Platanus × hispanica trees at three locations were chosen around Hasselt, Belgium for sampling based on the local urban characteristics and adjacent traffic patterns, which were observed visually. Hasselt is a city of 102.24 km² with a population of 77.000 inhabitants. Live air pollution data are monitored and available on the VMM website. The year average PM2.5 concentrations are 15 μg/m³ (VMM). Mothers and their newborns in the city of Hasselt and nearby city of Genk have the ability to participate in the ENVIRONAGE birth cohort study, in which the effect of UFP exposure and black carbon on human health is investigated (Janssen et al., 2017, see also Figure 1 therein for a Corine land cover and main road map of Hasselt). At each of the three locations, adult trees were sampled (see Google street view for the general tree morphology and environment). The “Rural” tree (50.969 N; 5.408 E) was located in the arboretum of Domein Bokrijk which has a low burden of urbanization and road traffic. The “Intersection” tree (50.930 N; 5.360 E) was located at a major intersection outside of the city proper; it represents lower urban intensity and has a high road traffic burden. The “Inner-city” tree (50.932 N; 5.347 E) was located within the city proper; it represents a higher urban intensity but has a lower road traffic burden compared to the Intersection. None of the trees chosen receive any type of annual maintenance such as pollarding or crowning.

In early May 2014 and 2016, we collected leaves with ethanol sterilized gloves from four equidistant points around each tree at an approximate height of 3–4 ms, which was judged by the length of our cutting tool and near to the bottom of all trees sampled. Each of the four replicate samples consisted of three leaves from the same stem and weighed approximately 2.5 g. The sample leaves were transferred into sterile falcon tubes with 20 ml of 10 mM MgSO₄ solution and transported to lab.

At each location, ultra-fine particles (UFP, < 1.0 μm diameter) were measured with an Aerasense Nanomonitor (Philips, Amsterdam, Netherlands) during the same hours as the leaf sampling took place and additional leaves were collected to determine their black carbon (BC) content. We took the land-use classifications for each location from the 2012 CORINE Land Cover database (European Union, 2018).

To release the epiphytic bacteria from the surface, sample leaves in 20 ml of 10 mM MgSO4 were vortexed on high for 2 min. The leaves were removed and the epiphyte solution was split into three aliquots which were passed through three sterile 0.2 μm nitrocellulose filters (Merck, Darmstadt, Germany). Each filter was transferred to a separate 2 ml bead beating tube with 0,1 mm glass beads (MoBio, Carlsbad, CA, United States) to release the epiphytes by shaking into Lysis Buffer P solution from the Invisorb Spin Plant Mini Kit (Stratec Molecular, Berlin, Germany). The lysis buffer was transferred to a new tube and gDNA was extracted with the kit according to manufacturer’s instructions, utilizing the Prefilter to remove excess glass beads and plant material.

Prior to delivery, sequences were trimmed by Macrogen with Scythe (v.994) and Sickle to remove the adapter sequence and reads less than 36 bp. Remaining reads were then quality filtered and chimera checked with DADA2 v 1.8 (Callahan et al., 2016). The resulting ASV table and representative sequences were imported as artifacts to Qiime 2 (Gregory Caporaso et al., 2010). To improve the accuracy of taxonomic classification, we extracted our 16S target region of 799F–1193R from the Silva v132 (Quast et al., 2013) 99% data set and used these optimized reference sequences to train a custom Naïve Bayesian classifier on the majority_taxonomy_7 SILVA database. FastTree 2 (Price et al., 2010) generated rooted and unrooted phylogenetic trees after masked alignment of the reference sequences. All samples were rarefied to the lowest equal number of reads and we checked for adequate sampling depth with the Good’s Coverage metric. Genera occurring in at least 75% of the replicate samples each year were included as part of each tree’s persistent microbiome.

All raw Illumina 16S rRNA gene sequencing data were submitted to NCBI SRA with project identifier PRJNA505409, and sample IDs SRX5015223 to SRX5015246.

To better understand the bacterial phyllosphere of Platanus × hispanica trees planted in urban centers and the effects which anthropogenic pressures may exert on these communities, we chose three locations of different urban density and traffic patterns from which to sample air quality and leaves. Hasselt is the capital city of the Limburg province in Belgium. According to open data on statbel.be, between 2014 and 2016, the municipal population rose from 76,000 to 77,000 inhabitants. Table 1 further describes the sampling sites, including the name we assigned each location to help differentiate between the trees, the coordinates and land use class in which each tree stands, and the distance from each tree to the nearest road. The average UFP per cm³ and average UFP size in nanometers are also listed in Table 1.

