study-species and sites
Ailanthus was introduced from China in 1784, and is now widely naturalized throughout much of the United States (Hu 1979). It produces very large numbers of wind-dispersed seeds (Hu 1979), grows quickly in high light (Bazzaz 1979; Feret 1985), and can reproduce asexually via root sprouts (Hu 1979). Although early studies considered it shade-intolerant and unable to invade closed forests, recent studies show that it can establish in intact forests when canopy gaps open (Kowarik 1995; Knapp & Canham 2000). It can reach mature size during a single period of release in a treefall gap, while most native species need several periods (Canham 1989; Knapp & Canham 2000). Although rapid growth and prolific reproduction (both sexually and vegetatively) undoubtedly contribute to its success as an invader, the arsenal of biochemical defences found in all Ailanthus tissues suggests allelopathy as another potential mechanism of invasion.
The experiment was conducted in three forest stands in northwestern Connecticut (USA), at elevations from 300 to 500 m. All sites had patchy distributions of Ailanthus within the stands. The three sites are approximately 20 km apart, one located near the village of Amesville (41°58′N, 73°27′W), the second located in the Dark Entry Forest (DEF) near the village of Kent (41°49′N, 73°23′W), and the third located in the village of New Milford (41°36′N, 73°25′W). Soils at Amesville (AM) and New Milford (NM) are Dystric Eutrochrepts derived from calcareous glacial outwash over limestone bedrocks, whereas soils at DEF are Typic Dystrochrepts derived from glacial till over mica-schist bedrocks (Hill et al. 1980). All sites are in second-growth stands (80–130 years) with a history of logging but no history of agriculture. The species composition of the stands included elements of the oak forests of southern New England and the northern hardwood forests of the northeastern United States and Canada. The dominant native tree species were A. saccharum Marsh. (sugar maple), F. americana L. (white ash), Q. rubra L. (northern red oak), Prunus serotina Ehrh. (black cherry), and Betula lenta L. (black birch). The relative basal area of Ailanthus was 31.2% in AM, 28.4% in DEF and 32.3% in NM. The mean DBH of Ailanthus trees was approximately 20 cm in the three sites (19.9 cm in AM, 21.6 cm in DEF and 24.6 cm in NM), and the maximum DBH ranged from 37.9 cm in DEF to 54.8 cm in AM.
seedling transplant experiment
In June 2005, seedlings of A. rubrum L., A. saccharum and Q. rubra were collected from the surrounding forests and planted at each of the three study sites. For each species, we selected seedlings of similar age (3–5 years) and height without evident signs of herbivory damage. Twenty planting locations were selected per site, stratified along a gradient of distance from and abundance of Ailanthus in the immediate neighbourhood (defined as a 25-m radius circle around each planting location). We identified and mapped every Ailanthus tree with a DBH ≥ 2 cm within each of the 60 neighbourhoods (n = 337 trees), using a laser rangefinder with a digital compass (Laser Technology Inc., Centennial, CO). Relative basal area of Ailanthus in the neighbourhoods varied between 0% and 70%.
At each planting location, two 20 × 60 cm plots were established, separated by a distance of 50 cm. The soil of one of the plots was dug to a depth of 20 cm, moved to a bucket where it was hand-mixed with AC (GC Powdered Activated Carbon, General Carbon Corp., Paterson, NJ) at a rate of 20 mL/L soil, and put back in the ground (AC treatment). The soil from the second plot was dug to the same depth and moved to a different bucket (to reproduce the disturbance caused to the soil in the AC treatment), and then put back into the ground without adding any chemicals (Control treatment). An additional amount of AC (0.125 L) was applied in May 2006 onto the soil surface of each AC treatment. AC is frequently used in allelopathy experiments because it acts as an adsorbant for many large organic compounds, therefore minimizing allelopathic effects while having minor impacts on nutrient dynamics (Cheremisinoff & Ellerbusch 1978). AC has been successfully used to test for allelopathic interactions in a large number of studies (see review in Hierro & Callaway 2003), and has been recommended as an effective approach for allelopathy studies in the field (Inderjit & Callaway 2003).
