Association of oral microbiota with lung cancer risk in a low-income population in the Southeastern United States

Abstract

Purpose

Oral microbiome plays an important role in oral health and systemic diseases, including cancer. We aimed to prospectively investigate the association of oral microbiome with lung cancer risk.

Methods

We analyzed 156 incident lung cancer cases (73 European Americans and 83 African Americans) and 156 individually matched controls nested within the Southern Community Cohort Study. Oral microbiota were assessed using 16S rRNA gene sequencing in pre-diagnostic mouth rinse samples. Paired t-test and the permutation multivariate analysis of variance test were used to evaluate lung cancer risk association with alpha diversity or beta diversity, respectively. Conditional logistic regression models were used to evaluate the association of individual bacterial abundance or prevalence with lung cancer risk.

Results

No significant differences were observed for alpha or beta diversity between lung cancer cases and controls. Abundance of families Lachnospiraceae_[XIV], Peptostreptococcaceae_[XI] and Erysipelotrichaceae, and species Parvimonas micra was associated with decreased lung cancer risk, with odds ratios (ORs) and 95% confidence intervals (CIs) of 0.76 (0.59–0.98), 0.80 (0.66–0.97), 0.81 (0.67–0.99), and 0.83 (0.71–0.98), respectively (all P<0.05). Prevalence of five pre-defined oral pathogens were not significantly associated with overall lung cancer risk. Prevalence of genus Bacteroidetes_[G-5] and species Alloprevotella sp._oral_taxon_912, Capnocytophaga sputigena, Lactococcus lactis, Peptoniphilaceae_[G-1] sp._oral_taxon_113, Leptotrichia sp._oral_taxon_225, and Fretibacterium fastidiosum, was associated with decreased lung cancer risk, with ORs and 95% CIs of 0.55 (0.30–1.00), 0.36 (0.17–0.73), 0.53 (0.31–0.92), 0.43 (0.21–0.88), 0.43 (0.19–0.94), 0.57 (0.34–0.99), and 0.54 (0.31–0.94), respectively (all P<0.05). Species L. sp._oral_taxon_225 was significantly associated with decreased lung cancer risk in African Americans (OR [95% CIs]: 0.28 [0.12–0.66]; P=0.00012).

Conclusion

Results from this study suggest that oral microbiota may play a role in the development of lung cancer.

Introduction

Lung cancer is the most common cause of cancer mortality and one of the most common cancers in both men and women worldwide [1, 2]. Cigarette smoking is attributed to approximately 80% of the lung cancer burden in men and more than 50% in women [3, 4]. Other lung cancer risk factors include alcohol drinking, unhealthy diet, physical inactivity, history of lung diseases, and environmental and occupational exposures. However, there is a need to better understand alternative risk factors, particularly with regard to how they interact with known risk factors as well as the underlying mechanisms of carcinogenesis.

Oral microbiome plays an important role in oral health and systemic diseases, including cancer [5–7]. Dysbiosis of oral microbiota has been proposed to contribute to cancer development via various mechanisms, including inducing chronic inflammation, inhibiting apoptosis, activating cell proliferation, promoting cellular invasion, and/or producing carcinogens [8, 9]. Regarding lung cancer, several lines of observational findings have suggested a possible association between oral microbiota and risk of the disease. First, an increased risk for lung cancer has been observed in patients with periodontal diseases [10–12], which is likely caused by oral microbial pathogens and dysbiotic oral microbiota [13]. Second, individuals carrying specific oral bacteria, such as Chlamydia pneumoniae, have an increased risk for lung cancer [14–16]. Third, differential abundance of certain oral bacteria were reported to be associated with lung cancer in case-control studies [17–20]. However, no study has prospectively investigated the association of oral microbiome with lung cancer risk in low-income populations, especially among African Americans (AAs).

In the present investigation, we conducted a case-control study nested within the Southern Community Cohort Study (SCCS), a low-income population living in the Southeastern United States, to systematically evaluate the associations between oral microbiota and lung cancer risk among AAs and European Americans (EAs).

