Lake Schreventeich is a small holomictic lake situated in northern Germany (54∘19′36.79′′ N, 10∘07′17.57′′ E). Its surface area covers approximately 0.38 km2 and it has an average depth of 1.4–1.6 m (maximum depth of 3.4 m). The lake has no tributaries and is solely fed by precipitation and ground water inflow.

Surface water samples for the analysis of HGs were taken from late July to the end of October 2013. Oxygen concentrations and surface water temperatures were measured at time of sampling using the portable oxygen measuring instrument Oxi 1970i coupled to a CellOx325 oxygen probe (WTW, Germany). The pH of all water samples was determined using a FG2-FiveGo (Mettler-Toledo, Germany) using a two-point calibration on certified reference solutions obtained from Hanna Instruments. Surface sediments (0–1 cm) from two locations within Lake Schreventeich were obtained in March 2014 using a Uwitec gravity corer (Uwitec, Switzerland). All sediments were freeze-dried and ground to a homogenous powder using pestle and mortar.

A total of 100 mL of surface water was collected during each sampling and filtered over a preweighed Whatman filter GF/C (1.2 µm, diameter 47 mm). After filtration, filters were manually inspected and non-phytoplankton biomass was removed using a pair of tweezers. All filters were subsequently dried in an oven at 105 ∘C for 24 h. Phytoplankton biomass was calculated as the weight difference between the preweighed and the oven-dried filters.

Measured volumes (e.g., 3–4 L) of surface water were filtered through a MN 85/70 BF (binder free) glass fiber filter with a pore size of 0.45 µm (Macherey-Nagel, Germany). All filters were freeze-dried and extracted following a modified Bligh and Dyer procedure as described by Rütters et al. (2002). Briefly, filters were cut into fine pieces with a solvent-cleaned scissor and ultrasonically extracted using a solvent mixture of methanol (MeOH), dichloromethane (DCM) and phosphate buffer (2:1:0.8; v/v/v). After centrifugation, the supernatant was collected and the residue extracted twice with the solvent mixture specified above. DCM and phosphate buffer were added to the pooled supernatants to achieve a ratio of MeOH / DCM / phosphate buffer of 1:1:0.9 (v/v/v), allowing the separation of two phases. The bottom layer, containing the organic fraction, was transferred to a glass vial and the remaining aqueous phase was extracted twice with DCM. The combined extracts were reduced under rotary vacuum, transferred to preweighed vials and dried under a gentle stream of N2. All Bligh and Dyer extracts were subsequently dissolved in DCM / MeOH (9:1; v/v) to a concentration of 2–4 mg mL-1 and filtered through a 0.45 µm pore size, regenerated cellulose filter (13 mm; LLG Labware, Germany) prior to analysis. In addition to water column filtrates, 0.5 g of freeze-dried core top sediments obtained from Lake Schreventeich were extracted using the procedure outlined above.

Heterocyst glycolipids were analyzed following the procedure described by Bauersachs et al. (2014) with some brief modifications. Separation of the target compounds was achieved using an Alliance 2690 HPLC (high-performance liquid chromatography) system (Waters, UK) fitted with a Luna Hilic 200A column (150 mm × 2 mm inner diameter; 3 µm particle size; Phenomenex, Germany) maintained at 30 ∘C. The following linear gradient was used with a flow rate of 0.2 mL min-1 : 95 % eluent A/5 % eluent B to 70 % A/30 % B in 10 min (held 20 min), followed by 70 % A/30 % B to 35 % A/65 % B in 15 min (held 15 min), then back to 95 % A/5 % B in 1 min (held 20 min) to re-equilibrate the column. Eluent A was hexane / isopropanol / formic acid / 14.8 M aqueous NH3 (79:20:0.12:0.04; v/v/v/v) and eluent B was isopropanol / water / formic acid / 14.8 M aqueous NH3 (88:10:0.12:0.4; v/v/v/v).