Air sampled at the urban locations, i.e., the inner-city and intersection sites, had significantly higher concentrations of UFP on average than the rural location (p < 0.001), but with no significant difference between each other. However, particle size did change based on location. Compared to the inner-city site, the rural position had significantly larger particles (p < 0.001) and the intersection location had significantly smaller particles (p < 0.001).

Similar to the UFP in the air, the urban leaves also had more BC particles than the rural leaves (Figure 1A). When accounting for the difference in surface area, the urban leaves also had a greater percentage of area covered by the BC particles (Figure 1B).

After quality filtering, our 16S amplicon data set consisted of 178,908 sequences, ranging from 2,907 to 34,895 sequences per sample. In an effort to keep all 24 samples in the dataset, rarefaction was set to the lowest number of reads and subsequently checked with Good’s coverage index. The average sample coverage was 0.99 (± 0.001 SE), which was deemed sufficient to continue with all samples included in further analysis.

The rarefied dataset represented bacterial communities from 17 phyla, 273 identified genera, and 1,502 ASVs. Overall, the most dominant phyla were Proteobacteria (71.5%), Deinococcus-Thermus (16.9%), Actinobacteria (4.5%), and Firmicutes (1.9%). These phyla were present on all trees in both sampling years and except for the Firmicutes which were missing in three of the eight Rural samples, these phyla were present in all samples.

Proteobacteria was the most abundantly identified phylum at all locations, however the phylum was more abundant in 2014 compared to 2016 (Figure 2). In 2014, Proteobacteria represented 93.4 and 86.8% of the rural and urban bacterial community, respectively. By 2016, the community structure had shifted to include more Deinococcus-Thermus bacteria, especially on the urban tree leaves (42.0%) compare to the rural tree leaves (14.0%). Even so, the Proteobacteria phylum still made up the majority of bacteria at each location. Other highly abundant phyla include the Actinobacteria (2.3–6.0%), Firmicutes (0.6–3.3%), Bacteroidetes (0.7–1.8%) and FBP (0.8–2.5%).

A closer look into the community structure at the level of genus illustrated not only the same temporal differences we saw on the phyla level, but also interesting similarities between the urban samples that are not in the rural samples (Figure 2). In 2014, the most abundant bacterial genus on the urban leaves was Escherichia-Shigella and in 2016 it was Deinococcus with an average relative abundance of 38.8% (±0.5 SE). The most abundant genus of the rural tree in 2014 was Sphingomonas (28.3%) and in 2016, unknown Enterobacteriaceae (25.3%). The top ten most abundant genera at each tree account for more than three-quarters (78.7% ± 2.5 SE) of the total genera.

At each tree location, bacterial genera that were persistently identified in the phyllosphere from one sampling year to another were compared between all three locations and we found eight genera are consistently shared across all trees (Figure 3). While this represents a small fraction of the overall diversity (28% of all ASVs) it comprises a large portion (60.9%) of the total reads.

The shared genera were Deinococcus^∗ (15.9% of total reads), Escherichia-Shigella (15.9%), Sphingomonas^∗ (12.1%), Pseudomonas (5.9%), 1174-901-12^∗ (Family: Beijerinckiaceae, 5.4%), Methylobacterium^∗ (3.3%), Acidiphilium (1.8%), and Chryseobacterium (0.6%). Of these six, only four genera, those denoted by “^∗,” were identified in 100% of the samples taken.

Between the intersection and inner-city sites, Friedmanniella was a persistently identified genus and between the intersection and rural leaves, Afipia was also persistent. The rural and inner-city locations did not share any more genera than those common to all three sites. However, the rural and inner-city leaves each had one genus unique to that location while the intersection leaves had four.

There was no meaningful difference in observed species or phylogenetic diversity between the locations (Table 2). Although in 2014 the inner-city had slightly less diversity than the rural leaves (Simpson and Shannon p = 0.0304), this was not the case in 2016. The phylogenetic diversity (Faith’s PD) and observed ASV counts did not differ by location in both years.

To compare the different bacterial composition of each leaf phyllosphere by location and between years, a Bray-Curtis dissimilarity matrix was calculated for each sample. PCoA plotting (Figure 4) revealed that the communities are differentiated by year (p = 0.001), as seen on the first axis with 30.5% of the variation. Since most of the variation is explained by the first axis, we employed a blocking strategy to separate the years before testing for significance of community dissimilarity between tree locations (p = 0.001). Pairwise PERMANOVA analysis (Table 3) of the 2014 samples revealed the urban bacterial communities of inner-city and intersection were dissimilar (p = 0.032). But in 2016, the two urban communities clustered together (p = 0.461). But what the urban leaves both had in common was that they were significantly different than the rural leaves in both sampling years (2014, p = 0.032 and 2016, p = 0.045).

