Evaluation of bacterial glycerol dialkyl glycerol tetraether and 2 H– 18 O biomarker proxies along a central European topsoil transect

. Molecular fossils, like bacterial branched glycerol dialkyl glycerol tetraethers (brGDGTs), and the stable isotopic composition of biomarkers, such as δ 2 H of leaf wax-derived n -alkanes ( δ 2 H n -alkane ) or δ 18 O of hemicellulose-derived sugars ( δ 18 O sugar ), are increasingly used


Introduction
Information about the variability and consequences of past climate changes is a prerequisite for precise predictions regarding the present climate change. Molecular fossils, socalled biomarkers, have great potential to enhance our understanding about variations in past climate and environmental changes. Lipid biomarkers in particular are increasingly used for paleoclimate and environmental reconstructions (e.g. Brincat et al., 2000;Eglinton and Eglinton, 2008;Rach et al., 2014;Romero-Viana et al., 2012;Schreuder et al., 2016). However strengths and limitations of respective proxies need to be known (Dang et al., 2016). For this, calibrations using modern reference samples are essential.
One famous and widely applied lipid biomarker group are terrestrial branched glycerol dialkyl glycerol tetraethers (brGDGTs). They are synthesised in the cell membranes of anaerobe heterotrophic soil bacteria (Oppermann et al., 2010;Weijers et al., 2010) and have great potential for the reconstruction of past environmental conditions (e.g. Coffinet et al., 2017;Schreuder et al., 2016;R. Zech et al., 2012), although some uncertainties exist. Calibration studies suggest that the relative abundance of the individual brGDGTs varies with mean annual air temperature (T MA ) and soil pH (Peterse et al., 2012;Weijers et al., 2007), at least across large, global climate gradients or along pronounced altitudinal gradients (Wang et al., 2017). However, in arid regions the production of brGDGT is limited, while isoprenoidal GDGTs (iGDGTs) produced by archaea provide the dominant part of the overall soil GDGT pool (Anderson et al., 2014;Dang et al., 2016;Dirghangi et al., 2013;Wang et al., 2013;Xie et al., 2012). The ratio of brGDGTs vs. isoprenoid GDGTs (BIT) can be used as an indication of whether a reconstruction of T MA and pH will be reliable. Moreover, Mueller-Niggemann et al. (2016) revealed an influence of the vegetation cover on the brGDGT producing soil microbes. From field experiments, it is known that vegetation type and mulching practice strongly effect soil temperature and moisture (Awe et al., 2015;Liu et al., 2014). Thus, multiple factors can be expected to influence soil microbial communities and GDGT production. So far, little is known about the variability of GDGT proxies on a regional scale, and a calibration study with a small climate gradient but with different vegetation types might be useful.
Concerning paleohydrology proxies, compound-specific stable hydrogen isotopes of leaf wax biomarkers, such as long-chain n-alkanes (δ 2 H n-alkanes ), record the isotopic signal of precipitation and therefore past climate and environmental conditions (Sachse et al., 2004(Sachse et al., , 2006. However, various influencing factors are known, e.g. the moisture source to leaf waxes Zhou, 2018 andSachse et al., 2012, for review). Next is the evapotranspiration of leaf water (Feakins and Sessions, 2010;Kahmen et al., 2013;Zech et al., 2015), which is strongly driven by relative air humidity (RH; e.g. Cernusak et al., 2016 for review). In addition, a strong precipitation signal is known to be incorpo-rated into long-chain leaf waxes (Hou et al., 2008;Rao et al., 2009;Sachse et al., 2004). In paleoclimate studies, it is often not feasible to disentangle between the evapotranspirative enrichment and the precipitation signal. Zech et al. (2013) proposed coupling δ 2 H n-alkane results with oxygen stable isotopes of hemicellulose-derived sugars (δ 18 O sugar ). Assuming constant biosynthetic fractionation (ε bio ) for the different compound classes (n-alkanes and hemicellulose sugars), this coupling enables the reconstruction of the isotopic composition of leaf water, RH, δ 2 H and δ 18 O of plant source water (≈ δ 2 H and δ 18 O of precipitation; Tuthorn et al., 2015). So far, a detailed evaluation of this approach on the European scale, as well as related effects concerning vegetation changes, is missing.
We analysed topsoil samples under coniferous, deciduous and grassland vegetation along a central European transect in order to estimate the variability of the biomarker proxies. More specifically, we aim to test whether i. the vegetation type has an influence on the brGDGT proxies, the δ 2 H n-alkane and the δ 18 O sugar stable isotopic composition, and the reconstructed δ 2 H source-water , δ 18 O source-water and RH; ii. the published brGDGT proxies used for reconstructing mean annual temperature and soil pH are sensitive enough to reflect the medium changes in temperature and soil pH along our transect; iii. the coupled δ 2 H n-alkane -δ 18 O sugar approach enables a δ 2 H and δ 18 O of precipitation and RH reconstruction along the transect.
2 Material and methods

Geographical setting and sampling
In November 2012, we collected 29 topsoil samples (0-5 cm depth) from 16 locations along a transect from southern Germany to southern Sweden (Fig. 1a). We distinguished between sites with coniferous forest (con, n = 9), deciduous forest (dec, n = 14) and grassland (grass, n = 6) vegetation cover (for more details see Schäfer et al., 2016, and Table S1 in the Supplement).

