Mice
We used an outbred mouse strain (CD1) that is less susceptible to generating papilloma or squamous cell carcinomas than more inbred mouse strains used in other studies11,12. Mice were generated by interbreeding mice carrying the following alleles: Krt14-CreER51 and FR-HrasG12V/+ (ref. 52), constitutive p21 (Cdkn1a) loss of function38 (JAX stock no. 016565), LSL-tdTomato53 (JAX stock no. 007909), Krt14-H2B–GFP23, Krt14-rtTA54 (JAX stock no. 008099), TRE-EGFR-DN55 (JAX stock no. 010575) and LSL-KrasG12D/+ (ref. 56). Mice expressing Krt14-H2B–GFP were used to track epithelial cells with the two-photon microscope. The tdTomato reporter line was used to visualize CreER-driven recombination upon tamoxifen injection. CreER/LoxP lines are well documented to exhibit a certain degree of tamoxifen-independent (leaky) Cre activity over time. To account for the leakiness of the system in our experiments and to correctly interpret the data, we always compared wild-type mosaic models to Ras-mosaic models as well as tracked the same cells in the same animals over time. Mice from experimental and control groups were randomly selected for either sex. No blinding was done. All procedures involving animal subjects were performed under the approval of the Institutional Animal Care and Use Committee (IACUC) of the Yale School of Medicine. The mice were sacrificed if tumours reached 1 cm3 (not allowed by IACUC) or if they presented signs of distress or weight loss. The tumour size limit was not exceeded in any of the experiments.
Tamoxifen induction and drug treatment of mice
To induce CreER-driven recombination, mice were administered a single dose of 100 μg (mosaic induction) or 2 mg (maximal induction) tamoxifen (Sigma T5648-5G in corn oil) at postnatal day 19 by intraperitoneal injection (this time is designated day 0 for experiments). To induce rtTA-driven induction of EGFR-DN, mice were administered 2% of doxycycline (Sigma D9891) and 2% sucrose (Sigma S9378) in drinking water. All time courses began 6 days after tamoxifen injection. Gefitinib (ZD1839-Selleckchem) was resuspended in water with 0.5% (w/v) methylcellulose and 0.2% (v/v) Tween-80 (vehicle) and was administered orally (200 mg kg−1 body weight) starting 2 days before wound induction until 14 days PWI (mice were not treated at day 7 and day 8 PWI).
Injury induction
At postnatal day 21, mice were anaesthetized by intraperitoneal injection of a ketamine and xylazine mix (100 mg kg−1 and 10 mg kg−1, respectively in phosphate-buffered saline). Once the anaesthetized mouse did not physically respond to a noxious stimulus, a punch biopsy was performed using a 4-mm-diameter punch biopsy tool (Integra Miltex Standard Biopsy Punches). The punch biopsy tool was used to make a circular full-thickness injury on the dorsal side of a mouse ear or back skin. The injury did not penetrate the entire ear but remained above the cartilage. The skin epithelium in the mouse ear was chosen in this study for its accessibility to two-photon imaging and revisits over time. For recovery from the wound procedure Meloxicam (Metacam Loxicom) was administered via subcutaneous injection (0.3 mg kg−1).
Lentiviral production and in utero injection
Large-scale production and concentration of lentivirus expressing CreER (LV-CreER) was performed as previously described57,58. Detailed descriptions of in utero-guided lentiviral transduction can be found elsewhere57,59. To induce LV-CreER-mediated recombination, a maximal dose of tamoxifen (2 mg) was intraperitoneal injected at postnatal day 19.
In vivo imaging
Mice were anaesthetized by intraperitoneal injection of ketamine and xylazine cocktail mix (100 mg kg−1 and 10 mg kg−1, respectively in phosphate-buffered saline) and then anaesthesia was maintained throughout the course of the experiment with the delivery of vaporized isoflurane by a nose cone as previously described60. Image stacks were acquired with a LaVision TriM Scope II (LaVision Biotec) laser scanning microscope equipped with a tunable Two-photon Chameleon Vision II (Coherent) Ti:Sapphire laser. To acquire serial optical sections, a laser beam 940 nm was focused through a 20× water immersion lens (NA 1.0; Zeiss) or a 25× water immersion lens (NA 1.1; Nikon) and scanned with a pixel size of 0.49 × 0.49 μm2 or 0.43 × 0.43 μm2 at 800 Hz. Z-stacks were acquired in 2–3 μm steps to image a total depth of 90–900 μm of the tissue. ImSpector v7.5.2 (LaVision Biotec) was used for 3D image acquisition. To visualize large areas, 5–12 tiles of optical fields were imaged using a motorized stage to automatically acquire sequential fields of view as previously described60. Mice were imaged at different time points after tamoxifen treatment and injury induction as indicated. To revisit the same area of the skin epidermis, organizational clusters of hair follicles and vasculature were used as landmarks.
