Thirst beats hunger – declining hydration during drought prevents carbon starvation in Norway spruce saplings



  • Drought-induced tree mortality results from an interaction of several mechanisms. Plant water and carbon relations are interdependent and assessments of their individual contributions are difficult. Because drought always affects both plant hydration and carbon assimilation, it is challenging to disentangle their concomitant effects on carbon balance and carbon translocation. Here, we report results of a manipulation experiment specifically designed to separate drought effects on carbon and water relations from those on carbon translocation.
  • In a glasshouse experiment, we manipulated the carbon balance of Norway spruce saplings exposed to either drought or carbon starvation (CO2 withdrawal), or both treatments, and compared the dynamics of carbon exchange, allocation and storage in different tissues.
  • Drought killed trees much faster than did carbon starvation. Storage C pools were not depleted at death for droughted trees as they were for starved, well-watered trees. Hence drought has a significant detrimental effect on a plant's ability to utilize stored carbon.
  • Unless they can be transported to where they are needed, sufficient carbon reserves alone will not assure survival of a drought except under specific conditions, such as moderate drought, or in species that maintain plant water relations required for carbon re-mobilization.


Increasing occurrences of drought- and heat-induced tree and forest mortality have incited scientific discussions about how exactly drought kills trees (McDowell & Sevanto, 2010; Sala et al., 2010). Drought affects both tree hydraulics and carbon balance because trees – like all vascular plants – respond to decreasing soil water availability and/or declines in leaf turgor with reductions in stomatal conductance, thereby reducing carbon assimilation rates (Brodribb & McAdam, 2011). In species with a conservative water-use strategy (i.e. isohydric or strong stomatal control), the reduction in carbon assimilation associated with stomatal closure occurs early during drought and may force trees into a negative carbon balance as conditions persist (McDowell et al., 2008). Stored carbon may then be mobilized to meet metabolic needs – until reserves are depleted and trees die from carbon starvation (McDowell et al., 2008). However, since prolonged droughts also cause declines in plant water potential, even in anisohydric species (with weak stomatal control) (Mitchell et al., 2013), phloem functioning will probably be negatively affected (Hölttä et al., 2009). This can in turn impede mobilization and translocation of stored carbon from source (storage) tissues to sink tissues (Sala et al., 2010) and hence may prevent trees from both using and depleting their carbon reserves (Hartmann et al., 2013).

Carbon storage pools increase in tissues of many plant species as a short-term response to drought (Muller et al., 2011), although it still needs to be elucidated whether this increase is the result of reduced carbon translocation from sources to sinks or whether it results from an active control of carbon storage pool size (Sala et al., 2012). Investigating whether storage pool control is active requires the assessment of genetic coordination of metabolic processes (Smith & Stitt, 2007) and may be very difficult to carry out in structurally and functionally complex perennial species such as trees. Similarly, in situ assessments of phloem functioning (transport) are currently limited to small trees and not practical under field conditions (Helfter et al., 2007) or can only be inferred from measurements of phloem shrinking in woody stems (Sevanto et al., 2011). Phloem transport can be investigated indirectly with tracers (e.g. 13C pulse labeling) (Ruehr et al., 2009), but the time window in which the tracer can be chased may not extend more than a few weeks (Högberg et al., 2008), not long enough for monitoring drought-induced tree mortality, especially if assimilation is curtailed during drought conditions.

A different indirect approach is to simulate drought-induced reductions in carbon assimilation (negative net carbon balance) while maintaining plant water potential required for phloem functioning. Comparing carbon use, survival time and dynamics of stored nonstructural carbohydrate (NSC) pools between simulated and real drought allows inference to be made about of the role played by carbon translocation during drought. Simulating drought-induced reductions in carbon assimilation can be achieved by reducing the energy necessary for the light reactions during photosynthesis with shading (S. Sevanto et al., unpublished) or by reducing ambient CO2 concentrations below the carbon compensation point (net daily carbon assimilation < daily respiration). Both methods force trees into a negative carbon balance and may cause carbon starvation if treatment is prolonged. However, the latter method may be advantageous, in that it manipulates directly the factor of interest, CO2 availability, without creating other nonnatural conditions (i.e. continuous darkness).

