The effect of plant density on the response of Agrostemma githago to herbicide

Authors

  • Roger W. Humphry,

    Corresponding author
    1. School of Biological Sciences, University of Liverpool, Liverpool L69 3BX, UK
      Roger W. Humphry, Epidemiology Unit, Scottish Agricultural College VSD, Drummondhill, Stratherrick Road, Inverness IV2 4JZ, UK (fax +44 1463 711103; r.humphry@ed.sac.ac.uk).
    Search for more papers by this author
  • Martin Mortimer,

    1. School of Biological Sciences, University of Liverpool, Liverpool L69 3BX, UK
    Search for more papers by this author
  • Rob H. Marrs

    1. School of Biological Sciences, University of Liverpool, Liverpool L69 3BX, UK
    Search for more papers by this author

Roger W. Humphry, Epidemiology Unit, Scottish Agricultural College VSD, Drummondhill, Stratherrick Road, Inverness IV2 4JZ, UK (fax +44 1463 711103; r.humphry@ed.sac.ac.uk).

Summary

  • 1The estimation of dose–response relationships is an integral and legally necessary part of the routine regulatory process for herbicides. For each herbicide, plants of both target and non-target species are exposed to different levels of the chemical, and effects on mortality and performance expressed as LD50 and ED50 values (respectively, the dose at which 50% of plants die and the dose at which plants show a 50% response to herbicide). Thereafter, LD50 and ED50 values may provide comparative information between herbicides and species. There is little published on the effects of plant density on dose–response relationships.
  • 2We used the herbicide 2,4-D amine and Agrostemma githago as a model system to investigate the effects of plant density on the dose–response relationship (ED50) between 2,4-D and A. githago biomass measured as fresh weight.
  • 3Plants were grown in a controlled environment room at both low and high densities (two and 64 plants per pot) for 2 weeks, sprayed with 2,4-D using a precision sprayer, and then harvested after a further 2 weeks. The ED50 values were significantly greater when the plants were grown at high density (616 g active ingredient ha−1) than those grown at low density (42 g active ingredient ha−1); a 15-fold difference.
  • 4Mathematically, it was shown that a simple multiplicative relationship exists between ED50 and dose received when plants are grown at different densities, all other conditions being equal.
  • 5To explore the underlying effects of competitive processes arising from different densities, we experimentally investigated three phases where competition could be influencing the response of plants: (i) before spraying, (ii) at the time of spraying and (iii) after spraying. We used a sequential series of experiments to determine which phase was contributing most to the observed difference between ED50 values obtained from plants grown at low and high density.
  • 6It was shown that the competition between plants after spraying was the most likely phase to have contributed to the observed difference in ED50 values between the two densities.
  • 7These results demonstrate that trials used in the pesticide regulatory process ought to test not only different doses of pesticide but also different densities of plants (both crop and weed).

Introduction

It is important in both ecotoxicology and agronomy to measure the sensitivity of a plant species to a herbicide and to understand those factors determining variation in the response. Such information is used in the regulatory process, as well as helping to determine recommended application rates under field conditions (Hance & Holly 1990) and in environmental assessment of the long-term effects of pesticides (Boutin, Freemark & Keddy 1995; Pratt et al. 1997).

The most common method used to compare the susceptibilities of a test species to a given herbicide is to produce a dose–response relationship (Seefeldt, Jensen & Fuerst 1995). Test plants, grown under controlled conditions, are exposed to increasing doses of herbicide and some measure of damage is assessed (mortality, mass reduction, colour change, height, etc.). Where applicable a response curve is fitted to these data from which secondary indicator values are derived, usually the LD50 for mortality and ED50 for other performance measures. The LD50 is defined as the dose at which there is 50% mortality of sprayed plants, and the ED50 is defined as the dose at which the plant response is halfway between the response when the dose is zero and the maximum response (i.e. the asymptotic yield as dose becomes increasingly high).

It is, of course, simplistic to use only one parameter to compare sensitivity to herbicide. However, if only one measure of susceptibility is to be used, then the LD50 or ED50 are, in general, the most appropriate because the steepest part of the curve generally lies around the 50% value and, thus the ED50 can be measured more accurately than, say, the ED90 (Seefeldt, Jensen & Fuerst 1995). In the present study, all work was concerned with performance measures (fresh weight) and thus the ED50 was used throughout.

