Microbial contribution to metabolic niche formation varies across the respiratory tract

SUMMARY

Variations in the airway microbiome are associated with inflammatory responses in the lung and pulmonary disease outcomes. Regional changes in microbiome composition could have spatial effects on the metabolic environment, contributing to differences in the host response. Here, we profiled the respiratory microbiome (metagenome/metatranscriptome) and metabolome of a patient cohort, uncovering topographical differences in microbial function, which were further delineated using isotope probing in mice. In humans, the functional activity of taxa varied across the respiratory tract and correlated with immunomodulatory metabolites such as glutamic acid/glutamate and methionine. Common oral commensals, such as Prevotella, Streptococcus, and Veillonella, were more functionally active in the lower airways. Inoculating mice with these commensals led to regional increases in several metabolites, notably methionine and tyrosine. Isotope labeling validated the contribution of Prevotella melaninogenica in generating specific metabolites. This functional characterization of microbial communities reveals topographical changes in the lung metabolome and potential impacts on host responses.

In brief

Wong et al. show that microbial activity varies along the respiratory tract, influencing metabolite concentrations and contributing to niche construction in the lower airways. The lung microbiome directly influences metabolic pathways with immunomodulatory effects, highlighting the importance of harnessing the complexity of microbial functions in the lower airways.

Graphical Abstract

INTRODUCTION

The airway microbiome has been associated with contributing to the inflammatory tone of the lower airways^1–3 as well as to pathogenesis and disease progression for several pulmonary diseases.^4–7 Alteration in the microbial community composition may trigger a host response, which, in turn, may lead to an up- or downregulation of inflammatory pathways. In particular, micro-aspiration of common oral commensals, such as Prevotella, Streptococcus, and Veillonella, have been linked to an increase in inflammatory tone of the lower airway, with upregulation of Th-17 pathways and expression of inflammatory cytokines.¹ Although 16S rRNA gene sequencing can be used to identify the presence of bacteria and predict bacterial function, microbial activity can be better characterized with newer technologies, such as whole-genome shotgun (WGS) metagenomic and RNA metatranscriptomic sequencing, as we have previously demonstrated.⁸ We have shown that functionally active microbes correlate with byproducts of microbial metabolism such as short-chain fatty acids (SCFAs).⁸ Importantly, these can directly activate host immune responses, both in an anti- and pro-inflammatory manner.^4,9

Microbial-host interactions are partially mediated by metabolites that can modulate signaling pathways and immune responses.¹⁰ Indeed, metabolites are being investigated as potential biomarkers for disease endotypes and disease progression.¹¹ For example, glutathione and methionine, both associated with airway oxidative stress, have been found to be elevated in sputum samples from patients with chronic obstructive pulmonary disease (COPD).^11,12 These changes in the metabolic environment of the respiratory tract could play a critical role in priming immunological response locally.^13,14 The potential link between metabolites and lung microbiota was further explored in a recent SPIROMICS study, assessing their combined roles in the clinical characteristics of patients with mild COPD.¹⁵ Notably, increased abundances of some Prevotella taxa in the lower airways are correlated with adenosine and polyamines, which are linked to a slower decline in lung function and fewer symptoms in COPD. On the other hand, in a murine model, airway colonization of Staphylococcus aureus correlated with levels of homocysteine in the lower airways, which led to neutrophilic immune responses, precipitating a decline in lung function.⁷ Varying microbial composition in the lung likely leads to differences in substrate utilization and subsequent production of diverse classes of metabolites, such as amino acids and SCFAs. These data suggest that the lower airway microbiome contributes to its metabolic environment. Here, we extend these findings by showing that the lower airway metabolic environment is associated with the genomic function of the lower airway microbiome, characterized by metagenomics and metatranscriptomics of bronchoscopy samples from healthy volunteers with a significant smoking history. With difficulties in isolating the effect of microbe functions and their direct contribution to the airway metabolic profiles, we validated our observations by demonstrating the mechanistic underpinnings in a pre-clinical model. We demonstrate that changes in the microbial composition directly affect the levels of metabolites in the lower airways, highlighting the contribution of the lower airway microbiome to metabolites, many with clear immunomodulatory effects.

RESULTS

Patient cohort

For this investigation, we utilized data generated from 56 subjects with a significant smoking history (>10 pack years), and no significant past medical history, in whom we performed elective research bronchoscopy and profiled the metabolome and microbiome in the upper and lower airways. Before bronchoscopy, all subjects underwent a screening visit and were questioned about respiratory symptoms. None of the subjects recruited for this study complained of new respiratory symptoms, including wheezing, dyspnea, and cough. Furthermore, no abnormalities were identified at bronchoscopy, with a normal airway inspection across all recruits. Most patients were male (41 out of 56) and current smokers (51 out of 56) at the time of bronchoscopy. Therefore, we did not control for smoking status in our analysis. Table 1 lists subject demographics and clinical characteristics.

Table 1.

Demographics of the subjects

Clinical data	Total (N = 56)
Age median [Q1, Q3]	56 [51, 59.25]
Gender
Female	15 (26.8)
Male	41 (73.2)
BMI median [Q1, Q3]	27.3 [23.15, 31.3]
Ethnicity
Asian	2 (3.6)
Black or African American	34 (60.7)
Caucasian	15 (26.8)
Hispanic	5 (8.9)
Pack_years
Median [Q1, Q3]	21.5 [15, 35]
FEV1_post
Median [Q1, Q3]	2.75 [2.39, 3.18]
Missing	1 (1.8)
FEV1_post_perc
Median [Q1, Q3]	82 [71.5, 94.5]
Missing	1 (1.8)
Comorbidities
HTN	11 (19.6)
COPD	26 (46.4)

Values presented in median intraquartile range (IQR) or N (proportion).

To focus on lower airway changes and to account for the low biomass nature of the lower airways, we utilized 60 bronchoalveolar lavage (BAL) samples, ten upper airway (UA) samples, and ten background (BKG) samples for metagenome (MG) and metatranscriptome (MT) analyses. Among these, ten BAL samples are paired with UA samples and BKG samples. Due to processing issues with two of the BAL and UA samples, the metabolome analysis contained only 58 BAL samples, 8 UA samples, and 10 BKG samples. Given the small number of paired samples (n = 10), we have conducted unpaired analyses.

Topographical differences of the mucosal metabolome

To determine whether metabolite profiles were different in the lower airways and UAs, we performed a liquid chromatography-mass spectrometry (LC-MS) global metabolomics on all samples, including BAL, UA, and BKG samples, using polar and hydrophobic approaches in parallel. In total, we identified 265 metabolites with Kyto Encyclopedia of Genes and Genomes (KEGG) pathway annotations. The most abundant metabolites in BAL included betaine, monoethylglycylxylidide, choline, niacinamide, 5-aminovaleric acid, creatinine, phosphoric acid, and glutamine (Figure S1). By intensity, the top metabolites in UA included vanillin, 5-aminovaleric acid, and phosphoric acid, whereas the top metabolites in BKG samples were phosphoric acid and sulfuric acid. Among these different sample types, there were significant differences in overall metabolomic profiles, based on Bray-Curtis dissimilarity (Figure 1A, permutational multivariate analysis of variance [PERMANOVA] p < 0.01). Top differentially enriched metabolites in the lower airways, identified with partial least squares discriminant analysis (PLS-DA), included myo-inositol, niacinamide, choline, betaine, and glycine but also included adenosine, glutamine, glutamic acid/glutamate, and homocysteine (Figure 1B; Table S1).

Figure 1. Topographical evaluation of metabolomes and functional microbiomes.

(A) Principal-component analysis (PCA) of metabolomic data comparing sample types, bronchoalveolar lavage (BAL, blue), upper airway (UA, yellow), and environmental background (BKG, gray) samples with permutational multivariate analysis of variance (PERMANOVA) p value.

(B) Partial least squares discriminant analysis (PLS-DA) was used to identify metabolites enriched in BAL (blue) and UA (yellow), displaying component 1 of PLS-DA analysis on the x axis.

(C) Differentially enriched functional metagenomic (MG) data based on fold change versus log₁₀ adjusted p value (false discovery rate [FDR]) using edgeR comparing BAL and UA. Pathways with a positive log fold change are enriched in BAL and those with a negative fold change are enriched in UA. Only pathways with an FDR < 0.2 are displayed. Bubble size is based on relative abundance.

(D) Differentially enriched functional metatranscriptomic (MT) data based on fold change versus FDR using edgeR comparing BAL and UA. Pathways with a positive log fold change are enriched in BAL and those with a negative fold change are enriched in UA. Only pathways with an FDR < 0.2 are displayed. Bubble size is based on relative abundance.

(E) Relative abundance of top lower airway microbial pathways in metatranscriptome data for BAL and UA.

