Systemic lupus erythematosus (SLE) is an autoimmune disease characterized by the presence of low-density granulocytes (LDGs) with a heightened capacity for spontaneous NETosis, but the contribution of LDGs to SLE pathogenesis remains unclear. To characterize LDGs in human SLE, gene expression profiles derived from isolated LDGs were characterized by weighted gene coexpression network analysis, and a 92-gene module was identified. The LDG gene signature was enriched in genes related to neutrophil degranulation and cell cycle regulation. This signature was assessed in gene expression datasets from two large-scale SLE clinical trials to study associations between LDG enrichment, SLE manifestations, and treatment regimens. LDG enrichment in the blood was associated with corticosteroid treatment as well as anti-dsDNA, low serum complement, renal manifestations, and vasculitis, but the latter two of these associations were dependent on concomitant corticosteroid treatment. In addition, LDG enrichment was associated with enrichment of gene signatures induced by type I IFN and TNF irrespective of corticosteroid treatment. Notably, LDG enrichment was not found in numerous tissues affected by SLE. Comparison with relevant reference datasets indicated that LDG enrichment is likely reflective of increased granulopoiesis in the bone marrow and not peripheral neutrophil activation. The results have uncovered important determinants of the appearance of LDGs in SLE and have emphasized the likely role of LDGs in specific aspects of lupus pathogenesis.

Systemic lupus erythematosus (SLE) is an autoimmune disease characterized by autoreactive B cell hyperactivity, autoantibody generation, and the presence of a type I IFN gene expression signature. SLE patients also manifest an increased population of low-density granulocytes (LDGs) in the peripheral blood that remains in the PBMC fraction after Ficoll density gradient separation rather than sedimenting with normal-density granulocytes (1). LDGs are known to appear in the circulation of subjects with a number of diseases, including rheumatoid arthritis, HIV infection, cancer, tuberculosis, and Plasmodium vivax infection (16). Although the presence of LDGs in these conditions tends to be associated with more severe disease, the physiologic effects of this population seem to be mediated by diverse proinflammatory and anti-inflammatory mechanisms. For example, LDGs may contribute to rheumatoid arthritis pathogenesis by exposing immunogenic citrullinated histones, whereas LDGs in HIV infection may aggravate disease by inhibiting CD4+ T cells via arginase 1 (3, 7).

In SLE, LDGs have been described as a proinflammatory subset of neutrophils with an enhanced capacity to release neutrophil extracellular traps (NETs) compared with autologous SLE neutrophils and healthy control (HC) neutrophils through a process called NETosis (8). During this process, neutrophils expel chromatin, antimicrobial agents, and immunostimulatory molecules into the extracellular space to trap and kill bacteria, but this process can also induce tissue damage (7, 9). LDGs expose dsDNA, oxidized mitochondrial DNA, LL-37, elastase, and IL-17, among other molecules, during NETosis, and increased NETosis by LDGs may be an important source of immunostimulatory molecules and autoantigens involved in the pathogenesis of SLE (8, 10).

The presence of LDGs in pediatric SLE patients is associated with increased lupus activity as measured by the SLE Disease Activity Index (SLEDAI) (11). LDGs have also been implicated in skin involvement and vascular damage in SLE, and netting neutrophils have been described in the glomeruli and skin of lupus patients, although it remains unclear whether the infiltrating cells were LDGs or normal-density neutrophils (8, 12).

Based on nuclear morphology and surface marker expression, LDGs have been hypothesized to be immature neutrophil precursors released from the bone marrow, perhaps related to stimulation by CSF, such as G-CSF or GM-CSF (4, 12). However, the specific origin of LDGs in SLE and, more importantly, the mechanisms by which they contribute to organ involvement and/or disease activity remain unclear. To gain more insight into LDGs in SLE, we have used a large-scale bioinformatics approach that combines gene expression data and clinical measurements. The goals of this research were to generate a transcriptomic signature that characterizes LDGs in SLE, to determine whether this signature can be detected in the blood and tissue of SLE patients, and to characterize the relationship between this signature and SLE disease manifestations.

Data were derived from publicly available datasets on Gene Expression Omnibus (https://www.ncbi.nlm.nih.gov/geo/) and collaborators. Raw data sources are as follows: LDGs (GSE26975 [9 HC neutrophils, 10 SLE neutrophils, 10 SLE LDGs]), PBMCs (GSE50772 [20 HC, 59 SLE], GSE81622 [25 HC, 30 SLE], FDABMC3 [6 HC, 43 SLE]), whole blood (WB) (GSE49454 [10 HC, 49 SLE], GSE88884 [17 HC, 1612 SLE]), kidney glomerulus and tubulointerstitium (TI) (GSE32591 [14 HC, 30 lupus nephritis (LN)]), skin (GSE52471 [3 HC, 7 discoid lupus erythematosus], GSE72535 [8 HC, 9 discoid lupus erythematosus]), synovium (GSE36700 [4 osteoarthritis, 4 SLE]), and bone marrow myeloid lineage cells (GSE19556 [6 promyelocytes (PM), 6 myelocytes (MY), 6 bone marrow polymorphonuclear neutrophils (PMN, 6 peripheral blood PMN]). Clinical data, when available, including disease activity assessed by SLEDAI, anti-dsDNA titers, and complement levels, were included in the analysis.

Statistical analysis was conducted using R and relevant Bioconductor packages. Nonnormalized arrays were inspected for visual artifacts or poor RNA hybridization using Affy quality control plots. To inspect the raw data files for outliers, principal component analysis plots were generated for all cell types available for each experiment. Datasets culled of outliers were cleaned of background noise and normalized using GeneChip robust multiarray averaging, resulting in log2 intensity values compiled into R expression set objects (E-sets). To increase the probability of identifying differentially expressed genes (DEGs), analysis was conducted using normalized datasets prepared using the native Affy chip definition files (CDFs), followed by custom BrainArray (BA) Entrez CDFs maintained by the University of Michigan Molecular and Behavioral Neuroscience Institute. The Affy CDFs include multiple probes per gene and almost twice as many probes as BA CDFs. Although Affy CDFs can provide the greatest amount of variance information for Bayesian fitting, the BA CDFs are used to exclude probes with known nonspecific binding and those shown by quarterly BLASTs (https://blast.ncbi.nlm.nih.gov/Blast.cgi) to no longer fall within the target gene. Illumina CDFs were used for the Illumina datasets (GSE49454, GSE81622).

The CDF-annotated E-sets were filtered to remove probes with very low-intensity values. This reduced the E-set dimensions and the degree of multiple hypothesis testing correction, which increased the statistical significance of the differential expression (DE) probes. Probes missing gene annotation data were also discarded. GeneChip robust multiarray averaging–normalized expression values were variance corrected using local empirical Bayesian shrinkage before calculation of DE, using the ebayes function in the Bioconductor limma package. Resulting p values were adjusted for multiple hypothesis testing using the Benjamini–Hochberg correction, which resulted in a false discovery rate (FDR). Significant Affy and BA probes within each study were merged and filtered to retain DE probes with an FDR <0.05, which were considered statistically significant. This list was further filtered to retain only the most significant probe per gene to remove duplicate probes.

