Immune cell transcript modules reveal leukocyte heterogeneity in synovial biopsies of seronegative spondylarthropathy patients

All data were publicly available and downloaded from the gene expression omnibus (GEO, http://​www.​ncbi.​nlm.​nih.​gov/​geo/​). The leukocyte (GSE1133 & GSE22886), mouse AS (GSE41039) and patient biopsy (GSE41038) datasets were downloaded and processed independently [14‐17]. For Affymetrix data, the raw .CEL files comprising each dataset were download and normalized using the Robust Multichip Algorithm (RMA) to generate probe set intensities [18]. For Illumina data, normalized data was downloaded directly from GEO. All human data were downloaded from a publicly available database, and were previously published. No new patients or samples were collected for this study.

PAM was downloaded as an excel add-in from http://​www-stat.​stanford.​edu/​~tibs/​PAM/​, and was used according to the available manual. Briefly, we used this algorithm to interrogate the human leukocyte series for the top 25 class genes where each different population of purified leukocytes was an independent class. Leukocyte class assignment was made as follows: Monocytes (GSM18871 & 18872), DCs (GSM18873 & 18874), NK cells (GSM18875 & 18876), CD4+ T cells (GSM18877 & 18878), CD8+ T cells (GSM18879 & 18880), and B cells (GSE18881 & 18882).

For cross-platform comparisons, datasets were collapsed to single genes based on highest mean average and mapped based on Unigene ID. To evaluate leukocyte metagenes, the expression values for each gene was transformed such that the mean and standard deviation were set to 0 and 1 in each dataset, respectively. A metagene score was calculated for each sample as follows:

where x is the transformed expression, n is the number of genes in class set P and P is the set of class marker genes for a particular metagene [19‐21].

T-tests were used to compare indices between control and spondylarthropathy samples in both murine and human datasets. All tests were two-sided and a p-value of 0.05 or less was considered statistically significant.

Whereas the underlying mechanisms behind AS and other SpA are known to be inflammatory in nature, the specific leukocyte cell populations that drive inflammation are unknown. Hence, we sought to measure infiltration of specific leukocyte cell populations within synovial biopsies of experimental and clinical models of AS and SpA. To this end, we identified leukocyte transcript modules, which could be implemented to measure leukocyte infiltration in patient biopsies for which the global gene expression profile is known. Based on publicly available global gene expression profiling data of the human transcriptome, we identified samples representing multiple purified leukocyte cell populations including monocytes, dendritic cells (DC), B cells, natural killer (NK) cells, CD4+ T cells and CD8+ T cells. We used PAM [22] to identify a gene expression based classifier that could discriminate between leukocyte cell type according to patterns in gene expression. Briefly, PAM is an efficient tool to identify molecular patterns that is applicable to gene expression data, and uses nearest shrunken centroids to identify subsets of genes that best characterize sample classes. We used PAM to identify a 150 gene classifier that comprised the top 25 marker genes associated with each leukocyte cell type (see Additional file 1: Table S1). This composite gene signature is known as a metagene, and gene clusters representing each leukocyte metagene is shown in Figure 1A. To test whether the expression of each leukocyte metagene was highly discriminatory between specific populations of leukocytes, we calculated immune cell metagene scores for each cell type across all cell types (Figures 1B and C). For example, the average monocyte metagene score was highest in the purified monocyte cell population, and negative in each other leukocyte population. Although we observed some cross-specificity between the CD4 and CD8 T cell signatures, the leukocyte metagene score was highest in the target cell population for each leukocyte cell type, supporting the overall validity of the approach. To further validate this approach we downloaded additional independent gene expression profiles from purified leukocyte populations and tested the capacity of the various metagenes to accurately discriminate between the cell types [17]. Importantly, the metagene signatures were sufficient to stratify these various leukocyte cell types (see Additional file 2: Figure S1). Based on these data, we concluded that leukocyte metagene signatures could be implemented to measure infiltration of specific leukocyte cell populations in patient biopsies.

Although the overarching objective of this work was to was identify populations of leukocytes that might mediate inflammation in human patients with either AS or SpA, we first sought to examine the expression of leukocyte metagenes in a mouse model of AS. In this case, the mouse model offers several advantages over the human samples, including samples being harvested from mice with identical genetic backgrounds, as well as whole spine samples being obtained for global gene expression profiles. Briefly, we downloaded publicly available gene expression profiles of control (n = 4) and experimental (n = 4) mouse spines based on the proteoglycan-induced AS model in BALB/c mice [23]. We observed expression differences between control and experimental mice for several of the leukocyte metagenes (Figure 2A). Specifically, we observed increased monocyte, DC, NK, and CD4+ T cell metagene expression within the spines of mice induced with experimental AS (Figure 2B). Hence, we conclude that in an experimental mouse model of AS, the disease is associated with increased monocyte, DC, NK cells, and CD4+ T cell infiltration relative to control animals.

To extend these findings in human patients, we examined the expression of leukocyte metagenes in synovial biopsies extracted from patients with seronegative SpA or in control biopsies. Briefly, we downloaded publicly available gene expression profiling data for the samples described above, and compared leukocyte metagene expression in biopsies derived from the 8 SpA patients (2 AS, 6 SpA) with 7 control biopsies (4 normal, 3 OA). Similar to the mouse model data, we observed differences in leukocyte metagene expression among the control and SpA patient biopsies (Figure 3A). Specifically, we observed increased DC and CD4+ T cell metagene expression in the SpA biopsies relative to control biopsies (Figure 3B). Interestingly, we did not observe increased expression of monocyte and NK cell metagene expression in the human SpA biopsies relative to control, whereas these metagenes were significantly more highly expressed in experimental mouse biopsies. A post-hoc analysis of individual genes comprising the differentially expressed leukocyte metagenes, revealed that metagenes were upregulated as a whole, and the observed increases were not driven by dramatic up-regulation relatively few genes (see Additional file 3: Figure S2). Importantly, this suggests that our observations are not confounded by dramatic changes in single genes, and likely represent true measurements of infiltration.

Taken together with our previous data, these findings strongly suggest that the joints of patients with AS/SpA contain increased leukocyte infiltrate that primarily comprises DC and CD4+ T cells relative to control patient joints.