Mice
For lineage tracing of resting fibroblasts and tracking of injury-responsive fibroblasts, we crossed Col1a2creERT2 mice (MGI 6721050, a gift from B. Zhou)61 with Rosa26tdT-Ai14 mice (R26-CAG-RFP, Jackson 007914) containing a flox-stop-flox sequence upstream of a CAG-RFP-WPRE cassette in the constitutively expressed ROSA26 locus. Tamoxifen induces RFP expression in Col1a2-lineage+ cells. Col1a2creERT2 mice were also crossed to Rosa26Sun1GFP mice (Jackson 030952), enabling fibroblast nuclear identification. For dMCAO experiments, a distinct Col1a2creERT allele was used (Jackson 029567). Additional stromal reporters used include Col1a1GFP mice (a gift from D. Brenner) to mark active Col1a1-expressing cells62, PdgfraGFP (PDGFRa-H2B-eGFP nuclear-localized GFP, Jackson 007669) to track fibroblasts and oligodendrocyte lineage cells63,64, Rosa26Tdt-Ai14 mice crossed to Gli1creERT2 mice (Jackson 007913) to track adventitial fibroblasts16, Twist2cre mice (Jackson 008712) to track fibroblasts and mural cells65, Acta2creERT2 mice to track myofibroblasts and smooth muscle cells66, Ng2creER mice (Jackson 008538) to mark pericytes and oligodendrocyte precursor cells67, Atp13a5creERT2 mice to mark pericytes68, and Cthrc1creER mice (a gift from D. Sheppard) to track myofibroblasts28.
To track immune cell subsets, we crossed Rosa26tdT-Ai14 mice to CD4cre mice (Jackson 022071) to track CD4+ and CD8+ T cells69, Ccr2creERT2 mice (a gift from B. Becher) to track monocyte-derived cells, P2ry12creERT2 mice (Jackson 034727) to track microglia-derived cells, Cx3cr1creER (Jackson 020940) to track macrophages and Pf4cre (Jackson 008535) to track BAMs. We also used TBET (Tbx21)-zsGreen transgenic mice (provided by J. Zhu)70 to track type 1 lymphocytes. Cx3cr1creER mice were additionally crossed with Tgfb1GFP mice (MGI 3719583)71 and Tgfb1flox mice (Jackson 033001) to drive deletion of Tgfb1 in macrophages (in Cx3cr1creER; Tgfb1GFP/flox mice). Controls were tamoxifen-induced, littermate Cx3cr1creER; Tgfb1flox/+ mice.
To conditionally delete fibroblast Tgfbr2, we crossed Col1a2creERT2 mice to Tgfbr2flox mice (both Tgfbr2-exon2flox, MGI 238451372 and Tgfbr2-exon4flox, Jackson 012603). To conditionally delete fibroblast Cxcl12, we crossed Col1a2creERT2 mice to Cxcl12flox mice (Jackson 021773). We used Itgb8tdT mice (Itgb8-IRES-tdT, provided by H. Paidassi)73 to visualize Itgb8 expression. We used Itgb8flox mice (MGI 3608910)74 crossed to Emx1cre mice (Jackson 005628) or hGfapcre mice (Jackson 004600), to delete Itgb8 from neurons and glial cells42,75,76; some hGfapcre mice were crossed to iSurecre (MGI 6361135, 77) to optimize Cre efficiency. For the above conditional-knockout strains, controls were tamoxifen-induced (when relevant), littermate Cre-negative or flox-heterozygous mice, co-housed non-littermate lineage tracer mice (for select experiments involving Col1a2-lineage+ fibroblast quantification), or age-matched, uninduced non-littermate controls (a portion of immunophenotyped resting mice and select tMCAO experiments). Additionally, to enable depletion of myofibroblasts, we crossed Cthrc1creER mice to Rosa26DTA mice; controls were littermate cre-positive vehicle-treated (saline or corn oil) or cre-negative tamoxifen-induced mice.
Mice of both sexes were backcrossed on C57BL/6 for at least ten generations, or on a mixed genetic background (Gli1creERT2, Cx3cr1creER, Emx1cre). If not otherwise stated, all experiments were performed with 7–21 week old male and female mice. All mice were bred and maintained in specific-pathogen-free conditions, at 25 °C and ambient humidity under a 12 h:12 h day:night cycle, at the animal facilities of University of California, San Franciso (UCSF) or University of California, San Diego (UCSD). Sample sizes were estimated based on standard power calculations (a = 0.05, 80% power) performed for similar published experiments. Mice were used in accordance with institutional guidelines and under study protocols approved by the UCSF or UCSD Institutional Animal Care and Use Committee (protocols AN193180-01J, AN195716-01B (UCSF) and s14044 (UCSD)).
Marmosets
Eleven outbred middle-aged marmoset monkeys (Callithrix jacchus; aged >5 years; median age ~7 years) were used in this study. No siblings were used. Animals were housed in family groups (12 h:12 h light:dark cycle, temperature 31 °C, humidity 65%). Experiments were conducted according to the Australian Code of Practice for the Care and Use of Animals for Scientific Purposes and were approved by the Monash University Animal Ethics Committee. Marmosets were obtained from the National Nonhuman Primate Breeding and Research Facility (Monash University, Australia).
Tamoxifen-induced Cre recombination
Mice were injected intraperitoneally with 200 μl of tamoxifen (Sigma-Aldrich) dissolved in corn oil at 10 mg ml−1. For transcranial activation, 4-OH-tamoxifen (Sigma-Aldrich), the active metabolite of tamoxifen78, was dissolved in acetone at 100 μg ml−1. One-hundred microlitres of 4-OHT was applied to a specific cranial location, determined stereotactically as described in ‘Photothrombotic injury’. Tamoxifen was applied via micropipette (approximately 5 μl at a time) and allowed to evaporate in between applications.
Photothrombotic injury
For PT injury surgeries79,80,81, mice were anaesthetized via inhaled isoflurane, shaved on the scalp, and stereotaxically fixed. After sterilization with iodine and 70% ethanol, 0.5% lidocaine was administered subcutaneously to the scalp. The cranium was surgically exposed, and a fibre optic white light is placed over the S1 cortex (3.0 mm, −0.5 mm (x, y) from bregma, with coordinates determined via stereotax). Mice were injected intraperitoneally with 8 mg kg−1 Rose Bengal dye; after 1 min, the cranium was exposed to high intensity white light (150 W) for 2 min. The scalp was sutured using nylon sutures and surgical glue, and buprenorphine was administered.
