Millennial-age GDGTs in forested mineral soils: C-based evidence for stabilization of microbial necromass

Understanding controls on the persistence of soil organic matter (SOM) is essential to constrain its role in the carbon cycle and inform climate-carbon cycle model predictions. Emerging concepts regarding formation and turnover of SOM imply that it is mainly comprised of mineral-stabilized microbial products and residues, however, direct evidence in support of this concept remains limited. Here, we introduce and test a method for isolation of isoprenoid and branched glycerol dialkyl glycerol tetraethers (GDGTs) – diagnostic membrane lipids of archaea and bacteria, respectively for subsequent natural abundance 5 radiocarbon analysis. The method is applied to depth profiles from two Swiss pre-alpine forested soils. We find that the ∆C values of these microbial markers markedly decrease with increasing soil depth, indicating turnover times of millennia in mineral subsoils. The contrasting metabolisms of the GDGT-producing microorganisms indicates it is unlikely that the low ∆C values of these membrane lipids reflect heterotrophic acquisition of C-depleted carbon. We therefore attribute the C-depleted signatures of GDGTs to their physical protection through association with mineral surfaces. These findings thus 10 provide strong evidence for the presence of stabilized microbial necromass in forested mineral soils.

the early Miocene (Heumann and Litt, 2002)(62.3% OC, bulk F 14 C = 0.003) were used for assessment and validation with regard to contamination.

GDGT isolation for radiocarbon measurement
Despite the relative ease of detection of GDGTs using modern HPLC-mass spectrometry (MS) techniques, one challenge in the radiocarbon analysis of GDGTs in soil and sediment samples is their low abundance, with ambient concentrations of 90 brGDGTs and isoGDGTs that are typically in the range of 10 to 1000 ng gdw −1 and 1 to 100 ng gdw −1 (grams dry weight) soil, respectively (Weijers et al., 2006b). A separation of individual GDGTs of the soil samples used in this study would require on average 3000 g of soil to reach the minimum recommended mass (∼ 15 µg C) for high-precision compound-specific radiocarbon analysis (Haghipour et al., 2018). As extracting several kg of material is impractical, focused instead on pooled isolation and 14 C measurement of isoprenoid GDGTs and branched GDGTs, respectively, at the compound class level, due 95 to the common putative biological percursors and biosynthetic formation pathways for each compound class (Schouten et al., 2013). For this study, the pooling of the GDGTs reduced the required initial sample size to a maximum of 500 gdw of soil. The extraction and purification of the compounds prior to HPLC analysis purification of the compounds prior to HPLC analysis followed a procedure that is similar to that applied to samples processed for quantification of GDGTs (Freymond et al., 2017).
In brief: 100 lipids were extracted from dried soil samples using a microwave (CEM MARS 5) or an Energized Dispersive Guided Extraction (CEM EDGE) system. No difference in performance was observed for the different extraction systems. Samples were processed in batches of roughly 15 to 20 g of material. For microwave extraction the samples were transferred to the extraction vessels and covered by a dichloromethane (DCM):methanol (MeOH) 9:1 (v/v, 25 ml) solvent mixture. Extraction temperature was programmed to ramp to 100 • C in 35 min and is subsequently held for 20 min. For EDGE extraction 25 ml DCM:MeOH 9:1 with 82% solvent B. The fraction collection is solely based on retention times with the isoprenoid fraction being collected from 14.5 to 26 min and the branched fraction from 33 to 43 min ( Figure 1). The retention time is recurrently monitored to avoid undetected drifts. The injection volume is set to 15 µl corresponding to total GDGT amounts of 100 to 300 ng ml −1 .
Each sample was injected 10 times and fractions were pooled afterwards. The isolated compound classes and the subset of the initial polar fraction set aside previously were analyzed for purity and quantification using the same HPLC system coupled to a quadrupole mass spectrometer (Agilent 6130) according to Hopmans et al. (2016). The isolated fractions were dried and 125 transferred into 0.025 ml tin capsules (Elementar 03951620). The capsules containing each sample were measured using an elemental analyzer coupled to a gas-ion-source equipped accelerator mass spectrometer (EA-AMS) (Haghipour et al., 2018) at the laboratory of Ion Beam Physics at ETH Zürich (Synal et al., 2007;Ruff et al., 2007). In all cases, sample sizes were > 15 µg C.