Database of instrumental climate variables and isotope composition of precipitation
Climate data were derived from close-by weather observation stations operated by regional institutions (Deutscher Wetterdienst (DWD) for Germany, Danmarks Meteorologiske Institut (DMI) for Denmark and the Sveriges Meteorologiska och Hydrologiska Institute (SMHI) for Sweden).
The DWD provides hourly data for each station (DWD Climate Data Center, 2018b), enabling not only the calculation of T MA , but also of the mean annual relative air humidity (RH MA ), mean temperature and relative air humidity during the vegetation period (T MV and RH MV ), and daytime temperature and relative humidity averages over the vegetation period (T MDV and RH MDV ). In addition, annual precipitation observations were used to derive the mean annual precipitation amount (P MA ; DWD Climate Data Center, 2018a). From the DMI, the respective climate variables were derived from published technical reports (Cappelen, 2002;Frich et al., 1997;Laursen et al., 1999).
The SMHI provides open data from which we derived the climate variables for the Swedish sites (Swedish Meteorological and Hydrological Institute, 2018). For more details about the climate database used for calculations and comparisons, the reader is referred to Table S2. For comprising German precipitation (δ 2 H and δ 18 O) along the transect, we realised a regionalisation (called δ 2 H GIPR and δ 18 O GIPR ) using online available data from 34 German GNIP stations, four Austrian ANIP stations and the Groningen GNIP station (van Geldern et al., 2014;IAEA/WMO, 2018;Stumpp et al., 2014;Umweltbundesamt GmbH, 2018), following the approach of Schlotter (2007). However, instead of the multivariate regression procedure applied by Schlotter (2007), we used a random forest approach (Hothorn et al., 2006;Strobl et al., 2007Strobl et al., , 2008 to describe the relationship of squared latitude, latitude, longitude and altitude vs. long-term weighted means of precipitation δ 2 H and δ 18 O and realised the prediction for each site (see the Supplement for more information). For the Danish and Swedish sites, such a procedure was not possible. Hence, the annual precipitation δ 2 H and δ 18 O values were derived from the Online Isotopes in Precipitation Calculator (OIPC, version 3.1), therefore called δ 2 H OIPC and δ 18 O OIPC (Bowen, 2018;Bowen and Revenaugh, 2003;IAEA/WMO, 2015). The final used δ 2 H GIPR,OIPC and δ 18 O GIPR,OIPC data are given in Table S1.

GDGTs and pH
A detailed description of sample preparation for lipid analysis can be found in Schäfer et al. (2016). Briefly, 1-6 g of freeze-dried and ground soil sample was microwave extracted with 15 mL dichloromethane (DCM) / methanol (MeOH) 9 : 1 (v : v) at 100 • C for 1 h. Extracts were separated over aminopropyl silica gel (Supelco, 45 µm) pipette columns. The nonpolar fraction (including n-alkanes) was eluted with hexane and further purified over AgNO 3 -coated silica pipette columns (Supelco, 60-200 mesh) and zeolite (GHGeochem Ltd.). The GDGT-containing fraction was eluted with DCM : MeOH 1 : 1 (v : v), redissolved in hexane / isopropanol 99 : 1 (v : v) and transferred over 0.45 µm PTFE filters into 300 µL inserts. For quantification, a known amount of a C46 diol standard was added after transfer. The samples were analysed at ETH Zurich using an Agilent 1260 Infinity series HPLC-atmospheric chemical pressure ionisation mass spectrometer (HPLC-APCI-MS) equipped with a Grace Prevail Cyano column (150 mm × 2.1 mm; 3 µm). The GDGTs were eluted isocratically with 90 % A and 10 % B for 5 min and then with a linear gradient to 18 % B for 34 min at 0.2 mL min −1 , where A is hexane and B is hexane / isopropanol (9 : 1, v : v). Injection volume was 10 µL and single ion monitoring of [M+H] + was used to detect GDGTs.
The pH of the samples was measured in the laboratory of the Soil Biogeochemistry group, Institute of Agronomy and Nutritional Sciences, Martin Luther University Halle-Wittenberg, using a pH meter in a 1 : 3 soil : water (w/v) mixture.

