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

Username: 

Password:  

 

Chapter 103

Similarities and Differences of Individual Bacteria within a Clone

ARTHUR L. KOCH

 

INTRODUCTION

Seemingly, two sister bacteria are biological entities as closely identical as can be conceived. In this chapter, their differences are scrutinized. The examination will extend to the differences between subcultures and subclones of Escherichia coli. Important concepts have also come from the study of other eubacteria. These differences presumably have a biological basis and can be subdivided under the major divisions listed in Table 1 of strategies to cope with environmental challenges. A few of the variations are due to cell systematic changes during the cell cycle and to idiosyncratic mutational events; the rest concern features that have evolved to allow the bacteria to cope with unpredictable variations in their environment. Importantly, these include mechanisms to generate nongenetic differences between sister cells to ensure that one or the other survives potential catastrophes and preserves the strain. Key among them is an ability to enter into a Dauer modification state, referred to by Dow et al. (15) as the shutdown state.

The global life strategy of a bacterium is characterized by its expression of a diversity which allows it to invade those niches and habitats obligate for its lifestyle. Its strategy can achieve the variability necessary in both the short and long terms to respond to a range of rare catastrophes, engage in genetic exchange to a controlled degree with other kinds of organisms, balance mutagenic forces with repair systems so that accumulation of mutations is consistent with both remaining the same and changing innovatively, and possibly enable invasion of habitats that are not necessary for the cell’s cycle of growth and continued propagation (such as opportunistic infections of a compromised host).

First a catalog of the types of variability found in nature is given. Then some theoretical ideas for the evolutionary processes in prokaryotes are presented. Finally, the cell cycle is analyzed to understand some of the variability in comparing two cells taken at random from an asynchronous culture.

 

CLONAL HETEROGENEITY

 

Presence of Duplicated Regions of the Chromosome

Although most of the bacterial chromosome is single-copy DNA, there are some duplicate genes even in the normal wild-type genome. The most notable cases are the occurrence of seven copies of the DNA coding for rRNA and for other components of the protein synthesis machinery in the standardized E. coli. These homologous regions permit unequal crossing over between sister chromosomes, leading to an equal number of deficiencies and duplications of the regions between ribosomal genes (1). Both of these recombinant cells are eventually lost, except under very selective conditions. For example, the genotypes with duplications can survive and experience selection, causing further recombination and multiplication of duplicated genes (35). Usually a growing population has large duplications (two or more copies) in about 1% of the cells (J. R. Roth, personal communication). These cells can have properties quite different from those of cells with the nonduplicated genome. Note that the cells with an abnormal composition usually grow more slowly and can be an additional source of the slowly growing component of the population that generates a skewed age-at-division distribution.

 

Inversion Switches

Phase variations in Salmonella typhimurium (official designation, Salmonella enterica serovar Typhimurium) and phage Mu are well-studied cases of special genetic regions that are selectively inverted to yield other gene products and different mutant phenotypes. A similar system exists in bacteriophage P1. These systems are related to each other and to transposon TnA (66). In these cases, a promoter initiates RNA transcription in an inverted or not normally oriented genetic structure, thus making different products. The value of such systems for long-term survival of the organism is in providing a mechanism to reversibly switch to an alternative phenotype, not simply convert by an all-or-none change to a different state irreversibly. Depending on the human host of the bacterium or the bacterial host for the phage, different states are positively selected. However, the alternate state appears with a high probability (10^–3 to 10^–5 per generation). Such systems need not have the same rate of inversion in the two directions, because the gene catalyzing the inversion is under the control of different promoters in the two states. Clearly, it would not benefit an organism to have inappropriate switching at inappropriate rates, as would happen if two such systems were to function within the cells of a bacterial culture (cross talk?). However, the number of such switching systems is probably quite limited because of the close genetic relationship among the systems that have been studied and because the protein catalyzing inversion in some cases can act on other systems. Again, this circumstance could explain a positive skewness of the age-at-division distribution if some proportion of the cells in the population were in a form that grew more slowly.

 

Radiation-Resistant Fraction of Populations

Early in the history of radiation biology, Luria and Laterjet (59) discovered that although the survival curves of microorganisms grown under different treatment conditions might have quite differently shaped shoulders on the standard semilogarithmic plots, in all cases the survivorship at high doses was unexpectedly high. At moderate doses, the curves exhibited a first-order region, but the straight line did not extend to very low survival values. Instead, at high doses the line curved, and survivorship was essentially independent of dose. Thus, a resistant fraction of cells remained at values between 10^–2 and 10^–4 of the initial number of viable cells. This response has been seen with UV light, X rays, and even tritium suicide (61). In all these cases, it appears that a fraction of the population is in a resistant and temporarily nongrowing state. This state of affairs makes good evolutionary and ecological sense: a certain proportion of the bacteria that are formed are somehow destined to be quiescent. In this state, they do not grow and are resistant to many environmental stresses. Eventually they do grow and produce normal cultures and colonies. It is as if the culture of gram-negative organisms allots a small fraction of its productivity during balanced growth to form the equivalent of the spore stage of gram-positive organisms. This process has been little studied, no doubt because it is so rare. But in many ways it is a strategy superior to the complete conversion of a culture upon starvation into endospores: it is an a priori calculated diversification to form products that are a hedge against radiation damage as well as other difficulties that might arise. This phenomenon is yet a third possibility to explain the positive skewness of the age-at-division distribution.

