Vitamin D and allergic airway disease shape the murine lung microbiome in a sex-specific manner

All experiments were performed according to the ethical guidelines of the National Health and Medical Research Council of Australia and with approval from the Telethon Kids Institute Animal Ethics Committee (AEC#238). Mice were purchased from the Animal Resources Centre, Western Australia. Mice were maintained under specific-pathogen-free conditions. Female 3 week-old BALB/c mice were fed semi-pure diets, which were either supplemented with vitamin D₃ (2,280 IU vitamin D₃/kg, SF05-34, Specialty Feeds, Perth, Western Australia) or not (0 IU vitamin D₃/kg, SF05-033, Specialty Feeds) as previously described [24, 25]. From 8 weeks of age, female mice were mated with male BALB/c mice for up to 2 weeks. These male mice were fed standard mouse chow until breeding (Specialty Feeds, containing 2,000 IU vitamin D₃/kg). Litter sizes from dams fed either diet were a mean of 5 pups per litter, with equal proportions of females and males. Offspring born were fed the vitamin D-replete or -deficient diets for the rest of the experiment, except in experiments when initially vitamin D-deficient mice were switched to a vitamin D-replete diet at 8 weeks of age (Fig. 1). Experiments were performed over a 12-month period between November 2011 and November 2012.

Serum and BALF 25(OH)D levels were measured using IDS EIA ELISA kits (Immunodiagnostic Systems Ltd, Fountain Hills, AZ) as described by the manufacturer (limit of detection was 5–7 nmol/L).

Following the BALF procedure (see below), the left lobe of each lung was minced using a sterile scalpel blade. A 2 mm³ sample of the lung was snap-frozen in liquid nitrogen. DNA was extracted from lung samples using the DNeasy Blood and Tissue DNA extraction kit as described by the manufacturer (Qiagen). Universal 16S rRNA primers (F primer = 5′-TCCTACGGGAGGCAGCAG T-3′; R primer = 5′-GGACTACCAGGGTATCTAATCCTGTT-3′ [26]) were used to detect bacteria in DNA samples using the Power SYBR Green PCR Master Mix as described by the manufacturer (Applied Biosystems, Calsbad, USA). These ‘universal primers’ have broad specificity to detect conserved regions of 16S rDNA from 34 bacterial species encompassing most bacterial groups [26]. An 18S rRNA endogenous control including a FAM-MGB probe was used as the internal control for this assay using conditions described by the manufacturer (Applied Biosystems).

To profile the microbial community of the lungs, we amplified the variable region V3-V4 of the 16S rRNA gene as previously reported [17], with modifications, using the primer pair of 341 F and 806R. The PCR mix was composed of 5 μl 5x Phusion buffer HF (7.5 mM MgCl₂, Finnzymes, Finland), 0.5 μl 10 mM dNTPs, 1.25 μl 10 μM of each primer, 0.25 μl DNA polymerase (Hotstart Phusion 540 L, 1 unit/μl Finnzymes) and 3.5 μl template. The PCR program started with 98 °C (2 min), followed by 35 cycles of denaturation at 98 °C (5 s), annealing at 55 °C (30 s) and strand elongation at 60 °C (60 s). The PCR was finalised by a single elongation step at 72 °C for 5 min and than cooled to 4 °C. The size of the PCR product was evaluated using gel electrophoresis. The fragment was then excised and purified using the Montage Gel Extraction Kit (Merck Millipore) with negative controls also excised at the position of the expected fragment size. Adaptors were added to the amplicons in a second PCR run under the same conditions as PCR I with a reduced cycle number of 15. The sequences were generated with GS FLX Titanium (454 Life Sciences, Roche). Two sequencing runs were performed, with the machine cleaned between runs (using standard protocols) and the second run then proceeding, with each run containing samples from the ‘naïve’ and ‘OVA-induced allergic airway disease’ datasets. The 1,475,452 reads were trimmed for low quality (minimal quality score = 25) using the Qiime pipeline version 1.5.0 [27]. Only sequences with a minimal length of 200 bp were considered, the sequences were denoised using the Ampliconnoise algorithm [28]. Chimeras were removed using the Uchime algorithm [29]. Operational taxonomic units (OTUs) were picked de novo from quality-checked reads and clustered at 97 % sequence similarity using Uclust. Taxonomy was assigned using the RDP classifier (version 2.2) method and Greengenes as reference database [30]. After data treatment, 280,699 reads from the ‘naïve dataset’, and 330,733 reads from the ‘OVA-allergic airway disease’ dataset, were used for down stream analysis. To adjust for differences in sequencing depth between samples the OTUs were normalised using the cumulative sum-scaling method available in the metagenomeSeq Bioconductor package [31].

