Weighted gene co-expression network analysis to identify key modules and hub genes associated with paucigranulocytic asthma

Asthma is a heterogeneous disease with different phenotypes that vary in natural history, severity of the disease and response to anti-inflammatory therapy [1]. According to the airway inflammation subtypes, asthma can be categorized into four distinct inflammatory phenotypes: eosinophilic asthma (EA), neutrophilic asthma (NA), mixed granulocytic asthma (MGA), and paucigranulocytic asthma (PGA) [2]. Recently, extensive attentions have been paid to EA and NA, which have been successfully applied to clinical research and asthma management. For instance, airway eosinophilic inflammation is somewhat related to atopy and EA has a good response to inhaled corticosteroids (ICS) [3‐6]. While airway neutrophilic inflammation is associated with the exposure to environmental pollutants (such as smoking) or the presence of bacterial or viral infection [7]. Additional therapy of macrolide may be more suitable for NA with respect to reducing airway neutrophilic inflammation [8].

However, as one of the most common phenotypes of asthma, PGA are still poorly understood and researches on PGA are limited [9]. Some studies considered PGA to be a special phenotype driven by macrophages or mast cells other than eosinophils or neutrophils [10, 11]. Other studies suggested that PGA may represent a non-inflammatory type or a phenotype with a low grade of eosinophilic inflammation [12]. The precise characteristics and pathobiology of PGA are not well delineated. It is urgent to unveil inflammatory and immune mechanisms underlying PGA.

The rapid development of microarray and high-throughput sequencing technologies facilitate the study of asthma in genetic level. An earlier study conducted a hierarchical cluster analysis based on the transcriptional profiles of asthma and identified three clusters that showed similarities with the inflammatory phenotypes of EA, NA and PGA [10]. However, there are no studies that specifically address the transcriptional features of PGA. The key gene modules or hub genes of PGA are still unknown. Traditional methods rely on differential expression detection to identify potential biomarkers or targets, but may miss useful genes. Weighted gene coexpression network analysis (WGCNA) is a bioinformatic method to explore complex interactions among gene expression profiles. According to expression similarity, WGCNA can transform gene expression data into potentially biologically relevant modules and reveal relationships between the gene modules and external clinical traits by using an intramodular hub gene or module eigengene [13]. It is quite helpful in identifying hub genes or therapeutic targets.

In this study, we sought to identify the hub genes located in the regulatory center of PGA using WGCNA and other bioinformatic methods. Additional biological functional analyses were also conducted to investigate the biological processes, related pathways and immune status of PGA. The results will help to shed light on hidden mechanisms and identify therapeutic targets of PGA.

In total, 47 samples (18 PGA and 29 non-PGA) from GSE45111 were used to perform WGCNA, DEGs analysis, functional enrichment analysis and PPI analysis. 54 samples from GSE137268 were used to validate the identified hub genes. The demographic and clinical data of patients in the two datasets were summarized in Additional file 1: Table S1. The flowchart of the study is presented in Fig. 1.

As one of the most common phenotypes of asthma, PGA accounts for the 31–51.7% of asthma [2, 26, 27]. However, unlike EA or NA, researches on PGA are limited and its characteristics have not been well delineated [9]. To the best of our knowledge, this is the first transcriptomics study on PGA to identify key gene modules and hub genes. In the present study, we investigated the transcriptome of 18 asthmatic patients with a phenotype of PGA and 29 controls of non-PGA. Using integrated analyses of DEGs, WGCNA and PPI, we identified and validated six hub genes of PGA, including ADCY2, CXCL1, FPRL1, GPR109B, GPR109A and ADCY3. In comparison with strategies focused on individual gene, network-based methods are more suitable to reveal global biological activity. WGCNA focuses on the correlations between the co-expression modules and the external clinical traits, not merely the differences in gene expression patterns, and thus the results are more reasonable [13]. Consequently, the analysis allows the identification of candidate genes and the modules potentially linked to the biological function of interest. The GO, KEGG and ssGSEA analyses were further performed to elucidate the potential biological process, pathways and immune functions that may be implicated in the pathogenesis of PGA. These results may enhance the current understanding of the mechanisms underlying PGA and provide potential therapeutic targets for newly developed treatments.

It has been previously reported that PGA most likely represents a “benign” phenotype of asthma and it is associated with a good response to anti-asthma treatment. Several studies have suggested that PGA has distinct inflammation features compared with EA or NEA [26, 28, 29]. In the enrichment analysis of our study, GO terms related to the inflammation response and immune regulation were significantly enriched, such as regulation of inflammatory response (GO:0050727), regulation of immune effector process (GO:0002697) and regulation of adaptive immune response (GO:0002819), indicating that the inflammatory and immunological characteristics were different between PGA and non-PGA. Ntontsi et al. found that patients with PGA express lower levels of inflammatory biomarkers in exhaled air and induced sputum supernatants compared with other inflammatory phenotypes, representing a less intense inflammatory process [26]. Demarche et al. also showed that PGA may display a low-grade airway inflammation [12]. The results of ssGSEA in our study showed that the scores of immune cell infiltration and immune functions were lower in PGA than non-PGA, which seems to support that PGA represents a less intense immune response and the viewpoint that PGA is somewhat a kind of phenotype with low degree of inflammation [12]. According to Ntontsi et al. in some patients with PGA, the “absence of inflammatory response” could possibly be the results of a pre-existing eosinophilic asthma adequately treated with ICS in which there is no neutrophilic inflammation. In other words, some PGA patients may be the result of the successful therapeutic intervention. The hypothesis may partly explain the low degree of immune response presented in PGA and its good response to anti-asthma treatment [26]. However, Deng et al. found that the phenotype of PGA was stable and that most patients with PGA had not undergo an inflammatory phenotype transition. Their study did not support the hypothesis that all subjects with PGA represent a cross sectional view related to disease activity or represent a treatment success. Instead, it indicated that most patients with PGA could constitute an independent phenotype [30]. More studies are required to address these concerns. Meanwhile, it should be noted that the immune scores of most immune cells were lower in PGA except for NK cell. Although the mechanisms of NK cells in the regulation of inflammation of asthma are not fully elucidated, recent studies have suggested that NK cells in asthma inflammation can be protagonistic or antagonistic, depending on the environmental agent that is used to elicit the disease (allergen, diesel exhaust particles and virus) and the phase of the disease (the sensitization phase, the effector phase and the resolution phase) [31‐34]. Therefore many factors, including the type of the environmental trigger, the phase of inflammation and the cytokine milieu between PGA and non-PGA, should be further investigated.

