Ruxolitinib Improves Immune-Dysregulation Features but not Epigenetic Abnormality in a Patient with STAT1 GOF

This research was reviewed and approved by the institutional review board of Samsung Medical Center (IRB No. 2018–07-140). Informed consent was obtained from the study participant.

Peripheral blood mononuclear cells (PBMCs) were separated by density gradient centrifugation using lymphocyte separation medium (Corning), and cryopreserved in fetal bovine serum (FBS; Corning) with 10% DMSO (Sigma-Aldrich).

Serum from the patient and healthy donors assessed for CXCL10 concentration using the human Recombinant Human CXCL10/IP-10 Protein ELISA kit (R&D) according to the manufacturer’s instructions.

Cryopreserved PBMCs were thawed and subjected to staining with fluorochrome-conjugated antibodies against surface markers for 10 min at room temperature. Dead cells were excluded from analysis using LIVE/DEAD red/aqua fluorescent reactive dye (Invitrogen, Waltham, MA, USA). Multicolor flow cytometry was carried out using an LSR II instrument (BD Biosciences, Franklin Lakes, NJ, USA), and the data were analyzed using FlowJo software (BD Biosciences).

PBMCs were labeled with antibodies targeting surface markers. Subsequently, cells were exposed to IFN-β (50 ng/mL) or PBS for 10 min at 37 °C. This was followed by treatment with IC fixation buffer (Invitrogen) for 10 min at room temperature and fixation using 100% cold methanol for 40 min at 4 °C. Afterward, cells were stained with anti-human STAT1 antibodies for 1 h at room temperature before undergoing flow cytometry analysis.

Single-cell CITE-seq and ATAC-seq libraries were generated using the Chromium Next GEM Automated Single Cell 3′ Library and Gel Bead Kit v3.1 (PN-1000147), 3ʹ Feature Barcode Kit (PN-1000262) and Chromium Next GEM Single Cell ATAC Kit v2 (PN-1000406), following the manufacturer’s instructions. Libraries were constructed and sequenced at a depth of approximately 20,000 reads per cell for gene, antibody-derived tag (ADT), and ATAC-seq fragment. The sequenced data were demultiplexed using the mkfastq function (Cell Ranger, 10X Genomics, v3.0.1) to generate fastq files. Demultiplexed fastq files of gene and ADT expression were aligned to the reference human genome (GRCh38; 10X Cell Ranger reference GRCh38 v3.0.0) and the barcode sequence of each ADT. Filtering, barcode counting, and UMI counting were performed to generate feature-barcode matrices using the count function (Cell Ranger, v3.0.1).

The feature-barcode matrices of RNA and ADT expression were analyzed as follows, using the Seurat R package (Seurat, v4.1.1) [13]. For basic quality control, we de-convoluted sample identity and filtered inter-individual multiplets using the demuxlet package [14]. Next, we filtered low-quality cells expressing mitochondrial genes in > 15% of their total gene expression, or in > 5,000 genes. We also excluded doublets that initially clustered with doublets annotated through the demuxlet package.

Subsequently, the sctransform method (SCTransform function from sctransform package, v0.3.3) [15] was used to perform normalization and variance stabilization of the RNA read count matrix with highly variable genes (n = 2,000), and of the feature-barcode read count matrix with the whole list of feature barcodes. Principal component analysis (PCA; RunPCA function) was carried out to dimensional reduction for each transformed data matrix. Next, the cells underwent unsupervised clustering according to the weighted nearest neighbor (WNN) graph—applying the FindMultiModalNeighbors function, using the top 20 principal components (PCs) from RNA, and 20 PCs from feature barcode, along with the FindClusters function (resolution = 0.4). The results were visualized by Uniform Manifold Approximation and Projection (UMAP) using the top 20 PCs from RNA for whole PBMCs. For sub-clustering analysis of the monocyte immune subset, the cells underwent unbiased clustering, performed using the top 25 PCs from RNA and 15 PCs (resolution = 0.3) from feature barcode, and the results were visualized by UMAP using the top 20 PCs from RNA. To identify marker genes, differentially expressed genes (DEGs) in each cluster relative to the other clusters were selected based on the Wilcoxon rank sum test, using the FindAllMarkers function (parameter; log fold change compared to the other clusters > 0.25, > 0.6 min.pct2 [minimum fraction of test genes detected in cells of other clusters], and Bonferroni-adjusted P < 0.05).

