Grazing triggers soil carbon loss by altering plant roots and their control on soil microbial community

Authors

  • Katja Klumpp,

    1. INRA, UR874 Grassland Ecosystem Research, 234, Av. du Brézet, F-63100 Clermont-Ferrand, France
    Search for more papers by this author
    • The first two authors have contributed equally to this work.

  • Sébastien Fontaine,

    Corresponding author
    1. INRA, UR874 Grassland Ecosystem Research, 234, Av. du Brézet, F-63100 Clermont-Ferrand, France
    Search for more papers by this author
    • The first two authors have contributed equally to this work.

  • Eléonore Attard,

    1. INRA, CNRS, Université de Lyon 1, Microbial Ecology, UMR 5557, 43 Bd du 11 novembre 1918, 69622 Villeurbanne, France
    Search for more papers by this author
  • Xavier Le Roux,

    1. INRA, CNRS, Université de Lyon 1, Microbial Ecology, UMR 5557, 43 Bd du 11 novembre 1918, 69622 Villeurbanne, France
    Search for more papers by this author
    • Present address: Fondation pour la recherche sur la biodiversité, 57 rue Cuvier, CP 41, 75231 Paris cedex 05, France.

  • Gerd Gleixner,

    1. Max Planck Institute for Biogeochemistry, Postbox 100164, D-07701 Jena, Germany
    Search for more papers by this author
  • Jean-Francois Soussana

    1. INRA, UR874 Grassland Ecosystem Research, 234, Av. du Brézet, F-63100 Clermont-Ferrand, France
    Search for more papers by this author

*Correspondence author. E-mail: fontaine@clermont.inra.fr

Summary

1. Depending on grazing intensity, grasslands tend towards two contrasting systems that differ in terms of species diversity and soil carbon (C) storage. To date, effects of grazing on C cycling have mainly been studied in grasslands subject to constant grazing regimes, whereas little is known for grasslands experiencing a change in grazing intensity. Analysing the transition between C-storing and C-releasing grasslands under low- and high-grazing regimes, respectively, will help to identify key plant–soil interactions for C cycling.

2. The transition was studied in a mesocosm experiment with grassland monoliths submitted to a change in grazing after 14 years of constant high and low grazing. Plant–soil interactions were analysed by following the dynamics of plant and microbial communities, roots and soil organic matter fractions over 2 years. After disturbance change, mesocosms were continuously exposed to 13C-labelled CO2, which allowed us to trace both the incorporation of new litter C produced by a modified plant community in soil and the fate of old unlabelled litter C.

3. Changing disturbance intensity led to a cascade of events. After shift to high disturbance, photosynthesis decreased followed by a decline in root biomass and a change in plant community structure 1.5 months later. Those changes led to a decrease of soil fungi, a proliferation of Gram(+) bacteria and accelerated decomposition of old particulate organic C (<6 months). At last, accelerated decomposition released plant available nitrogen and decreased soil C storage. Our results indicate that intensified grazing triggers proliferation of Gram(+) bacteria and subsequent faster decomposition by reducing roots adapted to low disturbance.

4.Synthesis. Plant communities exert control on microbial communities and decomposition through the activity of their living roots: slow-growing plants adapted to low disturbance reduce Gram(+) bacteria, decomposition of low and high quality litter, nitrogen availability and, thus, ingress of fast-growing plants. Our results indicate that grazing impacts on soil carbon storage by altering plant roots and their control on the soil microbial community and decomposition, and that these processes will foster decomposition and soil C loss in more productive and disturbed grassland systems.

Introduction

Understanding the impact of disturbances (grazing, cultivation, climate variability) on linkages between above-ground and below-ground biota is essential to predicting the consequences of land use and global change for soil C cycling. Over the last 20 years, theses questions have inspired various experiments in forests and grasslands (Pickett & White 1986; McNaughton, Banyikwa & McNaughton 1997; Bardgett & McAlistair 1999; Wardle et al. 2008). Hence, we are now able to define principles that are at least applicable to disturbances such as grazing and cultivation (Wardle et al. 2004; Bardgett et al. 2005). Depending on the intensity of disturbance, ecosystems tend towards two contrasting systems that differ in terms of functional diversity of biota, C storage and primary production. For example, intensively grazed grasslands are dominated by fast-growing plant species producing litter of high quality (low C/N ratio and lignin content) that are quickly decomposed by bacteria (Seastedt 1985; Berendse, Bobbink & Rouwenhorst 1989; McNaughton, Banyikwa & McNaughton 1997; Bardgett, Wardle & Yeates 1998). As a result, in these productive ecosystems, C storage in the soil– litter continuum is relatively low (C-releasing ecosystem) (Milchunas & Lauenroth 1993). Grasslands adapted to low grazing levels, on the other hand, are dominated by fungi and slow-growing plants species and exhibit larger C storage (C-storing ecosystem) and lower above-ground net primary productivity than productive ecosystems.

