Toll-like receptors, environmental caging, and lung dysbiosis

Abstract

Recent studies have implicated lung microbiota in shaping local alveolar immune responses. Toll-like receptors are major sensors of microbiota and determinants of local epithelial homeostasis. The impact of toll-like receptor deficiency on lung microbiota is unknown. To determine whether the absence of toll-like receptors results in altered lung microbiota or dysbiosis, we compared lung microbiota in wild-type and toll-like receptor-deficient experimental mice using 16S ribosomal RNA gene quantification and sequencing. We used a randomized environmental caging strategy to determine the impact of toll-like receptors on lung microbiota. Lung microbiota are detectable in toll-like receptor-deficient experimental mice and exhibit considerable variability. The lung microbiota of toll-like receptor-deficient mice are altered in community composition (PERMANOVA P < 0.001), display reduced diversity (t test P = 0.0075), and bacterial burden (t test P = 0.016) compared with wild-type mice with intact toll-like receptors and associated signaling pathways. The lung microbiota of wild-type mice when randomized to cages with toll-like receptor-deficient mice converged with no significant difference in community composition (PERMANOVA P > 0.05) after 3 wk of cohousing. The lung microbiome of toll-like receptor-deficient mice is distinct from wild-type mice and may be less susceptible to the effects of caging as an environmental variable. Our observations support a role for toll-like receptor signaling in the shaping of lung microbiota.

INTRODUCTION

Serial investigations over the last decade have reported the presence of microbiota in the respiratory tract of humans (1, 2). These dedicated studies have shown that lung microbiota are altered in acute and chronic lung disease, correlate with alveolar cytokines, and associate with clinical outcomes including mortality (3–8). In health and disease, changes in lung microbiota can induce variation in host immune responses (9, 10). We and others have previously shown that in healthy experimental mice, lung microbiota are detectable, correlate with alveolar cytokines, and are altered and causal in models of lung disease (4, 6, 11, 12). However, the mechanisms through which lung microbiota impact alveolar immunity remain poorly understood.

Innate immunity is charged with the detection and response of the host immune system to invading pathogen but also to the sensing of local commensal microbiota (13). Host interactions with microbiota are often dependent on the engagement of toll-like receptors (TLRs) (14). TLRs are pattern recognition receptors, germline-encoded proteins that sense and recognize highly conserved microbial products termed pathogen (PAMP) and microbe-associated molecular patterns (MAMPs) (15, 16). There are 10 reported human TLRs and 12 murine TLRs identified to date (14). They have relatively specific ligand interactions and can exist on the cell surface or in endosomes. For instance, surface-bound TLR2 senses and engages peptidoglycan (17), TLR4 engages lipopolysaccharide (18), and TLR5 engages flagellin (19). Within the lung, TLRs from TLR1-10 are expressed on epithelial cells and have critical roles in the pathogenesis of most acute and chronic inflammatory pulmonary disease (20–22). TLRs are also responsible for the sensing of molecular patterns associated with cellular damage and inflammation, so called danger-associated molecular patterns (DAMPs) (23, 24). As such TLRs occupy a key role in lung injury and acute inflammation mediated by noninfectious and environmental insults (25, 26).

The gut microbiota have a profound effect on human health and immunity (13). Several decades of study have identified that gut microbiota interact with the host through TLR engagement of microbiota-derived products to regulate physiological homeostasis (27). In recent years, studies have supported a regulatory role for gut microbiota in the activation of innate and adaptive pulmonary immunity through multifaceted mechanisms (28–31). The ability of TLRs to sense and engage commensal microbiota and differentiate between commensal microbiota and pathogen is crucial to intestinal homeostasis (32). Key features of the lung microbiome are significantly different to the gut microbiome and this has crucial implications for host-microbiota interactions within both organs (33–35). Relatively stable microbial communities reside within the cecum facilitating a potential multitude of regulatory interactions with the host (13). However, within the lung, complex communities of microbiota exist at considerably lower biomass and are dynamic, subject to several ecological pressures that govern species immigration, emigration, and replication within the lung environment (4). Given the role of TLRs in inflammatory lung disease, known correlations between lung microbiota and alveolar inflammation, it is biologically plausible that TLR expression would contribute to shaping lung microbial communities and contribute to alveolar inflammation through altered microbiota sensing, interactions, and subsequent dysbiosis.

We hypothesized that the lung microbiota of TLR-deficient (TLR^−/−) mice would demonstrate dysbiosis with altered community structure and altered diversity. TLRs may have a role in regulating the clearance and/or expansion of pathobionts within the lung microenvironment. We have previously shown that healthy mice from different vendors demonstrated converging lung microbiota when cohoused and that this occurred within 7 days (4). We also hypothesized that in the absence of specific TLR genes, randomized caging would not result in significant changes in the lung microbiota of TLR^−/− mice. To study these relationships, we leveraged our knowledge of the impact of environmental caging on lung microbiota. Here, we show that TLR^−/− mice have altered lung microbiota compared with wild-type mice with intact TLRs and that caging may have a significant impact on lung microbiota of wild-type mice when cohoused with TLR^−/− mice.

MATERIALS AND METHODS

Ethics Statement

All animal experiments reported in this manuscript were approved by the Institutional Animal Care and Use Committee at the University of Michigan. All laboratory animal care policies follow the Public Health Service Policy on Humane Care and Use of Laboratory Animals at the University of Michigan.