 

Phage Yield and Cell Size Variation

The first reference (of which I am aware) to the importance of biological variability was a report by Delbrück (13) in 1945. He assessed the variability of the burst size of viruses from one-step growth curves and found it to be much larger than the variability of cell lengths in the population. This finding implies that either the virus replication is highly variable or the bacteria are highly variable in characteristics other than length.

 

Nutritional Generation of Cellular Heterogeneity

The kinetics of lactose metabolism was quite confusing until it was recognized that one needed to distinguish between enzyme formation averaged over the entire culture and that within individual cells. Benzer (5) realized this and set out to measure the variability between bacteria within a culture that had been shifted to lactose as a carbon and energy source. He stopped enzyme induction by phage infection. Under this circumstance, only those bacteria that had on hand the enzymes necessary for lactose metabolism could support the growth of the virus and later lyse, liberating infectious particles and β-galactosidase. With this system as an assay, Benzer was able to show that after a culture of bacteria has been exposed to lactose for a few minutes (when the level of β-galactosidase is still low), the enzyme levels of individual cells are very different from each other. Only a few cells had become sufficiently induced to be able to use lactose as a carbon and energy source and thereby support normal phage reproduction. The bulk of the original population, which remained uninduced, supported a limited growth of phage, and that only after a delay. Benzer showed that after a shift to lactose, the heterogeneity of the population first increased and then decreased. When a poorly used substrate, methyl-β-galactopyranoside, was substituted for lactose, the response was uniform among the cells throughout the induction process.

How were the bacteria that were more easily induced different from the rest? With the knowledge accumulated since 1953, many of the factors involved can now be specified (3, 4, 73). It is most probable that in this small, easily induced portion of the population, an uninduced (random) transcription event of the lac operon had recently occurred. Although such untriggered mRNA syntheses are rare under noninducing conditions, each such event results in a number of translation products. Consequently, the background level of β-galactosidase and permease in these few cells is high, and thus the enzyme level in the population is quite heterogeneous. Lactose is accumulated much more rapidly in the few cells with higher than average levels of permease, and there the level of β-galactosidase is sufficiently high that enough transglycosylation events take place to form the actual inducers, allolactose (6-o-β-d-galactopyranosyl-d-glucose) (21) and glycerol-β-d-galactopyranoside (A. W. Hsie, personal communication). This starts the avalanche of events leading to increased transport, transglycosylation, induction, and production of galactose and glucose as carbon and energy sources. The process is autocatalytic in that it creates the cellular growth response, leading to an increase of the induced group of cells and further heterogeneity in the culture. However, after the lactose-metabolizing cells become the dominant part of the population, cell-to-cell variability decreases.

On this basis, the major source of variability comes from the chance event of the release of the bound repressor from the operator DNA of the lac operon. These kinetics have been well studied, both theoretically (30, 60, 73) and experimentally (2). The fact that there are few molecules of tetrameric repressor per cell is sufficient to explain the situation that creates the heterogeneity in β-galactosidase activity in partially induced populations.

 

Quasi-Genetic Variation of Growth Rate

In an early but well-designed experiment, Hughes (18) incubated a single cell for 3 h to obtain a clone. He then spread a portion of the clone on nutrient agar and after 3 h selected both a small and a large microcolony. From these second and third clones, he similarly selected a fourth and fifth, again on the basis of colony size. The diameter distributions obtained from 100 microcolonies derived from a further subcloning of each of these five colonies showed large variability within each subclone, while their mean sizes were significantly different, demonstrating short-term "inheritance" of colony size. It might not be surprising that one can select slowly growing variants. The fact that it is as easy to select faster-growing variants is possibly more surprising. However, the key point is that this inheritance is not permanent. But the degree of variation present in the cells of a microcolony that has had only 3 h at 37°C under nominally anaerobic conditions to grow from a single cell is quite surprising. In this time, one cell could have grown to at most 500 cells, and many of these were shown to have growth properties markedly, but temporarily, different from those of the progenitor cell. Axelrod et al. have used essentially the same technique to show that mouse cell lines exhibit this same phenomenon (2).

 

All-or-None Formation of Induced Enzymes

With the development of the gratuitous inducer thiomethyl-β-d-galactopyranoside (TMG) and the technique of chemostat culture, a kinetic study of conversion of cells to the induced state could be made. Novick and Weiner (60) designed a medium containing a level of TMG that failed to induce uninduced cells but maintained the state of induction of previously induced cells that had permease and could accumulate the TMG. Consequently, these authors were able to measure the rate at which cells first become induced. Note that their technique gives quite different information than does a simple estimate of the general level of lac operon products. They showed that the transformation to the induced state was a first-order process in the bacterial concentration but a high-order process (the 8.4 power of TMG concentration) for the inducer concentration. This means that while the pseudo-first-order rate constant for recruitment to enzyme production rises rapidly as the external TMG concentration is changed, from the cell’s point of view the process is a stochastic one like the decay of a radioactive atom; i.e., it depends on the rare chance of the binding of the inducer to the repressor, changing the latter’s conformation, and leading to an opening of the operator site. Once that happens, induction becomes autocatalytic until that cell and its descendants are fully induced.

 

Quiescent Cells in Low-Dilution-Rate Chemostats

Our test for this phenomenon (50) depended on two different assays for β-galactosidase. One was the usual assay in which the bacteria are lysed with sodium dodecyl sulfate; the other was the hydrolysis of o-nitro-β-d-galactopyranoside by dense suspensions of permease-negative intact bacterial cells.