In this paper, we report high-resolution data from a drought manipulation experiment performed under near-ambient and low-CO2 conditions. For the current study, we chose Norway spruce (Picea abies) because of its sensitivity to climate change and its economic importance (Hlásny et al., 2011a,b). The distributional range of Norway spruce is expected to shift during the upcoming decades, in part because of direct impacts from climate change (drought and heat spells) as well as indirect impacts from changes in pest dynamics (Dutilleul et al., 2000). Because Norway spruce closes stomata early during drought and is relatively vulnerable to cavitation (Bréda et al., 2006), it may be considered a rather isohydric species. We continuously monitored the carbon balance of Norway spruce saplings during induced lethal drought and carbon starvation and regularly assessed carbon pool loading. We hypothesized that: hydraulic limitations will kill droughted trees quickly, regardless of CO2 concentration; limited carbon transport as a result of declining plant hydration will cause trees to die without being able to mobilize C stores; and low CO2 concentrations (below compensation point) will also kill trees, but will take longer to do so because stored C can be mobilized throughout the plant. Hence C stores at death will be highest in the drought/high-CO2 plants and roughly equal to the drought/low-CO2 plants, while they will be lowest in the watered/low-CO2 plants. Furthermore, watered/high-CO2 plants will not die, their carbon storage loading being representative for nonstressed conditions.

Material and Methods

Growth chambers

We built a glasshouse facility of 12 glass growth chambers (80 cm high × 75 cm long × 45 cm wide, c. 250 l volume), allowing control of atmospheric [CO2]. Each chamber was subdivided into above- (c. 200 l volume) and below-ground compartments (made of PVC-U, c. 50 l volume). Both above- and below-ground chambers had an open bottom, and an airtight seal was accomplished by covering the glasshouse table with closed-cell rubber foam mats of ethylene propylene diene monomer (EPDM), which acted as a gasket to create an airtight seal. Chambers were flushed in parallel at a rate of 25 l min−1 (above ground) or 5 l min−1 (below ground), with plastic tubing inserted through the glasshouse table serving as air inlet and outlet pipes. Incoming air entered at the bottom, while outgoing air left the chamber through a pipe opening at the top (Fig. 1).

Figure 1.

Schematic view of one of the 12 growth chambers showing the ventilation system and the set of sensors. Chambers dimensions are 45 cm × 75 cm × 80 cm (W × B × H). PAR, photoysnthetically active radiation; TDR, time domain reflectometer.

In each chamber four 4-yr old half-sibling Norway spruce (P. abies (L.) H Karst.) saplings (c. 75 cm high) were grown with their pots in the below-ground compartment. Trees were grouped so as to homogenize biomass across chambers as best as possible using visual assessment. The pots contained a carbon-free 2 : 1 vermiculate : sand mixture (fertilized with Manna® Wuxal Super 8-8-6 with microelements and a slow-release conifer fertilizer Substral® Osmocote 11-8-17; Wilhelm Haug GmbH & Co. KG, Düsseldorf, Germany, and Scotts Celaflor GmbH, Mainz, Germany). As there was no organic matter initially in the growth substrate to contribute to CO2 fluxes by decomposition, we interpret carbon efflux from below-ground chambers as purely tree-derived. During installation, tree canopies were guided through a c. 4-cm-wide opening which was then sealed around the trunk with a 3-cm-thick EPDM closed-cell rubber plug. Gas tightness between compartments was checked by flushing chambers with air of strongly contrasting [CO2] (i.e. 0 vs 400 ppm). An absence of deviations in measured [CO2] from the preset [CO2] indicated the absence of air exchange between chamber compartments or with ambient air.


Controlling the atmospheric [CO2] was achieved by first removing all CO2 from ambient air with a molecular sieve and then mixing CO2 from a pressurized bottle back into the CO2-free air (Schnyder, 1992; Gamnitzer et al., 2009; see Supporting Information Fig. S1). We chose 350 and 75 ppm as predefined concentrations of [CO2] for the inlet air (IN) in control and carbon starvation treatments, respectively. The latter was determined in a preliminary study by progressively reducing [CO2] until the carbon balance of the trees was negative (daily cumulative sum of assimilated carbon < daily cumulative sum of respired carbon). Other sink activities (e.g. growth, storage) were not considered in defining starvation values of [CO2]. During the experiment, when droughted trees had already died, we reduced the [CO2] of the starvation treatment to 40 ppm and later to 20 ppm to increase the starving effect. These concentrations are within the range of conditions under which carbon starvation has been observed (Gerhart & Ward, 2010).

Drought was applied as a cross-treatment yielding four treatment groups: watered and high [CO2] (W + CO2); watered and low [CO2] (W – CO2); drought and high [CO2] (D + CO2); and drought and low [CO2] (D – CO2). Watered trees were given 200 ml of water each week and droughted trees were given 50 ml wk−1. We did not apply total drought conditions, because a preliminary study showed that, without additional watering, mortality occurred after only c. 5 wk, making it difficult to detect carbon dynamics during this short period. Treatments were arranged on the glasshouse table in three replicates in a randomized design. Treatments were carried out until death occurred in droughted/starved trees. Tree death was determined by the presence of the following three indicators: zero carbon/water fluxes; complete wilting of foliage; and cambial necrosis (bark removal).

