Biogeographic and disease-specific alterations in epidermal lipid composition and single-cell analysis of acral keratinocytes

Abstract

The epidermis is the outermost layer of skin. Here, we used targeted lipid profiling to characterize the biogeographic alterations of human epidermal lipids across 12 anatomically distinct body sites, and we used single-cell RNA-Seq to compare keratinocyte gene expression at acral and nonacral sites. We demonstrate that acral skin has low expression of EOS acyl-ceramides and the genes involved in their synthesis, as well as low expression of genes involved in filaggrin and keratin citrullination (PADI1 and PADI3) and corneodesmosome degradation, changes that are consistent with increased corneocyte retention. Several overarching principles governing epidermal lipid expression were also noted. For example, there was a strong negative correlation between the expression of 18-carbon and 22-carbon sphingoid base ceramides. Disease-specific alterations in epidermal lipid gene expression and their corresponding alterations to the epidermal lipidome were characterized. Lipid biomarkers with diagnostic utility for inflammatory and precancerous conditions were identified, and a 2-analyte diagnostic model of psoriasis was constructed using a step-forward algorithm. Finally, gene coexpression analysis revealed a strong connection between lipid and immune gene expression. This work highlights (a) mechanisms by which the epidermis is uniquely adapted for the specific environmental insults encountered at different body surfaces and (b) how inflammation-associated alterations in gene expression affect the epidermal lipidome.

Keywords: Dermatology; Skin.

Figure 1. Biogeographical variation of epidermal ceramides.

(A) Analysis of tape-stripping samples by targeted mass spectrometry. (B) The relative abundance of 351 lipids was simultaneously monitored across 12 different anatomic locations; representative ceramides are shown. Sites monitored included abdomen (AB), antecubital fossa (AC), alar crease (AL), axillae (AX), cheek (CK), dorsal surface of the hand (DH), glabella (GB), popliteal fossa (PF), plantar heel (PH), anterior proximal lower extremity (PLE), upper back (UB), and volar forearm (VF). Results are presented as box-and-whisker plots, where the upper and lower bars connected to each box indicate the boundaries of the normal distribution, and the box edges mark the first and third quartile boundaries within each distribution. The dark horizontal line represents the median. The relative abundances of the NH ceramide Cer(t20:1[6OH]/26:0), the NP ceramide Cer(t18:0/26:0), and the AH ceramide Cer(t18:1[6OH]/22:0[2OH]) were highly variable across different anatomic locations. (C) In contrast, the expression of the ADS dihydroxyceramide Cer(d18:0/18:0[2OH]) was largely invariable across all anatomic sites, a finding representative of all ADS ceramides monitored (Supplemental Table 1). Acyl-ceramides EOS ω-linoleoyloxy-Cer(d18:1/30:0) and EOH ω-linoleoyloxy-Cer(t20:1[6OH]/32:0) are also shown. Note the low expression of the EOS ceramide in PH epidermis, which was a typical finding for EOS ceramide expression (Supplemental Table 1).

Figure 2. Biogeography of epidermal lipids across 12 anatomical sites.

Anatomical sites used for skin adhesive tape sampling are shown. Two pie charts are shown for each anatomical location, one containing ceramides (left) and the other cholesterol and unsaturated fatty acids (right). For each pie chart, 100% equals the combined abundance of listed lipids. Lipids included in the pie charts are color coded and listed in the key. Across different sites, some pie charts look similar to one another (e.g., GB, AL, and CK), indicating similarities in lipid composition.

Figure 3. Clustering of different anatomical surfaces by similarities in epidermal lipid composition.

A cluster dendrogram was constructed to assess similarities in epidermal lipid composition between different anatomical surfaces. Anatomic locations that cluster together are depicted in the same color. The height of the link connecting different anatomical locations represents how similar their respective epidermal lipid compositions are. Note that PH does not cluster well with other anatomic sites.

Figure 4. Palmoplantar skin has a unique epidermal lipid composition.

