Introduction

Globally, inland waters are responsible for processing, transporting, and sequestering ~ 2.9 Pg of terrestrial carbon (C) annually (Tranvik et al. 2009). At landscape and global scales, inland waters (including rivers and streams) constitute a small portion of the land surface but are biogeochemically active and process more C than is expected based solely on area (Cole et al. 2007; Tranvik et al. 2009; Raymond et al. 2013). River ecosystems also have a large role in C export to oceans, contributing an estimated 1.06 Pg/year as particulate and dissolved forms (Battin et al. 2009; Aufdenkampe et al. 2011; Li et al. 2017). In lotic ecosystems, C cycling involves interplay of numerous factors, including land use, geomorphology, origin of C sources, C quality and quantity, microbial activity, and microbial community composition and diversity (Cole et al. 2007; Williams et al. 2010). Given the diverse factors and interactions between them, it is often difficult to determine how, and at what scale, these factors affect C cycling in fluvial networks (Amado et al. 2013).

Carbon metabolism by aquatic heterotrophic microbes is a critical process, contributing to overall ecosystem production, and representing a primary route of dissolved organic carbon (DOC) transformation (Cole et al. 2007). Multiple factors affect DOC processing rates, and the relationship between bacterial C metabolism and the concentration, sources, and composition of DOC is complex (Hitchcock and Mitrovic 2015; Figueroa et al. 2016). Bacterial production (BP), the rate which C is incorporated into biomass, and bacterial respiration (BR), the rate at which bacteria use C for cell maintenance, can exhibit a positive relationship with DOC concentration (Lennon and Pfaff 2005). However, BP and BR may not be strongly related to the size of the DOC pool, particularly if the DOC is refractory (Findlay 2003; del Giorgio and Davis 2003). Thus, increasing the proportion of refractory C within DOC affects bacterial metabolic function and leads to greater BR per unit BP, indicative of decreased bacterial growth efficiency (BGE; del Giorgio and Cole 1998). A critical question in aquatic ecosystem studies is the relative importance of allochthonous versus autochthonous C sources in aquatic food webs because autochthonous DOC is generally considered more labile than allochthonous (terrestrial) DOC and particulate organic matter (Sondergaard and Middelboe 1995; del Giorgio and Davis 2003; Raymond and Spencer 2015). Evidence indicates that autochthonous DOC sources are important in determining bacterial C metabolism because most labile C is generated by in situ primary producers (Wetzel 2001; Sondergaard and Middelboe 1995). However, DOC lability is also influenced by other factors such as DOC age (e.g., younger pools are more reactive), molecular structure (e.g., lignin content), photo-degradation, and anthropogenic inputs (e.g., wastewater) (Raymond and Spencer 2015).

Although DOC concentration, sources, and lability are important determinants of bacterial C metabolism, physicochemical processes also act upon DOC (e.g., ultraviolet radiation exposure, flocculation), but the relative importance of these processes can be difficult to distinguish from the effects of bacterial processing (del Giorgio and Pace 2008). Increased availability of inorganic nutrients can catalyze bacterial C utilization and stimulate greater bacterial production (Granéli et al. 2004; Ochs et al. 2010). In freshwater systems, phosphorus (P) can limit bacteria (as opposed to inorganic nitrogen or DOC), influencing bacterial growth rates, metabolism, and community composition (Granéli et al. 2004; Nelson and Carlson 2011; Peura et al. 2012). To understand how microbial communities process C in riverine ecosystems, it is necessary to examine how the combination and co-occurrence of physicochemical characteristics, dissolved inorganic nutrients, and C quality and quantity influence bacterial C metabolism.

Most research on landscape-level variation in river bacterioplankton C processing has been in temperate systems (e.g., Maranger et al. 2005; del Giorgio and Pace 2008; Guillemette and del Giorgio 2011), but studies have been conducted in boreal (Holmes et al. 2008; Comte and del Giorgio 2009), subtropical (Hadwen et al. 2010), and tropical (Amon and Benner 1996; Farjella et al. 2009) rivers. Arid and semiarid ecosystems comprise ~ 40% of the earth’s land surface (Middleton and Thomas 1997), but few studies address bacterial C metabolism to arid rivers. In situ gross primary production in arid rivers (via algae) is thought to be greater than that in temperate rivers due to higher light availability and water temperatures, lower flows, and more efficient nutrient cycling (Lamberti and Steinman 1997; Velasco et al. 2003). In arid rivers, adjacent terrestrial productivity is also relatively low; thus, in situ (autochthonous) algal primary production is expected to be the main C source for organisms, including bacteria (Finlay 2001). Consequently, it can be predicted that a greater relative amount of autochthonous C in arid river systems leads to greater DOC lability, more rapid bacterial C processing, and greater BGE when compared to more mesic rivers. Understanding the role of microbial C processing and the factors that influence rates of processing in arid river systems is relevant as rivers in arid landscapes are under increasing anthropogenic pressure (Kingsford 2000).

We assessed spatiotemporal variation in heterotrophic bacterioplankton C metabolism in a large, complex arid and semiarid riverine network in the southwestern USA: the Rio Grande/Río Bravo del Norte (hereafter the Rio Grande drainage). Specifically, we examined how bacterial C metabolism was affected by spatiotemporal variation in physicochemical conditions, DOC concentration, and DOC lability. We investigated whether bacterial abundance and C metabolism in the basin were more influenced by large-scale physicochemical gradients, or whether they were more influenced by the quantity and quality of C sources and bacterial stoichiometry. We hypothesized that spatiotemporal variation in physicochemical conditions across the drainage would affect rates of bacterial C metabolism, but that variation in the quantity and quality of DOC would be the primary driver of the bacterial C processing. We also hypothesized that bacterial DOC processing in more arid portions of the drainage would have greater reliance on autochthonous and more labile C sources.