Experimental Mice

Six-wk-old mice were sourced from a commercial vendor (Jackson) and transported in the same shipment to the University of Michigan Laboratory Animal Care facility. At the commercial vendor’s facility, these transgenic mice were housed in different rooms. To facilitate randomized caging, we limited use of experimental mice to female sex. Mice were housed under specific pathogen-free (SPF) conditions and received water and standard chow ad libitum. A total of 77 mice were used in the experiments described in this manuscript as follows: 20 wild-type C57BL/6 mice, 20 TLR2^−/− (B6.129-Tlr2tm1Kir/J) mice, 17 TLR4^−/− (B6(Cg)-Tlr4tm1.2Karp/J) mice, and 20 TLR5^−/− (B6.129S1-Tlr5tm1Flv/J) mice. All transgenic TLR mice were derived from a C57BL/6 background. These experimental mice were housed in 16 cages. Cage bedding was changed according to institutional timelines and protocols.

Ribosomal RNA Gene Sequencing

Deoxyribonucleic acid (DNA) was isolated, and the V4 region of the bacterial ribosomal ribonucleic acid (rRNA) gene amplified and sequenced using previously published protocols that employ multiple DNA isolation, procedural, and sequencing controls to control for the effects of contamination in low biomass specimens (4, 6). In brief, the V4 region of the 16S rRNA gene was amplified using published primers (36) and the dual indexing sequencing strategy was developed by the laboratory of Patrick D Schloss (37). Sequencing was performed using the Illumina MiSeq platform, using a MiSeq Reagent kit V2 (500 cycles), according to the manufacturer’s instructions with modifications found in the Schloss standard operating procedure (38). Primary PCR cycling conditions were 95°C for 2 min, followed by 20 cycles of touchdown PCR (95°C 20 s, 60°C 20 s, and decreasing by 0.3°C each cycle, 72°C 5 min), then 20 cycles of standard PCR (95°C for 20 s, 55°C for 15 s, and 72°C for 5 min), and finished with 72°C for 10 min.

Bacterial DNA Measurement

Bacterial DNA was quantified in experimental specimens and isolation and procedural controls using a previously published protocol (4) with a QX200 droplet digital PCR (ddPCR) system (Bio-Rad). Primers and cycling conditions were performed according to a previously published protocol (39). Specifically, primers were 5′- GCAGGCCTAACACATGCAAGTC-3′ (63 F) and 5′- CTGCTGCCTCCCGTAGGAGT-3′ (355 R). The cycling protocol was 1 cycle at 95°C for 5 min, 40 cycles at 95°C for 15 s, and 60°C for 1 min, 1 cycle at 4°C for 5 min, and 1 cycle at 90°C for 5 min all at a ramp rate of 2°C/s. The BioRad C1000 Touch Thermal Cycler was used for PCR cycling. Droplets were quantified using the Bio-Rad Quantisoft software. Two replicates were used per sample. Negative control specimens were used and were run alongside lung specimens.

Cage Randomization

Mice were either confined to the same cohort on shipment arrival and cohoused with mice of the same genotype, that is, wild-type mice with wild-type mice, or mice were randomized 1:1 to fill cages. This ensured that at least one TLR^−/− mouse on each background was located within each cage.

Mouse Lung Tissue Specimens

Mice were euthanized via CO₂ asphyxiation and iatrogenic pneumothorax as per institutional protocols. Instruments were rinsed with ethanol and flamed between each organ. Murine lungs were excised, placed in tubes containing 1 mL of sterile water and homogenized mechanically by using a Tissue-Tearor (Biospec Products, Bartlesville, OK). The tissue homogenizer was cleaned and rinsed in ethanol and water between each tissue specimen. Water control specimens from homogenization, and samples exposed to cleaned instruments, were included in sequencing as procedural controls. Lung specimens were homogenized immediately after extraction.

Statistical Analysis

16S sequence data were processed and analyzed using mothur v. 1.44.3 software according to the Standard Operating Procedure for MiSeq sequence data using a minimum sequence length of 250 E4 base pairs (37). For each experiment and sequencing run, a shared community file and a phylotyped (genus-level grouping) file were generated using operational taxonomic units (OTUs) binned at 97% identity generated using the dist.seqs, cluster, make.shared, and classify.otu commands in mothur. OTU numbers were arbitrarily assigned in the binning process and are referred to in association with their most specified level of taxonomy. Classification of OTUs was carried out using the mothur implementation of the Ribosomal Database Project (RDP) Classifier and the RDP taxonomy training set 14 (Trainset14_032015.rdp), available on the mothur website. The vegan and mvabund package were used to further analyze community sequencing data. For relative abundance and ordination analysis, samples were normalized to the percent of total reads and we restricted analysis to OTUs that were present at greater than 1% of the sample population. All OTUs were included in diversity analysis. Direct community similarity comparisons were performed using the Bray-Curtis similarity index. We performed ordinations using principal component analysis (PCA) on Hellinger-transformed OTU tables generated using Euclidean distances. We determined significance of differences in community composition using PERMANOVA (adonis) with 10,000 permutations using Euclidean distances. We compared means via Student’s t test and ANOVA as appropriate. All analysis was carried out in R v. 4.0.0. [R Core Team (2020). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. URL https://www.R-project.org/]. Sequences are available via the NCBI sequence read archive (accession number PRJNA695440). OTU, taxonomy, and metadata tables are available at https://github.com/dodwyer-lung/TLR_lung_microbiota_cage.