Controlling for difference between the years again, we were able to determine that land use class type had an effect on the relative abundance of two major bacterial classes in the phyllosphere. LDA inclusion threshold > 2.0, p < 0.05. The LEfSe cladogram of Figure 5 shows Actinobacteria and more specifically, Propionibacteriales, Propionibacteriaceae, and Friedmanniella, in addition to Flavobacteriaceae are more enriched on the urban leaves. Whereas Alphaproteobacteria, Methylocella, Sphingomonas, Acidobacteriales, Acidobacteriaceae Subgroup 1, and unknown Enterobacteriaceae are more enriched on leaves in the rural environment.

Across all samples, 6,846 orthologs and 304 pathways were identified using the KEGG reference database. The similarity between predicted KEGG orthologs was visualized on a PCoA plot (Figure 6). Comparable to the Bray-Curtis dissimilarity of bacterial communities (Figure 4), the first axis differentiates between the years and the second axis distinguishes variation between the locations.

In order to determine how the functional profile of the microbiome may have changed over time, we compared the percentage of orthologs with significantly different abundances between 2014 and 2016 (Table 4). The rural tree remained relatively stable with 10.75% of the total orthologs changing between the years. In contrast, the inner-city and intersection trees were more dynamic; roughly half of the orthologs had significantly different abundances (Inner-city, 52.05%; Intersection, 50.18%).

To determine if the functional profile of the microbiome was influenced by location, we looked at the inferred orthologs within each year (Table 5). The percentage of orthologs with significantly different abundances was consistent between years for the inner-city and rural trees (2014, 21.22%; 2016, 20.92%). But their relationship to the intersection tree had more variation.

To determine if the urban phyllosphere might be metabolically enriched in alternative carbon sources, such as polycyclic aromatic hydrocarbons, we used the LEfSe tool to analyze the 154 KEGG annotated metabolic pathways predicted by Piphillin. The enriched pathways (p < 0.05) and corresponding LDA effect size scores for each land use class type are displayed in Figure 7. The bacteria on leaves in the mixed forest environment were more associated with pathways concerning “metabolism of cofactors and vitamins” and “glycan biosynthesis and metabolism,” while the urban communities are predicted to contain more xenobiotics and lipid metabolism pathways.

From the perspective of microorganisms, the leaf habitat is vast in space but limited in nutrients and constantly exposed to environmental stresses like temperature fluctuations, UV radiation, and rainfall. In particular, leaf associated bacteria on trees in urban areas and close to roadways are recurrently exposed to aerosolized pollutants dispersed into the environment via vehicular exhaust (Weyens et al., 2015). As the majority of aerosolized particulate matter in urban environments is comprised of particles in the ultra-fine range of 100 – 10 nm, we chose to focus on this fraction as well (Morawska et al., 1998; Junker et al., 2000). Our results indicate the average concentrations of particles measured in this study (Table 1) are similar to those published in other European studies of urban and residential areas (Matson, 2005; Puustinen et al., 2007; Pey et al., 2009; Hofman et al., 2016).

In addition, we also measured BC directly on the leaf surface. BC is a common component of PM and UFP formed during incomplete combustion and aerosolized in automobile exhaust. Recently, the BC fraction of particulate matter has come under greater scrutiny for its negative effects on human health and the environment (Janssen et al., 2012; De Prins et al., 2014). Zhu et al. (2002) reported that UFP and BC concentrations were both highest next to roadways, therefore we expected to find more BC on the urban leaves as a result of their increase exposure to UFP and close proximity to roads (Table 1). This was confirmed by our results of the higher number of BC particles (Figure 1A), on urban leaves as well as the greater percentage of leaf coverage by the BC particles (Figure 1B). Although the Intersection has more traffic passing by than the Inner-city location, it did not make a significant difference in average UFP counts or BC accumulation on leaves. The abundance of smaller UFP at both of the urban locations and especially the more residential Inner-city location, is of concern as this size fraction is associated with negative effects on respiratory health (Leitte et al., 2011; Li et al., 2016).

Proteobacteria, Actinobacteria, Firmicutes, and Bacteroidetes, that were the dominant phyla observed in our samples (Figure 2), are often reported in culture-dependent and culture-independent studies of leaf epiphytes. As expected, Proteobacteria was the dominant bacterial phylum in all samples, which is in accordance to numerous other studies on the bacterial community composition of plant leaves (Redford et al., 2010; Bodenhausen et al., 2013; Kembel et al., 2014; Laforest-lapointe et al., 2016; Lymperopoulou et al., 2016; Smets et al., 2016; Gandolfi et al., 2017). It has been suggested that this is due to the speed at which many bacteria within the Proteobacteria phylum replicate in comparison to other phyla (Leff et al., 2015).