Chamber and environmental measurements

In each above-ground chamber, we measured air temperature (°C) and photon flux density (μmol m−2 s−1) of photoysnthetically active radiation (PAR) with custom-made high-stability silicon photovoltaic detectors equipped with acrylic diffusers. On one tree per chamber we measured soil temperature and installed a custom-made dendrometer (see Fig. S1). Average air temperature varied between 24 and 26°C during the day and was c. 20°C at night, while the soil temperature varied between 24 and 25°C during the day and between 20 and 22°C at night (data not shown). Owing to controlled temperature, water vapor and light conditions, vapor pressure deficit and daily photon flux integral were quite constant with overall means of 3.52 kPa (± 0.39 kPa) and 4.64 mol m−2 d−1 (± 0.72 mol m−2 d−1), respectively, throughout the experiment.

Volumetric soil water content was assessed with TDR (time domain reflectometer) sensors (CS645, 7.5 cm, three-rod probes with a TDR 100 connected to a SDMX50-series multiplexer, Campbell Scientific Inc., Logan, UT, USA) in two chambers per treatment. Volumetric soil water content at time i (SWi) was then converted to relative extractable soil water at time i (REWi) using the following equation:

display math(Eqn 1)

where SWmax and SWmin are the volumetric soil water content before the start of the experiment and the minimum volumetric soil water content after several wk of drought, respectively. All sensor readings were recorded with a Campbell CR23X (temperature, PAR, dendrometer) or a Campbell CR10X (TDR) micrologger. Although droughted trees were given small amounts of water each week (a quarter of control trees), the relative water content of the substrate declined rapidly to a minimum in late July (Fig. 2).

Figure 2.

Relative extractable soil water content of the four treatments: (a) watered and high [CO2] (W + CO2); (b) watered and low [CO2] (W – CO2); (c) drought and high [CO2] (D + CO2); and (d) drought and low [CO2] (D – CO2). Note that TDR (time domain reflectometer) readings were missing during early August and at the end of October because of a logger failure. Also, from the end of July onwards, the low soil moisture content produced noisy TDR estimates of soil water content in droughted trees (c, d) and these were replaced with the last reliable reading. TDR measurements were carried out in one pot per treatment and in one replicate only because of limited hardware availability.

Tree water status

Because of the small stature of the trees, measurements of branch hydraulic conductivity were not feasible. This would have required taking large branch sections and may have imposed an additional severe stress on the trees. We therefore collected small branch sections (branch and needles) regularly and determined their FW. We then oven-dried them at 70°C for 48 h and weighed them again. The difference between FW and DW approximates tissue water content (WC) which, when expressed as a ratio over the DW, yielded an estimate of the relative water content of tree tissues:

display math(Eqn 2)

Canopy gas exchange and root respiration

We measured [CO2] and [H2O] concurrently in above- and below-ground chambers with a Picarro 2131-i and 2101-i, respectively (Picarro Inc., Santa Clara, CA, USA). Within 1 h, all 12 below- and above-ground chambers were measured in 5 min intervals using a custom-built valve switching device controlled by a Campbell CR1000 micrologger to rotate the sample gas stream. Within each 5 min interval, another logger-controlled manifold switched from incoming (IN) air to outgoing (OUT) air and we used a core period of 30 s of stable measurements in each 2.5 min measuring interval to compute an average. Instantaneous canopy gas exchange, root respiration and canopy transpiration of the trees were then defined as:

display math(Eqn 3)

These were considered constant for the whole 1 h cycle and were converted to hourly carbon and water flux (Cj or H2Oj) at time j using the following equation:

display math(Eqn 4)

where Δ[CO2 or H2O] was the difference in [CO2] between IN and OUT at time j for a given chamber, VFR was the volumetric flow rate of air going through the above- and below-ground chambers (25 and 5 l min−1, respectively) and MW was the weighted molecular weight of carbon mol–1 CO2 or the molecular weight of water (18.02 g mol−1).

We computed the whole-chamber carbon balance on day i (Ccumi) as the differences in carbon assimilation and respiration. To do so, we summed the carbon fluxes (above- and below-ground) at hour j on a daily basis and then accumulated these over the duration of the experiment (n d):

display math(Eqn 5)

where the subscripts A and B in Cj denote above- and below-ground, respectively.

