Impaired diversity of the lung microbiome predicts progression of idiopathic pulmonary fibrosis

The subjects who were retrospectively included in this analysis were IPF patients who had visited Sapporo Medical University Hospital. The study was approved by the hospital’s ethics board (approval number, 292–35). Diagnosis of IPF was performed according to the ATS/ERS/JRS/ALAT statement 2011 [2]. In this study, drug history (discussed in the Results section) is not included in the exclusion criteria. The patient characteristics, results of pulmonary function tests, arterial blood gas analysis, and 6-min walk distance (6MWD) test, as well as general blood test and some biomarkers for IPF in serum at enrollment were investigated. The pulmonary function test (PFT), including determination of the FVC and diffusing capacity of lung carbon monoxide (DLco), was performed by using a CHESTAC-55 V system (CHEST M.I., Inc., Tokyo, Japan). The 6MWD test was performed in accordance with ATS guidelines. The GAP (gender, age, and physiological variables) point score was calculated by determining the subject’s gender, age in years, percent predicted FVC (%FVC), and percent predicted DL_CO (%DL_CO), as described previously [15]. PFT results at 12 ± 1 months from baseline and the change in FVC were evaluated. Deterioration was defined as a ≥ 10% relative decline in FVC from baseline and severe respiratory status in which PFTs were unable to be performed [16]. AE was defined according to the Japanese criteria [16]. These criteria are consistent with the criteria proposed by Collard and coworkers in 2007 [17].

Bronchoscopy was performed in patients with clinically diagnosed IPF. The bronchoscope was inserted via the mouth and through the vocal cords. After carefully observing the lumen of the bronchus, the bronchoscope was wedged into a targeted lobe using an intubation tube. Bronchoalveolar lavage (BAL) was performed in either the middle lobe or the lingula, irrespective of the imaging abnormality on high-resolution computed tomography (HRCT). BAL was performed with 3 instillations of 50 ml of warmed normal saline for a total of 150 ml with all possible returns collected and pooled. Samples for microbiome analysis were added to a falcon tube and stored at − 80 °C until the time of DNA extraction.

C57BL/6 J mice were purchased from Sankyo Labo Service Corporation Inc. (Tokyo, Japan). We ensured that they were equivalent in genetic quality and had stable phenotypes. Although non-littermate mice might have different microbial flora, we did not use littermates in this study to avoid a bias arising from the residing genetic inheritance. In addition to similar housing conditions provided to all animals, including food, water and co-house, the random selection ensured elimination of the bias as much as possible. The mice were maintained under specific pathogen-free conditions and used when they were 8 weeks of age. Thereafter, we divided the mice into two groups: bleomycin and normal saline groups. Bleomycin hydrochloride (Nippon kayaku Co., Tokyo, Japan) was dissolved in normal saline and continuously administrated via micro-osmotic pump (Alzet 1007D; DURECT Corporation, CA). Mice were anesthetized by inhalation of isoflurane, followed by Alzet pump insertion into the mid-back subcutaneous region. Alzet pumps contained either 100 μl of bleomycin hydrochloride (100 mg/kg) or normal saline as a control, and were designed to deliver their contents at 0.5 μL/h over 7 days. At 28 days after Alzet pump insertion, mice were sacrificed with an intraperitoneal injection of pentothal, and blood was immediately collected for obtaining serum samples. Lungs were lavaged with 1 ml of normal saline. For histological examination, lungs were fixed with 10% formalin phosphate and embedded in paraffin. Lung sections were stained with hematoxylin-eosin and Masson trichrome reagent, and examined with a microscope. Total and differential cells in the bronchoalveolar lavage fluid (BALF) were counted using the XT-1800iV (Sysmex, Japan). All experiments using mice were conducted according to the regulations of the Sapporo Medical University Animal Care and Use Committee.

The number of tuf gene expression was analyzed to estimate the number of bacteria in the BALF. Real-time PCR was performed on these samples by using a bacterial quantitative PCR kit (Takara, Japan) and a 7900 HT Fast Real-Time PCR System (Applied Biosystems). The cycling conditions were 1 cycle at 95 °C for 30 s, 35 cycles at 95 °C for 5 s, and 60 °C for 30 s.

