BAL Cell Gene Expression in Severe Asthma Reveals Mechanisms of Severe Disease and Influences of Medications

Abstract

Rationale: Gene expression of BAL cells, which samples the cellular milieu within the lower respiratory tract, has not been well studied in severe asthma.

Objectives: To identify new biomolecular mechanisms underlying severe asthma by an unbiased, detailed interrogation of global gene expression.

Methods: BAL cell expression was profiled in 154 asthma and control subjects. Of these participants, 100 had accompanying airway epithelial cell gene expression. BAL cell expression profiles were related to participant (age, sex, race, and medication) and sample traits (cell proportions), and then severity-related gene expression determined by correlating transcripts and coexpression networks to lung function, emergency department visits or hospitalizations in the last year, medication use, and quality-of-life scores.

Measurements and Main Results: Age, sex, race, cell proportions, and medications strongly influenced BAL cell gene expression, but leading severity-related genes could be determined by carefully identifying and accounting for these influences. A BAL cell expression network enriched for cAMP signaling components most differentiated subjects with severe asthma from other subjects. Subsequently, an in vitro cellular model showed this phenomenon was likely caused by a robust upregulation in cAMP-related expression in nonsevere and β-agonist-naive subjects given a β-agonist before cell collection. Interestingly, ELISAs performed on BAL lysates showed protein levels may partly disagree with expression changes.

Conclusions: Gene expression in BAL cells is influenced by factors seldomly considered. Notably, β-agonist exposure likely had a strong and immediate impact on cellular gene expression, which may not translate to important disease mechanisms or necessarily match protein levels. Leading severity-related genes were discovered in an unbiased, system-wide analysis, revealing new targets that map to asthma susceptibility loci.

At a Glance Commentary

Scientific Knowledge on the Subject

Global gene expression of BAL cells, which samples the cellular milieu within the lower respiratory tract, has not been well studied in severe asthma. We found BAL cell gene expression was heavily influenced by participant traits; cell proportions in samples; and recent medication use, notably β-agonist exposure. The influence of β-agonists on expression was recapitulated in a THP-1 cellular model. After identifying and accounting for these influences, we identified genes related to asthma severity measures. Many of these genes were found at well-known asthma susceptibility loci and related to antiinflammatory and/or repair pathways, providing new investigational targets.

What This Study Adds to the Field

This is the first study to examine global gene expression in BAL cells in a large cohort enriched for severe asthma and the first in the Severe Asthma Research Program. We found that β-agonist use, among other factors (age, sex, race, and cell populations), has a significant impact on gene expression that has not been previously considered. Nonetheless, we identified gene expression that tracked well with severity, and many of these genes related to antiinflammatory or repair pathways and mapped to well-known and validated asthma susceptibility loci.

Severe asthma (SA) constitutes approximately 10% of asthma cases, but accounts for approximately 50% of all disease-related costs (1). SA is a complex and poorly understood condition (2–5). Global gene expression in cells derived from nasal and bronchial epithelial brushings has helped identify underlying inflammatory mechanisms (6–21) and delineate asthma subphenotypes (7, 9, 10, 21). Global gene expression of BAL immune cells, which samples the cellular milieu within the lower respiratory tract, has not been well studied in SA, and a systems-level analysis may offer insights into prominent disease processes. In this study, BAL cells were obtained by bronchoscopy in 154 asthma and healthy control (HC) subjects from SARP (Severe Asthma Research Program), and global gene expression profiled. The goal was to identify new biomolecular mechanisms underlying SA by an unbiased, detailed interrogation of global gene expression.

An increased prevalence of goblet and immune cells in epithelial brushings of asthmatic airways means differential gene expression in comparison with healthy airways is unavoidably biased by quantitative differences in the cell populations present. Gene expression is also knowingly influenced by age, sex, race, and recent medication use, although these influences are seldomly interrogated. In this study, severity-related gene expression was determined in BAL cells by associating expression with clinical markers of increasing severity (e.g., low lung function, low quality-of-life scores), but only after carefully identifying and accounting for influences that may bias the results. This analytic pipeline was repeated in airway epithelial cell (AEC) gene expression (obtained during the same bronchoscopy) to relate AEC to BAL severity-associated gene expression, including a cluster analysis that combined datasets and identified SA subclusters with distinct clinical and gene expression characteristics.

Methods

Study Population

Samples were obtained from subjects in SARP under institutional review board supervision and following an informed consent. Bronchoscopies were performed in each subject. A single dose of a short-acting β-agonist (SABA) was administered 45 minutes before the procedure. Bronchial brushings and lavage were obtained on contralateral sides in accordance with SARP protocols (22–25). A cell pellet was derived from five sequential washings that were performed and combined into one solution. Some of the sample was banked and used subsequently for ELISAs. The median recovery from the instilled solutions was 56.5 ml (interquartile range, 44–91). Protocols can be found in detail in the online supplement. The expression dataset was made available to the public (GEO; Accession GSE130499).

Tools Used in BAL Gene Expression Analysis

Limma R package, weighted gene coexpression network analysis (WGCNA) (26–29), and PANTHER gene ontology were used for differential gene expression, coexpression network, and pathway enrichment analyses, respectively (30). One hundred participants had both BAL and AEC microarray data. The severity-associated gene expression in AECs has already been reported (19). In this study, networks from the BAL and AEC datasets were combined and clustered by K means to identify subphenotypes. K = 5 was used based on a minimization of error. A random forest model was used to permutate and validate the clusters identified. Out-of-bag estimates were extracted from the model to calculate overall error and cluster-specific error (see online supplement, Confusion matrix).

In Vitro Studies Using THP-1 Monocyte Cells and Alveolar Macrophages

Alveolar macrophages were collected from BAL in nondiseased lungs that were harvested but not used for transplant. Cultured THP-1 and alveolar macrophages were exposed to physiologically relevant concentrations of a nonselective β-agonist (BA; isoproterenol [ISO]) both acutely (30 min–2 h) and in a prolonged fashion (24 h). These cells were examined by RNAseq, immunoblot (monoclonal antibodies from Santa Cruz Biotechnology and Millipore), ELISA (Aviva Systems Biology), and cAMP measurement using cAMPglo (Promega). RNASeq of the THP-1 cells was analyzed using the CLC genomics workbench (QIAGEN).

