Hypersensitive interferon responses in lupus keratinocytes reveal key mechanistic determinants in cutaneous lupus

Abstract

Systemic lupus erythematosus (SLE) is a complex autoimmune disease in which 70% of patients experience disfiguring skin inflammation (grouped under the rubric of cutaneous lupus erythematosus, CLE). There are limited treatment options for SLE and no FDA-approved therapies for CLE. Studies have revealed that interferons (IFNs) are important mediators for SLE and CLE, but the mechanisms by which IFNs lead to disease are still poorly understood. We aimed to investigate how IFN responses in SLE keratinocytes contribute to development of CLE. A cohort of 72 RNA-sequencing samples from 14 individuals (7 SLE and 7 healthy controls) was analyzed to study the transcriptomic effects of type I and type II IFNs on SLE vs. control keratinocytes. In-depth analysis of the interferon responses was conducted. Bioinformatics and functional assays were conducted to provide implications for the change of interferon response. A significant hypersensitive response to IFNs was identified in lupus keratinocytes including genes (IFIH1, STAT1, and IRF7) encompassed in SLE susceptibility loci. Binding sites for the transcription factor PITX1 were enriched in genes that exhibit IFN-sensitive responses. PITX1 expression was increased in CLE lesions based on immunohistochemistry and by using siRNA knockdown we illustrated that PITX1 was required for upregulation of IFN-regulated genes in vitro. SLE patients exhibit increased interferon signatures in their skin secondary to increased production and a robust, skewed IFN response that is regulated by PITX1. Targeting of these exaggerated pathways may prove to be beneficial to prevent and treat hyperinflammatory responses in SLE skin.

I. Introduction

Systemic lupus erythematosus (SLE) is a chronic autoimmune disease that results in inflammatory organ damage. SLE has a complex genetic architecture, and it affects up to 0.4% of the population worldwide(1). Significantly, around 70% of SLE patients experience cutaneous manifestations, referred to as cutaneous lupus erythematosus (CLE)(2), and the flares of lesions can lead to significant loss of productivity and quality of life(3, 4). Despite a few commonly used, yet usually insufficient, treatment options for SLE, there is no FDA-approved therapy for CLE.

Previous studies have revealed the importance of even small amounts of interferons for priming cells to maintain homeostasis and respond to immune stimuli (5, 6). Importantly, type I interferons (IFNs) are mediators in the pathogenesis of SLE and CLE, and recent trials blocking type I IFN signaling show efficacy for skin lesions(7). Previous studies have illustrated that type I IFNs are highly up-regulated in CLE lesions(8), and are induced in keratinocytes by UV light(9), an important trigger for CLE flares (10–12). Recently, we identified IFNκ, a keratinocyte-produced type I IFN, to be over expressed in CLE lesions and in non-lesional SLE keratinocytes (13). In addition, IFNα, another type I IFN, is released from plasmacytoid dendritic cells (pDCs) recruited to the dermal-epidermal junction and is detected in CLE lesions after UVB-promoted chemotaxis of pDCs (14). The consequences of chronic type I IFN signaling in the skin are important for immune activation: they can induce chemokine and cytokine production from keratinocytes(9, 15, 16), promote UVB-mediated apoptosis (13), activate the adaptive immune system(17, 18) and promote inflammasome activation in monocytes (19). They also have important functions to promote auto-reactive B cell activation and class switching (20, 21). Although all type I IFNs signal through the type I IFN receptor, not all type I IFNs have identical effects on cells; for instance, IFNβ has been reported to have both pro-survival and anti-infective effects when compared to other type I IFNs (22, 23). Differential responses to type I IFNs have variable effects on cells, but how this relates to changes in keratinocyte function and inflammation in SLE and/or CLE is unknown.