To describe the characteristic of each subcluster, we performed gene set enrichment analysis by calculating the gene set module score (AddModuleScore in Seurat package), combined score (enrichR), [16] and enrichment score (GSEA v4.2.3, Broad Institute, CA, USA) using publicly available gene sets, including Gene Ontology: Biological process databases (GO.BP) [17] and LINCS L1000 data [18]. We filtered Gene Ontology (GO): Biological Process (BP) terms related to immune responses using the following inclusion criteria: "T_CELL", "IMMUNE", "INNATE", "ADAPTIVE", "INFLAM", "INTERLEUKIN", "INTERFERON", "NECROSIS", "APOPTOSIS", "SENESCENCE", "NATURAL", "LYMPHOCYTE", "LEUKOCYTE", "TRANSFORMING", "CHEMO", "CYTOTO", "CYTOKINE", "ANTIGEN", and "SIGNALING" and TRRUST TF. To investigate the dynamic changes of the monocyte immune subset, we exported cells from the monocyte immune subset for monocle’s standard analysis process (monocle v2.22.0) [19]. CellDataSet objects were built based on the normalized count (SCTransform), and then processed using the estimateSizeFactor and estimate-Dispersions functions (default option), detectGenes (default option), setOrdering-Filter and reduceDimension (default option), orderCells (default option), and plot_cell_tra-jectory (default option).

The peak-cell matrices were analyzed as follows, using the Signac R package (Signac, v1.8.0) [20]. For basic quality control, we calculated the nucleosome binding pattern, transcriptional start site (TSS) enrichment score that represents the accessibility signal over all TSS sites, and the ratio of reads in genomic blacklist regions. Next, we filtered low-quality cells using the ratio of mononucleosomal to nucleosome-free fragments (> 3), TSS enrichment (< 2), and ratio of reads in genomic blacklist regions (> 0.05). On the filtered matrix, we performed normalization and linear dimensional reduction using term frequency-inverse document frequency (TF-IDF) normalization, followed by singular value decomposition (SVD) using the top 25% of all peaks. Through this process, we calculated the latent semantic indexing (LSI) component (RunTFIDF, FindTopFeatures, and RunSVD function) [21]. To integrate and correct the batch effect across each sample, we found integration anchors and integrated the datasets using reciprocal LSI projection. We calculated the gene activity score, which predicts and quantifies the activity of each gene in the genome, by assessing the chromatin accessibility associated with each gene, using the GeneActivity function.

Collected PBMCs from healthy donors and a patient with STAT1 GOF mutation were cross-linked with 1% formaldehyde for 10 min and quenched with 125 mM glycine for 5 min at room temperature. The cross-linked PBMCs were lysed with 1% SDS buffer (1% SDS, 10 mM EDTA, 50 mM Tris pH 8.1) and chromatin was fractionated by sonication (Covaris, S220). The lysate was diluted with dilution buffer and incubated with pre-coated Dynabeads Protein A (Thermo, 10001D) with H3K27ac antibody (Active Motif, 39,133) and IgG as negative control overnight at 4ºC with rotation. A portion of diluted lysate was stored to quantify the total amount of DNA in each sample as input control. The immunoprecipitated chromatin was washed four times with RIPA buffer (140 mM NaCl, 1 mM EDTA, 0.5 Mm EGTA, 1% Triton X-100, 0.1% SDS, 0.1% sodium deoxycholate, 10 mM Tris pH 8.0) and incubated with RNase A (Thermo, EN0531) and proteinase K (NEB, P8107S) overnight at 65 ºC to reverse crosslink. The immunoprecipitated DNA was purified by AMPure XP beads (Beckman, A63881) and analyzed by Real-Time PCR using ChIP-qPCR primers (STAT1: FW – ACTCTGCGCAGGAAAGCGAA, RV – GGAACAGCCGCGTCTAATTG; IRF1: FW- AGCCAAGGAATTGCTCCAGT, RV – ACAGGGGGCTGAGTTCTCTT).

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Eui-Cheol Shin (ecshin@kaist.ac.kr).

CITE-seq data and ATAC-seq will be deposited at the NCBI GEO (https://​www.​ncbi.​nlm.​nih.​gov/​geo/​). This study did not generate any original code. Links to the publicly available R packages and codes used in the paper are listed in the key resources table. Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

