Introduction

Chronic respiratory diseases (CRDs) such as bronchiectasis [1, 2], chronic obstructive pulmonary disease (COPD) [2] and asthma [3] are characterized by perturbed airway microbiome which leads to heightened inflammatory responses [4]. Dysbiosis, characterized by expansion of pathogenic bacteria and reduction in microbial diversity, correlates with the disease severity of bronchiectasis [1] and inflammatory subtypes of COPD [5]. The microbial compositions could also predict the clinical trajectory and outcomes in CRDs such as bronchiectasis [1].

Lower airways are the niche of microbiome, the reservoir of antibiotic resistance genes (ARGs), particularly in patients with structural lung diseases [6,7,8,9,10], prolonged or repeated exposures to antibiotics [11, 12] or acute viral infections [13]. Previous studies have documented the high abundance of ARGs which correlated with bacterial loads in COPD [14] and the shared macrolide-related resistome among bronchiectasis, COPD and asthma [15, 16]. In bronchiectasis, prolonged exposures to azithromycin has been associated with the selection of drug-resistant Pseudomonas aeruginosa [17]. Meanwhile, the growing burden of antibiotic resistance because of extensive antibiotic exposures has become concerning globally [18, 19], which could readily result in blunted antimicrobial capacity and confer selective pressures on antibiotic-resistant bacteria [17]. The clinical correlates (disease severity, clinical and inflammatory subtypes) of ARGs in CRDs remain unclear.

Bronchiectasis is characterized by irreversible dilated bronchi which harbor abundant pathogenic bacteria [20]. Patients may experience recurrent exacerbations which frequently require antibiotics and readily acquire antibiotic-resistant bacteria [14, 15]. The clinical and biological characteristics of the human host, including the disease severity and airway inflammatory subtypes could shape the microbiome and ARGs profiles in bronchiectasis. However, no study has directly linked ARGs expression profiles to the clinical heterogeneity of bronchiectasis.

By performing an in-depth metagenomic shotgun sequencing, we profiled ARGs in adults with bronchiectasis via linking sputum microbiota to ARGs, and explored their clinical correlates. We hypothesized that the ARG profiles were influenced by the sputum microbiota taxa and the clinical metrics of bronchiectasis.

Methods

Study populations

We prospectively enrolled participants aged 18 years or greater between May 2019 and January 2021 from out-patient clinics of The First Affiliated Hospital of Guangzhou Medical University, a major tertiary hospital specialized in and a national clinical research center for respiratory medicine. All eligible bronchiectasis patients had to be symptomatic (chronic cough, daily sputum production, etc.), remained stable for 4 weeks, and had undergone high-resolution computed tomography within 12 months. We excluded patients with an exacerbation [21] or antibiotic use (except for maintenance low-dose macrolides) within 4 weeks. Exacerbation denoted a significant deterioration of three or more symptoms lasting for ≥ 48 h (cough; sputum volume and/or consistency; sputum purulence; breathlessness and/or exercise tolerance; fatigue and/or malaise; hemoptysis) that required immediate changes in treatment.

Through advertisement, we recruited healthy subjects who had normal chest X-ray and spirometry findings, no lower respiratory tract infections within 4 weeks or uncontrolled systemic diseases.

Ethics approval was obtained from the ethics committee of The First Affiliated Hospital of Guangzhou Medical University (Medical Ethics 2016, the 32th). All participants gave written informed consent.

Study design

We collected information pertaining to prior antibiotic exposures, co-existing respiratory diseases and concomitant medications from medical charts. We evaluated radiologic severity with modified Reiff score, performed spirometry and sputum cell differentials, and rated the bronchiectasis severity index (BSI) [22]. Co-existing asthma and COPD was diagnosed based on Global Initiatives for Asthma [23] and Global Initiatives for Obstructive Lung Disease [24], respectively. The BSI was stratified into mild (0–4), moderate (5–8) and severe (9 or higher) categories. Some patients who had experienced exacerbations donated sputum at follow-up visits (exacerbation cohort). Clinical assessments for healthy subjects consisted of clinical history and smoking status inquiry, spirometry, chest X-ray and sputum induction.

