Autoimmune neuroinflammation leads to neuronal death via MIF nuclease-mediated parthanatos

Abstract

Central nervous system inflammation is implicated in neurodegeneration across several disorders, including multiple sclerosis (MS). While marked therapeutic progress has been made in preventing relapses in MS, primary neuroprotection in this disease remains elusive. This, in part, is due to an incomplete understanding of the molecular pathways involved in immune-mediated neuronal death. Here we show that parthanatos, a recently described caspase-independent and DNA damage-induced cell death program, contributes to neuron death in the experimental autoimmune encephalomyelitis (EAE) mouse model of autoimmune neuroinflammation. We reveal that DNA damage increases in neurons during EAE, and that neurons are progressively lost over the disease course. Neurons in affected areas display intracellular hallmarks of the parthanatos cascade. Genetic or pharmacologic blockade of the final step in parthanatos, genomic fragmentation by macrophage migration inhibitory factor (MIF) nuclease, reduces neuron loss and disease severity. Transcriptomic characterization of neurons with and without MIF nuclease activity reveals parthanatos-dependent differences in response to EAE. Together, this work establishes parthanatos as a key mechanism of neuron cell death during neuroinflammation.

Neuroinflammation triggers DNA damage and subsequent parthanatos-mediated neuron death. Inhibiting parthanatos in a mouse model of autoimmune inflammation is neuroprotective and reduces disease-associated paralysis.

Main

Multiple sclerosis (MS) is a chronic neuroinflammatory disease characterized by inflammation, demyelination and neuronal injury, and is a major cause of progressive neurological disability in young adults. Currently approved MS therapeutics target the peripheral immune system and effectively reduce relapses, but do not markedly affect neurodegeneration during progressive MS¹. While many studies suggest a link between chronic central nervous system (CNS)-restricted inflammation and neurodegeneration in progressive MS^1–5, the biology underlying inflammation-mediated neuron death remains incompletely understood.

Recent advancements highlight a role for tissue-restricted, chronic inflammation in creating a neurotoxic microenvironment in MS. There are several upstream mechanisms, including mitochondrial dysfunction, chronic demyelination, iron deposition, immune cell-derived cytokines and reactive oxygen species (ROS)^6–8 that lead to oxidative stress and DNA damage^9–12. Notably, neurons in MS have increased DNA damage, but the consequences of these DNA breaks are unclear^11–13. Here, we investigate parthanatos, a form of regulated cell death initiated by DNA damage, in neurons during neuroinflammatory disease¹⁴.

In parthanatos, the nuclear enzyme poly(ADP-ribose) polymerase 1 (PARP1) senses DNA damage, activating PARP1 to generate branched poly(ADP-ribose) (PAR) polymers. When produced in excess, PAR escapes the nucleus and triggers the release of apoptosis-inducing factor (AIF) from mitochondria¹⁵. Cytosolic AIF, which contains a nuclear localization signal, binds the multifunctional protein macrophage migration inhibitory factor (MIF) and together AIF–MIF shuttle to the nucleus. Once in the nucleus, MIF fragments genomic DNA through intrinsic nuclease activity, culminating in cell death^14,16. Hallmarks of parthanatos-mediated neurodegeneration have been described in diseases, including Alzheimer’s disease, Parkinson’s disease^16,17, traumatic brain injury¹⁸ and other conditions^{12,16,17,19–22}, but had not previously been studied in autoimmune neuroinflammation. Of note, steps of the parthanatos cascade, including cytoplasmic AIF, upregulation of PAR and caspase-independent death have been reported in tissue from patients with MS and animal models^23–25.

Here, we provide evidence that key steps of parthanatos, including DNA damage, production of PAR and nuclear translocation of MIF, occur in spinal cord and retinal neurons during experimental autoimmune encephalomyelitis (EAE). Furthermore, using knock-in mice and pharmacologic inhibitors to prevent or limit MIF nuclease activity, we observe a neuroprotective effect by blocking parthanatos in EAE. These findings introduce a potentially targetable pathway linking autoimmune inflammation, DNA damage and neuron cell death^5,26,27.

Results

Neuronal DNA damage and loss of neurons in the spinal cord and retina during EAE

We first sought to determine the extent of oxidative stress and DNA damage that occurs within neurons during EAE, a mouse model of autoimmune neuroinflammation with axonal pathology and neuronal loss mirroring what is seen in MS^{2–4,28–31}. We focused our studies on the lumbar spinal cord and the retina, regions with inflammation-induced neuronal loss in EAE³². We established a systematic approach to quantify cells in the lumbar spinal cord by taking contiguous sections every 500 µm throughout the length of the tissue (typically about 14 mm long) and enumerating neurons (Methods and Extended Data Fig. 1a).

Extended Data Fig. 1. Spinal cord neuron quantification method and DNA damage within the spinal cord and retina.

(A) Depiction of lumbar spinal cord cell quantification method: 1) lumbar spinal cord is isolated from full spinal cord of mouse after postmortem dissection. 2) Following freezing and cryoprotection steps, transverse sections every are collected at 500 µm intervals onto a single slide, such that each lumbar spinal cord yields about 25 sections. 3) All sections per spinal cord of each mouse mounted onto a single slide. 4) Following immunofluorescence staining protocol, each intact slide is imaged on microscope. 5) Lumbar spinal cord cross-section, diagramming the ventral and dorsal roots that were quantified. The dorsal horns were defined as the gray matter area just ventral from dorsolateral fasciculus (laminas I/II). The ventral horns were defined as the gray matter region ventrally located from the central canal that is perpendicular to the dorsoventral axis. In both dorsal and ventral horns, the sensory and motor neurons of the somatic regions were quantified. 6) Representative image of lumbar spinal cord neurons in one section from a single slide. (B–E) Representative images and quantifications of DNA-damaged neurons from both the ventral (B–C) and dorsal (D–E) horns of the lumbar spinal cord in CFA and peak EAE to independently validate γH2AX staining patterns with an alternate marker of DNA damage, 53BP1. Peak EAE/CFA ventral horns N = 5, peak EAE/CFA dorsal horns N = 4. (F) Mouse eyeball diagram with representative cross-section showing the innermost RGC layer which was analyzed. Created in BioRender (https://BioRender.com/cl2mrss). (G) Representative cross-section images of mouse retina taken from CFA and peak EAE, with immunolabeling for γSyn (RGCs) and γH2AX (DNA damage). (H) γH2AX+ RGC quantifications in retinas from CFA only and peak EAE. N = 6 per group. (I) Representative cross-section images of mouse retina taken from CFA and peak EAE mice, with antibodies to detect γSyn+ RGCs and 8-OHdG (a marker of oxidative stress). (J) Fluorescent intensity quantifications in CFA only and peak EAE retinas. N = 6 per group. (K–L) Representative images and quantifications to independently validate γH2AX staining patterns in the retina with an alternate marker of DNA damage, 53BP1. N = 5 per group. Scale bars=50 µm. Data shown in each graph represents mean ± SD; each dot represents one mouse. Data in Extended Data Fig. 1C,E,L analyzed with two-way ANOVA followed by Tukey’s post hoc test. Data in Extended Data Fig. 1H,J analyzed with two-sided, unpaired t-tests.

Double-strand DNA breaks (DSBs) trigger activation of ataxia-telangiectasia mutated kinase, which modifies the histone H2AX to the phosphorylated form, termed γH2AX. In areas of DSBs, γH2AX rapidly assembles and is a sensitive marker for DNA damage³³. We found increased γH2AX signal in neurons of the lumbar spinal cord during peak EAE (day 15; Fig. 1a). Notably, γH2AX seemed to most prominently accumulate in the ventral horns. While complete Freund’s adjuvant (CFA)-only control mice had ~10% γH2AX⁺ ventral horn neurons, during peak and chronic (day 45) EAE, most ventral neurons (~80–90%) were γH2AX⁺ (Fig. 1b,c). Concomitantly, neuron density progressively decreased from peak to chronic EAE (Fig. 1d). In laminas I/II of the dorsal horns, there was a modest increase in the fraction of γH2AX⁺ neurons during peak and chronic EAE but no statistically significant change in the neuron number (Fig. 1e–g). We extended this key finding using an alternative marker of DSBs, 53BP1 (refs. ^33–35) (Extended Data Fig. 1b–e). We speculated that DNA damage in our model system was related to oxidative injury, which is known to contribute to pathology in MS and EAE^9,36. In support of this, we observed increased 8-hydroxy-2’-deoxyguanosine (8-OHdG), an indicator of oxidatively damaged nucleic acids³⁵, in lumbar spinal cord neurons at peak EAE (Fig. 1h,i).

Fig. 1. Lumbar spinal cord ventral neurons undergo substantial DNA damage and neurodegeneration during EAE.

a, Representative images of lumbar spinal cord sections from CFA and peak EAE mice stained for γH2AX and NeuN. White arrows indicate DNA-damaged neurons. b–g, Representative images and quantifications of neuronal DNA damage and survival in the ventral (b–d) and dorsal (e–g) spinal cord. Fraction of γH2AX⁺NeuN⁺ cells out of total NeuN⁺ cells throughout the ventral horns of the lumbar spinal cord in CFA only, peak EAE and chronic EAE (b,c). CFA n = 5, peak EAE n = 8, chronic EAE n = 9. NeuN⁺ cell numbers in the ventral horns of the lumbar spinal cord in CFA only, peak EAE and chronic EAE (d). CFA n = 5, peak EAE n = 16, chronic EAE n = 17. Peak EAE ventral horn NeuN⁺ cell counts pooled from two experiments, and chronic EAE ventral horn NeuN⁺ cell counts pooled from three experiments. Fraction of γH2AX⁺NeuN⁺ cells out of total NeuN⁺ cells in the dorsal horns of the lumbar spinal cord in CFA only, peak EAE and chronic EAE (e,f). n = 6 per group. NeuN⁺ cell numbers in the dorsal horns of the lumbar spinal cord in CFA, peak EAE and chronic EAE (g). n = 6 per group. h,i, Representative staining (h) and fraction of 8-OHdG⁺NeuN⁺ out of total NeuN⁺ cells (i) in the dorsal and ventral horns of the lumbar spinal cord from CFA and peak EAE mice. n = 6 per group. Insets: regions quantified (pink) and presented in representative images (red boxes). Scale bars, 1 mm (a), 50 µm (b,e,h). Data shown in each graph represent mean ± s.d.; each dot represents one mouse. Data in c, d, f and g were analyzed with one-way analysis of variance (ANOVA) followed by Tukey’s post hoc test. Data in i were analyzed with a two-way ANOVA followed by Tukey’s post hoc test.

We next examined retinal ganglion cells (RGCs) that die as a sequela of optic neuritis in EAE^37–39. RGCs, like spinal cord neurons, display increased DNA damage (γSyn⁺γH2AX⁺, NeuN⁺53BP1⁺) and oxidative stress (γSyn⁺8-OHdG⁺) in peak EAE compared to CFA only (Extended Data Fig. 1f–l).

Parthanatos markers are upregulated in neurons during EAE and MS

After establishing an increase in neuronal DNA damage, we assessed markers of parthanatos. Key steps of parthanatos include a DNA damage event, overproduction and cytoplasmic transport of PAR and AIF–MIF binding and translocation to the nucleus (Fig. 2a). Peak EAE spinal cord had less noncleaved genomic DNA compared to CFA, as measured via pulsed-field gel electrophoresis (Fig. 2b,c). Western blot analysis of lumbar spinal cord lysates showed substantial accumulation of PAR starting at day 10 and peaking at day 15 of EAE (Fig. 2d,e). Using confocal microscopy, we confirmed increased PAR in neurons specifically over this time course (Extended Data Fig. 2a–c). This induction was accompanied by a gradual decline in NeuN and an increase in γH2AX abundance (Fig. 2f–h). To assess the subcellular localization of neuronal PAR in vivo, we performed high resolution confocal imaging with co-labeling for PAR and NeuN. Cytosolic PAR in neurons was increased in peak EAE compared to CFA only, and partially recovered in chronic EAE (Fig. 2i,j). We next sought to assess nuclear accumulation of MIF and AIF–MIF binding. We validated MIF antibody labeling with MIF^−/− spinal cord tissue, finding high baseline MIF expression in neurons as previously reported⁴⁰ (Extended Data Fig. 2d,e). Furthermore, nuclear MIF was increased in neurons during peak EAE compared to CFA, and as with PAR, only partially recovered in chronic EAE (Fig. 2k,l). Neuron-specific AIF–MIF binding, total and nuclear, was also elevated in peak EAE as measured by proximity ligation assay (PLA; Fig. 2m,n). Finally, to experimentally validate the link between oxidative stress and neuron parthanatos, we treated neonatal-derived primary neurons with hydrogen peroxide to induce oxidative injury and observed rapid increase in both PAR (Extended Data Fig. 2f,h) and γH2AX (Extended Data Fig. 2g,i).

