Research Article Open Access | 10.1172/jci.insight.176963

Gabrielle Rabadam,1,2 Camilla Wibrand,3,4 Emily Flynn,5 George C. Hartoularos,6,7,8 Yang Sun,7 Chioma Madubata,4,5 Gabriela K. Fragiadakis,5,7 Chun Jimmie Ye,7,8,9,10,11,12 Susan Kim,4 Zev J. Gartner,2,11 Marina Sirota,10,13 and Jessica Neely4

1UC Berkeley-UC San Francisco Graduate Program in Bioengineering, and

2Department of Pharmaceutical Chemistry, UCSF, San Francisco, California, USA.

3Aarhus University, Aarhus, Denmark.

4Division of Pediatric Rheumatology, Department of Pediatrics,

5CoLabs,

6Graduate Program in Biological and Medical Informatics,

7Division of Rheumatology, Department of Medicine,

8Institute for Human Genetics,

9Department of Epidemiology and Biostatistics, and

10Bakar Computational Health Sciences Institute, UCSF, San Francisco, California, USA.

11Chan Zuckerberg Biohub, San Francisco, California, USA.

12Parker Institute for Cancer Immunotherapy, San Francisco, California, USA.

13Department of Pediatrics, UCSF, San Francisco, California, USA.

Address correspondence to: Jessica Neely, 550 16th Street, San Francisco, California 94158, USA. Email: jessica.neely@ucsf.edu.

Authorship note: GR and CW are co–first authors.

Find articles by Rabadam, G. in: JCI | PubMed | Google Scholar

1UC Berkeley-UC San Francisco Graduate Program in Bioengineering, and

2Department of Pharmaceutical Chemistry, UCSF, San Francisco, California, USA.

3Aarhus University, Aarhus, Denmark.

4Division of Pediatric Rheumatology, Department of Pediatrics,

5CoLabs,

6Graduate Program in Biological and Medical Informatics,

7Division of Rheumatology, Department of Medicine,

8Institute for Human Genetics,

9Department of Epidemiology and Biostatistics, and

10Bakar Computational Health Sciences Institute, UCSF, San Francisco, California, USA.

11Chan Zuckerberg Biohub, San Francisco, California, USA.

12Parker Institute for Cancer Immunotherapy, San Francisco, California, USA.

13Department of Pediatrics, UCSF, San Francisco, California, USA.

Address correspondence to: Jessica Neely, 550 16th Street, San Francisco, California 94158, USA. Email: jessica.neely@ucsf.edu.

Authorship note: GR and CW are co–first authors.

Find articles by Wibrand, C. in: JCI | PubMed | Google Scholar

1UC Berkeley-UC San Francisco Graduate Program in Bioengineering, and

2Department of Pharmaceutical Chemistry, UCSF, San Francisco, California, USA.

3Aarhus University, Aarhus, Denmark.

4Division of Pediatric Rheumatology, Department of Pediatrics,

5CoLabs,

6Graduate Program in Biological and Medical Informatics,

7Division of Rheumatology, Department of Medicine,

8Institute for Human Genetics,

9Department of Epidemiology and Biostatistics, and

10Bakar Computational Health Sciences Institute, UCSF, San Francisco, California, USA.

11Chan Zuckerberg Biohub, San Francisco, California, USA.

12Parker Institute for Cancer Immunotherapy, San Francisco, California, USA.

13Department of Pediatrics, UCSF, San Francisco, California, USA.

Address correspondence to: Jessica Neely, 550 16th Street, San Francisco, California 94158, USA. Email: jessica.neely@ucsf.edu.

Authorship note: GR and CW are co–first authors.

Find articles by Flynn, E. in: JCI | PubMed | Google Scholar

1UC Berkeley-UC San Francisco Graduate Program in Bioengineering, and

2Department of Pharmaceutical Chemistry, UCSF, San Francisco, California, USA.

3Aarhus University, Aarhus, Denmark.

4Division of Pediatric Rheumatology, Department of Pediatrics,

5CoLabs,

6Graduate Program in Biological and Medical Informatics,

7Division of Rheumatology, Department of Medicine,

8Institute for Human Genetics,

9Department of Epidemiology and Biostatistics, and

10Bakar Computational Health Sciences Institute, UCSF, San Francisco, California, USA.