Log2 normalized microarray expression values were used as input to weighted gene coexpression network analysis (WGCNA) to conduct an unsupervised clustering analysis, resulting in coexpression “modules,” or groups of densely interconnected genes, which could correspond to comparably regulated biologic pathways (13). For each experiment, an approximately scale-free topology matrix was first calculated to encode the network strength between probes. Probes were clustered into WGCNA modules based on topology matrix distances. Resultant dendrograms of correlation networks were trimmed to isolate individual modular groups of probes, labeled using semirandom color assignments, based on a detection cut height of 1, with a merging cut height of 0.2, with the additional use of a partitioning around medoids function. Final membership of probes representing the same gene into modules was based on selection of the greatest within-module correlation with module eigengene (ME) values. Expression profiles of genes within modules were summarized by an ME, the module’s first principal component. MEs act as characteristic expression values for their respective modules and can be associated with sample traits such as cell type, cohort (HC or SLE), or serological measurements. This was done by Welch t test. The correlation coefficient of each gene in a module with the ME (kME), a metric for module membership, was used to determine the association of individual genes with the expression of the module as a whole. The mean kME of all genes in a module was taken as a metric of overall module quality. If the genes in a module have low kMEs, it is indicative that a few highly variable genes have dominated the eigengene calculation. Modules with mean kMEs close to 1 were considered to be high quality, and modules with mean kMEs close to 0 were considered to be low quality. When analyzing multiple datasets, the grand mean was the mean of the mean kMEs for each dataset.

STRING (v10.5) was used to score protein–protein interaction networks, which were visualized using the Cytoscape (v3.5.1) software. The clusterMaker2 (v1.1.0) plugin application was used to create MCODE clusters of the most closely related genes.

The gene set variation analysis (GSVA) Bioconductor package was used as a nonparametric, unsupervised method for estimating the variation of predefined gene sets in patient and control samples of microarray expression datasets. The GSVA algorithm accepts a gene expression matrix of log2-transformed expression values and a collection of predefined gene sets as inputs. Enrichment scores are calculated nonparametrically using a Kolmogorov–Smirnov–like random walk statistic. The enrichment scores were the largest positive and negative random walk deviations from zero, respectively, for a particular sample and gene set. Individual patient gene expression sets are considered positively or negatively enriched for a given signature if they display a z-score of >2 or <2, respectively, relative to controls. Analysis of GSVA scores was carried out using Fisher exact test or Welch unequal variances t test, where appropriate.

The p values resulting from DE analysis were adjusted by the Benjamini–Hochberg FDR correction. Analysis of parametric data was done using a two-tailed Welch t test. Correlation analysis of continuous variables was done by Pearson correlation, and analysis of noncontinuous variables was done by Spearman rank correlation. Correlations are reported as Pearson r or Spearman rho, as appropriate. Odds ratio analysis was done by Fisher exact test, and odds ratios are accompanied by 95% confidence intervals.

Samples from GSE26975 were used to carry out DE analysis of LDG, SLE neutrophils, and HC neutrophils (Fig. 1). This approach identified 657 DEGs in LDGs compared with SLE neutrophils (173 upregulated, 484 downregulated) and 224 DEGs compared with HC neutrophils (145 upregulated, 79 downregulated) (Supplemental Material File 1). No DEGs were noted between SLE neutrophils and HC neutrophils. A total of 132 DEGs were found to be upregulated in LDGs compared with both SLE neutrophils and HC neutrophils.

FIGURE 1.

LDG DEGs relative to SLE neutrophils and HC neutrophils. Comparison of LDG upregulated genes versus SLE neutrophils or HC neutrophils by limma analysis. Genes were considered upregulated or downregulated if they had an FDR <0.05. (A) Comparison of LDG genes upregulated versus SLE neutrophils or HC neutrophils. (B) Comparison of LDG genes downregulated versus SLE neutrophils or HC neutrophils.

FIGURE 1.

LDG DEGs relative to SLE neutrophils and HC neutrophils. Comparison of LDG upregulated genes versus SLE neutrophils or HC neutrophils by limma analysis. Genes were considered upregulated or downregulated if they had an FDR <0.05. (A) Comparison of LDG genes upregulated versus SLE neutrophils or HC neutrophils. (B) Comparison of LDG genes downregulated versus SLE neutrophils or HC neutrophils.

Close modal

LDG DEGs included transcripts for granule proteins, cell cycle regulation, chromatin remodeling, cell adhesion, and cytoskeletal regulation. Upregulated genes also included many genes specific to platelets, and downregulated genes included many transcripts for TCR and BCR complexes, suggesting some contamination during neutrophil isolation.

Because of the potential effects of cellular contamination on the DEG analysis, a new computational approach was developed to identify highly discriminatory genes characteristic of LDGs that could be used to facilitate downstream analyses in blood and tissues. An overview of this process can be found in Supplemental Fig. 1. Identifying groups of coexpressed genes could minimize the effects of potential contamination and enable better characterization of LDGs by separating LDG-specific genes from lymphocyte- and platelet-specific groups of genes. Samples from GSE26975 were used to carry out unsupervised WGCNA of LDGs, SLE neutrophils, and HC neutrophils to identify modules of potentially informative genes based on coexpression rather than known experimental design (Fig. 2). This approach initially generated 56 modules of genes. Six of these had ME values that were significantly increased or decreased in LDG samples by Welch t test (p < 0.05) (Supplemental Material File 2). One module (midnightblue) was removed from consideration upon inspection because its ME values did not differ from SLE neutrophils in a majority of samples. Another module (mediumpurple3) was removed because it contained transcripts from B and T cells, an indication that the WGCNA approach could filter out contamination.

FIGURE 2.

WGCNA ME values separate LDGs from both SLE neutrophils and HC neutrophils. Samples from GSE26975 were used in two separate WGCNA analyses to examine LDGs and HC or LDGs and SLE neutrophils. Module colors are assigned by the WGCNA pipeline based on module size. Eigengene values separate LDGs from (A) HC neutrophils (n = 9 HC, 10 LDG) and (B) SLE neutrophils (n = 10 SLE, 10 LDG) by Welch t test (*p < 0.05).

FIGURE 2.

WGCNA ME values separate LDGs from both SLE neutrophils and HC neutrophils. Samples from GSE26975 were used in two separate WGCNA analyses to examine LDGs and HC or LDGs and SLE neutrophils. Module colors are assigned by the WGCNA pipeline based on module size. Eigengene values separate LDGs from (A) HC neutrophils (n = 9 HC, 10 LDG) and (B) SLE neutrophils (n = 10 SLE, 10 LDG) by Welch t test (*p < 0.05).

Close modal

The remaining four modules (pink, black, grey60, and greenyellow [Supplemental Material File 2]) were compared based on their ME values in the LDG expression data and the genes that comprised them. The pink and black modules had strongly correlated MEs (r = 0.99, p = 7.80 × 10−8) and shared 409 genes (Fig. 3A, 3C). Similarly, the grey60 and greenyellow modules had correlated MEs (r = 0.94, p = 4.87 × 10−5) and shared 92 genes (Fig. 3B, 3D). All other ME correlations were NS (all p > 0.6). Modules were then consolidated with the goal of acquiring a gene signature that could robustly set LDGs apart from both HC neutrophils and SLE neutrophils. The pink and black modules were combined to form module A, and the greenyellow and grey60 modules were combined to form module B. (Supplemental Material File 2) These modules were then subjected to functional analysis to identify a specific, robust LDG gene signature.

