nat. Neurosci. 27, 2354–2365 (2024).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

13. Schirmer, L. et al. Neuronal vulnerability and multilineage diversity in multiple sclerosis. Nature 573, 75–82 (2019).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

14. Green, A. J., McQuaid, S., Hauser, S. L., Allen, I. V. & Lyness, R. Ocular pathology in multiple sclerosis: retinal atrophy and inflammation irrespective of disease duration. Brain 133, 1591–1601 (2010).

    Article  PubMed  PubMed Central  Google Scholar 

15. Rodriguez, A. R., de Sevilla Müller, L. P. & Brecha, N. C. The RNA binding protein RBPMS is a selective marker of ganglion cells in the mammalian retina. J. Comp. Neurol. 522, 1411–1443 (2014).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

16. Petzold, A. et al. Diagnosis and classification of optic neuritis. Lancet Neurol. 21, 1120–1134 (2022).

    Article  PubMed  Google Scholar 

17. Mastellos, D. C., Hajishengallis, G. & Lambris, J. D. A guide to complement biology, pathology and therapeutic opportunity. Nat. Rev. Immunol. 24, 118–141 (2024).

    Article  CAS  PubMed  Google Scholar 

18. Edwards, A. O. et al. Complement factor H polymorphism and age-related macular degeneration. Science 308, 421–424 (2005).

    Article  ADS  CAS  PubMed  Google Scholar 

19. Haines, J. L. et al. Complement factor H variant increases the risk of age-related macular degeneration. Science 308, 419–421 (2005).

    Article  ADS  CAS  PubMed  Google Scholar 

20. Zetterberg, M. et al. Association of complement factor H Y402H gene polymorphism with Alzheimer’s disease. Am. J. Med. Genet. B 147B, 720–726 (2008).

    Article  CAS  Google Scholar 

21. Wang, J. H. et al. Modeling the cumulative genetic risk for multiple sclerosis from genome-wide association data. Genome Med. 3, 3 (2011).

    Article  PubMed  PubMed Central  Google Scholar 

22. Jordão, M. J. C. et al. Single-cell profiling identifies myeloid cell subsets with distinct fates during neuroinflammation. Science 363, eaat7554 (2019).

    Article  PubMed  Google Scholar 

23. Vainchtein, I. D. et al. Characterizing microglial gene expression in a model of secondary progressive multiple sclerosis. Glia 71, 588–601 (2023).

    Article  CAS  PubMed  Google Scholar 

24. Itoh, N. et al. Cell-specific and region-specific transcriptomics in the multiple sclerosis model: focus on astrocytes. Proc. Natl Acad. Sci. USA 115, E302–E309 (2018).

    Article  CAS  PubMed  Google Scholar 

25. Schattling, B. et al. Bassoon proteinopathy drives neurodegeneration in multiple sclerosis. Nat. Neurosci. 22, 887–896 (2019).

    Article  CAS  PubMed  Google Scholar 

26. Woo, M. S. et al. The immunoproteasome disturbs neuronal metabolism and drives neurodegeneration in multiple sclerosis. Cell 188, 4567–4585 (2025).

    Article  CAS  PubMed  Google Scholar 

27. McRae, B. L. et al. Induction of active and adoptive relapsing experimental autoimmune encephalomyelitis (EAE) using an encephalitogenic epitope of proteolipid protein. J. Neuroimmunol. 38, 229–240 (1992).

