Lake Greifen (Greifensee) is a small perialpine lake, located in the eastern fringes of the Zurich metropolitan area at 47∘21′ N and 8∘40′ E (Fig. 1). The lake has a surface area of 24 km2 and a maximum depth of 32 m. The lake is fed by three small brooks and has one main outlet, the Glatt canal. Lake Greifen experienced severe eutrophication in the mid-20th century (Hollander et al., 1992; Keller et al., 2008). Strict government regulations on nutrient inputs were imposed in the 1970s, and the water quality in the lake has since improved, but its deep water remains anoxic and nutrient levels in the upper water column are still elevated. Winter overturn in the lake brings additional nutrients to the surface water, resulting in large phytoplankton blooms in the spring and summer as temperature and light availability increase (McKenzie, 1982). All samples from Lake Greifen were collected from the northern part of the lake, near a permanent platform maintained by Eawag (at 47∘21.99′ N, 8∘39.89′ E).

Lake Lucerne (Vierwaldstättersee) is a large perialpine lake, located in central Switzerland at 47∘0′ N and 8∘30′ E (Fig. 1). The lake, which has a total surface area of 116 km2, is formed of seven distinct basins, of which the deepest is 214 m. The lake is fed by four alpine rivers: the Reuss, Muota, Engelberger Aa, and Sarner Aa, and its primary outflow is Reuss river from the northwest tip of the lake. Although Lake Lucerne experienced a mild eutrophication event in the 1970s, it is oligotrophic today (Bürgi et al., 1999; Bührer and Ambühl, 2001; Thevenon et al., 2012). All samples from Lake Lucerne were collected from the center of Kreuztricher basin (near 8∘21′ N, 47∘0′ E), with a water depth of 96 m.

Particulate material in each lake was collected at approximately monthly intervals throughout the spring and summer of 2015 (mid-April through early September). Surface water (∼0.5 m water depth) was filtered onto a pre-combusted 142 mm diameter GF/F filter (0.7 µm pore size) using a WTS-LV Large Volume Pump (McLane, Massachusetts, USA). Pumping began at 7 Lmin-1 and continued until the flow rate decreased to 4 Lmin-1 or until 25 min had passed. All filters were collected at midday on sunny or mostly sunny days. Filters were wrapped in combusted aluminum foil and stored in a cool box on ice until transport to the laboratory, where they were stored at -20 ∘C until analysis.

Water samples were collected from surface water before and after pumping began. Samples were collected in 4 mL screw cap vials, sealed with electrical tape, and stored at room temperature prior to analysis. Depth profiles of temperature, conductivity, pH, turbidity, and dissolved oxygen were collected for the upper 20 m of the water column each sampling day at the beginning and end of filtration using a multiparameter CTD probe (75M, Sea and Sun Marine Tech, Trappenkamp, Germany).

On the morning of each sampling day, 4×12.5 L of surface water was collected in acid-rinsed, autoclaved, transparent carboys for in situ incubations. In two of the four carboys, 1 mL of concentrated NaH13CO3 solution was added. The other two carboys were not isotopically labeled. Carboys were mixed and attached to a fixed, floating line so that they stayed in the upper 50 cm of lake water throughout the day. After 6 h, they were retrieved and the contents were filtered onto a pre-combusted 142 mm diameter GF/F filter using a peristaltic pump. Water samples for DIC analyses were collected in 12 mL exetainers prior to isotopic labeling, after labeling but before incubation, and after incubation. These samples were sterile filtered through a 0.2 µm syringe filter and stored in the dark at 4 ∘C prior to analysis.

Surface water isotope samples were filtered through a 25 mm syringe filter with a 0.45 µm polyethersulfone membrane to remove particulate matter. Water δ2H and δ18O values were measured by cavity ring-down spectroscopy (CRDS) on a L-2120i water isotope analyzer (Picarro, Santa Clara, CA, USA) at ETH Zurich. Each sample was injected seven times in sequence, and the first four values were discarded to avoid any memory effects from the previous sample. Three water standards with known δ2H values of ranging from -161 to 7 ‰ and δ18O values ranging from -22.5 to 0.9 ‰ were injected at the beginning and end of each sequence, as well as after every 10 samples. These standards were used to correct measured values to the VSMOW scale and to account for any instrumental drift over the course of the sequence. Average standard deviations (SDs) were 0.4 ‰ for hydrogen isotopes and 0.06 ‰ for oxygen isotopes.