It is well known that dose–response relationships are dependent on various factors, including the characteristics of the target species that govern interception, retention and uptake of a herbicide; the herbicide’s mode of action; the growth stage of the plant at the time of spraying (Garrod 1989; Ascard 1994); and the growing conditions, both edaphic and climatic (Garrod 1989; Fletcher, Johnson & McFarlane 1990; Skuterud et al. 1998; Sharma & Singh 2001). Mechanisms of delivery of foliar applied herbicides, including the median droplet size of the spray and the droplet size distribution spectrum (Enfält et al. 1997), also govern efficacy, as does the density of the target plants (Ascard 1994). Often in mixed communities, high volume rates, small droplet sizes and adjuvants (compounds added to herbicide to improve the droplet spectrum or impaction of droplets) are used to ensure sufficient coverage. There is, however, a lack of information about the effects of plant density on herbicide dose–response relationships. This is important if reduced rates of application are to be promoted and integrated with other forms of weed control through manipulation of crop competition. Boyle & Fairchild (1997) noted that, within plant communities, competitive processes will interact with herbicides in the sense that there may be a primary phytotoxic effect of the herbicide on the target species with a secondary, indirect, effect of the herbicide on the individual plant due to the altered competition from neighbours. At reduced rates of herbicide application, competitive processes may play a key role in determining the outcomes of herbicide application, and an understanding of the interaction is important in recommending reduced rates of application (Haas 1989; Rydahl & Thonke 1993; Christensen 1994).

In this paper we describe a series of experiments using a model system to test whether plant density affects the dose–response relationship, and, having demonstrated a relationship, to examine at what point in the course of the experimental process density has the greatest influence on this relationship.

Methods

common methods to all experiments

Agrostemma githago L. (corncockle) was used as the test species in all experiments because it germinates and grows quickly, responds to competition (Watkinson 1981), its leaf area is relatively easy to measure and plants are sensitive to 2,4-D amine. Seeds were collected from an existing experimental population raised, without exposure to herbicide, at Ness Botanic Gardens on the Wirral, Cheshire, UK (latitude 53·27N; longitude 3·05W; grid reference east 3303; north 3754; altitude 30 m). The seed stock had a germination capacity of 90%.

Seeds were sown in John Innes no. 2 potting compost in 10-cm diameter pots to establish two densities: two (occasionally three) and 64 plants per pot (denoted as ‘low’ and ‘high’, respectively). This was achieved by sowing four and 71 seeds per pot with subsequent thinning. The plants were grown in a controlled environment room at the University of Liverpool, Liverpool, UK at a temperature of 22 °C under a set of fluorescent strip-lights with a 16 : 8 h light : dark cycle and a light intensity of 100 µmol m−2 s−1.

An experimental cycle ran for 4 weeks. Plants were allowed to grow for 2 weeks after sowing, and then were sprayed. Two weeks after spraying, individual plants were harvested at random from each pot by excision just below the cotyledons and fresh weight was measured. Thus, both expanded cotyledons and first true leaves were measured. In one experiment, leaf area and the amount of spray intercepted were measured. Harvesting for each experiment was completed within a 36-h period.

The herbicide used throughout was the amine formulation of 2,4-D (490 g a.i. litre−1; MSS 2,4-D; Mirefield Sales Services Ltd, Dewsbury, UK; a.i. = active ingredient). The herbicide was applied using two passes of a Mar-Drive precision belt sprayer operated at a pressure of 2·2 × 105 Pa and working at a pass speed of 1 m s−1. The sprayer was fitted with a Lurmark 01F110 pink nozzle set at a height of 0·75 m above the plant, and delivered a spray volume of 118 l ha−1.

experiment 1: a test of the effect of plant density on sensitivity to the herbicide

The hypothesis tested in this experiment was that the ED50 of plants grown at high density was significantly different to that of plants grown at low density.

Preliminary experiments were used to estimate the approximate ED50 for 2,4-D for A. githago grown at low and high density. The results from these trials were then used to establish a range of 12 doses for use with each of the two densities so that some individual doses were positioned in the region of the anticipated ED50, enabling a more precise statistical estimate of the ED50 to be made (Table 1).

