Spatial and temporal variation in nitrogen fixation and its importance to phytoplankton in phosphorus-rich lakes

1. Limnological theory posits that phosphorus (P) limits primary production in fresh water lakes, in part because fixation of atmospheric nitrogen (N 2 ) can compensate for limitations in nitrogen (N) supply to phytoplankton. However, quantitative estimates of the degree to which N 2 fixation satisfies planktonic N demand are rare. 2. Here we used biweekly sampling during summer in seven lakes over 2 decades to estimate both planktonic N 2 fixation and phytoplankton N demand. We further assessed

To date, select whole-lake experiments have been instrumental in demonstrating that biological N 2 fixation can sustain lake eutrophication arising from moderate fertilisation with P alone (Higgins et al., 2017;Schindler et al., 2008). However, several lines of evidence suggest that N 2 fixation may be insufficient to meet N requirements of natural phytoplankton assemblages in some P-rich lakes. For example, mass-balance studies comparing biological N 2 fixation with other N sources suggest that planktonic cyanobacteria may supply only a small fraction (5-10%) of summer N influx to lakes in agricultural basins (Ferber, Levine, Lini, & Livingston, 2004;Leavitt et al., 2006;Patoine et al., 2006 but see Scott & Grantz, 2013). Second, soluble reactive phosphorus (SRP) is freely available (>50 μg/L) during and after intense blooms of diazotrophic cyanobacteria (Lathrop, 2007; McGowan, Leavitt, & Hall, 2005), suggesting that P limitation has not been re-established (Scott & Grantz, 2013). Third, N 2 fixation is energetically demanding (Flores & Herrero, 2005;Herrero, Stavans, & Flores, 2016) and can be limited by low light availability in turbid eutrophic lakes (Ferber et al., 2004;Mugidde, Hecky, Hendzel, & Taylor, 2003). Fourth, nitrogenase enzyme complex activity may be limited by supply of micronutrients Marino, Howarth, Chan, Cole, & Likens, 2003).
Fifth, fixed N is readily released from diazotrophic cyanobacteria (Glibert & Bronk, 1994;Patoine et al., 2006) and can be transformed and returned to the atmosphere via microbial denitrification before uptake by other phytoplankton, particularly in eutrophic lakes (Bruesewitz, Hamilton, & Schipper, 2011;David, Wall, Royer, & Tank, 2006;Seitzinger et al., 2006). These diverse observations, in conjunction with the observation that lakes differ by up to 100-fold in the availability of water-column P (Harke et al., 2016;Patoine et al., 2006), prohibit generalisation about the importance of fixed N 2 in sustaining lake production and P limitation.
In this study, we used two decades of biweekly experiments and monitoring data from seven lakes to quantify the importance of fixed N to satisfy the biological demand of phytoplankton in eutrophic hardwater lakes characteristic of the northern Great Plains and other continental interiors . In addition, we sought to identify the environmental conditions associated with elevated N 2 fixation and importance to phytoplankton. Study sites span a gradient from mesotrophic to hyper-eutrophic lakes that vary in the degree of N and P deficit (Finlay, Patoine, Donald, Bogard, & Leavitt, 2010;Hall, Leavitt, Quinlan, Dixit, & Smol, 1999), N 2 fixation , N and P availability (Bogard, Donald, Finlay, & Leavitt, 2012;Donald, Parker, Davies, & Leavitt, 2015) and urban pollution with N . We predicted that fixed N would supply a larger proportion of N demand in perennially N-limited downstream lakes (Smith & Schindler, 2009) and that fixed N would be more important to primary producers in years with elevated production and growth demand (i.e. that production and fixation would vary synchronously; Scott, Stanley, Doyle, Forbes, & Brooks, 2009). Finally, we anticipated that there would be strong longitudinal gradients of N 2 fixation and demand, as downstream eastern lakes in the Qu'Appelle River catchment exhibit lower ratios of N:P (Soranno et al., 1999) and historically higher abundance of diazotrophic cyanobacteria Vogt, Sharma, & Leavitt, 2018). Ultimately, the objective of this study was to ascertain whether retained fixed N is sufficient to meet phytoplankton N demand in P-rich hardwater lakes.