δ 2 H n-alkane
The hydrogen isotopic composition of the highest concentrated n-alkanes (n-C 25 , n-C 27 , n-C 29 , n-C 31 and n-C 33 ) was determined using a TRACE GC Ultra gas chromatography connected to a Delta V Plus isotope ratio mass spectrometer via a 2 H pyrolysis reactor kept at 1420 • C (GC-2 H-Py-IRMS; Thermo Scientific, Bremen, Germany) at ETH Zurich (Christoph et al., 2019). For more details about n-alkane quantification the reader is refereed to Schäfer et al. (2016). The compound-specific 2 H/ 1 H ratios were calibrated against an external standard with C 15 -C 35 homologues. External standard mixtures (A4 mix from Arndt Schimmelmann, University of Indiana) were run between the samples for multipoint linear normalisation. The H + 3 factor was determined on each measurement day and was constant throughout the periods of the sample batches. Samples were analysed in duplicates, and results typically agreed within 4 % (average difference = 1.4 %). All δ 2 H values are expressed relative to the Vienna Standard Mean Ocean Water (V-SMOW).

δ 18 O sugar
Hemicellulose sugars were extracted and purified using a slightly modified standard procedure (Amelung et al., 1996;Guggenberger et al., 1994;Zech and Glaser, 2009). Briefly, myo-inositol was added to the samples prior to extraction as the first internal standard. The sugars were released hydrolytically using 4 M trifluoroacetic acid for 4 h at 105 • C, cleaned over glass fibre filters and further purified using XAD and Dowex columns. Before derivatisation with methylboronic acid (Knapp, 1979), the samples were frozen and freezedried, and 3-O-methylglucose in dry pyridine was added as the second internal standard. Compound-specific hemicellulose sugar 18 O measurements were performed in the laboratory of the Soil Biogeochemistry group, Institute of Agronomy and Nutritional Sciences, Martin Luther University Halle-Wittenberg, using GC-18 O-Py-IRMS (all devices from Thermo Fisher Scientific, Bremen, Germany). Standard deviations of the triplicate measurements were 1.4 ‰ (over 29 investigated samples) for arabinose and xylose. We focus on these two hemicellulose-derived neutral sugars arabinose and xylose as they strongly predominate over fucose in terrestrial plants, soils and sediments (Hepp et al., 2016, and references therein). Rhamnose concentrations were too low to obtain reliable δ 18 O results. All δ 18 O values are expressed relative to the Vienna Standard Mean Ocean Water (V-SMOW).

Calculations used for the GDGT-based reconstructions
The branched and isoprenoid tetraether (BIT) index is calculated according to Hopmans et al. (2004); for structures see Fig. S5: The cyclopentane moiety number of brGDGTs correlates negatively with soil pH (Weijers et al., 2007), which led to the development of the ratio of cyclisation of branched tetraethers (CBT). CBT and the CBT-based pH (pH CBT ) were calculated according to Peterse et al. (2012): The number of methyl groups in brGDGTs correlates negatively with T MA and soil pH (Peterse et al., 2012;Weijers et al., 2007). Thus, the ratio of the methylation of branched tetraethers (MBT) and the CBT ratio can be used to reconstruct T MA . We use the equation given by Peterse et al. (2012):

Calculations and concepts used for the coupled δ 2 H-δ 18 O approach
The apparent fractionation is calculated according to Cernusak et al. (2016): The isotopic composition of leaf water (δ 2 H leaf-water and δ 18 O leaf-water ) can be calculated using ε bio for δ 2 H n-alkane (−160 ‰; Sachse et al., 2012;Sessions et al., 1999) and δ 18 O sugar (+27 ‰; Cernusak et al., 2003;Schmidt et al., 2001): Zech et al. (2013) introduced the conceptual model for the coupled δ 2 H n-alkane -δ 18 O sugar approach in detail. Briefly, the coupled approach is based on the following assumptions (illustrated in Fig. 8). (i) The isotopic composition of precipitation, which is set to be equal to the plant source water, typically plots along the global meteoric water line (GMWL; δ 2 H = 8 × δ 18 O + 10) in a δ 18 O vs. δ 2 H space (Craig, 1961).
(ii) Source water uptake by plants does not lead to any fractionation (e.g. Dawson et al., 2002), and significant evaporation of soil water can be excluded. (iii) Evapotranspiration leads to enrichment of the remaining leaf water along the local evaporation line (LEL; Allison et al., 1985;Bariac et al., 1994;Walker and Brunel, 1990), compared to the source water taken up by the plant. (iv) The biosynthetic fractionation is assumed to be constant. In addition, isotopic equilibrium between plant source water (∼ weighted mean annual precipitation) and the local atmospheric water vapour is assumed.
Further assumption concerns the isotope steady state in the evaporating leaf water reservoir. The coupled approach allows for reconstruction of the isotopic composition of plant source water (δ 2 H source-water and δ 18 O source-water ) from the reconstructed leaf water, by calculating the intercepts of the LELs with the GMWL (Zech et al., 2013). The slope of the LEL (S LEL ) can be assessed by the following equation (Gat, 1971): where ε * represents equilibrium isotope fractionation factors and C k represents kinetic fractionation factors. The latter equals 25.1 ‰ and 28.5 ‰ for C 2 k and C 18 k , respectively (Merlivat, 1978). The equilibrium fractionation factors can be derived from empirical equations (Horita and Wesolowski, 1994) Table S2).
In a δ 18 O-δ 2 H diagram, the distance of the leaf water from the GMWL defines the deuterium excess of leaf water (d leaf-water = δ 2 H leaf-water −8×δ 18 O leaf-water , according Dansgaard (1964) ; Fig. 8). To convert d leaf-water into mean RH during daytime and the vegetation period (RH MDV ), a simplified Craig-Gordon model can be applied (Zech et al., 2013): where d is the difference in d leaf-water and the deuterium excess of source water (d source-water ).

