Lethal drought leads to reduction in nonstructural carbohydrates in Norway spruce tree roots but not in the canopy

Authors


Correspondence author. E-mail: hhart@bgc-jena.mpg.de

Summary

  1. Heat waves and droughts are expected to increase in frequency and severity in many regions with future climate change, threatening the survival of a number of forest ecosystems. However, our understanding of the physiological processes and mechanisms underlying drought-induced tree mortality is incomplete. Here, we present results on the physiological response of young Norway spruce trees exposed to lethal drought stress.
  2. We applied three levels of drought treatment (control, drying–rewetting, complete drought) and monitored relevant physiological functions and processes of carbon and water relations at high temporal resolution until tree death occurred.
  3. Only trees subjected to continuous drought died in our experiment. Trees subjected to drying–rewetting cycles consistently recovered in their ability to transport water, indicating that these trees do not suffer permanent damage to the hydraulic system. In all cases, drought reduced carbon assimilation, caused changes in carbon allocation and appeared to have severely reduced phloem functioning and carbon translocation. Structural growth was sacrificed for carbon investment in maintenance respiration and osmoprotection. Severe drought caused trees to rely on stored carbon reserves but, in contrast to above-ground tissues, only root carbon pools were strongly reduced when trees died.
  4. Our results indicate that drought-induced changes in carbon allocation, use and transport differ between above- and below-ground tissues in trees. While root death may have been caused by carbon depletion, this was definitely not the case in above-ground tissues. Our findings indicate that mortality mechanisms are not defined at the organism level but rather within tree compartments.

Introduction

Drought-induced tree mortality has received much attention in the recent past (Zeppel, Adams & Anderegg 2011) and occurrences of increased tree and forest mortality from drought and heat have been observed within all major types of forest ecosystems (e.g. Elliot & Swank 1994; Condit, Hubbell & Foster 1995; Dutilleul, Nef & Frigon 2000; Williamson et al. 2000; Breshears et al. 2005; Körner, Sarris & Christodoulakis 2005; Bigler et al. 2007; van Mantgem et al. 2009; Brodersen et al. 2010; Peng et al. 2011), indicating emerging climate change risks for forests (Allen et al. 2010). Large-scale forest mortality can cause significant releases of CO2 into the atmosphere (Kurz et al. 2008) and alter the local energy balance (Bala et al. 2007) and thereby influences feedbacks between climate, forests and the global carbon cycle (Malhi, Meir & Brown 2002). Hence, there is a need for a thorough understanding of the processes and mechanisms underlying drought-induced tree mortality.

During drought (defined here as water demand >> water supply for extended periods), plants downregulate transpiration rates by reducing stomatal conductance (Bréda et al. 2006) but this also reduces the rate of CO2 diffusion to the sites of carbon fixation in leaves and thus carbon assimilation rates (Farquhar & Sharkey 1982). Because maintenance respiration cannot be sustainably reduced under prolonged drought (Meir et al. 2008), reduced CO2 diffusion into leaves can lead to negative carbon assimilation rates (Flexas et al. 2006; Güneralp & Gertner 2007). In response, trees may remobilize stored carbon reserves to fuel maintenance respiration during drought (McDowell et al. 2008; McDowell & Sevanto 2010).

Hypothetical frameworks of drought-induced tree mortality suggest that drought duration and intensity determine the cause of tree death depending on a tree's hydraulic strategy. Isohydric species close their stomata and maintain nearly constant leaf water potential while reducing CO2 diffusion into leaves early during drought (Tardieu & Simonneau 1998; Lambers, Chapin & Pons 2008). On the other hand, anisohydric species keep stomata open longer during drought and allow leaf water potential to fluctuate with evaporative demand, permitting higher CO2 diffusion rates into leaves. If drought and the resulting negative carbon balance are sufficiently lengthy, isohydric tree species relying on stored carbon reserves may be more susceptible to carbon starvation (McDowell et al. 2008).

An alternative explanation for tree mortality is hydraulic failure that occurs if drought intensity promotes irreversible xylem cavitation, especially in anisohydric species. The lack of water supply causes plant desiccation before carbon starvation can occur (McDowell et al. 2008). However, the validity of this conceptual framework has been challenged because other potential mechanisms, such as impeded carbon translocation with declining plant water potential, are not considered (Sala, Piper & Hoch 2010). Translocation of stored carbon from sources (storage organs) to sinks (respiring tissue) requires phloem loading and phloem transport – two processes that are negatively affected by decreasing plant water potential (Hölttä, Mencuccini & Nikinmaa 2009). Hence, declining plant water potential could impede the redistribution of stored carbon and cause local depletions of carbon reserves instead of whole-tree carbon starvation. Moreover, the hydraulic framework defines mortality mechanisms at the scale of the organism and does not consider potential differential responses across tree compartments (e.g. leaves, twigs, branches, stem, roots) although it has been shown that hydraulic failure during drought may be constrained in some species to distal parts of the crown (i.e. leaves, twigs, branches) allowing the rest of the tree (i.e. stem, roots) to maintain functional water relations (Rood et al. 2000). We are not aware of any studies reporting similar differential responses during drought-induced mortality with respect to the tree carbon budget.

Despite the increasing body of literature on observed impacts of drought on forest ecosystems, few experimental studies have evaluated changes in tree physiology during drought-induced mortality for individuals. Recent investigation of drought and temperature effects on tree mortality in Pinus edulis (Englem.) showed that elevated temperature during drought increased respiration rates and accelerated tree death (Adams et al. 2009), but direct evidence for carbon depletion was not found (Sala 2009). More recently, Anderegg et al. (2011) linked results from experimental manipulations of aspen saplings with assessments of widespread aspen mortality and discovered that hydraulic failure of roots was linked to landscape patterns of canopy and root mortality. Yet, because saplings did not die from drought during their study, strict experimental support for either mortality mechanism (carbon starvation, hydraulic failure, impeded carbon translocation) is still missing. In a recent study by Plaut et al. (2012), hydraulic constraints on gas exchange, rather than hydraulic failure per se, have been found to promote drought-induced mortality in P. edulis. Although taken together these findings seem to point towards carbon limitation (starvation) as mortality mechanism, no assessment of storage pool loading had been carried out (Plaut et al. 2012). Despite the recent increase in research on drought-induced tree mortality (Zeppel, Adams & Anderegg 2011), further studies specifically designed to investigate mortality processes by killing plants are needed (McDowell 2011). Moreover, because plant hydraulics and carbohydrate metabolism are interdependent (McDowell et al. 2011), experiments explicitly designed to distinguish between drought effects on tree water status and carbon dynamics are needed (McDowell & Sevanto 2010).