Table 1.  Application rates of 2,4-D applied to A. githago grown at low density and high density in (a) experiment 1 and (b) experiments 3 and 5
(a)
Dose (g a.i. ha−1)Low-density plantsHigh-density plants
   0
   1 
   2 
   4
   8 
   16 
   32
   64
  128
  256
  512
 1 024
 2 048 
 4 096 
 8 192 
16 384 
(b)
Dose g a.i. ha−1Log2 (dose) (g ha−1)Experiment 3Experiment 5
   0
   1 0
   4 2
   16 4
   32 5
   64 6 
  128 7
  256 8
  512 9
 1 02410
 2 04811
 8 19213
32 76815

Six replicate groups containing each dose × density combination were used. Three replicate groups were positioned at random on each of two shelves in the controlled environment (CE) room. Within each replicate group there were 72 pots. Of these pots, 12 contained plants at high density, with each pot receiving a different herbicide dose, while the remaining 60 pots contained plants at low density and comprised 12 sets of five pots, each set receiving a different dose. At harvest the fresh weights of 10 individual plants were measured. At high-density sowings, these 10 were randomly chosen from the 64 plants receiving a particular dose in one pot. At low density, 10 plants were taken from each set of five pots. Mean values of the 10 individual plant weights were used in the data analysis.

experiment 2: a test of the effect of plant density on spray receipt by plants

The hypothesis tested was whether the amount of spray received by a plant (on a per plant basis and on a per leaf area basis) differed as a consequence of planting density.

Preliminary experiments (Humphry 1999) showed that the dye, tartrazine, dissolved in the non-ionic wetter Agral (948 g l−1 alkyl phenol ethylene oxide; Zeneca, Jealott’s Hill, Berks, UK) solution could be used as a marker to indicate how much spray a plant receives. Tartrazine is a safe non-toxic dye that can be sprayed with or without Agral, and c. 95% of the tartrazine deposited on the leaves can be recovered by a washing procedure. The washing procedure involves shaking the leaves of each plant for 10 s in an aqueous solution of Agral (1 ml l−1) (Humphry 1999). The concentration of tartrazine in solution was measured using a spectrophotometer (at 428 nm). The relationship between concentration and absorbance was linear over the range of tartrazine concentrations used. The presence of 2,4-D in the tartrazine spray solution made no measurable difference to the amount of spray received by the plants when sprayed with 16 g l−1 tartrazine in 1 ml l−1 Agral solution (Humphry 1999).

Plants of A. githago were again raised at high and low density. Plants were organized into six blocks, each block containing 12 pots (six of high-density pots and six low-density pots). The 2-week-old plants were sprayed with the tartrazine solution (16 g l−1 tartrazine and 1 ml l−1 Agral) with spraying conditions as described in the common Methods. Plant leaves were cut and tartrazine receipt measured as described above. The leaves of each selected plant were also separated, carefully flattened and photocopied onto a white background and the leaf area of each plant measured using Delta T DIAS image analysis (Delta-T Devices, Burwell, UK).

experiment 3: a test of the effect of plant density on ed50 before and at the time of spraying

Three treatments were imposed to distinguish the effect of plant size on the resultant ED50 by manipulating density during the growth period. Treatments were coded by three letters, the first standing for the density in the period of growth before spraying, the second for the density at the time of spraying and the third for the density during the period after spraying. These treatments were: treatment LLL, plants grown at low density throughout the experiment; treatment HLL, plants grown at high density but thinned to low density immediately prior to spraying, to eliminate the shading effect for receipt of herbicide; treatment HHL, plants grown at high density, sprayed and then thinned immediately to eliminate the effects of competition after spraying.

These treatments, coupled with the observations from experiment 2 enabled two hypotheses to be tested. First, that population density prior to receipt of spray governs the size of the plant and in consequence the subsequent sensitivity to the herbicide, and secondly, that density at the time of spraying also influences sensitivity (as measured by ED50).

There were seven blocks. Each block contained 36 pots: three treatments × 12 doses of 2,4-D herbicide (Table 1b). Pots were randomly positioned within blocks. Blocks were arranged, sown and harvested sequentially.

experiment 4: a test of whether herbicide reduces competition

Agrostemma githago plants were grown at high density for 2 weeks. A single ‘focal’ plant was selected in each pot, marked around the stem with a strip of aluminium foil and the whole plant then covered with foil. Two treatments were then applied. Pots were either: (i) sprayed with herbicide (2,4-D) at a rate of 2000 g a.i. ha−1; or (ii) sprayed with water (the control).

