Two species of deciduous Quercus oak (Downy oak Quercus pubescens and Sessile oak Quercus petraea) that had been collected in Switzerland during 1952 and 2007 and have been stored in the University of Zürich Herbarium (n=8) under constant temperature and humidity conditions were dedicated to the retrospective analysis of this study (Fig. 1a). One leaf of the bunch was sampled for Q. pubescens, n=5, collected on 11 August 1952 (converted to decimal year: 1952.611, leap year), 15 August 1965 (1965.622), 14 September 1968 (1968.704, leap year), 3 July 1973 (1973.504), and 2 July 1982 (1982.502); and Q. petraea, n=3, collected on 30 May 1966 (1966.411), 25 May 1995 (1995.398), and 8 July 2007 (2007.518). The specimen labels provided us fragmentary information such as altitude and location where the samples were collected (Table 1). Approximately 3 mg of the leaf samples were cut by clean tweezers for the bulk Δ14C measurement. Remaining leaf samples were stored at -20 °C until the following analysis.

Chl a was extracted from each leaf sample using the modified method of Ishikawa et al. (2015) In brief, crude pigments were extracted from 100–200 mg of dried and crushed leaves using about 30 mL of acetone in a PTFE tube (Oak Ridge Centrifugal Tube, 3114-0050, Thermo Scientific, USA). The tubes were ultrasonicated for 15 min and were centrifugated at 4000 rpm for 30 min. The supernatant was transferred into a pre-combusted glass vial (ASE collection vial 60 mL, 048 784, Thermo) and was dried under the argon stream. About 2 mL of dimethylformamide (DMF) was added to the PTFE tube, ultrasonicated, centrifugated, and transferred into the 60 mL glass vial. The DMF extraction was repeated one more time to increase the recovery. After drying up the samples, they were transferred using dichloromethane (DCM) into a pre-combusted glass vial (4 mL screw vial, 5183-4448, Agilent Technologies, USA) and were dried using argon. Since the extracted Chl a is quickly degraded at room temperature in laboratory, about 1 mL of 2 molmL-1 hydrochloric acid was added to the vial to convert Chl a into pheophytin a (Pheo a) to increase stability. Therefore, the present study regards Pheo a as a surrogate of Chl a, and does not consider a potential difference in Δ14C values between Chl a and Pheo a. The only difference between Chl a and Pheo a is the presence or absence of magnesium at the center of the tetrapyrrole ring (Fig. 1b). About 1 mL of n-hexane was added to the vial, and the liquid–liquid extraction was made three times, and the organic layer was transferred into another 4 mL vial. After drying up, a 0.2 mL of DMF was added to the vial and the solution was passed through a membrane filter (Cosmospin Filter G, pore size 0.2 µm, 06549-44, Merck, Germany) and recovered in a pre-combusted 1.2 mL glass vial (Supelco 29658-U, Merck).

The acidified crude pigment dissolved in DMF was injected to a high-performance liquid chromatography (HPLC) system (1260 Infinity, Agilent Technologies) for the first separation using a reversed-phase column (Eclipse XDB-C18, 5 µm particle size, 4.6mm×250mm, P/N 990967-902, Agilent Technologies) with the corresponding guard column (5 µm particle size, 4.6mm×12.5mm, Agilent Technologies). All the solvent used in the following wet chemical operation was higher than the HPLC grade. The solvent gradient was programmed as follows: acetonitrile : ethyl acetate : pyridine =75:25:0.5 (v/v/v) held for 5 min, then gradually changed to 67.5:32.5:0.5 (v/v/v) in 15 min, followed by flushing (25:75:0.5, v/v/v for 5 min) and equilibration (75:25:0.5, v/v/v for 5 min). The flow rate and temperature were set constant at 1.0 mLmin-1 and 30 °C, respectively. Three injections were made per sample, and Pheo a and its allomer and epimer were collected using a fraction collector based on their retention times (14.5–16.0, 16.0–18.0 and 18.0–19.5 min, respectively) at the 660 nm wavelength of the diode array detector (DAD). The concentration of Pheo a and its derivatives (0.2–0.7 µgmg-1) was two orders of magnitude smaller than that typically found in fresh Quercus leaves (10–20 µgmg-1, Rodríguez-Calcerrada et al., 2008), probably due to significant amounts of degradation during the long-term storage. However, the degradation does not impact radiocarbon (Δ14C) results because any isotopic fractionation during the storage up to 70 years is internally corrected by δ13C values (see Sect. 2.5 Radiocarbon measurements). It should also be mentioned that the Chl a survived after all is obviously intact because non-photoautotrophs such as fungi potentially colonizing the leaf surface during the storage in the herbarium cannot synthesize Chl a. Pheo b and its derivatives (allomer and epimer) being also found in the chromatogram (Fig. 1c) were not collected because their concentrations were too small to implement the CSRA measurement. These fractions were combined in a pre-combusted 6 mL glass vials, dried under the argon stream, and re-dissolved in a 0.2 mL of DMF.

