Direct and Indirect Effects of Herbicides on Insect Herbivores in Rice, Oryza sativa

Densities of insect pests in agricultural communities may be affected by herbicides commonly used for weed management via several routes. First, herbicides may cause direct mortality to insects present both during and immediately following application. Second, herbicides may induce plant defenses that increase resistance to insect herbivores. Third, herbicides may alter the quantity and composition of weed populations, which in turn may change the structure of insect communities found subsequently in the crop. This study was designed to investigate the effects of an array of herbicides on the densities of several major pests found in rice in the southern United States. These pests included the rice water weevil, Lissorhoptrus oryzophilus Kuschel (Coleoptera: Curculionidae), the rice stinkbug, Oebalus pugnax (Fabricius) (Hemiptera: Pentatomidae), and a stemborer complex comprised of three lepidopteran species (Lepidoptera: Crambidae). Insects directly exposed to herbicides experienced high mortality; while those fed leaf material that had been exposed to herbicides did not. Herbicide application did not significantly increase resistance in rice to subsequent herbivore infestation. Results provided modest support for the third hypothesis represented by positive correlations between weed densities and insect pest densities.

www.nature.com/scientificreports www.nature.com/scientificreports/ instar nymphs cause damage to rice by removing the contents of developing grains with their piercing-sucking mouthparts, thereby negatively affecting rice yields and the quality of grains [26][27][28] . Because these insects do not enter fields until at least 60 days after most herbicides are applied, direct mortality from herbicides is extremely unlikely. Although herbicides have been shown to alter susceptibility of rice to piercing-sucking insects in past studies (such as the brown planthopper 29,30 ), the temporal separation between herbicide applications and stink bug infestation makes it most likely that rice stink bugs would be affected by changes in the weed community resulting from herbicide applications.
This study had three objectives. The first objective was to determine if any of seven herbicides commonly used in southern U.S. rice affected RWW mortality. Potential direct effects of herbicides on mortality of stemborers and rice stink bugs were not tested because the pattern of occurrence of these pests in rice fields strongly reduces the probability of direct exposure to herbicide applications. The second objective was to determine if these herbicides induced resistance in the rice plant thereby reducing subsequent pest densities of RWW, rice boring lepidopterans, or rice stinkbugs, and altering subsequent yield. Evidence from most prior studies suggests induction of the jasmonic acid pathway by herbicides in rice 9 . This pathway is associated with resistance to chewing herbivores, making it more likely that herbicide-induced resistance would affect RWW and stemborers. The third objective was to determine if reductions in density of weeds and alterations in weed communities due to herbicide application altered densities of insect pests.

Results
RWW mortality experiments (objective 1). Direct topical application of herbicides to RWW adults caused substantial mortality of insects (F = 16.68; df = 6,21; P < 0.0001) (Fig. 1A). Mortality was higher than in controls for all herbicide treatments. The herbicides 2,4-D, Bolero, Propanil, and Ricebeaux caused 100% mortality. The herbicides Command and Newpath caused 60-70% mortality, significantly higher than the control, but mortality caused by Command was significantly lower than most of the other herbicides.
Results of feeding assays in which RWW adults fed on rice leaves immediately after treating leaves with herbicide showed a significant overall effect of herbicide treatment on RWW mortality (F = 4.67; df = 6,21; P = 0.004), but mortality was only significantly higher than controls for RWW adults that fed on Bolero-treated material (Fig. 1B).

Field experiments (objectives 2 and 3).
Results from three small-plot field experiments conducted over two growing seasons were used to investigate the relative importance of herbicide-induced resistance (Objective 2) and herbicide-induced changes in weed communities (Objective 3) in influencing populations of insect pests in rice. Multivariate analyses including Canonical Correlation Analysis (CCA) and MANOVA were used to analyze results. Although these analyses could not definitively separate the two routes by which herbicide applications could indirectly affect pest populations, correlations of pest densities with herbicide injury and weed control were used to explore the most likely routes for each pest. In the presentation of results that follows, correlations of pest densities with herbicide injury (Objective 2) are presented before correlations of pest densities and weed control (Objective 3).
