Species diversity and varietal diversity within agricultural fields can increase crop resilience to plant pathogens and insect pests. The functional differences between two species in the same family or two varieties of the same species are often less apparent than the differences between species in interfamilial polycultures (e.g., cereal-legume mixtures), but can nevertheless result in yield advantages. Intrafamilial cereal mixtures are grown for their resilience to drought, weeds, and disease in parts of northern Africa, Asia, and Europe, though they were formerly widespread in those regions. Farmers plant multiple cereal species and varieties within the same field, treating the mixture as a single crop. In the northeastern United States, we created mixtures using two varieties of wheat (Triticum aestivum) and two varieties of barley (Hordeum vulgare) to test whether species and varietal diversity would reduce the prevalence of globally important pathogens, the yellow dwarf viruses (YDVs), and their aphid vectors. YDVs are consequential viruses in agriculture, with localized outbreaks causing significant yield loss in Europe, Africa, and North America during the second half of the 20^th century. This is the first experimental study of how varietal and species diversity within an intrafamilial mixture of crop species influences YDV infection prevalence. We found that the wheat varietal mixture had significantly less YDV infection than the average of the wheat varieties grown in monoculture. Aphid pressure was higher in barley monocultures and most mixtures that contained barley, though the yields of species mixtures resembled those of their higher-yielding component, wheat. Aphid pressure may not have been severe enough for us to observe intrafamilial species mixtures lowering viral prevalence, but mixing wheat and barley did not lead to higher disease prevalence or reduced yield. Our experimental results highlight the importance of diversity and identity within intrafamilial mixtures.

Experimental design

To examine the effect of intrafamilial species and varietal diversity on aphid populations and YDVs, we created mixtures using four components, including two hard red spring wheat varieties, Marquis and Glenn, and two two-row malting barley varieties, AAC Synergy and Newdale. We selected Marquis and Glenn wheat for their early maturity so that they would mature at approximately the same time as the barley varieties. All varieties were susceptible to barley yellow dwarf virus (McCallum & DePauw 2008, Mergoum et al. 2006, Legge et al. 2014, Legge et al. 2008).

We planted monoculture stands of the four components, all possible two-component mixtures, and the complete four-component mixture for a total of 11 treatments. This design resulted in: four monocultures; a two-component wheat varietal mixture; a two-component barley varietal mixture; four two-component species mixtures of 50% wheat and 50% barley (all possible combinations); and a four-component mixture composed of 25% of each variety of wheat and barley (Table 1). Hereafter, the mixtures planted in our experiment will be abbreviated to each mixture component’s first letter with a superscript denoting whether the component is a wheat (“Wh”) or barley (“Ba”). For example, the species mixture of Marquis wheat and Newdale barley will be referred to as “species mixture M^Wh + N^Ba”.

We conducted this experiment between May and August 2020 at the Cornell University Thompson Research Farm in Freeville, NY, USA. We planted treatments into 2x2m plots in a randomized block design with six replicates. Plots within a block were separated by 1.5m and experimental blocks were separated by 2m. All plots were seeded at the same overall rate (180kg/ha), with the mixtures composed of equal numbers of seeds of each component in the mixture (Woldeamlak 2001a).  To simulate typical planting methods by smallholder farmers, we broadcasted seeds by hand and tamped seeds into the soil using a land roller. Standard fertilizer for small grain crops in this location (NPK 13-13-13) was applied at a rate of 55.07kg/hectare 10 days before planting. The experimental plot was irrigated as needed. No herbicides, fungicides or pesticides were sprayed, and the plots were weeded by hand.

The growing season of summer 2020 was hot and dry, resulting in fast maturation times for all grain varieties. We planted on May 21, 2020 and harvested the experiment August 5-8, 2020.  Glenn wheat, which usually takes 64 days from planting to reach the heading stage (Mergoum et al. 2006), took only 40 days in our experiment.

Aphid surveys

We surveyed the number and species identity of aphids on every plant within the innermost 25 x 25cm quadrat of each plot. We recorded the number of plants per quadrat and the number of plants occupied by at least one aphid. We conducted four surveys, beginning as soon as aphids appeared on the young plants (3 June 2020) and surveying every other week. From these surveys, we defined two per-plot metrics of aphid pressure: number of aphids and percent of plants occupied by aphids. Percent plant occupancy was defined as the percentage of plants within a quadrat that had at least one aphid feeding on them.

YDV sampling

To estimate the proportion of plants infected with YDVs in each plot, we collected leaf samples on three sampling days (17 June, 6 July, 19 July 2020). We used a portable wooden grid to collect two plants without regard to symptoms from each 25cm x 25cm section of the center 1m² of each plot, for a total of 30 samples per plot. We excluded the innermost 25cm x 25cm of each plot from virus sampling to avoid disturbing aphids within the aphid survey area. Each sample consisted of approximately 5g of leaf tissue and was collected into plastic bags on ice, then stored in a -20°C freezer until being tested for YDV presence.