The Pheo a fraction after the above first separation was re-introduced to the HPLC for the second separation using another column (Eclipse PAH, 5 µm particle size, 4.6mm×250mm, P/N 959990-918, Agilent Technologies). The solvent gradient was programmed as follows: acetonitrile : ethyl acetate : pyridine =80:20:0.5 (v/v/v) held for 5 min, then gradually changed to 32:68:0.5 (v/v/v) in 18 min, followed by equilibration (80:20:0.5, v/v/v for 5 min). The flow rate and temperature were set constant at 1.0 mLmin-1 and 15 °C, respectively. Three injections were made per sample, and the Pheo a was re-collected based on their retention times (16.5–17.8 min for allomer, 17.8–19.2 min for Pheo a, and 19.2–20.5 min for epimer) at the 660 nm wavelength. After drying the Pheo a fractions, the liquid–liquid extraction was made using water :n-hexane : DCM (1:0.7:0.3, v/v) three times and the organic layer was transferred into a pre-combusted 1.2 mL glass vial. The vial was dried using argon and kept at -20 °C until the following analysis.

We determined stable carbon and nitrogen isotopic compositions (δ13C and δ15N) and C/N of bulk leaves and purified Pheo a using the elemental analyzer coupled to isotope ratio mass spectrometry (Delta Plus XP) with a Conflo III interface (Thermo Finnigan, Bremen, Germany) for ultra-small-scale analysis (nano EA/IRMS) system (Isaji et al., 2020; Ogawa et al., 2010). In brief, a small piece of leaves (50–70 µg dry weight) was used for the bulk measurement. The purified Pheo a was dissolved in a 400 µL of trichloromethane (TCM). A portion of the TCM solution corresponding to about 3 µg of Pheo a (10–42 µL, depending on the Pheo a concentration) was transferred into a pre-cleaned tin capsule using a pre-cleaned glass syringe on a hot plate set at 80 °C. The data were calibrated using three interlaboratory-consensus reference materials (standard name, δ13C, and δ15N: BG-A, -26.9 ‰, and -1.7 ‰; BG-P, -10.3 ‰, and +13.5 ‰; and BG-T, -20.8 ‰, and +8.7 ‰) (Tayasu et al., 2011) and three in-house reference materials (BG-LC-G, -13.4 ‰, and -5.4 ‰; BG-GC-G, -13.4 ‰, and -5.7 ‰; and SK-GC-V, +0.2 ‰, and +60.4 ‰). An in-house Chl a standard was also measured to assess reproducibility of the C/N ratios (n=3, mean and standard deviation, 12.1 ± 0.6) prepared for ultra-small-scale measurements (Isaji et al., 2020). The analytical errors of the δ13C and δ15N measurements obtained by the repeated analyses were less than ±0.37 ‰ for δ13C (n=22, 1.1–6.9 µgC) and less than ±0.64 ‰ for δ15N (n=19, 0.14–0.9 µgN).

