Phenotypic clines in herbivore resistance and reproductive traits in wild plants along an agricultural gradient

The conversion of natural landscapes to agriculture is a leading cause of biodiversity loss worldwide. While many studies examine how landscape modification affects species diversity, a trait-based approach can provide new insights into species responses to environmental change. Wild plants persisting in heavily modified landscapes provide a unique opportunity to examine species’ responses to land use change. Trait expression within a community plays an important role in structuring species interactions, highlighting the potential implications of landscape mediated trait changes on ecosystem functioning. Here we test the effect of increasing agricultural landscape modification on defensive and reproductive traits in three commonly occurring Brassicaceae species to evaluate plant responses to landscape change. We collected seeds from populations at spatially separated sites with variation in surrounding agricultural land cover and grew them in a greenhouse common garden, measuring defensive traits through an herbivore no-choice bioassay as well as reproductive traits such as flower size and seed set. In two of the three species, plants originating from agriculturally dominant landscapes expressed a consistent reduction in flower size and herbivore leaf consumption. One species also showed reduced fitness associated with increasingly agricultural landscapes. These findings demonstrate that wild plants are responding to landscape modification, suggesting that the conversion of natural landscapes to agriculture has consequences for wild plant evolution.


Introduction
Though habitat change is a natural part of global ecosystems, the scale and intensity of anthropogenic habitat disturbance in the last hundred years is unprecedented [1,2]. In much of the world, human modified landscapes are frequently characterized by expansive monocultures with intensive chemical and mechanical management. Within these heavily modified landscapes, semi-natural habitat patches such as field margins and riparian buffers can function as havens for natural populations by creating connecting corridors and refuges [3][4][5]. While studies have documented declines in species diversity in agriculture dominated landscapes [6][7][8], species that persist in these environments may also experience shifts in diversity on the level of individual trait expression resulting in phenotypic clines across land use gradients. Examining intraspecific trait responses to landscape modification may provide further insight into species sensitivity to environmental change and help predict the consequences for ecosystem function [9]. There is growing evidence that trait shifts associated with phenotypic clines may contribute substantially to community trait variation, with significant implications for ecosystem processes [9][10][11]. In plants, intraspecific trait variation has been found to account for 32% of total trait variation among communities [12]. This variation can be attributed to a combination of heritable differences as well as plasticity [13] which collectively are an important mechanism allowing plants to track changing environmental conditions [14,15]. Studies have found that wild plants experience fitness costs in landscapes dominated by agriculture [16,17] raising questions about wild plants' capacity to track the unprecedented speed of anthropogenic environmental change.
There is already some evidence of phenotypic clines corresponding with human mediated landscape modification. Collectively these studies documented larger floral displays, reduced defenses, and advanced phenology in plants within highly modified landscapes [18][19][20][21][22][23]. Though these studies found similar patterns of phenotypic change, the predicted mechanism driving this change was not consistent in all studies. For example, both Moreira et al. (2019) and Thompson et al. (2016) documented a reduction in chemical defenses for plants in urbanized landscapes, however, the former was explained by variation in herbivore damage, while the latter was explained by colder ground temperatures in winter. Shifts in floral traits were consistently attributed to pollinator visitation [18,20]. While these studies uncover important patterns and drivers of trait change associated with human mediated landscape modification, they are restricted to urbanization gradients, leaving patterns across agricultural gradients largely unexplored [24]. Given that agricultural landscapes cover over 40% of the earth surface [1,25], this represents a critical gap in our understanding of plant responses to landscape modification.
Like urban landscapes, agricultural landscapes are characterized by unique biotic and abiotic conditions with the potential to alter trait distributions within local plant populations. For example, agricultural landscapes have been associated with a reduction in the abundance and diversity of pollinating insects [26,27], soil degradation [28], and contamination by agrochemicals [29,30]. A phenotypic cline would provide evidence that plants are responding to these novel environmental conditions. It is important to understand trait responses of plants growing in semi-natural spaces within agricultural landscapes due to the critical ecosystem services they provide by supplementing floral resources for pollinators [31,32], creating reservoirs for natural enemies [33,34], and connecting fragmented habitat patches [35,36]. Therefore, trait changes have the potential not only to affect plant species persistence in modified landscapes, but also shape the community around them. Plant traits have been found to mediate intraspecific diversity in insects [37], highlighting the potential for shifts in intraspecific plant traits to have cascading effects on the broader community.
Though the direction and magnitude of insect responses to land use change can be highly variable by species [24], there are some generalizable patterns that we can use to formulate predictions for plant trait evolution across an agricultural gradient. Generally, land use change reduces the abundance and diversity of pollinators in a landscape [38][39][40][41][42][43]. Plants in landscapes with reduced pollinator abundances have been found to express both increased and decreased flower size [44,45], suggesting that pollinator limitation may result in plant reproductive strategies that either increase or decrease reliance on pollinators [46]. For herbivores, there is evidence that increasing agricultural land area with low crop diversity increases specialist herbivore abundances that can utilize the dominant crop and decreases generalist herbivore abundances [47][48][49]. Based on this evidence, we expect increasing agricultural landscape modification to reduce floral trait expression if plants are shifting away from a reproductive strategy reliant on pollinators. If the dominant herbivores in a landscape feed on non-crop plants as well, we expect to find an increase in resistance traits deterrent to herbivores. However, plants interact with pollinators and herbivores simultaneously, and the combined selective forces can be either opposing or synergistic [50][51][52][53][54].
In this study we test for evidence of a phenotypic cline across a gradient of agricultural landscape modification in three short-lived wild Brassicaceae: Barbarea vulgaris (Yellow Rocket), Thlaspi arvense (Field Pennycress) and Capsella bursa-pastoris (Shepherd's Purse). To evaluate trait changes in the absence of environmental variation, we collected seeds from plants growing in field margins along a gradient of increasing agricultural area in the surrounding landscape and grew their progeny together in a greenhouse common garden. Ultimately, this study seeks to examine wild plant responses in terms of herbivore resistance and reproductive traits to agricultural landscape modification, thus providing a foundation for formulating hypotheses and predictions about the mechanisms of landscape mediated trait change in future studies.

