Relationship between inflammatory status and microbial composition in severe asthma and during exacerbation

In T2‐mediated severe asthma, biologic therapies, such as mepolizumab, are increasingly used to control disease. Current biomarkers can indicate adequate suppression of T2 inflammation, but it is unclear whether they provide information about airway microbial composition. We investigated the relationships between current T2 biomarkers and microbial profiles, characteristics associated with a ProteobacteriaHIGH microbial profile and the effects of mepolizumab on airway ecology.


| INTRODUC TI ON
Characterisation of heterogeneous pathophysiological mechanisms within severe asthma 1 along with the development of multiple adjunctive therapies for the T2-mediated endotype have transformed the care of real-world severe asthma populations. Biologic agents targeting T2 cytokines, such as the anti-IL5 agent mepolizumab, have led to consistent improvements in asthma exacerbation rates 2,3 and reduced oral corticosteroid burden. 4 However, some patients demonstrate a suboptimal response to these therapies, with studies indicating that some 'loss-of-control' events are not mediated by T2 inflammation. 5 In a small proportion of asthmatics, there is a persistent absence of T2 inflammation. 6 With no targeted therapies currently licensed in the T2-low endotype, identifying and understanding non-T2 drivers of disease remains an unmet clinical need.
In severe asthma, drivers of non-T2 disease include airway infection and dysbiosis, 5,7 defined as a compositional or functional change in the airway microbiome compared to health. Molecular approaches to define the airway ecology typically include microbiomics and polymerase chain reaction (PCR) for specific airway pathogens. Previously we have proposed the ratio of the predominant phyla in the airway, Proteobacteria (including the genera Haemophilus, Moraxella and Pseudomonas) and Firmicutes (including Streptococcus), as a simple biomarker (P:F ratio) to reflect the overall pattern of the microbiome. 7 In subjects with asthma, there is heterogeneity in the composition of the microbiome. Cluster analyses have demonstrated two microbiological groups, with a smaller subgroup of ~20%-25% characterised by lower alpha diversity indices and predominance of potentially pathogenic organisms representing the Proteobacteria phylum (P:F high), especially H. influenzae, compared to the larger subgroup. 8,9 These differences could be important clinically as specific phyla or genera that differ between these groups are associated with adverse clinical features, such as worsening symptoms 10 or bronchial hyperresponsiveness. 11 Additionally, airway microbial composition may impact responsiveness to specific therapies, such as corticosteroids 12 or macrolides. 13 Inflammatory indices also differ between these microbiological subgroups. Associations between airway inflammation and airway microbial composition are recognised. [14][15][16] However, the relationship between inflammation, as indicated by more widely accessible T2 biomarkers, such as FeNO or blood eosinophils, and airway ecology is not fully established, especially in response to treatment. This has led to some concern that iatrogenic suppression of eosinophils could affect the immune response to pathogens 17,18 and increase the risk of clinical infection.
In this exploratory analysis of sputum samples collected in two trials previously published by the Medical Research Council Refractory Asthma Stratification Programme (RASP-UK) consortium, 5,6 we hypothesised that levels of T2 biomarkers (FeNO, sputum and blood eosinophils) are associated with aspects of airway microbial composition. We also aimed to (1) provide further characterisation of microbiological subgroups and (2) study longitudinal samples, where available, to explore the impact of mepolizumab on the airway microbiome.

