Genetic diversity is increasingly recognized as crucial for biodiversity and ecosystem services. In terrestrial ecosystems, plant genotypic diversity can enhance resistance to natural enemies, although mixed planting doesn't always yield positive results. Associational resistance (increased resistance in mixed plantings) and associational susceptibility (decreased resistance) involve complex ecological interactions, making it difficult to identify beneficial genotype pairs. While some studies have used genome-wide association studies (GWAS) to investigate stand-level properties, they were limited by the effort required for pairwise cultivation. This study aimed to predict key genotype pairs reducing herbivory by combining genome-wide single nucleotide polymorphisms (SNPs) in *A. thaliana* with a novel GWAS method, "Neighbor GWAS", which is inspired by the Ising model and accounts for neighbor genotype effects. This method offers a practical advantage by being applicable to randomized mixtures of many genotypes, enabling analysis of how neighbor genotypes influence herbivore damage across space. The study involved field experiments with replicated individuals of 199 *A. thaliana* genotypes at two sites, followed by Neighbor GWAS analysis and genomic prediction to identify beneficial pairs. These predictions were then validated through further field experiments.

Extensive research highlights the impact of genetic diversity on plant resistance to herbivores. However, the outcomes of mixed planting can be variable, ranging from enhanced resistance (associational resistance) to increased susceptibility (associational susceptibility). These phenomena arise from complex ecological interactions including plant-herbivore, plant-plant, and plant-carnivore interactions. Plant traits like odor and physical barriers influence herbivore behavior, and these traits can be modulated by plant-plant interactions through volatile communication or competition. Plant-carnivore interactions can also mediate associational resistance or susceptibility. Despite the importance of associational resistance in reducing insecticide use, few studies have leveraged genome-wide polymorphism data to investigate stand-level properties in biodiversity research. Previous studies using GWAS to dissect the genetic basis of stand-level growth in *A. thaliana* were often limited to smaller numbers of genotype pairs in controlled environments. This constraint hindered large-scale GWAS for identifying beneficial pairs in anti-herbivore resistance in field conditions.

This study employed 199 *A. thaliana* genotypes planted in a randomized block design at two field sites (Zurich, Switzerland, and Otsu, Japan) over two years. Herbivore damage (leaf holes in Zurich, leaf area loss in Otsu) and insect community composition (external chewers, other herbivores, total species) were assessed. Neighbor GWAS, a linear mixed model incorporating both focal and neighbor genotype effects, was used to quantify phenotypic variation explained by neighbor genotypes. A standard GWAS was conducted to examine focal genotype effects. The Neighbor GWAS model, structured similarly to the Ising model, allowed the identification of SNPs with positive (associational resistance) or negative (associational susceptibility) neighbor genotype effects. Genomic prediction using LASSO (least absolute shrinkage and selection operator) was employed to identify key genotype pairs for herbivore damage prediction. 756 neighbor-related SNPs were selected for this prediction. Finally, three genotype pairs with predicted positive effects were planted in mixtures and monocultures to validate the predictions in field experiments. Laboratory choice experiments with black flea beetles were also conducted to complement field observations. Statistical analyses included likelihood ratio tests, permutation tests, and analysis of variance.

Field experiments revealed quantitative phenotypic variation in herbivore damage and insect community composition among individual plants. Standard GWAS showed significant heritability for all four phenotypes, with significant effects of the *GL1* gene on herbivore damage in Zurich. Neighbor GWAS demonstrated that including neighbor genotypes significantly explained a portion of the phenotypic variation in herbivore damage, particularly affecting mobile herbivores (flea beetles in Zurich and thrips in Otsu). GWAS of neighbor genotype effects did not identify significant SNPs, suggesting a polygenic basis for neighbor effects. Analysis of SNPs with positive or negative neighbor genotype effects revealed that SNPs associated with associational resistance (positive effects) involved more minor alleles and showed more signatures of balancing selection than those associated with associational susceptibility. Genomic prediction using LASSO identified 823 genotype pairs predicted to reduce herbivory. Field experiments with three selected pairs confirmed a significant reduction (18–30%) in herbivore damage in mixed planting compared to monocultures, with the magnitude of reduction aligning with the predicted effect sizes. Laboratory choice experiments corroborated the field results, showing significant differences in herbivore damage between the high-effect pairs but not the low-effect pair.

This study successfully predicted key genotype pairs underlying associational resistance despite the prevalence of associational susceptibility. The use of genomic prediction with Neighbor GWAS overcame the challenges of distinguishing between these two phenomena, allowing for the identification of beneficial pairs in large-scale field experiments. This approach differs from previous studies that were limited to smaller numbers of genotype pairs in controlled environments. The genomic prediction strategy, similar to genomic selection in plant breeding, identifies elite genotypes without necessarily identifying the specific genes responsible. Gene ontology enrichment analysis suggested a potential role of jasmonate-induced defense in associational resistance, specifically highlighting the involvement of LOX genes known to influence volatile production. The study's findings highlight the importance of field data for understanding ecological interactions under naturally fluctuating conditions and the species-specific response to mixed planting, as evidenced by the differing effects on mobile versus sedentary herbivores. This approach provides a practical strategy for designing resistant mixtures in agriculture, offering a potential pathway to reduce insecticide use without compromising agronomic management.

This research demonstrates the successful prediction and validation of key genotype pairs that promote associational resistance to herbivory in *A. thaliana*. The combination of Neighbor GWAS and genomic prediction provides a powerful tool for identifying beneficial genotype mixtures. The findings highlight the potential of this approach for improving crop pest management by reducing the need for chemical interventions. Future research could focus on dissecting the precise mechanisms underlying these interactions and expanding this approach to other plant species and pest systems.

The study focused on a specific set of *A. thaliana* genotypes and herbivore communities. Generalizability to other species and environments requires further investigation. The polygenic nature of the effects makes it challenging to pinpoint specific genes responsible for the observed phenomena. The study predominantly addressed herbivore damage; the impact on the broader ecosystem and potential indirect effects require further exploration. Finally, while the laboratory experiments supported the field findings, they used a controlled setup that may not fully replicate the complexity of natural field conditions.