11Chan Zuckerberg Biohub, San Francisco, California, USA.

12Parker Institute for Cancer Immunotherapy, San Francisco, California, USA.

13Department of Pediatrics, UCSF, San Francisco, California, USA.

Address correspondence to: Jessica Neely, 550 16th Street, San Francisco, California 94158, USA. Email: jessica.neely@ucsf.edu.

Authorship note: GR and CW are co–first authors.

Find articles by Hartoularos, G. in: JCI | PubMed | Google Scholar

1UC Berkeley-UC San Francisco Graduate Program in Bioengineering, and

2Department of Pharmaceutical Chemistry, UCSF, San Francisco, California, USA.

3Aarhus University, Aarhus, Denmark.

4Division of Pediatric Rheumatology, Department of Pediatrics,

5CoLabs,

6Graduate Program in Biological and Medical Informatics,

7Division of Rheumatology, Department of Medicine,

8Institute for Human Genetics,

9Department of Epidemiology and Biostatistics, and

10Bakar Computational Health Sciences Institute, UCSF, San Francisco, California, USA.

11Chan Zuckerberg Biohub, San Francisco, California, USA.

12Parker Institute for Cancer Immunotherapy, San Francisco, California, USA.

13Department of Pediatrics, UCSF, San Francisco, California, USA.

Address correspondence to: Jessica Neely, 550 16th Street, San Francisco, California 94158, USA. Email: jessica.neely@ucsf.edu.

Authorship note: GR and CW are co–first authors.

Find articles by Sun, Y. in: JCI | PubMed | Google Scholar

1UC Berkeley-UC San Francisco Graduate Program in Bioengineering, and

2Department of Pharmaceutical Chemistry, UCSF, San Francisco, California, USA.

3Aarhus University, Aarhus, Denmark.

4Division of Pediatric Rheumatology, Department of Pediatrics,

5CoLabs,

6Graduate Program in Biological and Medical Informatics,

7Division of Rheumatology, Department of Medicine,

8Institute for Human Genetics,

9Department of Epidemiology and Biostatistics, and

10Bakar Computational Health Sciences Institute, UCSF, San Francisco, California, USA.

11Chan Zuckerberg Biohub, San Francisco, California, USA.

12Parker Institute for Cancer Immunotherapy, San Francisco, California, USA.

13Department of Pediatrics, UCSF, San Francisco, California, USA.

Address correspondence to: Jessica Neely, 550 16th Street, San Francisco, California 94158, USA. Email: jessica.neely@ucsf.edu.

Authorship note: GR and CW are co–first authors.

Find articles by Madubata, C. in: JCI | PubMed | Google Scholar

1UC Berkeley-UC San Francisco Graduate Program in Bioengineering, and

2Department of Pharmaceutical Chemistry, UCSF, San Francisco, California, USA.

3Aarhus University, Aarhus, Denmark.

4Division of Pediatric Rheumatology, Department of Pediatrics,

5CoLabs,

6Graduate Program in Biological and Medical Informatics,

7Division of Rheumatology, Department of Medicine,

8Institute for Human Genetics,

9Department of Epidemiology and Biostatistics, and

10Bakar Computational Health Sciences Institute, UCSF, San Francisco, California, USA.

11Chan Zuckerberg Biohub, San Francisco, California, USA.

12Parker Institute for Cancer Immunotherapy, San Francisco, California, USA.

13Department of Pediatrics, UCSF, San Francisco, California, USA.

Address correspondence to: Jessica Neely, 550 16th Street, San Francisco, California 94158, USA. Email: jessica.neely@ucsf.edu.

Authorship note: GR and CW are co–first authors.

Find articles by Fragiadakis, G. in: JCI | PubMed | Google Scholar |

1UC Berkeley-UC San Francisco Graduate Program in Bioengineering, and

2Department of Pharmaceutical Chemistry, UCSF, San Francisco, California, USA.

3Aarhus University, Aarhus, Denmark.

4Division of Pediatric Rheumatology, Department of Pediatrics,

5CoLabs,

6Graduate Program in Biological and Medical Informatics,

7Division of Rheumatology, Department of Medicine,

8Institute for Human Genetics,

9Department of Epidemiology and Biostatistics, and

10Bakar Computational Health Sciences Institute, UCSF, San Francisco, California, USA.

