An isotopic ( 1 14 C, (cid:14) 13 C, and (cid:14) 15 N) investigation of the composition of particulate organic matter and zooplankton food sources in Lake Superior and across a size-gradient of aquatic systems

. Food webs in aquatic systems can be supported both by carbon from recent local primary productivity and by carbon subsidies, such as material from terrestrial ecosystems, or past in situ primary productivity. The importance of these subsidies to respiration and biomass production re-mains a topic of debate. While some studies have reported that terrigenous organic carbon supports disproportionately high zooplankton production, others have suggested that phytoplankton preferentially support zooplankton production in aquatic ecosystems. Here we apply natural abundance radiocarbon ( 1 14 C) and stable isotope ( (cid:14) 13 C, (cid:14) 15 N) analyses to show that zooplankton in Lake Superior selectively incorporate recently ﬁxed, locally produced (autochthonous) organic carbon even though other carbon sources are readily available. Estimates from Bayesian isotopic modeling based on 1 14 C and (cid:14) 13 C values show that the average lake-wide median contributions of recent in-lake primary production and terrestrial, sedimentary, and bacterial organic carbon to the bulk POM in Lake Superior were 58 %, 5 %, 33 %, and 3 %, respectively. However, isotopic modeling estimates also show that recent in situ production contributed a disproportionately large amount (median, 91 %) of the carbon in mesozooplankton biomass in Lake Superior. Although terrigenous organic carbon and old organic carbon from resuspended sediments were signiﬁcant portions (median, 38 %) of the available basal food resources, these contributed only a small amount to mesozooplankton biomass. Comparison of zooplankton food sources based on their radiocarbon composition showed that terrigenous organic carbon was relatively more important in rivers and small lakes, and the proportion of terrestrially derived material used by zooplankton correlated with the hydrologic residence time and the ratio of basin area to water surface area.


Introduction
The role of terrigenous organic carbon in aquatic food webs is not yet well constrained. Some studies (Pace et al., 2004;Carpenter et al., 2005;Cole et al., 2006) have reported that terrigenous organic carbon supports disproportionately high zooplankton production in lakes. Others have suggested a smaller role for allochthony (the consumption of organic matter produced outside of the system of interest), and have noted that, rather, phytoplankton sustain disproportionately larger and/or most of the zooplankton production in aquatic ecosystems (Brett et al., 2009;Karlsson, 2007;Karlsson et al., 2012;Pace et al., 2007). Still others have reported seasonal shifts in the food resources supporting aquatic food webs such that autochthony (consumption of organic matter produced within the system of interest) is predominant during high within-lake phytoplankton productivity in summer, whereas allochthony (as well as heterotrophic bacterial biomass) is most important to zooplankton biomass during winter periods when within-lake primary and secondary production is minimal (Grey et al., 2001;Taipale et al., 2008;Karlsson and Sawstrom, 2009;Rautio et al., 2011).
There has been a realization that terrigenous organic matter exported from catchments is less refractory within aquatic systems than previously recognized, and can fuel microbial metabolism (Jones and Salonen, 1985;Tranvik, 1992), and that even the ancient (old according to radiocarbon measurements) component traditionally thought to be more recalcitrant could support bacterial (Cherrier et al., 1999;Petsch et al., 2001;McCallister et al., 2004), zooplankton (Caraco et al., 2010), and fish production (Schell, 1983). Accordingly, the notion that terrestrial carbon partially sustains food webs in aquatic systems has gained currency in the past few decades (Salonen and Hammar, 1986;Hessen et al., 1990;Meili et al., 1993;Pulido-Villena et al., 2005;Cole et al., 2011). Terrigenous carbon could be introduced and accumulated in aquatic food webs by zooplankton directly feeding on terrestrially derived detrital particles (Hessen et al., 1990;Cole et al., 2006;Brett et al., 2009), and/or feeding on heterotrophic organisms that consume terrestrially derived organic carbon (Jones, 1992;Lennon and Pfaff, 2005;Berggren et al., 2010).
Several studies in lakes have concluded that terrigenous food can support aquatic animal consumers (Cole et al., 2006;Karlsson and Sawstrom, 2009), and the relative importance of allochthony in lakes is thought to relate to factors such as lake color (indicating the amount of humic material present), trophic status, and size. Therefore, allochthony should be higher in small humic lakes, and lower in eutrophic lakes and/or clear-water lakes with less terrestrial influence on organic matter cycling (Jones, 1992;Pace et al., 2007;Cole et al., 2011). The relative significance of these factors has been difficult to test as neither lab-based studies (Salonen and Hammar, 1986;Brett et al., 2009), small-scale in situ enclosure studies (Hessen et al., 1990), nor whole-lake 13 Clabeled bicarbonate addition approaches (Cole et al., 2002(Cole et al., , 2006Carpenter et al., 2005;Pace et al., 2007;Taipale et al., 2008) are easily applied to large-lake or marine systems. Also, the use of whole-lake 13 C labeling techniques for estimating the proportion of terrigenous organic carbon supporting zooplankton in lakes is limited or challenged by the fact that unlabeled food particles incorporated by zooplankton could possibly be from metalimnetic phytoplankton or phytoplankton-derived material predating label introduction rather than from terrestrial sources (Brett et al., 2009). Further, quantification of zooplankton food sources using ambient stable carbon isotopic signatures is difficult because of the inherent difficulty in directly measuring the δ 13 C of phytoplankton, and the narrow and overlapping range of phytoplankton and terrigenous organic matter δ 13 C signatures, especially in freshwater systems (Hamilton et al., 2005). The dynamic range of 14 C (−1000 to ∼ +200 ‰) is much greater than that of δ 13 C in organic carbon (−32 to −12 ‰) (Petsch et al., 2001;McCallister et al., 2004;Wakeham et al., 2006), and provides a more sensitive means for differentiating the sources of organic carbon in the particulate organic matter (POM) matrix and organic carbon sustaining zooplankton secondary production. Also, while both δ 13 C and 14 C are linear quantities that can be used for isotopic mixing models, 14 C has the added advantage of being the same for consumers and their food source in a modern ecosystem (as the 14 C calculation corrects for biochemical fractionations) thereby eliminating the need for fractionation correction along trophic levels as is the case for δ 13 C (and δ 15 N).
In this study we examine the possible food sources of mesozooplankton in Lake Superior, the world's largest freshwater lake by surface area (Herdendorf, 1990), using natural abundance radiocarbon distributions. Recent investigations of Lake Superior, an oligotrophic system with low nutrient concentrations and primary productivity and a pronounced deep-chlorophyll maximum (Russ et al., 2004;Barbiero and Tuchman, 2004), have concluded that the lake appears to be net heterotrophic (McManus et al., 2003;Cotner et al., 2004;Russ et al., 2004;Urban et al., 2004Urban et al., , 2005. Terrigenous and resuspended sedimentary organic carbon sources have radiocarbon signatures that are unique and different from those of the lake's dissolved inorganic carbon and recently fixed primary production, hence providing the opportunity for better understanding the role of these possible food sources in mesozooplankton production and food web dynamics in the lake. We exploit the natural abundance of radiocarbon ( 14 C), stable isotope (δ 13 C and δ 15 N), and elemental compositions (atomic C : N ratio) of mesozooplankton to assess the role of different carbon sources in supporting mesozooplankton production, thereby providing a clearer picture of food web dynamics in Lake Superior. We also assess the putative food sources of zooplankton in a suite of other aquatic systems (riverine, smaller lakes, and oceanic) for a broader-scale understanding of zooplankton food sources in aquatic food webs.