Soil turnover model
130 Turnover times of the individual compounds are calculated based on a steady state two-pool box model (e.g., Trumbore et al., 1996;Torn et al., 2009;Schrumpf and Kaiser, 2015;van der Voort et al., 2019). This model assumes two homogenous pools with a first-order decay rate, a fast-cycling and a passive pool. For each of the pools the F 14 C is calculated independently (equation 1), where F 14 C pool(t) is the radiocarbon signal of the respective pool in the sampling year t, lag is the number of years between CO 2 fixation in plants and plant litter entering the soil, λ is the radioactive decay of 14 C (1/8267 years), and 135 k pool is the decomposition rate constant.
The fraction-weighted sum of the F 14 C of each of the pools is the modelled F 14 C of the sample and depends on the decomposition rate constants of each pool k 1 and k 2 , as well as the relative size of the two pools. The ∆ 14 C of atmospheric CO 2 was taken from Hua et al. (2013) from 1950to 1986and from Hammer and Levin (2017 for the years thereafter.

Method Validation
Repetitive preparation of samples with 10 injections each reveals a recovery efficiency of 0.85 ± 0.05. Analysis of isolated fractions on a quadrupole mass spectrometer operated in scan mode (Agilent 6130) for all masses between m/z 500 and 1500 reveals that more than 95% of compounds in either fraction are comprised of masses assigned to GDGTs (Figure 1).
The extraneous contamination added in the preparatory process is assumed to be of constant mass m c and radiocarbon signature F 14 C c . Therefore, the measured signal F 14 C m is a mixture of the sample and the contaminant according to equation 2: F 14 C s and m s are the true radiocarbon signal and carbon mass of the sample. The measured F 14 C m changes depending 150 on the mass of the sample, as smaller masses are more strongly affected by the constant contamination. We assume that in samples with a bulk radiocarbon signal that is either completely modern or does not contain any 14 C at all the compoundspecific radiocarbon value is similar to the bulk. Therefore, a radiocarbon-modern sample, i.e., the topsoil composite, and the radiocarbon-dead lignite were prepared and measured repeatedly with different concentrations. The best fit for F 14 C c and m c to match the observed F 14 C m for both sets of measurements is calculated according to Haghipour et al. (2018).

155
The blank assessment ( Figure 2) yields a contamination of 2.62 ± 0.79µg C with a fraction modern of 0.59 ± 0.18, which is in range of previously determined contamination introduced by HPLC separation of lipids (e.g., Shah and Pearson, 2007;Birkholz et al., 2013). The impact of the constant contamination decreases as the sample mass increases. Therefore, the limit towards large carbon masses of the fitted curve is equivalent to the radiocarbon signal of the sample unaffected by extraneously 160 introduced carbon. For both samples, this limit and hence the compound F 14 C differs from the bulk F 14 C of the initial material. In the topsoil reference the compounds are depleted in radiocarbon (F 14 C = 0.94) with respect to the source, in the lignite the GDGTs are enriched (F 14 C = 0.06).
The recommended sample size to reach a precision <5% varies depending on the age of the sample. For samples with a radiocarbon age < 1800 years (F 14 C > 0.8) a size of 20 µg C is sufficient to reach the desired precision, while samples older 165 than 6000 years (F 14 C < 0.5) require at least 50 µg C. These uncertainties are taken into account when considering the GDGT 14 C results for the soil samples measured in this study.