11Chan Zuckerberg Biohub, San Francisco, California, USA.

12Parker Institute for Cancer Immunotherapy, San Francisco, California, USA.

13Department of Pediatrics, UCSF, San Francisco, California, USA.

Address correspondence to: Jessica Neely, 550 16th Street, San Francisco, California 94158, USA. Email: jessica.neely@ucsf.edu.

Authorship note: GR and CW are co–first authors.

Find articles by Ye, C. in: JCI | PubMed | Google Scholar

1UC Berkeley-UC San Francisco Graduate Program in Bioengineering, and

2Department of Pharmaceutical Chemistry, UCSF, San Francisco, California, USA.

3Aarhus University, Aarhus, Denmark.

4Division of Pediatric Rheumatology, Department of Pediatrics,

5CoLabs,

6Graduate Program in Biological and Medical Informatics,

7Division of Rheumatology, Department of Medicine,

8Institute for Human Genetics,

9Department of Epidemiology and Biostatistics, and

10Bakar Computational Health Sciences Institute, UCSF, San Francisco, California, USA.

11Chan Zuckerberg Biohub, San Francisco, California, USA.

12Parker Institute for Cancer Immunotherapy, San Francisco, California, USA.

13Department of Pediatrics, UCSF, San Francisco, California, USA.

Address correspondence to: Jessica Neely, 550 16th Street, San Francisco, California 94158, USA. Email: jessica.neely@ucsf.edu.

Authorship note: GR and CW are co–first authors.

Find articles by Kim, S. in: JCI | PubMed | Google Scholar |

1UC Berkeley-UC San Francisco Graduate Program in Bioengineering, and

2Department of Pharmaceutical Chemistry, UCSF, San Francisco, California, USA.

3Aarhus University, Aarhus, Denmark.

4Division of Pediatric Rheumatology, Department of Pediatrics,

5CoLabs,

6Graduate Program in Biological and Medical Informatics,

7Division of Rheumatology, Department of Medicine,

8Institute for Human Genetics,

9Department of Epidemiology and Biostatistics, and

10Bakar Computational Health Sciences Institute, UCSF, San Francisco, California, USA.

11Chan Zuckerberg Biohub, San Francisco, California, USA.

12Parker Institute for Cancer Immunotherapy, San Francisco, California, USA.

13Department of Pediatrics, UCSF, San Francisco, California, USA.

Address correspondence to: Jessica Neely, 550 16th Street, San Francisco, California 94158, USA. Email: jessica.neely@ucsf.edu.

Authorship note: GR and CW are co–first authors.

Find articles by Gartner, Z. in: JCI | PubMed | Google Scholar |

1UC Berkeley-UC San Francisco Graduate Program in Bioengineering, and

2Department of Pharmaceutical Chemistry, UCSF, San Francisco, California, USA.

3Aarhus University, Aarhus, Denmark.

4Division of Pediatric Rheumatology, Department of Pediatrics,

5CoLabs,

6Graduate Program in Biological and Medical Informatics,

7Division of Rheumatology, Department of Medicine,

8Institute for Human Genetics,

9Department of Epidemiology and Biostatistics, and

10Bakar Computational Health Sciences Institute, UCSF, San Francisco, California, USA.

11Chan Zuckerberg Biohub, San Francisco, California, USA.

12Parker Institute for Cancer Immunotherapy, San Francisco, California, USA.

13Department of Pediatrics, UCSF, San Francisco, California, USA.

Address correspondence to: Jessica Neely, 550 16th Street, San Francisco, California 94158, USA. Email: jessica.neely@ucsf.edu.

Authorship note: GR and CW are co–first authors.

Find articles by Sirota, M. in: JCI | PubMed | Google Scholar |

1UC Berkeley-UC San Francisco Graduate Program in Bioengineering, and

2Department of Pharmaceutical Chemistry, UCSF, San Francisco, California, USA.

3Aarhus University, Aarhus, Denmark.

4Division of Pediatric Rheumatology, Department of Pediatrics,

5CoLabs,

6Graduate Program in Biological and Medical Informatics,

7Division of Rheumatology, Department of Medicine,

8Institute for Human Genetics,

9Department of Epidemiology and Biostatistics, and

10Bakar Computational Health Sciences Institute, UCSF, San Francisco, California, USA.

