Introduction

In weed management, research has repeatedly focused on understanding and manipulating the relative importance of interspecific plant competition for limiting resources (bottom-up effect) and herbivory (top-down effect)1,2,3,4. To compare the impact of plant competition and herbivory and analyze their interactive effects, Sheppard proposed three possible outcome categories: substitutive, additive (or multiplicative), and synergistic5. While synergistic outcomes, i.e., the impact of the combined application of plant competition and herbivory being higher than the multiplicative effect of the two single factors, are most preferred from a weed control perspective, they are rather uncommon. In most situations, multiplicative outcomes (i.e., additive on the log scale) are observed, with plant competition as the relatively more important factor5. Yet, there is a high variability in outcomes of studies combining plant competition and biocontrol herbivory5,6, and a mechanistic understanding of how factors interact and how they influence, individually or combined, the success of weed biological control is often lacking. This may explain why the success of such projects varies considerably and ranges from having a negligible impact to resulting in complete control7.

To advance our understanding of the interactive effects of plant competition and herbivory on weed performance or abundance, not only the direct impacts on the target plant, but also the interactions between competition and herbivory should be studied6 (see Fig. 1 for a conceptual illustration). For example, plant competition can mediate either herbivore load (e.g., through a change in plant apparency or micro-climatic conditions8) or herbivore impact (e.g., through a change in herbivore defense or tolerance9). Alternatively, moderate direct effects of herbivory can lead to significant changes in the competitive interactions between the target weed and the desirable plant community10,11. Moreover, herbivore impacts might depend on the size of the target weed, as plant herbivore resistance or tolerance might differ among growth stages. Evaluation of plant size and its inclusion as a covariate into statistical models can thus “reduce the unexplained variance and reveal agent effects that would otherwise have gone undetected”12. Yet, to assess plant size dependency of the interactive effects of plant competition and herbivory on weed performance, multi-factorial studies are required that allow investigating direct and indirect processes for a range of pre-specified plant sizes. Although repeatedly demanded (e.g., Morin et al.12, Willis et al.13) such studies are rare.

Figure 1
figure 1

Conceptual illustration of the direct and indirect effects of plant competition and herbivory on a target plant’s performance at two differing initial plant sizes. Hypothesized outcomes under no competition (a) and plant competition (b) are each displayed in yellow ellipse. In general, herbivory can be expected to impair weed growth across a range of plant sizes (a: ii). However, in the absence of interspecific plant competition, weeds may benefit from increased availability of abiotic resources, potentially allowing them to respond to non-lethal herbivory by investing in compensatory or even over-compensatory regrowth62. When exposed to the effects of both plant competition and herbivory (b: ii, iii), weed performance is expected to be more severely affected than by herbivory alone, mainly because resource availability will be limited. Moreover, effects of plant competition on herbivore behavior and survival may influence the top-down pressure on the target weed (b: iv). Where the effects of plant competition and herbivory are plant size dependent (i), the magnitude of the interactive effect of herbivory and plant competition would differ in smaller and larger weed plants (b: compare internal structure of arrows).

Full size image

Rumex obtusifolius L. (broad-leaved dock) is a perennial plant and is considered a problematic weed in both its native range (Eurasia) and its introduced ranges (North and South America, Southern Africa, Australia, New Zealand, Eastern Asia)14,15,16. Several characteristics of the plant facilitate successful propagation, including high production of seeds and their long viability15 as well as a taproot system with the ability for clonal growth17. Where R. obtusifolius is abundant, it can form large, persistent soil seed banks18 from which the species can potentially recruit for many years. In conventional farming, R. obtusifolius is often controlled by herbicides. However, in organic farming, where herbicides are not allowed, non-chemical control measures are highly demanded16,19.

Rumex obtusifolius is able to grow in intensively managed grasslands used for livestock production, as the species resists repeated defoliation16,20. These grasslands can be sown or permanent and, depending on pedo-climatic conditions, are fertilized and mown and/or grazed several times per year21,22. Under medium to high management intensity, weeds are suppressed by resource competition from the forage plants23,24,25. However, although high management intensities with frequent cutting (up to six times annually) reduce aboveground biomass of R. obtusifolius, the species cannot be eliminated by cutting alone16,20,26. In situations with very high R. obtusifolius abundance (> 5–8 small to big plants m−2), its negative impact on forage yield becomes so strong that grasslands are generally renewed by ploughing and reseeding27. All these aspects point to the need for a cost-effective control of the species.

