Gene expression analysis in asthma using a targeted multiplex array

Human lungs were donated with consent from the IIAM and used with approval from the University of British Columbia and St. Paul’s Hospital ethics committee. Diagnosis of asthma was determined through patient medical history and asthma medication usage as determined by family interview. Non-asthma donor deaths were primarily due to head trauma while 8 of the 12 donors with asthma died during exacerbations of their asthma. The other four donors with asthma died due to other, accidental causes (eg. head trauma). Subject demographics can be seen in Table 1 with full subject characteristics found in Additional file 1: Table S1. After surgical removal the lungs were flushed with Custodiol HTK solution (Odyssey Pharmaceuticals, East Hanover, NJ, USA) and transported on ice by plane. The time between harvesting and arrival at the University of British Columbia was 15–20 h. Tissues from the lungs have been used in previous studies [9‐11]. Inflated frozen lungs were processed into tissue cores for sectioning on a crytostat. A total of twenty 10 μm thick sections per core were cut and stored at −80 °C until RNA was isolated. Sections 1, 5, 10, 15 and 20 were stained with hematoxylin and eosin (H&E) for morphometric measurements. For the remaining 15 sections, airways and a small amount of surrounding parenchyma were macroscopically dissected using a scalpel (Fisher Scientific® No. 11) for RNA isolation. Samples were only used if the airways seen on the first section were continuous for the 20 sequential sections. Large vessels were avoided. Sample airway is seen in Fig. 1. RNA was isolated using the Qiagen® RNeasy Mini Kit according to manufacturers protocol.

Measurements of ASM, epithelial, collagen, and total wall area in addition to basement membrane thickness were measured to quantify the degree of airway remodeling. The five sections stained with H&E (Fig. 2) from each core were digitally scanned and airway wall compartments were quantified using the Aperio® system (Leica Biosystems, Germany). ASM area, epithelial area and total airway wall area were quantified using a point counting method, where a grid of 4000 points was overlaid onto each airway of interest and the points falling on the area of interest were counted (Image Pro Plus ®, Media Cybernetics, Maryland). The measurements for all airways were normalized to the internal perimeter (Pi) of the airway and averaged across all the sections for each subject. The Pi was measured by tracing along the luminal side of the epithelium. For the measurement of basement membrane thickness and collagen area, two additional sections were cut from each core (same as used for RNA isolation and other measurements) to stain with Masson’s trichrome (basement membrane) and Picrosirius red (collagen). To quantify the basement membrane thickness, a random series of line segments was placed over each image and the thickness of the basement membrane was measured at the points where any line segment crossed the basement membrane. All thickness measurements were made perpendicular to the epithelium. A minimum of 40 measurements were made for each airway. For the measurement of collagen content, Picrosirius red stained slides were visualized under polarized light where the Picrosirius red stain shows birefringence. The collagen appears red on a black background and the amount of collagen was quantified using color segmentation in Image Pro Plus®. All measurements were carried out in a blinded manner.

Expression of mRNA for the 334 candidate genes and 12 housekeeping genes was measured with the Nanostring® system using a custom codeset panel. The most stable housekeeping genes were selected by measuring 12 common housekeeping genes. By comparing the % Coefficient of variation (%CV) across the 12 housekeeping genes we were able to determine that the 5 most stable genes for data normalization were: RNA Polymerase R2A (POLR2A), TATA box binding protein (TBP), Ribosomal Protein L19 (RPL19), Guanine nucleotide-binding protein subunit beta-2-like 1 (GNB2L1), and β-Glucuronidase (GUSB). These 5 genes were also selected because they spanned a range of counts from low (average 198 counts for GUSB) to high (average 22,486 counts for RPL19). Data were normalized in accordance with Nanostring® guidelines. See Additional file 1: Table S2 in the online supplement for a complete list of genes in the panel.

We selected the candidate genes based on a priori hypotheses and grouped them based on their function and potential role in the pathogenesis of asthma and/or AHR. These are: ASM contraction and relaxation, structure and regulation of the cytoskeleton, epithelial barrier function, innate and adaptive immunity, fibrosis and remodeling and epigenetics. The rationale for the choice of groups of genes is provided below and the list of the genes by category is in Additional file 1: Table S2

Additional genes were added to the list given their reproducible association in asthma GWAS and observation that the SNP’s in these genes act as eQTLs.

