HAEMOPHILUS, ANTIBIOTIC THERAPY AND THE AIRWAY MICROBIOME IN CHRONIC OBSTRUCTIVE PULMONARY DISEASE

Chronic obstructive pulmonary disease (COPD) is a smoking-related illness affecting 64 million people worldwide. Airway infection drives recurrent exacerbations and lung function decline. Prophylactic antibiotics may prevent exacerbations but their use is a significant cause of population antimicrobial resistance. We characterised the sputum microbiome by 16S rRNA gene analysis using 138 samples collected during a randomised controlled trial of prophylactic antibiotics in 71 patients with stable COPD. On comparing the profile of the microbiome obtained by sequencing to the isolates grown from samples using standard culture, there were similarities overall, although with a much narrower spectrum of genera on culture with under-representation of certain genera including Veillonella and Prevotella . There was concordance in the most abundant genera within samples and the number of isolates cultured reflected the measured bacterial diversity.


Introduction
COPD is a debilitating chronic lung disease characterized by episodes of symptomatic worsening associated with disease progression, healthcare utilisation and mortality (1)(2)(3).
COPD is associated with structural and immune defects (4) and altered microbial communities within the airways.The presence of pathogenic airway bacteria is associated with exacerbation frequency and disease progression (5)(6).Studies of the COPD microbiome with sequencing technology have so far used invasive sampling from small numbers of severely diseased patients (7,8).There is an unmet need for larger studies relevant to the wider range of patients with COPD in the general population.
Long-term antibiotic therapy is one therapeutic option to prevent exacerbations (9).Longterm macrolide antibiotics may reduce the frequency (10, 11) and duration (11) of exacerbations and fluoroquinolones may be effective in selected patients (12).Macrolides also have potent anti-inflammatory effects (13), and there are no data to show their action on the microbiota in COPD.The airway bacterial load does not change with prophylactic antibiotic therapy (14), although there are significant increases in antimicrobial resistance in treated patients (10,14) and the general population (15).
In order to understand better the effects of prophylactic antibiotic therapy and to streamline its use, we have investigated in detail the composition of the airway microbiome in COPD, using spontaneously expectorated sputum samples from a randomised controlled trial of prophylactic antibiotic therapy in patients with stable disease (14).In keeping with clinical practice, we have studied sputum samples as a surrogate for events in the lower .CC-BY 4.0 International license under a not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.It is made available The copyright holder for this preprint (which was this version posted September 17, 2018.; https://doi.org/10.1101/419127doi: bioRxiv preprint 6 airways.We have also carried out a direct comparison of the microbiome profile obtained using sequencing to that obtained using traditional bacterial culture, in order to help delineate the relationship of these newer techniques to those that are currently used to guide clinical practice.Furthermore, six isolates were available from stored samples from siz different patients, allowing us to confirm the species, antibiotic resistance gene profile and potential virulence by whole genome sequencing. . CC-BY 4.0 International license under a not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.It is made available The copyright holder for this preprint (which was this version posted September 17, 2018.; https://doi.org/10.1101/419127doi: bioRxiv preprint

