Diet at disease onset determines tissue inflammatory memory

Mice

Mouse studies were conducted at the University of California, San Francisco (UCSF). Mice were housed in specific-pathogen-free facilities in temperature- and humidity-controlled animal facilities, with ambient room temperature maintained between 19 and 23 °C and relative humidity ranging from 40% to 60%. Mice were maintained under a 12 h–12 h light–dark cycle, with lights on from 07:00 to 19:00, and were group-housed with ad libitum access to food and water. When required, mice were purchased from The Jackson Laboratory. By default, mice fed at UCSF were given irradiated PicoLab Rodent Diet 20 5053. For experiments requiring the use of special diets, diets were irradiated and purchased from Research Diets. The HFD was 60 kcal% fat (D12492), and the control LFD was 10 kcal% fat (D12450J). The MFD was 45 kcal% fat (D08091803). The 30 kcal% fat diet was D20072301. The Western diet was D18053102, and the matched control diet was D18053101. Mice designated as obese in this study were fed the HFD for at least 10 weeks, starting at an age of 6–7 weeks, and weighed at least 38 g (as an average within a cage). For short-term HFD feeding, 6-week-old mice were given HFD for 7 days before the IMQ challenge unless otherwise indicated. Unless otherwise indicated, mice remained on their assigned diets throughout IMQ treatment. For HFDⁱⁿ feeding, lean mice were placed on HFD starting 3 days before IMQ challenge and remained on this diet until 1 day after initiation. For LFDⁱⁿ feeding, obese mice were placed on LFD starting 3 days before IMQ challenge and remained on LFD until 1 day after initiation. For HFD^out feeding, lean mice were maintained on LFD, except for two distinct periods when they were switched to HFD: first, from 7 to 3 days before IMQ initiation and, later, from 3 to 6 days after initiation during disease progression. For LFD^out feeding, obese mice were maintained on HFD, except for two distinct periods when they were switched to LFD: first from 7 to 3 days before IMQ initiation and, later, from 1 to 6 days after initiation during disease progression. For HFD^DODO feeding, mice were placed on HFD throughout the entire disease onset phase (spanning 7 days before to 6 days after IMQ initiation) and were then returned to LFD for the recovery and rechallenge phases. C57BL/6J (000664), B6.SJL-Ptprca Pepcb/BoyJ (002014)⁵⁶, Rag1^−/− (002216)⁵⁷, Tcra^−/− (002116)⁵⁸, B6.129S2-H2dlAb1-Ea/J (MHCII^Δ/Δ) (003584)⁵⁹, B2m^−/− (002087)⁶⁰, Tcrd^−/− (002120)⁶¹, Cd4^cre (022071)⁶² and Nlrp3^−/− (021302)⁶³ mice were purchased from The Jackson Laboratory. Il1r1^fl/fl (ref. ⁶⁴) and Cd4^creERT2Ai14 mice^65,66 were provided by the Scharschmidt Lab at UCSF. All of the mice used for the studies were male. All procedures involving animals were performed in accordance with protocols approved by the Institutional Animal Care and Use Committee (IACUC) of UCSF.

IMQ-induced acute type 17 skin inflammation model

IMQ-induced acute type 17 skin inflammation was generated in mice at various ages fed with either the LFD or HFD. In brief, mice were anaesthetised using isoflurane, and 15 mg cream containing 5% IMQ (TARO) was topically applied to the dorsal and ventral sides of the mouse ear daily for 6 days (ref. ²²). Ear thickness was measured daily with a digital micrometer (Mitutoyo). Then, 24 h after the last application, mice were euthanized, and ears were collected for flow cytometry, RNA extraction, histological analysis and other downstream applications. For rechallenge experiments, after the last day of IMQ treatment during the initial challenge, mice were allowed to recover. Ear thickness was measured weekly to evaluate recovery. Unless otherwise stated, after 25–30 days, mice were subjected to 15 mg cream containing 5% IMQ treatment for 4–6 consecutive days. To inhibit NLRP3 activation, MCC950 sodium (S7809, Selleckchem) was prepared in 50 mg ml⁻¹ in water and 75 μg was given topically to each mouse daily in sync with application of IMQ. For experiments involving dorsal back skin, mice were anaesthetised with isoflurane 1 day before IMQ treatment, and an approximately 2.5-cm² area was shaved and treated with Nair depilatory cream, which was promptly rinsed off with water. Under isoflurane anaesthesia, 50 mg of cream containing 5% IMQ (TARO) was applied topically to the prepared back skin daily for 6 consecutive days. For rechallenge experiments, shaving and depilation were performed only before the initial challenge and were not repeated before the second IMQ challenge.

