TrackGene: video-tracking insect resistance genes in field and greenhouse crop populations

TrackGene: video-tracking insect resistance genes in field and greenhouse crop populations

Organisatie-onderdeel

WR-cap TU

Projectcode

LWV19167

MMIP

Landbouw, Water, Voedsel>Sleuteltechnologieën LWV>Biotechnologie en Veredeling

Startdatum

01/04/20

Einddatum

01/04/24

Samenvatting

There is a rapidly increasing demand for robust crops with resistance to pests. On the one hand there is a direct economic benefit for the farmer if he can spend less on inputs to control pests, and on the other hand society, motivated by environmental and health concerns, imposes more and more limits on the use of pesticides such as neonicotinoids (ambition 2023 of this call, MMIP-A,B,D). Insect resistance is usually polygenic and can be introduced from wild germplasm if plant breeders have access to the best tools to phenotype plant populations in conjunction with powerful genetics tools like genomic prediction.
The aim of this project is to make the video tracking method of identifying resistance directly useable in plant breeding by direct association of the video tracking data to the genetic variation of the tested plant populations through the development of TrackGene, a package that performs statistical genetic association analysis and optimizes settings by machine learning with a graphic output to quickly assess results. The functionality will be tested by a unique international consortium of 3 plant breeders of horticultural and field crops, each contributing their own genotyped populations. The broad variety of crops including sweet pepper, maize, and rice and the broad variety of pests (thrips, maize aphids, plant hoppers, etc) guarantees a major impact on the industry when the project is successful both in terms of genomic prediction models for resistance in the various crops and in terms of QTL detection technology. The project, furthermore, has a strong dissemination component. All companies will be trained in the use of the software on their own populations and may purchase the platforms for their own use. The user company from India will be trained to apply the video tracking technique on site in India and share the experiences and data with the development team in Wageningen. Wageningen Plant Research will take up the further development of the EntoLab assay platform with dedicated assay plates for the different pest insects and technical optimizations.

Doel van het project

The extensive use of insecticides in agricultural production has been shown to be an unsustainable model. Instead the use of cultivars with genetic tolerance to insect pests represents a highly desirable trait which will benefit the environment, the farmer income and human health. There is a strong demand therefore for techniques to rapidly identify relevant germplasm and genetic loci that confer useful levels of insect control. Insect behavior is a very rapid and early indicator of host plant quality and can be monitored by videotracking insect behavior. These statistics insect pest plant phenotypes can be correlated directly to the plant’s genetics. This fits under the aim of Smart Technology ST2 Biotechnology and Breeding: Development of supporting technology to provide agriculture with robust insect resistant crops (missions A,B,D) by combining insect video tracking with statistical genetics and machine learning, to develop new approaches for genomic prediction and QTL-mapping (subprograms 1,2,3). This program has attracted the participation of Dutch/German, French and Indian plant breeding companies.

Relatie met missie (Motivatie)

Conventional assays to phenotype for plant resistance to insects are less effective because they mostly score population size only (no insight in the basis of the trait), take 2-3 weeks to complete (high cost and environmentally variable), are difficult to implement with quarantine organisms, and only generate single data points per plant (no robust information). Recently, video tracking (EntoLab) was introduced by WPR and Noldus IT as a promising alternative or complement which requires only a few hours of recording of multi-arena assay plates with insects placed on a leaf fragment taken from a plant/genotype in its regular field growth condition. Typically ~20 insects are independently video tracked per plant in their own arenas generating over 30 robust behavioral parameters in time bins or as assay average with strong statistical power for each plant. Assays can be run in both choice and no-choice formats and the large rich data sets were shown to be an excellent basis for the sensitive detection of novel resistance traits (Kloth et al. 2017, Jongsma et al. 2019). However, given the sensitive camera systems and often very subtle feeding behavior of insects depending on crops and traits, a major (technological) issue is the optimal setting of various movement and duration thresholds, like for example “above which speed/duration is a small change in position a movement”. Currently, the choice of such thresholds is manual and based on prior knowledge of the resistant genotype. Without this knowledge in the large-scale analysis of new crops, traits and insects it is highly desirable to automate these procedures using the genetic association success as a basis for machine learning, and to use all available data.

Geplande acties

After automatic parallel video tracking of insects on leaves in isolated arenas, the current EntoLab system (see Annex 2 for details) generates over 30 behavioral parameters per insect (e.g. distance moved, duration moving/halting, number of halting/moving events, etc.), and provides an extensive toolkit to compare behavior on different plant accessions. In this project for the first time, the system will be applied for genetic analyses of 3 different crops (rice, maize and pepper) and Arabidopsis. However, the system is not yet well designed for efficient genetic analyses in large populations, such as QTL-mapping or genomic prediction.

We aim to improve the design in two areas:

1. Experimental design. Given the biological/physiological constraints, such as geometry of the leaves and the available plant material, an efficient experimental design should be chosen that maximizes the power for genetic mapping studies and avoids spending valuable resources on unnecessary (pseudo-)replications. Guidelines are needed to construct efficient trial designs and for choosing the optimal numbers of genotypes, plants per genotype, number of leaves per plant, and insects per leaf. Designing assay plates to accommodate different insects and allow maximum efficiency is a second focus. (2020-2021)
2. Tuning and choice of behavioral parameter(s) most relevant to breeders. Hours of video-tracking data lead to an almost infinite dimensional phenotype. The current system converts these data to over 30 pre-defined parameters based on a number of threshold settings. The number/choice of parameters and the dependence on the settings raises two challenges for breeding:
• Development of algorithms for automatically selecting the most relevant settings and behavior parameters. This problem is crop and insect specific, making an automated procedure indispensable. Relevant parameters should have high heritability, which is the proportion of phenotypic variability that can be attributed to genetic factors. This directly determines the feasibility of genomic selection (Eeuwijk et al. 2018), as well the power of QTL-mapping, and is therefore of central importance for breeders. Similarly, behavioral parameters should have large genetic correlation with resistance phenotypes in the field, which requires separate checking (e.g. population build up, avoidance). (2021-2022)
• Construction of entirely new parameters from the raw tracking data, with advanced statistical and machine-learning methodology. Until now parameters and settings have been iteratively optimized based on datasets of resistant and susceptible varieties; however this knowledge may be incomplete, and in any case is insect, trait and crop specific. Recent work in several areas suggests that more meaningful parameters may be obtained, in particular using state-space models, deep learning models and from reverse QTL-mapping. (2022-2023)

Naam projectleider

Maarten Jongsma