The standard method of measuring pollen limitation is to add pollen to a number of flowers, preferably to a whole plant, and to compare fruit and seed set with that of naturally pollinated flowers on other plants. In 25 yr of research, this method has yielded valuable data, but it is difficult to use in large plants. This has caused a bias in the available data towards smaller, herbaceous plants with relatively few flowers. I argue that, in order to widen our knowledge of how pollen limitation affects plants, we should go beyond whole-plant pollen addition and change our concept of how a flowering plant functions. The traditional method does not take into account the variation in and dynamics of resource allocation and pollen availability. The concept of integrated physiological units (IPUs) does, but, although it has been applied to pollination biology, it has not received the attention it deserves. I use this article to present its merits again, to propose a step-by-step methodology for studying pollen limitation, and to examine factors influencing possible plant strategies.
The year 2006 marks the 25th anniversary of Andrew Stephenson's seminal review paper on the causes and function of flower and fruit abortion in the Annual Review of Ecology and Systematics (Stephenson, 1981), which set the stage for what has proved to be quite a fruitful study subject: pollen limitation. Pollen limitation, the degree to which pollinator services are sufficient to ensure maximum fruit or seed set in a plant, has been studied from several perspectives. While the first papers published dealt mainly with the fundamental question of whether reproductive success is limited by pollen or by resources, more recent studies have used the same tools in a more applied setting to look at the functioning of species and ecosystems in the face of a ‘pollination crisis’(Kearns et al., 1998).
Recently, the data collected in the past 25 yr have been considered in three reviews discussing the role of pollen limitation in plant populations (Ashman et al., 2004; Knight et al., 2005, 2006). However, although an impressive body of results has been obtained, the methods used to study pollen limitation have hardly changed. The basic technique is to add extra pollen to flowers on some plants and compare fruit and/or seed set with that of flowers on control plants subjected to natural pollination levels. In this paper I discuss the limitations of this approach, and argue that we should diversify our methods in order to increase the number of plant species and growth forms studied, and to improve our understanding of the dynamic interaction between pollen deposition and resource allocation.
A short history of the study of pollen limitation
The first works comparing reproductive output after hand pollination with that after natural pollination were published at the end of the 1970s, as a reaction to publications stating differences in resource availability as the main explanation for differences among populations in fruit or seed set. The first author to use the term ‘pollinator limitation’ in the title of an article was Paulette Bierzychudek, in a short communication in The American Naturalist entitled ‘Pollinator limitation of plant reproductive effort’ (Bierzychudek, 1981). This paper was followed by the review article by Andrew Stephenson (1981), who focused on plants that regularly abscise a large portion of their flowers and immature fruits, and tried to determine which factor was limiting fruit and seed production. He concluded that fruit abortion was an important factor in explaining low fruit set in these species, and that pollen limitation played a minor role.
A few years later, Michael Zimmerman and Graham Pyke published an experimental paper on pollen limitation in Polemonium foliosissimum (Zimmerman & Pyke, 1988), in which they warned that pollinating only a proportion of the flowers on a plant could induce reallocation within a plant, leading to spurious results. This problem had already been mentioned by Stephenson (1981, p. 259) and Bawa & Webb (1984), but neither of these publications had had much of an impact on the methods used in studying pollen limitation.
In 1994, Martin Burd published a review (Burd, 1994) in which he compiled data on studies using hand pollination for an impressive total of 258 species. Almost 60% of these species showed significant pollen limitation of female reproduction. In his article, Burd discussed the multiple scales of pollen limitation, ranging from ovules via flowers to whole plants and lifetime fitness. However, this distinction was not made in the meta-analysis, which combined data from articles using hand pollination at different levels, ranging from a few flowers to whole plants. Lumping data sets with different methodologies had the great advantage of allowing the presentation of data for many different species, which increased the value of the general conclusions of the paper. Burd showed that self-incompatible species suffer more from pollen limitation, that the degree of pollen limitation within populations usually varies throughout the season, and that fruit set is more frequently and more strongly affected than seed set. This last result is explained by variation in pollen deposition among flowers and subsequent abortion of the less visited flowers, and was confirmed in the model he published later on ovule packaging strategies as a response to stochastic variation in pollen and resources (Burd, 1995).
