An illustrated gardener's guide to transgenic Arabidopsis field experiments

Authors


Author for correspondence:
Stefan Jansson
Tel: +46-90 786 53 54
Fax: +46-90 786 66 76
Email: stefan.jansson@plantphys.umu.se

Summary

  • • Field studies with transgenic Arabidopsis lines have been performed over 8 yr, to better understand the influence that certain genes have on plant performance. Many (if not most) plant phenotypes cannot be observed under the near constant, low-stress conditions in growth chambers, making field experiments necessary. However, there are challenges in performing such experiments: permission must be obtained and regulations obeyed, the profound influence of uncontrollable biotic and abiotic factors has to be considered, and experimental design has to be strictly controlled.
  • • The aim here is to provide inspiration and guidelines for researchers who are not used to setting up such experiments, allowing others to learn from our mistakes.
  • •  This is believed to be the first example of a ‘manual’ for field experiments with transgenic Arabidopsis plants. Many of the challenges encountered are common for all field experiments, and many researchers from ecological backgrounds are skilled in such methods.
  • • There is huge potential in combining the detailed mechanistic understanding of molecular biologists with ecologists’ expertise in examining plant performance under field conditions, and it is suggested that more interdisciplinary collaborations will open up new scientific avenues to aid analyses of the roles of genetic and physiological variation in natural systems.

Introduction

Tremendous progress has been made in recent decades in plant sciences using Arabidopsis as a model species. Arabidopsis has many advantages as a model system, and the sequencing of the Arabidopsis genome has led to an explosion in the development of genetic and genomic resources (Hilson et al., 2003). This sequencing has facilitated the application of diverse reverse genetics techniques, notably enabling gene expression to be readily knocked out by RNA interference (RNAi), and large numbers of constitutive or conditional over-expression systems have been well established. Today Arabidopsis lines lacking most gene products can be easily identified and ordered from stock centres (Alonso et al., 2003). This means that the main challenge for Arabidopsis researchers today is seldom to get the ‘right genotype’, but rather to be able to make correct and informative phenotypic analyses of their plants.

The expressed phenotype of an organism is influenced by the environment, so plant-to-plant variation has often been minimized in experiments in order to explore the effects of genotypic variations by using climate chambers, growth rooms, or phytotrons to provide tightly controlled and highly reproducible growth conditions. However, this has a price: the growth conditions used in virtually all Arabidopsis experiments are much more constant than those that plants experience in the field, under which selection has been taking place for millions of years. From an evolutionary perspective, the genotypes with the highest Darwinian fitness are those that have survived selection in natural environments that typically show strong variations across all timescales, such as sunflecks under a canopy (Leakey et al., 2005) that change incident light intensities by orders of magnitudes in seconds, strong diurnal fluctuations in temperature, light intensity and humidity, and seasonal changes that make most terrestrial habitats cycle between conditions permissive and restrictive to plant growth. Further, there are enormous environmental variations that are more irregular and often highly stochastic: for example, attacks by pathogens and herbivores and more catastrophic events, such as extreme flooding, earthquakes and fire. The ability to perform well under all these conditions must be a major component of the fitness of a plant. Indeed, phenotypic plasticity and the ability to grow, develop and reproduce under nonoptimal conditions are likely to be much more important than the ability to do so under the kinds of conditions in which Arabidopsis experiments are typically performed. Consequently, these experiments are not likely to identify traits, genes and proteins that contribute to phenotypic plasticity, such as abilities to adapt to rapid changes, or growth under suboptimal conditions.

We believe that a way forward for the plant sciences in general, and Arabidopsis research in particular, is to better develop experimental procedures for studying the responses of genetically well-defined genotypes under field conditions (see also Ballare, 2001; Pigliucci, 2003; Weinig et al., 2003). This may sound trivial, but plant physiologists and molecular biologists have been slow in realizing the importance of doing this and performing relevant experiments. Symptomatically, we were among the first to publish a very simple experiment in which seed set was compared in Arabidopsis mutants and wild-type plants grown in the field, thus making it possible to quantify the fitness contribution of a single gene (Külheim et al., 2002). This experiment also demonstrated that mutations that do not have clear phenotypic consequences under laboratory conditions can significantly influence performance in the field, and that the ability to cope with environmental variation is important to plants. In order to perform field experiments, the typical molecular biologist has to go outside his or her ‘comfort zone’ and face new problems. In addition, the fact that many of these experiments are performed using transgenic plants may make them less attractive to most Arabidopsis researchers, owing to restrictive legislation and public hostility towards transgenic plants.