    Article  CAS  PubMed  Google Scholar 

28. Simmons, S. B., Pierson, E. R., Lee, S. Y. & Goverman, J. M. Modeling the heterogeneity of multiple sclerosis in animals. Trends Immunol. 34, 410–422 (2013).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

29. Friedmann Angeli, J. P. et al. Inactivation of the ferroptosis regulator Gpx4 triggers acute renal failure in mice. Nat. Cell Biol. 16, 1180–1191 (2014).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

30. Yan, H. et al. Ferroptosis: mechanisms and links with diseases. Signal Transduct. Target. Ther. 6, 49 (2021).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

31. Kamala, O. et al. Homodimeric minimal factor H: in vivo tracking and extended dosing studies in factor H deficient mice. Front. Immunol. 12, 752916 (2021).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

32. Yang, Y. et al. An engineered complement factor h construct for treatment of C3 glomerulopathy. J. Am. Soc. Nephrol. 29, 1649–1661 (2018).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

33. Schnabolk, G. et al. Delivery of CR2-fH using AAV vector therapy as treatment strategy in the mouse model of choroidal neovascularization. Mol. Ther. Methods Clin. Dev. 9, 1–11 (2018).

    Article  CAS  PubMed  Google Scholar 

34. Lucientes-Continente, L., Márquez-Tirado, B. & Goicoechea de Jorge, E. The factor H protein family: the switchers of the complement alternative pathway. Immunol. Rev. 313, 25–45 (2023).

    Article  CAS  PubMed  Google Scholar 

35. Dixon, S. J. & Olzmann, J. A. The cell biology of ferroptosis. Nat. Rev. Mol. Cell Biol. 25, 424–442 (2024).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

36. Van San, E. et al. Ferroptosis contributes to multiple sclerosis and its pharmacological targeting suppresses experimental disease progression. Cell Death Differ. 30, 2092–2103 (2023).

    Article  PubMed  PubMed Central  Google Scholar 

37. Woo, M. S. et al. STING orchestrates the neuronal inflammatory stress response in multiple sclerosis. Cell 187, 4043–4060 (2024).

    Article  CAS  PubMed  Google Scholar 

38. Rothammer, N. et al. G9a dictates neuronal vulnerability to inflammatory stress via transcriptional control of ferroptosis. Sci. Adv. 8, eabm5500 (2022).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

39. Zipfel, P. F. & Heidenreich, K. The 4 functional segments of factor H: role in physiological target recognition and contribution to disease. J. Immunol. 214, 2150–2164 (2025).

    Article  CAS  PubMed  Google Scholar 

40. Weismann, D. et al. Complement factor H binds malondialdehyde epitopes and protects from oxidative stress. Nature 478, 76–81 (2011).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

41. Seiler, A. et al. Glutathione peroxidase 4 senses and translates oxidative stress into 12/15-lipoxygenase dependent- and AIF-mediated cell death. Cell Metab. 8, 237–248 (2008).

    Article  CAS  PubMed  Google Scholar 

42. Branon, T. C. et al. Efficient proximity labeling in living cells and organisms with TurboID. Nat. Biotechnol. 36, 880–887 (2018).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

43. Alic, L. et al. A genome-wide association study identifies key modulators of complement factor H binding to malondialdehyde-epitopes. Proc. Natl Acad. Sci. USA 117, 9942–9951 (2020).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

44. Józsi, M. et al. Anti-factor H autoantibodies block C-terminal recognition function of factor H in hemolytic uremic syndrome. Blood 110, 1516–1518 (2007).

    Article  PubMed  Google Scholar 

45. Cañeque, T. et al. Activation of lysosomal iron triggers ferroptosis in cancer. Nature 642, 492–500 (2025).

    Article  ADS  PubMed  PubMed Central  Google Scholar 

46. Sassano, M. L. et al. Endoplasmic reticulum–mitochondria contacts are prime hotspots of phospholipid peroxidation driving ferroptosis. Nat. Cell Biol. 27, 902–917 (2025).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

47. Xu, L. et al. Live-cell imaging with fluorogenic radical-trapping antioxidant probes reveals the onset and progression of ferroptosis. Nat. Chem. https://doi.org/10.1038/s41557-025-01966-x (2025).

