Consequences of an ecosystem state shift for nitrogen cycling in a desert stream

Cessation of cattle grazing has resulted in the reestablishment of wetlands in some streams of the U.S. Southwest. Decades of cattle grazing prevented vascular plant growth in Sycamore Creek (Arizona, U.S.A.), resulting in stream reaches dominated by diatoms and filamentous green algae. Establishment of vascular plants can profoundly modify ecosystem processes; yet, the effects on nitrogen (N) cycling remain unexplored. We examined the consequences of this ecosystem state shift on N cycling in this N‐limited desert stream. We compared results from whole‐reach ammonium‐N stable isotope (15NH4+) tracer additions conducted before (pre‐wetland state) and 13 yr after (wetland state) free‐range cattle removal from the watershed. Water column estimations showed that in‐stream N uptake and storage were higher in the pre‐wetland than in the wetland state. N turnover was also higher in the pre‐wetland state, indicating that assimilated N was retained for shorter time in stream biomass. In addition, N uptake was mostly driven by assimilatory uptake regardless of the ecosystem state considered. Water column trends were mechanistically explained by the fact that the dominant primary uptake compartments in the pre‐wetland state (i.e., algae and diatoms) had higher assimilatory uptake and turnover rates than those in the wetland state (i.e., vascular plants). Overall, results show that the shift in the composition and dominance of primary producers induced by the cessation of cattle grazing within the stream‐riparian corridor changes in‐stream N processing from a dominance of intense and fast N recycling to a prevalence of N retention in biomass of primary producers.

(pre-wetland state) and 13 yr after (wetland state) free-range cattle removal from the watershed. Water column estimations showed that in-stream N uptake and storage were higher in the pre-wetland than in the wetland state. N turnover was also higher in the pre-wetland state, indicating that assimilated N was retained for shorter time in stream biomass. In addition, N uptake was mostly driven by assimilatory uptake regardless of the ecosystem state considered. Water column trends were mechanistically explained by the fact that the dominant primary uptake compartments in the pre-wetland state (i.e., algae and diatoms) had higher assimilatory uptake and turnover rates than those in the wetland state (i.e., vascular plants). Overall, results show that the shift in the composition and dominance of primary producers induced by the cessation of cattle grazing within the stream-riparian corridor changes in-stream N processing from a dominance of intense and fast N recycling to a prevalence of N retention in biomass of primary producers.
Ecosystem state shifts refer to changes in ecosystem structure that encompass time scales exceeding the range of natural variability after episodic or abrupt stochastic events, anthropogenic forcing, or due to the ascendance of key species commonly referred as ecosystem engineers. Stochastic events include weather extremes, fires, or pest outbreaks, whereas anthropogenic forcing is associated with land transformation, habitat fragmentation, and climate change (Scheffer et al. 2001;Steffen et al. 2011). Ecosystem engineers refers to organisms, such as several animals (e.g., beavers, salmons, and hippos) with the capacity to physically modify the abiotic environment (Moore 2006;Dutton et al. 2021). The resulting alternative ecosystem states are often characterized by a different physicochemical template and by assemblages of primary producers that differ from the previous ecological community in terms of standing stock, morphologic, and metabolic characteristics (McGowan et al. 2005;Law et al. 2016;Dutton et al. 2021). In addition, the persistence of new assemblages of primary producers is reinforced by positive feedbacks, such as anchoring of sediment by vegetation that prevents scour and removal of that vegetation (Heffernan 2008;Dong et al. 2016). Consequences of ecosystem state shifts have mainly been described in marine and terrestrial ecosystems, such as coral reefs, rocky macroalgal communities, woodlands, deserts, and lakes (Scheffer et al. 2001;Conversi et al. 2015;Boada et al. 2017); less so in streams and rivers (but see Ib añez et al. 2012;Brighenti et al. 2019).
A decade ago, Heffernan and co-authors documented an ecosystem state shift in a desert stream in the U.S. Southwest that corresponded to a change in the watershed management practices (Heffernan 2008;Heffernan et al. 2008). Cattle removal from the watershed, per an agreement between the U.S. Forest Service and the ranchers, was instigated by a shift in federal grazing policy in western United States as a result of the cessation of cattle grazing along the stream-riparian corridor, stream reaches dominated by algae on gravel substrata to a state with abundant emergent vascular plants, which slowed water flow rates and promoted wetland conditions within the stream channel. The resulting wetlands represent an alternative stable state of desert streams (Heffernan 2008). This ecosystem state shift and changes associated with it have since been intensively investigated in Sycamore Creek Dong et al. 2016Dong et al. , 2017.
Prior to cattle removal, high light levels reaching the stream bottom supported high rates of algal production (Busch and Fisher 1981;Grimm 1987). Under baseflow conditions, blooms of filamentous green algae were typical (Fisher et al. 1982;Grimm 1987), and patches of vascular plants were present but rare and limited in spatial extent (Dudley and Grimm 1994). The cessation of cattle grazing in the streamriparian zone caused an increase in the abundance of vascular plants in many areas of the stream channel, which came to cover a remarkable portion (up to 45%) of the main stem of Sycamore Creek, with > 80% vascular plant cover in some reaches (Dong et al. 2016(Dong et al. , 2019. Wetland establishment drives changes in the physical, chemical, and biological structure of desert streams, reducing surface-flow velocity and water exchange between the surface and subsurface, increasing standing crops of live biomass and organic detritus, and increasing spatial heterogeneity (Heffernan 2008;Heffernan et al. 2008;Dong et al. 2017).
