Pulse of dissolved organic matter alters reciprocal carbon subsidies between autotrophs and bacteria in stream food webs

Soils are currently leaching out dissolved organic matter (DOM) at an increasing pace due to climate and land use change or recovery from acidification. The implications for stream biogeochemistry and food webs remain largely unknown, notably the metabolic balance (biotic CO2 emissions) and carbon cycling between autotrophs and bacteria. We increased by 12% the flux of DOM in a stream for three weeks to mimic a pulse of natural DOM supply from soils rich in organic matter. We were able to track its fate into the food web through the use of a before and after control impact experimental design and the addition of DOM with a distinctive dC signature. We used whole-stream metabolism to quantify carbon fluxes. Both photosynthesis and heterotrophic respiration increased rapidly following C addition, but this was short lived, likely due to nutrient limitations. Carbon exchange between autotrophs and bacteria in the control stream accounted for about 49% of bacterial production and 37% of net primary production, under stable flow conditions. Net primary production relied partly (19% in the control) on natural allochthonous dissolved organic carbon via the CO2 produced by bacterial respiration, intermingling the green and brown webs. The preferential uptake of labile carbon by bacteria and excess bacterial CO2 relative to nutrients (N, P) for autotrophs shifted the reciprocal carbon exchange between bacteria and autotrophs to a predominantly one-way carbon flow from bacteria to autotrophs, increasing the C:N:P molar ratios of autotrophs, the latter likely to become less palatable to consumers. The bacterial response to sucrose addition shifted the metabolic balance toward heterotrophy increasing biotic CO2 emissions (+125%), shortened the average distance travelled by a molecule of organic matter ( 40%), and thus provided less organic matter of lower quality for downstream ecosystems. Even a small increase in labile dissolved organic matter supply due to climate and land use change could significantly alter in-stream carbon cycling, with large effects on stream food web and biogeochemistry in small streams draining catchments with soils rich in organic carbon.


INTRODUCTION
The global annual riverine flux of organic C (0.26-0.53 Pg C/yr) to the oceans is comparable to the annual C sequestration in soil (0.4 Pg C/yr), suggesting that terrestrially derived aquatic losses of organic C may contribute to regulating changes in soil organic carbon storage (Dawson 2013). Soils are currently leaching out their organic matter at faster rates to aquatic ecosystems due to climate and land use change, or recovery from acidification (e.g., Freeman et al. 2004, Monteith et al. 2007, Drake et al. 2015. The fate of this organic matter in riverine systems remains poorly understood at the global scale, notably CO 2 emissions resulting from the mineralization of the organic matter (Drake et al. 2018). The role of terrestrial organic matter is also highly debated regarding its contribution to aquatic food webs (e.g., Cole 2013, Brett et al. 2017, Tanentzap et al. 2017, Drake et al. 2018). This may partly stem from an evolving understanding of the nature and structure of organic matter delivered to aquatic ecosystems.
The persistence (recalcitrance) of the organic matter in soils is more an ecosystem property than a characteristic of the molecular structure (aromaticity) of the organic matter (Schmidt et al. 2011, Dungait et al. 2012. The existence of humic macromolecules in soils has not been verified by direct measurements, suggesting the presence of simple biomolecules (Schmidt et al. 2011). Old carbon can also hold a large proportion of labile molecules such as sugars or acetate (Schmidt et al. 2011, Drake et al. 2015. The emerging understanding is that degradability of organic matter is more constrained by the biotic and abiotic environment (Schmidt et al. 2011, Dungait et al. 2012, such as C:N:P ecological stoichiometry (Evans et al. 2017, Kominoski et al. 2018, change in redox potential (e.g., Boye et al. 2017), and hydrological connectivity (e.g., Wilson et al. 2013, Marin-Spiotta et al. 2014, Demars 2019. This can explain the rapid turnover of allochthonous organic matter in aquatic ecosystems once it leaves the soil matrix (e.g., Marin-Spiotta et al. 2014, Drake et al. 2015, Demars 2019.
In streams, numerous dissolved organic matter (DOM) additions with simple compounds (from trace amount to 20 mg C/L) have repeatedly shown DOM use by bacteria and its transfer through the food chain (e.g., Warren et al. 1964, Hall 1995, Hall and Meyer 1998, Parkyn et al. 2005, Wilcox et al. 2005, Augspurger et al. 2008, Collins et al. 2016. Fewer studies used leaf leachate material (e.g., Cummins et al. 1972, Friberg and Winterbourn 1996, Wiegner et al. 2005, 2015, which are known to be processed at a slower rate than simple organic compounds (Mineau et al. 2016). Many studies have also traced the flux of autochthonous DOM through to bacteria (e.g., Neely and Wetzel 1995, Lyon and Ziegler 2009, Risse-Buhl et al. 2012, Kuehn et al. 2014, Hotchkiss and Hall 2015. The potential benefit of bacterial CO 2 for primary producers has been hypothesized to explain a small increase (16-20%) in gross primary production following dissolved organic carbon addition and associated increase in bacterial activities (Robbins et al. 2017). This is a reasonable assumption (e.g., Redfield 1958, Butenschoen et al. 2016, Hasler et al. 2016 suggesting stream CO 2 limitation (at least within the boundary layer of photosynthesizing organisms and within biofilm, e.g., Hartley et al. 1996). Yet, the reciprocal carbon subsidies between autotrophs and bacteria in stream food webs have not been studied, to our knowledge, either empirically or theoretically. It raises the possibility that the carbon of the primary producers may be partly derived from allochthonous organic matter processed by the bacteria and intermingles the green and brown webs, similarly to the use of autotrophic DOM by bacteria.
The interactions between autotrophs and bacteria are, however, difficult to study (Amin et al. 2015), because primary producers, decomposers, and organic matter (allochthonous and autochthonous) are intricately connected both in benthic biofilms (Kamjunke et al. 2015, Battin et al. 2016) and pelagic aggregates (Grossart 2010). At the ecosystem scale, stream metabolism showed respiration to be generally tightly correlated to gross primary production across the globe, with respiration generally exceeding photosynthesis (Demars et al. 2016). Ecosystem respiration and primary production also tend to mirror each other throughout the year, although changes in canopy cover, hydrological events and pulses of allochthonous organic matter are known to affect this reciprocity (e.g., Uehlinger 2006, Bernhardt et al. 2018, Ulseth et al. 2018. The extent to which an increase in CO 2 availability from bacterial respiration can boost primary production may be constrained by nutrient (N, P) availability ( Fig. 1). Since natural DOM is generally poor in nutrients (Stutter et al. 2013) relative to bacterial tissue (Cotner et al. 2010), bacteria may shift from being nutrient regenerators to nutrient retainers (Cotner et al. 2010). Excess of CO 2 availability relative to N and P is expected to increase the C:N and C:P of autotrophs in nutrientpoor freshwater ecosystems (Hessen et al. 2004), possibly limiting their growth rate (Kahlert 1998, Hillebrand andSommer 1999) and eventually reduce autotrophic C subsidies to bacteria. When respiration and photosynthesis are in synchrony, nutrient cycling rate increases with metabolic activity (Huryn et al. 2014), this is when bacteria depend strongly on autotrophic C and act as regenerators of N and P. Whether this holds during periods of asynchrony remains uncertain due to changes in net retention by autotrophs and heterotrophs .
Here we test how a small addition of labile organic carbon (sucrose, C 12 H 22 O 11 ) affects the strength and dynamics of the autotroph-bacteria reciprocal carbon FIG. 1. Reciprocal carbon subsidies between autotrophs and bacteria in nutrient-poor stream food webs under contrasting supply of soil derived dissolved organic matter (allochthonous DOM). Arrows depict transfer of matter between the parts of the ecosystem. The size of the arrows is proportional to a priori hypotheses. subsidies in phosphorus-poor streams draining soils rich in organic carbon. The time scale of the experiment (three weeks) corresponded to a pulse of soil derived DOM during a period of hydrological connectivity (wet soils) known to drive stream heterotrophic respiration in the studied streams (Demars 2019). We expected sucrose addition to (1) weaken reciprocal carbon exchanges between autotrophs and bacteria ( Fig. 1); (2) shift the metabolic balance toward heterotrophy and thus increase biotic CO 2 emissions; (3) increase C : nutrient ratios in autotrophs and alter whole ecosystem nutrient (N, P) cycling rates; (4) reduce the organic carbon uptake length and speed up organic carbon uptake velocity; (5) reduce light use efficiency by primary production; and (6) increase organic carbon use efficiency by bacteria.
More fundamentally, we show how the overlooked carbon reciprocal transfers in streams affect food web ecology and biogeochemistry in the face of environmental changes.