| Study design and participants
All samples that were adequate for microbiota analysis, collected from two previously completed clinical trials (NCT03324230 and NCT02717689) were combined for the current analysis ( Figure S1).
Both trials recruited patients with GINA step 4/5 asthma, 19 aged between 18 and 80 years, from specialist severe asthma centres across the UK participating in the RASP-UK consortium. All participants provided written informed consent. Sputum induction was performed according to an established protocol. 20 The MEX trial 5 (NCT03324230) was a prospective observational trial performed across four UK severe asthma centres, using sputum samples to characterise exacerbations in subjects on mepolizumab.
In this trial, an exacerbation was defined as severe asthma symptoms worsening outside of a patient's normal daily variation and occurring any time after the initial dose of mepolizumab, at which point participants attended their clinical site for assessment before commencing rescue treatment. 5 Induced or spontaneous sputum samples were collected preceding treatment with high dose oral corticosteroids or antibiotics in the context of an acute asthma exacerbation, whereas stable sputum samples, collected prior to mepolizumab initiation and when clinically stable after at least 12 weeks of treatment, were induced. All participants were deemed suitable to receive mepolizumab by a severe asthma multi-disciplinary team in accordance with UK clinical guidance, with eligibility criteria including both blood eosinophilia and either maintenance oral corticosteroid use or an annual exacerbation rate of ≥4 (supplementary methods and previously published in full 5 ).
The RASP-UK biomarker stratification study 6 (NCT02717689) was a prospective parallel group trial across 12 UK severe asthma centres that compared strategies of corticosteroid adjustment, either using a composite score derived from three T2 biomarkers (blood eosinophils, fractional exhaled nitric oxide [FeNO] and periostin) or an algorithm based on symptoms, physiology and exacerbations. Participants were required to demonstrate a FeNO <45 ppb to enrich a population in whom steroid reduction was possible. Sputum samples collected were induced at baseline and spontaneous at exacerbation (supplementary methods and previously published in full 6 ). In this trial, exacerbation sputum samples were obtained including if rescue treatment had been commenced. Pre-processing of raw sequence reads to remove artefacts and chimeras, merge the paired ended reads and generate the amplicon sequence variants (ASV) feature table were done using DADA2 program. 22 All ASVs were aligned with mafft 23 (via q2-alignment) and used to construct a phylogeny with fasttree2 24 (via q2-phylogeny).

| Microbiological analysis
Sample rarefication at 9700 read depth was performed as it showed good coverage of the bacterial diversity across our samples without any loss of samples. Post-rarefaction, alpha-diversity metrics for within sample comparison, beta diversity metrics using UniFrac distance measure 25 and Principle Coordinate Analysis (PCoA) were estimated using q2-diversity. Taxonomy was assigned to ASVs using the q2-feature-classifier 26 using the classify-sklearn naïve Bayes method against the Greengenes 13_8 99% OTUs reference sequences. 27 PERMANOVA and PERMDISP analysis 28 were applied to beta diversity measures to explore differences in between and within-group variation (dispersion) between the biomarker groups, respectively, and FDR <0.05 was considered significant for multiple group comparison.
Proteobacteria to Firmicutes (P:F) ratio was calculated for each individual sample using the sequencing data, by dividing the total proportion of the sample belonging to the phylum Proteobacteria by the proportion belonging to the phylum Firmicutes. Sequence data are deposited at the National Center for Biotechnology Information Sequence Read Archive (Bioproject: PRJNA779201).

| Statistical analysis
This was a secondary analysis of pre-existing data, and hence no sample size calculation was conducted. Descriptive statistics are presented as means (standard error), medians (IQR) or counts (%) unless otherwise stated. Microbial composition in biomarker subgroups were evaluated on the basis of FeNO (≤20 ppb, 21-49 ppb and ≥50 ppb for low, mid and high FeNO subgroups, respectively) and serum C-reactive protein (CRP) (>5 mg/L and ≤5 mg/L for high and low CRP subgroups, respectively) in accordance with thresholds applied in the MEX trial, 5 sputum eosinophils (≥2% or <2% of total cells counted for eosinophilic and non-eosinophilic subgroups, respectively), 19 sputum neutrophils (≥65% or <65% for neutrophilic and non-neutrophilic subgroups, respectively) 29 and blood eosinophils (<0.15 × 10 9 /L, 0.16-0.29 × 10 9 /L, ≥0.3 × 10 9 /L for low, mid and high subgroups, respectively). 19,30 Univariate analyses were conducted using paired and unpaired t-tests, Mann-Whitney U and Wilcoxon signed-rank test, and chi-square tests for parametric, non-parametric and nominal data respectively. ANOVA (parametric) and Kruskal-Wallis (non-parametric) tests were additionally used to compare across three FeNO subgroups. As data for assessed exacerbations were collected prospectively during exacerbations, accompanying clinical data where sputum samples were obtained were rarely missing. Therefore, we conducted all analyses under a complete-case framework. Analyses were conducted using SPSS (Version 26.0; IBM Corp).

