Soil water samples were collected in March, May, and November of 2005 at three different long term monitoring sites in Germany. The first site is a deciduous old-growth forest in the Hainich National Park [51°04′46″N, 10°27′08 ″E, 440 m above sea level (a.s.l.)], Thuringia (Knohl et al., 2003; Gleixner et al., 2009; Tefs and Gleixner, 2012). The forest is unmanaged and dominated by European beech (Fagus sylvatica, 65%) and ash (Fraxinus excelsior, 25%). Understory vegetation includes Alium ursinium, Mercurialis perennis, and Anemone nemorosa. The soil column consists of a 5 to 15 cm deep A horizon which is followed by a clayish T horizon on 50 to 60 cm deep cambisol. The second site is a maple (Acer pseudoplatanus), pine and spruce (Pinus sylvestris and Picea abies) mixed forest stand near the village of Thann (49°19′24″N, 12°27′37″E), Bavaria. A 5 to 7 cm deep organic layer overlays a sandy cambisol developed on granite parent material (Sachse et al., 2009). The third site is the Wetzstein forest (50°27′13″N, 11°27′27″E, 785 m a.s.l.) in Thuringia. This managed forest consists mostly of ca. 50 year old Norway spruce (Picea abies) trees on an acidic and carbonate-free podzol (Schulze et al., 2005; Moyano et al., 2008; Kindler et al., 2011). Setup and sampling was conducted with permission of the national park administration, the land owner and the forest management administration, respectively, for the sites. The diversity in vegetation and soil parameters within the sample set results in a relatively wide coverage of top soil DOM conditions. Together with the replicated sampling throughout the growing season we aim at providing overarching trends in DOM composition not biased through site-specific effects.

The water samples were taken during a regular biweekly sampling scheme using glass ceramic suction plates with a pore size of 1–1.6 μm which were permanently installed at 5 cm soil depth. If samples were unavailable at 5 cm on a specific sampling location, we used samples from 10 cm depth. Bottles connected to the suction plates were evacuated to 20 kPa such that free flowing soil water was sucked through the ceramic plate into the bottle until pressure equilibrium was reached. As a result of the biweekly sampling scheme, each sampling time point integrates the water flow of 2 weeks. Water samples were immediately freeze-dried in the lab and stored at room temperature in the dark. After re-dissolution, DOM was subjected to solid phase extraction using PPL (styrene-divinylbenzene polymer) cartridges according to previously established protocols (Dittmar et al., 2008), with an average extraction efficiency of 63%. A procedural blank of ultrapure water (pH 2) was stored and extracted together with samples.

FT-ICR-MS analysis was performed at the University of Oldenburg, Germany (Roth et al., 2015). The FT-ICR-MS (Bruker Solarix, 15 T) measurement was conducted in negative electrospray ionization mode at a continuous injection flow of 120 μL/h and ESI needle voltage of -4 KV. 500 scans of m/z 150–2000 and 0.25 s ion accumulation time were combined into one spectrum. Peaks with signal/noise (S/N)> 3 and within 150–800 m/z were used for further analysis. Ions with m/z> 800 were not detected. Peaks found in the procedural blank were removed from the peak list of the samples. For molecular formula assignment the elements C, H, O, N, and S (N ≤ 2, S ≤ 1) were utilized in accordance with previous studies (Herzsprung et al., 2012; Riedel et al., 2012). Compounds with successful molecular formula assignment were considered for further data interpretation (Supplementary Table S4).

PPL extracts were dried into tin capsules for combustion to CO₂ in an elemental analyzer, graphitization and subsequent radiocarbon analysis on a 3 MV Tandetron ¹⁴C-AMS at the MPI-BGC in Jena, Germany together with size-matched modern (Oxalic Acid II) and ¹⁴C-depleted standard materials (Steinhof et al., 2004, 2010). The δ¹³C values of the samples were determined online in the AMS. Age calibration was performed on the background corrected fraction modern [F¹⁴C; (Reimer et al., 2004; Steinhof, 2013)] values using OxCal 4.3 (Bronk Ramsey, 2009) and the NH1 bomb 2013 curve (Hua et al., 2013). Resulting median calAD dates were subtracted from the year of sampling (2005) to give an age estimate.