Based on the observed C/N ratios of purified Chl a fractions by the nano EA/IRMS measurements, a mass-balance equation with respect to impurity being derived from sample matrix and/or procedural blank was written as follows. 1C or NObserved=C or NExpected+C or NImpurity Equation () was rewritten in terms of carbon (hereafter referred to as impurity carbon in percentage) as follows. 2CImpurityCObserved=CObserved-CExpectedCObserved=1-C/NExpected×NExpectedCObserved Under the condition where all the nitrogen detected on EA/IRMS is derived from Chl a (i.e., NImpurity=0), Eq. () was rewritten in terms of nitrogen as follows. 3NObserved=NExpected Substituting Eq. () for Eq. () yielded the following equation. 4CImpurityCObserved=1-C/NExpected×NObservedCObserved=1-C/NExpectedC/NObserved Given that C/N_Expected is 11.8 for Chl a (weight / weight, 55 carbon atoms and 4 nitrogen atoms) and the repeated nano EA/IRMS measurement of our in-house Chl a standard gives analytical error of C/N ±1.2 (2σ), the analytically permissible range of the impurity carbon percentage was from -11 % to 9 %, corresponding to C/N_Observed from 10.6 to 13.0, respectively. This was used as the criterion for sample Chl a purity in this study. It should be mentioned that the criterion is more relaxed compared to the stricter one when the EA/IRMS measurement was implemented at a larger scale (Isaji et al., 2020).

To identify and characterize impurity carbon in the purified Pheo a fractions, three additional assessments based on (i) diode array detector (DAD), (ii) Orbitrap MS, and (iii) GC/MS spectra were performed. Assessment (i) was subject to all eight samples, while assessment (ii) subject to 1952, 1968, 1973, 1982, and 1995 samples and assessment (iii) to 1952 and 1968 samples due to availability of leftover materials after CSRA. Experimental details and analytical settings of each assessment are found in Sect. S1 in the Supplement.

The radiocarbon content is reported as F14C (Reimer et al., 2004). The present study derived Δ14C (‰) from the reported F14C value as follows. 5Δ14C=F14C×eλ1950-x-1, where λ and x are the decay constant of 14C (1/8267=1.21×10-4) and the year when 14C was measured, respectively. Δ14C is expressed in ‰, which is also formulated as follows (Stuiver and Polach, 1977). 6Δ14C=δ14C-2(δ13C+0.025)(1+δ14C), where δ13C, δ14C, and Δ14C are expressed in ‰ and carbon isotopic fractionations are internally corrected by δ13C (Stuiver and Polach, 1977). The bulk leaf Δ14C (Δ14C_Leaf) values of a small piece of leaves (3–4 mg dry weight) were determined using an accelerator mass spectrometer (AMS) at the Institute of Accelerator Analysis (Kanagawa, Japan; AMS lab code IAAA) in which analytical errors (1σ) were better than 4.0 ‰. The compound-specific radiocarbon analysis (CSRA) of Chl a (Δ14C_Chl) were conducted according to Haghipour et al. (2019). In brief, 16–40 µgC of purified Chl a fractions (n=8) were submitted to CSRA. The dried Chl a samples were dissolved in 30 µL of dichloromethane. 15–30 µL of each sample was transferred into a pre-cleaned (washed with DCM three times) tin capsule (3 mm diameter, 6 mm height, and 25 µL volume, P/N 84.9906.26, Lüdi Swiss, Switzerland) using a pre-cleaned glass syringe on a hot plate set at 80 °C. The syringe transfer was repeated three times to increase recovery. The folded capsules were then placed on an autosampler, which is transferred into an elemental analyzer (Elementar, Handforth, UK) where they were combusted to gaseous CO2 before being sent to a gas ion source/miniature carbon dating system (GIS/MICADAS) at Ion Beam Physics Laboratory, ETH Zürich (lab code ETH) in which analytical errors (1σ) were better than 8.1 ‰. Although the Chl a standards that have modern 14C (i.e., Δ14C >0 ‰) and dead 14C (i.e., Δ14C ∼-1000 ‰) are commercially unavailable, we conducted a blank assessment for the entire procedure and found that the procedural blank has 0.32 ± 0.10 µgC (Fig. 2), which is smaller than that found in typical CSRA studies (e.g., Haghipour et al., 2019; Ishikawa et al., 2018). Even in the most extreme case where the wet chemistry blank Δ14C was -1000 ‰, the effect of the procedural blank on Chl a Δ14C correction is smaller than the AMS analytical error (±8 ‰, 1σ). The procedural blank assessment is detailed in Sect. S2.