(A) Results are displayed as box-and-whisker plots, as described in Figure 1. Representative lipids were chosen to highlight the differences in lipid composition between PH (red arrow) and other anatomic locations. Note the low abundance of AS ceramide Cer(d16:1/28:0[2OH]) in PH skin and the increased abundance of AH ceramide Cer(t18:1[6OH]/22:0[2OH]). This pattern was also observed for all monitored ceramides with similar length sphingoid and fatty acid moieties (Supplemental Figure 2 and Supplemental Table 1). The acyl-ceramide EOSω-linoleoyloxy-Cer(d18:1/30:0) was also decreased in PH epidermis. (B) Additional examples of differentially expressed ceramides in PH skin that follow the same trends as presented in A. (C) Principal component analysis of lipid-associated metabolic gene expression data revealed complete separation of samples by anatomic location. (D) Whole-tissue RNA-Seq performed on palm and trunk biopsy specimens identifies differentially expressed lipid-associated metabolic genes. Genes presented here are relevant to the synthesis of the differentially expressed ceramides in PH epidermis presented in A and B.

Figure 5. Differential expression of lipid-associated metabolic genes assessed by single-cell sequencing.

(A) Individual keratinocyte transcriptome data are presented using the uniform manifold approximation and projection (UMAP) method. Each dot represents an individual keratinocyte. Note that keratinocytes originating from the same anatomic location and epidermal layer cluster together (basal [palm, green; trunk, yellow], spinous [palm, blue; trunk, pink], and granular [palm, red; trunk, orange] layers). Cells expressing the noted lipid-associated are depicted in purple. Dashed circles are drawn around the keratinocyte population, in which the expression of the noted gene is most strongly upregulated. (B) Predicted alterations in ceramide lipid expression based on ceramide synthase gene expression in trunk versus palm skin. (C) Single-cell RNA-Seq data of palm and trunk epidermal granular layer keratinocytes presented as box-and-whisker plots. Genes selected for presentation are relevant to the observed lipid alterations in PH skin. Each individual data point represents the number of reads that mapped to the indicated gene in a unique granular layer keratinocyte.

Figure 6. Transcriptome alterations in acral keratinocytes provide insight into the distinctive epidermal features of palmoplantar skin.

(A) UMAP dimensionality reduction plots of single-cell RNA-Seq data. Single-cell RNA-Seq was performed on keratinocytes isolated from paired palm and trunk skin biopsies. Each dot represents an individual keratinocyte (basal [palm, green; trunk, yellow]; spinous [palm, blue; trunk, pink]; and granular [palm, red; trunk, orange] layers). Cells expressing the noted gene of interest are depicted in purple. Dashed circles are drawn around the keratinocyte population in which the expression of the noted gene is most strongly upregulated. Arrows to the left of gene names represents the directionality of gene expression in acral keratinocytes. (B) Single-cell RNA-Seq data presented as box-and-whisker plots. Each individual data point represents the number of reads that mapped to the indicated gene in a single granular layer keratinocyte. (C) Paired biopsies obtained from palm and trunk skin were evaluated by RNA-Seq. Results of select genes relevant to keratin citrullination and corneodesmosome degradation are presented as box-and-whisker plots.

Figure 7. Acral skin is associated with altered expression of keratin genes.

(A) UMAP dimensionality reduction plots of single-cell RNA-Seq data. Single-cell RNA-Seq was performed on keratinocytes isolated from paired palm and trunk skin biopsies as described in Figure 6. (B) Basal and granular layer single-cell keratin expression. Each individual data point represents the number of reads that mapped to the indicated gene in a single basal or granular layer keratinocyte. (C) RNA-Seq was performed on whole-skin biopsies obtained from palm and trunk. Box-and-whisker plots of representative keratin genes differentially expressed in palm skin are shown. (D) Box-and-whisker plot of KRT9 and KRT10 expression in palm skin. Scatter plot of KRT9 versus KRT1 expression in individual granular layer palm keratinocytes. Each dot represents a single granular layer keratinocyte. Note the strong correlation between KRT1 and KRT9 within each cell. (E) Correlation scatter plot of KRT15 and KRT78 in cultured keratinocytes, where each data point represents a different primary keratinocyte culture.