TBI
For TBI surgeries82, mice were anaesthetized and stereotaxically fixed, as above. A 3-mm craniotomy was performed over the right S1 centred at –1 mm posterior from bregma, +3 mm lateral from the midline. TBI was performed with a CCI device (Impact One Stereotaxic Impactor for CCI, Leica Microsystems) equipped with a metal piston using the following parameters: 3 mm tip diameter, 15° angle, 0.8 mm depth from the dura, 3 m s−1 velocity, and 100 ms dwell time. Sutures were administered as above.
tMCAO
Mice (age postnatal day (P)25–45) underwent focal ischaemia–reperfusion with tMCAO for 3 h, or sham surgery, as detailed previously33,83,84. In brief, the right internal carotid artery (ICA) was dissected, and a temporary ligature was tied using a strand of 6-0 suture at its origin. This ligature was retracted laterally and posteriorly to prevent retrograde blood flow. A second suture strand was looped around the ICA above the pterygopalatine artery and an arteriotomy was made proximal to the isolated ICA. A silicone coated 6-0 nylon filament from Doccol Corporation was inserted 6.5−7 mm to occlude the MCA and the second suture strand was tied off to secure the filament for the duration of occlusion. Injury was confirmed by severe left frontal/hindlimb paresis resulting in circling movements during the occlusion period. For reperfusion, each animal was anaesthetized and all suture ties and the occluding filament were removed. Avitene Microfibrillar Collagen Hemostat was placed over the arteriotomy and the skin incision was closed. According to pre-established criteria, cohorts displaying excessive bleeding at time of reperfusion were excluded. Cohorts with significantly underweight mice (75% or less of expected weight at time of surgery) were excluded from survival analysis. Mice that died during surgery were also excluded. Sham animals were anaesthetized for 15 min, equivalent to surgery procedure time; at the time of reperfusion, the sham animals were once again anaesthetized for 5 min, equivalent to the reperfusion procedure time for tMCAO animals.
dMCAO
dMCAO surgeries were performed as previously described85. In brief, mice were sedated with isoflurane and analgesic administered before surgery. An incision was made between the eye and ear, the temporal muscle retracted, and the skull removed above the distal middle cerebral artery (dMCA). The dMCA was then occluded with a bipolar coagulator which clots the artery, reducing blood flow to the ipsilateral cortex. Following this procedure, the incision was closed and mice monitored daily.
Marmoset stroke
Induction of focal stroke to the marmoset primary visual cortex (V1) was performed by vasoconstrictor-mediated vascular occlusion of the calcarine branch of the posterior cerebral artery (PCAca), as detailed previously86,87. In brief, following anaesthesia (Alfaxalone 5 mg kg−1; maintained with inspired isoflurane 0.5–4%), a craniotomy and dural thinning was performed, followed by intracortical injections of endothelin-1 (ET-1; 0.1 µl per 30 s pulse at 30 s intervals, totalling ~0.7 µl over 7 sites) proximal to the PCAca, which supplies operculum V1. The craniotomy was replaced, secured with tissue adhesive (Vetbond; 3 M) and the skin sutured closed. Monkeys recovered for 7 days (n = 2), 6 weeks (n = 2) and 1 year (n = 2; equal numbers of each sex).
Vitals monitoring
The MouseOx Pulse Oximeter system (STARR Life Sciences) was used to measure arterial oxygen saturation from awake mice. Mice were shaved at time of injury or prior to measurement. Mice were measured at 5 measurements per second for at least 5 min and at least 10 successful readings. All successful measurements (error code = 0) were averaged for each mouse.
For blood pressure and heart rate measurements, mice were restrained and placed on a warming platform. Measurements were taken using the CODA-HT4 Noninvasive Blood Pressure System (Kent Scientific), using default settings for sets, cycles, deflation time and failed cycle exclusion. Accepted cycles were averaged per mouse. Mean arterial pressure (MPA) was calculated as MAP = (systolic pressure + 2 × (diastolic pressure))/3.
Vitals were recorded on day 3 unless mice appeared to be decompensating, in which case measurements were taken daily, beginning days 1–2, to maximize data collection; the latest available (pre-death) measurement was taken for each mouse.
Liposome injection
Clodronate liposomes or empty control liposomes (Encapsula Nanoscience) were randomly assigned and administered intraperitoneally, beginning with injection 3 days prior to injury (200 μl) followed by injection on the day of injury (100 μl) and every 3 days subsequently (100 μl) until collection (14 dpi).
Antibody injection
ADWA1188,89 (generously provided by D. Sheppard) or IgG1 isotype control (InVivoMab) was randomly assigned, diluted in sterile DPBS to 10 mg kg−1, and injected intraperitoneally on the day of injury (0 dpi) and every week subsequently until collection (7 dpi, etc.).
EdU injection
To measure proliferation, 5-ethynyl-2′-deoxyuridine (EdU, Thermo Scientific) was reconstituted at 5 mg ml−1 in DPBS and injected at 50 mg kg−1 intraperitoneally. Mice were injected either every other day throughout the injury time course or 2 h prior to euthanasia.
Mouse tissue processing for imaging
Following CO2 euthanasia, mice were transcardially perfused with 10 ml DPBS and 10 ml of 4% paraformaldehyde (PFA) (Thermo Scientific). Skullcaps and/or brains were removed from bases. Skullcaps (for meningeal imaging) or skullcaps and brains were fixed overnight (4% PFA, 4 °C). Skullcaps/brains were washed (DPBS) and decalcified in 0.3 M EDTA (VWR) (1 week, 4 °C) followed by cryoprotection (30% sucrose). When removed from skullcaps, brains were cryoprotected directly after fixation. Brains were frozen in O.C.T. (Thermo Scientific) on dry ice and sliced to indicated thickness on a cryostat (Leica). Spinal cords processed similarly (decalcifying intact vertebra). For spatial transcriptomics, tissue processing was performed as above, without PFA perfusion and fixation. Brains were removed from skullcaps and directly frozen. For quantitative imaging of (blinded) PT lesions, 2× 14-μm sections were collected per 100 μm sliced, and sections representing a lesion’s maximal cross-sectional area were stained, imaged and quantified. Lesion size outliers (diameter <25% of normal, representing technical errors during injury) were excluded prior to unblinding, resulting in one exclusion from Fig. 4g,h, and one exclusion from Fig. 4p and Extended Data Fig. 8ab. Mice with hydrocephalus were also excluded.
Marmoset tissue processing for imaging
For immunofluorescence, naive controls (n = 2; equal numbers of each sex) and post-stroke marmosets were administered an overdose of pentobarbitone sodium (100 mg kg−1; intraperitoneal injection). Following apnoea, animals were transcardially perfused with 0.1 M heparinized PBS, followed by 4% PFA in PBS (0.1 M). Brains were dissected, post-fixed and cryoprotected, as outlined previously26,86,87. Following separation of the hemispheres, each hemisphere was bisected coronally at the start of the caudal pole of the diencephalon and frozen in liquid nitrogen at −40 °C. Tissue was cryosectioned in the parasagittal plane at −20 °C to obtain 40 µm for free-floating sections stored in cyroprotectant solution (50% PBS 0.1 M, 20% ethylene glycol, 30% glycerol).
Immunohistochemistry
Slide-mounted thin sections were thawed, washed (DPBS) and blocked (1 h, DPBS/0.4% Triton X-100/5% secondary host serum). For select antibodies, antigen retrieval was performed prior to blocking, involving: (1) incubation for up to 15 min in Liberate Antibody Binding Solution (Polysciences) followed by a 5 min wash in PBS (ASPA); or (2) incubation for 3 (ALDH1A2 (thin section only), ALPL, αSMA (thin section only), FGF13, LAMA1, SEMA3C, CDH18) or 5 min (cleaved caspase-3 (cCasp3)) in 0.01 M Na3C6H5O7 (heated to 95 °C in a water bath), followed by cooling to room temperature (20 min) and 3 washes in PBS (5 min each). Samples were then incubated in primary antibodies diluted in blocking solution (room temperature, 1 h, or 4 °C, overnight). Samples were washed (DPBS/0.05% Triton X-100, 5 min, 3 times) and incubated in secondary antibodies diluted 1:1,000 in blocking solution (room temperature, 45–60 min). Samples were washed, mounted in DAPI Fluoromount-G (Thermo Scientific), and imaged. Proliferation was measured using the Click-iT EdU Alexa Fluor 647 Imaging Kit (Thermo Scientific), according to the manufacturer’s instructions (in between primary and secondary staining steps). FluoroJade C was stained using the FluoroJade C Ready-to-Dilute Staining Kit (VWR), according to the manufacturer’s instructions.