The two congeneric sesiid moths, Pyropteron chrysidiforme (Esper) and P. doryliforme (Ochsenheimer) (Lepidoptera; Sesiidae) are native to Europe. Pyropteron chrysidiforme occurs throughout large parts of western and central Europe, while P. doryliforme is mainly found in the Mediterranean region28,29. The larvae of both species are root borers and are highly host-specific to the genus Rumex. They were considered as candidates in a classical biological control program in Australia against introduced Rumex species, including R. obtusifolius28,29, but only P. doryliforme was able to synchronize its univoltine life cycle to the southern hemisphere conditions. This classical biological control program was successful and resulted in a significant reduction in densities of introduced Rumex species30. Biological control has also been considered as a non-chemical management option against R. obtusifolius in the weed’s native range in Europe by implementing an inundative biological control approach19,31. The inundative biological control approach uses periodic releases of large numbers of control agents over a target weed population to control it by causing high levels of damage, while the build-up of antagonist populations is neither intended nor expected19,32. However, a recent study in which full-grown R. obtusifolius plants were inoculated with eggs of P. chrysidiforme under field conditions revealed an insufficient level of control33. Root biomass of large R. obtusifolius plants was only marginally affected by the herbivore and no mortality was observed33.

Given the results of Hahn et al.33, here we assessed the interactive effects of interspecific plant competition and root herbivory on R. obtusifolius to clarify the extent to which these effects and their relative importance depended on the size of the target weed. To this aim, we manipulated three factors in a field experiment: first, presence or absence of competition from a grass sward (Lolium perenne L.); second, root herbivory of larvae by application of the two Pyropteron species, P. chrysidiforme and P. doryliforme; and third, a wide range of root mass (i.e., plant size) of R. obtusifolius covering the natural range of size variation. Lolium perenne (perennial ryegrass) was chosen as a competitor because it is the most important grass species of temperate productive grasslands worldwide and was previously used to test for competitive effects on R. obtusifolius20,34. Artificial infestation of R. obtusifolius by the two Pyropteron species was used to clarify their potential for inundative biological control in the weed’s native range in Europe19,33. Finally, target plant size has relevance because large, flowering plants of R. obtusifolius are known to be strong competitors20, while seedlings or small rosettes are not35. Specifically, we addressed the following questions: (1) Does competition from a L. perenne sward affect infestation of R. obtusifolius by Pyropteron larvae (Fig. 1, iv)? (2) What is the relative importance of grass competition and herbivory on the performance of R. obtusifolius plants (Fig. 1, ii–iii)? (3) Does grass competition combined with herbivory have a substitutive, multiplicative, or synergistic effect on R. obtusifolius performance? (4) Do the interactive effects of competition and herbivory depend on the initial size of R. obtusifolius plants (Fig. 1, i)?

Results

Increased infestation of herbivores under plant competition

First, we assessed the probability of a R. obtusifolius plant being infested by Pyropteron after placing a toothpick with 30 eggs glued-on into the soil near the plant base. Infestation probability was significantly greater when R. obtusifolius was grown under competition from L. perenne (Fig. 2, Table 1: competition effect). For example, in spring 2020, the infestation probabilities were 0.70 and 0.57 for P. chrysidiforme and P. doryliforme, respectively, when grown under competition, but were 0.26 and 0.13 when grown alone. The degree of infestation was generally sustained over both harvest seasons (Table 1), but infestation was significantly greater for P. chrysidiforme than for P. doryliforme, with the differences becoming less pronounced in spring 2020 (Fig. 2, Table 1: harvest season × Pyropteron interaction). Noteworthy, the infestation probability did not differ between R. obtusifolius plants grown from initially smaller or larger roots (Table 1). Roots in the control treatment revealed no signs of herbivore attack, indicating that there was no occasional infestation by Pyropteron or other root-boring insects. Second, the total number of larvae retrieved per infested plant was similar under the two competition treatments and between harvest seasons, but was significantly greater for P. chrysidiforme than for P. doryliforme (Fig. 3, Table 1).

Figure 2
figure 2

Probability of a R. obtusifolius plant being infested by at least one larva measured in the two harvest seasons autumn 2019 (a) and spring 2020 (b) as affected by Pyropteron treatment (application of P. chrysidiforme [Pch] or P. doryliforme [Pdo]) under no competition and competition from a L. perenne sward. Only plants under the two Pyropteron treatments analyzed because there was no infestation of control plants (no inoculation with Pyropteron). Displayed are means ± standard error across all initial root sizes, calculated following Agresti and Coull63.

