Transcriptional signature for dysplastic epidermal development in human pre-malignant skin

To the Editor

Actinic keratoses (AKs) are prolific skin lesions, which occur in susceptible individuals due to chronic ultraviolet radiation exposure (Massone and Cerroni, 2015). The histopathologic features of AKs include a dysplastic epithelium that is frequently, but not always, spatially limited to the interfollicular epithelium. The follicular and acrosyringium epithelium appear relatively refractory to the development of dysplasia despite the fact that the surface of the upper follicular epithelium is equally susceptible to ultraviolet light exposure compared to the adjoining interfollicular epithelium. However, the cellular and molecular underpinnings underlying this clinical observation are lacking. To address this issue, we employed spatial transcriptomics to profile dysplasia-susceptible (interfollicular) versus -refractory (follicular) epithelium in AK tissue sections.

In AK sections, the overwhelming majority of normal appearing epithelium was, in fact, the upper follicular epithelium rather than interfollicular epithelium (Figure 1a–c; Supplementary Figure S1). We performed spatial transcriptomic profiling of five human AK lesions (AK1 – AK5) to assess whether a perturbed transcriptome may spatially correlate with the histopathologically distinct normal follicular versus dysplastic interfollicular epidermal regions. First, unsupervised clustering analysis successfully segregated epidermal from dermal compartments in all five AK specimens (Figure 1d; Supplementary Figure S1) thereby validating this approach. Interestingly, unsupervised clustering analysis also identified transcriptionally distinct regions of epithelium in four of the five AK lesions, AK1 – AK4, that spatially overlapped either normal follicular or dysplastic interfollicular regions in H&E-stained sections (Figure 1d, brown versus red clusters; Supplementary Figure S1). Spatial variable gene analysis of normal versus dysplastic epidermal clusters identified a total of 193 dysregulated genes (DEGs), 42 upregulated and 151 downregulated, in dysplastic interfollicular epithelium that were conserved across all four, AK1 – AK4, lesions (Figure 1e). Within the top 10% of common downregulated DEGs ranked by log2 fold change, we observed a high proportion of genes associated with epidermal terminal differentiation (Figure 1f), which is consistent with an epidermal dysplastic phenotype. However, the majority of top ranked common upregulated DEGs have little to no described role in epidermal differentiation or carcinogenesis (Figure 1f). To exclude less functionally relevant DEGs based on low RNA counts, we generated a higher dimensional view of top-ranking DEGs from dysplastic epidermis, integrating log2 fold change, P value, and average RNA counts, in AK1 – AK4 lesions. All highly significant downregulated DEGs also exhibiting higher average RNA counts were again related to epidermal terminal differentiation (Supplementary Figure S2). For upregulated DEGs, SLC1A5, SHISA2 and FLOT2 demonstrated relatively high RNA counts (Supplementary Figure S2) that were conserved across all four AK specimens (Figure 1f). A protumorigenic role for FLOT2 in the metastasis of murine SCC cells has been reported (Li et al, 2022). However, a role for FLOT2 in early epidermal dysplasia or SCC in human skin has largely been unexplored. SLC1A5 and SHISA2 are yet to be investigated in cutaneous photocarcinogenesis.

Figure 1. Spatial transcriptomic profiling in actinic keratosis. Figure 1. Spatial transcriptomic profiling in actinic keratosis.

(a-d) Representative images of H&E-stained (a-c) sections and mapping of assigned clusters (d) in human AK1. Bracketing demarcates follicular (FE) or interfollicular (IFE) epithelium. Right panels represent approximate 4–5X magnifications of normal (b) or dysplastic (c) epidermal regions from H&E panel a. Asterisks designate normal (b) or thinning (c) granular layer. Scale bar = 0.5 (a) or 0.1 (b-c) mm. (e) Venn diagrams showing upregulated and downregulated gene numbers in dysplastic versus normal epidermis that are conserved between AK1 - AK4 specimens. (f) Heat maps showing the top 15 upregulated and downregulated genes, ranked by P value (P<0.05), in dysplastic versus normal epidermis in AK1 – AK4 specimens. Abbreviations: d or dyspl, dysplastic; DEGs, dysregulated genes; epid, epidermis; n, normal; sup, superficial.

