Cytotoxic CNS-associated T cells drive axon degeneration by targeting perturbed myelinating oligodendrocytes in PLP1 mutant mice

Myelin defects lead to neurological dysfunction in various diseases and in normal aging. Chronic neuroinflammation often contributes to axon-myelin damage in these conditions and can be initiated and/or sustained by perturbed myelinating glia. We have previously shown that distinct mutations in the PLP1 gene result in neurodegeneration that is largely driven by adaptive immune cells. Here we characterize CD8+ CNS-associated T cells in these myelin mutants using single-cell transcriptomics and identify population heterogeneity and disease-associated changes. We demonstrate that early sphingosine-1-phosphate receptor modulation attenuates the recruitment of T cells and neural damage, while later targeting of CNS-associated T cell populations is inefficient and has no effect on neurodegeneration. Applying bone marrow chimerism and utilizing random X chromosome inactivation, we provide evidence that axonal damage is driven by cytotoxic, antigen specific CD8+ T cells that target mutant myelinating oligodendrocytes. These findings offer insights into neural-immune interactions and are of translational relevance for neurological conditions associated with myelin defects and neuroinflammation.


Introduction
The integrity of myelinated axons is essential for the proper function of the mammalian central nervous system (CNS) 1 . Due to their unique properties, myelinated axons are particularly susceptible to injury and their perturbation is a hallmark and early feature of various neurological diseases and of normal aging 2,3 . Recent findings have highlighted the important interplay between neural cells and immune cells in maintaining homeostasis of the axon-myelin unit [4][5][6][7][8] . Disturbances of this interaction can contribute to the initiation and perpetuation of neuroinflammation, demyelination, axonal damage, and neurodegeneration, contributing to functional decline and clinical impairment in distinct neurological diseases such as multiple sclerosis, Alzheimer's disease, Parkinson's disease, hereditary diseases, and in normal aging 9,10 .
In these conditions, macroglial (oligodendrocytes, astrocytes) and microglial cells exhibit changes in their gene and protein expression and adopt a chronic proinflammatory state 4,7,10,11 . Many of these glial pro-inflammatory changes are related to the communication with adaptive immune cells, including elevated expression of molecules implicated in antigen presentation, T cell receptor stimulation/costimulation, and T cell recruitment 8 . Along these lines, there is increasing evidence for a contribution of T lymphocytes and particularly CD8 + T cells to many neurological disorders including inflammatory and classical neurodegenerative diseases, often associated with aging [12][13][14] .
We have previously shown that secondary neuroinflammation acts as an important and targetable amplifier of neural damage in distinct genetically mediated CNS diseases 15 . In mice overexpressing normal or carrying mutant proteolipid protein (PLPtg and PLPmut mice, respectively), the major myelin protein of the CNS, adaptive immune cells accumulate in the white matter, drive axonopathic and demyelinating alterations, and contribute to functional impairment [16][17][18][19] . Pharmacological targeting of innate and adaptive immune reactions can attenuate disease progression in the respective models but has only limited potential to reverse functional impairment 20,21 . This has important implications for progressive forms of multiple sclerosis and leukodystrophies/hereditary spastic paraplegia, which are associated with chronic low-grade neuroinflammation and axon degeneration and can be related to primary oligodendrocyte perturbation 15,[22][23][24] . Moreover, we recently identified commonalities among normal aging and mice with myelin gene defects, underscoring the broad relevance of these processes for frequent agingrelated diseases. We showed that in normal aging (wildtype) mice without defined disease, cytotoxic CD8 + CNS-associated T cells drive axonal damage and neurodegeneration 25 . Similar as in PLPtg mice 19 , this process is dependent on the cytolytic effector protease granzyme B and cognate TCR specificity. Moreover, T cell-driven axon degeneration in aged mice can be aggravated by mimicking infection-related systemic inflammation 25 . Thus, also in aging, perturbation of white matter glial cells results in secondary myelin-related neuroinflammation and contributes to structural and functional decline of myelinated axons. The comparison of these processes between normal aging mice and models with cell type-specific myelin defects might help to clarify some of the involved pathomechanisms and identify targets for intervention.
Both our previous characterization of CD8 + T cells in the CNS of adult and aged mice and pharmacological treatment approaches in PLPmut mice indicated population heterogeneity and functional diversity of these cells 21,25 . However, the exact composition of CD8 + T cell populations and their disease-related changes in the CNS of PLPmut mice have not been analyzed. Moreover, their recruitment and maintenance as well as putative pathogenic effector mechanisms and target structures have not been characterized. These issues are of high relevance for strategies to attenuate deterioration of the nervous system by targeting chronic neuroinflammation. Here, we focus on these questions in the context of sphingosine-1-phosphate receptor (S1PR) modulation with fingolimod, an established diseasemodifying treatment for multiple sclerosis that sequesters lymphocyte subsets in secondary lymphoid organs 26 . This might offer novel explanations for its limited efficacy in progressive disease forms and lead to refined indications for immunomodulatory therapy in multiple neurological disorders.

