Introduction

Freshwater ecosystems are biogeochemical reactors that drive carbon processing and serve as central elements for the flux of dissolved organic matter (DOM) from land to the ocean1,2,3. Within this framework, bacterioplankton-mediated biogeochemical transformations play critical roles in aquatic food webs. Planktonic bacteria process DOM, a larger pool of organic carbon in aquatic ecosystems, transferring the carbon contained within their biomass to higher aquatic trophic levels4. Additionally, carbon can be transferred from DOM to the CO2 pool through bacterial respiration5,6,7. Despite their vital role in biogeochemical cycles and their rapid response to changes in environmental conditions, bacterioplankton are rarely included in monitoring and conservation strategies8,9. This exclusion is partly due to the unpredictable nature of bacterial communities responses to environmental conditions, which makes it difficult to empirically link their taxonomic composition to ecosystem function10. Although the biogeographical patterns of bacterioplankton communities in lakes and rivers imply that their distribution is not random11, we are only beginning to unravel how different factors and ecological processes shape the assembly of these communities.

Selection associated with changes in local aquatic conditions (e.g. pH, pollutants, and biotic interactions), is often attributed as the main factor determining variation in the composition of aquatic bacterial communities12,13,14. The species sorting hypothesis invokes the environment as a filter that selects taxa that grow under specific conditions15. Consequently, sites with similar environments are expected to have similar communities dominated by taxa that are well adapted to such conditions. Other ecological processes, such as dispersal, may increase similarity among communities through the flow of taxa or cells from sites where they are thriving (i.e. high rates of dispersal), which may overwhelm the effect of selection (i.e. mass effects). Given the unidirectional flow of water within draining basins, aquatic ecosystems are increasingly studied within the context of their aquatic networks rather than as a collection of isolated systems16,17. This perspective has expanded the recognition of bacterial metacommunities, which are a group of local communities connected by dispersal mediated by the directional flow of water within aquatic networks15,18,19. However, the observed decrease in bacterioplankton richness along the aquatic network and the presence of dominant soil taxa in the water have highlighted the importance of incorporating terrestrial ecosystems in understanding freshwater bacterioplankton dynamics14,20,21,22,23. This inclusion leads to a more complex conceptual framework, which includes aquatic and terrestrial factors that modulate dispersal and selection in the bacterioplankton metacommunity assembly (Fig. 1).

Fig. 1
figure 1

Conceptual diagram illustrating different perspectives on the relative importance of ecological processes in bacterioplankton metacommunity assembly along hydrological network. The colored circles represent local aquatic conditions. The arrows represent dispersal (width reflects relative importance and, color represents ecosystem connectivity: aquatic in white and soil-aquatic in black). The balance between ecological processes (selection and dispersal) reflects species sorting or mass effect change along the hydrological network. Font size corresponds to relative importance of each process along the network, and color is related to the potential source (water in blue and soil in red). The first perspective (A) predicts that (i) headwater communities are regulated by selection because they are isolated, thus community assembly is described by species sorting, whereas (ii) downstream communities are described by mass effects due to a surplus of dispersal. Therefore, β diversity is expected to decrease with Strahler order. The second (B) and third (C) perspectives add the terrestrial connections of aquatic network with the surrounding soil (dotted rectangle). However, B is a special case of C. B proposes that i) community assemblages in headwaters are predominantly shaped by mass effects, driven by the high dispersal of soil taxa into the water channel (this perspective overlooks terrestrial heterogeneity, assuming that dispersed soil bacteria are uniform across different headwaters), and ii) downstream communities are increasingly influenced by selection exerted by similar local aquatic conditions due to higher water retention time. Hence, β diversity is expected to be more similar in communities from low- and high-order Strahler sites. C proposes an increase in headwater selection due to local environmental changes resulting from landscape heterogeneity (colored patches in background). Therefore, β diversity patterns along the network are expected to be modulated by the degree of landscape heterogeneity (boxplot boxes size).