Our study suggested the different gene expression patterns between PGA and non-PGA. Six hub genes were identified based on the combination analyses of DEGs, WGCNA and PPI. Of these, ADCY3 was up-regulated in PGA, while the remaining five hub genes were down-regulated. The expression patterns were further validated in a separate dataset (GSE137268). The majority of the hub genes were involved in the regulation of immune response and inflammation. For example, the up-regulation of ADCY3 suggests an increase in cAMP formation, which could suppress inflammatory function in DCs [35]. FPRL1 was reported to be implicated in several immune processes, such as chemotactic migration and the production of reactive oxygen species (ROS) [36]. Several agonistic and antagonistic peptide sequences for the FPRL1 receptor have been investigated as drug candidates for inflammatory diseases including asthma [37]. The remaining hub genes were found to be involved in the migration of inflammatory cells. GPR109B participated in the migration of eosinophils to the sites of inflammation [38]. CXCL1 was found to be a chemoattractant for neutrophil recruitment during tissue inflammation [39]. GPR109A was expressed in many immune cells, including macrophages, monocytes, neutrophils and DCs. Activation of GPR109A has been found to be implicated in several diseases where inflammation contributes to the underlying pathophysiology such as obesity, colitis and neurodegenerative disorders [40]. But its role in asthma is still not elucidated. The expression patterns of these genes that related to the immune cells activation or migration may explain the decreased ssGSEA score of immune status in PGA. Difference in chemotaxis and migration of the immune cells between PGA and non-PGA may be an important factor that leads to the different inflammatory characteristics of the two asthma phenotypes.

Our study was different in many respects from the original study for GSE45111 [41]. First, the objective of the study was to find gene signatures that could discriminate eosinophilic asthma from other phenotypes and to investigate its predicted value for ICS treatment response. In our study, we focused on identifying the hub genes for PGA. Gene signatures in the original study were identified based on the differential expression analysis, while in our study the hub genes were identified by the combination analyses of DEGs, WGCNA and PPI. We conducted more comprehensive bioinformatic analyses such as GO and KEGG enrichment analysis, ssGSEA, WGCNA and PPI analysis. These bioinformatic analyses were not performed in the original study.

It should be mentioned that although the identification of asthma inflammatory phenotype in both datasets used in our study was based on the cross-sectional data of induced sputum, asthma inflammatory phenotypes identified by this method are proved to be stable by many studies [30, 42‐44]. Actually, induced sputum is currently the best available noninvasive assessment of bronchial inflammation in asthma, and it is regarded as the gold standard for asthma inflammatory phenotyping [45]. This method has been widely adopted in many of studies [46‐48]. GINA guideline also recommends to use the method to confirm asthma inflammatory phenotype [49]. Deng et al. particularly focused on the PGA and found the phenotype of PGA identified by induced sputum was stable and that majority of the patients with PGA had not undergone an inflammatory phenotype transition after one-month fixed anti-asthma treatment with ICS [30]. In our study, the subjects in GSE45111 were stable asthmatics. Those who with recent (past month) respiratory tract infection, asthma exacerbation, unstable asthma, change in therapy and current smoking were excluded [39]. Taking account of all these factors into consideration, sputum phenotypes in the dataset of GSE45111 could be considered as stable.

The present study had several limitations. Firstly, we analyzed a single platform of a dataset and the sample size relatively was small, which may affect the stability of our study. Although we have validated our findings in a separate dataset, the results should be interpreted carefully. Besides, more sociodemographic characteristics and some other important clinical traits, such as pulmonary function or exacerbation history were absent in the original datasets, so we cannot perform a more comprehensive analysis. Finally, our study is based on a in silico analysis, more studies aimed at elucidating the further mechanisms of the identified hub genes in PGA are desired in the future.

In summary, our study suggested that PGA and non-PGA were different in gene expression patterns, biological processes, related pathways and immune status. With comprehensive analyses of WGCNA and other related bioinformatic methods, we constructed a weighted co-expression network and identified a key gene module that associated with PGA. Finally, six hub genes of PGA were identified and then validated in a separate dataset. The different gene expression patterns, biological processes and immune status between PGA and non-PGA indicated the heterogeneity of asthma at level of molecular biology, suggesting the requirement of individualized treatment for different phenotypes of asthma. Our results may improve the understanding the heterogeneity of asthma and the underlying mechanism of PGA, providing potential therapeutic targets for patients with PGA.