Transcriptional signatures of CD8 + T cells associated with the healthy and myelin mutant CNS
To characterize CD8 + CNS-associated T lymphocytes in detail we used single-cell RNA sequencing (scRNA-seq) of CD8 + T cells isolated from the brains of adult (12month-old) wildtype (Wt) and PLP1 mutant (PLPmut) mice. Unsupervised clustering of the combined datasets identified nine different clusters ( Fig. 1 A), revealing population heterogeneity. Similar as previously observed when comparing cells from the same adult with aged mice 25 , two clusters resembled central memory T (TCM) cells (TCM1 and TCM2; Fig. 1 B). Moreover, we identified one cluster representing effector T (TEFF) cells and one cluster with a prominent interferon response signature (interferon-stimulated T cells (IST)). We previously localized these subsets in the blood, cerebrospinal fluid (CSF), leptomeninges and choroid plexus of adult and aged mice 25 . We also detected the presence of five groups of previously described CD8 + CNS-associated T cells (CAT1-5) with distinct transcriptional signatures. CAT1, previously found strongly enriched in white matter of aging brains and expressing inhibitory checkpoint molecules like Lag3 and Pdcd1, was similarly frequent in Wt and PLPmut mice. This difference in cluster allocation in comparison with aged mice is likely due to the similarity of these cells to CAT3 and CAT4 in PLPmut mice. Myelin disease-related changes in CD8 + T cells were primarily based on increased numbers of cells representing CAT2 to CAT5 and transcriptional changes within CAT2, e.g., increased expression of Ly6a and Klrc1 (Fig. 1 C-G).
CD8 + CNS-associated T cells showed variable expression of gene modules related to effector and memory function and high expression levels for markers of tissue recruitment and residency (Fig. 2 A). Flow cytometry of CD8 + T lymphocytes isolated from brains of Wt and PLPmut mice confirmed the presence of distinct populations and strongly increased numbers of Ly6A/E + CD103 + cells in PLPmut mice (Fig. S1). These correspond to CAT2, which expressed the highest amount of Gzmb and Itgae (encoding CD103) among CD8 + CAT (Table S1). We furthermore validated the heterogenous expression of different identified CD8 + T cell subset markers in the white matter of Wt and PLPmut mice by immunofluorescence (Fig. S2). Again, a disproportional increase of CD8 + CD103 + T cell numbers with an increased expression of Ly6A/E was detectable in the myelin mutants.
To compare CD8 + T cells accumulating in the brains of aged and PLPmut mice, we integrated the present and previous datasets of adult and aged (24-month-old) Wt, and adult PLPmut mice. We confirmed that the originally identified CAT1 population is specifically enriched in aging mice and that allocation of cells from CAT3 and CAT4 to this cluster occurs when comparing adult Wt and PLPmut mice due to their similar transcriptional profiles (Fig. 2 B). In summary, CD8 + CNS-associated T cells are heterogeneous and show commonalities but also differences in genetic myelin disease compared with normal aging. Their transcriptional signatures indicate cytotoxic effector memory function, activation, and long-term tissue residency 27 .

Distinct S1PR modulation regimens reveal recruitment dynamics and maintenance of CNS-associated T cells in myelin disease
Our unbiased analysis of CD8 + CAT indicated that they show relatively strong expression of genes associated with tissue recruitment (e.g., Cxcr6, Cxcr3, Gzmk) and residency (e.g., Cd69, Itga1, Itgae), while marker genes for tissue egress and recirculation (e.g., S1pr1, S1pr3, S1pr5) were mostly confined to TCM and TEFF clusters (Figs. 2 A and 3 A). This might reflect their adaption to maintenance in the brain and has implications for immunomodulatory approaches that antagonize S1PR signaling to target T cell-mediated neuroinflammation. To address this question, we performed pharmacological treatment of PLPmut mice using fingolimod (FTY720) in the drinking water and distinct treatment regimens. Based on our previous characterization of chronic T cell recruitment and disease progression in the myelin mutant model from around 2 months of age onwards 16 , we started early and late onset regimens at 4 and 10 months of age, respectively (Fig. 3 B). Fingolimod modulates S1PR signaling to sequester lymphocyte subsets in secondary lymphoid organs 26 , and we confirmed a strong depletion of circulating T cells in PLPmut mice by the treatment (Fig. S3 A and B). Lymphopenia was reflected by a significant reduction in relative spleen weight (Fig. S3 B and C).
Focusing on the optic nerve as a model white matter tract, we found that early onset (preventive) treatment for 150 days attenuated the recruitment of CD8 + and CD4 + T cells in PLPmut mice (Fig. 3 C-E). Termination of the treatment at half-time (after 75 days) resulted in restoration of relative spleen weight, reflecting a re-establishment of circulating lymphocytes (Fig. S3 C). However, numbers of CNS-associated T cells were still lower than in untreated mice, demonstrating a relatively slow recruitment without overshoot after cessation of immunomodulation in PLPmut mice (Fig. 3 C-E). Late onset therapeutic treatment of PLPmut mice with FTY720 resulted in a similar reduction of spleen weight as preventive treatment (Fig. S3 D) but had no effect on T cell numbers in optic nerves (Fig. 3 C-E). Neuroinflammation in the white matter of PLPmut mice is characterized by microglial activation, which orchestrates 20 but also reacts to adaptive immune reactions 16 and is reflected by increased expression of sialoadhesin (Sn, Siglec-1, CD169; Fig S4). Parenchymal Sn + microglia in PLPmut mice also expressed the established marker for disease-associated microglia CD11c and low but detectable levels of P2RY12, corroborating that they do not represent infiltrated monocyte-derived macrophages or substantial numbers of borderassociated macrophages 20 . Preventive treatment with fingolimod inhibited the numerical increase of CD11b + microglia in PLPmut mice and attenuated their Sn expression ( Fig. 3 C, F, and G). In contrast, microgliosis was restored after termination of treatment or unaffected by therapeutic treatment (Fig. 3 C, F, and G). Taken together, once CD8 + T cells become CNS-associated in white matter of PLPmut mice, they do not respond to S1PR modulation and sustain chronic neuroinflammation.