Methods

Study area and sampling design

Headwaters of the US portion of Rio Grande drainage originate in Colorado and New Mexico before entering Texas and forming the border between the USA and Mexico (Fig. 1). This study focused on major river catchments within the Texas portion of the drainage. The Rio Grande subcatchment originates in Colorado, with ~ 2000 river km (rkm) of the river running along the border. The Pecos and Devils Rivers are on the US side of the drainage and the Río Conchos is the major Mexican tributary in the study area. The Pecos River flows approximately 560 rkm from New Mexico and through west Texas before joining the Rio Grande at Amistad International Reservoir. The Devils River originates within Texas and flows ~ 160 rkm before joining the Rio Grande at Amistad International Reservoir. Along the upstream–downstream gradient of study area, the surrounding landscape represents several ecoregions, transitioning from the arid Chihuahuan Desert and Trans-Pecos of the upper Rio Grande and Pecos Rivers (mean annual rainfall < 400 mm year−1), to the semiarid Edwards Plateau in the region of the Devils River, to semiarid subtropical South Texas brush country in the lower portion of the drainage (mean annual rainfall ~ 700 mm year−1).

Fig. 1
figure 1

Map of the Rio Grande, Pecos River, and Devils River catchments in Texas, USA. Points indicate study site locations, and the designations RG, PR, and DR for sites indicate sites in the Rio Grande, Pecos River, and Devils River watersheds, respectively. Site designations and numbers correspond to sites in Tables 1 and 2. Locations of Amistad International Reservoir and Falcón Reservoir are also indicated

Full size image

Data were collected at 14 sites (Fig. 1) to assess differences in physicochemical and bacterial C dynamics along the drainage. All sites were in the mainstem of rivers; sampling sites were free flowing (i.e., not backwater areas or behind impoundments), not small order headwaters, and were wadeable. Nine sampling sites were on the Rio Grande mainstem: three sites were located above Amistad International Reservoir (designated RG1–3), four sites were distributed between Amistad International and Falcón International reservoirs (RG4–7), and two sites were located below Falcón International Reservoir (RG8 and 9). Four sites were located along the Pecos River (PR1–PR4), and one site was located on the Devils River (DR1). Sites RG1–RG3 are in the Chihuahuan Desert and Trans-Pecos ecoregions, with RG2 located immediately below the confluence of the Rio Grande and Rio Conchos. Sites RG4–9 are distributed through semiarid subtropical South Texas Brush ecoregion in the Lower Rio Grande River Valley. Flows in this portion of the Rio Grande tend to be of higher magnitude and more consistent due to releases from Amistad and Falcón International Reservoirs for agricultural needs and to fulfill international treaty requirements. Pecos River (PR) sites are found in the Trans-Pecos and along the edge of the Edwards Plateau; PR1 and PR2 exhibit relatively low flows and higher salinities (Trans-Pecos sites), but PR3 and PR4 (edge of Edwards Plateau) are strongly influenced by groundwater contribution (Edwards–Trinity aquifer) (Gregory and Hatler 2008). The landscape surrounding DR1 is characterized by a convergence of the three main ecoregions. The Devils River is remote and largely pristine (little anthropogenic impact), dominated by groundwater from the Edwards–Trinity aquifer.

Field sampling

Data were collected on a seasonal basis (Spring, Summer, and Fall) from March to December 2010. We define “Spring” as March to May, “Summer” as July to August, and “Fall” as October to December. Seasons were defined by the agricultural calendar because flows are managed for irrigation. Data collected included physicochemical parameters, total, dissolved and sestonic nutrients, and phytoplankton biomass. We additionally assessed bacterial abundance, bacterial C metabolic rates (i.e., BR and BP), and DOC quality at a subset of nine sites consisting of RG1, RG2, RG4, RG7, RG8, RG9, PR1, PR4, and DR1. This subset was selected to maximize variation and to serve as proxies for conditions across the drainage.

At each site, physicochemical data and water samples were collected from the thalweg at the water column mid-point. Dissolved oxygen (DO; mg L−1), specific conductivity (SpC; µS cm−1), pH, salinity (ppt) and water temperature (°C) were collected with a YSI 556 multiparameter instrument (YSI Inc., Yellow Springs, Ohio, USA). Water was collected as duplicate grab samples and stored in pre-cleaned 2-L opaque or brown Nalgene™ high-density polyethylene (HDPE) bottles that had been rinsed with site water prior to collection (Becker et al. 2017). At the subset of sites selected for bacterial C metabolism estimates, an additional 7.5 L of water was collected in pre-cleaned opaque HDPE Nalgene™ carboys that were field rinsed. Water samples were stored on ice in coolers until the return to Texas State University, where they were held at ambient river temperature. Samples were processed within 48 h of collection. We estimated hydrological conditions at each site prior to and at the time of sampling by calculating the mean daily discharge (Q9) for the 9-day period prior to the sampling date using data from United States Geological Survey (USGS) or International Boundary Waters Commission (IBWC) gauging stations located nearest each sampling site (Table 1).

Table 1 Study site catchment, location, and nearest USGS or IBWC gauging station site for 14 sites distributed across the Rio Grande drainage in Texas, USA
Full size table
Table 2 Biological, bacterial metabolism, and OC variables for the subset of 9 sites across the Rio Grande drainage in Texas, USA
Full size table

Laboratory analyses

Total P (TP), nitrate (NO3−), ammonium (NH4+), soluble reactive P (SRP), non-volatile suspended solids (NVSS), dissolved organic carbon (DOC), and suspended particulate nutrients (C, N and P) in the < 1 µm size (bacteria-sized) fraction, and suspended phytoplankton biomass were estimated in duplicate for each site on each sampling date. The amount of variation among duplicate samples for a site for each parameter was very small. TP was measured as PO43− following digestion with potassium persulfate and analysis with the molybdenum blue method (Wetzel and Likens 2000). Nitrate, NH4+, and SRP concentrations were determined in water samples filtered through pre-ashed Pall A/E filters (nominal pore size = 1 µm). Nitrate concentrations were determined with second-derivative UV spectroscopy (Crumpton et al. 1992) and SRP concentrations were measured as PO43− with the molybdenum blue method (Wetzel and Likens 2000). Ammonium concentrations were determined by the phenate method (Wetzel and Likens 2000). DOC samples were filtered through a pre-ashed Whatman GF/F filter and analyzed on a Shimadzu TOC-VCSH total organic carbon analyzer (Shimadzu Scientific Instruments, Columbia, Maryland, USA). DOC color on the same samples was assessed as absorbance at 440 nm and calculated and reported as absorbance coefficients (g; m−1) from Cuthbert and del Giorgio (1992). Spectrophotometry was performed on a Varian Cary 50 UV–Vis spectrophotometer (Agilent Technologies, Santa Clara, California, USA).