Materials and methods

Study participants and data collection

The SCCS is an ongoing prospective cohort study, with the aims to investigate risk factors of cancer and chronic diseases among a low-income population, described in detail elsewhere [21]. Briefly, approximately 86,000 adults, two-thirds of whom were AAs, were recruited between March 2002 and September 2009 from 12 Southeastern states of the United States. Approximately 86% were recruited from community health centers (CHCs), institutions providing basic health care and preventative services in underserved areas, so that the cohort included a large number of individuals of low income and educational status. The remaining 14% of cohort members were recruited through mail-based general population sampling. During enrollment, mouth rinse samples were collected from ~34,100 participants [22]. The SCCS was reviewed and approved by the institutional review boards at Vanderbilt University and Meharry Medical College. Written, informed consent was obtained from all study participants.

In the baseline survey, participants completed a comprehensive questionnaire, used to collect information on anthropometric characteristics, lifestyle factors, disease history, medication use, and other characteristics. Passive cohort follow-up by record linkage to state cancer registries operating in the 12-state study area and the national death index registry started immediately upon completion of the baseline survey. Active follow-up surveys started in 2008. In mailed or telephone follow-up surveys, participants were asked about their personal medical histories and medication use. Cigarette smokers were defined as those who reported smoking at least 100 cigarettes in their lifetime. Former smokers were defined as those who had quit smoking for at least one year, whereas those who had quit for less than one year were considered current smokers. Lung cancer cases were defined according to the International Classification of Diseases (ICD-10) for Oncology, Second Edition (ICD-O-2) and included all invasive cancers coded as C34.0-C34.9.

For this study, we selected 166 incident lung cancer cases and 166 individually matched controls who donated mouth rinse samples at enrollment. For each incident lung cancer case, one control was randomly selected and individually matched to cases by age of enrollment (±2 years), race, sex, smoking status at baseline (never/former/current smoker), date of mouth rinse sample collection (± 90 days), CHC recruitment site, and recruitment source. We excluded six participants who used antibiotics ten or more times during the prior year or took any dose of antibiotics during the past week before mouth rinse sample collection. Participants with a history of HIV infection, other cancers, diabetes, stroke, or myocardial infarction were also excluded.

During enrollment, mouth rinse samples were collected from participants, who were asked to swish vigorously for 45 seconds with 10 mL Scope mouthwash containing a 15 wt% alcohol content (Procter & Gamble, Cincinnati, OH), then expectorate into a specimen container. Mouth rinse samples were shipped to the Molecular Epidemiology Laboratory at Vanderbilt University Medical Center for processing and stored at −80°C until DNA extraction. The Mean time between mouth rinse collection and lung cancer diagnosis is 3.5 years.

DNA extraction and 16S rRNA gene sequencing

Total DNA, including bacterial DNA, was isolated from mouth rinse samples using the QIAmp DNA kit (Qiagen, Germantown, MD, USA). Sequencing libraries were prepared using the NEXTflex 16S V4 Amplicon-Seq Kit (Bioo Scientific 4201–05), following the manufacturer’s protocol. This kit was designed to sequence approximately 253 bp of the fourth hypervariable (V4) domain of the 16S rRNA gene [23, 24]. Pair-end reads of 250 bp were obtained using the Illumina HiSeq System. Each 96-well plate was sequenced with one negative control (i.e., distilled water) and two duplicate quality control (QC) samples. In this study, each of the two discrete QC samples comprising of mixed mouth rinse DNA samples of de-identified volunteers was sequenced four times. Comparable microbial profiles were observed for the same QC samples. For example, for alpha diversity measurements within each sample, using Shannon and phylogenetic diversity (PD) whole tree indexes, the average coefficients of variability among repeated QC samples were 2.0% and 4.8%, respectively. For individual species-level taxa, Pearson correlation coefficients among QC samples ranged from 97.9% to 99.9%, with a median of 99.6%.