However, the influence of disturbance on biotic communities and ecosystem functioning has mainly been studied in grasslands and forests either under constant disturbance (i.e. close to steady-state) and/or along chronosequences where many factors are likely to covary (soil mineralogy, soil depth, climate) (Wardle et al. 2004; Bardgett et al. 2005; Patra et al. 2006; but see Bardgett & McAlistair 1999). Consequently, the mechanistic basis of the transition between undisturbed C-storing and disturbed C-releasing ecosystems is less known. Plant communities, litter pools, decomposer communities and soil processes are linked in feedback loops which make it difficult to determine the major ecological drivers (see Bardgett et al. 2005). For example, it has been proposed that frequent grazing favours fast-growing plant species that allocate carbon to growth rather than to secondary defences (i.e lignin) and which decrease litter quality (e.g. Wardle et al. 2004). In that case, plant species would feedback on soil communities and processes through litter quality (Seastedt 1985; Berendse, Bobbink & Rouwenhorst 1989; Bardgett et al. 2005; De Deyn & Van der Putten 2005). However, in some cases, grazing has been shown to affect soil microbial communities before plant communities, suggesting that the shift in ecosystem functioning is not necessarily caused by a change in plant litter quality (Patra et al. 2005). Finally, recent studies have shown that grasses strongly affect litter decomposition via the presence of their living roots which modify the environment of decomposers (Van der Krift et al. 2001; Van der Krift, Kuikman & Berendse 2002; Personeni & Loiseau 2004). As grazing affects root biomass and activity (Holland & Detling 1990; Guitan & Bardgett 2000; Hamilton & Frank 2001) changes in soil communities and litter decomposition may be provoked by a change in living roots rather than litter quality.

Given the lack of knowledge, the role of plant–grazer–decomposer interactions on the control of ecosystem C-N fluxes is rarely included in current geochemical models, which partly explains discrepancies between predictions and empirical results (Fontaine & Barot 2005). A better understanding of the nature of links between plant and soil communities and ecosystem functioning after a change in disturbance will contribute to build new models including functional diversity of biota. These models may provide more accurate predictions on the impact of disturbance and diversity loss on ecosystem productivity and C storage.

A mesocosm study with semi-natural grasslands showed a trade-off between above-ground productivity (i.e annual net primary productivity, ANPP) and below-ground carbon storage when these grasslands were submitted to a change in disturbance (simulated grazing) after 14 years of constant high and low grazing intensity. Both pre-experimental (field) and experimental low disturbance favoured above- and below-ground C storage (Klumpp, Soussana & Falcimagne 2007a). Moreover, this trade-off was partly assigned to community aggregated leaf (specific leaf area, leaf dry-matter content) and root and rhizome traits (specific length, tissue density), which responded significantly to (changes in) disturbance (Klumpp & Soussana 2009).

Here, by using the same mesocosm experiment, we study the mechanistic basis of the transition between undisturbed C-storing and disturbed C-releasing grassland ecosystems. Given that plant species influence litter decomposition via the presence of living roots and that disturbance modifies root biomass and activity (e.g. Holland & Detling 1990; Van der Krift et al. 2001), we hypothesize that disturbance affects soil C cycling by altering plant roots and their control on microbial community and decomposition. To study these mechanisms, we analysed in detail the dynamics of plant and microbial communities, plant roots, soil organic matter fractions and N leaching over 2 years. After a change in disturbance, mesocosms were continuously exposed to 13C-labelled CO2, allowing to trace both the incorporation of new litter C produced by a modified plant community and the fate of old unlabelled litter C. Analysing the transition between undisturbed C-storing and disturbed C-releasing ecosystems will help identify key plant–soil interactions for C cycling and geochemical models.

Materials and methods

Experiment with grassland monoliths and management

The study was carried out with grassland monoliths from a permanent semi-natural mesic pasture that had been subjected to high- and low-intensity sheep grazing without fertilization for the last 14 years (Louault et al. 2005). High-intensity grazing, and therefore high disturbance (H) prior to the experiment, meant that grassland was cut once and sheep-grazed four times per year, whereas under low-intensity grazing [pre-experimental low disturbance (L)], grassland was grazed by sheep once a year only. Grassland under high disturbance was dominated by Holcus lanatus, Lolium perenne, Agrostis capillaris, Festuca arundinacea, Taraxacum officinale and Trifolium repens, while grassland under low disturbance was dominated by Elytrigia repens, Agrostis capillaris, Arrhenatherum elatius, Festuca rubra and Vicia cracca (data not shown).