Sputum sampling

We collected spontaneous sputum among all bronchiectasis patients, and induced sputum (3% hypertonic saline) in healthy subjects. Participants forcefully coughed up sputum after gargling mouth with distilled water. We sampled the most purulent portion for quality control. We evaluated the presence/absence of eosinophilia or neutrophilia [25] and stored freezing in − 80 degree freezers for metagenomic sequencing. See further details in Additional file 14.

Metagenomic sequencing

Bacterial genomic DNA was extracted using Qiagen DNA Mini kit, and homogenized using sterile zirconium/silica beads. Sequencing libraries were generated using NEB Next® Ultra™ DNA Library Prep Kit for Illumina® (New England Biolabs, MA, USA), followed by ultra-deep sequencing using Illumina NovaSeq 6000 (targeted >  = 40G sequences per sample, 2 × 150 bp paired-end sequencing). Two DNA extraction blank controls were sequenced.

Raw reads were processed using the Sunbeam pipeline [26], in which the adaptor sequences were removed by using Cutadapt (v2.5) [6], and quality filtered using Trimmomatic (v0.36) (setting: ILLUMINACLIP: NexteraPE-PE.fa:2:30:10:8:true LEADING:3TRAILING:3 SLIDINGWINDOW:4:15 MINLEN:36) [8]. Low-complexity sequences were filtered using Komplexity [26]. FastQC (v0.11.8) was used for quality check. Next, BWA (v0.7.17) was used to align quality-filtered reads to human reference genome (hg38) [7]. The sequencing depth was 40 G/sample. The average and median non-human read count across all samples was 16,792,475 and 7,852,636, respectively (range: 1,616,520–173,251,576). Taxonomic classification of non-human reads was performed using Kraken2 [9]. Raw data have been deposited in the Genome Sequence Archive in Beijing Institute of Genomics Data Center, under BioProject accession PRJCA013397 (https://ngdc.cncb.ac.cn/gsa-human/browse/HRA003749, No.: HRA003749).

Quantification of ARGs

Non-human reads were matched against the Structured Antibiotic Resistance Gene database (SARG) by using BLASTX [27], as described previously [28,29,30]. The reads which met BLASTX criteria (alignment length 75 bp, similarity 80%, e-value 10−7) were classified according to the SARG hierarchy. ARG abundance was calculated as reads per kilobase of reference sequence per million sample reads (RPKM) [31].

Statistical analysis

No sample size estimation was performed. Unsupervised cluster analysis was performed using the average linkage method of the hclust function in R stat package. The beta-diversity of microbiome and ARG profiles was presented using Principal Co-ordinate Analysis (PCoA), and ANalysis Of SiMilarities was used to test the statistical significance between groups. Permutational multivariate analysis of variance using distance matrices analysis (Adonis) was performed using R vegan package. To compare ARG abundance for paired stable-exacerbation samples, paired Wilcoxon rank-sum test was performed. Redundancy analysis was conducted to visualize the impact of clinical characteristics on ARG profiles. Spearman’s correlation was performed for continuous variables and Wilcoxon signed-rank test for categorical variables. Shannon index was calculated to evaluate the alpha diversity of ARGs. Procrustes analysis was performed to determine the agreement between resistome and microbiota compositions with Procrustes function in R vegan package.

For each variable, we performed FDR adjustment for P-values. The FDR-corrected P < 0.05 was deemed statistically significant. Analysis was performed using R (version 4.1.0), with the ggplot2 package for visualization. Heat maps and Venn diagram were produced using R with Pheatmap and VennDiagram package, respectively.