Neuroaxonal degeneration in PLPmut mice correlates with the response of CNS-associated T cells to S1PR modulation
Next, we investigated how FTY720 treatment affects axonal damage and neuron loss in PLPmut mice. Neuroinflammation-related degeneration of myelinated axons in these mice is preceded by focal block of axonal transport and formation of spheroids, which can be visualized using antibodies against non-phosphorylated neurofilaments 16 . Corresponding to the response of T cells, preventive treatment attenuated axonal spheroid formation in optic nerves of PLPmut mice, while therapeutic treatment had no effect on ongoing axonal damage ( Fig. 4 A and B). The slow recruitment of T cells after treatment termination translated into preserved axonal integrity compared with untreated mice. Axon degeneration in PLPmut mice culminates in neuron loss and can be monitored using non-invasive longitudinal readout measures of the retinotectal system like optical coherence tomography (OCT) 16 . Preventive and terminated treatment regimens significantly decreased loss of RBPMS + retinal ganglion cells and delayed the selective thinning of the inner retinal layers in OCT ( Fig. 4 A and C). In contrast, therapeutic treatment did not slow the progression of neuron loss or retinal thinning.
We assessed the beneficial effects of preventive FTY720 treatment on white matter integrity in PLPmut mice at the ultrastructural level using electron microscopy. The frequency of axons with abnormally thin (g-ratio ≥ 0.85) or no myelin was not significantly different between treated and untreated groups ( Fig. 5 A and B). However, corroborating our immunohistochemical observations, axons undergoing swelling and organelle accumulation, or Wallerian-like degeneration were less frequent after fingolimod ( Fig. 5 A and C). Thus, neurodegeneration in PLPmut mice can be attenuated by early preventive but not by late S1PR modulation, being in line with the effects on CNS-associated T cells.
To investigate putative direct neuroprotective effects of FTY720 treatment independent of adaptive immune reactions, we treated PLPmut/Rag1 -/mice that lack mature adaptive immune cells and show mitigated but still detectable neurodegeneration 16 . Preventive treatment did not cause any additional amelioration of axon damage, neuron loss, and retinal thinning compared with untreated PLPmut/Rag1 -/mice (Fig. S5).

Axonal damage in PLPmut mice is driven by cytotoxic CD8 + T cells that specifically target mutant myelinating oligodendrocytes
Our previous data suggested a detrimental impact of the CD8 + T cell compartment in PLPmut mice 16,20,21 and our unbiased analysis identified the strongest diseaserelated changes in CAT2, a cluster with the signature of tissue-resident memory and cytotoxic function. We therefore investigated the pathogenic role, putative cytotoxic effector mechanisms, and antigen/TCR dependency of CD8 + T cells by generating bone marrow chimeric PLPmut mice without confounding irradiation. PLPmut/Rag1 -/mice lack mature adaptive immune cells and show diminished axonal damage 16 , enabling us to use them as recipients for comparing the impact of different adaptive immune cell populations/effector molecules. PLPmut/Rag1 -/mice received bone marrow from various donor lines ( Fig. 6 A), which leads to efficient and persistent restoration of adaptive immune cells within host tissues including the CNS 17,19,25,28 . Transplantation of Wt bone marrow into Rag1 deficient PLPmut mice re-established T cell recruitment and the formation of SMI32 + axonal spheroids and thinning of the inner retina as assessed by OCT ( Fig. 6 B-F). In contrast, PLPmut/Rag1 -/mice reconstituted with Cd8 -/bone marrow retained diminished axonal damage despite the presence of CD4 + T cells in the white matter. Similarly, reconstitution with Gzmb -/or OT-I (TCR specificity against ectopic ovalbumin) bone marrow did not result in more axonal spheroid formation and retinal thinning than in PLPmut/Rag1 -/mice despite the restored recruitment of (genetically modified) T cells. In combination, these observations demonstrate a detrimental role of CD8 + T cells which damage axons in a granzyme B-and cognate TCR-dependent manner in PLPmut mice. Moreover, the normal recruitment but lack of impact of CD8 + T cells in BMC OT-I mice indicates white matter antigen specific activation.
Since neuroinflammation in PLPmut mice results from gene defects specifically affecting oligodendrocytes, we wondered if CD8 + T cells target axon segments enwrapped by mutant myelin-producing cells. Due to random X chromosome inactivation, heterozygous female mice contain both Wt and PLPmut oligodendrocytes in the same white matter tracts. We detected a similar accumulation of CD8 + T cells but decreased Sn expression on microglia in optic nerves of heterozygous compared with homozygous females (Fig. 7 A, B, and C). Moreover, SMI32 + axonal spheroids were approximately half as frequent in mosaic mice ( Fig. 7 D). PLP1 mutations result in ultrastructural myelin compaction defects, which can be appreciated by electron microscopy 16 . We observed that spheroids and degenerating profiles in heterozygous females were almost exclusively surrounded by myelin segments displaying compaction defects, suggesting their formation by mutant oligodendrocytes (Fig. 7 E; 85.4% mutant myelin, 4.2% wt myelin, 10.4% no myelin; n = 100 damaged axons in 5 mice). Since loss of RGCs and retinal thinning were also halved by oligodendrocyte mosaicism (Fig. 7 F and G), we conclude that cytotoxic CD8 + T cells specifically target mutant myelinating oligodendrocytes which indirectly causes axonal damage and neurodegeneration. This matched with increased expression of the cognate antigen-presenting complex for CD8 + T cellsmajor histocompatibility complex class I (MHC-I) -on oligodendrocytes and microglia, especially in homozygous PLPmut mice (Fig. S6, Fig 7 H and I). To further investigate direct interactions between CD8 + T cells and Wt and PLPmut oligodendrocytes, we analyzed citrullination of MBP, which is typically enhanced in perturbed and destabilized myelin and increases its susceptibility to inflammation 23 . Indeed, citrullinated MBP was strongly enriched in optic nerves of PLPmut mice and was less homogenously distributed in heterozygous Wt/PLPmut mice (Fig. 7 J). Using triple immunofluorescence, we found that an association of CD8 + T cells with SMI32 + damaged axons was almost exclusively detectable at segments ensheathed by PLPmut (citrullinated MBP + ) myelin ( Fig. 7 K and L).

