We all know the power of natural selection to drive evolution toward more successful characteristics, or phenotypes. At the level of DNA, however, most changes convey neither advantages nor disadvantages--they are neutral. Random drift can happen at the phenotype level as well, but at the genotype level it predominates.
The indifference of evolution to the vast majority of molecular changes was described by Japanese geneticist Mooto Kimura in 1968. Mutations that improve fitness are very rare, he argued. Those that make an organism worse off are more common, but if they are truly detrimental they will never be passed on. Most mutations, however, will make little or no difference to survival. These neutral mutations will be free to accumulate at random.
In the simplest mathematical model, mutations arise in each nucleotide of DNA with constant probability each generation. A larger population will include proportionally more mutations at each position. For a mutation to become "fixed," however, it must spread through the entire population, rather than dying out. The larger the population, the less likely this is; the fixation probability that is inversely proportional to the population size. Because the product is a constant, the rate at which mutations become fixed at any position doesn't depend on population size. (For a very clear recent introduction to this model in the context of geographically heterogeneous populations, see the July 2009 article in Physics Today by Oskar Hallatschek and David R. Nelson.)
The constant accumulation of fixed mutations underlies the powerful "molecular clock" technique, which allows researchers to estimate how recently two species diverged from one another by counting the number of differences that have accumulated in corresponding DNA sequences. Although the actual rates are more variable than would be expected from this model, molecular clocks provide a powerful quantitative window into our evolutionary past and into relationships between living species.
The background accumulation of mutations over evolutionary time also makes it possible to discern sequences that don't change. As a rule, this happens when changes in those sequences prevent survival or reproduction, so they never accumulate: they are constrained during evolution. Researchers often regard constraint (or the related property of sequence conservation) as a sign that a particular section of DNA has a critical function, even if they don't yet know what that function is.
The best known examples of the two types of mutation are those in sequences that code for amino acids in proteins. Mutations that change the amino acid can destroy the function of the protein, so they will be constrained if the protein itself is important. Researchers like Arend Sidow at Stanford have shown that the sequence is highly constrained in the active sites of proteins but less so in less critical regions.
By contrast, because the genetic code is redundant, some mutated groups of three bases still specify the same amino acid, so the protein chain will be unchanged. These "synonymous" mutations are often used to calibrate the background mutation rate. However, the exact base sequence can still have an effect, for example changing the sequences preferences among different alternative splicing arrangements of the final RNA.
Although constraint and function often go together, there are exceptions, as I discuss in my July 10 story in Science (subscribers only, but I have a pdf in the Clips section of my website, encrypted with the password "monroe"). Sometimes important sequences don't seem to be constrained, and sometimes constrained sequences don't seem to be important. Understanding when and why this happens is important as researchers look for new functions in the 98.5% or so of the human genome that doesn't code for proteins, which includes microRNAs and other regulatory regions.