Insecticides may facilitate the escape of weeds from biological control

Field sites

We established our three-year field experiment in two 1.5-ha fields at the Penn State Russell E. Larson Agricultural Research Center (Rocksprings, PA, 40°42′42″N, 77°57′51″W), and identified them as “North” and “South” fields, referring to their location relative to a main road that bisects the research farm. In 2016, the year before the experiment began, five of six blocks in the South field grew soybeans, and one block grew a combination of sunflower mixed with other harvestable forage crops. In the North field, three blocks grew wheat, and three blocks grew soybeans. In spring 2017, we established a factorial field experiment to quantify the interactive effects of pesticide seed treatments and grass cover crops. The experiment was established as a soy-corn-soy rotation on the North field and as a corn-soy-corn rotation on the South field. We divided each field into 12.2 × 33.5 m plots, laid out in a randomized complete block design (RCBD) with six treatments (three levels of pest management by two levels of cover cropping) each replicated six times in each field.

The pest management treatments consisted of a “preventative” pest management treatment (“PPM”), an integrated pest management treatment (IPM), and a control (“no pest management”: “NPM”). In the PPM treatment, seeds were treated with a neonicotinoid insecticide and mixture of fungicides prior to planting (Rowen et al., 2022, Table S1). In the IPM treatment, we scouted plots and treated fields if insect pest populations exceeded economic thresholds. During the three-year experiment, the IPM plots received a single insecticide application: an in-furrow application of tefluthrin (Force 3G, 5 kg ha⁻¹ [Syngenta]) at planting in 2018 to control white grubs (Scarabaeidae; mostly Japanese beetles Popillia japonica Newman). Thus, in 2017, pest management practices in the IPM and NPM treatments were identical, while in 2019, IPM and NPM treatments were both untreated but had different treatment histories.

Cover-crop treatments included oats (Avena sativa L.) planted in spring of 2017 and 2019, and cereal rye (Secale cereale L.) planted in the fall of 2017. We used spring-established oats when planting conditions in fall were not ideal for establishing cereal rye. One to two weeks before planting corn or soybeans, we applied herbicides (RoundUp PowerMax or a combination of Impact, Accent, Banvil and DegreeExtra; Table S1) to all plots to terminate cover crops and manage early-season weeds across both fields. The terminated cover crops were not removed. In 2017 and 2018, we applied herbicides a second time (RoundUp and Accent) at the end of June or early July (Table S1; also in Rowen et al., 2022).

Spring weed and cover crop biomass

We harvested weed and cover-crop biomass two to three weeks before planting (Table 1). We collected all above-ground plant biomass from three, randomly spaced 0.25-m² quadrats in each plot. Cover-crop biomass was collected and handled separately from weed biomass. We dried all biomass samples in a 60 °C drying oven for at least 5 days before weighing.

Weed-seed bank

We used direct-germination assays to assess changes to the weed-seed bank in response to our treatments. We sampled the germinable weed-seed bank in early May each year (Table 1), before terminating weeds and cover crops or planting corn or soy. For each plot, we pooled nine, evenly spaced soil samples that were collected using a bulb planter (five cm wide, 10 cm depth; Yard Butler, San Diego, CA, USA). We sieved (1-cm mesh) and homogenized these pooled samples, then transferred them to paper bags to air-dry in a greenhouse for at least five days.

We subsampled 946 cm³ of air-dried soil from each sample, which we spread across an equal amount of soilless potting media (Promix) in standard plastic planting trays (28 × 54 × 6 cm). After an initial watering, we watered these trays as needed (2–3x per week) depending on ambient temperature and weed emergence. We maintained assays for six months (July to January), regularly identifying, counting, then removing seedlings as they emerged. Species that were challenging to identify as seedlings were transplanted to pots to continue growing until they could be accurately identified.

Weed biomass in August

To understand the influence of our treatments on weed communities, we assessed weed biomass in August of each year (Table 1). When assessment could not be completed in a single day, we assessed weeds by block. For each plot, we identified and collected all aboveground weed biomass from three, randomly spaced 0.25-m² quadrats. In 2017, we harvested plants en-masse and brought them into to the lab for identification; for forbs, we identified plants to species, whereas we grouped all grasses together. In 2018 and 2019, we identified all forb and grass weeds to species in the field and collected each species into separate paper bags. Some plots had such small amounts of a particular species that we noted its presence but did not harvest it for collection. We dried harvested biomass for at least 5 days at 55 °C and weighed each species.

