TraitMill: serial cereal phenotyping

An enormous, high-security greenhouse in the countryside 20km north of Ghent in Belgium hints at the high-technology future of crop breeding in the 21st century.

TraitMill, designed and constructed by Belgian plant biotechnology company CropDesign, is a marvel of engineering and automation. It is the world's largest corporate facility for applied genomics research and development in cereals.

CropDesign was spun out of the Flanders Inter-University Institute for Biotechnology in 1998, then acquired by German agbiotech giant BASF Plant Science this year.

CropDesign's CEO, Dr Johann Cardoen, told the recent ABIC conference that the high-throughput TraitMill platform allows crop phenotyping - evaluation of the structural and functional effects of transgenes - to be performed on an industrial scale.

The process begins with routine gene discovery; researchers use statistical techniques to identify quantitative trait loci (QTLs) harbouring genes of potential commercial interest to plant-breeding companies.

QTLs identify chromosomal ballparks in which to search for specific genes making significant contributions to complex traits like increased yield, increased tolerance of environmental stresses such as drought, salinity and frost, increased biomass, and improved nitrogen-use efficiency. "We then construct different gene-promoter combinations, and carefully select which ones go into our pipeline," Cardoen said.

"On average, each gene is tested with two or three different promoters, and we make thousands of gene constructs per year, generating around 50,000 independent lines.

"We generate 50 independent transformation events, select five or six single-locus transformants for evaluation, and compare the transgenic versions with hybrid segregants - 50 per cent of the greenhouse is occupied by non-transgenic segregants."

Cardoen said the company looked at hundreds of thousands of transgenic plants, grown under greenhouse conditions, every year. "We look to simulate the field situation as closely as possible - we can control a range of environmental factors in the greenhouse, such as spacing between plants, plant responses to a variety of nutrient regimes, nutrient deficiency and a wide rang of pHs, from acid to alkaline."

How it works

The individual plants are grown in transparent pots so root development and architecture can be recorded, he said. Each pot is equipped with a tiny transponder that transmits its serial number to a computer as it passes an interrogation point.

The plants are randomly shuttled around the greenhouse on a network of intersecting, computer-controlled conveyor belts, negating any positional effects of temperature and sunlight on growth.

At nodes around the greenhouse, pots can be diverted onto other conveyor-belt loops for specific treatments, or for imaging by digital cameras that record growth habit, panicle development, root development, and estimate biomass.

Biomass calculation is done by rotating the plant in front of the camera, and taking six shots at 60-degrees of the above-ground tissues, silhouetted against a neutral background.

By subtracting the area of frame occupied by green pixels from the frame's total area, and applying a computer model of the crop's area-biomass relation, the plant's total leady biomass - a reliable proxy for total biomass - can be calculated with considerable accuracy.

Plant height, the number of tillers, and panicle size and architecture, are also extracted automatically from the images. Promising statistical outliers are flagged for real-world, expert inspection in the company's proprietary database. Cardoen said the imaging and data-collecting system generates 10 terabytes of data each year.

CropDesign is focusing on japonica rice, but does some work on maize, testing promising gene-promoter combinations from the rice program. Seed harvesting is the only manual operation.

The company has developed a way of representing the trait measurements graphically, in an easily interpreted spider-web diagram that shows how individual productivity traits are linked and contribute to the overall result.

The diagram displays parameters such as root biomass, green biomass, time to flower, total seed weight, weight of 1000 individual kernels, filling rate, the number of panicles, the number of flowers per panicle.

The diagrams also facilitate comparisons of the effects on a single breeding line of different nutrient regimes, nitrogen deficiency, salinity, water stress and other nutrient stresses and other environmental influences on a particular plant line.

"We can dissect out the contributions of the individual traits, and thus tell what genes are doing," he said.

Increased average yields

The approach works - spectacularly. The genes are proprietary secrets, but Cardoen gave the example of a gene that, in eight out of eight transformation events, increased average seed yields by 112 per cent - more than double that of its non-GM precursor.

"The effect of the gene on panicle size was most striking," he said. "We're very excited about taking some of our lead genes into maize - it will be around one in four of those that work in rice."

At any given time, more than 100 lead varieties will be going "through the mill", literally - or undergoing field verification.

"In stage 1, any lead validated in the glasshouse is taken to the field for validation, then earmarked as a candidate for commercialization," he said. "In stage 2, field efficacy is translated into commercial germplasm, and the product is developed as a candidate for regulatory approval. "In stage 3, we obtain regulatory clearance and release a market-ready product."

With high-throughput phenotyping capabilities, the company sees great potential synergies with other companies in the crop development business. "For example, we're working with a company that does high-throughput metabolic profiling, that we can correlate with out our phenotyping results."

Currently, CropDesign is only phenotyping genes that contribute to yield.

"We deliberately chose not to measure all the measurable parameters in a plant," Cardoen said. "We could evaluate many other traits, but the number of transformed plants, and the number of different treatments, would rapidly become unmanageable."