Results

Baseline characteristics

We enrolled 101 study participants, including 82 bronchiectasis patients and 19 healthy controls. Seventeen patients had paired stable and exacerbation samples, resulting in 118 sputum samples being analyzed. Baseline characteristics are shown in Table 1. 82 Bronchiectasis patients consisted of mostly never-smoked middle-aged adults with predominantly mild-to-moderate bronchiectasis (median BSI: 6.0). 15.8% had co-existing asthma while 7.1% had sputum eosinophilia (eosinophils ≥ 3%). 61.0% and 13.4% had blood eosinophil count greater than 100 and 300 cells per microliters, respectively. Overall, 20% had exposures to different classes of antibiotics within the previous 6 months, with a slightly higher frequency of beta-lactams (22.0% vs. 35.3%) and fluoroquinolones (23.2% vs. 41.2%) in the exacerbation cohort. The clinical characteristics of the exacerbation cohort were comparable with the whole bronchiectasis cohort, with nominally lower lung function, greater disease severity and more frequent antibiotic use in the exacerbation cohort. Furthermore, there were no remarkable differences in the age, gender and smoking between bronchiectasis patients and healthy controls (all P > 0.05).

Table 1 Baseline characteristics of bronchiectasis patients and healthy controls
Full size table

Overview of airway resistome and correlation with disease severity metrics

Comparison of the DNA extraction blank controls and sputum revealed marked differences in the metagenomic compositions that precluded the contamination by microbiota from the sequencing reagents or procedures (Additional file 1: Fig. S1). Metagenomic sequencing identified 272 bacterial species-level taxa with a relative abundance above 10–4 and 307 ARGs in all 118 sputum samples. Both the ARGs and the microbial compositions differed considerably between bronchiectasis patients and healthy controls. Furthermore, unsupervised clustering revealed three subgroups of microbiome (Fig. 1a): Pseudomonas-predominant subgroup (11 stable and 5 exacerbation samples, n = 16; Pseudomonas relative abundance: 73.0 ± 20.6%), Haemophilus-predominant subgroup (42 stable and 6 exacerbation samples, n = 48; Haemophilus relative abundance: 70.8 ± 15.3%); a balanced microbiome composition subgroup (29 stable and 6 exacerbation samples, plus samples from 19 healthy individuals, n = 54; no single dominant microbe, Additional file 8: Table S1). Distinct resistome profiles were observed among the three groups (Fig. 1a). ARGs of multi-drug resistance were dominant in the Pseudomonas-predominant subgroup, while ARGs of beta-lactam resistance were most highly abundant in Haemophilus-predominant subgroup (Additional file 2: Fig. S2). Adonis analysis suggests that the microbiome-based clusters were most strongly correlated with the ARG profile. Other than the microbiome clusters, gender, blood monocytes, quinolone use within the prior 6 months (analyzed as an independent variable instead of confounding variable), long-acting muscarinic antagonist use, and BSI were significantly associated with the ARG profile (Adonis P < 0.05, Fig. 1b, Additional file 2: Table S2). An overall balanced resistome profile was observed in the balanced microbiota subgroup. Pseudomonas-predominant subgroup had the highest ARG diversity and total abundance, while Haemophilus–predominant subgroup and balanced microbiota subgroup were lowest in ARG diversity and total abundance, respectively (Fig. 1c, Additional file 3: Fig. S3). A robust and significant correlation was observed between the microbiota and ARG profiles, as revealed by Procrustes analysis (Fig. 1d, Additional file 4: Fig. S4).