We established an experimental facility that can deliver high-resolution tree physiology data during drought-induced Norway spruce (Picea abies L.) sapling mortality and separately evaluate the carbon and water dynamics of different tree compartments (leaves, branches, roots). The general climatic trend in Europe is expected to cause important shifts in the distributional range of Norway spruce and a strong reduction in forest biomass of this economically very important tree species (Hlásny et al. 2011a). Also, anticipated increasing drought may enhance its susceptibility to secondary agents (Dutilleul, Nef & Frigon 2000) and, combined with climate change-driven changes in pest population dynamics (Hlásny et al. 2011b), poses another threat to the survival of Norway spruce in Central Europe. Cavitation occurs in this species at relatively high xylem water potentials while stomatal control against cavitation is considerably tight (Bréda et al. 2006) making Norway spruce a rather isohydric species.

Our experiments were designed to maintain water deficit until tree death occurred while continuously monitoring relevant physiological functions and processes, including (i) xylem transport, (ii) leaf carbon assimilation, (iii) root respiration and its isotopic signature, (iv) structural growth, (v) whole-tree carbon exchange and (vi) carbon storage (starch, sugars) loading and isotopic signature of needles, branches and roots. Our aim was to test whether carbon pools are reduced during severe drought stress and whether such a reduction was strong enough to explain drought-induced tree mortality. Furthermore, by assessing carbon dynamics across all tree compartments, we wanted to evaluate whether storage pools were affected similarly in all compartments. We discuss our findings with respect to proposed mechanisms and highlight the importance of impeded carbon translocation during drought-induced tree mortality.

Materials and methods

Study Location

The study was carried out at the Max-Planck Institute for Biogeochemistry in Jena, Germany (50°54′36″N, 11°33′59″E). Mean annual temperature was 9·3 °C (1961–1990), but it has increased in the last decade to c. 10·5 °C (1999–2010). The institute has an outdoor experimental site that was divided into two sections, one for a rain exclusion experiment and the other for whole-tree chambers (WTC). Considering the many problems in upscaling of ecological processes (Jarvis 1995), the latter served as an alternative for upscaling point measures of carbon metabolism (e.g. leaf gas exchange, root respiration; see below) to the tree level.

Rain Roof

We constructed a 10 × 10 m wooden structure covered with transparent acrylic roof sheeting (light transmittance factor c. 0·90 of photosynthetically active radiation, PAR, 400–700 nm). In early spring, we planted 18 half-sibling Norway spruce of compatible provenances from a nursery (c. 1·5 m tall from root collar, 7 years old, c. 65 mm diameter at stem base) in pots (c. 25 L volume, insulated with aluminium-coated styrofoam sheeting) in a 2:1 vermiculate–sand mixture. Nutrients were supplied by an instant fertilizer (Manna® Wuxal Super 8-8-6 with microelements, Wilhelm Haug GmbH & Co. KG, D\xFCsseldorf, Germany), combined with a slow-release conifer fertilizer (Substral® Osmocote 11-8-17, Scotts Celaflor GmbH, Mainz, Germany). We placed 12 trees under the roof in a completely randomized design and surrounded these with a row of unmonitored border trees to create similar neighbourhood conditions for all sample trees (Fig. S1, Supporting Information).

Whole-Tree Chambers

Six trees were placed in WTC. These chambers were approximately 2·5 m high and 1·3 m in diameter and were covered with 180-μm greenhouse film (Folitec® UV 4, Scotts Celaflor GmbH) attached to a bottom and top plate (Fig. S2, Supporting Information). They were continuously flushed with ambient air at a rate of 130 m3 h−1, which exchanged the air approximately every 1·5 min, preventing excessive heating. Even with this flushing, however, maximum daily temperatures within the WTC were c. 2 °C higher than ambient air temperatures during the course of the experiment.

Treatments

Watering system

Soil water potential measurements from tensiometers were automatically read once an hour and compared to a preset ‘field capacity’ threshold (−0·005 MPa). When the actual value decreased below this threshold for individual trees, they were watered with a magnetic valve–activated watering system. The amount of water added per period was proportional to the deviance from the threshold. This watering regime was applied to all trees in the pretreatment period, which lasted from 1 month (WTC) to 2 months (rain roof trees), and to trees of the control treatment throughout the experiment.

Watering regimes

Groups of four trees were randomly assigned to the following treatment levels: (i) sufficient watering (‘control’), (ii) drying–rewetting cycles (‘cycle’) and (iii) complete drought (‘drought’). The different watering regimes were initiated in July, once longitudinal growth had ceased for the year, at a time when natural droughts can occur in this region. Trees in the ‘control’ treatment were watered regularly to maintain soil moisture at field capacity. Trees in the ‘cycle’ treatment were left unwatered until the relative extractable soil water content (REW, see below) decreased to below 3%. Rewetting aimed at saturating the substrate and was carried out in two watering events. A fixed amount of water was added to the pots (3·5 L), and they were left for several hours before adding more water (1·5 L). After the second watering event, water temporarily accumulated in the saucer below the pot and was slowly absorbed by the substrate. Trees exposed to complete drought did not receive any water from the beginning of treatment until tree death occurred. Because there is no unequivocal definition for tree death, we assumed that trees were dead when vital signs were no longer detectable, that is, foliage had wilted and fallen completely and leaf gas exchange and root respiration rates remained at near-zero levels. These symptoms occurred in drought trees from 13 September onwards. However, continuous data collection and tissue sampling for all treatments continued until 23 September.

Environmental Conditions

Soil water availability was assessed with TDR sensors (CS645, 7·5 cm three-rod probes with a TDR 100 connected to a SDMX50-series multiplexer; Campbell Scientific Inc., Logan, UT, USA) on a subset of trees within each treatment group (two control, three cycle and three drought trees). Volumetric soil water content measured by the TDR (SWi) was converted to relative extractible soil water:

display math(eqn 1)

where SWmax and SWmin are the volumetric soil water content at field capacity and the minimum volumetric soil water content after several weeks of drought.

Photosynthetically active radiation photon flux density (μmol s−1 m−2) was measured with high-stability silicon photovoltaic detectors equipped with acrylic diffusers. Twelve of these sensors were distributed evenly under the roof, and three sensors were place at mid-height in each of the WTC. One more PAR sensor was installed above the roof to measure total incident radiation. All sensor readings were transmitted to a Campbell® AM16/32 relay multiplexer and then recorded with a Campbell® CR23X micrologger.