Discussion
Recent studies have renewed the appreciation of oligodendrocyte-lineage cells as immunocompetent glial cells that can initiate or perpetuate neuroinflammation 5,24,[29][30][31][32] . This has relevance for inherited and acquired diseases of myelin and supports the hypothesis of CNS-intrinsic perturbation as one possible mechanism (among multiple other factors) contributing to chronic adaptive immune reactions with a detrimental impact on white matter integrity [33][34][35] . We have previously shown that defects in myelin genes result in secondary neuroinflammation that contributes to axonal damage and functional impairment 16,17 . PLP1 mutant oligodendrocytes carry point mutations that have been identified in patients with multiple sclerosis and seem to acquire disease-associated characteristics at least partially shared (e.g., increased MHC-I expression) with other diseases and normal aging 11,32,36 . Moreover, normal aging is also associated with an accumulation of CD8 + T cells in the white matter that appear to target myelinated axons and contribute to neurodegeneration, cognitive, and motor decline 25 . These observations reflect the vulnerability of myelinated axons to immune-mediated perturbation upon various primary defects and underscore an active role of myelinating glia in these conditions. Moreover, targeting pro-inflammatory microglia in PLPmut mice attenuates T cell-driven axon damage 20 , supporting the hypothesis that disturbed glial interactions play a major part in the initiation of neuroinflammation.
Previous work has indicated that CD8 + T cells populating the CNS parenchyma resemble tissue-resident memory (TRM) cells which do not recirculate but respond to peripheral stimuli [37][38][39] . Our characterization of these cells in aging 25 and myelin mutant mice (this study) is in line with this and revealed that different subpopulations of these cells show signatures related to maintenance and long-term residency within the CNS. While our exploratory scRNA-seq approach required pooled analysis of rare CD8 + T cells from brains of multiple mice (limiting insights into variation), we validate our major conclusions using independent techniques and confirm the phenotype, heterogeneity, and disease-related changes of these cells. A hallmark of TRM cells is the downregulation of receptors involved in tissue egress and recirculation once they have infiltrated their respective niches 40,41 . Consequently, TRM cells become resistant to S1PR modulation and systemic depletion of T cells has little effect on frequencies and impact within the infected or inflamed tissue. Our distinct treatment regimens with fingolimod confirm that early (preventive) treatment causes lymphopenia and attenuates CNS recruitment of T cells, whereas late (therapeutic) treatment still depletes circulating T cells but has no effect on those associated with the CNS parenchyma. The relative stability of CD8 + T cell densities but accelerating accumulation of axonal damage at advanced disease stages might indicate either earlier compensatory/resilience mechanisms or a pronounced detrimental impact of long-term resident T lymphocytes with limited turnover. Recent observations support a local contribution of CD8 + TRM cells to chronic CNS autoimmunity 42,43 . In multiple sclerosis, active lesions are dominated by CD8 + T cells that infiltrate the parenchyma, partially acquire features of tissue-resident memory cells, and persist in the CNS 44 . Activated, clonally expanded CD8 + TRM cells are also present in the CSF from MS-discordant monozygotic twins with subclinical neuroinflammation (prodromal multiple sclerosis) 45 . Interestingly, these cells at least partially express CD103 and accumulate in normal-appearing white matter of MS patients, where they might sustain diffuse chronic inflammation and axonal damage especially in progressive forms of the disease [46][47][48] . Here we observed using multiple independent methods that a subset of CD8 + CAT with the highest expression of CD103 (CAT2) shows the most prominent disease-related changes in PLPmut mice, indicating that the myelin mutant model is suitable to study neural-immune interactions with relevance for chronic neuroinflammatory diseases.
The same population also showed the highest expression of Gzmb among parenchymal CD8 + T cells, indicating that like in aging and PLPtg mice 19,25 , axonal damage might depend on the cytolytic effector protease. By bone marrow transfer into Rag1-deficient PLPmut mice, we demonstrate that CNS-infiltrating CD8 + T cells require Gzmb expression and cognate TCR specificity (BMC OT-I) to drive axonal damage. In contrast, reconstitution of CD4 + T cells and B cells in the absence of CD8 + T cells (BMC Cd8 -/-) does not result in more axonal damage than in genuine PLPmut/Rag1 -/mice (displaying attenuated axonal damage). These reconstituted CD8lymphocytes are unable to damage myelinated axons despite otherwise similar characteristics (CNS infiltration, putative effector molecule expression) as in genuine PLPmut mice. Thus, axonal damage in PLPmut mice appears to be driven by antigen specific, autoreactive CD8 + T cells that use contact-dependent (cytotoxic granules) effector mechanisms. The similarity of these processes between distinct myelin mutants and normal white matter aging is a novel insight and indicates a conserved feature possibly shared with many other conditions. While the exact antigen(s) recognized by these cells remains to be determined in future studies, several observations lead us to speculate that they might target perturbed myelin. First, CD8 + T cells are predominantly accumulating in white matter tracts and associate with juxtaparanodal domains of mutant fibers 16 , similar as previously observed in PLPtg mice 18 . Second, axons showing focal signs of damage and degeneration are almost exclusively enwrapped by perturbed mutant myelin in heterozygous (mosaic) females. Third, myelin mutant oligodendrocytes express increased levels of MHC-I compared with healthy (Wt) myelinating glia, indicating that they might show increased (auto)antigen presentation to CD8 + T cells. It is conceivable that myelin gene defects (and aging) result in molecular changes and cell stress pathways in oligodendrocyte subsets that initiate their communication with immune cells and make them susceptible to immune-mediated damage, similar as recently shown for selective neuron populations in AD 49 . Among those glial changes is enhanced citrullination of MBP, as occurs in destabilized myelin after toxin-or myelin disease-related perturbation 23,50 . We found that CD8 + T cells in proximity to SMI32 + axons were almost exclusively associated with segments showing enhanced MBP citrullination, again indicating specificity for mutant myelin-related components. Such changes in myelin properties might lead to its recognition as neoantigen when presented to T cells.
Considering that a cytotoxic T cell attack on perturbed myelinating oligodendrocytes drives axonal damage, it remains to be clarified how this impairment is mediated before obvious demyelination. Collateral damage, changes in trophic/metabolic support of axons, or other detrimental reactions of myelinating oligodendrocytes in response to T cell-derived lytic granules might be among the responsible mechanisms [51][52][53] . In aged mice, scRNA-seq has revealed some of the transcriptional changes in oligodendrocytes upon being targeted by CD8 + T cells 32 . Both the frequencies of Serpina3n + (an inhibitor of proteases including granzyme B) and interferon-responsive oligodendrocytes in the white matter were decreased upon Rag1 deficiency, indicating that oligodendrocytes show distinct responses to neuroinflammation. We speculate that this might be related to the impact of Gzmb-vs Ifng-expressing CD8 + T cell populations and have differential effects on axon-glia interactions and axon degeneration. Our previous characterization of CD8 + CAT from aged mice revealed the strong increase of a specific population expressing checkpoint molecules that is not responding in PLPmut mice and a downregulation of Ifng 25 . Combined with the observation that axonal damage in aged and myelin mutant mice depends on granzyme B, we propose that the population of oligodendrocytes upregulating Serpina3n to counteract T cell cytotoxicity, demyelination, and cell death is more detrimental to axonal integrity. Indeed, axons that remain myelinated under inflammatory conditions have recently been proposed to be at higher risk for degeneration 54 . The fact that FTY720 treatment attenuated T cell recruitment and axonal damage but did not significantly affect myelin integrity in PLPmut mice also argues against the possibility of axon degeneration being a consequence of demyelination. Since a direct T cell attack on axons is also difficult to explain when considering the reduced damage in mosaic females, further studies should explore the detrimental T cell-driven reactions of PLPmut oligodendrocytes.
Our observations are of translational relevance for neurological conditions associated with myelin defects and chronic neuroinflammation and show that early onset of therapy might be critical for the efficacy of S1PR-modulation to prevent recruitment and CNS colonization of T cells. When these cells become resident, they downregulate receptors for tissue egress and appear to stay within the CNS for extended time periods without relying on high turnover from the circulation. Moreover, cessation of circulating lymphocyte ablation in PLPmut mice leads to a slow restoration of T cells within the CNS with sustained benefits on neural integrity, reflecting the low-grade accumulation of inflammatory damage typical of chronic disease. The lack of any beneficial effect upon therapeutic treatment regarding the progression of neurodegeneration argues against a direct protective effect of fingolimod in PLPmut mice. Indeed, FTY720 treatment of PLPmut/Rag1 -/mice (lacking adaptive immune cells and showing much milder -but still detectablepathology than genuine PLPmut mice) did not cause any additional amelioration of axon damage and neuron loss. This is supported by our previous treatment approaches in Rag1-deficient models of rare lysosomal storage diseases accompanied by T cell-driven axon degeneration, which also did not reveal major effects independent of immune cells 55 .
The resilience of CNS-associated T cells and compartmentalized inflammatory response might explain the limited efficacy of a previous clinical trial using fingolimod in primary progressive multiple sclerosis 56 . Furthermore, they implicate that neuroaxonal degeneration in such chronic conditions might still be related to lowgrade inflammation, going along with the poor efficacy of S1PR-modulating therapies. Interestingly, we previously observed that late onset therapeutic treatment with teriflunomide, another established drug for MS, showed beneficial effects and halted ongoing axon degeneration in PLPmut mice 21 . Teriflunomide is an inhibitor of dihydroorotate dehydrogenase and modulates mitochondrial respiration, T cell activation and migration, particularly within the CD8 + compartment 57 . Moreover, it induces a tolerogenic bias in immune cells of MS patients 58 . In the myelin mutants, therapeutic treatment with teriflunomide resulted in an increased frequency of CD8 + T cells expressing high levels of CD122 and PD-1 in the white matter 21 . Such cells have been shown to restrict inflammation and autoimmunity by suppressing effector T cells, and regulatory populations within the CD8 lineage have been described in mice and humans [59][60][61][62][63][64] . Combining our previous and present observations, we speculate that CD8 + CAT5 might represent such an anti-inflammatory population within the CNS. This cluster is enriched with marker genes of regulatory function (Fig. 2 A) and shows low expression of Txnip (enriched in CAT2; Table S1), a sensor of oxidative phosphorylation and glycolysis 65 , which might explain its resilience to inhibition of mitochondrial metabolism. Therapeutic strategies to modulate the activation or change the composition of CD8 + CNS-associated T cells towards tolerance might be preferred over more general immunosuppression. Moreover, our data indicate that treatment time point/disease stage and activity must be carefully considered when selecting specific immunomodulatory treatment approaches for chronic neuroinflammation. Nevertheless, we here emphasize that targeting neuroinflammation might be a feasible approach for chronic progressive degenerative diseases associated with myelin defects and aging.