RESULTS

Toll-Like Receptor Signaling, Environmental Caging, and Study Design

Previous work demonstrated that lung microbiota are detectable in the healthy murine lung and correlate with alveolar immunity (4). Lung microbiota are susceptible to the effects of caging, and lung microbial communities converge in healthy experimental mice cohoused together for 7  days (4). We chose to leverage these findings to improve our understanding of the relationship between lung microbiota and TLRs. A cohort of mice including TLR-deficient mice (TLR2^−/−, TLR4^−/−, and TLR5^−/−) and appropriate wild-type control mice on the same C57BL/6 genetic background were delivered within the same shipment from a commercial vendor. These experimental inbred mice were housed in different rooms within the commercial vendor facilities. These transgenic mice lack the receptor apparatus to sense and respond broadly to lipoprotein (TLR2^−/−), lipopolysaccharide (TLR4^−/−), and flagellin (TLR5^−/−), encompassing the ability to recognize and modify local innate immune responses to a large array of Gram-positive and Gram-negative microbiota (Fig. 1A). On shipment arrival, mice remained cohoused in their original cohorts by genotype or were randomized 1:1 to new cages (Fig. 1B). Mice remained in these cages for 3 wk at which time mice were euthanized and the lungs were extracted for DNA isolation and 16S rRNA gene sequencing for community sequencing. Three weeks was chosen to ensure adequate time for the effects of cohousing on lung microbiota to occur and to mitigate the potential impact of changes in gut microbiota that may occur. Murine lung DNA was also quantified using ddPCR to accurately measure bacterial DNA (Fig. 1C). On comparing bacterial burden in murine lung specimens to isolation and procedural control samples derived from the DNA extraction and sequencing process, we found that bacterial burden was significantly higher in the murine lung than isolation and procedural controls samples (Mann–Whitney P < 0.0001) (Fig. 1D), consistent with previous studies (4). Analysis of 16S rRNA gene community sequencing data demonstrated that murine lung specimens were distinct in community structure from isolation and procedural control samples (PERMANOVA P < 0.0001) (Supplemental Fig. S1; all Supplemental Material is available at http://doi.org/10.5281/zenodo.4739466). Murine lung specimen microbial communities were also richer (Shannon diversity index) (Mann–Whitney P < 0.0001) (Supplemental Fig. S2) and more even (PERMANOVA P = 0.001) (Supplemental Fig. S2) than isolation and procedural controls samples. Ranked relative abundance of murine lung samples and isolation and procedural controls can be found in the supplemental data section (Supplemental Fig. S3). We concluded that our sequencing protocols were adequate to detect bacterial DNA and analyze community structure and α diversity in the murine lung. All OTUs are included in the analysis, and no specimens were removed.

Figure 1.

Toll-like receptor engagement, randomized caging study design, and adequacy of low biomass murine lung specimen 16S rRNA gene sequencing. A: schematic representation of toll-like receptor 2 (TLR2), toll-like receptor 4 (TLR4), and toll-like receptor 5 (TLR5) and specific bacterial ligand interactions. TLRs dimerize to sense and signal when engaging a specific ligand. TLR2, TLR4, and TLR5 are cell surface-bound TLRs and engagement of ligand results in a proinflammatory downstream signaling cascade that modifies the immune response. B: our caging strategy involved keeping a cohort of mice caged by genotype and randomizing mice independent of genotype to separate cages in a 1:1 order. C: study design involved the receipt of experimental mice in the same shipment followed by 3  wk of cohousing as per the randomization strategy above. After 3 wk, mice were euthanized and murine lungs were extracted and processed for DNA isolation and amplification of the 16S rRNA gene for subsequent analysis of microbial communities. The DNA extracted form murine lungs and from isolation and procedural controls was also quantified using droplet digital PCR (ddPCR) platform. D: bacterial DNA was significantly greater as measured by ddPCR in all murine lungs specimens compared with isolation and procedural controls (****P < 0.0001, Mann–Whitney U test). DNA, deoxyribonucleic acid; rRNA, ribosomal ribonucleic acid; TLRs, toll-like receptors.