Results

Participant Characteristics by Traditional Asthma Severity

Subjects with SA were older and had a higher body mass index (BMI) than other groups except moderate asthma on inhaled corticosteroids (Table 1; full demographics are available in the online supplement). In comparison with subjects without SA, subjects with SA had lower BAL monocyte/macrophage cells (81% vs. 87%, respectively; P = 0.0008) and higher neutrophils (6.2% vs. 2.3%, respectively; P < 0.0001) (Table 2).

Table 1.

Basic Demographics of All Participants Grouped by American Thoracic Society–defined Asthma Severity^*

	HCs (1)	Mild–Mod No ICS (2)	Mild + ICS (3)	Mod + ICS (4)	Severe (5)	P Value	Significant Intergroup Differences (Bonferroni Corrected)†
Totals	37	40	18	15	44	NA	NA
Age, yr, mean ± SE	29 ± 1.8	29 ± 1.7	32 ± 2.6	36 ± 2.8	43 ± 1.7	<0.0001	(5) > (1)
							(5) > (2)
							(5) > (3)
Sex, M/F	18/19	12/28	6/12	7/8	15/28	NS	NS
Race, W/AA/O	24/4/8	23/12/5	13/4/0	5/8/2	30/10/3	0.03	NS
BMI, mean ± SE	26 ± 1.0	29 ± 0.9	27 ± 1.4	31 ± 1.5	32 ± 0.9	0.0003	(5) > (1)
							(5) > (2)‡
							(5) > (3)
Atopy, Y/N	20/17	28/4	13/2	11/3	34/6	0.003	(2) > (1)
							(5) > (1)
CRS, n (%)	0	1/37 (2.7)	3/16 (18.8)	2/14 (14.3)	10/43 (23.3)	0.005	(2) > (1)
							(3) > (1)
							(4) > (1)
							(5) > (1)
Pneumonia history, n (%)	3/37 (8)	5/38 (13)	9/15 (60)	5/14 (36)	28/43 (65)	<0.0001	(3) > (1)
							(3) > (2)
							(5) > (1)
							(5) > (2)
Juniper AQLQ (total score), mean ± SE	7.0 ± 0.2	5.6 ± 0.2	5.0 ± 0.3	4.2 ± 0.25	3.8 ± 0.1	<0.0001	(1) > (2)
							(1) > (3)
							(1) > (4)
							(1) > (5)
							(2) > (4)
							(2) > (5)
							(3) > (4)
							(3) > (5)
FEV1% pred, mean ± SE	99.8 ± 2.4	87.6 ± 2.3	93.9 ± 3.4	68.3 ± 3.7	64.2 ± 2.2	<0.0001	(1) > (2)
							(1) > (4)
							(1) > (5)
							(2) > (4)
							(2) > (5)
							(3) > (4)
							(3) > (5)
BDR %, mean ± SE	4.9 ± 3.4	14.9 ± 3.4	10.5 ± 5.1	24.0 ± 5.5	36.4 ± 3.1	<0.0001	(4) > (1)
							(5) > (1)
							(5) > (2)
							(5) > (3)
ED/Hosp in past 1 yr, n (%)	0	2/38 (5)	3/16 (19)	6/14 (43)	34/43 (79)	<0.0001	(3) > (1)‡
							(4) > (1)
							(4) > (2)
							(5) > (1)
							(5) > (2)
							(5) > (3)
IgE, IU/ml, median 25th–75th percentiles	17 (8–85)	106.5 (49–426)	139 (72–468)	141 (43–239)	132 (30–510)	0.0002§	(2) > (1)
							(3) > (1)
							(4) > (1)
							(5) > (1)
FeNO, ppb, mean ± SE	30 ± 5.7	47 ± 5.9	43 ± 9.9	47 ± 9.5	46 ± 5.6	0.21	(A) > (HCs)
Blood eos, cells/μl, mean ± SE (mean %)	143 ± 27 (2.6)	253 ± 29 (4.0)	230 ± 39 (3.5)	309 ± 46 (5.9)	265 ± 26 (3.8)	0.005‖	(4) > (1)
							(A) > (HC)

Definition of abbreviations: A = asthma; AA = African American; AQLQ = asthma quality-of-life questionnaire; BDR = bronchodilator response; BMI = body mass index; CRS = chronic rhinosinusitis; ED/Hosp = emergency department visits or hospitalizations; eos = eosinophils; Fe_NO = fractional exhaled nitric oxide; HCs = healthy control subjects; ICS = inhaled corticosteroids; Mod = moderate; mod + ICS = moderate asthma on inhaled corticosteroids; NA = not applicable; NS = not significant; O = other; W = white.

Bold indicates P values less than 0.05 (i.e. statistically significant).

^*44 (29%) were missing Fe_NO, and 20 (13%) were missing AQLQ data, otherwise sample traits had less than 10% missing data.

^†All significant intergroup differences listed achieved a Bonferroni-corrected P value of P ≤ 0.005.

^‡Highly significant P values, although not reaching stringent Bonferroni correction (specifically, P < 0.008).

^§According to a Wilcoxon test of significance. Several IgEs were markedly elevated, assuming normal distribution was not valid, and therefore Student’s t tests could not be performed.

^‖Based on differences in mean percentage.

Table 2.

BAL Characteristics of Participants Grouped by American Thoracic Society–defined Asthma Severity

	HCs (1)	Mild–Mod No ICS (2)	Mild + ICS (3)	Mod + ICS (4)	Severe (5)	P Value	Significant Intergroup Differences (Bonferroni Corrected)*
BAL cell count, cells × 104/ml	7.2 ± 1.1	9.9 ± 1.1	8.4 ± 1.7	10.6 ± 1.9	5.7 ± 1.1	0.06	NS
BAL mac, cells × 104/ml, mean ± SE (mean %)	6.3 ± 1.1 (87)	8.8 ± 1.1 (89)	7.2 ± 1.6 (85)	9.4 ± 1.7 (87)	4.8 ± 1.1 (81)	0.01†
							(2) > (5)
							(non-SA) > (SA)
BAL lymph, cells × 104/ml, mean ± SE (mean %)	0.7 ± 0.1 (9.6)	0.6 ± 0.1 (8.9)	0.9 ± 0.2 (10.5)	0.9 ± 0.2 (9.9)	0.5 ± 0.1 (11.2)	0.79	NS
BAL neu, cells × 104/ml, mean ± SE (mean %)	155 ± 51 (2.9)	210 ± 51 (1.8)	169 ± 75 (2.4)	187 ± 83 (1.9)	269 ± 51 (6.2)	0.0005†	(5) > (1)
							(5) > (2)
							(5) > (3)
							(5) > (4)
BAL eos, cells × 104/ml, mean ± SE (mean %)	65 ± 21 (0.7)	31 ± 21 (0.5)	92 ± 31 (1.8)	83 ± 34 (0.9)	87 ± 21 (1.8)	0.33	NS

Definition of abbreviations: eos = eosinophils; HCs = healthy control subjects; ICS = inhaled corticosteroids; lymph = lymphocytes; mac = macrophages/monocytes; Mod = moderate; Mod + ICS = moderate asthma on inhaled corticosteroids; neu = neutrophils; non-SA = subjects with nonsevere asthma; NS = not significant; SA = severe asthma.

