Analysis of the Upper Respiratory Tract Microbiotas as the Source of the Lung and Gastric Microbiotas in Healthy Individuals

For 12 study subjects, we compared matched samples from the oral wash sample with the first BAL fluid sample obtained and the second BAL fluid sample obtained from a right middle lobe segment. The objective was to confirm that research bronchoscopy did not systematically carry over oral contamination into the lower airways. After insertion of the bronchoscope past the epiglottis into the first lobe for the first BAL, the bronchoscope was retracted slightly without pulling the tip of the bronchoscope more proximal than the midtrachea (well below the epiglottis) and then repositioned into the other lobe for the second BAL. Within each subject, we determined the similarity of the bacterial communities between the oral wash and the first return BAL fluid sample and between the oral wash and the second return BAL fluid sample by calculating the θ_YC distance (1 − θ_YC) between the two comparisons. The oral-BAL fluid sample comparison showed that the first and second return BAL fluid samples did not differ significantly in terms of bacterial community composition (Fig. 2A). In addition, there was no difference in bacterial levels in the first and second BAL fluid sample as measured by quantitative PCR (qPCR) (P > 0.05, data not shown). The total bacterial load in the BAL fluid (10^3.5/5 ml) was greater than that found in the saline and preinsertion bronchoscope rinse controls (<10²/5 ml) (Fig. 2B). These results were entirely consistent with the serial bronchoscopy study by Segal et al., which also concluded that bronchoscopy is a viable option for sampling of the lung microbiome (13). Furthermore, we have previously reported that, despite significant differences between the oral and nasal microbiomes, the route of insertion of the bronchoscope (oral insertion versus nasal insertion) also did not affect the microbiota of BAL fluid (16). Our data are also consistent with a previously published study by Morris et al. that demonstrated a significant difference between oral and BAL fluid microbiota communities in healthy subjects (17). Thus, sampling of the lungs by bronchoscopy was not confounded by oral microbiome contamination.

We also compared the compositions of the microbial communities in the first and second return BAL fluid samples of individuals in our healthy cohort. There were no statistically significant differences between the bacterial communities of the first and second return BAL fluid samples (analysis of molecular variance [AMOVA] = 0.963, permutational multivariate analysis of variance [PERMANOVA] = 0.967, Table 1). Thus, the analyses in this study combined the sequence data from the two BALs to generate a single data set for each subject prior to subsampling.

Our first analysis comparing the four contiguous sites was for total bacterial levels measured by qPCR targeting levels of 16S rRNA-encoding genes in the DNA. Similar to previously published studies, the total bacterial signal level in the BAL fluid was 100- to 1,000-fold lower than that found in the oral wash fluid (Fig. 2B). The BAL fluid samples also contained 10- to 100-fold lower bacterial signal levels than the gastric aspirate samples (Fig. 2B). We could readily detect bacteria on the nasal swabs (~10⁴ 16S rRNA gene copies/swab). Although these measurements roughly correspond to the expected levels of total bacterial colonization in each of these sites, it would not be accurate to draw conclusions about the differences in bacterial densities at the different sites due to disparity in sample types and dilution effects. However, the samples do provide valid snapshots of the relative abundances of bacteria within a site and can be compared at that level. Thus, the oral and gastric compartments contained the highest levels of bacteria and the nasal cavity and lungs contained much lower levels.

We next determined the richness of each site by calculating the number of operational taxonomic units (OTUs) per sample after subsampling all of the samples to the same depth, 700 reads. The oral and gastric samples had the highest OTU richness, which was significantly greater than that of the lung samples (Fig. 2C). The OTU richness of the nasal samples was very similar to that of the BAL fluid samples.