Fig. 2. Accumulation of parthanatos markers in lumbar spinal cord neurons during EAE.

a, Parthanatos diagram: ROS accumulates in cell and induces DNA damage; H2AX is phosphorylated (to γH2AX) in response to DSBs; PARP1 recognizes the DSB and synthesizes PAR to tether DNA repair enzymes to the site of damage; PAR is overproduced and miss-localizes into the cytosol and releases AIF from the mitochondria; cytosolic AIF binds MIF and together shuttle into the nucleus where MIF fragments DNA. b,c, Pulsed-field gel electrophoresis assay on CFA and peak EAE lumbar spinal cords to assess genomic DNA fragmentation. Representative blot with asterisk indicates region quantified as noncleaved genomic DNA (b). Percent noncleaved genomic DNA relative to CFA (c). CFA n = 5, peak EAE n = 6. d–h, Representative western blots and pooled quantitative analyses for PAR (d,e), NeuN (f(i),g) and γH2AX (f(ii),h) in lumbar spinal cord over an EAE time course. CFA n = 3, EAE n = 5–6 per time point. i,k, Representative images of lumbar spinal cord ventral horn neurons in CFA, peak EAE and chronic EAE stained for PAR (i) and MIF (k) to measure intracellular localization. j, Integrated density quantifications of cytoplasmic PAR. l, Integrated density quantifications of nuclear MIF. n = 5 per group. m,n, Representative field of view (FOV) images (m) and quantifications of a PLA to assess MIF and AIF binding in full neuronal cells (n(i)) and in neuronal nuclei (n(ii)) within the ventral lumbar spinal cord of CFA and peak EAE mice. CFA n = 5, peak EAE n = 6. Insets: region quantified (pink) and presented in representative images (red boxes). Scale bars, 5 µm (i,k), 25 µm (m). Data shown in each graph represent mean ± s.d.; each dot represents one mouse. Data in e, g, h, j and l were analyzed with one-way ANOVA followed by Tukey’s post hoc test. Data in c and n were analyzed with two-sided, unpaired t-tests. For all variables with the same letter, the difference between the means is not statistically significant (P > 0.05) (e,g,h).

Extended Data Fig. 2. Supporting evidence for parthanatos in neurons during inflammation or oxidative stress.

(A) High-magnification co-localization images of a NeuN⁺ /MIF⁺ /Hoechst+ cell and a NeuN⁺ /PAR⁺ /Hoechst+ cell used in Fig. 2 chronic EAE representative images. (B–C) Representative images and fluorescent intensity quantifications of ventral horn lumbar spinal cord PAR+ NeuN+ cells in CFA, peak EAE, and chronic EAE. N = 5 per genotype. (D–E) Immunofluorescence confirmation that spinal cord neurons express MIF at baseline. Representative images and quantifications of ventral horn lumbar spinal cord MIF+ neurons in WT and MIF^−/− mice. N = 5 per genotype. (F–G) Representative western blots of primary neurons, derived from P17 WT mice pups, exposed to hydrogen 150 µM peroxide and blotted for PAR (E) and γH2AX (F). Dashed line indicates replicate experiment run on same blot. (H–I) Pooled quantitative analyses of PAR (G) and γH2AX (H) in primary neurons treated with hydrogen peroxide or DMSO control. N = 4 biological replicates per group, over two independent experiments. (J) Quantitative analysis of cell viability in WT and MIF-E22Q primary neurons treated with vehicle or hydrogen peroxide for 24 hours. (K) Brain slices obtained from a 41-year-old control donor (top row) and a 41-year-old multiple sclerosis patient (bottom row) were stained for NeuN and PAR. NeuN+ and PAR⁺ /NeuN+ cells (from three brain sections per donor) were counted in gray matter cortex, with the percentage of double-positive cells over the total number of cells displayed in the far-right pie chart. N = 1 control donor, N = 1 multiple sclerosis donor. Scale bars= 5 µm (Extended Data Fig. 2A); 50 µm (Extended Data Fig. B,D); 25 µm (Extended Data Fig. K). Data shown in each graph represents mean ± SD; each dot represents one mouse (Extended Data Fig. 2C,E) or one well (Extended Data Fig. 2H–J). Extended Data Fig. 2E analyzed with two-sided, unpaired t-tests. Extended Data Fig. 2C,H,I analyzed with one-way ANOVA followed by Tukey’s post hoc test. Extended Data Fig. 2J analyzed with two-way ANOVA followed by Tukey’s post hoc test. For all variables with the same letter, the difference between the means is not statistically significant (p > 0.05) (Extended Data Fig. 2H–I).

We also observed neuronal PAR staining in a limited sample of human postmortem cortical tissue. We found higher average neuronal PAR in a 41-year-old person with MS and recent MRI activity as compared to an age-matched non-neurological disease control (Extended Data Fig. 2j).

Inhibition of MIF nuclease activity is neuroprotective in EAE without impacting peripheral immune cell chemotaxis, reactive gliosis or myelin

Our group has previously developed a mouse line in which there is a point mutation in MIF (MIF-E22Q) that inhibits its nuclease activity without affecting cytokine or other enzymatic functions¹⁴. Using this knock-in strain, we generated primary neuronal cultures and treated them with hydrogen peroxide, finding attenuated death in MIF-E22Q neurons (Extended Data Fig. 2k).

Next, we induced EAE to assess the impact on neurologic disability and neuron survival in vivo. Both wild-type (WT) and MIF-E22Q had similar peak EAE severity, but MIF-E22Q mice had lower long-term disability as measured by motor scores in the chronic phase (Fig. 3a,b). Quantification of neurons in the ventral horns of the lumbar spinal cord showed a substantial neuroprotective effect in MIF-E22Q mice at peak and chronic EAE compared to WT (Fig. 3c,d) despite there being no differences in baseline neuronal density between the two genotypes (Extended Data Fig. 3a). To confirm that NeuN appropriately labeled neurons in the genotypes studied and was not downregulated due to neuronal stress during EAE, we verified our results with the alternate pan-neuronal marker: HuC/HuD (Extended Data Fig. 3b–e). Furthermore, MIF-E22Q mice had increased RGC density compared to WT at peak and chronic EAE (Fig. 3e,f) with no difference at baseline (Extended Data Fig. 3f).

Fig. 3. Genetic ablation of MIF nuclease activity in EAE is neuroprotective.

MIF-E22Q knock-in mice have a point mutation in MIF that specifically disrupts its nuclease activity. a, EAE was induced in MIF-E22Q and WT mice, and they were scored daily for clinical disease severity. b, Area under the curve (AUC) quantification for total (EAE days 1–45), peak (EAE days 13–19) and chronic (EAE days 25–45) time ranges. WT n = 8, MIF-E22Q n = 10, representative of 3 replicate experiments. c,d, Representative images (c) and quantifications (d) of mean lumbar spinal cord ventral horn neuron counts in MIF-E22Q compared with WT mice at peak and chronic EAE time points. Peak EAE WT n = 8, peak EAE MIF-E22Q n = 8, chronic EAE WT n = 7, chronic EAE MIF-E22Q n = 6. e,f, Representative images (e) and quantification (f) of RGCs (Brn3a⁺ cells) from whole-mount retinas in MIF-E22Q compared with WT at peak and chronic EAE time points. Peak EAE WT n = 9, peak EAE MIF-E22Q n = 9, chronic EAE WT n = 8, chronic EAE MIF-E22Q n = 9. Insets: region quantified (pink) and presented in representative images (red boxes). Scale bars, 50 µm (c,e). Data shown in each graph (other than a) represent mean ± s.d.; each dot (other than a) represents one mouse; a shows one dot representing the mean of all mice per genotype that day ± s.e.m. Data in b were analyzed with a two-sided Mann–Whitney nonparametric test. Data in d and f were analyzed with two-sided, unpaired t-tests.

Extended Data Fig. 3. Validation of neuronal quantification method and impact of genetic model on neuron count.

(A) NeuN+ neuron counts in the ventral horn of the lumbar spinal cord of WT naive and MIF-E22Q naive control mice. N = 5 per genotype. (B–E) Validation of NeuN as a comprehensive marker for neurons in EAE by measuring agreement with an alternate label, HuC/HuD. (B) Full lumbar spinal cord cross-section, marked with antibodies against NeuN (Bi) and HuC/HuD (Bii). (C) Quantification of NeuN+ cells and HuC/HuD+ cells in WT and MIF-E22Q mice during chronic EAE. (D) Merged image showing overlap between NeuN and HuC/HuD+ immunofluorescence. (E) Quantification of double-positive cells. (F) Brn3a+ RGCs in the retina of WT naive and MIF-E22Q naive control mice. N = 5 per genotype. Insets display region quantified (pink). Scale bars=1 mm (Extended Data Fig. 3B,D); 50 µm (Extended Data Fig. 3D inset). Data shown in each graph represents mean ± SD; each dot represents one mouse. Extended Data Fig. 3C was analyzed with two-way ANOVA followed by Tukey’s post hoc test. Extended Data Fig. 3A,E,F analyzed with two-sided, unpaired t-tests.

Peripheral immune cell infiltration is a prominent component at peak EAE that modulates the degree of neurologic injury. While the MIF-E22Q variant does not directly impact the cytokine function of MIF, it could have unexpected effects on MIF production, secretion or immune cell infiltration and death that impact the initial immune response during EAE. We therefore sought to confirm that the neuroprotection seen in MIF-E22Q animals does not merely reflect decreased immune cell infiltration or activation. Using flow cytometry to analyze the initial inflammatory response in MIF-E22Q and WT lumbar spinal cords, we quantified myeloid and T cell subsets. There were similar numbers of total CD11b⁺ myeloid cells, Ly6C^hi macrophages, Ly6C^lo macrophages and granulocytes (Ly6G+) infiltrating the lumbar spinal cord in MIF-E22Q and WT mice (Extended Data Fig. 4a–c). There was no difference in the infiltration of CD8⁺ T cells, CD4⁺ T cells or T regulatory cells (CD4⁺CD25⁺FoxP3⁺) in MIF-E22Q and WT (Extended Data Fig. 4d–f). Helper T cell subsets also displayed similar activation states, with no statistically significant difference in the fraction of cells producing IFNγ or IL-17 (Extended Data Fig. 4g,h). We next tested whether lymphocyte division was altered in MIF-E22Q animals. CD4⁺ T cells had a similar ability to expand in response to in vitro stimulation in MIF-E22Q and WT genotypes (Extended Data Fig. 4i,j).

Extended Data Fig. 4. MIF-E22Q knock-in mutation does not affect immune cell infiltration to the spinal cord during peak EAE.

(A–C) Myeloid lineage cell analysis in WT and MIF-E22Q spinal cord at peak EAE. Representative gating strategy (A), and representative plots comparing myeloid cell subsets in WT and MIF-E22Q animals (B). Total cell counts per spinal cord of myeloid cell subsets (C) N = 6 per genotype. (D–H) Lymphoid cell analysis in WT and MIF-E22Q spinal cord at peak EAE. Representative gating strategy (D). Representative plots comparing T cell populations between genotypes (E). Total counts of T cell subsets per spinal cord (F). Representative plots comparing IFNɣ and IL-17A production by T helper cells in the spinal cord at peak EAE (G), and quantification of cytokine-expressing T helper cells (H). N = 6 per genotype. (I–J) In vitro T cell proliferation assay. CD4⁺ T cells were purified and activated in vitro with cell stimulation cocktail for 72 hours. Number of cell divisions as measured by dilution of membrane bound cell trace violet dye (I). Summary of cell division number per genotype (J). N = 3 per genotype. Data shown in each graph represents mean ± SD; each point represents one mouse. All data analyzed with one-way ANOVA followed by Tukey’s post hoc test. ns=not significant.

Reactive gliosis occurs as a sequela of immune cell activation and infiltration into the CNS in EAE and MS, likely impacting neuron survival⁴¹. Therefore, we aimed to understand if the MIF-E22Q mutation impacted microglia and astrocyte proliferation or activation. Glial cell numbers were not altered in the ventral horns of the lumbar spinal cord during peak EAE (Extended Data Fig. 5a–c). MIF-E22Q mice also had no overt difference in reactive microglia or astrocytes markers within the optic nerves during peak EAE (Extended Data Fig. 5d–i). Finally, we assessed the impact of MIF-E22Q on myelination. Using FluoroMyelin staining, we quantified white matter myelination, finding that EAE reduced myelin similarly in both WT and MIF-E22Q animals at peak (Extended Data Fig. 5j,k) and chronic EAE (Extended Data Fig. 5j,l).