^*All significant intergroup differences listed achieved a Bonferroni-corrected P value of P ≤ 0.005.

^†Based on differences in mean percentage.

BAL Expression Caused by Differences in Cell-Type Proportions and Other Factors

Using an false discovery rate <0.05 cutoff, genes were found to positively correlate with the proportion of eosinophils (n = 16), neutrophils (n = 2,125), macrophages (n = 520), and lymphocytes (n = 1,414) in the BAL. Accounting for age, sex, race, and BMI reduced and perhaps improved the robustness of these correlated genes: 19 for eosinophils, 579 for lymphocytes, 112 for neutrophils, and 62 for alveolar macrophages (see online supplement). Interestingly, the top approximately 20 lymphocyte-correlated genes included two CD8 subunits, three granzyme genes, RANTES (CCL5), and several IFN-γ–induced genes (e.g., CXCL9), suggesting enrichment for IFN-γ–producing cytotoxic T cells (CTLs), which is in agreement with prior studies (31–36). Alternatively, tumor necrosis factor (TNF)-α and type 1 IFN-induced (IFN-α/β) genes (e.g., TNFAIP6, IFITM2, IFITM3, IFITM4P) were positively correlated with neutrophil abundance. Suggesting an increased prevalence of reparative and antiinflammatory properties in alveolar macrophages, genes CD11c (ITGAX) and two antiinflammatory growth factors, epiregulin (EREG) and heparin-binding epidermal growth factor–like growth factor (HBEGF), were most correlated with monocyte/macrophage abundance.

Excessive shedding of airway epithelium is known to occur in asthma (37). We therefore manually interrogated the % epithelial cells in 70 BAL samples (still available) to look for differences between cohorts. We found that 2.8% (median) epithelial cells were present in samples, but there was no difference in asthma (n = 47) versus HCs (n = 23), or in SA (n = 26) versus non-SA (n = 44). Furthermore, no genes significantly correlated with the proportion of epithelial cells (false discovery rate <0.05).

BAL genes correlated to race (n = 1,170) and 18% (n = 205) overlapped with genes (n = 2,537) that correlated to race in the AECs and five genes were the top-ranked genes in both datasets: DPF2, CRBB2, CRYBB2P1, RNF135, and PSPH. Interestingly, CRYBB2 and CRYBB2P1 have strongly correlated race in prior studies (19, 38). X and Y chromosome-specific genes (n = 52) correlated to sex, and a remarkable 33 (63%) overlapped with genes (n = 93) that correlated to sex in the AECs. Again, the top five genes (XIST, DDX3Y, RPS4Y1, RPS4Y2, and DDX3Y) were the same in both lists. Genes correlated with age (n = 412) and 41 (10%) overlapped with genes (n = 562) that correlated to age in the AECs. Genes correlated with BMI (n = 436), with 45 (10%) overlapping with genes (n = 1,658) that correlated with BMI in the AECs. No other available demographics (e.g., age of diagnosis) had correlated genes. Accounting for cell proportions did not strongly affect genes correlated to these traits.

BAL Expression Related to Traits Indicative of Increased Severity

Gene expression was then correlated to clinical measures indicative of more severe disease, including FEV₁% predicted prebronchodilator, bronchodilator response, increased oral corticosteroid use (daily or ≥3 bursts/yr), increased BA use (long-acting BA plus increased SABA use), number of emergency department (ED) visits or hospitalizations in the past year (ED/Hosp), and asthma quality-of-life questionnaires (AQLQ). Using log10 P values, genes were ranked according to their correlation to all these traits (Figure 1), creating a ranked severity-related gene list. Linear models were also used to account for age, sex, race, BMI, cell proportions, and medication use, further helping determine which top genes were likely unbiased and truly related to severity.

Figure 1.

Asthma severity gene set. BAL cell genes were correlated with clinical traits indicative of more severe disease (including low asthma quality-of-life questionnaires scores, low FEV₁, high bronchodilator response, oral and high-dose inhaled steroid use, high β-agonist use, high emergency department or hospitalization in the past year), and ranked according to their P value significance to these traits. Blue colored rows indicate genes that are inversely correlated or associated with clinical measures of worsening disease severity, whereas yellow colored rows indicate genes that are positively correlated or associated with these clinical measures. AEC = airway epithelial cell; ASL = asthma susceptibility locus; BDR = bronchodilator response; COPD = chronic obstructive pulmonary disease; Dx-Sev = a diagnosis of severe asthma as defined by the European Respiratory Society/American Thoracic Society Task Force definition; ED/Hosp = emergency department visits or hospitalizations in the past year; GABRIEL = A Multidisciplinary Study to Identify the Genetic and Environmental Causes of Asthma in the European Community; GS = glucocorticosteroid; GWAS = genome-wide association study; MAP = mitogen-activated protein; MCP = monocyte chemoattractant protein; MMP = matrix metalloprotinase; NAD = nicotinamide adenine dinucleotide; PDE = phosphdiesterase; PKA = protein kinase A; ROS = reactive oxygen species; SPT = skin prick test; Treg = T regulatory cell.

Among the top 20 genes and positively associated with increasing severity were ADRB2, the β₂-adrenoceptor gene; ZBTB16 and KLF9, two genes strongly associated with corticosteroid use (19, 39, 40); and KRT18P55 and KRT18, two keratin-related genes. ADRB2 correlated strongest to increased BA use and AQLQ. ZBTB16 and KLF9 both correlated strongest to high-dose inhaled corticosteroids, increased oral corticosteroids (≥3 bursts/yr), and bronchodilator response. KRT18P55 and KRT18 both correlated strongest to FEV₁ and bronchodilator response. KRT18P55 is a keratin pseudogene located within the genomic asthma susceptibility locus 17q11 (41) and KRT18 is a keratin gene located at another asthma susceptibility locus 12q13 (42).

