Rational design of a JAK1-selective siRNA inhibitor for the modulation of autoimmunity in the skin

Abstract

Inhibition of Janus kinase (JAK) family enzymes is a popular strategy for treating inflammatory and autoimmune skin diseases. In the clinic, small molecule JAK inhibitors show distinct efficacy and safety profiles, likely reflecting variable selectivity for JAK subtypes. Absolute JAK subtype selectivity has not yet been achieved. Here, we rationally design small interfering RNAs (siRNAs) that offer sequence-specific gene silencing of JAK1, narrowing the spectrum of action on JAK-dependent cytokine signaling to maintain efficacy and improve safety. Our fully chemically modified siRNA supports efficient silencing of JAK1 expression in human skin explant and modulation of JAK1-dependent inflammatory signaling. A single injection into mouse skin enables five weeks of duration of effect. In a mouse model of vitiligo, local administration of the JAK1 siRNA significantly reduces skin infiltration of autoreactive CD8⁺ T cells and prevents epidermal depigmentation. This work establishes a path toward siRNA treatments as a new class of therapeutic modality for inflammatory and autoimmune skin diseases.

Fig. 1. Identification of lead siRNA 3033 that potently silences JAK1 expression.

a Mechanism of action of RNAi for selective silencing of JAK1 to narrow down the spectrum of action on inflammatory cytokine pathways. b, c In vitro screening of siRNAs in human HeLa (b) and mouse N2a (c) cells. Cells were treated with fully modified cholesterol-conjugated siRNAs at 1.5 μM for 72 h. mRNA levels were measured by the QuantiGene 2.0 assay. siRNA number represents the 5’-position of the mRNA target site. UNT untreated control, NTC nontargeting control siRNA. Data are represented as percent of UNT (n = 3 biologically independent samples, mean ± s.d.). d, e Seven-point dose–response curves of lead siRNAs 3033, 883, 4470 in human HeLa cells (d) and lead siRNAs 3033, 3232, and 3785 in mouse N2a cells (e). M, molar concentration of siRNA (n = 3 biologically independent samples, mean ± s.d., ****P < 0.0001; two-way ANOVA). f Accession number and mRNA targeting site of lead compound 3033. g JAK1 protein expression in human HeLa and mouse N2a cells (in a single capillary cartridge using western capillary ProteinSimple assay) at 72 h post siRNA treatment (1.5 μM). h Quantification of JAK1 protein in (g). (n = 4 biologically independent samples, mean ± s.d. ****P < 0.0001; two-sided unpaired t test). Source data are provided as a Source Data file.

Fig. 2. Selective silencing of JAK1 by siRNA inhibits IFN-γ signaling.

a Activity of 3033 on JAK family enzymes. Human HEL 92.1.7 cells were treated with siRNA at 1.5 μM for 72 h, and mRNA levels of JAK1, JAK2, JAK3, and TYK2 were measured using the QuantiGene 2.0 assay (n = 7 biologically independent samples, mean ± s.d.; one-way ANOVA, ****P < 0.0001, ns: not significant). b Ruxolitinib inhibits JAK1 and JAK2 to prevent the pathogenesis of vitiligo driven by IFN-γ signaling. c mRNA expression of IFN-γ-inducible chemokines CXCL9, 10, and 11. Cells were first treated with siRNA or ruxolitinib at 1.5 μM for 72 h, and then stimulated with recombinant human IFN-γ for 24 h (n = 7 biologically independent samples, mean ± s.d.; one-way ANOVA, **P < 0.01, ****P < 0.0001). d, e Chemical engineering of 3033 for skin delivery. Terminal nucleotides were stabilized with phosphorothioate backbone modifications for nuclease stability (d). Human HeLa cells were transfected with DCA-siRNAs for 72 h through lipofectamine RNAiMax-mediated uptake and mRNA levels were measured using the QuantiGene 2.0 assay (n = 3 biologically independent samples, mean ± s.d.; two-way ANOVA multiple comparison, *P < 0.05). f Molecular modeling (PyMOL 2) of DCA-siRNA 3033 in scaffold 2, si3033. g–i ruxolitinib and si3033 both reduce luciferase activity in IFN-γ-JAK1/2-STAT1 HeLa reporter cells (n = 7 biologically independent samples, mean ± s.d.; one-way ANOVA, **P < 0.01, ****P < 0.0001). Cells were treated with ruxolitinib or siRNA for 72 h and then stimulated with IFN-γ for 18 h. si3033 was transfected to the HeLa reporter cells using lipofectamine RNAiMax. j JAK1 and JAK2 mRNA expression in si3033-treated and untreated HeLa reporter cells (n = 7 biologically independent samples, mean ± s.d.; two-sided unpaired t test, ****P < 0.0001, ns: not significant). Source data are provided as a Source Data file.