A system blank test was conducted using DOM from the Wetzstein site that was (i) non-extracted, (ii) PPL extracted, (iii) PPL extracted, dried and re-dissolved in ultrapure water once, and (iv) re-dissolved three times. For (i) DOM material from the Wetzstein site was de-carbonified by acidifying with HCl and bubbling with N₂ in order to remove inorganic carbon from affecting the ¹⁴C/¹²C DOM signal. For (ii) the same original material as in (i) was extracted onto PPL cartridges and eluted with methanol (Dittmar et al., 2008). For (iii) and (iv) aliquots of the eluate from (ii) were dried and re-dissolved in ultrapure water in order to break any methyl ester bonds that might have formed to the methanol solvent and would skew the ¹⁴C/¹²C isotopic ratio.

Spearman rank correlation was employed on relative intensities of mass peaks with respect to the ¹⁴C calibrated ages of the samples. Rank correlation was used to account for known competitive ionization efficiencies that affect the ion intensities of the individual masses (Hertkorn et al., 2008; Herzsprung et al., 2012). Only mass peaks present in at least 8 out of 10 samples were considered for correlation in order to identify bulk trends. Statistical data analysis and correlation was done in R 3.4.1 using the R package “vegan” 2.4–5. Principal component analysis (PCA) was performed on the matrix of relative intensities of m/z values per sample, which was centered and scaled by the standard deviations. The environmental parameters pH, cal ¹⁴C age, DOC concentration and length of the growing season were fitted to the PC values as vectors. For characterization of molecular entities matching of sum formulae with the ChEBI database was performed using the web interface at http://www.ebi.ac.uk/chebi/ (Hastings et al., 2016). We chose the ChEBI database for its focus on “small” chemical compounds that have biological relevance, its links and integration with similar chemical databases [(for example PubChem (Kim et al., 2016) and KEGG (Kanehisa et al., 2017)], free availability and general ease of use for the extraction of annotated information and chemical structures.

Solvent residues in carbon extracts resulting from elution steps can influence the ¹⁴C/¹²C ratio in organic matter samples. As for PPL extraction, the methanol solvent could esterify free carboxyl groups in the organic matter (McIntyre and McRae, 2005), thereby introducing ¹⁴C-depleted carbon to the sample, which persists even after evaporating to dryness. Depending on the abundance of carboxyl groups in the sample, the percentage of modern carbon would be expected to be significantly reduced especially if samples are comprised entirely of modern carbon whereas commercially bought methanol is usually produced from fossil carbon sources. Since methyl ester bonds are highly susceptible to hydrolysis, re-dissolution in water could help to remove potential solvent residues. Testing our extraction method for changes of ¹⁴C/¹²C before and after extraction as well as after hydrolysis of residual ester bound methanol yielded highly similar percentages of modern carbon (pMC; Figure 1). A one-way analysis of variance (ANOVA) revealed that there was no significant difference (α = 0.05) between the groups. We conclude that the ¹⁴C/¹²C ratios of SPE-DOM in our study are not influenced by residual ester bound MeOH and likely represent the ¹⁴C/¹²C ratios of bulk DOM.

Principal component analysis (PCA) using the exact masses of molecules with annotated sum formulae summarized 68.2% of the variability in the molecular data by the first two principal components (PC): PC1 49.2%, PC2 19.0%. The PCA produced a well-separated clustering reflecting the 3 sampling sites (Figure 2). The fitted environmental parameters DOC concentration and pH were correlated with PC1. Cal ¹⁴C age was correlated to PC2. In a previous assessment none of the considered environmental parameters were correlated to PC2 (Roth et al., 2015). Our results show that cal ¹⁴C age can help to interpret additional variability in DOM composition.