NSC concentrations

We analyzed glucose, fructose, sucrose and starch concentrations in leaf, branch and root tissue as the major physiologically important carbon storage compounds. Because repeatedly opening below-ground chambers during the experiment may have imposed a major disturbance to the tree root system, we sampled root tissues only before and at the end of the experiment. Branch and needle samples were collected once every second week during the first 2 months and then about every third week. We cut the samples with a sharp branch cutter, froze them immediately by immersion in liquid nitrogen and kept them on dry ice until they were placed in a freezer at −80°C for longer storage. For NSC extraction, frozen samples were vacuum freeze-dried for 72 h and milled with a ball mill (Retsch® MM200, Haan, Germany) to fine powder.

Water-soluble sugars

After grinding, 50 mg of the samples were added to 1 ml of distilled water. The mixture was vortexed, 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 and the procedure was repeated twice. The supernatants were pooled and stored frozen at −20°C for later measurement (Raessler et al., 2010).


The same amount of ground sample (50 mg) was added to 0.35 ml distilled water, vortexed for 1 min and treated for 10 min in a thermomixer at 65°C. For starch hydrolysis we then added 0.5 ml of 33% perchloric acid and let it incubate in an orbital shaker for 20 min. After centrifuging at 14 300 g for 6 min, the supernatant was removed with a pipette and the procedure repeated on the remaining pellet (Raessler et al., 2010). The supernatants from the two extractions were pooled and stored frozen at −20°C for later measurement.

NSC concentration measurements

Sugar and starch extracts were diluted (1 : 20 and 1 : 55, respectively) before measurement with high-pressure liquid chromatography pulsed amperometric detection (HPLC-PAD) on a Dionex ICS 3000 ion chromatography system equipped with an autosampler (Raessler et al., 2010). Starch concentrations were then computed as the differences in glucose concentration in the hydrolyzed extract minus the glucose and half of the sucrose concentration in the water-soluble sugar extract multiplied by a conversion factor of 0.9 (Sullivan, 1935).

Statistical analysis

We compared treatment means for each time step with an ANOVA (response ~ f (treatment)) followed by Tukey's honest significance test (Tukey's HSD, α < 0.05) after checking the assumption of heteroscedasticity across groups with a Levene test (Morton & Forsythe, 1974). Tukey's HSD test results are only reported when ANOVA was significant and variances were homoscedastic. The temporal trend in NSC concentrations from the beginning to the end of the experiment was statistically assessed with repeated-measures ANOVA, after checking for sphericity with Mauchly's test (Mauchly, 1940). 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).


Stem diameter did not increase in any of the treatments over the duration of the experiment (Fig. 3). However, droughted trees showed the most pronounced shrinkage (Fig. 3c,d) and CO2-deprived trees showed the highest variation between watering events (Fig. 3b,d). All droughted trees died within a 2 wk period, independent of CO2 concentrations. These trees were dead after c. 13 wk while W – CO2 trees died after only c. 20 wk and W + CO2 trees survived until the end of the experiment (data not shown). Relative water content was similar across treatments for a given tissue type at the beginning of the experiment (Table 1). Water content declined in all tissues of all treatments during the experiment, perhaps indicating seasonal tissue maturation. However, all tissues in droughted and CO2-starved trees were below 10% of relative water content at death, except for root tissues in W – CO2 trees (Table 1).

Table 1. Change in relative tissue water content (water content over DW, g g–1) of the four treatments during the experiment in Norway spruce (Picea abies)
SampleTreatmentWeeks of experiment
  1. W + CO2, watered and high [CO2]; W – CO2, watered and low [CO2]; D + CO2, drought and high [CO2]; D – CO2, drought and low [CO2].

  2. Please note that root tissues were sampled only before and at the end of the experiment. Droughted trees were dead at week 13, while W + CO2 trees died during week 20.

NeedlesW + CO21.891.961.411.04
W – CO21.902.301.220.04
D + CO22.041.880.02 
D – CO21.861.730.04 
BranchesW + CO21.262.010.640.55
W – CO21.301.460.450.09
D + CO21.621.420.03 
D – CO21.041.560.04 
RootsW + CO23.03  0.45
W – CO23.21  0.49
D + CO22.89 0.09 
D – CO23.14 0.10 
Figure 3.

Stem diameter variation of the four treatments – (a) watered and high [CO2] (W + CO2); (b) watered and low [CO2] (W – CO2); (c) drought and high [CO2] (D + CO2); and (d) drought and low [CO2] (D – CO2) – expressed as the change in diameter from the initial condition (beginning of the experiment) in young Norway spruce (Picea abies). Note the steep decline in W – CO2 at the end of October.