A shift in dominance of algae to vascular plants is also expected to strongly affect stream ecosystem processes, including nutrient cycling (Twilley et al. 1985;Levi et al. 2015;Riis et al. 2019). This effect may be especially relevant for nitrogen (N) cycling in desert streams of the U.S. Southwest because this element is the limiting nutrient for in-stream biota in these ecosystems (Grimm and Fisher 1986;Peterson and Grimm 1992). Uptake of N in streams is primarily driven by benthic organisms, which have recently been referred to as primary uptake compartments (PUCs) because they meet their N demand via direct uptake of N from the water column (Tank et al. 2018). Biotic compartments of N uptake include both autotrophic microorganisms (i.e., epilithic biofilm dominated by microalgae, filamentous green algae, vascular plants, and bryophytes) and heterotrophic microorganisms (i.e., bacteria and fungi in biofilms on fine benthic organic matter [FBOM], decomposing leaves, and wood). Autotrophs show large variation in nutrient uptake and turnover rates (Mulholland et al. 2000;Peipoch et al. 2014), and among them, vascular plants tend to have higher nutrient and energy requirements than filamentous algae or single-celled organisms such as diatoms (Dodds et al. 2004;Allen et al. 2005).
Therefore, the shift in the dominant primary producers within the stream associated with cattle removal may directly affect N cycling at the reach scale. In addition, the development of vascular plants in the stream channel may indirectly affect N cycling via changes in stream hydromorphology. For instance, the reduction of water exchange between surface and subsurface water associated with plant establishment may decrease the contribution of the hyporheic zone, a key N retention compartment, on in-stream N uptake (Hall et al. 2002;Wollheim et al. 2014;Drummond et al. 2016). Conversely, the presence of plants may also increase water residence time within the stream channel, which may enhance N retention (Salehin et al. 2003;Nikolakopoulou et al. 2018).
Here, we aimed to examine how changes in the composition and dominance of primary producers caused by the ecosystem state shift in desert streams affected in-stream N cycling. We compared results from whole-reach 15 N stable isotope tracer additions conducted before and after free-range cattle removal from the Sycamore Creek watershed. Specifically, we assessed changes in N cycling processes between a stream reach in pre-wetland state (i.e., dominated by filamentous green algae and diatoms) sampled in 1997, and a stream reach in wetland state (i.e., extensively colonized by emergent vascular plants) sampled in 2013. We hypothesized that differences in the structural and metabolic characteristics of the dominant primary producers between the two ecosystem states would be reflected in contrasting N cycling at the reach scale. Specifically, we predicted (1) lower N uptake and storage in the pre-wetland state, because algae and bacteria have smaller assimilative requirements for biomass than vascular plants and (2) higher N turnover in the pre-wetland state, because algae and bacteria have higher growth rates but shorter lifespans than vascular plants.

Study site and sampling design
Sycamore Creek is a~65-km-long desert stream located in the Sonoran Desert (32 km NE of Phoenix, Arizona, U.S.A.), which drains a 505 km 2 watershed of mountainous terrain where elevation ranges from 427 to 2164 m. Annual precipitation (average between 7 and 43 cm) occurs during the monsoon season (July-September), producing intense, isolated rainfall and flash flooding, and during winter (January-March), when rain and flooding are more prolonged (Schade et al. 2005). This precipitation seasonality coupled with high interannual variability in rainfall results in a disturbance regime that exhibits widely varying frequency, intensity, and timing of flash floods (Grimm and Fisher 1989). The hydrological regime greatly affects the temporal dynamics of stream ecosystem structure and function, the morphology of the stream channel, and the temporal variation in nutrient concentrations. The stream base flow (i.e.,~100 and~10 L s À1 in winter and summer, respectively) may increase by as much as Ribot et al.
Ecosystem state shift affects N cycling 4-5 orders of magnitude during flash floods, which maintains an active channel that is much wider than the wetted stream and light availability is usually not a limiting factor for development of photoautotrophs (Peterson and Grimm 1992). Concentration of dissolved inorganic nitrogen (DIN) is consistently very low (< 10 μg L À1 ) throughout the year, except during floods. Availability of N has been shown to be a limiting factor for the communities and the ecosystem function of this stream (Grimm and Fisher 1986 (Fig. 1). Wetland expansion began in the year 2000, initially with aquatic vascular plants such as Veronica spp. and Nasturtium officianale and the grass, Paspalum distichum (Heffernan 2008), later giving way to extensive stands of cattail, Typha spp. and bulrush, Schoenoplectus americanus (Dong et al. 2016). Wetland reaches are concentrated near locations of high water permanence (Dong et al. 2016(Dong et al. , 2017, which included the two sites selected for this study. Although wetland vegetation was present at both sites in 2013, the more downstream site (used in 1997) had a much wider channel owing to extensive flooding in 2010, and the wetlands did not extend across the channel. Furthermore, because Sycamore Creek has experienced overall flow reductions during the current decades-long drought in the U.S. Southwest, there was a risk that the 1997 (downstream) site would dry before we could complete the experiment. Thus, we opted to conduct the tracer additions in the upstream site, which supported bank-to-bank cover of S. americanus, to maximize the differences in wetland cover between the two experiments.