Vertical Distributions of GDGTs
In the pre-alpine soil from Beatenberg, concentrations of GDGTs are generally highest in the topsoil, where the isoprenoid and branched GDGTs are 10 and 38 µg gdw −1 respectively, whereas corresponding concentrations in the top soil layer of 170 the Lausanne soil are much lower, 0.6 and 2 µg gdw −1 , respectively ( Figure 3). The concentration of both groups of GDGTs decreases sharply with increasing soil depths, with approximately ten times the abundance of isoGDGTs and brGDGTs in the top 5 cm than a few centimeters below. In contrast, isoprenoid and branched GDGTs concentrations normalized to organic carbon (OC) content increase with depth in the Beatenberg soil from 47µg gOC −1 for isoGDGTs and from 175 µg gOC −1 for brGDGTs in the top 5 cm to 273 µg gOC −1 and 80 µg gOC −1 , respectively, between 20 and 40 cm depth. In the Lausanne soil 175 profile, the OC-normalized isoprenoid and branched concentrations drop from 10 µg gOC −1 and 39 µg gOC −1 , respectively, in the top 5 cm to 4 and 13 µg gOC −1 between 10 and 20 cm depth, and then increase to 14 and 12 µg gOC −1 between 60 and The relative abundance of the individual brGDGTs also changes with soil depth, as reflected in the MBT' 5M e and CBT' ratio (De Jonge et al., 2014). The MBT' 5M e index does not exhibit significant variability in either soil profile, while the CBT' index 180 increases with soil depth, especially in the Lausanne soil indicating a shift towards 6-methylated GDGTs ( Figure 3).

Radiocarbon variations
The GDGT fractions prepared for AMS measurement contained between 30 and 80 µg C, except for the brGDGTs in the 10 to 20 cm depth interval in Beatenberg and the iso-and brGDGTs from 60 to 80 cm the Lausanne soil, which range between 15 and 20 µg C. The results of the radiocarbon measurements are shown in Figure  to the free particulate organic carbon (free POC) and the high density fraction is interpreted as mineral-associated POC. Both fractions do not differ by more than 40 ‰ in the top 20 cm of either soil profile, but in the lowest depth interval the fractions diverge, with markedly lower ∆ 14 C values the for high density fraction, and values similar to DOC for the free POC fraction.
In both soils, the iso and brGDGTs exhibit similar or lower ∆ 14 C values than the high density fraction, mineral-associated organic matter fraction.

Radiocarbon derived turnover times of GDGTs
Turnover times of the compounds are calculated based on a two-pool model that requires three parameters to be fitted: the turnover time of the fast-cycling pool, the turnover time of the passive pool and the proportion of the fast-cycling pool. As only one radiocarbon measurement per compound and depth interval is available, two of the parameters need to be estimated, while one can be fitted accordingly. We use the proportion of the labile low-density fraction of the samples (Van der Voort et al., 2017) to constrain the size of the fast-cycling pool. The turnover time of the fast-cycling pool can be estimated accordingly as the single-pool turnover time of the light fraction. Alternatively, the GDGT turnover in topsoil based on stable carbon isotopes has been shown to be similar to short-chain fatty acids (Weijers et al., 2010;Huguet et al., 2017). Thus, the turnover time of