Weed-seed predator community

To understand the influence of our treatments on seed-eating invertebrates, we assessed weed-seed-predator communities twice a year, in late June/early July and late August/early September. We defined the weed-seed-predator community as granivorous carabid beetles and ants, which we captured using two, evenly spaced pitfall traps per plot. Pitfall traps were constructed from 946 mL cups installed flush in the ground so that ground-dwelling insects would pass over them and fall in. We used a 50:50 mix of propylene glycol and water (∼60 mL per trap) as a killing agent. When traps were in-use, we placed a 20-cm plastic plate (propped ∼5 cm above the soil surface using nails) over them to protect traps from rain. We left traps open for 72 hr in June/July and Aug/Sept (Table 2). We returned trapped specimens to the lab, rinsed them with water, and transferred them to ethanol (70%) for storage and identification. When not in-use, we attached tight-fitting lids to the pitfall traps to avoid capturing insects between sampling events. In 2019 in the South field, vertebrate pests, likely raccoons, destroyed the majority of our traps during both sampling events, so we were unable to include those 2019 South field data in our analysis.

We identified carabid beetles to species using keys in Bousquet (2010) and all other insects to order. Here, we report on species that are classified as predominantly weed-seed predators, including ants (Formicidae) and carabid beetles in the genera Anisodactylus, Amara, Harpalus, Notiobia, and Bembidion (Larochelle, 1990; Lundgren, 2009). Taxa that predominantly feed on insects and other invertebrates are reported and discussed in Rowen et al. (2022).

Weed-seed predation

We deployed sentinel seeds to measure weed seed predation by invertebrates in the field. In 2017, we deployed sentinel seed cards comprising 30 seeds of red-root pigweed (Amaranthus retroflexus L.) and 20 seeds giant foxtail (Setaria faberi L.) that we glued to a 4 × 9 cm piece of 60 grit sandpaper (Westerman et al., 2003). Because this method proved delicate and difficult to transport, we switched to a different deployment method in 2018 and 2019 and removed instances where we recovered more than the number of seeds we put out from the dataset. In 2018 and 2019, each trap comprised double-sided tape (Duck^® Brand Indoor Heavy Traffic Carpet Tape) attached to the bottom of an inverted Petri dish (5-cm diameter) with 30 pigweed and 20 giant foxtail seeds scattered across the surface. We then adhered sifted sand (Quickrete Play Sand) across the remaining sticky areas of the tape (Gallandt, 2005; Ward et al., 2011) to prevent predators from getting stuck to the dishes. We installed seed cards or seed dishes in fields, with each placed inside vertebrate exclusion cages made from hardware cloth (10 cm wide × 8 cm tall, with 1-cm mesh with a plastic lid). We deployed three weed-seed cards (2017) or dishes (2018 & 2019) per plot and left them in place for 48 hr, three times each summer (Table 2). After 48 h in the field, we collected seed cards into envelopes (2017) and dishes into plastic bags (2018 & 2019) and brought them back to the lab to count the remaining number of whole pigweed and foxtail seeds.

Statistical analyses

In general, we tested the interactions of pest management (PM), presence of a cover crop (CC), field (N: North, S: South) and year (2017, 2018, 2019) using generalized linear mixed models (GLMMs) with plot, and both directions of blocking from the Latin-square design as random intercepts using appropriate error distributions. For carabid and ant activity density, we included the interaction of month (June/July and Aug/Sept) in the model. Similarly, we included an interaction of month with all other categorical variables in modeling weed-seed predation (June, July, or August/September). For each model, we calculated estimated marginal means based on the full interaction model and plotted those means with 95% confidence intervals. To determine the relative effects of different factors, we then reduced the interactions in the models until they had the lowest AIC values. We never dropped a term from the model completely (e.g., even if they were not significant, we included field or cover crop in the model). We ran all analyses, except weed-seed predation and forb biomass, using a Poisson or a negative binomial error distribution. Weed-seed predation models used a binomial distribution where the fate of each seed was calculated separately, and each seed card/dish, as well as the plot, was included as a random intercept to account for non-independence. For forb biomass, we used a zero-inflated gamma error distribution with log link function. We conducted all GLMM analyses using the package ‘glmmTMB’ (Brooks et al., 2017) in R (v. 4.1.2; R Core Team, 2018). We used the package ‘emmeans’ to calculate estimated marginal means and to conduct pairwise post-hoc tests using a Tukey multiplicity adjustment (Lenth, 2019). To examine violations of homogeneity of variance and test the fit of our chosen distributions, we used the package ‘DHARMa’ (Hartig, 2018). To test for spatial autocorrelation of E. candensis in 2019, we used Moran’s I test of Spatial Autocorrelation using the package ‘ape’ (Paradis, Claude & Strimmer, 2004).