Sequence data analysis and quality control

Raw sequencing reads, ranging from 36,323 to 351,766 (with a mean of 114,420 and standard deviation of 37,394) among the participants, were trimmed and filtered to remove bases and reads of low quality using the Sickle tool [25]. An average of 9,827 reads (standard deviation=10,831) were filtered. Then, BayesHammer [26] was utilized for correcting sequencing errors, and PANDAseq [27] for stitching paired-end reads [28]. Clean reads were then clustered into Operational Taxonomic Units (OTUs) at 97% sequence identity, using the closed reference OTU picking strategy with the Human Oral Microbiome Database (HOMD) [29] as reference via the Quantitative Insights Into Microbial Ecology (QIIME), v1.9.1 [23]. Four participants with clean sequencing reads <20,000 were removed. We excluded ten matched case-control pairs from which any paired participant either used antibiotics (n=6) or had low sequencing reads (n=4). A total of 156 lung cancer case-control pairs were included in the final analysis.

Statistical analyses

The QIIME version 1.9.1 [23] was used to rarefy the OTU/species table at a sequencing depth of 20,000 and estimate observed bacterial OTUs and alpha diversity indices including Chao1, Shannon, and PD whole tree. Paired t-test was used to compare bacterial richness and alpha diversity between lung cancer cases and matched controls. Beta diversity (the total variance of an oral microbial composition) between lung cancer cases and controls was compared by the Permutation Multivariate Analysis of Variance (PERMANOVA-S) test implemented in the miProfile software [30]. This software can construct and produce P-values for each of the six beta diversity matrices (Jaccard, Bray-Curtis, presence-weighted UniFrac with parameter 1 or 0, unweighted UniFrac and weighted UniFrac) and a unified P-value by combining all abundance and presence-absence distances [30], adjusting for unmatched covariates including smoking pack-years, alcohol intake (none, light, moderate, or heavy; by sex), total energy intake, body mass index (BMI), last time of dentistry visit, and sequencing batch.

Associations between individual bacterial taxa at phylum, family, genus, and OTU levels and lung cancer risk were evaluated by two approaches based on taxonomic relative abundance and prevalence (carriage frequency). First, taxa with a median relative abundance >0.01% among control participants were selected for abundance association with lung cancer risk. There were 119 such taxa, including six phyla, 23 families, 33 genera, and 57 species. At each taxonomic level, the sequencing read counts for each taxon were normalized using centered log-ratio transformation after adding 1 as a pseudo-count [31]. Conditional logistic regression was conducted in the stratum for matched cases and controls, adjusting for the above-mentioned unmatched covariates for beta diversity associations. Second, taxa with a median relative abundance ≤0.01% and a prevalence of >20% among control participants were selected for prevalence association with lung cancer risk. For 219 such taxa, including three phyla, 19 families, 50 genera, and 147 species (total n=219), conditional logistic regression models were used to compare the prevalence of carriers of each taxon among lung cancer cases and controls, adjusted for the above-mentioned covariates and sequencing depth. In addition, taxonomic prevalence in associations with lung cancer risk were also tested for five pre-defined oral pathogens, including P. gingivalis, Treponema denticola, Tannerella forsythia, A. actinomycetemcomitans, and Prevotella intermedia [32, 33].

Stratified analyses by race, sex, smoking status (ever/never smoking), and time between mouth rinse sample collection and lung cancer diagnosis (<2 years: n=94; >2 years, n=62) were conducted for both abundance and prevalence associations. Adjusting for diet quality score (based on the healthy eating index 2010 edition [34]) rather than total energy intake in above logistic regression models did not materially change the results. Thus, results without adjustment for healthy eating index are reported. We did not collect detailed oral health information at baseline enrollment. We used time since last dentistry visit as a proxy variable for oral health, which was adjusted in the regression models for association analyses between oral microbiota and lung cancer risk.

Considering the fact that taxa of different taxonomic ranks are correlated, we used Galwey’s method [35], implemented in the function “meff” of the R package “poolR” [36], to estimate the effective number of independent tests, used for Bonferroni correction. All of the analyses were carried out using SAS statistical software (SAS Enterprise Guide 7.1, SAS Institute Inc.) or R version 3.4.3 unless otherwise indicated.

Results

Characteristics of the study subjects

Table 1 presents the distribution of selected demographic characteristics of the participants. Lung cancer cases and controls were well matched with respect to age, race, sex, and smoking status. Compared with controls, ever-smoking participants with lung cancer had significantly higher pack-years (P=7.1×10⁻⁴). No significant differences were observed in the distributions of alcohol drinking status, family history of lung cancer, dental visit, healthy eating index, or BMI between case and control groups.

Table 1.

