Introduction

Lower respiratory tract infections (LRTI), including pneumonia, are the most important cause of morbidity in adults, accounting for 6.2% of the total disability-adjusted life years (Mathers et al., 2008). Moreover, LRTIs are responsible for 7.1% of deaths worldwide each year and mortality rates of up to 44% in elderly aged >85 years have been reported (Kaplan et al., 2003; Tichopad et al., 2013).

The incidence of LRTI peaks in young infants and in adults over 60 years of age (Jansen et al., 2009; Millett et al., 2013). In children, this high incidence can be explained by the naivety of their immune system; however, in elderly, the mechanisms of the heightened susceptibility to pneumococcal disease are still poorly understood. Immunosenescence, defined as age-related deterioration of both innate and adaptive immunity, seems to impair elderly individuals to elicit effective immune responses against invading pathogens, at least partly explaining their vulnerability to LRTI (Krone et al., 2014).

Although the etiology of LRTIs in many cases remains unknown, the most important causative bacterial pathogen for LRTI and particularly pneumonia is Streptococcus pneumoniae (File, 2003). Typically, this pathogen populates the upper respiratory tract (URT) asymptomatically in coherence with a vast and diverse community of other potential pathogenic and commensal bacteria, fungi and viruses. In general, during health, the so-called microbiome is presumed to be at an equilibrium, providing beneficial functions to the host, such as pathogen resistance (Cho and Blaser, 2012). However, acquisition of new bacterial or viral pathogens, environmental factors or immunological perturbations can potentially disrupt the equilibrium, leading to dysbiosis, pathogen overgrowth and dissemination, resulting in symptomatic infections like pneumonia (Bogaert et al., 2011; Dethlefsen et al., 2007).

Only few studies have addressed the possible etiologic influence of the URT microbiome on the development of LRTIs. The available studies typically include only small numbers of subjects without a healthy control population (Zhou et al., 2010; Bousbia et al., 2012). Also, different microbial niches are studied (Sakwinska et al., 2014) and research is focused on specific risk groups like HIV (human immunodeficiency virus)-infected patients (Iwai et al., 2012) or patients infected with the pandemic H1N1 influenza virus (Chaban et al., 2013; Leung et al., 2013). Nevertheless, microbiota studies in chronic respiratory diseases like chronic obstructive pulmonary disease (COPD) and asthma suggests an important role for bacterial communities, both commensals and potential pathogens, in the pathogenesis of respiratory diseases (Hansel et al., 2013; Molyneaux et al., 2013).

As the oropharynx is the niche that can be regarded as the reservoir from which potentially pathogenic bacteria may spread to the lower respiratory tract, we hypothesized that pneumonia in adults and elderly is associated with imbalanced oropharyngeal bacterial communities leading to lack of containment and overgrowth of specific bacterial species (Morris et al., 2013). Therefore, we investigated the oropharyngeal microbiome in a large group of elderly with pneumonia (n=100) using 16S-based sequencing and compared this with healthy community-dwelling controls (n=91). To validate our findings and to explore the effect of age, we sought to replicate our results in a cohort of young adult pneumonia patients (n=27) and healthy controls (n=187).

Materials and methods

Study population

Patients aged ⩾18 years with a suspected community-acquired pneumonia (CAP) presenting in two general hospitals in Tilburg, The Netherlands, were invited to participate in the study (Central Committee on Research involving Human Subjects, NL18747.041.07). In- and exclusion criteria were previously described (Huijskens et al., 2012), see also Supplementary Information S1. We studied the microbiota composition of the oropharynx in 100 patients aged ⩾60 years (‘elderly pneumonia’) and 27 patients aged <60 years (‘adult pneumonia’). For all pneumonia patients, the pneumonia severity index (PSI) was assessed and used to stratify patients in mild (PSI 1–3), moderate (PSI 4) and severe pneumonia (PSI 5; Fine et al., 1997). We additionally studied the microbiota composition of oropharyngeal samples obtained from 91 healthy community-dwelling elderly living in the western part of the Netherlands who participated in a pneumococcal carriage study (‘healthy elderly’; https://clinicaltrials.gov, NCT00744263).

