Corticosteroid therapy and airflow obstruction influence the bronchial microbiome, which is distinct from that of bronchoalveolar lavage in asthmatic airways

Abstract

Background

The lung has a diverse microbiome that is modest in biomass. This microbiome differs in asthmatic patients compared to control subjects, but the effects of clinical characteristics on the microbial community composition and structure are not clear.

Objectives

We examined whether the composition and structure of the lower airway microbiome correlated with clinical characteristics of chronic, persistent asthma including airflow obstruction, use of corticosteroid medications, and presence of airway eosinophilia.

Methods

DNA was extracted from endobronchial brushings and bronchoalveolar lavage fluid collected from 39 asthmatic and 19 control subjects, along with negative control samples. 16S rRNA V4 amplicon sequencing was employed to compare the relative abundance of bacterial genera to clinical characteristics.

Results

Differential feature selection analysis revealed significant differences in microbial diversity between asthmatic and control brush and lavage samples. Lactobacillus, Pseudomonas, and Rickettsia were significantly enriched in asthmatic samples; while Prevotella, Streptococcus, and Vellonella were enriched in the control brushing samples. Generalized linear models (GLM) on brush samples demonstrated oral corticosteroid usage as an important factor affecting the relative abundance of the taxa significantly enriched in asthmatic patients. In addition, bacterial alpha-diversity in brush samples from asthmatic subjects was correlated with FEV1 and with the proportion of lavage eosinophils.

Conclusion

The diversity and composition of the bronchial airway microbiome of asthmatic patients is distinct from that of control, non-asthmatic patients and is influenced by worsening airflow obstruction and corticosteroid usage.

Introduction

Formerly thought to be sterile [1, 2], it is now clear that the lung is colonized by microbes from early infancy [3] and is exposed continuously to air as well as nasal, oropharyngeal, and gastrointestinal tract secretions. While normally low in biomass compared to other body sites such as the gastrointestinal tract, the ecology of the lung microbiome is diverse and complex [2, 4], and ecological dynamics of the microbial community, rather than just the presence of any individual species, may be an important component of disease pathogenesis.

Previous investigations suggest that the lung microbiome may contribute to the pathogenesis of asthma. Both bacterial exposure and greater diversity of environmental microbial exposures in early childhood diminish the risk of subsequent asthma or allergy [5, 6]. Commensal gastrointestinal microbiota may influence the development of atopy and asthma [7]. Infants whose upper airways were colonized with select organisms had an increased risk for asthma later in life [3]. Treatment of asthmatic patients with macrolide antibiotics may provide symptomatic relief for selected patients, though recent trials dispute this [8–13].

Recent reports suggest the microbiome in the lower airways may be different in patients with asthma. Hilty et al [14] demonstrated with endobronchial brushes that the genus Hemophilus, had a greater relative abundance, and the genus Prevotella had a reduced relative abundance in the bronchi of adult patients with either asthma or COPD compared to control subjects. Marri et al [15] demonstrated that three major phyla, Firmicutes, Actinobacteria, and Proteobacteria, accounted for over 90% of total 16S rRNA sequences in the sputum of subjects with mild asthma, and again Proteobacteria were significantly enriched compared to microbial communities in the sputum of control subjects [15]. Goleva et al [16] demonstrated that bronchoalveolar lavage (BAL) fluid of control subjects and subjects with corticosteroid ‘resistant’ or ‘sensitive’ asthma had significantly different relative abundances of many bacterial genera. Finally, using endobronchial brush samples previously collected from the ‘Macrolides in Asthma’ (MIA) study, Huang et al [17] demonstrated greater bacterial diversity in asthmatic samples compared to healthy control subjects, which correlated with bronchial hyperresponsiveness.

These data suggest that the lower airway microbiome may differ in asthma. However, differences in sample collection and sample location within the lung, phenotypes of asthmatic subjects, and differing use of corticosteroids are likely significant confounders that need to be considered to identify the microbial biomarkers of clinically relevant characteristics in asthma.

