Redox-dependent niche differentiation provides evidence for multiple bacterial sources of glycerol tetraether lipids in lakes

The North Basin of Lake Lugano is 288 m deep, mesotrophic, and meromictic, with a permanent oxycline at ∼100 m and a seasonal thermocline at 10 m to 20 m water depth (Fig. 2A). The brGDGT pool in suspended particulate matter (SPM) was dominated by noncyclic hexamethylated brGDGTs comprising three structural isomers (i.e., IIIa, IIIa′, and IIIa″; Fig. 1), which accounted for 46 to 94% of all brGDGTs in the water column (data for the less abundant brGDGTs are shown in SI Appendix, Figs. S2 and S3). The organic carbon-normalized total concentration of brGDGTs was low in near-surface waters (∼4 ng⋅g⁻¹ at 10 m depth), but increased to ∼100 ng⋅g⁻¹ just below the thermocline (Fig. 2B). Similarly, the fraction of intact polar brGDGTs that still retain the labile head group moieties present in living cells increased with depth from ∼30% at the base of the thermocline to >70% (Fig. 2B; see SI Appendix for further information). This is consistent with pronounced brGDGT synthesis in the deeper (anoxic) waters of the lake, which has previously been observed in other (seasonally) stratified lake systems (19–21).

We found profound changes in the relative contribution of brGDGTs IIIa, IIIa′, and IIIa″ throughout the water column (Fig. 2C). While brGDGT IIIa showed a continuous increase with depth across the redox transition zone (RTZ), IIIa′ was most abundant within oxygenated waters ∼20 m above the oxic−anoxic interface, and rapidly decreased below. In stark contrast to all other brGDGTs, the IIIa″ isomer, which was recently identified in anoxic sediments of another Alpine lake (22), occurred exclusively at depths below 90 m. Moreover, it was detected neither in nearby catchment soils nor in riverbed sediments collected from the lake’s watershed (SI Appendix, Fig. S5 and Table S1), suggesting that it is exclusively produced by anaerobic bacteria within the deeper waters (cf. ref. 22).

The spatially segregated production of these structurally similar brGDGTs in oxic and hypoxic/anoxic water masses is also reflected in their sinking fluxes, which we assessed by analyzing settling particles collected in sediment traps from three different depths throughout an annual cycle (Fig. 2C). Fluxes of IIIa′ increased ∼10-fold between the upper (20 m) and middle (85 m) trap but did not further increase below, indicating that it is predominantly biosynthesized in the lower part of the oxygenated water column. In contrast, IIIa″ was only present in settling particles below the RTZ (176 m), further attesting to its exclusive production within oxygen-deprived waters. IIIa, on the other hand, showed increasing fluxes across the RTZ, suggesting that it is produced by both aerobic and anaerobic, or facultative anaerobic, bacteria.

Distinct sources of these specific brGDGTs within Lake Lugano are also indicated by the stable carbon isotope signature of their alkyl moieties. In contrast to the bulk of brGDGT-derived alkanes that showed significantly decreasing δ¹³C values with depth (i.e., up to 10‰; Fig. 2D), the C isotope signature of alkane e that is exclusively contained in brGDGT IIIa′ (compare Fig. 1) remained unchanged across the RTZ, suggesting that it is only produced in the oxygenated part of the lake, from where it is exported to the sediment. To further explore the potential of brGDGT biosynthesis under oxic conditions, we amended near-surface SPM from Lake Lugano (10 m depth) with ¹³C-enriched organic substrates (see SI Appendix for details). Indeed, we observed pronounced production of the brGDGT IIIa′ and other C6-methylated isomers, providing evidence for an aerobic metabolism of their producers, whereas IIIa, IIIa″, and all other C5-methylated brGDGTs were not detectable at the end of the experiment (SI Appendix, Fig. S6).