DIC concentrations were measured on a TOC-LCSH/CHN total organic carbon analyzer (Shimadzu, Kyoto, Japan). Solutions with DIC concentrations ranging from 5 to 100 mgL-1 were injected at the beginning of the sequence to form a calibration curve, and one standard of 50 mgL-1 was run after every five samples. Samples were analyzed in triplicate.

Exetainers of 3.7 mL were prepared for δ13C measurements of DIC by adding 100 µL of concentrated H3PO4 and filling the headspace with He; 1 mL of sample water was added with a syringe through the septa of the exetainer. Samples were allowed to equilibrate overnight before analysis. Carbon isotope values were measured on an isotope ratio mass spectrometer (IRMS) (Isoprime, Stockport, UK). A standard of known isotopic composition was analyzed after every six samples. All samples were measured in duplicate.

An internal standard containing nC19-alkanol, nC19-alkanoic acid, and 5α-cholestane was quantitatively added to freeze-dried filters, which were extracted in 30 mL of 9:1 dichloromethane (DCM)/methanol (MeOH) in a SOLVpro microwave reaction system (Anton Paar, Graz, Austria) at 70 ∘C for 5 min (Randlett et al., 2017), centrifuged, and the supernatant containing the total lipid extract (TLE) was poured off and evaporated under a gentle stream of N2. The TLE was saponified with 3 mL of 1 N KOH in MeOH and 2 mL of solvent-extracted nanopure H2O for 3 h at 80 ∘C (Randlett et al., 2017), after which the neutral fraction was extracted with hexane. Subsequently, the aqueous phase was acidified to pH =2, and the protonated fatty acids were extracted with hexane.

Neutral fractions were further purified using silica gel column chromatography, following a scheme modified from Randlett et al. (2017). The sample was dissolved in hexane and loaded onto a 500mg/6mL Isolute Si gel column (Biotage, Uppsala, Sweden). N-alkanes were eluted in 4 mL of hexane, aldehydes and ketones in 4 mL of 1:1 hexane/DCM, alcohols in 4 mL of 19:1 DCM/MeOH, and remaining polar compounds in 4 mL of MeOH. The alcohol fraction was acetylated with 25 µL of acetic anhydride and 25 µL of pyridine for 30 min at 70 ∘C. The δ2H and δ13C values of the added acetyl group were determined by analyzing acetylated and unacetylated nC10-alkanol.

Further purification was necessary in order to obtain base line separation of brassicasterol for δ2H measurements. This was achieved by loading the acetylated alcohol fraction onto 500 mg of Si gel impregnated with AgNO3 (10 % by weight, Sigma Aldrich) in a 6 mL glass cartridge. The first fraction, containing n-alkanols and phytol, was eluted with 20 mL of 4:1 hexane/DCM; the second fraction, containing stanols and singly unsaturated sterols (such as cholesterol) with 20 mL of 1:1 hexane/DCM; the third fraction, containing most doubly unsaturated sterols including brassicasterol, with 16 mL of DCM; and the remaining compounds with 4 mL of ethyl acetate.

Fatty acid fractions were methylated with 1 mL of BF3 in MeOH (14 % by volume, Sigma Aldrich) for 2 h at 100 ∘C. After methylation, 2 mL of nanopure H2O was added to the sample and the fatty acid methyl esters (FAMEs) were extracted with hexane. The δ2H and δ13C values of the added methyl group were determined by methylating phthalic acid of known isotopic composition (prepared by Arndt Schimmelmann at Indiana University).