Detection of heterocyst glycolipids was accomplished using a Quattro LC triple quadrupole mass spectrometer (Micromass, UK). The positive electrospray ionization (ESI) conditions were as follows: capillary voltage, 3.2 kV; cone voltage, 25 V; source temperature, 120 ∘C; desolvation temperature, 200 ∘C; cone gas flow, 1 L min-1; and desolvation gas flow, 4 L min-1. To qualitatively determine the distribution of HGs in water column filtrates of Lake Schreventeich, all Bligh and Dyer extracts were analyzed in data-dependent mode with two scan events, where a positive ion scan (m/z 300–1000) was followed by a product ion scan of the base peak of the mass spectrum of the first scan event. Identification of HGs was based on comparison with published mass spectra (Bauersachs et al., 2009b). To improve the sensitivity of the measurement and therewith increase reproducibility, HGs were also detected via single ion recording (SIR) of their protonated molecules [M+H]+ (dwell time 234 ms) with m/z 575.5 (HG26 keto-ol), m/z 577.5 (HG26 diol), m/z 603.5 (HG28 keto-ol), m/z 605.5 (HG28 diol), m/z 619.5 (HG28 keto-diol) and m/z 621.5 (HG28 triol). Selected samples were analyzed in duplicate and fractional abundances of HGs as well as calculated HG ratios (e.g., HDI26, HDI28, HTI28) given in the text represent average values of these measurements. Quantification was done by integration of the peak area using the QuanLynx application manager.

Physical and biological data of Lake Schreventeich collected from late July to the end of October 2013 are summarized in Fig. 2. All investigated physical parameters (i.e., temperature, oxygen concentration and pH) show maxima in late July or at the beginning of August and gradually decline to yield minima in late October. Surface water temperatures ranged from 10.5 to 24.0 ∘C and were highest in late July (Fig. 2a). Oxygen concentrations in the surface waters varied from 2.5 to 7.6 mg L-1 with highest values occurring in late July and they subsequently declined over the investigated time interval to yield minimum values in late October (Fig. 2b). The pH values ranged from 7.18 to 7.79 and were comparatively high during the first half of the sampling campaign with values averaging 7.56 in August (Fig 2c). In contrast, the pH showed a significant drop by almost 0.2 units at the beginning of September and stayed around 7.32 throughout the first half of September before increasing again to values of ca. 7.50 at the beginning of October. Lake productivity was determined by measuring the amount of biomass present at time of sampling. Comparatively low amounts of biomass were found in late July with values of 11.6 mg L-1 that almost doubled in August with an average value of 20.7 mg L-1 (Fig. 2d). After a pronounced peak in the first half of September (maximum 50.1 mg L-1; average 35.3 mg L-1), biomass concentrations declined to an average value of 22.5 mg L-1 in October.

Heterocyst glycolipids were below detection limit in late July and early August. They were first identified in mid-August in relatively low abundances and gradually increased in late August to reach peak abundances in early to mid-September (Fig. 3). In late September, the relative abundance of HGs declined to reach comparatively low but constant values from mid- to late October. As shown in Fig. S1 in the Supplement, two structural isomers of 1-(O-hexose)-3,25-hexacosanediol (HG26 diol) and 1-(O-hexose)-3-keto-25-hexacosanol (HG26 keto-ol) generally dominated the HG pool and together they constituted 71–100 % (average 82.7 ± 7.2 %) of all heterocyst glycolipids over the investigated time interval. The early eluting HG26 diol, however, generally constituted only a minute fraction of all HGs (on average < 0.5 %). The heterocyst glycolipids 1-(O-hexose)-3,25,27-octacosanetriol (HG28 triol) and 1-(O-hexose)-3-keto-25,27-octacosanediol (HG28 keto-diol) were particularly abundant in late August with fractional abundances of up to 25 % but in general they contributed 6–17 % (average 12.3 ± 6.2 %) to the heterocyst glycolipid content of Lake Schreventeich. 1-(O-hexose)-3,27-octacosanediol (HG28 diol) and 1-(O-hexose)-3-keto-27-octacosanol (HG28 keto-ol) were below detection limit in water column filtrates taken before early September (Fig. 3) and they usually constituted a minor component of the total HG pool with fractional abundances of both compounds ranging from 0 to 13 % (average 4.9 ± 3.8 %).