Full size image

The network position hypothesis (NPH), which is frequently used in freshwater ecology, highlights the importance of position within the network, considering headwaters as isolated sites and downstream reaches as sites more connected (Fig. 1A)17,24,25. From this perspective, the strength of dispersal increases progressively toward downstream sites; thus, selection is expected to dominate headwater ecosystems and decrease its influence downstream25. On the other hand, empirical evidence, particularly for bacterioplankton (Fig. 1B), recognizes the role of hydrology (i.e., water residence time) and soil-water connectivity as factors that modulate the relative importance of local selection and soil-derived dispersal along aquatic networks20,22. Unlike NPH, mass effects (from soil bacteria communities) are expected to shape community structure in upstream sites, whereas species sorting should increase downstream. Both perspectives recognize variations in aquatic conditions across the network, and the second identifies a contribution from terrestrial taxa, but neither explicitly states the role of the surrounding terrestrial conditions, which may affect the structure of aquatic bacterial communities (Fig. 1C). Although aquatic microbial ecologists currently recognize the landscape as the physical template occupied by the metacommunity, the influence of terrestrial conditions in shaping bacterioplankton metacommunity remains poorly understood26,27.

An approach that integrates terrestrial landscape elements into freshwater microbial community ecology requires the measurement of landscape structure. To evaluate the relative importance of local aquatic conditions, terrestrial landscape composition and connectivity in structuring the bacterioplankton metacommunity along the network, we sampled bacterial communities across a small aquatic network in the upper part of the Guaviare River watershed and immersed in the natural reserve “Serranía de la Lindosa”. This area, located within the Colombian Amazon region, is characterized by several rocky outcrops that generate high levels of geological and floristic heterogeneity within a small spatial extent28,29, whereas other regionally variable factors (i.e. climate) are similar. This unique configuration allowed us to examine how landscape, local aquatic conditions, and within-network-connectivity change along aquatic networks.

Materials and methods

Fig. 2
figure 2

Study site – Serranía de La Lindosa, each point corresponds to a sampling site (A) Hydrological network (intense blue lines), with main caños, as part of Guaviare River basin. The color of points is related to the four sub-basins studied (AB: Agua-Bonita, AY: Arenales-Yamús, CL: Conuco-Lagunas, DLN: Delicias-LagunaNegra). (B) Land cover associated with hydrological network (colors in background) and, type of systems included in the network (shape of points: circles - oxbow lakes and triangles - streams).

Full size image

Study sites and sampling

The study site was Serranía de La Lindosa in Guaviare, Colombia. This region is characterized by a few rocky outcrops of the Colombian Amazon covering an area of ~ 120 km2. Due to its ecological importance, this region is protected as part of Distrito de Manejo Integrado Ariari-Guayabero29. Its territorial extension houses some small streams that are part of the Guaviare river basin such as caño Negro, caño El Retiro, caño La María, caño Agua Bonita and caño Yamús and oxbow lakes (“madreviejas”) called Laguna Negra, Laguna Chica, and Laguna Grande29. According to stream drainage, we identified four sub-basins (based on an elevation map of the region – https://asf.alaska.edu/datasets/daac/alos-palsar/, resolution: 12.5 m): Delicias-Laguna Negra (DLN), Conuco-Lagunas (CL), Arenales-Yamús (AY), and Agua Bonita (AB). Sampling sites were chosen based on the observed differences in the surrounding land cover, composition and structure, as well as on its accessibility. In total, 40 sites were selected: 13 within DLN, 15 within the CL, and 6 within the AY and AB (Fig. 2).

The survey was conducted in June 2021. In order to describe the local aquatic conditions, we sampled 1 L of the surface water and stored it at ~ 4 °C for water chemistry analysis and measured water in situ parameters in each site. To analyze bacterial community composition, 1 L of the surface water was concentrated through a polycarbonate filter (0.22 μm pore size). In order to identify soil-derived bacterial taxa (i.e. taxa shared between soil and water bacterial communities), we collected 250 g of composite topsoil sample from the surrounding area at each sampling site. The polycarbonate filters and the soil samples were stored at -80 °C until DNA extraction. Finally, to characterize the landscape composition, we registered the geographic location of each sample site.

