Carbon dynamics during long-term starving poplar trees - the importance of older 1 carbohydrates and a shift to lipids during survival

Abstract


Introduction
Trees require sufficient carbon (C) to build up new biomass (including reproductive structures), fuel respiration, use C for defense and allocate C to storage pools (Chapin et al., 1990;Lambers & Poorter, 1992;Sala et al., 2012).When the C supply from assimilation exceeds demand, trees can store substantial amounts of non-structural carbon (NSC).Those reserves may be used to maintain tree functions (e.g., respiration, osmoregulation, repair, biosynthesis of defense compounds) when C supply is reduced below requirements, like during periods of harsh environmental conditions (e.g., Regier et al., 2009;Hartmann et al., 2013;Hartmann & Trumbore, 2016;Zohner et al., 2019).Carbon storage compounds, including starch, sugars or lipids provide an essential buffer against C shortage and play an essential role in tree's resilience capability (Hartmann & Trumbore, 2016).Large NSC storage pools can be beneficial for the recovery of a tree after stress (e.g., insect herbivore defoliation, drought, fire) (Sala et al., 2010;Dietze et al., 2014;Piper & Paula, 2020).The dynamics of reserve use and their availability during periods of reduced C supply in mature trees, over the short-and long term, are still poorly understood (Gessler & Treydte, 2016;Hartmann & Trumbore, 2016).For a more comprehensive understanding of C storage and remobilization dynamics in trees, studies over several years are needed to improve predictions of tree and forest resilience over time (McDowell, 2011;Rosas et al., 2013;Gessler & Treydte, 2016).
In order to gain insights into C reserve use under stressful conditions, one can artificially produce a lack of photo-assimilate supply via stem girdling.When removing a circumferential band of bark, phloem and cambium of a tree, the C supply from the canopy to the lower stem section is interrupted, and only upward water transport through the xylem is maintained.The stem section below the girdle is isolated from the rest of the tree above and is forced to use C reserves from within the stem or from the root system to maintain metabolic activity beneath the girdle.To date, empirical evidence supporting substrate shifts in trees is scarce, but see Fischer et al. (2015) and Wiley et al. (2019).It is still unclear whether and to what degree all types of reserve compounds, including sugars, starch, and lipids, can be used as respiratory substrate when C supply is limited.Plant lipid metabolism is far less studied due to methodological challenges in quantifying neutral lipids (Fischer & Höll, 1991;Hoch et al., 2003;Fischer et al., 2015), but progress has been made (see Grimberg et al., 2018;Herrera-Ramírez et al., 2021).
The simultaneous measurement of CO2 and O2 allows calculating the ratio of CO2/-O2, an useful indicator for the respiratory substrate identity (cellular level: Respiratory Quotient (RQ)).Respiratory substrates differ in their stoichiometric ratios of C:O:H and in their degree of oxidation.Thus, during respiration, quantities of O2 required as electron acceptor vary depending on the respiratory substrate.During the break-down of carbohydrates, one molecule of O2 is consumed for each molecule of CO2 released, resulting in RQ ~ 1, while for the break-down of lipids more oxygen is needed, resulting in RQ ~ 0.7.While RQ refers to the respiratory processes in the strict sense (i.e.measured at the mitochondrion), the apparent respiratory quotient (ARQ, Angert & Sherer 2011) may imply post-respiratory processes also (see Trumbore et al., 2013 for a summary), and this is the case when measured away from the mitochondrion, e.g. at the tree stem.More precisely, highly soluble CO2 can be transported away from the respiration site (e.g., Teskey & McGuire, 2002;McGuire & Teskey, 2004) or refixation mechanisms during the day (woody tissue photosynthesis) (e.g., Pfanz et al., 2002;Wittmann et al., 2006) or during the night (phosphoenolpyruvatcarboxylase, PEPC hereafter) can fix CO2 locally and therefore reduce the CO2 efflux (ECO2) to the atmosphere, leading to ARQ values below 1 (Angert et al., 2012;Hilman & Angert, 2016).However, the potential role of fixation via PEPC has been investigated mainly in leaves and young green twigs of C3 plants 5 (Berveiller & Damesin, 2008), but might be relevant as a mechanism of local CO2 removal, as high potential PEPC activity has been measured in stem wood of mature beech trees (Helm et al., 2023).
Sugars, starch, and lipids can also be distinguished by their C isotope signal of respired CO2 (δ 13 C) which carry through the respiratory processes (Gleixner et al., 1993;Cernusak et al., 2003;Bowling et al., 2008;Brüggemann et al., 2011).Former studies showed that 2-yr old oak saplings shifted substrate for respiration from recently fixed carbohydrates to starch reserves (below the girdle) after girdling (Maunoury-Danger et al., 2010), deduced from a δ 13 C enrichment of CO2 respired by stems (Brugnoli et al., 1988;Tcherkez et al., 2004).In young Pinus sylvestris trees, reducing C assimilation by experimental shading triggered a shift from carbohydrate-dominated respiration to almost pure lipid-based respiration, indicated by lower δ 13 CO2, as well as lower RQ (Fischer et al., 2015).The δ 13 C signal can also reflect environmental conditions (stomatal closure to avoid water loss), as e.g., a change towards a more enriched δ 13 C signal could be explained by expected changes in photosynthetic discrimination (Farquhar et al., 1989;Högberg et al., 1995).
To enhance our understanding of NSC dynamics in trees, it is important to know how long these C reserves can be stored and how fast they can be used.The bomb-radiocarbon ( 14 C) approach allows determining the mean age of C assimilated by a plant, and thus can be used to estimate the age of substrates used for respiration by calculating the amount of time elapsed between fixation and use, and the time trees take to tap into their long term reserves (Levin et al., 2010;Trumbore et al., 2016).Amazonian tree stems below the girdle mobilized ~5 yr old C for respiration within one month of girdling, and of decade-old C about 1 yr postgirdling (Muhr et al., 2018).Up to 17 yr old C was used for stump sprouts after complete tree removal of Acer rubrum (Carbone et al., 2013) and maximum 16 yr old C was used for new fine root growth after hurricane damage in a seasonally dry tropical forest (Vargas et al., 2009).
In our study, we investigated responses of mature poplar trees to reduced C supply to stem sections.This species is a very common and fast-growing tree species that is known to store, besides sugars and starch, substantial amounts of lipids (Hoch et al., 2003).We acknowledge here the potential effect of root grafting during starvation, as C transfer between trees has been reported in mature poplar trees (DesRochers & Lieffers, 2001;Fraser et al., 2006;Jelínková et al., 2009) and could compensate for the lack of photo-assimilates.We investigated how reduced C supply of recent photo-assimilates via girdling affects respiratory substrate use and mobilization of storage pools in the isolated stem section.In particular, we tested the following hypotheses (Figure 1): H1.After the disruption of the supply of photo-assimilates, poplar trees initially mobilize NSCs (decrease in NSC concentration), increasingly digging into older C reserves (increase in Δ 14 C).
H2. Lipids contribute to metabolism maintenance during starvation, indicated by progressive mobilization and metabolization of lipids as starvation proceeds (decline in ARQ ratios, lower δ 13 CO2 signal).