| RE SULTS
We analysed all sputum samples that were adequate for microbiota analysis with paired biomarker data. The number of subjects and samples included in the analysis are shown in Figure S1. Clinical characteristics for the subjects with stable severe asthma and those studied during exacerbation are demonstrated in Table S1. Clinical characteristics of the participants included from each study are compared in Table S2. We compared microbial composition in those receiving or not receiving maintenance oral corticosteroids, with no difference in beta-diversity between these groups ( Figure S2).

| Microbial diversity and composition in biomarker subgroups at stable state
The number of subjects and samples for the biomarker subgroup analyses is shown in Table S3.
FeNO was available in 115 subjects with sequencing data at stable state and deemed to be low (≤20 ppb), mid (21-49 ppb) or high (≥50 ppb). 5 Clinical characteristics are demonstrated in Table 1.    Note: Data is shown as mean (SEM), median (IQR) and n (%) unless otherwise stated.
b Both trials permitted inclusion of subjects on background macrolide (or other antimicrobial) therapy.
c Geometric mean (95% CI). In these cases, statistical tests applied were t-test (including for geometric mean), Kruskal-Wallis, and chi-squared test respectively. group ( Figure 1A,B  Figure 1D).
95 subjects with sequencing data were able to provide sputum for differential cell count at stable state. There were 42 eosinophilic and 62 non-eosinophilic samples based on sputum eosinophil count ≥2% or <2% of total cells counted. There was no difference in UniFrac distance to centroid between sputum eosinophil subgroups. Alpha diversity measures or microbial composition at phylum or genus level did not differ between subgroups ( Figure S3). There were no notable differences in microbial composition according to the stable blood eosinophil-defined biomarker subgroups: low (<0.15 × 10 9 /L, n = 62), mid (0.16-0.29 × 10 9 /L, n = 35), high (≥0.3 × 10 9 /L, n = 31) ( Figure S4).

| Microbial diversity and composition in biomarker subgroups at exacerbation
The same low (≤20 ppb), mid (21-49 ppb) and high (≥50 ppb) FeNO classification was applied to FeNO measurements made at exacerbation. Clinical characteristics are presented in Table 2 Figure S5). Subjects with low, mid and high blood eosinophils at exacerbation had no differences in microbial diversity or composition ( Figure S6).
Microbial profiles were also compared between subjects with neutrophilic (sputum neutrophils ≥65%) and non-neutrophilic (sputum neutrophils <65%) exacerbations. There was no difference in beta-diversity between these 2 subgroups. Serum CRP level at ex- difference in beta-diversity between those with CRP >5 mg/L versus CRP ≤5 mg/L ( Figure S7).

| Clinical characteristics associated with a Proteobacteria-dominant microbial profile
The heterogeneity (dispersion) in microbial composition in the low and mid FeNO subgroups was driven by an increased Proteobacteria and reduced Firmicutes dominance in a number of samples in these subgroups. Therefore, we utilized the P:F ratio to determine the Proteobacteria HIGH samples and their associated clinical characteristics in the stable FeNO subgroups. The 75th quartile value in the low FeNO group was used as a threshold to identify Proteobacteria HIGH samples ( Figure S8)

| The effect of mepolizumab on airway microbial composition
Paired sputum samples before initiation of mepolizumab and after ≥12 weeks of treatment were available for 13 subjects from the MEX trial. Clinical characteristics are presented in Table 3, along with baseline and follow up symptom scores, physiology and biomarkers, demonstrating a reduction in blood eosinophils consistent with the known drug effect. From baseline to follow up visit, there was no change in total bacterial load as measured by total 16S rRNA gene copy numbers.