Study system
While our three study species originate from Eurasia, they have a global distribution and have been present in North America for over a century [55]. These species provide a useful system to study landscape-mediated trait adaptation because they are abundant even in landscapes dominated by agriculture. Because they persist across the entire landscape gradient, these species are exposed to the full range of variation imposed by human-mediated landscape modification. All three species utilize mixed mating systems, but the potential for self-pollination varies by species, with T. arvense and C. bursa-pastoris producing a high proportion of self-fertilized seeds [55,56], while B. vulgaris is primarily outcrossing [57]. Because all three species are considered naturalized in the landscape, no permits were required to collect seeds.

Landscape analysis
Parent seeds were collected from sites in the Finger Lakes Region of New York State, USA in the fall 2019. Land use composition at each site was evaluated using the 2019 National Agricultural Statistics Service Cropland Data Layer for New York (USDA 2019) in QGIS 3.16 (QGIS Development Team, QGIS Geographic Information System). Within a 500, 1000, and 1500 m radius of each collection site we calculated the proportion categorized as agriculture, pasture, urban development, natural open and natural forested land area (see S1 Table for detailed land cover classifications). This a well-established method for evaluating the broad scale effects of landscape modification on species interactions and ecosystem function [58]. To confirm that the landscape composition has been consistent for at least the last decade, allowing for adaptation to take place across the gradient of landscape modification, we regressed agricultural landscape composition in 2008 against each subsequent year based on the USDA Cropland Data Layer (Fig 1).

Seed collection from wild populations
At each site, parent plants were chosen haphazardly in semi-natural patches alongside cultivated crops (field margins, fallow fields, ditches). Sites were selected such that the existing variation in landscape composition at the broader scale from 500-1500m around each collection site represented a gradient with an increasing proportion of agricultural land cover (Fig 1). Seeds of B. vulgaris were collected at 13 sites from at least five parent plants at all sites except one where only four parent plants were present. Seed of T. arvense were collected from 15 sites, with at least five parent plants at all but two sites where only three parent plants were present. For C. bursa-pastoris, seeds were collected at 16 sites, with at least five parent plants at all sites except two where only four parent plants were present. Seeds were allowed to fully dry on the plant before collection and were kept separated by parent and stored in the freezer for at least a month to promote germination.