Study site and girdling treatment
The study site is located in the Thuringian forest, Germany (50°42´50´´N, 10°36´13´´E, site elevation 616 m a.s.l., north slope).Mean annual temperature is ~7°C and mean annual precipitation is 800 to 1200 mm (Bouriaud et al., 2016).Soil was formed on volcanic bedrock.
Our measurements were carried out in the growing season (May to September) of 2018 and all trees and measurements were conducted from 5 th of May to 20 th of September.Due to limited capacity, in 2019, chambers were installed on 3 control and 3 girdled trees from 2 nd of June to 2 nd of September.From 2 nd of July to 3 rd of August 2021, chambers were installed on 4 control and 4 girdled trees.Chambers were installed on the north side of the trees and were covered with reflective foil to prevent heating from direct solar radiation.For details about the chamber set-up and sensor specifications see Helm et al. (2021).Configuration settings were based on a repeated closed chamber-mode with 45-min measurement cycles (CO2 and O2 raw data were recorded every 10 sec).Each cycle was followed by a 15-min flushing period of the chamber with ambient air before a new measurement cycle started.
As a general requirement, O2 as a non-trace gas needs to be corrected for the dilution effect of changing H2O and CO2 concentrations (Helm et al., 2021), therefore we used the relative humidity sensor integrated in the COZIR non-dispersive infrared (NDIR) absorption sensor (Gas Sensing Solution GSS, Cumbernauld, UK) for the correction.Relative humidity was converted to [H2O] using the Magnus formula (see Helm et al., 2021).
Measurements of CO2/-O2 headspace concentrations over time are subsequently used to calculate CO2 and O2 fluxes.To this end, the linear increase of CO2 and decrease of dilution corrected O2 concentrations of the first 20-min were used, after removing the first 5-min period following flushing to avoid the influence of pressure fluctuations: Where ΔC/Δt is the change in gas concentration over time (ppm s −1 ) for CO2 and O2 (absolute value), respectively, V is the volume of the chamber (m 3 ), A is the stem surface area (0.0028 m 2 ), P is the barometric pressure (kPa; from LuminOx sensor), R is the molar gas constant (0.008314 m 3 kPa K −1 mol −1 ) and T the temperature (Kelvin).Volumes of the stem chambers ranged from 70 to 105 cm 3 and were determined after installation by injecting water with a calibrated syringe into the chamber headspace.To allow air-bleeding from headspace we inserted two syringe needles into the chamber headspace; one to inject water, the other to vent air from the chamber (Supplementary Table S1).

