Molecular analysis of the endobronchial stent microbial biofilm reveals bacterial communities that associate with stent material and frequent fungal constituents

Endobronchial stents are increasingly used to treat airway complications in multiple conditions including lung transplantation but little is known about the biofilms that form on these devices. We applied deep sequencing to profile luminal biofilms of 46 endobronchial stents removed from 20 subjects primarily with lung transplantation-associated airway compromise. Microbial communities were analyzed by bacterial 16S rRNA and fungal ITS marker gene sequencing. Corynebacterium was the most common bacterial taxa across biofilm communities. Clustering analysis revealed three bacterial biofilm types: one low diversity and dominated by Corynebacterium; another was polymicrobial and characterized by Staphylococcus; and the third was polymicrobial and associated with Pseudomonas, Streptococcus, and Prevotella. Biofilm type was significantly correlated with stent material: covered metal with the Staphylococcus-type biofilm, silicone with the Corynebacterium-dominated biofilm, and uncovered metal with the polymicrobial biofilm. Subjects with sequential stents had frequent transitions between community types. Fungal analysis found Candida was most prevalent, Aspergillus was common and highly enriched in two of three stents associated with airway anastomotic dehiscence, and fungal taxa not typically considered pathogens were highly enriched in some stents. Thus, molecular analysis revealed a complex and dynamic endobronchial stent biofilm with three bacterial types that associate with stent material, a central role for Corynebacterium, and that both expected and unexpected fungi inhabit this unique niche. The current work provides a foundation for studies to investigate the relationship between stent biofilm composition and clinical outcomes, mechanisms of biofilm establishment, and strategies for improved stent technology and use in airway compromise.


Introduction
Endobronchial stents are frequently implanted to treat airway complications, with lung transplantation being a particularly common setting for indications including dehiscence, stenosis, and malacia [1,2]. Stents are composed of various materials, including uncovered metal, covered metal, and silicone [3]. Despite their value in treating medical conditions, implanted biomaterials typically become colonized via biofilms. In other biomedical devices, biofilms have been linked to device failure, especially in immunosuppressed and critically ill hosts [4]. Bacterial biofilms may also contain potential pathogens, and organisms within biofilms are more resistant to host immunity and antimicrobials than free-floating planktonic forms [5]. Biofilms on central venous catheters, intra-cardiac devices, and orthopedic devices has been explored extensively [4,5]. Understanding the nature of biofilms that form on implanted devices is an essential first step in assessing biofilm impact on outcome and/or optimizing such devices.
Remarkably little is known about biofilms on endobronchial stents, however.
Emerging culture-independent molecular tools have revolutionized our ability to understand the composition of microbial communities [6][7][8]. Such sequence-based approaches can comprehensively define entire communities, are highly quantitative, and do not depend on a priori knowledge of suspected constituents or the ability to culture individual organisms. These methods have recently been applied to oral [9][10][11] and endotracheal tube biofilms [12][13][14][15][16]. Here we present the first sequence-based analysis of luminal biofilms of endobronchial stents used for clinical treatment of critical airway compromise, most of which were in the setting of lung transplant-associated airway complications.

Subjects and sample collection
Endobronchial stents were collected from March 2014 through November 2016 from subjects who were undergoing clinically-indicated bronchoscopic removal due to need for stent replacement or to resolution of the disease process. Stents were initially placed in the airway and subsequently retrieved through rigid bronchoscopy by a team of interventional pulmonary specialists. Following removal, stents were placed in sterile specimen cups containing sterile saline for transport to the lab (15-30 minutes) and then gently rinsed again to dissociate poorly adherent material. The lumen was then vigorously swabbed circumferentially (Copan Diagnostics), and swabs kept at -80˚C. In four cases, bronchoalveolar lavage (BAL) was available from the same bronchoscopy. Clinical data was extracted from the electronic medical record. All subjects gave written informed consent under protocols approved by the University of Pennsylvania IRB (protocol #823558, #812748, #817513, #820073).