Final Nanostring results were filtered to keep only genes that had an average count of at least 30. Normalized mRNA expression values were compared between asthmatic and non-asthmatic subjects using a linear model with a negative binomial distribution controlling for age, sex, and inhaled corticosteroid use. Differential gene expression data are presented as volcano plots as well as in a summary table showing the top differentially regulated genes. The level of expression of each transcript is not completely independent since there was strong co-expression between the 344 genes. To account for this, we employed the Matrix Spectral Decomposition analysis of Nyholt and Li et al. [27, 28] to identify the effective number of independent genes. This lead to the multiple comparison correction shown in Table 2 that is based on an effective n of 31. Adjusted p-values will be listed as p.adj and genes with a nominal unadjusted p < 0.05 will be listed as p.unadj.

In addition to the analysis of differential gene expression, we performed an analysis of differential co-expression in our data set using the analysis package CoXpress for R. Differential co-expression analysis identifies pairs of genes that are differentially co-expressed i.e. have opposite correlation patterns in cases vs. controls or show correlations in one condition only. Genes that were differentially co-expressed were entered into WebGestalt for pathway enrichment analysis and are presented in table form. Network analyst was used to understand the protein-protein interaction network [PMID: 25,950,236} of all nominally significant genes (p.unadj < 0.05). Data are plotted using GraphPad version 5.04 (La Jolla California USA).

The total number of airways analyzed was 52 in asthmatic and 53 in non-asthmatic subjects; with an average of 4.3 airways per subject (p > 0.05, asthmatic vs. non-asthmatic). The average internal perimeter (Pi) of asthmatic subjects was 5.1 ± 1.5 mm (geometric mean 4.0 ± 1.7 mm) and in non-asthmatic subjects was 5.3 ± 1.5 mm (geometric mean 4.9 ± 1.3 mm) (p > 0.05 for both arithmetic and geometric mean). Airway wall area per unit length of Pi was significantly greater in asthmatics (0.22 ± 0.024 mm²/mm) than in non-asthmatics (0.13 ± 0.019 mm²/mm, p < 0.01). There was also an increase in the ASM area per unit Pi (0.018 ± 0.0024 mm²/mm vs. 0.011 ± 0.0015 mm²/mm, p < 0.05) and in basement membrane thickness (6.9 ± 0.81 mm vs. 3.9 ± 0.73 mm, p < 0.01) in asthmatics versus non-asthmatics. There was no significant difference in the area of epithelium or collagen per unit Pi between donor groups (p > 0.05). Side by side comparison of asthmatic and non-asthmatic airway can be seen in Fig. 2.

Gene expression changes in all genes are summarized in Fig. 3 with the candidate gene hypothesis categories plotted in Fig. 4. In total there were 51 genes differentially expressed based on a threshold p-value of p < 0.05 and three genes that were significant after p-value correction (Table 2, p.adj). In brief, there were three genes whose significance reached the adjusted p-value cutoff, Collagen Type 1 Alpha 1 (COL1A1), COL3A1, and integrin beta 6 (ITGB6). The gene for COL1A1 was the most significantly up-regulated gene (1.83-fold increase, p.adj = 0.01) and integrin beta 6 (ITGB6) was the most significantly down-regulated gene (1.29-fold decrease, p.adj = 0.002) (Additional file 1: Table S2). COL1A1 expression was positively associated with the amount of collagen in the airway in asthmatics and non-asthmatics (Collagen/Pi, R² = 0.2221 and 0.2182 for asthma and non-asthma respectively, p < 0.01) and the thickness of the basement membrane in both groups combined (R² = 0.1313, p = 0.01). COL3A1 expression was positively associated with ASM/Pi (R² = 0.2089, p = 0.02) and Collagen/Pi (R² = 0.2083, p = 0.03) in asthma. ITGB6 expression was negatively associated with basement membrane thickness in asthma (R² = 0.1801, p = 0.04) and negatively associated with Epithelial area/Pi in both groups combined (R² = 0.1158, p = 0.02). There was an association between ITGB6 expression and Collagen/Pi in asthma that did not quite reach significance. (R² = 0.144, p = 0.06). In each hypothesis group there were a number of differentially expressed genes that did not reach significance after p-value adjustment, these included: contractile apparatus structure – Smoothelin (1.40 fold increase, p.unadj = 0.01); regulation of contraction – CD38 (1.66 fold decrease, p.unadj =0.003); cytoskeletal structure – ITGB6 (see above); cytoskeletal regulation – RAC1 (1.60 fold decrease, p.unadj =0.002); epigenetic regulation – PRMT5 (1.52 fold decrease, p.unadj =0.03); epithelial function – LAMC2 (1.59 fold decrease, p.unadj =0.006); fibrosis and remodeling – COL1A1 (see above); innate and adaptive immunity – PTGFR (2.18 fold decrease, p.unadj =0.004). The counts, p-values, and adjusted p-values for all significant genes can be seen in Additional file 1: Table S2. Of the 12 genes identified in GWAS or linkage analysis, only ADAM33 (1.56-fold increase, p.unadj =0.0057) was up-regulated. There were no pathways significantly enriched in the differentially up or down-regulated genes.