The next objective was to compare the bacterial community composition of each of the four sites to the other sites. We used redundancy analysis (RDA) to visualize how the bacterial communities at each site relate to each other (Fig. 3) and to test whether significant amounts of variation could be explained by the differences between sampling sites (Table 1). The model proved to be significant (P = 0.005), with significant amounts of variation explained on all three axes (Table 1). Nasal communities separated from all other samples along the first RDA axis, and BAL fluid communities separated along the second RDA axis (Fig. 3A). The greatest similarity was between the oral and gastric communities (Table 1; Fig. 3B). The same relationships of the four sites were confirmed when tested by AMOVA or PERMANOVA (Table 1). Overall, these analyses demonstrated that the BAL fluid bacterial communities were significantly different from the oral wash, gastric aspirate, and nasal swab bacterial communities.

We also examined the similarity of bacterial community at each site in a given subject to either the oral or nasal community in that individual as the source community (Fig. 3C). For each subject, we calculated the θ_YC distances (1 − θ_YC) among the bacterial communities in the BAL fluid, oral wash, gastric aspirate, and nasal swab samples from that subject (Fig. 3C). In 16/28 subjects (57%), the bacterial communities in the lung and oral samples were relatively dissimilar (θ_YC distance, >0.5), and in 9/28 subjects (32%), the θ_YC distance was >0.75. Conversely, in 4/28 subjects (14%), the lung and oral communities were highly similar (θ_YC distance, <0.3). These nearly bimodal data are consistent with the reports that approximately half of healthy test subjects aspirate oral/nasal secretions in a given 24-h window (6–8). In contrast, the oral wash and gastric aspirate bacterial communities tended to be more similar within a subject, with only 1/28 subjects (3.6%) having a θ_YC distance of >0.7, as predicted, due to swallowing of saliva (and bacteria) from the mouth. For the nasal bacterial communities, >75% of the samples had a θ_YC distance of >0.7, whether the comparison was to BAL fluid or gastric communities (or oral communities; data not shown). Overall, the θ_YC distances between lung and oral bacterial communities in a subject ranged from similar to highly dissimilar, whereas the nasal bacterial communities were largely dissimilar from the BAL fluid communities.

To gain further insights into the community memberships of all four sites, the we classified OTUs taxonomically and then compared the average abundances of each site's bacterial communities (Fig. 4). The most dominant bacterial OTUs (in descending relative abundances) in the mouth, stomach, and lungs were classified as a Prevotella species, two different Streptococcus species, an unclassified member of the family Pasteurellaceae, a Fusobacterium species, and a Neisseria species. In contrast, the bacterial communities of the nasal cavity were dominated by three OTUs, classified as a Staphylococcus OTU, a Corynebacterium OTU, and a Propionibacterium OTU. Of these, the Staphylococcus and Corynebacterium OTUs were found almost exclusively in the nasal cavities of healthy individuals, while the Propionibacterium OTU was found in the lungs and stomach but not the mouth. Thus, the bacterial communities of the healthy lungs shared significant membership with the mouth but not the nose (Fig. 4); however, the relative abundance of the bacterial OTUs in the healthy lungs was different from that in the mouth, resulting in significantly different community structures, as well as different from that in the nose (Table 1; Fig. 3).

In our final analysis, we compared microbial elimination of the stomach to that of the lungs. To accomplish this, we used a paired-specimen analysis of the Prevotella genus. We chose this genus because of its high relative abundance and broad distribution in the oral cavity, as well as the gastric and pulmonary compartments, and the presence of multiple Prevotella OTUs within a subject for comparison (Fig. 5). The relative abundances of the six dominant Prevotella OTUs in oral communities (the primary source community) were compared to the corresponding abundance of each OTU at the same subject's distal sites (stomach and lungs). This resulted in 168 data points in each analysis (28 subjects with six Prevotella OTUs per subject, Fig. 5). A 1:1 ratio of OTUs at the source and distal sites would be expected if elimination were balanced with immigration; in contrast, a lack of a 1:1 ratio would imply selectivity in elimination. Comparison of oral and gastric samples revealed no evidence of selective microbial elimination in the stomach. On average, the relative abundance of Prevotella OTUs in the stomach was comparable to that in the mouth, distributed symmetrically around the line of identity (Fig. 5A). While some OTUs within a subject were detected exclusively in the mouth, a comparable number of OTUs were detected exclusively in the stomach (Fig. 5A). Thus, we found no evidence of selective elimination of Prevotella from the stomach. In marked contrast, the relative abundance of Prevotella OTUs was not symmetrically distributed around the line of identity (Fig. 5B). This observation is consistent with the idea that selective elimination of Prevotella is occurring in the lungs (and inconsistent with the conclusion that BAL fluid communities are attributable to simple bronchoscopic carryover). Numerous Prevotella OTUs of moderate to high abundance in oral communities within a subject were below the limit of detection in BAL fluid communities (<0.5%), whereas the converse was not observed (Fig. 5B). Furthermore, when a Prevotella OTU could be detected in the oral wash sample, that same OTU was below detectable levels in 44% of BAL fluid samples but only 15% of gastric samples. If a Prevotella OTU was below the detectable level in an oral sample, it could be found 44% of the time in the gastric aspirate but only 4% of the time in the BAL fluid. Thus, the lung, unlike the stomach, exhibits evidence of selective elimination of Prevotella species, one of the most abundant members of the bacterial community in the aerodigestive tracts of healthy individuals.