Recently, bacteria within the phylum of Deinococcus-Thermus have also been described as typical phyllosphere residents (Redford et al., 2010; Laforest-lapointe et al., 2016). This might be due to a change in the methodology for studying aerial sections of the plant. Clone libraries and amplicon strategies not including the 799F primer for chloroplast exclusion (Chelius and Triplett, 2001) seem to be biased against this phylum (Redford and Fierer, 2009; Jackson and Denney, 2011; Beckers et al., 2016; Gandolfi et al., 2017). Indeed, the use of 799F–1115R primers found Deinococcus-Thermus on leaves of nearly all the 56 of the tree species studied (Redford et al., 2010).

Only two Deinococcus-Thermus genera, Deinococcus and Truepera, were identified in our samples. Of these, Deinococcus was more abundant on all leaves and in both years. This is not surprising as Deinococcus species are reportedly highly resistant to UV radiation and desiccation, which are fairly common environmental stressors to the phyllosphere. There are indications that Deinococcus might be useful for degradation of aromatic compounds. Deinococcus bacteria capable of phenanthrene degradation has been isolated from Ixora leaves in Thailand (Waight et al., 2007).

Similar to Deinococcus-Thermus, the candidate phylum FBP is also associated with extreme environments and often found in 16S rRNA gene sequencing of phyllosphere samples (Laforest-lapointe et al., 2016; Ruiz-Pérez et al., 2016). However, not much is known about these candidate bacteria since only one has been successfully cultured and studied (Tahon et al., 2018).

Despite the dynamic and complex nature of host-associated microbiomes, it has been theorized that a core portion of each community is persistent across time and habitats (Shade and Handelsman, 2012). Members of the core microbiome are credited with being an integral part for the optimal microbial functioning needed to promote the health of their host-species (Toju et al., 2018).

In this study, we identified which genera were persistently found at each location and then compared these to find the shared persistent bacteria. We chose to compare genera rather than ASVs in order to include the possibility of functional redundancy within the same genus. As deciduous trees grow new leaves every year and the order of bacterial colonization matters (Maignien et al., 2014), it may not be reasonable to expect the same ASVs but to consider that ASVs of the same phylogenetic lineage will share the desired functional trait.

Although it is likely that the eight shared genera (Figure 3) are connected to the core microbiome because they were detected the leaf bacterial community in two non-consecutive years, due to sampling limitations, it is still unknown if they are universal to all Platanus trees or they are localized to this region and climate. Only one other study has sought to evaluate the taxonomic composition of bacterial communities on Platanus leaves (Gandolfi et al., 2017) for which we can compare our results. Out of the eight genera identified in our study, only Sphingomonas, Pseudomonas, and Methylobacterium were also consistently identified on Platanus × acerifolia leaves in Milan (Italy) over one growing season. Therefore, these three genera stand out as the most probable members for the core microbiome of Platanus leaves. It should be noted that Platanus × acerifolia and Platanus × hispanica are historically synonymous names for the same tree (Besnard et al., 2002) and bacterial communities on leaves of the two other common Platanus species, orientalis and occidentalis, have not yet been reported.

The decrease in available green space in cities is often associated with negative effects on human health, mostly related to mental and physical wellbeing (Jim et al., 2018). In addition to the benefits received from eukaryotic diversity, exposure to a variety of prokaryotes is connected with establishing and maintaining a healthy immune system (Haahtela et al., 2013; Rook, 2013). We clearly see that the built environment limits plant and animal biodiversity in cities, so there is concern regarding the loss of microbial richness, which is more difficult to assess.

Phyllospheric communities frequently undergo disturbances such as changes in UV radiation, precipitation, and resource availability that disrupt the natural community succession (Redford and Fierer, 2009). The significant temporal difference in intra-tree diversity between 2014 and 2016 is likely to be a combination of multiple environmental factors. Therefore, within each year we sampled from the same trees located in close proximity to each other so meteorological changes would have less of an effect on location. Based on a comparison with the leaves of the trees in the urban and rural areas, our results indicated that the richness and diversity of bacteria on urban Platanus × hispanica tree leaves are not significantly impacted by increased exposure to UFP and BC (Table 2). This fits with other similar studies on urban phyllospheric bacterial diversity (Smets et al., 2016; Laforest-Lapointe et al., 2017).