A recent review on the quantitative effects of pollen supplementation experiments (Knight et al., 2006) confirmed that pollinating only a fraction of the flowers on a plant can effectively shunt resources away from nonpollinated flowers and cause a local increase in fruit or seed set, while decreasing reproductive output in neighbouring flowers. The effect size, the difference between treated and control flowers, in experiments using only individual flowers or inflorescences proved to be much higher than in experiments where whole plants were hand-pollinated.
Using whole plants: a solution and a problem
A simple solution to the problem of resource reallocation is to restrict the data set on pollen limitation to studies that applied hand pollination to whole plants. This was done by Tia-Lynn Ashman et al. (2004), who discussed the effects of pollen limitation at the whole-plant level and upwards on population, community and ecosystem processes. Their meta-analysis clearly demonstrates the occurrence of pollen limitation without the confounding effects of reallocation of resources, but the use of only whole-plant pollinations also introduces a number of biases. The data set used in this article contained only 22 species, all herbaceous: two annuals, two biennials, and 18 perennials, most of these species having relatively few flowers per plant, with a maximum of c. 50 flowers. Among the 18 perennials, nine species were monocots with underground storage organs such as rhizomes, bulbs or corms, with just a few, large flowers, often flowering in early spring, when pollinators are still scarce. These biases are a logical result of the methodology chosen, as whole-plant studies are much more likely to be performed on plants with a few flowers that all develop at the same time or at very short intervals. It is therefore no surprise that the conclusions regarding the importance of pollen limitation in this study differed profoundly from those of Stephenson (1981), who mainly focused on trees, which produce hundreds if not thousands of flowers in a single season.
The meta-analyses of Burd (1994), Ashman et al. (2004) and Knight et al. (2006) have given us many useful insights into the role of pollen limitation in plant reproduction. However, they had to work with the data available in the literature, which have usually been generated by one and the same method: additional hand pollination on parts of plants or whole plants. It is clear that this method alone is not going to answer our questions about the importance of pollen limitation in different taxonomic groups and growth forms. To make further progress, we need to use methods that allow the study of larger plants while controlling for the effects of dynamic resource allocation. In other words, we should move down the hierarchy, from whole plants to inflorescences and individual flowers, and study both resource and pollen availability at these levels. This requires a different view of how pollen deposition and resource availability interact, and I will explain here how the views behind the different methodologies differ.
A bouquet in a vase?
The problem behind the traditional methodology is that it regards a flowering plant as a collection of flowers that all have equal access to a common pool of nutrients, much like a bunch of flowers in a vase. Even the article by Zimmerman & Pyke (1988) on reallocation among flowers used the same concept: no details were given as to the spatial arrangement of the flowers chosen, and only temporal effects were taken into account. Martin Burd (1995), in his article on ovule packaging strategies, incorporated random pollination and fertilization, but regarded nutrients as a fixed pool available to all flowers. It is clear that the ‘bouquet in a vase’ concept is far from realistic: both spatial and temporal variations in resource availability occur, which mean that different flowers have different amounts of energy available for fruit development. I will present the current state of knowledge on both spatial and temporal variation in resource availability for individual flowers within a plant, and then argue that it is now time to move away from a whole-plant perspective towards a methodology based on individual flowers.
Spatial variation in resource availability
The fact that not all flowers are equal has been convincingly demonstrated in pollination biology, for example in terms of the architectural effects described by Pamela Diggle (1995). Flowers have inherently different capacities of fruit and seed production according to their place in the ontogeny of the inflorescence and, even if flowers are of the same size, flower position within an inflorescence or in the flowering order can greatly influence the chances of acquiring sufficient resources for fruit set (e.g. Ashman & Hitchens, 2000). Furthermore, the filling of fruits and seeds requires mostly photosynthates, and these are not produced in a single pool, but by many different green organs, often including the flowers and fruits themselves (Hansen, 1969; Watson & Casper, 1984). Developing fruits often utilize assimilates from adjacent leaves, and their proximity to these sources may influence their probability of developing a fruit.