Strategy
The core of the study is based on a flow food web approach (sensu Marcarelli et al. 2011) and a whole-ecosystem before-after-control-impact (BACI) experimental design, adding sucrose with a distinctive d 13 C signature. We first present the study area and the nutrient (C, N, P) context. Second, we show how the food web was derived from whole-stream metabolism and carbon source apportionment (food web links). Third, we quantify C fluxes: calculations of whole-ecosystem CO 2 emissions (net ecosystem production, autotrophic and heterotrophic respiration), whole-stream production (net primary production and bacterial production), carbon and energy supplies (sucrose, natural allochthonous DOM, photosynthetic active radiation) and whole-ecosystem carbon cycling.

Study area
We studied two heather moorland catchments with soils rich in organic carbon, within the Glensaugh research station of the James Hutton Institute in northeast Scotland (2°33 0 W, 57°55 0 N; Fig. 2). Soil water, groundwater, and stream water have very low soluble reactive phosphorus concentrations (5 lg P/L; Demars 2019) independently of discharge variability. The streams were about 0.8-1.0 m wide in the studied sections and their channels significantly undercut the banks by 30-46% of stream width. Brown trout (Salmo trutta fario, Salmonidae) was present in both streams. The management of the land includes regular heather burning (10-12% of surface area yearly target) for hill farming: mixed grazing of sheep and cattle. For further information see Demars (2019).

The control stream
The control stream (Birnie Burn) is part of the longterm monitoring of the UK Environmental Change Network (ECN, data available online). 6 There is monitoring of soil temperature and moisture on the hillslope of the Birnie Burn at 275 m elevation (Cooper et al. 2007). Volumetric soil moisture content is recorded every 30 minutes at 10 and 45 cm depth, corresponding respectively to the base of the O (organic layer) and B (subsoil) horizons of the humus iron podzol present. The stream is equipped with a flume for continuous monitoring of discharge (catchment area 0.76 km 2 ) and dip water samples are collected weekly. This stream showed substantial increase in annual flow-weighted mean concentrations of stream-water dissolved organic carbon (DOC, +0.28 mg CÁL À1 Áyr À1 during 1994-2007Stutter et al. 2011). Water samples were analysed for DOC, pH, nutrients (N, P), and major ions. See Cooper et al. (2007), Stutter et al. (2012), and Demars (2019) for further details.

The paired stream
The control stream was paired with a neighboring stream (Cairn Burn) in 2005, also part of a long-term monitoring scheme with samples collected every week or two for stream water quality. In the late 1970s and early 1980s, two areas covering 33 ha were improved (reseeded, limed, and fertilized) as part of sheep grazing experiments (Hill Farming Research Organisation 1983). The added facilities at the Cairn Burn (catchment area 0.9 km 2 ) included a calibrated flume, water level, water FIG. 2. Paired stream experiment at Glensaugh research station. Birnie Burn is the control stream and Cairn Burn the manipulated stream with DOM addition indicated by the arrow. The Cairn Burn also has a control upstream of the treatment reach. The symbols refer to flumes (open triangles), dissolved oxygen stations (solid circles), and soil moisture instrumentation (solid squares). The 50-m elevation contour lines are indicated. The catchment area is 0.99 km 2 (0.90 km 2 at the flume) for Cairn Burn and 0.76 km 2 for Birnie Burn. Inset shows the location of Glensaugh in Scotland. electric conductivity, water and air temperature, and barometric pressure. Data were recorded every 5 minutes (CR109 data logger; Campbell Scientific, Logan, Utah, USA). Data logger, battery, and barometric pressure were housed in a weather-resistant enclosure. Photosynthetic active radiations (PAR) were also recorded in air, 1 m above ground, at the same time intervals (LICOR, Lincoln, Nebraska, USA). For more information, see Demars (2019).
Terrestrial DOM: main source of organic carbon DOM was the dominant flux of organic carbon (98%) under stable flows with average concentrations of 9.3 AE 1.7 mg C/L in the two studied streams (Demars 2019). This DOM was of terrestrial origin as shown by d 13 C analyses of the natural DOM against terrestrial and aquatic plant material (Stutter et al. 2013). The median molar C:N:P stoichiometry of the DOM was 3201:103:1, with values ranging between 978:38:1 to 12,013:282:1 (Stutter et al. 2013). Chlorophyll a concentration in the water column was extremely low (<1 lg/L; Stutter et al. 2013). The pool of particulate organic carbon in the sediment is very small (Demars 2019), and coarse particulate organic carbon was less than 10 g C/m 2 .

DOM addition
A carboy was refilled every 2 d with 6 kg of sucrose (granulated pure cane sugar, Tate & Lyle PLC, London, UK) dissolved in over 60 L of stream water filtered through muslin square in a large funnel. The carboy was set as a Mariotte bottle to ensure a constant dripping rate of 22 mL/minute lasting 48 h (15 mg C/s). The ventilation tube was netted at the top to avoid insect contamination. The dripping rate was kept constant over the 22 d of sucrose addition (23 August-14 September 2007) and was initially set to increase stream DOC concentration by about 0.5 mg C/L at 30 L/s. Samples were collected in washed bottles and filtered on site with prewashed filters (0.45-lm Millipore PVDF membrane filter; MilliporeSigma Filtration, now part of Merck, Darmstadt, Germany). DOC was determined within 48 h of collection with a Skalar San++ continuous flow analyzer (Breda, The Netherland), using potassium hydrogen phthalate as standards and sodium benzoate for quality controls. The detection limit was 0.1 mg C/L.