Given the high frequency of skin disease among SLE patients, the prominence of Type I-IFN responses in lesional skin of CLE, and the elevated IFN production in SLE keratinocytes, this study aims to understand pathological mechanisms of CLE through profiling the interferon responses in keratinocytes. Specifically, we evaluated and compared the responses of keratinocytes from normal versus lupus patients upon IFN stimulation by setting up a cohort of 72 RNA-sequencing samples from 14 individuals to study the transcriptomic effects of different cytokine stimulations in keratinocytes derived from normal or lupus patients. Notably, we identified a significant hypersensitive response to IFNs in lupus keratinocytes. We also provided biological implications and evaluated potential regulatory mechanisms for the interferon-responding genes, and highlighted STAT1, IRF7, and IFIH1, which are within lupus-associated loci (24), to be key lupus-sensitive components participating in the pathology of the disease. Importantly, we revealed an enrichment of binding sites for the transcription factor PITX1 in genes that exhibit interferon sensitive responses, and we validated PITX1 expression and its regulator effect on IFN-regulated gene expression. The identification of lupus-sensitive interferon (LSI) response in our study provides implications for understanding the differentiated nature of IFN responses in keratinocytes between normal and SLE patients, and it will facilitate development of appropriately targeted, novel and specific therapies of downstream global interferon inhibition for CLE.

II. Methods

Patient population:

SLE patient keratinocytes were isolated from skin biopsy samples from participants of the Michigan Lupus Cohort under IRB #00066116. All subjects were treated according to the declaration of Helsinki, gave written, informed consent and fulfilled four or more of the American College of Rheumatology criteria for SLE(25). SLE patients were required to have a history of cutaneous disease to be included. Sex- and age-matched control subjects were identified via advertisement. All fresh biopsies were obtained from nonlesional, non-sun-exposed skin on the upper thigh. Cases of DLE and SCLE biopsies for microarray studies and histopathology were identified and acquired from the University of Michigan Pathology Database under Institutional Review Board (IRB) #HUM72843.

Keratinocyte isolation:

Two 6-mm punch skin biopsy samples were taken from each subject and placed in Hank’s balanced salt solution overnight. Keritinocytes were liberated via incubation in basal media (100 mmol/L NaCl, 22.5 mmol/L HEPES, 7.6 mmol/L glucose, 2.25 mmol/L KCl, 0.75 mmol/L Na₂HPO₄) with 0.17% trypsin for 2 hours at 37˚C. Keratinocytes cultured in Epilife (Gibco Thermo-Fisher Scientific, Waltham, MA) with keratinocyte growth supplement and passaged at 60% confluency to avoid differentiation. Media was replaced every 2–3 days. Keratinocyte culture purity was confirmed by morphology. For all experiments, cells were used at passage 3, and patient samples were chosen that had robust growth during the experimental phase.

Keratinocyte treatment, RNA extraction and RNA-Seq:

Keratinocytes from biopsy samples were treated with interferon α2 (1,000 U/mL), interferon β (1,000 U/mL), interferon γ (5 ng/mL), or interferon α6 (1,000 U/mL) for 6 hours at passage 3. Cells were harvested in Tripure (Sigma-Aldrich, St. Louis, MO). RNA was isolated from cell cultures using Qiagen RNeasy kit (Cat # 74136). Libraries for RNA-seq were generated from polyadenylated RNA and sequenced at six libraries per lane on Illumina HiSeq4000 with the assistance of the University of Michigan DNA sequencing core, utilizing single-end read with 50bp in length.

RNAseq transcriptome profiling:

We generated on average 48 million single-ended reads per sample. For each RNA-seq sample, we performed quality control (FastQC v0.11.7) (26) and adapter trimming (Trimmomatic-0.36) (27). We used STAR-2.5.2 (28) to align the reads to the human reference genome (build 37), with genes annotated in GENCODE version 24. We then used HTSeq-0.6.1 (29) to count the number of uniquely mapped reads for each gene (-a 50 -t exon -m union), and on average >80% of reads were uniquely mapped to gene regions. The gene expression level was modeled using a negative binomial distribution and normalized in DESeq2(30). Only genes with on average >=1 read/sample were used in the subsequent analysis. Principal component analysis was performed on DESeq2 normalized expression matrix with additional inverse normalization; COMBAT was used to remove any batch effect (31). The coefficients of biological variation for each condition are illustrated in Supplementary Table 1. We conducted differential expression analysis for each cytokine stimulation, conditioning on the patient specific effect, for keratinocytes derived from normal and lupus patients separately. The significant genes were declared as having False Discovery Rate (FDR)≤10% and |log2 (Fold Change)|≥1. The RNA-seq datasets are available at https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE124939.