The protective foil meant that the focal plant was not sprayed by either treatment but the neighbouring (‘associate’) plants were exposed to either treatment [herbicide in case (i) and water in case (ii)]. To allow for drying, foil covers were retained for 24 h and then removed.

Plants were then left to grow for 2 weeks before marked plants were harvested. There were 39 pots, 20 were allocated to the treatment with herbicide and 19 to the control.

This experiment tested the hypothesis that the competitive effects of neighbours on a focal plant were affected by herbicide receipt by the neighbours.

experiment 5: a test of whether competition after spraying has an effect on dose–response to herbicide

The hypothesis tested was whether plants grown at high density throughout the experiment have a different ED50 to plants grown at high density up to and including the time of spraying but thinned to low density thereafter.

Plants were grown in one of two treatments (coding nomenclature as for experiment 3): HHL, where plants were grown at high density, sprayed at high density and, immediately after being sprayed, the plants thinned to low density; HHH, where plants were grown at high density throughout. Thirteen doses of 2,4-D were applied (Table 1b). A randomized complete block design was used with seven replicates. Blocks were arranged, sown and harvested sequentially as described above. During the course of experiment 5 there was equipment failure resulting in less successful control of temperature than in the other experiments.

statistical analysis

Treatment differences in each experiment were tested by analysis of variance, and dose–response analysis (Seefeldt, Jensen & Fuerst 1995) used to estimate ED50 values. SAS statistical software was used in all analyses (SAS 1990). In experiments 1, 3 and 5 there was more than one plant measured per pot. To avoid pseudoreplication, mean values were calculated for each dose × treatment × block combination and the means were used in the ensuing analysis of variance and dose–response analysis.

To describe adequately dose–responses, the fit of both three- and four-parameter log-logistic models were considered. In all cases employing a log transformation (natural log) improved the homogeneity of error variance. In the four-parameter model:

image( eqn 1)

where y is the yield per plant, C is the yield at very high doses, K + C is the yield of unsprayed plants, x is dose, ED50 is the dose at which there is a 50% reduction in yield and b is a slope parameter.

In the three-parameter model C is dropped, the value being fixed at zero. Inspection of the data in all experiments indicated that there was insufficient reason to attempt estimation of parameter C. The data from experiments 3 and 5 showed that, at upper dose ranges, plant biomass was minimal, whereas in experiment 1 the measured plant response did not tend sufficiently towards an asymptotic value.

To compare dose–response curves between two treatments, residual mean square analysis was followed (Seefeldt, Jensen & Fuerst 1995). This method for comparing curves involves testing model parameters in turn, allowing values to be unconstrained between treatments (model I) or constrained to a common value for both (model II). An F-test (equation 2) is then employed to test whether there is a significant contribution to residual mean squares. Where there is not, then parsimony dictates the use of common parameter values. The order of successive choice of parameters for constraint was governed by magnitude of effect.

image( eqn 2)

where F signifies an F-ratio; DFe the error degrees of freedom; SSe the error sums of squares; I, model I in which the parameter is independent (i.e. the parameter is free to be different for each treatment); and II, model II in which the parameter is not independent (i.e. the parameter is forced to be the same for both treatments).

Results

experiment 1: a test of the effect of plant density on sensitivity to the herbicide

Analysis of variance of the response to density across common dosages (log-transformed fresh weight) showed that there were significant effects of plant density (F1,73= 479·5; P < 0·001) and herbicide dose (F7,73 = 47·9; P < 0·001) on plant yield and an interaction between dose × density (F7,73 = 9·01; P < 0·001). No significant effect of shelves within the controlled environment room was detected, but groups within shelves contributed a significant source of variation (F4,73 = 5·2; P < 0·001; Table 2).

Table 2. anova of log mean fresh weight of A. githago in response to herbicide dose rate at two planting densities. The data set was restricted to the eight doses common to both planting densities. The F-ratio denominator for the shelf term was groups within shelf. Two degrees of freedom were lost due to missing values
Treatmentd.f.MSF-ratioP Ho
  1. NS, not significant.

Shelf 10·002 < 0·1NS
Groups within shelf 40·057 5·20·0001
Density 15·37479·50·0001
Dose 70·537 47·90·0001
Density × dose 70·101 9·010·0001
Error730·011  

Dose–response analysis showed that, despite the high level of within-treatment variation, the estimated ED50 values were significantly different between the two densities of plants (Table 3; F1,134 = 91·4; P < 0·001). The ED50 for the plants grown at low density was 42 g a.i. ha−1, while at high density it was 616 g a.i. ha−1, a 15-fold difference (Fig. 1).