FIGURE 3.

LDG WGCNA modules can be grouped by eigengene values and constituent genes. LDG eigengene values for (A) pink and black modules or (B) grey60 and greenyellow modules demonstrate that the four WGCNA modules can be broken into two groups based on the behavior of their eigengenes from patient to patient. Pearson r and p values are shown. WGCNA modules with highly correlated eigengenes have many genes in common. (C) LDG module A was formed from the genes shared between the pink and black modules. (D) LDG module B was formed from the genes shared between the grey60 and greenyellow modules.

FIGURE 3.

LDG WGCNA modules can be grouped by eigengene values and constituent genes. LDG eigengene values for (A) pink and black modules or (B) grey60 and greenyellow modules demonstrate that the four WGCNA modules can be broken into two groups based on the behavior of their eigengenes from patient to patient. Pearson r and p values are shown. WGCNA modules with highly correlated eigengenes have many genes in common. (C) LDG module A was formed from the genes shared between the pink and black modules. (D) LDG module B was formed from the genes shared between the grey60 and greenyellow modules.

Close modal

Protein–protein interaction networks were generated for each consolidated LDG module in STRING and sorted to form clusters in Cytoscape using MCODE. Functional analysis of clustered genes showed that module B contained a strongly interconnected cluster of neutrophil granule genes and another cluster of genes associated with DNA synthesis and cell cycle regulation (Fig. 4). Overall, 30 of 92 genes in module B had the neutrophil degranulation Gene Ontology (GO) designation, including AZU1, CAMP (LL-37), CTSG, DEFA4, ELANE (ELA2), LCN2 (NGAL), LTF, MMP8, MPO, and RNASE3. Additionally, 21 of 92 genes in module B had the cell cycle GO designation. Of the 41 remaining genes, there were several genes encoding typical LDG surface proteins, including CD66b, and also genes involved in transcriptional regulation, but no overall function for the remaining genes could be determined. Because circulating neutrophils do not express granulopoietic genes and because SLE neutrophils do not differentially express any genes relative to HC neutrophils, the presence of this module of genes in the blood or tissue of SLE patients was likely to be attributable to LDGs and not merely neutrophil activity (14).

FIGURE 4.

STRING/MCODE functional analysis of LDG module B elucidates two major clusters characterized by cell cycle and neutrophil degranulation. MCODE clustering was used to identify the most strongly connected members of module B’s STRING protein-protein interaction network. The top cluster has many genes associated with the cell cycle by GO (diamonds). The bottom cluster is almost entirely composed of genes associated with neutrophil degranulation (squares). Cell cycle and neutrophil degranulation genes not connected to an MCODE cluster are shown on the right. The presence of neutrophil-associated genes in module B led to its selection as the module used to query blood and tissue gene expression data.

FIGURE 4.

STRING/MCODE functional analysis of LDG module B elucidates two major clusters characterized by cell cycle and neutrophil degranulation. MCODE clustering was used to identify the most strongly connected members of module B’s STRING protein-protein interaction network. The top cluster has many genes associated with the cell cycle by GO (diamonds). The bottom cluster is almost entirely composed of genes associated with neutrophil degranulation (squares). Cell cycle and neutrophil degranulation genes not connected to an MCODE cluster are shown on the right. The presence of neutrophil-associated genes in module B led to its selection as the module used to query blood and tissue gene expression data.

Close modal

To confirm that module B was not related to neutrophil activation, module B genes were compared with 809 genes differentially expressed by activated human neutrophils in experimental endotoxemia at three time points (15). Only 18 of the 809 (2.2%) activated neutrophil genes were found among the 92 genes in module B, including 13 of the 30 module B genes implicated in neutrophil degranulation (Supplemental Material File 3). Although module B bears some similarity to this activated neutrophil signature (18 of 92 genes), it retains a unique array of granule proteins (AZU1, CAMP, CHIT1, CTSG, DEFA4, ELANE, LTF, MPO, RNASE3), cell cycle proteins, and surface markers (CD24, CD66b, CD66c, CLEC12A, MS4A3) that set it apart. Furthermore, an analysis of LDG module B GSVA enrichment scores in GSE49454 WB showed only a minimal association between LDG enrichment and neutrophil count (r = 0.45, p = 0.0015), which lost significance when patients with extremely high or low neutrophil counts were excluded (r = 0.22, p = 0.18) (Supplemental Fig. 2). These results implied that module B genes did not reflect either neutrophilia or neutrophil activation, and therefore module B was chosen to query blood and tissue gene expression data for the presence of LDGs.

Module A contained many genes associated with intracellular signaling as well as genes specific for platelets that may have been coisolated with LDGs during separation (Supplemental Material File 2). In GSE49454 WB, module A enrichment scores showed a correlation with platelet counts (r = 0.33, p = 0.02) and no correlation with neutrophil counts (p > 0.6) (Supplemental Fig. 2). Although this module of genes could be informative for studying the biology of LDGs, it was not sufficiently specific to query blood and tissue expression data for the presence of LDGs.

We next wanted to determine whether the genes in LDG module B were coexpressed cohesively from patient to patient. WGCNA was used to construct ME for the module B genes in six WB and PBMC gene expression datasets as well as datasets from LN glomerulus and TI, skin, and synovium. The kME was used to assess the quality of the gene expression module in each dataset. To obtain points of reference, COXPRESdb (https://coxpresdb.jp) was queried for genes coexpressed with CD79A and ZAP70 and thus associated with BCR signaling and TCR signaling, respectively (Supplemental Material File 4). In the original LDG expression data, the module B genes were considered the standard, with a mean kME of 0.67. In blood datasets, the mean kME of the module B genes had a range of 0.41–0.52, with a grand mean of 0.46 (Table I). This was considered acceptable with regard to the original LDG expression data. For reference, the CD79A (BCR) and ZAP70 (TCR) modules exhibited grand mean kMEs across blood datasets of 0.55 and 0.65, respectively. In tissue datasets, however, the module B genes had mean kMEs ranging from 0.02 to 0.24, with a grand mean of 0.14, whereas the ZAP70 (TCR) and CD79A (BCR) modules both had grand mean kMEs of 0.72 (Table II). These results indicate that the module B genes acted as a cohesive module in blood expression data but not in tissue data. This implies that LDGs defined by module B expression are not present in the tissues, but further testing was done to assess this assumption.

