Movement of populations through time can be viewed as a sampling process. Part of that sampling is associated with survival—you can't be sampled as a living, biotic entity if you don't survive. The other part has to do with reproduction. That is, reproduction can be viewed as the amplification of a sample of a population. Of course, organisms have to survive in order to reproduce, so for simplicity one can view sampling entirely from the perspective of reproduction. It can be convenient, though, to consider sampling instead entirely from the perspective of survival, though this disregards differences in reproductive output.
So stated, this movement of populations through time can be seen as a deterministic process. That is, if you survive, then you reproduce at some rate. The reality however is that sampling also is a random, statistical process and therefore stochastic, that is, subject to chance. Furthermore, the smaller the sampling size, then the greater the error, in other words, the greater the randomness. Thus, with very large populations, in which most individuals survive to reproduce, there is little non-deterministic difference in the frequency of alleles going from one generation to the next, and thus chance has relatively little impact on allele frequency. On the other hand, when populations are small, or few individuals survive from generation to generation, then the likelihood is also small that allele frequencies will be constant rather than stochastically varying. That is, chance has a greater impact on allele frequencies in small versus large populations.
In evolutionary biology, sampling error can be described as genetic drift. What drift accomplishes is a modification of the frequency of existing alleles, within populations, and this modification occurs in a manner that is not deterministic, i.e., is not a selective mechanism, nor is the change a function of migration. In fact, selection and drift can be viewed as competing forces. That is, when selection is very strong or populations are very large, then drift plays comparatively little role in evolution. Alternatively, when selection is weak, or populations sizes are small, then drift can be play a larger role in evolution than selection.
Selection can also give rise to drift. In a process known as periodic selection, populations come to be dominated by alleles that initially were low in frequency but nonetheless came to provide a selective advantage to their bearers (i.e., as following a selective sweep and/or clonal expansion). As a consequence of their initially low frequency – and therefore initially small absolute initial population size, and assuming 100% genetic linkage among alleles within individuals (i.e., assume that there is no sex) – then whatever alleles were originally linked with the selected allele will also come to dominate the population (i.e., increase in frequency and perhaps become fixed). Furthermore, this dominance will occur whether or not the not-selected, i.e., hitchhiking alleles are beneficial, neutral, or even detrimental. Instead what is key is that the alleles, perhaps only by chance, are found within genotypes that otherwise provide the bearer with a superlative benefit. Thus, effectively randomly chosen alleles can come to dominate populations and do so to at least some degree independently of their contribution to the overall fitness of the genotype within which they are found, and this occurs more as a consequence of random mutation, and in some cases the randomness of what can be described as preadaptation (assuming that it is within a newly colonized environment that the periodic selection is occurring), than as a consequence of sampling error per se. This issue of periodic selection will be returned to subsequently.
For genetic drift to occur to a substantial extent then population numbers either must be continually small or sampling for reproduction must be made from only a small subset of a larger population. The first example can be described as an evolutionary bottleneck and can continue for many generations, ultimately to the detriment of the experiencing population as beneficial alleles are lost by chance from the population. The second example can be described as a founder effect. That is, a new population (in the example, the next generation) is founded by a small sample of a larger, parental population. Due to chance alone, i.e., because of drift, the allele frequency of the founded population is likely to be different from that of the parental population, resulting in evolution. Again, because such evolution is not adaptive, or at least isn't necessarily so given drift, the new population may display a relative dearth of beneficial alleles, a phenomenon that may be viewed equivalently as an increase in the frequency of detrimental alleles. Also by chance, however, the population may come to possess an increased frequency of beneficial alleles, though presumably selection alone, with time, would have given rise to the same result. The point, thus, is that small population sizes mean that it becomes more difficult to predict what alleles and therefore which adaptations will and will not survive.
Muller's ratchet represents one possible outcome of genetic drift, particularly given ongoing genetic bottlenecking. The idea is based on the premise that most mutations are deleterious, that the least mutated genome is most fit, and that sampling error can lead to the fixation of deleterious genotypes. That is, in small populations, particularly if mutation rates are high, there will be a tendency for loss, by chance, of genotypes that contain no or minimal deleterious mutations. Furthermore, in the absence of gene exchange among members of these populations, there will be no recombinational mechanisms available for reestablishment of a genotype containing no or minimal deleterious mutations. That is, the surviving population may possess wild-type alleles, i.e., non-deleterious alleles at every locus, but bringing all of those alleles back into a single individual, once such an individual has been lost due to genetic drift, simply cannot occur without either sexual processes or, alternatively, fortuitous mutational events. The latter includes reverse mutations. Wild-type levels of fitness, however, also might be restored or even maintained through accumulation of compensating mutations.
With time, and continued bottlenecking, there should be a compounded loss of most-fit, least-mutated individuals, perhaps eventually resulting in insufficient fitness within a population to evade overall extinction. Interestingly, the establishment of pure cultures, i.e., the laboratory formation of a clonal population from a single individual, represents a form of genetic bottlenecking, that is, the population is passaged through only a single individual. If the population consists predominately of wild type, then this single individual probably is representative of the wild-type genotype. It is likely that a population will consist predominately of wild type if mutation rates are low and as a consequence passage through a single individual is routinely forced on populations in the microbiology laboratory, as a component of pure culture technique. Furthermore, in the course of pure culture technique propagation following bottlenecking should not be excessive. Together these approaches help to avoid the evolution of more-fit individuals. In populations where mutation rates are sufficiently high, however, the probability of choosing a single individual that is representative of wild type is not great. Genetic drift consequently could operate simply as a consequence of employing pure culture technique. Such drift is seen particularly in high mutation-rate viruses such as those possessing RNA genomes, or in mutator strains of otherwise relatively low mutation-rate organisms.
Note that what makes Muller's ratchet different from fixation of deleterious alleles due simply to sampling error, that is, due simply to drift, is the loss of the wild-type genotype without fixation of deleterious alleles. That is, polymorphism is not or at least not necessarily lost. Thus, experimentally, demonstration of Muller's Ratchet should involve showing that wild-type alleles have not been lost from populations, but that the wild-type genotype has been lost, and furthermore, that the wild-type genotype has not been replaced with more-fit genotypes (that is, Muller's Ratchet essentially by definition is associated with fitness losses rather than fitness gains). In addition, if possible it should be shown that the proposed Muller's Ratchet condition does not occur given imposition of sexual processes, but does occur (if it does occur) when a population is unable to display sexual processes.
Thus, it is relatively trivial to show that a bottlenecked population will lose beneficial alleles, but much more difficult to show that these alleles have not been lost, that nonetheless that beneficial genotypes have been lost, and furthermore that the beneficial genotypes would not have been lost had sex and recombination been sufficiently prevalent. In Muller's ratchet experiments bottlenecking therefore should not consist of only a single organism since under those conditions loss of the wild-type genotype and fixation of the non-wild-type alleles must, by definition, coincide. If possible it should also be shown that the ratchet effect is associated particularly with an absence of sex.