Medium-thickness sections (30–50 μm) were blocked in 250 μl DPBS/0.25% Triton X-100/5% secondary host serum (1 h, room temperature). Samples were subsequently incubated in primary antibody diluted in blocking solution (4 °C, overnight), washed (DPBS/0.05% Triton X-100, 5 min, 4 times), and incubated in secondary antibodies diluted 1:500 in blocking solution (room temperature, 45–60 min). Samples were washed three times, mounted in DAPI Fluoromount-G, and imaged.
Meninges were blocked and stained before removal from skullcaps (block: 4 °C, overnight, in 2 ml DPBS/0.3% Triton X-100/5% FBS/0.5% BSA/0.05% NaN3). Samples were incubated in primary antibody diluted in 600 μl of staining solution (DPBS/0.15% Triton X-100/7.5% FBS/0.75% BSA/0.075% NaN3; 4 °C, 72 h), washed (DPBS/0.15% Triton X-100, 4 °C, 30 min, 3 times), and incubated in secondary antibodies diluted 1:400 in staining solution solution (4 °C, 24 h). Samples were washed and incubated in 10 μg ml−1 DAPI in PBS (room temperature, 1 h). Dural meninges were subsequently micro-dissected from skullcap, mounted in 50–100 μl of Refractive Index Matching Solution (RIMS, DPBS/Histodenz (133.33 g per 100 ml)/0.017% Tween-20/0.17% NaN3), and imaged.
Thick sections (100–200μm) were stained using the iDISCO protocol90 with 1–2 days of permeabilization, 1–2 days of blocking, 3 days of primary antibody and 3 days of secondary antibody.
For marmoset imaging, free-floating sections comprising the infarct and peri-infarct regions were selected and washed in PBS (0.1 M) and pre-blocked in a solution of 10% normal goat serum in PBS + 0.3% Triton X-100 (TX; Sigma) before incubation with primary antibodies overnight at 4 °C. Sections were rinsed in 0.1% PBS-Tween and incubated with secondary antibodies (1 h). After washes in PBS, sections were treated with DAPI, mounted in Fluoromount-G, and imaged.
For haematoxylin and eosin (H&E) imaging and liver cCasp3 immunohistochemistry, histology was performed by HistoWiz ( using a standard operating procedure and fully automated workflow. Samples were processed, embedded in OCT, and sectioned at 4 μm for H&E staining or cCasp3 immunohistochemistry. Immunohistochemistry was performed on a Bond Rx autostainer (Leica Biosystems) with enzyme treatment (1:1,000) using standard protocols. Bond Polymer Refine Detection (Leica Biosystems) was used according to the manufacturer’s protocol. After staining, sections were dehydrated and film coverslipped using a TissueTek-Prisma and Coverslipper (Sakura). Whole slide scanning (40×) was performed on an Aperio AT2 (Leica Biosystems).
Imaging antibodies
Primary antibodies used for mouse imaging include chicken anti-GFP (Aves Labs GFP-1020, 1:200), rabbit anti-dsRed (Takara 632496, 1:300), chicken anti-GFAP (Invitrogen PA1-10004, 1:200 or 1:500), rat anti-GFAP (2.2B10, Invitrogen 13-0300, 1:200), rat anti-ER-TR7 (Novus Biologicals NB100-64932, 1:200), rabbit anti-αSMA (Abcam ab5694, 1:300), rat anti-CD31 (MEC13.3, Biolegend 102514, 1:200), goat anti-Desmin (GenWay Biotech GWB-EV0472, 1:200), rat anti-PDGFRβ (APB5, Invitrogen 14-1402-82, 1:500), rabbit anti-NG2 (Millipore Sigma ab5320, 1:500), goat anti-Decorin (Novus Biologicals AF1060, 1:200), goat anti-collagen 1 (Southern Biotech 1310-01, 1:200 or 1:500), rabbit anti-collagen 6α1 (Novus Biologicals NB120-6588, 1:200), rat anti-periostin (345613, Novus Biologicals MAB3548, 1:200), rat anti-ICAM1 (YN1/1.7.4, Biolegend 116110, 1:200), Syrian hamster-anti-CD3ε (500A2, BD Biosciences 553238, 1:200), goat anti-S100A8 (R&D Systems AF3059, 1:200), chicken anti-NeuN (Millipore Sigma ABN91, 1:200), rabbit anti-IBA1 (Aif3, Fujifilm Wako 019-19741, 1:200–1:1,000), mouse anti-FGF13 (N235/22, Invitrogen MA5-27705, 1:100), goat anti-CD80 (R&D Systems AF740, 1:200), goat anti-ALPL (Novus Biologicals AF2910, 1:50), rabbit anti-LAMA1 (EPR27258-37, Abcam ab307542, 1:200), rabbit anti-SEMA3C (Invitrogen PA5-103168, 1:100), rabbit anti-CDH18 (Invitrogen PA5-112902, 1:50), rabbit anti-ALDH1A2 (Novus Biologicals NBP2-92915, 1:200), goat anti-SOX10 (R&D Systems AF2864, 1:300), rabbit anti-ASPA (Genetex GTX113389, 1:1,000), mouse anti-E-cadherin (Clone 36, BD Biosciences 610181), rat anti-I-A/I-E (MHCII, M5/114.15.2, eBioscience 14-5321-82), mouse anti-Ly76 (TER119, Biolegend 116232, 1:200), goat anti-mouse IgM (Invitrogen 31172, 1:200), and rabbit anti-cCasp3 (Cell Signaling Technology 9661T, 1:400; Cell Signaling Technology 9991, Histowiz). For marmoset imaging, rabbit anti-COL6 (Abcam ab6588, 1:500) was used.
Secondary antibodies were used at 1:500 (for thin sections) and 1:1,000 (for thicker sections), as specified in relevant Methods sections. Secondary antibodies used include donkey anti-rat IgG AF488 (Thermo Scientific A21208), donkey anti-rat IgG AF555 (Thermo Scientific A78945), donkey anti-rat IgG AF647 (Abcam ab150155), donkey anti-rabbit IgG AF488 (Thermo Scientific A21206), donkey anti-rabbit IgG AF555 (Thermo Scientific A31572), donkey anti-rabbit IgG AF647 (Thermo Scientific A31573), donkey anti-goat IgG AF488 (Thermo Scientific A11055), donkey anti-goat IgG AF555 (Thermo Scientific A21432), donkey anti-goat IgG AF647 (Thermo Scientific A21447), donkey anti-chicken IgG AF488 (Sigma, SAB4600031-250UL), donkey anti-chicken IgG AF647 (Thermo Scientific A78952), goat anti-rat IgG AF488 (Thermo Scientific A11006), goat anti-rabbit IgG AF555 (Thermo Scientific A21429), goat anti-rabbit IgG AF647 (Thermo Scientific A21245), goat anti-hamster IgG AF647 (Thermo Scientific A21451) and donkey anti-mouse IgG AF647 (Thermo Scientific A31571).
Confocal and wide-field microscopy
Confocal images (for thick sections, immune cell quantification and some thin sections) were imaged using a Nikon A1R laser scanning confocal including 405, 488, 561 and 650 laser lines for excitation and imaging with 16×/0.8 NA Plan Apo long working distance water immersion, 20×/0.95 NA XLUM PlanFl long working distance water immersion, or 60×/1.2 NA Plan Apo VC water immersion objectives. z-steps were acquired every 4 μm. Wide-field images (for thin sections, fibrosis quantification and lesion size quantification) were imaged using a Zeiss Axio Imager.M2 wide-field fluorescent microscope with a 10×/0.3 or 20×/0.8 air objective, or on a Leica Aperio Versa 8 Slide Scanner with a HC PL APO 20X/0.75 CS2 air objective. Marmoset images were acquired using the VS200 slide scanner (Olympus).