Among the top 20 genes and negatively associated with increasing severity were ID3, a transcriptional regulator that promotes T-regulatory cell delineation (43, 44); RAB7B, a negative regulator of IL-6 and TNF-α production in macrophages (45); OSM, a secreted regulator of IL-6 (46, 47), with a gene location near asthma susceptibility locus 22q12.3 (48); CD83, a marker of mature dendritic cells and that negatively attenuates dendritic cell ability to stimulate T lymphocytes (49, 50); and numerous regulators of cAMP-signaling, including transcriptional factor CREM and phosphodiesterase PDE4B.

Among the top genes in the severity-related gene set, many were within or near three asthma-susceptibility loci. Five of the top approximately 40 severity genes we found in region 5q12–32 (48, 51–53), including HBEGF (5q31), PLK2 (5q11), SH3PXD2B (5q35), C5orf30 (5q21), and ADRB2 (5q32). Three genes were found in 17q11–23 (54), including KRT18P55 (17q11), CCL14 (17q12), and KCNJ2 (17q24). Six genes were in 1p13–36, including ID3 (1p36.1), AMPD2 (1p13), CASP9 (1p36), C8A (1p32), PDE4B (1p31), and SLC44A3 (1p21).

Increased BA Use Linked to Depressed cAMP Signaling Networks

WGCNA identified 49 gene coexpression networks. Only one network (n = 132 genes) strongly related to all indicators of SA (Figure 2). A pathway enrichment analysis of the network genes showed “regulation of cAMP-dependent protein kinase activity” as a top pathway. A literature review of the 132 genes (see online supplement) showed the network to be enriched for genes related to cAMP signaling. ADRB2 (β₂ adrenoceptor), which activates cAMP signaling, was found in the severity network, but not the known steroid-induced genes (ZBTB16 and KLF9). The geometric mean of the network decreased with increasing BA use and increasing severity (American Thoracic Society–defined) (Figure 3). Thus, the network was named BAL + cAMP (BALcAMP).

Figure 2.

Heat map of relationships between gene module eigenvalues and participant clinical traits. Forty-nine gene coexpression networks were identified using Weighted Gene Coexpression Network Analysis. In the heat map, positive relationships are red and inverse relationships are blue. The network labeled “Sex” contained mostly Y-chromosome genes and showed the strongest correlation with sex (P = 7e−49). Modules of lymphocyte-specific and eosinophil-specific genes were identified (named Lymph and Eos, respectively) and correlated strongly with % lymphocytes and % eosinophils in the BAL. Only one module correlated strongly with all asthma severity measures: BALcAMP. Asthma dx = asthma diagnosis; AQLQ = asthma quality-of-life questionnaires; BA = β-agonist; BDR = bronchodilator response; BMI = body mass index; DC = dendritic cell; ED_or_Hosp_1 yr = emergency department visit or hospitalization in the past 1 year; Eos = eosinophils; Fe_NO = fractional exhaled nitric oxide; FEV₁_pre_per = FEV₁ prebronchodilator, percent predicted; FEV₁_pre_liter = FEV₁ prebronchodilator in liters; ICS = inhaled corticosteroids, liters; LABA = long-acting β-agonists; Lymph = lymphocytes; Mac = macrophages; Mono = monocytes;PMN = polymorphonuclear; Severe_asthma = diagnosis of severe asthma; WBC = white blood cell.

Figure 3.

BALcAMP expression in relation to β-agonist use and American Thoracic Society (ATS)-defined asthma severity. BALcAMP expression as represented by the geometric mean was inversely related to (A) β-agonist use and (B) ATS-defined severity classes (P = 0.0003 and P < 0.0001, respectively), suggesting a relationship to both. HC = healthy control subjects; ICS = inhaled corticosteroids; Mod = moderate.

The top 14 hub genes included the previously mentioned genes CREM, CASP9, RAB7B, CD83, and PDE4B and were all decreased in high-severity subjects. The closeness of genes to the network center, called “module membership,” was plotted against the correlation or association traits indicative of SA, called “trait significance” (Figure 4). The positive slopes seen in the graphs in Figure 3 indicated that the network genes were likely part of the same biologic process.

Figure 4.

Plots of module membership versus gene significance for asthma severity measures and BALcAMP. The closeness of a gene to the center of its module (i.e., eigenvalue) was quantified using a module membership variable. The correlation or association of a gene to a sample trait (e.g., total asthma quality-of-life questionnaire score) was quantified using a gene significance variable. For each gene in the BALcAMP module, module membership versus gene significance was plotted for (A) total Juniper asthma quality-of-life questionnaire score, (B) FEV₁% predicted, (C) rate of emergency department visit or hospitalization in the past year, (D) use of inhaled long-acting β-agonists, (E) use of high-dose inhaled corticosteroids, and (F) systemic corticosteroids. As shown in A–F, the genes closest to the BALcAMP module eigenvalue also had the strongest relationship to each trait. The gene CREM (highlighted by red box) is notable for being in the top right of each graph. AQLQ = asthma quality-of-life questionnaires; CS = corticosteroids; ED/Hosp = emergency department visits or hospitalizations; ICS = inhaled corticosteroids; LABA = long-acting β-agonists.

The network included many genes that were monocyte-specific and from the monocyte/macrophage-specific gene set described previously. The other cell type–specific genes segregated into other networks, which included neutrophil, eosinophil, lymphocyte, and inflammatory dendritic cell networks. In agreement with the linear model analysis mentioned previously, the lymphocyte network was also enriched for CD8⁺ CTL networks. Peters and colleagues (55) performed WGCNA using sputum cell gene expression of subjects with asthma and identified closely matching dendritic cell (42% shared genes, including CD1A, CD1B, CD1C, CD1D, CD207, and FCER1A) and CTL cell networks. Our CTL network was enriched with genes upregulated in M1 (e.g., CCR7, IL2RA, IL15RA, CCL19, CCL5, IRF1, CXCL9, CXCL10, and CXCL11) but not M2 macrophages (e.g., CD209, CCL17, CLEC7A, and ARG1) (56), suggesting an interaction between CTLs and classically polarized M1 macrophages.