Full size image
Table 1 Summary of generalized linear mixed-effects models for the effects of harvest season, competition from a L. perenne sward, Pyropteron application treatment, and initial root mass of R. obtusifolius on the probability of a R. obtusifolius plant being infested and the total number of larvae per infested plant.
Full size table
Figure 3
figure 3

Total number of larvae retrieved per infested plant of R. obtusifolius measured in the two harvest seasons autumn 2019 (a) and spring 2020 (b) as affected by Pyropteron treatment (application of P. chrysidiforme [Pch] or P. doryliforme [Pdo]) under no competition and competition from a L. perenne sward. Only infested plants included in analysis (compare n below panels). Displayed are means ± standard error across all initial root sizes.

Full size image

Synergistic interactions between plant competition and herbivores increase impact on small R. obtusifolius plants

Given that infestation probability (Fig. 2) and larval load (Fig. 3) were similar in autumn 2019 and spring 2020, all further analyses were performed using the data from the final harvest in spring 2020. Here, competition from the L. perenne sward reduced the aboveground biomass of R. obtusifolius 62.7-fold (Fig. 4a,b) and final root mass 7.4-fold (Fig. 4c,d; Table 2). Thus, on average across the Pyropteron treatments, aboveground biomass and final root mass of R. obtusifolius were 94.1 g (± 3.74 g SE) and 190.7 g per plant (± 9.80 g), respectively, when grown without competition, but only 1.5 g (± 0.17 g) and 25.7 g (± 2.33 g) when grown under L. perenne competition (Fig. 4). The competition effect from L. perenne was so strong that there was hardly any growth of R. obtusifolius roots under this treatment, irrespective of initial root mass (Fig. 4d: compare dotted line). By contrast, when grown without competition, R. obtusifolius plants from all initial root sizes grew up to the same size order of magnitude (Fig. 4a,c).

Figure 4
figure 4

Aboveground biomass and final root mass of R. obtusifolius plants in spring 2020 grown under no competition (a,c) and competition from a L. perenne sward (b,d) depending on the initial root mass and Pyropteron treatment (no application [Ctr], application of P. chrysidiforme [Pch] or P. doryliforme [Pdo]). Predicted lines (± standard error grey shaded) are based on generalized linear mixed-effects models (Table 2). No standard error is given for Pdo because it was largely overlapping with the other treatments. Unequal variances were modeled with a distinct dispersion function (see Supplementary Appendix S1 for the equation). Aboveground biomass is the sum over all five harvests. Dotted lines in panels (c) and (d) represent the 1:1 relation of final versus initial root mass, i.e., steady-state of growth and decay.

Full size image
Table 2 Summary of generalized linear mixed-effects models for the effects of competition from a L. perenne sward, Pyropteron application treatment, and initial root mass of R. obtusifolius on its aboveground biomass and final root mass in spring 2020.
Full size table

Pyropteron application negatively affected aboveground biomass and root mass of R. obtusifolius when grown under L. perenne competition, but not without (Fig. 4, Table 2: significant competition × Pyropteron interaction), indicating a synergistic effect of grass competition and Pyropteron application on the growth of R. obtusifolius plants. The different Pyropteron effects in the presence and absence of competition justified to split the data into the ‘L. perenne competition’ and ‘no competition’ treatments to reduce model complexity and allow for a clearer interpretation (see Supplementary Appendix S1 for details). By doing so, it turned out that only under competition were plants with initially smaller roots more negatively affected by Pyropteron than plants with initially larger roots (Supplementary Table S1: initial root mass × Pyropteron interaction), and that the impact was only significant for P. chrysidiforme (Fig. 4b,d; Supplementary Table S2). Thus, under L. perenne competition, comparing the responses over initial root mass between the control treatment and plants under P. chrysidiforme application (reflected in the fitted lines, Fig. 4b,d), a significant slope difference appeared for aboveground biomass (z = 2.3, P = 0.022) and final root mass (z = 1.9, P = 0.057; Supplementary Table S2), indicating a greater herbivore impact when R. obtusifolius was grown from smaller plants. With P. doryliforme application, no differences in slopes to the control were evident for aboveground biomass and final root mass (both z ≤ 0.155, P > 0.877).

Competition-mediated herbivore load per final root mass

Combining the results of number of larvae per plant (Fig. 3) and the effects of competition and Pyropteron application on root mass (Fig. 4c,d) suggested that herbivore loads per unit root mass would differ between small and large roots at the end of the experiment. Indeed, when the number of larvae retrieved was scaled by 100 g of final root mass, significant differences between the competition treatments were evident (Fig. 5, Table 3: competition effect). In spring 2020, on average 1.6 and 1.3 larvae per 100 g final root mass were retrieved for P. chrysidiforme and P. doryliforme, respectively, under no competition, but 15.5 and 7.4 larvae (per 100 g final root mass) under L. perenne competition. Thus, competition from the grass sward induced a ten- and sixfold higher herbivore load per unit final root mass for P. chrysidiforme and P. doryliforme, respectively, the difference in larval load between the two Pyropteron species being significant (Table 3). Furthermore, significantly more larvae per unit final root mass were retrieved from initially smaller roots than from larger roots when plants were grown under competition from L. perenne (Fig. 5b, Table 3: initial root mass × competition interaction).