Image analysis and quantification
z-stacks were rendered in 3D and quantitatively analysed using Bitplane Imaris v9.8 software package (Andor Technology). Individual cells (for example, lymphocytes, fibroblasts, etc.) or fibrotic and glial scars were annotated using the Imaris surface function, thresholding on fluorescent signal (based on antibody staining or reporters when available), along with additional co-stains (for example, CD45, IBA1 and CD3ε), DAPI staining and size or morphological characteristics; when helpful, background signal in unrelated channels was excluded. Colocalization was determined using ‘intensity mean’, ‘intensity min’ (for DAPI) or ‘intensity max’ (for select secreted factors). Parameter values were chosen for optimal signal-to-background balance. 3D distances between lymphocytes and stromal/glial cells were calculated using the Imaris Distance Transform Matlab extension. Proportions of immune or stromal cells within given cortical or lesional regions were determined by manually tracing regional borders (for example, GFAP–ER-TR7) and filtering cell surfaces based on inclusion. Additional surface statistics were calculated using Imaris. Lesion sizes were calculated in Fiji (ImageJ version 1) by tracing the fibroblast–astrocyte border (using ER-TR7 and/or GFAP). Fibroblast or myeloid cell coverage was determined by thresholding the relevant channel (for example, ER-TR7, IBA1 and tdT), using a consistent threshold for each slice except for cases of exceptionally high tissue background. Fibroblast or pericyte coverage after dMCAO (Extended Data Fig. 1) was determined using particle analysis, adjusting thresholds based on image background. In both cases, coverage was calculated as thresholded area normalized to lesion area, and data were subsequently unblinded. For pooled data, equivalent thresholds were applied across experiments when possible, with exceptions made for varying tissue background. Lineage-traced pericytes were manually counted. Border thickness was calculated in Fiji, using a macro to calculate the distance between each (manually traced) border edge at points 100 µm apart, followed by averaging these distances. Midline shift was calculated from coronal images by tracing: (1) a line connecting dorsal and ventral brain midpoints; (2) a curve tracking observed midline structures; and (3) a perpendicular line between lines 1 and 2. For analysis of cCasp3 puncta in liver immunohistochemistry images (Histowiz), five representative fields of view (10×) were captured per slide; for each, immunohistochemistry signal was isolated in ImageJ via colour deconvolution, cCasp3 was thresholded, and puncta were counted using the ImageJ particle analysis function.
Serum chemistries
Serum analysis was performed by the Unit for Laboratory Animal Medicine Pathology Core (University of Michigan). Whole blood was collected into serum separator tubes, allowed to clot, and separated into serum by centrifugation. Serum chemistries were run on an AU480 Chemistry Analyzer (Beckman Coulter) using the manufacturer’s provided reagents. For relevant analytes (for example, alanine transaminase), severely haemolysed samples were excluded according to Unit for Laboratory Animal Medicine and manufacturer guidelines.
Tissue processing for Visium
Tissue was collected from 1× resting mouse, 1× 2 dpi mice, 2× 7 dpi mice and 4× 21 dpi mice (including 2 Cthrc1creER; Rosa26DTA mice, discarded after initial clustering due to undetectable deletion; PT injury). Tissue was collected as above. Ten-micrometre slices were prepared via cryostat and directly mounted onto Tissue Optimization and Spatial Gene Expression slides (10X Genomics). Tissue optimization was carried out as per manufacturer’s recommendations, resulting in an optimal tissue permeabilization time of 12 min. Tissue mounted on the Spatial Gene Expression slide was processed per the manufacturer’s recommendations, imaged on a Leica Aperio Versa slide scanner at 20×, and transferred to the Gladstone Genomics Core for library preparation according to the manufacturer’s protocol. Samples were subsequently transferred to the UCSF Center for Advanced Technology for sequencing on the NovaSeq 6000 system.
Tissue processing for mouse nuclear isolation
Nuclear isolation was employed to overcome limitations of traditional flow cytometry and/or scRNA-seq, including ECM interference with fibroblast isolation and dissociation signatures that disproportionately affect stromal and immune cells91. For nuclear flow cytometry, tissue was collected from resting, 7 or 14 dpi (PT injury) Pdgfra-GFP mice. For snRNA-seq experiment 1 (wild-type time course), tissue was collected from 2 wild-type mice per time point (1 male and 1 female) at rest, 2 dpi, 7 dpi and 21 dpi. Two mice per time point with impaired TGFβ signalling (Col1a2creER; Tgfbr2flox and Cdh5creER; Tgfbr2flox) were collected at 7 and 21 dpi but were discarded after initial clustering and not analysed separately due to insufficient yield. For snRNA-seq experiment 2 (wild-type, Tgfbr2-cKO and ADWA11-treated mice), tissue was collected from 2 wild-type mice, 2 Col1a2creER; Tgfbr2flox mice, and 2 ADWA11-treated mice at 7 and 21 dpi (1 male and 1 female per time point or condition). For snRNA-seq experiment 3 (wild-type and Cxcl12-cKO), tissue was collected from 2 wild-type and 2 Col1a2creER; Cxcl12flox mice at 21 dpi.
Following CO2 euthanasia, mice were transcardially perfused (10 ml DPBS) and decapitated. Brains were removed from skullcaps and placed in iMED+ (15 mM HEPES (Fisher) and 0.6% glucose in HBSS with phenol red)92. For nuclear flow cytometry and snRNA-seq experiments 1 (time course) and 3 (Cxcl12-cKO), dura, lesion and perilesional cortex (with or without contralateral cortex) were micro-dissected and processed separately. For snRNA-seq experiment 2 (wild-type, Tgfbr2-cKO and ADWA11), only lesions were dissected. Microdissection involved meningeal/skullcap separation, removal of subcortical structures, and separation of lesions from skullcaps (lesions often separate from cortex during initial dissection but can be micro-dissected as necessary). For snRNA-seq, tissue from male and female mice within experimental conditions was combined.
Tissue was processed using ST-based buffer protocol91, with the following modifications: initial centrifugation was performed at 500g for 10 min. After lysis and initial centrifugation, nuclei were resuspended in 1 ml ST buffer (nuclear flow cytometry, RNA-sequencing experiment 2) or PBS/1% BSA/0.2 U μl−1 Protector RNAse inhibitor (Roche) (RNA-sequencing experiments 1 and 3), filtered through 35-μm cell strainers, and subsequently processed as below.
For nuclear flow cytometry and snRNA-seq experiment 2, nuclei were centrifuged for 5 min at 500 g, resuspended in FANS buffer (DPBS/1% BSA/0.1 mM EDTA)93 with 2 μg μl−1 DAPI and 0.2 U μl−1 RNase inhibitor (snRNA-seq), and stained (nuclear flow) or sorted (snRNA-seq).
For snRNA-seq experiments 1 and 3, cell counts were performed after initial centrifugation (NucleoCounter, Chemometic), and a maximum of 2 × 106 nuclei were multiplexed using CellPlex Multiplexing technology (10X Genomics) according to the manufacturer’s instructions (using protocol 1 for nuclear multiplexing, with only one wash after multiplexing to increase yield). Nuclei were resuspended in FANS buffer with 0.2 U μl−1 RNase inhibitor and 2 μg μl−1 DAPI and nuclear concentrations were determined. Immediately before sorting, multiplexed microanatomical regions (including lesion, parenchyma and dural meninges) from individual mice were combined at desired ratios (75% lesion, 25% parenchyma). Nuclei were sorted (forward and side scatter and DAPI) into pre-coated tubes containing 200 μl PBS/1% BSA/0.2 U μl−1 RNase inhibitor (BDFACSAria II sorting system, 100 μm nozzle size, 4 way-purity sort mode). Resting dural fibroblasts were enriched using Col1a2creER; Rosa26Sun1GFP and resting nuclei were combined after sorting (50% GFP+ meninges, 25% GFP− meninges, 25% parenchyma).