The radiocarbon activity of the samples ranged from 102 to 117 percent Modern Carbon (pMC). All samples were “modern” and the calibrated ages ranged between 0 and 49 years (Table 1). The within-site variability was higher than between sampling sites, suggesting that radiocarbon ages in this study are not biased through site-effects.

From here on we refer to negative/positive correlation with age of a molecular formula in cases when the rank relative intensity distribution of the respective sum formula was significantly descending/ascending between samples from low cal ¹⁴C age to samples with higher cal ¹⁴C age, respectively. We observed distinct differences in the m/z ranges as well as H/C and O/C elemental ratios between the group of negatively and positively correlated compounds (Figure 3). The masses of negatively correlated formulae ranged from 250 to 450 m/z while the positively correlated formulae ranged from 450 to 650 m/z. This indicates that inputs into soil water DOM containing recently assimilated carbon are likely of lower molecular weight compared to molecular entities comprised of ¹⁴C-older carbon. The full distribution of masses and respective rank correlation strength values further suggests a continuous mass-age relation rather than fractionated sub pools of young and old DOM (Supplementary Figure S1). The finding of lower molecular weight compounds being associated with more recent SPE-DOM is consistent with results from an assessment of marine SPE-DOM aging trends (Flerus et al., 2012). Previous studies of marine DOM size fractions have suggested decreasing molecular weight with increasing age (Kaiser and Benner, 2009; Benner and Amon, 2015). Yet, direct comparisons between SPE-DOM and studies using size-fractionation cannot be drawn due to the differences in sample preparation and resulting observed mass range.

In the forest environment one of the most prevalent sources of young carbon is plant litter. Consequently, lignin-derived molecules associated with the young carbon inputs are expected. In the van Krevelen diagram indeed many of the negatively correlated sum formulae can be estimated to represent lignin-type structures from their elemental composition [H/C: 0.6 – 1.7, O/C: 0.1 – 0.6; (Antony et al., 2014)]. A search within the ChEBI library revealed that the group of negatively ¹⁴C-age correlated formulae could possibly be mostly comprised of methoxylated phenylpropanoid-derivatives (Table 2 and Supplementary Tables S1, S2). Lignin dimers were prominent suggestions among flavonoids, coumarins and stilbenes, all of which result from the phenylpropanoid pathway (Vogt, 2010). This suggests that the group of negatively correlated compounds represents inputs of plant biomass derived molecules that feature recently assimilated carbon and are likely leaching from litter decay processes.

Positively ¹⁴C-age correlated sum formulae were more saturated and more oxygenated, their molecular weight was higher and in some cases they contained nitrogen or sulfur as heteroatoms (Table 2). From the elemental composition in the van Krevelen diagram the positively correlated sum formulae plot in the tannin-like area [H/C: 0.5 – 1.5, O/C: 0.6 – 1.2; (Antony et al., 2014)]. However, database search did not select tannin-type structures for positively correlated sum formulae even though tannin structures are annotated in the ChEBI library and in contrast structures of glycosylated phenylpropanoid derivatives were suggested (Supplementary Tables S1, S3). The tannin area in the van Krevelen diagram lies aside the lignin and carbohydrate areas [H/C: 1.5 – 2.2, O/C: 0.6 – 1.2; (Antony et al., 2014)] and consequently the structure suggestions are very likely (Figure 3).

The observed ¹⁴C-age related trends regarding the elemental composition and structure suggestions for the molecular entities differ from aging trends in marine DOM, likely due to largely different timescales of DOM aging. There, the composition converged toward carboxyl rich alicyclic molecules (CRAM) (Hertkorn et al., 2006; Lechtenfeld et al., 2014). Marker compounds related to marine DOM degradation state (Flerus et al., 2012) were entirely absent within the data set of this study.