Bacterial and fungal amplification, sequencing and analysis
DNA extraction, PCR amplification, and sequencing details are included in supplemental materials. Briefly, DNA was extracted from swabs by combined chemical and mechanical (BioSpec Mini-Beadbeater-16) methods using PowerSoil DNA isolation kit (MoBio) with an additional 95˚C incubation to improve DNA recovery from fungi. Extracted DNA was amplified using barcode-labelled primers (27F forward and 338R reverse) directed at the bacterial 16S rRNA gene V1-V2 and fungal ITS-1 regions (ITS1F forward and ITS2 reverse), and sequenced on the Illumina MiSeq platform with 250-bp paired-end reads [17]. ITS amplification was assessed using automated electrophoresis with Agilent 22000 TapeStation system (Agilent Technologies, Santa Clara, CA) and if no detectable amplification occurred, or if the amplification was less than that of primer dimers, these samples were not pooled for sequencing. As a control for environmental background, sterile swabs exposed to saline in collection cups were analyzed in parallel.
Bacterial 16S rRNA gene sequences were clustered into de novo operational taxonomic units (OTUs) at 97% sequence similarity using UCLUST in the QIIME 1.91 pipeline [18], and then aligned to the GreenGenes reference database (v13_8) using PyNAST. Samples with more than 1000 reads (which included all stent swab samples) and OTU's with 2 or more hits across samples were kept for downstream analysis. Fungal ITS sequence data were processed with the PIPITS pipeline, which uses ITSx software that employs hidden Markov models to extract known ITS sequences from the amplicons. This was followed by annotation of 97% percent identity OTU representative sequences with BROCC and manual BLASTn results [19,20]. Bacterial analysis was based on relative abundances, except when comparing BAL with bronchoscopic pre-wash samples where we also examined absolute read counts so that taxa within low microbial biomass pre-wash samples are not exaggerated [6,21]. Fungal analysis was carried out using absolute read numbers rather than relative abundances given the high variability in fungal quantity between samples, also to avoid exaggeration of taxa with high proportions in specimens with very low fungal content [22]. Sequence data are available in the NCBI Short Read Archives (Project SRP154880) and our code and metadata are available at: https://github.com/johnmcginniss/stent/.

Statistical analysis
All figures and statistical tests were conducted in R (v.3.4.4; http://www.r-project.org). Alpha diversity (richness, Shannon index) was calculated using the vegan package. Beta diversity (UniFrac distances) was calculated in the QIIME pipeline (v.1.91; http://qiime.org) and used to perform principal coordinate analysis (PCoA) [23]. Clinical data were summarized using range, median, and interquartile range. Pairwise Wilcoxon rank sum test was used to compare between-group differences with Bonferroni-Holm correction for multiple testing. Fisher's exact test was used to analyze categorical data. Permutational multivariate analysis of variance (PERMANOVA) with 999 permutations was used to calculate group differences in weighted UniFrac distances. We used the hclust function from the R stats package for hierarchical clustering and the pam function from the cluster package for partitioning around medoids analysis (PAM) of weighted UniFrac data [24], with further details in the S1 File.

Subjects and stent samples
We analyzed the luminal biofilm of 46 airway stents from 20 subjects (Table 1). Nineteen were treated for airway complication following lung transplantation: 3 for anastomotic dehiscence, and 14 for stenosis or dynamic airway collapse (indication was not available for 2 subjects). All transplant subjects were maintained on an immunosuppression regimen of multiple agents (generally tacrolimus, mycophenolate, and prednisone). One subject had non-transplant-associated dynamic airway collapse. Stents were silicone (n = 20), uncovered metal (n = 18) or covered metal (n = 8).

