Summary
Herbicide drift exposes nontarget plants to sublethal agrochemical doses, yet its ecological and evolutionary consequences remain poorly understood. Although hormesis – stimulatory responses to low doses of otherwise toxic compounds – has been documented following herbicide exposure, it has rarely been examined within an evolutionary ecology framework. Here, we examine herbicide‐induced hormesis in the agricultural weed Oxalis stricta across two field experiments.
We exposed replicated maternal lines to dicamba drift and quantified growth, reproductive traits, floral allocation, and pollinator visitation. Dicamba drift consistently increased flower production in both years, revealing a plastic shift toward reproductive allocation. Fitness consequences were context‐dependent: in 2022, increased flowering was associated with higher seed production through indirect, trait‐mediated pathways, whereas in 2023, increased flowering did not translate into detectable differences in reproductive output.
We detected little genetic variation in hormetic responses, suggesting limited potential for adaptive evolution. Dicamba drift also reduced individual flower size, indicating a shift toward larger floral displays composed of smaller flowers. These allocation shifts reduced pollinator visitation primarily through changes in flower number, linking herbicide exposure to trait‐mediated changes in plant–pollinator interactions.
Together, our results show that low‐dose herbicide exposure can generate compensatory responses that reshape ecological interactions, even when evolutionary responses are constrained.
Keywords: dicamba, eco‐evolutionary dynamics, herbicide drift, hormesis, overcompensation, reproductive allocation, tolerance
Conceptual framework linking herbicide dose–response to plant fitness outcomes under sublethal exposure.

Introduction
Hormesis, defined as the stimulatory effect of a subtoxic dose of a chemical that is toxic at higher doses (Calabrese et al., 2007), has been documented across a wide range of plant stressors, including pollutants and agrochemicals. In plants, hormetic responses are typically expressed as increased growth, reproduction, or altered allocation to specific traits following sublethal exposure to a chemical (Duke et al., 2005; Belz et al., 2011; Belz & Duke, 2017). Because these responses directly affect traits linked to fitness and species interactions, hormesis has the potential to influence both population dynamics and evolutionary trajectories. However, hormesis has rarely been examined through an evolutionary ecology lens and remains poorly integrated into frameworks linking sublethal stress exposure to trait evolution and downstream consequences for species interactions.
Hormesis is likely to be particularly important in agricultural landscapes, where nontarget plants are frequently exposed to herbicide drift via the off‐target movement of agrochemical droplets or vapor following application. Sublethal exposures to herbicides such as glyphosate and acetolactate synthase (ALS) inhibitors have been shown to elicit hormetic responses in various weed species (reviewed in Belz et al., 2022; Duke et al., 2025), with typical responses including an increase in biomass, plant size, or fitness‐related traits by 9–145% compared with nonexposed controls (Duke et al., 2025). Studies of low‐dose herbicide exposure also report highly variable effects that depend on species identity, herbicide formulation, and trait measured. Importantly, sublethal exposure can alter reproductive and phenological traits even when overall biomass or reproductive output is unaffected. Because sublethal herbicide exposure can enhance fitness‐related traits and alter phenology, hormesis has been hypothesized to contribute to shifts in weed community composition (Belz & Duke, 2014, 2017; Duke et al., 2025), the evolution of herbicide resistance (Belz & Duke, 2014; Belz et al., 2022), and changes in interactions with pollinators and other community members (Guedes et al., 2022; Iriart et al., 2022). However, empirical tests linking hormetic trait responses to longer‐term evolutionary outcomes remain limited, as does our understanding of how hormetic responses might alter interactions with other community members, such as pollinators.
As such, key questions remain regarding the ecological and evolutionary significance of hormesis. It is unclear whether hormetic responses are consistent across years, as dose–response outcomes can depend on environmental conditions that vary across growing seasons; whether they vary genetically within species; or whether exposure to low‐dose herbicide stress could drive evolutionary change through selection on hormetic responses or underlying allocation strategies. In addition, because hormesis can simultaneously alter multiple traits that contribute to fitness and species interactions–such as growth, reproduction, and floral display–it remains unknown whether sublethal herbicide exposure reshapes ecological outcomes indirectly by modifying how plants interact with pollinators.
These questions can be addressed using a well‐developed framework from evolutionary ecology focused on tolerance and overcompensation (Paige, 1987; Ramula et al., 2019), which provides a conceptually aligned approach. In this framework, tolerance is defined as the ability to maintain fitness under stress relative to benign conditions, regardless of damage incurred (following Strauss & Agrawal, 1999; Agrawal, 2000; Baucom & Mauricio, 2004; Iriart et al., 2022). As illustrated in Fig. 1(a), tolerance encompasses a range of fitness responses to stress. Undercompensation occurs when fitness declines under stress relative to control conditions, whereas complete tolerance reflects equivalent fitness across environments. Overcompensation, in which fitness increases under stress, represents a direct analogue to hormesis, capturing cases where low‐dose exposure stimulates growth or reproduction rather than suppressing it. Importantly, genetic lines may also differ in both baseline fitness and the direction or magnitude of their response to stress, revealing genetic variation in compensatory strategies and the potential for evolutionary change.
Fig. 1.

Conceptual framework linking herbicide dose–response to plant fitness outcomes under sublethal exposure. (a) Reaction norm panels showing the range of possible responses in fitness or fitness‐related traits across control and drift environments for multiple genetic lines (represented by lines of varying shading). Undercompensation: all lines show reduced fitness under herbicide relative to control plants, with lines differing in fitness. Tolerance: all lines maintain fitness across environments. Overcompensation: all lines exhibit increased fitness under herbicide exposure, again varying in fitness but not in the direction of response. Mixed response: genetic lines differ both in baseline fitness and in the direction and magnitude of their response, illustrating genetic variation for plasticity to sublethal herbicide exposure. (b) Biphasic dose–response curve illustrating the hormetic relationship between herbicide dose and fitness. At drift‐relevant doses (filled circle), fitness exceeds the no‐herbicide control plants, consistent with overcompensation; at agronomic field doses (open circle), fitness declines sharply below control plants. The dashed line indicates fitness in the absence of herbicide.
The extensive literature on overcompensation – most often studied in the context of herbivory damage – has generated a robust foundation linking tolerance, allocation, and evolutionary potential (Paige, 1987; Agrawal, 2000; Ramula et al., 2019). Hormesis following herbicide exposure, we argue, is a parallel instance of this phenomenon: overcompensation triggered by chemical stress rather than tissue loss. This reframing lets hormetic responses be treated as measurable traits with fitness consequences, heritable variation, and potential eco‐evolutionary feedbacks under sustained sublethal herbicide exposure. Yet despite the conceptual overlap, hormesis has not been examined through the evolutionary‐ecology lens of tolerance and overcompensation – a gap this study addresses directly.
Here, we use the framework of tolerance and overcompensation from evolutionary ecology to examine potential herbicide‐induced hormesis in the weed Oxalis stricta. Prior glasshouse and field studies show that O. stricta survives and often tolerates the herbicide dicamba at drift doses: plants exposed to dicamba drift showed no detectable reduction in growth or flower production (Iriart et al., 2022; Baucom et al., 2025), although visible injury still occurred (Baucom et al., 2025). Our work also suggests overcompensation at drift rates: we previously detected a significant increase in flowering duration following dicamba drift exposure in glasshouse settings (Iriart et al., 2022), and unpublished observations show elevated flower production at drift doses of dicamba in the field (Baucom et al., 2025). By contrast, the field dose of dicamba (561 g ae ha−1) causes complete mortality in field settings (Baucom et al., 2025), consistent with a biphasic, hormetic response in which low doses stimulate growth or reproduction while higher doses become increasingly toxic and ultimately lethal (Fig. 1b). The relevance of variation in hormetic responses has grown with the rapid expansion of dicamba‐resistant crop systems in the United States (US EPA, 2021), which has increased off‐target drift risk (Soltani et al., 2020). As a synthetic auxin herbicide, dicamba disrupts core developmental pathways, and even sublethal exposure may generate phenotypes that vary with factors such as dose, plant developmental stage, or growing conditions in non‐target plants.