Fig. 1
figure 1

Profiles of antibiotic resistance genes (ARGs) in bronchiectasis patients and healthy controls. A A heat map demonstrating the bacterial species-level taxa and ARGs in 118 sputum samples, with unsupervised clustering revealing the Pseudomonas predominant, Haemophilus predominant, and a balanced microbiome composition subgroup. The upper panel demonstrates the bacterial compositions whereas the lower panel displays the ARG categories. B Association between clinical characteristics and the ARG profile. C The box and dot plots revealing the ARG diversity and total abundance among the Pseudomonas-predominant subgroup (Pseudomonas relative abundance: 73.0% ± 20.6%), Haemophilus–predominant subgroup (Haemophilus relative abundance: 70.8% ± 15.3%) and balanced microbiota subgroup (no single dominant microbe). D Correlation between the sputum microbiota and ARG composition. ARG antibiotic resistance gene, LAMA long-acting muscarinic antagonist, BSI bronchiectasis severity index, mMRC modified Medical Research Council dyspnea scale, CT computed tomography, FEV1 forced expiratory volume in one second, RPKM reads per kilobase of gene per million. All P values for the statistical analyses have been corrected with the false discovery rate

Full size image

The microbiome clusters was overall comparable between the paired stable and exacerbation samples, with a transition from the Pseudomonas-predominant to the Haemophilus-predominant subtype in three patients (Fig. 2a). Notably, there was a reduction in the abundance of Pseudomonas aeruginosa at exacerbation onset compared with the stable state. The reduction of Pseudomonas aeruginosa relative abundance at exacerbation onset was non-significant (FDR = 0.265, paired Wilcoxon test). In addition, the decreased representation of Pseudomonas-predominant subgroup at exacerbation onset as compared with that of the stable state was non-significant (P = 0.599, Fisher’s exact test). A greater representation of ARGs of beta-lactam resistance over multi-drug resistance was observed during exacerbations (Additional file 5: Fig. S5). The microbial subtypes based on metagenomic sequencing highly concurred with bacterial culture results (Fig. 2b).

Fig. 2
figure 2

Microbial compositions of bronchiectasis patients when clinically stable and at exacerbation onset. A Transition patterns of the microbial subtypes between the paired stable and exacerbation samples. B Agreement between the microbial subtypes and the bacterial culture results. The 9 sputum samples that were listed as ‘others’ in culture results are: Moraxella catarrhalis (n = 2), Klebsiella pneumoniae (n = 4), Pseudomonas fluorescens (n = 1), Staphylococcus aureus (n = 1), Streptococcus pneumoniae (n = 1)

Full size image

We have compared the samples collected prior to and after the COVID-19 outbreak (cut-off: January 2020) with PCoA and ANOSIM, revealing no major differences in the microbial compositions and ARG profiles (Additional file 6: Fig. S6).

Differential association between resistome and clinical characteristics in Haemophilus- and Pseudomonas-predominant subtypes

Within samples from stable state, we identified differential ARGs in Haemophilus and Pseudomonas-predominant subtypes, compared with the balanced microbiota subtype. When clinically stable, the abundance of 12 ARGs were significantly enriched and 16 ARGs were depleted in the Haemophilus-predominant subtype as compared with the balanced microbiota subtypes (FDR P < 0.05, Fig. 3a, Additional file 10: Table S3). PBP-1A (beta-lactam), ksgA (kasugamycin) and emrB (multidrug) were most significantly enriched in Haemophilus-predominant subtype (Fig. 3b). BSI was most strongly associated with the Haemophilus-predominant ARG profile (Additional file 9: Table S2), followed by quinolone use in the previous 6 months and FEV1% predicted (Fig. 3c, Adonis P < 0.05). Most ARGs enriched in Haemophilus-predominant subtype yielded positive correlations with the BSI, fluoroquinolone use, and modified Reiff score, while they were mostly negatively correlated with the body-mass index and FEV1% predicted (Fig. 3d, Additional file 11: Table S4). Specifically, 10 ARGs significantly and positively correlated with BSI, including PBP-1A that was most significantly enriched in Haemophilus-predominant subtype (Fig. 3d).