Figure 4. Preventive but not therapeutic FTY720 treatment attenuates neuroaxonal degeneration in
PLPmut mice (A) Immunofluorescence detection of SMI32 + axonal spheroids (top; arrows) in the optic nerves and RBPMS + Brn3a + RGCs (bottom) in the retinae of Wt, PLPmut and FTY720-treated PLPmut mice using regimens indicated in Fig. 3 B. Scale bars, 20 μ m. (B) Quantification of SMI32+ axonal spheroids and (C) RGCs in Wt, PLPmut and FTY720-treated PLPmut mice (n = 5 mice per group). Preventive FTY720 treatment attenuates axonal damage and neuron loss in PLPmut mice which is maintained after termination at half time. Therapeutic FTY720 treatment has no effect on the progression of neurodegeneration in PLPmut mice. (D) OCT analysis of the innermost retinal composite layer (NFL/GCL/IPL) in peripapillary circle scans. Preventive FTY720 treatment attenuates retinal thinning in PLPmut mice which is partially maintained after termination at half time. Therapeutic FTY720 treatment has no effect on retinal thinning in PLPmut mice. B, C: one-way ANOVA with Tukey's multiple comparisons test. D: two-way ANOVA with Tukey's multiple comparisons test. Data are presented as the mean ± SD. All data represent at least three independent experiments.  0.85) and non-myelinated axons or (C) axonal spheroids and degenerating axons in Wt, PLPmut and FTY720-treated PLPmut mice (n = 5 mice per group). Preventive FTY720 treatment attenuates axonal damage in PLPmut mice but has no effect on myelin alterations. B, C: oneway ANOVA with Tukey's multiple comparisons test. Data are presented as the mean ± SD. All data represent at least three independent experiments.