Fig. 3. si3033 reduces JAK1-mediated T-cell recruitment to the skin and prevents depigmentation in a mouse model of vitiligo.

a Single-dose subcutaneous (S.C.) injection of si3033 at 20 mg/kg supports Jak1 silencing in tail skin for over 5 weeks (n = 5 animals, mean ± s.d.; two-sided unpaired t test, *P < 0.05, **P < 0.01, ***P < 0.001); QuantiGene 2.0 assay and presented as percentage of siNTC control. b Biodistribution of si3033 at skin local to the injection site and in systemic tissues. (n = 5 animals, mean ± s.d.). c Blood chemistry diagnostics at 24 h and 72 h post S.C. injection of 20 mg/kg si3033 (n = 8 animals, mean ± s.d. the current plotting is to support visual clarity, raw data values can be found in the supplied source data file). ALB albumin, ALP alkaline phosphatase, ALT alanine transaminase, AMY amylase, Ca²⁺ calcium, CRE creatinine, GLU glucose, Na⁺ sodium, K⁺ potassium, TP total protein, GLOB globulin, WBC white blood cell, LYM lymphocyte, MON monocyte, NEU neutrophil, RBC red blood cell, PLT platelet, HGB hemoglobin, HCT hematocrit, MCV mean corpuscular volume, MCH mean corpuscular hemoglobin, MCHC mean corpuscular hemoglobin concentration, RDWc red blood cell distribution width coefficient of variation, RDWs red blood cell distribution width standard deviation, MPV mean platelet volume, PCT procalcitonin, PDWc platelet distribution width coefficient of variation, PDWs platelet distribution width standard deviation. d 0.08 mg of si3033 provides Jak1 silencing in the footpad skin (right vs. left pad) for 2 weeks. Mouse Jak1 mRNA were measured over 5 weeks using the QuantiGene 2.0 assay (n = 16 animals, mean ± s.d.; two-sided paired t test, *P < 0.05, **P < 0.01, ***P < 0.001). e si3033 significantly reduces the expression of IFN-γ-inducible chemokines CXCL9 and 10 in an ex vivo skin model of IFN-γ signaling at week 2. IFN-γ signaling was induced using recombinant mouse IFN-γ in a 3-mm skin punch collected at the injection site of footpad. Mouse CXCL9 and CXCL10 were quantified in the ex vivo culture media using ELISA assays (n = 10 animals, mean ± s.d.; lines represent two-sided paired t test in the same mouse, *P < 0.05). f Schematic of a mouse model of vitiligo that mimics human skin depigmentation. g Skin infiltration of autoreactive PMEL CD8⁺ T cells (n = 12 animals, mean ± s.d.; two-sided paired t test, **P < 0.01). h Representative image reveals that si3033 prevents skin depigmentation in footpads. The locally injected area of left footpad exhibited less severity of depigmentation compared to right footpad treated with siNTC control compound. Source data are provided as a Source Data file.

