Airway Transcriptome Networks Identify Susceptibility to Frequent Asthma Exacerbations in Children

Abstract

Background

Frequent asthma exacerbators, defined as those experiencing >1 hospitalization in a year for an asthma exacerbation, represent an important subgroup of individuals with asthma. However, this group remains poorly defined and understudied in children.

Objective

To determine the molecular mechanisms underlying asthma pathogenesis and exacerbation frequency.

Methods

We performed RNA-sequencing of upper airway cells from both frequent and non-frequent exacerbators enrolled in the Ohio Pediatric Asthma Repository.

Results

Through molecular network analysis, we found non-frequent exacerbators display an increase in modules enriched for immune system processes, including type 2 inflammation and response to infection. In contrast, frequent exacerbators showed expression of modules enriched for nervous system processes, such as synaptic formation and axonal outgrowth.

Conclusion

These data suggest that the upper airway of frequent exacerbators undergoes peripheral nervous system remodeling, representing a novel mechanism underlying pediatric asthma exacerbation.

Capsule Summary:

This is the first report examining the molecular etiology of pediatric frequent asthma exacerbators. It implicates the peripheral nervous system in exacerbation pathobiology. Therapeutics targeting these neuronal processes may be beneficial in preventing/managing frequent exacerbators.

Graphical Abstract

INTRODUCTION

Asthma is a common and heterogenous allergic respiratory disease found within both adult and pediatric populations. Much work over the past two decades has revealed asthma to be a complex phenotype composed of multiple pathobiological subgroups regulated by interactions with both genetics and inhaled environmental exposures ¹. The most common endotype, comprised of Th2-driven cytokine release and subsequent eosinophil and mast cell activation, has been extensively examined at a mechanistic level ^2–4. Others, including exacerbations due to respiratory-virus infection and severe exacerbations driven by neutrophil-predominant disease are also beginning to be explored ^5,6. While these endotypes play differing roles in an individual’s specific disease, the common severe outcome is an exacerbation, representing the majority of healthcare costs and burden associated with asthma.

An exacerbation consists of an acute and progressive loss of lung function associated with smooth muscle hyperresponsiveness and mucus hypersecretion, leading to the bronchiolar constriction and mucus plugging characteristic of asthma. Severe exacerbations require hospitalization where oral corticosteroids and β-adrenergic agonists are utilized to dampen inflammation and relax the airway. Repeated exacerbations result in airway remodeling, with goblet and squamous cell hyperplasia leading to impairment of barrier defense and mucociliary clearance ⁷. Additionally, analysis from single-cell RNA-sequencing (scRNA-seq) studies of asthmatic airways identified a population of ciliated cells expressing genes required for mucus assembly and secretion, suggestive of metaplastic changes to the epithelium following airway inflammation ^2,8. Due to the costs, morbidity, and pathological airway changes associated with repeated asthma exacerbations, there is a need to understand the etiology underlying increased exacerbation frequency.

Epidemiological evidence has identified another phenotype of individuals with asthma based on the frequency of asthma exacerbation. Termed “frequent exacerbators,” this understudied group disproportionally affects Black and Hispanic populations and is responsible for a majority of asthma morbidity and healthcare costs ⁹. These individuals often exhibit classic signs of allergic inflammation, such as increased exhaled nitric oxide and serum IgE levels ^9,10; however, these markers are not specific and cannot reliably predict which patients may experience recurrent severe exacerbations, providing limited insight towards the mechanisms of exacerbation pathology. Additionally, studies examining this phenotype have largely focused on adult patients with asthma, creating a need for investigation within pediatric populations. To address these questions and others, the Ohio Pediatric Asthma Repository (OPAR) was created as the first statewide repository of children admitted for treatment of asthma exacerbations at any of the 6 Ohio major children’s hospitals, including Akron, Cincinnati, Cleveland, Columbus, Dayton, and Toledo ¹¹. In addition to health outcomes data, biospecimens, including nasal epithelial cells and blood, were collected from participants at the time of exacerbation and banked at Cincinnati Children’s Hospital, allowing the interrogation of molecular pathways which may differentiate frequent from non-frequent exacerbators.