A total of 1 ml BALF was collected for each mouse as described previously [32]. Levels of mBD2 in BALF were detected using an ELISA kit and method supplied by USCN Life Science Inc (Wuhan, China).

Ovalbumin (OVA) (Sigma Chemical Company, St Louis, MO, USA) was mixed with an aluminium hydroxide suspension (Alum, Serva, Heidelberg, Germany). This OVA/Alum solution was diluted in 0.9 % saline to sensitise and boost mice intraperitoneally (200 μl) with 1 μg OVA and 0.2 mg Alum as previously described, with mice sensitised at 12 weeks of age and boosted at 14 days of age [24]. For respiratory challenge, mice were placed in a Perspex box and inhaled a 1 % OVA-in-saline (1 mg/ml) aerosol delivered using an ultrasonic nebuliser (UltraNebs, DeVilbiss, Somerset, PA, USA) for 30 min. This was performed once, 7 days after the OVA/Alum boost. Lungs and BALF was obtained 24 h after the aerosol challenge.

BALF cells (5 × 10⁵) were spun onto glass slides using a cytocentrifuge and differential counts of inflammatory cells performed after staining cells with the DIFF-Quik Stain Set 64851 (Lab Aids, Narrabeen, NSW, Australia) as per the manufacturer’s instructions. At least 300 cells were counted for each sample from ≥3 independent fields of view (x100).

Blood was obtained from mice and immediately added to K₂EDTA-coated Microtainer tubes (BD, Franklin Lakes, NJ) to prevent clotting. The proportions of circulating neutrophils and monocytes within the white blood cell population was determined using the ADVIA® 120 Haematology System (Siemens Healthcare Diagnostics Inc, Tarrytown, NY).

Data were compared using one-way ANOVA or an unpaired two-way student’s t test or correlated (Spearman’s rank correlation) using the Prism 5 for Mac OS X statistical analysis program. Differences were considered significant with a p-value <0.05. We used the non-parametric Kruskal-Wallis test (p < 0.05) to test for significant differences in OTU abundance. Treatment effects on the complex microbial communities in the lung were observed by applying the vegdist function (part of the R vegan package) on the raw metagenomeSeq normalised OTU abundance table to generate the Bray-Curtis dissimilarity between samples.

The lungs of specific-pathogen-free mice contain little microbial DNA, and as recently described for microbiome studies [33] are of high risk of contamination by bacterial DNA present in molecular biology reagents used for high throughput sequencing. These can especially contribute towards incorrect interpretations of cluster analyses. However, simply removing OTUs from the abundance table might introduce bias by potentially removing ‘real’ OTUs present in the sample (e.g. Escherichia spp.). Therefore, the entire cluster analysis that was based on the metagenomeSeq normalised OTU table was performed twice; first with, and then without 126 OTUs detected in the blank reagents and listed in the OTU table that contained afterwards 5,635 OTUs for downstream analysis. Most (96 %) of the contaminating OTUs were taxonomically assigned to the halophilic marine genus Halomonas. The rest were Shewanella, Delftia and Stenotrophomonas. Unless stated differently we used the cleaned microbial OTU table describing the lung microbiome according to [34].

The Bray-Curtis dissimilarity test takes microbial diversity and relative abundance into account. Grouping of samples was visualised by using ordination applying non-metric multidimensional scaling (NMDS) generated in the R vegan package [35]. Microbial clustering was further evaluated with the analysis of similarity (ANOSIM), which takes the ranked dissimilarities (Bray-Curtis) of samples belonging to the same treatment group and compares it to the distance of samples of a different treatment [36]. The closer the ANOSIM generated R-value was to 1, the larger the variation between microbial treatments, whereas 0 indicates no difference. The results were than tested for significance by 999 permutations with a 5 % significance level. Individual variation of selected microbes was displayed with the Euclidean distance in a one-sided dendrogram (Heatmap) based on the log-transformed metagenomeSeq normalised OTU counts. The percentage of each OTU was relative to the number of OTUs for each sample.

Circulating serum levels of 25(OH)D were measured in 8-week old offspring born to vitamin D-replete and -deficient dams (Fig. 2a). Serum 25(OH)D levels were <20 or ≥50 nmol.L⁻¹ in offspring fed a vitamin D-deficient or -replete diet, respectively. When vitamin D-deficient offspring were fed a vitamin D-containing diet for four weeks, their serum 25(OH)D increased to levels equivalent to mice fed a vitamin D-containing diet throughout the experiment (Fig. 2a). BALF 25(OH)D levels were low, although greater than the detection limit of the ELISA and not affected by dietary vitamin D (Fig. 2b).

There was a trend similar to our previously published findings (24) for increased bacterial load in the lungs of vitamin D-deficient male mice, although this was not statistically significant (One-way ANOVA, F = 1.481, p = 0.245; Fig. 3). Bacterial load was quantified using a PCR with 16S rRNA primers capable of amplifying DNA of most bacterial species.