Three seedlings (one of each of the three native tree species) were planted in each plot (n = 360 seedlings). We measured the initial stem height, extension growth of the stem, basal diameter, number of leaves and diameter of each expanded leaf at the time of planting. In order to facilitate accurate remeasurement of stem height, a small mark was made on the stem of each seedling at the ground level. Within the initial pool of seedlings collected for transplanting, a random subsample (n = 30 per species) was taken to the laboratory at the beginning of the experiment for measurement of initial stem height, stem dry biomass and dry biomass, area and diameter of all leaves. The objective was to generate, for each species, regressions of (i) stem dry biomass as a function of stem height, (ii) leaf area as a function of leaf diameter (measured from the base to the tip of the leaves), and (iii) leaf dry biomass as a function of leaf area. These regressions (R2 > 0.9 in all cases, data not shown) allowed non-destructive estimations of initial stem biomass, initial leaf area and initial leaf biomass for each experimental seedling using field measurements (see below).
For each experimental seedling, survival, stem height, stem extension growth, number of leaves and diameter of each individual leaf were sampled twice, once at the end of the first growing season (September 2005) and once at the end of the second growing season (September 2006). Seedlings alive in September 2006 were harvested and taken to the laboratory where leaves were removed and measured for total area using a leaf area meter (LI-COR Inc., Superior St. Lincoln, NE). Roots were rinsed by hand, and the root and shoot systems were separated at ground level (using the marks on the stems). Roots, stems and leaves were dried for 48 h at 60 °C and weighed. The extension growth of the stem for the second growing season was separated from the rest of the stem and weighed separately. We considered the biomass of the extension growth as a more accurate estimation of the effects of AC on biomass allocation to stems than the total stem biomass. As some seedlings had finished their annual growth by the time the experiment started (June 2005), we decided to conduct statistical analyses using growth data only for the second growing season. Specifically, we considered six response variables: (i) survival after the two years of the experiment; (ii) extension growth in 2006; (iii) dry biomass of the 2006 extension growth; (iv) root dry biomass; (v) leaf dry biomass in 2006; and (6) leaf area in 2006. Due to the difficulties of excavating Q. rubra roots without losing a significant part of the root system, we decided not to consider root biomass as a response variable in the analyses for this species.
Because variation in light availability was expected to affect seedling growth and survival, we used fisheye photography to estimate a gap light index (GLI, Canham 1988) for each seedling plot. GLI is the percentage of ‘gap’ light (Canham 1988; i.e. photosynthetically active radiation transmitted through discrete openings in the canopy) that reaches a point in the understorey over the course of a defined growing season. Photographs were taken in the middle of each plot by placing the camera (with a fisheye lens) at approximately 30 cm over the ground. All pictures were taken on cloudy days during August 2005.
The transplant experiment was initially designed including Ailanthus as a fourth seedling species. Given the difficulty of finding natural seedlings of Ailanthus in our study sites, seeds were germinated in the glasshouse in May 2005 and Ailanthus seedlings transplanted to the field at the same time as the native species. Ailanthus seedlings were transplanted to individual 20 × 20 cm plots 30 cm away from the native plots (in order to avoid potential allelochemical interference among seedlings), using the same soil treatments used for the plots containing native tree seedlings. However, even though dead Ailanthus seedlings were replaced during the first 3 weeks of the experiment, no Ailanthus seedlings were alive at the end of the first growing season, and the species had to be excluded from the study.
seed sowing experiment
We conducted a seed sowing experiment at the same locations as the seedling transplant experiment (n = 20 locations per site). Two 30 × 30 cm quadrats were established at each location, one (AC quadrat) next to the seedling plot with AC and the second one (Control quadrat) next to the Control seedling plot. In the AC quadrat, AC was added at a rate of 20 mg/L soil to the first 5 cm of the soil. In the Control quadrat the soil was mixed by hand but no chemicals were added. In each quadrat, 10 seeds of A. rubrum and 10 seeds of Q. rubra were sown at 1 cm depth in four lines of five seeds, each line were 2 cm from each other and 5 cm from the border of the quadrat. Acer saccharum was not included in the experiment due to the unavailability of seeds during the years of the study. Seeds of A. rubrum were obtained commercially from regional seed sources (lot with 98% viability). Seeds of Q. rubra were collected in the surrounding forests, and non-viable acorns (empty or depredated by insects), identified by flotation in water, were excluded. To exclude seed predators, we built cages around each quadrat using 24-gauge, 1.5 cm mesh hardware cloth buried to a depth of 5 cm and extending 25 cm above-ground. Seeds were sown in November 2005, and emergence was monitored every 2 weeks during April–June 2006.