Extended Data Fig. 5. MIF-E22Q knock-in mutation does not promote reactive gliosis in the spinal cord and optic nerves during peak EAE.

Representative images (A) and quantifications of Iba1+ (B) and Sox9+GFAP+ (C) cell counts in the ventral horns of the lumbar spinal cord during peak EAE of WT and MIF-E22Q mice. N = 5 per group. Representative images (D) and mean fluorescent intensity quantifications of TMEM119 (E), Iba1 (F), CD68 (G), and GFAP (H–I) in the optic nerves during peak EAE of WT and MIF-E22Q mice. (J) Representative images of FluoroMyelin labeling, dotted line represents the area of white matter used for measurements. (K–L) FluoroMyelin quantification at peak (K) and chronic (L) disease time points of ventral horn white matter in CFA control, WT EAE, and MIF-E22Q EAE mice. Insets display regions quantified (pink) and presented in representative images (red boxes). Scale bars=50 µm (Extended Data Fig. 5A,D,H), 500µm (Extended Data Fig. 5J). Data shown in each graph represents mean ± SD; each dot represents one mouse. Extended Data Fig. 5B,C,E,F analyzed with two-sided, unpaired t-tests, Extended Data Fig. 5K,L analyzed with one-way ANOVA with Tukey’s post hoc test. ns=not significant.

Pharmacologic inhibition of MIF nuclease is neuroprotective in chronic EAE

PAANIB-1 is a recently developed small molecule that was identified following an unbiased screen for compounds that inhibit MIF nuclease¹⁶. Previous work established that it does not affect the tautomerase or oxidoreductase enzymatic functions of MIF, other forms of cell death or nuclease activity of structurally related nucleases. These studies also demonstrated that 10 mg kg⁻¹ daily dose leads to about 1 μM PAANIB-1 brain concentration, a concentration that is expected to yield >80% protection from parthanatos in cell culture and to be well tolerated¹⁶.

We first tested the impact of PAANIB-1 in a prevention paradigm, in which we started daily gavage immediately before EAE immunization. Mice had identical symptom onset and comparable peak disease severity, but PAANIB-1 treatment resulted in less disability at the chronic time points (Fig. 4a,b). We observed neuroprotection in the ventral lumbar spinal cord (Fig. 4c,d) and retina (Fig. 4e,f) in PAANIB-1 compared to vehicle-treated mice. We also tested PAANIB-1 in a treatment paradigm in which mice were randomized to begin daily PAANIB-1 or vehicle treatment on the day of EAE symptom onset. While there was no statistically significant difference in motor score between treatment and control in this paradigm (Extended Data Fig. 6a,b), there was neuroprotection both in the spinal cord (Extended Data Fig. 6c,d) and retina (Extended Data Fig. 6e,f) at the chronic EAE time point. Of note, we confirmed in our model that PAANIB-1 did not affect upstream steps of the parthanatos pathway itself, such as excess cytosolic PAR (Extended Data Fig. 7a–d) or AIF–MIF binding and translocation to the nucleus (Extended Data Fig. 7e,f). PAANIB-1 also had no effect on infiltrating myeloid and T cell populations within the spinal cord during peak EAE (Extended Data Fig. 7g–j).

Fig. 4. Pharmacologic inhibition of MIF nuclease activity in EAE is neuroprotective.

a, Pretreatment (beginning 2 days before EAE induction) with a small molecule MIF nuclease inhibitor (PAANIB-1) or vehicle. Mice were scored daily for clinical disease severity. Blue line represents duration of treatment administration. b, AUC quantification for total (EAE days 1–45), peak (EAE days 13–19) and chronic (EAE days 25–45) time ranges. Vehicle n = 15, PAANIB-1 n = 14, representative of 2 replicate experiments. c,d, Representative images (c) and average lumbar spinal cord ventral horns quantification (d) for NeuN⁺ cells in PAANIB-1-compared with vehicle-treated EAE mice at the chronic EAE time point. n = 6 per group. e,f, Representative images (e) and RGC (Brn3a⁺ cells) quantification (f) from whole-mount retinas in PAANIB-1-treated WT EAE mice compared with vehicle-treated WT EAE mice at the chronic EAE time point. n = 6 per group. Insets: regions quantified (pink) and presented in representative images (red boxes). Scale bars, 50 µm (c,e). Data shown in each graph (other than a) represent mean ± s.d.; each dot (other than a) represents one mouse; a shows one dot representing the mean of all mice per genotype that day ± s.e.m. Data in b were analyzed with a two-sided Mann–Whitney nonparametric test. Data in d and f were analyzed with two-sided, unpaired t-tests.

Extended Data Fig. 6. Pharmacologic inhibition of MIF nuclease post-EAE development is neuroprotective.

(A) Treatment after EAE onset (starting ~10 days for most mice, following EAE induction) with MIF nuclease inhibitor (PAANIB-1) or vehicle. Mice were scored daily for clinical disease severity. Blue line represents duration of treatment administration. (B) Area under the curve (AUC) quantification for total (EAE days 1–45), peak (EAE days 13–19), and chronic (EAE days 25–45) time ranges. (C–D) Representative images and average lumbar spinal cord ventral horns quantification for NeuN+ cells in PAANIB-1-treated EAE mice compared to vehicle-treated EAE mice at the chronic EAE time point. (E–F) Representative images and RGC (Brn3a+ cells) quantification from whole-mount retinas in PAANIB-1-treated WT EAE mice compared to vehicle-treated WT EAE mice at the chronic EAE time point. Vehicle N = 8, PAANIB-1 N = 9. Insets display regions quantified (pink) and presented in representative images (red boxes). Scale bars=50 µm. Data shown in each graph (other than Extended Data Fig. 6A) represents mean ± SD; each dot (other than Extended Data Fig. 6A) represents one mouse, Extended Data Fig. 6A shows one dot representing the mean of all mice per genotype that day ± SEM. Extended Data Fig. 6B analyzed with a two-sided Mann Whitney non-parametric test. Extended Data Fig. 6C–F analyzed with two-sided, unpaired t-tests.

Extended Data Fig. 7. MIF nuclease pharmacologic inhibitor does not affect upstream parthanatos pathway or immune infiltration.

(A) Pre-treatment (beginning 2 days prior to EAE induction) with a small molecule MIF nuclease inhibitor (PAANIB-1) or vehicle. Mice were scored daily for clinical disease severity. Blue line represents duration of treatment administration. (B) Peak EAE motor scores on day of euthanasia. N = 5 per group. (C–D) Representative images and Integrated density quantifications of cytoplasmic PAR in neurons within the ventral lumbar spinal cords of vehicle-treated and PAANIB-1-treated peak EAE mice. N = 5 per group. (E–F) Representative images and quantifications of a proximity ligation assay (PLA) to assess MIF and AIF binding in neuronal field of view (FOV) ROIs, quantified both in whole neuron cells (Fi) or neuronal nuclei (Fii) in the ventral lumbar spinal cords of vehicle-treated and PAANIB-1-treated peak EAE mice. Insets display regions quantified (pink) and presented in representative images (red boxes). Scale bars=50 µm (Extended Data Fig. 7C); 25 µm (Extended Data Fig. 7E). (G–J) Flow cytometry analysis of immune cell infiltration with PAANIB-1 pre-treatment. (G) Myeloid lineage cell analysis in vehicle and PAANIB-1 spinal cords at peak EAE with representative gating strategy and plots comparing myeloid cell subsets. (H) Lymphoid cell analysis in vehicle and PAANIB-1 spinal cord at peak EAE with representative gating strategy and plots comparing T cell populations. (I) Representative gating strategy and plots comparing IFNɣ and IL-17A production by T helper cells in the spinal cord at peak EAE. N = 5 per group. (J) Quantification of GM-CSF produced by T cell populations in peak EAE treated with vehicle or PAANIB-1. N = 11 per group. Data shown in each graph represents mean ± SD; each dot represents one mouse. All data analyzed with two-sided, unpaired t-tests. ns=not significant.

Upregulation of immune-related gene expression in neurons during EAE

To characterize the importance of MIF nuclease at the transcriptomic level, we performed single-nucleus transcriptomic profiling of the lumbar spinal cord from WT and MIF-E22Q mice in CFA only and peak EAE (day 15) groups (Fig. 5a). To minimize processing artifacts between samples, barcoded probes were hybridized to transcripts in each sample before pooling, followed by single-nucleus partitioning of pooled samples⁴². Poor-quality nuclei were excluded from further analysis (Supplementary Table 1), and following quality control, we obtained 72,712 nuclei across ten samples for downstream analysis (Supplementary Table 2). Cell types were annotated using the scType package⁴³ and cell identities confirmed with canonical markers (Extended Data Fig. 8a). There was substantial infiltration of peripheral myeloid cells, T cells and expansion of microglia into the spinal cord during EAE, which was comparable between genotypes (Fig. 5b–d and Extended Data Fig. 8b).

Fig. 5. Transcriptomic analysis of lumbar spinal cord during peak EAE reveals a proinflammatory shift in neurons that is enhanced by genetic ablation of MIF nuclease activity.

a, Experimental design schematic depicting single-nucleus transcriptomic experiment in WT and MIF-E22Q female mice at day 14 (peak) EAE and CFA immunized controls. b–d, UMAP plots of 72,712 high-quality nuclei across 10 animals labeled by cell type (b), group (CFA or peak EAE) (c) and genotype (WT or MIF-E22Q) (d). OL, oligodendrocyte; OPC, oligodendrocyte progenitor cell. e–g, UMAP plots of 34,413 neuronal nuclei (bottom right cluster in b) following subclustering and subsequent analysis labeled by neuronal family (as assigned by reference-based SeqSeek pipeline) (e), group (f) and genotype (g). h, PCA plot of neuronal snRNA-seq data pseudobulked at the sample level showing separation primarily by group (EAE versus CFA, principal component 1 on x axis) and genotype (MIF-E22Q versus WT, principal component 2 on y axis). Each dot represents the pseudobulked neuron data from one mouse, colored by genotype + group. Ellipses represent 95% CI (for groups with n > 2). i, Volcano plot highlighting DEGs when comparing pseudobulked WT EAE neurons to WT CFA. Each dot represents a gene. The x axis shows log₂FC of WT EAE neurons (n = 3) versus WT CFA neurons (n = 2). The y axis shows –log₁₀P for each gene derived from a Wald test (DESeq2). Red dots had log₂FC > 1 and P value < 1 × 10⁻⁵. Green dots met the FC cutoff, but not the P value. Blue dots met the P value cutoff, but not the FC. Gray dots met neither the P value cutoff nor the FC. NS, not significant. j, Barcode plot showing enrichment of genes in the WP gene sets ‘type II interferon signaling IFNγ’ (top, red) and ‘oxidative damage response’ (bottom, blue) in WT EAE neurons compared with WT CFA control neurons. The x axis is the signed F-statistic generated via a quasi-likelihood F-test by edgeR. Each vertical line represents a gene associated with that pathway whose position on the x axis indicates its probability of being differentially up- or downregulated. Genes at the extreme ends are more likely to be differentially expressed than those in the middle. The y axis and enrichment worm show relative enrichment computed by a moving average with tricube weights. Reported FDR values are result of the two-sided competitive gene-set test CAMERA (followed by Benjamini–Hochberg multiple-testing correction) ran on the WT EAE versus WT CFA comparison using the WP genes. k, Heatmap showing relative expression values for genes that were (1) associated with gene sets in j following pseudobulk analyses and (2) were also identified as differentially expressed by DESeq2 for WT EAE versus WT CFA (adjusted P < 0.05 and FC greater than 50%). Each row represents a gene, and each column represents a pseudobulked biological replicate. In addition, all ‘response to type II interferon’ genes in the heatmap were also identified as being differentially expressed when comparing MIF-E22Q EAE neurons to WT EAE neurons. Cell color indicates row z-score based on regularized log-transformed count values. Column coloring at the top represents genotype and group. Row coloring on the side indicates gene set with black being type II interferon signaling and gray indicating oxidative stress and redox pathway. l, Violin plots showing the expression of inflammatory genes in four different conditions (WT CFA, MIF-E22Q CFA, WT EAE and MIF-E22Q EAE) using single-nucleus data, where each column represents all biological replicates in that condition. The y axes represent expression values following library normalization (via log transformation) on a ln(n + 1) scale. m, UMAP plots colored by the expression of selected genes associated with the WP oxidative stress and redox pathway. The left column shows neurons from CFA mice (WT and MIF-E22Q). The right column shows neurons from EAE mice (WT and MIF-E22Q). Testing for statistical significance in k and P values in l were calculated using a two-tailed Wald test, and then corrected for multiple comparisons by the Benjamini–Hochberg procedure. The FDR values reported in j were also adjusted for multiple comparisons using the Benjamini–Hochberg procedure. Panel a created with BioRender.com.