FAMEs and brassicasterol were quantified by gas chromatography–flame ionization detection (GC-FID) (Shimazdu, Kyoto, Japan). Samples were injected by an AOC-20i autosampler (Shimadzu) through a split/splitless injector operated in splitless mode at 280 ∘C. The GC column was an InertCap 5MS/NP (0.25mm×30m×0.25 µm) (GL Sciences, Japan) and it was heated from 70 to 130 ∘C at 20 ∘Cmin-1, then to 320 ∘C at 4 ∘Cmin-1, and held at 320 ∘C for 20 min. FAMEs were identified by comparing their retention times to an external standard (fatty acid methyl ester mix from Sulpelco, reference no. 47885-U). Brassicasterol and phtyol were identified by comparing their retention times to those obtained by analyzing a subset of samples by gas chromatography–mass spectrometry (GC-MS) under identical conditions. In order to determine how much of the compound was in the original sample, peak areas were normalized to those of the internal standard. Peak areas were quantified relative to an external calibration curve in order to determine suitable injection volumes for isotopic analysis.

The stable isotope values of individual FAMEs and brassicasterol were measured by gas chromatography–isotope ratio mass spectrometry (GC-IRMS). A GC-1310 gas chromatograph (Thermo Scientific, Bremen, Germany) equipped with an InertCap 5MS/NP (0.25mm×30m×0.25 µm) (GL Sciences, Japan) was interfaced to a Delta Advantage IRMS (Thermo Scientific) with a ConFlow IV interface (Thermo Scientific). Samples were injected with a TriPlusRSH autosampler to a PTV inlet operated in splitless mode at 280 ∘C. The oven was heated from 80 to 215 ∘C at 15 ∘Cmin-1, then to 320 ∘C at 5 ∘Cmin-1, and then was held at 320 ∘C for 10 min. Hydrogen isotope samples were pyrolyzed at 1420 ∘C after they eluted from the GC column. Carbon isotope samples were combusted at 1020 ∘C after elution.

Raw isotope values were converted to the VSMOW (hydrogen) and VPDB (carbon) scales using Thermo Isodat 3.0 software and pulses of a reference gas that was measured at the beginning and end of each analysis. Sample δ2H and δ13C values were further corrected using the slope and intercept of measured and known values of isotopic standards (nC17,19,21,23,25,28,34-alkanes; Arndt Schimmelmann, Indiana University), which were run at the beginning and end of each sequence as well as after every six to eight sample injections. Offsets between measured and known values for these standards were used to correct for any drift over the course of the sequence or any isotope effects associated with peak area or retention time. The SD for these standards averaged 4 ‰ and the average offset from their known values was 2 ‰ for hydrogen isotopes. For carbon isotopes, the average SD of isotopic standards was 0.4 ‰ and the average offset from known values was 0.1 ‰ over the period of analysis.

An additional standard of nC29-alkane was measured three times per sequence, corrected in the same way as the samples, and used for quality control. The SD of these measurements was 4 ‰ for hydrogen and 0.5 ‰ for carbon over the period of analysis. The H3+ factor was measured at the beginning of each sequence and averaged 3.6±0.3 during the analysis period. Samples were corrected for hydrogen and carbon added during derivatization using isotopic mass balance, and reported errors represent propagated errors from replicate measurements and the uncertainties associated with the added hydrogen.

Lipid production rates were calculated using Eq. (1) (modified from Popp et al., 2006): Production rate=(δ13Cl-δ13Cn)/(δ13CDIC-δ13Cn)×(Ct/t), where δ13Cl is the δ13C value of the target compounds from labeled incubations, δ13Cn is that from unlabeled incubations, δ13CDIC is the δ13C value of DIC, Ct is the concentration of the lipid at the end of the incubation, and t is the duration of the incubation. Residence times – assuming a steady state, the amount of time needed to replace all molecules of a given lipid – were calculated by dividing Ct by the production rate, which reduces to Eq. (2): Residence time=t×(δ13CDIC-δ13Cn)/(δ13Cl-δ13Cn).