Figure 8. Lipid alterations in psoriasis skin.

(A) Abundance of epidermal lipids in different dermatologic diseases as determined by targeted mass spectrometry. Results are displayed as box-and-whisker plots (as described in Figure 1). Representative lipids were chosen to highlight the characteristic patterns of lipid expression in lesional psoriasis epidermis. Red, purple, and black arrows highlight the upregulation of NS ceramide Cer(d18:1/16:0) (upper left corner) in psoriasis, atopic dermatitis, and tinea corporis lesional epidermis, respectively. NS ceramide Cer(d18:1/16:0) is an example of how certain ceramides with 18-carbon sphingoid bases are upregulated in inflammatory skin. FA 24:1 illustrates the general upregulation of unsaturated fatty acids in psoriasis, and EOS ceramide ω-linoleoyloxy-Cer(d20:1/29:0) is an example of a differentially expressed EOS ceramide. (B) Additional examples illustrating the trends in ceramide expression in psoriasis lesional epidermis. Upper row presents lipid expression data from paired psoriasis lesional and nonlesional samples. Lower row presents data for psoriasis versus healthy controls. (Results for additional monitored ceramides demonstrating these same trends can be found in Supplemental Figure 7, and results for all monitored lipids can be found in Supplemental Table 5.) (C) Differential expression of EOS ceramides in psoriasis depends in part on the length of their sphingoid base. (D) Predicted alterations in ceramide synthesis in psoriasis skin based upon psoriasis-associated alterations in lipid-gene expression.

Figure 9. Similarities and differences between the lipid alterations seen in atopic dermatitis and psoriasis.

(A) Whole-tissue RNA-Seq of psoriasis lesional skin identifies differentially expressed lipid-associated metabolic genes. Results are presented as box-and-whisker plots. (Additional gene expression data is presented in Supplemental Figure 8.) (B) Pie charts of epidermal lipid expression in control healthy skin, lesional atopic dermatitis skin, and lesional psoriasis skin. Note that 100% equals the combined abundance for listed lipids, not all epidermal lipids. The pie charts on the left are of representative fatty acids. The relative abundances of cholesterol and ceramides subclasses are depicted in the pie charts in the center. Pie charts on the right are of representative ceramides. Note that the strong upregulation of NS ceramide Cer(d18:1/16:0) in psoriasis makes it difficult to appreciate changes in other ceramide structures. (C) Examples of differentially expressed epidermal ceramides in psoriasis versus atopic dermatitis.

Figure 10. Epidermal lipid expression can diagnose skin diseases.

