A pilot randomised controlled trial of oral doxycycline after endoscopic sinus surgery and its effects on the sinonasal microbiome*

Background: Oral antibiotics are commonly prescribed after endoscopic sinus surgery (ESS) despite minimal clinical data supporting this practice. We aim to assess the effect of post-ESS doxycycline on clinical outcomes and on the diversity and composition of the sinonasal microbiome. Methods: Samples from the middle meatus were collected from twelve patients undergoing ESS to treat chronic rhinosinusitis. Patients were double-blind randomised to receive either oral doxycycline or placebo in the post-operative period. Further samples were collected at two weeks and three months post-operatively. The sinonasal microbiome was characterized using 16S ribosomal RNA (rRNA) gene amplicon sequencing. SNOT-22 scores, Lund Mackay scores, and Modified Lund Mackay Endoscopic Scores (MLMES) were collected. Results: After ESS, bacterial diversity increased while SNOT-22 score decreased for both treatments. Microbiome composition diverged between treatments, and random forest analysis identified nine taxa that may distinguish treatment groups. There was no significant difference in SNOT-22 score, 3-month MLMES or bacterial diversity between the placebo and doxycycline groups. The trends for all of these measures favour placebo. Conclusion: In this pilot study, we detected no significant difference between placebo and antibiotic treatments in clinical outcome. As patient symptoms improved after ESS, we detected a concurrent increase in the diversity of the sinonasal microbiome. Our data highlight the need for and facilitate the design of future larger studies to explore the relationship between prophylactic antibiotics and post-ESS recovery.


Introduction
Our current understanding of the pathogenesis of chronic rhinosinusitis (CRS) and the optimal treatment for this disease is inadequate. Previous studies have suggested that the sinonasal microbiome plays a role in the development or maintenance of this disease (1) . The current research suggests bacterial dysbiosis (deviations from the typical microbial community) and a decrease in bacterial diversity are associated with this disease (2)(3)(4) .
There is currently level 1 evidence demonstrating the efficacy of macrolide and tetracycline antibiotics for the treatment of CRS (5,6) . However, it is noted that both of these classes of antibiotics have anti-inflammatory properties as well as antibacterial properties (7,8) . It has been suggested that antibiotic exposure may be harmful in the long term, increasing the risk of bacterial dysbiosis and reducing bacterial diversity (9) . Other questions relating to the microbiology of CRS remain unanswered, including what impact oral antibiotics have on the sinonasal microbiome (10) and what the optimal collection method for sinonasal samples is (11,12) .
There is increasing evidence demonstrating the efficacy of endoscopic sinus surgery (ESS) in CRS management (13) . Staphylococcus aureus was found to be associated with worse postoperative outcomes in patients undergoing ESS (14) , suggesting that the use of peri-operative anti-staphylococcal antibiotic may be beneficial. ESS, however, falls under the category of "cleancontaminated" surgery where there is generally considered to be no benefit in the use of post-operative antibiotics for the reduction in risk of surgical site infection (15) . A recent metaanalysis concluded that there is a lack of high-quality evidence to guide the decision around the use of antibiotics after ESS (16) . A further randomised trial that did not feature in the meta-analysis concluded that patients receiving placebo likely achieved equivalent results to patients provided with co-amoxiclav after ESS (17) . Significantly, however, in a 2015 study, 73.1% of surgeons reported routine use of antibiotics after ESS (18) . This is important given the frequency with which ESS is performed and the increasing need for antibiotic stewardship (19) .
Given the lack of certainty around the need for antibiotics after ESS, the potential beneficial role of anti-staphylococcal treatment, and the proven efficacy of doxycycline in the treatment of CRS (5,20) , this study aims to investigate the use of oral doxycycline after ESS. Particular focus is on the impact of doxycycline on overall clinical recovery from surgery and the impact doxycycline has on the sinonasal microbiome through the post-operative recovery period. This is intended as a pilot study to facilitate a larger, appropriately powered study allowing us to more definitively answer these questions. In so doing, we further describe the efficacy of ESS in treating CRS, the microflora of CRS both before and after ESS, and allow us to further investigate the optimal sampling method for sinonasal microbiome samples. Twelve patients fulfilling the EPOS criteria for CRS (1) undergoing ESS after failing medical treatment were prospectively recruited. Patients underwent a complete sphenoethmoidectomy and frontal recess dissection. Exclusion criteria included the following: prior ESS, underlying condition predisposing to CRS (e.g., vasculitis, cystic fibrosis, aspirin-exacerbated respiratory disease), unilateral CRS, Lund-Mackay (21) score (LMS) less than 10/24, any antibiotic usage in the 12 weeks prior to surgery, allergy to doxycycline, confirmed or possible pregnancy. On the day of surgery, demographic and clinical data was collected, including age, gender, presence or absence of nasal polyps, prior diagnosis of asthma, pre-operative Lund-Mackay score, SNOT-22 score, and completion of any adjunct procedures (e.g., septoplasty or inferior turbinoplasties) ( Table 1) Genomic DNA extraction and 16S rRNA gene PCR amplicons sequencing Total DNA was extracted from tissue and swab samples using assigned with the native implementation of the naive Bayesian classifier and a DADA2-formatted reference database for the SILVA v138 database (26) .