After sorting and pooling (snRNA-seq), samples were centrifuged (500g, 15 min), and supernatant was removed to leave a minimum final volume of 45 μl with a maximum of 16,500 nuclei. Final nuclear concentrations were acquired, and samples were transferred to the UCSF Genomics CoLab for library preparation. Up to 1.6 × 104 nuclei were loaded onto the Chromium Controller (10X Genomics). Chromium Single Cell 3′ v3.1 reagents were used for library preparation according to the manufacturer’s protocol. Libraries were transferred to the UCSF Institute for Human Genetics for sequencing on the NovaSeq 6000 system.
Marmoset tissue processing for nuclear isolation
For single-nuclei RNA sequencing (snRNA-seq), naive control marmosets (n = 3; 1 female, 2 male; median age 4 years) were administered an overdose of pentobarbitone sodium (100 mg kg−1; intraperitoneal). Following apnoea, frontal lobes were recovered and dissected under aseptic conditions in sterile ice-cold phosphate buffered saline (PBS; 0.1 M; pH 7.2). Tissues were and snap frozen in isopentane chilled in liquid nitrogen. The procedures and dissections were performed in chilled RNAase-free PBS with RNase-free sterilized instruments under RNase-free conditions. Approximate time from apnoea to snap-freezing ranged from 20–30 min. All six samples passed quality control. Nuclear isolation was performed as described previously26, involving pulverization in liquid nitrogen, lysis in lysis buffer, dounce homogenization, and gradient purification. After isolation, nuclei were counted and diluted to 1 million per ml with sample-run buffer (0.1% BSA, RNAse inhibitor (80 U ml−1), 1 mM DTT in DPBS). snRNA-seq was performed on the 10x Genomics Chromium System. Cellranger commercial software was utilized to conduct initial data processes including sequence alignment to the marmoset genome (CalJac3).
Tissue processing for mouse single-cell isolation
Single-cell suspensions were prepared from tissues including brain, spinal cord, meninges, blood and spleen. Immediately following CO2 euthanasia, spleens were removed into RPMI/10% FBS and peripheral blood was collected through the right ventricle into heparin tubes. Mice were subsequently transcardially perfused through the left ventricle with 10 ml DPBS, decapitated, and brains were carefully removed from skullcaps and placed in iMED+ as above92. For select experiments, spinal cords were carefully dissected from vertebra. Cortex, lesion and meninges were dissected as above. Brain was weighed and subsequently homogenized in iMED+ homogenized using a 2-ml glass tissue grinder (VWR; 6 plunges, followed by filtration through a 70-μm filter, addition of 2 ml iMED+ and 6 more plunges). Filtered suspensions were centrifuged at 220g for 10 min and resuspended in 5 ml of 22% Percoll (GE Healthcare) in Myelin Gradient Buffer (5.6 mM NaH2PO4•H2O, 20 mM Na2HPO4•2H2O, 140 mM NaCl, 5.4 mM KCl, 11 mM glucose in H2O)92. PBS (1 ml) was layered on top of Percoll. Samples were centrifuged at 950g for 20 min at 4 °C with no break to separate myelin and resuspended in fluorescence activated cell sorting (FACS) buffer. Dissected meninges were incubated in digestion medium (RPMI/10% FBS/80 μg ml−1 DNase I/40 μg ml−1 Liberase TM (Roche)). Tissue was subsequently mashed through 70-μm filters, followed by centrifugation and resuspension in FACS buffer. Spleens were prepared by mashing tissue through 70-μm filters without tissue digestion, followed by centrifugation. Red blood cells were lysed for 2 min using 1× Pharm-Lyse and the remaining cell pellet were resuspended in FACS buffer. Blood samples were centrifuged for 5 min at 500g. Pellets were resuspended in 1× Pharm-Lyse 5 min at room temperature, followed by centrifugation and suspension in FACS buffer. For cytokine restimulation assays, samples were transferred to U-bottom plates and incubated in stimulation medium (RPMI supplemented with 10% FBS, 1% penicillin/streptomycin, 1× Glutamax (Thermo Scientific), 1× HEPES buffer (Fisher), 1× non-essential Amino Acids (Thermo Scientific), 1 mM NaC3H3O3 (Thermo Scientific), 55μM b-mercaptoethanol, 1× Cell Stimulation Cocktail (Tonbo), and 1× Brefeldin A (Thermo Scientific)) at 37 °C for 3 h, followed by centrifugation and transfer to a V-bottom plate.
Flow cytometry
Resuspended samples were stained in 96-well V-bottom plates. Surface staining was performed at 4 °C for 45 min in 50 μl staining volume. For experiments involving intracellular staining (including transcription factor and cytokine staining), cells were fixed and permeabilized using Foxp3 Transcription Factor Staining Buffer Set (eBioscience) followed by staining at 4 °C for 1 h in 50 μl staining volume. All samples were acquired on a BD LSRII Fortessa Dual or a BD FACSAria II for cell sorting. Live cells or nuclei were gated based on their forward and side scatter followed by Zombie NIR fixable (Biolegend 423106), Fixable Viability Dye eF780 (eBioScience 65086514), Draq7 (Biolegend 424001) or DAPI (40,6-diamidine-20-phenylindole dihydrochloride; Millipore Sigma D9542-10MG) exclusion (or inclusion for nuclei). Lineages were subsequently identified as follows:
Oligodendrocyte-lineage nuclei were identified as DAPI+Pdgfra-GFPhiOlig2+. Fibroblast nuclei were identified as DAPI+Pdgfra-GFPint or DAPI+ Col1a2creER; Rosa26Sun1GFP+ (snRNA-seq experiment 1, resting dural meninges). Bulk nuclei were identified as DAPI+. Global lymphocytes were defined as CD45+Thy1+. T cells were identified as CD45+CD11b−CD19−NK1.1−CD3ε+CD4+ (CD4 T cells; further subset as FOXP3+ (regulatory T cells) or FOXP3− (conventional CD4 T cells)) CD8α+ (CD8 T cells) or TCRγδ+ (γδ T cells) and were further defined as CD44+CD69+ (resident memory T cells (TRM)), CD62L+ (naive T cells), CD62L−CD44+ (activated T cells) or CTVdiluted (proliferating T cells). Additionally, CD4 T cells were defined as TBET+ (TH1 cells), GATA3+ (TH2 T cells) or RORγt+ (TH17 T cells), and various T cell subsets were defined as cytokine-positive or negative (IFNγ, IL-17A or IL-10). Neutrophils were defined as CD45+CD11b+Ly6G+ (and optionally Thy1−CD19−NK1.1−). Monocytes were defined as CD45+CD11b+Ly6G−Ly6C+ (and optionally Thy1−CD19−NK1.1−Siglec F−). Microglia were defined as CD45intCD11b+. Macrophages were defined as CD45+Ly6G−Ly6G−CD64+ (optionally MERTK+). Microglia/macrophages were further defined as DAM/SAM (CD9+ and CD63+/TREM2+). cDCs were identified as CD45+Ly6G−Ly6C−CD64−MHCII+CD11c+, and were further defined as cDC1s (CD11blo, optionally SIRPα−) or cDC2s (CD11bhi, optionally SIRPα+). B cells were defined as CD45+Thy1−CD19+. Eosinophils were defined as CD45+Thy1−CD19−NK1.1−Ly6G−CD11b+Siglec F+. Populations were backgated to verify purity and gating.
