Sample sites In our study, we used water samples from four small streams in the catchment of the Bode River in Central Germany, which is part of the TERENO environmental observatory network (Wollschläger et al. 2017 ). Water samples were taken from the 1st order tributaries Getel, Sauerbach, Asse, and Ströbecker Fließ, with surface hydrological catchment areas of 6.8, 2.0, 30.2, and 17.4 km2 , respectively. The catchments of the sampled streams were mainly or partially used for arable farming and urban land use. Arable land covered 85.3, 28.0, 87.9, and 88.9% of the catchment area, and urban land covered 14.6, 9.8, 5.0, and 5.7% of the stream site catchments for Getel, Sauerbach, Asse, and Ströbecker Fließ, respectively. We extracted the data from the 2018 Corine land cover map, and the digital elevation model of the shuttle radar topographic mission was used to determine the hydrological catchment area in QGIS, version 3.12.3.
Experiment preparation We used a DOC solution produced from freshly fallen Alder leaves (Alnus glutinosa (L.) Gaert.). The leaves were rinsed, dried at 60 °C and leached for three days in ultrapure water (Milli-Q synthesis A10) in the dark at 4 °C. The leachate was filtered first through a pre-combusted (500 °C, 4 h) glass-fiber filter (Whatman GF/A) and then through a 0.2 µm pore size filter (Nalgene SFCA, surfactant-free cellulose acetate). Aliquots were stored frozen (− 22 °C) (Attermeyer et al. 2015 ).
For each stream site, we sampled ten liters of stream water for the incubation and one liter of stream water for the bacterial inoculum into acid-washed polyethylene containers. We took samples on 27th July 2018 from Getel (gt) and Sauerbach (sb) and on 13th August 2018 from Asse (as) and Ströbecker Fließ (sf). Water samples were stored at 4 °C and transported to the laboratory for immediate filtration. The bacterial inoculum samples were filtered with GF/C filters (Whatman) with an approximate cutoff of 1.4 μm to remove algae and zooplankton as much as possible, without losing bacteria. These filters were rinsed beforehand with 1 L of deionized water (Milli-Q®, Merck, Darmstadt, Germany) to remove potential residual production-related DOC from the filters (Yoro et al. 1999 ). The water samples for incubation were filtered through pre-rinsed 0.22 μm polycarbonate membrane filters (Millipak 60 Gamma Gold Capsule, Durapore membrane, Merck) to remove any bacteria. We took an aliquot from each stream water sample for DOC and nutrient analysis, and started the experiments within two days of the sampling.
Experiment setup We first prepared a stock DOC solution extracted from Alder leaves with concentrations of DOC, nitrate-N, nitrite-N, ammonium-N, DON, SRP, and DOP of 3780 mg L−1 , < 0.42 mg L−1 , < 0.06 mg L−1 , 11.0 mg L−1 , 4.3 mg L−1 , 20.2 mg L−1 , and 0.3 mg L−1 respectively (see below for measurement and determination of solutes).
For each of the four sites (as, gt, sb, sf), we then mixed stream water, bacterial inoculum (5% vol.), and the DOC solution at a fixed volume. To reach six different DOC:nitrate-N and DOC:SRP start ratios for each of the four stream sites (as, gt, sb, sf), we added Alder DOC stock solution at different dilutions, using deionized water for the dilution. For each of the stream sites, the final solutions contained 0.02% (equivalent to 0.8 mg L−1 ), 0.2% (7.6 mg L−1 ), 0.4% (15.2 mg L−1 ), 1.0% (38.2 mg L−1 ), 2.0% (77.1 mg L−1 ) or 4% (157.5 mg L−1 ) of the original concentration of the Alder DOC stock solution. With that and the different nutrient concentrations of the stream water samples, we reached a range of different DOC:nitrate-N (0.3 to 694.6) and DOC:SRP (22.3 to 1026.2) molar ratios.
We established one start solution for all replicates of each DOC solution treatment and stream. We split this start solution into aliquots of 150 mL per replicate (three replicates per stream-treatment combination). We ran the experiment in 150 mL brown-glass vials, washed thoroughly with distilled water before the experiment. The 150 mL brown-glass vials were constantly shaken at 89 revolutions per minute in the dark at 18 °C. In the experiment, we measured DOC and nutrient concentrations at the start of the experiment, and after 60 h, a time at which previous, similar experiments have proven maximum bacterial reaction to Alder leaf leachate DOC (Graeber et al. 2018 ).
In an additional control treatment, we measured how the inoculum influenced the experimental nutrient and DOC concentrations. We measured inoculum concentrations of DOC and nutrients without Alder leaf leachate for each stream at both sampling times with three replicates. To separate inoculum effects from the treatment effects, we subtracted the DOC or nutrient concentration means of the inoculum control from the values of the respective stream-date combination of each treatment.