Sensor calibration
In 2018, CO2 sensors initially were calibrated every 6 weeks.Upon noticing substantial sensor drift beyond 3 weeks since calibration (Helm et al., 2021) we excluded all data recorded more than 3 weeks since last calibration, and from 2019 on, sensors were calibrated every ~ 3 weeks.For O2 sensors, electronic storing of calibrated parameters was not possible, therefore stability/validity has been checked regularly by evaluating possible drift (predefined limit of the slope: 1±0.03) using different reference gas concentrations (Westfalen AG, Münster, Germany).For more in-depth information about the calibration procedure and calibration unit see Helm et al. (2021)..

Sap flow rate
Sap flux density (l cm −2 h −1 ) was monitored during the growing season 2018 (May to September), only.We measured the sap flow with Sap Flow Meter SFM1 sensors from ICT International installed below the glass flask chamber (see 2.6) at approx.0.5 m stem height.
Sap flux density was recorded every 20-min and converted to sap flow rate (l h -1 ) by using the software Sap Flow Tool (ICT International, University Ghent).Sapwood depth was assumed to be 50% of the xylem radius.

Non-structural carbohydrate analysis
The effect of girdling on storage reserves in the sapwood was evaluated by seasonal NSC measurements of stem cores.In 2018 we sampled stem cores twice a year, before the girdling (DOY 172) and 90 days after the girdling (DOY 262).In 2019 and 2021 we sampled once a year  DOY 147 and DOY 195).Stem cores from all 12 trees were immediately placed in a cooler (0-5°C) for transport to the laboratory, where they were dried for 72 h at 60°C within 4 h after the core collection in the field.Cores were sanded with sand paper to facilitate identification of annual rings under a light microscope (Stemi 2000-C, Carl Zeiss Microscopy GmbH, Göttingen, Germany).Defining the outermost (most recent) annual ring as ring 1, we then cut the cores into pieces consisting of rings 1-6 (without bark) and 7-14.Wood material was ground to a fine powder in a ball mill.Aliquots of 30 mg homogenized wood material were analyzed for concentrations of sugars (glucose, fructose, sucrose) and starch according to protocols S1 and S2 from Landhäusser et al. (2018).In short, ethanol (80% v/v) was used as solvent for sugar extraction.After vortexing for 1 min, incubating at 90°C for 10 min and centrifuging at 13,000g for 1 min, supernatants were analyzed by a High-Performance Liquid Chromatography coupled to a Pulsed Amperometric Detection (HPLC-PAD).Concentrations are expressed in glucose equivalents per dry wood mass.Starch was extracted from the remaining pellet from soluble sugar extraction using two digestive enzymes: alpha-amylase and amyloglucosidase (Sigma-Aldrich).The glucose hydrolysate was measured by HPLC-PAD.
2.6 14 C and 13 C signatures of respired CO2 We repeatedly collected gas flask samples for δ 13 CO2 and Δ 14 CO2 measurements by means of additional stem chambers that were installed in close proximity below the respiration chamber (Supplementary Picture S2).A glass flask chamber consisted of polypropylene plate equipped with three connectors for sampling flasks and a foam frame (2.4 cm thick; 14 C neutral material) placed between the stem and the plate to ensure airtight sealing.Chambers for sampling isotopes were installed temporarily for sampling campaigns using 4 rachet straps for fastening the chamber on the stem.Three flasks were connected to the chamber and opened.Each of these incubation periods lasted approx. 1 week to ensure sufficient amounts 11 of CO2 for 13 C and 14 C analysis and establishment of steady-state conditions.Then, flask inlets were closed and glass flasks removed from the stem.The sampling flasks were custom-built, made of glass and with a volume of 115 ml.Glass flasks were evacuated prior to sampling and inlets were equipped with a Louwers O-ring high-vacuum valve (Louwers H.V. glass valves, Louwers Glass and Ceramic Technologies, Hapert, Netherlands) (Muhr et al., 2018).We conducted three pre-girdling samplings.Following girdling, sampling took place at approx.monthly intervals from July to October in the same year, from June to September in 2019 and from July to September in 2021.Leaks in the field or problems during extractions repeatedly resulted in smaller number of replicates than intended (n = 6 for control and girdling each) (Supplementary Table S3, S4 and S5 for 13 C and 14 C sampling).Flask samples were brought to the laboratory at the MPI-BGC in Jena for analysis.