In the pre-wetland state, the addition was done in a 180-m open reach where the wetted channel was dominated by diatoms and filamentous green algae (Chlorophyta), such as Cladophora glomerata and Zygnematales (Zygnema spp. and Spirogyra spp.) In the wetland state, the addition was done in the upstream reach (170 m long), where the wetted channel was dominated by riverine aquatic emergent vascular plants (mainly S. americanus and Typha domingensis, with some P. distichum). Mature willows (Salix goodingii) and cottonwoods (Populus fremontii) dominated the areas immediately lateral to the wetted channel, providing some shade throughout the growing season (February-July). The stable isotope additions started on 01 May 1997 and 04 June 2013 for the pre-wetland and wetland states, respectively. Both additions were done under base flow conditions and after more than 1 month since the last flood, thus the in-stream communities were well developed and representative of each ecosystem state.

Field methods
To assess N cycling at the reach scale and the role of different primary producers under the two alternative states, we conducted constant-rate additions of 15 N (as 15 NH 4 + ) for seven consecutive days in each reach. To determine the most abundant PUCs in the pre-wetland and wetland state stream reaches, we measured the percent coverage of each PUC in the reach. We defined 6-8 cross-sectional transects as sampling stations, which were evenly distributed along each study reach. In addition, we defined one sampling station upstream from the addition points used as reference for background conditions. In the wetland state reach, we installed four PVC piezometers (internal diameter: 2.5 cm) in the stream thalweg at 20, 45, and 60 m downstream and 5 m upstream of the addition point to collect samples of subsurface water. Piezometers reached a depth of~25 cm below the sediment surface. Prior to the start of the 15 NH 4 + addition, we collected samples of surface and subsurface water at the sampling stations to measure ambient water chemistry, electrical conductivity (EC), and δ 15 N signatures of NH 4 + and NO 3 À . To collect subsurface water, we purged the piezometers twice before taking

Ribot et al. Ecosystem state shift affects N cycling
the sample with a 100-mL syringe connected to silicone tubing inserted into the piezometer. We also collected samples from the most abundant PUCs at each station to estimate N standing stocks (mg N m À2 ) and background δ 15 N signatures.
In the case of vascular plants, at each station we collected samples of aboveground (i.e., stem and leaves separately) and belowground (i.e., roots) biomass to estimate δ 15 N signatures. We repeated the sampling of water and PUCs for nutrient chemistry and isotopic 15 N signatures at the same stations after 7 d, under conditions when the 15 N tracer addition had reached plateau along the reach, as verified with a conservative tracer (bromide; see below). We filtered all water samples in the field through ashed Whatman (Maidstone, U.K.) GF/F glass-fiber filters. We stored the filtrate on ice in the field and kept it frozen in the laboratory until further processing and analysis. We stored samples for PUCs in plastic bags and brought them to the laboratory for further processing. We sampled the following PUCs in the pre-wetland state reach: periphyton developed on fine gravel (predominantly diatoms), filamentous green algae (C. glomerata and Zygnematales), cyanobacteria (Nostoc spp.), and FBOM; and the following PUCs in the wetland state reach: periphyton developed on cobbles (predominantly diatoms), S. americanus, T. domingensis, FBOM, and submerged roots in the stream channel from S. goodingii growing on the stream margins.
We sampled periphyton on fine gravel, filamentous green algae, and cyanobacteria using a 24.6-cm 2 core. We collected periphyton on cobbles by scraping randomly collected rocks and filtering the slurry onto ashed and pre-weighed Whatman GF/A glass-fiber filters. We estimated cobble surface area by tracing the rock onto weighing paper and calculating the surface area compared to a known area of weighing paper. We sampled FBOM using a syringe to collect an aliquot of the material suspended by manual agitation of a known water volume contained in a plastic corer of 78.5 cm 2 . We filtered the FBOM aliquot onto ashed and pre-weighed Whatman GF/F glass-fiber filters. We sampled submerged willow roots by collecting all material found within a quadrat of 81 cm 2 . For vascular plants, we used a nondestructive approach to estimate the mean aboveground biomass for each wetland species. We averaged percent cover, stem width, and stem length throughout the entire reach (area = total reach length Â average reach width). We then applied established species-specific height-todry mass (DM) relationships to estimate total reach biomass (N. B. Grimm unpubl.). We estimated the belowground biomass of vascular plants using the above-to-belowground biomass ratio of a study conducted in a constructed treatment wetland located~80 km from the Sycamore watershed, which included the same wetland species (Weller et al. 2016). Despite different nutrient conditions, previous studies suggest that the availability of nutrients has a low effect on the above-tobelowground biomass ratio on different wetland species of the same genus (Svengsouk and Mitsch 2001;Kearney and Zhu 2012).
Once background samples were collected, we added at the head of the reach a solution containing 15 NH 4 + in the form of either 15 NH 4 Cl (in 1997) or ( 15 NH 4 ) 2 SO 4 (in 2013) and bromide as a conservative tracer in the form of NaBr. We added the enriched solution at a constant rate using a batterypowered Fluid Metering peristaltic pump. To achieve a target δ 15 N enrichment of 1000‰ in the water column, we calculated the appropriate concentration of 15 NH 4 + based on stream discharge. To verify plateau conditions along the reach during the 15 NH 4 + addition, we collected water samples at each sampling station 1 d before and at days 1, 3, and 7 after the starting of the 15 N tracer addition to measure Br-concentration. We used longitudinal changes in Br-concentration at plateau to calculate the percentage of dilution along the reach as an estimate of the lateral inflow from subsurface water. During each addition, we also measured hydraulic parameters such as discharge and water velocity by conducting slug additions of diluted NaCl (Gordon et al. 2004) on four separate days (i.e., À1, 3, and 7 d after the starting of the 15 N tracer addition). We measured EC using a Yellow Springs Instruments conductivity meter. The 15 NH 4 + tracer additions only slightly increased background NH 4 + concentrations (i.e., 0.2% and 3.2% in the pre-wetland and wetland state, respectively).