If the induction response is heterogeneous, then the population will consist of a few cells that contain most of the β-galactosidase and a majority of cells that contain very little. Although all of the enzyme molecules will function at a high rate in the lysis assay because the o-nitro-β-d-galactopyranoside is available at nearly saturating concentrations, they will not function in the whole-cell assay because the reaction rate of the enzyme within the cell will be limited by the passive diffusion of substrate through the membrane of these (permease-negative) organisms. On the other hand, if after a short time from the inducer addition (allowing for the penetration of the inducer, the start of transcription, and the completion of the first translation products) the enzyme molecules are uniformly distributed among all cells, then the hydrolysis rate of the whole cells will be maximal because the permeation of o-nitro-β-d-galactopyranoside through a combined total surface area of all the cells is much greater than it would be through the surfaces of a smaller subset of these cells.

With this test, it was shown that all cells in chemostat populations growing with doubling times of up to 13 h had a uniform and rapid response. On the other hand, cultures growing more slowly were heterogeneous. The kinetic results were interpreted as follows. In low-dilution-rate cultures, each cell varies in its instantaneous rate of protein synthesis. A particular cell may be temporarily quiescent (possibly accumulating reserves) but will at some time self-activate and synthesize proteins at the normal rate. In fact, it was found that the same step time was characteristic of both cells from a faster-growing chemostat culture and cells growing in unlimited medium, otherwise of the same composition. For a 24-h chemostat culture, only about one-third of the cells are inactive at any instance of time, but all cells are active over a 3-h period. This type of heterogeneity could not be due to inadequacy of mixing in the chemostat culture. It is probably controlled by minor fluctuations in the energy resources available to individual organisms and the time since the last activation of protein synthesis (also see the discussions in references 22, 23, and 31).

 

Individuality of Cellular Chemotaxis

It is now well understood that bacteria collectively and individually swim in an ecologically favorable direction by the following process. They proceed in a more or less constant direction for a time (69); however, sooner or later they change direction, more or less at random. The frequency of the "tumbles" depends on the chemoreception process and is usually responsive to whether the conditions are satisfactory for the cell’s needs. The operant philosophy of the motility system apparently is, if things are satisfactory, don’t rock the boat (i.e., don’t change course).

The tumbling response is controlled by the level of methylation of key proteins and is an all-or-none response which is caused by reversing the direction of rotation of the cell’s flagella (69). Spudich and Koshland (68) observed that upon a shift to favorable conditions, i.e., the presence of 0.5 mM serine as an attractant, individual cells of a tumbling mutant strain exhibited a lag period before a tumbling event. Eventually the cells engaged in a few tumbling events and then tumbled continuously. The key point of the Spudich and Koshland study was that individual cells had significantly different lag times and that the lag time did not vary systematically with cell size; the lag appeared to be a characteristic of an individual cell. The same individuality was shown with wild-type cells, although the assay is more difficult. In repeated trials, particular cells maintained a quite constant time until tumbling.

Mechanistically the cells respond to the stimulus by activating the methylating (CheR) and demethylating (CheB) enzymes that function until the protein substrate occupies enough of the critical sites to control a sufficient number of flagella. If the rate constant of the process is k and if there are r critical sites per cell that must be altered out of a total of n, the fraction of cells that have not yet tumbled is described by the binomial distribution, with p = e ^–kt and q = 1 – e ^–kt. This is basically the model that Rahn (64) used to describe bacterial division kinetics. The binomial expression would describe the case if k, r, and n were exactly the same for every cell in the population. In fact, the number of functional flagella does vary, and thus r and n may vary from cell to cell. It could also be the case that the number of enzymes affecting the state of methylation is small and varies stochastically from cell to cell.

 

Variability of the Growth Rate in Broth

With a computer-linked turbidimetric system constructed to monitor bacterial growth accurately and continuously, Wang and Koch (74) observed a phenomenon which probably could not be detected in any other way. The experimental measuring system was capable of measuring the growth rate to better than 1% in a 3-min period. The conclusion of their paper is important because without knowledge of this phenomenon, false conclusions would be made about what might appear to reflect bacterial individuality. It was found that the growth rate of wild-type bacteria on ordinary Difco nutrient broth exhibited a large and systematic, but temporary, slowing when a particular bacterial density was reached. It was shown that the cultures had depleted something in the growth medium and therefore became "shifted down." The cells, however, quickly readjusted physiologically so that the growth rate increased to nearly the original value. The implication is that if samples are removed from a culture which is apparently in balanced growth and growing in rich medium, they could have radically different instantaneous properties if taken at slightly different time points.

 

FORMULATION OF THE NEED FOR VARIABILITY OF MANY DIFFERENT KINDS

 

The First Cell Had No Versatility and No Backup Mechanisms

The earliest living organism must have had a repertoire of catalytic abilities that was limited but sufficient to enable it to survive and multiply (38, 45, 54). The tool chest of the earliest cells capable of adaptive evolution had to include three processes: first, a mechanism to duplicate information-bearing substances, including, possibly, clays, mineral surfaces, inorganic and organic membranes, proteinaceous polymers, RNA, and DNA; second, a mechanism to transduce energy into a form that it could exploit from an exergonic chemical reaction using reactants present in the environment; and third, a mechanism to foster cell division. This earliest cell may have depended on the environment for other needs (38).