Tape strippings were performed on lesional skin and control healthy skin. Each column represents a stratum corneum tape-stripping sample (grouped by diagnostic category). (A) Heatmap of lipid abundance in psoriasis (NN = normal healthy skin [gray, n = 20], PN = nonlesional psoriasis skin [blue, n = 16], and P = lesional psoriasis skin [red, n = 37]). Each row represents a monitored lipid, with red representing increased expression and blue representing decreased expression. Rows were sorted by standard mean difference (SMD). To construct heat maps to compare lipid abundances, lipid peak intensity values were preprocessed using the “scale and center” function in R, which subtracts the mean value of the analyte and divides the result by the standard deviation for that analyte. (B) Heatmap of lipid abundance in atopic dermatitis (AN = nonlesional atopic dermatitis skin [green, n = 10] and AD = lesional atopic dermatitis skin [purple, n = 9]). (C) Principal component analysis of relative lipid abundance data. Each dot represents 1 epidermal sample. Each color represents a diagnostic group (normal healthy skin [light gray], psoriasis lesional [red], psoriasis nonlesional [light blue], atopic dermatitis lesional [purple], atopic dermatitis nonlesional [green], actinic keratosis lesional [yellow], seborrheic keratosis lesional [dark blue], and tinea lesional [dark gray]). (D) Cluster dendrogram with Euclidian distance represented on the horizontal axis (AD, atopic dermatitis lesional; AK, actinic keratosis lesional; AN, atopic dermatitis nonlesional; NN, normal healthy skin; PN, psoriasis nonlesional; PP, psoriasis lesional; TI, tinea lesional). (E) Left: Receiver operating characteristic curve (ROC) for the single analyte classifier, NH ceramide Cer(t18:1[6OH]/30:0), demonstrates its ability to distinguish psoriatic lesional skin (PP) from all other diagnostic groups (NN, PN, PP, AK, SK, TI, and AD) combined (AUC, 0.96). Right: ROC for the 2-analyte classifier, NH ceramide Cer(t18:1[6OH]/30:0) + AS ceramide Cer(d16:1/28:0[2OH]), capable of distinguishing psoriatic lesional skin from all other diagnostic groups combined. The AUC for the 2-analyte classifier was 0.98 ± 0.02 (5-fold cross-validated).

Figure 11. Analysis of epidermal lipid expression uncovers highly significant lipid-lipid expression patterns.

(A) Correlation matrix depicting every possible pairwise lipid-lipid correlation among the 351 monitored lipids, with 123,201 combinations in total. The 351 columns and rows represent the monitored lipid analytes. The intensity of the color at the intersect between a column and row represents the strength of the correlation for that particular lipid-lipid combination (positive correlation, red; negative correlation, blue; no correlation, yellow). The checkerboard pattern indicates a consistent pattern of intraclass and interclass lipid correlations. (B) Scatter plots of representative lipid-lipid correlations. Intrasubclass ceramides with the same sphingoid base and similar fatty acid moieties positively correlated with one another. For example, the AH ceramide Cer(t18:1[6OH]/20:0[2OH]) positive correlated with AH ceramide Cer(t18:1[6OH]/22:0[2OH]) (r = 0.99, FDR = 2.5 × 10^–86). Likewise, interclass ceramides with the same length sphingoid base and the same or similar fatty acid moieties positively correlated with one another. Shown here, the AS ceramide Cer(d18:1/20:0[2OH]) positively correlated with AH ceramide Cer(t18:1[6OH]/22:0[2OH]) (r = 0.93, FDR = 2.9 × 10^–44). Also, 18-carbon sphingoid base ceramides negatively correlated with 20- and 22-carbon sphingoid base ceramides, usually with dissimilar length fatty acids. Also shown, the NP ceramide Cer(t22:0/26:0) negatively correlated with the AH ceramide Cer(t18:1[6OH]/22:0[2OH]) (r = –0.93, FDR = 1.5 × 10^–46). (C) Unsaturated fatty acids of similar length tended to positively correlate with one another. Shown here, FA 22:1 positively correlated with FA 24:1 (r = 0.95, FDR = 2.9 × 10^–55). FA 24:1 (and to a lesser extent FA 22:1 and sometimes FA 20:1) positively correlated with NS(C18) and NDS(C18) ceramides. Also shown, FA 24:1 positively correlated with the NS ceramide Cer(d18:1/20:0) (r = 0.88, FDR = 2.2 × 10^–33) and the NDS ceramide Cer(d18:0/20:0) (r = 0.87, FDR = 3.0 × 10^–32). (D) Bar graphs illustrate the percent of ceramides that follow the patterns described in B.

Figure 12. Lipid expression patterns.