Data were analysed using FlowJo software (TreeStar) and compiled using Prism (Graphpad Software). Cell counts were performed using flow cytometry counting beads (CountBright Absolute; Life Technologies) per manufacturer’s instructions.
Flow cytometry antibodies
Antibodies used for flow cytometry include rabbit anti-OLIG2 (Thermo Scientific P21954, 1:100), anti-CD45 (30-F11, BD Biosciences 564279 or Biolegend 103132 or 103104, 1:400), anti-CD90.2 (Thy1, 53-2.1, Biolegend 140327, BD Biosciences 553004, 1:200), anti-CD11b (M1/70, Biolegend 101224 or BD Biosciences 563015, 1:400), anti-CD19 (6D5, Biolegend 115554, 1:400), anti-NK1.1 (PK136, Biolegend 108736, 1:200), anti-CD3ε (17A2, Biolegend 100216, 1:200), anti-CD4 (RM4-5, Biolegend 100557 or GK1.5, BD Biosciences 563050, 1:200), anti-CD8α (53-6.7, Biolegend 100750, 1:200), anti-CD44 (IM7, Biolegend 103030, 1:200), anti-CD69 (H1.2F3, Biolegend 104505, 1:200), anti-CD62L (MEL-14, Biolegend 104407, 1:200), anti-TBET (4B10, Biolegend 25-5825-80, 1:100), anti-GATA3 (TWAJ, eBioscience 12-9966-41, 1:100), anti-RORγt (B2D, eBioscience 17-6981-82, 1:100), anti-Ly6G (1A8, Biolegend 127624, 1:200), anti-IFNγ (XMG1.2, Biolegend 505810, 1:100), anti-IL-17A (TC11-18H10.1, Biolegend 506922, 1:100), anti-IL-10 (JES5−16E3, eBioscience 12-7101-81, 1:100), anti-TCRγδ (Biolegend 118118, 1:200 (extracellular) or 1:400 (intracellular)), anti-FOXP3 (eBioscience 53-5773-82, 1:100), anti-Ly6C (HK1.4, Biolegend 128011 or 128035, 1:400), anti-CD64 (X-54-5/7.1, Biolegend 139323 or BD Biosciences 558539, 1:200), anti-MERTK (DS5MMER, eBioscience 46-5751-80, 1:200), anti-CD9 (KMC8, BD Biosciences 564235, 1:200), anti-TREM2 (237920, R&D systems FAB17291A, 1:200), anti-CD63 (NVG-2, Biolegend 143904, 1:200), anti-I-A/I-E (MHCII, M5/114.15.2, BD Biosciences 748845, 1:400), anti-CD11c (N418, Biolegend 117339 or 117318, 1:200), anti-CD172a (SIRPα, P84, eBioscience 12-1721-80, 1:200), anti-Siglec-F (E50-2440, BD Biosciences 740956, 1:200), anti-podoplanin (gp38, 8.1.1, Biolegend 127412, 1:200), anti-CD31 (390, Biolegend 102404 or 102408, 1:200), anti-EpCAM (G8.8, Biolegend 118230, 1:200), anti-PDGFRα (APA5, Biolegend 135908, 1:200), anti-Sca-1 (Ly-6A/E, D7, Biolegend 108131, 1:200), anti-phospho-SMAD3 (EP823Y, Abcam ab52903, 1:50) and anti-CD16/32 (2.4G2, BD Biosciences 553142, 1:100 or 1:250).
Ex vivo coculture
For ex vivo coculture experiments, lesions and contralateral cortex were dissected as described above and divided into halves or equivalently sized sections (contralateral cortex). Tissue was added to 96-well round bottom plates in 100 μl of R10 (RPMI supplemented with 10% FBS, 1% penicillin/streptomycin, 1× Glutamax (Thermo Scientific), and 55 μM β-mercaptoethanol).
For myeloid cell cocultures, single-cell suspensions from perilesional (ipsilateral) cortex were generated and stained as above. Homeostatic microglia (negative for DAM markers (CD9 and CD63)) were sorted into pre-coated tubes containing 3 ml R10 (BDFACSAria II sorting system, 100 μm nozzle size, 4 way-purity sort mode) and checked for purity post-sort. Cells were subsequently counted, and a maximum of 1 × 107 cells were labelled with 5 mM CellTrace CFSE in PBS (Thermo Scientific) for 10 min, followed by washing with R10, centrifugation, and resuspension at 4.5 × 104 cells per ml. One-hundred microlitres of suspension (containing 45,000 microglia) was subsequently plated with lesions or alone.
For T cell cocultures, T cells were isolated from spleens and cervical/inguinal lymph nodes using EasySep Magnetic Bead negative selection (Stem Cell), according to the manufacturer’s instructions (using 31.5 μl of vortexed selection beads). A maximum of 1 × 107 T cells were labelled with 5 mM CellTrace Violet in PBS (Thermo Scientific) for 10 min, followed by washing with R10, centrifugation, and resuspension at 106 cells per ml. One-hundred microlitres of suspension (containing 100,000 T cells) was subsequently plated with lesions, control tissue, or alone. As a positive control, anti-CD3/CD28 T cell activating DynaBeads (Thermo Scientific) were magnetically washed in 1 ml DPBS and added to T cells at a 1:1 ratio.
For purified fibroblast cocultures, meninges were processed as above. Lungs were perfused, dissected, and digested in PBS with Dispase II (15 U ml−1), collagenase 1 (22,500 U ml−1), and DNAse 1 (10 mg ml−1) for 30 min at 37 °C, followed by centrifugation, filtration, and RBC lysis as above. Endothelial and hematopoietic cells were negatively selected using magnetic beads per manufacturers’ instructions. After staining as above, fibroblasts were sorted as Lin−PDGFRα+gp38+ (meningeal) or PDGFRα+gp38+Sca-1+ (lung adventitial), plated at up to 15,000 per well in a flat bottom 96-well plate, and cultured for 7 days prior to initiation of lymphocyte coculture, as above.
Plates were incubated at 37 °C, 5% CO2 for 72 h. After coculture completion, wells were mixed by pipetting and cells were transferred to a V-bottom plate, and blocked, stained, and analysed as above. For cocultured wells, Cell Trace Violet (CTV) or carboxyfluorescein diacetate succinimidyl ester (CFSE) staining was used to identify plated (versus lesion-resident) T cells or microglia; given the low rates of proliferation observed, gates were gates chosen to maximize exclusion of unlabelled cells while still including proliferating or CTV-diluting cells (up to at least three rounds of division). Figure 3s and Extended Data Fig. 6t–v include pooled control data from experiments where wild-type lesions were cultured with diphtheria toxin (100 ng ml−1; any lesions expressing Rosa26DTR were excluded), empty liposomes, or IgG1 isotype control (BioXCell). For direct comparisons between 7 dpi and 21 dpi lesions, media refeeding was performed daily (replacing 100 µl, 2×) to mitigate rapid media acidification by 7 dpi lesions.
Vascular permeability and haemorrhage analysis
Vascular permeability was measured using Evans Blue extravasation94. In brief, Evans Blue (1% in saline) was injected intraperitoneally (8 ml kg−1) 3 h prior to euthanasia. Following euthanasia, mice were transcardially perfused (10 ml DPBS), and mice that did not show appropriate liver clearing were excluded. Brains were dissected as above, whole hemispheres were weighed and added to 250 or 500 μl formamide, and tissue was incubated for 44–48 h to extract Evans Blue. Evans Blue fluorescence was measured on a SpectraMax microplate reader (excitation 620 nm, emission 680 nm). A standard curve (four-point logistic regression) was used to calculate extravasated mass, which was subsequently normalized to tissue weight.