Procedures to select and extract grassland monoliths are described by Klumpp, Soussana & Falcimagne (2007a). Briefly, in June 2002, 48 monoliths (L 0.5 × W 0.5 × H 0.4 m) were sampled from the two grassland plots. Following extraction, 24 monoliths of each pre-experimental treatment were placed in eight mesocosms closed by transparent canopy enclosures (L 1.5 × W 0.5 × H 0.75 m), each containing three monoliths of the same disturbance treatment. Mesocosms were placed in natural light and outdoor temperatures. To reduce light transmission from the side, dark plastic edges of 10 to 20 cm height were placed around each monolith throughout the experiment (see Klumpp, Soussana & Falcimagne 2007a). Air humidity was adjusted to outdoor conditions. Monoliths were watered one to three times a week to target a soil water potential of c.−30 kPa.

After the transfer into mesocosms, ‘high’ grazing disturbance (H) was simulated by cutting at 5 cm height and applying artificial urine (5 g N m−2, see Klumpp, Soussana & Falcimagne 2007a) five times per year. The low-disturbance (L) monoliths were neither cut nor fertilized during the experiment. After 6 months of acclimatization (from now on referred as t0), half of the monoliths of each treatment were switched to the opposite disturbance treatment, resulting in four replicate mesocosms each with constant low disturbance (LL), constant high disturbance (HH), a shift to high disturbance (LH) and a shift to low disturbance (HL).

All 16 mesocosms were henceforth continuously exposed to 13C-labelled CO2. For δ13C labelling, outdoor air was scrubbed from H2O and CO2, which was then replaced by fossil-fuel derived CO2 depleted in 13C (δ13C −34.7 ± 0.03‰). Thereafter, the 13C-labelled air was humidified, mixed with temperature-regulated cooled outdoor air (30%) and distributed to the mesocosms. During night time, mesocosms were provided with unlabelled air. The mean CO2 concentration inside the mesocosms was held close to the outdoor CO2 concentration (425 ± 39 μmol mol−1; mean difference of 13.2 ± z9.5 mol mol−1) and had on average a δ13C signature of −21.5 (±0.27 ‰) during the two-year experiment. Air flow and CO2 concentration in each mesocosm was monitored continuously during the two-year experiment (see Klumpp, Soussana & Falcimagne 2007a for full details on the method).

Plant community structure and nitrogen nutrition index

Analyses of plant community structure comprising botanical composition and plant functional traits were carried out after monolith extraction in September 2002 and then five times during the course of the experiment (1, 6, 12, 18 and 24 months after the start of the experiment). Measured shoot and root and rhizome traits were: vegetative height, specific leaf area (SLA), leaf dry matter content (LDMC) and leaf nitrogen content (LNC), specific length (SL) and tissue density of roots and rhizomes (DENS) (for methodologies see Klumpp, Soussana & Falcimagne 2007a; Klumpp & Soussana 2009).

The Nitrogen Nutrition Index (NNI) evaluates N availability of the vegetation and was calculated according to Lemaire & Salette (1983) from nitrogen content of live biomass (in N%) and green above-ground biomass (in t DM ha−1) as: (N%/(4.8 DM−0.32).

Soil organic matter fractions

Soil was sampled once before the start of 13C labelling (t0) and then five times during the experiment (at 1.5, 6, 12, 18 and 24 months). At each soil harvest, half a monolith per mesocosm was sampled. The remaining half was sealed with a stainless steel plate and the empty space was filled with sand. Vertical soil slices (L 0.4 × l 0.06 × H 0.3 m) of the sampled monoliths were split into three horizontal layers (0–10, 10–20 and 20–30 cm). The soil layers were air-dried and free organic matter (OM) fractions were separated with water by passing the soil through a series of three brass sieves of decreasing mesh sizes (1.0, 0.2 and 0.05 mm) (wet sieving). The remaining material in each sieve was separated into the organic and mineral fraction by density flotation in water (Loiseau & Soussana 1999). Organic fractions containing roots (R), rhizomes (Rh), coarse (>1 mm) and fine (0.2 to 1 mm) particulate organic matter (POM) were oven-dried and analysed for their C content and δ13C. Hence forward, root and rhizome were added together (R + Rh).

Genetic structure of the eubacterial community

For each soil sampling date, DNA was extracted using the FastDNA® SPIN Kit for Soil (BIO 101® Systems, Qbiogene, Carlsbad, CA, USA) and bacterial community structure was characterized by Ribosomal Intergenic Spacer Analysis (ARISA) using a modified method from Ranjard et al. (2000) (for details see Attard et al. 2008). The ARISA method, which gives a molecular fingerprint of the soil community, is a rapid method to set up and was used to follow temporal changes in microbial community structure. However, this method does not give any information on the microbial community and its activity.