We address four related questions. First, does O. stricta exhibit increased growth, flower production, or reproductive output under dicamba drift, consistent with overcompensation or hormesis, and are such responses consistent across years? Second, do these responses vary among maternal lines, indicating genetic variation and potential for evolutionary change? Third, does dicamba drift alter patterns of floral allocation, such as trade‐offs between flower number and per‐flower investment (e.g. flower size)? Finally, because herbicide drift may simultaneously influence multiple components of floral display, and because the consequences for pollinator behavior remain poorly understood, we ask whether dicamba drift alters pollinator visitation, and whether any changes are mediated primarily through flower number, flower size, or their combined effects. To integrate hormesis within an evolutionary ecology framework, we test for genotype × environment interactions in fitness and use structural equation modeling (SEM) to evaluate whether dicamba influences fitness and pollinator visitation directly or indirectly through correlated growth and floral allocation pathways. Together, this approach allows us to assess both the evolutionary potential of hormetic responses and the potential mechanistic links connecting drift exposure to plant fitness and plant–pollinator interactions.
Materials and Methods
Study system
Oxalis stricta L. (wood sorrel; Supporting Information Fig. S1) is a wide‐ranging, short‐lived perennial plant that is native to both North America and Eurasia and is commonly found in agricultural fields, roadsides, cosmopolitan areas, and as herbaceous understory vegetation in temperate forests (Rokhlova, 2013; Imamura & Hariu, 2020). Although O. stricta exhibits the morphological remains of a tristylous breeding system, the species is a facultative self‐pollinator and is also able to reproduce vegetatively through rhizome production (Doust et al., 1985; Marshall, 1987). The species is commonly visited by syrphid flies and honeybees largely for nectar and pollen rewards (see Results). Though O. stricta mostly self‐pollinates, with viable seed produced as a result, it is also able to produce seed when outcrossing and does not undergo apomixis (Doust et al., 1981, 1985). The leaves of O. stricta contain oxalic acid and are often added to food by humans to impart a sharp flavor (Doust et al., 1981, 1985). The species is considered a problematic weed in nurseries, turf, and agriculture (Doust et al., 1981, 1985; Marshall, 1987) and is relatively difficult to remove via both mechanical and chemical means (Marble et al., 2013).
Seed collection
We collected seeds from 10 populations of O. stricta growing near soybean or fallow fields in Kentucky and Tennessee, USA (Fig. S1) in the summers of 2018 and 2019 for field experiments. At each collection site, seeds from individual plants were sampled every 1 m along a 20 m transect.
Field Experiment 1
We performed two field experiments to examine the potential for trait level (e.g. increased growth or flowering) responses and how that may or may not lead to overcompensation at the fitness level (e.g. seed or fruit production). Throughout this study, we distinguish hormetic responses expressed at the trait level (e.g. increased growth or flowering) from overcompensation at the fitness level, recognizing that enhanced trait expression does not necessarily translate into increased reproductive output.
For the first field experiment, on 14 June 2022, we planted replicate seeds from 31 maternal plants sampled from natural populations in a common garden at Matthaei Botanical Gardens (Table S1). Four replicate seeds per maternal line were planted across two blocks, with two treatments (control, dicamba drift) randomly assigned to subplots within each block (4 replicates × 2 blocks × 2 treatments × 31 maternal lines = 496 individuals); all plants within a given subplot received the same dicamba treatment. We additionally germinated seeds from each maternal line in a growth chamber in the Biological Science Building's plant growth facility under an 18 h light : 6 h dark photoperiod at 26°C and replaced nongerminants in the field with transplants on 18 July 2022. Individuals that never germinated or did not survive transplant were excluded from analyses.
After 22 d of growth in the field (9 August 2022), plants were sprayed with either a control or herbicide treatment. Treatment plants were sprayed with a dicamba solution corresponding to c. 0.7% of the recommended field dose (561 g ae ha−1), defined based on the equivalent proportion of the labeled field‐rate tank concentration and representing a particle drift exposure (Felsot et al., 2011), plus 0.1% (v/v) Preference surfactant (nonionic surfactant blend, WinField Solutions, St. Paul, MN, USA). Control plants were sprayed with Preference at the same concentration (0.1% v/v) to control for any effect of the surfactant carrier independently of dicamba. Plants were sprayed in a single pass until just wet using a handheld multipurpose sprayer set to a medium‐fine mist at an operating pressure of 40–45 PSI. We used a handheld sprayer rather than a CO2‐pressurized sprayer with a constant application rate to better simulate the heterogeneity characteristic of off‐target herbicide drift under field conditions. Because applications were made using a handheld sprayer rather than a calibrated boom system, exposure is reported as a proportion of the recommended field dose rather than as an absolute area‐based application rate.
We recorded the number of true leaves for each plant immediately before herbicide treatment and 3 wks posttreatment. We then calculated plant growth as prespray leaf number subtracted from leaf number 3 wks postspray, divided by the number of days between leaf number collection (21). We likewise recorded the number of leaves exhibiting dicamba damage – that is, leaf cupping or yellowing (Foster & Griffin, 2018) – for an estimate of herbicide damage. We recorded the date of first flowering for each plant (in Julian dates) and recorded the number of open flowers per plant twice a week until the first hard frost, which occurred on 17 November 2022. We collected ripe fruits from plants in twice‐weekly collection rounds beginning in September and continuing until the first frost. We weighed the total seed mass for each individual and the mass of a subsample of 100 seeds from that same individual. We then used this known seed mass to convert total seed weight into an estimated seed count for each individual.
Field Experiment 2
On 21 April 2023, we planted seeds from 16 maternal lines drawn from eight populations in the glasshouse at Matthaei Botanical Gardens and maintained plants under ambient glasshouse conditions. These maternal lines were randomly selected as a subset of those used in Field Experiment 1 (Table S1). Replicate seeds were obtained either from plants grown in a controlled growth‐room environment or from plants grown in the 2022 field experiment, resulting in multiple parental environments represented among experimental individuals. Preliminary analyses indicated that parental environment did not influence overall treatment effects for any trait considered and, as such, we dropped this effect from the final reported models. Once plants produced at least five true leaves, they were transplanted to the field on 13 June 2023. As in the previous field experiment, we established two environmental blocks, each containing both dicamba and control treatments, and planted between six and 24 replicate individuals of each maternal line in every block–treatment combination (16 maternal lines × 2 blocks × 2 treatments × 6–24 replicates (average c. 20 replicates per line) = 689 total individuals).
Approximately 3 wks after transplanting (7 July 2023), plants were exposed to the same dicamba and control treatments used previously. We recorded the same vegetative growth measurements as in the prior year and monitored flower production throughout the experiment. Flower number was recorded repeatedly through early September, and flower width, another component of floral display, was recorded 1 month after spraying (9 August–8 September 2023) by measuring the diameter of five flowers per plant with digital calipers, defined as the distance from the tip of one petal to the tip of the opposing petal. In total, 3430 flowers were measured; mean flower width per individual was used in subsequent analyses. At the end of the season, we counted fruits per plant and used fruit number as our estimate of fitness, as fruit number was positively correlated with total seed production in the earlier field experiment (r = 0.53, P < 0.001).
Pollinator observations began on 6 September 2023 and were conducted using a subset of experimental plants. Individuals within one block were grouped into spatially proximate clusters of 2–10 plants (arrays) that could be observed simultaneously. Ten arrays per treatment were selected, yielding 20 arrays total. Randomly chosen arrays were observed across 7 d (rounds) conducted between September and October 2023. General weather patterns were recorded (sunny, partly sunny, cloudy, partly cloudy). Each array was observed for 20 min per round; the number of open flowers per plant was recorded at the start of each observation, followed by recording the number of pollinator foraging events, defined as instances in which an insect visitor entered a flower. Floral visitors were categorized into functional groups (honeybee, bumblebee, syrphid fly, wasp, and fly). Across rounds, dicamba‐exposed and control plants were observed for a total of 380 and 300 min, respectively.