Table I.
LDG module B genes are coexpressed in SLE blood
StudyModule B Mean kMECD79A COXPRESdb Mean kMEZAP70 COXPRESdb Mean kME
GSE49454 WB 0.51 0.58 0.66 
GSE88884 ILLUMINATE-1 WB 0.44 0.52 0.67 
GSE88884 ILLUMINATE-2 WB 0.41 0.51 0.66 
GSE50772 PBMC 0.46 0.64 0.69 
GSE81622 PBMC 0.52 0.44 0.56 
FDABMC3 PBMC 0.42 0.62 0.69 
Blood grand mean 0.46 0.55 0.65 
GSE26975 LDGs 0.67 NA NA 
StudyModule B Mean kMECD79A COXPRESdb Mean kMEZAP70 COXPRESdb Mean kME
GSE49454 WB 0.51 0.58 0.66 
GSE88884 ILLUMINATE-1 WB 0.44 0.52 0.67 
GSE88884 ILLUMINATE-2 WB 0.41 0.51 0.66 
GSE50772 PBMC 0.46 0.64 0.69 
GSE81622 PBMC 0.52 0.44 0.56 
FDABMC3 PBMC 0.42 0.62 0.69 
Blood grand mean 0.46 0.55 0.65 
GSE26975 LDGs 0.67 NA NA 

Mean kMEs, the correlation of each gene in module B with the overall eigengene, are shown for blood gene expression data. Mean kME values in LDG gene expression data are shown for reference as well as mean kME values for genes coexpressed with CD79A and ZAP70 as determined from COXPRESdb.

Table II.
LDG module B genes are not coexpressed in SLE-affected tissues
StudyModule B Mean kMENeutrophil Degranulation Mean kMECell Cycle Mean kMEOther Mean kMECD79A COXPRESdb Mean kMEZAP70 COXPRESdb Mean kME
GSE32591 kidney glomerulus 0.21 0.25 0.73 0.12 0.77 0.70 
GSE32591 kidney TI 0.24 0.10 0.47 0.27 0.50 0.47 
GSE52471 skin 0.19 0.09 0.54 0.17 0.84 0.82 
GSE72535 skin 0.05 0.04 0.55 0.23 0.81 0.89 
GSE36700 synovium 0.02 0.29 0.62 0.09 0.66 0.74 
Tissue grand mean 0.14 0.15 0.58 0.18 0.72 0.72 
GSE26975 LDGs 0.67 0.83 0.73 0.61 NA NA 
StudyModule B Mean kMENeutrophil Degranulation Mean kMECell Cycle Mean kMEOther Mean kMECD79A COXPRESdb Mean kMEZAP70 COXPRESdb Mean kME
GSE32591 kidney glomerulus 0.21 0.25 0.73 0.12 0.77 0.70 
GSE32591 kidney TI 0.24 0.10 0.47 0.27 0.50 0.47 
GSE52471 skin 0.19 0.09 0.54 0.17 0.84 0.82 
GSE72535 skin 0.05 0.04 0.55 0.23 0.81 0.89 
GSE36700 synovium 0.02 0.29 0.62 0.09 0.66 0.74 
Tissue grand mean 0.14 0.15 0.58 0.18 0.72 0.72 
GSE26975 LDGs 0.67 0.83 0.73 0.61 NA NA 

Mean kMEs, the correlation of each gene in module B with the overall eigengene, are shown for tissue gene expression data. Mean kME values in LDG gene expression data are shown for reference as well as mean kME values for genes coexpressed with CD79A and ZAP70 as determined from COXPRESdb. Submodules based on GO designations display different behaviors from the module as a whole.

Module B was further broken down into three submodules of genes based on GO designations to account for the possibility that the noise of the tissue environment could be masking the behavior of the module. The three submodules were made up of neutrophil degranulation genes, cell cycle genes, and genes that did not have either designation (other) (Supplemental Material File 2). In the original LDG expression data, the neutrophil degranulation, cell cycle, and other submodules had mean kMEs of 0.83, 0.73, and 0.61, respectively. Across tissue datasets, they had grand mean kMEs of 0.15, 0.58, and 0.18, respectively (Table II). These results show that cell cycle–related genes behave cohesively in the tissues, but the rest of the genes in module B do not, suggesting that cells other than LDGs convey the signature of cell cycle–related genes in lupus tissues. Overall, the gene coexpression results indicate that although LDGs are enriched in the blood of SLE patients, LDGs are not enriched in SLE-affected organs. Because the LDG module B genes were coexpressed in blood but not in tissue, further analyses were carried out to evaluate the presence of LDGs in SLE peripheral blood.

GSVA was used to query lupus WB gene expression data from GSE88884 for the enrichment of LDG module B genes in 1612 SLE patients. GSVA was performed separately on the data derived from the two clinical trials (ILLUMINATE-1 and ILLUMINATE-2) contained within this dataset. LDG enrichment was modestly but significantly correlated with increasing SLEDAI (Spearman rho = 0.192, p = 6.59 × 10−15). Welch unequal variances t test was used to determine whether LDG enrichment scores were significantly different in patients with and without each component of the SLEDAI score or patients receiving any of four classes of drugs (Table III). LDG enrichment was significantly greater in patients with anti-dsDNA seropositivity (p = 2.14 × 10−25), those with low serum complement (p = 9.02 × 10−23), and those taking corticosteroids (p = 1.26 × 10−33). LDG enrichment was also greater in patients with hematuria, proteinuria, pyuria, pericarditis, vasculitis, or leukopenia and those taking immunosuppressives (all p < 0.05). LDG enrichment was decreased in patients taking nonsteroidal anti-inflammatory drugs (NSAIDs) or antimalarials and those with arthritis or mucosal ulcers (all p < 0.05).

Table III.
LDG enrichment is associated with treatment, SLE disease manifestations, and SLEDAI
Estimatet-Statisticp Value
Antimalarials (n = 1091) 0.041 −2.24 0.025 
Corticosteroids (n = 1184) 0.212 12.6 1.26 × 10−33 
Immunosuppressants (n = 670) 0.067 3.90 1.00 × 10−4 
NSAIDs (n = 506) −0.086 −4.83 1.59 × 10−6 
Alopecia (n = 982) −0.001 −0.06 0.956 
Anti-dsDNA (n = 953) 0.174 10.6 2.14 × 10−25 
Arthritis (n = 1413) 0.103 −4.00 8.17 × 10−5 
Fever (n = 31) 0.091 1.33 0.195 
Hematuria (n = 44) 0.128 2.44 0.019 
Leukopenia (n = 125) 0.098 3.27 1.32 × 10−3 
Low complement (n = 748) 0.164 9.98 9.02 × 10−23 
Mucosal ulcers (n = 563) 0.042 −2.38 0.018 
Myositis (n = 17) −0.067 −0.83 0.420 
Pericarditis (n = 28) 0.140 2.23 0.034 
Pleurisy (n = 110) 0.018 0.53 0.597 
Proteinuria (n = 46) 0.145 2.81 7.18 × 10−3 
Pyuria (n = 79) 0.184 4.92 4.11 × 10−6 
Rash (n = 1133) 0.020 1.10 0.272 
Thrombocytopenia (n = 29) 0.087 1.23 0.230 
Vasculitis (n = 119) 0.137 0.67 0.549 
Visual disturbance (n = 26) 0.119 3.66 3.63 × 10−4 
SLEDAI (range 6–40, mean 10.4 ± 3.8) 0.192  6.59 × 10−15 
Estimatet-Statisticp Value
Antimalarials (n = 1091) 0.041 −2.24 0.025 
Corticosteroids (n = 1184) 0.212 12.6 1.26 × 10−33 
Immunosuppressants (n = 670) 0.067 3.90 1.00 × 10−4 
NSAIDs (n = 506) −0.086 −4.83 1.59 × 10−6 
Alopecia (n = 982) −0.001 −0.06 0.956 
Anti-dsDNA (n = 953) 0.174 10.6 2.14 × 10−25 
Arthritis (n = 1413) 0.103 −4.00 8.17 × 10−5 
Fever (n = 31) 0.091 1.33 0.195 
Hematuria (n = 44) 0.128 2.44 0.019 
Leukopenia (n = 125) 0.098 3.27 1.32 × 10−3 
Low complement (n = 748) 0.164 9.98 9.02 × 10−23 
Mucosal ulcers (n = 563) 0.042 −2.38 0.018 
Myositis (n = 17) −0.067 −0.83 0.420 
Pericarditis (n = 28) 0.140 2.23 0.034 
Pleurisy (n = 110) 0.018 0.53 0.597 
Proteinuria (n = 46) 0.145 2.81 7.18 × 10−3 
Pyuria (n = 79) 0.184 4.92 4.11 × 10−6 
Rash (n = 1133) 0.020 1.10 0.272 
Thrombocytopenia (n = 29) 0.087 1.23 0.230 
Vasculitis (n = 119) 0.137 0.67 0.549 
Visual disturbance (n = 26) 0.119 3.66 3.63 × 10−4 
SLEDAI (range 6–40, mean 10.4 ± 3.8) 0.192  6.59 × 10−15 