| Study sites
The seven study lakes are located in the Qu'Appelle River drainage basin (ca. 52,000 km 2 ) in southern Saskatchewan, Canada ( Figure 1) and are part of the Qu'Appelle Long-term Ecological Research network (QU-LTER) as described by Vogt, Rusak, Patoine, andLeavitt (2011), Vogt et al. (2018). The Qu'Appelle River flows eastward from headwaters near Lake Diefenbaker through a series of six productive lakes (including Buffalo Pound, Pasqua, Katepwa and Crooked) to its confluence with the Assiniboine River in Manitoba.
Wascana Lake and Last Mountain Lake drain into the Qu'Appelle River mid-reach near the city of Regina. Nitrogen-rich urban effluent from Regina first enters Pasqua Lake before being conveyed to eastern downstream lakes (Katepwa, Crooked). Land use within the Qu'Appelle watershed is comprised mainly of agriculture (75%), along with natural grasslands (12%), the urban centres (5%) of Moose Jaw and Regina, and surface waters (8%; Finlay et al., 2015;Hall et al., 1999). The effective drainage area of each lake, defined as the region supplying water to a lake during years of median river flow, increases from west to east for lakes along the mainstem of the Qu'Appelle River ( Figure 1). Southern Saskatchewan's climate is classified as cool-summer humid continental (Köppen Dfb classification), with short summers (mean 19°C in July), cold winters (mean -16°C in January), high evaporation (ca. 60 cm/year) relative to precipitation (ca. 30 cm/year), and ca. 75% of runoff occurring during spring snowmelt Pham, Leavitt, McGowan, Wissel, & Wassenaar, 2009). Study lakes range from mesotrophic Lake Diefenbaker to hypereutrophic Wascana Lake and vary >10-fold in most physical, chemical and biological properties (Table 1). Organic and inorganic C content is high in all lakes, reflecting regional geology, soil characteristics and high C influx from terrestrial ecosystems (Finlay, Leavitt, Patoine, & Wissel, 2010). Lakes are polymictic in most years, with pronounced deepwater anoxia by late summer in central and eastern lakes (Vogt et al., 2011). Mean summer concentrations of Chl-a are elevated in all lakes except mesotrophic Lake Diefenbaker, while ratios of dissolved N:P decline from the mid-reaches of the Qu'Appelle drainage east to the Manitoba border. Diverse N 2 -fixing and non-N 2 -fixing cyanobacteria are common in all of the lakes (e.g. Aphanizomenon, Anabaena, Gloeotrichia, Microcystis), as are surface blooms during July-September (Donald, Bogard, Finlay, Bunting, & Leavitt, 2013;Leavitt et al., 2006). Summer residence time was calculated for each basin as the lake volume divided by inflow minus evaporation. Groundwater was not included in these calculations as it is expected to be a minor component of the water budget in these lakes. River inflow to each lake was estimated from two sources: gauge-measured flows from data collated by the Government of Canada's Water Office (https://wateroffice.ec.gc.ca/index_e.html) and projections from the Saskatchewan Water Security Agency Water Resources Management Model.
Inflow for each lake was calculated as the sum of gauged inflows (May-August). Lake volume was calculated as the average of volume estimates calculated from area capacity curves and daily measurements of lake level.