Landscape and local aquatic data

Landscape information was provided by the Instituto Amazónico de Investigaciones Científicas – SINCHI, which generated a land cover classification for the Amazon region (including Serranía de La Lindosa) in 2020, following the Corine Land Cover (CLC) classification (https://datos.siatac.co/pages/coberturas). For each site, we delimited circular buffers with a 30-m radius to calculate the percentage of each land cover class. This radius was chosen by considering the digital elevation model and selecting only the region draining into the stream of interest. Consequently, the samples closest to each other (~ 30 m) had the same buffer. Land cover proportion was calculated using QGIS 3.28 with the LecoS plugin (Landscape Ecology Statistics). We selected level 3 of the CLC classification for the analyses (categories included within the study region can be observed in Fig. 2 and Supplementary file 1-Table S1). Additionally, we measured dendritic distance (i.e. distance between sampling points considering the water flow) in QGIS.

The physical and chemical analysis of water samples included the quantification of total acidity (TAc), organic matter percentage (perOM), biochemical oxygen demand (BOD - mg/L), chemical oxygen demand (COD – mg/L), total suspended solids (TSS - mg/L), total solids (TS – mg/L), and concentrations of total organic carbon (TOC - mg/L), calcium (Ca – mg/L), chlorine (Cl – mg/L), total phosphorus (TP – mg/L), total iron (TFe – mg/L), magnesium (Mg – mg/L), nitrate (NO3 – mg/L), nitrite (NO2 – mg/L), sulfate (SO4 – mg/L), and sulfur (Sul – mg/L). In addition to these variables, environmental characterization incorporated the variables measured in situ: temperature (Temp - °C), pH, percentage of oxygen saturation (perDO), conductivity (Cond - µS/cm), total dissolved solids (TDS – mg/L), redox potential (roP - mV), and turbidity (Turb - NTU) using a multi-parameter sonde. Although we quantified 23 physicochemical variables, only 14 were used to describe the local aquatic environment. We excluded some variables from the study because their measurements were below the detection limit (NO3, NO2, and Sul) or because they were highly correlated with other variables. We considered this set of variables to refer to the local aquatic environment.

Aquatic bacterial community composition

The bacterial community structure was generated by sequencing of the 16 S rRNA gene of the aquatic and soil DNA samples. First, bacterial community DNA was extracted from the stored filter using the QIAGEN DNeasyPowerWater kit and from the stored soil samples using QIAGEN DNeasyPowerSoil kit, both according to the manufacturer’s instructions. The quality and quantity of extracted DNA were measured using a NanoDrop OneC spectrophotometer and a Qubit fluorometer. DNA integrity was evaluated via electrophoresis (data not shown). Then, the hypervariable region V3–V4 of the 16 S rRNA gene were amplified and sequenced in Macrogen, Korea. Amplification was achieved using the primers 341 F and 805R30 and, the sequencing was performed using an Illumina MiSeq platform with a paired-end protocol (300 × 300pb). Finally, the raw sequences were processed using the QIIME2 suite (version 2019.4)31. To obtain high-quality sequences, we used the DADA2 method32, and the deblur plugin was used for OTU aggregation with a 99% similarity threshold. Taxonomic assignment of each OTU was performed using the SILVA132 database33. We filtered and removed the sequences from the Archaea kingdom and we rarefied the aquatic bacterial community dataset to 10.032 reads per sample. We used the raw soil table in order to increase the probability of detecting any OTU shared between aquatic and soil samples.

Statistical analysis

Terrestrial landscape analysis was based on the land cover composition obtained from a 30-m circular buffer at each site, and heterogeneity was measured using the Shannon index. We also described the variation in local aquatic environmental conditions using principal component analysis (PCA) and used a redundancy analysis (RDA) to correlate the aquatic conditions with the land cover composition. As the number of response variables (i.e. local aquatic conditions)14 exceeded the number of explanatory variables (i.e. land cover proportions)8, we reduced the local aquatic conditions to 8 based on collinearity among the variables observed in PCA. We used non-metric multidimensional scaling analysis (NMDS) based on compositional differences among sites, using the Bray-Curtis index, to determine bacterial taxonomic variation across sites. For subsequent analysis, we standardized the datasets as follows: (i) landscape composition was transformed with the “Hellinger” option from the decostand function from the vegan v. 2.6-4 package, (ii) local aquatic conditions were standardized with the scale function, and (iii) OTU abundances were log – transformed.