In the pre-wetland (1997) experiment, we measured dissolved oxygen (DO) concentration and water temperature at hourly intervals at the end of the reach using the Winkler titration method (APHA 1995) over 24 h. In the wetland (2013) experiment, we recorded DO concentration and percentage saturation, and water temperature at the end of the reach at 15-min intervals over four 24-h periods using a Yellow Springs Instruments model 556 multiparameter sonde (sensor accuracy: AE 2% of the reading, resolution: 0.02 mg L À1 , and calibrated prior its use following manufacturer's instructions). We used these data to estimate reachscale metabolism using the open-system, single-station approach (Hall and Hotchkiss 2017). We calculated daily rates of gross primary production (GPP) and ecosystem respiration (ER) by integrating the DO fluxes (corrected for reaeration) estimated at the bottom of the reach over 24 h. Reaeration coefficients (K oxy ) estimated with the nighttime regression method (Kosinski 1984) were 19.6 d À1 and 10.5 AE 0.2 d À1 (mean AE SE) for pre-wetland and wetland state, respectively. The reach lengths associated with the turnover of DO along the stream, estimated from Hall and Hotchkiss (2017), were 3.8 and 5.8 km for pre-wetland and wetland state, respectively. We extrapolated instantaneous respiration rates at night to 24 h to estimate ER. We multiplied daily rates of GPP and ER by the mean water column depth in the reach to obtain areal estimates.

Laboratory methods
In the pre-wetland state, we analyzed concentrations of NO 3 À and soluble reactive phosphorus (SRP) on a Bran and Luebbe TRAACS ® 800 autoanalyzer. We manually analyzed injection analysis (Lachat QC8000). We analyzed dissolved organic carbon (DOC) and total dissolved nitrogen (TDN) using a total organic carbon/nitrogen analyzer (Shimadzu TOV-C). We calculated dissolved organic N (DON) by subtracting DIN (calculated as NO 3 À + NH 4 + ) from TDN. We analyzed bromide concentrations on grab water samples using ion chromatography (Dionex ICS 2000). We reported all nutrient concentrations in units of elemental mass/volume (e.g., mg N L À1 ). We determined isotopic enrichment of 15 NH 4 + in stream water using the ammonia-diffusion procedure (Holmes et al. 1998). We used the same procedure for isotopic enrichment of 15 NO 3 À after reduction of 15 NO 3 À to 15 NH 4 + with Devarda's alloy (Sigman et al. 1997 where R sample is the 15 N : 14 N ratio in the sample and R standard is the 15 N : 14 N ratio in atmospheric N 2 (R standard = 0.0036765). We measured patch-specific DM (mg) of each PUC by oven drying the samples at 60 C to constant weight, then weighing to the nearest 0.1 mg. We estimated the total DM of each PUC in the reach using the patch-specific DM and its percent cover. We divided this DM by the surface area of the reach (length Â average width) to estimate reach-scale DM (mg m À2 ) of each PUC. We estimated reach-scale N standing stocks (mg N m À2 ) for each PUC by multiplying the reach-scale DM standing stock by the percentage of N in DM.
We prepared PUC pre-and plateau-addition samples for 15 N analysis by oven drying at 60 C to constant weight, and then processing dried samples according to specific isotope facility protocols. 15 NO 3 À , 15 NH 4 + , and PUC samples were analyzed at the University of California Stable Isotope Facility (Davis, California, U.S.A.). The N content (as a percent of DM) and stable isotope ratios were measured by continuous flow isotope-ratio mass spectrometry (20-20 mass spectrometer; PDZEuropa, Northwich, U.K.) after sample combustion in an on-line elemental analyzer (PDZEuropa ANCA-GSL).

Calculation of N uptake metrics
We calculated reach-scale and PUC-specific N uptake parameters following the procedures described in Mulholland et al. (2000) and Tank et al. (2018). We calculated the surface water 15 N-NH 4 + flux at each station by multiplying background-corrected δ 15 NH 4 + signatures at plateau 7 d by the station-specific discharge estimated based on the Br-dilution along the reach during plateau. We estimated the uptake rate per unit distance (k w , m À1 ) in the reach from the slope of the regression of the ln-transformed tracer 15 N fluxes vs. site distance from the top of the reach. We calculated the uptake length (S w , m) as the inverse of k W and converted it to uptake velocity (V f ; mm min À1 ) by dividing the stream-specific discharge (i.e., discharge/wetted width) by S w (Stream Solute Workshop 1990). We calculated the areal uptake (U; μg N m À2 s À1 ) by multiplying V f by the mean ambient NH 4 + concentration. We calculated the uncertainty for each nutrient retention metric (i.e., S w , V f , and U) by first running Monte Carlo simulations (n = 10,000) based on the mean and the standard deviation (SD) of each parameter used (i.e., discharge, k w , NH 4 + concentration, and wetted width). Then, we recalculated each retention metric but substituting the parameters by their simulations. The resulting SD was reported as uncertainty.
