Figure 1.
The epidermal gene expression landscape: Regions repressed and activated in psoriasis lesions.
We identified 235 gene modules based upon co-expression patterns in KC and epidermis microarray samples. These 235 modules were clustered, with distance between modules proportional to Euclidean distance between module medoids. For each module, a “median of medians” approach was used to calculate a score reflecting the typical PP/PN fold-change among member genes (horizontal axis). Labels for selected modules are listed in the right margin, with modules showing the most PP-increased or PP-decreased bias listed in the left-most columns. Red labels denote PP-increased modules (i.e., median PP/PN fold-change greater than 1) and blue labels denote PP-decreased modules (i.e., median PP/PN fold-change less than 1). Modules with significant bias towards PP-increased or PP-decreased expression are indicated by an asterisk symbol (FDR <0.05 by GSEA with median FC >1.25 or FC <0.80). In the right margin, red shade boxes denote landscape regions with a concentration of PP-increased modules, and blue shade boxes denote regions in which PP-decreased modules are concentrated.
Figure 2.
Differentially expressed module (DEM) cluster analysis.
We identified 30 modules with bias towards increased or decreased expression in PP skin compared to PN skin. Modules were clustered based upon expression patterns in KC and epidermis microarray samples, with distance between modules proportional to the Euclidean distance between module medoids. Grey boxes span the middle 50% of fold-change estimates (PP/PN) among module genes, and whiskers span the middle 80% of fold-change estimates (PP/PN) among module genes. The right margin lists the median fold-change (PP/PN) among genes belonging to each module, along with a p-value testing whether module genes were disproportionately increased or decreased in psoriasis lesions (FDR <0.05 in each case; GSEA).
Figure 3.
DEMs are biased towards increased or decreased expression in psoriasis epidermis (laser capture microdissection).
The 30 DEMs were evaluated to determine if member genes were disproportionately induced or repressed in LCM-dissected PP epidermis (GSEA). (A) GSEA detection rate curve area statistics for each of the 30 DEMs (red labels, PP-increased DEMs; blue labels, PP-decreased DEMs). Significant statistics are denoted by an asterisk symbol (FDR <0.05, yellow asterisk). Parts (B) and (C) show GSEA results for LCN2-33 and PKD1-53, respectively. Human genes were ranked according to their expression difference in LCM-dissected PP epidermis relative to LCM-dissected PN epidermis (horizontal axis). Low ranks were assigned to genes elevated in LCM-dissected PP epidermis (left, red region), while high ranks were assigned to genes repressed in LCM-dissected PP epidermis (right, blue region). Yellow hash marks (top) denote placement of DEM genes with respect to each ranking, and the curve in each figure tracks the cumulative overlap of DEM genes from left to right (vertical axis). Enrichment of DEM genes among genes elevated in LCM-dissected PP epidermis is indicated by a cumulative overlap curve above the diagonal (i.e., positive area statistic; Figure B). Enrichment of DEM genes among genes repressed in LCM-dissected PP epidermis is indicated by a cumulative overlap curve below the diagonal (i.e., negative area statistic; Figure C).
Figure 4.
Psoriasis DEMs are associated with regulatory axes connecting cytokine signals to transcription factors (STAT1, E2F, RUNX1 and NF-κB) and the SETDB1 histone methyltransferase.
The 30 DEMs were evaluated to assess whether member genes were regulated in cytokine-treated KCs (left) and to identify transcription factor binding sites enriched in sequences adjacent to genes within each DEM (right). Left: Cytokine-DEM relationships were investigated based upon transcriptional responses observed in KCs following treatment with one of the 18 listed cytokines. Red arrows indicate that DEM genes were disproportionately increased by cytokine treatment (P<10−3 by both GSEA and Fisher’s exact test). Blue arrows indicate that DEM genes were disproportionately decreased by cytokine treatment (P<10−3 by both GSEA and Fisher’s exact test). Arrow thickness corresponds to the strength of the cytokine-DEM association (thin arrows, P<10−3; middle thickness, P<10−6; thickest arrows, P<10−9). Right: DEMs were evaluated to identify the single motif most highly enriched in non-coding intergenic regions adjacent to member genes. For each DEM, the motif showing the highest enrichment is listed in the right margin, with significant motifs shown in magenta font (FDR <0.10). The Z statistic quantifies the degree of motif enrichment, based upon a comparison of motif frequency in intergenic regions near DEM genes versus frequency in all other intergenic regions associated with epidermis-expressed genes (semiparametric generalized additive logistic model; see Methods). Within each bar, the text indicates the fraction of DEM genes potentially regulated by the motif listed in the right margin, i.e., the fraction of DEM genes for which at least one site matching the indicated motif was identified within flanking intergenic sequence.