11Chan Zuckerberg Biohub, San Francisco, California, USA.

12Parker Institute for Cancer Immunotherapy, San Francisco, California, USA.

13Department of Pediatrics, UCSF, San Francisco, California, USA.

Address correspondence to: Jessica Neely, 550 16th Street, San Francisco, California 94158, USA. Email: jessica.neely@ucsf.edu.

Authorship note: GR and CW are co–first authors.

Find articles by Neely, J. in: JCI | PubMed | Google Scholar

Authorship note: GR and CW are co–first authors.

Published May 14, 2024 - More info

Published in Volume 9, Issue 12 on June 24, 2024
JCI Insight. 2024;9(12):e176963. https://doi.org/10.1172/jci.insight.176963.
© 2024 Rabadam et al. This work is licensed under the Creative Commons Attribution 4.0 International License. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/. Published May 14, 2024 - Version history
Received: November 7, 2023; Accepted: May 8, 2024 View PDF Abstract

Juvenile dermatomyositis (JDM) is one of several childhood-onset autoimmune disorders characterized by a type I IFN response and autoantibodies. Treatment options are limited due to an incomplete understanding of how the disease emerges from dysregulated cell states across the immune system. We therefore investigated the blood of patients with JDM at different stages of disease activity using single-cell transcriptomics paired with surface protein expression. By immunophenotyping peripheral blood mononuclear cells, we observed skewing of the B cell compartment toward an immature naive state as a hallmark of JDM at diagnosis. Furthermore, we find that these changes in B cells are paralleled by T cell signatures suggestive of Th2-mediated inflammation that persist despite disease quiescence. We applied network analysis to reveal that hyperactivation of the type I IFN response in all immune populations is coordinated with previously masked cell states including dysfunctional protein processing in CD4+ T cells and regulation of cell death programming in NK cells, CD8+ T cells, and γδ T cells. Together, these findings unveil the coordinated immune dysregulation underpinning JDM and provide insight into strategies for restoring balance in immune function.

Graphical Abstract Introduction

Juvenile dermatomyositis (JDM) is part of a broad group of childhood-onset autoimmune conditions characterized by a type I IFN gene signature and specific autoantibodies ranging from related systemic conditions such as systemic lupus erythematosus (SLE) to endocrine-specific disorders such as type 1 diabetes (1–3). Despite a shared IFN signature, JDM is associated with pathognomonic rashes and proximal muscle weakness resulting in distinct clinical phenotypes. The etiology of JDM is not fully understood, but studies have shown that JDM is autoimmune mediated and associated with a combination of genetic and environmental risk factors (4). While mortality is low with corticosteroid treatment, long-term patient follow-up studies have reported that 60%–70% of patients have cumulative tissue damage, with the risk of damage increasing almost linearly for each year after diagnosis (5–7). This finding highlights the importance of early disease intervention and the need for a personalized approach to disease management to improve upon these outcomes.

Clinical management of JDM currently relies on compiled empirical metrics such as physician global visual analog scale (VAS) of disease activity and muscle strength quantified via the childhood myositis assessment scale (CMAS) or manual muscle testing (MMT) (8). However, how these clinically observable phenotypes are rooted in disease immunopathology remains insufficiently understood. The presence of myositis-specific antibodies (MSA) that correspond to distinct clinical phenotypes and recent work showing that MSAs may be pathogenic suggest the involvement of B cells (9–11). The expansion of naive B cells in JDM has been highlighted by 3 independent studies using flow cytometry, mass cytometry, and single-cell RNA-Seq (12–14). The adaptive arm of the immune system is further implicated in disease pathogenesis by several large immunophenotyping studies that demonstrated the expansion of extrafollicular Th2 memory cells and central memory B cells (15, 16). Additionally, the innate immune system has emerged as a contributor in JDM with increased macrophages in skin and NK cell dysfunction described peripherally (13, 17, 18). Together, these findings highlight the involvement of both the adaptive and innate immune compartments in JDM in blood and disease-affected tissues. However, it also raises the question of whether the cause of JDM lies in a single cell type or is a manifestation of broadly dysregulated cellular interactions across the immune system.