Finally, we studied the bacterial community profiles of 187 healthy young adult parents (‘healthy adults’) participating in a bacterial carriage study and living in the western part of the Netherlands (https://clinicaltrials.gov, NCT00189020; van Gils et al., 2009). All samples were taken after the introduction of the seven-valent pneumococcal vaccine in the national immunization program in newborns (June 2006) and before the start of the CAPiTA-trial evaluating the efficacy of the 13-valent pneumococcal vaccine in elderly for prevention of CAP. Furthermore, all four groups were matched for season (that is, all inclusions took place between November and February).

Sampling and storage methods

Oropharyngeal samples of patients were taken before hospitalization at the emergency ward and samples of controls were taken during routine visits. Samples were obtained and transported according to protocol (Huijskens et al., 2012). Samples were subsequently frozen at −70 °C until further analyses.

Microbiota composition analysis

Bacterial DNA isolation and quantification

DNA was isolated from 200 μl sample as previously described (Wyllie et al., 2014), eluded in two aliquots of 25 μl elution buffer and stored at −20 °C until further analyses. Bacterial DNA was quantified for each sample by quantitative PCR (qPCR) as previously described (Bogaert et al., 2011; Biesbroek et al., 2012).

Amplicon library preparation

Generation of a PCR amplicon library was performed by amplification of the 16S ribosomal RNA gene V5–V7 hypervariable region as previously described (Bogaert et al., 2011), except templates containing ⩽10 pg μl−1 DNA were cycled 35 times instead of 30 times for adequate amplicon recovery. The library was sequenced in three runs with the 454 GS-FLX-Titanium Sequencer (Life Sciences, Roche, Branford, USA). These sequence data have been submitted to the NCBI GenBank database under accession number SRP055536.

Data processing

The raw sequences obtained were processed and classified using QIIME version 1.7 (Caporaso et al., 2010). Barcodes and primers were trimmed off, chimeric sequences were identified and removed using chimeraslayer, and the sequences were checked for their quality. Sequences with ⩾1 error in the primer or >1 error in the barcode, a sequence length of <200 and >1000 base pairs (bp), average quality window score <25, >6 ambiguous bp or >6 homopolymers were removed from the database. Sequences were clustered into operational taxonomic units (OTUs) using UCLUST at 97% similarity. Aligned sequences were taxonomically classified using the Bayesian RDP classifier. For each sample, α-diversity was determined by calculating diversity indices (number of observed species and Shannon) at a rarefaction depth of 1100 sequences (Magurran, 2004).

Nucleic acids extraction and qPCR for viruses and bacteria

qPCR was used to test for the presence of respiratory viruses and bacteria, including human bocavirus, polyomaviruses WU and KI, respiratory syncytial virus A and B, human influenza virus A and B, parainfluenza virus 1–4, human rhinoviruses, adenoviruses, human coronavirus OC43, NL63, HKU and 229E and human metapneumovirus. Nucleic acids were extracted from the oropharyngeal samples as previously described (van de Pol et al., 2007; 2009). Details on the probes, primers and PCR assay are described elsewhere (Huijskens et al., 2012). For the identification of Streptococcus pneumoniae, a qPCR was used targeting a conserved region of the autolysin (lytA) gene (Carvalho et al., 2007).

Statistical analysis

Differences in microbial composition between (sub)groups were assessed using significance analysis of microarrays in TIGR MeV (Tusher et al., 2001). Subsequent analyses were performed in SPSS or R using either Student’s t-tests or Mann–Whitney U-tests for continuous data and χ2-tests for categorical data. A multivariate linear regression model was used to explore the independent association between microbiome composition and pneumonia, correcting for potential drivers of microbiota changes including comorbidity (COPD, heart failure and immunocompromised status (ICS)) and smoking (Wang et al., 2012). The correlation between relative abundance of OTUs and age was assessed using corrected Spearman rank coefficients.

Nonmetric multidimensional scaling plots based on various distance measures were used to visualize differences between groups and statistical significance of these differences in the overall microbial community composition was calculated using PERMANOVA. Hierarchical clustering, average linkage dendrograms were built and visualized using iTol (Letunic and Bork, 2011) and the optimal number of clusters was estimated using the Silhouette index (Maechler et al., 2014).

Random forest analysis was performed to select microbial community members most important in discriminating between health and disease. Selected OTUs were included in a decision tree model; specificity and sensitivity were assessed over all study groups, for the elderly and adult cohorts separately and a k-fold cross-validation was performed.