The University of Michigan and Veterans Affairs Ann Arbor health care system institutional review boards approved this study. All subjects provided written consent. The subjects were clinically well and could not have received antibacterials or corticosteroids in the 3 months prior to sampling. Detailed exclusion criteria were described previously (17). The demographics of the 28 subjects included in this study are presented in Table 2 (FEV₁% = FEV₁/FVC ratio, where FEV₁ is the volume that has been exhaled at the end of the first second of forced expiration and FVC is the forced vital capacity [the vital capacity from a maximally forced expiratory effort]).

The same bottle of freshly opened sterile 0.9% saline (Baxter Healthcare Corporation) was used for all sample collection from a given participant. Two saline control samples were collected for each bronchoscopy in specimen cups: an aliquot of saline directly from the bottle (neat saline) and a 10-ml sample of saline that was collected after injection through the bronchoscope with a sterile syringe (scope saline).

Oral washes were performed at the start of the bronchoscopy study visit, before any topical anesthesia or sedation, by having the participant gargle for 30 s with 20 ml sterile saline and then immediately expectorate into a sterile specimen cup. The scope saline, neat saline, and oral wash samples were placed on ice. After the oral wash was completed, subjects were asked to gargle with 10 to 20 ml of Listerine antiseptic mouthwash for 20 s.

Bronchoscopy was performed under moderate conscious intravenous sedation conducted in accordance with standard safety and monitoring procedures for clinically indicated bronchoscopy, with some modifications to reduce possible contamination of the lower respiratory tract during sample collection. Participants were premedicated with diphenhydramine, midazolam, and fentanyl, followed by topical lidocaine. The bronchoscope was advanced via the mouth to a position just above the vocal cords. Lidocaine (4%) was slowly injected around the glottis as 4 aliquots of 1 ml each, attempting when possible to have the participant oppose the cords by phonating the letter E to maximize exposure to the vocal cords and minimize penetration of the medication below the cords. The bronchoscope was then advanced rapidly, without suction and as much as possible without additional lidocaine in the lower airways, into the wedged position in a subsegment of the right middle lobe and lingula.

BAL was performed by installation of saline in 30-ml aliquots that were immediately aspirated under gentle manual suction to a total of 300 ml (150 ml for each lung segment). After insertion of the bronchoscope past the epiglottis into the first lobe (usually the left one) for the first BAL, the bronchoscope was then retracted slightly without pulling the tip of the bronchoscope more proximal than the midtrachea and then repositioned into the other lobe for the second BAL. During BAL, the participant underwent gentle mechanical percussion with a pneumatic chest percussor to maximize the yield of alveolar cells for ancillary analysis of the host immune response. BAL fluid specimens from the two lung segments were collected in separate specimen cups and stored on ice.