In September 2006, all live seedlings in the seed sowing experiment were harvested and taken to the laboratory, where they were measured using the same procedures for seedlings in the transplant experiment (see above). Response variables from the sowing experiment were: (i) emergence, estimated as the percentage of seeds with shoots growing beyond the ground surface by the time of the last emergence census (June 2006); (ii) survival, estimated as number of emerged seedlings in June that were alive at the end of the experiment in each quadrat; (iii) stem dry biomass, estimated as the mean stem biomass of all alive seedlings in each quadrat; (iv) leaf dry biomass, estimated as a mean per quadrat; and (v) leaf area, also estimated as a mean per quadrat. As for the transplant experiment, we decided not to consider root biomass as a response variable in Q. rubra.
neighbourhood analyses of seedling emergence, survival and growth
We used a neighbourhood approach to the study of allelopathy in which seed emergence, seedling survival and seedling growth were analyzed as a function of: (i) the study site; (ii) the initial size of the seedling (only in the case of transplanted seedlings); and (iii) the size, abundance and spatial distribution of Ailanthus in the neighbourhood. The models were run separately for each of the study species and response variables in each of the two experiments. For each response variable (Y), our basic allelopathy model takes the form:
- ( eqn 1)
The first term in the model, Sitej, is an estimated parameter that represents the average potential seedling performance (i.e. survival, root biomass, leaf area, etc., per unit effect of plant size) in the absence of neighbouring Ailanthus for j = 1 ... 3 study sites. The second term, Sizeλ, controls for the effects of initial plant size on seedling performance in the transplant experiment, as a function of the parameter λ, which scales the response variable to size as a power function. We used different measures of plant size depending on the response variable analyzed in the model: (i) initial stem height was used as the size estimator for survival and extension growth; (ii) initial stem biomass (calculated indirectly using regressions, see above) as the size estimator for extension biomass and root biomass; (iii) initial leaf biomass (calculated using regressions) as the size estimator for (final) leaf biomass; and (iv) initial leaf area (calculated using regressions) as the size estimator for (final) leaf area. The range of variation among individuals was relatively small because of their similar age.
The third term in the model X, captures the neighbourhood effects of Ailanthus on individual seedling performance. If Ailanthus has no effect on seedling performance then X = 1, if the effect is negative then 0 ≤ X < 1, and if the effect is positive then X > 1. We assumed that the neighbourhood effects vary monotonically as a function of an Ailanthus neighbourhood index (ANI):
- ( eqn 2)
ANIi is the Ailanthus neighbourhood index for seedling i of the target species (equation below), and ANImax is the maximum value of ANI for all seedlings of the target species. The use of ANImax standardizes the neighbourhood effects term (0 ≤ ANIi/ANImax ≤ 1) and facilitates comparisons across seedling species. To compute ANI we used a simple additive index of the abundance of Ailanthus within the immediate neighbourhood, as a function of the size and the distance to Ailanthus neighbours. Thus, for i = 1 ... n Ailanthus trees ≥ 2 cm DBH within a 25-m radius around the target seedling, the ANI is given by:
- ( eqn 3)
In order to keep the number of parameters manageable and to avoid parameter trade-offs, we allowed β to vary and be estimated by the analyses but tried alternative models setting the value of α either to α = 2 or α = 0. A value of α = 2 indicates that the influence of Ailanthus scales approximately with tree biomass (i.e. DBH2), whereas a value of α = 0 means that the influence of Ailanthus varies as a function of density, regardless of size.