Fig. 4. si3033 efficiently downregulates JAK1-dependent inflammatory signaling in human skin ex vivo.

a Intradermal injection of 200 µM (0.13 mg in 50 µL of PBS) fluorescently (Cy3)-labeled si3033 into human skin explant. b Fluorescent microscope images of cross-sectioned skin. Skin was cultured for 24 h before microscopy. Nuclei were stained with DAPI in cyan, and si3033 was in red. Left to right: image 1: scale bar length 10 mm; image 2 to 4: scale bar length 100 μm. c Relative uptake efficiency of si3033 in skin cell types of human skin (n = 4 biologically independent samples; Min to max values; box edges represent quartiles and mean). 0.013 mg of Cy3-si3033 in 50 µL of PBS (20 µM) was intradermally injected to reduce the excess fluorescence intensity of siRNA for flow cytometry analysis. d JAK1 silencing in whole skin and e in epidermis and dermis separately. Skin biopsies (8 mm) were injected with 0.13 mg of si3033 for 4 days; epidermis and dermis were separated by 1 h of dispase digestion. JAK1 mRNA level was quantified by QuantiGene 2.0 assay (n = 4 biologically independent samples, mean ± s.d.; two-sided unpaired t test, *P < 0.05, ***P < 0.001). f mRNA and g protein levels of IFN-γ-inducible chemokines CXCL9, CXCL10, and CXCL11. Skin biopsies were stimulated with 10 ng/mL of recombinant IFN-γ and 10 ng/mL of TNF-α (for synergy) for 24 h, mRNAs were quantified by QuantiGene 2.0 assay and proteins were measured by enzyme-linked immunosorbent assay (n = 4 biologically independent samples, mean ± s.d.; one-way ANOVA, *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, ns not significant). Source data are provided as a Source Data file.

Fig. 5. si3033 alleviates IFN-γ-induced inflammatory responses in human skin explant.

Human skin biopsies were injected with PBS, 0.13 mg siNTC or si3033 and cultured for 4 days ex vivo (unstimulated; naive: non-injected control); IFN-γ signaling in a separate set of samples was induced with 10 ng/mL of recombinant IFN-γ and 10 ng/mL of TNF-α for 24 h (stimulated). a Volcano plot showing differentially expressed genes with IFN-γ stimulation. PBS and siNTC-treated samples were analyzed together, controlling for treatment status. (Genes colored: blue if FDR ≤ 0.05 and log2FoldChange ≤ −1, red if FDR ≤ 0.05 and log2FoldChange ≥ 1, gray if either FDR > 0.05 or −1<log2FoldChange < 1, n = 4 biologically independent samples per treatment). padj: Benjamini–Hochberg-adjusted P value as calculated by a two-sided Wald Test implemented within DESeq2. b Gene set enrichment analysis for the differential expression analysis in (a). Gene ontology terms were clustered by semantic similarity and nodes were colored by the normalized enrichment score. c Heatmap showing k-means cluster results of genes with a significant interaction between IFN-γ signaling stimulation and si3033 treatment (Likelihood ratio test, FDR ≤ 0.05, n = 4 biologically independent samples per treatment). d Most significant terms from gene ontology analysis of genes in each cluster from k-means clustering in (c). e Enrichment analysis of genes in the Hallmark IFN-γ Response in each cluster from k-means clustering in (c) with Enrichment defined as the number of genes in a cluster belonging to Hallmark IFN-γ Response divided by the number of genes in that cluster not belonging to Hallmark IFN-γ Response. Statistical analysis for the enrichment of Hallmark IFN-γ Response genes in each cluster was conducted using a hypergeometric test in R (phyper, lower.tail=FALSE) with Benjamini–Hochberg correction for multiple comparisons. Statistical analysis to compare enrichment between clusters was performed using a binomial family generalized linear model with a logit link, and multiple test-corrected P values were calculated by a two-sided Tukey HSD test using the emmeans R package.