Sampling
Cruises were undertaken on the R/V Blue Heron to sample Lake Superior in May-June and August-September 2009 during isothermal (mixed) and thermally stratified water conditions, respectively. Site locations, water depths, and sampling depths are given in Fig. 1 Williams (1990, 1991) 3 William et al. (1987) and Williams (1990) 4 William et al. (1987) and Druffel et al. (1996) 5 McCallister and del Giorgio (2008)    collected mesozooplankton using 50 m vertical tows through the water column using a 300 µm plankton net. At each of the nearshore stations (ONT and BR), the depth of tow was modified to a maximum depth of 4 to 10 m above the sediment water interface. The biomass was rinsed with lake water into the cod end of the net, and duplicate samples were filtered onto glass-fiber filters (precombusted GF/F filters, 0.7 µm pore size), and stored frozen. Although we did not separate mesozooplankton into different groups in this study, a recent survey in the lake shows that copepods are the most dominant zooplankton in the surface waters of offshore Lake Superior (Yurista et al., 2009). In this extensive study at 31 sites over a 3-yr period, Yurista et al. (2009) reported ∼ 90 % (by biomass) of the crustacean zooplankton in the offshore sites (> 100 m water depth region) were copepods, and most of these (∼ 80 %) were concentrated in the surface 50 m of the lake water column, which is the depth over which we sampled our zooplankton in the offshore lake. Within the copepods, the taxa calanoids, dominated by Diaptomus copepodites and Limnocalanus macrurus, were more abundant in the lake than the cyclopoids, which were mostly Cyclops bicuspidatus thomasi and Cyclops copepodites (Yurista et al., 2009). The calanoid copepods contributed ca. 70 % of the biomass of crustacean zooplankton in Lake Superior ( (Yurista et al., 2009). Sediment cores were taken from the open lake sites using an Ocean Instruments multi-corer. Recovered cores were sectioned at 2 cm resolution and kept frozen until further analysis, and the surface sediments (top 0-2 cm inclusive of the flocculant layer) were used in this study.
We collected dissolved inorganic carbon (DIC), dissolved organic carbon (DOC), and particulate organic carbon (POC) samples from surface waters (≤ 5 m water depth). Water samples were drawn using twelve 8 l Niskin bottles mounted on the CTD rosette. DIC samples were collected directly from the Niskin bottles via pre-cleaned (10 % HCl v/v, then ultra pure water (Millipore Milli-Q Plus)) silicone tubing into previously acid-cleaned and combusted (450 • C for 4 h) 500 ml amber Pyrex bottles. The bottles for DIC were rinsed three times with sample and then overflowed with two volumes of the unfiltered water. As quickly as possible after collection, a small aliquot of water was removed, and the samples were preserved with saturated mercuric chloride solution, sealed airtight with glass stoppers coated with Apiezon M grease and stored at room temperature in the dark. POC and DOC samples were obtained by filtering lake water through precombusted Whatman GF/F glass fiber filters (450 • C for 4 h; 0.7 µm nominal pore size) via nitrogen-pressurized stainlesssteel canisters. Approximately 1 l of DOC sample from the resulting filtrate was collected into an acid-cleaned and combusted glass bottle, acidified to pH 2 using 6 N HCl (American Chemical Society Plus grade) and refrigerated. After ∼ 10 l of lake water had passed through a GF/F filter, the filter with retained particulate matter (POC sample) was placed in previously combusted aluminum foil and stored frozen until analysis.
For comparison with our Lake Superior study, we collated similar data for a suite of aquatic ecosystems of various sizes and residence times. Data from five northern small lakes sampled between June and September 2004 in southern Quebec were adapted from McCallister and del Giorgio (2008); these small lakes include Bran-de-Scie, Des Monts, Stukely, Bowker, and Fraser Lakes. Zooplankton biomass and water samples for DOC, DIC, and POC and their isotopic signatures were collected at a depth of 0.5-1.0 m using a diaphragm pump connected to an acid-rinsed (10 % HCl) plastic hose (McCallister and del Giorgio, 2008). Zooplankton were collected by passing at least 200 l of water through a 50 µm mesh screen, subsequently washed from the screen and stored overnight in deionized water at 4 • C to evacuate gut contents prior to isotopic analysis (McCallister and del Giorgio, 2008). The zooplankton samples were dominated by cladocerans and copepods. Cladocerans were primarily comprised of the genus Daphnia, most notably by Daphnia mendotae and Daphnia catawba, while copepods were dominated by Diacyclops bicuspidatus, Mesocyclops edax, and Leptodiaptomus minutus. We also collated existing data from the Pacific Ocean (including Pacific coastal ocean, North Central Pacific, and North Eastern Pacific sites) and the Hudson River (eastern New York, USA). Data from the Hudson River were adapted from Caraco et al. (2010). Pacific Ocean zooplankton data included crustaceans and fishes, and were adapted from William et al. (1987), Druffel and Williams (1990Williams ( , 1991, and Druffel et al. (1996).

