The study was performed in the temperate deciduous forest of Compiègne, located ~60 km north of Paris, in the Picardy region, northern France (49°22′N, 2°54′E; 32–152 m elevation; 677 mm mean annual precipitation, 10°C mean annual temperature, Chabrerie et al., 2008). The dominant soils are leached brown soils (luvisols, cambisols) but, locally, podzolised soils (podzols) and calcareous soils (leptosols) also occur. The soils developed in sedimentary substrates (palaeogeneous sands, cretaceous chalk sand, and limestone) variably covered by quaternary loess or alluvial deposits (Closset-Kopp et al., 2007). The forest, with a total surface area of ca. 15,000 ha, is mostly intensively managed as even-aged high forest. The dominant tree species are European beech (Fagus sylvatica L.), oaks [Quercus robur L. and Q. petraea (Matt.) Liebl.], European hornbeam (Carpinus betulus L.) and, on sandy soils, Scots pine (Pinus sylvestris L.). The silvicultural management units (stands) are homogeneous or mixed species even-aged stands, each stand covering on average (± SE mean) 13.34 ± 0.14 ha. The prevailing silvicultural systems are clearcutting and seed-tree shelterwood, in cycles of 100–220 years with thinning once every 4–15 years (Closset-Kopp et al., 2007; Chabrerie et al., 2008). The changes in light environment associated with clearcutting and a number of severe storm events have left ample opportunity for the rapid and widespread invasion of gaps by P. serotina (Closset-Kopp et al., 2007). It was locally reported as an invasive species for the first time in the early 1970's. Since then, it has spread over the entire forest and is present in more than half of the total area of the forest (Chabrerie et al., 2008).

Using available distribution maps of P. serotina and examining published data from 64 plots of 25 × 25 m in the study area (Chabrerie et al., 2008), we stratified the forest in invaded and uninvaded stands and randomly selected 44 of the 64 existing plots for further study (see Supplementary Material for the protocol and illustrations of the respiration experiment). We established six additional plots at intermediate P. serotina invasion stage (both tree and shrub cover of P. serotina ≈30%; Chabrerie et al., 2008). Following our own field surveys of 2014, we omitted two plots due to partial missing data, thus totalizing 48 plots. Furthermore, we had to reclassify the plots as the invasion by P. serotina had noticeably progressed. We defined plots as invaded by P. serotina if the species' basal area exceeded the threshold of 0.05 m² ha⁻¹. By applying this threshold instead of using pure P. serotina presence data we avoided misclassifying plots that were only in a very early stage of invasion (very few seedlings present only). The final sampling design comprised 20 uninvaded plots and 28 invaded plots with basal areas of P. serotina ranging between 0.09 and 14.38 m² ha⁻¹. The selected plots were on average separated by 543 m (SE 42 m) from their nearest neighbor (distance between centroids) and were well-distributed throughout the forest (see Supplementary Material for a Google Earth Datafile containing the locations of the sample plots).

In each 25 × 25 m plot we recorded vegetation data and collected soil, litter and canopy leaf samples. For all live trees (woody species > 6 m tall) and shrubs (woody species between 2 and 6 m tall) we recorded species, girth at breast height (GBH; m) and the proportion (PR; %) of their vertical crown projection overlapping the plot.

We collected soil samples by pooling a number of subsamples gathered in each plot (“individual soil samples” sensu Reinhart and Rinella, 2016). A first sample comprised 10 soil cores which were 15-cm deep (without litter (L) and ectorganic OF horizon) obtained from 10 random locations within the plot; this sample was used for soil chemical analysis. A second sample comprised five analogous soil cores and was used for the respiration experiment. Similarly, a litter sample (L and OF horizons) was composed from four subsamples from random 0.25 × 0.25 m subplots. Soil and litter samples were air-dried. The air-dried soil samples for the decomposition experiment were stored at 1°C awaiting further analysis. Soil samples for chemical analysis and litter samples were oven-dried (soil: 72 h at 40°C; litter: 72 h at 60°C). The oven-dried litter samples were weighed and the available litter on the forest floor was then quantified as tons dry litter per ha. Soil and litter samples were collected in July 2014.

