Comprehensive Multi-Species and Multi-Omics Approach Details Skin Tissue Analysis

A series of studies meticulously detailed extensive methodologies for collecting, processing, and analyzing skin tissue samples across a wide range of species, including humans and various animals. These methods encompassed traditional histological techniques, advanced spatial and single-cell omics technologies, and sophisticated computational pipelines.

The primary goal of these studies was to investigate cellular composition, spatial organization, and cell-cell communication within the skin.

Tissue Sample Collection and Preservation

Skin samples were acquired from a diverse array of adult and neonatal species. Animal subjects included naked mole rats, rhesus macaques, common marmosets, bottlenose, long-beaked common, and short-beaked common dolphins, North American grizzly bears, various mouse strains (WT C57BL/6, K14-Cre;Lef1fl/fl, K14-CreERT;Bmpr1afl/fl, K14-CreERT), and multiple pig types (E90, P3, P10, 6mo, EDA-KO, Yucatan miniature hairless, Hanford miniature, Mangalitsa).

Human samples were collected from gestational and adult donors, including de-identified autopsy samples (post-mortem interval < 20 hours, no skin disease history), surgical discards from Mohs reconstruction, and panniculectomy procedures. Samples were obtained from various anatomical sites such as back skin, trunk skin, dorsal rump, digits, and abdominal skin.

All animal studies adhered to approved Institutional Animal Care and Use Committee (IACUC) protocols, and human tissue collection followed Institutional Review Board (IRB) approved protocols with informed maternal or patient consent.

Tissue preservation involved fixation in 4% paraformaldehyde (PFA) or 10% neutral buffered formalin, followed by storage in 70% ethanol, or cryopreservation in Optimal Cutting Temperature (OCT) compound at -80 °C. For specific human samples, a 10 nM Ribonucleoside-Vanadyl Complex in PBS on ice was used before fixation or OCT embedding.

Histological and Cellular Analysis

Tissue processing for histological examination involved sectioning paraffin-embedded tissues at 5 μm or 10 μm, and cryo-preserved samples at 10-15 μm. Sections were stained using Hematoxylin and Eosin (H&E) or Herovici’s polychrome. Coverslips were mounted using Permount or DPX.

Imaging was conducted using a variety of microscopes, including Nikon Eclipse E600, Keyence BZ-X810, Leica cryostat, Olympus FV3000, Leica SP5/SP8 confocal, Leica DMI8 systems, Hamamatsu Nanozoomer S210, and Leica upright widefield microscopes.

Measurements included:

• Epidermal and stratum corneum thickness
• Rete ridge density
• Apical ridge length
• Hair density
• Scar size
• MKI67+/BrdU+ cell counts in different epidermal layers

Specialized Porcine Studies

• Wound Healing: Neonatal pigs received 2.5 × 2.5 cm full-thickness wounds. Wound size was periodically assessed, and wound sites were collected for histological analysis at 28, 43, and 58 days post-wounding (dpw).
• BrdU Labeling: Neonatal pigs were injected intraperitoneally with 50 mg/kg of BrdU from P5 to P7. Tissues were collected at P8, P12, and P16 for histological analysis to evaluate cell proliferation.

Immunofluorescence and RNA FISH

Both cryo-preserved (10-60 μm sections) and paraffin-embedded (10 μm sections) tissues underwent standard immunofluorescence protocols, including antigen retrieval for paraffin sections. A panel of primary antibodies targeting specific proteins (e.g., ITGA6, LEF1, αSMA, PDGFRA, KRT10, MKI67, BrdU, SMAD1/5, PECAM1, PDGFC, activated Caspase-3) was utilized. Secondary antibodies included various Alexa Fluor conjugates, and DAPI was used for nuclear counterstaining. Imaging was primarily conducted using Leica SP5/SP8 confocal microscopes, with additional imaging on Leica DMI8 fluorescence and Olympus FV3000 confocal microscopes.

Tissues were prepared using the RNAscope protocol for fixed frozen tissue. Sections (10 μm) were placed on glass slides, and target RNA transcripts (e.g., Hs-CCL19-C1, Hs-PDGFRA-C2) were detected. Fluorescence signal development used RNAscope VIVID dyes with DAPI counterstain, and imaging was performed on a Leica upright widefield microscope. Quantification involved manually counting specific cell types (e.g., PDGFRA+/CCL19+ cells).

Human Skin Explant Culture and qPCR

Full skin thickness samples from the abdomen of de-identified panniculectomy donors were cultured as 8-mm punch biopsies on Surgifoam in 24-well plates with DMEM supplemented with 1% FBS and 1% penicillin–streptomycin, maintaining an air–liquid interface. After 24 hours, media was replaced with recombinant human TNF (10 ng ml−1) or PBS + 0.1% BSA vehicle control for another 24 hours before collection for histology, RNA FISH, immunofluorescence, and qPCR.