Radiocarbon ( 14 C) analysis
14 C measurements for Lake Superior samples were performed at the National Ocean Sciences Accelerator Mass Spectrometry Facility (NOSAMS) at the Woods Hole Oceanographic Institution (WHOI). POC, zooplankton, and sediments were freeze-dried, acid fumigated (12 N HCl) overnight, and redried, and then converted to CO 2 by combustion in a modified Carlo Erba NA 1500 elemental analyzer (Zigah et al., 2011). DOC and DIC samples were converted to CO 2 by ultraviolet irradiation and phosphoric acid volatilization, respectively. The evolved CO 2 was cryogenically separated and reduced to graphite with H 2 over Fe catalyst (Zigah et al., 2011). The graphite produced was analyzed by accelerator mass spectrometry (AMS) along with primary and secondary standards, and combustion and graphitization process blanks.
Radiocarbon values are reported as 14 C, the part per thousand deviation of the sample's 14 C : 12 C ratio relative to a 19th century wood standard that has been corrected to the activity it would have had in 1950 and a δ 13 C of −25‰.
14 C was corrected for fractionation using δ 13 C of samples according to the convention of Stuiver and Polach (1977). Instrumental precision of the 14 C analysis is based on the error of standards or multiple analyses on a target.

Stable isotopes (δ 13 C and δ 15 N) and C : N analysis
Carbon and nitrogen contents of bulk POM and zooplankton were measured on a Costech ECS 4010 elemental analyzer (EA) coupled to a Finnigan DELTAplus XP isotope ratio mass spectrometer (IRMS) at the Large Lakes Observatory (LLO). δ 13 C of samples (DOC, POC, sediment OC and zooplankton) were determined at NOSAMS using an Optima IRMS on subsamples intended for radiocarbon analyses. Stable nitrogen isotope ratios (δ 15 N) and a set of samples for δ 13 C of POM and δ 13 C of zooplankton were measured at LLO using a Finnigan DELTAplus XP IRMS with Conflo III interface (Thermo Fisher Scientific Inc., Waltham, MA) coupled to a Costech ECS 4010 EA. Typical instrumental precisions of δ 15 N and δ 13 C based on analyses of multiple external standards were 0.17‰ and 0.2‰, respectively. The stable isotope ratios ( 13 C : 12 C and 15 N : 14 N) are reported as δ 13 C and δ 15 N, respectively, which are the per mil difference relative to Vienna Pee Dee Belemnite carbonate and atmospheric air standards.

Bayesian MixSIR mixing model for multiple endmembers
The Bayesian isotopic modeling software MixSIR (Version 1.04) (Moore and Semmens, 2008;Semmens et al., 2009) was used to partition the proportional contributions of potential OC sources to the bulk POC and to zooplankton diet based on their 14 C and δ 13 C signatures. The MixSIR model works by determining probability distributions of sources contributing to the observed mixed signal while accounting explicitly for the uncertainty in the isotopic signatures of the sources and fractionation. The uncertainties of δ 13 C and 14 C values used for modeling here are the analytical uncertainties based on analyses of multiple external standards or multiple analyses of graphite targets in the case of 14 C. Since isotopic fractionation is already corrected for in the calculation of 14 C values, radiocarbon fractionation was not used in the model (thus specified as zero). Isotopic fractionation of +1 was used for δ 13 C (Fry and Sherr, 1984). Prior information was not used in the model; hence, all possible source combinations were equally possible contributions to the observed mixed signal. The number of iterations used was 10 000 000 (and 100 000 000 when the posterior draws were less than 1000). For each potential source, we report the median and the 5 % and 95 % confidence percentiles estimates of the proportional contribution of the sources to the measured (observed) value.

