Natural selection covers a diversity of phenomena but basically represents the impact of environments, broadly defined, on populations. The response of individuals to this environmental impact is quantified using the concept of fitness, a.k.a., Darwinian fitness. Changes in allele frequencies can lead to increases in the frequencies of certain alleles, and potentially of certain genotypes, at the expense of others, though ultimately allele fixation, at least as defined strictly, also involves stochasticity that is in addition to this deterministic evolution. Natural selection also need not result in changes in allele frequencies but instead can stabilize the existence of current allele frequencies. Natural selection, as considered by evolutionary biologists, thus involves a diversity of phenomena, though at its basis it still remains favoring the retention by populations of certain traits – adaptations – over the retention of others.

Darwinian Fitness

The common measure of the impact of natural selection on organisms is Darwinian fitness. The most straightforward way of conceptualizing Darwinian fitness (or, simply, fitness) is that it is a measure of the reproductive output of an organism, that is, reproductive success. Thus, for example (and simplistically), a lifetime reproductive output of two offspring could be considered to bestow a Darwinian fitness of two. This value of two can be described as an organism's absolute fitness. Often one determines fitness not for individual organisms but instead for genotypes or, more commonly, for specific alleles. Obtaining such information for microorganisms can be easier than for larger, especially obligately sexual organisms, because often it is possible to work with isogenic strains, that is, two lineages of microorganisms that differ genetically either solely in terms of one allele or otherwise in a limited, well-defined manner.

While absolute fitness is simple in concept, often it can be difficult to actually apply. This is because there is more to Darwinian fitness than just lifetime reproductive output. For example, generation time can be important as well as the number of reproductive episodes, clutch size, and whether or not (or how many) progeny survive to produce progeny of their own. (Indeed, things can become even more complicated to the extent that different individuals can affect each other's fitness.) Thus, while absolute fitness provides a simple means for understanding just what fitness is, experimentally or even conceptually fitness can be challenging to determine. The alternative to absolute fitness is relative fitness. Relative fitness is the reproductive output of an organism that is determined in comparison to another organism or organisms, usually a conspecific. Here one can simply compare the reproductive output after a certain amount of time, such as one generation, many generations, or some absolute amount of time, where the latter can be useful especially when organisms differ in terms of their generation times. Even here, though, measured fitness can differ depending upon conditions including in terms of when or even how reproductive output is measured, whether strains are competing within the same versus parallel environments, or even as a function of replication history prior to the point of fitness determination.

With microorganisms it is common to determine relative fitness following head-to-head competition between two strains. Thus, for example, two bacteria may be compared in terms of their reproductive output following growth within the same culture, where differences in starting versus ending ratios – or better yet, multi-point, that is, kinetic analysis – can be used to infer differences in fitness between the two competitors. Key, and potentially complicating in such experiments, however, is that the two strains must be relatively easily distinguished. Often this is accomplished by including a marker, ideally something that is selectively neutral, that is associated with one of the two strains, e.g., such as a default, marked strain against which the relative fitness of all other strain or strains is determined.

In experimental evolution studies, one often can determine fitness using the same environmental type as the original selection experiment was performed in. In this way one can determine how fitness has improved, either during or following the completion of the experiment as specifically in response to the selection imposed. This approach doesn't always work, however, especially if selection is sufficiently strong that the fitness of the default or parental strains are effectively reduced to zero within the selective environment. This can be the case, for example, given selection for resistance to some environmental degradant (e.g., acid) or when selecting for modification of host range in symbiotic organisms.

All else held constant, organisms with shorter generation times or higher fecundities (number of progeny produced per capita) are more fit than organisms with longer generation times or lower fecundities. Tradeoffs complicate these conclusions, however. For example, an organism that invests more in progeny may pay a cost in overall fecundity but may reap the benefit of higher progeny survival. So too with generation times, i.e., shorter generation times may be associated with lower fecundities or lower progeny survival. Alternatively, in the course of reducing generation times or increasing fecundities, organisms may become less able to survive until reproductive maturity. There may be, for example, tradeoffs between improving the robustness or capabilities in one life stage and maintaining capabilities or robustness in another, such as tradeoffs in parasites between replication rates and transmission rates.