Fig. 3
figure 3

The distribution of antibiotic resistance genes in bronchiectasis patients with Haemophilus-predominant microbial profiles when clinically stable. A Abundance profiles of the enriched (log2 fold-change > 0) or depleted ARGs (log2 fold-change < 0) in Haemophilus-predominant versus balanced microbiota subtypes. Only ARGs with relative abundance greater than 0.005 are shown for display purpose; B A heat map demonstrating the profiles of differential ARGs between stable samples within Haemophilus-predominant and the balanced microbiota subgroups. Only ARGs with relative abundance greater than 0.005, with absolute fold-change greater than 2.0, and with FDR P < 0.05 are shown; C Association between the core clinical parameters and the Haemophilus-predominant ARG profile; D A heat matrix displaying the strength of association between the ARGs and the core clinical metrics. Only ARGs with relative abundance greater than 0.001 are shown for display purpose. ARG antibiotic resistance gene, ABO asthma-bronchiectasis overlap, BMI body-mass index, BSI bronchiectasis severity index, mMRC modified Medical Research Council dyspnea scale, CT computed tomography, FEV1 forced expiratory volume in one second. All P values for the statistical analyses have been corrected with the false discovery rate

Full size image

Thirty-four and 27 ARGs were significantly enriched or depleted in Pseudomonas-predominant versus the balanced microbiota subtype (FDR P < 0.05, Fig. 4a, Additional file 10: Table S3), with most enriched ARGs bearing multidrug resistance (Fig. 4a, b). Blood neutrophil and lymphocyte percentages were most strongly associated with the Pseudomonas-predominant ARG profile (Fig. 4c, Additional file 9: Table S2). Specifically, they exhibited significant inverse correlations with multiple ARGs of multi-drug resistance, including emrB, mexXY, and a multi-drug transporter (Fig. 4d, Additional file 11: Table S4). Therefore, there existed differential sets of clinical parameters which were associated with the resistome in Haemophilus- and Pseudomonas-predominant subtypes in stable bronchiectasis, respectively.

Fig. 4
figure 4

The distribution of antibiotic resistance genes in bronchiectasis patients with Pseudomonas-predominant microbial profiles when clinically stable. A Abundance profiles of the enriched (log2 fold-change > 0) or depleted ARGs (log2 fold-change < 0) in Pseudomonas-predominant versus balanced microbiota subtypes. Only ARGs with relative abundance greater than 0.005 are shown for display purpose; B A heat map demonstrating the profiles of differential ARGs between stable samples within Pseudomonas-predominant and the balanced microbiota subgroups. Only ARGs with relative abundance greater than 0.005, with absolute fold-change greater than 2.0, and with FDR P < 0.05 are shown; C Association between the core clinical parameters and the Pseudomonas-predominant ARG profile; D A heat matrix displaying the strength of association between the ARGs and the core clinical metrics. Only ARGs with relative abundance greater than 0.001 are shown for display purpose. ARG antibiotic resistance gene, ABO asthma-bronchiectasis overlap, BMI body-mass index, BSI bronchiectasis severity index, mMRC modified Medical Research Council dyspnea scale, CT computed tomography, FEV1 forced expiratory volume in one second All P values for the statistical analyses have been corrected with the false discovery rate

Full size image

Validation of associations between resistome and clinical parameters

Finally, we analyzed data in an independent cohort (Singapore, United Kingdom, Italy, Malaysia) [11], with the same analytical procedure and pipeline. Unsupervised clustering similarly revealed three microbiome clusters, with an over-representation of the balanced microbiota subgroup compared with our cohort (the Guangzhou cohort, Additional file 7: Fig. S7). The number of Klebsiella-predominant was 2 in our dataset (Fig. 1) and 4 in the validation dataset (Additional file 7: Fig. S7). Due to the small sample size, it was difficult to statistically assess the clinical characteristics for this particular subgroup of patients. Of the 61 and 28 ARGs associated with Pseudomonas-predominant and Haemophilus-predominant subgroups in the Guangzhou cohort respectively, 52 and 20 ARGs exhibited the same directionality of changes in the validation dataset, with 40 and 6 ARGs being statistically significant (FDR P < 0.05). In the validation dataset, the microbiome cluster, macrolides and ICS use were associated with the ARG profile (FDR P < 0.10, Additional file 12: Table S5). Macrolides use and age were associated with the ARG profile in Pseudomonas-predominant subgroup, while geography, macrolides use, gender and BSI were associated with the ARG profile in Haemophilus-predominant subgroup (FDR P < 0.25, Additional file 12: Table S5). There were 105, 23 and 71 correlations between ARG and the clinical features that were shared between the two datasets in all samples, Pseudomonas-predominant, and Haemophilus-predominant subgroups, respectively (FDR P < 0.10). Of these figures, 72, 18 and 45 correlations in all samples, Pseudomonas-predominant, and Haemophilus-predominant subgroups exhibited the same directionality in the validation dataset (Additional file 13: Table S6).