The clear vertical segregation of the isomers IIIa′ and IIIa″, as well as their distinct stable carbon isotope signatures, strongly suggests that they are produced by distinct (groups of) microorganisms with different redox requirements. To elaborate links between brGDGT synthesis and microbial niche differentiation, we investigated vertical trends in the bacterial population by 16S rRNA gene sequencing. The change in brGDGT composition across the RTZ was accompanied by a marked shift in the bacterial community, with only 11% of the operational taxonomic units (OTUs) being shared between oxic (10 m to 80 m) and anoxic (90 m to 275 m) water masses. Furthermore, we compared OTU-specific DNA concentrations with the concentration of each individual brGDGT in the water column (see SI Appendix for details). A substantial fraction of the bacterial community (i.e., 30% of the OTUs) showed a strong correlation (r > 0.75) with the abundance of at least one of the major brGDGTs, whereas this correlation was much weaker when all brGDGTs were considered as a whole (r ≲ 0.5; SI Appendix, Fig. S7). Moreover, the bacterial clades that correlated with individual brGDGTs were often taxonomically distinct from one another (Fig. 3). In particular, the OTUs correlating well with brGDGT IIIa′ had taxonomic affiliations distinct from those corresponding to brGDGTs IIIa″, in support of brGDGT synthesis by multiple (groups of) bacteria that reside in different ecological niches (e.g., oxic vs. hypoxic/anoxic).

Acidobacteria are a plausible biological source of brGDGT, due to their membrane lipid composition (15). In Lake Lugano, 58% of all acidobacterial OTUs correlated well (r > 0.75) with one or more of the major brGDGT, and the diversity between the acidobacterial groups associated with different brGDGTs varied strongly (Fig. 3). This attests to a marked niche partitioning among members of the Acidobacteria that seems to resemble the differential vertical trends of certain brGDGTs in the water column (SI Appendix, Fig. S8). However, the vertical distribution of brGDGT also corresponded to OTUs belonging to many other bacterial phyla (Fig. 3 and SI Appendix, Fig. S7), suggesting that the trait of brGDGT synthesis extends beyond the phylum of Acidobacteria. Because ecological niches within the water column are often shared by diverse groups of microbes due to syntrophic relationships (23) or similar metabolic lifestyles (17), we cannot draw definite conclusions on the identity of brGDGT producers in Lake Lugano.

In Lake Lugano’s water column, the ¹³C content of brGDGTs (assessed by their alkyl moieties) provides insight into the potential C substrates utilized by their bacterial producers. Coinciding with the marked shift in the brGDGT distribution, the ¹³C content of most brGDGTs (i.e., the δ¹³C values of the alkyl moieties a, b, and c) decreased from −33 to −42‰ across the RTZ (Fig. 2D). This may, in part, reflect the vertical δ¹³C gradient of the C substrate utilized by brGDGT producers, and/or may be due to a distinct biosynthetic ¹³C fractionation expressed by different (groups of) source organisms. To elucidate this further, we measured ¹³C in the inorganic and organic C pools that may serve as substrates for brGDGT production within the water column. In fact, bulk suspended particulate organic matter (POM) showed vertical ¹³C trends similar to the brGDGTs (i.e., a decline across the RTZ; Fig. 2D), whereas δ¹³C values of dissolved inorganic and dissolved organic C did not change substantially with depth (SI Appendix, Fig. S9). We therefore conclude that brGDGT-producing bacteria in Lake Lugano are likely involved in the heterotrophic breakdown and assimilation of particulate organic C in the water column, which is in agreement with the known metabolic capabilities of acidobacterial isolates (24), and analogous to what has previously been suggested for peat bogs and soils (25, 26).

The more negative δ¹³C_POM values encountered at the RTZ, in turn, are predominantly due to the presence of methanotrophic bacteria that oxidize and incorporate ¹³C-depleted methane (−60‰) at and below the anoxic interface (27). We thus argue that at least some of the brGDGT producers thriving within the RTZ are part of a methane-based microbial food web (22, 28), in which they assimilate organic matter that is derived either directly from methanotrophic bacteria and/or from other organisms that have previously incorporated methanotroph-derived carbon (29–31).