“Estimate” denotes the change in LDG enrichment score or Spearman rho (SLEDAI only). Urinary casts, organic brain syndrome, lupus headache, seizure, psychosis, cranial nerve disorder, and cerebrovascular accidents appeared in fewer than five patients each and were excluded from this analysis. Significant estimates are bolded (p < 0.05).

Based on the results of these tests and trends in the current literature, a smaller panel of characteristics was selected to study in more depth. Corticosteroid treatment was used to divide patients, as it appeared to have strong effects on LDG enrichment. In addition, anti-dsDNA and low complement were selected as manifestations of interest because of their strong associations with LDG enrichment. Vasculitis and the presence of any renal manifestation (proteinuria, hematuria, pyuria, or urinary casts) were also selected for further analysis. Although their p values were modest compared with those of other characteristics, there is a body of recent literature exploring the links between neutrophil-like gene signatures and vasculitis or renal disease in lupus patients (7, 1618).

Welch unequal variances t test was used to determine whether LDG module B enrichment scores were significantly different in patient subpopulations with and without the manifestations of interest. Gene signatures from plasma cells and cytotoxic T cells/NK cells/NKT cells were used as positive and negative controls, respectively, as plasma cells should be clearly associated with anti-dsDNA, as previously reported (19), and cytotoxic cells are not known to be associated with any of the manifestations of interest (Supplemental Material File 5).

When using all patients, all manifestations of interest were significantly associated with increases in the LDG enrichment score (p < 0.001) (Table IV). Among corticosteroid users, results closely resembled those acquired with all patients. Among corticosteroid nonusers, anti-dsDNA (p = 1.10 × 10−4) and low complement (p = 3.36 × 10−4) remained modestly associated with increased LDG enrichment scores, whereas vasculitis and renal manifestations were no longer associated with increased enrichment scores (p > 0.3). Similar tests with other drugs, including antimalarials, immunosuppressives, and NSAIDs, showed that overall associations between LDG enrichment and SLE manifestations were only minimally affected by the presence or absence of these classes of drugs, with the exception of NSAIDs in patients with renal manifestations (Supplemental Table I).

Table IV.
LDG enrichment is associated with different manifestations depending on corticosteroid treatment
CorticosteroidsEstimatet-Statisticp Value
Anti-dsDNA All patients 0.174 10.6 2.14 × 10−25 
Yes 0.144 7.07 3.28 × 10−12 
No 0.111 3.92 1.10 × 10−4 
Low complement All patients 0.164 9.98 9.02 × 10−23 
Yes 0.131 6.67 3.92 × 10−11 
No 0.113 3.65 3.36 × 10−4 
Renal manifestations All patients 0.142 4.58 9.44 × 10−6 
Yes 0.125 3.75 2.62 × 10−4 
No 0.045 0.69 0.497 
Vasculitis All patients 0.120 3.66 3.63 × 10−4 
Yes 0.115 3.27 1.42 × 10−3 
No 0.065 0.95 0.351 
CorticosteroidsEstimatet-Statisticp Value
Anti-dsDNA All patients 0.174 10.6 2.14 × 10−25 
Yes 0.144 7.07 3.28 × 10−12 
No 0.111 3.92 1.10 × 10−4 
Low complement All patients 0.164 9.98 9.02 × 10−23 
Yes 0.131 6.67 3.92 × 10−11 
No 0.113 3.65 3.36 × 10−4 
Renal manifestations All patients 0.142 4.58 9.44 × 10−6 
Yes 0.125 3.75 2.62 × 10−4 
No 0.045 0.69 0.497 
Vasculitis All patients 0.120 3.66 3.63 × 10−4 
Yes 0.115 3.27 1.42 × 10−3 
No 0.065 0.95 0.351 

Shown are t test results for manifestations of interest in patients grouped by corticosteroid treatment. “Estimate” denotes the change in LDG enrichment score. “Renal manifestations” denote at least one of hematuria, proteinuria, pyuria, or urinary casts. The subset of patients not taking corticosteroids did not show significant differences in LDG enrichment related to renal manifestations or vasculitis. Significant estimates are bolded (p < 0.05).

As expected, plasma cell enrichment was strongly associated with anti-dsDNA irrespective of corticosteroid treatment, and cytotoxic T cell/NK cell/NKT cell enrichment was not associated with any manifestations of interest, save for a mild association with renal manifestations among corticosteroid users (p = 0.015) (Supplemental Tables II, III).

Further analyses of the links between LDG enrichment and disease manifestations among different patient populations were undertaken to determine whether binary (yes/no) enrichment of LDGs could be used as a diagnostic or proxy test for other clinical traits or gene signatures potentially involved in SLE pathogenesis. WB gene expression data from GSE88884, including HC subjects, were analyzed with GSVA as described above, using LDG module B, the IFN gene signature (IGS), and genes induced by TNF (20). Patients with a z-score >2 relative to controls were considered positive for differential enrichment of the gene signature in question. LDG differential enrichment was compared with available clinical traits and the IGS and TNF signatures in all patients and in the previously mentioned subgroups based on corticosteroid treatment. Testing for associations between LDG differential enrichment and traits was done by Fisher exact test.

Differential LDG enrichment was found in 55% (891 of 1612) of SLE patients, IGS differential enrichment was found in 75% (1216 of 1612) of patients, and TNF response differential enrichment was found in 44% (704 of 1612) of patients. Strong associations with LDG differential enrichment were found for IGS and the TNF response in all patients and in both subgroups of patients by Fisher exact test (p < 1 × 10−10) (Table V). Remarkably, LDG enrichment and TNF response had the strongest association in patients not taking corticosteroids, with an odds ratio of 8.3. Associations between LDG differential enrichment and clinical traits of interest were similar to those found by t tests, as LDG differential enrichment was not associated with renal manifestations or vasculitis in patients not taking corticosteroids (Supplemental Table IV).