In the analysis for each of the three experiments, the first canonical correlation was significant or marginally significant, but subsequent canonical correlations were not significant in any of the three analyses. In the experiment in 2015, the first canonical correlation was 0.839 (F = 2.53; df = 12, 45.27; P = 0.012), with 94.13% of variance explained by the two variates. The canonical cross loading values indicated that in three of four core samplings (RWW1, RWW2, and RWW4) densities of RWW were negatively correlated with injury at seven days post-application (Injury1) (Objective 2) and weed control at both rating dates (Weed1 and Weed2) (Objective 3) ( Table 1). www.nature.com/scientificreports www.nature.com/scientificreports/ In the first experiment of 2016, the first canonical correlation was 0.630 (F = 2.07; df = 20,100.45; P = 0.001), with 48.2% of variance explained by the two variates. Consistent with results from 2015 the canonical cross loading values indicated that injury measured at seven and 21 days post-application (Injury1 and Injury 2) (Objective 2) was negatively correlated with RWW densities (RWW1 and RWW2). Canonical correlation values indicated that injury was positively correlated with whitehead (Whitehead) and rice stink bug (Stinkbug) densities. The relationship between RWW densities and densities of weeds was more complex (Objective 3). RWW densities were negatively correlated with weed control at seven days post-application, but a positive correlation was found for weed control ratings 21 days post-application. Densities of whiteheads and rice stink bug were positively correlated with weed control at seven days post-application, but were negatively correlated with weed control at 21 days post-application. Finally, canonical correlation values indicated that injury at seven and 21 days post-application (Injury1 and Injury 2) was negatively correlated with yield. Weed control at seven days post-application was negatively associated with yield while the opposite relationship was found using ratings of weed control at 21 days post-application (Table 1).
Simple linear regressions showed the relationships between insect densities and weed control (Objective 3) were somewhat consistent among the three experiments. In 2015 and the first experiment of 2016, RWW densities were negatively associated with weed control. In 2015, the relationship between weed control rating at seven days post-application (Weed1) and densities of RWW on the first core sampling date (RWW1) was significant (F = 7.23; df = 1,23; P = 0.013) (Fig. 2). Similarly, densities of RWW immatures in the first core sampling (RWW1) were negatively correlated with weed control at 21 days post-application (Weed2) (RWW1 F = 4.38; df = 1,23; P = 0.048). All other combinations of core samplings and weed control ratings in 2015 showed non-significant relationships.
In the first experiment of 2016, densities of RWW in the second core sampling and percent weed control (Objective 3) at 21 days post-application were significantly positively correlated (F = 12.88; df = 1,44; P = 0.008). Densities of RWW at the first core sampling, rice stink bug densities, and whitehead densities showed no significant relationship to weed control ratings. In the second experiment of 2016 there were no significant regressions among weed control ratings and insect densities.
Univariate analyses of the effects of herbicide treatment on the following response variables were performed: injury rating (Injury1 and Injury2), weed control rating (Weed1 and Weed2), insect densities (RWW, Whitehead, and Stinkbug) and Yield (Yield). Results showed significant injury (Objective 2) to rice in all herbicide treated plots in all three experiments at the first rating date seven days post application (Injury1); 2015 (F = 11.56, df = 4,20; P < 0.001), 2016(1) (F = 15.43; df = 6,24; P < 0.0001), 2016(2) (F = 29.58; df = 6,42; P < 0.0001) ( Table 3). No injury was observed at the second rating (Injury2) in 2015 indicating plants in all treatments recovered from injury incurred by herbicide treatments by 21 days post application. Conversely, in the 2016 experiments the effect of herbicide treatment on level of injury was still significant at 21 days post application; 2016(1) (F = 12.08; df = 6,24; P < 0.0001), 2016(2) (F = 2.59; df = 6,42; P = 0.031). Plots treated with Ricebeaux and 2,4-D showed consistently higher injury relative to control plots over the three experiments. The injury level in Bolero treated plots was significantly higher than controls at the second injury rating 21 days post-application (Injury2) in the first experiment of 2016.