Data analysis
Analysis of plant traits: Field Experiment 1
All analyses were conducted in R (v.4.3.1; R Core Team, 2023) using R Studio (Posit team, 2024). We analyzed postspray vegetative growth, date of first flower, total flower number, and seed number from the 2022 field experiment. To improve normality and variance structure, postspray growth and flowering date were rank‐normalized using the orderNorm transformation in the BESTNORMALIZE package (Peterson, 2021), and total flower number was square‐root transformed. Seed number, which exhibited strong right skew and heteroscedasticity, was analyzed on its original count scale using a generalized linear mixed effects model with a Tweedie error distribution and log link, implemented in GLMMTMB (Brooks et al., 2017). For postspray growth, flowering date, and total flower number, we fit linear mixed effect models using LMER in the lme4 package (Bates et al., 2011), with treatment specified as a fixed effect and prespray leaf number included as a covariate where relevant.
In all models, random effects were included to account for experimental blocking and genetic structure. Population and maternal line (nested within population) were included as random intercepts, and block was included as a random intercept to account for spatial structure in the field. To test for genetic variation in treatment responses (i.e. maternal‐line‐specific reaction norms), we evaluated an additional random effects structure allowing treatment to vary among maternal lines, implemented as a random treatment slope within maternal line (nested within population). Baseline models took the general form:
with reaction‐norm models additionally including the term (1 + treatment | maternal line:population). Block effects were retained only when they explained nonzero variance. Model fit for all generalized linear mixed effect models was assessed using the DHARMA package (Hartig, 2024). Fixed effect significance was assessed with Satterthwaite degrees of freedom using lmertest (Kuznetsova et al., 2017), and random effects were tested using likelihood ratio tests. Estimated marginal means were determined using EMMEANS (Russell, 2025) and back‐transformed to the original measurement scales for interpretation.
Analysis of plant traits: Field Experiment 2
We applied similar mixed effects modeling approaches to postspray vegetative growth, total flower production for the season, flower number during the first month following spraying, and fruit production in the 2023 field experiment. Because many plants had already initiated flowering before dicamba drift exposure (see Results), we did not analyze treatment effects on the date of first flowering. Models followed the same general structure used for Experiment 1, with treatment specified as a fixed effect and population, maternal line nested within population, and block included as random intercepts.
Continuous response variables (growth and flower number) were analyzed using linear mixed‐effects models. Fruit count was analyzed using generalized linear mixed‐effects models implemented in GLMMTMB (Brooks et al., 2017), with a negative binomial (NB1) error distribution and a log link. Transformations were applied to continuous variables as needed to satisfy assumptions of normality and homoscedasticity. Model fit for all generalized linear mixed effects models was assessed using the DHARMA package (Hartig, 2024). As in Experiment 1, we evaluated models including a random treatment slope among maternal lines to test for genetic variation in treatment responses.
Structural equation models of growth‐ and flower‐mediated overcompensation
We used SEM to evaluate whether dicamba exposure influenced reproductive output directly or indirectly through changes in growth and floral allocation. SEM allows simultaneous evaluation of direct and indirect effects within an integrated causal framework (Lefcheck, 2016; Grace & Irvine, 2020). We adopted a model‐selection approach to SEM (MSA‐SEM), comparing alternative causal structures using AIC values from component models fitted relative to a fully saturated model (Garrido et al., 2022). As with all structural equation models, these analyses evaluate support for alternative, biologically motivated causal hypotheses rather than establishing definitive mechanistic pathways (Methods S1).
SEMs were fit separately for each field experiment using the PIECEWISESEM package in R (Lefcheck, 2016) to allow for year‐specific differences in reproductive traits while maintaining a similar model structure. For both years, dicamba treatment was specified as a predictor variable, with hypothesized effects on postspray vegetative growth and flower number. Growth was modeled as a predictor of flower production, and reproductive output (seed number in 2022; fruit number in 2023) was modeled as a function of downstream allocation traits. Block was included as a random effect in all component models.
Because reproductive output differed in distribution between years, we used year‐specific error structures. In 2022, seed number was modeled using a Tweedie generalized linear mixed model with a log link, whereas in 2023, fruit number was modeled using a negative binomial (NB1) distribution. Growth and flower number were modeled using linear mixed effect models. Continuous predictors were standardized to facilitate comparison of effect sizes. We manually estimated standardized coefficients for non‐Gaussian responses on the link scale, following established recommendations for generalized responses (Schielzeth, 2010).
For each year, we compared a small set of biologically motivated SEMs that differed in whether reproductive output was mediated by flower number, postspray growth, or both, and whether direct effects of dicamba on reproduction were included (Fig. S2). Model selection was based on tests of directed separation and comparison of AIC values. Full details of candidate models, model comparisons, and goodness‐of‐fit statistics are provided in the Supplemental Methods.
Floral allocation given dicamba drift
We analyzed variation in flower width to test whether dicamba drift altered per flower allocation. Flower width was averaged per individual and square‐root transformed to improve normality. Linear mixed effect models were fit with treatment as a fixed effect, with population and maternal line modeled as random effects. To test whether maternal lines differed in their response to dicamba exposure, we included a random treatment slope among maternal lines, allowing maternal‐line‐specific reaction norms. The final model structure was:
Block was evaluated in preliminary models but excluded because the effect explained negligible variance. Fixed effects were assessed using Satterthwaite‐approximated F‐tests (lmertest; Kuznetsova et al., 2017) and type III Wald chi‐squared tests (Fox & Weisberg, 2018), and random effects were evaluated using likelihood‐ratio tests (Bates et al., 2011). Estimated marginal means were obtained from the final model and back‐transformed for presentation.
To test whether dicamba drift altered floral allocation strategies, we modeled average flower width as a function of flower number, treatment, and their interaction using a linear mixed effects model. Maternal line was included as a random effect. Fixed effect significance was assessed using type III tests (Fox & Weisberg, 2018).
Pollinator visitation
We analyzed pollinator visitation using generalized linear mixed effect models to account for the nonnormal and right‐skewed distribution with many zero values. Visitation rate was quantified as the number of pollinator visits per open flower per minute. Models were fit using the GLMMTMB package in R (Brooks et al., 2017) with a Tweedie error distribution and a log link function. Dicamba treatment was included as a fixed effect. To account for repeated observations and hierarchical structure in the data, population, observation round, weather category, and experimental array (array × treatment combination) were included as random effects. Model assumptions were assessed using residual diagnostics implemented in the DHARMA package (Hartig, 2024). Fixed effects were evaluated using type III Wald chi‐squared tests. For interpretation, estimated marginal means were back‐transformed to the response scale using the EMMEANS package (Russell, 2025), yielding visitation rates expressed as visits per flower per minute with associated 95% confidence intervals.
Structural equation models of floral trait–mediated pollinator visitation
We used piecewise SEM to evaluate whether dicamba drift influenced pollinator visitation directly or indirectly through changes in floral traits. Models were fit using the PIECEWISESEM package in R (Lefcheck, 2016) and constructed from a set of component models. Dicamba treatment was specified as an exogenous driver with potential direct effects on flower width, flower number, and pollinator visits (visits/min). The goal of this analysis was to determine whether any indirect effects of dicamba on visitation were mediated primarily through changes in individual flower size (flower width) or through changes in floral display size (flower number).
Based on preliminary analyses and tests of directed separation, we evaluated models that differed in whether flower width influenced flower number and whether flower width exerted a direct effect on pollinator visitation (Methods S2). Flower width and flower number were modeled using linear models, while pollinator visit time, which contained many zero values and was strongly right‐skewed, was modeled using GLMMTMB (Brooks et al., 2017) with a Tweedie error distribution and a log link function. Models were compared using tests of directed separation, Fisher's C, and summed AIC values to assess support for alternative pathways. For non‐Gaussian components, predictor‐standardized effects were used to facilitate comparison of effect sizes.
