Rapid and Stable Microbial Community Assembly in the Headwaters of a Third-Order Stream

Headwater streams are critical introduction points of microbial diversity for larger connecting rivers and play key roles in the establishment of taxa that partake in in-stream nutrient cycling. We examined the microbial community composition of a first- through third-order stream using fine-scale temporal and spatial regimes. Our results show that the bacterioplankton community develops rapidly and predictably from the headwater population with increasing total stream length. Along the length of the stream, the microbial community exhibits substantial diversity loss and enriches repeatedly for select taxa across days and years, although the relative abundances of individual taxa vary over time and space. This repeated enrichment of a stable stream community likely contributes to the stability and flexibility of downstream communities.

ABSTRACT

Small streams and their headwaters are key sources of microbial diversity in fluvial systems and serve as an entry point for bacteria from surrounding environments. Community assembly processes occurring in these streams shape downstream population structure and nutrient cycles. To elucidate the development and stability of microbial communities along the length of a first- through third-order stream, fine-scale temporal and spatial sampling regimes were employed along McNutt Creek in Athens, GA, USA. 16S rRNA amplicon libraries were constructed from samples collected on a single day from 19 sites spanning the first 16.76 km of the stream. To provide context for this spatial study and evaluate temporal variability, selected sites at the stream’s upper, mid, and lower reaches were sampled daily for 5 days preceding and following the spatial study. In a second study, three sites at and near the creek’s headwaters were sampled daily for 11 days to understand initial bacterioplankton community assembly. Both studies revealed decreasing alpha and beta diversity with increasing downstream distance. These trends were accompanied by the enrichment of a small fraction of taxa found at low abundance in headwater-proximal sites. Similar sets of taxa consistently increased in relative abundance in downstream samples over time scales ranging from 1 day to 1 year, many of which belong to clades known to be abundant in freshwater environments. These results underpin the importance of headwaters as the site of rapid in-stream selection that results in the reproducible establishment of a highly stable community of freshwater riverine bacteria.

IMPORTANCE Headwater streams are critical introduction points of microbial diversity for larger connecting rivers and play key roles in the establishment of taxa that partake in in-stream nutrient cycling. We examined the microbial community composition of a first- through third-order stream using fine-scale temporal and spatial regimes. Our results show that the bacterioplankton community develops rapidly and predictably from the headwater population with increasing total stream length. Along the length of the stream, the microbial community exhibits substantial diversity loss and enriches repeatedly for select taxa across days and years, although the relative abundances of individual taxa vary over time and space. This repeated enrichment of a stable stream community likely contributes to the stability and flexibility of downstream communities.

INTRODUCTION

Riverine systems act as an interface between many distinct habitat types, connecting hillslope and bottomland soils to downstream bodies of water and controlling the flow of bacteria and nutrients from one environment into the next. In many fluvial systems, alpha and beta diversity both decrease along the length of a stream or stream network (1–11). This suggests that headwater streams are a critical source of microbial biodiversity. The primary source of microbes entering headwater bacterial communities appears to be soil and soil waters (1, 2, 8, 12), with pelagic stream community assembly resulting from the enrichment of bacteria present in these environments at low abundance. From the headwaters, the bacterioplankton community continues to develop as it travels through the stream network (10, 11, 13).

Pelagic stream communities are renewed continuously via the paired forces of in-stream selection and bulk transport of organisms from upstream and upslope environments. This raises the possibility of rapid and significant shifts in community composition within streams resulting from fluctuations in physicochemical conditions, flow rates, and the composition or origin of microbes entering the stream. Temporal trends in community profiles have been described in many riverine systems (3, 5–7, 12–18). In several of these studies, the same taxa were found over time, varying in abundance based on shifts in physiochemical factors or following landscape-level disturbances (5, 10, 13, 17–20). This reoccurrence of taxa has led several groups to suggest the presence of a core community that is of particular importance in shaping riverine community dynamics (2, 9, 10, 13, 21–23). Conversely, multiple studies have also found substantial variation in microbial community structure and function over time in a wide range of study systems (3, 5, 6, 10, 15–18, 20, 22, 24). A temporal study of headwater streams and an adjoining higher-order stream by Portillo et al. (24) revealed seasonal variation among samples taken from the same site. Despite their proximity, samples clustered by site regardless of collection day (24). These findings indicate either that the fluvial core community fluctuates over time or that changing environmental conditions lead to alterations in the relative abundance of core versus transient community members.