Bacterial 16S rRNA gene sequence analysis of biofilm communities
Stent biofilm samples were subject to bacterial 16S rRNA gene amplification using V1-V2 primers followed by Illumina sequencing. After quality filtering we had approximately 6.88x10 6 reads in total, with a median read count per stent biofilm sample of 53,471 (IQR 46,252-64,614). After 97% clustering this yielded 13,769 de novo OTUs in the swab samples. For each sample, the OTU-based taxa were collapsed into the lowest taxonomic assignment available in our analysis. These were then ranked on absolute read count.
Reported are the top two taxa per sample. If the cultured organism was not in the top two it is reported with its respective rank. D Culture is from bronchoalveolar lavage or tissue. NA denotes no culture taken at the time of stent removal. E + denotes sample that underwent ITS sequencing; samples without sufficient ITS amplicon for sequencing are indicated as below threshold (BT).
The amount of 16S amplicon, as a surrogate for bacterial biomass [22,25], was markedly greater in stent biofilms compared to several types of controls (p<0.001 comparing biofilm samples to laboratory background and bronchoscope prewash; S1 Fig Subjects transplanted for suppurative lung disease had lower within-sample diversity (alpha-diversity as measured by the Shannon index) compared to those transplanted for ILD (p = 0.0134) or COPD (p = 0.0033) ( Fig 1A). In contrast, alpha-diversity did not differ across the three stent materials ( Fig 1B).
We examined between-sample diversity (beta-diversity) using a principal coordinate analysis (PCoA) of weighted UniFrac distances (Fig 2). UniFrac compares communities based on , and within-sample bacterial diversity was calculated using the Shannon Index. Each dot represents a sample, the boxplot represents the median, 75th and 25th percentiles, and points outside the whiskers are outliers. Subjects with underlying suppurative lung disease (cystic fibrosis (CF) and non-CF bronchiectasis) had lower alpha diversity compared to COPD and ILD (p < 0.05, pairwise comparison Wilcoxon rank sum test). Stent material did not impact alpha diversity.
https://doi.org/10.1371/journal.pone.0217306.g001 shared phylogenetic lineages of constituent taxa, and the weighted approach accounts for relative abundances of constituent taxa. The taxa present in greater >5% mean relative abundance driving the distribution on the PCoA were Corynebacterium, Staphylococcus, Pseudomonas, Prevotella, and Streptococcus. There was a statistically significant relationship between stent material and composition on the weighted UniFrac analysis (Fig 2B; R 2 = 0.09, p = 0.016; PER-MANOVA;). In contrast, there was only a nonsignificant trend towards clustering by diagnosis (Fig 2A; R To examine whether alternative analytical approaches might change the nature of these results, we carried out the same analysis after raising the threshold for inclusion of an OTU to �10 reads across all samples (reducing the total to 3199 OTUs) and excluding OTUs that could potentially be derived from contamination sources [26] (leaving 3167 OTUs). We also applied rarefaction to 10,000 reads per sample. However, as shown in S3 Fig, this alternative analysis did not change the associations between biofilm diversity and clinical or stent features, nor impact the ordination on the PCoA plot. Thus, these findings are robust to different methodological approaches. Subsequent analyses therefore employed the more complete dataset.
When we examined taxa at the 1% mean relative abundance level (S4 Fig), Corynebacterium continued to define one dimension, suggesting a dominant effect on these communities, whereas multiple additional taxa contributed to other dimensions of the PCoA plot. We repeated this analysis without 10 specimens from one highly sampled subject, #0099, which revealed a similar pattern (S5 Fig), indicating that this relationship is not driven by subject oversampling.
Within the Corynebacterium genus there were 865 de novo OTUs with more than 10 reads assigned. However, >90% of reads were assigned to two OTUs, accounting for 71.8% and 18.9%, respectively. The top hits by BLASTn search of the NCBI 16S rRNA database for these OTUs were C. striatum (99% coverage and 100% identity) and C. xerosis (99% coverage and 99% identity), respectively; however, there is a high degree of homology in this region among Corynebacterium, and other species also had slightly lower but still >97% coverage and identity for these two OTU's, thus precluding definitive assignment at the species level. Within the Staphylococcus genus there were 3 de novo OTUs that accounted for >95% of reads. The two most prevalent Staphylococcal OTUs could not be assigned at the genus level, while the top five BLASTn hits for the third most prevalent OTU (21% of Staphylococcal reads) were to S. aureus (100% coverage and 99% sequence identity), with other Staphylococcal species having lower coverage and/or identity.