Death, Stochasticity, and Fixation

Natural selection, as measured by fitness, gives rise to changes in allele frequencies that are biased, i.e., deterministic rather than stochastic (predictable rather than unpredictable). This means that in an infinite population, fitness will be the sole determinant of whether a genotype increases or decreases in frequency. Eventually, in real populations, increases in frequency can have an end point of allele fixation (ignoring, for now, frequency dependent selection). In infinite populations, however, in fact the less-fit alleles would never be lost. This point illustrates the importance of death and sampling in natural selection. That is, the only way that an allele can be lost from a population is if the organisms carrying that allele are killed off or, instead, otherwise never reproduce.

In experimental populations, failure to reproduce can occur because periodic sampling of populations occurs. This periodic sampling, however, is inherently stochastic, that is, who is or is not sampled is subject to chance. Thus, we can view the process driven by natural selection that ultimately results in the fixation of alleles as having three steps during which drift, then selection, then drift dominate. The initial step is the initial formation of a beneficial allele, at which time the allele is susceptible to extinction due to drift that results from its initially low frequency/number. If the allele survives this early extinction, then with larger allele numbers selection can drive increases in frequency with greater certainty. Eventually, however, the alleles against which this beneficial allele is competing will be reduced to such low frequencies/numbers that once again drift will serve as a stronger force than selection, at least in terms of those low-frequency alleles. Thus, the fixation of beneficial alleles, while importantly biased by natural selection, ultimately occurs as a function of drift as well, with drift affecting both initial likelihood of survival and later loss of the allele that is being replaced. Allele loss otherwise would never occur given an overall infinite population size, unless the fitness associated with those now rare alleles somehow was reduced to zero.

Hard versus Soft Selection

Natural selection can be distinguished into so-called hard selection versus soft selection. These can be viewed in terms of impact on population survival, where hard selection has an explicit and negative impact whereas soft selection may not have any impact at all on the population's overall reproductive output. Similarly, hard selection can drive a population to extinction, whereas soft selection directly will have no such effect. These differences can be viewed in terms of absolute versus relative fitness, where hard selection impacts the absolute fitness of the entire population, i.e., its potential to survive and reproduce, whereas soft selection does not. Both soft and hard selection can negatively impact the relative fitness of certain members of the population, and positively impact others, but with soft selection the impact may be seen particularly in terms of relative fitness. In other words, with hard selection either every member of a population may lose, though with potentially some losing more than others, or, at best, those that don't lose may display no or only minor reductions in absolute fitness (rather than gains). By contrast, with soft selection individual population members may display a gain in relative fitness, particularly at the expense of the fitness of other members of the same population, though for the population as a whole there may be no change in absolute fitness. In human terms, this difference is equivalent to a depression (or environmental collapse) that lowers all fortunes (hard selection) versus the typical state of capitalism where some win while others lose (soft selection).

Hard selection may be viewed as an insurmountable (or nearly so) environmental constraint placed on a population. For example, the climate may change to sufficiently colder, wetter, drier, warmer, etc., such that most or all members of a population will not be able to survive or reproduce. Alternatively, this could be attack by a body's immune system against an infecting pathogen, imposing hard selection on the latter, or it could be the loss of a specific host organism (i.e., as equivalent to the loss or degradation of a niche) such that survival is possible only for those individuals that can adapt to using a new host. Certain genotypes, if present within a population, may however be able to survive despite these new challenges, resulting in adaptation of the population to the new challenge. Of course, the challenge need not be so dire that most or all of the population is lost in a single generation. Nonetheless, a population-wide absolute fitness of less than one – i.e., less than that required to replace parents with offspring – indicates a population in decline which, if not addressed in some manner, ultimately will result in extinction.