Systems-level studies based on single-cell measurements are required to reveal how dysregulated cell populations act individually or cooperatively to produce the observed inflammation. Accordingly, several groups have turned to next-generation sequencing, as it enables unbiased profiling of tissues at a single-cell resolution. We previously described a pan–cell type IFN gene signature overexpressed in treatment-naive JDM that was most strongly correlated with disease activity in cytotoxic cell types (14). This signature has since been independently identified in the peripheral blood of treatment-naive patients (19). However, these studies have utilized small cohorts and lack pediatric controls. Thus, it has been challenging to determine which of these findings are specific to JDM compared with healthy children, how these disease-specific dysregulated cell states are coordinated with one another, and which of these states cooperatively change in response to treatment.

In this study, we addressed this challenge by profiling JDM across several stages of disease activity using multiplexed Cellular Indexing of Transcriptomes and Epitopes by sequencing (CITEseq) of peripheral blood mononuclear cells (PBMCs) from 15 patients with JDM, totaling 22 samples, and 5 healthy controls (HC). Compositional analysis of immune populations identified a disease activity–associated imbalance of naive and mature lymphocytes, corroborated by distinct immunophenotypes in treatment-naive disease. To move beyond the identification of disease-associated cell populations and toward an understanding of immune-scale dysregulation in JDM, we applied a recently developed computational method DECIPHERseq to infer networks of coordinated cell states from large cohorts of single-cell data (20). Importantly, this unsupervised method takes advantage of the biological heterogeneity in the entire data set, improving upon previous work that relied on pairwise comparisons of subsetted disease groups. Among other signatures previously masked by traditional single-cell analysis, this approach revealed cooccurring cell states in CD4+ T and B cell populations, suggestive of extrafollicular responses. A subset of these CD4+ T signatures implicates disruption of protein targeting and immune tolerance processes; notably, these cell states persist even in patients in remission off medication. Furthermore, we show that the hyperactive type I IFN response in disease is paralleled by impaired cell death processes in cytotoxic immune cells, highlighting the functional imbalance across immune compartments that typifies this complex autoimmune disease. This broadened understanding of the underlying immune dysregulation in disease can inform precision treatment strategies for JDM.

Results

JDM is associated with immunophenotypic differences in B and CD4+ T cell compartments. To gather a data set with appropriate controls and limited confounding, patients were selected according to disease activity and medication status (Figure 1A and Supplemental Table 1; supplemental material available online with this article; https://doi.org/10.1172/jci.insight.176963DS1). Of the 15 patients with JDM, serial samples were collected from 6 individuals, totaling 22 samples. Detailed information on sample numbers from patients can be found in Figure 1A. To minimize confounding by immune suppression, the study included 9 treatment-naive samples as well as 6 samples from patients with inactive disease off medication. CITEseq was performed on PBMCs to generate single-cell libraries (Figure 1B). Surface protein expression was measured using antibody-derived tags (ADT). Following preprocessing steps, we identified 29 clusters, which comprised 21 distinct immune cell populations across 105,827 cells (Figure 2A). Clusters were annotated using canonical RNA (Figure 2B) and protein markers (Supplemental Figure 1, A and B) within all major mononuclear immune cell compartments.

Figure 1

Study design and analysis strategy for profiling PBMCs from 27 samples. (A) Overview of clinical characteristics of study cohort. Individuals are labeled by donor ID (JDM 1–15, HC 16–20). Longitudinal samples were collected from the following donors: JDM1 (n = 2), JDM2 (n = 2), JDM4 (n = 3), JDM8 (n = 2), JDM13 (n = 2), JDM15 (n = 2). Icon shapes denote disease activity group, and shades of blue denote medication regimen. (B) Analysis strategy for CITEseq data from PBMCs. n = 22 JDM, n = 5 HC.