TLR Deficiency is Associated with Alterations in Lung Microbiota

TLRs are major determinants of the sensing and response to MAMPs from commensal bacteria. TLRs are known to play a role in gut epithelial homeostasis but our current understanding of the role TLRs play in shaping lung microbiota if any is limited. TLR2, TLR4, and TLR5 will broadly sense Gram-positive and Gram-negative bacteria. Here, we analyzed experimental mice that were cohoused by genotype for 3  wk. We initially examined community structure using PCA of Hellinger-transformed data and visualized these data by ordination (Fig. 2A). The community composition of all TLR^−/− mice is significantly different to that of wild-type mice by permutational multivariate analysis of variance (PERMANOVA) (P < 0.001) (Fig. 2A). We next examined the ranked relative abundance of OTUs in all wild-type mice (Fig. 2B) and the relative abundance of lung microbiota in TLR^−/− mice ranked in order of abundance in wild-type mice (Fig. 2C). Although several abundant OTUs were shared between groups, several OTUs were not detectable in TLR^−/− mice. Ranked abundance of OTUs based on genotype is presented in the supplement data (Supplemental Figs. S4, S5, and S6). We next used an unbiased regression-based model (mvabund) to determine the significant OTUs contributing to this altered community composition in wild-type and TLR^−/− experimental mice. There was a significant difference in lung community composition by mvabund (P < 0.001). After accounting for multiple comparisons, Sphingomonas OTU0044 (P = 0.001) (Fig. 2D), Aurantimonas OTU0082 (P < 0.001) (Fig. 2E), Pseudomonas OTU0091 (P < 0.001) (Fig. 2F), Brevundimonas OTU0196 (P = 0.007) (Fig. 2G), and Novosphingobium OTU0191 (P = 0.018) (Fig. 2H) were found at increased abundance in wild-type mice compared with TLR^−/− mice. OTU0091 was detectable in 1 of 17 control specimens at a relative abundance of less than 1%. All of the other OTUs referenced above were undetectable in isolation, procedural, and sequencing control specimens (Supplemental Table S1). We examined the α diversity of lung microbial communities in experimental mice. We found that wild-type mice demonstrated greater diversity with increased richness as measured by rarefaction (PERMANOVA P = 0.009) (Fig. 2I) and community evenness as measured by Shannon diversity index (SDI) (t test P = 0.0075) (Fig. 2J). Finally, we examined the lung bacterial burden (total bacterial DNA) between cohoused wild-type mice and cohoused TLR^−/− mice. We found that cohoused wild-type mice had significantly greater detectable bacterial DNA than cohoused TLR^−/− mice (t test P = 0.016) (Fig. 2K). We concluded that cohoused TLR^−/− mice exhibited lung microbiota that were altered in community composition, less diverse, and less abundant compared with wild-type mice with preserved TLRs. However, whether these observations were the result of caging or TLR-deficiency was undetermined.

Figure 2.

TLR deficiency is associated with alterations in lung microbiota. A: principal component analysis (PCA) of 16S rRNA data from wild-type and toll-like receptor-deficient (TLR^−/−) experimental mice cohoused by genotype demonstrate that when clustered by genotype status (wild-type vs. TLR^−/−), lung microbial communities are significantly different in compositional heterogeneity (PERMANOVA P < 0.001). B and C: ranked relative abundance of operational taxonomic units (OTUs) in wild-type mice ranked by relative abundance and TLR^−/− mice (ranked by the relative abundance of OTUs in wild-type mice). D–H: Mvabund negative binomial regression model identifies several OTUs that are significantly more abundant in wild-type than TLR^−/− and contributing to community differences. I: rarefaction curves of sequencing data show that lung microbiota from wild-type mice are significantly different and more even in comparison to TLR^−/− mice (PERMANOVA P = 0.009). J: Shannon diversity indices are significantly higher in cohoused wild-type mice compared with cohoused TLR^−/− mice (t test P = 0.0075). K: in addition, the total lung bacterial DNA quantified in wild-type mice was significantly greater than that quantified in TLR^−/− mice (t test P = 0.016). n = 37, all female experimental mice. Statistical testing by PERMANOVA, Mann–Whitney U test or unpaired t test (**P < 0.01, *P < 0.05). DNA, deoxyribonucleic acid; PC1, principal component 1; PC2, principal component 2; PERMANOVA, permutational multivariate analysis of variance; rRNA, ribosomal ribonucleic acid; TLR, toll-like receptor.

Caging as an Environmental Variable and Its Impact on Lung Microbiota

Having established that the lung microbiota of TLR^−/− experimental mice were broadly altered compared with wild-type mice from the same vendor and on the same genetic background, we next sought to determine the impact of randomized caging on this observation. Here, mice were randomized to cage independent of genotype in a 1:1 order. After 3  wk of randomization, we found that lung microbiota in wild-type and TLR^−/− mice were not significantly different when compared by PCA, visualized by ordination, and tested statistically by PERMANOVA (P = 0.37) (Fig. 3A). A binomial regression model (mvabund) was used and while overall the model was statistically significant, after accounting for multiple comparisons the model did not identify differentially abundant OTUs between wild-type and TLR^−/− mice (Supplemental Table S2). We next examined α diversity between randomized wild-type and TLR^−/− mice and found no significant difference in SDI (Mann–Whitney P = 0.70) (Fig. 3B) or by rarefaction curves (PERMANOVA P = 0.67) (Fig. 3C). Finally, we examined bacterial DNA burden quantified by ddPCR in wild-type and TLR^−/− mice and found no significant difference (Mann–Whitney P = 0.18) (Fig. 3D). We concluded that upon random housing, lung microbiota composition in wild-type and TLR^−/− mice broadly converged, with no statistically significant differences in overall composition, α diversity, or bacterial burden. This in turn supports a significant role for environment in shaping lung microbiota.

Figure 3.

