Joint Management of Wheat Stem Sawfly, Fusarium Crown Rot, and Weeds: Assessing the Ecological Basis of a Total Systems Approach to Pest Management Strategies

Results from our experiments indicate that yield differences can be readily and fully explained as a result of the effects of management on pests and multi-pest interactions, rather than just by the direct effect of any particular management scheme on yield. Different pest interactions under different management schemes and environments (sites, years) make a total systems approach to the analysis imperative and highlight the need to update current single-pest management recommendations.

The low disease incidence and low cheatgrass pressure observed during our study period at the Ulm precluded us to assess the joint response of pathogens and weeds to management practices utilizing data gathered at this site. Nevertheless, we were able to accurately analyze the joint response of Fusarium and cheatgrass to management practices with data obtained at the Amsterdam (2008) and Havre (2009) sites. Cheatgrass biomass and cover increased the odds of high Fusarium intensity for all management practices but had a less consistent effect across management practices on Fusarium infection (Fig. 2).

Because Fusarium primarily develops during periods of water stress, disease severity is expected to increase with increasing plant biomass, and cultural controls are often related to moisture management such as reduction in seeding density (Cook, 1980; Paulitz et al., 2002). Therefore, we expected that disease severity to be lower at reduced wheat seeding densities. Unfortunately, this strategy may not be suitable for wheat productions systems in the NGP, as reducing seeding rates is counter to weed control strategies. In contrast, high initial seeding rates may help suppress cheatgrass emergence and plant vigor, mitigating the effect of cheatgrass on Fusarium. Thus, to properly model the effects wheat cultivar and seeding rate on Fusarium, it is necessary to incorporate both the direct and indirect effects of the presence of cheatgrass on disease abundance and severity.

The joint evaluation of the impact of management practices (wheat variety and seeding rate) on disease and weed pressure indicated that the drought tolerant cultivar, also resistant to Fusarium, significantly suppressed cheatgrass at high seeding rates (Table 1), which explains the unexpected significant reduction in Fusarium intensity observed in that treatment. The significant increase in Fusarium in the WSSF tolerant cultivar could be explained by similar logic, except in this case the WSSF tolerant cultivar is a weak competitor and failed to suppress cheatgrass (Table 1). Although the magnitude of the effect of cheatgrass on Fusarium reported in Table 1 may seem low, it is the effect of every one gram increase in cheatgrass. Thus, for example, the odds ratio of high intensity Fusarium infection in the vicinity of a 100 gram plant of cheatgrass is approximately 6 (e100*.018), with a 95 % highest posterior density of 1.5 – 14.

The Fusarium – cheatgrass interaction results presented in Table 1 were obtained under extreme levels of WSSF infestation and refer to uncut (lodged) stems only due to our inability of visually determining Fusarium status for lodged stems. To assess the accuracy of these results and further investigate the joint impact of pathogens and insects on crop yield, we collected 46 paired of WSSF cut and standing samples of 20 stems each from a range of crop variety and seeding rate conditions. These samples were visually scored for Fusarium as well as thru quantitative PCR. Pair-wise comparisons of cut versus uncut samples showed no statistical differences in Fusarium populations (t = -1.10, p = 0.278). This result indicates that, first, Fusarium populations within cut stems are similar to those described by visual assessments of neighboring uncut stems. Second, the Fusarium – cheatgrass interactions detected in Table 1 are not confounded by the presence of WSSF. Finally, this experiment showed differential measurement errors in visual Fusarium assessment by treatment Agreement between the two methods to assess disease status (visual and qPCR) was high for WSSF tolerant cultivar, and low for the other two (Fig. 3). This result suggests that WSSF larval burrowing could suppresses the symptoms of Fusarium, but further tests are required).