Univariate analysis revealed that herbicides significantly controlled weeds relative to controls at each rating date (Objective 3). In 2015 percent weed control at both rating dates was significantly higher in all herbicide treatments when compared to controls. Plots treated with Newpath and Ricebeaux showed significantly higher control than plots treated with 2,4-D or Command at both rating dates (Weed1 F = 188.5; df = 4,20; P < 0.001) (Weed2 F = 48.11; df = 4,20; P < 0.001) ( Table 4).
In the first experiment of 2016, percent weed control (Objective 3) in plots treated with Bolero, Newpath, Propanil, and Ricebeaux was significantly higher than in control plots at both rating dates. At the second rating date, Command-treated plots showed higher percent weed control than untreated plots. There were also significant differences among herbicide treatments, with Ricebeaux, Newpath, and Propanil consistently providing higher weed control than other treatments. 2,4-D did not control weeds when compared to untreated plots in this experiment at the first or second rating (Weed1 F = 8.19; df = 6,42; P < 0.0001) (Weed2 F = 19.40; df = 6,42; P < 0.0001) ( Table 4).     (2)). Values in the same column accompanied by different letters represent means that differed significantly with Tukey's HSD test at α < 0.050. *N/A = Treatment was not applied in that year.
Univariate analysis of data from core samples showed that densities of RWW (Objectives 2 and 3) in herbicide-treated plots differed significantly from controls in 2015 but not in 2016, though there were differences among treatments in 2016. In 2015, RWW densities were significantly lower in Ricebeaux-treated plots (Herbicide F = 3.54; df = 4,76; P = 0.011), and RWW densities differed significantly among core sampling dates with the third date the highest and first date the lowest (Date F = 35.77; df = 3,76; P < 0.001). The interaction between herbicide treatment and core sampling date was not significant (Herbicide*Date F = 1.31; df = 12,76; P = 0.232) (Fig. 5).
In the first experiment of 2016, there was a significant effect of herbicide on RWW densities, but none of the herbicide treatments differed from the control (Herbicide F = 3.88; df = 6,68; P = 0.002) Plots treated with 2,4-D showed the lowest RWW densities and densities in these plots were significantly lower than in Bolero-treated plots, which had the highest densities (Fig. 6). There was no significant effect of core sampling date or the interaction of herbicide and core sampling date (Date F = 0.09; df = 1,68; P = 0.770; Herbicide*Date F = 0.46; df = 6,68; P = 0.838).
Densities of whiteheads (Objectives 2 and 3) were measured in the two experiments in 2016, and differed significantly among herbicide treatments in both experiments, although no differences from controls were noted (2016 (1) F = 2.46; df = 6,42; P = 0.039) (2016 (2) F = 2.35; df = 6,42; P = 0.048). In the first experiment of 2016, plots treated with Ricebeaux (2.57 ± 0.43 whiteheads/plot) showed the lowest densities of whiteheads, and densities in these plots were significantly lower than in Command-treated plots (6.29 ± 1.23 whiteheads/plot), which    www.nature.com/scientificreports www.nature.com/scientificreports/ had the highest densities. However, neither was significantly different from densities in control plots (3.71 ± 0.75 whiteheads/plot). In the second experiment of 2016 there were no significant differences in whitehead densities among treatments after Tukey adjustment.
Finally, yield was significantly affected by herbicide treatments in the first experiment in 2016 (F = 2.69; df = 6,42; P = 0.027), but not in the second (F = 1.12; df = 6.42; P = 0.365). In the first experiment, results show that Bolero-treated plots (9.79 ± 0.25 kg/plot) had the lowest yield and Command-treated plots (12.49 ± 1.38 kg/plot) the highest. These treatments differed from each other, but not from the control (11.48 ± 0.16 kg/plot).