Figure 5
figure 5

Number of larvae retrieved per 100 g final root mass of infested R. obtusifolius plants in spring 2020 grown under no competition (a) and competition from a L. perenne sward (b) depending on the initial root mass and Pyropteron treatment (application of P. chrysidiforme [Pch] or P. doryliforme [Pdo]). Only infested plants included in analysis. Predicted lines (± standard error grey shaded) are based on a linear mixed-effects model (Table 3).

Full size image
Table 3 Summary of linear mixed-effects model for the effects of competition from a L. perenne sward, Pyropteron application treatment, and initial root mass of R. obtusifolius on the number of larvae scaled by final root mass of infested R. obtusifolius plants in spring 2020.
Full size table

Synergistic interactions increase root decay

The proportion of root decay was significantly higher in R. obtusifolius plants grown under L. perenne competition than in those grown without competition, and it was significantly increased by Pyropteron application (Fig. 6, Table 4). Furthermore, competition from L. perenne amplified root decay caused by Pyropteron application (Table 4: marginally significant competition × Pyropteron interaction), again pointing to a synergistic effect between the two factors. The average proportions of root decay under no competition were 0.07, 0.23, and 0.11 for control, P. chrysidiforme, and P. doryliforme treatments, respectively, but under competition from L. perenne were, respectively, 0.31, 0.78, and 0.64. Initially smaller roots revealed significantly greater root decay than initially larger roots only under the competition treatment (Table 4: initial root mass × competition interaction, no figure shown).

Figure 6
figure 6

Proportion of root decay of R. obtusifolius in spring 2020 as affected by Pyropteron treatment (no application [Ctr], application of P. chrysidiforme [Pch] or P. doryliforme [Pdo]) under no competition and competition from a L. perenne sward. Displayed are means ± standard error across all initial root sizes. Symbols indicate significance of Tukey range test within each competition treatment, following a generalized linear mixed-effects model (Table 4). ***P ≤ 0.001, **P ≤ 0.01, †P ≤ 0.1, ns not significant.

Full size image
Table 4 Summary of a generalized linear mixed-effects model for the effects of competition from a L. perenne sward, Pyropteron application treatment, and initial root mass of R. obtusifolius on its proportion of root decay in spring 2020.
Full size table

The proportion of root decay was positively related to the number of larvae retrieved under both competition treatments (z = 4.9, P < 0.001, inference for slopes in Fig. 7). However, per unit number of larvae, greater proportional root decay was caused in plants grown under L. perenne competition than those grown without competition (z = 8.7, P < 0.001, inference for average level of root decay, Fig. 7).

Figure 7
figure 7

Proportion of root decay of R. obtusifolius in spring 2020 under no competition (a) and competition from a L. perenne sward (b) depending on the total number of larvae per infested plant. Predicted lines (± standard error grey shaded) are from a beta regression model with an overall R2 value of 0.31. Fitted slopes were not significantly different between competition treatments (χ2 < 0.1, df = 1, P = 0.995). Circles are scattered horizontally to improve their visibility.

Full size image

No compensatory growth of R. obtusifolius plants due to herbivory

Competition from L. perenne significantly suppressed the formation of auxiliary rosettes of R. obtusifolius (Supplementary Fig. S1 and Table S3) and reduced the number of roots (Supplementary Fig. S2 and Table S3), while Pyropteron application caused no effects on either parameter. Thus, despite the impact of the Pyropteron larvae on growth of aboveground biomass and root mass of R. obtusifolius (Fig. 4) and proportion of root decay (Fig. 6), R. obtusifolius plants did not respond with compensatory growth.

Discussion

Using a multi-factorial experimental design, we found that interspecific competition and targeted application of biological control candidates affected the growth of R. obtusifolius in an interactive, synergistic way. Plant competition had a higher direct impact on R. obtusifolius growth than herbivory and also indirectly affected plant growth by increasing the probability of infestation by the biological control candidates and, consequently, the level of root decay. By integrating the range of naturally occurring plant sizes, we could show that an additional amount of the observed variation in interactive effects of competition and herbivory can be explained by plant size. Smaller plants were affected by both competition and herbivory by P. chrysidiforme, while larger plants were mainly affected by competition. These results provide evidence for a significant plant size dependency in the magnitude of the effects of competition and herbivory on the growth of R. obtusifolius and have implications for the management of this problematic weed.