Choice of endmembers
To determine carbon sources to POC and food sources supporting mesozooplankton biomass, we chose isotopic endmembers based on identifiable unique sources of OC to the POC pool in the lake (Zigah et al., 2011(Zigah et al., , 2012. Because our modeling is based upon natural abundance stable carbon and radiocarbon distributions, these endmembers vary from those generally used in labeling experiments (e.g., Taipale et al., 2008) or natural abundance stable isotope modeling (e.g., Karlsson et al., 2003). Based upon their unique 14 C values, the potential components of POC in the lake include carbon derived from recent photosynthesis (here described as "algal carbon" although it also includes herbivore biomass supported by recent primary production), bacterial OC, terrestrial OC, and resuspended-sediment OC. As this is a novel suite of endmembers relative to previous work, we discuss our rationale further below.
Lake-wide primary production in Lake Superior is estimated at 9.73 Tg C per year (Sterner, 2010), although most of the POC input from phytoplankton photosynthesis is thought to be mineralized rapidly (Maier and Swain, 1978;Urban et al., 2005) and does not persist in the lake. The POC pool in the lake is only ∼ 1 Tg C (compared to ∼ 15 Tg DOC and ∼ 122 Tg DIC; Zigah et al., 2012). We assigned a δ 13 C value of −30 ± 1 ‰ as representative of algal C (Sierszen et al., 2006). We used a 14 C of DIC as the 14 C of al-gal carbon from recent photosynthesis as DIC-incorporation is the starting point for algal biomass production (McNichol and Aluwihare, 2007;McCarthy et al., 2011). For calculating 14 C values, measured radiocarbon values are normalized to remove mass-dependent isotopic fractionation such that 14 C values reflect only time (age) or mixing (variably aged components). To normalize the sample, fractionation between 14 C and 12 C is assumed to be approximately twice that between 13 C and 12 C since the mass difference between 14 C and 12 C is twice that between 13 C and 12 C (Donahue et al., 1990;McNichol and Aluwihare, 2007). Therefore, in a modern system, the 14 C of algal carbon tracks that of DIC that was incorporated.
Bacterial carbon is another identifiable component of POC in the lake (Cotner et al., 2004). For 14 C and δ 13 C of bacterial carbon, we used the 14 C and δ 13 C of DOC, the main microbial food source, as we do not have direct measurement of bacterial biomass 14 C and δ 13 C. We acknowledge that this is only a first-order approximation of the 14 C and δ 13 C of bacteria in Lake Superior, and look forward to further refining this model endpoint when better data become available.
Radiocarbon values of bulk POC in the lake suggest that they contain a pre-aged carbon source that may result from sediment resuspension and further that this resuspension can impact surface water samples as well as deeper samples (Zigah et al., 2011(Zigah et al., , 2012. This finding is consistent with previous studies showing the importance of sediment resuspension in Lake Superior (Urban et al., 2005;Churchill and Williams, 2004;Flood, 1989;Flood and Johnson, 1984). In our Lake Superior work (see below), the OC in the surface sediments (0-2 cm) at the various study stations across the lake has 14 C values that are older ( 14 C-depleted) than recent algal OC from lake photosynthesis. The physical mechanisms transporting such old OC from the lake sediments into the lake surface water are not well known. However, Lake Superior is dimictic; thus, there is density-driven vertical mixing of the water column twice each year. Hence, organic materials resuspended into the deep waters due to strong bottom currents could be introduced into the surface waters during the lake overturn. In our lake surface (top 0-2 cm) sediment samples, the δ 13 C values of OC were −27.3 ‰, −26.9 ‰, −27.0 ‰, −26.2 ‰, and −27.0 ‰, and the 14 C values were −20±3 ‰, −37±2 ‰, −23±2 ‰, −117±2 ‰, and −36 ± 2 ‰ at sites CM, SM, NM, WM, and EM, respectively, and the corresponding site-specific values were used as the sediment OC endmember for each site in the lake. We note that this endmember assignment is a first-order approximation as lateral advection of old OC from shallower depths, especially at the WM site, is also possible.
The choice of endmember for terrestrial OC was challenging because there are both "old" and "recent" terrestrial OC components. The radiocarbon value of recently synthesized terrestrial OC reflects the radiocarbon value of atmospheric CO 2 . We determined the radiocarbon value of 3668 P. K. Zigah et al.: POM composition and zooplankton food sources atmospheric CO 2 using the radiocarbon content of corn leaves (annual plant) collected in the western watershed of the lake ( 14 C = 38 ± 2 ‰, Zigah et al., 2011). While this approach does not cover the entire watershed of the lake, we do not think there would be considerable differences across the basins because most variations in atmospheric 14 C occur at a larger spatial scale. The remoteness of the lake from big industrial plants or big cities, the uniformity of surface 14 C-DIC across the lake, and the absence of considerable soot (black) carbon in the POC pools across the lake (Zigah et al., 2012) suggest little regional variation in atmospheric 14 C around Lake Superior. Terrigenous POC δ 13 C of −27.3 ‰ (Zigah et al., 2011) was used in the model. To account for the fact that terrestrially produced OC could spend some time in the soil before delivery to the lake, we performed additional model runs replacing the corn 14 C value with that of POC collected during high flow conditions in June 2008 from Amity Creek ( 14 C = 11 ± 2 ‰; δ 13 C = −27.3 ‰, Zigah et al., 2011), a primarily forested watershed north shore stream which drains into western Lake Superior. The choice of highflow data was because most terrestrial influx in streams and rivers occurs during storm flows. While we note that using data from one stream within the watershed might not be representative, the similarity of high flow Amity Creek POC 14 C to nearshore POC 14 C ( 14 C range of 7-17 ‰) from both the southern and northern nearshore regions of the lake that we sampled implies that our terrestrial endmember POC-14 C is a good first approximation.

Zooplankton allochthony based on 14 C
Zooplankton allochthony in Lake Superior was estimated using a binary (terrigenous and autochthonous) mixing model as follows: where f is the fraction of terrestrial OC in the zooplankton biomass, (1 − f ) the fraction of algal-derived carbon in the zooplankton biomass, and the subscripts T err and Algal refer to terrestrial and algal-derived, respectively. We used 14 C of DIC as the algal-derived OC endmember. For the terrestrial endmember, we used the atmospheric CO 2 14 C and 14 C of POC from high flow Amity Creek in separate model runs for sensitivity analysis.

Statistical analyses
We used SigmaPlot 9.0 (Systat Software Inc., San Jose, California, USA) for all statistical analyses. Relationships among samples were tested via correlation analyses in which case we report the Pearson's correlation coefficient (r), probability (p), and number of samples (n). The difference between isotopic composition of zooplankton and that of DIC, POC and DOC was determined using paired t-tests, and for these we reported the two-tailed probability value (p), and the number of samples (n). Significance difference or correlation was tested at 95 % confidence level (α = 0.05).

