Horsenettle (Solanum carolinense L., Solanaceae) is a common herbaceous weed in its native range of the southeastern United States, as well as in parts of Europe, Asia, South America, and Australia (Bassett and Munro 1986; Imura 2003; Follak and Strauss 2010). Horsenettle spreads vegetatively via perennial lateral roots in disturbed areas, such as pastures and old agricultural fields (Sylvester 1946; Albert 1960; Ilnicki et al. 1962). It reproduces sexually through the production of flowers that mature acropetally on racemes that bear an average of 7–8 flowers each (Wise et al. 2008). Large ramets (i.e. stems) may produce more than a dozen racemes, while small ramets may produce none (Wise et al. 2008). The flowering period for horsenettle in the southeastern U.S. is roughly June through August, while a typical ramet remains in flower for about one month (Wise et al. 2008). Most of the flowers are perfect (i.e. cosexual or hermaphroditic), but a small proportion are male, and this proportion varies among genets (Elle 1998; Wise and Cummins 2007; Wise and Hébert 2010). Horsenettle is obligately out-crossing, and pollen is transferred mainly by buzz-pollinating insects such as bumblebees (Richman et al. 1995; Quesada-Aguilar et al. 2008). Fertilized ovaries may mature into round yellow berries with a diameter commonly between 1 and 2 cm (Wise and Sacchi 1996; Wise and Cummins 2007). Many fertilized ovaries abort, apparently due to a lack of resources available to fill the fruits (Solomon 1985; Steven et al. 1999; Wise 2018).

The most-abundant leaf-feeding specialists (folivores) of horsenettle in the geographic area of this study include the eggplant flea beetle (Epitrix fuscula Crotch, 1873), the false potato beetle (Leptinotarsa juncta (Germar, 1824)), the eggplant tortoise beetle (Gratiana pallidula (Boheman, 1854)), the eggplant leafminer (Tildenia inconspicuella Murtfeldt, 1883), and the eggplant lace bug (Gargaphia solani Heidemann, 1914) (Wise 2007b). In the field in which the current study was performed, flea beetles were the most damaging of the folivores (Wise 2010; Wise and Rausher 2013). Adult flea beetles feed on leaves from plant emergence to senescence, producing a characteristic shot-gun pattern of small holes in nearly every leaf (Wise and Weinberg 2002).

Flowers of horsenettle are also susceptible to high levels of herbivore damage. In a large study including 929 horsenettle plants in the same field site as the current study, plants lost a mean ± 1 standard deviation of 51 ± 20% of their flowers to florivory (Wise and Rausher 2013). Ten percent of the plants lost more than three-quarters of their flowers, while 10% lost fewer than one-quarter of their flowers. The majority of flower damage was caused by two beetle species: the potato bud weevil (Anthonomus nigrinus Boheman, 1843) destroyed a mean of 31 ± 18% of the flower buds per plant, and the false potato beetle destroyed a mean of 12 ± 16% of the flowers per plant (Wise and Rausher 2013). Both of these herbivores are present through the entire flowering period of horsenettle (Tuttle 1956; Wise 2007b; Wise et al. 2008).

This study involved ten horsenettle genets (genetic individuals) that were originally collected as roots in the spring of 1997 from an oldfield population at Blandy Experimental Farm (39°03’43”N, 78°03’49”W) in Boyce, Virginia, USA. These ten genets were part of a larger research program, and more details on propagation methods can be found elsewhere (Wise 2007a).

The field-collected roots were planted in 18.9-liter (5-gallon) pots filled with commercial growing medium (Wesco Growing Media III, Wetsel Seed Company, Harrisonburg, VA, USA). These pots were placed on wooden pallets in full sunlight in a semi-protected propagation area at Blandy Farm, where the plants were allowed to grow through senescence in autumn. The roots were placed in cold storage over winter, and cuttings of new root growth were used to perpetuate the genets through 2002 using similar procedures. The two main goals of these propagation procedures were to generate numerous clonal replicates of the genets for a series of experiments and to purge the plants of potential maternal or carryover effects due to variation in the microhabitats of the original source fields.