    Article  PubMed  PubMed Central  Google Scholar 

48. Kiss, M. G. et al. Cell-autonomous regulation of complement C3 by factor H limits macrophage efferocytosis and exacerbates atherosclerosis. Immunity 56, 1809–1824 (2023).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

49. Hammond, J. W. et al. Complement-dependent synapse loss and microgliosis in a mouse model of multiple sclerosis. Brain Behav. Immun. 87, 739–750 (2020).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

50. Kolev, M. et al. Diapedesis-induced integrin signaling via LFA-1 facilitates tissue immunity by inducing intrinsic complement C3 expression in immune cells. Immunity 52, 513–527 (2020).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

51. Tobimatsu, S. & Kato, M. Multimodality visual evoked potentials in evaluating visual dysfunction in optic neuritis. Neurology 50, 715–718 (1998).

    Article  CAS  PubMed  Google Scholar 

52. Porciatti, V. & Sartucci, F. Retinal and cortical evoked responses to chromatic contrast stimuli. Brain 119, 723–740 (1996).

    Article  PubMed  Google Scholar 

53. Meltzer, E. et al. Retinal architecture and melanopsin-mediated pupillary response characteristics. JAMA Neurol. 74, 574 (2017).

    Article  PubMed  PubMed Central  Google Scholar 

54. La Morgia, C. et al. Melanopsin retinal ganglion cell loss in Alzheimer disease. Ann. Neurol. 79, 90–109 (2016).

    Article  PubMed  Google Scholar 

55. Rezola Artero, M. et al. Intracellular complement factor H promotes tumor progression through modulation of cell cycle and actin cytoskeleton. Commun. Biol. 9, 524 (2026).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

56. Fleckenstein, M. et al. Age-related macular degeneration. Nat. Rev. Dis. Primers 7, 31 (2021).

    Article  PubMed  PubMed Central  Google Scholar 

57. Zou, Y. et al. Cytochrome P450 oxidoreductase contributes to phospholipid peroxidation in ferroptosis. Nat. Chem. Biol. 16, 302–309 (2020).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

58. Yan, B. et al. Membrane damage during ferroptosis is caused by oxidation of phospholipids catalyzed by the oxidoreductases POR and CYB5R1. Mol. Cell 81, 355–369 (2021).

    Article  CAS  PubMed  Google Scholar 

59. Lorenz, S. M. et al. A fin-loop-like structure in GPX4 underlies neuroprotection from ferroptosis. Cell 189, 287–306 (2026).

    Article  CAS  PubMed  Google Scholar 

60. Absinta, M. et al. A lymphocyte–microglia–astrocyte axis in chronic active multiple sclerosis. Nature 597, 709–714 (2021).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

61. Zheng, G. X. Y. et al. Massively parallel digital transcriptional profiling of single cells. Nat. Commun. 8, 14049 (2017).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

62. Fleming, S. J. et al. Unsupervised removal of systematic background noise from droplet-based single-cell experiments using CellBender. Nat. Methods 20, 1323–1335 (2023).

    Article  CAS  PubMed  Google Scholar 

63. Hao, Y. et al. Dictionary learning for integrative, multimodal and scalable single-cell analysis. Nat. Biotechnol. 42, 293–304 (2024).

    Article  ADS  CAS  PubMed  Google Scholar 

64. Germain, P.-L., Lun, A., Garcia Meixide, C., Macnair, W. & Robinson, M. D. Doublet identification in single-cell sequencing data using scDblFinder. F1000Res. 10, 979 (2022).

    Article  Google Scholar 

65. Yu, G., Wang, L.-G., Han, Y. & He, Q.-Y. clusterProfiler: an R package for comparing biological themes among gene clusters. OMICS 16, 284–287 (2012).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

66. Love, M. I., Huber, W. & Anders, S. Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biol. 15, 550 (2014).

    Article  PubMed  PubMed Central  Google Scholar 

67. Aibar, S. et al. SCENIC: single-cell regulatory network inference and clustering. Nat. Methods 14, 1083–1086 (2017).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

68. Bankhead, P. et al. QuPath: open source software for digital pathology image analysis. Sci Rep. 7, 16878 (2017).

    Article  ADS  PubMed  PubMed Central  Google Scholar 

69. Rodríguez, C. I. et al. High-efficiency deleter mice show that FLPe is an alternative to Cre-loxP. Nat. Genet. 25, 139–140 (2000).

