Calibration of a probabilistic model of oilseed rape fertility to analyze the inter-variety variability in number of seeds

Xiaojun Wang, Amélie Mathieu, Paul-Henry Cournède, J. M. Allirand, Alexandra Jullien, Philippe Verchère de Reffye, B.g. ZhangAgritrop 2010

Abstract

Predicting plant yield from genetic information requires models with parameters tied to strong genetic determinants. This study calibrates a probabilistic model of flower fertility to analyze seed production in four oilseed rape varieties (Mendel, Gamin, Exocet, and Pollen). By simulating ovule formation, pollination, fertilization, and abortion, the researchers identified variety-specific parameters. Results indicate that while pollination and resource competition generally impact seed production similarly, the variety 'Gamin' exhibits a unique strategy: high rates of fertilized ovule abortion compensated by significantly higher individual seed weights.

Introduction

In modern agriculture, scaling from genotype to phenotype remains a significant challenge. Understanding the variability in the number of seeds per pod is crucial for predicting overall yield. Seed production is a multi-stage biological process involving ovule formation, pollination, fertilization, and potential abortion due to resource constraints. This research aims to develop a mathematical framework to distinguish these factors and identify biological parameters inherent to specific plant varieties.

Methodology

The researchers developed a probabilistic model treating each stage of seed development as a statistical distribution:

  • Ovule Formation: Modeled using a Binomial distribution, where $N$ represents the maximum potential ovules.
  • Pollen Arrival: Modeled via a Lognormal distribution to account for the stochastic nature of pollination.
  • Fertilization: Defined as the minimum of available ovules and pollen grains.
  • Seed Viability: Modeled as a Bernoulli process, where a fertilized ovule has a probability $p$ of reaching maturity.

The model was calibrated using data from four oilseed rape varieties grown in Grignon, France. To eliminate position-dependent effects, 15 pods were collected from ranks 11 to 40 on the main stem of ten plants per variety. Measurements included the number of successful seeds, aborted seeds, and total seed dry weight.

Results and Discussion

The model successfully captured the variability between varieties. The maximum number of ovules ($N$) ranged from 36 in Mendel to 45 in Exocet. Interestingly, even under identical field conditions, pollen landing parameters varied, suggesting that genetic traits like flower architecture may influence pollination efficiency.

Observed and computed frequency distributions of the number of seeds per pod across four oilseed rape varieties.

A key finding was the distinct behavior of the Gamin variety. While other varieties showed high seed viability (e.g., Pollen at 0.916), Gamin had a significantly lower seed viability probability ($p \approx 0.573$) and a low ratio of seeds to fertilized ovules (approximately 52%). However, this lower seed count was offset by a significantly higher mean seed weight ($P < 0.001$).

Comparison of mean seed weight per grain, illustrating the compensatory strategy of the Gamin variety.

This suggests a resource allocation trade-off: Gamin invests more heavily in fewer seeds. Across all varieties, pollination issues typically accounted for a 10% loss in potential seeds, though this rose to 18% for Gamin.

Conclusion

This probabilistic approach allows for the separation of environmental factors (like pollination) from internal resource competition (abortion). Future work will expand the model to include the entire plant architecture, comparing the main stem's behavior with that of its branches to provide a more comprehensive prediction of total plant yield.

Summary updated

Abstract

With the objective of using plant models as predictive tools scaling from genotype to phenotype, model parameters should have a strong genetic determinant. For this purpose, the modeling process involves assessing the differences in model parameters between varieties. In this study, a model of flower fertility is used to explain the observed behaviors and to identify the variety related parameters relevant to seed production. The model simulates the steps of seed production: ovule formation, landing of pollen grains on a flower, fertilization of ovule by pollen grains, possible abortion of the fertilized ovules. The aims of this study are to assess the differences of estimated parameters and identify the factors that can explain observed differences among varieties in the number of seeds per pod. Four varieties of oilseed rape (Mendel, Gamin, Exocet and Pollen) were grown at the experimental station of Grignon, France. Ten plants were marked and 15 pods from rank 11 to 40 (to eliminate the effect of position) on the main stem were collected on each plant for each variety. The numbers of seeds and aborted seeds were recorded. The total seed dry weight for each plant was measured. The maximum number of ovules per flower is different among the four varieties with the range of 36-45. The landing of pollen grains on a flower was significantly different among the varieties, although they were grown in the same field. The estimation result of model allows us to conclude that pollination and resource competition have the similar impact on the ovule and seed abortion. However, for the variety Gamin, the probability of fertilized ovules to abort was quite large in agreement with the smaller number of seeds measured but mean seed weight was higher than others (P<0.001, ANOVA). The data analysis indicated that the small number of seeds per pod was compensated by a higher seed weight. The abortion of seeds could result from insufficient pollination and resource competition. Current work aims at quantifying more precisely the roles of these factors and investigating other ones. We intend to use the model to distinguish the effects of different factors on seed production, such as the plant architecture. One way is to compare the behaviors of the main stem, the ramifications and the plants.
BibTeX
@inproceedings{wang2010calibration,
  author = {Xiaojun Wang and Amélie Mathieu and Paul-Henry Cournède and J. M. Allirand and Alexandra Jullien and Philippe Verchère de Reffye and B.G. Zhang},
  title = {{Calibration of a probabilistic model of oilseed rape fertility to analyze the inter-variety variability in number of seeds}},
  booktitle = {Agritrop},
  year = {2010},
}
Figure