We estimated a separate γ parameter in eqn 2 for each of the two treatments (i.e. for AC vs. the control). The parameter γ is an exponential decay coefficient, and defines the sign and steepness of the variation in the neighbourhood effects (X), and therefore in seedling performance (Y), due to an increment in ANI. Positive values of γ would indicate a positive effect of the presence of Ailanthus neighbours relative to the mean effects of the native neighbours, whereas negative γ values would indicate a negative effect of the presence of Ailanthus relative to the presence of native neighbours. The difference in the γ values among the AC and control treatments measures the magnitude of the allelopathic effects of Ailanthus on seedling performance.
In order to test for the possibility of any AC effects, independent of the presence of Ailanthus, we ran a modified version of the basic allelopathy model in which the average potential seedling performance in the absence of neighbouring Ailanthus at each site (i.e. term Sitej in eqn 1) was estimated separately for AC and Control seedlings. Different Sitej terms for the two groups of seedlings would indicate that AC affected seedling performance even in the absence of Ailanthus. This could reflect either the presence of other allelopatic species or some other unintended effect of the addition of AC. However, the modified basic allelopathy model was never a better fit to the data than the simpler basic allelopathy model (see Appendix S1 in Supplementary Material), indicating that effects in AC treatments were directly linked to the presence of Ailanthus. The absence of side-effects (i.e. not related to the presence of Ailanthus) of the AC is also supported by the lack of significant differences in seedling performance among treatments (AC vs. Control) when only seedlings without Ailanthus neighbours were considered in the analyses (see Appendix S2).
We also explored the effect of light on seedling performance by adding a fourth term (GLIδ) to the basic allelopathy model (eqn 1). However, as the resulting models were never a better fit to the data, this term was dropped from the analyses (results not shown for simplicity). The absence of a light effect on seedling performance was probably a consequence of the limited variation in light levels experienced by seedlings in the understorey (GLI = 3–6% in 90% of the cases).
In order to analyze whether the allelopathic effects of Ailanthus varied among sites, we tested a modified version of the basic allelopathy model in which the value of γ (eqn 2) for the AC treatment was allowed to vary as a function of the site (Site-specific allelopathy model). The value of γ for the Control treatment was not allowed to vary among sites due to limitations in the number of parameters permitted by our sample size. The basic and site-specific allelopathy models were compared to a null model in which seedling performance was predicted just as a function of the site and the seedling initial size (i.e. setting the multiplier X to 1 in eqn 1). By doing this, we assessed whether including the neighbouring effects of Ailanthus into a model significantly improved its explanatory power.
parameter estimation and comparison of alternate models
We used simulated annealing, a global optimization procedure, to determine the most likely parameters (i.e. the parameters that maximize the log-likelihood) given our observed data (Goffe et al. 1994). We used three different error structures depending on the nature of the response variables. We analyzed survival of individual transplanted seedlings using a logistic regression in which the probabilistic scientific model provided the likelihood function. For emergence and survival in the sowing experiment we assumed a binomial error structure. For growth variables we used a normal error structure with the variance as a power function of the mean. This required estimating an additional parameter to determine the scaling of the variance to the mean. Details on the likelihood functions and the software code used for the simulated annealing algorithm are provided in Appendix S3. Alternate models were compared using the Akaike Information Criterion (AICc) corrected for small sample sizes (Burnham & Anderson 2002). Models with a difference in AICc < 2 units are considered to have equivalent empirical support. When the difference in AICc between two models is > 2, the model with the lowest AICc is considered to have larger empirical support. We used asymptotic two-unit support intervals to assess the strength of evidence for individual maximum likelihood parameter estimates (Edwards 1992). These are simply the range of parameter estimates for which ‘support’ (log-likelihood) is within two units of the maximum log-likelihood, and were determined by incrementally varying parameter estimates above and below the maximum likelihood estimate until log-likelihood had dropped by two units. The R2 of the regression (1 – SSE/SST, sum of squares error (SSE); sum of squares total (SST)) of observed vs. predicted was used as a measure of goodness-of-fit. All analyses were done using software written specifically for this study using Delphi for Windows (Version 7, Borland Software Corp., Cupertino, CA).