METHODS

Study Characteristics

All patient samples used in this study came from the Ohio Pediatric Asthma Repository (OPAR), described in detail previously ¹¹. Briefly, OPAR is a collaborative among 6 major Ohio children’s hospitals to prospectively enroll children between the ages of 5 and 19 years old admitted to any of the participating hospitals for treatment of an acute asthma exacerbation. Clinical, demographic, and environmental data are collected during each child’s index hospitalization and data are collected regarding subsequent hospitalizations from the electronic medical record. Biospecimens were collected on a subset of participants. Frequent asthma exacerbators (FEs) were defined as patients experiencing two or more hospitalizations for an asthma exacerbation within a calendar year; non-frequent exacerbators (non-FEs) were hospitalized no more than once in a 12-month period. Annual exacerbations were tracked with the following categories: 1, 2, 3, 4–5, or 6–7 exacerbations per year. Nasal epithelial cells were obtained at the index visit from 144 subjects by brushing the anterior turbinate with a cytology brush; this was submerged in RLT plus lysis buffer and β-mercaptoethanol and stored at −80°C until extraction. For this study, we included 139 subjects who could be categorized as FEs or non-FEs.

RNA sequencing and gene expression analysis

Total RNA was isolated from the nasal epithelial cells using RNeasy Microarray Tissue Mini Kit (Qiagen) according to manufactures’ instructions. Samples were then submitted for RNA-sequencing (RNA-seq) at the Genomics, Epigenomics, and Sequencing Core at the University of Cincinnati. Bioanalyzer (Agilent, Santa Clara, CA, USA) was used to determine RNA-quality and NEBNext Poly(A) mRNA Magnetic Isolation Module (New England BioLabs, Ipswich, MA, USA) was used to isolate polyA RNA. NEBNext Ultra II Directional RNA Library Prep Kit (New England BioLabs) was used for library preparation and the RNA was sequenced using HiSeq 1000 sequencer (Illumina, San Diego, CA, USA). Each sample was sequenced at a 50 million read depth at 150 bp long. Quality trimmed reads were mapped to the human hg19 genome, quantified using RSEM ¹² and mapped with Bowtie 2 ¹³ using default thresholds within the computational suite for bioinformaticians and biologists tool (CSBB v. 3).

To detect viral presence or absence, RNA was subjected to cDNA synthesis and sequencing using the SMARTer^® Stranded Total RNA-Seq Kit v3 - Pico Input Mammalian reagents (Takara Inc). Briefly, RNA was fragmented and cDNA was synthesized using random hexamers. Thereafter, Illumina-compatible indexed adapters were added and host rRNA was then depleted enzymatically. A second round of amplification was used to generate the final sequence library. Samples were pooled and then subjected to sequencing on the NovaSeq 6000 machine (Illumina Corp) to 50 million 150 nucleotide paired end reads per sample. Raw RNA sequence reads were aligned against a comprehensive viral genome database (downloaded from: https://ftp.ncbi.nlm.nih.gov/refseq/release/viral/) using Bowtie2. All aligned reads where then assembled into contigs using Spades ¹⁴. Assembled contigs were then re-aligned to the viral genome database with BLASTN using cutoff value of 1e-10. Each contig was assigned to a single genome demonstrating the top scoring alignment. Virus-positivity was defined as any subject with a strain of adenovirus, influenza virus, parainfluenza virus, respiratory syncytial virus, or rhinovirus (RV) detected. Because rhinovirus was by far the most prevalent of the respiratory viruses detected, we also counted RV load as the sum of different strains detected.

Of the 144 participants with airway epithelial RNA-sequencing available, 139 could be defined as frequent or non-frequent exacerbators. Raw counts generated from Poly-A selected RNA were used to assess outlier samples; 7 were determined to be more than 2.5 standard deviations from the mean and were subsequently removed from the analysis. Genes were then filtered for those that had a minimum of 0.1 counts per million in at least 10 samples. voomWithQualityWeights from the limma R package ¹⁵ was used to transform counts to log2 counts per million with observation level weights. The final dataset included 132 samples and 22,158 genes, 17,719 coding and 4,439 non-coding.