From our assessment of the suggested structures of molecular entities (Supplementary Tables S2, S3) that were associated with ¹⁴C young and old carbon samples we identified that plant-derived phenylpropanoid compounds were core structural units for both groups, although it is probable that there are other structural units that may be suppressed during ionization (Nebbioso et al., 2010). Molecular entities associated with ¹⁴C-young or old DOM were characterized by a high degree of methoxylation and glycosylation, respectively. This finding is consistent with a previous investigation of plant-derived SOM (Gleixner et al., 2002).

While demethylation is a long known processes during lignin degradation (Jin and Kirk, 1990), it has recently been reported that glycosylation of phenylpropanoids greatly affects their bioavailability (Le Roy et al., 2016). In this study we cannot disentangle if the observed glycosylated phenylpropanoids remain from the original substrate or are synthesized in the environment. Yet, a combination of both processes is probable. Glycosylated phenylpropanoids are widely present in plants and glycosylated lignin oligomers can be stored in leaf vacuoles (Dima et al., 2015). Due to the increased solubility in water compared to their respective aglycones, the glycosides might be preferentially leached during biomass decomposition and accumulate in DOM over time. Furthermore, glycosylation of phenylpropanoids increases their stability by weakening their nucleophilic character as well as their tendency to polymerize. Because polymerization and reactions of phenylpropanoids with biomolecules are potentially harmful if they happen in cells, active glycosylation of these compounds could be a way for organisms to tolerate elevated levels of otherwise toxic lignin-derived molecules. This detoxification effect and the solubility increase are especially crucial for lignin decomposers since degradation products resulting from the activity of extracellular enzymes must be taken up by the mycelium (Jeffries, 1994; Sanchez, 2009).

The introduction of ¹⁴C young molecules derived from lignin degradation, their detoxification and stabilization can provide a framework for ¹⁴C age related marker compounds. The differing evolution of individual molecular entities over time is highlighted in the relative intensity distributions of two possible ferulic acid derivatives (Figure 4). Their final identification needs of course further structural investigation. However, the preliminary identification of C₂₀H₂₀O₇ and C₂₂H₃₀O₁₄ is likely as ferulic acid is a key component of lignocellulose (de Oliveira et al., 2015). On the one hand glycosmisic acid, a lignin dimer comprised of a ferulic acid and a coniferyl alcohol substructure showed linearly decreasing relative intensities with respect to increasing age of the samples, suggesting continuous degradation. On the other hand ferulic acid in its form as a disaccharide showed strongly increasing relative intensities. From not being detected in two samples with zero and one year of cal ¹⁴C-age, the ferulic acid disaccharide entity built up to stable levels within ∼10 years. The observed build-up of persisting lignin derivatives supports recent advances in dissolved lignin analysis, suggesting lability-based pools of lignin-derived molecules (Feng et al., 2017).

The decrease in the relative intensities of the feruloyl-coniferyl-dimer highlights the potential usability of lignin dimer molecules as fast flow markers for ¹⁴C young, surface-derived DOM inputs in the CZ. High abundances of this new marker in ground water could indicate fast transport of water and DOM through the subsurface. As aromatic organic matter is known to interact with mineral surfaces, the potential applicability of the proposed molecular markers in groundwater will depend on the mineral structure of the respective subsurface environment (Klotzbücher et al., 2016). Recent investigations show that a considerable proportion of lignin-derived phenols might be irreversibly sorbed to minerals (Kaiser and Guggenberger, 2000; Hernes et al., 2013), thereby limiting the potential traceability of lignin-derived markers into groundwater. Low pH conditions, as are present in sandy forest soils, can further favor strong OM sorption to mineral surfaces (Kleber et al., 2015). Therefore further studies focusing on subsurface DOM are required in order to decide which of the proposed potential marker compounds are applicable in the respective pedologic settings.