Weed biomass in May/June

Weed biomass measured before cover crops were terminated and cash crops were planted differed among years and between fields (Field × Year: χ² = 36, df = 2, P < 0.0001; Fig. 1, Table 1). As expected, weed biomass was often lower in plots where cover crops were planted (CC: χ² = 6.0, df = 1, P = 0.01), though the magnitude of difference varied depending on the field (Field x CC: χ² = 4.3, df = 1, P = 0.04). In the North field in 2017 and 2019, we observed reductions in weed biomass due to cover crops (2017: P = 0.015, 2019: P = 0.002). Pest management treatments had no effect on weed biomass before cash crop planting (χ² = 2.8, df = 2, P = 0.25).

When including cover-crop biomass, total plant biomass in spring was consistently higher in cover-cropped plots compared to non-cover-cropped plots (CC: χ² = 29.3, df = 1, P < 0.0001), although the effect of cover on total biomass depended on the field and year (Field × CC: χ² = 7.7, df = 1, P = 0.005; Year × CC: χ² = 12.0, df = 2, P = 0.003; Fig. S1).

Weed-seed bank

The weed-seed bank was dominated by forbs from the North field and grasses in the South field (Fig. S2). Further, we saw different communities in different blocks in the North and South fields that corresponded strongly with previous field history (between blocks 3 & 4 in the North and block 6 from the other five blocks in the South).

We predicted that weed-seed predators and the fungicide in the pesticidal seed treatment would act on the weed-seed bank to alter the weed community. The forb weed-seed bank varied between fields and among years (Field × Year: χ² = 28.1, df = 2, P < 0.0001, Fig. 2A, Table 1), and pest-management treatment decreased forb abundance (PM: χ² = 6.4, df = 2, P = 0.04). We did not detect an interaction between pest management and field or year, but we did observe that the IPM treatment in 2017 and 2018 were lower than the other pest management treatments (Fig. 2A). Because the IPM treatment was not implemented until after the weed seed bank was sampled in 2018, this indicates that despite our Latin-square design, the abundance of seeds in the weed-seed bank was significantly different across the different pest-management treatments at the start of the project and was not attributable the treatments we imposed. As there do not appear to be consistent differences in plots year-to-year (i.e., plots with high weed abundances in 2017 do not necessarily have high abundances in 2018), we re-analyzed our data without 2017 in the model. When we looked at 2018 and 2019, we found that cover crops, rather than pest management treatment, decreased forb abundance in the weed-seed bank by 20% (χ² = 5.02, df = 1, P = 0.025; Fig. 2B).

Similarly for grass seeds, we found that, again, the seed bank varied between fields and among years (Field × Year χ² = 9.4, df = 2, P = 0.009, Fig. 3A). We also found that grass weed seed abundance was 25% higher when a cover crop was planted (χ² = 5.3 df = 1, P = 0.02), although this appears to be again largely driven by differences in 2017. When 2017 was removed from the model, cover crops no longer affected the weed seed bank (χ² = 0.86, df = 1, P = 0.77; Fig. 3B).

Because we expected weed seeds of different species would not be equally susceptible to pathogens, predators, or the suppressive effect of cover crops, we analyzed the effect of pest-management treatment and cover crops on species richness in the weed-seed bank. We found that species richness varied across fields and years (Field × Year: χ² = 43.5, df = 2, P < 0.0001 Fig. S3A). We again found that species richness varied among pest-management treatments and was consistently lower in the IPM treatments compared to the PPM or NPM treatments, due to differences in 2017 and 2018, particularly in the North field (PM: χ² = 10.8, df = 2, P = 0.005, Fig. S3B). Thus, as with abundance of weed seeds, the richness of the weed-seed bank started with different communities in the IPM treatments compared to other treatments despite our Latin-square design.

Weed biomass in August

Because weeds compete with cash crops throughout the growing season, we measured weed biomass in August. When we examined the community composition of weeds in August, we saw patterns in weed communities depending on year and field. In the North field, the weed community had high abundances of dandelion (Taraxacum sp.), yellow woodsorrel (Oxalis stricta L.), and C. album in 2017, which was gradually replaced by marestail (Erigeron canadensis L.) and grasses by 2019. The South field had more grasses than other weeds in 2017 and 2018, and in 2019, dandelion was relatively common compared to grass (Fig. S4).