We repeated WGCNA using only BAL samples with greater than 90% alveolar macrophages (n = 68 samples). This analysis recapitulated the BALcAMP network despite lowering the total sample number and statistical power by approximately half (see Figure E3 in the online supplement). Taken together, these findings strongly indicated that the BALcAMP network was found in BAL monocyte/macrophages.

Identifying Gene Expression Caused by BA Exposure

BALcAMP expression was inversely related to both BA use and American Thoracic Society–defined severity classes (P = 0.0003 and P < 0.0001, respectively) (Figure 3), suggesting a relationship to both. As BAs increase intracellular cAMP and all subjects were exposed to a SABA, we next delineated the effect of BA exposure on monocyte/macrophages gene expression. Using a cell line model system, cultured monocyte cells (THP-1) were stimulated with a nonspecific BA (ISO), gene expression measured by RNAseq, and cAMP signaling pathways interrogated by functional assays. We found acute exposure to ISO increased intracellular cAMP levels, but this response was lost with prolonged (24 h) ISO exposure and diminished on rechallenge (24 + 1 h ISO), indicating a desensitization to subsequent acute BA exposure (Figure 5A). In human alveolar macrophages, we observed decreases in BA receptor 2 protein levels after prolonged ISO treatment with a concomitant induction of β-Arrestin, which shuttles the β₂ adrenoreceptor from the membrane for degradation (Figure 5B) and plays a role in BA desensitization. Acute ISO treatment of THP-1 cells resulted in dephosphorylation of the protein kinase A regulatory subunit (i.e., activated protein kinase A) (Figure 5C); increased phosphorylation of CREB (via activated protein kinase A [57]); and induced the c-Fos protein, whose gene transcription is also activated by CREB (58). All these effects were lost on cellular desensitization by prolonged ISO treatment and could not be induced with ISO retreatment.

Figure 5.

cAMP induction and signaling in THP-1 cells following acute and chronic β-agonist exposure. (A) cAMP levels were induced with acute isoproterenol exposure (ISO 1 h), suppressed with chronic ISO (ISO 24 h), and had a dampened response to chronic followed by acute exposure (ISO 24 h+ 1 h). (B) Human primary alveolar macrophages showed induction of β₂-agonist receptor (B2AR) with ISO 1 hour, but suppression of B2AR and induction of B2AR-negative regulator β-Arrestin after 24 hours ISO. (C) ISO 24 hours blocked dephosphorylation of the PKA regulatory subunit (p-PKA Reg), preventing activation and PKA-dependent phosphorylation of CREB (p-CREB). Downstream gene products for c-Fos and HBEGF were induced at ISO 1 hour but suppressed by ISO 24 hours and ISO 24 hours + 1 hour. (D) cAMP induction by acute and prolonged ISO exposure in Beas-2B epithelial and Jurkat T cell lines shows less intense and delayed action, respectively, compared with THP-1 cells. A and D, results are average cAMP values ± SEM relative to untreated cells from three experiments performed on different days. Blots shown are representative from two (B) and three (C) separate experiments from different days. **P < 0.01, by ANOVA compared with other treatments of the same cell type; ***P < 0.001 compared with UnTx and ISO 1 hour. PKA = protein kinase A.

We also interrogated BA response and desensitization in cell lines for human T cells (Jurkat) and airway epithelia (Beas-2B). Beas-2B cells showed a modest (∼50 vs. ∼400 nM for THP-1 cells) upregulation of cAMP after acute ISO exposure, with congruent desensitization to subsequent ISO. Jurkat cells displayed a minimal cAMP induction acutely but maintained a robust induction after prolonged ISO exposure that was unchanged 1 hour after reexposure (Figure 5D).

Corresponding to cAMP pathway activation in THP-1 cells, 7 of the top approximately 20 genes upregulated by acute ISO overlapped with BALcAMP, including the hub genes NR4A3, NR4A2, OSM, FOSB, AREG, SOC3, and PDE4B. In fact, most BALcAMP hub genes (NR4A2, NR4A3, AREG, OSM, EREG, CREM, and PDE4B) were differentially expressed between acute versus chronic ISO (see online supplement). Thus, the cellular model recapitulated many of the gene expression findings observed in vivo (i.e., BALcAMP network), and indicated it was caused, at least in part, by BA exposure. Interestingly, it also suggested that these genes were not being downregulated in SA, but simply more robustly upregulated in subjects without SA, who were also given SABAs just before bronchoscopy. It also indicated that BAL immune cells in subjects with SA were likely desensitized to BA, although unknown if simply caused by BA use alone.

Examination of BALcAMP Proteins in BAL Cell Lysates

Lysates of all BAL cells from a subset of HCs (n = 8), mild asthma (n = 4), and subjects with SA (n = 8), with a range of BA use, were chosen for a protein validation (see online supplement for clinical data). Immunoblots of BAL cell lysates were technically challenging and nonconclusive because of excess degradation and/or extracellular proteins in SA samples with highly variable band density despite equal protein mass run in gels. As an alternative, GAPDH, CREM, FOS, and HBEGF were measured by ELISA. GAPDH protein/volume by ELISA was found to be consistent among all 20 samples (Figure 6A), and was used to normalize the other proteins. In disagreement with gene expression findings, CREM protein abundance was not different between HC/mild compared with SA BAL samples (Figure 6B) and FOS protein was significantly higher in SA compared with non-SA (Figure 6C). HBEGF protein levels were strongly decreased in lysates from patients with SA (in agreement with gene expression). Furthermore, HBEGF protein levels showed a positive relationship to FEV₁ and AQLQ, whereas CREM and FOS demonstrated an inverse relationship with these clinical parameters (see Figure E4).

Figure 6.

Protein expression of BALcAMP genes in asthmatic BAL lysates. (A) GAPDH protein levels were similar between control/mild patients’ BAL lysates and those from patients with severe asthma. CREM (B), FOS (C), and HBEGF (D) protein quantities from BAL samples were standardized to GAPDH for each sample, with variable correlation of protein abundance with gene expression; only HBEGF protein levels correlated with mRNA. P values shown are from two-tailed Student’s t tests. Ctrl = control; Mod = moderate.