Results
Hormetic responses
Dicamba drift induced hormetic responses in some but not all traits (Fig. 2; Table S2). In Field Experiment 1 (2022), growth increased by c. 44% in drift‐treated plants (F 1,171.13 = 4.62, P = 0.033; Fig. 2a), such that drift‐treated plants produced, on average, 4.46 postspray leaves compared with 3.10 leaves in control plants. Similar to growth, total flower number increased under dicamba drift, rising by c. 46% (139 vs 95 flowers in drift vs control; F 1,121.01 = 12.23, P = 0.001; Fig. 2c). The increase in flower production was evident across the majority of sampling dates (Fig. 3a). Flowering phenology (day of first flower), however, showed no difference between treatments (F 1,106 = 0.50, P = 0.479; Fig. 2b). Interestingly, we found an c. 18% increase in seed number in treated plants (1.9 × 105 vs 1.6 × 105 seeds produced on average in drift vs control; Fig. 2d), but this effect was not significant (F 1 = 0.58, P = 0.446), indicating that the variance in seed number among individuals was likely high. In fact, the coefficient of variation for seed number in the drift environment was 1.05, whereas the coefficient of variation for flower number was lower (0.58).
Fig. 2.

Effects of dicamba drift on the growth and reproductive output of Oxalis stricta assessed across two field experiments. Trait means (large, filled circles) ± SE (error bars) are shown for control (blue) and dicamba drift–exposed plants (yellow). Individual data points (open circles) represent all measured plants from each experiment. Statistical results (ANCOVA) are provided in Supporting Information Table S1. Panels show (a) growth rate in Experiment 1 (2022); (b) date of first flowering (Julian day; Exp. 1); (c) total seasonal flower number (Exp. 1); (d) total seed production (thousands; Exp. 1); (e) growth rate in Experiment 2 (2023); (f) total seasonal flower number (Exp. 2); (g) flower number within the first month following dicamba drift exposure (Exp. 2); (h) total fruit number (Exp. 2).
Fig. 3.

Seasonal flowering dynamics of Oxalis stricta given dicamba drift exposure. Daily flower production for control (blue) and dicamba drift–exposed (yellow) plants in (a) Experiment 1 (2022) and (b) Experiment 2 (2023). Open circles show the number of flowers produced per plant on each census day, and solid lines represent daily mean flower production for each treatment. The red vertical line marks the date on which drift‐dose dicamba was applied to treated plants (9 August 2022 in Experiment 1; 7 July 2023 in Experiment 2).
In comparison, in Field Experiment 2 (2023), we found no evidence for a treatment effect on growth (F 1,676.05 = 0.02, P = 0.883; Fig. 2e), but like that of Field Experiment 1, we again found that dicamba drift increased the total number of flowers produced across the season (F 1,670.29 = 8.18, P = 0.004; Fig. 2f). This effect appeared strongest in the first month postspray (Fig. 3b), with drift‐exposed plants producing c. 29% more flowers, on average, during the first month after exposure (180 vs 140 flowers; F 1,671.53 = 45.43, P < 0.001; Figs 2g, 3b). We found no evidence that fruit number differed between treatment environments (χ 2 = 0.64, P = 0.426; Fig. 2h), mirroring our findings examining seed number in the prior experiment.
SEM of reproductive output
We used piecewise SEM to test whether dicamba exposure influenced reproductive output indirectly through correlated changes in growth and floral allocation across two field experiments (Fig. 4; Table S3). In both years, the best‐supported models converged and adequately captured the hypothesized causal relationships based on tests of directed separation and relative model fit.
Fig. 4.

Structural equation models describing the effects of dicamba drift on plant growth and reproduction of Oxalis stricta across two field experiments. Path diagrams show the hypothesized causal relationships among dicamba drift treatment, plant growth, and reproductive output in Field Experiment 1 (a) and Field Experiment 2 (b). Boxes represent observed variables and arrows indicate directional causal paths. Standardized path coefficients are shown above each arrow; significant paths are displayed as solid black arrows, whereas nonsignificant paths are indicated with grey dotted arrows. The r 2 values within each response variable denote the proportion of variance explained by the full model. AIC‐based comparisons of alternative pSEMs are reported in Supporting Information Table S3. Direct, indirect, and total effects for all pathways are provided in Table S4.
In Experiment 1 (2022; Fig. 4a), dicamba exposure increased reproductive output entirely through indirect effects on floral allocation. Dicamba positively influenced flower number both directly and indirectly via enhanced postspray growth, and flower number was the dominant predictor of seed production. Once flower number was included in the model, neither dicamba nor growth exerted detectable direct effects on seed output. Standardized path coefficients revealed two positive indirect pathways linking dicamba to seed production: a growth‐mediated pathway (dicamba → growth → flowers → seeds; 0.39 × 0.59 × 0.68 = 0.16) and a direct floral allocation pathway (dicamba → flowers → seeds; 0.48 × 0.68≈0.33). Together, these pathways produced a substantial positive total indirect effect (0.49), indicating that hormetic responses in 2022 were driven by increased floral allocation rather than direct enhancement of seed production.
In Experiment 2 (2023; Fig. 4b), dicamba effects on reproduction were absent. Although dicamba increased flower number, neither flower number nor postspray growth mediated any effect of dicamba on fruit production. Both indirect pathways linking dicamba to fruit output – via flower number (0.22 × 0.02≈0) and via a direct dicamba → fruit pathway (0.01) were negligible, yielding a total effect of zero. Fruit production was instead associated with growth independently of treatment, and models including dicamba–reproduction pathways were unsupported. Thus, in 2023, dicamba altered floral display without translating into increased reproductive output, highlighting strong context dependence in the expression of hormetic responses.
Genetic variation in fitness‐related traits
We found little evidence for genetic variation in treatment‐specific responses in Field Experiment 1 (2022). Neither maternal line nor the maternal line × treatment interaction explained significant variance for growth, flowering phenology, flower number, or seed production (all χ2 ≤ 0.19, all P ≥ 0.750 for maternal line; all χ2 ≤ 0.28, all P ≥ 0.870 for maternal line × treatment; Table S2). Maternal line variance for seed production was negligible (χ2 = 0.10, P = 0.748), providing no evidence for underlying genetic structure in reproductive output in this year.
By contrast, several traits in Field Experiment 2 (2023) showed clear maternal line effects (Table S2). Maternal line significantly explained variation in total flower number (χ2 = 14.43, P < 0.001) and early‐season flower number (χ2 = 10.19, P = 0.001) and showed a marginal effect on fruit production (χ 2 = 3.45, P = 0.063). These results indicate substantial among‐line variation in reproductive traits in 2023. However, the maternal line × treatment interaction was nonsignificant for all traits in both years (all P ≥ 0.230), indicating limited evidence that maternal lines differed in the magnitude of their response to dicamba drift. To assess whether inclusion of population as a random effect reduced power to detect maternal line × treatment interactions, we re‐ran all models without population as a random effect; results were unchanged across all traits (maternal line × treatment: all χ2 ≤ 0.50, all P ≥ 0.870), confirming that the lack of genetic variation in hormetic responses is robust to model structure. Collectively, these results provide no evidence for genetic variation underlying hormetic responses in any fitness‐related trait examined.
Floral allocation
Dicamba drift significantly reduced flower width in the 2023 field experiment (Fig. 5a; Table S5). Flowers produced under drift conditions were significantly smaller than those produced in the control environment (F 1,15.8 = 11.41, P = 0.004), with mean flower width declining from 8.73 mm (95% CI: 8.01–9.47 mm) in control plots to 8.09 mm (95% CI: 7.43–8.77 mm) under dicamba drift, representing an approximate 0.64 mm reduction in size. Importantly, maternal lines differed significantly in their response to dicamba exposure (= 7.46, P = 0.024), providing evidence for genetic variation in tolerance or plasticity of flower width. Population‐level variation in flower width was weaker ( = 3.05, P = 0.081).
Fig. 5.

Effects of dicamba drift on flower width and its relationship to reproductive traits in Oxalis stricta. (a) Mean flower width (± SE) for control (blue) and dicamba drift–exposed (yellow) plants, with individual flowers shown as open circles. (b) Relationship between mean flower width and total seasonal flower number, with treatment‐specific regression lines. (c) Daily flower width measurements across the flowering period, with individual flowers shown as open circles and smoothed lines depicting temporal trends for each treatment. All data are from Experiment 2.