PRISM software (Graphpad Software Inc., La Jolla, CA, USA) was used to carry out all statistical analyses. Ordinary least-squares regression was used to determine relationships among fractionation factors, temperature, and lipid production rates. Regression lines are only shown where the slope of the regression was significantly different from 0 at the p<0.05 level. The results of all linear regression analyses are presented in Table 2. Differences between the slopes of various regressions were assessed using a two-tailed test of the null hypothesis that both slopes are equal. Differences in the mean values of replicate measurements were determined using an unpaired, two-tailed t test and were considered significantly different for p<0.05.

Lipid concentrations increased significantly in Lake Greifen from April to July and then declined slightly from July to September, except for phytol and nC161:1 fatty acid, which had increasing concentrations into the late summer (Fig. 2a). nC16:0 fatty acid had the highest concentrations, while those of nC16:1 fatty acid were usually the lowest, except in April–May, when nC14:0 was the least abundant fatty acid. Brassicasterol concentrations were 1–2 orders of magnitude smaller than those of fatty acids and phytol concentrations were intermediate (Fig. 2a). Lipid concentrations in Lake Lucerne were generally an order of magnitude lower than in Lake Greifen (Fig. 2b). Fatty acid concentrations increased significantly from April to May in Lake Lucerne and were then relatively stable throughout the rest of the time series (Fig. 2b). Again, nC16:0 fatty acid had the highest concentrations and nC16:1 was the least abundant fatty acid. Phytol concentrations were typically an order of magnitude lower than those of fatty acids in Lake Lucerne and increased slightly over the course of the time series. Brassicasterol concentrations were an order of magnitude lower still and reached a maximum in June (Fig. 2b).

In both lakes, fatty acid production rates were highest for nC16:0 (palmitic acid), followed by nC18:x (unsaturated C18 fatty acids, primarily nC18:1n9c, or oleic acid), nC14:0 (myristic acid), and nC16:1 (palmitoleic acid) (Fig. 2c and d). Phytol and brassicasterol production rates were 2–3 orders of magnitude lower in both lakes than those of fatty acids (Fig. 2c and d). Lipid production rates were up to three times higher in Lake Greifen than in Lake Lucerne (Fig. 2c and d). Lipid production rates generally increased from May to July and then remained high in Lake Greifen, while in Lake Lucerne they were relatively constant throughout the study period (Fig. 2c and d).

Residence times – or the amount of time necessary to replace all molecules of a given compound assuming steady state – of individual lipids were calculated according to Eq. (2) (Sect. 2.7) and were typically shortest for nC14:0, nC16:0, and nC18:x fatty acids, with values as low as 9±1 h in Lake Greifen in May and as low as 12±3 h in Lake Lucerne in August (Table 3). Of the fatty acids, nC16:1 had the longest residence times, reaching 60±13 h in Lake Greifen in May and 62±27 h in Lake Lucerne in June (Table 3). Brassicasterol residence times were the longest of any lipid in the first part of the time series but were exceeded by phytol for the last two sampling dates (Table 3).

In both lakes lipid δ2H values typically decreased over the spring and summer (Fig. 2e and f). This effect was most pronounced for fatty acids in Lake Greifen. For example, nC16:0 fatty acid δ2H values declined by 133 ‰ (from -172 to -305 ‰) from April to August in Lake Greifen, while they only declined by 53 ‰ (from -249 to -302 ‰) over the same time period in Lake Lucerne (Fig. 2e and f). During the same time period water δ2H values increased slightly in Lake Greifen (from -73 to -65 ‰) and were relatively constant in Lake Lucerne (fluctuating between -82 and -86 ‰) (Supplement Sect. S2). Changes in the fractionation factor between fatty acids and surface water (αlipid-water) were therefore linked closely to changes in fatty acid δ2H values. In Lake Greifen, αlipid-water for nC16:0 fatty acid decreased from 0.891 to 0.743 from April to August (Fig. 2g), while in Lake Lucerne it decreased from 0.821 to 0.763 (Fig. 2h). Similar patterns were observed for αlipid-water for nC14:0, nC16:1 and nC18:X fatty acids (Fig. 2g and h). Values for αlipid-water were less variable for phytol and brassicasterol than for fatty acids, although they also declined from April to May. Brassicasterol was always 2H-depleted relative to fatty acids in Lake Greifen and was depleted in 2H relative to all fatty acids except nC14:0 in Lake Lucerne (Fig. 2g and h). Phytol δ2H values were the most 2H depleted of any lipid measured in either lake (Fig. 2g and h).