Few studies have evaluated the extent of short-term variability between the microbial communities of headwater streams as well as how daily fluctuations in upstream community composition may impact downstream community structure and function. Further investigation into the assembly and stability of small stream bacterioplankton populations on both fine temporal (days) and spatial scales is necessary to advance the current understanding of the initial development of freshwater pelagic communities. Fundamental questions remain, including (i) to what extent are fine-scale spatial patterns of community diversity and composition reproducible over time, and (ii) do in-stream communities exhibit a temporal memory or distance-decay relationship and at what time scale(s)? To directly assess this knowledge gap, two multiday studies on a small creek in Athens, GA, USA, were performed. In the first, we conducted an intensive geographical study consisting of 19 sampling sites along an ∼16-km transect of the stream to evaluate spatial patterns of community diversity. To evaluate temporal variability in microbial community assembly, three of these sites (at upper, middle, and lower reaches of the stream) were sampled for the 5 days preceding and following the spatial study. As rapid changes in the community composition were observed in the upper reaches, a second study was conducted in which three headwater-proximal sites were visited daily for 11 days to more completely characterize temporal variability in this region of the stream. Three additional sites in this upper section of the creek were sampled at the beginning, middle, and end of the second study to provide a detailed snap shot of spatial variation.

Our chosen study site, McNutt Creek (Fig. 1A), has consistently demonstrated a loss of diversity with increasing dendritic distance as well as a corresponding increase in common freshwater taxa across seasons, as seen in larger riverine systems and higher-order streams (25). The sampling sites along the creek capture bacterial community development at first- through third-order locations (first order, no confluences with smaller streams; second order, positions downstream of a confluence with a first-order tributary; third order, positions downstream of a confluence between two second-order streams). These characteristics make McNutt Creek an ideal study system to address these questions regarding bacterial community assembly in headwater streams. In this work, a rapid rise in dominance of freshwater-associated bacterial taxa along the stream flow path was observed. Community assembly was renewed daily and quickly recovered following landscape-level disturbance, revealing robust community stability across the stream’s flow path.

FIG 1.

Transitions in bacterial communities from the headwaters to downstream sites on McNutt Creek. (A) McNutt creek is shown in dark blue, and all samples were taken along its length (at numbered collection sites). Tributaries are shown in light blue. (B) The relative abundances of OTUs at each location. All OTUs representing 1.5% or more of the community in any sample in the full study (including samples not shown here) are displayed. The legend lists the highest taxonomic level assigned to each OTU and is sorted and colored by phyla (greens, Actinobacteria; blues, Bacteroidetes; purples, Firmicutes; reds, Proteobacteria; yellows, Verrucomicrobia). OTU numbers are constant across all figures. All taxa present at <1.5% of all samples are grouped into the “other” category (gray). Shannon diversity (C) and OTU richness (D) are plotted against log cumulative dendritic distance. (E) PCoA, with samples colored according to their location along the stream path. Inset displays PCoA 1 plotted against log cumulative dendritic distance. Maps created using ArcMap 10.5.

RESULTS

Physiochemical results.

Two studies of temporal and spatial dynamics in microbial community composition in McNutt Creek were conducted in 2016 and 2017. The initial study (in 2016) examined community progression and temporal stability along the length of the creek. To capture a high-resolution spatial picture of community progression, 19 sites along the entire creek length were sampled on a single day. To evaluate the historical context and temporal progression of observed patterns, water was also collected at three sites (sites 3, 12, and 18) for the 5 days immediately preceding and following the 19-site survey. Table S2 in the supplemental material presents environmental metadata associated with the samples, including dissolved oxygen (DO), water temperature, land use, and land cover parameters. Most measures did not show significant trends along the length of the stream, with the exception of temperature (rho = 0.66, P = 3 × 10⁻³). There was measurable precipitation on the 11th (0.33 cm), 12th (0.31 cm), 14th (1.56 cm), and 17th (0.20 cm) of June for a total of 2.39 cm of rain over the course of this study. All samples collected on 14 June were taken before the rainfall, with the exception of site 4.

A second study was completed in 2017 to investigate community composition change at the headwaters and focused on the six upstream-most sites (sites 1 to 6). Sites 1, 3, and 6 were sampled daily for 11 days, and sites 2, 4, and 5 were sampled on days 1, 6, and 11 to contextualize these results. Average daily air and water temperatures were slightly lower than observed during the 2016 study, while total dissolved nitrogen (TDN), total dissolved phosphorus (TDP), and dissolved organic carbon (DOC) were measured at site 3 only and found to be comparable to measurements taken in 2016 (Table S2). A total of 3.03 cm of rainfall fell during the 2017 study, with measurable precipitation occurring on 30 August (0.43 cm) and 31 August (2.59 cm).

Longitudinal study of McNutt creek.

In the 2016 longitudinal study, alpha diversity (Shannon’s) significantly decreased with increasing cumulative dendritic distance (CDD; P < 1 × 10⁻⁴, rho = −0.63) (Fig. 1C). This trend was confirmed using Simpson’s index (P = 0.02, rho = −0.54). Richness, on the contrary, exhibited no significant trend over the stream length when calculated using total number of unique operational taxonomic units (OTUs; in rarefied data) or Chao1 index (Fig. 1D).