Histology of lesional ear skin

Ear skin was collected and submerged in 4% paraformaldehyde (PFA) overnight. Tissues were then embedded, sectioned and stained with H&E by the Histology and Light Microscopy Core at Gladstone Institutes. Images were taken using a Zeiss microscope. The scale bar was added by converting pixels to μm using ImageJ.

Serum cytokine, hormone and IL-1β assays

Serum was collected and submitted to Eve Technologies for quantification using Luminex-based assays. For tissue IL-1β measurements, individual whole ears were collected and processed separately. Each ear was minced in 400 µl lysis buffer at 4 °C and homogenized using a Precellys Cryolys tissue homogenizer. Homogenates were clarified by centrifugation, and 100 µl of clarified lysate from each sample was submitted to Eve Technologies for Luminex-based cytokine quantification. IL-1β concentrations were calculated from standard curves and are reported as pg ml⁻¹ of clarified lysate generated from one whole ear homogenized in a fixed volume of lysis buffer.

Glucose tolerance test

Fasting was induced for 2 h. Glucose (2.0 g kg⁻¹, intraperitoneally) was injected and serum was collected for glucose measurements. Blood glucose was monitored using a Nova Max Plus glucometer.

Semaglutide injections

Mice were administered semaglutide (Selleck Chemicals, S9697) by daily intraperitoneal injection at a dose of 120 µg per kg. Treatment was initiated at the onset of exposure to either HFD or LFD and continued daily until the end of IMQ treatment. Semaglutide was dissolved in a vehicle consisting of PBS:kolliphor:DMSO at a ratio of 18:1:1. Control mice received daily intraperitoneal injections of vehicle alone. Injection volume did not exceed 150 µl per mouse.

Ear immune cell suspensions

Dissected mouse ear skin was separated into dorsal and ventral halves and incubated in 1 ml of digestion buffer. Mouse skins were incubated in 1 ml digestion buffer (RPMI-1640 medium with 2% FCS and 1× penicillin–streptomycin–glutamine solution (10378016, Roche) containing 0.2 mg ml⁻¹ liberase (05401020001, Roche), 0.2 mg ml⁻¹ DNase I (DN25-100MG, Sigma-Aldrich) and 0.5 mg ml⁻¹ hyaluronidase (LS002592, Worthington). After incubation for 30 min, the skin samples were minced into fine pieces (1–2 mm³) and incubated for another 30 min. The suspension was then passed through a 40 μm mesh to remove undigested clumps and debris. The flow-through was centrifuged at 500 rcf for 5 min. The pellet containing the stromal vascular fraction was washed once in 10 ml RPMI, and the resultant isolated cells were prepared for fluorescence-activated cell sorting (FACS) analysis directly or first stimulated with PMA and ionomycin in the presence of brefeldin A (Cell Activation Cocktail; BioLegend) for 1.5 h at 37 °C for subsequent intracellular cytokine staining and then prepared for FACS analysis.

In vivo antibody treatments

The following antibodies were used: anti-mouse CD4 (BioXCell, GK1.5), isotype control for anti-mouse CD4 (rat IgG2b, BioXCell, LTF-2), anti-mouse Ly6G (BioXCell, 1A8) and isotype control for anti-mouse Ly6G (rat IgG2a, BioXCell, 2A3). For the anti-CD4 experiments: antibodies were injected intraperitoneally 2 days before, on the day of, and 2 days after the initiation of the IMQ challenge at a dose of 0.1 mg per mouse. For the rechallenge experiment, anti-CD4 or isotype control antibodies were injected intraperitoneally on days 29, 32 and 35 after the initial IMQ challenge using the same dosage. For the anti-Ly6G experiments: antibodies were injected intraperitoneally 3 days before, 1 day before, 1 day after, 3 days after, and 5 days after the initiation of the IMQ challenge at a dose of 0.1 mg per mouse. Injection volumes never exceeded 150 µl. For the LPS-induced lung inflammation model, antibodies were injected intraperitoneally 3 days before, 1 day before, 1 day after and 3 days after the LPS challenge at a dose of 0.1 mg per mouse. Injection volumes never exceeded 150 µl.