Whole-mount, OCT section and epidermal preparation immunostaining and imaging
To prepare whole mounts of mouse ear, skin was separated from connective tissue using forceps and incubated in 4% paraformaldehyde (PFA) at 37 °C for 4 h. To prepare the epidermal preparation, the skin separated from the connective tissue was incubated with Dispase (5 mg ml−1, Roche) for 10 min at 37 °C and then the intact sheet of epidermis was gently peeled away from the dermis. The epidermal preparations were fixed in 4% PFA for 45 min at room temperature. Immunostaining was performed on whole mounts and epidermal preparations blocked with 5% normal goat serum/1% BSA/2% Triton X-100/PBS at room temperature. For tissue-section analysis, mouse ears were fixed in 4% PFA for 1 h at room temperature and then embedded in optimal cutting temperature (OCT; Tissue Tek). Frozen OCT blocks were sectioned at 10 μm. The skin preparations were incubated with primary antibodies (active caspase-3 (AF835-R&D Systems) 1:300, p-histone H3 (06-570-Millipore) 1:300, keratin-6A (905701-BioLegend) 1:500, p21 (also known as Cdkn1a) (ab188224-Abcam) 1:50; p-p44/42 MAPK(Erk1/2) (4370-Cell Signaling) 1:300, Alexa Fluor 647 Phalloidin (A22287-Thermofisher) 1:200 and keratin-10 (03-GP-K10-ARP) 1:200 diluted in blocking buffer (∼66 h at 37 °C for whole mounts and 4 °C overnight for epidermal preparations and OCT sections). Secondary antibodies (A-21071-Invitrogen goat anti-rabbit IgG (H + L) secondary antibody, Alexa Fluor 633, A-21105-Invitrogen goat anti-guinea pig IgG (H + L) highly cross-adsorbed secondary antibody, Alexa Fluor 633, A-21206, donkey anti-rabbit IgG (H + L) highly cross-adsorbed, Alexa Fluor 488 and A-10042 donkey anti-rabbit IgG (H+L) highly cross-adsorbed secondary antibody, Alexa Fluor 568) were diluted 1:300 in blocking buffer and applied to skin sections (∼66 h at 37 °C for whole mounts or 1 h at room temperature for epidermal preparations and OCT sections). DAPI was added to label nuclei. Image stacks of whole-mount, epidermal preparation and OCT section immunostaining were acquired with the two-photon microscope described above with an additional tunable two-photon Chameleon Discovery (Coherent) Ti:Sapphire laser. To acquire serial optical sections, laser wavelengths of 800 nm, 880 nm (Vision II) and 1080 nm (Discovery) were focused through a 20× water immersion lens (NA 1.0; Zeiss) or a 25× water immersion lens (NA 1.1; Nikon) and scanned with a pixel size of 0.49 × 0.49 μm2 or 0.43 × 0.43 μm2 at 800 Hz. Image stacks of epidermal preparation immunolabelled for p-ERK1/2 and phalloidin were acquired with confocal microscope Zeiss LSM 980, objective Zeiss 20× (0.8 NA Dry) and scanned with a pixel size of 0.20 × 0.20 μm2 with Software ZEN (blue edition). Z-stacks were acquired in 1-μm (whole mounts) or 2-μm (epidermal preparations and OCT sections) steps to image total depth of the samples.