Downstream analysis of expression data:

We conducted functional enrichment analysis using the hypergeometric test, and we examined the enrichment for pathways/functions annotated from Gene Ontology (32), Reactome (33), KEGG (34), and Biocarta (35), compiled from the GSEA (36). We investigated the 5,218 functions/pathways with ≥10 and ≤300 annotated genes that are expressed in our dataset. Significant functions were declared as having FDR≤10% and Observed/Expected (O/E) ratio ≥1.5.

To compute the statistical significance for observing X interferon-responding genes with consistent higher response in lupus keratinocyte upon different interferon stimulation, we first computed the probability d_i that an interferon-responding gene would have higher effect in lupus keratinocytes under the stimulation of interferon i. The probability that an interferon-responding gene would have higher effect in lupus keratinocytes for all the different interferons being considered would be: $a={\prod }_i^I{d}_i.$ We could then compute the probability of observing X among all Y interferon-responding genes by: $p={\sum }_{x=X}^Y\left(\begin{array}{c}Y\\ x\end{array}\right)a^x{(1-a)}^{Y-x}$. We then used mixed-effect regression to model the effect size of interferon response: $y=\left(SLE\right){\beta }_1+\mathit{X}{\mathit{\beta }}_2+\mathbf{\text{Pu}}+\mathbf{\in },$ where β represents the fixed effects regression coefficient (for lupus or different interferons), u is the random effect for patient, and ∊ is a vector of random errors. The R library “lmer” was used for the modeling(37). Wilcoxon rank sum test was used to compare the lupus effect values estimated for the interferon-responding genes versus those for the non-responding genes. We applied MEAGA, which uses graphical algorithm to integrate interaction information, to evaluate functional (i.e. LSI response in this case) enrichment among the established lupus loci (38). We used the interaction data from STRING (39) for generating the interactome, and the genetic variants from 1000 genomes as background (40).

We compared the cytokine stimulated effects measured in our gene expression experiments with a previously published, independent microarray dataset measuring global transcriptomic differences in skin biopsies for 47 discoid lupus erythematosus (DLE) and 43 subacute cutaneous lupus erythematosus (SCLE) (GEO GSE81071). We performed the motif enrichment by screening the motif of every transcription factor across the 2k bp upstream region from the transcription start site (41); we then evaluated the enrichment of each motif in the LSI responding genes, by setting the interferon-responding genes as background.

PITX1 immunohistochemistry, knock down, qRT-PCR, and Western blot:

We obtained 5um thick slides from formalin fixed, paraffin embedded skin tissue specimens. PITX1 (Lifespan Biosciences #LS-C100923/55637) staining was performed at 1:50 dilution and at pH6 along with isotype controls. N/TERTs⁽⁴²⁾, an immortalized keratinocyte line, were plated in 96 well plate (30,000 cells/well) and incubated at 37°C with 5% CO2 overnight. 100 μM Accel siRNA (PITX1; Dharmacon # E-017246-00-0005) was prepared in 1x siRNA buffer (Dharmacon# B-002000-UB-100). 1 μl of 100μM siRNA or control siRNA was diluted with 100 μl accel delivery medium (Dharmacon # B-005000) for each well of 96 well plate. Growth medium was removed from the cells and 100 μl of the appropriate delivery mix with siRNA was added to each well and the plate was incubated at 37°C with 5% CO2. Accell Non-targeting Control siRNA (Dharmacon # D-001910-01-05) was used as a negative control. After 48 Hrs, cells were stimulated with 5 ng/ml of each cytokine (IFN-α, R&D Systems # 11100-1; IFN-β, R&D Systems # 8499-IF; IFN-κ, R&D Systems # 11165–1) separately for 24 hours and then harvested for RNA preparation. RNA was isolated from cell cultures using Qiagen RNeasy plus kit (Cat # 74136). Reverse transcription was performed using High Capacity cDNA Transcription kit (ThermoFisher # 4368813). qPCR was performed on a 7900HT Fast Real-time PCR system (ThermoFisher) with TaqMan Universal PCR Master Mix (ThemoFisher # 4304437) using TaqMan primers (ThermoFisher Scientific; PITX1: Hs00267528_m1, MX1: Hs00895608_m1). RPLP0 (ThermoFisher # Hs99999902_m1) was used as a loading control as it is not regulated by interferon stimulation. For Western blot, protein was harvested 72 hours after siRNA treatment in RIPA buffer with protease inhibitors. Total protein was measured using BCA Assay. Protein from each sample was separated on 10% acrylamide gel and transferred to the Amersham™ Protran™ 0.2μM NC nitrocellulose membrane (GE Healthcare). The membranes were blocked with 5% milk and the incubated with rabbit-anti human PITX1(1:2500 dilution in PBS+5%BSA) at 4^oC overnight followed by HRP-conjugated anti-rabbit IgG (both of the antibodies were from Abcam, Cambridge, MA). The same membrane was also blotted with anti-human β-actin (1:1000 dilution, purchased from Cell Signaling, Danvers, MA) following the same procedure. Protein expression bands were detected by chemiluminescence using WesternBright™ Quantum Western blot detection reagent (Advansta, Menlo Park, CA), and the protein bands were imaged by Omega Lum C (Gel Company, San Francisco, CA).