Very heavy WSSF infestations were observed at all three Montana sites in all years. To assess WSSF levels, lodged stems in the sample rings (total 729 rings per year) were separated from the rest of the sample, counted and threshed for grain yield. Once Fusarium scoring was completed, individual stems were dissected to assess WSSF larva infections and threshed for grain yield. WSSF data were scored in four categories; 1) uni: a clean stem were no larva is found, 2) inf: a stem were a dead WSSF larva is found, 3) par: a stem where a parasitoid of WSSF was found, i.e. the stem was attacked by the WSSF but the larva or eggs were killed by a parasitoid, and 4) cut: the WSSF completed the life cycle and caused the stem to lodge before harvest. In this study, WSSF cutting was assumed to be equivalent to a 40% yield loss, as 10-20% loss is attributed to the larva burrowing through the stem during the growing season (Holmes 1977) or losses may be greater when plants are nutrient or water deficient (Delaney et al 2010) and approximately 25% results from failure to harvest due to stem lodging (Ainslie 1920, Beres et al 2007). Under low cheatgrass pressure, we found WSSF attack rates and cutting rates significantly lower on the solid stem WSSF tolerant cultivar (p < 0.001) (Table 2). However, these differences did not translate into significant yield increases at the plot level. This result may be attributed to our research size combine collecting more lodged stems than a commercial one could or the inherently higher yields of the drought tolerant and high yielding cultivars under dry conditions, which compensated for the increased cutting and burrowing.

The impact of WSSF on crop yield was modified by the presence of grassy weeds. Specifically, cheatgrass modified WSSF oviposition choices (Fig. 4) and cutting rate (Fig. 5). WSSF oviposit in cheatgrass present at low densities in wheat fields despite low success rates for completing its life cycle (Perez-Mendoza et al. 2006). Consequently, we observed a decrease in WSSF attack rates on wheat as well as an increase in attack rate variability as cheatgrass levels increase (Fig 4). This effect was dependent on cultivar, with the WSSF tolerant variety having the lowest mean attack rates under low cheatgrass pressure but exhibiting the most change in the shape of the posterior distribution with increasing cheatgrass levels. In contrast, the drought tolerant variety, which is the more competitive cultivar and thus less impacted by increasing cheatgrass levels, exhibited a decreasing mean attack rate and increasing variability.

The observed differences in cheatgrass impact on wheat yield across cultivars suggest that WSSF choice was not modified simply due to the increased frequency of an alternative host (cheatgrass). Since WSSF is known to oviposit based on stem diameter, a “proxy” for fitness based on the fact that haplodiploid hymenopteran species produce more females from wide stems and more males from narrow ones (Wall, 1952, Morrill et al. 2000, Carcamo et al. 2005) reference), it is likely that plant competition with cheatgrass resulted in a higher frequency of smaller and narrower wheat stems, which are subsequently less preferred by WSSF females (Buteler et al. 2009). Further evidence of the role of plant competition in mediating WSSF oviposition choices is the fact presence of cheatgrass competition reduces wheat fitness (Fig. 6), which resulted in increased lodging in WSSF tolerant and high yielding cultivars, especially at high seeding rates, but not for the more resilient drought tolerant cultivar (Fig. 5).

When jointly considering the three pest groups (insects, grassy weeds and pathogens), the presence of biological interactions occurring at different trophic levels may, in turn, lead to non-intuitive outcomes of management practices. Despite these difficulties, our decision models accounted for the presence of multiple direct and indirect interactions of management treatments on crop and pests.

Path analysis can be a useful tool in natural resource management where indirect effects can have significant impacts on outcomes. A path model consists of all direct and indirect effects in the system, but current methodology relies on calculating indirect effects by post hoc tracing all direct effects along every path. For complex networks with a large number of connected indirect paths, this process, as well as extracting accurate standard errors for path coefficients, becomes cumbersome. We developed a model fitting procedure which, when implemented in a Bayesian Decision Network context, directly derives total effects (direct and indirect) and their variance estimates for all variables in the path model. Our procedure also overcomes problems of convergence and bias which are common in complex network analysis (Fig. 7) and allows easy derivation of the estimate error and construction of confidence interval for path coefficients. Also, since all paths (direct and indirect) are derived in one step, it may help expand implementing total systems models to more complex networks and make more accurate predictions of the long term effects of treatments on crop in agronomic settings.

