Measuring spatial variation in natural selection using randomly-sown seeds of Arabidopsis thaliana

We describe a simple method of measuring spatial variation in fitness using randomly sown seeds. Given a random initial distribution of seeds, any spatial variation in the final distribution of genotypes must be the result of spatial variation in natural selection. Departures from spatial randomness are tested using a modification of join-count statistics, based on the probability that two randomly chosen plants separated by a given distance carry the same genetic marker. Monte Carlo simulations showed that for the randomly sown seeds design the join-count test had higher power to detect spatial variation in fitness than analysis of variance. The power was relatively independent of the number of genotypes used or the number of microsites sampled. The method was tested in the field using three standard lines of Arabidopsis thaliana. At the end of the life cycle, adjacent plants were significantly more likely to be the same line than randomly chosen pairs and the spatial dependence extended to distances of nearly 50 cm. Simple experiments like this may be especially useful for ''comparative ecological genetics'', allowing systematic studies of natural selection among many species or habitats.