We estimated the fractional uptake rate per unit distance for nitrification (k NIT ) by fitting the 15 NO 3 À tracer fluxes vs. distance to a two-box model as proposed by Mulholland et al. (2000) using a nonlinear regression model with two unknown parameters (i.e., k 2 as 15 NO 3 À uptake rate and k NIT ).
We calculated the reach-scale areal uptake for nitrification (U NIT , μg N m À2 s À1 ) from k NIT , as previously described for U. We calculated the uncertainty for U NIT by first running Monte Carlo simulations (n = 10,000) based on the mean and SD of each parameter used (i.e., discharge, k NIT , NH 4 + water concentration, and wetted width). Then, we recalculated U NIT but substituting the parameters by their simulations. The resulting SD was reported as uncertainty. Furthermore, we calculated the reach-scale assimilatory uptake (U ASSIMÀWATER , μg N m À2 s À1 ) by subtracting U NIT from U. Finally, we calculated the uncertainty for U ASSIMÀWATER by running Monte Carlo simulations (n = 10,000) with the mean and SD values of each parameter and by using the same equation but with the simulation results. The resulting SD was reported as uncertainty.
To estimate the PUC-specific NH 4 + assimilatory uptake of periphyton, filamentous green algae, Cyanobacteria, and FBOM, we divided the reach-weighted mass of backgroundcorrected 15 N tracer per m 2 measured in each biotic compartment by the total 15 N addition time in seconds and the fraction of 15 N in the stream surface water flux of NH 4 + . To estimate NH 4 + assimilatory uptake of vascular plants, we used the same calculation approach but considering 15 N-NH 4 + in the subsurface water. In addition, we calculated separately NH 4 + assimilatory uptake from aboveground (sum of stem and leaves) and belowground (roots) biomass to infer where assimilated N was stored in vascular plants. For this calculation, we only considered data from the two uppermost sampling stations downstream of the addition point (i.e., 25 and 40 m, respectively) to avoid confounding effects of N regeneration along the reach. We then estimated the areal NH 4 + assimilatory uptake by PUCs at the reach scale (U ASSIMÀPUC , μg N m À2 s À1 ) as the sum of the mean compartment-specific NH 4 + assimilatory uptake of each PUC. We calculated the uncertainty for U ASSIMÀPUC by running Monte Carlo simulations (n = 10,000) with the mean and SD values of each PUC U ASSIM and then summing up U ASSIM simulations of each PUC. The resulting SD was reported as uncertainty.
For each PUC, we also calculated the biomass-specific N uptake (mg N mg N À1 d À1 ), a surrogate of N turnover rate, by dividing PUC-specific NH 4 + assimilatory uptake by the N content in PUC biomass (Dodds et al. 2004). Likewise, we estimated the biomass-specific N uptake (mg N mg N À1 d À1 ) at reach scale, a surrogate of N turnover rate for each stream reach, as U ASSIMÀPUC divided by the sum of the N content in all PUCs. We calculated the uncertainty for biomass-specific N uptake at reach scale by substituting the parameters in the equation by their corresponding simulations computed with the Monte Carlo method (n = 10,000 runs). The resulting SD was reported as uncertainty.
We calculated reach-scale N storage in the PUCs as the percentage of 15 N added during the addition that was estimated in the biomass of the different PUCs at the addition plateau. For each PUC, the total mass of stored 15 N was based on an integration of the downstream decline in compartment-specific 15 N-biomass along the reach. If the slope of the regression of 15 N biomass vs. distance was not significant (p > 0.05), we used the mean 15 N biomass for the entire reach. For this calculation, we used a reach length equal to five times the measured S w to standardize estimates for variable reach lengths across the pre-wetland and wetland reaches. This standardization ensured that > 99% of the 15 N tracer was removed; so our calculations fully encompass the reach length where biota was exposed to 15 NH 4 + from the water column (Mulholland et al. 2000). We performed the computations in R version 1.4.1106 (R Core Team 2018). Statistical results were evaluated at the α = 0.05 significance level.

Environmental characteristics of surface water in the two alternative states
The pre-wetland state reach was wider and shallower than the wetland state and had a higher discharge throughout the 15 N tracer additions ( Table 1). Results of NaCl additions from day À1 and day 7 indicated that discharge decreased by 15% and 27% in the pre-wetland and wetland state reaches, respectively, due to extreme weather conditions. Dilution factors along the two reaches calculated from longitudinal patterns of bromide concentrations at plateau on day 7 were 5% and 1% for pre-wetland and wetland state, respectively. The two reaches had similar water temperature, whereas EC was lower in the pre-wetland state and concentration of DO was higher in the pre-wetland state (Table 1). Ambient NO 3 À and NH 4 + concentrations were similarly low between the two study reaches ( Table 1). Concentration of DON was an order of magnitude higher than that of DIN and similar between stream states ( Table 1). Concentration of SRP was four times lower in the pre-wetland state; however, low DIN : SRP ratios (< 3) indicated potential N limitation in the two study states (Table 1). Concentration of DOC was slightly lower in the pre-wetland state reach (Table 1). The daily GPP rate was higher in the prewetland state, and ER was similar in both states. Thus, GPP : ER ratio shifted from 1 in the pre-wetland state to < 1 in the wetland state, indicating a more heterotrophic system in the latter state. It is worth noting that the open-channel method used to measure metabolism relies on water-column changes in DO concentration and cannot adequately estimate the photosynthesis and respiration rates of emergent vascular plants. Therefore, these results reflect only the metabolic activity associated with submersed benthic organisms and we Table 1. Physicochemical characteristics and daily rates of reach-scale metabolism for the pre-wetland and wetland state study reaches during the two 15 NH 4 + additions. Data for physical parameters (except temperature) are the mean (AE standard error; SE) of the four NaCl additions conducted during each addition period. Data for water temperature and DO are the mean (AE SE) of the 24-h records used to infer metabolism. Data for EC and nutrient concentrations are the mean (AE SE) from samples collected at the different sampling stations along the study reaches at the pre-sampling and the 7-d plateau, respectively. Daily rates of reach-scale GPP and ER were estimated on a single date on the pre-wetland addition and on four dates (mean AE SE) on the wetland addition. cannot draw conclusions about total system production or respiration.