Table V.
LDG differential enrichment is associated with the IGS and genes induced by TNF
CorticosteroidsOdds Ratio (95% CI)p Value
IGS All patients 5.0 (3.9, 6.6) <2.2 × 10−16 
Yes 4.4 (3.2, 6.1) <2.2 × 10−16 
No 5.0 (3.0, 8.4) 1.37 × 10−12 
TNF All patients 7.5 (5.9, 9.5) <2.2 × 10−16 
Yes 7.0 (5.3, 9.4) <2.2 × 10−16 
No 8.3 (5.2, 13.4) <2.2 × 10−16 
CorticosteroidsOdds Ratio (95% CI)p Value
IGS All patients 5.0 (3.9, 6.6) <2.2 × 10−16 
Yes 4.4 (3.2, 6.1) <2.2 × 10−16 
No 5.0 (3.0, 8.4) 1.37 × 10−12 
TNF All patients 7.5 (5.9, 9.5) <2.2 × 10−16 
Yes 7.0 (5.3, 9.4) <2.2 × 10−16 
No 8.3 (5.2, 13.4) <2.2 × 10−16 

Fisher exact test results in patients grouped by corticosteroid treatment.

CI, confidence interval.

Samples from GSE19556 were used to compare PM, MY, and bone marrow PMN (bmPMN) to peripheral blood PMN by DE analysis (Supplemental Material File 6, Table VI). A total of 68 of the 92 genes in LDG module B were differentially expressed in PM (overlap p value = 1.4 × 10−6) compared with 71 in MY (p = 1.8 × 10−18) and 28 in bmPMN (p = 8.5 × 10−12). In contrast, bmPMN did not differentially express the cell cycle portion of module B found in PM and MY, indicating that LDGs are transcriptionally similar to these more immature precursors.

Table VI.
LDG module B is enriched in neutrophil precursors normally found in the bone marrow
Cell TypeUpregulated Genes versus pbPMNUpregulated Genes in LDG Module BOverlap p ValueNeutrophil Degranulation Genes in OverlapCell Cycle Genes in Overlap
PM 4951 68 of 92 1.4 × 10−6 25 of 30 18 of 21 
MY 3267 71 of 92 1.8 × 10−18 28 of 30 18 of 21 
bmPMN 690 28 of 92 8.5 × 10−12 20 of 30 2 of 21 
Cell TypeUpregulated Genes versus pbPMNUpregulated Genes in LDG Module BOverlap p ValueNeutrophil Degranulation Genes in OverlapCell Cycle Genes in Overlap
PM 4951 68 of 92 1.4 × 10−6 25 of 30 18 of 21 
MY 3267 71 of 92 1.8 × 10−18 28 of 30 18 of 21 
bmPMN 690 28 of 92 8.5 × 10−12 20 of 30 2 of 21 

Overlap p values were calculated using Fisher exact test and a universe of 10,000 genes to account for the fact that low-intensity genes were filtered out in both experiments, and this results in more conservative p values than the use of 20,000 genes. bmPMN differentially express only a small portion of the cell cycle signature compared with PM and MY.

pbPMN, peripheral blood PMN.

Analysis of LDGs, SLE neutrophils, and HC neutrophils revealed hundreds of genes significantly differentially expressed by LDGs and initially identified granulopoietic and proliferative signatures as potentially descriptive of LDGs. Given that circulating neutrophils do not express granulopoietic genes and that SLE neutrophils did not differentially express any genes relative to HC neutrophils, it was posited that the detection of these signatures in SLE blood could be attributed to LDGs (14). However, the DE approach seemed to be challenged by contamination from platelets and lymphocytes. LDGs were isolated from PBMC by negative selection, using a mixture of biotinylated Abs to human CD molecules, whereas HC and SLE neutrophils were isolated by dextran sedimentation of RBC pellets (8, 12). Although the purity of LDG and neutrophil isolates was high, the low baseline levels of transcription in neutrophils may allow even small amounts of contamination to affect microarray results strongly, so further refinement was needed to extract a robust LDG gene expression signature.

The coexpression-based unsupervised clustering method of WGCNA was able to dissect the gene expression landscape down into several modules of genes that separated LDG samples and neutrophil samples. One of these modules captured what seemed to be the pattern of lymphocyte contamination in the original expression data, and another set of modules, which were merged to form module A, contained many of the platelet genes identified in the original DE analysis. Functional analysis narrowed the WGCNA modules down to one final module of genes, which contained neutrophil granule genes and cell cycle regulation genes. The presence of granule genes indicated that the module is neutrophil lineage–specific, whereas the presence of cell cycle genes after coexpression network construction suggested that the cell cycle signature is likely descriptive of LDGs and not an artifact of the isolation protocol. The combination of neutrophil lineage–specific granule genes along with cell cycle genes appears to identify the unique signature of LDGs. This module of genes was strongly coexpressed in SLE blood expression data but not in lupus-affected tissue, including LN glomerulus, TI, lupus skin, and synovium, which indicated that the LDG gene expression signature can be recovered from blood but not from tissue. Although netting neutrophils have been described in SLE-affected glomerulus and skin, the current results suggest that infiltrating neutrophils are either normal-density neutrophils or LDGs with an altered transcriptional program (8). More studies will be necessary to shed light on this, as LDGs did not differentially express any homing receptors or activation markers associated with the ability to infiltrate tissues.

It was initially surprising not to find transcriptional evidence for LDGs in SLE-affected kidneys or a strong association between LDG enrichment and renal involvement, as a similar group of neutrophil genes was previously reported to be enriched in the blood of LN patients compared with lupus patients without nephritis (17). In that work, the claim of an association with neutrophils was based on a gene module, M5.15, derived from modular repertoire analysis and consisting of 24 neutrophil-specific genes, 14 of which overlap with LDG module B. Notably, both LDG module B and M5.15 contain a core signature of 10 granulopoiesis-related genes that are not part of the endotoxemia-induced neutrophil activation signature described previously (AZU1, CAMP, CEACAM6, CEACAM8, CTSG, DEFA4, ELANE, LTF, MPO, MS4A3). This would suggest that module M5.15 may not describe neutrophil activation but rather the presence of LDGs. One limitation of the current study is that the presence of rapidly progressive or severe renal disease excluded patients from the ILLUMINATE trials, so an association of active renal disease with enrichment of LDGs may have been missed. Therefore, enrichment of LDG genes cannot yet be ruled out as a potential biomarker for LN. It is notable that an association between the LDG signature in the blood and renal involvement in the current study was only noted in those receiving corticosteroids. Whether the usage of corticosteroids is a surrogate for disease activity in this circumstance cannot be further delineated, but it does suggest that LDG module B and similar signatures may be of diagnostic use to identify those with LN only in the subset of patients taking corticosteroids.

By taking a large-scale transcriptomics approach to quantify the enrichment of the LDG signature in SLE blood gene expression data, it was possible to draw associations between LDG enrichment and clinical measurements of disease manifestation by studying both relative enrichment scores and binary LDG enrichment. LDG enrichment was associated with increased disease activity estimated by SLEDAI, decreased complement levels, and the presence of anti-dsDNA, suggesting that LDGs can act as markers of serological disease activity. Because complement levels and anti-dsDNA are components of the SLEDAI score, it is possible that these measurements account for the association with increased SLEDAI, as the associations with anti-dsDNA and low complement were stronger than the association with SLEDAI score.