Patients and sample collection
The taxonomic data and ASV data were imported in R with phyloseq (v1.32.0) (27) . Contaminant taxa were identified with the decontam pipeline (v1.8.0) (28) Table 1). Bacterial community composition was transformed with a centred log-ratio transformation. Euclidean distances were calculated between samples and a Principal Coordinates Analysis (PCoA) was performed. After confirming that samples did not differ significantly in their dispersion using an analysis of multivariate homogeneity (PERMDISP), a PERMANOVA was performed to check for differences over time and between treatments, both performed in vegan (v2.5-6) (29) . A procrustes analysis, also performed in vegan, was used to correlate datasets obtained by tissue or swab sampling. A random forest analysis was conducted to model ASV potentially associated with each treatment group (random forest (v4.6-14) (30) ). Pearson correlation coefficients were employed to assess the correlation between the centred log-ratio transformed ASV abundance and time. ggplot2 (v3.3.3) (31) was used to visualise the data.

Bioinformatics and statistical analysis
Bioinformatic and statistical analyses are described in detail in the supplementary methods. Briefly, the raw FASTQ files were imported into R (v4.0.2) using 'DADA2' package (v.1.16.0) (25) for amplicon sequence variant (ASV) inference. Taxonomy was Table 1. Clinical and demographic data collected from patients across 3 months during this study. Patients 1 and 4 did not significantly improve their symptoms after recovering from surgery according to their pre-op and second post-op SNOT-22 scores. Low SNOT-22 scores (total score range = 0-110), LMS scores (total score range = 0-24), and MLMES scores (total score range = 0-100) indicated better surgical outcomes.  Figures 3 and 4).  Clinical and demographic data for included patients are shown in Table 1. Patient 11 was excluded from the dataset throughout this analysis since this patient received Augmentin as a treatment for another illness during this study (Table 2).

Clinical outcomes
Both Three patients reported side-effects from the medication ( Table   2), and they were all receiving doxycycline. No side-effects were reported for the placebo group. Reported side-effects correspond to known side-effects of the medication (32) .

Microbiome analysis
Diversity of the bacterial communities was calculated using the  Table   2), but a significantly higher diversity was observed between the pre-op and second post-op timepoints for the placebo treatment (p-value = 0.045). There was no significant difference between the remaining timepoint comparisons for the placebo treatment (Supplementary Table 3).
The bacterial community composition was analysed using a PCoA based on Euclidean distances, which is an acceptable proxy for identifying differences/similarities in the community The effect of the specimen collection method was investigated by comparing tissue and swab samples collected during

Discussion
Several studies have investigated whether post-operative antibiotics improve outcomes in CRS patients undergoing ESS but the answer remains unclear (17,(33)(34)(35) . In this pilot study, we has shown that a more diverse microbiome is associated with improved health (36,37) .

Clinical outcomes
Our study showed a non-significant trend favouring placebo with less reduction in SNOT-22 scores (p-value = 0.052) and a higher 3-month endoscopic score in the doxycycline group ( Figure 1). Acknowledging the small sample size, our data do raise the possibility that doxycycline may result in a worse outcome than placebo. This is especially relevant when taking into account the sinonasal microbiome data. There was also a noted difference between treatment groups in regards to side-effects suffered by patients. The difference suggested by this study is concordant with the potentially deleterious impact of antibiotics on other mucosal sites, including on the gastro-intestinal and urinary tract (38,39) .