III. Results

Shared and distinct responses to different interferons in keratinocytes

We obtained and cultured keratinocytes derived from skin biopsies of 7 normal and 7 lupus patients from non-lesional skin (Supplementary Table 2). We stimulated the keratinocytes using different Type I (i.e. IFNα2, IFNβ, IFNα6) and Type II (IFNγ) interferons and/or IL-18, a non-Jak-STAT signaling cytokine with implications in CLE(43). RNA-sequencing (RNA-seq) was conducted to profile the expression of unstimulated and cytokine-stimulated conditions, and in total we assayed 72 distinct transcriptomes (Supplementary Table 3) and detected 27,574 genes being expressed. We first investigated the transcriptomic differences between the various conditions and skin origin by principal component (PC) analysis. Interestingly, we did not observe systematic differences between the keratinocytes derived from normal versus lupus skin under the unstimulated (or stimulated) condition when using only the top two principal components; however, the analysis illustrated a strong contrast between the Type I interferon stimulations and unstimulated or other-stimulated conditions, including IFNγ (Figure 1a). Notably, the response under IL-18 and IFNα co-stimulation is more similar to the other Type I Interferon responses than IFNγ stimulation.

Figure 1. Transcriptomic analysis on keratinocytes upon IFN stimulations.

a) Principal component analysis (top 2 PCs are shown) illustrating the pronounced transcriptomic changes during Type I IFN stimulations when comparing against other cytokine stimulations; b,c) large overlap between genes being differentially expressed by IFNα2 (b) and IFNγ (c) stimulations in normal and lupus keratinocytes; c) Venn diagram illustrating the overlap between differentially expressed genes in different IFN stimulations in keratinocytes derived from normal (d) and lupus (e) patients; f) functional enrichment test results (in -log10(FDR)) for genes differentially expressed only in IFNγ (x-axis) or only in Type I IFN (y-axis). Significant functions are colored in red (only significant in IFNγ), blue (only significant in Type I IFN), or green (significant in both type I and II IFNs)