Resource availability
Lead contact Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Janos Groh (groh_j@ukw.de).

Materials availability
This study did not generate new unique reagents.

Data and code availability
• The single cell RNA sequencing data in this publication have been deposited in the Gene Expression Omnibus (GEO): GSE138891. Other data that support the findings of this study will be shared by the lead contact upon request. • This paper does not report original code.
• Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.  68 mice were on a uniform C57BL/6J genetic background; they were bred, regularly backcrossed and aged in-house. Since we did not detect obvious differences between male and female mice in the analyses presented in the current study, mice of either sex were used for most of the experiments (except for analyzing heterozygous vs homozygous females). Genotypes were determined by conventional PCR using isolated DNA from ear punch biopsies.

Method details Flow cytometry and cell sorting
Mice were euthanized with CO 2 (according to the guidelines by the State Office of Health and Social Affairs Berlin) and blood was thoroughly removed by transcardial perfusion with PBS containing heparin. Brains including optic nerves, leptomeninges and choroid plexus were dissected, collected in ice-cold PBS and cut into small pieces. Tissue was digested in 1 ml of Accutase (Merck Millipore) per brain at 37 °C for 30 min and triturated through 100-μm cell strainers, which were rinsed with 10% FCS in PBS. Cells were purified by a linear 40% Percoll (GE Healthcare) centrifugation step at 650 g without brakes for 25 min and the myelin top layer and supernatant were discarded. Mononuclear cells were resuspended in fluorescenceactivated cell sorting buffer (1% BSA and 0.1% sodium azide in PBS) and isolated cells were counted for each brain. single viable cells were gated and CD45 high CD8 + cells were analyzed using a FACSLyric (BD Biosciences) and FlowJo (version 10; LLC). Contribution of the different subsets in absolute numbers was calculated by extrapolating their frequencies to the number of CD45 high CD8 + T cells per brain. Circulating leukocytes were quantified in peripheral blood samples. Before transcardial perfusion, blood was collected from the right atrium of the heart using a heparinized capillary and coagulation was prevented by adding PBS containing heparin. Erythrocytes were lysed and the remaining cells were washed and analyzed by flow cytometry. Total leukocytes were gated based on forward and side scatter, myeloid cells were stained using PE-conjugated antibodies against CD11b (1:100, catalog no. 557397; BD Biosciences), and T lymphocytes were stained using antibodies against CD4 and CD8 (1:100, catalog nos. 553049 and 553032; BD Biosciences). At least 1 × 10 5 leukocytes per mouse were analyzed and their amount per microliter of blood was calculated.

Single-cell RNA sequencing (scRNA-seq) and data processing
Around 15,000 CD45 high CD8 + single cells were sorted per sample using a FACSAria III (BD Biosciences) before being encapsulated into droplets with the Chromium Controller (10x Genomics) and processed according to the manufacturer's specifications. Briefly, every transcript captured in all the cells encapsulated with a bead was uniquely barcoded using a combination of a 16-base pair (bp) 10x barcode and a 10-bp unique molecular identifier (UMI). Complementary DNA libraries ready for sequencing on Illumina platforms were generated using the Chromium Single Cell 3′ Library & Gel Bead Kit v2 (10x Genomics) according to the detailed protocol provided by the manufacturer. Libraries were quantified by Qubit 3.0 Fluorometer (Thermo Fisher Scientific) and quality was checked using a 2100 Bioanalyzer with High Sensitivity DNA kit (Agilent Technologies). Libraries were pooled and sequenced with a NovaSeq 6000 platform (S1 Cartridge; Illumina) in paired-end mode to reach a mean of 75,412 reads per single cell. A total of 5,017 and 6,467 cells were captured and a median gene number per cell of 1,325 and 1,283 could be retrieved for adult Wt and PLPmut cells, respectively. Data were demultiplexed using the CellRanger software v.2.0.2 based on 8 bp 10x sample indexes; paired-end FASTQ files were generated. The cell barcodes and transcript unique molecular identifiers were processed as described previously 69 . The reads were aligned to the University of California, Santa Cruz mouse mm10 reference genome using STAR aligner 70 v.2.5.1b. The alignment results were used to quantify the expression level of mouse genes and generate the gene-barcode matrix. The cellranger aggr command of CellRanger was used to aggregate different libraries. Subsequent data analysis was performed using the R package Seurat 71 v.2.4 and 4.0. Doublets and potentially dead cells were removed based on the percentage of mitochondrial genes (cutoff set at 5%) and the number of genes (cells with >800 and <2,200 genes were used) expressed in each cell as quality control markers. The gene expression of the remaining cells (4,338 and 5,110 cells from Wt and PLPmut mice, respectively) was log-normalized. Highly variable genes were detected with Seurat and the top 1,000 of these genes were used as the basis for downstream clustering analysis after regressing out mitochondrial expression per cell. Principle component analysis was used for dimensionality reduction and the number of significant principal components was calculated using the built-in JackStraw function. Cells were clustered based on the identified principal components (16) with a resolution of 0.6; uniform manifold approximation and projection was used for data visualization in two dimensions. A minimal contamination of myeloid cells was removed based on marker gene expression. Contribution of the samples to each cluster in absolute numbers was calculated by extrapolating their frequencies to the number of CD45 high CD8 + T cells per brain. Differentially expressed genes were identified with min.pct = 0.25 and a cutoff of p_val_adj > 0.05. Complete lists of differentially expressed genes are included in Table S1. Marker gene scores for feature expression programs were calculated using the AddModuleScore function in Seurat.