Discussion
Applications of herbicides may affect herbivores in agroecosystems directly, by toxifying them, or indirectly, by inducing plant defenses or by altering the availability of non-crop plant hosts 13 . This study is one of only a few that has investigated direct and indirect effects of herbicides on herbivores simultaneously. Overall, the results from this series of experiments indicated that herbicides are capable of altering pest densities of both weeds and insects in rice, and the resulting changes in these communities may have impacts on yields.
The first objective of this study was to determine if applications of seven herbicides commonly used in rice in the southern U.S. caused mortality in RWW adults that were exposed by topical application or by feeding on leaves that had been recently treated with the herbicides. RWW adults are present and feeding in rice fields both before and after flooding of rice fields in the southern U.S, and are therefore present at the times that most herbicides are applied to rice fields in the southern U.S 20 . The herbicides caused significant mortality to RWW adults when applied topically at maximum label rates. This level of direct and sustained exposure is unlikely in the field as RWW adults are rarely found at the top of rice canopies, drop from plants when disturbed, and are capable of flying and swimming.
The feeding assays with herbicide-treated foliage likely better represent the type of exposure to which RWW adults are subjected in a field setting. RWW adults that were fed on rice with herbicide residue for the most part did not experience significant mortality. The exception to this was RWW adults fed on rice treated with Bolero. These insects experienced significant mortality when compared to RWW adults on non-treated rice tissue. The effect of consuming Bolero-treated rice tissue on RWW adults, however, did not translate to a reduction of immature RWW densities in Bolero-treated plots in the field experiments. In fact, Bolero-treated plots showed a trend toward higher densities of RWW immatures than the control plots. Thus, while herbicides, such as Bolero, may cause direct mortality of adults, this did not translate into reduced larval populations later in the season. This is likely due to the natural patterns of infestation of rice fields by RWW adults: relatively few adult weevils are present in fields before flooding, as they are still migrating from overwintering sites, and most arrive after flooding 20 .
The second objective was to determine if herbicide applications reduced pest densities by inducing resistance in the host plant. While evidence from this study does not provide substantial support for induction of defenses to all three insects by herbicides, it does provide support of low levels of resistance induced by herbicides to RWW. Of the herbivores studied, the RWW was most likely to be affected by herbicide-induced resistance, because it is the only insect of the three present in high numbers early in the growing season, when herbicides are applied 20 . RWW densities in herbicide-treated plots were significantly lower than control plots in one experiment (Ricebeaux-treated plots in 2015). However, there were instances in which densities of RWW in the various www.nature.com/scientificreports www.nature.com/scientificreports/ herbicide-treated plots differed significantly from each other, and this may have been related to the level of injury to the plants caused by herbicide treatments. Levels of induction of defenses in many plants are correlated with the amount of injury plants receive 31 . Over the three experiments in the current CCA suggested a negative correlation between injury and RWW density on five of eight core sampling dates. The herbicides 2,4-D, Bolero, and Ricebeaux consistently caused significant injury to rice plants. Ricebeaux-treated plots had lower RWW densities than many other treatments in most core samplings throughout the three experiments. In the first experiment of 2016 RWW densities in 2,4-D-treated plots were significantly lower than densities than several other herbicide-treated plots including Bolero and Ricebeaux. The injury caused by 2,4-D in this experiment was likely due to inappropriate timing of this herbicide, which is labeled for use before seeding or after the permanent flood.
The herbicides used in these experiments have different modes of action, and were used because mode of action could conceivably influence toxicity to insects, induction of plant defenses, and effectiveness of weed control. Bolero and Ricebeaux, two of the herbicides with the largest effect on RWW, both contain photosynthesis inhibitors, which reduce electron transport and have various effects on host plant quality. In rice, for example, pesticide-induced susceptibility to brown planthopper was correlated with reduced photosynthesis 30 . Thus, herbicide injury can affect plant physiology in ways that do not involve induction of resistance-related traits, and these changes in plant physiology may be relevant to herbivore performance.