Extended Data Fig. 8. snRNA transcriptomic analyses of EAE lumbar spinal cord neurons in WT and MIF-E22Q mice.

(A) Dotplot showing expression of canonical cell type markers (X-axis) in each cell type (Y-axis) shown in Fig. 5b supporting cell type assignment. (B) Stacked barplot showing proportion of nuclei in each sample identified as respective cell type, using same cell types from Fig. 5b. Each column is a single sample, and the height of each color indicates the proportion of nuclei from that sample that were identified as respective cell type. (C) Dotplot showing expression of identifying genes (X-axis) for each neuron family (Y-axis) shown in Fig. 5e supporting neuron family assignment by SeqSeek algorithm. (D) Dotplot as in (A) and (C) showing genes associated with dorsal neurons and mid/ventral neurons supporting anatomic assignment of reference-based neuron families. (E) UMAP plots showing single-nucleus neuron data from Fig. 5 labeled by predicted lamina (based on family assignment) (Ei), anatomic location (Eii), and neuronal subtype (Eiii). (F) UMAP plot showing expression of neurotransmitters associated with inhibitory (Fi,ii), excitatory (Fiii), and cholinergic (Fiv) supporting assignment in (Eiii). (G–J) Stacked column plots as in (B) showing proportion of neurons for each sample colored by lamina (G), anatomic location (H), neuron subtype (I), and neuron family (J). (K–L) PCA plot of expression data following pseudobulking of the mid/ventral excitatory (K) or inhibitory (L) neuron nuclei data at the sample level showing a similar pattern as Fig. 5h (all neurons). (M) Barcode plots showing enrichment of genes associated with Wikipathway “Type II Interferon Signaling” (top, red) and “Oxidative Damage Response” (bottom, blue) in the upregulated genes when comparing mid/ventral inhibitory pseudobulked WT EAE vs WT CFA samples. Barcode plots follow the same structure as those in Fig. 5j. Reported FDR values are results of competitive gene-set testing algorithm CAMERA for named comparison using the Wikipathway Gene Sets.

We next investigated the effect of EAE and MIF-E22Q specifically on neuronal nuclei (34,413 nuclei). We annotated neuronal subsets using the previously published SeqSeek neural network pipeline which subdivided neurons into families. These family identities map to specific spinal cord lamina, functional subset and gene expression profile⁴⁴. The SeqSeek neuronal families were manually confirmed with marker gene expression (Extended Data Fig. 8c,d). The neurons in our dataset clustered based on anatomic location. Neuronal families located in the dorsal lamina formed distinct clusters around the periphery of the uniform manifold approximation and projection (UMAP), whereas families located in the deep-dorsal (mid) and ventral lamina formed a large mixed cluster toward the center of the UMAP (Fig. 5e, Extended Data Fig. 8e(i,ii)). Neuronal subtype also drove clustering. Excitatory and inhibitory neurons within each region (dorsal, mid/ventral) clustered together (Extended Data Fig. 8e(iii),f). All biological replicates had neurons present from each lamina, family, anatomic location and subtype (Extended Data Fig. 8g–j). Transcriptional differences based on disease condition and genotype also became apparent when examining the subset of neuronal nuclei, indicating effects of EAE and MIF nuclease genetic ablation on the lumbar spinal cord neuron transcriptomes (Fig. 5f,g).

We performed pseudobulk analysis where each sample’s gene expression was summed across all neurons^45,46. Principal component analysis (PCA) of the pseudobulked neuron samples revealed clear clustering by disease condition and genotype (Fig. 5h). We then assessed differential expression using DESeq2 (ref. ⁴⁷). Using a fold change (FC) cutoff of 50% and false discovery rate-adjusted P value threshold of 0.05, 593 genes were found to be upregulated in the WT EAE compared to WT CFA, and 643 genes were downregulated in WT EAE compared to WT CFA (Fig. 5i and Supplementary Table 3).

To identify which pathways and biological processes were most altered in lumbar spinal cord neurons during EAE, we next performed competitive gene-set testing using the CAMERA algorithm⁴⁸ with the WikiPathways (WP) database⁴⁹ and Gene Ontology-Biological Process (GO-BP)^50,51 terms (Supplementary Table 4). Competitive gene-set testing revealed that pathways related to inflammation, such as ‘WP type II interferon signaling IFNγ’, were strongly upregulated (Fig. 5j, top) in neurons following EAE induction. The genes that make up these pathways include interferon response genes (Irf1), genes related to antigen presentation (H2-Aa), as well as genes which help recruit peripheral immune cells such as chemokines (Ccl5 and Ccl9) (Fig. 5k,l). Of note, Cd74, which serves as both the invariant chain in the major histocompatibility complex (MHC) class II complex as well as the receptor for soluble MIF, was one of the top upregulated genes in the WT EAE to CFA comparison. In addition to the pathways and genes associated with immune response modulation, pathways related to oxidative stress also were enriched in the genes upregulated during EAE (Fig. 5j, bottom), such as Mgst1 and Mgst2, which neutralize ROS⁵² as well as Cyba and Cybb, which contribute to ROS generation⁵³ (Fig. 5k,m). This increased expression of genes related to oxidative stress is consistent with the observed increased 8-OHdG staining described above and is an expected downstream consequence of CNS inflammation⁵⁴.

MIF-E22Q neurons exhibit increased expression of a subset of immune pathways at peak EAE

Notably, comparing the MIF-E22Q EAE to WT EAE samples revealed higher expression of a subset of inflammatory genes (Fig. 5k,l). Indeed, competitive gene-set testing of MIF-E22Q EAE versus WT EAE neurons showed several inflammatory pathways upregulated in MIF-E22Q EAE above WT EAE (Supplementary Table 4). Oxidative damage response pathways, however, were not differentially enriched in MIF-E22Q relative to WT EAE (Supplementary Table 4 and Fig. 5k).

Given the possibility that changes in proportions within the neuronal subsets could result in differentially expressed genes (DEGs) and pathways, we also performed identical pseudobulk analyses on the major neuronal subsets (dorsal excitatory, dorsal inhibitory, mid/ventral excitatory and mid/ventral inhibitory). As demonstrated in the mid/ventral inhibitory subset, we observed similar effects within the subsets (Extended Data Fig. 8k–m), suggesting that the inflammatory response is occurring in neurons throughout the spinal cord.

We next assessed whether pharmacological inhibition of MIF nuclease activity would result in the same neuronal transcriptomic changes. We performed an additional single-nucleus transcriptomic experiment from the lumbar spinal cord of mice at the peak of EAE that had been treated daily with PAANIB-1 or vehicle starting 2 days before immunization (Extended Data Fig. 9a–e). We again applied the SeqSeek algorithm to defined neuron subtypes (Extended Data Fig. 9f–h). We observed similar EAE-induced changes in the neurons of these mice as compared to the previous experiment (Extended Data Fig. 9i–k); however, PAANIB-1 had little effect on the neuronal transcriptome at this time point, unlike the MIF-E22Q mutation (Extended Data Fig. 9l–o). We reasoned that this may be explained by its partial blockade of parthanatos for less than 20 days, as compared to innate blockade in MIF-E22Q, and the lack of observed neuroprotection in PAANIB-1 mice at the peak time point (Extended Data Fig. 9o). Thus, we next chose to focus on the effects of genetic inhibition of parthanatos.

Extended Data Fig. 9. PAANIB-1 pre-treatment was not neuroprotective and did not alter neuronal transcriptome in the same manner as that of the MIF-E22Q knock-in mutation at peak EAE.

(A) Stacked column plot showing proportion of nuclei in each sample (1 column per sample) identified as respective cell type (indicated by color) after nuclei QC and doublet identification.(B) Dotplot of canonical cell type markers (on x-axis) confirming cell type assignment (y-axis) where size of dot represents what percentage of nuclei labeled that cell type express the gene in question and dot color indicating average expression within nuclei of the specified gene after normalization for per-cell library size. (C-E) UMAP plots of all high-quality nuclei from PAANIB-1 treatment experiment colored by cell type (C), group (CFA or Peak EAE) (D), and treatment (Vehicle or PAANIB-1, (E). (F–H) UMAP plots of only the neuronal nuclei colored by neuronal family (F), group (G), and treatment (H). (I) PCA plot of pseudobulked neuronal nuclei from PAANIB-1 experiment. Ellipses indicate 95% confidence for respective group. (J–O) Barcode plots showing enrichment for genes associated with respective term when comparing pseudobulked neurons from EAE mice compared to CFA controls for vehicle-treated mice (top row) and PAANIB-1 treated mice (bottom row). Reported FDR values are calculated by the competitive gene-set test CAMERA. (P) Representative images and average lumbar spinal cord ventral horns quantification for NeuN+ cells in PAANIB-1-treated EAE mice compared to vehicle-treated EAE mice at the peak EAE time point. Inset display regions quantified (pink) and presented in representative images (red box). Scale bar=50 µm. Data shown Extended Data Fig. 9P graph represents mean ± SD; each dot represents one mouse. Extended Data Fig. 9P analyzed with two-sided, unpaired t-tests.

EAE suppresses genes related to neuronal function and survival in WT but not MIF-E22Q mice

In addition to the upregulation of inflammatory genes in the neurons of the EAE mice, there were many downregulated genes in WT EAE relative to WT CFA neurons. Competitive gene-set testing revealed many of those downregulated genes were associated with pathways related to essential neuronal functions, such as the GO-BP terms for neuropeptide signaling pathway (false discovery rate (FDR) 9.7 × 10⁻⁵) and regulation of postsynaptic membrane potential (FDR 2.5 × 10⁻⁴), glutamate receptor signaling (FDR 4.7 × 10⁻⁶) and GABA signaling (FDR 5.12 × 10⁻⁵) (Fig. 6a and Supplementary Table 4). Genes associated with these terms that were downregulated during EAE, largely include neuronal function and survival gene families such as Adcyap1 (pituitary adenylate activating polypeptide 1 (PACAP)), a potent neuroprotective peptide⁵⁵, and brain-derived neurotrophic factor (Bdnf), a canonical neuron growth factor⁵⁶. Of note, these genes and gene sets were preserved in MIF-E22Q EAE (Supplementary Tables 3 and 4). Competitive gene-set testing of MIF-E22Q EAE neurons compared to WT EAE neurons confirmed the relative preservation of gene sets related to neuronal functioning and survival, including neurotransmitter signaling (Fig. 6b,c), neuropeptide signaling (Fig. 6d,e) and regulation of membrane potentials (Fig. 6f,g). Adcyap1, whose expression was almost halved in WT EAE neurons compared to WT CFA neurons (~46% decrease in WT EAE versus WT CFA, Padj = 9.1 × 10⁻¹⁰), was not substantially altered in MIF-E22Q EAE versus MIF-E22Q CFA (~13% increase in MIF-E22Q EAE versus MIF-E22Q CFA, Padj = 0.49). The effect on Bdnf was similar (WT EAE versus WT CFA: log₂FC −0.94, Padj = 9.5 × 10⁻⁹, E22Q EAE versus E22Q CFA: log₂FC 0.01, Padj = 0.98). Again, to ensure these effects were not solely driven by a change in neuronal subset proportions, we performed the same analyses in each anatomic and subtype specific subset, observing similar trends in neuronal subsets (Extended Data Fig. 10a–d). We validated a selection of these transcriptomic findings with immunofluorescent antibody staining. CD74 and MHC class I were increased in WT EAE compared to CFA neurons, and even more increased in MIF-E22Q neurons (Extended Data Fig. 10e–g). BDNF and PACAP were decreased in WT EAE neurons compared to CFA and MIF-E22Q neurons (Extended Data Fig. 10h–j). Finally, we confirmed that MIF expression is not altered by either genetic or pharmacologic MIF nuclease inhibition (Extended Data Fig. 10k,l).