| Limnological monitoring
Lakes were sampled biweekly between 1 May (day of year, DOY, 121) and 31 August (DOY 243) following standard protocols (Vogt et al., 2011(Vogt et al., , 2018. Diefenbaker, Wascana, Last Mountain, Katepwa, and Crooked lakes were sampled 1996-2014, while Pasqua Lake was sampled 2004-2014. Most lakes were sampled between 09:00 hr and 13:00 hr at a single, standard site located at the deepest point of the lake. In contrast, Wascana Lake was sampled at a standard location; however, as the lake was deepened in 2004, this site was the deepest point in the lake only until 2004. Profiles of dissolved oxygen, pH, conductivity and temperature were measured at 1-m intervals using a YSI-85 multi-probe meter (YSI, Inc., Yellow Springs, OH, USA). Depth-integrated water samples were collected by pooling F I G U R E 1 Qu'Appelle River drainage basin in southern Saskatchewan (inset), Canada, showing location of study lakes. Watershed is indicated in grey with a dashed line. Wascana Lake is located in an urban park with the City of Regina samples from a 2.2-L Van Dorn water bottle deployed at 0.5-m intervals (Vogt et al., 2011). Depth-integrated water was screened through a 243μm mesh net to remove zooplankton and stored at 4°C in a dark bottle until processed. An aliquot of 100 ml of depthintegrated water sample was preserved with Lugol's IKI solution for microscopic analysis of the phytoplankton community composition (Donald et al., 2013).
Samples for analysis of Chl-a, particulate organic matter (POM), phytoplankton pigments, and stable N isotopes were filtered onto Whatman GF/C glass fibre filters (nominal pore size 1.2μm), wrapped in aluminum foil, placed in a dark film canister and frozen (-12°C). Depth-integrated water was filtered through 0.45μm pore size membrane filter and the filtrate was stored until nutrient analysis, as described by Vogt et al. (2011). We estimated primary production using standard dark and light bottle techniques for eutrophic lakes (Rice, Baird, Eaton, & Clesceri, 2012) following Finlay, Leavitt, Wissel, and Prairie (2009). Briefly, triplicate samples of 243μm screened, depth-integrated lake water were incubated in a growth chamber in either transparent or darkened 250-ml glass bottles. Bottles were incubated at ambient lake temperature and under a 12-hr light/dark cycle with 450 μmol quanta m -2 s -1 , a photon flux comparable to that recorded in situ at Secchi depth using a profiling radiometer . Oxygen concentrations in each bottle were measured at the start of the experiment and after incubation for 24 hr.
Heterocyte enumeration was restricted to samples collected over 7 years from Crooked and Katepwa lakes, the two basins with the highest rates of N 2 fixation in preliminary studies .
Concentrations of biomarker pigments (nmoles pigment/L) were estimated for compounds characteristic of total algal abundance (Chl-a, pheophytin a, β-carotene), chlorophytes and cyanobacteria (lutein-zeaxanthin), total cyanobacteria (echinenone), colonial cyanobacteria (myxoxanthophyll), and Nostocales cyanobacteria (canthaxanthin; Leavitt & Hodgson, 2001). In this HPLC system, structural isomers from chlorophytes (lutein) and cyanobacteria (zeaxanthin) were not separated and were instead used to estimate abundance of summer bloom-forming taxa (Hall et al., 1999; Leavitt  , 2006). The proportion of phytoplankton biomass in the surface waters (surface to integrated Chl-a) was calculated by dividing the Chl-a from a surface grab sample by the Chl-a concentrations of a depth-integrated water sample.
The elemental composition (%C, %N, C:N mass ratio) and stable nitrogen isotope value (δ 15 N) of particulate organic material (POM) from depth-integrated water samples were estimated by combustion and isotope-ratio mass spectrometry following Savage, Leavitt, and Elmgren (2004). POM was scraped from frozen filters into preweighed tin capsules, dried to constant weight, and combusted in a NC2500 Elemental Analyzer (ThermoQuest; CE Instruments) coupled to a Thermoquest (Finnigan-MAT) Delta Plus XL IRMS. Nitrogen stable isotope ratios are reported in the conventional δ notation with respect to atmospheric N 2 .
Nutrient content of lake waters was estimated using standard

QU-LTER protocols at the Biogeochemical Analytical Service
Laboratory, University of Alberta, Edmonton, Alberta, Canada (Vogt et al., 2011(Vogt et al., , 2018). Depth-integrated water was screened (243μm mesh) then filtered through an 0.45μm pore size membrane filter within 3 hr of collection before analysis for total dissolved phosphorus and orthophosphate (SRP), both as μg P/L, as well as NO 3 -, NH 4 + , and total dissolved nitrogen (TDN), which includes organic and inorganic fractions of dissolved nitrogen (all mg N/L) using standard analytical procedures (Stainton, Capel, & Armstrong, 1977).