In order to study significant, nonrandom copresence or mutual exclusion of various OTUs, a network was inferred by CoNet (Cytoscape plugin; Faust et al., 2012), the network was subsequently clustered using the Markov clustering algorithm in ClusterMaker (Cytoscape plugin) and analyzed using NetworkAnalyzer (Cytoscape plugin). For details on the statistical analysis, see Supplementary Information S1.

Results

Characterization of study population

A total of 405 subjects were analyzed; characteristics of the study population are depicted in Table 1. In general, patient characteristics were similar between groups, except for several risk factors known to be associated with pneumonia, such as smoking (in elderly and young adults), COPD, heart failure (in elderly) and age (in adults; Jackson et al., 2004).

Table 1 Demographics and clinical parameters of elderly and adult controls and pneumonia cases
Full size table

Characterization of microbiota analysis

A total of 1 528 288 sequences were obtained over all four cohorts (mean 3774±1179 sequences per sample), classified into 243 OTUs (excluding singletons), representing 25 taxonomic phyla, with most prominent Firmicutes (58.8%), Actinobacteria (18.0%), Bacteroidetes (10.9%), Proteobacteria (8.4%) and Fusobacteria (3.1%), together accounting for 99.1% of all sequences. Sequence depth was sufficient given a Good’s coverage of 0.938 (median 0.949, range 0.833–0.991). We observed a core microbiome (defined as OTUs present in >80% of individuals) of 12 out of 243 OTUs (Supplementary Table S1).

Microbiota composition in pneumonia patients and healthy controls

We observed significant differences in the overall microbiota composition between elderly patients with pneumonia and elderly controls (Bray–Curtis dissimilarity, unadjusted R2=0.096, P<0.0005), as well as between adult pneumonia patients and their healthy controls (unadjusted R2=0.081, P<0.0005). These differences were also observed using Jensen–Shannon and weighted Unifrac as distance measures (P<0.0005 for all comparisons; Figure 1, Supplementary Figure S1) and were related to shifts in relative abundance of multiple predominant OTUs (Supplementary Table S1).

Figure 1
figure 1

Two-dimensional nonmetric multidimensional scaling (nMDS) plot of the oropharyngeal microbiome composition in adult and elderly pneumonia patients and healthy controls based on the Weighted Unifrac distance measure. Each dot represents the total oropharyngeal microbiome composition of a single individual and was colored according to group; healthy elderly in blue, elderly pneumonia patients in yellow, healthy adults in green and adult pneumonia patients in red. Ellipsoids represent the standard deviation per cohort. Data points are positioned in such a manner that the two-dimensional plot represents the multidimensional data structure the best way possible (that is, (dis)similarities in microbiome structure between individuals are conserved). The microbiome structure was well captured in this nMDS visualization (stress=0.15).

Full size image

We observed a significantly higher relative abundance of Streptococcus (pseudo)pneumoniae, several Streptococcus OTUs, Rothia and a lower abundance of Gemellales, Prevotella melaninogenica, Veillonella dispar, Parascardovia and Leptotrichia in elderly pneumonia patients compared with elderly controls. Highly similar differences between health and disease were observed for the young adult cohort. Using a multivariate analysis, we confirmed that these differences in microbiota profiles associated with health status were independent of other explanatory variables (comorbidities and smoking; Supplementary Table S2). Additional independent associations between microbiota composition and explanatory variables were only observed for ICS, which was related to increased abundance of Haemophilus, Parascardovia and Clostridiaceae (Supplementary Table S2). The data set lacked the statistical power to assess the associations between disease severity and microbiome composition.

Given the low resolution of 16S-based pyrosequencing regarding discrimination of streptococcal species, we verified the contribution of Streptococcus pneumoniae in our study cohorts by qPCR; we observed a positive CT-value (<40 cycles) for lytA in 56% of elderly pneumonia patients respectively, vs 17% in healthy elderly (χ2-test, P<0.0005).