Despite this reduction in flower size, the slope of the relationship between flower number and flower width did not differ between treatments. Flower width increased significantly with flower number in both environments ( = 176.66, P < 0.001; Fig. 5b), and the flower number × treatment interaction was not significant ( = 1.19, P = 0.275), indicating that dicamba drift did not alter how floral display size scaled with flower abundance. Thus, although individual flowers were smaller in the drift environment, plants did not shift their allocation strategy according to flower number.
The reduction in flower size persisted across sampling dates. When daily flower width measurements were plotted over the month following dicamba application (Fig. 5c), drift‐treated plants produced smaller flowers throughout the sampling period, with no evidence of convergence toward control values. Flower size changes were parallel between treatments, further reinforcing that dicamba drift consistently reduced per‐flower investment over time.
Pollinator observation
Pollinator visitation rate differed significantly by treatment ( = 9.79, P = 0.002; Fig. 6a, Table S6), with lower visitation under dicamba drift (0.0278 visits per flower per minute; 95% CI: 0.020–0.039) than in control plants (0.0415; 95% CI: 0.030–0.057). This corresponds to an c. 33% reduction in visitation under drift exposure, with pollinators visiting control flowers 1.49 times more frequently than drift‐exposed flowers (control/drift ratio = 1.49).
Fig. 6.

Dicamba drift indirectly alters pollinator visitation to Oxalis stricta through changes in flower number. (a) Pollinator visitation rate (visits per flower per minute) under control and dicamba drift treatments. Points show individual observations; horizontal bars indicate treatment means ± SE. (b) Mean visitation rates by pollinator functional group under control and dicamba drift treatments (± SE). P‐value reflects the functional group × treatment interaction; an asterisk above a functional group's bars indicates a significant control‐vs‐drift difference within that group (post‐hoc pairwise comparison, P < 0.05). (c) Structural equation model summarizing direct and indirect effects of dicamba drift on pollinator visitation. Standardized pathway coefficients are shown next to arrows; arrow thickness is proportional to effect size. The direct effect of dicamba drift on visitation (dotted arrow) and pathways involving flower width were not significant. The r 2 values within each response variable denote the proportion of variance explained by the full model.
Pollinator functional groups differed in their floral visitation rates (Treatment × functional group interaction: = 18.77, P < 0.001; Fig. 6b). Visitation by flies and honeybees was significantly reduced under dicamba drift, with flies visiting control flowers more than four times as frequently as drift‐exposed flowers (control/drift ratio = 4.32, P < 0.001) and honeybee visitation reduced by c. 33% under drift exposure (control/drift ratio = 1.49, P = 0.002). By contrast, bumblebee visitation increased slightly under drift (P = 0.049), while syrphid fly visitation did not differ between treatments (P = 0.462).
SEM of pollinator visitation
We performed SEM to determine which component of floral display – flower number or flower size – most strongly influenced pollinator visits, quantified as visits per minute. We compared alternative structural equation models to evaluate whether dicamba effects on pollinator visits were mediated primarily through changes in floral display size or individual flower size, using a fully saturated model as a reference. The best‐supported model, which fit the data well (Fisher's C = 0.13, df = 2, P = 0.937; Table S7), indicated that dicamba influenced pollinator visitation indirectly through effects on floral traits rather than through a direct pathway (Fig. 6C). Specifically, dicamba significantly reduced flower width (β = −0.22, P < 0.001) while increasing flower number (β = 0.13, P = 0.021), and flower number had a strong positive effect on pollinator visitation time (β = 0.53, P < 0.001). In contrast, dicamba had no detectable direct effect on visitation after accounting for floral traits (β = −0.10 ± 0.10 SE, P = 0.313), and flower width did not significantly predict either flower number (P = 0.37) or pollinator visitation; accordingly, models including a direct flower width‐to‐visitation pathway were less well supported (ΔAIC ≈ 2). The resulting indirect effect of dicamba on visitation mediated by flower number was positive (β indirect ≈ 0.07; Table S8). This apparent contrast reflects differences in scale between visitation metrics, with SEM capturing whole‐plant visitation rather than per‐flower rates.
Discussion
By linking herbicide‐induced hormesis to damage‐induced overcompensation, a response long studied within the evolutionary ecology of plant defense, this study is the first to place hormesis within the broader framework of tolerance. Using the weed O. stricta as a model system, we show that drift‐relevant doses of dicamba elicited changes in growth, flowering, and allocation traits, but the magnitude, direction, and downstream consequences of these responses varied between years. SEM clarified why: trait shifts induced by hormesis did not consistently influence fitness, sometimes leading to increased reproductive output and other times reshaping floral display and pollinator visitation without translating to higher fitness. Genetic variation in these responses was limited, suggesting constrained potential for evolutionary change. Together, these results show that low‐dose chemical stress can restructure trait expression and plant–pollinator interactions even in the absence of strong underlying genetic variation, extending the evolutionary ecology framework long applied to overcompensation to the case of hormesis.
Hormesis as compensatory flowering under herbicide drift
Our prior work showed that O. stricta does not survive or maintain seed production when exposed to the recommended field dose of dicamba (Baucom et al., 2025), indicating that this species is neither resistant nor tolerant to this herbicide at agronomic rates. By contrast, flowering responses under drift‐level dicamba exposure were consistently positive. Across both prior studies and the present experiments, dicamba drift increased flower number and, in some cases, altered flowering timing or duration relative to non‐exposed controls (Iriart et al., 2022; Baucom et al., 2025). Extending these observations, our two field experiments provide clear evidence of compensatory responses: in both years, dicamba exposure consistently increased flowering, whereas effects on downstream components of fitness were more variable. In 2022, enhanced flowering was accompanied by increased vegetative growth and propagated indirectly to higher seed production, whereas in 2023, flowering increases did not translate into greater fruit output. This contrast indicates that flowering is a robust hormetic response to dicamba drift in O. stricta, while the fitness consequences of that response depend strongly on environmental or developmental context.
One likely contributor to these year‐specific outcomes is the timing of herbicide exposure relative to plant development. In the first experiment, dicamba was applied after 22 d of growth, when only 17% of plants had initiated flowering, whereas in the second experiment, 78% of individuals had flowered within that same interval, such that plants were at a more advanced reproductive stage at the time of exposure. Developmental timing has been shown to shape the expression of hormesis in other plant systems (Velini et al., 2008; De Carvalho et al., 2013; Belz & Duke, 2014), with earlier exposure often eliciting stronger compensatory growth or reproductive responses. Although both experiments were conducted in the same field location, additional year‐to‐year variation in environmental conditions (e.g. temperature, precipitation) may also have influenced which traits most strongly determined fitness in a given year. Together, these factors likely explain why consistent increases in flowering were associated with higher seed production through indirect trait pathways in 2022, but not in 2023, when increased flowering did not translate into detectable differences in reproductive output. Such variability is common in hormesis studies and has complicated efforts to predict or exploit low‐dose herbicide responses in agricultural systems (Duke et al., 2025). Future experiments that explicitly manipulate exposure timing and environmental context will be essential for resolving how and when compensatory flowering responses lead to fitness gains.
Further, although we have not yet conducted a full dose–response experiment, the contrast between field‐dose lethality and drift‐dose survival, combined with increased flowering across years, offers compelling evidence that hormesis operates in this species and underscores the importance of evaluating herbicide drift effects under ecological field conditions. Our findings align with the broader literature: while dicamba‐specific hormesis studies in weedy plants remain limited, our results parallel hormetic responses to sublethal doses of herbicides such as glyphosate and ALS inhibitors (summarized in Belz et al., 2022; Duke et al., 2025).
Our use of SEM to examine hormesis is, to our knowledge, novel, providing a framework to evaluate how low‐dose herbicide exposure influences fitness through interconnected traits. By explicitly modeling causal relationships among growth, flowering, and reproductive output, we were able to distinguish consistent trait responses from year‐specific fitness outcomes. In 2022, dicamba had no detectable direct effect on seed production; instead, fitness gains emerged indirectly through a cascade of trait responses in which enhanced vegetative growth increased flower production, which in turn drove higher seed output. By contrast, although dicamba consistently increased flowering in 2023, neither flower number nor growth translated into increased fruit production, and no direct treatment effect on reproduction was supported. These contrasting pathway structures indicate that while flowering responses to dicamba drift are robust, their consequences for fitness depend on which traits most strongly limit reproduction in a given year.