Our asthma clinical research program has been collecting lower airway samples from both carefully characterized asthma patients and from control subjects [18, 19]. To overcome previous sampling inconsistencies, we decided to elucidate and model the variability in the microbiome of different lung regions; specifically, the bronchial (endobronchial brushing, EB) and small (bronchoalveolar lavage, BAL) airways. We show significant differences in the relative abundance of bacterial taxa between asthmatic and control samples, and between EB and BAL samples, and we demonstrate that the asthmatic EB microbiome correlates with the degree of airflow obstruction. In addition, we highlight anatomical localization and corticosteroids usage as important factors influencing the relative abundance patterns of differentially abundant taxa. Our data, combined with previous studies, help set the stage for longitudinal studies, that can answer important questions about the role of the lower airway microbiome in asthma.

Methods

Subjects

This is a cross-sectional, retrospective study. Adult asthmatic and control subjects were recruited from among the participants in previous asthma genetic and airway biology studies that were originally conducted from 2011 to 2013. Approval for the retrospective use of samples generated from these subjects was obtained from the Institutional Review Board at the University of Chicago. All subjects provided written, informed consent at the time of their recruitment. Additional information including details of negative brush and reagent control sample collection is provided in an online data supplement.

Bronchoscopy was done using standard methods and conscious sedation. Because of the nature of this study with retrospective identification of subjects, no oral or nasal control samples were available for microbial analysis. Sample collection at bronchoscopy is described in detail in the online data supplement.

Sample processing

DNA was extracted using methods detailed in the online supplement. Primers specific for the V4 region (515–806 bp) of the 16S rRNA encoding gene were used to generate amplicons. Samples with sufficiently high DNA loading after amplification were sequenced in a paired-end 150 bp run using the Illumina MiSeq at the National Laboratory for High-Throughput Genome Analysis Core at Argonne National Laboratory (Argonne, IL). Paired end reads were quality trimmed and processed for OTU (operational taxonomic units) picking using UPARSE [20] pipeline, set at 97% sequence identity cutoff. Taxonomic status was assigned to high quality OTUs (<1% incorrect bases) using ‘parallel assign taxonomy’ script from QIIME software [21]. Multiple sequence alignment and phylogenetic reconstruction was performed using PyNast and FastTree [21]. Phyloseq package [22] was used for the ordination, alpha and beta diversity analysis.

Statistical analysis

Using the Phyloseq package [22], the OTU matrix was processed to remove OTUs containing less than 5 reads in order to reduce the PCR and sequencing based bias; then the OTU table was rarified to the minimum numbers of reads present in the smallest library.

For ANOVA and p-value correction, we performed post-hoc test using the Tukey Kramer test with effect size of “eta squared” and corrected for multiple test using the method of Benjamini and Hochberg. MetagenomeSeq [23] was used to identify the differentially abundant taxons across groups. Clinical data are expressed as the mean ± standard error of the mean. Two-group analysis was performed using Welch’s t-tests followed by Story’s FDR correction. We then used generalized linear regression model (GLM) to examine the contribution of patient demographic data on the abundance patterns of the differentially abundant bacterial genera.

Results

In this study we used specimens collected from 3 asthmatic and 19 control subjects for lung microbiome evaluation. Study subject characteristics are summarized in Table 1. As expected, asthmatic subjects had a lower Forced Expiratory Volume (FEV1) compared to controls (P=0.001), higher concentration of exhaled nitric oxide (P=0.05), and a larger percentage of eosinophils in BAL fluid (P=0.001). There were no identified differences based on gender, ancestry, or in the proportion of subjects with a serum IgE level > 100 U/ml, though asthmatic subjects were significantly older than the control population (44.2 vs 34.3 years old; P < 0.004; Table 1).

Table 1.