The carbon starvation treatment prevented carbon assimilation almost immediately (Fig. 4a). Canopy net carbon exchange approached zero in D – CO2 trees after c. 1 wk, while W – CO2 trees maintained low carbon assimilation fluxes until early September. Drought alone (D + CO2) reduced carbon assimilation progressively and prevented carbon uptake from c. 10 wk onwards. Droughted trees started wilting during the second half of September and their canopy died within 2 wk, whether CO2-deprived or not, while the canopy of trees in the W – CO2 treatment survived much longer, almost until the end of November as indicated by maintained night-time respiration rates (Fig. 4b). Root respiration was similar among the three treatments until mid-August, and then declined rapidly in droughted trees but remained high in W – CO2 (Fig. 4c). This apparent persistence of respiration in W – CO2 trees towards the end of November (when foliage had wilted, needles had fallen and night-time above-ground respiration and transpiration had ceased) could reflect a shift from autotrophic (root) to heterotrophic respiration, that is, microbes feeding on senescent root matter in these wet soils. Canopy transpiration was highest in W + CO2 trees and lowest in D + CO2 and D – CO2 trees. From mid-September onwards, both drought treatments stopped transpiring. Transpiration in W + CO2 trees responded prominently to the watering events, while such a response vanished in W – CO2 trees from early October onwards (Fig. 4d).

Figure 4.

Weekly averages of daily carbon assimilation, A (a), dark (night-time) respiration (R) in leaves (b), total R (c) and transpiration for the four treatments in young Norway spruce (Picea abies): watered and high [CO2] (W + CO2, blue line); watered and low [CO2] (W – CO2, blue dashed line); (c) drought and high [CO2] (D + CO2, red line); and (d) drought and low [CO2] (D – CO2, red dashed line). Error bars are ± 1 SE. Occurrences of significant differences between treatments are indicated by symbols. Circles show significant differences (< 0.05, Tukey's honest significance test (HSD) following significant ANOVA) between carbon concentrations within a given drought treatment, and dots indicate significant differences between drought intensities within a given CO2 treatment. The asterisk marks the date when the chambers with D + CO2 and D – CO2 were disassembled after death of the trees.

W + CO2 trees assimilated on average three times, and D + CO2 trees c. 1.6 times, more carbon than required for respiration, while W – CO2 and D – CO2 trees assimilated on average only 33 and 7%, respectively, of respiration demand (Fig. 5a). Water-use efficiency (WUE) was higher in D + CO2 than in W + CO2 trees, especially during early drought. On average, trees in high-CO2 treatments assimilated 3.0 × 10−2 or 3.3 × 10−1 mg carbon, respectively, for each g of water transpired. This increase was mainly the result of highly varying WUE values during the period of severe water stress in August. Under low-CO2 conditions, trees assimilated only 0.31 × 10−1 mg (W – CO2) or 0.12 × 10−1 mg of carbon g–1 of water transpired (D – CO2)  (Fig. 5a). Carbon dynamics during drought showed three distinct phases: carbon surplus (assimilation/respiration (A/R) > 1) with high WUE; carbon deficiency (A/R < 1) with declining WUE; and (3) carbon limitation (A/R c. 1, but very low A) and increasing WUE (Fig. 5).

Figure 5.

(a) Ratio of weekly assimilation (A) over weekly respiration (R, ± 1 SE) (a) and daily water-use efficiency (WUE, ± 1 SE) for the four treatments in young Norway spruce (Picea abies): watered and high [CO2] (W + CO2, blue line); watered and low [CO2] (W – CO2, blue dashed line); (c) drought and high [CO2] (D + CO2, red line); and (d) drought and low [CO2] (D – CO2, red dashed line). The horizontal line in the upper panel indicates where assimilation equals respiration in upper panel; arrows indicate the different phases of carbon dynamics (see text). Occurrences of significant differences between treatments are indicated by symbols. Circles show significant differences (< 0.05, Tukey's honest significance test (HSD) following significant ANOVA) between carbon concentrations within a given drought treatment, and dots indicate significant differences between drought intensities within a given CO2 treatment. The asterisk marks the date when the chambers with D + CO2 and D – CO2 were disassembled after death of the trees.

The cumulative carbon gain of W + CO2 continuously increased during the experiment and trees of D + CO2 died before their cumulative carbon balance became negative, that is, they did not respire all carbon assimilated during the experiment (Fig. 6). Both W – CO2 and D – CO2 lost carbon from the beginning of the starvation treatment until death, but the rate of loss slowed in the last weeks before death (Fig. 6).

Figure 6.

Cumulative carbon gain/loss (± 1 propagated SD) for the four treatments in young Norway spruce (Picea abies): watered and high [CO2] (W + CO2, blue line); watered and low [CO2] (W – CO2, blue dashed line); (c) drought and high [CO2] (D + CO2, red line); and (d) drought and low [CO2] (D – CO2, red dashed line). The horizontal line indicates a zero cumulative carbon balance, that is, all carbon gained during the experiment would be consumed by respiration. Note that the low-CO2 treatment caused a continuous negative carbon balance throughout the experiment. The asterisk marks the date when the chambers with D + CO2 and D – CO2 were disassembled after death of the trees.