From 1 to 4 May 2000, I removed from refrigeration the horsenettle roots that were grown in 1999 to begin the experiment that is the focus of this paper. I created equal-sized root cuttings by dipping a root into a 100-mL graduated cylinder containing 98 mL of water and cutting the root at the point at which it displaced exactly 2 mL of water. I planted 30 root cuttings for each of 10 genets separately in 3.8-L (1-gallon) plastic pots in Wesco Growing Media III. The pots were then placed onto wooden pallets outdoors in a semi-protected propagation area, where I watered them and monitored plant growth for 5–6 weeks. Once a ramet emerged, I attached a Fibe-Air^TM plant sleeve (Kleen Test Products, Brown Deer, WI, USA) to its pot to prevent herbivory. If more than one ramet emerged, I clipped all but the largest down to the soil surface. On 13 Jun. 2000, I selected 20 healthy ramets from each of the 10 genets to participate in the experiment. Some ramets had initiated racemes, but no flowers had opened yet.

The experiment employed a three-way factorial design, with a folivory treatment (four levels) crossed with a florivory treatment (five levels), crossed with plant genet (10 genets). Each of the 20 damage combinations was assigned randomly to one ramet from each genet. Each of the 200 ramets was then randomly assigned to a position where it would be transplanted into an oldfield at Blandy Farm (39°04’00”N, 78°03’40”W) within an existing horsenettle population. The transplanting positions consisted of seven rows, two meters apart, with two meters between positions within each row.

The 200 ramets were transplanted into the field site on 15–17 Jun. 2000. I blended the transplants into the field setting by covering the growing medium with field soil and leaf litter. A plastic plant label was placed in the soil in front of each transplanted ramet, and a flag was staked near each location with the flag colour indicating the folivory treatment.

The insecticide regime was effective at creating a range of folivory levels. With ambient levels of folivory (no spraying), the mean flea-beetle-damage index was 0.74 (Fig. 2A). Spraying twice, four times, and 11 times over the experiment decreased the FB index to means of 0.67, 0.44, and 0.17, respectively (standard errors of these means are shown in Fig. 2).

The simulated-florivory treatments were generally within 1–2% of their targeted means (Fig. 2B). The exception is that the mean percentage of buds lost in the 0% treatment was actually 4%. This discrepancy was due to the fact that a few flowers were accidentally damaged or were pollinated too late to set fruit. Such flowers were treated as “cut”. For the other treatment levels, any accidentally “cut” flowers were compensated for by cutting fewer than the targeted number of remaining flower buds. Such an accommodation was not possible in the 0% treatment group because no buds would have been targeted for cutting in that group.

Folivory had a largely negative effect on seed production (Fig. 3A), with the most-frequently sprayed plants producing a least-squares mean ± SEM of 711 ± 77 seeds, compared to 316 ± 90 seeds for plants experiencing ambient levels of folivory. However, the effect of folivory was noticeably nonlinear, with a substantial decrease in seed production once the FB index exceeded 0.17, and then a levelling off of impact up to the highest FB index of 0.74 (i.e. at ambient damage level). The main effect of folivory (FB index) was not statistically significant when the interaction between the FB index and florivory was included in the ANCOVA model (F_1,114 = 1.7346, p = 0.19, Table 1); however, when that interaction term was not included in the model, the main effect of folivory was found to be statistically significant (F_1,113 = 6.2076, p = 0.014, Table 2). In contrast, simulated florivory caused a statistically significant reduction in seed production even when this interaction term was included in the model (F_1,114 = 46.4658, p < 0.0001, Table 1). The impacts of flower damage on seed production also showed evidence of nonlinearity (Fig. 3B). Specifically, a negative impact of simulated florivory was not evident until the plants lost at least 40% of their flowers. Plants that lost 80% of their flowers produced an average of 41% as many seeds as plants in the 0%-florivory group. The statistical significances of the folivory and florivory main effects must be interpreted with caution, however, because the interaction between the main effects was significant (F_1,114 = 4.2726, p = 0.041, Table 1).

When flower damage was considered to be caused by two different species of simulated florivores, damage by either florivore alone caused significant reductions in log-transformed seed production (Table 2). However, the reduction when both were modelled to feed together differed significantly from what would be predicted by adding together their individual impacts (Florivore 1-by-Florivore 2 interaction: F_1,113 = 11.2746, p = 0.0011, Table 2). The same ANCOVA run on the untransformed seed numbers provided a very similar inference regarding the interaction (F_1,113 = 5.2287, p = 0.024). The negative value of the coefficient of the interaction term (Table 2) indicates that the impacts combined synergistically, consistent with the prediction of Hypothesis 1. For example, at mean levels of florivory by both simulated species (i.e. 20% each), a plant would lose 59 more seeds than would be predicted by the additive model. Moreover, at maximum damage levels of both species (40% each), a plant would lose 238 more seeds than an additive model would predict.