NVSS was determined by filtering water onto pre-combusted and pre-weighed Pall A/E filters. Filters were dried at 50 °C for 48 h and were re-weighed to determine total suspended solids (TSS) and subsequently combusted at 500 °C for 4 h and re-weighed to determine NVSS. C, N, and P content of the bacterial size fraction (< 1 µm) of the seston was determined by filtering samples through pre-ashed Pall A/E filters, and filtrate was subsequently filtered onto pre-ashed Whatman GF/F filters and dried at 50 °C for 48 h. The < 1 μm C and N concentrations were measured on a CE Elantech Flash EA 1112 CN analyzer (CE Elantech Inc., Lakewood, New Jersey, USA). For < 1 μm particulate P samples, filtrate was filtered onto Whatman GF/F filters, digested with HCl at 100 °C and measured as PO43− with the molybdenum blue method (Wetzel and Likens 2000). Suspended phytoplankton biomass was estimated using chlorophyll a (Chla) concentration. Water was filtered onto Pall A/E filters and stored at − 20 °C until extraction with acetone for 8 h in the dark and measurement on a Turner Designs Trilogy fluorometer (Turner Designs, Inc., Sunnyvale, California, USA).

Bacterial abundance (BA) from the subset of sites was determined by preserving 60 mL of whole water clean glass bottles with 4% filtered (< 0.2 μm) formalin. Duplicate samples for each site on each sampling date were kept in the dark at 4 °C until analysis. Samples (1–6 mL) were filtered onto black 0.2-μm Nucleopore membrane filters, stained with 4′,6-diamidino-2-phenylindole. Immediately after staining, slides were mounted with Citifluor™ AF1 solution and examined with a Nikon Eclipse 80i microscope, fitted for epifluorescence microscopy with a mercury lamp (Nikon; X-Cite™ 120) and UV-2E/C filter (Nikon Instruments, Inc., Melville, New York, USA). Cells were counted at 1000 × magnification, and 20 fields of view (grid area = 1.0 × 10−4 mm2) were counted for each duplicate sample.

At the subset of sites, BP was measured using the microcentrifuge 3H-leucine method (Pace et al. 2004). On each sampling date, whole (unfiltered) water from the well-mixed 7.5-L carboy was allocated into four “live” and two “killed” 2-mL microcentrifuge tubes. 3H-leucine (~ 60 Ci/mmol leucine) was added to all tubes, but the “killed” tubes received cold 50% trichloroacetic acid (TCA) prior to 3H-leucine addition; each tube received 1.4 μCi of 3H-leucine (Smith and Azam 1992; Kirchman 1993). Tubes were incubated for 45–60 min in the dark at site-specific water temperature, after which activity in “live” tubes was stopped with the addition of cold 50% TCA and all tubes were centrifuged for 10 min at 14,000 rpm. Supernatant was aspirated, cold 5% TCA was added, and tubes were re-centrifuged. Supernatant was removed, scintillation fluor was added, and 3H activity was measured on a Beckman LS 60001C scintillation counter and bacterial production was expressed as µC L−1 h−1. The same microcentrifuge tube type and brand were used throughout (Pace et al. 2004).

BR at the subset of sites was estimated via biological oxygen demand (BOD) (Roland et al. 1999; Williams and del Giorgio 2005). To estimate BR, we conducted 48-h BOD experiments; the 48-h incubation time was utilized because initial BOD incubations found that this incubation length was adequate to measure changes in DO in systems. Incubations were performed in clean and acid-washed 60-mL Whatman BOD bottles with stoppers. For each site on each sampling dates, five replicate bottles were filled with whole (unfiltered) water. Initial DO concentrations were measured on Day 0 on duplicate bottles using a spectrophotometric Winkler method (Roland et al. 1999). Remaining replicate bottles were incubated in the dark at in situ site temperatures. After 48 h, bottles were removed and processed as above. Hourly O2 consumption rate (mg O2 L−1 h−1) was calculated as the difference between initial and final DO, and O2 consumption values were converted to C respired (µg C L−1 h−1) using a respiratory quotient of 1 (del Giorgio et al. 2006). To be consistent with BP methods, we used whole water for BR estimates; use of whole water includes respiration by free-floating and particle attached bacteria, as well as larger planktonic organisms (del Giorgio et al. 2006; Vidal et al. 2011). Abundance of larger planktonic organisms (e.g., zooplankton) is low in flowing river systems such as the Rio Grande (Basu and Pick 1996); thus, these organisms likely did not have substantial influence on respiration rates. Use of filtered (e.g., < 1 μm) water can lead to underestimates of BR (del Giorgio et al. 2006) because a substantial portion of the bacterial community is attached to particles in river systems (Ochs et al. 2010), and filtration removes particle-associated and some free-floating bacteria (del Giorgio and Pace 2008). Thus, BR measurements in this study are likely representative of the maximum potential BR rate of the microbial community (Becker et al. 2017). In addition, use of whole water BR estimates could lead to lower estimates of BGE than if we used filtered water for incubations. Measurements of short-term BR were coupled with the concurrently measured BP estimates in order to calculate BGE from del Giorgio and Cole (1998) as