Next, we performed CellChat analysis (Jin et al, 2024) on AK1, AK2 and AK4 specimens, which employs a ligand-receptor database to identify significant cell-cell communication events that may be enriched between spatially localized cell populations. AK3 was not deemed suitable for CellChat analysis due to a lack of sufficient dermal tissue. For all three AK specimens, cluster interaction weight/strengths were highest in communication pathways between dermal and dysplastic epidermal clusters (Figure 2a–b). Interaction strength represents the summed probability of ligand-receptor signaling, highlighting the robust communication between cell types and suggests that perturbations in dermal-epidermal communication may be early drivers of epidermal dysplasia. Highly enriched ligand-receptor cell-cell communication pathways from dysplastic epidermis to the dermis were identified as MIF (keratinocyte-derived ligand)-CD74 (dermal-derived receptor) and MIF (keratinocyte-derived ligand)-CD44 (dermal-derived receptor) (Figure 2c). Highly enriched dermal origin pathways were identified as type I collagen (dermal-derived ligand)-CD44 and -SDC1 (keratinocyte receptors) (Figure 2d). To further validate these ligand-receptor candidates we performed immunofluorescence labeling in tissue sections from AKs and normal skin. We observed strong epidermal immunoreactivity for CD44 and MIF in AK and normal skin (Figure 2e–f; Supplementary Figure S3) and, interestingly, CD44 expression was selectively detected in the dermis of AK lesions but not in normal skin dermis (Figure 2e; Supplementary Figure S3). These observations validate our CellChat candidate pathways at the protein level and identify a predicted soluble, epidermal-derived MIF- dermal derived CD44 receptor paracrine pathway that is unique to AK lesions compared to normal skin and may be critical for early epidermal dysplastic development.

Figure 2. Inferred cell-cell communication networks in actinic keratosis lesions.

(a-b) Circle plots illustrating numbers (a) and strength (b) of significant ligand-receptor interactions between all clusters using CellChat analysis. (c-d) Bubble plots showing communication probability and statistical significance of the top enriched ligand-receptor pairs with either an epidermal (c) or dermal (d) origin. (e-h) Representative immunofluorescent detection of top cell-cell communication candidates CD44 (e), MIF (f), CD74 (g) and COL1A1 (h) in histological sections of AK. DAPI counterstain (blue) was conducted to visualize nuclei. Hashed lines demarcate the boundary between epidermis (epid) and dermis (derm). Scale bar = 50 μm.

Collectively, to our knowledge our findings elaborate a previously unreported transcriptional signature for early dysplastic epidermal development in human pre-malignant skin. Furthermore, the enriched transcriptional signature in dysplastic clusters provides much needed insight into the longstanding clinical observation of susceptible (interfollicular) versus refractory (follicular) epithelium to ultraviolet radiation-induced dysplasia. Our findings are consistent with recent lineage tracing studies in UV-treated skin showing that early mutant clones are primarily localized in the interfollicular epithelium rather than the follicular epithelium (Roy et al, 2020). Interestingly, UV-induced mutant clone size correlated with proximity to follicular epithelium, suggesting that, while the follicular epithelium is relatively refractory to dysplasia, there may be an indirect role for follicular epithelium in influencing the molecular nature of UV-induced interfollicular dysplasia.

Data Availability Statement

Datasets related to this article can be found at https://www.ncbi.nlm.nih.gov/sra/PRJNA1224987, hosted at NCBI SRA, accession number PRJNA1224987 and analysis code can be accessed at https://github.com/YvonneChen22/ak_data_analysis.git hosted by GitHub, accession ak_data_analysis.