The effect on log-transformed seed production that was caused by flower damage depended on the amount of leaf damage the plants experienced, and vice versa. In other words, the individual impacts of the two types of damage combined in a nonadditive fashion, as evidenced by the significant interaction between the proportion of flower buds cut and the FB index (F_1,114 = 4.2726, p = 0.041, Table 1). Notably, the same ANCOVA run on the untransformed seed numbers provided a nearly identical inference regarding the interaction (F_1,114 = 5.6349, p = 0.019). The positive value of the coefficient of the interaction term (Table 1) indicates that the impacts combined in a subadditive manner, consistent with the prediction of Hypothesis 2. For example, at mean levels of flea beetle folivory and flower damage, a plant would lose 107 fewer seeds than would be predicted from the sum of the individual impacts that the same levels of folivory and florivory would have had if the herbivores were feeding separately (i.e. from an additive-impact model). Moreover, at maximal levels of both types of herbivory, a plant would lose 347 fewer seeds than the additive model would predict.

Flower damage by one simulated species of florivore had a statistically significant negative impact on horsenettle’s fitness only at the highest level of flower damage by the other simulated species of florivore (Fig. 4, Table 3). Specifically, the identical amount of damage by Florivore 1 had either negligible impact, slight but non-significant impact, or highly significant impact on horsenettle’s fitness on ramets that respectively had either 0%, 20%, or 40% of its flowers damaged by Florivore 2 (Table 3).

Simulated florivory had by far its most significant negative impact when leaf damage was at its lowest—that is, with the most frequent spraying of insecticide (Fig. 5A, Table 4). At higher levels of leaf damage, flower damage had less of an impact, and the impact was only marginally significant at the medium and highest levels of leaf damage (Table 4). Similarly, leaf damage had a statistically significant negative impact on horsenettle’s fitness only when there was no flower damage (Fig. 5B, Table 5). Leaf damage had a slightly lower, but only marginally significant impact when 20% of the flowers were destroyed by simulated florivory. Once floral damage reached 40% or greater, leaf damage had essentially no impact on horsenettle’s fitness (Table 5).

This result can be explained by the Limiting Resource Model (LRM) of plant tolerance with a consideration of source-sink dynamics (Geiger and Servaites 1991; Wise and Abrahamson 2005). Specifically, seed production in the absence of herbivory was apparently not sink-limited; that is, plants produced more flowers than could be matured into fruits, so flowers were not a limiting “resource”. In particular, the plants could tolerate the loss of up to 20–40% of their flowers to simulated florivory without a significant reduction in seed production. The amount of damage caused by just one species of florivore was not severe enough to cause the seed production to become sink limited. However, the combined damage by two species caused the plants to become sink limited, thus crossing the threshold at which plants could no longer tolerate the combined damage of two florivores without a concomitant reduction in seed production.

This pattern of synergistic impact of two herbivores feeding on the same resource is envisaged to result from a nonlinear relationship between damage level and fitness (Gould 1988). In other words, rather than each unit of damage causing the same decrement in plant fitness, the decrements depend on how much damage the plants have already suffered. Specifically, if plants can completely tolerate small amounts of damage, then the decrements caused by a unit of damage may be zero when there is little cumulative damage, but nonzero once cumulative damage levels pass the tolerance threshold. When viewed with this perspective, it becomes clear that the synergistic impact of two herbivores is a result of the same phenomenon that can cause a nonlinear relationship between the damage level of a single species of herbivore and its host-plant’s fitness. Synergistic impact may be common in nature simply because two (or more) species of herbivores that feed on the same type of tissue tend to cause more damage than just one herbivore, and thus the combined feeding is more likely to overwhelm the plant’s ability to tolerate the damage.

If the plants had already been sink-limited in the absence of florivory, then any amount of flower damage would have reduced seed production—that is, there would have been no tolerance threshold. If the damage-fitness relationship was linear, then each unit of herbivore damage would cause an equal fitness decrement, regardless of the cumulative amount of damage. As such, the combined impact of two species of florivores would be expected to be equal to the sum of the individual impacts of each herbivore species in isolation. Such an additive scenario of combined impact is likely to be quite common in nature when two or more species of herbivores feed on the same type of tissue and thus affect the acquisition of the same, limiting resource. The current study did not test this additive-impact hypothesis, however. Such a test would have required the plants to be sink (flower) limited in the absence of florivory. As detailed above, seed production in the experimental plants did not become sink limited until the damage level exceeded ~40% of the flowers.