Caging alters lung microbiota. A: principal component analysis (PCA) of 16S rRNA data from wild-type and toll-like receptor-deficient (TLR^−/−) mouse lungs demonstrate that on randomized caging for 3  wk, the lung microbiota of wild-type and TLR^−/− mice converge (PERMANOVA P = 0.37). B: no differences noted in Shannon diversity indices of lung microbiota in randomized wild-type and TLR^−/− mice (Mann–Whitney P = 0.70). C: rarefaction curves demonstrating no significant differences in evenness of lung microbial communities’ in wild-type and TLR^−/− mice (PERMANOVA P = 0.67). D: total lung bacterial DNA is not significantly different in randomized wild-type and TLR^−/− mice (Mann–Whitney P = 0.18). n = 40 animals, all female experimental mice. Statistical testing by PERMANOVA, Mann–Whitney U test. DNA, deoxyribonucleic acid; PC1, principal component 1; PC2, principal component 2; PERMANOVA, permutational multivariate analysis of variance; rRNA, ribosomal ribonucleic acid.

Caging as an Environmental Variable on the Lung Microbiome in TLR^−/− Models

Cage randomization results in changes to lung microbiome composition, α diversity, and bacterial burden. To better understand whether the mechanisms promoting these changes were dependent on TLR signaling or wholly dependent on cage environment, we next sought to examine changes in lung microbiota of mice cohoused by genotype (nonrandomized) and mice cohoused randomly. We hypothesized that if TLRs contributed to the regulation of lung microbiota, then no significant changes would occur on cohousing with other mice, i.e., lung microbial communities in cohoused TLR^−/− mice would not be significantly different to lung microbiota communities in randomly housed TLR^−/− mice on the same genetic background. When we selected the randomized and cohoused wild-type mice, we found that on analyzing the data by PCA and visualized by ordination, there was a significant difference in lung microbial communities between randomized wild-type and cohoused wild-type mice (PERMANOVA P < 0.001) (Fig. 4A). To confirm this, we next examined Bray-Curtis dissimilarity between experimental groups (examining dissimilarity by the presence and absence of shared OTUs and relative abundances). We used non metric multidimensional scaling (NMDS) to visualize these comparisons, where objects that are similar are ordinated closer together and objects that are dissimilar are placed further apart. In wild-type mice, lung microbial communities are dissimilar in both mice cohoused together and housed randomly (Supplemental Fig. S7). However, cohoused wild-type mice are broadly more similar to each other than randomly housed wild-type mice (PERMANOVA P < 0.001). To identify OTUs that are significantly different in relative abundance between experimental groups, we used a binomial regression model (mvabund). This model identified significant differences in the relative abundance of OTUs overall (P = 0.005) and selected OTU0082 as significantly different between groups of wild-type mice (P = 0.02). We then analyzed data from randomized and cohoused TLR2^−/− mice. When analyzed by PCA and visualized by ordination, randomized and cohoused TLR2^−/− mice clustered together and there was no statistically significant difference between lung microbiota in these experimental mice (PERMANOVA P = 0.97) (Fig. 4B). Both cohoused TLR2^−/− mice and TLR2^−/− mice housed randomly demonstrate considerable dissimilarity (Bray-Curtis) but no significant difference is found between the groups (PERMANOVA P = 0.39) (Supplemental Fig. S8). PCA and ordination of lung microbiota from randomly housed and cohoused TLR4^−/− mice show that these mice groups clustered together and lung microbiota were not significantly different (PERMANOVA P = 0.29) (Fig. 4C). Again, in cohoused TLR4^−/− mice and TLR4^−/− mice housed randomly, there is considerable dissimilarity (Bray-Curtis) but no reported significant differences between both groups (PERMANOVA P = 0.28) (Supplemental Fig. S9). We also report no significant differences in lung microbiota on comparing randomized and cohoused TLR5^−/− experimental mice by PCA (PERMANOVA P = 0.38) (Fig. 4D). Cohoused TLR5^−/− mice and TLR5^−/− mice housed randomly, lung microbiota are dissimilar but no significant differences are noted between groups (PERMANOVA P = 0.89) (Supplemental Fig. S10). We next examined α diversity between these groups of randomized and cohoused mice. There were no significant differences in Shannon diversity index between randomized and cohoused wild-type mice (t test P = 0.68) (Fig. 4E). There was a significant difference in Shannon diversity index on comparing randomized and cohoused TLR2^−/− mice (Mann–Whitney P = 0.0029) (Fig. 4F), with increased α diversity noted in randomized TLR2^−/− mice. We report no significant differences on comparing Shannon diversity indices of lung microbiota in randomized and cohoused TLR4^−/− mice (t test P = 0.79) (Fig. 4G). We found a significant difference in Shannon diversity index between randomized and cohoused TLR5^−/− mice (t test P = 0.019) (Fig. 4H) with increased diversity in randomized TLR5^−/− mice. Measurement of bacterial DNA in murine lungs was not noted to be significantly different between cohoused and randomly housed mice (Supplemental Fig. S11). The α diversity of lung microbiota in TLR^−/− mice can increase on randomized caging but this is not accompanied by an increase in bacterial burden. Our observations suggest that the compositional heterogeneity of lung microbiota in TLR^−/− mice is changed minimally by environmental caging compared with wild-type mice. In turn, we found that the lung microbiota of mice demonstrate considerable dissimilarity as previously published (4). Wildtype mice exhibit less community dissimilarity in lung microbiota compared with TLR^−/−. Although random housing promotes increasing dissimilarity in wild-type mice, it does not promote significant changes in dissimilarity or similarity in TLR^−/− mice, generally supporting our observations using PCA.