We previously reported the case of a 23-year-old woman with a heterozygous STAT1 GOF variant (NM_007315.3:c.800C > T, p.Ala267Val), which was a known pathogenic variant, who presented with bronchiectasis and recurrent respiratory infections (Supplementary Fig. 1A and B) [22]. The variant has previously been reported to cause both delayed dephosphorylation of phosphorylated STAT1 (pSTAT1) [23] and increased STAT1 expression levels [24, 25]. We confirmed that the STAT1 expression levels were increased in PBMCs or monocytes from the patient compared to in healthy donors (Supplementary Fig. 1C). She exhibited autoimmune symptoms, with severe Raynaud’s phenomenon in her hands and feet and tenosynovitis in multiple joints (Fig. 1A and B) [26]. Her rheumatoid factor level was elevated to 84.3 IU/mL (reference range, < 14 IU/mL), and she was weakly positive for fluorescent antinuclear antibody intermittently. Hydroxychloroquine treatment was administered for 22 months, but her autoimmune symptoms persisted. We next initiated treatment with the oral JAK1/2 inhibitor ruxolitinib (10 mg per dose twice a day; body weight, 33 kg), which yielded dramatic disappearance of the autoimmune symptoms. Notably, Raynaud’s phenomenon did not develop for seven months after starting ruxolitinib treatment.

Peripheral blood mononuclear cells (PBMCs) were longitudinally collected from the patient before and after ruxolitinib treatment. PBMCs were also obtained from four healthy donors. We used the Chromium system (10 × Genomics) to perform CITE-seq, which allows simultaneous profiling of protein markers and RNA, and ATAC-seq (Fig. 1C). First, we performed pseudo-bulk analysis, using CITE-seq data to identify transcriptional features of whole PBMCs from the patient with STAT1 GOF before and after ruxolitinib treatment. We identified the differentially expressed genes (DEGs) in the patient’s PBMCs before treatment (day 0) compared to the healthy controls’ PBMCs (Fig. 1D and Supplementary Fig. 2A-C). The pre-treatment DEGs featured IFN-stimulated genes (ISGs) (e.g., STAT1, ISG15, MX1, and OAS1) and immune activation genes (e.g., NFKBIA and JUNB), and the expression levels of these genes substantially decreased after ruxolitinib treatment (Fig. 1E). Next, we performed gene set enrichment analysis (GSEA) using the Gene Ontology Biological Process and TRRUST TF databases. The pre-treatment DEGs were significantly associated with type I and type II IFN pathway-related gene sets and transcription factors (e.g., STAT1 and IRF1), and the combined scores of these gene sets decreased after ruxolitinib treatment as early as 3 days (Fig. 1F and G). We also calculated differentially expressed proteins (DEPs) in our patient at day 0 compared to in healthy donors, to identify cell surface proteins that were upregulated in the patient with STAT1 GOF. At day 0, we found significant upregulation of chemokine receptors, such as CX3CR1 and CD183 (CXCR3), and co-stimulatory molecules, such as CD80 and CD86. These proteins were downregulated in the patient after JAKi treatment. (Supplementary Fig. 3A and B). We also performed ELISA to determine the serum levels of CXCL10, a representative chemokine that is inducible by type I or II IFNs. Compared to healthy donors, our patient with STAT1 GOF exhibited a marked increase of CXCL10 transcript and CXCL10 serum levels, which gradually decreased after JAKi treatment. (Supplementary Fig. 3C and D). Collectively, the PBMCs from a patient with STAT1 GOF exhibited aberrantly activated inflammatory features, especially IFN-related features, and these aberrant features were effectively down-modulated by ruxolitinib treatment.

Next, we performed cell clustering, and analyzed each immune cell cluster using the WNN method, which enables the simultaneous consideration of both surface protein expression and transcriptional expression [27]. According to the expressions of genes and markers stained by antibody-derived tags (ADTs), we identified the following nine immune cell types (Fig. 2A): granulocytes, cDCs, monocytes, NK cells, innate-like T cells, CD8 T cells, CD4 T cells, B cells, and proliferating cells (Fig. 2B). We examined which immune cell types mainly exhibited aberrantly activated inflammatory features before treatment (day 0). To this end, we calculated the gene set module scores of each immune cell type, using the DEGs of PBMCs from the pre-treatment time-point, and found that the monocyte population showed the highest enrichment of the DEGs from pre-treatment PBMCs (Fig. 2C).

We subsequently focused on the dynamic transcriptomic changes of the monocyte population after ruxolitinib treatment. In pseudotime analysis, we identified a pseudotime trajectory with monocytes from pre-treatment and day 3 aligned at an earlier pseudotime, and monocytes from healthy controls and the other post-treatment groups (days 14, 28, 64, and 92) aligned at a later pseudotime (Fig. 2D). These findings validated ruxolitinib-induced transcriptomic changes among monocytes. The DEGs of pre-treatment monocytes featured genes related to IFN response (STAT1, IRF1, and ISG15), monocyte activation and differentiation (S100A8, S100A9, IL1B, and FCGR3A), cell cycle and apoptosis regulation (CDKN1C and BCL2A1), and tissue infiltration (FCN1 and VCAN). The expressions of these genes substantially decreased after ruxolitinib treatment (Fig. 2E and F).