(F) Microbial functions significantly upregulated in lower airways samples represented as a delta of MT to MG by Wilcoxon rank sum test, p < 0.05. Data is represented as box-whisker plot with interquartile range.

(G) Relative abundance of top upper airway microbial pathways in metatranscriptome data for BAL and UA.

(H) Microbial functions significantly upregulated in upper airways samples represented as a delta of MT to MG by Wilcoxon rank sum test, p < 0.05. Data is represented as box-whisker plot with interquartlie range.

Topographical analysis of the functional genomic profile of the microbiome

Having identified differences in the metabolite profiles of the lower airways and UAs, we evaluated differences in the functional profile of the microbiome using MG and MT sequencing data. Microbial gene annotation of the metagenomic data was used to show gene composition differences by α-diversity and β-diversity across sample types (Figures S2A and S2B). Differential analysis using edgeR identified several metagenomic pathways that were enriched in BAL samples. The top pathways include flavone and flavanol biosynthesis and fatty acid elongation (Figure 1C), as well as pathways associated with tryptophan metabolism (Table S1).

For the MT, there were no statistically significant differences in α- or β-diversity metrics (Figures S3A and S3B). In the differential analysis of metatranscriptome pathways, the most enriched pathways in BAL samples, such as biosynthesis of enediyne antibiotics, fatty acid elongation, linoleic acid metabolism, and biosynthesis of sesquiterpenoid and triterpenoid, were also confirmed to be significantly enriched in the corresponding MG. On the other hand, tyrosine metabolism was noted to be enriched in the UA samples for both MG and metatranscriptome (Table S1). Similarly, the modules associated with ornithine and glutamate biosynthesis were enriched in both datasets of BAL samples (Figure 1D; Table S1). Looking specifically at KEGG orthologs (KOs), some KOs associated with methionine (KO03825, KO08351, KO14287, and KO24042) and glutamate (KO10004, KO21949, KO05826, and KO24034) were also found to be enriched in both datasets of the lower airways (Table S1).

We next compared the potential microbial function (as measured by metagenomics) to active microbial function (as measured by metatranscriptomics) and calculated the difference in read counts between the MT and MG datasets for individual pathways (hereafter referred as MT-MG). With this analysis, we identified several pathways highly transcribed in BAL samples (as compared with UA samples), including fructose, mannose, and galactose metabolism, as well as valine, leucine, and isoleucine biosynthesis (Figures 1E–1H).

To better understand the relationship between metabolites and microbial function, we performed HeatWave analysis, a network-based visualization and propagation analysis that integrates transcriptomic pathway and reaction data with metabolomic data (Figure 2). This network-based differential analysis on microbial gene expression and metabolite levels across the upper and lower airways is projected onto a reaction network generated from the KEGG database. From our analysis, we used 164 measured metabolites, with a median absolute fold change of 2.69, and 2,376 measured orthologs, with a median absolute fold change of 2.19. The final network presents 262 nodes, 105 metabolites, 587 orthologs, and 799 edges. The HeatWave algorithm is agnostic to direction of change. It takes the absolute value of the metabolite (or gene) fold change to identify components of the network that are changing, regardless of whether the change is up or down. This approach allows for focusing on areas of significant activity within the network.

Figure 2. Multi-omic analysis.

(A) Schematic of data used for network HeatWave analysis.

(B) HeatWave analysis based on differential expressions for metabolomic and microbial functional metatranscriptomic datasets for human UA and BAL samples. Highlighted nodes include several metabolite nodes with microbial function correlations, such as inosine, adenosine, glutamic acid/glutamate, homocysteine, niacinamide, and tyrosine. Color gradient of light blue (extreme downregulation) to light red (extreme upregulation), with a limit of 8-fold change or more in either direction. Circles represent metabolites whereas round cornered squares represent enzymes (orthologs). An interactive version of the main network is available in the supplement (Data S1).

Several of the metabolites, such as glutamate, niacinamide, and tyrosine, with potential immunomodulatory effects^15–17 were highlighted to showcase their biochemical linkages to microbial genes. Many of these genes are involved in pathways identified in our differential analysis of MT pathways. For example, glutamic acid/glutamate, found enriched in the lower airways, is linked to genes, such as GCLC and GCLM, involved in biosynthesis of glutathione (map00480), a metabolite that also displays increased levels in the lower airways. Additionally, niacinamide metabolites identified as enriched in the lower airways were shown to be linked to microbial genes associated with KEGG pathways also enriched in the lower airways. Specifically, niacinamide is linked to pyrazinamidase/nicotinamidase 1 (PNC1), which catalyzes the conversion of nicotinamide to nicotinamide adenine dinucleotide (NAD), an agent involved in oxidation and reduction reactions necessary for biosynthesis of alkaloids (map00996) and elongation of fatty acids (map00062). Interestingly, tyrosine levels were generally higher in the UAs and significantly correlated with several genes (yhdR, GOT2, and aspB) involved in the enriched tyrosine metabolism pathway (map00350), as revealed by MG and MT differential analyses.

Topographical analysis of the taxonomic microbiome

Having identified differences in microbial community function comparing the BAL and UA, we next evaluated functional differences within different taxa by sample type. Consistent with our previous work,⁸ there were clear differences in taxonomic annotation by sample type in the MG (Figure S4A) and the MT (Figure S4B). In a taxa-specific MT-MG analysis, several taxa, such as Streptococcus pneumoniae, Bacillus cereus, and Streptococcus salivarius, were found to be more functionally active, with higher MT read counts compared with MG, in BAL when compared with UA (Figure 3). Interestingly, some taxa, like common oral commensals such as Veillonella atypica and Prevotella jejuni, were present at lower relative abundance in the BAL compared with the UA but had higher MT read counts compared with MG, suggesting that they had greater functional activity in the lower airways.

Figure 3. Topographical evaluation of the microbial taxonomy.

(A) Relative abundance of top lower airway taxa in metatranscriptome (MT) data for BAL (blue) and UA (yellow).

(B) Taxa significantly upregulated in lower airways samples represented as a delta of MT to metagenome (MG) by Wilcoxon rank sum test, p < 0.05. Data is represented as a box-whisker plot with interquartlie range.

(C) Relative abundance of top upper airway taxa in metatranscriptome data for BAL and UA.

(D) Taxa significantly upregulated in upper airways samples represented as a delta of MT to MG by Wilcoxon rank sum test, p < 0.05. Data is represented as a box-whisker plot with interquartlie range.

To identify which microbes could be contributing to the production of metabolites enriched in the lower airway, we further analyzed both MG and MT functional datasets. We focused on taxa possessing genes under KOs involved in metabolite production. In the MG analysis, Streptococcus, Veillonella, and Prevotella emerged as the top contributing genera for KOs associated with niacinamide, glutamic acid/glutamate, tyrosine, adenosine, inosine, and methionine, in both the upper and lower airways. In the MT analysis, Streptococcus and Veillonella remain top contributors for KOs related to niacinamide, methionine, adenosine, inosine, and glutamic acid/glutamate in both airway regions. Notably, Streptococcus was the leading genus in the lower airway, accounting for more than 40% of the contribution for those metabolites. Although Streptococcus, Veillonella, and Prevotella are all common taxa found in the oral cavity, their significant contribution in functional orthologs involved in the production of these metabolites in the lower airway suggests that they play a crucial role in the enrichment of metabolites in the region (Figure S5).

Functionally active microbes in the lower airways lead to a change in the airway metabolome

Having shown an association between microbial function and metabolite levels in the lower airway environment of human subjects as well as increased microbial function of some taxa, we sought to experimentally test the effects of functionally active bacteria on the level of metabolites in the lungs in a preclinical mouse model. Given that Prevotella, Streptococcus, and Veillonella were among the top contributors to some of the metabolites enriched in the lower airway of our human subjects, and considering their increased functional activity in this region, we cultured the three taxa commonly found in the upper and lower airways of humans, Prevotella melaninogenica, Streptococcus mitis, and Veillonella parvula. Good recovery of the functional genomic data for these taxa, both in terms of MG and MT, was observed, with expected differences based on Bray-Curtis dissimilarity (Figures S6A–S6E). Mice were then exposed to a mixture of these oral commensals (MOC) inoculated into the oral cavity (MOC_Oral) or into the lung (MOC_Lung), with some mice receiving phosphate-buffered saline (PBS) as a negative control. At sacrifice, tongue and lung samples were examined by metagenomics, metatranscriptomics, and metabolomics (Figure 4A).

Figure 4. Pre-clinical mouse model.

(A) Murine experimental model and exposures.