$${\text{BGE}} = {\text{BP/}}\left( {{\text{BP}} + {\text{BR}}} \right)$$

DOC lability at the subset of sites was evaluated through use of long-term BOD incubations on water that was first filtered through Pall A/E filters (< 1 μm); water was filtered in order to remove the effect of decomposition of particulate matter and to target free-floating bacteria (Ostapenia et al. 2009; Sullivan et al. 2010). Concentration of labile DOC (OCL) was estimated by interpreting plateaus in DO consumption over long-term BOD incubation time interval (Coffin et al. 1993). We used a first-order decomposition kinetics model:

$${\text{BOD}}_{t} = {\text{BOD}}_{\text{ult}} \left( {1 - {\text{e}}^{ - kt} } \right)$$

where BODt is the biological oxygen demand at time t, BODult is the total BOD possible for the bacterioplankton community (asymptotic at infinity, ∞), and k is the reaction constant of aerobic decomposition of OC (Ostapenia et al. 2009). The breakdown rate of OCL (k) can be estimated, and using this estimated k, BODult can be inferred, providing the concentration of labile DOC (Ostapenia et al. 2009; Sullivan et al. 2010). We conducted 20-day incubations that were held in the dark in pre-cleaned and acid-washed 120-mL Whatman glass BOD bottles with glass toppers at in situ site temperatures. DO concentrations were measured on days 0, 2, 4, 8, 16, and 20 from triplicate bottles with the modified Winkler method. Using Eq. 1 from Ostapenia et al. (2009), we calculated the breakdown rate of labile C (k), and using estimated k, BODult is inferred, providing an indication of the size of the labile DOC pool. BOD kinetics and OCL concentration were calculated with Microsoft Excel Solver Tool, which employs generalized reduced gradient nonlinear optimization code and assumes first-order kinetics. Use of first-order kinetics calculates a decay rate based on the concentration of a single reactant (in this case, OCL) (Ostapenia et al. 2009; Sullivan et al. 2010). Estimates for BODult and k were calculated by using O2 consumption curves and solving for the minimized residual sum of squares using Eq. 1 (Coffin et al. 1993; Ostapenia et al. 2009). Estimates of OCL were obtained by multiplying the 20-day O2 consumption by 0.30 (Ostapenia et al. 2009).

Data analysis

To assess spatiotemporal patterns in physicochemical characteristics across all sites in the drainage, we used principal component analysis (PCA). To avoid inclusion of too many variables in a single PCA (McCune et al. 2002), we performed data reduction procedures. Specific conductivity and salinity were highly correlated (Pearson r = 0.992, p < 0.001); thus, we only used salinity. We also excluded pH because it exhibited little variation across sites. Remaining physiochemical and nutrient variables (i.e., Q9, water temperature, DO, NO3−, NH4+, SRP, TP, DOC, NVSS, salinity, Abs440, and season; Table 1) were log10 or log10(x + 1) to meet assumptions of normality and z-score transformed prior to analysis. It was not possible to determine Q9 for two sites on one sampling date (i.e., PR3 and PR4 sites during Summer sampling) because gauging stations were inoperable; thus, mean Q9 across all sites and seasons was substituted (McCune et al. 2002).

Using PCA, we identified groups of sites that exhibited similar physicochemical and nutrient conditions. We examined how bacterial C metabolism, bacterial nutrient stoichiometry, BA, and Chla varied with among these site PCA-defined site groups with multivariate analysis of variance (MANOVA) on the subset of nine sites. In the MANOVA, site group and sampling season (Spring, Summer, and Fall) were used as fixed independent factors and BA, Chla, BR, BP, BGE, k, OCL, Abs440, < 1 μm C:N, < 1 μm C:P and < 1 μm N:P were used as dependent variables in the analysis. If a significant overall effect of a dependent factor was detected, we then performed individual ANOVAs with post hoc Tukey HSD tests. Data were log10-transformed before analysis to meet assumptions of normality and homoskedasticity as needed, and we utilized Type III sum of squares because of unequal n among site groups. Significance was inferred at α ≤ 0.05, and analyses were performed in SPSS version 24.

To test the hypothesis that C quality–quantity would exert a greater influence than physicochemical conditions across the study drainage, we utilized redundancy analysis (RDA). RDA allows for examination of biological responses to environmental gradients and can be used to assess the relative importance of groups of predictor variables on response variables (e.g., proportion of variance explained; Legendre and Legendre 2012). RDAs were performed using the subset of nine sites where biological and bacterial C metabolic responses were measured. Specifically, we were interested in determining whether landscape-level variation in (1) physicochemical and nutrient conditions or (2) C quality–quantity and bacterial nutrient stoichiometry was more important in determining patterns of bacterial C metabolism across the drainage. Thus, we examined the relationship between biological and metabolic responses (BR, BP, BGE, Chla, BCD, and k) and variables we classified as either physicochemical (i.e., Q9, water temperature, DO, NO3−, NH4+, SRP, TP, NVSS, and salinity) or (2) C quality–quantity and bacterial stoichiometry variables (i.e., DOC, OCL, Abs440, and < 1 μm C:N, C:P and N:P). Season (Spring, Summer, and Fall) was initially included as a predictor in RDAs but was not significant and was therefore excluded. Initially, there were a large number of predictor variables relative to the number of samples, so we performed data reduction on predictor variables to avoid excessive variable-to-sample ratios (McCune et al. 2002). We used a Pearson correlation matrix among predictor variables in each group and excluded variables that were highly correlated (|r| = 0.70). After variable reduction, the physicochemical predictor data set contained NO3−, NH4+, TP, NVSS, and Q9, and the C quality–quantity and bacterial stoichiometry data set contained DOC, OCL, Abs440, < 1 μm C:N, and < 1 μm N:P.