Fig. 6. EAE suppresses expression of genes associated with neuronal function in the lumbar spinal cord which is attenuated by genetic ablation of MIF nuclease activity.

a, Gene-Concept network plot showing genes and processes downregulated in WT EAE versus WT CFA neurons as calculated by pseudobulk analyses of snRNA data (the same as in Fig. 5). Central (yellow) nodes are a selection of GO-BP terms related to essential neuronal function that were downregulated (using hyper-geometric analyses). Edges connect downregulated genes associated with each term. Each gene in the plot had >50% reduction with an FDR adjusted P < 0.05. Gene nodes are colored by the log₂FC of MIF-E22Q EAE neurons compared with WT EAE neurons, showing all selected genes were more highly expressed to some degree in MIF-E22Q EAE relative to WT EAE. The size of central term nodes represents the number of genes associated with that term. b, Barcode plots showing enrichment of genes associated with the GO-BP terms ‘glutamate receptor signaling pathway’ (top, red) and ‘GABA signaling pathway’ (bottom, blue) in the downregulated genes when comparing WT EAE versus WT CFA (i) and in the MIF-E22Q EAE versus MIF-E22Q CFA (ii) comparisons showing downregulation in WT but not MIF-E22Q. The x axes represent signed F-statistic as calculated by a quasi-likelihood F-test of pseudobulk expression data by edgeR. Each vertical line represents a gene associated with that pathway whose position on the x axis indicates its probability of being differentially up- or downregulated. Genes at the extreme ends are more likely to be differentially expressed than those in the middle. The y axes and enrichment worm show relative enrichment computed by a moving average with tricube weights. Reported FDR values are result of the two-sided competitive gene-set test CAMERA ran on the comparison using the GO-BP term gene sets. c, Heatmap showing relative expression for genes associated with terms in b that were differentially expressed in WT EAE versus WT CFA. Each row is a gene, and each column is a pseudobulked biological replicate. Color indicates row z-score based on regularized log-transformed count values. The top column coloring indicates which genotype (E22Q and WT) and group (EAE and CFA) each sample belongs to. Side row coloring indicates which GO-BP term gene is associated with black for glutamatergic signaling pathway and gray for the GABAergic signaling pathway. d,f, Barcode plots as in b but showing only a single pathway per plot ‘neuropeptide signaling pathway’ (d) and ‘regulation of postsynaptic membrane potential’ (f), where the top plot shows the WT EAE versus WT CFA comparison results (d(i), f(i)) and bottom shows the MIF-E22Q EAE versus MIF-E22Q CFA comparison results (d(ii),f(ii)). e,g, Heatmaps as in c but showing DEGs associated with terms ‘neuropeptide signaling pathway’ (e) and ‘regulation of postsynaptic membrane potential’ (g). h–k, Reanalysis of human MS snRNA-seq dataset initially published by Macnair et al. Neurons from GMLs in PPMS and SPMS were pseudobulked and compared with control gray matter (GM). DEGs were identified using Wald test (DESeq2) for PPMS (h) and SPMS (i). The DEGs were used for gene set overrepresentation analysis and presented as dotplots for PPMS (j) and SPMS (k). FDR values reported in b were adjusted for multiple comparisons using the Benjamini–Hochberg procedure.

Extended Data Fig. 10. Pseudobulked snRNA transcriptomic analysis of only mid/ventral inhibitory neurons shows similar pattern of neuronal function suppression as analyses of all neurons.

(A,C) Barcode plots of GO-BP terms “Neuropeptide Signaling Pathway” (A) and “Regulation of Postsynaptic Membrane Potential” (C) in the WT EAE vs WT CFA comparison of mid/ventral inhibitory neurons (top) and the same comparison in MIF-E22Q EAE vs MIF-E22Q CFA samples (bottom). Barcode plots follow same structure as Fig. 6d,f. (B,D) Heatmaps showing relative expression of genes associated with GO-BP terms “Neuropeptide Signaling Pathway” (B) and “Regulation of Postsynaptic Membrane Potential” (D) of just the mid/ventral inhibitory neurons. The heatmap structure is the same as those in Fig. 6e,g. (E–J) Validation of select genes of interest on a protein basis using immunofluorescence staining for markers of immune activation (E–G) and neuronal functioning (H–J) in neurons. (K–L) MIF expression levels in MIF-E22Q EAE vs WT EAE (genetic experiment) and vehicle EAE vs PAANIB-1 EAE (treatment experiment). N = 5 per group (Extended Data Fig. 10F,G,I,J); N = 3 per group (Extended Data Fig. 10K); N = 5 vehicle, N = 6 PAANIB-1 (Extended Data Fig. 10L). Error bars in Extended Data Fig. 10F,G,I,J represents mean ± SD, Extended Data Fig. 10K,L represents mean ± SE; each dot represents one mouse. Extended Data Fig. 10F,G,I,J analyzed with one-way ANOVA followed by Tukey’s post hoc test. Extended Data Fig. 10K,L analyzed with multiple two-tailed Student’s T-tests.

We next sought to compare neuronal gene changes seen in EAE with MS. We reanalyzed previously published single-nucleus RNA sequencing (snRNA-seq) data⁵⁷ from gray matter lesions (GMLs) in progressive MS. We pseudobulked the neurons from either primary progressive MS (PPMS) or secondary progressive MS (SPMS) cases by sample and identified DEGs using a FC cutoff of 50% and false discovery rate-adjusted P value threshold of 0.05. We found 84 upregulated and 83 downregulated genes in PPMS versus control (Fig. 6h and Supplementary Table 5) and 55 upregulated and 65 downregulated genes in SPMS versus control (Fig. 6i and Supplementary Table 6). While we did not see differential expression of genes related to acute inflammation (which is not surprising given that the initial inflammation occurred years or decades earlier), gene sets related to basic neuronal functions, including ion transport, maintenance of membrane potential and neurotransmitter signaling were consistently downregulated in MS neurons (Fig. 6j,k).

Discussion

Here, we establish parthanatos as a new cell death pathway in immune-mediated neuron death. We present evidence that the trigger for parthanatos, DNA damage, increases and persists during peak and chronic EAE. Progressive neuron loss over the course of EAE is paired with hallmarks of parthanatos such as oxidative stress, large-scale DNA fragmentation, cytoplasmic PAR, AIF–MIF binding and nuclear MIF translocation. Genetic and pharmacologic blockade of parthanatos was neuroprotective in EAE. Transcriptomic pathway analysis revealed preserved expression of gene sets related to a variety of essential neuronal functional and survival pathways, including neuropeptide signaling, maintenance of membrane potential and various signaling pathways in MIF-E22Q neurons relative to WT during peak EAE.

Several forms of cell death are proposed to contribute to neurodegeneration in MS, including apoptosis, ferroptosis and necroptosis^58–62. Earlier studies from MS patient samples and EAE mouse tissues found PARP1 overactivation, AIF nuclear translocation, nuclear shrinkage and caspase-independent DNA fragmentation, features that would be consistent with parthanatos, though these studies pre-date its discovery^24,63–65. Furthermore, techniques classically used to measure apoptosis, such as TUNEL staining or counting pyknotic nuclei, are nonspecific and could detect parthanatos-mediated DNA fragmentation^66,67. Notably, cell death pathways may not occur in isolation, especially in the complex neuroinflammatory milieu of MS or EAE. Parthanatos could act as a precursor to, or in parallel with, other cell death pathways^{60,61,67–69}. In support of this possibility, while broadly protective in the acute setting, there was substantial neurodegeneration in chronic EAE despite genetic blockade of parthanatos. Components of the parthanatos cascade also may interact with other pathways, having broad implications for inflammation, cell survival or death, and disease outcomes. MIF nuclease, for example, could be a source of host DNA fragments that activate innate nucleic acid sensors such as cGas-STING, AIM2 or TLR9 and promote inflammation or pyroptosis^62,70. Overall, while various cell death pathways may occur in EAE, we propose that DNA damage-induced parthanatos occurs early and has a nonredundant impact on neuron death. In future studies of human tissue, specific markers such as cleaved caspase-3 and MIF nuclear translocation should be used to distinguish forms of cell death.

Our findings of increased MHC class II transcript expression in neurons during EAE represent a departure from the conventional understanding of MHC class II only being expressed by professional antigen-presenting cells⁷¹; however, recent studies have demonstrated MHC class II expression on other cells, including astrocytes, oligodendrocytes and oligodendrocyte progenitor cells, in inflammatory conditions^72–75. Still, reports of neuronal MHC class II expression are rare. Neural stem cells upregulate MHC class II in response to IFNγ⁷⁵, and neurons in the dorsal root ganglion express MHC class II after peripheral nerve transection⁷⁶. As compared to other studies in EAE, our single cell experiment was optimized to collect neuronal transcriptomic data by using single-nucleus sequencing and our choice to study the spinal cord. Further studies are imperative to confirm and expand upon this observation, assessing the protein levels, kinetics and consequences of neuronal MHC class II expression.

Of note, we found that MIF-E22Q neurons had increased inflammatory gene pathway expression, particularly IFNγ-related pathways, compared to WT neurons in EAE. Neurons protected in MIF-E22Q EAE likely drive these genotype-level differences, and perhaps they reflect a transitional senescence-like state in the recovering neurons. Inflammatory signaling may also directly affect neuronal recovery in cells protected from parthanatos. For example, IFNγ signaling is known to promote neuronal survival through ERK1/2 activation and has a multitude of effects supporting neuronal connectivity and conduction^77–79. IFNγ also induces MHC class I expression, which may protect cells from natural killer-cell mediated cytotoxicity⁸⁰. Various gene pathways related to essential neuronal functioning were downregulated in WT neurons, but preserved in MIF-E22Q neurons, during EAE. Several of these genes have been independently associated with neuroprotection; for example, PACAP was recently proposed as a new therapeutic target for neurodegeneration^81–83.

While susceptibility to MS has predominantly been linked to immune gene variants, a recent large-scale genome-wide association study identified that the variant rs10191329 in the DYSF-ZNF638 locus was implicated in CNS resilience and potentially neurocognitive reserve in MS⁸⁴, underscoring the critical role of neuroprotection in MS clinical outcomes. While this is an intergenic variant, it is closest to the dysferilin (DYSF) promoter and QTL data⁸⁴ suggest that it may play a role in regulating its expression. In our neuronal single-nucleus data, we found that Dysf was upregulated in WT EAE, but not in MIF-E22Q EAE (Supplementary Table 3). Dysferilin is thought to play an essential role in plasma membrane repair⁸⁵, and consistent with this role, we found other genes (Arl8b, Chmp7, Myh9 and Myh10) associated with membrane repair to also be differentially expressed in neurons at peak inflammation of EAE. One possible explanation for this observation is that during EAE, dying neurons are attempting to repair themselves, whereas the surviving MIF-E22Q neurons do not upregulate repair pathways to the same extent. We also carried out the same transcriptomic experiments in mice treated with an MIF nuclease inhibitor (PAANIB-1) at peak EAE. Control neurons had similar transcriptional changes in response to neuroinflammation; however, there was no substantial impact of PAANIB-1 treatment on neurons. This result aligns with the less potent impact on neuroprotection and behavioral recovery in PAANIB-1 relative to MIF-E22Q. Indeed, our subsequent measurement of neuronal survival in PAANIB-1-treated mice at peak EAE showed no protection at peak disease, whereas there was a protective effect at the chronic time point.

One limitation of our study is the lack of cell-specific MIF nuclease blockage. While we strove to confirm that no confounding impacts on immune cell infiltration, cytokine production, reactive gliosis or myelination were present in MIF-E22Q, global parthanatos inhibition may have had other indirect effects on neuronal survival. Furthermore, the myelin oligodendrocyte glycoprotein 35–55 (MOG_35–55) EAE mouse model used in this study recapitulates only a limited set of pathologies present in MS. Nonetheless, we establish that acute inflammation is a trigger for parthanatos in neurons, which may have relevance for MS-specific findings such as gray matter atrophy, axonal pathology, RGC layer thinning and elevated levels of serum neurofilament from disease onset^86–89.

Our data together demonstrate that neurons accumulate DNA damage and undergo parthanatos in acute inflammation. Our transcriptomic findings combined with the improved EAE motor scores suggest that the protected neurons remain functional. Ultimately, these findings, paired with new discoveries from other laboratories about neuronal susceptibility in MS²⁷, could have direct relevance to MS and related neuroinflammatory disorders, in which current therapies predominantly reduce peripheral immune cell activity without directly promoting neuroprotection.