Clustering Subjects by Type 2 and Severity-related Networks

The current clinical practice is to divide subjects with SA into “type 2 high” (T2-hi) versus “type 2 low” (T2-lo), because T2-hi SA is marked by eosinophilia, elevated fractional exhaled nitric oxide, and an enhanced response to novel biologic therapies (9, 19). Persistently increased T2 inflammation in SA is also associated with worse clinical outcomes (59, 60). The relationship between T2 inflammation and leading severity-associated networks identified in the AECs (previously reported [19]) and BAL was next identified through clustering. Specifically, a previously validated T2 signature identified in AECs and leading severity-associated networks discovered in the BAL and AEC were clustered into five subject clusters (SCs) using K means and a minimization of error (Figure 7). Of note, the leading AEC severity network was identified by correlating gene networks to the same asthma severity factors and was enriched for genes related to epithelial growth and repair (19). Cross-validation and prediction using a random forests permutation model found a model accuracy high at approximately 80% (see Figures E5 and E6 and online supplement).

Figure 7.

Cluster of combining airway epithelial cell (AEC) and BAL gene expressions. In a previous study, an AEC T2 gene signature has differentiated “T2-hi” asthma subjects from “T2-lo” subjects and healthy control subjects (19). An epithelial growth and repair (EGR) module had separated severe disease from milder forms. In a combinatorial analysis, 17 hub genes from the severity module of the BAL (BALcAMP) module were combined with 8 EGR hub genes and 8 T2 hub genes in the AEC expression. Using entropy-weighted K-means clustering, participants clustered into five subject clusters, objectively, using minimization of error to determine k. In the heat map, upregulated gene expression is yellow and downregulated gene expression is blue. HC = healthy control; ICS = inhaled corticosteroids; Mild–Mod no ICS = mild–moderate asthma/no inhaled corticosteroids; Mild + ICS = mild asthma with ICS; Mod + ICS = moderate with ICS; SC = subject cluster; Severe = severe asthma.

SC1 contained 80% of the HCs and only subjects with mild asthma (Table 3). SC2, SC3, and SC4 had significantly increased T2 gene expression in AECs with corresponding elevations in T2 inflammatory markers, including blood/BAL eosinophilia and elevated fractional exhaled nitric oxide. SC3 (n = 18), marked by significantly lower epithelial growth and repair in AECs and the lowest BALcAMP, contained the most severe patients of the three T2-hi clusters (61% severe). In fact, SC3 contained only patients with asthma, and it had the lowest AQLQ (μ = 3.5), lowest FEV₁ (61.6% predicted), and highest ED/Hosp rates (71%) of any SC. Despite high T2 signals, 44% were using daily oral steroids. SC5 (n = 13) was characterized by significantly lower gene expression across all three modules (T2, epithelial growth and repair, and BALcAMP) and high clinical severity. All SC5 participants has asthma and 69% were severe. SC5 participants had the longest disease duration (32.9 yr), low AQLQ (μ = 4.1), and low but reversible FEV₁ (μ = 69.5% predicted). Sixty-seven percent were in the ED/Hosp in the previous year. Forty-two percent had chronic rhinosinusitis, whereas 67% had a history of pneumonia. Corresponding to the low T2 module expression, blood eosinophils (μ = 175) and BAL eosinophils (0.2%) were low. SC5 had the highest BAL neutrophils (2.7%), significantly greater than SC3. Despite this, 90% were atopic (median of three positive skin tests) and SC5 had the highest IgE level of any cluster (μ = 409 IU/ml). Recently, it has been demonstrated that dupilumab, an anti-IL-4Rα monoclonal antibody that blocks IL-4 and IL-13, decreases IgE over time (60, 61). Thus, it seems likely that subjects labeled as T2-lo (i.e., SC5) may have prior or hidden T2 inflammation, perhaps masked by steroid use. Notably, Peters and colleagues recently showed in sputum that CTL expression positively correlated with T2 inflammation. In disagreement, our CTL module was higher in the T2-lo cluster (SC5) in comparison with T2-hi clusters (SC2 and SC3) (see Figure E7) (55). In fact, our CTL module was significantly higher in clusters with milder disease, SC1 and SC4, in comparison with T2-hi clusters, SC2 and SC3, possibly because of lower corticosteroid exposure, as recently demonstrated in natural killer cells derived from subjects with asthma (62).

Table 3.