(B) In vivo relative abundance in the metagenomic dataset of P. melaninogenica (orange), S. mitis (green), and V. parvula (purple) in tongue and lung samples for the three different inoculations, including PBS to the lung (PBS_Lung), MOC to the oral cavity (MOC_Oral), and MOC to the lung (MOC_Lung). Comparisons and p values based on Wilcoxon rank sum test. Only statistically significant comparisons are shown. Data is represented as a box-whisker plot with interquartile range.

(C) In vivo relative abundance in the metatranscriptomic dataset of P. melaninogenica, S. mitis, and V. parvula in tongue and lung samples for the three different inoculations, including PBS_Lung, MOC_Oral, and MOC_Lung. Comparisons and p values based on Wilcoxon rank sum test. Only statistically significant comparisons are shown. Data is represented as a box-whisker plot with interquartlie range.

(D) Differentially enriched functional metagenomic data from murine lung samples based on fold change versus FDR using edgeR comparing two different exposures. Pathways with a positive log fold change are enriched in mice inoculated with MOC_Lung (blue) and those with a negative fold change are mice inoculated with PBS_Lung (gray). Only pathways with an FDR < 0.2 are displayed. Bubble size is based on relative abundance. Pathways not associated with the three MOC taxa (P. melaninogenica, S. mitis, and V. parvula) were italicized and colored gray.

(E) Differentially enriched functional metatranscriptomic data from murine lung samples based on fold change versus FDR using edgeR comparing two different exposures. Pathways with a positive log fold change are enriched in mice inoculated with MOC_Lung and those with a negative fold change are mice inoculated with PBS_Lung. Only pathways with an FDR < 0.2 are displayed. Bubble size is based on relative abundance. Pathways not associated with the three MOC taxa were italicized and colored gray.

We first evaluated the relative abundance of the three MOC taxa in the metagenomic and metatranscriptomic data (Figures 4B and 4C). Interestingly, for the mice exposed to MOC in the UA (MOC_Oral), we could not identify a significant difference in the abundance of these taxa in either the MG or MT data when compared with the PBS group. The data suggest that these oral commensals failed to seed the UA of mice or that the level of seeding was below the ability to detect their genomic material. In contrast, for the mice where MOC was exposed to the lung, we identified a much higher relative abundance of three taxa in the lungs in both the MG and MT datasets. These data support that there was efficient seeding of these three microbes to the lower airways, at least during the time frame evaluated here (2 h post inoculation), with evidence of persistent transcripts from the introduced human oral commensals (Figures 4B and 4C).

Having identified these findings in lung samples that were inoculated, we assessed how the functional annotation of the microbial genomic data changed in the lung (MOC_Lung) as compared with PBS. In the MG data, several functional pathways were enriched in the MOC_Lung samples. The top pathways included zeatin biosynthesis, acarbose and validamycin biosynthesis, and polyketide sugar unit biosynthesis (Figure 4D). Other pathways of interest, associated with metabolites that were enriched in human lower airway samples that were significantly increased in MOC_Lung samples, included glutamate metabolism, glycolysis, and gluconeogenesis, as well as cysteine and methionine metabolism (Table S2). In the MT dataset, we identified several pathways that were enriched in MOC_Lung; the top pathways included nitrotoluene degradation, indole alkaloid biosynthesis, and atrazine degradation (Figure 4E), as well as glutathione metabolism, important in the production of glutamic acid/glutamate (Table S2). Importantly, there was some overlap between the MG and MT datasets for pathways enriched in MOC_Lung samples. These include nicotinate and nicotinamide metabolism (associated with niacinamide), ferroptosis (associated with iron utilization), and folate biosynthesis and pyrimidine metabolism (associated with methionine and homocysteine through the one-carbon metabolism pathway¹⁸) (Table S2). Metabolic pathways not associated with the three MOC taxa were italicized in Figures 4D and 4E to distinguish them from the known metabolic pathways of these taxa.

Having identified differences in microbial function, after inoculation of MOC in a mouse model, we investigated the changes in the metabolic micro-environment. In total, there were 211 KEGG-annotated metabolites identified by our LC-MS approach. By PLS-DA, we identified numerous metabolites, including uridine, methionine, and tyrosine, as enriched in MOC_Lung samples compared with PBS (Figure 5A; Table S2). Some of these metabolites, such as niacinamide, glutamic acid/glutamate, adenosine, and inosine, were similarly found to be enriched in the lower airways of human subjects.

Figure 5. Metabolomic features in an in vivo mouse model of airway dysbiosis.

(A) PLS-DA was used to identify metabolites, known to be produced by the three MOC taxa, enriched in murine lung samples inoculated with MOC_Lung (blue) and PBS_Lung (gray).

(B) Taxonomic contribution of P. melaninogenica (orange), S. mitis (green), and V. parvula (purple) in the metagenomic dataset for niacinamide, methionine, inosine, adenosine 5-monophosphate, glutamic acid/glutamate, adenosine, and tyrosine.

(C) Taxonomic contribution of P. melaninogenica,S. mitis, and V. parvula in the metatranscriptomic dataset for niacinamide, methionine, inosine, adenosine 5-monophosphate, glutamic acid/glutamate, adenosine, and tyrosine.

We further examined the MG and MT datasets to identify which organisms could be contributing to these metabolites enriched in the MOC_Lung. For the MG dataset, Prevotella melaninogenica, Veillonella parvula, and Streptococcus mitis were the predominant contributors to KOs related to niacinamide, glutamic acid/glutamate, tyrosine, adenosine, inosine, and methionine in MOC_Lung samples, although none of these taxa were identified as contributors in the PBS MG (Figure 5B). For many of these metabolites, Prevotella melaninogenica seemed to be the top contributor (>20% contribution based on relative abundance of metagenomic reads to KOs related to those metabolites). Interestingly, in the MT dataset, we also identified transcripts from Prevotella melaninogenica, Veillonella parvula, and Streptococcus mitis annotated to KOs related to niacinamide, glutamic acid/glutamate, tyrosine, adenosine, inosine, and methionine (Figure 5C). However, the distribution was different, with Veillonella parvula contributing at the highest level, particularly for KOs related to methionine. Together, these data suggest that common human oral commensals introduced to the lower airways continue to actively transcribe genes associated with metabolic pathways, along with their metabolic byproducts. It is, however, possible that some of the metabolic changes are directly from the host.

Isotopically labeled metabolites in an ex vivo and pre-clinical model

To further confirm that the identified metabolic signatures were produced by the bacteria introduced into the lower airways, we labeled bacterial metabolites by growing Prevotella melaninogenica (P. melaninogenica) in an isotopically labeled carbon-13 (¹³C) glucose-enriched media to serve as a metabolic substrate (carbon source) for metabolite production. Importantly, P. melaninogenica was selected for this experiment as we were able to confirm appropriate growth curves in a ¹³C glucose-enriched media. Using LC-MS, we measured metabolites present in the culture media enriched with ¹³C glucose P. melaninogenica as compared with those found in culture media without ¹³C glucose (¹²C) enrichment (Figure 6A). These data showed that many metabolites found in the ¹³C glucose media where P. melaninogenica grew were ¹³C labeled (Table S3). For example, there was a significant increase in the relative intensity of ¹³C-labeled fructose 1,6-bisphosphonate and adenosine compared with ¹²C culture media (Figure 6B). In addition, similar significant increases were seen with inosine and glutamic acid/glutamate but, surprisingly, not with methionine.

Figure 6. ¹³C Isotope glucose labeling to identify microbial contribution to measured metabolites.

(A) Experimental design for ex vivo ¹³C isotope glucose labeling model.

(B) Ex vivo comparison of C13 labeling intensity for several metabolites, including fructose 1,6-bisphosonate, adenosine, inosine, glutamic acid/glutamate, and methionine. *indicates a significant (p < 0.05) difference between C12 and C13 experiments for each metabolite based on Wilcoxon signed ranged test. Data is represented as a barplot with an error bar.

(C) Experimental design for in vivo model.

(D–F) (D) PCA of in vivo murine metabolomic data comparing mice inoculated with (C13) and without (C12) a C13-labeled glucose P. melaninogenica, sacrificed at 1 and 6 h, with PERMANOVA p value. Comparison of C13-labeled percentage adenosine, inosine, glutamic acid/glutamate, and methionine in the C13 (turquoise) and C12 (green border) samples at (E) 1 h and (F) 6 h. * indicates a significant (p < 0.05) difference between C12 and C13 experiment for each metabolite based on Wilcoxon signed ranged test. Data is represented as a barplot with an error bar.

(G) Mass spectrometry analysis of three metabolites, adenosine, inosine, and methionine, showing intensity and weight in the first panel followed by time and intensity curves for each peak identified.