RESULTS
The 71 patients who were sampled before and after treatment had moderately severe COPD, with a mean (SD) age of 69 (9) years, 52 (72%) male, mean (SD) forced expiratory volume in 1 second (FEV 1 ) of 50 ( 14) % predicted and moderate airway obstruction (Table 1).We carried out quantitative polymerase chain reaction (qPCR) of the V4 region of the 16S rRNA and used identical primers to sequence the same region on the Illumina MiSeq platform.After quality control measures 138 of the original 142 samples were included for analysis, comprising a total of 9,113,146 sequence reads from 2259 distinct OTUs and 114 genera.The median sequencing depth was 61088 reads per sample (range 4,241 -248,309), and samples were rarefied to a minimum sequencing depth of 4,241 reads unless otherwise specified.
Haemophilus dominance of the microbiome defines a distinct cluster of patients with higher airway inflammation and lower airway bacterial diversity We examined the profile of the microbiome in these patients at baseline.Streptococcus was the most abundant genus in sputum samples taken before antibiotic therapy, accounting for 42% of all sequence reads, followed by Haemophilus (19%).Sequences from the genera Veillonella, Actinomyces, Granulicatella and Prevotella comprised 17% of sequence reads (Supplementary Table 1 and Supplementary Fig. 4).
We carried out an unsupervised hierarchical clustering analysis of OTU frequencies which identified clear subsets of patients, the largest of which (n=42) had Streptococcus as the proportionally most abundant genus by number of sequence reads (Fig. 1a).We observed a .CC-BY 4.0 International license under a not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.It is made available The copyright holder for this preprint (which was this version posted September 17, 2018.; https://doi.org/10.1101/419127doi: bioRxiv preprint 8 cluster of 17 (24%) patients in whom sequence reads from Haemophilus OTUs were most abundant.Two samples, also related to this cluster, were dominated by Moraxella.A single sample dominated by Pseudomonas clustered separately.
The mean (SD) bacterial load by 16S rRNA gene qPCR was 9.2 (0.8) log 10 copies/gram of sputum.Sputum bacterial load was correlated with the inflammatory cytokine IL-1β (r=0.311,p = 0.009) but not FEV 1 , health status, exacerbation frequency, other cytokines, or inhaled corticosteroid use.
Correspondingly, there were strong correlations between the proportion of sequencing reads comprised by Haemophilus (proportional abundance) and the three inflammatory markers (IL-1b, r=0.56, p<0.001;IL-8, r=0.52, p < 0.001; IL-6, r=0.28, p = 0.02).We saw no .CC-BY 4.0 International license under a not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.It is made available The copyright holder for this preprint (which was this version posted September 17, 2018.; https://doi.org/10.1101/419127doi: bioRxiv preprint significant differences in demographics, FEV 1 , previous exacerbation frequency or health status between groups (Supplementary Table 2).
The predominant Haemophilus OTU within the COPD microbiome is H. influenzae Of the 306 different Haemophilus OTUs present at baseline (1,208,824 sequence reads), 84% were from a single OTU (Haemophilus_1393, Fig. 2a).This was also the most abundant OTU in each of the Haemophilus dominated samples.Sequence alignment and phylogenetic analysis confirmed that this OTU clustered strongly within the branch containing H. influenzae and H. haemolyticus (Fig. 2b).
Oligotyping is a supervised computational technique that allows the detection of meaningful sub-populations within closely related sequences, for example those assigned to a single OTU, and offers improved resolution over standard clustering techniques.Oligotyping of the 1,019,560 sequence reads within Haemophilus_1393 identified two base-pair positions of high entropy within the sequence alignment (Fig. 2c).Three distinct oligotypes, defined by the base differences C-G, T-A and T-G at these two positions, comprised 99% of the sequence reads (Fig. 2d).Notably, in approximately one quarter of patients there was more than one of these oligotypes present within the OTU.These are likely to represent different strains of the species.
In order to provide further confirmation of the species of this OTU, we examined the corresponding base-pair positions from the 16S rRNA gene sequences of 42 high-quality H. influenzae reference strains, finding C-G was present at these positions in 27 (64%), T-A in 13 (31%) and T-G in 2 (5%).Comparison of these positions to 96 other representative Haemophilus species confirmed that, while the C-G and T-G combinations were seen in .CC-BY 4.0 International license under a not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.It is made available The copyright holder for this preprint (which was this version posted September 17, 2018.; https://doi.org/10.1101/419127doi: bioRxiv preprint other Haemophilus species, T-A was only present in reference strains of H. influenzae.Furthermore, H. influenzae was cultured from 9/17 (53%) of the samples where Haemophilus was the most abundant genus (compared to 3/54 [6%] where it was not) and was the only Haemophilus isolate from any of these samples.We were therefore confident that this OTU represents H. influenzae, comprises most of the Haemophilus present in the microbiome of our COPD patients, and that different strains of Haemophilus may co-exist within the microbiome.