    Article  PubMed  Google Scholar 

70. Groh, J. et al. Microglia activation orchestrates CXCL10-mediated CD8⁺ T cell recruitment to promote aging-related white matter degeneration. Nat. Neurosci. 28, 1160–1173 (2025).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

71. Joly, S. et al. B cell-dependent EAE induces visual deficits in the mouse with similarities to human autoimmune demyelinating diseases. J. Neuroinflamm. 19, 54 (2022).

    Article  CAS  Google Scholar 

72. Schattling, B. et al. TRPM4 cation channel mediates axonal and neuronal degeneration in experimental autoimmune encephalomyelitis and multiple sclerosis. Nat. Med. 18, 1805–1811 (2012).

    Article  CAS  PubMed  Google Scholar 

73. Woo, M. S. et al. The NR4A2/VGF pathway fuels inflammation-induced neurodegeneration via promoting neuronal glycolysis. J. Clin. Invest. 134, e177692 (2024).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

74. Chan, K. Y. et al. Engineered AAVs for efficient noninvasive gene delivery to the central and peripheral nervous systems. Nat. Neurosci. 20, 1172–1179 (2017).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

75. Woo, M. S. et al. Neuronal metabotropic glutamate receptor 8 protects against neurodegeneration in CNS inflammation. J. Exp. Med. 218, e20201290 (2021).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

76. Mishima, E. et al. A non-canonical vitamin K cycle is a potent ferroptosis suppressor. Nature 608, 778–783 (2022).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

77. Hughes, C. S. et al. Single-pot, solid-phase-enhanced sample preparation for proteomics experiments. Nat. Protoc. 14, 68–85 (2019).

    Article  CAS  PubMed  Google Scholar 

78. Mayer, C. et al. Data for ‘Intracellular complement factor H protects neurons during CNS inflammation’. Figshare https://doi.org/10.6084/m9.figshare.33075182 (2026).

79. Perez-Riverol, Y. et al. The PRIDE database at 20 years: 2025 update. Nucleic Acids Res. 53, D543–D553 (2025).

Download references

Author notes

1. These authors contributed equally: Christina Mayer, Marcel S. Woo

Authors and Affiliations

1. Institute of Neuroimmunology and Multiple Sclerosis (INIMS), University Medical Center Hamburg-Eppendorf, Hamburg, Germany

   Christina Mayer, Marcel S. Woo, Jana K. Sonner, Lars Binkle-Ladisch, Felix Fischbach, Fernando Lucas-Ruiz, Darwin Manteufel, Kuno M.-J. Mattern, Lena Kristina Pfeffer, Albert Miguela, Vanessa Vieira, Nina Meurs, Simone Bauer, Jan Broder Engler & Manuel A. Friese

2. Hamburg Center for Translational Immunology (HCTI), University Medical Center Hamburg-Eppendorf, Hamburg, Germany

   Christina Mayer, Marcel S. Woo, Jana K. Sonner, Lars Binkle-Ladisch, Felix Fischbach, Fernando Lucas-Ruiz, Darwin Manteufel, Kuno M.-J. Mattern, Lena Kristina Pfeffer, Albert Miguela, Vanessa Vieira, Nina Meurs, Simone Bauer, Jan Broder Engler & Manuel A. Friese

3. Division of Neuroimmunology, Department of Neurology, Medical Faculty Mannheim, Heidelberg University, Mannheim, Germany

   Patricia Sekol, Sophia Schwarz & Lucas Schirmer

4. Departments of Neurology and Neuroscience, Johns Hopkins University School of Medicine, Baltimore, MD, USA

   Matthew D. Smith & Peter A. Calabresi

5. Group of Experimental Ophthalmology, Instituto Murciano de Investigación Biosanitaria Virgen de la Arrixaca (IMIB-Arrixaca) and Ophthalmology Department, Universidad de Murcia, Murcia, Spain