Next, we used an in vivo model where mice were inoculated with P. melaninogenica grown in a ¹³C-glucose-enriched media or with ¹²C glucose-enriched media. We expected that metabolites produced by P. melaninogenica in the lung would carry the ¹³C label and measured these using LC-MS at 1 and 6 h post inoculation (Figure 6C). Comparison of overall metabolic composition of the lung by Bray-Curtis dissimilarity-based beta diversity analyses showed that there were significant differences between lung samples exposed to P. melaninogenica grown in the ¹³C-glucose-enriched media as compared with ¹²C controls at 1 and 6 h (Figure 6D). At 1 h after P. melaninogenica inoculation, there was a significantly higher relative intensity of ¹³C-labeled adenosine, inosine, and glutamic acid/glutamate in the lungs of mice exposed to the P. melaninogenica grown in the ¹³C-glucose-enriched media (Figure 6E; Table S3). After 6 h post P. melaninogenica inoculation we continued to observe a significant increase in the relative intensity of ¹³C-labeled adenosine, inosine, and glutamic acid/glutamate, as well as ¹³C-labeled methionine in the lung samples of mice exposed to P. melaninogenica grown in the ¹³C-glucose-enriched media as compared with ¹²C controls (Figure 6F; Table S3). We then used spectrometric analysis to identify the mass shift that corresponds to the incorporation of ¹³C into a given metabolite to confirm identity (Figure 6G). For example, among the lung samples inoculated with P. melaninogenica grown in the ¹³C-glucose-enriched media, there was an increase in relative intensity of adenosine at M + 5, corroborating the isotopically labeled adenosine with five ¹³C atoms. The M + 0 peak has the same retention time as the M + 5 peak, confirming the identity of the metabolite as adenosine with different isotopic compositions. Similar verifications on the isotopic-labeled metabolites were seen for inosine and methionine. We were not able to confirm this with glutamic acid/glutamate. These data confirm that active microbial metabolism from the inoculated P. melaninogenica contributes to the levels of different metabolites in the lungs of these mice.

In parallel, we examined the blood samples of these mice. Compared with lung samples, there were significantly fewer metabolites with traceable microbial contribution with ¹³C labeling. The lung samples revealed 211 unique metabolites with potential traceable ¹³C labeling from the microbial source at 1 h, and 252 metabolites at 6 h, whereas blood samples had only 92 metabolites at 1 h and 29 metabolites at 6 h (Table S3). For the blood samples, there was a significantly higher relative intensity of ¹³C-labeled valine-methionine (val-met), a dipeptide combining valine and methionine, at 6 h. As highlighted, methionine was a metabolite with detectable microbial contributions in the lung.

DISCUSSION

With the advance of culture-independent techniques, we have started to uncover the importance of the human microbiome in different mucosa, including the lungs, in both health and disease.^19–21 However, the presence or absence of microbes in the lung environment does not truly capture the activity of these microbes and the actual impact microbial products have on the host. Active microbial metabolism, for example, leads to the production of SCFAs, which are known to be immunomodulatory, a phenomenon that has been studied extensively in the gut.^22–24 In some of our prior work, we used metagenomic and metatranscriptomic sequencing to demonstrate that microbial function in the lower airways correlates with levels of SCFAs found in that environment.⁸ However, the full impact of lower airway microbes, which have much lower biomass compared with other mucosae sites, on the metabolic environment of the lower airways remains unclear. In this study, we employed a multi-omics approach, encompassing metabolomics, metagenomics, and metatranscriptomics, to investigate a broad metabolic profile along the respiratory tract and estimate the contribution of microbial metabolism to the metabolic environment. In a group of subjects with a significant smoking history, we found that the metabolic environment differs between the upper and lower airways. In parallel, microbial functions also varied along the respiratory tract and their metabolic pathways correlated with the mucosal levels of several metabolites. This was further evaluated in a pre-clinical model, whereby inoculating a mixture of common human oral commensals into the lungs of mice induced a change in the lower airway metabolic environment that also correlated with changes in microbial function. As a proof of principle, we showed that human oral commensals, such as Prevotella melaninogenica, Veillonella parvula, and Streptococcus mitis, were directly associated with the production of specific metabolites, such as adenosine, glutamic acid/glutamate, niacinamide, tyrosine, and methionine, affecting their levels in the lower airways. This is of importance, given the potential immunomodulatory effects and critical relevance for several of these metabolites in affecting host metabolic processes.

Several studies, including some from our group, have shown that bacterial biomass is much higher in the UAs compared with the lower airways.^25–27 The respiratory tract, although a continuum, harbors very different microbial communities in the nasal cavity, mouth/oropharynx, trachea, and lower airways. Frequently, microbes commonly considered commensals in the UAs are identified in the lower airways, potentially due to micro-aspiration, which can occur in healthy adults but also in patients with pulmonary conditions such as COPD, potentially facilitating the spread of these microbes.^28–31 However, even when the same microbes are found in both the upper and lower airways, their function may differ due to exposure to distinct environments with different nutrients. We have previously shown that oral commensals are functionally active in the lower airways of patients with physiological evidence of COPD and, as mentioned, these oral commensals are associated with the production of immunomodulatory metabolites.⁸ In addition, their microbial metabolic activity can also contribute to shaping their surrounding environment and have potential physiological roles through interactions with the host, such as maintaining host metabolite balance and modulating host cell signaling.^32,33 Here, beyond showing marked taxonomic differences between the microenvironment of the upper and lower airways, our multi-omics approach allowed us to also uncover topographical differences in microbial function and metabolism. Our data support the notion that the metabolic environment is compartment specific. Other investigations have also uncovered differences between the lower airways and systemic circulation. In a prior metabolomic study on COPD, it was noted that a larger subset of metabolites detected in BAL samples correlated with severity of airway obstruction than those detected in plasma.³⁴ Consequently, although lower airway samples are more difficult to obtain, they likely provide a different and possibly more realistic perspective in lung disease.

Having seen that microbial abundance gradient seemed specific to various compartments, as previously suggested,³⁵ we found that not all taxa at lower abundance in the lower airways had lower functional levels in the same compartment based on metatranscriptomic reads. In our analysis, we applied a surrogate metric (the difference between metagenomic and metatranscriptomic reads for microbial function pathways) to better assess active microbial pathways, integrating both genetic potential (MG) and genetic expression (MT). Through this analytical approach of orthogonal data, we identified that some oral commensals, e.g., Prevotella and Veillonella, were present at higher relative abundance in the UAs but more functionally active in the lower airways (higher MT-MG). These findings extend those of prior studies,^1,26 where enrichment of the lower airways with oral commensals was associated with elevated inflammatory tone. Thus, the increased microbial function of these oral commensals within the lower airways supports a mechanism by which these taxa can interact with the host immune response.

Previous studies combining metabolomics and microbial sequencing data have primarily relied on correlation between taxonomic and metabolomic datasets,^36,37 inferring potential microbial influence on the metabolic environment without establishing direct causation. To better elucidate the functional link between microbial genes and metabolites, more integrative multi-omics approaches have been implemented to link genomic, transcriptomic, and metabolomic data to map out potential biochemical pathways and interactions.³⁸ Similarly, to determine whether microbial functional differences between the upper and lower airways contribute to a distinct topographical metabolic signature, we used an integrative method (HeatWave analysis) based on an orthogonal approach that combines metabolic, (meta)-transcriptomic and reaction datasets, leveraging on KEGG reaction network. We identified several metabolites of interest that were clearly associated directly with microbial function, such as glutamic acid/glutamate, previously suggested to be involved in different inflammatory pathways.^39,40 We also found microbial genomic function coregulates with levels of adenosine and its derivative, adenosine 5′-monophosphate, as well as methionine, both previously shown to be elevated in COPD patients and associated with bronchoconstriction and regulation of airway inflammation.^41,42 As a proof of concept, our pre-clinical model extended these associations demonstrating how levels of specific metabolites could be modified by introducing live oral commensals to the lower airways. Our findings support the active role of functional microbes in the variation of these metabolites in the respiratory tract.