Figure 5.
STAT1-57 is an interferon-regulated DEM repressed by the TP63/ZNF750/KLF4 differentiation axis.
Part (A) shows the expression profile of the 57 genes belonging to STAT1-57 (KC and epidermis microarray samples). The red line represents median expression among the 57 genes, dark grey outlines expression for the middle 50% of genes, and light grey outlines expression for the middle 80% of genes. Figures (B) – (E) show GSEA results evaluating whether the 57 genes are disproportionately increased or decreased in (B) KCs treated with IFN-γ versus untreated control KCs (GSE2737) (C) KCs treated with IFN-α versus untreated control KCs (GSE36287), (D) KCs with siRNA knockdown of TP63 versus scambled siRNA (GSE33495) or (E) KCs with siRNA knockdown of ZNF750 versus scambled siRNA (GSE32685). In each figure (B – E), genes were first ranked according to how their expression is altered in the indicated comparison (horizontal label). Red background denotes genes increased in each comparison while blue background denotes genes decreased in each comparison. Yellow hash marks (top) denote placement of the STAT1-57 genes with respect to each ranking, and the curve in each figure tracks the cumulative overlap of STAT1-57 genes with top-ranked genes from left to right (vertical axis). Enrichment of STAT1-57 members among genes increased in each comparison is indicated by a cumulative overlap curve above the diagonal (i.e., positive area; Figures B – E).
Figure 6.
STAT1-57 genes are embedded in genome regions with increased density of an ISRE-like motif recognized by multiple transcription factors (STAT1, IRF1 and the ISGF3 complex).
(A) Ten motifs most enriched in non-coding intergenic sequences adjacent to STAT1-57 genes. (B) Ten motifs most enriched in conserved regions 1 KB upstream of STAT1-57 genes. (C) Sequence logo for STAT1|GRAANNGAAAST. (D) Sequence logo for Isgf3g|RAAWCGAAACT|UP00074. (E) Potential regulatory “hotspots” in 2 KB upstream regions upstream of the ten STAT1-57 genes most strongly elevated in psoriasis lesions. Upstream regions were scanned to identify loci with matches to motifs highly enriched with respect to the complete set of STAT1-57 genes (dark grey = best 400 BP window; yellow = best 200 BP window; orange = best 100 BP window; red = best 50 BP window). Blue circles correspond to STAT1|GRAANNGAAAST binding sites. (F) Zoomed view of potential regulatory hotspot 50-130 BP upstream of the IFI27 transcription start site (hg19, chr14, 94576948–94577028).
Figure 7.
STAT1 and IRF9 mRNA are elevated in psoriasis lesions with increased abundance of pSTAT1(ser727) and IRF9 protein in the psoriatic epidermis.
Figures (A), (D) and (G) show the PP/PN fold-change distributions for STAT1, IRF1 and IRF9, respectively (n = 215 patients). Figures (B), (E) and (H) show staining for pSTAT1(ser727), IRF1 and IRF9 in lesional (PP) skin, respectively. Figures (C), (F) and (I) show staining for pSTAT1(ser727), IRF1 and IRF9 in uninvolved (PN) skin, respectively.
Figure 8.
Proposed model for STAT1-57-driven activation of inflammatory and hyperproliferative responses in psoriasis lesions.
STAT1-57 is a cytokine hub module consisting of 57 genes co-expressed in KCs and epidermal isolates. In cultured KCs, the 57 genes are biased towards increased expression following treatment with IL-1α, IFN-α, IFN-γ, TNF, IL-36γ and OSM. Similarly, expression of STAT1-57 genes is elevated in hyperproliferative states, such as during wounding and in squamous cell carcinoma (SCC), basal cell carcinoma (BCC) and actinic keratosis (AK). In contrast, in cultured KCs, the 57 genes are biased towards decreased expression following stimulation by IL-17A, IL-4, or glucocorticoid (dexamethasone). RNAi experiments using cultured KCs were also consistent with repression of STAT1-57 genes by the TP63/ZNF750/KLF4 differentiation axis. These upstream signals influence STAT1-57 transcription, potentially by modulating interactions between an ISRE-like motif (GRAANNGAAAST) and the IGSF3 complex (STAT1, STAT2 and IRF9). Activation of STAT1-57 leads to upregulation of mRNAs encoding proteins that promote inflammation (ISG15 and DDX58), inhibit apoptosis (IFI6), or which enable complete activation of EGF-dependent signaling (PLSCR1). Arrows represent activation and round-tipped lines denote inhibition.