For Δ 14 C of CO2, gas samples (~0.5 mg of C) were cryogenically purified, graphitized and analyzed with an accelerator mass spectrometer (Lowe, 1984;Vogel et al., 1984;Steinhof et al., 2017;Muhr et al., 2018).Radiocarbon data are reported as Δ 14 C (‰), i.e. the per mil deviation from the 14 C to 12 C ratio of oxalic acid standard in 1950.Accounting for any massdependent fractionation effects, Δ 14 C is corrected to a δ 13 C value of -25‰ (Stuiver & Polach, 1977).Detailed calculation can be found in Trumbore et al. (2016).The ∆ 14 C of any given sample can be used for estimating the ´age´ of respired CO2 by calculating the difference to the atmospheric Δ 14 C of the study site at the time of sampling.
with Δ 14 C values clearly below atmospheric Δ 14 C (< 5‰) were discarded, as those samples might reflect influence of CO2 from local fossil sources.
We used the following formula to estimate the mean age of respired CO2 (yr) according to Hilman et al. (2021): Δ 14 Csample is the measured value from the gas sample, Δ 14 Catmosphere is the signature of the current atmospheric CO2, and 4.7‰ is the mean annual decline in atmospheric Δ 14 C.The estimate for the atmospheric Δ 14 C during the growing seasons was +2.3‰ (2018), -2.4‰ (2019) and -5.4‰ (2021), respectively.
For δ 13 CO2 measurement two aliquots (50 µl) from each gas sample were analyzed with an isotope ratio mass spectrometer (Delta+ XL; Thermo Fisher Scientific, Bremen, Germany) coupled to a modified gas bench with a Conflow III and GC (Thermo Fisher Scientific).δ 13 CO2 samples were analyzed against a laboratory air standard on the Vienna Pee Dee Belemnite scale realized by the Jena Reference Air Set-06 (JRAS-06) (Wendeberg et al., 2013).The values obtained were corrected using the Davidson equation (Davidson, 1995) to account for fractionation effects: where Cs is the CO2 concentration of respired CO2 in the flask (ppm), δs is the isotopic composition of respired CO2 (‰), Ca is the ambient air concentration of CO2 (assumed 400 ppm) and δa is the isotopic composition of ambient air (assumed -9‰).
Besides δ 13 C of respired CO2, we measured δ 13 C also for soluble sugars and neutral lipids following a modified protocol (Bligh & Dyer, 1959;White et al., 1979) and liquid chromatography (Schwab et al., 2019).Increment cores from all 12 trees were extracted at breast height using a standard 5.15 mm diameter increment borer (Haglöf Company Group, Sweden) in 2021 (DOY 236, ring 1-14).Wood material was ground to a fine powder in a ball mill (MM 400, Retsch, Haan, Germany) and in a next step phase-separated: water-soluble C was analyzed as proxy for soluble sugars and C extractable in methanol:chloroform solution (total lipids) was transferred to silica gel column.The lipids that eluted by chloroform were regarded as "neutral" and analyzed (further details see Supplementary Method S1).Aliquots from the extractions were put into tin cups, dried and afterwards the measurement was performed with a Finnigan MAT DeltaPlus XL EA-IRMS (ThermoFinnigan GmbH, Bremen, Germany), coupled to an autosampler (Koppenaal et al., 1991).We used the pictures to quantify the percentage of the aerial surface covered by neutral lipid droplets using ImageJ (Schneider et al., 2012).We quantified the percentage of lipid coverages in small regions of interest (ROIs) of 0.25 mm 2 randomly generated by the automatic script used for Image J (Anexx 1).We divided the images in sections corresponding to 3 mm of wood counted from bark to pith and in each 3 mm wood section we measured 50 ROIs, leading to a total of 500 ROIs along the 3 cm of wood.We estimated the percentage of the aerial surfaced covered by neutral lipids in the wood as the average between all the measured ROIs along the wood sample.