We used Δ14C data of atmospheric CO2 and tree rings during 1950 and 2019 (monthly resolution, i.e., 12 data per year, n=833) in the Northern Hemisphere Zone 1 (NH1), which covers aerial Switzerland, provided by Hua et al. (2022). The timeseries dataset (t=1,2,…,833) consists of decimal-year time points and Δ14C values (hereafter referred to as Δ14C_Atm (t)). The decimal years nearest to the times when Quercus samples were collected were identified (t=32: 1952.625; t=188: 1965.625; t=197: 1966.375; t=225: 1968.708; t=283: 1973.542; t=391: 1982.542; t=545: 1995.375; and t=691: 2007.542). The difference in decimal years between the Quercus sample collection time and their nearest time t was 0.008 ± 0.03 (n=8, equivalent to 2.9 ± 9.9 d).

To interpret observed Δ14C values of Chl a, a two-pool model developed for the soil carbon pool (Koarashi et al., 2012) was applied to our dataset with a modification. We considered two different carbon pools as follows. 7FQ(t+1)=FQ(t)(1-1/TQ-λ)+1/TQFAtm(t),8FS(t+1)=FS(t)(1-1/TS-λ)+1/TSFAtm(t), where F is the F14C value, and FQ(t), TQ, FS(t), and TS are the F14C at time t and turnover time of Quercus leaf and soil carbon pools, respectively. We assumed that the Chl a compound is a mixture of these two carbon pools, and its F value is formulated as follows. 9FChl(t)=PQFQ(t)+PSFS(t), where FChl(t) is F14C of Chl a at time t and PQ and PS are proportional size of Quercus leaf and soil organic matter, respectively (PQ+PS=1). Quercus is a deciduous tree with faster and less variable carbon turnover (i.e., 1–2 years, Ichie et al., 2013) than the soil (van der Voort et al., 2019). Therefore, in the model, TQ was set at constant 1.5 years and TS was allowed to vary between 0 and 3000 years. PS was allowed to vary between 0 % and 30 %. The stepwise model was run with intervals of 100 years and 0.1 % for TS (31 models) and PS (301 models), respectively (31×301=9331 models in total) to reconstruct radiocarbon trajectories for each model from 1950 to 2019.

To constrain TS (i.e., turnover time of the soil pool contributing to Chl a) and PS (percentage of the soil pool contributing to Chl a) at time t using the observed and modelled Δ14C_Chl(t) (Δ14C_{Chl, observed(t)} and Δ14C_{Chl, modelled(t)}, respectively) values, we computed their absolute difference at each of the 8 years (i.e., 1952, 1965, 1966, 1968, 1973, 1982, 1995, and 2007) for mutually independent models as follows. 10ΔΔ14C=|Δ14CChl, observed(t)-Δ14CChl, modelled(t)|. The closer to ΔΔ14C=0, the more plausible the TS and PS expected. Therefore, the TS and PS values that gave the smallest ΔΔ14C values were explored for each of the 8 years among the 9331 models (van der Voort et al., 2019).

All the statistical analyses and graphing were performed using MATLAB 2025b (MathWorks, USA).