Solute sample analysis After sampling, we filtered all solute samples with pre-rinsed (1 L DI water) 0.22 μm polycarbonate membrane filters (Membrapure, Gelman). Subsequently, we analyzed the concentrations of ammonium-N, nitrite-N, nitrate-N, and SRP by segmented-flow analysis (Skalar) photometrically, with quantification limits of 0.011, 0.006, 0.042, and 0.003 mg L−1 , respectively. We measured DOC concentration as CO2 in the near infrared after acidification stripping out the DIC by high-temperature catalytic oxidation (DIMATOC 2000 from DIMATEC, Germany), with a quantification limit of 0.50 mg L−1 .
We determined total dissolved N (TDN) after persulfate oxidation photometrically after reduction as nitrite-N complex in the segment-flow analysis with a quantification limit of 0.042 mg L−1 . Here, we also determined total dissolved phosphorus (TDP) after oxidation of all P containing compounds to SRP as organo-phosphorous complex photometrically (Hach DR 5000) with a quantification limit of 0.006 mg L−1 .
We measured DOM composition by EEMs (Horiba Aqualog, USA) in 1-cm Quartz glass cuvettes. We used an excitation range from 255 to 600 nm, and an emission range from 240 to 621 nm with an integration time of 0.72 s, medium CCD gain, an excitation increment of 5 nm, an emission increment of 0.82 nm, and an excitation slit width of 10 nm. With the same instrument, we measured the light absorbance from 255 to 600 nm to correct the inner-filter effect (Kothawala et al. 2013 ; Murphy et al. 2013 ). The absorbance was always below 1 cm−1 at 240 nm, a range at which the inner-filter effect can be efficiently removed (Kothawala et al. 2013 ).
Data processing We conducted all data processing, PARAFAC modeling, statistics, and data plot building in R, version 4.0.2 (R Core Team 2020 ).
For solute concentrations below the quantification limit (see above for quantification limits), we randomly chose one number from 0 to the respective quantification limit, based on a uniform distribution (function runif in R).
We calculated DON as TDN—(nitrate-N + ammonium-N + nitrite-N) and DOP as TDP—SRP. The indirect determination of DON and DOP can result in considerable DON and DOP concentration uncertainty (as exemplified for DON, Graeber et al. 2012a ). We used a DIN:TDN versus DON diagnostic plot or SRP:TDP versus DOP diagnostic plot to assess this uncertainty. We based this approach on Graeber et al. (2012a ), where it has been shown that at DIN:TDN < 0.6 DON concentrations can be trusted; that at DIN:TDN ratios = 0.6 to 0.8 systematic deviations of the estimated DON concentrations from the true DON concentrations can occur; that DIN:TDN ratios > 0.8 are indicative of high random and systematic uncertainty; and that DIN:TDN ratios > 1 indicate an overestimation of DIN, resulting in negative DON concentrations. We assume the same thresholds for DOP and SRP:TDN. Based on the diagnostic plots, we found that most water samples had considerably high DON concentrations and exhibited DIN:TDN ratios < 0.6 (Fig. S1). In contrast, DOP concentrations were generally low, or even negative, with most water samples having SRP:TDP ratios > 0.8. Based on these diagnostic plots, we assume DON concentrations to be relatively trustworthy, while this is not the case for DOP concentrations. Furthermore, we found considerable DON but not DOP in the Alder leaf leachate (see above) and stream water samples (Table 1 ). Based on the probable high uncertainty of determined DOP concentrations and its concentration irrelevance compared to SRP, we excluded DOP from any further data analysis and calculations of stoichiometric ratios. We also assumed that SRP represented the reactive P pool sufficiently due to the low apparent DOP concentrations.
Table 1 Dissolved macronutrient composition of samples at the start of the experiment, before additions of Alder leaf leachate To prepare the EEMs for the PARAFAC, we subtracted the blank water fluorescence and removed the inner-filter effect with the Horiba Aqualog software package (version 4.0) using the absorbance-based approach (Kothawala et al. 2013 ). We converted the EEM measurements to Raman units within the staRdom package (Pucher et al. 2019 ), based on the approach by Lawaetz and Stedmon (Lawaetz and Stedmon 2009 ). Here, we used an excitation wavelength of 350 nm and an emission range of 371 to 428 nm. To remove Rayleigh and Raman scatter, we used the scatter removal and interpolation approach in staRdom (Pucher et al. 2019 ), which usually results in the best extractable fluorophore spectra (Bahram et al. 2006 ). Here, we removed a width of 15 nm for the 1st order Rayleigh, and Raman scatter and a width of 20 nm for the 2nd order Rayleigh and Raman scatter. Subsequently, we interpolated the removed fluorescence data by spline interpolation (Pucher et al. 2019 ).
To chemometrically extract the DOC fluorophores from the EEMs, we established a PARAFAC model in the staRdom package (Pucher et al. 2019 ) based on the rules outlined in Pucher et al. (2019 ) and Murphy et al. (2013 ). In short, we developed the PARAFAC model based on split-half validation, spectral shapes of the fluorophores, assessment of sample and wavelength model leverage, residual fluorescence assessments, and global minimum model error assessment based on 200 random model starts. We attributed the fluorophore by spectral shape comparison to the OpenFluor database (Murphy et al. 2014 ) and other literature.