Adoptive transfer of CD4⁺ T cells

CD4⁺ T cells were isolated from the spleens of WT mice using the Mouse CD4⁺ T Cell Isolation Kit (StemCell) and 100,000 cells were transferred intravenously into Rag1^−/− mice. After an 8-week reconstitution period, mice were fed LFD or HFD for 1 week. Mice were then subjected to IMQ treatment for 6 days to induce acute type 17 skin inflammation. Disease severity was assessed by ear thickness measurements, and immune responses were analysed by flow cytometry.

Flow cytometry

Fc Block (2.4G2, 553141, BD) was diluted in FACS buffer and added to each sample 5 min before performing surface staining. eBioscience Fixable Viability Dye eFluor780 was used to distinguish live from dying or dead cells. Cell surfaces were stained for 30 min at 4 °C. For intracellular staining, cells were treated with fixation and permeabilization reagents from eBioscience and labelled with the appropriate antibodies before being analysed. Data were analysed using Aurora Cytek (v.4.0.1) and FlowJo software (v.10 and v.11, FlowJo). Cells were sorted on the Cytek Aurora CS System. The following antibodies were used: CD45.2 (104, 109839, BioLegend), CD197 (4B12, 120120, BioLegend), CD11c (N418, 117322, BioLegend), CD11b (M1/70, 101224, BioLegend), CD44 (IM7, 103037, BioLegend), TCRγδ (GL3, 563993, BD), CD62L (MEL-14, 20-0621-U025, TONBO), CD69 (H1.2F3, 104506, BioLegend), Ly6G (1A8, 127672, BioLegend), CD3ε (145-2C11, 100349, BioLegend), CD4 (RM4-5, 100528, BioLegend), CD8α (53-6.7, 100744, BioLegend), Ki-67 (SolA15, 367-5698-82, Invitrogen), RORγt (Q31-378, 562684, BD), FOXP3 (FJK-16s, 46-5773-82, Invitrogen), IL-17A (TC11-18H10.1, 506914, BioLegend), IL-17A (TC11-18H10.1, 506925, BioLegend), IL-17F (9D3.1C8, 517008, BioLegend), CD121a (35F5, 564387, BD), CD121a (JAMA-147, 113505, BioLegend), IFNγ (XMG1.2, 505806, BioLegend), IL-22 (Poly5164, 516409, BioLegend). Antibodies against Fc Block, CD45.2, CD197, CD11c, CD11b, CD44, TCRγδ, CD62L, CD69, Ly6G, CD3ε, CD4 and CD8α were used at a 1:200 dilution; antibodies against Ki-67, RORγt, FOXP3, IL-17A, IL-17F, CD121a (clones 35F5 and JAMA-147), IFNγ and IL-22 were used at a 1:400 dilution. eBioscience Fixable Viability Dye eFluor780 (65-0865-14) was used at a 1:1,000 dilution.

Lesional CD4⁺ T cell scRNA-seq

CD4⁺ T cells from mouse ear skin were FACS-sorted, and mRNA isolation, cDNA synthesis and library preparation were then performed based on the protocol provided by PIPseq T2 3′ Single Cell RNA Kit v4.0PLUS (Fluent BioSciences). In brief, cells were sorted by FACS (sorting on live CD45.2⁺CD3ε⁺CD4⁺ cells on a Cytek Aurora CS System) in cell suspension buffer (provided in the kit) with 1 U μl⁻¹ RNase inhibitor. Then, 4 μl cell suspension (5,000 cells maximum) with 20 U of RNase inhibitor was added to each PIP tube reaction, followed by vortexing to generate single bead to cell emulsion. We performed 16 cycles of PCR for cDNA amplification after cDNA synthesis, and 25% of each cDNA sample was carried into transcriptome library preparation. PIPseq T2 3′ Single Cell gene expression libraries were subject to standard Illumina paired-end sequencing targeting 20,000 reads per input cell using the NovaSeq X system.