Soft selection, by contrast, is a consequence of competition between conspecifics. This competition can be direct or indirect. That is, it can involve actually antagonistic interactions versus competitive utilization of resources, respectively, where resource use by one individual results in lack of resource availability to another (exploitative competition). The result of these interactions is the evolution of increased competitive ability. This increased competitive ability does not necessarily improve either survival or reproductive ability in the face of abiotic or interspecific interactions, but does improve fitness in terms of intraspecific interactions.

Classically one can see soft selection in sexual selection, that is, where competition among conspecifics is over access to mates. While crucial to individual reproductive success, sexual selection results in only relative improvements by more successful individuals rather than overall improvements in the fitness of the population . Among microorganisms one can see soft selection, for example, in the production of bacteriocins that are narrowly targeted towards conspecifics. Bacteriocins have the effect of reducing the fitness of sensitive conspecifics, if they happen to be residing within the same microenvironment as bacteriocin producers, since bacteriocins have the effect of killing these sensitive bacteria. Bacteriocins, however, are also costly to produce or release, thereby imposing a fitness cost on producers. Overall, the average fitness of a population declines due to bacteriocin production, whether due to death of sensitive bacteria or costs associated with their production. Producers, however, may still gain a competitive advantage by eliminating sensitive bacteria. By producing such bacteriocins, producers thus can increase their relative fitness while simultaneously display a decreasing in absolute fitness, i.e., soft selection.

Purifying versus Directional Selection

Purifying selection, also known as stabilizing or negative selection is the elimination of low-fitness alleles from populations. Its description as purifying refers to the consequence of this negative selection especially within populations that are already well adapted to their environment. Thus, new mutations in particular, or those hiding as deleterious but recessive alleles, will tend to be lost from lineages, thereby "purifying" the already more prevalent and presumably fitter alleles found in the population. The genetic makeup of the population as a result remains stable, thus purifying selection also being referred to as stabilizing selection.

Purifying selection presumably is the most common form that natural selection takes as it involves the ongoing loss of lower-fitness alleles from populations and is seen even within populations, as found within stable environments, that do not otherwise appear to be evolving. Directional selection, a.k.a., positive selection, is that form of natural selection that is most commonly associated in people's minds with natural selection, however. Here evolutionary change is observed, including in terms of changes in allele frequencies (versus maintenance, via selection, of existing allele frequencies). Such change, however, is most readily observed within the context of changing environmental conditions, since it is particularly in response to environmental change that otherwise non- prevalent alleles within populations will come to display levels of fitness that are in excess of those alleles that were most prevalent prior to this imposition of environmental change.

Note that within the context of microbiology the concept of negative selection is slightly different from its more general use. In all cases, negative selection can be viewed as decreasing the frequency of alleles – or simply eliminating them – that natural selection does not "like" whereas positive selection increases the frequency of those alleles that natural selection does "like". Nonetheless, in microbiology negative selection actually means selection for (versus against) those alleles that in fact display a reduced fitness, a process, which you might imagine, is not as easily achieved as selection for alleles that display a higher fitness, and which typically involves a technique known as replica plating, though other approaches are possible as well. Negative selection thus can be confusing topic to consider especially in terms of microbial evolution unless one keeps in mind that in microbiology its meaning is not quite identical to its meaning in evolutionary biology.

Note also that in a very real sense these two ideas of purifying versus directional or negative versus positive selection are identical, though, representing views from opposite perspectives. Thus, decreasing the relative prevalence of one set of alleles (negative selection) will have the effect of increasing the relative prevalence of another set (positive selection). The difference arises, however, when the selected alleles in fact are already prevalent, in which case emphasis is placed on the loss of less fit alleles (purifying, stabilizing, or negative selection), versus the selected alleles not already prevalent, in which case emphasis is placed instead on the gain of more fit alleles (directional or positive selection). In the above figure note that allele A is being subject to positive selection while allele a is being subject to negative selection. In terms of stabilizing or directional selection, environment 1 and environment 2 can be replaced with allele frequencies prior to the imposition of selection and allele frequencies afterward.