Biomass standing stocks of PUCs in the two alternative states
The total N biomass in PUCs was lower in the pre-wetland reach than in the wetland reach: 4750 (1190) and 7813 (1689) mg N m À2 , respectively (mean and SD value based on Monte Carlo method). However, the type and relative contribution of the PUCs to total N biomass clearly differed between the two states ( Fig. 2A). In the pre-wetland state, algae and detrital compartments accounted for the majority of reach-scale N biomass. In particular, C. glomerata (36.1%) and FBOM (26.1%) contributed the most to the total N biomass, followed by periphyton (18.9%), Zygnematales (10.5%), and Nostoc spp. (8.4%). In the wetland state reach, the N biomass at the reach scale was dominated by emergent vascular plants, particularly S. americanus, accounting for 78.3% of N biomass, of which 58% was accumulated in belowground biomass (Fig. 3A). The other PUCs contributing to reach-scale N biomass in the wetland state were roots from riparian S. goodingii (15.5%), periphyton (3.1%), and FBOM (2.5%). T. domingensis accounted for the remaining 0.6% of N biomass in the reach, mostly accumulated in aboveground biomass (Fig. 3A).

N uptake in the two alternative states
In the pre-wetland state, S w was two times shorter, V f was one order of magnitude higher, and U was 2.5 times higher than in the wetland state, indicating that the pre-wetland state was more efficient in retaining N and had a higher N uptake capacity (Table 2). Nitrification (U NIT ) accounted for 25.2% of U in the pre-wetland state. In the wetland state, U NIT was nil since no significant increase of 15 NO 3 À was detected along the reach during the 15 N addition. Therefore, in the two study reaches, U at the reach scale was dominated by assimilatory uptake (U ASSIMÀWATER ; Table 2). The sum of assimilatory uptake by PUCs (U ASSIMÀPUC ) at the reach scale was three-to fourfold higher than that measured from the water column (i.e., U ASSIMÀWATER ) in both study reaches. In addition, in the pre-wetland state, U ASSIMÀPUC was twofold higher than in the wetland state ( Table 2). The relative contribution of each PUC to U ASSIMÀPUC varied between the two state reaches (Fig. 2B). In the prewetland reach, C. glomerata was the PUC with the highest N demand, accounting for 41.8% of U ASSIMÀPUC , followed by periphyton (22.8%) and Zygnematales (20.5%). The contribution of Nostoc spp. and FBOM to U ASSIMÀPUC was similar (7.5% and 7.3%, respectively). In the wetland reach, willow roots and S. americanus showed the highest contribution to U ASSIMÀPUC , accounting for 85% of U ASSIMÀPUC . Periphyton and FBOM accounted for 10.7% and 4.1% of U ASSIMÀPUC , respectively, whereas T. domingensis accounted for only 0.2%.
The portion of added 15 N stored in PUCs along the reach differed between the two stream states (Table 2; Fig. 2C). In Fig. 2. Reach-scale N biomass (A), assimilatory NH 4 + uptake (U ASSIMÀPUC ) (B), and N storage as percentage of total added 15 N (C) in each of the two study reaches (pre-wetland and wetland state). Different fill in each bar represents the relative contribution of PUCs to reach-scaled totals. FBOM refers to fine benthic organic matter. For clarity, we separated PUCs in three different classes: microbes and algae, FBOM, and vascular plants.
the pre-wetland reach, 100% of the 15 N added was stored in the sampled PUCs, whereas in the wetland reach, only 33.5% of the 15 N added was stored in the sampled PUCs (Table 2). In addition, the relative contribution of each PUC to N storage also differed between the study state reaches (Fig. 2C). In the pre-wetland reach, 53.6% of the total 15 N stored was found in C. glomerata, followed by periphyton and Zygnematales (24.7% and 11.0%, respectively). Nostoc spp. and FBOM stored 2.0% and 9.1%, respectively. In the wetland reach, most N storage occurred in S. americanus (59.9%) and in S. goodingii roots (27.2%). Periphyton (8.5%), FBOM (4.0%), and T. domingensis (0.4%) accounted for the remaining 15 N stored (Fig. 2C). Regarding the location of N stored in vascular plants, 71.1% of total 15 N assimilated by S. americanus was stored in belowground biomass, whereas in T. domingensis, N was mostly stored in aboveground biomass (78.3%; Fig. 3B).