The Bayesian network approach to path analysis allowed us to provide probabilistic answers to outcomes of management decisions. It also allowed us to make predictions based on specific management goals, such as maximizing yield in the presence of multiple pests occurring at different trophic levels. For example, WSSF tolerant varieties should be planted at a low seeding rate under high WSSF pressure. However, this variety should be replaced by a more competitive and drought tolerant cultivar at high seeding rates as cheatgrass levels increase, despite the persisting WSSF infestation (Fig. 8).

• Figure 3. Frequency of stems in each Fusarium level category of the visual method, sorted by number of Fusarium DNA copies determined with qPCR (red curve). For the WSSF tolerant, except for an odd observation in position six, visual estimates generally follow the qPCR results while the other cultivars do not, especially at high levels.
• Figure 4. Fraction of WSSF infested wheat stems in Amsterdam, MT 2009. Samples were grouped by cultivar (rows) and quartiles of cheatgrass visual estimate of volume (m-3) (columns). The overlay beta distribution curve represents the posterior oviposition rate with a non-informative prior.
• Figure 6. Wheat yield in the sampling rings (not adjusted for losses due to WSSF burrowing) relative to cheatgrass abundance at two sites in 2009 for all cultivar by initial seeding rate combinations. A similar pattern was observed in 2008 (data not shown).
• Figure 7. Four MCMC chains in a simulated simple triangle network (X2 = ßX1+ex ; Y = ?1X1 +?2X2+ey) with normally iid errors. The fit was done through a “classical” path analysis procedure (bottom row) or with the new procedure (top row). All simulated direct effects have a true value of 1, thus the combined direct and indirect effect is equal 2 (top left panel). The superiority of fitting the new estimators a’s over the old ?’s is demonstrated by: 1) convergence is achieved faster, Gelman and Rubin diagnostic Rhat is close to 1 (less then .05) after 50 iterations, 2) Any bias introduced by collinearity in ?’s is less apparent in a, 3) The algorithm is better at “sharing the information” contained in X1 and X2 to get joint estimates, as apparent in the convergence path of chain number 1 (solid black line), and 4) an estimate of direct and indirect effects is immediately derived with correct variance in a1 with no need to calculate it from separate direct path coefficients (top left panel).
• Table 1. Response of cheatgrass and Fusarium infection and intensity to wheat management practices. Models parameterized using values obtained at the Havre, MT location in 2009.
• Figure 2. Effect of 1 g increase in cheatgrass on the odds ratio of Fusarium infection (left panel) and Fusarium intensity (right panel) at the Amsterdam field site in 2008. Distributions are 95% highest posterior densities for every high yield (HY), drought tolerant (DT) and WSSF tolerant (SFT) cultivar by seeding density (low – l, medium – m and high – h) combination. Letters on the right side of left panel denote significant (>95%) overlap between treatment combinations. All posteriors overlapped in the right panel. Posteriors were generated independently for each treatment combination.
• Figure 5. Lodging (WSSF cutting) rate modified by cheatgrass plotted for cultivar and initial seeding rate. Size of the point character represents number of stems available, and thus also precision of the estimate.
• Figure 8. Posterior predicted yields of WSSF tolerant and drought tolerant at low and medium seeding rates. Models were parameterized using pest and crop values obtained in Amsterdam, MT in 2008. WSSF levels were high and assumed a 40% loss of yield from cut stems. Fusarium levels were low and assumed zero in this model. Percent cheatgrass referrers to a fraction of the maximum cheatgrass biomass observed in this field.
• Odds ratio* for rates of WSSF oviposition rate and lodging by cultivar for cheatgrass free sub-sub-plots at all three Montana sites over two years. Values reported are of the mode and 95% highest posterior density interval of 10,000 draws.