Potential PEPC activity in woody tissue
For potential PEPC activity measurements, we collected stem cores from all 12 trees in August 2019 (DOY 236).Cores were immediately frozen in liquid nitrogen in order to avoid any further metabolic activity, transported to the laboratory and stored at -80°C freezer.We cut the first 2 cm of stem material (bark to xylem) and ground the wood to a fine powder with a mortar and pestle in liquid nitrogen.A discontinuous assay was performed following the steps of Bénard and Gibon (2016) in order to quantify potential PEPC activity.We used 20 mg of woody  Aliquots, together with 500 µL of extraction buffer, were shaken for extraction.Extracts were centrifuged for 7 min (3000 g, 4°C) before the extracts were diluted by a factor of 2000 (w/v).
NAD + standards were prepared in the before mentioned extraction buffer (ranging from 0 to 1 nmol per well).Afterwards, those standards and the diluted extracts were incubated for 20 min in a 20 µl medium (100 mM Tricine-KOH pH 8.0, 20 mM MgCl2, 1 unit ml -1 malate dehydrogenase, 10 mM NaHCO3, 0.1 mM NADH, 1% w/v polyvinylpyrrolidone, phosphoenolpyruvate 0 (blanks) or 2 mM (maximal activity)).In order to stop the reaction 0.5 M HCl (20 µl) was used.In order to destroy NADH, the 96-well microplate was sealed and incubated for 10 min at 95°C.In a next step, the microplate had to acclimate to room temperature, and a neutralization step with NaOH 0.5M (20 µl) and 0.2 M Tricine-KOH followed to adjust the pH to 9.0.Together with 6 units ml -1 alcohol dehydrogenase, 100 mM Tricine-KOH pH 9.0, 4 mM EDTA, 0.1 mM PES, 0.6 mM MTT, and 500 mM ethanol, NAD+ was quantified.The absorbance at 570 nm was measured at 30°C in a filter-based microplate reader (SAFAS MP96).To calculate the amount of NAD + formed during the first step of the assay, the reaction rates (mOD min -1 ) were used.For further details see Bénard and Gibon (2016).

Statistics
All analyses were performed using R software (R Development Core Team, 2019).We used R package climatol for walter lieth climate graph.We used pad function from the padr package for linear interpolation of the flux data to fill flux data gaps shorter than 2h.Flux data were discarded if R 2 of the slope of the linear regression was < 0.96 and relative humidity > 99% (after filtering 2018: 89%; 2019: 81%; 2021: 65% used).The ARQ ratio was calculated as the slope of changing CO2 concentration over time divided by the negative slope of changing O2 concentration over time (slope CO2/-slope O2).For ARQ values we applied an outlier removal function, accepting only ARQs between 25% quantile -1.5*Inter Quartil Range (IQR) and 75% quantile + 1.5*IQR.Data were averaged over 6h time intervals (net efflux of CO2 (ECO2), net influx of O2 (IO2) and ARQ) for raw data plotting.We computed daily mean values only when data for the whole 24h period exist.
We used the lme function (nlme package; Pinheiro et al., 2017) to perform linear mixed-effect models.We analyzed if treatment influenced ECO2 and IO2, ARQ, NSC, δ 13 CO2 and Δ 14 CO2 month-wise (in 2018 three-week-average).Treatment was considered as a fixed factor, while tree and if applicable sensor ID, to account for the effect of different sensors being installed in chambers across years, were considered as random factors.An autocorrelation structure was included into the models to account for temporal correlation.The model's normality of residuals was checked visually (quantile-quantile (Q-Q) plots).All results were expressed as mean ± standard deviation (SD).

CO2 efflux, O2 influx and ARQ
ECO2 and IO2 during the pre-girdling period did not differ between treatments (p = 0.48 and p = 0.53, for ECO2 and IO2, respectively, Figure 3).After the girdling event in 2018, a significant difference was observed in ECO2 between treatments for the measurement period in August (p < 0.01).One yr after girdling, control and girdled trees differed significantly (p = 0.03 and p = 0.02, for ECO2 and IO2 respectively) with a marked decline of ECO2 and IO2 in girdled trees.
However, in 2019 fluxes in control trees were also ca.40% lower than in 2018.Daily maximum values of 6.7 (control) and 2.4 (girdling) µmol m −2 s −1 were recorded for ECO2, while for IO2 daily maximum values reached 9.7 (control) and 2.9 (girdling) µmol m −2 s −1 in 2019.In 2021, differences between treatments increased (p < 0.001, p < 0.0001 for ECO2 and IO2).Control fluxes were twice as high as in 2019, roughly the same as in 2018.Differences between ECO2 and IO2 were significantly different in all 3 yr (p < 0.001).