The use of manually simulated damage in this experiment made the connection between nonlinearity, tolerance thresholds, and synergistic impact even clearer. That is, the damage could be envisioned as having been caused by just one hypothetical species of herbivore. The plants in this experiment tolerated damage ranging up to about 40% of flower buds with only a negligible loss of seed production; however, above that 40% threshold, seed production was reduced precipitously. Thus, there was an overall nonlinear relationship between florivory and plant fitness. However, a study of natural herbivory in this same field site revealed that the two most-damaging horsenettle florivores (the potato bud weevil, and the false potato beetle) destroy an average of only 31% and 12% of the flowers, respectively, while other florivores combined to destroy an average of 8% of the flowers (Wise and Rausher 2013). None of these florivores feeding alone would commonly cause the plants to cross the tolerance threshold. However, feeding by any one species in effect lowers the tolerance threshold for feeding by any of the other species of florivores, making synergistic impact increasingly likely in multiple-herbivore communities.

Although manually simulated herbivory has some advantages in experimental studies (Tiffin and Inouye 2000; Lehtilä 2003; Lehtilä and Boalt 2004), additional insight would be gained by studying actual herbivory. In this system, the two main florivorous species of horsenettle are beetles whose chewing destroys the flowers rapidly, and their damage tends to be spread across racemes. However, the potato bud weevils kill flowers in the bud stage by chewing through the pedicels after laying eggs in the buds, while false potato beetles consume flowers either in the bud stage or after a flower is open (Wise 2007b). In addition, meadow voles often destroy entire racemes by chewing their peduncles, and damage by larvae of the moth Frumenta nundinella (Zeller, 1873) (Gelechiidae) causes the maturation of parthenogenetic, seedless fruits (Wise 2007b). While the manner in which flower damage was simulated in this study may do a fair job at representing the general impact of the loss of sinks, a next step would be to investigate how subtleties in the damage patterns and manners of feeding of the different species might influence plant responses vis-á-vis the plant’s tolerance thresholds.

The results of this experiment were also consistent with the prediction of Hypothesis 2: the impacts of the two different types of herbivore damage combined in a subadditive fashion to affect horsenettle’s seed production. In other words, horsenettle expressed greater tolerance of each type of damage when damage by the other herbivore was relatively high than when damage by the other herbivore was low or absent. This pattern can also be explained by the LRM. Specifically, leaf feeding (folivory) likely decreased the supply of carbon available for seed production. With increasing levels of folivory, horsenettle’s reproduction became relatively more source (leaf) limited and less sink (flower) limited. With more folivory, plants could tolerate a higher loss of flowers to herbivores because a greater proportion of the flowers would have aborted anyway due to a lack of photoassimilates needed to fill fruits.

It is worth re-emphasizing that the regression slopes of relative fitness on damage in the analyses of this paper should not be interpreted as natural-selection differentials (or gradients) acting on resistance because the damage levels were not the result of any particular plant trait. Instead, the damage levels were assigned randomly by the experimenter. As such, the nonadditivity of impact found in this study cannot provide direct evidence of nonadditive selection acting on resistance, and thus it cannot provide evidence that diffuse (co)evolution is occurring in a natural horsenettle population. Nevertheless, the results of the ANCOVAs do show that the impact of damage by one herbivore can depend on the amount of damage caused by another herbivore (without influencing the amount of damage caused by the other herbivore). Moreover, the regression analyses clearly show that subadditive and synergistic impacts have opposite effects on the damage-fitness relationship—a relationship that is integral to the magnitude of a selection differential or gradient.

In a more natural setting, allocation and ecological costs of resistance traits may reduce the benefits of a reduction in damage and thus make it more difficult to detect the effects that nonadditivity of impact may have on selection for resistance. The advantage of a more controlled, factorial experiment is that it removes potentially obscuring factors from the picture and thus provides a more transparent view of how damage levels themselves affect plant fitness. This clearer view shows that nonadditivity may be a result of something as simple and fundamental as the balance of sources and sinks in plant reproduction. Such a fundamental balance would be expected to apply not only in a factorial experiment, but in nature as well. Thus, nonadditivity of combined fitness impacts of two herbivores on a shared host plant will contribute a diffuse component to selection for resistances to the herbivores, even if countervailing factors cause the overall gradients for selection on resistance to lose the signal of nonadditivity. In this sense, it is reasonable to conclude that nonadditivity of impact is a sufficient (though not necessary) criterion for diffuse (co)evolution in plants attacked by multiple species of herbivores.