Selected characteristics of lung cancer cases and controls in the Southern Community Cohort Study

Characteristics	Cases (n=156)	Controls (n=156)	P a
Race (n [%])
European American	73 (46.8%)	73 (46.8%)	1.000
African American	83 (53.2%)	83 (53.2%)
Sex (n [%])
Male	82 (52.6%)	82 (52.6%)	1.000
Female	74 (47.4%)	74 (47.4%)
Age at sample collection (n [%])
40–49 years	35 (22.4%)	38 (24.4%)	0.964
50–59 years	55 (35.3%)	56 (35.9%)
60–69 years	45 (28.8%)	43 (27.6%)
≥70 years	21 (13.5%)	19 (12.2%)
Body mass index (n [%])
<18.5	9 (5.8%)	3 (1.9%)	0.260
18.5–25	54 (34.6%)	51 (32.7%)
25–30	54 (34.6%)	54 (34.6%)
≥30	39 (25.0%)	48 (30.8%)
BMI (kg/m2)b	27.3 ± 6.8	28.3 ± 6.5	0.176
Smoking status (n [%])
Current smoker	90 (57.7%)	90 (57.7%)	1.000
Former smoker	43 (27.6%)	43 (27.6%)
Never smoker	23 (14.7%)	23 (14.7%)
Smoking pack-yearsb	39.7 ± 34.4	29.0 ± 22.8	7.1×10−4
Alcohol drinking (n [%])
None	82 (52.6%)	82 (52.6%)	0.477
Light, <1 drink per day	35 (22.4%)	46 (29.5%)
Moderate, 1 –2 drink/day	16 (10.3%)	13 (8.3%)
Heavy, >2 drinks per day	20 (12.8%)	15 (9.6%)
Missing	3 (1.9%)	0 (0%)
Total energy-intake (kcal/day)b	2517±1463	2368 ± 1243	0.296
Healthy eating indexb	55.2 ± 12.3	56.8 ± 12.8	0.436
Family history of lung cancer (n [%])
Yes	22 (14.1%)	16 (10.3%)	0.299
No	134 (85.9%)	140 (89.7%)
Last visit to a dentistry (n [%])
<6 months	34 (21.8%)	42 (26.9%)	0.418
6–12 months	25 (16.0%)	25 (16.0%)
12–36 months	24 (15.4%)	29 (18.6%)
≥36 months	62 (39.7%)	49 (31.4%)
Missing	11 (7.1%)	11 (7.1%)
Teeth lost due to decay or gum disease (n [%])
None	6 (3.8%)	6 (3.8%)	0.045
1 to 10	14 (9.0%)	44 (28.2%)
>10, not all	8 (5.1%)	30 (19.2%)
All	14 (9.0%)	17 (10.9%)
Missing	114 (73.1%)	59 (37.8%)

^aChi-squared test for categorical variables and paired t-test for continuous variables.

^bMean ± standard deviation (SD); smoking-packyears were calculated among ever-smokers.

Association of overall oral microbiome composition with lung cancer risk

Observed OTUs among lung cancer cases and controls are summarized in Figure 1A, separately for EA cases, EA controls, AA cases and AA controls. Among 708 OTUs, 534 were observed in all four groups, while 32, 29, 30, and 27 were unique among EA cases, EA controls, AA cases and AA controls, respectively. Compared with controls, lung cancer cases had lower observed OTUs (a mean ± standard deviation of 172.0 ± 54.9 vs 181.3 ± 56.5; P=0.09), Chao1 (195.2 ± 57.7 vs 204.8 ± 58.2; P=0.08), PD whole tree (8.5 ± 2.4 vs 8.9 ± 2.4; P=0.06), and Shannon (3.8 ± 0.8 vs 3.9 ± 0.6; P=0.31), respectively (Figure 1B). No significant differences were observed in any of the six distance matrices of beta diversity between lung cancer cases and controls (all P>0.1). Results of principal coordinate analyses (PCoA) of Bray-Cutis dissimilarity are presented in Figure 1C.

Fig. 1.