Temporal dynamics of resource allocation
A very insightful model of how a plant can adjust its investment in (female) reproduction over time was published by David Lloyd (1980), before the pollination vs resource limitation debate had even started. He considered three stages: flower production, ovary development (to determine if a flower will be hermaphrodite or male) and fruit development. Adjustment in the last two stages can only be achieved by a decrease from the limit set by the number of flowers. He listed 10 factors that could influence the way a plant adjusts its investment in reproduction in these three stages, including the effects of a simultaneous or staggered development of flowers, the predictability of resources and services such as pollination, and positional effects. He was already stressing the importance of obtaining more precise data on the spatial and temporal patterns of resource allocation to plant reproduction.
Not all flowers are equal
That ignoring inherent differences between flowers at different positions can lead to spurious results in the study of pollen limitation is illustrated in the work by Javier Guitián and colleagues on Petrocoptis grandiflora (Guitián & Sánchez, 1992; Guitián et al., 1994; Guitián & Navarro, 1996; Navarro Etxeberria, 1996). Petrocoptis grandiflora has inflorescences with a terminal flower that opens first, followed by the flowers on the two first-order lateral branches, and flowers on second-order laterals branches will open last (Fig. 1). Initially, the terminal flower on each inflorescence was used for additional hand pollination, while the remaining flowers on the same inflorescence and flowers on different plants were used as controls. All hand-pollinated flowers set fruit (35 out of 35), while only 162 flowers out of 245 without additional hand pollination on the same plant did so, which is a highly significant difference (Guitián et al., 1994; J. Guitián, pers. comm.). However, it turned out that the central flower always has a higher probability of setting fruit under natural conditions: 98.5% fruit set for the terminal flower vs 71% for flowers on first-order lateral branches and 43% for those on second-order laterals (Guitián & Navarro, 1996). Comparison of hand-pollinated terminal flowers with the appropriate control, i.e. nonpollinated terminal flowers, shows that hand pollination increased fruit set in terminal flowers from 98.5% (68 out of 69 flowers) to 100%, which is not a significant difference. Hand pollinations were not performed on lateral flowers, but removal of the terminal flower increased fruit set in flowers on first-order laterals to 94% (Guitián & Navarro, 1996), thus showing indirectly that these flowers were not pollen limited.
It is clear that lack of knowledge of how resource allocation influences plant reproduction may lead to erroneous results in pollen limitation studies. In order to understand patterns of seed and fruit production, we need to adopt a different concept of what constitutes a flowering plant.
An assemblage of semiautonomous physiological units
We do not need to go far to find a more appropriate concept for understanding the roles of resources and pollen in shaping patterns of fruit and seed set; the framework already exists. In an article published in 1993, Brenda Casper & Richard Niesenbaum introduced the concept of integrated physiological units (IPUs) to pollination biology (Casper & Niesenbaum, 1993). Instead of regarding a plant as a collection of flowers with equal access to nutrients and all connected to one another, this concept essentially regroups flowers into units within which reallocation is possible, while exchange of nutrients between units does not occur. They wrote their article in response to that of Haig & Westoby (1988), who argued that selection acts to favour increases or decreases in display size towards an equilibrium where life-time seed production is limited by pollen and resources simultaneously. Casper and Niesenbaum made the assumption that a plant can experience both pollen and resource limitation within a single season, but at the single flower level. Pollen deposition is variable from one flower to another, and resources cannot be redistributed among all flowers. This will cause pollen limitation of seed production in some flowers, while in other flowers it is limited by resource availability.
The first article describing IPUs, by Maxime Watson and Brenda Casper, was published in the Annual Review of Ecology and Systematics (Watson & Casper, 1984). They defined an IPU as an identifiable array of morphological subunits that together function as relatively autonomous structures with respect to the assimilation, distribution, and utilization of carbon. Young seedlings and many monocots may consist of a single IPU, while larger plants become progressively more subdivided. A developing branch first uses assimilates from the main stem but, once it produces its own leaves, it becomes photosynthetically independent. IPUs generally reflect the vascular architecture of the plant, in which carbon export is limited to plant parts connected to the same vascular strands (Watson & Casper, 1984).