Overall, fatty acid αlipid-water values were negatively correlated with lake surface temperature (LST) in both lakes (R2=0.32, p=0.004 in Lake Greifen; R2=0.24, p=0.01 in Lake Lucerne) (Fig. 3; Table 2). The slope of the relationship was significantly steeper (p=0.03) in Lake Greifen than in Lake Lucerne (m=-0.006±0.002 in Lake Greifen and -0.003±0.001 in Lake Lucerne). Significant correlations were observed between LST and αlipid-water values for most fatty acids but not for nC16:1 in Lake Greifen and nC14:0 in Lake Lucerne (Fig. 3; Table 2). Significant relationships between LST and αlipid-water values were not observed in either lake for brassicasterol or phytol (Fig. 3; Table 2).

Fatty acid production rates were not correlated with αlipid-water values in either lake (Table 2). Among individual fatty acids, only nC16:1 fatty acids from Lake Greifen had a significant negative correlation between αlipid-water values and production rate (R2=0.84; p=0.03) (Table 2). Brassicasterol and phytol production rates were not correlated with αlipid-water in either lake (Table 2), although brassicasterol αlipid-water values from Lake Lucerne cluster as a significantly higher group than in Lake Greifen (p=0.0004).

Hydrogen isotope fractionation for short-chain fatty acids varies significantly among organisms with different metabolisms (Zhang et al., 2009a; Osburn et al., 2011; Heinzelmann et al., 2015a). In general, fatty acids from heterotrophs grown on tricarboxylic acid (TCA) cycle precursors are most enriched in 2H, followed by heterotrophs grown on sugars, then photoautotrophs, and finally chemoautotrophs (Zhang et al., 2009a; Osburn et al., 2011; Heinzelmann et al., 2015a). This variability can be greater than 500 ‰ and has led to the suggestion that the δ2H values of ubiquitous compounds such as palmitic acid can be used as a proxy of net community metabolism. For example, at a coastal site in the North Sea, fatty-acid-chain-weighted average δ2H values declined by more than 40 ‰ during the spring phytoplankton bloom, which was attributed to increased contributions from photoautotrophs (Heinzelmann et al., 2016).

In lakes Lucerne and Greifen, large decreases in fatty acid δ2H values from April to May coincide with increases in fatty acid concentrations of 1–2 orders of magnitude (Fig. 2). It is therefore possible that the 2H-enriched April samples represent a wintertime background of mixed heterotrophic and autotrophic derived compounds. As phytoplankton productivity ramped up with warmer temperatures, water column stratification, and longer daylight hours in the spring, newly produced fatty acids from photoautotrophs could have overwhelmed the heterotrophic signature, causing the net fatty acid δ2H values to decrease. The increase in phytoplankton cell density (Fig. 4) from April onward in both lakes is supportive of increased contributions of fatty acids from phytoplankton as the study period progressed.

For Lake Greifen, a simple isotopic mass balance indicates that 31 % of the total nC16:0 fatty acid would need to come from heterotrophs with αlipid-water values of 1.200 (the maximum observed by Zhang et al., 2009a) in mid-April if the remaining nC16:0 fatty acid was derived from phytoplankton with αlipid-water values of 0.750 (assuming mid-summer αlipid-water values represent the phytoplankton end member). Similar calculations suggest that 16 % of nC16:0 fatty acid would need to come from heterotrophic bacteria in mid-April in Lake Lucerne in order to account for the 50 ‰ decrease in nC16:0 fatty acid δ2H values over the course of the summer. These calculations assume that all heterotrophic bacteria use the highest αlipid-water value ever observed for heterotrophs and that they primarily use the TCA cycle rather than glycolysis. If, as is likely, there is diversity in αlipid-water values for short-chain fatty acids produced by the heterotrophic bacteria, and at least some of the heterotrophs are relying primarily on glycolysis, the portion of these fatty acids from heterotrophic sources in April would need to be even higher than the values calculated above. This would necessitate a proportionally larger winter heterotrophic contribution of fatty acids than was observed in the coastal North Sea (Heinzelmann et al., 2016), and it seems likely that other variables may contribute to the springtime decline in fatty acid δ2H values.