Differences in gene expression were assessed between frequent and non-frequent exacerbators by ANOVA using a weighted linear model (limma) appropriate for RNA-seq data with normalized gene expression as the outcome variable and exacerbation status as the primary predictor. Because sequencing batch contributed significant variability to the data, it was included as a blocking variable in the primary model. Other covariates (including the clinical characteristics described in Table 1, rhinovirus detection, and rhinovirus load) did not significantly contribute to data variability. P-values were obtained from 2-sided tests. For gene selection, multiple testing correction was performed using the Benjamini-Hochberg method and genes with a lenient FDR cutoff of 0.3 were incorporated for module generation (n = 2569). Genes were associated into modules using weighted gene correlation network analysis (WGCNA) ¹⁶. Parameters used were: power = 4, the lowest power for which the scale-free topology fit index reached 0.8; networkType and TOMType = “signed”; minModuleSize = 15 and maxBlockSize = 600; and deepSplit = 4. Module level expression values were calculated using the mean expression of genes assigned to a module, and differential expression of the modules between phenotypes was determined with the same weighted linear model as used for differential gene expression analysis. Multiple testing correction was performed using the Benjamini-Hochberg method, and modules with an FDR cutoff <0.05 were considered significant, with fold change differences relative to non-frequent exacerbators. Module expression values were similarly compared between multiple annual exacerbation categories by replacing the phenotype term with a sample’s respective category. To assess trends of module expression over increasing annual exacerbations, Spearman’s correlation appropriate for ordinal data was employed. An absolute value of rho between 0.2 and 0.4 indicated a mild to moderate relationship, and p <0.05 was considered statistically significant. A linear regression model was fit to observe the trend of module expression over increasing annual exacerbations. All analyses were performed using R.

Table 1.

Study population stratified by exacerbation phenotype

	Frequent Exacerbator (N=49)	Non-frequent Exacerbator (N=83)	p-value *
Demographic
Mean age (±SD) (yrs)	10.92 (3.90)	10.22 (3.76)	0.32
Male sex	25 (51.0%)	43 (51.8%)	0.99
Black race	36 (75.0%)	44 (53.7%)	0.02
Public insurance	41 (85.4%)	56 (68.3%)	0.04
Site			0.22
Cincinnati	23 (46.9%)	48 (57.8%)
Cleveland	9 (18.4%)	7 (8.4%)
Columbus	15 (30.6%)	27 (32.5%)
Dayton	2 (4.1%)	1 (1.2%)
Baseline Characteristics
Symptom Frequency Score			0.67
1	35 (74.5%)	61 (75.3%)
2	6 (12.8%)	11 (13.6%)
3	4 (8.5%)	3 (3.7%)
4	2 (4.3%)	6 (7.4%)
Baseline Severity			0.10
Intermittent/Mild	20 (42.6%)	48 (58.5%)
Moderate/Severe	27 (57.4%)	34 (41.5%)
Baseline Albuterol Use			0.282
Never	10 (34.5%)	30 (56.6%)
Weekly	17 (37.0%)	28 (34.6%)
Multiple Times Weekly	9 (31.0%)	11 (13.6%)
Daily	10 (34.5%)	12 (14.8%)
Exacerbation Characteristics
Mean no. of days with symptoms prior to exacerbation (SD)	3.00 (1.08)	2.30 (0.99)	<0.01
ED RAD			0.87
1	2 (10.0%)	6 (9.8%)
2	9 (45.0%)	23 (37.7%)
3	9 (45.0%)	32 (52.5%)
IP RAD			0.09
1	3 (14.3%)	16 (25.8%)
2	11 (52.4%)	16 (25.8%)
3	6 (28.6%)	29 (46.8%)
Mean ED LOS (±SD) (Hrs)	30.04 (99.53)	49.46 (24.05)	0.10
Mean IP LOS (±SD) (Hrs)	41.19 (17.48)	44.19 (23.64)	0.46
Mean ED PRD (±SD) (Hrs)	15.91 (97.87)	35.37 (19.11)	0.09
Mean IP PRD (±SD) (Hrs)	27.05 (15.99)	30.07 (18.53)	0.36
Time to biospecimen collection (±SD) (Hrs)	28.62 (11.44)	27.51 (12.91)	0.64
Respiratory Virus Detection
Rhinovirus + (%)	19 (43.2)	39 (50.6)	0.37
RSV + (%)	0 (0.0)	1 (1.29)	0.61
Adenovirus + (%)	1 (2.27)	0 (0.0)	0.37
Influenza + (%)	2 (4.35)	2 (2.60)	0.66
Total IgE (±SD)	653.44 (724.81)	1024.90 (1600.89)	0.10

Values are counts dispalying percentages (%) or means ± standard deviation (±SD). Because some data may be missing values for certain variables, the denominators may not be the same as the total N for each exacerbation category.