PLFA analyses

To better describe the microbial community and its role on soil processes, phospholipid fatty acids (PLFA) and their δ13C signature were determined before (t0) and 18 months after disturbance change from fresh soil samples (depth 0–10 cm, sieved at 2 mm). PLFA and their δ13C signature were analysed according to Zelles (1999) and Kramer & Gleixner (2006). The PLFA-methyl esters were corrected for methyl carbon (MeOH) introduced during methylation to obtain δ13C values of PLFAs. PLFA nomenclature was used as described by Frostegard & Baath (1996) and Zelles (1999). Saturated, branched-chain fatty acids (14:0iso, 15:0iso, 15:0anteiso, 16:0, 16:0n, 17:0br, 17:0n) were associated with Gram-positive bacteria. Monounsaturated and cyclopropyl-substituted PLFAs (16:0br, 16:1ω7br, cy17:0, 17:1ω8br, 18:1ω7 and cy19:0) were associated with Gram-negative bacteria (Ratledge & Wilkinson 1988) and 18:1ω9 and 18:2ω6 with saprophytic fungal lipids. C16:1ω5, 20:3 and 20:4 indicated arbuscular mycorrhizal fungi (Jabaji-Hare 1988; Olsson 1999) and the PLFAs 10Me-C16:0 were used to indicate soil actinomycetes (Kroppenstedt, Greinermai & Kornwendisch 1984; Brennan 1988). Individual PLFAs and communities (fungal, gram(+), gram(−), mycorrhizal fungi and actinomycetes) are expressed in percentage peak area.

13C terminology and analyses

The fraction of ‘old’ unlabelled C (f Cold) and ‘new’ C derived from 13C-labelling (f Cnew) in roots and rhizomes, coarse and fine POM, and individual PLFA were calculated by a mass balance equation:

image( eqn 1)
image( eqn 2)

where δ13Csample is the δ13C of the sample, δ13Ccontrol is the δ13C value before the start of labelling (t0) and δ13Cinput is the mean δ13C value of the fully labelled green leaves during the experiment (−40.4‰ ± SD 2.7). δ13Cinput was more depleted in winter than in spring (< 0.05), but did not differ between disturbance treatments (> 0.1) (see Klumpp, Soussana & Falcimagne 2007b for more details).

Model approach

Root mortality and decomposition rates of particulate organic matter (POM) were determined by a carbon flux model. The model was constrained to measured quantities of old (unlabelled) and new (labelled) carbon in root biomass and particulate organic matter in the upper 0–10 cm soil layer during the 2-year experiment (see Model flowchart, Fig. 1). The model comprises six compartments (expressed in mg C g−1 soil): new C and old C in roots (Rootnew and Rootold), coarse particulate organic matter (cPOMnew and cPOMold) and fine particulate organic matter (fPOMnew and fPOMold). The Rootnew compartment is supplied by a fraction (s) of the daily net plant photosynthesis (Anet in mg C g−1 soil) measured by gas exchange. Root turnover (mortality, m) releases particulate organic matter supplying carbon with a fraction (β) to the fPOM and a fraction (1- β) to the cPOM compartment. The cPOM and fPOM compartments have distinct decomposition rates, (kc) and (kf). The decomposition of coarse POM releases CO2 with a fraction (1-α) and supplies the fPOM compartment with a fraction (α). Therefore, the model comprises six equations allowing to calculate six parameter unknowns (s, m, β,α, kc and kf):

image( eqn 3)
image( eqn 4)
image( eqn 5)
image( eqn 6)
image( eqn 7)
image( eqn 8)
Figure 1.

 Model flowchart. New and old C in roots (Rootnew and Rootold), coarse particulate organic matter (cPOMnew and cPOMold) and fine particulate organic matter (fPOMnew and fPOMold). (s) is fraction of the daily net plant photosynthesis (Anet) going to Rootnew; (m) root turnover (mortality); (kc) and (kf) decomposition rates of cPOM and fPOM compartments; (β) and (1- β) fraction of decomposed roots going to fPOM and cPOM; (α) and (1-α) fraction of decomposed cPOM going to fPOM and released as CO2.

The model was fitted to measured data of four replicate mesocosms per treatment by using Berkeley-Madonna software. The model was run in thermal time (Tair, in degree days) by using daily temperature means in mesocosms. Parameters (s, m, β,α, kc and kf) are therefore expressed in (degree days−1). Mean residence times (MRT, in years) of carbon in roots and rhizomes, coarse POM and fine POM was calculated as 1/m, 1/kc and 1/kf divided by a mean (for the 2 years of the experiment) annual temperature sum of 4500 °C.

Data analysis

anovas were performed for particulate organic matter fractions and PLFAs to test differences between constant low disturbance (LL), shift to high disturbance (LH), constant high disturbance (HH) and shift to low disturbance (HL). Differences were tested with a Fisher-LSD post hoc test. When necessary, data was log-transformed prior to analysis to satisfy Shapiro–Wilk’s test of normality. Statistical analyses were performed with the statistica 6 package (StatSoft Inc., Tulsa, OK, USA). With the primer software (PRIMER-E Ltd., Plymouth, UK) rank similarity matrices were computed and used to construct ‘maps’ highlighting the similarity or dissimilarity of bacterial and plant community structure among samples by non-metric multidimensional scaling (MDS) (Kruskal & Wish 1978). Similarity percentages (SIMPER) were computed to quantify the percentage of dissimilarity between treatments, whereas one-way analyses of similarity (ANOSIM) were performed to test differences between treatments at each sampling by using the primersoftware package (Plymouth Routines In Multivariate Ecological Research).