Lake Superior isotopic distributions
The bulk POC in the lake (including both stratified and isothermal surface samples) had a mean 14 C value of 10 ± 29 ‰ (range −55 ‰ to 39 ‰, n = 14) ( Fig. 2a and b; Table 2), and the 14 C of DOC in the lake was 38 ± 21 ‰ (range −10 ‰ to 74 ‰, n = 13) ( Fig. 2a and b). 14 C of DIC varied from 36 ‰ to 38 ‰ at NB and ONT sites (both nearshore regions) to 62‰ at CM and SM (both offshore regions) ( Fig. 2a and b). At each site 14 C of mesozooplankton and 14 C of DIC were similar ( Fig. 2a and b), and a paired t-test showed no significant difference in their values (p = 0.96, n = 13). In contrast, 14 C of mesozooplankton was significantly more positive (thus, 14 C-enriched) than 14 C of either POC (t-test, p < 0.0001, n = 13) ( Fig. 2a and b) or DOC (t-test, p = 0.03, n = 13) ( Fig. 2a and b).
The δ 15 N and δ 13 C values of consumers reflect both the isotopic composition of the incorporated food plus biochemical fractionations. Movement across trophic levels imposes additional fractionation on the resulting biomass, with consumers exhibiting 13 C-enriched values of ∼ 0.5-1 ‰ (Fry and Sherr, 1984) and 15 N-enriched values of 2-3 ‰ (Fry, 1991) relative to their food source. In Lake Superior, the mesozooplankton were 15 N-enriched by an average of ∼ 4 ‰ relative to bulk POM during isothermal condition, and 15 Nenriched by ∼ 6 ‰ relative to bulk POM during stratification (Table 2). In contrast to δ 15 N values, mesozooplankton were 13 C-depleted by an average of ∼ 1 ‰ relative to bulk POM during stratification in August (Table 2)    The C : N values of mesozooplankton ranged from 5.8 to 8.7 (mean 7.1, n = 14), and were consistently and significantly lower than the C : N values (C : N, mean 8.2, range 7.0 to 9.5) of bulk POM (t test, p = 0.001, n = 14) ( Table 2).

POC sources
The contribution of potential source materials to the bulk POC was estimated using the Bayesian MixSIR model based on source 14 C and δ 13 C signatures. Based on the model results, the median (and 5 % and 95 % confidence percentiles) contribution of algal carbon to the bulk POM varied from 10 % (5-14 %) at the EM site during isothermal condition to 85 % (77-93 %) at the NM site during isothermal condition (Table 3). The median contribution of terrestrial carbon to bulk POM ranged from 1 % (0.1-5 %) at EM site during isothermal condition to 19 % (2-47 %) at the WM site during stratification (Table 3). Sedimentary OC influence on bulk POM varied from a median of 10 % (2-15 %) during stratification at WM site to 87 % (84-91 %) during isothermal condition at EM site (Table 3). The average lake-wide (including both seasons) median contributions of algal, terrestrial, sedimentary, and bacterial OC to the bulk POM were 58 %, 4 %, 34 %, and 2 % (Table 3), and the corresponding values with creek POC as terrestrial endmember were 58 %, 6 %, 32 %, and 3 %, respectively (Table 3). It should be noted that these model estimates change to 23 %, 28 %, 15 %, and 25 %,  respectively, if only 14 C values (rather than both 14 C and δ 13 C) are used in the MixSIR modeling.

Sources of mesozooplankton diet
After estimating the relative contributions of potential basal food resources to the bulk POM, we used the Bayesian MixSIR mixing model to evaluate the relative contributions of these basal foods to mesozooplankton production in Lake Superior. Algal-based food contributed a disproportionately large amount to the mesozooplankton biomass in Lake Superior, with a median contribution (average of both isothermal and stratified season at all sites) of 91 % (range: 85-95 %).
If only 14 C values are used in the MixSIR modeling, the algal-based contribution changes to 40-89 %. Based upon MixSIR modeling with both δ 13 C and 14 C as inputs, mesozooplankton in Lake Superior also appear to gain a lake-wide median of 3 % (2-4 %) of their biomass carbon from consumption of bacterial biomass (Table 4). Although making up a considerable portion of the bulk POC in the lake, OC from the sediment and terrestrial POC contributed minimally (median: 3 % each) to mesozooplankton carbon (Table 4). Based on a two-endmember mixing model using 14 C values representing recent in-lake primary production and terrestrial inputs as the endmembers, mesozooplankton allochthony varied across the lake sites (Table 5), ranging from 0-54 % (with corn leaves used as the terrestrial endmember) or 0-25 % (with creek POC as the terrestrial endmember). The mesozooplankton autochthony estimates from both mul-tiple endmember, multiple isotope (Table 4) and binary endmember radiocarbon-based (Table 5) models were comparable with a lake-wide average offset of ∼8 % or 3 % if the isothermal SM site with large difference is excluded. The offset is only ≤ 1 % when the creek POC is used as the terrestrial endmember. The mesozooplankton allochthony estimate from the binary model varied from the terrestrial contribution from the multiple endmember Bayesian model with a lakewide average offset of ∼ 18 % or 11 % (if isothermal SM is excluded). The offset, however, is ≤ 6 % when creek POC is used as terrestrial endmember.