Nutrient cycling studies
Nutrient cycling studies were run in the control and manipulated streams before and during sucrose addition to test whether the expected increase in nutrient demand may be satisfied by faster cycling rate of N and P (hypothesis 3). Nutrient cycling rates were derived from continuous in situ nutrient addition experiments where a conservative tracer is also included (Stream Solute Workshop 1990, Demars 2008. This method tends to overestimate the nutrient uptake length S w , average distance travelled by a nutrient molecule in the water column before river bed uptake. This is due to the addition of nutrient compared to isotopic tracer studies (e.g., Mulholland et al. 2002). However, preliminary tests in the Cairn Burn showed that the bias can be kept small (10-15%) with small nutrient additions (Demars 2008). Nitrate (as KNO 3 ) and phosphate (as KH 2 PO 4 ) were continuously added together with NaCl as conservative tracer (cf Schade et al. 2011). The target molar stoichiometric ratio for the addition was N:P % 20, similar to the periphyton optimal ratio for growth (Hillebrand and Sommer 1999) but differential uptake of N and P in the mixing zone led to slightly higher average added molar N:P ratios (34 and 70 in the control and treatment reach, respectively). This was similar to the average background molar ratio of the treatment reach at the time of the nutrient additions (N: P = 58), but somewhat lower than the control (N: P = 206). When the plateau phase was reached, water samples were collected at about 10-m intervals along the reach and filtered on site (pre-washed 0.45 lm Millipore PVDF membrane filter; see Demars 2008). The samples were kept cool at 4-10°C. The nutrients (PO 4 and NO 3 ) were determined within 48 h by colorimetry using a Skalar San++ continuous flow analyzer (Breda, The Netherland) and chloride by ion chromatography (Dionex DX600, Sunnyvale, California, USA). The limits of detections were 0.001 for NO 3 and PO 4 and 0.003 mg/L for Cl. In order to provide a more comparable indicator of nutrient cycling for different hydrological conditions, the uptake velocity v f was also calculated as follows: v f = (u 9 z)/S w , with u being average water velocity and z average depth. Short uptake lengths and fast uptake velocities indicate fast cycling rates (high exchange rates between water and benthos).

Flow food web
We determined whole-stream metabolism to test the effect of DOM addition on net ecosystem production (NEP, corresponding to biotic CO 2 emissions; hypothesis 2). This approach also gives gross primary production (GPP) and ecosystem respiration (ER) from which net primary production (NPP) and heterotrophic production (HP) can be derived knowing the carbon use efficiency and bacterial growth efficiency, respectively. This, combined with a knowledge of C source apportionment, allows to build the flow food web from which the C reciprocal subsidy hypothesis between autotrophs and bacteria can be tested.

Whole-stream metabolism
Whole-stream metabolism was estimated by the open channel two-station diel oxygen method of Odum (1956) modified by Demars et al. 2011, 2015and Demars (2019. Many tracer studies (using NaCl and propane) were carried out as detailed in Demars et al. (2011) to estimate lateral inflows, mean travel time, and reaeration coefficient as a function of discharge within the range of stable flows (up to 32 L/s). The relationships with discharge in our three stream reaches were very strong (Appendix S1: Fig. S1, Demars 2019) allowing accurate parameterization of metabolism calculations under varying flow conditions, similarly to previous studies (e.g., , Beaulieu et al. 2013. The high oxygen reaeration coefficient of those streams (0.05-0.24 per minute) required very accurate dissolved O 2 data. Oxygen concentrations were measured with optic sensors fitted on multiparameter sondes TROLL9500 Professional (In-Situ Inc., Fort Collins, Colorado, USA) and Universal Controller Sc100 (Hach Lange GMBF, D€ usseldorf, Germany), the latter powered with two 12-V DC (75 mA) car batteries per sensor kept charged with two 20 W solar panels (SP20 Campbell Scientific). The sensors were calibrated to within 1% dissolved oxygen saturation. Four sondes were deployed in the Cairn Burn at 0, 84, 138, and 212 m upstream of the flume to include an extra control reach (138-212 m) upstream of the manipulated reach (0-84 m). Another two sondes were set in the control stream Birnie Burn at an 88-m interval (60-148 m upstream of the ECN flume). The distances between oxygen stations corresponded to 80-90% of the oxygen sensor footprints (3u/k 2 ), with u/k 2 entirely independent of discharge (R 2 = 0.0005), which allowed the manipulated reach to be independent of the control reach. The DOM injection point was 28 m upstream of the top station of the manipulated reach, and this distance corresponded to 69% of the oxygen sensor footprint of the top station. All sondes were deployed from May to October 2007, logging at 5-minute intervals.
The net metabolism was only calculated for stable flow conditions (3-32 L/s), as follows (Demars 2019): with NEP t , net ecosystem production at time t (g O 2 Ám À2 Áminute À1 ); C AV , average dissolved oxygen (g O 2 /m 3 ) of the two stations at time t + Dt and t (minutes); Dt, time interval (minutes); k 2 , oxygen exchange coefficient (per minute); C s , saturated oxygen concentration (g O 2 /m 3 ); h, the proportion of lateral inflows (Q g /Q); z, average stream depth (m); and C g , oxygen concentration in lateral inflows (g O 2 /m 3 ). The latter was calculated as follows (from base flow analysis of stream hydrographs): with Q, discharge; and a and b, constants; permitting to correct for base flow (first term of the equation) and soil water (second term of the equation) lateral inflows; see Demars (2019). The proportion (mean AE SE) of total lateral inflows relative to discharge (Q g /Q) was 10.7% AE 0.6%, 6.6% AE 0.5%, and 2.3% AE 0.4% for the Birnie Burn control, Cairn Burn control, and treated reach, respectively, independently of discharge in the range 3.8-32.5 L/s (stable flows). All calculations were run in Excel (Microsoft, Redmond, Washington, USA) using a preformatted spreadsheet (Demars 2019). The overall uncertainties in daily stream metabolism, including cross-calibration errors, individual parameter uncertainties, spatial heterogeneity (through the average of diel O 2 curves), and correction for lateral inflows, were propagated through all the calculations using Monte Carlo simulations (Demars 2019). The corrections for lateral inflows amounted to about 6% of ER for the treated reach (Cairn Burn), 19% and 16% in the control reaches, Cairn Burn and Birnie Burn, respectively.