^*Comparisons between frequent and non-frequent exacerbators are done using Fisher’s exact test for count data and linear model ANOVA.

ED = Emergency Department; IP = Inpatient; RAD = Respiratory rate, accessory muscle use, decreased breath sounds; LOS = Length of stay

Pathway Analysis

The biological function of modules was investigated using the ToppGene⁸ suite. Each module was submitted to these databases and enrichment was considered significant if one or more categories displayed an FDR <0.05. STRING version 11.5 ¹⁷, a database of known and predicted protein-protein interactions was used to determine interaction networks for each module. All interaction sources were used, with a minimum interaction score of .2 required for a gene to be included in a network. Cytoscape version 3.9.1 ¹⁸ was used to draw interaction networks according to a prefuse force directed layout using the combined interaction score exported from STRING. Networks were filtered for up to the 30 most interconnected genes in each module. The size of each gene was made proportional to the number of interactions in the network as a percentage of the maximum number of interactions of any node in that network and edges were scaled to interaction score. Unconnected nodes were excluded. Summary annotations of modules were derived from manual inspection of the functional enrichment categories and the genes comprising the final interaction network.

Module Expression Heatmap

Average module expression values for each sample were scaled and both modules and samples were subjected to hierarchical clustering. The heatmap and related dendrograms were plotted in R using the complexheatmap package ¹⁹.

Cell-Specific Deconvolution

Raw counts from the OPAR bulk RNA-sequencing dataset as well as the multi-subject single cell profiles of nasal swabs from Ziegler et al²⁰ were input into the MuSiC algorithm²¹ implemented in R. Cell type identifications were based on prior categorizations defined in Ziegler et al. Genes in the scRNA-seq dataset were filtered to match those present in the OPAR bulk read matrix; these filtered profiles were then used to estimate cell proportions using the music_prop function. The estimated proportions were normalized to sum 1 across included cell types.

Data Availability

The raw RNA-sequencing FASTQ data and minimum information about a high-throughput nucleotide SEQuencing Experiment (MINSEQE) have been deposited to the National Center for Biotechnology Information Gene Expression Omnibus (GEO) with accession number GSE211158.

Code Availability

The R code for all analyses in this manuscript has been annotated and deposited as open-source code in GitHub at https://github.com/kphelan13/OPAR

RESULTS

Study Population

The goal of the current study was to evaluate clinical and molecular features specific to children prone to recurrent severe asthma exacerbations, termed frequent exacerbators. We defined frequent exacerbators (FEs) as those participants having 2 or more hospital admissions for treatment of an asthma exacerbation within a 12-month period and compared these children to non-frequent exacerbators (non-FEs) who had a single index admission within 12-months. Figure 1 summarizes the overall study design and definitions of clinical variables obtained from this cohort.

Fig 1. Study Design and Demographic/Clinical Characteristics.

132 children with asthma admitted to any of the 6 major pediatric medical centers in Ohio for an exacerbation were consented and enrolled into the Ohio Pediatric Asthma Repository. At the time of admission biospecimens, including blood and nasal epithelial cells, were collected and banked at Cincinnati Children’s Hospital. Children with another admission for an asthma exacerbation 12 months before or after their index admission were defined as frequent exacerbators. Baseline disease characteristics were collected based on parent-administered questionnaire. Exacerbation characteristics were obtained from the electronic medical record. This figure was generated at Biorender.com.

Table 1 displays the clinical and demographic information for the 132 participants analyzed in the current study (49 frequent exacerbators and 83 non-frequent exacerbators). FEs were more likely to be Black (75.0% vs 53.7%; p = 0.02) and have public health insurance (85.4% vs 68.3%; p = 0.04), but there were no significant differences in distribution of age, sex, or sample collection site between the two groups. Baseline disease characteristics at least four weeks prior to the index hospitalization were assessed using parent-administered questionnaires. There were no statistically significant differences in baseline symptom score or baseline disease severity between frequent and non-frequent exacerbators (see Figure 1 for definitions of these variables). There was also no significant difference in the proportion of rhinovirus-positive participants. Exacerbation characteristics measured relevant clinical parameters obtained during the acute exacerbation. Frequent exacerbators were more likely to have a longer period of symptoms prior to the exacerbation requiring hospital admission (3.00 vs. 2.30 days; p < 0.01); however, acute disease severity measured by RAD (respiratory rate, accessory muscle use, decreased breath sounds) score ²², length of stay in both the emergency and inpatient setting, and physiologic readiness for discharge in both the emergency and inpatient setting did not differ between the two groups. Additionally, the time between steroid administration in the emergency department and biosample collection after admission to the inpatient setting did not significantly differ.