Raw fastq files were mapped to the mouse transcriptome (UCSC mm10) using PIPseeker⁶⁷ (Fluent BioSciences, v.3.2) pipeline that performs barcode whitelisting, alignment with STAR, and molecular index correction and generates barcode-feature counts matrices. After alignment and sensitivity calling (sensitivity_5 was chosen for all three samples), 2,103 cells were recovered from the LFD condition and 5,232 cells were recovered from the HFD condition—two T2 PIPseq reactions were used, labelled HFD1 and HFD2. Subsequent analysis was performed using Scanpy (v.1.10.4)⁶⁸. The three datasets were concatenated into one AnnData object. Low-quality cells (number of genes < 200) and genes (expressed in fewer than three cells) were removed. Doublets were identified and removed using the scrublet function. Data were normalized to 10,000 reads per cell and logarithmized before clustering. Clusters were calculated using the Leiden algorithm and visualized using UMAP. Three clusters (clusters 8, 11 and 12) were excluded due to contamination (monocytes and stromal cells) or high mitochondrial content. Cell types were annotated based on the expression level of known genes and differential gene expression analysis, using the rank_genes_group function. After quality-control filtering (for doublets and non-CD4⁺ T cell contamination), 1,769 CD4⁺ T cells remained in the LFD condition, and 4,450 CD4⁺ T cells remained in the HFD condition, with over 1,500 median genes per cell. To assess T_H17 pathogenicity, we calculated a T_H17 pathogenicity score based on published gene signatures associated with pathogenic T_H17 cells^9,31. Gene sets from both studies, comprising key regulators and effector genes that define pathogenic T_H17 cells, were curated and used to evaluate the pathogenic potential of T_H17 cells in our dataset. Using the score_genes function in Scanpy, we calculated the T_H17 pathogenicity score for each cell in the T_H17 cluster. To compare differentially expressed genes in T_H17 cells in HFD compared to LFD samples, differential gene expression analysis was performed using Scanpy’s rank_gene_groups function, with tie correction for Wilcoxon scores. To remove background noise of genes expressed by few cells, genes expressed in fewer than 100 cells were excluded. Genes with log₂[fold change] > 1 and P < 0.0001 were displayed. Plotting was performed using matplotlib (v.3.9.3) and seaborn (v.0.13.2).

Whole-ear bulk RNA-seq

Total RNA was extracted from whole mouse ear skin before, at the end of or 26 days after IMQ challenge using the RNeasy Fibrous Tissue Kit according to the manufacturer’s protocol (74704, Qiagen). RNA quantity and quality were determined by NanoDrop spectrophotometers and Qubit 4 fluorometer. In total, 10 μg RNA from each sample was sent to Genewiz for library preparation and sequencing services (RNA-seq). Read alignment and junction finding were accomplished using STAR⁶⁹ using UCSC mm10 as the reference sequence. Differential gene analysis and data visualization were performed in R using the DESeq2⁷⁰ package.

Lesional immune-cell scRNA-seq

Single-cell gene expression libraries were generated using the Chromium Single Cell Fixed RNA Profiling (Flex) platform (10x Genomics) according to the manufacturer’s instructions. Fixed single-cell suspensions were encapsulated using the Chromium Controller to generate barcoded gel beads in emulsion, followed by reverse transcription and library construction per the 10x Genomics Flex protocol. Libraries were sequenced on an Illumina NovaSeq X platform to obtain paired-end reads. Resulting raw fastq files were mapped to the reference genome (GRCm39) using the Cell Ranger’s multi pipeline⁷¹. Resulting matrix, barcodes and features files were analysed using Scanpy (v.1.10.4). A total of 27,738 cells was analysed (14,990 HFD and 12,748 LFD). The two datasets were concatenated into one AnnData object. Low-quality cells (number of genes < 200) and genes (expressed in fewer than three cells) were removed. Doublets were identified and removed using the scrublet function. Data were normalized to 10,000 reads per cell and logarithmized before clustering. Clusters were calculated using the Leiden algorithm (resolution = 0.5) and visualized using UMAP. Differentially expressed genes were determined using the rank_genes_groups function. Using the PanglaoDB database⁷² and decoupler’s OmniPath module⁷³, cell type scoring was performed to help assign cell type labels. Using this, together with differentially expressed genes and expression levels of known markers, 17 different cell types were identified. Differentially expressed genes between conditions were calculated using rank_genes_groups. Plotting was performed using matplotlib (v.3.9.3) and seaborn (v.0.13.2).