Senescence-associated β-galactosidase activity
Senescence was measured in epidermal preparation of uninjured and injured (day 3 PWI) mouse ear skin and in mouse pancreas and kidney, as positive controls, with a Beta Galactosidase Assay Kit from Abcam (ab287846). This kit uses the fluorogenic fluorescein digalactoside galactosidase substrate, which, upon cleavage by β-galactosidase, generates a fluorescent product that can be measured with an ELISA plate reader. In brief, ~10 mg of the tissues described above was lysed with protein lysis buffer (included in the kit) and then the kit guidelines were followed to prepare samples, positive controls provided by the kit and the standard curve. The fluorescence in each sample was quantified with a GloMax Plate reader (Promega) at 475 nm excitation and 500–550 nm emission at two different time points (30 min apart). β-Gal levels in each sample were calculated using a β-galactosidase standard curve.
Two-photon image analysis
Raw two-photon image stacks were analysed in ImageJ (NIH Image, 1.53c) or IMARIS (v. 9.9.1, Oxford Instruments). ImageJ was used to draw the boundaries between tdTomato+ and tdTomato− areas in the basal stem cell layers of the skin epidermis and to measure the percentage of coverage of these areas. The average of the percentage of coverage of tdTomato+ cells of three areas (294 μm2 each) taken at different distances from the wound and randomly in the uninjured condition were calculated for each mouse at every time point. Then, the percentage increase of tdTomato+ area over time was represented in the graphs. To be reproducible in the measurements, the areas quantified in the injured ears were not localized close to the wound edge because of the increased thickness of the epithelium in that region that prevents an accurate isolation of the basal stem cell layer due to reduced image resolution. tdTomato+ and tdTomato− regions for each uninjured or injured ear were revisited and quantified over time.
We measured the number of events (mitotic figures, nuclear fragmentation events, or cells positive for p-histone H3, keratin-10 and active caspase-3 immunostaining) per unit of surface (1 μm2) of tdTomato+ or tdTomato− areas and then multiplied that value by the total surface (294 μm2) to compare wild-type and mutant cell populations. ~900 μm2 were analysed for each uninjured or injured mouse ear. We measured the number of cells positive for p-ERK1/2 per unit of surface (1 μm2) of tdTomato+ or tdTomato− areas and then multiplied that value by the total surface (150 μm2) to compare wild-type and mutant cell populations. An area of approximately 600 μm2 was analysed for each uninjured or injured mouse ear.
To quantify the thickness of the skin epithelium, we used IMARIS (v. 9.9.1). By utilizing the second harmonic collagen signal in the dermis and, when absent near the wound, the basal epithelial cell layer, we created a surface to approximate the basement membrane of the epithelium. When unable to visualize the basal–dermal interface near the wound due to excessive epithelia thickness, such as in the case of HrasG12V/+ max or KrasG12D/+ max, we created a surface along the bottom of the stack to ensure our thickness measurements were stringent in that and reflected only as much tissue as we could effectively image. From this surface we could extract the distance to the top of the cornified layer around the injured area of the epithelium and visually depict the tissue thickness with an intensity heat map. Using MatLab (v. R2018a) we could extract individual pixel intensity values from this heat map that directly correlate to tissue thickness and plot them based on relative distance from the wound edge.
To quantify the nuclear signal of p21 in the basal stem cell layer of the skin epidermis, we used IMARIS (v. 9.9.1). In IMARIS, surfaces were created using the Krt14-H2B–GFP signal to isolate a mask of the p21 signal within the basal epidermal stem cell nuclei. A maximum intensity projection of this mask was used to quantify the mean fluorescence intensity of the p21 signal within each individual nucleus. The mean fluorescence intensity of the p21 signal of each cell was normalized for background by subtracting the average p21 fluorescent intensity of mitotic cells within the field of view, as these cells would be negative for p21.
scRNA-seq sample preparation and data analysis
After the sacrifice of wild-type, HrasG12V/+ mosaic and HrasG12V/+ max models at 6 days after tamoxifen injection (3 days PWI), the uninjured and injured ears of each mouse were cut in small pieces with a punch biopsy of 8 mm in diameter (the wound was kept at the centre the 8 mm biopsy to mostly isolate cells involved in injury repair). The ear epidermis was dissociated from the dermis and incubated in 0.25 % Trypsin at 37 °C for 30 min. The epidermal preparation was placed in a 70 μm cell strainer, smashed with a piston and rinsed three times with PBS + 0.04% BSA. The flow-through was subsequently filtered through a 40 μm cell strainer, spun down and resuspended in 300–400 μl of PBS + 0.04% BSA. The viability of the cell suspension was determined using trypan blue. To prepare the single-cell library, the cellular suspensions were counted and diluted to a final concentration of 1,200 cells per μl in PBS/0.04% BSA and then loaded on a Chromium Controller to generate single-cell gel bead emulsions, targeting 3′. Single-cell 3′ RNA-seq libraries were generated according to the manufacturer’s instructions (Chromium Single Cell 3′ Reagent v3 Chemistry Kit, 10X Genomics). Libraries were sequenced to an average depth of ∼20,000 reads per cell on an Illumina Novaseq 6000 system.