Statistics
In the statistical analysis we checked sample distributions for normality (Shapiro and Wilk, 1965) and for equal variance (Levene, 1960). If normality and equal variances are given, we perform an analysis of variance (ANOVA). If that is not the case, we conduct the non-parametric Kruskal-Wallis test. ANOVA or Kruskal-Wallis are used to find significant differences (a = 0.05) between the vegetation types (deciduous, conifer and grass). In order to describe the relation along a 1 : 1 line, the coefficient of correlation (R 2 ) was calculated as R 2 = 1 − (modeled − measured) 2 / (measured − measured mean) 2 . The small r 2 is taken as coefficient of correlation of a linear regression between a dependent (y) and explanatory variable(s). The root-mean-square error (RMSE) of the relationships was calculated as All data plotting and statistical analysis was realised in R (version 3.2.2; R Core Team, 2015).
3 Results and discussion

GDGT concentrations
GDGT Ia has the highest concentration under all vegetation types, followed by GDGT IIa and GDGT IIIa (Fig. 2). GDGT Ib, IIb and Ic occur in minor amounts and GDGT IIc and IIIb only in trace amounts. GDGT IIIc was below the detection limit in most of the samples (Table S3). Although other studies document an influence of the vegetation cover on soil temperature and soil water content, which control the microbial community composition in soils (Awe et al., 2015;Liu et al., 2014;Mueller-Niggemann et al., 2016), we find no statistically different pattern of the individual brGDGTs. Total concentrations of brGDGTs range from 0.32 to 9.17 µg g −1 dry weight and tend to be highest for the coniferous samples and lowest for the grasses (Fig. 3a, Table S3). Mean concentrations of individual brGDGTs as percentage of all brGDGTs for the three investigated types. Abbreviations: con: coniferous forest sites (n = 9); dec: deciduous forest sites (n = 14); grass: grassland sites (n = 6).
Bulk brGDGT concentrations lie within the range of other studies examining soils of mid-latitude regions (Huguet et al., 2010a, b;Weijers et al., 2011). Similar concentrations in coniferous and deciduous samples imply that brGDGT production does not strongly vary in soils below different forest types. The grass samples show lower brGDGT concentrations compared to the forest samples, but this is probably mainly due to ploughing of the grass sites in former times and hence admixing of mineral subsoil material. The differences in brGDGT concentrations are not significant (p value = 0.06).

BIT index
Most of the samples have a BIT index higher than 0.9 ( Fig. 3b and Table S3). The BIT values are typical for soils in humid and temperate climate regions (Weijers et al., 2006). However, outliers exist. The most likely source of iGDGTs in soils are Thaumarchaeota, i.e. aerobe ammonia oxidising archaea producing Crenarchaeol and its regioisomer (Schouten et al., 2013 and references therein), when the precipitation amount drops below 700-800 mm (Dang et al., 2016;Dirghangi et al., 2013). The P MA data of our sampling sites mostly show precipitation > 550 mm (Fig. 1b), but one has to be aware that these data are based on the climate station closest to the respective sampling locations and that microclimate effects, such as sunlight exposure, canopy cover or exposition, might have a pronounced influence on the brGDGT vs. iGDGT distribution. Mueller-Niggemann et al. (2016) found higher BIT indices in upland soils compared to paddy soils and stated that the management type also influences BIT values in soils. Along our transect, grass sites tend to have slightly lower BIT values than forest sites, probably due to the absence of a litter layer and hence no isolation mechanism preventing evaporation of soil water. Differences between vegetation types are not significant (p value = 0.32).