Multi-factorial approaches enable identification of the relative effects of several factors in a single ecological context and whether substitutive, multiplicative or synergistic dynamics are associated with the combination of factors. In our study, interspecific plant competition was the single most important factor impairing growth of R. obtusifolius plants, which is consistent with previous studies comparing the effect of competition and herbivory on weed performance5,36,37. Lolium perenne, a characteristic grass species of productive grassland systems in temperate regions worldwide, was able to effectively suppress growth of R. obtusifolius and to increase root decay. In a similar study assessing the combined effects of grass competition and root herbivory on Centaurea maculosa Lam.38, the strong effect of the grass competitor was attributed to belowground competition for nutrients and water, as well as competition for light of the rosette plants in the grass sward. In our experiment, we specifically tested interactive effects of interspecific plant competition and herbivory at two fixed levels (i.e., presence vs. absence). Theoretically, the effects of both plant competition and herbivory may vary with plant density/biomass or herbivore load. In the case of interspecific plant competition, however, applying different sowing densities of L. perenne to intensively managed grasslands (as reflected in our experiment) would not translate into differences in yield or resource use21,39. Therefore, manipulating sowing density of L. perenne cannot be expected to result in differential competitive effects on R. obtusifolius34.

Herbivory affected the performance of R. obtusifolius to a lesser extent than plant competition. The clearest effect of herbivory was on proportion of root decay, which increased through feeding by P. chrysidiforme larvae in both the presence and absence of plant competition (Figs. 6 and 7). Root decay is expected to translate into a reduction in plant performance40 and in our study, increase in root decay coincided with reduced belowground and aboveground biomass production (compare Figs. 4 and 6). Despite the percentage of root decay being high, R. obtusifolius plants did not respond with compensatory growth to Pyropteron feeding (Supplementary Table S3, Figs. S1, S2). Thus, a plant with 70% decayed root mass (approx. mean of Pyropteron applications under competition, Fig. 6) can be assumed to be severely degraded if not dead, which should result in a sustained herbivory effect in the long run. It should be noted that in our study the effect of herbivory on root biomass was underestimated, as the root biomass measured at the end of the experiment also included degraded root mass. Infestation probability and impact were higher with P. chrysidiforme than with P. doryliforme, which may indicate differences in climatic suitability of the two species. P. chrysidiforme has its distribution range in central and western Europe and is better adapted to the relatively cool and humid conditions of the study site than P. doryliforme, which has a Mediterranean distribution range29. Finally, the number of larvae per plant (Fig. 3) was in the same range as found in previous studies33,41, indicating that P. chrysidiforme and P. doryliforme build up comparable herbivore loads under varying field conditions.

Under natural conditions, plant competition and other factors may also affect oviposition behavior and thus increase variation in number of eggs deposited on R. obtusifolius plants. Here, we deliberately applied a standardized number of eggs to reflect a biocontrol management approach using targeted inoculations of individual plants and by this, aspects of host-finding and female oviposition preferences depending on host size and occurrence of non-host species have been excluded. The number of eggs applied were in the range of those used in the biological control program in Australia42 and in a previous inundative biological control experiment in Switzerland33. Moreover, preliminary tests revealed that applying a different number of eggs per plants did not result in significantly different infestation rates (U. Schaffner, unpublished data). This may be partly explained by interference competition among larvae as observed in laboratory bioassays and pot experiments43.

In addition to the direct effects of plant competition, we found that plant competition also indirectly affected growth of R. obtusifolius by increasing the probability of infestation by Pyropteron larvae (Fig. 2, up to 70% infested plants in spring 2020), indicating that plants suppressed by competition experienced a stronger herbivore effect. Theoretically, neighboring plants may negatively (associational resistance) or positively (associational susceptibility) interact by altering the behavior or numerical responses of their natural enemies1,44,45. Shabbir et al. found that a biological control agent for the weed Parthenium hysterophorus L. induced more galls when the target weed was grown in competition with other plant species than when grown alone46; the authors proposed that the increased herbivore load on plants under competition may be due to a fertilizer effect of the competing plants or due to altered female oviposition behavior. In our study, altered female oviposition behavior could not explain the increased herbivore load and root decay, as all plants subjected to herbivory were inoculated with the same number of eggs. The grass sward surrounding the inoculated plants may, however, have improved the micro-climatic conditions for neonate larvae thus increasing their survival from egg hatching to entering the roots, resulting in higher infestation (Fig. 2). In an experimental study, herbivore damage on Solanum carolinense increased with increasing frequency of a plant competitor, and soil moisture and other microclimate variables were (negatively) correlated with herbivore damage47, suggesting that the susceptibility was at least partly mediated by effects of plant competition on microclimatic conditions. An alternative explanation for associational susceptibility is that plant competition and the associated reduction in resource availability may impair the target plant’s investment in chemical defenses48. In our study, the increased infestation rate of R. obtusifolius by the herbivores under L. perenne competition constitutes a mechanism that partially explains the synergism between the two suppressive factors.