To explore changes in microbial community structure and beta diversity, principal-coordinate analysis (PCoA) was completed using rarified samples (Fig. 1E). It is of note that principal-component analysis (PCA) and nonmetric multidimensional scaling (NMDS) ordinations were also run for all instances in which PCoAs are reported and yielded highly similar results. Log-transformed total CDD correlated strongly with the position of samples along the first principal coordinate (Fig. 1E, inset) (P < 1 × 10⁻⁴, R² = 0.87). Upstream samples are scattered but downstream sites are clustered more tightly, suggesting a decrease in variance with increasing CDD. The incorporation of physiochemical, land use, and nutrient data suggests that tree cover type (deciduous upstream and evergreen downstream), dissolved oxygen levels, water temperature, and urbanization factors (impervious surface level and ground cover type) may influence microbial community structure (see Fig. S1). However, given that only a single linked flow path was examined, these relationships are difficult to interpret and may be autocorrelative.

To directly assess distance-decay relationships along the stream length, Bray-Curtis dissimilarity was calculated between each sample site and plotted against the length of stream between them (Fig. S2). These relationships were calculated in terms of upstream and downstream proximity. For both weighted and unweighted measures, 12 relationships were significant (P < 0.05). All sites downstream of site 13 displayed significant dissimilarity between adjacent sites, indicating a downstream increase in similarity decay.

In addition to the analysis of community-level trends, we examined longitudinal trends in the abundance of individual taxa (OTUs). Among the 250 most abundant taxa, a statistically significant negative correlation (Spearman’s, false discovery rate [FDR]-adjusted P < 0.05) was found between the relative abundances of 75 OTUs and CDD (Fig. 2). Together, negatively correlated taxa represent 48.83% of sequences in the headwater site and 6.53% of sequences at site 19 (Fig. 2B). In contrast, a statistically significant positive correlation was found between 35 OTUs and total CDD (Fig. 2A). Positively correlated taxa include OTUs belonging to the freshwater-associated Actinobacteria clade Luna1-A OTU2, and Bacteroidetes clades bacIII-A OTU3 and bacI-A OTU6 (nomenclature as reported by Newton et al. [26]) (Fig. 2A). All positively correlated taxa together represent 4.07% of sequences in the furthest upstream site (site 1) and 45.74% of sequences in the furthest downstream site (site 19).

FIG 2.

Changes in the abundances of positively and negatively correlated OTUs along the stream length. The relative abundances of all significantly positively (A) or negatively (B) correlated taxa comprising at least 1% of any sample are displayed in color. All additional taxa below this threshold are reported as “others” in gray, and any taxa that are unclassified beyond the specified level are labeled with the abbreviation “un.” OTU numbers are reported to the right of all taxa for reference.

Daily fluctuations in community compositions at the upper, middle, and lower reaches of McNutt Creek.

Daily assessment of sites 3, 12, and 18 during June 2016 revealed that the taxonomic profile of the community at each site was relatively stable at all taxonomic levels, including at the level of individual OTUs (Fig. 3A). For site 3, 675 OTUs (of 14,860 observed) were shared across the entire 11-day time series, representing an average of 96.11% (range, 95.12% to 97.20%) of recovered sequences. For site 12, 513 OTUs (of 15,022) were shared across the sampling period, representing an average of 96.75% (range, 95.47% to 97.64%) of recovered sequences. For site 18, 413 OTUs (of 15,122) were shared across the 11-day time series, representing an average of 95.17% (range, 88.30% to 97.47%) of recovered sequences. Across all dates and times, 344 OTUs were shared, representing 93.61% of recovered sequences.

FIG 3.

Daily analysis of bacterioplankton communities across the length of the creek. (A) The relative abundances of OTUs present at >1.5% of any one sample are displayed (all other OTUs are grouped under “other”). (B) Shannon diversity plotted against day of sampling for each of the three selected sites, with sites 3, 12, and 18 shown by red, green, and blue dots, respectively. This color scheme was used to show sample ordination in the PCoA plot (C), in which envfit was used to calculate loadings of the parameters listed in Table S1 in the supplemental material. (D) Bray-Curtis dissimilarity was determined for and between all samples taken from each site. Dissimilarity calculated between sites is denoted in the format of site number versus site number, i.e., “3 v 12.” Blue dots represent individual samples and open circles represent outliers. *, significant difference between sample sets.

Alpha diversity measurements fluctuated at all sites throughout the study (Fig. 3B). The bacterial community at site 3 was the most diverse (Shannon index, P < 0.01 versus sites 12 and 18). When individual days were considered, site 3 had a higher diversity (Shannon) than the other two sites on 9 of 11 sampling dates. The differences in Shannon indices between sites 12 and 18 were smaller and not statistically significant. The relationship between these sites on individual days was also less predictable, with site 18 exhibiting a Shannon of less than 12 for only 4 of 11 days.