Drought had different effects on NSC concentrations, depending on carbon availability. In general, starch and sucrose accumulated in needles and branches of the D + CO2 trees, while glucose and fructose concentrations decreased (Fig. 7). By contrast, droughted and starved trees (D – CO2) showed decreases in starch and sucrose concentrations in branches but not in needles (Fig. 7). Starved trees generally had the lowest NSC concentrations and/or the steepest decline in all tissues and the depletion was stronger, almost complete, in most tissues of W – CO2 trees. Contrary to above-ground tissues, all measured NSC compounds in roots showed a seasonal decline. However, this decline in root NSC was significant in all treatments but not in control trees (repeated-measures ANOVA, P < 0.05, data not shown) and drought combined with starvation (D – CO2) reduced NSC more than starvation only, again contrary to above-ground tissues (Fig. 7).

Figure 7.

Nonstructural carbohydrate (NSC) concentrations (from top to bottom, in mg g−1 of dry biomass ± 1 SE: starch, total mobile, sucrose, glucose, and fructose) in needles (a–e), branches (f–j) and roots (k–o) of young Norway spruce (Picea abies). Closed symbols indicate significant differences (< 0.05, Tukey's honest significance test (HSD) following significant ANOVA) between droughted and/or starved trees (watered and low [CO2] (W – CO2, blue dashed line), drought and high [CO2] (D + CO2, red line), drought and low [CO2] (D – CO2, red dashed line)) from the control (watered and high [CO2] (W + CO2, blue line)). Closed symbols for dates with only two groups (week 20 for W + CO2, W – CO2 in all tissues, week 13 for D + CO2 and D – CO2 for roots) indicate significant differences (< 0.05, Tukey's HSD following significant ANOVA) between these two groups only.


Major findings

The C-starvation treatment in our experiment forced trees into a negative carbon balance. These trees had to rely on stored carbon to survive and our results indicate that the hydration status (watered vs drought) had a strong influence on carbon storage remobilization and use, as well as on survival. Trees grown under carbon deficiency but with sufficient water supply survived longer than droughted trees. After c. 13 wk, droughted trees died, whereas carbon-deprived trees survived for c. 20 wk. NSC concentrations were generally lower in all tissues of watered, CO2-deprived trees than in droughted trees, whether starved or not, indicating that the use of stored carbon was inhibited by drought-induced declines in hydration. Hence our study is the first to successfully demonstrate that the ability to move C reserves within the tree helps it survive persistent drought.

Carbon dynamics

We observed three distinct phases of carbon dynamics during drought (Fig. 5a): carbon surplus, carbon deficit and carbon limitation. During the first period, carbon assimilation was not yet strongly affected by drought and exceeded respiration. The second period was characterized by a tradeoff between carbon gain and water loss. Here trees required more water per unit C fixed to maintain assimilation at high rates (Fig. 5b). During the last period, respiration decreased with vanishing assimilation and was seemingly limited by carbon supply. The apparent increase in A/R and WUE may be explained by very low carbon and water fluxes rather than by physiological recovery.

In contrast to observations in many different species and plant types (Adams et al., 2009, 2013), we did not find more rapid declines in respiration than in carbon assimilation during drought. The droughted trees maintained assimilation rates at control values for c. 2 wk before these declined and approached zero after c. 4 wk. Although respiration also declined to some extent, it never declined as dramatically as assimilation rates (Fig. 4b,c). The rapid decline in respiration during early drought usually results from a turgor-mediated decrease in growth rates (Muller et al., 2011), but this apparently did not occur in our trees. We started the experiment only after the longitudinal growth flush had already been accomplished and trees did not show any secondary growth (Fig. 3), probably because of the relatively low light intensities. Hence the trees in our study did not experience a sudden decline in carbon requirement from rapid turgor-mediated growth reduction, which may explain why carbon pools did not increase during early drought.

Our findings thus challenge the conceptual framework of carbon dynamics during drought, which predicts that growth respiration declines faster than assimilation, leading to an initial increase in carbon surplus and hence to an accumulation of NSC (McDowell, 2011; McDowell et al., 2011). This concept is appropriate when drought stress occurs during a period of strong sink activity (e.g. early season longitudinal growth, high secondary growth rates), but may not hold as a general model for plant response to drought. The timing of drought and the phenology of the affected species can thus strongly influence carbon dynamics during drought.