Microbiota composition in young and elderly controls

Because of the increased susceptibility of elderly to respiratory infections in general, we hypothesized that the trends observed between health and disease would possibly also exist when comparing young vs elderly healthy individuals. We indeed confirmed the negative association between the relative abundance of Prevotella, Veillonella, Leptotrichia, and positive associations between Rothia and Lactobacillus and elderly age when comparing healthy adults and elderly (Figure 2). The observed age-related dynamics of the above species could be verified within the healthy elderly cohort, confirming moderate–strong correlations between the relative abundance of Prevotella, Leptotrichia and Rothia, and age (Spearman rank r= −0.19, −0.35 and 0.25, P=0.07, 0.0008 and 0.02, respectively).

Figure 2
figure 2

Patterns in relative abundance of frequently occurring OTUs in elderly and adult controls and pneumonia patients. Relative abundance (percentage) is depicted per individual for Gemellales, Streptococcus (pseudo)pneumoniae, Neisseria, Rothia, Veillonella, Prevotella melaninogenica, Haemophilus, Leptotrichia and Lactobacillus. Statistically significant differences between groups (healthy elderly vs elderly pneumonia (circles), healthy adults vs adult pneumonia (triangles) and healthy adults vs healthy elderly) were calculated using significance analysis of microarrays with false discovery rate correction for multiple tests. *q⩽0.05; **q⩽0.01; ***q⩽0.001; ****q⩽0.0001; NS, not significant. Bars represent geomean and 95% CI of geomean per group. OTUs are color coded based on the phylum level; Firmicutes, red; Actinobacteria, yellow; Bacteroidetes, green; Proteobacteria, blue; and Fusobacteria, purple.

Full size image

Diversity and bacterial density in relation to health and pneumonia

With regard to ecological diversity measures, species richness was decreased in elderly pneumonia patients compared with healthy elderly (mean 120 (95% confidence interval (CI); 113–127) vs 159 (95% CI; 148–170), respectively, Student’s t-test P<0.0001), whereas Shannon diversity indices were increased (mean 4.5 (95% CI; 4.4–4.6) vs 4.0 (95% CI; 3.7–4.2), respectively, P=0.0002). These differences were not detected in the young adult cohort.

As an additional measure of dysbiosis, we determined the overall bacterial density of the samples, which was significantly higher in elderly pneumonia patients as opposed to elderly controls (geomean 413 vs 193 pg μl−1, P=0.001), supporting our hypothesis that bacterial replication is less controlled during pneumonia. A similar, but stronger difference in density was observed for the younger adult cohorts, with ultimately the lowest bacterial density being observed in healthy young adults (geomean 38 pg μl−1; Figure 3). In addition, within the pneumonia groups, we observed a significantly higher bacterial density in patients suffering from a moderate/severe pneumonia compared with mild pneumonia cases (geomean 517 vs 246 pg μl−1, P=0.02) supporting bacterial overgrowth being associated with (severity of) disease (Supplementary Figure S2).

Figure 3
figure 3

Bacterial density in the oropharynx of elderly and adult controls and pneumonia patients. For each group, the bacterial density per sample is depicted in pg μl−1. Bars represent geomean and 95% CI of geomean per group. Statistically significant differences between groups (healthy elderly vs elderly pneumonia (circles), healthy adults vs adult pneumonia (triangles) and healthy adults vs healthy elderly) were calculated using Mann–Whitney U-tests; *P⩽0.05; **P⩽0.01; ***P⩽0.001; ****P⩽0.0001. Healthy subjects are depicted in light gray, patients are shown in dark gray.

Full size image

Sample clustering analysis based upon phylogeny

We used weighted Unifrac distance measures to cluster samples from all four cohorts based on phylogenetic (dis)similarity (Figure 4). Only clusters comprising more than four samples were considered for subsequent analyses, resulting in 11 (sub)clusters including 95% (386/405) of samples. We observed three microbiota profiles that were strongly associated with pneumonia (χ2-test P<0.05); one smaller cluster (IV, n=11) was characterized by a high mean relative abundance of lactobacilli (relative abundance 45% vs 1% in the remaining clusters), with 10/11 (91%) individuals within this cluster suffering from pneumonia (including nine severe pneumonia cases). A second and larger cluster (Va, n=51) was characterized by relative overgrowth of Rothia (relative abundance 34% vs 7% in the remaining clusters), with 40/51 (78%) individuals having pneumonia. A third cluster (VII) was typified by predominance of Streptococcus (pseudo)pneumoniae (relative abundance 56% vs 13% in the remaining clusters)—4/42 (81%) of the individuals in this cluster were pneumonia patients. Subsequently, three clusters were correlated with health (χ2-test, P<0.05); two clusters (IIIb, n=76 and IIIa, n=7) showed high relative abundance of Prevotella melaninogenica (respectively, 16% and 28%, vs 3% in the remaining clusters). In addition to a high relative abundance of Prevotella ssp, cluster IIIb showed a relative increase of Leptotrichia (4% vs 1%). Within these two clusters, 80/83 (96%) subjects were healthy individuals. A third large cluster (VIII, n=100) showed a more even profile, although this cluster was also relatively enriched for Prevotella and Leptotrichia spp.; 98/100 (98%) of the individuals in this cluster were healthy. Five remaining clusters dominated by Neisseria (Ia, n=6 and Ib, n=24), Gemellales (VI, n=38), Actinomyces (II, n=11) and another Rothia spp. (Vb, n=20), respectively, showed no clear association with either health or disease (χ2-test, P⩾0.05).

