Understanding the patterns and drivers of biodiversity is central to predicting ecosystem responses to environmental change. This is particularly true for microbes because of the key roles they play in global biogeochemical cycles. Despite recent advances in sequencing technologies, identifying the mechanisms that underlie microbial diversity remains a major challenge. This challenge is particularly significant for highly dynamic ecosystems such as flowing-waters where temporal and spatial variability in flows are often dramatic (Poff et al., 2006). These environments host mosaics of habitat patches including surface and subsurface water, sediment, and epilithic biofilms that differ in their environmental conditions (Winemiller et al., 2010) and set the stage for local adaptation and patch scale species sorting by microbes (Adams et al., 2014). However, the pool of dispersing microbes available to colonize these patches is highly dynamic (Zeglin, 2015)—for example, bacteria can be suspended into stream water following streambed disturbances and many microbes appear to enter stream water from watershed sources (Crump et al., 2003, 2007, 2012).

The combination of high habitat heterogeneity and a large dispersal potential in running-water systems has led researchers to suggest that frameworks from landscape ecology and metacommunity theory may be useful in studies of the diversity and composition of stream microbial communities (Battin et al., 2007). Specifically, water column bacteria represent a pool of microbes available to colonize benthic habitats after which local adaptation and patch-scale species sorting can occur; both dispersal and local environmental conditions may influence microbial diversity, and composition albeit at different proportions in different habitat types (Crump et al., 2007, 2012; Besemer et al., 2013). For microbes in environments with longer residence times and decreased colonization rates, such as stream bed environments, environmental sorting has a stronger influence on microbial composition (Or et al., 2012; Adams et al., 2014; Handley et al., 2014) except in cases where mass effects are strong and continual (Souffreau et al., 2014).

Watershed land use, including urbanization, influences microbial diversity, and composition (Belt et al., 2007; Wang et al., 2011) through both dispersal and by changing the environment at the patch scale. Urbanized landscapes are likely the sources of novel microbial taxa not found in undisturbed stream ecosystems, including taxa from sewage and septic systems, water distribution systems, and stormwater management ponds. Urbanization also changes the local physical and chemical milieu of stream habitat patches by, for example, subjecting stream reaches to sediment erosion or deposition, elevated conductivity, temperature, nutrients, altered organic matter quality, or other stressors (Walsh et al., 2005; Hosen et al., 2014). Flow extremes exacerbated by urban development (Paul and Meyer, 2001) may directly alter the composition of bacterial communities via scouring and dispersal, and indirectly change communities by altering sediment size—as has been shown for denitrifying taxa (Perryman et al., 2011b). These flow-related effects on biological diversity are well-known for larger stream organisms such as macroinvertebrates (Moore and Palmer, 2005; Wenger et al., 2009), but the impact of urbanization on bacterial community composition is less clear.

Most urbanization studies on microbial communities have focused on pathogenic taxa such as fecal coliform bacteria (Nagy et al., 2012; Daly et al., 2013; Kapoor et al., 2014), denitrifying bacteria (Hale and Groffman, 2006; Knapp et al., 2009; Perryman et al., 2011b; Harrison et al., 2012), or unicellular algae and diatoms (Hill et al., 2000; Elsdon and Limburg, 2008). The handful of studies that were not limited to coliforms, denitrifiers, or algae/diatoms have suggested a large difference between bacterial communities in biofilms (Lear and Lewis, 2009; Lear et al., 2011), streambed sediments (Jackson and Weeks, 2008; Perryman et al., 2011a; Wang et al., 2011), and the water column (Belt et al., 2007; Or et al., 2013). This body of work provides important insights, but these studies used relatively coarse measurements of microbial diversity such as denaturing gel gradient electrophoresis (DGGE), automated ribosomal intergenic spacer analysis (ARISA), and terminal restriction fragment length polymorphism (T-RFLP) analysis. The coarse unit of analysis for these fingerprinting techniques is inadequate for testing of alpha diversity (Dunbar et al., 2000), community interactions, and potential functions (Lozupone and Knight, 2007; Hamady and Knight, 2009). Our goal was to determine how urbanization influences stream microbial diversity in both the water column and sediments using a high-throughput sequencing approach that provides higher taxonomic resolution and allows for direct cross-study comparisons.

