Characterisation of microbiota in saliva, bronchoalveolar lavage fluid, non-malignant, peritumoural and tumour tissue in non-small cell lung cancer patients: a cross-sectional clinical trial

Despite the advancements in its detection and treatment, lung cancer (LC) is still the leading cause of death by cancer worldwide [1]. Non-small cell lung cancer (NSCLC) is diagnosed in 85–90% of LC cases and presents the most frequent type of lung cancer. Unlike small cell lung cancer, NSCLC is operable in 20–25% of cases. This concerns mostly early stage tumours (stage I and II), sometimes locally advanced disease (stage III) and rarely oligometastatic disease (stage IV). Other treatments, such as chemotherapy, radiotherapy and until recently immunotherapy, are often associated with surgery as multimodality treatment. Even though surgery is recognised as the most effective initial treatment of NSCLC, the 5-year survival rates remain however low (~ 90% for stage IA1, and ~ 12% for stage IIIC) [2, 3]. Therefore, the tumour staging is used as the important prognostic tool based on tumour size, lymph node invasion and metastatic status [2]. Curiously, tumour lobe location has also been associated to tumours’ aggressiveness, with tumours in lower lobes (LL) showing worse term and 5-year survival after resection than the ones in upper lobes (UL), still without a clear explanation [4‐6].

Increasing interest in the interaction between host and its microbiota revealed its potential implication in health and disease, but also in tumour immunology and physiology [7‐11]. Certain bacteria, such as genus Bifidobacterium [8] or species Enterococcus hirae [10], have been found to improve the efficiency of chemotherapy or immune-checkpoint inhibitors if administered orally in animal models. This phenomenon has been explained by their translocation from the gut to mesenteric lymph nodes, the priming of the upstream regulatory immune cells, such as dendritic cells, and causing increased reactivity against tumour epitopes [12, 13]. Moreover, administration of Bifidobacterium cocktail alone has been proved equally effective as the anti-PD-1 (Programmed cell Death protein 1) antibody in abolishing tumour growth in the animal melanoma model [8]. Finally, faecal transplantation from the patients responding (enriched in Bifidobacterium, Akkermansia, Faecalibacterium) or not (enriched in Bacteroidales) to the anti-cancer treatment has induced the same kind of response in receiving tumour-bearing animals [8, 14]. These findings have been one of the most elucidating in terms of interactions between the host’s immune system, gut microbiota and cancer.

Unlike local and systemic effects of the gut microbiota, the lung microbiota and its effects remain scarcely explored, being only recently accepted as one of the resident microbiota (and not only present during infection) [15, 16]. Since then, an emerging number of studies turned to its exploration, notably in the context of cystic fibrosis, asthma and chronic obstructive pulmonary disease (COPD), interstitial lung disease, and lung transplantation [17‐22]. Despite its impact on global cancer-related death, lung cancer studies were surprisingly few and started to emerge only a few years ago. However, they confirmed that lung microbiota interacts with local immunity and modifies tumour properties. The microbial dysbiosis in antibiotic-treated or germ-free animals influenced growth of injected lung tumour cells [23, 24] while usage of penicillin, cephalosporins, or macrolides showed increasing risk of lung cancer in human subjects [25]. In lung cancer patients, lung microbiota from bronchoalveolar lavage fluid (BAL) enriched with supraglottic taxa was associated with pro-inflammatory profile and stimulation of Th17 cells with protumourigenic effect [26‐28], and also exhibited different abundance and metabolic profiles compared to those of healthy subjects [29, 30]. Interestingly, salivary microbiota was also found to show cancer specific profile, with genera Veillonella and Capnocytophaga more abundant in saliva of lung cancer patients [31]. At the present, only two studies analysed lung tissue microbiota in lung cancer. One found increased alpha diversity in non-malignant tissue compared to tumours as well as in adenocarcinoma compared to squamous cell carcinoma [32], while the other showed association between increased diversity of the non-malignant tissue (but not tumour) and decreased recurrence-free and disease-free survival [33]. Among studies on lung microbiota, those on BAL are the most numerous, since it remains the sample with acceptable ratio of contamination risk by upper airways, precision in lung microbiota sampling and invasiveness. However, this has been a potential source of contradictory information since varying characteristics of BAL and tissue microbiota, as a result of samples’ different nature, have been previously suggested [21]. Therefore, there has been an increasing necessity to characterise the ground differences between different lung microbiota in NSCLC patients to enable better comprehension of the obtained results depending on the initial lung sample.