Recovered-skin CD4⁺ snATAC–seq

snATAC libraries were prepared using the Chromium Single Cell ATAC v2 kit (10x Genomics) according to the manufacturer’s protocol. Nuclei were isolated, transposed using Tn5 transposase and barcoded by droplet-based partitioning on the Chromium Controller, followed by library amplification and construction. Libraries were sequenced on an Illumina NovaSeq X platform to generate paired-end reads for downstream alignment and peak calling. Resulting raw fastq files were mapped to the reference genome (GRCm39) using Cell Ranger’s atac count pipeline⁷¹. Subsequent analysis was performed using ArchR⁷⁴ (v.1.0.3). In the first step, arrow files were created from the fragment files using the barcodes that passed the cellranger pipeline. Analysis was performed using the mouse GRCm39 genome. After initial quality control, where fragment size distribution, TSS enrichment and number of fragments per cell were visualized, peaks were called using the addGroupCoverages function and addReproduciblePeakSet using MACS2⁷⁵. Initial clustering was performed using the addIterativeLSI and addClusters functions, and visualized using UMAP. Top markers for each cluster were calculated using getMarkerFeatures (using the GeneScoreMatrix) and significant markers were defined as markers with a false-discovery rate of <0.05 and Log2FC ≥ 1 or ≤ −1. To add imputation weights and smooth the signal across nearby cells for improved visual interpretation, the addImputeWeights function was used and data were visualized using the plotEmbedding function. Differentially accessible regions were calculated using the getMarkerFeatures using the PeakMatrix. Data were visualized using base R plotting functions or in-built ArchR functions, such as plotMarkers or plotBrowserTrack.

LPS-induced lung inflammation model

Acute lung injury was induced by intranasal instillation of lipopolysaccharide (LPS; Escherichia coli O111:B4, Sigma-Aldrich, L3012). Mice were deeply anaesthetised with inhaled isoflurane to achieve a reduced respiratory rate with augmented tidal breathing. LPS (60 μg in 50 μl sterile PBS) was delivered dropwise to the nares while the mice were held in an upright position. During deep inspiratory efforts, animals were gently rotated to alternate the orientation of the main bronchi and promote more uniform distribution of the inoculum.

Lung immune cell suspensions

Mice were euthanized 4 days after LPS administration. Bronchoalveolar lavage fluid and lung tissues were collected. Lung tissues were weighed before further processing. Dissected mouse lung was collected and finely minced in 1 ml of digestion buffer. Mouse lungs were incubated in 1 ml digestion buffer (RPMI-1640 medium with 2% FCS and 1× penicillin–streptomycin–glutamine solution (10378016, Roche) containing 0.2 mg ml⁻¹ liberase (05401020001, Roche), 0.2 mg ml⁻¹ DNase I (DN25-100MG, Sigma-Aldrich) and 0.5 mg ml⁻¹ hyaluronidase (LS002592, Worthington). After 1 h incubation, the suspension was passed through a 40 µm mesh to remove undigested clumps and debris. The flow-through was centrifuged at 500 rcf for 5 min. The pellet was washed once in 5 ml RPMI, and the resultant isolated cells were stimulated with PMA and ionomycin in the presence of brefeldin A (Cell Activation Cocktail, BioLegend) for 1.5 h at 37 °C for subsequent cytokine staining and then prepared for FACS analysis.

Human psoriasis scRNA-seq analysis

Publicly available scRNA-seq datasets from psoriasis vulgaris lesions were obtained from EGA (EGAS00001007373 and EGAS00001005271; n = 15 samples) as reported previously^76,77. Count matrices were loaded into R using Read10X (Seurat) and processed with Seurat⁷⁸ to generate one Seurat object per sample. Putative doublets were identified and removed using scDblFinder⁷⁹. Cells were filtered before downstream analysis to retain those with 100–6,000 detected genes, <20% mitochondrial transcripts and <50% ribosomal transcripts. Filtered datasets were merged and normalized in Seurat using standard workflows. Cell type labels were assigned by reference mapping to an annotated psoriasis immune atlas⁷⁶ using FindTransferAnchors, TransferData, AddMetaData and MapQuery. Marker genes were identified using FindAllMarkers, and visualization was performed using FeaturePlot and VlnPlot.

FTY720 treatments

Mice received daily intraperitoneal injections of 100 µl per mouse of FTY720 (1 mg per kg) or water (vehicle), starting 1 day before the IMQ challenge and continuing daily through its duration. This treatment was applied during either the initial disease onset or the rechallenge phase.

Thick-section 3D confocal microscopy

Tissue processing for 3D confocal imaging: after CO₂ euthanasia, the dorsal back skin was excised and floated dermis side down in 2% PFA (Thermo Fisher Scientific) for 1 h at 4 °C, then fully submerged in fresh 2% PFA for an additional 6 h at 4 °C. Tissue was subsequently cryoprotected in 30% sucrose overnight at 4 °C. One-centimetre-squared sections were embedded in OCT compound (Thermo Fisher Scientific), snap-frozen on dry ice and stored at −80 °C. Using a cryostat, 200-µm thick sections were cut, washed in PBS and blocked for 24 h at 37 °C in DPBS containing 0.3% Triton X-100, 1% host serum (matched to the secondary antibody) and 1% BSA. The sections were incubated with primary antibodies diluted in blocking buffer for 48–72 h at 37 °C, washed three times for 2 h each in PBS and incubated with appropriate secondary antibodies for 24 h at 37 °C. The sections were then washed three times for 8 h total and cleared using Ce3D clearing solution (40% n-methylacetamide) before imaging.