We used high throughput sequencing of 16S rRNA genes to quantify bacterial community composition in 11 streams (hereafter, “sites”) in watersheds in which the dominant land cover was forest with varying levels of urbanization—as measured by percent watershed impervious cover. To explore potential links between composition and environmental factors we also measured physicochemical parameters known to influence aquatic diversity. Our specific objectives were to: (1) quantify alpha diversity in different habitat types (water column vs. sediment) and landscape urbanization—as determined by percent watershed impervious cover; (2) quantify beta diversity and taxa co-occurrence patterns across habitats and across a gradient of urbanization; and, (3) identify the environmental factors that explain variation in community composition.

Communities of larger stream organisms—including macroinvertebrates and fish—show compositional differences among stream habitat types and across watershed land uses (Morgan et al., 2007; Campbell and McIntosh, 2013). We anticipated similar patterns among microbial communities including large differences between habitat types, and correlations between microbial alpha- and beta-diversity and the magnitude of landscape urbanization. We hypothesized that water column communities are more strongly related to watershed land use than sediment communities because delivery of microbes from the landscape is a major driver of community composition for microbes suspended in stream water columns (Crump et al., 2012). By contrast, we hypothesized that sediment communities are more strongly linked to environmental conditions within stream reaches because these habitats are more stable and bacterial residence time is longer. Thus, we expected to find evidence that mass effects strongly influence community composition in the water column, but that species sorting would mute the impact of mass effects in sediment environments.

Bacterial community composition in both sediment and water column was significantly correlated with urbanization, as measured by watershed impervious cover. However, the underlying factors driving microbial community response to urbanization were different for water column vs. sediment. Water column communities showed a stronger connection to watershed urbanization than sediment communities, providing evidence that water column communities may have been more strongly influenced by dispersal and growth of organisms from sites of anthropogenic disturbance on the landscape. By contrast, sediment communities showed a stronger connection to environmental conditions within the stream reach sampled including discharge, conductivity, and nutrient levels, suggesting that species sorting had a greater impact on sediment bacterial communities. While we do not have direct evidence of the ultimate source of water column microbial taxa, our finding that microbial community composition in the water column is more tightly linked to landscape conditions than sediment communities is consistent with previous studies suggesting that most bacteria in headwater streams may originate from upstream watersheds, however the taxonomic composition of these communities is subsequently modified by environmental sorting within the streams (Crump et al., 2012; Adams et al., 2014; Souffreau et al., 2014). In contrast to these patterns in beta-diversity, microbial alpha-diversity was not strongly connected to urbanization, but the microbial co-occurrence network for urbanized sites was smaller and less connected than for forested samples. We hypothesize that urbanized sites are subjected to an increased magnitude or frequency of disturbance events that is driving a loss of keystone-like taxa in urbanized streams.

Microbial beta-diversity was significantly related to urbanization, measured as log-transformed percent watershed impervious cover, regardless of the habitat type—water or sediment. Urbanized stream sites included taxa that have been associated with eutrophic conditions, human activity, and other impacts of urban infrastructure. Polynucleobacter was identified as an indicator of urbanized streams and was highly abundant in those samples—accounting for 1.65% of sequences in urbanized stream samples compared to only 0.58% of sequences in forested stream samples. This OTU is a genus of picobacteria that has been linked to high levels of planktonic autotrophic activity and warm temperatures and is more often associated with larger rivers rather than headwater streams (Hahn, 2003; Boenigk et al., 2004; Wu and Hahn, 2006). Hence, the presence of Polynucleobacter suggests the establishment of a more planktonic microbial community, possibly in response to increased nitrogen levels in these streams. Other taxa identified as indicators of urbanized streams included Albidiferax species, which were found to be ubiquitous in groundwater contaminated with tetraclorethene (Kotik et al., 2013), and Gallionella sp., which are responsible for corrosion in water distribution systems (Ridgway et al., 1981). Thus, the manner in which urbanization shapes stream bacterial communities can be multifold—representing both local environmental impacts and regional processes. Our data suggest that nutrient inputs favor development of a eutrophic community within the stream environment, while infrastructure such as leaky pipes within a watershed deliver novel taxa directly to urbanized streams.