To evaluate the relative importance of environmental factors (i.e. land cover composition and local aquatic conditions) and connectivity (i.e. dendritic distance) in explaining variations in bacterial community composition, we applied a Generalized Dissimilarity Model (GDM) using the gdm v. 1.5.0-9.1 package34. We used (i) dendritic distance and differences of NMDS scores between sites from (ii) land cover composition and (iii) local aquatic conditions ordinations, as predictors of bacterial community dissimilarity within the GDM model. The weight of each predictor was assessed using the gdm.partition.deviance() function. In addition, we quantified the number of shared OTUs between sites, also as proxy to connectivity between communities within network. Finally, we explored the influence of the terrestrial landscape heterogeneity on bacterial community composition via differential abundance analysis using the ANCOM-BC method35 - ancombc v.1.4.0 package. For this analysis, we considered landscape heterogeneity clustering to probe the differential abundance of bacterial classes and used a canonical analysis of principal coordinates (CAP) to demonstrate the relationship between the abundance of selected classes and the dissimilarity of bacterial communities. For all analyses, we used R software (version 4.0.2) and the statistical significance at a p-value < 0.05.

Results

Terrestrial landscape heterogeneity

The terrestrial landscape surrounding the aquatic sites was variable and heterogeneous although grassland and pastureland usually dominated the landscape (Fig. 3a). In addition, we identified three site clusters with different levels of landscape heterogeneity: High, Medium and Low (Supplementary file 1 – Figure S1).

Fig. 3
figure 3

Terrestrial landscape heterogeneity surrounding aquatic sites in the Serranía de La Lindosa. (a) Comparison of land cover composition across sites by sub-basin (AB, AY, CL and DLN); colors represent different types of land cover. (b) PCA bi-plot of sites according to land cover proportions. All variables (represented by arrows) are expressed as percentages (per).

Full size image

Variation in local aquatic conditions

Variation in local aquatic conditions was more related to the type of system and land cover composition than to sub-basin membership (Fig. 4). PC1 clearly separated oxbow lakes and streams, with the former having higher conductivity (17 µS/cm), temperatures (30.87 °C), and pHs (pH 6) and lower oxygen levels than streams. Meanwhile, PC2 separated some sites of the AB and YL sub-basins associated with high BOD and turbidity or sulfates and acidity (Fig. 4a). Other conditions, such as Fe, P, and Ca concentration showed low loading and explained little variation along PC1 and PC2 components (Supplementary file 1 - Table S2). Finally, we observed correlations between: (i) the proportion of grassland and secondary vegetation cover and the concentrations of SO4, Mg, and OD; (ii) the proportion of fragmented forest and TFe concentration; and (iii) the proportion of water surface area and conductivity (Fig. 4b).

Fig. 4
figure 4

Description of the local aquatic conditions in the hydrological network of Serranía de La Lindosa. (a) PCA bi-plot of sites colored by sub-basin (AB, AY, CL, DLN). (b) RDA model (p < 0.05) explaining changes in environmental variables (red arrows) according to land cover proportions (black arrows).

Full size image

Variation in the bacterial community composition (BCC) within hydrological networks

In total, we obtained 8688 OTUs with their taxonomic assignments after rarefaction. We observed differences in the taxonomic composition of the bacterial communities among the aquatic sites (Fig. 5). There was a clear segregation between oxbow lakes and streams (ANOSIMType: p < 0.01, R = 0.82) although, the bacterial community composition tended to be more similar between lakes than between streams. In contrast, we found no evidence of separation in BCC between sub-basins (ANOSIMsubBasin: p = 0.23, R = 0.36) (Fig. 5a). Additionally, we identified a declining pattern of bacterial community dissimilarity with increasing hydrological - Strahler - order (from upstream to downstream sites in the network) (Fig. 5b).

In general, Gammaproteobacteria was the most dominant class with > 25% relative abundance across sites (Supplementary file 1 – Figure S2). Taxonomic differences in bacterial communities between oxbow lakes and streams were related to an increase in the relative abundance of: Parcubacteria and Acidobacteriia toward lower river orders and Actinobacteria, Oxyphotobacteria, and Acidimicrobiia toward lakes.