In recent years, we have performed field experiments with Arabidopsis mutants and transgenics, not only to study seed production, which in Arabidopsis is likely to be a reasonably accurate proxy of Darwinian fitness (Külheim et al., 2002; Andersson et al., 2003; Ganeteg et al., 2004; Frenkel et al., 2007), but also to look at physiological processes (Külheim & Jansson, 2005) in conjunction with microarray and metabolomic analyses (Frenkel, 2008). We would by no means argue that all Arabidopsis experiments should be performed under field conditions. However, we believe that field studies can be highly informative. Indeed, many important questions cannot be readily addressed without examining the performance of organisms in natural or close to natural environments. Our aims here are to allow others to learn from our ample mistakes and provide some inspiration and guidelines for researchers who wish to address such questions but are not used to setting up field experiments.

Setting up a field experiment

Obtaining permission to do the experiment

Most, if not all, countries where research with transgenic plants is performed have a regulatory legislative framework for such activities. These regulations differ between countries; even within Europe, which is covered by the European Union (EU) Directive 2001/18/EC (European Union, 2001), there is no consensus between countries. Differences in legislation are beyond the scope of this article, but since we were the first to obtain a permit to grow transgenic Arabidopsis in the field in Europe, our experiences may be used as a precedent for experiments in other countries. In Sweden, permits to grow transgenic plants in the field (‘deliberate release’ in European legislative terms) are given by the Board of Agriculture (Jordbruksverket). The permit application procedure requires a SNIF (summary notification information format for the release of genetically modified higher plants) to be completed (a copy of a SNIF for one of our applications can be seen in Supporting Information, Text S1). More information is given in the appendix to the Directive 2001/18/EC. The current fee in Sweden is 38 000 SEK (c.ı4200, US$6000 at April, 2008) per application. All applications have to be scrutinized by national authorities and sent to corresponding authorities in the other EU countries that are allowed to comment. The authorities should give an answer within 120 d, but handling could take longer if additional information is required. One application could cover a large number of transgenic lines, which typically have to be described one by one. However, since it is the introduced DNA fragment that is of major concern to the authorities, a collection of insertional lines would be described similarly. For instance, we have a very useful permit that covers the use of T-DNA or transposon knockout lines of all genes coding for photosynthetic proteins; this permit is also valid for genes that are not involved in the primary photosynthetic reaction. Once a permit is given we are obliged to provide: (i) a map of the field site (in our case a plot covering c. 10 m2 in Umeå University's experimental garden); (ii) a copy of the written instructions provided to the field personnel (for an example see Text S2); (iii) a copy of the notification provided to the local authorities (a simple letter that we will perform a field experiment according to a certain permit); and (iv) a copy of an advertisement in a local newspaper notifying the public of the experiment. In addition to the cost of the application, the cost of an inspection to ensure that we follow our written instructions also has to be met.

One potential concern is that planned experiments could be disrupted by activities of organizations or individuals opposed to experiments with transgenic plants. Although the public has been notified about our experiments by both newspaper advertisements and, on two occasions, quite long newspaper articles (written at the initiative of local journalists with an interest in the research at the university), we have never experienced negative public reactions to our experiments, in either words or actions. We believe that the acceptance of transgenic field experiments depends on the trait modified and the plant species used; local and national opinion among both the public and politicians; and the public's attitudes regarding the credibility of the organization performing the experiments. The species may be the least important factor, but even in this respect Arabidopsis is a good choice. Albeit a weed, it does not show invasive properties and is not a species that raises much concern outside the scientific community, in contrast to tobacco, rice or soybean, for instance. The credibility of the organization is, in our opinion, likely to be a major factor. A public body with government funding, such as a university, probably has more credibility than a research centre that is partially funded by private industry. A national company will have even less credibility, and multinational companies have the least credibility in this respect.