The effects of urbanization were also demonstrated by the CCA, which showed that bacterial communities at forested sites were associated with higher phosphorus levels, while communities at urbanized sites were linked to higher nitrogen levels. This analysis provided further evidence that elevated nitrogen in urbanized streams fuels a shift toward planktonic communities. According to the CCA analysis, Polynucleobacter (OTU 4) and Candidatus Planktophila limnetica (OTU 18) were both positively associated with higher levels of total dissolved nitrogen in the water column (Figure 8C). The latter taxon is a planktonic actinobacterium (Jezbera et al., 2009) that, like Polynucleobacter, was also highly abundant in urbanized samples. An OTU from the genus Paucibacter (OTU 62099) was also linked to higher nitrogen levels. Species from this genus degrade toxic peptides produced by cyanobacteria and may be linked to phytoplankton blooms (Rapala et al., 2005). Forested and urbanized stream sites were located at similar positions within the stream network and had similar water velocity and discharge during baseflow conditions, yet urbanized streams had a large number of taxa that are typically associated with water bodies with relatively high water residence times. Increased nutrient levels with urbanization have been linked to increased stream primary production (Alberts et al., 2017) and CCA revealed a link between “planktonic” taxa and higher nitrogen levels (Figure 8). Thus, we argue that increased nutrient levels in combination with a more open canopy at urbanized sites may be stimulating primary production and the rapid growth of a community that would otherwise not be able to develop in a headwater streams with low hydrologic residence times. We acknowledge that the implications of the link between planktonic and nitrogen levels is limited by the observational nature of this study and recommend that experiments under controlled conditions be conducted to evaluate the correlation identified here.

In contrast to urbanized streams, a number of indicator taxa for forested streams were methanotrophic or methylotrophic. The top indicator for forested streams was an OTU from the genus Hyphomicrobium—which is composed entirely of facultative methylotrophs (Scheulderman-Suylen and Kuenen, 1985; Rissanen et al., 2016). Rhizobiales (OTU 23477) and Crenothrix (OTU 5) were two additional taxa linked to forested streams that are likely methanotrophic (Stoecker et al., 2006; Dieser et al., 2014). By contrast, only one top indicator of urbanized streams—an OTU from the order Methylococcales—was identified as methanotrophic (Bowman, 2005; Kato et al., 2013). Previous work in these systems (Febria et al., 2015) and elsewhere (Beck et al., 2013) has demonstrated a strong link between anaerobic methanogens and methane-oxidizing methylotrophs and methanotrophs. Considering this evidence, the relatively high diversity of methane-consuming taxa in forested samples suggests that forested stream microbial communities are structured by inputs from groundwater and the hyporheic zone. Though we lack data on the source of water in urbanized vs. forested streams, the difference in community composition between the two stream types suggests a switch from deep to shallow water flowpaths, suggesting that perhaps watershed urbanization disconnects headwater streams from groundwater supplies.

Forested streams also supported a variety of denitrifying microbes including Hyphomicrobium and Rhizobiales spp., and had high numbers of nitrifiers including Nitrospira which oxidizes nitrite and supports anaerobic ammonia oxidizing (anammox) bacteria that convert ammonia to dinitrogen gas (Daims et al., 2001; Park et al., 2015). In contrast, these taxa were not favored in urbanized streams, providing further evidence of decreased nitrogen removal capacity in urbanized streams. This finding is consistent with other studies that have demonstrated a loss of stream denitrifiers in urbanized streams (Perryman et al., 2011b; Wang et al., 2011).

Microbial co-occurrence networks provide a way to assess the degree to which microbial communities are integrated, and to identify taxa that are important drivers of these relationships (Barberán et al., 2011). Interactions can include competition for resources, parasitism, or mutualistic interactions such as co-metabolism (Parter et al., 2007; Faust et al., 2012). The forested co-occurrence network was significantly larger than the urbanized network, indicating a loss of keystone-like microbial taxa in urbanized streams (Berry and Widder, 2014; Figure 6, Supplementary Datasheet 3). The urbanized network also demonstrated lower average path length and transitivity than the forested network, indicating less interaction across taxa in urbanized microbial communities (Faust et al., 2012).

Microbial co-occurrence network structure is responsive to environmental disturbance (Ruiz-Moreno et al., 2006; Parter et al., 2007), which may explain the difference in size and structure between the forested and urbanized networks. In urbanized streams, environmental conditions (e.g., temperature, hydrology) were likely more variable and more extreme than in forested streams (Nelson and Palmer, 2007; Coleman et al., 2010; Stanley et al., 2010). Previous studies examining the impact of disturbance on microbial co-occurrence networks observed an increase in modularity with increasing disturbance, so long as disturbances were short-term in nature (Ruiz-Moreno et al., 2006; Parter et al., 2007). In this study, modularity was roughly equivalent in urbanized and forested networks, but the number of keystone-like taxa was drastically reduced in the urbanized network. This would be the case if less-adapted taxa become completely inactive due to the long-term nature of urbanization's effect on watersheds. Our measurement of diversity can include taxa that have high sequence abundance, but are functionally unimportant or largely dormant (Shi et al., 2012; Stewart et al., 2012). Thus, these microbes may persist in the environment, but no longer interact with other microbes—in which case these taxa would no longer appear in microbial co-occurrence networks. We propose that the smaller size of the urbanized network reflects a reduction of bacterial taxa that exhibit keystone-like behavior in urbanized streams as some previously well-adapted taxa become less functionally important under urbanized conditions.