Identification of carbon sources and pathways
Macrophytes.-The percentage cover of bryophytes and filamentous green algae was measured with a ruler across transects taken every 2 m along the stream reaches. Young shoots of filamentous green algae and bryophytes were collected by hand along both studied reaches before and after sucrose addition. All samples were freeze dried and milled prior to analyses for C, N, d 13 C, and d 15 N. The main source of inorganic carbon for primary producers was assumed to be CO 2 because of the low alkalinity (remaining below 0.5 meq HCO 3 /L under low flows). The fractionation factor for CO 2 assimilation into macrophyte tissue is known to vary with pCO 2 and growth rate, and was set at À25.5& AE 3.5& within the range of Rubisco forms IA and IB in the absence of carbon concentrating mechanism and transport limitation (22-29&, e.g.; Raven et al. 1994, McNevin et al. 2007, Boller et al. 2015. Net primary production was assumed to be driven by filamentous green algae and biofilm autotrophs. Bryophytes were not used to calculate the flow food webs. Inorganic carbon.-The d 13 C of dissolved CO 2 was estimated indirectly under low flows using the fractionation factor of Rubisco À25.5& AE 3.5& and d 13 C of bryophytes (strict CO 2 user and no CO 2 transport limitation). In our study, the average d 13 C of bryophytes was À36.4&, and assuming the above fractionation coefficient of À25.5&, Glensaugh d 13 C of dissolved CO 2 would be À10.9&. This is similar to the d 13 C of DIC (6-14& under low flows; Waldron et al. 2007) and d 13 C of bryophytes (À33.3&; Palmer et al. 2001) reported for the Brocky Burn (a stream running on the other side of the hill). The proof of concept comes from another Scottish stream (Dighty Burn), where Raven et al. (1994) reported the d 13 C of dissolved CO 2 as À14.7& and bryophytes as À39.2&, suggesting a fractionation coefficient of À24.5& by difference. In our calculations under low flow conditions, we therefore assumed d 13 C of dissolved CO 2 as À11& AE 3&.
To quantify the reciprocal subsidies between autotrophs and bacteria, it remained to partition the overall stable isotope signature of stream dissolved CO 2 into the allochthonous (groundwater, soil water and atmospheric exchange) and autochthonous (respiration by heterotrophs and autotrophs) sources. The allochthonous signature, d 13 C CO2-allochthonous , can be deduced from rearranging the following three-ended mixing model: where F CO2 represents CO 2 fluxes (g CÁm À2 Ád À1 , with all fluxes expressed as positive values) and d 13 C the isotope signature (&) of the different sources of CO 2 . We averaged the d 13 C of the autotrophs (filamentous green algae and biofilm primary producers). The d 13 C CO2-allochthonous was only calculated for the control stream under low stable flows and assumed to apply to both streams. Uncertainties were propagated in quadrature using standard deviation dx for sums, and relative uncertainties dx/x for the division.
Sucrose.-The d 13 C of sucrose from sugar cane is similar to that of dissolved CO 2 (À12& AE 1&; Jahren et al. 2006, Augspurger et al. 2008, Kankaala et al. 2010, de Castro et al. 2016), but sucrose uptake by autotrophs was assumed to be without isotopic discrimination (Wright and Hobbie 1966). The proportion of carbon derived from added sucrose (F S ) in organic matter, primary producers and bacteria was calculated from their d 13 C in the control (C) and treatment (T) reaches, before (subscript B) and after (subscript A) sucrose addition as follows: with all uncertainties propagated in quadrature using standard deviation dx for sums, and relative uncertainties dx/x for the division. The standard error of the mean was calculated as SEM ¼ dx= ffiffi ffi n p with n average number of samples in C B , C A , T B , T A .
Allochthonous organic carbon.-The d 13 C of the DOM (mean AE SD) was available from a previous study from the same catchment and showed it was of terrestrial origin, i.e., not autochthonous (d 13 C = À28.5& AE 0.3&; Stutter et al. 2013). Coarse particulate organic matter (CPOM) was also collected by hand along both studied reaches before and after sucrose addition. Since there was hardly any difference in d 13 C between DOM and CPOM (d 13 C = À27.4& AE 0.7&, Appendix S1: Table S1), we used the d 13 C of CPOM determined in this study as the signature for allochthonous organic carbon.
Periphyton.-Periphyton (or biofilm) samples represent a mixture of primary producers (algae and cyanobacteria), bacteria and fine particulate organic matter. The samples were collected before and at the end of the sucrose addition from the flumes and stones with a toothbrush, funnel, and bottle. All samples were freezedried and milled.
We also placed six pairs (with/without petroleum jelly) of unglazed ceramic tiles (10 9 10 cm) fixed on bricks and deployed along the studied reaches in both streams three weeks before the start of the manipulation. Vaseline was applied around half the tiles to prevent grazing by invertebrates. After three weeks, there was hardly any growth on the tiles, and so the tiles were left in the stream until thze end of the manipulation. One brick in the control stream was lost. At the end of the experiment, the tiles were frozen at À20°C, later freeze-dried, and the biofilm was scraped with a razor blade. Since there was little biomass per tile (about 1 g C/m 2 ), the biofilm was pooled by stream and grazer treatments (leaving two samples per stream). Very little grazing activity was observed on the tiles during the six weeks and unsurprisingly no difference in biofilm dry mass emerged due to grazer exclusion (paired t tests on lntransformed mass; Birnie, t 4 , P = 0.13; Cairn, t 5 , P = 0.26).
Phospholipid fatty acids (PLFAs) were extracted from the biofilm samples from the tiles and derivatized to their methyl esters (FAMES) following the procedure of Frosteg ard et al. (1993) using the modified extraction method of Bligh and Dyer (1959), as detailed in Certini et al. (2004). Quantification and d 13 C values of the PLFAs were both determined by Gas Chromatography-Combustion-Isotope Ratio Mass Spectrometry (GC-C-IRMS) as described by Main et al. (2015), and averaged for each stream (d 13 C, Appendix S1: Table S2). Only PLFAs up to 19 carbon chain length were determined (excluding some long-chain essential polyunsaturated fatty acids, e.g., Muller-Navarra et al. 2000, Gladyshev et al. 2011. Bacterial carbon.-The bacterial PLFAs were identified as those most affected by sucrose addition in the treated reach and information derived from the literature (Appendix S1: Table S2). This also allowed us to determine the d 13 C of bacteria in the control reach. We assumed a fractionation factor of À3& for the d 13 C of bacterial fatty acids relative to bulk tissue samples (Hayes 2001). This is within the range of observed values in other studies (Boschker andMiddelburg 2002, Bec et al. 2011). We used the same fractionation factor of À25.5& AE 3.5& for the assimilation of CO 2 coming from bacterial respiration into green algal tissue. Since we only had comparative data for the treatment period, the fraction of sucrose within PLFAs was calculated as follows: Biofilm autotroph carbon.-The carbon from cyanobacteria and algae was identified from a specific PLFA (a-linolenic acid, Risse-Buhl et al. 2012) using the same fractionation factor as above between PLFA and bulk tissue (Appendix S1: Table S2), knowing that this fractionation factor can vary between individual PLFAs and taxonomic groups (Taipale et al. 2015).
Analytical methods.-The total carbon and total nitrogen concentrations and the d 13 C and d 15 N natural abundance isotope ratios of the milled samples were determined using a Flash EA 1112 Series Elemental Analyzer connected via a Conflo III to a Delta Plus XP isotope ratio mass spectrometer (all Thermo Finnigan, Bremen, Germany). The isotope ratios were traceable to reference materials USGS40 and USGS41 (both L-glutamic acid); certified for d 13 C (& VPDB) and d 15 N (& air N 2 ). The carbon and nitrogen contents of the samples were calculated from the area output of the mass spectrometer calibrated against National Institute of Standards and Technology standard reference material 1547 peach leaves, which was analyzed with every batch of 10 samples. Long-term accuracies for a quality control standard (milled flour) were as follows: total carbon, 40.3% AE 0.35%; d 13 C, À25.4% AE 0.13&; total nitrogen, 1.7% AE 0.04%; and d 15 N, 1.90% AE 0.26& (mean AE SD, n = 200). Data processing was performed using Isodat NT software version 2.0 (Thermo Electron, Bremen, Germany) and exported into Excel. Total P was determined after a 30-minute digestion in 50% nitric acid at 120°C (see Demars and Edwards 2007) for CPOM, bryophytes, biofilm, and green filamentous algae in order to test for expected changes in C:N:P stoichiometry (hypothesis 3).
Data analyses for carbon sources and pathways.-Most studies use d 13 C and d 15 N to identify the flow path in the food web. Here the BACI experimental design allowed us to calculate the proportion of sucrose (Fs) in all parts of the food web. Fs was used as a tracer in addition to d 13 C to determine the sources of carbon for bacteria and algae in the treatment reach after 22 d of sucrose addition. Thus, the carbon pathways were identified with carbon tracers. While F s in the sources was determined from the change in d 13 C due to sucrose addition, the sources had different initial d 13 C and different coefficients of isotopic discrimination (or TEF, trophic enrichment factor), and this decoupled F s from d 13 C as shown in the isospace plots (Appendix S1: Figs. S2 and S3). End-member mixing analyses were used to determine the proportion of C sources and their uncertainties in primary producers and bacteria. We provided the numerical solutions given by a Bayesian stable isotope mixing model using MixSIAR 3.1.9 (Parnell et al. 2010, Stock and Semmens 2016 in R version 3.5.0 (R Core Team 2018). We checked that the mixture data were within the range of the sources (see raw data in Appendix S1: Table S3 and S4 and illustrations in Appendix S1: Figs. S2 and S3). The numerical solutions converged rapidly according to the Markov chain Monte Carlo convergence diagnostics (Gelman-Rubin and Geweke). The numerical solutions of Mix-SIAR were also very similar to the analytical solutions of IsoError 1.04 (Phillips and Gregg 2001) and suggested that the results were not biased (cf Brett 2014).

