Why are some hoverflies poor mimics?

Merodon equestris, the Narcissus fly, a bumblebee mimic

Eristalis tenax, a dronefly (top left) and honeybee

Volucella zonaria, a large and colourful hoverfly that is an excellent mimic of the european hornet Vespa crabro

Helophilus, a wasp mimic 

 Sericomia silentis, a wasp mimic

The photos above illustrate that hoverflies are amongst the best examples of mimicry. Although harmless, their colours, patterns, level of hairiness, and detailed morphological and behavioural features often matches species of bees, bumblebees and wasps, making them such wonderful impostors that they fool people into thinking they really are stinging bees or wasps. At close range, anybody can learn to tell mimics and models apart (the shape and size of the antenna is a giveaway). But the fantastic mimicry of hoverflies is shown by the repeated identification failures of natural history photographers, and even specialised editors (a book cover on bees featuring a hoverfly). These are evident when you have a look at many natural history photography websites. As an example I googled "wasp flower" and the first hit was a hoverfly.

The selective pressure for mimicry is predation. Many birds are consummate fly predators, but they will avoid bees and wasps. Although birds could benefit nutritionally from preying upon these succulent, soft flies there is a heavy cost to pay if they get it wrong: a painful and possibly debilitating sting. Given this, shouldn’t natural selection perfect mimicry? Why are there some hoverflies that only have a vague resemblance to bees and wasps?
 Here is an example:

Syritta pipiens, an example of a poor mimic

 There had been many hypotheses put forward to explain why there are poor mimics:

1. 'Eye of the beholder': Imperfect mimics are only imperfect to the human eye, they are really much better mimics to bird eyes. The poor resemblance would be a reflection of our own perception bias.
2. 'Multimodel' Maybe mimics could be safer when imitating roughly several models.
3. 'Kin selection' Flies that are abundant might be surrounded by relatives, and kin selection might have selected for these imperfect mimics.
4. 'Trade offs' mimetic perfection cound be trading off with camouflage ability, or thermoregulation.
5. 'Constraints' mimicry could be limited by developmental or phylogenetic constraints.
6. 'Relaxed selection': small hoverflies – which are less profitable to eat – could have lower predation even if they are poor mimics, which will mean that natural selection for improved mimicry will be weaker.

Heather Penney and co-authors tested these hypotheses using a large set of European hoverfly species. They obtained morphological data and human volunteer rankings of mimicry fidelity to bee, bumblebee and wasp models and phylogenetic data to test these hypotheses.

First, they carried out a detailed morphological analysis on a large set of hoverlfy species which shows that human perception of mimicry fidelity indeed corresponds to objective assessments of similarity between mimic and model based on their morphological analysis. This rules out the 'eye of the beholder hypothesis" and also the 'multimodel' as poor mimics did not fall in between different potential models.

They also performed a meta-analysis showing that poor mimics were not more abundant than accurate mimics and therefore rules out the importance of kin selection in the evolution of imperfect mimicry.

The constraints hypothesis, they did not explicitly test even when they had phylogenetic data to evaluate the importance of phylogeny as a constraint.

The core result of the paper was that the degree of fidelity to the model is strongly correlated with hoverfly body size even when correcting for phylogenetic relationships, as shown in their figure:

Relationship between an estimate of body size and human ratings of mimetic fidelity (fHR). Lines 

show the fitted linear regressions. Filled triangles, wasp mimics; open triangles, bee mimics; filled circle, the non-mimetic syrphid Cheilosia vernalis (from Penney et al 2012)

As the benefit of predating a hoverfly (a meal for the bird) is directly related to the size of the fly, while the cost of error should be constant, the higher risk of predation when the hoverfly is large imposes a higher selective pressure to be a good mimic. Therefore, a hoverfly doesn’t need to be a good mimic if it is small, as the benefit of eating a small fly for a bird is too insignificant to risk the potential cost, therefore resulting in a lower predation pressure for small hoverflies. Birds will err on the side of caution and avoid even the poor mimics when they are small. The other way round: Large hoverflies provide a bigger benefit, so the predator might risk being stung and therefore they have been selected for more precise mimicry. 

Although the results are solid and the hypothesis makes intuitive sense, I found two problems with their design: (1) the limited choice of models and (2) the geographically biased choice of some models, both of them resulting in an underestimation of the degree of mimicry of some hoverflies.

First, not all potential models or even model types were available and this will create an alternative explanation for the evolution of "poor" mimicry, if you can't compare the mimic with its model, they it will by necessity result in a poor mimic. There are hundreds of European species of bees, wasps and bumblebees with a broad range of colours and patterns, but they only used ten. Syritta pipiens, the little hoverfly shown above was rated as a poor mimic in their study, but who says it doesn't mimic a small solitary wasp or bee not included as a model?
In addition, of the three bumblebee species they included two were American bumblebees (B. affinis and B. impatiens), which seemed strange given that their hoverfly sample was European. A single European bumblebee, B. lucorum was included. Given that bumblebees differ extensively in coat colour, this might create biases in the evaluation of mimicry fidelity in bumblebee mimics: they cannot be regarded as excellent mimics if their models are not available for comparison. This is likely to have caused a bias in the evaluation of, for example, the bumblebee mimic Merodon equestris, the Narcissus fly, also illustrated above, which is highly polymorphic for coat colour pattern and mimicks different bumblebee species. They included just a single colour form of this hoverfly, which imitates the all-brown Carder Bee B. pascuorum, but not a single all-brown bumblebee was used as a model. This predictably resulted in (1) the misclassification of M. equestris as a honeybee mimic and (2) his rating as a poor mimic, when this species is in fact a fantastic bumblebee mimic which has tricked me in more than one occasion.
   I wonder up to what point the narrow selection of hymenoptera could result in the assessment of many hoverflies as poorer mimics than they really are. We might not really know some hoverfly species are really imitating because the models they mimic might not relevant to us such as large bees and wasps are.
 Although I think they are valid concerns, I don't think the conclusions of this paper would be affected, but they might have ever got stronger results if their model choice had been wider.

More information

Penney, H., Hassall, C., Skevington, J., Abbott, K., & Sherratt, T. (2012). A comparative analysis of the evolution of imperfect mimicry Nature, 483 (7390), 461-464 DOI: 10.1038/nature10961