Germination and propagation
For each species, twenty-five seeds from each of five randomly selected parents (or the maximum number of parents available when less than five parents were found at a site) from each collection site were weighed together to the nearest .01mg to test for differences in seed mass from the field. Seeds were then planted in trays with Cornell Mix (see soil preparation in Supporting Information) in a greenhouse at approximately 20˚C with a 12-hour photoperiod and biweekly fertilization with Jack's Professional 1 water-soluble fertilizer 21-5-20 as 300 PPM (JR Peters, Inc., Allentown, PA USA). Total germination was recorded after 14 days. Seeds germinated from all B. vulgaris parent plants, while eight out of 71 T. arvense parent plants and 35 of 72 C. bursa-pastoris parent plants resulted in no germination. Replanting C. bursa-pastoris seeds to account for the low germination resulted in 1-25 germinated seedlings per site. Five randomly selected seedlings per parent (or the maximum number germinated when less than five seedlings from one parent germinated) were transplanted into individual pots (4 x 4 x 3.5 in) with Cornell Mix resulting in 320 B. vulgaris, 282 T. arvense, and 231 C. bursa-pastoris seedlings. The seedlings were arranged randomly in the greenhouse to account for any possible variation in light, temperature, or watering.
As biennials, it is common for brassica species to require a cold period to initiate flowering. All but two C. bursa-pastoris plants and 72 T. arvense plants bolted without vernalization. The remaining 210 T. arvense plants and all 320 B. vulgaris plants were moved into a cold growth chamber to vernalize. The temperature was set to 4˚C with a 10-hour photoperiod (corresponding to median winter day length) and the plants were watered only as needed to prevent wilting. Plants were moved back into the greenhouse after approximately 4 months. Out of 319 B. vulgaris plants, 308 individuals successfully bloomed after vernalization. All but 8 of the 282 individual T. arvense and 2 of the 231 C. bursa-pastoris plants successfully bloomed.

Floral traits
For all species, we removed one of the first 5 flowers that opened on a plant. All petals were removed and fixed to a piece of paper with transparent tape. Petal length and width were measured using a microscope at 12.5 magnification and CellSens software (Olympus SZX10 stereo microscope using the digital measurement tool in the cellSens software (Olympus Corp. Tokyo, Japan)) to the nearest hundredth mm. Anther-stigma distance was evaluated by measuring the length of the stamen and pistil and then subtracting the two values (stamen-pistil). The longest petal was used to measure petal length and width. To estimate petal area, we used the formula for an oval (½ length x ½ width x π). We evaluated self-pollination once the plants had finished blooming by recording the proportion of fertile seed pods out of 30 on three stalks (for a total of 90 per plant) when counting from the base. A seed pod was classified as fertile if at least one seed was developing inside. Plants were spaced in the greenhouse so that no individuals were touching, to prevent any neighboring plants from exchanging pollen.

Plant harvest/seed collection
For C. bursa-pastoris and T. arvense, once seed pods had dried but not shattered, we collected the first 15 unshattered seed pods from one of the most central stalks. After the seeds were collected, plants were cut back to the soil surface and placed in paper bags. The bags were left in the greenhouse to dry and moved into a drying oven for 48 hours at approximately 38˚C. The plants were weighed immediately after they were removed from the drying oven. The seed pods that were collected from the plants were weighed separately. For C. bursa-pastoris, seeds and seed pods were weighed together due to difficulty separating the small seeds from shattered pod debris. For T. arvense, the seeds were separated easily from the seed pods, and therefore only the seeds themselves were weighed. Due to the COVID19 pandemic related campus closures, B. vulgaris seeds were collected before the pods were ripe. The green pods were dried in a drying oven for 3 days before the seeds and seed pods were weighed together. This method was validated using a subset of B. vulgaris seed pods from a separate experiment confirming that seed pod mass and seed mass are highly correlated (Pearson's r = 0.90, P < 0.001).

Bioassay
We conducted no choice bioassays using the generalist herbivore, Trichoplusia ni. Bioassays were conducted independently for each species following the same protocol. Seeds from the same parent plants used to evaluate reproductive traits were sown in a growth chamber with a temperature of 21˚C, 15 h fluorescent light per day and daily watering. Once the seedlings had produced at least four true leaves, we randomly selected and transferred two seedlings from each parent to individual pots (4 x 4 x 3.5 in) with Cornell Mix and randomized the position of the pots among rows and columns within the growth chamber. Approximately 2-4 weeks after transplanting (depending on the species) two leaves were removed from each plant (126 B. vulgaris, 150 C. bursa-pastoris, and 124 T. arvense plants) starting with the first fully expanded leaf and continuing around the rosette. Each leaf was placed in an individual florist tube with water and enclosed in a plastic cup to contain the caterpillar with the leaf. Each leaf received one caterpillar that had fed on Southland Products Cabbage Looper Diet for two days to maximize caterpillar survival. Caterpillars were weighed immediately prior to placement and remained on the leaves for 3 days in a growth chamber with a temperature of 25˚C and 15 h of fluorescent light per day. On the third day, we weighed the caterpillars to the nearest .01mg and measured the total leaf area and leaf area consumed using LeafByte [59].