It is interesting to note that the fractional abundance of all HG diols and triols declined over the investigated time interval, while the fractional abundance of their corresponding keto-ol and keto-diol varieties showed a concomitant increase (Fig. 3). For example, the fractional abundance of the HG26 diol was highest (> 70 %) in late August to early September and thereafter declined gradually to yield values around 50 % at the end of October. Over the same time period, the fractional abundance of the HG26 keto-ol significantly increased from 9 % at the end of August to 25 % in late October. Overall, similar trends were also observed for the HG28 triol and HG28 keto-diol as well as for the HG28 diol and HG28 keto-ol. It should be pointed out though that for the HG28 diol and the HG28 keto-ol this trend was less apparent, which may be due to the low analytical response and the resulting uncertainties in determining the contribution of both components to the total HG pool.

The distribution of HGs in surface sediments of Lake Schreventeich largely resembled those observed in the water column filtrates with the HG26 diol and HG26 keto-ol being most abundant (Fig. 3). Both components constituted ca. 81 % of the total HG pool with HG26 diol and HG26 keto-ol accounting for 64 and 17 % of all HGs, respectively. HG28 triol and HG28 keto-diol were the second most abundant types of HGs in Lake Schreventeich sediments, contributing 7 and 4 % of all HGs. Similar to the distribution of HGs in the water column filtrates, HG28 diol and HG28 keto-ol constituted only minor components of the HG pool with 5 and 3 %, respectively. Fractional abundances of HGs found in the core top sediments of Lake Schreventeich are thus well in line with those observed in water column filtrates, in particular with those obtained in mid-September.

Heterocyst glycolipids were below the detection limit in water column filtrates collected throughout July to early August and were first observed in mid-August. In general, the distribution of heterocyst glycolipids in Lake Schreventeich is well in line with those previously reported from nostocalean cyanobacteria (Bauersachs et al., 2009a; Wörmer et al., 2012) and likely suggests that members of the genera Anabaena and/or Aphanizomenon were part of the phytoplankton community of Lake Schreventeich in late summer 2013. This agrees well with microbiological studies of the phytoplankton community of other lakes in northern Germany, for which representatives of both genera have indeed been reported in abundance (Arp et al., 2013). The simultaneous increase in total HG abundances and aquatic biomass in early to mid-September (Figs. 2 and 3) may also suggest that heterocystous cyanobacteria constituted a significant component of the lake's phytoplankton.

We observed systematic changes in the distribution of heterocyst glycolipids in water column filtrates of Lake Schreventeich over the time interval investigated. The most apparent was a systematic decline in the fractional abundances of HG diols and the HG triol from mid-August to late October, which was significantly positively correlated with surface water temperature (Table 1). On the contrary, fractional abundances of HG keto-ols and keto-diols gradually increased from late August to the end of the sampling campaign. This increase in fractional abundances was significantly negatively correlated with changes in surface water temperatures (Table 1). Similar changes in the fractional abundances of HG26 and HG28 diols and keto-ols with growth temperature have previously been described from cultures of the N2-fixing heterocystous cyanobacteria Anabaena CCY9613 and Nostoc CCY9926 (Bauersachs et al., 2009a, 2014) and been explained as a physiological adaptation to compensate for greater gas diffusion rates of O2 at higher temperatures in order to keep the entry of atmospheric gases into the heterocyst at a minimum, which is considered a prerequisite for optimum N2 fixation. To quantitatively express these structural changes of the heterocyst cell envelope, Bauersachs et al. (2009a) introduced the HG26 index (heterocyst glycolipid index of 26 carbon atoms), which is defined as HG26=HG26keto-ol/(HG26diol+HG26keto-ol).

