Research over the last 20 years has shown that a large fraction of insect species has intimate associations with symbiotic bacteria that often have major effects on their biology. Some symbionts are mutualists increasing their host's fitness, whereas in other cases the host obtains no benefit or even suffers from carrying the micro-organism. Bacteria that do not benefit their carriers need other tricks to spread themselves through the host population. A particularly subtle strategy is cytoplasmic incompatibility (CI), in which, through a mechanism that is described below, the bacteria spread not by benefiting their host but by reducing the fitness of non-carriers. Until recently a single type of bacterium, Wolbachia, was thought to spread by CI. In 2003, Molly Hunter's group at Tucson discovered a second, unrelated bacterium, Cardinium, which could do the same thing (Hunter et al., 2003). In this issue, Harris et al. (2009) provide a rare quantitative test of the theory of bacterial spread by CI. They find that Cardinium spreads more readily than theory predicts and explore why this might be so.

The manner in which CI works in Wolbachia (and presumably in Cardinium, about which less is known) is that infected males in some way modify their sperm so that they can be utilized only by females that also carry the bacteria (Werren, 1997). Despite intensive investigation (and the availability of several complete Wolbachia genome sequences), it is still not understood how this modification/rescue system works mechanistically, although it seems to have something to do with the decondensation of the paternal chromosomes in the zygote. Females carrying Wolbachia can therefore mate successfully with any male, whereas uninfected females will have lower fitness because, whenever they mate with an infected male, their eggs will remain unfertilized. Wolbachia spread is therefore an example of what has been called ‘evolutionary spite’—carriers obtain a relative fitness advantage by harming non-carriers. Note that when Wolbachia is rare, and non-carriers mate only occasionally with infected males, their relative fitness disadvantage will be small and the bacterial infection will increase only slowly. But as Wolbachia increases in frequency, uninfected females will suffer progressively higher costs, and the fraction of infected individuals will accelerate until everyone carries it.

The argument as just stated assumes that carrying the CI bacteria has no fitness consequences for the host. In fact, there is plenty of evidence that carriage may be costly, and also that the prohibition on the use of modified sperm by uninfected females may not be absolute. In these circumstances, theory predicts that there is a threshold CI bacteria frequency that has to be exceeded before the infection begins its autocatalytic spread. We can calculate that threshold (denoted by p) by recognizing that at this unstable equilibrium the fitness of infected and uninfected females must be exactly the same. Let us assume that carrying Wolbachia reduces the host fitness from 1 to (1−sf). The fitness of uninfected females is reduced whenever they mate with an infected male (probability p) by an amount equal to the probability that they cannot use the modified sperm (denoted by sh), giving (1−p sh). Putting (1−sf)=(1−p sh) and solving for p gives the threshold, p=sf/sh.

Based on previous studies and new experiments, Harris et al. (2009) were able to estimate these parameters. The strength of CI was not as high as often observed in Wolbachia, and uninfected females lost only about two-thirds of their offspring (sh=0.62) when mating with infected males. Carrying Cardinium reduced the fecundity by about a fifth (sf=0.18). Our calculation suggests that the threshold frequency that the Cardinium must attain before it can begin to spread is 0.18/0.62=0.29. Harris et al. (2009) actually did a more sophisticated calculation (Turelli and Hoffmann, 1991; Vavre et al., 2000), taking into account the rare loss of the symbiont during maternal transmission and the haplodiploid genetics of the host (a parasitoid hymenopteran with a bizarre life history, but that is another story), and predicted the threshold was somewhat higher at ∼0.33, averaged across the sexes.

To test this prediction, Harris et al. (2009) set up population cages for which the initial frequency of infection was 0.15, ∼0.33 and 0.55, respectively. They expected that in the first case Cardinium would definitely invade, in the last case it definitely would not, while the middle case could go either way. To their surprise, Cardinium invaded in all cases. Why?

This is the first study of the dynamics of Cardinium, but related work on Wolbachia offers some clues as to what might be happening. The effect of Wolbachia on host fitness parameters can depend very subtly on the conditions under which they are measured (Hoffmann et al., 1990), and may also be affected by the internal dynamics of the bacterial population within the host, especially as it influences bacterial densities. A statistical analysis of the data suggests that they are best explained by assuming weaker CI and less costs (or even benefits). Perhaps the conditions in the population cages (and possibly also in the field) were more propitious to the bacterium than those in the laboratory experiments used to estimate the parameters. Over the last few years a series of fascinating examples of how some insect endosymbionts provide benefits to their host have been discovered, and in many cases the benefits are realized only when the host faces an abiotic or biotic challenge. It is likely that we are still some way from fully understanding all of the complexities of the mutualistic interaction. It could also be that in the interval between measuring the parameters and performing the experiment the bacteria have evolved. One of the best-studied endosymbiont interactions in the field involves a strain of Wolbachia attacking Drosophila simulans in California. Weeks et al. (2007) have recently shown that over the 20 years that it has been studied, it has evolved to be less costly to its host.

Only further study will resolve these issues. A better understanding of insect–symbiont biology will require not only mechanistic explorations of how the two parties interact, but also more cage and field population experiments to explore their population and genetic dynamics. Harris et al.'s (2009) blend of theory and experiments is a fine example of how this should be done.