The present study indicates that Chl a, which is an essential compound for photosynthesis in a variety of autotrophs, involves carbon atoms that are not directly routed from atmospheric CO2 (Fig. 7). Atmospheric 14CO2 concentrations in Northern Hemisphere tend to be higher in winter and spring when the troposphere is well-mixed with the stratosphere to which a significant amount of bomb-derived 14C was injected during the 1950s and 1960s (Randerson et al., 2002). Among the 8 years when our leaf samples were collected, intra-year variations in Δ14CO2 values reached up to 100 ‰ in 1965, followed by 60 ‰ in 1966, and <30 ‰ in 1968 (Hua et al., 2022), all of which were excluded from our model in the interest of the Chl a purity. The bomb 14C seasonality is overlapped but not perfectly concurrent with Quercus oak phenology where the leaf growth and Chl a production are maximal in spring and summer (Mészáros et al., 2007). If our Quercus leaf samples, which had been collected from late spring (May) to late summer (September), were predominantly made from 14C-enriched carbon available in the springtime, their Δ14C_Leaf values should be higher than atmospheric Δ14C values at time t (Δ14CAtm(t)) of the respective year. Nevertheless, the Δ14C_Leaf was rather always lower than Δ14C_Atm(t), which is not explainable with the single carbon source. The Chl a compound brings carbon being depleted in 14C by 33.4 ‰–147.1 ‰ relative to the bulk leaf. Since the Chl a concentration in the Quercus oak is at most 1 %–2 % of the total weight in their leaves (Rodríguez-Calcerrada et al., 2008), its contribution to the difference between Δ14C_Leaf and Δ14C_Atm(t) is approximately <30 %. Therefore, it is demonstrated that Chl a is not the only compound that lowers Δ14C_Leaf values. By employing the high-resolution CSRA methodology, we found that there exists other 14C-depeleted compound(s) that cannot be unveiled by conventional bulk radiocarbon measurements.

The Δ14C_Chl value consistently lower than the Δ14C_Leaf value throughout the 50-year-long chronology suggests that very old carbon is derived from elsewhere. Indeed, the difference between Δ14C_Chl and Δ14C_Leaf values is greater in the 1960s and 1970s (when the bomb-derived radiocarbon remained in the atmosphere at high concentrations) than in the others. The results cannot be explained without considering another source of carbon that has turnover time longer than the atmosphere. The Δ14C_Chl values after the 1970s suggest the scale of the turnover time of this additional carbon source. If the source's turnover time was decadal to centennial scales, its Δ14C value should have become higher than atmospheric Δ14C values at some point onward. For example, the source's turnover time 10, 50, and 100 years makes its Δ14C endmember higher than atmospheric Δ14C after 1973, 1989, and 1999 CE, respectively (Gaudinski et al., 2000). Under the condition where this additional carbon source contributes to the Chl a, the Δ14C_Chl values should be always higher than atmospheric Δ14C. Obviously, this was not the case in the present study, nor consistent with our previous study where Japanese blue oak Quercus glauca collected in 2013 showed a Δ14C_Chl value 37 ‰ lower than its Δ14C_Leaf (Ishikawa et al., 2015). Our data collectively suggest that the additional carbon source has turnover time longer than 100 years. Although Quercus oak trees can live >100 years, forests accommodating such a long-living tree are rare in Europe due to frequent human disturbances (Martin-Benito et al., 2021). Therefore, such a carbon source with >100-year turnover time is, most likely, in rhizosphere.

The most plausible (i.e., smallest ΔΔ14C) two-pool models estimated contributions of soil carbon to Chl a in the Quercus leaf (i.e., PS) significantly greater than 0 % (mean and standard deviation, 17 ± 2 %). We designed our indeterministic model not to deduce a unique algebraic solution from the differential equations, but to induce the most parsimonious and the least unlikely constraint from available data. Although the 4 years (1973, 1982, 1995, and 2007 CE) that met our purity criterion did not hold as large a Δ14C offset as the other 4 years (1952, 1965, 1966, and 1968 CE), all of them showed consistent PS values. Even if our estimates were affected by currently unconsidered factors, the most plausible PS values would not be <10 % in all the 4 years (Fig. 6) unless the soil turnover time TS was extended to longer than 3000 years, which is biogeochemically improbable. We acknowledge that the TS values (1000–3000 years) were less constrained than PS in our model, leaving a key question “how deep and old carbon in soils is incorporated into plants” open to debate.