The Casper & Niesenbaum (1993) paper describing the application of IPUs to pollination biology was published in a relatively obscure journal (Current Science India), which is probably the reason why it has been much less cited than the Zimmerman & Pyke (1988) paper, which appeared in The American Naturalist: only 25 times vs 113 times for Zimmerman & Pyke in the same period, 1994–2006 (ISI Web of Science, accessed on 12 October 2006, http://portalisiknowledge.com/portal.cgi). The first article on IPUs (Watson & Casper, 1984) was published in the Annual Review of Ecology and Systematics, and it has been cited 378 times, mainly in articles dealing with clonal integration and effects of herbivory or defoliation, and much less often in pollination studies.
To clarify the differences between the two concepts, I have drawn schematic representations of each. The principle is that each flower can potentially set a given number of seeds, depending on the number of ovules. The ‘bouquet in a vase’ view of a flowering plant (Fig. 2) assumes that all flowers that flower simultaneously have equal access to resources. Before flowering, the plant has spent some resources on flower production, and the remainder will serve to fill the developing seeds. If female reproduction is pollen limited (Fig. 2, centre), the ‘ceiling’ imposed by low levels of natural pollination is lifted by applying supplemental pollen, and flowers will set more seed. If hand pollination does not increase the number of seeds produced, the upper limit to seed set is resource availability (Fig. 2, bottom).
The dynamic view, based on individual flowers, is best illustrated by a flow chart (Fig. 3). It starts by looking at the individual flower, which will have its ovules fertilized to a greater or lesser extent depending on pollen quantity and quality, and which will set fruit depending on the amount of resources available. Resource availability depends on other flowers in the IPU, the sink strength of the developing fruit and local production of photosynthates.
The main differences between the two concepts are (1) that not all flowers are equal, there being variation in pollen deposition and resource availability depending on environmental conditions, flower position and timing, and (2) that resources, especially carbon-based assimilates, do not freely move around among all flowers, but are restricted in space to IPUs. In order to make progress in understanding the role of pollen limitation, we need more studies that take within-plant patterns of resource availability into account. I have used the existing literature to compile techniques that have proved their worth in this field into a stepwise methodology that can be used as a guideline for future studies of pollen limitation.
A comprehensive methodology
In order to stimulate the study of pollen limitation using a dynamic concept of plant reproductive allocation, I here propose a framework for studies of pollination biology in general and pollen limitation in particular. The relevant question to ask is whether pollen limitation plays a role in determining the patterns of reproductive output observed. Total seed production may or may not be affected, depending on the severity of pollen limitation and the extent to which a plant can react to low pollen deposition by shifting around resources or by producing additional flowers.
However, before outlining a methodology, a further exploration of what IPUs are and how we can use them in pollination ecology is warranted. Firstly, IPUs are in themselves dynamic. As the plant develops and more and more structures are added, existing IPUs may divide into several different units. Leaves near the apex export their nutrients upwards to the apex, while assimilates from older leaves near the base of the stem are transported downwards, to the root sink. As a leaf becomes older and its distance to the apex increases, its assimilates will change destination (Watson & Casper, 1984). Secondly, the boundaries between IPUs are not fixed, and even within IPUs there will be a gradient in which plant parts will be most affected by the local addition of nutrients or the removal of competing sinks. The distance at which effects can be observed will also be dependent on relative resource levels. A strong sink will pull nutrients from farther away than a weak sink, and removal of a strong source will affect more neighbouring structures than the removal of a weak source. Experiments with 14C in several species (cited in Watson & Casper, 1984) have shown that photosynthates are exported from the leaf that assimilated the labelled carbon to nearby leaves in the same vascular strand and to a lesser extent to adjacent strands. This leads to vertical integration in a number of IPUs that is related to the geometric arrangement of leaves on the stem. Finally, growth form influences physiological integration in other ways than via vascular patterns, as illustrated by the fact that the leaves of rosette plants keep receiving nutrients from other, younger leaves, while this is not the case in plants with an elongate growth axis (Watson & Casper, 1984).