The association between corticosteroid use and LDG enrichment was notable. Patients taking corticosteroids had significantly higher LDG enrichment than those not taking corticosteroids, and some disease manifestations were only associated with LDG enrichment in patients taking corticosteroids. It is unknown at this time whether increased LDG enrichment among patients using corticosteroids is related to increased granulopoiesis in the bone marrow or demargination of LDGs from the endothelium. Most literature suggests that the major effect of corticosteroids on distribution of cells of the neutrophil lineage relates to demargination, although this is not known for LDGs (21). However, the findings suggest that at least one component of the appearance of increased LDGs in the blood of lupus patients relates to corticosteroid-induced demargination. It has been suggested that LDGs play a role in SLE vascular pathology (22). It is possible, therefore, that LDGs home to the endothelium and contribute to local vascular inflammation. In this situation, corticosteroid-induced demargination could be therapeutically useful by dissociating LDGs from the vascular endothelium. The relationship between circulating LDGs and vascular pathology may be complex, and a better understanding of whether corticosteroid use stimulates LDG production or alternatively causes demargination of LDGs is therefore essential to resolve this conundrum.

The presence of LDG-specific genes in bone marrow myeloid precursors supports the hypothesis that LDGs are related to early neutrophil precursors (PM or MY) released from the bone marrow in response to cytokine challenge. A recent report has suggested that there may be two populations of LDGs in tumor-bearing mice and humans: one originating from the bone marrow and the second from peripheral neutrophils as a result of TGF-β stimulation (23). Similar to that report, we found that LDGs overexpress CD66b (CEACAM8), but we found no evidence of upregulation of the TGF-β signaling pathway (data not shown). These results are most consistent with the conclusion that the LDGs expanded in SLE are most similar to early neutrophil precursors and not TGF-β–stimulated mature neutrophils. Taken together with the strong association between LDG enrichment and TNF response, these results suggest that another component of the increased appearance of LDGs in the blood of lupus patients could relate to their enhanced release from the bone marrow as a result of chronic TNF-induced production of G-CSF (24, 25). The associations between LDG enrichment and both low complement levels (indicative of complement consumption, presumably owing to the presence of immune complexes) and a TNF response suggest that LDGs are part of an acute phase-like response in SLE (26, 27). Autoantibodies to dsDNA were present in ∼73% of patients with positive LDG enrichment, and an IFN signature was seen in 98% of patients with LDGs. These results are consistent with a role for autoantibodies and/or autoantibody containing immune complexes in the appearance of LDGs in the circulation either directly or through the induction of cytokines, such as type I IFN or TNF. Alternatively, LDGs could play a role in the induction of autoantibodies, as LDG NETs may be autoantigenic and interferogenic (10). Further experiments will be necessary to sort out these possibilities.

The current studies analyzed bulk RNA from blood and various lupus-affected tissues and, as a result, could not explore the possible heterogeneity of LDGs at the single-cell level. Single-cell transcriptomic studies of LDGs in SLE may be useful to further elucidate the characteristics of this cell population and whether a related population is present in lupus-affected tissues. A deeper understanding of any subtypes of LDGs and how they differ in composition among SLE patients may offer unique insights into disease processes and therapeutic options for patients with circulating LDGs.

The results presented in this paper suggest that LDGs are not directly involved in inflammation in SLE-affected organs, but they may act as biomarkers of processes that can in parallel result in tissue damage or vascular damage. As LDGs are associated with anti-dsDNA, low serum complement, and the presence of an IGS, they may indirectly lead to increasingly severe disease in afflicted patients. However, one cannot dismiss the possibility that factors such as treatment regimens may contribute to the presence of LDGs because of their association with increased disease activity, highlighting the complexity of the association of LDGs with disease manifestations in SLE. Further study of LDGs will help us to understand the links between corticosteroid treatment, LDG enrichment, and SLE pathogenesis.

We are grateful to M.D. Linnik for providing access to the ILLUMINATE datasets.

This work was supported by the RILITE Foundation.

The online version of this article contains supplemental material.

Abbreviations used in this article:

BA

BrainArray

bmPMN

bone marrow PMN

CDF

chip definition file

DE

differential expression

DEG

differentially expressed gene

E-set

R expression set object

FDR

false discovery rate

GO

Gene Ontology

GSVA

gene set variation analysis

HC

healthy control

IGS

IFN gene signature

kME

correlation coefficient of each gene in a module with the ME

LDG

low-density granulocyte

LN

lupus nephritis

ME

module eigengene

MY

myelocyte

NET

neutrophil extracellular trap

NSAID

nonsteroidal anti-inflammatory drug

PM

promyelocyte

PMN

polymorphonuclear neutrophil

SLE

systemic lupus erythematosus

SLEDAI

SLE Disease Activity Index

TI

tubulointerstitium

WB

whole blood

WGCNA

weighted gene coexpression network analysis.

