Mass transit systems are BEs critical to the everyday lives of a vast number of people, and its potential role in the transmission of infectious diseases as well as bioterrorism risk cannot be understated [6, 12, 13]. With the advent of molecular assay techniques—more recently high-throughput sequencing—we are no longer restricted to culture-based techniques and can better unravel microbial diversity in mass transit systems. Characterization of microbial diversity in such environments is vital to understand the dynamics of antimicrobial resistance and enables the detection and monitoring of potential pathogens and bioterrorism threat agents. Furthermore, it is essential for the understanding of how our own microbiomes interact with the microbiomes that surround us, and how this ultimately may affect our health and wellbeing [3, 9]. A vital step in this effort is to explore the variability of mass transit microbiomes across sample matrices and temporal scales, and identify important drivers of such variation. In this study, we described the biological background in both air and surfaces from 16 subway stations in Oslo, Norway—a smaller and more northerly city compared with other cities where subway microbiomes have previously been mapped. We provide a direct comparison of surface and air communities, and an assessment of seasonal variation in subway microbiomes.
Taxonomy, relative abundances, and ecology
In the entire dataset, over 300 K unique ASVs were identified. This is substantially higher than comparable studies [5, 7, 20, 21]; however, direct comparisons of studies are not feasible since differences in sampling and wet lab protocols, and sequencing depths may strongly influence results. Further, the use of different taxonomic classifiers with different sensitivities will have substantial effects on the number of OTUs/ASVs reported [39].
The top five most abundant phyla in both surface and air samples (Table 1; Fig. 1a) matched the top five phyla in the Mexico City subway (station and train surfaces) [7] perfectly, the only difference being their ordering by relative abundance. Further, three genera in the top five overlapped between our surface samples and the Mexico City study: Staphylococcus, Streptococcus, Corynebacterium. Major phyla identified in the subway studies from Hong Kong subway [20] and New York [21] were also the same as those identified in the present study.
We found that many less abundant ASVs were unique to specific seasons or sample matrices, while abundant groups were, for the most part, ubiquitous across seasons, and surface and air samples; importantly, a very low filtering cutoff was sufficient to remove almost all taxa only present in single seasons or a single sample matrix (Fig. 2). In surface and air samples, the top 20 most abundant phyla were the same and ordered identically. Two families were highly abundant in air but not in surface samples: Rubrobacteriaceae and Pseudonocardiaceae with relative abundances of 3.52% and 2.14% in air samples. Our findings were similar at the genus level: the radiotolerant [40] Rubrobacter (which constituted the Rubrobacteriaceae contribution in its entirety; 3.52%) had a uniquely high abundance in air, along with Pseudonocardia (most notably producers of antibiotics for use in pest control by fungus farming ants [41]; 1.10%), and Nesterenkonia (ubiquitous, and in extreme environments, opportunistically pathogenic [42] 0.92%). We have no explanation for why these particular bacteria were so abundant in air, but not on surfaces. One might expect that all bacteria in air eventually settle on surfaces; however, the chemical and biological properties, and size of bacteria [43], along with environmental variables in air, can affect both deposition and resuspension rates, which introduces a high level of complexity in the relationship between air and surface microbiomes.
We observed that three Verrucomicrobia families (Verrucomicrobiaceae, Rubritaleaceae, and Chthoniobacteraceae) varied in abundance across seasons, showing the highest abundance during summer (Fig. 1b). Verrucomicrobia, which is part of the PVC superphylum, is ecologically diverse, often highly abundant and present in a range of different environments [44].
Among the three investigated surface types—kiosks, benches, and railings—we found more congruency among the highly abundant taxa (Additional file 1: Table S5), as compared with the level of differentiation observed between air and surface (Table 1).