Discussion

We performed metagenomic shotgun sequencing on spontaneous sputum from bronchiectasis patients, revealing distinct profiles of ARGs associated with three microbial subgroups. The ARG profile was associated with multiple clinical metrics. The PBP-1A, ksgA and emrB genes were enriched in Haemophilus-predominant subgroup, whereas the emrB, mexXY genes were depleted in Pseudomonas-predominant subgroup. These findings could be partly validated (~ 68.6% of the ARG-clinical correlations) in an independent international cohort [11, 32, 33], indicating the generalizability to other study populations.

Concerns have been raised regarding the accumulation of ARGs in bronchiectasis. By comparison with healthy controls, Mac Aogain and colleagues documented the prevalent ARGs associated with fluoroquinolone, beta-lactam and tetracycline which were independent of prior antibiotic exposures and the clinical status of CRDs [11]. Congruent with these observations, our study has further identified the ARGs of beta-lactam resistance in the Haemophilus-predominant subgroup, multi-drug resistance in the Pseudomonas-predominant subgroup. The two studies, however, differed in patient populations (three CRDs vs. Bronchiectasis) and study objective (profiling resistome/microbiome and the inhaler microenvironment vs. profiling resistome according to microbiome and clinical characteristics). Our findings pertaining to ARGs associated with Pseudomonas spp. were clinically relevant. Sputum microbiota with predominance of Pseudomonas spp. has been associated with a low microbial diversity and greater severity of bronchiectasis [34]. Other studies [7, 8, 35] have reported the predominance of Pseudomonas associated with multi-drug resistance in cystic fibrosis (CF). Other cellular mechanisms (e.g., active reflux, porin alterations) might have also contributed to the acquired resistance to antibiotics for Pseudomonas spp. [36]. Furthermore, the ampC (encoding beta-lactamase) and ftsl gene (encoding penicillin-binding protein 3) mutations were frequently detected in P. aeruginosa [27], and both fusA1 and pmrB mutations conferred polymyxin resistance in CF [9].

The Haemophilus-dominant taxa was common among bronchiectasis patients [1, 37]. As per international guidelines [38], beta-lactams are first-line antibiotics for bronchiectasis patients isolated with H. influenzae at exacerbation onset. The ARGs including blaTEM-1D (ampicillin resistance) and tetB genes (tetracycline resistance) have also been detected, despite the lower prevalence (6.1%) in patients with other airway diseases [39]. Apart from beta-lactam resistance, a rarer but high-level ciprofloxacin-monoresistant H. influenzae (mainly involving gyrA, parC and parE mutations) isolate has been identified [40], indicating the mutational capacity of ARGs in H. influenzae. Consistent with findings from COPD [14], ARGs were largely determined by the microbial compositions. The balanced subgroup was characterized by the intermediate ARG profile, which has also been shaped by other microbiota taxa. Antibiotic resistance associated with oral commensals such as Prevotella spp., which was generally resistant to amoxicillin (96%) and occasionally resistant to multiple drugs (10%), has been detected despite the less prominent (24%) resistance to moxifloxacin [41].