An Ion 16S Metagenomics Kit, which is designed for rapid analyses of bacterial samples using Ion Torrent sequencing technology, was used for the sequencing analysis of the bacterial 16S rDNA gene. This kit includes 2 primer sets that selectively amplify the corresponding hypervariable regions of the 16S region in bacteria: V2–4-8 and V3–6, 7–9. The primary PCR cycling conditions were 95 °C for 10 min, followed by 27 cycles of standard PCR (95 °C for 30 s, 50 °C for 30 s, and 72 °C for 45 s), and finished with 72 °C for 7 min. Then, sequence analyses were performed by using the Ion Torrent sequencing platform (Ion Torrent PGM™). Raw sequencing data were processed by using the mother software package, as described in the Ion Torrent standard operating procedure. After sequencing, raw signal data were analyzed using Torrent Suite version 5.0. The pipeline included signaling processing, base calling, low quality read removal, and adapter trimming. Sequencing data in BAM format were further processed using Ion Reporter software v5.2. These files, with the taxonomic information for the operational taxonomic units (OTUs), were imported and further analyzed using Explicet v2.10 software (Robertson, http://​www.​explicet.​org). We restricted the principal components and regression analyses to OTUs that were present at > 1.0% of a given sample’s population and taxonomically identifiable.

Continuous variables are shown as the median with interquartile ranges. For comparisons between two groups, continuous variables were examined by using the Mann–Whitney U test on GraphPad Prism v7 software (GraphPad, Inc., San Diego, CA, USA). Alpha-diversity metrics were used to characterize the bacterial diversity within each sample. Alpha-diversity metrics, including the Shannon diversity index and Simpson diversity (1-D) index, were computed by using Explicet v2.10 software. Spearman’s correlation analysis and JMP 13.0 (SAS Institute, Cary, NC, USA) were used to evaluate the coefficients of determination (ρ), residuals, and significance (p) to identify associations between the microbiome and parameters. Principal coordinates analysis (PCoA) of the Bray-Curtis dissimilarity between samples was generated using Past v3.13 software (Hammer O. Past 3.× 2017. Available from: http://​folk.​uio.​no/​ohammer/​past/​).

The subjects included in this analysis are shown in Table 1. Thirty-four patients with suspected IPF were enrolled in this study from July 2013 to December 2015 and underwent bronchoscopy. A diagnosis of IPF was made by ≥3 pulmonologists on the basis of HRCT. All 34 patients had a consistent usual interstitial pneumonia pattern. None of 34 patients received surgical lung biopsy for definitive diagnosis. The 34 subjects with IPF were predominantly men (73.5%). The median age was 69 years old. According to the GAP score, 22 patients had stage I, 8 had stage II, and 4 had stage III disease, and the subjects were considered to have had mild or moderate disease at enrollment. The incidence of an AE within 12 months from entry was 14.7%. There were 4 deaths reported among the 34 study participants in the initial 12 months. Before the time of assessment, the participants were undergoing treatment with clarithromycin (n = 1), inhaled corticosteroids (n = 1) or antacids (n = 10), respectively. None of the participants were taking systemic immunosuppressants, N-acetylcysteine or antifibrotic agents at the time of entering this study. During the observation period, the participants received treatment with antifibrotic agents (n = 7), systemic corticosteroids (n = 4), cyclophosphamide (n = 2) and N-acetylcysteine (n = 5) at some point. Table 1 also summarises the information of the total cell counts in BALF. Although the alteration in alveolar microbial flora is often accompanied by changes in BALF cell counts, no correlation was established between BALF cell counts and the microbiome in this study.

The number of tuf gene was measured instead of the number of bacteria. Figure 1a shows the overall total bacterial load measured in 1 ml of BALF, and values ranged from 601 to 4554 (Fig. 1a). There was no statistically significant difference regardless of the history of AE (Additional file 1: Figure S1). Overview of DNA sequencing data are shown in Additional file 1: Table S1. The number of OTUs and sequences reads per BALF sample ranged from 56 to 93 and from 29,956 to 132,216 in the family level classification, respectively. Rarefaction curves were calculated to measure how well the amount of sequencing performed for a library represented the biodiversity in the library, and the measured values ranged from 0%–100%, with 100% indicating that all expected OTUs were observed, which would suggest that deeper sequencing would be unlikely to identify additional OTUs. The rarefaction curves in this analysis indicated that sufficiently deep sequencing had been achieved (Additional file 1: Figure S2). To evaluate the microbial diversity in these subjects, the Shannon and Simpson diversity indexes were examined. The values in the Shannon diversity index ranged from 2.817 to 4.623. Poor OTU diversity was obtained in three patients, which had values of 2.817, 3.281, and 3.347. The Simpson diversity index ranged from 0.727 to 0.931, which were parallel to the indices of the Shannon diversity index (Fig. 1b and c). To examine whether the diversity of the pulmonary microbiome influenced disease progression, we compared the diversity with the clinical data of 30 patients in the non-deterioration group (n = 22) and the deterioration group (n = 8; 7 had a − 10% or greater decrease in FVC and 1 could not perform the PFT) (Fig. 1d). In all patients within the deterioration group (n = 8), the diversity index tended to be smaller than that in the patients of the non-deterioration group. Additionally, the diversity index was significantly smaller in the early death populations than in the survival populations (Fig. 1e). These results suggested the possibility that the diversity of the pulmonary microbiome could be used to predict prognosis for IPF pathogenesis. The relative abundances of bacterial phyla and family OTUs with > 1.0% relative abundance per sample are shown in Fig. 1f. The most dominant lung phyla found were Firmicutes, Proteobacteria, Bacteroidetes, and Actinobacteria (37.4%, 29.4%, 19.8%, and 11.7%, respectively). Additionally, we observed Cyanobacteria and Fusobacteria in the lung. The largest and second largest family OTU in the Firmicutes were the Streptococcaceae and Veillonellaceae (14.9% and 9.7%, respectively). The second largest family OTU was Prevotellaceae (12.1%), which is a member of the Bacteroidetes. Other common OTUs identified in the family included Propionibacteriaceae, Shewanellaceae, Pseudomonadaceae, and Staphylococcaceae. Considering the diversity more deeply, we presumed that the decrease in diversity was due to an increase in Firmicutes and a decrease in Proteobacteria. Particularly concerning Firmicutes, involvement of the Streptococcaceae and Veillonellaceae families was shown (Table 2).