CBT-derived pH
The CBT ratio shows a pronounced variation independent of vegetation type with values between 0.03 and 2.16 (Fig. 3c). The coniferous samples tend to be highest, but the differences between vegetation types are not significant (p value = 0.48). The CBT index can be related to pH in acidic and/or humid soils (e.g. Dirghangi et al., 2013;Mueller-Niggemann et al., 2016;Peterse et al., 2012;Weijers et al., 2007) but might be an indicator of soil water content and hence precipitation in more arid and alkaline soils (e.g. Dang et al., 2016). There is a pronounced correlation between CBT and soil pH (Fig. 4), which is in good agreement with other studies from mid-latitude regions where precipitation is relatively high (Anderson et al., 2014, and references therein). Moreover, the CBT-to-pH relationship in terms of slope and intersect in our dataset (CBT = −0.47 × pH + 3.5, r 2 = 0.7, p value < 0.0001, n = 29) is comparable to the correlation described for the global calibration dataset of Peterse et al. (2012) However, there are some outliers in the CBT-pH correlation, which need further examination (see locations grass L04, dec L10 and dec L12 as marked in Figs. 4 and 5). The outliers show lower BIT indices (< 0.85, Table S3). Even though the data from the nearest climate station suggest no abnormal P MA . Local effects such as differences in the amount of sunlight exposure, nutrient availability for brGDGT-producing organisms or most likely soil water content might influence the brGDGT production at these locations (Anderson et al., 2014;Dang et al., 2016). A lower BIT index as well as a lower CBT occur when soil water content decreases (Dang et al., 2016;Sun et al., 2016) or when aeration is high and less anoxic microhabitats for GDGTproducing microbes exist (e.g. Dirghangi et al., 2013).
As the CBT and pH are similarly correlated in our dataset and the global dataset of Peterse et al. (2012), the CBTderived pH correlated well with the actual pH ( Fig. 5a; R 2 = 0.3). Expressed as pH (CBT-derived pH -measured pH), there is a tendency that the GDGTs result in an overestimation of the real pH for the forest sites (Fig. 5b). However, a Kruskal-Wallis test shows no statistically significant difference between the vegetation types, with a p value of 0.13. The overall pH of 0.6 ± 0.6 shows that the reconstruction of soil pH using brGDGTs works well along this transect.

MBT -CBT-derived T MA reconstructions
The MBT shows high variability with values ranging from 0.17 to 0.67 and no statistical differences between vegetation types (p value = 0.54; Fig. 3d, Table S3). When comparing reconstructed (MBT -CBT-derived) T MA with climate station T MA , the data plot close to the 1 : 1 line and fit well into the global dataset of Peterse et al. (2012) (Fig. 6a). The T MA values reveal an overall offset of 0.5 • C ± 2.4, and  there is no statistical difference between vegetation types (Fig. 6b). The standard deviation in T MA of ±2.4 is well in line with the RMSE of 5.0 for the global calibration dataset (Peterse et al., 2012).

Potential impact of the used liquid chromatography method on pH and T MA reconstructions
The GDGT data presented in this study are not acquired with the up-to-date method ( (Peterse et al., 2012) and their newly developed T MA calibration and state that they plot around a 1 : 1 line. They furthermore state that, especially for arid areas, larger deviations can be expected. Finally, they conclude that the use of the newly developed calibrations will improve the T MA and pH reconstructions for areas with arid climate conditions. Because our study transect spans from southern Germany to southern Sweden, representing temperate and humid climate conditions, we argue that the usage of the older liquid chromatography method does not introduce a systematic error in our T MA and pH reconstructions. Still, a higher variability/scatter could be associated with the calibration of Peterse et al. (2012) and therefore also present in our T MA and pH reconstructions.

Apparent fractionation of δ 2 H and δ 18 O in the different vegetation types
δ 2 H values could be obtained for the n-alkanes C 27 , C 29 and C 31 in all samples and additionally at two locations for n-C 25 and at six other locations for n-C 33 . The δ 2 H n-alkane values, calculated as the mean of n-C 25 to n-C 31 δ 2 H, range from −156 ‰ to −216 ‰. Pooled standard deviations show an overall average of 3.6 ‰. The δ 18 O sugar values, calculated as the area-weighted means for arabinose and xylose, range from +27.7 ‰ to +39.4 ‰. The average weighted mean standard deviation is 1.4 ‰. The compound-specific isotope data are summarised along with the calculations in Table S4. Apparent fractionation (ε n-alkane/precipitation ) is on the order of −120 ‰ to −150 ‰, i.e. a bit less than the biosynthetic fraction of −160 ‰. This implies that evapotranspirative enrichment is ∼ 10 ‰ to 40 ‰ (Fig. 7a). ε n-alkane/precipitation is lower for grass sites compared to the forest sites. Differences are significant between deciduous and grass sites (p value = 0.005). This finding supports the results of other studies (Kahmen et al., 2013;Liu and Yang, 2008;McInerney et al., 2011) and can be named signal damping. Grasses  and lower (25 %) quartiles, lowest whisker still within 1.5 IQR of the lower quartile, and highest whisker still within 1.5 IQR of the upper quartile; dots mark outliers. Abbreviations: con: coniferous forest sites (n = 9); dec: deciduous forest sites (n = 14); grass: grassland sites (n = 6).
do not only incorporate the evaporatively enriched leaf water but also unenriched xylem water in the growth and differentiation zone of grasses (Gamarra et al., 2016;Liu et al., 2017).
The grass-derived hemicellulose sugar biomarkers do not fully record the evapotranspirative enrichment of the leaf water, either, as indicated by lower apparent fractionation (ε sugar/precipitation ) in Fig. 7b. The differences are significant between forest and grass sites (p value < 0.005). This is in agreement with a study on cellulose extracted from grass blades (Helliker and Ehleringer, 2002), and again the signal damping can be explained with incorporation of enriched leaf water and non-enriched stem water.
Based on the comparison of evapotranspirative enrichment between forest and grass sites, the signal damping can be quantified to be ∼ 31 % for the hemicellulose sugars and ∼ 49 % for the n-alkanes. This is in agreement with other studies that reported a loss of 22 % of the leaf water enrichment for hemicellulose sugars (Helliker and Ehleringer, 2002) and 39 % to 62 % loss of the leaf water enrichment for n-alkanes (Gamarra et al., 2016).