Unlike RWW, stemborers and rice stink bugs are typically present in rice fields in high numbers later in the season and are less likely to be affected by herbicide-induced resistance, which, like other types of induced resistance, probably decays over time [32][33][34] . The results of the field experiments were consistent with a lack of herbicide-induced effects on these insects, as the only significant relationship between injury ratings and whitehead densities occurred in 2016 when injury was extremely low, and no significant relationship between injury and stink bug densities were found. Overall, our results suggest that the herbicides used in this study may have induced low levels of resistance to the RWW, but not to the other insect pests. It may be that high levels of induction were not achieved because overall injury from herbivores was low. The low levels of injury from herbicides was not surprising, since these herbicides were developed with the intent of causing little injury to the crop.
The third objective of this study was to determine if changes in plant (weed) populations caused by herbicide applications altered densities of subsequent insect pest populations. The herbivores involved in this study preferentially utilize grasses, including common weeds in rice. Densities of stemborers and rice stink bug were expected to be more affected by changes in weed densities, than by the other two routes of herbicide impact as these insects are not present in early season rice when herbicide applications are most frequently made. Rather, they attack rice in mid-to late-season, well after herbicide applications are made and when differences in weed densities are most pronounced.
Results provide modest support that alteration in weed densities affected densities of these pests. The CCA showed that in twelve of sixteen core sampling dates and rating combinations the relationship between weed control and RWW densities was negative; that is, RWW densities decreased as weed densities decreased. This was not further supported by simple linear regressions as only two of the sixteen combinations showed a significant negative relationship on the univariate level. In contrast, relationships among weed densities and densities of mid-to late-season pests were stronger. The CCA showed that overall densities of whiteheads and rice stink bugs were positively correlated with weed densities in six of eight insect samplings and weed rating combinations. While simple linear regressions showed that densities of whiteheads and rice stink bugs were not significantly correlated with weed densities in experiments performed in 2016, there was some support at the univariate level. Command-treated plots had the lowest weed control rating (aside from the control) at seven days post-application and the highest whitehead count, while Ricebeaux-treated plots had the highest weed control rating and the lowest whitehead count, suggesting that plots with higher weed densities had higher densities of borers. The relationship did not hold, however, for the second weed rating at 21 days post-application. In all three experiments, for five of six weed control rating dates, all herbicides were shown to be capable of significantly controlling weed pests. Therefore, overall weed densities in the herbicide-treated plots were low, and this may have prevented a robust investigation of the effect of weed density on rice stink bug and stemborer densities.
Differences in yields were observed among herbicide treatments. CCA of both experiments in 2016 showed a negative relationship between injury and yield. Results of simple linear regression in both experiments in 2016 support this result and showed a significant negative relationship between level of injury to rice and yield. This suggests that herbicide injury may have contributed to yield reduction. In the second experiment of 2016, injury was greatest in plots treated with 2,4-D, from 13-22%, and yields from 2,4-D-treated plots were the lowest of all groups. The first experiment of 2016 also showed that Bolero-treated plots had significantly lower yields than Command-treated plots. This may have been attributable to the fact that Bolero-treated rice received as much as 85% more injury than Command-treated plots in this experiment. It may also be explained by the high rates applied and inappropriate timings of applications.
In the second experiment of 2016, CCA showed a positive correlation between weed control and yield for three of four rating dates. The simple linear regressions supported this result and showed a significant increase in yield with increasing weed control. This may have been due to reductions in plant competition, but may also be attributed to lower insect pest densities in plots with lower weed densities. The combination of herbicide injury and insect herbivory has previously been shown to have negative effects on plant health. For example, weed management was shown to be more effective when volunteer potatoes, considered to be weeds, were defoliated by Colorado potato beetles, Leptinotarsa decemlineata Say, after herbicide-stress, relative to control of weeds by beetles on plants that were not herbicide-stressed 35 .