As its primary objective, this cross-sectional pilot study analysed lung microbiota from four different samples in 18 NSCLC patients eligible for surgery without neoadjuvant therapy. Lung microbiota was analysed in BAL, non-malignant tissue, peritumoural tissue and tumour, as each sample should have different architectural and physiological characteristics. Unlike what was previously seen, in this study BAL was obtained directly from the excised lobe without passing through the upper airways to decrease the contamination risk. In addition to lung microbiota, salivary microbiota was characterised for each patient and used as an extra-pulmonary sample to put in perspective the relation with and between lung samples. As a second objective, we investigated whether tumour location in the UL or LL yields significant changes in these microbiota.

Following the last advancement on the interaction between gut microbiota, immune system and the tumour environment [8‐11], there has been a growing interest in studying this concept in other physiological environments involving extraintestinal tumours. However, before exploring the effect of the gut microbiota, there has been an increasing necessity to investigate the effect of the local microbiota on the tumour as well [42]. Despite the emerging number of studies on the lung microbiota in different malignancies [17, 43‐46], its involvement in lung cancer is in its promising beginnings [26, 30, 32, 33, 47]. However, for the moment there is still no study considering the ground difference between different lung samples and their microbiota, while it is suggested that those could harbour microbiota with varying characteristics [21], and therefore, have diverging interactions with local immunity and tumour.

This study is, to our knowledge, the first to characterise the lung microbiota originating from four different lung samples (BAL, non-malignant tissue, peritumoural tissue and tumour), accompanied with the characterisation of salivary microbiota in NSCLC patients. We hypothesised that samples’ nature, “architecture”, physiological functions and environment, will influence characteristics of associated lung microbiota. Therefore, BAL should represent “planktonic” bacterial population found within the bronchial lumen or associated with biofilms or mucus [48], sampled along with the hydrodynamic force of the instilled liquid. Non-malignant tissue from the same lobe but taken on the opposite side from the tumour should represent a sample with a normal lung architecture, with well-defined small alveolar spaces and single-cell epithelial layer. In majority, it should harbour the biofilm-, mucus- and cell-associated lung microbiota. On the contrary, tumour represents a tissue with disrupted architecture, varying in form and obstruction degree relative to its type and grade [2]. Tissue modelling could also involve overproduction of the mucus as seen in certain subtypes of adenocarcinoma [49], but also different reaction of the immune system [50]. Finally, peritumoural tissue represents a non-malignant tissue in the direct contact with the tumour, separated as based on different histological properties. In literature it is addressed as tumour microenvironment and harbours different roles in stimulation or suppression of tumour metabolism [51]. Therefore, we presumed that its characteristics would be different both from the distal non-malignant tissue and the tumour. Except for the number of different lung samples, the particularity of this study is also the way of obtaining BAL. It was obtained directly on the excised lobe containing tumour without using the bronchoscope, to minimise contamination risk of upper airways and increase the precision in characterisation of “true” BAL microbiota in the tumour proximity.

We reported that the four lung samples significantly clustered versus oral microbiota based on beta diversity, a confirmation of a previous result in healthy subjects that lung microbiota is distinct from other communities [32]. Salivary microbiota had several lower alpha diversity metrics compared to lung microbiota, as well as high dominance of genus Streptococcus and overall phylum Firmicutes. Along with its homogenous core microbiota (again prevalence of Streptococcus OTUs) that represented almost ¾ of total relative abundance per patient, this indicated its inter-subject stability and lower complexity compared to the one of the lung samples. In addition, the significant difference in abundance including the majority of the taxa detected in samples clearly separated the four lung samples versus oral microbiota. We identified typically oral taxonomic groups more abundant in saliva in all comparisons, such as Fusobacteria, Spirochaetes, Synergistetes, Erysipelotrichi, Epsilonproteobacteria, Bacilli and Neisseriales, most of them being in concordance with previous literature [52, 53]. On the other hand, phyla Acidobacteria, Cyanobacteria, Nitrospira, Verrucomicrobia and Planctomycetes were detected as strictly lung-associated (detected in each lung sample) with no representatives in saliva.