Beta diversity analyses suggest that site 3 hosted a distinct community from downstream sites. Weighted Bray-Curtis analyses show the median dissimilarities between site 3 and the two downstream sites are markedly higher than the median dissimilarity between sites 12 and 18. Similarly, a PCoA shows a clear separation of site 3 from the downstream samples (Fig. 3C), while samples from sites 12 and 18 cluster together, which may be driven by differences in water temperature (P < 1 × 10⁻³). Permutational multivariate analyses of variance (PERMANOVAs) revealed that sites but not sampling days had a significant effect on community composition (P = 1 × 10⁻³ and 0.28, respectively) and that sites and days together were also significant (P = 1 × 10⁻³ and 4 × 10⁻³, respectively). To assess community progression at individual sites, Bray-Curtis dissimilarities between samples collected 1 to 5 days apart at the same site were calculated. These analyses show temporal trends in community similarity over time for abundance-weighted (P = 0.01 for sites 3 and 18) but not unweighted (presence/absence) measures (see Fig. S3). No significance was found for either measure at site 12.

To understand which taxa consistently increase or decrease downstream, downstream trends in the abundance of the 250 most-abundant taxa were analyzed. Two hundred thirteen were found to decrease downstream on at least 1 day. ZB2 OTU208, ABY1 OTU280, unclassified Rhodospirillaceae OTU38, and bacII-A OTU4 all decreased downstream on 10 of the 11 sampling days. In contrast, 94 of the 250 were identified as increasing on at least 1 day, although only 12 increased on at least 5 days. BacI-A OTU6 and bacIII-A OTU3 were found to increase between sites 3, 12, and 18 during multiple days. Alphaproteobacterium Alf-V OTU193 was the most consistent OTU of the top 250, increasing down the stream length for 9 of the sampling days. Other organisms, such as Luna1-A OTU2, bacV OTU10, betI-A OTU1, and betIV OTU11, that increased along the full flow path in the 19-site study only exhibited increases across these three sites for a few of the individual sampling days.

Temporal and spatial trends in community assembly at headwater-proximal sites.

The 2017 spatial and temporal study examined community assembly near the stream headwaters in greater temporal and spatial detail. Three longitudinal studies of the 6 most uppermost stream sites were performed at 5-day intervals to evaluate spatial and temporal trends in community composition (Fig. 4). Across these three collections, qualitatively similar trends in taxonomic composition were observed (Fig. 4A). This result was supported by the outcome of PERMANOVA used to test the effects of site and day on headwater community composition. The effect of site but not collection day was significant (P = 0.04 and 0.12, respectively), and the effect of site and day together was significant (P = 0.02 and 0.04, respectively). Alpha diversity was significantly negatively correlated with log CDD on the second (but not the first or third) longitudinal sample collection (P = 4.5 × 10⁻³) (Fig. 4B).

FIG 4.

Community analysis of all headwater site samples collected at three time points in 2017. (A) The relative abundances of taxa present in each sample are displayed according to the parameters defined in the legend for Fig. 1. (B) Diversity for each sample plotted according to the CDD measurement for each site.

Daily fluctuations in microbial community composition were evaluated at sites 1, 3, and 6 from 24 August to 3 September 2017 (Fig. 5A). Alpha and beta diversity were consistently greater at site 1 and lower at sites 3 and 6, in a manner similar to that of the whole stream (Fig. 1C). Site 1 was found to be significantly more diverse than sites 3 and 6 (Shannon, P = 5.7 × 10⁻³ and <1 × 10⁻⁴, respectively), which were not significantly different. Day-to-day beta dissimilarity was assessed for each focus site (site 1, 3, and 6) to understand community change over time (Fig. 5D). The median weighted Bray-Curtis dissimilarity across time was higher at site 1 than either downstream site and significantly different from site 3 by presence-absence measures (P = 0.02) and from site 6 by both measures (P = 3.4 × 10⁻³ and <1 × 10⁻⁴). At site 1, 197 OTUs of 15,338 were shared across the entire 11-day time series, representing an average of 87.11% (range, 81.83% to 94.35%) of sequences recovered from this site. For site 3, 97 (of 15,438) were shared across the sampling period, representing an average of 86.31% (range, 71.42% to 98.70%). This was substantially lower than the 675 OTUs shared at this site during the 2016 daily study. For site 6, 34 OTUs (of 15,501) were shared across the 11-day time series, representing an average of 91.06% (range, 76.18% to 98.65%) of recovered sequences.

FIG 5.