Figure 4
figure 4

Weighted Unifrac (phylogenetic) average linkage hierarchical clustering analysis in elderly and adult healthy controls and pneumonia patients. The weighted Unifrac distance matrix was used to construct a circle dendrogram including all 405 individuals; healthy elderly (blue branches), elderly pneumonia patients (yellow), healthy adults (green) and adult pneumonia patients (red). Adjacent to the dendrogram branch ends, viral data and information on disease severity are visualized and stacked bar charts show the relative abundance of the 15 highest ranked OTUs. Color coding: presence of influenza virus and non-influenza virus is shown in purple and orange respectively; dark purple/orange, virus present in moderate-high amount (CT-value <35 cycles); moderate light purple/orange, virus present in low amount (CT-value 35–45 cycles); light purple/orange, viral qPCR performed and tested negative (CT-value >45 cycles); white, no viral qPCR performed. Clusters (defined as more than four samples; bright) are designated by dotted gray lines originating from the center of the dendrogram. Groups of samples comprising less than four samples are shown opaque. Three clusters were significantly associated with health; these clusters showed predominance of Prevotella melaninogenica (IIIa and IIIb) and Leptotrichia (IIIb) or showed a balanced microbial community (VIII). Clusters that were correlated with disease showed high relative abundance of either lactobacilli (IV), Rothia (Va) or Streptococcus (pseudo)pneumoniae (VII). Cluster IV was most strongly associated with the presence of influenza virus (50%). Five clusters showed no association with either health or disease, these clusters showed high relative abundance of Neisseria (Ia and Ib), Actinomyces (II), Gemellales (VI) or Rothia (Vb). Per cluster, the mean relative abundance of the 15 highest ranked oropharyngeal OTUs and the residual bacteria separated by phylum is given in circle graphs that were size-scaled according to the number of individuals within each cluster. PSI, pneumonia severity score.

Full size image

Association between respiratory viruses and microbiota clusters

Overall, more viruses were present in the young adult and elderly pneumonia cohorts as compared with the elderly controls (15% and 35% vs 6%, χ2-test, P<0.0005 and P=0.075, respectively; Table 2). Viral presence was more frequently observed in the pneumonia-associated clusters, that is, the Lactobacillus-dominated cluster IV (5/10 positive for any virus (50%), one missing), Rothia-dominated cluster Va (15/40 positive (38%), 11 missing) and the Streptococcus (pseudo)pneumoniae-dominated cluster VII (6/31 positive (19%), 11 missing; Figure 4).

Table 2 Viral detection rates in oropharyngeal samples taken from elderly and adult controls and pneumonia patients
Full size table

Network analysis

A network was constructed to study the co-occurrence and -exclusion of various OTUs, which yielded a total of 11 distinct OTU clusters, the largest of which consisted of 125 nodes (79% of total number of nodes), interconnected with 777 edges (95% of total number of edges; cluster 1). Within this cluster, we observed strong negative associations between important community members associated with health (Prevotella and Leptotrichia; subclusters 1A) and species associated with pneumonia (Streptococcus (pseudo)pneumoniae, Rothia and Streptococcus, subclusters 1B–D (Supplementary Figure S3).