these compounds based on a single-pool box model can also be used to constrain turnover time of the fast-cycling pool of GDGTs. For simplicity, a lag-term addressing the time between atmospheric carbon fixation and input into the soil is not used, 215 as it is shorter than a decade (Solly et al., 2018) and its potential influence is hence already covered by the range of the turnover Compared to prior methods to achieve individual isoGDGT separation by HPLC (Smittenberg et al., 2002;Ingalls et al., 2006), the introduced method isolates GDGTs only at the compound-class level, hence potential radiocarbon variations among GDGT isomers are not discernable. However, previous analyses of stable carbon isotopic as well as radiocarbon analysis of GDGTs on a molecular level do not show significant differences between the individual isoprenoid or branched GDGTs, respectively (e.g., Ingalls et al., 2006;Shah et al., 2008;Oppermann et al., 2010;Weber et al., 2015). This implies similar metabolisms 230 for brGDGT-producing organisms and also for microbial communities that synthesize isoGDGTs. Consequently, pooling of isomers within a compound class according to their respective microbial domain (bacteria, archaea) seems reasonable, particularly given the practical constraints imposed by their low abundance in many terrestrial (and aquatic) environments. The introduced method requires only a single normal-phase isolation step using the same columns that are used for quantification of GDGTs (Hopmans et al., 2016), minimizing the time required for sample preparation and without extensive adjustments to the 235 analytical HPLC set-up. The calculated contamination is in range of the blank assessment by Ingalls et al. (2006), but higher than the extraneous carbon observed by Birkholz et al. (2013). However, the blank assessment in Birkholz et al. (2013) is based only on a modern non-GDGT standard (cholesterol), potentially leading to an underestimation of the sample preparation blank.
The GDGT-specific ∆ 14 C values of the top soil and lignite samples used as "modern" and "fossil" endmembers for blank 240 assessment did not yield values that fully matched those expected given their age. In case of the soil, different ∆ 14 C values of the GDGTs compared to bulk OC are to be expected due to the heterogeneous nature of soil organic matter, however for lignite sample that is of geologic age (> 30 Ma), all components would be expected to be radiocarbon-dead. A preliminary batch of lignite that was extracted yielded 18 µg C of isoGDGTs and 48 µg C of brGDGTs, with corresponding ∆ 14 C values of the resulting isolated compounds of of -960‰ and -980‰, respectively. The second batch of lignite used to assess constant con-245 tamination was prepared 4 months later and shows ∆ 14 C values consistently higher than -950‰ (figure 2). This shift towards higher ∆ 14 C values likely reflects contamination resulting from sample-to-sample carry-over on the HPLC. Although this is adressed in the blank assessment, this highlights the importance of repeated blank assessment in order to control for variations in carry-over and other potential sources of contamination (e.g., column bleed) over time. Careful assessment of compound purity is also important to ensure robust isotopic determination.