Analysis of daily water samples collected from three headwater-proximal sites across multiple days. (A) Relative abundances of all taxa present in each sample are displayed according to parameters defined in the legend for Fig. 1 (B) Community diversity observed at each site over the course of the study, with samples from sites 1, 3, and 6 represented by red, green, and blue points, respectively. (C) Ordination of all daily samples in the resulting PCoA plot. (D) Bray-Curtis dissimilarity was determined for and between all samples taken from each site. Dissimilarity calculated between sites is denoted in the format of site number versus site number, i.e., “1 v 3.” Blue dots represent individual samples and open circles represent outliers. *, significant difference between sample sets.

In a PCoA of these data (Fig. 5C), headwater samples (site 1) clustered separately from those from sites 3 and 6, with the exception of samples taken after the rain event (days 8 and 9). A PERMANOVA revealed a significant effect of site but not sampling day on community composition (P = 1 × 10⁻³ for site and 0.12 for day); however, the effect of sampling day and sites together was significant (P = 0.03). In addition, dissolved oxygen (DO) content was found to correlate significantly with the PCoA (P = 6 × 10⁻³ and 0.02 for DO% and DO mg/liter, respectively). No individual sites exhibited significant temporal distance-decay relationships in Bray-Curtis dissimilarity, with samples taken 24 h apart showing similar levels of dissimilarity as samples taken up to 5 days apart (see Fig. S4).

Multiple OTUs were present across the time series. All three sites were consistently dominated by betI-A OTU1 during the study period, and no consistent trend in its abundance was observed between sites. However, other Proteobacteria showed positional trends in representation. For example, alfVI OTU26 and betIV OTU11 were found to increase on 5 days, though they showed decreases on a single study day. Other proteobacterial OTUs were noted to decrease longitudinally on several days over the study period, including Telmatospirillum OTU46, Bdellovibrio OTU17, and Geobacter OTU47. As was observed in 2016, Luna1-A OTU2, and bacIII-A OTU3 increased across the three sites on 8 or more of the study days. Interestingly, bacII-A OTU4 exhibited a positive trend with total CDD on 6 of the 11 sampling days in 2017. These results are in stark contrast to the negative trends OTU4 exhibited for 10 of 11 days in 2016.

The 2.54-cm rain event captured between day 7 and 8 (30 and 31 August) had variable effects on the abundances of prominent taxa at each site. As mentioned above, several Proteobacteria, which increased along the sampled flow path on other days, decreased on either day 8 or 9. This was also the case for bacV OTU10, though this OTU had inconsistencies in trend direction in the 2016 study. Taxa that typically showed positive trends with CDD (e.g., Luna1-A OTU2, bacIII-A OTU3, and bacII-A OTU4) did not on days 8 or 9 but returned to exhibiting trends in increasing on day 10.

Comparisons of community variation across daily and annual time scales at a single site.

Site 3 was sampled in common across the two studies and used to evaluate community variability at both daily and annual time scales. A clear difference in the relative abundances of multiple bacterial phyla was observed across the two time series (Fig. 4 and 5). In particular, Proteobacteria were found at higher abundances in 2017, while Bacteroidetes and Actinobacteria were found at lower abundances (P < 0.01 for all comparisons). At the OTU level, 756 of the 770 taxa found in 90% of the 2017 site 3 samples were also found in 90% of samples collected in 2016. While only 25 of the 8,943 OTUs detected across all samples in 2017 were also present in all samples collected in 2016, this included 24 of the 100 most abundant OTUs.

The 2017 daily time series exhibited significantly greater day-to-day dissimilarity by both weighted and unweighted measures than observed in 2016 (t test, P = 0.01 and P < 1 × 10⁻⁴, respectively) (Fig. 6). In fact, day-to-day variability in 2017 was similar in scale to cross-year variability between 2016 and 2017. To account for the potential effects of the rain event on day 8 of the 2017 study, significance was reevaluated by removing days 8 and 9 from the data set and rerunning dissimilarity calculations. Weighted dissimilarity was still significantly different although less so (P = 0.03). Unweighted dissimilarity, however, was no longer significant (P = 0.97). A principal-coordinate analysis showed some degree of clustering within years, with the 2017 samples largely separating along dimension 1 and 2016 samples separating along dimension 2, which may be driven by significant differences in water temperature between the two samples sets (P < 1 × 10⁻³).

FIG 6.

Comparison of site 3 across years. (A) Resulting ordination plot of PCoA. Samples collected in 2016 and 2017 are shown in red and blue, respectively. The envfit function was run with daily metadata, but no values were significant, with the single exception of temperature. (B) Bray-Curtis dissimilarity was calculated for samples from each year and between years. *, significant difference between sample sets.

DISCUSSION

This work examined the bacterial community assembly at a daily resolution in a third-order stream located in a temperate urban watershed. While pelagic freshwater bacterioplankton communities were previously described around the world, the stability and renewal of these populations in headwater streams remains in question. In particular, we aimed to elucidate, on fine temporal and spatial scales, the degree to which lower-order streams mimicked the community assemblages and diversity trends observed in higher-order rivers and whole watersheds (1, 2, 4, 8, 9, 11).