Temporal dynamics and subject-level factors influencing biofilms
We asked whether stent biofilms characteristics were linked to the duration that stents were in place. Duration of stent was not related to biofilm total bacterial biomass (S6 Fig; r = 0.18, p = 0.24), nor PCoA axes 1 or 2 (axis 1: r = 0.06, p = 0.69; axis 2: r = -0.01, p = 0.97). This suggests that factors other than stent duration determine community composition.
Seven subjects had �2 stents available, including two with contemporaneous bilateral stents (Fig 3A). Sequential stents often showed consistent biofilm composition, but sometimes this was interrupted by gradual (#0991) or abrupt (#0099, #0988) compositional changes. In contrast, when bilateral stents were present, the communities were highly concordant (e.g. #0099 days 143, 151, 253, 262; #0988 day 85). We also visualized the 4 subjects with �3 stents on the PCoA plot (Fig 3B). Most showed considerable change. Together these results indicate that subject-level variables such as pre-transplant lung disease, and temporal factors such as time post-transplant or stent duration, did not have a primary role in determining stent biofilm composition. type 2 was clearly distinct, while types 1 and 3 reflected more of a continuum between community types (Fig 4A). The three PAM biofilm types had similar bacterial biomass based on 16S amplicon concentrations and were markedly greater than background (S1B Fig), with the Corynebacterium-dominated type 2 being slightly but non-significantly greater (p = 0.07, ANOVA). This result indicates that none of the biofilms reflect background or contamination. We queried the taxa present above 1% mean relative abundance within each PAM biofilm type (Table 2). Corynebacterium was widely present. Furthermore, Corynebacterium dominated biofilm type 2, accounting for 86% of all reads. In contrast, PAM types 1 and 3 had more taxa above 1%. Concordantly, biofilm type 2 had lower diversity than the other types (Fig 4B). Stent material was significantly associated with biofilm type (Fig 4C; p = 0.008, Fisher's exact test). Specifically, covered metal was associated with type 1 (Staphylococcus), silicone with type 2 (Corynebacterium), and uncovered metal with type 3 (polymicrobial). In contrast, biofilm type did not correlate with pre-transplant diagnosis.

Stent biofilm relationship to the bronchoalveolar lavage (BAL) microbiome and environmental controls
Four subjects had BAL performed during the stent removal bronchoscopy. We compared the stent biofilm, BAL, and washes of the bronchoscope working channel prior to the procedure (S7 Fig). The stent biofilm and BAL bacterial composition were similar, and distinct from the environmental controls. In biofilms with high Corynebacterium, this bacterium was abundant in the BAL but not in environmental controls, indicating it was authentically present in the stent biofilm and lower respiratory tract, and not derived from environmental contamination.

Fungal microbiome in stent biofilms
We used ITS sequencing to identify fungal taxa in the stent biofilm. Of the 46 specimens, 21 had no definitive fungal amplification using ITS primers and were not pooled for sequencing. We first visualized fungal taxa in the positive samples on a heat map (Fig 5). Because the presence of fungi is highly variable in lower respiratory tract specimens [22,27], we analyzed fungi by read count numbers rather than relative abundances, which can exaggerate the appearance of taxa in samples with few fungal reads [22].
Several stents yielded high numbers of ITS reads aligning to Aspergillus and Candida, and in some samples, Scedosporium, Stereum, and Sarcinomyces (Fig 5, Table 3). Candida species included C. albicans, C. glabrata, and C. tropicalis. Among samples with Aspergillus, three were confidently assigned to A. fumigatus, while the remainder could only be assigned to Aspergillus at the genus level. Scedosporium is occasionally recognized as an opportunistic pathogen. However, Stereum and Sarcinomyces were unexpected because they are typically considered environmental contaminants, yet they were present at high read numbers within biofilms and absent from our environmental controls. Environmental control specimens had very low numbers of fungal reads, suggesting that fungi identified at high read count in stent biofilms are authentic constituents rather than contamination-derived.
In an analysis of bacterial/fungal co-variation, Candida showed a significant inverse correlation with the most abundant bacterial genus, Corynebacterium (S8 Fig; ρ = -0.48, p = 0.017, Spearman's rank correlation). In contrast, there was no relationship between Candida and Streptococcus or other prevalent taxa.