Time scale

Not all measures can be conducted at the same frequency because of methodological constrains. The frequency of each analysis is reported in Appendix S1 in Supporting Information.

Results

Description of grasslands maintained under constant low and high disturbance

Grassland monoliths maintained under constant low (LL) and high (HH) disturbance had a significantly different botanical composition (Fig. 2a). The plant communities of the HH and LL treatments exhibited divergent community aggregated morphological traits (see dissimilarity Fig. 2b). The HH plant community was characterized by high specific leaf area (SLA), high leaf nitrogen content (LNC) and low mean plant height (mHeight) indicating a fast growth strategy (i.e. nutrient exploitation). The LL plant community had a low SLA, low LNC and an increased LDMC and mHeight, characterizing a slower plant growth strategy (i.e. nutrient conservation) than the HH plant community (data not shown).

Figure 2.

 Percentage of dissimilarity for plant (a) and functional traits of leaves (b) and bacterial community (c) between treatments submitted to constant disturbance (LL versus HH,•), shift to high disturbance (LH versus LL,○) and shift to low disturbance (HL versus HH,▽). Data points are means of four replicates per treatment for percentage of dissimilarity being significant (*, P ≤ 0.05) or not (NS). In all three graphs, the two constant disturbance treatments (HH versus LL,•) were significantly dissimilar (P ≤ 0.05) throughout the experiment. For plant species composition, percentage of dissimilarity was determined for species (n = 30) with a presence throughout the experiment (>1%). Dissimilarity of plant functional traits was determined by using specific leaf area, leaf dry matter content and leaf nitrogen content.

Differences in plant community structure were mirrored by differences in soil C dynamics and microbial community. The amount of coarse and fine particulate organic matter (POM) was higher at low (LL) compared to high (HH) disturbance (Fig. 3). Moreover, during the 13C labelling experiment, the amount of old (unlabelled) carbon in roots and rhizomes and in coarse and fine POM declined at a slower rate in LL than HH soils, indicating a slower microbial decomposition of POM in the LL treatment (Fig. 3 & Table 1). The bacterial community structure, characterized by the ARISA method, differed significantly between soils of high and low disturbance (Fig. 2c). This difference was also supported by PLFA data. Low- compared to high-disturbance soil showed a higher fraction of fungal (including arbuscular mycorrhizal fungi) and individual Gram(−) PLFAs, and a lower fraction of Gram(+) and actinomycetes PLFA (Table 2).

Figure 3.

 Total and old C in roots and rhizomes (R + Rh), coarse particulate organic matter (cPOM) and fine particulate organic matter (fPOM) during 13C-labelling experiment in the 0–10 cm soil layer. Four treatments were studied: constant low (LL,•) and high (HH, bsl00072) experimental disturbance, and shift to low (HL,▽) and high (LH,○) disturbance. Data are means of four replicate mesocosms. Letters indicate significant differences (P ≤ 0.05) between treatments for a given date.

Table 1.   Mean residence times (MRT, in years; Flowchart Fig. 1) of carbon in roots and rhizomes, coarse (cPOM) and fine (fPOM) particulate organic matter in pasture monoliths with constant low disturbance (LL), constant high disturbance (HH), shift to low (HL) and shift to high (LH) disturbance. Results are means of four replicates per treatment. For a given compartment, letters indicate significant differences (P ≤ 0.05) between treatments; n = 3 as one replicate was randomly left out each time. When appropriate, data was log-transformed prior to analysis to conform to the assumption of normality
TreatmentCompartmentMRT (years)
LLRoot & rhizomes2.3a
LH 1.0bc
HH 0.9b
HL 1.8ac
LLcPOM0.4a
LH 0.2b
HH 0.3b
HL 0.5a
LLfPOM0.9ac
LH 0.7b
HH 1.0c
HL 1.3c
Table 2.   Mean relative abundance of individual PLFAs and percentage (%) of new C in PLFAs of fungal, Gram(−) and Gram(+) bacteria, mycorrhizal fungi and actinomycetes in pasture monoliths with constant low disturbance (LL), shift to high disturbance (LH), constant high disturbance (HH) and shift to low disturbance (HL). For a given date, letters indicate significant difference (P-values ≤ 0.05, bold) between treatments. Data were arcsine-transformed prior to analysis to conform to the assumption of normality
Time communityPLFA t0t18 month
LLHHLLLHHHHL
Fungal18:1ω921a19b27a17b20ab20ab
18:2ω66a5a11a3b7ab3b
Total 27a24b38a20b27b23b
%new C   9a13a9a13a
Gram(−)16:1ω7 br11a9a7a7a11b8a
17:1ω84a3b5ab2a2a8b
18:1ω713a12a15a11ab12ab8b
cy 17:0br2a2a1a2b2b2b
cy 19:0 7a13b7a11b11b12b
Total 37a39a35a33a38a38a
%new C   1a2a4c3bc
Gram(+)16:0n9a11b6a12b8b10b
17:0br4a4a1a5b4b5b
 n1a2b0a2b1b2b
a 15:0 8a10b6a10bc8ac10bc
i 15:0 4a6b3a7b4ab3a
Total 26a33b17a34b26c29c
%new C   2a3b2a4c
Mycorrhizal fungi16:1ω57a2b6a6a5a7a
20:3 0.4a0b1a1a0.6a0.4a
20:4 2a1a4a5a3ab1b
Total 9a3b11a12a8a9a
%new C   1a1a2b2b
Actinomycetes16:010Me1a2b0a0.5a1ab2b
%new C   0 a0 a0 a0.1b