Statistical evaluations of the experiment To determine the relations between DOC:nitrate-N, DOC:reactive N, or DOC:SRP molar ratios and relative nitrate or SRP uptake for the first experiment, we used a logistic model derived from the conceptual models of Taylor & Townsend (Taylor and Townsend 2010 ) and Stutter et al. (2018 ):
$$U_{nitrateN\% } = \frac{C}{{1 + a \cdot e^{{ - k \cdot DOC:nitrate - N_{start} }} }}$$
(1)
Here, UnitrateN% is the uptake of nitrate-N relative to the start value of nitrate-N, and C, a and k are constants, where C indicates the maximum uptake. This form is analogous for SRP and reactive N. Hence, UnitrateN% is substituted with USRP% and DOC:nitrate-Nstart is substituted with DOC:SRPstart . We calculated the inflection point as log10(a)/k, which equals the steepest slope in logistic models. To assess the fit of the logistic model, we calculated the R2 based on the total sum of squares (TSS) and the sum of squared errors (SSE), as:
$$R^{2} = 1 - \frac{SSE}{{TSS}}$$
(2)
To link DOC composition and the amount of bioavailable bDOC or bioavailable bDON, we investigated the correlation between the uptake of DOC or DON and the uptake of the PARAFAC components. Here, a positive correlation indicates the dependence of DOC uptake on the uptake of a specific PARAFAC component, a negative correlation indicates the accumulation or production of a PARAFAC component during DOC processing, and no correlation indicates independence of DOC uptake from a given PARAFAC component. To construct and evaluate linear correlations between DOC uptake and a specific PARAFAC component, we used the summary. lm and lm() functions within R. The samples used for linear regression were independent of each other, and the residuals were homogeneously distributed; hence the assumptions of the linear models were valid. Furthermore, the assumptions of the logistic growth model were valid, as the model represented the distribution of the data well, and underlying samples were independent of each other.
If a specific PARAFAC component exhibited a consistent positive correlation with DOC or DON uptake for all stream waters, we deemed it representative of a bioavailable moiety of DOC or DON. Based on this assumption, we calculated the concentration of bDOC for each sample separately as:
$$bDOC = \frac{{\sum\nolimits_{i = 1}^{n} {Cb_{i} } }}{{\sum\nolimits_{i = 1}^{m} {Comp_{i} } }} \cdot DOC$$
(3)
Here Cb is the fluorescence (in R.U.) of a bioavailable PARAFAC component, n is the total number of bioavailable PARAFAC components, Comp is the fluorescence of every PARAFAC component, and m is the total number of PARAFAC components. This calculation assumes that the ratio of the fluorescence of the bioavailable components to the total fluorescence equals the ratio of the concentration of bioavailable DOC to total DOC. In other words, we assume that EEM-PARAFAC based bDOC represents all bDOC. Based on the bDOC calculated from EEM-PARAFAC, we subsequently assessed with a linear correlation how well it represents the DOC uptake during the experiment.
For bioavailable DON (bDON), we conducted an analogous calculation:
$$bDON = \frac{{\sum\nolimits_{i = 1}^{n} {Cb_{i} } }}{{\sum\nolimits_{i = 1}^{m} {Comp_{i} } }} \cdot DON$$
(4)
Similarly to bDOC, we assumed that the uptake of the PARAFAC components was representative of the bDON contribution to bulk DON. Again, we assessed this assumption in the same way as for bDOC; hence, how well the calculated bDON correlated to the measured DON uptake. Finally, we calculated the total reactive N pool as follows:
$$reactiveN = (nitrate - N + nitrate - N + ammonium - N) + bDON$$
(5)
Here, we assumed that all DIN is reactive and that the total dissolved reactive N pool is the sum of DIN and bDON. We further assumed that all reactive P equals SRP due to only very low, uncertain concentrations of DOP (see also explanation above in the Data Processing section). Finally, we determined the link between bDOC:reactive N or bDOC:SRP ratios and nitrate-N or SRP uptake with the same non-linear model approach used for bulk DOC above.
We constructed ternary plots for which we normalized the bDOC, reactive N and P calculated above with the Godwin-Cotner molar ratio (GCR) of 68C:14N:1P. The GCR is the median C:N:P of 137 isolates of lake planktonic bacteria grown under non-limiting conditions (Godwin and Cotner 2018 ). For the normalization, we used the bDOC, reactive N and P concentrations at the start of the experiment. For the GCR-normalized ternary plots, we calculated the contributions of molar concentrations (mM = mmol L−1 ) of bioavailable DOC, reactive N, and reactive P as follows:
$$\% bDOC = \frac{bDOC/68}{{(bDOC/68) + (reactive/14) + reactiveP}}$$
(6)
$$\% reactiveN = \frac{reactiveN/14}{{(bDOC/68) + (reactiveN/14) + reactiveP}}$$
(7)
$$\% reactiveP = \frac{reactiveP/14}{{(bDOC/68) + (reactiveN/14) + reactiveP}}$$
(8)