Methods

Animals

C57BL/6J (RRID:IMSR_JAX:000664) WT mice were obtained from The Jackson Laboratory. MIF-E22Q knock-in mice were generated as previously described^14,16 on the same C57BL/6J WT mice background. MIF knockout(RRID:IMSR_JAX:003830) mice were obtained from The Jackson Laboratory. All animals were housed in temperature- and humidity-controlled rooms, maintained on a fixed light–dark cycle, and age- and sex-matched in each experiment. Adult mice tissues were collected when mice were 12–17 weeks old. All procedures on animals were approved by the Johns Hopkins Institutional Animal Care and Use Committee.

Experimental autoimmune encephalomyelitis

For EAE experiments, mice were randomized into control (CFA only) and experimental (EAE) groups. EAE was induced in mice at 10–12 weeks old. Male and female mice were used for all experiments, other than the transcriptomic analyses that only probed female mice to eliminate any possible sex bias. EAE was induced via active immunization; in brief, an emulsion was prepared by mixing equal volumes of MOG_35–55 peptide (2 mg ml⁻¹ in PBS) and CFA (8 mg ml⁻¹ Mycobacterium tuberculosis in incomplete Freund’s adjuvant) with syringes attached to a stopcock for 10 min. A total 75 μl of emulsion were injected subcutaneously in each side of the abdomen for a total of 150 µg MOG_33–55 peptide per mouse. Pertussis toxin (500 ng) was injected intraperitoneally on the day of immunization and 2 days later. Animals were labeled with a unique identifier and two blinded experimenters weighed and scored the mice each day based on the Hooke Laboratory Scoring Rubric of Experimental Autoimmune Encephalomyelitis Mice criteria: 0.5, tip of tail limp; 1, limp tail; 1.5, limp tail and wobbly gait; 2, limp tail and hindlimb weakness; 2.5, limp tail and dragging of one or both hindlimbs; 3, limp tail and complete hindlimb paralysis; 3.5, limp tail and complete hindlimb paralysis and unable to right itself or flat hindquarters; 4, complete hind and partial front limb paralysis with minimal movement around the cage; and 5, death⁹⁰. Mice were killed at peak or chronic EAE stages.

Flow cytometry

Mice were killed with isoflurane then cardiac perfused with chilled HBSS. Spinal cords were flushed from the columns using hydrostatic pressure then minced into fine pieces with a razor and transferred to digestion buffer (1× EBSS, 22.5 mM D(+)-glucose, 2.2 g l⁻¹ NaHCO3, 3 mM CaCl₂ in water at pH 7.4) with 2 mg ml⁻¹ collagenase IV (Worthington Biochemicals, cat. no. LS004188) and 100 U ml⁻¹ DNase I (Worthington Biochemicals, cat. no. LS002007). Tissue was dissociated by slow rotation at 37 °C for 25 min with manual trituration using fire-pulled Pasteur pipettes after 10 min and 15 min. Dissociated tissue was passed over 100 μM cell strainer and one drop of counting beads (Thermo Fisher, cat. no. C36950) were spiked into each sample at this time. Tissue was washed with 10 ml HBSS, centrifuged at 300g for 10 min. The supernatant was then removed and the pellet resuspended in 7 ml PBS. A total of 3 ml isotonic Percoll (90% Percoll, 10% 10× PBS) was added for a final concentration of 30% isotonic Percoll. Samples were mixed by pipetting, then centrifuged at 500g for 25 min with no brake. Supernatant was removed and the cell pellet was washed in PBS and centrifuged again at 300g for 10 min with the brake on. Supernatant was removed and cells were transferred to a 96-well V-bottom plate then pelleted again at 500g for 5 min. Supernatant was decanted and cells were resuspended in PBS then split into two aliquots (one for a T cell panel and one for a myeloid panel). The T cell aliquot was pelleted again then resuspended in 200 μl complete IMDM (10% FBS, 100 U ml⁻¹ penicillin–streptomycin, 2 mM GlutaMAX (Thermo Fisher, cat. no. 35050061), 55 μM 2-mercaptoethanol) with a 1:500 dilution of Cell Stimulation Cocktail with Protein Transport (Thermo Fisher, cat. no. 00-4975-03). The T cell panel was incubated for 5 h at 37 °C to allow for cytokine production. The myeloid panel was processed immediately. For both panels, staining began with Zombie Fixable NIR Staining (1:2,000, BioLegend, cat. no. 423106) and TruStain FcX Plus Fc block (1:200, BioLegend, cat. no. 163404) in 200 μl PBS for 15 min at room temperature. Cells were then pelleted and resuspended in surface-stain antibodies (listed below) in staining buffer (2% FBS, 2 mM EDTA, 5% Super Bright Complete Staining Buffer (Thermo Fisher, cat. no. SB-4401-75) in PBS) and incubated for 30 min at room temperature then washed in FACS buffer (2% FBS, 2 mM EDTA in PBS). The myeloid panel was fixed in 200 μl of Intracellular Fixation Buffer (Thermo Fisher, cat.no. 00-8222-49) mixed 1:1 with FACS buffer and incubated for 30 min at 4 °C. The T cell panel was fixed in 200 μl FoxP3 Fixation/Permeabilization Buffer (Thermo Fisher, cat. no. 00-5523-00) for 20 min at room temperature then washed in Permeabilization Buffer (Thermo Fisher, cat. no. 00-8333-56) followed by intracellular staining (ICS) with cytokine and transcription factor antibodies for 1 h at room temperature in permeabilization buffer. Next, cells were washed in 200 μl permeabilization buffer, washed again in 200 μl FACS buffer, then resuspended in 100 μl FACS buffer and ran on a three-laser (violet/blue/red) Aurora Spectral Cytometer (Cytek Biosciences). The myeloid panel was run on the same day, while the T cell panel was left at 4 °C overnight then ran the next morning. The following antibodies were used: CD45 BV605 (RRID:AB_2562342, 1:400 dilution), CD11b BV510 (RRID:AB_2561390, 1:800 dilution), CD11b Pacblue (RRID:AB_755986, 1:800 dilution), CD11c BV570 (RRID:AB_10900261, 1:200 dilution), IA/IE PerCP (RRID:AB_2191073, 1:200 dilution), Ly6C FITC (RRID:AB_1186135, 1:400 dilution), Ly6G BV785 (RRID:AB_2566317, 1:400 dilution), Clec12a APC (RRID:AB_2564264, 1:100 dilution), CD4 PE/Cy7 (RRID:AB_469578, 1:200 dilution), CD8a ef450 (RRID:AB_1272198, 1:200 dilution), CD25 PE/ef610 (RRID:AB_2574542, 1:200 dilution), FoxP3 PE (RRID:AB_465936, 1:200 dilution), TCRβ PerCP/Cy5.5 (RRID:AB_1575173, 1:150 dilution), IFNγ FITC (RRID:AB_315400, 1:200 dilution), IL-17A APC (RRID:AB_536017, 1:200 dilution) and GM-CSF PE-eFluor 610 (RRID:AB_2814428, 1:200 dilution). Raw data were unmixed in SpectroFlow (v.2.2.0.3, Cytek Biosciences) software with single stained beads (UltraComp eBeads Plus Compensation Beads, Thermo Fisher, cat. no. 01-2222-42) for antibodies as well as unstained and single stained cells for permeability marker. Flow data were analyzed using FlowJo software (BD). Fluorescence-minus-one staining controls were used to help guide manual gating and analysis.

In vitro T cell proliferation assay

CD4⁺ T cells were collected from cervical and axillary lymph nodes of WT and MIF-E22Q animals for an in vitro proliferation assay. Lymph nodes were dissected, passed through a 70-µm cell strainer, and CD4⁺ T cells were purified via a negative selection magnetic bead kit following the manufacturer’s recommendation (BioLegend, cat. no. 480005). T cells were counted and labeled with CellTrace Violet (Cell Signaling, cat. no. 48444). Cells were then plated at a density of 500,000 cells per ml into a 96-well plate. Plates were precoated with anti-CD3 and soluble anti-CD28 was added to each well. After 72h, cells were collected and analyzed via flow cytometry. Unstimulated control was used to define baseline cell trace violet brightness, and the number of cell divisions was measured by dye dilution.

Tissue preparation and immunostaining

Mice were anesthetized with isoflurane and intracardially perfused with PBS followed by 4% paraformaldehyde (PFA). Eyes and spinal cords were dissected, post-fixed in 4% PFA for 4 h (eyes) or 48 h (spinal cords) and transferred to 30% sucrose for at least 48 h.

Each lumbar spinal cord was identified as the caudal 14 mm of the full spinal cord (typically about 65 mm long) and dissected with the aid of a spinal cord matrix (Electron Microscopy Sciences, cat. no. 69085-CS). The lumbar spinal cord was then embedded in OCT and sectioned in 20-µm transverse slices. Tissue sections were collected every 500-µm intervals onto a single glass slide, for a total of about 25 sections per slide (Extended Data Fig. 1a). Before staining, antigen retrieval was performed by incubating twice in sodium citrate solution (0.01% sodium citrate, pH 6.0, boiled immediately before the incubation) for 10 min. Slices were blocked for 1 h at room temperature in blocking buffer (5% normal donkey serum and 0.1% Tween-20 in PBS) followed by overnight incubation in primary antibody at 4 °C. The following primary antibodies were used: anti-PAR binding reagent (RRID:AB_2665467, 1:300 dilution), mouse anti-HuC/HuD (RRID:AB_221448, 1:250 dilution), rabbit anti-NeuN (RRID:AB_10807945, 1:1,000 dilution), chicken anti-NeuN (RRID:AB_11205760, 1:1,000 dilution), chicken anti-NeuN (RRID:AB_2937041, 1:1,200 dilution), rabbit anti-MIF (RRID:AB_10709860, 1:500 dilution), mouse anti-γH2AX (Ser139) (RRID:AB_559491, 1:500 dilution), mouse anti-8-OHdG (RRID:AB_1857195; 1:500 dilution), rabbit anti-53BP1 (RRID:AB_2890610; 1:500 dilution), rabbit anti-CD74 (RRID:AB_2924345, 1:100 dilution), mouse anti-MHC class I (RRID:AB_302029, 1:100 dilution), mouse anti-PACAP (RRID:AB_2897693, 1:750 dilution), and rabbit anti-BDNF (RRID:AB_2792869, 1:500 dilution). Slices were then washed and incubated for 1 h at room temperature with the appropriate Alexa-conjugated secondary antibodies (1:1,000 dilution, donkey anti-chicken/mouse/rabbit Alexa-488/555/647, Thermo Fisher Scientific). Slides were counterstained with Hoechst 33342 (RRID:AB_3675235, 1:10,000 dilution) and mounted with Prolong Gold antifade mounting medium (Thermo Fisher Scientific, cat. no. 18606-20).

Fixed eyes were embedded in OCT and three 16-µm-thick transverse retina and optic nerve sections were collected onto glass slides. Before staining, slides were warmed on a slide warmer at 50 °C for 10 min. Antigen retrieval was performed as described above. Slices were blocked for 1 h at room temperature in blocking buffer followed by overnight incubation in primary antibody at 4 °C. The following antibodies were used for retina immunostaining: mouse anti-γH2AX (Ser139), mouse anti-8-OHdG and rabbit anti-γSyn (custom via Covance, 1:200 dilution). The following antibodies were used for optic nerve staining: chicken anti-Iba1 (RRID:AB_2923485, 1:1,000), rabbit anti-Iba1 (RRID:AB_2735228, 1:1,000), rabbit anti-TMEM119 (RRID:AB_2800343, 1:100), rat anti-CD68 (RRID:AB_2572857, 1:100), goat anti-Sox9 (RRID:AB_2194160, 1:500), and chicken anti-GFAP (RRID:AB_2313547, 1:1,000). Alexa-conjugated secondary antibody staining was performed as described above.

Whole retinas were dissected and washed in permeabilization buffer (3% Triton X-100 in PBS) three times for 1 h each. Retinas were then incubated in 5% normal donkey serum and 1% Triton X-100 in PBS for 2 h at room temperature. Mouse anti-Brn3a (RRID:AB_2737037, 1:1,000 dilution) was added directly into the wells and incubated for 48 h at 4 °C. Samples were then washed three times for 1 h followed by secondary antibody staining in 5% normal donkey serum and 1% Triton X-100 in PBS at 4 °C overnight. After three 1-h washes, whole retinas were mounted with Prolong Gold antifade mounting medium.

Image acquisition and quantification

Epifluorescent and confocal images were acquired at ×20 magnification with an Axio Observer.Z1 microscope (Zeiss) and an LSM900 (Zeiss) using Zen v.3.1 software.

