To access the private area of this site, please log in.

Username: 

Password:  

 

Chapter 146

Population Genetics: an Introduction^†

JOHN MAYNARD SMITH

 

WHAT POPULATION GENETICS IS ABOUT

Population geneticists study the genetic variability of natural populations and try to explain this variability in terms of mutation, selection, genetic recombination, migration, and the chance changes that occur because populations are finite. The subject was developed because people wanted to understand evolution: natural selection is effective only if there is heritable variation.

Because the subject was developed to provide a theoretical understanding of evolution and because it involves fairly sophisticated mathematics, there has been an understandable tendency to regard it as of academic interest only and of little relevance to practical people. I think that bacteriologists make a serious mistake if they hold this view. Because of the immense population sizes and short generation times of bacteria, evolutionary changes in bacteria can be very rapid. The evolution of insecticide resistance in many insects is of great practical importance, yet the generation time of insects is long compared to that of bacteria. The comparable event in bacteria has been the spread of resistance to antibiotics. At least some bacteriologists think that we are close to the end of the era in which infectious disease can be controlled by antibiotics. Probably we would have reached this stage sooner or later, however careful we had been, but a use of antibiotics more influenced by evolutionary considerations would have postponed the evil day.

Whenever we want to apply our knowledge of bacteriology, I think that evolutionary considerations will be paramount. If we use bacteria to carry out industrial processes, for example, the disposal of pollutants, we should be aware that the populations we use to do the job will not stay unchanged. If we release genetically engineered organisms, we should know that the genes may not stay in the organisms in which they were released. Thus, far from being of academic interest only, population genetics is the practical side of bacteriology. However, it is only fair to add that my own interest in bacteria arose because I wanted to understand evolution in general and the evolution of sex in particular: bacteria are the only organisms in which things happen fast enough.

 

DO WE NEED A SPECIAL POPULATION GENETICS FOR BACTERIA?

A large body of theory and observation about the population genetics of sexual diploids already exists. Cannot we just apply the theory to bacteria? Indeed, bacteria should be easier to study, because they are haploid, so we do not have to worry about dominance and overdominance and can usually deduce the genotype directly from the phenotype. Unhappily, no such direct transfer is possible. In a sexual population, it is roughly true that we can treat each gene locus as an independent entity. If genetic recombination is common enough, any particular allele, say, A, will associate randomly with alleles at other loci: it will not be true that A is usually found with B but not with b. Technically, A will be in "linkage equilibrium" with alleles at other loci, and it will increase or decrease in fitness depending on its own effect on fitness (averaged over environments and over genotypes in the population). Of course, much effort has gone into calculating the effects of linkage and finite population size, but the simple model, treating each locus as an independent entity, is usually a good guide to what will happen.

No such simple assumption holds for bacteria, in which recombination may be a rare event. Often, papers about bacteria tacitly make the opposite assumption, that recombination is wholly absent. A sample of bacterial strains is analyzed electrophoretically or by some other technique, and the resulting data and some standard package are used to construct a phylogenetic tree. This construction is justified only if the objects being classified have arisen by a process of branching without recombination. This is often near enough the truth, just as the assumption of independent assortment is often near enough the truth for sexual populations. But, as I shall argue in the next section, it is by no means always true.

The population genetics of bacteria is difficult because recombination is neither absent nor very common.

 

HOW CLONAL ARE BACTERIA?

The serious study of bacterial population genetics began with the measurement of electrophoretic variability in Escherichia coli by Milkman (19). At that time, the burning question about isozymes was whether they were selectively neutral or were maintained in populations by selection for functional differences. Milkman hoped to answer this question by studying bacteria. The logic was as follows. Bacterial populations are immense. Therefore, unless there is selection, the amount of electrophoretic variability should likewise be immense. Unfortunately, as Levin (13) pointed out, the argument cannot safely be applied to populations with little or no recombination because of the phenomenon of "recurrent selection" (1). If a favorable mutation occurs in a single individual and spreads to fixation, the resulting population will be homogeneous for the alleles at all loci that happened to be present in the original individual. Even with some recombination, this process of "hitch-hiking" (16) can greatly reduce the amount of variability. Milkman, therefore, was not able to settle the selection-versus-neutral debate, but his work did open up the field of bacterial population genetics.

