Paths on Fitness Landscapes

Another point: Ongoing natural selection in two populations can allow evolutionary events to occur that would be impossible in a single well-mixed population, since it allows for simultaneous exploration of divergent paths. Natural selection is short-sighted: Alleles increase in frequency because of their current advantage, not because they might someday be useful. Think of various possible solutions of some problem as hills, with higher hills corresponding to better solutions. Natural selection climbs up the first hill it chances upon; it can’t see that another solution has greater possibilities in the long run. Not only that: Since the environmental conditions of Europe and Africa were significantly different, evolution could try solutions in Europe that couldn’t be explored in Africa, because the initial step along that path had negative payoffs in Africa. In Europe, for example, you had to worry about staying warm enough, whereas Africans faced heat stress: These issues were important considerations in the evolution of larger brains. It may be that the relative unimportance of heat stress in Europe opened up some evolutionary pathways that had greater long-term possibilities than the ones that developed in Africa.

Consider an analogy from the history of technology. Somewhere back in late classical times, the use of the camel was perfected—a better saddle was developed, for example, one that allowed camels to carry heavy loads efficiently. Throughout most of the Middle East and North Africa, camels were (after those developments) a superior means of land transportation: They were cheaper than ox-drawn wagons and not dependent upon roads. Over a few centuries, people in areas where camels were available abandoned wheeled vehicles and roads almost entirely.19 You can still see the effects in the oldest sections of some cities in the Arab world, where the alleys are far too narrow to have ever passed a cart or wagon. Europeans, not having camels, had to stick with wheeled vehicles, which were clearly more expensive, given the infrastructure they required. But as it turned out, wheeled vehicles—in fact, the whole road/wheeled vehicle system—could be improved. Back then, when camels seemed so much better, who knew that someday there would be horse collars and nailed horseshoes, then improved bridge construction, suspensions that reduced road shock, macadamized roads, steam power, internal combustion engines, and ultimately the nuclear Delorean. The motto here is that sometimes the apparently inferior choice has a better upgrade path: Evolution can’t know this, and we aren’t particularly good at recognizing it ourselves. On the genetic level, it translates as follows: Natural selection may solve the same problems differently in different populations, and what appears to be the most elegant solution at the time may not in fact turn out to be the one that works best in the long run. The seemingly inferior choice may come out on top down the road. It is easy to think of plausible cases: Imagine, for example, that excess heat production limited the trend toward larger brains in Africa, while in the climate of Europe heat was not much of a problem. Later, as evolution fine-tuned the physiology of large brains, much of the heat problem was solved—and so the new brain could then spread in Africa as well.

The idea of a fitness landscape was introduced by Sewall Wright (1932), and it has become a standard imagination prosthesis for evolutionary theorists. It has proven its value in literally thousands of applications, including many outside of evolutionary theory. In Artificial Intelligence, economics, and other problem-solving domains, the model of problem-solving by incremental hill-climbing (or “gradient ascent”) has been deservedly popular. It has even been popular enough to motivate theorists to calculate its limitations, which are severe. For certain classes of problems—or, in other words, in certain types of landscape—simple hill-climbing is quite impotent, for an intuitively obvious reason: the climbers get stuck on local second-rate summits instead of finding their way to the global summit, the Mount Everest of perfection. (The same limitations beset the method of simulated annealing.) The Local Rule is fundamental to Darwinism; it is equivalent to the requirement that there cannot be any intelligent (or “far-seeing” ) foresight in the design process, but only ultimately stupid opportunistic exploitation of whatever lucky lifting happens your way.

What Eigen has shown is that this simplest Darwinian model of steady improvement up a single slope of fitness to the optimal peak of perfection just doesn’t work to describe what goes on in molecular or viral evolution. The rate of adaptation by viruses (and also of bacteria and other pathogens) is measurably faster than the “classical” models predict—so fast that it seems to involve illicit “look-ahead” by the climbers. So does this mean that Darwinism must be abandoned? Not at all, for what counts as local depends (not surprisingly) on the scale you use.

Eigen draws our attention to the fact that when viruses evolve, they don’t go single-file; they travel in huge herds of almost identical variants, a fuzzy-edged cloud in the Library of Mendel that Eigen calls a “quasi-species.” We already saw the unimaginably large cloud of Moby Dick variants in the Library of Babel, but any actual library is likely to have more than one or two variant editions of a book on its shelves, and in the case of a really popular book like Moby Dick it is also likely to have multiple copies of the same edition. Like actual Moby Dick collections, then, actual viral clouds include multiple identical copies but also multiple copies of minor typographical variants, and this fact has some implications, according to Eigen, that have been ignored by “classical” Darwinians. It is the shape of the cloud of variants that holds the key to the speed of molecular evolution.

A classical term among geneticists for the canonical version of a species (analogous to the canonical text of Moby Dick) is the wild type. It was often supposed by biologists that among the many different genotypes in a population, the pure wild type would predominate. Analogous would be the claim that in any library collection of copies of Moby Dick, most copies will be of the received or canonical edition—if there is one! But this doesn’t have to be the case for organisms any more than for books in libraries. In fact, the wild type is really just an abstraction, like the Average Taxpayer, and a population may contain no individuals at all that have exactly “the” wild-type genome. (Of course, the same is true of books—scholars might debate for years over the purity of a particular word in a particular text, and until such debates were resolved, nobody could say exactly what the canonical or wild-type text of that work was, but the identity of the work would hardly be in jeopardy. James Joyce’s Ulysses would be a good case in point.)

Eigen points out that this distribution of the “essence” over a variety of nearly identical vehicles turns out to make that essence much more movable, much more adaptable, especially in “rugged” fitness landscapes, with multiple peaks and few smooth slopes. It permits the essence to send out efficient scouting parties into the neighboring hills and ridges, ignoring wasteful exploration of the valleys, and thereby vastly (not Vastly, but enough to make a huge difference) enhancing its capacity to find higher peaks, better optima, at some distance from its center, where the (virtual) wild type sits.

The reasons it works are summarized by Eigen as follows:

Functionally competent mutants, whose selection values come close to that of the wild type (though remaining below it), reach far higher population numbers than those that are functionally ineffective. An asymmetric spectrum of mutants builds up, in which mutants far removed from the wild type arise successively from intermediates. The population in such a chain of mutants is influenced decisively by the structure of the value landscape. The value landscape consists of connected plains, hills, and mountain ranges. In the mountain ranges, the mutant spectrum is widely scattered, and along ridges even distant relatives of the wild type appear with finite [that is, not infinitesimal] frequency. It is precisely in the mountainous regions that further selectively superior mutants can be expected. As soon as one of these turns up on the periphery of a mutation spectrum the established ensemble collapses. A new ensemble builds up around the superior mutant, which thus takes over the role of the wild type___This causal chain results in a kind of ‘mass action’, by which the superior mutants are tested with much higher probability than inferior mutants, even if the latter are an equal distance away from the wild type. [Eigen 1992, p. 25.]

最后一段跟群选择无关，因为涉及的可能是很多物种，这是个关于生态系统的观点。