After 18 months of 13C labelling, only a small quantity of new C (1–3% of total) was incorporated in Gram(−) and Gram(+) bacteria PLFA irrespective of disturbance treatments (Table 2). This result indicated that those decomposers either fed on pre-experimental soil organic matter or most of them were in a dormant stage. In contrast, soil fungi depended more on the newly fixed plant C as the higher percentages of new C (9–13% of total) compared to bacteria indicated. Overall, the percentages of new C in total PLFA biomarkers were higher in HH than in LL, demonstrating a higher microbial biomass turnover under high disturbance.

Dynamics of ecosystem functioning after a disturbance change

Hours and days following the cut of above-ground biomass, plant photosynthesis rate decreased in monoliths previously adapted to low disturbance (LH). This was followed by a net emission of CO2 during daytime for several days, indicating that plant shoot respiration exceeded photosynthesis (for details see Klumpp, Soussana & Falcimagne 2007a).

1.5 month after a shift to high disturbance (LH versus LL), plant community structure changed significantly (Fig. 2a), showing a shift from high-stature plant species (i.e. Elytrigia repens, Arrhenatherum elatius, Holcus lanatus, Cirsium sp., Vicia sp.) to low-stature plant species (i.e. Agrostis capillaris, Dactylis glomerata) supporting frequent cuts (data not shown). Species replacement resulted in considerable changes of community aggregated traits (Fig. 2b) that were related to modified growth conditions (i.e. light, nutrients, regrowth) and defoliation intensity. For example, a shift from LL to LH increased specific leaf area (SLA) and leaf nitrogen content (LNC). However, a shift to low disturbance (HL versus HH) lead to non significant changes in plant species community, indicating, a shift to less frequent disturbance, modifies botanical composition only slowly (Fig. 2a).

Six months after shift to high disturbance (LH versus LL), less old C was found in roots and rhizome due to an increased mortality (and reserve mobilization) of root and rhizome tissues (Fig. 3 and Table 1). Both, increased root mortality and lower transfer of new C to roots and rhizomes resulted in a reduction of total root and rhizome mass (Fig. 3). Despite the greater input of old C into soil, caused by the increased root and rhizome mortality, the amount of old C in cPOM and fPOM remained unchanged (Fig. 3). Model results indicate that microbial decomposition of old POM was accelerated after disturbance was intensified (LH) (Table 1). The shift to reduced disturbance (HL versus HH) increased root and rhizome biomass through a reduction of root mortality (Fig. 3 and Table 1). Despite the reduced transfer of old C from roots to POM, the amount of C in POM remained constant (Fig. 3), indicating a reduced decomposition rate in this fraction (Table 1). Changes in decomposition rates of old C in the LH and HL treatments were parallel to a significant change in bacterial community revealed by the ARISA method (Fig. 2c).

Twelve months after a shift to high disturbance (LH), markedly less old C was detected in coarse and fine POM, confirming a substantial acceleration of the decomposition of pre-experimental POM (Fig. 3). Modelling shows that decomposition rates of cPOM and fPOM were accelerated by 50% and 80%, respectively, after intensified disturbances (LH), leading to a reduction of total particulate organic C (Fig. 3). In contrast, decomposition rates of cPOM and fPOM were decelerated by 70% and 30%, respectively, after reduced disturbance (HL), increasing total particulate organic C. The acceleration of POM decomposition in the LH treatment released large amounts of nutrients. Considering differences in organic nitrogen of R + Rh and POM within the 0–30 cm soil layer between the LL and LH treatment, an amount of 14.3 g N m−2 (143 kg N ha−1) was released through accelerated microbial decomposition (Table S2 in Supporting Information). Most released N was retained by grassland monoliths as loss of N through leaching was low (0.8 g N m−2 year−1). Together with N fertilization this additional N contributed to a higher nitrogen nutrition index (Lemaire & Salette 1983) in the LH (0.76 ± 0.05) compared to the LL treatment (0.56 ± 0.05). In the HL treatment, the deceleration of POM decomposition sequestrated 3.8 g N m−2 (38 kg N ha−1).