   Fernando Lucas-Ruiz

6. III. Department of Medicine, University Medical Center Hamburg-Eppendorf, Hamburg, Germany

   Yubing Guo, Anne Katrin Mühlig & Tilo Freiwald

7. Hamburg Center for Kidney Health (HCKH), University Medical Center Hamburg-Eppendorf, Hamburg, Germany

   Yubing Guo, Anne Katrin Mühlig & Tilo Freiwald

8. Section Mass Spectrometry and Proteomics, University Medical Center Hamburg-Eppendorf, Hamburg, Germany

   Bente Siebels

9. Neurodegeneration and Neuroinflammation Group, Girona Biomedical Research Institute (IDIBG-CERCA), Salt, Spain

   Albert Miguela

10. Institute of Pathology, Nephropathology Section, University Medical Center Hamburg-Eppendorf, Hamburg, Germany

    Sonia Wulf & Thorsten Wiech

11. University Children’s Hospital, University Medical Center Hamburg-Eppendorf, Hamburg, Germany

    Anne Katrin Mühlig

12. University Children’s Research@Kinder-UKE, University Medical Center Hamburg-Eppendorf, Hamburg, Germany

    Anne Katrin Mühlig

13. Institute of Metabolism and Cell Death, Molecular Targets and Therapeutics Center, Helmholtz Munich, Neuherberg, Germany

    Marcus Conrad

14. Translational Redox Biology, Technical University of Munich (TUM), TUM Natural School of Sciences, Garching, Germany

    Marcus Conrad

15. Institute of Clinical Neuroimmunology, University Hospital LMU Munich, Munich, Germany

    Florence M. Bareyre

16. Institute of Neuropathology, University Medical Center Hamburg-Eppendorf, Hamburg, Germany

    Susanne Krasemann & Markus Glatzel

17. Department of Infection Biology, Leibniz Institute for Natural Product Research and Infection Biology, Jena, Germany

    Peter F. Zipfel

18. Mannheim Center for Translational Neuroscience, Medical Faculty Mannheim, Heidelberg University, Mannheim, Germany

    Lucas Schirmer

19. Mannheim Institute for Innate Immunoscience, Medical Faculty Mannheim, Heidelberg University, Mannheim, Germany

    Lucas Schirmer

20. Interdisciplinary Center for Neurosciences, Heidelberg University, Heidelberg, Germany

    Lucas Schirmer

Contributions

C.M. and M.A.F. designed the study. C.M., M.S.W. and M.A.F. planned most experiments. C.M. conducted bioinformatical analyses. C.M. and M.S.W. performed and analysed most in vitro and in vivo experiments with additional contributions from J.K.S., L.B.-L., F.F., F.L.-R., D.M., K.M.-J.M., L.K.P., A.M., B.S., V.V., N.M., S.B. and S.W.; P.A.C., M.D.S. and L.S. provided human retinas, tissues from C3-reporter mice and expertise. P.S., S.S. and L.S. performed smFISH experiments. A.K.M., Y.G. and T.F. provided Cfh-floxed mice. S.K. and M.G. performed immunostainings of post-mortem tissues. F.M.B., T.W., M.C., J.B.E. and P.F.Z. provided material and expertise. C.M., M.S.W. and M.A.F. wrote the initial version of the manuscript. M.A.F. initiated, supervised and funded the study. All of the authors contributed to the editing and discussion of the manuscript and approved the final version.

Corresponding author

Correspondence to Manuel A. Friese.

Competing interests

The authors declare no competing interests.

Peer review information

Nature thanks Christopher J. Binder, Mate Kiss, Alexandre Prat and the other, anonymous, reviewer(s) for their contribution to the peer review of this work.

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.

Reprints and permissions