NSC dynamics

Our findings underscore the importance of translocation in carbon dynamics during drought (Sala et al., 2010) and emphasize that availability rather than mere carbon pool size determines the ability to draw on stored carbohydrates during drought (Sala et al., 2012). Droughted starved trees were not able to redistribute available carbon from sources (above-ground tissues) to sinks (roots) in order to counteract carbohydrate depletion. Analogously, Mitchell et al. (2013) showed that tree species with a conservative water-use strategy, that is, species controlling water loss with early stomatal closure, survived longer and had more strongly depleted carbohydrate pools in leaves, stems and roots (only starch declined in roots) than ‘water spenders’, which died sooner and with less reduced NSC pools. In contrast to Mitchell et al. (2013), we were able to maintain plant hydration at functional levels for carbon translocation during storage dependency within the same species. Hence our results are not affected by species-specific storage use behavior (Piper, 2011) as a potential driver of the observed differences.

It is surprising that above-ground tissues of droughted trees showed increases in both sucrose and starch concentration in the absence of an apparent drought-related decline in sink strength. Drought stress is known to enhance starch degradation and increases sucrose synthesis under conditions of strong sink activity, such as grain filling in rice (Yang et al., 2001). Although the continuous respiration demand of heterotrophic tissues, mainly roots, was a strong sink activity in the droughted trees, we did not observe an increase in starch degradation (i.e. lower starch concentration). Sucrose was seemingly synthesized as the main compound for long-distance carbohydrate transport (Holbrook & Zwieniecki, 2005), but declining hydration may have impeded its export to sinks and hence may have led to this accumulation. The cause of starch accumulation and degradation in perennial species like trees is still not well understood (Stitt & Zeeman, 2012), but the observed accumulation in above-ground tissues of droughted trees may be the result of declines in enzymatic activity for starch degradation (Sala et al., 2010).

Even more surprising is the fact that starch and sucrose concentration also increased, after an initial decline, in needles of droughted and starved trees towards the end of the experiment (Fig. 7). These trees were not assimilating any carbon during this period (Fig. 4) and the increase in starch and sucrose may have been actively allocated. NSC accumulation during drought stress is not uncommon but may result from decreased sink activity (Muller et al., 2011) and hence be a passive process. In our study, sink activity relative to carbon availability did not decrease during that period (Fig. 5b,c), making such a passive accumulation unlikely. Also surprising is the fact that starch and sucrose and not the monosaccharides accumulated in needles. Increases in glucose and fructose as means for osmoregulation (Dichio et al., 2009) have been observed 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 et al., 1991). However, starch is osmotically inactive and the conversion of sucrose into glucose/fructose would be osmotically more efficient. A recent study documented that glucose and fructose concentrations increased in droughted Norway spruce saplings at the expense of sucrose and starch (Hartmann et al., 2013), but others have also found NSC reductions in foliar tissues during drought (Adams et al., 2013). In our study, the decreasing water content in needles and branches may explain why both sucrose and starch were not depleted. Conversion of starch to glucose or of sucrose to glucose/fructose requires the presence of water for hydrolysis during invertase and amylase catalytic reactions (Stitt & Zeeman, 2012). However, this does not explain the observed increase in both compounds during advanced drought and further research is necessary to elucidate these issues.

In contrast to above-ground tissues, the concentration of all NSC species declined in roots during drought. Similar results were shown in an earlier drought study and were explained by the uncoupling of the root system from above-ground tissues, seemingly as a result of carbon translocation failure (Hartmann et al., 2013). As a consequence, roots became entirely dependent on local carbohydrates, and continued relatively good hydration in roots (compared with above-ground tissues) may have allowed them to remobilize any locally stored compounds. By contrast, in above-ground tissues exposed to dry air, tissue desiccation was probably faster than in roots, thus preventing carbohydrate mobilization and use. Our data show that water content was higher in roots than in above-ground tissues, even at tree death (Table 1), and may have been higher all along during the experiment, thereby setting more adequate conditions for carbon mobilization and use.

Causes of tree mortality

Drought-induced tree mortality is thought to be mediated by hydraulic failure, carbon starvation or impeded carbon translocation (McDowell & Sevanto, 2010; Sala et al., 2010). In this ‘hydraulic framework’ (McDowell et al., 2008), hydraulic failure is defined as irreversible catastrophic xylem cavitation interrupting water transport to the canopy, carbon starvation as the decrease of available carbon below metabolic requirements, and impeded carbon translocation as the inability to move carbon from sources (photosynthetic tissues, storage tissues) to sinks (heterotrophic tissues). These definitions have been generally accepted by the scientific community concerned with drought-induced tree mortality, although they are not very descriptive of the mechanisms involved. For example, for the trees in the W – CO2 treatment it was probably not irreversible catastrophic xylem cavitation but rather declining water content that impeded cellular functioning until tissue senescence and – maybe – tree death. Here we argue that more mechanistic definitions of drought effects on water and carbon relations in trees are needed.