Demographics, Clinical, and Inflammatory Characteristics of Asthma SCs

	SC1	SC2	SC3	SC4	SC5	P Value	Significant Intergroup Differences (Bonferroni Corrected)*
Totals	37	14	18	18	13	NA	NA
Asthma, n (%)	22 (59)	13 (93)	18 (100)	15 (83)	13 (100)	0.0006	SC3 > SC1, SC5 > SC1†
Severe asthma, n (%)	5 (13.5)	4 (29)	11 (61)	2 (11)	9 (69)	<0.0001	SC3 > SC1, SC3 > SC4, SC5 > SC1, SC5 > SC4
Severity (HC/mild–mod no ICS/mild + ICS/mod + ICS/severe)	16/9/7/0/5	1/6/1/2/4	0/1/1/5/11	3/8/2/3/2	0/1/2/1/9	<0.0001	SC3 > SC1 (severity), SC3 > SC4, SC5 > SC1
Age, mean ± SE	34.7 ± 2.0	29.6 ± 3.3	40.3 ± 2.9	29.3 ± 2.9	41.5 ± 3.4	0.011	SC5 > SC4†, SC3 > SC4†
Duration, mean ± SE	20.4 ± 2.6	20.8 ± 3.3	28.8 ± 2.9	19.3 ± 3.0	32.9 ± 3.4	0.008	SC5 > SC4, SC5 > SC1
Sex, M/F	14/23	8/6	6/12	4/14	5/8	0.37	NS
Race, W/AA/O	26/4/2	7/6/0	10/4/3	6/5/1	7/3/2	0.19	NS
BMI, mean ± SE	28.0 ± 1.1	29.6 ± 1.7	32.0 ± 1.5	27.6 ± 1.5	32.3 ± 1.8	0.08	NS
Atopy, Y/N	24/11	13/1	14/4	16/1	10/1	0.11	NS
Positive allergy skin tests, n (mean ± SE)	2.6 ± 0.5	4.6 ± 0.9	4.9 ± 0.8	5.1 ± 0.8	3.7 ± 1.0	0.039	NS
Diagnosis of CRS, n (%)	10/34 (29)	3/14 (21)	10/17 (59)	5/18 (28)	5/12 (42)	0.16	NS
Pneumonia history, n (%)	10/34 (29)	2/14 (14)	11/17 (65)	8/18 (44)	8/12 (67)	0.008	SC3 > SC2, SC5 > SC2†
Juniper AQLQ, mean ± SE	5.9 ± 0.2	4.8 ± 0.4	3.5 ± 0.3	5.4 ± 0.3	4.1 ± 0.4	<0.0001	SC1 > SC3, SC4 > SC3, SC1 > SC5, SC4 > SC5
FEV1% pred, mean ± SE	88.7	76.4	61.6	83.9	69.5	<0.0001	SC1 > SC3, SC4 > SC5, SC1 > SC5
FVC% pred, mean ± SE	93.7 ± 2.8	86.6 ± 4.5	78.2 ± 4.0	91.7 ± 4.0	78.0 ± 4.7	0.005	SC5 > SC1, SC3 > SC1
BDR %, mean ± SE	13.7 ± 4.2	20.2 ± 6.9	49.2 ± 6.0	14.5 ± 5.9	27.7 ± 7.2	<0.0001	SC3 > SC1, SC3 > SC4, SC3 > SC2
ED/Hosp in past 1 yr, n (%)	3/34 (9)	6/14 (43)	12/17 (71)	4/18 (22)	8/12 (67)	<0.0001	SC2 > SC1†, SC3 > SC1, SC3 > SC4
IgE, IU/ml, median 25th–75th percentiles	59 (16–110)	279 (146–852)	141 (47–731)	154 (69–414)	409 (36–632)	0.005	SC2 > SC1, SC4 > SC1, SC5 > SC1
Blood eos, mean ± SE (med %)	158 ± 24 (2)	354 ± 39 (4)	336 ± 34 (5)	250 ± 33 (5)	175 ± 41 (2.5)	<0.0001	SC2 > SC1, SC2 > SC5, SC3 > SC1, SC3 > SC5
BAL mac, mean ± SE (med %)	6,294 ± 1,080 (86.7)	8,727 ± 1,707 (91.4)	5,865 ± 1,505 (89.4)	7,534 ± 1,549 (87.3)	8,839 ± 1,843 (85.2)	0.52	NS
BAL lymph, mean ± SE (med %)	778 ± 128 (9)	517 ± 202 (5.8)	413 ± 179 (6.3)	1,080 ± 184 (10.6)	593 ± 219 (7.9)	0.09	NS
BAL neu, mean ± SE (med %)	149 ± 51 (1.7)	179 ± 82 (1.6)	180 ± 72 (1.9)	203 ± 74 (1.7)	497 ± 89 (2.7)	0.02	SC5 > SC1, SC5 > SC3
BAL eos, mean ± SE (med %)	52 ± 23 (0.3)	74 ± 37 (0.5)	84 ± 33 (0.7)	88 ± 33 (0.5)	44 ± 39 (0.2)	0.84	NS
FeNO, ppb, mean ± SE	23.0 ± 4.0	52.9 ± 6.9	71.5 ± 5.6	52.3 ± 6.0	22.9 ± 6.3	<0.0001	SC3 > SC5, SC3 > SC1, SC2 > SC5, SC2 > SC1, SC4 > SC1, SC4 > SC5
BA use, n (%)‡	13/29 (45)	10/10 (100)	11/12 (92)	14/17 (82)	9/9 (100)	<0.0001	SC2 > SC1, SC3 > SC1, SC5 > SC1
Daily BA use, n (%)‡	7/29 (24)	7/10 (70)	11/12 (92)	4/17 (24)	6/9 (67)	<0.0001	SC2 > SC1†, SC3 > SC1, SC3 > SC4
LABA use, n (%)‡	6/29 (21)	6/10 (60)	12/12 (100)	4/17 (24)	8/9 (89)	<0.0001	SC3 > SC1, SC5 > SC1, SC3 > SC4, SC5 > SC4
ICS high dose, n (%)	5/37 (13.5)	3/14 (21)	11/18 (61)	2/18 (11)	9/13 (69)	<0.0001	SC3 > SC1, SC3 > SC4, SC5 > SC1, SC5 > SC4
OCS, n (%)‡	3/36 (8)	2/14 (14)	8/18 (44)	1/18 (6)	5/13 (39)	0.035	NS
Omalizumab treatment, n (%)	1/36 (3)	0	3/18 (17)	0	3/13 (23)	0.026	NS
AEC T2 expression, geomean ± SE	1,563 ± 562	4,150 ± 715	5,402 ± 608	5,934 ± 666	1,793 ± 715	<0.0001	SC4 > SC1, SC4 > SC5, SC3 > SC1, SC3 > SC5, SC2 > SC1†
AEC EGR expression, geomean ± SE	896 ± 45	555 ± 57	275 ± 49	448 ± 53	479 ± 57	<0.0001	SC1 > SC3, SC1 > SC4, SC1 > SC5, SC1 > SC2, SC2 > SC3, SC5 > SC3†
BALcAMP expression, geomean ± SE	1,929 ± 106	1,397 ± 134	1,230 ± 114	2,120 ± 125	1,288 ± 134	<0.0001	SC4 > SC3, SC4 > SC5, SC4 > SC2, SC1 > SC3, SC1 > SC5, SC1 > SC2
BAL lymphocyte module, geomean ± SE	762 ± 140	438 ± 178	543 ± 151	1,183 ± 165	1,080 ± 178	0.008	SC4 > SC2, SC4 > SC3†

Definition of abbreviations: AA = African American; AEC = airway epithelial cell; AQLQ = asthma quality-of-life questionnaire; BA = β-agonist; BDR = bronchodilator response; BMI = body mass index; CRS = chronic rhinosinusitis; ED/Hosp = emergency department visits or hospitalizations; EGR = epithelial growth and repair; eos = eosinophils; Fe_NO = fractional exhaled nitric oxide; HCs = healthy control subjects; ICS = inhaled corticosteroids; LABA = long-acting β-agonist; lymph = lymphocytes; mac = macrophages/monocytes; med = median; mod = moderate; mod + ICS = moderate asthma on inhaled corticosteroids; NA = not applicable; neu = neutrophils; NS = not significant; O = other; OCS = oral corticosteroid; SC = subcluster; W = white.

SCs were determined by K means clustering of BAL and AEC gene expression. Differences in clinical characteristics were determined and reported here.

^*Significant intergroup differences listed achieved a Bonferroni-corrected P value of P ≤ 0.005.

^†Highly significant P values, although not reaching stringent Bonferroni correction (i.e., P ≤ 0.009).

^‡23% of participants were missing BA and LABA use data.