Following the classification of Sheppard on the type of outcomes5, the combination of plant competition and root herbivory enhanced the negative effect on R. obtusifolius in a synergistic way (Figs. 4 and 6, Tables 2 and 4). Synergism between the two factors can also be demonstrated by calculating the effect size based on the single, isolated effects (i.e., the effect of competition without P. chrysidiforme application and of P. chrysidiforme application in the absence of competition) and comparing this value with the observed impact under the combined treatment. Doing so for aboveground biomass grown from the two pooled largest categories of initial root mass, we expected a 29-fold reduction from the multiplied isolated effects, but observed a 50-fold reduction with the combined treatment. The same figure for aboveground biomass grown from the two pooled smallest initial root categories predicted a 196-fold reduction from the multiplied isolated effects, but we observed a 1067-fold reduction with the combined treatment. Our findings are consistent with the simpler type of synergy where the factor with the lower interaction strength (usually herbivory) is observed to have an impact only in the presence of the factor with the higher interaction strength (usually plant competition4,5,40). In our experiment, the number of larvae per plant was not affected by the competition treatment (Fig. 3, Table 1). However, once established in the root, these larvae had a generally higher impact on R. obtusifolius plants suppressed by the L. perenne sward (Fig. 4), which can be explained by the higher larval load per unit final root mass (Fig. 5). Surprisingly, field data for the existence of synergistic effects of herbivory and plant competition are relatively rare and, even if demonstrated, it is not clear whether these synergistic effects are strong enough to effectively control native weeds or invasive alien plant species (but see Santamaría et al.49, Zhang et al.50).

By deliberately incorporating a gradient of a trait of the target weed, i.e., initial plant size, we have shown that the interactive effects of competition and herbivory are subject to non-linear dynamics (Fig. 4). We chose initial plant size as a variable because it reflects a major source of variation in plant communities and because literature underlines the role of plant size in mediating plant competition20,51 and herbivory4. In R. obtusifolius, root size relates to carbohydrate storage capacity and therefore to the potential to resist sustained stress17,52, and in our study, larger plants appeared to have, at least in the short-term, a higher tolerance to herbivory even when subjected to competition. The observed effects imply that interspecific competition for soil resources and light, in combination with root herbivory, exhausted small plants to the degree that root growth was no longer possible (Fig. 4d) and plants would eventually die (J. Klötzli, personal observation). Our results also demonstrate that incorporating species traits and other biotic or abiotic variables can help increase the predictability of the interactive effects of herbivory and competition on plant performance. Plant size dependency has rarely been systematically investigated using multifactorial experiments that evaluate different biotic and abiotic factors on the joint effects of herbivory and (interspecific) plant competition. In our model, initial plant size explained an additional amount of variation in the interactive effects of competition and herbivory on R. obtusifolius growth, and by ignoring incorporation of initial plant size into the model, erroneous conclusions could have been drawn, thereby affecting management recommendations.

Implications for management

Regarding practical management, combining competition from grass swards and herbivory to manage grassland weeds in general and particularly R. obtusifolius has received interest in the past34,53 and follows the strategies of integrated weed management37,54. In the case of perennial weeds, herbivores attacking the storage organs belowground represents a great option to successfully reduce weed performance55. In our study, for example, the percent reduction of final aboveground biomass and root mass of R. obtusifolius induced by P. chrysidiforme was -85% and -80%, respectively, for the two pooled smallest categories of initial root mass. Such a strong impact suggests that P. chrysidiforme inoculated using an inundative approach in an established grassland could be a promising control option. This approach is also compatible with organic agriculture, where measures for the control of R. obtusifolius are urgently needed19. However, P. chrysidiforme would need to be repeatedly applied because a single application will most probably not permanently change the herbivore density to a degree as to result in an enduring impact on R. obtusifolius population densities. Rather, the results of this study suggest that the combination of interspecific competition and herbivory primarily damage smaller R. obtusifolius plants, which in the short-term would lead to a change in the size distribution of R. obtusifolius populations. The control effect will arise over time, when establishment of younger plants is impeded by repeatedly applied herbivores and mature plants naturally die or are regulated by other means. Our study thus indicates that smaller plants should be targeted using this inundative approach. It remains to be shown whether several years of competition and herbivory will further increase root decay, thus also weakening large R. obtusifolius plants.