Relationship between biofilm composition, airway cultures, and clinical features
Ten subjects had airway cultures concurrent with stent removal; six cultures revealed respiratory pathogens (Table 1). In five out of the six positive cultures, the cultured bacteria were identified in the stent biofilm. Importantly, the respiratory pathogen was not the most abundant 16S rRNA sequence, but rather was part of a community that included anaerobic or upper respiratory-type bacteria that are not usually cultured. One subject (#0895) grew Mycobacterium avium-intracellulare (MAI), which was not identified by sequencing. This suggests that either MAI does not participate in stent biofilm communities even if present in the lower respiratory tract, or these molecular methods are less sensitive than culture for mycobacteria [27], even though our previous work suggests that this pipeline can detect mycobacteria [17].
Conversely, we detected Enterococcus by sequencing but not culture (#0099, day 267) five days after a prior positive airway culture for this organism.
Three subjects had stents placed for anastomotic dehiscence, and two of these had the highest numbers of Aspergillus reads in our study (#0777, #0988; Fig 5 and Table 3). Subject 0077 (day 98) had a contemporaneous fungal culture, which was positive for Aspergillus. Candida spp. were detected by sequencing in many stents, including several at high levels. Candida spp. are typically reported in respiratory cultures as yeast with no further identification by our clinical microbiology lab. The stents that had Scedosporium, Stereum, and Sarcinomyces did not have contemporaneous BAL cultures. Twenty-five biofilm swab samples and seven control samples had fungal ITS amplification sufficient for sequencing. In the heat map, rows reflect fungal taxa at the lowest level assignment possible (down to the species level). Each column is a unique biofilm sample, grouped by subject. The name is in the format of: subject ID-side of stent-day post-transplant. Because the total fungal biomass is highly variable among samples, the color scale reflects absolute read counts rather than relative abundances to avoid exaggeration of taxa in low fungal content specimens. Taxa with fewer than 10 reads in any sample are excluded. https://doi.org/10.1371/journal.pone.0217306.g005 Endobronchial stent biofilm microbiome Taxa with �3000 reads in a sample are shown, along with the number of reads. All but one subject was on at least one antibiotic at the time of stent removal and most were also exposed to anti-fungal medications (S1 Table). The ubiquity of the exposure and between-subject heterogeneity limits statistical analysis of the impact of specific antimicrobials on biofilm composition, but indicates that, overall, biofilm formation is robust to antimicrobial treatment. While the goal of this study was to understand the composition of luminal biofilms established, we also queried the relationship to clinical outcome, data for which was available in 13 subjects. We asked if biofilm type (limited to the final stent in subjects with >1 stent) was related to outcome categorized as resolution of the airway problem motivating stent placement, presence of excessive granulation tissue, and stent mucus plugging. Within these categories, there was no significant relationship between stent biofilm type and outcomes (p = 1, Fisher's Exact Test), though the number of subjects in any individual outcome group was small in this cohort. Thus, the impact of biofilm types identified here on outcomes will require future prospective studies.