Organizing the field site and planting

In order to reduce the risk of unintentional release of transgenic material, and to maximize the value of the results obtained from our experiments, we have organized the field site in a particular way. Seeds are sown in pots (usually with sizes of c. 6 × 6 cm, each containing a few seeds with colour-coded markers in the pot for the genotypes; pot supplier VEFI A/S, Larvik, Norway), vernalized at +4°C for 2–4 d in the cold room, and then germinated in the laboratory. When seedlings appear, all but one in each pot is removed and the pots are carefully taken to the field site to reduce the risk of unintentionally releasing transgenic seeds. As we have been mainly interested in the expression of phenotypes and effects on seed production, we have performed most of our experiments with individual plants in different pots. However, if resource competition is the main focus of the experiment, individuals could be planted together at this point. At the site, 10–20 cm of the topsoil is removed and the remaining soil is then covered by a plastic water-permeable mat normally used for weed control. About 5 cm of soil, for which we use a standard garden variety (‘Yrkesplantjord’, Weibull trädgård AB, Hammenhög, Sweden), is replaced on top of the mat and pots are arranged in trays, each with a capacity of 30 pots, that also contain soil. These are then placed in the field site (Fig. 1). The whole site is surrounded by a low fence in order to prevent it being unintentionally disturbed. Covering the site with an insect net is a requirement by the authorities in order to prevent unintentional insect-borne pollination of plants. Hence, when plants subsequently bolt, the whole site is covered by an insect net supported c. 50 cm above the ground, which is sufficient to allow the plants to develop undisturbed (Fig. 1, inset). Arabidopsis is an almost obligate self-pollinator, but the rationale behind these preventative measures is that chewing insects could potentially mediate cross-pollination if they mimicked laboratory procedures for making Arabidopsis crosses (i.e. removing petals, carpels and stamens from closed flower buds that have not yet been pollinated while leaving the pistil intact). Any researcher who tries to make a cross between two Arabidopsis genotypes will find that the probability of producing unintended crosses is vanishingly small, but this easy measure also provides better control over insect herbivory (see later). One effect of the insect net is that it shades the plants (the net we use reduces the light intensity by almost 50%), but peak light intensities may nevertheless approach 1000 µmol, much higher than those normally used in growth chambers, and the rapid and unpredictable light variation that characterizes natural growing conditions is not affected. The permit also stipulates that the area around the experiment (up to 10 m) should be regularly inspected and all Arabidopsis and Cardaminopsis plants found should be destroyed (Cardaminopsis-Arabidopsis hybrids have been described; Chen et al., 1998).

Figure 1.

Schematic illustration of the field site. Inset: photo showing pots and trays covered by insect netting to prevent accidental pollination.

After each experiment, all transgenic material must be removed. We treat all soil on the mat as potentially containing transgenic material, so it is packed on the site in plastic bags that are sealed and transported to the laboratory, where it is treated together with other transgenic waste according to our local regulations.

Seedling age

Many Arabidopsis genotypes from temperate areas are winter annuals and germinate late in the growing season, overwinter as rosettes, and flower early in the next season. This life cycle is not easily mimicked as the chances for an approval of an application to overwinter transgenic plants will inevitably be reduced, for several reasons, notably because it is more difficult to keep the site under control for longer periods, and it is difficult to make sure that no transgenic material is transported to the surroundings if the site is covered by snow for part of the year. For these reasons, we start our experiments in the beginning of the growing season and do not plant out seedlings until the risk of frost is over (in the harsh Umeå climate, this corresponds to early or mid-June). We typically move seedlings as young as possible into the field, to let them develop from an early age under field conditions, but sometimes we grow them indoors and move the pots to the site when the plants are bigger. This inevitably influences the outcome of the experiments. In Fig. 2, growth characteristics and seed set of plants (wild-type Columbia C-0) are shown. The plants were planted out either just after germination (7 d after sowing, ‘early’ plants; n = 29) or 2 wk after growth in a climate chamber under long day conditions (21 d after sowing, ‘late’ plants; n = 15). In the field, plants were grown next to each other in similar conditions. As can be seen in Fig. 2, initially the growth rates of these two sets of plants were very similar; the sizes of ‘late’ plants on the day they were moved to the field site were almost identical to ‘early’ plants grown in the field from germination, but this did not hold true later in the experiment. This is probably because long day conditions in the field triggered flowering immediately in ‘early’ plants at the time they were developmentally ready to perceive it. Meanwhile ‘late’ plants were still in the climate chamber not perceiving the signal, and the onset of flowering (after the necessary lag phase) was therefore earlier in the former plants. Thus, vegetative growth was prolonged in ‘late’ plants so they were bigger before they set seed and produced higher seed yields (Fig. 2, inset). These differences clearly illustrate the necessity of treating all genotypes within an experiment in an identical way. The immediate initiation of flowering in the field, once plants have reached a developmental stage at which they can produce flowers, is not specific for growth at high-latitude locations, such as Umeå. Most laboratory strains of Arabidopsis have a critical photoperiod of less than 12 h and thus flowering is induced by summer conditions anywhere in the world.