After both BALs were completed and while the subject was still sedated, the bronchoscope was withdrawn to a position immediately superior to the glottis, where it was stabilized at the mouth with the vocal cords in constant view. An 18-gauge sterile gastric tube was introduced via the mouth, and the participant's head was flexed gently. Passage of the tube posterior to the larynx into the esophagus was observed directly to exclude entry into the trachea. The gastric tube was advanced to 40 to 45 cm, and its correct placement was confirmed by auscultation of air injected with a syringe. The bronchoscope was then withdrawn. Next, 50 ml saline was instilled through the tube and immediately aspirated with the syringe. The return was collected and placed on ice, and the gastric tube was removed.

Finally, while the participant was still sedated, the posterior nasopharynx was swabbed gently with a single sterile swab introduced via the nares without lubrication or topical anesthetic. The tip of the swab was cut with sterilized scissors and placed into a 2-ml UltraClean Fecal DNA Isolation kit (Mo Bio Laboratories, Inc.).

Samples were transferred to the laboratory on ice. For saline controls, oral wash samples, BAL fluid samples, and gastric aspirate samples, 1 ml was transferred to a dry bead tube (Mo Bio Laboratories, Inc.), the tubes were centrifuged for 2 min at ~16,000 × g, and the supernatant was removed. This was repeated until 5 ml saline control, 5 ml BAL fluid, 5 ml oral wash, or 10 ml gastric aspirate had been transferred to the dry bead tube. Samples were then stored at −80°C until DNA isolation.

For DNA isolation, 750 µl PowerSoil DNA kit bead solution (Mo Bio) and 60 µl PowerSoil DNA kit solution C1were added to each Dry Bead Tube (containing a nasal swab or the pellet from 5 ml saline control, 5 ml BAL fluid, 5 ml oral wash, or 10 ml gastric aspirate). Samples were bead beaten for 2 min at the Homogenize setting of a Mini-BeadBeater-8 (BioSpec Products) and centrifuged for 30 s at 10,000 × g. We then continued with the PowerSoil DNA Isolation kit protocol (Mo Bio) starting with step 7 (transfer supernatant to a clean 2-ml collection tube) or transferred samples into a 1-ml collection plate to continue with the PowerSoil-htp 96 Well Soil DNA Isolation kit protocol (Mo Bio) starting with step 10 (add 250 µl solution C2 to each well).

Quantification of bacterial 16S rRNA-encoding genes was performed by real-time PCR with TaqMan hydrolysis probes on a Roche 480 LightCycler. The primers used, targeting the V1 and V2 regions of the 16S rRNA-encoding gene, were 5′ AGAGTTTGATCCTGGCTCAG 3′ (forward) and 5′ CTGCTGCCTYCCGTA 3′ (reverse), and the probe used was 5′-FAM-TA+ACA+CATG+CA+AGTC+GA-BHQ1-3′ (where FAM is 6-carboxyfluorescein, locked nucleic acid nucleotides are indicated by preceding plus signs, and BHQ1 is Black Hole Quencher 1) (9, 36). After an initial denaturation for 5 min at 95°C, 40 cycles of amplification for 30 s at 94°C, 30 s at 50°C, and 30 s at 72°C were performed. A final elongation step of 72°C for 2 min was performed. Final cooling was performed at 4°C.

Libraries of V3V5 16S rRNA gene amplicons were constructed on the basis of the HMP protocol (http://www.hmpdacc.org/doc/16S_Sequencing_SOP_4.2.2.pdf) as described previously (37, 38), except under the following PCR conditions. The PCR started with 2 min at 95°C, which was followed by (i) 20 cycles of a touchdown PCR for 20 s at 95°C, 30 s at the annealing temperature (which was 60°C in the first cycle and dropped 0.5°C with each cycle), and 5 min at 72°C and then (ii) 20 cycles of a standard PCR with 20 s at 95°C, 30 s at 50°C, and 5 min at 72°C. Large-volume Lib-L emPCRs (Roche 454) were performed, and 454 sequencing was performed at the University of Michigan with the GS FLX titanium platform (Roche) in accordance with the manufacturer's instructions.