Sap flow rate and ARQ in 2018
Sap flow rate (l h -1 ) clearly decreased after the girdling event (Supplementary Figure S1).
When looking at daily patterns of ARQ in 2018, the ratio was significantly higher during the night (~8pm -~4am; 0.93 for control, 0.83 for girdling) compared to daytime (~8am to ~4pm; 0.85 for control and 0.79 for girdling), when sap flow rate is maximal (Figure 4).Negative correlation was found between ARQ and sap flow rates (Pearson correlation, r 2 = -0.57,p < 0.01 for both control and girdling).

Non-structural carbohydrates and neutral lipids
Pre-girdling sampling of the outer stem segment (0-6 yr) showed no differences in soluble sugar concentration of the xylem (glucose, fructose, sucrose; mg g -1 ) (p = 0.3) and starch concentration (mg g -1 ) between treatments (p = 1.0; Figure 5).In 2018 and 2019, starch concentration was lower than 2 mg g -1 , independent from treatment.Soluble sugar concentration increased from 1.0 to 5.1 mg g -1 after the growing season (09/2018) in control trees.Finally, in 2021, soluble sugar concentration and starch concentration varied significantly between treatments (p < 0.001 and p < 0.001, respectively) with mean soluble sugar concentration of 17.5±3.5mg g -1 and mean starch concentration of 11.0±3.0mg g -1 for control trees, while for girdled trees soluble sugar concentration and starch concentration remained low (3.6±4.2 and 1.4±2.2mg g -1 , respectively).The concentrations of soluble sugars and starch in the second stem segment to a maximum depth of ring 14 did not show significant differences in concentrations in all 3 yr (Supplementary Figure S2).
Neutral lipids, analyzed in 2021, were 0.76±0.1 and 0.56±0.2%area in control and girdled trees, respectively without a notable treatment effect (Wilcoxon test, p = 0.4) and high variability in girdled trees (for individual trees, Supplementary Table S6).For visualization of histological slices see Supplementary Picture S3.
3.4 14 C-based estimates of respired CO2 age and 13 C signature of stem-respired CO2 Mean age of respired CO2 from the pre-girdling sampling was 1.4 yr±1.1 (control) versus 1.5 yr±1.3 (girdling) (Figure 6).For the control trees, C age reached its highest value of 4.0 yr±1.2 in October 2018, after leaves had senesced.By contrast, C age from girdled trees increased up to 15.1±11.8 in 2021.In 2018 and 2019, C age between control and girdled trees was significantly different for certain time points with mean differences of all sampling dates in 2019 of 3.5 yr and 2021 of 7.5 yr (for individual trees, Supplementary Table S4 and S5).

Potential PEPC activity in woody tissue
At the end of the growing season in 2019, in vitro PEPC activity (±SD) was 568.2±149.2and 267.3±94.7 nmol g FW min -1 for control and girdled trees, respectively with a notable treatment effect (t-test, p < 0.001) (for individual trees see Supplementary Table S7).

Discussion
Our 3-yr experimental study indicates that the use of a mixture of respiratory substrates with a late contribution of increasingly older reserves provides a mechanism of tree resilience to strong reduction in C supply in poplar trees.Our data suggest that lipid metabolism, indicated by changes in 13 C of respired CO2 (Figure 7), may allow poplar trees to ride out periods of C starvation, yet, further dedicated studies on lipid metabolism will be helpful.Tree decline may take much longer than the duration of our study, as most of the trees were still alive after the 3-yr girdling treatment.