Decision tree model to differentiate healthy controls from pneumonia patients

A random forest analysis including the 25 highest ranking OTUs was performed to identify bacterial community members most potent in partitioning the total study population in healthy individuals and pneumonia patients. Seven OTUs, Prevotella melaninogenica OTU rank 5, Leptotrichia, Streptococcus OTU rank 8, Veillonella dispar, Parascardovia, Gemellales and Prevotella OTU rank 20, appeared most discriminative, and were therefore included in a decision tree model. The decision tree was trained on the total study cohort and the final model was based on five OTUs; Prevotella melaninogenica, Leptotrichia, Streptococcus, Gemellales and Parascardovia (two OTUs were not included; Figure 5). We calculated a specificity of 95% and sensitivity of 84% for the decision tree based on the total population. To rule out skewing of specificity by the relatively high contribution of correctly classified healthy adults, the decision tree was validated separately for the young adult and elderly study population. Also after stratification, we confirmed that for both the within-elderly cohort and the within-young adult cohort analysis, a high specificity was retained (Table 3).

Figure 5
figure 5

Decision tree model to distinguish healthy individuals from pneumonia patients based on URT microbiota composition. On the basis of random forest analysis, seven variables (Prevotella melaninogenica OTU rank 5, Leptotrichia, Streptococcus OTU rank 8, Gemellales, Parascardovia, Prevotella OTU rank 20 and Veillonella dispar) were found to be most important in the discrimination between healthy controls and pneumonia patients. Measures of impurity (Gini- and information indices) were used to identify the most discriminative OTUs based on the random forest analysis. Subsequently, these OTUs were used to construct a decision tree model, which was based on the relative abundance of Prevotella melaninogenica, Leptotrichia, Streptococcus, Gemellales and Parascardovia. From the root of the tree, each branch represents a division of the initial group based on the relative abundance (shown in small font) of an OTU. The numbers (regular font size) neighboring the branches depict the number of individuals before each split. Adjacent to the branch ends are circle graphs illustrating the number of individuals (represented by the surface of the circles, colored by the study group they originate from) that are stratified by the prior split. Light green, healthy adults; dark green, healthy elderly; light red, adult pneumonia and dark red, elderly pneumonia. The numbers below the graphs represent the exact proportions. The combined presence and abundance of Prevotella and Leptotrichia appears to be the strongest determinant for respiratory health.

Full size image
Table 3 Results decision tree model
Full size table

Decision tree models are known for their risk of overfitting (Breiman, 1984) and therefore a fivefold cross-validation was used to more accurately estimate the discriminative power of the model, resulting in a specificity of 89% and sensitivity of 73% (Table 3). A cross-validation estimate for the specificity and sensitivity per age group was not performed due to the low numbers of individuals in especially the adult pneumonia cohort.

Discussion

The URT can be regarded as an ecological niche in which a dynamic interplay takes place between both commensals and potential disease causing bacteria. If balanced, the microbiome is generally advantageous to the host by direct or immune-mediated colonization resistance, which refers to the resilience against invasion and dissemination of pathogens (Buffie and Pamer, 2013). We postulate that the disequilibrium of oropharyngeal microbiota contributes to decreased colonization resistance and reduced containment of potential pathogens, leading to dissemination and subsequent development of pneumonia.

We chose to type the oropharyngeal microbiota in both elderly and young adult pneumonia patients and asymptomatic community-dwelling controls using high-throughput sequencing as it is generally assumed that the pathogenesis of pneumonia is initiated in the URT due to successive microaspiration (Krone et al., 2014). Although Streptococcus pneumoniae is the most well-known pathogen responsible for pneumonia, our study suggests that Rothia may additionally have a role in the pathogenesis of pneumonia. Rothia is unknown as a classical pathogen for CAP, though is capable of causing a diverse array of disease entities among which pneumonia, particularly in immunocompromised hosts (Ramanan et al., 2014). Furthermore, the presence and density of Rothia in the URT has been linked to an increased risk of otitis media in children (OR 1.35; 1.03–1.77; Laufer et al., 2011), also suggesting a potential role for this bacterium in the pathogenesis of respiratory infections in general.