The ARG profiles in bronchiectasis did not differ significantly when comparing clinically stable samples and those at exacerbation onset, which mirrored the findings from COPD [14]. The microbial compositions during clinical stability and exacerbation mostly remained stable. In COPD, antibiotics therapy targeting at ameliorating exacerbations did not enrich any particular microbial taxa or ARG [14]. Hence, the ARG profiles may be independent of the clinical status.

We identified reproducible correlations between certain ARG profiles and the clinical metrics of bronchiectasis. The cardinal factors shaping ARGs consisted of sex, blood monocytes, quinolone or LAMA use, and the BSI. Blood neutrophil and lymphocyte percentages correlated strongly with the Pseudomonas-dominant ARG profile. The association between blood inflammatory cells and ARG profiles in bronchiectasis merits further comments. A cross-sectional study reported a marked difference in ARG profiles between neutrophil-dominant COPD (dominated by multidrug resistance genes related to Ralstonia and Pseudomonas) and eosinophil-dominant COPD (macrolide-high resistome predominantly associated with Campylobacter and Aggregatibacter) [42]. Concurrent profiling of the metabolic pathways and inflammatory biomarkers will help to shed mechanistic insights into our observations. The precise mechanisms underlying the association between LAMA use and ARG profiles are not entirely clear, whether this relates to the anti-inflamamtory actions of LAMA that could shape the microbial compositions merits further investigation.

Our study has profiled the ARGs abundance, in a well characterized cohort of bronchiectasis patients, by linking ARG profiles to disease severity, inflammatory subtypes and prior antibiotic exposures. Metagenomic sequencing has minimized the risk of biased analysis (e.g., polymerase chain reactions). Profiling of the dominant microbiota taxa and key clinical metrics (e.g. inflammatory subtypes, disease severity) will enable a better understanding of the factors associated with microbial resistance, for instance, revealing the patient subgroups most likely harboring extensive ARGs.

However, our findings was limited by the inability to evaluate dynamic changes (e.g. repeated assessments when clinically stable) within an individual. Direct comparison of spontaneous sputum in bronchiectasis patients vs. induced sputum in healthy controls has been problematic; however, obtaining bronchoalveolar lavage would be constrained by the invasiveness and cost, reducing the feasibility with much larger sample sizes. Despite the substantial difference in microbiota compositions detected with 16 s rRNA sequencing between induced sputum and spotaneous sputum in patients with COPD [43], data regarding the comparison with metagenomic sequencing are lacking. Although not directly comparable due to the notable differences in specimen processing and sequencing techniques and demographic characteristics, our principal findings could be validated in a multinational cohort, which has partly circumvented the single-country study design. The overal cohort is relatively mild with a mild-moderate BSI score and a low median exacerbation frequency the previous year, and hence extrapolation of our conclusion should be taken cautiously. We employed a previously validated read-based approach to profile ARGs, however, future validation with assembly-based algorithms would be needed. We have standardized the relative abundance of ARGs based on the sequencing throughput which allowed for direct comparisons among different samples, however, the batch effects related to nucleic acid extract and sequencing cannot be fully addressed. There was no control for Pseudomonas ARGs for healthy controls; however, the lack of Pseudomonas- or Haemophilus-dominant ARG subtype precluded our further analysis. This might be circumvented by the comparison with ARGs in other CRDs. We noted the distinct resistome in different microbial-dominant subgroups, albeit the sample size for the Pseudomonas-dominant subgroup was insufficient to fully confirm our findings. Finally, although ARGs were associated with clinical severity and macrolide use, the prognostic implications are not entirely clear. Integration of ARG profiling with other measures (e.g. metabolome) may enable further clues as to how ARGs interact with host immune response particularly during exacerbation dynamics.

In conclusion, we have unraveled the differentially abundant ARG profiles enriched or depleted in bronchiectasis patients with distinct predominant microbial species. Future studies should enable the scientific community to understand longitudinal evolution/acquisition, and whether this is related to environmental niche, the effect from antibiotics and the impact of broad spectrum antibiotics in the context of exacerbation.