We tested all samples for BYDV-PAV, BYDV-MAV, and CYDV-RPV using a triple-antibody sandwich, enzyme-linked immunosorbent assay (TAS-ELISA) (Agdia Inc, Elkhart, IN, USA) following D’Arcy and Hewings (1986). We homogenized leaf samples (5g) with 5mL of phosphate-buffered saline (pH 7.4) (PBS) using a leaf extraction press. We coated microtiter plates (2uL antibody/ mL coat buffer) and incubated them overnight at 4°C. After washing the coated plates twice with PBS-Tween (0.5% Tween 20 + PBS), we loaded the sap of homogenized samples into the plates alongside infected and healthy control sap and incubated the loaded plates overnight at 4°C.  Following another wash with PBS-Tween, we added alkaline phosphatase-conjugated immunoglobulin (2uL + 2uL antibody / mL conjugate buffer) to the wells and incubated the plates for four hours at 37°C. Each well received 100uL of substrate (1mg p-Nitrophenyl phosphate disodium salt hexahydrate substrate powder / mL substrate buffer) and was incubated at 37°C for 50 minutes before measuring reactions using a microtiter plate autoreader. At 405 nm, we considered samples with optical density values at least 1.75 times the mean of the OD of healthy control wells to be positive for BYDV or CYDV.

Yield

We harvested aboveground biomass from all plants within the innermost 1m² of each plot once the entire experiment had reached maturity. We separated out the components of every mixture and separated grain heads from straw. We were unable to clearly distinguish between Newdale barley and AAC Synergy barley in mixture (Fig. 7). Since their yields did not differ significantly in monoculture, we assumed that the two varieties had equivalent yields when planted in mixture and split mixtures of the two barleys evenly. We allowed plants to air dry for 6 months before recording grain and straw weights. We also counted the number of seed heads to calculate the average weight of one seed head.

Statistical analysis

We conducted all analyses in R, version 4.0.4 (R Development Core Team 2021). All generalized linear mixed models (GLMMs) were fit using the glmmTMB function from the glmmTMB package (Brooks et al. 2022). For each of our metrics -number of aphids, percent plant occupancy, percent plant infection, yield measures- we created two separate models, one with treatment as the predictor variable and one with mixture type as the predictor variable. We grouped treatments to generate the different mixture types (see Table 1). For example, the “monoculture wheat” type included the Marquis wheat monoculture and Glenn wheat monoculture, and the “2 species mixture” mixture type included all the two-species mixtures: M^Wh+A^Ba, M^Wh+N^Ba, G^Wh+A^Ba, and G^Wh+N^Ba. In all models, the predictor variable of treatment included all possible mixtures and monocultures. To create consistent lettering within Fig. 4, we excluded the two varietal mixtures from our analysis for only that one model. None of the significant differences between treatments changed.

To analyze percent of plant occupancy by aphids, we used a binomial GLMM with a logit link. Our first model included sampling period as an interaction with percent occupancy, but since the interaction between sampling period and treatment was not significant in our ANOVA (Type II Wald Chi-square), we based all comparisons on a final model that included sampling period as an additive fixed effect. In this final model, block and plot number nested in block were also random effects, and we accounted for variation in plant density among plots by including the number of plants per quadrat as an offset. The number of aphids was analyzed using a negative binomial distribution GLMM with a log link. Sampling period had a non-significant interaction with the number of aphids when assessing the initial model with ANOVA so it was included as an additive fixed effect. The final model included random effects of block and plot nested within block, and the number of plants per quadrat as an offset. Since aphid species abundance varied across sampling periods, we also ran occupancy and aphid abundance models separately for the two most abundant aphid species.

We analyzed the percent of YDV-infected plants sampled from the innermost 1m² of each plot using a binomial GLMM with a logit link and sampling period as a fixed effect. Our model included block and plot number nested in block as random effects and an offset by the number of plants per quadrat. To analyze yield metrics (grain, straw, number of seed heads, seed head weight), we used a Gaussian LMM with a log link and included block as a random effect.

We tracked the outcome of planting each variety in different mixture contexts, and we adjusted the yield of each variety to the equivalent of a half-plot area (0.5m²). Since the initial seeding rates were the same for every variety, this allowed us to compare the yield of an equal number of seeds of each variety grown over the same land area among monocultures and mixtures. We compared the monoculture yield means to the combined yield mean of the three species mixtures for each variety (Fig. 7). For each variety, we also compared the monoculture yield means to the individual yield means of the three species mixtures separately (Fig. S1).

For each of these models, we used the emmeans package (Lenth 2023) to conduct pairwise comparisons between our Tukey-adjusted Estimated Marginal Means (least-squares means). We used the contrast() function from emmeans to build our custom contrasts so that we could compare each mixture to the unweighted average of its components. For example, to determine whether mixtures had higher or lower percent aphid occupancy than their components grown in monoculture, we tested whether the mean percent occupancy in a mixture is significantly different than the unweighted average of its two components’ mean percent occupancy.