Beside several bacterial OTUs associated with disease, we showed a decreased abundance of several Gram-negative anaerobic bacteria including Prevotella, Veillonella and Leptotrichia and the Gram-positive genus Parascardovia in pneumonia patients. Prevotella and Leptotrichia commonly reside in the oral cavity (Zaura et al., 2009), gut (Wu et al., 2011) and vagina (Eribe and Olsen, 2008; Shipitsyna et al., 2013). Prevotella is associated with a reduced risk of hospital-acquired pneumonia in adult ICU-patients (Bousbia et al., 2012), and was reported to be reduced in the oropharynx of adults and children with asthma or COPD (Hilty et al., 2010). Leptotrichia is a lactic acid-producing bacterium of the Fusobacteriaceae and is more commonly observed in healthy nonsmoking individuals (Charlson et al., 2010). We observed Veillonella was more prevalent in healthy controls, which is in line with observations regarding periodontal health (Kumar et al., 2005). Parascardovia has been isolated from saliva, gut and human breast milk (Beighton et al., 2008; Solís et al., 2010) and belongs to the family of Bifidobacteriaceae, members of which have been linked to the health-promoting effects of breast milk (Solís et al., 2010).

Interestingly, we observed a similar decline in anaerobic colonization with increasing age when only healthy individuals were analyzed, suggesting that the absence of these anaerobic bacteria might be associated with age-related susceptibility to LRTI. Although no human studies are yet available to confirm the age-related changes in (upper) respiratory microbiota, previous research in mice supports these findings; young mice showed a higher relative abundance of Bacteroidetes spp. in the URT compared with elderly mice, and additionally showed a faster and larger Bacteroidetes expansion in response to colonization and clearance of Streptococcus pneumoniae (Krone et al., 2014).

Although the similar age-related shifts in microbiota composition support the hypothesis that our findings are related to susceptibility rather than consequence of disease, we can not fully rule out the alternative hypothesis, that is, that the observed microbiome profile changes observed both in pneumonia patients and with increasing age might be the consequence rather than the cause of infection. This could, for example, be explained by inflammation-induced acidification of the local environment of the oropharyngeal niche (Lardner, 2001), which has been related to a reduction of members of the Bacteroidetes phylum and outgrowth of acidophilic species like lactobacilli (Duncan et al., 2009). Therefore, new, preferably longitudinal studies should be executed to elucidate a possible role of the observed species in relation to increased or decreased risk of developing respiratory infections.

Strengths of our study include the large cohort size including pneumonia patients of different age groups with season- and age-matched asymptomatic controls, and sampling executed according to identical protocols. Also our data did not seem to be skewed owing to antibiotic usage before admission, probably as a consequence of the restrictive antimicrobial prescription policy in the Netherlands. Finally, we were able to execute age-dependent analyses within the subgroups allowing for studies in both the general population as well as a major risk group for pneumonia: elderly age.

Potential limitations of our study include the relatively high number of ‘healthy’ non-pneumonia elderly suffering from one or more comorbidities, although this seems to reflect the normal population distribution of comorbidities in elderly, and, moreover, we thoroughly assessed the independent effect of pneumonia, comorbidities and smoking on the oropharyngeal microbial composition using a multivariate linear regression model. Our healthy adult cohort consisted solely of parents of young children, potentially limiting generalizability to the general population. Nevertheless, oropharyngeal microbiota profiles in our young adult cohort largely resembled those observed in the Human Metagenome Project (Segata et al., 2012). Owing to the 24/7 setting, sampling was performed by multiple different caregivers in different hospitals and home-based settings, which could have introduced a sampling bias; however, we tried to prevent this type of bias by confirming our observations using within-group analyses. Last, as stated, the cross-sectional study design precludes solid statements on the causal relationship between microbiome composition and development of pneumonia.

In conclusion, this is the first explorative study on correlations between oropharyngeal microbiota composition and health and pneumonia in large cohorts of elderly and young adult individuals. We showed that pneumonia in elderly patients is not only associated with increased abundance of well-known pathogens such as Streptococcus pneumoniae, but also with Rothia and Lactobacillus. Furthermore, we observed an even more profound association between the absence of anaerobic commensals and pneumonia in both young and elderly adults. In addition, we observed an increased bacterial density in oropharyngeal samples of pneumonia patients in general, and decreased bacterial richness in elderly pneumonia patients, suggesting a dysbiotic ecological niche with lack of pathogen containment. Whether the lack of ‘protective’ bacteria predisposes to the observed imbalance and whether the resulting dysbiosis is part of the causative mechanism as posed by Dickson et al. (2014), a consequence of disease, or an epiphenomenon, has yet to be determined.