The study of pollen limitation within the context of a dynamic resource allocation can be divided into three major phases: description, observation and manipulation. In the descriptive phase, basic information on breeding system and pollinator guild is collected, as well as on the spatial arrangement and temporal development of flowers. The observational phase looks at naturally occurring patterns of fruit and seed set, and how they relate to plant architecture and phenology. In the third phase, experimental manipulations of pollen deposition, resource availability or both will allow the testing of hypotheses generated by the observations in the previous phase.
Phase 1: description
Breeding system and pollinators Information about breeding system and pollinators is the bread and butter of every pollination study. A researcher should at least know the breeding system of the plant species studied (self-compatible or self-incompatible) and whether or not the plant is capable of autonomous pollen deposition in the absence of pollinators, or even apomixis, the formation of seeds without fertilization. If this knowledge is available before the study, it should be mentioned. The standard suite of manipulations applies here: all experimental flowers should be shielded from natural pollination, one set being hand-pollinated with self pollen and a second with outcross pollen, while the two remaining sets are not pollinated, one without and one with emasculation to check for auto-pollination and apomixis, respectively. If seeds are formed in flowers without pollinators and with emasculation, the stigma should be checked for accidental pollen deposition, either during flower manipulation (when the anthers were dehisced during emasculation) or by wind or tiny insects (e.g. thrips) that may still be able to enter the flower even when it is shielded from larger pollinators. It is also useful to check for possible negative effects of emasculation (as a result of damage to the flower) by comparing hand-pollinated emasculated flowers with hand-pollinated intact flowers (e.g. Eckert & Schaefer, 1998). Note that the choice of which flowers to pollinate should also take positional and temporal effects into account (see next section). For self-incompatible plants, the incompatibility system (sporophytic or gametophytic) should be known, because this influences the way in which pollen tubes can be used to quantify actual pollen deposition (see ‘Patterns of pollen deposition’).
Information about pollinators is also indispensable (Thomson, 2001), and nothing replaces actual observations in the population under study. Information should be gathered on the identity and visitation frequency of the animals, and whether or not they are effectively pollinating the flowers (Bloch et al., 2006). Wind-pollinated species are seriously underrepresented in pollen limitation studies, and, unless an effort is made to fill this gap, comparisons between animal- and wind-pollinated species will not be possible.
Spatial and temporal arrangement: plant architecture and phenology Unlike breeding systems, patterns in space and time are not always well described in the literature, except in articles specifically dealing with architectural effects. Firstly, to describe spatial arrangements, it is certainly not enough just to specify inflorescence morphology (e.g. corymbose cyme). The description should include if and how flowers are arranged in inflorescences, if the inflorescences are indeterminate or determinate (capable of continuous flower production), how many flowers there are per inflorescence (mean and range), and how inflorescences are placed in relation to each other (on a branch, terminal or axillary, as separate stalks coming from one basal rosette, as stalks coming from different rosettes, etc.). The presence and size of leaves or bracts and their placement relative to the flowers or inflorescences are also important to note, as they can be important sources of photosynthates. It may be useful to define a nested hierarchy of the different levels (e.g. flower–inflorescence–twig–branch–plant, or flower–twig–branch–plant, or flower–inflorescence–plant, or even simply flower–plant, for single-flowered plants), which can later help to determine the IPUs. Secondly, information on temporal arrangements should be given, comprising information on the flowering order of flowers within inflorescences (e.g. acropetal or basipetal) and of inflorescences in relation to their location within the plant. Another important factor is the relative timing of anthesis and fruit development. Do all flowers reach anthesis before the first fruits start to develop, or is there a large overlap between fruit formation of the first flowers and anthesis of the later flowers? Finally, the number of ovules per flower should be known, not just an average for random flowers, but the range and variation in relation to flower position, both in space and in the flowering order (Diggle, 1995).