The reach-scale N turnover rate was four times higher in the pre-wetland state than in the wetland state (0.03 and 0.007 mg N mg N À1 d À1 , respectively; Table 2). Within each state, the N turnover rate varied considerably among PUCs (Fig. 4). Filamentous green algae (C. glomerata and Zygnematales) and periphyton showed the highest N turnover rates, and the two vascular plants had the lowest rates (Fig. 4). Turnover rates of periphyton and FBOM were similar for the two states.

Discussion
We compared N uptake, storage, and turnover, derived from 15 N-NH 4 + tracer additions, between two stream reaches representative of two alternative ecosystem states in Sycamore Creek, a desert stream in Arizona, U.S.A. (Fig. 1). The reach in the pre-wetland state was characterized by a gravel-bed reach dominated by Chlorophyta and microbial  communities on streambed sediments (i.e., periphyton) and detritus (i.e., FBOM). The reach in the wetland state was characterized by a dominance of emergent vascular plants and riparian trees and shrubs, a result of cattle cessation along the stream-riparian corridor (Heffernan 2008;Heffernan et al. 2008). Based on the characteristics of the dominant primary producers, we expected significant differences in whole-stream metabolic activity and thus, in the N uptake between the two ecosystem states. Since vascular plants tend to have higher biomass and nutrient and energy requirements, and longer lifespan than filamentous green algae or diatoms, we predicted a lower N uptake demand and N storage capacity but a higher turnover rate in the prewetland state than in the wetland state. Results only partially matched our predictions. Measured S w and V f in the two states (Table 2) were in the high range of stream values in terms of N uptake efficiency (i.e., short S w values) and N demand (i.e., high V f values) of streams elsewhere, which is expected for N-limited streams (Ensign and Doyle 2006;Ribot et al. 2017). Nevertheless, S w was threefold shorter and V f was one order of magnitude higher in the prewetland state than in the wetland state (Table 2). These results indicate that in-stream efficiency of N retention is higher in the pre-wetland state than in the wetland state. Unexpectedly, we observed that areal N uptake (i.e., U) and storage in the pre-wetland state were both higher than in the wetland state (Table 2). Interestingly, a previous study conducted in the same stream in the 80s (i.e., pre-wetland state) showed that the N storage capacity can be even higher than that measured in our study (Grimm 1987). Nonetheless, according to our predictions, the turnover rate in the pre-wetland state was lower than in the wetland state. Altogether, our results indicate that the two ecosystem states cycled N efficiently, but through different biogeochemical pathways. Higher N uptake and turnover indicates fast in-stream N cycling in the pre-wetland state, whereas both lower N uptake and N turnover indicates that in-stream N cycling is slowed down in the wetland state and that it was mostly characterized by more persistent N retention in vascular plants (as summarized in Fig. 5).
There are several mechanisms that may explain observed differences in U between the two stream ecosystem states. Reach-scale NH 4 + uptake is mainly driven by nitrification and assimilatory uptake in streams (Peterson et al. 2001;Ribot et al. 2017). Nitrification, the aerobic oxidation of NH 4 + to Fig. 5. Summary of N uptake, storage and turnover in the pre-wetland and wetland reaches. Percentages in boxes reflect the relative contribution of each PUC class to reach-scaled N biomass (above) and the 15 N storage into biomass (below). Percentages in vertical arrows reflect the relative contribution of each PUC class to assimilatory NH 4 + uptake from 15 N tracer into biomass (U ASSIMÀPUC ). N storage percentage refers to the total 15 N retained in biomass with respect to total added. Uncertainty for each parameter is shown in brackets as the standard deviation calculated based on the Monte Carlo method (see main text for details).

NO 3
À carried by some specialized microbes, can comprise a substantial portion of NH 4 + uptake in some streams (Peterson et al. 2001;Ribot et al. 2017). However, we observed a low contribution of nitrification to U (≤ 25%) in the two study stream reaches, especially in the wetland state (Table 2). Low nitrification rates may be explained by the high light availability reaching the stream channel (for pre-wetland only) which can inhibit nitrifiers (Merbt et al. 2012) or because photoautotrophic and heterotrophic microorganisms outcompete nitrifiers for NH 4 + under high environmental C : N ratios such as those expected in an N-limited stream (Verhagen and Laanbroek 1991;Strauss and Lamberti 2000).
In streams, it is common that assimilatory uptake accounts for the largest fraction of U (Peterson et al. 2001;Ribot et al. 2017;Tank et al. 2018). Our results support these previous findings (Table 2). However, it is worth noting that assimilatory uptake estimated from the sum of PUC-specific uptakes (U ASSIMÀPUC ) was considerably higher than that estimated from N uptake in the water column (U ASSIMÀWATER ) in both study reaches. This result highlights the importance of primary producers in N cycling in this N-limited, open-channel stream. Nevertheless, the mismatch between the U ASSIM estimates may be potentially explained by two methodological issues: (1) the most active parts of each PUC (i.e., hotspots of NH 4 + uptake) were sampled disproportionately compared to other parts of these compartments (Peipoch et al. 2016); or (2) the spike added in water samples to bring the N mass to the minimum required for the mass spectrometric analysis may have underestimated 15 N in the water. Dilution of 15 N signal in the sample may result in an underestimation of the 15 NH 4 + , with a minor effect on water-column estimations (i.e., based on the slope of longitudinal 15 N fluxes) but a greater impact on water-to-PUC 15 N ratio used to calculate assimilatory uptake. Alternatively, reach-scale biogeochemical processes that may counterbalance NH 4 + uptake, such as mineralization, may lead to an underestimate of U from declines in water column fluxes of NH 4 + along the reaches (Tank et al. 2018). In spite of this, the observation that U was mostly driven by assimilatory uptake suggests that differences in N cycling between the two ecosystem states are likely related to the intrinsic metabolic and structural characteristics of the dominant PUCs, in this case, primary producers. Assimilatory N uptake by the most abundant PUCs in the pre-wetland state (i.e., C. glomerata, Zygnematales and periphyton) was clearly higher than assimilation by PUCs in the wetland state (i.e., vascular plants) (Fig. 2B). Furthermore, vascular plants showed lower N turnover rates; findings that align with the general trends observed at the reach scale (Fig. 4). These results may mechanistically explain the observed reach-scale differences in N cycling. In addition, establishment of vascular plants in the stream channel can also indirectly contribute to lower reach-scale N uptake by reducing the water exchange between the surface stream and the hyporheic zone, as reported in Heffernan et al. (2008).