Along the length of McNutt Creek, an inverse relationship between alpha diversity and CDD was observed, which was previously reported in other river systems (9, 10, 12, 23). The patterns of decreasing alpha diversity with downstream distance traveled were most apparent across the full stream length (Spearman’s R² = 0.63, P = 5 × 10⁻³), however, site-to-site fluctuations were apparent in both daily studies. It is also of note that alpha diversity calculations for the 2017 study were typically lower than those recorded in 2016, although the cause of this diversity loss is unknown.

Downstream trends in site-to-site beta diversity comparisons were present but relatively weak, primarily exhibiting differences between the headwater and furthest sites downstream. For example, significant unweighted beta diversity relationships were found between site 3 and both 12 and 18 in the 2016 study (Fig. 3D) (P < 1 × 10⁻⁴ for both). When dissimilarity was assessed along the entire stream length on a site-by-site basis, the majority of sites were found to have a significant positive correlation with distance, particularly at the extreme ends of the stream (see Fig. S3 in the supplemental material). During 2017, the only significant beta diversity relationships were between site 1 and both downstream sites, 3 and 6. These results confirm earlier findings in the broader Upper Oconee watershed, where three of five seasons exhibited trends in beta diversity loss with increasing cumulative dendritic distance (10). This pattern of dissimilarity decreases with increasing hydrologic distance is not unique to the Upper Oconee watershed and was also observed across the Ybbs river network in Austria (2).

The effects of time between samples collected on community structure were apparent at some but not all sites, suggesting a limited impact of the previous community composition on later ones. A similar trend in diversity was recorded in rock biofilms in a headwater stream, in which weekly samples from the same site were less similar than those taken from other locations along the creek, although site locations were considerably closer together than in the present work (27). As dispersal in streams is primarily unilateral as microbes are passively transported, this speaks to the strength of renewal of these communities and their resilience to immigrant taxa from neighboring environments.

Water temperature is the sole environmental factor that was found to have a statistically significant correlation with community composition over both time and space, suggesting that water temperature may play a key role in shaping stream community composition. None of the other physiochemical parameters measured in this study showed consistent statistically significant correlations with community progression or composition. These results are not entirely surprising given the results of a previous study on the greater Upper Oconee watershed, which showed that the position within the watershed had far more significant impacts on community composition than individual physicochemical factors (10). In addition, studies in other fluvial and freshwater systems have revealed substantial temporal and spatial variability in the relative importance of different physiochemical parameters in shaping community composition (25, 28). Land use parameters associated with development were significantly associated with community composition across sites (Fig. S1), and similar relationships were also noted by de Oliveira and Margis (13). However, in this study, these relationships are more likely to be a coincidental finding due to the fact that downstream portions of this stream happened to be more highly developed.

During the 2017 sampling period of headwater sites, a storm event resulting in 2.54 cm of rainfall occurred. Rainfall was previously shown to introduce taxa into freshwater systems and increase diversity (22, 29, 30). While our results are consistent with this, it is interesting that each of the three sites sampled appeared to be affected differently by the influx of rainwater. At site 1, we observed a notable loss of alpha diversity paired with the introduction of Bacteroidetes and Actinobacteria, while site 3 exhibited a noticeable increase in total diversity, and site 6 exhibited a decrease in diversity and an increase in the fraction of Proteobacteria (Fig. 2). All sites returned to pre-rain community compositions within 48 h (Fig. 4). Several other minor precipitation events occurred during both studies; however, no strong disturbance was detected in the community data. While further studies are merited to better understand both the consistency in disruption and recovery of bacterioplankton populations along the entire stream reach, these data suggest that headwater bacterial communities are highly resilient to rainwater influx.

Throughout both 2016 and 2017 studies, a small set of taxa found at low abundance in the headwaters of the stream was enriched in downstream sites. This enrichment corresponded with the significant decreases of many OTUs prevalent at the headwater sites. These findings support the hypothesis that freshwater streams function as a major site of selection and species sorting (2, 8, 10, 12). The set of taxa that were specifically enriched in downstream environments was relatively consistent over time. Multiple OTUs were identified as significantly positively correlated with stream length across days. Many of these OTUs belonged to well-known “typical” freshwater bacterial clades (10, 23, 25, 26), including Luna1-A (OTU2), bacIII-A (OTU3), bacVI (OTU90), bacV (OTU10), betI-A (OTU1), betIV (OTU11), and bacI-A (OTU6). It is of note that in the 2016 study, a shift in prevalent Bacteroidetes members from bacII-A OTU4 in upstream environments to bacI-A OTU6 and bacIII-A OTU3 further downstream occurred, suggesting that certain Bacteroidetes clades are better adapted either for life in different regions of the stream or for long-term exposure in freshwater environments.