Tree Measurements

All trees

On each tree, we installed a steel wire dendrometer to measure stem diameter variation. The steel wires of the dendrometers wrapped around a portion of the tree stem at c. 1/3 tree height and were attached to a push-rod potentiometer (Fig. S3, Supporting Information). We used variation in stem diameter as an indicator for phloem hydration and carbon allocation (stem growth). Although changes in plant hydration affect both phloem and xylem tissues, most of the initial shrinking and swelling occur in the nonlignified cells of the phloem making stem diameter variation a reliable indicator for phloem hydration (Zweifel, Zimmermann & Newbery 2005). Variations in bark thickness included in stem diameter variation were neglected because bark tissue is very thin in these young trees.

We also installed a Granier-type heat dissipation sapflow sensor (3 cm long) on each tree. The two probes of the sapflow sensors were installed c. 10 cm apart vertically on the lower portion of the stems with a slight horizontal offset. We computed a sapflow index from the temperature differences between the two probes (eqn 2):

display math(eqn 2)

where Ki is the sapflow index at time i, ΔTmax is the maximum temperature difference between the two sensors measured on a daily basis to account for drought-induced drifts of Tmax (Lu, Urban & Zhao 2004) during an assumed zero-flow (0h - 5h) period, and ΔTi is the temperature difference at time i. Hence, Ki is dimensionless and indicates relative sapflow velocity (Wang & Stutte 1992). While the assumption of night-time zero-flow may not hold when equilibrium between plant and soil moisture takes a long time to establish (e.g. large trees, dependency on deep soil water pools) (Regalado & Ritter 2007), our experimental design (i.e. small trees in pots) resulted in much shorter required time periods for night-time water recharge.

To avoid physically damaging the experimental trees, we measured leaf water potential with a pressure bomb (Skye®, model SKPM 1400/80, Skye Instruments Ltd, Powys, UK) on separate trees undergoing soil drying to below 3% REW. We did not measure branch conductivity to assess hydraulic failure because to do so would require harvesting of large branch portions that would have substantially affected tree hydraulic functioning and carbon balance in the experimental trees. Instead, we assumed that the absence of functional hydraulic failure (irreversible and complete xylem cavitation) at the whole-tree level could be inferred from the recovery of sapflow in the cycle treatment.

Rain roof trees

Each rain roof tree was equipped with a branch and a root chamber. The transparent cylindrical branch chambers (c. 0·8 L volume) were made of methyl methacrylate resin and suspended from the roof. We inserted a terminal branch section of each tree into these chambers through a round opening at the chamber base. To maintain air tightness, the branches were sealed using closed-cell foam plugs slit open to accommodate the branch and inserted tightly into the chamber openings. The chambers were flushed with ambient air at a rate of 3 L min−1 to prevent depletion of CO2 during times of high photosynthetic activity. However, at this flow rate, both δ13C and night-time respiration of branches/needles could not be measured with adequate precision and are not reported here.

Root chambers were made by sealing off the head space of the pots with a stainless steel cover. This head space was also continuously flushed with ambient air at 3 L min−1 to prevent accumulation of CO2 respired by roots. Flow rates were kept constant with electronically controlled mass flow controllers (M+W Instruments®, model D-6211, M+W Instruments GmbH, Leonhardsbuch, Germany). Each branch and root chamber was also equipped with a thermocouple for measuring air and soil temperature, respectively.

Leaf gas exchange, root respiration rates and δ13C

Each tree was measured during a 5-min period per hour, so the whole cycle of all 12 trees was completed after 1 h, with timing controlled by a Campbell® CR10X micrologger and a custom-made valve switching unit. Within each 5-min interval, a second valve switching unit rotated in 100-s intervals among inlet streams of ambient air and outlet streams of branch and root chambers and directed these to a Picarro® G2101-i isotopic CO2 cavity ring-down spectrometer gas analyser (Picarro, Santa Clara, CA, USA). Within each 100-s measuring interval, a core period of 20 s was used to compute an average [CO2] for ambient air and for each sample air type. We then computed leaf gas exchange and root respiration rates as the difference in [CO2] of branch ([CO2]leaf) or root ([CO2]root) chambers minus ambient air within each 5-min cycle.

The Picarro® G2101-i measures [12CO2] and [13CO2] individually at a very high temporal resolution (c. 1 data point/2 s) and computes δ13C (relative to the Vienna Pee Dee Belemnite standard) from these. However, because of short-term variations in source δ13C (i.e. of ambient air) due to fossil fuel emissions in this suburban area, we report δ13CSample (root respiration) as deviations (Δδ13C) from the source (ambient air), hence:

display math(eqn 3)

We computed δ13Croot directly as:

display math(eqn 4)

where [CO2]root was computed for both isotopic species individually.

The nominal volumetric flow rate of ambient air through the chambers (VFR, 3 L min−1) was corrected to account for variations in ambient temperature, air pressure and partial water vapour pressure:

display math(eqn 5)

where DAir is the actual air density (g L−1) computed from data collected at the institutional meteorological station and DAir_max is the maximum density of dry air at normal temperature (0 °C) and pressure at sea level (1·293 g L−1). This flow rate was then used to compute the net carbon exchange (Cnet) by the branch section or the roots within the chamber:

display math(eqn 6)

where ∆[CO2] was measured in p.p.m. and represents the changes in the [CO2] in the air flow due to photosynthesis or respiration. The molar fraction of carbon in this flow difference multiplied by the molecular weight of carbon then yielded the net carbon exchange by the branch section or the roots in the chamber per minute. These estimates were considered constant for each 5-min interval and integrated over the whole 1-h cycle.

Leaf area contained in the branch chambers was estimated with digital images at the end of the experiment. Fresh and oven-dried needles from each chamber were placed on a light table and photographed. We estimated the projected leaf area as the count of pixels beyond a specified RGB threshold (i.e. dark area against bright background). The shrinking factor from fresh to dry state in green needles was applied as a ‘growing’ factor for needles from dead trees, and this allowed estimating their projected leaf area before the onset of drought stress. The area was calibrated with an object of known area (1 Euro coin) and then used to estimate specific leaf gas-exchange measurements for each chamber. Because longitudinal growth had ceased before the start of the experiment, we assumed needle biomass (and leaf surface) to be constant over the time course of the experiment.

Whole-Tree Net Carbon Exchange

A valve switching unit controlled by a Campbell® CR10X micrologger was also used for directing the air stream of each of the six individual WTC to a LI-COR® 6262 IRGA (LI-COR Environmental, Lincoln, NE, USA). The reference cell was constantly flushed with N2, yielding absolute [CO2] measurements. As for the branch and root chambers on rain roof trees, nominal flow rates were corrected for temperature, air pressure and water vapour and converted to net carbon exchange rates. We report whole-tree net carbon exchange of the two drought levels (cycle, drought) as relative deviations from control trees. Trees were removed from WTC after the experiment, separated into structural/functional units (leaves, branches, stems, roots), completely dried at 70 °C and weighed.