Radiocarbon constraints on the origin and turnover of GDGTs in soils
Our study reveals low ∆ 14 C values, with corresponding radiocarbon ages of up to 6000 years for GDGTs in forested soils.
These 14 C characteristics are similar to those of the mineral-associated OM (from density fractionation), as well as long-chain, higher plant wax-derived n-fatty acids and n-alkanes. As GDGTs are microbial membrane lipids, these findings reveal the presence of 14 C-depleted, millennial age microbial residues as a component of organic matter in deeper soils. There are two 255 possible pathways leading to these old apparent radiocarbon ages: (1) active GDGT-producing heterotrophic soil microbial communities in deeper soils are utilizing pre-aged SOM as a carbon source, and accrue this signal with continuous community turnover. Alternatively, (2) upon cell death these microbial lipids are stabilized for millenia, likely via interaction with soil minerals. We first consider the first explanation: The ∆ 14 C values of living organisms, and their constituent lipids, directly reflect that of their metabolic carbon source 260 as, unlike stable isotopes, they are impervious to biological fractionation effects (Ingalls and Pearson, 2005). Upon death of the organism, radioactive decay leads to depletion in 14 C contents. Consequently, the 14 C contents of iso-and brGDGTs should reflect that of the carbon source of their biological precursors. IsoGDGTs are known to be produced by Thaumarcheota and Euryarcheota (Schouten et al., 2013). The specific microbes that produce brGDGTs are yet to be identified, there is strong evidence that the precursor organisms are heterotrophic bacteria (Pancost and Damsté, 2003;Weijers et al., 2010), 265 with Acidobacteria amongst the candidate phyla . For heterotrophic bacteria, potential carbon sources include DOC leached from the organic layer, exudates from root systems or organic matter that has accumulated during soil development. The activity of soil microbial communities has often been assayed using phospholipid-fatty acids (PLFAs), as phospholipids are only found in living cells and thus serve as biomarkers for viable microbial communities (e.g., Tunlid and White, 1991). Compound-specific radiocarbon analyses of PLFAs have shown that soil microbes can use a variety of carbon acids that likely reflect active microbial communities (Figure 4, 5). The markedly lower ∆ 14 C values of both isoGDGTs and brGDGTs at depth would require that both groups of precursor organisms, i.e., Archaea and Bacteria, occupy specific niches using metabolic strategies that enable them to utilize stabilized, aged carbon. The precursor organisms of isoGDGTs 280 in soils are known to be mainly comprised of crenarchaeota, i.e., chemoautotrophic nitrifiers using soil CO 2 as substrate (Leininger et al., 2006;Urich et al., 2008;Weijers et al., 2010;Damsté et al., 2012) and acetotrophic methanogens (Weijers et al., 2010). Contributions from the latter organisms in the studies soils are likely minor as the soils are not strictly anaerobic (Walthert, 2003)). This is also supported by GDGT-0/Crenarchaeol ratios, that differ sharply from those in soils and sediments dominated by methanogens (Blaga et al., 2009;Weijers et al., 2010;Naeher et al., 2014). Soil-respired CO 2 has relatively 285 high ∆ 14 C values (Gaudinski et al., 2000;Liu et al., 2006), and thus it seems highly unlikely that 14 C-depleted signatures of isoGDGTs in the deeper soils results from metabolism of an old C substrate by active soil microbial communities. By analogy, the 14 C-depleted characteristics of brGDGTs is difficult to reconcile with heterotrophic consumption of pre-aged C.
Overall, the contrasting metabolisms of the GDGT precursor organisms (primarily autotrophy for isoGDGTs and heterotrophy for brGDGTs), yet similar (and low) ∆ 14 C values for both compound classes, argue against an origin of the GDGT signals 290 from microbial growth at depth.We therefore conclude that uptake of pre-aged carbon by active soil microbial communities is unlikely to be the cause for the 14 C-depleted GDGT signatures.
We next consider the long-term stabilization microbially-derived carbon as the source of 14 C-depleted GDGT signatures.
This implies that microbial residues persist in soils for millennia, lending support to emerging concepts that microbial necromass comprises an important component of older SOM (e.g., Lehmann and Kleber, 2015;Liang et al., 2017). Long-term 295 persistence of GDGTs could arise from their stabilization by soil minerals at greater soil depths. The amphiphilic nature of lipids such as GDGTs, with both polar and hydrophobic components, promotes the association with mineral surfaces, and therefore may afford physical protection from degradation (Jandl et al., 2004;Kleber et al., 2007;von Lützow et al., 2008;Van der Voort et al., 2017). By comparison, in surface soils with high organic matter contents and less availability of reactive mineral surfaces, GDGTs are continuously produced and degraded, which results in a younger mean radiocarbon age and evi-300 dence for turnover on decadal timescales (Weijers et al., 2010). This explanation agrees with conceptual models of soil organic matter dynamics whereby older SOM in deeper soils primarily consists of microbial metabolites that are stabilized by their interaction with mineral surfaces (Schmidt et al., 2011;Lehmann and Kleber, 2015). Given the structural resemblance between brGDGTs and isoGDGTs, and hence similar propensity to associate with mineral surfaces , we consider this a more likely explanation for their similarly old 14 C ages than "niche metabolisms" of different precursor organisms. The older GDGT age 305 in the the Cambisol at Lausanne compared to the subalpine Podzol at Beatenberg with a bleached eluvial horizon also supports this conclusion. The higher contents of clay and highly reactive amorphous Fe and Al-oxides and hydroxides of the former (Table 2) are known to play a key role in the sorptive stabilization of SOM (Kaiser and Guggenberger, 2003;Kleber et al., 2007) Overall, 14 C characteristics of iso and brGDGTs and the inferred turnover times that are far longer than those of discrete POM 310 (free light density fraction) and signature lipids of active microbial communities (short-chained fatty acids), but similar to those of plant-derived long-chain n-alkanes and fatty acids ( Figure 5), serve as strong evidence for the presence of mineral-stabilized microbial necromass in the studied forested mineral soils.