| Estimation of N demand and N 2 fixation
Nitrogen demand was calculated from bioassay estimates of phytoplankton productivity (mg C fixed m -3 hr -1 ), in situ estimates of light penetration , measured day-length, and estimates of phytoplankton C:N ratios from POM samples Patoine et al., 2006). This approach assumed that: there was a constant stoichiometry between the rate of oxygen evolution and carbon fixation (Sala, Jackson, Mooney, & Howarth, 2000); phytoplankton acquired sufficient N to meet the measured rate of C uptake and maintain observed cellular ratios of C:N (Ferber et al., 2004); and POM was composed mainly of phytoplankton (Donald et al., 2013). Microbial respiration (R) was estimated as the rate of decline in oxygen in the dark bottle, whereas net primary productivity (NPP) was estimated from increased oxygen concentrations in light bottles, and gross primary productivity (GPP) was calculated as the sum of NPP and R (all as mg O m -3 hr -1 ). We assumed that R includes both autotrophic and heterotrophic respiration by microbes and that R in light and dark were similar, such that GPP represented mainly autotrophic processes (Rice et al., 2012). Metabolic rates were converted to carbon equivalents (mg C m -3 hr -1 ) assuming a photosynthetic quotient of 1.24 (Sala et al., 2000). Nitrogen demand (mg N m -3 hr -1 ) was estimated by multiplying GPP by the C:N of POM, assuming that phytoplankton maintained their cellular N quota at the observed C:N ratios. We used GPP to estimate N demand because dark and light bottle assays more accurately estimate GPP than NPP and because bottle-based estimates of GPP correlated strongly with phytoplankton standing stock in these lakes ).
Whole-lake and summer-long estimates of N demand in these polymictic lakes were determined from biweekly estimates of lake transparency, day length, and gross planktonic productivity described above Finlay et al., 2009).
Estimates of whole-lake planktonic N demand were calculated by multiplying lab-based rates of N demand with the fraction of the day spent in the euphotic zone, defined as the ratio of the euphotic volume (as Secchi depth) to total lake volume. We assumed that our summer-long whole-lake estimates of phytoplankton productivity captured a high proportion of spatial and temporal variation in GPP attributable to changes in light regimes. We made this assumption because incubator irradiance was equivalent to that measured at Secchi depth, Secchi depth varied seasonally from ca. 10 cm to ca. 8 m, and rates of photosynthesis are typically independent of irradiance at higher photon fluxes (MacIntyre, Kana, Anning, & Geider, 2002). Further discussion of these assumptions, and their comparison to other methods of estimating lake production, is presented in Finlay,  and Finlay et al. (2009).
Two approaches were used to estimate rates of nitrogen fixation: we compare the more commonly used heterocyte-based method  to demonstrate the accuracy of our isotopebased method . First, we used two forms of the natural abundance method (NAM) developed for both freshwater ecosystems  and oceanographic studies (Karl et al., 1997). NAM mixing models assume that summer declines in the δ 15 N of lake nitrogen arise mainly from N 2 fixation, a process that is offset, in part, by a similar magnitude of isotopic enrichment due to denitrification Patoine et al., 2006;Peterson & Fry, 1987). As such, NAM integrates N transformations and represents the residual flux (gross N 2 fixation -denitrification) of fixed N to primary producers. This is the equivalent to the proportion of total fixed N that remains in the water column after denitrification and is referred to as net N 2 fixation hereafter. Because over 75% of total annual inflow occurs some 4-7 weeks prior to monitoring (Pham et al., 2009), we assumed that in each year most allochthonous N had already entered the lake by the start of our estimates of N fixation or demand Pham et al., 2009) and that phytoplankton mainly used this dissolved N pool, which includes both inorganic and organic forms of nitrogen, for growth.
In the first NAM method, and throughout this paper, we modelled the influx of N into the particulate pool by measuring changes in δ 15 N of POM following Patoine et al. (2006). In addition, we compared these results to a second NAM method that modelled the influx of to carbon by multiplying by a standard Chl-a to carbon conversion ratio (Patoine, Graham, & Leavitt, 2006). We converted from POM-C to POM-N by multiplying by POM C:N then multiplied by wholelake volume to estimate the standing stock. The fractional increase in standing stock attributable to N 2 fixation was then calculated as: before the total mass of fixed N was estimated from standing stock of POM-N (PON ss ; or TDN): Mean summer rates of retained fixed N accumulation (mg N m -3 hr -1 ) were calculated from estimates of whole-summer fixed mass by dividing the summer total by the product of lake volume and the number of daylight hours in the growing season (mean 12 hr/day and 104 days/year).
In the second approach to estimate N 2 fixation, gross rates of N 2 fixation by cyanobacteria were calculated by applying in vitro estimates of heterocyte-specific N 2 fixation rates to microscopic determinations of in situ heterocyte density in Katepwa and Crooked lakes, following procedures developed at the Experimental Lakes Area (ELA) Higgins et al., 2017). Cell-specific and volumetric (per μl) rates of N 2 fixation were obtained from a comprehensive search of the literature of studies that quantified cyanobacterial N 2 fixation under natural conditions (Supporting Information Table S1), and were slightly higher than those used at ELA . This mean (n = 7) heterocyte-specific rate (8.845 × 10 -17 g N heterocyte -1 s -1 ) was applied to biweekly enumerations of heterocyte density in Crooked and Katepwa lakes, sites that were known to have elevated densities of N 2fixing cyanobacteria Patoine et al., 2006). As shown at ELA, such heterocyte-based methods provide estimates of N 2 fixation that are a strong linear function (r 2 > 0.90) of those derived from the commonly used acetylene reduction protocol and newer 15 N uptake techniques Higgins et al., 2017).
Finally, estimates of N 2 fixation from both NAM and heterocyte approaches were compared to each other and to literature values to assess potential bias. In this case, we focused on estimates of N 2 fixation in eutrophic systems most similar to lakes of the Qu'Appelle River drainage, as well as unpublished estimates from our study lakes based on 15 N protocols.