The first important conclusion from electrophoretic studies was that many bacterial populations are essentially clonal (26, 33): that is, they consist of a set of asexually reproducing lineages between which recombination is rare or absent. The idea was first proposed by Orskov et al. (23), who noted that independent outbreaks of pathogenic bacteria are often caused by apparently identical strains, but it was the electrophoretic data that established the point. In effect, three lines of evidence can be adduced in favor of clonal structure.

i) One piece of evidence is the repeated occurrence of the same multilocus electrophoretic type (ET). For example, suppose that 20 loci are analyzed and that at each locus, the probability that two randomly chosen individuals are indistinguishable is 0.5. Then, if loci assort independently, the probability that two individuals are the same at all loci is 0.5²⁰, or approximately one in a million. Hence, if the same ET occurs repeatedly in a sample, particularly in isolates separated in time and space, that is evidence that loci do not assort independently.

This line of argument must be used with caution for two reasons. First, it is not sufficient that an ET should recur: one must show that it occurs more often than one would expect by chance. For example, in a sample of 227 isolates of Neisseria gonorrhoeae analyzed at nine loci, O’Rourke and Stevens (22) found that the ET carrying the commonest allele at each locus occurred 35 times. This is not evidence of clonality, because the expected number, given random assortment, was 32.2. Obviously, the risk of error is lower if more loci are analyzed. The second possibility is that the population displays an "epidemic" structure (17). That is, extensive recombination is occurring, but one particular ET has recently increased explosively; given time, this ET will be lost by recombination. This appears to be the explanation for the recurrence of ETs in a sample of Neisseria meningitidis analyzed by Caugant et al. (4).

(ii) Stronger evidence for clonal structure can be obtained by measuring linkage disequilibrium between pairs of loci. Essentially, this is just a more general method of looking for nonrandom assortment between alleles at different loci. Whittam et al. (33) found a highly significant disequilibrium in E. coli. Their method is described below, when evidence for recombination is discussed.

(iii) The weakness of the two methods just described is that they cannot demonstrate clonality but only that recombination is too infrequent to generate random assortment. Stronger evidence can come from a third method: the comparison of phylogenetic trees based on different parts of the genome. If a population is strictly clonal, then all genes have the same ancestral history, and trees based on the sequences of different genes should be similar. Nelson et al. (21) found that the phylogenetic tree of 16 Salmonella strains obtained by sequencing the gapA gene was similar to that previously derived from multiple locus electrophoresis. Desjardins et al. (4a) compared phylogenetic trees of E. coli based on three kinds of data (restriction fragment length polymorphism, random amplified polymorphic DNA, and rRNA sequence) and found them to be similar.

I turn now to evidence pointing the other way, showing that recombination has affected the structures of bacterial populations. The most direct evidence comes from comparing the gene sequences of related bacteria. Stoltzfus et al. (29) (for the trp operon), Dubose et al. (6) (for the phoA locus), and Dykhuisen and Green (7) (for the gnd locus) all found that the sequences of different E. coli strains were very similar except in particular regions, where there was substantial nucleotide divergence. This is exactly what would be expected if short pieces of DNA had been inserted from an unknown source. A still clearer picture emerged from the analysis of a gene coding for a penicillin-binding protein in Streptococcus (5) and Neisseria (28) spp. In Neisseria spp. the naturally penicillin-susceptible pathogenic species have evolved resistance by acquiring from several naturally resistant commensal species pieces of DNA on the order of 100 to 1,000 nucleotides long that differ by up to 23% of their nucleotides.

The picture in Streptococcus and Neisseria spp. is particularly clear for two reasons. First, both genera are naturally competent for transformation. Second, the PBP gene has been under strong selection for change but only for the last 50 years, so there has not been time for subsequent mutation to obscure the mosaic structure. Horizontal transfer has also played a role in adaptive evolution in Salmonella spp. and E. coli, although the mechanism has presumably been transduction rather than transformation. For example, Smith et al. (27) found that the sequence of the central, antigen-determining part of the phase I flagellin gene (fliC) differed by 19% of its nucleotides in the closely related species Salmonella typhimurium (official designation, Salmonella enterica serovar Typhimurium) and Salmonella heidelberg but that the sequence in S. typhimurium differs by only 2.2% from that in a much more distantly related serovar in Salmonella subgenus IIIb.

There is no contradiction between this evidence for recombination derived from the mosaic structure of gene sequences and the evidence described above for strong linkage disequilibrium between electrophoretic markers. Recombination in bacteria is usually local, involving at most a few thousand nucleotides. Such events, if they are not too common, will not bring allele frequencies into linkage equilibrium (15).