Next, we conducted differential analysis to compare the expression profiles between the unstimulated condition and each of the cytokine-stimulated conditions, in normal or lupus keratinocytes separately (Supplementary Table 4). We observed >400 differentially expressed genes (DEGs; False Discovery Rate, FDR≤10% and |log₂Fold Change|≥1) in each of the IFN stimulations for both control and lupus keratinocytes; in contrast, the IL-18 stimulation, which served as a non-IFN control, did not result in any significant transcriptional changes. In concordance with the PC analysis, the genes differentially expressed (DE) upon IFN stimulations in normal keratinocytes also tended to be DE in lupus keratinocytes (Supplementary Figure 1; Figure 1b,c). While there was substantial overlap between the DEGs under different IFN stimulations either in normal (Figure 1d) or lupus (Figure 1e) keratinocytes (271 and 279 genes differentially expressed under all four interferons in normal and lupus keratinocytes, respectively), we consistently observed a large number of genes differentially expressed only by Type I Interferon but not IFNγ (368 in normal and 418 in lupus keratinocytes), which highlighted the shared and unique downstream signaling cascades for the type I and II Interferon stimulations on keratinocytes (Figure 1d,e). In addition, we identified 67 and 112 functions/pathways being significantly enriched (FDR≤10% and Observed/Expected, O/E ratio≥1.5) among genes that were only regulated by Type I or Type II interferons, respectively, of which only 16 overlapped (Supplementary Table 5; Figure 1f). The distinct sets of functions for Type I Interferon include MHC class I protein (p=7.1×10⁻⁹ and O/E=30.2) and regulation of NFkB activity (p=2.3×10⁻⁷ and O/E=5.8); for type II interferon, the functions included G-protein coupled receptor ligand binding (p=1.1×10⁻⁶ and O/E=5.5) and MHC class II proteins (p=6.6×10⁻¹⁰ and O/E=47.8). While genes regulated only by IFNα are not enriched with particular biological functions/pathways, interestingly, IFNβ-only regulated genes are enriched with 17 functions such as epithelium elongation (p=2.3×10⁻⁶ and O/E=18.8) and extracellular matrix regulation (p=8.5×10⁻⁵ and O/E=3.6).

Lupus-sensitive interferon (LSI) responses

We observed consistently higher numbers of DEGs in keratinocytes from lupus patients upon different IFN stimulations, when compared with the same stimulations in normal keratinocytes (Supplementary Figure 1). Furthermore, the IFN-driven gene-regulation tended to be more statistically significant in lupus keratinocytes (Supplementary Figure 2), even with the same number of samples as normal keratinocytes in the differential expression comparison (Supplementary Table 3). We thus hypothesized that there are differential IFN responses between keratinocytes derived from lupus patient versus healthy individuals. We first compared the effect sizes (i.e. fold changes, FCs) of cytokine stimulation under each keratinocyte type (Figure 2). Interestingly, for genes that were induced (i.e. up-regulated) by IFNs, they had a stronger effect in keratinocytes from lupus patients; the increase in effect size was consistently observed across different IFN stimulations (both type I and type II), and not observed under IL-18 stimulation. Indeed, when evaluating the most significant DEGs under the IFN stimulations, the distributions of the effect size differences between normal and lupus keratinocytes tended to be right skewed (Figure 3a; Supplementary Figure 3), indicating a stronger IFN stimulation effect for lupus patients. We were also able to illustrate that the stronger the “baseline” effect after IFN stimulation in normal keratinocytes, the larger the magnitude of the change in effect size in lupus keratinocytes (Figure 3b and Supplementary Figure 4).

Figure 2. The Effect sizes (i.e. log₂FC) under different cytokine stimulations in normal (x-axis) and lupus (y-axis) keratinocytes.

Diagonal line represents what would be expected if the effect sizes are correlated between the two keratinocyte types.

Figure 3. The identification of genes with lupus sensitive interferon (LSI) responses.

(a) the changes in effect sizes between normal and lupus keratinocytes upon IFNα2 stimulation; (b) associations between different effect size cutoffs upon IFNα2 in normal keratinocytes and the magnitude of the change in effect size in lupus keratinocytes; (c) the changes in effect sizes upon general interferon in lupus keratinocytes (when comparing against normal keratinocytes) for all genes and cytokine-regulated genes.