Table 3.  Estimates of parameters from fitting a dose–response curve to the mean fresh weight data for the plants grown at two different densities (experiment 1)
ParameterEstimateLower 95% confidence limitUpper 95% confidence limit
Plants grown at high density
K (g) 0·212 0·170  0·254
ED50 (g ha−1)616 01382
b 0·391 0·277  0·506
Plants grown at low density
K (g) 1·18 0·989  1·37
ED50 (g ha−1) 42·011·7 72·3
b 0·618 0·494  0·741
Figure 1.

Dose–response relationship of A. githago plants grown at two different densities (low = 2 plants per pot; high = 64 plants per pot) subjected to a range of doses of the herbicide 2,4-D (experiment 1). An improvement of fit was obtained by unconstraining (i) K (F1,135 = 267), (ii) ED50 (F1,134 = 91·4) and (iii) b (F1,133 = 6·13). The models based on this fit are presented.

It was argued that the experimental cycle could be broken into three periods where density could have exerted an effect. These were (i) the period between sowing and spraying; (ii) the time of spraying itself; and (iii) the period between spraying and harvesting. Experiments 2–5 were designed to test the relative importance of these three periods for the effect of plant density on dose–response.

experiment 2: a test of the effect of plant density on spray receipt by plants

The leaf area of individual plants grown at high density was much smaller than plants grown at low density (Fig. 2). Moreover, it was observed that the leaves of the plants grown at high density were generally less horizontal than the leaves of plants grown at low density. Unsurprisingly, there was a positive relationship between the amount of tartrazine received and leaf area (Fig. 2a). The amount of spray received per plant was 0·189 µg tartrazine (95% confidence limits = 0·175–0·202 µg) for plants grown at low density compared with 0·0437 µg tartrazine (0·0347–0·0528 µg) for plants grown at high density. There was therefore an approximately fourfold difference in spray received per plant. If the amount of herbicide received per plant is a good predictor of response to herbicide, then we would expect only a fourfold difference in ED50 due to plant density at the time of spraying (see the Appendix).

Figure 2.

(a) Tartrazine captured by plants in relation to leaf area and (b) the spray received per unit leaf area of plants, with plants grown at two different densities (experiment 2). Diamonds, two plants per pot; squares, 64 plants per pot.

An important result was that a few of the plants grown at high density were as large as the smallest plants grown at low density. In this overlapping region of size the amount of tartrazine received by the leaves of plants at low density was greater than that received by plants at high density.

To assess the differential amount of spray intercepted by the plants at the different densities, the spray received per unit leaf area was calculated for each density (Fig. 2b). The low-density plants received approximately 2·5 times as much tartrazine per unit leaf area compared with the high-density plants [mean ± SE (95% confidence intervals) units mg tartrazine m−2: low density = 243 mg m−2 ± 11 (220–266); high density = 95·6 mg m−2 ± 7·7 (80–111)]. From the rationale in the Appendix, if the amount of herbicide received per leaf area is a sole predictor of response to herbicide, we would expect only an approximately 2·5-fold difference in ED50 due to competition at the time of spraying.

experiment 3: a test of the effect of plant density on ed50 before and at the time of spraying

Analysis of variance of the mean fresh weight per pot showed that there were significant effects (P < 0·05) of block, treatment, dose and dose × treatment (Table 4). The growth of plants in LLL at control and very low doses was noticeably greater (K = 1·7 g) than that of plants grown at high densities before spraying (HLL and HHL K = 1·0 g) (Fig. 3), reflecting the effect of competition during the first 2 weeks of growth.

Table 4.  Results of anova on the dose–response data of A. githago grown in three treatments (HLL, LLL HHL), where each treatment was treated with 12 doses with seven blocks (experiment 3)
Treatmentd.f.MSF-ratioProbability > F
Block 6 0·487 2·53    0·022
Treatment 2 6·69 34·8< 0·0001
Dose 1128·3147< 0·0001
Dose × treatment 22 0·32 1·66    0·037
Error210 0·051  
Figure 3.