Lumbar and dorsal spinal cord cell counts

About 20 intact sections per mouse were utilized for quantifications. For every section on each slide, two regions of interest (ROIs) were defined. The dorsal ROI corresponded to the gray matter area where the sensory nuclei (somatic neuronal region) are located, corresponding to laminas I/II. The ventral ROI was defined as the gray matter region ventral to the central canal. Within each ROI, ventral and dorsal horn NeuN⁺Hoechst⁺ or HuC/HuD⁺Hoechst⁺ cells were automatically identified in ImageJ (v.2.9.0) with a priori-defined parameters (eight-bit image, threshold set to 6–10% granularity, circularity ≥ 0.2 and pixel size ≥ 0.001 µm² per particle) and each cell-count image was confirmed by manual review. Cell density was calculated as cell number per unit area, and the mean cell density (cells per mm²) for each sample was recorded. Iba1+ microglia and Sox9+GFAP+ astrocyte densities were quantified the same way.

DNA damage or oxidative stress in neurons

Spinal cord tissue sections were stained for γH2AX (DNA damage), NeuN, and Hoechst or 8-OHdG (oxidative stress), NeuN and Hoechst. Retina tissue sections were stained for γH2AX, γSyn and Hoechst or 8-OHdG, γSyn and Hoechst. For each image, a threshold was applied based on the background fluorescence of γH2AX and 8-OHdG, respectively, in CFA samples. γH2AX⁺NeuN⁺, 8-OHdG⁺NeuN⁺ (spinal cord) γH2AX⁺γSyn⁺ (retina) cells and 8-OHdG intensity (retina) were analyzed using ImageJ and each count was confirmed manually. For the spinal cord, all intact sections (about 20) per slide were averaged. For cross-section of the retina, three equally spaced sections were analyzed and averaged per mouse.

Fluorescence intensity by area analysis

Lumbar spinal cord sections were stained for PAR, NeuN and Hoechst or MIF, NeuN and Hoechst or MHC class I, CD74, NeuN and Hoechst or BDNF, PACAP, NeuN and Hoechst followed by slide imaging. ImageJ was used to draw an ROI based on the Hoechst signal (nucleus) and a second ROI was drawn based on the NeuN signal (total). The integrated density (ID) of PAR was measured in the nucleus ROI and in the total ROI. Cytosolic PAR ID was calculated by subtracting the nuclear PAR from the total PAR signal. MIF ID was measured in the nuclear ROI. Total neuronal PAR, MHC class I, CD74, BDNF and PACAP IDs were measured using the NeuN signal. For each spinal cord section, all ventral horn neurons were analyzed across three equally spaced sections (2.5 mm apart) per mouse were averaged. Mean fluorescence intensity (MFI) was calculated in ImageJ for microglia and astrocyte markers in each of the full optic nerve sections.

Retinal ganglion cell counts

For quantification of RGCs in the whole retina, a tiled image of the whole-mount retina was acquired. Brn3a⁺ cells were counted as previously described using a custom MATLAB algorithm³⁷. In brief, 12 ROIs of a known area, at standardized distances from the optic disk, were selected and had all RGCs within the ROI counted. The RGC density was calculated by averaging the density of the 12 regions.

Human tissue

Postmortem human brain sections from one 41-year-old male who lived with MS and one age-matched control were obtained from the JHU pathology Department. Cortical gray matter areas from lesioned areas or non-neurological disease controls were deparaffinized and antigen retrieval was performed by boiling the samples in Trilogy (920P-05, Sigma-Aldrich). Brain slices were blocked for 1 h at room temperature with 5% normal goat serum in TBS. Immediately after blocking, the samples were incubated with the following antibodies in 0.05% Triton X-100, 5% normal goat serum in TBS: chicken anti-NeuN (1:100 dilution); human anti-PAR (custom via Dawson Laboratory¹⁷; 1:250 dilution) overnight at 4 °C. After washing, slices were incubated with the appropriate Alexa-conjugated goat anti-chicken or anti-human secondary antibodies. After a 1-h incubation at room temperature, samples were counterstained with Hoechst, washed with TBS and mounted using Prolong Glass (Invitrogen). Brain slices were imaged at ×20 in a LSM900 confocal microscope and three different sections per brain were quantified from the control case (gray matter cortex) and MS (gray matter cortical lesion) patient tissues.

FluoroMyelin quantification

To label myelin, five equally spaced free-floating spinal cord sections were permeabilized for 20 min in PBS containing 0.1% Triton X-100, incubated at room temperature for 20 min in PBS containing FluoroMyelin Red Fluorescent Myelin Stain (Invitrogen, cat. no. F34652), rinsed three times for 10 min in PBS, mounted on glass slides and cover slipped with Fluormount-GTM mounting medium (Invitrogen, cat. no. 00-4958-02). The ventral horn white matter was then quantified using a percentage area calculation.

Western blot

Spinal cords from WT mice were isolated at 5, 10, 15, 20 and 25 days after EAE induction. Immediately after dissection, the whole spinal cord was flash frozen and stored in liquid nitrogen. Tissue was homogenized in RIPA buffer (50 mM Tris, pH 7.5, 150 mM NaCl, 1% IGEPAL CA-630, 0.5% sodium deoxycholate and 0.1% SDS and supplemented with Complete Protease Inhibitor Cocktail and PhosSTOP (Roche, cat. no. 4906845001)) using a Dounce homogenizer. Primary cortical neurons were lysed in RIPA buffer supplemented with protease and phosphatase inhibitors after treatment with 150 μM H₂O₂ and lysates were stored at −80 °C. Then 30 µg of protein were separated by electrophoresis in 4–20% PAA gels and transferred overnight to Immobilon-FL membranes (Millipore, cat. no. IPFL00005). After blocking with Intercept Blocking Buffer (LI-COR Biotech, cat. no. 927-60001) membranes were incubated with the following primary antibodies: human anti-PAR (custom via Dawson Laboratory; 1:2,500 dilution), rabbit anti-NeuN (RRID:AB_10807945, 1:1,000 dilution), mouse anti-γH2A.X (Ser139) (RRID:AB_559491, 1:1,000 dilution) and mouse anti-actin (RRID:AB_2242334, 1:5,000 dilution). NeuN and γH2AX western blots were incubated with the appropriate IRDye secondary antibody (1:10,000 dilution, LI-COR Biotech) for 1 h at room temperature, and PAR western blots were incubated with goat anti-human HRP (RRID:AB_10673462, 1:2,000 dilution) for 1 h at room temperature. The infrared signal was acquired using an Odyssey M instrument (LI-COR Biotech) and the chemiluminescent signal was acquired using an Amersham Imager 600 (Cytiva). Signal intensity of the target proteins was normalized to the intensity of β-actin. Signals were quantified with Empiria Studio Software (LI-COR Biotech).

Proximity ligation assay

To investigate the interaction between MIF and AIF in spinal cord tissue from control and EAE mice, we used the NaveniFlex Tissue MR Atto647N PLA kit (NaveniFlex, cat. no. NT.MR.100.Atto). The 20-µm-thick lumbar spinal cord sections were subjected to heat-induced antigen retrieval using antigen retrieval buffer (eBioscience, cat. no. 00-4955-58) and permeabilized with 0.3% Triton X-100 in PBS. The interaction between MIF and AIF was detected using MIF (RRID:AB_10709860, 1:500 dilution) and AIF (RRID:AB_626654, 1:250 dilution) antibodies. NeuN was included in the primary antibody incubation to identify neurons. Blocking, primary and secondary antibody incubations, and PLA reactions were performed following the manufacturer’s instructions. Negative controls were included to validate the specificity of the PLA signal: (1) sections without primary antibodies and (2) sections incubated with only mouse anti-AIF. Two sections from each group per experiment were processed for each negative control. Following the PLA reaction and Hoechst nuclear staining, sections were mounted. The z-stack images (20 µm total thickness, 1-µm step-size) were acquired using a Zeiss LSM900 confocal microscope. Ten FOV images were captured per lumbar spinal cord section from each group. A minimum of five sections were used per mouse. Image processing and quantification were performed using Arivis Vision 4D (v.4.1.2) software. Using an intensity threshold segmentation algorithm, PLA signals were identified and quantified after removing imaging artifacts. Nuclei (Hoechst) and neurons (NeuN) were segmented using the ‘Detect cells or particles’ pipeline. The number of PLA dots per FOV, the number of neurons and the number of nuclei exhibiting PLA signals were quantified.

DNA fragmentation assay

CFA and day 16 EAE (average EAE motor score of ~2.5) mice were anesthetized with isoflurane and intracardially perfused with PBS. Lumbar spinal cords were immediately dissected and flash frozen in liquid nitrogen. Tissue was lysed by repetitive pipetting in lysis buffer (100 mM NaCl, 10 mM Tris-HCl, pH 8.0, 25 mM EDTA and 0.5% SDS) supplemented with 0.2 mg ml⁻¹ proteinase K (Roche Diagnostics, cat. no. 3115887001). Samples were incubated overnight at 55 °C and additionally kept at 85 °C for 45 min (this step inactivates proteinase K). The fragmented genomic DNA was then separated in a 1.2% agarose (pulsed-field certified agarose, BIO-RAD, cat. no. 1620137) gel in 0.5× TBE buffer with an initial switch time of 1.5 s and a final switch time of 3.5 s for 12 h at 6 V cm⁻¹ as previously described¹⁶. Finally, the gel was stained with 0.5 mg ml⁻¹ ethidium bromide for 2 h. UV light was used to visualize the gel. Noncleaved genomic DNA was quantified as the largest kb DNA band and normalized such that the average CFA noncleaved genomic DNA was equal to 100%.

Cell viability assay

Embryonic (E17) cortical neurons from WT and MIF-E22Q embryos were cultured as previously described¹⁶. Neurons were treated with 150 µM hydrogen peroxide (Fisher Scientific, cat. no. 7722-84-1) or vehicle solution. Twenty-four hours after treatment, cells were immediately incubated with alamarBlue Cell Viability Reagent (Bio-Rad, cat. no. BUF012A) to assess their viability. The alamarBlue fluorescence was measured using an excitation wavelength of 570 nm and an emission wavelength of 600 nm.

Pharmacological compound and gavage administration

PAANIB-1 was prepared as previously described¹⁶ and resuspended in vehicle (1% Tween 80, 10% polyethylene glycol and 1% hydroxy propyl methyl cellulose). PAANIB-1 was administered daily at 10 mg kg⁻¹ via oral gavage (200 µl per dose) using 20-gauge needles (Kent Scientific, cat. no. FNC-22-1.5-2). For all experiments, mice were separated into different cages based on treatment groups to prevent cross-contamination. In prevention paradigm experiments, vehicle- and PAANIB-1-treated groups were separated into different cages from the experiment’s outset. In treatment paradigm experiments, at the onset of symptoms, mice were separated into two groups with similar average motor scores and randomly assigned to receive PAANIB-1 or vehicle.