Thus, many required items had to be present in the environment at the right level for the cells to survive. Such a primitive cell, however, could engage in Darwinian evolution, thus making it possible for mutants advanced in these three properties to supplant the prior population. Qualitatively new processes could also have arisen, and thus the cells became more versatile and were less rigidly dependent on the environment. These improvements would have led to an expansion of world biomass and a wider range of habitat.

 

Elaboration, Refinement, and Development of Survival Mechanisms

The vast bulk of the thousands of steps involved in the cell processes common to all cells, no matter to which kingdom or domain they belong, apparently evolved before stable diversity developed (Fig. 1). The logical explanation of this unity is that evolution was essentially monophyletic throughout this period, and although there were many offshoots at any one time, all but one evolutionary branch died out. Thus, the world biosphere was essentially a monoculture up to the time of the last universal ancestor. The initial split into three kingdoms (or domains) must have occurred because the founder organisms used quite different, noncompeting strategies (45). This chapter focuses on only certain enteric organisms, but they have hidden in their genomes processes inherited from the earlier monophyletic time. Significant here are the systems to deal with challenges, the last item in the center section of Fig. 1.

Although the generation of stable diversity occurred late compared with the development of a highly developed cell physiology, many mechanisms to cope with environmental fluctuations developed much earlier. The heat shock system, for example, is ubiquitous, and the genes are homologous across all phyla. Conversely, resistance to environmental fluctuations in water activity developed after stable diversity was well established. For this adaptation, some organisms perfected sporulation, some developed mechanisms to generate or sequester osmoprotectants, etc. The sporadic and unique occurrence of these protection mechanisms suggests that the dependence of life on water is so essential that ways to overcome its lack were very difficult to devise and developed late in specialist organisms. Similarly, ways to survive and grow at a range of temperatures, salinities, and pH values had only feeble starts before diversity became well established, but then specialists in narrower, more extreme, environments did arise. It has been argued that high-temperature forms, particularly of the domain Archaea, came first (75, 76), but one would have thought that such an origin would have left its mark on many proteins from organisms now growing at more moderate temperatures.

Enteric organisms, whose biomass is thought to be largely restricted to the colon and feces of animals, have developed an astounding ability to adapt and to cope with the necessary fluctuations during the host cycle (40). I will use the term "host cycle" to include the circular process by which coliforms move from one gastrointestinal tract to the external environment and then establish themselves in another colon. The term "host cycle," of course, includes the "culture cycle," i.e., the process set in motion when a stationary culture is diluted into fresh rich medium and grows again into a stationary culture. That term, in turn, includes the "cell cycle," i.e., the process set in motion when a cell divides and each newborn cell grows and divides again into two newborn cells. In addition, we need something to include longer-term changes, like response to an ice age or to a time when certain antibiotics are temporarily abundant in the ecosystems. Thus, layers of regulatory systems must be coded for.

Consequently, to think about the variability of a modern prokaryotic species, including its inheritance from the past and the strategies that it uses to survive today, we must consider the time to express that variability (44). Some cellular controls must function within a small fraction of a second, while others depend on reactivating cellular processes evolved many millions of years ago that are needed only occasionally. To consider them, I will first discuss some thought experiments that, if carried out, would eliminate such mechanisms, and I will then examine cases of living organisms that have lost certain of these mechanisms.

 

TWO THOUGHT EXPERIMENTS

 

The Glucose-Limited "Perennial" Chemostat

Imagine that a single cell of a modern bacterial species with no plasmids or cryptic viruses (i.e., truly gnotobiotic) was inoculated into a constant, continuously mixed environment and cultured for a very long time (say, billions of years). Specifically, imagine a "perennial" chemostat culture (39, 41, 47, 48) with one starting organism growing continuously limited by one nutrient—say, E. coli with a low level of glucose in a minimal salts medium. (Imagine that unspecified means are then used to prevent adherence of the organism to the walls of the vessel.) In this perennial austerity, the culture would remain as a monoculture because there is only a single nonsubstitutable resource available. Therefore, in this truly extreme case, the competitive exclusion principle of Gause will ensure that any favorable mutation will displace the parental kind. Thus, the system would remain as a monoculture, although there would be population turnovers and temporarily there would be transients with mixed genotypes present. Under such conditions, diversity could not develop. Moreover, there would be no development of symbiosis or antibiosis. Plasmidlike entities might develop as permanent parts of the cell, but development of the ability to transmit host genes to other cells would not serve a useful purpose. This is because transmission of genes between cells would not be selected for because of the occurrence of population turnovers which would regenerate a homogeneous population and by periodic selection (see reference 32) eliminate diversity, and thus the movement of genes from cell to cell would not increase fitness.