(A) Correlation matrix representing the patterns of positive correlations among different lipid subclasses. The intensity of the color at the intersect between a column and row represents the average correlation coefficient for that particular lipid subclass combination. The size of the circle within each colored box represents the percent of lipids that positively correlated. Hierarchical clustering was used to order the lipid subclasses based on the similarities between their patterns of correlation. (B) Ceramides were grouped by the length of their sphingoid bases. Lipid expression across different groups was then assessed, and a correlation matrix was constructed. The intensity of color at the intersect between a column and row represents the average positive correlation coefficient for that particular group comparison. The size of the circle within each colored box represents the percent of lipids that correlated. (C) Schematic seesaw diagrams depicting common patterns of lipid-lipid correlations.

Figure 13. The ELOVL4 expression correlates with immune and skin barrier genes.

(A) Hierarchical clustering of lipid genes by their expression yields 2 clear clusters, one centered on ELOVL4 and the other on ELOVL6. The lipid genes coexpressed with ELOVL4 include ELOVL1, ELOVL3, ELOVL7, and CERS3. These genes negatively correlated with the lipid-genes within the ELOVL6 cluster (ELOVL5, ELOVL6, CERS2, CERS5, CERS6, and ELOVL2). The size of the circle within each box is proportional to the significance of the intersecting lipid-lipid correlation, while the color represents the correlation coefficient of the comparison. (B) The t-SNE nonlinear dimensionality reduction method was used to create a 2-dimensional plot of the keratinocyte transcriptome from RNA-Seq data obtained from 50 primary human keratinocytes cell lines. Within this plot, each point represents a keratinocyte-expressed gene, and the distance between the points is inversely related to how strongly the genes correlated with one another. Representative genes that cocluster with ELOVL4 are listed on the right, and they include various inflammatory mediators (e.g., IL36B, in blue). (C) Individual gene expression scatter plots reveal strong correlations between the expression of ELOVL4 (x axis) and representative coclustering genes (y axis). In these plots, each dot represents a unique in vitro cultured primary human keratinocyte cell line and culture condition (50 unique primary human keratinocyte cell lines were each cultured under 8 different conditions; see Supplemental Methods). ELOVL4 strongly correlated with lipid genes involved in ceramide and fatty acid synthesis (pink), as well as select keratin (green) and corneodesmosone-related (yellow) genes. Note the strong correlation between ELOVL4 and various immune-related genes (blue).

Figure 14. Altered ELOVL4 expression impacts expression of immune genes in keratinocytes.

(A) RNA-Seq data sets of psoriasis lesional skin were mined for gene expression of ELOVL4, IL36B, PDE4A, CHMP2B, CDK7, and S100A13. These genes were selected because they were highly coexpressed with ELOVL4 in primary human keratinocytes and are known inflammatory mediators that are differentially expressed in psoriasis lesional skin. A metaanalysis was performed to combine the ELOVL4 coexpression results across 4 independently acquired RNA-Seq data sets. Forest plots are presented, and P values for the final model are shown. (B) Keratinocyte RNA-Seq data sets were parsed into 3 groups based on the ELOVL4 allele they expressed (0/0 representing the ELOVL4 reference allele, and 0/1 and 1/1 representing heterozygosity and homozygosity for the ELOVL4 variant, rs62407622). Box plots of ELOVL4 expression revealed that keratinocytes homozygous for rs62407622 expressed significantly reduced levels of ELOVL4. Keratinocytes homozygous for the ELOVL4^lo variant also expressed significantly lower levels of CDK7, CHMP2B, IL36B, and MAP3K8 and significantly increased levels of PDE4A, S100A13, and TLR3. (C) HaCat and N/TERT immortalized keratinocyte cell lines were treated with ELOVL4 siRNA or control scrambled RNA. ELOVL4 siRNA knockdown significantly downregulated expression of CDK7, CHMP2B, DNAJA2, and MAP3K8. It also increased expression of PDE4A in N/TERT cells and TLR3 in HaCaT cells. (D) ELOVL4 coexpression network. Line thickness is directly proportional the correlation coefficient for the coexpression of the connecting genes. Red lines indicate a positive correlation, and blue lines indicate a negative correlation.