Cross-system comparisons of isotopic distributions (Hudson River, small lakes, Lake Superior, and the Pacific Ocean)
Zooplankton in the Hudson River had pre-aged radiocarbon content ( 14 C = −236 ‰; Caraco et al., 2010), and were 14 C-depleted relative to recent terrestrial OC, algae (based on 14 C DIC ), POC, and DOC (Table 6). The 14 C of zooplankton in small lakes (Bran-de-Scie, Des Monts, Stukely, Bowker, and Fraser Lakes) ranged from −2 ‰ in Bowker Lake to 40 ‰ in Des Monts Lake (Fig. 3a), and was consistently 14 C-depleted relative to concurrent POC, DOC, and DIC (with the exception of Fraser Lake with a DIC 14 C of −91 ‰ vs. a zooplankton 14 C of 10 ‰, Table 6; Fig. 3a). It is evident from 14 C DIC-Zoop ( 14 C DIC − 14 C Zoop ) vs. 14 C POC-Zoop ( 14 C POC − 14 C Zoop ) that the 14 C values of zooplankton in the small lakes are more similar to   .88 (0.80-0.95) 0.04 (0.003-0.12) 0.05 (0.01-0.11 (Table 6). In the coastal Pacific Ocean, 14 C DIC of 100 ‰ and 14 C POC of 100 ‰ were identical and slightly 14 C-enriched compared to 14 C zoop of 76 ‰ ( Fig. 3b; Table 6). A similar trend was observed in the north central Pacific Ocean where 14 C zoop of 124 ‰ was slightly 14 C-depleted relative to the 14 C DIC of 132 ‰ and 14 C POC of 139 ‰ ( Fig. 3b; Table 6). In contrast, the zooplankton in the northeast Pacific Ocean had 14 C values identical to 14 C of   Williams et al., (1987), Williams, (1990, 1991), Druffel et al., (1996).  Druffel et al. (1996) POC, but different from the 14 C of DIC as evident in the 14 C DIC-Zoop and 14 C POC-Zoop of 69 ‰ and −4 ‰, respectively ( Table 6).
The entire dataset was pooled to assess inter-system trends (thus small-to-large water body ecosystems) in 14 C zoops vs. 14 C DIC , and 14 C zoops vs. 14 C POC . There was a positive correlation between 14 C values of zooplankton and DIC (excluding Hudson River) in the pooled 14 C data (r = 0.82, p < 0.0001, n = 14) (Fig. 4a), with ∼ 67 % of the variation in 14 C of zooplankton accounted for by the changes in 14 C of phytoplankton utilizing in situ DIC (based on coefficient of determination (r 2 ) of 0.67, Fig. 4a). In contrast, 14 C of zooplankton was not correlated to 14 C of bulk POC (r = 0.03, p = 0.92, n = 14) (Fig. 4b).
There was a positive correlation between 14 C DIC-Zoop and the ratio of basin area to lake surface area (correlation, r = 0.88, p = 0.047) (Fig. 5a). Although marginally significant (correlation, r = −0.84, p = 0.078), there was a negative correlation between the hydrological residence time of the lakes and 14 C DIC-Zoop (Fig. 5b).

Composition of bulk POM, and putative food sources for consumers in Lake Superior
Isotopic signatures of baseline food resources can be used to assess their relative importance in the diet of their animal consumers. Food-source tracking using isotopic signatures works if a measurable contrast exists between the potential food resources. Mesozooplankton in Lake Superior could obtain their diet from recent primary production (perhaps cycled through an additional small herbivore first), bacterial biomass, terrestrial OC, or OC from resuspended sediments.
Our results indicate that, in Lake Superior, the proportional median contribution of recent primary production to bulk POC was 58 % (Table 3). This is not surprising for a large cold temperate oligotrophic lake with low levels of autochthonous primary production (Cotner et al., 2004;Sterner, 2010). Although the estimated annual lake-wide primary production is 9.73 Tg C (Sterner, 2010), the OC input from phytoplankton photosynthesis is thought to be mineralized rapidly (Maier and Swain, 1978;Urban et al., 2005) and does not persist in the lake. Consequently, the POC pool in the lake is estimated at only ∼ 1 Tg C (compared to ∼ 15 Tg DOC and ∼ 122 Tg DIC; Urban et al., 2005;Zigah et al., 2012).
Our model estimates show that the combined proportions of terrestrial OC and resuspended-sediment OC can constitute a considerable fraction of the basal food available to consumers in the lake (Table 3). These estimates are consistent with published values from previous studies in the lake. Urban et al. (2004) reported that resuspended sedimentary OC contributed 10-35 % of OC in sinking POC off the Keweenaw Peninsula at the depth of 25-35 m in the lake.

Stable isotopes and C : N ratios
Although bulk POM was 13 C-enriched and 15 N-depleted during stratification in August relative to isothermal conditions in June, the mesozooplankton in the lake did not exhibit such  seasonal changes in their δ 13 C and δ 15 N signatures (Table 2). Mesozooplankton in the lake were generally 13 C-depleted relative to the bulk POM, especially in the productive surface waters during stratification in August. Such 13 C-depletion of mesozooplankton compared to bulk POM has been reported by several researchers (Del Giorgio and France, 1996;Karlsson et al., 2003;Pulido-Villena et al., 2005;Matthews and Mazumder, 2006;McCallister and del Giorgio, 2008), and suggests that the mesozooplankton were primarily supported by a subsurface algal food with 13 C-depleted values, or a baseline algal food source within the surface POM with 13 C-depleted values, as typical trophic-level enrichments for δ 13 C are +0.5 to +1 ‰ (Fry and Sherr, 1984). Another possibility is the accumulation and/or storage of lipids by the mesozooplankton from their food, thus making their en-  Although marginally significant correlation (p = 0.08) is observed between 14 C DIC-Zoop and hydrologic residence time, these two plots generally illustrate that zooplankton support by allochthonous organic carbon is related to variables that indicate terrestrial influence. tire biomass or whole body more 13 C-depleted than their food source as lipids are more 13 C-depleted than other biochemicals in their biomass (Deniro and Epstein, 1978;Mc-Connaughey and McRoy, 1979;Kling et al., 1992;Matthew and Mazumder, 2005;Smyntek et al., 2007). Mesozooplankton in Lake Superior do exhibit an increase in C : N values during stratification in August relative to isothermal conditions in June, which is consistent with increasing accumulation and storage of lipids during the more productive and warmer season (McConnaughey and McRoy, 1979;Kiljunen et al., 2006;Smyntek et al., 2007;Syvaranta and Rautio, 2010).

