The clearest sites are inhabited by two different species: colour and preference distributions are strongly bimodal with few or no intermediates; colour is strongly associated with microdistribution and with alleles at the long wavelength sensitive opsin gene (LWS): Red males and females that prefer mating with red males live deep and have fixed an LWS allele that makes them sensitive to the red-shifted ambient light in their habitat. Blue males and females that prefer mating with blue males live shallow in rocky crevices, and have fixed an LWS allele that is less strongly red-shifted. They also have different diets and are significantly differentiated at many neutral loci.
2. The mate choice traits that mediate assortative mating at the differentiated islands (female preferences and male nuptial colours) are already variable at the non-differentiated islands. Pundamilia is not unusual in this regard. There is a large number of cichlid populations in Lake Victoria that segregate for the different nuptial colours.
3. Populations from islands with intermediate water clarity are intermediate in that they show strongly bimodal frequency distributions of the mate choice traits, male nuptial colour and female preference, strongly differentiated allele frequencies at the LWS opsin gene, and weakly but significantly different allele frequencies at neutral loci. These populations also show an intermediate level of association between colour and microdistribution; reds have a much wider depth range, while blues are restricted to shallow waters. Most females have strong mating preferences, and these are based on the male coloration.
4. The data are compatible with models of speciation by sensory drive, where divergent light conditions at different water depths and divergent adaptations of the visual system affect mating preferences and the fitness of male nuptial colour morphs.
5. The same nuptial colour cues that females use for choosing mates are also used by males when defending territories against competitors. Evidence gathered from a number of different experiments are consistent with the hypothesis that colour-mediated competition between males contributes a negative frequency-dependent selection component but it is asymmetric and hence unsuitable to stabilize the coexistence of the incipient species without contributions of other mechanisms.
6. Parasitological data have yet to be collected from intermediate sites on the transect to ask whether sexual selection for parasite resistance in different parts of the environment may contribute to divergent selection on mating preferences and colours. Imprinting experiments ought to be conducted with fish from intermediately differentiated and non-differentiated populations to better understand the role of maternal imprinting in the early stages of speciation.
7. The accumulation of divergence in a successively larger number of traits along the Pundamilia ‘speciation continuum’ suggests increasing dimensionality of niche divergence. One potentially important observation then is that the dimensionality of niche divergence can grow rapidly with decreasing gene flow or vice versa, gene flow can rapidly decrease with increasing dimensionality of niche divergence. That divergent daptation and gene flow can be reciprocally constraining has been shown for other incipient species systems too, suggesting a feedback loop between gene flow and selection. If the analogy of the different stages of differentiation on the different islands with successive stages in speciation is valid, the observations in Pundamilia may imply that dimensionality of niche divergence and gene flow restriction co-evolve. A minimum dimensionality of niche divergence may be required for divergent selection to be sufficiently strong to initiate speciation (here the occupation of a sufficiently large range of water depth to be exposed to different selection regimes). However, as soon as gene flow becomes sufficiently reduced, divergence in other traits becomes possible, which may be under less strong divergent selection. The increased dimensionality of niche divergence, in turn, is likely to reduce gene flow further.