Sequences were processed with mothur v.1.30.0 (39, 40) as described previously (37, 38, 41) through the step used for removal of sequences classified as chloroplast, mitochondria, Archaea, Eukaryota, or unknown kingdom. However, the fasta, names, and groups files (six of each, two from each sequencing run) were concatenated into a single fasta, names, and groups file, respectively, after the trim.seqs step. To control for equipment- and reagent-derived DNA that could contaminate the low-bacterial-biomass BAL fluid samples, we collected and sequenced controls from (i) scope saline, (ii) neat saline, and (iii) reagents used for DNA extraction. A customized R script based on the neutral model (42) was then used to identify and remove individual BAL fluid sequences that that likely arose by contamination from neat saline, scope saline, DNA isolation reagents, or PCR reagents (17). This model uses the relative abundance of a sequence read in the controls to calculate the probability of its detection in BAL fluid samples because of carryover. This analysis was performed prior to clustering of the sequences into OTUs at 97% similarity. Sequence reads falling within the confidence intervals of the model (neutrally distributed from controls) and outside the lower bounds of the confidence intervals (enriched in controls) were considered to be contaminating sequences and were removed from the analysis of all BAL fluid samples. Only sequences outside the upper bounds of the confidence interval with the controls as the source (overrepresented in BAL fluid samples) and unique to BAL fluid (not detected in controls) were retained for subsequent analyses.

Sequences were clustered by mothur into OTUs defined as ≥3% difference between OTUs using the average-neighbor algorithm. Consensus taxonomic classifications were generated for each OTU based on modified Ribosomal Database Project reference files. The make.Shared command was used to produce a table (shared file) of the number of sequence reads assigned to each OTU in each sample. Seven hundred sequences from each sample were subsampled, and samples with fewer than 700 sequences were discarded. A complete sample set (BAL1, BAL2, oral wash, gastric aspirate, and nasal swab), with a minimum of 700 sequences/sample, was required for each subject included in the analysis. The shared file was used to calculate the θ_YC distance (1 − θ_YC) between bacterial communities.

For analyses in which BAL1 and BAL2 were combined (identified as BAL), the sequences of each OTU in the BAL1 and BAL2 samples were added together for each subject prior to subsampling at 700. Seven hundred sequences per sample were then subsampled from the shared file, and samples with fewer than 700 sequences were discarded. A complete sample set (BAL fluid, oral wash, gastric aspirate, and nasal swab), with a minimum of 700 sequences/sample, was required for each subject included in the analysis. This shared file was used to calculate the number of OTUs observed in each sample, AMOVA based on θ_YC, PERMANOVA, redundancy analysis (RDA), relative abundances of OTUs and the θ_YC distances (1 − θ_YC) between bacterial communities in all of the samples and within each subject.

Statistical analyses were performed with Prism 5 (GraphPad Software) for analysis of variance (ANOVA), t test, and regression analysis; mothur for AMOVA (43); and vegan and R for all diversity, rank abundance, and ordination analyses. The θ_YC distance (1 − θ_YC) between bacterial communities measures differences in community structure (44) and was selected because it weighs rare and abundant OTUs more evenly than other metrics such as Bray-Curtis or Morisita-Horn (39). Using the shared file, we calculated θ_YC distances (1 − θ_YC) between communities with mothur (http://www.mothur.org/wiki/Thetayc), where S_T is the total number of OTUs in samples P and Q, p_i is the relative abundance of OTU i in sample P, and q_i is the relative abundance of OTU i in sample Q. The θ_YC distance (1 − θ_YC) incorporates the relative abundances of shared and nonshared OTUs of two communities. For data display by ordination, RDA (the constrained form of principal-component analysis) was used to focused on the variation that can be explained by the question of interest (are the microbial communities different at different sampling locations). ANOVA-like permutation testing of constrained ordinations, including RDA, was performed with the anova.cca function in the R package vegan. Significant differences in community membership identified via constrained ordination were confirmed by PERMANOVA via the Adonis function in vegan.

The sequences determined in this study are available at the NIH Sequence Read Archive (http://www.ncbi.nlm.nih.gov/bioproject) under BioProject accession numbers PRJNA263948 and PRJNA269493.