Statistics
We used R programming software (v.4.1.0) for statistical analyses [60]. We first computed Pearson correlation coefficients evaluating how the landscape composition across sites for each year correlated with the composition in the earliest year available (2008) to confirm the landscape composition has been stable across the last decade. Given the high number of landscape variables and scales (S1 Table), we then conducted a principal component analysis combining the landscape classifications found to be the most predictive for each individual species across scales to produce PC1 and PC2 values that summarized the relative composition of each source site (Fig 2). The PC1 value represents increasing agricultural land cover, explaining 47.6% of the variation in landscape composition, and was used for all species in all further analyses. We consider these results to be highly conservative given that we found each species showed the strongest response to a unique landscape type and scale when analyzed independently (S1-S3 Figs; S3-S5 Tables). We present the results from the principal component landscape variable here to uncover generalizable patterns of trait change associated with agriculture.
To evaluate changes in trait expression across the land use gradient we fit separate linear mixed models using the 'lme4' R package with each trait as the response variable and the PC1 landscape variable described in the previous analysis as a fixed effect [61]. We fit individual linear mixed-effects models for each species and trait combination (field collected seed mass, germination rate, petal area, anther-stigma distance, plant mass, self-pollination rate, selfpollination seed mass, caterpillar consumption efficiency, leaf area consumed, and caterpillar relative growth rate). Each model included a random effect of parent plant to control for multiple offspring from each parent which was nested within collection site to control for non-independence of parents within a site. Models for the parent-level traits of field collected seed mass and germination rate only included collection site as a random effect. For models evaluating petal size, we included plant mass as a predictor to control for variation in plant size. In each model evaluating T. arvense reproductive traits (petal area, seed mass, plant mass) we included vernalization as a random effect. We also ran a logistic regression to test the extent that the landscape gradient explains the probability that a T. arvense plant required vernalization to bloom.
To evaluate defensive traits, we used a logistic regression to test if caterpillar survival correlated significantly with the landscape gradient. Given there was low mortality that did not correlate with the landscape gradient, we excluded any caterpillars that did not survive to the end of the experiment in all future models. In the models evaluating defensive traits from the bioassay (caterpillar consumption efficiency, leaf area consumed, and caterpillar relative growth rate), we included bolting status (a binary variable describing whether a plant had developed reproductive structures as the time of the experiment) as a random effect for T. arvense and C.
bursa-pastoris to account for variation in plant ontogeny. For all models, we evaluated the significance of individual trait responses using a type three ANOVA. Mantel tests indicated spatial autocorrelation in the residuals of the model testing self-pollination rate in B. vulgaris, and the site coordinates were therefore included as fixed effects in this model (Dray & Dufour, 2007 [62]; S2 Table). There was no significant spatial autocorrelation in the residuals for all other models (S2 Table).