A redox-dependent differentiation of the brGDGT-producing bacterial community as we observed in meromictic Lake Lugano may be a common feature in other stratified lake systems. Indeed, in 20% of the lake sediments investigated in this study, the characteristic alkyl moiety derived from brGDGT IIIa′ (alkane e) had substantially higher δ¹³C values (≥4‰) than those originating from brGDGTs IIIa and IIIa″ (alkanes d and f; Fig. 4A). This isotopic offset further seems to be associated with lakes that are more eutrophic and have relatively lower δ¹³C_d,f values. We therefore assume that the production of brGDGT IIIa′ in these lakes is also spatially separated from that of the bulk of the other major brGDGTs, similar to what we show for Lake Lugano (compare Fig. 2D). Moreover, in >50% of the lake sediments, the most common alkyl moieties of brGDGTs (i.e., alkanes a, b, and c) were depleted in ¹³C relative to total organic carbon (TOC) (i.e., ≥5‰ lower δ¹³C values) (Fig. 4B), again indicating a substantial contribution of brGDGTs produced in methanotrophic water and/or sediment layers, as we propose for Lake Lugano (see also SI Appendix, Fig. S5). Remarkably, these low δ¹³C_(a,b,c) values are consistently associated with an elevated trophic state of the lake (Fig. 4C), which, in turn, gives rise to an enhanced export production, as well as anoxia (Fig. 4D) and methane accumulation in bottom waters (32). Eutrophication and redox stratification thus likely promote the growth of brGDGT-producing microbes at the RTZ where methanotrophy may play an important role in microbial food webs. As indicated by the δ¹³C data, these deeper water-derived brGDGTs substantially contribute to the sedimentary brGDGT pool.

In Lake Lugano, the production of brGDGT IIIa″ is restricted to hypoxic/anoxic water (compare Fig. 2C). Corroboratively, we detected this compound preferentially in sediments of mesotrophic to eutrophic lakes (9 out of 11 cases; SI Appendix, Table S1), where biological oxygen consumption is likely to cause anoxia within the bottom water and/or the top layer of the surface sediment. The presence of brGDGT IIIa″ may thus be used as a redox indicator and, indirectly, as a proxy for the trophic state. Taken together, redox-dependent differentiation of brGDGT-producing bacterial communities, as shown here for Lake Lugano, seems to be prevalent in a substantial fraction of lake systems. Compositional shifts in sedimentary brGDGT records may thus not only reflect the ability of bacteria to adapt the chemical properties of their membrane in response to temperature and pH (5, 6), but may also be strongly dependent on changes in the trophic state (33) and the microbial community composition.

Observed correlations between brGDGT distributions in recent lake sediments and MAT gave rise to transfer functions that allow for the quantitative reconstruction of paleoclimatic conditions from lacustrine sediment archives (e.g., ref. 9). For instance, the Methylation Index of brGDGTs (MBT′_5me, reflecting the relative amount of C5-methyl branches; ref. 4) has been successfully applied to East African lake sediments, where it showed a strong correlation with MAT (r = 0.96; ref. 34). However, this correlation was poor for the surface sediments of the lakes investigated here (r = 0.33, n = 36). Several factors may affect the fidelity of brGDGT-based temperature proxies in lake sediments. One major concern is that terrigenous brGDGTs may contribute to the sediment through soil organic matter inputs. The brGDGTs from soils have a composition that is markedly distinct from those produced within lakes of the same climate regime (8), and, as a consequence, their admixture to the aquatically produced brGDGT pool can substantially bias temperature estimates derived from lake deposits. Another factor relates to the vertical temperature distribution within the water column. Besides a possible input of soil-derived brGDGTs, the water depth from where the in situ-produced brGDGTs are predominantly transferred to the sediments will ultimately determine the “temperature imprint” of brGDGT in the surface sediment. In Lake Lugano, the largest part of the brGDGT sinking flux can be attributed to production in subthermocline (i.e., hypolimnetic) water (compare Fig. 2C), and thus the pool of brGDGTs finally reaching the sediment reflects isothermal deeper water, rather than seasonal surface water conditions (refs. 19 and 35 and SI Appendix, Fig. S10). On the other hand, in lakes with less-pronounced deeper water production, the sedimentary brGDGTs may more dominantly derive from surface water, shifting the proxy record toward higher temperatures (i.e., increased MBT index values). The relative contributions of soil-, (near) surface water-, and hypolimnion-derived proxy signals to the sediments may consequently vary largely among different study sites, depending on hydrochemical parameters (e.g., trophic state/redox conditions), morphological characteristics (i.e., maximal depth), and the depositional environment (i.e., soil organic matter input), which is likely to confound the quantitative relationships between atmospheric MAT and the sedimentary brGDGT record. Knowledge of the predominant brGDGT sources, therefore, is of prime importance for the use of brGDGTs in paleoenvironmental reconstructions.