Using Network Analyst, we identified a minimum protein-protein interaction network and key nodes from our nominally differentially expressed genes (Fig. 5). Green nodes indicate down-regulated genes, red nodes indicate up-regulated nodes (both relative to non-asthmatics), and grey nodes indicate first order interactions. This network highlights a number of key nodes in our data set including: Mitogen-Activated Protein Kinase 1 (MAPK1, degree = 24), c-FOS (degree = 22, and Calmodulin 3 (CALM3, degree = 21). Using this network, we were able to identify pathways significantly associated with up and down-regulated nodes (Table 3). These included key pathways in collagen degradation and remodeling and alterations to cell-cell communication.

We identified groups of genes that were differentially co-expressed between asthmatics and non-asthmatics. In this analysis, genes are clustered together based on how their expression values correlate with each other. The analysis was performed twice with the comparison group being the non-asthmatics or asthmatics in the different analyses. Clusters of genes that are significantly co-expressed in one condition and not in the other are said to be differentially co-expressed. Figure 5 shows an example; expression of genes in cluster 11 changes from subject to subject in asthmatics (left) and non-asthmatics (right). Each line represents one gene in the group. Genes in cluster 11 follow a similar pattern of expression in asthmatics (p < 0.001) but not in non-asthmatics (p = 0.32). The rest of these figures can be seen in Additional file 1: Fig. S3 and S4.

In non-asthmatics, there were 3 groups of genes that were differentially co-expressed compared to asthmatics. These clusters (10, 35, and 53) had an average co-expression correlation coefficient of 0.772 (p < 0.0001) in non-asthmatics and 0.223 (p > 0.05) in asthmatics. The genes in these groupings are summarized in Additional file 1: Table S3. We performed pathways analysis on each group of genes with Webgestalt. These results are shown in Table 4. There were no pathways that were significantly enriched for genes differentially co-expressed in non-asthmatics. Further analysis identified specific gene pairs from cluster 10 that were both significantly positively correlated in non-asthmatics and significantly negatively correlated in asthmatics (Table 5). These included: chitinase 3-like 1(CHI3L1) and GSDMB (R = 0.760 and R = −0.707); CHI3L1 and histone deacetylase 10 (HDAC10) (R = 0.731 and R = −0.676); HDAC10 and thymocyte antigen 1 (THY1 or CD90) (R = 0.727 and R = −0.596); and indoleamine 2,3-dioxygenase 1 (IDO1) and nuclear factor of activated T-cells, cytoplasmic 2 (NFATC2) (R = 0.659 and R = −0.604), in non-asthmatics and asthmatics respectively. All showed significant but opposite direction of correlation (p < 0.05 Fig. 6).

In asthmatic samples there were 6 clusters of genes that were found to be differentially co-expressed. These clusters had an average correlation coefficient of 0.728 (p < 0.0001) in asthmatics and 0.169 (p > 0.05) in non-asthmatics. The genes in these 6 clusters are summarized in Additional file 1: Table S3. Each of the clusters was also analyzed with Webgestalt (Table 4). In brief, asthmatic co-expressed genes were significantly enriched in pathways for cytoplasmic virus pattern recognition signaling (p = 3.0 × 10⁻⁴), positive, and negative regulation of type 1 interferon production (p = 3.9 × 10⁻³, and p = 2.0 × 10⁻³ respectively). Within the cluster of genes that were differentially co-expressed, there were 4 pairs of genes whose expression were positively and significantly correlated in asthmatics but negatively and significantly correlated in non-asthmatics (Table 5). In brief: laminin B2 (LAMB2) and matrix metallopeptidase 9 (MMP9) (R = 0.727 and R = −0.686); BPI fold containing family A (BPFIA1) and DEAD box polypeptide 58 (DDX58 or RIG-1, retinoic acid inducible gene 1 protein) (R = 0.814 and R = −0.646); bone morphogenetic protein 2 (BMP2) and NFATC2 (R = 0.707 and R = −0.660); and G-protein alpha stimulating (GNAS) and NFATC2 (R = 0.665 and R = −0.798), in asthmatics and non-asthmatic respectively. All were significantly correlated but in opposite directions (p < 0.05, Fig. 6).

All eight pairs of differentially co-expressed genes (Table 5) were entered into the ENCODE ChIP-SEQ significance tool. All but one of the genes (IDO1) from the pairs was regulated by the transcriptional repressor CCCTC-binding factor or CTCF. Since only 47% of the candidate genes have a CTCF binding site this represents a significant enrichment (p = 0.0035, Chi-squared test).