Frequent exacerbators display a unique increase in gene signatures associated with nervous system processes

Airway epithelial RNA-sequencing (RNA-seq) was available for all participants described in Table 1. After normalizing raw reads, differential gene expression analysis followed by weighted gene correlation network analysis (WGCNA) identified 9 discrete gene expression modules composed of 2342 genes which displayed significantly distinct patterns of expression between frequent and non-frequent exacerbators (Supplementary Table 1). To confirm our definition of a frequent exacerbator, we examined average module expression over increasing reported exacerbations and found no significant differences between individuals reporting two annual exacerbations or those reporting more than two (Supplementary Table 2). These modules were then analyzed using ontological tools (methods) to determine their functional enrichments. Subsequently, STRING ¹⁷ and Cytoscape ¹⁸ were employed to generate networks of known protein interactions allowing the identification of highly-interconnected genes likely driving module enrichment.

In frequent exacerbators, five modules had significantly higher expression compared to non-frequent exacerbators. Interestingly, two of these (M1 and M5) were enriched for genes involving neuronal processes, such as synaptic formation and axonal outgrowth. The module annotated as “Synapse Formation and Axon Projection” (M5) had an average 1.52 fold increase in frequent exacerbators over non-frequent exacerbators, representing the module with the greatest difference in expression between these two groups (false discovery rate [FDR] < 0.01) (Fig 2A). Furthermore, increasing expression of M5 correlated with increasing annual exacerbations (AEs) (rho = 0.309, p < 0.001) (Fig. 2B). M5 contained genes important for glutaminergic synapse formation (NRXN1, BSN, PPFIA2, LRRTM3), transport along microtubules in axons (KIF1A, CLIP3) (Fig 2C), and, additionally, olfaction (CNGA2, OR5A1, OR5A2), highlighting these aspects of neuronal enrichment. In a similar manner, module M1, annotated as “Synapse Formation and Cell-Cell Junction”, displayed an average 1.15-fold higher expression in frequent exacerbators compared to non-frequent exacerbators (FDR <0.01) (Fig 2D); M1 also showed increasing expression as the number of AEs increased (rho = 0.335, p <0.0001) (Fig 2E) and was enriched for genes involved in synapse formation, including DLG2, LRR4C, and TENM2 (Fig 2F), along with genes important for forming cellular junctions (SSX2IP, ITGAV, SSPN). M1 further included genes important for neuronal development, such as the transcription factor SOX2 and downstream effector proteins of the TGFβ and Wnt/β-Catenin signaling pathways (SMAD9, TCFL5). Collectively, these results demonstrate abnormal peripheral nervous system regulation, which is relatively specific to the upper airway of frequent exacerbators.

Fig 2. RNA-seq based network analysis displays neuronal transcriptional profiles increased in frequent exacerbators.

A) Average expression of module M5. The bar indicates the median expression level and the black square indicates the mean expression level. B) M5 module expression over self-reported annual exacerbations. Significance was calculated using Spearman’s correlation. C) STRING networks for genes comprising module M5. Nodes are scaled to the number of interactions in the module and width of edges is scaled for interaction score (0.2–1). D-F) Similar to A-C, but for module M1.

Immune system modules are increased in non-frequent exacerbators

Four modules showed higher expression in non-frequent exacerbators as compared to frequent exacerbators and showed a remarkably different pathway enrichment. The module with the greatest expression difference, M2, contained 407 genes enriched for immune system processes (FC = 1.38, FDR < 0.01) (Fig 3A) and displayed decreasing expression as AEs increased (rho = −0.257, p = 0.003) (Fig 3B). These included genes responsible for leukocyte activation (FCGR3A, CD2, IL2RG, BTK, CSF1R), immune cell chemotaxis (CCR5, CCR2, CCL5, ITGAL), and transcription factors important for immune cell function (IKZF1, RUNX3, MAFB) (Fig. 3C). Genes within this module are implicated in both innate and adaptive processes, displaying an activation of both arms of the immune system in non-frequent exacerbators. Interestingly, this module also possesses genes important for the response to infection, such as TLR7, a toll-like receptor responsible for the detection of single-stranded RNA in endosomes and LY96, an important adaptor molecule for TLR4-mediated detection of bacterial lipopolysaccharide ^23,24. Interferon-inducible genes, such as IFITM1, IFI27, and IRF9, were also present in this module, suggesting a putative barrier response to infection.