| Numerical analyses
Candidate predictor variables to be used in modelling N processes were selected based on 25 years of previous research in these systems Hall et al., 1999;Leavitt et al., 2006;Vogt et al., 2018).  (Marra & Wood, 2011). The analysis was made robust to outliers by assuming that the response variable followed a scaled t-distribution rather than a Gaussian distribution (Wood, Pya, & Säfken, 2016). Extreme canthaxanthin concentrations were removed from annual time series to improve spline fits in 2 lake-years.
Analysis of temporal coherence was used to assess whether N demand and estimates of fixed N were responding to similar environmental drivers. Synchrony, S, was estimated as the average of all pair-wise Pearson correlations between lakes and years, with values ranging from 1 (perfectly and positively coherent) to -1 (perfectly synchronous but out of phase) through 0 (completely asynchronous). Comparison of S values among a diverse set of environmental, physico-chemical and biological parameters in these lakes demonstrates that variables with similar degrees of temporal coherence are often subject to common regulatory mechanisms (Vogt et al., 2011).
All statistical analyses were performed using R version 3.3.3 (R Core Team, 2017).

| N demand and retained fixed N
Mean summer algal N demand varied by two orders of magnitude between <0.1 and 9.2 mg N m -3 hr -1 over 117 lake-years of study ( Figure 2 and Supporting Information Figure S1). On average, N demand was significantly lower in the mesotrophic Lake Diefenbaker  Table S2). Comparison of the N 2 fixation rates by POM with estimates of N demand by pelagic primary producers revealed that fixed N accounted for a median of 3.5% (mean ± standard deviation; 11.5 ± 21.6%) of phytoplankton requirements across all Qu'Appelle study lakes (Figure 3). Fixed N was a trivial fraction of phytoplankton N requirements in mesotrophic Lake Diefenbaker (mean < 0.0%) and N-polluted Pasqua Lake (mean = 1.1%). In contrast, fixed N contributed between 15.0% (Crooked) and 36.3% (Katepwa) of the total planktonic N demand in downstream eutrophic lakes. In these  (Donald et al., 2013;McGowan et al., 2005). Excluding N-polluted Pasqua Lake, there was a significant increase in the importance of fixed N to phytoplankton nutrition with lake longitude (α = 0.1), with a greater magnitude of fixed N in downstream eastern lakes (Supporting Information Figure S2).
Lake-specific estimates of the mass of fixed N varied from 0.0% to 12.2% of spring standing stock of N ( Figure 4). Overall, fixed N was a greater proportion of N standing stock during spring in downstream Katepwa (mean = 12.2%) and Crooked lakes (11.5%), with a significant relationship (r 2 = 0.59, p = 0.07) with landscape position if Pasqua is excluded (Figure 4 and Supporting Information Table   S2). Fixed N represented the lowest fraction of vernal N pools in mesotrophic Lake Diefenbaker (0.0%) and N-polluted Pasqua Lake (0.3%). Despite significant increases in spring N stock through time in Last Mountain, Pasqua, Katepwa and Crooked lakes (Supporting Information Figure S3), only easternmost Crooked Lake exhibited a progressive decline in the fraction of spring N attributable to retained fixed N (Supporting Information Table S2).

| Environmental predictors of N 2 fixation
Only in situ abundance of Nostocales cyanobacteria (as canthaxanthin) was retained as a significant (p < 0.001) predictor of the rates of N 2 fixation in the final GAM ( Figure 5). GAM analyses showed that N 2 fixation rates increased in a saturating relationship with the watercolumn concentration of canthaxanthin, although there was high variation in the estimated relationship at elevated Nostocales abundances ( Figure 5). No significant N 2 fixation was recorded when canthaxanthin was absent from the phytoplankton pigment assemblage.