Thus, long-term selection would produce cells that have achieved an effective and efficient glucose transport system so that in the final state, neither more units of phosphotransferase in the membrane nor a higher specific activity (or a qualitatively different mechanism) would improve their ability to scavenge glucose from the environment. The steady-state concentration of free glucose in the medium would be extremely small. I chose this example because the extant coliforms have almost reached this state. Actually, modern E. coli is not far from the limiting state in which every glucose molecule that diffuses to it is consumed (27, 34, 56). Even so, the most rate-limiting step in the uptake and metabolism of glucose is the diffusion through the porins in the outer membrane during the steady state of uptake (48, 56), and thus the outer membrane would be discarded during our continuous culture. As a second example, modern enteric organisms change size depending on the growth rate, which in turn depends on the nutrients in the culture medium; the cells are smaller in poor medium. To increase its surface-to-volume ratio, under extended chemostat growth, the cells should evolve to be as small as possible under constraints fixed by other cellular processes. Also, the cells should become as elongated and narrow as mechanistically possible to increase further their surface-to-volume ratio. Furthermore, long-term evolution in the chemostat will favor a mechanism whereby the sister cells will not adhere to each other but rapidly separate after cell division. If a cell separates itself from others, it will maximize its rate of glucose uptake. Motility and chemotaxis in principle can increase the rate at which glucose impinges on the cell, but mathematical analysis showed (39) that if the cells remain well separated, motility of the isolated cell would not help the bacterium scavenge a low-molecular-weight substrate in its environment because diffusion of the nutrient is rapid compared with the speed of chemotaxis. The modern organism has adapted to the fluctuations in its environment at considerable cost compared with the fixed needs in the hypothetical chemostat.

 

The Perpetual Turbidostat

Now imagine that the same experiment had been carried out in a perfect device that continuously measured the turbidity (or biomass in some other way) of a well-stirred culture, added growth medium of constant composition in response to cell growth, and removed an equivalent volume of culture simultaneously to maintain the cell density. The same starting organism during the long evolutionary time would learn how to grow faster and faster in that given medium. In a rich medium under favorable conditions, it would grow very rapidly. In the language of the ecologist, the evolved strain would be a pure r-strategist; i.e., all of its systems would become adapted and dedicated to growing as fast as possible. Basically, this strategy entails austerity in order to survive with smaller amounts of protein per new genome and to increase efficiency in order to streamline the process of protein synthesis so that it is faster per ribosome (32, 39, 42, 47). Consequently, the "doodads" that are inherently present in the ribosome to protect it against antibiotic agents that are no longer in the environment of the organism could be eliminated with significant simplification. With respect to the function of protein synthesis itself, there is a balance point between simplifying or retaining the changes built in during the organism’s history even though they could protect against a rechallenge that in the chemostat would never happen.

 

The Evolutionary Product of Long-Term Continuous Culture

Genes That Would Be Deleted.

Now let us focus on the cellular changes that would develop independent of the details of how the constant continuous culture was carried out. Our hypothetical organism would lose many chromosomal regions present in modern organism; these are listed in Table 2. Certain systems would be refined to a sophisticated degree; these are listed in Table 3. Changes that could possibly arise in the regulation of protein synthesis are grouped together in Table 4.

The genome would become much smaller, largely because many open reading frames can be omitted (Table 1). Furthermore, in an unchanging environment, the organism does not need to possess alternative metabolic pathways and could eliminate currently utilizable inducible catabolic enzymes and systems, such as the lac operon, and cryptic, inactivated genes, such as the evolved β-galactosidase gene. It would need no genes to protect itself from temporary nutritional starvation or overfeeding, from lowered or elevated temperature, or from dehydration. Host cycle genes now needed for survival during the necessary transit from one gastrointestinal tract to the next could also be eliminated. Selection for smaller genomes would allow shorter genome replication times. This may appear as a minor issue because the amount of a cell’s resources devoted to chromosomal synthesis is small. However, any saving in time would be positively selected. Those mechanisms maintained and refined as listed in Table 3 would not need much more coding capacity. The simplification of the control of translation, itemized in Table 4, would also lead to a decrease in the size of the genome. Conservatively, we can estimate that the genome could be less than 1/10 the size of that in the modern "worldly" and "cautious" organisms. It would have a genome considerably smaller than that of the most streamlined mycoplasma found thus far (47).

Processes That Would Be Refined during Long-Term Continuous Culture.

The cardinal aspect of the organisms generated by these selection processes is that they would be essentially perfectly adapted to their environment. The bacteria would be stultified; they would carry out only those processes constantly needed but would do so very efficiently. They therefore would need to have amplified and refined conservative mechanisms preventing genetic change. Because of the way our hypothetical continuous culture system was set up, the usual genetic mechanisms for transfer of genes from organism to organism were eliminated from the start by commencing with a true gnotobiotic system containing only the host chromosome and no plasmids or activatable lysogenic viruses. Once evolution had reached the point that physical and chemical conditions in the environment had become rate limiting in all essential aspects, then further genetic improvements could no longer be selected and any mutations would create only deleterious genotypes. Consequently, the cells would evolve so that mutational events would be prevented as far as possible and the effects of unavoidable mutations would be eliminated, again as much as physically and chemically possible. The development of mechanisms for more accurate repair and prevention of mutations would be a very slow process because (i) improvement in the repair systems is not directly selected and (ii) as the accuracy of repair improves, the mutation rate would become smaller and change in the repair system, as well any other system, would be slow. Changes in the DNA would continue to appear as a result of ionizing or nonionizing electromagnetic radiation unless the long-term cultures were carried out deep in the earth in the dark. There would also be biochemical mistakes. Radioactive decay of naturally occurring potassium would cause unavoidable mutations. Mutations so produced could be minimized only by repair of the genetic damage or outgrowth of the defective cells.

Streamlining the Control of Protein Synthesis.

Protein synthesis is subject to a very large number of regulations and controls that make the kinds of proteins being formed attuned to the external and internal environments. The continuous-culture organism would need no adjustable regulation and a much simpler physiology would suffice. Table 3 lists some possibilities. It may be that an encyclopedic knowledge of the entire text of this volume would allow a better judgment of which regulatory mechanisms would survive the winnowing of these varieties of control mechanisms.