Physiological drought can be defined as the discrepancy between water demand and supply causing hydraulic limitations to plant development and survival (Passioura, 1996). For individual plants, water deficit during drought imposes hydraulic limitations on cellular carbon metabolism (respiration and growth), carbon balance (carbon acquisition, storage and remobilization) and carbon translocation (phloem functioning). Severe hydraulic limitations may induce a failure of these processes. They are mutually dependent; for example, decreased carbon acquisition and transport will reduce carbon availability for maintaining cellular metabolism. Similarly, reduced cellular functioning will impede carbon acquisition, remobilization and translocation within the plant. Hence, this more mechanistic view precludes the view of three mutually exclusive tree mortality mechanisms, and instead emphasizes how these mechanisms are interdependent (McDowell et al., 2011).

While our experiment was not designed strictly to test the validity of mortality mechanisms, our results give a detailed picture of the carbon metabolism during lethal drought and, as such, allow careful inferences to be drawn on mortality mechanisms. Given the very low water content in needle and branch tissue of droughted trees, we assume that stems and branches of these trees experienced extensive cavitation during advanced drought. The fact that droughted trees, regardless of CO2 supply, died earlier than starved trees and that only the droughted high-CO2 trees maintained available NSC pools at death indicates that hydraulic failure of cellular metabolism was probably the ultimate cause of mortality in droughted trees. On the other hand, both droughted and nondroughted starved trees depleted their carbon storage reserves at similar rates during the early part of the experiment (weeks 0–7), making carbon limitation the main stress factor during this period, while water limitations became important only afterwards. Carbon limitation on phloem functioning was apparently not the cause of death, because the amount of NSC varied for the different CO2 treatments when death occurred in droughted trees. Although carbon starvation does not necessarily require carbon pools to be completely depleted (Sala et al., 2012) and minimum requirements for maintenance of hydraulic and phloem functioning may not be a fixed constant (McDowell et al., 2011), one would expect the same ‘lethal’ amounts of NSC if carbon limitation alone was the cause of mortality. Our results corroborate findings by S. Sevanto et al. (unpublished), who showed that both hydraulic failure and carbon starvation may act in parallel during drought-induced tree mortality and that the relative dominance of different mortality mechanisms ultimately determines the cause of death.

On the other hand, trees in the watered low-CO2 treatment probably died of starvation. Concentrations of all NSC were very low, almost zero, in all tissues at death, apparently causing cell metabolism to collapse. This breakdown may have occurred at the end of October, as indicated by a sudden collapse of stem diameter in these trees (Fig. 3). Throughout the whole experiment, both low-CO2 treatments showed a much greater response of stem diameter to watering events than trees in the high-CO2 treatment (Fig. 3). Reduced phloem NSC concentration may have impeded maintenance of phloem turgor with declining hydration, causing phloem to shrink (Zweifel et al., 2000; Sevanto et al., 2011). In the watered low-CO2 treatment, the depletion of NSC may have caused an osmotic collapse of phloem turgor and phloem functioning, observed elsewhere as a precursor of drought-induced tree mortality (S. Sevanto et al., unpublished). The very low NSC we observed in the watered low-CO2 treatment also indicates that minimum carbohydrate requirements for hydraulic and phloem functioning can be very small in young Norway spruce trees.

Implications for drought-induced tree mortality

This experimental study provides mechanistic insights into links between carbon and water dynamics during drought. However, these insights must be carefully extrapolated with respect to drought-induced tree mortality in natural ecosystems. For example, it has been hypothesized that isohydric species (tighter stomatal control) are potentially more vulnerable to carbon starvation than anisohydric species, especially during moderate but prolonged drought (McDowell et al., 2008). Our findings support this hypothesis by demonstrating the importance of translocation for storage use. Only trees grown with adequate hydration to maintain phloem functioning were able to strongly (almost completely) reduce carbon pools in our study. Severe water limitations had a stronger and more prominent effect on tree survival than carbon supply alone.


This work was supported by a research grant from the German Science Society to H.H. (DFG own position, HA 6400). We thank Annett Boerner, Anett Enke, Agnes Fastnacht, Olaf Kolle, Iris Kuhlmann, Michael Hielscher, Savoyane Lambert, Michael Raessler, Martin Strube, Bernd Schloeffel, Janine Schmidt, René Schwalbe and Frank Voigt for technical support during the implementation of the experiment and for sample processing.