Tobacco Smoking and the Fecal Microbiome in a Large, Multi-ethnic Cohort

Recent work suggests that cigarette smoke may alter the microbiome (1–5). Cigarette smoke is a source of numerous toxicants (6), which can perturb the microbial ecology of the gut via antibiotic effects, oxygen deprivation, or other potential mechanisms (7–10). Loss of beneficial gut species due to smoking may then lead to pathogen colonization and ultimately to disease (11). Yet, these studies describe only phylum-level taxonomic shifts in the Firmicutes to Bacteroidetes ratio, and theoretical physiologic effects. Thus, the relationship between cigarette smoking and alterations in specific microbial taxa and functional characteristics of the gut microbiome remains under-investigated. Given the importance of both smoking and the gut microbiome in modifying cancer risk, understanding their inter-relationships may be critical to modification of those risks.

Thus, we hypothesized that smoking influences the fecal microbiome, resulting in shifts in specific taxonomic abundances and changes in imputed microbial metabolism. To study this hypothesis in a large multi-ethnic cohort of 803 adults from the NYU Food and Microbiome Longitudinal Investigation (FAMiLI), we examined the overall structure of the human gut microbiome and specific taxon abundances in current, former, and never smokers and identified related imputed microbial metabolic pathways.

Participants were recruited from the New York University Food and Microbiome Longitudinal Investigation study cohort (12). Briefly, participants over the age of 40 were recruited in the New York Metropolitan area, including nearby areas of New Jersey and Connecticut. Study participants submitted samples of stool, and completed a questionnaire describing demographic and lifestyle characteristics (13). From the recruited population, we selected an initial subset of 873 participants for microbiome sequencing who confirmed no use of antibiotics in the 2 weeks prior to provision of stool samples (12). Participants with missing or unknown demographic data (age, race, or gender; n = 45), BMI (n = 16), or smoking status (n = 8) were excluded from analysis. Some samples were excluded for missing data in multiple categories. One sample was excluded for insufficient richness, resulting in a final sample size of 803 subjects (Table 1), including 357 White, 98 Black, 269 Asian, and 79 other subjects.

Stool samples were provided by all participants and stored in our biorepository in a −80°C freezer. As described previously, samples underwent 16S rRNA gene sequencing at the Environmental Sample Preparation and Sequencing Facility at Argonne National Laboratory (12). DNA was extracted using the PowerSoil (MO BIO Laboratories) DNA Isolation Kit, following the manufacturer's protocol. The V4 region of the 16S rRNA gene was PCR amplified with the 515F/806R primer pair, which included Illumina flow cell adapter sequences with sample-specific barcodes (14). Sequence reads were processed using the QIIME2 pipeline (15). Briefly, sequence reads were demultiplexed and paired-end reads were joined, followed by quality filtering (16). Next the Deblur workflow was applied, which uses sequence error profiles to obtain putative error-free sequences, referred to as “sub” amplicon sequence variant (s-ASV; ref. 17). s-ASVs were assigned taxonomy using a naïve Bayes classifier pretrained on the Greengenes 13_8 99% ASVs (18), where the sequences have been trimmed to only include 250 bases from the 16S rRNA gene region V4, bound by the 515F/806R primer pair (14). A phylogenetic tree was constructed via sequence alignment with MAFFT (19), filtering the alignment, and applying FastTree (20) to generate the tree.

We identified current, former, and never smokers, and assessed number of cigarettes per day and duration of use in ever-smokers. The relationship between smoking status (never smokers, former smokers, or current smokers) and overall fecal microbiome community structure (β diversity) at the species level was assessed by analysis of Bray–Curtis dissimilarity (21), computed using the QIIME2 pipeline (15). Principal coordinate analysis plots were generated using the nonmetric multidimensional scaling rank method (22), and labeled according to smoking status. Within the vegan R package, permutational MANOVA (“adonis” function, refs. 23–25) of the Bray–Curtis distance was used to test differences in overall microbiome composition across the categories of smoking status, adjusting for age, sex, race, and BMI. Bray–Curtis distance from each smoking group centroid, both within and between each subgroup was also calculated (“meandist” function, vegan R package, refs. 23, 24) and analyzed by ANCOVA controlling for age, sex, race, and BMI. Race was categorized as White, Black, Asian, or Other. BMI was categorized as underweight (BMI < 18.5), normal weight (18.5–25), or overweight/obese (>25; ref. 26). Overweight/obese was defined as BMI > 23 for Asian subjects per global consensus guidelines based on cardiovascular risk (27).