Discussion

In this hypothesis-generating study and the first to examine global gene expression in BAL cells in a large cohort enriched for SA, severity-related gene expression in BAL immune cells was determined in an unbiased manner by identifying and carefully deconvoluting gene expression changes caused by influences seldomly considered. Indeed, we found that gene expression was strongly influenced by cell-type proportions, age, sex, and race, and many of these findings were shared across AEC and BAL datasets and supported by prior studies. Moreover, a major finding was the discovery that BA exposure may have the strongest impact on gene expression, which may not relate to disease mechanisms or immediately matched by protein expression. Specifically, BA exposure strongly upregulates cAMP-related gene expression in those with low or no prior BA exposure, whereas high and prolonged BA exposure in SA likely desensitizes BAL monocyte/macrophage cells, affecting gene expression and potentially cellular function, a likely shared phenomenon in T2-hi and T2-lo SA phenotypes.

Woodruff and colleagues (20) showed that KLF9 and TFCP2L1 expression increases in AECs exposed to budesonide. Leigh and colleagues (63) likewise demonstrated that ZBTB16 increases in whole blood following budesonide inhalation, and that KLF9, TFCP2L1, and ZBTB16 are induced by budesonide in bronchial biopsies. We found all three of these regulatory transcriptional factors in the top approximately 30 genes in our asthma-severity gene set, indicating medication adherence in our population. In those with persistently high T2 gene expression and corresponding inflammatory characteristics (e.g., blood eosinophilia), these genes may serve as biomarkers for steroid resistance.

Similarly, BAL severity-related gene expression was strongly enriched for genes related to cAMP-signaling, as supported by gene pathway enrichment, biologic understanding of protein function (e.g., cAMP regulators CREM and PDE4B), and reproduction of the gene signature in a THP-1 cell model. In further support, Zhang and colleagues (64) also showed that many of these same genes were the most strongly induced by acute cAMP signaling in islet cells. Using a THP-1 cell model, we found the cAMP-related gene signature strongly increased with acute BA exposure and intracellular cAMP, but dropped to levels just above untreated cells after prolonged exposure. Thus, the difference in cAMP-related gene expression between subjects with SA and subjects with non-SA observed in vivo was likely caused by upregulation in nonsevere or BA-naive subjects (rather than a decline in cAMP among SA). Clinically, this finding made sense because every subject received an inhaled SABA just before sample collection. Alternatively, these findings also suggested that BAL monocyte/macrophage cells in SA are being desensitized by BAs. In support, Bachelet and colleagues (65) also showed that macrophages from subjects with asthma on long-acting BAs had lower levels of cAMP and were less reactive to SABA.

Yet, it is still to be determined whether immune cell BA desensitization is detrimental or beneficial to the SA disease state. Moriyama and colleagues (66) recently showed that signaling through the β₂ adrenoceptor in ILC2 cells negatively regulates the T2 inflammatory response and thereby airway inflammation. Alternatively, many genes in the cAMP gene signature have protective or antiinflammatory protein functions. For example, NR4A2 is rapidly induced by prostaglandin E₂ (67, 68) and is cytoprotective during oxidative stress (69). OSM is a secreted regulator of IL-6 (70). AREG is upregulated in the airway after injury and in subjects with asthma during an exacerbation, helping orchestrate tissue repair (71, 72). SOCS3 inhibits STAT3 activation, a proinflammatory transcription factor (73).

Beyond gene expression caused by medications and other factors, we identified many genes related to asthma severity that are of likely importance. Strongly increased in SA, KCNJ2 (#7) has been tied to eosinophilic inflammation (74). Keratin proteins KRT18P55 (#18) and KRT18 (#37) are strongly related to lung function in both AEC and BAL data sets, suggesting an important role in tissue remodeling. Moreover, KRT18P55 is located at the asthma susceptibility locus 17q12 (54). Alternatively, ID3 (#1 ranked in the BAL severity gene list), CD83 (#2), and RAB7B (#9) all encode important immune regulators that are strongly downregulated in SA. ID3 is a transcription factor that controls differentiation of CTLs and has a critical role in Foxp3 expression and delineation of tissue-resident T-regulatory cells (75, 76). RAB7B is an important regulator of IL-6-driven inflammation (45).

Clinically, we found that subjects with SA had higher BAL neutrophils, and neutrophil abundance correlated with elevated TNF-α and type 1 IFN-induced gene expression. Alternatively, patients without SA had comparatively higher alveolar macrophages, whereas gene expression suggested these cells were CD11c^high (alveolar) and producing higher levels of both EREG and HBEGF, two vital growth and antiinflammatory growth factors. Interestingly, HBEGF, which had gene expression findings validated by protein levels, is located at an asthma susceptibility locus near PDE4D, TSLP, IL4, IL13, and CD14 (5q31.3) (52), suggesting a dual role of genetics and BA use on expression/protein levels. Because HBEGF, EREG, and other epidermal growth factors are vitally important to reducing inflammation, preserving barrier function, and protection from acute lung injury (77, 78), augmenting its action may be a new mechanism for preventing, disrupting, or ameliorating asthmatic airway inflammation.

There are important limitations to this study. Adherence to prescribed medications was not formally addressed. Yet, corticosteroid-induced genes were strongly elevated with reported increasing steroid dosages (19). All data were collected and analyzed in cross-section. Thus, causative mechanisms for SA cannot be determined and proposed mechanisms here are therefore speculative and meant to be hypothesis-generating and to direct future research efforts. Although we believe our ELISA data to be of good quality, the detection of proteins from banked biospecimens of varied age presents significant technical challenge (e.g., variable protein content, quality, and degradation), which may be compounded by disease factors beyond the specific elements under investigation. Our results also demonstrate that changes in gene expression do not necessarily correlate with the abundance or activity of those genes’ translated proteins, so we must temper the assumptions made about how gene expression data might relate to pathophysiology at the cellular level. Finally, ELISAs were performed on only a small cohort of samples and may not be generalizable to the entire population, especially given asthma heterogeneity.

In summary, future gene expression research should consider all factors that affect gene expression, particularly the influence of any medication used before sampling. Prolonged and high BA exposure likely alters lung immune cell gene expression, and the impact of these gene changes on disease state is uncertain. Yet, our results indicate that secretion of antiinflammatory and repair-aiding growth factors (i.e., HBEGF and EREG), with implications for airway remodeling, may be modulated by asthma disease processes, therapy, or both.