Significant differences in carbohydrate pools between treatments developed only over time
In our experiment, we combined short-and longer-term responses of poplar trees to a girdling treatment.We could not confirm an initial decrease in NSC concentration (H1) as concentration in both treatments was very low (starch < 2 mg g -1 ) in 2018 (Figure 5).Girdled trees apparently downregulated their metabolism in concert with sugar supply, as suggested by the strong reduction in respiration rates in 2019 (Figure 3).The downregulation of respiration and growth can be a strategy to maintain certain NSC concentrations in aboveground organs in order to ensure tree survival (Huang et al., 2019).Reduced growth respiration may explain why sugar concentrations initially remained stable.Girdled trees ceased growth after the girdling event (Supplementary Figure S3, Supplementary Method S2), in accordance with other studies reporting cessation of stem growth below the girdle (Maunoury-Danger et al., 2010;De Schepper & Steppe, 2011;Oberhuber et al., 2017) and reduced growth in chilled mature red maple trees below the phloem restriction (Rademacher et al., 2022).Transport of NSC from neighbouring trees via root grafts has been shown to be critical for survival of root suckers in poplar trees (DesRochers & Lieffers, 2001;Jelínková et al., 2009), however, net exchange between trees usually is very low (Klein et al., 2016) and may not explain why girdled trees were able to maintain NSC concentrations.In other studies on C limitation, no complete depletion of starch reserves had been observed (e.g., Hoch, 2015;Weber et al., 2019) and NSC concentrations of drought-stressed Picea abies, also strongly C limited, did not differ to control trees in aboveground organs, whereas only starch reserves in roots strongly declined under drought (Hartmann et al., 2013).The depletion of starch in mature trees under drought stress may take many years (Peltier et al., 2023).However, with regard to stem girdling, various studies showed that NSC concentrations usually decrease below the girdle and/or in the roots, with a concomitant accumulation of NSC above the girdle (Jordan & Habib, 1996;Maunoury-Danger et al., 2010;Regier et al., 2010;De Schepper & Steppe, 2011;Mei et al., 2015).
In our study, significant differences in carbohydrate pools between treatments developed only over time.In 2021, control trees showed significantly greater carbohydrate concentrations with a 5-to 10-fold increase in starch and soluble sugar respectively, potentially because climate conditions had normalized, after the 2018 and 2019 dry years (see . Surprisingly, concentrations of soluble sugars in two of the 6 girdled trees increased in 2021.We hypothesize a remobilization of NSCs from deeper stem layers or from the roots to the section below the girdle in the stem.Overall NSC concentrations in girdled trees were more or less stable and remained at a low level over the 3 years, despite the lack of new photoassimilate provision.Some of the observed differences between 2018/2019 and 2021 may be due to seasonal variation of NSCs, as sampling dates differed somewhat between years. Concentrations typically decrease after bud break and then increase in the late growing season (Hoch et al., 2003;Richardson et al., 2013;Scartazza et al., 2013;Martínez-Vilalta et al., 2016), however, seasonal variability of NSCs has been shown to be only 10% in stem sapwood of deciduous trees (Hoch et al., 2003), much less than in our study.Also, as our stem cores were not microwaved, this may have resulted in loss of NSCs to respiration during the initial stage of oven drying (Landhäusser et al., 2018).

Slow mobilization of older carbohydrate storage pools after girdling
In accordance with our hypothesis (H1), poplar trees accessed older C pools once the supply of fresh assimilates was disrupted.In a girdling study in the Amazon rainforest, trees that were presumably older than 100 yr used ~ 6 yr old C, already two months after stem girdling (Muhr et al., 2018) while our 14 C data indicated a delayed use of older C reserves (Figure 6).Girdled trees respired slightly older C than control trees starting in late summer 2018, except for the date in October, when leaf shedding was almost complete and both treatments used older stored C. In girdled trees, the age of respired CO2 increased up to a maximum of 15 years (average 2021: 7.5 yrs).These values were in accord with previous studies showing maximum C age of 14 yr after girdling of tropical trees (Muhr et al., 2018) and decade old stored C in temperate trees (Richardson et al., 2013).Control trees relied mostly on recent photo- assimilates throughout the 3-yr period (Figure 7).But, mixing of young and old NSCs for stem respiration was also reported in undisturbed mature oak trees (Trumbore et al., 2015).

The contribution of lipids during starvation
We found evidence supporting our second hypothesis (H2), that poplar trees mobilize and metabolize lipids after starvation via girdling.In August 2018, δ 13 CO2 values close to -30‰ (Figure 6 and 7) indicate that a substantial amount of CO2 originates from lipid catabolism (δ 13 C of neutral lipids ~-32‰, data not shown) but see also 4.4.We found substantial average differences between treatments in δ 13 CO2 (2.5‰), starting already within one month after girdling, indicating that control and girdled trees did not use the same respiratory C source mixture.For control trees, we assume that poplar use a mixture of carbohydrates and lipids supporting respiration (Figure 7; mean ARQ: ~ 0.85, mean δ 13 CO2 ~-28.5‰),similar to what have been found in Pinus sylvestris (Fischer et al., 2015).In general, we observed high seasonal variations in control trees (e.g., -29.22‰ (05/2018) to -24.17‰ (08/2018); see Supplementary Table S3), most likely with increased values during stem growth (Damesin & Lelarge, 2003) and seasonal variations due to phenological changes (e.g., leaf growth, senescence) (Seibt et al., 2008).
The observed decline in ARQ values after the girdling treatment in autumn 2018 was followed  Europe, during which C supply may have been impaired also in control trees from reduced C assimilation.A tree ring analysis showed ~50% reduced tree ring width in 2019, but also in 2020, which can be seen as a legacy effect of drought (e.g., Miller et al., 2023) (Supplementary