For the primary analysis of bacterial relative abundance by smoking subgroups, the ASV tables of raw counts across all phylogenetic levels were normalized to relative abundances. We filtered taxa with fewer than two copies or present in fewer than 25% of subjects, similar to previous taxonomic analyses (5). Relative abundance of bacterial taxa were compared across the three categories of smoking status, current, former, or never, using the analysis of communities method (ANCOM-II package, R v1.1.463, refs. 28, 29), adjusting for age, sex, race, and BMI. ANCOM-II accounts for the compositionality and sparse nature of microbiome data matrices, in which nearly 80% of the matrix values may be zero (30), to analyze differential abundance between two or more groups with greater power and lower FDR than similar methodologies (29). Taxa were considered statistically significant if they were differentially abundant, as measured by rejection of the pairwise null hypothesis, from more than 90% of taxa at the same phylogenetic level (FDR < 0.001, ref. 28). Differences in the relative abundance of taxa between subgroups was termed “differentials.”

Analysis of differential taxonomic abundance by time since smoking cessation and cigarettes smoked daily was performed with the subset of taxa which were found to be significant by ANCOM analysis. Taxonomic abundances were normalized by centered log-ratio (CLR) transformation (31, 32). This accounts for the compositional nature of microbiome data by taking the ratio of each observation with the geometric mean of the whole composition, thus producing more precise relative abundances than other methodologies (31). CLR normalized abundance of taxa in former and current smokers was compared with never smokers. For analysis of smoking cessation, subjects were binned into current smokers, and former smokers who quit smoking less than 10 years ago, 10 to 29 years ago, or greater than 29 years ago (30+). For analysis of cigarette dose, subjects were binned into smokers of 1 to 10 cigarettes daily or greater than 10 cigarettes daily (11+).

Bacterial metabolic genes and pathways were imputed from 16S rRNA gene-based microbial compositions using the PICRUSt2 algorithm (33), with reference to the MetaCyc pathway catalog (https://metacyc.org/; ref. 34). A total of 391 MetaCyc pathways were imputed across all of our samples. Relative imputed abundance of MetaCyc pathways across the three categories of smoking were analyzed by ANCOM-II, controlling for age, sex, race, and BMI (28). We noted that these pathways were largely clustered within the Biosynthesis and Degradation/Utilization/Assimilation (“Oxidation/Degradation”) MetaCyc superpathways, with the few pathways which fell outside these classes termed “Other.” The correlation between imputed pathways and taxonomic abundance was evaluated by Spearman correlation analysis and visualized using the pheatmap R package (https://CRAN.R-project.org/package=pheatmap).

Our population includes 803 subjects, of which 79 (9.9%) were current smokers, 181 (22.5%) were former smokers, and 543 (67.6%) were never smokers (Table 1). Within these groups, we found that former and current smokers were more likely to be men than women, compared with never smokers. We also found that current smokers were less likely to be overweight or obese, and more likely to identify as Black.

There were no significant differences in subject α-diversity between smoking subgroups (Supplementary Table S1). However, we found significant differences in β-diversity by smoking status (Fig. 1A and B). This differential is suggested by PCoA visualization (Fig. 1A) of Bray–Curtis dissimilarity, and confirmed by PERMANOVA analysis by smoking status (Fig. 1B). We found that β-diversity, as measured by Bray–Curtis dissimilarity, differed significantly between never smokers and both current (P = 0.0006) and former smokers (P = 0.0186). No significant changes were noted when additionally controlling for ethnicity or fiber in our model (Supplementary Table S1). We then evaluated microbiome dissimilarity from centroid, which acts as a measure of the actual microbial variation demonstrated by β-diversity, and found that it varied both within and between smoking subgroups. Specifically, current and never smokers showed significantly greater Bray–Curtis dissimilarity than former smokers (Fig. 1B). We observed that the dissimilarity between current and never smokers was greater than that between former and current smokers or never and former smokers (Fig. 1C).