Quantification of carbon fluxes
To assess trophic transfer efficiencies in the food web, our production estimates were all standardized to g CÁm À2 Ád À1 . Respiration and photosynthesis rates in oxygen were converted to carbon using a respiratory and photosynthetic quotient of 1 (Williams and del Giorgio 2005, but see Berggren et al. 2012). Total CO 2 emissions (F CO2 ) were estimated to derive d 13 C CO2-allochthonous in the three-ended mixing model presented above and to estimate the relative proportion of biotic CO 2 emissions. In order to calculate carbon and light use efficiencies of the food webs (hypotheses 5 and 6), we determined net biotic productions and fluxes of carbon and energy supplies (sucrose, natural allochthonous DOM, photosynthetic active radiation). Finally, we could derive carbon uptake length and uptake velocity from heterotrophic production and DOM supply (hypothesis 4).
Total CO 2 emissions.-In the absence of direct measurements, the excess partial pressure of CO 2 (EpCO 2 ) of the streams was estimated from three measured parameters: pH, alkalinity and temperature (Neal et al. 1998; as applied in Demars et al. [2016] with atmospheric CO 2 = 384 ppm, data available online). 7 Our EpCO 2 estimates had high uncertainties (AE50%; Demars 2019). EpCO 2 is the concentration of free CO 2 in the stream water (C t at time t) relative to the atmospheric equilibrium free CO 2 concentration (C SAT ): C SAT was calculated from published CO 2 solubility in pure water at equilibrium with atmospheric CO 2 in the temperature range 0°À90°C (Carroll et al. 1991) and Henry's law (Stumm andMorgan 1981, Butler 1982). C t was calculated as EpCO 2 9 C SAT . The flux of CO 2 (F CO2 , g C m À2 d À1 ) at the interface between water and the atmosphere was calculated as for O 2 following Young and Huryn (1998): with k CO2 , reaeration coefficient of CO 2 (per d); C SAT and C t (mg C/L or g C/m 3 ); s, mean travel time of the stream reach (d); Q, average water discharge (m 3 /d); and A, surface water area of the stream reach (m 2 ). The reaeration coefficients between CO 2 and O 2 can be simply related as follows , Demars 2019: based on the molecular diffusivity (D m ) of CO 2 and O 2 measured at three different temperatures within the same study (Davidson and Cullen 1957), assuming dissolved gas solubility played a negligible role in the studied streams with little gas exchange area due to air bubbles (cf Hall and Madinger 2018). The flux of CO 2 was then related to discharge within the range of low stable flows for which stream metabolism was processed (Cairn n = 47, R 2 = 0.81; Birnie n = 65, R 2 = 0.77) to provide daily estimates. For more details, see Demars (2019).
Biotic CO 2 emissions.-These were simply calculated as the net ecosystem production (NEP), gross primary production (GPP) plus ecosystem respiration (ER, a negative flux) expressed in g CÁm À2 Ád À1 . Bacterial respiration of DOM was calculated as heterotrophic respiration (HR, a negative flux) from HR ¼ ER þ aGPP with a ¼ AR=GPP with AR, autotrophic respiration, and ER, ecosystem respiration, both negative fluxes (oxygen consumption) and GPP, positive flux (producing oxygen). We partitioned ER into auto and heterotrophic respiration with a = 0.5 (see  and calculated uncertainties using a = 0.2 and a = 0.8. Bacterial respiration of the added sucrose was calculated as the difference in heterotrophic respiration between the treatment and a control reach during sucrose addition, after standardizing for site differences using the control period. Whole-stream production.-We calculated net ecosystem primary production (NPP) by assuming a 50% carbon use efficiency (e = 1 À a), that is one-half GPP (reviewed in Demars et al. , 2017 and computed NPP for e = 0.2 and e = 0.8 to provide uncertainties in our estimates. Heterotrophic production (HP, g CÁm À2 Ád À1 ) was calculated from heterotrophic growth efficiency HGE and heterotrophic respiration (HR, negative flux expressed in g CÁm À2 Ád À1 ) as follows: HP ¼ ÀHR Â HGE 1 À HGE : Heterotrophic production was estimated for low (5%) and moderate (20%) bacterial growth efficiencies (e.g., Berggren et al. 2009, Fasching et al. 2014, Berggren and del Giorgio 2015. Allochthonous organic matter.-The overall flux at the outlet of both streams was calculated as discharge times DOC concentration using the weekly data from the long-term ECN monitoring collected in 2007-2008 under low stable flows. The DOC flux was then related to discharge (Cairn n = 74, R 2 = 0.85; Birnie n = 100, R 2 = 0.83), to provide an estimate for the days for which DOC was not measured (within the same range of flows).
Sucrose.-The flux of added sucrose over the treatment reach was determined at the top of the reach (84 m upstream of the flume of Cairn Burn) using the average observed increase in DOC concentrations (28 August, 5 September, 11 September 2007) multiplied by discharge (Appendix S1: Fig. S4).

Ecosystem efficiencies
With all fluxes expressed in g CÁm À2 Ád À1 , we calculated the light use efficiency (net primary production/ photosynthetic radiations), organic carbon use efficiency (bacterial production/DOC supply), and the proportion of biotic CO 2 emissions (net ecosystem production/total CO 2 emissions).

Data analyses of the experiment (stream reach scale)
We calculated an effect size (i.e., proportional changes) of sucrose addition on our response variables (e.g., nutrient cycling, stoichiometric ratios, metabolic fluxes, ecosystem efficiencies) using the values of the control (C) and treatment (T) reaches, before (subscript B) and after (subscript A) sucrose addition as follows: with all uncertainties propagated in quadrature using standard deviation dx for sums, and relative uncertainties dx/x for the division. Since the overall design was unreplicated, we simply reported the effect size in relation to the relative uncertainties or one standard deviation (equivalent to 68% confidence interval of our measurement errors). We did not calculate a standard error of the mean because our samples were pseudo-replicated (random collection was within one plot) or temporally autocorrelated (daily stream metabolism; see Hurlbert 1984). We therefore did not report P values. The interpretation of the results will be based on the strength of the effect size, knowledge from previous experiments addressing individual aspects of our study, and hypothetical standard errors based on varying the number of replicated studies. The control period for the metabolic parameters and trophic transfer efficiencies was limited to the 2 d prior to sucrose addition (21-22 August), so the whole data series was within a single period of stable flows, i.e., not interrupted by peak flows, producing more comparable results across sites.