Methods

Study system

Eggs of the two Pyropteron species used in this study originated from rearing colonies maintained at the CABI Switzerland Centre, Delémont. The rearing colony of P. chrysidiforme was established in 2010 with insects collected in southwestern Switzerland, while the colony of P. doryliforme was established in 2018 with insects collected in southern Spain. Rumex obtusifolius plants used in the experiment were dug up from managed, productive grasslands in northern Switzerland, where neither of the two Pyropteron species occur naturally. The use of Pyropteron spp. and R. obtusifolius plants in the present study complies with international, national and institutional guidelines, and we had the consent from the farmers who owned and managed the grasslands to sample R. obtusifolius plants.

Site conditions and experimental layout

A field experiment was carried out at Agroscope Zürich-Reckenholz, in central Switzerland (47°43′79″ N, 8°52′81″ E, 486 m a.s.l). The soil at the site is classified as calcic cambisol with a topsoil composition of 37.6% sand, 39.2% silt, 21.0% clay, containing 2.2% humus, and with a pH of 6.6. Experimental plots were established in spring 2019 and were maintained until May 2020. In 2019, mean annual temperature was 10.6 °C and annual precipitation 972 mm.

The experimental layout followed a split-split plot design with three factors to test the effects of plant competition, application of herbivore species, and initial root size of R. obtusifolius on its performance. Swards of Lolium perenne L. cultivar Allodia and plots with bare soil were established on sixteen main-plots of 1.8 m × 5 m (main-level factor), termed ‘L. perenne competition’ and ‘no competition’ treatment, respectively (see below for details on the treatments). Within each main-plot, three Pyropteron treatments testing the herbivore species were randomly assigned to sub-plots of 1.8 m × 1.66 m: application of P. chrysidiforme to R. obtusifolius plants, application of P. doryliforme, and a control with no application of either herbivore (split-level factor). Within each sub-plot, nine roots of R. obtusifolius of different size were randomly assigned to a 3 × 3 grid with a spacing of 0.4 m (split-split level factor), thus preventing plant-to-plant movement of herbivores and assuring that L. perenne remained the most abundant plant component in the competition treatment. This resulted in a total of 432 planted roots. Main-plots were arranged according to a randomized complete block design on the site (8 blocks, each containing a L. perenne competition and a no competition treatment).

Treatments and management

In the competition treatment, swards of L. perenne were sown after soil tilling in mid-August 2018 at a density of 25 kg seeds ha−1. Plots with bare soil representing the no competition treatment were tilled, but not seeded and were regularly weeded thereafter to prevent the growth of plants other than R. obtusifolius.

In early February 2019, R. obtusifolius plants of different sizes were dug up from grasslands (locations as noted), and roots were carefully washed free of soil and stored in a dark, cold room at 4 °C until early June. Thereafter, all roots were weighed, cut to a maximum length 15 cm, and labeled. The mass of these roots followed a log normal distribution, and the total sample was split into nine approximately equal groups based on log root mass. In mid-June 2019, roots were planted into the soil close to the surface: one root randomly selected from each size category was (randomly) arranged into the 3 × 3 grid of a sub-plot, for a total of nine plants per grid.

In early July 2019, after a three weeks period during which the roots were allowed to settle to the soil, all R. obtusifolius plants in the respective treatments were infested with P. chrysidiforme or P. doryliforme by placing a toothpick with 30 eggs glued-on into the soil near the plant base, following a procedure successfully implemented by Fisher et al.42. The toothpicks were removed after 2 weeks. The mean hatching rate per toothpick, determined by evaluating the eggs, was 0.85 and 0.75 for P. chrysidiforme and P. doryliforme, respectively, with no difference between the competition treatments.

All plots received phosphorus (100 kg P ha−1 year−1) and potassium fertilizer (100 kg K ha−1 year−1) in early spring 2019, following local fertilization recommendations for intensively managed grasslands. Moreover, plots were fertilized at a rate of 150 kg nitrogen ha−1 year−1 over three equal applications in mid-August 2019, mid-September 2019, and mid-March 2020. Applications in August and September were 3 to 5 days after cutting aboveground biomass of R. obtusifolius plants and mowing L. perenne swards to a height of 7 cm with a plot harvester (Wintersteiger Cibus).

Measurements

Aboveground biomass of all R. obtusifolius plants was repeatedly cut at a height of 5 cm aboveground, when the largest plants reached the flowering stage: in August, September, and November 2019, as well as in April and May 2020. The harvested biomass was dried at 105 °C for 24 h and weighed. For each plant, aboveground biomass was summed over harvests to obtain the cumulative aboveground dry biomass.