We used Model 55 Goose Guns (Marlin Firearms, Madison, NC) and a fully choked Select Sporting II 12M 12-gauge shotgun (Winchester Repeating Arms, Morgan, UT) with Buckshot 27 ammunition (27 × 6.2 mm lead pellets) to shoot 325 sets of leaf samples from 307 individual trees in 48 plots (see Supplementary Material for a video). We used the leaf sampling by shotgun method for a number of reasons: tree trunk climbing was not permitted in the forest; leaf sampling by shotgun is the standard method in the protocol of the French National Network for Long-term Forest Ecosystem Monitoring (RENECOFOR); and leaf sampling by shotgun was found to be much more time- and cost-efficient than any other protocol, including those that use boom trucks or throw-line launchers (see e.g., Youngentob et al., 2016). In each plot we aimed to collect leaves from three individuals per species. For broadleaved tree species, one set of 10–15 undamaged leaves was collected from whole branches or twigs shot down from the upper, sun-exposed part of the crown. For P. sylvestris two sets of needles were collected per tree: one from new shoots (year N) and one from 1-year-old shoots (year N−1). Each set of leaves was put in a labeled paper bag. Paper bags with leaf samples were put in plastic zip-loc bags per species and per plot and stored in a cooler. The leaf samples were oven-dried (48 h at 80°C). Canopy leaf samples were collected in July 2015.

The soil samples were sieved (<2 mm) to remove stones and root biomass. For each sample for the chemical analysis, we determined the average of two measurements of soil potential acidity using a NH₄Cl extraction solution and a glass electrode; soil nitrogen (wt%) using a C/N-analyzer (TruSpec CN, LECO Corporation, St Joseph, MI, USA); soil phosphorus (mg kg⁻¹) using the lactate extractable phosphate method; and concentrations of exchangeable bases Ca, K, Mg, and Na (cmol kg⁻¹) by extraction with 1M NH₄Cl. Soil organic matter content (SOM; wt%) was determined for a subsample of the soil collected for the respiration experiment by using the loss on ignition method in a muffle furnace (Hoogsteen et al., 2015).

Oven-dried leaf samples were milled prior to chemical analysis. Foliar carbon (C) (g kg⁻¹) and nitrogen (N) (g kg⁻¹) contents were determined using the Dumas method in a vario MACRO device (Elementar Analysensysteme, Hanau, Germany). Foliar phosphorus (P) (g kg⁻¹) content was determined using an inductively coupled plasma-optical emission spectrometer (ICP-OES) (Varian 725ES, Varian Inc., Palo Alto, CA, USA).

To determine soil and litter respiration, we conducted an ex-situ respiration experiment, using 96 experimental microcosms in an air-conditioned incubation room. For each of the 48 plots, we set up two soil environments by transferring 40 g air-dried soil to each of two air-tight glass jars (287 mL) fitted with two three-way valves to allow air sampling from the headspace. Soil was compacted to a bulk density of 1.5 g cm⁻³ and the water filled pore space was set to 60 wt% by adding demineralized water. The jars were incubated for 12 days at 25°C with open valves to allow free soil respiration. After this initial incubation, an amount of oven-dried, milled litter proportional to the litter available on the forest floor (tons ha⁻¹; see above) was transferred to one of each set of two jars. Jars with only soil were used to determine heterotrophic soil respiration (R_S; release of carbon from soil organic matter and from inorganic carbon sources from the soil by microorganisms). Jars with soil and litter were used to determine total respiration (R_LIT+S; litter decomposition + soil respiration). Valves were closed and jars were placed at 25°C in a dark incubation room. R_S and R_LIT+S were determined by periodically measuring the CO₂-concentration in the headspace of the jars by use of a LI-820 CO₂ infrared gas analyzer (LI-COR Biosciences, Lincoln, Nebraska USA) attached in closed circuit to each jar separately. The gas in the closed circuit stream passes through a Mg(ClO₄)₂ (Sercon, UK) absorptive water trap to remove water vapor from the air sample. Between measurements, the circuit was flushed with CO₂-free air by looping in a CO₂ trap (Carbosorb, Sercon, UK). After each measurement, jar lids were removed for 5 min to allow CO₂ concentrations to drop to background levels (see Kerré et al., 2015 for a similar protocol and see Supplementary Presentations 1 and 2 for the protocol and illustrations of the respiration experiment). A minimum of 12 and maximum of 16 measurements spread over a period of 25 days were taken for each jar. Each measurement represented the realized respiration over ~24 h.