Experimental DOM addition
Stable flows were present during the whole three weeks of carbon addition (Fig. 3). The target concentration was achieved at the top of the studied reach, 28 m downstream of the injection point, where the DOM addition averaged 0.52 mg C/L above background concentration (Appendix S1: Fig. S4).

Stream metabolism continuous monitoring
Gross primary production (GPP) peaked to 7.6 g O 2 Ám À2 Ád À1 10-11 d after the start of sucrose addition in the treatment reach before decreasing sharply down to an average 2.4 g O 2 Ám À2 Ád À1 during the last 4 d of the carbon addition, and this despite high photosynthetic active radiations. In contrast, GPP remained relatively constant in the control reaches Birnie Burn (about 1.2 g O 2 Ám À2 Ád À1 ) and Cairn Burn (3.2 g O 2 Ám À2 Ád À1 for the first two weeks declining to 2.1 g O 2 Ám À2 Ád À1 during the last week g O 2 Ám À2 Ád À1 ; Fig. 3). Bryophytes covered 4, 11 and 17% of the river bed in the Birnie control, Cairn control and Cairn treatment reach, respectively. Filamentous green algae (mostly Microspora sp., Microsporaceae) percentage cover increased during the period of sucrose addition from 1% to 19%, 11% to 33%, and 4% to 37% of channel width in the Birnie control, Cairn control, and Cairn treatment reach, respectively.
Peaks in ecosystem respiration (ER) to À20 g O 2 Ám À2 Ád À1 (Birnie Burn) and À35 g O 2 Ám À2 Ád À1 (Cairn Burn) were more visible for the controls than the treatment reach, with respiration activity inversely related to soil hydrological connectivity, as recorded by soil moisture continuous monitoring (Fig. 3). There was an increase in ER in the treatment reach at the start of the sucrose addition, despite the continuing loss of hydrological connectivity. While ER mirrored GPP with a 4-d time lag (ER t+4 d = À1.06 AE 0.12 9 GPP t À4.23 AE 0.64 g O 2 Ám À2 Ád À1 ; R 2 = 0.79), heterotrophic respiration activity associated to sucrose addition peaked sharply 15 d after the start of the addition, independently of GPP and autotrophic carbon use efficiency, processing up to 59% of the daily sucrose flux (Appendix S1: Fig. S5). On average 35% AE 20% of the added sucrose was respired during the addition over just 84 m (or 15 minutes mean travel time). Heterotrophic production ranged between 2% and 10% of the sucrose flux, based on bacterial growth efficiencies of 0.05 and 0.20, respectively.

Nutrient cycling studies and stoichiometry
The background concentrations of nitrate and phosphate were 180 and 90 lg N/L and 2 and 4 lg P/L in the Birnie control and Cairn treatment reach, respectively. The added geometric mean of N and P were on average 471 lg N/L and 24 lg P/L. The addition of sucrose had no effect on nitrate and phosphate nutrient uptake length and uptake velocity (Fig. 4, Appendix S1: Table S5). The phosphate uptake length was highly related to discharge and became very short, down to 31 m in the treatment reach toward the end of the sucrose addition. Phosphate uptake velocity was about 0.2 mm/s and an order of magnitude faster than nitrate with uptake lengths in the kilometer range.
The molar C:N:P stoichiometric ratios of coarse particulate organic matter and bryophytes remained stable throughout the experiment. Sucrose addition exerted strong effects on filamentous green algae and periphyton stoichiometry (Fig. 5, Appendix S1: Table S6). While the molar C:N:P stoichiometric ratios decreased in the control stream, they increased sharply in the treatment reach following sucrose addition: from 330:29:1 to 632:49:1 in filamentous green algae and from 262:29:1 to 428:38:1 in periphyton.

Identification of carbon sources and pathways under stable flows (end of experiment)
The autotrophs (filamentous green algae and periphyton autotrophs) derived 63% AE 14% of their CO 2 from allochthonous sources (groundwater and atmosphere) and 37% AE 14% from autochthonous bacteria in the control reach. In the treatment reach, autotrophs derived a similar proportion of CO 2 from allochthonous sources (61% AE 8%), some osmotrophic uptake of sucrose (13% AE 8%), and relatively less from bacteria (26% AE 11%). The bacteria used equally autotrophic carbon (49% AE 15%) and allochthonous organic carbon FIG. 3. Continuous data monitoring and stream metabolism in the control reach (Birnie Burn) and treatment reach (Cairn Burn). Ecosystem respiration (ER) is negative (consuming O 2 ) and gross primary production (GPP) is positive (producing O 2 ). The period of DOM (sucrose) addition (23 Aug-14 Sep) is indicated by gray shading. The depths for soil moisture correspond to the depths of the organic soil (10 cm) and subsoil (45 cm).

Ecosystem carbon fluxes
Photosynthetic active radiation (light) decreased slightly from 106 to 80 and 118 to 90 g CÁm À2 Ád À1 in the control and treatment respectively. Total CO 2 emissions (range 10-29 g CÁm À2 Ád À1 ) were higher before than during the treatment period, and higher in the control than in the treatment, reflecting differences in hydrological connectivity with soil water between sites and period of time. The flux of natural allochthonous organic matter (DOM) was more than halved from 177 to 85 and 110 to 50 g CÁm À2 Ád À1 in the control and treatment due to the soils becoming dryer (Fig. 3). This was reflected by a general reduction in the organic carbon uptake length 3,214 to 2,531 m and 4,257 to 1,886 m in the control and treatment, respectively, independently of the carbon addition. The organic carbon uptake velocity decreased in the control from 0.82 to 0.55 m/d but increased in the treatment from 0.53 to 0.76 m/d. FIG. 4. Stream nutrient uptake length (S w ) and uptake velocity (v f ) before and during sucrose addition in the control (gray symbols) and treatment (black symbols). Error bars represent SE, with some error bars smaller than the symbols. Note the different magnitudes on the y-axes between NO 3 -N and PO 4 -P.

Effects of carbon additions
The small carbon addition had large effects on carbon fluxes and metabolic efficiencies (see Fig. 8, Appendix S1: Tables S7 and S8). While GPP increased slightly over the whole treatment period (12%), there was a small relative increase in light use efficiency (37%). ER intensified by 70%, and the net ecosystem production (NEP) became relatively more negative by 125%, i.e., 125% relative increase in biotic CO 2 emissions. Heterotrophic respiration and production increased by 89%, and this was reflected by a shorter (À40%) uptake length (S w-OC ) and faster mineralization velocity (92%) of organic carbon. The proportion of natural allochthonous DOC flux respired (range 2.2-5.3%) and organic carbon use efficiency by bacteria (range 0.1-0.3%) increased by 112% at the reach scale.

DISCUSSION
The use of a before and after control impact experiment together with the addition of a deliberate tracer with a distinctive d 13 C signature allowed not only to trace the fate of the added carbon into the treated reach but also to build the flow food web of the control reach, unravel C reciprocal subsidies between autotrophs and bacteria, and demonstrate the potential for some natural allochthonous organic matter to feed the primary producers via bacterial respiration. Our experiment showed that a small continuous addition of labile DOM (0.52 mg C/L as sucrose, 12% of total DOC, Appendix S1: Fig. S1) can profoundly alter wholeecosystem behavior (see also Warren et al. 1964, Lutz et al. 2012, Robbins et al. 2017.