Observed operational taxonomic units (OTUs) and oral OTU-level bacterial diversity in lung cancer cases and controls. (A) Shared and unique OTUs observed among lung cancer cases and controls of European Americans and African Americans. (B) Comparing mean observed OTUs and three alpha diversity indexes (Chao1, PD whole tree, and Shannon) between lung cancer cases and controls (paired t-test, all P>0.05). (C) Principal coordinates analysis (PCoA) of Bray-Curtis beta diversity (P=0.93 from PERMANOVA-S, permutation number=1,000,000).

Associations of oral bacterial taxa with lung cancer risk

We evaluated the differences of abundance for 119 oral taxa between lung cancer cases and controls. Six taxa, all of them are within phylum Firmicute, were associated with reduced lung cancer risk at a P<0.05 (Table 2). Family Peptoniphilaceae, and its highly correlated descendants, genus Parvimonas and species P. micra (Pearson correlation coefficient >0.99), were associated with reduced lung cancer risk, with similar ORs (0.82–0.83). Three other families, including Lachnospiraceae_[XIV], Peptostreptococcaceae_[XI], and Erysipelotrichaceae were also associated with decreased lung cancer risk, with ORs (95% CIs) of 0.76 (0.59–0.98), 0.80 (0.66–0.97), and 0.81 (0.67–0.99), respectively. No significant associations were shown after Bonferroni correction for 41 independent tests for the 119 taxa. When tests were stratified by race, sex, or cancer diagnosed within 2 years after mouth rinse sampling, most of the associations largely remained (Supplementary Table 1). No significantly different associations between lung cancer subgroups were detected by formal tests of multiplicative interactions (data not shown).

Table 2.

Abundance association of oral bacterial taxa with lung cancer risk^a

Taxon	Median relative abundance (%)		OR (95% CI)b	P b
	Cases (n=156)	Controls (n=156)
Phylum Firmicute
Family Lachnospiraceae_[XIV]	0.205	0.226	0.76 (0.59–0.98)	0.037
Family Peptoniphilaceae	0.012	0.030	0.82 (0.69–0.97)	0.023
Genus Parvimonas	0.012	0.028	0.83 (0.70–0.98)	0.030
Species P. micra	0.012	0.028	0.83 (0.71–0.98)	0.031
Family Peptostreptococcaceae_[XI]	0.067	0.108	0.80 (0.66–0.97)	0.022
Family Erysipelotrichaceae	0.014	0.018	0.81 (0.67–0.99)	0.038

^aThe taxa had median relative abundance >0.01% among control subjects.

^bOdds ratios (ORs), 95% confidence intervals (CIs), and P values were calculated from conditional logistic regression on normalized taxa counts using centered log-ratio transformation, adjusted for unmatched covariates including smoking pack-years, alcohol drinking status, total energy intake, BMI, last time of dentistry visit, and sequencing batch.

Among 219 taxa for prevalence comparison between cases and controls, three genera and six species were associated with lung cancer risk at a P<0.05. These taxa showed around 10% to 15% lower prevalence in lung cancer cases compared with controls; however, none of the associations were significant after Bonferroni correction for 70 independent tests (Table 3). As shown in Supplementary Table 2, a stronger association was observed between the presence of species Leptotrichia sp._oral_taxon_225 and lung cancer risk in AAs (OR=0.28, 95% CI=0.12–0.66; P=1.2×10⁻⁴; Bonferroni-corrected P=0.006 for 48 independent effective tests), showing a significant interaction with race (P=0.019). The association of species Capnocytophaga sputigena was possibly caused by cancer development, as a more evident association was found for cases diagnosed within two years after mouth rinse sample collection (OR= 0.26, 95%CI=0.11–0.61; P=7.2×10⁻⁴; Bonferroni corrected P=0.034) compared with cases who were diagnosed more than two years after mouth rinse sampling (OR=1.06, 95%CI=0.43–2.57; Supplementary Table 2).

Table 3.