Using remaining predictors, we conducted partial RDAs, where the analysis is performed on one set of predictor variables while controlling for the effect of the other. We then performed a global analysis (both predictor sets combined) and used variance partitioning to evaluate the combined and pure effects of the two predictor sets (Peres-Neto et al. 2006; Legendre and Legendre 2012). Variance partitioning provides the amount of variation explained by each set of predictors (physiochemical/nutrient vs. C quality–quantity/bacterial stoichiometry), as well as the amount of variation shared between the two groups of predictors. Permutation tests (n = 999, α = 0.05) were used to assess significance of partial and global effects models (Legendre and Legendre 2012). For all RDA models, we present the first two axes corrected by the \(R_{\text{adj}}^{2}\) (Peres-Neto et al. 2006; Legendre and Legendre 2012). All multivariate analyses were performed using R 3.3.2 (R Development Core Team 2013) using the R statistical packages “vegan,” “car,” and “psych” (Oksanen et al. 2012; Revelle 2019; Fox et al. 2016).

Results

Spatial variation in physicochemical gradients

Physicochemical conditions across the drainage exhibited substantial spatial variation (Table 1; Fig. 2a). The first two principle components (PC 1 and PC 2) explained 54.4% of the total variation. PC 1 (29.3% of variation explained) generally described an upstream–downstream gradient in which the more upstream sites in the drainage (sites RG1 through RG3, PR1, and PR2; Fig. 2b) exhibited higher salinities, DO and NH4+ concentrations, while the more downstream sites (RG4 though RG9) were characterized by higher mean discharge and warmer water temperatures. PC 2 (25.1% of variation explained) generally described a gradient of sites dependent on surface flows (all RG sites and PR1, PR2) to sites dominated by groundwater contributions (PR3, PR4, and DR). Sites influenced by surface water were characterized by higher NVSS, TP, DOC, and SRP concentrations and greater Abs440 values, whereas the groundwater-dominated sites were characterized by lower P and NVSS and higher water clarity and NO3− concentrations. Groundwater-dominated sites also exhibited limited seasonal variation in conditions as indicated by relatively tight clustering of measurements from a site in the ordination. In contrast, the furthest upstream site in the Rio Grande (RG1) showed the greatest seasonal variability, exhibiting extremely high NVSS, TP, and DOC concentrations during Summer (Table 1; Fig. 2b). However, season had minimal overall influence on the data set (Fig. 2a). This site and this sampling date strongly influence PC 2; however, when this point was removed prior to analysis, the influence of the physicochemical variables along PC 1 and 2 did not change and the spatial distribution of sites in the ordination does not change substantially.

Fig. 2
figure 2

Principle component analysis (PCA) of physicochemical and nutrient variables across n = 14 study sites over three sampling seasons in the Rio Grande, Pecos River and Devils River drainages. a Multivariate relationships among physiographic variables across all sites. b Ordination of sampling sites. Drainage designations (RG, PR, and DR) and site numbers correspond to those in Fig. 1, Tables 1 and 2

Full size image

Overall, ordination of sites revealed a segregation of sites into three general groups based upon physicochemical and nutrient conditions. The first group was composed of sites in the upper portion of the drainage (hereafter called the UD group), which included the upper Rio Grande sites (RG1 through 3) as well as the two upper Pecos sites (PR1 and 2). A second group was made up of sites from the mainstem of the Rio Grande in the lower portion of the drainage (hereafter, LD group) below and between Amistad International Reservoir and Falcón International Reservoir (RG4 and RG9). The last group was composed of groundwater-influenced sites (hereafter, GR group) and included the two lower Pecos sites (PR3 and 4) and the Devils River (DR1). These physicochemically defined site groups were used to assess spatial patterns in biological and metabolic responses across the drainage.

Spatiotemporal variation in biological and bacterial C metabolism responses

BA, Chla, bacterial C metabolism and C processing responses varied among site groups (UD, LD, and GW; Pillai’s Trace: F18, 22 = 2.18; p = 0.049) and with sampling season (Pillai’s Trace: F18, 22 = 4.37; p = 0.001), and the interaction effect between these factors was significant (Pillai’s Trace: F44, 44 = 1.74; p = 0.035). When the effect of site groups was assessed on individual response variables, BA, Chla, DOC concentration, BP, and OCL concentration differed among site groups (Table 3), with GW sites exhibiting lower BA (Fig. 3a), Chla (Fig. 3b), and BP rates (Fig. 3c) than the UD and LD site groups. DOC concentration differed among the site groups (Fig. 3f), with URG sites having the highest and GW sites having the lowest concentrations. OCL concentration was also lower in the GW group than the UD group (Fig. 3g). In contrast, seasonal effects on BA, Chla, and bacterial C metabolism were substantially more limited. Bacterial C:N varied among seasons, with the C:N in the Spring (mean across sites = 5:1) being greater than bacterial C:N in the Fall (mean across sites = 4:1) (Table 3). In addition, OCL concentration was the only variable to exhibit a site group × season interaction (Table 3); OCL concentrations in the URG and LRG groups were substantially higher than those in the GW group in the Spring, but OCL similar among groups by Summer. Note that these season and interaction effect results are not presented in a figure due to the limited nature of these effects.

Table 3 Analysis of variance (ANOVA) results examining the effect of site group, season, and the site × season interaction on biological and C quantity–quality response variables at the subset of nine sites across the Rio Grande drainage in Texas, USA
Full size table
Fig. 3
figure 3

Mean a bacterial abundance, b phytoplankton biomass (measured as Chla), c rate of bacterial production, d rate of bacterial respiration, e BGE, f DOC concentration, g OCL concentration, h breakdown rate of OCL (k), i Abs440, (j) < 1 μm C:N (molar), k < 1 μm C:P, l < 1 μm N:P at the three site groups as determined from principle component analysis. Site groups are composed of the subset of sites (n = 9) at which bacterial metabolism estimates were obtained and the sites groupings were the upper drainage sites, lower drainage sites, and groundwater-dominated sites (indicated by UD, LD, and GW, respectively). Bars represent the mean of each variable for all sites in each site group across sampling dates, and errors represent ± 1 SE. Capital letter designations above site group bars indicate that there is a significant difference among site groups for a variable and capital letters indicate homogenous groups as indicated by post hoc Tukey tests