Fecal taxonomic composition analysis identified 48 taxa differentially prevalent in current smokers versus never smokers, 33 differentially prevalent taxa in former smokers versus never smokers, and 19 taxa in current versus former smokers (Supplementary Table S2). We noted that the Tenericutes, Lachnospira, Veillonellaceae, and Prevotella taxa were significantly altered at the highest ANCOM significance threshold (0.9) between all smoking subgroups. Analysis of mean differences in abundances following centered log-ratio normalization confirmed differences in taxonomic abundance between these taxa, which is visualized using cladogram across three smoking groups and ethnic subgroups (Fig. 2). We noted that the relative abundance of the Firmicutes genus Lachnospira was lower in current relative to never smokers (Fig. 2A and D), whereas the relative abundance of members of the Bacteroidetes genera Prevotella, and the Clostridia family Veillonellaceae were higher in current smokers. These relative differentials in abundance were largely preserved when comparing current smokers to former smokers (Fig. 2C), with depletion of Lachnospira and enrichment of Veillonellaceae and Prevotella members. However, smaller relative differentials were noted between former and never smokers (Fig. 2B) in most significantly differential taxa. Of note, unlike current smokers, the microbiome of former smokers demonstrated even lower relative abundance of Prevotella and Veillonellaceae Phascolarcobacterium than never smokers.

Shifts in relative abundance across significant taxa were largely consistent across all ethnic and racial subgroups, defined as non-Hispanic White or Black, Asian, and Hispanic (Fig. 2D). Notably, although we observed similar directional shifts in taxa between current smokers and never smokers in Asian subjects (Supplementary Table S3), these differences were more muted than in other ethnic subgroups. Post hoc subgroup ANCOM analysis showed only Phascolarcobacterium genus and Eubacterium biforme species to be significantly differentially abundant between current and never smokers in Asian subjects, with increased abundance in current smokers (Fig. 2D). Shifts in taxonomic abundance with smoking were directionally similar across sexes, BMI subgroups, and fiber intake subgroups (Supplementary Table S4).

We evaluated whether differences in taxonomic abundance varied by smoking dose or time since smoking cessation, selecting taxa noted to be significantly differentially abundant by our ANCOM analysis (Fig. 3; Supplementary Table S2). When we examined the normalized differential taxonomic abundance between either current or former smokers and never smokers based on years since quitting (Fig. 3A; Table 2), we found that the taxa which were relatively enriched in the microbiota of current smokers, depleted to nearly the level of nonsmokers in those who reported smoking cessation of over 30 years. This included depletion in Veillonellaceae taxa Acidaminococcus and Phascolarcobacterium, and Prevotella copri. Only an ASV identified within Oscillospira remained persistently enriched in all former smokers. Similarly, we found that taxa which were depleted in current smokers relative to never smokers, like Lachnospira and Erysipelotrichaceae, enriched to near nonsmoker abundances in formers smokers who reported cessation of over 30 years. The Tenericutes phylum remained persistently depleted in all former smokers.

Similar differentials in relative taxonomic abundance were noted when comparing current smokers by cigarettes smoked per day with nonsmokers (Fig. 3B; Table 2). We again noted depletion of Veillonellaceae and Prevotellaceae, and enrichment of Lachnospiraceae and Erysipelotrichaceae with decreasing cigarettes-per-day dosing. Oscillospira and Peptostreptococcaceae abundance did not differ significantly with cigarettes smoked daily (Table 2).

We performed an ANCOM analysis of imputed metabolic pathways relative to smoking status to further understand how alterations in gut microbiome affected gut function. In current smokers versus former, or never smokers, we found relative enrichment of oxidative and degradative metabolism superpathway members, such as lactose and galactose degradation or formaldehyde oxidation, and depletion of biosynthesis superpathway members, such as peptidoglycan biosynthesis (Fig. 4A). Pathways outside of these superpathway classes, like glycolysis and reductive acetyl coenzyme A pathway, were also increased in current smokers relative to former/never smokers.

We then correlated normalized imputed pathways with normalized taxonomic abundance of taxa previously identified as significantly differentially abundant by ANCOM (Fig. 4B), as in our smoking cessation and dose analysis. We found that taxa depleted in current smokers relative to never or former smokers negatively correlated with imputed oxidative/degradative superpathway members and unclassified pathways. An ASV from Peptostreptococcaceae did not conform to this trend in pathway correlation, despite showing significant depletion in current smokers. Of smoking depleted taxa, only the order Pasteurellales demonstrated significant positive correlation with biosynthetic pathways. Taxa enriched in current smokers, specifically an Oscillospira ASV and the genus Acidaminococcus, demonstrated positive correlation with oxidative/degradative superpathway members and unclassified pathways, including allantoin degradation and glycolysis (Fig. 4B).