Phase 2: observation
Patterns of fruit and seed set After collecting the basic information, it is time to start asking the question of whether pollen limitation plays a role in determining the patterns of fruit and seed set observed in nature. To answer this question, the first step is to record patterns of fruit and/or seed set in space and time under natural conditions. By fruit set, I mean the probability that a fruit will fully develop, and seed set is the proportion of ovules that develop into seeds. Do all flowers within an inflorescence have an equal chance of setting fruit and seed, or are there consistent within-inflorescence patterns? Are there differences among inflorescences in the probability of fruit set or the percentage seed set? Is there a relationship between the spatial position of the flower/inflorescence and its fruit or seed set, and do flowers at certain positions always have a higher fruit/seed set? Is there a relationship with temporal patterns, i.e. do early flowers/inflorescences have a higher reproductive success than later flowers/inflorescences? In general, the first flowers to bloom and start fruit formation have an advantage over later developing flowers (e.g. Ladio & Aizen, 1999). Studies in plants with inflorescences that develop acropetally (e.g. Vallius, 2000) generally show that the basal flowers, which are first to develop, act as strong sinks for resources, which in turn diminishes fruit and seed set in later, more apical flowers within the inflorescence. The patterns in basipetally developing inflorescences, in which the first flowers to develop are farthest away from the peduncle, are more mixed, with position sometimes prevailing over flowering order (Vaughton, 1993) and sometimes not (Brunet, 1996).
Patterns of pollen deposition To be able to answer questions about pollen limitation, we need information about patterns of pollen deposition (Washitani et al., 1994; Huang & Guo, 2002; Waites & Ågren, 2004) and the relation between actual pollen deposition and seed production for individual flowers (Wesselingh & Arnold, 2003). This requires the collecting of stigmas and the counting of conspecific and heterospecific pollen grains, and the counting of pollen tubes growing down the style for self-incompatible species. Again, this should not be treated as an average for a plant, but specified on a per flower basis, in relation to the position of the flower within the inflorescence/plant and in the flowering sequence (Casper & Niesenbaum, 1993). It should become customary to report distributions of pollen deposition or number of pollen tubes, rather than just mean values, as recommended by Burd (1995). Data of this kind can show whether individual flowers are pollen limited or not. For instance, in Iris fulva, I found that the topmost flower on an inflorescence had a high resource availability and the highest probability of setting a fruit (50–70%), while the bottom-most flower had low resource availability and a fruit set of only 5–10% (Wesselingh & Arnold, 2003). Although the mean pollen deposition rates for the two types of flower were not significantly different, the occurrence of a few low deposition rates caused pollen limitation of fruit set in the topmost flowers, while fruit set in the bottom-most flowers was resource limited.
Phase 3: experimentation
The first two phases can only yield descriptive answers; correlations between patterns of fruit and seed set and pollen deposition, flowering order, or plant architecture. The next step is to manipulate either resource availability or pollen deposition, or both, and to observe the effects on fruit and seed set in individual flowers. These experiments should clarify not only the role of pollen limitation but also the extent to which a plant can shift its resources around among flowers. This will tell us what the IPUs of the plants are. The techniques described below can be used at different levels of integration: at the individual flower level to study the effects on neighbouring flowers, usually within the same inflorescence, but also at the inflorescence or branch level, to see to what extent resources can be shifted from one inflorescence to another or from one branch to the next. The differences between these manipulations are largely in scale: in order to detect effects at the branch level, the manipulations will have to be bigger (in terms of biomass removed or numbers of hand-pollinated flowers) than when looking at individual flowers.
General principles It is difficult to give precise instructions as to what manipulations should be performed, because much will depend on the architecture of the plant, which can vary widely from one species to another. I will just list the different possibilities, and suggest that the experimenter uses the observed patterns of fruit and seed set as a guide for determining which manipulations are appropriate.
Plants that undergo experimental manipulations should always be matched with the appropriate control plants (Zimmerman & Pyke, 1988). Plant of different sizes often have widely different resource levels and subsequently different patterns of allocation to flower and fruit formation (Klinkhamer & de Jong, 1987). Plant size should be taken into account when observing natural fruit set and when performing manipulative experiments. This can be done by measuring either plant size or a corollary, such as display size (the number of flowers), and this measure can then be used as a covariate in comparisons (Lawrence, 1993) or to choose control plants and experimental plants that match each other in size/number of flowers, so that pairwise comparisons can be made (Fox, 1992; Wesselingh & Arnold, 2003).
It is a well-known fact that, in general, flowers produced late in the flowering sequence have a lower chance of producing seeds. As weather conditions can greatly affect pollinator activity, flowers that develop at different times will have experienced a different pollination environment. Control and treated plants should be at the same developmental stage and flower at the same time, to prevent confounding effects. Again, a pairwise experimental set-up would be ideal.