Immunomodulatory treatment
Fingolimod (FTY720, provided by Novartis, Basel, Switzerland) was dissolved in autoclaved drinking water at 3 μ g/mL and provided ad libitum. With an approximate consumption of 5 ml/day and 30 g body weight, this corresponds to a dose of 0.5 mg/kg body weight/day. This concentration is based on previous animal experiments 72,73 and approximately corresponds to doses used for human multiple sclerosis patients, when a dose conversion scaling is applied 74 . Non-treated controls received autoclaved drinking water and the water with or without FTY720 was changed weekly. Mice were treated for 75 or 150 days and monitored daily regarding defined burden criteria and phenotypic abnormalities. No obvious side effects or significant changes in body weight were detected upon the treatment.

Histochemistry and immunofluorescence
Mice were euthanized with CO 2 (according to the guidelines by the State Office of Health and Social Affairs Berlin), blood was removed by transcardial perfusion with PBS containing heparin and tissue was fixed by perfusion with 2% paraformaldehyde (PFA) in PBS. Tissue was collected, postfixed, dehydrated, and processed as described previously 16 . Blood was collected before transcardial perfusion from the right atrium using a heparinized capillary and smears were air-dried overnight. Immunohistochemistry was performed on 10-μm-thick longitudinal optic nerve or spleen cryo-sections and blood smears after postfixation in 4% PFA in PBS or ice-cold acetone for 10 min. Sections were blocked using 5% BSA in PBS and incubated overnight at 4 °C with 1 or an appropriate combination of up to 3 of the following antibodies: For indirect detection, immunoreactive profiles were visualized using fluorescently labeled (1:300; Dianova) secondary antibodies, streptavidin (1:300; Thermo Fisher Scientific) or biotinylated secondary antibodies (1:100; Vector Laboratories) and streptavidin-biotin-peroxidase (Vector Laboratories) complex using diaminobenzidine HCl and H 2 O 2 ; nuclei were stained with 4,6-diamidino-2-phenylindole (DAPI) (Sigma-Aldrich). Light and fluorescence microscopy images were acquired using an Axio Imager M2 microscope (ZEISS) with ApoTome.2 structured illumination equipment, attached Axiocam cameras and corresponding software (ZEN v.2.3 blue edition) or a FluoView FV1000 confocal microscope (Olympus) with corresponding software (v.2.0). Images were minimally processed (rotation, cropping, addition of symbols) to generate figures using Photoshop CS6 and Illustrator CS6 (Adobe). For quantification, immunoreactive profiles were counted in at least three nonadjacent sections for each animal and related to the area of these sections using the cell counter plugin in Fiji/ImageJ v.1.51 (National Institutes of Health). To quantify RGCs, perfusion-fixed eyes were enucleated, and specific markers of the inner retinal cell types were labeled in free-floating retina preparations. Fixed retinae were frozen in PBS containing 2% Triton X-100, thawed, washed and blocked for 1 h using 5% BSA and 5% donkey serum in PBS containing 2% Triton X-100. Retinae were incubated overnight on a rocker at 4 °C with appropriate combinations of the following antibodies: guinea pig anti-RBPMS (1:300, catalog no. ABN1376; Merck Millipore); goat anti-Brn3a (1:100, catalog no. sc-31984; Santa Cruz Biotechnology); immune reactions were visualized using fluorescently labeled (1:500; Dianova) secondary antibodies, retinae were flat-mounted, and the total retinal area was measured. RGCs were quantified in three images of the middle retinal region per flat mount using the cell counter plugin in Fiji/ImageJ v.1.51 (National Institutes of Health). Images were taken at a fixed distance (~1 mm) and magnification from the optic nerve head in three different quadrants of the flat mounts.

Electron microscopy
The optic nerves of transcardially perfused mice were postfixed overnight in 4% PFA and 2% glutaraldehyde in cacodylate buffer. Nerves were osmicated and processed for light and electron microscopy; morphometric quantification of neuropathological alterations was performed as published previously 16 using a LEO906 E electron microscope (ZEISS) and corresponding software iTEM v.5.1 (Soft Imaging System). At least 10 regions of interest (corresponding to an area of around 5% and up to 3,000 axons per individual optic nerve) were analyzed per optic nerve per mouse. The percentages of axonal profiles showing spheroid formation or undergoing degeneration were identified individually by their characteristic morphological features in electron micrographs and related to the number of all investigated axons per optic nerve per mouse. Genetically perturbed myelin in heterozygous females was identified by myelin compaction defects at high resolution. Images were processed (rotation, cropping, addition of symbols and pseudocolor) to generate figures using Photoshop CS6.

Spectral domain optical coherence tomography (OCT)
Mice were subjected to OCT imaging with a commercially available device (SPECTRALIS OCT; Heidelberg Engineering) and additional lenses as described previously 16,75 . Mice were measured at different ages for longitudinal analysis and the thickness of the innermost retinal composite layer comprising the nerve fiber layer (NFL), GCL and inner plexiform layer (IPL) were measured in high-resolution peripapillary circle scans (at least ten measurements per scan) by an investigator unaware of the genotype and treatment condition of the mice using HEYEX v.1.7.1.