Imaging antibodies

Primary antibodies used for mouse imaging included rabbit anti-dsRed (Takara, 632496, 1:400), rat anti-CD49f–AF488 (BioLegend, 313608, 1:200), rat anti-α-smooth muscle actin–CF405S (Biotium, BNC040665, 1:100) and goat anti-FABP5 (R&D Systems, AF1476-SP, 1:400). Secondary antibodies conjugated to Alexa Fluor 555 or Alexa Fluor 594 (Thermo Fisher Scientific) were used as appropriate.

Confocal microscopy

Cleared thick sections were imaged using the Leica SP8 upright laser-scanning confocal microscope equipped with a 405 nm laser and tuneable white light laser, using a 20× water-immersion objective. z-Stacks were acquired under identical acquisition settings within experiments to allow qualitative comparison across conditions.

Full-thickness skin transplantation

Donor mice were anaesthetised with isoflurane. Anaesthesia depth was confirmed by toe pinch. The left dorsal back skin of the donor was shaved with an electric razor and disinfected with povidone-iodine and 70% alcohol. Using sterile scissors and aseptic technique, full-thickness left dorsal back skin was collected using blunt dissection at the level of the areolar connective tissue. After graft collection, the donor was euthanized by cervical dislocation. Under a dissecting microscope, connective tissue, fat and the panniculus carnosus were removed from the graft using fine scissors. Individual grafts measuring around 15 × 15 mm were cut and stored on sterile PBS-soaked gauze in a Petri dish on ice. A single donor yielded two grafts for two recipient mice.

Recipient mice were anaesthetised with isoflurane (induction at 3%, maintenance at 2%) and given bupivacaine (local) and Ethiqa XR (subcutaneous) for analgesia. The upper back (thoracic) region was shaved and disinfected with 10% povidone-iodine. A graft bed of matching size (15 mm square) was created by carefully excising skin while preserving the panniculus carnosus and its superficial vasculature. The donor graft was placed on the bed, aligned without folds and secured with eight interrupted sutures at the corners and midpoints of each edge. Once suturing was complete, the thorax was wrapped with an adhesive bandage. The bandage was adjusted to allow free breathing and thoracic excursion.

The recipient mice were returned to the housing facility after they could move freely. On postoperative day 7, the bandages were removed under brief anaesthesia by cutting only the ventral side of the dressing to avoid disturbing the graft. The grafts were inspected for signs of vascularization; contraction, scabbing or firmness indicated technical failure. For analgesia, Ethiqa XR was administered pre-operatively for sustained pain control. Additional analgesics were provided as required under veterinary guidance.

Statistics and reproducibility

Statistical analyses were performed with Prism (GraphPad, v.10 and v.11). P values were calculated using Welch’s t-tests unless otherwise noted. Statistical tests were two-sided. When applicable, mice were randomly assigned to treatment or control groups. Experimental data were not excluded from the statistical analyses, except for exclusions due to technical errors or loss of confidence in appropriate controls. Measurements were taken from distinct biological samples, except for ear thickness, which was measured longitudinally in the same mice at multiple time points. The investigators were not blinded in the studies unless otherwise stated. No estimate of variance was made between each group. All in vivo experiments were conducted with at least two independent cohorts unless otherwise specified. Unless otherwise noted, data from one instance of the repeated experiments are represented in this Article. The recurrence experiment in which mice were allowed to recover for 6.5 months before rechallenge with IMQ was only conducted once. The targeted lipidomics analyses were conducted once. Bulk RNA-seq, scRNA-seq and snATAC–seq experiments were conducted once using multiple biological samples per group (as indicated in the figure legends). Histological analyses were performed on tissues from a single experimental cohort for each comparison. Unless otherwise indicated, at least five mice per group were examined; for each mouse, three serial sections were assessed at each of at least two separately cut tissue levels. Thick-section confocal imaging was performed once using two mice per group, with three 200 µm sections imaged per mouse.

Reporting summary

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