Measured CO₂ concentrations in ppm were corrected for atmospheric background concentrations (which were measured in three blank jars per batch) and then converted to instantaneous respiration rates (g CO₂ ha⁻¹ h⁻¹) via the ideal gas law. Instantaneous respiration rates were converted to total CO₂ released (kg ha⁻¹) by integrating the instantaneous respiration rate curves over time. The slope of a linear regression fitted over the total CO₂ released curve was then used as an estimator for the average respiration rate (kg CO₂ ha⁻¹ h⁻¹) realized during the experiment in each jar. Because of expected (see e.g., Wang et al., 2016) and observed positive priming effects related to the sudden addition of biomass to the jars containing soil and litter, the first four measurements per jar were not taken into account when fitting the linear regressions.

For all 48 plots, tree, and shrub GBH data were transformed to a measure of species dominance by converting GBH to basal area per hectare (BA; m²). The resulting data matrix was split into one vector describing the BA of P. serotina and one 25 × 48 matrix for the BA of the 25 indigenous woody species recorded during the field surveys. We used detrended correspondence analysis (DCA) to reduce the latter, multivariate data matrix to two DCA axes which reflect differences in species composition and may represent underlying environmental gradients. We used DCA because the structure in the dataset was too weak for non-metric multidimensional scaling (NMS). We used the DCA axes to represent differences in indigenous litter species composition and quality (via species and their traits; Makkonen et al., 2012), with DCA1 (41.3% variance explained; eigenvalue 0.873; gradient length 2.288) representing a turnover from broadleaved species to pines and DCA2 (16.4% variance explained; eigenvalue 0.330; gradient length 2.017) a turnover from F. sylvatica to C. betulus. Although these are rather short gradients, DCA has been found to produce robust results in similar cases (Ejrnaes, 2000). Similarly, we used principal component analysis (PCA) with a varimax rotation to reduce the soil chemical data matrix (soil pH, soil P, soil Ca, K, Mg, and Na content) to two PCA axes which reflect differences in soil properties, with PCA1 (45.2% variance explained) representing a gradient of soil pH (r = 0.900, P < 0.001) and base cations (soil Ca r = 0.973, P < 0.001; soil Mg r = 0.775, P < 0.001), and PCA2 (25.1% variance explained) a soil nutrient or NPK gradient (soil N r = 0.910, P < 0.001; soil P r = 0.734, P < 0.001; soil K r = 0.824, P < 0.001; soil Mg r = 0.384, P < 0.001).

The variables that have an effect on heterotrophic respiration are related to plant species composition and climate, and include soil temperature, soil moisture, amount, and quality of litter and soil organic matter, soil pH, soil nutrients and soil disturbance (e.g., Cornwell et al., 2008; DeForest et al., 2009; Wang et al., 2016). Under our controlled microcosm conditions, there was only minimal random variance in soil temperature, moisture, and disturbance between cases. We accounted for differences in soil properties between uninvaded and invaded stands to separate invasion effects from effects of soil chemical properties (existing or caused by P. serotina), as well as effects from litter quantity and quality. To that end, we built general linear models for the response variables R_S and R_LIT+S using invasion by P. serotina (0/1) as the fixed factor, and litter mass (tons ha⁻¹), SOM content (wt%), DCA1 and DCA2 scores (measures for species composition, as proxy for litter species composition and litter quality; Cornwell et al., 2008), and PCA1 and PCA2 scores (for soil properties) as covariates. After evaluation of the initial full models, we built reduced general linear models for R_S and R_LIT+S using invasion by P. serotina (0/1) as the fixed factor, the significant (P < 0.05) covariates from the full models, and their interaction terms with the fixed P. serotina factor. The main effects of P. serotina on heterotrophic respiration rates were estimated by calculating estimated marginal means and 95% confidence intervals for R_S and R_LIT+S for stands invaded and not invaded by P. serotina based on these reduced general linear models.