Reciprocal subsidies between autotrophs and bacteria
The use of bacterial CO 2 by autotrophs has long been part of lake, marine, and earth-scale carbon cycles (Hutchinson 1948, Redfield 1958, Kuznetsov 1968), but has been overlooked in streams. Here we show that bacterial CO 2 can also be important in turbulent headwater streams, accounting for 37% of the CO 2 fixed by autotrophs in the control reach. This is possible within the intricate matrix of the biofilm (e.g., Kamjunke et al. 2015, Battin et al. 2016 or an algal mat, as when the CO 2 is released in the water column it is very quickly degassed to the atmosphere (here 5-15 minutes; Demars 2019). This is supported by a preferential uptake of bacterial CO 2 in autotrophs relative to total flux of CO 2 : 1.4 AE 1.1 and 1.7 AE 1.4 in the control and treatment reach, respectively (calculated as bacterial CO 2 relative to total CO 2 fixed by autotrophs compared to heterotrophic respiration relative to total CO 2 emissions).
The results suggest a significant proportion of autotrophic C (19 AE 9%) was derived from allochthonous DOM via bacterial respiration in the control (Fig. 6). The bacterial CO 2 flux to autotrophs assumed its d 13 C signature was the same as that of the bacteria (as in Collins et al. 2016), and the proportion of C sources the FIG. 5. Proportional changes in C, N, P (% w/w) and molar C:N:P stoichiometric ratios in the basal food web resources due to sucrose addition based on the before and after control impact experimental design. Error bars represent SE; CPOM, coarse particulate organic matter. same as for bacterial production. The controlled experiment allowed us to calculate the bacterial respiration of the added sucrose as 2.16 g CÁm À2 Ád À1 over the three weeks of sucrose addition. This represented 82% (53-184%) of the estimated treatment bacterial respiration (2.64 AE 1.45 g CÁm À2 Ád À1 , Appendix S1: Table S8) and the proportion of sucrose in bacteria was estimated at 62% AE 7% (Fig. 6). So the estimates were still within measurement errors. The proportion of DOM channeled to autotrophs via bacterial respiration was similar in the treatment (21% AE 8%), more from sucrose (16% AE 7%) than natural DOM (5% AE 3%). This microbial loop allows part of the allochthonous DOM processed by bacteria (brown web) to be incorporated into autotrophs (green web), likely increasing the quality of food available to consumers (Brett et al. 2017).
The use of autotrophic C by bacteria was reduced by the sucrose addition despite a small increase in GPP. Bacterial production in the treatment relied more on sucrose than autotrophs and natural allochthonous organic matter, assuming a similar bacterial growth efficiency for all sources (range 5-20%), which is plausible under nutrient limitations (del Giorgio and Cole 1998) and corresponded to previous studies (e.g., Berggren et al. 2009, Fasching et al. 2014, Berggren and del Giorgio 2015. Overall, sucrose addition shifted the reciprocal carbon exchange between bacteria and autotrophs to a predominantly one-way flow from bacteria to autotrophs.

Boom and bust: role of nutrients
Gross primary productivity (GPP) appeared to be stimulated by the addition of sucrose (contrary to expectations under nutrient limitations, Fig. 1) but this was short lived, despite sustained light availability during the addition period (Fig. 3). Heterotrophic respiration of sucrose (independent of NPP) peaked after two weeks but crashed within days while the supply of sucrose was continuously flowing through the reach (with sucrose concentration increasing from 0.22 to 0.88 mg C/L with falling discharge, Appendix S1: Fig. S4). This was in sharp contrast to the peaks in respiration followed by a more sustained response of ecosystem (mostly heterotrophic) respiration to hydrological connectivity with soil water in the control reaches ( Fig. 3; see Demars FIG. 6. Carbon source partitioning of autotrophs (black arrows) and bacteria (gray arrows) in the control (top) and treated reach (bottom) after three weeks of sucrose addition. Numbers represent the proportion (%) of the sources and their uncertainties (SD). The calculations were based on carbon stable isotope analyses (d 13 C) and the fraction of sucrose in the different compartments (Appendix S1: Tables S3 and S4). 2019). This boom and bust in the treatment reach was likely due to nutrient limitations for both autotrophs and eventually heterotrophs (e.g., Peterson et al. 1985, Kominoski et al. 2018, by both N and P according to the changes in filamentous green algae and periphyton C:N:P stoichiometry (C:N:P = 632:49:1 and 428:38:1, respectively), and relative to the optimal ratio for growth (C:N:P = 119:17:1 to 158:18:1; Kahlert 1998, Hillebrand andSommer 1999).
The shortfall of nutrients (N, P) was not compensated by faster nutrient cycling rates, as observed in previous studies with much higher concentrations of labile DOM additions (e.g., Bernhardt and Likens 2002), possibly due to increased bacterial retention at ecosystem scale . This may also be due to limitation of phosphate uptake by NH 4 availability (long term median in both streams 7 lg N/L) in streams relatively rich in nitrate (see Oviedo-Vargas et al. 2013).
The observed pulse of ER due to sucrose addition seemed to mirror GPP with a 4-d time lag, but heterotrophic respiration was much larger and mostly independent of NPP (GPP and carbon use efficiency by autotrophs, Appendix S1: Fig. S5). So, while the small peak in GPP was neither predicted (under the assumption of nutrient limitation, Fig. 1) nor observed in the control reaches (Demars 2019), sucrose addition did shift the metabolic balance toward heterotrophy. The shift in heterotrophy was not so obvious in the comparative flow food webs (see Fig. 7), but differences in size effects with the more robust BACI design were clear: only +12% for GPP but +70% for ER and +89% for heterotrophic respiration (Fig. 8). FIG. 7. Flow food webs and biotic CO 2 emissions under stable flows: in-stream biotic carbon fluxes (g CÁm À2 Ád À1 , black arrows) in the control (top) and treatment reach (bottom) after three weeks of sucrose addition, based on source partitioning using stable isotopes (end of experiment) and production estimates (average during sucrose addition). The C fluxes going to autotrophs and bacteria represent rates of biomass accrual, i.e., net primary production and heterotrophic production. Bacterial respiration and net ecosystem production (biotic CO 2 emissions) were similar in the control and the treatment (unlike expectations in Fig. 1) because the stream water of the control reach was hydrologically still connected to soils at the beginning of the treatment period (Fig. 3), unlike the treatment reach, which had negligible lateral inflows but see Fig. 8 for the effects. Two estimates were given for bacterial production based on heterotrophic growth efficiencies (HGE) of 0.05 (low) and 0.2 (moderate). The uncertainties represent SD (68% confidence interval, see Discussion, Uncertainties for interpretation of uncertainties in unreplicated studies).
A small increase in GPP following sucrose addition has been observed before under similar experimental conditions (Robbins et al. 2017). This transient response may be explained by the increase in supply of CO 2 by bacteria and an initial sufficiency of nutrients (N, P).