Nonstructural Carbohydrate Concentrations and Mobile NSC δ13C

We considered glucose, fructose, sucrose and starch to be the main physiologically important carbon storage compounds and determined their concentrations in leaf, branch and root tissue. Root tissues were sampled before and at the end of the experiment; branch and needle samples were collected twice a week before and during the experiment. Samples were cut from trees with a sharp branch cutter, immediately deep-frozen in liquid nitrogen and kept on dry ice until they were placed in −80 °C freezer for longer storage. For nonstructural carbohydrate (NSC) extraction, frozen samples were vacuum freeze-dried for 72 h and milled with a ball mill (Retsch® MM200; Haan, Germany) to fine powder (Raessler et al. 2010).

Water-soluble sugars

We added 50 mg of the ground samples to 1 mL distilled water and vortexed the mixture until it was homogenized. This mixture was incubated for 10 min at 65 °C in a thermomixer and then centrifuged for 15 min at 2300 g. The supernatant was removed with a pipette and stored on ice. The procedure was repeated twice, and supernatants were pooled. Water extracts were stored frozen at −20 °C for later measurement (Raessler et al. 2010).

Starch

We added 50 mg of the ground sample to 0·35 mL distilled water, vortexed this mixture for 1 min and treated it for another 10 min in the thermomixer at 65 °C (1050 rpm). We then added 0·5 mL of 33% perchloric acid and let it incubate in an orbital shaker for 20 min. The mixture was then centrifuged at 14 300 g for 6 min and the supernatant removed with a pipette. This procedure was then repeated on the remaining pellet, and the supernatant from the two extractions was pooled.

NSC concentration measurements

An aliquot of the sugar/starch extract was diluted (1:20 for soluble sugar and 1:55 for starch extracts) before measurement of sugars by high-pressure liquid chromatography–pulsed amperometric detection (HPLC – PAD), using a Dionex® ICS 3000 ion chromatography system equipped with an autosampler (Thermo Fisher GmbH, Idstein, Germany; Raessler et al. 2010).

Mobile NSC δ13C

An aliquot of the sugar extracts was pipetted into tin cups and assayed with a Finnigan MAT DeltaPlus XL EA-IRMS (ThermoFinnigan GmbH, Bremen, Germany), coupled to an autosampler (Koppenaal, Tschaplinski & Colombo 1991). Measurement error of δ13C analyses was <0·1‰ based on measurements of an internal laboratory standard (NBS 22: –30·03‰ on VPDB scale) (Coplen et al. 2006). However, residuals of perchloric acid in the hydrolysed starch extracts prohibited their analysis by mass spectrometry, so we have no data for stable isotopic signatures of this fraction.

Statistical Analysis

Where appropriate, we compared treatment means for each time step with an anova (response c. treatment) followed by Tukey's honest significance test (α < 0·05) to detect significant differences between the treatment levels. We used the Levene test to check for heteroscedasticity across groups (Kozlowski & Pallardy 2002) and report significant test results of pairwise comparisons only when variances were homoscedastic. The temporal trend in NSC concentrations from beginning vs. end of the experiment was statistically assessed with repeated measures anova, after checking for sphericity with Mauchly's test. A significant (P < 0·05) treatment*date interaction indicated that observed differences between groups developed over the duration of the experiment. All analyses were carried out with r (v. 2.13.0, R Foundation for Statistical Computing 2011).

Results

Treatment and Environmental Conditions

For the duration of the experiment, light availability, ambient temperature and water vapour pressure deficit were quite variable. The cumulative daily sum of PAR photon flux density was, on average, 29·4 mol m−2 day−1 during the experimental period and varied between 4·6 and 52·7 mol m−2 day−1. Average daily maximum temperature was 21·7 °C and varied between 13·1 and 33·1 °C, while average daily vapour pressure deficit was 5·8 hPa and varied between 1·0 and 14·0 hPa (Fig. 1a).

Figure 1.

Climate forcing (a) and relative extractable soil water (b) during the experimental period. Shown in (a) are the cumulative daily sum of PAR photon flux density (PAR, mol m−2 day−1, line only), the daily maximum temperature (TEMP, °C, line with dots) and water vapour pressure deficit (VPD, hPa, line with crosses). Relative extractable soil water (b, REW, see text) for the three treatments. Arrows indicate dates when cycle trees were given water. Note that REW never regained field capacity following rewetting.

The drought treatment started on 1 July (last day of watering for cycle and drought trees) and caused rapid declines in REW. REW of the control treatment remained at c. 80% (=field capacity), while REW in the drought treatment declined to <3% after c. 3·5 weeks. Cycle trees were given water four times during the experiment. However, between rewetting events, REW declined to values similar to those of complete drought (Fig. 1b).

Treatment Effects on Sapflow and Stem Diameter Variation

Sapflow was strongly affected by both drought conditions. Shortly after the beginning of the experiment, sapflow declined in both cycle and drought trees and was near-zero after c. 2 weeks. Sapflow rates never went down to true zero, which may be due to overestimations of low sapflow rates with thermal dissipation probes (Burgess et al. 2001). Rewetting events in cycle trees caused a rapid recovery of sapflow but showed an increasing delay with each rewetting cycle (Fig. 2a). Stem diameter increased in control trees by c. 50%, while cycle trees maintained their initial stem diameter during the experiment. Complete drought caused stem shrinkage of c. 33%, indicating a severe decline in phloem and xylem hydration (Fig. 2b). Daily stem diameter variation was highest in cycle trees, peaking during rewetting events but also during periods of declining REW and sapflow (Fig. 2c). Both stem diameter and stem diameter variation showed significant temporal trends (repeated measures anova, P < 0·05), and stem diameter in both drought treatments was significantly smaller than in control trees (anova, followed by Tukey's HSD, P < 0·05) from c. 1 week after the start of the experiment onwards. Midday leaf water potential under nondrought conditions ranged between −0·8 and −2·2 MPa but decreased to less than −3·5 MPa when soil moisture was at a minimum and even decreased to less than −4·5 MPa on hot and sunny days (data not shown).

Figure 2.

Smoothed daytime sapflow index (a), diameter index (b) and diurnal diameter index variation (c) of the three treatment groups (control – grey line, cycle – black line, drought – black line and round symbols). The sapflow index is proportional to xylem flux velocity, and the diameter index represents the cumulative diameter variation from initial diameter at the beginning of the experiment. Diurnal diameter index variation is the daily variation in stem diameter and is therefore indicative for stem capacitance. Arrows indicate dates when cycle trees were given water. Note the apparent negative sapflow indices during periods of drought stress for both drought and cycle tree.