Prevalence association of oral bacterial taxa with lung cancer risk^a

	N (%) carriage
Taxon	Cases (n=156)	Controls (n=156)	OR (95% CI)b	P b
Phylum Bacteroidetes
Genus Bacteroidetes_[G-5]	39 (25.0)	55 (35.3)	0.55 (0.30–1.00)	0.044
Species Alloprevotella sp._oral_taxon_912	41 (26.3)	56 (35.9)	0.36 (0.17–0.73)	0.017
Species Capnocytophaga sputigena	66 (42.3)	89 (57.1)	0.53 (0.31–0.92)	0.021
Phylum Firmicute
Genus Lactococcus	23 (14.7)	39 (25.0)	0.43 (0.21–0.88)	0.017
Species L. lactis	23 (14.7)	39 (25.0)	0.43 (0.21–0.88)	0.017
Genus Peptoniphilaceae_[G-1]	17 (10.9)	34 (21.8)	0.43 (0.21–0.88)	0.028
Species P_[G-1]. sp._oral_taxon_113	17 (10.9)	34 (21.8)	0.43 (0.19–0.94)	0.028
Phylum Fusobacteria
Species Leptotrichia sp._oral_taxon_225	67 (42.9)	88 (56.4)	0.57 (0.34–0.99)	0.041
Phylum Synergistetes
Species Fretibacterium fastidiosum	46 (29.5)	66 (42.3)	0.54 (0.31–0.94)	0.026

^aThe taxa had median relative abundance ≤0.01% and carriage >20% among control subjects.

^bOdds ratios (ORs), 95% confidence intervals (CIs), and P values were calculated from conditional logistic regression with non-carriers as the reference, adjusted for unmatched covariates including smoking pack-years, alcohol drinking status, total energy intake, BMI, last time of dentistry visit, sequencing batch, and sequencing depth.

Associations of pre-defined periodontal pathogens with lung cancer risk

Among the five periodontal pathogens, prevalence of P. intermedia and P. gingivalis was correlated with a Spearman correlation coefficient of 0.72. The prevalence of species A. actinomycetemcomitans was 10.3% in lung cancer cases and 5.1% in controls, whereas a lower prevalence of the other four pathogens was found in lung cancer cases compared with controls. However, none of the differences were statistically significant (Table 4). Compared to controls, a nominally higher prevalence of A. actinomycetemcomitans was observed in lung cancer cases among males (13.4% vs 4.9%; P=0.041) and ever-smokers (9.8% vs 3.8%; P=0.021), and a lower prevalence of T. forsythia was observed in AA lung cancer cases (56.6% vs 74.7%; P=0.020). However, none of these differences were significant after Bonferroni correction (Supplementary Table 3).

Table 4.

Prevalence association of pre-defined periodontal pathogens with lung cancer risk

	N (%) carriagea
Periodontal pathogens	Cases	Controls	OR (95% CI)b	P b
	n=156	n=156
Aggregatibacter actinomycetemcomitans	16 (10.3)	8 (5.1)	2.35 (0.92–6.00)	0.066
Porphyromonas gingivalis	101 (64.7)	115 (73.7)	0.60 (0.34–1.05)	0.069
Prevotella intermedia	102 (65.4)	108 (69.2)	0.80 (0.47–1.38)	0.424
Tannerella forsythia	83 (53.2)	97 (62.2)	0.72 (0.40–1.31)	0.280
Treponema denticola	77 (49.4)	92 (59.0)	0.67 (0.37–1.21)	0.177

^aNumber and percentage of cases and controls carrying the periodontal pathogen.

^bOdds ratios (ORs), 95% confidence intervals (CIs), and P values were calculated from conditional logistic regression with non-carriers as the reference, adjusted for unmatched covariates including smoking pack-years, alcohol drinking status, total energy intake, BMI, last time of dentistry visit, sequencing batch, and sequencing depth

Discussion

In this nested case-control study among a low-income population, we found that abundance of six taxa and prevalence of nine taxa at OTU, genus, or family phylogenetic levels were nominally associated with decreased lung cancer risk. The prevalence of species L. sp._oral_taxon_225 was associated with decreased risk of lung cancer among AAs, after Bonferroni correction. Our findings warrant validation in independent large-scale studies to further understand the role of oral microbiota in lung cancer development.