While there have been many studies investigating the effects of herbicides on insect communities, the vast majority focused on direct effects on insect mortality or fecundity and indirect effects due to changes in weed populations 36 . The literature is not, however, devoid of investigations of herbicide-induced resistance to herbivores. One study suggested 2,4-D induced susceptibility in corn, Zea mays, to two insects and one pathogen 37 . (2019) 9:6998 | https://doi.org/10.1038/s41598-019-43361-w www.nature.com/scientificreports www.nature.com/scientificreports/ Additionally, studies investigating herbicide-induced stress in the plant and subsequent effects of herbivory have been conducted. Phosphonate herbicides have been shown to have a dual effect as both herbicides and insecticides to the black bean aphid, Aphis fabae Scopoli (Hemiptera: Aphididae), when applied systemically in broad bean, Vicia faba 38 . There are few if any other studies that have combined investigations of all three routes by which herbicides may alter insect communities. There is a need for additional studies investigating direct and indirect effects simultaneously in order to make clearer inferences.
The use of herbicides is ubiquitous in modern agroecosystems, and thus it is important to understand effects of herbicide applications on arthropod communities in addition to their effects on the weeds they are intended to control. This study is one of the few studies to simultaneously investigate multiple routes by which herbicides could potentially affect arthropods and crop yields. Use of herbicides had minor and inconsistent impacts on insect pest densities in rice, whereas rice yields were negatively impacted by herbicide injury and presence of weeds. Although further studies in other crops are needed to investigate the relative importance of the various routes by which herbicides can impact populations of herbivores and crop yields, this study overall reinforces the importance of appropriate weed management in agricultural systems, as increases in weed and insect densities can negatively affect yield, as can crop injury resulting from improper use of herbicides.

Materials and Methods
Study site and rice culture. All 39 . Plots were fertilized with 134 kg/ha N applied pre-flood and 67 kg/ha P and K applied in the previous fall. Fertilization and disease control practices followed recommendations for drillseeded rice in southwest Louisiana 40 . Dates of all planting, fertilization, and flooding activities can be found in Supplemental Table 1. Plots were harvested at grain maturity with a mechanical combine and yield was adjusted to 12% moisture.

Rice water weevil mortality experiments (objective 1). To investigate direct effects of herbicides
on survival of adult RWW, adults were exposed either by direct topical application or by exposure to leaves treated with herbicides. RWW adults were collected from untreated border plots and used in experiments the same day as collection. For both experiments all herbicides were applied at their highest label rate. Herbicide treatments used in the 2015 experiment included 'Weed Rhap A-4D' (2,4-D), 'Command 3ME' (Command), 'Newpath' (Newpath), 'Ricebeaux' (Ricebeaux), 'Propanil 4SC' (Propanil), 'Bolero 8EC' (Bolero) and a control (Control), which was sprayed with water only. Information concerning the trade names, manufacturers, active ingredients, modes of action, rates, and typical timings for applications of each of these herbicides is provided in Supplemental Table 1. To test the effects of topical exposure, four petri dishes for each herbicide, each containing ten field-collected RWW adults, were placed on the ground and lids removed. Herbicides were applied using a backpack sprayer pressurized with CO 2 and calibrated to deliver 140 L/ha at 207 kPa through three 1002 flat fan nozzles TeeJet ® at 51 cm spacing. Mortality was assessed 24 hours after spraying. To assess mortality, weevils in each petri dish were observed for 15-30 minutes. RWW adults that did not display coordinated movement within that time were removed from the dish, and recorded as dead 41 .
To test the effects of exposure of RWW adults to leaves treated with herbicide, weevils were fed on herbicide-treated leaf material immediately after treating plants with herbicides. This was done by spraying one field plot by backpack sprayer with each herbicide at the same rate as described above. Rice foliage from these plots was immediately cut from the plants with scissors, and placed in a petri dish lined with moistened filter paper for adult feeding. Approximately 20 leaves were placed in each dish to ensure that weevils would not be food-limited. Ten RWW adults were then added to each of four dishes for each herbicide. The petri dishes were then moved to a colony rearing room with 16:8 L:D, 22 °C, 70% RH until mortality was assessed. Mortality was assessed one day after herbicide applications described above.