Whole genome sequencing of six Haemophilus influenzae isolates
Six Haemophilus influenzae isolates were archived during the study, coming from patients [ADD WHICH PATIENTS AND WHAT TREATMENT ARM THEY WERE IN].These were whole genome sequenced, assembled and annotated.They were all identified as non-typeable Haemophilus influenzae owing to absence of both capsular regions I and II.Antibiotic resistance genes were annotated using the CARD database [REF] revealing that all six strains had at least one macrolide resistant gene, with a variable profile of additional resistances (figure 2e).The NTHI genomes were searched for a list 40 known virulence determinants [REF] revealing an identical set of 19 virulence genes to be present in all six isolates (Supplementary tables 1 and 2).This might suggest selection of particular NTHI strains adapted to the COPD lung.Two further analyses were therefore carried out to determine whether this was the case, a phylogenetic tree based on 40 universally conserved singlecopy marker proteins and a pan-genome analysis of core (present in 90% of isolates in both these isolates and the Chiara et al 2014 genomes) and accessory genes.The six genomes did not cluster together by either method.A tanglegram comparison of a phylogeny built from a concatenation of the 19 virulence genes with the marker protein tree demonstrated .CC-BY 4.0 International license under a not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.It is made available The copyright holder for this preprint (which was this version posted September 17, 2018.; https://doi.org/10.1101/419127doi: bioRxiv preprint agreement, indicating that selection does not appear to working at the strain level, though may still be occurring at the level of presence or absence of genes given the identical set of 19 virulence genes across the isolates (supplementary figure XX).
Fluoroquinolone administration is associated with a particular decrease in Haemophilus taxa, while Haemophilus-dominated patients show an improved inflammatory response to antibiotic therapy The patients were randomised to receive 13 weeks of treatment with azithromycin 250mg three times/week (18 patients), doxycycline 100mg daily (20 patients), moxifloxacin 400mg daily for 5 days every four weeks (18 patients) or 1 placebo capsule daily (14 patients).As previously reported (14), overall airway bacterial burden by 16S rRNA gene qPCR did not change significantly between the start and the end of treatment in any arm.
We examined whether treatment with antibiotics caused differences in overall community composition between treatment groups by performing permutational multivariate analysis of variance (PERMANOVA) at baseline and at the end of treatment.Prior to starting antibiotics we saw no significant effect of treatment allocation, but a treatment arm effect was present after the trial (R 2 = 0.11, p<0.001) (Supplementary Fig. 6).There were no significant changes in within-sample diversity by the Shannon diversity index and number of observed OTUs in any treatment arm modeled against placebo.
Azithromycin treatment resulted in a decrease from baseline in eight non-Haemophilus OTUs, whilst seven OTUs (two from the genus Haemophilus) increased (Fig. 3a and .CC-BY 4.0 International license under a not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.It is made available The copyright holder for this preprint (which was this version posted September 17, 2018.; https://doi.org/10.1101/419127doi: bioRxiv preprint 12 Supplementary Fig. 7).Treatment with doxycycline resulted in a decrease in 29 OTUs, four of which were from the genus Haemophilus; this included a decrease of 85% in H. influenzae (Fig. 3b and Supplementary Fig. 7).The most profound changes were seen in the moxifloxacin treatment arm where a total of 107 OTUs decreased significantly, 80 (75%) of which were from the genus Haemophilus, including a decrease of 99% in H.influenzae.There was a corresponding increase in 37 OTUs, 15 (41%) of which were from the genus Veillonella (Fig. 3c, Supplementary Fig. 7).In the placebo arm, the only change was in one Pseudomonas OTU that increased significantly in abundance (Fig. 3d).These OTU-level changes were also seen in the overall profile of the microbiome after treatment, where there was near-total eradication of Haemophilus OTUs from the microbiome after moxifloxacin therapy but not following azithromycin or doxycycline (Supplementary Fig. 8).