We next sought to provide quantitative assessment of the above observations and implemented statistical models to evaluate lupus specific effects for genes regulated by type I and type II interferons. We studied two conditions: i) all the type I IFN stimulations and ii) all the general IFN (type I and type II) stimulations. There were 579 and 233 genes differentially expressed in both normal and lupus keratinocytes under the two conditions, respectively; 276 and 121 genes among them had a consistently higher magnitude in lupus keratinocytes under each of the interferon stimulations (Supplementary Figure 5), this consistency was highly significant (p=5.7×10−³³ and 7.1×10−³² for type I and all IFNs, respectively; see Methods). We then used mixed-effects regression to model the effect sizes of the IFN stimulation to estimate the significance of the lupus effect after controlling for the different patient and interferon effects. While individually no gene was significant by itself, the estimated effect size of lupus keratinocytes for interferon-response genes as a group was significantly higher than the other expressing transcripts (p=1.2×10−¹² and 1.3×10−¹⁸ for type I and all IFNs, respectively; Figure 3c; see Methods), concordant with the above results. These results illustrate the hypersensitive interferon responses in keratinocytes for lupus patients. We thus deemed these 119 interferon-response genes with higher effect in SLE vs. normal keratinocytes (from both random effect model and differential expression analysis for every IFN stimulation) as lupus-sensitive interferon (LSI) responses (Supplementary Table 6; Figure 4; examples in Table 1). There are 259 LSI response genes if we only consider Type I interferons using the same definition.

Figure 4. LSI genes having higher interferon responses in lupus patients.

(a) heatmap illustrates the increase in log₂ fold change (darker the color, higher the increase in fold change) upon different IFN stimulations in keratinocytes from lupus patients comparing with those from healthy individuals for the 119 LSI genes; (b) the log₂ fold changes for three LSI genes from lupus susceptibility regions illustrate keratinocytes from lupus patients have modestly elevated values.

Table 1.

Effect size and statistical significance level for notable genes having higher IFN responses in keratinocytes from lupus patients.

	IFNα stimulation				IFNβ stimulation				IFNγ stimulation
	normal		lupus		normal		lupus		normal		lupus
	log2(FC)	p-value	log2(FC)	p-value	log2(FC)	p-value	log2(FC)	p-value	log2(FC)	p-value	log2(FC)	p-value
IDO1	5.96	8.78E−10	7.80	1.14E−18	7.36	6.11E−29	8.51	2.12E−11	7.78	2.85E−08	9.89	1.09E−03
GBP7	5.43	3.59E−09	6.82	5.89E−16	5.39	7.76E−08	7.07	3.55E−12	5.88	2.57E−03	7.60	1.67E−04
IFI44	5.77	3.53E−14	6.60	2.93E−27	5.34	5.45E−13	7.19	2.43E−32	1.39	5.14E−03	2.49	2.46E−04
BATF2	7.01	2.72E−45	7.84	3.79E−64	8.05	8.66E−83	8.84	1.42E−54	6.47	5.31E−12	7.59	2.18E−10
IRF7	4.18	1.13E−23	4.56	1.70E−55	4.27	2.97E−18	4.75	2.32E−45	1.00	1.34E−03	1.20	5.13E−05

Pathologic implications for LSI responses

We then aimed to characterize the LSI responses, and attempted to provide pathologic implications. We first investigated the functions that are enriched among these genes. Although it is expected that the LSI responding genes are enriched among functions/pathways for interferon signaling (Supplementary Table 5), we identified functions related to the regulation of DDX58/IFIH1 signaling that were significant (e.g. negative regulation of DDX58/IFIH1 signaling: p<2.3×10⁻⁸) only among the LSI genes, but not significant in the functional analysis for genes only regulated by Type I or Type II interferon from the above analysis. Since IFIH1 is within one of the lupus -associated loci (24), we next investigated the overlap between the genes with LSI responses and the lupus susceptibility regions using a systems biology approach, MEAGA (38). By considering the functional enrichment and the coherence in gene-gene interactions, we found that genes within the +−100kb intervals of the lupus-associated regions were enriched in LSI response (p=3.7×10⁻²). Specifically, IFIH1, STAT1, and IRF7 from three lupus susceptibility loci exhibited LSI responses (Figure 4), and they also interacted with other LSI responding genes including IRF9, DDX58, and ISG15. These results provide prioritization for the candidate genes from the lupus associated loci; in addition, genes exhibiting the LSI responses for functional analysis provide insight on the possible relevance of IFIH1/DDX58 signaling cascade may contribute to the pathology of CLE. To validate the relevance of LSI genes in disease, we then performed a direct comparison against an independent transcriptomic cohort of cutaneous lupus comparing healthy control versus CLE lesional tissue. Figure 5 illustrates that the genes regulated by our interferon experiments overlap with genes that are dysregulated in the subacute cutaneous erythematosus (SCLE) tissue, and we also observed a modest correlation between the effect sizes in the lesional skin and the interferon stimulation in lupus vs. normal keratinocytes, thus concurring with our above findings (Supplementary Figure 6 shows similar results for discoid lupus erythematosus, DLE). Notably, we found significant enrichment of our LSI responding genes among the dysregulated genes in SCLE tissue, where 51 (p=1.5×10⁻⁴²) and 40 (p=9.1×10⁻⁴³) out of 258 SCLE dysregulated genes exhibit LSI response for Type I interferons and all interferons, respectively, including IFIH1 and STAT1. This analysis provided translational implication for our in vitro and in silico findings.