Herbicide treatments for field experiments (objective 2 and 3).
Herbicide treatments used in the 2015 experiment included 'Weed Rhap A-4D' (2,4-D), 'Command 3ME' (Command), 'Newpath' (Newpath), 'Ricebeaux' (Ricebeaux), and an unsprayed control (Control). A randomized complete block design with five blocks and one replicate of each treatment within each block was used. Herbicides were applied one day prior to flooding, when plants had reached the four-leaf stage. Although this timing of application differs from standard timing for herbicide applications in rice, which generally include a pre-emergence application followed by one or more post-flood applications at or near the time of flooding, this timing was used to standardize treatment applications and facilitate comparison of effects on insects. All herbicides were applied at their highest label rate. Information concerning the trade names, manufacturers, active ingredients, mode of action, rates, and typical timings for applications of each of these herbicides is provided in Supplemental Table 1. Herbicides were applied using a backpack sprayer pressurized with CO 2 and calibrated to deliver 140 L/ha at 207 kPa through three 1002 flat fan nozzles TeeJet ® at 51 cm spacing.
Treatments in 2016 included all herbicides used in 2015. In addition to these herbicides, the herbicides 'Propanil 4SC' (Propanil) and 'Bolero 8EC' (Bolero) were also used. The two field experiments conducted in 2016 followed the same experimental design as the 2015 experiment, with the exception of the addition of the Propanil and Bolero treatments and two blocks, for a total of seven blocks in each experiment, and one replicate of each treatment in each block. The active ingredients of Propanil and Bolero are both present in Ricebeaux. Therefore, www.nature.com/scientificreports www.nature.com/scientificreports/ in order to determine if either alone was responsible for the effects of Ricebeaux observed in 2015, both were used in 2016 in addition to Ricebeaux. Information regarding dates of application and flooding can be found in Supplemental Table 2.

Insect sampling (objective 2 and 3). All experiments relied on natural infestations of the insects under
study. Insect densities were not measured prior to herbicide applications. Pre-treatment assessments of RWW populations were not made because infestations of larvae occur only after rice is flooded, and pre-and post-flood densities of RWW adults are not associated with post-flooding larval densities 20,42 . Pre-treatment densities of stemborers and rice stink bugs were not assessed because these insects do not infest rice fields until much later in the growing season (see description of pest life cycles above).
The procedure for determining densities of RWW larvae and pupae on roots after flooding involves removing soil/root core samples from plots and counting larvae and pupae associated with roots and has been described in detail elsewhere 43 . Larval densities are expressed as the numbers of immatures per core sample. Stemborer and rice stink bug densities were measured in the 2016 experiments only. Densities of whiteheads found in plots shortly after heading were used to estimate stemborer infestation levels. Whitehead counts were made by visual inspections within one week of anthesis, approximately two months after herbicide applications. Populations were a mixture of the three stemborer species mentioned above, but larvae were not identified to species. The procedure used for estimating rice stink bug density consisted of sweeping plots with 38 cm sweep nets. The length of each plot was swept with ten consecutive sweeps across the plot and numbers of adult and immature bugs captured in nets were counted. The total number of rice stink bugs per plot was used for analysis. Sweeps were made within one week of anthesis, approximately two months after herbicide application.

Rating of rice injury and weed control (objective 2 and 3). Visual assessments of rice injury and weed
control resulting from herbicide treatments were made one and three weeks after application of herbicide. Ratings for injury were based on chlorosis and necrosis of foliage and plant height on a scale of 0 to 100, with 0 = 0 injury and 100 = complete plant death. The scale for weed control was also 0 to 100, with control plots representing the standard for zero weed control.
Statistical analyses. All analyses were performed in SAS version 9.4 (SAS Institute 2013). Means, standard deviations, and standard errors were determined in PROC MEANS. Poisson distribution and log link were used for all count data. All continuous data were tested for normality with PROC UNIVARIATE and the Shapiro-Wilk test (P < 0.05).
