Skip to main content

Normalized linear counts from NanoString autoimmune profiling panel and summary of statistical analyses


Heine, Lauren (2022), Normalized linear counts from NanoString autoimmune profiling panel and summary of statistical analyses, Dryad, Dataset,


Though dependent on genetic anomalies, clinical manifestations of the human autoimmune disease systemic lupus erythematosus (lupus) can be triggered by environmental exposures including inhalation toxicants such as crystalline silica dust (cSiO2), tobacco smoke, and ambient air particles.  Prednisone, a glucocorticoid (GC), is a keystone therapy for managing lupus flaring and progression, however, long-term use is associated with many adverse side effects. Here, we characterized the dose-dependent immunomodulation and toxicity of prednisone in a preclinical model that emulates onset and progression of cSiO2-triggered lupus. Two cohorts of 6-wk-old female NZBWF1 mice were fed either control AIN-93G diet or one of three AIN-93G diets containing prednisone at 5, 15, or 50 mg/kg diet which span human equivalent oral doses (HED) currently considered to be low (PL; 5 mg/d HED), moderate (PM; 14 mg/d HED), or high (PH; 46 mg/d HED), respectively. At 8 wk of age, mice were intranasally instilled with either saline vehicle or 1 mg cSiO2 once weekly for 4 wk. The experimental plan was to 1) terminate one cohort of mice (n=8/group) 14 wk after the last cSiO2 instillation for pathology and autoimmunity assessment and 2) to maintain a second cohort (n=9/group) to monitor glomerulonephritis development and survival. Mean blood concentrations of prednisone’s chief active metabolite, prednisolone, in mice fed PL, PM, and PH diets were 27, 105, 151 ng/ml, respectively, which are consistent with levels observed in human blood ≤ 12 h after single bolus treatments with equivalent prednisone doses. Results from the first cohort revealed that consumption of PM but not PL diet significantly reduced cSiO2-induced pulmonary ectopic lymphoid structure formation, nuclear-specific AAb production, and inflammation/autoimmune gene expression in the lung, splenomegaly, and glomerulonephritis in the kidney. Relative to GC-associated toxicity, PM but not PL diet elicited muscle wasting, but these diets did not affect bone density or cause glucosuria. Importantly, neither PM nor PL diet influenced latency of cSiO2-accelerated death. PH-fed mice in both cohorts displayed robust GC-associated toxicity including body weight loss, reduced muscle mass, and hyperglycemia 7 wk after the final cSiO2 instillation requiring their early removal from the study. Taken together, our results demonstrate that while moderate doses of prednisone can reduce certain pathological endpoints of cSiO2-induced autoimmunity in lupus-prone mice, these ameliorative effects come with unwanted GC toxicity and, crucially, none of these three doses extended survival time.


NanoString Autoimmune Profiling

RNA was extracted from lungs, kidneys, and blood with RNeasy Mini Kits with DNase treatment (Qiagen, Valencia, CA). RNA was dissolved in nuclease-free water, quantified with Qubit (Thermo Fisher Scientific), and integrity verified with a TapeStation (Agilent Technologies). Samples (RNA integrity > 8) were analyzed with NanoString Autoimmune Gene Expression assay (XT-CSO-MAIP1-12, NanoString Technologies, Seattle, WA) at the MSU Genomics Core. Assays were performed and quantified on the nCounter MAX system, sample preparation station, and digital analyzer (NanoString Technologies) according to the manufacturer’s instructions.

Raw gene expression data were analyzed using NanoString’s software nSolver v3.0.22 with the Advanced Analysis Module v2.0. Background subtraction was performed using the eight negative controls included with the module. Genes with counts below a threshold of 2σ of the mean background signal were excluded from subsequent analysis. Data normalization was performed on background-subtracted samples using internal positive controls and selected housekeeping genes that were identified with the geNorm algorithm (

Differential gene expression analyses were performed using the nSolver Advanced Analysis Module, which employs several multivariate linear regression models (mixture negative binomial, simplified negative binomial, or log-linear model) to identify significant genes.  Resulting p values were adjusted using the Benjamini-Hochberg (BH) method to control the false discovery rate. A statistically significant difference in gene expression was defined as 1.5-fold change in expression (log2 > 0.58 or < -0.58) with BH q < 0.05.  Four pairwise comparisons within each time point for each tissue examined were determined a priori, as follows: cSiO2/P0 vs VEH/P0, cSiO2/PL vs cSiO2/P0, cSiO2/PM vs cSiO2/P0, and cSiO2/PM vs cSiO2/PL. Venn diagrams of significant differentially expressed genes were generated using BioVenn. 

To assess the impact of experimental diets on annotated gene sets, global and directed significance scores were calculated for each pathway at each time point. The global score estimates the cumulative evidence for the differential expression of genes in a pathway.  Directed significance scores near zero indicate that a pathway may have many highly regulated genes, but no apparent tendency for those genes to be over- or under-expressed collectively.  As a complementary method for comparing pathways and discriminating between experimental groups, pathway Z scores were calculated as the Z-scaled first principal component of the pathway genes’ normalized expression. ClustVis was used to perform unsupervised hierarchical cluster analyses (HCC) and principal components analyses (PCA) using log2 transcript count data for DEGs. Spearman rank correlations were performed to examine overall patterns in the gene expression profiles using the pathway Z score compared to other biomarkers of disease in lung or kidney tissues at 14 weeks PI.  A significant correlation was inferred when ρ > 0.5 or <-0.5 and p < 0.05. Network analyses for interactions among significant genes were performed using STRING database version 11.5 (, with a minimum interaction score > 0.05 and cluster identification using the Markov Cluster (MCL) algorithm with inflation parameter of 1.5. Networks generated by STRING were visualized with Cytoscape v. 3.9. 

The NanoString nSolver Advanced Analysis software employs the method described by Danaher to measure the abundance of various immune cell populations using marker genes that are expressed stably and specifically in particular cell types. Cell type scores were calculated as the average log-scale normalized expression of their characteristic genes. Relative cell type measurements were based on the total population of infiltrating lymphocytes, which is useful in a sample of heterogenous mix of cell types. Only cell types that exceeded the quality control analysis for correlation of marker gene expression are reported.

Statistical Analysis

All data were analyzed, and statistical tests were performed using Prism 9 (GraphPad Prism v 9.2, San Diego, CA) except for the NanoString gene expression data discussed above. Data were assessed for outliers using the Grubb’s outlier test (with Q = 1%) and for normality using the Shapiro-Wilk test (p < 0.01). Data of histopathological endpoints were analyzed using an unpaired one-tailed t-test to detect cSiO2-induced inflammation and autoimmunity in lupus-prone mice (VEH/P0 vs cSiO2/P0) and a One-Way ANOVA with Dunnett’s post-hoc test to address our hypothesis that dietary prednisone would dose-dependently suppress cSiO2-triggered responses (cSiO2/P0 vs cSiO2/PL or cSiO2/PM). Non-normal and semi-quantitative data were analyzed using the nonparametric Mann-Whitney U test (for VEH/P0 vs cSiO2/P0) and the nonparametric Kruskal-Wallis test with a Dunn’s post-hoc test (cSiO2/P0 vs cSiO2/PL or cSiO2/PM). Data are presented as mean ± standard error of the mean (SEM), with a p-value ≤ 0.05 being considered as statistically significant. 

Usage notes

Microsoft Excel


National Institute of Environmental Health Sciences, Award: ES027353