Eighteen months after management change, analyses of PLFAs allowed a better description of changes in the microbial community and their consequences on decomposition processes. Intensified disturbance (LH) decreased the fraction of fungal PLFA by a factor of 2 and increased the fraction of Gram(+) bacteria by a factor of 2 (Table 2). The percentage of new labelled C in Gram(+) bacteria PLFA was less than 5%, indicating that these microorganisms mostly decomposed old POM deposited prior to disturbance change. Reduced disturbance (HL) did not significantly change microbial communities (Table 2). Thus, as supported by analyses of microbial DNA (Fig. 1), the ‘recovery’ of a fungi-dominated soil microbial community after reduced disturbance is much slower than the proliferation of Gram(+) after intensified disturbance. Arbuscular mycorrhizal fungi (e.g. 16:1ω5, 20:3 and 20:4) and actinomycetes (16:0 10Me) were not affected by disturbance change.

Discussion

The description of grassland monoliths maintained under constant low and high grazing disturbance for the last 14 years was consistent with the commonly observed pattern (e.g. De Deyn, Cornelissen & Bardgett 2008; Bardgett et al. 2005; Wardle et al. 2004 etc). Grassland mesocosms exposed to low disturbance were characterized by high-stature plant species with a slow-growth strategy, fungi-dominated soil community and slow soil C cycling (i.e. C-storing grassland) (Fig. 2, Tables 1 and 2). In contrast, grassland mesocosms exposed to high grazing were dominated by small-stature plant species with a fast-growth strategy, bacteria-dominated soil community, higher above-ground productivity and lower soil carbon storage (i.e. C-releasing grassland) (see Klumpp, Soussana & Falcimagne 2007a). Hence, the studied mesocosms are a useful proxy to investigate the mechanistic basis of the transition between undisturbed C-storing and disturbed C-releasing ecosystems.

How does the transition occur?

Changing the frequency of disturbance led to a cascade of effects which tend to be opposed between intensified and reduced disturbance treatments. Shift to high disturbance led, within hours, to a sharp decrease in canopy photosynthesis during daytime and net emission of CO2 for several days. During this time, plant respiration exceeded plant photosynthesis, and reserve remobilization and C starvation were likely to increased root mortality (Table 1). 1.5 months after a shift to high disturbance (LH versus LL), plant and soil community structure changed significantly. Changes in plant community structure were assigned to a replacement of slow-growing high-stature plant species by fast-growing low-stature plant species, which do support frequent defoliations.

These changes induced a decrease in root biomass and an acceleration of old (pre-experimental) POM decomposition within 6 months following the shift to high disturbance (Fig. 2). The accelerated mineralization of old POM was concurrent to a proliferation of Gram(+) bacteria and a decline in the fungal and Gram(−) communities (Table 2). The fact that only a small amount of new labelled C (<5%) was found in Gram(+) bacteria indicated that these microorganisms mostly decomposed old POM. Hence, the accelerated mineralization of POM was provoked by a proliferation of Gram(+) bacteria. Changes in mineralization could not be ascribed to mineral nitrogen supply as the increase in decomposition rate (50 and 80% for coarse and fine POM, respectively, Table 1) was ten times higher than the observed effects of N addition in grasslands (Van der Krift, Kuikman & Berendse 2002; Manning et al. 2008). Moreover, it is well established that N% of about 1.7 is sufficient to cover the requirements of microbes (Waksman & Tenney 1927; Melin 1930; Parton et al. 2007). Thus, although decomposition of roots and coarse POM may require an additional source of nitrogen (N% was 1.1% and 1.4% for roots and fine POM, respectively, see Table S1 in Supporting Information), N addition was not likely to be the main driver of the accelerated decomposition of fine POM (N% was 1.6). Consequently, intensified grazing triggers proliferation of Gram(+) bacteria and subsequent faster decomposition by reducing plant roots adapted to low disturbance.

These results support our hypothesis that disturbance affects soil C cycling by altering plant roots and their control on microbial community and decomposition. At last, the accelerated POM mineralization decreased soil carbon storage (Table 1) and released plant available nitrogen, which contributed to the higher above-ground primary production of these LH monoliths (Klumpp, Soussana & Falcimagne 2007a). However, N losses through leaching were doubled (0.8 g and 0.4 g N m−2 years−1 in the LH compared to the LL treatment), providing evidence that less frequently disturbed ecosystems better retain nutrients.