Fig 3. Non-frequent asthma exacerbators show an increase in co-expression modules related to immune system processes.

A) Average expression of module M2. The bar indicates the median expression level and the black square indicates the mean expression level. B) M2 module expression over self-reported annual exacerbations. Significance was calculated using Spearman’s correlation. C) STRING networks for genes comprising module M2. Nodes are scaled to the number of interactions in the module and width of edges is scaled for interaction score (0.21). D-F) Similar to A-C, but for module M6.

The module with the second-greatest increase in expression in non-frequent exacerbators, M6, is enriched for processes related to cell-cycle initiation and basophil differentiation (FC = 1.27, FDR < 0.01) (Fig. 3D) and decreases in expression as AEs increase (rho = −0.289, p < 0.001) (Fig. 3E). Accordingly, GATA1 and TAL1 are expressed within this module; these together form a transcriptional complex regulating hematopoietic differentiation, including that of basophilic progenitors ²⁵. Interestingly, many genes which regulate the cell cycle and initiate DNA replication (CDC45, CDC6, CDCA7, MCM6, and MCM10) are found within this module, suggesting an additional proliferative response specific to activated immune cells in non-frequent exacerbators. In addition, markers of airway eosinophil activation and type II inflammation (IL1RL1, SIGLEC8, CPA3, and DPP4) are present within this module, displaying a component of allergic immunity highlighted in previous studies of asthma ^26,27 (Fig. 3F). Corresponding to the known effect of type II inflammation on ciliogenesis and ciliary processes ²⁸, non-frequent exacerbators show a significant decrease in three modules enriched for these genes (Supplementary Fig. 1). Two other modules showed an increase in expression in non-frequent exacerbators. M4 showed enrichment for genes involved in endoplasmic reticulum stress and the unfolded protein response (Supplementary Fig. 2A), including the central regulator XBP1 (Supplementary Fig 2C). M9, annotated as the “Antigen Presentation and NK Cell Function Module,” showed enrichment for HLA genes indicative of interaction between antigen presenting cells and effector cells of the immune system (Supplementary Fig. 2F). In total, the transcriptomic modules associated with non-frequent exacerbators are enriched for pathways involved in canonical immune system mechanisms previously implicated in asthma, particularly genes involved in the type II- and viral-immune response.

Distinct patterns of airway gene expression differentiate frequent and non-frequent exacerbators

To discern global patterns of gene expression between the two asthmatic phenotypes, we performed hierarchical clustering of each OPAR participant using individual scaled values of module expression. This separated the modules into two distinct groups, with cluster 1 (modules M1, M3, M5, M7, M8) associated with frequent exacerbators and cluster 2 (modules M2, M4, M6, M9) with non-frequent exacerbators (Fig 4), highlighting the stark contrasts in module expression between the two groups. As expected, the cluster associated with frequent exacerbators contained modules for neuronal and ciliary processes. Non-frequent exacerbators displayed enrichment for immune modules and cellular stress.

Fig 4. Clustered heatmap of co-expression modules highlights transcription divergence between frequent and non-frequent asthma exacerbators.

Individual average module expression was scaled and hierarchically-clustered to show associations with a clear separation between modules increased in frequent and non-frequent exacerbators. Module expression levels are shown as scaled mean expression for each individual (columns) with red representing higher relative expression and blue representing lower expression. Participants were also clustered to identify patterns in module expression, with frequent and non-frequent exacerbators largely clustering together. The “Status” bar indicates an individual’s phenotype.

Cell-type deconvolution and viral sub-analyses

To estimate cell proportions in the bulk RNA-sequencing data, we leveraged a single-cell RNA-sequencing (scRNA-seq) dataset of nasal swabs ²⁰ and the MuSiC deconvolution algorithm²¹. Concordant with a previous scRNA-seq analysis of human nasal brushings²⁹, the majority of cell proportions were estimated to be from secretory and ciliated epithelial cells (Supplementary Fig. 3). OPAR participants did display a higher estimated proportion of immune cell types than what has been reported in healthy airways, highlighting the immune cell signature associated with an acute exacerbation.