Figure S3
).While in 2018, trees either were still able to cope with the dry and hot conditions, where our study was located.ECO2 and IO2 were reduced by ~40% in 2019 compared to 2018 in control trees, which could potentially be explained by the drought effect in 2019.Similar results (50% decline in ECO2) were observed in Quercus ilex when soil predawn water potential decreased (Rodríguez-Calcerrada et al., 2014).Drought-induced decline in stomatal conductance would also lead to increased δ 13 C values of CO2 (due to changes in photosynthetic discrimination; Farquhar et al., 1989;Högberg et al., 1995), as most likely observed in 2018 and 2019.Compared to wetter years, an increase in δ 13 CO2 of control trees might also be due to starch hydrolysis, as an alternative and more enriched C source that causes a shift towards a more enriched CO2 pool (Maunoury-Danger et al., 2010).In 2021, control trees showed no longer an enriched δ 13 C, possibly because fresh C no longer had a drought signal in δ 13 C. Overall, we critically note that we did not assess tree water relations, and that we lack evidence of how severe the drought stress was.

Outlook
Several studies have highlighted shortcomings in the representation of tree stress responses and resilience, mediated by C storage, in vegetation models (Ogle & Pacala, 2009;DeSoto et al., 2020;Peltier & Ogle, 2020;Hartmann et al., 2022).The common assumption that assimilated C via photosynthesis is directly respired to the atmosphere needs to be updated for model improvement (Sierra et al., 2022) as it is contradictory to our study results and previous empirical evidence (Vargas et al., 2009;Carbone et al., 2013;Muhr et al., 2013;Muhr et al., 2018).Combining empirical studies on remobilization and metabolization of C with C dynamic models may help improve predictions and constrain model parameters (Sierra et al., 2022).

2. 7
Quantification of neutral lipids in stem-woodsFor the visualization and quantification of lipids, we took stem cores in 2021 (DOY 147) from 3 randomly selected trees from each treatment.To quantify neutral lipids in the stem wood, we used a histological method based on the protocols proposed byMehlem et al. (2013) andHerrera-Ramírez et al. (2021).We took histological slides (30 µm thick) from the first 3 cm, from bark to pith.The slices were washed with distilled water and then placed in a petri box.Wood histological slices were stained with Oil Red O (ORO) to visualize neutral lipids.ORO stock solution was prepared adding 2.5 g of ORO to 400 ml of 99% (vol/vol) isopropyl alcohol and mixing the solution for 2 h at room temperature.ORO working solution was prepared by adding 1.5 parts of ORO stock solution to one part of distilled water, shaking it for 5 min, letting it stand for 10 min at room temperature and filtering it through a 45 µm filter to remove the precipitates.ORO working solution was added into the petri box until completely Downloaded from https://academic.oup.com/treephys/advance-article/doi/10.1093/treephys/tpad135/7370247 by Max-Planck-Institut für Biogeochemie user on 07 November 2023 slices.We closed the petri box to avoid drying and precipitating of the ORO solution, and let the sample incubate for 20 min at room temperature.Then, we rinsed the samples with running distilled water for ca.15 min, changing water every 5 min.The histological slices were mounted on glass slides using water as a mounting medium, and placed under a coverslip.We took pictures of each histological slice within one hour after mounting them on the glass slide.After that time water started to dry out and the ORO solution started to precipitate.Panoramic photos of the wood slides were taken using an optical digital microscope with large depth of field (Keyence, VHX-6000, USA) at x500 magnification.
pipetting steps were performed using a 96-head robot (Hamilton Star).
by a more drastic decline in ARQ values during the summer months in 2019 (~ 0.63, Figure 3 and 7), indicating the contribution of lipids to maintain metabolism during starvation.During shading-induced starvation, Fischer et al. (2015) found evidence for lipid metabolism, with ARQ ratios ~ 0.7 in Pinus sylvestris trees.In May/June (2018, 2019) we observed highest ARQ, which could indicate the use of carbohydrates caused by mobilization of reserves.The treatment effect in ARQ and δ 13 CO2 decreased then towards the end of the seasons, likely because remobilization of storage decreased at the onset of dormancy.Despite the high Downloaded from https://academic.oup.com/treephys/advance-article/doi/10.1093/treephys/tpad135/7370247 by Max-Planck-Institut für Biogeochemie user on 07 November 2023 not yet severely affected by the 2018 summer drought at the north slope