Figure 2.

Mean growth and seed set (inset) of Arabidopsis plants moved to the field 7 d after planting (‘early’, left-hand arrow, closed circles) and 21 d after planting (‘late’, right-hand arrow, open circles). The variation in seed set is shown by the standard deviation.

The effect of soil quality

Plants grown under climate chamber or glasshouse conditions are regularly watered and typically fertilized in order to keep the water and nutrient status of their growth media nearly constant. In the field this may be less easy, and doing so may not be desired if natural conditions are to be simulated as closely as possible. Although field-grown plants are typically smaller when flowering is initiated than corresponding laboratory-grown plants, their root systems are generally very well developed and their roots often extrude from their pots (Fig. 3) from early stages in the experiments. Not surprisingly, under such conditions, soil quality will have a major impact on plant performance. Figure 4(a) shows phenotypic differences between plants grown in soils with different nutrient regimes (wild-type Columbia). Their rosette diameters varied considerably (Fig. 4b), and seed set varied by several orders of magnitude (the average numbers of seeds produced by plants grown in the most nutrient-rich and nutrient-poor soils were > 10 000 and 4, respectively; see Fig. 4(c), note the logarithmic scale). Again, it is obvious that all genotypes in an experiment must be treated in a similar fashion so they have access to the same amounts of resources and are subject to the same stresses.

Figure 3.

An Arabidopsis individual (51 d after sowing and 21 d in the field) showing strong root growth protruding from the pot.

Figure 4.

Arabidopsis plant performance on different soils, with a fertility gradient created by using mixtures with varying proportions of rich commercial (Yrkesplantjord, Weibull AB) and nutrient-poor soil (Kaktusjord, Weibull AB). Based on the content information on the soil packaging, the nitrogen (N) contents calculated in four sets of pots with ‘high’, ‘intermediate-high’, ‘intermediate-low’ and ‘low’ fertile soil were, respectively, 22.5, 14, 7, and 3.5 mg N per pot. (a) Photograph of flowering plants. (b) Rosette size after 18 d in the field. (c) Seed set at the end of the experiment (note logarithmic scale).

Seed counting

In our studies we have mainly focused on genes involved in photosynthesis and metabolism. Minor alterations in these processes could perhaps be compensated for during the vegetative phase of growth (in which photosynthesis may even be limited by low sink strength; for review see Paul & Foyer, 2001). However, even minute reductions in parameters such as photosynthetic rates may reduce seed set, since the plant's photosynthetic capacity is more likely to be limiting during the reproductive phase because of the strong sinks created by the developing reproductive structures. Thus, seed set data can provide sensitive indications of differences in Darwinian fitness associated with genotypic variations affecting these parameters, and we have frequently observed that genotypes that seem to have the same growth rates during the vegetative phase could differ in seed set.