Figure 4.

Lung microbiota in TLR^−/− mice and the environment. A: principal component analysis (PCA) and ordination of 16S rRNA data from wild-type mice both cohoused (housed with other wild-type mice) and cage randomized demonstrate significant differences in the composition of lung microbiota (PERMANOVA P < 0.001). B: lung microbiota from cohoused and cage randomized toll-like receptor 2 deficient (TLR2^−/−) mice are not significantly different when analyzed by PCA, visualized by ordination, and clustered by caging status (PERMANOVA P = 0.97). C: lung microbiota from toll-like receptor 4 deficient (TLR4^−/−) mice are not significantly different in cohoused TLR4^−/− mice and cage randomized TLR4^−/− mice. (PERMANOVA P = 0.29). D: lung microbiota are again not significantly different in cohoused toll-like receptor 5 deficient (TLR5^−/−) mice and cage randomized TLR5^−/− mice (PERMANOVA P = 0.38). E–H: Shannon diversity is not significantly different in cage randomized and cohoused wild-type mice (t test P = 0.68) and cage randomized and cohoused TLR4^−/− mice (t test P = 0.79). However, cage randomization in TLR2^−/− and TLR5^−/− mice does result in increased Shannon diversity in these animals compared with cohoused TLR2^−/− (Mann–Whitney P = 0.002) and cohoused TLR5^−/− mice (t test = 0.02). n = 10 animals per group, except for cohoused TLR4^−/− where n = 7, all female experimental mice. Statistical testing by PERMANOVA, unpaired t test, or Mann–Whitney U test where appropriate (**P < 0.01, *P < 0.05). PC1, principal component 1; PC2, principal component 2; PERMANOVA, permutational multivariate analysis of variance; rRNA, ribosomal ribonucleic acid; SDI, Shannon diversity index.

Caging as an Environmental Variable and Lung Microbiota Drift in Wild-Type Mice

Given the impact of randomization on lung microbiota composition, diversity, and burden, we next sought to examine whether the lung microbiota of wild-type mice were predictably different from TLR^−/− mice. We initially selected cohoused wild-type mice and compared lung microbiota in these mice to lung microbiota in cohoused TLR2^−/− experimental mice. These data, analyzed by PCA and visualized by ordination (Fig. 5A), show that lung microbiota composition is significantly different between wild-type and TLR2^−/− mice (PERMANOVA P < 0.001). Using a regression model (mvabund P = 0.006), we identified three OTUs that were differentially abundant in wild-type mice compared with TLR2^−/− mice including OTU0044 (P = 0.01), OTU0082 (P = 0.03), and OTU0091 (P = 0.03) after accounting for multiple comparisons. We next examined lung microbiota in cohoused wild-type mice and cohoused TLR4^−/− mice and found that when analyzed by PCA, clustered by TLR status, and visualized by ordination, lung microbiota were significantly different between these two groups of experimental mice (PERMANOVA P < 0.001) (Fig. 5B). A regression model identified significant differences between lung microbiota in cohoused wild-type and TLR4^−/− mice (mvabund P = 0.015) with OTU0044 selected as differentially abundant in wild-type mice after accounting for multiple comparisons (P = 0.003). In cohoused wild-type mice and cohoused TLR5^−/− mice, again clustered by TLR status and visualized by ordination, the lung microbiota were significantly different between experimental groups (PERMANOVA P < 0.001) (Fig. 5C). A regression model (mvabund P = 0.002) identified OTU044 (P = 0.001), OTU0082 (P = 0.016), and OTU0091 (P = 0.021) as differentially abundant in wild-type mice compared with TLR5^−/− mice. To confirm our findings, we studied Bray–Curtis dissimilarity in these experimental groups, visualized using NMDS. Cohoused wild-type mice are significantly more similar than cohoused TLR2^−/− mice (PERMANOVA P = 0.001) (Supplemental Fig. S12), cohoused TL4^−/− mice (PERMANOVA P = 0.001) (Supplemental Fig. S13), and cohoused TLR5^−/− mice (PERMANOVA P < 0.001) (Supplemental Fig. S14). We next analyzed lung microbiota in randomized wild-type mice compared with lung microbial community composition in randomized TLR2^−/−, TLR4^−/−, and TLR5^−/− mice. Using PCA and ordination to visualize the data, we report no significant differences between the lung microbiota composition of randomly housed wild-type mice and randomly housed TLR2^−/− mice (PERMANOVA P = 0.97) (Fig. 5D), TLR4^−/− mice (PERMANOVA P = 0.29) (Fig. 5E), and TLR5^−/− mice (PERMANOVA P = 0.38) (Fig. 5F). Importantly, there were no statistically significant differences in the composition of lung microbial communities when comparing between randomized TLR2^−/− to TLR4^−/−, TLR2^−/− to TLR5^−/−, and TLR4^−/− to TLR5^−/− groups, respectively (PERMANOVA P > 0.05). These findings are confirmed by Bray-Curtis dissimilarity scores, whereby the relative similarity observed in cohoused wild-type mice is lost on cage randomization with TLR-deficient mice (Supplemental Figs. S15, S16, and S17). We next examined α diversity by measuring the Shannon diversity index of lung microbiota in experimental groups. In cohoused mice, there is significantly greater diversity in wild-type mice compared with TLR2^−/− mice (P < 0.01) (Supplemental Fig. S18). In randomized cages, the difference in Shannon diversity between wild-type and TLR2^−/− mice was not maintained (Supplemental Fig. S18). There was no significant difference in the Shannon diversity index of lung microbiota in cohoused wild-type mice and TLR4^−/− mice (Supplemental Fig. S18) and randomized wild-type and TLR4^−/− mice (Supplemental Fig. S18). We report a significant difference between cohoused wild-type mice and cohoused TLR5^−/− mice in Shannon diversity index (P < 0.05) (Supplemental Fig. S18). However, on randomization, this difference in Shannon diversity index was not maintained (Supplemental Fig. S18). Analysis of bacterial burden demonstrated significant differences in TL5^−/− mice compared with wild-type only. In cohoused wild-type and TLR5^−/− mice, bacterial burden was significantly greater in wild-type mice (P < 0.05) (Supplemental Fig. S19) and in randomized wild-type and randomized TLR5^−/− mice, bacterial burden was greater in TLR5^−/− mice compared with wild-type mice (P < 0.05) (Supplemental Fig. S19). We concluded that cohoused wild-type mice had different lung microbiota compared with TLR^−/− mice but these differences were driven by similar changes in community composition and were not different or dependent on specific TLR gene deficiency. Given the absence of significant changes between cohoused and randomized TLR^−/− mice, these observations support changes in the lung microbiome of wild-type mice on randomized caging with increasing dissimilarity and altered community composition.