Leaf Gas Exchange

Carbon assimilation ceased rapidly after the onset of drought in both drought and cycle trees. Average daily carbon assimilation rates approached zero after <2 weeks in these trees (Fig. 3a). Rewetting events caused instantaneous increases in carbon assimilation of cycle trees lasting c. 2 weeks, while carbon assimilation in drought trees ceased completely after c. 6 weeks. Early morning assimilation rates were similar between all treatment levels and remained positive until c. August 10 (Fig. 3b). Furthermore, during mild, cloudy and humid days, midday (12–14 h) assimilation rates remained remarkably high in drought trees even after c. 4 weeks of drought stress (e.g. July 30, Fig. 3c).

Figure 3.

Daily (a), early morning (6–8 h, b) and midday (12–14 h, c) averages (±1SEM) of specific carbon assimilation rates (g m−2 h−1) of the three treatment groups (control – grey line, cycle – black line, drought – black line). Filled symbols in cycle and drought trees (triangles and circles, respectively) indicate dates where differences are significant from control trees (anova and Tukey's HSD, α < 0·05). Arrows indicate dates when cycle trees were given water, and the asterisks (*) indicate when trees in the complete drought treatment apparently died.

Root Respiration and δ13C of Root-Respired CO213Croot)

Root respiration in control trees was driven largely by temperature (correlation coefficient 0·46, Fig. 4a). In cycle and drought trees, respiration rates were less controlled by temperature (correlation coefficient 0·15 and 0·16, respectively). Respiration rates in cycle trees increased after each rewetting event and also on mild, cloudy and humid days in drought trees. These were also the dates when carbon assimilation rates increased (Fig. 3). δ13Croot increased in both cycle and drought treatments beginning just after the start of the experiment and continuing until tree death (drought trees) or the end of the experiment (cycle trees, Fig. 4b). In cycle trees, δ13Croot decreased temporarily by c. 5‰ after the second and third watering event indicating downward transport of fresh assimilates. In contrast, δ13Croot in drought trees remained enriched but became more variable from August 25 onwards (Fig. 4b).

Figure 4.

Smoothed hourly averages in root respiration rates (a, g h−1) and root δ13C (b, adjusted for mixing ratio in air from root chamber, in ‰, see eqn 5) of the three treatments (control – grey line, cycle – black line, drought – black line). Symbols on lines in cycle and drought trees (triangles and circles, respectively) indicate where differences are significant from control trees (anova and Tukey's HSD on raw unsmoothed data, α < 0·05). Arrows indicate dates when cycle trees were given water, and the asterisk (*) indicates when trees in the complete drought treatment apparently died.

Whole-Tree Net Carbon Exchange

To make comparison easier, we report the differences in whole-tree net carbon exchange for cycle and drought conditions relative to control trees. Positive deviations (smaller whole-tree C loss compared to control trees) occurred mainly during night-time in drought trees, indicating that respiration was reduced by drought (Fig. 5a). However, cycle trees showed increases in night-time respiration after watering and overall lost more carbon at night than control trees (Fig. 5a).

Figure 5.

Average cumulative carbon loss (g, ±1 propagated SEM) for cycle (grey line) and drought (black line) relative to control trees during night-time (upper panel, 18 h00–6 h00) and daytime (lower panel, 6 h00–18 h00). Arrows indicate dates when cycle trees were given water, and the dotted line indicates zero carbon loss. Note that each group represents only two individual WTC. WTC, whole-tree chambers.

During the daytime, however, carbon exchange rates were strongly and negatively affected by complete drought and to a lesser degree in cycle trees (Fig. 5b). Overall, the negative daytime deviation in net carbon exchange rates more than compensated for smaller losses during the night in drought trees, and this led to a substantial overall carbon loss. Rewetting events in cycle trees dampened the negative effect but only for c. 1 week (Fig. 5b). Root development was also strongly affected by drought. Control trees had well-developed root systems at the end of the experiment, while root systems of both cycle and drought trees were smaller and shallower (Table S1, Fig. S4, Supporting Information).

Nonstructural Carbohydrate Pools and Mobile NSC δ13C

Carbon pools in the above-ground tissues were not affected by drought. NSC concentrations in droughted needles and branches were similar to those of control trees (Tables 1 and 2). In cycle trees, this was also the case for sucrose. However, glucose and fructose concentrations in branches and needles were higher in cycle trees than in control trees at the end of the experiment (Tables 1 and 2). Root NSC concentrations, especially sucrose, were reduced in the drought treatment (Table 3). However, starch was almost completely depleted in roots of droughted trees while showing large increases in cycle and control trees over the course of the experiment (Table 3). Total NSC concentrations in needles were not affected by drought treatments (repeated measures anova, P > 0·05, Fig. 6a) but significantly increased in branches of cycle trees (anova, followed by Tukey's HSD, P < 0·05, Fig. 6b) and decreased in roots of trees in the complete drought treatment (Fig. 6c).

Table 1. Needle nonstructural carbohydrate (NSC) concentrations (in mg g−1 of dry biomass) for glucose (glc), sucrose (suc), fructose (fruc) and starch during the beginning (early July) and towards the end (mid-September) of the experiment
Date (dd/mm)NSCTreatment mean (SE)Comparison (P)
ControlCycleDroughtCT-CYCCT-DRCYC-DR
  1. Pairwise comparison between treatments (Tukey's HSD following anova) were considered significant when < 0·05. For easy reading, P-values > 0·05 were replaced with a period (‘.’).

01/07Gluc11·49 (1·82)14·69 (0·36)15·49 (3·72)...
Fruc8·71 (1·76)11·82 (0·65)14·01 (5·03)...
Sucr22·86 (7·64)15·82 (7·57)11·15 (3·77)...
Starch10·32 (3·44)7·15 (3·41)5·04 (1·7)...
05/07Gluc19·73 (2·47)16·76 (3·15)21·76 (1·93)...
Fruc12·84 (2·1)13·12 (3·1)16·76 (2·8)...
Sucr25·91 (4·34)21·99 (5·05)14·89 (5·85)...
Starch11·7 (1·96)9·93 (2·28)6·73 (2·64)...
08/07Gluc18·97 (0·73)13·71 (2·61)23·03 (1·29)..0·011
Fruc15·61 (1·93)13·68 (2·63)18·56 (1·97)...
Sucr14·24 (3·58)9·16 (2·46)7·75 (5·28)...
Starch6·44 (1·62)4·15 (1·11)3·52 (2·38)...
09/09Gluc18·37 (1·37)28·04 (2·15)25·3 (1·87)0·012..
Fruc11·66 (1·54)23·57 (4·27)19·56 (2·42)...
Sucr19·19 (5·63)11·98 (6·02)18·13 (4·83)...
Starch8·68 (2·54)5·44 (2·71)8·2 (2·17)...
13/09Gluc21·11 (1·12)29·16 (2·26)23·33 (0·86)0·012..
Fruc13·87 (1·79)24·63 (2·78)17·35 (0·95)0·01..
Sucr15·09 (4·32)5·19 (0·85)18·02 (5·76)...
Starch6·83 (1·95)2·37 (0·38)8·16 (2·59)...
16/09Gluc21·47 (1·86)30·88 (3·44)22·17 (1·73)...
Fruc15·01 (3·19)24·7 (2·82)15·53 (2·06)...
Sucr18·02 (8·00)3·13 (0·77)19·86 (6·53)...
Starch8·16 (3·60)1·45 (0·35)8·98 (2·94)...
Table 2. Branch nonstructural carbohydrate (NSC) concentrations (in mg g−1 of dry biomass) for glucose (gluc), sucrose (sucr), fructose (fruc) and starch during the beginning (early July) and towards the end (mid-September) of the experiment
Date (dd/mm)NSCTreatment mean (SE)Comparison (P)
ControlCycleDroughtCT-CYCCT-DRCYC-DR
  1. Pairwise comparison between treatments (Tukey's HSD following anova) were considered significant when < 0·05. For easy reading, P-values > 0·05 were replaced with a period (‘.’).