Fate of carbon
By and far, the largest quantity of carbon processed by bacteria was lost as CO 2 emission. Heterotrophic respiration over the treated reach respired (on average) 35% of the added sucrose. Bacterial production averaged 2-10% of the sucrose flux within the studied reach. Heterotrophic respiration of natural allochthonous DOM was about 3% in the control stream and 2.2% prior to sucrose addition in the treatment. The fact that sucrose was processed 10 times faster than natural stream DOM was well reflected by the shortening of the organic uptake length (S w-OC ) and increased mineralization velocity (v f-OC ). This was expected (see, e.g., Marcarelli et al. 2011, Mineau et al. 2016. However the average rates of mineralization of natural DOM at the soil-water interface in both streams (scaled up to the first-order catchment) were relatively high: 22% AE 10% (Birnie) and 31% AE 15% (Cairn) under stable flows, with peaks over 50% (Demars 2019). The changes in organic uptake lengths in the control and the differences between control and treatment prior to sucrose addition can be explained by the loss of hydrological connectivity with soil water, as indicated by the changes in soil moisture, and the difference in lateral inflows between the control (10.7%) and the treatment (2.3%). The shortening of the organic uptake length (S w-OC ) reflected more hydrological changes as indicated by the different direction of change in organic carbon uptake (v f-OC ) between the control and the treatment. In the control v f-OC declined by 32% against a 52% decline in DOM supply. In contrast, v f-OC increased by 43% in the treatment with the addition of labile carbon, despite a 55% fall in DOM supply (similar to the control). Overall, the uptake lengths of natural DOM (i.e., excluding those influenced by sucrose addition) were longer (2.5-4.3 km) than the length of the first-order streams studied here (1 km), and so a large part of the carbon is released to downstream ecosystems as previously observed (e.g., Wiegner et al. 2005), especially during times of loss of hydrological connectivity with the soil of the catchment (Demars 2019).
The treatment reach did not show the large peaks in respiration (outside the period of sucrose addition) that the control reaches showed when the catchment was hydrologically connected, as indicated by soil moisture (Fig. 3; Demars 2019). This is likely because the treatment reach is a more constrained reach largely disconnected from the land. It was initially chosen to avoid lateral inflows, which were very small (2.3%) and mostly from a spring fed flush (i.e., groundwater), rather than seepage from organic and riparian soils known to stimulate bacterial activity (e.g., Brunke andGonser 1997, Pusch et al. 1998). Interestingly, the average bacterial respiration was similar in the control and treatment reaches over the three weeks of sucrose addition (Fig. 7), suggesting part of the organic matter respired in the control was relatively labile and comparable to the 0.5 mg C/L of added sucrose. The dynamic of this respiration within the three weeks was very different however, decreasing with the progressive loss of hydrological connectivity with soils in the control, and peaking after two weeks in the treatment (Fig. 3), also independently of GPP (Appendix S1: Fig S5).
Sucrose is very labile and is well known to promote the growth of filamentous bacteria Sphaerotilus natans ("sewage fungus"), even at relatively low concentrations (0.25-1.00 mg/L) in a stable flow forested stream FIG. 8. Relative effect size of sucrose addition (change relative to the mean, 1 = 100%) on selected ecosystem properties at reach scale based on the before and after control impact experimental design. Error bars are AE SD (equivalent to 68% confidence interval), not the usual standard error of the mean because the study was unreplicated and without P values. Many of the responses from individual ecosystem properties followed expectations from previous studies, hence error bars could be considerably reduced by replicating the experiment (see Discussion, Uncertainties). See Appendix S1: Tables S7 and S8 for individual values. (Warren et al. 1964). This was not observed in this study, likely because the moorland streams studied here were more open and colonized by Microspora, a common genus of filamentous green algae in Scottish streams (Kinross et al. 1993). Microspora was likely able to uptake sucrose by osmotrophy (Wright and Hobbie 1966) but this accounted for only 13% of C uptake by Microspora. Overall daily uptake of sucrose by autotrophs represented 1.6% of the average daily flux of sucrose.

Uncertainties
We strived to quantify and propagate uncertainties in all our calculations. These uncertainties may appear large, particularly for the effect size of ecosystem properties presented in Fig. 8. It was clear from the discussion that previous studies produced similar results for individual effects, and thus error bars could be considerably reduced by replicating this experiment allowing the calculation of a standard error (SE) and statistical significance (P value). Since SE ¼ dx= ffiffi ffi n p with dx standard deviation and n the number of replicates, just four replicates could potentially divide the error bars by two and make 80% of the effects presented in Fig. 8 significant (assuming independent experiments produced the same results).
Much of the C flux uncertainties in the flow food webs (Fig. 7) did not come from C source partitioning (see Fig. 6), but rather from the estimation of net autotrophic and net heterotrophic ecosystem production (Appendix S1: Table S8). The estimation of net autotrophic production will likely remain challenging (e.g., ) but bacterial growth efficiency should be determined in future studies.
The autotrophs did not include bryophytes in our flow food web calculations because their contribution to primary production was thought to be negligible over the few weeks of the experiment, and particularly toward the end of the experiment when filamentous green algae were covering bryophytes. The lack of bryophyte growth under very low phosphorus concentrations (here 2-4 lg P/L of soluble reactive P) combined with shading by epiphytes has been well documented (e.g., Finlay and Bowden 1994). This may also explain the lack of changes in bryophyte C:N:P stoichiometry (Fig. 5).

CONCLUSIONS
Part of the carbon derived from allochthonous organic matter can feed the autotrophs via the CO 2 produced by stream bacterial respiration, intermingling the green and brown webs. The preferential uptake of labile carbon (sucrose) by bacteria and excess bacterial CO 2 relative to nutrients (N, P) for autotrophs shifted the reciprocal carbon exchange between bacteria and autotrophs to a predominantly one-way carbon flow from bacteria to autotrophs. Light use efficiency by autotrophs increased, but so did the C:N:P molar ratios of autotrophs, the latter likely to become less palatable to consumers. The bacterial response to sucrose addition shifted the metabolic balance toward heterotrophy, increasing organic carbon use efficiency and biotic CO 2 emissions, shortening carbon uptake length (average distance travelled by a molecule of organic matter) and providing less organic matter of lower quality for downstream ecosystems.
Even a small increase in labile dissolved organic matter supply due to climate and land use change could alter instream carbon cycling, with large effects on stream food web and biogeochemistry in small streams draining catchments with soils rich in organic carbon. These effects could be accrued by climate change, notably temperature increase, thawing of permafrost and rainfall increase in boreal ecosystems with soils rich in organic matter.

ACKNOWLEDGMENTS
We thank Carol Taylor and Helen Watson for managing the long-term monitoring, Yvonne Cook and Susan McIntyre for running water chemical analyses, Claire Abel for the phospholipid fatty acid extraction, Maureen Procee for running the compound specific isotope ratio analysis, Gillian Martin for preparing and running the samples for stable isotope ratio analysis, Glensaugh farm manager Donald Barrie for hosting BOLD during the experiment and facilitating the work, Baptiste Marteau and two anonymous referees for comments on the manuscript. This study was funded by the Scottish Government Rural and Environmental Science and Analytical Services (RESAS), with additional funding support as part of the UK Environmental Change Network (ECN), and NERC Macronutrient Cycles Program. The writing up was partly funded by the Norwegian institute for water research (NIVA). The authors acknowledge the provision of data forming part of the ECNwide data set, https://catalogue.ceh.ac.uk/documents/456c24dd-0fe8-46c0-8ba5-855c001bc05f LITERATURE CITED