It is also important to recognize that many metabolites are affected by both microbial and host metabolism.⁴³ Here, we conducted a mouse experiment where we attempted to distinguish the impact of functionally active microbes in the lower airway environment. In this pre-clinical model, we introduced live oral commensals (MOC) into the oral cavity and lung of mice. Interestingly, the selected human oral commensals, Prevotella melaninogenica, Streptococcus mitis, and Veillonella parvula, seemed to seed the lower airways better than the UAs of mice. This could have resulted from enhanced clearance of these microbes in the UAs, as compared with the lower airways, or the resilience of abundant oral microbiota creating a barrier effect preventing new inductions in the oral cavity. It is also possible that the lower bacterial abundance in the lower airways of these mice may provide a more favorable environment to facilitate the establishment of these microbes.^44,45 This location-dependent functional difference of MOC in our experimental model provides evidence that, in many ways, mirrors our observations in human samples. Although the artificial introduction of MOC into the lower airways of mice may not be reflective of real-life physiological process, this transient exposure of MOC not only altered the microbial functional pathways but also changed the metabolic environment in the lower airways. The functional activities of other microbes also likely changed indirectly due to the presence of the MOC taxa, as some of the metabolic pathways not associated with MOC taxa were also found to be enriched in the lower airways. The intensity of different metabolites, such as glutamic acid/glutamate and methionine, known to be produced by the three taxa as part of MOC,⁴⁶ increased as MOC was introduced into the lower airways. Importantly, there were several metabolites affected by our experimental model that overlap with the associations found in the human data, such as adenosine, glutamic acid/glutamate, etc. Interestingly, homocysteine, a precursor to methionine, had a strong correlation with microbial function in the lower airways of humans in our data, supporting the translational aspect of the pre-clinical model. Furthermore, upon inoculation of MOC into the lung, we saw increased transcriptional signals for different metabolites across MOC taxa, particularly with Veillonella parvula. Reduced functional expressions from Streptoccucus mitis and Prevotella melaninogenica may be attributed to their relatively rapid clearance from the lower airways of mice as compared with Veillonella parvula, something we have previously shown.²

To strengthen the robustness of these findings, we used ¹³C stable isotope glucose labeling⁴⁷ of metabolites produced by Prevotella melaninogenica. This is not a perfect system that would label all metabolites produced by Prevotella melaninogenica; rather, it labels a proportion where carbons present in the labeled glucose contribute to the metabolites produced by this bacterium. A prior study used a similar approach through ¹³C-labeled dietary fibers to detect microbial metabolites attributed to different fiber sources.⁴⁸ In our ex vivo model, we were able to trace in the lower airways the labeled carbon of several isotopic metabolites, such as adenosine (M + 5) and glutamic acid/glutamate (M + 3), among those mice exposed to the Prevotella melaninogenica grown in ¹³C glucose-enriched media. Adenosine (M + 5), for example, received five ¹³C atoms via the metabolism of ¹³C glucose by Prevotella melaninogenica through glycolysis and the pentose phosphate pathway. We extended the ex vivo findings to our in vivo model and identified an increased proportion of isotopic ¹³C labeling over an extended period of time (up to 6 h in this experiment) for many metabolites, such as glutamic acid/glutamate and methionine, in the lower airways. Furthermore, we were able to identify derivatives of methionine in blood, which raises the possibility that metabolic changes in the lower airways may influence the systemic metabolic profile through potential absorption of metabolites into the blood stream and subsequent enzymatic modifications into the metabolite derivatives. Unsurprisingly, there appears to be a stronger local effect of microbial function on the metabolic profile of the lower airways, supported by a significantly higher number of traceable metabolites in the lung samples relative to the blood samples. It is important to note that these experiments only offer a proof of concept that microbes contribute to the metabolic environment of the lower airways. Additionally, we were unable to confirm whether some of the metabolites of interest (such as niacinamide and glutamic acid/glutamate), are bacteria-derived through ¹³C stable isotope glucose labeling. However, it is quite possible that stable isotopic labeling of other carbon sources (other nutrients) would have been needed to effectively label different metabolites.

Better characterization of functionally active microbial communities has become increasingly important, especially given the recent interest in therapeutic approaches to altering the airway microbiota.⁴⁹ In a recent study by Liang et al., the authors proposed that targeted therapy of airway dysbiosis using bacteriophages against Staphylococcus aureus may restore pulmonary function among emphysema modeled mice.⁷ Interestingly, this approach aimed to regulate levels of homocysteine, a metabolite produced by Staphylococcus aureus,⁴⁶ given its potential role in neutrophil proliferation.⁵⁰ Similarly, in our study, we saw that homocysteine was strongly correlated with several microbial pathways in the lower airways. Many other metabolites affected by microbial metabolism, such as SCFAs and tryptophan/kynurenine, are known to play important roles in inflammatory and host immune responses.^51,52 The microbial functions affecting the levels of amino acids such as methionine and tyrosine are likely to alter the host’s uptake of these amino acids and their downstream metabolism.⁵³ For instance, exogeneous supplementation of methionine for macrophages has been shown to decrease production of pro-inflammatory mediators, such as tumor necrosis factor alpha (TNF-α) and interleukin (IL)-6, after exposure to lipopolysaccharides.⁵⁴ Thus, dissecting the contribution of lower airway microbes to their levels is likely important for understanding disease pathogenesis and will potentially uncover novel therapeutic approaches.

There are inherent limitations to our study. First, although we saw several significant differences between upper and lower airway samples, it is important to recognize that the number of samples with quality data on three orthogonal platforms was relatively small, and further studies exploring functional aspects on the lung microbiome should be considered using larger sample sets. In addition, with a larger cohort, we would hope to perform paired analysis with different sample types across the respiratory tract. Second, similar to prior studies,^8,55 we did not remove from the metagenomic and metatranscriptomic data sequence reads that we identified as potential contaminants. We based this decision on the compositional nature of the data and difficulties with precisely establishing the contaminant nature of different signals.⁵⁶ However, we believe that, through identification of overlapping signals across orthogonal platforms, we focus on those signatures that are more likely to be real rather than those introduced by contaminants or stochastic noise.^57,58 Additionally, for the purpose of this manuscript, we used one method of alignment that is commonly used (Kraken v2.1.2 and Bracken v2.5) but appreciate there are different methods available, which may result in different results. Employing a global metabolomic approach allowed us to systematically detect a large number of metabolites. However, it may have included metabolites from acellular sources like culture media, and from potential contaminants or cellular debris. Acknowledging the challenge of defining the impact of contaminants, we used ¹³C isotope labeling to accurately assess the direct contributions of microbes of interest. Because microbes are likely to utilize various sources of carbon, future investigations should consider using other compounds, such as amino acids and fatty acids, to better trace microbial metabolic pathways. Third, we focused solely on demonstrating direct microbial contributions to the metabolic signatures in the respiratory tract, without exploring the potential host’s involvement, recognizing the complex metabolic interplay between microbes and the host.¹⁰ Follow-up investigations should incorporate techniques such as host transcriptomics using bulk RNA sequencing (RNA-seq) to assess how these identified signatures impact the host. This approach will help determine how microbiome-induced metabolic changes directly affect host gene expression and host-specific metabolic pathways. Because microbial metabolism likely changes over time, longitudinal studies incorporating host transcriptomics would better capture the evolving metabolic environment. Fourth, we were unable to evaluate the impact of cigarette smoke exposure on the microbiome and metabolome, as most of our patients were current smokers and their smoking habits were self-reported. Given potential alteration of the microbial composition and metabolite production in the respiratory tract by cigarette smoking exposure, future studies could perform subgroup analyses based on smoking status with a larger cohort or sensitivity analyses adjusting for biomarkers of tobacco smoke exposure, such as serum cotinine and urine 4-(methylnitrosamino)-1-(3-pyridyl)-1-butanol (NNAL).^59–62 Lastly, our analysis did not focus on comparing patients with different demographic and clinical characteristics, and the predominance of males in our human cohort prevented us from performing sex and gender-based analyses. Similarly, the mouse model involved only female mice, which could introduce sex-specific biases. Although we identified variations in metabolic signatures due to active microbial function, the clinical relevance of the observed differences on disease pathogenesis and clinical phenotype should be investigated further. Future studies could therefore investigate how changes in microbial metabolites impact host metabolism and immune response in both healthy individuals and those with pulmonary diseases. Employing experimental methodologies to manipulate microbial composition will be essential to identify microbial-associated metabolic signatures specific to a particular disease and discern any resulting effects on clinical outcomes.

In summary, our study demonstrated the direct impact of the airway microbiome on the metabolic landscape of the lower airways. We show that the functional activity of different taxa varies depending on the environment and contributes to the metabolic profiles throughout the respiratory tract. Separating our assessment of the microbial and host metabolism is therefore likely important to better understand microbial-host interactions and dissect the pathogenesis of pulmonary diseases.

RESOURCE AVAILABILITY

Lead contact

Further information and requests for resources and codes should be directed to and will be fulfilled by the lead contact, Imran Sulaiman, MD, PhD (sheikmohammadImran.sulaiman@nyumc.org).

Materials availability

This study did not generate new unique reagents.

Data and code availability

• Sequencing data are available in NCBI’s Sequence Read Archive under project numbers PRJNA870929, PRJNA1182825, and PRJNA1182915.

• Codes used for the analyses in this manuscript are available at https://github.com/segalmicrobiomelab/Microbial_Metabolism.

• Any additional information required to reanalyze the data reported in this work paper is available from the lead contact upon request.