Full size image

Drivers of biological and bacterial C metabolism responses

The RDA model investigating solely the effect of physicochemical and inorganic nutrient gradients on biological and bacterial metabolic responses was significant (p = 0.006) and explained 22.3% of the variation in the data (19.3% explained by the first two axes; Fig. 4a). NH4+ (p = 0.016), TP (p = 0.024), and Q9 (p = 0.038) were significant individual predictors in the model. RDA 1 described a gradient of sites with higher NH4+ and Q9 to sites with lower discharge and higher NVSS and TP. RDA 2 largely described a gradient of groundwater-dominated sites (higher NO3−, but lower TP and NVSS) to more turbid, nutrient-rich sites. BP rates, BGE, BA, and Chla were positively correlated with higher Q9 and NH4+ concentrations, while BR rates increased with NVSS. In addition, BR and BGE were negatively correlated with each other indicating that variation in BR was largely driving variation in BGE (Pearson r between log10BR and log10BGE = − 0.842); BP rate responses were orthogonal not related to BR and BGE (Pearson r between log10BP and log10BGE and log10BGE = 0.295 and 0.229, respectively).

Fig. 4
figure 4

Partial redundancy analysis (RDA) plots of relationships between predictor variable sets and biological and bacterial metabolism responses in the Rio Grande drainage. a Relationships between physicochemical and nutrient predictors and biological and bacterial metabolism response variables after accounting for the influence of DOC quantity–quality and bacterial stoichiometric predictors. b Relationships between DOC quantity–quality and bacterial stoichiometric predictors and response variables after accounting for the influence of physicochemical and nutrient predictors. c Relationships between the global (complete) predictor variable sets (physicochemical and nutrient predictors, DOC quantity–quality and bacterial stoichiometric predictors) and response variables. Response variables are in italics and in boxes in all RDA plots. Bacterial (< 1 μm) C:N and N:P are not presented because they have little predictive power and were plotted near the origin. See Tables 1 and 2 for variable abbreviations. Percent of the variance in response variables explained by each RDA axis (RDA 1 and 2) is provided in parentheses in axis titles

Full size image

The RDA model examining biological and bacterial C metabolic responses to C quality–quantity predictors was significant (p = 0.002), explaining 25.6% of the variation in the data (first two axes explaining 20.8% of the variation; Fig. 4b). Abs440 (p = 0.002) and DOC concentration (p = 0.013) were the significant predictors in the model. In this model, BA, Chla, and bacterial production were correlated with DOC and OCL, whereas BR was correlated with Abs440. Again, BR and BGE were negatively correlated with each other (with BP responses being orthogonal), indicating that variation in BGE was being primarily driven by changes in BR.

The global RDA model which included both physicochemical/inorganic nutrient and C quality–quantity and bacterial stoichiometry predictors was significant (p = 0.001) and explained a substantial portion (39.5%) of the variation in biological and bacterial C metabolic responses across the Rio Grande drainage, with the first two axes explaining 16% of the variation (Fig. 4c). In the global model, NH4+ (p = 0.008), TP (p = 0.004), Q9 (p = 0.012), NVSS (p = 0.026), and Abs440 (p = 0.007) were significant predictors. There was a substantial covariation among physiochemical and C quantity–quality predictors across the drainage. In the global model, BP rates, BA, and Chla were greater at sites with higher NH4+ and DOC concentrations, whereas BR rates increased with NVSS, TP, and Abs440. Variance partitioning indicated that C quality–quantity predictors accounted for slightly more (17.2%) of the variation than physiochemical variables (13.9%). Covariation among variables in the predictor variable data sets is portrayed by the amount of shared variation between the two groups of predictor variables and 8.0% shared variance was explained.

Discussion

The Rio Grande basin exhibited substantial spatial variation in physicochemical characteristics, divided into two main gradients across the drainage. The primary physicochemical gradient was an upstream-to-downstream gradient wherein the upstream sites in the more arid portion of the drainage had higher salinity and nutrient concentrations, and downstream sites had lower salinity and higher discharge and water temperatures. In many United States Gulf Coast drainages, nutrients and turbidity generally increase in an upstream-to-downstream direction with river order and river gradient changes (e.g., Becker et al. 2014, 2017), but the presence of two large reservoirs in the Rio Grande drainage appeared to strongly modify upstream–downstream physicochemical gradients. Impoundments typically lead to increased water clarity, changes in the dominant nutrient forms, and alteration of OM sources supporting food webs below dams (Wetzel 2001; Hoeinghaus et al. 2007; Proia et al. 2016). LD sites in the Rio Grande drainage were located between and below Amistad and Falcón International Reservoirs and nutrient and suspended solid concentrations were substantially lower at LD sites, indicating that these impoundments serve as a primary driver of physicochemical and water quality differences observed along the upstream–downstream gradient in the drainage. In addition, the effect of impoundments is apparent when examining variation in physicochemical conditions within some of the individual site groupings in the drainage. Within the UG group, the Río Conchos from Mexico discharges into the Rio Grande below site RG1 and accounts for up to 40% of the flow below this site (Douglas 2009). The Río Conchos originates in the Sierra Madre Occidental and the drainage contains seven reservoirs; below the confluence of the Río Conchos and the Rio Grande (between sites RG1 and RG2), NVSS concentrations are reduced in the Spring and Summer (mean reduction ~ 65%). However, despite inflow of the Río Conchos, NVSS and TP concentrations at the RG2 site are still elevated when compared to LD sites (Table 1). It is also noteworthy that upstream sections of the Pecos River and Rio Grande above the furthest upstream study sites also contain impoundments (i.e., Red Bluff Reservoir at the Texas–New Mexico border on the Pecos River and Elephant Butte and Caballo Reservoirs in New Mexico on the Rio Grande). However, these reservoirs are substantially smaller than Amistad and Falcón (Caballo = 4500 ha, Red Bluff = 4530 ha, Elephant Butte = 15,000 ha, Amistad = 26,000 ha, Falcon = 33,000 ha), are relatively distant from sampling sites, and thus likely had marginal influence on patterns observed in this study.