Dose–response relationships and fitted curves comparing treatment HLL with LLL as well as treatment HLL with HHL (experiment 3). See text for explanation of treatment codes. HHL vs. HLL: an improvement of fit was obtained by unconstraining the ED50 (F1,164 = 32·5) and no further improvement of fit could be made. HLL vs. LLL: an improvement of fit was obtained by unconstraining K (F1,164 = 56·3) and no further improvement of fit could be made. The reason for the two slightly different models for treatment HLL is that these are the best fit models compared with the two alternative treatments.

Treatment HLL vs. treatment LLL: a test of whether different densities before spraying affect the dose–response relationship

Dose–responses in treatments HLL and LLL were compared and found to differ only in the estimate of K (Fig. 3). This result indicated that a difference in plant density before spraying had no difference on the ED50 for this particular combination of herbicide, plant species and conditions.

Treatment HLL vs. treatment HHL: a test of whether different densities at the time of spraying affect the dose–response relationship

The ED50 for the treatment HHL was approximately threefold that for treatment HLL (Table 5; F1,164 = 33). Thus density at the time of spraying contributed approximately threefold of the overall 15-fold difference found in the ED50 values for plants grown at low and high density.

Table 5.  Parameter estimates for the dose–response curves for treatments HLL and HHL constraining parameters b and D to common values but allowing the ED50 to vary (experiment 3)
ParameterEstimateLower 95% confidence limitUpper 95% confidence limit
D 0·986 0·813 1·158
ED50 HLL20·05 3·908 36·20
ED50 HHL62·8914·01111·8
b 0·518 0·463 0·574

experiment 4: a test of whether herbicide reduces competitive effects amongst plants

The mean fresh weight of the protected focal plants was significantly greater (planned one-tailed t-test, P < 0·001) than the mean from the control treatment (Fig. 4). Therefore, it could be concluded that, at this level of herbicide, the phytotoxic effect of herbicide reduces competition significantly.

Figure 4.

The effect of herbicide vs. control (water) on fresh weight of unsprayed (focal) plants grown at high density. The treatment was sprayed onto the neighbouring plants but not the plants that were weighed. H = herbicide treatment; W = water treatment (experiment 4).

experiment 5: a test of whether competition after spraying has an effect on dose–response to herbicide

Analysis of variance showed that there were no significant block effects (F6,147 = 0·83; P > 0·5) but significant dose (F12,147 = 6·16; P < 0·001) and treatment effects (F1,147 = 106; P < 0·001) and a dose–treatment interaction (F12,147 = 6·99; P < 0·001).

Dose–response analysis demonstrated that parameters K and ED50 differed between the two treatments (Fig. 5).

Figure 5.

Dose–response relationship and fitted curves comparing treatment HHL with HHH (experiment 5). See text for explanation of treatment codes. Improvement of fit was obtained by unconstraining (i) K (F1,187 = 84·9), (ii) ED50 (F1,186 = 63·8) and no further improvement of fit could be made.

The estimates for the ED50 values differed greatly between the two treatments, with ED50 for treatment HHL being approximately 10 times the ED50 for treatment HHH (Table 6). This strongly supported the hypothesis that competition acts during the growth period after spraying to alter the final estimate of the ED50.

Table 6.  Parameter estimates for the dose–response models for the two treatments in experiment 5: HHL, plants were grown at high density before and at the time of spraying and at low density after spraying; HHH, plants were grown at high density throughout. Parameter b was constrained to a common value for both treatments as a consequence of fit analysis
ParameterEstimateLower 95% confidence limitUpper 95% confidence limit
D HHL  0·765  0·662  0·868
ED50 HHL 272·7 108·7 436·7
D HHH  0·296  0·266  0·325
ED50 HHH268213114052
b  0·580  0·511  0·650

It should be noted that the ED50 values found for treatment HHL in experiment 5 (Table 6) were much higher than in the equivalent treatment in experiment 3 (Fig. 3; ED50 = 63 g a.i. ha−1) and that the ED50 values found for treatment HHH were much higher than in the equivalent treatment in experiment 1 (616 g a.i. ha−1; Table 3). The difference in ED50 between equivalent treatments in different experiments may have been due to problems with equipment (see the Methods). It is important to note, given the equipment failure during experiment 5, that we cannot assume these results would be replicated if the experiment was carried out in more controlled conditions.

Discussion

Knowledge of dose–response relationships between a pesticide and a range of species is an integral part of the regulatory process (Hance & Holly 1990). For herbicides, the dose–response to the chemical is usually tested on a range of species, including those of economic importance (crops and major weeds) and a range of non-target species that might be affected by it during use (Boutin, Freemark & Keddy 1995).