We further analyzed the expressions of two different sets of ISGs. Type I IFN stimulation deliver signals via phosphorylated ISGF3 (p-ISGF3) formed by p-STAT1, p-STAT2, and IRF9 [28]. On the other hand, prolonged type I IFN stimulation leads to STAT1 upregulation, which can form ISGF3 even without phosphorylation, so-called unphosphorylated ISGF3 (u-ISGF3) [29, 30]. Downstream genes of p-ISGF3 and u-ISGF3 are known as p-ISGs (ISGs uniquely induced by p-ISGF3) and u-ISGs (ISGs commonly induced by p-ISGF3 and u-ISGF3), respectively [29‐31]. Based on this background, we examined which IFN pathway was more associated with aberrantly activated IFN-related features. In pre-treatment monocytes, we detected significantly enrichment of a gene set of u-ISGs (Fig. 3A). In contrast, after ruxolitinib treatment, a gene set of u-ISGs was substantially down-regulated. Notably, STAT1 itself is part of a gene set of u-ISGs, and STAT1 expression showed the same pattern observed for u-ISG expression (Fig. 3B and C). Next, we evaluated the expression levels of u-ISGF3 components in pre-treatment monocytes. We first examined the transcript levels of STAT1, STAT2, and IRF9, and found that monocytes from the patient exhibited upregulation of all three on day 0 (before JAKi treatment) compared to healthy donors (Supplementary Fig. 4A). We also examined the STAT1 protein levels using flow cytometry analysis. These results confirmed that the STAT1 protein levels were much higher in monocytes from the patient on day 0, compared to in healthy donors. Upon ex vivo IFN-β treatment, the STAT1 protein levels were greatly increased at 3 h, and then returned to the baseline level at 12 h, which was still higher than the levels in healthy donors (Supplementary Fig. 4B and C). These data indicated that STAT1 GOF led to persistent STAT1 upregulation, resulting in aberrant IFN features and that ruxolitinib treatment could successfully decrese the aberrant IFN-features.

Finally, we used ATAC-seq data to examine gene accessibility in before (day 0) and after (day 92) ruxolitinib treatment. Gene accessibility analysis provides important information regarding cellular differentiation and transition, because certain genes are not transcribed despite being highly accessible. In Fig. 4A, we present the gene expression levels of representative u-ISGs (including STAT1, IRF1, MX1, and OAS1) from PBMCs obtained during the course of treatment. The expression levels of STAT1, IRF1, MX1, and OAS1 were dramatically reduced by ruxolitinib treatment (Fig. 4A); however, the gene accessibility scores of these genes were not decreased even at 92 days after starting ruxolitinib treatment (Fig. 4B). Coverage plots also demonstrated highly open chromatin regions near STAT1, IRF1, MX1, and OAS1 in PBMCs of the patient with STAT1 GOF, both before (day 0) and after (day 92) ruxolitinib treatment (Fig. 4C). Furthermore, we used chromatin immunoprecipitation-qPCR (Chip-qPCR) to assess the differential occupancy of H3K27ac in target sites. H3K27ac is an epigenetic modification of histone H3, which is associated with enhanced transcription activation, and is thus defined as an active enhancer. Compared to healthy donors, PBMCs from the patient with STAT1 GOF at day 0 exhibited a higher signal intensity of H3K27ac in both the STAT1 and IRF1 regions, which was decreased after JAKi treatment (Fig. 4D). Overall, these results indicated that genetic regions of ISGs were highly accessible both before and after ruxolitinib treatment, although their transcriptional levels were no longer upregulated after ruxolitinib treatment. This suggested that discontinuation of ruxolitinib may enable the gene expression of ISGs to increase again, and aggravate autoimmune features.

Seven months after starting ruxolitinib treatment, the patient developed rapidly progressing pneumonia. Considering the possibility of pulmonary invasive fungal infection, liposomal amphotericin B was started and switched to posaconazole later. During this event, ruxolitinib was temporarily discontinued to avoid drug-drug interactions. One month after antifungal treatment initiation and ruxolitinib discontinuation, pneumonia resolved; however, the patient redeveloped Raynaud’s phenomenon in both ankles. When ruxolitinib treatment resumed, Raynaud’s phenomenon disappeared. The aggravation of autoimmune features following ruxolitinib discontinuation may be explained by the persistence of epigenetic abnormality, e.g., sustained high accessibility of ISG regions.