As RV-induced exacerbations have been predicted to cause the majority of severe pediatric asthma exacerbations^30,31, we performed subgroup differential gene expression analyses to determine the effect of rhinovirus in frequent and non-frequent exacerbators separately. In either case, there were no significant differentially expressed genes between RV+ and RV- individuals (Supplementary Table 3).

DISCUSSION

Our data demonstrate that the pathogenesis of asthma exacerbations in frequent exacerbators is characterized by modules related to neuronal processes in contrast to non-frequent exacerbators in whom exacerbations are characterized by immunological pathways previously described in asthma. These data suggest that neuronal remodeling plays an important role in the etiology of the frequent exacerbator phenotype, representing a distinct pathogenesis from the classic immunologic pathways involved in the non-frequent exacerbator phenotype. Our findings highlight fundamental differences in these exacerbations and suggests that therapy targeting neuronal pathways may be necessary to adequately prevent and/or treat frequent exacerbations.

In our study, frequent exacerbators were defined as those having two or more admissions for an asthma exacerbation within a 12-month period. Previous analyses of frequent exacerbators in both adult and pediatric populations have defined this phenotype as those with ≥2 ^9,32 or ≥3 ^33,34 exacerbations per year. We evaluated the gene expression differences between these definitions and found that while there were marked differences in gene expression between those with 1 exacerbation versus 2 or more exacerbations, children with 2 exacerbations were similar to those with more than 2 exacerbations. Collectively, our data suggests that the definition of frequent exacerbator (2 or more exacerbations in a 12-month period) is not just a clinical definition, but a biologically relevant definition that reveals a new asthma endotype with distinct pathogenesis. This combined with the moderate association of neuronal module expression with annual exacerbations indicates increased airway neuronal gene expression predisposes to more frequent exacerbations.

Previous studies of adult frequent asthma exacerbators have found an association between frequent asthma exacerbations and a history of sinusitis and smoking, increased corticosteroid use, and poor symptom control ^9,35,36; however, heterogeneity in both the definition of a “frequent exacerbator” and the patient populations studied makes these epidemiologic findings difficult to generalize to a pediatric population. In the OPAR cohort, frequent exacerbators were more likely to be black, have public insurance, and experience longer days of symptoms prior to arrival in the emergency department. While race was not associated with significant variability to the RNA-sequencing data, it is important to acknowledge racial and socioeconomic disparities in asthma exist and have been well documented, likely due to disproportionate allergen exposure in the home environment and systemic inequalities which pervade all facets of life, including access to healthcare^37,38. Recent genome wide association studies have implicated novel loci with asthma severity and drug responsiveness specific to minority populations ^39,40, underscoring a genetic regulatory component in addition to early life exposures. As such, an important next step is to determine if certain allergens or other exposures such as pollution present in urban environments or allelic variants more prevalent in black populations are associated with the gene expression modules defined for frequent exacerbators in this study. Interestingly, baseline disease severity and inpatient disease severity (assessed through RAD score) did not differ between the two phenotypes. Thus, baseline severity does not predict the frequent exacerbator phenotype. Further, this suggests more frequent asthma exacerbations do not lead to more severe exacerbations; rather, the airway is primed to recapitulate a similar exacerbation event and represents a distinct pathogenesis separate from endotypes previously described ^26,41,42.

Few prior studies have attempted to assess the transcriptomic changes underlying increased exacerbation frequency which could help pinpoint potential targets for further mechanistic study. Herein, we characterized the transcriptome of frequent (and non-frequent) exacerbators, identifying distinct signatures associated with both phenotypes. Compared with frequent exacerbators, non-frequent exacerbators have higher expression of multiple immune system pathways, including leukocyte activation and chemotaxis, interferon response, and type 2 (T2) inflammation. T2 inflammation is a complex interplay between environmental allergens and the epithelium, with subsequent cytokine signaling to eosinophils, basophils, and mast cells which mediate immune responses ^27,43. We identified increased expression of a module containing the T2-related genes IL1RL1, SIGLEC8, and CPA3. IL1RL1 is the immune-cell receptor for IL33 and has been studied in the context of epithelial alarmin signaling to immune effector cells ^4,44. SIGLEC8 and CPA3 are genes associated with mast cells, eosinophils, and basophils ^45,46, suggesting enrichment for these cell functions in non-frequent exacerbators. These findings align with previous transcriptomic studies of asthma ^2,47, showing a conserved role for T2 inflammation and networks of immune cell activation and epithelial alarmin response which are specific to nonfrequent exacerbators.