Single-cell data from each sample—that is, all wild-type and HrasG12V/+ mosaic and HrasG12V/+ max, uninjured and injured conditions (24 independent samples, Extended Data Figs. 4a, 6a and 7a) were first processed with SoupX61 (https://github.com/constantAmateur/SoupX) to remove barcodes that most probably represent ambient RNA as opposed to whole cells, using the algorithm’s automated method. The resulting matrix was then processed with the Seurat package62 (v.3, https://satijalab.org/seurat/index.html), to retain genes or features that are detected in at least 3 cells and include cells for which at least 200 genes or features are detected. Additionally, cells expressing greater than 12.5% of mitochondrial transcripts were filtered out as possible dead or dying cells. According to Seurat’s normal workflow, the data was log-normalized and scaled. Linear dimensionality reduction was carried out using principal component analysis and the first 15 principal components were chosen for the downstream analysis steps. Clustering was carried out using Louvain algorithm, for resolution of 0.1. Non-linear dimensionality reduction was carried out by running UMAP. Next, the DoubletFinder63 package (https://github.com/chris-mcginnis-ucsf/DoubletFinder) was used to get rid of barcodes that may represent possible doublets. The resulting cell matrix was normalized, scaled, and re-clustered, using the same steps mentioned above with the Seurat package. Three replicates for each group were then integrated using Seurat’s canonical correlation analysis (CCA). The data were re-clustered as described previously22, and the clusters were annotated using the top 5–10 highly expressed genes in each cluster. To further remove doublets that were not identified by the DoubletFinder algorithm and other contaminating cell populations, infundibulum cells, immune cells, and red blood cells (RBCs) were removed. In brief, using a chosen set of features, each cell in the Seurat object was assigned a score using the AddModuleScore function. The features used for each type of cells are listed below. Infundibulum-specific features: “Sostdc1”, “Aqp3”, “Ptn”, “Fst”, “Aldh3a1”, “Postn”, “Krt17”,“Alcam”, “Apoe”, “Sox9”, “Vdr”, “Nfib”, “App”, “Gsn”, “Hmcn1”, “Cspg4”, “Efnb2”, “Nedd4”, “Adh7”, “Defb6”, “Mgst1”, “Krt79”. Immune cell-specific features: “H2-Aa”, “H2-Ab1”, “H2-Eb1”, “Cd74”, “Ptprc”. RBC-specific features: “Hba-a1”, “Hbb-bs”, “Hba-a2”, “Hbb-bt”, “Bpgm”, “Hebp2”. For infundibulum signature, cells with scores higher than 0.4 were removed, while for immune and RBCs, cells with scores higher than 0.5 were removed. The following cell numbers (per sample and biological replicate (R)) passed QC and constitute the final dataset: HMU: 11,437 (R1), 9,864 (R2), 9,650 (R3); HMW: 9,590 (R1), 10,281 (R2), 8,922 (R3); WTMU: 11,856 (R1), 9,916 (R2), 8,416 (R3); WTMW: 9,181 (R1), 8,747 (R2), 8,320 (R3); HFU: 13,278 (R1), 10,711 (R2), 6,997 (R3); HFW: 9,613 (R1), 8,999 (R2), 7,825 (R3); WTFU: 13,379 (R1), 11,701 (R2), 6,433 (R3); WTFW: 9,649 (R1), 9,374 (R2), 6,350 (R3).