 

TRUE OLIGOTROPHS

 

Description

Mycoplasmas and VBNC (viable but not culturable) organisms may be almost the equivalent of the organisms that would arise after long-term continuous culture. A particularly important study of marine ultramicroorganisms (oligobacteria) has been carried out by Button’s group (9). These organisms appear to die when grown in rich media containing even as little as 5 mg of amino acids per liter. Because biomasses are very small, special circumstances are needed to show that they are alive. This was done by diluting a sample into sterilized seawater and observing that after a long time at a low temperature, the number of particles possessing double-stranded nucleic acid (measured by a flow cytometer) increased to the steady-state level of the original sample. With this technique, pure strains could be isolated. One of these, oligobacterium RB1, had 1,120 kb of DNA per genome, compared with the 4,700 kb of DNA in the E. coli genome. It would appear that the ecosystem that is the highly oligotrophic oceans is a very constant environment with a heterotrophic biota that has adapted by producing a very small genome size and has retained almost no ability to cope with a surfeit of amino acids. Clearly this is an extreme case of "substrate-activated death" (62). Mycoplasmas, particularly phytoplasmas, can have a genome size of only 450 kb, or 10% of that of E. coli. They live intracellularly in plant tissues and must be able to survive both in an insect vector and in plant tissue. Without this need and if plant growth were continuous and phytoplasma dispersal were effective, we can imagine that the genome would be smaller still.

 

Effects of the SOS System and Other Mechanisms Protecting the Genome

One of the number of genetic regulatory systems that modify the rules of the timing of cell division is the SOS system. The sfiA (sulA) gene of the SOS system (19) inhibits cell division when DNA has been damaged or when the chromosome cannot replicate. In addition, sfiC, a non-SOS gene, and genes such as the ccd gene of the low-copy-number F plasmid (20) have similar effects. All of these genes are normally repressed, but a basal level of activity could function upon occasion either inappropriately or even appropriately when some accidental damage to the DNA had occurred during normal balanced growth. The unusual activation of these genetic functions would cause a few cells in a growing population to "filament," i.e., to continue elongation without making constrictions or dividing. If the stimulus does not continue, such temporarily inhibited cells eventually recover, become repaired, and belatedly divide, giving rise to a few cells that take an extraordinarily long time to divide.

The frequency of these special cells in a bacterial population can be estimated from the frequency of nonnucleated cells. Normally, nonnucleated cells with approximately normal birth length constitute about 0.1% of the population in a culture in balanced growth. Recent experiments (19) have established that when the sfiA system does not work, 0.5% nonnucleated cells are found, and when both the sfiA and sfiC systems are inoperative, 0.7% nonnucleated cells are found. These results imply that in wild-type bacteria, 0.5 to 0.6% of scheduled cell divisions do not actually take place until some later time because of unusual SOS action, thereby distorting the distributions of age at division and size at division at their high ends. Consequently, a sufficient explanation of the skewness of the age-at-division distribution is the occasional operation of the SOS system acting in a temporary fashion even when not triggered by sufficient DNA damage to cause mass blockage of cell division in the population.

 

SYSTEMATIC CHANGES WITH GROWTH

I will now turn to the question of how cells differ from each other even though they originate from the same recently cloned culture.

 

The Canonical Case

When a cell divides, the two daughter cells are different from the parental cell from which they arose. Let us first consider the canonical model, in which every cell behaves like every other. This model has been used a great deal in physiological studies of various aspects of the microbial cell cycle. Although the model, predicated on precise division, is a poor representation of actual cultures, it would probably apply quite well to the hypothetical products of the continuous cultures considered above because evolution would select for even more precision in the evenness of division of cells. This idealized model has been used for two purposes: (i) calculating the average content of cell constituents per cell (e.g., the average DNA content per cell) and (ii) calculating the average time spent in a cell cycle phase from the fraction of the cells in the population showing a certain morphological or autoradiographic character. To cut through the mathematical complications (33, 43, 53, 48), in this ideal case a single cell taken at random from a growing population will have a mean age of 0.44 (with a standard deviation of ± 0.29) of the age at division (or the doubling time, symbolized by T ₂). Many cellular functions, such as the rate of protein synthesis, increase continuously during the cell cycle. Other functions, such as those having to do with chromosomal replication or cell division, change discontinuously. Consequently, if mass growth is continuous and increases in a monotonic fashion, the distribution of cell masses or sizes will have a mean size of 0.693 (with a standard deviation of ±0.14) of the size at division (critical size) (43, 53). The DNA content depends on the age in the cell cycle when chromosome replication starts, at age B, and when it ends, at age B+C, where B is the interval before initiation of chromosome replication and C is its duration. The age when DNA synthesis starts can also be expressed as T ₂ – D – C, where D is the interval between chromosome completion and division. In sum, even with clocklike precision there is considerable variability, in age, size, and DNA content, from cell to cell in a balanced growing population.

The canonical age and cell size distributions are applicable to a population of cells that grow according to some deterministic law (28) (i.e., every cell of a given size and age class does what every other cell of that class does according to same explicit rule) and in which every cell divides at a critical age into two identical daughter cells. The distribution of ages was first derived by Euler in the 1700s and independently rediscovered many times since (see reference 30). Its formula is given by

φ(a) da = 2μ e ^–μa da:  0 ≤ a ≤ ln 2/μ = T ₂

where μ is the specific growth rate, a is the age since birth, and φ(a) da is the frequency of cells whose ages are between a and a + da. The canonical age distribution is depicted in Fig. 2; its main characteristic is that the frequency decreases exponentially twofold from birth to division. This twofold factor results because one cell divides into exactly two daughters. The logic is simple: in cells that are just about to divide, whatever their number, there will be twice as many at a time infinitesimally later when they have divided into two newborn cells.