Sinonasal microbiome outcomes
A higher bacterial diversity was observed when comparing the pre-op and second post-op samples (Figure 2a). This trend, however, was only statistically significant in the placebo treatment group (p-value = 0.046). Previous observations have indicated that antibiotic therapy reduces bacterial diversity (40) , noting that a decrease in bacterial diversity may be associated with CRS (2) .
The bacterial community composition also appeared to differ between the treatment groups over time (Figure 2b). Bacterial composition was similar between treatment groups at the preop stage but differed over time post-ESS. This difference may be attributed to doxycycline's impact on the patient's microbiome, preventing some bacteria from repopulating the sinuses (38) . As  (41)(42)(43) . The stronger decrease in potentially pathogenic bacteria in the placebo group also provides some evidence towards doxycycline not having a superior outcome. It should, however, be noted that our results showed a low Pearson correlation coefficient, and that other studies have found Corynebacterium to be associated with healthy patients and this genus may not play a pathogenic role but rather a beneficial role in CRS (3,44) .

Collection methods
This study provided further evidence that tissue and swab samples may have distinct microbiomes (11,12) . Our results found some similarities between the bacterial community composition when comparing the collection method. We also note a correlation between the ordination of the tissue and swab sample communities (r=0.6), indicating overlap between these communities; however, this was not extensive (Supplementary Figure 6). In contrast to bacterial composition, bacterial diversity differed significantly between collection methods (p-value = 0.0114). On average, tissue samples had higher diversity than swab samples.
This difference may be due to tissue samples also representing the bacteria present in biofilms and/or bacteria within the epithelium (11) . Our study indicates that tissue samples capture a greater bacterial diversity, but it should be noted that tissue sample collection is invasive and logistically challenging. Our study did not find a significant difference between the position of the samples (right v left) for bilateral CRS patients (p-value = 0.298), suggesting that the nostril of the sample did not impact the sinonasal microbiome significantly.

Strengths
A particular strength of this study was the molecular assessment that occurred alongside the clinical assessment, as this provided some suggestions as to why antibiotics may generate worse outcomes for the patient. A further strength of this study was that patients were excluded if they had had any antibiotics 12 weeks prior to enrolment to allow the microbiome to recover from the impacts of previously consumed antibiotics (45) . This study has also provided an opportunity to further describe the microbial ecology of CRS in patients at the time of ESS without the influence of antibiotics or prior surgery.

Limitations and further research
Patients in our study were not a priori separated into groups according to their clinical state, e.g., CRSwNP, asthma, etc. Chimeras were removed with the removeBimeraDenovo function using the method "consensus".
Contaminant taxa were identified with decontam from extraction blanks and the PCR negative control using the decontam package (Supplementary Figure 1). Contaminated taxa were identified according to the prevalence-based method. This method used the prevalence of ASV present in the extraction blanks to calculate the prevalence of known contaminants in clinical samples (2) . This method accounts for competing template DNA (from the ASV present in the sinonasal microbiome) in the clinical samples, which would reduce the abundance or prevent some contaminants from present in these samples (2) .
Using chi-square statistics (presence-absence table) or Fisher's exact tests, the probability that a taxon is a contaminant or non-contaminant is calculated (2) . 'decontam' identified 889 taxa as contaminants at a 0.5 threshold (higher sensitivity compared with the default threshold of 0.1 (2) . This threshold was selected since the extraction blanks had a high number of reads. The taxa identified as contaminants (all relatively rare taxa in patient samples) were then removed. Samples with fewer than 25,000 reads appear to be under-sampled based on diversity indices (data not shown) so were considered anomalous and removed (Supplementary Figure 2 and Table 1).

Supplementary methods
Genomic DNA extraction and 16S rRNA gene sequencing before being added to the column, and the elution buffer was incubated on the column for longer [additional 3 minutes].

Bioinformatics and statistical analysis
The raw FASTQ files from the Illumina MiSeq were imported into R (v4.0.2). ASV inference and initial filtering were performed using the 'DADA2' package (v.1.16.0) (1) . Briefly, the forward and reverse reads' quality profiles were visually examined, and it was determined that truncLen trimming was unnecessary. Reads