Taken together, the SEM results demonstrate that dicamba exposure reshaped growth and allocation traits in both experiments, but only under certain ecological or developmental contexts did these changes propagate to reproductive output. This strong interannual context dependence underscores the value of multivariate, causal approaches for understanding hormesis, revealing how low‐dose chemical stress can reorganize trait relationships even when its effects on fitness are inconsistent across environments.
No evidence for genetic variation in hormetic responses
Despite consistent effects of dicamba on flowering and other allocation traits, we found little evidence for genetic variation underlying hormetic responses in any fitness‐related trait. Maternal line × treatment interactions were nonsignificant in both years, indicating that lines did not differ in their plastic responses to dicamba drift and suggesting limited potential for the evolution of increased hormesis in O. stricta. Similar patterns have been reported in other systems: Trifolium pratense exposed to synthetic auxin drift likewise detected strong trait‐level responses but little evidence for line × treatment interactions in growth‐related traits, consistent with predominantly plastic responses to sublethal exposure (Iriart et al., 2025). By contrast, studies in barley (Hordeum vulgare; Belz & Sinkkonen, 2019, 2021) and weedy plants Tripleurospermum perforatum (Belz & Sinkkonen, 2016) and Abutilon theophrasti (Ethridge et al., 2023) strongly suggest that genetic variation in hormetic responses can occur at low herbicide doses.
In contrast to the absence of genetic variation in hormesis, we did detect significant among‐line variation in reproductive traits such as flower number, indicating ample genetic variation in baseline fitness traits themselves. Thus, while reproductive performance may evolve, the plastic responses to dicamba drift exposure likely will not, and hormesis is unlikely to drive adaptive change in this species. While we detected little evidence for genetic variation in responses at the drift dose used, we cannot exclude the possibility that genetic variation in hormesis may emerge at other sublethal exposure levels or under different environmental conditions. Thus, an important next step will be to evaluate hormetic responses across multiple sublethal doses, as recommended by (Duke et al., 2005), to determine whether genetic variation emerges at particular dose thresholds, which would alter expectations about the evolutionary trajectory of hormesis in this species.
Dicamba drift reshapes floral allocation and display traits
Given the absence of genetic variation underlying hormetic responses in O. stricta, these responses are unlikely to evolve, and their consequences are more likely to be expressed through altered ecological interactions rather than adaptive change in Oxalis. Across both field experiments, dicamba drift consistently increased floral display, reinforcing prior evidence that sublethal exposure promotes elevated flower production in this species (Iriart et al., 2022). We therefore asked whether this increase in flower number was accompanied by shifts in floral allocation. Consistent with this expectation, dicamba exposure significantly reduced flower width throughout the flowering period, even as flower production increased. This pattern suggests a compensatory allocation strategy in which plants produce more flowers at the expense of per‐flower investment, resulting in a larger but finer‐scaled floral display. Such size reductions have important ecological implications, as floral morphology strongly influences pollinator attraction and pollen transfer efficiency (Caruso et al., 2019); persistent declines in flower size under drift could alter pollination dynamics and, potentially, patterns of mating and gene flow. Similar patterns of altered floral display under dicamba drift have been documented in other species, including shifts in flower number, phenology, and pollinator visitation (Egan et al., 2014; Bohnenblust et al., 2016; Baucom et al., 2025), indicating that auxin‐mimicking herbicides can restructure floral traits in ways that extend beyond simple biomass responses.
Consequences for plant–pollinator interactions
Consistent with these allocation shifts, dicamba drift significantly altered pollinator visitation patterns, indicating that low‐dose herbicide exposure reshapes plant–pollinator interactions rather than uniformly suppressing pollinator activity. Overall visitation rates declined under drift exposure, but responses differed strongly among functional groups, revealing a treatment × pollinator interaction rather than a generalized reduction in visitation. Visitation by flies and honeybees declined substantially in the drift environment, whereas bumblebee visitation increased slightly and syrphid fly visitation was unaffected. These group‐specific responses suggest that pollinators respond differently to dicamba‐induced changes in floral display, likely reflecting variation in foraging strategies and sensitivity to display size rather than uniform avoidance of drift‐exposed plants. The slight increase in bumblebee visitation under drift is consistent with this interpretation, suggesting that larger floral displays may be particularly attractive to pollinators that respond strongly to display size, even when individual flowers are smaller.
To our knowledge, this is the first study to use SEM to identify the mechanistic pathway by which herbicide drift alters pollinator visitation, revealing that effects are mediated through hormetic changes in floral display rather than direct effects on visitation. Dicamba had no detectable direct effect on pollinator visitation once floral traits were included, nor did changes in flower width contribute to variation in visitation. Instead, dicamba influenced visitation predominantly through its hormetic effect on flower number, which had a modest but positive indirect effect on visit rate. This apparent contrast with the mixed effect results reflects differences in scale and response metrics: mixed effect models quantified visitation as visits per flower per minute, whereas the SEM evaluated visitation at the plant level, integrating the expanded floral display (i.e. visits/minute rather than visits/flower/minute). Together, these results indicate that dicamba‐exposed plants receive similar or slightly greater total pollinator attention due to increased flower production (as summarized by Fig. 6c), but that this attention is distributed across more flowers, resulting in reduced visitation rate per flower (evident in Fig. 6a). Thus, dicamba drift alters pollinator behavior primarily by redistributing pollinator effort across an increased floral display rather than through changes in individual flower size, with potentially important consequences for pollen transfer efficiency and mating outcomes.
Implications for weed ecology and eco‐evolutionary dynamics
Overall, our findings align with the tolerance–overcompensation framework (Fig. 1) but reveal important context dependence in how hormetic responses translate to fitness. Across both experiments, O. stricta consistently increased flower production under dicamba drift, indicating a compensatory allocation response rather than a uniform increase in reproductive success. However, this enhanced floral allocation resulted in overcompensation – defined as increased fitness under stress – only in 2022, when dicamba's effect on seed output was mediated indirectly through increased flower production. By contrast, in 2023, increased flowering did not translate into increased reproduction, underscoring that compensatory trait responses do not necessarily imply overcompensation in fitness. Because we detected no heritable variation in the magnitude of these responses, hormesis in O. stricta appears to reflect plastic tolerance rather than an evolving overcompensatory strategy.
While our study focuses on a single species and herbicide, the trait‐mediated pathways identified here are likely relevant to other non‐target plants exposed to synthetic auxin drift. The patterns we document are especially timely given recent shifts in US agriculture, where the widespread adoption of auxin‐resistant crop varieties has driven substantial increases in the use of synthetic auxin herbicides such as dicamba and 2,4‐D since the early 2000s (Singh & Jhala, 2025). Although regulatory changes have recently redirected use away from dicamba (Singh & Jhala, 2025), auxin drift remains a widespread and emerging feature of agricultural landscapes, particularly with the predicted increase in 2,4‐D use in the coming decade (Freisthler et al., 2022).
This shift toward synthetic auxin herbicides is ecologically consequential because auxins are central regulators of plant development (Woodward & Bartel, 2005) and influence interactions with herbivores and pollinators through effects on defense chemistry, volatile signaling, and floral traits (Ramos et al., 2021, 2023; Johnson et al., 2023). Accordingly, synthetic auxin drift has the capacity to reshape not only plant growth and reproduction but also plant–insect interactions. The combination of increased floral display and reduced flower size under drift that we document in O. stricta suggests that low‐dose exposure may alter traits important for pollinator attraction, pollen transfer efficiency, and mating outcomes, even in the absence of consistent fitness overcompensation.
These responses have implications for weed dynamics in agroecosystems. Plastic, compensatory increases in reproduction may promote persistence under low‐dose exposure, favoring species capable of such responses and potentially reshaping weed community composition. Likewise, auxin‐induced modification of floral traits may influence pollination networks and gene flow even in the absence of evolutionary change in hormesis itself (Iriart et al., 2022). Future work should evaluate whether these plastic responses affect pollinator visitation, seedling recruitment, and longer‐term population trajectories, and whether drift exposures of synthetic auxins do in fact facilitate weed shifts.