Implications for application of GDGTs as molecular proxies and soil tracer biomolecules
In addition to the insights into soil carbon turnover, the observed 14 C signatures of GDGTs in the two soil profiles carry impli-315 cations for their application as proxies of environmental conditions and as tracers of soil carbon input to aquatic environments.
Several studies have shown that soils comprise a significant, and often the dominant, component of terrestrial organic carbon exported in the suspended load of rivers to lake and ocean sediments (e.g., Tao et al., 2015;Vonk et al., 2019;Hein et al., 2020).
Prior analyses of branched GDGTs in sedimentary archives have revealed older GDGT ages than depositional ages (Smittenberg et al., 2005;Birkholz et al., 2013) that may reflect intermittent storage during transport or export of deeper mineral soil 320 carbon suggesting a lag between production and deposition. Our findings suggest that this may be a consequence of protracted storage in and mobilization from deeper mineral soils. This reinforces the value of brGDGTs as tracers of soil carbon, but implies that much of the signal may be sourced from deeper soil layers, with corresponding GDGT proxy signals reflecting environmental conditions at the time that they were microbially produced.

325
We modified and validated a normal-phase HPLC method to isolate isoprenoid and branched GDGTs at the compound class level for radiocarbon analysis. Although further refinements in the method would be desirable, this new approach yields reliable GDGT 14 C measurements on sample sizes > 20 µg C that have enabled novel questions to be addressed concerning the provenance and turnover of this key suite of microbial lipids. In addition to its application to questions of soil C cycling, the streamlined method opens up new opportunities to further explore the biogeochemical and paleoclimate significance of this 330 intriguing yet enigmatic class of lipids.
Application of the method to depth profiles for two well-studied sub-alpine soil profiles in Switzerland reveals a marked decrease in 14 C contents of both isoGDGTs and brGDGTs with depth, with resulting model estimates for GDGT turnover times of 2000 to 6000 years in deeper mineral soils. These old ages for archaeal and bacterial membrane lipids provides compelling evidence for stabilization of microbial necromass in soils that contributes to the long-term C storage. Through 335 comparison with parallel 14 C data for soil density fractions and other hydrophobic lipid biomarkers, we attribute the stability of GDGTs to protection via association with reactive mineral surfaces underlining the crucial role of microbial processes in soil C cycling and stabilization.
Our findings also provide motivation for further work to validate our interpretations and assess the broader significance of the current limited suite of observations. For example, comparison of the proportions and isotopic signatures of intact polar lipid 340 GDGTs relative to the "core" lipids measured here could shed light on the significance active GDGT-producing communities residing at a specific soil depth versus remnants of past microbial activity (necromass). Furthermore, while concentrations of isoprenoid as well as branched GDGTs commonly decrease with increasing soil depth (Huguet et al., 2010;Yamamoto et al., soil profiles that exhibit such sub-surface concentrations peak would be informative and provide context for our observations in the two Swiss soil profiles. Further insights into the provenance and turnover of brGDGTs might be gained from in-depth assessment of molecular distributions and associated proxy indices (De Jonge et al., 2014), that may reflect changes in current or past microbial communities, or imply differences in susceptibility to degradation. Despite the presence of GDGTs as trace constituents of SOM, their unequivocal microbial origin, distinctive chemical structures, and environmental properties that 350 their distributions encode render them powerful tracer compounds and molecular proxies. Here, we demonstrate that when also constrained with natural abundance 14 C , these compounds provide a new window into the role of microorganisms in soil carbon cycling.
Code and data availability. The data set and script for the turnover model used in this study is available at https:  I, II and III, corresponding to tetra-, penta-and hexamethylated GDGTs are labelled)(For molecular structures see Figure A1)      discernable: those with ∆ 14 C values higher than the bulk OC, and thus with a more rapid turnover (in most samples, this includes shortchain (C16-C22) FA and the low density fraction (free POM) and those with lower ∆ 14 C values implying longer turnover times (including long-chain n-alkanes (in the Beatenberg soil C29-alkane, in the Lausanne soil the C27), and fatty acids (C26, C28 FA) and the GDGTs.

isoprenoid GDGTs branched GDGTs
Name m/z Name 6-methyl isomer m/z Supplementary Figure 1: Structures of GDGTs analyzed in this study. Figure A1. Molecular structures of GDGTs analyzed in this study