Discussion
Here we report the first molecular investigation of the airway stent biofilm. In this predominantly post-transplant cohort, we identified three bacterial community types-one low diversity and dominated by Corynebacterium, while the other two have greater diversity and are characterized by Staphylococcus or by a mixed population including Pseudomonas, Prevotella, and Streptococcus. Biofilm type correlated with stent material. When individuals had serial stents, the biofilm communities frequently shifted, suggesting that device-related, procedural and/or contemporaneous clinical factors, but not pre-transplant lung disease, influence biofilm formation. Fungal sequences were frequently detected, sometimes at high read counts.
Corynebacterium was the most prevalent bacteria. It dominated biofilm type 2, and was present at substantial albeit lower abundance in other biofilm types. Corynebacteria are common skin flora and are generally not considered pathogens in the respiratory tract [6,28], although there have been reports of C. striatum respiratory tract infection in immunocompromised hosts[29-32]. One previous study identified Corynebacterium in airway cultures from 4.6% of patients after lung transplantation, and within this population the presence of airway stents was associated with Corynebacterium persistence [33]. Our data suggests a potentially important role for Corynebacterium is its interactions and competition with other bacteria and fungi in the biofilm. Supporting this notion, Corynebacterium spp. have been shown to antagonize Staphylococcus aureus and Streptococcus pneumoniae in the anterior nares [34][35][36]. Furthermore, Corynebacterium spp. have recently been identified as foundational taxa in dental biofilms, serving as a nidus of nucleation for other bacteria in dental plaque [11]. Thus, in the post-transplant stent biofilm Corynebacterium may be derived from the oral cavity and thrive within the biofilm, provide a scaffold for some taxa to persist while competing with others, and play a central role in endobronchial stent biofilm structure and composition.
Despite their importance for treating airway compromise, stents may themselves increase risk of lower respiratory tract complications, including infection, granulation tissue formation and mucous plugging [37,38]. Stent colonization has been postulated to contribute to these consequences [39][40][41], and so understanding biofilms that form on these devices could be important to optimizing their development and use. We found stent material was significantly correlated with biofilm type, with silicone stents favoring the Corynebacterium-dominated biofilms, covered metal favoring Staphylococcus-type biofilms, and uncovered metal associated with polymicrobial biofilms. Although the association was imperfect and explained only part of the biofilm composition, this information if further substantiated could help guide stent design or selection for the purposes of minimizing infectious and inflammatory complications.
Such knowledge could be important in reducing the risk of colonization by potential pathogens. Future studies could address how antimicrobial-impregnated stent materials impact biofilm development.
Fungi were common within stent biofilms, with Candida being the most frequently identified, including pathogenic species C. albicans, C. glabrata, and C. tropicalis. While Candida are often considered contaminants in respiratory cultures, they have a propensity to establish biofilms [4,42,43]. Our observation of high read numbers of Candida spp. suggest they are bona fide inhabitants of these post-transplant airway stent biofilms. Aspergillus is often suspected as a cause of anastomotic complications requiring stent insertion [1], and we found a preponderance of Aspergillus reads in dehiscence cases. The presence of Aspergillus in these biofilms is not only consistent with such an association, but raises the possibility that the biofilm might also contribute to fungal persistence. Several other fungi were found in individual samples at high abundance, including Scedosporium, Stereum, and Sarcinomyces. Scedosporium is recognized as an opportunistic pathogen. In contrast, the latter two fungi are generally not considered human pathogens. Further study is needed to determine whether these biofilm inhabitants are linked to anastomotic tissue infections, either a cause or a consequence, or might perpetuate inflammation, impede healing, or cause stent/airway complications.
The fact that stent biofilms were markedly higher in bacterial biomass and differed in composition from environmental controls (S1 and S2 Figs) indicates that these findings authentically reflect biofilm and not contamination. In addition, the major findings are robust to different analytical approaches. Nevertheless, our study has several limitations. The study was observational so there was inherent heterogeneity in exposures, although accurately reflects clinical practice in complex patients. Sampling was done at the time of stent removal, so it is difficult to make inferences into the how the stent microbiome changed in situ. We investigated a relatively large number of stents (n = 46), but the number of any single stent material, underlying lung disease, and outcome group was modest. Finally, our study was done in a predominantly post-transplant population, which has distinct exposures to antibiotics and immunosuppression that may limit generalizability to stent biofilms that form in non-transplant populations.
In summary, this is the first study to comprehensive, unbiased molecular methods to interrogate the biofilm of airway stents. We identify a central role for Corynebacterium in the stent biofilm, three broad types of biofilm communities, evidence for unexpected uncultured fungi and fungal/bacterial community co-variation, a possible link between fungi and anastomotic dehiscence, and an association between stent material and biofilm type. Understanding the nature and composition of airway stent biofilms sets the stage for future studies to determine how luminal biofilms influence outcomes, identification of low-risk or high-risk biofilms and better understanding of mechanisms involved in their establishment. This knowledge will offer opportunities for improved stent technology and decision-making around stent selection and management. Four subjects had bronchoalveolar lavage (BAL) carried out contemporaneous with stent removal, with samples shown as heatmaps, along with matched bronchoscope pre-wash samples as a background control. Each column is a sample, each row is a bacterial taxon, and each group represents a different subject. The left-hand panel shows the number of matched reads per sample; because the environmental controls are low biomass samples we used absolute read counts rather than relative abundances so taxa with low numbers of reads within a low microbial biomass sample are not exaggerated. The right-hand panel shows the relative abundances within each sample. Within each subject grouping, the first column is a BAL sample (B), second is a stent biofilm (S), and third is a bronchoscope pre-wash environmental control (P). (TIF) S8 Fig. Fungal-bacterial covariation. For all samples with at least 500 fungal reads (n = 22), the absolute reads on a log10 scale of Candida within the ITS dataset (x-axis) were plotted against the proportion Corynebacterium and Streptococcus (y-axis). The line is fit with a general linear regression model with 95% confidence intervals shown in gray. There was a significant inverse relationship between Candida and Corynebacterium (ρ = -0.48, p = 0.017, Spearman's rank correlation) but no correlation with Streptococcus relative abundance. The regression line is derived from a generalized linear regression model with 95% confidence interval in gray. (TIF) S1 Table. Anti-microbial exposure at the time of stent removal. (DOCX)