Radiocarbon-based estimation of mesozooplankton food sources in Lake Superior
Taken together, and without seasonal comparison, the stable C and N isotope values do not distinguish mesozooplankton from the bulk POM pool from which they feed, especially during isothermal conditions in the lake. Adding radiocarbon information allows for the estimation of mesozooplankton dependence on food sources other than that year's in situ primary production (and its immediate consumers), and also significantly refines the relationship between POM and mesozooplankton. Mesozooplankton in Lake Superior in both isothermal and stratified conditions have 14 C values that track those of co-occurring DIC rather than bulk POM (Fig. 1), indicating that the mesozooplankton in this system are preferentially feeding on food resources resulting from contemporary photosynthesis rather than indiscriminately upon bulk POM. Bayesian MixSIR modeling results generally show that most of the mesozooplankton biomass in the entire lake, and in both seasons (medians 85-95 %; Table 4), came from incorporation of recent primary production. These results are generally consistent with mesozooplankton autochthony estimates from binary isotopic mixing modeling with the exception of SM site during isothermal conditions (range 61-100 % or 75-100 % depending on choice of terrestrial endmember as shown in Table 5). Both approaches show considerable enrichment in mesozooplankton biomass relative to the proportion of "algae" in bulk POC (median, 58 %; Table 3). That algal carbon dominantly supports mesozooplankton biomass production was not surprising as algal-derived food is generally known to be labile and the most preferred food option for secondary producers (Brett et al., 2009). Our results agree with previous studies in other lakes (Del Giorgio and France, 1996;Cole et al., 2002;McCallister and del Giorgio, 2008;Mohammed and Taylor, 2009;Karlsson et al., 2012) and rivers (Sobczak et al., 2002;Thorp and Delong, 2002;Meersche et al., 2009) that reported that zooplankton were sustained disproportionately and/or largely by phytoplankton biomass.
Mesozooplankton dependence on organic carbon subsidies (sedimentary and terrestrial OC) in Lake Superior was small (Table 4), although these organic carbon resources make up a considerable fraction of the bulk POC in the lake (Table 3). Contrary to our results, other studies have reported larger use of non-algal food by zooplankton in some aquatic systems based on either natural abundances of 14 C (Schell, 1983;Caraco et al., 2010), δ 13 C and δ 15 N (Meili et al., 1996;Jones et al., 1998;Karlsson et al., 2004;Matthews and Mazumder, 2006), or whole lake addition of 13 C-labeled bicarbonates (Carpenter et al., 2005;Pace et al., 2007;Taipale et al., 2008).
Differently aged components (modern vs. ancient) of organic carbon subsidies may have different fates in aquatic ecosystems. The relative ages of the non-algal OC that sup-ports heterotrophic microbial communities and the upper trophic levels of food webs are not well known, although this knowledge is essential in understanding food web dynamics. In Lake Superior, although pre-aged organic carbon from the sediment was a putative food option in the lake, and constituted a median proportion of as much as 87 % (84-91 %) of the available food carbon (POC) during isothermal (mixedlake water) condition at EM site and 84 % (78-87 %) during stratified condition at SM site (Table 3), mesozooplankton in the lake only incorporated trace amounts (median: 3 % (1-7 %)) of this old carbon into their biomass (Table 4). This observation could be due to a general decrease in palatability of considerably aged organic carbon or could be due to the extensive amount of reworking this material has experienced in Lake Superior. Some studies have suggested that modern terrestrial organic carbon supports heterotrophic respiration (Mayorga et al., 2005), whereas ancient terrestrial components could be important food sources for heterotrophic microbes (McCallister et al., 2004) and animal consumers (Ishikawa et al., 2010) in certain aquatic systems. In contrast to Lake Superior, studies of the Hudson River food web (Caraco et al., 2010) and bacterial biomass production in the Hudson and York River systems (McCallister et al., 2004) have shown that both mesozooplankton and bacteria can use considerably aged reduced carbon as a food source. Also, in the open ocean in eastern North Pacific, radiocarbon studies show that bacteria assimilate both modern and ancient organic carbon (Cherrier et al., 1999). Schell (1983), in a study of the Colville River and coastal Alaskan Beaufort Sea, reported that old carbon from peat in the catchment was introduced primarily into food webs in the freshwater portions of the system, i.e., anadromous fish and ducks feeding in these areas. While it is still not clear which aquatic variables drive the relative utilization of ancient vs. modern food sources in these systems, some studies have indicated that terrestrial materials from the catchment are less refractory than previously thought (Hessen, 1992;Tranvik, 1992), and others have suggested the addition of newly synthesized algal food could act as co-metabolic primer facilitating the use of the aged (potentially refractory) organic material (Horvath, 1972;Mc-Callister et al., 2004;Goni et al., 2006;Aller et al., 2008).
It should be noted that, while this study is one of the most extensive isotopic (particularly radiocarbon) investigations of the ecosystem of any of the great lakes in the world, our results represent a general large-scale view of the ecosystem functioning of Lake Superior since the spatial and temporal coverage of this study is limited to 8 sampling sites covering nearshore and offshore locations, and visited twice, during thermal stratification and mixed-lake condition. A high resolution spatial and temporal sampling scheme would be needed for a more detailed understanding of the feeding habits and ecology of the mesozooplankton in the lake.