This notation, however, is somewhat counterintuitive as values of the HG26 index decline with increasing growth temperature. Therefore, we here used the HDI26 (heterocyst diol index of 26 carbon atoms), which in contrast to the HG26 index is positively correlated with temperature and defined as given below. It should be pointed out though that the HG26 index and the HDI26 have the same statistical significance. HDI26=HG26diol/(HG26keto-ol+HG26diol)HDI26=0.0224×SWT+0.4381;r2=0.93

In Lake Schreventeich, HDI26 values ranged from 0.89 in mid-August to 0.66 in late October (Fig. 4) and closely followed variations in surface water temperatures (Fig. S2 in the Supplement). For example, HDI26 values gradually declined over the investigated time period until mid-October and afterwards slightly increased again, in agreement with a rise in measured surface water temperature in late October. Least squares analysis of the data showed that variations in HDI26 values are strongly linearly correlated with surface water temperatures. As the HG28 diol and keto-ol as well as the HG28 triol and keto-diol showed similar changes in fractional abundances compared to the HG26 diol and the HG26 keto-ol, we also employed the HDI28 (heterocyst diol index of 28 carbon atoms) and the HTI28 (heterocyst triol index of 28 carbon atoms) in order to quantitatively determine changes in HG distributions with environmental parameters. Both indices were calculated as given in the following equations: HDI28=HG28diol/(HG28keto-ol+HG28diol),HDI28=0.0405×SWT+0.0401;r2=0.70,HTI28=HG28triol/(HG28keto-diol+HG28triol),HTI28=0.0288×SWT+0.2292;r2=0.78.

Similar to the HDI26, the HDI28 and the HTI28 closely followed measured surface water temperatures with absolute values of these indices gradually declining over the investigated time period from 0.82 to 0.42 and from 0.81 to 0.49, respectively (Fig. 4). Least squares analysis of the data demonstrates that both indices are significantly correlated with surface water temperatures, although correlations are generally less strong as compared to the HDI26. All three HG indices, however, seem to track temperature changes in the lake's surface waters in a similar fashion, albeit with slight differences in absolute values and trends between the individual indices (see Fig. S2 in the Supplement). One explanation for the slight offsets between the individual indices may be the contribution of heterocyst glycolipids from different cyanobacterial sources. Bauersachs et al. (2009a, 2014) as well as Wörmer et al. (2012) noticed that fractional abundances of heterocyst glycolipids may vary between different genera of heterocystous cyanobacteria and even within heterocystous cyanobacteria belonging to the same genus. Moreover, Bauersachs et al. (2014) observed that fractional abundances of HG26 and HG28 diols and keto-ols changed differently in Anabaena CCY9613 and Nostoc CCY9926, resulting in slightly deviating HG26 and HG28 values for each of the investigated species. As multiple members of heterocystous cyanobacteria (e.g., Anabaena and Aphanizomenon), adapting the composition of the heterocyst cell envelope in slightly different fashions, likely contributed to the total pool of HGs in Lake Schreventeich, absolute values of the different HG indices may have varied depending on the amount of heterocyst glycolipids contributed by each individual cyanobacterium. In this context it is interesting to note that the different HG indices show a similar trend with surface water temperatures but that HDI26 values are generally higher compared to HDI28 and HTI28 values, resulting in a deviation from the 1:1 line as shown in Fig. S3. HDI28 and HTI28 values, on the contrary, are very similar to each other and fall close to the 1:1 line, indicating that they may have the same biological origin.

When the different HG indices are plotted against environmental parameters other than surface water temperatures (Fig. 4), it is apparent that the HDI26 (p < 0.001; r2= 0.64) and the HTI28 (p < 0.05; r2= 0.42) are positively correlated with decreasing oxygen concentrations and that the HDI26 (p < 0.05; r2= 0.35) and the HDI28 (p < 0.05; r2= 0.35) also show a weak positive correlation with pH. However, these correlations are generally less significant and not as strong as observed for the correlation with surface water temperatures. It should also be noted that oxygen concentrations and pH are strongly correlated with surface water temperatures and that both parameters show a positive correlation with each other (Table 1). Therefore, the observed correlations between the different HG indices and oxygen concentrations as well as pH are likely indirect rather than indicating a statistically significant relationship between the individual environmental parameters and changes in the heterocyst glycolipid distribution. However, Kangatharalingam et al. (1992) reported that the heterocyst cell envelope of Anabaena flos-aquae increased in thickness when this cyanobacterium was grown under increased levels of oxygen stress and it can therefore not be excluded that environmental factors other than growth temperature may affect the distribution of heterocyst glycolipids in heterocystous cyanobacteria (although these authors did not analyze changes in the chemical structure of the heterocyst cell envelope). Additional investigations employing culture-dependent approaches and studying the effect of environmental parameters other than growth temperature will be needed to elucidate whether and to which extent oxygen concentrations and pH exert a control on the structural composition of the heterocyst cell envelope of heterocystous cyanobacteria.