Our genome-wide unbiased approach revealed increased expression of genes enriched for neuronal pathways in frequent asthma exacerbators, suggesting peripheral nervous system dysregulation plays a role in increased exacerbation frequency. While novel in the context of asthma transcriptomics, nervous system dysregulation and remodeling has been increasingly implicated in a variety of allergic and inflammatory conditions, including atopic dermatitis, allergic rhinitis, and inflammatory bowel disease ^48–50. Previous studies examining the neuronal-airway interface have shown that neuropeptides activate the Th2 response from resident immune cells ⁵¹ or directly stimulate goblet cell hyperplasia in the conducting airways ^52,53. In mouse models of eosinophilic asthma, increased eosinophils lead to increases in peripheral nerve density and irritant sensitivity in the airway ⁵⁴. Additionally, it has been increasingly recognized that psychosocial stress intensifies asthma pathology⁵⁵. In particular, dysregulated neuroendocrine activity and glucocorticoid signaling have been hypothesized to affect the onset and severity of asthma⁵⁶. While the exact mechanisms mediating the interaction between frequent asthma exacerbations and the nervous system remain to be elucidated, our data shows an increase in genes specific for axonal outgrowth and targeting, suggesting increased nerve density is present in frequent exacerbators and may constitute an important part of airway remodeling after multiple exacerbations. An important next step will be to collect and interrogate scRNA-sequencing of FEs to establish which cells of the airway may contribute to the dysregulated gene expression described in our bulk sequencing data.

Our study has several limitations. Because lower airway specimens are largely inaccessible in most pediatric cohorts, we have leveraged nasal epithelial swabs which may not fully recapitulate the lower conducting airway. However, previous studies have shown the transcriptomes of the nasal and bronchial epithelium are highly similar, with >90% overlap in expressed genes between the tissues, making the upper airway a useful model for asthma pathology ^57,58. We also lacked access to peripheral blood mononuclear cells (PBMCs) from OPAR participants. While the nasal transcriptome has been shown to display more biological variation than blood in asthma ^5,59, this tissue may have provided us with a source of baseline gene expression. Additionally, patients in OPAR received oral corticosteroids in the emergency department prior to nasal swabbing in the inpatient unit. Corticosteroids have been shown to alter expression of genes related to eosinophil activation and epithelial differentiation ⁵, potentially dampening aspects of the inflammatory response. Finally, while we have shown a trend of increasing neuronal gene expression as annual exacerbations increase, we lack baseline samples which would allow us to establish chronic expression of these genes and pathways over time. Future studies examining the longitudinal impact of frequent asthma exacerbations will be needed to fully support this phenomenon. Nonetheless, our study represents the first report implicating the nervous system in recurrent exacerbation pathobiology and provides further evidence for the definition of frequent exacerbators as those having two or more exacerbations in a 12-month period. Further investigation regarding the transcriptional regulators of these neuronal gene modules and mechanisms underlying interactions between the epithelium of the upper airway and the peripheral nervous system in the setting of asthma exacerbations are warranted.

Key Messages.

• Pediatric frequent asthma exacerbators are a recognized phenotype of asthma, but the molecular etiology of this important subgroup has yet to be examined.

• The transcriptome of frequent exacerbators is characterized by an increase in co-expression networks enriched for neuronal genes, indicating peripheral nervous system remodeling may play a role in frequent asthma exacerbations.

• This is the first report implicating the role of the nervous system in recurrent exacerbation pathology and highlights frequent exacerbators as a biologically distinct endotype in children.

Abbreviations

OPAR

Ohio pediatric asthma repository

FE

Frequent exacerbator

Non-FE

Nonfrequent exacerbator

AEs

Annual Exacerbations

FDR

False discovery rate

T2

Type 2

RV

Rhinovirus

RAD

Respiratory rate, accessory muscle use, decreased breath sounds

RNA-seq

RNA-sequencing

scRNA-seq

single-cell RNA sequencing