Next, we evaluated correlations between clinical data and microbiomes. The FVC was used as an index to understand the progress of IPF, and we first focused on the relationship between the FVC and pulmonary microbiome. We observed no significant association between bacteria burden and FVC (ρ = − 0.153; P = 0.395). However, we found that the FVC decreased in inverse proportion to the relative abundance of Firmicutes. On the other hand, the relative abundance of Proteobacteria closely paralleled the FVC results. At the family level, the relative abundances of Streptococcaceae and Veillonellaceae, the largest and second largest single OTU in a member of the Firmicutes, was significantly associated with a decline in FVC (Fig. 2a–d). Furthermore, the Shannon diversity index was positively correlated with the FVC (Table 3). The diversity index from microbial communities are also correlated to the serum level of surfactant protein D (SP-D), lactate dehydrogenase (LDH), and 6MWD, which are the clinical biomarkers or physiological assessment of progression in IPF (Table 3). The families of the Firmicutes phylum also correlated with 6MWD and serum level of SP-D (Fig. 3a and b). Bacterial burden demonstrated an inverse correlation with 6MWD (ρ = − 0.347; P = 0.051). However, we established no other obvious relationship between bacterial burden and the clinical biomarkers or physiological assessment. Taken together, these results showed that a reduction in diversity caused by an increase in Firmicutes (Streptococcaceae and Veillonellaceae families) and a decrease in Proteobacteria correlated with disease worsening, which indicated a possible contribution of these organisms to IPF pathogenesis.

Microbiomes are thought to be affected by various confounding factors such as smoking history, drugs, etc. [18]. To investigate the clear relationship between microbiome and pathological conditions, we examined the microbiome in mice with the bleomycin-induced lung fibrosis. BAL was performed at 28 days post bleomycin, then representative photomicrographs are presented in Additional file 1: Figure S5. Histopathological assessment showed that diffuse intra-alveolar fibrosis and focally dense fibrosis at the sub-pleural region resulted in increases in the number of cells, especially alveolar macrophages, in BALF (Additional file 1: Figure S5). Although the degree of fibrosis was slightly different in individual mice, all mice treated with bleomycin were included in this analysis, irrespective of the degree of fibrosis. There was no difference in total load of tuf gene between the bleomycin-treated and saline-treated mice (Fig. 4a). Conversely, the Shannon diversity index was significantly smaller in bleomycin-treated mice compared to saline-treated mice. (Fig. 4b). The most dominant lung phyla found were Firmicutes, Proteobacteria, Bacteroidetes, and Actinobacteria in saline-treated mice lungs (49.9%, 20.9%, 10.9%, and 16.0%, respectively). In particular, a significant increase was confirmed in Firmicutes in BALF from bleomycin-treated mice (p < 0.05) (Fig. 4c). These results indicate that the alternation of the microbiomes in phylum on fibrotic changes is quite similar between the human samples and mice. Principal coordinate analysis (PCoA) was performed to establish relationships among all samples. The PCoA graph demonstrated that the OTUs detected in BALF of bleomycin-treated mice were well separated from that of saline controls, which indicates that the microbial communities found in bleomycin-treated mice were different from those in the saline controls (Fig. 4d).