Contributions from heterotrophs are also an improbable explanation for 2H-enriched brassicasterol and phytol in April. Brassicasterol is a sterol that is commonly used as a biomarker for diatoms, although it has also been detected in some non-diatom eukaryotic phytoplankton (Volkman et al., 1998; Volkman, 2003; Rampen et al., 2010; Taipale et al., 2016) and occasionally in plant oils (Zarrouk et al., 2009). Since brassicasterol is not produced by bacterial sources, it seems improbable that the 25 ‰ (Greifen) and 19 ‰ (Lucerne) decreases in its δ2H values from April to May could be due to proportionately greater heterotrophic contributions during the winter, as suggested for fatty acids. The 46 ‰ April–May decrease in δ2H values of Lake Greifen phytol, the side chain of chlorophyll molecules, is also unlikely to be caused by heterotrophic contributions in the early spring. Although phytol is produced by some photoheterotrophs, these typically have similar αlipid-water values to photoautotrophs (Zhang et al., 2009a).

However, seasonal changes in the phytoplankton community composition alone could be a significant source of variability in αlipid-water over the course of the study period. Hydrogen isotope fractionation for nC16:0 fatty acid has been demonstrated to vary by over 100 ‰ among five different species of freshwater green algae grown in laboratory batch cultures (Zhang and Sachs, 2007). Such variations may be due to different enzymes involved in lipid synthesis among different species or to the colony-forming behavior of the two species with higher αlipid-water values. Smaller species-dependent variations in αlipid-water have been observed in cultures of haptophytes (∼30 ‰ offset between alkenones in G. oceanica and in E. huxleyi; Schouten et al., 2006). Since there are limited data from culturing experiments, it is not possible to say how widespread such interspecies variability is. It is possible that most phytoplankton display similar magnitudes of hydrogen isotope fractionation during lipid synthesis under similar conditions. However, it is equally possible that lipid hydrogen isotope fractionation varies among species in ways that are not yet understood.

Given this uncertainty, and the significant changes in abundance of different phytoplankton taxa in Lake Greifen over the course of 2015 (Fig. 4a), contributions of lipids from different species of algae with different magnitudes of hydrogen isotope fractionation could account for some or all of the seasonal variability in αlipid-water. A comparable data set of algal species counts does not exist for Lake Lucerne from 2015, but bimonthly data have been compiled from 2014 (Fig. 4b). Some changes in relative distributions of taxa are similar between the two lakes; for example, both Lake Greifen in 2015 and Lake Lucerne in 2014 experienced a peak in golden algae (Chrysophyceae) in June and elevated abundance of cyanobacteria in late summer and late autumn (Fig. 4). Other trends differ starkly between the two lakes. Green algae (Chlorophyceae) are largely absent from Lake Lucerne, while they make up a significant portion of the algal community in Lake Greifen. Notably, the relative abundance of green algae steadily increased from May to September in Lake Greifen (Fig. 4), during which time αlipid-water values declined at a greater rate for most compounds than they did in Lake Lucerne (Fig. 2). If green algae tend to have lower αlipid-water values than other algal taxa, their greater abundance in Lake Greifen throughout the summer could account for the greater decline in αlipid-water over the course of the time series than in Lake Lucerne.