Figure 5. Comparing interferon response with dysregulation in lupus skin.

The changes in effect sizes upon interferon stimulation (y-axis) in normal (left) and lupus (right) keratinocytes against dysregulated genes in SCLE (x-axis). Red color represents genes differentially expressed in both conditions; orange color represents genes differentially expressed only in lupus skin; blue color represents genes differentially expressed only under interferon response.

To further understand the regulatory mechanism of these LSI responses, we screened the promoter regions of LSI genes to reveal transcription factor binding sites that were enriched, using all the IFN-regulated genes’ promoter regions as background. Notably, we identified the binding motif of PITX1, a homeoprotein transcription repressor, to be significantly enriched (p=7.2×10⁻⁵; Figure 6a); PITX1 has been shown to physically interact with IRF3 and IRF7 on IFNA promoters (44), but its function in regulation of IFN response genes is unknown. Immunohistochemical analysis of normal and CLE skin identified PITX1 as expressed in the epidermis and globally elevated in CLE lesions (Figure 6b). To determine whether PITX1 has a regulatory role in type I IFN-mediated transcriptional changes, we evaluated whether downregulation of PITX1 expression was sufficient to interrupt type I IFN-induced expression of MX1. As shown in Figure 6c–e, baseline and type I IFN-mediated expression of MX1, a known type I IFN-regulated gene (and importantly, also a LSI gene), was significantly repressed when expression of PITX1 was inhibited via siRNA. Interestingly, the inhibition of PITX1 also has inhibiting effect on IRF3/IRF7/OASL expression (Supplementary Figure 7). These results implicate PITX1 as an important regulator of interferon response, including LSI genes, and a potential target for further investigation.

Figure 6. PITX1 is a regulator for LSI response.

(a) p-value enrichment for transcription factor binding sites among LSI responding genes; (b) immunohistochemical analysis of normal and CLE skin for PITX1; (c-e) Knock-down of PITX1 was achieved through siRNA in N/TERT keratinocytes (c), and it significantly repressed expression of MX1 upon type I IFN stimulation. (e) Western blot showing efficient knockdown of PITX1 protein expression. *<0.05; **<0.005; ***<0.0005

IV. Discussion

This work has identified a novel mechanism by which skin from lupus patients is prone to inflammation. Not only does lupus skin, even in non-lesional tissue, produce increased type I IFNs (likely a mix of IFNα and IFNκ (9, 45)), we now also reveal that lupus keratinocytes are primed for hypersensitive responses to both type I and type II interferons and that the LSI genes link to genetic risk for SLE and are identified as over-expressed in cutaneous lupus lesions, giving credence to their pathologic relevance. Further, we have identified upstream regulators, including PITX1,that may serve as important targets for downregulation of LSI responses.

Lupus is a complex autoimmune disease, and previous genetic studies (24, 46) have revealed multiple susceptibility regions associated with the disease. However, similar to other complex conditions, most of these signals reside within non-coding regions and affect the regulation of nearby genes in a context-specific pattern (47). While the most promising candidate cell types for the complex autoimmune conditions are immune cells (47), we and others have identified a strong interferon signature in lesional and non-lesional lupus keratinocytes (9, 45, 48). We thus hypothesized that the keratinocytes in lupus patients would exhibit an enhanced response to interferons. Previous SLE genomewide association studies also revealed that most signals have modest effect size, and therefore it is not too surprising that the subtle elevated lupus-specific interferon response would not be revealed for each individual gene in our dataset; however, using interferon-responding genes as a unit we identified a significant lupus-specific effect (p<5×10⁻¹²).