Manipulating resource availability Resource availability can be decreased locally by removing leaves (Obeso & Grubb, 1993) or increased by removing developing fruits (Lauri & Terouanne, 1999) or infructescences. The resource status of the plant as a whole can also be increased, for instance by supplementing nutrient solution (Stephenson, 1984; de Jong & Klinkhamer, 1989). Another possibility is to increase light availability on a part of the plant or on the whole plant in case this is the limiting factor for photosynthesis (in forest understory plants, for example).
Manipulating pollen deposition Supplemental hand pollination, achieved by adding ample pollen from a mix of outcross donors, is only one way to manipulate pollen availability. In order to mimic a more natural increase in pollination services, pollinators could be added by enclosing them in cages around the plants, as is sometimes done in agricultural seed production (Carre et al., 1998; Thomson, 2001). Sperens (1996) added pollen to one inflorescence and then checked for a decrease in fruit set in its neighbour and in inflorescences further away to investigate dynamic resource allocation and pollen limitation in Sorbus aucuparia, and Parker & Haubensak (2002) applied the method proposed by Zimmerman & Pyke (1988) to study pollen limitation in two broom species (Cytisus scoparius and Genista monspessulana). A second way to manipulate pollen deposition is to exclude pollinators from one or more inflorescences (Fox, 1992; Karoly, 1992), thereby increasing local resource availability for the remaining inflorescences by taking away a sink. A combination of additional pollination on certain inflorescences and pollinator exclusion on others (Fox, 1992) will increase the change in source–sink relations and should produce a more pronounced effect. One problem that can be caused by pollinator exclusion is an accumulation of nectar in the unvisited flowers, which will render them very attractive once the exclosure is removed. Relatively short exclosure times, or the use of partial exclosures with a certain mesh size, which allow the passage of smaller, nonpollinating visitors while excluding the larger pollinators (Hingston et al., 2004), could be a solution to this problem.
Manipulations and controls Of course, a combination of pollen and resource manipulations will probably be most useful (Stephenson, 1984; de Jong & Klinkhamer, 1989), if only to keep one factor constant (usually pollen deposition rates) while investigating the effect of the second (resource availability). In all these manipulations, there should be appropriate controls on unmanipulated plants, as Zimmerman & Pyke (1988) already pointed out. I would prefer to modify their experimental set-up, which uses one control on the manipulated plant and one on the control plant. The ideal situation would be a pairwise comparison of two plants on which two units each are marked, of comparable sizes and positions and at the same distance from each other. On the control plant, each marked unit would serve as the control for its counterpart on the treated plant, on which one unit is manipulated to detect its influence on the second. In this way, positional effects can be controlled for, which is not the case when comparisons are made between the single control unit on the control plant and both unmanipulated and manipulated units on the treated plant.
The use of stable isotopes A powerful technique to detect the movements of resources in a plant is the use of stable isotopes, especially 14C, to trace the destination of photosynthates produced by a specific leaf at a specific moment (e.g. Hansen, 1969; Watson & Casper, 1984). The patterns observed often follow the vertical links formed by vascularization patterns, in such a way that subtending leaves or bracts on one side of a stem only feed the inflorescences or branches on the same side (Watson & Casper, 1984; Carroll & Delph, 1996). Although the results from such a trace study really belong in the second, observational phase, I mention the technique here because it may not be readily available to, and applicable by, the average pollination ecologist, but used in combination with manipulative experiments it could be a great help in elucidating the patterns of local resource allocation.
Pollen limitation and the dynamic concept of resource allocation: perspectives
Casper & Niesenbaum (1993) made a number of predictions based on known patterns of resource and pollen distribution. Pollen is more likely to be the limiting factor for seed production early in the season, when resource availability is still high and pollen deposition might be lower than later in the season. Seed and fruit abortions are expected to be more frequent towards the end of the flowering season, but less important when pollen deposition is low.