Demographic and clinical characteristics of study subjects

	Asthma (39)	Control (19)	p
Age, years (mean ± SE)	44.2 ± 1.8	34.3 ± 3.0	0.004
Sex (male : female)	11 : 28	8 : 11	—
Race (EA : AA : other)	21 : 17 : 1	7 : 10 : 2	—
Inhaled CS use (yes : no)	29 : 10	0 : 19	0.001
Oral CS use (yes : no)	17 : 22	0 : 19	0.001
Serum IgE, IU/ml (mean ± SE)	242 ± 61	200 ± 109	—
Serum IgE > 100 IU/ml (yes : no)	19 : 20	5 : 14	—
ENO, ppb (mean ± SE)	36.8 ± 7.6*	21.9 ± 3.2	—
BAL Eosinophils (proportion, mean ± SE)	4.2 ± 0.4	0.6 ± 0.3	0.001
BAL Neutrophils (proportion, mean ± SE)	5.4 ± 0.3	5.1 ± 0.5	—
FEV1 predicted (mean ± SE)	77.2 ± 3.1	94.4 ± 2.6	0.001
≥ 80% predicted (N, %)	16 (41)	18 (95)
60 – 80% predicted (N, %)	16 (41)	1 (5)
≤ 60% predicted (N, %)	7 (18)	0 (0)

Abbreviations are as given in the text.

• This data-point has one missing value.

Airway microbial community in endobronchial brushes vs. bronchoalveolar lavage

We sequenced bacterial 16S rRNA of EB and BAL samples from 39 subjects with asthma and 19 control subjects, 5 negative brush samples, and 7 negative reagent samples. A total of 1.1 million 16S rRNA V4 amplicon sequence reads (average sample depth=14,918) were generated, which following quality control (<0.1% incorrect bases) were clustered into 2,134 OTUs. Using a strict quality control, OTUs (n=403) with variance greater than 0.000001 were kept for further analysis.

Using the GreenGene database, 38 phyla were identified in our dataset; of these, six (Firmicutes, Proteobacteria, Bacteriodes, Fusobacteria, Acidobacteria and Actinobacteria) accounted for 85.9% of all sequences, unclassified sequences accounted for 9.1% of sequences, and the remaining phyla accounted for 1.3% of sequences. Similarly, while 303 genera were identified, the top 25 represented 64.7% of sequences (Figure 1).

Figure 1.

Relative abundance (%) of bacteria at the (A) phylum and (B) genera level identified in each sampling group.

Microbial community alpha-diversity was significantly different between EB and BAL samples within both specimen groups at the genus level using the Simpson index (diversity), Chao index (richness) and the Shannon index (evenness); where EB samples were significantly more diverse than BAL samples (Figure 2A). Microbial community structure variance (beta diversity) was greater between sampling location (EB vs BAL) compared to within each location (Figure S1). Figure 2B depicts the beta-diversity of EB and BAL samples for both asthma and control subjects combined. Two-group analysis (Welch’s t-test with Story’s FDR correction) demonstrated a significant difference between the microbial community beta-diversity in the EB and BAL samples for both the asthma patient group (P < 0.005) and the control patient group (P < 0.005) (Figure 2B). Significant differences in the relative abundance of several genera were observed between EB and BAL samples. Lactobacillus and Pseudomonas were significantly less abundant while Prevotella and Streptococcus were significantly more abundant in EB-Asthma compared to BAL-Asthma samples (Figure 2C, Table 2). Similar trends were observed when comparing EB-Control and BAL-Control samples (Figure 2D, Table 3), suggesting that this lung biogeography is absolute.

Figure 2.

Analysis of microbial communities by sampling location. A. Alpha diversity of all samples according to location.. B. Ordination plot of principle component analysis of beta-diversity of EB, BAL, and negative control samples. C. and D. Significantly, differentially abundant genera in EB-Asthma versus BAL-Asthma (C) and EB-Control versus BAL-Control (D) samples.

Table 2.

Significantly differentiated relative abundance proportions of genera of EB-Asthmatic and BAL-Asthmatic subjects

Genus	BAL-Asthma	EB-Asthma
Lactobacillus	38.21 ± 10.88	23.26 ± 13.12*
Pseudomonas	24.64 ± 8.29	15.73 ± 9.95*
Streptococcus	1.81 ± 4.35	10.40 ± 10.48*
Prevotella	2.95 ± 9.25	11.09 ± 10.46*
Fusobacterium	0.29 ± 0.85	3.93 ± 6.36*
Rickettsia	7.67 ± 4.01	4.45 ± 3.30*
Veillonella	0.87 ± 4.35	3.07 ± 3.03*
Actinomyces	0.15 ± 0.34	1.80 ± 1.98*
Haemophilus	0.16 ± 0.40	1.53 ± 2.05*
Leptotrichia	0.08 ± 0.42	1.37 ± 2.24*
Rothia	3.14 ± 2.14	2.28 ± 1.43*

Proportions represented as mean ± std. dev.

