Though the term ‘coevolution’ was coined less than 50 years ago (Ehrlich and Raven, 1964), over 100 years prior to that Darwin first described the process to account for the matching of tongue and tube lengths of bee species and the clover species they visited (Darwin, 1859). He expounded on this process when describing the exceptionally long spur of Angraecum sesquipedale and predicted a moth pollinator with an equally long tongue (Darwin, 1862). Darwin envisioned a ‘race’ in which long-tongued individuals had a fitness advantage in gaining food from exceptionally long-spurred flowers and long-spurred flowers had a fitness advantage in reproduction due to increased pollination. Without a countervailing force, the tongue length of the pollinator and the spur length of the plant would increase due to the reciprocating fitness advantages. When traits of two interacting species such as tongue and floral tube length match tightly, it is easy to assume that Darwin's race, that is, coevolution, is responsible.

We recently tested predictions of Darwin's coevolutionary race model and an alternative, the ‘pollinator shift’ model (Wasserthal, 1997), as applied to the evolutionary history of spurs among species in the North American Aquilegia clade (Whittall and Hodges, 2007). We concluded that the pollinator shift model is a better fit for the majority of spur-length evolution, that is, species of Aquilegia have primarily evolved to fit the already established tongue lengths of their pollinators. Ennos (2008) provides an excellent summary of our results, and then offers an alternative hypothesis to account for them that combines aspects of both Darwin's coevolutionary race and the pollinator shift hypotheses. He envisions that pollinator shifts do occur but that the process begins when a new pollinator has a similar tongue length as the original pollinator, and thus both are effective pollinators. Then, Darwin's race causes the evolution of longer tongues in the new pollinator class but the original pollinator does not evolve because of some countervailing evolutionary force (for example, body size constraints). Darwin's race then ensues between the plant and the new pollinator until another counterbalancing force prevents the lengthening of the new pollinator's tongue. This counterbalancing force then causes stasis of tongue and spur-length evolution until yet another pollinator class appears with a similar tongue length to carry on Darwin's coevolutionary race.

Ennos’ model predicts pollinator shifts followed by stasis in spur length as does the pollinator shift model, which envisions that plants adapt to fit the pre-established tongue length of a pollinator with no reciprocal evolution by the pollinator. Ennos (2008) goes on to claim that the pollinator shift hypothesis ‘requires a rather implausible ecological scenario’ because it requires that the original pollinator must be absent from the population for many generations. We disagree. Instead, the pollinator shift model only requires that in some part of the plant species’ range a longer tongued pollinator becomes the predominant visitor and the cause of most plant reproduction (Whittall and Hodges, 2007). First, bees and hummingbirds do not have to have similar tongue lengths to be similarly effective at pollen transfer of ‘bee-adapted’ flowers (Castellanos et al., 2003) as Ennos claims. Thus changes in the abundance of these pollinator classes could substantially change the selection regime for floral morphology (Castellanos et al., 2003). Furthermore, temporal and spatial variation in pollinator abundances have been found to be substantial and is thus certainly not implausible (for example, Herrera, 1996; Herlihy and Eckert, 2005; Price et al., 2005; Brunet and Sweet, 2006). Changes in pollinator abundance may occur over long time periods owing to a number of factors, including changes in pollinator migration patterns, extinction of pollinators, changes in community composition or adaptation of plants to a new habitat (see Thomson and Wilson, 2008). Though such situations may be rare, this does not make them unimportant to plant evolution.

In addition to ecological changes, floral characters can affect the frequency of pollinator visitation. In Aquilegia, hawkmoths do not discriminate between control flowers and those with artificially shortened spurs, but variation in flower orientation and color have very large effects on visitation (Fulton and Hodges, 1999; Hodges et al., 2002). Similarly in two species of Mimulus, introgression of alternate quantitative trait locus alleles (QTL) for flower color caused dramatic changes in the frequency of visitation by bees and hummingbirds (Bradshaw and Schemske, 2003). Thus, changes in traits other than nectar spurs may precipitate shifts in pollinator visitation rates. Such changes would be especially likely to cause shifts when the new pollinator was already particularly abundant (Bradshaw and Schemske, 2003). Once the setting for more frequent visitation is created then there would be very strong selection for other floral traits, such as nectar spur length, to become optimized for plant reproduction via the new pollinator (Thomson and Wilson, 2008).