These data parallel findings in the existing literature in which a consistent freshwater stream microbiome has been proposed (2, 9, 10, 13, 19, 21, 31), though the exact definition of this population and the nomenclature describing this phenomenon have been highly variable. For example, the “core” freshwater community of stream biofilms and sediment was defined by Besemer et al. as taxa found in at least 50% of all samples (2). A stricter definition was employed in de Oliveira and Margis’ seasonal study of an entire river length, describing the core community as taxa that persisted across all samples and time points (13). In contrast, Ruiz-Gonzalez et al. (21) defined the “core seed bank” of bacterial taxa as those that were more abundant in downstream than upstream sites versus “restricted” taxa that decreased in abundance with increasing downstream distance. (21). It is clear that a select and consistent set of taxa dominate freshwater habitats in fluvial systems at all scales (23, 25, 28, 31). However, it is also clear that the abundances of these microbes can vary substantially, both within and between fluvial networks (28). While these variations are often posited to be driven by physicochemical variables, the relationships found between specific taxa and different environmental variables have often varied across studies (4, 16) and, in some cases, even across time within a single system (19). With these and other works, it becomes increasingly apparent that the concept of a “typical” freshwater microbiome (25) extends across all levels of the freshwater continuum, with even a single stream exhibiting strong and reproducible selection for a small subset of microbial taxa that exhibit variable abundances over time.

Conclusions.

It is well documented that, globally, freshwater ecosystems are dominated by a consistent set of taxa known as typical freshwater bacteria (26). Headwater streams act as a primary entry point for soil and soil water bacteria into river networks and were previously shown to host elevated diversity levels compared to those of higher-order water bodies. In this work, we demonstrate temporally stable trends in bacterial diversity and taxonomic representation along the length of the creek. While longitudinal trends in bacterial diversity across watersheds were previously documented, this study demonstrated consistent enrichment of sequences belonging to multiple OTUs (97% identity sequence clusters) across time scales ranging from days to years. Furthermore, we found striking community stability at individual stream sites, with little evidence for a temporal distance-decay relationship in community composition at individual sites, and interannual variation that was comparable in scale to daily variation. These findings reinforce the hypothesis that entry into freshwater fluvial systems serves as a strong selective filter resulting in the enrichment of a select set of bacterial taxa. In addition, this work suggests that downstream ecosystems are strongly structured by rapid community assembly processes taking place in headwater-proximal streams.

MATERIALS AND METHODS

Study site description.

McNutt Creek is a 20-km-long stream in Athens, GA, USA, that flows through a mixed land use area, spanning agricultural, residential, and forested sections (Fig. 1A, see also Table S2 in the supplemental material). The stream width ranges from 1.10 m (site 1) to 5.18 m (site 19). McNutt Creek is part of the Upper Oconee Watershed, a temperate urban watershed that provides drinking water for the city of Athens and the surrounding area.

Sampling schemes.

Pelagic water samples were collected from McNutt Creek, Athens, GA, USA (33°55′50.2″N, 83°30′30.91″W). Nineteen sampling locations were selected based on their location in the stream network (headwaters versus main stem) and the ability to access these sites, as some were located on private property and required the consent of the owners. Water samples were collected in 2016 and 2017 to assess community assembly and stability along the headwater (1st order) to main stem (3rd order) longitudinal gradient. In 2016, three streams (sites 3, 12, and 19) were sampled daily from 9 to 19 June. On 14 June, all nineteen sites were sampled within 7 h. Three groups of three to four volunteers each were sent out to survey a third of the creek, working from the most upstream site to those downstream.

At each site, water was collected mid-stream from the water column approximately at mid-depth in 4-liter acid-washed Cubitainers. Each container was rinsed with stream water three times before the final sample was collected. Water was processed either immediately on site or within 1 h of collection following the sampling methods outlined by Hassell et al. (10). Samples were run through sterilized tubing with in-line filtration through a 5.0-μm 47-mm-diameter SVPP prefilter (Millipore) to capture particulate matter. The effluent was then run through a 0.22-μm 2-ml Sterivex filter (Millipore). A total of 500 ml of water was filtered for each sample. Filters were preserved at −80°C until DNA extraction. In addition to the filtered water samples, a YSI Professional Plus meter was used to measure temperature, pH, dissolved oxygen percentage, and conductivity in the field (Table S2). Dissolved organic carbon (DOC), total dissolved nitrogen (TDN), and total dissolved phosphorus (TDP) were measured for selected filtered water samples by the Stable Isotope Ecology Lab at the University of Georgia (Table S2). Daily air temperature measurements were available through Weather History for Athens (https://www.wunderground.com/history/airport/KAHN/), and daily precipitation measures were available through the National Weather Service (https://www.weather.gov/ffc/rainfall_scorecard).