Figure 2

Cell types associated with JDM in peripheral blood. (A) UMAP constructed using weighted nearest neighbors (wnn) clustering colored by cell type. pDCs, plasmacytoid DCs; cDCs = classical DCs; PBs, plasmablasts; B_mem, memory B cells. (B) Heatmap with top 2 markers per cluster. (C) Box plot shows cell type proportion by disease group, using Kruskal-Wallis test with Dunn’s post hoc test comparing TNJDM with HC, TNJDM with inactive JDM, and inactive with HC (Holm’s, Padj < 0.05; *Padj < 0.05, **Padj < 0.01). The dot plot above shows the Spearman correlation between corresponding cell type proportion in box plot and Physician Global VAS, where the size of the dot indicates the correlation, the color indicates the direction of the correlation, and the border weight indicates significance (Padj < 0.05). (D) Heatmap with selected ADT protein markers. Asterisks mark significant comparisons between TNJDM and HC per cell type with an absolute LFC > 0.5 and Padj < 0.05.

We first characterized global changes to cell composition across disease states comparing treatment naive JDM (TNJDM), inactive JDM, and HC (Figure 2C). Within the T cell compartment, the proportion of Tregs (CD45RO+, IL2R+, FOXP3+) was increased in patients with TNJDM (P = 0.02) consistent with previous findings (14). CD4+ effector T cells (CD45RO+) and γδ T cluster 2 (TRDC, TRGC) were significantly increased in patients with inactive JDM, and the proportion of cells from these populations negatively correlated with disease activity measures (P < 0.05). There was an overall decrease in innate populations in TNJDM compared with HC and inactive JDM, and the proportion of these cell types also correlated negatively with disease activity (P < 0.05).

Compared with HCs and patients with inactive disease, treatment-naive patients had higher proportions of multiple naive B cell populations, herein referred to as “B_naive,” and their UMAP cluster number, including B_naive1 (IgM+IgD+CD38+CD24+CD10+) corresponding to an immature naive B population, B_naive2 (IgM,+IgD+CD38loCD24lo), and B_naive3 (IgM+IgD+CD38+CD24+), and the proportion of these populations positively correlated with multiple disease activity measures (P < 0.05) (Figure 2C). The proportion of B_mem cells, characterized by TNFRSF13B expression, negatively correlated with the muscle VAS score (P < 0.05). The immature naive B population had higher expression of CD38 (both RNA and protein) and MZB1, 2 genes essential for plasma cell differentiation, compared with all other B cell clusters (21, 22).

Given the observed imbalance of lymphocytes in treatment-naive JDM, we next sought to immunophenotype B cell and CD4+ T cell subsets in JDM at the proteomic level to gain molecular insight into cell states (Figure 2D and Supplemental Table 2). Differential protein analysis of immature naive B cells comparing treatment-naive JDM to HC identified increased expression of MICA-MICB and decreased expression of CD1C, BAFF-R, and PD-L1 (Figure 2D). Within the CD4+ T compartment, CD4+ Tregs from TNJDM had higher expression of Tim-3, ICOS, CD164, and CD38 and downregulation of CD101 a molecule which decreases proinflammatory T cell responses (23). CD4+ Teff in patients with TNJDM had higher surface expression of CD164 and PD-1 and showed downregulation of KLRG1, an inhibitory molecule (Figure 2D). The overexpression of PD-1 on the cell surface suggested that peripheral Th cells might be present in JDM (24, 25). However, while ICOS expression was higher (P < 0.05), no difference was found in surface expression of CXCR5 between CD45ROhiPD-1hiCD4+ T cells and CD45ROloPD-1loCD4+ T cells, and these cells were not significantly expanded in JDM (Supplemental Figure 2).

SIGLEC-1 expression is a composite measure of the IFN gene signature in JDM. We next compared gene and protein expression between treatment-naive JDM and HC samples in all cell types based on the hypothesis that certain cell types may not be altered in composition but may be functionally altered at the molecular level. Monocytes displayed the highest number of differentially expressed genes and proteins in this analysis, including CD169 (SIGLEC-1), CD107a (LAMP-1), and CD164 (Supplemental Figure 3). SIGLEC-1 is a monocyte-restricted IFN-induced protein that was recently identified as a potential biomarker in JDM (26). Both CD107a and CD164 are cell adhesion molecules involved in trafficking of activated PBMCs and adhesion to vascular endothelium (27).