δ 2 H source-water and δ 18 O source-water reconstructions
The δ 2 H versus δ 18 O diagram shown in Fig. 8 graphically illustrates the reconstruction of δ 2 H leaf-water and δ 18 O leaf-water (coloured dots) from δ 2 H n-alkane and δ 18 O sugar (crosses), as well as the reconstruction of δ 2 H source-water and δ 18 O source-water (black dots). For reconstructing δ 2 H source-water and δ 18 O source-water , LELs with an average slope of 2.8 ± 0.1 (Eq. 10) can be generated through every leaf water point and the intercepts of these LELs with the GMWL. and lower (25 %) quartiles, lowest whisker still within 1.5 IQR of the lower quartile, and highest whisker still within 1.5 IQR of the upper quartile; dots mark outliers. Abbreviations: con: coniferous forest sites (n = 9); dec: deciduous forest sites (n = 11 and 14 for n-alkanes and sugars, respectively); grass: grassland sites (n = 4 and 6 for n-alkanes and sugars, respectively). The figure conceptually illustrates the effect of biosynthetic fractionation and evapotranspirative enrichment as well as signal damping. Figure 8. δ 2 H vs. δ 18 O diagram illustrating the coupled δ 2 H n-alkane -δ 18 O sugar approach: measured δ 2 H n-alkane and δ 18 O sugar values, reconstructed δ 2 H leaf-water and δ 18 O leaf-water (according Eqs. 8 and 9), and reconstructed δ 2 H source-water and δ 18 O source-water in comparison to GIPR and OIPC-based δ 2 H precipitation and δ 18 O precipitation . Abbreviations: con: coniferous forest sites (n = 9); dec: deciduous forest sites (n = 11); grass: grassland sites (n = 4).
The reconstructed δ 2 H source-water and δ 18 O source-water results can be compared with the δ 2 H GIPR,OIPC and δ 18 O GIPR,OIPC data (Fig. 9). This comparison reveals that the coupled δ 2 H n-alkane -δ 18 O sugar approach yields more accurate δ 2 H source-water and δ 18 O source-water compared to single δ 2 H n-alkane approaches. However, the range of the reconstructed δ 2 H source-water and δ 18 O source-water values is clearly larger than in δ 2 H GIPR,OIPC and δ 18 O GIPR,OIPC values. δ 2 H is systematically underestimated by ∼ 21 ‰ ± 22 ‰ (Fig. 9b) and δ 18 O by ∼ 2.9 ‰ ± 2.8 ‰ (Fig. 9d). The type of vegetation seems to be not particularly relevant (p value = 0.18 for δ 2 H and p value = 0.34 for δ 18 O). Nevertheless, the systematic offsets tend to be lowest for the deciduous sites ( δ 2 H and δ 18 O are closer to zero with ∼ −5 ‰ ± 15 ‰ and ∼ −1.1 ‰ ± 2.1 ‰), followed by grass sites (∼ −14 ‰ ± 20 ‰ and ∼ −2.1 ‰ ± 2.6 ‰). In comparison, the coniferous sites show the largest offsets (∼ −23 ‰ ± 26 ‰ for δ 2 H and ∼ −3.0 ‰ ± 3.3 ‰ for δ 18 O). Differences are, however, not statistically significant. The systematic offset and the large variability might have more specific reasons, and we suggest that this is related to the type of vegetation. Deciduous trees produce lots of leaf waxes and sugars (e.g. Prietzel et al., 2013;M. Zech et al., 2012a), and all biomarkers reflect and record the evapotranspirative enrichment of the leaf water (e.g. Kahmen et al., 2013;Tuthorn et al., 2014). By contrast, coniferous trees produce quite low amounts of n-alkanes (Diefendorf and Freimuth, 2016;M. Zech et al., 2012a), while sugar concentrations are as high as in other vascular plants (e.g. Hepp et al., 2016;Prietzel et al., 2013). For the coniferous soil samples this means that the n-alkanes stem most likely from the understorey whereas the sugars originate from grasses and coniferous needles. When the understorey is dominated by grass species, the n-alkane biomarkers do not record the full leaf water enrichment signal, whereas the sugars from the needles do. The reconstructed leaf water for the coniferous sites is therefore too negative concerning δ 2 H, and reconstructed δ 2 H source-water and δ 18 O source-water values thus also become too negative (Fig. 8). Concerning the grass sites, the following explanation can be found. Correcting for signal damping makes the reconstructed leaf water points more positive and shifts them in Fig. 8 up and to the right. As the Figure 9. Correlation of reconstructed δ 2 H source-water and δ 18 O source-water vs. precipitation δ 2 H GIPR,OIPC and δ 18 O GIPR,OIPC (a, c). Black lines indicate the 1 : 1 relationship. Differences between reconstructed source water and precipitation ( δ 2 H = δ 2 H source-water − δ 2 H GIPR,OIPC , δ 18 O = δ 18 O source-water −δ 18 O GIPR,OIPC ) for the three different vegetation types (b, d). Box plots show the median (red line), interquartile range (IQR) with upper (75 %) and lower (25 %) quartiles, lowest whisker still within 1.5 IQR of the lower quartile, and highest whisker still within 1.5 IQR of the upper quartile. Abbreviations: con: coniferous forest sites (n = 9); dec: deciduous forest sites (n = 11); grass: grassland sites (n = 4). signal damping is stronger for δ 2 H than for δ 18 O, the corrected leaf water points would now plot above the uncorrected ones. The corrected leaf water points lead to more positive reconstructed δ 2 H source-water and δ 18 O source-water values for the grass sites. However, Gao et al. (2014) and Liu et al. (2016) showed that the ε bio (regarding 2 H between nalkanes and leaf water) of monocotyledon plants could be larger than those of dicotyledonous ones. This would therefore also cause a more negative ε n-alkane/precipitation for grasses compared to trees. We observe that the ε n-alkane/precipitation is indeed more negative for the grass sites compared to the forest sites ( Fig. 7 and Sect. 3.6). Therefore, effects of signal damping vs. variable ε bio along with vegetation types are indistinguishable here. As an outlook for a future study, we therefore strongly recommend a comparison between the here measured δ 2 H n-alkane values with modelled ones using, for example, the new available model approach from Konecky et al. (2019), which could provide insights if such vegetation effects on ε bio of 2 H in n-alkanes are describable.