We observed that abundance of six taxa were lower in lung cancer cases compared to controls. Similar patterns were observed in subgroup analyses stratified by race, sex, or whether lung cancer cases were diagnosed within 2 years after enrollment. The lower abundance of families Peptoniphilaceae (including its genus Parvimonas and species P. micra), Peptostreptococcaceae_[XI], and Erysipelotrichaceae were more evident among ever-smoking cases, whereas enriched in never-smoking cases compared with corresponding controls (Supplementary Table 1), which suggests a strong impact of cigarette smoking [37–39]. Studies have shown that the genus Parvimonas and its species P. micra was more abundant in smokers’ oral cavity than in non-smokers [38, 40, 41]. Species P. micra within the genus Parvimonas and family Peptoniphilaceae, a Gram-positive anaerobic cocci species, is commensal in the oral cavity and can cause periodontitis [42]. It is also related to infections of other organs including lung abscesses [43]. Oral Parvimonas and P. micra levels have been reported to be increased in oral squamous cell carcinoma [44–46], and decreased in colorectal cancer [47]. Although it is possible that oral bacteria such as P. micra can translocate to lung and cause cancer, additional studies are warranted to investigate the underlying mechanisms of lung carcinogenesis.

Decreased prevalence of several taxa were also observed in lung cancer cases. Such association pattern remained for most of these taxa when stratifying the participants by race, sex, time to diagnosis, or smoking status. Interestingly, the prevalence of the species L. sp._oral_taxon_225 was more significantly decreased in AA lung cancer cases compared to controls (44.6% vs 67.5%, P=1.2×10⁻⁴), whereas it was slightly decreased in EA cases (41.1% vs 43.8%, P=0.57). These results may suggest a possible race-specific role of this species in lung cancer development. In addition, L. sp._oral_taxon_225 was reportedly significantly abundant in dental caries-free children [48], and its affiliated genus Leptotrichia was abundant in pancreatic cancer- and liver cancer-free controls [49, 50]. Further studies are needed to confirm and disentangle the possible protective role of L. sp._oral_taxon_225 in lung cancer.

Increasing evidence has shown that oral pathogens can cause chronic inflammatory periodontal diseases, which are associated with an increased risk for cancers of oral cavity and other body sites [8, 9]. Among five pre-defined oral pathogens, lung cancer cases had a higher prevalence of A. actinomycetemcomitans compared to controls (10.3% vs 5.1%). A higher prevalence of A. actinomycetemcomitans in oral mouthwash samples has been found to be associated with increased risk of pancreatic cancer [49]. As a keystone pathogen for aggressive periodontitis, A. actinomycetemcomitans has been hypothesized to initiate Toll-like receptor signaling pathways [51], which are involved in inflammatory tumorigenesis [52–54]. We observed lower prevalence of other four periodontal pathogens in lung cancer cases. The lower prevalence of these pathogens in cases was unlikely to be caused by the cancer diagnosis, as similar results were also observed in those diagnosed within 2 years after sample collection. However, the associations of these four pre-defined oral bacterial pathogens with lung cancer risk were not statistically significant.

This prospective study, to date, is the largest in sample size to investigate the role of oral microbiome in the development of lung cancer. Other strengths of this study include the nested case-control study design, participants from a low-income population of both African and European ancestries, as well as the ability to match multiple sociodemographic factors and smoking behavior. Several limitations in this study should be pointed out. First, although lung cancer risk association patterns were the same for most microbial bacteria in both EAs and AAs, our study with 312 samples may be underpowered. Second, the present study lacked a systematic assessment of oral health at the baseline examination, when samples were collected. In this study, we used time since last dentistry visit as a surrogate measurement to control for confounding oral hygiene and/or oral disease. Studies among participants with more detailed oral health, including tooth loss, may further clarify the role of oral microbiota in lung cancer. Third, the 16S rRNA gene sequencing method and using QIIME V1 instead of QIIME V2 may have a limited accuracy to profile taxa at the OTU/species level [55, 56]. Further studies using metagenomics approaches and state-of-the-art analysis pipeline are warranted to confirm our findings.

In summary, we found that multiple oral bacterial families, genera, and species may be associated with risk of lung cancer in a low-income population, although most of the associations were not statistically significant after controlling multiple testing. Further studies with larger sample sizes and using metagenomics approaches, as well as in vitro and in vivo studies are warranted to investigate the role of oral microbiota in the development of lung cancer.

Data availability

Data and statistical codes used in the present study can be requested through the SCCS Online request System (https://ors.southerncommunitystudy.org).