Dose–response relationships often show high variability, and it is essential to minimize this as far as possible. These relationships have been shown to be affected in the field by wind speed and rainfall (Garrod 1989; Fletcher, Johnson & McFarlane 1990), and in both field and laboratory by temperature, size of plant (Ascard 1994), physiological state of the plant (Garrod 1989), time of spraying (Skuterud et al. 1998) and the density of the plants being tested (Ascard 1994). Clearly, if this is so and the aim is to isolate the effects of the herbicide on plant performance and to produce accurate results, then it is essential to understand how experimental conditions influence the process.

In order to minimize experimental error in this study we used seeds from the same source, carried out our experiments using a precision sprayer and grew our plants in a controlled environment room. However, even with these precautions the variability was high, and there were important block effects noted in some experiments. In experiment 1, for example, block effects within shelf were found although there were no consistent trends found in these block effects, and there were no significant effects of shelf. These results suggest that high replications, coupled with appropriate blocking to take account of the small-scale variations in environmental conditions even within a controlled environment room, are needed to minimize variation and hence increase the signal relative to noise. The high level of variation is of unknown source and may be a result of any combination of influences, e.g. small-scale environmental heterogeneity, genetic variation, variation in plant size for whatever reason, variation between plants in herbicide deposition and competition increasing the magnitude of variation from other sources.

effects of plant density on sensitivity to herbicide

It is surprising that there have been few attempts to assess the effects of plant density on dose–response relationships, given the well-known effects that density has on the outcome of competition (Begon & Mortimer 1986; Firbank & Watkinson 1990; Goldberg & Barton 1992) and the general acceptance of its importance in commercial plant growing. The limited information that is available comes from a study on flame weeding showing that plant density was not as important as the size of plant in determining a plant’s response to flame weeding (Ascard 1994). However, this conclusion cannot be extrapolated directly to herbicides because: (i) size was varied by varying age, which might correlate with variation in other variables such as cuticle thickness or hairiness that, in turn, could affect the response to flame weeding more or less than the response to herbicide; (ii) it is difficult to compare the two different variables (size and density) because the change in plant response depends on the relative range of the two different variables tested, the higher the range tested the greater the effect on the dose–response. Therefore, it is possible that the conclusion of Ascard (1994) results from the range of densities tested being low compared with the range of sizes tested.

However, in this study we have unequivocal evidence (experiment 1) that the density of A. githago plants has a significant effect on the dose–response to the herbicide 2,4-D amine. The data for the high densities (64 plants per pot) were variable but still gave an ED50 of approximately 15 times the ED50 calculated for the low densities (two plants per pot). This suggests that competition is important in determining ED50 values.

isolating the phase in which competition affects dose–response

It is possible that the main effect is due to competition before spraying, where, for example, plants grown at high densities grow with more vertical leaves and hence receive less herbicide than plants grown at low density. The comparison of plants grown at high density prior to spraying but reduced to low density at the time of spraying, with plants grown at low density throughout (experiment 3), showed that there was no significant difference in the ED50 values. Therefore, we conclude that the effects of competition before spraying were not responsible for the observed difference in ED50 values.

Another possibility is that the main effect of competition occurs at the time of spraying itself. For example, it is possible that plants grown at high density shade each other for receipt of herbicide, and consequently on average receive less herbicide per unit leaf area than plants grown at low density. This study shows that the amount of spray per unit leaf area received by plants at low density was approximately 2·5 times the amount received by plants grown at high density and that the amount of spray received per plant at low density was approximately four times the amount received by plants grown at high density (experiment 2).

We also showed that, where the difference in density occurs only at the time of spraying itself, the ED50 is approximately three times higher in plants grown at high density (experiment 3). We therefore conclude that the effects of high density vs. low density at the time of spraying cause a c. threefold difference on the ED50 because either (i) plants at high density receive 2·5 times less herbicide per unit leaf area than plants grown at low density, or (ii) plants at high density receive four times less herbicide per plant than plants grown at low density. It would therefore be interesting to do further work to examine which of the two measures of herbicide receipt (dose per plant, dose per leaf area) is a better predictor of response to herbicide. There was a high level of variation in the amount of spray received per plant and per unit leaf area. The extent to which herbicide uptake and subsequent effect is influenced by deposition on stems as opposed to leaves deserves further attention.