The regression analyses suggest that nonadditivity of combined impact results in alterations of the strength of selection for resistance in shared host plants. In combination with other factors—such as amount of genetic variation for resistance traits, genetic correlations among resistance traits, and ecological interactions among herbivores—the manner in which fitness impacts combine will affect the expected evolutionary trajectory of resistance. Specifically, if the fitness impacts of multiple herbivores combine in a synergistic fashion on a shared host plant, the pace at which the plant population evolves resistance will accelerate (as long as the plant population possesses genetic variation for resistance mechanisms and the cost of resistance is not prohibitive). In contrast, if the fitness impacts combine in a subadditive manner, then the pace of evolution of resistance to at least some of the herbivores is likely to be slowed. Such evolutionary effects on resistance would then have feedback effects on the ecology and behaviour of the herbivores, which would affect the selective regime for resistance traits in the plant population, and so on.

Previous research on horsenettle and its herbivores provides a basis for speculating on implications of the current results to the ecoevolutionary dynamics of this system. Consider, for instance, the horsenettle-specialist moth Frumenta nundinella, which in a large field study destroyed an average of only ~3% of horsenettle ramets’ flowers (Wise and Rausher 2013). On its own, a minor florivore such as F. nundinella probably has negligible impact on horsenettle’s fitness, and thus there is likely to be no selective advantage for increased resistance to these florivores in horsenettle populations located near the study area. However, if potato bud weevils are also present, then horsenettle is likely to face much stronger selection for resistance against florivory in general. Thus, F. nundinella may be at a disadvantage in the coevolutionary arms race with horsenettle due to selection for increased resistance to florivory brought about by damage caused by potato bud weevils. The survival and fecundity of individuals of F. nundinella may decrease, or their behaviour may change to focus on less-resistant plant species. Therefore, indirect ecological effects (namely, synergistically combining fitness impacts) would lead to evolutionary changes that cause further ecological effects in the plant-herbivore community.

Subadditively combining impacts are more likely to have the opposite effect, slowing the plant’s evolution of resistance against members of its herbivore community. Consider that natural populations of horsenettle are often heavily damaged by folivory, especially by flea beetles and potato beetles, but also by lace bugs, leaf-mining caterpillars, tortoise beetles, blister beetles, and others (Wise 2007b). Controlled experiments indicate that feeding by at least the first three species can cause significant impact on horsenettle’s reproduction (Wise and Sacchi 1996; Wise and Cummins 2006; Wise and Mudrak 2021), and thus we might expect strong directional selection to increase resistance to these folivores. However, selection for resistance against leaf-feeding species was found to be relatively weak in a large field study on horsenettle (Wise and Rausher 2013). In that field study, florivores destroyed an average of roughly half of the plants’ flowers, and selection for resistance against the florivores was quite strong. The findings of the current factorial experiment suggest that the fitness effects of the leaf feeders in that field experiment was negligible because the flower feeders caused reproduction in the plants to be limited by the number of sinks (flowers) rather than the supply of sources (photosynthates) (Wise and Rausher 2016).

In the face of strong selection for increased resistance against florivory, some factors must be constraining an evolutionary response to that selection for damage levels to flowers to remain so high. Most simply, horsenettle populations may possess a rather small amount of genetic variation for resistance against floral herbivory (Wise 2007a). If that is the case, before the plant population can begin to evolve resistance to leaf feeders, the plant population must wait for one of the following phenomena: for mutations that increase resistance to floral herbivory, for gene flow from populations with different alleles for florivory-resistance traits, or for a reduction in the florivore population due to natural enemies or an abiotic disturbance. Horsenettle’s evolution of resistance to potato bud weevils may also be constrained by internal (e.g. allocation) costs of resistance (Wise and Rausher 2016), negative genetic correlations with resistances to folivory by flea beetles and frugivory by meadow voles (Wise and Rausher 2013), and competitive interactions that likely result in the flowers that are saved from the weevils ending up being lost to feeding by potato beetles and meadow voles (Wise 2009).

With all of these potential genetic and ecological constraints, horsenettle’s evolution of resistance against potato bud weevils may be at a stalemate. Such a stalemate would allow other species of herbivores to maintain large populations on horsenettle, unimpeded by selection for increased resistance. In this manner, subadditivity of combined-herbivore impact may play an important role in maintaining the persistence of the diverse multiple-herbivore community of horsenettle.