Discussion
We have shown that, in patients with stable COPD, dominance of the microbiome by the pathogen H. influenzae defines a group of patients who have more active disease with higher airway bacterial load and airway inflammation and lower bacterial diversity.
Importantly, these patients also exhibited an improved inflammatory response to antibiotic therapy.
This expands on previous data linking bacterial colonization to exacerbation frequency (5) and disease progression (6).Defining 'dominance' of the microbiome by the most abundant genus provides a means of stratifying complex data and this has been shown in non-cystic fibrosis bronchiectasis to be clinically meaningful (17).We found agreement between the dominant genus and the most abundant isolate on culture, indicating that these DNA sequences originate from viable bacteria and that this dominance is clinically important.
While studies published previously are too small to infer patterns individually, detailed profiles of the sputum microbiota of 92 clinically stable COPD patients have been reported across six studies (18)(19)(20)(21)(22)(23).Twenty-eight (30%) of these were dominated by the genus Haemophilus or its parent phylum Proteobacteria (Supplementary Table 4).This is entirely consistent with our results.
Expectorated sputum was used during this study and carries a risk of contamination by oral microflora.Previous studies that have attempted to avoid this by using invasive specimens .CC-BY 4.0 International license under a not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.It is made available The copyright holder for this preprint (which was this version posted September 17, 2018.; https://doi.org/10.1101/419127doi: bioRxiv preprint 14 obtained bronchoscopically (7) or from lung explants (8) are necessarily small and lacking statistical power.The ubiquitous use of sputum microbiology in clinical practice indicates that minor contamination does not prevent expectorated sputum from providing a valid means of assessing the thoracic airway microbiome.
The quarter of patients in whom H. influenzae dominated the microbiota exhibited much higher levels of inflammatory cytokines and bacterial load than other cases.This is consistent with previous reports (24) using species-specific techniques but has not previously been shown in the full context of the airway microbiome.Airway inflammation predicts exacerbation frequency (25) and lung function decline (26), and these patients are therefore at high risk.
Haemophilus abundance was also associated with a narrowed diversity of species within the airway, and when the burden of Haemophilus reduced following moxifloxacin treatment there was a corresponding increase in the abundance of other bacterial genera.That there was no corresponding increase in measured diversity within the microbiota of the moxifloxacin-treated patients may reflect that the expanding taxa were from genera already present at baseline.These dynamic changes suggest that Haemophilus is able to dominate the microbiome, leaving a vacant ecological niche when removed.The fluoroquinolone moxifloxacin exhibited particularly potent activity against Haemophilus species.While this is in line with previous reports (27), the near-total eradication of Haemophilus from the airway microbiome has not previously been demonstrated.Oligotyping was successfully used in order to improve the 16S rRNA gene sequencing identification of Haemophilus spp.OTUs as Haemophilus influenzae, despite short sequences being available.It also revealed co-.CC-BY 4.0 International license under a not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.It is made available The copyright holder for this preprint (which was this version posted September 17, 2018.; https://doi.org/10.1101/419127doi: bioRxiv preprint infection with multiple strains of H. influenzae, which may have implications for antibiotic therapy. The relatively minor effect of azithromycin on the airway microbiome is surprising, particularly as it has efficacy against Haemophilus (28).Macrolides may therefore prevent exacerbations by attenuating bacterial outgrowth following viral infection (29, 30), or via unmeasured effects on bacterial function.Alternatively, macrolides have a wide spectrum of immunomodulatory effects (13) that may underlie the reduction in exacerbation frequency.
The dose used here however is lower than that used in subsequently published trials (10) and this may have resulted in a reduced effect on airway bacterial communities.Although doxycycline also had a lesser effect than moxifloxacin it did result in the attenuation of several OTUs including H. influenzae.
Long-term antibiotic therapy is not without risk, particularly the induction of antimicrobial resistance (14,15).We identified at least one macrolide resistance gene in each of six isolates that were whole genome sequenced, though this does not necessarily translate to in vivo resistance.A better understanding of antibiotic effects on the airway microbiome will allow therapy to be tailored to those who will benefit most.Importantly, vaccination against non-typeable Haemophilus influenzae may spare antibiotic use in selected patients (16).
A restricted set of 19 virulence genes of the 40 screened for was found in each of the six isolates, all of which were from six different COPD patients.This suggests that there may be some selection of particular virulence strategies in NTHI in COPD, many of these genes are involved in adhesion.However, this is small opportunistic set of genomes and systematic .CC-BY 4.0 International license under a not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.It is made available The copyright holder for this preprint (which was this version posted September 17, 2018.; https://doi.org/10.1101/419127doi: bioRxiv preprint 16 sampling and sequencing of a much larger number of NTHI isolates from COPD patients is required to confirm or refute this.Molecular characterisation of whole sputum communities allowed us to define Haemophilus dominance in our patients in a manner that was not possible with classical culture.Our 16S analyses have given novel insights into the response of all the airway bacteria to antibiotics, and we suggest that community analyses should underpin future therapeutic trials of antimicrobial therapy in COPD. . CC-BY 4.0 International license under a not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.It is made available The copyright holder for this preprint (which was this version posted September 17, 2018.; https://doi.org/10.1101/419127doi: bioRxiv preprint this study has been uploaded to the European Nucleotide Archive, accession number PRJEB11966.