We used high-throughput sequencing to determine how vitamin D might affect the composition of bacteria in the lung microbiome. After data treatment (see methods) there was an average read distribution of 4,803 sequences per naïve animal. We first evaluated if dietary vitamin D had a significant impact on the microbial composition enumerating the dissimilarity of the OTUs. As shown in NMDS plots, there was no clustering by diet (Fig. 4a), sex (Fig. 4b), or the combination of both (Fig. 4c). We then evaluated the relative abundance of OTUs with 23–112 OTUs identified/mouse (Fig. 5). Neither dietary vitamin D nor sex significantly affected bacterial diversity in the lungs as defined by the relative number (Fig. 5, Kruskal-Wallis Test, p > 0.05) of OTUs. In Fig. 6, a heat map was used to show the most frequent OTUs identified from the lungs of male mice. There was very little effect of dietary vitamin D on the relative proportion of each OTU (Fig. 6), with similar findings in female mice (data not shown), and little difference in bacterial variation between treatments identified using ANOSIM analysis, except in mice fed a vitamin D-containing diet throughout the experiment (Table 1).

We observed a significant separation (NMDS cluster) between female and male mice that were fed the vitamin D-supplemented diet throughout the experiment (Fig. 7). This difference was attributed to increased representation of an Acinetobacter OTU in female mice, which varied significantly between the mice (Wilcoxon Rank Sum test, p = 0.02), together with a single low-abundant OTU that could only be annotated as bacteria. Reduced carriage of the Acinetobacter OTU was observed in male mice (2/10 mice; ≤5 % of sequences in the observed sequences; mean = 0.7 %), compared to female mice (7/10 mice; ≤40 % of sequences in the observed sequences; mean = 7.0 %) fed a vitamin D-containing diet throughout the experiment.

In both male and female mice, dietary vitamin D increased ß-defensin-2 (mBD2) protein levels in BALF, with the effects of early-life vitamin D-deficiency reversed by dietary vitamin D (Fig. 8a). There was no effect of dietary vitamin D on serum mBD2 levels (Fig. 8b). Levels of mBD2 were significantly greater in serum than BALF. These results suggest that the sex-dependent effects of vitamin D on Acinetobacter in the lungs were not due to differences in the expression of mBD2 as dietary vitamin D increased levels to a similar degree in the BALF of male and female mice.

In a separate analysis, we identified an inverse correlation between serum 25(OH)D levels and the total number of OTUs (Fig. 9, Spearman’s rho = −0.592, p = 0.001). When each sex was considered separately, some evidence for an inverse correlation was observed between total OTUs in the lungs and serum 25(OH)D levels in females (Spearman’s rho = −0.642, p = 0.052) but not males (Spearman’s rho = 0.463, p = 0.238). When examining individual OTUs in the lungs, there was a strong negative correlation between 25(OH)D levels in serum and an unspecified Pseudomonas OTU (Spearman’s rho = −0.556, p = 0.016), with a marginal evidence observed in females (Spearman’s rho = −0.611, p = 0.081) but not males (Spearman’s rho = −0.343, p = 0.366). Some evidence for a negative correlation was observed between serum 25(OH)D and all Pseudomonas OTUs (Spearman’s rho = −0.429, p = 0.07). However, there was no evidence of a relationship between total OTUs or the unspecified Pseudomonas OTUs in the lung and mBD2 levels in BALF (total OTUs; Spearman’s rho = 0.109, p = 0.429; Pseudomonas OTU Spearman’s rho = 0.202, p = 0.138 for BALF) or serum (total OTUs; Spearman’s rho = 0.114, p = 0.673, Pseudomonas OTU Spearman’s rho = 0.114, p = 0.673).

We hypothesised that allergic sensitisation and challenge with OVA (to induce allergic airway disease) would significantly change the microbiota composition of the lungs. The relative effects of the induction of OVA-induced allergic airway versus that of vitamin D deficiency (alone) on the lung microbiome are unknown. To better understand the magnitude of the observed effect of dietary vitamin D on the lung microbiome, we compared its capacity to regulate the lung microbiome to the effects of the induction of allergic airway disease. Mice were intraperitoneally sensitised and boosted with ovalbumin (OVA) (1 μg) and Alum hydroxide suspension (Alum, adjuvant, 0.2 mg) at 12 and 14 weeks of age (respectively), which was followed by a respiratory challenge at 15 weeks of age with aerosolised OVA-in-saline (1 mg/ml) [24]. Lungs were sampled 24 h after the challenge. We compared the overall microbial community from the lungs of the naïve mice characterised above (n = 58; 4,803 sequences per naïve mouse) with the lungs of mice with OVA-induced allergic airway disease (n = 68; 4,469 sequences per mouse). As shown in Fig. 10a, the lung samples of naïve mice and mice with allergic airway disease clustered significantly apart, which was confirmed using a compositional analysis to compare the difference in numbers of observed bacterial OTUs (Bray-Curtis, R = 0.334, p = 0.001). A NMDS analysis confirmed that observed differences were not induced by performing two sequencing runs (data not shown). Bacterial diversity was not affected with the mean number of OTUs for naïve mice or mice with allergic airway disease (naïve, 138 + 5.1; allergic airway disease, 128 + 5.4; mean + SEM, p = 0.132), respectively.