Similarly, Dong et al. (2017) observed that restriction of surface-subsurface water exchange was likely a major effect of the ecosystem state change to wetland dominance that influenced longitudinal patterns of stream nutrient concentration. The hyporheic zone is considered a relevant in-stream compartment of N uptake (Jones et al. 1995;Hall et al. 2002); thus, reducing the water exchange with this bioreactive zone may decrease reach-scale N uptake. Nevertheless, both bromide concentrations and δ 15 NH 4 + measured at 20, 45, and 60 m downstream of the addition point during the tracer solute additions in the wetland reach were very similar in surface and subsurface samples (mean AE SE = 0.69 AE 0.009 and 0.69 AE 0.009 mg L À1 Br and 90.7 AE 14.1 and 83.9 AE 15.3 δ 15 NH 4 + , respectively), which indicates that there was vertical hydrological exchange to some extent. We also observed a disproportionate contribution of S. goodingii roots to N uptake with respect to their N biomass, in comparison to the aboveground biomass of S. americanus or T. domingensis ( Fig. 2A,B). These results suggest that the presence of riparian vegetation may substantially contribute to in-stream N uptake in the wetland state. This is in agreement with the findings observed in the pre-wetland state that show the relevance of riparian trees for N retention in this stream (Schade et al. 2005). We also found that the shift from algae to vascular plants as dominant primary producers in the study stream may not only affect reach-scale N uptake, but also the time that the assimilated N is stored in biomass before it is released back to the water column as inorganic N (i.e., reach-scale N turnover) ( Table 2). In general, vascular plants live longer than diatoms and filamentous green algae, causing lower N turnover rates in the former than in the latter (Peipoch et al. 2014). The lower N turnover rate in vascular plants may explain why temporary N retention in wetland state biomass was four times longer when compared to the pre-wetland state reach (i.e., 142 and 33 d, respectively). Differences between states in the time over which N is immobilized in biomass may have important implications for N export from the watershed, especially considering the hydrologic regime of desert streams, which is characterized by extreme events. The frequent and severe flash floods and extended dry periods (Fisher et al. 1982;Stanley et al. 1997;Bunn et al. 2006) may affect the capacity of N retention in the two ecosystem states differently. Floods easily mobilize organic material grown and stored in the streambed sediments, resulting in high N export to downstream reaches (Grimm and Fisher 1989). Nevertheless, previous studies indicate that primary producers in desert streams under prewetland state conditions are highly resilient, and rapidly recover their biomass and nutrient uptake capacity after floods (Grimm and Fisher 1989;Marti et al. 1997). Wetland vascular plants, on the other hand, may be better adapted to withstand floods than filamentous algae and biofilm, due to their anchored root systems (Lytle and Poff 2004;Zhu et al. 2012). Furthermore, streambed establishment of vascular plants produces a suite of additional changes that reinforce their stability (Dent et al. 2002;Heffernan 2008). Therefore, higher resistance of vascular plants to floods in the wetland state may result in lower export of assimilated N to downstream reaches in desert streams during storm events. On the other hand, the specific plant location of stored N within each vascular plant species may also have consequences for N export during storms and plant senescence (i.e., loss of aboveground biomass at the end of the reproductive period) (Fig. 3; Vernescu et al. 2005).
In conclusion, our results indicate that the two alternative stream ecosystem states, characterized by different primary producers, cycle N differently in the study stream (Fig. 5). More specifically, we showed that N spiraling in this N-limited stream changed from a dominance of intense and fast recycling to a prevalence of N retention in biomass of primary producers. The pre-wetland state, dominated by diatoms and filamentous green algae, shows higher capacity to take up and store N than the wetland state, which was dominated by emergent vascular plants (Fig. 2). However, the wetland state showed lower turnover rates, indicating that assimilated N is retained in the stream for longer time as organic N in biomass (Fig. 4). These reach-scale contrasting patterns are mostly explained by the differences in assimilatory uptake by the dominant primary producers present in each stream state and the environmental hydromorphological changes that they determine or induce. Differences in N cycling between ecosystem states may have further implications within the context of the highly variable hydrologic regime of desert streams (i.e., alternation of floods and dry events). Establishment of vascular plants in the wetland lowered N uptake capacity with respect to algal dominance in the pre-wetland state; however, higher resistance by vascular plants to episodic high floods and drying may result in a more stable N uptake over a wider range of flow conditions vs. a more cyclic and variable behavior in the pre-wetland state. In addition, the wetland state may reduce downstream N export during high-flow conditions due to the higher temporal N retention in biomass.