A common finding across all cell types when comparing treatment-naive JDM and HC samples was overexpression of genes enriched in type I IFN processes, which was previously reported in bulk expression data and confirmed in single-cell studies (Supplemental Figure 4) (14, 19, 28). Using an IFN gene score derived from the transcriptional data (Supplemental Figure 5), we plotted the per-patient average score in each cell type (Figure 3A). This approach did not detect IFN gene expression to persist beyond the treatment-naive state, and 2 patients with TNJDM had negligible IFN gene signatures as quantified by this method. This heterogeneity of the IFN gene signature was partly explained by disease activity level (Figure 3B), as a bulk IFN gene score correlated with disease activity (r = 0.69). However, the remaining unexplained heterogeneity of this IFN score exemplifies a limitation of utilizing gene scores identified through pairwise comparisons between subsets of the data.

Figure 3

Type I IFN–induced gene and protein expression is associated with disease activity in JDM in CD14+ monocytes. (A) Heatmap of average IFN score per cell type and sample. Hierarchical clustering was performed using Euclidean distance and the complete clustering method. IFN score was calculated based on average expression of IFN module across all cells per sample. (B) Spearman correlation between IFN score and Physician Global VAS colored by disease group. (C) Scatter plot showing Spearman correlation between CD169 (SIGLEC-1) expression in CD14+ monocytes and Physician Global VAS. (D) Scatter plot showing Spearman correlation between IFN score and Physician Global VAS. (E) Scatter plot showing Spearman correlation between CD169 expression and IFN score in CD14+ monocytes.

Given that SIGLEC-1 is a type I IFN–induced protein, we investigated whether patterns of type I IFN–stimulated gene expression were reflected at the protein level, as protein biomarkers are more amenable for clinical lab-based testing. SIGLEC-1 expression in CD14+ monocytes also correlated with disease activity to a similar degree as the IFN gene signature (Figure 3, C and D), and SIGLEC-1 expression was itself highly correlated with the IFN gene signature in monocytes (Figure 3E). This suggests that SIGLEC-1 expression in CD14+ monocytes is a representative composite measure of the IFN gene signature in JDM. These results underscore the potential of SIGLEC-1 as a biomarker of IFN responses in JDM that may be useful for tracking disease activity.

Unsupervised network analysis reveals coordinated cell states shared among immune cells in JDM. We next turned to a systems-level approach to better understand the coordination of immune cell gene programs in JDM relative to HCs and in relation to disease activity level. We applied an unsupervised network inference method, DECIPHERseq, to the 6 major cell types annotated in the data set: B cells, CD4 T, CD8 T, NK cells, γδ T cells, and myeloid cells (Figure 4A). DECIPHERseq relies on nonnegative matrix factorization (NMF) to first break the data set down into gene sets that represent distinct states of biological activity, or “activity programs,” and then constructs a network of gene expression programs (GEPs) based on how expression of the programs covaries across patient samples (Figure 4A) (29). After outlier filtering, NMF identified 76 activity programs (Figure 4B).

Figure 4

DECIPHERseq extracts gene expression programs from single-cell RNA-Seq data in JDM. (A) Overview of the DECIPHERseq workflow. (B) Heatmap showing 6 major clusters of GEPs identified by DECIPHERseq (Pearson). GEPs are clustered into modules, with isolated GEPs filtered out (grayscale).

Next, a force-directed network graph from the correlation matrix of activity programs was constructed in which each node represents a program and each edge represents a statistically significant (P < 0.05) positive correlation between 2 nodes (Figure 5A). Using DECIPHERseq’s community detection algorithm, we identified 6 hubs of interconnected activity programs or “modules.” All modules contained multiple cell types, highlighting that many biological processes in JDM are coordinated across several immune cell types (Figure 4B and Figure 5A). We annotated each node using gene set enrichment analysis (GSEA) of Gene Ontology (GO) terms on each program’s ranked marker gene list (Supplemental Tables 3–4 and Supplemental Figures 6–11) (30, 31).

Figure 5

Network of coordinated biological activity inferred from GEPs in peripheral blood. (A) Network constructed from correlated GEPs in PBMCs from patients with JDM and HCs. Nodes represent programs in the given cell types, and edges represent positive significant correlations (Pearson, P < 0.05). (B) Dot plot showing selected gene sets found to be enriched within specific modules compared with the rest of the network. Color corresponds to module enrichment P value, and size corresponds to a set’s rank in the list of significantly enriched gene sets for that given module ordered by ascending module enrichment P value (network permutations, GSEA, FDR < 0.01). All gene sets shown fall in the top 10 terms for their respective modules (total gene sets: 626).