What, then, can be deduced about the extent of recombination from data, electrophoretic or otherwise, on allele frequencies in natural populations? Whittam et al. (33) analyzed their E. coli data by using the following method, first proposed by Brown et al. (3) in a different context, to test for an association between loci. Suppose that n strains have been characterized at m polymorphic loci. The difference, d_ab, between a pair of strains, a and b, is the number of loci at which they differ. There are n(n – 1)/2 pairs of strains: it is easy to calculate d and V _obs, the mean and variance, respectively, of the n(n –1)/2 values of d. It is also possible to estimate V _exp, the expected value of the variance of d if the alleles at different loci are independent. In fact,

where h_j is the probability that two random individuals are different at the jth locus. By comparing V _obs and V _exp, one can test for an association between loci.

Maynard Smith et al. (17) used the statistic

I_A = V _obs/V _exp – 1

to compare different bacterial species. If loci assort independently, the expected value of I_A is zero. However, if there is an association, the expected value depends on, among other things, the number of loci used in the analysis. The absolute value of I_A therefore cannot be used to measure the degree of association.

Table 1 gives values of I_A for several species. In N. gonorrhoeae isolates collected worldwide over 25 years, alleles at different loci assort independently, whereas Salmonella spp. are highly clonal. The named "species" are now regarded as serovars of a single species, S. enterica (10). The striking fact is that although the reduced genetic distance within serovars indicates that these are groups of closely related strains (but see comments about Salmonella derby below), the genetic structure within serovars is still clonal.

The data on Rhizobium meliloti indicate that this "species" consists of two populations, within each of which recombination is frequent but between which it is rare. Bacillus spp. are competent for transformation: in the laboratory, genes are transferred both within species and, at a reduced rate, between species. The distinction between Bacillus subtilis and Bacillus licheniformis was made biochemically: since the mean genetic distance is almost as great within these "species" as it is in the whole sample, it is likely that the groups so recognized have little reality. The distinction between groups B and D of B. subtilis is based on the electrophoretic data: the I_A values indicate that recombination within these groups is almost frequent enough to generate independent assortment between loci. (See Table 1.)

The value of this method of analysis is that it can indicate whether recombination is rare or absent (Salmonella spp., within and between serovars; Rhizobium and Bacillus spp., between subgroups) or is frequent enough to generate independent assortment of alleles (N. gonorrhoeae; within subgroups of R. meliloti and B. subtilis).

How frequent must horizontal transfer be, relative to mutation, to generate apparent linkage equilibrium? Simulation suggests that if transformation at a locus is 20 times as common as mutation, the population structure will be indistinguishable from a random mating one (J. Maynard Smith, J. Evol. Biol., in press). In E. coli and Salmonella spp., linkage equilibrium is highly significant. Is there, then, any statistical evidence for recombination? Hedrick and Thomson (11) and Whittam and Ake (32) used the distribution of alleles in natural populations to show that recombination is probably occurring, albeit infrequently.

Despite differences in the frequency of recombination, all bacteria interrupt asexual division with occasional horizontal transfer. This has led Milkman and Bridges (20) to propose a "clonal frame" model of the bacterial chromosome (strictly, they were writing about the E. coli chromosome, but their model is of wider applicability; a similar model was suggested by Maynard Smith et al. [15] on the basis of a study of Neisseria and Streptococcus spp.). The idea is that the chromosomal DNA of a set of related bacterial strains consists of a clonal frame that has been unaffected by recombination, that (if it could unambiguously be recognized) could be used to construct a phylogenetic tree of the strains, and that is peppered by short inserted sequences. In Neisseria spp., in which it is easier to recognize insertions, phylogenetic trees of seven named species that were based only on clonal-frame DNA were constructed for two gene loci, PBP and argF. The trees, with one minor exception, were isomorphic: had they been based on total DNA, including inserts, they would be different for the two genes and would depend on which strains of each species were chosen.

 

ARE THERE BACTERIAL ECOTYPES?

One of the few general ideas in ecology is that of "competitive exclusion" or "Gause’s principle." The idea is as follows. If two species are to coexist in the same region, their ecology must be different: more precisely, they must be limited by different factors. If their ecological requirements are identical, it is inevitable that one will be a slightly better competitor and will displace the other. But if one feeds on large seeds and the other on small ones, or if one nests in holes and the other on the ground, they may be able to coexist. Does this idea apply to bacteria?