After revealing lupus-specific interferon responses, we integrated computational approach and in vitro validation to identify PITX1 as an intriguing candidate in the interferon response. PITX1 is able to bind to IRF3 and IRF7, which results in negative regulation of various IFN genes (44). Importantly, interactions with IRF3 and IRF7 can overcome the repressive activity of PITX1 (44), suggesting that in chronic IFN high states where IRF3 and IRF7 levels are elevated, such as cutaneous lupus, PITX1 may be an important switch to skew increased transcription of IFN-regulated genes. Intriguingly, PITX1 is required for IFN-driven expression of IRF3 and IRF7, which suggests that it may also be important for enhanced IFN priming in SLE keratinocytes. Importantly, the role of PITX1 in regulating key upstream signaling and transcriptional mediators involved in type I IFN pathways, including IFNκ production in keratincocytes, should also be investigated. Among other transcription factors with their binding sites enriched among the promoters of the LSI genes (Figure 6a), it is also noteworthy to know that FOXO3 can modulate pro-inflammatory cytokine production and serve as IKK-ε regulated check-point of interferon regulatory factor (49), and GSX2 is within a genomic region outside MHC showing strongest association with lupus nephritis among systemic lupus erythematosus patients (50).

We also identified unique keratinocyte responses for each type I and type II IFN stimulation in normal and SLE keratinocytes. All type I IFNs, including IFNα and IFNβ, signal through a common type I IFN receptor (IFNAR), yet it remains unknown why the molecular responses of these unique interferons differ. IFNβ has been identified to have a higher affinity for IFNAR1 which may promote prolonged signaling despite negative regulatory mechanisms (51). Other explanations such as degradation vs. recycling of the receptor have also been explored (52). However, the contribution of downstream signaling molecules in differential IFN responses is unknown. Certainly it will be of interest to identify whether regulators of LSI responses participate in differential IFN responses as well.

Previous studies have illustrated the importance of low levels of type I IFN for maintaining immune responses (6), and increased levels can drive chronic IFN priming effect in lupus patients (13). In this study, we have also conducted differential expression analysis to compare the untreated keratinocytes from healthy individuals versus the keratinocytes from the lupus patients (Supplementary Table 7). Despite fewer significantly differential expressed genes in the normal versus SLE keratinocytes comparison, we did observe significant positive correlations (Supplementary Figure 8) against the effect sizes upon IFN stimulation, indicating a modest priming effect in the keratinocytes among SLE patients. In contrast, a previous study using in vivo data (comparing 240 lupus nephritis (LN) patients versus 89 controls) illustrates keratinocytes (not cultured) from skin biopsies in LN patients have elevated IFN response signature (45), and this result can be contributed by both intrinsic effect or previous exposure to IFN. As we cultured the keratinocytes from the skin biopsies of normal and lupus patients for our study prior to RNA sequencing, this may explain the reason for more subtle priming effects in our study.

In summary, cutaneous manifestations of SLE affect a majority of the lupus patients, and a better understanding of its pathology is essential to achieve effective treatment of patients. Using RNA-seq and bioinformatics approaches, we have confirmed a skewed IFN response in keratinocytes from lupus patients and have identified PITX1 as a potential targetable regulator of this response. Continued work to understand the LSI responses in the skin will lead to novel and better therapies for SLE and potentially other skin diseases characterized by dysregulated IFN responses.

Abbreviations

CLE:

cutaneous lupus erythematosus

DE:

differentially expressed

DEGs:

differentially expressed genes

DLE:

discoid lupus erythematosus

FDR:

false discovery rate

LSI response:

lupus-sensitive interferon response

O/E:

observed/expected

pDCs:

plasmacytoid dendritic cells

SCLE:

subacute cutaneous lupus erythematosus

SLE:

Systemic lupus erythematosus