Rooting depths specific to vegetation type could partly cause the overall high variability in reconstructed δ 2 H source-water and δ 18 O source-water . Deep-rooting species most likely use the water from deeper soil horizons and/or shallow groundwater, which is equal to the (weighted) mean annual precipitation (e.g. Herrmann et al., 1987). Shallowrooting plants take up water from upper soil horizons, which is influenced by seasonal variations in δ 2 H precipiation and δ 18 O precipiation and by soil water enrichment (Dubbert et al., 2013). Thus, the overall assumption that the source water of the plants reflects the local (weighted) mean precipitation might be not fully valid for all sites. Moreover, a partial contribution of root-derived rather than leaf-derived sugar biomarkers in our topsoil samples is very likely. This does, by contrast, not apply for n-alkanes, which are hardly produced in roots (M. Zech et al., 2012b, and the discussion therein).
Moreover, the high variability within the vegetation types could be caused by variability in ε bio of 2 H in n-alkanes, as well as 18 O in sugars. There is an ongoing discussion about the correct ε bio for 18 O in hemicellulose sugars (Sternberg, 2014vs. Zech et al., 2014, and ε bio is probably not constant over all vegetation types. This translates into errors concerning leaf water reconstruction and thus reconstructing δ 2 H source-water and δ 18 O source-water values (Eq. 9 and Fig. 8). Likewise, the ε bio values reported in the literature for 2 H of nalkanes can be off from −160 ‰ by tens of per mille (Feakins and Sessions, 2010;Tipple et al., 2015;Feakins et al., 2016;Freimuth et al., 2017). The degree to which hydrogen originates from NADPH rather than leaf water is important, because NADPH is more negative (Schmidt et al., 2003). The wide range in biosynthetic 2 H fractionation factors, which can be even larger, is therefore also related to the carbon and energy metabolism state of plants (Cormier et al., 2018).
Too low reconstructed RH MDV values for the coniferous sites make sense in view of the previously discussed option that soils contain n-alkanes from the understorey (which is dominated by grass species), while sugars stem from needles and grasses. As explained earlier already, the signal damping leads to too negative reconstructed δ 2 H leaf-water (whereas δ 18 O is affected less by the signal damping), and too negative δ 2 H leaf-water translates into overestimated d-excess and underestimated RH values. In Fig. 8, a correction for this requires moving the coniferous leaf water data points upwards towards more positive δ 2 H values; thus the distance between the leaf water and the source water is reduced. It should be noted that here variable ε bio (regarding 2 H between n-alkanes and leaf water) along with vegetation types could also not be distinguished from signal damping effects.
The underestimation of RH for the deciduous and grass sites could be partly associated with the use of the GMWL as a baseline for the coupled δ 2 H n-alkane -δ 18 O sugar approach. The deuterium excess of local meteoric water lines (LMWLs) is generally lower than the +10 ‰ of the GMWL, while the slopes of the LMWLs are comparable to the GMWL (Stumpp et al., 2014). In addition, if soil water evaporation occurred before water uptake by the plants, this would lead to an underestimation of biomarker-based RH MDV values, too. It can be furthermore assumed that plant metabolism is highest during times with direct sunshine and high irradiation, i.e. during noon on sunny days. The relevant RH could therefore be lower than the climate-station-derived RH MDV . Indeed, already climate station RH MDV is considerably lower than RH MA and RH MV (Table S1 in the Supplement).
The uncertainty of reconstructed RH MDV values is large for all three investigated vegetation types, and again these uncertainties are probably also related to ε bio , which is most likely not constant as assumed for our calculations. Moreover, microclimate variability is underestimated in our approach. As mentioned in Sects. 2.4.2 and 3.7, in the coupled approach not only the source water of the plants is equated with (weighted) mean annual precipitation, but also an isotopic equilibrium between the source water and the (local) atmospheric water vapour is assumed. However, in areas with distinct seasonality this might be not fully valid. To account for this lack of equilibrium between precipitation and local atmospheric water vapour, apparent ε values can be calculated with data from Jacob and Sonntag (1991). As shown by Hepp et al. (2019) those values can be used to achieve alternative RH reconstructions based on the coupled δ 2 H n-alkaneδ 18 O sugar approach. Such calculated RH MDV values are on average 1.5 % more negative than the original values. How-ever, this difference in RH is far below the analytical uncertainties of the compound-specific biomarker isotope analysis.
Finally, the integration time of the investigated topsoils has to be discussed. Unfortunately, no 14 C dates are available for the soil samples. However, most likely the organic matter has been built up over a longer timescale than the available climate data, which are used for comparison. In combination with vegetation changes/management changes throughout that period, this could surely lead to a less tight relationship of the reconstructions compared to the climate station data. Root input of arabinose and xylose seems to be of minor relevance in our topsoil samples. Otherwise, the reconstructed δ 18 O sugar values would be too negative, resulting in RH MDV overestimations, which are not observed.