STAR★METHODS

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

Human Cohort

This was a prospective observational study of healthy research subjects, as part of an ongoing project that has been approved by the Institutional Review Board of New York University (IRB#S14–01546), titled “Lung Microbiome and Inflammation in Early COPD”. All subjects were recruited through advertisements in local newspaper with the intent to enroll individuals with significant smoking history and gave written informed consent. Subjects with more than 10-pack years smoking history were invited for further screening. At in-person assessment, subjects underwent an initial screening visit that included blood sample collection, 12-lead EKG, chest X-ray and pulmonary function testing (PFTs). Exclusion criteria for subjects included severe airflow obstruction with FEV1<50% predicted, recent use of systemic corticosteroids or antibiotics within the past two months, known cardiovascular disease or abnormal EKG, diabetes mellitus, known cancer, significant liver or renal disease, severe coagulopathy (INR > 1.4, PTT > 40 seconds, and platelet count < 150 × 10^3 cells/L), pregnancy and high self-reported alcohol consumption with more than 6 beers or 4 mixed drinks daily. Those seemed suitable were then invited for elective bronchoscopy specifically for research purposes, with no other clinical indications.

Pre-Clinical Model

C57BL/6 Mice were kept in their respective vivariums under monitored temperature and environmental control. The mice received 12-hour day and night cycles and were offered food and water on an ad libitum basis. Female mice aged 8–10 weeks were used for consistency with previous studies and consistency of biological sex.^8,68 To control for cage effect, mice were grouped between those who received microbial challenges and those receiving the PBS control within the same cage. All animal studies were performed with the approval from the Institutional Animal Care and Use Committee of Columbia University School of Medicine and New York University Grossman School of Medicine. Further detailed description of the mouse model methods have been previously published.⁶⁹ Lungs and tongues from euthanized mice were harvested.

METHOD DETAILS

Lower airway and upper airway sampling procedure

Every subject underwent a single research bronchoscopy to obtain one set of samples, including environmental background (BKG), upper airway (UA) and bronchoalveolar lavage (BAL) samples. BKG and UA samples were obtained prior to bronchoscopy, as previously described.¹ BKG samples were obtained by passing sterile saline through the suctioning channel of the bronchoscope prior to the procedure. UA samples were collected by suctioning the oral cavity using a Yankauer suction tip prior to performing bronchoscopy. BAL samples were obtained from the right middle lobe and lingula. Samples were transported on ice to the laboratory where aliquots of whole BAL and BAL cell pellet were generated and stored at −80°C until further processing.

DNA/RNA isolation, library preparation and sequencing

This study did not generate new unique reagents and there are no restrictions to the availability. DNA and RNA were isolated from BAL, UA and BKG controls in parallel using zymoBIOMICS^™ DNA/RNA Miniprep Kit (Cat: R2002) as per manufacturer’s instructions. For DNA library prep was performed using Nextera DNA Flex Library Prep Kit (Illumina, Cat# 20018704). DNA obtained from whole BAL underwent untargeted sequencing (whole genome sequencing, WGS) on a NovaSeq platform (Illumina). For RNA library prep was performed on a Beckman Coulter Biomek FXp workstation using the Tecan Trio RNA-Seq Library Preparation Kit (Tecan, Cat# 0506-A01). RNA obtained from whole BAL underwent RNA Sequencing for metatranscriptome. In both datasets human reads were removed and analysis was performed using Trimmomatic v0.36,⁶⁴ Bowtie2 v2.3.4.1,⁶³ Kraken v2.0.7,⁶⁵ Bracken v2.5,⁶⁶ FMAP v0.1517,⁶⁷ as previously described.²⁶

Liquid Chromatography-tandem Mass Spectrometry (LC-MS) extraction

BALF, UA and BKG samples were extracted using the Oasis HLB 3cc/60mg Waters solid phase extraction column. 4mL of sample was loaded into conditioned columns and then eluted using 1.6mL of 80% methanol with 0.1% TFA. Samples were then dried down using a speed vacuum and reconstituted in 40μL of LC-MS grade water. Other samples (from the pre-clinical mouse models) were mixed with 80% methanol at concentrations of 10mg/mL for solid tissue, 25 μL/mL for plasma, and 200 μL/mL for media. Samples were then homogenized using a BeadBlaster (Benchmark Scientific) and centrifuged (21kg for 3 minutes at 4°C). 450μL of supernatant was then dried down via speed vacuum concentration and reconstituted in 50μL of LC-MS grade water.

LC-MS/MS with the hybrid metabolomics method

Samples were subjected to an LCMS analysis to detect and quantify known peaks. A metabolite extraction was carried out on each sample based on a previously described method.⁷⁰ The LC column was a MilliporeTM ZIC-pHILIC (2.1 ×150 mm, 5 μm) coupled to a Dionex Ultimate 3000TM system and the column oven temperature was set to 25°C for the gradient elution. A flow rate of 100 μL/min was used with the following buffers: A) 10 mM ammonium carbonate in water, pH 9.0, and B) neat acetonitrile. The gradient profile was as follows; 80–20%B (0–30 min), 20–80%B (30–31 min), 80–80%B (31–42 min). Injection volume was set to 2 μL for all analyses (42 min total run time per injection).

LC-MS/MS with the hydrophobic metabolomics method

Samples were subjected to an LCMS analysis to detect and quantify putatively identified peaks and features. The LCMS parameters were adapted from a previously described method.⁷¹ The LC column was a WatersTM BEH-phenyl (2.1 ×150 mm, 1.7 μm) coupled to a Dionex Ultimate 3000TM system and the column oven temperature was set to 25°C for the gradient elution. A flow rate of 200 μL/min was used with the following buffers: A) 0.1% formic acid in water, and B) 0.1% formic acid in acetonitrile. The gradient profile was as follows; 0–35%B (0–10 min), 35–75%B (10–15 min), 75–99%B (15–15.25 min), 99–99%B (15.25–16.5 min), 99–0%B (16.5–16.75 min), 0–0%B (16.75–20 min). Injection volume was set to 2 μL for all analyses (20 min total run time per injection).

LC-MS/MS Analysis

MS analyses were carried out by coupling the LC system to a Thermo Q Exactive HFTM mass spectrometer operating in heated electrospray ionization mode (HESI). Method duration was 20 min with a polarity switching data-dependent Top 5 method for both positive and negative modes. Spray voltage for both positive and negative modes was 3.5kV and capillary temperature was set to 320°C with a sheath gas rate of 35, aux gas of 10, and max spray current of 100 μA. The full MS scan for both polarities utilized 120,000 resolution with an AGC target of 3e6 and a maximum IT of 100 ms, and the scan range was from 95–1000 m/z. Tandem MS spectra for both positive and negative mode used a resolution of 15,000, AGC target of 1e5, maximum IT of 50 ms, isolation window of 0.4 m/z, isolation offset of 0.1 m/z, fixed first mass of 50 m/z, and 3-way multiplexed normalized collision energies (nCE) of 10, 35, 80. The minimum AGC target was 1e4 with an intensity threshold of 2e5. All data were acquired in profile mode.

For each batch, the resulting ThermoTM RAW files were converted to SQLite format using an in-house python script to enable downstream peak detection and quantification. The available MS/MS spectra were first searched against the NIST17 MS/MS,⁷² METLIN⁷³ and respective Decoy spectral library databases using an in-house data analysis python script adapted from our previously described approach for metabolite identification false discovery rate control (FDR).⁷⁴ Putatively identified metabolites were filtered for any duplicated metabolites names to generate a list of metabolites with unique names. Next, the decoy hits in the resulting list were dropped, and the peak heights for each putative metabolite hit were extracted from the sqlite3 files based on the metabolite retention time ranges and accurate masses in the above-mentioned metabolite list.

Metabolism-Focused Data Extraction for Multi-Omics Analyses

To focus on microbial metabolism, only KEGG pathway associated with Metabolism were included in the differential analysis for metagenomic and metatranscriptomic pathway datasets.⁷⁵ We also identified metabolic pathways associated with the three MOC taxa (Veillonella parvula, Prevotella melaninogenica, Streptococcus mitis) based on the KEGG database.

For metabolomic analysis, only KEGG-annotated metabolites involved in metabolic pathways were included. The Clinical Translation Service was utilized to convert InChIKey IDs to KEGG Compound IDs. Given that metabolites can be utilized or produced, we did not subtract background (BKG) contributions to BAL or UA samples for metabolic analyses, given it could result in missing important metabolic activities. However, metabolites that were undetected in more than 70% of the samples were filtered out.

Identification of Potential Microbial Contaminants

Background samples served as negative controls to identify potential contaminants in UA and BAL samples. To ensure sufficient sequencing depth and taxonomic data for background samples, we used an increased amount of template DNA/RNA compared to UA and BAL samples.