Comparison of zooplankton food sources in small-to-large aquatic systems
To gain cross-system insight into the food sources supporting animal consumers in aquatic systems, we compared the food sources of zooplankton in the Hudson River, five separate small northern lakes, and different sites in the North Pacific Ocean, to the food resources supporting zooplankton in a large lake (Lake Superior). This cross-system dataset is not representative of global lake diversity and/or variability and is only from North America as we cannot find radiocarbon compositions of zooplankton in aquatic systems in other parts of the world. Trends observed and discussed here give a broad picture of ecosystem functioning across lake size gradient in the US and Canada. A more globally distributed dataset is needed to ascertain whether the trends observed in this study are consistent with the global view of the relationship between lake size and zooplankton ecology. In Bran-de-Scie, Des Monts, Stukely, and Bowker Lakes, the zooplankton biomass was generally largely supported by in-situ primary production (and its immediate consumers) as evidenced by smaller values of 14 C DIC-Zoop relative to 14 C POC-Zoop (Table 6). However, the 14 C-depletion of zooplankton biomass relative to the putative autochthonous food sources (Table 6) indicates the use of some aged allochthonous food source by the zooplankton for their dietary needs. Zooplankton incorporation of aged allochthonous food in these small lake systems contrasts with observations in Lake Superior, where the mesozooplankton preferentially and heavily depended on in situ primary production. Conservative estimates based on 14 C DIC-Zoop indicate that the proportion of allochthonous food supporting zooplankton in the small lakes (except Fraser Lake) was larger than that in open Lake Superior (Table 6). This is also consistent with the observed relationship between zooplankton dependence on allochthonous food resources and variables such as ratio of catchment area to lake surface area (Fig. 5a), and water residence time (Fig. 5b). The ratio of basin area to surface area of a lake gives an indication of potential terrestrial subsidy to the lake's ecosystem. As the basin area-to-surface area ratio increases, suggesting potentially higher terrestrial influence, the difference between 14 C Zoop and 14 C DIC also increases as reflected in the positive correlation between 14 C DIC-Zoop and the ratio of basin area to lake surface area (Fig. 5a). Hydrologic residence time is a variable that is related to lake size. Small lakes tend to have shorter water residence times, whereas large lakes usually hold water for longer time periods (Table 1). There was generally a negative relationship between the hydrological residence time of the lakes and 14 C DIC-Zoop (Fig. 5b), implying the difference between 14 C of zooplankton and 14 C of DIC decreases with an increase in lake water residence time and, by extension, with lake size.
In the oceanic sites, 14 C values of zooplankton and bulk POM were similar at all sites (Fig. 3b), suggesting that either the bulk POM was almost entirely derived from that year's primary production, or that the zooplankton were indiscriminately feeding on the bulk POM. It is worth noting however that estimating zooplankton food sources in the oceanic sites is complicated by the considerable differences in 14 C values of DIC with depth and laterally, such that water mass movements and migratory feeding of zooplankton (and upper trophic organisms) could significantly mask the actual radiocarbon relationships between zooplankton, DIC and POC. The pooled data from the small lakes, Lake Superior and the Pacific Ocean show strong correlation between 14 C values of zooplankton and DIC, but poor correlation between 14 C of zooplankton and bulk POM (Fig. 4a, b) indicating that, in most aquatic ecosystems, recent in-situ primary production is the most preferred food resource for zooplankton.
It is worth stating that different zooplankton groups do have different feeding and/or ecological strategies, and the observed zooplankton food preferences and the relationship between lake size and allochthony of zooplankton discussed above could be influenced by this. While the higher mesozooplankton autochthony seen in Lake Superior relative to the smaller lakes could be attributed to the specific filter feeding style of the predominant calanoid copepods (∼ 70 % of crustacean zooplankton) in the lake, the small offset between 14 C of algae (based on 14 C-DIC) and 14 C of the bulk mesozooplankton suggests the remaining zooplankton groups in Lake Superior (∼ 30 %) including cyclopoid copepods and daphnids were also largely feeding on algae (or herbivores that ate algae), although these zooplankton groups are adapted to utilize other food options such as detritus, protists, bacteria and other zooplankton. Since cladocerans, such as daphnids, and cyclopoid copepods are typically more abundant in small nutrient-enriched aquatic systems (Gannon and Stemberger, 1978;Balcer et al., 1984), it could be argued that their adaptation for feeding on non-algal food options is responsible for the relatively high zooplankton allochthony seen in the smaller lakes. However, the zooplankton composition itself could be coupled to nutrient (N and P) availability (DeMott and Gulati, 1999;Schulz and Sterner, 1999;Conde-Porcuna et al., 2002) and hence terrestrial influence, consistent with the observed pattern of increasing zooplankton allochthony with terrestrial influence as seen in this study.

Conclusions
Our isotopic investigation shows that intermediate trophiclevel mesozooplankton in Lake Superior prefer to incorporate fresh autochthonous food, despite the availability of other organic carbon sources, and that upper trophic levels are likely not supported by terrestrial and/or resuspendedsediment OC subsidies to the carbon cycle. A similar trend is apparent in our oceanic sites and selected small lakes, although the small lakes do exhibit a higher degree of zooplankton dependence on allochthonous food resources. This provides real-world support to lab studies showing preferential incorporation of phytoplankton fatty acids and POC into herbivorous zooplankton offered mixed diets of terrestrially derived and phytoplankton-derived particulate organic matter (Brett et al., 2009). This study also confirms, in a large lake and similar clear-water systems such as the open oceans, the observation from unproductive small lakes that zooplankton selectively incorporate fresh autochthonous organic carbon (Karlsson, 2007;Karlsson et al., 2012). Our results suggest that, if spatial or temporal subsidies of organic carbon fuel the net heterotrophy seen in Lake Superior, their effects are limited to the microbial loop and lower trophic levels, and do not extend to mesozooplankton and higher trophic levels. Further research should focus upon catabolic metabolism of mesozooplankton and both anabolic and catabolic metabolism in the microbial loop to further our understanding of such subsidies in the carbon cycle and energy transfer.