DECIPHERseq’s module enrichment analysis identified consensus biological themes for each module in an unsupervised manner (Figure 5B and Supplemental Table 5). Module 1 was enriched for type I IFN response programs such as Response to Virus. Module 2 was enriched in ribosomal processes including Translational Initiation. Module 3 included lymphocyte programs and was significantly enriched for cell adhesion and migration. Module 4 represented cells’ steady state processes as it was enriched for gene sets like Circadian Rhythm. Module 5 was annotated as a Stress Response module and enriched for Regulation of Cell Death and Cellular Response to Chemical Stress. Module 6 contained few unique gene sets and was enriched for programs intrinsic to eukaryotic cells like DNA Packaging.

JDM CD4+ T cells and B cells display persistent alterations in gene expression in both active disease and remission. Next, we aimed to interpret the annotated network in the context of GEPs associated with JDM compared with HCs irrespective of disease activity. We first focused on Module 1, which was enriched in type I IFN responses. Many programs in this module were increased in TNJDM, as expected (Figure 6, A and B). All 6 major cell types expressed IFN gene programs that were highly correlated to one another, as shown by the closely connected hub at the center of Module 1 (Figure 6A). This IFN hub was associated with JDM as compared with HCs (P < 0.05) (Figure 6, B and C). IFN modules identified by NMF were highly expressed in all treatment-naive patients as well as some patients with active disease, inactive disease, and a HC (Figure 6B), in contrast to the signature of IFN gene expression previously detected by differential gene expression in Figure 3A. This highlights the strength of this method to more accurately reflect the low-dimensional space of gene expression where measurement of many genes working together may be needed to detect underlying biological processes (32–34).

Figure 6

JDM is associated with a central IFN hub and cell-specific gene programs in the B and CD4 T compartments. (A) Zoomed-in graph of Module 1. GSEA results for Response to type I IFN GO term shown with each node colored according to FDR. Padj value of module enrichment is also shown (network permutations, Methods). (B) Heatmap showing significant differences in expression of selected programs between HC (n = 5) and patients with JDM (n = 22), with columns annotated by P values (P < 0.05) of case-control (t test) and disease activity association (4-group 1-way ANOVA). (C) Network graph showing case-control analysis of each program’s expression, with node size scaled according to P value and colored according to strength of the association between disease status and program expression (t test).

We next identified gene programs in Modules 1–3 expressed more highly in all patients with JDM compared with HCs (P < 0.05). These included B cell (5 and 14) and CD4T (1, 10, 17) programs, and their expression persisted even in patients with inactive JDM who previously achieved remission off medication (Figure 6B and Supplemental Figure 12). Patients with JDM more highly expressed 2 B cell programs; B5 in Module 1 was enriched in mRNA metabolic processing, RNA splicing, chromatin organization and modification, and cell cycle regulation, and B14 in Module 3 was enriched in chromatin remodeling and cytoskeletal organization (Supplemental Figure 6 and Supplemental Figure 8). These enriched biological processes suggest that a subpopulation of B cells are more transcriptionally active and undergoing epigenetic regulation in JDM relative to HCs.

In Module 3, correlated to B14, CD4T1 (enriched in cell migration, adhesion, activation, and secretion) was expressed more highly in JDM and in the region of the Uniform Manifold Approximation and Projection (UMAP) corresponding to CD4+ Teff (Supplemental Figure 13). This CD4T1 program expressed by CD4+ Teff contained genes (GATA3, CCR4, PRDM1) that indicate possible skewing toward a Th2 subset, while expression of PRDM1 (Blimp-1) suggests participation in extrafollicular reactions (Figure 6B). Th2 CD4+ T cells were previously found to be expanded in JDM and associated with extrafollicular B cell–T cell help (15, 16). We observed similar expression of Th2 genes (GATA3, CCR4, PRDM1) in CD4T10, a Treg program (FOXP3, IKZF2, IL2RA) expressed more highly in JDM (Figure 6B). CD4T1 and CD4T10 included genes for costimulatory molecules OX40 (TNFRSF4) and GITR (TNFRSF18), both of which have been described to promote survival and proliferation of CD4+ Teff and have bee