The strength of the difference between the lung microbiomes of naïve mice, and, mice with allergic airway disease was quantified by ANOSIM. A significant difference was identified (Table 2). As for naïve mice (Fig. 7), a significant separation (NMDS cluster) was observed between female and male mice that were fed the vitamin D-supplemented diet throughout the experiment with OVA-induced allergic airway disease (Fig. 10b; Table 3 for ANOSIM bacterial variance). There was a significant effect of allergic airway disease on the relative proportion of OTUs detected in mice. The most abundant OTUs detected, which each accounted for >1 % of the sequences after normalization, are shown for the naïve mice, or, mice with allergic airway disease in Fig. 11a. Proportions of the environmental microbial group GNO2 (Naïve, mean = 9.5 %; OVA, mean = 0.0 %) and an unassigned bacterial OTU (Naïve, mean = 1.1 %; OVA, mean = 0.0 %) were reduced in mice with OVA-induced allergic airway disease, while Acinetobacter (Naïve, mean = 1.29 %; OVA, mean = 1.8 %), Chloroplast (Naïve, mean = 0.1 %, OVA, mean = 5.0), Comamonadaceae (Naïve, mean = 0.1 %; OVA, mean = 4.0.%), Cyanobacteria (Naive, mean = 0.1 %; OVA, mean = 1.0 %) Staphylococcus (Naïve, mean = 0.5 %; OVA, mean = 1.2 %), Micrococcaceae (Naïve, mean = 0.1 %; OVA, mean = 1.8 %) and Verrucomicrobiaceae (Naïve, mean = 0.1 %; OVA, mean = 7.6 %) OTUs, and an OTU assigned to Peptostreptococcaceae (Naïve, mean = 1.0 %; OVA, mean = 1.5 %) were increased by the induction of allergic airway disease. The Acinetobacter OTU significantly increased in the OVA samples was not the same Acinetobacter OTU identified as responsible for largely driving the difference between male and female mice fed a vitamin D-containing diet throughout the experiment. There was no significant difference in the number of any Pseudomonas OTUs detected in the lungs of naïve mice, or, mice with allergic airway disease. Together, these results suggest that the effects of allergic airway disease on the lung microbiome of mice far exceeded those of vitamin D deficiency.

Even though OVA-induced allergic airway disease substantially altered the lung microbiome, the sex-specific effect on the abundance of Acinetobacter OTUs was still apparent in mice fed exclusively a vitamin D-containing diet (Table 3). Acinetobacter OTUs were more frequent in OVA-induced allergic airway disease than naïve mice (Fig. 11b). There were a further 11 OTUs varied in their frequency when the lung microbiomes of female and male mice with OVA-induced allergic airway disease were compared. However, these were present at a very low frequency (<0.01 % of the total normalised sequences). There was no effect of OVA-induced allergic airway disease on the proportions of the dominant Acinetobacter OTU in female mice; however, induction of OVA-induced allergic airway disease increased the proportions of this OTU in male mice. For example, for male mice fed the vitamin D-containing diet throughout the experiment, the percentage of this OTU increased from 0.69 + 0.51 (mean + SEM) in naïve mice to 4.7 + 1.5 in mice with OVA-induced allergic airway disease (*p < 0.05, n = 10–13 mice/treatment). There was no significant correlation between the number of Acinetobacter OTUs and 25(OH)D or mBD2 levels in BALF or serum (Spearman p > 0.05, data not shown).

The effects of vitamin D deficiency and the induction of OVA-induced allergic airway disease on numbers of neutrophils and macrophages/monocytes in the BALF or blood were then compared. Substantially increased BALF neutrophil numbers were observed in both male (Fig. 12a) and female (Fig. 12e) mice with OVA-induced allergic airway disease as expected; however, there was only a marginal effect of vitamin D deficiency. There was also a significant increase in neutrophil numbers in the blood of female mice with OVA-induced allergic airway disease, with no effect of vitamin D deficiency (Fig. 12f). Neither treatment significantly modulated macrophage levels in BALF (Fig. 12b, f), nor monocytes levels in blood (Fig. 12d, h). These data suggest that there is an important interplay between respiratory inflammation and the lung microbiome.