The distribution of cell sizes or masses in many ways is more useful than the distribution of ages (43, 49, 53). For the usual case of exponential growth, where the rate of biomass formation increases continuously in proportion to the biomass, a cell would increase in size twice as fast when it was about to divide as when it was just born. Consequently, within a given fixed mass range, dm, fewer cells will be found in a cell size class approaching division than in the class of newly separated cells. Therefore, there would be an additional twofold decrease because the rate of passage through a given size increment for those large and rapidly growing cells about to divide, relative to the rate of passage through the same size increment for newborn cells, is twice as great. The formula for this canonical distribution where the frequency drops fourfold between birth and division is given by Θ(m) dm = c/m ² dm as long as c/2 ≤ m ≤ c, where m is the cellular biomass and c is the critical size at cell division and all division events produce two identical daughters of mass c/2. This inverse-square distribution is shown in Fig. 3.

 

Variability in the Age, Size, and Evenness at Division

The age-at-division distribution is measured by watching cells arise, grow, and divide under the microscope or examining time-lapse films. If one could examine a population of cells during balanced growth at an instant of time and follow each cell forward and backward in time, one could construct both an extant age distribution and the momentary age-at-division distribution equivalent to those shown in Fig. 2. The extant size distribution is directly observable, and the size-at-division distribution is approximated by the distribution of cells showing constrictions. Figure 4 shows an example taken from the very many available in the publications from Nanninga’s laboratory in Amsterdam (58). The evenness of the division process culled from the many studies of the Amsterdam group has been brought together in the thesis of Trueba (70) (see also reference 17).

As far back as 1932 (24), it was clear that the age at division varied a good deal from one cell to another in a population of bacteria in good exponential growth. Typically the coefficient of variation (CV) of the age at division of random cells is about 20% (6, 29, 53, 65, 71). This high variability and the detailed shape of the distribution, particularly the positive skewness of the distribution, were much studied in the early literature and have been used in numerous speculative discussions. Some of the variability may be genetic, but most is something else. A little may be due to death and dying, but physiologic Daurer modifications or a change to the shutdown cell state is more important. In these early studies, however, the variation was presumed to be entirely stochastic. It was assumed that the broad skewing of available distributions reflected a small number of random discrete events that were needed to complete the cell cycle. Thus, it was suggested that for E. coli, there were 25 independent gene duplication events that had to be completed for a cell to be capable of division (64) or that there were 20 sequential processes that had to be completed (25, 26). Although from the modern perspective such interpretations seem naive, these papers were important in the development of the theory of stochastic processes and may be important for understanding the consequences of some of the phenomena discussed below.

At a later time, when cellular biochemical processes were better understood, a deterministic model for cell growth (53) was proposed as a simple extension of the canonical model discussed above. The extension assumed that the size at division of cells varied by a small amount as a result of inherent "biological variation" and that this generated a larger variation of age at division. Thus, the CV of the age at division is larger than the CV of the size at division because two cell division events (birth and division) determine the age at division, and only one event, division, determines the size at division.

Subsequently, the deterministic model has been modified in several minor ways and has been often renamed (11, 12, 36, 72). It has served as a useful way to study the cell cycle without the problems attendant to synchronization procedures (51, 52). A recent version of this model accounts for the age and size distributions as well as for correlations between the ages at division of mother and daughter cells and of sister cells (46). Other related properties of the population have also been calculated, but the key issue is that the variability of the size at division, no matter what its source, is the direct determinant of the broadness of the age-at-division distribution. However, positive skewness, which was the other frequently observed salient feature of the age-at-division distribution observed in most, but not all, of the relevant studies, is not so easily explained. Analytical and computer studies have shown that no matter whether the size-at-division distribution is approximately Gaussian (and, therefore, presumably due to many sources of variation) or due to a single stochastic event (negative exponential distribution), the resultant age-at-division distributions and the extant mass and extant age distributions change very little (43, 46). Control of the instant of cell division may trivially be dependent on fluctuations in the shear forces due to local stirring of cultures in aerated suspension or on the fluctuations in the mechanical stresses that develop in cells which grow when affixed to an agar surface. Variation in measured cell size at division may be trivially attributable to artifacts due to the cell’s orientation when viewed in the microscope and not to factors important or real with respect to the cell (63). These various possibilities affect the shape of the size-at-division distribution directly. However, because of the central limit theorem of statistics, the other two relevant distributions, the size at birth of cells and the distribution of ages in the entire population, are less affected.

The distribution of cell sizes at the instant of cell division is called the momentary distribution, and its CV is here designated q. The actual distribution could be measured directly only if a number of individual growing cells were followed through the cycle and their size, volume, or mass was noted at the exact instant of division. Practically, q is estimated from the distribution of cells in a population of cells showing evidence of impending division (46, 51, 52).