The second physicochemical gradient in the drainage was between groundwater-influenced and the surface water-dominated sites, where groundwater-influenced sites exhibited lower nutrient concentrations (except NO3−), DOC, and suspended matter. The two sampling sites in the downstream portion of the Pecos River (PR3 and PR4) and the Devils River site are highly influenced by groundwater contributions from the Edwards–Trinity aquifer, which is characterized by high water clarity, greater NO3− concentrations, and low P concentrations (Mace et al. 2004). The Devils River is almost entirely supported by groundwater contributions during base flow conditions and exhibits little temporal variation in physicochemical conditions (i.e., consistent high water clarity and low nutrients; see Fig. 2). However, substantial groundwater contributions to rivers whose upstream sections are dominated by surface water can augment and dilute flows, leading to upstream–downstream declines in nutrients and turbidity. For example, groundwater contributions along the Pecos River lead to a notable reduction in nutrients and DOC between PR1 and PR4 (Table 1). Thus, groundwater contributions play an important role in influencing spatial variation in physicochemical characteristics within the Rio Grande drainage through dilution of more turbid and nutrient-rich surface waters and by providing more stable base flows. The Rio Grande drainage is in an arid and semiarid landscape with limited surface water, and groundwater overexploitation and extraction have the potential to destabilize base flows and decrease water quality in portions of the drainage.

Consistent with our first hypothesis, we found that spatial variation in hydrology and physicochemical conditions in the Rio Grande drainage led to differences in biological and bacterial C metabolism responses. In general, Chla, BA, bacterial C metabolism varied with both the upstream–downstream and surface water–groundwater gradients in the drainage. Sites dominated by groundwater contributions had lower Chla, BA, and BP rates. This spatial variation is likely in response to the nutrient and DOC gradients in the drainage; increased nutrients (in this case P and NH4+) and DOC supply can increase plankton production and biomass (Murrell 2003; Holmes et al. 2008; Guenet et al. 2010).

Labile OC concentration differed among site groups in the Rio Grande drainage, with the UD and GW sites having the highest and lowest concentrations, respectively. (LD sites were intermediate and did not differ from UD and GW sites.) DOC concentration varied among all site groups; however, the fraction of the DOC pool composed of OCL (OCL/DOC) did not differ across site groups. Across study sites, OCL comprised 8–83% of the bulk DOC pool (mean = 31%) and this is in general agreement with previous studies across aquatic habitats (i.e., range 5–30% of the bulk DOC pool; Sondergaard and Middelboe 1995; Amon and Benner 1996). In the present study, the mean percent of bulk DOC composed of OCL is at the higher end of this range, but most observations fall within the reported range of values. The proportion of the DOC pool composed of relatively labile C sources is equivocal and can depend upon a variety of factors; Sondergaard and Middelboe (1995) state that riverine systems contain relatively high amounts of labile DOC, whereas others contend that riverine DOC contains a smaller fraction of labile C than lacustrine systems (del Giorgio and Davis 2003). Mean concentration of DOC in the Rio Grande drainage (4.23 ± 0.28 mg L−1) was approximately half that of the lakes and rivers examined by del Giorgio and Davis (2003), but OCL/DOC values are similar to OCL/DOC values for lakes. Thus, in comparison with other studies, DOC concentrations in the Rio Grande are relatively low, but the fraction of the bulk DOC pool composed of labile DOC in the Rio Grande drainage is high. In addition, the concentration of OCL was not related to DOC concentration across the Rio Grande drainage (Pearson r = 0.33, p = 0.094), further corroborating cross-system analyses that conclude that bulk DOC concentration does not reliably predict the labile fraction of the pool (del Giorgio and Davis 2003).

Environmental drivers of bacterial C metabolism responses

Bacterial C metabolism responded to environmental gradients across sites in the Rio Grande drainage and responses were correlated with both physicochemical/inorganic nutrient and C quantity–quality variables. The percent of variation in biological and metabolic responses across the drainage was approximately equivalent when physicochemical and C quality–quantity predictors were compared (13.9% vs. 17.2%, respectively). This finding did not support our hypothesis that variation in C quality–quantity would be the primary driver of bacterial C metabolism across the drainage. In addition, there was a substantial amount of covariation among many of the individual physicochemical and C quantity–quality predictors, leading to ~ 8% of the variation in biological and metabolic responses being shared between the two predictor sets. For example, DOC concentration was positively correlated with NH4+ (Pearson r = 0.54, p = 0.004) and TP concentration (Pearson r = 0.40, p = 0.039). In addition, Abs440 was positively correlated with TP (r = 0.80, p < 0.001) and NVSS concentration (r = 0.80, p < 0.001). Thus, it appeared that physicochemical/inorganic nutrient and C quality–quantity conditions were responding similarly to the overall gradients across the Rio Grande drainage.