By zooming in on within-plant patterns rather that just using whole-plant pollinations, we should be able to acquire more insight into the role of pollen limitation in determining spatial and temporal patterns of fruit and seed set in plants with widely varying growth forms and flowering strategies. Severe pollen limitation will have an effect on total seed production, but varying levels of pollen deposition can influence resource allocation patterns. As plants differ in their ability to match resource supply with pollen deposition (and thus minimize the effects of pollen limitation), I expect the following factors to play an important role in defining the possible strategies and patterns in different plant species.
Determinate vs indeterminate flowering
Some plants develop all their flower buds well before the flowering season, without the opportunity to produce more flowers or inflorescences later on. This is true for many trees, such as apples (Malus) and pears (Pyrus), but also for a number of alpine and arctic perennial herbs, some of which may need 4 yr to develop an inflorescence from initiation to maturity (e.g. Diggle, 1997). It is to be expected that these plants, in order to cope with the unpredictability of pollen and resource availability, will invest proportionally more in flower production relative to the amount of resources available for fruit and seed production (Lloyd, 1980), and adjust fruit production by increasing the level of fruit abortion, as predicted by Burd (1995) for the optimum number of ovules per flower. The disadvantage of having to produce more flowers (or ovules) could be offset by increased opportunities for selection for good pollen donors (female choice).
Synchronous vs sequential flowering
When all flowers develop simultaneously, no adjustment of flower number is possible in response to different levels of pollen deposition. When flower production is more spread out, a plant can react to a previous episode of low pollen deposition and subsequent low fruit or seed set by investing more in subsequent flowers that can take advantage of the unused resources. If pollination levels are sufficient for the first fruit, subsequent flower production is suppressed (Stephenson et al., 1988). Again, plants with just a single flowering flush will produce more flowers and have more fruit abortion than plants with possibilities for subsequent flower production. Stephenson (1984) used the analogy of birds that produce a single clutch per season vs those that can have multiple clutches. The latter can adjust the size and number of subsequent clutches as a response to the energetic costs of the first clutch. Determinate flowering and synchronous flowering will often be coupled, as are continuous flowering and indeterminate flowering (Fig. 4).
IPU size and number
The degree to which a plant is subdivided into IPUs will influence resource dynamics and hence the ability to transport resources to the flowers that received most of the pollen. In Salix species, catkins only 5–10 cm apart appear to be physiologically independent (Fox, 1992), whereas in Sorbus aucuparia inflorescences were found to influence each other at a distance of 60 cm, but not at 120 cm (Sperens, 1996). If pollen deposition varies greatly among flowers, a plant with many, small IPUs will experience more variation in average pollen deposition per IPU, as this depends on the pollen deposition in a limited number of flowers within the IPU. There will thus be a higher probability of finding pollen limitation in plants with these IPUs than in plants that have larger IPUs, where the higher flower number reduces the effect of a few flowers with low pollen deposition.
Presence of leaves and bracts
Because green plant parts are important local contributors of assimilates, the ability of a flower or inflorescence to function independently from others depends on the amount of photosynthetically active tissue in its direct environment. I predict that individual flowers interspersed with leaves, or inflorescences with bracts, will be more independent (and thus form separate, smaller IPUs) than dense inflorescences without any bracts, which depend on leaves under the inflorescence for assimilates. This makes it more likely that flowers within such an inflorescence function as one unit, although there can still be an effect of vascular strands. In Silene latifolia, Carroll & Delph (1996) found that the two branches in the inflorescence were largely independent: one was connected to one of the two leaves under the inflorescence, while the other branch depended on the other leaf of the pair.
I hope that by showing a different way to consider the role of pollen and resources in plant reproduction, I have stimulated others to undertake more studies within this framework, which in the long term should allow us to draw further conclusions and make comparisons among different growth forms and flowering strategies.
The comments of the participants of the 2003 meeting of the Scandinavian Association of Pollination Ecologists (SCAPE) on a talk I gave during the meeting were very useful in developing my ideas in their early stages. I thank Javier Guitián Rivera for sending me a copy of the PhD thesis of Luis Navarro on Petrocoptis grandiflora, and three anonymous reviewers whose pertinent remarks helped to markedly improve an earlier version of this paper. This is publication number BRC104 of the Biodiversity Research Centre of the Université catholique de Louvain.