Competing interests
None declared.
Author contributions
ASB and RSB designed the research. ASB, ECK, BMC, and OAT collected data. ASB and RSB analyzed the data. ASB and RSB interpreted results and wrote the manuscript. ASB, ECK, BMC, OAT, and RSB edited the manuscript.
Disclaimer
The New Phytologist Foundation remains neutral with regard to jurisdictional claims in maps and in any institutional affiliations.
Supporting information
Fig. S1. Oxalis stricta sampling locations and species image.
Fig. S2. Conceptual path diagrams illustrating the full saturated structural equation models evaluated for comparison with reduced models in Field Experiment 1 (2022) and Field Experiment 2 (2023).
Methods S1. Structural equation model selection – reproductive output.
Methods S2. Structural equation model selection – pollinator visitation.
Table S1. Locations of the 10 populations from which seeds were sampled.
Table S2. Mixed effects model results for Experiments 1 (2022) and 2 (2023).
Table S3. Comparison of piecewise structural equation models evaluating alternative pathways linking dicamba treatment (trt), post‐spray growth, floral allocation, and reproductive output in Field Experiments 1 (2022) and 2 (2023).
Table S4. Indirect, direct, and total effects of dicamba drift on fitness from the structural equation models in Fig. 4.
Table S5. Results of mixed effects models testing the effects of dicamba treatment on average flower width.
Table S6. Results of mixed effects models testing the effects of dicamba treatment on pollinator visitation rate (visits per flower per minute).
Table S7. Comparison of alternative piecewise structural equation models evaluating hypothesized pathways linking dicamba treatment, flower width, flower number, and pollinator visitation time.
Table S8. Decomposition of total treatment effects on pollinator visitation time based on the best‐supported piecewise structural equation model.
Please note: Wiley is not responsible for the content or functionality of any Supporting Information supplied by the authors. Any queries (other than missing material) should be directed to the New Phytologist Central Office.
Acknowledgements
We thank Mike Palmer and Jeremy Moghtader and Matthaei Botanical Gardens for assistance with field preparation and maintenance. We thank Yasmina Zimmer, Julie Sheldon, and Robin Pitilon for help with data collection and entry and thank Grace Zhang and the Baucom lab for feedback on the manuscript. During the preparation of this work, the author(s) used Claude.ai (Sonnet 5) in 2026 to edit manuscript text for clarity and generate ALT TEXT for the visually impaired. After using this tool, the author(s) reviewed and edited the content as needed and take full responsibility for the content of the published article. This work was supported by the University of Michigan, and the College of Literature, Science, and Art's Meet the Moment initiative.
Data availability
Data are available at Dryad: https://doi.org/10.5061/dryad.gqnk98t4d.
References
- Agrawal AA. 2000. Overcompensation of plants in response to herbivory and the by‐product benefits of mutualism. Trends in Plant Science 5: 309–313. [DOI] [PubMed] [Google Scholar]
- Bates D, Maechler M, Bolker B. 2011. lme4: linear mixed‐effects models using S4 classes.
- Baucom RS, Iriart V, Soble A, Armstrong MR, Ashman T‐L. 2025. Off‐target drift of the herbicide dicamba disrupts plant‐pollinator interactions via novel pathways. New Phytologist 247: 850–862. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Baucom RS, Mauricio R. 2004. Fitness costs and benefits of novel herbicide tolerance in a noxious weed. Proceedings of the National Academy of Sciences, USA 101: 13386–13390. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Belz RG, Carbonari CA, Duke SO. 2022. The potential influence of hormesis on evolution of resistance to herbicides. Current Opinion in Environmental Science & Health 27: 100360. [Google Scholar]
- Belz RG, Cedergreen N, Duke SO. 2011. Herbicide hormesis – can it be useful in crop production? Weed Research 51: 321–332. [Google Scholar]
- Belz RG, Duke SO. 2014. Herbicides and plant hormesis. Pest Management Science 70: 698–707. [DOI] [PubMed] [Google Scholar]
- Belz RG, Duke SO. 2017. Herbicide‐mediated hormesis. In: ACS Symposium Series. Washington, DC, USA: American Chemical Society, 135–148. [Google Scholar]
- Belz RG, Sinkkonen A. 2016. Herbicide hormesis to segregate a weed population? – a case study with Tripleurospermum perforatum (Mérat) Lainz. In German Conference on Weed Biology and Weed Control. Julius Kühn‐Institut, 103–110.
- Belz RG, Sinkkonen A. 2019. Low toxin doses change plant size distribution in dense populations – Glyphosate exposed Hordeum vulgare as a greenhouse case study. Environment International 132: 105072. [DOI] [PubMed] [Google Scholar]
- Belz RG, Sinkkonen A. 2021. Low glyphosate doses change reproduction and produce tolerant offspring in dense populations of Hordeum vulgare . Pest Management Science 77: 4770–4784. [DOI] [PubMed] [Google Scholar]
- Bohnenblust EW, Vaudo AD, Egan JF, Mortensen DA, Tooker JF. 2016. Effects of the herbicide dicamba on nontarget plants and pollinator visitation: dicamba and pollinator visitation. Environmental Toxicology and Chemistry 35: 144–151. [DOI] [PubMed] [Google Scholar]
- Brooks M, Kristensen K, van Benthem K, Magnusson A, Berg C, Nielsen A, Skaug H, Mächler M, Bolker B. 2017. GLMMTMB balances speed and flexibility among packages for zero‐inflated generalized linear mixed modeling. R Journal 9: 378. [Google Scholar]
- Calabrese EJ, Bachmann KA, Bailer AJ, Bolger PM, Borak J, Cai L, Cedergreen N, Cherian MG, Chiueh CC, Clarkson TW et al. 2007. Biological stress response terminology: integrating the concepts of adaptive response and preconditioning stress within a hormetic dose–response framework. Toxicology and Applied Pharmacology 222: 122–128. [DOI] [PubMed] [Google Scholar]
- Caruso CM, Eisen KE, Martin RA, Sletvold N. 2019. A meta‐analysis of the agents of selection on floral traits: meta‐analysis of selection on floral traits. Evolution 73: 4–14. [DOI] [PubMed] [Google Scholar]
- De Carvalho LB, Alves PLCA, Duke SO. 2013. Hormesis with glyphosate depends on coffee growth stage. Anais da Academia Brasileira de Ciências 85: 813–821. [DOI] [PubMed] [Google Scholar]
- Doust LL, Doust JL, Cavers PB. 1981. Fertility relationships in closely related taxa of Oxalis, section Corniculatae. Canadian Journal of Botany 59: 2603–2609. [Google Scholar]
- Doust LL, MacKINNON A, Dousp JL. 1985. Biology of canadian weeds: 71. Oxalis stricta L., O. corniculata L., O. dillenii Jacq. ssp. dillenii and O. dillenii Jacq. ssp. filipes (Small) Eiten. Canadian Journal of Plant Science 65: 691–709. [Google Scholar]
- Duke SO, Belz RG, Carbonari CA, Velini ED. 2025. Understanding herbicide hormesis: evaluating its positive and negative aspects with emphasis on glyphosate. Advances in Weed Science 43: e020250104. [Google Scholar]
- Duke SO, Cedergreen N, Velini ED, Belz RG. 2005. Hormesis: is it an important factor in herbicide use and allelopathy? Outlooks on Pest Management 16: 35–36. [Google Scholar]
- Egan JF, Bohnenblust E, Goslee S, Tooker JF, Mortensen DA. 2014. Effects of simulated dicamba herbicide drift on the structure and floral resource provisioning of field edge and old field plant communities. Agriculture, Ecosystems & Environment 185: 77–87. [Google Scholar]
- Ethridge SR, Chandra S, Locke AM, Everman WJ, Jordan DL, Owen MDK, Leon RG. 2023. Changes in the herbicide sensitivity and competitive ability of Abutilon theophrasti over 28 years: implications for hormesis and weed evolution. Pest Management Science 79: 4048–4056. [DOI] [PubMed] [Google Scholar]
- Felsot AS, Unsworth JB, Linders JBHJ, Roberts G, Rautman D, Harris C, Carazo E. 2011. Agrochemical spray drift; assessment and mitigation‐‐a review. Journal of Environmental Science and Health. Part. B, Pesticides, Food Contaminants, and Agricultural Wastes 46: 1–23. [DOI] [PubMed] [Google Scholar]
- Foster MR, Griffin JL. 2018. Injury criteria associated with soybean exposure to dicamba. Weed Technology 32: 608–617. [Google Scholar]
- Fox J, Weisberg S. 2018. An R companion to applied regression. Christchurch, New Zealand: Sage Publications. [Google Scholar]
- Freisthler MS, Robbins CR, Benbrook CM, Young HA, Haas DM, Winchester PD, Perry MJ. 2022. Association between increasing agricultural use of 2,4‐D and population biomarkers of exposure: findings from the National Health and Nutrition Examination Survey, 2001‐2014. Environmental Health: A Global Access Science Source 21: 23. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Garrido M, Hansen SK, Yaari R, Hawlena H. 2022. A model selection approach to structural equation modelling: a critical evaluation and a road map for ecologists. Methods in Ecology and Evolution 13: 42–53. [Google Scholar]
- Grace JB, Irvine KM. 2020. Scientist's guide to developing explanatory statistical models using causal analysis principles. Ecology 101: e02962. [DOI] [PubMed] [Google Scholar]
- Guedes RNC, Rix RR, Cutler GC. 2022. Pesticide‐induced hormesis in arthropods: towards biological systems. Current Opinion in Toxicology 29: 43–50. [Google Scholar]
- Hartig F. 2024. DHARMA: residual diagnostics for hierarchical (multi‐level/mixed) regression models. CRAN: contributed packages.