There are further aspects of Darwin's coevolutionary race model, alone or as part of Ennos's hybrid model, which seem to us to make it an unlikely explanation for the majority of spur-length evolution in the North American Aquilegia clade (or more generally, floral tube-length evolution). For example, pollinators such as hummingbirds likely already had tongue lengths similar to those found today prior to their first association with bee-adapted Aquilegia. One clade of hummingbirds, called bee-hummingbirds, is comprised of two sister clades, one including all North American species and the other including primarily South American species (McGuire et al., 2007). Phylogenetic analysis indicates that the ancestor of bee-hummingbirds occurred in South or Central America (McGuire et al., 2007). This ancestor also likely had a bill morphology similar to extant bee-hummingbird species because species in both sister clades have similar bill morphology (around 1.5 to less than 2.0 cm; Colwell, 2000). Thus, bill size of hummingbirds was likely established prior to their invasion of North America and an interaction with Aquilegia (or other North American species that became adapted to hummingbirds). Furthermore, molecular dating indicates that bee-hummingbirds originated about 6 mya (Bleiweiss, 1998), while the origin of hummingbird pollination in Aquilegia was likely to have been far more recent at around 1.7 mya or less (Hodges et al., 2004; Kay et al., 2006). Taken together, these results strongly suggest that bill morphology was established prior to an association with Aquilegia and that it did not substantially change thereafter. We have already made a similar argument concerning hawkmoth tongue lengths (Whittall and Hodges, 2007).

Another reason we doubt coevolutionary hypotheses provide a major explanation for most spur-length evolution is because the interaction between plant and pollinator is likely to be very asymmetrical. Hummingbirds and hawkmoths visit a large number of species (for example, Grant, 1983, 1994) and thus they do not have a tight dependency on any one plant species. For example, the longest spurred Aquilegia species, A. longissima, belongs to a guild of unrelated species from six distinct genera in the desert (southwest USA), all with floral tubes greater than 9 cm, which by and large share the same hawkmoth pollinators (Grant, 1983; Grant and Grant, 1983). Thus, the ability of the longest tongued individuals to reach a bit more nectar from the longest spurred individuals of a single species would provide a rather small fitness advantage for the pollinator, at best. In contrast, reproduction in most plants is dependent on the visitation of pollinators and thus selection will be quite strong to optimize floral morphology to maximize pollen transfer. Thus, unrelated plant species visited by the same pollinator will likely converge on a similar floral morphology, such as tube or spur length. Similar arguments have been made during a previous discussion of Darwin's coevolutionary race and the pollinator shift hypothesis (Jermy, 1999). Finally, Ennos’ model requires a new species of hummingbird or hawkmoth for each transition to a new pollinator so that Darwin's race can be run each time anew. Given the large number of shifts from bee to hummingbird and from hummingbird to hawkmoth pollinated plant species in North America (for example, Grant, 1994; Whittall and Hodges, 2007; Thomson and Wilson, 2008) and the much smaller number of hummingbird and hawkmoth species, this seems unlikely as a general explanation for the evolution of tube and tongue lengths.

While we believe that the pollinator shift hypothesis accounts for the majority of spur-length evolution in Aquilegia, and likely many other species as well, some spur and tongue lengths may well be due to Darwin's coevolutionary race. How would one identify such instances? We agree with Ennos that complementary, and more specifically, phylogenetic studies of plants and pollinators will be needed. If evolution among species or populations of both plants and pollinators indicates increasing lengths during the same time period and in the same geographical area, then this would provide strong evidence for Darwin's race. Even still, it will be necessary to establish that the respective fitness advantages are due to the plant–pollinator interaction rather than some other factor (for example, Wasserthal, 1997; Borrell, 2005; Strauss and Whittall, 2006).