For both forested and urbanized networks, OTU node degree was positively correlated with the proportion of sequences for that OTU found in sediment samples (Figure 7). This means that taxa that interact with many other microbial taxa are more likely to be found in sediment than water column environments, providing additional support for the claim that stream sediment communities are more strongly linked to environmental conditions within a stream reach than are water column communities. In light of the lines of evidence presented in this study, we argue that the impact of urbanization on bacterial community composition lends evidence to the assertion that different metacommunity processes drive bacterial community composition in sediment vs. water column habitats (Leibold et al., 2004; Souffreau et al., 2014). The relationship with watershed percent impervious cover was stronger for the water column microbes than sediment communities (Figure 4B). On the other hand, sediment bacterial communities were more strongly linked to the environmental milieu—including conductivity, sediment grain size, and sediment carbon content—than nearby water column communities (Figure 8). All of the top indicators of water column microbial communities (Figure 5, Table 3) were taxa that have been found to be associated with groundwater or landscape sources in previous studies. Candidate Division OP3 and OD1 are both linked to groundwater environments (Glöckner et al., 2010; Nelson and Stegen, 2015) and members of Oxalobacteraceae are often associated with soils and plants as well as aquatic habitats (Baldani et al., 2014). This provides further evidence that actively dispersing water column microbes are akin to a regional pool of taxa, controlled in large part by mass effects of microbes imported from the surrounding watershed. In relatively stable stream bed sediment environments, species sorting by local environmental factors determines which subset of the dispersed bacterial taxa can become established (Beisner et al., 2006; Souffreau et al., 2014). Our findings suggest that urbanization impacts different parts of the stream environment in fundamentally different ways: water column communities are directly influenced by urbanization while sediment microbial communities are indirectly impacted via alterations to the stream environment and via changes in the inoculating pool of organisms.

In laboratory experiments, Souffreau et al. (2014) demonstrated that bacterial community composition was controlled by species sorting, except in cases where bacteria immigration rates were high. The water column bacteria assemblage in our study streams likely represents such a high-turnover community. All stream sites sampled were located within 250 m of the stream origin. Thus, it is not unreasonable that most bacteria sampled in the water column during baseflow originate within the watershed, which would explain why water column community composition was more strongly related to watershed impervious cover than sediment community composition. Ultimately, as water flows downstream, we anticipate that the selective pressures of the aquatic environment will result in a more typical pelagic community (Adams et al., 2014). In the headwaters, the water column microbial community appears to be controlled by land use activity in the watershed. This correlation between watershed land cover and the dispersing community has an impact on the sediment community as well, causing a cascade of effects that shape the functional capacity of stream ecosystems.

Bacterial community composition was most strongly related to stream habitat (water vs. sediment) but within a habitat type urbanization was significantly correlated with composition, particularly for the water column community. In contrast, the sediment community composition was more strongly linked to local environmental conditions (e.g., conductivity, sediment C:N). We suggest that these results support using a metacommunity framework to describe how watershed urbanization changes stream bacterial community composition as has been done for microbial biogeography in other aquatic systems (Crump et al., 2007). With such a framework, community composition in the water column is driven by regional factors related to watershed land use, while sediment bacterial community composition is more strongly controlled by local physicochemistry. Bacterial co-occurrence network analysis showed that urbanization can have substantial implications for microbial community interactions—and potentially functional diversity—even though broader measures of diversity showed no differences. The result was an overall loss in the number of keystone-like taxa in urbanized streams, which implies a loss of functional capacity in headwater streams with increasing watershed impervious cover. It is worth noting however, that due to the nature of land use patterns within the Parkers Creek watershed, the urbanized sites were clustered in the western portion of the watershed which could have resulted in some confounding effects. It is very difficult to find watersheds in which urban sites and forested sites are evenly distributed, but future studies may wish to explicitly consider how spatial arrangement of land use influences stream microbial communities.

JH designed and carried out the work with input from BCC, CF, and MAP. CF assisted with field and laboratory work. BCC assisted with genetic analysis. All authors contributed to drafts and the final manuscript.