Statistical Analysis
Read numbers were rarefied (randomly subsampled) to a uniform sequencing depth equal to that of the minimum retained sample after quality control, 4,241 reads.Summary variables were expressed as mean ± standard deviation (SD) or median (interquartile range) and log-transformed as appropriate.Hierarchical clustering was performed using the unweighted pair group method with arithmetic mean and the Bray-Curtis measure of between-group (beta) diversity.The relative biomass of Haemophilus within a sample was estimated by multiplying the qPCR bacterial load by the proportion of the sequencing reads that were from the genus Haemophilus.Continuous data were compared using Pearson's correlation coefficient and significant variables used to fit a multiple regression model against IL-1β levels at baseline.Changes in Haemophilus load against placebo were assessed using a multivariate regression model adjusted for baseline Haemophilus load and baseline % predicted FEV 1 .Phylogenetic tree construction and sequence comparisons used ARB version 6.0.1 (www.arb-home.de)with the SILVA SSU reference database version 119.1 (www.arb-silva.de).Oligotyping (37) was carried out using version 2.0 of the pipeline (www.merenlab.org).
Differential OTU counts before and after treatment were modelled using the negative binomial distribution in the DESeq2 package (38) version 1.6.3 and the false discovery rate controlled using the Benjamini-Hochberg correction (39).Statistical analysis was performed .CC-BY 4.0 International license under a not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.It is made available The copyright holder for this preprint (which was this version posted September 17, 2018.; https://doi.org/10.1101/419127doi: bioRxiv preprint 20 using R statistics version 3.1.3(www.r-project.org) and the phyloseq package (40) version 1.10.0.

Genome sequencing of isolates
Details of this can be found in supplementary materials.Briefly, 6 available isolates of Haemophilus influenzae had DNA extracted and submitted for sequencing at the Wellcome Trust Sanger Institute, genomes were assembled and annotated for virulence genes, antibiotic resistance cassettes and phylogenetic trees constructed to compare these isolates to other non-typeable H. influenzae.
. CC-BY 4.0 International license under a not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.It is made available The copyright holder for this preprint (which was this version posted September 17, 2018.; https://doi.org/10.1101/419127doi: bioRxiv preprint Haemophilus was the most abundant genus had higher levels of 16S rRNA gene copies, lower bacterial diversity, and higher levels of all three cytokines in sputum than others.Between-group differences were assessed using one-way ANOVA, Chi-squared test, or Kruskal-Wallis test as appropriate.There were no significant between-group .CC-BY 4.0 International license under a not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.It is made available The copyright holder for this preprint (which was this version posted September 17, 2018.; https://doi.org/10.1101/419127doi: bioRxiv preprint 33 differences at the significance level of 0.05 except for baseline FEV 1 and FEV 1 % predicted (p = 0.033 and p = 0.009, respectively).
. CC-BY 4.0 International license under a not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.It is made available The copyright holder for this preprint (which was this version posted September 17, 2018.; https://doi.org/10.1101/419127doi: bioRxiv preprint a.

Figure 1 :
Figure 1: Profile of the COPD airway microbiome at baseline.

Figure 2 :
Figure 2: Characterizing Haemophilus species at baseline.a. Profile of individual Haemophilus operational taxonomic units (OTUs) in the microbiota of

Figure 3 :
Figure 3: Changes in community composition and inflammatory cytokines following 13

Figure 4 .
Figure 4. represents a phylogenetic tree based on concatenated alignment of 40 peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.It is made available peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.It is made availableThe copyright holder for this preprint (which was this version postedSeptember 17, 2018.