In the current study, BA and BP were positively related to NH4+ and DOC concentrations across the Rio Grande drainage. BP responds to a variety of biotic and abiotic factors, making it difficult to predict the relative importance of individual drivers influencing rates (del Giorgio and Cole 1998; Bergstrom and Jansson 2000). Numerous studies have found that BP can be limited by inorganic nutrients or DOC, depending on in situ conditions (e.g., Makino et al. 2003; Lennon and Cottingham 2008). BP responses to increasing bulk DOC concentration are equivocal; some studies find no relationship between BP and DOC concentrations (e.g., del Giorgio and Davis 2003; Judd et al. 2006; Becker et al. 2017), while others found that BP increases with DOC supply (e.g., Lennon and Pfaff 2005; Lennon and Cottingham 2008). Bulk DOC pools in natural systems are complex and composed of C sources of different quality and decomposition rates (Sondergaard and Middelboe 1995) and BP response to DOC is also dependent upon the timing of DOC supply (e.g., pulsed vs. continuous; Findlay 2002; Lennon and Cottingham 2008). In addition, supply of inorganic nutrients, specifically P is often a determinant of bacterial growth (e.g., Makino et al. 2003; Lennon and Pfaff 2005). In the present study, BA and BP were not related to TP or SRP concentrations, indicating that external P supply likely did not strongly affect spatial patterns of BA and BP across the drainage. However, we also found that BP concomitantly increased with Chla across the Rio Grande drainage (Pearson r = 0.73, p < 0.001), suggesting that either (1) bacteria and phytoplankton populations were simultaneously responding to the same physicochemical/inorganic nutrient gradients, or (2) that suspended phytoplankton was positively associated with increased inorganic nutrient supply (i.e., NH4+) and bacteria were responding to changes in phytoplankton biomass through utilization of labile C algal cell exudates (e.g., Wolter 1982). BP rates were not related to OCL concentrations across the Rio Grande drainage, indicating that specific mechanisms behind relationships between BP, DOC, inorganic nutrients, and phytoplankton remain to be elucidated.

In contrast to BP, BR rates responded to environmental drivers linked to specific in situ conditions across the Rio Grande drainage. BR rates can be affected by a diversity of factors, including temperature (del Giorgio and Cole 1998), DOC quantity and quality (del Giorgio and Cole 1998; Lennon and Pfaff 2005; Kritzberg et al. 2005), bacterial biomass (Hamdan and Jonas 2006), and community composition (Judd et al. 2006). In the present study, higher BR rates were correlated with Abs440 (water color), TP concentration, and NVSS concentration. Water color is often used as an indicator of DOC allochthony, and highly colored DOC is considered refractory (Kritzberg et al. 2006; Jones et al. 2009). Respiration rates of bacterial communities increase with concentrations of allochthonous and refractory DOC, regardless of the availability of OCL in the overall DOC pool (del Giorgio and Davis 2003; Berggren et al. 2007). In addition, BR rates across the Rio Grande drainage were positively correlated with NVSS concentrations, suggesting that BR was likely responding to localized terrestrial runoff contributions, and therefore, allochthonous C sources and not necessarily to the bulk DOC pool.

Differential responses of BP and BR to OC supply and lability serve as one of the mechanisms for widespread occurrence of net heterotrophy in aquatic systems (Cole and Caraco 2001). Decoupling of BR and BP has been attributed to a number of different factors including a lack of BR responsiveness to C quality (Lennon and Cottingham 2008) and the presence of localized “hot spots” of increased bacterial C metabolism in areas where aquatic and terrestrial ecosystems interface (del Giorgio et al. 2006). In the Rio Grande drainage, BP and BR responded to different environmental factors: BP increased with increased concentrations of labile and total DOC and NH4+ (which were also correlated with Chla), while BR increased with water color and NVSS. These findings suggest that different aspects of bacterial C metabolism (BR vs. BP) in the Rio Grande drainage are responding to different mechanisms of nutrient and OC supply that exist across the drainage: BP is likely primarily responding to autochthonous C supply and inorganic nutrients, while BR is responding to localized supplies of allochthonous particulate matter and DOC.

In the Rio Grande drainage, BGE did not differ among site groups and there was no obvious relationship with hydrological and physicochemical gradients. The range of observed BGE values in the Rio Grande drainage (range 0.01–0.79) generally falls within those of riverine BGE values reported by del Giorgio and Cole (1998) (range 0.03–0.46), but BGE values across the drainage were more variable than those reported by del Giorgio and Cole (1998). In addition, relationships between BP, BR and BGE were different from those observed in several other studies. In other systems, BGE increases with BP (Maranger et al. 2005; Amado et al. 2013), but in the present study BGE was not related to BP (Pearson r = 0.05, p = 0.785), but instead declined with increasing BR (Pearson r = − 0.50, p = 0.008). Calculation of BGE requires estimates of both BR and BP, but BR is often cited as being less responsive than BP to changes in environmental conditions (del Giorgio and Cole 1998). This was not the case in the Rio Grande: BR rates exhibited substantially more variation (range 0.77–114.29 μg C L−1 h−1) than BP rates across the drainage (range 0.40–7.39 μg C L−1 h−1). Other studies have similarly found that BGE varied directly with BR (and not BP) and suggested that variation in BGE across a subtropical riverine network was a more related to temperature responses of BR rather than DOC or inorganic nutrients (Becker et al. 2017). However, in the Rio Grande drainage neither BGE (Pearson r = 0.11, p = 0.595) nor BR (Pearson r = 0.05, p = 0.819) were related to water temperature. Thus, in the Rio Grande drainage BGE was driven by conditions that affected BR rates, namely DOC color and allochthonous sediment loading and BR was more sensitive than BP to changes in landscape position and environmental conditions.

Conclusions

Arid river systems are thought to be largely dependent on autochthonous C (Bunn et al. 2003), but the results of this study show that allochthonously derived C also influenced spatial variation in bacterial C metabolism across the Rio Grande drainage. Although bacteria are the greatest single contributor to ecosystem respiration in many aquatic systems, widely used riverine conceptual models do not incorporate bacterial DOC processing into frameworks (e.g., Vannote et al. 1980; Junk et al. 1989). The present study joins the growing body of work which demonstrates the importance of aquatic bacteria communities in the sequestration and transformation of both allochthonous and autochthonous C and the crucial role that bacterial C metabolism plays in ecosystem function at the landscape scale (e.g., Becker et al. 2017). Finally, this study suggests that anthropogenic alteration of arid river hydrology (e.g., reservoir construction) and inputs of groundwater affect biogeochemical processes at the riverscape scale, which has implications for C and nutrient sequestration, transformation, and transport within the drainage, and ultimately delivery to the ocean.