Even for lipids produced in controlled cultures of eukaryotic algae, several factors have been shown to influence hydrogen isotope fractionation, including salinity, light availability, temperature, and growth rate (summarized in Table 1) (Sachs, 2014 and sources therein; Chivall et al., 2014; M'boule et al., 2014; Heinzelmann et al., 2015b; Sachs and Kawka, 2015; van der Meer et al., 2015; Wolhowe et al., 2015; Maloney et al., 2016; Sachs et al., 2016, 2017). Of these, salinity can be excluded as a source of variability freshwater lakes such as Greifen and Lucerne. The effect of light availability on αlipid-water has only been detected at low light levels (below 250 µmol photons m-2s-1) (van der Meer et al., 2015; Wolhowe et al., 2015; Sachs et al., 2017). Although photosynthetically available radiation (PAR) was not measured as part of the present study, all samples were collected from lake surface water at a midlatitude Northern Hemisphere site during boreal spring and summer, and it is unlikely that PAR was less than 250 µmol photons m-2s-1 at any sampling date (Pinker and Laszlo, 1992), meaning that the effect of light intensity is unlikely to be a source of the observed seasonal variability in αlipid-water in lake surface water.

LST varied from 11 to 27 ∘C in Lake Greifen and from 11 to 25 ∘C in Lake Lucerne over the study period (Fig. 2i and j) and therefore may have contributed to the seasonal changes in lipid δ2H values. In laboratory cultures of eukaryotic algae, αlipid-water values for acetogenic lipids have been shown to decrease by 0.002–0.004 ∘C-1, resulting in more depleted δ2H values (Schouten et al., 2006; Zhang et al., 2009b; Wolhowe et al., 2009, 2015). Increased hydrogen isotope fractionation at higher temperatures has been attributed to (i) changes in the relative activity of different enzymes involved in lipid synthesis at different temperatures, (ii) changes in the relative amount of NADPH from the pentose phosphate cycle as temperature changes, and (iii) the potential for hydrogen tunneling at higher temperatures as substrate–enzyme complex vibrations increase (Sachs, 2014, and references therein). The relationship between temperature and hydrogen isotope fractionation in cultures is similar to that observed for fatty acids in Lake Lucerne, where αlipid-water decreases by 0.003±0.001 ∘C-1 (Fig. 3; Table 2). The relationship between αlipid-water and temperature for fatty acids in Lake Greifen (-0.006±0.002 ∘C-1) (Fig. 3; Table 2) is much steeper than that observed in culturing studies.

If the influence of temperature on hydrogen isotope fractionation is consistent among laboratory cultures and lakes, warmer temperatures can account for the entire seasonal change in αlipid-water for fatty acids in Lake Lucerne. However, increasing temperatures would only be able to explain part of the decrease in fatty acid δ2H values in Lake Greifen over the course of the spring and summer. At most, temperature could account for half of the decrease in αlipid-water in Lake Greifen, assuming a consistent relationship to that observed in cultures.

Brassicasterol and phytol αlipid-water values do not have strong relationships with LST. There is no correlation between LST and αlipid-water values for phytol in either lake nor for brassicasterol in Lake Lucerne. In Lake Greifen temperature and αbrassicasterol-water are negatively correlated, although the relationship is not quite significant (Fig. 3f; Table 2). The slope of the relationship between αbrassicasterol-water and LST in Lake Greifen is significantly shallower than that observed for fatty acids (-0.002±0.001 for brassicasterol vs. 0.008-±0.002 for nC16:0 fatty acid). Given these relatively weak relationships, it seems unlikely that temperature influences hydrogen isotope fractionation of either brassicasterol or phytol. The negative correlations between temperature and αlipid-water values for fatty acids may therefore be an artifact of the probable increase in photoautotrophically derived compounds as it became warmer throughout the spring (Sect. 4.1). However, the relationship between temperature and αlipid-water values in cultures has only been observed for n-alkanoic acids and alkenones, both of which are acetogenic lipids. It is thus possible that temperature is partially responsible for the decreases in αlipid-water values as temperature increased in lakes Lucerne and Greifen but that the responsible mechanism is specific to lipids produced acetogenically and does not affect isoprenoids, such as sterols and phytol.