For interfollicular epidermis (IFE) keratinocytes, all mosaic samples were integrated and annotated using Scanpy (1.6-1.9)64. In brief, raw counts for the selected IFE cells were log-normalized and cell cycle stages were scored (sc.pp.score_genes_cell_cycle) based on a gene list from65. Biological replicate batches were corrected with bbknn (1.4.1)66. Next, the selected cells were scored for stress, immune and infundibulum related gene expression signatures (see the notebooks on GitHub: https://github.com/kasperlab/Gallini_et_al_2023_Nature), classified with a Gaussian mixture model (scikit-learn, 0.24.267) and positive cells were filtered out. Similarly, classification was performed to annotate cells based on tdTomato and GFP expression. The remaining healthy IFE keratinocytes were then mapped (sc.tl.ingest) onto the characterized IFE differentiation trajectory and annotated accordingly based on the basal-suprabasal status and commitment that have been previously defined22. Finally, appropriate IFE groups were integrated with CCA as described above, using the Seurat package. Differential gene expression analysis was carried out between corresponding datasets and cell types.
For fibroblast and immune cell characterization, all mosaic and max datasets were analysed in Scanpy, with similar preprocessing and batch effect removal as for keratinocytes. Mixed cell populations were removed based on shared gene expression signatures. Wound-related cells were annotated based on Leiden clustering, sample type and wound-related gene expression signatures with additional confirmation by differential abundance testing using miloR (1.2.0)68. In brief, a Milo graph was built using the integrated dataset with the following parameters: k = 20, d = 30 and differential abundance was tested for the injury condition. Differential gene expression analysis was performed with scanpy.tl.rank_genes_groups function using Wilcoxon rank-sum test and Holm–Sidak correction for multiple comparisons. Gene set enrichment analysis was performed using the enrichr method in GSEAPY package (v0.12) with Gene Ontology biological process 2021 gene sets69,70,71,72,73.
Histology
Uninjured and injured sections of ear skin were fixed in 10% neutral formalin for 24 h and stored in 70% ethanol until paraffin embedding. Haematoxylin and eosin (H&E)-stained skin sections were used for histopathology analysis. Images were taken using an Olympus BX61 microscope equipped with a SPOT flex 15.2 64-Mp shifting pixel camera, 4×, 10× and 20× objectives, and SPOT v 5.2 software.
Western blot analysis
Uninjured and injured ear skin were lysed with ice-cold RIPA buffer (Pierce) supplemented with cOmplete Protease Inhibitor Cocktail and PhosSTOP (Sigma) and centrifuged at maximum speed for 30 min to collect lysates. Protein concentration was measured with the BCA protein assay (Pierce). An aliquot of 20–30 μg of total protein per sample was loaded into 7.5 or 10% Mini-PROTEAN TGX Precast Protein Gels (BioRad) and separated by SDS–PAGE. Proteins were transferred to PVDF membranes (BioRad). The following rabbit primary antibodies were used at the given concentrations; p-p44/42 MAPK (ERK1/2) (Thr202/Tyr204) (1:500, Cell Signaling 9101), p44/42 MAPK (ERK1/2) (1:500, Cell Signaling 4695), p-EGFR (Tyr1068) (1:100, Cell Signaling 2234), EGFR (1:100 Cell Signaling 4267; Extended Data Fig. 8b), EGFR (1:100, Cell Signaling 2232; Fig. 5b), p-AKT (Ser473) (1:200, Cell Signaling 4060), AKT (1:200, Cell Signaling 9262) and GAPDH (14C10) (1:500, Cell Signaling 2118). An anti-rabbit IgG HRP (1:500, Cell Signaling 7074) secondary antibody was used. Western blot analyses were performed on whole ear skin at 6 days after tamoxifen injection (3 days PWI).
Statistics and reproducibility
Statistical analyses were performed using an unpaired, two-tailed Student’s t-test for comparison between different groups of mice. Paired, two-tailed Student’s t-tests were used for comparison between tdTomato+ and tdTomato− populations in the same group of mice. Unpaired, ordinary one-way ANOVA was used for comparison between mice with three different genotypes. Statistical analyses were performed using Prism (v. 9) as indicated in the figure legends. Gene expression differences between different conditions from scRNA-seq data were performed with Student’s t-test and Holm–Sidak correction for multiple comparisons. P values of less than 0.05 were considered statistically significant (*P < 0.05, **P < 0.005, ***P < 0.0005, ****P < 0.0001). n is defined for each experiment, and always indicates the number of mice used for each condition examined. Box plots within violin plots denote the 25th, 50th and 75th quartiles, with whiskers depicting the minima and maxima of the data, excluding outliers that are beyond 1.5× the interquartile range.
Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.