Our data from Lake Lugano show that the brGDGTs produced at the RTZ—likely by utilization of methane-derived C compounds—are much more depleted in ¹³C than those synthesized in both the oxic part of the water column (≳35‰; compare Fig. 2D) and soil [−25 to −31‰ (25, 26); see also SI Appendix, Table S1]. Such ¹³C-depleted sedimentary brGDGT pools in eutrophic lakes may thus be indicative of proxy signals that almost exclusively reflect variations in deeper water temperatures, and may hence be largely unaffected by above-mentioned differential brGDGT inputs from surface water and/or soils. To further investigate the temperature response of sediments with a deeper water-dominated brGDGT source, we reassessed the MAT–brGDGT correlation in our dataset for different δ¹³C domains. We found that, in mesotrophic to eutrophic lakes with low δ¹³C_(a,b,c) values (−45 to −36‰), the correlation between MBT′_5me (and related proxy indices; SI Appendix, Fig. S11) and MAT was indeed significant (r = 0.67, n = 15, P < 0.01), and much stronger than in oligo- to mesotrophic lakes, where the δ¹³C_(a,b,c) was higher (r = 0.16, n = 20, P = 0.5). This correlation was again substantially improved when only shallow lakes with a maximal depth of <40 m were considered (r = 0.87, n = 10, P < 0.01; Fig. 5B), in which the RTZ may overlap with the thermocline, and the low-¹³C brGDGTs thus probably reflect thermocline water temperatures. In contrast, MBT′_5me proxy values were more or less constantly low among the deeper lakes (>40 m) (Fig. 5D), which we attribute to the decoupling of atmospheric air and hypolimnetic water temperatures in lakes of the cool-temperate climate zone (i.e., MAT ≲ 11 °C; ref. 36). In warmer climates, on the other hand, the temperature of hypolimnetic water is, in fact, linearly related to that of the atmosphere (37, 38), and hypolimnion-derived brGDGTs from deeper lakes will therefore be more responsive to interannual MAT variations than in the Alpine lakes investigated here.

We thus conclude that deeper water communities of brGDGT-producing bacteria thriving at the RTZ “reset” the sedimentary brGDGT “thermometer” to (near-)bottom water temperature, thereby mitigating the interfering effects of surface water- and/or catchment soil-derived brGDGT signatures. Depending on climate zone and water depth, such settings, which can be identified by the C isotopic imprint of brGDGTs inherited from biogenic methane, are expected to yield reliable records of climate variability, and are preferable sites for the application of brGDGT-based paleotemperature proxies in lakes.

SPM was collected from Lake Lugano (northern basin) close to the deepest location (SI Appendix, Fig. S12) in September 2014 by large-volume in situ filtration of lake water (∼50–300 L), using a double layer of glass fiber (GF) filters (2.7 and 0.7 µm; Whatman GF/D and GF/F). The filters were frozen on dry ice immediately after sampling and stored at −80 °C until further processing. Settling particles were collected at the same location, using cylindrical sediment traps (d = 10 cm) deployed in 1990/1991 at three depths (20, 85, and 176 m), and were sampled up to three times per month.