Figure 5.

Impact of cage environment on the lung microbiota of wild-type mice. Lung microbiota were analyzed by 16S rRNA gene sequencing followed by principal component analysis (PCA) and ordination. A: selection of lung microbial communities from cohoused wild-type and cohoused toll-like receptor 2 deficient (TLR2^−/−) mice demonstrates significantly different lung microbial communities (PERMANOVA P < 0.001). B: cohoused wild-type and cohoused TLR4^−/− mice with PCA of 16S rRNA data and ordination demonstrate significant differences in lung microbiota (PERMANOVA P < 0.001). C: PCA and ordination of lung microbial communities form cohoused wild-type and cohoused TLR5^−/− mice demonstrate significantly different lung microbiota (PERMANOVA P < 0.001). D–F: cage randomization of wild-type and TLR2^−/− (PERMANOVA P = 0.97), TLR4^−/− (PERMANOVA P = 0.29), and TLR5^−/− (PERMANOVA P = 0.38) mice results in a converging of lung microbiota without statistically significant differences noted. n = 10 animals per group, all female experimental mice. Statistical testing by PERMANOVA. PERMANOVA, permutational multivariate analysis of variance; rRNA, ribosomal ribonucleic acid.

DISCUSSION

In this study, using caging as an environmental variable to determine the relationship between local TLR expression and lung microbiota, we show that the lung microbiome of TLR^−/− mice is distinct from that of wild-type mice. The lung microbiota of TLR^−/− mice have altered composition (β diversity), reduced α diversity, greater dissimilarity, and reduced burden compared with wild-type mice. When these TLR^−/− mice are randomized to cages with wild-type mice, lung microbiota converge with no statistically significant differences in community composition after 3  wk. These changes are driven largely by alterations in features of the lung microbiome in wild-type mice. These data support a role for TLRs in regulating lung microbiota and this may be particularly relevant when mice are cohoused based on underlying genotype. TLR-deficient mice demonstrate persistent community dissimilarity that is not altered significantly by caging, supporting a dysregulated lung microbiome in these animals. However, in wild-type mice, cage environment has a key impact on community structure over time. As lung microbiota correlate with alveolar inflammation and are altered in models of lung disease, our observations have potential implications for the design and implementation of studies of pulmonary immunobiology and microbiota-host interactions within the respiratory tract.

We and others have previously shown that lung microbiota of experimental mice are detectable, distinct from low biomass control samples, and correlate with local measures of pulmonary inflammation (4, 6, 7). In these studies, we have previously shown that lung microbiota in healthy mice are susceptible to caging and will converge with other healthy cohoused mice over a period of 7 days (4). We leveraged these experimental observations to show that TLR^−/− mice may have a distinct lung microbiome. We report similar compositional heterogeneity of lung microbiota within the lung in TLR2^−/−, TLR4^−/−, and TLR5^−/− mice, suggesting that TLR deficiency promotes dysregulated local physiological and immunological homeostasis that is independent of the specific underlying TLR deficiency. Importantly, lung microbiota in these TLR^−/− mice is unchanged in different environments (cohoused vs. randomly housed). However, detectable differences were seen in the composition of lung microbiota in wild-type mice exposed to different environments (cohoused vs. randomly housed). This may reflect a generalized dysregulation of mucosal immunity and host-microbiota interactions in TLR^−/− mice. TLR expression within the respiratory tract is broad and present within both the proximal and distal airway (22). TLR^−/− mice exhibit considerable community dissimilarity sharing few OTUs, and this persists irrespective of environment. In contrast, wild-type mice demonstrate less dissimilarity but are subject to increased community dissimilarity on random housing reflecting possibly a greater susceptibility to the effects of environment. However, it remains unknown whether the lung dysbiosis observed in these TLR^−/− experimental mice is dependent on altered epithelial cell TLR expression and microbiota interactions or promoted by aberrant interactions between macrophage TLRs and respiratory microbiota. Further work is required to understand the key cellular and molecular interactions that occur between epithelial cell and lung microbiota interactions.