To quantify haemorrhage (bleeding), photographs were taken of frozen brains during slicing (1 photograph per 100–150 μm). Representative photographs were chosen from across the lesion volume for quantification (3 photographs per brain). Regions of overtly visible blood (corresponding to erythroid cell accumulation, as visible by microscopy) were traced in Fiji, followed by normalization to lesion area (traced via tissue discoloration) and averaging per animal.
Visium data processing
Sequencing data were aligned to mouse genome mm10 with SpaceRanger version 2.0.0 (10x Genomics). Data were processed using the Seurat R package, version 4.2.195. Individual capture areas were processed using Seurat’s SCTransform function to normalize data, select variable features for dimensionality reduction, and scale data. Capture areas were subsequently merged for further analysis. Principle components were calculated using Seurat’s RunPCA function, followed by graph-based clustering using Seurat’s FindNeighbors (dims = 1:30) and FindClusters (res = 0.8) functions and 2D visualization using Seurat’s RunUMAP function (dims = 1:30). Feature and spatial feature plots, violin plots, and UMAP plots were generated using Seurat. We used FindAllMarkers to identify markers for each cluster (test.use = MAST, min.pct = 0.05, logfc.threshold = 0.2) and generated resulting dot plots using Seurat. Fibroblast-containing clusters were identified via expression of Col1a1 and further investigated using feature plots and spatial feature plots. One large cluster (cluster 10) was subclustered using Seurat’s FindSubCluster function (res = 0.5); subcluster 10_0 was identified as a 7 dpi fibroblast-enriched cluster and selected for further analysis based on expression of Pdgfra. Cluster 8 was identified as a 21 dpi fibroblast-enriched cluster.
Analysing wild-type tissue, we used Seurat’s FindMarkers function (min.pct = 0.05, logfc.threshold = 0.2, test.use = MAST) to determine markers for cluster 10_0 and 8 and used the EnhancedVolcano R package to visualize relevant differentially expressed genes (DEGs) (EnhancedVolcano function, minfc = 1.15, alpha = 0.05)96. To calculate fibroblast TGFβ scores, we utilized a previously generated bulk sequencing dataset of primary lung fibroblasts treated with TGFβ (1 ng ml−1) or PBS97 for 48 h. TGFβ-upregulated genes were identified from normalized data (log2(fold change (FC)) > 0, q < 0.05). Module scores were generated using Seurat’s AddModuleScore function. To calculate proliferation scores, we utilized a proliferation signature composed of 23 genes (Ccnb1, Ccne1, Ccnd1, E2f1, Tfdp1, Cdkn2b, Cdkn1a, Plk4, Wee1, Aurkb, Bub1, Chek1, Prim1, Top2a, Cks1, Rad51l1, Shc, Racgap1, Cbx1, Mki67, Mybl2, Bub1 and Plk1)98,99.
Mouse snRNA-seq data processing
Sequencing data were aligned to mouse genome mm10 (snRNA-seq experiments 1 (wild-type time course) and 2 (wild-type, Tgfbr2-cKO and ADWA11)) or Grcm39 (snRNA-seq experiment 3 (wild-type and Cxcl12-cKO)) with CellRanger version 7.1.0 (experiment 1), version 7.2.0 (experiment 2), or version 9.0.0 (experiment 3) (10x Genomics). Data were processed using the Seurat R package, version 4.2.1 (experiment 1), 5.0.1 (experiment 2), or 5.2.1 (experiment 3). We excluded cells with high mitochondrial gene expression and low or high unique molecular identifier (UMI) and feature counts, using the bottom and top 2.5 percentiles as our cutoff. We used Seurat’s SCTransform function to normalize data, select variable features for dimensionality reduction (with the removal of certain sex-related and mitochondrial genes), and scale data. Principle components were calculated using Seurat’s RunPCA function, followed by graph-based clustering using Seurat’s FindNeighbors (dims = 1:30) and FindClusters (res = 0.5 (experiments 1 and 2) or 0.25 (experiment 3)) functions and 2D visualization using Seurat’s RunUMAP function (dims = 1:30). We used FindAllMarkers (test.use = MAST, experiment 1) or RunPrestoAll100 (experiments 2 and 3) to identify markers for each cluster (min.pct = 0.05, logfc.threshold = 0.2) and annotated clusters using select common and well-validated lineage marker genes (for example, Col1a1, Col1a2, Pgdfra (fibroblasts); Cspg4 (mural); Itgam (myeloid); P2ry12, Sall1 (microglia); Cd3e, Cd4, Cd8 (T cells); Rbfox3 (neurons); Dcx, Prom1 (neural progenitors); Gfap, Aldh1l1 (astrocytes); Mbp, Olig2, Sox10 (oligodendrocytes); Olig2, Sox10, Pdgfra (oligodendrocyte precursor cells); and Pecam1 (endothelial cells)). We identified and excluded 2 clusters made up of likely doublets based on gene expression (experiments 1 and 2) and excluded 1 sample due to insufficient nuclear yield (experiment 1), visualizing resultant clusters on UMAPs and dot plots using Seurat. Final datasets comprised 28,187 cells including 8,096 fibroblasts; 189 mural cells; 4,568 myeloid cells/microglia; 548 T cells; 11,216 neurons; 2,026 astrocytes; 537 oligodendrocytes; 94 oligodendrocyte precursor cells; 470 endothelial cells; 259 neural progenitor cells; and 184 unassigned nuclei (experiment 1); 60,070 cells including 18,455 fibroblasts; 496 mural cells; 24,302 myeloid cells/microglia; 2,271 T cells; 5,994 neurons; 2,781 astrocytes; 1,669 oligodendrocytes; 2,399 oligodendrocyte precursor cells; and 1,703 endothelial cells (experiment 2); and 19,668 cells including 7,954 fibroblasts; 3,792 myeloid cells/microglia; 180 T cells; 5,862 neurons; 1,213 astrocytes; 390 oligodendrocytes; 218 endothelial cells; and 59 unassigned nuclei (experiment 3). We subsequently subset the data to wild-type cells for downstream analysis (experiment 1). FindMarkers was used to identify differentially expressed genes between time points or genotypes/conditions (test.use = wilcox). Additional feature plots, UMAP plots, dot plots and heat maps were generated using Seurat. Joint densities for gene combinations (including published meningeal layer signatures)35 were visualized using the NebulosaPlot R package (plot_density function)101 and modified using the ScCustomize R package (Plot_Density_Custom function)102.