To assess the effect of P. serotina invasion on canopy leaf chemical signatures, we built linear mixed models for foliar C, N, and P, and for foliar C:N and N:P ratios, for C. betulus, F. sylvatica, Quercus spp., and P. sylvestris using invasion by P. serotina (0/1) as fixed factor, and soil pH plus the relevant variables from the set SOM (wt%), soil N (wt%), and soil P (mg kg⁻¹) as covariates. We used the plot ID as random grouping variable to account for replication of tree species within plots. The main effects of P. serotina on foliar nutrient contents and nutrient ratios were estimated by calculating estimated marginal means (i.e., controlling for the covariates) with 95% confidence intervals for foliar nutrient contents and ratios in C. betulus, F. sylvatica, Quercus spp., and P. sylvestris based on these linear mixed models. We specifically tested for significance in differences (i) between indigenous species and P. serotina; and (ii) within indigenous species between invaded and uninvaded stands.

Spatial analyses (calculating areas and distances between stands) were performed in QGIS 2.2.0. All statistical analyses were performed in IBM SPSS Statistics 20.

Soils in invaded stands were characterized by significantly lower soil pH and lower concentrations of base cations (Table 1). The literature on P. serotina effects on forest soils is ambiguous. Some studies report decreases in soil acidity and increases in concentrations of soil nutrients through the perceived positive litter effect of P. serotina (Vanderhoeven et al., 2005; Dassonville et al., 2008). Other studies indicate that the high biomass production of P. serotina requires an increased uptake of soil nitrogen and base cations, resulting in increased soil acidity (Starfinger et al., 2003). Thus, depending on the relative contributions of the positive litter effect and the negative N and cation depletion effects, invasions of P. serotina may yield decreases or increases in soil acidity (and thus improve or degrade soil quality). It is important to note that P. serotina has been shown to possess a competitive advantage on poor, sandy soils, and its wide distribution in pine forests throughout Europe has been attributed to it (Lorenz et al., 2004). Thus, the higher soil acidity in invaded stands may at least partially reflect the initial high soil acidity of the stands prior to invasion, which therefore may have been easier to invade. The resulting invasion by P. serotina may have caused a positive feedback, further increasing soil acidity and decreasing soil cation concentration through rapid growth. These results underscore that it is important to adequately account for environmental variability when assessing P. serotina effects on the ecosystem.

P. serotina had higher foliar N and P concentrations than the studied indigenous species (Figure 3). The leaf carbon concentration and the C:N ratio of P. serotina were higher than those observed in its native range (Heberling et al., 2016). Such differences may stem from the ability of invasive species to adapt their resource use strategy in the invasive range (Heberling et al., 2016) and hereby gaining the ability to use the available resources more efficiently than the co-occurring native species (e.g., Baruch and Goldstein, 1999; Rothstein et al., 2004). Some evidence indeed suggests that P. serotina taps from a larger nutrient pool than the indigenous species on the same site (Dassonville et al., 2008). The resulting high foliar N and P may be responsible for the fast decomposition of P. serotina litter and such leaf properties may improve overall litter dynamics and humus formation (Lorenz et al., 2004) as well as accelerate nutrient circulation at the level of the ecosystem (Carreño-Rocabado et al., 2012; Aragón et al., 2014). In contrast, the lower foliar N content observed in the indigenous tree species (Figure 3), and in particular F. sylvatica, provides support for the hypothesis that the invasive P. serotina is more efficient in exploiting soil N resources than its indigenous competitors (Vilà and Weiner, 2004; Dassonville et al., 2008), and may effectively short-circuit the forest's nutrient cycles in response to its high N demand (Lorenz et al., 2004). This may have important consequences for the ecosystem in terms of humus quality, as the dominant species F. sylvatica and Quercus spp. are species with an intrinsic relatively poor chemical litter quality (compared to other, indigenous deciduous species such as C. betulus and Tilia cordata; Jacob et al., 2010). A decrease in their foliar N, and thus increase in foliar C:N ratio (Figure 4), further deteriorates the quality of the beech and oak litter, and this may result in ecosystem degradation through soil acidification (contributing to the above mentioned positive feedback loop) and impacts on humification. The degradation of the overstory litter quality may therefore partially offset the effects of the nutrient-rich and fast decomposing P. serotina litter (Lorenz et al., 2004; Ashton et al., 2005). This may also explain why dense, nutrient-rich shrub layers with high litter quality, including litter from P. serotina, could not mitigate soil acidification in pine and oak forests on poor sandy soils in NE Belgium (Van Nevel et al., 2014).