| Comparison of isotopic-and heterocyte-based estimates of N 2 fixation
Heterocyte abundance varied with season and year in both Katepwa and Crooked lakes ( Figure 6). Heterocytes were present in all years, particularly after July (data not shown), with similar mean (± standard deviation) summer densities in Katepwa (0.31 ± 0.58 heterocytes/L) and Crooked basins (0.76 ± 2.9 heterocytes/L). Rates of gross N 2 fixation based on enumerated in situ heterocyte density and physiological estimates of heterocyte activity (Supporting Information   Table S1) ranged from 0.02 to 1.78 mg N m -3 hr -1 (0.30 ± 1.14 mg N m -3 hr -1 ), slightly lower than that observed for the NAM approaches during the same time intervals (0.66 ± 70 mg N m -3 hr -1 , range -0.01 to 2.40 mg N m -3 hr -1 ). In most lakes, NAM estimates of N 2 fixation based on changes in the mass and isotopic composition of whole F I G U R E 3 Ratio of mean summer N 2 -fixation to phytoplankton nitrogen demand (both mg N m -3 hr -1 ) for May-August of each year. Values were positive when N demand was greater than N 2 -fixation and values were negative when N 2 -fixation was negative, potentially indicating greater denitrification than N 2 -fixation. Lake organisation as Figure 2 filtered water were lower than those based on POM, but were similar to heterocyte-derived determinations in Katepwa and Crooked lakes (Supporting Information Figure S4).
Supply of fixed N to primary producers did not appear to re-establish P limitation of the phytoplankton. For example, phytoplankton biomass was correlated positively (r = .42, p < .001) with concentrations of SRP, indicating that primary producers did not deplete SRP pools in summer (Supporting Information Figure   S5a). Similarly, the degree of phytoplankton growth limitation by P in bioassays was correlated negatively (r = -.32, p < .001) with Chla concentrations (Supporting Information Figure S5b), with little evidence of strong P limitation over a wide range of Chl-a values (10-100 μg/L). Finally, there was no relationship between rates of N 2 fixation and mean annual SRP, mean spring SRP, annual summer biomass, or mean summer P limitation status from bottle bioassays (Supporting Information Figure S5c-f).

| D ISCUSS I ON
The role of fixed N in meeting phytoplankton demand, alleviating seasonal N limitation and re-establishing P limitation in lakes is contentious in modern limnology Schindler et al., 2008Schindler et al., , 2016Scott & McCarthy, 2010). Although N 2 fixation meets planktonic N demands in a small boreal lake fertilised with moderate concentrations of P Higgins et al., 2017), to this point, it has been unclear whether diazotrophic processes can meet pelagic demands in systems with up to 100-fold higher concentrations of TP (Bunting et al., 2016;Leavitt et al., 2006). Here we demonstrate that estimates of fixed N (Figures 2, 6) represent a median of 3.5% (mean 11.5%) of the nutritional N requirements of phytoplankton in 117 of lake-year of study (Figures 2, 3). The magnitude F I G U R E 4 Ratio of the total mass of fixed N 2 (kg/summer) to total dissolved N in each lake on first sampling date in May. Values were positive when N demand was greater than N 2 -fixation and values were negative when N 2 -fixation was negative, potentially indicating greater denitrification than N 2 -fixation. Lake organisation as in Figure 2  (Ferber et al., 2004;Scott & Grantz, 2013;Smith, 1983). Although rates of fixed N were comparable to gross N 2 fixation estimates recorded in other eutrophic lakes Vrede et al., 2009), diazotrophic N supply was only a small fraction of phytoplankton demand, and was insufficient to deplete water-column concentrations of SRP during summer blooms or initiate P limitation of phytoplankton growth (Donald, Bogard, Finlay, & Leavitt, 2011;Donald et al., 2015;.

| Magnitude of N 2 fixation by phytoplankton
Mean estimates of fixed N accumulation based on NAM approaches (0.31 ± 0.64 mg N m -3 hr -1 ) were similar to values observed for gross N 2 fixation determined for eutrophic lakes using acetylene reduction techniques (Higgins et al., 2017;Scott & Grantz, 2013;Torrey & Lee, 1976;Vrede et al., 2009), heterocyte densities Higgins et al., 2017;Schindler et al., 2008) (Bruesewitz et al., 2011;David et al., 2006;Seitzinger et al., 2006), including N 2 fixed by heterotrophic organisms ) and non-heterocystous fixation (Bergman, Gallon, Rai, & Stal, 1997). Further work is needed to refine estimates of N 2 fixation in aquatic ecosystems (Jankowski et al., 2012), including validation of the acetylene reduction protocols, which is known to inhibit the action of nitrogenase (Fulweiler et al., 2015). However, the close similarity of results from two NAM protocols, a heterocyte-based method, and literature estimates of acetylene reduction (Supporting Information Figure S4) strongly supports the conclusion that fixed N is only a small fraction of phytoplankton N demand that does not re-establish P limitation in these SRP-rich study lakes.