There are several potential ways of obtaining seed set data, including efficient commercial systems for collecting Arabidopsis seed in which whole plants with inflorescences are enclosed, and their seeds fall down into an acrylic funnel. However, placing a field-grown plant in such an environment is not compatible with growing plants under conditions that are as natural as possible (for instance, the conditions for pathogens and herbivores inside a tube of cellophane would be highly artificial). We have tried estimating ‘true’ lifetime seed set by continuously harvesting ‘ripe’ siliques (those that have turned yellow), but that is both time-consuming and labour-intensive since plants have to be inspected and harvested every 2–3 d and siliques from each plant have to be kept apart. In addition, this procedure cannot really be followed if the risk of distributing transgenic material is to be minimized, since it involves handling transgenic seeds under field conditions (e.g. in rain and wind). Instead, we have adopted a procedure in which we terminate the experiment at a given date and then estimate seed production at this point. We typically terminate the experiment and begin to count seeds when the first plants in a tray show siliques that are just about to spread their seeds (at which time the rest of the individuals in the tray have also started to produce seeds). When this happens, the whole tray is transferred to a box, to avoid spreading seeds during transport, and brought into the laboratory for seed counting. At this point, many of the siliques, if not most, are still not ripe, and opening green siliques to count the seeds inside would be very time-consuming. Instead, we approximate total seed set by first harvesting a number (usually three to five) of randomly chosen dessicated (yellow) siliques per plant, then opening them on a piece of paper and manually counting the number of seeds inside. The total number of siliques on each plant is then manually counted and total seed set is estimated by multiplying the average number of seeds counted in the randomly chosen yellow siliques by the total number of siliques. There are drawbacks to this procedure since if, for example, elimination of a particular gene's expression reduces leaf longevity, its ‘true’ consequences for Darwinian fitness are likely to be underestimated, but we have found no good alternatives to date.

There are several potential pitfalls in approximating Darwinian fitness from seed set. For instance, variations, such as seeds from one genotype being larger or having a higher germination frequency, will not be captured. However, when we have measured seed weights or germination rates (which is not a trivial task, since such measurements are only valid if all of the seeds compared are mature and have been identically treated) we have not observed any significant variation between genotypes or tendencies for genotypes that produce fewer seeds to have higher germination frequencies. If anything, we believe that seed counting in the way we do tends to under-estimate, rather than over-estimate, the ‘true’ differences in Darwinian fitness between genotypes.

Variability

Plant-to-plant variation

One of the most obvious differences between experiments performed in the field and in climate chambers or glasshouses is that plant-to-plant variations are higher in the field, as a result of both differences in microclimate, which may even vary between pots, and stochastic events, such as infections by pathogens and grazing by herbivores. To illustrate this, in Fig. 5(a) we show seed set data obtained for three batches of c. 30 Arabidopsis plants of wild-type Columbia, and c. 30 of wild-type Nössen in the summer of 2007. The data show that the number of seeds produced by plants of the same genotype growing under the same conditions may vary 10-fold. There was a strong correlation between rosette diameter and seed number (see also, for example, Heidel et al., 2004), with larger plants unsurprisingly producing more seeds (R2 = 0.70, Fig. 5b), but other factors also influenced seed set.

Figure 5.

Variation in seed sizes of two different Arabidopsis wild-type lines (Columbia and Nössen) within an experiment. The Columbia seeds came from three different laboratories. (a) Variation in seed set for each seed source and wild-type. (b) Variation in seed set in relation to diameter size before flowering. (c) As (a), but showing the mean and variation (SE) for each seed source and wild-type.

If the variation is so high, is there a justification for estimating the seed set of a given genotype? We believe so, as illustrated by the average numbers of seeds, with standard errors, shown in Fig. 5(c) of the three batches of c. 30 wild-type Columbia plants for which total numbers of seeds produced are presented in Fig. 5(a). The data presented in these graphs are the results from three parallel experiments. In the summer of 2007, we performed studies in collaboration with workers in other laboratories who provided us with seeds of mutants and T-DNA knockout plants for us to assess their fitness. These collaborators also provided wild-type controls: in three cases Columbia-0, and in one case Nössen. We treated the different batches of wild-type plants as independent genotypes, but when the average seed sets of the Columbia genotype in the separate experiments were compared, the results showed that the values were almost identical. By contrast, Nössen produced on average 74% more seeds than Columbia. This illustrates, in our opinion, that a mutant/transgenic plant can only be compared with plants with a corresponding genetic background; and Arabidopsis Columbia grown in these three laboratories had not been strongly influenced by independent propagation of seed sources.

Normally we find a standard deviation of c. one-third to one-half of the total seed set. In the experiments of 2005, for example, the mean seed set was c. 35 000 seeds, with a standard deviation of c. 10 000 seeds. A sample size of 30 plants (which is our normal number of replicates) would then enable us to significantly detect a mean difference of c. 8000 seeds with a power of 0.80. Furthermore, the variation in seed set between plants grown in the field is comparable to that of plants grown under controlled conditions. For instance, mean values of coefficients of variation (CVs) for seed set by plants grown in the field and the laboratory in a number of our experiments were 0.46 and 0.36. Although the variation in the field seems to be larger, there are no significant differences between any of the CVs we have measured (according to variance ratio tests).