Further, we showed that BAL microbiota, even though undoubtedly belonging to lung microbiota cluster, had significantly different features from tissues. First, multidimensional representation of the beta diversity placed BAL samples on the side of the lung cluster towards salivary microbiota (Fig. 1a). Even though presence of more abundant salivary taxa was very limited in both tissue samples and BAL, Streptococcus, Prevotella and Veillonella (the three most abundant genera in saliva) were found elevated in BAL compared to tissue samples. It was previously suggested that these and other typically oral bacteria are found in low abundance in healthy lungs, due to their constant elimination [54]. Since BAL represents microbiota of the bronchial lumen, which undergoes constant influx of upper airways particles by respiration and is also the first “space” influenced by microaspiration [55], those could explain increased presence of these supraglottic taxa in BAL compared to the lung tissues. This position “in the middle” was also seen when looking at the differential abundance between all five samples (Fig. 5). The taxa that were more abundant in saliva when compared to the four lung samples almost perfectly corresponded to taxa that were more abundant in BAL when compared to the three lung tissues. Second, lung samples had Proteobacteria and Firmicutes as the two most abundant phyla, however with different ratios. While BAL was dominated by Firmicutes, almost twice as abundant as Proteobacteria, lung tissues were dominated by Proteobacteria (as previously reported [32]), or at best, had equal abundance of the two (peritumoural tissue). Here it is however important to note that the high abundance of phylum Firmicutes in lung samples was due to members of the class Clostridia, unlike in saliva, where it was due to highly abundant class Bacilli. This emphasises one of the essential differences between oral and lung microbiota often omitted by selective presentation of only phylum or genus level. A third feature was that BAL did not differentiate from lung tissues only by difference in taxa abundance, but also by the presence of BAL-specific bacteria, such as genus Capnocytophaga, or absence of tissue specific-bacteria, such as genus Kocuria or Clostridium. The two latter genera were, however, detected in BAL, but with relative abundance 175-fold and 520-fold lower (~ 0.03%), respectively, than in the tissues. This detection is possibly due to their presence in host’s cells during the DNA extraction, since cellular BAL was used for BAL analysis [48]. Lastly, no differences in alpha diversity metrics were detected between four lung samples, suggesting two different but equally “rich” lung microbiota populations. All this supports the hypothesis that BAL indeed represents a unique lung microbiota in lung cancer and that concerns of diverging results due to different samples have been justified [21].

Conversely to our hypothesis, there was no significant difference detected between the three lung tissues when analysing the totality of samples. Although increased diversity of non-malignant tissue versus tumour has been previously suggested [32], we did not find the same. This could be due to the fact that our study group was more balanced in patients with less and more advanced tumour stages. In a referenced study, the majority of subjects had tumours in stages I and II, showing decreased diversity compared to higher stages according to authors.

We also reported on the existence of the core OTUs and that these were mostly shared between lung tissues and partially with BAL, consisting in the majority of the members of Proteobacteria. Interestingly, one of the core OTUs was uniquely associated to species Variovorax paradoxus, an ambiguous aerobic organism found in many niches, also in oral microbiome, and known for its potent biofilm creation [56, 57]. It was detected as the core OTU in all four lung samples, while conversely, it was not found in saliva (where it was originally detected). This could suggest its potential role in biofilm creation in lower airways and would be an interesting topic for future verification and investigation of potential role in pathogenesis of lung cancer [58].

Even though still under debate, several studies connected lower lobe tumours with worse prognostics [5, 59, 60]. Therefore, we examined the characteristics of the lung, but also salivary microbiota, if the tumour was found in upper or lower lobes. Indeed, we noted several changes in alpha and beta diversity but they were focused on changes in BAL and peritumoural tissue (with lower and higher alpha diversity, respectively, in LL compared to UL). We reported that in all four lung samples, lower lobes had decreased abundance of phylum Proteobacteria into favour of Firmicutes. Increased abundance of phylum Firmicutes was previously seen in BAL of LC patients compared to patients with benign-mass lesions [47], but also in patients with COPD [21]. In our study, 62% of patients in the LL were diagnosed with stage III tumours, against 30% in the UL. Even though this remains a small study, this finding supports that “shift” from Proteobacteria to Firmicutes might have a role in lung cancer progression.

We showed that location significantly influenced the abundance of two clades, Actinobacteria and Flavobacteriia, in the inverse manner between BAL and extratumoural tissues. While both are more abundant in UL BAL, their abundance in UL extratumoural tissues was 2–10-fold lower, with exact inverse situation in LL. Both genus Chryseobacterium sp. (Flavobacteriia) and Kocuria rhizophila (Actinobacteria) detected as more abundant in LL tissues have been previously reported as uncommon human pathogens [61, 62]. However, their increased presence selectively in LL tissues and their important overall abundance (~ 4 and 12%, respectively) suggest that there might be an important communication between different lung environments depending on local conditions or malignancy status, influencing preferential bacterial growth in one type of considered environment.