Overall, changes in plant and soil community structure and related processes were slower with reduced disturbance than intensified disturbance. This difference could be explained by contrasting growth rates of plant and microbial communities in disturbed and undisturbed grasslands. After intensified disturbance, new conditions created by defoliation (high availability of light and nutrients) allow fast-growing plant species to dominate quickly, leading to a rapid change in soil processes. In contrast, after reduced disturbance, an exclusion of fast-growing plant species is slow and depends on the depletion of key resources by slow-growing plant species. Moreover, the slower transition may be exacerbated by the soil community since soil fungi commonly have a slower growth rate than bacteria (Swift, Heal & Anderson 1979). The transition between fast- and slow-growing communities therefore explains why, after cessation of disturbance, the C storage capacity recovers more slowly than this capacity is lost after intensified disturbance (Fig. 3).

Above-ground and below-ground interactions controlled through living root activity

It has been proposed that plant communities control soil microbial communities and subsequent processes through litter quality (Seastedt 1985; Berendse, Bobbink & Rouwenhorst 1989; Merrill, Stanton & Hak 1994; Bardgett, Wardle & Yeates 1998). Although plant species may greatly differ in their decomposability, these differences are unlikely to affect decomposition at ecosystem scale, as no correlation has been found between species decomposability and fast- and slow-cycling ecosystems (Van der Krift et al. 2001; Manning et al. 2008). Our results provide support that plant communities exert control on soil microbial communities and related processes through their living roots, which modify soil resources (moisture, inorganic N, labile C) and provide rhizodeposits (Personeni & Loiseau 2004; De Deyn & Van der Putten 2005). Indeed, growth of Gram(+) bacteria and decomposition of pre-experimental POM were reduced in the presence of living roots of slow-growing plant communities. Although the advantage of these reductions for slow-growing plants remain unclear, maintaining a dense root system, slow decomposition rates and, hence, low nutrient availability may be seen as a strategy of those plants to hamper the ingress of fast-growing plants in undisturbed ecosystems. Moreover, mediating soil organic matter decomposition through the activity of living roots will have a larger impact than the production of low-quality litter. Indeed, as shown here, the activity of living roots controlled litter decomposition independently of their quality (Table 1 and S2). However, further experiments, under controlled conditions, are necessary to directly assess effects of plant species (slow- versus fast-growing communities) on soil community and processes.

Implications for primary production, nutrient retention and carbon storage in grasslands

Our results indicate that grazing triggers fast carbon and nutrient cycling by altering plant roots and their control on microbial community and decomposition. Furthermore, they signify that slow and fast nutrient cycling have complementary effects on biomass production. Intensified grazing stimulates microbial decomposition, releases the previously accumulated nutrients and increases primary production (Table 1 and S2). However, despite returns of urine and faecal material by herbivores, a depletion of stored nutrient in fast-cycling (C-releasing) grasslands will occur over time, owing to lower nutrient retention (N leaching was doubled in LH compared to the LL treatment) and exportation by herbivores. Consequently, to maintain primary production without fertilization in the long-term, a return to slow-cycling (C-storing) ecosystem is required to build new organic reserves (Table 1 and S2). Indeed, slow-cycling grasslands have a greater retention of nutrients due to higher root biomass (Fig. 3 and Table S2).

In order to compensate climate change and food requirements of a growing human population, changes in land use towards more productive (but disturbed) systems are likely to become more frequent in the coming decades (FAO 2008). Our results show that more productive and disturbed grassland systems will foster POM decomposition and soil C loss. This concurs with a report that, for a range of European grassland sites, net ecosystem carbon storage declines with increasing disturbance by grazing and cutting (Soussana et al. 2007). Provided that microbial decomposition is limited to the POM compartment, soil C loss should be low given that POM represents a small fraction of total soil organic matter. However, studies have shown that nutrient loss of ecosystem favours the activity of mining microbes and accelerates the decomposition of recalcitrant soil organic matter (Carreiro et al. 2000; Fontaine et al. 2004; Waldrop, Zak & Sinsabaugh 2004; Fontaine & Barot 2005). Therefore, higher N losses of ecosystems exposed to high disturbance may lead to a long-term soil C loss. However, further work is needed to investigate the impact of ecosystem disturbance on the loss of recalcitrant soil organic carbon.

Acknowledgements

We thank Patrick Pichon (UR 874), Claire Commenaux (UMR5557), Emanuelle Personeni, and Christiane Kramer for their technical and scientific contributions. We also thank Juliette Bloor and Nicolas Gross, Vincent Maire and anonymous referees for helpful comments on the manuscript. This work was supported by a Marie Curie Individual Fellowship (EVK2-CT2002-50026) to K.K., a PhD Scholarship (INRA – Région Poitou Charentes) to E. A., the European FP5 ‘GREENGRASS’ project (EVK2-CT2001-00105) and the Project ANR BIOMOS.

Ancillary