The third period during which competition may have a significant effect on the dose–response to herbicide is competition after spraying and before harvesting. This may have two explanations. (i) There is, in effect, a trade-off between the phytotoxic effect of the herbicide reducing growth and the effect of reduced competition from the neighbours that increases growth. Thus the effect of herbicide is less than where there is no competition, so the ED50 is greater at higher plant density. (ii) Competition after spraying may decrease the rate of growth and thus decrease the effect of the herbicide 2,4-D, which acts as a growth inhibitor (Royal Society of Chemistry 1991).

We have shown that there was a 15-fold difference in ED50 between plants of high and low density (experiment 1), of which most was due to competition during the period after spraying (experiment 5). We did not investigate how much this was due to the two explanations set out above (i and ii). However, we were able to show that application of herbicide at a high concentration in dense populations caused decreased competition (experiment 4). This is the result expected if the first explanation (i) was important, i.e. that at high density competition is reduced by the effect of herbicide and therefore the net effect of herbicide is less than for those plants without competition. However, it is possible that the large effect of competition after spraying demonstrated in experiment 5 may have been altered by the different conditions in that experiment due to equipment failure. Future work should safeguard against changes over time by a fully complete experiment containing all treatments (LLL, HHH, HHL, HLL) with replication over time to achieve sufficient power.

In these experiments we have demonstrated that density affects plants’ responses to herbicide. However, we were unable to explore the relationship between density and ED50. Future work might test a number of different densities and attempt to describe, mathematically, the relationship between ED50 and density.

the agricultural context of these results

The work presented here has used A. githago as a model species because it has many attributes suitable for competition studies. However, this species is no longer of economic importance as a weed as a consequence of improved seed cleaning. The densities tested here differed greatly (32-fold difference) and it is not known whether such densities were found in the field when A. githago was a common weed. It is known, however, that similar densities of A. githago have been used in competition experiments (Firbank & Watkinson 1985). While these results need to be tested on other species, the conclusions are likely to be applicable to crops and weeds used to assess pesticide toxicity.

The regulatory process demands that the toxicity of a pesticide to both weeds and crops be assessed (Hance & Holly 1990). The results presented here demonstrate that not only should a range of doses be applied but that the effects of plant density should be taken into account. For example, it may be that, as in our experiment, weeds at low density are particularly sensitive to a pesticide. This potential positive feedback system might lead to local extinction and this should not be ignored, particularly if the weed is of conservation value. Furthermore, although it is beyond the immediate scope of this study, we believe that our findings suggest that non-target neighbour species may strongly affect the response of a plant to herbicide (Humphry 1999). Therefore future work should include testing combinations of species, e.g. crop and weed.

In summary we have shown that a high density of plants increased the ED50 and that the period during which density had most effect was the period after spraying. It follows that single plant dose–response experiments may not be sufficient to predict the effect of herbicide on a community (either semi-natural or sown; Brain et al. 1999). This is because, within a community, the densities of plants are greater than those found in single-plant dose–response trials. It is also possible that the effect of density on a species’ response to herbicide depends on the species in question. Further work to test this might involve comparing the effect of density on dose–response amongst a range of species.

Acknowledgements

This work was carried out at the University of Liverpool with financial support from the Natural Environment Research Council. We thank Hazel Lewis for technical support. Kathy Humphry is thanked for assistance in the preparation of this paper.

Appendix

Proof that, for a given delivery rate, if the amount of spray received by a plant changes due to changes in the canopy structure at high density, for example, then the ED50 changes by the same factor.

Plants are grown at a given density, sprayed with a range of doses (x) and their individual plant response (y) is measured some time after spraying. This response is described as:

image

where K, ED50 and b are parameters as previously defined.

Now consider plants grown under exactly the same environmental conditions but at different densities such that, when they are sprayed with an application of herbicide, individuals plants that would have previously received a dose (x), now receive a dose of (ax) due to changes in canopy structure (shading for example). The response in this second scenario is called (y2):

image

As for every dose (x) the plants now receive a dose of (ax):

image

therefore:

image

This must be true for all x including x = 0, therefore:

image

Thus:

image

This is true for all x including x = ED50_2, therefore with substitution:

image

this implies (and is implied by):

image

or

image

As it must be true when b ≠ 0, then:

image

A similar proof for the four-parameter model confirms the same result (Humphry 1999).

Ancillary