RNA was isolated from fresh human abdominal skin and explants using QIAzol Lysis Reagent and purified with the Quick-RNA Microprep Kit. Reverse transcription was performed using the iScript cDNA Synthesis Kit, followed by qPCR using Luna Universal qPCR Master Mix in technical quadruplicates on a LightCycler 480 System. Relative gene expression was determined using the 2−ΔΔCt method.

Advanced Omics Technologies

Single-cell suspensions were generated from pig skin (E90, P3, P10, and 6 mo) through enzymatic digestion. Libraries were prepared with the 10x Genomics scRNA-seq 3′ V3 Kit and sequenced on an Illumina NovaSeq PE150. Publicly available human and mouse scRNA-seq datasets were also reanalyzed.

PFA-fixed frozen cryo skin samples from P3, P10, and 6 mo pigs were sectioned at 10 μm for Stereo-seq using a Complete Genomics T FF v.1.2 kit and DNBSEQ-T7 sequencing.

Human skin samples were prepared following Vizgen's MERSCOPE user guide with modifications to improve tissue adherence, decrosslinking, digestion, clearing, photobleaching, and imaging (e.g., weighted gel embedding, extended DAPI staining). Gene panels included canonical cell-type markers, highly expressed ligand-receptor pairs, and overlapping genes for integration.

Publicly available Visium data from six studies, covering normal/nonlesional skin and various skin diseases (e.g., BCC, SCC, HS, AD, psoriasis), were downloaded and processed.

Computational Methods and Data Analysis

Extensive computational methods were employed for data processing, analysis, and integration across various platforms.

Utilized 10x Genomics Cell Ranger for alignment to the Sscrofa.11.1 genome, and the Seurat package in R for quality control, normalization (SCTransform), dimensional reduction (UMAP), clustering (SLM algorithm), and cell type annotation based on canonical markers (e.g., keratinocytes, fibroblasts, pericytes, vascular cells, sweat gland cells). Pseudotime analysis was performed using Monocle3. Public scRNA-seq data were integrated using Harmony after quality control, normalization, scaling, and PCA.

Involved SAW for gene expression matrix generation and alignment. Analysis utilized Stereopy in Python for quality control, normalization (sctransform), clustering (Leiden algorithm), and spatial cell type assignment based on canonical markers and spatial localization.

Segmentation was performed with Vizgen postprocessing tool (VPT) using Cellpose with DAPI and Cellbound3 stain or DAPI alone, further optimized with Baysor. Quality filters removed cells with volume <100 μm3 or <10 detected transcripts. Tissue compartments (dermis, epidermis, subcutis) were manually annotated, and areas calculated using alphashape and geopandas.

Global integration of MERFISH samples used scVI for batch correction, followed by Leiden clustering, curation into 18 broad cell types, and further subclustering using Harmony, resulting in 45 distinct cell types. Missing gene expression in the MERFISH panel was imputed using Tangram with an scRNA-seq dataset as a reference.

CellChat and Spatial CellChat were employed to infer pathway and ligand–receptor interactions from both scRNA-seq and Stereo-seq datasets, respectively, utilizing the human ligand–receptor database. Analyses focused on core basal and dividing keratinocyte, papillary fibroblast, pericyte, and blood vessel clusters.

Spatial CellChat constrained interactions to biologically realistic distances, distinguishing between secreted and contact-dependent signaling. L–R expression was visualized using a k-nearest-neighbors smoothing approach and analyzed for pseudobulked coexpression. CellChat was also run on scRNA-seq samples simulating neighborhood interactions transcriptome-wide to cross-validate MERFISH hits.

Integrated latent space embeddings of the MERFISH dataset were adjusted with scANVI before running CellCharter to identify multicellular neighborhoods. The optimal number of neighborhoods was selected based on local maxima of the stability metric. Neighborhoods were annotated based on cellular composition and spatial localization. CellCharter and squidpy functions were used to determine spatially proximal cell types.

Crumblr, a mixed linear model framework for compositional data using centered log ratio, and Dream were used to determine differentially abundant cell types per sample on a per-anatomic site basis for MERFISH data. Donor sex, donor ID, and imaging batch were included as random effects, with donor age and dermal, epidermal, and subcutis compartment areas as fixed effects. For neighborhood differential abundance, total tissue area was used. Visium spot clusters or annotated neighborhoods were similarly analyzed for differential abundance relative to normal skin samples for various diseases, with study and donor ID as random effects.

The scRNA-seq object was filtered and used to train the cell2location model, which was then run on each Visium tissue sample. To map MERFISH-defined neighborhoods onto Visium data, differential expression analysis identified top genes per neighborhood, and Seurat's AddModuleScore() calculated a neighborhood score for each Visium spot.

Statistical Analysis

Statistical analyses were conducted using R (v.4.2.2 and v4.4.0) and Python (v3.12.9). Methods included one-way analysis of variance (ANOVA) with post hoc Tukey’s HSD, Welch’s two-sample t-test, and linear regression. A P-value of < 0.05 was considered statistically significant.

No data points were excluded from statistical analysis. All available samples were included, and no statistical method predetermined sample size.