A significant limitation to advances in biomedical science has been a lack of reproducibility in often well designed and carefully crafted biomedical experiments (40). The gut microbiota and the effects of environmental caging have previously been implicated as contributors to overall experimental heterogeneity and irreproducibility (41, 42). Animals models are an important tool in the study of plausibility for important biological finings. Although many factors may play a key role in a lack of reproducible experimental results, our data here suggest that lung microbiota dysregulation in transgenic TLR-deficient mice may also play a role in studies of lung disease. Indeed, multiple experiments designed to understand the role of TLRs in chronic lung disease have provided conflicting results. Experiments studying the role of TLR4 in pulmonary fibrosis have reported both a beneficial role for intact TLR4 signaling and a detrimental role (20, 43). It is possible that the lung dysbiosis associated with TLR deficiency may have a significant impact on any outcome measures that include indices of alveolar inflammation. It is also possible that common caging strategies and experimental design which include or fail to cater for the possible impact of caging may also produce irreproducible or heterogeneous results. Further work will be required to determine the exact impact of environmental caging in animal models of inflammatory lung disease.

This study has several notable limitations. We include multiple procedural and isolation controls during the DNA processing and sequencing stages to account for potential contamination, in addition we accurately quantify our bacterial DNA in both experimental murine lung specimens and procedural controls. However, we cannot absolutely exclude all elements of contamination and their effects on our results and the interpretation of the data. Even with methods that cater for all sources of potential contamination, 16S gene sequencing can result in inaccurate results through stochastic effects and kit contamination (44, 45). In this study, the bacterial burden in murine lung specimens was considerably greater than in all isolation and procedural controls when measured using an accurate and sensitive platform (ddPCR). Although the taxa identified in our statistical models are not detectable in isolation and procedural control samples, and have for the most part been reported in previous animal models, Aurantimonas spp and Brevundimonas spp appear less plausible (45). Aurantimonas spp have been previously cultured from human skin, which speaks to its potential as a taxon of importance and may reflect the influence of human handlers on mouse lung microbiota (46). The exact mechanism through which the lung microbiota converges within cohoused animals remains unknown. Although environmental caging has an obvious impact on the compositional heterogeneity of lung microbiota, the effect of caging on wild-type mice may be related to the greater numbers of TLR^−/− mice in these experiments. The key feature is a loss of differentially abundant bacteria in wild-type mice on randomized caging which suggest that these changes were independent of TLRs and grossly dependent on environmental caging. However, in TLR^−/− mice from different vendor facility rooms, and housed in different cages at our facility, there were no significant differences in compositional heterogeneity of lung microbiota supporting a potentially broad role for TLRs in shaping lung microbiota. The changes seen in α diversity remain unexplained. In comparison to cohoused wild-type mice, both cohoused TLR2^−/− and TLR5^−/− mice exhibit reduced diversity. However, on randomized caging, the diversity of lung microbiota in both these TLR deficit mice increases without changes in bacterial burden. In these individual mice, there are changes in α diversity, but the overall effect does not significantly contribute to changes in community composition when comparing groups of mice by genotype. To facilitate the randomized experimental design, we were limited to female mice in this study, and this is an important limitation as sex has reported associations with gut microbiota and will require further work. The impact of sex on lung microbiota in mice is unknown.

In conclusion, we have observed a broadly dysregulated lung microbiome in animal models of impaired TLRs, further validating our work on the potential impact of lung microbiota on pulmonary immunity in experimental mice. Cohousing has a significant impact on lung microbiota and the effect of study design on lung microbiota should be considered in biological models particularly when results demonstrate heterogeneity. Further detailed study of the intricate relationship between lung microbiota and host immunity including work investigating lung microbiota in TLR-deficient mice from varied barrier facilities and its impact on the pathogenesis of lung disease are required.

SUPPLEMENTAL DATA

Supplemental Figs. S1–S19 and Supplemental Tables S1 and S2: http://doi.org/10.5281/zenodo.4739466.

GRANTS

This work was supported by funding provided by NIH National Heart, Lung, and Blood Institute Grants R00HL139996 (to D.N.O’D.), R35HL144481 (to B.B.M.), R01HL121774 (to G.B.H.), and National Institute of Allergy and Infectious Diseases Grant R01AI138348 (to G.B.H.).

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the authors.

AUTHOR CONTRIBUTIONS

G.B.H., J.R.E.-D., B.B.M., and D.N.O’D. conceived and designed research; J.H.L., N.R.F., and D.N.O’D. performed experiments; J.H.L. and D.N.O’D. analyzed data; J.H.L. and D.N.O’D. interpreted results of experiments; D.N.O’D. prepared figures; B.B.M. and D.N.O’D. drafted manuscript; J.H.L., G.B.H., J.R.E.-D., R.P.D., B.B.M., and D.N.O’D. edited and revised manuscript; J.H.L., G.B.H., J.R.E.-D., R.P.D., B.B.M., and D.N.O’D. approved final version of manuscript.