^*p < 0.01

Table 3.

Significantly differentiated relative abundance proportions of genera of EB-Control and BAL-Control subjects

Genus	BAL-Control	EB-Control
Streptococcus	38.23 ± 15.09	18.05 ± 10.43*
Veillonella	4.93 ± 8.04	18.56 ± 10.96*
Actinomyces	20.05 ± 8.97	10.51 ± 6.79*
Pseudomonas	0.39 ± 0.74	3.51 ± 2.84*
Prevotella	1.59 ± 2.67	4.54 ± 2.53*
Lactobacillus	3.26 ± 2.00	0.85 ± 0.68*

Proportions represented as mean ± std. dev.

^*p< 0.01

In EB samples, two-group analysis revealed that Pseudomonas was significantly more abundant in asthmatics compared to controls, while Actinomyces and Prevotella were significantly less abundant in asthmatic samples (Figure 3A, Table 4). In BAL samples, Rickettsia, Staphylococcus, Marinobacter, and Novosphingobium were enriched while Sphingobium was significantly less in asthmatics compared to controls (Figure 3B, Table 5).

Figure 3.

Differentially abundant genera across asthmatic and control subjects. A. Significantly differential genera between EB-Asthmatics and EB-control samples. B. Significantly differential genera between BAL-Asthmatic and BAL-Control samples.

Table 4.

Significantly differentiated relative abundance proportions of genera of EB-Control and EB-Asthma subjects

Genus	EB-Asthma	EB-Control
Prevotella	8.67 ± 9.18	14.84 ± 10.13*
Pseudomonas	10.16 ± 5.76	7.25 ± 3.97*
Actinomyces	1.34 ± 1.57	2.76 ± 2.37*

Proportions represented as mean ± std. dev.

^*p < 0.01

Table 5.

Significantly differentiated relative abundance proportions of genera of BAL-Control and BAL-Asthma subjects

Genus	BAL-Control	BAL-Asthma
Rickettsia	4.94 ± 3.04	7.66 ± 4.01*
Sphingobium	0.99 ± 0.86	0.47 ± 0.63*
Staphylococcus	0.07 ± 0.18	0.40 ± 0.60*
Marinobacter	0.01 ± 0.06	0.26 ± 0.61*
Unclassified	0.03 ± 0.11	0.28 ± 0.54*
Novosphingobium	0.00 ± 0.00	0.12 ± 0.32*

Proportions represented as mean ± std. dev.

^*p < 0.01

Clinical influences on relative abundance of specific genera

Moving forward, we chose to focus solely on EB samples as they had the most robust difference compared to negative controls (both brush and reagent). To explore the relationship between clinical characteristics and the lung microbiome, we used simple linear regression comparing the microbiome community and clinical variables in EB samples. A significant decrease in the relative abundance of Firmicutes, Bacteroidetes, and Actinobacteria was demonstrated for EB samples in subjects with the lowest FEV1 levels (Table 5). Also at the genus level, Streptococcus, Prevotella, Gemella and Veillonella were significantly reduced in EB samples from the patients with the lowest FEV1 levels (Table 5). Finally, the beta-diversity between samples in the lowest FEV1 group was significantly greater than for the highest FEV1 group. This suggests that FEV1 levels influence the diversity and structure of the EB-associated microbiome.