Phytoplankton productivity and biomass also increased with temperature in the spring and summer in lakes Greifen and Lucerne, with a more marked effect in nutrient-rich Greifen (Fig. 2a–d, Fig. 4). This trend could also partially explain the increase in 2H fractionation and decrease in lipid δ2H values that co-occurred with rising temperatures. Increased nutrient availability and higher growth rates have been shown to result in lower αlipid-water values during lipid synthesis for eukaryotic algae in laboratory settings, with lipids more 2H depleted relative to source water as growth rate increases (Schouten et al., 2006; Zhang et al., 2009b; Sachs and Kawka, 2015; Wolhowe et al., 2015). This relationship is most likely caused by increased contributions of hydrogen from relatively enriched NADPH from the oxidative pentose phosphate cycle under low-growth, nutrient-stressed conditions, at the expense of relatively depleted hydrogen from photosystem I (Schmidt et al., 2003; Sachs and Kawka, 2015). If April samples include a higher proportion of lipids from organisms in a low-growth maintenance phase, they should therefore be relatively enriched in 2H. As light availability and water column stratification became more amenable to photosynthesis later in the spring, relatively 2H-depleted lipids produced with NADPH from photosystem I would be expected to become more abundant, bringing the net δ2H and αlipid-water values down.

Bottle incubations to determine lipid production rates were unfortunately not conducted for the first sampling in April, when the most enriched lipid δ2H values were measured. For the remaining five sampling dates, there were not significant correlations between lipid production rate and αlipid-water, with the exception of nC16:1 in Lake Greifen (Table 2). Lipid concentrations and production rates are not a direct proxy for growth rate, as a higher percentage of algal biomass is typically allocated to lipids under low nutrient and slow growth conditions (Roessler, 1990; Williams and Laurens, 2010). However, higher lipid production rates for the whole community (rather than on a per cell basis) will co-occur with higher growth rates. The large increase in fatty acid concentrations from April to May in both lakes, as well as smaller increases in brassicasterol concentrations, may indicate that the greatest community-wide change in growth rate occurred between those 2 months and contributes in part to the decrease in fatty acid, phytol, and brassicasterol δ2H values from April to May.

For phytol, nC14:0, nC16:0, and nC18:x fatty acids, there was no significant difference in αlipid-water between the oligotrophic and eutrophic lake. However, significant differences in αlipid-water do exist for brassicasterol (0.045±0.008; p=0.0004) and nC16:1 fatty acid (0.058±0.022; p=0.02) (Fig. 2). For brassicasterol, αlipid-water is lower in Lake Greifen (0.712±0.006) than in the less productive Lake Lucerne (0.757±0.005). This result would be consistent with decreased hydrogen isotope fractionation (higher α values) for sterols in more nutrient-limited systems, as predicted by culturing experiments (Zhang et al., 2009b; Sachs and Kawka, 2015). The fact that this strong difference in fractionation between the two lakes is observed only for brassicasterol may be because it is the most source specific of the biomarkers that were analyzed. Phytol and short-chain fatty acids are produced by all photoautotrophs and may be dominated by phytoplankton that are optimized to grow under the nutrient regimes of each system, while relatively more of the brassicasterol may come from taxa that are nutrient-stressed and relying more on the pentose phosphate pathway than photosystem I.

Alternatively, the difference in αbrassicasterol-water values between the two lakes could be due to variable contributions of brassicasterol from different phytoplankton sources. Even though brassicasterol is produced by fewer organisms than short-chain fatty acids and phytol, it still has multiple sources (Volkman et al., 1998; Volkman, 2003; Rampen et al., 2010; Taipale et al., 2016). Species-specific differences in hydrogen isotope fractionation have not been observed for sterols but have been reported for fatty acids and alkenones (Schouten et al., 2006; Zhang and Sachs, 2007), making this an unconstrained possibility that could be responsible for the difference in αbrassicasterol-water between the oligotrophic and eutrophic lake. Different sources could also account for the difference in αlipid-water for nC16:1 fatty acid, which displays higher αlipid-water values in the more productive lake, and therefore cannot be explained by the nutrient effect observed in cultures.