01/07Gluc11·283 (2·12)16·094 (4·87)10·41 (2·79)...
Fruc10·149 (2·65)16·686 (6·04)10·86 (2·65)...
Sucr28·329 (3·99)23·006 (6·78)24·67 (5·24)...
Starch12·775 (1·80)12·606 (4·31)10·55 (3·87)...
05/07Gluc14·673 (1·46)10·535 (1·69)11·21 (1·00)...
Fruc16·134 (1·34)9·26 (1·37)11·76 (2·04)0·036..
Sucr30·219 (3·65)26·902 (2·55)22·55 (3·80)...
Starch13·617 (1·65)12·132 (1·15)9·84 (2·35)...
08/07Gluc16·257 (1·50)12·963 (1·52)16·29 (0·71)...
Fruc15·741 (1·13)13·64 (2·45)17·39 (2·99)...
Sucr28·122 (1·47)27·016 (0·75)34·75 (3·32)...
Starch12·668 (0·94)11·888 (0·20)14·5 (0·12)...
09/09Gluc10·442 (0·87)18·23 (1·43)9·73 (0·51)0·001.0·001
Fruc6·096 (1·40)17·745 (2·35)2·29 (1·27)0·003.<0·001
Sucr28·517 (2·31)31·21 (1·96)25·26 (3·05)...
Starch12·864 (1·04)14·086 (0·88)11·4 (1·38)...
13/09Gluc11·925 (0·56)21·144 (0·73)11·19 (1·05)<0·001.<0·001
Fruc8·481 (0·72)21·794 (0·96)5·77 (1·21)<0·001.<0·001
Sucr26·046 (2·97)27·443 (1·20)18·92 (3·28)...
Starch11·76 (1·34)11·96 (0·46)8·54 (1·48)...
16/09Gluc11·65 (0·90)20·10 (2·33)11·69 (1·67)0·018.0·019
Fruc7·28 (1·29)17·20 (2·29)4·83 (1·18)0·006.0·001
Sucr25·65 (4·11)23·76 (1·39)23·81 (1·34)...
Starch11·58 (1·85)10·73 (0·63)10·75 (0·61)...
Table 3. Root nonstructural carbohydrate (NSC) concentrations (in mg g−1 of dry biomass) for glucose (gluc), sucrose (sucr), fructose (fruc) and starch before (17/06/2011) and at the end (28/09/2011) of the experiment
Date (dd/mm)NSCTreatment mean (SE)Comparison (P)
ControlCycleDroughtCT-CYCCT-DRCYC-DR
  1. Pairwise comparison between treatments (Tukey's HSD following anova) was considered significant when < 0·05. For easy reading, P-values > 0·05 were replaced with a period (‘.’).

17/06Glc10·98 (2·01)15·79 (1·62)9·75 (4·90)...
Suc29·81 (3·20)22·99 (3·32)26·06 (3·20)...
Fruc9·02 (0·93)12·97 (1·82)8·49 (3·00)...
Starch13·47 (1·44)9·42 (0·93)11·77 (1·45)...
28/09Glc8·58 (1·63)10·72 (1·76)5·25 (1·40)...
Suc38·57 (3·97)31·58 (2·23)5·41 (2·30).<0·001<0·001
Fruc10·22 (2·07)9·82 (2·23)4·15 (1·90)...
Starch17·40 (1·79)14·95 (0·73)2·44 (1·06).<0·001<0·001
Figure 6.

Total (glucose + fructose + sucrose + starch) concentrations (mg g−1 dry biomass) in needles (a), branches (b) and roots (c) of the three treatments (control – squares, cycle – triangles, drought – circles). Repeated measures anova showed a significant (P < 0·01) temporal trend in total NSC concentrations of branches and roots (not shown). Filled symbols indicate dates where differences are significant from control trees (anova and Tukey's HSD, α < 0·05). Note that actual sampling dates of root tissues (06/17, 09/28) are reported here for simplicity as 07/01 and 09/16, respectively. NSC, nonstructural carbohydrate.

Mobile NSC δ13C of needles and branches showed a clear treatment effect (repeated measures anova, P > 0·001). Over the growing season, the initial δ13C values in needles and branches (c. −25·5‰ and −26·5‰, respectively) decreased in control trees to c. −29‰ and −27·5‰, respectively, while they increased slightly in needles of both drought treatments (c. −26·5‰) and somewhat more in branches (c. −25·5‰ in cycle trees, c. −26‰ in drought trees, Table 4). Mobile NSC δ13C remained constant in roots of cycle trees and increased by c. 0·7‰ in drought trees, while it decreased in control trees by almost 2‰ (Table 4). However, there was no overall significant treatment effect over the run of the experiment (repeated measures anova, P > 0·05).

Table 4. Mobile nonstructural carbohydrate (NSC) δ13C (in ‰) in different functional units (FU) at the beginning (June–July) and at the end (September) of the experiment
FUDate (dd/mm)Treatment mean (SE)Contrasts
ControlCycleDroughtCT-CYCCT-DRCYC-DR
  1. Pairwise comparison between treatments (Tukey's HSD following anova) were considered significant when < 0·05. For easy reading, P-values > 0·05 were replaced with a period (‘.’).