The impact of weather and catastrophic events

Obviously, weather conditions will strongly influence the outcome of field experiments. Weather in temperate areas varies substantially between days, making it necessary to monitor weather forecasts before deciding when to sow, when to transfer seedlings to the field, and when to terminate an experiment. Even so, the Arabidopsis garden will be strongly influenced by the weather. For example, ‘catastrophic events’, such as the very strong rainfall that occurred during 27–28 August 2007 at our field site, could completely destroy an experiment (Fig. 6a) unless adequate measures are taken to avoid all the plants dying (Fig. 6b). Another example of such an event at our field site was an unexpected herbivore attack of extremely high intensity by aphids and diamondback moths (Plutella xylostella), which occurred in mid-August 2006, almost completely destroying the experiment. Weather and other factors that may affect the plants are often associated to varying degrees; for instance, diamondback moths do not overwinter in the area but migrate every year from many hundreds of kilometres away (Talekar & Shelton, 1993), so their abundance depends, inter alia, on the prevailing wind conditions during certain periods. However, regardless of the interactions between these factors, a practical consequence of their potentially catastrophic effects is that plants must be monitored almost daily, so locating field sites at a position close to the usual working environment is strongly recommended.

Figure 6.

Photographs showing the effect of an extreme rainfall event. (a) Flooding at the field site. (b) The trays were placed on boxes to lift them out of the water.

Year-to-year variation

Natural environmental variations such as those mentioned in the previous section will lead to large variations in seed set of the same genotype between seasons. To illustrate the variation encountered between years, the average seed set of Arabidopsis wild-type Columbia in our field experiments in each of the years from 2000 to 2007 is shown in Fig. 7. Although some of the variation could be the result of factors such as inconsistencies in sowing time and herbivore treatments, the vast majority is caused by variation in the abiotic environment. In 2003 we performed experiments at two sites, c. 50 m apart (one in partial shade and the other in full sunlight) that we have used during other years (Ganeteg et al., 2004). As can be seen, the plants in the shade produced c. 90% fewer seeds than those placed in full sunlight (Fig. 7). In a ‘good’ year, Arabidopsis Columbia plants grown during the summertime (and thus in long day conditions) at Umeå can nevertheless produce on average over 35 000 seeds each, whereas in other years they may produce less than 2000. Clearly, therefore, it is difficult to compare seed set between genotypes unless they are grown at the same time and place.

Figure 7.

Variation in mean seed set of Arabidopsis wild-type plants over different years and different experiments. Note that the plants were divided into sun-exposed and shaded plants (under nets) in the 2003 experiment. All other experiments were performed in full sunlight.

Herbivores and pathogens

In the first year that we performed these experiments, our aim was to mimic natural conditions as closely as possible. We therefore only watered the plants sufficiently to keep them alive and we did not try to prevent grazing by herbivores. In one of the experiments that year, this approach was informative (Frenkel, 2008), although in other years we would not have been able to obtain any seed set data because of very strong herbivory (see Fig. 8a for an example of a grazed plant). Therefore, in most years it has been deemed necessary to use insecticides (e.g. Pyrex N, Wikholm & Co Eftr AB, Borås, Sweden). Again, almost daily inspections are needed in order to observe and effectively treat herbivores. For instance, aphids can rapidly infest particular inflorescences (Fig. 8b) and diamondback moth larvae (Fig. 8c) can have major effects on the leaf mass of a plant very rapidly. Slugs, such as the netted slug (Deroceras reticulata; Fig. 8d), are also sometimes common at our site, especially after rain events. Of course, the herbivore fauna at other field sites may be different.

Figure 8.

Photographs of herbivores and herbivore damage at the site. (a) A grazed plant; (b) aphids on an inflorescence; (c) larva on an Arabidopsis leaf disc and adult (inset) diamondback moth (Plutella xylostella); (d) Slug (Deroceras reticulatum).