To evaluate the immediate infestation success of the herbivore species, one third of the R. obtusifolius roots were excavated and dissected following the third harvest of aboveground biomass in November 2019. This was used to verify whether treatment effects observed in spring 2020 could be assigned to the presence of larvae and not to other factors. Roots excavated in autumn 2019 were randomly chosen from all blocks, treatments, and initial root size groups to provide a representative sub-sample. The remaining two thirds of roots were excavated in spring 2020 at the end of the experiment. All roots were placed in labeled plastic bags and stored in the dark at 4 °C until dissection, which took place right after the harvests in autumn 2019 and spring 2020.

Prior to dissection, roots were washed free of soil, weighed, and the number of primary and secondary roots and number of rosettes present at the root collar were counted. We used fresh weight of roots as the response variable because it allowed to relate final root mass directly to initial root mass in the analysis (see below) and to determine the 1:1 relation of steady-state of growth and decay (compare Fig. 4). Roots were then dissected to determine the probability of R. obtusifolius plants being infested by Pyropteron larvae. The presence of living and dead larvae was recorded, from which the total number of larvae was calculated (see Supplementary Appendix S1 for details on the determination of infestation probability). To receive a measure of herbivore load in relation to the final plant size of R. obtusifolius, the total number of larvae was divided by final root mass. Following dissection, the proportion of the root that was decayed (dead material, structure porous, color of material blackish, brownish) was visually estimated and assigned to one of six proportion categories (0, 0.05, 0.15, 0.50, 0.85, 1). We used categories because, after root dissection, a determination to precise proportions cannot be done anymore.

Data analysis

Data were analyzed with (generalized) linear mixed-effects models. Response variables were infestation of R. obtusifolius plants by Pyropteron spp., total number of Pyropteron larvae retrieved per plant, total number of larvae per unit final root mass, aboveground biomass and number of rosettes per R. obtusifolius plant, root mass and number of roots per R. obtusifolius plant, and proportion of root decay. Predictor variables were harvest time (fixed factor with two levels), competition from L. perenne (fixed factor with two levels), Pyropteron treatments (fixed factor with three levels), and initial root mass. For initial root mass, we used the initially measured masses to increase accuracy and allow root mass to be a continuous variable. To account for the multilevel grouping structure of the experiment, block, main-plot, and sub-plot were each modeled as random factors (random intercepts). Given the different response variables and their underlying distributions, a variety of link functions were employed to appropriately model the data (see Supplementary Appendix S1 for all model equations). Note that a significant interaction between L. perenne competition and Pyropteron application in a generalised linear (mixed) model proves a synergism, i.e., an impact of the combined application of both factors being higher than the multiplicative effect of the two single factors.

Regarding predictors, the models had three different structures. First, we evaluated infestation of R. obtusifolius plants by Pyropteron ssp. in autumn 2019 and the following spring 2020. Thus, probability of infestation and total number of larvae per plant were modeled as a function of harvest time, L. perenne competition, Pyropteron treatments, and initial root mass (as defined above). Second, given that infestation with Pyropteron was confirmed and was similar in 2019 and 2020, all further response variables were analyzed for the final harvest in May 2020. Thus, aboveground biomass and final root mass of R. obtusifolius, number of larvae per unit final root mass, number of rosettes per R. obtusifolius plant, number of R. obtusifolius roots, and proportion of root decay each were regressed on L. perenne competition, Pyropteron treatments, and initial root mass. Third, visual inspection and preliminary tests indicated a significant competition × Pyropteron interaction on both aboveground biomass and final root mass of R. obtusifolius plants, meaning that the herbivore effect on R. obtusifolius would differ depending on L. perenne competition. Moreover, a second-order polynomial on initial root mass was needed to appropriately model the data. Therefore, after first testing for the main effects of L. perenne competition, Pyropteron treatment and their interaction, data for aboveground biomass and root mass was split into the L. perenne and no competition treatment to reduce model complexity and allow for a clearer interpretation. Here, predictors were Pyropteron treatments and a linear and quadratic term of initial root mass.

Final models included the two-way interactions of factor variables, while the inclusion of higher order interactions and interactions between factors and (continuous) initial root mass was assessed by the second-order Akaike Information Criterion (AICc56). Inference on main effects was achieved with single term deletion from the main effects model (each effect in turn) and subsequent likelihood ratio tests; interactions were similarly tested, but from a model that contained all interactions in that respective order. The marginal and conditional R2 of final models were calculated following Nakagawa and Schielzeth57 and Nakagawa et al.58. Differences in estimates between the factor levels of variables were tested post-hoc using the Tukey range test59. All data was analyzed with the statistical software R, version 4.2.260, using the glmmTMB package for generalized linear mixed-effects models61 and the multcomp package for Tukey range tests59.