To identify genera that were highly divergent among seasons, surface and air, and surface types, we performed random forest classification analyses, where genera were scored by their ability to bin samples in their correct category (season/sample matrix/surface type). The two genera with the highest importance for classifying samples by season, namely Psychrobacter, and Cryobacterium (Table 2) are both psychrophilic (cold tolerance or preference towards colder temperatures) [45, 46]. Psychrobacter was most abundant during winter and Cryobacterium during spring (Table 2; Additional file 1: Figure S6). For correctly binning surface and air samples, Ralstonia and Streptomyces were the most important genera, both being more abundant in air samples (Table 3; Additional file 1: Figure S7). Ralstonia are environmental opportunistically pathogenic bacilli [47], while Streptomyces is a species-rich genus, highly abundant in soil where they play an important role in carbon cycling [48]. We note that Ralstonia has been identified as a common contaminant in sequence library preparation steps [49] and that such contaminants may introduce stronger bias in sequence data from low-biomass samples, such as air [50]. The random forest classification of samples by surface type performed very poorly (Additional file 1: Table S6), which indicated that genus level taxonomic composition is not strongly diverged among surface types. Thus, we conclude that taxonomic representation is much more similar across surface types, than across air/surface or different seasons.
Diversity
Analyses of within-sample diversity (Shannon’s diversity index) and among-sample diversity (ordinated Bray–Curtis dissimilarity distances) revealed several interesting patterns. We analyzed diversity with some hitherto untried predictors (season, surface/air, indoor/outdoor stations), and some that have been included in previous subway studies (temperature, humidity, time of day, surface types).
We found no evidence for within-sample diversity differing among surface types (kiosks, railings, and benches). Analysis of among-sample diversity did reveal a significant association (Additional file 1: Figure S9), although only a relatively small proportion of the variance in among-sample diversity was explained by surface type (~ 2%). This latter finding is congruent with that of Hsu et al. [5], who found that microbial community structure varied significantly across surface types in the Boston metropolitan transit system.
Previous studies have found time of day to be a highly important variable for understanding subway microbiome fluctuations, with peak and non-peak commuting hours showing marked differences [15, 17]. We found time of day to be a significant predictor of among-sample diversity (Table 5; Fig. 4), but that it explained a relatively small proportion of the total variance in diversity, as compared with the other predictors. This may partly be due to the huge difference in number of commuters between Oslo, and Hong Kong and Beijing, and that the present study sampled outside peak commuting hours. Furthermore, the study design used here is not suited to properly gauge the importance of time of day—since this would require within-day repeated sampling for single locations.
Temperature was a highly significant predictor of both within-sample (Table 4; Fig. 3) and among-sample diversity (Table 5; Fig. 4), whereas humidity was only significant for within-sample diversity (Table 4; Fig. 3). Note that a very small proportion of the among-sample diversity total variance was explained by temperature (Table 5), while the effect size on within-sample diversity was pronounced (Fig. 3). Leung et al. [20] also found temperature and humidity to influence microbial diversity in the Hong Kong subway; note however, these associations were only significant when including outdoor stations. Our results show that humidity had a weak negative impact on diversity (Fig. 3), which is not congruent with Leung et al. who found a positive association. This incongruity may be explained by the variability and non-linear nature of the association between humidity and bacterial survival rates [51], which may give rise to different results across geographical areas and temporal scales. Humidity ranged from approximately 50 to 80% in the Leung et al. study, while our data ranged from 29.8 to 76.3%. Leung et al. found a negative association between temperature and diversity, the opposite of what we observe. Again, this is perhaps explained by the lack of overlap in the temperature range in the two studies (Leung et al.: approximately 24–30 °C; our study: − 5.45–24.91 °C).