Many of the asthmatic subjects in this study used an inhaled corticosteroid on a daily basis, and 48% were using oral corticosteroids at the time of bronchoscopy. Subjects were stratified according to no corticosteroid (CS) use, use of inhaled CS (ICS) only, or both ICS and oral CS (OCS) use. No subjects used OCS without ICS. In the EB samples, there was a decrease in the relative abundance of both Bacteroidetes and Fusobacteria, and an increase in Proteobacteria, based on OCS use (Table 6). At the genus level, generalized linear modeling (GLM) predicted oral corticosteroids as being a significant influence in the relative abundance proportions of Pseudomonas, Rickettsia, Prevotella, Lactobacillus, and Streptococcus (Figure 4). There was a decreased relative abundance of Prevotella and an increase of Pseudomonas based on increasing CS use (Table 7). These data suggested that OCS use was associated with significant changes in the relative abundance of the EB microbial community. Additionally, GLM revealed a significant relationship between relative abundance proportion of Rickettsia and exhaled nitric oxide (ENO) and BAL eosinophils (Figure 4).

Table 6.

Relative abundance (proportion) of phyla and genera in asthmatic endobronchial brush samples stratified by FEV1.

Phyla, FEV1 as % predicted	> 80%	60 – 80%	< 60%
Firmicutes	33.6 ± 1.4	32.3 ± 2.2	24.0 ± 1.3*
Proteobacteria	30.7 ± 3.5	37.4 ± 3.3	43.8 ± 6.8
Bacteroidetes	15.1 ± 2.3	11.4 ± 2.1	6.7 ± 2.7*
Fusobacteria	5.0 ± 1.3	3.1 ± 1.0	5.6 ± 5.3
Actinobacteria	4.7 ± 0.7	3.7 ± 0.5	2.5 ± 0.4†

Genera, FEV1 as % predicted	> 80%	60 – 80%	< 60%
Lactobacillus	11.9 ± 1.7	16.3 ± 1.8	18.6 ± 2.9*
Streptococcus	10.1 ± 1.5	8.2 ± 2.6	2.2 ± 1.3*
Prevotella	9.7 ± 2.1	7.4 ± 2.2	1.9 ± 1.5†
Pseudomonas	9.1 ± 1.4	9.8 ± 1.4	12.8 ± 2.8
Veillonella	3.2 ± 0.6	2.0 ± 0.6	0.3 ± 0.2*
Gemella	1.7 ± 0.3	1.1 ± 0.4	0.1 ± 0.0*

^*P < 0.05;

^†P < 0.01; versus high FEV1 by Metastats.

Figure 4. Generalized linear model fitting analysis across significantly important taxa (Genera as predicted by MetagenomeSeq) across Asthma samples.

For each genus GLM model was constructed, validated and analyzed using analysis of variance (ANOVA). * P < 0.05.

Table 7.

Relative abundance (proportion) of phyla and genera in asthmatic endobronchial brush samples stratified by corticosteroid use.

Phyla, corticosteroid (CS) use	No CS	ICS only	ICS & OCS
Proteobacteria	28.4 ± 2.6	39.1 ± 2.7*	37.4 ± 3.5*
Bacteroidetes	17.6 ± 1.8	12.0 ± 1.6*	10.1 ± 2.1†
Fusobacteria	6.0 ± 1.3	3.0 ± 1.1*	2.2 ± 0.9*

Genera, CS use	No CS	ICS only	ICS & OCS
Prevotella	12.1 ± 1.7	6.2 ± 1.3*	6.6 ± 2.1*
Pseudomonas	7.2 ± 0.9	10.4 ± 1.8	11.0 ± 1.3*
Veillonella	3.1 ± 0.4	1.4 ± 0.4*	2.4 ± 0.7

^*P < 0.05;

^†P < 0.01; versus no CS use by Metastats.

Clinical influences on endobronchial microbiome diversity

We used mixed model regression to determine what clinical factors associated with changes in diversity in EB asthmatic samples. Inverse Simpson and Shannon indices in EB samples correlated with a FEV1 in the lowest quartile, oral corticosteroid use, European ancestry, and BAL eosinophil proportion in the highest quartile. In this model one subject was excluded as an outlier after testing via jackknife residual and Cook’s distance; doing so corrected skew significantly without affecting fit. A best-fit model for each index is shown in Figure 5. Neutrophil proportion in the BAL fluid was not a significant predictor of diversity or evenness.

Figure 5.

Mixed model regression demonstrating a relationship between Shannon (top panel) and Inverse Simpson (bottom panel) measurements of diversity and FEV1, corticosteroid use, white ancestry, and BAL eosinophils for EB-Asthmatic samples only.