PLOS ONE
Phenotypic clines in wild plant traits along an agricultural gradient

Landscape stability and field collected data
There was a strong positive correlation between the landscape cover across sites in 2008 and the landscape composition for each subsequent year, providing evidence for landscape stability in the last decade (r > 0.98, P < 0.0001). Of the seeds collected from the field, neither the seed mass nor germination rates correlated significantly with the land use gradient except for C. bursa-pastoris, where a lower germination rate was found for seeds collected from sites with greater agricultural habitat modification (χ 2 = 6.7344, P < 0.01, Fig 3A-3C, Table 1).
Results of individual linear mixed-effects models investigating the effects of agricultural landscape modification (PC1) on field collected seed mass and germination rate in three wild brassica species. Agricultural landscape modification refers to the extent of agricultural land cover in a radius surrounding the collection site. Collection site was included as a random effect in all models. Statistically significant predictors are indicated in bold (P < 0.05).
We found no significant relationship for any species between stigma-anther distance, plant height, plant mass, or auto-fertility and agricultural landscape modification. In C. bursa-pastoris, self-pollinated seed mass declined significantly with increasing agricultural land cover ( Table 2, Fig 3G). There was a marginal negative correlation for B. vulgaris self-pollination rate and a marginal positive correlation for T. arvense seed mass with increasing agricultural landscape modification (B. vulgaris: χ 2 = 366, P = 0.056, T. arvense: χ 2 = 3.46, P = 0.063, Table 2, Fig 3H and 3I).
Results of general linear mixed models investigating the effects of agricultural landscape modification (PC1) on petal size, stigma-anther distance, plant mass, auto-fertility, and mass of self-fertilized seeds in three wild brassica species. Agricultural landscape modification refers to the extent of agricultural land cover in a radius surrounding the collection site. Collection site and parent plant were included as a random effect in all models. The site coordinates are included in the model of self-pollination rate in B. vulgaris to account for spatial autocorrelation in the residuals of the initial model. For T. arvense, vernalization status was also included as a random effect to account for the proportion of plants that underwent a cold treatment to initiate bloom. Statistically significant predictors (P < 0.05) are indicated in bold and marginal predictors (P < 0.1) are italicized.

Herbivore resistance assessment
In the herbivore no-choice bioassay, 96.  (Table 3, Fig 3J-3L). For all species, the leaf area and T. ni initial mass were positively correlated with consumed leaf area

PLOS ONE
Phenotypic clines in wild plant traits along an agricultural gradient (Table 3). Leaf area was positively correlated with caterpillar relative growth rate for all species, while agricultural land cover had no effect (Table 3). Caterpillar consumption efficiency was significantly positively correlated with agricultural land cover in C. bursa-pastoris, but not in B. vulgaris and T. arvense (Table 3, Fig 3M-3O). Caterpillar initial mass was a significant positive predictor of consumption efficiency for B. vulgaris and negative predictor for T. arvense. Leaf area was a significant negative predictor of consumption efficiency for C. bursa-pastoris.
Results of general linear mixed models investigating the effects of agricultural landscape modification (PC1) on caterpillar leaf area consumption, relative growth rate, and consumption efficiency (mass gained per unit area eaten) in a no choice bioassay with a generalist herbivore (Trichoplusia ni) feeding on three wild brassica species. Agricultural landscape modification refers to the extent of agricultural land cover in a radius surrounding the collection site. Collection site and parent plant were included as a random effect in all models. For C. bursa-pastoris and T. arvense, bolting status was also included as a random effect as a significant proportion of the plants began to bloom at the onset of the experiment. Statistically significant predictors (P < 0.05) are indicated in bold and marginal predictors (P < 0.1) are italicized.

PLOS ONE
Phenotypic clines in wild plant traits along an agricultural gradient

Discussion
The consequences of agricultural landscape modification on biodiversity and ecosystem function are at the forefront of ecological research [63][64][65]. However, the cascading effects of this landscape modification on individual trait responses have, until now, been largely unexplored. Recently, Mitchell et al. (2021) [66] found that proximity to cultivated sunflowers resulted in homogenized selection on floral traits for wild sunflowers. Here, we build on these findings by documenting phenotypic clines in both defensive and reproductive traits across three related plant species. These landscape mediated changes in plant phenotypes have the potential to shape the trait diversity of the broader community [37]. Our findings document a reduction in floral display size in plants from agriculture dominant landscapes for two of the three species examined. This pattern is counter to the findings along urbanization gradients documenting increased floral display size associated with urban environments [18,20], highlighting the unique conditions imposed by each landscape type. Mitchell et al. (2021) [66] found evidence of selection for higher ray length in wild sunflowers further from cultivated sunflower, aligning with our findings here in Brassicaceae. There are a number of potential biotic and abiotic mechanisms that may be mediating these changes in floral traits [67], however, insect pollinators are frequently identified as key drivers. For example, Brys and Jacquemyn (2012) found that pollinator depleted environments are associated with reduced flower size and increased autonomous self-pollination. There is strong evidence that agricultural landscapes support a reduced abundance and diversity of pollinators [26], making this mechanism a priority for exploration in future studies. A lack of response by the