Interestingly, we found significant differences in abundance of salivary microbiota between patients with tumour in either upper or lower lobe. All differences considered taxa elevated in the case of LL tumours and included families previously associated with bacterial exacerbations and infections (Moraxellaceae, members of Bacilli as Streptococcus, Staphylococcus, etc.) [63] or pulmonary complications (Propionibacteriaceae, Enterococcaceae) [64‐66]. However, these taxa were not found elevated in LL lung microbiota. This, however, does not exclude their implication in malignant changes or being its consequence, and will be an interesting topic for further consideration.

Finally, we found that peritumoural tissue showed higher similarity to BAL in UL both in beta diversity and in abundance, while in LL it shared characteristics with other tissue samples. Moreover, we found that alpha diversity of peritumoural tissue increased in LL to the level of other tissues, while the ones of BAL decreased. Peritumoural tissue is the tissue in direct contact and interaction with the tumour. In the history of cancer research, the role of extracellular matrix (ECM) surrounding tumour has been extensively studied [67]. Tumour cells are found to be able to directly influence the rearranging of connective units (such as collagen) and degradation of ECM to create a tumour-permissive environment and enabling metastatic progression [68‐70]. In our case, it is possible that the observed changes in peritumoural tissue composition reflect certain remodelling of ECM, which could therefore be either more or less permissive for microbial attachment.

Curiously, of all samples, tumour was almost completely uninfluenced by location. Rather than from the outside, it is possible that intratumoural conditions, such as oxygen availability, density, necrosis and other factors [71, 72] independent of the external conditions are more likely to be the influencing ones. This is, however, a matter for further research.

The major strongpoint of this study is the analysis of the microbiota from four different lung samples covering the major physiologically different environments in NSCLC patients that has not been characterised previously, with the addition of saliva as the sample of oral microbiota. The representation using taxonomical trees also gives better insight into sample’s composition, especially important in characterisation of this type of as yet, scarcely defined microbiota. Moreover, this study reports higher numbers of detected OTUs as well as diversity indexes than previously suggested for saliva, non-malignant tissue, tumour and BAL [29, 32, 33]. The reason could be due to the higher quality of obtained samples (only 6 from 85 samples lower than 1000 reads) and direct sampling of the BAL in the excised lobe. Also, the observed study group and stratification by tumour lobe location were well balanced in various clinical and demographic factors, minimising the result’s bias.

The major drawback of this study is a low number of subjects, due to strict inclusion criteria for the overall protocol and a limited recruiting time. Additional inconvenience is that, even though initially requested, forming a control group has not been authorised by the Ethics Committee due to the invasiveness of the sampling techniques. Certain studies on lung cancer microbiota have adapted different strategy to include a non-malignant control group in their protocols, e.g. prospective studies analysing BAL from patients with pulmonary nodules that are next characterised as either benign or malignant (i.e. lung cancer) [47]. However, patients diagnosed with benign pulmonary nodules represent the minority, in most cases they do not require any treatment and partial resection is undertaken only in rare occasions [73, 74]. With the objective to analyse the lung microbiota from multiple lung samples (BAL in situ and different tissues), the partial resection in benign lung nodules would exclude both the possibility of performing BAL directly on the excised lobe and obtaining different types of lung tissue relative to the distance from the nodule. Moreover, it is unclear for the moment whether the presence of benign pulmonary nodule also modifies the local lung microbiota, introducing a potential bias in interpretation without previous investigation. Considering other surgically treatable lung diseases, such as interstitial lung disease [75], collapsed lung [76], emphysema or bronchiectasis [77], they represent malignancies with a distinct physiological condition and were not considered for use as a potential control group. The ideal control group would be the one including lobectomy in “healthy” patients, as seen in certain transplantation cases [78].

The limiting factor was also the decision to obtain BAL directly from the excised lobe to improve the precision in characterisation, but disabling the possibility of sampling the complementary non-tumour lobe by bronchoscopy due to difference in sampling technique.

Finally, it is important to note that the surgical resection considers fewer than 20% of all lung cancer patients and includes less severe cases [79]. Therefore, our results cannot be extrapolated to the overall lung cancer community, especially the most advanced cases relying mostly on multimodality treatment [2].