| Importance of fixed N to phytoplankton and lakes
Increased interest in the importance of fixed N in lake ecosystems arises in part because of current debate concerning the need to regulate N and P pollution to control eutrophication Schindler et al., 2016). In boreal lakes, N 2 fixation by planktonic cyanobacteria appears to compensate for experimental reduction in N influx (Schindler et al., 2008;Paterson et al., 2011but see Scott & McCarthy, 2010, 2011. However, in other regions, N influx has been demonstrated to increase lake production up to four-fold in low-and replete-P lakes, alike (Bunting et al., 2007;Deininger, Faithfull, & Bergström, 2017;Leavitt et al., 2006). In particular, our findings suggest that despite abundant diazotrophic cyanobacteria (McGowan et al., 2005;Patoine et al., 2006;Vogt et al., 2018) and rates of fixed N retention similar to fixation rates typical of other eutrophic lakes Scott & Grantz, 2013;Vrede et al., 2009), biological N 2 fixation may not alleviate N limitation in Prich lakes . In these lakes, the combination of geological substrate (Hall et al., 1999;Klassen, 1989) and over 75 years of agricultural fertilisation with P (Bennett et al., 2001;Carpenter, 2005) leads to elevated P export to lakes and rivers (Bennett et al., 2001;Carpenter et al., 1998;Stoddard et al., 2016) and has resulted in TP and SRP levels 100-fold greater than in other lake regions Vogt et al., 2018).
Several lines of evidence suggest that rates of N 2 fixation were insufficient to satisfy N demands in most years in these productive lakes (Figures 2, 3, 6). First, fixed N represented only 3.5% of total N supply to phytoplankton ( Figure 3) and <5% of spring N standing stock ( Figure 4) in most lakes and years, consistent with previous whole-lake mass budgets (Donald et al., 2015;Leavitt et al., 2006;Patoine et al., 2006) and enriched δ 15 N values of POM in all lakes (Supporting Information Figure S6). As noted by Ferber et al. (2004) Figure S5). Instead, SRP concentrations remained elevated throughout July and August in all basins except Lake Diefenbaker, the period when Chl-a and diazotrophic cyanobacteria are abundant (Donald et al., 2011;Patoine et al., 2006). Third, net N efflux (N losses exceed N 2 fixation) was common only in headwaters (Figure 3) (Hammer, 1971;Patoine et al., 2006) has been insufficient to alter the nutrient status of these lakes.
N 2 fixation alone was sufficient to completely meet total phytoplankton N demands in ca. 20% of surveyed years in Katepwa Lake, and commonly met 25% of requirements in eutrophic Crooked Lake (Figures 2, 3). However, in most other instances, including all but 4 years in hyper-eutrophic Wascana Lake, N 2 fixation rarely exceeded 5% of planktonic N demand, a value which was often statistically-indistinguishable from zero (Supporting Information Figure S7). Given the role of Nostocales cyanobacteria in the prediction of N 2 fixation ( Figure 6 and below), we infer that the nutritional demands of the phytoplankton are met by atmospheric N 2 only during years in which limnological conditions favoured extreme densities of colonial diazotrophic species (see also Patoine et al., 2006).
In principle, Nostocales and other potentially N 2 -fixing cyanobacteria are promoted by elevated water temperature (Jöhnk et al., 2008;Paerl & Scott, 2010), intense grazing by large-bodied cladocerans (Elser et al., 2000), low flushing rates (Elliott, 2010), sufficient micronutrient availability Marino et al., 2003), elevated light regimes (Ferber et al., 2004;Mugidde et al., 2003) and low N:P ratios (Levine & Schindler, 1999;Smith, 1983). In general, densities of colonial cyanobacteria in Qu'Appelle lakes are correlated positively to surface water temperature and negatively to wind speed and the El Niño-Southern Oscillation index (Vogt et al., 2018), although the high proportion of unexplained variance (>40%) suggests a high degree of site-specific variation (Dröscher et al., 2009). Taken together, these findings suggest that while N 2 fixation is rarely a substantial source of N to natural phytoplankton assemblages in Qu'Appelle lakes (Figures 2, 3), factors that selectively increase surface blooms of cyanobacteria may increase the importance of fixed N 2 in the future (Paerl & Scott, 2010).
Cross-validation with multiple methods shows that estimates of N demand and fixed N are robust for the study systems. Our bottle assays of GPP are based on standard protocols used in eutrophic systems for >70 years (Rice et al., 2012) and provide a reliable metric of lake productivity similar to those derived from whole-lake C budgets , CO 2 flux estimates derived from gaseous stable isotopes (Quiñones-Rivera et al., 2015), and Chl-a based estimates of phytoplankton standing stock McGowan et al., 2005). Similarly, community analysis using HPLC and microscopic enumeration provide highly correlated estimates of phytoplankton abundance (Donald et al., 2013) suggesting that detrital material is not a substantial component of POM in these lakes. Often, C:N ratios are elevated in detrital material (Moe et al., 2005), reflecting the preferential mineralisation of N from dead phytoplankton; however, mean C:N (mass) values for POM in Pasqua (6.0 + 1.7 standard error) and Crooked lakes (7.6 + 2.7) encompassed the expected Redfield ratio for live phytoplankton (5.7). Further, this observation means that any bias in our calculations, which used C:N ratios slightly above Redfield values, resulted in an underestimate of total N demand by phytoplankton and, thereby, reinforces our principal conclusions.
We infer that our NAM approach does not underestimate the quantity of fixed N for several reasons. First, our POM-based estimates of N 2 fixation were higher than those derived from heterocytes, even though the latter method is highly correlated to independent estimates with both acetylene reduction and 15 N uptake techniques Higgins et al., 2017). Second, our rates of fixed N accumulation in the POM are like those measured independently (0.04-4.45 mg N m -3 hr -1 ) in these Qu'Appelle Lakes during 2017 using a 15 N-uptake assay (L. Boyer and H. Baulch, University of Saskatchewan, unpublished data). Third, estimates of N 2 fixation based on changes in δ 15 N of the TDN pool provide values very similar to those derived from heterocyte determinations (Supporting Information Figure S4). Fourth, rate of hydrologic exchange (residence time) was not retained in any statistical model, making it unlikely that changes in N influx lead to an underestimation of N 2 fixation (Jankowski et al., 2012). Additionally, we note that rates of N 2 fixation in closed-basin Last Mountain Lake (Supporting Information Figure S8) were similar to those in other more highly flushed lakes. Fifth, in most lakes and years, retained fixed N is near zero and in 52% of samples the upper and lower bounds of NAM estimates include zero (Supporting Information Figure S7). Thus, while debate remains on the exact proportion of N demand required to be met by N 2 fixation to sustain P-limited growth of phytoplankton (Ferber et al., 2004), we suggest that in at least half of all 117 lakeyears, diazotrophic supply represents a trivial source of N to natural pelagic assemblages in lakes where decadal averages of summer SRP concentration routinely exceed 100 μg/L (Table 1) even during major blooms (Vogt et al., 2018).