Fig. 5
figure 5

Beta diversity of aquatic bacterial communities in the hydrological network of Serranía de La Lindosa. (a) NMDS of BCC using the Bray-Cutis index. (b) Bacterial community dissimilarity among sites with same hydrological Strahler order. OL denotes the oxbow lakes.

Full size image

Land cover as a driver of bacterioplankton community composition

Our approach allowed us to determine the relative importance of environmental factors (i.e. land cover and local aquatic conditions) and connectivity (i.e. dendritic distance) in the explanation of bacterial community dissimilarity. Considering that we identified a clear separation of BCC between lakes and streams (Fig. 5a), we present both GDM results for lakes and streams (i.e. all sites) and those that excluded lakes (i.e. only streams). The two GDM models were statistically significant and explained 49.02% (all sites – Fig. 6a) and 52.62% (only stream sites – Fig. 6b) of the deviance. In the model including all sites, the interaction between aquatic conditions and land cover composition explained the largest deviance (10.65%), followed by land cover composition alone (10.05%). In this case, the dendritic distance was the least important factor (2.64% of the deviance). In contrast, in the model that only included stream sites, land cover composition alone explained most of the deviance (14.34%), followed by the intersection between land cover composition and dendritic distance (12.96%), and the least important factor was aquatic conditions alone.

Fig. 6
figure 6

Deviance partition in GDM model for BCC from Serranía de La Lindosa. The values correspond to the percentage of deviance explained. (a) All sites (b) Only stream sites.

Full size image

Discussion

We analyzed the changes in bacterioplankton composition in relation to dendritic distance, local aquatic conditions, and terrestrial landscape across different sub-basins located on the upper part of the Guaviare River in the Colombian Amazon. Our results support the notion that environmental factors (i.e. land cover composition and local aquatic conditions) rather than connectivity-related factors (i.e. dendritic distance) shape the bacterioplankton community composition within this aquatic network. Although the relative influence of local aquatic conditions and land cover composition on bacterioplankton community structure changed along the network, the magnitude of their deviance was always higher than that of connectivity (Fig. 6). In the lotic portion of the aquatic network, the terrestrial landscape was the main driver of the bacterioplankton community structure, whereas in complete network, aquatic conditions and their interaction with the terrestrial landscape explained the differences in bacterioplankton communities. This work highlights the role of terrestrial landscape heterogeneity on bacterioplankton dissimilarity through the modulation of local aquatic conditions (Figs. 4b and 6)36. Also, it adds landscape heterogeneity (Fig. 1C) to previous evidence of the importance of hydrology (i.e. water residence time – WRT) and water chemistry (Scenario Fig. 1B) in the structuring of freshwater bacterioplankton communities14,22,37.

We show a decreasing pattern in community dissimilarities from upstream to downstream sites (Fig. 5b), a pattern that fits the expectation of scenarios A and C of our perspectives in Fig. 1. However, GDM revealed that the beta diversity pattern was better predicted by the landscape and local aquatic environment than by dendritic distance (Fig. 6). The last implies that the lower compositional dissimilarities observed downstream cannot be explained only by stronger dispersal, as suggested by NPH (Fig. 1A); in fact, we found that only between 46 and 257 OTUs persisted along each sub-basin (Supplementary file 1 - Table S4 and Figures S3 and S4). Instead of that, this decreasing pattern may be explained by the combination of longer WRTs with more similar local environments at downstream sites14 and larger differences in the aquatic environmental conditions toward the upper parts of the network16, both of which favor selection processes. The analysis of aquatic environmental conditions also indicated that these conditions are related to changes in land cover composition (Fig. 4b). Therefore, our analysis of the GDM model revealed that land cover composition seems to capture environmental or spatial changes in unmeasured environmental gradients associated with differences in bacterial community composition between upstream and downstream sites. This finding highlights the importance of including landscape variables and opens a new avenue for exploring the hierarchical relationships between regional (e.g. climate and land cover) and local environmental variables to better understand bacterioplankton dynamics in the aquatic networks38.

