One of the big questions in evolutionary biology is about reversibility. That is, once an organism evolves down a path of adaptation to a particular climate or biological community, how easy is it for natural selection to make a U-turn and go back to a less specialized state? Many evolutionary changes are probably irreversible - an idea that was classically expressed in "Dollo's Law": "An organism is unable to return, even partially, to a previous stage already realized in the ranks of its ancestors." But many evolutionary changes may not be irreversible - and it's not always easy to predict which ones those are.
answer this question [$-a] for a group of flowers in the genus Ruellia. The authors, Tripp and Manos, use a phylogeny to reconstruct the evolutionary history of pollination syndromes, groups of floral traits like color, nectar tube length, scent, &c, that are associated with pollination by different groups of animals.
For instance, bright red flowers with longish, narrow nectar tubes, not much scent, and large volumes of dilute nectar (like the Ruellia species in the upper figure), are almost always pollinated by hummingbirds; blue flowers with short nectar tubes, stronger scent, and small volumes of concentrated nectar (like Ruellia brittoniana in the lower figure) tend to be pollinated by bees or other insects. Other Ruellia species are pollinated by hawkmoths (white flowers, very long nectar tubes) or bats (yellow flowers, short nectar tubes, lots of dilute nectar, strong scent). Generally, syndromes associated with a single, small group of pollinators (hummingbirds, hawkmoths, or bats) are considered "specialized", while syndromes associated with many more different pollinators (bees and insects) are not.
With a phylogeny of the genus Ruellia, Tripp and Manos use the pollination syndromes of currently existing Ruellia species to estimate what pollination syndromes their ancestors may have had. Then they determine how common transitions between pollination syndromes have been in the history of Ruellia, and whether any pollination syndromes are "dead ends" - that is, whether Ruellia species that evolve to specialize on, say, hummingbird pollination are "stuck" that way.
Surprisingly, Tripp and Manos found that some specialized pollination syndromes are dead ends, but one, the hummingbird syndrome, isn't. Hawkmoth- and bat-pollinated species tended to have evolved from ancestors with the bee/insect syndrome, and they seem to be "stuck" once they get there. But in several cases, hummingbird-specialized ancestors have given rise to bee/insect-pollinated species. This has never been seen before in other, similar groups of plants. Hummingbirds are generally thought to be more efficient pollinators than bees, so while it makes sense for flowers to evolve from using bees to using birds, it's not clear how natural selection would work in the opposite direction.
Tripp, E.A., Manos, P.S. (2008). Is Floral Specialization an Evolutionary Dead-End? Pollination System Transitions in Ruellia (Acanthaceae). Evolution, 62(7), 1712-1737. DOI: 10.1111/j.1558-5646.2008.00398.x