Redford and Fierer (2009), Laforest-lapointe et al. (2016), and Gandolfi et al. (2017) all found that over one growing season, the influence of site was a stronger driver of inter-tree compositional diversity for the phyllosphere than temporal changes. We chose to study a longer separation of time, where the trees have lost and regained their leaves twice, to address the impact of site on the establishment of new bacterial communities rather than its effect on the community succession.

Our results indicate that the taxonomic composition of epiphytic bacterial communities is influenced by urbanization. It was apparent that each community saw a large temporal shift in bacterial taxa, but within each year the general relationships between communities remained nearly the same (Figure 4).

Consistent in both sampling years, the urban trees shared a similar percentage of the same dominant genus (Figure 2). In 2014, this was Escherichia-Shigella and in 2016, it changed to be Deinococcus. Interestingly, when considering changes in relative abundance at the urban trees, Deinococcus was infrequently identified in 2014 and Escherichia-Shigella was, in turn, similarly, scarce in 2016. The rural tree had a distinctly different pattern of abundant bacteria compared to the urban trees in both sampling years.

The bacteria revealed by LEfSe analysis to be enriched at the urban and rural locations in both years are mostly those that are typically found in environmental samples (Figure 5). The abundance of Propionibacteriaceae on the urban leaves stands out however, as this family is often associated with human skin. Lehtimäki et al. (2017) found Friedmaniella was enriched on the forearms of children living in urban areas compared to those in rural homes. Also of note is that although the Sphingomonas genus was persistently identified on all Platanus × hispanica leaves (Figure 3), it was shown here to be more commonly identified in the rural samples.

We wanted to know if the temporal and spatial differences in taxonomic composition (Figure 4) would also be reflected in the inferred functional composition (Figure 6). For this purpose, we employed Piphillin because it has the flexibility to interpret inferences from ASV tables rather than needing a closed reference OTU picking strategy. One of the unfortunate disadvantages is that Piphillin is less accurate than PICRUSt and Tax4fun when used on environmental samples (Iwai et al., 2016). Hopefully this will change in the future when more genomes of environmental microbes are fully sequenced and added to databases. It is also important to note, that although the functional orthologs and pathways are directly inferred from the taxonomic composition of each bacterial community, Piphillin uses a reference database to correct for variation in 16S copy number and therefore the functional differential abundance cannot be directly compared to the taxonomic differential abundance.

The KEGG orthologs predicted for 2014 were compared to those predicted for 2016 at each location in order to bring a greater understanding of how the functional aspect of each bacterial community changed over time (Table 4). The urban trees saw a larger change in ortholog composition between the sampling years than the rural tree. This suggests that even though the taxonomic composition of the rural tree was significantly different, the predicted functional profile of the bacteria remained similar.

The difference in ortholog composition between locations (Table 5) also mirrors the differences seen in taxonomic compositions. In both sampling years, roughly 21% of the rural and inner-city orthologs had significantly different abundances. However, the dissimilarity between each of those bacterial communities to the intersection community was more dynamic across both years. In terms of geographic distance, rural and inner-city are also further away from each other and experiencing other environmental conditions (neighboring plants versus more artificial city surfaces), so a higher percentage of differently expressed KEGG orthologs suggest that the communities exert different metabolic functions. In order to confirm this, a shotgun sequencing approach, metabolomics study or individual-based microbial ecology (IBME) should be performed to quantify the performance of microorganisms in their natural habitats. An IBME approach has confirmed that heterogeneity in fructose availability has an important role in the reproductive success of bacterial immigrants to the leaf surface (Remus-Emsermann and Leveau, 2010). How atmospheric car-exhaust related air pollutants might affect carbon substrates in the phyllosphere and therefore sustain metabolically different microbial communities remains to be explored.

We limited the LEfSe analysis (Figure 7) to include only the metabolic pathways as our interest was mainly in determining if the urban trees had a greater inclination toward accumulating bacteria with pathways to degrade common traffic related compounds, such as polycyclic aromatic hydrocarbons. While many of these were not directly enriched, LEfSe did reveal that overall degradation of xenobiotic compounds was predicted to be more common in the urban phyllosphere bacterial communities.

The benzoate degradation pathway is also enriched in the urban samples. This pathway is intimately involved with the degradation of monoaromatic hydrocarbons such as benzene and toluene, as well as degradation of polycyclic aromatic hydrocarbons metabolites. Although these results do not necessarily mean that the enzymes in this pathway are more active in the urban bacterial communities. It has been observed previously that bacteria with hydrocarbon degradation potential are inhabiting the phyllosphere of city plants, e.g., we isolated previously hydrocarbonoclastic Rhodococcus (Stevens et al., 2017b) and Bacillus (Stevens et al., 2017a) on Hedera helix leaves.