| CON CLUS IONS
While fixed N was an occasionally important source of N to phytoplankton, there was limited evidence that N 2 fixation alleviated N deficits or initiated P limitation of phytoplankton growth in these highly P-rich ecosystems at catchment or decadal scales.
While our conclusions may not apply to oligotrophic boreal lakes in undisturbed catchments, our conclusions should generalise well to other lakes in continental landscapes with similar climatic, edaphic, and limnological characteristics-a region estimated to cover >8 × 10 6 km 2 . For example, Qu'Appelle lakes exhibit similar snowmelt water sources (Pham et al., 2009), carbon fluxes , nitrogen biogeochemistry (Bogard et al., 2012), phosphorus dynamics (Donald et al., 2015) and land-use practises (Hall et al., 1999;Pham, Leavitt, McGowan, & Peres-Neto, 2008) as 100 other lakes within a 235,000 km 2 region of southern Saskatchewan. These basins have been shown to be good models for other large regional prairie lakes, including lakes Winnipeg and Manitoba (Bunting et al., 2016;Maheaux, Leavitt, & Jackson, 2016). Finally, similar to conclusions of Higgins et al. (2017), we note that low temporal coherence (S = -0.04) and high inter-annual variation in the magnitude of N supply by fixation ( Figure 2) suggests that limnologists should be cautious in interpreting short annual or sub-decadal records as evidence of the role of N in supporting lake eutrophication.

ACK N OWLED G M ENTS
We thank members of the Limnology Laboratory for assistance with data collection since 1994. K. Hodder generated the map of our study sites. We also thank D. Program. We acknowledge that this research was conducted in Treaty 4 territory.

CO N FLI C T O F I NTE R E S T
The authors declare no conflict of interest.