Pathogens (e.g. fungi and bacteria) may also affect the fitness of Arabidopsis plants (e.g. Goss & Bergelson, 2007) and we have sometimes seen signs of infection on our plants, although not to the extent that it seemed to influence seed set. However, we have not rigorously examined its effects, and thus we cannot exclude the possibility that some of the observed variation in seed set is the result of pathogen infections. Therefore, we would like to draw the gardener's attention to the possibility of infections affecting the fitness of experimental plants, which means that pathogens will have to be treated in a similar way to herbivores.

Experimental design

As we have shown, many factors can influence, and complicate the interpretation of, the results of field experiments involving different Arabidopsis genotypes. In this respect, we see huge potential in combining the detailed mechanistic understanding of molecular biologists with the expertise of ecologists in examining plant performance under natural conditions, and we suggest that more interdisciplinary collaborations will open up new scientific avenues. It is essential to make sure that no systematic bias is introduced by using (and carefully implementing) an appropriate experimental design. Some key issues in experimental design that we have found to be important are the following. Firstly, all genotypes must get an identical start, and thus seed batches that germinate at significantly different rates under the vernalization/germination conditions employed cannot be used. Secondly, the genotypes should, as far as possible, be anonymous to the people involved, by using (as a minimum measure) colour coding. An alternative, which we have not employed, would be to label each pot with a unique number that is not decoded until the experiment has been completed. Thirdly, a randomized block design, with the different genotypes randomized within each block, generally increases the analytical power by identifying heterogeneities in the conditions across the site, and when a block design is not (or cannot be) used, different genotypes must be randomized within the experiment. That said, experimental design is a large research field in itself, and we advocate consulting a statistician or someone trained in experimental design for guidance.

Concluding remarks

We believe that experiments in which plants are grown under field conditions are essential for elucidating the full complexity of plant metabolism, development, gene function and stress tolerance, which are all strongly influenced by the environment. However, field experiments present challenges that need to be tackled. Data showing strong variations between years (Fig. 7), or between replicates of the same genotype (Fig. 5), may make many researchers who are used to growing plants in climate chambers believe that field studies can never be used to draw significant conclusions (at least, some reviewers of our submitted papers seem to believe so). Such misgivings are, to a large extent, grounded in the notion that has served science well for several centuries, that in order to obtain robust data regarding relationships between variables, all other variables should be strictly controlled. However, advances, inter alia, in computer power and experimental design naturally lead to rejection of ‘change-one-variable-at-a-time’ approaches in diverse contexts (e.g. drug design, chemometrics and ‘-omic’ analyses). Instead, approaches in which multiple variables are changed simultaneously to explore both their single and interactive effects should be favoured.

Over the years, ecologists have acquired much better skills than molecular biologists in analysing field data with complex patterns of variation, so collaboration among physiologists, molecular biologists and ecologists can be very fruitful for analysing large datasets reflecting complex genotype–environment interactions. We have often found that people who have been trained in the laboratory would like us to present the ‘true fitness differences’ between two genotypes under our field conditions. Unfortunately, however, there is no such thing as ‘true fitness’. People trained in field studies know that the fitness of an organism is a product of the interaction between its genotype and the environment. Thus, they are less alarmed by (or may even expect) large between-experiment variations in a genotype's seed set or the substantial variations we sometimes observe in the relative fitness of two genotypes (Külheim et al., 2002; Ganeteg et al., 2004). In fact, in some cases one mutant has been significantly ‘less fit’ than the corresponding wild type one year, but significantly ‘fitter’ in another year. Instead of merely regarding this as a complication, we believe that such observations are a first step towards a better understanding of the interactions between plant genotypes and their environment. Consequently, we strongly believe that data obtained in experiments performed under natural, uncontrolled conditions will provide a better understanding of plants as highly dynamic organisms, their evolution and the processes involved.

Acknowledgements

We are indebted to those who, over the years, have taken part in discussing and developing the practices and routines discussed here. A few who deserve special mention are Carsten Külheim, Jon Ågren, Daniel Eriksson, Mike Bastian, Ulrika Ganeteg and Frank Klimmek. This work was supported by grants from the Swedish Research Council (VR) and the Swedish Research Council for Environment, Agricultural Sciences and Spatial Planning (FORMAS).

Ancillary