An important long-term effect of reduced foliar N contents in the presence of P. serotina may have consequences that exceed the ecosystem boundaries. As foliar N content is related to the leaf light-saturated rate of photosynthesis and the leaf respiration rate and thus an important driver of photosynthetic capacity (Wright et al., 2004; Hikosaka et al., 2016), reductions in foliar N content in the long-lived indigenous trees may reduce their net primary productivity, and thus carbon sequestration capacity. Because net primary production is usually higher in invaded ecosystems (Liao et al., 2007; Vilà et al., 2011), potential losses in carbon storage in indigenous tree species following invasion would likely be compensated by the fast-growing P. serotina, either in the soil or in its biomass. Carbon storage in stem wood accounts for 80% of net carbon sequestration by forests in Europe (de Vries et al., 2006), and here P. serotina may be less efficient than the indigenous species in the long term. P. serotina is relatively short-lived compared to the indigenous broadleaved species, and its wood is currently not used in forest products with long life cycles. Therefore, it is unlikely that P. serotina can serve as a long term carbon store as efficiently as the oaks and beech trees could do.

Biotic invasions alter and often reduce the functional diversity of ecosystems (Chabrerie et al., 2010) and reductions of litter types, in turn, have been linked to slower litter carbon cycling (Handa et al., 2014). In our experiment, heterotrophic respiration was significantly higher in stands invaded by P. serotina when comparing respiration from soil and litter combined (R_LIT+S), but not when comparing respiration from soil only. In an in vitro carbon mineralization experiment, Koutika et al. (2007) also found little evidence for an effect of P. serotina on carbon mineralization from soil, and therefore, our observed P. serotina effect on respiration may primarily be mediated by litter (see also DeForest et al., 2009). In annual grasslands, it has been demonstrated that rapidly decomposing litter from an exotic grass accelerated the decomposition of native litter in litter mixtures, hereby enhancing soil respiration rates and accelerating carbon cycling (Zhang et al., 2014), rather than slowing down litter carbon cycling through effects on community functional diversity. In the same grassland ecosystem, exotic forbs also increased soil respiration via their high amounts of rapidly decomposing litter (Zhang et al., 2016).

Increased respiration in the presence of litter does not imply that our observed additional respiration R₊ (calculated as R_LIT+S − R_S) can be entirely allocated to litter and thus equals R_LIT. Nutrients present in litter biomass have an effect on the soil microbial communities and their activity (Wang et al., 2016). Because of a positive priming effect of these litter nutrients and because of synergistic effects observed elsewhere in mixtures of native (Handa et al., 2014) and of native and exotic litter (Rothstein et al., 2004; Hickman et al., 2013; Zhang et al., 2014), the respiration from soil may be higher in the presence of litter (DeForest et al., 2009), and in particular in the presence of large quantities of exotic P. serotina litter. In situ measurements would increase our understanding of the effect of P. serotina on carbon turnover, as soil microclimate, roots and microbial communities also have an influence on respiration rates. But as our measured ex situ soil and litter incubation is a good indicator of SOM and litter quality, we can conclude that invasion by P. serotina in any case increases the proportion of labile litter in the forest.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.