The accuracy with which surface water temperatures of a given aquatic environment can be reconstructed is essential for any novel lipid thermometer. Based on replicate analysis of individual water column filtrates and surface sediments, the average analytical precision with which the HDI26 can be determined is ±0.006. Using the respective temperature calibration (see Eq. 3), this equals a standard error in temperature estimates of ±0.27 ∘C. The determination of HDI28 (±0.012) and HTI28 (±0.010) values is slightly less accurate than for HDI26, which may be due to the lower abundance of HG28 diols, triols, keto-ols and keto-diols in the analyzed water column filtrates, with the standard error in temperature estimates being ±0.30 ∘C for the HDI28 and ±0.34 ∘C for the HTI28. However, the overall analytical precision in the analysis of the different HG indices is of the same order of magnitude or even slightly better when compared to other well-established temperature proxies, such as the TEX86 and U37K′, and indicates that reconstructions of surface water temperatures using the HDI26 and other HG indices may be achieved with a relatively high accuracy. This is also suggested by analysis of the residual errors of the HG-estimated SWTs (calculated SWTs - measured SWTs), which are generally < 2 ∘C with mean standard errors of 0.97, 1.62 and 1.69 ∘C for HDI26-, HDI28- and HTI28-reconstructed SWTs, respectively, and without following a clear trend with SWT (see Fig. S4 in the Supplement).

In order to determine if the heterocyst glycolipid signal observed in the water column filtrates is transferred to the sedimentary realm, we also analyzed two surface sediments collected from Lake Schreventeich for their HG content. Sedimentary HG distributions were indeed very similar to those observed in water column filtrates with HG26 diol and HG26 keto-ol dominating over smaller quantities of HG28 triol and HG28 keto-diol as well as HG28 diol and HG28 keto-ol. It is interesting to note that the distribution of HGs in the two surface sediment samples most closely resembled the one observed during the period of maximum lake productivity and peak abundances of HGs in early to mid-September (Figs. 2, 3), suggesting that the preserved HGs were mainly produced during maximum activity of heterocystous cyanobacteria in Lake Schreventeich. HDI26 values of surface sediments from Lake Schreventeich averaged 0.791 ± 0.008. Using the temperature calibration obtained from the analysis of the water column filtrates, the HDI26 value translates into an average surface water temperature of 15.8 ± 0.3 ∘C. Considering the current accuracy for the detection of heterocyst glycolipids, the HDI26-based temperature reconstructed for Lake Schreventeich largely agrees with surface water temperatures measured from early to mid-September and thus during the time period of highest productivity of heterocystous cyanobacteria. Likewise, reconstructed surface water temperatures based on HDI28 (0.575 ± 0.018) and HTI28 (0.637 ± 0.012) values, obtained from the analysis of surface sediments of Lake Schreventeich and using their respective temperature calibrations, are 13.1 ± 0.4 and 14.1 ± 0.3 ∘C, respectively. Although slightly lower than the HDI26-based SWT estimates, both values again agree well with surface water temperatures measured during mid-September. Together, these observations suggest that the analysis of sedimentary HGs may allow reconstructing summer surface water temperatures in Lake Schreventeich and possibly also other lacustrine environments with sufficient export and incorporation of cyanobacterial-derived organic matter into the sediment.

Despite the good agreement between measured and reconstructed surface water temperatures, it should be pointed out that the recovered surface sediments most likely not only contained HGs produced during the investigated time interval but HG distributions probably reflect a time-integrated signal that covers several years. In addition, surface water temperatures of Lake Schreventeich are expected to vary over the course of a day and the obtained temperatures (though always recorded at the same time of the day) provide only a snapshot of the actual temperature variance of the lake. Parts of the uncertainties in the correlation of HG indices and surface water temperatures may in fact be related to the low number of diurnal temperature measurements but may be improved by continuous temperature logging of the lake's surface waters in future studies. As discussed above, contributions of HGs from heterocystous cyanobacteria with slightly different HG distribution patterns and absolute abundances of HGs may also result in the observed offsets between the HG-based SWT calculations. Nonetheless, the overall good agreement of HG distributions in surface sediments and water column filtrates seems to indicate that (1) HGs in Lake Schreventeich are largely produced in late summer, coinciding with blooms of heterocystous cyanobacteria, and that (2) HG-reconstructed surface water temperatures primarily reflect a summer signal in this temperate lake.