Single-nucleus transcriptomic library preparation

Two single-nucleus transcriptomic experiments are described in this work. For the genetic MIF nuclease inhibition experiment (Figs. 5 and 6 and Extended Data Figs. 8 and 10), lumbar spinal cord from CFA controls, WT, MIF-E22Q female mice was collected and flash frozen in liquid nitrogen after PBS perfusion at day 16 of EAE. Tissue was later retrieved and moved into nuclei extraction buffer (Miltenyi, 130-128-024) with 0.2 U μl⁻¹ RNasin Plus RNase Inhibitor (Promega, N2615). Tissue was lysed in a C tube (Miltenyi, cat. no. 130-093-237) on a gentleMACS automated dissociator (Miltenyi, cat. no. 130-093-235) using the 4c_nuclei_1 program with a gentleMACS Octo Cooler (Miltenyi, cat. no. 130-130-533) placed over the C tube to keep the tissue cold during lysis. After dissociation was complete, lysed tissue was filtered over a 100 μM SmartStrainer (Miltenyi, cat. no. 130-098-463) into a prechilled 5-ml conical tube. The filter was then washed with 2 ml Nuclei Extraction Buffer. The lysed tissue was centrifuged at 400g for 5 min and supernatant removed. The pellet was resuspended in 1 ml nuclei separation buffer (NSB; 0.04% bovine serum albumin, 14% Nuclei Extraction Buffer in PBS with 40 U ml⁻¹ RNasin RNAse inhibitor). Tissue was dissociated across three batches on the same day with even representation of genotype and time point across the batches. Following tissue lysis, nuclei were manually counted on a hemocytometer with 5 μg ml⁻¹ acridine orange. Nuclei concentration was adjusted to 2 × 10⁶ per ml for samples that had greater than 2 × 10⁶ nuclei with additional NSB. Anti-nucleus microbeads (Miltenyi, cat. no. 130-132-997) were then added at a ratio of 50 μl beads per 1 × 10⁶ nuclei and pipette-mixed then incubated at 4 °C for 15 min. Next, 2 ml NSB was added to each sample and samples were applied to LS Columns (Miltenyi, cat. no. 130-042-401) in QuadroMACS magnet (Miltenyi, cat. no. 130-090-976). Columns with labeled nuclei were washed twice with 1 ml NSB and then removed from magnet and eluted in 1 ml NSB. Nuclei were positively selected across two batches on the same day with equal representation of time point and genotype in both batches. Column separation was carried out in a cold room. Positively selected nuclei were then centrifuged at 500g for 5 min and supernatant was removed. Nuclei were processed for single-nucleus transcriptomics using 10x Genomics Flex reagents and protocols. The nuclei pellet was resuspended in 1 ml Fix/Perm buffer (10% Conc. Fix/Perm buffer (10x Genomics, cat. no. PN-2000517), 4% formaldehyde in water) and then incubated overnight at 4 °C. The next morning, fixed nuclei were retrieved and centrifuged at 850g for 5 min at room temperature. The supernatant was removed and the pellet was resuspended in 1 ml quenching buffer (12.5% Conc. Quenching Buffer (10x Genomics, cat. no. PN-2000516) in water) and pipette-mixed five times. Nuclei were counted again with 5 μg ml⁻¹ acridine orange on a Countess 3 Hemocytometer (Thermo Fisher, cat. no. A49865). Quenched nuclei were centrifuged at 850g for 5 min at room temperature and supernatant was removed. Each pellet was then resuspended in 80 μl Hyb Mix (12.5% Enhancer (10x Genomics PN-2000482) in Hyb Buffer B (10x Genomics, cat. no. 2000485)) and 20 μl barcoded probes (Mouse WTA Probes, 10x Genomics, cat. no. PN-2000703-2000718) were added to each sample. Samples were pipette-mixed ten times and then incubated for 20 h at 42 °C. Samples were diluted in 900 μl Post-Hyb Wash Buffer (5% Conc. Post-Hyb Buffer (10x Genomics, cat. no. 2000533), 5% Enhancer in water) and pipette-mixed five times then incubated at 42 °C for 10 min. Samples were centrifuged at 850g for 5 min and supernatant removed. Pellets were resuspended in 500 μl post-hyb wash buffer and the previous 42 °C incubation and wash were repeated twice more. Following final centrifugation, nuclei were resuspended in 500 μl Post-Hyb Resuspension Buffer (5% Conc. Post-Hyb Buffer in water), pipette-mixed 20 times then passed over a 30 μM Pre-Separation Filter (Miltenyi, cat. no. 130-041-407). Nuclei were again counted with 5 μg ml⁻¹ acridine orange. Next, all samples were equally pooled based on these counts into a single tube and again filtered across a 30 μm Pre-Separation Filter and centrifuged at 850g for 5 min.The supernatant was removed, the pellet was resuspended in 150 μl Post-Hyb Resuspension Buffer, and nuclei were counted. Approximately 200,000 nuclei (expected recovery of 128,000) were loaded onto a single lane of a Chip Q (10x Genomics, cat. no. PN-2000518) according to manufacturer’s recommendations. Nuclei were partitioned into gel beads in emulsion using a Chromium iX device (10x Genomics, cat. no. 1000331). A single-nucleus transcriptomic library was then generated according to manufacturer’s recommendation (Chromium Fixed RNA Profiling Reagent Kits for Multiplexed Samples, cat. no. CG000527 Rev E) using nine amplification cycles during Sample Index PCR. The library was then sequenced on a NovaSeq 6000 S4 flow cell in 1 lane (150 bp paired end) with 10% PhiX spike-in. For the pharmacologic (PAANIB-1) inhibition experiment (Extended Data Fig. 9), the experiment was performed as described for the genetic dataset above, except for the following differences: lumbar spinal cord was collected from vehicle CFA controls (n = 2 female), PAANIB-1 treated CFA controls (n = 2 female), vehicle EAE mice (n = 3 female), and PAANIB-1 treated EAE mice (n = 3 female) 16 days post-immunization, the nuclei were prepared across two equal batches on the same day, and the library was sequenced on one lane of a NovaSeq X Plus 10B flow cell in 28 × 10 × 10 × 90 configuration with 10% PhiX spike-in at the Genetic Resources Core Facility (RRID#913SCR_018669), Johns Hopkins Department of Genetic Medicine.

Single-nucleus transcriptomic profiling analysis

Each sample went through identical quality control processing steps. Cell Ranger (v.7.2.0 for MIF-E22Q library, v.9.1.0 for PAANIB-1 library) was used to align the raw sequencing data using mm10-2020A reference transcriptome and Chromium Mouse Transcriptome Probe Set v.1.0.1. Filtered feature barcode expression matrices were loaded and downstream analysis conducted primarily using Seurat (v.5.1.0)⁹¹. An adaptive quality control approach was used to remove outlier nuclei based on number of reads, unique molecular identifiers and mitochondrial RNA ratio. Outliers were calculated on a per-sample basis using Scater (v.1.30.1)⁹² and any nuclei >3 median absolute deviations were excluded from further analysis (Supplementary Table 1). For the purposes of dimensionality reduction and unsupervised clustering, data normalization and stabilization of sequence depth variance were performed on each sample using SCTransform with default parameters⁹³. For the purposes of plotting gene expression (in UMAP and Violin Plot form), the data was log normalized with Seurat. Data was then annotated and combined into a merged object for downstream quality control and analysis. Dimensionality reduction was performed on all nuclei using PCA, followed by computation of shared nearest neighbors using the first 20 principal components and cluster identification (resolution 0.8). The cell types were identified using a combination of scType⁴³, SingleR (v.2.6.0)⁹⁴ (both automated annotation algorithms) and manual inspection of canonical cell type markers. Doublets were identified using DoubletFinder (v.2.0.4)⁹⁵ in the MIF-E22Q library and scDblFinder (v.1.18.0)⁹⁶ in the PAANIB-1 library with additional manual inspection based on known canonical cell markers and removed from further analyses. The neuronal nuclei were then subsetted and the above analysis pipeline (SCTransform, PCA, nonlinear dimensionality reduction using the first 20 principal components) was repeated with just the neuronal nuclei. Detailed spinal neuron annotation was performed with the SeqSeek software pipeline⁴⁴ (https://github.com/ArielLevineLabNINDS/SeqSeek_Classify_Full_Pipeline; accessed 27 July 2024). Our single-nucleus RNA-seq query dataset was loaded into a Seurat object, quality controlled and normalized as described above. Reference data was downloaded from the SeqSeek GitHub portal and updated to a Seurat v.5 object. Label transfer was performed using the Seurat FindTransferAnchor and TransferData functions. Count matrices for the query and reference datasets were input into the SeqSeek neural_net.py script, using a Tensorflow neural network to annotate the query neuronal nuclei. The resulting annotation was manually completed based on previously defined markers⁴⁴, relabeling <10% of nuclei that were not successfully annotated by the SeqSeek pipeline. Anatomic location of neurons was inferred from SeqSeek derived families which as part of the initial publication were validated using in-situ-hybridization⁴⁴. For identifying DEGs and pathways, the neuron data were pseudobulked using the aggregateAcrossCells function from the scuttle package, which sums read accounts across all cells with a particular label⁹². Any pseudobulked sample that originated from fewer than ten nuclei was not included in the differential expression analyses. For analyses in Figs. 5 and 6, neurons were pseudobulked per sample. For analyses in Extended Data Figs. 8 and 10, data were pseudobulked at the sample, region (dorsal versus mid/ventral) and subtype (excitatory versus inhibitory). Differential gene expression between groups was attained using DESeq2 (v.1.42.1)⁴⁷ on the pseudobulk datasets. Unless otherwise specified, cutoffs for DEGs were Padj < 0.05 and absolute FC > 1.5. Using the DEGs identified by DESeq2, we tested for enrichment of GO terms in the DEGs (either up or downregulated)^36,37,39 using clusterProfiler (v.4.10.1)⁹⁷. For the competitive gene-set testing, the pseudobulk data were analyzed with the pseudoBulkDGE function from scran⁹⁸, which uses edgeR to normalize library size then estimates negative binomial dispersion, followed by quasi-likelihood dispersion and genewise quasi-F-tests. These results were used as the input into the CAMERA gene-set test⁴⁸. Gene set testing was performed using gene sets from MSigDb, including the M2 WP collection and the M5 GO-BP gene sets^49,99,100.

Plots were generated and statistical tests calculated using R (v.4.3.2) and Rstudio (v.2024.04.2+764). PCA plots were produced with DESeq2; volcano plots were produced with EnhancedVolcano (v.1.20.0)¹⁰¹; heatmaps were produced with pHeatmap (v.1.0.12) and enrichplot (v.1.22.0)¹⁰², cnetplots were produced with enrichplot, barcode plots were produced with limma (v.3.60.6)¹⁰³; dimensionality reduction plots and violin plots were produced with Seurat and scCustomize¹⁰⁴.

Statistics

Statistical analyses were carried out using GraphPad Prism v.7. No statistical methods were used to predetermine mouse censuses, but our sample sizes are aligned with those reported in previous publications¹⁰⁵. Mice were randomized or blocked, respectively, given each experiment. WT mice were randomly selected to receive EAE or CFA only. EAE mice were randomly picked to receive PAANIB-1 or vehicle. All mice, unless specifically noted (western blot time point experiments in Fig. 2), that got sick (EAE score > 0) were included in analyses. Each experimenter was blinded to the conditions of the experiments during data collection and analysis. All experiments, other than single-nucleus transcriptomic analyses, were reproduced at a minimum of twice. In each figure, data are presented as mean ± s.d. unless noted otherwise. A two-sided Mann–Whitney nonparametric test was used to measure data that were not normally distributed (EAE motor scores). All other data distributions were assumed to be normal; however, this was not formally tested. An unpaired two-sided Student’s t-test was used to compare differences between two groups. A one-way ANOVA followed by Tukey’s post hoc test was used to compare differences between three or more groups. A two-way ANOVA followed by Tukey’s post hoc test was used to analyze the effect that two independent variables had on a single dependent variable. For single-nucleus transcriptomic profiling analysis, all reported P values adjusted for multiple comparisons were corrected using the Benjamini–Hochberg method¹⁰⁶, unless otherwise specified.

Reporting summary

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

Online content

Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at 10.1038/s41593-026-02201-7.

Author contributions

J.W.M., S.P.G., P.A.C., T.M.D. and V.L.D. conceived of and designed the study; J.W.M., E.S.S., T.G. and M.G. performed the pilot experiments for the paper; J.W.M., S.P.G., M.D.S., D.G., M.L. and B.G.K. performed the primary experiments and collected the data for the paper; J.W.M., M.D.S., M.R. and B.S. carried out the revision experiments; J.W.M., S.P.G., M.D.S. and D.G. performed the data analysis; J.W.M. and S.P.G. developed the lumbar spinal cord cell quantification method; M.D.S. and S.P.G. developed the single-nucleus sequencing method; C.A.P. provided the human brain specimens; J.W.M. and S.P.G. wrote the original paper; J.W.M., S.P.G., M.D.S. and P.A.C. wrote the final paper; J.W.M., S.P.G., P.A.C., D.G., T.M.D., V.L.D. and A.J.G. reviewed and edited the paper; J.W.M., S.P.G., P.A.C., T.M.D. and V.L.D. acquired funding to support this study; P.A.C., T.M.D. and V.L.D. supervised this study.

Peer review

Peer review information

Nature Neuroscience thanks John R Lukens, V. Wee Yong, and the other, anonymous, reviewer(s) for their contribution to the peer review of this work.

Data availability

Raw sequencing data and raw and filtered barcode × feature matrices have been submitted to the Gene Expression Omnibus under SuperSeries accession code GSE304068 with the genetic (MIF-E22Q) dataset being available at GSE282120 and the PAANIB-1 dataset at GSE304067. The datasets generated during the current study are available within the paper or from the corresponding author upon reasonable request. Any additional information on the data reported in this paper is available from the lead contact upon request. Source data are provided with this paper.

Code availability

Previously published software and standard analysis packages were used to perform data analysis as described in Methods. No new code was generated in this study.

Competing interests

T.M.D. and V.L.D. have filed patents on the use of MIF nuclease inhibitors to treat neurologic disorders. All other authors declare no competing interests.

Extended data

is available for this paper at 10.1038/s41593-026-02201-7.

Supplementary information

The online version contains supplementary material available at 10.1038/s41593-026-02201-7.