Discussion

In this study we compared the microbial communities in the central airways and peripheral lung in both asthmatic and control subjects. We examined the relationship of corticosteroid use and degree of airflow obstruction, two key clinical parameters of asthma, on these communities. We demonstrate that both the diversity and relative abundance the microbiome in peripheral airways (EB) differs significantly from that of the samples taken from central airways (BAL) and negative brush and reagent controls in both asthmatics and non-asthmatics. The BAL specimens showed slight differences between the negative control samples. In EB samples of asthmatic subjects there are significant differences in diversity and relative abundance based on both corticosteroid use and FEV1. These data demonstrate that the lower airway microbiome in asthma differs by location, disease severity, and corticosteroid use.

We demonstrated a greater number of bacterial taxa present in our samples than that noted in previous studies. In contrast to the study of Huang, et al [17], we report a significant number of taxa, identified sequences, and observed unique species in control subjects without demonstrable airways disease. We note that approximately two-thirds of asthmatic patients in the Huang study had not used corticosteroids within 2 years of study. Whether this difference accounts for the observed differences is not clear. Goleva et al [16] demonstrated equal bacterial numbers in the BAL fluid of both normal and asthmatic subjects that was approximately one-half log lower than that observed in our study with similar coverage, richness and diversity.

There are differences in our study versus previously published work relating to the effect of corticosteroids on the microbiome. Goleva et al [16] compared asthmatic subjects who were (“sensitive”) or were not (“resistant”) responsive to a course of treatment with oral corticosteroids for seven days compared to normal subjects. Their study demonstrated a reduction in commensal genera such as Prevotella and Veillonella and an increase in both Actinobacteria and Proteobacteria in the asthmatic subjects. We demonstrated significant differences based on corticosteroid treatment, particularly the combination of ICS and OCS, in alpha and beta diversity, with an increased abundance of Proteobacteria and the genus Pseudomonas, and decreased abundance of Bacteroidetes, Fusobacteria and of Prevotella. However, we did not specifically test CS responsiveness, and the Goleva study did not segregate data based on current or recent CS use. Though our studies may not be directly comparable, both suggest that corticosteroid therapy in asthmatic subjects is associated with changes in relative abundance in the lower airway microbiome.

We also demonstrated changes in the lower airway microbiome in asthmatics based on FEV1 as a marker of airflow obstruction. Subjects with an FEV1 < 60% predicted had lower alpha diversity, higher beta diversity, and a lower relative abundance of the phyla Firmicutes, Bacteroidetes and Actinobacteria, and of the potential pathogenic genus Streptococcus and the commensal genera Veillonella and Prevotella, compared to asthmatic subjects with an FEV1 > 80% predicted. There was a corresponding increase in the phylum Proteobacteria and of the genus Pseudomonas but these did not reach statistical significance. Huang et al [17] demonstrated that increased diversity in EB community composition was correlated with greater bronchial hyperresponsiveness. Their data and ours taken together suggest that airway reactivity and airflow obstruction, both cardinal features of chronic asthma, are associated with changes in microbial burden, diversity and membership.

Previous studies of the airway microbiome in asthma have examined microbial 16S-rRNA presence in sputum [15], BAL [4, 14, 16] and/or EB [14, 17]. In these studies five major phyla, Firmicutes, Proteobacteria, Bacteroidetes, Actinobacteria, and Fusobacterium, account for >90% of identified sequences, a finding we also observed in both EB and BAL samples. Marri et al [15] suggested, based on sputum analysis, that the microbiome of patients with mild asthma is similar to that of patients with more severe disease, and demonstrate a greater abundance of Proteobacteria, and lower abundance of both Firmicutes and Actinobacteria, in sputum from asthmatic subjects. Our data from lower airway samples suggest otherwise: for both EB and BAL samples, the five major phyla are comparable between asthmatic and control subjects though both are different than the negative controls. To date no comprehensive study has been done to examine the relation of sputum or the oral microbiome to lower airway microbiomes in asthmatic subjects.