- Imamura A, Hariu K. 2020. Withering corolla remains on the calyx tube enhance reproductive success in Oxalis stricta L. Plant‐Environment Interactions (Hoboken, N.J.) 1: 95–101. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Iriart V, Baucom RS, Ashman T‐L. 2022. Interspecific variation in resistance and tolerance to herbicide drift reveals potential consequences for plant community co‐flowering interactions and structure at the agro‐eco interface. Annals of Botany 130: 1015–1028. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Iriart V, Kubota N, Ashman T‐L. 2025. Can the right partner mitigate harm? Rhizobial strains vary in their mediation of herbicide stress in a plant‐rhizobia mutualism. Evolution Letters 10: 54–64. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Johnson N, Zhang G, Soble A, Johnson S, Baucom RS. 2023. The consequences of synthetic auxin herbicide on plant–herbivore interactions. Trends in Plant Science 28: 765–775. [DOI] [PubMed] [Google Scholar]
- Kuznetsova A, Brockhoff PB, Christensen RHB. 2017. lmertest package: tests in linear mixed effects models. Journal of Statistical Software 82: 1–26. [Google Scholar]
- Lefcheck JS. 2016. PIECEWISESEM: piecewise structural equation modelling in for ecology, evolution, and systematics. Methods in Ecology and Evolution 7: 573–579. [Google Scholar]
- Marble SC, Gilliam CH, Wehtje GR, Samuel‐Foo M. 2013. Early postemergence control of yellow woodsorrel (Oxalis stricta) with residual herbicides. Weed Technology 27: 347–351. [Google Scholar]
- Marshall G. 1987. A review of the biology and control of selected weed species in the genus Oxalis: O. stricta L., O. latifolia H.B.K. and O. pes‐caprae L. Crop Protection (Guildford, Surrey) 6: 355–364. [Google Scholar]
- Paige KN. 1987. Overcompensation in response to mammalian herbivory: the advantage of being eaten. The American Naturalist 129: 407–416. [Google Scholar]
- Peterson R. 2021. Finding optimal normalizing transformations via bestNormalize. The R Journal 13: 310. [Google Scholar]
- Posit team . 2024. rstudio: integrated development environment for R. Posit Software, PBC, Boston, MA, USA.
- R Core Team . 2023. R: A language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing. [WWW document] URL https://www.R-project.org/. [Google Scholar]
- Ramos SE, Bakhtiari M, Castañeda‐Zárate M, Iriart V, Ashman T‐L. 2023. Herbicides and their potential to disrupt plant–insect chemical communication. Journal of Systematics and Evolution 61: 550–560. [Google Scholar]
- Ramos SE, Rzodkiewicz LD, Turcotte MM, Ashman T‐L. 2021. Damage and recovery from drift of synthetic‐auxin herbicide dicamba depends on concentration and varies among floral, vegetative, and lifetime traits in rapid cycling Brassica rapa . The Science of the Total Environment 801: 149732. [DOI] [PubMed] [Google Scholar]
- Ramula S, Paige KN, Lennartsson T, Tuomi J. 2019. Overcompensation: a 30‐year perspective. Ecology 100: e02667. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Rokhlova E. 2013. Biological characteristic of Oxalis stricta L. in South Karelia. Hortus Botanicus 8: 51–55. [Google Scholar]
- Russell L. 2025. EMMEANS: estimated marginal means, aka least‐squares means. R package version.
- Schielzeth H. 2010. Simple means to improve the interpretability of regression coefficients. Methods in Ecology and Evolution 1: 103–113. [Google Scholar]
- Singh M, Jhala AJ. 2025. Perspectives on the economics of herbicide usage in corn and soybean production in the United States. Weed Technology: A Journal of the Weed Science Society of America 39: e48. [Google Scholar]
- Soltani N, Oliveira MC, Alves GS, Werle R, Norsworthy JK, Sprague CL, Young BG, Reynolds DB, Brown A, Sikkema PH. 2020. Off‐target movement assessment of dicamba in North America. Weed Technology: A Journal of the Weed Science Society of America 34: 318–330. [Google Scholar]
- Strauss SY, Agrawal AA. 1999. The ecology and evolution of plant tolerance to herbivory. Trends in Ecology & Evolution 14: 179–185. [DOI] [PubMed] [Google Scholar]
- US EPA . 2021. Dicamba 2021 report on Dicamba incidents. US EPA.
- Velini ED, Alves E, Godoy MC, Meschede DK, Souza RT, Duke SO. 2008. Glyphosate applied at low doses can stimulate plant growth. Pest Management Science 64: 489–496. [DOI] [PubMed] [Google Scholar]
- Woodward AW, Bartel B. 2005. Auxin: regulation, action, and interaction. Annals of Botany 95: 707–735. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Fig. S1. Oxalis stricta sampling locations and species image.
Fig. S2. Conceptual path diagrams illustrating the full saturated structural equation models evaluated for comparison with reduced models in Field Experiment 1 (2022) and Field Experiment 2 (2023).
Methods S1. Structural equation model selection – reproductive output.
Methods S2. Structural equation model selection – pollinator visitation.
Table S1. Locations of the 10 populations from which seeds were sampled.
Table S2. Mixed effects model results for Experiments 1 (2022) and 2 (2023).
Table S3. Comparison of piecewise structural equation models evaluating alternative pathways linking dicamba treatment (trt), post‐spray growth, floral allocation, and reproductive output in Field Experiments 1 (2022) and 2 (2023).
Table S4. Indirect, direct, and total effects of dicamba drift on fitness from the structural equation models in Fig. 4.
Table S5. Results of mixed effects models testing the effects of dicamba treatment on average flower width.
Table S6. Results of mixed effects models testing the effects of dicamba treatment on pollinator visitation rate (visits per flower per minute).
Table S7. Comparison of alternative piecewise structural equation models evaluating hypothesized pathways linking dicamba treatment, flower width, flower number, and pollinator visitation time.
Table S8. Decomposition of total treatment effects on pollinator visitation time based on the best‐supported piecewise structural equation model.
Please note: Wiley is not responsible for the content or functionality of any Supporting Information supplied by the authors. Any queries (other than missing material) should be directed to the New Phytologist Central Office.
Data Availability Statement
Data are available at Dryad: https://doi.org/10.5061/dryad.gqnk98t4d.