Bone marrow transplantation
Bone marrow was transferred according to previously published protocols 25,76 . Briefly, bone marrow was isolated from the femur and tibia of donor mice and 1 × 10 7 cells were injected intravenously into anaesthetized PLPmut/Rag1 -/mice; this provides a niche for engraftment and long-term reconstitution of adaptive immune cells without confounding irradiation 28 . PLPmut/Rag1 -/mice were reconstituted at 2 months of age and analyzed at 6 months of age. Successful chimerism was controlled by flow cytometry of splenocytes and immunohistochemistry on optic nerve sections. Engraftment of transplanted bone marrow led to a frequency of the respective T lymphocyte types in the PLPmut/Rag1 -/hosts which was similar to PLPmut/Rag1 +/+ mice.

Quantification and statistical analysis
All quantifications and analyses were performed by blinded investigators who were unaware of the genotype and treatment group of the respective mice or tissue samples after concealment of genotypes with individual uniquely coded labels. Animals were randomly placed into experimental or control groups according to the genotyping results using a random generator (http://www.randomizer.org). For biometrical sample size estimation, the program G*Power v.3.1.3 was used 77 . Calculation of appropriate sample size groups was performed using an a priori power analysis by comparing the mean of 2 to 4 groups with a defined adequate power of 0.8 (1 -beta error) and an α error of 0.05. To determine the prespecified effect size d or f, previously published data were considered as comparable reference values 16,20,21 . This resulted in large prespecified effect sizes ranging from 1.20 to 3.56 for our primary outcome measures (densities of T cells, SMI32 + axonal spheroids, RBPMS + RGCs, OCT of inner retinal thinning). The number of individual mice per group (number of biologically independent samples) for each experiment and the meaning of each data point are indicated in the respective figure legends. All data (except the scRNA-seq experiment) represent at least three independent experiments. For this, we quantified a specific cell type/structure in multiple different sections/samples of a respective tissue and averaged the measurements into one single data point. No animals were excluded from the analyses. In the scRNA-seq experiment, we had to pool the brains of 4-5 mice for each age/genotype group due to the low number of T cells in the CNS. Statistical analysis was performed using Prism 8 (GraphPad Software). The Shapiro-Wilk test was used to check for the normal distribution of data and the F test was used to check the equality of variances to ensure that all data met the assumptions of the statistical tests used. Comparisons of two groups were performed with an unpaired Student's t-test (parametric comparison) or Mann-Whitney U-test (nonparametric comparison). For multiple comparisons, a one-way analysis of variance (ANOVA) (parametric) or Kruskal-Wallis test (nonparametric) with Tukey's post hoc test were applied and adjusted P values are presented. P < 0.05 was considered statistically significant; exact P values are provided whenever possible in the figures and/or figure legends. Figure S1. Flow cytometry validates the heterogeneity and accumulation of CD8 + T lymphocyte subsets in the myelin mutant CNS (A) Representative plots of flow cytometric analysis of single, viable CD45 high CD8 + T cells freshly isolated from brains of 15-month-old Wt (top) and PLPmut (bottom) mice. Gated CD8 + T cells are analyzed for expression of CXCR4, CXCR6, Ly6A/E, and CD103. Percentages and identity of the respective cells are indicated in the quadrants. (B) CD8 + T cells comprise similar proportions of CXCR6 + CXCR4 + cells (CAT5) and CXCR6 + CXCR4cells (CAT1-4, IST) when comparing Wt and PLPmut mice (n = 6 mice per group). (C) Among CD8 + CXCR6 + CXCR4 -T cells, Ly6A/E + CD103 + cells (CAT2) but not Ly6A/E + CD103cells show an increased frequency. (D). Total numbers of CD45 high CD8 + T cells and (E) the different subsets per brain reflect a significant accumulation of most populations with a disproportionally increased number of CAT2 cells. B-E: unpaired Student's t-test. Data are presented as the mean ± SD. All data represent at least three independent experiments Figure S2. Immunohistochemistry validates the heterogeneity and activation of CD8 + T cell subsets in the myelin mutant CNS (A) Immunofluorescence detection of CD8, LAG3, and CXCR4 or CD8, CD103, and Ly6A/E in the optic nerves of 9month-old Wt and PLPmut mice. CD8 + T cells (arrows) show heterogenous expression of these markers. Scale bar, 10 μ m. (B) Quantification of LAG3 + (CAT1), CXCR4 + (CAT5), CD103 + (CAT2) subsets among CD8+ T cells as well as Ly6A/E immunoreactivity of CAT2 (n = 5 mice per group). There is an increased frequency of CD103 + cells with increased Ly6A/E expression detectable among CD8 + T cells in PLPmut mice. B: unpaired Student's t-test. Data are presented as the mean ± SD. All data represent at least three independent experiments. Figure S3. Sphingosine-1-receptor modulation with FTY720 depletes circulating T cells in PLPmut mice (A) Flow cytometric quantification of leukocytes, CD11b + myeloid cells, CD8 + T cells, and CD4 + T cells per µl blood from Wt, PLPmut and FTY720-treated PLPmut mice (n = 5 mice per group) after the preventive regimen shown in Fig. 3 B. (B) Representative immunohistochemical detection of CD8 + T cells in blood and spleen of PLPmut and FTY720-treated PLPmut mice. Scale bars, 20 µm (top) and 40 µm (bottom). Numbers of CD8 + T cells and spleen volume are strongly decreased by FY720. (C) Analysis of the relative spleen weights in Wt, PLPmut and FTY720-treated PLPmut mice (n = 5 mice per group) using regimens indicated in Fig. 3 B. Preventive FTY720 treatment reduces spleen weight in PLPmut mice which is restored after termination at half time. Therapeutic FTY720 treatment also reduces spleen weight. C: one-way ANOVA with Tukey's multiple comparisons test. Data are presented as the mean ± SD. All data represent at least three independent experiments.    Table S1. Complete lists of cluster-specific marker and differentially expressed genes for the scRNA-seq data.