Fibroblasts and immune cells were subset and reclustered as above (immediately prior to wild-type subsetting (experiment 1); res = 0.5 (experiment 1), 0.2 (experiment 2) or 0.3 (experiment 3) for fibroblasts, res = 0.15 for immune cells (experiment 1), res = 0.25 for myeloid cells (experiments 2 and 3), res = 0.5 for T cells (experiments 2 and 3), res = 0.25 for neurons (experiment 3)) and DEGs were recalculated. Fibroblast clusters expressing high levels of myeloid or neuronal genes were excluded as likely contaminants and removed from global UMAP and bar plots. For experiment 1, resting dural meningeal fibroblasts were removed (by microanatomical metadata and cluster) for downstream analysis. For experiments 2 and 3, clusters were annotated by expression of previously defined CNS fibroblast or myeloid cell ‘signatures’ generated from experiment 1 DEG marker lists (visualized using AddModuleScore and violin plots). Dural fibroblasts were subclustered (FindSubCluster function) to identify lymphocyte-interactive and dural subsets (experiment 2). Inhibitory and excitatory neurons were identified by expression of Gad1/Gad2 and Slc17a6/Slc17a7, respectively. SAM were subclustered to identify subsets that changed across genotypes (experiment 3). Relative abundance across time was calculated for each fibroblast, myeloid, or T cell subcluster after normalizing for time point or condition sample size (total number of nuclei). Gene ontology analysis was performed using the ClusterProfiler R package for gene set testing (enrichGO function)103. Ligand–receptor and ligand–signalling network interactions underlying myofibroblast emergence were interrogated using the NicheNetR R package104 (experiment 1). Beginning with the global Seurat object, fibroblasts were treated as ‘receiver’ cells, with resting fibroblasts as the ‘reference condition’ and 7 dpi myofibroblasts as the ‘condition of interest’. Potential ‘sender’ cells were any cells present at rest, 2 or 7 dpi. Integration of single nuclear and Visium data was performed using the SpaceXR package105 (to deconvolute individual spots) and using Seurat’s AddModuleScore function (to score each Visium cluster with marker sets for single nuclear fibroblast clusters from experiment 1). Multi-gene scores were calculated as follows and generated using AddModuleScore. Myofibroblast (fibroblast TGFβ) scores were calculated as above. Profibrotic SAM scores were generated as published previously, using a combination of six genes (Trem2, Cd9, Spp1, Gpnmb, Fabp5 and Cd63)22. DAM scores were generated using the top 30 published markers specific to DAM (P < 0.001, ranked by logFC)31. Dysmaturity scores were generated using genes significantly upregulated in microglia from Emx1cre; Itgb8flox mice (relative to controls; bulk sequencing data accessed via the GEO and analysed with originally described parameters)41. IFNγ response scores for myeloid cells were generated using genes significantly upregulated (q < 0.05, log2FC > 1.5) in microglia that were sort-purified 22 h after intraventricular injection of IFNγ (100 ng) into juvenile (P9) mice54. IFNγ response scores for neurons were generated using genes significantly upregulated (q < 0.05, log2FC > 1.5) in neurons that were sort-purified 72 h after intraparenchymal injection of IFNγ (40 ng) into the ventral midbrain of adult mice106. Macrophage-fibroblast ligand–receptor interactions were interrogated and visualized using CellPhoneDB (version 4, statistical method; experiment 1)107. Pseudotime trajectories for myeloid cells were generated using the monocle3 R package v1.3.4 (learn_graph and order_cells functions), with homeostatic microglia and monocytes chosen as 2 possible roots (experiment 2). Phylogenetic trees were calculated using Seurat’s BuildClusterTree function.
Marmoset snRNA-seq data processing
Processed data from 7 dpi marmosets (ET-1-induced stroke)26 and resting marmosets (unpublished) were generously provided by the J. Bourne laboratory. Fibroblasts from both datasets were identified via COL1A1 expression, subset and integrated using the Seurat FindIntegrationAnchors and IntegrateData functions (dims = 1:30). The integrated data were subsequently rescaled, principal components were calculated, clusters were identified, and marker genes were selected as above. UMAP plots, bar plots, and heat maps of selected fibrosis-related genes were generated using Seurat. Myofibroblast (fibroblast TGFβ) scores were calculated as above.
Human TBI snRNA-seq data processing
Human TBI snRNA-seq data were accessed via the Gene Expression Omnibus (GEO)27. Datasets representing individual patients were merged, and quality control was performed on mitochondrial expression, UMI counts, and gene counts, as above. We used Seurat’s SCTransform function to normalize data, select variable features for dimensionality reduction (with the removal of certain sex-related and mitochondrial genes), and scale data. Principle components were calculated using Seurat’s RunPCA function, followed by batch correction using Harmony108, graph-based clustering, and cluster identification as above (res = 0.1). After cluster annotation, fibroblasts were identified via COL1A1 expression and subset. Heat maps and violin plots were generated using Seurat. Myofibroblast (fibroblast TGFβ) scores were calculated as above.
Human GBM scRNA-seq data processing
Human GBM data were generously provided by the M. Aghi lab12. QC was performed in accordance with the original publication (excluding cells with mt.percent > 20% and fewer than 200 or more than 20,000 UMIs). We used Seurat’s SCTransform function to normalize data, select variable features for dimensionality reduction (with the removal of certain sex-related and mitochondrial genes), and scale data. Principle components were calculated using Seurat’s RunPCA function, followed by graph-based clustering and cluster identification as above (res = 0.1). After cluster annotation, fibroblasts were identified via COL1A1 expression, subset, and reclustered as above (res = 0.6). UMAP plots, feature plots, and heat maps were generated using Seurat. Myofibroblast (fibroblast TGFβ) scores were calculated as above.
Other software
Python coding (using Python v3.11.0) was performed in Jupyter Notebook. R coding (using R version 4.3.2) was performed in RStudio. Additional R packages used include Presto, DESeq2, dplyr, ply, ape, cowplot, Matrix, variancePartition, MAST, HGNChelper, openxlsx, RColorBrewer, gridExtra, ggpubr, ComplexHeatmap, tidyverse, tibble, biomaRt, data.table, glmGamPoi, SeuratWrappers, patchwork, magrittr, s2, gplots, stringr, ggnewscale, ggbreak, coin and dunn.test.
Illustrations
Illustrations were created using BioRender as follows. Molofsky, A. (2025) (Fig. 1a–d and Extended Data Fig. 8s); (Fig. 1b); (Figs. 1c,d, 5a and Extended Data Fig. 1t,u); (Fig. 1e); (Fig. 2a); (Fig. 2r and Extended Data Fig. 5x); (Fig. 2h); (Fig. 3e); (Fig. 3r); (Fig. 3m–o); (Figs. 4a,d,j,o and 5i and Extended Data Figs. 1k,m,o,r,t,u, 5d,g and 7n); (Fig. 4d); (Fig. 4r and Extended Data Fig. 8ah); (Fig. 5i); (Fig. 5r and Extended Data Fig. 9a); (Fig. 5x); (Extended Data Figs. 1a,z and 4f); (Extended Data Fig. 1y); (Extended Data Fig. 5c); (Extended Data Fig. 6l); (Extended Data Fig. 7n); (Extended Data Fig. 8g); (Extended Data Fig. 9x); and (Supplementary Fig. 2).
Statistical analysis
All data were analysed by comparison of means using unpaired (unless otherwise noted) two-tailed Student’s t-tests; for multiple comparisons, one-way ANOVA (with Tukey post hoc test) or two-way ANOVA (with Sidak’s post hoc test, applied over within-subject and between-subject comparisons) were used as appropriate (Prism, GraphPad Software). Graphs display mean ± s.d.unless otherwise noted. When possible, results from independent experiments were pooled. All data points reflect individual biological mouse replicates, unless otherwise noted. Select experiments were performed once, for reasons including breeding, cost, and technical constraints, including the following: tMCAO in Cthrc1creER; R26tdT mice; analysis of spinal cords of Tgfbr2-cKO mice; analysis of cortical scar-associated macrophages in Tgfbr2-cKO mice; meningeal fibroblast coculture (using fibroblasts sorted from 18 mice); quantification of recombination in Ng2creER mice; and quantification of Col1a1GFP+ fibroblasts after ADWA11 treatment. RNA-sequencing experiments contained two pooled mice per sample. All other experiments were performed at least 2 times with successful reproduction.
Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