We applied the Wilcoxon rank-sum test (p<0.05) to compare the relative abundance of taxa between the background samples and the UA and BAL samples (Table S4). In the metagenomic data of BAL samples, Burkholderia dolosa, Xanthomonas citri, and Pseudoalteromonas sp. scap25 were identified as the most abundant potential contaminants, while the metatranscriptomic data revealed Malassezia restricta, Cutibacterium acnes, and Talaromyces rugulosus as the most prevalent potential contaminants. All potential contaminants were highlighted in red for identification purposes in all figures but were not removed from the analyses, due to the compositional nature of the data.

For reference, in addition to Wilcoxon rank-sum test, we also utilized decontam, a commonly used statistical classification technique for identifying contaminants in metagenomic and metatranscriptome data, with a test threshold of 0.5. We provided tables listing all taxa identified as potential contaminants using the different methods—Wilcoxon rank sum test, decontam-prevalence, decontam-frequency and decontam-combined (Table S4).

Integration and Visualization of Metatranscriptomic and Metabolomic Data (“HeatWave Analysis”)

HeatWave analysis is a visualization framework which attempts to highlight areas in the metabolic reaction network where there appears to be a high degree of change across the conditions being compared. The network is comprised of nodes corresponding to either orthologs or metabolites and they are connected if the metabolite is either a product or a substrate of the given ortholog in a reaction it is known to catalyze. Here change simply corresponds to fold change and we refer to it as “heat” and measure it as the absolute value of the base two logarithm of the fold change (i.e. abs(log₂(fc)) ). The only input required is a list of metabolites (technically molecular skeletons or “InChIK”s – see below) and orthologs (listed as KEGG KO numbers) with their corresponding fold changes. The resulting visualization filters the data showing only those nodes that are sufficiently “hot” i.e. have a value of abs (log₂(fc)) above a predefined threshold. The framework enables a heuristic whereby heat is iteratively spread (in so called “HeatWaves”) across nodes to support the visualization of connected modules or pathways despite the potential presence of individual nodes which either were not measured or did not appear to be changing across the conditions being compared. In this report, the heat-spread heuristic was not utilized but the capability is delivered along with the source code (see below).

The first step in the process is the generation of the reaction network. The entirety of this step is implemented by a one Python script (create_metagenomic_network.py) which downloads reactions from the KEGG website using the official RESTful API and linking each reaction to the orthologs that catalyze the reaction. Additionally, for each metabolite participating in a reaction, the official InChIKey is downloaded from PubChem (when available). This enables the unambiguous naming of metabolites according to their potentially shared chemical core structure, which is encoded as the first 14 characters of the InChIKey (which we refer to as an “InChIK”). This allows the system to refer to metabolites purely in terms of their equivalence class when studied by mass-spectrometry, since it is typically not possible to resolve enantiomers. Note that, when visualizing results, one specific metabolite name will be chosen as label, but this is purely for readability. All the information downloaded by the network generation scripts is cached and given a unix timestamp in order to unequivocally date the underlying data sources. The output of the script is a tab-delimited representation of the network, where each row corresponds either to a node definition or an edge, with the first field being used to distinguish between the two. For rows representing nodes, the next two fields specify whether the node is a reaction or a metabolite and the node’s ID. For rows representing edges, the next two fields represent the source and destination of the edge. Note that HeatWave treats all edges as undirected, so the order of these two fields is immaterial.

The actual HeatWave analysis is done by another python script (analyze_metagenomic_data.py) which takes as input the previously generated reference network along with two input data files: one representing transcriptomic data and the other representing metabolomic data. In both cases, the first column in the data file represents an identifier that matches those used in the network definition (KEGG ortholog numbers and InChIKs, respectively). The second column represents the log₂ ratio of conditions being compared, and the third column provides a user-friend label for use in the interactive network visualization, since both InChIKs and KEGG ortholog IDs are essentially opaque to the end-user. The resulting data then undergoes an option iterative process of data processing (called “waves”), whereby evidence of activity is spread along neighboring edges. This is done by storing in each node the absolute value of the log₂ fold changes for each node (the “heat” of the node) and iteratively sending a proportion of that heat (which is controlled by an input parameter) along each edge. The heat is then summed, and the process is repeated. The final step in the process is a filtering of only those nodes that contain heat above a certain threshold (also a user-controlled parameter). If no heatwaves are applied, then this corresponds to a straightforward ratio thresholding. The resulting network is then exported both as an interactive HTML-based graph with editing and export capabilities, and as a tab-delimited file that reports, for each node, its ID, label, original fold-change, and final heat. Depending on the user-specified parameter, singleton nodes (without any neighbors that pass the heat filter) are removed from the network. The script also reports network statistics including: total network metabolites, total network orthologs, total network reactions (i.e. network edges), number of metabolites with empirical values, number of ortholog (transcripts) with empirical values, percentile of metabolites/orthologs above the cutoff parameter, median absolute fold change (BAL vs UA).

In building the HeatWave network, we used a reference library with 10,546 network nodes, 4,548 network metabolites, 5,998 network orthologs and 31,534 network edges. The HeatWave analysis visualizations generated in this study were done with a wavenumber of zero, meaning that no heat was transferred across network edges. Hence, all visible nodes were originally detected in either metabolomic or meta-transcriptomic data. All code and data for the HeatWave Analysis workflow is available on GitHub: https://github.com/NYUMetabolomics/UA_vs_BAL.

In particular, we included the reference network used for this publication as well as the results of the analysis. The analyze_metagenomic_data.py script can be run with no user specified parameters, and it will regenerate the exact result used in the paper as long as the provided reaction_network.tsv file is used. Running the create_metagenomic_network.py script will overwrite this file which may modify the resulting output reflecting changes in the curation of the underlying metabolic resources (in this case KEGG and PubChem).

Human Oral Commensals

As done before, we used a mixture of oral commensals (also known as MOC consisting of Veillonella parvula, Prevotella melaninogenica, and Streptococcus mitis; ATCC 17742, ATCC 25845, and ATCC 49456, respectively). From the vendor, the bacteria were grown in anaerobic conditions (Bactron 300, Shel Labs, Cornelius, OR), then stored in 20% glycerol tryptic soy broth at −80°C. To prepare the oral commensal inoculum the bacteria strains were thawed and streaked on anaerobic PRAS-Brucella Blood agar plates (Anaerobe Systems, Morgan Hill, CA). The plates were incubated at 37°C in an oxygen-free environment (tri-mix: 5% carbon dioxide, 5% hydrogen, and 90% nitrogen) in the anaerobic chamber for 24–48 hours. The colonies were collected from the plate and re-suspended in 1 ml of sterile PBS. The OD620 was measured to calculate the approximate concentration prior to use to ensure equal distribution of the three taxa within the inoculum. The MOC was dispensed via an intra-tracheal challenge in a 50 μL aliquot using a gel-loading tip and micropipette. MOC species were chosen based on identification of these genus and strains in previously published data of oral and lower airway microbiota.^{1,8,25,68,76,77}

Intra-tracheal microbial challenges

Mice were anesthetized using isoflurane via VetFlow Anesthesia Machine (Kent Scientific, Torrington, CT) sedated to 10–15 breaths per minute and monitored for any distress. While anesthetized, mice were then placed on an intubation platform with blunt forceps, with their tongue gently pulled ventrally until the pharynx was exposed.⁷⁸ A human otoscope with a 2 mm ear cone (Welch Allyn 3.5V Hill-Rom, Inc., Skaneateles Falls, NY Model #20200) was introduced into the oral airway to visualize the murine vocal cords. A gel-loading pipette tip loaded with a 50 μL aliquot was introduced through otoscope cone and the murine vocal cords of the mouse. Then, the aliquot was deployed into the lower airway. The mouse was removed from the platform to recover on a heat pad. The procedure is described in detail previously.^8,68,79,80

As a first step to identify active microbial functions in the lower airways we inoculated MOC into the oral cavity (MOC_Oral) or into the lung (MOC_Lung), with some mice receiving PBS as a negative control (PBS_Lung) as displayed in Figure 4A. This exposure was repeated twice over a 24hr period followed by sacrifice 2 hours after the last exposure. Harvested tongue and lungs were then used for metagenome, metatranscriptome and metabolome assessment performed as described above.