This work demonstrates that the β-diversity of the fecal microbiome differs in current and former smokers from never smokers. We found that previously observed phylum-wide shifts in Bacteroidetes and Firmicutes were broadly similar across racial subgroups. These data suggest that smoking associated shifts persist much longer than previously hypothesized, with most taxa normalizing only after approximately 30 years cessation of tobacco use. These suggest underlying long-term changes in the microbial niche, which have not previously been described. Further, these data demonstrated novel smoking differentials in fecal taxonomic abundance, especially in the Veillonellaceae family and Tenericutes phylum.

Enrichment in oral Veillonellaceae in current smokers has been reported in multiple studies (2, 35, 36), yet no prior work had demonstrated changes in this taxa in the fecal microbiome. Our results are consistent with oral-fecal bacterial transfer, especially in the presence of environmental stressors such as tobacco use (37). Prior studies have also demonstrated an enrichment in Veillonellaceae family members in colon and lung cancer tissue (38, 39), suggesting an important role as a marker of dysbiosis and a possible role in carcinogenesis. The relationship between tobacco smoking and Tenericutes abundance is less clear. Two prior studies found enrichment of this taxa with smoking (3, 40), whereas we observed depletion. Although this may be due to differences in populations, further studies are necessary to clarify the effect of smoking on Tenericutes phylum abundance, and to determine whether these shifts play a role in smoking-associated gut pathogenesis.

Although several studies have demonstrated the importance of race and immigrant acculturation to microbiome composition (12, 41, 42), few have evaluated effect of smoking on the microbiome in a multi-ethnic population. Our findings, which show that smoking leads to similar shifts across these populations, suggest a commonality of mechanisms by which smoking impacts the gut microbiome. That these shifts are present in all racial and ethnic subgroups, with documented variations in diet and culture (43, 44), reinforces the magnitude of smoking-associated microbial differentials. Thus, common mechanisms of tobacco smoking related shifts may include the influence of smoking-associated nitrous oxide compounds (4), smoking-associated immune changes (45), and possible pathogenic content of cigarette tobacco (8).

Evaluation of the imputed metabolic pathways in current and former smokers suggests that relative taxonomic differentials may arise due to exposure to smoking-associated toxins and free-radical species. Taxa enriched in current smokers, specifically the genus Acidaminococcus and an Oscillospira ASV, were positively correlated with formaldehyde and allantoin degradation, and glycolysis V. Given the known formaldehyde content of cigarette smoke (46), documented elevations in plasma allantoin in smokers (47), and upregulation of glycolysis in response to reactive oxygen species (48, 49), our data suggest a potential selection mechanism for these taxa based on available metabolic inputs. These suggest that smoking-related toxins ROS may be present in the colon in sufficient quantities to affect microbial ecology, and that these taxa may play a role in the known association between smoking, ROS, and carcinogenesis (50).

A limitation of our investigation, as a cross-sectional study, is that the temporal relationship between smoking-related exposures and gut microbial outcomes cannot be directly tested. However, given that smoking is a behavior and the gut microbiome is an observed state, it is unlikely that alteration of the gut microbiome precedes or causes smoking. The study may also have a limitation in that smoking behavior is ascertained through questionnaire, which may be subject to recall and reporting bias, and that racial/ethnic differentials were evaluated with relatively small numbers in certain groups.

In summary, our research demonstrates that smoking is associated with profound and long-lasting differences in the gut microbiome, across a diverse population, which likely persist following smoking cessation. The observed differentials may also be associated with microbial upregulation of pro-inflammatory and carcinogenic pathways, and downregulation of biosynthetic pathways and homeostatic taxa. Longitudinal and direct-transcriptional follow-up studies are warranted to evaluate the mechanistic implications of our findings.

R.B. Hayes reports grants from NCI during the conduct of the study. No disclosures were reported by the other authors.

A. Prakash: Conceptualization, data curation, formal analysis, investigation, visualization, methodology, writing–original draft, writing–review and editing. B.A. Peters: Conceptualization, supervision, methodology, writing–review and editing. E. Cobbs: Resources, data curation, writing–review and editing. D. Beggs: Resources, data curation, writing–review and editing. H. Choi: Resources, data curation, writing–review and editing. H. Li: Methodology, writing–review and editing. R.B. Hayes: Conceptualization, resources, supervision, funding acquisition, project administration, writing–review and editing. J. Ahn: Conceptualization, resources, supervision, funding acquisition, methodology, project administration, writing–review and editing.

The authors would like to acknowledge Caroline Um for provision of fiber intake values for the FAMiLI study. This work was supported by the grant numbers P20 CA252728 and P30 CA016087 (to J. Ahn) and R01 CA159036 (to R.B. Hayes and J Ahn).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.