The presence of bacteria in the lower airways raises the possibility that either specific microbes or the combined effects of the microbial community could contribute to airway inflammation in asthma via modulation of the innate immune system. In support of the latter, some asthmatic patients have a significant proportion of neutrophils in bronchial wall biopsies or in BAL fluid [24, 25] or in sputum [26–28]. However, we did not find a correlation between the neutrophil proportion in BAL and changes in either diversity or in relative abundance in asthma EB or BAL microbes at any level.

Heterogeneity in asthma is well described and there are multiple proposed asthma phenotypes based on age, the presence of concurrent allergies, elevation in serum IgE, the cellularity of BAL fluid (eosinophil- or neutrophil-enriched), response to corticosteroid therapy, and obesity [29, 30]. Our study examines two distinct groups of asthmatic subjects: those with relatively mild airflow obstruction who were well controlled on either ICS therapy or with no CS therapy at all, and those with more severe airflow obstruction and a need for OCS therapy. Asthmatic subjects with the greatest degree of obstruction and those using OCS therapy had differences in relative abundance of potentially pathogenic taxa compared to asthmatic subjects with milder disease. Further, eosinophil proportion, but not neutrophil proportion, in BAL fluid correlated with microbial diversity in the EB microbiome in asthmatics. Whether these differences persist over time or change with therapy or airflow obstruction is not clear.

Our study has several limitations. We note that changes in microbial communities do not establish causality. While our study is about equal to or greater in size compared to previous studies [14–17], it is still small and does not include patients with the full spectrum of asthma phenotypes. It is not clear how remote use of antibiotics, corticosteroids, and other therapies for asthma influence the airway microbiome. Other features of the airway such as the composition or rheology of mucous in the sol layer, the presence and activity of macrophages, the presence of inflammatory mediators, and the secretion of local host-defense factors may also influence the composition of the microbial communities and would not be detected in our study.

One important consideration is that the majority of asthmatic patients in the group with greatest airway obstruction, as measured by FEV1, are patients treated with oral corticosteroids. This partial concordance may account for similarities within the correlation of our data. In our asthmatic patient cohort, FEV1 and corticosteroid use did not significantly correlate though was very close (p = 0.57). For this reason we chose to treat the FEV1 and corticosteroid variables separately in our data analyses.

We also note that there may be organisms that participate significantly in the lower airway microbial community that would not be considered in our analysis given the observational and analytical thresholds applied. This is a common issue in microbiome studies [31] and awaits advances in understanding of how to assess the functional significance of sparse microbiota. Also, our study was performed at a single institution in patients drawn from a single major metropolitan center. Differences in geography, household exposures, pet ownership and environment all may lead to differences in the microbial communities recovered [2] and account for some of the differences between our and previous studies.

The retrospective nature of our study precluded collection of upper airway and oral samples. Previous studies have demonstrated significant commonality between upper and lower airway microbiomes [4, 14, 32, 33], and we have no reason to believe that our study would differ in this regard. Likewise, Segal et al recently demonstrated in normal subjects, using a two-bronchoscope method in which the first instrument samples the supraglottic airway and the second the lower airways, that little carry-over of organisms occurs from the bronchoscope [33].

In summary, we demonstrate clear differences in the microbiome recovered from endobronchial brushes and bronchoalveolar lavage in both asthmatic and control subjects. Asthmatic subjects differ in their EB microbiome based on corticosteroid use and degree of airflow obstruction. Understanding the role of the microbiome in asthmatic airways and the changes over time may lead to novel, targeted therapies that improve asthma control.

As ours is a mechanistic article, we provide the following bulleted sentences:

• The microbiome of central airways in asthmatic subjects has a lower diversity and greater abundance of key bacterial pathogens such as Pseudomonas compared to control subjects

• Both diversity and relative abundance changes of pathogens are related to corticosteroid use and worsening airflow obstruction as measured by FEV1

• The central airway microbiome from endobronchial brushes is substantially different than that of the peripheral airways from bronchoalveolar lavage

Abbreviations

EB

endobronchial brushing

BAL

bronchoalveolar lavage

FEV

forced expiratory volume

ICS

inhaled corticosteroid

OCS

oral corticosteroid

CS

corticosteroid

ENO

exhaled nitric oxide