van der Voort et al. (2019) showed that the typical turnover time of surface soil (0–5 cm) in Switzerland is 14–410 years. Considering the Quercus trees grow their root down to 700 cm below ground level (David et al., 2013), it is most likely that they acquire carbon below the soil organic layer via root uptake for synthesizing the Chl a. The turnover time of such soil carbon (i.e., TS) was no shorter than 1000 years in our most plausible model, which roughly corresponds to the soil deeper than 20 cm (van der Voort et al., 2019). It might be somewhat surprising that carbon in Chl a is partly originated from such a deep soil layer, as organic carbon content generally decreases with soil depth (van der Voort et al., 2019). The carbon pool with turnover times on the order of millennium is believed to be tightly stabilized by minerals and hardly accessed by plants or microbes. Therefore, such carbon is apparently the last candidate of a building block for the Chl a compound in Quercus oak among all other carbon available in the rhizosphere. The present study shows the very first but preliminary evidence of the millennial-aged carbon contributing to Chl a, which contrasts sharply with the current pedological paradigm. However, our observation is associated with analytical (lack of working standards) and methodological (unconstrained TS values) limitations as mentioned earlier. Furthermore, it is challenging to estimate the exact soil depth where the old carbon is sourced because it would depend on temperature, precipitation, soil type, and aboveground vegetation, all of which are highly uncertain and beyond the scope of this study. In fact, δ13C and δ15N values varied greatly among the eight Quercus leaf samples, suggesting that their growing condition was quite different from each other. This would also be one of the reasons the estimated PS and TS varied. Further investigations are needed to demonstrate the validity of our results, the potential significance of this process, and the broader relevance with respect to carbon cycling.

Direct uptake of Chl a from soil through root would be unlikely due to its hydrophobicity, phototoxicity, and instability outside the cell (Matile et al., 1999). Quercus is one of the oak trees that develops a symbiosis with ectomycorrhizal fungi (Smith and Read, 2008). It is possible that the ectomycorrhizal symbiosis plays a critical role in breaking down organo-mineral complex in soils (Landeweert et al., 2001) and mobilizing carbon as inorganic forms such as HCO3- or as hydrophilic compounds such as phytol or amino acids that can permeate fine roots of their host plants (Jones et al., 2005, 2009). Although these organic compounds are mainly derived from photosynthetic products, their Δ14C values near the interface between fine roots and ectomycorrhizal fungi might be extremely low (Trumbore, 2000). It is reported that amino acids and monoterpenes are translocated via xylem with water to synthesize a variety of organic compounds including storage proteins or secondary metabolites (Martin et al., 2002; Nabais et al., 2005). Therefore, there is no reason to conclude that Chl a is the only organic compound to which soil-derived carbon is incorporated, as discussed earlier with respect to Δ14C_Atm(t), Δ14C_Leaf, and Δ14C_Chl values. If this was the case, radiocarbon age and provenance within and among compounds in a single tree would be more diverse than previously thought.

Despite no direct evidence on what kind of soil carbon, either organic or inorganic, contributes to the Chl a biosynthesis in chloroplast, previous isotope-labeling studies using 13C, 14C, and 15N showed that plants do take up organic carbon and nitrogen from the soil (Moran-Zuloaga et al., 2015; Rasmussen et al., 2010). Given that some organic nitrogen in plants is derived from soil (Näsholm et al., 1998), so is carbon would be unsurprising. However, microbial remineralization of the labile and labeled organic matter such as amino acids before root uptake particularly evident in agricultural soils is controversial (Farzadfar et al., 2021). Furthermore, it is highly uncertain about the fate of such soil-derived organic matter in the plant metabolism in which Chl a plays a significant role (Masuda and Fujita, 2008). As reported in Cress Arabidopsis thaliana (Ischebeck et al., 2006) and cyanobacteria Synechocystis sp. (Vavilin and Vermaas, 2007), the side chain (phytol, 20 carbon compound) of the Chl a (55 carbon compound) could be salvaged from its catabolic pathway. It is believed that the remaining chlorophyllide a (35 carbon compound) is not recycled because the compound is photo-toxic for plant cells (Matile et al., 1999). This leads us to a hypothesis that plants uptake nitrogen-rich amino acids such as glutamine from rhizosphere, use its amide as a nitrogen source, and transfer the resulting glutamic acid or carbon skeleton to the Chl a biosynthesis (Fig. S1). Although speculative, it is possible that approximately 30 % of chlorophyllide a or half of phytol derived from the rhizosphere explain the PS value (17 ± 2 %) constrained in our two-pool model.