For the two RWW mortality experiments (Objective 1), each petri dish containing ten weevils was considered a replicate and the percentage of dead weevils in each dish was calculated and used for analyses. Mortality data were analyzed by generalized linear mixed models using PROC GLIMMIX, with herbicide treatment as the independent variable. Means were tested for significant differences using Tukey's HSD (P < 0.05). This analysis tested the hypothesis that applications of herbicides directly reduced RWW survival.
Each of the three field experiments were analyzed separately. The models described below tested the hypotheses that herbicide-induced resistance indirectly altered RWW densities, borer densities, rice stink bug densities, and yields (Objective 2) or that herbicide-induced changes in in weed communities altered densities of these insects and yields (Objective 3), but the analyses do not allow these two hypotheses to be distinguished.
For the 2015 experiment, multivariate approaches were utilized followed by appropriate univariate tests. Response variables included RWW densities in each of the four core samplings (RWW1, RWW2, RWW3, and RWW4), two ratings of weed density (Weed1 and Weed2), and two ratings of rice injury (Injury1 and Injury2). Each core sampling was maintained as a separate response variable to allow for analysis of effects of time. Canonical correlation analysis (CCA) 44,45 in PROC CANCOR was first performed to determine if significant relationships existed between response variables. Biological relationships were inferred from the canonical cross loadings for individual variables from the two sets of variates. To analyze the twov2016 experiments, CCA was conducted as for the 2015 experiment, but with the addition of three response variables, whitehead densities (Whitehead), rice stink bug densities (Stinkbug), and plot yields (Yield). CCA was performed with one group including response variables RWW1, RWW2, Whitehead, Stinkbug, and Yield, and the second group including response variables Injury1, Injury2, Weed1, and Weed2. One outlier was removed from the first experiment in this year. While CCA is a very informative and appropriate analysis for this experimental design, biological effects are difficult to portray visually. Due to this weakness, and in order to provide additional support for the results of the CCA, simple linear regressions were performed using PROC REG for both 2015 and 2016 experiments to reveal the nature of direct relationships between variables of interest. These regression analyses were performed under circumstances in which neither variable was measured without error because they were both response variables of the herbicide treatments. The CCA and simple linear regression analyses test potential effects of both induced resistance and of altered weed community. In each analysis (CCA and simple linear regression), the level of injury to rice and level of weed control were compared to the subsequent densities of RWW. As CCA and simple linear regression analyses utilize only the response variables in the dataset there is no analysis of effects of herbicide treatments involved in either. One outlier was removed from the CCA in 2015. Simple linear regression was performed as described above for both experiments in 2016 even though the canonical correlation analysis was only marginally significant in the second experiment in this year.
Additional multivariate analyses consisting of a MANOVA followed by pairwise contrasts of each herbicide treatment to the control were performed separately for each experiment in PROC GLM (Objectives 2 and 3). Block was used as a random effect and herbicide as a fixed effect in these analyses. Results of MANOVA and pairwise contrasts were considered significant when the p-value from Wilk's Lambda test was less than 0.05. (2019) 9:6998 | https://doi.org/10.1038/s41598-019-43361-w www.nature.com/scientificreports www.nature.com/scientificreports/ Significant results in the pairwise contrasts indicated that one or more response variables showed significant differences from the control.
Response variables significant at the multivariate level were analyzed by univariate generalized linear mixed models in PROC GLIMMIX. Means, standard deviations, and standard errors were determined in PROC MEANS. The Poisson distribution and log link were used for all count data (insect densities). RWW densities were determined on four separate core sampling dates for 2015 and two separate core sampling dates in 2016. In order to analyze these data, block was used as a random effect and core date, herbicide and the interaction of core date and herbicide were used as fixed effects. For densities of whiteheads and rice stink bug, block was used as a random effect and herbicide as a fixed effect. The identity link and Gaussian distribution were used for continuous data (yields), and block was used as a random effect and herbicide as fixed effect.

Data Availability
The datasets generated and/or evaluated during the current study are available from the corresponding author on request.