Isotopic Labeling

To prepare 13C-labeled Prevotella melaninogenica, bacteria were initially cultured in M9 medium. The bacteria were grown in liquid media with the addition of [u-13C6] glucose (Cambridge Isotope Laboratories) as another carbon source to try to trace metabolites produced by this organism. To this end, we prepared liquid cultures in anaerobic tubes by adding 0.5 grams of normal glucose or U-13glucose to 2 ml of RPMI 1640 medium. Then, using a sterile inoculating loop or pipette, transfer a small amount of the isolated Prevotella melaninogenica into the liquid culture medium in 15 ml tubes with a tight cap and kept within the anaerobic chamber and kept it in the anaerobic incubator for 48hrs. After that, in order to separate the supernatant from bacteria, we centrifuged at 800g × 10 min. Supernatant was then stored for ex vivo measurement of microbial metabolites while pellet with bacteria was used for our inoculation experiments with labeled bacterial metabolites (Figure 6A). For this later experiment, bacteria harvested at OD600 0.5 – 1.0 were washed with PBS and resuspended in cold PBS (2 × 10^9 CFU per ml). For intratracheal inoculation, 10^8 CFUs per animal were delivered into WT B6 mice. Tissue harvesting was performed 1 and 6 h post-inoculation for further analysis, including collection of samples of plasma and lung tissue. Plasma samples were collected with BD Microtainer and stored at −80°C, while lung tissue was washed in ice-cold PBS and immediately frozen in liquid nitrogen. Further untargeted metabolomics analysis with liquid chromatography/tandem mass spectrometry was conducted on the collected samples at the Metabolomics Core Resource Laboratory of NYU.

For isotope labeling analyses, we calculated the theoretical m/z value for each isotopologue based on the metabolite formula, ion type, and combination of isotopes, including no labeling through maximum labeling. These metabolite peaks were then extracted based on the theoretical m/z of the expected ion type (with or without labeling), e.g., [M+H]+, with a 15 part-per-million (ppm) tolerance and a ± 0.2 min peak apex retention time tolerance within an initial retention time search window of ±0.5 min. The resulting data matrix of metabolite intensities for all the samples and blank controls was processed using an in-house python script (https://github.com/NYUMetabolomics/plz), and the final peak detection was calculated based on a signal-to-noise ratio (S/N) of 3× compared to the blank controls, with a floor of 10,000 (arbitrary units). For the samples where the peak intensity was lower than the blank threshold, the metabolites were annotated as not detected and were imputed with either the blank threshold intensity for statistical comparisons so as to enable an estimate of the fold change, as applicable, or zeros for the median metabolite intensity calculation of a sample. For isotope labeling analyses, we further converted the resulting peak intensities to relative abundance to the monoisotopic peak (defined as 100%) for each metabolite and converted to a percentage value. Unlabeled control samples were used to assess isotope labeling enrichment and background levels of natural isotopes.

QUANTIFICATION AND STATISTICAL ANALYSIS

Statistical Analysis

For human samples, we applied Bray-Curtis dissimilarity indices to determine the community metabolite composition difference among lower airway, upper airway and background samples. We utilized R package vegan (version 2.6.4) to calculate Bray-Curtis dissimilarity indices to determine the community metabolite composition as well as construct Principal Coordinate Analysis (PCoA). To identify differentiating metabolites between lower airway and upper airway, we utilized Partial Least Square Discriminant Analysis (PLS-DA) from mixOmics (version 6.22.0) R package. For differential expression of KEGG pathways in metagenome and metatranscriptome between lower airway and upper airway, we applied edgeR (version 3.40.2) R package, with FDR cut-off 0.2. We chose an FDR cutoff of 0.2 for our differential expression analyses to capture a broader range of differentially expressed genes and pathways of potential interest. This higher threshold increases sensitivity, allowing us to identify trends that might be missed with lower cutoffs.

To compare metagenomic data to metatranscriptomic data, we evaluated read counts for each dataset with matched taxonomic annotation and matched functional (KO) annotation, creating a delta between functional capacity (MG) and active functional transcribing (MT): MT-MG. We applied a Wilcoxon rank-sum test to compare BAL and UA samples, retaining only ones that with a p-value below 0.01. We then sorted based on the top median difference across datasets for BAL and UA separately to determine the most upregulated microbial function and taxa for lower airway and upper airway, respectively.

Similar to the approach with human samples, with in preclinical in-vivo model, we again applied edgeR to identify top differentiating KEGG pathways in metagenome and metatranscriptome for lower airway samples exposed with phosphate-buffered saline (PBS) and mixed oral commensals (MOC). We also utilized PLS-DA to identify differentiating metabolites between lower airway samples exposed with PBS and MOC. To determine the microbial contribution to specific metabolites, we first identified the KEGG Orthologies (KO) which encode information on enzymes and biochemical transformations for the metabolites of interest from the KEGG Orthology database.⁸¹ We then utilized Kraken v2.1.2 and Bracken v2.5 to identify all microbes with functional annotation in both metagenomic and metatranscriptomic sequences for those KOs. Lastly, we calculate the percent contribution of Veillonella parvula, Prevotella melaninogenica, and Streptococcus mitis among all of the identified microbes with functional annotation for KOs associated with each metabolite. Additionally, we used the PathwayGenome Database from Biocyc to identify compounds associated with Prevotella melaninogenica (ATCC 25845), Streptococcus mitis strain NCTC 12261 (ATCC 49456) and Veillonella parvula (ATCC 17742).

Supplementary Material

SUPPLEMENTAL INFORMATION

Supplemental information can be found online at https://doi.org/10.1016/j.chom.2025.06.002.

KEY RESOURCES TABLE.

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Biological samples
Human Background Samples	This Paper	BKG
Human Upper Airway Samples	This Paper	UA
Human Bronchoalveolar Lavage Samples	This Paper	BAL
Murine Lungs	Jackson Laboratories (C57BL/6J)	Lung
Murine Tongues	Jackson Laboratories (C57BL/6J)	Tongue
Murine Plasma	Jackson Laboratories (C57BL/6J)	Plasma
Chemicals, peptides, and recombinant proteins
Formic Acid	Sigma Aldrich	Cat#5330020050
Acetonitrile	Fisher Scientific	Cat# A955-4
Methanol	Fisher Scientific	Cat# A452-4
Water	Fisher Scientific	W5-4
[u-13C6] glucose	Cambridge Isotope Laboratories	CDLM-3813-1
Trifluoroacetic acid (TFA)	Sigma Aldrich	Cat#302031-10X1ML
Ammonium carbonate	Spectrum Chemical	Cat# A1913-500GM
Ammonium hydroxide	Sigma Aldrich	Cat# 338818-100ML
Acetonitrile	Fisher Scientific	A955-4
Acquity UPLC BEH Phenyl 1.7um, 2.1×150mm column	Waters	186003378
Formic Acid	Sigma Aldrich	5330020050
D-Glucose	Sigma Aldrich	Cat# G-7528
Critical commercial assays
ZymoBIOMICS DNA/RNA Miniprep Kit	Zymo Research	R2002
Trio RNA-Seq™ library preparation kit	Tecan	Cat# 0506-A01
Nextera DNA Flex Library Prep Kit	Illumina	Cat# 20018704
Deposited data
Human Metagenome and Metatranscriptome	https://www.ncbi.nlm.nih.gov/sra	PRJNA870929
Mouse Metagenome and Metatranscriptome	https://www.ncbi.nlm.nih.gov/sra	PRJNA1182825
Ex Vivo and Mouse In Vivo Experiments	https://www.ncbi.nlm.nih.gov/sra	PRJNA1182915
Experimental models: Organisms/strains
Human Veillonella parvula	American Type Culture	ATCC 17742
Human Prevotella melaninogenica	American Type Culture	ATCC 25845
Human Streptococcus mitis	American Type Culture	ATCC 49456
Software and algorithms
Bowtie2 v2.3.4.1	Langmead and Salzberg63	http://bowtie-bio.sourceforge.net/bowtie2/index.shtml
Trimmomatic v0.36	Bolger et al.64	https://github.com/usadellab/Trimmomatic
Kraken v2.0.7	Wood et al.65	https://github.com/DerrickWood/kraken2/blob/master/docs/MANUAL.markdown
Bracken v2.5	Lu et al.66	https://ccb.jhu.edu/software/bracken/index.shtml
FMAP v0.1517	Kim et al.67	https://github.com/jiwoongbio/FMAP
HeatWave Analysis	This Paper	https://github.com/NYUMetabolomics/UA_vs_BAL
Other
PRAS-Brucella Blood agar plates	Anaerobe Systems	Cat# AS-141
Oasis HLB 3cc/60mg SPE column	Waters	Cat# WAT094226
SeQuant ZIC-pHILIC (2.1 ×150 mm, 5 μm) column	Millipore	Cat# 15046000001
Acquity UPLC BEH Phenyl 1.7um, 2.1×150mm column	Waters	Cat# 186003378

Highlights.

• Microbial activity and metabolite production vary across the respiratory tract

• Isotope labeling reveals microbial contributions to the metabolic lung environment

• Changes in microbial metabolism contribute to airway metabolic niche construction

• Immunomodulating metabolites are influenced by the lung microbiome