Lake surface sediments (SI Appendix, Fig. S13 and Table S1) were obtained by gravity coring during multiple field campaigns (2008 to 2014). The core tops (upper ∼5 cm) were then subsampled, homogenized, and freeze-dried before extraction. We further collected eight to ten top-soil samples (0 cm to 10 cm) within a perimeter of ∼500 m at 19 of the 36 investigated lakes. The individual soil samples from each lake site were pooled, freeze-dried, sieved (2 mm), and homogenized with a mortar before extraction. Additionally, we collected sediment samples (0 cm to 10 cm) from two main tributaries (Cuccio and Casarate) of Lake Lugano at locations close to the river mouths (SI Appendix, Fig. S12), which were processed in the same way as the soil samples.

Sediments and soils were extracted three times by accelerated solvent extraction (ASE300; Thermo Fisher Scientific) (5). For SPM samples, we employed a sequential solvent extraction and acid hydrolysis procedure to assure complete recovery of brGDGTs from living cells (intact polar brGDGTs; see also SI Appendix), and to avoid losses and biases associated with modified Bligh−Dyer extraction (39). Briefly, the freeze-dried SPM was extracted by ultrasonication with methanol and dichloromethane as solvents, and the residual material was subsequently subjected to acid hydrolysis. To maximize lipid recovery for compound-specific stable C isotope analysis of brGDGTs, SPM from selected depths (representing ∼150 L of filtered water) was subjected to acid hydrolysis before ultrasonic solvent extraction.

The brGDGT-containing lipid fractions were prepared as previously described (39), spiked with an internal standard (40), and analyzed by ultra-high-performance liquid chromatography (UHPLC) positive ion atmospheric pressure chemical ionization mass spectrometry in selected ion monitoring mode. For all samples from Lake Lugano, analytical separation of GDGTs was achieved using two UHPLC columns in series (ACQUITY UPLC BEH HILIC, 130 Å, 1.7 µm, 2.1 mm × 150 mm; Waters) (41), whereas, for all remaining sites, we used an array of four HPLC columns (Alltima Silica, 100 Å, 3 µm, 2.1 mm × 150 mm; W. R. Grace & Co.) (4). Both “high-resolution” UHPLC setups reliably separate C5- and C6-methylated brGDGT isomers (41). Response was monitored regularly using a known mixture of the internal standard and crenarchaeol.

The δ¹³C content of TOC and POM were determined after acidification of the sample (42) by elemental analysis followed by isotope ratio mass spectrometry (IRMS) (22). For stable carbon isotope analysis of brGDGTs, we subjected bulk brGDGT fractions to ether cleavage (57% HI), and measured δ¹³C values of the released alkyl moieties by gas chromatography combustion (GC/C) IRMS as previously described (22), with a few modifications (SI Appendix).

DNA was extracted from fractions of GF filters (43), equivalent to 2 L to 7 L of lake water. A two-step PCR approach (Nextera; Illumina Corp.) was employed for library preparation, using the universal primers 515F-Y (5′-GTGYCAGCMGCCGCGGTAA) and 926R (5′-CCGYCAATTYMTTTRAGTTT-3′) (44) that target the V4 and V5 regions of the 16S rRNA gene. The individual samples were indexed, pooled at equimolar concentrations, and sequenced on an Illumina MiSeq V2 platform using the 2 × 250 bp paired-end protocol (Microsynth Inc.). After initial quality control and bioinformatical processing (SI Appendix), taxonomy was assigned to OTUs by comparison with the SILVA 128 SSU reference database (https://www.arb-silva.de/documentation/release-128/). To compare vertical trends of brGDGTs and microbial taxa, relative OTU abundances were converted to OTU-specific DNA concentrations (nanograms per liter) by multiplication with the total amount of extracted DNA, and subsequently subjected to correlation analysis in R (see SI Appendix for details).

Abundance-filtered 16S rDNA sequences are deposited at GenBank (National Center for Biotechnology Information) under accession numbers MH111698 through MH113143. Raw bacterial OTU abundances are provided in Dataset S1. The brGDGT data from Lake Lugano and other Alpine lake sediments and catchment soils are included in Dataset S2. R code for correlation analysis is available at https://github.com/yukiweber/phylo.lipids.