A peculiar feature of bacterial clones compared to animal and plant species is that their distribution is often worldwide. For example, Whittam et al. (33) wrote that "enteric populations of E. coli in human hosts in widely separated regions of the world exhibit little genetic differentiation." In contrast, Salmonella serovars do sometimes have restricted geographical distributions, but this may be because they are confined to host species with restricted distributions. The existence of several clones cannot therefore be explained by geographical differentiation, as one would explain the existence of several subspecies of a sexual species that replace one another geographically. It would therefore seem to require either that the clones are selectively equivalent and coexist only because they are equivalent or, more plausibly, that they are indeed adapted to different ecological niches, as are sexual species that inhabit the same region.

Since the term "species" is of uncertain meaning when applied to bacteria, I will use the term "ecotype," meaning a group of genotypes adapted to a particular niche. The importance of ecotypes for population structure is as follows (14, 24). If, in E. coli, for example, there were no ecotypes and no recombination, then the spread of a single favorable mutation to fixation would make all E. coli genetically identical. But suppose there are ecotypes

A, B, C, . . . . A favorable mutation occurs in an individual of type A. It will spread by selection to all members of type A, rendering that type genetically homogeneous, but type A, even with the favorable mutation, will not replace types B, C, . . . because it lacks the necessary ecological adaptations. The further spread of the favorable mutation requires horizontal transfer into other ecotypes. The final result, after horizontal transfer, will be that the mutation will spread to all E. coli but will not make the species genetically homogeneous, as would be the case if there were no ecotypes. This scenario is of course oversimplified, but the qualitative effect of ecotypes in preserving genetic variability is real.

Ecotypic structure, then, would be important in maintaining genetic variability. But does such a structure exist? In Salmonella spp., the evidence that different serovars are adapted to different hosts is clear. For example, Salmonella typhi is confined to humans. Usually, we know of Salmonella serovars specific to animals other than humans because they cause disease. For example, 10 of 60 infants with salmonellosis in Puerto Rico had been exposed to a pet turtle within 2 weeks of the onset of illness. It was later found that all turtles in pet shops in Puerto Rico were positive for Salmonella spp., with 89% yielding Salmonella pomona (30). The effect is not always so extreme. The serovar S. derby has been divided into two rather distantly related groups of clones, I and II (what this means, presumably, is that there has been horizontal transfer of the antigen-determining gene). Type I is found commonly in birds and rather rarely in mammals; type II is rare in birds and common in mammals (2). Ecological adaptation may also be to different sites within a host rather than to a particular kind of host, although evidence for this is harder to come by.

Ecotypic structure, particularly by host specialization, is less obvious in E. coli. Whittam (31) found a statistical difference between the clones present in chickens and those present in the surrounding litter and soil. Almost certainly, this difference reflects a difference between clones in the ability to survive and reproduce in the two environments. Presumably, however, all clones cycle between living in a host and in the external environment. It is therefore unclear what "niches" the different clones are adapted to.

The evidence for adaptation to particular hosts is stronger for Salmonella spp. than for E. coli. I am left with the following question. Is the difference only apparent, arising because we have more data about Salmonella spp. in nature, essentially because Salmonella spp. cause disease? Or is the difference a real one, arising because Salmonella spp., being pathogens, invoke stronger immunological responses from their hosts and have therefore been forced to differentiate into host races?

 

ARE THERE BACTERIAL SPECIES?

Bacteriologists have adopted the Linnean practice of grouping their strains into species, which are given latinized binomial names. Are these species real genetic entities? Is there a better way of classifying and naming bacteria?

Before I try to answer these questions, it is helpful to describe the use made of the species concept in higher organisms. One cannot decide on a sensible method of naming and classifying unless one has some idea of what one wants to do with the system. It is convenient to start with the species concept as it applies to birds: ornithologists (particularly Mayr [18]) played an important part in developing our present species concept, mainly because birds are large, conspicuous, and diurnal and recognize one another by the same senses (sight and hearing) that we use to recognize them. Let our discussion start with empirical considerations and then move to more theoretical ones.

(i) A competent bird-watcher expects to place every bird seen into a species. This determination of species is based partly on plumage and song, partly on habitat, and partly on geographical location. That is, members of a species share a common phenotype, a common ecology, and a defined geographical range.