Three of the 16 stations included in this study were outdoor subway stations. Indoor/outdoor was borderline significant in a univariate test (p = 0.08) of within-sample diversity; however, there was no significant association in the final multivariate model. The temperatures at outdoor stations will vary significantly throughout the seasons and even throughout the day, which may drive the (nearly significant) association between indoor/outdoor and within-sample diversity. When removing temperature from the final model of within-sample diversity, indoor/outdoor was again borderline significant (p = 0.07), which leads us to conclude that temperature outcompetes indoor/outdoor in our model (Table 4). Much like for temperature, we found indoor/outdoor to be a significant predictor of diversity in air samples (univariate; p = 0.04), but not in surface samples (univariate; p = 0.29). Reiterating the observed dynamic between indoor/outdoor and temperature mentioned above, a model with indoor/outdoor and temperature as predictors of air sample diversity only supported temperature (p = 0.23, p = 5 × 10−10, respectively). Although outdoor air is known to be a major source for indoor microbiomes [25], one would expect commuters, another important source [20], to be a more significant contributor in indoor environments. Hence, the lack of significance in univariate tests of indoor/outdoor as a predictor of diversity is an unexpected finding. One possible explanation is that there are relatively few commuters in Oslo, making human sources less dominant, or that effective air exchange reduces the differences between indoor and outdoor air.
A major aim of this study was to compare subway air and surface microbiomes, and we found air/surface to be a highly significant predictor of both within-sample and among-sample diversity (Tables 4 and 5; Figs. 3 and 4). Importantly, the effect of this association was dependent on temperature; we found air to have lower within-sample diversity at low temperatures, and higher diversity at high temperatures (Additional file 1: Figure S8). This can be explained by microbial diversity in air being more sensitive to temperature, as compared with surface. To evaluate this hypothesis, we ran post hoc univariate analyses of Shannon’s diversity index scores and temperature on air and surface samples separately, which found temperature to be a non-significant predictor for surface samples, (R2 = 0.01; p = 0.08), but highly significant for air samples (R2 = 0.52; p = 4.05 × 10−11). It appears that the diversity differences in air and surface microbiomes to a large extent are driven by differential effects of temperature. One explanation for this observation is the association between temperature and air circulation regimes, which can strongly influence air microbiome composition [52].
We found significant differences in within-sample and among-sample diversity across seasons (Tables 4 and 5; Figs. 3 and 4). Within-sample diversity was highest during spring and summer (Fig. 3). Apart from subway station, seasons explained the largest amount of among-sample diversity of all included predictors (R2 = 0.11; Table 5). Seasonal variation has not previously been evaluated in subways using culture-independent methods; however, Patel et al. [28] cultured bacteria and fungi from dust collected at railway stations in England and Scotland, and Heo et al. [14] measured concentrations of culturable bacteria in underground subway stations through spring and autumn. Both studies are congruent with the results presented here; bacterial numbers increased from spring through summer and decrease towards winter. Several studies have observed seasonality in atmospheric microbiome composition [22,23,24]. With the outdoors being an important source for BE microbiomes [25], this suggests that seasonal variations in subway microbiomes may be influenced, at least partly, by seasonal changes in atmospheric microbial communities.
Subway station was a highly significant predictor of among-sample diversity, explaining 15% of the total variance (Table 5). However, when inspecting the clustering of PCoA ordinated values in Fig. 4, there are no clear patterns. We suspect that this result is mainly a consequence of including a categorical predictor with too many levels. Hence, we must refrain from concluding on the importance of subway station as a predictor of microbiome composition in our study. Sequence run was also a significant predictor of among-sample diversity and explained 2% of the total variance. We propose that this stems from an unbalanced partitioning of samples from different seasons, sample matrices, or other variables into the four sequence runs. Alternatively, the association with sequence run may be explained by predictors not included. Both these explanations are congruent with the qPCR results, which show higher bacterial load in the samples from sequence run 3 (Additional file 1: Table S4 and Figure S2).
Caveats
In our study, seasonality was assessed by sampling on single days within seasons without accounting for the variation in shorter time periods (e.g., weekly variation) or repeatability across years. While patterns such as differential abundance of certain taxa in spring and summer, compared with autumn and winter are convincing, a higher resolution sampling scheme should be implemented in the future to distinguish between variations that occur on different timescales. Although we provide a relatively high level of geographical resolution in the present study, we recommend that future studies address seasonal and air/surface variability across cities, countries, and continents using standardized methods.