Regarding the role of hydrology and soil-derived bacteria in structuring aquatic metacommunities (Fig. 1B), previous studies have indicated that water residence time interacts with local environmental conditions to drive bacterial community composition14,37. Although we did not explicitly measure water residence time, we observed the largest divergence in bacterial community composition between streams and oxbow lakes (Fig. 5a; Supplementary file 1- Figure S2). This pattern was related to a lower prediction power of local aquatic conditions on bacterial community composition in the GDM model that excludes oxbow lakes (Fig. 6b). We argue that this observation is related to an increasing relative importance of environmental selection toward downstream sites10,22, associated with the largest differences in local aquatic conditions between stream and lakes (Fig. 4a). These differences may impose spatially variable selection pressures resulting in distinct bacterial communities toward downstream sites21,37. On the other hand, we expected an increasing relative importance of dispersal of soil-derived taxa (i.e. mass effects) toward upstream sites (Fig. 1B) but measuring this process may be a challenging task. However, we found a low number of shared OTUs between aquatic and surrounding terrestrial samples along the whole network (329 OTUs; ~5% of total soil community – Supplementary file 1-Figure S5). The last indicates that dispersal of soil-derived bacteria into aquatic environments is occurring, although our data seem not enough to match the expected numbers to support the mass effects previously proposed10,20,22,23. We acknowledge that our sampling design and sequencing depth (about 10.000/sample) may not be optimal for capturing the rarest portion of the bacterial communities. Nonetheless, soil-water interface is a strong environmental filter that prevents most taxa from reaching the aquatic environment22,23, which may reconcile the expected large flow of bacteria from soil to water with the small proportion of shared OTUs shown here. Regardless of these limitations, the larger deviance explained by dendritic distance in the only-streams GDM model (Fig. 6) suggests that dispersal with the network may be relatively more important upstream than in downstream sites. This is partially consistent with the expectations of our second scenario (Fig. 1B), which proposes a gradual change from upstream sites dominated by mass effects10,22 to downstream sites dominated by species sorting.

The influence of terrestrial landscape on bacterioplankton community composition has been largely neglected39 but recently started to be examined by a few authors. The conclusions of these studies have not always been consistent40,41,42, because some of them still emphasize the dominance of local aquatic conditions whereas others highlight the role of the terrestrial landscape. We attribute these contrasting results to differences in the spatial extent (i.e. draining areas) and to the position of the sampling sites within the networks (e.g. upstream or downstream), which represent partial views of the processes driving bacterial communities in aquatic networks and therefore may influence the relative role of landscape. In this regard, we decided to study a network that includes aquatic systems with low and high WRT (i.e. streams and lakes), with a known variable geological and floristic heterogeneity29, which is unusual for such a small area (120 km2). The increase in the deviance explained by land cover composition in the only-stream sites compared to the all-sites GDM (Fig. 6) suggests that the relative role of the landscape decreases toward downstream sites, regardless of the heterogeneity of the landscape. The last, probably due to changes in the water: soil ratio along the network, which is related to the interchange and the dilution of whatever is moving from soil to water. However, we acknowledge that how the bacterial community responds to the influence of landscape along the network may be modulated by how heterogeneous the landscape is. In fact, the landscape heterogeneity did not increase or decrease with the flow of water in our dataset. Instead, we found that a small portion of bacterial taxa (22 out 87 Classes) were actually changing in abundance, as a response to our groups of landscape heterogeneity (High, Medium and Low in Fig. 3b) (Supplementary file 1 – Figure S6 and Table S3), which contained sites from different sub-basins (Supplementary file 1- Figure S1). Thus, although we added evidence of how hydrological and environmental gradients influence bacterioplankton community changes along the aquatic networks, we also showed that they seem to be responding to landscape heterogeneity and, in general, to how the landscape is structured across space.

Conclusion

In this study, we highlight the role of landscape in the structuring of aquatic bacterial communities, and how this role may change along aquatic networks. We propose that landscape composition captures unmeasured aquatic environmental gradients that drive changes in bacterial community composition in aquatic networks. These aquatic environmental gradients may result from the flow of water across the landscape, which gather and convey several soil properties that promote selection of bacterial communities within aquatic ecosystems. The relative importance of these landscape-mediated properties vary depending on the position within the network, in a close relationship with changes in hydrology. This landscape perspective may improve our understanding of bacterial community patterns and microbes-mediated biogeochemical processes in aquatic ecosystems. This approach may also promote the use of this biological component in monitoring processes that may extend beyond the boundaries of aquatic ecosystems.