Genus-level microbial composition data and FeNO subgroup
is shown for the 13 individuals in Figure 3. There was no significant increase in pathogenic organisms, specifically Haemophilus or Moraxella, in response to mepolizumab. Similarly, there was no significant change in P:F ratio, and no cases in which a baseline Proteobacteria LOW subject became Proteobacteria HIGH at follow up.

| DISCUSS ION
In this analysis, the airway microbiome was evaluated in a wellcharacterised, real-life severe asthma population. There was no marked difference in microbial composition comparing subgroups with high or low sputum or blood eosinophils; however, high FeNO level both when stable and at exacerbation indicated a Proteobacteria-low microbial profile with high alpha-diversity and low intersubject variability. Proteobacteria HIGH individuals tended to have a longer duration of asthma but no other defining clinical characteristics. Additionally, we were able to demonstrate, in a limited subset, that mepolizumab treatment did not lead to an increase in airway bacterial load or relative abundance of pathogenic organisms.
Previous studies have demonstrated an association between inflammatory phenotype and airway microbiome, with the greatest differences observed between neutrophilic and eosinophilic phenotypes. 14 Lower alpha-diversity is widely reported in neutrophilic asthma, with a relative abundance of pathogenic organisms including Haemophilus and Moraxella. 15,16,32 Associations between microbial composition and eosinophilic inflammation are less clear and appear to differ in relation to disease severity. 10 [35][36][37][38] FeNO is an attractive biomarker of T2 inflammation in view of its wide accessibility and relative ease of measurement and has recently demonstrated potential utility in determining inflammatory phenotype at acute exacerbation in severe asthma. 5 FeNO has previously been shown to correlate with the alpha-diversity of the mycobiome, but was unrelated to the microbiome, in asthma. 38 In our study, a FeNO ≥50 ppb indicated a subgroup with low intersubject variability in microbial profiles, with a low relative abundance of pathogenic organisms, including Haemophilus; increased abundance of which has been shown to predict response to macrolide therapy. 13 39 and in studies of T2 and T17 epithelial gene expression clusters, FeNO varied more between clusters than blood, sputum and tissue eosinophils, although between-cluster differences in sputum eosinophils were also significant. 40  Abbreviations: ACQ-5, asthma control questionnaire-5; BDP, beclometasone diproprionate equivalent; BMI, body mass index; FEV 1 , forced expiratory volume in 1 s; FeNO, fractional exhaled nitric oxide; FVC, forced vital capacity; ICS, inhaled corticosteroid; LABA, long acting betaagonist; LAMA, long acting muscarinic-antagonist; LTRA, leukotriene receptor antagonist. a Geometric mean (95% CI). In these cases, statistical tests used were paired t-tests or Wilcoxon signed-rank tests respectively. exploration of this in future trials. In contrast to the high FeNO subgroup, microbial profiles in those with lower T2-biomarkers (FeNO ≤50 ppb) were highly dispersed, with only ~20% meeting our definition of Proteobacteria HIGH . This suggests that T2-low asthma encompasses various pathobiological mechanisms, with the best strategies to identify and target these remaining unclear.
Characterisation of microbial profiles in poorly controlled asthmatics is increasingly important in the pursuit of precision medicine, as evidence suggests that these may influence the response to specific therapeutic strategies, such as low dose macrolides. 13 As observed in other studies, our results indicate that a Proteobacteriadominant microbial profile was associated with sputum neutrophilia, 14,15 but also a longer duration of disease. In a previous cluster analysis, subjects with more pathogenic organisms at stable state had a younger age at asthma diagnosis 8   suggesting that serum biomarkers may be less reflective of inflammatory changes in the airway compartment.
Notwithstanding the limited sample size, we found that mepolizumab treatment was not associated with an increase in total airway bacterial load or increased relative abundance of potentially patho-  for supporting the RASP-UK Consortium.