As mentioned previously, N2-fixing heterocystous cyanobacteria are a common component of the phytoplankton community in contemporary freshwater and brackish environments of polar to tropical latitudes, where they may form massive blooms during summer (Whitton, 2012). Likewise, HGs seem to be widely distributed in modern freshwater and brackish environments. They have been reported from surface sediments of several European and African lakes including Lake Ohrid, Lake Malawi and Lake Challa (Bauersachs et al., 2010) as well as in phytoplankton collected from a number of Spanish freshwater reservoirs (Wörmer et al., 2012). HG distributions dominated by HG26 diols and HG28 triols have been reported from core top sediments recovered from the Landsort Deep, Baltic Sea (Bauersachs et al., 2010). They have also been described in several microbial mats growing along the coast of the southern North Sea (Bauersachs et al., 2011; Bühring et al., 2014) and western Spitsbergen (Rethemeyer et al., 2010) as well as in an Icelandic hot spring (Bauersachs et al., 2013). A suite of HG26–HG28 diols, triols, keto-ols and keto-diols was detected in suspended particulate matter in the surface waters of 23 oligotrophic and eutrophic lakes in Minnesota and Iowa, USA (Schoon, 2013), while HG26–HG28 diols and keto-ols were present in variable abundances and distributions in microbial mats recovered from Shark Bay, Western Australia (Bauersachs et al., unpublished data).

The remarkably strong linear correlations found for the distribution of HGs in water column filtrates of Lake Schreventeich and its surface water temperatures indicate that HG distributions, in form of the HDI26 and other HG indices, may be well suited to track changes in water temperatures of the photic zone in freshwater environments. The generally good agreement of HG indices obtained from core top sediments of Lake Schreventeich with summer surface water temperatures furthermore suggests that the distribution of sedimentary HGs may also record surface water temperatures of lacustrine settings over time. In addition, it may indicate that no or only little selective degradation of HGs (e.g., diols vs. keto-ols) upon sinking and transport through the water column as well as during the incorporation into the sediment record occurred in this shallow lake system. At this point, however, it cannot be ruled out that microbial reworking may bias the initially synthesized HG signal in deeper lakes. Hence, additional studies determining degradation rates of individual HGs as well as changes in the overall HG distribution patterns with water depth will be necessary in order to elucidate whether and to which extend the HG inventory of lakes experiences early diagenetic alteration. Likewise, only limited information on the preservation potential of HGs over geological timescales exists. These components have been reported from Pleistocene Mediterranean sapropels as well as lacustrine deposits from the Oligocene Lake Enspel and the Eocene Messel oil shale (Bauersachs et al., 2010), indicating that they may readily preserve in the sediment record. However, detailed studies investigating the preservation of HGs in sedimentary sequences and their stability under varying environmental conditions are currently missing but will be essential to determine the robustness of HGs as lipid paleothermometers.

It has previously been demonstrated that changes in the distribution of HGs as a function of growth temperature can vary significantly between different cyanobacterial species as reported for Anabaena CCY9613 and Nostoc CCY9928 (Bauersachs et al., 2014). A finding that we confirmed for other nostocalean cyanobacteria such as Aphanizomenon sp. and Nodularia sp. in recent culture experiments (Bauersachs et al., unpublished data). The application of HDI26 and other HG-based indices may thus potentially be biased in lakes that are characterized by simultaneous growth of multiple species of heterocystous cyanobacteria, each modifying the composition of the heterocyst cell envelope in a slightly different fashion. It should also be pointed out that core top calibrations may not be applicable to accurately determine surface water temperatures in lake environments, in which the cyanobacterial community gradually changed over time.