(ii) The characteristics of a species are thought to depend on sexual reproduction: members of a species exchange genes with one another but only very rarely with members of other species.

(iii) A species has a common gene pool. One consequence of this gene pool is that a favorable mutation can spread to fixation in a species but will be confined to that species. Because of frequent recombination, the fixation of a favorable mutation does not substantially reduce genetic variability.

(iv) Species are monophyletic: that is, they include all the descendants and only the descendants of some common ancestral population. The same is true of higher taxonomic categories, i.e., genus, family, order, and so on.

(v) It is thought that the splitting of a single species into two usually requires a period of geographical isolation, during which sufficient genetic difference can accumulate to ensure reproductive isolation when the two populations meet again.

These are not universal truths, even for birds. Perhaps the most important qualification is that if one looks at similar birds from different geographical regions, they usually differ, and there is no objective way of deciding whether they should be placed in different species or merely in different subspecies of the same species. Things can be much worse in other taxonomic groups. In flowering plants, for example, even the first point does not hold. Often one cannot confidently ascribe a specimen to a particular species, partly because hybrids between species are not uncommon and partly because there are many asexually reproducing plants. But we do at least try to give specific names only to groups of which the above statements are approximately true.

I do not think there is any way of classifying bacteria into species that would carry such a heavy load of assumptions. One difference is easily stated. As discussed earlier, most bacteria have a worldwide distribution: we can forget about point v.

More important, I do not think we can identify anything corresponding to a common evolving gene pool. In Neisseria spp., the evolving unit is wider than the named species: the evolution of penicillin resistance has involved gene transfer across the whole genus. In contrast, in Salmonella spp. the evolving unit is the clone, complicated by the fact that occasionally, as in the case of fliC, a gene is transferred between rather distant relatives. No common system of naming can ensure that a given term has the same meaning and implications for all bacteria.

All the same, there are discontinuities in the pattern of variation. There are no intermediates between E. coli and Salmonella spp. (although some species that have been placed in the genus Escherichia are more distant from E. coli than the latter is from Salmonella spp.) and no evidence of gene exchange between them. To take two much more closely related groups, N. meningitidis and N. gonorrhoeae are different populations, even though they exchange genes. At one time, I suspected that the only reason the two were given different names was that they cause different symptoms in humans, but it seems that I was wrong: isolates of N. meningitidis resemble one another genetically more closely than they do isolates of N. gonorrhoeae and vice versa.

Two proposals to regularize the naming of bacterial species have been made. One is that strains should be placed in different species, genera, etc., according to the degree of DNA difference between them. I think this proposal has little merit. The purpose of naming things is to name a set of entities that have something in common, not to draw an arbitrary division through what may be a continuous variable. The rate of genetic exchange varies more or less continuously with genetic distance (25). The difficulty becomes apparent when one sees that it could lead us to place A and B in the same species and B and C in the same species but A and C in different species.

Dykhuisen and Green (8) suggest that a set of strains should be placed in a single species if the phylogenetic trees of the strains are different for different genes (because of horizontal transfer between strains) but should be placed in different species if the phylogenetic trees for different genes are the same (as will be the case in the absence of horizontal transfer). The proposal has the merit of corresponding to the "biological species concept" in higher organisms (in effect, a single species has a single gene pool) but has the disadvantage that the species category would have very different meanings in different bacterial taxa. In Salmonella spp., each clone would be a species; in Neisseria spp., the whole genus would be a species.

I suspect that species will continue to be named according to their phenotypes: there is little point in naming something unless it has characteristics of interest to us. However, in naming bacteria, some criteria are worth bearing in mind.

(i) Named taxa (species, genera, etc.) should be monophyletic. We should not name polyphyletic "species" like Shigella (which consists of two or more unrelated lineages that are phylogenetically within E. coli [26]) or S. derby, discussed above.

(ii) Named species in different bacterial taxa will have very different amounts of genetic recombination within and between them.

(iii) There is a lot to be said for using names that lie outside the Linnean system, particularly for groups that are not monophyletic. Bridled guillemots are recognizable variants of the species Uria aalge. At a higher level, fish, reptiles, and protists are none of them monophyletic groups, but biologists talk about them all the time. There would be no harm in giving a name to those E. coli that do not ferment lactose and cause invasive disease: just don’t write the name in italics.

^† A similar treatment of this subject is to appear in the Fifty-Second Symposium of the Society for Microbiology, Cambridge University Press, Cambridge, United Kingdom, 1995.