Needle01/07−27·39 (0·17)−27·08 (0·19)−27·1 (0·25)...
05/07−27·5 (0·15)−27·2 (0·71)−27·5 (0·4)...
08/07−27·39 (0·34)−26·77 (0·45)−27·12 (0·45)...
09/09−28·62 (0·15)−26·53 (0·68)−26·59 (0·29)0·0020·001.
13/09−28·78 (0·37)−26·23 (0·3)−26·52 (0·22)0·0010·001.
16/09−28·94 (0·2)−26·24 (0·55)−26·46 (0·17)<0·001<0·001.
Branch01/07−26·41 (0·47)−26·52 (0·36)−25·86 (0·21)...
05/07−26·32 (0·39)−25·57 (0·56)−26·31 (0·58)...
08/07−26·65 (0·55)−26·03 (0·86)−26·47 (0·19)...
09/09−27·55 (0·14)−25·47 (0·59)−25·99 (0·29)0·0010·003.
13/09−27·42 (0·26)−25·41 (0·23)−25·99 (0·21)<0·0010·003.
16/09−27·76 (0·34)−25·2 (0·56)−25·99 (0·21)0·0010·006.
Root17/06−25·94 (0·57)−24·81 (0·39)−25·6 (0·4)...
28/09−27·88 (0·33)−24·87 (0·54)−26·34 (1·04)0·012..

Discussion

Drought Effects on Tree Hydraulic Functioning

We found no indication for irreversible xylem cavitation in cycle trees although their sapflow declined before rewetting events as much as in drought trees (Fig. 2a). Recovery of sapflow after rewetting indicated that trees seemingly refilled cavitated vessels (Sperry et al. 1994; Hacke & Sauter 1996) although we observed an increasing delay it took to recover. Xylem refilling under tension, as occurred in our rewetted cycle trees, is an energy-driven process (Zwieniecki & Holbrook 2009) and imposes an additional burden on the carbon budget, potentially triggering a negative feedback on xylem functioning (McDowell et al. 2011). However, cycle trees showed high soluble sugar concentrations in branch tissue at the time of the last rewetting event (Table 2), which makes it very unlikely that the increasing delay in sapflow recovery was driven by carbon limitation, but instead could result from a water deficit impeding the signalling for xylem refilling (Brodersen et al. 2010).

We observed strong diameter variation especially in cycle trees during periods of severe water deficit (i.e. before rewetting events) and explain this observation with decreased phloem hydration and the use of stored stem water (capacitance) (Offenthaler, Hietz & Richter 2001; Meinzer et al. 2009). In our experiment, this phenomenon may have contributed to the avoidance of runaway embolism and catastrophic (complete and irreversible) xylem cavitation in cycle trees by buffering fluctuations in xylem pressure. However, because a sustained use of capacitance requires the xylem to be recharged (Meinzer et al. 2008), its functionality as a mechanism of drought tolerance or even as a survival strategy was dysfunctional for complete drought trees where no soil water was available for xylem refilling.

Conclusions about the absence of a functional hydraulic failure (i.e. complete and irreversible xylem cavitation) in drought trees cannot be drawn from sapflow and diameter variation dynamics in cycle trees. Moreover, even the absence of a functional hydraulic failure would not exclude the possibility that droughted trees suffered from an effective hydraulic failure, that is, permanently discontinued water transport to the canopy. Both the lack of soil water (Fig. 1b) and near-zero sapflow rates (Fig. 2c) indicate a complete loss of water transport to the canopy and hence the possibility of an effective hydraulic failure in above-ground tissues.

Drought Effects on Tree Carbon Metabolism

Carbon starvation may occur when availability of free (i.e. not bound to metabolic needs like respiration) mobile NSC fails to meet the amount needed to maintain osmotic potential and cell turgor (McDowell 2011). The exact threshold where NSC availability falls below demand and causes death is not defined, but is expected to result from a simultaneous decrease in carbon acquisition and maintained carbon expenditure (McDowell 2011). As expected, carbon assimilation decreased in the droughted trees quite rapidly after the onset of drought. While drought trees reduced respiration and took advantage of favourable environmental conditions (early mornings, mild cloudy days, Fig. 3b,c) for optimizing carbon assimilation in relation to water supply (Chaves et al. 2002), there was a net carbon loss compared to control trees overall (Fig. 5). The fact that cycle trees did not show any increase in diameter during the growing season may be interpreted as an indication for either strong carbon limitation (no carbon left for stem growth) or a shift in carbon allocation from structural growth to maintenance respiration (Maunoury-Danger et al. 2010).

The increase in the isotopic signature of root respiration could reflect two different processes: a decrease in the discrimination during CO2 fixation in drought and cycle trees and/or a shift to the increased use of NSC that are more enriched in 13C as substrates for respiration. We expect decreased fractionation during photosynthesis because of reduced stomatal conductance and ensuing declines in intercellular [CO2] (Farquhar & Sharkey 1982). During periods of active phloem transport, the increase in δ13Croot (isotopic ratio of root-respired CO2) would then indicate the use of fresh 13C-enriched assimilates for respiration (Duranceau et al. 1999). However, sapflow and stem diameter strongly decreased in drought trees from mid-August onward and in cycle trees before the third rewetting event, which suggests that stem water potential was very low during these periods (Offenthaler, Hietz & Richter 2001). Given the link between xylem water potential and phloem functioning (Hölttä, Mencuccini & Nikinmaa 2009) and considering that c. 90% of the diurnal variation in stem diameter (minus radial growth) can be attributed to changes in phloem water deficit (Zweifel, Zimmermann & Newbery 2005), it is likely that phloem functioning was severely impeded in drought trees from mid-August onward and maybe this similarly applied to cycle trees when stem shrinkage was greater than c. 10% (Fig. 2b). A lack of phloem transport would impede carbon translocation, and thus, the source of 13C-enriched respiration from roots would not reflect changes in the signature of C fixed in leaves. This idea is further supported by an observed decoupling of photosynthetic fractionation from the isotopic signal of respiration and mobile NSC in roots. Increases observed in the δ13C of mobile NSC in needles and branches (indicating less discrimination) were not mirrored in root mobile NSC of drought trees (Table 4), and when cycle trees were watered, δ13Croot decreased suddenly and substantially (Fig. 4b) indicating the use of fresh assimilates and hence a relief of drought-induced limitation of carbon transport from above- to below-ground.

Starch compounds are more enriched in 13C than free sugars (Badeck et al. 2005), so the increase in δ13Croot during drought, combined with the probable lack of transport of new photosynthetic products through the phloem to the roots, likely indicates starch remobilization and use (Gessler et al. 2007; Maunoury-Danger et al. 2010). Very low starch and total NSC concentrations in droughted tree roots at the end of the experiment (Table 3, Fig. 6) also suggest that roots were metabolizing 13C-enriched starch compounds. Because carbon assimilation (and apparently carbon transport) decreased during drought, roots had to rely progressively more on in situ carbon reserves for respiration, and during the following weeks of increasingly severe drought stress, trees seemed to have switched to alternative respiration substrates. We observed sudden negative shifts of up to c. 10‰ in δ13Croot in droughted and cycle trees (during drought periods) compared to controls (Fig. 4b), which may indicate the use of lipids during severe stress (Tcherkez et al. 2003). Further research on the identity and availability of putative substrates and their isotopic signature is necessary to assess the dynamics of storage use under drought stress.