 

Possible Causes of the Positive Skewness of the Age-at-Division Distribution

In analyses of the cell cycle from the statistical viewpoint by Rahn (64) and Kendall (25, 26), the focus was on the observation that the age-at-division distribution was positively skewed in the samples of data available to them. With the development of the transition probability theory by Smith and colleagues (6, 67), another representation of the age-at-division distribution, called the alpha plot, became popular. An alpha plot is a semilogarithmic plot of the fraction of cells that have yet to divide. A linear tail is a precise prediction of the transition probability model in its original form that assumes that there is only one stochastic process involved. Many sets of data that did have a linear tail were found. Of course, any age-at-division distribution that is moderately positively skewed translates into an alpha plot that has a linear tail. Therefore, many instances of a positively skewed age-at-division distribution have been presented in the literature.

Since positive skewness is not predicted by the deterministic theory and its variants, it is important to establish the basis of the more commonly observed skewness. One suggestion (36) is that a small fraction of the cellular population is abnormal, i.e., pathological or defective because of a temporary error in the control of cell processes, and does not divide according to the same rules as does the bulk of the population. These could also be cells in an altered state that in the older literature were called Daurer modifications. Because of their rareness, such unusual events would be hard to study directly, but there are many phenomena observed in special ways that would cause the age-at-division distribution to have positive skewness.

 

Correlation of the Ages of Relatives

The age at division of one sister cell can differ from that of the other under presumably identical conditions. Experimentally, the correlation coefficient of the sister:sister age at division, CC_ss, is much less than unity; it is typically +0.5 to +0.8. Similarly, the correlation of the age at division of a mother, CC_md , with either of her daughters ranges from –0.5 to a slightly positive value. More distantly related cells in a clone are even more divergent in age at division and their correlation coefficients are closer to 0. For the theoretical case where the rate of formation of cell biomass is directly proportion to amount of cell biomass, these correlation coefficients are CC_ss = +0.5 and CC_md = –0.5. Further studies of how the correlation coefficients depend on evenness of the division process can be found in reference 46.

 

Distribution of the Oldest Cell Parts

Certain parts of the cell are subject to little or no turnover or regeneration. For example, a DNA strand created during semiconservative replication serves as a template in subsequent generations but remains inviolate in that act. Although the action of gyrase opens the chain, it also closes it without material change of the residues involved. On the other hand, repair systems (except photoreactivation) replace the material in portions of the chain. Excision repair replaces parts of the chain with new material, while recombination and postreplication repair exchange material from other strands. Nevertheless, the greatest portion of the DNA remains intact.

The other candidate for immortality is the cell wall. For Streptococcus spp., it is unquestionable that wall growth takes place in zones; after a pole is formed, it remains unchanged and does not turn over constituent molecules (10). (At least, no autolytic activity occurs unless the cell is stressed [e.g., temporary inhibition with chloramphenicol followed by removal of the chloramphenicol].) Although it was thought that E. coli represented a different case (55), it is now clear that the outer two layers of the pole wall are metabolically inert (57).

The distribution of poles or DNA strands of various ages is readily derived for the canonical case. If the formed strands were completely stable and if the culture had undergone balanced asynchronous growth for a long time, then the number of cells, N_T, that arose by division and formed a new pole at a prior time, T, would be N_T = Ne ^–μT, where N is the number of cells per unit volume at the time the culture is sampled. Since a frequency distribution must sum to unity, the frequency of cells in which the oldest pole or DNA strand has an age of T is φ(T) = μ ce ^–μT. Notice that μ replaces the 2μ of the canonical extant age distribution.

A different formula is necessary for cases in which cell senescence is slow and cells with very old poles or old DNA strands die. There are other cases in which the material in the cells is replaced gradually, but eventually the original cell part is no longer identifiable by morphological or radiochemical criteria. If this occurs after T_c doublings, then μ(T) has a zero value outside the range 0 ≤ T ≤ T_c and μ(T) = μ e ^–μT/(1 – e ^–μTc).

All of these distributions are positively skewed. The presence of poles or DNA strands of disparate ages could be important for certain cellular properties but not for others. These phenomena are clearly evident in the cellular process of fission yeasts and bacteria with specialized life cycles, such as Caulobacter crescentus (7, 16) and Rhodomicrobium vannielli (15).

Donachie (14) had argued quite persuasively that wall growth in E. coli is zonal and parallels the behavior of the fission yeast Schizosaccharomyces pombe. But my coworkers and I (37, 55), Burman et al. (8), and, most critically, Woldringh et al. (77) have demonstrated that wall growth is dispersive under the usual growth conditions. However, as Crawford Dow has pointed out (personal communication), there is no convincing evidence of dispersive growth during wall elongation of E. coli growing more slowly than a doubling time of 1 h. He points out that under conditions of very slow growth, E. coli may grow in a zonal manner and we not know it.

 

MECHANISMS FOR LONG-TERM ADAPTATION WITHOUT GENETIC LOSS

The last universal ancestor evidently had a very elaborate and refined cell physiology, and its descendants have improved on it. Bacteria, because of their short cell cycles, have therefore created for themselves an especially difficult problem that no amount of refinement of the cellular machinery to cope with a single environment will solve. Conditions change, and during the many generations of growth that occur under a new set of constraints, mutation and selection will lead to adaptation; however, seemingly most of the potential evolutionary advances would delete the improvements suitable to the previous set of conditions and, upon reestablishment of these cells, would be at a dead end. In responding to the need for change, and the much longer term needed to enable a return to the original state, the organisms must have evolved, in a still longer time frame, techniques that would allow such mutational reversibility. A number of the possibilities are presented in Table 5 and outline the next phases of the study of the basic strategy that bacteria use for long-term survival.