In 2017, sampling was focused on headwater-proximal sites (sites 1 to 6) to determine the reproducibility of initial community assembly. Sites 1, 3, and 6 were sampled daily from 24 August to 3 September. On 24 and 29 August and 3 September, samples were collected from all six uppermost sites (sites 1 to 6). The 2017 samples were collected using the same methods as used in 2016.

Watershed characteristics.

Geographic information system (GIS) analysis was used to determine physical differences between each site’s watershed, drainage network, and land use/land cover. ArcMap 10.5 was used to calculate each sample site’s watershed area, total cumulative dendritic distance (CDD), land use/land cover characteristics, and impervious surface area. Each sample point’s watershed was delineated using “hydrology” tools in the Spatial Analyst toolbox and the national elevation data set’s 30-m digital elevation model (DEM) (32). Flow direction and flow accumulation rasters were created from a hydrologically corrected DEM (“fill” tool was used to correct the DEM). After the sampling points were associated with the flow accumulation raster (using the Snap Pour Point tool), each watershed was delineated using the “watershed” tool. Total dendritic distance was calculated by using each sampling point’s watershed to clip the high-resolution national hydrography data set’s stream layer (33). Similarly, each watershed was used to extract land use/land cover characteristics and impervious surface cover from the 2011 national land cover database (34, 35).

DNA extraction.

DNA was extracted from the Sterivex filters according to the methods outlined by Hassell et al. (10). Filters were thawed, and then 1 ml of lysis buffer (40 mM EDTA, 50 mM Tris [pH 8.3], 0.73 M sucrose) and lysosome dissolved in lysis buffer (2.11 mg/ml final concentration) were added and incubated at 37°C for 30 min while rotating. Proteinase K dissolved in lysis buffer (0.79 mg/ml final concentration) and 200 μl 10% SDS were added for a second incubation step at 55°C for 2 h. The lysate was extracted and mixed with an equal volume of phenol/chloroform/isoamyl alcohol (25:24:1; pH 8.0). Samples were centrifuged for 5 min at 3,500 × g, and the top phase was saved. Next, 0.04× volume 5 M NaCl and 0.7× volume isopropanol were added, mixed, and incubated at room temperature for 10 min. Samples were centrifuged for 15 min at 17,000 × g, and the supernatant was discarded. DNA was resuspended in 400 μl elution buffer (Omega Biotek, Norcross, GA, USA) and incubated at 65°C while rotating for 10 min. DNA was then processed using Omega Bio-tek’s E.Z.N.A. Water DNA kit according to the manufacturer’s protocol from step 13 through completion (Omega Bio-tek, May 2013 version).

PCR, sequencing, and analysis.

According to the methods outlined by Tinker and Ottesen (36), the V4 region of the 16S rRNA gene was amplified in each DNA sample. The resulting library was submitted to the Georgia Genomics Facility for sequencing (Illumina MiSeq 250 × 250 bp; Illumina, Inc., San Diego, CA).

Returned reads were processed using the mothur software package (37) according to the MiSeq standard operating protocol with the following modifications: reads that fell outside 200 to 275 bp were excluded from contig generation; SILVA reference database release v123 was used for sequence alignment; primers GTGCCAGCMGCCGCGGTAA and GGACTACHVGGGTWTCTAAT were used to perform in silico PCR; the VSEARCH algorithm was used to identify chimeric sequences (38); taxonomic classification was completed following the TaxAss workflow (39–41), using the FreshTrain data set to first classify as many sequences as possible using the freshwater-associated database and then classifying remaining sequences using the 5 May 2013 release of the Greengenes reference database, version 13_8. During taxonomic assignment, a bootstrap value of 70 was used; an additional remove.lineage command was run to ensure that sequences from cyanobacteria, chloroplasts, and unknown taxa were removed. Operational taxonomic units (OTUs) were called at 97% or greater sequence similarity. From the 91 samples that were collected from these studies, a total of 6,119,579 16S rRNA gene sequences passed quality filtering steps, resulting in an average of 67,248 sequence reads and 1,562 OTUs per sample (see Table S1). Rarefaction curves were calculated for each study and are shown in Fig. S5.

Statistical analysis was completed in R using the vegan package (42). When necessary for analysis, samples were rarefied to a depth of 3,291 sequences. The envfit function was used to calculate significant (P ≤ 0.05) physiochemical parameters and watershed characteristics that are displayed as loadings on ordination plots. The cor.test function was utilized to test for significance of Spearman’s correlations. t tests were used to determine significance between boxplots. When appropriate, the Benjamini-Hochberg procedure was used to adjust P values for false discovery rate correction.

Accession number(s).

The raw sequences from these experiments are available under the accession numbers SRP155540 in the NCBI Sequence Read Archive.