We did not find any indication for soluble sugar depletion in above-ground tissues but rather a general increase in branch glucose and fructose concentrations (Tables 1 and 2). Increases in fructose concentrations have been explained in olive plants as a mechanism of osmotic adjustment to increase drought tolerance (Dichio et al. 2009), and laboratory experiments confirm that reducing sugars (e.g. glucose and fructose) play an important role in cell osmotic adjustment during water stress (Handa et al. 1983). While the mechanisms as well as the compounds involved in osmotic adjustment seem to be highly species specific (Kozlowski & Pallardy 2002), a concurrent decrease in starch and sucrose and increases in glucose and fructose concentrations indicated that monosaccharides were the most important osmotica for adjustment in leaves of Malus domestica (Borkh.) (Wang & Stutte 1992) as well as in shoots and roots of Pinus banksiana (Lamb.) and Picea glauca (Moench) Voss) (Koppenaal, Tschaplinski & Colombo 1991). The fact that soluble sugar concentrations were generally higher in cycle than in control trees seems to corroborate their importance as osmotica; concentrations of both glucose and fructose increased in needles and branches; in needles, this seemed to have occurred at the expense of sucrose (Table 1). We speculate that the cycle trees may have acclimated to drought stress in earlier drying cycles and optimized carbon partitioning for osmotic adjustment. Given an average increase of c. 20 mg of glucose and fructose per gram of dry needle and branch biomass (Tables 1 and 2), a 1:3 ratio of dry/fresh weight (data not shown) and assuming a symplastic water content of c. 75% (Gross & Koch 1991), the osmotic pressure would have been increased by 0·18 MPa at full turgor in cycle trees. We are currently investigating the temporal dynamics of this water-driven carbon partitioning in more detail in a separate study.

Overall, our NSC and starch data suggest that above-ground tissues and roots respond differently to drought stress in terms of carbon allocation and balance. While above-ground tissues were not carbon depleted even when the drought trees died, roots in droughted trees showed severe declines in carbon pools (Fig. 6c), especially starch (Table 3), and likely lost biomass (or at least gained very little compared to control trees). This finding is interesting because other studies have found that severe drought increased root NSC concentrations in poplar seedlings (Galvez, Landhäusser & Tyree 2011) and in many other tissues of droughted plants (Muller et al. 2011).

Sink limitation to growth can explain high NSC concentrations in trees even in stressful environments (Körner 2003) because growth declines faster than photosynthesis and causes substrate accumulation (Muller et al. 2011). Increased glucose, fructose and total NSC concentrations occurred in above-ground tissues of cycle trees during periods of substantial source activity (Fig. 3) and concurrent near-zero growth sink activity in stems (Fig. 2b) and roots (Fig. S4, Supporting Information). Although other sink activities (e.g. maintenance respiration) are less affected by drought (Flexas et al. 2006) and heterotrophic tissues in distal parts from sources, like stems and roots, were still in need of metabolic energy, periods of reduced phloem functioning may have prevented carbon import from tissues close to source activity (i.e. needles, branches) and hence to local carbon accumulation.

Moreover, the particular conditions in our experiment (e.g. limited rooting space without access to deeper soil water pools, extreme drought) not only decreased carbon assimilation but seemingly also carbon translocation from above- to below-ground in the complete drought treatment (Ruehr et al. 2009). The drought may have been too severe for carbon storage remobilization, and this may have prevented sustained storage use in above-ground components. The trees' metabolism collapsed in these tissues seemingly before storage pools could be substantially decreased, and studies with less severe but prolonged drought should be undertaken to allow storage remobilization. On the other hand, stress periods may have been too short in cycle trees to cause strong carbon storage dependency, which may also explain, besides sink limitation, why these trees did not show symptoms of above-ground carbon depletion but instead increases in carbon pools. In our experiment, reduced phloem functioning clearly caused roots to be isolated from carbon sources and made them dependent on locally stored carbon for metabolic needs. Reduced root growth (Fig. S4, Supporting Information) seemingly did not offset the reductions in carbon supply and reductions in carbon pools ensued.

Summary and Outlook

In above-ground tissues of cycle trees, hydraulic constraints were not irreversible as indicated by the recovery of xylem transport in tree stems and of carbon assimilation in needles when they were watered. Nevertheless, hydraulic failure from a lack of available water cannot be refuted as mortality mechanism in above-ground tissues of drought trees. In the roots of these trees, the continued use and depletion of stored carbon suggest carbon starvation as a potential cause of death. Stored carbon fuelled maintenance respiration from locally available carbon pools but ultimately root respiration rates dropped to near-zero and death ensued. Decoupled isotopic signatures of respired and mobile NSC in leaves and branches vs. roots indicate that carbon translocation was also impeded but further studies will be necessary to validate its role during drought stress.

Care should be used in the extension of our results to trees growing in a natural setting. Our trees could not avoid drought by means of accessing deeper soil water (Dawson 1996) or by taking advantage of hydraulic lift from deep-rooting neighbours (Dawson 1993). Because the effects of drought on tree physiology are influenced by the specific set of environmental condition and tree characteristics (Hartmann 2011), the results obtained here may have been impractical in a study with naturally growing trees. There are many potential physiological causes for tree decline (Franklin, Shugart & Harmon 1987), and the interplay of predisposing, inciting and contributing factors makes tree mortality a complex process (Manion 1991). In the light of this, we underscore that neither of the hypothesized mortality mechanism (carbon starvation, hydraulic failure or carbon translocation failure) acts over the entire organism, whether individually or interactively. Our results suggest that physiological responses to drought stress (and hence maybe also mortality mechanisms) are not defined at the level of species or organisms (e.g. iso- vs. anisohydric species) but instead within tree compartments (needles/branches vs. roots). Different mechanisms are not only interacting (McDowell et al. 2011) but also occur concurrently in different plant compartments. These findings need to be examined in more detail with both field experiments and observational studies in natural ecosystems.

Acknowledgements

This work was supported by a research grant from the German Science Society to H.H. (DFG own position). We thank A.B., A.E., A.F., O.K., I.K., M.H., M.R., M.S., B.S., J.S., R.S. and F.V. for technical support during the implementation of the experiment and for sample processing. A special thanks to T.M. for helpful comments and to three anonymous reviewers for their insightful suggestions. We declare to have no conflict of interest whether financial or otherwise.

Ancillary