1
Hacbarth
,
E.
,
A.
Kajdacsy-Balla
.
1986
.
Low density neutrophils in patients with systemic lupus erythematosus, rheumatoid arthritis, and acute rheumatic fever.
Arthritis Rheum.
29
:
1334
1342
.
2
Wright
,
H. L.
,
F. A.
Makki
,
R. J.
Moots
,
S. W.
Edwards
.
2017
.
Low-density granulocytes: functionally distinct, immature neutrophils in rheumatoid arthritis with altered properties and defective TNF signalling.
J. Leukoc. Biol.
101
:
599
611
.
3
Cloke
,
T.
,
M.
Munder
,
G.
Taylor
,
I.
Müller
,
P.
Kropf
.
2012
.
Characterization of a novel population of low-density granulocytes associated with disease severity in HIV-1 infection.
PLoS One
7
:
e48939
.
4
Scapini
,
P.
,
O.
Marini
,
C.
Tecchio
,
M. A.
Cassatella
.
2016
.
Human neutrophils in the saga of cellular heterogeneity: insights and open questions.
Immunol. Rev.
273
:
48
60
.
5
Deng
,
Y.
,
J.
Ye
,
Q.
Luo
,
Z.
Huang
,
Y.
Peng
,
G.
Xiong
,
Y.
Guo
,
H.
Jiang
,
J.
Li
.
2016
.
Low-density granulocytes are elevated in mycobacterial infection and associated with the severity of tuberculosis.
PLoS One
11
:
e0153567
.
6
Rocha
,
B. C.
,
P. E.
Marques
,
F. M. S.
Leoratti
,
C.
Junqueira
,
D. B.
Pereira
,
L. R. D. V.
Antonelli
,
G. B.
Menezes
,
D. T.
Golenbock
,
R. T.
Gazzinelli
.
2015
.
Type I interferon transcriptional signature in neutrophils and low-density granulocytes are associated with tissue damage in malaria.
Cell Rep.
13
:
2829
2841
.
7
Jorch
,
S. K.
,
P.
Kubes
.
2017
.
An emerging role for neutrophil extracellular traps in noninfectious disease.
Nat. Med.
23
:
279
287
.
8
Villanueva
,
E.
,
S.
Yalavarthi
,
C. C.
Berthier
,
J. B.
Hodgin
,
R.
Khandpur
,
A. M.
Lin
,
C. J.
Rubin
,
W.
Zhao
,
S. H.
Olsen
,
M.
Klinker
, et al
.
2011
.
Netting neutrophils induce endothelial damage, infiltrate tissues, and expose immunostimulatory molecules in systemic lupus erythematosus.
J. Immunol.
187
:
538
552
.
9
Brinkmann
,
V.
,
U.
Reichard
,
C.
Goosmann
,
B.
Fauler
,
Y.
Uhlemann
,
D. S.
Weiss
,
Y.
Weinrauch
,
A.
Zychlinsky
.
2004
.
Neutrophil extracellular traps kill bacteria.
Science
303
:
1532
1535
.
10
Lood
,
C.
,
L. P.
Blanco
,
M. M.
Purmalek
,
C.
Carmona-Rivera
,
S. S.
De Ravin
,
C. K.
Smith
,
H. L.
Malech
,
J. A.
Ledbetter
,
K. B.
Elkon
,
M. J.
Kaplan
.
2016
.
Neutrophil extracellular traps enriched in oxidized mitochondrial DNA are interferogenic and contribute to lupus-like disease.
Nat. Med.
22
:
146
153
.
11
Bennett
,
L.
,
A. K.
Palucka
,
E.
Arce
,
V.
Cantrell
,
J.
Borvak
,
J.
Banchereau
,
V.
Pascual
.
2003
.
Interferon and granulopoiesis signatures in systemic lupus erythematosus blood.
J. Exp. Med.
197
:
711
723
.
12
Denny
,
M. F.
,
S.
Yalavarthi
,
W.
Zhao
,
S. G.
Thacker
,
M.
Anderson
,
A. R.
Sandy
,
W. J.
McCune
,
M. J.
Kaplan
.
2010
.
A distinct subset of proinflammatory neutrophils isolated from patients with systemic lupus erythematosus induces vascular damage and synthesizes type I IFNs.
J. Immunol.
184
:
3284
3297
.
13
Langfelder
,
P.
,
S.
Horvath
.
2008
.
WGCNA: an R package for weighted correlation network analysis.
BMC Bioinformatics
9
:
559
.
14
Borregaard
,
N.
,
O. E.
Sørensen
,
K.
Theilgaard-Mönch
.
2007
.
Neutrophil granules: a library of innate immunity proteins.
Trends Immunol.
28
:
340
345
.
15
de Kleijn
,
S.
,
M.
Kox
,
I. E.
Sama
,
J.
Pillay
,
A.
van Diepen
,
M. A.
Huijnen
,
J. G.
van der Hoeven
,
G.
Ferwerda
,
P. W. M.
Hermans
,
P.
Pickkers
.
2012
.
Transcriptome kinetics of circulating neutrophils during human experimental endotoxemia.
PLoS One
7
:
e38255
.
16
Banchereau
,
R.
,
S.
Hong
,
B.
Cantarel
,
N.
Baldwin
,
J.
Baisch
,
M.
Edens
,
A. M.
Cepika
,
P.
Acs
,
J.
Turner
,
E.
Anguiano
, et al
.
2016
.
Personalized immunomonitoring uncovers molecular networks that stratify lupus patients. [Published erratum appears in 2016 Cell 165: 1548–1550.]
Cell
165
:
551
565
.
17
Jourde-Chiche
,
N.
,
E.
Whalen
,
B.
Gondouin
,
C.
Speake
,
V.
Gersuk
,
B.
Dussol
,
S.
Burtey
,
V.
Pascual
,
D.
Chaussabel
,
L.
Chiche
.
2017
.
Modular transcriptional repertoire analyses identify a blood neutrophil signature as a candidate biomarker for lupus nephritis.
Rheumatology (Oxford)
56
:
477
487
.
18
Lewandowski
,
L. B.
,
M. J.
Kaplan
.
2016
.
Update on cardiovascular disease in lupus.
Curr. Opin. Rheumatol.
28
:
468
476
.
19
Jacobi
,
A. M.
,
M.
Odendahl
,
K.
Reiter
,
A.
Bruns
,
G. R.
Burmester
,
A.
Radbruch
,
G.
Valet
,
P. E.
Lipsky
,
T.
Dörner
.
2003
.
Correlation between circulating CD27high plasma cells and disease activity in patients with systemic lupus erythematosus.
Arthritis Rheum.
48
:
1332
1342
.
20
Waddell
,
S. J.
,
S. J.
Popper
,
K. H.
Rubins
,
M. J.
Griffiths
,
P. O.
Brown
,
M.
Levin
,
D. A.
Relman
.
2010
.
Dissecting interferon-induced transcriptional programs in human peripheral blood cells.
PLoS One
5
:
e9753
.
21
Nakagawa
,
M.
,
T.
Terashima
,
Y.
D’yachkova
,
G. P.
Bondy
,
J. C.
Hogg
,
S. F.
van Eeden
.
1998
.
Glucocorticoid-induced granulocytosis: contribution of marrow release and demargination of intravascular granulocytes.
Circulation
98
:
2307
2313
.
22
Carlucci
,
P. M.
,
M. M.
Purmalek
,
A. K.
Dey
,
Y.
Temesgen-Oyelakin
,
S.
Sakhardande
,
A. A.
Joshi
,
J. B.
Lerman
,
A.
Fike
,
M.
Davis
,
J. H.
Chung
, et al
.
2018
.
Neutrophil subsets and their gene signature associate with vascular inflammation and coronary atherosclerosis in lupus.
JCI Insight
3
:
e99276
.
23
Sagiv
,
J. Y.
,
J.
Michaeli
,
S.
Assi
,
I.
Mishalian
,
H.
Kisos
,
L.
Levy
,
P.
Damti
,
D.
Lumbroso
,
L.
Polyansky
,
R. V.
Sionov
, et al
.
2015
.
Phenotypic diversity and plasticity in circulating neutrophil subpopulations in cancer.
Cell Rep.
10
:
562
573
.
24
Christensen
,
A. D.
,
C.
Haase
,
A. D.
Cook
,
J. A.
Hamilton
.
2016
.
Granulocyte colony-stimulating factor (G-CSF) plays an important role in immune complex-mediated arthritis.
Eur. J. Immunol.
46
:
1235
1245
.
25
Semerad
,
C. L.
,
F.
Liu
,
A. D.
Gregory
,
K.
Stumpf
,
D. C.
Link
.
2002
.
G-CSF is an essential regulator of neutrophil trafficking from the bone marrow to the blood.
Immunity
17
:
413
423
.
26
Walport
,
M. J.
2002
.
Complement and systemic lupus erythematosus.
Arthritis Res.
4
(
Suppl. 3
):
S279
S293
.
27
Gruys
,
E.
,
M. J. M.
Toussaint
,
T. A.
Niewold
,
S. J.
Koopmans
.
2005
.
Acute phase reaction and acute phase proteins.
J. Zhejiang Univ. Sci. B
6
:
1045
1056
.

B.J.K., P.E.L., and A.C.G. have filed a provisional patent application on technology described in this paper. The other authors have no financial conflicts of interest.