PLOS ONE
Phenotypic clines in wild plant traits along an agricultural gradient third species (T. arvense) may be explained by the fact that in the individual analyses, trait variation was best predicted by the proportion pasture in the landscape rather than agriculture. However, it may also indicate resilience or that this species is responding by varying another trait not measured in this study such as floral rewards or number of open flowers. Future studies should evaluate how variation in different floral traits affect total plant fitness and whether this is predicted by insect interactions or abiotic factors. We also found that the land use gradient predicted metrics of plant defense for two species. T. ni caterpillars consumed less leaf area and developed more efficiently on plants from agriculturally dominant landscapes. Given that all plants were grown under equivalent conditions, these observed differences in consumption and assimilation are likely due to increased plant defenses in agricultural landscape, but further work is necessary to confirm this mechanism. There is a large body of evidence that herbivore pressure and plant defenses vary predictably along environmental gradients such as elevation, latitude, or plant diversity [68][69][70][71]. Once again, studies examining plant defenses in urban landscapes have found the opposite pattern to what we observed here, demonstrating a decrease in plant defenses within modified urban landscapes [21,72]. The few studies that have tested these patterns of herbivory in wild plants growing alongside agriculture have shown mixed results, where some report reduced herbivore pressure associated with agriculture [73], while others document augmented herbivore damage on wild plants [74]. There are a number of possible factors contributing to these varied results, including the natural history of the major insect herbivore, the relative attractiveness of adjacent crops compared to the wild plant community, potential insecticide spillover onto wild plants in intensively managed landscapes, and other abiotic factors like temperature [23, 48,[75][76][77].
We examined metrics of fitness through seed mass and germination rate in the wild parent plants and through seed mass and self-fertilization rate in the common garden plants. Of the field collected data, only C. bursa-pastoris demonstrated a significant decline in seed germination in agriculture dominant landscapes. This is consistent with existing studies comparing wild plant reproduction across agricultural gradients [16,78] and may also be the result of a dormancy mechanism adaptive in high disturbance agricultural environments [79]. Of the common garden plants, again C. bursa-pastoris was the only species that expressed a fitness cost with plants from agriculture dominant landscapes producing lighter seeds through self-fertilization. The combination of low germination rate in seeds from the parent plants and low seed weight in the common garden seeds may indicate inbreeding depression [80], potentially due to population fragmentation associated with homogenized landscapes and reduced pollinator availability to connect isolated populations. The absence of a fitness cost for the other two species could provide evidence of adaptation to the local environment in agricultural landscapes.
In this study we document phenotypic clines across a landscape modification gradient, however, with these data we cannot disentangle the relative degree that plant phenotypic differences can be explained by genetic differences, plasticity, and maternal effects. Given the importance of plasticity and maternal effects in enabling plants to track rapidly changing environmental conditions [81], examining overall trait responses provides a more complete estimate of wild plant responses to agricultural landscape modification. Future studies should test the relative contribution of plasticity and heritable change and examine whether these changes in trait values are adaptive. These studies should evaluate how different plant phenotypes perform across the landscape to test for evidence of selection and document insect interactions and abiotic conditions to identify the mechanisms driving plant trait change.
To date, much of the literature examining the consequences of agricultural landscape modification on plant communities has focused primarily on changes in species diversity. Here we demonstrate the potential for intraspecific trait change in species that persist within highly modified landscapes, highlighting the need to incorporate trait-based analyses into landscape scale studies. Given that wild plant communities play a critical role in habitat connectivity and stability in highly fragmented, human altered landscapes, understanding how they respond to land use change can inform predictions about the responses and resilience of entire communities [37,82]. We hope this work will spur further research examining plant trait change across species and land use types to expose broader patterns of plant trait change in human modified landscapes.

Conclusions
Across three Brassicaceae species, we found evidence for trait changes associated with increasing agricultural landscape modification in two species. For these species, plants originating from sites with greater agricultural modification in the surrounding landscape produced smaller flowers and received less damage in an herbivore no-choice bioassay than plant originating from landscapes with greater natural land cover. Because pollinators often impose selection on larger floral traits and herbivores on increased resistance, these findings may suggest greater pollinator mediated selection in natural landscapes and greater herbivore mediated selection in agricultural landscapes. However, because the outcome of selection is ultimately a result of the combined contribution of pollinators and herbivores, future studies should evaluate the relative role that pollinators and herbivores play in driving landscape mediated changes in plant traits.