Conclusions
We were able to show the following.
i. The vegetation type does not significantly influence the brGDGT concentrations and proxies, yet the coniferous sites tend to have higher brGDGT concentrations, BIT indices and CBT-MBT ratios, while grass sites tend to be lowest.
ii. CBT faithfully records soil pH with a median pH of 0.6 ± 0.6. The CBT overestimates the real pH in particular at the forest sites.
iii. CBT-MBT -derived T MA values reflect the climatestation-derived T MA values with a median T MA of 0.5 • C ± 2.4, but again slightly too high reconstructions for the forest sites were observed. iv. Differences in the apparent fractionation between the investigated vegetation types can be explained with signal damping.
v. The reconstructed δ 2 H source-water and δ 18 O source-water reflect the δ 2 H GIPR,OIPC and δ 18 O GIPR,OIPC with a systematic offset for δ 2 H of ∼ −21 ‰ ± 22 ‰ and for δ 18 O of ∼ −2.9 ‰ ± 2.8 ‰ (based on overall medians of δ 2 H, δ 18 O). This is caused by too negative reconstructions for coniferous and grass sites. For coniferous sites, this can be explained with n-alkanes originating from understorey grasses. As for the grass sites, the signal damping or variable ε bio along with vegetation types affect δ 2 H more than δ 18 O. This leads to too negative reconstructed δ 2 H leaf-water values and thus to too negative δ 2 H source-water and δ 18 O source-water reconstructions.
vi. Reconstructed (biomarker-based) RH MDV values tend to underestimate climate-station-derived RH MDV values ( RH MDV =∼ −17 % ± 12 %). For coniferous sites the underestimations are strongest, which can be explained with understorey grasses being the main source of n-alkanes for the investigated soils under coniferous forests.
Overall, our study highlights the great potential of brGDGTs and the coupled δ 2 H n-alkane -δ 18 O sugar approach for more quantitative paleoclimate reconstructions. Taking into account effects of different vegetation types improves correlations and reconstructions. This holds particularly true for the coupled δ 2 H n-alkane -δ 18 O sugar approach, which is affected by signal damping of the grass vegetation or variable ε bio (regarding 2 H between n-alkanes and leaf water) along with vegetation types. By contrast, vegetation-related effects do not strongly influence the brGDGT-derived reconstructions. Assuming constant ε bio is likely a considerable source of uncertainty and should be further addressed in future field and/or modelling studies. Climate chamber experiments are most promising to further evaluate and refine the coupled δ 2 H n-alkane -δ 18 O sugar approach, because uncertainties related to microclimate variability can be reduced. Field experiments like ours suffer from the fact that biomarker pools in the sampled topsoils may have been affected by past vegetation and climate changes and by the rather small range covered by the sampled transect. Both make the comparison between reconstructions and observations more difficult compared to large datasets and well-defined conditions. Data availability. All the data used in this work are available via the Supplement.
Author contributions. JH and IKS wrote the paper. MZ, RZ and JH acquired financial support. VL was responsible for field sampling and lipid analysis, and JH was responsible for sugar and pH analy-sis. IKS and JH acquired metadata and did the data evaluation. JF carried out the statistical analysis. VL, MB, KR, BG, MZ, TIE and RZ contributed to the discussion of the data and commented on the paper.