Examining the forb community, we found that mid-season forb biomass varied across field and years (Field × Year: χ² = 15.8, df = 2, P = 0.003; Fig. S5, Table 1), but contrary to our hypothesis, did not respond to pest-management treatment (χ² = 1.3 , df = 2, P = 0.5) nor cover crop (χ² = 1.9, df = 1, P = 0.17). One weed species E. canadensis, stood out as an increasing problem in the North field during the experiment (Fig. 4). In 2017, in the North field, we did not collect any E. canadensis in our samples. In the North field in 2018, we collected an average of 8.3 g E. canadensis m⁻² (95% CI [4.4–15.7]) and by 2019, we found an average of 93.4 g m⁻² (95% CI [61.4–142]). Erigeron canadensis in our trial did not respond to glyphosate application in 2019. In 2019, treatments without a cover crop that used an insecticide (IPM or PPM) had significantly higher E. canadensis biomass than the other treatments (CC × PM: χ² = 6.9, df = 2, P = 0.03, Fig. 4). Because E. canadensis seeds are wind-distributed, we tested for spatial autocorrelation within E. canadensis biomass in these plots. We found no evidence of autocorrelation (Moran’s I = 0.016, P = 0.18).

Grasses were an important part of the weed community in the South field in 2017 and 2018, and in the North field in 2019. We analyzed each field year separately because there was almost no grass in the North field in 2018. In 2017, in both the North and South fields, planting a cover crop suppressed biomass of grass weeds into August (North: χ² = 4.8, df = 1, P = 0.03; South: χ² = 48.9, df = 1, P < 0.0001, Fig. 5, Table 1). In addition, in the South field in 2017, we found grass weed biomass was higher in the IPM plots, likely due to initial differences in the weed-seed bank (PM: χ² = 9.9, df = 2, P = 0.007). In 2019 in the North field, we again found that cover crops marginally suppressed grass (CC: χ² = 3.5, df = 1, P = 0.06). In the South field in 2018 and 2019, however, planting a cover crop increased grass biomass in August (2018: χ² = 8.7, df = 1, P = 0.003; 2019: χ² = 3.4, df = 1, P = 0.07). In 2018, the magnitude of this effect depended on the pest management treatment, with the greatest effect in the NPM plots (CC × PM: χ² = 6.7, df = 2, P = 0.04).

Weed-seed predators (ants and herbivorous carabids) in June and September

Across the experiment, we collected and identified 17 species of granivorous carabids (Fig. S6). The weed-seed predator community was dominated by Harpalus pennsylvanicus (1,331 individuals, 81% of all granivorous carabids in both collection periods combined). We collected this species throughout the experiment, particularly in August and September (92% of all seed predator carabids captured). Early in the season, the carabid community was more even, when other Harpalus species (H. affinus, H. erraticus, H. rubripes, and H. faunus), Amara species (A. neoscotica and A. aenea) and Anisodactylus species (A. carbonaris, A. rusticus, and A. sanctaecrucis) were more common. One common granivorous species, Notiobia sayi, was only captured in August and September (Fig. S6).

Activity-density of granivorous carabids was higher in August/September than June (χ² = 24, df = 1, P < 0.001, Table 2), although the strength of this effect depended on the year and field (Season × Field × Year: χ² = 15.01, df = 2, P = 0.001; Figs. 6A, 6B). Overall, activity-density of granivorous carabids was higher in non-cover cropped plots than those planted with a cover crop (χ² = 9.9, df = 1, P = 0.002; Figs. 6C, 6D). We detected no effect of insecticide use on granivorous carabids (χ² = 0.4, df = 1, P = 0.8).

The other granivorous group we caught in pitfall traps, ants, were affected by neither the presence of a cover crop (χ² = 0.27, df = 1, P = 0.60) nor insecticides (χ² = 2.2, df = 2, P = 0.34), although ants were more abundant in June than in August/September (χ² = 117.3, df = 1, P < 0.001, Table 2), their abundance varied by year and field as well (Year × Field: χ² = 104.9, df = 2, P < 0.001, Fig. S7).

Weed-seed predation

To understand the potential of weed-seed predators to control weed seeds, we measured predation of two species of weed seeds (pigweed and foxtail) at three time points throughout each season. Total weed-seed predation rate depended on season and year and was affected intermittently by the presence of a cover crop and by insecticides (PM × CC × Year × Field × Season: χ² = 38.2, df = 8, P < 0.0001, Fig. 7, Table 2). In five of six field-years, weed-seed predation increased over the season (Fig. 7). The effect of cover crops and pest management treatment was inconsistent among years and seasons and between fields.